Communication Engineering Principles

Telegram: @Persian_IEEEComSoc
 Communication Engineering Principles

2nd Edition

Ifiok Otung
University of South Wales,
Pontypridd, UK
Telegram: @Persian_IEEEComSoc
 This edition first published 2021
© 2021 John Wiley & Sons Ltd

Edition History
2001 (1e, Palgrave Macmillan)

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any
means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by law. Advice on how to obtain permission
to reuse material from this title is available at http://www.wiley.com/go/permissions.

The right of Ifiok Otung to be identified as the author of this work has been asserted in accordance with law.

Registered Offices
John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, USA
John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, UK

Editorial Office
The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, UK

For details of our global editorial offices, customer services, and more information about Wiley products visit us at www.wiley.com.

Wiley also publishes its books in a variety of electronic formats and by print-on-demand. Some content that appears in standard print
versions of this book may not be available in other formats.

Limit of Liability/Disclaimer of Warranty

While the publisher and authors have used their best efforts in preparing this work, they make no representations or warranties with
respect to the accuracy or completeness of the contents of this work and specifically disclaim all warranties, including without
limitation any implied warranties of merchantability or fitness for a particular purpose. No warranty may be created or extended by
sales representatives, written sales materials or promotional statements for this work. The fact that an organization, website, or
product is referred to in this work as a citation and/or potential source of further information does not mean that the publisher and
authors endorse the information or services the organization, website, or product may provide or recommendations it may make. This
work is sold with the understanding that the publisher is not engaged in rendering professional services. The advice and strategies
contained herein may not be suitable for your situation. You should consult with a specialist where appropriate. Further, readers
should be aware that websites listed in this work may have changed or disappeared between when this work was written and when it
is read. Neither the publisher nor authors shall be liable for any loss of profit or any other commercial damages, including but not
limited to special, incidental, consequential, or other damages.

Library of Congress Cataloging-in-Publication Data

Names: Otung, Ifiok, author.
Title: Communication engineering principles / Ifiok Otung, University
of South Wales, Pontypridd, UK.
Other titles: Digital communications
Description: 2nd edition. | Hoboken, NJ : John Wiley & Sons, Inc., USA,
[2021] | Revised edition of Digital communications : principles and
systems / Ifiok Otung. 2014. | Includes bibliographical references and
index.
Identifiers: LCCN 2020020271 (print) | LCCN 2020020272 (ebook) | ISBN
Telegram: @Persian_IEEEComSoc

9781119274025 (cloth) | ISBN 9781119273967 (adobe pdf) | ISBN

9781119274070 (epub)
Subjects: LCSH: Digital communications. | Digital communications–Equipment
and supplies–Design and construction.
Classification: LCC TK5103.7 .O88 2021 (print) | LCC TK5103.7 (ebook) |
DDC 621.382–dc23
LC record available at https://lccn.loc.gov/2020020271
LC ebook record available at https://lccn.loc.gov/2020020272

Cover Design: Wiley

Cover Image: © greenbutterfly/Shutterstock

Set in 9.5/12.5pt STIXTwoText by SPi Global, Chennai, India

10 9 8 7 6 5 4 3 2 1
 In memory of the short long lives of my parents Charles and Sylvia
and the long short lives of my sisters Alice, Theresa, and Lucy.

Go back in time until you can go no further,

For trillions of years if you must.
What do you see at this beginning?
You see a particle, or you see God.
Neither is more scientific, neither more religious.
Both involve a leap of faith beyond reason.

This book is dedicated to God, Who

In the beginning created the heavens and the earth.
Telegram: @Persian_IEEEComSoc
 vii

Contents

Preface xxi
Acknowledgements xxiii
About the Companion Website xxv

1 Overview of Communication Systems 1

1.1 Introduction 1
1.2 Nonelectrical Telecommunication 2
1.2.1 Verbal Nonelectrical Telecommunication 2
1.2.2 Visual Nonelectrical Telecommunication 3
1.2.2.1 Flags, Smoke, and Bonfires 3
1.2.2.2 Heliography 4
1.2.2.3 Semaphore 4
1.2.2.4 Demerits of Visual Nonelectrical Telecommunication 5
1.3 Modern Telecommunication 5
1.3.1 Developments in Character Codes 7
1.3.1.1 Morse Code 7
1.3.1.2 Baudot Code 7
1.3.1.3 Hollerith Code 8
1.3.1.4 EBCDIC Code 9
1.3.1.5 ASCII Code 9
1.3.1.6 ISO 8859 Code 10
1.3.1.7 Unicode 11
1.3.2 Developments in Services 13
1.3.2.1 Telegram 13
Telegram: @Persian_IEEEComSoc

1.3.2.2 Telex 14
1.3.2.3 Facsimile 14
1.3.2.4 The Digital Era 15
1.3.3 Developments in Transmission Media 16
1.3.3.1 Copper Cable 17
1.3.3.2 Radio 18
1.3.3.3 Optical Fibre 19
1.4 Communication System Elements 21
1.4.1 Information Source 21
1.4.1.1 Audio Input Devices 22
1.4.1.2 Video Input Devices 23
 viii Contents

1.4.1.3 Data Input Devices 23

1.4.1.4 Sensors 23
1.4.2 Information Sink 24
1.4.2.1 Audio Output Device 24
1.4.2.2 Visual Display Devices 26
1.4.2.3 Storage Devices 28
1.4.3 Transmitter 29
1.4.4 Receiver 31
1.5 Classification of Communication Systems 32
1.5.1 Simplex Versus Duplex Communication Systems 32
1.5.2 Analogue Versus Digital Communication Systems 33
1.5.3 Baseband Versus Modulated Communication Systems 35
1.5.3.1 Analogue Baseband Communication System 35
1.5.3.2 Discrete Baseband Communication System 36
1.5.3.3 Digital Baseband Communication System 41
1.5.3.4 Modulated Communication Systems 44
1.5.4 Circuit Versus Packet Switching 47
1.5.4.1 Circuit Switching 48
1.5.4.2 Packet Switching 50
1.6 Epilogue 53
References 53
Review Questions 53

2 Introduction to Signals and Systems 57

2.1 Introduction 57
2.2 What Is a Signal? 58
2.3 Forms of Telecommunication Signals 58
2.4 Subjective Classification of Telecommunication Signals 60
2.4.1 Speech 60
2.4.2 Music 62
2.4.3 Video 63
2.4.4 Digital Data 64
2.4.5 Facsimile 64
2.4.6 Ancillary and Control Signals 65
2.5 Objective Classification of Telecommunication Signals 65
2.5.1 Analogue or Digital 65
Telegram: @Persian_IEEEComSoc

2.5.2 Periodic or Nonperiodic 67

2.5.3 Deterministic or Random 68
2.5.4 Power or Energy 69
2.5.5 Even or Odd 69
2.6 Special Waveforms and Signals 71
2.6.1 Unit Step Function 74
2.6.2 Signum Function 74
2.6.3 Rectangular Pulse 75
2.6.4 Ramp Pulse 76
2.6.5 Triangular Pulse 77
2.6.6 Sawtooth and Trapezoidal Pulses 77
 Contents ix

2.6.7 Unit Impulse Function 78

2.6.8 Sinc Function 79
2.7 Sinusoidal Signals 81
2.7.1 Qualitative Introduction 82
2.7.2 Parameters of a Sinusoidal Signal 83
2.7.2.1 Angle 86
2.7.2.2 Amplitude 87
2.7.2.3 Angular Frequency 87
2.7.2.4 Frequency 87
2.7.2.5 Period 88
2.7.2.6 Wavelength 88
2.7.2.7 Initial Phase 88
2.7.2.8 Phase Difference 89
2.7.3 Addition of Sinusoids 92
2.7.3.1 Same Frequency and Phase 93
2.7.3.2 Same Frequency but Different Phases 93
2.7.3.3 Multiple Sinusoids of Different Frequencies 97
2.7.3.4 Beats Involving Two Sinusoids 97
2.7.4 Multiplication of Sinusoids 99
2.8 Logarithmic Units 99
2.8.1 Logarithmic Units for System Gain 101
2.8.2 Logarithmic Units for Voltage, Power, and Other Quantities 102
2.8.3 Logarithmic Unit Dos and Don’ts 104
2.9 Calibration of a Signal Transmission Path 107
2.10 Systems and Their Properties 109
2.10.1 Memory 109
2.10.2 Stability 111
2.10.3 Causality 112
2.10.4 Linearity 113
2.10.5 Time Invariance 116
2.10.6 Invertibility 118
2.11 Summary 121
Questions 122

3 Time Domain Analysis of Signals and Systems 127

3.1 Introduction 127
Telegram: @Persian_IEEEComSoc

3.2 Basic Signal Operations 128

3.2.1 Time Shifting (Signal Delay and Advance) 128
3.2.2 Time Reversal 130
3.2.3 Time Scaling 132
3.3 Random Signals 134
3.3.1 Random Processes 134
3.3.2 Random Signal Parameters 135
3.3.3 Stationarity and Ergodicity 138
3.4 Standard Distribution Functions 139
3.4.1 Gaussian or Normal Distribution 139
3.4.2 Rayleigh Distribution 143
 x Contents

3.4.3 Lognormal Distribution 148

3.4.4 Rician Distribution 153
3.4.5 Exponential and Poisson Distributions 157
3.5 Signal Characterisation 162
3.5.1 Mean 162
3.5.2 Power 163
3.5.3 Energy 167
3.5.4 Root-mean-square Value 168
3.5.5 Autocorrelation 171
3.5.6 Covariance and Correlation Coefficient 176
3.6 Linear Time Invariant System Analysis 180
3.6.1 LTI System Response 181
3.6.2 Evaluation of Convolution Integral 186
3.6.3 Evaluation of Convolution Sum 190
3.6.4 Autocorrelation and Convolution 196
3.7 Summary 197
References 198
Questions 198

4 Frequency Domain Analysis of Signals and Systems 203

4.1 Introduction 203
4.2 Fourier Series 205
4.2.1 Sinusoidal Form of Fourier Series 206
4.2.1.1 Time Shifting 211
4.2.1.2 Time Reversal 212
4.2.1.3 Even and Odd Functions 212
4.2.1.4 Piecewise Linear Functions 214
4.2.2 Complex Exponential Form of Fourier Series 222
4.2.3 Amplitude and Phase Spectra 224
4.2.3.1 Double-sided Spectrum 227
4.2.3.2 Single-sided Spectrum 228
4.2.4 Fourier Series Application to Selected Waveforms 235
4.2.4.1 Flat-top-sampled Signal 235
4.2.4.2 Binary ASK Signal and Sinusoidal Pulse Train 243
4.2.4.3 Trapezoidal Pulse Train 248
4.3 Fourier Transform 253
Telegram: @Persian_IEEEComSoc

4.3.1 Properties of the Fourier Transform 257

4.3.1.1 Even and Odd Functions 257
4.3.1.2 Linearity 258
4.3.1.3 Time Shifting 258
4.3.1.4 Frequency Shifting 258
4.3.1.5 Time Scaling 258
4.3.1.6 Time Reversal 259
4.3.1.7 Complex Conjugation 259
4.3.1.8 Duality 259
4.3.1.9 Differentiation 259
4.3.1.10 Integration 260
 Contents xi

4.3.1.11 Multiplication 260

4.3.1.12 Convolution 260
4.3.1.13 Areas 260
4.3.1.14 Energy 261
4.3.2 Table of Fourier Transforms 263
4.3.3 Fourier Transform of Periodic Signals 268
4.4 Discrete Fourier Transform 270
4.4.1 Properties of the Discrete Fourier Transform 275
4.4.1.1 Periodicity 275
4.4.1.2 Symmetry 276
4.4.2 Fast Fourier Transform 277
4.4.3 Practical Issues in DFT Implementation 283
4.4.3.1 Aliasing 283
4.4.3.2 Frequency Resolution 284
4.4.3.3 Spectral Leakage 285
4.4.3.4 Spectral Smearing 285
4.4.3.5 Spectral Density and Its Variance 288
4.5 Laplace and z-transforms 291
4.5.1 Laplace Transform 291
4.5.2 z-transform 292
4.6 Inverse Relationship Between Time and Frequency Domains 295
4.7 Frequency Domain Characterisation of LTI Systems 297
4.7.1 Transfer Function 297
4.7.2 Output Spectral Density of LTI Systems 301
4.7.3 Signal and System Bandwidths 302
4.7.3.1 Subjective Bandwidth 303
4.7.3.2 Null Bandwidth 303
4.7.3.3 3 dB Bandwidth 304
4.7.3.4 Fractional Power Containment Bandwidth 306
4.7.3.5 Noise Equivalent Bandwidth 308
4.7.4 Distortionless Transmission 311
4.7.5 Attenuation and Delay Distortions 313
4.7.6 Nonlinear Distortions 314
4.8 Summary 316
References 317
Questions 317
Telegram: @Persian_IEEEComSoc

5 Transmission Media 327

5.1 Introduction 327
5.2 Metallic Line Systems 328
5.2.1 Wire Pairs 328
5.2.2 Coaxial Cable 332
5.2.3 Attenuation in Metallic Lines 333
5.3 Transmission Line Theory 334
5.3.1 Incident and Reflected Waves 337
5.3.2 Secondary Line Constants 338
5.3.3 Characteristic Impedance 340
 xii Contents

5.3.4 Reflection and Transmission Coefficients 342

5.3.5 Standing Waves 345
5.3.6 Line Impedance and Admittance 347
5.3.7 Line Termination and Impedance Matching 353
5.3.8 Scattering Parameters 359
5.3.9 Smith Chart 363
5.4 Optical Fibre 365
5.4.1 Optical Fibre Types 367
5.4.1.1 Single-mode Step Index 368
5.4.1.2 Multimode Step Index Fibre 368
5.4.1.3 Multimode Graded Index 369
5.4.2 Coupling of Light into Fibre 369
5.4.3 Attenuation in Optical Fibre 371
5.4.3.1 Intrinsic Fibre Loss 371
5.4.3.2 Extrinsic Fibre Loss 375
5.4.4 Dispersion in Optical Fibre 376
5.4.4.1 Intermodal Dispersion 376
5.4.4.2 Intramodal Dispersion 377
5.5 Radio 380
5.5.1 Maxwell’s Equations 382
5.5.2 Radio Wave Propagation Modes 384
5.5.2.1 Ground Wave 386
5.5.2.2 Sky Wave 386
5.5.2.3 Line-of-sight (LOS) 387
5.5.2.4 Satellite Communications 387
5.5.2.5 Mobile Communications 388
5.5.2.6 Ionospheric Scatter 388
5.5.2.7 Tropospheric Scatter 388
5.5.3 Radio Wave Propagation Effects 388
5.5.3.1 Ionospheric Effects 388
5.5.3.2 Tropospheric Attenuation 390
5.5.3.3 Tropospheric Scintillation 393
5.5.3.4 Depolarisation 394
5.5.3.5 Tropospheric Refraction 395
5.5.4 Reflection and Refraction 397
5.5.5 Rough Surface Scattering 406
Telegram: @Persian_IEEEComSoc

5.5.6 Diffraction 408

5.5.6.1 Diffraction Configuration and Terms 408
5.5.6.2 Fresnel Zones 410
5.5.6.3 Knife-edge Diffraction Loss 411
5.5.7 Path Loss 416
5.5.7.1 Free Space Path Loss 416
5.5.7.2 Plane Earth Propagation Path Loss 418
5.5.7.3 Terrestrial Cellular Radio Path Loss 421
5.5.8 Radio Frequency Allocation 424
5.6 Summary 424
References 425
Questions 426
 Contents xiii

6 Noise in Communication Systems 431

6.1 Introduction 431
6.2 Physical Sources of Random Noise 432
6.2.1 Thermal or Johnson Noise 432
6.2.2 Quantisation Noise 433
6.2.3 Radio or Sky Noise 433
6.2.4 Shot Noise 435
6.2.5 Partition Noise 435
6.2.6 Quantum Noise 435
6.2.7 Flicker or 1/f Noise 436
6.3 Additive White Gaussian Noise 437
6.3.1 Gaussian PDF of Noise 438
6.3.2 White Noise 439
6.3.3 Canonical and Envelope Representations of Noise 444
6.4 System Noise Calculations 448
6.4.1 Available Noise Power 448
6.4.2 Equivalent Noise Temperature 450
6.4.3 Noise Figure of a Single System 451
6.4.4 Noise Figure of Cascaded Systems 454
6.4.5 Overall System Noise Temperature 457
6.4.6 Signal-to-noise Ratio 459
6.5 Noise Effects in Communication Systems 462
6.5.1 SNR in Analogue Communication Systems 462
6.5.2 BER in Digital Communication Systems 465
6.6 Summary 469
References 470
Questions 470

7 Amplitude Modulation 473

7.1 Introduction 473
7.2 AM Signals: Time Domain Description 474
7.2.1 AM Waveform 474
7.2.2 Sketching AM Waveforms 475
7.2.3 Modulation Factor 476
7.3 Spectrum and Power of Amplitude Modulated Signals 480
7.3.1 Sinusoidal Modulating Signal 480
Telegram: @Persian_IEEEComSoc

7.3.2 Arbitrary Message Signal 482

7.3.3 Power 485
7.4 AM Modulators 488
7.4.1 Generation of AM Signal 488
7.4.1.1 Linearly-varied-gain Modulator 488
7.4.1.2 Switching and Square-law Modulators 489
7.4.2 AM Transmitters 491
7.4.2.1 Low-level Transmitter 491
7.4.2.2 High-level Transmitter 492
7.5 AM Demodulators 492
7.5.1 Diode Demodulator 493
7.5.2 Coherent Demodulator 496
 xiv Contents

7.5.3 AM Receivers 498

7.5.3.1 Tuned Radio Frequency (RF) Receiver 498
7.5.3.2 Superheterodyne Receiver 499
7.6 Merits, Demerits, and Application of AM 501
7.7 Variants of AM 502
7.7.1 DSB 502
7.7.1.1 Waveform and Spectrum of DSB 502
7.7.1.2 DSB Modulator 504
7.7.1.3 DSB Demodulator 507
7.7.1.4 DSB Applications 509
7.7.2 SSB 510
7.7.2.1 Merits and Demerits of SSB 511
7.7.2.2 SSB Modulators 514
7.7.2.3 SSB Demodulator 516
7.7.2.4 Applications of SSB 517
7.7.3 ISB 518
7.7.3.1 ISB Modulator 518
7.7.3.2 ISB Demodulator 518
7.7.3.3 ISB Merits, Demerit, and Application 520
7.7.4 VSB 520
7.7.4.1 VSB Modulator 521
7.7.4.2 VSB Demodulator 522
7.8 Summary 524
Questions 525

8 Frequency and Phase Modulation 529

8.1 Introduction 529
8.2 Basic Concepts of FM and PM 530
8.2.1 Frequency Modulation Concepts 531
8.2.2 Phase Modulation Concepts 535
8.2.3 Relationship Between FM and PM 537
8.2.3.1 Frequency Variations in PM 537
8.2.3.2 Phase Variations in FM 540
8.3 FM and PM Waveforms 543
8.3.1 Sketching Simple Waveforms 543
8.3.2 General Waveform 544
Telegram: @Persian_IEEEComSoc

8.4 Spectrum and Power of FM and PM 549

8.4.1 Narrowband FM and PM 549
8.4.1.1 Frequency Components 549
8.4.1.2 Comparing AM, NBFM, and NBPM 551
8.4.1.3 Amplitude Variations in NBFM and NBPM 556
8.4.2 Wideband FM and PM 557
8.4.2.1 Spectrum 558
8.4.2.2 Power 560
8.4.2.3 Bandwidth 563
8.4.2.4 FM or PM? 567
8.5 FM and PM Modulators 567
 Contents xv

8.5.1 Narrowband Modulators 567

8.5.2 Indirect Wideband Modulators 569
8.5.3 Direct Wideband Modulators 572
8.5.3.1 LCO Modulator 573
8.5.3.2 VCO Modulator 575
8.6 FM and PM Demodulators 576
8.6.1 Direct Demodulator 577
8.6.1.1 Filter-based Demodulator 577
8.6.1.2 Digital Demodulator 577
8.6.2 Indirect Demodulator 577
8.6.2.1 PLL Demodulation Process 579
8.6.2.2 PLL States 580
8.6.2.3 PLL Features 580
8.6.3 Phase Demodulator 580
8.6.4 Frequency Discriminators 581
8.6.4.1 Differentiators 581
8.6.4.2 Tuned Circuits 583
8.7 FM Transmitter and Receiver 584
8.7.1 Transmitter 584
8.7.2 SNR and Bandwidth Trade-off 586
8.7.3 Pre-emphasis and De-emphasis 586
8.7.4 Receiver 588
8.8 Noise Effect in FM 588
8.9 Overview of FM and PM Features 594
8.9.1 Merits 594
8.9.2 Demerits 594
8.9.3 Applications 595
8.10 Summary 595
Questions 595

9 Sampling 599
9.1 Introduction 599
9.2 Sampling Theorem 599
9.3 Proof of Sampling Theorem 600
9.3.1 Lowpass Signals 602
Telegram: @Persian_IEEEComSoc

9.3.2 Bandpass Signals 603

9.3.3 Sampling at Nyquist Rate 606
9.4 Aliasing 607
9.5 Anti-alias Filter 613
9.6 Non-instantaneous Sampling 615
9.6.1 Natural Sampling 616
9.6.2 Flat-top Sampling 618
9.6.3 Aperture Effect 622
9.7 Summary 623
Questions 624
Reference 625
 xvi Contents

10 Digital Baseband Coding 627

10.1 Introduction 627
10.2 Concept and Classes of Quantisation 628
10.3 Uniform Quantisation 634
10.3.1 Quantisation Noise 635
10.3.2 Dynamic Range of a Quantiser 636
10.3.3 Signal-to-quantisation-noise Ratio (SQNR) 636
10.3.4 Design Considerations 639
10.3.5 Demerits of Uniform Quantisation 640
10.4 Nonuniform Quantisation 641
10.4.1 Compressor Characteristic 642
10.4.2 A-law Companding 644
10.4.3 𝜇-law Companding 645
10.4.4 Companding Gain and Penalty 647
10.4.5 Practical Nonlinear PCM 650
10.4.6 SQNR of Practical Nonlinear PCM 657
10.5 Differential PCM (DPCM) 661
10.5.1 Adaptive Differential Pulse Code Modulation (ADPCM) 664
10.5.2 Delta Modulation 664
10.5.2.1 Quantisation Error 664
10.5.2.2 Prediction Filter 666
10.5.2.3 Design Parameters 666
10.5.2.4 Merits and Demerits of DM 666
10.5.2.5 Adaptive Delta Modulation (ADM) 668
10.5.2.6 Delta Sigma Modulation 668
10.6 Low Bit Rate Speech Coding 668
10.6.1 Waveform Coders 671
10.6.2 Vocoders 671
10.6.2.1 IMBE 672
10.6.2.2 LPC 672
10.6.2.3 MELP 673
10.6.3 Hybrid Coders 673
10.6.3.1 APC 673
10.6.3.2 MPE-LPC 673
10.6.3.3 CELP 673
10.7 Line Codes 674
Telegram: @Persian_IEEEComSoc

10.7.1 NRZ Codes 674

10.7.2 RZ Codes 675
10.7.3 Biphase Codes 676
10.7.4 RLL Codes 676
10.7.5 Block Codes 677
10.8 Summary 680
Reference 681
Questions 681

11 Digital Modulated Transmission 683

11.1 Introduction 683
 Contents xvii

11.2 Orthogonality of Energy Signals 687

11.3 Signal Space 689
11.3.1 Interpretation of Signal-space Diagrams 690
11.3.2 Complex Notation for 2D Signal Space 693
11.3.3 Signal-space Worked Examples 694
11.4 Digital Transmission Model 699
11.5 Noise Effects 701
11.6 Symbol and Bit Error Ratios 703
11.6.1 Special Cases 705
11.6.2 Arbitrary Binary Transmission 708
11.7 Binary Modulation 712
11.7.1 ASK 712
11.7.2 PSK 714
11.7.3 FSK 715
11.7.3.1 Generation 715
11.7.3.2 Spectrum 716
11.7.3.3 Frequency Spacing and MSK 716
11.7.4 Minimum Transmission Bandwidth 718
11.8 Coherent Binary Detection 719
11.8.1 ASK Detector 719
11.8.2 PSK Detector 721
11.8.3 FSK Detector 721
11.9 Noncoherent Binary Detection 723
11.9.1 Noncoherent ASK Detector 725
11.9.2 Noncoherent FSK Detector 727
11.9.3 DPSK 727
11.10 M-ary Transmission 730
11.10.1 Bandwidth Efficiency 730
11.10.2 M-ary ASK 732
11.10.2.1 M-ary ASK Modulator 732
11.10.2.2 M-ary ASK Detector 734
11.10.2.3 BER of M-ary ASK 734
11.10.3 M-ary PSK 737
11.10.3.1 QPSK Modulator and Detector 738
11.10.3.2 M-ary PSK Modulator and Detector 740
11.10.3.3 BER of M-ary PSK 743
Telegram: @Persian_IEEEComSoc

11.10.4 M-ary FSK 746

11.10.4.1 M-ary FSK Modulator and Detector 746
11.10.4.2 BER of M-ary FSK 746
11.10.4.3 Noise-bandwidth Trade-off in M-ary FSK 748
11.10.5 M-ary APSK 749
11.10.5.1 16-APSK 749
11.10.5.2 BER of Square M-ary APSK 752
11.11 Design Parameters 754
11.12 Summary 757
Reference 758
Questions 759
 xviii Contents

12 Pulse Shaping and Detection 763

12.1 Introduction 763
12.2 Anti-ISI Filtering 765
12.2.1 Nyquist Filtering 767
12.2.2 Raised Cosine Filtering 769
12.2.3 Square Root Raised Cosine Filtering 771
12.2.4 Duobinary Signalling 774
12.2.4.1 Cosine Filter 774
12.2.4.2 Signal Power Trade-off 777
12.2.4.3 Sine Filter 778
12.2.4.4 Polybinary Signalling 779
12.3 Information Capacity Law 780
12.4 The Digital Receiver 787
12.4.1 Adaptive Equalisation 787
12.4.2 Matched Filter 787
12.4.2.1 Specification of a Matched Filter 788
12.4.2.2 Matched Filter by Correlation 790
12.4.2.3 Matched Filter Worked Examples 791
12.4.3 Clock Extraction 797
12.4.4 Eye Diagrams 799
12.5 Summary 799
References 801
Questions 801

13 Multiplexing Strategies 805

13.1 Introduction 805
13.2 Frequency Division Multiplexing 809
13.2.1 General Concepts 809
13.2.2 Demerits of Flat-level FDM 812
13.2.3 Future of FDM Technology 813
13.2.4 FDM Hierarchies 814
13.2.4.1 UK System 816
13.2.4.2 European System 820
13.2.4.3 Bell System 821
13.2.4.4 Nonvoice Signals 822
13.2.5 Wavelength Division Multiplexing 823
Telegram: @Persian_IEEEComSoc

13.3 Time Division Multiplexing 825

13.3.1 General Concepts 825
13.3.2 Plesiochronous Digital Hierarchy 827
13.3.2.1 E1 System 827
13.3.2.2 T1 and J1 Systems 832
13.3.2.3 PDH Problems 838
13.3.3 Synchronous Digital Hierarchy 838
13.3.3.1 SDH Rates 839
13.3.3.2 SDH Frame Structure 839
13.3.3.3 SONET 844
13.3.4 ATM 846
 Contents xix

13.3.4.1 ATM Layered Architecture 847

13.3.4.2 ATM Network Components 850
13.3.4.3 ATM Cell Header 851
13.3.4.4 ATM Features Summary 852
13.3.4.5 ATM Versus IP 852
13.4 Code Division Multiplexing 853
13.4.1 Types of Spread Spectrum Modulation 853
13.4.2 CDM Transmitter 856
13.4.3 CDM Receiver 858
13.4.4 Crucial Features of CDM 863
13.4.4.1 Synchronisation 863
13.4.4.2 Cross-correlation of PN Codes 864
13.4.4.3 Power Control 864
13.4.4.4 Processing Gain 866
13.5 Multiple Access 867
13.5.1 FDMA 868
13.5.2 TDMA 869
13.5.3 CDMA 871
13.5.4 Hybrid Schemes 872
13.6 Summary 873
Questions 874

Appendix A Character Codes 877

Appendix B Trigonometric Identities 883

Appendix C Tables and Constants 885

C.1 Constants 885
C.2 SI Units 886
C.3 Complementary Error Function erfc(x) and Q function Q(x) 887

Index 891
Telegram: @Persian_IEEEComSoc
 xxi

Preface

If nature gives you a free lunch, you pay with your dinner.

This second edition of Communication Engineering Principles is a painstaking and comprehensive revision of the
original publication, including several new chapters and a complete rewrite of some of the old chapters. I have
remained faithful to the approach and philosophy that made the first edition so successful. It is an engineering-first
approach inspired by an engineering-is-fun philosophy. I have left no stone unturned to ensure complete clarity
and to break complex concepts into their simple bite-sized components for the benefit and enjoyment of every
reader and tutor.
Communication Engineering Principles is aimed at undergraduate courses in communication engineering, digi-
tal communications, and signals and systems analysis. It is also suitable as preparatory material for MSc students
and for researchers and practising engineers wishing to fully understand and apply the concepts and principles of
the subject in their area of work. The book prioritises clarity and engineering insight above mathematical rigour,
although maths is an essential tool that is used when necessary. Analogies, graphs, heuristic arguments, and
numerous worked examples are used to deepen the reader’s insight and hone their skills in problem solving and
the correct interpretation and application of key concepts and principles.
Chapter 1, Overview of Communication Systems, is a nonmathematical overview of communication systems
that erects crucial knowledge pegs needed to hang a more detailed treatment of the subject in subsequent chapters.
It also presents a carefully selected review of our journey from telegraphy in 1837 to 5G in 2020 and a discussion of
the main communication system elements and processes. This is an extensive update of the first chapter of the first
edition. In addition to a detailed update to reflect the state of the art of telecoms in 2020, new material has been
added on circuit and packet switching, character coding, developments in transmission media, and the digital era.
It is in this chapter that we discover that ATM is a slow-start sprinter, whereas IP is an instant-start jogger and we
learn the different attitudes of each technique towards sharing the community cake.
Telegram: @Persian_IEEEComSoc

Chapter 2, Introduction to Signals and Systems, is a new chapter that retains some of the material in the old
Chapter 2, which was titled Telecommunication Signals. It is a must-read for everyone, including those who
already have some familiarity with some of the topics discussed. We lay a great foundation for dealing with signals
and systems in engineering. This includes an exhaustive treatment of sinusoidal signals, the building blocks of all
other signals, and an introduction to various system properties. It is also in this chapter that we learn 10 logarith-
mic dos and don’ts. For example, did you know that you should never add together two dBW values, although you
may subtract them?
Chapter 3, Time Domain Analysis of Signals and Systems, is a new chapter that deals with various signal oper-
ations from time reversal and delay to convolution and autocorrelation. We use a graphical approach and various
worked examples to make it easy to fully master these important operations. Random signals are also discussed and
the statistical distributions that are most used for telecom systems and services analysis and modelling are fully
 xxii Preface

covered. The last part of the chapter is then devoted to learning how to characterise and analyse linear systems in
the time domain.
Chapter 4, Frequency Domain Analysis of Signals and Systems, is new. Using a mix of heuristic, graphical and
mathematical approaches, we explore the topics of Fourier series, Fourier transform, and discrete Fourier trans-
form at a depth and breadth that are considered complete for the needs of modern engineering. We explore new
applications of the tools and at all points emphasise the correct interpretation of results. The chapter ends with
careful coaching on the use of a frequency domain approach in system characterisation and analysis.
Chapter 5, Transmission Media, is also new. A nonmathematical discussion of the characterisation, signal
impairments, and applications of metallic lines, optical fibre, and radio is followed by a more in-depth analysis to
develop the tools needed to calculate signal strength at various points in each medium. Transmission line theory
is also covered in full.
Chapter 6, Noise in Communication Systems, is an update of the old Chapter 9 that went by the same title.
The update includes new worked examples and improvements in presentation and discussion. We acquire a good
grounding in the quantification of random noise and the assessment of their impact on digital and analogue com-
munication systems. The design parameters that affect SNR in analogue systems and BER in digital systems are
explored in detail.
Chapter 7, Amplitude Modulation, is an update of the old Chapter 3 that was similarly titled. It gives a compre-
hensive treatment of amplitude modulation and all its variants.
Chapter 8, Frequency and Phase Modulation, retains much of the material of the old Chapter 4 titled Angle
Modulation. The treatment of noise effects using a phasor approach is improved. New worked examples are also
included.
Chapter 9, Sampling, retains much of the old Chapter 5 that was similarly titled. The treatment of bandpass
sampling is improved, and new graphical illustrations are employed.
Chapter 10, Digital Baseband Coding, is an extensive revision of the previous Chapter 6 titled Digital Baseband
Transmission. The treatment of quantisation and PCM is improved.
Chapter 11, Digital Modulated Transmission, is an extensive revision of the old Chapter 7 that was similarly
titled. New material is introduced on signal orthogonality, signal-space diagrams, bandwidth efficiency, design
parameters, and bit error ratios. New worked examples are also introduced.
Chapter 12, Pulse Shaping and Detection, is new. We develop various filtering measures for mitigating inter-
symbol interference (ISI), evaluate the Shannon–Hartley information capacity law, and derive the matched filter
for optimum detection of a signal in additive white Gaussian noise. Various worked examples are also presented.
Chapter 13, Multiplexing Strategies, is an extensive revision of the previous Chapter 8 that was similarly titled.
A new section on multiple access is introduced and the treatment of all topics, including wavelength division
multiplexing, is updated and improved. New worked examples are also added.
The entire Chapter 1 and up to Section 2.5 of Chapter 2 is nonmathematical. This is in keeping with our
engineering-first approach. We wanted engineering, rather than maths, to be our gatekeeper to welcome you
Telegram: @Persian_IEEEComSoc

to the beauty and fun of telecoms as presented in this volume. Beyond Section 2.5 it is assumed that you
have a knowledge of calculus, although a lot of explanation of mathematical manipulations is provided as
deemed necessary. In all cases, however, we approach every concept and every problem solving by starting with
engineering, bringing in maths if necessary, and then ending with engineering through a careful interpretation
of any mathematical results.
I hope that you will enjoy using this book as much as I enjoyed writing it. I look forward to hearing how this book
has helped your work, whether as student or tutor. Please visit my website at https://professorifiokotung.com/ for
further support, including video clips and presentations that could help make your study easier and even more
exciting.
 xxiii

Acknowledgements

The communication engineering principles, technologies, and standards covered in this book are the culmination
of the efforts of many people and organisations over several generations. This book owes its very existence to these
pillars of our subject and to the stellar work of the International Telecommunication Union (ITU) in developing
many of the technical standards reflected within.
I am grateful to Simon Haykin, whose writings played a significant role in guiding my first steps into commu-
nication systems theory in the 1980s. Since then, my journey in the subject has been further shaped through the
contributions of others too numerous to mention. However, this book brings a unique approach born out of many
years of teaching the subject matter to international cohorts of students and engineers with diverse mathematical
abilities. The book’s style and attention to detail are motivated by a strong belief in simplicity and the necessity
of clarity, and an uncompromising dedication to training competent engineers with a complete understanding of
the underlying principles of the subject as well as excellent skills in communication system analysis and design.
I am indebted to generations of my undergraduate and postgraduate students, short course participants, and
users of the first edition of the book whose feedback and learning requirements helped in no small measure to
shape the style and content of this second edition. I thank my colleagues Dr Ali Roula and Professor Jonathan
Rodriguez for their support with some administrative and research responsibilities while the book was in prepa-
ration. I also thank my research student Ms Jinwara Surattanagul for her secretarial contributions to some parts
of the later chapters.
This book could not have materialised without Sandra Grayson, the Wiley commissioning editor for electrical
and computer engineering. I thank her for her faith in the project from conception to completion and for her
professional support and patience during the many months it took to combine my university responsibilities with
writing. I also thank Jayaprakash Unni and the production team at Wiley who worked diligently to take the book
from manuscript through to final publication.
I thank my family for their patient support, wholehearted encouragement, and infectious faith in the completion
of the project.
Telegram: @Persian_IEEEComSoc

Finally, I am grateful to God for granting me the privilege and ability to make this contribution to the education,
training, career, and reading pleasure of many.

March 2020 Ifiok Otung

South Wales, United Kingdom
 xxv

About the Companion Website

This book is accompanied by a companion website:

www.wiley.com/go/otung

The website includes:

1. Solutions manual featuring solutions to selected end-of-chapter questions.
2. Telecoms software laboratory for a non-mathematical study of Fourier analysis and synthesis.
3. MATLAB code to support simulation-based exercises.
4. Full-colour diagrams/slides.
Scan this QR code to visit the companion web site:
Telegram: @Persian_IEEEComSoc
 1

Overview of Communication Systems

Success is not the absence of failure but the triumph of improved attempts.

In this Chapter
✓ A quick overview of nonelectrical telecommunication techniques highlighting their gross inadequacies for
today’s communication needs. You will find, however, that these ancient techniques are still indispensable
in certain situations.
✓ A brief historical sketch of developments in modern (electrical) telecommunication from telegraphy to the
Internet and 5G. Key developments in binary codes for data transmission, electronic components, trans-
mission media, signal processing, and telecommunication services are summarised.
✓ A discussion of the elements of a communication system. Modern telecommunication systems may vary
widely in complexity and applications, but they are all accurately represented by one block diagram. You
will become conversant with the elements of this generic diagram and the roles and signal processing
tasks that they perform. Information sources, information sinks, transmitters, and receivers are all briefly
introduced.
✓ A detailed overview of the classification of communication systems. Every system is simplex or duplex,
analogue or digital, baseband or modulated, circuit-switched or packet-switched. You will learn the features
of each of these systems and find out just why we have undergone a transformation to digital and IP
dominance.

1.1 Introduction
Telegram: @Persian_IEEEComSoc

This first chapter provides a panoramic view of communication systems. It is intended to give a lucid and compre-
hensive introduction to the subject of communication engineering. We follow a nonmathematical approach and
attempt to show you how each telecommunication concept fits into the overall picture. Armed with this knowl-
edge, it is hoped that you will have enough inspiration and motivation to go on to the remaining chapters, where
the principles and terminology are treated in more detail. If you do not have the time to study the entire book, it
is advised that you work carefully through this chapter and the introductory sections of the remaining chapters.
The material thus covered is suitable for a short course that presents a survey of modern telecommunications.
To drive home the fact that telecommunication generally means communication at a distance, we begin with
a quick overview of various nonelectrical means of telecommunicating. The use of the word telecommunication
is then narrowed to apply exclusively to the electrical techniques. After an important review of selected signif-
icant historical developments from telegraphy to 5G, we present a block diagram that adequately describes all

Communication Engineering Principles, Second Edition. Ifiok Otung.

© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 2 1 Overview of Communication Systems

communication systems. Each component and the processes that it contributes to the overall performance of the
system are then briefly discussed in a nonmathematical way.
Different types and classifications of communication systems are discussed. Digital and analogue communica-
tion systems are compared, and analogue baseband systems are discussed in some detail. The features of digital
baseband transmission are introduced. We show that modulated systems are essential to exploit the radio spec-
trum and the optical fibre medium and discuss the features of this class of systems. Our discussion includes brief
references to some of the main modern applications of telecommunications, namely television systems, com-
munication networks, telephony, satellite communications, optical fibre communication systems, and mobile
communication systems. We also discuss and compare the different switching technologies of space switching,
time switching, circuit switching, connection-oriented (CO) packet switching, and connectionless packet switch-
ing (CL) and explain why CL has become the dominant switching technology of the twenty-first century.
On completing this chapter, you will be well equipped to make informed decisions regarding the suitability of
different types of system blocks (including transmission media) and various classes of communication systems
for different applications. You will also have a clear overall picture of telecommunication and a good foundation
on which you can build a more detailed knowledge of the subject. I hope that working through this chapter will
inspire you towards a career in telecommunication or a very enjoyable further study of this exciting field.

1.2 Nonelectrical Telecommunication

This book is concerned with modern telecommunication or communication over a distance, which relies prin-
cipally on electrical means to convey information from one point S, the sending end, to another point R, the
receiving end. After this section, we will use telecommunication, communication systems, and other similar terms
to refer exclusively to these modern technologies. However, before the advent in 1837 of telegraphy, the forerun-
ner of modern telecommunications, many forms of nonelectrical ‘telecommunications’ existed. We will briefly
introduce these nonelectrical communication systems and discuss their demerits, before turning our attention to
various types and classifications of modern communication systems.

1.2.1 Verbal Nonelectrical Telecommunication

The most basic nonelectrical telecommunication was verbal, in which, for example, a town crier used a combina-
tion of a gong and their voice to broadcast information in an African village. Trumpets were also blown to summon
to war, call a truce, or announce victory, depending on the distinctive sounds made. The transmitter in this case
is the human vocal system and other suitable instrumental sound sources such as the gong, drum, and trumpet.
The signal is a pressure wave called an acoustic signal, the transmission medium is air, and the receiver is every
human ear within range of hearing. This form of ‘telecommunication’ is fraught with problems, which include
Telegram: @Persian_IEEEComSoc

the following.
● Interference: the ear also receives other unwanted sound signals in the environment, which serves as a common
transmission medium. These sounds, which may be artificial or natural background noise, or the intelligible
communication of other people, interfere with and hence corrupt the wanted signal. In the town crier example,
an African woman would have to hush her chattering children, bleating goats, and barking dogs, or move away
from the noisy zone in order to minimise interference with the town crier’s message.
● Nuisance: the signal is received by those who do not want it. If you had to use this verbal means to ‘telecommu-
nicate’ with a friend a few blocks away, your neighbours would not be pleased.
● Huge losses: the signal suffers a lot of attenuation, or reduction in the amplitude of the pressure wave as it is
reflected at material boundaries and as it spreads out with distance over a wider wave front. For this reason, it
is difficult to hear the town crier from within a closed room or from afar.
 1.2 Nonelectrical Telecommunication 3

● Limited range: communication can only take place over small distances (i.e. small separations between S and R).
To overcome the above losses, sound signals of high SPL (sound pressure level) must be used to extend com-
munication range. However, this increases the nuisance effect and endangers the hearing of those close to the
sound source. Besides, the human vocal system and other practical sound sources are severely limited in the
SPL that they can generate.
● Masking effect: the wanted signal can be easily masked by louder signals in the medium. This makes hearing (i.e.
detection) of the wanted signal impossible. The threshold of hearing increases with ambient noise. This means
that the ear–brain system will be completely oblivious to the presence of one sound if there is another louder
sound of about the same frequency.
● Lack of privacy: privacy is an important requirement that this means of communication is incapable of providing.
Everyone (within range) hears your message and is potentially able to understand and make use of it. Therefore,
this means of communication is only suitable for the broadcast of public information and would not be suitable
for the communication of private, military, commercial, or classified information.
● Delay: even if we could overcome all the problems discussed above, propagation delay, the time it takes for the
signal to travel from S to R, would still make this type of telecommunication unacceptable. Sound travels at a
speed v = 330 m/s in standard air. Over very short distances, it seems as though we hear at the same instant as the
sound is produced. However, imagine that S and R are separated by, say, d = 2376 km, a realistic international
and in some cases national distance. The propagation delay is
d 2376000 m
t= = = 7200 s = 2 h
v 330 m∕s
Thus, if you made an acoustic ‘phone call’ over this distance, the person would hear each utterance a staggering
two hours after it was made. And if the person said, ‘Pardon?’ you would complete your call and hang up without
knowing they wanted you to say it again. Real-time interactive communication would only be possible over short
distances. This is in sharp contrast to the case of modern telecommunications using electromagnetic waves, which
travel at the speed of light (v ≡ c = 3 × 108 m/s). The propagation delay for the distance d = 2376 km is t = 8 ms.
Barring other delays, you would hear each other practically instantaneously.

1.2.2 Visual Nonelectrical Telecommunication

Various methods of nonverbal telecommunication were developed as a vast improvement on the basic technique
discussed above. Here, the information to be conveyed is coded in the form of visually observable events. The
transmitter generates these events. Light signals convey the information through a transmission medium that is
ideally a vacuum but is in practice air. The receiver is a human eye within visibility range along an unobstructed
line-of-sight path. The problems of interference, nuisance, propagation delay, and masking by other signals are
negligible compared to the verbal technique. Furthermore, the communication range is extended, especially in
Telegram: @Persian_IEEEComSoc

clear non-foggy weather. However, the receiver must have an unobstructed view of the signalling event and the
major problem of lack of privacy remains. Visual nonelectrical telecommunication systems were identical in trans-
mission medium and receiver, differing only in the type of transmitter. The most common techniques are now
discussed.

1.2.2.1 Flags, Smoke, and Bonfires

The practice of raising a highly visible object, or performing complex coded motions with the object, has had
universal application in warfare and ordinary social interactions over the centuries. Hand signals remain an indis-
pensable means of communication in modern society. For example, hand signals are used by traffic wardens to
control traffic. Hoisting a red flag or red light is still used to warn of danger, and raising a white flag in warfare
communicates a message of surrender over distances that would be difficult to reach by verbal means.
 4 1 Overview of Communication Systems

An interesting record of a military application of this method of visual nonelectrical telecommunication around
1405 BC is given in the Bible. Hebrew soldiers divide into two parties to attack a city. One party waits in ambush
outside the city gates while the other attacks the city and fakes defeat in order to draw the city guards away in
pursuit. At a good distance from the city, the fleeing party raises a javelin from a vantage point, thus signalling to
the ambush party to take the now defenceless city. This party takes the city and quickly starts a bonfire, which
‘telecommunicates’ their success to the retreating party and signals the beginning of a bidirectional attack on the
guards, who are in shock at the sight of their burning city.

1.2.2.2 Heliography
Heliography was a system of communication developed in the nineteenth century in which rays of the sun were
directed using movable mirrors on to distant points. Some links for heliograph communication were set up and the
technique was reliably used, for example in 1880 by General Stewart to give a battle account over some distance.
Reading a heliogram (a message sent by heliography) was said to cause eye fatigue.
A technique for transmitting information using variations in the pattern of light had been developed by the
Greeks as far back as the second century BC. Different combinations and positions of torchlight signals were
employed to represent the letters of the Greek alphabet. In this way messages could be coded and transmitted.

1.2.2.3 Semaphore
A semaphore consists of an upright post with one or more arms moving in a vertical plane. Beginning in the
eighteenth century, different positions of the arms or flags were used to represent letters. Table 1.1 gives the con-
ventional alphanumeric semaphore codes. The semaphore codes for letters A, Q, and V are illustrated in Figure 1.1.
Note that some of the letters of the alphabet are shared with numerals 0–9. A numerical message is distinguished
by being preceded with the code (0∘ , 45∘ ). Code AR is used to indicate the end of signal and code R to acknowledge
reception. In Table 1.1, the arm positions are given in degrees measured clockwise from vertical.

Table 1.1 Semaphore codes.

Positions of Positions of
Symbol semaphore flags Symbol semaphore flags

A 1 225∘ N (135∘ , 225∘ )

B 2 270∘ O (270∘ , 315∘ )
C 3 315∘ P (270∘ , 0∘ )
D 4 0∘ Q (270∘ , 45∘ )
E 5 45∘ R (270∘ , 90∘ )
Telegram: @Persian_IEEEComSoc

F 6 90∘ S (270∘ , 135∘ )

G 7 135∘ T (315∘ , 0∘ )
H 8 (225∘ , 270∘ ) U (315∘ , 45∘ )
I 9 (225∘ , 315∘ ) V (0∘ , 135∘ )

J (0∘ , 90∘ ) W (45∘ , 90∘ )

K 0 (zero) (0∘ , 225∘ ) X (45∘ , 135∘ )

L (45∘ , 225∘ ) Y (90∘ , 315∘ )

M (90∘ , 225∘ ) Z (90∘ , 135∘ )

Angle is measured clockwise from vertical.

 1.3 Modern Telecommunication 5

Figure 1.1 Semaphore codes for letters A, Q, and V.

A Q V

Some governments built large semaphore towers on hilltops along strategic routes. For example, during the
Napoleonic wars (1792–1815), a relay of semaphore towers was built in England from London to Portsmouth. By
the year 1886, semaphore had been almost universally adopted for signalling on railways. Semaphore is still used
today at sea by warships preferring radio silence and for mechanical railway signalling.

1.2.2.4 Demerits of Visual Nonelectrical Telecommunication

The major disadvantages of visual nonelectrical telecommunication include the following.
● Low signalling speed: the rates of signal generation at the transmitter and detection at the receiver are extremely
low by modern standards. For example, if the torchlight or semaphore flag patterns are manipulated quickly
enough to complete the representation of one letter every second, we have a maximum data rate of one character
(or byte) per second. In modern systems, a modem that operates at 10 000 times this speed is considered slow.
Note, however, that propagation delay is negligible, and the signal is received almost at the same instant as it is
generated at the transmitter.
● Manual operation: a trained signaller is required at the transmitter to code the information and at the distant
receiver to view and decode the signal patterns as they are transmitted. Automation of both transmission and
reception is impractical.
● Limited range: the maximum separation that allows visibility between transmission and reception points may
range from less than 50 m in thick fog to a few miles in obstruction-free terrain under very clear weather. How-
ever, this range is very low compared to distances involved in modern communications.
● Lack of privacy: even with the use of encrypted codes, the communication process is carried out in full view of
all within the communication range.
● Limited application: a combination of the above factors means that visual nonelectrical telecommunication
remains a system that can only accommodate a very few users (low capacity) for very limited applications. This
is grossly inadequate to meet modern requirements of broadband communication.
Telegram: @Persian_IEEEComSoc

1.3 Modern Telecommunication

Modern telecommunication began in 1837 with the invention of telegraphy. Sir Charles Wheatstone (1802–1875)
and Sir William Fothergill Cooke (1806–1879) built the world’s first commercial telegraph system in England
in 1837. They laid the telegraph lines, consisting of six iron wires, along a 2.4 km railway track from Euston to
Camden Town stations. Their invention was first publicly demonstrated on 24th July 1837 and described as a ‘most
extraordinary apparatus’ in an advertisement. The receiver at the receiving station consisted of five galvanometer
needles arranged in a row across the face of a grid of letters, as shown in Figure 1.2. Each needle (G1, G2, …, G5)
could be deflected to the left or right by current flowing in the corresponding wire (W1, W2, …, W5), not shown
in the diagram. A sixth wire, W6, was needed to provide a common return path for each of the five circuits. To
 6 1 Overview of Communication Systems

B D

E F G

H I K L

G1 M G2 N G3 O G4 P G5

R S T

V W

Figure 1.2 Wheatstone-Cooke 5-needle electric telegraph display grid.

transmit a letter, current flow was sent from the transmitter down two wires causing two of the needles to deflect
in opposite directions and point to the right letter. For example, to transmit the letter A, current flow is established
in wires W1 and W5 so that G1 deflects clockwise and G5 counterclockwise from vertical. But to transmit the letter
Y, current flow is established in the opposite direction in the two wires so that G1 deflects counterclockwise and
G5 clockwise from vertical.
This five-needle telegraph system was simple enough to allow unskilled operation, but it required six wires to
make the connection between transmitting and receiving stations and only 20 letters could be transmitted, which
made it necessary to transmit some words with a modified spelling. It was soon replaced in 1838 by a cheaper
two-needle telegraph, which was also eventually superseded in 1845 by a single-needle system. The single-needle
telegraph coded each letter using a unique combination of movements of one needle to the left and right. By
arranging for the needle to strike small metal pipes that emitted sounds of different pitches, each letter was heard
as a unique melody at the receiver.
At about the same time that Wheatstone and Cooke developed their needle-deflecting telegraph, two Americans,
Telegram: @Persian_IEEEComSoc

Samuel Finley Breese Morse (1791–1872) and Alfred Vail (1807–1859), devised a more efficient telegraph system,
which was eventually adopted all over the world. Morse and Vail demonstrated their system on 24th May 1844 by
sending the message ‘WHAT HATH GOD WROUGHT’ over the first telegraph link in America from the Capitol
at Washington to Mount Clare Depot in Baltimore, a distance of 64.5 km. The Morse–Vail telegraph consisted of
a key or switch that was pressed to complete a circuit containing a battery connected by a single pair of wires to
a distant sounder. The key and battery constituted the transmitter, the pair of wires the transmission medium,
and the sounder the receiver. Transmission of textual information was achieved using 39 characters (26 uppercase
letters of the alphabet, 10 numerals, and 3 punctuation marks: comma, full stop, and question mark).
In what follows, we introduce in chronological order the various coding schemes which were developed to
represent characters and then we discuss some of the most significant developments in telecommunication since
the advent of telegraphy.
 1.3 Modern Telecommunication 7

1.3.1 Developments in Character Codes

1.3.1.1 Morse Code
Each character in the Morse–Vail telegraph was represented by a pattern of dots (•) and dashes (—) called Morse
code, invented in 1838. The innovative and enduring principle behind the Morse code, which has sustained its
application into the twenty-first century (e.g. in amateur radio), is that characters that occur frequently are rep-
resented using shorter codes, whereas less frequent characters are assigned longer codes. For example, the Morse
code for letter E (the most frequent in the English alphabet) is a single dot •; the code for T (the next most frequent
letter) is a single dash —; but the code for letter Q (one of the least frequent letters) is — — • —. Morse code has
been extended over the years to include other characters and control signals. An international version is shown in
Table A.1 of Appendix A.
In actual transmission using Morse code, a dot is a voltage pulse of duration one unit (say T) generated by
pressing a key to close the transmitter circuit for this duration, a dash is a voltage pulse of duration 3T, and dots
and dashes within a Morse code for a character are separated by a silence interval of duration T. Characters are
separated by a silence of duration 3T, and words by 7T. At the receiver the incoming message is identified by ear,
using the sounds produced. For example, the letter E sounds like ‘dit’, the letter T sounds like ‘dah’, and the letter
Q sounds like ‘dahdahdidah’. The value of T used depends on the transmission speed (words per minute) at which
the receiving operator can copy accurately. The spacing between characters and the spacing between words can
be increased above 3T and 7T, respectively, to accommodate a slow operator.
In the original Morse–Vail telegraph system the incoming message was recorded on a moving strip of paper by
using an electromagnet to move a marker which sketched the code waveform comprising a sequence of short (dot)
and long (dash) pulses. As an example, the statement (excluding quotes) ‘THANK GOD’ would be conveyed by
the voltage pulse sequence shown in Figure 1.3. The system, however, developed into receiving by ear from the
clicks of the receiver. This was more convenient and faster for a trained operator who could handle 40–50 words
per minute. By 1914, automatic operation involving tape-reader machines and multiplexing (transmitting more
than one message at a time over a single wire) allowed telegraph transmission speeds of 400 words per minute to
be achieved.
The development of automated teleprinter communication technology in the 1920s marked the beginning of
the end of manual landline (i.e. wire-connected) Morse–Vail telegraphy. Its commercial use was discontinued in
1932 by the British Post Office, and in the 1960s on railroads in America and Australia. Landline application has,
however, continued on railroads in some developing countries and amongst a few die-hard enthusiasts in the US.
Furthermore, Morse code continues to be used in radio telegraphy by radio amateurs worldwide and by a few ships
at sea as a low-cost alternative to satellite communications.

1.3.1.2 Baudot Code

Efforts to improve the telegraph system led to the invention in 1871 of the first device for time division multiplexing
Telegram: @Persian_IEEEComSoc

(TDM) by the French engineer Jean-Maurice-Émile Baudot (1845–1903). The multiplexer consisted of a copper

t
T H A N K G O D .

Figure 1.3 Morse code’s sequence of voltage pulses for THANK GOD.
 8 1 Overview of Communication Systems

ring divided into several equal sectors (say N). Each sector had five contacts which could be opened or closed,
giving 32 (= 25 ) possible combinations for coding characters. A brush arm rotated in a circle around the copper
ring and sequentially picked up the code combinations from each sector. Connecting this arm to a transmission
line allowed N messages to be simultaneously sent.
The existing variable-width Morse code was not suitable for the automatic transmission and reception operation
of Baudot’s multiplexer, so he devised a new fixed-length code which was patented in 1874. The Baudot code was
modified in 1901 by Donald Murray (1865–1945) – a New Zealand born farmer, journalist, telegraph engineer,
and entrepreneur. After further modification in 1932 by the International Telegraph and Telephone Consultative
Committee (CCITT) – now International Telecommunications Union, Telecommunications Sector (ITU-T), the
Baudot–Murray code became known as the International Telegraph Alphabet No. 2 (ITA-2).
Table A.2 of Appendix A lists the ITA-2 code. Each character is represented using five symbols drawn from a
binary alphabet, namely a mark (voltage pulse) and a space (no voltage pulse), corresponding to binary digits or bits
1 and 0, respectively. With fixed-width codes like this, there is no need for the short pause (between characters) and
long pause (between words), as in the Morse code. Rather a codeword ‘space–space–mark–space–space’ or 00100
is used to separate words. Characters need no separation between them, except that before actual transmission
each character is framed by preceding the 5-bit character code with a start bit (usually a space) and terminating
with a stop bit (usually a mark), giving seven transmitted bits per character.
Using 5-bit codes, only 25 = 32 different characters can be represented. This is insufficient to cover 52 uppercase
and lowercase letters of the alphabet, 10 numbers, and various punctuation and control characters. Raising the
number of bits above five was not an option, owing to hardware constraints in the electromechanical technology
employed at the time. To work around this 5-bit constraint, lowercase letters were not represented and two ‘shift
codes’ were used. More specifically, a letter shift code 11111 is transmitted to indicate that all subsequent characters
are uppercase letters until a figure shift code 11011 is encountered, which causes all following characters to be
interpreted as numbers or punctuation marks. Baudot can therefore be credited with inventing the use of an escape
code – still essential in modern computer systems – which is a code that changes the way the system interprets
subsequent codes. Thus, 26 of the 32 codes were used to represent uppercase letters; 4 were used to designate
blank, space, carriage return and line feed; and the remaining 2 were used for the letter and figure shift codes.
Issuing a figure shift code allowed the 26 codes to be reused for representing numbers and punctuation marks, as
shown in Table A.2.
The ITA-2 code was adopted in teleprinters and the telex (telegraph exchange) network operating at a transmis-
sion speed of 50 symbols per second. By the middle of the twentieth century it had replaced Morse as the most
widely used telegraph code and was still used in telex networks operating at speeds ≤300 symbols/second until the
turn of the century. In other data communication systems Baudot code was replaced by American Standard Code
for Information Interchange (ASCII) and extended binary coded decimal interchange code (EBCDIC) codes, which
could represent a larger number of characters. The memory of Baudot, however, lives on today, transmission speed
in symbols/second being called baud (Bd) in his honour.
Telegram: @Persian_IEEEComSoc

1.3.1.3 Hollerith Code

In 1881, Herman Hollerith (1860–1929), a young American statistician working for the US Census Bureau, devised
a means of recording census data on punched card for automatic machine reading. The punched card was later
standardised to 186 mm × 82 mm in size, with a clipped top left-hand corner and 12 rows and 80 columns. A
column was used to represent one character by punching holes in a selection of the 12 row positions available.
One card could hold 80 characters. Recorded data were read electronically by moving the punched card between
brass pins in a ‘tabulating machine’. At positions with holes, the pins on opposite sides of the card made contact,
thereby completing an electrical circuit and registering a value.
The pattern of holes for each character was defined by the Hollerith code. With 12 binary digits – i.e. 12
hole-or-no-hole positions – a total of 212 = 4096 characters can be represented. However, only 69 characters (52
 1.3 Modern Telecommunication 9

upper and lowercase letters, 10 numerals, and 7 punctuation marks and symbols) were defined. The use of 12 rows
(rather than, say, 7) allowed each character to be uniquely represented using only a few holes. Bear in mind that the
holes were initially manually punched. Besides, many holes in one row would have made the card more liable to
tear. Thus, all 10 numerals and 2 letters were represented with just one hole, the remaining letters were represented
using two holes, and two or more holes were used to code the less frequent punctuation marks and symbols.
Hollerith’s punch card technology was first employed in 1887 for calculating mortality statistics, but it gained
popularity when it was used for the 1890 US census. Thanks to the new technology, it took just six weeks to analyse
the 1890 census data more thoroughly than the previous 1880 census, which took seven years of toil by hand. To
market his highly successful invention Hollerith founded the Tabulating Machine Company in 1896, which in
1911 merged with two other machine manufacturers to form the Computer-Tabulating-Recording Company. In
1924, the company was renamed International Business Machines Corporation (IBM) and it dominated the world
of computing until the 1970s.
The Hollerith code continued to be used for data representation until the 1960s, when IBM developed a new
character code, EBCDIC, for its mainframe computers. But punched cards remained the primary means of data
input to computers until the early 1970s, and the technology continued to be used in some establishments up to
the early 1980s. Many today regard Herman Hollerith as the father of information processing.

1.3.1.4 EBCDIC Code

The extended binary coded decimal interchange code (EBCDIC – pronounced ebb-sea-dick) is a proprietary 8-bit
character code developed in the 1960s by IBM for use on its computing machines. It provides 28 = 256 codewords
for the representation of 256 different characters including a wide range of control characters. Table A.3 gives a
list of EBCDIC codewords. Note the following features.
● The first 64 codewords (hexadecimal 00 → 3F) specify control characters.
● The codewords for uppercase and corresponding lowercase letters differ in only one bit position (bit 7). For
example, the codewords for letters ‘m’ and ‘M’ are 10010100 and 11010100, respectively.
● The 62 alpha-numeric characters (52 uppercase and lowercase letters and 10 numbers) occupy 7 noncontiguous
blocks in the code table. Their codewords are restricted to those whose lowest four bits (i.e. first hexadecimal
digit) have decimal values in the range 0–9. It is for this reason that the term binary-coded-decimal (BCD) features
in the name of the coding scheme. A drawback of this noncontiguous arrangement of letters is that it makes
computer programs for manipulating textual information more complex.
The EBCDIC character code has always been perceived as an IBM ploy for locking in its customers. Other
computer manufacturers and the rest of the world adopted the ASCII code in the 1960s. The rapid growth in
personal computers (PCs) starting in the late 1970s brought IBM’s domination of the computer industry to an end.
EBCDIC’s use has therefore since become confined to a vanishing minority of historical computer systems. It is
worth emphasising that EBCDIC is inherently incompatible with ASCII. A PC requires code conversion to be able
Telegram: @Persian_IEEEComSoc

to communicate with an IBM mainframe computer that uses EBCDIC, and a PC software program that manipu-
lates letters of the alphabet by the relative decimal values of their ASCII codewords would not run properly on an
EBCDIC system.

1.3.1.5 ASCII Code

The ASCII code was first developed in 1963, and updated in 1968 as ANSI X3.4, by the American National Stan-
dards Institute (ANSI) – formerly the American Standards Association. ASCII uses seven bits, which allows rep-
resentation of 27 = 128 characters covering 26 lowercase letters, 26 uppercase letters, 10 numbers, punctuation
marks and mathematical symbols, and a wide range of control characters.
Table A.4 of Appendix A gives a complete listing of 7-bit ASCII codes. To use this table, note that the number
in the top left-hand corner of each character entry is the decimal value of the code. The first and second columns
 10 1 Overview of Communication Systems

of the table give, respectively, the first hexadecimal (hex) digit and the least significant four bits b4 b3 b2 b1 of the
ASCII code for the character in the corresponding row. Similarly, the first and second rows give the second hex
digit and the most significant three bits b7 b6 b5 of the code. Thus, the ASCII code for the character M is decimal
77, hex 4D, or binary 1001101. We may write these, respectively, as 7710 , 4D16 or 10011012 .
The following features of the ASCII coding scheme may be observed in Table A.4.
● The three most significant bits indicate the type of character. For example, the C0 control characters are in the
first two columns b7 b6 b5 = 000 and 001; uppercase letters of the alphabet are in the fifth and sixth columns
b7 b6 b5 = 100 and 101; and lowercase letters are in the seventh and eighth columns b7 b6 b5 = 110 and 111.
● The allocation of codewords to numbers (0–9) and letters of the alphabet (A–Z and a–z) follows a binary progres-
sion. For example, the codewords for the numbers 0, 1, and 2 are respectively 0110000, 0110001, and 0110010, the
last four bits being the binary equivalent of each decimal number. The codewords for the letters R, S, and T are
1010010, 1010011, and 1010100, respectively. Mathematical operations can therefore be performed on the code-
words that represent numbers and alphabetisation can be achieved through binary mathematical operations on
the codewords for letters. For this reason, ASCII codes are described as computable codes.
● For ease of generation on a keyboard, lowercase and uppercase letters differ only at the sixth bit position (b6 ).
For example, the ASCII codes for letter ‘A’ and ‘a’ are 1000001 and 1100001, respectively. This bit position (b6 )
is changed on the keyboard by holding down the shift key.
ASCII was an instant success and became a de facto international standard. However, as you can see from Table
A.4, the code was inadequate even to cover international characters such as é, ü, ô, etc. based on the Latin alphabet.
It was therefore necessary to internationalise ASCII. This task was undertaken by the International Organization
for Standardization (ISO) leading to an international standard ISO 646 in 1972 and a subsequent revision in 1991.
The ASCII code table shown in Table A.4 was designated the international reference version (IRV) and identified as
ISO-646-IRV, which is synonymous with the US ASCII version ISO-646-US. Provision was then made for national
versions to be created by substituting other graphic characters in place of the least needed 10 characters, namely @
[ \ ] ^ ‘ { | } ∼. For example, the German version (ISO-646-DE) is identical to Table A.4 except that these 10 characters
are replaced by the characters § Ä Ö Ü ^ ‘ ä ö ü ß, respectively. Similarly, the French version (ISO-646-FR) has the
characters à ∘ ç § ^ ‘ é ù è ̈ , respectively; the Italian version (ISO-646-IT) has the characters § ∘ ç é ^ù à ò è ì,
respectively; and so on. Furthermore, allowance was made for versions to be created in which the IRV was altered
in two other character entries by substituting the pound sign £ for the number sign # and/or substituting the
currency sign ⦻ ⚫ for the dollar sign $.
The process of internationalisation of the 7-bit ASCII code was extended to include languages that are not based
on the Latin alphabet, such as Arabic, Greek, and Hebrew. As a result, up to 180 ASCII-based character code ver-
sions were eventually registered. The main drawback of a multiplicity of versions is that the character represented
by a given code number (say decimal 91) is not unique but depends on the ASCII version of the source system.
This poses a nuisance in information interchange between computers using dissimilar versions. National variants
Telegram: @Persian_IEEEComSoc

of ISO-646 are now obsolete having been replaced by less problematic schemes.

1.3.1.6 ISO 8859 Code

A more interoperable character code scheme was developed by the European Computer Manufacturers Asso-
ciation (ECMA) in the mid-1980s by extending the codeword from seven to eight bits. This allowed up to 256
characters to be represented. The new scheme was endorsed by ISO as ISO-8859. To maintain compatibility with
the well-established ISO-646-IRV the 8-bit code table was designed as shown in Figure 1.4, with the left half (most
significant bit, or MSB, = 0) identical to the ISO-646-IRV table and the right half (MSB = 1) providing an extra
32 control characters and 96 graphic characters. A US-ASCII character with code b7 b6 b5 b4 b3 b2 b1 was therefore
assigned an 8-bit code 0b7 b6 b5 b4 b3 b2 b1 under the new scheme.
 1.3 Modern Telecommunication 11

0 1 2 3 4 5 6 7 8 9 A B C D E F
0 SP
1
2
ISO-646- IRV Control (C0) Characters

ISO-646-IRV Graphic Characters

Regional Graphic Characters

4

Control (C1) Characters

5
6
7
8
9
A
B
C
D
E
F DEL

ISO-646-IRV (Table A.4)

Figure 1.4 Layout of ISO 8859 (8-bit) character code table. The column label is the second hex digit (or most significant 4
bits) and the row label is the first hex digit.

Since there are more than 96 foreign graphic characters, it was necessary to develop multiple character sets, one
set for a specified geographical region. This gave rise to more than 10 sets of 8-bit codes, which differed only in
the area labelled Regional Graphic Characters in Figure 1.4. Escape sequences were defined for switching between
character sets – analogous to changing typeheads on a typewriter. Thus, there was ISO-8859-1 or Latin Alphabet
No. 1 for the needs of Western Europe. This is listed in Table A.5 of Appendix A, where the number in the top
left-hand corner of each cell is the hex code of the character. So, for example, the 8-bit code for the ligature æ is
hex E6 or binary 11100110 and the pound sign £ has code 10100011 (equivalent to decimal 163). Other charac-
ter sets in the scheme included ISO-8859-2, which covered Eastern European languages (Albanian, Hungarian,
Romanian, and Slavic), ISO-8859-3 to -8, which respectively covered Southern Europe, Northern Europe, Cyrillic
(i.e. Bulgarian, Macedonian, Russian, Serbian, Ukrainian), Arabic, Greek, and Hebrew, and so on.
The 8-bit ISO-8859 scheme was an improvement on the 7-bit ASCII. It reduced the number of character sets
involved in global information exchange to a smaller number of regional variants. However, it still involved switch-
Telegram: @Persian_IEEEComSoc

ing between character sets and did not cover all the world’s languages, most notably the East Asian languages.

1.3.1.7 Unicode
In 1987, three engineers from Apple Computer and Xerox – Joe Becker, Lee Collins, and Mark Davis – started a
project to develop the ultimate character code: one that would be universal, uniform, and unique. Becker coined
the term Unicode to describe this scheme, which aimed to cover all of the world’s languages, assign a unique char-
acter to each code number or bit sequence, employ fixed-width codes (i.e. same number of bits in every codeword),
and wherever possible maintain some compatibility with existing standards (e.g. ISO-646-IRV and ISO-8859-1).
This was a mammoth task the success of which required the support of influential global computer companies
such as Microsoft Corporation and Sun Microsystems.
 12 1 Overview of Communication Systems

The group did gain support and grew into an incorporated consortium in 1991. And in 1992 Unicode merged
with and subsumed IS0/IEC 10646, a multilingual character code independently developed by ISO and at that time
incompatible with Unicode. Version 13.0 of Unicode was released on 10th March 2020 as ISO/IEC 10646:2020, con-
taining a staggering 143 859 characters covering almost all modern and historical writing systems in the world.
Unicode has ample room for all graphic characters, format characters, control characters, and user-defined charac-
ters as well as future growth. The graphic characters include a wide range of symbols (technical, optical character
recognition, Braille pattern, dingbats, geometric shapes, emoji, etc.) as well as ideographic characters used in
China, Japan, Korea (CJK), Taiwan, Vietnam, and Singapore. Duplicate encoding is avoided by assigning a single
code to equivalent characters irrespective of usage or language of occurrence. There are ongoing efforts to identify
and define additional characters for inclusion in future versions of the standard, subject to a policy of not altering
the codes already assigned to characters in previous versions, and not encoding logos, graphics, and font variants,
nor ‘characters’ deemed idiosyncratic, novel, private-use, or rarely exchanged.
To fully cater for all the world’s written languages, Unicode makes provision for 1 114 112 characters using inte-
gers or code points in the range 0 → 10FFFF16 . A code point is referred to by its numeric hex value prefixed by U+.
For compatibility with US-ASCII and ISO-8859-1, the first 256 code points (i.e. 0000 → 00FF) are assigned as in
Tables A.4 and A.5 of Appendix A. Each encoded character has a unique official Unicode name. For example, the
character G is assigned code point U + 0047 and named LATIN CAPITAL LETTER G, and the character Ç, assigned
code point U + 00C7, is named LATIN CAPITAL LETTER C WITH CEDILLA.
It is convenient to think of the 1 114 112 Unicode code points as divided into a total of 17 planes, each containing
216 code points. Almost all common-use characters of all the world’s languages are contained in the 65 536 code
points (0000 → FFFF) available in the basic multilingual plane (BMP) or Plane 0. These Unicode characters can
be represented in computer systems using fixed-width 16-bit codewords, unlike the seven bits of ASCII and the
eight bits of ISO-8859. However, to cover all the characters in all 17 planes using unique (non-overlapping) code-
words, the Unicode Standard specifies the following three distinct encoding forms, called Unicode Transformation
Formats (UTF):
● UTF-32: each character is expressed using one 32-bit code unit which has the same value as the code point for
the character. This is a fixed-width 32-bit character format, which quadruples the size of text files compared to
the 8-bit ISO-8859. But, importantly, all characters of the world are uniquely covered using a single code unit
access. UTF-32 is the preferred encoding form on some Unix platforms.
● UTF-16: most common-use characters of all the modern scripts of the world lie in the BMP (U + 0000 → U +
FFFF). These characters can be represented by a single 16-bit code unit. This was the original design of the
Unicode Standard, to be a fixed-width 16-bit character code. However, as the Standard evolved it became clear
that not all characters could be fitted into code points ≤ U + FFFF. Thus, the code points’ range was extended
to include the supplementary planes (1 → 16). An area of the BMP that was unallocated up until the extension
was then set aside for surrogates, and a pair of 16-bit code units (called surrogate pair) in this area was used to
represent (i.e. point to) a code point in a supplementary plane. Note that neither of the code units of a surrogate
Telegram: @Persian_IEEEComSoc

pair is used to represent a character in the BMP. Thus UTF-16 is a variable-width encoding form that uniquely
codes characters by using one 16-bit code unit for each of the more frequent BMP characters, and two 16-bit
code units for each of the much less frequent characters of the supplementary planes. A code unit conversion
is required to convert a surrogate pair to the Unicode code point (> U + FFFF) and hence the supplementary
character represented by the pair. UTF-16 typically requires half the memory size of UTF-32 and is the preferred
encoding form in most implementations of Unicode other than on Unix platforms.
● UTF-8: this is the preferred encoding form for the Internet. It was prescribed to provide transparency of Uni-
code implementation in ASCII-based systems. Each Unicode code point in the range U + 0000 → U + 007F is
represented as a single byte in the range 00 → 7F having the same value as the corresponding ASCII character.
Then, using 8-bit code units or bytes of value >7F, the remaining code points are represented by two bytes (for
U + 0080 → U + 07FF), three bytes (for U + 0800 → U + FFFF), and four bytes (for U + 10 000 → U + 10FFFF).
 1.3 Modern Telecommunication 13

Thus UTF-8 is also a variable-width encoding form that uses codewords ranging in width from one byte for the
first 128 (ASCII) characters to four bytes for the supplementary Unicode characters. The UTF-8 form works
automatically as follows: if a byte – let’s call it Byte 1 – starts with bit 0 then it represents a character; or else if
Byte 1 starts with bits 110 then the character is represented by the concatenation of Byte 1 and the next byte (i.e.
Byte 2); or else if Byte 1 starts with bits 1110 then the character is represented by the concatenation of Byte 1 and
the next two bytes (i.e. Byte 2 and Byte 3); or else if Byte 1 starts with bits 11110 then the character is represented
by the concatenation of Byte 1 and the next three bytes (i.e. Byte 2, Byte 3, and Byte 4). Thus, a 1-byte codeword
has prefix 0; a 2-byte codeword has prefix 110; a 3-byte codeword has prefix 1110; and a 4-byte codeword has
prefix 11110. Furthermore, all Bytes 2, 3, and 4 have prefix 10, and this distinctively identifies them as not Byte
1, which prevents the error that would otherwise have been possible if one, for example, started reading a multi-
byte codeword from other than its first byte. Therefore, picking any byte at random inside a UTF-8 file, one can
correctly read the character as follows: if the byte’s prefix is 0 then that’s the entire 1-byte character; or else the
byte is part of a multibyte codeword formed by concatenating byte(s) to its left and/or right until the longest
possible multibyte (≤4 bytes) codeword is formed that has a first byte starting with 11 and a last byte starting
with 10.

Unicode has been described as the ultimate character code for the twenty-first century. It has gained worldwide
acceptance, including in East Asia, where it has the capacity to fully cater for all the ideographic characters in JIS
(Japan Industrial Standards) and other East Asian standards (e.g. Chinese GB 2312-1980, Korean KS C 5601-1992,
and Taiwanese Big-5 and CNS 11643-1992). Unicode may not be as efficient in handling a language as its regional
character code set (e.g. ISO-8859-8 for Hebrew), but Unicode’s multilingual capability is a compelling advantage
in an era of globalisation.

1.3.2 Developments in Services

Data transmission by telegraphy was soon followed by voice transmission with the invention of the telephone by
Alexander Graham Bell in 1876. Initially, voice communication was only possible over short distances between
two people using permanently linked telephone sets. The novelty of switching made it possible for two subscriber
terminals connected to the same switching office or local exchange to be manually linked by an operator. In 1897,
an undertaker, A. B. Strowger, developed a step-by-step electromechanical switch that made automatic switching
possible. The technology was thus in place for the communication of voice and data over wire line connections,
although the speed and quality of transmission left much to be desired by today’s standards.
A wide range of digital communication services, too many to discuss here, have been developed since the
1830s and especially during the last five decades. Here we briefly introduce a few of the most significant services
in roughly chronological order. Some services, such as the telex and facsimile, were largely limited to business
applications, whereas others, such as the telegram and email, made significant penetration into both home and
Telegram: @Persian_IEEEComSoc

business.

1.3.2.1 Telegram
Digital communication services began with the telegram in the late 1830s when members of the public sent
messages on the Wheatstone–Cooke telegraph from Paddington railway station to Slough at one shilling per mes-
sage – a large sum in those days. With the widespread use of the Morse–Vail telegraph all over the world, the
telegram became a well-established communication service starting from 1845.
The military used the telegram to communicate intelligence, deploy troops, and disseminate battle news; railway
companies used it to regulate trains, and this significantly extended railway capacity; industry used it for a wide
range of business communication (e.g. transmitting racing results, stock prices, urgent memos, etc.); and the police
quickly adopted it in their fight against crime. A much-publicised early application was the arrest of John Tawell
 14 1 Overview of Communication Systems

for the murder of his mistress on 1st January 1845. Tawell fled the crime scene and got on a train departing Slough
to London where he hoped to make good his escape. However, his failure to reckon with the power of the new
telecommunication invention proved his undoing. A telegram giving a description of the murder suspect beat him
to the London station and the police were waiting when his train pulled in. Tawell was arrested, later tried, found
guilty of murder, and hanged.
The telegram was also a great success with the general public. Messages – charged according to length – were
transmitted via telegraph, received and printed on paper at the destination telegraph station, and delivered by
hand to the recipient’s address. By 1874, the United Kingdom had over 650 000 miles of telegraph wire, and more
than 20 000 towns and villages were part of the UK network. However, by the second half of the twentieth century,
the telegram service was forced into steady decline due to developments in other communication services – both
analogue (e.g. telephony) and digital. The UK inland telegram service was eventually discontinued in 1982.

1.3.2.2 Telex
The telex, started in 1932 by the British Post Office, was another major digital communication service to be intro-
duced. A teleprinter (or teletypewriter) in the premises of one subscriber was connected to another subscriber’s
teleprinter via the public switched telephone network (PSTN) that had evolved since the advent of analogue tele-
phony in 1876. A text message, typed in at the keyboard of the sending teleprinter, was represented in 5-bit Baudot
(ITA-2) code, and transmitted down the line using on (mark) – off (space) – keying (OOK) of a 1500 Hz carrier volt-
age at a speed of 50 Bd. The message was received and automatically printed on paper at the destination teleprinter.
Telex quickly grew into a global service particularly popular with government departments and a wide range of
institutions and business organisations. Documents sent by telex had legal status. However, although the technol-
ogy behind telex was steadily improved over the years – from manual to automatic switching, from 50 Bd to 300 Bd
transmission speed, and from mechanical teleprinters to PCs – demand for the service began to decline steadily in
favour of facsimile in the 1980s and email in the 1990s.

1.3.2.3 Facsimile
Facsimile service, or fax for short, allows paper documents – handwritten, printed, drawings, photographs, etc. – to
be electronically duplicated over a distance. In 1980, the CCITT specified the first digital transmission standard
for fax, called Group 3 (G3) standard. The connection between the sending and receiving fax machines is in most
cases via the PSTN, and sometimes by radio. The G3 fax standard became the most widely used and could transmit
an A4 page in less than one minute, typically 15–30 seconds. A later G4 standard required a digital-grade telephone
line and could transmit an A4 page in less than five seconds with a resolution of 400 lines per inch (lpi).
The G3 standard has a resolution of 200 lpi, which means that the paper is divided into a rectangular grid of
picture elements (pixels), with 200 pixels per inch along the vertical and horizontal directions, amounting to 62
pixels per square millimetre. Starting from the top left-hand corner, the (A4) paper is scanned from left to right,
one grid line at a time until the bottom right-hand corner of the paper is reached. For black-and-white-only repro-
Telegram: @Persian_IEEEComSoc

duction, each pixel is coded as either white or black using one bit (0 or 1). For better-quality reproduction in G3
Fax, up to five bits per pixel may be used to represent up to 25 = 32 shades of grey. The resulting bit stream is com-
pressed at the transmitter and decompressed at the receiver in order to increase the effective transmission bit rate.
At the receiver, each pixel on a blank (A4) paper is printed black, white, or a shade of grey according to the value
of the bit(s) for the pixel location. In this way the transmitted picture pattern is reproduced. Early fax machines
used a drum scanner to scan the paper directly, but later machines formed an image of the paper onto a matrix of
charge-coupled devices (CCDs), which build up a charge proportional to incident light intensity.
Fax had its origin in chemical telegraphy invented in 1842 by Alexander Bain (1810–1877) and involved damp
electrolytic paper. Fax did not, however, come into any serious use until 1902 when a German, Dr Arthur Korn
(1870–1945), developed a suitably sensitive photoelectric cell, which allowed a mechanism to be devised for
converting a photographic negative of the picture into an analogue electrical signal. In 1924, AT&T (American
 1.3 Modern Telecommunication 15

Telephone and Telegraph Corporation) demonstrated telephotography, as fax was then known, by transmitting
pictures over a telephone line from Cleveland to New York.
By the end of the 1920s, pictures of distant events for publication in newspapers were being routinely sent by fax.
In 1935, the service began to come into more widespread use amongst businesses when the fax machine became
more affordable. This came about through eliminating the need for a photographic negative and producing the
electrical signal using light reflected directly off the picture being scanned.
At the turn of the century, the popularity of fax for business communication began to be challenged by email,
which is particularly convenient for sending documents that are in electronic (word-processed) form as it can
be done without the sender leaving their desk or feeding any paper into a machine. And by first scanning the
documents into electronic form and sending them as file attachments, email may even be used for transmitting
hard copies and documents that must be supported by a signature and/or headed paper.

1.3.2.4 The Digital Era

Digital communication services have grown at an astounding rate during the last five decades. Computer net-
working started in 1971 with the Advanced Research Projects Agency Network (ARPANET). By the end of the
1990s, this simple project had evolved into an Internet revolution having enormous global impact on every aspect
of business and social life, and spurning completely new jobs, business methods, work patterns, social pastimes,
vocabulary, crimes, etc. Social media, search engines, e-commerce, video/music streaming, e-learning, and online
transactions are just a few of the countless services made possible by the Internet, which is now a global network
of computers, devices, and (in short) things.
A cellular telephone concept, which was first demonstrated by Motorola in 1972, quickly caught on and
led to a global mobile communication revolution, driven by steady improvements in successive generations of
nonbackward-compatible terrestrial wireless communication technologies from 1G in 1979 to 5G in 2019. The
major areas of improvements included changing (i) from analogue technology in 1G (launched in 1979) to digital
technology in 2G (launched in 1991); (ii) from circuit switching to packet-switched high-speed data network of
data rate ≥ 144 kb/s in 3G (launched in 2001); (iii) to an all-IP (Internet Protocol) mobile broadband network of
peak data rate ≥ 100 Mb/s in 4G (launched in 2011); (iv) to a high-mobility (≤ 500 km/h), low-latency (≥1 ms),
ultra-broadband network of peak data rate ≥10 Gb/s in 5G (launched in 2019).
At the same time as mobile communication technology was evolving, there was enormous progress in the
development of the transceiver device or user terminal (enabling access to the network), culminating by the
year 2000 in the smartphone, which is unarguably the most versatile and consequential pocket-sized device or
gadget in humankind’s entire history. Think of it! The smartphone is a telephone handset, data terminal, video
terminal, video camera, still camera, calculator, audio recorder and player, clock and alarm, route finder (i.e. nav-
igational aid), personal digital assistant (e.g. notebook, diary, calendar, reminder), gateway to the Internet, and
much more; yet it is not just portable (i.e. untethered to one location) but pocket-sized. The smartphone has
become a must-have item for work, leisure, and travel, and its global penetration has even surpassed the all-time
Telegram: @Persian_IEEEComSoc

great success stories of the car and washing machine. For example, on 29th February 2020, the website http://live-
counter.com [1] estimated that there were 1.318 billion cars in the world. This figure should be compared to an
estimate of 3.5 billion smartphone users worldwide in 2020 by http://statista.com [2] and a report in http://gartner
.com [3] of total global smartphone sales of 1.524838 billion units in the year 2019 alone.
In 1973, the US department of defence began the development of a global positioning system (GPS) comprising a
space segment of 24 satellites, a control segment of several earth stations, and a user segment. The system eventu-
ally became operational in 1995. GPS was designed primarily for the US military to provide estimates of position,
velocity, and time of a GPS receiver and hence of the person or unit bearing the receiver. However, civil applica-
tions of the service have since grown to the extent that satellite navigation (enabled primarily by the US GPS) is
now indispensable to aeronautical and maritime transportation, mining, surveying, and countless other activities;
and is also an important tool for route determination (called satnav) by drivers and the general population around
 16 1 Overview of Communication Systems

the world. In addition to the US GPS, other less dominant global navigation satellite systems (GNSS) have also been
launched, including Europe’s Galileo (which went live in 2016), Russia’s GLONASS (the development of which
started in 1976 but full global coverage was only attained in 2011), and China’s BeiDou (which started limited
regional coverage in 2000 and attained full global coverage in 2018).
By the early 1970s, advances in integrated circuit technology had made digital transmission a cost-effective way
of providing telephony, which was hitherto an exclusively analogue communication service. Pulse code modulation
(PCM), a signal processing technique devised by British engineer Alec Reeves (1902–1971) back in 1937, was
employed to digitise the analogue speech signal by converting it into a sequence of numbers. Digital exchanges and
TDM began effectively to replace analogue technology, i.e. analogue exchanges and frequency division multiplexing
(FDM) – in the PSTN during the 1970s, although the first TDM telephony system had been installed as far back as
1962 by Bell Laboratories in the USA. The momentum of digitalisation of the entire transmission system increased
in the 1990s, until by the turn of the century the telecommunication networks of many countries had become
nearly 100% digital. Globally, in 2020, the only remaining analogue portion of the telecommunication network is
the local loop connecting, via copper wire pair, a subscriber’s landline telephone handset to the local exchange, or
a street cabinet beyond which transmission is entirely digital.
Digital audio broadcasting began in September 1995 with some field trials by the British Broadcasting
Corporation (BBC). Although analogue amplitude modulation (AM) and frequency modulation (FM) radio
broadcast will continue into the foreseeable future, digital radio now provides the best in broadcast sound quality,
especially in the presence of multipath distortion. Terrestrial digital television broadcast was first launched in the
UK in 1998 and provides a more efficient utilisation of the radio spectrum and a wider service choice than its
analogue counterpart. Transition to terrestrial digital television broadcasting and a total shutdown of analogue
television broadcasting (a process known as digital switchover) has been completed in many countries around
the world. For example, analogue TV broadcasting was fully terminated in the UK on 28th November 2013 and
in China on 14th May 2016. Also, TV broadcasts via satellite have been migrated from analogue transmission
using FM to digital transmission using a selection of bandwidth-efficient digital modulation techniques. It is
therefore likely that analogue television broadcasting, first started by the BBC in 1936, will cease to exist within
this decade.
By the turn of the century, supported by a variety of reliable mass data storage media, the digital era had fully
arrived, and the convergence of communication, broadcasting, computing, and entertainment had become a real-
ity. This convergence is best epitomised at the device level by the smartphone and at the network level by the
Internet. Both support multimedia streaming, audio and video (IP) telephony, mobile computing, and gaming.
In view of the rapid pace of development and diverse range of services, a detailed study of all telecommunication
applications would be an extremely challenging and inadvisable task. Fortunately, the underlying principles do
not change and the best way to acquire the necessary expertise in this exciting field is to first become grounded in
its principles. It is for this reason that this book will focus on a thorough yet accessible treatment of the principles
of communication engineering, drawing on modern examples and applications as necessary to ensure a complete
Telegram: @Persian_IEEEComSoc

understanding of the concepts presented as well as a better appreciation of the current state of the art.

1.3.3 Developments in Transmission Media

Much of the growth in telecommunications would have been impossible without significant developments in
transmission media which provide suitable paths for signals to travel from source to destination. The first telegraph
circuit built by Wheatstone and Cooke between Euston and Camden stations consisted of iron wires insulated with
cotton and laid in iron pipes buried beside the railway line. The insulation and hence the line failed whenever the
wires became wet. This forced Wheatstone and Cooke to abandon the idea of buried cables and opt rather for
suspending the wire from telegraph posts using glass insulators. From this very humble beginning, transmission
medium technology developed to encompass in chronological order copper cables, and radio and optical fibre,
 1.3 Modern Telecommunication 17

all of which have been used for short- to long-distance links. We will briefly review the historical development of
these three media. But note that other transmission media suitable only for very short links ranging from a few
centimetres to a few metres have also been developed over the years. These include microstrip lines in printed
circuit boards, metallic waveguides used, for example, to connect an outdoor antenna to an indoor receiver unit,
and infrared radiation used, for example, in the remote control of gadgets.

1.3.3.1 Copper Cable

The telegraph poles that crisscrossed America and Europe in the early days of telegraphy carried iron or steel wires,
one wire for each circuit with the ground providing the return path. It was known back then that copper was a
much better electrical conductor than iron. In fact, the first ever undersea telegraph cable, laid between Dover
(UK) and Calais (France) in 1851, consisted of four copper conductors insulated with a coating of gutta-percha
(the natural latex of the native Malaysian Palaquium tree brought to the UK in 1843) and protected by iron sheath-
ing. Copper was also the conducting material for the first successful trans-Atlantic telegraph cable laid in 1866
(after three previous failed attempts in 1857, 1858, and 1865) between Valentia Island (Ireland) and Newfound-
land (USA), a distance of 1852 miles. Copper, however, is a soft metal and the annealing technique necessary to
make it strong enough to support its own weight from pole to pole had not yet been discovered. This situation
changed in 1877, when Thomas Doolittle (1839–1921) invented a suitable copper wire manufacturing process.
As a result, in the 1880s copper wire replaced iron as the standard transmission medium, and the single wire
circuit (with ground return path) was replaced by a two-wire (i.e. wire pair) circuit, which significantly reduced
noise. Further improvement was obtained through the introduction of twisted-pair cables in which the two wires
comprising one circuit are independently insulated and twisted around each other. Multipair cable was introduced,
consisting of a lead tube containing up to 100 copper wires, each insulated with paraffin-impregnated cotton or
gutta-percha.
By 1900, to avoid the unsightly congestion of aerial wires, short-distance cables began to be laid underground
in conduits made initially of creosoted wood and then of vitrified clay. The first long-distance underground cable
was laid in 1912 and carried analogue telephone signals between Philadelphia and Washington DC. Long-distance
underground wire pairs were replaced by coaxial cables in the 1940s. Underground cable cores bearing the twisted
wire pairs of the PSTN local loop are a significant global investment of the twentieth century which could not be
simply discarded in the digital era. In the 1990s, technological advances (especially in very large scale integration
(VLSI) circuits to perform the required complex signal processing) allowed the copper lines of the local loop to be
used as a vital medium for digital transmission known as the digital subscriber loop (DSL). From 2003, there was
a rapid global uptake of DSL-enabled broadband connection to the home.
AT&T invented the coaxial cable and installed the first experimental coaxial cable system between New York
and Philadelphia in 1936, followed in 1941 during World War II by the first commercial installation between
Minneapolis and Stevens Point. A coaxial cable (or coax for short) consists of an insulation-covered, tube-shaped
outer conductor (sometimes of braided design) with a second conductor fixed into the centre of the tube and
Telegram: @Persian_IEEEComSoc

separated from it by other insulation. It provides better protection from interference and larger bandwidths than
twisted-pair configurations. The early L1 coaxial cable installations of the 1940s could carry 480 telephone calls,
whereas the L5 systems of the 1970s were vastly superior and could carry 132 000 simultaneous calls. The laying
of the first transatlantic coaxial cable (TAT 1) spanning 2240 miles between Oban in Scotland and Clarenville in
Newfoundland was completed in 1956. With an initial capacity of 36 telephone circuits, TAT 1 remained in service
for 22 years, although several other TAT coax cables were laid in the ensuing years to provide more capacity. The
last transatlantic coaxial cable to be laid was TAT 7 in 1978 with an initial capacity of 4000 telephone circuits. In
the 1980s, coaxial cables were superseded by optical fibre and most long-distance coaxial cable installations have
now been retired.
As the above evolution of the transmission medium from iron wire to copper coax took place, improvements
were also made in insulation material, from cotton through gutta-percha and paper to finally (synthetic) plastic.
 18 1 Overview of Communication Systems

1.3.3.2 Radio
James Clerk Maxwell (1831–1879), an outstanding university professor by age 25 and widely regarded as the most
brilliant British mathematician and physicist of the nineteenth century, made a theoretical prediction of the exis-
tence of radio waves in 1864. His theory is embodied in four equations, known as Maxwell’s equations, which when
combined yield two wave equations – one for a magnetic field and the other for an electric field – propagating at
the speed of light. Simply put, Maxwell’s theory stipulates that a changing electric field generates a changing mag-
netic field in the surrounding region, which in turn generates a changing electric field in the surrounding region,
and so on. In this way a coupled electric and magnetic field travels out in space at the speed of light (≈ 3 × 108 m/s),
and the resulting wave is known as an electromagnetic wave.
The significance of Maxwell’s theory for telecommunications is that if we can somehow generate an electro-
magnetic wave having one of its parameters – amplitude, frequency, or phase – varied in sync with the variations
of an information-bearing voltage signal then we can transmit information from one point to another at the speed
of light and without the need for a cable connection. This is radio or wireless communication. For example, to
implement wireless telegraphy we can switch the wave on or off to signal a mark or space, respectively, and at some
distance recover the sequence of marks and spaces by detecting the presence and absence of the wave during each
signalling interval.
A German physicist called Heinrich Hertz (1857–1894) provided an experimental verification of Maxwell’s wave
theory in 1888. Using a wire connected to an induction coil to generate the electromagnetic waves and a small loop
of wire with a spark gap to detect them, he measured their wavelength and showed that the waves were reflected
and refracted in a similar way to light.
The most significant contribution to the establishment of radio communications at the turn of the twentieth
century was made by Guglielmo Marconi (1874–1937), who was born and grew up in Bologna, Italy but moved
to Britain in 1896 in search of financial support for his radio experiments. In 1897, he sent signals by radio over
eight miles from Lavernock Point to Brean Down, made the first radio transmission from ship to shore while on a
visit to Italy, and formed the Wireless Telegraph & Signal Company. In 1899, he put his invention to work for the
international yacht races being held off New York harbour at that time. He followed the yachts around the course
and reported the races for two newspapers using wireless telegraph from yacht to shore. This event helped bring
wireless telegraphy to prominence in the US. In December 1901, Marconi achieved the first transatlantic radio
transmission covering 2800 km from Poldhu in Cornwall (UK) to Signal Hill in Newfoundland (Canada); and by
1907 he had established a transatlantic wireless telegraph service.
Radio communication started in 1897 with point-to-fixed-point wireless telegraphy using Morse code – a dig-
ital transmission that was simple enough to implement using the technology of the time by switching the radio
signal on and off. It was not long, however, before the broadcast and mobility potential of radio became obvious.
In 1904, Marconi started a service of Morse-coded news broadcasts to subscribing transoceanic ships, although
any Morse code literate person with the right equipment could receive the news for free. The exploitation of radio
communication grew very rapidly during the twentieth century to a point today where modern society is inextri-
Telegram: @Persian_IEEEComSoc

cably dependent on radio communication. Some of the most notable early developments that made this possible
are summarised below.
Reginald Fessenden (1866–1932) invented amplitude modulation (AM) in 1906, which for the first time allowed
the amplitude of a radio carrier signal to be continuously varied according to variations of an analogue (voice) sig-
nal. This paved the way for audio broadcasts and radio telephony once the design of vacuum tube transmitters and
receivers was perfected in the mid-1910s. The first AM radio broadcast was the transmission of the US presiden-
tial election returns from the Westinghouse station at Pittsburgh in November 1920. Commercial radio broadcasts
started in 1922 and stations proliferated in the US after it was finally figured out that the service could be financed
through advertising. In the UK, the BBC was established and started its broadcasts in the autumn of that year, paid
for through licence fees – a practice which has continued till today. The first transatlantic radio telephone service
 1.3 Modern Telecommunication 19

was established in 1927 between London and New York using a 60 kHz radio carrier. But with a capacity of only
one call at a time the service was extremely expensive, attracting a tariff of £15 for a three-minute call.
Edwin Howard Armstrong (1890–1954) invented the frequency modulation (FM) technique in 1934 and FM
radio broadcasts started in 1939. The FM technique varies the frequency of the radio carrier in proportion to the
variations of the voice signal and yields a superior received signal quality compared to AM. In 1940, the Connecti-
cut State police put into operation the first two-way FM mobile radio system designed by Daniel Noble (1901–1980),
a professor of electrical engineering in the University of Connecticut.
In 1936, radio broadcast of television was started by the BBC in the UK. The first (black-and-white) television
sets cost as much as a small car. Television had been invented in the 1920s, with significant contribution from
John Logie Baird (1888–1946) – who invented mechanical television and successfully demonstrated a working
prototype on the 26th of January 1926 to members of the Royal Institution in London. However, it was Philo
Farnsworth (1906–1971) who demonstrated the first all-electronic television system in 1928.
In the 1940s, radio began to be used to provide high-capacity, long-distance transmission links for telephone and
television signals. Called microwave radio relay, the system consisted of a relay of line-of-sight radio transceivers
arranged on towers and had the advantage of lower construction and maintenance costs compared to coaxial cable
transmission systems. It provided an alternative to cable and by the 1970s carried most of the voice and TV traffic;
but it was surpassed in the 1980s by optical fibre.
In 1945, Arthur C. Clarke, famous as a science fiction writer, suggested in an article in the British radio magazine
Wireless World that a long-distance radio link could be established using a satellite located in an equatorial orbit
with a period of 24 hours. In this orbit known as geostationary orbit (GEO), the satellite would appear stationary
when viewed from any point on the earth’s surface. A signal transmitted from an earth station to the satellite could
be retransmitted by the satellite back to earth and received at all points on the visible earth. This would make pos-
sible broad-area coverage and long-distance radio communication that was more reliable than the only method
available at the time via HF radio transmission. This vision of satellite communications became a reality 20 years
later when the first commercial satellite, Early Bird (later renamed INTELSAT 1), was launched into GEO and
commenced operations on the 28th of June 1965. The first spacecraft (Sputnik I) had been launched by the USSR
(now Russia) in 1957, but it was the satellite Telstar I, built by the Bell Laboratories and launched in 1962 into a
158-minute low earth orbit (LEO), that relayed the first television signals between USA and Europe. Demand for
satellite communication has grown steadily since its inception. In the 1970s and 1980s, satellite systems were used
for international and domestic telephony and television distribution. The arrival of superior optical fibre transmis-
sion systems in the 1980s shifted telephone traffic from satellite to optical fibre links. A bold attempt in the 1990s
to provide global mobile communication using a constellation of satellites in LEO or medium earth orbit (MEO)
ended in commercial failure due in part to the unforeseen runaway success of terrestrial cellular mobile com-
munications. But satellite communication is now well established as the leading means of television programme
distribution directly to homes, and of communication at sea and in remote, undeveloped or disaster areas. The
GPS service is also a satellite communication application. Satellite communication is also used to provide broad-
Telegram: @Persian_IEEEComSoc

band Internet access for subscribers in remote areas or in regions without adequate terrestrial communication
infrastructure.

1.3.3.3 Optical Fibre

The use of optical fibre as a transmission medium was first proposed in 1966 by Charles K. Kao of Standard
Telecommunications Laboratories, in the UK. An optical fibre is a dielectric waveguide about 125 μm in diam-
eter made from high-purity silica glass. It consists of a glass core surrounded by a glass cladding of lower refractive
index. A plastic sheath covers the cladding to give mechanical protection. An optical cable contains several fibres
used in pairs – one for each transmission direction. Light waves propagate within the inner core over long distances
 20 1 Overview of Communication Systems

through total internal reflection at the core-cladding interface. This light can be switched on and off to signal bits
1 and 0, respectively, and thereby convey information.
At the time of Kao’s proposal the best available fibre material had a loss of about 1000 dB/km, but Kao was
convinced that the high loss was due to absorption by impurities in the glass, and that losses ≤20 dB/km could be
achieved to make long-distance optical fibre transmission links possible. The first breakthrough in reducing fibre
attenuation came in 1970, when Robert Maurer of Corning Glass Works used fused silica to make a fibre having
an attenuation of 16 dB/km at a wavelength of 633 nm. Since then, improved fabrication technology and a move to
higher wavelength regions – called windows – offering lower optical losses have resulted in dramatically reduced
attenuation. There are four such windows at 850, 1310, 1550, and 1625 nm. The first-generation fibre systems
were installed in 1977 and operated at 850 nm with an attenuation of 3 dB/km. Most operational systems today
use the second window at 1310 nm with attenuation 0.5 dB/km and the third window at 1550 nm with attenuation
0.2 dB/km. The technique of Raman amplification enables use of the fourth window at 1625 nm (of attenuation
around 0.3 dB/km) especially in ultra-dense wavelength division multiplexing (WDM) systems. The attenuation
figures quoted here refer to intrinsic losses only. This is the loss inherent in the fibre material due to Rayleigh
scattering and absorption by impurities. Optical fibre is also subject to external sources of attenuation known as
extrinsic losses (e.g. due to sharp bending, coupling, and splicing), which may sometimes account for the bulk of
the total attenuation.
Optical fibre has numerous advantages over radio and copper lines as a transmission medium, but the most
significant are its low attenuation (stated above) and large bandwidth. This small attenuation allows a wider spac-
ing of repeaters in optical fibre transmission links than is possible in copper lines, leading to reduced costs. For
example, the last (and best) transatlantic copper cable TAT-7 had 662 repeaters compared to only 109 repeaters in
the first (and poorest) transatlantic optical fibre cable TAT-8.
Optical fibre conveys an optical carrier of very high frequency – about 200 000 GHz. So, assuming a transmission
bandwidth of about 5% of the supported carrier frequency, we see that optical fibre offers a potential bandwidth
of about 10 000 GHz. The technology exists today – using soliton laser, erbium-doped fibre amplifier (EDFA) and
dense wavelength division multiplexing (DWDM) – to allow long-distance transmission at several terabits per
second on optical fibre without regeneration.
Starting from the 1980s, optical fibre has replaced coaxial cables in all long-distance and undersea transmission
lines, and in the inter-exchange and metropolitan area networks. The first international submarine optical fibre
system was installed in 1986 between the UK and Belgium. In 1988, the first transatlantic optical fibre system,
TAT-8, was installed between Europe and the USA with a capacity of 560 Mb/s. Several more transatlantic optical
fibre cables have been installed since then, up to the latest TAT-14, which became operational in 2001 with a
capacity of 3.15 Tb/s.
Two transmission media frontiers have withstood the optical fibre revolution started in the final years of the
twentieth century. First, the local loop – the so-called last mile of the PSTN – continues under the dominance of
copper wire pairs. The reason for this is largely economic. Fibre in the local loop makes ultra-broadband connec-
Telegram: @Persian_IEEEComSoc

tion to the home possible, but it involves significant extra costs for cable installation and new terminal equipment,
costs which (unlike those of the long-distance segment) cannot be spread over several subscribers. Furthermore,
the power supply required to operate the subscriber unit is currently sent from the local exchange along the same
copper wire pair that carries the message signal, whereas fibre being an insulator imposes the burden of an alter-
native power supply arrangement. But these problems are not insurmountable and optical fibre is now beginning
to replace copper in the local loop and a twenty-first century broadband network is emerging as a natural successor
to the PSTN legacy of the last century.
The other frontier, the only one that is impregnable to optical fibre, is the connection or access provided by
radio for mobile communication units on land, sea, and air, and in remote, undeveloped or disaster-stricken areas.
The mobility and rapid deployment capabilities of radio in such applications are unique qualities that cannot be
delivered by optical fibre.
 1.4 Communication System Elements 21

1.4 Communication System Elements

Modern communication systems vary widely in their applications and complexity, but they are all accurately rep-
resented by the block diagram shown in Figure 1.5. A variety of information sources feed a transmitter with the
message signal to be transmitted. The transmitter transforms the message signal into a form that is compatible
with the type of communication system and that is suitable for passage through the transmission medium with
an acceptably small distortion. The output of the transmitter, known as the transmitted signal, is placed into the
transmission medium, which conveys it to a receiver located at the intended destination. The received signal at
the output of the transmission medium and input of the receiver is a distorted version of the transmitted signal.
Noise, distortion, and reduction in strength have been introduced by the medium. The receiver has the task of
removing (as far as possible) the transmission impairments and undoing each operation performed by the trans-
mitter. It then delivers an exact or close copy of the original message to a user or information sink. The receiver
is also selected to match the characteristics of the transmission medium and to be compatible with the type of
communication system.
In what follows, we give a brief overview of information sources and sinks. A nonmathematical introduction
to the processing tasks of the transmitter and receiver is also presented, which lays a good foundation for a more
detailed treatment in the rest of the book. However, an introduction to transmission media is postponed until the
first section of Chapter 5, which is devoted to a detailed treatment of the topic.

1.4.1 Information Source

The information source or input device acts as an interface between the communication system and the outside
world and provides the message signal that is processed by the transmitter. There are four main classes, namely
audio, video, and data input devices, and sensors.

Communication system

Transmitted Transmission Received

signal medium signal

Message Transmitter Receiver Estimate of

signal message
signal
Telegram: @Persian_IEEEComSoc

Audio input Audio output

device device
Data input Visual display
device device

Video input Storage

device device

Information Information
source sink

Figure 1.5 Block diagram of a communication system showing its major elements.
 22 1 Overview of Communication Systems

1.4.1.1 Audio Input Devices

The microphone is the most basic audio input device. It converts acoustic (sound) pressure wave into an electrical
signal (current or voltage) of similar variations, which serves as the message signal input of the communication
system. A voltage supply is often required for the operation of the microphone, or for the amplification of the weak
electrical signal it generates. Other audio input devices may include a music keyboard, musical instruments digital
interface (MIDI), audio players, and other digital music instruments. There are different types of microphones
based on the principle of their operation.
Dynamic microphone: the dynamic or electromagnetic microphone is based on Faraday’s law of induction,
which stipulates that a changing magnetic field will induce voltage in a conductor lying within the field. In the
moving coil dynamic microphone shown in Figure 1.6, a wire coil enclosing a fixed permanent magnet is physically
attached to a diaphragm. Pressure waves from a sound source fall on the diaphragm, causing it to vibrate along
with the coil. The motion of the coil in the field of the permanent magnet induces an electromotive force (electrical
output) in the coil, which varies in sync with the sound pressure waves. A moving magnet dynamic microphone
is also possible in which the coil is fixed and the magnet is attached to the diaphragm. In this case, sound waves
moving the diaphragm also move the magnet, causing a change in the magnetic field in which the coil lies and
hence inducing electromotive force in the coil. A type of dynamic microphone with much better fidelity at high
frequencies is the ribbon microphone. Here a light (low-inertia) aluminium ribbon is suspended edgewise between
the poles of a magnet. Electromotive force is induced in the ribbon when it is moved in the magnetic field by air
flowing past it from sound waves.
Piezoelectric microphone: when a piezoelectric material is subjected to mechanical strain, voltage is induced
between its ends. The operation of a piezoelectric microphone is based on this important property. If one end of
such a material is fixed and the other end is attached to a diaphragm, sound waves falling on the diaphragm will
cause a strain in the material. This induces a voltage that varies according to the strain and hence according to the
sound pressure.
Carbon microphone: in a variable-resistance or carbon microphone, carbon granules are packed in a chamber
formed by one fixed and one movable electrode. The movable electrode is attached to a diaphragm. Sound waves
falling on the diaphragm vary the packing density of the granules and hence the resistance of the chamber. The
variable resistance R is converted to a varying current I by connecting a constant voltage V across the chamber
since, by Ohm’s law, I = V/R.
Electret microphone: an electret microphone is of variable capacitance design. A capacitor is formed using two
metallic plates, one fixed and the other movable. The movable plate is a thin metallic layer diaphragm. The fixed
plate is a metal plate covered over with a film of Teflon-like plastic that has a permanent electric charge Q, hence
the name electret. Air pockets are trapped between the electret and the fixed electrode due to irregularities on the
surface of the electret. When sound waves fall on the movable diaphragm, the size of the air pockets and hence
the capacitance C of the capacitor is varied. This produces a variable voltage V according to the relation V = Q/C.
An alternative variable capacitance design is the condenser microphone in which the charge Q is maintained by
Telegram: @Persian_IEEEComSoc

an external voltage source.

Coil, physically attached

to diaphragm ʋ(t)
Electrical
output t
Sound wave
input N S

permanent magnet
Diaphragm

Figure 1.6 Moving coil dynamic microphone.

 1.4 Communication System Elements 23

1.4.1.2 Video Input Devices

A video input device presents a video message signal to the communication system. The video signal is a vari-
able voltage that originates from a visual signal, which consists of light intensity (or luminance) and colour (or
chrominance) information. The most basic video input devices are the video camera (e.g. digital camera, analogue
camcorder, and webcam) for dynamic three-dimensional (3D) images and the scanner (e.g. image scanner, 3D
scanner, barcode reader, and fingerprint scanner) for still images. Secondary input devices include those that read
pre-recorded video signals such as the video cassette recorder (VCR), which reads analogue video signals stored
on magnetic tape in two-reel cassettes, and digital video recorders.

1.4.1.3 Data Input Devices

There is a wide variety of data input devices, which generate data serving as the message signal input of a com-
munication system. Many of these devices work in conjunction with a computer system or other suitable data
terminal equipment (DTE). A few examples of data input devices are given below.
● A keyboard generates the Unicode or ASCII code corresponding to the key pressed.
● Several devices are used to convert finger and hand movement into computer commands, which may involve
data generation under the control of a computer program. Examples of such devices include the mouse, trackball
(an upside-down mouse), joystick, touchscreen, touchpad, pointing stick, and virtual reality systems.
● There are different types of devices that read stored data and present these to the communication system for
processing and/or transmission. A barcode reader or laser scanner scans a barcode, which encodes information
(e.g. the price of an item) in black lines of varying width. A magnetic-ink character reader reads information
written on a document using magnetic ink characters. A magnetic strip reader reads data stored magnetically
on magnetic strips, such as those found on credit cards, security access cards, etc. Optical character readers and
optical mark readers are used (in conjunction with software) to read printed characters and to detect and decode
marks on paper. Punched card readers read information stored as an array of small holes and spaces on paper.
Other data-reading input devices include the optical disk drive for reading data from an optical disk, the disk
drive for reading data from a floppy or hard disk, and the tape reader for reading data from a magnetic tape.
● A digitising tablet is used to convert graphics information (drawn on the tablet) into computer data. The drawing
is displayed on a computer screen and the associated data may be processed by a communication system.
● A smart pen is used to capture handwritten or other graphics information, which may then be converted into
data for processing in a communication system.

1.4.1.4 Sensors
Sensors measure physical quantities such as temperature, pressure, mass, etc., and convert the measurement
into an electrical signal that serves as the message signal input for the communication system. Sensors are used
in telemetry systems to obtain information from remote or inaccessible locations, e.g. monitoring the status of
Telegram: @Persian_IEEEComSoc

equipment on a spacecraft or detecting the accumulation of ice on the wings of an aircraft. They are also used
in numerous systems including automatic data-logging systems, security systems, safety systems, traffic control
systems, manufacturing systems, and process control systems. For example, a tipping-bucket rain gauge generates
a voltage pulse each time a standard-size cup tips after filling with rainwater. The irregular sequence of voltage
pulses serves as the message signal, which may be transmitted and analysed to obtain the rainfall rate. A security
system may use a sensor to detect movement. An alarm (or message) signal is then generated and transmitted to
a console and used to initiate appropriate action, such as the ringing of a bell.
In the most general sense, every input device can be described as a sensor of some sort. The camera senses
reflected light, the microphone senses sound pressure, the keyboard senses key press, the mouse senses hand
movements, etc. However, the classification of input devices presented here is useful in identifying the type of
information provided by the device to the communication system, whether audio, video, data, or a more general
measurement of a physical quantity.
 24 1 Overview of Communication Systems

1.4.2 Information Sink

An information sink is the ultimate destination of a transmitted signal. It may serve as an interface between
human users and the communication system, making the transmitted information understandable through sight
or sound. It may also serve as a repository, storing the transmitted information for later processing, re-transmission,
or display. There are therefore three types of information sink, namely audio output devices, visual display devices,
and storage devices.

1.4.2.1 Audio Output Device

Generation of sound: the loudspeaker is an audio output device that converts the electrical signal output of a
communication system into mechanical vibrations at the same frequencies. Figure 1.7 shows the basic operation
of a moving coil loudspeaker. A cone is physically attached to a wire coil, which encloses a permanent magnet
lying along its axis. The electrical signal is fed as current into the coil. This current sets up a magnetic field that
varies proportionately with the current and has maximum strength along the coil axis. The interaction of the field
of this electromagnet with the field of the permanent magnet results in a force that varies in sync with the field
strength of the electromagnet and hence with the input current. This varying force causes the coil along with the
attached cone to vibrate, thereby producing sound waves that follow the input current variations. The vibrations
also generate sound waves travelling in the backward direction, which are reflected by the loudspeaker casing,
and may interfere in a destructive manner with the forward waves. To prevent this, the casing is padded with
sound-absorbent material. It is usual to have separate loudspeakers, a woofer for the low frequencies and a tweeter
(with a smaller cone) for the high frequencies.
Sound pressure level (SPL): a sound signal or sound wave is an oscillation in pressure, stress, particle dis-
placement, or particle velocity in an elastic medium such as air. When this sound wave is received and processed
by the ear–brain mechanism, the sensation of hearing is produced, provided two conditions are met: (i) the oscil-
lation or vibration frequency must be within the audible range; and (ii) the SPL produced by the vibrations must
be high enough, i.e. above the threshold of hearing.
The ear only perceives sound signals of frequency in the range 20 Hz to 20 kHz. This is the theoretical audible
range of frequencies. However, in practice most people have a narrower hearing range, with the top end decreasing
with age after about age 30 years. Sound pressure may be expressed in dyne/cm2 , microbar, or Newton/m2 , where

1 dyne∕cm2 = 1 microbar = 0.1 N∕m2 (1.1)

The threshold of hearing for an average person below age 30 when the sound signal is at a frequency of 1 kHz is
0.0002 dyne/cm2 . That is, sound of 1 kHz in frequency will be heard (by this standard listener) only if the vibrations

Cone, physically
Telegram: @Persian_IEEEComSoc

attached to coil
Electrical
coil
input
ʋ(t)
Sound
S N
output
t

Permanent
magnet

Figure 1.7 Audio output device: a loudspeaker.

 1.4 Communication System Elements 25

produce (in the ears) a sound pressure of at least 0.0002 dyne/cm2 . This value has been adopted as a reference PREF
for expressing SPL in decibel (dB). Thus, the SPL of a sound of pressure P dyne/cm2 is expressed in dB above PREF as
( )
P
SPL = 20log10 dB,
PREF
where, PREF = 0.0002 dyne∕cm2 (1.2)
The SPL of ambient noise in a quiet home is about 26 dB, that of formal conversation is about 62 dB, whereas
that of jet planes at take-off is about 195 dB. The threshold of hearing increases irreversibly with age above 30 and
is raised to a higher level in the presence of ambient noise. The second effect is transient and is referred to as noise
masking. The dynamic range of the ear is typically 120 dB, bounded at the bottom by the threshold of hearing and
at the top by the threshold of pain. Sounds of SPL < 0 dB are inaudible, whereas sounds of SPL ≥ 120 dB will cause
pain and may result in an immediate and irreversible loss of hearing.
Frequency response of the ear: whereas SPL is an objective measure of the amplitude of sound vibrations,
another closely related term, loudness, is not. Loudness is the response of the human ear to the amplitude of sound
waves. It is a subjective attribute that allows us to place a given sound at a point on a scale from soft to loud. The
loudness level of a sound is expressed in phon, which is numerically equal to the SPL (in dB) of a 1 kHz reference
tone. The sensitivity or response of the human ear to sound varies markedly with frequency. A 1 kHz sound at an
SPL of 130 dB is deafeningly loud, but a 22 kHz sound at the same SPL is not loud at all. In fact, you would not
hear the latter (because it is outside your audible range), although both vibrations produce the same high level of
sound pressure.
The ear acts like a band pass filter, shutting out all frequencies outside the audible band. Within the audible band
itself, the response of the ear is far from uniform. Figure 1.8 is the ISO-226 standard curves of the SPL to maintain
constant loudness at various frequencies. It shows, for example, that to produce sound of loudness 40 phon, one
needs SPL of 40 dB at 1 kHz and a higher SPL of 77 dB at 30 Hz. We observe also that the ear’s frequency response
is more variable when dealing with softer sound, but reasonably flat with loud sounds (SPL > 100 dB).

120
110 115 120 phon
110
100 100
Sound pressure level (SPL), dB

90 90
80 80
70 70
60 60
Telegram: @Persian_IEEEComSoc

50 50
40 40
30 30
20 20
10 10
0 4.2

20 50 100 200 500 1k 2k 3k 7k 10k

Frequency, Hz 12.5k

Figure 1.8 Contours of equal loudness (phon) as a function of frequency and sound pressure level.
 26 1 Overview of Communication Systems

1.4.2.2 Visual Display Devices

Three broad classes of visual display devices are common as information sinks in communication systems.
Cathode-ray tube (CRT) and flat panel displays present what is usually called a soft copy of the received message
signal, whereas printers produce a hard copy of the message.
Cathode-ray tube (CRT): cathode ray tubes display information (pictures, text, graphics, etc.) by using an
accelerated beam of electrons to excite electrons in phosphor atoms, which emit visible energy as they return
to their unexcited state. The basic parts of a CRT are shown in Figure 1.9. The CRT consists of an electron gun, a
focusing system, a deflection system, and a phosphor-coated screen. Electrons are boiled off a hot cathode surface,
a process called thermionic emission. The number of electrons emitted by the gun and hence the brightness of
the display is controlled by a negative voltage on the control grid. A focusing system is needed to prevent beam
spreading due to mutual repulsion among the electrons and to focus the electron beam to a small point on the
screen. The focusing system may be an electrostatic lens formed by the electric field of a positively charged metal
cylinder, or a magnetic lens formed by the magnetic field of a current-carrying coil mounted around the outside
of the CRT envelope.
The deflection system is used to direct the beam to any spot on the screen. Figure 1.9 shows an electrostatic
deflection system. Magnetic deflection may be obtained using current in a pair of coils mounted on the top and bot-
tom of the neck of the CRT envelope for horizontal deflection and another pair of current-carrying coils mounted
left and right for vertical deflection.
The phosphor atoms on the screen at the spot impacted by the accelerated electron beam will glow for a short
time. The duration or persistence of the glow and its colour depends on the type of phosphor. To maintain an image
without flicker, the entire screen is refreshed at a regular rate by repeatedly moving the electron beam along rows
of the screen from top to bottom, while simultaneously varying the beam intensity to reproduce the desired image.
For a colour display, each spot has three phosphor dots that glow with a red, green, and blue colour. Three
electron guns are used, one for each colour dot. Any desired colour can be produced at each spot by (additive)
mixing (of) these three primary colours in the right intensity. For example, a yellow colour is obtained by exciting
the green and red phosphor dots equally without exciting the blue phosphor dot – that is, the electron gun aligned
with the blue phosphor dot is switched off.
Despite its excellent display quality, the weight, bulk, and power consumption of a CRT make it unsuitable for
portable devices (e.g. pocket calculators) and the large displays required for example in high-definition television
(HDTV). After the turn of the century, CRTs began to be superseded by flat panel displays.
Flat panel displays: flat panel displays have smaller volume, weight, and power consumption than a CRT. They
are now the preferred technology for use in a wide range of applications requiring monitors, e.g. TV, computers,

Horizontal
Cathode Phosphor-
deflection plates
coated
Telegram: @Persian_IEEEComSoc

screen

Electron
beam
Heating filament Focusing Accelerating Vertical
anode anode deflection plates
Control grid

Electron Focusing Deflection

gun system system

Figure 1.9 Basic components of a CRT.

 1.4 Communication System Elements 27

mobile telephone units, etc. There are two main classes of flat panel displays. One class, called emissive display,
works by converting electrical energy into light. The other class, called non-emissive display, uses an arrangement
that modulates (i.e. blocks or passes to varying degrees) light from an external or internal source, thereby creating
a graphics pattern.
Examples of an emissive flat-panel display include a light-emitting diode (LED) display, plasma-panel (or gas
discharge) display, and thin-film electroluminescent display.
In a LED display, a matrix of diodes is arranged to form the screen spots or picture elements (called pixels) and
a picture is displayed by applying a voltage to light the diode at each required location. Recent developments have
led to widespread use of organic light-emitting diode (OLED) displays in which the electroluminescent substance
is an organic compound that emits light in response to electric current. Sony produced the world’s first TV set
using an OLED display screen in 2007. In addition to TV and computer screens, OLED displays are now used in a
variety of portable devices, including the smartphone.
In plasma-panel and electroluminescent displays, the volume between two glass panels is filled with a suitable
substance – a mixture of neon and some other gases for the former, and a manganese-doped phosphor for the
latter. One panel carries a series of vertical conducting ribbons, and the other carries a series of horizontal ribbons.
The intersection between a vertical ribbon and a horizontal ribbon defines a display pixel. To light a pixel, firing
voltages are applied to the pair of ribbons that intersect at that point. This breaks down the gas at the pixel to
form glowing plasma, or, in the case of the electroluminescent display, causes the manganese atoms at the pixel
to absorb electrical energy, which makes them glow. The firing is refreshed at a regular rate (e.g. 60 Hz) using the
picture definition stored in a refresh buffer.
A liquid crystal display is non-emissive. In this case, the volume between two glass plates is filled with a liquid
crystal compound (LCC), a substance that has a crystalline arrangement of molecules but flows like a liquid. One
of the glass plates contains rows of transparent conductors and a light polariser. The other plate contains columns
of transparent conductors and a light polariser at right angles to the polariser in the other plate. The intersection
of a row conductor with a column conductor defines a pixel. A pixel is turned on when the LCC molecules at its
location are unaligned and is turned off when these molecules are forced to align by the application of a voltage.
This happens because the unaligned molecules at the pixel twist the polarised light, causing it to pass through
both polarisers. On the other hand, if a voltage is applied to two conductors it aligns the LCC molecules at their
intersection and prevents the twisting of the light so that it cannot pass through both polarisers, thereby becoming
blocked.
Printers: impact printers produce hard copies by pressing the face of formed characters against an inked ribbon
onto the paper. The characters are usually formed using a dot-matrix print head, which has a rectangular array of
protruding pins, by retracting appropriate pins. The quality of the print depends on the dot size and on the number
of dots per inch (or lines per inch) that can be displayed.
Various nonimpact printing techniques have been devised. Laser printers create a charge distribution on a rotat-
ing drum using a laser beam. Ink-like toner particles are then emitted from a toner cartridge onto the drum. These
Telegram: @Persian_IEEEComSoc

particles adhere only to the charged parts of the drum. The particles are then attracted from the drum to the paper
and pressed into the paper by heated rollers. Ink-jet printers use an electric field to deflect an electrically charged
ink stream to produce dot-matrix patterns on paper. Electrostatic printers charge the paper negatively at the right
spots. These spots then attract a positively charged toner to create the required picture or text. Electrothermal
printers apply heat in a dot-matrix print head to create the required pattern on a heat-sensitive paper.
Colour printing is obtained by mixing the right quantity of the colour pigments cyan, magenta and yellow. For
example, equal mixture of magenta and cyan gives blue and equal mixture of all three pigments gives black. Laser
colour printers deposit the three pigments on separate passes to produce a range of colour patterns, while ink-jet
colour printers shoot the three colours simultaneously.
Finally, we should mention that 3D printing has become commonplace in recent years. If viewed as an additive
manufacturing process then a 3D printer is not a display device. However, if seen as a process to produce a physical
 28 1 Overview of Communication Systems

3D object from information stored on a computer then a 3D printer qualifies as a display device. A 3D printer
produces a physical object through layer upon layer deposition of material. Information stored on the computer
specifies the model to be produced, including the material to be deposited at each voxel (i.e. 3D pixel).

1.4.2.3 Storage Devices

Many communication services are designed to store the received message signal for visual or audio display later.
In email, the text message is stored at the recipient’s email server until the recipient accesses the server to obtain
a soft copy (screen display) or hard copy (printout) of the message. Voicemail service provides for the storage of
digitised speech message, which is played back later to the authorised recipient at their request. In these services,
the information sink (in Figure 1.5) is an appropriate storage medium.
The desirable characteristics of a storage medium include large storage capacity, high access and data-reading
speeds, portability, durability, low cost, data integrity, and reusability. Four types of storage media are most used
as information sinks in communication systems.
Magnetic tape: this is made of a thin coating of a film of magnetic metal or ferromagnetic particles (embedded
in a binding material) on a plastic strip of width 6–50 mm and length up to 1.5 km. The overall thickness of the
tape (magnetic coating and substrate strip) is about 15–60 μm. The tape is contained in a two-reel cassette and is
read from and written to in a tape drive or deck. The drive winds the tape from one reel to the other, causing it
to move past a read/write head. Magnetic tapes offer large storage capacity and come in a variety of incompatible
formats. The leading brand, known as Linear Tape-Open (LTO) Ultrium, offers a capacity of up to 30 terabytes
(TB) when data are compressed. Other advantageous features of the magnetic tape include that they are reusable
and easy to edit, compact, and portable. However, they have several demerits, which limit their use mostly to data
archiving and hard disk backup. They are easily damaged through breaking or deterioration if exposed to heat or
humidity. Recorded information degrades with continuous use, due to tape wear. Information can also be ruined
by strong magnetic interference. Furthermore, access to data on the tape is slow and sequential.
Magnetic disk: magnetic disks may be fixed or removable. A flat-disk substrate is coated with a thin film of
magnetic material. The disk is either a floppy disk, with a flexible substrate, or a hard disk, also known as a hard
disk drive (HDD), with a rigid substrate. The surface of the disk is divided into several evenly spaced concentric
tracks and each track is divided into sectors. Data are stored in these sectors. During a read/write operation, the
disk is rotated at constant speed and the read/write head is moved in a radial direction to position it at the desired
track. This radial movement is referred to as seeking. Floppy disks are magnetically coated on both sides and
contained within a plastic with a spring-loaded door that is forced open when the disk is inserted in its drive. An
HDD, including the disk and read/write head, is hermetically sealed to prevent contamination by dust particles
and the head glides on a thin film of air without making actual contact with the disk. A PC hard drive usually
consists of several disks (with a head for each disk) mounted on the same rotating axle. In this case, the set of
tracks at the same radius on all the disks is called a cylinder. HDDs have two standardised form factors. The 3.5 in.
disk has dimensions (length × width × height) = 146.99 mm × 101.85 mm × 26.11 mm, whereas the 2.5 in. disk
Telegram: @Persian_IEEEComSoc

has standard length 100 mm and width 69.85 mm but they have been produced in a variety of thicknesses ranging
from 5 to 19 mm. HDD storage capacities have grown significantly over the years and is currently as high as 14 TB
and 5 TB, respectively, for the 3.5 in. and 2.5 in. drives.
Magnetic disks have several advantages. Access time is less than 28 ms in high-speed drives. Random data access
is possible using an index or directory (stored at a specific location on the disk) that lists the physical location of all
data objects. Magnetic disks are reusable, cheap, and easy to write to, read from, or edit. The removable hard disks
and floppy disks are also portable. However, the magnetic disks can fail, data on the disk becoming unreadable if
certain critical sectors are damaged or otherwise corrupted.
Floppy disk drives are much slower than hard drives and have very limited storage. Floppy disks came in two
standard sizes of 5.25 in. (with an initial storage capacity of 360 kB, later improved to 1.2 MB) and 3.5 in. (with
a storage capacity of 1.44 MB). They were extremely popular and ubiquitous as a portable storage medium for
 1.4 Communication System Elements 29

computer files in the 1980s and 1990s, but at the turn of the century they were superseded by more rugged and
higher-capacity portable storage media such as the compact disk and the USB memory stick, and from around
2006 computers were no longer manufactured with floppy disk drives. To find a higher-capacity and compatible
replacement for the floppy disk, efforts were made in the 1990s to introduce the 3.5 in. floptical disk, which used an
optical tracking mechanism to improve head positioning and with a data storage capacity of up to 120 MB. These
efforts ultimately failed in the market due primarily to the success of the optical disk.
Optical disk: optical disks include CD-R, magneto-optical, DVD-R, and DVD-RAM discussed below. The
compact disk (CD) is based on a 12 cm plastic disk that stores data in the variation of the reflectivity of densely
packed spots on the disk surface. The disk is read by bouncing a laser beam off the surface. Both the compact
disk recordable (CD-R) and the digital versatile disk recordable (DVD-R) are write-once, read-many. On the other
hand, the magneto-optical disk and the DVD random-access memory (DVD-RAM) can be erased and rewritten.
CDs and DVDs have numerous advantages. They provide a very large storage capacity, 650 MB for a CD and
up to 17 GB for a double-sided, dual-layer DVD. Data stored on a CD-R cannot be altered later, which ensures
the integrity of archived information. Optical disks are very durable as the reading process is without physical
contact with the disk and so does not cause any mechanical wear. Stored information can be read countless times
with practically no degradation. There is ample provision for error correction and the error rate is below 10−15 .
The disk is cheap, portable, compact, and not easily damaged, and the recording technique is immune to electric
and magnetic fields. The main disadvantages of optical disks are that most of them, except magneto-optical
disks and DVD-RAM, are not reusable. Once data have been written, they cannot be changed. Secondly,
unlike magnetic disks that use the same head for reading and writing, a special disk writer is required to write
information onto an optical disk. Thirdly, CD access times are about 10–20 times longer than that of magnetic
hard disks.
Solid state disk (SSD): an SSD, also called a flash drive, stores data in non-volatile semiconductor memory
cells made from metal-oxide-semiconductor field-effect-transistor (MOSFET) and fabricated as integrated circuit
(IC) chips. Unlike magnetic hard disks, they have no moving parts, run silently, and have much quicker access
times. The 2.5 in. SSD has storage capacity which may be as high as 4 TB. Since around 2006, SSDs have grown in
popularity as the preferred storage medium in high-end notebooks, laptops, and PCs.

1.4.3 Transmitter
The transmitter is essentially a signal processor. Its primary role is to transform the message signal into a form
that best suits the transmission medium, complies with any regulations governing the communication service,
and meets system design objectives. How the transmitter does this depends on the type of information source,
transmission medium, and communication system. For example, if the transmission medium is radio then the
signal processing tasks of the transmitter would necessarily include modulation and radiation. The transmitter
would process the signal in such a way as to ensure that it is transmitted in the authorised frequency band and
Telegram: @Persian_IEEEComSoc

at the permitted transmitted power level. If the main design objective is to minimise the cost of receivers then a
simple AM would be preferred to a more complex digital signal processing technology.
Subject to the above considerations, the signal processing performed by the transmitter may include one or
more of the following processes. A detailed discussion of most of these processes can be found in later chapters of
the book.
Source coding: also called source encoding, this deals with the efficient representation of the message signal.
Several processes may be involved, depending on the type of communication system. The following are examples
of source coding.

● Lowpass filtering to limit the bandwidth of the message signal.

● Multiplexing to combine several message signals for simultaneous transmission.
 30 1 Overview of Communication Systems

● Message formatting to represent the message in digital format. This includes character coding of text and graphics
using a suitable character code, such as Unicode, and analogue to digital conversion (ADC) of an analogue
message signal, such as voice and video.
● Data compaction to minimise the number of symbols or bits that represents the message. Examples of data
compaction include variable-length codes (e.g. Morse code and Huffman code), which assign shorter codewords
to more frequent source symbols, and the Lempel–Ziv algorithm, which uses a codebook built from the message.
● Lossy data compression to reduce the number of bits used for representing the message by removing some details
considered insignificant because a human observer will not perceive their absence. Examples of lossy data com-
pression include predictive coding (e.g. differential PCM and delta modulation), low bit rate speech coding,
transform coding, motion compensation, subsampling, colour table, and truncation.
Encryption: the message – in this context referred to as plaintext – is processed at the transmitter and converted
into a ciphertext, which is disguised in some way to ensure security. The process may be accomplished by scram-
bling the bits of the plaintext using an encryption key. Only authorised receivers will have the correct decryption
key with which to decipher or decrypt the ciphertext back to its original plaintext. This process is illustrated in
Figure 1.10.
Channel coding: also called channel encoding, this process is necessary to ensure that the message signal is
compatible with the transmission medium (or channel). It attempts to protect the message signal against chan-
nel distortion and transmission errors. It may add some redundancy to the transmitted symbols. Examples of
channel coding include pre-emphasis/de-emphasis in analogue systems (e.g. FM), line coding in digital baseband
systems, and provision for error detection and correction at the receiver. In fact, the process of carrier modulation
introduced next is also a form of channel coding, which is required for some transmission media such as radio.
Channel coding is, however, often defined very narrowly as just error control coding, a simple example of which
is single parity check in which the message bits are taken k bits at a time and one extra bit (called parity bit) is
added to produce a codeword of k + 1 data bits having an even number of binary 1 (for even parity check) or an
odd number of binary 1 (for odd parity check). With this scheme, the receiver can detect the occurrence of a single
bit error or any odd number of bit errors in a codeword. However, if two or more even numbers of bit errors occur
in a codeword, this will go undetected. Nevertheless, this simple scheme does lead to a significant reduction in the
number of undetected errors. For example, if a system incurs on average one character error per transmitted page

Encryption key Figure 1.10 Encryption at the transmitter and decryption at the
receiver.

Encryption
Telegram: @Persian_IEEEComSoc

Ciphertext
Plaintext

Decryption

Decryption key
 1.4 Communication System Elements 31

Location S Location R

Location S Location R

(a) Transmitter Receiver

(d)

(e)
(b) Transceiver Transceiver

transmitting receiving OR
OR receiving transmitting

(f)
Transmitter Receiver
(c)
Receiver Transmitter

transmitting receiving AND

AND receiving transmitting

Figure 1.11 Communication systems: (a) Simplex; (b) Half-duplex; (c) Full duplex; (d) Point-to-point;
(e) Point-to-multipoint; (f) Multipoint-to-multipoint.

of text, the introduction of single parity check with k = 7 leads to the number of undetected errors being reduced
to one character in about 2500 pages of text on average.
A large variety of error control coding schemes have been devised over the years with capabilities that far exceed
those of the single parity check, allowing not only the detection of a wide pattern of bit errors but also the reliable
correction of one or more bit errors in a codeword.
Carrier modulation: this process translates the message signal frequencies from baseband into a frequency
band that is suitable for and allows efficient exploitation of the transmission medium.
Spread spectrum modulation: this involves the transmitted signal being deliberately spread out over a
wide frequency band. This technique was previously designed for military communications to protect against
frequency-selective fading, interference, and intentional jamming; but it is now also employed as a multiplexing
and multiple access technique in non-military communications.
Radiation: in a wireless communication system, the message signal must be processed and radiated as electro-
Telegram: @Persian_IEEEComSoc

magnetic waves in the desired direction. The radiating element required depends on the type of wireless system. In
general, antennas of various designs are used for radio communication systems and light-emitting diodes or laser
diodes are employed in infrared and optical communication systems.

1.4.4 Receiver
A receiver is required at the other end of the transmission medium (i.e. destination of the transmitted signal) to
process the received signal to obtain a close estimate of the message signal and deliver this to the information
sink. Note that the receiver’s input comes from the original message signal after it has been processed by two
 32 1 Overview of Communication Systems

blocks of the communication system, purposely in the transmitter and undesirably in the transmission medium.
The receiver undoes the processing of the transmitter in a series of steps performed in reverse order to that of the
transmitter. This may include one or more of the following tasks.
● Radio reception using a receive-antenna to convert the received electromagnetic waves back to a voltage signal.
● Spread spectrum demodulation to remove the carrier spreading.
● Carrier demodulation to translate the signal back to baseband frequencies.
● Clock extraction and synchronisation to recover the clock signal (if any) used at the transmitter in order to use
the same timing intervals for operations at the transmitter and the receiver.
● Channel decoding to correct errors, remove redundancy, etc.
● Decryption to recover the plain and undisguised symbol sequence.
● Source decoding to recover the original message signal from the symbol sequence. This may involve demultiplex-
ing – which breaks up a composite (multiplexed) signal into the components belonging to each of the multiple
users – digital-to-analogue conversion (DAC), lowpass filtering, etc.

1.5 Classification of Communication Systems

There are four major ways of classifying the wide variety of communication systems in operation today. If we con-
sider the direction of transmission then the communication system will be either simplex or duplex. If we consider
the type of signal transmitted by the system then we have either an analogue communication system or a digital
communication system. Comparing the frequency content of the transmitted signal to that of the original signal
leads to a classification of the system as either a baseband communication system or a modulated (or bandpass) com-
munication system. Finally, in terms of how a transmission path is established to convey the signal from source to
destination, we may have either a circuit-switched or packet-switched system.

1.5.1 Simplex Versus Duplex Communication Systems

If information flows in only one direction, the communication system is referred to as simplex (SX). Typically,
information originates from one transmitter and has one (or more) receiver as its destination. These receivers do
not have the capability of responding through the same communication link.
By the above explanation, interactive digital television broadcast is a simplex communication system. Although a
customer can respond (via a telephone link), information transmissions from the television house to the customer
and from the customer to the television house are carried out on two different links. Radar, a system that measures
electromagnetic signals reflected off objects (such as ship, aircraft, missile, storm, hydrometeor, etc.) in order to
determine their range and velocity, is also a simplex system. In this case, although the signal may be reflected by
the object back towards the transmitter S, giving an appearance of bi-directional information flow, there is strictly
Telegram: @Persian_IEEEComSoc

speaking only one transmitter, which is co-located with one receiver. The reflecting object is neither a transmitter
nor a receiver.
Other examples of simplex systems include audio broadcast (AM radio, FM radio, music services, and digital
radio), television broadcast (satellite, cable, and terrestrial), paging services, telemetry, and remote control.
When information can flow in both directions on the same link, the communication system is referred to as
duplex. The communication equipment at both locations S and R is equipped with transmission and reception
capabilities and is therefore referred to as a transceiver – a name that is a portmanteau of the words transmitter
and receiver. The system may be designed in such a way that simultaneous communication in both directions is
possible. This is a full duplex (FDX) system and requires a separate channel (e.g. a different band of frequencies,
a different time slot, or a different wire pair) being allocated for each direction of communication. If, on the other
hand, information can only flow in one direction at a time then the system is referred to as half -duplex (HDX).
 1.5 Classification of Communication Systems 33

The most common example of a FDX system is public telephony (both fixed and mobile). Broadband
communication via DSL in the PSTN is also FDX, as is computer interconnection in local area networks (LANs).
An example of an HDX system is the walkie-talkie used for wireless voice communication between two locations.
Transmission in either direction uses the same radio frequency band. The handset at each location can be
switched between transmit and receive modes, so that at any given time one location transmits while the other
receives.
In both simplex and duplex systems, if communication takes place between only two transceivers then the sys-
tem may be further described as a point-to-point communication system. If there is one transmitter or transceiver
communicating with several receivers or transceivers, we have a point-to-multipoint system. If there are many
intercommunicating transceivers (as in a LAN, or a video conference system linking more than two locations)
then we have what is called a multipoint-to-multipoint communication system. In this last case, information flow
between two transceivers or nodes is essentially bi-directional and therefore a simplex multipoint-to-multipoint
system is not possible. The Internet is a multipoint-to-multipoint communication system. A radio or television
broadcast system is a good example of a point-to-multipoint simplex system. See Figure 1.11.
An important limitation of simplex systems is that there is no return path for the receiver to automatically
request re-transmission in the event of an error. Thus, there are two options to deal with the problem of errors: (i)
ignore them in noncritical systems or (ii) include sufficient redundancy in the transmitted signal so that, in the
event of an error, the receiver can make a reliable guess of the transmitted information. The guess may occasionally
be wrong. This system of error correction is called forward error correction (FEC). Duplex systems are not under
this limitation and can also use the more reliable technique of error correction called automatic repeat request
(ARQ), in which the receiver automatically requests retransmission once an error is detected.

1.5.2 Analogue Versus Digital Communication Systems

An important classification of a communication system, from the point of view of the technology involved in
its implementation, is based on the type of information signals conveyed, whether analogue or digital. Analogue
communication systems convey analogue signals, while digital systems convey digital signals. However, note that
a digital communication system may employ an analogue waveform called a carrier signal to convey the digital
signal, as in the use of modems for data transmission over public telephone networks, which were originally
designed for analogue speech signals. We will defer an introduction to analogue, digital, and other types of signals
until the next chapter and focus here on discussing the advantages of digital over analogue communication, which
are so great that even analogue signals such as speech and video are transmitted digitally by first converting them
to a digital format at the transmitter. This analogue-to-digital conversion (ADC) at the transmitter is followed at
the receiver by the reverse process of digital-to-analogue conversion (DAC).
The advantages of digital communication over its analogue counterpart include:
● Low cost: inexpensive digital circuits may be used to implement the signal processing tasks required in digital
Telegram: @Persian_IEEEComSoc

communication. With continuing advances in semiconductor technology, the cost of VLSI circuits will drop even
further, making digital communication systems cheaper than their analogue counterparts despite the simplicity
of the latter. Cost is an important factor that determines whether a new technology is successfully assimilated
into society. The digital revolution of the 1990s, which we referred to earlier, was driven to a large extent by the
falling costs of digital circuits of increasing computational power.
● Privacy and security: increased reliance on telecommunication systems for private, business, and military com-
munications and the sale of entertainment and information services calls for secrecy, authenticity, and integrity.
The first requirement ensures that the information is received only by an authorised user, whereas the last two
requirements assure the receiver that there has not been any impersonation of the sender and that the informa-
tion has not been deliberately or accidentally altered in transit. Digital communication permits data encryption
to be easily implemented on the information bit stream in order to satisfy these requirements.
 34 1 Overview of Communication Systems

● Dynamic range: the dynamic range of a communication system refers to the amplitude ratio between the
strongest and weakest signals that the system can process with an acceptably low level of impairment. Signals
of a wider range of values (from very small to very large) than is possible with analogue systems can be accu-
rately represented and transmitted with negligible distortion in digital communication systems. The dynamic
range may be increased as much as desired by increasing the number of bits used to represent each sample of
the analogue signal during the ADC process. The penalty, of course, is increased bandwidth requirements.
● Noise immunity: the number of errors in the received data, or the bit error ratio (BER), may be very small even in
the presence of a significant amount of noise in the received signal. Although the precise value of the received
signal will be changed by additive noise, the change will only rarely be large enough to force the signal value
beyond the range that represents the transmitted bits.
● Regeneration: digital communication systems allow the possibility of regenerating (at sufficiently closely spaced
regenerative repeaters) clean new symbols or pulses, which are free from all impairment effects and are (ide-
ally) an exact replica of the original transmission. Thus, unlike analogue systems, noise does not accumulate
from repeater to repeater and no further signal distortion occurs beyond that which was introduced at the
analogue-to-digital conversion stage. Digital signals may also be stored in various storage media (e.g. optical
or magnetic disks) and processed or re-transmitted later without loss of fidelity.
● Error correction: it is possible to detect and even correct errors in the received data using various coding tech-
niques, which generally insert some redundancy in the transmitted data.
● Flexibility: the signal processing tasks of a digital communication system may be readily reconfigured simply
by changing the software program, without any need to change the hardware. Modification of system functions
can therefore be implemented more cheaply and speedily.
● Integrated services: voice, video, and data can all be represented in a common bit stream format and transmitted
simultaneously in a common communication system. Multimedia communication, the Internet, and a host of
other modern communication services are only feasible through digital technology.
Digital communication also has several disadvantages. However, in most cases, these disadvantages are under
the system designer’s control, and their effects may be reduced as much as desired by making a suitable trade-off.
It should therefore be noted that the following disadvantages are far outweighed by the advantages discussed
above, and this accounts for the transformation of global telecommunication into an all-digital network at the
turn of the century.
● Large bandwidth: digital communication systems may require more bandwidth than analogue systems. For
example, a 4 kHz bandwidth is adequate for analogue speech transmission, whereas digital speech transmis-
sion using standard (64 kb/s) PCM requires a minimum bandwidth of 32 kHz. The spectrum of all transmission
media, especially radio, is limited. Transmission techniques that minimise the required bandwidth are therefore
preferred in order to increase the number of services, users, and bit rate per user that can be accommodated. Var-
ious low bit rate speech coding and data compression techniques have been devised to reduce the bandwidth
Telegram: @Persian_IEEEComSoc

requirements of digital audio and video transmission and storage at the price of a somewhat reduced signal
quality.
● Complexity: digital communication systems generally perform more complex processing operations on the input
signal and require more sophisticated circuitry. Synchronisation usually must be maintained between receiver
and transmitter. However, advances in semiconductor technology make circuit complexity a less significant dis-
advantage. Most of the signal processing tasks may be performed in a single highly reliable and affordable VLSI
unit, which can be easily replaced in the unlikely event of a malfunction. Furthermore, some digital transmission
techniques, such as the asynchronous transfer mode (ATM) – discussed in Chapter 13 – make synchronisation
a less challenging issue.
● Quantisation distortion: analogue signals such as speech must be converted to digital form prior to transmission
or processing in a digital communication system. This conversion introduces an irreversible quantisation dis-
tortion. However, this distortion may be made as small as the system designer wishes by increasing the number
of quantisation levels. The price for this improvement is increased bandwidth requirements.
 1.5 Classification of Communication Systems 35

1.5.3 Baseband Versus Modulated Communication Systems

Communication systems may also be classified as either baseband or modulated, depending on how the informa-
tion signal is prepared for the transmission medium. When the transmission medium can pass signals of frequency
around 0 Hz and upwards, it is described as a lowpass medium. The original information signal (called the baseband
signal since it usually contains frequencies near 0 Hz) may be conveyed to the receiver by being placed directly
into the lowpass medium without any frequency translation. Such a communication system is referred to as a
baseband communication system.
The only practical lowpass media are wire pair and coaxial cable. Thus, baseband systems are always wireline
connected systems. Although the radio spectrum stretches all the way down to frequencies near 0 Hz and it is
theoretically possible to convey a baseband signal (e.g. speech signal) by radio, this is not feasible in practice.
Because it contains low-frequency components, the baseband signal cannot be efficiently placed into the radio
medium at the transmitter or extracted from it at the receiver. We will have more to say on this later.

1.5.3.1 Analogue Baseband Communication System

The simplest example of baseband transmission is the voice intercom system. The baseband voice output signal
of a microphone is conducted along a wire pair to a loudspeaker at some distance. Other examples of analogue
baseband communication systems include the fixed (i.e. nonmobile) telephone connection between two local sub-
scriber terminals in the PSTN and a closed-circuit television (CCTV) system.
It could be argued that the communication system block diagram of Figure 1.5 does not apply in this case, since
there is neither transmitter nor receiver and the transmission medium links the message signal directly from
information source to information sink. However, in most cases a transmitter is incorporated that performs very
basic signal processing such as frequency-dependent amplification to compensate for attenuation and distortion in
the transmission medium. The identifying feature of a baseband system is that any type of frequency translation is
specifically excluded. The system usually includes a receiver that performs lowpass filtering to minimise the effect
of noise, and amplification to provide for output volume control.
A block diagram of a CCTV system is shown in Figure 1.12. A lens in the video camera forms an image of the
scene on a light-sensitive camera sensor, which converts the light intensity into an electrical signal called the
video signal. This has a peak-to-peak value of 1 V and contains frequencies in the range 0–10 MHz, depending
on the horizontal resolution of the camera. It is obvious that the video signal is a baseband signal. For a large
distance between camera and console sites, a video amplifier is inserted to boost the video signal level. If there
is only one camera then its output is connected directly via coaxial cable to a monitor. Otherwise, the outputs
of several cameras are connected to a video switcher. This is an electronic device that selects different cameras
automatically or manually for display on one or more monitors. The monitor is like a television set but has no
tuning/demodulation circuits and therefore cannot receive a television broadcast off air. The monitor is simply
a CRT or other type of visual display device that converts the baseband video signal into a visible display on its
Telegram: @Persian_IEEEComSoc

screen. A VCR may also be used for a permanent record of the video signal, and a printer to obtain a hard copy of
a selected scene.
There is no frequency translation in this CCTV system, and therefore it is a baseband system. There are some
implementations of a CCTV system that use radio, infrared, or optical fibre connection between camera and mon-
itor. Such systems must necessarily implement some form of carrier modulation and therefore are not baseband
systems.
One of the main difficulties with analogue baseband transmission is that a separate transmission medium (wire
pair or coaxial cable) is required for each signal. One medium cannot be shared simultaneously by multiple users
without interference, since each user’s signal is continuous in time and occupies (or at least partly overlaps) the
same frequency band. The cost would be prohibitive to provide a separate wire pair for all anticipated simultaneous
telephone calls between two exchanges in the PSTN. If the analogue signal is transformed into a discrete-time func-
tion then an analogue baseband transmission system can be implemented that has the capability to accommodate
multiple users on one link.
 36 1 Overview of Communication Systems

Camera sites Console site

Video camera Video

(Location 1) amplifier
Monitor 1

Video camera Video Video

(Location 2) amplifier switcher
Monitor 2

Video
Printer

Video camera Video

(Location N) amplifier
VCR

Figure 1.12 Closed-circuit television system (CCTV). Example of analogue baseband transmission system.

1.5.3.2 Discrete Baseband Communication System

If we take regular samples of an analogue signal v(t) at intervals of T s to obtain a sequence of continuous-value
samples v(nT s ) then we can perfectly reconstruct the original signal v(t) from the samples v(nT s ) by passing these
samples through a suitable lowpass filter (LPF), provided that T s is no more than half the period of the highest
frequency component of v(t). This is a statement of the sampling theorem. A discrete baseband communication
system transmits v(t) by sending one voltage pulse at the instance of each sample v(nT s ). The value of each sample
may be conveyed in the amplitude, duration, or position of this pulse.
PAM, PWM, and PPM: if the pulses are sent with equal width at regular intervals T s but the height of the
nth pulse (occurring at time t = nT s ) is varied in proportion to the sample value v(nT s ) then we have what is
referred to as pulse amplitude modulation (PAM). If the pulses are sent with equal height at regular intervals T s
but the duration or width of the nth pulse is varied in proportion to v(nT s ), we refer to this as pulse duration
modulation (PDM), also called pulse width modulation (PWM). Finally, sending the pulses with equal height and
Telegram: @Persian_IEEEComSoc

width but at irregular intervals such that the time of occurrence or position of the nth pulse is delayed relative to
the nth sampling instant by an amount that is proportional to v(nT s ) gives rise to pulse position modulation (PPM).
Figure 1.13 illustrates these waveforms. In Figure 1.13d, the sampling instants are indicated by thin vertical lines.
Note from this figure that the longest delay 𝜏 max occurs at the third pulse where the sample value is maximum. The
first pulse corresponds to the smallest sample value and has the shortest delay 𝜏 min . The remaining pulses have
delays ranging between 𝜏 min and 𝜏 max . In this way, information regarding the sample values is correctly conveyed
by the positions of transmitted pulses.
The block diagram of a PAM generator and receiver is shown in Figure 1.14 and a suitable arrangement for
generating PDM and PPM waveforms is given in Figure 1.15. Signal waveforms have been sketched at various
points of these block diagrams in order to clarify the function of each element. The waveform of v(t) is as earlier
shown in Figure 1.13a and has been omitted. The effect of the transmission medium is also not shown. Note in
 1.5 Classification of Communication Systems 37

v(t)

(a)

vPAM(t)

(b)

t
Ts
vPDM(t)

(c)

vPPM(t)

(d)

t
τmin τmax

Figure 1.13 Analogue signal and its discrete representation as PAM, PDM, & PPM.
Telegram: @Persian_IEEEComSoc

v(t) Sample vPAM(t) vPAM(t) Lowpass v(t)

Transmission medium
and hold filter

Clock
(Ts)

Figure 1.14 Block diagram of a PAM generator and receiver.

 38 1 Overview of Communication Systems

vPAM(t)
v(t)

Clock Sample
(Ts) Ts
and hold

vPAM(t)
vtrn(t)

Triangle
waveform
generator

vtrn(t)
vtPAM(t)

Vr
Σ

vtPAM(t)
Vr
vPDM(t)
– +
Comparator

vPDM(t)

vPPM(t)

Monostable
multivibrator

vPPM(t)

Figure 1.15 Block diagram of PDM and PPM generator.

Figure 1.14 the simplicity of the discrete baseband system, especially the PAM receiver, which is just an LPF.
This filter, often referred to as a reconstruction filter, may have a frequency response that is shaped in such a way
as to correct for a small distortion due to the action of the sample-and-hold circuit. PWM and PPM signals can
be received (i.e. converted back to the original analogue signal) by using an integrator to convert pulse width or
Telegram: @Persian_IEEEComSoc

pulse position to voltage level. This process essentially converts PWM and PPM to a PAM waveform, which is then
processed in an LPF to recover the original signal.
The following salient features of a discrete baseband system should be noted.
● Although it is sometimes erroneously viewed as a form of modulation employing a pulse train as the carrier,
a discrete baseband communication system is actually a baseband transmission because the spectra of PAM,
PDM, and PPM signals contain frequencies down to 0 Hz. In fact, the spectrum of an instantaneously sampled
PAM signal is the spectrum of the original (baseband) signal plus exact duplications at regular intervals along
the frequency axis.
● A discrete baseband system is an analogue communication system since the parameter of the pulse train that is
varied may take on a continuum of values in a specified range. For example, the precise value of the pulse height
is significant in a PAM system and any variation due to noise will distort the received signal.
 1.5 Classification of Communication Systems 39

● The bandwidth required to transmit a discrete baseband signal (PAM, PDM, or PPM) far exceeds the bandwidth
of the original analogue signal. An instantaneously sampled PAM signal has infinite bandwidth. A discrete
baseband signal fills up all the bandwidth available on the transmission medium. To share the medium among
multiple signals, each must be sent in a separate time slot.
● PAM is very susceptible to noise. PDM and PPM have a better noise performance, like that of FM systems but
are inferior in this respect to digital baseband transmission systems. If the pulses were perfectly rectangular
then PDM and PPM would be completely immune to additive noise, as this would only alter the unused height
parameter, without affecting the zero crossings of the pulse which determine pulse width or pulse location.
Unfortunately, perfectly rectangular pulses, with zero rise time, are not only impossible to generate, they are
impossible to maintain in a lowpass transmission medium.
● PDM is wasteful of power compared to PPM. Long pulses in PDM expend more power but carry no additional
information. PPM on the other hand transmits pulses of equal energy.
● The main advantage of a discrete baseband system is that transmission medium sharing by multiple users,
known as multiplexing, is possible. The intervals between the samples of one signal are used to transmit samples
of other signals. This type of multiplexing is known as TDM and is further discussed below.
TDM: Figure 1.16 shows the block diagram of an N-channel TDM system that allows simultaneous transmission
of several independent PAM signals over a single transmission medium. The commutator is shown as a rotating
arm simply for the purpose of an easier illustration of the sampling and interleaving process. It is an electronic

Message inputs

LPF LPF LPF Band-limiting filters

ʋ1(t) ʋ2(t) ʋN(t)

Commutator

ʋTDM(nTs/N)
Transmission medium
synchronisation
Telegram: @Persian_IEEEComSoc

ʋTDM(nTs/N)

Decommutator
ʋN ((n + (N – 1)/N)Ts)
ʋ1 (nTs) ʋ2 ((n +1/N)Ts)
LPF LPF LPF Reconstruction filters

ʋ1 (t) ʋ2 (t) ʋN (t)

Message outputs

Figure 1.16 Block diagram of an analogue TDM system.

 40 1 Overview of Communication Systems

(a) ʋ1(0) ʋ2(Ts/3) ʋ3(2Ts/3) ʋ1(Ts) ʋ2(4Ts/3) ʋ3(5Ts/3)

time, nTs/3
One frame
Duration = Ts
ʋ1 (t)
ʋ1 ʋ1 (nTs)

(b)

ʋ2 (t)
ʋ2 ʋ2 (( n + 1 ) Ts)
3

(c)

ʋ3 ʋ3 (t)
ʋ3 (( n + 2
3 ) Ts )
(d)

ʋTDM (nTs/3)

(e)

0 Ts 2Ts 3T s 4Ts 5Ts 6Ts 7Ts

time

Figure 1.17 3-Channel TDM.

switching circuit that samples each input at a rate f s = 1/T s and interleaves the N samples inside the sampling
interval T s . LPFs remove insignificant high-frequency components of the input signals and limit their bandwidth
to at most f s /2. These bandlimited analogue waveforms can then be correctly reconstructed at the receiver by
Telegram: @Persian_IEEEComSoc

passing their respective sequence of samples through an LPF, as shown in Figure 1.16. The waveforms referred
to in Figure 1.16 are sketched in Figure 1.17 for N = 3. It must be emphasised that this is an analogue system. Its
main advantage of simplicity is outweighed by its susceptibility to noise and distortion and therefore the analogue
TDM system shown in Figures 1.16 and 1.17 is rarely used in practice. However, this system forms the basis for
the TDM of digital signals, which has now become a ubiquitous technique in telecommunication systems. Some
of the features of TDM, whether of PAM or digital signals, include:
● The bandwidth requirement of an N-channel TDM signal expands by a factor N, the number of multiplexed
signals or channels. This happens because N samples are squeezed into one sampling interval, reducing the
sampling pulse period by a factor N and hence increasing its frequency by the same factor. In practice, the
bandwidth increase will be slightly larger than the factor N since some time slots must be reserved for system
management and synchronisation.
 1.5 Classification of Communication Systems 41

● The transmitter and receiver must be synchronised in order that the interleaved samples in the TDM signal are
correctly distributed by the decommutator to their respective channels.
● TDM is sensitive to transmission medium dispersion, which arises because the transmission medium differently
attenuates or delays various frequency components of the transmitted pulse. As a result, the pulse may broaden
out sufficiently to overlap adjacent time slots, an undesirable situation known as intersymbol interference (ISI).
● TDM is, however, immune to system nonlinearity as a source of crosstalk between independent signals, since at
any given time instant only one signal is present. This feature is an important advantage that allows amplifiers
to be operated near their maximum rating, a typically nonlinear region.

1.5.3.3 Digital Baseband Communication System

Digital signals, originating from coded textual information or from digitised analogue signals, may also be trans-
mitted at baseband. This is a more common type of baseband transmission as it has all the advantages of the digital
communication systems discussed earlier. The baseband transmitter performs line coding on the input bit stream,
and the baseband receiver has the task of decoding the received (and channel-distorted) coded waveforms back
into the original bit stream with minimum error. Before looking at these two operations, let us first consider how
an analogue signal is converted into a digital bit stream.
Analogue-to-digital conversion (ADC): various techniques have been devised to convert analogue signals
to digital. The earliest and still widespread technique is pulse code modulation (PCM). Figure 1.18 shows four
signal processing steps involved in converting an analogue signal to a PCM signal. The LPF removes nonessential
frequency components in the message signal and limits the highest frequency component to f max . This serves to
reduce transmission bandwidth and to ensure that the sampling theorem is satisfied when the message signal
is sampled (in the sample-and-hold circuit block) at a manageable rate f s ≥ 2f max . Next, the continuous-value
sample sequence (PAM signal) is converted to a discrete-value sequence. This process is known as quantisation.
It introduces irrecoverable errors as each sample is approximated to the nearest of a set of quantisation levels.
In a practical system with many quantisation levels, the errors are referred to as quantisation noise since their
effect is like that of white noise. If quantisation levels are uniformly spaced, we refer to this as uniform or linear
quantisation and to the signal conversion process simply as uniform ADC. The term PCM is usually applied to
the case of nonuniform quantisation, which spaces the quantiser levels nonuniformly in such a way as to make
the signal-to-quantisation-noise ratio (SQNR) approximately constant over all analogue signal values. Finally, the
encoder converts each quantised signal level to a unique binary word of length k bits. Note that k bits can be used
to uniquely represent N = 2k quantisation levels. Taking f s samples of the signal per second and representing each
sample with k bits, the number of bits generated each second, or the bit rate, is given by
Bit rate = kf s bits per second (b∕s) (1.3)
In PCM telephony, f s = 8000 samples per second (or Hz) and k = 8 bits/sample, yielding a bit rate of 64 kb/s. In
certain bandwidth-limited communication services such as mobile communications, this PCM bit rate requirement
Telegram: @Persian_IEEEComSoc

is excessive.
Reducing the redundancy inherent in a PCM signal allows the bit rate and hence bandwidth and storage
requirements to be significantly reduced. In a technique called differential pulse code modulation (DPCM), it is
the difference e(nT s ) between the actual sample and a predicted value that is quantised and encoded, rather than

Analogue signal PAM signal Digital signal

Message Anti-alias Sample Encoding PCM signal

Quantisation
signal LPF and hold (Bit stream)

Figure 1.18 Digitisation of an analogue signal: PCM signal generation.

 42 1 Overview of Communication Systems

the sample itself. If the predictor is properly designed and an adequate sampling rate (f s = 1/T s ) is used then the
range of e(nT s ) will be very small, allowing fewer quantisation levels and hence a smaller k (bits/sample) to be
used to achieve a SQNR comparable to that of a PCM system. Assuming the same sampling rate as in PCM, we
see from Eq. (1.3) that the bit rate of a DPCM system will be lower than that of a PCM system of comparable
SQNR. The ITU-T has adopted for voice telephony a 32 kbit/s DPCM system, obtained by using only k = 4 bits
to code each sample taken at the rate f s = 8 kHz. A 64 kbit/s DPCM system has also been adopted for wideband
audio (of 7 kHz bandwidth). This uses k = 4 and f s = 16 kHz.
Further bit rate reduction can be achieved by using sophisticated data compression algorithms and, for the digi-
tisation of speech, a range of methods known as low bit rate speech coding. These techniques are inherently lossy.
They exploit the features of the message signal, and the characteristics of human hearing and vision, to eliminate
redundant as well as insignificant information and produce a modified signal of greatly reduced bit rate. After this
signal has been decompressed or otherwise processed at the receiver, a human observer finds it acceptably close
to the original in quality and information content.
Line coding: whatever its origin, whether from Unicode-coded textual information or from a digitised analogue
signal, we have a bit stream to be transmitted. A digital baseband transmitter chooses suitable voltage symbols (e.g.
rectangular or shaped pulses) to represent the string of 1’s and 0’s, a process known as line coding, and places these
symbols directly into a transmission line system. Figure 1.19 shows the line codes used in Europe for connections
between equipment (often within one exchange building) in the interfaces of the digital transmission hierarchy.
Line codes are designed to have certain desirable characteristics and to fulfil several important functions.
● The spectral characteristics of coded data must be matched to the characteristics or frequency response of the
transmission medium. A mismatch may result in significant distortion of the transmitted voltage pulses. In
particular, the code should have no DC offset. Line transmission systems are easier to design when different
parts of the system are capacitor or transformer coupled to separate their DC bias voltage levels. These coupling
elements pass higher-frequency (AC) voltages but block zero-frequency (DC) voltages. The coded data must
therefore be void of DC content to prevent droop and baseline wander, whereby the received waveform drifts
significantly relative to the decision threshold, which is 0 V for the case of a bipolar binary code. See Figure 1.20.
● The line code must combine data and timing information in one signal. It would be very expensive if a separate
wire pair or coaxial cable had to be employed to carry the timing information needed at the receiver for setting
decision or sampling instants. Furthermore, the line code must have a reasonable amount of clock content: the

Bit stream 1 0 1 0 0 0 0 0 1 1 0 0 0 0 1 1 0 0 0 0

+V
AMI 0
–V
Telegram: @Persian_IEEEComSoc

+V
CMI 0
–V

+V
HDB3 0
–V

Figure 1.19 Examples of Line Codes: Alternate mark inversion (AMI), coded mark inversion (CMI), and high density bipolar
with 3 zero maximum (HDB3).
 1.5 Classification of Communication Systems 43

g(t), volts

1 1 1 1 1 1 1 0 1 1 1 1 1 0 0 0 Bit stream

+A
Ideal waveform

Drooping waveform

Ideal baseline

–A
Wandering
baseline

Figure 1.20 Droop and baseline wander in a binary line code.

timing content of a code is the maximum number of symbols that can occur together without a level transition – a
small number indicating a high timing content. Ideally, there should be at least one transition in every symbol,
but the major penalty is an increased bandwidth requirement.
● Vulnerability of the data to noise and ISI must be minimised. Sudden changes in a signal imply high fre-
quencies in its spectrum. A rectangular pulse (with sharp transitions) transmitted through a lowpass trans-
mission medium will spread out, with a potential for ISI. Thus, pulse shaping is frequently employed to reduce
high-frequency components, which also reduces crosstalk since higher frequencies are more readily radiated.
Pulse shaping also reduces the bandwidth necessary to correctly transmit the coded waveforms. The larger this
bandwidth, the larger will be the amount of noise power that the receiver inevitably ‘admits’ in the process of
receiving the waveforms.
● The line code should allow some amount of error detection. This usually involves the use of redundancy in
which some codewords or symbol patterns are forbidden. A received codeword that violates the coding rule
in force would then indicate some error.
● The line code should maximise code efficiency to allow a lower symbol rate to be used for a given bit rate. In
long-distance cable systems, a lower symbol rate allows increased repeater spacing and reduces overall system
cost. It turns out, however, that codes of high efficiency may lack certain other desirable characteristics. The
code selected in practice will involve some compromise and will depend on the priorities of the system. Code
efficiency is the ratio of actual information content (or bits) per code symbol to potential information content per
Telegram: @Persian_IEEEComSoc

code symbol. Potential information content per code symbol being given by log2 (Code radix), where code radix is
the number of signalling levels or voltage levels used by the code symbols. For example, the potential information
content per code symbol of a binary code (radix = 2) is log2 (2) = 1 bit, that of a ternary code (radix = 3) is
log2 (3) = 1.585 bits, and that of a quaternary code (radix = 4) is log2 (4) = 2 bits, etc. Codes with higher radix can
therefore convey more information per symbol, but there is increased codec complexity and a higher probability
of error. Although multilevel codes (of radix ≥4) are very common in modulated communication systems to cope
with restricted bandwidth, only codes with radix ≤4 are employed in baseband systems.
One example of a quaternary code is the 2B1Q line code, which was adopted by ANSI in 1986 for use on basic
ISDN lines. It is also the line code used on DSL local loops. As the name suggests, the 2B1Q code represents two
binary digits using one quaternary symbol, i.e. one of four voltage levels. More specifically, the dibits 00, 01, 11,
and 10 are represented by the voltage levels −3 V, −V, +V and +3 V, respectively.
 44 1 Overview of Communication Systems

Transmitter Transmission Receiver

medium

Data in Line Transmission Equaliser Pulse Data out

coder line and LPF detector
+

Noise and
Information distortion Clock Information
source recovery sink

Figure 1.21 Digital baseband transmission system.

● Finally, the complexity of the encoder and decoder circuits (codec) should be kept to a minimum in order to
reduce costs. In general, line codes that can be implemented by simple codecs are used for short links, whereas
more efficient but complex and costly codecs are used for long-distance links because they can work with fewer
repeaters and hence reduce overall system cost.

Line decoding: the transmission medium distorts the transmitted pulses by adding noise and by differently
attenuating and delaying various frequency components of the pulses. A baseband receiver or repeater takes the
distorted pulses as input and produces the original clean pulses at its output. It does this through three ‘R’ opera-
tions. First, a reshaping circuit comprising an equaliser and an LPF is used to reshape the pulse and ensure that its
spectrum has a raised cosine shape. This operation is important to minimise ISI. Next, a retiming circuit recovers
the clock signal from the stream of reshaped pulses. Level transitions within the pulse stream carry the clock infor-
mation. Finally, a regenerating circuit detects the pulses at the sampling instants provided by the recovered clock
signal. Occasionally, an error will occur when a noise voltage causes the pulse amplitude to cross the detection
threshold. The frequency of occurrence of such errors, or BER, can be maintained at an acceptable level by ensur-
ing that the noisy pulses are detected before the ratio of pulse energy to noise power density falls below a specified
threshold. Figure 1.21 is a block diagram of a digital baseband system showing the basic operations discussed here.

1.5.3.4 Modulated Communication Systems

In very many situations it is necessary to translate the baseband signal (without distorting its information content)
to a frequency band centred at a frequency f c that is well above 0 Hz. A communication system in which the infor-
mation signal undergoes this modulation process before being placed into the transmission medium is referred
to as a modulated communication system. There are numerous examples of modulated systems. All satellite com-
munication systems, mobile communication systems, radio and TV broadcast systems, radio relay systems, and
optical communication systems are modulated systems.
Telegram: @Persian_IEEEComSoc

Formally, we define modulation as the process of imposing the variations (or information) in a lower-frequency
electrical signal (called the modulating or baseband or message signal) onto a higher frequency signal (called the
carrier). The carrier signal is usually a sinusoidal signal of frequency f c . It effectively gives the message signal a
‘ride’ through the transmission medium because, for several reasons, it is impossible or undesirable for the message
signal to make the ‘journey’ on its own.
Role of modulation: there are several reasons why modulation is extensively used in modern communication:

● Modulation is used to obtain a more efficient exploitation of the transmission medium by accommodating more
than one user in the same medium at the same time. In most cases, the bandwidth that is available in the medium
is much larger than what is required by one user or message signal. For example, the bandwidth available on
a coaxial cable is more than 10 000 times the bandwidth of one telephone speech channel. The bandwidth of
an optical fibre medium exceeds that of an analogue TV signal by a factor of up to one million; and the radio
 1.5 Classification of Communication Systems 45

spectrum is much wider than the bandwidth required by one radio station for its broadcasts. Modulation allows
the implementation of FDM in which each user’s signal is placed in a separate frequency band by modulating
an appropriate carrier. If the carrier frequencies are sufficiently far apart, the different signals do not interfere
with each other. A signal can be recovered at the receiver by filtering (to exclude the unwanted channels) and
demodulation (to detect the message signal in the carrier signal). Providers of radio services can transmit and
receive within the bands allocated to them by using a suitable modulation technique.
● Modulation allows us to select a frequency that is high enough to be efficiently radiated by an antenna in radio
systems. The power radiated by an antenna may be expressed as P = Irms 2 R , where I
r rms is the root mean square
(rms) value of the current signal fed into the antenna and Rr is the antenna’s radiation resistance. It turns out
that Rr depends on the size of the antenna measured in wavelength units. In general, the size of the antenna
must be at least one-tenth of the signal wavelength if the antenna is to radiate an appreciable amount of power.
Consider the minimum size of antennas required to radiate signals at three different frequencies, 3 kHz, 3 MHz,
and 3 GHz. The wavelengths of these signals (given by the ratio between the speed of light 3 × 108 m/s and the
signal’s frequency) are 100 km, 100 m, and 10 cm, respectively. Thus, if we attempted to radiate a 3 kHz speech
signal, we would need an antenna that is at least 10 km long. Not only is such an antenna prohibitively expensive,
it is hardly suited to portable applications such as in handheld mobile telephone units. If, on the other hand,
we use our 3 kHz speech signal to modulate a 3 GHz carrier signal then it can be efficiently radiated using very
small and hence affordable antennas of minimum size 1 cm.
● The use of modulation to transmit at higher frequencies also provides a further important advantage. It allows
us to exploit the higher bandwidths available at the top end of the radio spectrum in order to accommodate more
users or to transmit signals of large bandwidth. For example, an AM radio signal has a bandwidth of 10 kHz.
Thus, the maximum number of AM radio stations that can be operated at low frequency (LF ≡30–300 kHz) in
one locality is given by
300 − 30
Maximum number of AM stations at LF = = 27
10
Observe that there is a tenfold increase in the number of AM radio stations when we move up by just one band
to medium frequency (MF ≡ 300−3000 kHz)
3000 − 300
Maximum number of AM stations at MF = = 270
10
The NTSC television signal requires a radio frequency bandwidth of 6 MHz. Frequency bands at MF and below
can therefore not be used for TV transmission because they do not have enough bandwidth to accommodate
the signal. The high frequency (HF ≡ 3−30 MHz) band can accommodate a maximum of only four such TV
channels. However, as we move to higher bands, we can accommodate 45 TV channels at very high frequency
(VHF ≡ 30−300 MHz), 450 TV channels at ultra high frequency (UHF ≡ 300−3000 MHz), 4500 TV channels at
super high frequency (SHF ≡ 3−30 GHz), 45000 TV channels at extra high frequency (EHF ≡ 30−300 GHz); and
the optical frequency band (300 GHz − 860 THz) can accommodate literally millions of TV channels.
Telegram: @Persian_IEEEComSoc

● Another important function of modulation is that it allows us to transmit at a frequency that is best suited to
the transmission medium. The behaviour of all practical transmission media is frequency dependent. Some fre-
quency bands are passed with minimum distortion, some are heavily distorted, and some are blocked altogether.
Modulation provides us with the means of placing the signal within a band of frequencies where noise, signal
distortion, and attenuation are at an acceptable level within the transmission medium. Satellite communication
was pioneered in C-band inside the 1–10 GHz window where both noise (celestial and atmospheric) and prop-
agation impairments are minimum. Ionospheric reflection and absorption become increasingly significant the
lower you go below this band, until at about 12 MHz the signal is completely blocked by the ionosphere. Further-
more, attenuation by tropospheric constituents such as rain, atmospheric gases, fog, cloud water droplets, etc.,
becomes significant and eventually very severe at higher-frequency bands. For this reason, modulation must be
used in satellite communication to translate the baseband signal to a congenial higher-frequency band.
 46 1 Overview of Communication Systems

● Another example of a bandpass transmission medium is the optical fibre medium. This blocks signals at radio
wave frequencies, but passes signals in the near-infrared band, particularly the frequencies around 194 and
231 THz. Thus, the only way to use this valuable medium for information transmission is to modulate an optical
carrier signal with the baseband information signal.

Types of modulation: there are three basic methods of modulation depending on which parameter of the
carrier signal is varied (or modulated) by the message signal. Consider the general expression for a sinusoidal
carrier signal

vc (t) = Ac cos(2𝜋fc t + 𝜙) (1.4)

There are three parameters of the carrier, which may be varied in step with the message signal. The (unmod-
ulated) carrier signal vc (t) has a constant amplitude Ac , a constant frequency f c , and a constant initial phase 𝜙.
Varying the amplitude according to the variations of the message signal, while maintaining the other parameters
constant, gives what is known as amplitude modulation (AM). Frequency modulation is obtained by varying the
frequency f c of a constant-amplitude carrier, and phase modulation is the result of varying only the phase 𝜙 of the
carrier signal.
An analogue modulating signal will cause a continuous variation of the carrier parameter, the precise value
of the varied parameter being significant always. It is obvious that this is then an analogue modulation and the
resulting system is an analogue modulated communication system.
A digital modulating signal (consisting of a string of binary 1’s and 0’s) will, on the other hand, cause the carrier
parameter to change (or shift) in discrete steps. Information is then conveyed, not in the continuous precise value
of the parameter but rather in the interval of the parameter value at discrete decision (or sampling) instants. This
is therefore digital modulation and the resulting system is a digital modulated communication system having all
the advantages of digital communication. In this case, the three modulation methods are given the special names
amplitude shift keying (ASK), frequency shift keying (FSK), and phase shift keying (PSK) to emphasise that the
parameters are varied in discrete steps. The number of steps of the parameter generally determines the complexity
of the digital modulation scheme. The simplest situation, and the one most robust to noise, is binary modulation
(or binary shift keying) where the carrier parameter can take on one of two values (or steps). This is shown in
Figure 1.22.
To transmit a bit stream using binary ASK, we take one bit at each clock instant and transmit a sinusoidal carrier
of frequency f c for the duration of the bit. The carrier frequency f c is chosen to suit the transmission medium. The
carrier amplitude is set to A1 for bit ‘1’ and to A0 for bit ‘0’. If either A1 or A0 is zero, we have a special type of binary
ASK known as on–off keying (OOK), which is the digital modulation scheme used in optical fibre communication.
In binary FSK, same-amplitude and same-phase carriers of frequencies f 1 and f 0 are transmitted for bit 1 and bit
0, respectively. In binary PSK, the carrier amplitude and frequency are fixed, but the carrier is transmitted with
a phase 𝜙1 for bit 1 and 𝜙0 for bit 0. Usually, the difference between the two phases is 180∘ in order to use the
Telegram: @Persian_IEEEComSoc

maximum possible spacing between them.

A useful way to look at digital modulation is to treat each transmitted carrier state as a code symbol. Thus, OOK
represents bits 1 and 0 with the symbols SOOK1 and SOOK0 shown in Figure 1.22d; FSK uses the symbols SFSK1 and
SFSK0 shown in Figure 1.22e; and PSK uses the symbols SPSK1 and SPSK0 shown in Figure 1.22f. The symbol rate (or
signalling rate) is limited by the transmission bandwidth available. The theoretical maximum symbol rate is twice
the transmission bandwidth. In a telephone channel with a bandwidth of about 3.1 kHz, the maximum possible
symbol rate is therefore 6200 Bd.
The main drawback of binary shift keying is that bit rate and symbol rate are equal since each symbol represents
one bit. In many applications, a much higher bit rate is desired than can be obtained by increasing the symbol
rate, which is limited by the available bandwidth. A higher bit rate can be achieved by using multilevel (also
called M-ary, pronounced em-aary) digital modulation, in which M distinct carrier states are transmitted. Binary
 1.5 Classification of Communication Systems 47

Bit Stream 0 1 1 0
Voltage

t
(a) (d) SOOK1 SOOK0

Voltage
(e) SFSK1 SFSK0
t
(b)

Voltage
(f) SPSK1 SPSK0
t
(c)

Symbol
duration

Figure 1.22 Binary digital modulation schemes: (a) On-Off Keying (OOK), a special type of ASK; (b) Frequency Shift Keying
(FSK); (c) Phase Shift Keying (PSK); (d) OOK symbols; (e) FSK symbols; (f) PSK symbols.

modulation is the special case M = 2. To do this, we take a group of k bits at a time and represent them with a
unique carrier state or code symbol, where k = log2 M, or M = 2k . For example, taking k = 3 bits at a time, we have
M = 23 = 8 possible states, namely 000, 001, 010, 011, 100, 101, 110, and 111, which must each be represented by
a unique carrier state. Each symbol now carries three bits of information and the bit rate is therefore three times
the symbol rate. In general, M-ary modulation increases the bit rate according to the relation
Bit rate = log2 M × (Symbol rate) (1.5)
However, the symbol states are closer together than in binary modulation, making it easier for noise effects to
shift one state sufficiently close to an adjacent state to cause a symbol error. For example, the phase difference
between adjacent states in an M-ary PSK is (360/M)∘ , which for M = 16 is only 22.5∘ . Bearing in mind that these
PSK states all have the same energy, carrier amplitude being constant, we see that there is great potential for
error if the transmission medium is prone to phase distortion. A combination of amplitude and phase shift keying
(APSK), also called quadrature amplitude modulation (QAM), is often used, which increases the phase difference
between symbol states. For example, it is possible to design for the minimum phase difference between states of
equal energy in 16-APSK to be 90∘ , compared to only 22.5∘ in 16-PSK.
Telegram: @Persian_IEEEComSoc

1.5.4 Circuit Versus Packet Switching

A transmission path is required for information to flow between two users (i.e. between source and destination)
in a communication system. It is only in a tiny minority of simple communication systems that this path is a
permanent setup. For the rest, the communication system is a network of transceivers or nodes, and a path is a
connected sequence of links which must be found as needed to convey information from the sending node through
one or more intervening nodes to the destination node. We may classify communication systems either as circuit
switched or packet switched according to how such a path is established.
In both cases, switching is required at various points in the network (e.g. at a local exchange in a PSTN) to
establish the concatenation of links that constitute the transmission path. This switching may be implemented in
space or time or as a combination of the two. In space switching, a message arriving on one input port of the switch
is routed or connected in its entirety to a designated output port or line of the switch without buffering. In time
 48 1 Overview of Communication Systems

Bufferless
1 1
(a)
2 2
3 3
4 4

Controller

Time slots: 1 2 3 4 1 2 3 4

H a b c d T H d a b c T
(b) Buffers
Input frame Output frame

Controller

Switching table:
1→ 2
(c) 2→ 3
3→ 4
4→ 1

Figure 1.23 (a) Space switching; (b) Time switching; (c) Switching table; (d) Concentrator switching; (e) Route switching.

switching, the input and output of the switch are multi-slot data frames. The input data frame is buffered, and the
contents of its time slots are switched to designated respective time slots in the output frame. The specification of
input and output pairings is usually given in a routing table. Furthermore, space switching may be either concen-
trator switching, in which the switch has a larger number of input than output lines, or it may be route switching,
where the switch has the same number of input and output lines. A concentrator switch is typically used to switch
subscriber lines at a local exchange. It improves line occupancy but introduces the possibility of lost or blocked calls
if an input message arrives at the switch while all output lines are engaged. Route switching on the other hand is
typically used when switching trunk lines between exchanges. Figure 1.23 illustrates space and time switching in
(a) and (b), respectively, under the control of the same switching table given in (c). The slots labelled H and T in
the data frames of (b) are the header and trailer bits of each frame. Note, for example, that the contents of time slot
1 in the input frame are identified with the letter ‘a’ and these contents are switched to time slot 2 in the output
frame according to the specification of the routing table. Concentrator and route switching are also illustrated in
Telegram: @Persian_IEEEComSoc

Figure 1.23d and e, respectively.

1.5.4.1 Circuit Switching

In a circuit switched communication system, before communication begins between two users, a dedicated chan-
nel or trunk or transmission path is set up end-to-end for the exclusive use of the two users for the entirety of their
connection. The dedicated channel may be a separate frequency band or time slot or code sequence or combina-
tions of these, depending on the channel access plan in operation. If a free path or circuit cannot be set up for a
new connection between a pair of users then the requesting user is either denied access (i.e. blocked) or is placed
on a queue until a free circuit becomes available.
 1.5 Classification of Communication Systems 49

Figure 1.23 (Continued)

Fewer output lines

Many input lines
(d)

N output lines
N input lines
(e)

The circuit switching approach described above has several advantages.

● Once a path has been established, i.e. once the call setup phase has been completed, a message will transit
quickly through the system. There are no extra delays at intermediate nodes arising from queuing and the time
taken to absorb portions of the message or its packets into a buffer at each node.
● The transmission system or network can manage congestion arising from excessive demand for simultaneous
connections. New connection requests are simply denied or placed on a queue to protect the quality of active
connections.
● The message has a smooth transit through the network from source to destination. There is no jitter arising from
variations in the transit time of portions of the message, which would be the case if sections of the path were
shared with other active connections. Such jitter is not desirable in real-time voice and video communications.
● There is no need for extra bits (called overhead bits) to be added to each portion of the message to convey source
and destination addresses and sequencing information (for use in reassembling the message in the right order
at the destination). As a result, the number of bits conveyed through the network is significantly reduced.
Telegram: @Persian_IEEEComSoc

● The overall computational and processing demand on the network is low since the message will transit seam-
lessly through the network without the need for any further processing at intermediate nodes once a path has
been established.
However, circuit switching has several significant shortcomings.
● It can lead to a highly inefficient utilisation of network resources. The dedicated channel cannot be used for
another call during idle intervals of the active call. An interactive phone conversation between two people will
be punctuated by periods of silence and an Internet browsing session may contain long intervals of inactivity, for
example while examining downloaded content. The circuit switching approach locks away and wastes channel
capacity during such idle intervals.
 50 1 Overview of Communication Systems

● The call setup phase, when a free channel is being arranged prior to commencement of the call, may be
responsible for a significant portion of the delay from the time a request is made for a channel to the time the
message is fully delivered at the destination. In fact, for short messages the call setup phase will account for
the bulk of the delay, which may lead to an unacceptably low data throughput – this being the number of bits
successfully delivered divided by the total time taken.
● The dedicated connection provides for a constant data rate between the two users, and this makes it difficult to
interconnect users with a variety of data rates.
● The impact of a technical failure or breakdown in any portion of the path or route from source to destination is
catastrophic and results in the termination of the active call. Once a call has been set up, circuit switching has
no mechanism to modify the route mid-call. The probability of call termination is therefore increased on (e.g.
long-distance) connections involving the concatenation of many links since the failure of any one link will lead
to call termination.
The features itemised above make circuit switching the ideal approach to ensure the best service quality for
real-time voice and video communication through a network with reliable constituents. The approach is also
suitable for transmitting sufficiently long messages of whatever content or type through such networks. How-
ever, a circuit switched network could be likened to a community that generously divides its cake among the few.
Packet switching offers the possibility of a different approach in which everyone can have a smaller piece of the
community cake.

1.5.4.2 Packet Switching

In a packet-switched communication system or network, each message (which must be in digital form) is first
broken into one or more packets of bits prior to transmission. Overhead bits in the form of a packet header are
added to each packet to fulfil a variety of purposes, such as identification (e.g. source and/or destination address
and/or route ID), error control, and transmission and network management functions. A data packet therefore
consists of two sections, namely a header made up of overhead bits and a payload made up of source bits. Note
that the source bits will include message bits, any redundant bits inserted by the channel encoder for error control,
and overhead bits from any previous round of data packetisation or framing.
If the packets are of fixed length (usually specified in bytes ≡ 8 bits) they are referred to as cells. An example
of this fixed-length packet approach is ATM switching. An ATM cell is of length 53 bytes, which consists of a
5-byte header and a 48-byte payload. In other cases, such as IP (Internet protocol) switching, the packets are of
variable length. IPv4 (version 4) has a header length that is typically 24 bytes but may vary from 20 to 60 bytes
depending on header options. Since a path may be a concatenation of different link types, each with its own
maximum transmission unit (MTU) which is the maximum number of bytes in a packet that the link can support,
and the lowest MTU is typically 576 bytes for all link types, the recommended maximum IPv4 packet length
is therefore 576 bytes to avoid packet fragmentation in transit. The latest IP version is IPv6, which has a fixed
40-byte header and a variable payload length constrained by a lowest MTU spec of 1280 bytes. IPv4 makes a 32-bit
Telegram: @Persian_IEEEComSoc

provision for source address (and similarly for destination address), which is enough to uniquely identify 232 ≈ 4.3
billion items such as nodes. In the early days of computer networking this provision was thought to be more than
enough, but with the explosion of the Internet, including IoT (the Internet of Things), that early thinking proved
extremely short-sighted. IPv6 therefore makes a 128-bit address provision, which is enough to uniquely identify
340 undecillion items—one undecillion being one trillion trillion trillion, i.e. one followed by 36 zeroes. IPv6 is
thus considered the ultimate IP version with enough capacity for all future needs, but its uptake has been slow
due to the need to invest in new Internet routers.
There are two modes of packet switching, namely connection-oriented (CO) packet switching, also called virtual
circuit, and connectionless (CL) packet switching, sometimes called datagram.
Connection-oriented (CO) packet switching: in CO packet switching, a free but nonexclusive path is
first set up through the network from source to destination and all user packets follow this path and arrive at
the destination strictly in the same order as they left the source. It is important to observe that, unlike circuit
 1.5 Classification of Communication Systems 51

switching, the path set up prior to the commencement of transmission in this virtual-circuit approach is not
a dedicated path. Other packets from other users may simultaneously use portions of the same path to reach
separate destinations. Therefore, packets may need to be queued at each node until their designated outgoing
link is free. Each packet header carries a logical connection identifier or virtual circuit identifier (VCI), which
identifies a predefined route to be followed by the packets from source to destination. Intervening nodes do not
therefore make any routing decisions.
Examples of connection-oriented packet switching include X.25 – the first public data network launched in
the 1970s, Frame Relay – a wide area network (WAN) protocol adopted in the 1980s that largely replaced X.25,
and ATM – launched in the 1990s with a lot of hype but now, due to the runaway global success of IP, mainly
confined to use by telephone carriers for high-speed internal data transport. CO packet switching addresses some
of the shortcomings of the circuit-switching approach and delivers the following improvements (when compared
to circuit switching):
● CO packet switching allows node-to-node links to be dynamically shared by multiple users over time so that
network resources are used much more efficiently.
● It supports the exchange of packets between two users operating at different data rates.
● It allows new calls to be accepted during heavy traffic, although with an increased packet transfer delay, and
hence reduced data throughput, due to a higher incidence of queuing at intermediate nodes. Circuit switching
would simply block new calls during heavy traffic if a free path cannot be established.
● A prioritisation scheme may be efficiently implemented whereby important packets (e.g. those conveying net-
work status information) are catapulted to the front of the queue at each node so that they have a much faster
transit through the network. The only way to implement such prioritisation in a circuit-switched approach
would be to dedicate some paths for exclusive use by priority packets, but this would be inefficient and waste-
ful because such packets are few and far between. Think of it this way: you wouldn’t build exclusive roads for
ambulance vehicles and fire engines because there are too few of them for this to be cost-effective. However, the
very fact that there are only a few of them means they can be prioritised and even allowed to interrupt other
road users’ green lights at road junctions. In this way, their journey time is massively reduced, yet the added
delay to other road users is negligible.
However, in addition to retaining the drawbacks of circuit switching associated with requiring an initial call
setup phase and the impact of failure of any portion of the path after call commencement, CO packet switching
introduces the following disadvantages (when compared to circuit switching):
● CO packet switching increases the transit time of messages due to queuing delays at network nodes. As a mini-
mum, a node delays each packet by the time taken to fully receive the packet into the node’s buffer before routing
to the designated output port can commence.
● It introduces a variation in overall packet delay (called jitter), which is not desirable for real-time voice and video.
Telegram: @Persian_IEEEComSoc

Although all packets of one message connection follow the same route, the transit time of each of these packets
will be a function of time as the queue state and workload of shared nodes change due to varying network
conditions.
● It increases the number of bits transported in the network because a message is broken into packets, and to each
of these packets a header is added. For example, in ATM, around 10% of all transported bits are overhead bits
from the header.
● There is increased processing at network nodes. Although nodes do not perform independent routing, each
node must at least read the VCI of each packet in order to correctly forward the packet according to the route
established when the circuit was set up.
Nevertheless, because CO packet switching solved what was the biggest shortcoming of circuit switching,
namely the inefficient utilisation of network resources, it was an important step towards building a global
high-capacity integrated digital network.
 52 1 Overview of Communication Systems

Connectionless (CL) packet switching: in CL packet switching, there is no initial call or connection setup
phase. Packets are sent as individual datagrams, each containing source and destination IDs in its header. The
packets are independently routed through the network at each router, which makes its own decisions on a
packet-by-packet basis according to a dynamic routing table that is regularly updated as network conditions
change.
IP is a CL packet switching implementation which has become the single most dominant networking technology
and has left its closest rivals, such as ATM, in relative obscurity. As the name implies, IP is the routing protocol
that powers the entire Internet and since 4G in 2011 it was also adopted for mobile broadband connectivity. In
the 1990s, phrases like voice over Internet protocol (VoIP) and television over Internet protocol (TVoIP) were coined,
but such phrases have become superfluous because it is now everything over IP. CL packet switching addresses
some of the shortcomings of the virtual circuit packet switching approach, including the following improvements
(when compared to virtual circuit switching):
● CL packet switching has no call setup phase, so message transmission through the network is an instant-start
process. Therefore, the delay associated with circuit setup is eliminated and transmission is much quicker for
users sending only a few packets, which is often the case in social media interactions and SMS (short messaging
service).
● It is much more flexible and agile in reacting to network conditions and, unlike virtual circuits, can do so mid-call
or mid-connection. If congestion develops at a given node, subsequent packets from the source can be routed
away from that node, whereas in virtual circuits all packets of a call must follow a predefined route, which may
include the congested node.
● It is also inherently more reliable. If a node fails then only packets that are lost in the crash are affected. A route
may be easily found through the network for subsequent packets to bypass the failed node. In virtual circuits,
all connections set up prior to the node failure and that include the failed node in their paths would continue
to attempt to pass through the node after its failure and would therefore be terminated.
However, CL packet switching does introduce the following disadvantages (when compared to virtual circuit
switching):
● CL packet switching causes packets to transit the network more slowly since routing decisions must be made
at every node. It is therefore a slower mechanism for users sending many packets. It could be said that CL is an
instant-start jogger, whereas CO is a slow-start sprinter.
● Packets may follow different routes to the same destination and therefore may arrive out of order. This necessi-
tates extra provision in the CL packet header and processing at the destination to allow received packets to be
sorted into the right order to form the correct message.
● It is inherently difficult in CL packet switching to implement measures that guarantee a desired quality of service
(QoS) and achieve congestion control. CL being an instant-start algorithm, it means that any source can instantly
Telegram: @Persian_IEEEComSoc

start sending packets through the network at any time. Each such new entrant can contribute towards both
network congestion and degrading the QoS of existing users. In contrast, QoS and congestion control can be
more readily managed in virtual circuit switching by negotiating the allocation of enough resources in advance
for each connection during the call setup phase.
Despite the above shortcomings, CL’s unmatched routing flexibility and elimination of the significant delay
overhead in call set-up have made it the undisputed technology of choice for the ubiquitous broadband commu-
nication networks of the twenty-first century, including the Internet and 4G mobile networks and beyond. The
significant weakness of CL packet switching in relation to congestion control and QoS have been cleverly miti-
gated by overlaying it with smart oversight and control algorithms such as transmission control protocol (TCP),
real-time transport protocol (RTP), and user datagram protocol (UDP). For example, TCP provides the congestion
and flow control mechanisms with which IP is supervised to ensure a smooth and reliable data transfer between
sender and receiver.
 Review Questions 53

1.6 Epilogue
Our detailed overview of communication systems is now complete. We had a twofold purpose in this chapter.
First, we wanted to carefully and nonmathematically lay a foundation of communication engineering and erect
important knowledge pegs on which you can hang the more detailed study presented in subsequent chapters.
Second, we wanted to make a brief and highly selective historical sketch of the journey of telecom from telegraphy
in 1837 to ubiquitous broadband Internet and 5G mobile communication in 2020. We believe that your study of
communication engineering should begin with allowing yourself to be informed and inspired in equal measure by
this history. We hope that seeing such potential for seemingly endless progress will stir in you a genuine hunger
for competence in the subject and for a complete mastery of its principles and concepts. We hope that you will go
on to become one of the architects of a yet-unseen broadband communication future.
Back in the 1940s during the heyday of copper line communication, who saw the rise of optical fibre communi-
cation using glass as a transmission medium? In the 1950s when the launch of Sputnik 1 set off the space race, who
saw today’s satnav ubiquity? When computer networking began with ARPANET in 1971, who imagined broad-
band Internet and its social media offspring? And today, who sees the future of telecom? Is there anything beyond
today’s smartphone? Will it be an artificial intelligence device (AID)? Is there anything beyond today’s multimedia
broadband Internet with its predominantly audio-visual capability? Will it be a multisensory Internet that adds
tactile and olfactory functionality so that after 150 years of telecoms it can finally directly cater to our senses of
touch and smell as much as to our sight and hearing?
We hope that this chapter has inspired you to freely imagine our telecom future and that subsequent chapters
will equip you with the skills to help shape it. We will now embark on the latter task and begin a more in-depth
study of the subject in the next chapter with a detailed introduction to signals and systems in the context of telecom-
munications.

References

1 Live-Counter.com (2020). Worldwide Automobile Productions. https://www.live-counter.com/number-of-cars

(accessed on 11th June 2020).
2 Statista (2020). Number of smartphone users worldwide from 2016 to 2021 (in billions). https://www.statista
.com/statistics/330695/number-of-smartphone-users-worldwide (accessed on 11th June 2020).
3 Gartner (2020). Gartner says worldwide smartphone sales will grow 3% in 2020. https://www.gartner.com/en/
newsroom/press-releases/2020-01-28-gartner-says-worldwide-smartphone-sales-will-grow-3 (accessed 11th June
2020).
Telegram: @Persian_IEEEComSoc

Review Questions
. Discuss the drawbacks of verbal and visual nonelectrical telecommunications.
1.1 (a)
(b) In spite of its many drawbacks, nonelectrical telecommunication remains indispensable in society. Dis-
cuss various situations in which nonelectrical telecommunication has some important advantage over
(modern) electrical telecommunications. In your discussion, identify the type of nonelectrical telecom-
munication and the advantage it provides in each situation.

1.2 Sketch the flags that represent the following messages in semaphore code. Remember to include an end of
signal code in each case.
(a) NO
(b) TAKE 5.
 54 1 Overview of Communication Systems

1.3 Sketch the voltage pulse sequence for the Morse code representation of each of the following complete
messages.
(a) WE WON 2-0
(b) I LOVE YOU.

1.4 Baudot code (Appendix, Table A.2) requires seven transmitted bits per character, which includes a start
bit (binary 0) and a stop bit (binary 1). Write out the complete bit sequence for the transmission (least
significant bit (LSB) first) of the following messages using Baudot code.
(a) DON’T GIVE UP
(b) 7E;Q8.

1.5 In asynchronous transmission of ASCII-coded data, characters are transmitted one frame at a time starting
with the LSB of the character. A frame is formed as follows: (i) take the 7-bit ASCII code for the character;
(ii) insert one odd parity check bit in the MSB position to form a byte; (iii) write the byte from LSB to MSB,
so transmission will start from LSB then insert bit 0 as a frame header (i.e. start bit) and bits 11 as a frame
trailer (i.e. stop bits) to complete the frame.
Repeat Question 1.4 for asynchronous transmission of the same messages using ASCII code. Compare the
number of bits required by Baudot and ASCII codes in each case and comment on your results. Calculate
the transmission efficiency of both coding schemes in (a) and (b).

1.6 The last century witnessed revolutionary developments in telecommunications far beyond what could have
been contemplated by the pioneers of telegraphy. Discuss four key areas of these developments.

1.7 Draw a clearly labelled block diagram that is representative of all modern communication systems. List 12
different devices that could serve as the information source of a communication system. Identify a suitable
information sink that may be used at the receiver in conjunction with each of these sources.

1.8 With the aid of a suitable diagram where possible, discuss the operation of the following information
sources or sinks.
(a) Dynamic microphone
(b) Loudspeaker
(c) CRT
(d) Plasma-panel display
(e) Liquid crystal display.

1.9 Figure 1.8 shows the ISO standard 226 contours of equal loudness as a function of frequency and SPL.
Telegram: @Persian_IEEEComSoc

(a) What is the loudness of sound of frequency 200 Hz and SPL 40 dB?
(b) If a tone of SPL 55 dB is perceived at a loudness 50 phon, what are the possible frequencies of the tone?
(c) Determine the amount by which the vibration (i.e. SPL in dB) of a 50 Hz tone would have to be increased
above that of a 1 kHz tone in order that both tones have equal loudness 10 phon.

1.10 Discuss the signal processing tasks performed by the transmitter in a communication system. Indicate why
each process is required and how it is reversed at the receiver to recover the original message signal.

1.11 Audio and television broadcasting have both gone digital and there has been a virtually complete digi-
talisation of the telecommunication networks of most countries. Examine the reasons for this trend of
 Review Questions 55

digitalisation by presenting a detailed discussion, with examples where possible of the advantages and
disadvantages of digital communication. Indicate in your discussion how the impact of each of the disad-
vantages can be minimised in practical systems.

1.12 Give two examples of each of the following types of communication systems:
(a) Simplex system
(b) Half-duplex system
(c) Duplex system
(d) Analogue baseband
(e) Analogue modulated
(f) Digital baseband
(g) Digital modulated.

1.13 .(a) With the aid of suitable block diagrams, discuss the operation of any two examples of an analogue
baseband communication system.
(b) What are the most significant disadvantages of an analogue baseband communication system?

1.14 With the aid of suitable block diagrams, discuss the generation of the following discrete baseband signals
starting from an analogue message signal.
(a) PAM
(b) PDM
(c) PPM.

1.15 Compare the discrete baseband systems of Question 1.14 in terms of noise performance, bandwidth require-
ment, power consumption, and circuit complexity.

1.16 Explain how three independent user signals can be simultaneously carried on one transmission link using
discrete baseband techniques. What are the major drawbacks of this system?

1.17 Using a suitable block diagram, discuss the steps involved in the ADC process. Identify the major parame-
ters that must be specified by the ADC system designer and explain the considerations involved.

1.18 Give a detailed discussion of the desirable characteristics of line codes, which are used to electrically rep-
resent bit streams in digital baseband communication systems.

1.19 Discuss the three ‘R’ operations of a digital baseband receiver.

Telegram: @Persian_IEEEComSoc

1.20 Discuss the roles of modulation in communication systems. Hence, identify the transmission media that
can be used for voice communication in the absence of modulation.

1.21 Sketch the resulting waveform when the bit stream 10111001 modulates a suitable carrier using:
(a) OOK
(b) Binary FSK
(c) Binary PSK.
Assume that the carrier frequency is always an integer multiple of 1/T b , where T b is the bit interval.
 56 1 Overview of Communication Systems

1.22 If the following digital communication systems operate at the same symbol rate of 4 kBd, determine the bit
rate of each system.
(a) OOK
(b) 2B1Q
(c) Binary FSK
(d) 16-APSK.

1.23 Discuss the advantages of circuit switching over packet switching and identify the scenarios where circuit
switching would be the preferred technique.

1.24 Discuss the mechanisms used by TCP to ensure a reliable service even though the IP network underneath
may be unreliable.
Telegram: @Persian_IEEEComSoc
 57

Introduction to Signals and Systems

Complexity is merely a multiplicity of simplicity.

The art of solving a complex problem is therefore to break it into its simple parts.

In this Chapter
✓ What is a signal?
✓ Forms of telecommunication signals.
✓ Objective and subjective classification of telecommunication signals.
✓ Specification of various special and standard waveforms.
✓ Qualitative introduction to sinusoidal signals.
✓ Quantitative characterisation and parameters of sinusoidal signals.
✓ Logarithmic units and transmission path calibration.
✓ Basic system properties and their applications.
✓ Worked examples and end-of-chapter questions.

2.1 Introduction
The field of telecommunication deals with the transfer or movement of information from one point to another by
electronic or electromagnetic means. The information to be transferred has first to be represented as a telecommu-
nication signal. In approaching this important field of study, one must therefore have a thorough understanding of
telecommunication signals and how they are manipulated in telecommunication systems. The presentation in the
Telegram: @Persian_IEEEComSoc

next three chapters has been carefully designed to help you acquire this crucial grounding using an approach that
emphasises applications and a graphical appreciation of concepts. It will pay you rich dividends to work diligently
through these three chapters, even if you have prior familiarity with some of the concepts discussed. You will gain
important insights into crucial fundamental principles and acquire analysis skills which will serve you very well
in your subsequent studies and future career.
We start with understanding what constitutes a signal in telecommunications, the various forms in which sig-
nals may exist or are produced, how signals are classified both subjectively and objectively, and the mathematical
specification of several special waveforms that serve as building blocks or analysis tools for general or arbitrary sig-
nals. Our treatment of special waveforms gives deserved emphasis to the sinusoidal signal using first a qualitative
introduction followed by various quantitative characterisations and manipulations. We devote some time to the
subject of logarithmic units, establishing its concepts and highlighting short cuts and common pitfalls, in order
Communication Engineering Principles, Second Edition. Ifiok Otung.
© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 58 2 Introduction to Signals and Systems

to give you complete mastery of this fundamental engineering tool. Basic systems properties are also introduced
with examples drawn from both continuous-time and discrete-time operations.

2.2 What Is a Signal?

A signal is a variable quantity which may be expressed as a function of one or more parameters which conveys
information about a physical phenomenon, activity, or event. Telecommunication signals may convey a range of
audio-visual, olfactory, tactile, and sensor measurement information, or they may serve to fulfil system control
roles. An important feature of an information-bearing signal is that its strength or level or value, howsoever
defined, must contain a variable component. Although a constant signal may be useful in conveying energy, such
a signal carries no information. For example, a green traffic light conveys an ‘okay’ message only if it can change
to red (for example) that conveys ‘not okay’, or amber that conveys ‘perhaps okay or soon okay’. Conversely, a
permanently green light that has constant brightness and intensity may be decorative or illuminating but has no
information content.
A signal is described as one-dimensional (1D) if it is a function of only one variable such as time and is said
to be multidimensional if it depends on two or more variables. There are numerous examples of signals: when
you speak into a microphone, your vocal tract activity produces acoustic pressure waves which are converted by
the microphone into a voltage waveform whose value varies with time in synchrony with the impinging sound
pressure. This is a speech signal and is 1D. Other 1D signal examples include audio (which is the voltage output of
a microphone in response to mechanical vibrations in general, covering a broader frequency range than speech),
and ambient temperature, humidity, and atmospheric pressure at a fixed point (all being the voltage or current signal
produced by a suitable sensor placed at the point), etc.
A still image, on the other hand, is a two-dimensional (2D) signal since the value of the picture element (or pixel
for short) depends on two variables, namely the x and y coordinates that specify location within the image. A video
signal is three-dimensional (3D), being a time sequence of still images, and hence a function of three variables
(x, y, t); although it is usually rendered as a 1D function of time through a left-to-right, top-to-bottom sequen-
tial scanning of the scene. Finally, radar can be employed to obtain detailed information about the variation
with time of a physical quantity such as rain rate within a volume in 3D space (x, y, z), which therefore yields
a four-dimensional signal. Can you think of a five-dimensional signal? In this book we will only consider signals
that are 1D single-valued functions of time. This is not as restrictive as first seems, because the variations (and hence
information) contained within any multidimensional signal can be fully captured and represented as a 1D function
of time through sequential scanning.
Telegram: @Persian_IEEEComSoc

2.3 Forms of Telecommunication Signals

Table 2.1 lists five different forms of signals, the media in which they can exist, and the signal parameter that
varies to convey information. Where a signal exists in acoustic or visual form, the first operation of a telecommu-
nication system is to convert the signal into electrical form (using a suitable transducer) for ease of processing and
transmission or further conversions before transmission.
Figure 2.1 shows various examples of transducers often employed to convert a telecommunication signal from
one form to another.
The microphone converts an acoustic (or sound) signal such as music and speech into an electrical signal,
whereas the loudspeaker performs a reverse process to that of the microphone and converts electrical signal into
sound.
 2.3 Forms of Telecommunication Signals 59

Table 2.1 Forms of signals.

Form Medium Varying Parameter

Electrical Wire (e.g. twisted wire pair in local Current or voltage level
subscriber loop and coaxial cable in CATV)
Electromagnetic Space (e.g. broadcast TV and radio) Electric and magnetic fields
Acoustic Air (e.g. interpersonal communication) Air pressure
Light Optical fibre and free space On–off switching of light from injection laser
diode (ILD) or light-emitting diode (LED)
Visual Electronic or mechanical display device Reflected light intensity
(e.g. paper for print image)

Acoustic Electrical
Microphone

Loudspeaker

Visual TV camera Electrical

or Scanner

Display
screen
or Printer

Light Photodetector Electrical

e.g. PIN or APD

LED
or LD
Telegram: @Persian_IEEEComSoc

Electromagnetic Electrical
Receive
antenna

Transmit
antenna

Figure 2.1 Signal conversions in telecommunication.

 60 2 Introduction to Signals and Systems

The scanner or television camera converts a visual signal into an electrical signal. The scanner works only on
printed (2D and still) images, whereas the television camera can handle physical scenes (3D and movable) as well.
The reverse process of converting from an electrical to a visual signal is performed using a suitable display device
such as the screen of a smartphone or a printer connected to a computer.
A light detector or photodetector converts light energy into electric current. Examples include the PIN diode
and the avalanche photodiode (APD). In both cases, the diode consists of an intrinsic (I) semiconductor layer
sandwiched between heavily doped layers of p-type and n-type semiconductors, hence the name PIN. The diode
is reverse-biased, and this creates a charge-depleted layer in the intrinsic region. Light falling on this layer creates
electron–hole pairs that drift in opposite directions to the diode terminals (electrons towards the positive-voltage
terminal and holes towards the negative), where they register as current flowing in the same direction. The APD
uses a large reverse-bias voltage so that the photo-induced electrons acquire enough kinetic energy to ionise other
atoms, leading to an avalanche effect.
A light-emitting diode (LED) and a laser diode (LD) perform the reverse process of converting electric current
to light. The optical radiation results from the recombination of electron–hole pairs in a forward-biased diode. In
the laser diode, there is a threshold current above which the stimulated emission of light of very narrow spectral
width commences.
An antenna may be regarded as a type of transducer. Used as a transmitter, it converts electrical signals to electro-
magnetic waves launched out into space in the desired direction. When used as a receiver, it converts an incoming
electromagnetic radiation into a current signal.
Signals may be classified subjectively according to the type of information they convey, or objectively depending
on their waveform structure. The waveform or wave shape of a signal is a plot of the values of the signal as a
function of time. Note that the following discussion employs terms such as frequency and bandwidth, which are
explained fully in subsequent chapters.

2.4 Subjective Classification of Telecommunication Signals

2.4.1 Speech
Speech sound is the response of the human ear–brain system to the sound pressure wave emitted through the lips
or nose of a speaker. The elements involved in speech production are illustrated in Figure 2.2. Air is forced from
the lungs by a muscular action that is equivalent to pushing a piston. The air stream passes through the glottis,
the opening between the vocal cords or folds. For voiced sounds, the vocal cords are set into vibration as the air
stream flows by, and this provides a pulse-like and periodic excitation to the vocal tract – the air passage from
the vocal cords to the openings of the mouth and nose. For unvoiced (or voiceless) sounds, there is no vibration
of the vocal cords. The vocal tract, comprising the oral and nasal cavities, is a tube of nonuniform cross-section
Telegram: @Persian_IEEEComSoc

beginning at the glottis and ending at the lips and nose. The nasal cavity is shut off from the vocal tract by raising
the soft palate (also called the velum) and coupled by lowering the velum. The vocal tract acts as a sound modifier.
Its shape is changed to determine the type of sound that is produced. Different vowel sounds are generated by the

Glottis Figure 2.2 Elements involved in human speech production.

Velum

Vocal folds
Lungs
Mouth Nose
 2.4 Subjective Classification of Telecommunication Signals 61

resonance of the vocal tract under different shapes. Vowels have strong periodic structures and higher amplitudes.
On the other hand, different consonant sounds are produced by constriction of the vocal tract at different points.
Air stream from the lungs flows through this point of constriction at a high velocity, giving rise to turbulence.
Consonants have weaker amplitude and a noise-like spectrum. Some sounds, such as the non-vowel part of zee,
are generated by mixed excitation. In this case the turbulent airflow at a point of constriction is switched on and
off by the closing and opening of the glottis due to the vibration of the vocal cords.
A microphone converts acoustic signal into an electrical signal, referred to as the speech signal. Knowledge of
the time domain and frequency domain characteristics of speech signals is very useful in the design of speech
transmission systems. Note that the exact details of these characteristics will vary significantly depending on the
speaker’s sex, age, emotion, accent, etc. The following summary is therefore intended as a rough guide.
A typical speech waveform is the sum of a noise-like part and a periodic part. Figure 2.3 shows 100 ms segments
of the voiced sound ‘o’ in over, the unvoiced sound ‘k’ in kid and the mixed sound ‘z’ in zee. Voiced sound is
strongly periodic and of relatively large amplitudes compared to unvoiced sound which is noise-like and of small
amplitudes. For example, the voiced sound in Figure 2.3a has a strong periodic component of about 356 Hz. Even
for a single speaker, speech signals tend to have a large dynamic range of about 55 dB. The dynamic range is given
by the ratio between the largest amplitude (which may occur during intervals of syllabic stress) and the smallest
amplitude in soft intervals measured over a period of 10 minutes or more. The ratio of the peak value to the
root-mean-square (rms) value, known as the peak factor, is about 12 dB. Compared to a sinusoidal signal that has
a peak factor of 3 dB, we see therefore that speech signals have a preponderance of small values. It is evident from
Figure 2.3 that most of these small values would represent consonants and they must be faithfully transmitted to
safeguard intelligibility. Over a long period of time the large amplitudes of a speech signal follow what is close to
an exponential distribution, whereas the small amplitudes follow a roughly Gaussian distribution.
The short-term spectrum of speech is highly variable, but a typical long-term spectrum is shown in Figure 2.4.
This spectrum has a lowpass filter shape with about 80% of the energy below 800 Hz. The low-frequency
components (50–200 Hz) enhance speaker recognition and naturalness, whereas the high-frequency components
(3.5–7 kHz) enhance intelligibility such as being able to differentiate between the sounds of ‘s’ and ‘f’. Good

(a) 0

–1
0 10 20 30 40 50 60 70 80 90 100
Relative Value

(b) 0
Telegram: @Persian_IEEEComSoc

–1
0 10 20 30 40 50 60 70 80 90 100
1

(c) 0

–1
0 10 20 30 40 50 60 70 80 90 100
time, ms

Figure 2.3 Speech waveforms of (a) voiced sound ‘o’ in over, (b) unvoiced sound ‘k’ in kid, and (c) mixed sound ‘z’ in zee.
 62 2 Introduction to Signals and Systems

–10
Relative power, dB

–20

–30

–40
0.5 1 2 4 8
Frequency, kHz

Figure 2.4 Typical power spectrum of speech signal.

subjective speech quality is, however, obtained in telephone systems with the baseband spectrum limited to
the range 300–3400 Hz. This bandwidth is the ITU-T standard for telephony, although a smaller bandwidth
(500–2600 Hz) has been used in the past on some international networks to increase capacity.

2.4.2 Music
Music is a pleasant sound resulting from an appropriate combination of notes. A note is sound at a specific fre-
quency, which is usually referred to as the pitch of the note. Pitch may be expressed in hertz (Hz) or using a
notation based on a musical scale. Western musical scale consists of notes spaced apart in frequency at the ratio of
21/12 (= 1.059463). For example, if the middle A of a piano is of frequency 440 Hz then subsequent notes up to the
next A, one octave higher, are 466, 494, 523, 554, 587, 622, 659, 698, 740, 784, 831, and 880 Hz. You will observe, as
illustrated in Figure 2.5, that there is a doubling of frequency in one octave, the space of 12 notes.
Sounds of the same note played on different instruments are distinguishable because they differ in a character-
istic known as timbre. The note of each musical instrument comes from a peculiar combination of a fundamental
frequency, certain harmonics and some otherwise related frequencies, and perhaps amplitude and frequency

466 554 622 740 831

Telegram: @Persian_IEEEComSoc

A 494 587 698 Ai

440 523 659 784 880
One octave

Figure 2.5 Fundamental frequency (in Hz) of various notes on a piano keyboard.
 2.4 Subjective Classification of Telecommunication Signals 63

variations of some of the components. Although sounds from different musical instruments can be differenti-
ated over a small bandwidth of, say, 5 kHz, a much larger bandwidth is required to reproduce music that faithfully
portrays the timbre. High-fidelity music systems must provide for the transmission of all frequencies in the audi-
ble range from 20 Hz to 20 kHz. Note therefore the significant difference in bandwidth requirements for speech
and music transmission. The maximum bandwidth required for speech transmission is 7 kHz for audio confer-
ence and loud-speaking telephones. This is referred to as wideband audio to distinguish it from the normal speech
transmission using the baseband frequencies 300–3400 Hz.

2.4.3 Video
The video signal is in general the electrical representation of movable 3D scenes as transmitted in television sys-
tems. An image signal is a special case that arises when the scene is a still 2D picture. Information conveyed by
video signals includes the following.
● Motion: video signals must contain the information required for the display system to be able to create the
illusion of continuous motion (if any) when pictures of the scene are displayed. To do this the camera must
take snapshots of the scene at a high enough rate. Each snapshot is called a frame. Observations show that a
frame rate of about 30 (or a little less) per second is adequate. In motion pictures (or movies) a frame rate of
24 frames/s is used, but at the display each frame is projected twice to avoid flicker. This means a refresh rate
of 48 Hz. Increasing the frame rate increases the amount of information and hence the required transmission
bandwidth.
● Luminance: information about the brightness of the scene is contained in a luminance signal, which is a
weighted combination of the red, green, and blue colour contents of the scene. The weighting emphasises
green, followed by red, and lastly blue, in a way that reflects the variation of perceived brightness with the
colour of light. For example, given red, green, and blue electric bulbs of the same wattage, the green light
appears brightest followed by the red, and the blue light appears dullest.
● Chrominance: the colour content of the scene must be conveyed. There is an infinite range of shades of colours,
just as there is an infinite range of shades of grey, or an infinite set of real numbers between, say, zero and one.
However, (almost) any colour can be produced by adding suitable proportions of the three primary additive
colours: red, green, and blue. Colour information is conveyed by chrominance signals from which the display
device extracts the proportion of each of the primary colours contained in the scene. Because the eye is not very
sensitive to changes in colour, the fine details of colour changes are usually omitted to allow the use of a smaller
bandwidth for the chrominance signals.
● Audio: the sound content of the recorded scene or other superimposed sound information is conveyed in a
bandwidth of 15 kHz. This allows high-fidelity sound reproduction.
● Control signals: the television receiver or display device needs information to correctly construct a 2D picture of
the scene from the 1D (time function) video signal. Thus, synchronisation pulses are included that control the
Telegram: @Persian_IEEEComSoc

rate at which the display device draws the scene line by line in step with the image scanning operation of the
camera.
How the above signals are combined to form the composite baseband video signal depends on the type of tele-
vision system. However, it is obvious that video signals contain much more information than music or speech
signals and will therefore require a larger transmission bandwidth. The now obsolete 625/50 PAL analogue TV
standard required a bandwidth of 8 MHz. The digital TV standard DVB-T2 (Digital Video Broadcasting – 2nd Gen-
eration Terrestrial), adopted in 2009, supports compressed information bit rates ranging from 7.44 to 50.32 Mb/s
and uses bandwidths of 1.7, 5, 6, 7, 8, or 10 MHz, depending on a number of factors, including image resolu-
tion (e.g. standard definition television, SDTV, or high definition television, HDTV), and signal processing mode
(e.g. modulation scheme, such as QPSK, 16APSK, 64APSK, or 256APSK, and code rate, such as 1/2, 3/5, 2/3, 3/4,
4/5, or 5/6).
 64 2 Introduction to Signals and Systems

2.4.4 Digital Data

Digital data may originate from textual information generated by PCs or other data terminal equipment and
encoded using the universal encoding standard Unicode, which assigns a unique integer number (called a code
point) to each character in the world’s writing systems, unlike the American Standard Code for Information Inter-
change (ASCII) which covered Latin characters only. The first 128 code points of Unicode correspond to ASCII in
Table A.4 (Appendix A). A good introduction to ASCII and Unicode is given in Section 1.3.1.
Coding transforms the message text into a stream of binary digits. For example, the message ‘Take $1’ is coded
in 8-bit ASCII as

An important consideration in the transmission of textual information is that there are no insignificant details
that may be ignored or modified. For example, changing even just one bit in a stream of digital data can have
far-reaching consequences. To verify this, check the effect of changing the 53rd bit in the above 56-bit digital data.
Elaborate schemes have been devised to detect and, in some cases, correct transmission errors in digital data by
adding extra bits to the message bits. Any data compression techniques adopted to reduce the size of the message
bit stream must be lossless. That is, the receiver must be able to expand the compressed data back to the original
bit stream.
Speech, music, video, and myriads of sensor signals are usually converted to a digital representation to exploit the
numerous advantages of digital transmission and storage. This analogue-to-digital conversion (ADC) at the trans-
mitter is followed at the receiver by the reverse process of digital-to-analogue conversion (DAC). Although such an
ADC output signal is a bit stream like the digital representation of textual information, there is an added flexibility
that both lossless and lossy compressions can be applied to reduce the number of bits required to represent the
analogue signal. Lossy compression eliminates signal details that are subjectively insignificant in order to save
bandwidth. For example, barring other distortions, speech can be accurately transmitted (with intelligibility and
speaker recognition) by using nearly lossless representation at 128 kbits/s, or using lossy compression to reduce
the bit rate to 4 kbits/s or even lower. There is a loss of subjective quality in the latter, but the consequences are not
catastrophic, as would be the case if lossy compression were applied to digital data conveying textual information.

2.4.5 Facsimile
A facsimile signal, usually abbreviated fax, conveys the visual information recorded on paper, including printed or
handwritten documents, drawings, and photographs. The transmission medium is in most cases a telephone line
and sometimes radio. There are several fax standards such as the Group 3 (G3) standard, which has a standard
resolution of 200 lines per inch (lpi), meaning that the paper is divided into a rectangular grid of picture elements
Telegram: @Persian_IEEEComSoc

(pixels), with 200 pixels per inch along the vertical and horizontal directions. That is, there are 62 pixels per square
millimetre. The Group 4 (G4) standard has an ultrafine resolution of 400 lpi.
The fax signal is generated by scanning the paper from left to right, one grid line at a time, starting from the top
left-hand corner until the bottom right-hand corner of the paper is reached. For black-and-white only reproduc-
tion, each pixel is coded as either white or black using one binary digit (0 or 1). For better-quality reproduction in
G3 Fax, up to five bits per pixel may be used to represent up to 25 = 32 shades of grey. The resulting bit stream is
compressed at the transmitter and decompressed at the receiver in order to increase the effective transmission bit
rate. At the receiver, each pixel on a blank paper is printed black, white, or shade of grey according to the value of
the bit(s) for the pixel location. In this way the transmitted pattern is reproduced.
The popularity of fax around the world reached its peak in the 1980s and 1990s. Since then fax has been in
decline, although many large businesses and organisations around the world still maintain a fax machine and
 2.5 Objective Classification of Telecommunication Signals 65

associated dedicated phone number. The emergence of smartphone devices, broadband wireless communication,
and the Internet has completely obsoleted fax technology in all social applications. The reason is simple: if you
can scan the document or picture using your smartphone camera and transmit it in high quality over a wireless
network to its destination using a myriad of apps on your device at no extra cost to your mobile subscription, why
would you need to invest in a separate and comparatively more cumbersome fax machine? Fax is still being used
for business communication, but even here the technology is rapidly approaching extinction since legally binding
documents may now be electronically completed online and digitally signed. Also, manually produced and signed
documents may be scanned and exchanged using email or other Internet-based applications.

2.4.6 Ancillary and Control Signals

Ancillary and control signals perform system control functions and carry no user-consumable information. There
are numerous examples. Synchronising pulses are transmitted along with a video signal to control the start of each
line and field scan. Pilot signals are transmitted along copper line systems to monitor variation in attenuation. Extra
symbols or bits are inserted in digital signals for error detection/correction, synchronisation, and other system
management functions.

2.5 Objective Classification of Telecommunication Signals

2.5.1 Analogue or Digital
A signal is said to be analogue if it can take on a continuum of values and is defined at a continuum of time instants
within a specified duration. Thus, an analogue signal g(t) is a continuous function of a continuous independent
variable t, drawn from a set of real numbers. We therefore describe an analogue signal as a continuous-value,
continuous-time signal. Most naturally occurring signals are analogue, e.g. speech, ambient temperature, etc. Dig-
ital signals on the other hand can only take on a set of values (e.g. 0 and 1, or V and −V for binary signals) at a
set of uniformly spaced time instants (e.g. t = 0, ±T s , ±2T s , ±3T s , …, where T s is called the sampling interval).
A digital signal is thus a discrete function of a discrete independent variable (drawn from a set of integers), and
hence a discrete-value, discrete-time signal. Note that a continuous-value signal that is defined only at discrete time
instants is not a digital signal; rather such a signal is a sampled signal or a discrete-time signal or simply a discrete
signal. Also, a discrete-value signal that is defined throughout a continuum of time instants within a specified
duration is neither digital nor discrete but is a quantised signal, often referred to as a staircase signal in view of its
shape, which does simplify certain computations on the signal – as will become obvious in the next chapter.
Figure 2.6 illustrates the four signal types introduced above. The analogue signal g(t) is defined for the duration
from tmin to tmax , with a range of values from Amin to Amax . Note that there is an infinite number of time instants
Telegram: @Persian_IEEEComSoc

between tmin and tmax and g(t) is defined at every one of them. Also, g(t) can have any of the infinite number of
values between Amin and Amax . The sampled or discrete signal g(nT s ) is obtained by taking the values of g(t) only
at discrete time instants that are integer multiples of the sampling interval T s . This process generates a time series
or sequence of values g[n], which is usually depicted diagrammatically, as shown in Figure 2.6b using a stem plot
in which a vertical line of height g(nT s ) is drawn at t = nT s to represent the value of the discrete signal at that time
instant.
A clarification of notation is in order here. We will use square brackets to denote the sequence of a discrete signal
as g[n], and round brackets to denote the value of the sequence at the nth sampling instant as g(n). Thus

g[n] = {g(n), n ∈ ℤ}; g(n) ≡ g(nT s )

where ℤ = {…, −3, −2, −1, 0, 1, 2, 3, …} is the set of integer numbers and T s is the sampling interval.
 66 2 Introduction to Signals and Systems

g(t)
Amax
tmin
(a) t
tmax
Amin

g(nTS) ≡ g(n)
tmin
(b) t = nTs
TS tmax

gq(t)
A4
tmin
A3
(c) t
A2 tmax
A1
gq(n)
A4
tmin A3
(d) t = nTs
A2 TS tmax
A1

Figure 2.6 (a) Analogue signal is continuous-value, continuous-time; (b) Sampled or discrete signal is continuous-value,
discrete-time; (c) Quantized signal is discrete-value, continuous-time; (d) Digital signal is discrete-value, discrete-time.

We will assume that it is understood that the sampling interval – though not written explicitly in g(n) – is T s
so that, for example, g(−2) is the value g(−2T s ), i.e. the value of g(t) at time t = −2T s . Clearly the independent
variable of g[n] is discrete time nT s , which is for convenience written simply as n (without the factor T s ), but you
must always remember that the value n = 3, for example, corresponds to the time instant t = 3T s . Note therefore
that a sampled signal or sequence g[n] is a continuous function of a discrete independent variable, hence we
describe it as a continuous-value, discrete-time signal.
Figure 2.6c shows the quantised or staircase signal gq (t), where the subscript q indicates that the values of the
signal are quantised (i.e. restricted to a discrete set of values). This signal is obtained from g(t) by rounding g(t) at
each of the continuum of time instants to the nearest allowed level. In this example there are four allowed levels or
values or states A1 , A2 , A3 , and A4 . A staircase signal is therefore a discrete function of a continuous independent
variable; hence we describe it as a discrete-value, continuous-time signal. Finally, Figure 2.6d shows a digital signal
gq [n] obtained from the sampled signal g[n] by approximating (i.e. rounding) each value to the nearest allowed
level. This process is known as quantisation. It introduces an irreversible rounding or quantisation error in each
quantised sample. The example shown in Figure 2.6d is a quaternary digital signal since gq [n] has four possible
Telegram: @Persian_IEEEComSoc

values or states. If there are only two possible states, the signal is referred to as a binary digital signal, which is the
most common type of digital signal used in digital communications. Ternary digital signals, with three possible
states, are also used for line coding to represent data as voltage levels in a baseband transmission system. In general,
a digital signal may have M possible states (where M is usually an integer power of 2) and is described as an M-ary
digital signal.
Two features of digital signals make them more suitable for representing information in practical communica-
tion systems. First, unlike an analogue signal, the precise value of a digital signal is not important. Rather, what
matters is the range within which the signal value lies. For example, if a binary digital signal is transmitted using
+12 V to represent binary 1 and −12 V to represent binary 0 then any received signal value above 0 V would be
interpreted as binary 1 and any below zero as binary 0. It would then take an impairment effect exceeding 12 V to
cause an error in the detection or interpretation of the received signal. Second, unlike analogue signals, only the
 2.5 Objective Classification of Telecommunication Signals 67

sampling instants are significant. This means that the detected value of a digital signal is not impaired by any dis-
tortions or noise outside the decision instants. Thus, a careful choice of sampling instants allows the digital signal
to be detected at the instants of minimum distortion to its values. More significantly, the gaps between samples
can be used to transmit other user signals, allowing multiple users to be simultaneously accommodated in the
transmission system. This popular technique is known as time division multiplexing and is only possible if the user
signals are discrete or digital.
Subsequently, and throughout this book, a continuous-value, continuous-time signal will be referred to simply
as a continuous signal or analogue signal and denoted such as g(t), x(t), y(t), etc., whereas a continuous-value,
discrete-time signal will be referred to simply as a discrete signal and denoted such as g[n], x[n], y[n], etc. with its
nth sample denoted g(n), x(n), y(n), etc. respectively. Furthermore, it will be assumed that a signal is continuous
unless otherwise specified.

2.5.2 Periodic or Nonperiodic

A periodic signal consists of a sequence or pattern of values that repeats over and over in time. The waveform of a
periodic signal will therefore have a repetitive fundamental shape. The duration of the shortest fundamental shape
is called the period, T of the waveform. Thus, if g(t) is a periodic signal of period T, then
g(t) = g(t ± nT); n = 0, 1, 2, 3, … (2.1)
The smallest duration of time T for which Eq. (2.1) is satisfied is the period of g(t). It means, for example, that
the signal has the same value at the time instants t = 0, T, 2T, 3T, …, which we may write as
g(0) = g(T) = g(2T) = g(3T) = · · ·
Strictly speaking, a periodic signal is eternal, having neither a beginning nor an end. It exists from t = −∞ to
+∞. In this respect, it would seem impossible to generate a truly periodic signal. However, in practical systems,
the duration of interest is always limited. Thus, if a signal’s waveform has a fundamental shape that repeats over
and over within the finite interval of interest then the signal is said to be periodic.
A nonperiodic (or aperiodic) signal has no repetitive pattern and hence does not satisfy Eq. (2.1) for any T > 0.
Some signals (e.g. some speech waveforms) may have a slowly changing repetitive pattern. They are referred to
as quasiperiodic signals. Figure 2.7 shows examples of periodic, nonperiodic, and quasiperiodic waveforms. The
fundamental shape of the periodic waveform is emphasised.
The fundamental frequency f o of a periodic signal is the rate at which the fundamental shape repeats itself. This
is the number of cycles per second, which is usually expressed in Hz. Since one cycle or fundamental shape has a
duration T (the period), it follows that the fundamental frequency is given by
1
fo = (2.2)
T
Telegram: @Persian_IEEEComSoc

In Eq. (2.2), if the period T is in seconds, the frequency is in Hz; when T is in milliseconds (ms), the frequency is
in kHz; when T is in microseconds (μs), the frequency is in MHz; and when T is in nanoseconds (ns), the frequency
is in gigahertz (GHz), etc.
By analogy between the cycles of circular motion and the cycles of periodic signal variation, we note that there
are 2𝜋 radians or 360∘ in one cycle. Thus, the number of radians per second, called angular frequency and denoted
𝜔, of a periodic signal of fundamental frequency f o is
2𝜋
𝜔 = 2𝜋fo = (in radians per second) (2.3)
T
In a similar manner, we consider a discrete signal g[n] to be periodic if
g(n) = g(n ± N) (2.4)
 68 2 Introduction to Signals and Systems

No repetitive pattern

(a)

Fundamental shape
Value

(b)

T
Slowly changing repetitive pattern

(c)

Time

Figure 2.7 Waveforms: (a) Nonperiodic; (b) Periodic with period T; (c) Quasiperiodic.

The smallest positive integer N for which Eq. (2.4) is satisfied is the period of the discrete signal, indicating that
the signal completes one cycle in N samples. In line with Eq. (2.3) for a periodic continuous-time signal, such a
periodic discrete signal has a fundamental angular frequency (expressed in radians per sample and denoted Ω)
defined by
2𝜋
Ω= (2.5)
N
Ω is often referred to simply as fundamental frequency. It is important to note that the unit of (fundamental)
frequency for a discrete signal is radian/sample, whereas that of the corresponding parameter of a continuous-time
signal is rad/s.
In addition to the fundamental frequency, every periodic signal (except sine waves, which we discuss later)
contains other frequencies, called harmonic frequencies, at integer multiples of the fundamental frequency. The
nth harmonic frequency is
Telegram: @Persian_IEEEComSoc

fn = nf o ; n = 1, 2, 3, · · · (2.6)
Note that the first harmonic frequency, f 1 = 1 × f o = f o , is the same as the fundamental frequency f o . This is
different from conventional usage in music where the first harmonic is twice the fundamental frequency.

2.5.3 Deterministic or Random

A deterministic signal is exactly predictable and can therefore be expressed as a completely specified function of
time or other variable such as location in space. For example, a signal described by the functional relationship
g(t) = 20cos(100𝜋t) is deterministic. A random signal on the other hand always has some element of uncertainty
in its value and can therefore only be described in terms of the probability that its instantaneous value will lie in
some specified range. Thus, whereas there is always some uncertainty about the value of a random signal before
 2.5 Objective Classification of Telecommunication Signals 69

it occurs, there is no uncertainty whatsoever about the value of a deterministic signal at any point or time past,
present, and future. Examples of a random signal include noise voltage in an electrical conductor and the signal
r(t)cos(100𝜋t + 𝜓(t)) in which the envelope r(t) and phase 𝜓(t) vary randomly with time t.

2.5.4 Power or Energy

If a signal has a finite nonzero energy E (i.e. 0 < E < ∞), it is said to be an energy signal. If, on the other hand, a
signal has a finite nonzero power P (i.e. 0 < P < ∞), it is described as a power signal. Power and energy are discussed
in more detail under signal characterisation in the next chapter, but for now it will suffice to note that normalised
average power P of a signal is the mean square value of the signal and is also the energy E of the signal per unit
time. Thus, for a signal of duration 𝕋 we may write
[ ]
E
P = lim ; E = lim [P•𝕋 ] (2.7)
𝕋 →∞ 𝕋 𝕋 →∞

We see that if E is finite then P = 0; and if P is nonzero and finite then E → ∞. That is, an energy signal has finite
energy but zero average power, whereas a power signal has finite power but infinite energy. Every signal will be
either an energy signal or a power signal, but not both. A realisable finite-duration signal (called a pulse or symbol)
is always an energy signal, whereas periodic signals and random signals are power signals.

2.5.5 Even or Odd

A real-valued function g(t) of a real variable t is said to be even if and only if its value is unchanged when t changes
sign. On the other hand, the function is said to be odd if its value changes sign without a change in magnitude when
t changes sign. Thus, even signals are symmetric about the vertical axis, whereas odd signals are antisymmetric.
Expressed mathematically
{
g(−t) for all t ⇒ Even signal
g(t) = (2.8)
−g(−t) for all t ⇒ Odd signal
Similarly, a discrete signal g[n] is even if g(n) = g(−n) for all n, and is odd if g(n) = −g(−n) for all n. Some signals,
such as g(t) in Figure 2.6a, are neither odd nor even, but they can always be expressed as a sum of even and odd
signals. That is, given an arbitrary signal g(t), we can write
g(t) = ge (t) + go (t)
where ge (t) is even and go (t) is odd. To derive expressions for these even and odd components of g(t), substitute
−t for t in the above equation, and make use of the definition of even and odd functions given in Eq. (2.8) to obtain
g(−t) = ge (−t) + go (−t)
Telegram: @Persian_IEEEComSoc

= ge (t) − go (t)
Thus, we have two equations: g(t) = ge (t) + go (t), and g(−t) = ge (t) − go (t), which when added together yields
1
[g(t) + g(−t)]
ge (t) = (2.9)
2
and when subtracted yields
1
go (t) =[g(t) − g(−t)] (2.10)
2
As an example, consider the deterministic signal
g(t) = t2 (1 − t) − 10
 70 2 Introduction to Signals and Systems

g(t) = ge(t) + go(t)

72

(a) 0 t

–64
–4 –3 –2 –1 0 1 2 3 4
ge(t)
6

0 t
(b)

–10
–4 –3 –2 –1 0 1 2 3 4
go(t)
64

(c) 0 t

–64
–4 –3 –2 –1 0 1 2 3 4

Figure 2.8 (a) Arbitrary signal g(t); (b) Even component; (c) Odd component.

which is shown in Figure 2.8a and is clearly neither even nor odd. However, we may express this signal as the sum
of an even signal ge (t) and an odd signal go (t) obtained using Eqs. (2.9) and (2.10) as follows
1
ge (t) = [g(t) + g(−t)]
2
1
= [t2 (1 − t) − 10 + (−t)2 (1 − (−t)) − 10]
2
1
= [t2 (1 − t) − 10 + t2 (1 + t) − 10]
2
= t2 − 10
1
go (t) = [g(t) − g(−t)]
Telegram: @Persian_IEEEComSoc

2
1
= [t2 (1 − t) − 10 − {(−t)2 (1 − (−t)) − 10}]
2
1
= [t2 (1 − t) − 10 − t2 (1 + t) + 10]
2
= −t3
These two component signals ge (t) and go (t) are shown in Figure 2.8b and c.
Even and odd functions possess several useful properties which may be exploited to simplify signal analysis.
Using their definition (Eq. (2.8)), it is a straightforward matter to show that the following properties hold for even
and odd signals:
1. The product or quotient of two even signals is even, and the product or quotient of two odd signals is also even.
2. The sum of two even signals is even, and the sum of two odd signals is odd.
 2.6 Special Waveforms and Signals 71

3. The quotient or product of an even and an odd signal is odd.

4. The derivative of an even signal is odd, whereas the derivative of an odd signal is even.
5. The integral of an odd signal g(t) in any range that is symmetrical about the vertical axis (i.e. from t = −T to
t = T) is zero. It therefore follows that odd signals always have zero average value.
6. The integral of an even signal from t = −T to t = T is twice the integral from t = 0 to t = T.
7. The Maclaurin series of an even signal comprises only even powers, whereas that of an odd signal comprises
only odd powers.
8. The Fourier series of a periodic even signal comprises only cosine terms, whereas the Fourier series of a periodic
odd signal comprises only sine terms. We discuss the Fourier series in Chapter 4.
To illustrate how the knowledge of some of the above properties can greatly simplify computations on a signal,
consider the integration
𝜋∕3
t2 sin(5t)dt
∫−𝜋∕3
We know that t2 is even and sin(5t) is odd. Thus t2 sin(5t) is an odd signal (by property 3 above), and hence the
above integration is zero (by property 5). Notice how we have obtained a solution without a lengthy calculation
that would involve integration by parts.

Discussion Example

Classify each of the following signals in terms of its form, type of variation, and number of dimensions:
(a) The number of customers in a supermarket queue.
(b) The singing of birds in a forest.
(a) The number of customers in the supermarket queue is a variable quantity that is capable of changing contin-
uously with time, is observed at one location, takes on values that are restricted to positive integers, and in
its raw form (i.e. before being captured and processed by a suitable sensor such as a digital camera) is visual.
It is therefore a quantised 1D visual signal.
(b) The singing of birds in a forest is obviously an acoustic signal that can be captured using a microphone sensor
and converted into an electrical voltage signal representing music. The signal will vary continuously both in
time t and in value (specified in terms of loudness of the entire sound or of a specific pitch). Furthermore, the
observed value at any given time t will vary with location (x, y, z) within the forest volume. Thus, this is an
analogue 4D acoustic signal.
Telegram: @Persian_IEEEComSoc

2.6 Special Waveforms and Signals

Figure 2.9 shows some common waveforms in telecommunication systems. The trapezoidal pulse is made up of
rising, flat, and falling portions, and is often used to closely approximate realisable laboratory pulses. It reduces to
rectangular, triangular, sawtooth, and ramp pulses as special cases. A periodic pulse train may be characterised by
its duty cycle (sometimes expressed as a percentage) defined by

Duration of pulse 𝜏
Duty cycle = = (2.11)
Period of waveform T
A rectangular pulse train having a 50% duty cycle is usually simply described as a square wave. Note that the
triangular and sawtooth waveforms shown in Figure 2.9 are pulse trains having 100% duty cycle. A triangular pulse
has equal rise and fall times, whereas a sawtooth pulse has unequal rise and fall times. Sawtooth signals are used,
 72 2 Introduction to Signals and Systems

Trapezoidal pulse
train

Rectangular pulse
train

Triangular pulse
train

Sawtooth pulse
train

Sinusoidal waveform

Rectangular pulse

Random waveform

Figure 2.9 Common waveform types in telecommunications.

for example, in oscilloscopes to sweep an electron beam across the face of the cathode-ray tube (CRT) during the
rising portion of the waveform and to provide a quick flyback of the beam during the falling portion which has
a much shorter duration. Also shown in Figure 2.9 are the sinusoidal waveform, which constitutes the building
blocks of all other signals, and the random waveform, which (in the form of noise) is an ever-present unwanted
addition or companion to all other signals in practical transmission systems. These two signals are so crucial, we
dedicate separate sections (in this chapter and the next) to their treatment.
Telegram: @Persian_IEEEComSoc

Figure 2.10 shows three examples of a pulse train with different pulse shapes and duty cycles determined using
Eq. (2.11). Other fundamental waveforms and their manipulations to build other signals are introduced after these
worked examples.

Worked Example 2.1

Sketch a rectangular waveform of amplitude 5 V that has a duty cycle 20% and a 3rd harmonic frequency 6 kHz.

We are given f 3 = 6 kHz and duty cycle = 0.2. To sketch the waveform, the period T and pulse duration 𝜏 must
first be determined. Equation (2.6) gives the fundamental frequency
f3 6 kHz
fo = = = 2 kHz
3 3
 2.6 Special Waveforms and Signals 73

g1(t)
d = 0.2

t
τ = 25 μs
T = 125 μs
g2(t)
d = 0.8

t, ms
0 4 5 10 15
g3(t)
d = 0.4
t, ns
0 4 8 20 40 60

Figure 2.10 Duty cycle (d) of various pulse trains.

The period is then obtained using Eq. (2.2)

T = 1∕fo = 1∕(2 kHz) = 0.5 ms
Equation (2.11) then gives the pulse duration
𝜏 = T × Duty cycle = 0.5 × 0.2 = 0.1 ms
A sketch of the rectangular waveform is shown in Figure 2.11

ʋ(t), V
5

t, ms
0 0.1 0.5 0.6 1.0

Figure 2.11 Two periods of the rectangular waveform of Example 2.1.

Worked Example 2.2

Telegram: @Persian_IEEEComSoc

Determine the first three harmonic frequencies of the periodic waveform shown in Figure 2.12a.

The crucial first step here is to identify the fundamental shape or cycle of the waveform. This is done in Figure 2.12b
and the fundamental shape is emphasised using a bold line. We see that there are three repetitions of the shape in
a time of 6 μs, giving a period T = 2 μs. Equation (2.2) yields the fundamental frequency of the waveform
fo = 1∕T = 1∕(2 𝜇s) = 0.5 MHz.
Therefore
First harmonic frequency f1 = 1 × fo = 0.5 MHz
Second harmonic frequency f2 = 2 × fo = 1 MHz
Third harmonic frequency f3 = 3 × fo = 1.5 MHz
 74 2 Introduction to Signals and Systems

(a)
t, μs
0 6

(b)
t, μs
0 6
One cycle

Figure 2.12 Periodic waveform of Example 2.2.

2.6.1 Unit Step Function

The continuous and discrete versions of the unit step function, u(t) and u[n], respectively, are shown in Figure 2.13a
and are defined by
{ {
1, t ≥ 0 1, n ≥ 0
u(t) = ; u(n) = (2.12)
0, t < 0 0, n < 0

Staircase (i.e. discrete value) waveforms can be constructed as a sum of unit step signals of appropriate
amplitudes and delays, as illustrated in Worked Example 2.4 for a rectangular pulse. Also, we can learn a
lot about the characteristic of a system by observing the system’s response to a unit step signal applied at
its input.

2.6.2 Signum Function

The signum function sgn(t) is shown in Figure 2.13b and is defined by

⎧1, t>0
⎪
sgn(t) = ⎨0, t=0 (2.13)
⎪
⎩−1, t<0
Telegram: @Persian_IEEEComSoc

Noting that u(t) = 1 for t ≥ 0, which means that u(−t) = 1 for −t ≥ 0, i.e. u(−t) = 1 for t ≤ 0, we see that the signum
function can be written in terms of the unit step function as

sgn(t) = u(t) − u(−t) (2.14)

The discrete signum function or sequence sgn[n] is similarly defined as follows

⎧1, n>0
⎪
sgn(n) = ⎨0, n=0 (2.15)
⎪
⎩−1, n<0
 2.6 Special Waveforms and Signals 75

u(t) u(n)
1 1
(a) ... ...
t n
–4 –3 –2 –1 0 1 2 3 4

sgn( t)
1

(b) t

–1

rect(t/τ)
1
(c)
t
–τ/2 τ/2

rmp(t/τ) rmp(–t/τ)
1 1
(d) (i) (ii)
t t
–τ/2 0 τ/2 –τ/2 0 τ/2

trian(t/τ)
1
(e)
t
–τ/2 0 τ/2

gr(t)
1
(f)
t
–τr 0 τf
Telegram: @Persian_IEEEComSoc

Figure 2.13 (a) Unit step function; (b) Signum function; (c) Rectangular pulse; (d) Ramp pulse; (e) Triangular pulse;
(f) Sawtooth pulse; (g) Trapezoidal pulse; (h) Impulse function; (i) Sinc function.

2.6.3 Rectangular Pulse

The unit rectangular pulse, shown in Figure 2.13c and denoted rect(t/𝜏), is zero everywhere except in the interval
−𝜏/2 ≤ t ≤ 𝜏/2 of duration 𝜏, where it has unit amplitude. Thus
{ 𝜏 𝜏
1, − ≤ t ≤
rect(t∕𝜏) = 2 2 (2.16)
0, elsewhere
 76 2 Introduction to Signals and Systems

g(t)
1
(g)
t
τr τc τf

δ(t) δ(n)
1
(h) ... ...
t n
–4 –3 –2 –1 0 1 2 3 4
sinc(t/ Ts)
1

(i)

t
–5Ts –4Ts –3Ts –2Ts –Ts Ts 2Ts 3Ts 4Ts 5Ts

Figure 2.13 (Continued)

A little thought (explained in Worked Example 2.4) will show that we may write the rectangular pulse in terms
of the unit step function as
( ) ( ) ( )
t 𝜏 𝜏
rect =u t+ −u t− (2.17)
𝜏 2 2

2.6.4 Ramp Pulse

The unit ramp pulse, shown in Figure 2.13d(i) and denoted rmp(t/𝜏), is zero everywhere except in the interval
−𝜏/2 ≤ t ≤ 𝜏/2, within which it increases linearly from 0 at t = −𝜏/2 to rmp(t/𝜏) = 1/2 at t = 0, and finally rmp(t/𝜏) = 1
at t = 𝜏/2. Thus
⎧
Telegram: @Persian_IEEEComSoc

( ) ⎪ t + 1, −𝜏 ≤ t ≤ 𝜏
t
rmp = ⎨𝜏 2 2 2 (2.18)
𝜏 ⎪0, elsewhere
⎩
The ramp pulse can be expressed in terms of the rectangular pulse as
( ) ( ) ( )
t t 1 t
rmp = + rect (2.19)
𝜏 𝜏 2 𝜏
It is worth noting that the linearly decreasing pulse shown in Figure 2.13d(ii) is a time-reversed ramp pulse
rmp(−t/𝜏) which is related to the reference ramp pulse by

rmp(−t∕𝜏) = rect(t∕𝜏) − rmp(t∕𝜏) (2.20)

 2.6 Special Waveforms and Signals 77

2.6.5 Triangular Pulse

The unit triangular pulse, shown in Figure 2.13e and denoted trian(t/𝜏), is zero everywhere except in the interval
−𝜏/2 ≤ t ≤ 𝜏/2 where it increases linearly from zero at |t| = 𝜏/2 to 1 at t = 0. Thus
⎧
( ) ⎪1 − 2 |t|, −𝜏∕2 ≤ t ≤ 𝜏∕2
t
trian =⎨ 𝜏 (2.21)
𝜏 ⎪0, elsewhere
⎩
You may be able to observe that the triangular pulse trian(t/𝜏) is really a sum of two ramp pulses of width 𝜏/2,
one of them starting earlier by 𝜏/4 relative to rmp(t/𝜏), and the other time reversed and starting later by 𝜏/4 relative
to rmp(t/𝜏). Thus
( ) ( ) ( )
t t + 𝜏∕4 −t + 𝜏∕4
trian = rmp + rmp (2.22)
𝜏 𝜏∕2 𝜏∕2
To understand how to evaluate the two ramp pulses on the right-hand side of Eq. (2.22), simply follow the
definition of ramp pulses given in Eq. (2.18) to obtain
( ) ⎧ t + 𝜏∕4 1
t + 𝜏∕4 ⎪ + , −𝜏∕4 ≤ t + 𝜏∕4 ≤ 𝜏∕4
rmp = ⎨ 𝜏∕2 2
𝜏∕2 ⎪0, elsewhere
⎩
⎧ t
⎪2 + 1, −𝜏∕2 ≤ t ≤ 0
=⎨ 𝜏
⎪0, elsewhere
⎩
( ) ⎧ −t + 𝜏∕4 1
−t + 𝜏∕4 ⎪ + , −𝜏∕4 ≤ −t + 𝜏∕4 ≤ 𝜏∕4
rmp = ⎨ 𝜏∕2 2
𝜏∕2 ⎪0, elsewhere
⎩
⎧ t
⎪−2 + 1, 0 ≤ t ≤ 𝜏∕2
=⎨ 𝜏
⎪0, elsewhere
⎩
An alternative construction of the triangular pulse may be obtained by using Eq. (2.20) to express the second
term on the right-hand side of Eq. (2.22) as
( ) ( ) ( ) ( )
−t + 𝜏∕4 −(t − 𝜏∕4) t − 𝜏∕4 t − 𝜏∕4
rmp ≡ rmp = rect − rmp
𝜏∕2 𝜏∕2 𝜏∕2 𝜏∕2
which therefore converts Eq. (2.22) to
( ) ( ) ( )
Telegram: @Persian_IEEEComSoc

( ) t + 𝜏∕4 t − 𝜏∕4 t − 𝜏∕4

t
trian = rmp − rmp + rect (2.23)
𝜏 𝜏∕2 𝜏∕2 𝜏∕2

2.6.6 Sawtooth and Trapezoidal Pulses

The sawtooth pulse shown in Figure 2.13f is a generalisation of the triangular pulse in the sense that the triangular
pulse has equal rise and fall times 𝜏/2, whereas the sawtooth pulse has rise time 𝜏 r and fall time 𝜏 f , which are not
necessarily equal. Following Eq. (2.22), this pulse can be realised as
( ) ( )
t + 𝜏r ∕2 −t + 𝜏f ∕2
gr (t) = rmp + rmp (2.24)
𝜏r 𝜏f
 78 2 Introduction to Signals and Systems

An all-inclusive generalisation of the pulses discussed so far, namely rectangular, ramp, triangular, and sawtooth,
is given by the trapezoidal pulse shown in Figure 2.13g. The trapezoidal pulse provides a more realistic piecewise
approximation of practical pulses with a rising edge of duration 𝜏 r , flat or constant portion of duration 𝜏 c , and
falling edge of duration 𝜏 f . With the flat portion centred about the y axis as shown in Figure 2.13g, the trapezoidal
pulse is given by the expression
( )
⎧ 1 + 𝜏c + t , −(𝜏 + 𝜏 ∕2) ≤ t ≤ −𝜏 ∕2
⎪ 2𝜏r 𝜏r r c c
⎪
⎪1, −𝜏c ∕2 ≤ t ≤ 𝜏c ∕2
g(t) = ⎨( 𝜏c
) (2.25)
⎪ 1+ − 𝜏t , 𝜏c ∕2 ≤ t ≤ 𝜏f + 𝜏c ∕2
⎪ 2𝜏f f

⎪
⎩0, elsewhere
We may also express g(t) in terms of ramp and rectangular pulses as
( ) ( ) ( )
t + (𝜏r + 𝜏c )∕2 t −t + (𝜏f + 𝜏c )∕2
g(t) = rmp + rect + rmp (2.26)
𝜏r 𝜏c 𝜏f

Note that, as expected, the trapezoidal pulse reduces to each of the other pulses under the following conditions
𝜏r = 0; 𝜏f = 0; 𝜏c ≠ 0; ⇒ Rectangular pulse
𝜏r ≠ 0; 𝜏f = 0; 𝜏c = 0; ⇒ Ramp pulse
𝜏r = 𝜏f ≠ 0; 𝜏c = 0; ⇒ Triangular pulse
𝜏r ≠ 0; 𝜏f ≠ 0; 𝜏c = 0; ⇒ Sawtooth pulse (2.27)

2.6.7 Unit Impulse Function

The unit impulse (also called the Dirac delta) function 𝛿(t) is zero at all time instants except at t = 0, where its value
is infinitely large such that the total area under the signal’s waveform is unity (i.e. a value of 1). This area is often
referred to as the weight of the function. Thus, the unit impulse function is fully characterised by the equations
𝛿(t) = 0 for t ≠ 0
∞
𝛿(t)dt = 1 (2.28)
∫−∞
The waveform of 𝛿(t) is usually represented by an upward pointing vertical arrow or spike located at t = 0 as
shown on the left-hand side of Figure 2.13h. Note that the unit impulse can be realised as a limiting case of the
rectangular pulse
Telegram: @Persian_IEEEComSoc

[ ]
1
𝛿(t) = lim rect(t∕𝜏) (2.29)
𝜏→0 𝜏

The unit impulse is an even function since, by the above definition

𝛿(t) = 𝛿(−t) (2.30)
and it obeys the sifting property (sometimes referred to as the sampling property) given by
∞
g(t)𝛿(t − Ts )dt = g(Ts ) (2.31)
∫−∞
The above property can be seen to hold by noting that 𝛿(t − T s ) is zero everywhere except at t = T s , which means
that the only value of g(t) that counts in the integral is g(T s ) which does not depend on t and so can be taken out
 2.6 Special Waveforms and Signals 79

of the integral sign so that

∞ Ts+
g(t)𝛿(t − Ts )dt = g(Ts ) 𝛿(t − Ts )dt = g(Ts )
∫−∞ ∫−T −
s

where Ts− and Ts+ refer to time t just marginally below and above T s , respectively.
A relationship between the impulse function 𝛿(t) and the unit step function u(t) shown in Figure 2.13a may be
discovered by observing that the slope of u(t) is zero everywhere except at t = 0 where it is infinite in view of the
step change at this point. Furthermore, the total area under 𝛿(t) from t = −∞ to t = 0− (i.e. just under 0) is zero,
but from t = −∞ to t = 0+ (i.e. just above 0) and beyond this area is unity. Thus
d
𝛿(t) = u(t)
dt
t
u(t) = 𝛿(𝛾)d𝛾 (2.32)
∫−∞
The impulse function is a seemingly strange signal that turns out to be a very useful idealisation in systems
analysis. For example, (i) instantaneous sampling of an analogue signal is readily analysed when viewed as the
product of the analogue signal and a train of unit impulses, (ii) the Fourier transform of periodic signals can be
specified if impulse functions are introduced, and (iii) an arbitrary signal x(t) can be represented as the (contin-
uous) sum of variously weighted and delayed impulse functions. An important outcome of the last application is
that the output of a linear time-invariant system is realised as the convolution of the input and the system’s impulse
response. These concepts and applications will be discussed in due course.
The impulse function discussed so far is applicable only to continuous-time signals and systems. For application
to discrete-time signals and systems, we define a discrete version of the unit impulse function 𝛿[n] as a sequence,
shown on the right-hand side of Figure 2.13h, that is zero everywhere except at n = 0 where it has a value of 1.
Thus
{
1, n = 0
𝛿(n) = (2.33)
0, n ≠ 0
A discrete equivalent of the sifting property represented by Eq. (2.31) may be derived by noting that
{
1, n − k = 0
𝛿(n − k) =
0, Otherwise
Therefore, an arbitrary discrete-time signal, such as g[n] in Figure 2.6b, may be expressed as
g[n] = · · · + g(−2)𝛿(n + 2) + g(−1)𝛿(n + 1) + g(0)𝛿(n)
+ g(1)𝛿(n − 1) + g(2)𝛿(n − 2) + · · ·
Telegram: @Persian_IEEEComSoc

∑∞
= g(k)𝛿(n − k) (2.34)
k=−∞

2.6.8 Sinc Function

The sinc function is defined by
sin(πt)
sinc(t) = (2.35)
πt
We will see in due course the crucial role of the sinc function and its relationship to pulses and pulse trains
which are ubiquitous in digital communications. It is worthwhile to take a moment to become familiar with its
 80 2 Introduction to Signals and Systems

features introduced here. You may verify using your calculator that the sine of very small angles (2∘ or less) is
approximately equal to the value of the angle (in rad), the approximation becoming exact at 0∘ . That is

sin 𝜃 → 𝜃 as 𝜃 → 0

So, we may write

sin(𝜋t) 𝜋t
→ = 1 as 𝜋t → 0
𝜋t 𝜋t
Or, more formally
sin(𝜋t) 𝜋t
sinc(0) = lim = =1
t→0 𝜋t 𝜋t
Second, recall that the sine of integer multiples of 𝜋 is zero. Therefore
sin(n𝜋) 0
sinc(n) = = = 0, n = ±1, ±2, ±3, · · ·
n𝜋 n𝜋
It means that
⎧1 t=0
⎪
sinc(t) = ⎨0 t = ±1, ±2, ±3, · · · (2.36)
⎪
⎩sin(𝜋t)∕𝜋t t≠0

Figure 2.13(i) shows the waveform of sinc(t/T s ), featuring a main lobe of duration 2T s and unit peak value,
several side lobes of decaying peak absolute values, and nulls at t = nT s , where n is an integer ± 1, ±2, ±3, …

Worked Example 2.3 Even and Odd Components of a Signal

The unit step signal (Figure 2.13a) and the ramp pulse (Figure 2.13d) are neither even nor odd. In this worked
example we want to express each of these two signals as a sum of even and odd signals, and to sketch these
component waveforms.

Applying Eqs. (2.9) and (2.10) to the unit step signal, we obtain its even component ue (t) and odd component
uo (t) as follows
1 1
ue (t) = [u(t) + u(−t)] =
2 2
1 1
uo (t) = [u(t) − u(−t)] = sgn(t)
2 2
Telegram: @Persian_IEEEComSoc

To understand the time-reversed signal u(−t), recall that by definition in Eq. (2.12), u(t) = 1 for t ≥ 0, and is zero
elsewhere. Thus u(−t) = 1 for −t ≥ 0, which means for t ≤ 0, and is zero elsewhere. The sketch of ue (t) and uo (t) is
thus straightforward, as shown in Figure 2.14a.
Similarly, the even and odd components of the ramp pulse are given by
1 1
ge (t) = [rmp(t∕𝜏) + rmp(−t∕𝜏)] = rect(t∕𝜏)
2 2 ( ) ( )
1 t 1 t
go (t) = [rmp(t∕𝜏) − rmp(−t∕𝜏)] = rmp − rect
2 𝜏 2 𝜏
where we have made use of Eq. (2.20). Since the ramp pulse rmp(t/𝜏) increases linearly from zero at t = −𝜏/2 until
it reaches 1 at t = 𝜏/2, it follows that its time-reversed counterpart rmp(−t/𝜏) decreases linearly from 1 at t = −𝜏/2
to zero at t = 𝜏/2. Thus, the waveforms of ge (t) and go (t) are as shown in Figure 2.14b.
 2.7 Sinusoidal Signals 81

u(t) ue(t) uo(t)

1
½
(a) ½
t
= + ½
t t
−½

rmp (t/τ)
1 ge(t) = 0.5rect(t/τ) go(t)
(b) = ½
½
t
½
t
+ –τ/2 t
–τ/2 τ/2 –τ/2 τ/2 τ/2
−½

Figure 2.14 Worked Example 2.3.

Worked Example 2.4 Rectangular Pulse in Terms of Unit Step Signals

Express the rectangular pulse in terms of unit step signals.

It can be seen from Figure 2.15 that if we subtract a unit step signal that starts (i.e. assumes a value of 1) at
t = 𝜏/2 from another unit step signal that starts earlier at t = −𝜏/2, then the result will be zero everywhere except
in the interval −𝜏/2 ≤ t ≤ 𝜏/2 where the result is 1. This result fits the definition of a rectangular pulse of duration 𝜏.
Noting that, by definition, u(t) starts at t = 0, it follows that u(t + to ) starts at t + to = 0, i.e. at t = −to . Thus, u(t + 𝜏/2)
starts at t = −𝜏/2, and u(t − 𝜏/2) starts at t = 𝜏/2, and we may express the above observation mathematically as
follows
( ) ( ) ( )
t 𝜏 𝜏
rect =u t+ −u t−
𝜏 2 2
Note that the above result is the earlier stated Eq. (2.17).

rect(t/τ) u(t + τ/2) u(t – τ/2)

1 1 1
= −
t t t
–τ/2 τ/2 –τ/2 τ/2

Figure 2.15 Worked Example 2.4.

Telegram: @Persian_IEEEComSoc

2.7 Sinusoidal Signals

Sinusoidal signals play a major role in telecommunications for several reasons:

● Every type of telecommunication signal can be obtained as the sum of sinusoidal signals. This is the subject of
Fourier analysis and synthesis, which we address fully in Chapter 4.
● If we input a sinusoidal signal into a linear time-invariant system, the output will be another sinusoidal signal
of the same frequency, although there may be a change in amplitude and initial phase. These parameters are
discussed shortly.
 82 2 Introduction to Signals and Systems

● If the transmission system is linear then the (received) signal at the output of the system in response to an arbi-
trary (transmitted) signal at the system input is obtained straightforwardly as the sum of the system’s response
to individual sinusoidal components of the transmitted signal.
● Many concepts and processes in telecommunications such as frequency, signal bandwidth, transfer function of
a transmission medium, frequency response of a system, distortion, dispersion, modulation, etc., are fundamen-
tally about the sinusoidal components of signals and how they are modified in systems.
A complete understanding of sinusoidal signals and their manipulations is therefore essential in mastering the
foundational concepts of communication engineering. The next two subsections give a comprehensive introduc-
tion to sinusoidal signals followed by a detailed treatment of various sinusoidal signal operations in the subsequent
two subsections.

2.7.1 Qualitative Introduction

A sine waveform is generated by oscillatory or rotational motion without energy losses. A good example is the
displacement of a pendulum swinging in vacuum through small angles, e.g. in the range ± 2∘ or less, as shown
in Figure 2.16a. Figure 2.16b is obtained by plotting the displacement d(t) of the pendulum from its rest position
as a function of time. The pendulum starts at the rest position (displacement d = 0) at time t = 0, i.e. the initial
displacement d(0) = 0. The standard convention is to treat displacements to the right of the rest position as positive
and displacements to the left as negative. The pendulum first swings rightward and reaches a maximum displace-
ment (called amplitude) dm at time T 1 . It then swings leftward, reaching the rest position (d = 0) at time 2T 1 and
a minimum displacement (−dm ) at time 3T 1 . It reverses direction yet again and swings rightward to reach the rest
position (d = 0) at time 4T 1 . This completes one cycle of oscillation. Therefore, the period of the displacement
waveform is T = 4T 1 . Subsequent oscillation cycles continue, and the pendulum repeats the above sequence over
and over, but with time constantly increasing in equal amounts such that the second cycle is completed at time
t = 2T, the third cycle at t = 3T, etc. The displacement d(t) is a sinusoidal signal, often referred to simply as a
sinusoid.

Initial swing

dm d (t )
0 → dm
(b) 0
–dm
dm d1(t)
dm → 0
(c) 0
Telegram: @Persian_IEEEComSoc

–dm
dm d2(t)
–dm 0 dm –dm → 0
(d) 0
Displacement
–dm
(a) d (t)
dm 3
0 → –dm
(e) 0
–dm
0 T1 2T1 3T1 T = 4T1 2T 3T time

Figure 2.16 Generation of sinusoidal waveform by a swinging pendulum.

 2.7 Sinusoidal Signals 83

The oscillation of a pendulum with period T and amplitude dm can differ in the initial phase, which indicates
both the initial displacement and the initial direction of swing. Three different situations d1 (t), d2 (t), and d3 (t)
will be discussed. In Figure 2.16c the pendulum starts at the maximum displacement, i.e. d1 (0) = dm , and swings
leftward. It reaches the rest position (d1 = 0) at t = T 1 and the minimum displacement (d1 = −dm ) at t = 2T 1 .
At this point, it changes direction and swings rightward to reach d1 = 0 at t = 3T 1 and d1 = dm at t = 4T 1 = T,
thereby completing one cycle of oscillation. Consider the waveform d(t) of Figure 2.16b and note that it takes
a one-quarter period (or cycle) for d(t) to reach dm – the starting point of d1 (t). We say that waveform d(t) lags
behind d1 (t) by a one-quarter cycle or, equivalently, that d1 (t) leads d(t) by a one-quarter cycle. Setting d1 (t) as
the reference waveform and bearing in mind that there are 360∘ in one cycle (and hence 90∘ in a one-quarter
cycle), the conventional way to describe the fact that d(t) lags the reference waveform by a one-quarter cycle is
to state that it has an initial phase of −90∘ . It should be emphasised that, when talking about the initial phase
of a sinusoidal waveform, there must always be a reference waveform of the same period. In this discussion our
reference waveform is d1 (t) in Figure 2.16c. The initial phase of the reference waveform is always taken as 0∘ .
In Figure 2.16d the pendulum starts at minimum displacement d2 (0) = −dm and swings rightward (of course it
could not go in any other direction). It passes through the points d2 = −dm , 0, dm , 0, and −dm at respective time
instants t = 0, T 1 , 2T 1 , 3T 1 , and 4T 1 to complete one cycle of oscillation. Since it takes d2 (t) one-half of a period to
reach the starting point of the reference waveform d1 (t), we say that d2 (t) lags d1 (t) by 180∘ , or that it has an initial
phase of −180∘ .
Finally, Figure 2.16e shows the displacement waveform generated when the pendulum starts at d3 = 0 and
swings leftward. To complete one cycle, it passes through the points d3 = 0, −dm , 0, dm , 0 at respective time instants
t = 0, T 1 , 2T 1 , 3T 1 , and 4T 1 . Seeing that it takes three-quarter cycle for waveform d3 (t) to reach the starting value
dm of the reference waveform d1 (t), you would be mathematically correct to say that d3 (t) has an initial phase of
−270∘ . However, it is the convention to always keep the value of an angle (including phase) within the range −180∘
to +180∘ . To do this for the phase of d3 (t), observe that if the cycle of values of d3 (t) in Figure 2.16e is continued to
the left beyond t = 0 then we find that d3 (t) was at dm earlier than the reference waveform by a one-quarter cycle.
Thus d3 (t) leads the reference waveform by 90∘ and therefore has an initial phase of +90∘ . In general, a quick
way to convert any angle of value outside the range (−180∘ , 180∘ ) to a value within this range is to simply add or
subtract as many cycles (i.e. 360∘ ’s) as necessary to bring the angle into range. So, in this case adding 360∘ to −270∘
yields the initial phase of 90∘ for d3 (t).

2.7.2 Parameters of a Sinusoidal Signal

For a detailed quantitative discussion, consider in Figure 2.17 the generation of a sinusoidal signal by the rotational
motion of a radial arm OP of length A which starts initially at an angle 𝜙 with the +x axis and rotates counterclock-
wise about the origin O at a constant rate of f cycles per second. In this scenario OP is called a phasor, denoted A∠𝜙
to indicate its length A and initial orientation 𝜙 with the reference +x direction. The orientation 𝜃 of OP at some
Telegram: @Persian_IEEEComSoc

later time t may be expressed in the four units of measurement of angle shown in Table 2.2. The gradian (often
abbreviated grad), in which each quarter cycle or quadrant or right angle is assigned 100 grad, is a commonly used
unit of angle in surveying but is no longer widely supported in scientific calculators.
If OP starts from the +x axis direction and rotates to the −x axis direction then 𝜃 has increased from 0∘ to 180∘ .
Alternatively, we may say that 𝜃 has increased by half a cycle or that it has increased by half the circumference of
the circle swept by OP, which is
1
× 2𝜋 × Radius = 𝜋R
2
In the special case where radius R = 1, then the unit of angle measured as the length of the arc swept by OP is
called the radian. Thus, we have the following relationships between the four units
1
180∘ = cycle = 𝜋 radian = 200 gradian (2.37)
2
 84 2 Introduction to Signals and Systems

y OP rotates counterclockwise at
f cycles/sec, starting initially at ϕ
By definition: P
cos θ = g/A A ∠ 0°
h ≡ Acos(2πft)

A
sin θ = h/A
θ Reference direction
g x
O A ∠ –90°
f cycles/sec

≡ Asin(2πft)

Figure 2.17 Generation of a sinusoidal signal by a phasor A∠𝜙.

Table 2.2 Different units of angle.

Degree Cycle Radian Gradian

360∘ 1 2𝜋 400
0∘ 0 0 0
45∘ 1/8 𝜋/4 50
90∘ 1/4 𝜋/2 100
180∘ 1/2 𝜋 200
720∘ 2 4𝜋 800

Radian is the SI unit (International System of Units) of angle and this is the unit that is implicitly assumed unless
some other unit is explicitly indicated. We will, however, for convenience often quantify angle in degrees (∘ ) and
convert the degree value to radians (by multiplying by the factor 𝜋/180) when required in any manipulation. To
state the obvious, be always careful not to mix degrees and radians. For example, the first term on the right-hand
side of Eq. (2.38) below is in radians, so if the second term 𝜙 is in degrees you must convert it to radians before
adding. You would not add three apples to two oranges to obtain five apples, would you?
Since OP is constantly rotating and completes f cycles each second and each cycle is 2𝜋 radians, it follows that
𝜃 is a function of time, increasing by 2𝜋f radians each second starting with 𝜃 = 𝜙 at t = 0, so that after t seconds
𝜃 is
𝜃(t) = 2𝜋ft + 𝜙 (2.38)
The number of cycles completed each second is the frequency f of the oscillation, expressed in units of hertz (Hz)
Telegram: @Persian_IEEEComSoc

in honour of the German physicist Heinrich Rudolf Hertz (1857–1894) who in 1886 gave the first ever experimental
proof of the existence of electromagnetic waves. The rate of change of angle, expressed in radians per second
(rad/s), is the angular frequency 𝜔 of the oscillation. Since there are f cycles per second and each cycle contains
2𝜋 radians, it follows that there are 2𝜋f rad/s. That is
𝜔 = 2𝜋f (2.39)
As 𝜃 changes with time, the projection g of OP on the x axis becomes a function of time denoted g(t). It follows
by the definition of the cosine function (cos) stated in Figure 2.17 that g = A cos(𝜃), and hence from Eq. (2.38) that
g(t) = A cos(2𝜋ft + 𝜙)
= A cos(𝜔t + 𝜙) (2.40)
 2.7 Sinusoidal Signals 85

g(t)
Period, T
A

)
πft

Amplitude, A

Peak-to-peak Amplitude, App

← ϕ = –90°

s(2
←ϕ=0

o
Ac
0 t

)
ft
2π
in(
As
–A
0 T/4 T/2 3T/4 T 5T/4 3T/2 7T/4 2T →t
0° 90° ±180° –90° 0° 90° ±180° –90° 0° →θ

Figure 2.18 The horizontal projection of OP (in Fig. 2.17) is a sinusoidal signal g(t) = Acos(2𝜋ft + 𝜙).

which is a sinusoidal signal with waveforms as shown in Figure 2.18, featuring two plots, the solid curve for 𝜙 = 0
and the dashed curve for 𝜙 = −90∘ . To see how each waveform arises, note that for the case 𝜙 = 0, the rotating
radial arm OP points in the reference direction (+x axis) at t = 0 so its horizontal projection has maximum value A
at t = 0. That is, g(t) = A at t = 0, which we simply write as g(0) = A. It takes OP one-quarter of a cycle to rotate to the
+y direction where it has zero horizontal projection, so that g(T/4) = 0, where T is the duration of one cycle. After
half a cycle, OP lies along the −x direction, which gives a horizontal projection g(T/2) = −A. After three-quarters
of a cycle, OP points in the −y direction with zero horizontal projection so that g(3T/4) = 0. After one full cycle OP
is back to the +x direction where it has maximum horizontal projection g(T) = A. The rotation carries on endlessly
and produces the values of horizontal projection g(t) shown in the solid curve in Figure 2.18.
For 𝜙 = −90∘ , OP is initially along the −y axis where it has zero horizontal projection so that g(0) = 0. OP then
rotates and reaches the +x axis, +y axis, −x axis, and −y axis directions at respective time instants t = T/4, T/2,
3T/4, and T where the values of horizontal projection are, respectively, g(T/4) = A, g(T/2) = 0, g(3T/4) = −A, and
g(T) = 0. The rotation continues ad infinitum, leading to the function g(t) = Acos(2𝜋ft − 90∘ ) plotted in the dashed
line. Note that this dashed waveform is identical to the vertical projection h = Asin𝜃 of the rotating arm OP when
the arm is initially oriented along the +x axis (i.e. 𝜙 = 0) so that the vertical projections are h(0) = 0, h(T/4) = A,
h(T/2) = 0, h(3T/4) = −A, h(T) = 0, and so on, which yields the dashed waveform h(t) = Asin𝜃 = Asin(2𝜋ft). Thus

A sin(2𝜋ft) = A cos(2𝜋ft − 90∘ ) (2.41)

What this means is that, as indicated on the right-hand side of Figure 2.17, the horizontal projection of phasor
A∠0∘ corresponds to the signal Acos(2𝜋ft), which is always taken as the reference signal when specifying the phase
of other sinusoids of the same frequency, whereas the horizontal projection of phasor A∠−90∘ corresponds to the
signal Asin(2𝜋ft) which lags the reference signal by 90∘ or a one-quarter cycle, as is obvious from Figure 2.18.
Telegram: @Persian_IEEEComSoc

Some literature will equate the phasor A∠𝜙 with Acos(2𝜋ft + 𝜙); however, it is important to understand a crucial
difference between the two. A rotating phasor has always both horizontal and vertical projections. The horizontal
projection is described as the real component, which may be either positive (denoted with an optional ‘+’ sign
or factor, e.g. +5 or 5 for short) if it lies along the 0∘ (i.e. positive x axis) direction or negative (denoted with a
mandatory ‘−’ sign or factor, e.g. −5) if it lies along the 180∘ (i.e. negative x axis) direction. The vertical projection
is described as the imaginary component and is identified with a factor j (e.g. j5) if it lies along the 90∘ (i.e. positive
y axis) direction. Thus, the factor j corresponds to a counterclockwise rotation (called the phase shift) through 90∘ .
Since a further shift of 90∘ (i.e. second multiplication by j) yields a total phase shift of 180∘ which corresponds to
a factor of −1, it follows that
√
j × j = j2 = −1; hence j = −1 (2.42)
 86 2 Introduction to Signals and Systems

In general, a signal undergoes a phase change of√180∘ when it is multiplied by −1, but a phase change of 90∘
when it is multiplied by the imaginary number j = −1, which has the following peculiar but correct identities.
j2 = −1
1 1 j j
= × = 2 = −j
j j j j
j3 = j2 × j = −j
j4 = j2 × j2 = −1 × −1 = 1
j5 = j4 × j = j (2.43)
Returning to Figure 2.17, OP represents a complex quantity g + jh, and the rotating phasor A∠𝜙 is therefore
equivalent to
A∠𝜙 ≡ A cos(2𝜋ft + 𝜙) + jA sin(2𝜋ft + 𝜙) (2.44)
But Euler’s formula relates exponential and sinusoidal functions as follows
ej𝜃 = cos(𝜃) + j sin(𝜃) (2.45)
Substituting −𝜃 for 𝜃 yields e−j𝜃 = cos(−𝜃) + jsin(−𝜃), or (since the cosine function is even whereas the sine
function is odd)
e−j𝜃 = cos(𝜃) − j sin(𝜃) (2.46)
Adding together Eqs. (2.45) and (2.46) yields
1 j𝜃
cos(𝜃) = (e + e−j𝜃 ) (2.47)
2
whereas subtracting Eq. (2.46) from (2.45) produces
1
sin(𝜃) = −j (ej𝜃 − e−j𝜃 ) (2.48)
2
With 𝜃 replaced by 2𝜋ft + 𝜙, Eq. (2.47) leads to the result
A j(2𝜋ft+𝜙) A j(2𝜋(−f )t−𝜙)
A cos(2𝜋ft + 𝜙) = e + e (2.49)
2 2
Consider therefore Eqs. (2.44) and (2.45). Note that the rotating phasor A∠𝜙 corresponds to an exponential func-
tion Aexp[j(2𝜋ft + 𝜙)], whereas, from Eq. (2.49), a sinusoidal function (of amplitude A, frequency f , and initial phase
𝜙) is the sum of two counter-rotating phasors (as illustrated in Figure 2.19). Both phasors have amplitude A/2. One
starts initially at orientation 𝜙 (relative to the +x axis) and rotates counterclockwise at f cycles/second. This is the
positive frequency component of the sinusoid. The other starts at −𝜙 and rotates clockwise at f cycles/second,
which corresponds to a negative frequency −f . Note that in this scenario the imaginary parts of both phasors are
Telegram: @Persian_IEEEComSoc

always equal in magnitude but opposite in sign, so that they cancel upon addition of the two phasors, yielding a
function that is real at all times as expected of a sinusoidal function.
We are now able to define the parameters of sinusoidal signals with references to Eq. (2.40) and Figures 2.17
and 2.18.

2.7.2.1 Angle
Angle 𝜃 (= 𝜔t + 𝜙) is the value of the argument of the cosine function, i.e. the term enclosed within the brackets
on the right-hand side of Eq. (2.40) at any time instant t. It is usually expressed in radians. Since the sinusoidal
signal can be mapped to circular motion, every point of a sinusoidal waveform corresponds to an angle 𝜃 in the
range −180∘ to +180∘ . This is shown along the bottom row of Figure 2.18 for selected points. The crest of the
reference waveform (solid curve) is at 𝜃 = 0∘ , the trough is at 𝜃 = ±180∘ , the downward zero crossing point is at
 2.7 Sinusoidal Signals 87

Counterclockwise rotation at 2πf

rad/s, starting at θ = ϕ

2
A/ b
ϕ g(t)/2
g(t)/2 –ϕ
A/ –b
2

Clockwise
rotation at
2πf rad/s,
starting at
A j(2πft + ϕ) A j(2π(–f)t – ϕ) θ = –ϕ
Acos(2πft + ϕ) = e + e
2 2
Positive frequency Negative frequency
component component

Figure 2.19 Illustration of g(t) = Acos(2𝜋ft + 𝜙) as the sum of two counter-rotating phasors. The sketch shows the initial
orientation of the phasors.

𝜃 = 90∘ , and the upward zero crossing point is at 𝜃 = −90∘ . Note that the first half of the waveform’s cycle spans
𝜃 = 0 to 180, whereas the second half of the cycle spans range −180∘ to 0∘ , so that one full cycle spans a range
of 360∘ .

2.7.2.2 Amplitude
Amplitude A is the maximum value of the signal. This may be read from the mathematical expression for the
sinusoidal signal as the absolute value of the number multiplying the cosine or sine term. For example, the signals
g1 (t) = 4cos(100t), g2 (t) = −4cos(500t + 𝜋/2), g3 (t) = 4sin(t + 15∘ ) all have amplitude A = 4. It is important to note
that amplitude is always a positive real number. The amplitude of a sinusoidal signal may also be read from a
waveform display of the signal (for example on an oscilloscope) as half of the peak-to-peak amplitude App of the
signal, which is the range from the trough of the waveform to its crest, as labelled in Figure 2.18

2.7.2.3 Angular Frequency

Angular frequency 𝜔 is the rate of change of the angle of the sinusoidal signal in rad/s. Given a mathematical
Telegram: @Persian_IEEEComSoc

expression for the signal then 𝜔 may be read straightforwardly as the coefficient of t. For example, the signals g1 (t),
g2 (t), and g3 (t) in the previous paragraph have respective angular frequencies 𝜔 = 100, 500, and 1 rad/s. Angular
frequency may also be determined from a graphical display of the signal’s waveform, as discussed below.

2.7.2.4 Frequency
Frequency f is the number of cycles (of values) completed by the signal each second. It is expressed in Hz and is
related to 𝜔 by Eq. (2.39). The angular frequency 𝜔 is sometimes simply referred to as the frequency, it being taken
for granted that the reader knows that the actual signal frequency is 𝜔/2𝜋. Frequency can therefore be read from a
mathematical expression for the signal by dividing the coefficient of t in the expression by 2𝜋. For example, in the
three signals given above, f = 50/𝜋, 250/𝜋, and 0.5/𝜋 Hz, respectively. Frequency may also be read from a display
or plot of the waveform, as discussed below.
 88 2 Introduction to Signals and Systems

2.7.2.5 Period
Period T is the time taken by the signal to complete one cycle and is therefore the reciprocal of frequency, as in
Eq. (2.2). You may wish to verify that the previous three signals have respective periods T = 62.832 ms, 12.566 ms,
and 6.283 s. On a graphical display of the signal waveform T may be read as the time separation between adjacent
corresponding points on the waveform, for example from one crest to the next crest (as indicated in Figure 2.18), or
from one trough to the next trough, or from one downward zero crossing to the next downward zero crossing, and
so on. Once the period is read in this way then the signal frequency f follows as f = 1/T and the angular frequency
as 𝜔 = 2𝜋f = 2𝜋/T.

2.7.2.6 Wavelength
If the signal represents a travelling or propagating wave then wavelength 𝜆 is the distance (in metres) travelled by
the wave during a time of one period. Thus 𝜆 = vT = v/f , where v is the speed of the wave (in m/s). For example,
if the wave is an acoustic signal then v ≈ 330 m/s in air, 1460 m/s in water, and 5000 m/s in steel. If the wave is an
electromagnetic signal then v ≈ 3 × 108 m/s in air or vacuum. Since the speed of a wave is usually a known constant
in a given medium, it is common practice, especially with electromagnetic waves, to specify the wavelength as an
alternative to frequency. For example, a 10 mm radio wave refers to a sinusoidally varying electromagnetic signal
of frequency
3 × 108 m∕s
f = = 30 GHz
10 × 10−3 m
2.7.2.7 Initial Phase
The initial phase 𝜙 of a sinusoidal signal (often referred to simply as the phase, which we will do from now) is
the angle of the signal at the initial time t = 0. Recall that a phase is always defined with reference to cos(2𝜋f t).
Therefore to read the phase of any sinusoidal signal from its mathematical expression, convert it to the standard
cosine format |A|cos(2𝜋ft + 𝜙), making sure to transfer any phase shift that is hidden in √amplitude to the angle.
For example, a factor of −1 in amplitude implies a ±180∘ phase shift, whereas a factor of −1 implies a phase shift
of +90∘ . So the signal −50cos(200𝜋t + 30∘ ) becomes 50cos(200𝜋t + 30∘ −180∘ ) = 50cos(200𝜋t − 150∘ ) and thus has
a phase of −150∘ ; the signal sin(200𝜋t + 45∘ ) becomes cos(200𝜋t + 45∘ −90∘ ) = cos(200𝜋t ∘
√ − 45 ), where we have
∘ ∘
√ Eq. (2.41) to change sine to cosine. So, the phase of this signal is 𝜙 = −45 . And −10 cos(10 𝜋t − 60 ) =
5
used
∘ ∘
10 cos(10 𝜋t − 60 + 90 ), which means that it has a phase of 30 .
5 ∘
The phase of a sinusoidal signal may also be easily read from a graphical plot of its waveform using the following
steps: (i) look at the crest of the signal that is nearest to the y axis (where t = 0). Note that the choice of the y axis
location is completely arbitrary in an oscilloscope display. (ii) If this crest is to the left of the y axis then the signal
‘started early’ and has a positive phase; but if it is to the right of the y axis then the signal ‘started late’ and therefore
has a negative phase. (iii) Express the separation or gap between the crest and y axis as a fraction of signal period
and multiply this fraction by 360∘ to obtain the magnitude of 𝜙. Equivalently, express this gap as a fraction of one
Telegram: @Persian_IEEEComSoc

cycle and convert to degrees or radians for 𝜙.

For a quick illustration of how to apply these steps, look at Figure 2.20, which features two waveforms, g1 (t) and
g2 (t). Note that Figure 2.20a contains four divisions (div) per cycle, whereas Figure 2.20b has 6 div/cycle. Waveform
g1 (t) has its nearest crest located half a division (= a one-eighth cycle) to the left of the y axis, therefore it ‘started
early’. That is, it ‘jumped the gun’ by a one-eighth cycle when compared to the reference waveform, which starts
‘bang on time’ at t = 0. This means that g1 (t) has ‘an advantage’ manifested in the form of a positive phase 𝜙1
given by
1 degrees
𝜙1 = cycle × 360 = 45∘
8 cycle
Similarly, waveform g2 (t) is ‘disadvantaged’ by starting ‘after the gun’ by two-sixths = a one-third cycle, so it has
a negative phase 𝜙2 = −120∘ . Note that using the nearest crest to the y axis is essential to ensure that the phase
 2.7 Sinusoidal Signals 89

g1(t)
10 V/div

(a) t

Time base = 5 ms/div

g2(t)
1 V/div

(b) t

Time base = 10 μs/div

Figure 2.20 Reading a sinusoidal waveform display.

determined lies in the range −180∘ to +180∘ . If a waveform has two equidistant crests either side of the y axis then
it does not matter which one you choose, since in that case it means that the phase is −180∘ , which is the same
angle as +180∘ .

2.7.2.8 Phase Difference

Phase difference is the difference between the phases of two sinusoidal signals of the same frequency (and period).
Phase difference can be compared to the distance or gap between two cars travelling on a motorway. Unless the two
cars travel at the same velocity, the gap between them changes with time. If two sinusoidal signals have different
frequencies then their phase relationship changes continuously. The higher frequency sinusoid ‘overtakes’ the
lower frequency sinusoid repeatedly along the ‘circular motorway’, and phase difference is therefore not defined.
When the two waveforms have the same frequency then the phase difference between them is defined and is fixed
at the same value irrespective of any arbitrary choice of y axis location (the point at which time t is set to 0). Phase
Telegram: @Persian_IEEEComSoc

difference provides a means of timing the oscillations of one sinusoidal signal relative to another, allowing us to
say whether one signal leads or lags the other and by how many degrees.
Given the mathematical expressions of two sinusoidal signals of the same frequency, it is a straightforward
matter to determine their phase difference by subtracting their phases. The signal with the larger phase is the
leading signal. Phase difference may also be read from a graphical plot or oscilloscope display of both waveforms.
In this case it is given by the angular separation between the closest corresponding points (e.g. crests) of the two
waves. The leading signal is the one whose crest is to the left since this is the signal that ‘started earlier’. As an
example, consider the three waveforms B, C, and D displayed in Figure 2.21. The y axis is deliberately not shown
since its location does not affect the reading of phase difference. Using the closest separation of crests and noting
that there are four divisions per cycle (i.e. per 360∘ ), we read the phase differences as follows: waveform D leads B by
45∘ , B leads C by 90∘ , and D leads C by 135∘ . Of course, you may express these readings equivalently as ‘waveform
 90 2 Introduction to Signals and Systems

D B C

Figure 2.21 Phase difference.

B lags D by 45∘ ’, and so on; but in every case, you must not just specify the magnitude of phase difference but also
identify which is the leading or lagging signal.

Worked Example 2.5

Given the voltage signal v(t) = 20cos(314.2t − 30∘ ) volt, determine

(a) The period of v(t).
(b) The instantaneous value of v(t) at t = 10 ms.
(c) The phase difference between v(t) and the current signal.

i(t) = 200sin(314.2t + 70∘ ) mA

(a) Comparing v(t) with the standard sinusoidal expression
v(t) = 20 cos(314.2t − 30∘ ) ≡ A cos(𝜔t + 𝜙)
we see that 𝜔 = 314.2 rad/s. Therefore
T = 1∕f = 2𝜋∕𝜔 = 20 ms
(b) Before evaluating the cosine function to obtain the instantaneous value of voltage, we need to express t in
seconds and the argument of the cosine function in radians.
t = 10 ms = 0.01 s; 30∘ = 30 × +𝜋∕180 rad = 𝜋∕6 rad
Therefore
v(0.01) = 20 × cos(314.2 × 0.01 − 𝜋∕6)
= 20 × cos(2.618) = 20 × (−0.866)
= − 17.32 V
Telegram: @Persian_IEEEComSoc

(c) To determine the phase difference between v(t) and i(t), we first obtain their phases with respect to a common
reference waveform, which of course must have the same frequency. Choosing cos(314.2t) as the reference, we
must convert i(t) from sine format to cosine using the fact that the sine function lags the cosine by 90∘ . Thus
i(t) = 200 sin(314.2t + 70∘ ) = 200 cos(314.2t + 70∘ − 90∘ )
= 200 cos(314.2t − 20∘ )
v(t) = 20 cos(314.2t − 30∘ )
That is
Phase of i(t) = −20∘
Phase of v(t) = −30∘
Phase difference = −20 − (−30) = 10∘ , with i(t) leading v(t).
 2.7 Sinusoidal Signals 91

Worked Example 2.6

Sketch three cycles of the sinusoidal waveforms

(a) v1 (t) = 10 sin(2𝜋t − 𝜋/2).
(b) v2 (t) = 20 sin(4𝜋t + 18∘ ).

Rough accuracy is enough in this case, so we may follow three straightforward steps to produce a quick sketch of
the waveforms without evaluating the sine functions: (i) first, sketch three cycles of sinusoidal variation having
the correct amplitude and period; (ii) place the y axis into your sketch at the correct displacement from the middle
crest of your sketch so that the waveform has the correct phase; (iii) label the horizontal axis with time values
starting with t = 0 at the location of the y axis. That’s all.
(a) v (t) = 10 sin(2𝜋t − 𝜋∕2) = 10 cos(2𝜋t − 𝜋), so it has amplitude A = 10, period T = 1 s and phase 𝜙 = −180∘ .
1 1 1 1
Figure 2.22a shows the first step, a sketch of three cycles of a sinusoid having amplitude 10 and completing
one cycle in 1 s. Figure 2.22b shows the second step in which the y axis is placed half a cycle to the left of
the penultimate crest to give the waveform its 180∘ ‘late start’, i.e. phase of 𝜙1 = −180∘ . The choice of which
crest to use is guided solely by the need to centre the y axis as much as possible. The final step is shown in
Figure 2.22c, where the time axis is labelled either side of the y axis, guided by the fact that one cycle is 1 s so
half a cycle is 0.5 s. Grid lines have been included in Figure 2.22c. This is optional but very helpful for reading
the waveform.
(b) v2 (t) = 20 sin(4𝜋t + 18∘ ) = 20 cos(4𝜋t − 72∘ ), so, it has amplitude A2 = 20, period T 2 = 0.5 s, and phase
𝜙2 = −72∘ . We proceed as in (a) using the new values of amplitude and period, but this time we place the y
axis 72∘ (= 72/360 of a cycle = one-fifth of a cycle = 0.1 s) to the left of the second crest to correctly represent
T = 1s
10

(a) 0 t

–10
T = 1s ʋ1(t)
10

(b) 0 t

–10
ʋ1(t)
10
Telegram: @Persian_IEEEComSoc

(c) 0 t

10
–1.5 –1.0 –0.5 0 0.5 1.0 1.5 → t(s)
ʋ2(t) T = 0.5s
20

(d) 0 t

–20
–0.4 –0.15 0 0.1 0.35 0.6 0.85 1.1 → t(s)

Figure 2.22 Worked Example 2.6.

 92 2 Introduction to Signals and Systems

the fact that v2 (t) started late by 72∘ . We then label the time axis either side of the y axis by 0.25 s per half
cycle. The resulting sketch is shown in Figure 2.22d.
The above method does not allow multiple waveforms to be sketched on the same graph since the location of
the y axis (t = 0) differs for each waveform. To have a common y axis located on the left of the sketch (at t = 0) for
each waveform, use the following alternative method to sketch n cycles of a sinusoidal signal g(t) of amplitude A,
period T, and phase 𝜙: (i) sketch the sinusoidal cycles, starting from its crest (y-value = A) at t = (|𝜙∘ |/360)T
if 𝜙 is negative, or t = (1 − 𝜙∘ /360)T if 𝜙 is positive, continuing the cycles to the right until you reach the point
(t, y) = (nT, g(nT)). (ii) Continue the cycles to the left until you intercept the y axis at (t, g(0)), and label the
graph axes.
Figure 2.23 illustrates the use of the steps of this second method to sketch the above waveforms in the interval
t = 0 to 3T.

T2 T1 T1

T1 = 1s; T2 = 0.5s

Step (i) t

T2/5 T1/2 3T2 3T1

20
ʋ2(t) = 20sin(4πt + 18°)
ʋ1(t) = 10sin(2πt – π/2)
10

Step (ii) 0 t

–10

–20
0 0.5 1.0 1.5 2.0 2.5 3.0 → t(s)

Figure 2.23 Worked example 2.6 – Alternative method.

Telegram: @Persian_IEEEComSoc

2.7.3 Addition of Sinusoids

The addition of two or more signals (in a summing device or receiver) occurs very often in telecommunication.
We consider four possibilities that may arise in the addition of sinusoidal signals:
(i) Sinusoids with the same frequency and phase.
(ii) Sinusoids with the same frequency but different phases.
(iii) Multiple sinusoids with different frequencies.
(iv) Two sinusoids producing a beat frequency.
 2.7 Sinusoidal Signals 93

2.7.3.1 Same Frequency and Phase

When two or more sinusoids of the same frequency and phase are added together, the result is another sinusoid of
the same frequency and phase, but of amplitude equal to the sum of the amplitudes of the added sinusoids. This
is by far the simplest scenario. The task is in fact trivial. Mathematically
A1 cos(𝜔t + 𝜙) + A2 cos(𝜔t + 𝜙) + · · · = [A1 + A2 + · · ·] cos(𝜔t + 𝜙)
It is, however, important to avoid a common mistake and note that
A1 cos(𝜔t + 𝜙) + A2 sin(𝜔t + 𝜙) ≠ [A1 + A2 ] cos(𝜔t + 𝜙)
The sinusoids on the left-hand side clearly differ in phase by 90∘ .

2.7.3.2 Same Frequency but Different Phases

The result of adding two or more sinusoids of the same frequency f but different phases is another sinusoid of
frequency f . That is
A1 cos(2𝜋ft + 𝜙1 ) + A2 cos(2𝜋ft + 𝜙2 ) + · · · + AM cos(2𝜋ft + 𝜙M ) = A cos(2𝜋ft + 𝜙)
The task here is to determine the amplitude A and phase 𝜙 of the resultant sinusoid on the right-hand side. Since
each component sinusoid on the left-hand side oscillates at the same frequency, their relative phase differences
are fixed in time. This means that we may carry out the summation by combining the phasors Ai ∠𝜙i , i = 1, 2, 3,
…, M of the component sinusoids (which show the amplitude and phase of each sinusoid) to obtain a resultant
phasor A∠𝜙 that represents the amplitude and phase of the required resultant sinusoid. A graphical approach may
be used for phasor addition and subtraction, as described in Worked Example 2.7, for the case of two sinusoids.
More than two sinusoids may be added by following the same procedure and combining two sinusoids at a time.
The graphical method, though useful for underlining fundamental principles, is quite cumbersome when there
are many component sinusoids. An alternative approach which we now discuss treats the phasors as complex
numbers and can be readily implemented as a computer program to handle any number of components.
The combination (through addition or subtraction) of M (≥ 2) sinusoids of the same frequency to obtain a single
resultant sinusoid may be easily done by taking the following steps:
(i) Express each of the M sinusoids in cosine format. That is, change any sinusoid in the sine format
Ak sin(2𝜋ft + 𝛽 k ) to Ak cos(2𝜋ft + 𝛽 k − 90∘ ). This makes it obvious that the phase of this sinusoid is actually 𝛽 k
− 90∘ .
(ii) Change subtraction to addition by adding 180∘ to the phase 𝛼 k of any subtracted sinusoid if 𝛼 k is negative or
subtracting 180∘ if 𝛼 k is positive. Note that it does not make any difference to your final result if you introduce
the 180∘ by adding or by subtracting. We are simply being so specific in order to adhere to the convention of
keeping angles in the range −180∘ to +180∘ . Once you have done this, treat the sinusoid as an added term.
This means that −Ak cos(2𝜋ft + |𝛼 k |) becomes + Ak cos(2𝜋ft + |𝛼 k | − 180∘ ), so it is now simply a case of adding
Telegram: @Persian_IEEEComSoc

a sinusoid of amplitude Ak and phase 𝜙k = |𝛼 k | − 180∘ . Similarly, −Ak cos(2𝜋ft − |𝛼 k |) becomes + Ak cos(2𝜋ft −
|𝛼 k | + 180∘ ). These first two steps simplify the overall task to that of evaluating the left-hand side of Eq. (2.50)
below to obtain the single sinusoid on the right-hand side
∑
M
Ak cos(2𝜋ft + 𝜙k ) = A cos(2𝜋ft + 𝜙) (2.50)
k=1

(iii) Express each sinusoid obtained at the end of step (ii), i.e. each sinusoid on the left-hand side of Eq. (2.50) as a
phasor Ak ∠𝜙k and resolve each phasor into its in-phase and quadrature components, AkI and Akq , respectively.
This step is illustrated in Figure 2.24a.
 94 2 Introduction to Signals and Systems

(iv) Sum all the in-phase components to obtain the in-phase component AI of the resultant sinusoid. Similarly,
sum the quadrature components to obtain Aq .
(v) Obtain the required resultant amplitude A and resultant phase 𝜙 on the right-hand side of Eq. (2.50) using
the following equations, making sure to place 𝜙 in the right quadrant depending on the signs (positive
or negative) of AI and Aq , as clearly illustrated in Figure 2.24b. Note that in the following equations we
use Pythagoras’s theorem to obtain A, and we define an acute positive angle 𝛼 based on the magnitudes
(i.e. absolute values) of AI and Aq , denoted |AI | and |Aq |, respectively. We are then guided by the signs of
AI and Aq to place 𝛼 in the right quadrant to obtain the required resultant phase 𝜙. Note also that angles
are positive going counterclockwise from the reference +x direction (where angle is 0∘ ) and negative
when going clockwise from +x. Thus, angles in the first and second quadrants have a positive value
in the range 0∘ to 180∘ , whereas angles in the third and fourth quadrants have a negative value in the
range −180∘ to 0∘ .

√
A= A2I + A2q
∑
M
∑
M
where, AI = Ak cos 𝜙k ; Aq = Ak sin 𝜙k (2.51)
k=1 k=1

⎧ ∘
⎪0 , Aq = 0, AI = 0
⎪90∘ , Aq > 0, AI = 0
⎪
⎪−90∘ , Aq < 0, AI = 0
⎪
𝜙 = ⎨𝛼, Aq ≥ 0, AI > 0
⎪
⎪180 − 𝛼, Aq ≥ 0, AI < 0
⎪𝛼 − 180, Aq < 0, AI < 0
⎪
⎪−𝛼, Aq < 0, AI > 0
⎩
where, 𝛼 = tan−1 (|Aq ∕AI |) (2.52)

As demonstrated in Worked Example 2.8, this method is quicker and more straightforward than its equations
might suggest at first glance.

Worked Example 2.7

Obtain the sum of the sinusoidal voltages v1 (t) = 3sin(𝜔t) and v2 (t) = 4sin(𝜔t − 𝜋/3) volts
Telegram: @Persian_IEEEComSoc

The sum voltage is

v(t) = 3 sin(𝜔t) + 4 sin(𝜔t − 𝜋∕3)
= A sin(𝜔t ± 𝛼).
To determine A and 𝛼, we first represent v1 (t) as a phasor A1 ∠𝜙1 – a straight line of length equal (or scaled) to
the amplitude A1 of v1 (t) and of orientation equal to the phase 𝜙1 of v1 (t). Since v1 (t) = 3sin(𝜔t) = 3cos(𝜔t − 90∘ ),
it follows that A1 = 3 V and 𝜙1 = −90∘ . Similarly, v2 (t) = 4sin(𝜔t − 𝜋/3) = 4cos(𝜔t − 60∘ − 90∘ ) = 4cos(𝜔t − 150∘ ),
so that v2 (t) is represented by phasor A2 ∠𝜙2 of amplitude A2 = 4 V and phase 𝜙2 = −150∘ . Next, we draw the
phasor A2 ∠𝜙2 starting at the end of A1 ∠𝜙1 . The phasor A∠𝜙 obtained by joining the starting point of A1 ∠𝜙1 to
the endpoint of A2 ∠𝜙2 gives the resultant voltage v(t) in both amplitude A and phase 𝜙. This phasor diagram is
shown in Figure 2.25a. Note from the diagram that the desired phase is 𝜙 = −(90 + 𝛿)∘ .
 2.7 Sinusoidal Signals 95

(a)

Ak
Akq = Ak sin (ϕk )

ϕk
AkI = Ak cos (ϕk )

q
(b)
2nd quadrant: AI negative, Aq positive 1st quadrant: Both AI and Aq positive

A A
Aq ϕ Aq
AI ϕ
AI I
AI
AI ϕ
Aq ϕ Aq
A A

3rd quadrant: Both AI and Aq negative 4th quadrant: AI positive, Aq negative

Figure 2.24 (a) Resolving a phasor Ak ∠𝜙k into its in-phase component AkI and quadrature component Akq ; (b) Determining
resultant amplitude A and phase 𝜙.

ϕ
δ
ϕ′
A′
3
A
(a) (c) 3 –A2∠ϕ2
β 4
A1∠ϕ1
150°
Telegram: @Persian_IEEEComSoc

4
β = 360 – (150 + 90)
= 120°

C
(b) b d

D B
c

Figure 2.25 (a) Phasor diagram for v(t) = v 1 (t) + v 2 (t) in Worked Example 2.7; (b) Naming convention for angles and sides in
Eq. 2.44; (c) Phasor diagram for v 1 (t) − v 2 (t).
 96 2 Introduction to Signals and Systems

To solve the triangle for A and 𝜙, we first use the cosine rule to obtain A and then the sine rule to obtain 𝜙.
The cosine rule is used to solve a triangle (as in this case) where two sides and an included angle are known.
The sine rule is simpler and is applicable in all other situations where up to three parameters of the triangle are
known, including at least one side and at least one angle. In terms of the naming convention shown in Figure 2.25b,
we have
Cosine Rule∶ d2 = b2 + c2 − 2bc cos(D)
sin(B) sin(C) sin(D)
Sine Rule∶ = = (2.53)
b c d
Thus
A2 = 32 + 42 − 2 × 3 × 4 × cos(120∘ )
= 37
A = 6.08 volts
And
sin 𝛿 sin(120∘ )
=
4 6.08
which evaluates to 𝛿 = 34.73∘ and therefore 𝜙 = −90 − 34.73 = −124.73∘ .
Hence, the resultant voltage is the sinusoid
v(t) = 6.08 cos(𝜔t − 124.73∘ ) = 6.08 sin(𝜔t − 34.73∘ ) volts
It is worth noting that this graphical approach may be used just as easily for the subtraction of sinusoids by
reversing the direction (but maintaining the same line) of the phasor of the sinusoid to be subtracted. To illustrate,
Figure 2.25c shows the phasor diagram for subtracting v2 (t) from v1 (t) to obtain a resultant amplitude A′ and
phase 𝜙 . Notice that in this case we move from the end of phasor A1 ∠𝜙1 in the opposite direction to A2 ∠𝜙2 ,
′

which amounts to subtracting A2 ∠𝜙2 .

Worked Example 2.8

Using the analytic approach of Eqs. (2.51) and (2.52), obtain a sinusoidal expression for g(t) = 10sin(20t + 45∘ )
− 20cos(20t) + 12cos(20t + 120∘ ).

To solve this problem, we simply follow the steps of the approach, as discussed above and identified below.
g(t) = 10 sin(20t + 45∘ ) − 20 cos(20t) + 12 cos(20t + 120∘ )
= 10 cos(20t − 45∘ ) − 20 cos(20t) + 12 cos(20t + 120∘ ) Step (i)
∘ ∘ ∘
= 10 cos(20t − 45 ) + 20 cos(20t + 180 ) + 12 cos(20t + 120 ) Step (ii)
Telegram: @Persian_IEEEComSoc

AI = 10 cos(−45∘ ) + 20 cos 180∘ + 12 cos 120∘ ⎫

⎪
= 7.0711 − 20 − 6 = −18.929 ⎪
⎬ Steps (iii) & (iv)
Aq = 10 sin(−45∘ ) + 20 sin 180∘ + 12 sin 120∘ ⎪
= −7.0711 + 0 + 10.3923 = 3.321 ⎪
⎭
√
A = A2I + A2q = 19.22 ⎫
⎪
𝛼 = tan−1 (|A ∕A |) = 9.95∘
q I ⎬ Step (v)
∘ ⎪
𝜙 = 180 − 𝛼 = 170.05 , since AI < 0, Aq > 0 ⎭
Thus, g(t) = 19.22 cos(20t + 170.05∘ )
 2.7 Sinusoidal Signals 97

ʋ1(t) = 3sin(2πf1t)

ʋ(t) = ʋ1(t) + ʋ2(t) is periodic

but not sinusoidal

ʋ2(t) = sin(6πf1t)

Figure 2.26 The sum of harmonically related sinusoids is a periodic signal.

2.7.3.3 Multiple Sinusoids of Different Frequencies

When two or more sinusoids of different frequencies are added together, the resulting waveform is no longer a
single sinusoid, as in the previous cases. There is no general analytic method for obtaining the resultant waveform
in this case. The sinusoids may be added graphically by plotting them over the desired time interval and adding
the instantaneous values of the sinusoids at each point to obtain the value of the resultant waveform at that point.
However, if the sinusoidal signals are harmonically related (i.e. the frequency of each of the sinusoids is an integer
multiple of one fundamental frequency) then the resultant waveform is a periodic signal. Figure 2.26 shows an
example of adding two sinusoids at frequencies f 1 and 3f 1 and respective amplitudes A1 > A2 . The case of A1 < A2
would also produce a periodic waveform, albeit of a different shape.

2.7.3.4 Beats Involving Two Sinusoids

The addition of two sinusoids at different frequencies f 1 and f 2 is a very special case that produces a phenomenon
known as beats which has extensive applications in medicine, music, law enforcement, telecommunications, etc.
In law enforcement, for example, the police may determine the speed of a moving vehicle by observing the beats
between a sinusoidal radio signal transmitted by their radar gun and the sinusoidal signal received back by reflec-
tion from the target. The difference in frequency is in general referred to as beat frequency, but in scenarios
involving relative motion between a signal source and a signal observer, it is usually called Doppler frequency
in honour of the Austrian physicist Christian Andreas Doppler (1803–1853) who, in 1842, explained the impact of
relative speed on observed frequency.
To fully understand the phenomenon of beats it is helpful to derive a general expression for the sum of two
sinusoids having equal amplitude but different frequencies. Consider the trigonometric identity in Eq. (B.6) of
Telegram: @Persian_IEEEComSoc

Appendix B
1
cos A cos B = [cos(A + B) + cos(A − B)] (i)
2
Let us make the following substitutions in the right-hand side

A + B = 𝜃1 (ii)
A − B = 𝜃2 (iii)

Adding together equations (ii) and (iii) yields

𝜃1 + 𝜃2
A= (iv)
2
 98 2 Introduction to Signals and Systems

whereas subtracting (iii) from (ii) yields

𝜃1 − 𝜃2
B= (v)
2
Substituting Eq. (ii) to (v) into (i) yields a trigonometric identity for the sum of two sinusoids as follows
( ) ( )
𝜃1 − 𝜃2 𝜃1 + 𝜃2
cos(𝜃1 ) + cos(𝜃2 ) = 2 cos cos
2 2
Now replacing 𝜃 1 with 2𝜋f 1 t and 𝜃 2 with 2𝜋f 2 t we obtain an expression for the sum of two sinusoidal signals of
frequencies f 1 and f 2
[ ( )] ( )
f1 − f2 f1 + f2
cos(2𝜋f1 t) + cos(2𝜋f2 t) = 2 cos 2𝜋 t cos 2𝜋 t (2.54)
2 2
If f 1 and f 2 are close in value then the right-hand side will appear as a sinusoid of frequency equal to the
average value (f 1 + f 2 )/2 having a slowly varying amplitude given by the term in square brackets. This term is
therefore referred to as the envelope of the sum signal. Variation in the envelope occurs due to a constantly chang-
ing phase difference between the two sinusoids. The higher frequency sinusoid oscillates faster, ‘overtaking’ the
lower-frequency sinusoid by exactly f 1 − f 2 times each second. At the point of ‘overtaking’, their two phasors
are momentarily coincident so that the two signals directly add to create a strong resultant signal. On the other
hand, when the two phasors are half a cycle apart, the two signals cancel through direct subtraction, giving rise
to a weak resultant signal. This interaction produces a periodic variation in resultant amplitude or intensity, or
perceived loudness known as beat at a frequency equal to f 1 − f 2 .
Figure 2.27 shows the beats produced when two sinusoids of frequencies 50 Hz and 46 Hz and equal amplitude
are summed. Notice how the envelope varies periodically and the peak points are spaced apart by 0.25 s which is
therefore the beat period, so that the beat frequency is 4 Hz, i.e. 50 – 46 Hz as expected.
To comment further on applications, the phenomenon of beat frequency is relied upon in music for tuning
instruments. To tune an instrument producing a tone at frequency f 1 to a slightly different tone at frequency
f 2 , you play the two sounds together and listen to the beats while tuning. You know that the two sounds have
equal frequency when the beats disappear. However, for the variation in sound volume (i.e. beats) to be audibly
discernible the frequency difference must be small, for example no more than about 20–30 Hz; otherwise, only a
tone at the average frequency will be perceived.

H L H L H L H L H
Beat Sum
envelope signal
Telegram: @Persian_IEEEComSoc

0 0.25 0.5 0.75 1 → t(s)

Beat period

Figure 2.27 Summing two sinusoids at frequencies f 1 = 50 Hz, f 2 = 46 Hz produces a beat frequency f 1 − f 2 = 4 Hz due to
the amplitude of the sum signal varying periodically between high (H) and low (L).
 2.8 Logarithmic Units 99

2.7.4 Multiplication of Sinusoids

Multiplication of sinusoids finds numerous applications in telecommunication and takes place, for example, in a
mixer circuit that translates a signal from one radio frequency to another. It is important to become thoroughly
familiar with the trigonometric identities presented in Appendix B, since they will feature prominently in our sub-
sequent study. Note the identities (B.5) to (B.7) for the product of two sinusoids. Therefore, consider two sinusoidal
signals at frequencies f 1 and f 2
v1 (t) = A1 cos(2𝜋f1 t)
v2 (t) = A2 cos(2𝜋f2 t)
Their product follows from Eq. (B.6) with A = 2𝜋f 1 t and B = 2𝜋f 2 t
v(t) = v1 (t)v2 (t)
AA
= 1 2 {cos[2𝜋(f2 − f1 )t] + cos[2𝜋(f2 + f1 )t]}
2
This result shows that multiplying two sinusoids generates two new frequencies at the sum and difference of
the original frequencies. A study of some of the applications of this principle is covered in later chapters.

2.8 Logarithmic Units

A signal is subject to gains and losses at various stages of transmission through a communication system. To deter-
mine the signal power at a given point in the system, one multiplies the input signal power by all the power gains
and divides by all the power losses experienced by the signal up to that point. This is a linear-units procedure
involving multiplication and division. Signal power is in watt (W) and gains and losses are dimensionless num-
bers greater than 1. In Figure 2.28, an input power Pin = 3.16 mW enters the transmission system. The power at
various points in the system is
PA = 3.16 mW × 10 = 31.6 mW
PB = 3.16 mW × 10 ÷ 63.1 = 0.50 mW
Pout = 3.16 mW × 10 ÷ 63.1 × 10 = 5.01 mW
We can simplify the process of signal power computation in a transmission system by adopting units of mea-
sure of power, gains, and losses that transform a complex series of multiplication and division into additions and
subtractions. Logarithmic units furnish this transformation.
The logarithm of a number (to base 10), denoted log10 , is the power to which 10 is raised to obtain that number.
For example, because 102 = 100, we say that the logarithm of 100 is 2 and because 100 = 1, we say that the logarithm
of 1 is zero, which we write as log10 100 = 2, and log10 1 = 0.
Telegram: @Persian_IEEEComSoc

Consider two positive numbers A and B whose logarithms (to base 10) are x and y, respectively. We may write
log10 (A) = x (a)
log10 (B) = y (b) (2.55)

Pin PA PB
Gain Loss Gain
Pout
× 10 ÷ 63.1 × 10
3.16 mW

Figure 2.28 Gains and losses in transmission system: Linear-units procedure.

 100 2 Introduction to Signals and Systems

It follows that
10x = A (a)
y
10 = B (b) (2.56)

Multiplying Eq. (2.56) (a) and (b) together

10x 10y = 10x+y = AB

And it follows by the above definition that

log10 (AB) = x + y = log10 (A) + log10 (B)

Note that we have made use of Eq. (2.55).

Now dividing Eq. (2.56) (a) by (b)
10x A
= 10x−y =
10y B
It similarly follows that
( )
A
log10 = x − y = log10 (A) − log10 (B)
B
To summarise, the following relations hold for logarithms to any base
log(AB) = log(A) + log(B)
( )
A
log = log(A) − log(B) (2.57)
B
Putting A = 1
( )
1
log = log(1) − log(B) = 0 − log(B)
B
= − log(B) (2.58)

Observe that multiplication (A × B) is replaced by the addition of logarithms, division (A/B) by subtraction of
logarithms, and inversion (1/B) by changing the sign of the logarithm. For example
log10 (103 ) = log10 (10 × 10 × 10) = log10 (10) + log10 (10) + log10 (10)
= 3 × log10 (10) = 3

In general

log(Ab ) = b log(A) (2.59)

Telegram: @Persian_IEEEComSoc

where b is any real number.

In logarithmic units therefore, the output signal power of a transmission system is obtained by adding the system
gains to the input power and subtracting the system losses. Of course, every quantity, including both the input and
output powers, must be expressed in logarithmic units. The most used logarithmic unit in the field of engineering
is the decibel (dB). The decibel unit is one-tenth of a bel (B), a logarithmic unit that is no longer in use.
It is common to use gain as a generic term for both a boost in signal strength (which is actual gain) and a
reduction in signal strength (which is actual loss). In logarithmic units such as the decibel, a positive gain then
indicates an increase in signal strength, whereas a negative value for gain indicates a loss or reduction in signal
strength. So, for example, we may refer to a gain of 18 dB or a gain of −10 dB. The latter means a loss of 10 dB, but we
would not normally refer to the former as a loss of −18 dB even though such a description would be mathematically
correct.
 2.8 Logarithmic Units 101

Figure 2.29 System Gain. I1 I2

V1, P1 G V2, P2

Z1 Z2

2.8.1 Logarithmic Units for System Gain

Figure 2.29 represents an arbitrary system of total gain G and input power P1 , input current I 1 , input voltage V 1 ,
output power P2 , output current I 2 , and output voltage V 2 . The input and output resistances of the system are Z 1
and Z 2 , respectively.
Various gains of the system, expressed as dimensionless ratios, are
P2
Power gain =
P1
I
Current gain = 2
I1
V2
Voltage gain = (2.60)
V1
The power gain of the system in dB is defined as
( )
P2
G = 10log10 dB (2.61)
P1
If dealing with normalised power, or if the input and output resistances of the system are equal (i.e. Z 1 = Z 2 ),
then
P2 V2 I2
= 22 = 22
P1 V1 I1
Substituting in Eq. (2.61), we obtain
( 2) ( )2
V2 V2
G = 10log10 2
= 10log10
V V1
( 1)
V2
= 20log10
V
( 1)
I
= 20log10 2 (2.62)
I1
Telegram: @Persian_IEEEComSoc

Equations (2.61) and (2.62) show that, for power gain in dB, the constant of multiplication is 10, whereas for
current and voltage gains in dB, the constant of multiplication is 20. This difference is extremely important and
must always be remembered to avoid errors. It is worth emphasising that Eq. (2.61) for power gain does not depend
on the values of system resistances, whereas Eq. (2.62) for voltage and current gains holds only if the system’s input
and output resistances are equal.
A less commonly used logarithmic unit of gain is the neper (Np), defined as the natural logarithm of the ratio
of output to input. This is logarithm to base e = 2.718281828459· · ·, denoted ln. It follows that for the system in
Figure 2.29
( ) ( ) ( )
V2 I 1 P2
Gain in neper (Np) = ln = ln 2 = ln (2.63)
V1 I1 2 P1
 102 2 Introduction to Signals and Systems

To obtain the relationship between Np and dB, note in Eq. (2.63) that a current or voltage gain of 1 Np, implies
that
( )
V2 V
loge = 1; or 2 = e1 = e
V1 V1
From Eq. (2.62), the corresponding dB gain is
( )
V2
20log10 = 20log10 (e) = 8.686 dB
V1
Similarly, a power gain of 1 Np means that
( )
1 P2 P
loge = 1, or 2 = e2
2 P1 P1
The corresponding dB gain follows from Eq. (2.61)
( )
P2
10log10 = 10log10 (e2 ) = 20log10 (e) = 8.686 dB
P1
The logarithmic units neper and decibel are therefore related as follows

Voltage, current, or power gain of 1 Np = 8.686 dB (2.64)

2.8.2 Logarithmic Units for Voltage, Power, and Other Quantities

The decibel and neper are units of relative measure. For example, in Eq. (2.61) the decibel value gives a measure
of P2 relative to P1 . By selecting a universally agreed reference level, we can express any signal power or voltage
in dB relative to the agreed reference. In this way, absolute power level can be measured in dB and Np. Standard
power reference levels are 1 W and 1 mW. Power expressed in dB relative to 1 W is said to be measured in dBW,
whereas power expressed in dB relative to 1 mW is said to be measured in dBm.
Logarithmic units for voltage measurement are designated dBV for a 1 V reference and dBu for a 775 mV refer-
ence. The 775 mV reference, often used in telephony, is the voltage that gives a power dissipation of 1 mW across
a 600 Ω resistance. Thus
( )
P
P (watt) = 10log10 dBW = 10log10 P dBW
(1W )
P
= 10log10 dBm = 30 + 10log10 P dBm (2.65)
1 × 10−3 W
And
Telegram: @Persian_IEEEComSoc

V(volt) = 20log10 (V) dBV

( )
V
= 20log10 dBu
775 × 10−3
= 2.214 + 20log10 V dBu (2.66)

Note that to convert power expressed in dBW to dBm, you simply add 30 to the dBW value; and to convert from
dBV to dBu, you add 2.214 to the dBV value.
Many other quantities may be expressed in logarithmic units in order to simplify computations. This practice is
particularly common in communication link analysis and design where signal and noise powers are products of
contributory parameters. By expressing every parameter in logarithmic units, these powers are more conveniently
computed through summation; and the computation process may be laid out in a tabular format known as a link
 2.8 Logarithmic Units 103

power budget. For example, we learn in Chapter 6 that noise power is given by

Pn = kTB (2.67)

where k = 1.38 × 10−23 J/K or 1.38 × 10−23 W/Hz/K ≡ 1.38 × 10−20 mW/Hz/K is Boltzmann constant, T is equivalent
noise temperature in kelvin (K), and B is bandwidth in Hz. Noise power per unit bandwidth, denoted N o , is given
by
Pn
No = = kT (W∕Hz) (2.68)
B
Expressing k in dB relative to 1 W/Hz/K gives it a value k = 10log10 (1.38 × 10−23 ) = −228.6 dBW/Hz/K. If
expressed in dB relative to 1 mW/Hz/K, then its value is k = −198.6 dBm/Hz/K. Similarly, T may be expressed
in a unit of dB relative to 1 K, called dBK, and bandwidth in dB relative to 1 Hz, known as dBHz. For example, if
T = 500 K and B = 2 MHz, then in logarithmic units

T = 10log10 (500) = 27 dBK

B = 10log10 (2 × 106 ) = 63 dBHz
Pn = k + T + B = −228.6 + 27 + 63 = −138.6 dBW

Note that if the Boltzmann constant is expressed in dBm/Hz/K, then the resulting noise power is in dBm, which
in this example would be −108.6 dBm.
As a further application, the energy per bit Eb at the output of a receive-antenna in a free space line-of-sight
communication link scenario is given by the expression
Pt Gt Gr
Eb = W∕(bit∕s) (2.69)
Ls La Rb
We may express all the above parameters in logarithmic units, namely transmit signal power Pt in dBW,
transmit-antenna gain Gt in dB, receive-antenna gain Gr in dB, free space path loss Ls in dB, additional losses La
in dB, and link bit rate Rb in dBbit/s to obtain Eb in dBW/(bit/s) and hence the important ratio Eb /N o in dB as
Eb
= Pt + Gt + Gr − Ls − La − k − T (dB) (2.70)
No
It is important to note that we are here following the standard practice of employing the same notation for a
parameter whether it is expressed in linear or logarithmic units. For example, transmit signal power is denoted Pt
whether it is in watts (W) as in Eq. (2.69) or in dBW as in Eq. (2.70). You must therefore be careful to recognise,
based on each usage context, which unit (linear or logarithmic) is in play. In general, if an expression involves
the product of multiplicative parameters, as in Eq. (2.69) then the units are linear, but if it is the sum of such
parameters, as in Eq. (2.70) then logarithmic units are involved.
Telegram: @Persian_IEEEComSoc

A psophometer is often used in speech telephony to measure the amount of power in noise and crosstalk.
Although the noise spectrum spans the entire receiver bandwidth, the human ear is less sensitive to some of the
spectral components, which will therefore have a less annoying effect. Human ear sensitivity is greatest between
about 500 and 2000 Hz. The psophometer weights the noise spectrum to take account of the non-flat frequency
response of the ear and the receiving equipment. It reduces noise power at each spectral frequency point in pro-
portion to the reduced sensitivity of the ear at that point. The weighting has a peak at 800 Hz and gives a noise
measurement that is smaller than would be the case in the absence of weighting, but which gives a better indica-
tion of how a human recipient perceives the noise. When the psophometer is used, a suffix ‘p’ is added to whatever
unit is employed. Thus, we may have dBmp, pWp, etc. for psophometrically weighted dBm and picowatt, respec-
tively. Psophometrically weighted noise power is less than unweighted white noise power by 2.5 dB over a 3.1 kHz
bandwidth and by 3.6 dB over a 4 kHz bandwidth. White noise is discussed later in the book.
 104 2 Introduction to Signals and Systems

2.8.3 Logarithmic Unit Dos and Don’ts

The following relations apply to the logarithmic units for voltage and power
PdBm = PdBW + 30
VdBu = VdBV + 2.214
PdBmp = PdBm − 2.5, over 3.1 kHz
PdBmp = PdBm − 3.6, over 4 kHz (2.71)
The way to interpret the above equations is, for example, that you add 30 to the dBW value of power to obtain
its value expressed in dBm, and so on.
Based on Eqs. (2.65) and (2.66), we can convert a power value PdBW (in dBW) to PW (in watts); a power value
PdBm (in dBm) to PmW (in mW), and a voltage value V dBV (in dBV) to V v (in volts), using the following relations
PW = 10(PdBW ∕10)
PmW = 10(PdBm ∕10)
Vv = 10(VdBV ∕20) (2.72)
It is worth taking the time to familiarise yourself with the following checklist of the dos and don’ts of logarithmic
units to avoid common pitfalls:
● Do not add together two dBW (or dBm) values. Doing so would amount to multiplying two powers to obtain a
quantity in square watts, which does not exist. If you wish to obtain the sum or difference of two powers, you
must first express each power in linear units (e.g. W or mW) before combining them.
● Do not multiply together (or divide) two logarithmic values. Doing so would amount to raising one quantity
to the power of the logarithmic value (or to the power of the reciprocal of the logarithmic value) of the other
quantity, which likely is not your intended operation.
● Do not multiply the logarithmic value of a quantity by a constant, say n, unless your intention is to raise that
quantity to power n. Similarly, do not divide by n unless you wish to raise the quantity to power 1/n. However,
if, for example, your actual intention is to share a power PdBW in dBW into n equal parts then perform this
operation as PdBW − 10log10 n to obtain the desired 1/n portion of the power in dBW. Alternatively, convert PdBW
to PW in watts before dividing by n.
● Do not subtract two powers that are expressed in dissimilar logarithmic units. For example, do not subtract a
dBm value from a dBW value.
● Do not pluralise the dB unit. In line with the convention of using logarithmic units to express absolute values,
the notation dBs means dB relative to one second. The unit of decibel should therefore always be singular to
avoid this confusion. Thus, it is a gain of 20 dB, not 20 dBs.
You may add together or subtract various logarithmic values if your intention is to obtain a product or division
Telegram: @Persian_IEEEComSoc

of the quantities represented by those values. For example, noise power, given by the formula Pn = kTB, may be
calculated by adding together the logarithmic values of k, T, and B. The logarithmic value of Eb /N o may be calcu-
lated by subtracting the logarithmic value of N o from that of Eb . And the figure of merit G/T of a wireless receiver
system may be found by subtracting the receiver’s system noise temperature in dBK from the receive-antenna gain
in dB.
You may subtract or add dB values to a dBW (or dBm, etc.) value to obtain a result in dBW (or dBm, etc.). This
operation amounts to scaling the quantity by loss or gain factors.
You may subtract two dBW (or two dBm, etc.) values to obtain a result in dB. This operation amounts to obtain-
ing the ratio between the two quantities. For example, subtracting input power in dBW from output power in
dBW gives the gain of the system in dB. And subtracting noise power in dBW from signal power in dBW yields a
signal-to-noise ratio in dB.
 2.8 Logarithmic Units 105

Always use the correct logarithmic unit for each quantity. Specifically, note that the unit of dB is applicable
only to gains, losses, and the ratio between two quantities having the same linear unit. The logarithmic units for
bandwidth, equivalent noise temperature, power, wireless receiver system’s figure of merit, etc. are not dB but
rather dBHz, dBK, dBW, dB/K, etc., respectively.
You may calculate the logarithm of a positive number X to any arbitrary base b (where b > 0) in terms of logarithm
to base 10 as follows
log10 X
logb X = (2.73)
log10 b
This equation comes in handy in situations where your calculator does not have a function to perform logarithm
to your desired base. For example
log10 81 1.9085
log3 81 = = =4
log10 3 0.4771
log10 2048 3.3113
log2 2048 = = = 11
log10 2 0.3010

Worked Example 2.9

Let us now apply logarithmic units to the transmission system of Figure 2.28 to determine power levels at
various points in the system.
( )
3.16mW
Pin = 3.16 mW = 10log10 = 5.0 dBm
1mW
Gain of 1st element = 10 (ratio) = 10log10 (10) = 10 dB
Loss of 2nd element = 63.1 (ratio) = 10log10 (63.1) = 18 dB
Gain of 3rd element = 10 (ratio) = 10log10 (10) = 10 dB
The power levels PA , PB , and Pout now follow by simple addition (subtraction) of the gains (losses) of the relevant
elements. See Figure 2.30.
PA = 5 dBm + 10 = 15 dBm
PB = 5 dBm + 10 − 18 = −3 dBm
Pout = 5 dBm + 10 − 18 + 10 = 7 dBm

Pin Gain PA Loss PB Gain

10 dB 18 dB 10 dB Pout
5 dBm
Telegram: @Persian_IEEEComSoc

Figure 2.30 Gains and losses in transmission system: Logarithmic-units procedure.

You may verify that these results agree with those obtained earlier using linear units. For example, using
Eq. (2.72), PB = −3 dBm = 10(−3∕10) mW = 0.5 mW, as obtained earlier. Note that we could have converted
−3 dBm to mW without using a calculator by observing that −3 dBm means 3 dB below 1 mW, which means a
factor of 2 below 1 mW, which means 0.5 mW.

Worked Example 2.10 Decibel Without Calculators

We wish to learn three simple steps which may be employed to obtain quick conversions between logarithmic
and linear values in a wide range of cases without the need for a calculator. It will serve you well in future if
you take a moment to learn these tricks.
 106 2 Introduction to Signals and Systems

Table 2.3 Conversion of ratio to dB.

Ratio or Number Expressed as Factors Converted to dB

8 2 × 2 × 2 or 23 3 + 3 + 3 or 3 × 3 = 9 dB
n
2 3n dB
200 100 × 2 20 + 3 = 23 dB
60 10 × 2 × 3 10 + 3 + 4.77 = 17.77 dB
1/2 2−1 3 × (−1) = −3 dB
500 1000 ÷ 2 30 – 3 = 27 dB
5 × 10−23 7 + (−230) = −223 dB
1/800 1 ÷ (100 × 8) 0 − (20 + 9) = −29 dB

Step 1:
Note that the dB value of any ratio (or number) that is a power of 10 does not require a calculator since it is
simply given by

10n (ratio) = 10n (dB) (2.74)

For example

1 (ratio) ≡ 100 (ratio) = 10 × 0 (dB) = 0 dB

100 (ratio) ≡ 102 (ratio) = 10 × 2 (dB) = 20 dB
1∕10000 (ratio) ≡ 10−4 (ratio) = 10 × (−4) (dB) = −40 dB

Step 2:
Know by heart the dB values of a few prime factors. You do not need to memorise them. Just interact with them
for a little while and you will know them just as you know your own name without having to claim that you
memorised it. Here are the few you need to know by heart

2 (ratio) = 3 dB
3 (ratio) = 4.77 dB
5 (ratio) = 7 dB
7 (ratio) = 8.45 dB (2.75)
Telegram: @Persian_IEEEComSoc

Step 3:
You may then convert any number to dB (without needing a calculator) if you can express that number as the
product of factors with known dB values. You then simply add up the dB values of those factors to obtain the
desired dB value of the number. Note that there will sometimes be more than one way of factorising. For example,
500 may be written as 100 × 5 or as 1000 × 1/2, but the dB result will be the same.
These steps may be applied in reverse to convert a dB value to its ratio equivalent as follows: write the dB
value as the sum of two or more terms each of whose ratio equivalent is known. The desired result is simply the
product of these component ratio equivalents. A few examples are tabulated below. Table 2.3 gives ratio-to-dB
conversions, whereas Table 2.4 shows dB-to-ratio conversions. The character ÷ in the tables denotes division
operation.
 2.9 Calibration of a Signal Transmission Path 107

Table 2.4 Conversion of dB values to ratio.

dB Value Expressed as a Sum Converted to Ratio

14 7+7 5 × 5 = 25
4 7–3 5 ÷ 2 = 2.5
44 40 + 7 – 3 104 × 5/2 = 2.5 × 104
1 10 – 9 10 ÷ 8 = 1.25
−53 7 – 60 5 × 10−6
−111.55 8.45 – 120 7 × 10−12
87 7 + 80 5 × 108

2.9 Calibration of a Signal Transmission Path

A transmission path imposes continual attenuation on a signal, which necessitates the placement of amplifiers
(called repeaters in analogue systems) at regular intervals along the path to boost the signal strength. To trace the
power level of a signal from its source through various points along a transmission path, we may calibrate the
transmission path relative to a chosen reference point in order to exploit the computational advantage offered by
logarithmic units. This point is called the zero-level reference point (ZRP) and is usually set around the input to the
link. For a four-wire circuit, the ZRP is usually the two-wire input to the hybrid transformer.
The transmission path is calibrated by assigning to every point on the link a transmission level (in dB) rela-
tive to the ZRP, hence the unit dBr. As illustrated in Figure 2.31, the assigned dBr value is the algebraic sum of
the dB gains (a loss being accounted as negative gain) from the ZRP to that point. This applies to points lying
beyond the ZRP in the forward path direction. The ZRP is usually chosen at the input, and therefore all points
would normally fall into this category. However, if for reasons of accessibility the ZRP is located other than at
the input then the dBr of each point lying before the ZRP is obtained by negating the algebraic sum of all gains
from that point to the ZRP. For example, a point lying beyond the ZRP and separated from it by a 10 dB gain
amplifier is assigned 10 dBr. A point lying beyond the ZRP and separated from it by 18 dB loss (G = −18 dB) and
15 dB gain (G = +15 dB) is marked −18 + 15 = −3 dBr. A point located before the ZRP and separated from it by
10 dB loss (G = −10 dB) and 25 dB gain is marked −(−10 + 25) = −15 dBr. The ZRP itself is marked 0 dBr since
it is the reference. Absolute signal power measured in dBm or dBW at the ZRP is expressed as dBm0 or dBW0,
respectively.
Telegram: @Persian_IEEEComSoc

0 dBr +10 dBr –8 dBr 2 dBr

ZRP + – +
L = 18 dB
G = 10 dB or G = 10 dB
G = −18 dB

+5 dBm0 +15 dBm –3 dBm +7 dBm

Figure 2.31 Transmission levels and the zero-level reference point (ZRP).
 108 2 Introduction to Signals and Systems

The absolute power level PdBm of a signal (in dBm) at any point on the link is determined by adding the link’s
dBr mark at that point to the signal’s dBm0 value, i.e. the power level of the signal at ZRP, denoted PdBm0 . That is
PdBm = PdBm0 + dBr (2.76)
where
PdBm = Signal power at the given point
PdBm0 = Signal power at the ZRP
dBr = Relative transmission level of the given point.
Equation (2.76) may also be used to determine the power of a signal at the entry point into the link (i.e. PdBm0 ).
This is given by the difference between the signal power at an arbitrary point along the link (i.e. PdBm ) and the dBr
value of the point. For example, in Figure 2.31, if we measure the signal level at the −8 dBr point and find that it
is −3 dBm, then we know that the level of the signal at the entry (i.e. ZRP) point is −3 − (−8) = 5 dBm0.

Worked Example 2.11

A transmission system consists of the following gain and loss components in the listed order:
(1) Loss = 30 dB
(2) Gain = 50 dB
(3) Loss = 8 dB
(4) Loss = 12 dB
(5) Gain = 25 dB
Draw a block diagram of the transmission system and calibrate it in dBr with the ZRP located at
(a) The input of the first component
(b) The input of the fourth component

A block diagram of the system is shown in Figure 2.32. Note that we entered the loss as negative gain in order
to simplify the algebraic summation involved in the calibration. The calibration for ZRP at the input of the first
component is shown on the upper part of the block diagram and that for ZRP at the input of the fourth component
is shown on the lower part. The procedure used is as earlier described. Points lying beyond the ZRP have a dBr
value equal to the algebraic sum of the gains up to that point. Points lying before the ZRP – in this case, the
first three components in (b) – have a dBr value equal to the negated algebraic sum of the gains from the ZRP
to the point.

(a) = Upper calibration

0 dBr
ZRP –30 dBr +20 dBr +12 dBr 0 dBr +25 dBr
Telegram: @Persian_IEEEComSoc

G = –30 dB G = 50 dB G = –8 dB G = –12 dB G = 25 dB

–12 dBr –42 dBr +8 dBr 0 dBr –12 dBr +13 dBr
ZRP
(b) = Lower calibration

Figure 2.32 Worked Example 2.11.

 2.10 Systems and Their Properties 109

2.10 Systems and Their Properties

A system may be defined very broadly as a functional group of interacting entities that is demarcated by a
boundary from its external environment. Systems may be physical artefacts such as the human body or a
household boiler, they may be concepts such as cultural or economic systems, or they may be processes such as
software algorithms. This book is concerned with a very narrow functional view of a system as an arrangement
or mathematical operation that maps or transforms an input signal into an output signal. And within this narrow
view, we focus mainly on transmission systems, an overview of which is presented in Chapter 1, that comprises
three parts, namely transmitter, channel, and receiver. Each of these parts is a system as well as a subsystem of
the overall communication system.
A system (in this narrow context) therefore manipulates an input signal or excitation x(t) to produce an output
signal or response y(t). This operation is illustrated in Figure 2.33 and will be denoted as
R
x(t) −−→ y(t) (2.77)
which is read ‘x(t) yields response y(t)’.
The system may be a continuous-time system which processes a continuous-time (CT) input signal to produce a
CT response; or, as also illustrated in Figure 2.33, it could be a discrete-time system which processes a discrete-time
(DT) input sequence x[n] to produce a DT response y[n]. An error control encoder is an example of a DT system
which transforms a discrete sequence of message bits into a discrete sequence of coded bits that includes some
redundant bits to aid error correction at the receiver. An AM (amplitude modulation) transmitter is a CT system
which transforms an analogue message signal into a transmitted analogue signal at higher frequency and power.
Note, however, that a system may feature both CT and DT signals. For example, an ADC system transforms a CT
input signal into a digital output signal representing a discrete sequence; and a digital modulator may process a
discrete input sequence of bits to produce analogue output pulse signals.
It will be useful for our tasks of transmission system analysis and design in later chapters to identify some of the
basic properties of the system in Figure 2.33.

2.10.1 Memory
A system is classed as memoryless if its output at any instant t depends at most on the input values at the same
instant. A system whose output at time t depends exclusively on the input at a past instant t − Δt is also regarded
as memoryless. To distinguish between these two, the first is described as memoryless instantaneous-response,
whereas the latter is referred to as memoryless delayed-response. If the output, however, depends on the current
input as well as one or more past or future input values then the system is said to have memory. The system also

Figure 2.33 System operation.

Telegram: @Persian_IEEEComSoc

x(t) Continuous- y(t)

time
system

x[n] Discrete- y(n)

time
system
 110 2 Introduction to Signals and Systems

System System
i(t) i(t)
Excitation

Excitation
Response

Response
t
1
R ʋR(t) = Ri(t) C ʋC (t) = C
i(t)dt
–∞

(a) (b)
System
i(t)
Excitation

Response di(t)
(c) L ʋL (t) = L
dt

Figure 2.34 Systems (a) with memory and (b) and (c) without memory.

has memory if the present output does not depend on the present input but depends on two or more past or
future inputs.
Figure 2.34 shows three single-component CT systems, namely (a) a resistor, (b) a capacitor, and (c) an inductor.
In each case the system excitation is the input current i(t) and the system response is the voltage drop produced
across the circuit element. From basic circuit theory, we have

vR (t) = Ri(t)

1
t
1 ∑
k=t∕Δt
vC (t) = i(t)dt ≡ lim i(kΔt)Δt
C ∫−∞ C Δt→0
k=−∞
[ ]
di(t) i(t) − i(t − Δt)
vL (t) = L ≡ L lim (2.78)
dt Δt→0 Δt
Thus, a resistor is a memoryless instantaneous-response system, whereas a capacitor, with its response voltage
obtained as a scaled accumulation of all past excitation currents, is a system with memory. The inductor also has
Telegram: @Persian_IEEEComSoc

memory since its response (voltage) is based on the difference between its present and immediate past excitation
(current).
All the discrete-time systems that perform the operations specified below, where p > 0 is an integer, have memory

1 ∑
p
y(n) = x(n − k) (Moving average)
p + 1 k=0
∑
p
y(n) = ak x(n − k) (Weighted sum)
k=0
y(n) = x(n) − x(n − 1) (Backward difference)
y(n) = x(n) − x(n + 1) (Forward difference) (2.79)
 2.10 Systems and Their Properties 111

Note, however, that the CT and DT systems described by the input–output relations
y(t) = Kx2 (t − t0 )
y(n) = ax(n − n0 )
where t0 , n0 > 0; K, a ≠ 0 (2.80)
are memoryless delayed-response systems. But the systems described by
y(t) = ax(t) + bx2 (t)
y(n) = cx2 (n) + dx3 (n)
where |a| + |b| > 0, |c| + |d| > 0 (2.81)
are memoryless instantaneous-response systems.
In a memoryless instantaneous-response system, the output signal is always in step with the input signal and an
excitation at the input is felt instantaneously (i.e. without delay) at the system output. Thus, such a system does not
cause any phase shift, signal delay, or phase distortion. A memoryless delayed-response system also does not cause
any phase distortion, but it introduces a fixed delay between excitation and system response which is akin to the
time it takes for a signal to propagate through the system, every frequency component of the signal experiencing
an equal amount of delay. It should be noted, however, that memoryless systems are an idealisation. Practical
transmission systems will have memory to varying extents since there will always be some amount of residual
capacitive and inductive effects in any conductors or semiconductors within the system, which will introduce
energy storage and hence some memory capacity. Therefore, when we treat a practical system as memoryless, we
simply mean that its memory capacity is small enough to be ignored in the context of the application.

2.10.2 Stability
A system is described as bounded-input, bounded-output (BIBO) stable if and only if, for every absolutely bounded
input |x(t)| < K i < ∞, the output is also absolutely bounded so that |y(t)| < K o < ∞.
The CT system
R
x(t) −−→ x2 (t)
is BIBO stable since, if |x(t)| < K i < ∞ then |y(t)| < K i 2 < ∞. That is, any finite input will produce a finite output.
However, the CT system
R
x(t) −−→ log(x(t))
is unstable since if |x(t)| = 0 then |y(t)| = ∞, and thus a bounded input produces an unbounded output.
All the DT systems represented in Eq. (2.79) are BIBO stable, but the DT system with input–output relation
(or difference equation) given by
Telegram: @Persian_IEEEComSoc

y(n) = x(n − 1) + 𝛼y(n − 1) (2.82)

is BIBO stable only if |𝛼| < 1, and is unstable otherwise. Notice that this system derives the current output by adding
a portion 𝛼 of the previous output to the previous input. If this portion is at least equal to the previous output
(i.e. |𝛼| ≥ 1) then the magnitude of the output will increase indefinitely with time, which gives rise to instability.
Stability is a critical system property which must always be carefully assessed and assured during design. Insta-
bility in an electrical system such as an amplifier or a digital filter might cause excessive current and power to be
delivered to sensitive circuit components and devices, causing malfunction or serious damage. And if a mechan-
ical system such as a suspension bridge is unstable, the combined excitation by strong winds and usage loading
might create excessive vibrational response in the bridge structure which could be enough to cause catastrophic
collapse. The Angers Bridge collapse of 16th April 1850, which happened when a battalion of French soldiers was
 112 2 Introduction to Signals and Systems

marching across the bridge in strong winds, killing over 200 of the soldiers, is a well-known example of the danger
posed by system instability.

2.10.3 Causality
A causal system is one in which the response at any given time depends only on the present or past excitations.
If the system’s response is influenced in any way by a future excitation then the system is said to be non-causal.
You may wonder how it can ever be possible for a system to use future inputs (which have not yet occurred)
in order to produce the present output. Well, this is not possible in real-time, so non-causal real-time systems
cannot be designed. However, a non-causal system can be implemented to operate in non-real-time on recorded
data where all ‘future’ input data are already available in storage and can be accessed as required to produce the
‘present’ output.
The moving average filter, with difference equation given in line 1 of Eq. (2.79), calculates the present output
as the average of the present input and p immediate past inputs. This operation is causal. The backward differ-
ence system in Eq. (2.79) is also causal. However, the centred moving average (MA) filter with difference equation
given by
1 ∑p
y(n) = x(n − k) (centred MA) (2.83)
2p + 1 k=−p

obtains the present output by averaging the present input x(n) along with p immediate past inputs {x(n−1), x(n−2),
…, x(n−p)} and p immediate future inputs {x(n + 1), x(n + 2), …, x(n + p)}. This operation is therefore non-causal.
The forward difference system in Eq. (2.79) is also non-causal.
Figure 2.35 shows plots of the responses of two CT systems to an impulse function excitation 𝛿(t). The impulse
responses of the systems are specified by
R
𝛿(t) −−→ h1 (t) = sinc(t) (Fig. 2.35b)
R
𝛿(t) −−→ h2 (t) = e−t u(t) (Fig. 2.35c) (2.84)

δ(t)

(a) Impulse excitation

t
0
(b) Response: Non-causal system (c) Response: Causal system
Telegram: @Persian_IEEEComSoc

h1(t) = sinc(t) h2(t) = e–tu(t)

1 1

0 0

–0.3 t t
–4 –3 –2 –1 0 1 2 3 4 –4 –3 –2 –1 0 1 2 3 4

Figure 2.35 Example of system excitation and responses.

 2.10 Systems and Their Properties 113

Notice how the system in Figure 2.35b has an output at time t < 0 in response to an input which is only applied
at time t = 0. In contrast, the response of the system in Figure 2.35c begins only after the input is applied at t = 0.
Thus, the first system is non-causal whereas the second is causal. We revisit the concept of a system’s impulse
response in Chapter 3 and apply it to the analysis of a certain class of systems.

2.10.4 Linearity
A linear system is one that obeys the principle of superposition, which states that if excitation x1 (t) produces
response y1 (t) and excitation x2 (t) produces response y2 (t) then excitation a1 x1 (t) + a2 x2 (t) will produce response
a1 y1 (t) + a2 y2 (t), where a1 and a2 are arbitrary constants. A linear system is therefore both additive and homoge-
neous. A system is said to exhibit the property of additivity if the response of the system to a sum of two or more
excitations is simply the sum of the responses of the system to each excitation applied alone. And a system is said
to be homogeneous if when an excitation is scaled by a constant factor the response is scaled by the same factor.
Employing the notation of Eq. (2.77) allows us to state the above definitions compactly as follows
Given that
R
x1 (t) −−→ y1 (t)
and
R
x2 (t) −−→ y2 (t)
then
R
x1 (t) + x2 (t) −−→ y1 (t) + y2 (t) (additivity)
R
a1 x1 (t) −−→ a1 y1 (t) (homogeneity)
R
a1 x1 (t) + a2 x2 (t) −−→ a1 y1 (t) + a2 y2 (t) (linearity) (2.85)
If a system disobeys any one of the properties of additivity and homogeneity then the system is said to be nonlin-
ear. The property of linearity is widely applied in system analysis to obtain the response of a system to an arbitrary
excitation if that excitation can be expressed as a linear combination (i.e. amplitude scaling and summation) of
signals whose responses are already known.
Letting  denote an operator that encapsulates the transformation performed by a system on its input x(t) to
obtain its output y(t) such that we can write
y(t) = {x(t)} (2.86)
it follows that if  (and hence the system) is linear then for a group of N amplitude-scaled inputs the output y(t)
is given by
{N }
∑
Telegram: @Persian_IEEEComSoc

y(t) =  ak xk (t)
k=1

∑
N
= ak yk (t)
k=1
∑N
= ak {xk (t)} (2.87)
k=1
Notice that the first line represents the system operating on a group of N amplitude-scaled inputs, whereas the
last line represents the system first transforming each input xk (t), k = 1, 2, 3, …, N, to yield its individual response
yk (t) and then scaling those responses and adding the scaled responses together to obtain the group response.
That is, the order of manipulation in the first line is (1) scaling, (2) summation, and (3) operator action, whereas
 114 2 Introduction to Signals and Systems

in the last line the order is (1) operator action, (2) scaling, and (3) summation. This means that if an operator is
linear then the order of scaling and operation, or summation (which includes integration) and operation, can be
interchanged as convenient. We expand upon this reasoning later to develop an extremely useful tool for system
analysis in the time domain (Chapter 3) and in the frequency domain (Chapter 4).

Worked Example 2.12

Determine which of the properties of additivity, homogeneity, and linearity is obeyed by the systems with the
following input–output relations:
(i) y(t) = 𝛽x(t) + A, where 𝛽 and A are constants.
(ii) y(n) = 𝛽x(n − 1) + 𝛾x2 (n), where 𝛽 and 𝛾 are constants.
(iii) The centred moving average filter of Eq. (2.83).

For convenience, let us adopt the following convention for naming the response of a system to various excitations
R
xk (t) −−→ yxk (t)
R
axk (t) −−→ yaxk (t)
R
x1 (t) + x2 (t) −−→ yx1 +x2 (t)
With this convention it follows that a system is additive if and only if
yx1 +x2 (t) = yx1 (t) + yx2 (t) (2.88)
A system is homogeneous if and only if
yax (t) = ayx (t) (2.89)
And a system is linear if it is both additive and homogeneous so that
ya1 x1 +a2 x2 (t) = a1 yx1 (t) + a2 yx2 (t) (2.90)
We are now ready to tackle the problems at hand.
(i) For this system
R
x1 (t) −−→ 𝛽x1 (t) + A ≡ yx1 (t)
R
x2 (t) −−→ 𝛽x2 (t) + A ≡ yx2 (t)
R
x1 (t) + x2 (t) −−→ 𝛽[x1 (t) + x2 (t)] + A ≡ yx1 +x2 (t)
Telegram: @Persian_IEEEComSoc

We see that the sum of individual responses to x1 (t) and x2 (t) is

yx1 (t) + yx2 (t) = 𝛽x1 (t) + 𝛽x2 (t) + 2A
whereas the response to the sum of x1 (t) and x2 (t) is
yx1 +x2 (t) = 𝛽x1 (t) + 𝛽x2 (t) + A
Since yx1 +x2 (t) ≠ yx1 (t) + yx2 (t), we conclude (in view of Eq. (2.88)) that the system is not additive. In general,
a system whose operation includes adding a constant term to the excitation (producing a nonzero response
under conditions of zero excitation) will not be additive.
 2.10 Systems and Their Properties 115

To assess the system’s homogeneity, we note that

R
x(t) −−→ 𝛽x(t) + A ≡ yx (t)
R
ax(t) −−→ 𝛽[ax(t)] + A ≡ yax (t)

We see that yax (t) = 𝛽ax(t) + A, whereas ayx (t) = 𝛽ax(t) + aA and these are not equal. That is, applying an
amplitude-scaling factor to the input does not yield the same result as applying this factor to the system
output. In view of Eq. (2.89), the system is therefore not homogeneous.
The lack of additivity or homogeneity demonstrated above is enough for us to conclude that the system is
nonlinear.
(ii) This DT system specification leads to
R
x1 [n] −−→ 𝛽x1 (n − 1) + 𝛾x12 (n) ≡ yx1 (n)
R
x2 [n] −−→ 𝛽x2 (n − 1) + 𝛾x22 (n) ≡ yx2 (n)
R
x1 [n] + x2 [n] −−→ 𝛽[x1 (n − 1) + x2 (n − 1)] + 𝛾[x1 (n) + x2 (n)]2 ≡ yx1 +x2 (n)

We see that the sum of individual responses to x1 [n] and x2 [n] is

yx1 (n) + yx2 (n) = 𝛽x1 (n − 1) + 𝛽x2 (n − 1) + 𝛾x12 (n) + 𝛾x22 (n)

and is not equal to the response to the sum of x1 [n] and x2 [n] which is

yx1 +x2 (n) = 𝛽x1 (n − 1) + 𝛽x2 (n − 1) + 𝛾x12 (n) + 𝛾x22 (n) + 2𝛾x1 (n)x2 (n)

Thus, the system is not additive.

To assess the system’s homogeneity, observe that
R
x[n] −−→ 𝛽x(n − 1) + 𝛾x2 (n) ≡ yx (n)
R
ax[n] −−→ 𝛽[ax(n − 1)] + 𝛾[ax(n)]2 ≡ yax (n)

Since

yax (n) = a[𝛽x(n − 1) + a𝛾x2 (n)] ≠ ayx (n)

we conclude that the system is not homogeneous. The system is also nonlinear in view of its lack of additivity
and homogeneity. In general, any system whose input–output relation is a polynomial of order 2 or higher is
nonlinear.
Telegram: @Persian_IEEEComSoc

(iii) For the centred MA system, let there be a second input sequence denoted v[n]. It follows from Eq. (2.83) that

R 1 ∑p
x[n] −−→ x(n − k) ≡ yx (n)
2p + 1 k=−p

R 1 ∑p
v[n] −−→ v(n − k) ≡ yv (n)
2p + 1 k=−p

R 1 ∑p
x[n] + v[n] −−→ [x(n − k) + v(n − k)] ≡ yx+v (n)
2p + 1 k=−p
 116 2 Introduction to Signals and Systems

We see that
1 ∑p
yx+v (n) = [x(n − k) + v(n − k)]
2p + 1 k=−p

1 ∑p
1 ∑p
= x(n − k) + v(n − k)
2p + 1 k=−p 2p + 1 k=−p
= yx (n) + yv (n)

which means that the system is additive.

To assess homogeneity, we observe that

R 1 ∑p
ax[n] −−→ [ax(n − k)] ≡ yax (n)
2p + 1 k=−p

Since
1 ∑p
yax (n) = [ax(n − k)]
2p + 1 k=−p
[ ]
1 ∑p
=a x(n − k)
2p + 1 k=−p
= ayx (n)

it follows that the system is homogeneous. And since the system is both additive and homogeneous, we can
conclude that it is linear.
Here are questions for you to ponder: can an additive system ever fail to be homogeneous, or a homogeneous
system fail to be additive? Are additivity and homogeneity different statements of the same principle? Is additivity
a necessary and sufficient condition for linearity?

2.10.5 Time Invariance

A system is said to be time invariant if its operations do not depend on the time of application of the excitation.
Thus, the only effect (on system response) of a time shift in the excitation is simply a corresponding time shift in
the response. By analogy with the homogeneity property in which amplitude scaling has the same impact whether
applied at the input or at the output, we see that time-invariance means that a time shift has the same impact on
the system whether applied at the input or at the output. If this is not the case, the system is said to be time variant.
A formal statement of the time invariance property is that a system is time invariant if
Telegram: @Persian_IEEEComSoc

given that
R
x(t) −−→ y(t) (CT system)
or
R
x[n] −−→ y(n) (DT system)
then
R
x(t − to ) −−→ y(t − to ) (time invariant CT)
R
x[n − no ] −−→ y(n − no ) (time invariant DT) (2.91)
 2.10 Systems and Their Properties 117

To present a simple check for time invariance, let us introduce the delay operator 𝔻 such that 𝔻[x(t), to ] ≡
x(t − to ) denotes a delay of x(t) by to . A system is therefore time invariant if and only if
y𝔻[x(t), to ] (t) = 𝔻[y(t), to ] (2.92)
The left-hand side of this equation is the response to a delayed version of the excitation, whereas the right-hand
side is a delayed version of the response to a non-delayed excitation. Letting {•} denote the system operation, we
may write Eq. (2.92) more completely as
{𝔻[x(t), to ]} = 𝔻[{x(t)}, to ] (2.93)
This makes it explicitly clear that the order of system operation and time-shift may be interchanged in a
time-invariant system. That is, you will obtain the same output if you apply a time-shift to the signal before
transmitting it through the system or you transmit the signal through the system before applying the time shift to
the system’s response.

Worked Example 2.13

Assess each of the following systems for time invariance:

(i) The system of Figure 2.34a if the system resistance R is a function of time, denoted R(t).
(ii) A DT system with input–output relation y(n) = Kx(Mn), where K ≠ 0 is a constant and M > 0 is an integer.
(iii) The backward difference system of Eq. (2.79).

(i) This is a simple system with output (voltage) R(t)x(t) in response to an input (current) x(t). Employing
Eq. (2.93), we write
{x(t)} = R(t)x(t)
𝔻[{x(t)}, to ] = 𝔻[R(t)x(t), to ] = R(t − to )x(t − to )
The above is the result of first passing the signal through the system at time t and then delaying the response.
If, on the other hand, we pass a delayed version x(t − to ) of the signal through the system at time t, we obtain
{𝔻[x(t), to ]} = {x(t − to )} = R(t)x(t − to )
Since R(t − to ) ≠ R(t), it follows that {𝔻[x(t), to ]} ≠ 𝔻[{x(t)}, to ]. Therefore, the system is time variant. In
general, a system with time-dependent components will be time-variant.
(ii) Passing the signal through this system at time interval count n and then delaying the response by no yields
𝔻[{x[n]}, no ] = 𝔻[Kx(Mn), no ] = Kx(M(n − no )) = Kx(Mn − Mno )
If, on the other hand, we pass a delayed version x(n − no ) of the signal through the system at interval n, we
Telegram: @Persian_IEEEComSoc

obtain
{𝔻[x[n], no ]} = {x[n − no ]} = Kx(Mn − no )
Notice that interchanging the order of time shift and operator action produces different results Kx(Mn − Mno )
and Kx(Mn − no ). Thus, the system is time variant. Figure 2.36 presents a graphical illustration of this solution
for input signal
{
n, n = −10, −9, · · · , 0, 1, 2, · · · , 10
x(n) =
0, Otherwise
and parameters K = 1, M = 2, and no = 3. Figure 2.36a–c show the implementation of the right-hand side of
Eq. (2.93) in which the excitation signal x[n] is first passed through the system and then the delay is applied
 118 2 Introduction to Signals and Systems

x(n)
10

(a) 0 n

–10
yx(n) = x(2n)
10

(b) 0 n

–10
D[x(2n), 3] ≡ yx(n – 3)
10

(c) 0 n

–10
–10 –9 –8 –7 –6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6 7 8 9 10

Figure 2.36 Worked Example 2.13(ii): (a) Excitation x(n); (b) System’s response y x (n) to x(n); (c) Delay of system’s response;
(d) Excitation x(n); (e) Delayed excitation x(n − 3); (f) System’s response to x(n − 3).

to the system’s response. Figure 2.36d–f, where a plot of the excitation signal x[n] is repeated in (d) for con-
venience of comparison, show the execution of the left-hand side of Eq. (2.93) in which the excitation is first
delayed and the delayed version is then passed through the system. The outcomes of these two operations are
shown in parts (c) and (f) of the figure. The resulting sequences are clearly different.
(iii) First passing the signal through the system and then delaying the response yields
𝔻[{x[n]}, no ] = 𝔻[x(n) − x(n − 1), no ] = x(n − no ) − x(n − no − 1)
If, on the other hand, the signal is first delayed before this delayed version is passed through the system, we
obtain
{𝔻[x[n], no ]} = {x[n − no ]} = x(n − no ) − x(n − no − 1)
Telegram: @Persian_IEEEComSoc

We see that the two results are identical. Interchanging the order of time shift and operator action does not
alter the outcome. The system is therefore time invariant.

2.10.6 Invertibility
A system is said to be invertible if the excitation signal can be recovered without error or ambiguity from the
response signal. If there is a one-to-one mapping between input and output then the system is invertible. If, on
the other hand, two or more different inputs can produce the same output then the system is not invertible. An
example of a noninvertible system is the square operation y(t) = x2 (t). In this system, given, for example, y(t) = 4,
it is not possible to recover the input x(t), since one cannot be sure whether x(t) = 2 or x(t) = −2. Another
 2.10 Systems and Their Properties 119

x(n)
10

(d) 0 n

–10
ʋ(n) ≡ x(n – 3)
10

(e) 0 n

–10
10

(f) 0 n

–10
–10 –9 –8 –7 –6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6 7 8 9 10

Figure 2.36 (Continued)

√
example is y(t) = cos(x(t)), since given, for example, y(t) = 1∕ 2 one cannot know if x(t) = (−45 + 360n)∘ or
x(t) = (45 + 360n)∘ , where n is an integer. In the process of analogue-to-digital conversion (ADC), a quantiser
is employed to convert a continuous-value input sample into a discrete-value output sample by mapping all input
values falling within one quantisation interval to a single output value. Once this has been done, information
about the precise value of the input is lost and it is not possible to go from quantiser output back to the original
input. The quantiser is therefore a noninvertible system.
If a system whose operation is represented by the operator  is invertible then we can define an inverse system
with operator  inv such that
x(t) =  inv {y(t)} =  inv {{x(t)}}
=  inv {x(t)} = {x(t)}
= x(t)
Telegram: @Persian_IEEEComSoc

The operators of a system and its inverse therefore satisfy the relationship
 inv  =  (2.94)
where  is an identity operator that performs the identity mapping y(t) = x(t).
Consider a memoryless delayed-response system having linear input–output relation
y(n) = Kx(n − no ) (2.95)
where K ≠ 0 is a constant and no > 0 is an integer. This system is invertible, and we may derive an inverse system
function as follows. First, rearrange to make x() the subject of the equation
1
x(n − no ) = y(n)
K
 120 2 Introduction to Signals and Systems

Next, substitute m = n − no (and hence n = m + no )

1
x(m) = y(m + no )
K
Finally, replace m by n to return to our standard use of n to denote sampling interval count
1
x(n) = y(n + no ) (2.96)
K
Eq. (2.96) is the inverse input–output relation by which the excitation x(n) may be recovered from the response
sequence y[n] of the system specified by Eq. (2.95). Notice that this inverse system is non-causal, whereas the
original system is causal. The system operator  and its inverse  inv may be written as follows
 = K(𝔻[x(n), no ])
1
 = (𝔻[y(n), −no ])
inv
(2.97)
K
It should be obvious from our opening comments on invertibility that the system
y(n) = Kx2 (n − no )
is noninvertible since its inverse is
√
1
x(n) = ± y(n + no )
K
and the uncertainty in sign makes unambiguous recovery impossible.
If the system is a transmission channel then the excitation x(t) is the transmitted signal, the response y(t) is the
received signal, and the effects of the channel are represented by the operator . We usually need to recover or
closely estimate x(t) from y(t), and this requires connecting in cascade with the channel an inverse system known
as an equaliser having an operator  inv that satisfies Eq. (2.94). This process is illustrated in Figure 2.37. We explore
the subject of equalisers in more detail in Chapter 4 (Section 4.7).

Worked Example 2.14

We wish to determine the inverse of a backward difference system that operates on a causal input signal. Note
that, strictly speaking, causality applies only to systems. However, the term is sometimes also applied to signals,
so that a causal signal means a signal that has zero value at time t < 0 (or n < 0 for discrete signals).

A backward difference system has input–output relation given earlier in Eq. (2.79) and repeated below for
convenience.
y(n) = x(n) − x(n − 1) (2.98)
Telegram: @Persian_IEEEComSoc

x(t) y(t) x(t)

Figure 2.37 Invertible system.

 2.11 Summary 121

Figure 2.38 shows a block diagram of this system. Let us write out the output sequence starting at n = 0, for
which y(0) = x(0) − x(−1) = x(0) (since the signal is causal and therefore x(−1) = 0)
y(0) = x(0)
y(1) = x(1) − x(0)
y(2) = x(2) − x(1)
⋮
y(n − 1) = x(n − 1) − x(n − 2)
y(n) = x(n) − x(n − 1)
Since the left and right-hand side of each of the above equations are equal, the sum of the left-hand side of all
these equations will also be equal to the sum of the right-hand side. Observe that, in summing the right-hand side,
all terms cancel out except x(n). Thus
x(n) = y(0) + y(1) + y(2) + · · · + y(n)
∑
n
= y(n) (2.99)
k=0

The inverse of the backward difference system of Eq. (2.98) is therefore the accumulator operation of Eq. (2.99).
Backward difference operation is common in data compression systems in which the encoder only transmits
the difference between the present sample and the immediate previous sample. This worked example therefore
shows that the decoder will be able to recover the original samples through the accumulation operation spec-
ified by Eq. (2.99), in which the present sample x(n) is the sum of all received differential samples up to the
present.

x(n) + y(n)
Σ
–
x(n – 1)
D[x(n),1]

Figure 2.38 Backward difference system.

2.11 Summary
Telegram: @Persian_IEEEComSoc

This chapter presented a detailed introduction to telecommunication signals and systems. As in fact in the rest
of the book, the topics covered, and the approach and depth of treatment, have been carefully selected to strike a
delicate balance between comprehensive rigour and succinct simplicity. The aim is to minimise needless mathe-
matical hurdles and present material that is fresh and accessible for newcomers, insightful, and informative for
everyone, and free of knowledge gaps in supporting further and more advanced studies.
 122 2 Introduction to Signals and Systems

We started with a nonmathematical discussion of signals and their various features and types, using both sub-
jective and objective classifications. Recognising and understanding such signal features will inform the methods
selected for analysis and design and can greatly simplify various computations on the signal. We then introduced
and characterised a range of special waveforms which can be used as building blocks to model other signals. Doing
so allows, for example, results obtained for such special signals to be quickly extended and applied to practical or
arbitrary signals without the need to start from first principles. To give a simple example, if you already know the
response of a linear transmission system to a unit step function then by expressing a rectangular pulse in terms
of unit step functions you can quickly obtain the response of the system to a rectangular pulse without having
to undertake lengthy calculations. We discussed the sinusoidal signal in exhaustive detail because of its unique
centrality to telecommunication and most of the basic concepts of the subject area.
We also discussed logarithmic units and their applications and emphasised some of the short cuts to exploit and
pitfalls to avoid, which you will do well to review again as required until you have complete mastery of what is an
indispensable engineering tool. Basic system properties were also introduced, and the discussion was supported
with worked examples covering both continuous-time and discrete-time systems. You are now well equipped to
be able to assess the properties of a system given its input–output relation, an essential step in selecting the correct
tools and approach for analysis and design.
In the next chapter we build on this introduction and delve a little deeper into more general and arbitrary signals,
including random signals, exploring various signal operations and characterisations, as well as signals and systems
analysis tools in the time domain.

Questions

2.1 Determine which of the following signals is an energy signal and which is a power signal:
(a) Impulse function 𝛿(t)
(b) Sinc function sinc(t)
(c) Unit step function u(t)
(d) Signum function sgn(t)
(e) Complex exponential function exp(j𝜔t).

2.2 By obtaining values at three time instants, namely the start, mid-point, and end of each pulse, verify the
correctness of the respective expressions on the right-hand side of Eqs. (2.19) and (2.20) for the ramp pulse
and inverse ramp pulse.

2.3 Figure Q2.3 shows a rectangular pulse g(t), bipolar pulse g1 (t), and triangular pulse g2 (t).
Telegram: @Persian_IEEEComSoc

g(t) = Arect(t/τ) g1(t) g2(t) = Atrian(t/τ)

A A
A
τ/2
t t t
–τ/2 τ/2 –τ/2 –τ/2 τ/2

–A

Figure Q2.3 Question 2.3.

 Questions 123

(a) Express g1 (t) in terms of rectangular pulses.

(b) Express g2 (t)
(i) in terms of g1 (t), and hence
(ii) in terms of rectangular pulses.
(Hint: consider the role of derivatives and integrals in solving (b).)

2.4 Show that the trapezoidal pulse of Figure 2.13g can be expressed as given by Eq. (2.26) in terms of ramp
and rectangular pulses.

2.5 Determine the average value of each of the following signals:

(a) −10u(t) volts.
(b) Rectangular pulse train of 30 V amplitude and 25% duty cycle.
(c) Triangular pulse train of amplitude 50 V, duty cycle 20%.

2.6 Determine the average value of a trapezoidal pulse train that has the following parameters:
(a) Amplitude A = 100 V.
(b) Duration of rising edge of pulse 𝜏 r = 10 ms.
(c) Duration of flat or constant portion of pulse 𝜏 c = 20 ms.
(d) Duration of falling edge of pulse 𝜏 f = 15 ms.
(e) Duration of no pulse in each cycle 𝜏 0 = 55 ms.

2.7 Determine the period, angular frequency, peak-to-peak, and rms values of the signal g(t) = −sin(t).

2.8 Determine
(a) The phase difference; and
(b) Delay between the signals g1 (t) = −25 cos(200𝜋t + 70∘ ); and g2 (t) = 5 sin(200𝜋t − 30∘ ), where t is time
in ms. (NOTE: you must specify which signal leads the other.)
(c) The wavelength (in air) of the resulting sound if g1 (t) in (b) were applied as input to a loudspeaker.

2.9 Figure Q2.9 shows oscilloscope displays of sinusoidal waveforms g1 (t), g2 (t), g3 (t), and g4 (t). Write down
the sinusoidal expression of each waveform in the form Acos(2𝜋ft + 𝜙).

2.10 Given the voltage signals

v1 (t) = 20 sin(t)
Telegram: @Persian_IEEEComSoc

v2 (t) = −5 cos(t − 60∘ )

v3 (t) = sin(2𝜋 × 103 t)
v4 (t) = 2 sin(4𝜋 × 103 t − 60∘ )

(a) Determine the period of each signal.

(b) Determine the phase difference between v1 (t) and v2 (t).
(c) Make clearly labelled sketches of v3 (t) and v4 (t) over three cycles starting at t = 0 and hence explain
why there is no defined phase difference between sinusoids of unequal frequencies.
(d) If v3 (t) is applied as input to a loudspeaker, what is the wavelength (in air) of the resulting audible
sound?
 124 2 Introduction to Signals and Systems

g1(t) g2(t)

10 V/div
10 V/div

t t

Time base = 50 s/div Time base = 62.5 ms/div

g3(t) g4(t)

10 V/div
10 V/div

t t

Time base = 500 μs/div Time base = 25 ms/div

Figure Q2.9 Question 2.9.

2.11 Given the signals

g1 (t) = 30 cos(200πt)
g2 (t) = 40 sin(200πt)
g3 (t) = −50 sin(200πt + 50∘ )
determine the signal g(t) = g1 (t) − g2 (t) + g3 (t) expressed in the form
g(t) = A cos(200πt + 𝜙)
The amplitude A and phase 𝜙 of g(t) must be specified.

2.12 Determine the output signal v(t) of each of the summing devices shown in Figure Q2.12a and b, expressed
Telegram: @Persian_IEEEComSoc

in the form
v(t) = A cos(𝜔t + 𝜙)
The amplitude A and phase 𝜙 of v(t) must be specified.

2.13 The solid curve of Figure Q2.13 is a sketch of a voltage waveform v(t), which consists of a DC component
and two sinusoids. Write out the full expression for v(t) in the form
v(t) = A0 + A1 cos(2𝜋f1 t + 𝜙1 ) + A2 cos(2𝜋f2 t + 𝜙2 )
explaining clearly how you arrive at the values of A0 , A1 , A2 , f 1 , f 2 , 𝜙1 , and 𝜙2 .
 Questions 125

ʋ1(t) ʋ1(t)

+ ʋ(t) + + ʋ(t)
(a) Σ (b) ʋ2(t) Σ
– –
ʋ2(t) ʋ3(t)
ʋ1(t) = 4cos(100t – π/3) ʋ1(t) = 30sin(ωt)
ʋ2(t) = 3sin(100t) ʋ2(t) = 40sin(ωt – π/4)
ʋ3(t) = 50sin(ωt + π/6)

Figure Q2.12 Question 2.12.

18

15

12

9
ʋ(t) (volts) →

–3
–6

–9
0 0.5 1 1.5 2 2.5 3
Time, t (ms) →

Figure Q2.13 Question 2.13.

Table 2.5 Question 2.15.

Power (W) Power (dBm) Power (dBW) Power (dBmp, 3.1kHz) Volts (dBV) Volts (dBu)

100
Telegram: @Persian_IEEEComSoc

20
30
10
10

2.14 Given two voltage signals v1 (t) = 10 + 4 sin(2𝜋 × 103 t) volts and v2 (t) = 25 sin(4𝜋 × 104 t) volts, sketch the
waveform of the product signal g(t) = v1 (t)v2 (t).

2.15 Fill in the blank cells in Table 2.5. Assume that the input and output resistances of the system are equal to
1 Ω, which also means that we are dealing with normalised power.
 126 2 Introduction to Signals and Systems

2.16 A transmission system consists of the following gain and loss components in the listed order:
(1) Gain = 20 dB
(2) Gain = 50 dB
(3) Loss = 95 dB
(4) Gain = 30 dB
(5) Loss = 12 dB.
(a) Draw a block diagram of the transmission system and calibrate it in dBr with the ZRP located at the
input of the second component.
(b) A signal monitored at the ZRP has a level of 60 dBm0. What will be the absolute power level of this
signal at the output of the transmission system?

2.17 A discrete-time system is described by the following difference equation

y(n) = x(n − 1) + 2y(n − 1)
(a) By preparing a tabular layout of the response h(n) of this system to a unit impulse excitation
𝛿[n] = {1, 0, 0, 0, …} for values of n from 0 to 6 show that h(n) grows indefinitely with n and hence
that the system is unstable.
(b) It is easy to erroneously conclude that this is a memoryless delayed-response system by thinking that
the current output depends exclusively on the input at a single past instant. Show that the output of
this system depends on multiple past inputs and that in fact every previous input has influence on the
current output, which means that the system has memory.

2.18 A system performs the accumulation operation specified by the input–output relation
∑
n
y(n) = x(k)
k=−∞

(a) Assess the properties of this system in relation to memory, stability, causality, linearity, time invariance,
and invertibility.
(b) Derive the difference equation (or input–output relation) of an inverse system which can be employed
to recover the excitation x[n] of this system from its response y[n].
Telegram: @Persian_IEEEComSoc
 127

Time Domain Analysis of Signals and Systems

Impossible is a lie peddled on the back of a wrong method.

In this Chapter
✓ Basic signal operations, including a detailed treatment of time shifting, time reversal, and time scaling.
✓ Random signals analysis.
✓ Standard statistical distribution functions and their applications in telecommunications.
✓ Signal characterisation in the time domain.
✓ System characterisation and analysis in the time domain.
✓ Autocorrelation and convolution operations.
✓ Worked examples involving a mix of heuristic, graphical, and mathematical approaches to demonstrate the
interpretation and application of concepts and to deepen your insight and hone your skills in engineering
problem solving.
✓ End-of-chapter questions to test your understanding and (in some cases) extend your knowledge of the
material covered.

3.1 Introduction
Signals and systems can be characterised both in time and frequency domains. This chapter focuses on a time
domain description and analysis of signals and systems. For signals, we may specify the waveform structure of
the signal – through a waveform plot as might be displayed on an oscilloscope, or a mathematical expression that
Telegram: @Persian_IEEEComSoc

gives values of the signal as a function of time, or a statistical description in the case of random signals. From
these, various parameters of the signal may be determined, such as amplitude, peak-to-peak value, average value,
root-mean-square (rms) value, period, duty cycle, instantaneous values and their rates of change, power, energy,
autocorrelation, etc. We may specify the input–output relation or transfer characteristic of any system. However,
for a special class of systems which are linear and time invariant, the response of the system to a unit impulse
input (known as the system’s impulse response) is the most versatile characterisation tool. The response of such a
system to any arbitrary input may be obtained by convolving the input signal with the system’s impulse response.
We start with a discussion of basic signal operations on which are built other more complex signal manipula-
tions in the time domain, e.g. autocorrelation, convolution, etc. Basic signal operations include time shifting, time
reversal, time scaling, addition, subtraction, multiplication, division, differentiation, and integration. However, we
discuss only the first three operations. These modify the signal through a direct manipulation of the independent
Communication Engineering Principles, Second Edition. Ifiok Otung.
© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 128 3 Time Domain Analysis of Signals and Systems

variable (which in this chapter is time t). It is assumed that the reader has a good grounding in the other basic
operations. These modify the signal through a direct manipulation of the dependent variable (which is the value
of the signal).
We then turn our attention to random signals and develop the basic time domain tools necessary to deal with
this very important class of signals. This is followed by a presentation of various standard distribution functions
with an emphasis on how the random variations modelled by each function arise in practical communication
systems. A detailed treatment of the most common time domain signal analysis and characterisation processes
applicable to both random and deterministic signals is then presented. This includes discussion of signal mean,
signal power, and rms value, signal energy, signal autocorrelation, and the covariance and correlation coefficient
between two signals. The rest of the chapter is then devoted to learning the craft of time domain analysis of linear
time invariant (LTI) systems with an emphasis on gaining a mastery of the convolution operation as applied to
continuous-time and discrete-time systems.

3.2 Basic Signal Operations

For convenience and clarity, but without any loss of generality, we discuss the operations of time shifting, time
reversal, and time scaling using the signal g(t) whose waveform is shown in Figure 3.1a. This signal has a value of
1 in the interval from t = −2𝜏 to 0, decreases linearly from a value of 1 to a value of 0 in the interval t = 0 to 𝜏, and
is zero everywhere else. Thus
⎧1, −2𝜏 ≤ t ≤ 0
⎪
g(t) = ⎨1 − t∕𝜏, 0≤t≤𝜏 (3.1)
⎪0, Elsewhere
⎩

3.2.1 Time Shifting (Signal Delay and Advance)

To uncover the waveform corresponding to the signal g(t − 𝜏), we visit the definition of g(t) in Eq. (3.1) and replace
t by t − 𝜏 wherever it occurs in that equation. Thus
⎧1, −2𝜏 ≤ t − 𝜏 ≤ 0
⎪ t−𝜏
g(t − 𝜏) = ⎨1 − , 0≤t−𝜏 ≤𝜏
⎪ 𝜏
⎩0, Elsewhere
Adding 𝜏 to each section of the inequalities on the right-hand side and simplifying 1 − (t − 𝜏)∕𝜏 to 1 − (t∕𝜏 − 1) =
2 − t∕𝜏 yields
Telegram: @Persian_IEEEComSoc

⎧1, −𝜏 ≤ t ≤ 𝜏
⎪ t
g(t − 𝜏) = ⎨2 − , 𝜏 ≤ t ≤ 2𝜏 (3.2)
⎪ 𝜏
⎩0, Elsewhere
This means that g(t − 𝜏) has a value of 1 in the interval t = (−𝜏, 𝜏), decreases linearly from a value of 1 to a value
of 0 in the interval t = (𝜏, 2𝜏), and is 0 everywhere else, which corresponds to the waveform shown in Figure 3.1b.
It is clear that g(t − 𝜏) is the original waveform g(t) shifted to the right through 𝜏 so that it starts (𝜏 units of time)
later. For positive 𝜏 (as implied in Figure 3.1), g(t − 𝜏) is the signal g(t) manipulated only by delaying it by 𝜏 without
altering its waveform size (i.e. amplitude), shape, or span (i.e. duration). The effect of delaying an arbitrary signal
g(t) by 𝜏 is thus to produce the signal g(t − 𝜏) in which every event (e.g. peak, trough, transition, etc.) that occurs
in g(t) at some time t = to also occurs within g(t − 𝜏) but at the later time t = to + 𝜏.
 3.2 Basic Signal Operations 129

g(t)
1
(a)
t
–3τ –2τ –τ 0 τ 2τ 3τ

g(t – τ)
1
(b)
t
–3τ –2τ –τ 0 τ 2τ 3τ

g(t + τ)
1
(c)
t
–3τ –2τ –τ 0 τ 2τ 3τ

Figure 3.1 Time-shift operations on signal g(t) in (a), including time-delay by 𝜏 in (b) and time-advance by 𝜏 in (c).

Similarly, we obtain the expression for the signal g(t + 𝜏) as

⎧1, −2𝜏 ≤ t + 𝜏 ≤ 0
⎪ t+𝜏
g(t + 𝜏) = ⎨1 − , 0≤t+𝜏 ≤𝜏
⎪ 𝜏
⎩0, Elsewhere
⎧1, −3𝜏 ≤ t ≤ −𝜏
⎪ t
= ⎨− , −𝜏 ≤ t ≤ 0 (3.3)
⎪ 𝜏
⎩0, Elsewhere
The waveform of g(t + 𝜏) specified in the above equation is plotted in Figure 3.1c, which shows that g(t + 𝜏) is
the original waveform g(t) shifted to the left through 𝜏, so it starts earlier. In general, regardless of the type of
functional dependence on t, the signal g(t + 𝜏) is simply the signal g(t) advanced in time by 𝜏 so that every event
in g(t) also occurs within g(t + 𝜏) but at 𝜏 units of time earlier, and with all waveform features of shape, size, and
span remaining unchanged.
The basic signal operations of time delay and time advance are very common and in fact vital in communication
systems. For example, sophisticated (digital) signal filtering is achieved by combining variously weighted and
Telegram: @Persian_IEEEComSoc

time-delayed versions of a signal to obtain a suitably filtered output. And if the system is non-real-time then the
combined components may even involve time-advanced versions of the signal since these ‘future’ samples are in
storage and therefore are available for use to produce an output at the present time.
Signal delay arises naturally in signal transmission through a medium. In this context it may be formally defined
as the time separation between an event in one signal and a corresponding event in a reference signal. The signal
in question y(t) may be the observation at one point of a transmission system (e.g. the received or output signal),
whereas the reference x(t) may be the same signal observed at an earlier point (e.g. input) of the system. The
delay then specifies the time taken for the signal to propagate between the two points. If the transmission system
is distortionless, there is a one-to-one correspondence between events in y(t) and x(t); each pair of events being
separated by a constant time 𝜏, which is the group delay. In this case

y(t) = Kx(t − 𝜏) (3.4)

 130 3 Time Domain Analysis of Signals and Systems

where K is a positive real constant scaling factor. That is, y(t) is a scaled (i.e. boosted or attenuated) and delayed
version of x(t). Equivalently, albeit less conventionally, by replacing t with t + 𝜏 wherever it occurs in the above
equation, we may write
1
x(t) = y(t + 𝜏)
K
which states that x(t) is the signal y(t) advanced by time 𝜏 and scaled by factor 1/K.
The angle of a sinusoidal signal increases at a rate of 2𝜋f rad/s and hence by 2𝜋f𝜏 rad in a time interval 𝜏.
Thus, there is a relationship between phase difference Δ𝜙 (in radian) and delay 𝜏 (in seconds). If x(t) and y(t) are
sinusoidal signals having the same frequency f (and period T = 1/f ) the two parameters are related by
( )
Δ𝜙 Δ𝜙
𝜏= = T (3.5)
2𝜋f 2𝜋
In a multipath transmission medium, such as terrestrial microcellular wireless environments, the signal reaches
the reception point via N paths – comprising one primary path (usually the direct and hence shortest path) and
(N − 1) secondary paths. In this case, delay is referenced to the primary signal. The delay of the ith path is referred
to as excess delay 𝜏 i . Clearly, the excess delay of the primary path is zero. In such a transmission medium, one
transmitted narrow pulse becomes a closely spaced sequence of N narrow pulses – effectively one broadened
pulse – at the receiver. This pulse broadening, or dispersion, places a limit on the symbol rate that can be used for
transmission in the medium if an overlap of adjacent symbols, a problem known as intersymbol interference (ISI),
is to be avoided. The duration of each pulse is extended by an amount equal to the medium’s total excess delay,
which is simply the excess delay 𝜏 imax of the last received significant path.
The delay profile of a multipath transmission medium is often characterised by two parameters, namely average
or mean delay 𝜏 avg and rms delay spread 𝜏 rms
∑
N
𝜏avg = 𝛼i 𝜏i
i=1
√
√N
√∑
𝜏rms = √ (𝛼i 𝜏i2 ) − 𝜏avg
2
(3.6)
i=1

where 𝛼 i is a ratio giving the power received via the ith path as a fraction of the total received power. Note that the
count of paths in Eq. (3.6) includes the primary or direct path for which 𝜏 i = 0. ISI is negligible, and the medium
is treated as narrowband if symbol duration T s > > 𝜏 rms ; otherwise, the medium is wideband and will be a source
of frequency-selective fading.

3.2.2 Time Reversal

The operation of time reversal is executed mathematically on a signal g(t) by replacing the independent variable t
Telegram: @Persian_IEEEComSoc

by −t wherever it occurs in the defining equation for g(t). Doing this for the signal g(t) defined earlier in Eq. (3.1),
whose waveform is plotted again in Figure 3.2a for convenience, we obtain
⎧1, −2𝜏 ≤ −t ≤ 0
⎪ −t
g(−t) = ⎨1 − , 0 ≤ −t ≤ 𝜏
⎪ 𝜏
⎩0, Elsewhere
Multiplying each section of the inequalities on the right-hand side by −1 flips each inequality and yields
⎧1, 0 ≤ t ≤ 2𝜏
⎪ t
g(−t) = ⎨1 + , −𝜏 ≤ t ≤ 0 (3.7)
⎪ 𝜏
⎩0, Elsewhere
 3.2 Basic Signal Operations 131

g(t)
1
(a)
t
–3τ –2τ –τ 0 τ 2τ 3τ
g(t – τ)
1
(b)
t
–3τ –2τ –τ 0 τ 2τ 3τ
g(t + τ)
1
(c)
t
–3τ –2τ –τ 0 τ 2τ 3τ
g(–t)
1
(d)
t
–3τ –2τ –τ 0 τ 2τ 3τ
g(–t – τ)
1
(e)
t
–3τ –2τ –τ 0 τ 2τ 3τ
g(–t + τ)
1
(f)
t
–3τ –2τ –τ 0 τ 2τ 3τ

Figure 3.2 Various time-shift and time-reversal operations on signal g(t).

This equation states that g(−t) has a value of 1 in the interval t = (0, 2𝜏), decreases linearly from a value of 1 at
t = 0 to a value of 0 at t = −𝜏 in the interval t = (−𝜏, 0), and is zero everywhere else. This variation corresponds to the
waveform shown in Figure 3.2d, from which a graphical process for time reversal is readily seen: the time-reversed
signal g(−t) has a waveform that is simply a mirror image of g(t), with the double faced mirror lying along the y
axis. That is, the waveform of g(−t) is the result of flipping the waveform of g(t) about the y axis. It follows from our
Telegram: @Persian_IEEEComSoc

definition of even and odd signals (see Eq. (2.8)) that an even signal is time-reversal invariant (i.e. it is unchanged
by the operation of time reversal), whereas an odd signal is changed only by a factor of −1 when subjected to time
reversal.
Time reversal and time shift are often combined in more advanced signal operations (e.g. convolution) whereby
a time-reversed signal is subsequently time-shifted, or a time-shifted signal is subsequently time-reversed. It is
therefore of practical interest to explore the forms of the signals g(−t − 𝜏) and g(−t + 𝜏). Substituting −t − 𝜏 for t
wherever it occurs in the definition of g(t) in Eq. (3.1), we obtain

⎧1, −2𝜏 ≤ −t − 𝜏 ≤ 0
⎪ −t − 𝜏
g(−t − 𝜏) = ⎨1 − , 0 ≤ −t − 𝜏 ≤ 𝜏
⎪ 𝜏
⎩0, Elsewhere
 132 3 Time Domain Analysis of Signals and Systems

Adding 𝜏 to each section of the inequalities on the right-hand side before multiplying through by −1 yields
⎧1, −𝜏 ≤ t ≤ 𝜏
⎪ t
g(−t − 𝜏) = ⎨2 + , −2𝜏 ≤ t ≤ −𝜏 (3.8)
⎪ 𝜏
⎩0, Elsewhere
This waveform is sketched in Figure 3.2e from which, by comparing Figure 3.2d and e, it can be seen that
g(−t − 𝜏) is the signal g(−t) advanced by 𝜏. That is, g(−t − 𝜏) is the result of time-reversing g(t) followed by a
time advance through 𝜏. Alternatively, by comparing Figure 3.2b and e, we see that g(−t − 𝜏) is the time reversal
of g(t − 𝜏). Thus, g(−t − 𝜏) may also be obtained from g(t) by first delaying g(t) by 𝜏 followed by a time reversal. In
general, time reversal of any signal followed by a time advance of 𝜏 is equivalent to time-delaying the signal by 𝜏
followed by time reversal. There is yet one other helpful way of viewing the combined operations of time reversal
and delay necessary to obtain g(−t − 𝜏) from g(t). Noting that the −t axis increases to the left, and by analogy with
g(t − 𝜏) which represents a shift of g(t) by 𝜏 to the right (i.e. in the direction of increasing t, which implies intro-
ducing a delay relative to t), we can state that g(−t − 𝜏) is the result of delaying g(−t) by 𝜏 relative to −t (i.e. shifting
g(−t) leftwards through 𝜏 in the direction of increasing −t).
It should now be a straightforward matter for you to proceed as above to obtain
⎧1, 𝜏 ≤ t ≤ 3𝜏
⎪t
g(−t + 𝜏) = ⎨ , 0≤t≤𝜏 (3.9)
⎪𝜏
⎩0, Elsewhere
This waveform is sketched in Figure 3.2f from which we see, by comparison with Figure 3.2d, that g(−t + 𝜏) is the
signal g(−t) delayed by 𝜏; or (by comparison with Figure 3.2c) that g(−t + 𝜏) is the time reversal of g(t + 𝜏). Thus,
the signal g(−t + 𝜏) may be obtained from g(t) either by first applying a time reversal followed by a delay of 𝜏, or
by first applying a time advance of 𝜏 followed by time reversal. In general, the time advance of any signal through
𝜏 followed by time reversal is equivalent to time reversal of the signal followed by a delay of 𝜏. Note again, since
the −t axis decreases to the right, it follows that g(−t + 𝜏) is an advance of g(−t) through 𝜏 relative to −t achieved
by shifting g(−t) rightward (in the direction of decreasing −t).
Letting 𝕋 ℝ[x] denote time reversal of signal x, 𝔻[x, 𝜏] denote time delay of x by 𝜏, and 𝔻[x, −𝜏] denote the time
advance of x by 𝜏, we may summarise the interrelationships and effects of various combinations of time shift and
time reversal operations as follows
g(t − 𝜏) = 𝔻[g(t), 𝜏] = 𝕋 ℝ[g(−t − 𝜏)]
g(t + 𝜏) = 𝔻[g(t), −𝜏] = 𝕋 ℝ[g(−t + 𝜏)]
g(t) = 𝔻[g(t + 𝜏), 𝜏] = 𝔻[g(t − 𝜏), −𝜏] = 𝕋 ℝ[𝕋 ℝ[g(t)]]
g(−t) = 𝕋 ℝ[g(t)] = 𝔻[g(−t + 𝜏), −𝜏] = 𝔻[g(−t − 𝜏), 𝜏]
Telegram: @Persian_IEEEComSoc

g(−t + 𝜏) = 𝔻[g(−t), 𝜏] = 𝕋 ℝ[g(t + 𝜏)] = 𝔻[𝕋 ℝ[g(t)], 𝜏] = 𝕋 ℝ[𝔻[g(t), −𝜏]]

g(−t − 𝜏) = 𝔻[g(−t), −𝜏] = 𝕋 ℝ[g(t − 𝜏)] = 𝔻[𝕋 ℝ[g(t)], −𝜏] = 𝕋 ℝ[𝔻[g(t), 𝜏]] (3.10)

3.2.3 Time Scaling

Time scaling is performed on a signal g(t) by multiplying the independent variable t, wherever it occurs in the
defining equation for g(t), by a real scale factor 𝛽 > 0. Carrying out this mathematical process on the signal g(t)
defined in Eq. (3.1), whose waveform is repeated in Figure 3.3a for ease of comparison, we obtain
⎧1, −2𝜏∕𝛽 ≤ t ≤ 0
⎪ t
g(𝛽t) = ⎨1 − 𝛽 , 0 ≤ t ≤ 𝜏∕𝛽, 𝛽>0 (3.11)
⎪ 𝜏
⎩0, Elsewhere
 3.2 Basic Signal Operations 133

g(t)
1
(a)
t
–6τ –5τ –4τ –3τ –2τ –τ 0 τ 2τ 3τ 4τ 5τ 6τ
g(βt), β = 1/3
1
(b)
t
–6τ –5τ –4τ –3τ –2τ –τ 0 τ 2τ 3τ 4τ 5τ 6τ
g(βt), β = 2
1
(c)
t
–6τ –5τ –4τ –3τ –2τ –τ 0 τ 2τ 3τ 4τ 5τ 6τ
0.5τ
g(βt), β = –1/3
1
(d)
t
–6τ –5τ –4τ –3τ –2τ –τ 0 τ 2τ 3τ 4τ 5τ 6τ
g(βt), β = –2
1
(e)
t
–6τ –5τ –4τ –3τ –2τ –τ 0 τ 2τ 3τ 4τ 5τ 6τ
–0.5τ

Figure 3.3 Time scaling of signal g(t).

The waveform of g(𝛽t) is sketched in Figure 3.3b for 𝛽 = 1/3, and in Figure 3.3c for 𝛽 = 2. Notice that the effect
of time scaling is to change the duration of the signal and the spacing (and hence rate of occurrence) of events
within the signal. For example, an event such as the transition from a value of 1 to a value of 0 that takes 𝜏 units of
time to complete in g(t) now takes 𝜏/𝛽 units of time to complete in the time-scaled signal g(𝛽t). In general, and for
any arbitrary signal g(t), when 𝛽 < 1 (as illustrated in Figure 3.3b) then g(𝛽t) is expanded in time by the factor 1/𝛽
when compared to g(t). This means that the duration of corresponding events (e.g. pulse width, cycle period, etc.)
is longer in g(𝛽t) than in g(t) by this factor (1/𝛽). Equivalently, we may state that the rates of occurrence of events
Telegram: @Persian_IEEEComSoc

(and hence the frequencies and bandwidth) of g(𝛽t) are those of g(t) reduced by a factor of 1/𝛽. Conversely, if 𝛽 > 1
(as illustrated in Figure 3.3c) then g(𝛽t) is compressed in time, durations of corresponding events are shorter in
g(𝛽t) than in g(t) by a factor of 𝛽, events occur faster, and the bandwidth of g(𝛽t) is 𝛽 times the bandwidth of g(t).
For a quick example, if a sinusoidal signal
x(t) = A cos(2𝜋ft + 𝜙)
is time-scaled by a factor 𝛽, we obtain another sinusoidal signal y(t) given by
y(t) = x(𝛽t) = A cos(2𝜋𝛽ft + 𝜙)
Notice that the frequency of y(t) is 𝛽 times that of x(t), whereas the period of y(t) is 1/𝛽f , which is 1/𝛽 times the
period of x(t). This observation is consistent with the above discussions on the impact of time scaling on frequency
and event durations.
 134 3 Time Domain Analysis of Signals and Systems

There is one important exception to the impact of time scaling, and that is when it is applied to a zero-duration
signal such as the impulse function 𝛿(t). To discover how 𝛿(𝛽t) is related to 𝛿(t), recall Eq. (2.29), which
defines the impulse function as a limiting case of the rectangular pulse. Substituting 𝛽t for t in this equation
yields
[ ( )] [ ( )]
1 𝛽t 1 t
𝛿(𝛽t) = lim rect = lim rect
𝜏→0 𝜏 𝜏 𝜏→0 𝜏 𝜏∕𝛽
The term in square brackets is a rectangular pulse of height 1/𝜏 and duration 𝜏/𝛽, which therefore has a constant
area of
1 𝜏 1
× = as 𝜏→0
𝜏 𝛽 𝛽
Since 𝛿(t) has a unit area and is an even function, it follows that
1
𝛿(𝛽t) = 𝛿(t), 𝛽≠0 (3.12)
|𝛽|
Thus, time scaling of 𝛿(t) by 𝛽 uniquely translates into amplitude scaling by 1/|𝛽|.
As earlier noted, the scale factor 𝛽 will be a nonzero and positive real number. But what if it is negative? To
explore this scenario, note that if 𝛽 < 0, then −𝛽 = |𝛽| so that Eq. (3.11) becomes

⎧1, 0 ≤ t ≤ 2𝜏∕|𝛽|
⎪ t
g(𝛽t) = ⎨1 + |𝛽| , −𝜏∕|𝛽| ≤ t ≤ 0 , 𝛽<0 (3.13)
⎪ 𝜏
⎩0, Elsewhere

The waveform of this signal is sketched in Figure 3.3d for 𝛽 = −1/3 and in Figure 3.3e for 𝛽 = −2,
from which we see that time-scaling using a negative scale factor introduces an extra operation of time
reversal.

3.3 Random Signals

We have so far dealt mostly with deterministic signals. However, random signals play a central role in digital
communications. For example, information signal is always treated as a random bit stream since the receiver does
not in general know ahead of time what bit will be in each interval. Furthermore, noise from a variety of sources
is inherent in all communication systems and is by nature random. We wish to provide in this section the basic
tools and concepts necessary for characterising random signals in communication systems.
Telegram: @Persian_IEEEComSoc

3.3.1 Random Processes

A random process is a collection or ensemble of sample realisations as illustrated in Figure 3.4. Each sample
realisation xk (t), k = 1, 2, 3, …, N, is a function of time and is the result of a trial of the random experiment
(i.e. observation of the random process) over a time interval 𝕋 . In general, no two sample realisations have an
identical sequence of values or waveform. Taking the value of each of the sample realisations at a fixed time
t = t1 within the observation interval results in a set of numbers {x11 , x21 , x31 , x41 , …, xN1 } which constitutes
a random variable, say X 1 . Repeating the exercise at another time t = t2 produces another random variable
X 2 = {x12 , x22 , x32 , x42 , …, xN2 }.
 3.3 Random Signals 135

x1(t)
x11
1st trial t
x12
x2(t) x21
2nd trial t
x22
x3(t)
x32
3rd trial t

x4(t) x31

4th trial t
x41 x42

Sample functions:
x1(t), x2(t), …, xN(t)

Random variables:
xN(t) X1 = {x11, x21, x31, x41, …, xN1}
X2 = {x12, x22, x32, x42, …, xN2}
xN2
Nth trial t
xN1
T
t = t1 t = t2
τ

Figure 3.4 Random process.

3.3.2 Random Signal Parameters

We cannot know with certainty in advance what value a random variable will take, but we can put forward a pos-
itive number on a scale from 0 to 1 that expresses the likelihood (i.e. probability) of the value being in a specified
range. A probability of 0 or 1 expresses certainty, with 0 indicating that the value of the random variable is certain
not to be in the given range, whereas 1 indicates that it is certain to be. More generally there is some uncertainty,
and this is indicated in the extent of the departure of the probability from the extreme ends of the scale. The prob-
ability Pr(X ≤ x) that a random variable X takes on some value less than or equal to x is known as the cumulative
distribution function (CDF) of the random variable, denoted PX (x). In this notation PX (x), the subscript X is the
Telegram: @Persian_IEEEComSoc

random variable, whereas the argument x is a dummy variable or parameter representing a realisation of X. Note
that the dummy variable could be any letter and does not have to be a lower case of the letter denoting the random
variable. The probability that X will have a value lying in the range (x1 , x2 ), where x2 > x1 , is therefore
Pr(x1 ≤ X ≤ x2 ) = PX (x2 ) − PX (x1 ) (3.14)
Considering an infinitesimal range (x, x + dx) and dividing Pr(x ≤ X ≤ x + dx) by the size dx of the range yields
the derivative of PX (x) which we denote as pX (x)
Pr(x ≤ X ≤ x + dx) PX (x + dx) − PX (x) dPX (x)
= ≡ ≡ pX (x) (3.15)
dx dx dx
The parameter pX (x) is known as the probability density function or probability distribution function (PDF) of the
random variable X. The PDF of X is in general not constant but varies with x in a way that reflects the relative
 136 3 Time Domain Analysis of Signals and Systems

likelihood of each x value. Equation (3.15) shows that pX (x)dx is the probability that X will have a value around x
in an infinitesimal range of size dx. That is
Pr(x ≤ X ≤ x + dx) = pX (x)dx
which is the area of a rectangle of height pX (x) and width dx. Thus, if we plot a graph of pX (x) versus x, then the
probability that X lies in a sizeable range (x1 , x2 ) will be the sum of the areas of all such rectangles from pX (x1 )dx
at x1 to pX (x2 )dx at x2 , which is simply the area under the PDF curve in that range. In other words
x2
Pr(x1 ≤ X ≤ x2 ) = pX (x)dx (3.16)
∫x1
Thus
x1
PX (x1 ) = Pr(−∞ < X ≤ x1 ) = pX (x)dx (3.17)
∫−∞
The CDF of a random variable is therefore the integral of its PDF, which is equivalent to the statement in
Eq. (3.15) that pX (x) is the derivative of PX (x). Since, by definition, probability is a positive number between 0
and 1, and we are certain that X has a value in the range (−∞, ∞), we may make the following characterising
statements about pX (x) and PX (x)
(i) pX (x) ≥ 0
∞
(ii) pX (x)dx = 1
∫−∞
(iii) 0 ≤ PX (x) ≤ 1
(iv) PX (∞) = 1
(v) PX (−∞) = 0
(vi) PX (x2 ) ≥ PX (x1 ) for x2 ≥ x1 (3.18)
Equation (3.18)(ii) indicates that the total area under a PDF curve is always 1, and Eq. (3.18)(vi) means that
PX (x) is monotonically non-decreasing.
Another distribution function of interest is the complementary cumulative distribution function (CCDF) or
exceedance distribution function. Denoted PX (x), this is the probability Pr(X > x) that the random variable X takes
on some value that exceeds x. Long-term statistical data on rainfall rate R (in mm/hr) and rain-induced path
loss L (in decibel, dB) are often analysed to obtain their CCDF, which gives the probability (usually expressed as
a percentage of time in an average year) that the random variable exceeds a specified level. The CCDF is often
presented as a graphical plot from which the value of the random variable exceeded with a given probability may
also be read. For example, the level of rain attenuation exceeded for 0.1% of time in an average year is needed to
design a super high frequency (SHF) link that can withstand rain impairments for 99.9% of the time in an average
Telegram: @Persian_IEEEComSoc

year. Using Eq. (3.18)(ii) we obtain the following relationship between the CCDF and CDF of a random variable
∞ ∞ x1
CCDF ≡ PX (x1 ) = pX (x)dx = pX (x)dx − pX (x)dx
∫x1 ∫−∞ ∫−∞
= 1 − PX (x1 ) ≡ 1 − CDF (3.19)
We may characterise a random variable X using various aggregate quantities, called moments of the random
variable. The nth moment of X is the expected value or expectation of X n , denoted E[X n ], which is the average
of all the realisations of X n in an infinite number of observations of X (or trials of the chance experiment). It is
worth emphasising the difference between sample mean of a quantity and the expectation of that quantity. Sample
mean is an estimate of the true mean obtained by averaging the quantity over a random subset or sample of the
population. Expectation is the true mean of the quantity obtained by averaging over the entire population. Thus
 3.3 Random Signals 137

E[X n ] is obtained by adding all the possible values of X raised to power n, each addition being weighted by the
relative likelihood of occurrence of the value. That is
∞
E[X n ] = xn pX (x)dx (3.20)
∫−∞
The first two moments are the most useful and respectively give the mean 𝜇X and mean-square value (i.e. total
power) of the random variable. The power of signals is discussed further in Section 3.5.
∞
E[X] ≡ 𝜇X = xpX (x)dx
∫−∞
∞
E[X 2 ] ≡ Total power = x2 pX (x)dx (3.21)
∫−∞
Another important characterising parameter is the second central moment of the random variable X. This is the
expected value of the squared deviation of X from its mean, i.e. the expected value of (X − 𝜇X )2 , which is more
commonly called variance and denoted 𝜎X2 or Var[X]. Since the expectation E[⋅] operator is linear, we note that
𝜎X2 = E[(X − 𝜇X )2 ] = E[X 2 − 2X𝜇X + 𝜇X2 ]
= E[X 2 ] − 2𝜇X E[X] + 𝜇X2 = E[X 2 ] − 2𝜇X2 + 𝜇X2
= E[X 2 ] − 𝜇X2 = E[X 2 ] − (E[X])2
≡ Total power − DC power (3.22)

Noting that total power comprises AC power and DC power components, it follows that the second moment
is the total power of the random variable, whereas the variance specifies AC power and the square of the mean
gives the DC power. The square root of variance is called the standard deviation 𝜎 X of the random variable X. To
further emphasise the physical significance of the variance parameter, note that if 𝜎X2 is zero it means that X has
a constant value 𝜇 X , and its PDF is therefore a unit impulse function located at 𝜇X . A nonzero value of 𝜎X2 is an
indication that X does take on values other than 𝜇 X . And the larger the value of 𝜎X2 , the more is the spread of the
values of X around its mean, and hence the broader is the PDF curve. Also, it follows from Eq. (3.22) that when a
random variable has zero mean then its variance equals the total power of the random variable.
In general, we can specify the expectation of any function of X, and not just its powers X n . Thus, denoting a
function of X as g(X), we have
∞
E[g(X)] = g(x)pX (x)dx (3.23)
∫−∞
In the special case where g(X) is the exponential function exp(j𝜔X), this expectation is known as the character-
istic function of X, denoted 𝜓 X (𝜔). Thus
∞
E[exp(j𝜔X)] ≡ 𝜓X (𝜔) =
Telegram: @Persian_IEEEComSoc

pX (x)ej𝜔x dx (3.24)
∫−∞
The characteristic function has many interesting properties and applications. To see one such application, let us
take the nth derivative of the above equation with respect to 𝜔, noting that it is only ej𝜔x that is a function of 𝜔 on
the right-hand side
∞ ∞
dn dn dn
𝜓X (𝜔) = pX (x)ej𝜔x dx = pX (x) n (ej𝜔x )dx
d𝜔 n n
d𝜔 −∞∫ ∫−∞ d𝜔
∞
= jn xn pX (x)ej𝜔x dx
∫−∞
∞
= jn xn pX (x)dx ≡ jn E[X n ] at 𝜔 = 0
∫−∞
 138 3 Time Domain Analysis of Signals and Systems

Thus
( )|
1 dn 𝜓X (𝜔) |
E[X n ] = | (3.25)
jn d𝜔n |
|𝜔=0
It will become obvious in the next chapter that if ℙX (𝜔) denotes the Fourier transform (FT) of pX (x), then the
right-hand side of Eq. (3.24) and hence 𝜓 X (𝜔) is simply ℙX (−𝜔). Thus, given its PDF pX (x), the characteristic
function of the random variable X can be read from widely available FT tables. The nth moment of X can then
be obtained by differentiation of 𝜓 X (𝜔) as in Eq. (3.25), which avoids the more difficult integration involved in
Eq. (3.20).
A real-valued random variable X that can take on values in the infinite interval (−∞, ∞) does not have a finite
peak value or amplitude since any value up to infinity is theoretically possible. However, we may define the peak
value Ap of such a random variable as the point at which its CDF is less than, say, 0.99. That is, PX (Ap ) ≤ 0.99, which
means that X takes on values less than this ‘peak’ with a probability of 0.99. Thus the ‘peak value’ so defined will
be exceeded 1% of the time on average.

3.3.3 Stationarity and Ergodicity

The task of characterising a random process is greatly simplified if the process is stationary. A random process is
said to be strict-sense stationary if its distribution functions do not change with time. However, if only its first two
moments are time-independent then the process is said to be wide-sense stationary (WSS). Defining the autocorre-
lation function of the random process as the expected value of the product of two random variables of the process,
e.g. X 1 and X 2 in Figure 3.4, a WSS random process therefore has a constant mean and an autocorrelation function
RX (𝜏) that depends only on the time shift 𝜏 = t2 − t1 between the samples but not on time t1 . Thus, if the random
process of Figure 3.4 is WSS, then

E[X1 ] = E[X2 ] = Mean of process

E[X1 X2 ] = RX (𝜏) (3.26)

A strict-sense stationary process may have a further attribute whereby, although its sample realisations cannot
be identical in waveform, it may turn out that they have identical statistical characterisations, and we can obtain
the statistical properties of the random process by observing any one of its sample realisations. Such a random
process is said to be ergodic. In this case we can replace the more difficult task of ensemble averaging (to obtain,
say, the mean of the process) with time averaging over any one of the observed sample realisations. For example,
if the random process of Figure 3.4 is ergodic then the mean 𝜇X and autocorrelation function RX (𝜏) – discussed in
Section 3.5.5 – can be obtained as follows
Telegram: @Persian_IEEEComSoc

𝕋 ∕2
1
𝜇X = lim xk (t)dt
𝕋 →∞ 𝕋 ∫−𝕋 ∕2
𝕋 ∕2
1
RX (𝜏) = lim xk (t)xk (t + 𝜏)dt
𝕋 →∞ 𝕋 ∫−𝕋 ∕2

k = 1, 2, 3, · · · , or N (3.27)

Notice that the random process can be treated simply as a nonperiodic power signal, allowing us to apply the
analysis tools that are developed in Section 3.5. We will henceforth assume that all random processes we deal with
are ergodic. This assumption applies to additive white Gaussian noise (AWGN), the type of random noise signal
usually assumed in communication systems.
 3.4 Standard Distribution Functions 139

3.4 Standard Distribution Functions

The performance of telecommunication systems and radio wave propagation is subject to randomness in the trans-
mission medium, intractability of a wide variety of contributing factors, and the nondeterministic behaviour of a
population of users. A statistical approach is therefore often necessary in the analysis and design of communi-
cation systems. In this section we discuss six of the most common standard distribution functions and how they
are employed to model and characterise random noise, signals, and events in various telecommunication system
scenarios.

3.4.1 Gaussian or Normal Distribution

When a random variable X is the result of a sufficiently large number of statistically independent and random
additive contributions, the central limit theorem stipulates that X will follow a Gaussian distribution. Also called
the normal distribution (because it occurs very frequently in real life), this distribution is fully characterised by
the mean 𝜇 X and variance 𝜎X2 of the random variable and is often denoted as  (𝜇X , 𝜎X2 ). Before delving into the
details of the normal distribution, it would be helpful to understand two example scenarios where it might arise.
Consider a random variable X defined as the count of the number of heads obtained when a fair coin is tossed
n times. Clearly, the value of X is random and can be any integer number from 0 to n. It is also the result of n
statistically independent and random contributions since the outcome of each toss is both random and unaffected
by other tosses. When n is very large, all the conditions of the central limit theorem are satisfied, and X will thus
follow a Gaussian distribution with mean 𝜇 X = n/2. However, by far the most relevant and important example to
our study is that of noise voltage as a random variable. Thermal noise, for example, results from the additive con-
tribution of many electrons, each moving randomly about an equilibrium position under thermal agitation. There
is no general drift of electrons in one direction in the conductor, this requiring some externally applied voltage.
If we observe the conductor over a significant period, we will find that at some instants the random motions lead
to a slight depletion of electrons in the upper half of the conductor, which gives a positive noise voltage. At other
instants there is a slight surplus of electrons in the upper half of the conductor, giving a negative noise voltage.
The average or mean of the noise voltage samples measured in the observation interval is zero since there is no
reason for either a surplus or depletion of electrons to be preferred. All the conditions of the central limit theorem
are satisfied, and the observed noise voltage will be a random variable X having a normal distribution with mean
𝜇 X = 0 and variance 𝜎X2 equal to the noise power.
The PDF of a Gaussian random variable  (𝜇X , 𝜎X2 ) is given by
( )
1 (x − 𝜇X )2
pX (x) = √ exp − (3.28)
2𝜋𝜎X2 2𝜎X2
Telegram: @Persian_IEEEComSoc

This PDF is plotted in Figure 3.5 for various values of mean 𝜇X and variance 𝜎X2 . Notice that it has a maximum
√
value of 1∕ 2𝜋𝜎X2 at x = 𝜇 X (i.e. the mean is the most likely value or mode of X). Furthermore, it decreases
symmetrically away from 𝜇 X in a log-quadratic or bell-shaped fashion, the rate of decrease being tampered by the
variance such that the PDF goes from being quite flat at 𝜎X2 → ∞ to being a unit impulse function 𝛿(x − 𝜇 X ) at
𝜎X2 → 0. Note that the value of pX (x) does not depend on the absolute value of x but on the deviation |x − 𝜇X | of
x from the mean. Note also that, as the PDF broadens with increasing variance, its height or peak value reduces
since its total area is always equal to 1. A standard Gaussian random variable Z is one that has zero mean and unit
variance. Thus, substituting 𝜇 X = 0 and 𝜎X2 = 1 in Eq. (3.28) gives the PDF of  (0, 1) as
1
pZ (z) = √ exp(−z2 ∕2) (3.29)
2𝜋
 140 3 Time Domain Analysis of Signals and Systems

0.4

0.3
pX(x)

0.2

0.1

0
–25 –20 –15 –10 –5 0 5 10 15 20 25
x

Figure 3.5 PDF pX (x) of Gaussian random variable  (𝜇X , 𝜎 X 2 ) for various values of mean 𝜇X and variance 𝜎 X 2 .

The CDF and CCDF of a Gaussian random variable cannot be obtained in closed-form, but we may express them
in terms of the Q-function Q(x) as follows by using Eq. (3.17) in line 1 below, the substitution z ≡ (x − 𝜇X )/𝜎 X in
line 2, and property (ii) of Eq. (3.18) in the last line
( )
x1
1 (x − 𝜇X )2
PX (x1 ) = √ exp − dx
∫−∞ 2𝜎X2
2𝜋𝜎X2
(x1 −𝜇X )∕𝜎X
1 2
= √ exp(−z ∕2)dz
∫−∞ 2𝜋
∞ ∞
1 2 1 2
= √ exp(−z ∕2)dz − √ exp(−z ∕2)dz
∫−∞ 2𝜋 ∫(x1 −𝜇X )∕𝜎X 2𝜋
( )
x1 − 𝜇X
=1−Q (3.30)
𝜎X
Thus,  (𝜇X , 𝜎X2 ) has CDF and (from Eq. (3.19)) CCDF given by
( )
x − 𝜇X
CDF ≡ Pr(X ≤ x) ≡ PX (x) = 1 − Q
𝜎X
( )
x − 𝜇X
CCDF ≡ Pr(X > x) ≡ PX (x) = Q (3.31)
𝜎X
Telegram: @Persian_IEEEComSoc

where
∞
1 2
Q(z1 ) = √ exp(−z ∕2)dz (3.32)
∫z1 2𝜋
is the probability that a zero-mean, unit-variance Gaussian random variable has a value that exceeds z1 . Values of
Q(x) may be read from widely available tables such as provided in Appendix C.3 for positive values of x. To obtain
values of Q(x) for negative arguments, use the relation
Q(x) = 1 − Q(−x) (3.33)
For example, Q(−2) = 1 − Q(2) = 1 − 0.0228 = 0.9772. We know that the most likely value of X is the mean 𝜇X .
So, what is the probability that X will take on a value that exceeds the mean by, say, m standard deviations, where
 3.4 Standard Distribution Functions 141

pX(x)

Pr(X ≤ μX – x1)
Area = 1 – 2a =
= Pr(X > μX + x1) Pr(X > μX + x1)
Pr (μX – x1 ≤ X ≤ μX + x1)

Area = a Area = a
x
μX – x1 μX μX + x1

Figure 3.6 Even symmetry of Gaussian PDF pX (x) about its mean 𝜇X .

m is any real number? This is Pr(X > 𝜇 X + m𝜎 X ), which is obtained using Eq. (3.31) as
( )
𝜇X + m𝜎X − 𝜇X
Pr(X > 𝜇X + m𝜎X ) = Q = Q(m)
𝜎X
The table of Q-function values (in Appendix C) gives Q(1) = 0.15866, Q(2) = 0.02275, Q(3) = 0.00135. Consid-
ering Figure 3.6 and the even symmetry of the normal distribution about its mean, it means that X lies within m
standard deviations of the mean with probability 1 − 2Q(m). Thus, 68.27%, 95.45%, and 99.73% of the samples of
X respectively lie within one, two, and three standard deviations of the mean.
In addition to the Q-function, the Gaussian CDF and CCDF are also often expressed in terms of the complemen-
tary error function erfc(x), which is defined by the integral
∞
2
erfc(x) = √ exp(−y2 )dy (3.34)
𝜋 ∫x
To obtain these equivalent expressions, consider the following extract from the third and fourth lines of Eq. (3.30)
( )
x − 𝜇X ∞
1
Q 1 = 2
√ exp(−z ∕2)dz
𝜎X ∫(x1 −𝜇X )∕𝜎X 2𝜋
√
Making the substitution y = z∕ 2 in the right-hand side yields
( ) √
x1 − 𝜇X ∞
1 1 2
∞
Q = √ exp(−y2 ) 2dy = √ exp(−y2 )dy
𝜎X ∫ x1 −𝜇√X 2𝜋 2 𝜋 ∫ x1 −𝜇√X
𝜎X 2 𝜎X 2
( )
1 x 1 − 𝜇 X
= erfc √
2 𝜎 2
Telegram: @Persian_IEEEComSoc

X
Thus
( )
1 x √
Q(x) = erfc √ ; erfc(x) = 2Q( 2x) (3.35)
2 2
It follows from Eqs. (3.31) and (3.35) that
( )
1 x − 𝜇X
CDF ≡ Pr(X ≤ x) ≡ PX (x) = 1 − erfc √
2 2𝜎X
( )
1 x − 𝜇X
CCDF ≡ Pr(X > x) ≡ PX (x) = erfc √ (3.36)
2 2𝜎X
 142 3 Time Domain Analysis of Signals and Systems

The complementary error function is also extensively tabulated in Appendix C and will be our preferred function
when discussing bit error probability in digital transmission systems. For negative arguments (not covered in the
complementary error function table), use the relation
erfc(x) = 2 − erfc(−x) (3.37)
The tabulated values of erfc(x) and Q(x) provided in Appendix C are accurate. However, if you prefer a direct
calculation or if your x value is not covered in the table then you may use the following formulas which give results
that are remarkably accurate to within 0.275%. For Q(x) the recommended formula is [1]
exp(−x2 ∕2)
Q(x) = √ √ , x≥0 (3.38)
2𝜋(0.661x + 0.339 x2 + 5.51)
which may be manipulated, using Eq. (3.35), into the following formula for erfc(x)
exp(−x2 )
erfc(x) = √ √ , x≥0 (3.39)
𝜋(0.661x + 0.339 x2 + 2.755)
For x > > 1, these formulas simplify to the following approximations
exp(−x2 ∕2) exp(−x2 )
Q(x) ≅ √ , erfc(x) ≅ √ ; x≫1 (3.40)
2𝜋x 𝜋x

Worked Example 3.1 Gaussian Noise

Determine the following for a Gaussian noise voltage vn (t) of standard deviation 𝜎 = 2.5 mV:
(a) Noise power Pn in dBm.
(b) The probability that a sample of vn (t) lies between 𝜎 and 2𝜎.
(c) The probability that a sample of vn (t) exceeds −3.0 mV.
( )
6.25 × 10−6
Pn = 𝜎 2 = (2.5 × 10−3 )2 = 6.25 𝜇W = 10log10 dBm
(a) 1 × 10−3
= −22 dBm
(b) Denoting vn (t) ≡ X and 𝜎 ≡ 𝜎 X for convenience, the required probability is Pr(𝜎 X ≤ X ≤ 2𝜎 X ). We know from
Eqs. (3.14) and (3.30) that
( ) ( ( ))
x2 − 𝜇X x1 − 𝜇X
Pr(x1 ≤ X ≤ x2 ) = PX (x2 ) − PX (x1 ) = 1 − Q − 1−Q
𝜎 𝜎X
( ) ( )X
x1 − 𝜇X x2 − 𝜇X
=Q −Q
𝜎X 𝜎X
Substituting 𝜇 X = 0, x1 = 𝜎 X , and x2 = 2𝜎 X yields the required probability as
( ) ( )
𝜎X 2𝜎X
Pr(𝜎X ≤ X ≤ 2𝜎X ) = Q −Q = Q(1) − Q(2)
𝜎X 𝜎X
Telegram: @Persian_IEEEComSoc

= 0.1587 − 0.0228 = 0.1359

where we have read values of Q(1) and Q(2) from the Q-function table.
(c) What we are to determine here is the CCDF Pr(X > −3.0 mV), which, from Eqs. (3.31) and (3.33), with 𝜇X = 0,
x = −3 mV, and 𝜎 X = 2.5 mV, is
( )
x − 𝜇X
Pr(X > x) = Q
𝜎X
( )
−3 mV
Pr(X > −3 mV) = Q = Q(−1.2)
2.5 mV
= 1 − Q(1.2) = 1 − 0.1151 = 0.8849
 3.4 Standard Distribution Functions 143

3.4.2 Rayleigh Distribution

The Rayleigh distribution arises in terrestrial wireless transmission media when the receiver does not have a direct
line-of-sight to the transmitter. In this scenario, the received signal R will be the resultant of n multipath signals or
rays, each of which is produced by a distinguishable group of scatterers in the medium and may be resolved into
an in-phase component as well as a quadrature component. Because of random variations in the magnitude and
phase of each ray due to small fluctuations in attenuation and delay, these n in-phase and quadrature components
add to produce zero-mean random variables X and Y , respectively, which have equal power and are the real and
imaginary parts of R. For a sufficiently large number of scatterers n, the central limit theorem is satisfied so that
X and Y are independent zero-mean equal-variance Gaussian random variables, and the envelope of the received
signal will follow a Rayleigh distribution.
It is worth emphasising here that, in practice (and in the absence of a dominant direct ray), the received signal
envelope will be acceptably well modelled by a Rayleigh distribution even for a relatively small number n of mul-
tipath signals or scattered rays. To demonstrate this, Figure 3.7 shows the results of a simulation experiment in
which n sinusoids having equal amplitude and a uniformly distributed random phase were combined. The PDF
of the resultant magnitude or envelope R is plotted along with a Rayleigh PDF for comparison. Notice that for
n ≥ 5, the Rayleigh PDF gives a reasonably good fit to the observed distribution of the received signal envelope.
Therefore, this model is usually assumed for the received signal envelope in situations (commonly encountered
in terrestrial macrocellular wireless communication systems) where there is no line-of-sight between a transmit-
ter and the receiver. As a useful exercise, by counting the rectangular grids estimate the area under each PDF in
Figure 3.7. Is your result as expected?
One scenario in which the occurrence of the Rayleigh distribution is absolutely assured is in the envelope of
thermal noise voltage in conductors. Noise in communication systems is actually a complex random variable
R = X + jY , where X is an in-phase zero-mean Gaussian random variable of variance 𝜎X2 that gives rise to a real or
active noise power 𝜎X2 , and Y is a quadrature-phase zero-mean Gaussian random variable of variance 𝜎Y2 that pro-
duces a reactive or imaginary noise power 𝜎Y2 . This is simply as a result of thermal agitation (for example) producing
a random noise voltage that is (to randomly varying degrees) out of step with the cycles of a reference wanted sig-
nal and may therefore be resolved into an in-phase component which is either in-step or exactly half-cycle out
of step, and a quadrature component that is exactly one-quarter cycle (i.e. ±90∘ ) out of step. Clearly, there is no
reason for either phase shift to be preferred in the random agitation of countless electrons, so X and Y are of equal
power and we adopt 𝜎X2 = 𝜎Y2 ≡ 𝜎 2 . The relationship between R, X, and Y is illustrated in Figure 3.8, where r is the
magnitude of a sample of the complex noise, commonly referred to as the natural envelope (or simply envelope) of
the noise, 𝜓 is the phase of that sample and is the value of another random variable Ψ which specifies the phase
of the complex noise, and x and y are the in-phase and quadrature components of R and therefore the respective
values of X and Y . Thus
√
r = x2 + y2 ; 𝜓 = tan−1 (y∕x) (3.41)
Telegram: @Persian_IEEEComSoc

We are interested in determining the PDFs pR (r) and pΨ (𝜓) of the envelope and phase of the complex noise.
Since the random variables X and Y are independent, it is intuitively obvious that R and Ψ are also independent.
In Figure 3.8, the probability that a sample of the complex noise R will lie in the shaded elemental area dA =
dxdy = rd𝜓dr is therefore
Pr(x ≤ X ≤ x + dx, y ≤ Y ≤ y + dy) = Pr(x ≤ X ≤ x + dx) ⋅ Pr(y ≤ Y ≤ y + dy)
= pX (x)dx ⋅ pY (y)dy
= Pr(r ≤ R ≤ r + dr, 𝜓 ≤ Ψ ≤ 𝜓 + d𝜓)
= Pr(r ≤ R ≤ r + dr) ⋅ Pr(𝜓 ≤ Ψ ≤ 𝜓 + d𝜓)
= pR (r)dr ⋅ pΨ (𝜓)d𝜓
 pR(r)
1.3 0.5 0.4

n =3 n =4 n =5
1.0 0.3

0.25 0.2
0.5

0.1

0 0 0
0 1 2 3 0 1 2 3 4 0 1 2 3 4 5
pR(r)
0.28 0.2

0.3 n = 10 n = 20
n =6
0.2

0.2
0.1

0.1
0.1

0 0 0
0 1 2 3 4 5 5.8 0 2 4 6 8 0 2 4 6 8 10 12
r→ r→ r→
Telegram: @Persian_IEEEComSoc

Figure 3.7 PDF pR (r) of the resultant amplitude or envelope when n sinusoids of equal amplitude and uniformly distributed random phase are combined. A
Rayleigh PDF is shown in dotted line for comparison.
 3.4 Standard Distribution Functions 145

Figure 3.8 Envelope r, phase 𝜓, in-phase component x, and

quadrature component y of complex noise.

Imaginary axis

rdψ
dA
dy
r dx
dψ y
ψ
x Real axis

Substituting the Gaussian PDFs of X and Y as given in Eq. (3.28) into the second line above, bearing in mind
that 𝜇 X = 𝜇 Y = 0 and 𝜎X2 = 𝜎Y2 ≡ 𝜎 2 , we obtain
( ) ( )
1 x2 1 y2
pR (r)dr ⋅ pΨ (𝜓)d𝜓 = √ exp − 2 dx ⋅ √ exp − 2 dy
2𝜋𝜎 2 2𝜎 2𝜋𝜎 2 2𝜎
( 2 )
1 x + y2
= 2
exp − dxdy
2𝜋𝜎 2𝜎 2
( )
1 r2
= exp − rdrd𝜓
2𝜋𝜎 2 2𝜎 2
( )
r r2 1
= 2 exp − 2 dr ⋅ d𝜓
𝜎 2𝜎 2𝜋
Comparing corresponding terms on the left-hand side and the last line of the right-hand side yields
( )
r r2
pR (r) = 2 exp − 2 , r ≥ 0
𝜎 2𝜎
1
pΨ (𝜓) = , −𝜋 ≤ 𝜓 ≤ 𝜋 (3.42)
2𝜋
pR (r) specifies the PDF of a Rayleigh distribution, whereas pΨ (𝜓) is a uniform distribution. The conditions r ≥ 0
and −𝜋 ≤ 𝜓 ≤ 𝜋 are written into Eq. (3.42) as an important reminder that the Rayleigh distribution is only appli-
cable to a positive real random variable and that 𝜓 is an angle in the range −180∘ to 180∘ . Alternatively, we may
invoke the unit step function (Eq. (2.12)) to cover this constraint on the Rayleigh distribution by writing
( )
r r2
pR (r) = 2 exp − 2 u(r) (3.43)
𝜎 2𝜎
The Rayleigh distribution is completely characterised by a single scale parameter 𝜎, which is the standard devia-
tion of the underlying constituent Gaussian distributed in-phase and quadrature random variables. Regarding the
Telegram: @Persian_IEEEComSoc

uniform distribution, it is worth pointing out that, since the area under a PDF curve is 1, the PDF of a uniformly
distributed random variable U that takes on continuous values in the range (u1 , u2 ) is given by
1
pU (u) = , u1 ≤ u ≤ u2 (3.44)
u2 − u1
Integrating the expression for pR (r) yields the Rayleigh CDF as
r r ( )
1 z2
PR (r) = pR (z)dz = 2 z exp − 2 dz
∫0 𝜎 ∫0 2𝜎
To evaluate this integral, recall that
d −ar2 2
e = −2are−ar
dr
 146 3 Time Domain Analysis of Signals and Systems

which when both sides are integrated and a = 1∕2𝜎 2 is substituted, gives the useful result
2 ∕2𝜎 2 2 ∕2𝜎 2
re−r dr = −𝜎 2 e−r (3.45)
∫
that we employ to evaluate the integral for PR (r) to obtain
[ ( )]|r ( )0
1 z2 | z2 ||
PR (r) = 2 −𝜎 2 exp − 2 | = exp − 2 |
𝜎 2𝜎 | 2𝜎 ||r
|0
Thus
( )
r2
CDF ≡ Pr(R ≤ r) ≡ PR (r) = 1 − exp − 2
2𝜎
( )
r2
CCDF ≡ Pr(R > r) = 1 − CDF = exp − 2 (3.46)
2𝜎
The characteristic values of the Rayleigh distribution (derived below in Worked Example 3.2) are as follows
√
Median = 𝜎 2 ln 2 = 1.17741𝜎 (3.47)
√
Mean = 𝜎 𝜋∕2 (3.48)
Mean square value ≡ Power = 2𝜎 2 (3.49)
√
Rms value = 2𝜎 (3.50)
( )
𝜋
Variance = 2 − 𝜎2 (3.51)
2
Mode = 𝜎 (3.52)
1 −1∕2 0.60653
Max PDF value = e = (3.53)
𝜎 𝜎
Note that the median of a random variable is the value at which its CDF = 0.5, which means that half of the
population of the samples of the random variable will be less than its median and the other half will be larger.
Figure 3.9a and b show the Rayleigh distributions for 𝜎 = 1 and 𝜎 = 2, including the PDF in (a) and the cumulative
and exceedance distribution functions in (b). You may again wish to use a count of the tiny rectangular grids in
Figure 3.9a to estimate the area under each PDF and to compare your finding with the expected result.

Worked Example 3.2

Determine the median, mean, mean square value, variance, mode, rms, and maximum PDF value for a Rayleigh
distributed random variable R in terms of the scale parameter 𝜎 of the distribution. This is an optional worked
example.
̃ R , and noting that CDF = 0.5 at the median, it means that PR (𝓂
Denoting the median value of R as 𝓂 ̃ R ) = 0.5,
so that Eq. (3.46) gives
Telegram: @Persian_IEEEComSoc

( ) ( ) ( )
𝓂̃ 2R 𝓂̃ 2R 1 𝓂̃ 2R
1 − exp − 2 = 0.5; ⇒ exp − 2 = ; ⇒ exp =2
2𝜎 2𝜎 2 2𝜎 2
Taking the natural logarithm of both sides gives 𝓂
̃ 2R ∕2𝜎 2 = ln 2, from which we obtain the desired expression
√
̃ R = 𝜎 2 ln 2
for median as: 𝓂
The mean 𝜇 R is given by the equation
∞ ∞
r −r2 ∕2𝜎 2
𝜇R = r ⋅ pR (r)dr = r⋅ e dr
∫−∞ ∫0 𝜎2
r −r 2 ∕2𝜎 2
Using integration by parts ∫ u ⋅ dv = uv − ∫ du ⋅ v with u ≡ r, dv ≡ 𝜎2
e
[( )|∞ ) ]
∞(
1 −r 2 ∕2𝜎 2 | −r 2 ∕2𝜎 2
𝜇R = 2 r⋅ re dr | − re dr dr
𝜎 ∫ | ∫0 ∫
|0
 3.4 Standard Distribution Functions 147

0.6
σ=1

0.5

Mode

Mean
0.4 Median
pR (r)

0.3
σ=2
0.2
Mean
Mode

0.1
Median

0
0 1 2 3 4 5 6 7 8
r→
(a)
1

CDF
0.8

0.6
Probability

0.4

CCDF
0.2

0
0 1 2 3 4 5 6 7 8
r→
(b)

Figure 3.9 (a): Probability distribution function (PDF) pR (r) of Rayleigh distribution for 𝜎 = 1, 2; (b): Cumulative distribution
function (CDF) and exceedance distribution function (CCDF) of Rayleigh distribution for 𝜎 = 1, 2.

Using Eq. (3.45) to evaluate the indefinite integrals yields

[ ( )∞ ]
2 |
∞ ∞
1
Telegram: @Persian_IEEEComSoc

𝜇R = 2 − r𝜎 2 e−r ∕2𝜎 || + 𝜎 2
2 2 2 2 2
e−r ∕2𝜎 dr = 0 + e−r ∕2𝜎 dr
𝜎 |0 ∫0 ∫0
( 2 )
∞
But ∫−∞ √1 exp − 2𝜎r 2 dr = 1, being the area of a zero-mean Gaussian PDF. Also, considering the even sym-
𝜎 2𝜋
metry of this PDF, it follows that
∞ ( ) ∞ ( )
1 r2 1 r2
√ exp − dr = 2 √ exp − dr = 1
∫−∞ 𝜎 2𝜋 2𝜎 2 ∫0 𝜎 2𝜋 2𝜎 2
( )
𝜎√
∞
r2
⇒ exp − 2 dr = 2𝜋
∫0 2𝜎 2
√
Hence, mean 𝜇R = 𝜎 𝜋∕2
Mean square value is the expectation of R2 and is obtained as
∞ ∞
r −r2 ∕2𝜎 2
E[R2 ] = r 2 ⋅ pR (r)dr = r2 ⋅ e dr
∫−∞ ∫0 𝜎2
 148 3 Time Domain Analysis of Signals and Systems

Using integration by parts (as previously) and then Eq. (3.45), we obtain
[( )|∞ ( ) ]
∞
1 2 2 | 2 2
E[R2 ] = 2 r2 ⋅ re−r ∕2𝜎 dr | − 2r ⋅ re−r ∕2𝜎 dr dr
𝜎 ∫ | ∫0 ∫
|0
∞
2 2 |∞ 2 2 2 2 |0
= −(r 2 e−r ∕2𝜎 )| + 2 re−r ∕2𝜎 dr = 0 +(2𝜎 2 e−r ∕2𝜎 )|
|0 ∫0 |∞
= 2𝜎 2
Variance is given by Eq. (3.22) as
𝜎R2 = E[R2 ] − 𝜇R2
Therefore, using the previous two results, we obtain
( )
𝜋 𝜋
𝜎R2 = 2𝜎 2 − 𝜎 2 = 𝜎 2 2 −
2 2
Mode 𝓂 ̂ R is the most likely value of the random variable. This is therefore the value of r at which pR (r) is
maximum and may be determined by setting the derivative of pR (r) to zero and solving for r. That is
( ) [ ( )]
d d r −r2 ∕2𝜎 2 r −r2 ∕2𝜎 2 2r 1 2 2
pR (r) = e = e − + 2 e−r ∕2𝜎
dr dr 𝜎 2 𝜎2 2𝜎 2 𝜎
At r = 𝓂 ̂ R , this derivative is zero. Thus
[ ( )]
𝓂 ̂ R −𝓂̂ 2 ∕2𝜎 2 2𝓂 ̂R 1
+ 2 e−𝓂̂ R ∕2𝜎 = 0
2 2
e R −
𝜎 2 2𝜎 2 𝜎
( )
2 𝓂
̂ 2
1
⇒ e−𝓂̂ R ∕2𝜎 − 4R + 2 = 0
2 2

2𝜎 𝜎
𝓂
̂ 2R 1 𝜎4
⇒ = ; ⇒ 𝓂 ̂ 2R = 2 = 𝜎 2
𝜎4 𝜎 2 𝜎
Hence, mode 𝓂 ̂R=𝜎 √
Rms value (denoted Arms ) is simply the square root of the mean square value obtained above. Thus, Arms = 2𝜎
By substituting r = 𝓂 ̂ R into the PDF expression, we obtain the maximum value of the Rayleigh PDF as
( )
𝜎 𝜎2
pR (r)max = pR (𝓂 ̂ R ) = pR (𝜎) = 2 exp − 2
𝜎 2𝜎
( )
1 1 0.60653
= exp − =
𝜎 2 𝜎

3.4.3 Lognormal Distribution

In the absence of terrain features and scatterers, the average power Pav of a received radio signal decreases inversely
Telegram: @Persian_IEEEComSoc

with distance d far from the transmitter according to d−m , where m = 2 in free space and m = 4 in a plane-earth-only
scenario. This would mean that all locations at the same radial distance d from the transmitter receive the same
power, which is not the case in practice. The presence of terrain features, including hills, trees, buildings, and other
obstacles will, one after the other, interfere with the signal or expend a portion of its energy through absorption,
diffraction, or scattering. This introduces several levels of variation of the received signal in a terrestrial wireless
environment.
The received signal power averaged at a small locality at distance d from the transmitter, called the local-mean
power, will vary between locations that have the same distance d from an isotropic transmit-antenna since the
environmental interventions experienced by the signal will be location dependent and no longer just distance
dependent.
If we define Pav as the long-term mean power averaged over all outdoor locations at a distance d from an
isotropic transmit-antenna, we find m > 4 so that Pav decreases more rapidly with distance than in the free space
and plane-earth-only environments.
 3.4 Standard Distribution Functions 149

The local-mean power will exhibit a long-term random variation about the above-average power, an effect known
as shadow fading. This is because of several contributory randomly varying multiplicative loss factors imposed by
multiple objects in the signal’s transmission path.
Reflections and scattering from these objects will give rise to multiple received rays, which adds an extra dimen-
sion of variation in received signal power on a small-scale of a few wavelengths (called multipath fading or fast
fading) due to random short-term fluctuations in the instantaneous received signal envelope, which may be mod-
elled by a Rayleigh distribution if there is no dominant direct ray, as discussed in the previous subsection.
The impact of shadow fading on local-mean power may be accounted for by expressing this power (in watts) as
the product of the average (d−m -dependent) power and n multiplicative independent random loss factors l1 l2 l3 …ln .
When expressed in logarithmic units, we see that the dBm power is the result of additive contributions from n
independent random variables L1 , L2 , L3 , …, Ln , where Lk = 10log10 lk . For large n, all conditions of the central
limit theorem are satisfied and the dBm power values will have a Gaussian distribution.
In general, if when we take the logarithm (to whatever base b) of the values x of a random variable X we find that
logb (X) follows a Gaussian distribution then X is said to have a lognormal distribution. In addition to the lognormal
variation of local-mean power highlighted above, any natural phenomenon in which the observed quantity is the
result of a build-up of numerous and independent small percentage changes (such as gain or loss factors) may be
modelled by the lognormal distribution. This distribution has therefore been widely used to model variations in
signal power, time duration of complex events, rates, length, and size in many fields, including biology, chemistry,
medicine, hydrology, social sciences, and engineering.
The PDF pX (x) of a lognormal random variable X may be derived straightforwardly as follows. If a random
variable X has a lognormal distribution, it means that the random variable Y ≡ ln(X), where ln denotes logarithm
to the base of the constant e = 2.718281828459…, has a Gaussian distribution  (𝜇Y , 𝜎Y2 ) which we denote as
 (𝜇ln X , 𝜎ln
2
X
) to emphasise that the mean and variance are computed on log values of X. Thus
pX (x)dx ≡ Pr(x ≤ X ≤ x + dx)
= Pr(ln x ≤ Y ≤ ln(x + dx)) = Pr[ln x ≤ Y ≤ ln(x(1 + dx∕x))]
= Pr(ln x ≤ Y ≤ ln x + ln(1 + dx∕x)) = Pr(ln x ≤ Y ≤ ln x + dx∕x)
ln x+dx∕x
= pY (y)dy
∫ln x
1
= pY (ln x)dx
x
where, in line 3 above, we have used the relation ln(1 + a) ≈ a for a ≪ 1, and in the penultimate line we have
evaluated the integral by noting that the integration range of size dx/x is so small that the integral is simply the
area of a rectangle of height pY (lnx) and width dx/x. Comparing the left-hand side and the last line of the right-hand
side
1
pX (x) = pY (ln x)
Telegram: @Persian_IEEEComSoc

x
Now replacing pY (lnx) using its PDF pY (y) as given in Eq. (3.28), we obtain
( ( )2 )
1 1 ln x − 𝜇ln X
pX (x) = √ exp − u(x) (3.54)
x𝜎ln X 2𝜋 2 𝜎ln X
This is the PDF of a lognormal random variable X, where the unit step function u(x) has been introduced
above to explicitly state that pX (x) = 0 for x < 0. A lognormal random variable X can only have positive val-
ues in the range (0, ∞) and its distribution is fully characterised by two parameters 𝜇lnX and 𝜎 lnX , which are,
respectively, the mean and standard deviation of the log-values of X. The CDF and CCDF of X are obtained as
follows
( ( )2 )
1 ln y − 𝜇ln X
x
1 1
CDF ≡ PX (x) = √ exp − dy
𝜎 ln X 2𝜋 ∫0 y 2 𝜎ln X
 150 3 Time Domain Analysis of Signals and Systems

Substituting z ≡ (ln y − 𝜇ln X )∕𝜎ln X , we have dz∕dy = 1∕y𝜎ln X or dy = y𝜎ln X dz, and the limits of integration
y = (0, x) become z = (−∞, (ln x − 𝜇ln X )∕𝜎ln X ), since ln 0 = −∞. The above integration therefore simplifies to
(ln x−𝜇ln X )∕𝜎ln X ( 2)
1 z
CDF ≡ PX (x) = √ exp − dz
∫
2𝜋 −∞ 2
∞ ( 2) ∞ ( 2)
1 z 1 z
= √ exp − dz − √ exp − dz
2𝜋 ∫−∞ 2 2𝜋 ∫(ln x−𝜇ln X )∕𝜎ln X 2
( )
ln x − 𝜇ln X
=1−Q
𝜎ln X
Hence
( ) ( ) ( )
ln x − 𝜇ln X 1 ln x − 𝜇ln X 1 ln x − 𝜇ln X
CDF = 1 − Q = 1 − erfc √ = erfc − √
𝜎ln X 2 𝜎ln X 2 2 𝜎ln X 2
( ) ( )
ln x − 𝜇ln X 1 ln x − 𝜇ln X
CCDF = 1 − CDF = Q = erfc √ (3.55)
𝜎ln X 2 𝜎 2
ln X
Some of the characteristic values of a lognormal random variable X are derived in the next worked example. The
results are summarised below for convenience
Median = exp(𝜇ln X ) (3.56)
( )
𝜎ln
2
X
Mean = exp 𝜇ln X + (3.57)
2
Mean square value = exp[2(𝜇ln X + 𝜎ln
2
X )] (3.58)
Variance = exp(2𝜇ln X + 𝜎ln
2 2
X )[exp(𝜎ln X ) − 1] (3.59)
Mode = exp(𝜇ln X − 𝜎ln
2
X) (3.60)
Rms value = exp(𝜇ln X + 𝜎ln
2
X) (3.61)
exp(−𝜇ln X + 𝜎ln
2
X
∕2)
Max PDF value = √ (3.62)
𝜎ln X 2𝜋
Figure 3.10 shows various plots of the lognormal PDF to illustrate the impact of its two model parameters 𝜇lnX
and 𝜎 lnX . The skewness of the distribution is controlled entirely by 𝜎 lnX . As 𝜎 lnX decreases (left to right of bottom
row of Figure 3.10), we notice the following trends: (i) the lognormal PDF becomes increasingly symmetrical about
the mode; (ii) the mean and median coalesce towards the mode; (iii) the spread of values reduces and becomes con-
centrated around the mean, which means that the PDF decreases more sharply away from the mode and its peak
Telegram: @Persian_IEEEComSoc

value increases to maintain a unit area. If 𝜎 lnX is fixed then skewness remains unchanged, and as 𝜇 lnX increases
(left to right of top row of Figure 3.10) the lognormal PDF broadens, its peak value reducing as it expands hori-
zontally away from the origin, making larger values of X relatively more likely to occur. For example, as indicated
in the top row of Figure 3.10, where 𝜎 lnX is fixed at 𝜎 lnX = 1, the probability that the lognormal random vari-
able X will take on a value larger than eight increases from just under 2% to nearly 60% as 𝜇lnX increases from
𝜇 lnX = 0 to 2.3026. Thus, we may conclude that 𝜎 lnX is a shape parameter, whereas 𝜇 lnX is a location parameter of
the lognormal probability distribution.

Worked Example 3.3

Determine the median, mean, mean square value, variance, mode, rms, and maximum PDF value for a lognormal
random variable X in terms of its distribution parameters 𝜇lnX and 𝜎 lnX . This is an optional worked example.
 pX(x)
0.14
0.6 0.06
μlnX = 0; σlnX = 1 μlnX = 1.6094; σlnX = 1 μlnX = 2.3026; σlnX = 1
0.1
0.4 0.04
Mode

Mode
Pr(X > 8) = 0.0188 Pr(X > 8) = 0.3192 Pr(X > 8) = 0.5883
Mode

0.05
0.2 0.02
Median

Median

Median
Mean

Mean

Mean
0 0 0
0 2 4 6 8 0 5 10 15 20 0 10 20 30 40
pX(x)
0.8 4
Mode,
median,
μlnX = 1.6094; μlnX = 1.6094; mean μlnX = 1.6094;
σlnX = 0.5
0.6 σlnX = 0.1 3 σlnX = 0.02

0.1
0.4 2

0.2 1
Mode,
median,
mean
0 0 0
0 5 10 15 20 3 4 5 6 7 4.5 5 5.5
x→ x→ x→

Figure 3.10 Probability distribution function PDF pX (x) of lognormal distribution for various values of distribution parameters 𝜇lnX and 𝜎 lnX .
Telegram: @Persian_IEEEComSoc
 152 3 Time Domain Analysis of Signals and Systems

The median 𝓂 ̃ X is the value of X at which the CDF in Eq. (3.55) equals 0.5. Thus, replacing x in the right-hand
side of Eq. (3.55) by 𝓂 ̃ X and equating to 0.5 allows us to solve for 𝓂
̃X
( ) ( )
ln 𝓂̃ X − 𝜇ln X ln 𝓂
̃ X − 𝜇ln X 1
1−Q = 0.5; ⇒ Q = ;
𝜎ln X 𝜎ln X 2
ln 𝓂̃ X − 𝜇ln X
⇒ =0
𝜎ln X
since Q(0) = 1/2. Thus, ln 𝓂
̃ X − 𝜇ln X = 0, and hence
𝓂
̃ X = exp(𝜇ln X )

The mean E[X] ≡ 𝜇 X and mean square value E[X 2 ] may be determined by noting that Y = ln(X) is a Gaussian
random variable, and X = eY , so that the kth moment of X is given by
E[X k ] = E[(eY )k ] = E[ekY ]
∞ ∞ (y−𝜇ln X )2
1 −
ky ky 2𝜎 2
= e pY (y)dy = √ e e ln X dy
∫−∞ 𝜎ln X 2𝜋 ∫−∞
[ ]
1
∞ 2
y2 − 2(𝜎ln X
k + 𝜇ln X )y + 𝜇ln
2
X
= √ exp − dy
2𝜋 ∫−∞
2
𝜎ln X 2𝜎ln X

The numerator of the integrand is in the form y2 − 2by + c, which we may express in the form (y − b)2 + c − b2 .
Doing this and simplifying leads to
[ ]
k 1
∞ (y − (𝜇ln X + 𝜎ln
2
X
k))2 + 𝜇ln
2
X
− (𝜇ln X + 𝜎ln 2
X
k)2
E[X ] = √ exp − dy
𝜎ln X 2𝜋 ∫−∞
2
2𝜎ln X
[ ]
1
∞ (y − (𝜇ln X + 𝜎ln
2
X
k))2 2k𝜇ln X + k2 𝜎ln 2
X
= √ exp − + dy
𝜎ln X 2𝜋 ∫−∞
2 2
2𝜎ln X
( ){ [ ] }
2k𝜇ln X + k2 𝜎ln
2
X 1
∞ (y − (𝜇ln X + k𝜎ln 2
X
))2
= exp √ exp − dy
2𝜋 ∫−∞
2 2
𝜎 2𝜎ln X
ln X

But the term inside the curly brackets is the total area under a Gaussian PDF of variance 𝜎 lnX and mean 𝜇ln X +
2
k𝜎ln X
, so this term equals 1. Hence
( )
k
2k𝜇ln X + k2 𝜎ln
2
X
E[X ] = exp
2
Telegram: @Persian_IEEEComSoc

Substituting k = 1 for mean and k = 2 for mean square value yields the desired results
( )
𝜎ln
2
X
Mean ≡ E[X] ≡ 𝜇X = exp 𝜇ln X +
2
Mean square value ≡ E[X 2 ] = exp[2(𝜇ln X + 𝜎ln
2
X )]

The variance follows straightforwardly from the above two results

Variance ≡ 𝜎X2 = E[X 2 ] − (E[X])2

X ) − exp(2𝜇ln X + 𝜎ln X )
2 2
= exp(2𝜇ln X + 2𝜎ln
= exp(2𝜇ln X + 𝜎ln
2 2
X )[exp(𝜎ln X ) − 1]
 3.4 Standard Distribution Functions 153

Mode 𝓂 ̂ R is the value of x at which pX (x) is maximum, which corresponds to the point at which dpX (x)/dx = 0.
Taking the derivative of pX (x)
( [ ( )2 ])
d d 1 1 ln x − 𝜇ln X
p (x) = √ exp −
dx X dx x𝜎 2 𝜎ln X
ln X 2𝜋
[ ( )2 ]
1 1 ln x − 𝜇ln X
=− √ exp −
x2 𝜎ln X 2𝜋 2 𝜎ln X
[ ( )2 ] ( )( )
1 1 ln x − 𝜇ln X ln x − 𝜇ln X 1
+ √ exp − −
x𝜎 2𝜋 2 𝜎ln X 𝜎ln X x𝜎ln X
{ ln X [ ( )2 ]} ( )
1 1 ln x − 𝜇ln X 𝜇ln X − ln x
= √ exp − −1
x2 𝜎ln X 2𝜋 2 𝜎ln X 𝜎ln
2
X

At x = 𝓂 ̂ X , this derivative is zero. Notice that there are two factors on the right-hand side of the above equation.
The first factor (in curly brackets) can only be zero at x = ∞ which is not an acceptable solution for the mode.
Therefore, it is the second factor that will be zero at x = 𝓂 ̂ X . Thus
𝜇ln X − ln 𝓂
̂X
−1=0
𝜎ln
2
X
̂ X = 𝜇ln X − 𝜎ln
⇒ ln 𝓂 2
X

Hence, (upon taking the exponential of both sides)

Mode ≡ 𝓂
̂ X = exp(𝜇ln X − 𝜎ln
2
X)

Rms value (denoted Arms ) is simply the square root of the mean square value, the formula of which was derived
above. Thus

Arms = exp(𝜇ln X + 𝜎ln

2
X)

By substituting x = 𝓂
̂ X into the PDF expression, we obtain the maximum value of the lognormal PDF as

pX (x)max = pX (𝓂̂ X ) = pX (exp(𝜇ln X − 𝜎ln

2
X ))
[ ( )2 ]|
1 1 ln x − 𝜇ln X |
= √ exp − |
𝜎 |
x𝜎ln X 2𝜋 2 |
ln X
|x=exp(𝜇ln X −𝜎ln2 X )
( )2
⎡ 𝜇 − 𝜎 2
− 𝜇 ⎤
1 1
√ exp(𝜎ln X − 𝜇ln X ) exp ⎢− ⎥
2 ln X ln X ln X
=
𝜎ln X 2𝜋 ⎢ 2 𝜎ln X ⎥
Telegram: @Persian_IEEEComSoc

⎣ ⎦
exp(−𝜇ln X + 𝜎ln
2
X
∕2)
= √
𝜎ln X 2𝜋

3.4.4 Rician Distribution

It is often the case, especially in microcellular wireless transmission environments, that the received signal R′ is
the resultant of a dominant direct signal of magnitude A combined with n significantly weaker multipath sig-
nals. Following our earlier discussion (see Section 3.4.2), and using the direct signal as reference, the situation is
now that of combining a constant in-phase component A with random in-phase and quadrature components x
 154 3 Time Domain Analysis of Signals and Systems

Imaginary axis
dxʹ
rʹdϕ
dy dA
dϕ drʹ
ϕ rʹ y = rʹ sin ϕ
A x Real axis
xʹ = rʹ cos ϕ

Figure 3.11 Combination of line-of-sight signal of amplitude A with random multipath-induced in-phase, and quadrature
components x and y to produce a resultant signal having magnitude r ′ and phase 𝜙.

and y, which are respective samples of zero-mean Gaussian random variables X and Y having equal variance 𝜎 2 .
Figure 3.11 shows that the envelope or resultant magnitude of this received signal is given by
√
r ′ = (A + x)2 + y2 (3.63)
which will obviously have some random variation. We are interested in determining its PDF pR′ (r ′ ).
Notice that we are now dealing with two independent random variables, namely Y , which is Gaussian with
mean 𝜇Y = 0 and variance 𝜎Y2 = 𝜎 2 , and X ′ = A + X, which is Gaussian with mean 𝜇X ′ = A and variance 𝜎X2 ′ = 𝜎 2 .
Thus
( )
1 y2
pY (y) = √ exp − 2 ;
2𝜋𝜎 2 2𝜎
( )
1 (x′ − A)2
pX ′ (x′ ) = √ exp − (3.64)
2𝜋𝜎 2 2𝜎 2
A sample of the received signal is characterised by its magnitude r ′ given in Eq. (3.63) and phase 𝜙 shown in
Figure 3.11, which are respective samples of random variables R′ and Φ. However, these random variables are not
Telegram: @Persian_IEEEComSoc

independent since the amount 𝜙 by which the phase of R′ deviates from the reference 0∘ direction of the dominant
direct signal (due to the addition of the random quadrature component y) depends on the value of A and hence
of R′ . This observation has an important implication in that we therefore cannot express the joint probability
distribution pR′ ,Φ (r ′ , 𝜙) of R′ and Φ as the product of their respective individual PDFs pR′ (r ′ ) and pΦ (𝜙).
The following relationships are obvious from Figure 3.11 and are used in the derivation below.
x′ = A + x; dx′ = dx
dA = dx′ dy = dxdy = r ′ dr ′ d𝜙
x′ = r ′ cos 𝜙; y = r ′ sin 𝜙
2 2
x′ + y2 = (r ′ cos 𝜙)2 + (r ′ sin 𝜙)2 = r ′
 3.4 Standard Distribution Functions 155

The probability that a sample of R′ will lie in the shaded elemental area dA is given in terms of the joint proba-
bility distribution of R′ and Φ as
pR′ ,Φ (r ′ , 𝜙)dr ′ d𝜙 = Pr(r ′ ≤ R′ ≤ r ′ + dr ′ , 𝜙 ≤ Φ ≤ 𝜙 + d𝜙)
= Pr(x′ ≤ X ′ ≤ x′ + dx′ , y ≤ Y ≤ y + dy)
Pr(x′ ≤ X ′ ≤ x′ + dx′ ) ⋅ Pr(y ≤ Y ≤ y + dy)
= pX ′ (x′ )dx′ ⋅ pY (y)dy = pX ′ (x′ )pY (y)dxdy
= pX ′ (x′ )pY (y)r ′ dr ′ d𝜙
Using the expressions for pX ′ (x′ ) and pY (y) given in Eq. (3.64) yields
( )
1 (x′ − A)2 + y2
pR′ ,Φ (r , 𝜙)dr d𝜙 =
′ ′
exp − r ′ dr ′ d𝜙
2𝜋𝜎 2 2𝜎 2
( ′2 )
1 r + A2 − 2Ar ′ cos 𝜙 ′ ′
= exp − r dr d𝜙
2𝜋𝜎 2 2𝜎 2
Thus
( ′2 )
r′ r + A2 − 2Ar ′ cos 𝜙
pR′ ,Φ (r , 𝜙) =
′
exp − (3.65)
2𝜋𝜎 2 2𝜎 2
This is the joint PDF of R′ and Φ, which, as earlier noted, cannot be equated to pR′ (r ′ )pΦ (𝜙) because the two
random variables are not independent. To obtain our desired PDF of R′ , observe that pR′ (r ′ )dr ′ is by definition
the probability that R′ will have a value in the infinitesimal interval (r ′ , r ′ + dr ′ ). This is the probability that a
sample of the received signal will have a magnitude lying anywhere within the annular area demarcated by the
two dotted circles in Figure 3.11. We therefore obtain pR′ (r ′ )dr ′ by summing the probability pR′ ,Φ (r ′ , 𝜙)dr ′ d𝜙 that
R′ lies within all the elemental areas dA as 𝜙 goes full circle from 0 to 2𝜋 to cover the entire annular area. That is
2𝜋
pR′ (r ′ )dr ′ = pR′ ,Φ (r ′ , 𝜙)dr ′ d𝜙
∫0
( ′2 ) 2𝜋 ( ′ )
r ′ dr ′ r + A2 Ar
= exp − exp cos 𝜙 d𝜙
2𝜋𝜎 2 2𝜎 2 ∫0 𝜎2
The integral in the above equation cannot be evaluated in closed form but may be expressed in terms of the
well-known modified Bessel function of the first kind of zero order I 0 (x) which is given by

1
2𝜋 ∑∞
(x∕2)2m
I0 (x) = exp(x cos 𝜙)d𝜙 = (3.66)
2𝜋 ∫0 m=0
(m!)2
Thus, we finally obtain
( ′2 ) ( ′)
Telegram: @Persian_IEEEComSoc

r′ r + A2 Ar
pR′ (r ′ ) = 2 exp − I0 (3.67)
𝜎 2𝜎 2 𝜎2
This is the Rician distribution, named in honour of Stephen Oswald Rice (1907–1986), an American scientist
and author of the classic papers on mathematical analysis of noise [2, 3]. As expected, this distribution reduces
to the Rayleigh distribution of Eq. (3.42) when A = 0, since I 0 (0) = 1. Furthermore, in the limit A2 ∕2𝜎 2 ≫ 1, the
random components are small compared to A, and the quadrature component y may be ignored so that r ′ ≈ A + x,
which is a Gaussian random variable of mean A. Thus, the Rician distribution tends towards Rayleigh in the limit
A2 ∕2𝜎 2 ≪ 1 when there is no distinct line-of-sight signal, and towards Gaussian in the opposite limit A2 ∕2𝜎 2 ≫ 1
when the line-of-sight signal is overwhelmingly dominant.
The Rician distribution depends on two parameters, namely the magnitude A of the reference constant compo-
nent of power A2 (see Eq. (3.108)) and the power 2𝜎 2 in the random multipath contributions. The ratio between
 156 3 Time Domain Analysis of Signals and Systems

power in the line-of-sight component and power in the multipath components is known as the Rician K-factor,
often expressed in dB (but converted to its non-dB ratio when used in formulas)
( 2)
A2 A
K = 2 ; K = 10log10 dB (3.68)
2𝜎 2𝜎 2
The K-factor is often employed as a parameter to describe and characterise the Rician distribution. It has been
reported [4] that measurements in microcellular environments show K in the range 6–30 dB.
We may further simplify the Rician PDF by expressing it as pV (v), where v is a dimensionless quantity specifying
the factor by which the sample r ′ exceeds 𝜎. That is, introducing
r′ A
v= ; a= ; ⇒ r ′ = 𝜎v; A = 𝜎a (3.69)
𝜎 𝜎
it follows that

pV (v)dv = Pr(v ≤ V ≤ v + dv)

= Pr(𝜎v ≤ R′ ≤ 𝜎(v + dv)) = Pr(𝜎v ≤ R′ ≤ 𝜎v + 𝜎dv)
= pR′ (𝜎v)𝜎dv = 𝜎pR′ (r ′ )dv

And hence

pV (v) = 𝜎pR′ (r ′ ) (3.70)

Using these substitutions in Eq. (3.67) gives the normalised Rician PDF
( 2 )
v + a2
pV (v) = v exp − I0 (av) (3.71)
2
In terms of the K-factor this becomes
[ ( )] √
v2
pV (v) = v exp − K + I0 (v 2K) (3.72)
2
The CDF and CCDF of a Rician random variable are given by
v
CDF ≡ Pr(R′ ≤ 𝜎v) = Pr(V ≤ v) = p (z)dz
∫0 V
v [ ( )] √ √
z2
= z exp − K + I0 (z 2K)dz = 1 − Q1 ( 2K, v) (3.73)
∫0 2

CCDF ≡ Pr(R′ > 𝜎v) = 1 − CDF

√
Telegram: @Persian_IEEEComSoc

= Q1 ( 2K, v) (3.74)

where Q1 (c, d) is the Marcum Q-function given by

∞ ( 2 )
z + c2
Q1 (c, d) = z exp − I0 (cz)dz
∫d 2
( 2 ) ∞
c + d2 ∑
= exp − (c∕d)n In (cd) (3.75)
2 n=0

and I n is the modified Bessel function of order n given by

∑∞
(x∕2)2m+n
In (x) = (3.76)
m=0
m!(m + n)!
 3.4 Standard Distribution Functions 157

0.6 K = –∞ dB

K = 0 dB
K = 6 dB K = 10 dB
0.4
pRʹ(rʹ)

0.2

0
0 1 2 3 4 5 6 7 8
1

0.1
Pr(V > ʋ)

K K K
K

= = =
=

0 6d 10
–∞

dB B dB
dB

0.01

0.001
0 1 2 3 4 5 6 7 8
Normalised signal level, ʋ = rʹ/σ

Figure 3.12 Rician PDF pR′ (r ′ ) and CCDF Pr(V > v) at various values of K-factor.

The mean square value of a Rician distributed random variable is its total power
2
Mean square value ≡ E[R′ ] = A2 + 2𝜎 2
= 2𝜎 2 (K + 1) (3.77)
And the mean is given by
√ ( )[ ( ) ( )]
𝜋 K K K
E[R ] ≡ 𝜇R′ = 𝜎
′
exp − (K + 1)I0 + KI 1 (3.78)
2 2 2 2
where I 0 and I 1 are as given in Eq. (3.76) for n = 0, 1, respectively.
Figure 3.12 shows the PDF and CCDF (exceedance distribution function) of a Rician random variable at various
K-factors. The value r ′ of the random variable R′ is normalised by 𝜎 as stated in Eq. (3.69). Notice how the PDF
gradually morphs from Rayleigh at K = −∞ dB (and hence A = 0 as per Eq. (3.68)) towards bell-shaped (i.e.
Gaussian) as K increases.
Telegram: @Persian_IEEEComSoc

3.4.5 Exponential and Poisson Distributions

Consider again a Rayleigh random variable R representing the magnitude of the resultant received signal in a
multipath environment in the absence of a dominant direct ray. The instantaneous power of R is defined as the
square of each sample r of the random variable and will itself be a random variable. Since half of this power is
reactive, let us define a new random variable W = R2 /2 representing the active instantaneous power of R. We wish
to determine the distribution function of W, and will make use of the following relationships between w and r,
respectively the samples of W and R
w = r 2 ∕2; r 2 = 2w; r = (2w)1∕2
 158 3 Time Domain Analysis of Signals and Systems

The PDF pW (w) of W is, by definition, given by

pW (w)dw ≡ Pr(w ≤ W ≤ w + dw)
= Pr((2w)1∕2 ≤ R ≤ [2(w + dw)]1∕2 )
√
= Pr( 2w ≤ R ≤ (2w)1∕2 (1 + dw∕w)1∕2 )
√ √ √
= Pr( 2w ≤ R ≤ 2w + dw∕ 2w)
√ √
2w+dw∕ 2w
= pR (r)dr
∫√2w
√ √
= pR ( 2w)dw∕ 2w

where we have simplified the third line above by noting that (1 + a)n ≈ 1 + na for a ≪ 1, and thus (1 + dw/w)1/2 =
1 + dw/2w since dw is infinitesimally small. Comparing the left-hand side with the last step of the right-hand side,
we see that
√
pR ( 2w)
pW (w) = √
2w
√ √
which means that pW (w) is obtained by evaluating pR (r) (in Eq. (3.43)) at r = 2w and dividing by 2w. Thus
√
2w ( )
2w 1
pW (w) = exp − ⋅√
𝜎2 2𝜎 2 2w
( )
1 w
= 2 exp − 2 u(w) (3.79)
𝜎 𝜎
This is the exponential distribution, applicable to a positive random variable that takes on continuous values
in the range (0, ∞). The unit step function u(w) is introduced above for the sole purpose of emphasising that
pW (w) = 0 for w < 0 and may be omitted, its presence being implied by the nature of the distribution and more
specifically the fact that the random variable W is always positive. The exponential distribution is often written in
the form

pW (w) = 𝜆e−𝜆w (3.80)

and is characterised by a single parameter 𝜆(≡ 1/𝜎 2 in Eq. (3.79)). In terms of 𝜆, its CDF and CCDF are obtained
as
w
CDF ≡ Pr(W ≤ w) = pW (z)dz
∫0
w
= 𝜆e−𝜆z dz = −e−𝜆z |w0
Telegram: @Persian_IEEEComSoc

∫0
= 1 − e−𝜆w (3.81)

CCDF ≡ Pr(W > w) = 1 − CDF

= e−𝜆w (3.82)

And it has the following characteristic values

Mean ≡ E[W] = 1∕𝜆 (3.83)

Mean square value ≡ E[W ] = 2∕𝜆 2 2
(3.84)
√ √
Rms value ≡ E[W 2 ] = 2∕𝜆 (3.85)
 3.4 Standard Distribution Functions 159

Variance ≡ 𝜎W
2
= E[W 2 ] − (E[W])2 = 1∕𝜆2 (3.86)
ln 2
Median = (3.87)
𝜆
We see from Eqs. (3.79), (3.80), and (3.83) that the received instantaneous active power in a multipath environ-
ment (with no dominant direct ray) is exponentially distributed with average value 𝜎 2 .
Although we have derived it in connection with the distribution of received signal power in a multipath envi-
ronment, the exponential distribution arises more naturally in the distribution of the time interval 𝜏 between
successive events or arrivals (also known as inter-arrival time) in a Poisson process, in which case the distribution
parameter 𝜆 represents the average arrival rate (which must be constant over the observation interval of interest).
A Poisson process is one that satisfies the following fundamental properties:

1. The probability of one event or arrival in an infinitesimally small interval Δt is 𝜆Δt, where 𝜆 is a specified
constant and 𝜆Δt ≪ 1.
2. The probability of zero arrival during the interval Δt is 1 − 𝜆Δt.
3. Arrivals are memoryless. This means that an arrival or event in any given time interval is independent of events
in previous or future intervals.
4. There cannot be more than one arrival in the interval Δt. This means that two or more arrivals cannot occur
at the same time instant; so that in an infinitesimally small interval Δt there are only two possibilities, namely
either there is no arrival or there is exactly one arrival.

The Poisson process is thus a special case of a Markov process, where the probability of an event at time t + Δt
depends only on the probability at time t. A discrete random variable X is said to have a Poisson distribution if it
takes on integer values 0, 1, 2, 3, … in accordance with a Poisson process. The probability mass function (PMF)
of X, denoted Pr(X = k), gives the probability that X is exactly equal to some value k. We wish to determine the
Poisson PMF, and specifically the probability that in a Poisson arrival process there will be exactly k arrivals in
some finite interval of duration D.
Dividing the interval D into n infinitesimally small slots Δt, so that D = nΔt and Δt = D/n, then Δt → 0 in the
limit n → ∞. In this limit, there will be either one arrival in each slot Δt with probability p = 𝜆Δt = 𝜆D/n or zero
arrival with probability 1 –p. Figure 3.13 shows the subsets of observations having exactly k arrivals in D. There
are n Ck (pronounced ‘n-choose-k’ or ‘n-combination-k’) such subsets, corresponding to the number of ways to
select k slots out of n available. Noting that in selecting the k slots, the first pick may be any of the n slots, followed
by the second pick, which may be any of the remaining n − 1 slots, and so on up to the kth pick which may be
any of the remaining n − (k − 1) slots, we see that there are n(n−1)(n−2)…(n−k + 1) possibilities. However, every
k! = k(k−1)(k−2)…1 of these are identical selections, being merely repeats of the same set of k slots in a different
order. For example, if k = 3 then every 3! = 3 × 2 × 1 = 6 sets, such as slots {2, 3, 4}, {2, 4, 3}, {3, 2, 4}, {3, 4, 2}, {4,
2, 3}, {4, 3, 2}, will occur as repeats of the same selection of slots 2, 3, and 4. The number of ways to select k slots,
Telegram: @Persian_IEEEComSoc

regardless of order, from n available is therefore

( )
n n(n − 1)(n − 2) · · · (n − k + 1)
n
Ck ≡ =
k k(k − 1)(k − 2) · · · 1
n!
= (3.88)
k!(n − k)!
Each of the observation subsets shown in Figure 3.13 occurs as a result of there being one arrival (with prob-
ability p) in some first slot AND one arrival (with probability p) in some second slot … AND one arrival (with
probability p) in some kth slot AND zero arrival (with probability 1 − p) in some (k + 1)th slot … AND zero arrival
(with probability 1 − p) in the nth slot. Since events in the slots are independent, the probability of each subset is
therefore the product of these probabilities, i.e. pk (1 − p)n−k . And because these subsets are mutually exclusive,
 160 3 Time Domain Analysis of Signals and Systems

Finite interval D
k slots n – k slots
Δt

nC Probability of
k
subsets each subset
= pk(1 – p)n – k

k slots
= One arrival with probability p
Legend:
= No arrival with probability 1 – p

Figure 3.13 Subsets of observations with exactly k arrivals in a finite interval D.

their probabilities add to give the following probability of there being exactly k arrivals in interval D
( ) ( ) ( )k ( )
n k n 𝜆D 𝜆D n−k
Pr(X = k) = lim p (1 − p)n−k = lim 1−
n→∞ k n→∞ k n n
[ ( ) ( ) ]
n(n − 1)(n − 2) · · · (n − (k − 1)) (𝜆D) k
𝜆D −k 𝜆D n
= lim 1 − 1 −
n→∞ k! nk n n
[( )( ) ( )( )−k ( )n ] (𝜆D)k
1 2 k−1 𝜆D 𝜆D
= lim 1 − 1− ··· 1− 1− 1−
n→∞ n n n n n k!
[( ) ]
𝜆D n (𝜆D)k
= lim 1 −
n→∞ n k!
To evaluate the first term on the right-hand side of the above equation, consider the binomial expansion of
(1 + x/n)n in the limit n → ∞
lim (1 + x∕n)n = lim [1 + nx∕n + n(n − 1)(x∕n)2 ∕2! + n(n − 1)(n − 2)(x∕n)3 ∕3! + · · ·]
n→∞ n→∞
= 1 + x + x2 ∕2! + x3 ∕3! + x4 ∕4! + · · ·
= ex
In view of this definition for the exponential function, it follows that the term in question is exp(−𝜆D) and hence
Telegram: @Persian_IEEEComSoc

(𝜆D)k
Pr(X = k) = exp(−𝜆D) (3.89)
k!
This is the PMF of the Poisson distribution. The mean E[X] (i.e. expected number of arrivals in interval D), mean
square value E[X 2 ], and variance 𝜎X2 of the distribution are given by (see Question 3.4)
E[X] = 𝜆D
E[X 2 ] = (𝜆D)2 + 𝜆D
𝜎X2 = E[X 2 ] − (E[X])2 = 𝜆D (3.90)
And since the average number of arrivals in interval D is 𝜆D, it follows that the distribution parameter 𝜆 is the
average rate of arrivals, which may be estimated as N/D by observing the number of arrivals N in a sufficiently large
 3.4 Standard Distribution Functions 161

interval D, where the value of D is chosen such that 𝜆D > > 1. In practice, 𝜆 may exhibit some recurring variation
and so D must be chosen to be just large enough to satisfy this condition while fully spanning only the interval of
interest in the cycle. For example, the number of customers arriving at a supermarket checkout each minute will
vary with time of day and even day of week, and the number of telephone calls arriving at a telephone exchange
will also vary with time of day, so 𝜆 would need to be estimated during peak intervals to allow the service system
to be designed to cope with busy hour traffic. Note that the quantity 𝜆D is a dimensionless number and therefore
it may be added to its square as done in Eq. (3.90) for E[X2 ] without any dimensional inconsistency.
The Poisson process is extensively employed in telecommunications network design to model random behaviour
or events from a population of users, such as the arrival of telephone calls at a local exchange or data packets
at a queue. It is shown below that a Poisson arrival process gives rise to an exponential distribution for the time
between successive arrivals. Furthermore, a Poisson departure process, such as telephone call completion or packet
departure from a queue, is produced by an exponential distribution of service duration (e.g. telephone call duration,
known as call holding time, and how long it takes for a packet to be served at a queue (excluding the time spent
waiting in the queue for service to commence)). In this case the exponential distribution parameter represents the
service rate, i.e. the average number of customers (e.g. telephone calls or packets) served per unit time.
Let us examine the times 𝜏 between successive arrivals in a Poisson process that generates a random variable X
having the Poisson distribution given in Eq. (3.89). In Figure 3.14, we start at time t = 0 to observe these Poisson
arrivals, the first of which occurs at time t = T. Consider then some arbitrary time t = 𝜏 into our observation. If
𝜏 < T, then there is no arrival within the interval (0, 𝜏) and thus

Pr(𝜏 < T) = Pr(X = 0)

= exp(−𝜆𝜏)

where we have used Eq. (3.89) with observation interval D ≡ 𝜏 and number of arrivals k = 0. Regarding the value
of the resulting 0!, recall that n! is, by definition, the number of ways to order a set with n elements, and there is
only one way to order a set with no elements, hence 0! = 1.
Since Pr(T ≤ 𝜏) = 1 − Pr(𝜏 < T), it follows that

Pr(T ≤ 𝜏) ≡ PT (𝜏) = 1 − exp(−𝜆𝜏) (3.91)

This is the CDF of the random variable T, which upon differentiation according to Eq. (3.15) yields the corre-
sponding PDF pT (𝜏) as
d
pT (𝜏) = P (𝜏) = 𝜆 exp(−𝜆𝜏) (3.92)
d𝜏 T
which is an exponential distribution. Therefore, as earlier stated, a Poisson arrival process gives rise to an exponen-
tial distribution for the time 𝜏 between successive arrivals. Similarly, it is easy to show that a Poisson departure
Telegram: @Persian_IEEEComSoc

process arises out of an exponential service time distribution. Equation (3.83) gives the mean of T as 1/𝜆, which
is the average inter-arrival time. An exponential PDF for the distribution of inter-arrival times of a Poisson arrival
process is shown in Figure 3.15 along with tabulated cumulative and exceedance probabilities. We see that on
average 63% of next arrivals will be within the mean inter-arrival duration and only 5% of next arrivals will exceed
three times the mean inter-arrival duration.

First arrival
time, t
t=0 t=τ t=T

Figure 3.14 Observation of Poisson arrivals.

 162 3 Time Domain Analysis of Signals and Systems

τ CDF ≡ Pr(T ≤ τ) CCDF ≡ Pr(T > τ)

3λ/4 1/λ 0.6321 0.3679
PDF, pT(τ) = λexp(–λτ)

2/λ 0.8647 0.1353

3/λ 0.9502 0.0498
λ/2 4/λ 0.9817 0.0183
5/λ 0.9933 0.0067
6/λ 0.9975 0.0025
1/λ ≡ Mean inter-arrival time
λ/4

0
0 1/λ 2/λ 3/λ 4/λ 5/λ
Time between successive arrivals, τ →

Figure 3.15 PDF of an exponential distribution along with tabulated cumulative and exceedance probabilities at selected
values of inter-arrival time 𝜏.

3.5 Signal Characterisation

Since signals are necessarily time-varying, it is often useful to extract from the signal a single feature or parame-
ter that quantifies or summarises the level, effectiveness, strength, or variability of the signal. Some of the basic
characterisation parameters include the mean level, root-mean-square (rms) value, power, energy, and autocorre-
lation of the signal. The similarity or otherwise between two signals may also be specified using measures of their
correlation and covariance.

3.5.1 Mean
The mean or average value of a signal is the level around which the instantaneous values of the signal vary with
time. The signal may be an electric current or voltage, in which case the mean value is referred to as the DC (for
direct current) value. We know that the mean A0 of a set of sample values such as {4, 4, 6, 8, 10, 10, 10, 12} is the
sum of the samples divided by the number of samples. Rearranging this expression for A0 leads to
Telegram: @Persian_IEEEComSoc

4 + 4 + 6 + 8 + 10 + 10 + 10 + 12
A0 =
8
2 1 1 3 1
= 4 × + 6 × + 8 × + 10 × + 12 ×
8 8 8 8 8
which indicates that the mean is obtained by summing the product of each sample value and the fraction of the
set composition corresponding to that value. This fraction represents the probability of occurrence of each value
within the set. In general, the mean A0 of a signal that takes on N possible values {g0 , g1 , g2 , …, gN−1 } with respective
probabilities {p0 , p1 , p2 , …, pN−1 } is
∑
N−1
A0 = g0 p0 + g1 p1 + g2 p2 + · · · + gN−1 pN−1 ≡ gn pn (3.93)
n=0
 3.5 Signal Characterisation 163

This equation is applicable to a wide range of signal types. For example, if the signal is a sequence of samples
{g0 , g1 , g2 , …, gN−1 } taken in respective nonuniform intervals {𝜏 0 , 𝜏 1 , 𝜏 2 , …, 𝜏 N−1 }, then the probability of the nth
sample gn is

∑
N−1
pn = 𝜏n ∕𝕋 , where 𝕋 = 𝜏n
n=0

is the total duration of the signal.

If it is a discrete signal g(nT s ) consisting of a sequence of N samples g(0), g(T s ), g(2T s ), …, g([N−1]T s ) taken at
a uniform interval T s , then the probability pn of each sample is simply the fraction of the total signal duration NT s
that is occupied by each of the samples, so pn = T s /NT s = 1/N, and gn ≡ g(nT s ) ≡ g(n).
For an infinitesimally small sampling interval T s , denoted dt, the discrete time instants nT s become continuous
time t, the samples merge into a continuum of values, and the signal gn ≡ g(nT s ) becomes a continuous function
of time g(t) having duration NT s ≡ 𝕋 and probabilities pn = Ts ∕NT s = dt∕𝕋 . The summation in Eq. (3.93) thereby
reduces to integration over the entire signal duration.
If g(t) is periodic, with period T, then there is no need to carry out the averaging process beyond an interval of
one cycle since the signal’s repetitive structure ensures the same result within each interval (b, b + T), where b is
any real number.
For a periodic staircase (or quantised) signal that takes on flat levels or steps g0 , g1 , g2 , …, gN−1 having respective
widths 𝜏 0 , 𝜏 1 , 𝜏 2 , …, 𝜏 N−1 that sum up to one period T of the signal, we do the averaging over one cycle of the signal.
The probability of the nth step gn is pn = 𝜏 n /T = dn . We may describe this signal as an N-step staircase waveform,
and dn as the duty cycle of the nth step.
Thus
⎧ ( / N−1 ) N−1
∑ ∑
⎪ lim 1 𝜏n gn 𝜏n , (non-uniformly sampled signal)
⎪N→∞
⎪ n=0 n=0
⎪ ∑
N−1
⎪ lim 1 g(n), (discrete or uniformly sampled signal)
⎪N→∞ N
⎪ n=0
⎪ 𝕋 ∕2
A0 = ⎨ lim 1 g(t)dt, (continuous-time (CT) signal) (3.94)
⎪𝕋 →∞ 𝕋 ∫−𝕋 ∕2
⎪ b+T T∕2
⎪1 1
⎪T ∫ g(t)dt = g(t)dt, (periodic CT signal)
T ∫−T∕2
⎪ b
⎪ N−1 ∑ ∑
N−1
⎪1 gn 𝜏n = gn dn , (periodic staircase signal)
⎪ T n=0
⎩ n=0
Telegram: @Persian_IEEEComSoc

The last expression of Eq. (3.94) applies to an N-step staircase periodic waveform, an example of which is shown
in Figure 3.16 for N = 4. Note that the mean value is computed within one cycle of the signal and that dn = 𝜏 n /T
is the fraction of time within each cycle during which the signal has value or level gn .

3.5.2 Power
The normalised average power P of a signal, often referred to simply as the power of the signal, is the mean square
value of the signal. In what follows we assume a real-valued signal; however, the expressions developed here for
power (and in the next subsection for energy) may be applied to complex-valued signals by replacing the signal
value g(t) wherever it occurs by its absolute value |g(t)|. We obtain P simply by averaging the square of the signal
 164 3 Time Domain Analysis of Signals and Systems

g(t)
g2
g0 τ2
τ0 τ3
g3 τ1 t
g1
T

Figure 3.16 A 4-step staircase periodic waveform of period T.

values or samples. It follows from the previous discussion, by replacing the signal value in Eq. (3.94) with its
square, that
⎧N−1
∑
⎪ gn2 pn , (finite discrete set)
⎪ n=0
⎪ ( / N−1 ) N−1
⎪ ∑ ∑
lim 1
⎪N→∞ 𝜏n gn 2 𝜏n , (nonuniformly sampled signal)
⎪ n=0 n=0
⎪ ∑
N−1
⎪ lim 1 g2 [n], (discrete or uniformly sampled signal)
⎪N→∞ N
P=⎨ n=0 (3.95)
⎪ 𝕋 ∕2
1
⎪𝕋lim g2 (t)dt, (continuous-time (CT) signal)
→∞ 𝕋 ∫−𝕋 ∕2
⎪
⎪1 b+T
1
T∕2
⎪ g2 (t)dt = g2 (t)dt, (periodic CT signal)
⎪ T ∫b T ∫−T∕2
⎪ N−1
⎪1 ∑ 2 ∑
N−1

⎪T gn n𝜏 = gn 2 dn , (periodic staircase signal)

⎩ n=0 n=0

where b is any real number, pn , n = 0, 1, 2, …, N−1 is the probability that a signal that takes on values drawn only
from the discrete set {g0 , g1 , g2 , …, gN−1 } will take on the value gn .
The normalised average power defined above is the heat that would be dissipated in a pure resistor of resistance
1 ohm (Ω) if the signal were an electric current flowing through the resistor or a voltage drop across the resistor.
Unless otherwise specified, we will work exclusively with normalised average power P. Whenever desired, P may
be scaled to obtain the average power PR dissipated in a pure resistor of resistance R as follows
{
P∕R, g(t) is an electric voltage signal
PR = (3.96)
PR, g(t) is an electric current signal
Telegram: @Persian_IEEEComSoc

The SI unit for dissipated power is joule per second (J/s) which is called watt (W) in honour of the British inventor
James Watt (1736–1819) for his contribution to the design of the steam engine. It is worth noting that there are
other power definitions in electrical engineering. For example, in an electric circuit, instantaneous power P(t) at
a given point in the circuit is defined as the product of instantaneous voltage v(t) and instantaneous current i(t)
at that point. For a sinusoidal voltage v(t) = V m cos(𝜔t) and current i(t) = I m cos(𝜔t − 𝜑vi ), where 𝜑vi is the phase
difference by which the voltage leads the current signal, it follows that
P(t) = Vm cos(𝜔t)Im cos(𝜔t − 𝜑vi )
1
= Vm Im [cos(𝜑vi ) + cos(2𝜔t − 𝜑vi )]
2
1
= Vm Im [cos(𝜑vi ) + cos(2𝜔t) cos(𝜑vi ) + sin(2𝜔t) sin(𝜑vi )] (3.97)
2
 3.5 Signal Characterisation 165

In the above manipulation, we employed trigonometric identities in line 2 and line 3, namely Eqs. (B.6) and
(B.4), respectively, from Appendix B. Thus, we may write
P(t) = Pa + Pa cos(2𝜔t) + Q sin(2𝜔t)
1
where, Pa = Vm Im cos(𝜑vi )
2
1
Q = Vm Im sin(𝜑vi ) (3.98)
2
Let us take a moment to appreciate this important result. Since the sinusoidal function has zero mean, it follows
from line 1 of Eq. (3.98) that
1
Pa = V I cos(𝜑vi )
2 mm
is the mean of the instantaneous power P(t). Pa is the (average) real power or active power. Measured in watts, it
represents the total power that is dissipated as heat in the resistive element R of the circuit or load and depends on
a factor cos(𝜑vi ) called the power factor, where 𝜑vi (also known as the power factor angle) is the phase difference
between current and voltage. The term Pa + Pa cos(2𝜔t) is the instantaneous active power which varies between 0
and 2Pa , completing each cycle at twice the signal frequency. The term
1
Q= V I sin(𝜑vi )
2 mm
is the reactive power, whereas Qsin(2𝜔t) is the instantaneous reactive power. Reactive power is sometimes called
imaginary power and its unit of measurement is volt-ampere reactive (var) to differentiate it from real power in
watts. Reactive power accounts for the rate of energy storage in the capacitive and inductive elements of the load.
Energy alternates between being stored in the load during one-quarter of the signal cycle and being returned
unchanged to the signal source during the next quarter cycle. That is, there is no absorption (or dissipation) of
reactive power by the load. There will, however, inevitably be some loss in the resistance of the line connecting the
load and the source during this alternating exchange or flow of reactive power between load and source. Finally,
the quantity
√
1
S = Pa2 + Q2 = Vm Im (3.99)
2
is the apparent power, also called the complex power, measured in volt-ampere (VA). The relationship between
apparent power (VA), active power (W), and reactive power (var) is illustrated in Figure 3.17.
If the load is a pure resistance then voltage and current will be in phase and the power factor angle 𝜑vi = 0. This
leads to reactive power Q = 0 so that there is no stored energy and the power is entirely active and dissipative. If
the load has zero resistive element then 𝜑vi = ±90∘ and active power Pa = 0 so that there is no power dissipation
in the load but only energy storage and release back to source in alternate quarter cycles as reactive power. If the
load is a complex impedance having a real resistance part and an imaginary reactance part then there are two
Telegram: @Persian_IEEEComSoc

i(t) = Im cos (ωt – φʋi)

Line
r
we
e n t po
par Im
Ap = ½V
m
Q= 1 VmIm sin φʋi
Source ʋ(t) = Vm cos(ωt) Load S 2
φʋi Reactive power

Pa = 1 VmIm cos φʋi

2
Active power

Figure 3.17 Apparent power (S), active power (P a ), and reactive power (Q).
 166 3 Time Domain Analysis of Signals and Systems

g(t), volts
10
(a)
5
t

–15 ms
–6 1 3 4 6 12

g(t), volts
10
(b)
5
t

–15 ms
–6 1 3 4 6 12

Figure 3.18 Worked Example 3.4.

possible scenarios. If the reactance is inductive then voltage leads current by an acute angle (0∘ < 𝜑vi < 90∘ ) and
reactive power Q is positive. However, if the reactance is capacitive then voltage lags current by an acute angle so
that −90∘ < 𝜑vi < 0∘ , which leads to a negative reactive power. That Q is positive when the load is inductive and
negative when the load is capacitive is merely indicative of the fact that energy flow in a capacitor and an inductor
is always in opposition, with one storing while the other supplies; and the flow is conventionally referenced to the
inductor.
In general, a signal comprises a DC component (which is the average value A0 of the signal) and several har-
monics (which are the sinusoidal components of the signal at frequencies f > 0). It is therefore common practice
to define DC power Pdc as the power in the DC component and AC power Pac as the power in the harmonics. Thus,
with total power denoted P
Pdc = A20
Pac = P − A20 (3.100)
Telegram: @Persian_IEEEComSoc

Worked Example 3.4

The periodic staircase voltage waveform g(t) shown in Figure 3.18a is measured across a 50-ohm (Ω) resistor.
Determine:
(a) The average value of the signal.
(b) The normalised average power of the signal.
(c) The average power dissipated by the voltage signal in the resistor.

We first identify one cycle of g(t) and its period T, as done in Figure 3.18b using the thicker line. Hence T = 6 ms.
Next, we identify the discrete values gn of g(t) and their respective probabilities pn . These are gn = {10, −15, 5, 0}
and pn = {1/6, 2/6, 1/6, 2/6}. We are now ready to calculate the required parameters.
 3.5 Signal Characterisation 167

(a) The average value of g(t) is obtained using Eq. (3.93) as

∑
N−1
A0 = gn pn
n=0
1 2 1 2
= 10 × + (−15) × + 5 × + 0 ×
6 6 6 6
= −2.5 V

(b) The normalised average power of g(t) is obtained using the last line of Eq. (3.95) as

∑
N−1
P= gn2 pn
n=0
1 2 1 2
= 102 × + (−15)2 × + 52 × + 02 ×
6 6 6 6
= 95.833 W

(c) The average power dissipated in the resistor follows from Eq. (3.96) for a voltage signal as

P 95.833
Pav = = = 1.917 W
R 50

3.5.3 Energy
Since power is defined as energy per unit time, it follows that the energy E of a signal is its average power multiplied
by the duration of the signal. Thus multiplying the expressions for P in Eq. (3.95) by NT s in the case of a discrete
signal, and by 𝕋 for a continuous signal yields
⎧ ∑
N−1
⎪ lim gn 2 𝜏n , (non-uniformly sampled signal)
⎪ N→∞
n=0
⎪
⎪ ∑
N−1
E = ⎨ lim T g2 [n] (discrete or uniformly sampled signal) (3.101)
s
⎪N→∞ n=0
⎪ 𝕋 ∕2
⎪
⎪𝕋lim g2 (t)dt, (continuous-time (CT) signal)
→∞ ∫−𝕋 ∕2
⎩
Telegram: @Persian_IEEEComSoc

A more general definition of energy that applies to both real-valued and complex-valued signals is
∞ ∞
E= |g(t)|2 dt = |G(f )|2 df (3.102)
∫−∞ ∫−∞
where G(f ) is the FT of g(t), which is discussed in the next chapter. Equation (3.102) is a statement of Parse-
val’s theorem (also known as Rayleigh’s energy theorem), which provides an alternative method of determining
the energy of a signal using a frequency domain representation of the signal. This is a very common flexibil-
ity in signal analysis and characterisation – that either a time domain or a frequency domain approach may be
followed.
Note that the energy E of a signal g(t) is the total area under the squared waveform g2 (t) of the signal, whereas
power P is the average of g2 (t).
 168 3 Time Domain Analysis of Signals and Systems

3.5.4 Root-mean-square Value

The mean value A0 of a signal is not a very useful parameter for characterising the effectiveness or strength of a
signal in those situations when the signal can take on both positive and negative values. This is because, when
positive and negative samples are aggregated, they cancel out each other’s contribution which therefore gives rise
to a reduced mean value. For example, an alternating electric current signal will be first positive and then negative
in alternate half-cycles as the direction of current flow is reversed in each half-cycle. And although the capacity of
the signal to do work (such as heat dissipation) is unaffected by this cyclical reversal in direction of current flow,
the resulting change in sign of the signal will directly reduce its mean value. In fact, in the case of a sinusoidal
signal, the mean will be zero regardless of signal amplitude.
We may define an average parameter that more correctly represents the effective value of the signal and its
capacity to do work by squaring each sample before averaging. In this way positive and negative samples will
both add to the accumulation process rather than cancelling each other’s contribution. Once the squares have
been averaged, we take the square root of the result to produce a parameter that has the same unit as the signal it
represents. The parameter so obtained is very appropriately called the root-mean-square (rms) value of the signal,
which we will denote Arms . Recalling that signal power was defined as the mean-square value of the signal, it
follows that Arms is the square root of normalised average power P. In fact, the rms parameter is defined only for
power signals and is given by
√
Arms = P (3.103)
where P is as given in Eq. (3.95) for various categories of signals, including discrete signals, sampled signals, general
continuous signals, and periodic continuous and N-step staircase signals. The rms value of a signal is an extremely
useful parameter which may be computed using Eq. (3.103) if the signal is deterministic, or may be measured
directly in the laboratory using an rms ammeter or voltmeter if the signal can be produced or modelled in electrical
form. Once Arms is known, it is a trivial matter to obtain normalised power P as
P = A2rms (3.104)
and hence some of the other power definitions such as in Eq. (3.96).
The following expressions for the rms values of special waveforms are derived in the next worked example.
A rectangular pulse train of amplitude A and duty cycle d (Figure 3.19a) has rms value
√
Arms = A d (3.105)
A triangular pulse train of amplitude A and duty cycle d (Figure 3.19b)
√
Arms = A d∕3 (3.106)
A sinusoidal signal of amplitude A and any frequency and phase (Figure 2.18)
√
Telegram: @Persian_IEEEComSoc

Arms = A∕ 2 (3.107)
A DC signal of constant value A
Arms = A (3.108)
A complex exponential function Aexp[j(2𝜋ft + 𝜙)]
Arms = A (3.109)
A sinusoidal pulse train of amplitude A and duty cycle d containing an integer number n of half-cycles within
the pulse duration 𝜏 (Figure 3.19c)
√
Arms = A d∕2 (3.110)
 g(t)
A
(a)
d = τ/T

t
τ
T
g ( t)
(b) A
d = τ/T

τ t
T

(c) g(t) d = τ/T

A
τ = n/F, n = 1, 2, 3, …

1/F
τ τr ≡ Rise time of pulse;
T
τf ≡ Fall time of pulse;
(d) g(t) τc ≡ Flat time of pulse;
A
dr = τr/T ≡ Rise time duty cycle;
df = τf /T ≡ Fall time duty cycle;
t dc = τc/T ≡ Flat time duty cycle;
τ τr τc τf τ = τr + τf + τc ≡ Pulse duration;
T = period of waveform
d = dr + df + dc ≡ Duty cycle of pulse train

Figure 3.19 Selected periodic waveforms.

Telegram: @Persian_IEEEComSoc
 170 3 Time Domain Analysis of Signals and Systems

A trapezoidal pulse train of amplitude A and duty cycles dr (rising portion), dc (constant portion), and df (falling
portion) (Figure 3.19d)
√
Arms = A dc + dr ∕3 + df ∕3 (3.111)

Worked Example 3.5

Derive expressions for the rms values of the following periodic signals:

(a) The trapezoidal pulse train shown in Figure 3.19d.

(b) The sinusoidal pulse train shown in Figure 3.19c.
(c) A complex exponential function Aexp[j(2𝜋ft + 𝜙)].
Since all the signals are periodic continuous-time signals, we employ the fifth line of Eq. (3.95) with b = 0 and
Eq. (3.104) to obtain the required rms expressions by evaluating
T
1
A2rms = g2 (t)dt (3.112)
T ∫0
(a) The trapezoidal pulse comprises three portions, namely rising portion of duration 𝜏 r , flat or constant portion of
duration 𝜏 c , and falling portion of duration 𝜏 f . Introducing amplitude A into the expressions for these portions
of the pulse given in Eq. (2.25) and integrating yields

⎧ [( ) ]2 [( ) ]2 ⎫
𝜏c ∕2 𝜏f +𝜏c ∕2
A2 ⎪ −𝜏c ∕2 𝜏c t 𝜏c t ⎪
A2rms = ⎨∫ 1+ + dt + 1 ⋅ dt + 1+ − dt⎬
T ⎪ −(𝜏r +𝜏c ∕2) 2𝜏 r 𝜏 r ∫−𝜏c ∕2 ∫𝜏c ∕2 2𝜏 f 𝜏 f ⎪
⎩ ⎭
Making the substitutions t′ = t + 𝜏r + 𝜏c ∕2 in the first integral (which shifts the rising portion to start at t′ = 0
along the t′ axis) and for a similar reason t′′ = t − 𝜏f − 𝜏c ∕2 in the third integral, and evaluating the second
integral straightforwardly to 𝜏c , we obtain
{ }
𝜏r 0
A2 1 ′2 ′ 1 ′′ 2 ′′
2
Arms = 𝜏c + 2 t dt + 2 t dt
T 𝜏r ∫0 𝜏 ∫−𝜏f f

⎧ ( ) t′ =𝜏r ( )|t′′ =0 ⎫
A2 ⎪ 1 t′ 3 || 1 t′′ 3 || ⎪
= ⎨𝜏c + 2 3 || + 2 | ⎬
T ⎪ 𝜏 𝜏 3 |
|t′ =0 |t′′ =−𝜏f ⎪
r f
⎩ ⎭
{ } ( )
A2 𝜏r 𝜏 f 𝜏 c 𝜏 r
𝜏f
= 𝜏c + + = A2 + +
T 3 3 T 3T 3T
Telegram: @Persian_IEEEComSoc

= A2 (dc + dr ∕3 + df ∕3)
√
Thus, Arms = A dc + dr ∕3 + df ∕3 for a trapezoidal pulse train.
The rectangular pulse train (Figure 3.19a) is a special case of the trapezoidal pulse train with dr = df = 0, 𝜏 c = 𝜏,
√
and dc = d, so that Arms = A d, as given in Eq. (3.105).
Also, the triangular pulse train (Figure 3.19b) is a special case of trapezoidal with dc = 0, 𝜏 r = 𝜏 f , and 𝜏 r + 𝜏 f = 𝜏
so that
𝜏r 𝜏f 𝜏 d
+ = =
3T 3T 3T 3
√
and hence, Arms = A d∕3, as given in Eq. (3.106).
 3.5 Signal Characterisation 171

(b) One cycle of the sinusoidal pulse train is defined by

{
A cos(2𝜋ft + 𝜙), 0 ≤ t ≤ 𝜏
g(t) =
0, 𝜏<t<T
One condition on the frequency f is that there must be an integer number of half-cycles in the pulse interval
𝜏. This means that if the angle 𝜃 of the sinusoid is 𝜃 = 𝜙 at t = 0, then 𝜃 = n𝜋 + 𝜙 at t = 𝜏, where n = 1, 2,
3, …, since 𝜋 radian is one half-cycle. Thus, 2𝜋f𝜏 + 𝜙 = n𝜋 + 𝜙, whence f = n/2𝜏. Therefore a mathematical
expression for the sinusoidal pulse train g(t) is as follows
{ ( )
n
A cos 𝜋 t + 𝜙 , mT ≤ t ≤ 𝜏 + mT
g(t) = 𝜏
0, Otherwise
m = 0, ±1, ±2, · · · , n = 1, 2, 3, · · · (3.113)
Substituting this expression into Eq. (3.112) and evaluating within the first cycle (for which m = 0), we obtain
𝜏[ ( )]2
1 n
A2rms = A cos 𝜋 t + 𝜙 dt
T ∫0 𝜏
2 𝜏
A
= [1 + cos(2𝜋nt∕𝜏 + 𝜙)]dt
2T ∫0
( )
A2 t=𝜏 sin(2𝜋nt∕𝜏 + 𝜙) ||t=𝜏
= t|t=0 − |
2T 2𝜋n∕𝜏 |t=0
( )
A2 sin(2𝜋n + 𝜙) − sin 𝜙
= 𝜏−
2T 2𝜋n∕𝜏
A2 A2
= (𝜏) = d
2T 2
where we have used the identity cos2 A = (1 + cos2A)/2 in the second line above, and sin(2𝜋n + 𝜙) = sin(𝜙)
to obtain the last line. Taking the √square root of both sides, we obtain the expression for the rms value of a
sinusoidal pulse train as Arms = A d∕2, which for a sinusoidal signal √ where d = 1 (because no portion of a
cycle is switched off, so that 𝜏 = T and d = 1), reduces to Arms = A∕ 2, as given in Eq. (3.107).
(c) Because the given function is complex, it is its absolute value that is squared in Eq. (3.112) to obtain the rms
expression for the complex exponential function as
T
1
A2rms = |A exp[j(2𝜋ft + 𝜙)]|2 dt
T ∫0
T
1 1 A2 T
= A2 dt = (A2 t|t=T )=
T ∫0 T t=0
T
= A2
Telegram: @Persian_IEEEComSoc

and hence Arms = A for a complex exponential function of amplitude A, irrespective of frequency or phase. The
DC signal of constant value A is a special case of such exponential with f = 0 and 𝜙 = 0, and so its Arms = A,
as stated in Eq. (3.108).

3.5.5 Autocorrelation
It is often of interest to discover and characterise the variabilities and repeating patterns (such as periodic compo-
nents) in a signal g(t), especially in situations where such features might be masked by noise. One way of doing this
is through a quantity which gives some measure of the match or similarity between the signal and a delayed version
of itself. Specifically, we compare g(t) and g(t − 𝜏), using either the average value of their product g(t)g(t − 𝜏) if g(t)
 172 3 Time Domain Analysis of Signals and Systems

is a power signal or the total area under this product (if g(t) is on the other hand an energy signal) as a measure of
their similarity. This measure is known as the autocorrelation function of the signal, which we previously defined
in Section 3.3.3 for a random process as the expected value of the product of two random variables of the process,
namely X 1 and X 2 being, respectively, the two sets of random numbers obtained by observing (analogous to tak-
ing a snapshot of) the random process at separate time instants t1 and t2 , where t2 − t1 = 𝜏 is a shift or scanning or
searching parameter.
It is important to appreciate the physical significance of the autocorrelation function. If we carry out the above
computation at 𝜏 = 0 then we are obviously comparing the signal to its identical self and we will get a maximum
value equal to the power of the signal, if it is a power signal, or energy otherwise. If the autocorrelation function
decreases rapidly towards zero as 𝜏 increases, it is indicative of high variability in the signal and specifically that
two samples of the signal separated by 𝜏 have little in common. Occupying one extreme of variability is white
noise (arising, for example, from thermal agitation of electrons in conductors) for which a sample at time t is a
random value that has absolutely no relationship to the previous or next sample, even for an infinitesimally small
sampling interval. Therefore, the autocorrelation function will be an impulse function of the scanning parameter 𝜏,
meaning that it will be nonzero for 𝜏 = 0 and zero for 𝜏 > 0. At the other extreme is a DC signal with no variability.
Its autocorrelation function will be a constant value, indicating that adjacent samples are identical, whatever the
spacing or interval between samples. Additionally, if the autocorrelation function is found to exhibit local peaks at
𝜏 = nΔ, for n = 1, 2, 3, …, then this is indicative of the presence of a periodic component in g(t) having fundamental
frequency f = 1/Δ.
The autocorrelation function of a signal therefore provides important information about the variability and
hence spectral content or the frequency spectrum of the signal. The autocorrelation function of a DC signal is
a constant, which indicates zero variability and hence a spectrum that is zero everywhere except at frequency
f = 0. The autocorrelation function of a white noise signal is an impulse, which indicates the maximum possible
variability and hence a flat spectrum that contains an equal amount of all frequencies from f = 0 to f = ∞. In fact,
it is for this reason that this type of noise is described as white in analogy with white light that contains an equal
amount of all colours. In general, a clear picture of the spectral content of any signal is given by the power spectral
density of the signal which happens to be the FT of the autocorrelation function of the signal, as is explored further
in the next chapter.
We may now formally define the autocorrelation function Rg (𝜏) of a signal g(t) for various signals, including
nonperiodic power signals (e.g. ergodic random process)
⎧ ∞
⎪ g(t)g(t − 𝜏)dt, Energy signal
⎪∫−∞
⎪ T∕2
⎪1
⎪ g(t)g(t − 𝜏)dt, Periodic power signal
Rg (𝜏) = ⎨ T ∫−T∕2 (3.114)
⎪ 𝕋 ∕2
Nonperiodic power signal
⎪ lim 1
Telegram: @Persian_IEEEComSoc

g(t)g(t − 𝜏)dt,
⎪𝕋 →∞ 𝕋 ∫−𝕋 ∕2 and ergodic random process
⎪
⎪E[X(t)X(t − 𝜏)], Stationary random process X
⎩
The autocorrelation function Rg (𝜏) of a real-valued signal g(t) has the following important properties:
● Rg (𝜏) has an even symmetry about 𝜏 = 0. That is, Rg (𝜏) = Rg (−𝜏). When comparing g(t) with a delayed version
of itself, what matters is the amount of shift, not the direction. Thus, we could have defined Rg (𝜏) by replacing
g(t − 𝜏) on the right-hand side of Eq. (3.114) with g(t + 𝜏).
● Rg (𝜏) has a maximum value at 𝜏 = 0 equal to the energy E of the energy signal or power P of the power signal
{
E, Energy signal
Rg (𝜏)max = Rg (0) = (3.115)
P, Power signal
 3.5 Signal Characterisation 173

● Rg (𝜏) exhibits the same periodicity in 𝜏 as is contained within g(t). If, however, g(t) has zero mean and no periodic
component then Rg (∞) = 0. This is just stating the obvious, that is g(t) and g(t + 𝜏) will be more strongly alike
whenever 𝜏 is an integer multiple of the period of g(t); but g(t) and g(t + 𝜏) become completely uncorrelated for
large 𝜏 if there is no repeating pattern (including a DC component) or periodicity in g(t).
● Rg (𝜏) and the spectral density of g(t) form a FT pair. This property is discussed further in the next chapter. It is
particularly useful in digital communications, providing a means to determine the spectral content of a signal
(such as a random power signal) that is not directly Fourier transformable. This is done by deriving Rg (𝜏) using
the statistical description of the signal and then taking the FT of Rg (𝜏) to obtain the spectral density, which is a
nonnegative, real and even function of frequency.

Worked Example 3.6 Autocorrelation Function of Selected Waveforms

We wish to determine the autocorrelation function of the following signals:
(a) A sinusoidal signal Acos(2𝜋ft + 𝜙).
(b) A rectangular pulse train of amplitude A and duty cycle d (Figure 3.19a).
(c) A bipolar random binary waveform in which a rectangular pulse of duration T b is sent with amplitude +A for
binary 1 and –A for binary 0, and binary 1’s and 0’s are equally likely to occur.
(a) In this case, g(t) = A cos(2𝜋ft + 𝜙); g(t − 𝜏) = A cos(2𝜋ft − 2𝜋f 𝜏 + 𝜙); and the signal is periodic with period
T = 1/f . Thus, the second line of Eq. (3.114) yields
T∕2
1
Rg (𝜏) = A cos(2𝜋ft + 𝜙) ⋅ A cos(2𝜋ft − 2𝜋f 𝜏 + 𝜙)dt
T ∫−T∕2
1∕2f
A2 f
= [cos(4𝜋ft − 2𝜋f 𝜏 + 2𝜙) + cos(2𝜋f 𝜏)]dt
2 ∫−1∕2f
[ ]
A2 f sin(4𝜋ft − 2𝜋f 𝜏 + 2𝜙) ||t=1∕2f t=1∕2f
= − | + cos(2𝜋f 𝜏)t|t=−1∕2f
2 4𝜋f |t=−1∕2f
[ ]
A2 f sin(−2𝜋 − 2𝜋f 𝜏 + 2𝜙) − sin(2𝜋 − 2𝜋f 𝜏 + 2𝜙) cos(2𝜋f 𝜏)
= +
2 4𝜋f f
A2
= cos(2𝜋f 𝜏)
2
Notice that cos(2𝜋f𝜏) and hence Rg (𝜏) has maxima at 2𝜋f𝜏 = 2𝜋n, i.e. when the shift 𝜏 = n/f , n = 0, 1, 2, 3, …
This indicates that g(t) contains a periodic component of fundamental frequency f , which is in fact what we
expect since in this case g(t) is a sinusoid of frequency f .
(b) Figure 3.20 shows the rectangular pulse train g(t) along with its shifted versions g(t − 𝜏). Note that pulse width
is here denoted w to distinguish it from 𝜏, which now represents the shift parameter. Although g(t) is a periodic
power signal and we could again use the second line of Eq. (3.114) to solve for Rg (𝜏), it is more insightful to
Telegram: @Persian_IEEEComSoc

employ the third line of that equation and compute over an interval 𝕋 = NT spanning N pulses, and let N → ∞
to satisfy that equation. Also, a positive value of 𝜏 corresponds to a shift to the right, whereas a negative value
is a shift to the left, but both have the same effect on the computation, so we will disregard the sign of 𝜏 by
using |𝜏|.
Figure 3.20a illustrates a shift in the range 0 ≤ |𝜏| ≤ w in which all N pulses of g(t) overlap with their corre-
sponding pulses of g(t − 𝜏) in the shaded area of width W = w − |𝜏|. The product g(t)g(t − 𝜏) = A2 . Thus, the
total area under g(t)g(t − 𝜏) contributed by this overlap is
( ) ( )
1 2
NA2 W = NA2 (w − |𝜏|) = NA2 w 1 − |𝜏| = NA2 w 1 − |𝜏|
( ) w 2w
𝜏
= NA2 w trian
2w
 174 3 Time Domain Analysis of Signals and Systems

(a) 0 ≤ |τ| ≤ w: W g(t) g(t – τ)

A
d = w/T
t
τ –T/2 w T/2
T
(b) w ≤ |τ| ≤ T – w:

τ t

(c) T – w ≤ |τ| ≤ T + w:

τ t

(d) Rg(τ)
A2d

τ
–T –w 0 w T

Figure 3.20 Worked Example 3.6(b): Rectangular pulse train with various shifts in (a)–(c) and its autocorrelation function
in (d).

where we have used the definition of the triangular pulse in Eq. (2.21).
Figure 3.20b shows a shift in the range w ≤ |𝜏| ≤ T − w, where there is no pulse overlap and therefore no
contribution to the total area.
Any shift beyond |𝜏| = T − w will simply repeat this pattern, as illustrated in Figure 3.20c, so that there is
recurrently an overlap over a shift range 2w followed by no pulse overlap over a shift range T − 2w.
In general, there is overlap of pulses only when the shift 𝜏 is in the range

mT − w ≤ 𝜏 ≤ mT + w, m = 0, ±1, ±2, ±3, · · ·

causing the nth pulse of g(t − 𝜏) to overlap with the (n + m)th pulse of g(t). The width W of the overlap increases
linearly from W = 0 at 𝜏 = mT − w to W = w at 𝜏 = mT, and then decreases linearly from W = w at 𝜏 = mT
to W = 0 at 𝜏 = mT + w. Thus, W is a train of triangular pulses spaced along the 𝜏 axis with peaks at 𝜏 = mT,
and amplitude w and width 2w. That is
( )
𝜏 − mT
W = w trian , m = 0, ±1, ±2, ±3, · · ·
Telegram: @Persian_IEEEComSoc

2w
Multiplying W by number of pulses N and height g(t)g(t − 𝜏) = A2 yields the required integration result for
the autocorrelation function
𝕋 ∕2
1
Rg (𝜏) = lim g(t)g(t − 𝜏)dt
𝕋 →∞ 𝕋 ∫−𝕋 ∕2

1 A2 W
= lim [NA2 W] =
N→∞ NT T
( )
2
Aw 𝜏 − mT
= trian
T ( 2w )
𝜏 − mT
2
= A d trian , m = 0, ±1, ±2, ±3, · · ·
2w
 3.5 Signal Characterisation 175

Random
(a) 1 0 0 1 0 1 1 1 0 0 0 0 1 1 0 1 …
bit stream
+A

g(t – τ)
g(t)
τ t

–A

Tb

(b) Rg(τ)
A2

τ
–Tb Tb

Figure 3.21 Worked Example 3.6(c): Bipolar random binary waveform g(t) in (a) and its autocorrelation function in (b).

Figure 3.20d provides a sketch of this autocorrelation function. The following interesting features of Rg (𝜏)
should be noted. First, it has peaks at 𝜏 = mT, m = 0, ±1, ±2, …, which indicates that g(t) contains a periodic
component with fundamental frequency 1/T. Also, each nonzero portion or lobe of Rg (𝜏) has width 2w, which
becomes narrower (signifying higher variability and spectral content in g(t)) as the width w of the pulses of
g(t) decreases. Finally, Rg (0) = A2 d, which (from Eq. (3.105)) is the power of g(t) as expected.
(c) The waveform of g(t) is shown in Figure 3.21a for a random sequence of binary 1’s and 0’s. The solution steps
are broadly like (b) above but with important differences that reflect the randomness of this case. We obtain
Rg (𝜏) using the third line of Eq. (3.114) and the statistical information provided. To do this we span N bits and
an interval 𝕋 = NT b and let N → ∞ to satisfy Eq. (3.114). When we shift g(t) by |𝜏| ≤ T b , the kth bit (denoted
bk ) of g(t) overlaps the kth bit of g(t − 𝜏) over a width T b − |𝜏| shaded in Figure 3.21a for the first few bits. In
Telegram: @Persian_IEEEComSoc

each overlap region, the product g(t)g(t − 𝜏) is (A) × (A) = A2 for bk = 1, and (−A) × (−A) = A2 for bk = 0, and
the area is A2 (T b − |𝜏|). Thus, total area under g(t)g(t − 𝜏) contributed by the overlap of corresponding bits in
g(t) and g(t − 𝜏) isNA2 (Tb − |𝜏|).
The above shift also creates an overlap between bk in g(t − 𝜏) and bk+1 in g(t) over an interval 𝜏. The product
g(t)g(t − 𝜏) will be A2 if bk = bk+1 and will be –A2 if bk ≠ bk+1 . Since these two bits are equally likely to be the
same or different, the total contribution to the area from the overlap of adjacent bits is zero. Furthermore, any
shift by 𝜏 > T b will only create overlaps between noncorresponding bits bk in g(t − 𝜏) and bk+n in g(t), where
n = 1, 2, 3, …, and this also makes zero total contribution to the area since bk and bk+n are equally likely to be
the same (+A and + A, or −A and −A) or different (+A and −A, or −A and + A). Thus
𝕋 ∕2
1
Rg (𝜏) = lim g(t)g(t − 𝜏)dt
𝕋 →∞ 𝕋 ∫−𝕋 ∕2
 176 3 Time Domain Analysis of Signals and Systems

⎧ 1
⎪ lim NA2 (Tb − |𝜏|), |𝜏| ≤ Tb
= ⎨N→∞ NT b
⎪0, otherwise
⎩
{ 2
A (1 − |𝜏|∕Tb ), |𝜏| ≤ Tb
=
0, otherwise
( )
𝜏
= A2 trian
2Tb
The autocorrelation function of a bipolar random binary waveform of amplitude A and bit duration T b is
therefore a triangular pulse having duration 2T b and amplitude A2 , as shown in Figure 3.21b. Notice that
Rg (𝜏)max = Rg (0) = power of g(t); the duration 2T b of Rg (𝜏) is directly proportional to bit duration T b and will
become narrower (signifying higher variability and spectral content) as bit duration decreases or, conversely,
bit rate increases; and there is no sequence of local maxima in Rg (𝜏), which indicates that g(t) does not contain
any periodic components.

3.5.6 Covariance and Correlation Coefficient

It is often necessary to obtain an objective estimate of the similarity between two signals using a quantity that
gives an indication of the level of synchronisation in the variability of the two signals. For example, during each
signalling interval the receiver in a binary digital transmission system may compare an incoming signal with a
standard waveform (e.g. the signal designated for conveying binary 0) in order to establish the identity of the
received signal. Covariance is the most used measure of joint variability and is applicable to both random and
deterministic signals. When normalised to have a value in the range (−1, 1) it is referred to as the correlation
coefficient of the two signals.
The covariance of two signals X and Y , denoted cov[X, Y ], is the expectation of the product of the deviations of
the two signals from their respective means 𝜇 X and 𝜇 Y . Thus

cov[X, Y ] = E[(X − 𝜇X )(Y − 𝜇Y )] (3.116)

Since the expectation operation is linear, we may expand the above equation to express covariance of two signals
as the expectation of the product of the two signals less the product of the means (i.e. expected values) of the signals

cov[X, Y ] = E[(X − 𝜇X )(Y − 𝜇Y )]

= E[XY − 𝜇Y X − 𝜇X Y + 𝜇X 𝜇Y ]
= E[XY ] − 𝜇Y E[X] − 𝜇X E[Y ] + 𝜇X 𝜇Y
Telegram: @Persian_IEEEComSoc

= E[XY ] − 𝜇Y 𝜇X − 𝜇X 𝜇Y + 𝜇X 𝜇Y
= E[XY ] − 𝜇X 𝜇Y

That is, we have an expression for covariance in terms of uncentred moments, namely

cov[X, Y ] = E[XY ] − E[X]E[Y ] (3.117)

Two signals are said to be uncorrelated if their covariance is zero. You will be able to show in Question 3.10 that if
two signals X and Y are independent random variables then E[XY ] = E[X]E[Y ] = 𝜇X 𝜇 Y . It therefore follows from
Eq. (3.117) that the covariance of two independent random signals is zero. Thus, if X and Y are independent then
they are also uncorrelated. Note, however, that the converse is not necessarily true. That is, two independent signals
are always uncorrelated, whereas two uncorrelated signals are not necessarily independent. See Question 3.11.
 3.5 Signal Characterisation 177

The covariance of two signals gives an indication of the linear relationship between the two signals. If both
signals vary in sync such that one signal is large when the other is large and small when the other is small then
their covariance is positive, whereas if one is small when the other is large (and vice versa) then their covariance
is negative. An indication of both the strength and the trend of the association between two signals is obtained by
dividing their covariance by the product of the standard deviations 𝜎 X and 𝜎 Y of the two signals to give a value in
the range (−1, 1), known as the Pearson correlation coefficient of the two signals X and Y , which we will denote as
r X,Y . Thus

cov[X, Y ] E[(X − 𝜇X )(Y − 𝜇Y )]

rX,Y = = √ √
𝜎X 𝜎Y E[(X − 𝜇X )2 ] E[(Y − 𝜇Y )2 ]
E[XY ] − E[X]E[Y ]
= √ (3.118)
(E[X 2 ] − (E[X])2 )(E[Y 2 ] − (E[Y ])2 )

where we have employed Eq. (3.117) for covariance and Eq. (3.22) for variance.
We will often only have a finite set of data comprising N corresponding values of each of X and Y , namely {x1 ,
x2 , …, xN } and {y1 , y2 , …, yN }. In such cases, we perform the above averaging operations over this dataset to obtain
unbiased estimates of covariance and Pearson correlation coefficient, respectively referred to as sample covariance
𝓬𝓸𝓿[X, Y ] and sample Pearson correlation coefficient 𝓻X,Y

1 ∑
N
𝓬𝓸𝓿[X, Y ] = (x − X)(yk − Y )
N − 1 k=1 k
∑
N
(xk − X)(yk − Y )
k=1
𝓻X,Y = √ √
∑
N
∑
N
(xk − X)2 (yk − Y )2
k=1 k=1

1 ∑ 1 ∑
N N
where, X = x ; Y= y (3.119)
N k=1 k N k=1 k

Note that if the population means E[X] ≡ 𝜇 X and E[Y ] ≡ 𝜇 Y are known then these should be used in the above
expressions in place of the sample means X and Y , in which case the factor 1/(N − 1) in the first line is replaced
with 1/N.
The Pearson correlation coefficient is computed on deviations of the raw data from the mean and is widely used
Telegram: @Persian_IEEEComSoc

in the field of statistics. However, in communication systems applications correlation coefficient is defined slightly
differently, based on directly aggregating the signal values (rather than their deviations from the mean). We will
refer to this variant definition simply as the correlation coefficient of signals X and Y , denoted as 𝜌X,Y . and given
by the expression

E[XY ]
𝜌X,Y = 1
(3.120)
2
(E[X 2 ] + E[Y 2 ])

The manner of carrying out the required averaging in the above equation is dictated entirely by the type of signals
involved. Denoting the signals as g1 (t) and g2 (t), the averaging is carried out over the entire time duration of the
signals if they are energy signals, power signals, or ergodic random processes, or over one cycle if the signals are
 178 3 Time Domain Analysis of Signals and Systems

periodic. Thus, we have for energy signals of duration T s

1 T
Ts
∫0 s g1 (t)g2 (t)dt
𝜌g1 (t),g2 (t) = ( )
1 1 T 1 T
2 Ts
∫0 s g12 (t)dt + Ts
∫0 s g22 (t)dt
T
∫0 s g1 (t)g2 (t)dt
= ( )
1 T T
2
∫0 s g12 (t)dt + ∫0 s g22 (t)dt

From the definition of energy in Eq. (3.101), we see that the denominator in the last line is the average of the
energies E1 and E2 of g1 (t) and g2 (t), respectively. Thus
T T
∫0 s g1 (t)g2 (t)dt 2 ∫0 s g1 (t)g2 (t)dt
𝜌g1 (t),g2 (t) = = (3.121)
Average energy E1 + E2

Similarly, if g1 (t) and g2 (t) are periodic signals having period T and respective powers P1 and P2 then
T∕2
2 ∫−T∕2 g1 (t)g2 (t)dt
𝜌g1 (t),g2 (t) = (3.122)
(P1 + P2 )T

And if the signals are nonperiodic power signals or ergodic random processes with respective powers P1 and P2 ,
then
𝕋 ∕2
lim 2 ∫−𝕋 ∕2 g1 (t)g2 (t)dt
𝕋 →∞ 𝕋
𝜌g1 (t),g2 (t) = (3.123)
P1 + P2

Two signals X and Y are said to be orthogonal if the expectation of their product is zero, i.e. if E[XY ] = 0. It
follows from Eq. (3.120) that orthogonal signals have a zero correlation coefficient, and conversely that if two sig-
nals have a zero correlation coefficient then they are orthogonal. We will see in later chapters that the principle
of signal orthogonality and the process of signal correlation both play a central role in signal transmission and
detection. Throughout this book we make use of the correlation coefficient as defined in Eq. (3.120) and applied
to special cases in Eqs. (3.121) to (3.123). For comparison, Table 3.1 gives a summary of the properties and special
values of the two correlation coefficients discussed in this section. In the table and Eq. (3.124) below, a, b, c, d are
constants and b > 0, d > 0. There is an important feature that is worth highlighting. In the third row of entries we
see that the standard correlation coefficient distinguishes between two signals X and aX that differ only in ampli-
tude, whereas the Pearson correlation coefficient does not. This is an important feature that makes the standard
correlation coefficient a more suitable parameter in telecommunications where it is often necessary to distinguish
between signals that differ only in amplitude (e.g. amplitude shift keying (ASK) and amplitude and phase shift
Telegram: @Persian_IEEEComSoc

keying (APSK) signals). The fourth row also shows that, unlike the standard correlation coefficient, the Pearson
correlation coefficient does not distinguish between two signals that differ by a constant value. In general, the
Pearson correlation coefficient is invariant under changes to a signal by a constant additive term or multiplicative
factor, whereas the standard correlation coefficient is sensitive to any of these changes. That is

ra+bX,c+dY = rX,Y
𝜌a+bX,c+dY ≠ 𝜌X,Y (3.124)

We also see from the last two rows of the table that the Pearson coefficient is more reliable for checking for the
independence of random signals, whereas the standard coefficient is better suited to detecting the orthogonality
of deterministic signals.
 3.5 Signal Characterisation 179

Table 3.1 Selected properties and values of correlation coefficients.

Correlation coefficient

Signal 1 Signal 2 Pearson, Eq. (3.118) Standard, Eq. (3.120)

X X r X,X = +1 𝜌X,X = +1
X −X r X,−X = −1 𝜌X,−X = −1
X bX r X,bX = +1 𝜌X,bX = 2b/(1 + b2 )
X a+X r X,a+X = +1 𝜌X,a+X =
aE[X] + E[X 2 ]
aE[X] + E[X 2 ] + a2 ∕2
X Y orthogonal to X r X,Y = 0 if and only if 𝜌X,Y = 0
i.e. E[XY ] = 0 E[X] = 0 or E[Y ] = 0
X Y and X independent r X,Y = 0 𝜌X,Y = 0 if and only if
i.e. E[XY ] = E[X]E[Y ] E[X] = 0 or E[Y ] = 0

Worked Example 3.7

In M-ary phase shift keying (M-PSK) modulation scheme, data bits are transmitted in groups of log2 M bits at a
time using a set of M unique sinusoidal symbols defined by
( ) ( )
t − 12 Ts 2𝜋k
gk (t) = Arect cos 2𝜋fc t + 𝜃o +
Ts M
k = 0, 1, 2, 3, · · · , M − 1; fc = n∕Ts
One of the M symbols is selected and transmitted in each signalling interval of duration T s depending on
the identity of the log2 M bits within that interval. Each symbol is a sinusoidal carrier of amplitude A, phase
𝜃 o + 2𝜋k/M (where 𝜃 o is a constant phase offset), and frequency f c = n/T s , where n is a positive integer,
which ensures that (with a period T s /n) the carrier completes an integer number of cycles in each signalling
interval.
Derive expressions for:

(a) Symbol energy Ek of each transmitted symbol gk (t).

(b) The correlation coefficient of adjacent symbols (i.e. two symbols that differ in phase by the smallest amount)
in the transmission scheme.
Telegram: @Persian_IEEEComSoc

(( ) )
(a) The rectangular pulse factor rect t − 12 Ts ∕Ts in the expression for symbol gk (t) ensures that the symbol
has duration T s . See Eq. (2.16) for the definition of a rectangular pulse. Thus, each symbol is a sinusoid of
amplitude A and duration T s , which yields energy Ek given by
Ek = Power × Duration
2 A2 T
= A2 × Ts = 2 s
All symbols in a PSK transmission scheme therefore have the same energy given by the above
expression.
(b) These are energy signals, so Eq. (3.121) yields the correlation coefficient of adjacent symbols gk (t) and gk + 1 (t),
k = 0, 1, 2, …, M − 2 as
 180 3 Time Domain Analysis of Signals and Systems

T
2 ∫0 s gk (t)gk+1 (t)dt
𝜌gk (t),gk+1 (t) =
Ek + Ek+1
Ts ( ) ( )
2 2𝜋k 2𝜋(k + 1)
= 2 A cos 2𝜋fc t + 𝜃o + ⋅ A cos 2𝜋fc t + 𝜃o + dt
A Ts ∫0 M M
Ts ( ) ( )
2 2𝜋k 2𝜋k 2𝜋
= 2 A2 cos 2𝜋fc t + 𝜃o + ⋅ cos 2𝜋fc t + 𝜃o + + dt
A Ts ∫0 M M M

Using Eq. (B.6) in Appendix B for the product of cosines in the integrand, and noting that f c T s = n and that
sin(4𝜋n + 𝛽) = sin(𝛽), we obtain
Ts [ ( ) ( )]
2 1 2𝜋 4𝜋k 2𝜋
𝜌gk (t),gk+1 (t) = 2 ⋅ A2 cos + cos 4𝜋fc t + 2𝜃o + + dt
A Ts 2 ∫0 M M M
( ) t=Ts
⎡ 4𝜋k 2𝜋 ⎤||
( ) sin 4𝜋f t + 2𝜃 + + M ⎥|
1 ⎢ 2𝜋 c o
⎥||
M
= ⎢ t cos +
Ts ⎢ M 4𝜋fc ⎥||
⎣ ⎦|t=0
( ) ( )
⎡ 4𝜋k 2𝜋 4𝜋k 2𝜋 ⎤
⎢ ( ) sin 4𝜋fc Ts + 2𝜃o + M + M − sin 2𝜃o + M + M ⎥
1 2𝜋
Ts ⎢⎢ s ⎥
= T cos +
M 4𝜋fc ⎥
⎣ ⎦
1
= [Ts cos(2𝜋∕M)]
Ts
= cos(2𝜋∕M)
Dropping the subscripts for convenience, we see that 𝜌 = −1 for binary phase shift keying (BPSK, M = 2); 𝜌 = 0
for quaternary PSK (QPSK, M = 4); 𝜌 = 0.7071 for 8PSK; 𝜌 = 0.9239 for 16PSK; etc. Thus, 𝜌 increases rapidly
towards +1 as M increases, and this has great significance in the design of communication systems, necessitating
a trade-off between transmitted power and transmission bandwidth, an issue we explore further in later chapters.

3.6 Linear Time Invariant System Analysis

A linear time invariant (LTI) system is one that is both linear and time invariant. Let the transformation performed
by a system on its input x(t) to obtain its output y(t) be represented by the operator  such that we may write
y(t) = {x(t)}. An LTI system therefore obeys the following rules
Telegram: @Persian_IEEEComSoc

Given that
y1 (t) = {x1 (t)}
and
y2 (t) = {x2 (t)}
then
a1 y1 (t) + a2 y2 (t) = {a1 x1 (t) + a2 x2 (t)} (i)
y1 (t − 𝜏) = {x1 (t − 𝜏)} (ii) (3.125)
where a1 and a2 are arbitrary constants. Rule (i) is known as the principle of superposition, and states that the
system produces an output by always doing the same thing to every input and then adding the results together.
 3.6 Linear Time Invariant System Analysis 181

Rule (ii) expresses time-invariance: what the system does is not dependent on the time the input is applied. That
is, the only change to the system output in response to a delayed input is a delay of the same amount. In practice
the system will not be indefinitely time-invariant, and it will be acceptable to treat the system as time-invariant
if it is time-invariant only for the duration of a call or connection. In this case there will be a new challenge to
obtain the system or channel characterisation (a task known as channel estimation) at the start of each call since
the channel behaviour may vary from call to call. We will also assume that the system is causal, which means that
there is no system response before an input is applied. Refer to Section 2.10 in Chapter 2 for a detailed discussion
of basic system properties.
It will be useful in the ensuing discussion to note the following alternative expression of Eq. (3.125) which uses
R
the notation x(t) −−−−→ y(t) to mean ‘x(t) yields response y(t)’

Given that
R
x1 (t) −−−−→ y1 (t)
and
R
x2 (t) −−−−→ y2 (t)
then
R
a1 x1 (t) + a2 x2 (t) −−−−→ a1 y1 (t) + a2 y2 (t) (i)
R
x1 (t − 𝜏) −−−−→ y1 (t − 𝜏) (ii) (3.126)

3.6.1 LTI System Response

An LTI system may be fully characterised in the time domain by its impulse response h(t), which is the output
of the signal when the excitation or input signal is the unit impulse function 𝛿(t), as illustrated in Figure 3.22 for
both continuous-time (CT) and discrete-time (DT) LTI systems. We are interested in the response y(t) of a CT LTI
system to an arbitrary excitation x(t) and in the response y(n) of a DT LTI system to an arbitrary input sequence
x[n], as shown in Figure 3.23. A clarification of notation is important here. We will use square brackets such as
in x[n], y[n], h[n], g[n], etc., to refer to the entire sequence of a DT signal, and round brackets such as in y(n) to
denote the nth sample of the sequence. So, x(n) is the present sample of the input, x(n − 1) is the immediate past
input sample, x(n + 1) is the immediate future or next input sample, and so on. Also, the present output sample
y(n) will in general be computed using a mathematical operation that involves one or more input samples (i.e.
x[n]) rather than just the present input sample x(n).
Considering first the CT case, Figure 3.24a shows a sample (of small thickness Δ𝜏) of an arbitrary input signal
x(t) taken at time t = 𝜏. This sample has height x(𝜏) and hence area x(𝜏) ⋅ Δ𝜏, and may be represented as x(𝜏) ⋅ Δ𝜏 ⋅
Telegram: @Persian_IEEEComSoc

𝛿(t − 𝜏), which is an impulse of weight x(𝜏) ⋅ Δ𝜏 located at t = 𝜏. In view of Eq. (3.126), the response of the system

Figure 3.22 Impulse response.

δ(t) CT h(t)
LTI
System

δ[n] DT h(n)
LTI
System
 182 3 Time Domain Analysis of Signals and Systems

Figure 3.23 Response to arbitrary input.

x(t) CT LTI y(t)
System
h(t)

x[n] DT LTI y(n)

System
h[n]

x(t)
x(τ) · Δτ · δ(t – τ) LTI x(τ) · Δτ · h(t – τ)
Δτ System
(a) h(t)
t
τ
x(t)
∞ ∞
Σ x(kΔτ) · Δτ · δ(t – kΔτ) LTI Σ x(kΔτ) · Δτ · h(t – kΔτ)
k = –∞ k = –∞
(b) System
t h(t)
kΔτ

x(t)
LTI
System
(c) ∞ h(t) ∞
t x(t) = ∫ x(τ)δ(t – τ)dτ y(t) = ∫ x(τ)h(t – τ)dτ
τ –∞ –∞
= x(t) * h(t)

Figure 3.24 Response y(t) of an LTI CT system to an arbitrary input x(t).

to this weighted and delayed impulse will be a similarly weighted and delayed impulse response. That is
R
x(𝜏) ⋅ Δ𝜏 ⋅ 𝛿(t − 𝜏) −−−−→ x(𝜏) ⋅ Δ𝜏 ⋅ h(t − 𝜏) (3.127)
Figure 2.24b shows the entire signal x(t) being approximated as a sequence of samples, each of thickness Δ𝜏,
taken at t = kΔ𝜏, where k takes on integer values from −∞ to ∞, as necessary in order to span the entire duration
of x(t). The kth sample occurs at t = kΔ𝜏, has height x(kΔ𝜏) and thickness Δ𝜏, and is therefore the impulse x(kΔ𝜏) ⋅
Δ𝜏 ⋅ 𝛿(t − kΔ𝜏). The input signal x(t) is therefore approximated by the sum of these impulses from k = −∞ to k = ∞
as
∑
∞
x(t) ≈ x(kΔ𝜏) ⋅ Δ𝜏 ⋅ 𝛿(t − kΔ𝜏) (3.128)
Telegram: @Persian_IEEEComSoc

k=−∞

In line with Eq. (3.126), the response to this sum of weighted and delayed impulses will be a sum of similarly
weighted and delayed impulse responses. That is
∑
∞
R ∑
∞
x(kΔ𝜏) ⋅ Δ𝜏 ⋅ 𝛿(t − kΔ𝜏) −−−−→ x(kΔ𝜏) ⋅ Δ𝜏 ⋅ h(t − kΔ𝜏) (3.129)
k=−∞ k=−∞

In the limit Δ𝜏 → 0, the approximation of x(t) in Eq. (3.128) becomes exact, as shown in Figure 3.24c, the discrete
instants kΔ𝜏 become a continuous variable 𝜏, and the summations in the Eq. (3.129) (and in Figure 3.24b) become
integrations with Δ𝜏 replaced by d𝜏, so that
∞ ∞
R
x(t) = x(𝜏)𝛿(t − 𝜏)d𝜏 −−−−→ y(t) = x(𝜏)h(t − 𝜏)d𝜏 ≡ x(t) ∗ h(t)
∫−∞ ∫−∞
 3.6 Linear Time Invariant System Analysis 183

Notice that the left-hand side of the arrow is simply the sifting property of the impulse function encountered
earlier in Eq. (2.31), taking the even property of the impulse function into account. The right-hand side is called
the convolution integral, the convolution operation being denoted as above with an asterisk. This is a very important
result. It states that the output y(t) of an LTI system in response to an arbitrary input x(t) is obtained by convolving
x(t) with the impulse response h(t) of the system
∞
y(t) = x(𝜏)h(t − 𝜏)d𝜏 ≡ x(t) ∗ h(t) (3.130)
∫−∞

Noting that t is a constant as far as the integration is concerned, and substituting t − 𝜆 = 𝜏 (so that d𝜏 = −d𝜆, 𝜆
= t − 𝜏 = −∞ when 𝜏 = ∞, and 𝜆 = ∞ when 𝜏 = −∞) yields
−∞ ∞
y(t) = − x(t − 𝜆)h(𝜆)d𝜆 = h(𝜆)x(t − 𝜆)d𝜆
∫∞ ∫−∞

Restoring our use of 𝜏 (in place of 𝜆) yields

∞
y(t) = h(𝜏)x(t − 𝜏)d𝜏 ≡ h(t) ∗ x(t) (3.131)
∫−∞

Thus, the convolution operation is commutative, which means that the same result is obtained whatever the
ordering of the signals being convolved. The convolution operation is also associative and distributive as sum-
marised in the next equation.

x(t) ∗ h(t) = h(t) ∗ x(t) Commutative law

x(t) ∗ [h1 (t) ∗ h2 (t)] = [x(t) ∗ h1 (t)] ∗ h2 (t) Associative law
x(t) ∗ [h1 (t) + h2 (t)] = x(t) ∗ h1 (t) + x(t) ∗ h2 (t) Distributive law (3.132)

These properties have important practical implications. For example, it implies that the order of arrangement
of a cascade connection of LTI stages does not affect the overall output, and that the impulse response of a system
comprising two LTI stages is the convolution of the impulse responses of the two stages. Note that a cascade
connection means a series connection in which there is impedance matching between stages so that there is no
signal reflection and the output of one stage is entirely delivered to the next stage and equals the input of that
stage.
If the LTI system is causal then h(t) = 0 for t < 0, since h(t) is the response to a signal 𝛿(t) that is applied at t = 0.
Since the integrand in Eq. (3.131) has h(𝜏) as a factor, the integral will be zero in the interval 𝜏 = (−∞, 0) where
h(𝜏) = 0, and hence
∞ 0 ∞
Telegram: @Persian_IEEEComSoc

y(t) = h(𝜏)x(t − 𝜏)d𝜏 = h(𝜏)x(t − 𝜏)d𝜏 + h(𝜏)x(t − 𝜏)d𝜏

∫−∞ ∫−∞ ∫0
∞
= h(𝜏)x(t − 𝜏)d𝜏
∫0

Now making the substitution t − 𝜏 = 𝜆 in the last integral, which means

𝜏 = t − 𝜆, d𝜏 = −d𝜆, 𝜆 = t when 𝜏 = 0, 𝜆 = −∞ when 𝜏 = ∞, we obtain
∞ −∞ t
y(t) = h(𝜏)x(t − 𝜏)d𝜏 = − h(t − 𝜆)x(𝜆)d𝜆 = x(𝜆)h(t − 𝜆)d𝜆
∫0 ∫t ∫−∞
t
≡ x(𝜏)h(t − 𝜏)d𝜏
∫−∞
 184 3 Time Domain Analysis of Signals and Systems

The response y(t) of a causal LTI system having impulse response h(t) to an input signal x(t) is therefore given
by the expressions
∞
y(t) = h(𝜏)x(t − 𝜏)d𝜏
∫0
t (Causal system) (3.133)
= x(𝜏)h(t − 𝜏)d𝜏
∫−∞

Notice that the second integral indicates that only values of the input occurring prior to time t in the interval
(−∞, t) contribute to the output y(t) at time t, and specifically that future input values occurring in the interval (t,
∞) do not contribute to y(t). This is a manifestation of the system’s causality. If both the system and the input signal
are causal (i.e. h(t), x(t) = 0 for t < 0), then the second integral, which contains the factor x(𝜏) in its integrand, will
be zero in the interval 𝜏 = (−∞, 0) where x(𝜏) = 0, so that we have
t
y(t) = x(𝜏)h(t − 𝜏)d𝜏
∫0 (Causal input signal
t (3.134)
and causal system)
= h(𝜏)x(t − 𝜏)d𝜏
∫0

where we have used the substitution t − 𝜏 = 𝜆 (as previously done) to derive the last integral from the first. The
first integral indicates that, because the system is causal, only past inputs (starting in this case at t = 0 in view of
the input signal’s causality) contribute to y(t).
Now turning our attention to a DT system that is characterised by impulse response h[n], let the system operation
be denoted by the operator , so that

h(n) = {𝛿(n)} (3.135)

Recall from Eq. (2.34) that an arbitrary DT signal x[n] may be expressed as a sum of delayed weighted impulses
as
∑
∞
x[n] = x(k)𝛿(n − k)
k=−∞

where x[n] denotes the entire DT signal, whereas x(k) is the value of x[n] at the kth sampling instant. By virtue of
linearity, the response y(n) of the system to x[n] will be
{ ∞ }
∑
y(n) = {x[n]} =  x(k)𝛿(n − k)
k=−∞
∑
Telegram: @Persian_IEEEComSoc

∞
= x(k){𝛿(n − k)}
k=−∞

In view of time invariance, it follows from Eq. (3.135) that

{𝛿(n − k)} = h(n − k)

Therefore, if an LTI system is characterised by impulse response h[n] then its output y(n) in response to an
arbitrary input x[n] is given by the convolution sum

∑
∞
y(n) = x(k)h(n − k) = x[n] ∗ h[n] (3.136)
k=−∞
 3.6 Linear Time Invariant System Analysis 185

x(k)
12

10
8
(a)
6
4
2
0 k
–5 –4 –3 –2 –1 0 1 2 3 4 5 6

h(n)
2

(b)

n
–3 –2 –1 0 1 2 3 4 5 6 7

Figure 3.25 Finite duration sequences: (a) Input sequence; (b) Finite impulse response.

This convolution sum is the discrete equivalent of the convolution integral of Eq. (3.130). In the above summa-
tion, substituting n − k = m (and hence k = n − m, m = −∞ when k = ∞, m = ∞ when k = −∞) yields
∑
−∞
∑
∞
y(n) = x(n − m)h(m) = h(m)x(n − m)
m=∞ m=−∞
∑∞
≡ h(k)x(n − k) = h[n] ∗ x[n] (3.137)
k=−∞

Thus, y(n) = x[n] ∗ h[n] = h[n] ∗ x[n], which means that the convolution of DT signals is commutative just as
is the convolution of CT signals. In fact, the commutative, associative, and distributive laws discussed earlier for
convolution of CT signals are also applicable in the DT case.
It is often the case that the input sequence x[n] is of finite duration so that x(k) in Eq. (3.136) is zero outside the
interval k = (k1 , k2 ). For example, in Figure 3.25a x(k) is zero outside the range k = (−2, 3), so k1 = −2 and k2 = 3.
The summation in Eq. (3.136) now only needs to be performed within this range to obtain system response
Telegram: @Persian_IEEEComSoc

∑
k2
y(n) = x(k)h(n − k), x(k) = 0 for k < k1 or k > k2 (3.138)
k=k1

Making the substitution m = n − k in the above summation, and subsequently restoring k as the index of sum-
mation (in place of m) yields

∑
n−k2
y(n) = h(k)x(n − k)
k=n−k1

If in addition to the input sequence x[k] being confined to the range k = (k1 , k2 ), the impulse response h[k] is also
zero outside the range k = (n1 , n2 ) then the factor h(k) in the above summation will be zero when k (= n − k1 ) is
 186 3 Time Domain Analysis of Signals and Systems

less than n1 , i.e. n < n1 + k1 , and when k (= n − k2 ) exceeds n2 , i.e. n > n2 + k2 . This means that the output sequence
y[n] will be zero for n < n1 + k1 or n > n2 + k2 , and so we may write

When
x(k) = 0 for k < k1 or k > k2
And
h(n) = 0 for n < n1 or n > n2
Then
⎧∑k2
⎪ x(k)h(n − k), n = n1 + k1 , · · · , →, · · · , n2 + k2
y(n) = ⎨k=k (3.139)
⎪ 1

⎩0, Otherwise
Figure 3.25b shows a plot of h[n] for a causal finite impulse response (FIR) system in which n1 = 0 (as is always
the case for a causal system) and n2 = 4.
Convolution is a key operation in signal transmission, so it is important to have complete clarity about its mean-
ing and evaluation. We do this in the next two sections through further discussions and worked examples that
emphasise both a graphical interpretation and direct mathematical computation in the case of CT signals and a
tabulated approach for DT signals. You may at this point wish to review the basic signal operations of time shifting
and time reversal discussed in Section 3.2.

3.6.2 Evaluation of Convolution Integral

The convolution y(t) = x(t) ∗ h(t) of two CT signals x(t) and h(t) has a value y(to ) at time t = to that is computed
through the following steps.

(a) Time reverse h(t) to obtain h(−t).

(b) Delay h(−t) by to to obtain h(−t + to ) ≡ h(to − t). These two steps can be denoted as h(to − t) ≡ 𝔻[𝕋 ℝ[h(t)], to ],
where 𝕋 ℝ[g(t)] denotes time reversal of signal g(t) and 𝔻[g(t), to ] denotes delay of g(t) by to .
(c) Multiply x(t) by h(to − t) to obtain the product waveform x(t)h(to − t).
(d) The desired value y(to ) is given by the total area under x(t)h(to − t). This is the integration of x(t)h(to − t) with
respect to t in the range t = (−∞, ∞).

The above steps describe a graphical approach to obtain the value of the convolution of two CT signals x(t) and
h(t) at any given time t = to . Mathematically, these steps correspond to determining the output at any arbitrary
time t by evaluating the integral
Telegram: @Persian_IEEEComSoc

∞
y(t) = x(𝜏)h(t − 𝜏)d𝜏 (3.140)
∫−∞

Worked Example 3.8 Graphical Evaluation of Convolution Integral

A system has the causal finite impulse response h(t) shown in Figure 3.26a. Determine the output y(t) of this
system when the input is the rectangular pulse x(t) also shown. Discuss in what ways the input signal x(t) has
been modified by this system.
We solve this problem by evaluating
∞
y(t) = Area under x(𝜏) × 𝔻[𝕋 ℝ[h(𝜏)], t] = x(𝜏)h(t − 𝜏)d𝜏
∫−∞
 3.6 Linear Time Invariant System Analysis 187

h(t)
3 x(t)
2
(a)
t, sec t, sec
0 4 –5 0 5

D[TR[h(τ)], t] ≡ h(t – τ), t ≤ – 5 h(t – τ), t ≥ 9

3
2 x(τ)
(b)
τ, sec
t –5 0 5 t–4 t

h(t – τ), –5 ≤ t ≤ –1 h(t – τ), 5 ≤ t ≤ 9

3 x(τ) 3 x(τ)
2 2
(c) a1
a2 a
τ, sec τ
t–4 –5 t 5 –5 t–4 5 t
b d
h(t – τ), –1 ≤ t ≤ 5
3
2 x(τ)
(d)
τ, sec
–5 t–4 0 t 5

Figure 3.26 Worked Example 3.8: (a) Impulse response h(t) and input signal x(t).

at enough values of t to obtain a smooth waveform of the output y(t). In carrying out this evaluation, it is convenient
to partition t into distinct regions within which the area under the product waveform is derived from the same
geometric shape.
Figure 3.26b shows that for t ≤ −5 and for t ≥ 9 there is no overlap between x(𝜏) and h(t − 𝜏), so the product
waveform x(𝜏)h(t − 𝜏) = 0. Thus

y(t) = 0, t ≤ −5
y(t) = 0, t ≥ 9

Next, the left-hand side of Figure 3.26c shows that for −5 ≤ t ≤ −1 the shaded portion of h(t − 𝜏) overlaps with
x(𝜏). The area under the product waveform is the area of a vertical trapezium (of base b and parallel sides a1 and
a2 ) multiplied by x(𝜏) = 2. Notice that b is the gap between t and − 5, so b = t − (−5) = t + 5. Also, a1 = 3, and the
Telegram: @Persian_IEEEComSoc

time taken by the linear slope to rise to level a2 is the gap between −5 and t − 4, which is −5 − (t − 4) = −(1 + t).
Since this slope rises through 3 units in 4 seconds, it means that a2 = −(1 + t) × 3∕4. Therefore, in this region
1
y(t) = (a1 + a2 ) × b × 2
(2 )
3
= 3 − (1 + t) (t + 5)
4
= 3(3 − t)(t + 5)∕4, −5 ≤ t ≤ −1

Another region of overlap, 5 ≤ t ≤ 9, is shown on the right-hand side of Figure 3.26c. Here the area under
the product waveform is the area of a triangle (of base d and height a) multiplied by x(𝜏) = 2. Note that
d = 5 − (t − 4) = 9 − t and a is the level reached in a time of d seconds by the above linear slope (which rises by
 188 3 Time Domain Analysis of Signals and Systems

12
y(t)
10

x(t)
2

0 t, sec
–5 –4 –3 –2 –1 0 1 2 3 4 5 6 7 8 9 10

Figure 3.27 Output y(t) of the system in Worked Example 3.8 along with the input x(t) for comparison.

3 units in 4 seconds), so a = 3d/4. Thus, in this region

1 3
y(t) = a × d × 2 = d2
2 4
3
= (9 − t)2 , 5≤t≤9
4
Figure 3.26d shows one more region of overlap in which h(t − 𝜏) lies fully within the interval of x(𝜏). The area
under the product waveform is therefore the area of a triangle (of base 4 and height 3) multiplied by x(𝜏) = 2. Thus,
in this region
1
y(t) = × 4 × 3 × 2 = 12, −1 ≤ t ≤ 5
2
To summarise, the output of the system is given by

⎧0, t ≤ −5
⎪3(3 − t)(t + 5)∕4, −5 ≤ t ≤ −1
⎪
y(t) = ⎨12, −1 ≤ t ≤ 5
⎪3(9 − t)2 ∕4, 5≤t≤9
⎪
⎩0, t≥9
Telegram: @Persian_IEEEComSoc

This output is plotted in Figure 3.27 along with the input for ease of comparison. We see that the system has
changed the input signal in several ways: (i) scaling: the amplitude of y(t) is six times that of x(t); (ii) delay: y(t)
reaches its peak 4 s after x(t) reaches peak; (iii) smoothing: a rectangular input waveform x(t) with abrupt transi-
tions to/from peak is smoothed into an output y(t) having a gradual transition to/from peak; (iv) spreading: the
duration of y(t) is 4 s longer than that of x(t). Note that only the last two changes constitute a distortion.

Worked Example 3.9 Direct Evaluation of Convolution Integral

A system has the impulse response h(t) = e−𝛽t u(t), 𝛽 > 0. Determine the output y(t) of this system when the input
is the sinusoidal signal x(t) = cos(𝜔t). Discuss in what ways the input signal x(t) has been modified by this system
and identify the type of system.
 3.6 Linear Time Invariant System Analysis 189

This problem requires evaluation of Eq. (3.140) with x(t) = cos(𝜔t) and h(t) = e−𝛽t u(t), and hence x(𝜏) = cos(𝜔𝜏)
and h(t − 𝜏) = e−𝛽(t − 𝜏) u(t − 𝜏)
∞
y(t) = cos(𝜔𝜏)e−𝛽(t−𝜏) u(t − 𝜏)d𝜏
∫−∞
But
{
1, 𝜏≤t
u(t − 𝜏) =
0, 𝜏>t

so the integral simplifies to

t
y(t) = e−𝛽t cos(𝜔𝜏)e𝛽𝜏 d𝜏
∫−∞

To avoid needing to use integration by parts, let us invoke Euler’s formula

ej𝜔𝜏 = cos(𝜔𝜏) + j sin(𝜔𝜏), and hence

1
cos(𝜔𝜏) = (ej𝜔𝜏 + e−j𝜔𝜏 )
2
1 1
sin(𝜔𝜏) = (ej𝜔𝜏 − e−j𝜔𝜏 ) = − j(ej𝜔𝜏 − e−j𝜔𝜏 )
2j 2
to transform the integrand into exponentials that are readily integrated
t
1 −𝛽t
y(t) = e [e(𝛽+j𝜔)𝜏 + e(𝛽−j𝜔)𝜏 ]d𝜏
2 ∫−∞
[ ]t [ (𝛽+j𝜔)t ]
1 −𝛽t e(𝛽+j𝜔)𝜏 e(𝛽−j𝜔)𝜏 || 1 e e(𝛽−j𝜔)t
= e + | = e−𝛽t + −0
2 𝛽 + j𝜔 𝛽 − j𝜔 ||−∞ 2 𝛽 + j𝜔 𝛽 − j𝜔
[ j𝜔t ] [ j𝜔t ( ) ( )]
1 e e−j𝜔t 1 e 𝛽 − j𝜔 e−j𝜔t 𝛽 + j𝜔
= + = × + ×
2 𝛽 + j𝜔 𝛽 − j𝜔 2 𝛽 + j𝜔 𝛽 − j𝜔 𝛽 − j𝜔 𝛽 + j𝜔
[ ]
1 1 j𝜔t 1 j𝜔t
= 𝛽 (e + e ) − 𝜔 j(e − e )
−j𝜔t −j𝜔t
𝛽 2 + 𝜔2 2 2
1
= [𝛽 cos(𝜔t) + 𝜔 sin(𝜔t)]
𝛽 2 + 𝜔2
Combining the cosine and sine terms as learnt in Section 2.7.3 finally yields
1 √ 2
y(t) = 2 𝛽 + 𝜔2 cos(𝜔t − tan−1 (𝜔∕𝛽))
𝛽 + 𝜔2
Telegram: @Persian_IEEEComSoc

1
= √ cos(𝜔t − tan−1 (𝜔∕𝛽))
𝛽 + 𝜔2
2

We see that the response of this system to a sinusoidal input is also a sinusoidal signal of the same frequency but
of different amplitude and phase. This describes the behaviour of LTI systems in general towards an input signal
that comprises one or more sinusoidal signals. The output frequencies will always be the same as input frequencies
and the only changes effected by the system on the signal will be some modification of input amplitudes through
the system’s gain response and some phase shift through the system’s phase response. We will learn more about
this in the next chapter. In this case, with input and output amplitudes given, respectively, by
1
Ain = 1, Aout = √
𝛽2 + 𝜔2
 190 3 Time Domain Analysis of Signals and Systems

the attenuation L (in dB) of the system is

( )
Ain
L = 20log10
Aout
√
= 20log10 ( 𝛽 2 + 𝜔2 )
= 10log10 (𝛽 2 + 𝜔2 ) = 10log10 [𝛽 2 (1 + 𝜔2 ∕𝛽 2 )]
= 20log10 𝛽 + 10log10 (1 + 𝜔2 ∕𝛽 2 )

This attenuation increases with (angular) frequency 𝜔 from a minimum Lmin = 20log10 𝛽 dB at DC (𝜔 = 0).
The system is therefore a lowpass filter, and its cut-off frequency f 1 (or 3 dB bandwidth B3dB ) is defined as the
frequency in hertz (Hz) at which its attenuation is 3 dB higher than Lmin . Clearly, this is the frequency at which
10log10 (1 + 𝜔2 /𝛽 2 ) = 3 dB, which means 𝜔2 /𝛽 2 = 1, so 𝜔 = 𝛽. Hence (since frequency is 𝜔/2𝜋)
𝛽
f1 = B3dB =
2𝜋

3.6.3 Evaluation of Convolution Sum

We will discuss the evaluation of convolutions sums by learning how to determine the output of two classes of
systems, namely recursive and non-recursive discrete-time (DT) systems.
A non-recursive DT system is one whose output y(n) depends only on the present and past values of the input
sequence x[n]. Thus, the output of this system is given by the linear difference equation
∑
N
y(n) = ak x(n − k) (nonrecursive DT system) (3.141)
k=0

Comparing this equation with Eq. (3.137) for the output of a DT LTI system having impulse response h[n]
∑
∞
∑
N
y(n) = h[n] ∗ x[n] = h(k)x(n − k) ≡ ak x(n − k)
k=−∞ k=0

we see that h(k) = ak in the above line, which means that the impulse response h[n] of the system corresponds to
the multiplier coefficients ak in Eq. (3.141). That is

h[n] = {a0 , a1 , a2 , · · · , aN } (3.142)

Figure 3.28 shows an illustrative sketch of this impulse response along with a block diagram of the system, where
the delay operation 𝔻[x(n), 1] that delays its input by one sampling interval is now denoted in short form simply
as 𝔻. The convention used to represent information in the block diagram is as follows:
Telegram: @Persian_IEEEComSoc

● The direction of signal flow along a branch is indicated by an arrow.

● Branches leading into the delay elements are labelled with their signal values. For example, the input of the first
delay element is x(n); the input of the second delay element is x(n − 1), and so on.
∑
● Branches leading into the summing device are labelled with a branch transmittance or coefficient that spec-
ifies a multiplication factor by which the branch modifies a signal flowing through it. For example, moving left
to right, the first branch feeds signal x(n)h(0) into the summer, the second branch feeds x(n − 1)h(1) into the
summer, and so on.

Equation (3.141) and Figure 3.28 represent an N th order finite impulse response (FIR) system, also described as
an FIR filter or processor, since its impulse response is a sequence of finite length N + 1, where N is the number
of delay elements in the processor.
 3.6 Linear Time Invariant System Analysis 191

h(n)
a3
a0
aN
(a) a1
a2

n
–2 –1 0 1 2 3 N N+1 N+2

x(n) x(n – 1) x(n – 2) x(n – 3) x(n – N + 1) x(n – N)

D D D D

h(0) h(1) h(2) h(3) h(N)

(b)

Σ h(k) = ak
y(n)

Figure 3.28 Non-recursive DT system or finite impulse response (FIR) filter: (a) Impulse response h(n), and (b) block
diagram of Nth -order FIR filter.

An alternative and straightforward way to obtain the impulse response h[n] of an FIR system from its linear
difference equation is by noting that h(n) is the output when the input x[n] is the unit impulse sequence 𝛿[n].
Thus, replacing x[n] by 𝛿[n] in Eq. (3.141) yields
∑
N
h(n) = ak 𝛿(n − k)
k=0
= a0 𝛿(n) + a1 𝛿(n − 1) + a2 𝛿(n − 2) + a3 𝛿(n − 3) + … + aN 𝛿(n − N)
⎧a0 , n=0
⎪
⎪a 1 , n=1
⎪a , n=2
=⎨ 2
⎪⋮
⎪a N , n=N
⎪0, n > N, or n < 0
⎩
An FIR system is guaranteed always to be stable since a necessary and sufficient condition for the stability of an
Telegram: @Persian_IEEEComSoc

LTI system is that

∞
|h(t)|dt < ∞, Stable CT LTI System
∫−∞
∑∞
|h(n)| < ∞, Stable DT LTI System (3.143)
n=−∞

and in this case

∑
∞
∑
N
|h(n)| = |an | < ∞
n=−∞ n=0

provided all coefficients ak are finite, which will be the case if the system is realisable.
 192 3 Time Domain Analysis of Signals and Systems

A recursive DT system is one whose output y(n) depends on past and present values of both the input x[n] and
output y[n]. A recursive implementation always involves feedback, and in general may depend on the present
input, N past values of the input, and M past values of the output. A general form of the linear difference equation
of a recursive DT system is therefore
∑
N
∑
M
y(n) = bk x(n − k) − ak y(n − k) (3.144)
k=0 k=1

Obviously, it is essential that at least one of the ak coefficients is nonzero for the system to be recursive (i.e.
for the present output to depend on past output values). When that is the case, the impulse response will be an
unending (i.e. infinite) sequence. For this reason, a recursive DT system is called an infinite impulse response (IIR)
filter. The stability of an IIR filter is not guaranteed but depends on the values of the coefficients being carefully
chosen in order for the resulting impulse response h[n] to satisfy Eq. (3.143).

Worked Example 3.10 Convolution Sum for a FIR System

We wish to determine the response of a causal FIR system to the input sequence x[n] shown in Figure 3.29a and
defined by
⎧4, n = −2
⎪
⎪8, n = −1
⎪12, n=0
x(n) = ⎨ (3.145)
⎪8, n=1
⎪4, n=2
⎪0, Otherwise
⎩
The impulse response h[n] of the system, shown in Figure 3.29b, is defined by
⎧10, n=0
⎪6, n=1
⎪
h(n) = ⎨3, n=2 (3.146)
⎪2, n=3
⎪
⎩0, Otherwise
The output of this FIR system is obtained by evaluating the convolution sum specified by Eq. (3.139)
⎧∑k2
⎪ x(k)h(n − k), n = n1 + k1 , · · · , →, · · · , n2 + k2
y(n) = ⎨k=k
⎪ 1

⎩0, Otherwise
Telegram: @Persian_IEEEComSoc

noting that the input sequence x[k] is zero outside the interval k = (−2, 2), so k1 = −2 and k2 = 2; and the impulse
response sequence h[n] is zero outside the interval n = (0, 3), so n1 = 0 and n2 = 3. Therefore y(n) is nonzero only
within the interval n = (n1 + k1 , n2 + k2 ) = (−2, 5). This means that we only need to compute y(−2), y(−1), y(0),
y(1), y(2), y(3), y(4), and y(5), the other output samples being zero.
What the above convolution sum operation means is that the output sample y(n) is the sum of the product of
corresponding elements of the sequences x[k] and h[n − k]. Figure 3.29c–g should provide complete clarity about
this important operation, so let us consider them one by one.
The sequence h[−k] is shown in Figure 3.29c, obtained by time-reversing the sequence h[k] in Figure 3.29b.
The sequence h[n − k] for n = −3 is required to compute y(−3) and is shown in the bottom half of Figure 3.29d,
with x[k] shown in the top half for ease of matching corresponding elements. We see that at least one element in
every pair of corresponding elements is zero. For example, at k = −6, h(n − k) = 2 but x(k) = 0; at k = 1, x(k) = 8
 3.6 Linear Time Invariant System Analysis 193

x(k)
12 h(k) h(–k)
10
8
(a) (b) 6 (c)
4 3 2
k k k
–4 –3 –2 –1 0 1 2 3 4 –1 0 1 2 3 4 –4 –3 –2 –1 0 1
x(k)
12
8
4
(d) 0 k
10 h(n – k), n = –3
6
3
0 k
–8 –7 –6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6 7 8

x(k)
12
8
4
(e) 0 k
10 h(n – k), n = 6
6
3
0 k
–8 –7 –6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6 7 8

x(k)
12
8
4
0 k
(f) h(n – k), n = 0
10
6
3
0 k
–8 –7 –6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6 7 8

x(k)
12
8
4
(g) 0 k
h(n – k), n = 2
10
6
Telegram: @Persian_IEEEComSoc

3
0 k
–8 –7 –6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6 7 8

Figure 3.29 Worked Example 3.10.

but h(n − k) = 0; and at k = 4, both x(k) and h(n − k) are zero. Thus, every product of corresponding elements is
zero and hence y(−3) = 0. From Figure 3.29d it is easy to see that this will be the outcome for all n ≤ −3. That is,
y(n) = 0 for n ≤ −3.
Figure 3.29e shows h[n − k] for n = 6 in the bottom half and x[k] in the top half, from which we see that there is
no overlap between the two sequences when n ≥ 6. That is, the product of every pair of corresponding elements is
zero. Thus, y(n) = 0 for n ≥ 6.
 194 3 Time Domain Analysis of Signals and Systems

Figure 3.29f shows h[n − k] for n = 0 in the bottom half and x[k] in the top half. We see that there is some overlap
of the two sequences, which yields

y(0) = (4 × 3) + (8 × 6) + (12 × 10) = 180

Figure 3.29g shows h[n − k] for n = 2 in the bottom half and x[k] in the top half. Multiplying corresponding
elements within the overlap region and summing these products, we obtain

y(2) = (8 × 2) + (12 × 3) + (8 × 6) + (4 × 10) = 140

Carrying on in this way, we compute y(−2), y(−1), …, y(5) to obtain the output sequence y[n] with the complete
set of values

⎧40, n = −2
⎪104, n = −1
⎪
⎪180, n=0
⎪184, n=1
⎪
y(n) = ⎨140, n=2
⎪72, n=3
⎪
⎪28, n=4
⎪8, n=5
⎪
⎩0, Otherwise

Convolution sum evaluation involves summing the sequence x(k)h[n − k] from k = k1 to k = k2 to obtain y(n).
This operation may be conveniently and more compactly carried out using a tabular layout in which the first
row contains the elements x(k1 )h[n − k1 ], the second row is x(k1 + 1)h[n − (k1 + 1)], and so on to the last row
x(k2 )h[n − k2 ]. The first column of the table is for n = n1 + k1 , the second column for n = n1 + k1 + 1, and so on to
the last column n = n2 + k2 . The rows are therefore

x(k1 ){h(n1 + k1 − (k1 )), · · · , h(n2 + k2 − (k1 ))}

x(k1 + 1){h(n1 + k1 − (k1 + 1)), · · · , h(n2 + k2 − (k1 + 1))}
⋮
x(k2 ){h(n1 + k1 − (k2 )), · · · , h(n2 + k2 − (k2 ))}

and the output y(n) is obtained as the sum of each column n, for n = n1 + k1 to n = n2 + k2 . In this example, with
k1 = −2, k2 = 2, n1 = 0, n2 = 3, making use of the specifications of x(n) and h(n) in Eqs. (3.145) and (3.146), the
Telegram: @Persian_IEEEComSoc

first row is

x(−2){h(0), · · · , h(7)} = 4 × {10, 6, 3, 2, 0, 0, 0, 0}

the second row is

x(−1){h(−1), · · · , h(6)} = 8 × {0, 10, 6, 3, 2, 0, 0, 0}

and so on until the last row k2 , which is

x(2){h(−4), · · · , h(3)} = 4 × {0, 0, 0, 0, 10, 6, 3, 2}

 3.6 Linear Time Invariant System Analysis 195

The tabular layout is shown below.

n→ −2 −1 0 1 2 3 4 5

x(−2)h[n − (−2)] 4 × 10 4×6 4×3 4×2 4×0 4×0 4×0 4×0

x(−1)h[n − (−1)] 8×0 8 × 10 8×6 8×3 8×2 8×0 8×0 8×0
x(0)h[n] 12 × 0 12 × 0 12 × 10 12 × 6 12 × 3 12 × 2 12 × 0 12 × 0
x(1)h[n − 1] 8×0 8×0 8×0 8 × 10 8×6 8×3 8×2 8×0
x(2)h[n − 2] 4×0 4×0 4×0 4×0 4 × 10 4×6 4×3 4×2
y(n) 40 104 180 184 140 72 28 8

Note that the last row y[n] is the sum of the products in each column.

Worked Example 3.11 IIR System

(a) Determine the impulse response of a causal first-order recursive system governed by the linear difference
equation

y(n) = x(n) + 𝛼y(n − 1)

(b) Hence determine the output y(n) of this system at instant n = 8 when the input is the step sequence 5u[n]
and 𝛼 = 0.85. Comment on the stability of this system.

(a) Impulse response h(n) is the output y(n) when the input x(n) is the unit impulse 𝛿(n). Substituting in the given
linear difference equation yields
h(n) = 𝛿(n) + 𝛼h(n − 1)
Let us write out the output sequence starting at n = −2, recalling that h(n) = 0 for n < 0, since the system is
causal; and 𝛿(k) = 0 for k ≠ 0, 𝛿(k) = 1 for k = 0
For n = −2 ∶
h(−2) = 𝛿(−2) + 𝛼h(−2 − 1) = 0 + 𝛼 × h(−3) = 0
For n = −1 ∶
h(−1) = 𝛿(−1) + 𝛼h(−1 − 1) = 0 + 𝛼 × h(−2) = 0
For n = 0 ∶
h(0) = 𝛿(0) + 𝛼h(0 − 1) = 1 + 𝛼 × h(−1) = 1
For n = 1 ∶
Telegram: @Persian_IEEEComSoc

h(1) = 𝛿(1) + 𝛼h(1 − 1) = 0 + 𝛼 × h(0) = 𝛼 × 1 = 𝛼

For n = 2 ∶
h(2) = 𝛿(2) + 𝛼h(2 − 1) = 0 + 𝛼 × h(1) = 𝛼 × 𝛼 = 𝛼 2
For n = 3 ∶
h(3) = 𝛿(3) + 𝛼h(3 − 1) = 0 + 𝛼 × h(2) = 𝛼 × 𝛼 2 = 𝛼 3
⋮
Therefore, the system has impulse response given by
h(n) = 𝛼 n u(n) (3.147)
which is clearly an unending or IIR.
 196 3 Time Domain Analysis of Signals and Systems

(b) The output of this system is obtained by evaluating the convolution sum

∑
∞
y(n) = x(k)h(n − k)
k=−∞

with h(n − k) obtained by substituting n − k for n in Eq. (3.147)

{ n−k
𝛼 , k≤n
h(n − k) = 𝛼 u(n − k) =
n−k
0, Otherwise
Furthermore
{
5, k≥0
x(k) = 5u(k) =
0, Otherwise
Substituting these expressions and constraints into the summation yields
∑
n
∑
n
∑
n
y(n) = 5 𝛼 n−k = 5 𝛼 n ∕𝛼 k = 5𝛼 n (1∕𝛼)k
k=0 k=0 k=0
= 5𝛼 n [1 + (1∕𝛼) + (1∕𝛼)2 + · · · + (1∕𝛼)n ]
The term in square brackets is the sum Sn+1 of the first n + 1 terms of a geometric series having first term = 1
and constant ratio = 1/𝛼, which is given by
1 − (1∕𝛼)n+1
Sn+1 =
1 − (1∕𝛼)
Therefore
[ ]
1 − (1∕𝛼)n+1 𝛼 n − 1∕𝛼
y(n) = 5𝛼 n =5
1 − (1∕𝛼) 1 − 1∕𝛼
For n = 8, 𝛼 = 0.85, we obtain y(8) = 25.6128.
If the coefficient 𝛼 > 1, then y(n) → ∞ as n → ∞, so the output increases indefinitely, and the system is unstable.
If 𝛼 < 1, then y(n) → 5/(1 − 𝛼) as n → ∞ and the system is stable. The system is also unstable if 𝛼 = 1.

3.6.4 Autocorrelation and Convolution

We discuss the mathematical operations of autocorrelation and convolution in Sections 3.5.5 and 3.6.2 along with
their defining equations (repeated below for convenience)
∞
Autocorrelation
x(t)x(t − 𝜏)dt,
Telegram: @Persian_IEEEComSoc

Rx (𝜏) =
∫−∞ of x(t)
∞
Convolution of
x1 (t) ∗ x2 (t) = x1 (𝜏)x2 (t − 𝜏)d𝜏,
∫−∞ x1 (t) and x2 (t)
Autocorrelation is an operation on a single signal, whereas convolution involves two signals. However, the math-
ematical expressions of these two operations bear some similarity that is worth exploring to establish a relationship
which we may exploit in system analysis.
Consider the convolution of x(t) with a second signal which is a time-reversed version of x(t). As discussed in
Section 3.6.2, computing the value of x(t) ∗ x(−t) at t = 𝜏 involves the following steps:
1. Time-reverse the second signal x(−t). This yields x(t).
2. Delay the time-reversed signal by 𝜏. This yields x(t −𝜏).
 3.7 Summary 197

3. Take the area under the product of the first signal and the time-reversed and 𝜏-delayed second signal. This is
the integration of x(t)x(t − 𝜏) between the limits t = −∞ and t = ∞, which is simply the autocorrelation of x(t).

Therefore, the autocorrelation of a signal x(t) is equivalent to convolving the signal with a time-reversed version
of itself, and we may write

Rx (𝜏) = x(𝜏) ∗ x(−𝜏) (3.148)

We return to this relationship between autocorrelation and convolution in the next chapter after learning about
FTs. By taking the FT of both sides of the above equation, we will obtain a very important result that states that
the FT of the autocorrelation function of a signal is the spectral density of the signal.

3.7 Summary

Telecommunication is primarily concerned with the processing of information and the transfer of information-
bearing signals from one point to another through a transmission medium. Chapters 2 and 3 have provided a
comprehensive introduction to signals and systems and their characterisation and analysis in the time domain.
The topics covered, and the approach and depth of treatment, were carefully selected to strike a delicate
balance between comprehensive rigour and succinct simplicity. The aim was to minimise needless mathemat-
ical hurdles and present material that is fresh and accessible for newcomers, insightful and informative for
everyone, and free of knowledge gaps in supporting further study and the material presented in subsequent
chapters.
Randomness, arising from an intractable interplay of deterministic laws and causes, is an ever-present phe-
nomenon in real life, and telecommunication is no exception. Whether in the form of noise as an unwanted
addition to the wanted signal, undesirable haphazard environmental interventions in a wireless transmission
medium, sequence of binary 1’s and 0’s in an information-bearing bitstream, or the behaviour and service demands
of a sizeable population of users, random signals are a constant presence in telecommunications. We discussed the
most important aspects of the basic tools needed to characterise and analyse random signals and treated in some
detail six of the standard statistical distributions that are most frequently encountered in communication systems
design and analysis.
This chapter began with a treatment of basic signal operations and continued into a discussion of various mea-
sures of a signal’s strength and patterns of variability applicable to both deterministic and random signals. The
concepts of active and reactive power were also briefly introduced, and we explored, with the help of worked
examples and consideration of implications, various approaches to determine expressions for the power, autocor-
relation functions, and the correlation coefficient of different signals.
Telegram: @Persian_IEEEComSoc

We also discussed the analysis of LTI systems, an important class of CT and DT systems which obey the princi-
ple of superposition in addition to having characteristics that do not change with time. We characterised an LTI
system in terms of its impulse response, which is the output of the system when the input is a unit impulse. Armed
with this impulse response, the response of the system to an arbitrary input signal may be obtained through time
domain analysis by convolving the input signal with the system’s impulse response. We distinguished between
non-recursive or FIR and recursive or IIR DT systems, and learnt various graphical, mathematical, and tabu-
lar approaches for evaluating convolution integrals (in the case of CT systems) and convolution sums (for DT
systems).
In the next chapter, we turn our attention to frequency domain representation and analysis of signals and sys-
tems. We will develop the tools and sound understanding of concepts needed to exploit all the simplifications
afforded by the frequency domain in the analysis and design of communication systems.
 198 3 Time Domain Analysis of Signals and Systems

References

1 ITU-R. (2007). Recommendation P.1057-2: probability distributions relevant to radiowave propagation modelling.
https://www.itu.int/rec/R-REC-P.1057-2-200708-S/en (accessed 12th June 2019).
2 Rice, S.O. (1944). Mathematical analysis of random noise. Bell System Technical Journal 23 (3): 282–332.
3 Rice, S.O. (1945). Mathematical analysis of random noise. Bell System Technical Journal 24 (1): 46–156.
4 Schwartz, M. (2005). Mobile Wireless Communications. Cambridge: Cambridge University Press.

Questions
3.1 A 925 MHz radio signal is received via a primary path and three secondary paths of excess delays 100, 150,
and 326.67 ns and respective attenuations (relative to the primary path) 1.2, 3.5, and 4.2 dB.
(a) Determine the total excess delay, the mean delay, and the rms delay spread of the multipath transmis-
sion medium.
(b) If the power arriving via the primary path is −50 dBm, determine the total received power at the
receiver.

3.2 A 1 GHz radio signal is received via a primary path and one secondary path. The primary signal power is
−100 dBm, and the secondary path is attenuated by 3 dB and delayed by 100.25 ns relative to the primary
path. Determine the received signal power.

3.3 A random variable Θ is uniformly distributed in the interval (−𝜋, 𝜋).

(a) Determine the mean and variance of Θ.
(b) Determine the characteristic function of Θ.
(c) Use the characteristic function obtained in (b) to evaluate the mean and variance of Θ and compare
these results with those of (a).

3.4 Show that the mean E[X] (i.e. expected number of arrivals in interval D) and mean square value E[X 2 ] of
a Poisson arrival process X are as given by the expressions in Eq. (3.90).

3.5 Using the definition of energy in Eq. (3.101) or otherwise, determine the energy of the following signals:
(a) g1 (t) = Arect(t∕𝜏)
(b) g2 (t) = A cos(2𝜋ft)rect(t∕𝜏), where 𝜏 = n∕f and n is an integer ≥ 1
(c) g3 (t) = Atrian(t∕𝜏)
(d) g4 (t) = Asinc(t∕Ts )
Telegram: @Persian_IEEEComSoc

3.6 Derive an expression (in terms of amplitude A and duty cycle d) for the rms value of the raised cosine pulse
train g(t) shown in Figure Q3.6. One cycle of the pulse train is defined by
{A [ (
2𝜋
)]
𝜏 𝜏
1 + cos t , − ≤t≤
gT (t) = 2 𝜏 2 2
0, Otherwise

3.7 Derive an expression for the autocorrelation function of a unipolar random binary waveform in which a
rectangular pulse of duration T b is sent with amplitude +A for binary 1 and amplitude 0 for binary 0, and
binary 1’s and 0’s are equally likely to occur. (HINT: follow the same steps as in Worked Example 3.6c with
only one change: use a voltage level of 0 V for binary 0, rather than a voltage level of −A used in that worked
example.)
 Questions 199

g(t)
d = τ/T

t
–T/2 –τ/2 0 τ/2 T/2

Figure Q3.6 Raised cosine pulse train of amplitude A and duty cycle d.

3.8 Derive expressions for the autocorrelation function of the following signals:
(a) g1 (t) = Arect(t∕𝜏)
(b) g2 (t) = A cos(2𝜋ft)rect(t∕𝜏), where 𝜏 = n∕f and n is an integer ≥ 1
(c) g3 (t) = Atrian(t∕𝜏)
(d) g4 (t) = Asinc(t∕Ts )

3.9 The alternate mark inversion (AMI) line code represents binary 1 using alternate-polarity half-width rect-
angular pulses of amplitude A, and binary 0 with no pulse (i.e. a pulse of zero amplitude). A half-width
pulse is a pulse of amplitude A and width T b /2 at the start of the bit interval followed by no pulse (i.e.
zero amplitude) for the remaining half of the bit interval, where T b is bit duration. If binary 1’s and 0’s are
equally likely, show that the autocorrelation function of the AMI line code (random waveform) is
( ) ( )
A2 𝜏 A2 𝜏 ± Tb
RAMI (𝜏) = trian − trian
4 Tb 8 Tb

3.10 The expectation of the product of two random signals X and Y is defined as
∞ ∞
E[XY ] = ∫−∞ ∫−∞ xypX,Y (x, y)dxdy
where pX,Y (x, y) is the joint PDF of X and Y .
Show that if X and Y are independent then

E[XY ] = E[X]E[Y ]

3.11 Consider a random signal X which is uniformly distributed in the interval (−a, a), where a is some positive
real constant. Let us define another random signal Y = 2X 2 . Clearly, X and Y are not independent. Show
that X and Y are uncorrelated, which proves that two uncorrelated signals are not necessarily independent.

3.12 Show that the Pearson correlation coefficient is invariant under changes to one or both signals by a mul-
Telegram: @Persian_IEEEComSoc

tiplicative constant factor or additive constant term. (Note: this requires you to show that the Pearson
correlation coefficient of signals X and Y , denoted r X,Y , is the same as the Pearson correlation coefficient
of a + bX and c + dY , denoted r a+bX, c+dY , where a, b, c, d are constants and b > 0, d > 0.)

3.13 M-ary amplitude shift keying (M-ASK) employs the following set of symbols for transmission
( )
t − 12 Ts
gk (t) = kArect cos(2𝜋fc t)
Ts
k = 0, 1, 2, 3, · · · , M − 1; fc = n∕Ts

where A is some constant and n > 0 is a positive integer.

 200 3 Time Domain Analysis of Signals and Systems

Derive an expression for the correlation coefficient of the largest-amplitude adjacent symbols gM−1 (t) and
gM−2 (t) and examine how this coefficient varies with M. [Note that M is always an integer power of 2, so it
may take on only the values M = 2, 4, 8, 16, 32, 64, …].

3.14 M-ary frequency shift keying (M-FSK) employs the following set of symbols for transmission
( )
t − 12 Ts
gk (t) = Arect cos[2𝜋(k + 1)(fc ∕2)t]
Ts
k = 0, 1, 2, 3, · · · , M − 1; fc = n∕Ts
where A is some constant and n > 0 is a positive integer.
Show that every pair of symbols gm (t) and gp (t), m ≠ p, in the set is orthogonal over a symbol duration in
the interval (0, T s ) and hence has the correlation coefficient 𝜌gm (t), gp (t) = 0.

3.15 A causal CT system has impulse response

( )
t 1
h(t) = 5rect −
6 2
where t is time in seconds.
Use a graphical convolution approach to determine and plot a smooth waveform of the output of the system
)
for each of the (following input signals
t
(a) x(t) = rect
12( )
t 1
(b) x(t) = 10trian −
8 4
(NOTE: you may wish to refer to Sections 2.6.3 and 2.6.5 for definitions of the rect() and trian() pulses used
above.)

3.16 The impulse response h[n] of an LTI discrete-time system is shown in Figure Q3.16a.
(a) Determine and fully sketch the output sequence y[n] of this system when the input x[n] is the
discrete-time signal shown in Figure 3.31b.

h(n)
12
10

(a) 8
6
4
Telegram: @Persian_IEEEComSoc

2
n
–3 –2 –1 0 1 2 3 4 5 6 7 8

x(n)
5
(b)

n
–6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6

Figure Q3.16 Impulse response h[n] and input sequence x[n] for Question 3.16.
 Questions 201

(b) Based on your result in (a), discuss all the effects of this system on a signal passed through it, and
identify what type of filter the system might be.

3.17 If the input sequence of Figure 3.25a is transmitted through a finite impulse response (FIR) filter with
impulse response as shown in Figure 3.25b, determine and fully sketch the resulting output sequence y[n]
of the system.

3.18 Calculate the energy or power (depending on type) of each of the following signals.
(a) Impulse function 𝛿(t)
(b) Sinc function sinc(t)
(c) Unit step function u(t)
(d) Signum function sgn(t)
(e) Complex exponential function exp(j𝜔t).

3.19 Obtain expressions for the

(a) average value,
(b) rms value, and
(c) power
of a full-wave rectified sinusoidal signal g(t) = |Acos(2𝜋ft + 𝜙)|. Your formulas should be applicable at all
values of initial phase 𝜙 and not only at 𝜙 = −90∘ .
How do your expressions compare with those for an unrectified sinusoidal signal Acos(2𝜋ft + 𝜙)?
Telegram: @Persian_IEEEComSoc
 203

Frequency Domain Analysis of Signals and Systems

There is a smart way and a hard way in every engineering problem-solving.

The fun is in doing it smart.

In this Chapter
✓ Fourier series: a step-by-step treatment of the frequency domain representation of periodic signals with
application to the analysis of important telecom signals, including sinusoidal pulse trains, binary amplitude
shift keying, staircase waveforms, flat-top sampling, and trapezoidal pulse trains.
✓ Fourier transform (FT): frequency domain analysis of nonperiodic waveforms and energy signals.
✓ Discrete Fourier transform (DFT): an extension of frequency domain analysis techniques to discrete-time
(DT) signals and systems, along with richly illustrated discussions of practical issues such as fast Fourier
transform (FFT), frequency resolution, spectral leakage, spectral smearing, periodograms, etc.
✓ Laplace transform (LT) and z-transform (ZT): a brief introduction to give you a complete picture of frequency
domain tools and their relationship with other transform techniques.
✓ Frequency domain characterisation and analysis of linear systems.
✓ Worked examples: a wide selection to further deepen your insight and help you master problem-solving
application and interpretation of important concepts.
✓ End-of-chapter questions: an essential test of your grasp of the subject matter.

4.1 Introduction
Telegram: @Persian_IEEEComSoc

Telecommunication signals and systems can be characterised in both the time and the frequency domains. Time
domain description is presented in the previous two chapters. In the frequency domain we specify the frequency,
phase, and relative amplitude of each of the sinusoidal components that constitute the signal, either through a
spectral plot as displayed on a spectrum analyser or through a mathematical expression giving the Fourier series
or FT of the signal. From these the bandwidth, spectral density, and other spectral characteristics of the signal can
be determined, as well as signal energy or power. For a system, we specify its gain response and phase response,
which may be combined into a single complex quantity known as the transfer function of the system. This func-
tion describes how the system scales the amplitude and shifts the phase of a sinusoidal signal that is transmitted
through it. The transfer function also leads to a specification of the bandwidth of the system.
We hope that it is your ambition to have a sound understanding of the principles of communication engineering
and to establish a robust foundation and the confidence necessary for a successful and pleasurable engagement
Communication Engineering Principles, Second Edition. Ifiok Otung.
© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 204 4 Frequency Domain Analysis of Signals and Systems

with telecoms, either in your career or in further studies and research. In that case, we recognise that this chapter
is pivotal for you. So, we have taken great care to present the material in a step-by-step manner, using appropriate
worked examples at every stage to help you see how each key concept is interpreted in the right context and applied
to engineering problem solving. It is strongly recommended that you give yourself plenty of time and perhaps set
yourself small section-by-section targets as a motivation to work diligently and carefully through the entire chapter.
It is much better to take a little more time to reach a place of excellence and confidence than to rush through this
chapter and risk coming away with hazy information or, worse, a misplaced confidence in your knowledge and
ability.
We start our discussion with the Fourier series applicable to continuous-time (CT) periodic signals. Using a mix
of heuristic, graphical, and mathematical approaches, we explore the topic of Fourier series at a depth and breadth
that are considered complete for the needs of modern engineering. We emphasise how to derive the Fourier
series either from first principles by evaluating integrals or by applying some of its properties. We develop in
full the relationships between the sinusoidal and complex exponential forms of the Fourier series and link these
to double-sided and single-sided amplitude and phase spectra. We then employ the tool of Fourier series to anal-
yse flat-top sampling, sinusoidal pulse trains, and binary amplitude shift keying (BASK) with very interesting and
insightful results. We also derive Fourier series expressions for the most general specification of a staircase wave-
form and for the trapezoidal pulse train (which reduces to rectangular, triangular, sawtooth, and ramp pulse trains
as special cases).
We then derive the FT applicable to continuous-time nonperiodic signals as a limiting case of the Fourier series
when signal period T tends to infinity. We learn how to calculate the FT of a signal from first principles, discuss
the properties of the FT and how they are leveraged in problem solving, and present a tabulated list of standard
FT pairs for reference purposes. To illustrate the relevance and power of the FT as a system analysis and design
tool, we undertake the evaluation of digital transmission pulses from both time domain and frequency domain
perspectives and come away with interesting findings on system capacity implications.
Today’s world is increasingly driven by digital technology and dominated by discrete signals. So, it is only a
logical step for us to extend the powerful tools of Fourier series and FT to be applicable to discrete-time (DT)
signals in the respective forms of discrete-time Fourier series (DTFS) and discrete-time Fourier transform (DTFT),
which we merge into one tool known simply as the discrete Fourier transform (DFT). We discuss the DFT in detail,
including the basics of an algorithm for its efficient computation known as the fast Fourier transform (FFT). We
endeavour to fully equip you to avoid common pitfalls in the use of FFT for data analysis, and to be able to correctly
select the values of parameters needed for reliable spectral analysis especially of nondeterministic signals, and to
correctly interpret results with full knowledge of their limitations.
The scope of this book precludes a detailed treatment of other transform techniques, but for the sake of com-
pleteness and, importantly, to help you see how they fit into the frequency domain picture, we briefly introduce
the Laplace transform (LT) and z-transform (ZT). The LT is presented as a generalisation of the FT. It is a versatile
tool in the design and analysis of continuous-time systems. The ZT is a generalisation of the DTFT for application
Telegram: @Persian_IEEEComSoc

in the design and analysis of DT systems, most popularly in the field of digital filters and digital signal processing
(DSP). It is hoped that readers who are involved with control systems, electrical circuits, DSP, etc., will find the
interrelationships among FT, LT, and ZT presented in this chapter both illuminating and of practical use.
We briefly outline the inverse relationships between time and frequency domains primarily to help you gain an
important early appreciation for the inherency of trade-offs in system design. The brief discussion of this issue
is also aimed at establishing how operations and forms (e.g. periodic versus nonperiodic and continuous versus
discrete) in one domain translate into the other.
Finally, we turn our attention to using FT tools to characterise and analyse linear systems. The main parameter of
such a system is its transfer function, which we derive and link to the impulse response of the previous chapter. We
attempt to bring clarity to the oft-misused terminology of bandwidth. And we rely on the transfer function param-
eter as we explore the concepts of distortionless transmission, equalisation, amplitude and phase distortions, and
 4.2 Fourier Series 205

the calculation of the output signal and output spectral density of linear systems. For completeness, we also briefly
discuss and demonstrate the harmonic and intermodulation distortions that arise due to nonlinear operation.

4.2 Fourier Series

We learnt in connection with Figure 2.26 that adding together harmonically related sinusoids yields a periodic
signal. By appropriately selecting the amplitude, frequency, and phase of such sinusoids, any arbitrary periodic
signal can be realised as the sum of these harmonically related sinusoids. This is in fact the Fourier theorem, which
states:

Any periodic function or waveform g(t) having period T can be expressed as the sum of sinusoidal signals
with frequencies at integer multiples (called harmonics) of the fundamental frequency f o = 1/T, and with
appropriate amplitudes and phases.

Figure 4.1 illustrates the realisation of an arbitrary periodic staircase waveform g(t) by adding together its har-
monic sinusoids. This process is known as Fourier synthesis. In Figure 4.1a the synthesised waveform gs (t) is the
sum of a DC component of value 1.05, a first harmonic – a sinusoid of amplitude 2.83 volts (V), frequency f o ,
and phase −98∘ , and a second harmonic – a sinusoid of amplitude 0.62 V, frequency 2f o , and phase 47.4∘ , where
f o = 1/T, and T is the period of g(t) as indicated on the figure. That is
g (t) = 1.05 + 2.83 cos(2𝜋f t − 98∘ ) + 0.62 cos(4𝜋f t + 47.4∘ )
s o o

Using the techniques learnt in Section 3.5, the normalised powers of g(t) and gs (t) are readily calculated as
5.65 and 5.3 W respectively, which means that the DC and first two harmonics of this particular waveform g(t)

gs(t) = 1.05 + 2.83 cos(2π fot – 98˚) + 0.62 cos(4π fot + 47.4˚); fo = 1/T
4V
g(t)
DC +
two harmonics
(a) 0 t

–2V
T
4V
DC +
15 harmonics
(b) 0 t
Telegram: @Persian_IEEEComSoc

–2V
4V
DC +
80 harmonics

(c) 0 t

–2V

Figure 4.1 Fourier synthesis of periodic staircase waveform g(t) using (a) DC + first two harmonics; (b) DC + first 15
harmonics; and (c) DC + first 80 harmonics.
 206 4 Frequency Domain Analysis of Signals and Systems

contain 93.83% of its power. The approximation of g(t) by gs (t) is improved as more harmonic sinusoids of the
right amplitude and phase are added to gs (t). This is illustrated in Figure 4.1b in which gs (t) consists of the DC and
the first 15 harmonics, making up 98.53% of the power. In Figure 4.1c the DC and first 80 harmonics are summed
to produce an even closer approximation to g(t) containing 99.71% of the power. Notice that ripples have been
markedly reduced in the synthesised waveform from Figure 4.1b–c. However, overshoots will persist at points of a
jump discontinuity (where, for example, an ideal rectangular pulse train (RPT) changes level instantaneously from
one value to another). This problem is known as Gibbs phenomenon, which refers to the oscillation of a Fourier
synthesised waveform near a jump, with an overshoot that does not die out but approaches a finite limit as more
and more harmonics are added.
One lesson to draw from this brief exercise in Fourier synthesis is that a waveform transmitted through a trans-
mission medium or system will be noticeably distorted unless the system has enough bandwidth to pass all signif-
icant frequency components of the waveform. We return to this important thought later in the chapter. It should
be pointed out that the summation of harmonic sinusoids as stipulated in the above Fourier theorem is only guar-
anteed to converge (i.e. to approximate the function g(t) better and better as more harmonics are added) if, over
one cycle, the function (i) is absolutely integrable, (ii) is of bounded variation, and (iii) has a finite number of finite
discontinuities. These three conditions are known as Dirichlet conditions and are satisfied by the telecommunica-
tion signals that we will deal with, so this caveat is more for information than for any serious practical caution.
We may therefore proceed with confidence to the task of the Fourier analysis of periodic waveforms to determine
the amplitudes and phases of their harmonic components.

4.2.1 Sinusoidal Form of Fourier Series

The Fourier theorem stipulates that a periodic signal g(t) may be expanded as a Fourier series of the form
∑
∞
∑
∞
g(t) = A0 + an cos(2𝜋nf o t) + bn sin(2𝜋nf o t) (4.1)
n=1 n=1

where

● f o = 1/T is the fundamental frequency, T being the period of g(t).

● nf o is the nth harmonic frequency; n = 1, 2, 3, …
● A0 is the DC or average value of g(t).
● an is the cosine coefficient and bn the sine coefficient.

The amplitude and phase of the nth harmonic of g(t) are not obvious in Eq. (4.1) because in this form of the
Fourier series the harmonic at each frequency is split into a cosine component an cos(2𝜋nf o t) and a sine component
bn sin(2𝜋nf o t). We know from Section 2.7.3 that we can combine these two components (since they have the same
Telegram: @Persian_IEEEComSoc

frequency nf o but differ in phase by 90∘ ) into a single resultant sinusoid An cos(2𝜋nf o t + 𝜙n ). The Fourier series
of g(t) therefore takes the more compact and informative form
∑
∞
g(t) = A0 + An cos(2𝜋nf o t + 𝜙n ) (4.2)
n=1

where An and 𝜙n are, respectively, the amplitude and phase of the nth harmonic component of g(t). Figure 4.2
shows the geometry and expression for An and 𝜙n in terms of an and bn for all combinations of the magnitude and
sign of the coefficients an and bn . We see that to determine the phase 𝜙n we first compute an acute angle 𝛼 using
the absolute values of an and bn as

𝛼 = tan−1 (|bn |∕|an |) (4.3)

 4.2 Fourier Series 207

Condition: an > 0, bn > 0 an > 0, bn < 0 an < 0, bn > 0 an < 0, bn < 0

an an
ϕn ϕn bn ϕn bn ϕn
bn bn
an an

bn bn bn bn
ϕn = –tan–1 ϕn = tan–1 ϕn = tan–1 –180˚ ϕn = 180˚ –tan–1
an an an an

An = an2 + bn2

Condition: an > 0, bn = 0 an < 0, bn = 0 an = 0, bn > 0 an = 0, bn < 0

an an ϕn
bn ϕn
bn ϕn

ϕn = 0 ϕn = 180˚ ϕn = –90˚ ϕn = 90˚

An = an An = bn

cos (Reference)
Phasor relationships:

sin

Figure 4.2 Amplitude An and phase 𝜙n of nth harmonic in terms of cosine and sine coefficients an and bn .

The phase of the nth harmonic is then given by

⎧
⎪𝛼, bn ≤ 0, an ≥ 0
⎪
⎪−𝛼, bn > 0, an ≥ 0
𝜙n = ⎨ (4.4)
⎪180 − 𝛼, bn ≤ 0, an < 0
⎪
⎪𝛼 − 180, bn > 0, an < 0
⎩
And the amplitude is
√
An = a2n + b2n (4.5)
A plot of the amplitudes An of the harmonic sinusoids against their frequencies nf o is called the magnitude or
amplitude spectrum, whereas a plot of 𝜙n versus nf o is called the phase spectrum. The normalised DC power Pdc of
Telegram: @Persian_IEEEComSoc

the signal, the normalised power Pnfo in the nth harmonic of the signal (for n > 0), and the AC power Pac (which
is the total normalised power in all the harmonics of the signal, i.e. power in all the sinusoidal components of the
signal having frequency f > 0) are given by
Pdc = A20
1
Pnfo = A2n
2
1∑ 2
∞
Pac = A (4.6)
2 n=1 n
Using the last line of the above equation to determine AC power is understandably cumbersome since it involves
summing a large and theoretically infinite number of components. It is usually more straightforward to obtain the
 208 4 Frequency Domain Analysis of Signals and Systems

1st harmonic
of frequency t
fo = 1/T

2nd harmonic
of frequency 2fo t

3rd harmonic
t
of frequency 3fo

4th harmonic
t
of frequency 4fo

–T/2 –T/4 0 T/4 T/2

Figure 4.3 Total area under the waveform of a harmonic sinusoid in an interval T spanning one cycle of the fundamental
waveform (of frequency f o ) is zero.

AC power of g(t) by first calculating the total power Pt of g(t) in the time domain (from its waveform structure,
using equations such as (3.104) to (3.111)) and then determining AC power as

Pac = Pt − Pdc (4.7)

It is a trivial matter to determine the frequency of each of the harmonic sinusoids present in a periodic signal
g(t). The frequency f n of the nth harmonic is simply nf o , which is n times the reciprocal of the period of g(t)

fn = nf o = n∕T (4.8)

Note that the frequency of the first harmonic (n = 1) is the same as the fundamental frequency f o of the signal.
Determining the amplitude An and phase 𝜙n of the nth harmonic, for n = 1, 2, 3, …, in general requires first cal-
culating the cosine and sine coefficients an and bn in Eq. (4.1) and then employing Eqs. (4.4) and (4.5) or Figure 4.2
to obtain An and 𝜙n . So how do we determine these Fourier coefficients? An important observation for this task
is illustrated in Figure 4.3, namely that the area of any harmonic sinusoid in a span of one cycle of the funda-
mental waveform is zero. Figure 4.3 shows only the first four harmonic sinusoidal waveforms, but the scenario
will be similar for all other harmonics: the nth harmonic completes precisely n positive half-cycles and n negative
Telegram: @Persian_IEEEComSoc

half-cycles within the interval T, so their contributions to the total area cancel exactly.
A mathematical proof of the above observation is straightforward. The sinusoids

cos(2𝜋mf o t) and sin(2𝜋mf o t), m = 1, 2, 3, · · ·

are harmonics of fundamental frequency f o and fundamental period T = 1/f o (which means that f o T = 1). The
area of cos(2𝜋mf o t) in an interval (−T/2, T/2) spanning one fundamental cycle is given by the integral
T∕2
sin(2𝜋mf o t) |T∕2
cos(2𝜋mf o t)dt = |
∫−T∕2 2𝜋mf o ||−T∕2
sin(2𝜋mf o T∕2) − sin(2𝜋mf o (−T∕2))
=
2𝜋mf o
 4.2 Fourier Series 209

2 sin(𝜋mf o T)
=
2𝜋mf o
sin(m𝜋)
= (since fo T = 1)
𝜋mf o
=0 (since sin(m𝜋) = 0, m = 1, 2, 3, · · ·)

Similarly, the area of sin(2𝜋mf o t) in this interval is

T∕2
− cos(2𝜋mf o t) |T∕2
sin(2𝜋mf o t)dt = |
∫−T∕2 |
2𝜋mf o |−T∕2
[ ]
cos(2𝜋mf o T∕2) − cos(2𝜋mf o (−T∕2))
=−
2𝜋mf o
=0 since cos(𝜃) = cos(−𝜃)

Armed with this knowledge, we make the following observations in Eq. (4.1) repeated here for convenience with
the index n replaced by m

∑
∞
∑
∞
g(t) = A0 + am cos(2𝜋mf o t) + bm sin(2𝜋mf o t) (4.9)
m=1 m=1

Since the right-hand side consists of A0 and harmonics, taking the area of (i.e. integrating) both sides of Eq. (4.9)
over an interval of one cycle eliminates the contribution of each harmonic and leaves us with an expression
for A0 .
By trigonometric identity, multiplying cos(2𝜋mf o t) or sin(2𝜋mf o t) by cos(2𝜋nf o t) creates harmonics at frequen-
cies (m + n)f o and (m − n)f o , except when m = n. Therefore, multiplying both sides of Eq. (4.9) by cos(2𝜋nf o t)
before integrating over an interval of one cycle eliminates the contribution of every newly created harmonic and
leaves us with an expression for an .
Similarly, multiplying both sides of Eq. (4.9) by sin(2𝜋nf o t) before integrating over an interval of one cycle leads
to an expression for bn .
Thus, for the DC component A0 , integrating both sides of Eq. (4.9)
[∞ ] [∞ ]
T∕2 T∕2 T∕2 ∑ T∕2 ∑
g(t)dt = A dt + a cos(2𝜋mf o t) dt + b sin(2𝜋mf o t) dt
∫−T∕2 ∫−T∕2 0 ∫−T∕2 m=1 m ∫−T∕2 m=1 m

Interchanging the order of integration and summation on the right-hand side

[ ] ∞ [ ]
T∕2 T∕2 ∑
∞ T∕2 ∑ T∕2
Telegram: @Persian_IEEEComSoc

g(t)dt = A0 dt + am cos(2𝜋mf o t)dt + bm sin(2𝜋mf o t)dt

∫−T∕2 ∫−T∕2 m=1
∫−T∕2 m=1
∫−T∕2

The second and third terms on the right-hand side are zero, being the sums of areas of various harmonic sinu-
soidal waveforms in an interval of one fundamental cycle. Therefore
T∕2 T∕2 ( )
T∕2 T T
g(t)dt = A0 dt = A0 t|−T∕2 = A0 − A0 − = A0 T
∫−T∕2 ∫−T∕2 2 2

which yields
T∕2
1
A0 = g(t)dt (4.10)
T ∫−T∕2
 210 4 Frequency Domain Analysis of Signals and Systems

For the cosine coefficient an , we multiply Eq. (4.9) through by cos(2𝜋nf o t) before integrating
[∞ ]
T∕2 T∕2 T∕2 ∑
g(t) cos(2𝜋nf o t)dt = A cos(2𝜋nf o t)dt + a cos(2𝜋mf o t) cos(2𝜋nf o t)dt
∫−T∕2 ∫−T∕2 0 ∫−T∕2 m=1 m
[∞ ]
T∕2 ∑
+ b sin(2𝜋mf o t) cos(2𝜋nf o t)dt
∫−T∕2 m=1 m

Recognising that on the right-hand side the first term evaluates to zero, and interchanging the order of integra-
tion and summation in the second and third terms yields
T∕2 ∑
∞ T∕2
g(t) cos(2𝜋nf o t)dt = [am cos(2𝜋mf o t) cos(2𝜋nf o t) + bm sin(2𝜋mf o t) cos(2𝜋nf o t)]dt
∫−T∕2 m=1
∫−T∕2

Using trigonometric identities (Appendix B), the integrand on the right-hand side expands to
am b
[cos(2𝜋(m − n)fo t) + cos(2𝜋(m + n)fo t)] + m [sin(2𝜋(m − n)fo t) + sin(2𝜋(m + n)fo t)]
2 2
These are harmonic sinusoids which all integrate to zero over interval (−T/2, T/2), except at m = n, where the
first term is
an a a
cos(2𝜋(n − n)fo t) = n cos(0) = n
2 2 2
and the third term is zero, since
bn b
sin(2𝜋(n − n)fo t) = n sin(0) = 0
2 2
Therefore, the only surviving term in the integrand is an /2, so that
T∕2 T∕2
an T
g(t) cos(2𝜋nf o t)dt = dt = an
∫−T∕2 ∫−T∕2 2 2
which yields
T∕2
2
an = g(t) cos(2𝜋nf o t)dt (4.11)
T ∫−T∕2
Finally, for the sine coefficient bn , multiply Eq. (4.9) by sin(2𝜋nf o t) and integrate
[∞ ]
T∕2 T∕2 T∕2 ∑
g(t) sin(2𝜋nf o t)dt = A sin(2𝜋nf o t)dt + a cos(2𝜋mf o t) sin(2𝜋nf o t)dt
∫−T∕2 ∫−T∕2 0 ∫−T∕2 m=1 m
[∞ ]
Telegram: @Persian_IEEEComSoc

T∕2 ∑
+ b sin(2𝜋mf o t) sin(2𝜋nf o t)dt
∫−T∕2 m=1 m

On the right-hand side, ignore the first term (which evaluates to zero), and interchange the order of integration
and summation in the second and third terms
T∕2 ∑
∞ T∕2
g(t) sin(2𝜋nf o t)dt = [am cos(2𝜋mf o t) sin(2𝜋nf o t) + bm sin(2𝜋mf o t) sin(2𝜋nf o t)]dt
∫−T∕2 m=1
∫−T∕2

Using trigonometric identities, the integrand on the right-hand side expands to

am b
[sin(2𝜋(n − m)fo t) + sin(2𝜋(n + m)fo t)] + m [cos(2𝜋(m − n)fo t) − cos(2𝜋(m + n)fo t)].
2 2
 4.2 Fourier Series 211

These are harmonic sinusoids which all integrate to zero over interval (−T/2, T/2), except at m = n, where the
first term is zero and the third term is bn /2. Therefore, the only surviving term in the integrand is bn /2, so that
T∕2 T∕2
bn T
g(t) sin(2𝜋nf o t)dt = dt = bn
∫−T∕2 ∫−T∕2 2 2
which yields
T∕2
2
bn = g(t) sin(2𝜋nf o t)dt (4.12)
T ∫−T∕2
We have now derived expressions for all three Fourier coefficients, which reveal that the DC component A0 is
the average of g(t); the cosine coefficient an is twice the average of g(t)cos(2𝜋nf o t), and the sine coefficient bn is
twice the average of g(t)sin(2𝜋nf o t).
To summarise, any periodic signal g(t) of period T may be expressed in a sinusoidal form of the Fourier series as
∑
∞
g(t) = A0 + An cos(2𝜋nf o t + 𝜙n )
n=1
where
T∕2
1
A0 = g(t)dt
T ∫−T∕2
√
An = a2n + b2n ; n = 1, 2, 3, · · ·
⎧𝛼, bn ≤ 0, an ≥ 0
⎪ ( )
⎪−𝛼, bn > 0, an ≥ 0 |bn |
𝜙n = ⎨ ; 𝛼 = tan−1
⎪180 − 𝛼, bn ≤ 0, an < 0 |an |
⎪𝛼 − 180, bn > 0, an < 0
⎩
T∕2
2
an = g(t) cos(2𝜋nf o t)dt
T ∫−T∕2
T∕2
2
bn = g(t) sin(2𝜋nf o t)dt (4.13)
T ∫−T∕2
The Fourier series stated in full in Eq. (4.13) is not a mere mathematical indulgence. Rather, it has important
practical applications, giving a complete picture of the frequency content or spectrum and hence bandwidth of
g(t). This information is vital in the design and analysis of signal transmission systems and will be emphasised
throughout this chapter.
It is not always necessary to evaluate all the integrals in Eq. (4.13) in order to derive the Fourier series of g(t). The
Telegram: @Persian_IEEEComSoc

following properties will sometimes apply and may be exploited to significantly simplify or reduce computation.

4.2.1.1 Time Shifting

By substituting t − to for t wherever it occurs in the Fourier series equation, we obtain the Fourier series of the
signal g(t − to ) as
∑
∞
g(t − to ) = A0 + An cos(2𝜋nf o (t − to ) + 𝜙n )
n=1
∑ ∞
= A0 + An cos(2𝜋nf o t + 𝜙n − 2𝜋nf o to )
n=1
 212 4 Frequency Domain Analysis of Signals and Systems

We see that time shifting by to changes only the phase spectrum by adding a shift of −2𝜋nf o to to the phase of
the nth harmonic, but the amplitude spectrum is unchanged. Letting 𝔸x(t),n and Φx(t),n , respectively, denote the
amplitude and phase spectra of signal x(t), we may write
𝔸g(t−to ), n = 𝔸g(t), n
Φg(t−to ), n = Φg(t), n − 2𝜋n𝛼; 𝛼 = to ∕T (4.14)
where we have expressed the time shift to as a fraction 𝛼 of the waveform’s period and invoked f o T = 1.
If signal x(t) is the result of a combination of horizontal and vertical shifts of another signal whose Fourier series
is already known then this time shifting property may be exploited to obtain the Fourier series of x(t) without the
need to work from first principles involving the evaluation of the integrals in Eq. (4.13). Note that a downward
(upward) vertical shift corresponds to subtracting (adding) a DC component. For example, in Figure 4.4 the bipolar
rectangular waveform g1 (t) is the result of delaying the unipolar RPT g(t) by to = T/8 and subtracting A1 = 10 V.
So (with 𝛼 = 1/8 in this case), if the Fourier series of g(t) is known, which means that expressions for An and 𝜙n
as functions of n are known for g(t), then the Fourier series of g1 (t) may be quickly derived as
∑
∞
g1 (t) = g(t − to ) − A1 = A0 − 10 + An cos(2𝜋nf o t + 𝜙n − 2𝜋n𝛼)
n=1
∑
∞
= A0 − 10 + An cos(2𝜋nf o t + 𝜙n − n𝜋∕4)
n=1

As another example of this application, the bipolar triangular waveform g3 (t) in Figure 4.4d is derived from the
unipolar triangular pulse train g2 (t) in Figure 4.4c by setting 𝜏 = T (which implies duty cycle d = 1), subtracting a
DC component A1 , and applying a time shift to = −T/5. So, if the Fourier series of g2 (t) is known, we may obtain
the Fourier series of g3 (t) simply by setting d = 1 in the Fourier series for g2 (t), subtracting A1 , and adding 2n𝜋/5
to the phase spectrum expression.

4.2.1.2 Time Reversal

By substituting −t for t wherever it occurs in the Fourier series equation, we obtain the Fourier series of the signal
g(−t) as
∑
∞
g(−t) = A0 + An cos(2𝜋nf o (−t) + 𝜙n )
n=1
∑∞
= A0 + An cos(−(2𝜋nf o t − 𝜙n ))
n=1
∑∞
= A0 + An cos(2𝜋nf o t − 𝜙n ) since cos(𝜃) = cos(−𝜃)
n=1
Telegram: @Persian_IEEEComSoc

Therefore, the time domain operation of time reversal does not alter the amplitude spectrum, but it changes the
phase spectrum by a factor of −1. That is
𝔸g(−t), n = 𝔸g(t), n
Φg(−t), n = −Φg(t), n (4.15)

4.2.1.3 Even and Odd Functions

If g(t) is an even function, i.e. g(−t) = g(t), then the integrand g(t)sin(2𝜋nf o t) in the integral for bn is an odd
function and its area in the range (−T/2 ≤ t ≤ 0) is the same magnitude but of opposite sign as its area in the range
(0 ≤ t ≤ T/2), which leads to bn = 0. In that case, we see from Figure 4.2 that the phase spectrum will take on only
two possible values, namely 0∘ if an is positive or 180∘ if an is negative.
 4.2 Fourier Series 213

g(t), volts

(a) A = 30 V τ d = τ/T
T
t
g1(t), volts
to τ = T/4; to = T/8;
(b) A2 = 20 V τ d = τ/T = ¼
T
t
A1 = 10 V to

g2(t)
A d = τ/T
(c) T fo = 1/T

τ t
g3(t), volts
(d)

A2 = 40 V
τ1
t
τ2
A1 = 20 V
T

g4(t)
A d = τ/T
τ/2
(e) t
τ
–A
T

g5(t) dk = τk/T, k = 1, 2, 3, …
A3
A1 τ3
τ1
(f)
τ2 t
A2
T
Telegram: @Persian_IEEEComSoc

Figure 4.4 Waveform examples for Fourier Series.

If, on the other hand, g(t) is an odd function, i.e. g(−t) = −g(t), then the integrand g(t)cos(2𝜋nf o t) in the integral
for an is an odd function, and this leads to an = 0. Under this condition, the phase spectrum will also have only
two possible values, namely −90∘ if bn is positive or 90∘ if bn is negative. To summarise
{
0, an positive
bn = 0, 𝜙n = g(t) even
180∘ , an negative
{
−90∘ , bn positive
an = 0, 𝜙n = g(t) odd (4.16)
90∘ , b negative
n
 214 4 Frequency Domain Analysis of Signals and Systems

Therefore, when deriving the Fourier series of a given signal g(t) from first principles, if the function is not
already either even or odd, it is prudent to check whether a horizontal shift of g(t) by some amount to would result
in either an even or an odd function g(t − to ). If that is the case then only one coefficient (an if even or bn if odd)
needs to be calculated to obtain the Fourier series of g(t − to ), which significantly reduces computational effort.
The correct Fourier series for g(t) is then obtained by invoking the time shifting property and adding 2𝜋nf o to to
the phase of the derived Fourier series for g(t − to ).

4.2.1.4 Piecewise Linear Functions

If the waveform g(t) is piecewise linear or a combination of standard geometrical shapes (e.g. rectangular, trian-
gular, sawtooth, ramp, trapezoidal, etc.) then A0 is the mean value of g(t) obtained by dividing the net area within
one cycle of g(t) by the period T of g(t). This net area may be easily calculated from the dimensions of the shape
of the waveform without the need for any integration. Consider again the waveforms g(t), g1 (t), g2 (t), and g3 (t) in
Figure 4.4. The net area in one cycle of g(t) is A𝜏, so the DC component of g(t) is A0 = A𝜏/T = Ad, where d is the
duty cycle of the waveform (as discussed in Section 2.6).
The net area in one cycle of g1 (t) is A2 𝜏 − A1 (T − 𝜏). Dividing by T and using the values A2 = 20, A1 = 10, and
𝜏 = T/4 shown on the figure yields the DC component of g1 (t) as
A2 𝜏 − A1 (T − 𝜏) 1 3
A0 = = A2 − A1 = 5 − 7.5 = −2.5 V
T 4 4
The net area in one cycle of g2 (t) is A𝜏∕2. Dividing this by T yields A0 = Ad/2.
To determine the DC component of g3 (t), we note that 𝜏 1 + 𝜏 2 = T, and (since the upward and downward slopes
of the triangle are equal), A2 /A1 = 𝜏 2 /𝜏 1 . Dividing the net area in one cycle of g3 (t) by T yields A0 as
1 (A2 𝜏2 − A1 𝜏1 ) 1 (A2 𝜏2 − A1 𝜏1 ) 1 (A2 𝜏2 ∕𝜏1 − A1 )
A0 = = =
2 T 2 𝜏2 + 𝜏1 2 𝜏2 ∕𝜏1 + 1
2 2
1 (A2 × A2 ∕A1 − A1 ) 1 (A2 − A1 )
= =
2 A2 ∕A1 + 1 2 A2 + A1
If, for example, A2 = 40 V and A1 = 20 V, then A0 = 20 V.
The next three worked examples will apply the concepts we have discussed so far in this section and are designed
to give you important skills in Fourier analysis and synthesis. In the first worked example, we derive the Fourier
series of rectangular and triangular pulse trains from first principles. In the second, we learn how to quickly derive
the Fourier series of a signal by manipulating a known Fourier series of a related standard waveform. In the third,
we learn to interpret the Fourier series and extract some of its frequency domain information for use in Fourier
synthesis. There are many other useful Fourier series exercises in the end-of-chapter questions for you to work on
to gain confidence in your mastery of the topic.
Telegram: @Persian_IEEEComSoc

Worked Example 4.1 Derivation of Fourier Series from First Principles

We wish to derive from first principles the Fourier series of the centred unipolar RPT g(t) shown in Figure 4.4a
and the centred unipolar triangular pulse train g2 (t) of Figure 4.4c. This means that we will evaluate the various
Fourier series expressions given in Eq. (4.13).
(NOTE: this worked example involves evaluation of integrals, including integration by parts. The next two
worked examples will focus on the use and interpretation of the results obtained here without need for calculus.)
Deriving the Fourier series of a waveform simply involves calculating its DC component A0 , and obtaining
expressions for the amplitude An and phase 𝜙n of the nth harmonic, n = 1, 2, 3, …
The unipolar RPT (g(t) in Figure 4.4a) has amplitude A, pulse width 𝜏, waveform period T, and duty cycle
d = 𝜏/T. We observe that this waveform is an even function, so the sine coefficient bn in Eq. (4.13) is zero, and the
 4.2 Fourier Series 215

Fourier series of g(t) in Eq. (4.1) simplifies to

∑
∞
g(t) = A0 + an cos(2𝜋nf o t)
n=1

which means that only the DC component A0 and the cosine coefficient an need to be calculated. As discussed
above, the RPT has DC component A0 = Ad. The cosine coefficient an is obtained by evaluating the integral in
Eq. (4.11), noting that within the integration interval (−T/2, T/2) the waveform has a constant value of A in the
range (−𝜏/2, 𝜏/2) and is zero elsewhere. Thus
T∕2 𝜏∕2
2 2
an = g(t) cos(2𝜋nf o t)dt = A cos(2𝜋nf o t)dt
T ∫−T∕2 T ∫−𝜏∕2
[ ] [ ]
𝜏∕2
2A sin(2𝜋nf o t) || 2A sin(𝜋nf o 𝜏) − sin(−𝜋nf o 𝜏)
= =
T 2𝜋nf o ||−𝜏∕2 T 2𝜋nf o
[ ] [ ]
2A 2 sin(𝜋nf o 𝜏) 2A sin(𝜋nf o 𝜏) 𝜏
= = ×
T 2𝜋nf o T 𝜋nf o 𝜏
[ ]
2A𝜏 sin(𝜋nf o 𝜏)
=
T 𝜋nf o 𝜏
= 2Ad sinc(nd), (since d = 𝜏∕T; fo 𝜏 = d)

where we have introduced the sinc function, sinc(x) = sin(𝜋x)/𝜋x, discussed in Section 2.6.8. The Fourier series
of a centred unipolar RPT of amplitude A, waveform period T, and duty cycle d is therefore given by
∑
∞
g(t) = Ad + 2Ad sinc(nd) cos(2𝜋nf o t); fo = 1∕T (4.17)
n=1

A unipolar square wave is a special case of the above pulse train with d = 1/2 for which the formula for the
amplitude of the nth harmonic sinusoid is

An = 2Ad sinc(nd) = A sinc(n∕2)

( )
sin(n𝜋∕2) 2A 𝜋
=A = sin n
n𝜋∕2 n𝜋 2
This simplifies to

⎧0, n even
⎪
An = ⎨ (4.18)
2A
⎪ (−1)(n−1)∕2 , n odd
Telegram: @Persian_IEEEComSoc

⎩ n𝜋
Turning our attention to the centred triangular pulse train g2 (t) (Figure 4.4c) of amplitude A, pulse width 𝜏,
waveform period T, and duty cycle d = 𝜏/T, we note that one cycle g2,T (t) of g2 (t) lies in the range (−T/2, T/2) and
is defined by

⎧
⎪A(1 − 2|t|∕𝜏), |t| ≤ 𝜏∕2
g2,T (t) = ⎨
⎪0, Otherwise
⎩
This waveform is also an even function, so only A0 and the cosine coefficient an need to be calculated. As dis-
cussed earlier, this triangular pulse train has DC component A0 = Ad/2. The cosine coefficient an is obtained by
 216 4 Frequency Domain Analysis of Signals and Systems

evaluating the integral in Eq. (4.11) using the above expression for one cycle of the waveform. Thus
T∕2 𝜏∕2
2 2
an = g (t) cos(2𝜋nf o t)dt = A(1 − 2|t|∕𝜏) cos(2𝜋nf o t)dt
T ∫−T∕2 2 T ∫−𝜏∕2
𝜏∕2 𝜏∕2
2 4A |t|
= A cos(2𝜋nf o t)dt − cos(2𝜋nf o t)dt
∫
T −𝜏∕2 ∫
T −𝜏∕2 𝜏
Notice that the first integral on the right-hand side has already been evaluated above as 2Adsinc(nd), and the
second integral may be doubled and evaluated in the half-interval (0, 𝜏/2). Thus
𝜏∕2
8A
an = 2Ad sinc(nd) − t cos(2𝜋nf o t)dt
T𝜏 ∫0
The integral on the right-hand side is of the form ∫ f ′ gdt, which equals fg − ∫ f g′ dt, with g ≡ t, f′ ≡
cos(2𝜋nf o t); g′ = 1, f = sin(2𝜋nf o t)∕2𝜋nf o . Employing this (integration by parts technique) yields

t cos(2𝜋nf o t)dt = t sin(2𝜋nf o t)∕2𝜋nf o − [sin(2𝜋nf o t)∕2𝜋nf o ]dt

∫ ∫
= t sin(2𝜋nf o t)∕2𝜋nf o + cos(2𝜋nf o t)∕(2𝜋nf o )2

and hence
8A 𝜏∕2
an = 2Ad sinc(nd) − [t sin(2𝜋nf o t)∕2𝜋nf o + cos(2𝜋nf o t)∕(2𝜋nf o )2 |0 ]
T𝜏
( ) ( )
⎡𝜏 𝜏 𝜏 ⎤
sin 2𝜋nf cos 2𝜋nf − 1⎥
8A ⎢ 2 o 2 o 2
T𝜏 ⎢⎢ ⎥
= 2Ad sinc(nd) − +
2𝜋nf o (2𝜋nf o )2 ⎥
⎣ ⎦
2A sin(𝜋nd) 2AT [1 − cos(𝜋nd)]
= 2Ad sinc(nd) − +
𝜋n 𝜏 (𝜋n)2
Multiplying the second term by d/d (in order to make its denominator equal to the sine function argument in
its numerator, which allows us to introduce the sinc function), and noting that in the third term: 2AT/𝜏 = A/(d/2),
1 − cos(𝜋nd) has trigonometric identity 2 sin2 (𝜋nd/2); and multiplying the third term by (d/2)/(d/2), we obtain
2
2A sin(𝜋nd) d A [2sin (𝜋nd∕2)] d∕2
an = 2Ad sinc(nd) − × + ×
𝜋n d d∕2 (𝜋n)2 d∕2
[ ]2
sin(𝜋nd∕2)
= 2Ad sinc(nd) − 2Ad sinc(nd) + Ad
(𝜋nd∕2)
= Ad sinc2 (nd∕2)
Telegram: @Persian_IEEEComSoc

Therefore, the desired Fourier series (of a centred unipolar triangular pulse train g2 (t) (Figure 4.4c) of amplitude
A, waveform period T, and duty cycle d) is
∑
∞
g2 (t) = Ad∕2 + Ad sinc2 (nd∕2) cos(2𝜋nf o t); fo = 1∕T (4.19)
n=1

Worked Example 4.2 Derivation of Fourier Series by Manipulating Standard Waveforms

The Fourier series of a wide selection of signals may be derived, without evaluating any integrals, simply by manip-
ulating already known Fourier series of standard waveforms. This worked example discusses such an approach.
We derive the Fourier series here and leave its interpretation and application for Fourier synthesis to the next
worked example.
 4.2 Fourier Series 217

Equations (4.17) and (4.19), respectively, give the Fourier series of the centred unipolar RPT (g(t) in Figure 4.4a)
and centred unipolar triangular pulse train (g2 (t) in Figure 4.4c). Based on these equations, determine the Fourier
series of each of the following periodic waveforms:

(a) A bipolar rectangular waveform (g1 (t) in Figure 4.4b).

(b) A bipolar triangular waveform (g3 (t) in Figure 4.4d).
(c) A bipolar rectangular pulse train (RPT) (g4 (t) in Figure 4.4e).
(d) An arbitrary periodic staircase waveform having amplitude Ak in the kth step of width 𝜏 k , k = 1, 2, 3, …
(g5 (t) in Figure 4.4f).

(a) As earlier discussed g1 (t) = g(t − to ) − A1 . Using this relation in the Fourier series of g(t) given in Eq. (4.17),
by replacing t with t − to wherever it occurs in that equation and subtracting A1 from the series, we obtain the
desired Fourier series of g1 (t) as

∑
∞
g1 (t) = g(t − to ) − A1 = Ad − A1 + 2Ad sinc(nd) cos(2𝜋nf o (t − to ))
n=1
∑
∞
= Ad − A1 + 2Ad sinc(nd) cos(2𝜋nf o t − 2𝜋nf o to )
n=1

Putting A = 30, A1 = 10, to = T/8, and d = 1/4, and recalling that f o T = 1, yields

1∑
∞
1
g1 (t) = 30 × − 10 + 60 × sinc(nd) cos(2𝜋nf o t − 2𝜋nf o T∕8)
4 4 n=1
∑
∞
= −2.5 + 15 sinc(nd) cos(2𝜋nf o t − n𝜋∕4)
n=1

(b) The triangular waveform g3 (t) is a special case of the triangular pulse train g2 (t) of Figure 4.4c with d = 1, ampli-
tude A = A2 + A1 , DC component A1 subtracted (because it has been shifted vertically downwards through
A1 ), and time shifted by to = −T/5 (since it has been shifted horizontally to the left by T/5 – a value determined
by counting the grids on the graph of g3 (t)). This means that g3 (t) has been advanced and made to start earlier.
Therefore, with the Fourier series of g2 (t) given by Eq. (4.19) as

Ad ∑∞
g2 (t) = + Ad sinc2 (nd∕2) cos(2𝜋nf o t)
2 n=1

the desired Fourier series of g3 (t) is obtained as

Telegram: @Persian_IEEEComSoc

g3 (t) = g2 (t − to ) − A1
Ad ∑∞
= − A1 + Ad sinc2 (nd∕2) cos(2𝜋nf o t − 2𝜋nf o to )
2 n=1
∑
∞
= 10 + 60 sinc2 (nd∕2) cos(2𝜋nf o t + 2n𝜋∕5)
n=1

where we substituted d = 1, to = −T∕5, A = 60, A1 = 20.

(c) The waveform g4 (t) has bipolar pulse width 𝜏, waveform period T and duty cycle d = 𝜏/T. In Figure 4.5, we
show that g4 (t) is the sum of two RPTs g4a (t) and g4b (t). Since g4a (t) has amplitude A, duty cycle d/2, and is
time-shifted relative to the centred RPT g(t) by to = 𝜏/4, the Fourier series of g4a (t) is obtained by replacing d
 218 4 Frequency Domain Analysis of Signals and Systems

g4(t)
A
d = τ/T
τ/2
t
τ
–A
T

=
g4a(t)
A
τ/2
t
T
+
g4b(t)
t
τ/2
–A
T

Figure 4.5 Worked Example 4.2(c).

with d/2 and t with t − 𝜏/4 in Eq. (4.17), noting that f o 𝜏 = f o dT = d. Thus
∑
∞
g4a (t) = Ad∕2 + Ad sinc(nd∕2) cos(2𝜋nf o t − 2𝜋nf o 𝜏∕4)
n=1
∑∞
= Ad∕2 + Ad sinc(nd∕2) cos(2𝜋nf o t − nd𝜋∕2)
n=1

The Fourier series of g4b (t) is similarly obtained from Eq. (4.17) by replacing d with d/2, replacing t with t + 𝜏/4
(since g4b (t) is advanced by 𝜏/4 relative to g(t)), and replacing A with −A (since g4b (t) has amplitude −A). Thus
∑
∞
g4b (t) = −Ad∕2 − Ad sinc(nd∕2) cos(2𝜋nf o t + nd𝜋∕2)
n=1
Telegram: @Persian_IEEEComSoc

The desired Fourier series of g4b (t) is the sum of the two series
∑
∞
g4 (t) = Ad sinc(nd∕2)[cos(2𝜋nf o t − nd𝜋∕2) − cos(2𝜋nf o t + nd𝜋∕2)]
n=1

The term in square brackets is of the form cos(D) − cos(E) which has a trigonometric identity
( ) ( )
E+D E−D
cos(D) − cos(E) = 2 sin sin
2 2
Substituting E ≡ 2𝜋nf o t + nd𝜋∕2; D ≡ 2𝜋nf o t − nd𝜋∕2 yields
( ) ( )
∑∞
d d
g4 (t) = 2Ad sinc n sin 𝜋n sin(2𝜋nf o t)
n=1
2 2
 4.2 Fourier Series 219

Multiplying the sin(𝜋nd/2) factor by (𝜋nd/2)/(𝜋nd/2) allows us to convert it into sinc(nd/2) and this leads to
∑
∞
g4 (t) = 𝜋Ad2 n sinc2 (nd∕2) sin(2𝜋nf o t)
n=1
∑∞
= 𝜋Ad2 n sinc2 (nd∕2) cos(2𝜋nf o t − 90∘ )
n=1

The form of this Fourier series is as expected since g4 (t) is an odd function, so its Fourier series consists only of
sine harmonics sin(2𝜋nf o t) having a phase of ±90∘ ; and it has no DC component. In this case all harmonics will
have the same phase of −90∘ since the amplitude of the nth harmonic, which is the factor 𝜋Ad2 n sinc2 (nd∕2)
in the above series, will always be positive.
(d) This task is a generalisation of (c). A periodic staircase waveform can be treated as the sum of RPTs with
appropriate delays. For example, the waveform g5 (t) in Figure 4.4f is the sum of (i) RPT of amplitude A1 and
delay 𝜏 1 /2 having pulse width 𝜏 1 and duty cycle d1 = 𝜏 1 /T; (ii) RPT of amplitude A2 and delay 𝜏 1 + 𝜏 2 /2 having
pulse width 𝜏 2 and duty cycle d2 = 𝜏 2 /T; (iii) RPT of amplitude A3 and delay 𝜏 1 + 𝜏 2 + 𝜏 3 /2 having pulse width
𝜏 3 and duty cycle d3 = 𝜏 3 /T; and so on.
Referring to each component RPT as a step within one cycle of the staircase waveform, the Fourier series of
the kth RPT (of amplitude Ak and pulse width or step duration 𝜏 k ) is
[ ( )]
∑∞
dk ∑
k−1
gk (t) = Ak dk + 2Ak dk sinc(ndk ) cos 2𝜋nf o t − 2𝜋n + di
n=1
2 i=1

The Fourier series of an m-step staircase periodic waveform is therefore

{ [ ( )]}
∑m
∑∞
dk ∑
k−1
g(t) = Ak dk + 2Ak dk sinc(ndk ) cos 2𝜋nf o t − 2𝜋n + di
k=1 n=1
2 i=1
𝜏 1
dk = k ; fo = (4.20)
T T
This result is completely general and applies to an arbitrary staircase waveform with m steps in one cycle,
including flat-top-sampled signals, in which case the steps will be of the same duration equal to the sampling
interval. It is, however, important to strictly follow the correct convention in identifying the steps: the m steps
are counted starting at t = 0 within the interval (0, T). If the first step straddles the y axis then it must be
counted as two steps, the first being the portion of the step beyond t = 0 and the second being the remaining
portion of the step ending at t = T.

Worked Example 4.3 Fourier Synthesis

The Fourier series of the bipolar rectangular waveform g1 (t) shown in Figure 4.4b was derived in the previous
Telegram: @Persian_IEEEComSoc

worked example as
∑
∞
g1 (t) = −2.5 + 15 sinc(nd) cos(2𝜋nf o t − n𝜋∕4)
n=1

For waveform period T = 0.1 ms

(a) Determine the frequency, amplitude, and phase of the DC component and first 12 harmonic components
of g1 (t). Present your result in tabular form.
(b) Synthesise g1 (t) using its DC component and first five harmonics.
(c) Write a MATLAB code to synthesise and plot g1 (t) using its DC component and first 20 harmonics.
 220 4 Frequency Domain Analysis of Signals and Systems

(a) We start by simply comparing the Fourier series of g1 (t) with the standard form of Fourier series in Eq. (4.2)
and matching corresponding terms
∑
∞
∑
∞
−2.5 + 15 sinc(nd) cos(2𝜋nf o t − n𝜋∕4) ≡ A0 + An cos(2𝜋nf o t + 𝜙n )
n=1 n=1

This shows that:

• DC component of g1 (t): A0 = −2.5 = 2.5∠180∘
• Amplitude of nth harmonic of g1(t): An = 15 sinc(nd)
• Phase of nth harmonic of g1 (t): 𝜙n = −n𝜋∕4 = −n × 45∘
Notice how we treated the negative sign of the DC component as a phase shift of 180∘ which will be taken
to the phase spectrum, leaving just the magnitude of the DC component in the amplitude spectrum. This
is done because the amplitude spectrum An must have only positive real values. All phase information in the
amplitude (e.g. a factor of −1 for 180∘ , or a factor of j for 90∘ , or any other angle) is transferred and consigned
to the phase spectrum.
The amplitude of each harmonic is calculated using the above formula for An with d = 1/4 (as spec-
ified in Figure 4.4b), for n = 1, 2, …, 12. For example, A1 = 15sinc( 1/4) = 15sin(𝜋/4)/(𝜋/4) = 13.505;
A2 = 15sinc( 1/2) = 9.55; and so on. If an amplitude turns out negative then 180∘ is added to the phase of that
harmonic (to account for the factor of −1) and the absolute value is retained as amplitude.
The phase of each harmonic is computed using the above formula for 𝜙n and then converted to a value in
the range −180∘ to 180∘ by adding or subtracting an integer number of cycles (i.e. 360∘ ) as necessary. For
example, the phase of the 10th harmonic is 𝜙10 = −10 × 𝜋/4 = −450∘ ≡ −450 + 360 = −90∘ . The phase of the
fifth harmonic is the phase given by the above formula plus 180∘ transferred from the negative amplitude.
Thus, 𝜙5 = −5 × 45∘ + 180∘ = −225∘ + 180∘ = −45∘ . If the amplitude of a harmonic is zero then its phase is by
convention set to zero irrespective of the value given by the phase formula. For example, 𝜙4 = 0∘ because A4 = 0.
The fundamental frequency of the waveform is f o = 1/T = 1/(0.1 × 10−3 ) = 10 kHz. So, the nth harmonic has
frequency nf o = 10n kHz. The full set of results is tabulated in Table 4.1.

Table 4.1 Harmonic components of g1 (t) in Figure 4.4b.

n Frequency = nf o , kHz 15 sinc(n/4) −n × 45∘ An 𝝓n , deg

0 0(DC) — — 2.5 180

1 10 13.50 −45 13.50 −45
2 20 9.55 −90 9.55 −90
Telegram: @Persian_IEEEComSoc

3 30 4.50 −135 4.50 −135

4 40 0 −180 0 0
5 50 −2.70 −225 2.70 −45
6 60 −3.18 −270 3.18 −90
7 70 −1.93 −315 1.93 −135
8 80 0 −360 0 0
9 90 1.50 −405 1.50 −45
10 100 1.91 −450 1.91 −90
11 110 1.23 −495 1.23 −135
12 120 0 −540 0 180
 4.2 Fourier Series 221

(b) The amplitude and phase of each sinusoidal component of g1 (t) is displayed in the table above up to the 12th
harmonic. Using these values, the synthesised waveform obtained by summing the DC and first five harmonics
is the signal

gs (t) = −2.5 + 13.50 cos(2𝜋fo t − 45∘ ) + 9.55 cos(4𝜋fo t − 90∘ )

+ 4.50 cos(6𝜋f t − 135∘ ) + 2.70 cos(10𝜋f t − 45∘ )
o o

where f o = 10 000 Hz. This signal is plotted in Figure 4.6a in the interval t = (−150, 150) μs to cover three cycles
of 100 μs (= 0.1 ms) each. Synthesising and plotting a waveform as we have done provides a good check that
the amplitudes and phases of the harmonics were correctly calculated.
(c) The following MATLAB code generates and plots the synthesised waveform by adding the DC component and
first 20 harmonics. The plot is shown in Figure 4.6b after some tidying up to include labels, grid, etc.

T = 0.1e-3; %Period of waveform

fo = 1/T; %Fundamental frequency
d = 1/4; %Duty cycle
Nh = 20; %Number of harmonics summed
t = (-1.5*T:T/200:1.5*T); %Time instants for synthesised signal gst
gst = -2.5 + zeros(size(t)); %Initialise gst with DC component
for n = 1:Nh %Now add the harmonics, one at a time.
gst = gst + 15*sinc(n*d)*cos(2*pi*n*fo*t - n*pi/4);
end
plot(t, gst);

20

10

(a)
0

–10

–150 μs –100 μs 0 100 μs 150 μs

20
Telegram: @Persian_IEEEComSoc

10
(b)
0

–10

–150 μs –100 μs 0 100 μs 150 μs

Figure 4.6 Worked Example 4.3: synthesis of g1 (t) of Figure 4.4b using DC and (a) 5 harmonics; (b) 20 harmonics.
 222 4 Frequency Domain Analysis of Signals and Systems

4.2.2 Complex Exponential Form of Fourier Series

The Fourier series discussed in the previous section expresses a periodic signal g(t) as a sum of harmonic sinusoids.
This format directly represents a single-sided spectrum view of g(t) having amplitude and phase spectral lines of
respective heights An and 𝜙n at frequencies nf o , n = 0, 1, 2, 3, … A complex exponential form of the Fourier
series, on the other hand, expresses g(t) as a sum of complex exponentials and directly represents a double-sided
spectrum view of g(t). If g(t) is real then the complex exponentials summed will occur as complex conjugate pairs
so that their imaginary parts cancel when added, thereby producing a real-valued signal. The implication of this
observation is that the amplitude spectrum of a real signal will always be an even function of frequency, whereas
the phase spectrum will always be an odd function.
The complex exponential form of the Fourier series may be derived from the sinusoidal form of the series by
invoking Eqs. (2.47) and (2.48) (based on Euler’s formula) to express cos(2𝜋nf o t) and sin(2𝜋nf o t) in terms of com-
plex exponentials as
1 j2𝜋nf o t
cos(2𝜋nf o t) = (e + e−j2𝜋nf o t )
2
1
sin(2𝜋nf o t) = −j (ej2𝜋nf o t − e−j2𝜋nf o t ) (4.21)
2
Note that ej2𝜋nf o t and e−j2𝜋nf o t are complex conjugates. When these are added together on the right-hand side of
the first line of Eq. (4.21), their imaginary parts (which are equal and of opposite sign) cancel, whereas the real
part is doubled. Subtracting the complex conjugates (as done in the second line of the above equation) cancels
out their real parts (which are equal) and doubles their imaginary part, which is then converted to a real value by
the factor −j. Replacing cos(2𝜋nf o t) and sin(2𝜋nf o t) in the Fourier series of Eq. (4.1) using the right-hand side of
Eq. (4.21) yields
∑
∞
g(t) = A0 + [an cos(2𝜋nf o t) + bn sin(2𝜋nf o t)]
n=1
∑1
∞
= A0 + [an (ej2𝜋nf o t + e−j2𝜋nf o t ) − jbn (ej2𝜋nf o t − e−j2𝜋nf o t )]
n=1
2
∞ { }
∑ 1 1
= A0 + (a − jbn )ej2𝜋nf o t + (an + jbn )e−j2𝜋nf o t
n=1
2 n 2

Note that A0 , an , and bn in the above equation are real coefficients if g(t) is real, and they are given by the
following equations (derived in the previous section)
T∕2 T∕2 T∕2
1 2 2
A0 = g(t)dt; an = g(t) cos(2𝜋nf o t)dt; bn = g(t) sin(2𝜋nf o t)dt
T ∫−T∕2 T ∫−T∕2 T ∫−T∕2
Telegram: @Persian_IEEEComSoc

The factor 12 (an − jbn ) in the first term of the above summation for g(t) will in general be a complex coefficient
(which we will denote as Cn ) given by
1
Cn = (a − jbn )
2 n
T∕2 T∕2
1 1
= g(t) cos(2𝜋nf o t)dt − j g(t) sin(2𝜋nf o t)dt
T ∫−T∕2 T ∫−T∕2
T∕2
1
= g(t)[cos(2𝜋nf o t) − j sin(2𝜋nf o t)]dt
T ∫−T∕2
T∕2
1
= g(t)e−j2𝜋nf o t dt
T ∫−T∕2
 4.2 Fourier Series 223

Setting n = 0 in this equation, we see that

T∕2 T∕2
1 1
C0 = g(t)e−j2𝜋×0×fo t dt = g(t)dt = A0
T ∫−T∕2 T ∫−T∕2
Therefore, substituting Cn into the summation for g(t) yields
∞ { }
∑ 1 1
g(t) = A0 + (an − jbn )ej2𝜋nf o t + (an + jbn )e−j2𝜋nf o t
n=1
2 2
∑
∞
∑
∞
1
= Cn ej2𝜋nf o t + (an + jbn )e−j2𝜋nf o t
n=0 n=1
2
Making the substitution n = −m in the second summation (which means that m = −n and therefore at the
summation limits when n = 1, m = −1, and when n = ∞, m = −∞)
∑
∞
∑
−∞
1
g(t) = Cn ej2𝜋nf o t + (a + jb−m )ej2𝜋mf o t
n=0 m=−1
2 −m
To determine how a−m and b−m are, respectively, related to am and bm , we simply use the expressions for these
coefficients given above, whence we find that
T∕2 T∕2
2 2
a−m = g(t) cos(2𝜋(−m)fo t)dt = g(t) cos(2𝜋mf o t)dt = am
∫
T −T∕2 T ∫−T∕2
T∕2 T∕2
2 2
b−m = g(t) sin(2𝜋(−m)fo t)dt = − g(t) sin(2𝜋mf o t)dt = −bm
T ∫−T∕2 T ∫−T∕2
That is
a−m = am
b−m = −bm (4.22)
and hence
∑
∞
∑
−∞
1
g(t) = Cn ej2𝜋nf o t + (a + jb−m )ej2𝜋mf o t
n=0 m=−1
2 −m
∑
∞
∑
−∞
1
= Cn ej2𝜋nf o t + (am − jbm )ej2𝜋mf o t
n=0 m=−1
2
∑
∞
∑
−∞
= Cn ej2𝜋nf o t + Cm ej2𝜋mf o t
n=0 m=−1

Returning to our (preferred) use of index n (rather than m) in the second summation, we finally obtain
Telegram: @Persian_IEEEComSoc

∑
∞
∑
−∞
g(t) = Cn ej2𝜋nf o t + Cn ej2𝜋nf o t
n=0 n=−1
∑∞
= Cn ej2𝜋nf o t
n=−∞

which completes our work.

So, what have we achieved? We have established that any periodic signal g(t) of period T may be expressed as a
sum of complex exponentials
∑
∞
∑
∞
g(t) = Cn ej2𝜋nf o t = |Cn |ej2𝜋nf o t+𝜙n (4.23)
n=−∞ n=−∞
 224 4 Frequency Domain Analysis of Signals and Systems

where 𝜙n is the angle of the complex coefficients Cn determined from g(t) by integration over one cycle as
T∕2
1
Cn = g(t)e−j2𝜋nf o t dt; n = · · · , −2, −1, 0, −1, −2, · · · (4.24)
T ∫−T∕2
This coefficient is related to the coefficients an and bn of the sinusoidal Fourier series (discussed in the previous
section) by
1
Cn = (an − jbn ) (4.25)
2
Since the summation for g(t) in Eq. (4.23) is from −∞ to ∞, the index n occurs in pairs n = ±1, ±2, ±3, …, and
so do the coefficients (C−1 , C1 ), (C−2 , C2 ), (C−3 , C3 ), … Considering, for example, n = 2, the pair of coefficients is
1
C2 = (a2 − jb2 )
2
1 1
C−2 = (a−2 − jb−2 ) = (a2 + jb2 ) = C2∗
2 2
where we have made use of Eq. (4.22), and the asterisk denotes complex conjugation. In general
1
C−n = Cn∗ = (a + jbn ), n = 1, 2, 3, · · · (4.26)
2 n
Corresponding coefficients on either side of the summation in Eq. (4.23) are therefore complex conjugates. This
means that we only need to calculate the coefficients for the positive side of the summation (n = 1, 2, 3, …). The
corresponding coefficients for the negative side (n = −1, −2, −3, …) then follow simply by complex conjugation
(i.e. changing the sign of the imaginary part). It is useful to note that replacing n with −n in Eq. (4.24) has the effect
of complex conjugating the right-hand side. This provides alternative proof of the complex conjugate relationship
between Cn and C−n for real g(t). Note also that if g(t) is an even function then bn = 0 (according to Eq. (4.16))
and Cn will be exclusively real-valued. But if g(t) is odd then an = 0 and Cn is exclusively imaginary-valued, with
C0 = 0. That is
Cn = C−n = an ∕2, g(t) even
Cn = −C−n = −jbn ∕2, g(t) odd (4.27)
It is worth emphasising that the Fourier series technique provides alternative methods of describing the same
periodic signal, in the time domain by specifying the waveform g(t) or in the frequency domain by specifying the
Fourier coefficients Cn . Once we have a specification in one domain, the specification of the signal in the other
domain may be fully determined. So, given g(t), we may obtain Cn using Eq. (4.24), a process known as Fourier
analysis; and given Cn , we may fully reconstruct g(t) using Eq. (4.23), a process known as Fourier synthesis.

4.2.3 Amplitude and Phase Spectra

Telegram: @Persian_IEEEComSoc

Insight into the utility of the sinusoidal and complex exponential forms of the Fourier series and how each form
is interpreted and visualised to extract practical information about the frequency content, bandwidth, and power
of the signal may be gained by employing the concept of phasors, which was introduced in Chapter 2.
Figure 4.7a shows a circle of unit radius |EG| = 1, centred in the complex plane with real and imaginary axes
shown. Considering the triangle EFG with angle 𝜃 at E, it follows by definition of the cosine and sine functions that
|EF| = cos 𝜃, |FG| = sin 𝜃. A few comments on basic arithmetic are in order here. Every number (in the complex
plane) has both magnitude and direction or angle. For example, +5 has magnitude 5 and angle 0∘ , denoted 5∠0∘
(pronounced ‘5 angle 0’); −5 = 5∠180∘ ; j5 = 5∠90∘ ; and (in Figure 4.7a) EG = |EG|∠𝜃; etc. The correct addition
and subtraction of numbers always respects their angles, although phasor addition (which is basically the same
process) is often treated as if it were something special. It isn’t! For example, 4 – 3 = 1 because we add 3∠180∘ to
4∠0∘ by moving (from the origin) through 4 units in the direction of 0∘ to get to a point, and then moving from that
 4.2 Fourier Series 225

Counterclockwise
rotation at
f cycles/sec,
starting at θ = ϕ
Imaginary
G jθ
ejθ Ae
=
jsin θ (t)
θ F g1 jAsin θ
Real Acos θ
E cos θ
Acos θ
–jAsin θ

g2
(t)
=
Ae
–j
θ
Clockwise rotation
at f cycles/sec,
starting at θ = –ϕ
(a) (b) (c)

Figure 4.7 Concept of positive and negative frequencies: (a) Unit radius in complex plane; (b) Phasor or complex
exponential function at positive frequency; (c) Phasor at negative frequency.

point through 3 units in the direction of 180∘ to complete the operation, which takes us to the point 1∠0∘ as the
answer. Without this respect for the angle of each number, we could have ended up with the wrong result, such
as 7∠0∘ or 5∠36.87∘ (if −3 were added as if it had angle 0∘ or 90∘ ). Notice therefore in Figure 4.7a that
|EG|∠𝜃 = 1∠𝜃 = |EF|∠0∘ + |FG|∠90∘
= cos 𝜃 + j sin 𝜃
= ej𝜃 (by Euler’s formula)
This means that Aej𝜃 , or A exp(j𝜃), represents a complex number A∠𝜃 having magnitude A and angle 𝜃. In
Figure 4.7b and c, this idea is extended to a complex exponential function by making 𝜃 a function of time as
follows: in (b), the radial arm or phasor of magnitude A rotates counterclockwise at a rate of f cycles per second,
starting initially at angle 𝜙 (radian) at time t = 0. Thus, 𝜙 is the phase (i.e. initial angle) of the function. Angular
displacement is by convention positive in the counterclockwise direction, so this phasor is said to have a positive
frequency f Hz (or positive angular frequency 2𝜋f rad/s), the angle of the phasor at any time instant t will be
𝜃 = 2𝜋ft + 𝜙, and this variation produces a complex exponential function
g1 (t) = A ej(2𝜋ft+𝜙) = A cos(2𝜋ft + 𝜙) + jA sin(2𝜋ft + 𝜙) (4.28)
which has constant magnitude A, constant positive frequency f , constant phase 𝜙, and sinusoidally varying real
and imaginary parts.
Telegram: @Persian_IEEEComSoc

In Figure 4.7c the rotation is clockwise at f cycles/sec starting initially at −𝜙, so (since clockwise angular dis-
placement is negative), the angle of the phasor at time t is − (2𝜋ft + 𝜙) ≡ 2𝜋(−f )t − 𝜙, and this variation produces
a complex exponential function
g2 (t) = A ej(2𝜋(−f )t−𝜙) = A cos(2𝜋ft + 𝜙) − jA sin(2𝜋ft + 𝜙) (4.29)
which has constant magnitude A and negative frequency −f . The two complex exponentials g1 (t) and g2 (t), pro-
duced by phasors counter-rotating at equal rate, are complex conjugates which when combined (i.e. by adding
Eqs. (4.28) and (4.29)) yields
A j(2𝜋ft+𝜙) A j(2𝜋(−f )t−𝜙)
A cos(2𝜋ft + 𝜙) = e + e (4.30)
2 2
Figure 4.7b and c and Eq. (4.30) show that
 226 4 Frequency Domain Analysis of Signals and Systems

Function: Amplitude spectrum: Phase spectrum:

A exp( j(2π ft + ϕ)) A/2 ϕ

(a)
2
f Freq., Hz f Freq., Hz

A/2
(b) A exp( j(–2π ft – ϕ))
2 –f
–f Freq., Hz Freq., Hz
–ϕ

A/2 ϕ
(c) A cos(2π ft + ϕ)
–f
–f f Freq., Hz f Freq., Hz
–ϕ
A
Single-sided spectrum of ϕ
(d)
A cos(2πft + ϕ)
f Freq., Hz f Freq., Hz

Figure 4.8 Frequency domain view of sinusoidal and complex exponential functions.

● The complex exponential function A2 ej(2𝜋ft+𝜙) has at all times a real and an imaginary component, and corre-
sponds to a single phasor of amplitude A/2 that rotates counterclockwise at f cycles per second in the complex
plane, starting initially at angle (i.e. orientation relative to the +x axis reference direction) equal to 𝜙. This func-
tion is said to have phase 𝜙 and positive frequency f (in view of the convention for angular measure that regards
counterclockwise angular displacement as positive and clockwise angular displacement as negative). Figure 4.8a
gives a frequency domain view of this complex exponential, depicting its amplitude spectrum which contains a
single vertical line of height A/2 located at point f along the frequency axis, and its phase spectrum which has
a single vertical line of height 𝜙 at f .
● The complex exponential function A2 e−j(2𝜋ft+𝜙) corresponds to a single phasor of amplitude A/2 that rotates clock-
wise at f cycles per second in the complex plane, starting initially at angle −𝜙. This function is said to have phase
−𝜙 and (negative) frequency −f . Its amplitude spectrum contains a single vertical line of height A/2 located at
point −f along the frequency axis, and its phase spectrum contains a single vertical line of height −𝜙 at frequency
−f . See Figure 4.8b.
The sinusoidal function A cos(2𝜋ft + 𝜙) consists of two counter-rotating phasors or complex exponentials, each
Telegram: @Persian_IEEEComSoc

of amplitude A/2. One phasor has phase 𝜙 and rotates counterclockwise at f cycles/second (which corresponds
to positive frequency f ). The other phasor has phase −𝜙 and rotates clockwise at f cycles/second (which corre-
sponds to negative frequency −f ). When these two phasors are added, their imaginary parts cancel, producing a
resultant function that is always real-valued. Therefore, as shown in Figure 4.8c, the amplitude spectrum of the
sinusoid A cos(2𝜋ft + 𝜙) contains two lines of height A/2 at locations ±f ; and its phase spectrum also contains
two lines, one of height 𝜙 at f , and the other of height −𝜙 at −f .

It should therefore be stressed that negative frequency is not merely an abstract mathematical concept but
a practical quantity with physical meaning. Negative frequency is the rate of change of angle in the clockwise
direction exhibited by the clockwise-rotating phasor or complex exponential exp(−j2𝜋ft) – assuming f > 0,
whereas positive frequency is the rate of change of angle in the counterclockwise direction exhibited by the
 4.2 Fourier Series 227

counterclockwise-rotating phasor or complex exponential exp(j2𝜋ft). A sinusoidal signal A cos(2𝜋ft + 𝜙) is

composed of two complex exponentials of amplitude A/2 that vary in sync at frequencies ±f so that their real
parts reinforce at all times and their imaginary parts cancel. This requires the complex exponentials to have
respective phases ±𝜙. The correct and complete spectrum of this sinusoid is therefore as shown in Figure 4.8c for
amplitude and phase. However, the spectrum of the sinusoid A cos(2𝜋ft + 𝜙) is often reduced for convenience to
the single-sided format shown in Figure 4.8d, where its amplitude spectrum is depicted as a single spectral line
of height A at positive frequency f , and its phase spectrum as a single line of height 𝜙 at f .

4.2.3.1 Double-sided Spectrum

A double-sided spectrum provides amplitude and phase information on both the positive and negative frequency
content of a signal, which means all the complex exponentials that constitute the signal. This double-sided spec-
tral representation is really the accurate and unabridged frequency domain view of a signal and leads to a lot of
mathematical simplification when dealing, for example, with passband transmission systems in which the signal
is translated to a different frequency band.
The complex exponential form of the Fourier series, given by Eq. (4.23), indicates that a periodic signal g(t) of
period T is made up of a DC component C0 and pairs of complex exponentials

Cn ej2𝜋nf o t ; n = ±1, ±2, ±3, · · · ; fo = 1∕T

where Cn is in general complex-valued, having magnitude |Cn | and angle 𝜙n , so it may be expressed as

Cn = |Cn |∠𝜙n = |Cn |ej𝜙n

which means that the complex exponentials are of the form

|Cn |ej𝜙n ej2𝜋nf o t = |Cn |ej(2𝜋nf o t+𝜙n ) ; n = ±1, ±2, ±3, · · ·

revealing that their amplitude is |Cn |, phase is 𝜙n , and frequency is nf o . From Eqs. (4.26), (4.25), and (4.13), it
follows that these amplitudes and phases satisfy the relations
√
1 1
|Cn | = |C−n | = a2n + b2n = An ; n > 0 (4.31)
2 2
where An ≡ amplitude of the nth harmonic in the sinusoidal form of Fourier series, and
{ }
1
𝜙n = Angle of Cn ≡ (an − jbn )
2
⎧𝛼, bn ≤ 0, an ≥ 0
⎪ ( )
⎪−𝛼, bn > 0, an ≥ 0 |bn |
=⎨ ; 𝛼 = tan−1
⎪180 − 𝛼, bn ≤ 0, an < 0 |an |
Telegram: @Persian_IEEEComSoc

⎪𝛼 − 180, b > 0, a < 0

⎩ n n
{ }
1
𝜙−n = Angle of C−n = (an + jbn ) = −𝜙n (4.32)
2
Note that this phase 𝜙n is the same as the phase of the nth harmonic in the sinusoidal form of Fourier series. We
may now summarise the features of the double-sided spectrum of a periodic signal g(t) as follows:

● The double-sided spectrum consists of a DC component and pairs of spectral lines at negative and positive
frequencies ±nf o , for n = 1, 2, 3, …
● The spectrum is a discrete line spectrum, with a constant gap or spacing between adjacent spectral lines equal
to the fundamental frequency f o of the periodic signal. This spacing depends only on the period of the signal
according to f o = 1/T.
 228 4 Frequency Domain Analysis of Signals and Systems

Cn
A0 , n=0
C1 C0 C
C2 1
C2 Cn = 1
C3 C3 A , n≥1
(a) 2 n
nfo
–3fo –2fo –fo 0 fo 2fo 3fo

ϕn ϕ2 0˚, C0 ≥ 0
ϕ1 ϕ0 =
–ϕ3 180˚, C0 < 0
(b) –2f o –f o ϕ0 3f o
nf o
–3f o 0 fo 2f o
–ϕ2 –ϕ1 ϕ3

An A1 C0 , n=0
An =
A0
A2 2 Cn , n≥1
ϕn ϕ2
A3 ϕ1
ϕ0 3f o
nf o nf o
0 fo 2f o 3f o 0 fo 2f o
ϕ3
(c) (d)

Figure 4.9 (a) Double-sided amplitude spectrum; (b) Double-sided phase spectrum; (c) Single-sided amplitude spectrum;
(d) Single-sided phase spectrum.

● Each spectral line in a double-sided amplitude spectrum (DSAS) represents a complex exponential component
of the signal g(t). The height and location (along the frequency axis) of a spectral line respectively specify the
amplitude and frequency of the complex exponential that the line represents. The DSAS is therefore a plot of
|Cn | against nf o for n = …, −3, −2, −1, 0, 1, 2, 3, …, which, in view of Eq. (4.31), is an even function of frequency
if the signal g(t) is real.
● The double-sided phase spectrum (DSPS) of g(t) is a plot of the phase 𝜙n of the complex exponential components
of g(t) against nf o , n = …, −3, −2, −1, 0, 1, 2, 3, … If g(t) is a real signal then 𝜙−n = −𝜙n and the phase spectrum
is therefore an odd function of frequency. Note that the phase 𝜙0 of the DC component C0 will be either 0∘ if
C0 ≥ 0 or 180∘ if C0 is negative.
Figure 4.9a and b show the double-sided amplitude and phase spectra, respectively. Since each spectral line in
Figure 4.9a is a complex exponential of normalised power |Cn |2 , according to Eq. (3.109), the total normalised
Telegram: @Persian_IEEEComSoc

power of g(t) may be determined in the frequency domain by summing the powers of all the complex exponential
components represented in its amplitude spectrum, or in the time domain (discussed in Chapter 3) by averaging
the squared signal over one cycle. Thus
∑
∞
1
T∕2
P= |Cn |2 = g2 (t)dt (4.33)
n=−∞
T ∫−T∕2

4.2.3.2 Single-sided Spectrum

We emphasise from the outset that the single-sided spectrum is a contraction of the true spectrum of real sig-
nals that is commonly used simply for convenience. Because the amplitude spectrum of real signals is even and
the phase spectrum is odd, the right-hand side of the spectrum (for n ≥ 0) contains all the amplitude and phase
 4.2 Fourier Series 229

information needed to fully represent a real signal in the frequency domain. However, we cannot simply delete
the left half of a DSAS of (Figure 4.9a), as this would eliminate half of the ac power content of the signal. To
ensure that it contains the full power as well as the spectral information of the signal, a single-sided spectrum is a
representation of the sinusoidal Fourier series
∑
∞
g(t) = A0 + An cos(2𝜋nf o t + 𝜙n )
n=1

using one spectral line per harmonic sinusoid. A single-sided amplitude spectrum (SSAS) is a plot of the values of
|An | in the above equation against nf o for n = 0, 1, 2, 3, …, as shown in Figure 4.9c. The SSAS is a discrete line
spectrum covering only positive frequencies with a constant gap or spacing between adjacent spectral lines equal
to the fundamental frequency f o of g(t). Using Eq. (4.31), the height of the spectral line at f = nf o in a SSAS is
related to that of a DSAS by
{
|C0 |, n=0
|An | = (4.34)
2|Cn |, n ≥ 1
An SSAS, shown in Figure 4.9d, is a discrete line plot of the values of 𝜙n (confined to the range −180∘ to 180∘
and including an additional phase shift of 180∘ if A < 0) against nf for n = 0, 1, 2, 3, …
n o
From the above discussion and Figure 4.9, we see that a double-sided spectrum may be converted into a
single-sided spectrum by transferring all spectral lines at negative frequency locations in the amplitude spectrum
to add to their counterparts at positive frequency locations, and by discarding all negative frequency lines in the
phase spectrum. Conversely, we may convert a single-sided spectrum into double-sided by taking each spectral
line at f = nf o , n = 1, 2, 3, … (representing the nth harmonic sinusoidal component of g(t) and having amplitude
An and phase 𝜙n ) and dividing it equally to a pair of locations on the frequency axis, one at the positive frequency
f = nf o having amplitude An /2 and phase 𝜙n ; the other at the negative frequency f = −nf o having amplitude
An /2 and phase −𝜙n . Note that, although the DC component A0 is untouched in this conversion, it may in fact be
viewed as a pair of components A0 /2 at f = +0 and A0 /2 at f = −0.
For later application to signal bandwidth and power analysis, we note here that the total normalised power PN in
the DC and first N harmonics of a signal may be determined from the SSAS (where each line represents a sinusoid
of normalised power A2n ∕2) or from the DSAS (where each line represents a complex exponential of normalised
power |Cn |2 ), except for the DC component with normalised power A20

1∑ 2 ∑
N N
PN = A20 + An = C02 + 2 |Cn |2 (4.35)
2 n=1 n=1
Finally, it is worth noting that the relationship between double-sided and single-sided spectra may be
heuristically demonstrated as follows, without involving complex exponential functions as done hitherto. Since
cos(𝜃) = cos(−𝜃), we may write
Telegram: @Persian_IEEEComSoc

A A
An cos(𝜃) = n cos(𝜃) + n cos(−𝜃)
2 2
Employing this identity in the sinusoidal form of the Fourier series, with 𝜃 ≡ 2𝜋nf o t + 𝜙n , leads to
∑
∞
g(t) = A0 + An cos(2𝜋nf o t + 𝜙n ); (single-sided)
n=1
∞ [ ]
∑ An An
= A0 + cos(2𝜋nf o t + 𝜙n ) + cos(2𝜋(−nf o )t − 𝜙n )
n=1
2 2
∑∞
An
= A0 + cos(2𝜋nf o t + 𝜙n ); 𝜙−n = −𝜙n ; (double-sided) (4.36)
n=−∞; 2
n≠0
 230 4 Frequency Domain Analysis of Signals and Systems

which, with the convention of using one spectral line per sinusoid, shows that the sinusoid An cos(2𝜋nf o t + 𝜙n ) is
converted from a single spectral line of height An at f = nf o in a SSAS to a pair of spectral lines of height An /2 at
f = ±nf o in a DSAS. In its phase spectrum, the sinusoid is also converted from a single line 𝜙n at f = nf o to a pair
of lines of respective values ±𝜙n at f = ±nf o .
A note of caution is necessary here to avoid a common misunderstanding. Treating the first line of Eq. (4.36) as
leading to a single-sided spectrum is not mathematically sound. This is a practice adopted simply for convenience
in which the amplitude spectrum of the sinusoid An cos(2𝜋nf o t + 𝜙n ) is represented as a single spectral line of
height An at frequency f = nf o ; and its phase spectrum as a single spectral line of height 𝜙n at f = nf o . A correct
and mathematically rigorous interpretation has been discussed at length in the preceding pages, namely (i) a real
signal has a double-sided spectrum; (ii) each spectral line represents a complex exponential function, rather than
a sinusoidal function; (iii) the amplitude spectrum of the sinusoid An cos(2𝜋nf o t + 𝜙n ) consists of a pair of spectral
lines of value An /2 at frequencies f = ±nf o ; and its phase spectrum has corresponding values ±𝜙n at f = ±nf o .

Worked Example 4.4 Parameters of Complex Exponential Functions

The purpose of this worked example is to use a series of simple exercises to ensure you have complete clarity about
complex exponential functions. Let us determine the following:
√
(a) The frequency and phase of g(t) = −10( exp[j(t
) − 𝜋∕3)].
(b) The frequency and phase of g(t) = exp j 𝜋2 t .
(c) The frequency of g(t) = 20e−j(200𝜋t−𝜋∕6) and the angle, magnitude, and real part of this function at time
t = 2.5 ms.
(d) The parameters of ej .
(e) The parameters of g(t) = B0 e-𝜆t .
(f) The parameters of g(t) = 50 exp[(j30𝜋 − 2)t].
√
(a) We start with a straightforward manipulation to move all phase information from the coefficient −10 into
the angle of the complex exponential function g(t)
√
g(t) = −10 exp[j(t − 𝜋∕3)]
√ √
= j 10ej(t−𝜋∕3) = ( 10∠𝜋∕2)ej(t−𝜋∕3)
√ √
= ( 10ej𝜋∕2 )ej(t−𝜋∕3) = 10ej(t+𝜋∕2−𝜋∕3)
√
= 10ej(t+𝜋∕6)
Now comparing the above expression to the standard form of a complex exponential function Aej(2𝜋ft+𝜙) , where
A > 0 ≡ Amplitude
Telegram: @Persian_IEEEComSoc

f ≡ Frequency
𝜙 ≡ Phase
2𝜋ft + 𝜙 ≡ Angle at time t
we see that, for the given signal, 2𝜋f = 1, which yields frequency f as
1
f = Hz = 159.2 ms
2𝜋
And the phase is read as 𝜙 = 𝜋/6 rad = 30∘ .
(b) The given function compares as follows with the standard form
( ) [ ( )]
𝜋 𝜋
g(t) = exp j t = exp j t + 0 ≡ A exp[j(2𝜋ft + 𝜙)]
2 2
 4.2 Fourier Series 231

Thus, phase 𝜙 = 0, and frequency f is obtained from

𝜋
2𝜋f = ; and hence f = 0.25 Hz
2

(c) Again, the given function compares as follows with the standard form

g(t) = 20e−j(200𝜋t−𝜋∕6) = 20ej(−200𝜋t+𝜋∕6) ≡ Aej(2𝜋ft+𝜙)

Thus
Frequency f is given by 2𝜋f = −200𝜋, which implies f = −100 Hz
Angle at t = 2.5 ms

𝜃(t)|t=2.5×10−3 = −200𝜋 × 2.5 × 10−3 + 𝜋∕6

= −200𝜋 × 2.5 × 10−3 + 𝜋∕6
= −𝜋∕2 + 𝜋∕6
= −𝜋∕3 radian = −60∘

The magnitude of a complex exponential function is constant and in this case is |A| = 20 at all times.
The real part of Aej(2𝜋ft+𝜙) is A cos(2𝜋ft + 𝜙), which for the given function at t = 2.5 ms is 20 cos(−200𝜋 × 2.5 ×
10−3 + 𝜋∕6) = 20 cos(−60∘ ) = 10
(d) The term ej is not a function of time. It is simply a complex number in the same form as ej𝜃 which equals 1∠𝜃.
So, in this case

ej = 1∠1 rad = 1∠(180∕𝜋)∘ = 1∠57.3∘

= cos(57.3∘ ) + j sin(57.3∘ )
= 0.5403 + j0.8415

(e) The function g(t) = B0 e−𝜆t is not a complex exponential function, seeing there is no j factor in its exponent.
Rather, g(t) is a real function that decays exponentially with time from an initial value g(0) = B0 at t = 0 to a
final value g(∞) = 0 at t = ∞. An exponentially decaying function is often characterised by its time constant
𝜏, which is the time taken for the function to reduce to 1/e (or e−1 ) times its initial value. Here, 𝜏 = 1/𝜆, since
g(𝜏) = B0 e−𝜆𝜏 ≡ B0 e−1 . Note that e ≈ 2.718281828459046, so 1/e = 36.788%.
(f) The given function can be rearranged as

g(t) = 50e(j30𝜋−2)t = 50e−2t ej30𝜋t ≡ Aej(2𝜋ft+𝜙)

Telegram: @Persian_IEEEComSoc

which reveals that it is an amplitude-modulated complex exponential function. Its amplitude decays exponen-
tially from an initial value of 50 with a time constant of 0.5 s, whereas its frequency and phase are constant
(i.e. unmodulated) and, respectively, f = 15 Hz, 𝜙 = 0∘ .

Worked Example 4.5 Amplitude and Phase Spectra of Pulse Train

We wish to sketch the single-sided and double-sided amplitude and phase spectra of the unipolar RPT x(t) shown
in Figure 4.10 up to the 12th harmonic and to discuss its features.
The Fourier series of the centred unipolar RPT g(t) in Figure 4.4a was derived in Worked Example 4.1 as given
in Eq. (4.17). Since the given waveform x(t) is the centred waveform g(t) delayed by half a pulse width 𝜏, we make
 232 4 Frequency Domain Analysis of Signals and Systems

x(t), volts

τ d = τ/T = ¼
A = 100 V
T T = 1 ms
t

Figure 4.10 Worked Example 4.5.

use of the relationship x(t) = g(t − 𝜏∕2) in Eq. (4.17) to obtain

∑
∞
x(t) = g(t − 𝜏∕2) = Ad + 2Ad sinc(nd) cos(2𝜋nf o (t − 𝜏∕2))
n=1
∑
∞
= Ad + 2Ad sinc(nd) cos(2𝜋nf o t − n𝜋d), since fo 𝜏 = d
n=1

Substituting the signal parameters (A = 100, d = 1/4) given in Figure 4.10 yields
∑
∞
x(t) = 25 + 50 sinc(n∕4) cos(2𝜋nf o t − n𝜋∕4)
n=1
fo = 1∕10−3 = 1000 Hz

Therefore

A0 = 25
𝔸n = 50 sinc(n∕4)
An = |𝔸n |
⎧−45n∘ , 𝔸n > 0
⎪
𝜙n = ⎨180∘ − 45n∘ , 𝔸n < 0
⎪ ∘
⎩0 , 𝔸n = 0

Notice that (as discussed earlier) the amplitude spectrum An is defined to be positive and an extra phase shift of
180∘ is added to the phase of the nth harmonic whenever its amplitude is negative. Also, the phase of a harmonic is
set to zero if its amplitude is zero. Computed values are tabulated in Table 4.2 up to the 12th harmonic. To obtain
the values listed in the last three columns of the table for the doubled-sided spectrum, the values of An obtained
as above were divided in two to give the positive and negative frequency coefficients Cn and C−n , and the phase
of the positive frequency was negated to obtain the phase of the negative frequency component. Take care to see
Telegram: @Persian_IEEEComSoc

how the values of 𝜙n in column six of the table are derived from column four. For example, for the 11th harmonic,
360∘ is added to the value of −495∘ in column four to bring it into the required interval of −180∘ to 180∘ ; and for
the fifth harmonic, 180∘ is added to the value in column four because the amplitude (in column three) is negative
at that harmonic.
Figure 4.11 shows a discrete line plot of 𝜙n and 𝜙−n against ±nf o to produce the double-sided phase spectrum.
The SSAS is shown in Figure 4.12. This is a discrete line plot of An against nf o . A plot of Cn and C−n against ±nf o
gives the DSAS shown in Figure 4.13.
We can now make the following observations regarding the spectrum of an RPT.
Spectral Line Spacing: spectral lines occur at frequencies f = 0, ±f o , ±2f o , ±3f o , …, where f = f o = 1/T is the
fundamental frequency component, f = nf o is the nth harmonic frequency, and T is the period of the pulse train.
The question is often asked whether the spectral line spacing f o is affected by the pulse width 𝜏 or duty cycle d of
 4.2 Fourier Series 233

Table 4.2 Worked Example 4.5: spectral values for rectangular pulse train of Figure 4.10.

Frequency,
n nf o , kHz An = 50sinc(n/4) −n×45∘ An = |An | 𝝓n , deg C n = An /2 C −n = An /2 𝝓−n = −𝝓n

0 0 — — 25 0 C0 = 25 0
1 1 45.02 −45 45.02 −45 22.51 22.51 45
2 2 31.83 −90 31.83 −90 15.92 15.92 90
3 3 15.01 −135 15.01 −135 7.50 7.50 135
4 4 0 −180 0 0 0 0 0
5 5 −9.00 −225 9.00 −45 4.50 4.50 45
6 6 −10.61 −270 10.61 −90 5.31 5.31 90
7 7 −6.43 −315 6.43 −135 3.22 3.22 135
8 8 0 −360 0 0 0 0 0
9 9 5.00 −405 5.00 −45 2.50 2.50 45
10 10 6.37 −450 6.37 −90 3.18 3.18 90
11 11 4.09 −495 4.09 −135 2.05 2.05 135
12 12 0 −540 0 0 0 0 0

135
90
Phase ϕn, deg

45
0
–45
–90
–135
–12 –10 –8 –6 –4 –2 0 2 4 6 8 10 12
Frequency nfo, kHz

Figure 4.11 Worked Example 4.5: Double-sided phase spectrum of waveform x(t) in Figure 4.10 having amplitude 100 V,
duty cycle 1/4, period 1 ms, and delay 1/8 ms relative to centred waveform.

the pulse train. It is true that we may write

1 d
=fo =
T 𝜏
Telegram: @Persian_IEEEComSoc

from which it may be noted that keeping d fixed while changing 𝜏, or keeping 𝜏 fixed while varying d, would alter
f o . However, f o is changed only because each of these actions changes T. In other words, f o will change if and only
if T changes; and changing d and/or 𝜏 will influence f o only if that change affects T. So, yes, spectral line spacing
f o depends exclusively on period T.
Spectral Nulls: the amplitude of the nth harmonic sinusoidal component is An = 2Ad sinc(nd), n = 1, 2, 3, …
However, since (see Section 2.6.8)
sinc(1) = sinc(2) = sinc(3) = · · · = 0
harmonics corresponding to nd = 1, 2, 3, …, (hence n = 1/d, 2/d, 3/d, …) will have zero amplitude. Therefore,
every (1/d)th harmonic will be absent from the spectrum. For example, a square wave is an RPT with d = 0.5, so
it will not contain every (1/0.5)th harmonic. That is, even harmonics 2f o , 4f o , 6f o , etc. are absent from a square
 234 4 Frequency Domain Analysis of Signals and Systems

46

40

30
Amplitude An,V

spectral line

20

10

0
0 1 2 3 4 5 6 7 8 9 10 11 12
Frequency nfo, kHz

Figure 4.12 Worked Example 4.5: Single-sided amplitude spectrum of rectangular pulse train of amplitude 100 V, duty
cycle 1/4, period 1 ms.

25
spectral envelope

20
Amplitude |Cn|, V

15

10

0
Telegram: @Persian_IEEEComSoc

–12 –10 –8 –6 –4 –2 0 2 4 6 8 10 12
Frequency nfo, kHz

Figure 4.13 Worked Example 4.5: Double-sided amplitude spectrum of rectangular pulse train of amplitude 100 V, duty
cycle 1/4, period 1 ms.

wave. In the current example (Figure 4.10), d = 1/4, and this explains why there is a null in the amplitude spectrum
(Figure 4.12) at every fourth harmonic 4f o , 8f o , 12f o , etc.
Density of Spectral Lines in each Lobe: since nulls occur at every 1/d harmonic count, and spectral line spacing
is f o , it means that one lobe (between adjacent nulls) of the spectral envelope has width f = f o (1/d) = 1/𝜏. The
number of spectral lines contained in one lobe therefore increases as duty cycle d decreases. If 1/d is an integer
 4.2 Fourier Series 235

then the relationship is exactly

1
Number of spectral lines in one lobe = −1 (4.37)
d
Keeping pulse width 𝜏 constant also fixes the spectral lobe width. Under this condition, if the period T of the
RPT is increased (meaning that d = 𝜏/T decreases) then spectral line spacing (f o = 1/T) decreases and hence the
number of spectral lines increases.
Null Bandwidth: the width of the main lobe in the positive frequency range of the amplitude spectrum is known
as the null bandwidth Bnull of the pulse train. This is the range from f = 0 to the first null in the spectrum, which
occurs at n = 1/d or f = nf o = (1/d)f o = 1/𝜏. Thus
1
Bnull = (4.38)
𝜏
and the bandwidth of the pulse train is inversely proportional to the pulse width. Narrower pulses are often
required, for example, in time division multiplexing to pack several independent (information-bearing) pulses into
a fixed sampling period T. We see therefore that such capacity increase is achieved at the expense of (at least) a pro-
portionate increase in bandwidth requirements. The constant of inverse proportionality is 1 for rectangular-shaped
pulses, as in Eq. (4.38), but will be a different value for other pulse shapes. Equation. (4.38) is a special case of what
is actually a more general phenomenon of the inverse relationship between time domain and frequency domain
representations. A shorter duration event (such as a pulse) in the time domain is represented in the frequency
domain using a wider bandwidth (or frequency range) and vice versa.

4.2.4 Fourier Series Application to Selected Waveforms

Fourier analysis is applicable to a very wide range of signals that may be generated in the laboratory or encountered
in telecommunications, and it gives valuable insight into the frequency content, bandwidth, and distribution of
power among the various frequency components of the signal. Apart from the Dirichlet conditions discussed ear-
lier (which will always be satisfied by such signals), the only other requirement for the applicability of the Fourier
series technique is that the signal must be periodic. This later condition can be satisfied by focusing on the signal
interval of interest and assuming that outside of this interval the signal has the same repetitive waveform.
In this section we wish to briefly demonstrate the versatility of the Fourier series tool by applying it to a selection
of three special waveforms, namely the flat-top-sampled signal, the BASK signal, and the trapezoidal pulse train.
Other interesting applications are left as exercises at the end of this chapter.

4.2.4.1 Flat-top-Sampled Signal

Sampling is an important signal processing task in telecommunications and Chapter 9 is devoted to its detailed
treatment. Here we wish to demonstrate how the Fourier series technique may be employed to gain useful insight
into the frequency domain effect of two types of sampling, namely flat-top sampling and instantaneous sampling.
Telegram: @Persian_IEEEComSoc

For this purpose, consider one cycle of the signal (in Figure 4.14a)
g(t) = 1 + 5 sin(2𝜋fo t) + 3 sin(4𝜋fo t) + 2 sin(8𝜋fo t)
= 1 + 5 cos(2𝜋fo t − 90∘ ) + 3 cos(4𝜋fo t − 90∘ ) + 2 cos(8𝜋fo t − 90∘ ) (4.39)
which contains frequency components of respective amplitudes 1, 5, 3, and 2 units at DC, f o , 2f o , and 4f o . This
signal is sampled at regular intervals T s corresponding to a sampling frequency F s = 1/T s chosen to be four times
the maximum frequency component of g(t) or F s = 16f o . There are therefore 16 samples in one cycle of g(t).
Figure 4.14a depicts instantaneous sampling, which requires a switch to operate momentarily at each sampling
instant nT s , n = 0, 1, 2, …, to produce an impulsive sample of weight equal to g(nT s ), which we may express in
terms of the unit impulse as
g𝛿 (n) = g(nT s )𝛿(t − nT s ); n = 0, 1, 2, · · · ; Ts = T∕16; T = 1∕fo
 236 4 Frequency Domain Analysis of Signals and Systems

8 g(t)

gδ(t)

(a)
0 t

Ts
–6
8
g(t)

gп(t)

(b)

0 t

–6

Figure 4.14 One cycle of signal g(t) and its (a) Instantaneously sampled signal g𝛿 (t); and (b) Sample-and-hold signal gΠ (t).

Figure 4.14b shows a more realisable sample-and-hold signal gΠ (t) in which the value of g(t) at each sampling
instant is held constant until the next sampling instant, thereby producing a staircase waveform gΠ (t) as shown.
The Fourier series of gΠ (t) is given by Eq. (4.20), repeated below for convenience
{ [ ( )]}
∑
m
∑
∞
dk ∑
k−1
gΠ (t) = Ak dk + 2Ak dk sinc(ndk ) cos 2𝜋nf o t − 2𝜋n + di
k=1 n=1
2 i=1
𝜏k 1
dk = ; fo =
T T
with number of steps m = 16, pulse width 𝜏 k = T s for all k, and hence dk = T s /T = 1/16, Ak = g((k − 1)Ts ) =
g((k − 1)∕16fo ). For example, using Eq. (4.39), A2 is calculated as
Telegram: @Persian_IEEEComSoc

A2 = g(Ts ) = g(1∕16fo )
( ) ( ) ( )
1 1 1
= 1 + 5 sin 2𝜋fo × + 3 sin 4𝜋fo × + 2 sin 8𝜋fo ×
16fo 16fo 16fo
( ) ( ) ( )
𝜋 𝜋 𝜋
= 1 + 5 sin + 3 sin + 2 sin
8 4 2
= 1 + 1.9134 + 2.1213 + 2 = 7.0347

The Fourier series for the sample-and-hold signal gΠ (t) in Figure 4.14b is therefore
{ ( ) ( ) ∞ [ ( ) ]}
( )
1 ∑ k−1 ∑
16
k−1 n 2k − 1
gΠ (t) = g + 2g sinc cos 2𝜋nf o t − n𝜋 (4.40)
16 k=1 16fo 16fo n=1 16 16
 4.2 Fourier Series 237

This equation states that gΠ (t) has DC component

( )
1 ∑
16
k−1
A0 = g (4.41)
16 k=1 16fo

and that the amplitude An and phase 𝜙n of the nth harmonic of gΠ (t) is obtained by adding m sinusoids, each of
frequency nf o , having respective amplitudes An,k and phases 𝜙n,k given by
( ) ( ) ( )
1 k−1 n 1 − 2k
An,k = g sinc ; 𝜙n,k = n𝜋;
8 16fo 16 16
k = 1, 2, 3, · · · , 16 (4.42)
Using the method for sinusoidal addition in Section 2.7.3, it is straightforward to obtain the results tabulated in
Table 4.3 for the frequency components of gΠ (t) up to the fourth harmonic. For ease of comparison, the table also
includes the frequency components of g(t) from Eq. (4.39), with their amplitudes and phases denoted as Ano and
𝜙no merely for distinction from those of gΠ (t). The aperture distortion loss (ADL), defined as 20 log10 (Ano ∕An ) dB,
for each harmonic is also included in the last column of the table. The following vital insight into the effect of
sample-and-hold operation is unveiled.
● The sample-and-hold operation introduces a distortion known as aperture distortion that affects both the ampli-
tude and phase of the frequency components of gΠ (t) when compared to those of the original signal g(t). It may
be seen in the last column of Table 4.3 that the amplitude of the nth harmonic of gΠ (t) is reduced from that of
the corresponding harmonic in g(t) by a factor equal to the ADL, which increases with harmonic frequency nf o .
It will become clear in Chapter 9 that this increase of ADL with frequency is according to the relation
( ( ))
| nf o ||
|
ADL = −20 log10 |sinc | dB (4.43)
| Fs ||
|

Using values from Table 4.3, the synthesised or reconstructed signal gΠs (t), based on the DC and first four har-
monics is
gΠs (t) = 1 + 4.97 cos(2𝜋fo t − 101.25∘ ) + 2.92 cos(4𝜋fo t − 112.5∘ ) + 1.8 cos(8𝜋fo t − 135∘ )
This signal is shown in Figure 4.15 along with the original and sample-and-hold signals g(t) and gΠ (t). The signal
gΠs (t), reconstructed from all the frequency components contained in the sample-and-hold signal gΠ (t) within the

Table 4.3 Harmonic components of signal g(t) compared to those of

sample-and-hold signal gΠ (t) derived by Fourier analysis.
Telegram: @Persian_IEEEComSoc

g𝚷 (t) g(t)

n An 𝝓n Ano 𝝓no ADL = 20 log(Ano /An ), dB

0 1 0 1 0 0
1 4.9679 −101.25 5 −90 0.056
2 2.9235 −112.5 3 −90 0.224
3 0 0 — — —
4 1.8006 −135 2 −90 0.912

The sampling rate employed to produce gΠ (t) is four times the maximum frequency
component of g(t).
 238 4 Frequency Domain Analysis of Signals and Systems

8
Sample-and-hold gп(t)

Original g(t)
Reconstructed gпs(t)

0 t

–6

Figure 4.15 Aperture effect causes a distortion in the signal gΠs (t) that is reconstructed using all the frequency
components contained in the sample-and-hold signal gΠ (t) within the frequency band of the original signal g(t).

bandwidth of the original signal g(t), can be seen to be altered both in amplitude and phase (note, for example,
the discernible delay) when compared to g(t). This distortion is due entirely to aperture effect introduced by the
sample-and-hold operation.
● Employing Eq. (4.42) to obtain more harmonics (n = 4, 5, 6, …) of the sample-and-hold signal gΠ (t) leads to
a more faithful synthesis of gΠ (t), as shown in Figure 4.16 for synthesis of gΠs (t) obtained by adding the DC
and the first 36 harmonics of gΠ (t). Notice that gΠs (t) does not approach closer to g(t) but to gΠ (t). Aperture
distortion cannot be remedied by passing more and more harmonics in the sample-and-hold signal gΠ (t) through
to contribute to the signal reconstruction. What that does is merely better reconstruct the sample-and-hold
waveform, but not the original signal.
● Figure 4.17 shows the single-sided amplitude and phase spectra of gΠ (t) up to the 36th harmonic. We notice
a very interesting feature: the only frequency components in gΠ (t) are the band of frequencies in the original
signal g(t), a baseband we denote as ±f m , plus replications of this band at F s , 2F s , and other integer multiples of
the sampling frequency F s . All frequency components are reduced in amplitude by the ADL of Eq. (4.43), so the
higher replicated bands are more severely attenuated. For example, the amplitude of the frequency component
f o is 5 units in the original signal g(t), whereas within gΠ (t) the amplitude of this component is reduced as

8
Sample-and-hold gп(t)

Original g(t)
Telegram: @Persian_IEEEComSoc

Reconstructed gпs(t)

0 t

–6

Figure 4.16 Signal gΠs (t) synthesised using DC and the first 36 harmonics of sample-and-hold signal gΠ (t).
 4.2 Fourier Series 239

Baseband Fs ± fm 2Fs ± fm
5
1st replicated band 2nd replicated band
Amplitude, An

fo
0 nfo
0 Fs 2Fs
0 nfo
Phase, ϕn (deg)

–45

–90

–135

Figure 4.17 Single-sided amplitude and phase spectra of sample-and-hold signal gΠ (t) of Figure 4.16 up to the 36th
harmonic. Sampling rate F s = 4f max .

follows: in the baseband where its frequency is f o , this component is reduced in amplitude by only 0.056 dB to
4.97 units. In the first replicated band where its frequency is F s + f o , the same component is reduced in amplitude
by 24.665 dB to 0.29 units. And in the second replicated band where its frequency is 2F s + f o , the component is
reduced in amplitude by 30.426 dB to 0.15 units.
● The maximum ADL within the frequency band of g(t) occurs at the maximum frequency component f max of
g(t), and is given by Eq. (4.43) as

ADLmax = −20 log10 (|sinc(fmax ∕Fs )|) dB

Since |sinc(x)| → 1 as x → 0, it means that 20 log(|sinc(x)|) → 0 as x → 0. Therefore, the way to reduce ADL and
hence aperture distortion in sample-and-hold signals is by choosing a sampling rate F s > > f max . To illustrate, the
maximum frequency component in our current example is 4f o , so let us repeat the above exercise using a sampling
rate F s = 160f o . The results are presented in Table 4.4, where it can be seen that the ADL is negligible and phase
shifts are very small, so the waveform reconstructed using the frequency components in the baseband of gΠ (t) will
be closely matched with g(t), as shown in Figure 4.18. This solution to ADL comes at a high price, however, since
it will lead to a significant increase in bandwidth. A minimum sampling rate of 2f max is required to avoid what is

Table 4.4 Harmonic components of signal g(t) compared to those of sample-and-hold signal
Telegram: @Persian_IEEEComSoc

gΠ (t) with sampling rate equal to 40 times the maximum frequency component of g(t).

g𝚷 (t) g(t)

n An 𝝓n Ano 𝝓no ADL = 20 log(Ano /An ), dB

0 1 0∘ 1 0∘ 0
1 4.9997 −91.125∘ 5 −90∘ 0.00056
2 2.9992 −92.245∘ 3 −90∘ 0.00223
3 0 0∘ — — —
4 1.9979 −94.5∘ 2 −90∘ 0.00893
 240 4 Frequency Domain Analysis of Signals and Systems

8
Original, sample-and-hold and
reconstructed signals

0 t

–6

Figure 4.18 Sample-and-hold operation at high sampling rate F s = 40f max . Aperture distortion is negligible and there is a
close match between the original signal g(t) and reconstructed signal.

8
Original g(t)

Ts d = τ/Ts

Flat-top-sampled gп(t)
τ

0 t

–6

Figure 4.19 Flat-top sampling at sampling rate F s = 1/T s using sampling pulses of fractional width d = 1/4.

known as alias distortion, which we explore further in Chapter 9. At five times this minimum (with F s = 10f max ),
the maximum ADL is only 0.14 dB.
● The sample-and-hold operation discussed above is a special case of a more general class of sampling known as
Telegram: @Persian_IEEEComSoc

flat-top sampling. In sample-and-hold, the sampling operation is implemented using full-width sampling pulses
that hold the value of g(t) constant for the entire sampling interval. More generally, fractional-width sampling
pulses may be used that produce gΠ (t) by holding the output constant at g(nT s ) for only a fraction d at the start
of each sampling interval. For the remaining fraction (1 − d) of the sampling interval the output is set to zero.
The result is a flat-top-sampled signal such as shown in Figure 4.19 for d = 1/4 using the analogue signal g(t)
earlier specified in Eq. (4.39). Note therefore that instantaneous sampling (Figure 4.14a) is flat-top sampling in
the limit d → 0, whereas sample-and-hold (Figure 4.14b) is flat-top sampling with d = 1.
The Fourier series of Eq. (4.20) may be applied in a straightforward manner to the flat-top-sampled signal gΠ (t) of
Figure 4.19. In this case, the m steps occur in pairs of amplitudes [g((k − 1)Ts ), 0] with corresponding pulse widths
[dT s , (1 − d)Ts ] for k = 1, 2, 3, …, m, where m = T/T s , T being the period of g(t), and T s the sampling interval.
With this in mind, we may proceed as previously discussed to evaluate the Fourier series of gΠ (t) and obtain a
 4.2 Fourier Series 241

2.5

(a) d = 0.5

0 nfo
fmax Fs 2Fs 3Fs 4Fs
0.5
(b) d = 0.1

0.25

0
BB RB1 RB2 RB3 RB4
0.05
(c) d = 0.01
Fs – fmax Fs + fmax
0.025

0 nfo
fmax Fs 2Fs 3Fs 4Fs

Figure 4.20 Single-sided amplitude spectrum of flat-top-sampled signal at sampling rate F s using sampling pulses of
fractional width d = 0.5, 0.1, and 0.01. (BB ≡ baseband; RB1 ≡ 1st replicated band, etc.)

tabulated list of the amplitude and phase of each of its sinusoidal components up to any desired harmonic n. We
have done this for gΠ (t) produced using sampling rate F s = 4f max and three fractional pulse widths d = 0.5, 0.1,
and 0.01. Figure 4.20 shows the SSAS of gΠ (t) up to the 68th harmonic for each value of d, from which we can see
several interesting features.
We notice (again) that the only frequency components present in a flat-top-sampled signal gΠ (t) are those in
the original analogue signal g(t), referred to as baseband, plus replications of this baseband at regular intervals
F s along the frequency axis. There is a scale factor of d on the amplitude of each harmonic since in this sampling
process the signal is held for only a fraction d of the time. In addition to this scale factor, there is also ADL, as earlier
discussed. However, we now see that the width of the sampling pulse (expressed as a fraction d of the sampling
interval) is a factor in the amount of ADL experienced: ADL reduces as d decreases until it becomes negligible at
d → 0, which corresponds to instantaneous sampling. Our Fourier analysis therefore reveals that flat-top sampling
with d → 0 (or instantaneous sampling) of an analogue signal g(t) at a sampling rate F s produces a sampled signal
Telegram: @Persian_IEEEComSoc

gΠ (t) that contains the entire and undistorted spectrum of g(t) plus replications of this undistorted spectrum at
a regular spacing F s . Provided F s ≥ 2f max so that the first replicated band (which contains frequencies down to
F s − f max ) does not overlap into the baseband of the sampled signal gΠ (t) (which contains frequencies up to f max )
then adding together all the sinusoidal components within the baseband of the sampled signal gΠ (t) will perfectly
reconstruct g(t). This is illustrated in Figure 4.21, where a perfect synthesis of g(t) is achieved with only the DC and
the first four harmonics in the flat-top-sampled signal gΠ (t) at d = 0.01. Note that the synthesised signal must be
multiplied by 1/d to restore it to the scale of the original signal g(t). The situation suggests that this reconstruction
may be achieved simply by putting gΠ (t) through a lowpass filter (LPF) that passes all frequencies up to f max
and blocks all frequencies ≥ F s − f max , thereby blocking all the replicated bands in gΠ (t). Of course, adding all
the sinusoidal components in gΠ (t), including those in the replicated bands, will synthesise gΠ (t), but interest is
usually in reconstructing the original analogue signal g(t) from which the samples were taken.
 242 4 Frequency Domain Analysis of Signals and Systems

8
gп(t) Synthesised
Original g(t)

–6

Figure 4.21 Original signal g(t), flat-top-sampled signal gΠ (t) (sampled at rate of F s = 4f max using sampling pulses of
fractional width d = 0.01), and synthesised signal (obtained by adding only the baseband components in gΠ (t)).

We are now able to state the complete expression for the ADL experienced by a harmonic of frequency nf o in a
flat-top-sampled signal that is generated at a sampling rate F s using sampling pulses of fractional width d
( ( ))
| nf o d ||
|
ADL = −20 log10 |sinc | dB (4.44)
| Fs ||
|
Note that the previously discussed Eq. (4.43) is a special case of the above equation when d = 1. Also, when nf o d
is an integer multiple of F s , then sinc(nf o d/F s ) = 0 and its negative logarithm and hence the ADL in Eq. (4.44) will
be infinitely large. In Figure 4.20a with F s = 16f o and d = 0.5, this condition is satisfied at f = 32f o , 64f o , …; and
in Figure 4.17 with F s = 16f o and d = 1, this condition is satisfied at f = 16f o , 32f o , … It is therefore this infinite
attenuation of certain frequency components due to aperture loss that accounts for the absence of spectral lines
at all integer multiples of F s in Figure 4.17 and at even integer multiples of F s in Figure 4.20a, although in both
cases the DC component in the baseband would have been replicated at these locations.
We earlier discussed and demonstrated in Figure 4.18 a way to reduce ADL by using a large sampling rate F s
and noted that this solution is wasteful in bandwidth. Equation. (4.44) suggests a cheaper alternative solution by
implementing flat-top sampling using fractional pulse width d → 0. This solution is demonstrated in Figure 4.21.
Telegram: @Persian_IEEEComSoc

In flat-top sampling, the maximum ADL within the frequency band of g(t) occurs at the maximum frequency
component f max of g(t), and is given by
( ( ))
| fmax d ||
|
ADLmax = −20 log10 |sinc | dB (4.45)
| Fs |
| |
The worst-case scenario occurs at the minimum allowed sampling rate F s = 2f max when the above expression
reduces to
ADLmax = −20 log10 (|sinc(d∕2)|) dB
So, at d = 1, 0.5, 0.1, and 0.01, ADLmax is respectively 3.9, 0.9, 0.04, and 0.004 dB. This effect is apparent in
the trend of reduction in spectral distortion in Figure 4.17 and Figure 4.20a–c. This would suggest that spectral
distortion due to aperture effect is negligible in flat-top sampling with d ≤ 0.1 and may therefore be ignored.
 4.2 Fourier Series 243

gm(t)
A d = τ/T
m=1
(a)
t
τ
T
A
m=2

(b) t

–A
A
m=3

(c) t

–A
A
m=4

(d) t

–A

Figure 4.22 Sinusoidal pulse train with m half cycles per pulse.

4.2.4.2 Binary ASK Signal and Sinusoidal Pulse Train

Modulation is a fundamental and indispensable signal processing operation in telecommunications and
Chapters 7, 8, and 11 are devoted to its detailed treatment. Here we wish to demonstrate the versatility of the
Fourier series method by applying it to analyse a simple digital modulation scheme to gain valuable insight into
the frequency content (and hence bandwidth) of the signals produced by the scheme.
In BASK, a sequence of data bits is conveyed using a sinusoidal carrier signal by changing or shifting only the
amplitude of the carrier between two values to signal binary 1 or 0. The best performance of the scheme in the
presence of noise is achieved by setting one amplitude state to zero and the other to some nonzero value A, and
this is commonly described as on–off keying (OOK). The highest frequency content (and bandwidth requirement)
Telegram: @Persian_IEEEComSoc

of a BASK signal will be when the data bits follow the fastest-changing sequence of 101010…, which produces
a periodic BASK waveform that may therefore be analysed using the Fourier series method. It is a requirement
of the scheme that the sinusoidal carrier must complete an integer number m of half-cycles in each bit interval,
so the waveform will be a sinusoidal pulse train, as shown in Figure 4.22 for m = 1, 2, 3, 4. The pulse interval 𝜏
corresponds, say, to binary 1, the no-pulse interval corresponds to binary 0, and the duty cycle d = 𝜏/T of the pulse
train will have a value d = 1/2 for the sequence 101010… However, the analysis that follows will be carried out for a
general value of d to allow the results to be applicable to slower-changing regular sequences such as 100100100…,
where d = 1/3, 110110110… with d = 2/3, 10001000… with d = 1/4, etc. Note that in all cases 𝜏 is the bit duration
and therefore the transmission bit rate is
1
Rb = (4.46)
𝜏
 244 4 Frequency Domain Analysis of Signals and Systems

The sinusoidal carrier completes m half-cycles in 𝜏 seconds, or m/2 full cycles per 𝜏 seconds, which means that
carrier frequency f c = m/(2𝜏). Since 𝜏 = dT and T = 1/f o (where f o is the fundamental frequency of the periodic
waveform), the relationships among carrier frequency f c , number m of half-cycles per bit interval 𝜏, bit rate Rb ,
and fundamental frequency f o are given by

m m mf o
fc = = Rb = (4.47)
2𝜏 2 2d
Let us now turn our attention to the important task of analysis of the waveforms gm (t) in Figure 4.22. One cycle
of this pulse train, in the interval 0 ≤ t ≤ T, is defined by

⎧
⎪A sin(2𝜋fc t), 0≤t≤𝜏
gm,T (t) = ⎨ (4.48)
⎪0, elsewhere
⎩

The rms value Arms , mean value (i.e. DC component) Ao , and Fourier coefficients an and bn of this pulse train
may be straightforwardly evaluated as follows from their respective definitions, using f c 𝜏 = m/2 from Eq. (4.47)
and f o 𝜏 = d, where necessary.
T 𝜏
1 1
A2rms = g2 (t)dt = A2 sin2 (2𝜋fc t)dt
T ∫0 mT T ∫0
𝜏
A2 1
= [1 − cos(4𝜋fc t)]dt, since sin2 𝜃 = (1 − cos 2𝜃)
2T ∫0 2

A2 𝜏 sin(4𝜋fc t) |𝜏 A2 sin(4𝜋fc 𝜏)
= − | = d−
2T 4𝜋fc ||0 2 4𝜋fc

A2 sin(2𝜋m)
= d−
2 4𝜋fc

A2
= d, since sin(2𝜋m) = 0
2
The rms value therefore depends neither on frequency f c nor on number of half-cycles m in a pulse, and is given
by
√
Arms = A d∕2 (4.49)
Telegram: @Persian_IEEEComSoc

The condition
√ d = 1 corresponds to a regular sinusoidal signal, in which case the above equation reduces to
Arms = A∕ 2 as expected.
The DC component is evaluated as
T 𝜏
1 1
A0 = g (t)dt = A sin(2𝜋fc t)dt
T ∫0 mT T ∫0

A cos(2𝜋fc t) ||
0
=
T 2𝜋fc ||𝜏
[ ] [ ]
A cos(0) − cos(2𝜋fc 𝜏) A 1 − cos(2𝜋m∕2)
= =
T 2𝜋fc T 2𝜋m∕2𝜏
 4.2 Fourier Series 245

Replacing 𝜏/T by d yields

Ad
A0 = [1 − cos(m𝜋)]
m𝜋
{
0, m even
= (4.50)
2Ad∕m𝜋, m odd

To evaluate the Fourier coefficients, we advance (i.e. shift leftward) each waveform in Figure 4.22 by 𝜏/2 so that
′ (t) in the
the pulse is centred at t = 0. A little thought will show that the main cycle of the resulting waveform gm
interval −T/2 ≤ t ≤ T/2 is now given by the following separate expressions for odd and even m
{
′ (−1)(m−1)∕2 A cos(2𝜋fc t), −𝜏∕2 ≤ t ≤ 𝜏∕2
gm,T (t) = ; m = 1, 3, 5, · · ·
0, elsewhere
{
(−1)m∕2 A sin(2𝜋fc t), −𝜏∕2 ≤ t ≤ 𝜏∕2
= ; m = 2, 4, 6, · · · (4.51)
0, elsewhere
′
Notice that gm,T (t) is an odd function when m is an even integer, whereas it is an even function when m is an
odd integer. Therefore, in view of Eq. (4.16), we need to compute only an when m is odd, and only bn when m is
even. It was of course specifically for the benefit of this simplification that we carried out the time shift. As a side
note, applying the trigonometric identities

sin(𝜃) = cos[𝜃 + (4k + 3)𝜋∕2] ⎫

⎪
⎪
− sin(𝜃) = cos[𝜃 + (4k + 1)𝜋∕2] ⎪
⎬; k = 0, 1, 2, 3, · · · (4.52)
cos(𝜃) = cos[𝜃 + (4k)𝜋∕2] ⎪
⎪
− cos(𝜃) = cos[𝜃 + (4k + 2)𝜋∕2] ⎪
⎭
′
we see that, for all m, gm,T (t) may be represented more simply by the following single expression

⎧ ( )
𝜋
′ ⎪ A cos 2𝜋fc t + (m − 1) 2
, −𝜏∕2 ≤ t ≤ 𝜏∕2
gm,T (t) = ⎨ ; m = 1, 2, 3, · · · (4.53)
⎪0, elsewhere
⎩
Now using Eq. (4.51) in Eq. (4.11) to compute an for odd m yields
𝜏∕2
2
an = (−1)(m−1)∕2 A cos(2𝜋fc t) cos(2𝜋nf o t)dt
T ∫−𝜏∕2
Telegram: @Persian_IEEEComSoc

(−1)(m−1)∕2 A 𝜏∕2
= [cos(2𝜋fc + 2𝜋nf o )t + cos(2𝜋fc − 2𝜋nf o )t]dt
T ∫−𝜏∕2
[ ]
(−1)(m−1)∕2 A 2 sin(2𝜋fc 𝜏∕2 + 2𝜋nf o 𝜏∕2) 2 sin(2𝜋fc 𝜏∕2 − 2𝜋nf o 𝜏∕2)
= +
T 2𝜋fc + 2𝜋nf o 2𝜋fc − 2𝜋nf o
[ ]
(m−1)∕2 𝜏 sin 𝜋(fc 𝜏 + nf o 𝜏) sin 𝜋(fc 𝜏 − nf o 𝜏)
= (−1) A +
T 𝜋(fc 𝜏 + nf o 𝜏) 𝜋(fc 𝜏 − nf o 𝜏)

= (−1)(m−1)∕2 Ad[sinc(nd + m∕2) + sinc(nd − m∕2)], m = 1, 3, 5, · · ·

 246 4 Frequency Domain Analysis of Signals and Systems

Similarly, using Eq. (4.51) in Eq. (4.12) to compute bn for even m yields
𝜏∕2
2
bn = (−1)m∕2 A sin(2𝜋fc t) sin(2𝜋nf o t)dt
T ∫−𝜏∕2
(−1)m∕2 A 𝜏∕2
= [cos(2𝜋fc − 2𝜋nf o )t − cos(2𝜋fc − 2𝜋nf o )t]dt
T ∫−𝜏∕2
[ ]
(−1)m∕2 A 2 sin(2𝜋fc 𝜏∕2 − 2𝜋nf o 𝜏∕2) 2 sin(2𝜋fc 𝜏∕2 + 2𝜋nf o 𝜏∕2)
= −
T 2𝜋fc − 2𝜋nf o 2𝜋fc + 2𝜋nf o
[ ]
𝜏 sin 𝜋(fc 𝜏 − nf o 𝜏) sin 𝜋(fc 𝜏 + nf o 𝜏)
= (−1)m∕2 A −
T 𝜋(fc 𝜏 − nf o 𝜏) 𝜋(fc 𝜏 + nf o 𝜏)
= (−1)m∕2 Ad[sinc(nd − m∕2) − sinc(nd + m∕2)], m = 2, 4, 6, · · ·
To summarise, the Fourier coefficients are given by
}
an = (−1)(m−1)∕2 Ad[sinc(nd − m∕2) + sinc(nd + m∕2)]
, m = 1, 3, 5, · · ·
bn = 0

}
an = 0
, m = 2, 4, 6, · · · (4.54)
bn = (−1)m∕2 Ad[sinc(nd − m∕2) − sinc(nd + m∕2)]
Finally, using the relations provided in Eq. (4.13) to determine the amplitude An and phase 𝜙n of the nth har-
monic based on an and bn , and including specifically Eq. (4.14) to obtain the correct phase of gm (t), bearing in
mind that Eq. (4.54) applies to the time-shifted gm ′ (t), we obtain the following Fourier series for the sinusoidal

pulse trains gm (t) shown in Figure 4.22. Note that we have used Ao given in Eq. (4.50), and that the following
result applies for all m, although Figure 4.22 only shows m up to 4.
∞ [ ( ) ( )]
⎧A + (−1) m−1 ∑
2 Ad sinc nd − m2 + sinc nd + m2 cos(2𝜋nf o t − 𝜋nd), m odd
⎪ 0
n=1
gm (t) = ⎨ m ∑
∞ [ ( ) ( )] ( ( ))
⎪A0 + (−1) 2 Ad sinc nd − m2 − sinc nd + m2 cos 2𝜋nf o t − 𝜋 nd + 12 , m even
⎩ n=1
Ad(1 − cos m𝜋)
A0 = (4.55)
m𝜋
Employing again the trigonometric identities of Eq. (4.52) and the fact that (−1)m = 1 when m is even and
equals −1 when m is odd, we obtain the following single expression for this Fourier series, which applies for all m
(whether even or odd)
∑
∞
gm (t) = A0 + An cos(2𝜋nf o t + 𝜙n );
n=1
[ ( ) ( )]
Telegram: @Persian_IEEEComSoc

m m
An = Ad sinc nd − − (−1)m sinc nd + ;
2 2
𝜋
𝜙n = (m − 1 − 2nd) ;
2
Ad(1 − cos(m𝜋))
A0 = ;
m𝜋
m = 1, 2, 3, 4, · · · (4.56)
Before considering what Eq. (4.56) reveals about BASK, it is useful to highlight a few special cases. The sinusoidal
pulse train g2 (t) of Figure 4.22b corresponds to m = 2 in Eq. (4.56). Making this substitution yields
∑
∞
g2 (t) = Ad [sinc(nd − 1) − sinc(nd + 1)] cos(2𝜋nf o t − 𝜋(nd − 1∕2)) (4.57)
n=1
 4.2 Fourier Series 247

g(t)
A
d = τ/T
(a)

t
–T –T/2 –τ/2 0 τ/2 T/2 T
T

g1(t)
A d = τ/T = 1
(b)

t
τ=T

g1s(t) = DC + 1st three harmonics

A

(c)

0 t

Figure 4.23 (a) Centred cosine-shaped pulse train; (b) Full wave rectifier waveform; (c) Full wave rectifier waveform
synthesised with DC and first three harmonics.

The centred cosine-shaped pulse train g(t) of amplitude A and duty cycle d shown in Figure 4.23a is a special
case of Eq. (4.56) with m = 1 and a time shift to = −𝜏/2 (= − d/2f o ). We substitute m = 1 in Eq. (4.56) and add
2𝜋nf o to = 𝜋nd to the phase to obtain the Fourier series for the centred cosine-shaped pulse train as

2Ad ∑∞
g(t) = + Ad [sinc(nd − 1∕2) + sinc(nd + 1∕2)] cos(2𝜋nf o t) (4.58)
𝜋 n=1

Finally, the full wave rectifier waveform g1 (t) of Figure 4.23b corresponds to Eq. (4.56) with m = 1 and d = 1.
Making these substitutions yields
∞ [ ( ) ( )]
2A ∑ 1 1
g1 (t) = +A sinc n − + sinc n + cos(2𝜋nf o t − n𝜋) (4.59)
𝜋 n=1
2 2
You may wish to verify that the DC and the first three harmonics of this series produce the synthesised waveform
Telegram: @Persian_IEEEComSoc

g1s (t) = 0.6366A − 0.4244A cos(2𝜋fo t) − 0.0849A cos(4𝜋fo t) − 0.0364A cos(6𝜋fo t)

which is sketched in Figure 4.23c and shows a close resemblance to the full wave rectifier waveform.
Now turning our attention to applying Eq. (4.56) to BASK, we note that the waveform of a BASK system (imple-
mented as OOK) for the fastest-changing sequence 101010… has d = 1/2, so Eq. (4.56) gives the spectral values
[ ( ) ( )]
A n−m n+m
An = sinc − (−1)m sinc
2 2 2
𝜋
𝜙n = (m − 1 − n)
2
A(1 − cos m𝜋)
A0 = (4.60)
2m𝜋
 248 4 Frequency Domain Analysis of Signals and Systems

A/2

fo
A/4

Bn = 4fo

A/20

0 20fo 40fo
↑
fc

Figure 4.24 Amplitude spectrum of BASK signal with m = 20.

If, for example, such a system operates at bit rate Rb = 50 kb/s and employs a carrier of frequency f c = 500 kHz, the
BASK waveform (for the fastest-changing sequence) will have 𝜏 = 1/Rb = 20 μs, period T = 2𝜏 = 40 μs, fundamental
frequency f o = 1/T = 25 kHz, and (from Eq. (4.47)) m = 2df c /f o = 20, so that f c = 20f o . Substituting m = 20 in
Eq. (4.60) yields the amplitude and phase of any desired harmonic n.
Figure 4.24 shows the amplitude spectrum of this BASK signal up to the 40th harmonic We see that the main
lobe of the spectrum is centred on f c with a null bandwidth Bn = 4f o (which is 4/T = 2/𝜏 = 2Rb ). The required band-
width for transmitting at bit rate Rb using BASK is therefore 2Rb . It turns out that other values of m also produce
similar results, namely a bandwidth 2Rb centred on f c , with a maximum amplitude at the carrier frequency. Phase
spectrum is not shown in Figure 4.24, but you may wish to check that Eq. (4.60) gives respective values of 0∘ , −90∘ ,
and −180∘ for harmonics n = 19, 20 (i.e. the carrier), and 21 used below. A BASK signal is usually passed through
an appropriate bandpass filter having a passband centred at f c to significantly reduce the amplitudes of remnant
frequency components outside the main lobe. Synthesising the BASK waveform using only the three frequency
components within the main lobe (i.e. bandwidth of 2Rb centred on f c ) produces the signal
gs (t) = 0.3265A cos(38𝜋fo t) + 0.5A cos(40𝜋fo t − 90∘ ) − 0.3105A cos(42𝜋fo t)
which is sketched in Figure 4.25 along with the original OOK waveform for comparison. The signal gs (t) is clearly
adequate for the detection of the data sequence at the receiver. Since the identified bandwidth is enough for the
transmission of the fastest-changing BASK sequence, it is also enough for all other slower-changing sequences.

4.2.4.3 Trapezoidal Pulse Train

The trapezoidal pulse was introduced in Section 2.6 as a useful waveform shape for approximating realisable labo-
Telegram: @Persian_IEEEComSoc

ratory pulses because of its separate intervals of rising, constant, and falling edges. It is also the general pulse shape
from which other popular pulses, such as rectangular, triangular, sawtooth, ramp, etc., are derived as special cases.
The conditions for each of these special cases are specified in Eq. (2.27). The rms value of the trapezoidal pulse
train is stated in Eq. (3.111) and derived in Worked Example 3.5. Here we wish to evaluate the Fourier series of
the trapezoidal pulse train for the simple reason of providing a set of equations from which the Fourier series of
a wide range of related waveforms may be quickly extracted without the need for fresh calculations starting from
first principles.
Figure 4.26 shows a unipolar trapezoidal pulse train g(t) of amplitude A and period T that is centred on the
constant portion of the pulse. The pulse train is expressed as the sum of three periodic signals g1 (t), g2 (t), and g3 (t)
of the same amplitude and period. The Fourier series of g(t) will be the sum of the series of these three component
signals. We wish to determine the required Fourier coefficients.
 4.2 Fourier Series 249

Bit interval
A

OOK 0

–A

gs(t) 0

–A
1/fo

Figure 4.25 OOK signal and gs (t) synthesised using three harmonics centred on carrier.

g(t) = g1(t) + g2(t) + g3(t)

τ = τr + τc + τf
g(t) A d = τ/T = dr + dc + df

τr τc τf t
τ
T

g1(t) A dc = τ c / T

τc t

A df = τf / T
g2(t)
t
τf
Telegram: @Persian_IEEEComSoc

g3(t) A dr = τr / T
τr
t
–T/2 –τc/2 τc/2 T/2

Figure 4.26 Trapezoidal pulse train g(t) expressed as a sum of three periodic signals.
 250 4 Frequency Domain Analysis of Signals and Systems

The DC component of g(t) is given by the area of the pulse divided by the period of the waveform. Thus
( )
1 1 1
A0 = A𝜏r + A𝜏f + A𝜏c
T 2 2
= A(dr ∕2 + df ∕2 + dc ) (4.61)

where dr , df , and dc are the respective duty cycles of the rising, falling, and constant pulse edges as defined in
Figure 4.26.
The component signal g1 (t) is a centred unipolar RPT whose Fourier coefficients are derived in Worked
Example 4.1. The results are

an1 = 2Adc sinc(ndc )

bn1 = 0 (4.62)

Note that we have introduced an extra subscript to identify the signal to which the coefficients pertain.
Next, we consider the component signal g2 (t) for which both the cosine and sine coefficients an2 and bn2 must
be calculated since g2 (t) is neither even nor odd. An equation for the waveform of g2 (t) within one cycle (−T/2,
T/2) may be determined by seeking an expression in the form y = 𝛾 + 𝛽t, where 𝛽 is the slope and 𝛾 is the y axis
intercept (i.e. value at t = 0). To determine 𝛽 and 𝛾, note that the pulse of g2 (t) falls from a value A at t = 𝜏 c /2 to 0 in
𝜏 f seconds, so its slope is −A/𝜏 f , which means that at t = 0 its value was higher than A by 𝜏 c /2 × A/𝜏 f . Therefore,
the expression for g2 (t) within one cycle (−T/2, T/2) is
( )
⎧ 𝜏c t 𝜏c 𝜏
⎪A 1 + − , ≤ t ≤ 𝜏f + c
g2,T (t) = ⎨ 2𝜏f 𝜏f 2 2 (4.63)
⎪0,
⎩ Otherwise

Substituting this expression in Eq. (4.11) allows us to evaluate an2 as follows

( )
𝜏f +𝜏c ∕2
2
T∕2
2 𝜏c t
an2 = g (t) cos(2𝜋nf o t)dt = A 1+ − cos(2𝜋nf o t)dt
T ∫−T∕2 3,T T ∫𝜏c ∕2 2𝜏f 𝜏f
( )
𝜏f +𝜏c ∕2 𝜏f +𝜏c ∕2
2A 𝜏c 2A
= 1+ cos(2𝜋nf o t)dt − t cos(2𝜋nf o t)dt
T 2𝜏f ∫𝜏c ∕2 T𝜏f ∫𝜏c ∕2

The first integral on the right-hand side is straightforward and the second is evaluated in Worked Example 4.1
using integration by parts. Taking each integral in turn
( ) ( )
Telegram: @Persian_IEEEComSoc

2A 𝜏 𝜏f +𝜏c ∕2
2A 𝜏c sin(2𝜋nf o t) |𝜏f +𝜏c ∕2
1+ c cos(2𝜋nf o t)dt = 1+ |
T 2𝜏f ∫𝜏c ∕2 T 2𝜏f 2𝜋nf o ||𝜏c ∕2
[ 𝜏f +𝜏c ∕2
]
2A t sin(2𝜋nf o t) cos(2𝜋nf o t) ||
𝜏f +𝜏c ∕2
2A
t cos(2𝜋nf o t)dt = + |
T𝜏f ∫𝜏c ∕2 T𝜏f 2𝜋nf o (2𝜋nf o )2 ||𝜏 ∕2
c

Substituting the integral limits for t and simplifying (with a little patience, using when necessary f o T = 1,
f o 𝜏 f = df , f o 𝜏 c = dc , sin(𝜋x)/𝜋x ≡ sinc(x), and the trigonometric identities cos(A + B) ≡ cos A cos B − sin A sin B
and 1 − cos2A ≡ 2 sin2 A) leads to

an2 = Adf cos(n𝜋dc ) sinc2 (ndf ) + Adc sinc(ndc )[sinc(2ndf ) − 1] (4.64)

 4.2 Fourier Series 251

Next, to determine the sine coefficient bn2 of g2 (t), we use Eq. (4.63) in Eq. (4.12)
( )
𝜏f +𝜏c ∕2
2 𝜏c t
bn2 = A 1+ − sin(2𝜋nf o t)dt
T ∫𝜏c ∕2 2𝜏f 𝜏f
( )
𝜏f +𝜏c ∕2 𝜏f +𝜏c ∕2
2A 𝜏c 2A
= 1+ sin(2𝜋nf o t)dt − t sin(2𝜋nf o t)dt
T 2𝜏f ∫𝜏c ∕2 T𝜏f ∫𝜏c ∕2
( ) [ ]|𝜏f +𝜏c ∕2
A 𝜏c 𝜏c ∕2 A sin(2𝜋nf o t) |
= 1+ cos(2𝜋nf o t)|𝜏 +𝜏 ∕2 − − t cos(2𝜋nf o t) |
n𝜋 2𝜏f n𝜋𝜏f 2𝜋nf o |
|𝜏c ∕2
f c

Again, following the same simplification steps used for an2 leads to
A
bn2 = Adf sin(n𝜋dc ) sinc2 (ndf ) + cos(n𝜋dc )[1 − sinc(2ndf )] (4.65)
n𝜋
We do not need to calculate the Fourier coefficients for the remaining component signal g3 (t) from first principles
if we observe that g3 (t) is a time-reversed version of g2 (t) with 𝜏 f (and hence df ) replaced by 𝜏 r (and dr ). We know
from Eq. (4.15) that time reversal alters only the phase spectrum by a factor of −1, which is a complex conjugation
operation achieved by multiplying the sine coefficient bn by −1 while leaving the cosine coefficient an unchanged.
Thus

an3 = Adr cos(n𝜋dc ) sinc2 (ndr ) + Adc sinc(ndc )[sinc(2ndr ) − 1]

A
bn3 = cos(n𝜋dc )[sinc(2ndr ) − 1] − Adr sin(n𝜋dc ) sinc2 (ndr ) (4.66)
n𝜋
The cosine and sine coefficients of the trapezoidal pulse train g(t) is the sum of the respective coefficients derived
above for the component signals. That is

an = an1 + an2 + an3

bn = bn1 + bn2 + bn3 (4.67)

Therefore, including the DC component, the Fourier series of a trapezoidal pulse train has the coefficients
A
A0 = (d + 2dc + df )
2 r
an = A cos(𝜋ndc )[dr sinc2 (ndr ) + df sinc2 (ndf )]
+ Adc sinc(ndc )[sinc(2ndr ) + sinc(2ndf )]
A
bn = cos(𝜋ndc )[sinc(2ndr ) − sinc(2ndf )]
n𝜋
+ A sin(𝜋ndc )[df sinc2 (ndf ) − dr sinc2 (ndr )] (4.68)
Telegram: @Persian_IEEEComSoc

This is a very important and versatile result. It is left as an exercise for you to show that Eq. (4.17) for the unipolar
RPT and Eq. (4.19) for the unipolar triangular pulse train follow straightforwardly from this result when the right
conditions are placed on the duty cycles dr , dc , and df . Further practice is provided in the end-of-chapter questions
to use this result to quickly derive the Fourier series of other trapezoid-related waveforms and pulse trains. This
saves the effort required to solve from first principles. Here is a quick worked example to get you started.

Worked Example 4.6 Fourier Series of a Periodic Ramp Waveform

We wish to use Eq. (4.68) to quickly derive the Fourier series of the periodic ramp waveform x(t) shown in
Figure 4.27a and to test the correctness of our derived series by synthesising x(t) using the DC and the first three
harmonics of the series.
 252 4 Frequency Domain Analysis of Signals and Systems

x(t), V
100

(a)

t, μs
–30 –20 –10 0 10 20 30

100
(b) xs(t)
= DC + first
three harmonics
0 t, μs
–30 –20 –10 0 10 20 30

100
(c) Synthesis
with DC + first
20 harmonics
0 t, μs
–30 –20 –10 0 10 20 30

Figure 4.27 Worked Example 4.6: Periodic ramp waveform.

This waveform is a special case of the trapezoidal pulse train in Figure 4.26 with A = 100, 𝜏 r = 𝜏 c = 0, and
𝜏 f = 𝜏 = T = 10 μs; so that dr = dc = 0, df = 𝜏/T = 1, and f o = 1/T = 100 kHz. Introducing these conditions into
the expressions for the Fourier coefficients of a trapezoidal pulse train given in Eq. (4.68) yields
100
A0 = d = 50 × 1
2 f
= 50

an = A cos(0)[0 + df sinc2 (ndf )] + 0

= sinc2 (n)

=0

100
bn = cos(0)[sinc(2ndr ) − sinc(2ndf )] + 0
n𝜋
100
= [sinc(0) − sinc(2n)]
Telegram: @Persian_IEEEComSoc

n𝜋
100
=
n𝜋
This condition (an = 0, bn > 0) corresponds to the third condition in the bottom half of Figure 4.2, from which it
follows that An = |bn | = 100/n𝜋, and 𝜙n = −90∘ . The required Fourier series of the periodic ramp waveform x(t) is
thus
∑∞
x(t) ≡ A0 + An cos(2𝜋nf o t + 𝜙n )
n=1
∑
∞
100
= 50 + cos(2𝜋nf o t − 90∘ ), fo = 100 kHz (4.69)
n=1
n𝜋
 4.3 Fourier Transform 253

We see that x(t) contains a DC component of 50 V and has an amplitude spectrum that is inversely proportional
to harmonic number n, with every harmonic component having the same phase of −90∘ .
It is always advisable to ascertain the correctness of a derived Fourier series by summing the first few terms to
compare the synthesised waveform with the original. Doing so here and summing the DC and first three harmonics
on the right-hand side of the above equation yields the synthesised waveform
xs (t) = 50 + 31.83 sin(2𝜋nf o t) + 15.92 sin(4𝜋nf o t) + 10.61 sin(6𝜋nf o t);
fo = 100 kHz
which is sketched in Figure 4.27b and compares well with x(t). The derived series is therefore correct. Adding
more harmonics (not required) would of course produce a closer fit as demonstrated in Figure 4.27c by summing
the DC and the first 20 harmonics.

4.3 Fourier Transform

The Fourier transform (FT) is simply an extension of the Fourier series to nonperiodic signals and will exist if a
signal satisfies the Dirichlet conditions earlier stated for Fourier series. In fact, if g(t) is an energy signal, which
means that
∞
E= |g(t)|2 dt < ∞
∫−∞
then its FT will exist. However, note that the Dirichlet condition regarding integrability is more relaxed and simply
requires that g(t) be absolutely integrable, which means that it should satisfy the condition
∞
|g(t)|dt < ∞ (4.70)
∫−∞
Figure 4.28 illustrates to scale the gradual evolution of a signal from being a periodic waveform gT (t) to a non-
periodic pulse g(t). The amplitude spectrum of the signal at each stage is also shown, and we see how the spacing
of spectral lines gradually reduces until the spectrum eventually ceases to be discrete. Let us now formalise these
observations and derive an expression for the FT of nonperiodic signals.
Recall from Eqs. (4.23) and (4.24) that a periodic waveform gT (t) may be expressed as
∑
∞
gT (t) = Cn ej2𝜋nf o t (i)
n=−∞
T∕2
1 1
Cn = g(t)e−j2𝜋nf o t dt, fo = (ii) (4.71)
T ∫−T∕2 T
Telegram: @Persian_IEEEComSoc

which means that gT (t) contains one complex exponential of magnitude |Cn | in each frequency interval of size
f o centred at nf o , for n = …, −3, −2, −1, 0, 1, 2, 3, … As shown in Figure 4.28, a nonperiodic waveform g(t)
arises from gT (t) in the limit T → ∞. In this limit, f o = 1/T becomes infinitesimally small and is denoted df ; the
discrete frequencies nf o become a continuum of frequencies, denoted f ; and g(t) therefore contains a continuum of
complex exponentials. The FT of g(t), denoted G(f ), is the coefficient (having magnitude |G(f )| and angle ∠G(f ))
of the totality of complex exponentials contained in g(t) per unit frequency. In other words, the amplitude |G(f )|
of the FT of g(t) equals the amplitude of the complex exponential component of g(t) lying in an infinitesimally
small frequency interval −df ∕2 ≤ f ≤ df ∕2 divided by the size df of the interval. And the angle ∠G(f ) of the FT is
the phase of that complex exponential. This definition of the FT implies that
Cn
G(f ) = lim = lim Cn T (4.72)
T→∞ f0 T→∞
 254 4 Frequency Domain Analysis of Signals and Systems

Waveform: Amplitude spectrum:

gT(t), T = τ

fo = 1/T
t
τ nfo →
gT(t), T = 5τ

t
τ T nfo →

gT(t), T = 20τ

nfo →
g(t) = gT(t), T → ∞

t
τ –2/τ f→ 2/τ

Figure 4.28 Evolution of a signal from periodic to nonperiodic.

But as T → ∞, f o → df , so we may also express the relationship between Cn and G(f ) as

Cn
G(f ) = ; ⇒ Cn = G(f )df (4.73)
df
Substituting Eq. (4.71)(ii) for Cn into Eq. (4.72), we obtain an expression for G(f ) as
[ T∕2 ]
1 −j2𝜋nf o t
G(f ) = lim Cn T = lim T g(t)e dt
T→∞ T→∞ T ∫−T∕2
∞
= g(t)e−j2𝜋ft dt (4.74)
∫−∞
Given the time domain representation g(t) of a nonperiodic signal, Eq. (4.74) provides a means of determining
its FT G(f ), which is the frequency domain representation of the signal. G(f ) is in general a complex number. Its
magnitude |G(f )| gives the amplitude spectrum of g(t), and its angle ∠G(f ) gives the phase spectrum of g(t), denoted
𝜙g (f ). Note that G(f ) has units of amplitude spectral density, which means that if the signal g(t) is measured in volts
Telegram: @Persian_IEEEComSoc

then its FT will have units of volts per hertz (V/Hz).

To establish a means of deriving the time domain signal g(t) from a frequency domain specification G(f ) of the
signal, a process known as inverse Fourier transform (IFT), we note that in the limit T → ∞, the periodic signal gT (t)
in Eq. (4.71)(i) becomes the nonperiodic signal g(t), the coefficient Cn in that equation becomes G(f )df , and the
summation in the same equation of a discrete set of complex exponentials of frequency nf o becomes integration
of a continuum of complex exponentials of frequency f . Thus
∑
∞
g(t) = lim gT (t) = lim Cn ej2𝜋nf o t
T→∞ T→∞
n=−∞
∞
= G(f )ej2𝜋ft df (4.75)
∫−∞
 4.3 Fourier Transform 255

Equation (4.74) gives the FT of g(t), whereas Eq. (4.75) gives the IFT of G(f ). The two functions g(t) and G(f ) are
said to form a FT pair, a relationship often denoted as

g(t) ⇌ G(f ) (4.76)

which states that the function on the right-hand side is the FT of the left-hand side; and that the left-hand side
is the IFT of the right-hand side. Be careful, however, to avoid the not uncommon mistake of writing g(t) = G(f ).
The operations of Fourier and IFTs are also sometimes expressed as

G(f ) = F[g(t)]
g(t) = F−1 [G(f )] (4.77)

Worked Example 4.7 Fourier Transform Derived from First Principles

We wish to determine the FT of the following pulses

(a) g(t) = Arect(t∕𝜏),

( ) shown ( in
) Figure 4.29a
(b) g1 (t) = cos m𝜋
𝜏
t rect t
𝜏
, m = 3, shown in Figure 4.30a

and to sketch their amplitude spectra.

g(t) = Arect(t/τ)

(a) A

t
–τ/2 τ/2

|G( f )|
Aτ

Bn = 1/τ
(b)

0 f
–5/τ –4/τ –3/τ –2/τ –1/τ 0 1/τ 2/τ 3/τ 4/τ 5/τ
Telegram: @Persian_IEEEComSoc

∠G(f), deg
180

(c) 0 f

–180
–5/τ –4/τ –3/τ –2/τ –1/τ 0 1/τ 2/τ 3/τ 4/τ 5/τ

Figure 4.29 Worked Example 4.7: Rectangular pulse and its amplitude and phase spectra.
 256 4 Frequency Domain Analysis of Signals and Systems

m t
A g1(t) = cos τ t rect τ ,m=3
(a)
–τ/2 τ/2
t

–A τ
(b)
0.509Aτ

0.25Aτ

0 f
–5/τ –4/τ –3/τ –2/τ –1/τ 0 1/τ 2/τ 3/τ 4/τ 5/τ

Figure 4.30 Worked Example 4.7: Sinusoidal pulse and its amplitude spectrum.

(a) The given signal g(t) is a unipolar rectangular pulse of amplitude A and duration 𝜏 centred at the origin, as
shown in Figure 4.29a. A straightforward evaluation of Eq. (4.74) using the given specification of g(t) yields
G(f ) as follows
∞
G(f ) = g(t)e−j2𝜋ft dt
∫−∞
{
𝜏∕2
−j2𝜋ft A, −𝜏∕2 ≤ t ≤ 𝜏∕2
= Ae dt, since g(t) =
∫−𝜏∕2 0, otherwise
𝜏∕2
e−j2𝜋ft |
| A j2𝜋f 𝜏∕2
=A = [e − e−j2𝜋f 𝜏∕2 ]
−j2𝜋f ||−𝜏∕2 j2𝜋f
A
= 2j sin(2𝜋f 𝜏∕2), (using Euler’s formula)
j2𝜋f
( )
Telegram: @Persian_IEEEComSoc

sin(𝜋f 𝜏) 𝜏
= A𝜏 , multiplying by
𝜋f 𝜏 𝜏
= A𝜏 sinc(f 𝜏)

The amplitude spectrum |G(f )| is sketched in Figure 4.29b and the phase spectrum 𝜙g (f ) = ∠G(f ) in
Figure 4.29c. We see that a rectangular pulse of duration 𝜏 has a sinc-shaped amplitude spectrum with nulls
at f = ±1/𝜏, ±2/𝜏, ±3/𝜏, … In this case in which the pulse is centred on the y axis, the spectrum is entirely
real, so the phase spectrum is 0∘ when G(f ) is positive and is ±180∘ when G(f ) is negative, which happens in
alternate lobes of the spectrum. Notice that the amplitude spectrum is an even function of frequency, whereas
the phase spectrum is an odd function of frequency, and this feature will always be the case whenever the
signal g(t) is real.
 4.3 Fourier Transform 257

(b) The given signal g1 (t) is a sinusoidal pulse of amplitude A and duration 𝜏 centred at the origin and completing
an odd number m of half-cycles within the interval 𝜏, as shown in Figure 4.30a. In this case m = 3, but we will
derive the FT expression for a general m = 1, 3, 5, 7, …
Substituting the given specification of g1 (t) in Eq. (4.74) and using Euler’s formula to expand the integral yields

∞ 𝜏∕2 ) (
m𝜋
G1 (f ) = g1 (t)e−j2𝜋ft dt = t e−j2𝜋ft dt
A cos
∫−∞ ∫−𝜏∕2 𝜏
𝜏∕2 ( ) 𝜏∕2 ( )
m𝜋 m𝜋
=A cos t cos(2𝜋ft)dt − jA cos t sin(2𝜋ft)dt
∫−𝜏∕2 𝜏 ∫−𝜏∕2 𝜏

The second integral is zero (since the integrand is an odd function) and the first integrand is even so the integral
may be doubled and evaluated in the half-interval (0, 𝜏/2) to obtain
𝜏∕2 ( )
m𝜋
G1 (f ) = 2A cos t cos(2𝜋ft)dt
∫0 𝜏
𝜏∕2
=A [cos((2𝜋f + m𝜋∕𝜏)t) + cos((2𝜋f − m𝜋∕𝜏)t)]dt
∫0
[ ] 𝜏∕2
sin((2𝜋f + m𝜋∕𝜏)t) sin((2𝜋f − m𝜋∕𝜏)t) ||
=A + |
2𝜋f + m𝜋∕𝜏 2𝜋f − m𝜋∕𝜏 |
[ ] |0
sin(𝜋(f 𝜏 + m∕2)) sin(𝜋(f 𝜏 − m∕2))
=A +
𝜋(2f + m∕𝜏) 𝜋(2f − m∕𝜏)
Multiplying the top and bottom of the right-hand side by 𝜏/2 allows us to introduce the sinc function and obtain
the result
𝜏
G1 (f ) = A [sinc(f 𝜏 + m∕2) + sinc(f 𝜏 − m∕2)]
2
The amplitude spectrum |G1 (f )| is shown in Figure 4.30b for m = 3. The phase spectrum ∠G1 (f ) (not shown)
is 0∘ and 180∘ in alternate lobes, starting with 0∘ in the highest-amplitude lobe.

4.3.1 Properties of the Fourier Transform

The FT has several properties which are often exploited in communication system analysis and to reduce the
computations needed to derive the FT of related signals. It is important that you familiarise yourself with these
features and properties which we summarise below, mostly without proof, along with examples showing how
some of them are used in practice. In what follows, it is given that
g(t) ⇌ G(f )
Telegram: @Persian_IEEEComSoc

g1 (t) ⇌ G1 (f )
g2 (t) ⇌ G2 (f )
and that a, a1 , a2 , and to are constants.

4.3.1.1 Even and Odd Functions

Using Euler’s formula, the FT may be expressed in terms of sine and cosine functions as
∞
G(f ) = g(t)e−j2𝜋ft dt
∫−∞
∞ ∞
= g(t) cos(2𝜋ft)dt − j g(t) sin(2𝜋ft)dt
∫−∞ ∫−∞
 258 4 Frequency Domain Analysis of Signals and Systems

Referring to the properties of even and odd functions listed in Section 2.5.5, and noting that the cosine function
is even whereas the sine function is odd, we see that the first integral will be zero when g(t) is an odd function of
time, and similarly the second integral will be zero when g(t) is even. Thus
{ ∞
2 ∫0 g(t) cos(2𝜋ft)dt, g(t) even
G(f ) = ∞ (4.78)
−2j ∫0 g(t) sin(2𝜋ft)dt, g(t) odd
This result indicates that:

● The FT of an even function of time is a real function of frequency, with its phase spectrum therefore limited in
values to only ∠G(f ) = 0∘ or ± 180∘ or both.
● The FT of an odd function of time is an imaginary function of frequency, with its phase spectrum therefore
limited in values to only ∠G(f ) = ±90∘ .

4.3.1.2 Linearity
a1 g1 (t) + a2 g2 (t) ⇌ a1 G1 (f ) + a2 G2 (f ) (4.79)

Equation (4.79) indicates that the FT obeys the principle of superposition. This means that FT is a linear operator
whose output in the presence of multiple simultaneous inputs is the result of treating each input as if they were
present alone and then adding their individual outputs together.

4.3.1.3 Time Shifting

g(t − to ) ⇌ G(f ) exp(−j2𝜋fto ) = |G(f )|∠[∠G(f ) − 2𝜋fto ] (4.80)

This indicates that the time domain action of delaying a signal g(t) by time to does not alter the signal’s amplitude
spectrum in any way; but it adds −2𝜋fto to the phase spectrum of g(t).

4.3.1.4 Frequency Shifting

g(t) exp(j2𝜋fc t) ⇌ G(f − fc ) (4.81)

Therefore, multiplying a signal by exp(j2𝜋f c t) in the time domain has the effect in the frequency domain of trans-
lating its entire spectrum rightward by f c along the frequency axis. Similarly, multiplying by exp(−j2𝜋f c t) will
translate the spectrum leftward through f c along the frequency axis. Furthermore, since
1
cos(2𝜋fc t) = [exp(j2𝜋fc t) + exp(−j2𝜋fc t)]
2
it follows from Eq. (4.81) and the linearity property of Eq. (4.79) that
Telegram: @Persian_IEEEComSoc

1 1
g(t) cos(2𝜋fc t) ⇌ G(f − fc ) + G(f + fc ) (4.82)
2 2
The effect of multiplication by cos(2𝜋f c t), an operation carried out, for example, in a mixer circuit, is therefore
to translate the signal’s spectrum by ±f c along the frequency axis, applying a scale factor of one-half in the process.

4.3.1.5 Time Scaling

1
g(at) ⇌ G(f ∕a) (4.83)
|a|
This indicates that compressing a signal in the time domain (e.g. by shortening its duration) causes its spectrum
to expand by the same factor, and vice versa. This is a statement of the inverse relationship that exists between the
description of a signal in the time and frequency domains.
 4.3 Fourier Transform 259

4.3.1.6 Time Reversal

g(−t) ⇌ G(−f ) (4.84)
Equation (4.84) states that reversal in the time domain corresponds to reversal in the frequency domain. We know
that if g(t) is real then its amplitude spectrum is even – and hence unchanged by a reversal operation – whereas
its phase spectrum is odd and therefore changed in sign by a reversal operation. Thus, time reversal does not alter
the amplitude spectrum of a real signal but multiplies its phase spectrum by a factor of −1.

4.3.1.7 Complex Conjugation

g∗ (t) ⇌ G∗ (−f ) (4.85)
Thus, complex conjugation in the time domain is equivalent to complex conjugation plus reversal (i.e. flipping
about f = 0) in the frequency domain. If g(t) is real then complex conjugation has no effect on the signal in the
time domain and therefore no effect in the frequency domain, so the spectrum of g(t) and the spectrum of g* (t)
given above must be identical. Similarly, if g(t) is real then g(−t) = g* (−t), which means that the spectrum of
g(−t) given in Eq. (4.84) as G(−f ) must be identical with the spectrum of g* (−t), which, according to Eq. (4.85), is
obtained by complex conjugation and reversal of G(−f ) to produce G* (f ). In summary

G∗ (−f ) = G(f ) ⎫
⎪
G(−f ) = G∗ (f ) ⎬ , g(t) real (4.86)
g(−t) ⇌ G∗ (f )⎪ ⎭

4.3.1.8 Duality
G(t) ⇌ g(−f ) (4.87)
Equation (4.87) states that if a signal g2 (t) = G(t) is identical in shape to the spectrum of g(t), then the spectrum
of g2 (t) will be G2 (f ) = g(−f ), having a shape that is identical to a time-reversed version of g(t). For example,
we established in Worked Example 4.7 that a rectangular pulse has a sinc spectrum. It follows from this duality
property that a sinc pulse will have a rectangular spectrum. To be more specific, since
( )
t
rect ≡ g(t) ⇌ 𝜏 sinc(f 𝜏) ≡ G(f )
𝜏
Equation (4.87) implies that
G(t) = 𝜏 sinc(t𝜏) ⇌ g(−f ) = rect(−f ∕𝜏)
1
sinc(t𝜏) ⇌ rect(f ∕𝜏), (since rect() is even)
𝜏 ( )
1 f
sinc(2Bt) ⇌ rect
2B 2B
Telegram: @Persian_IEEEComSoc

Thus, a sinc pulse with nulls at intervals of t = 1/2B has a brick wall spectrum of bandwidth equal to B.

4.3.1.9 Differentiation
d
g(t) ⇌ j2𝜋fG(f )
dt
n
d
g(t) ⇌ (j2𝜋f )n G(f ) (4.88)
dtn
Differentiation in the time domain is therefore a form of high-pass filtering operation. It boosts the high-frequency
components by multiplying the amplitude spectrum by a factor of 2𝜋f , which increases with frequency. This rela-
tively reduces the amplitudes of the low-frequency components. Furthermore, the factor j indicates that the phase
of each frequency component is advanced by 90∘ .
 260 4 Frequency Domain Analysis of Signals and Systems

4.3.1.10 Integration
t
G(f ) G(0)
g(𝜆)d𝜆 ⇌ + 𝛿(f ) (4.89)
∫−∞ j2𝜋f 2
Integration in the time domain is therefore a lowpass filtering operation. The amplitude of each frequency
component is multiplied by a factor of 1/2𝜋f , which is inversely proportional to frequency. This attenuates the
high-frequency components relative to the low-frequency components. The phase of each component is also
reduced by 90∘ .

4.3.1.11 Multiplication
∞
g1 (t)g2 (t) ⇌ G1 (𝛾)G2 (f − 𝛾)d𝛾 = G1 (f ) ∗ G2 (f ) (4.90)
∫−∞
Multiplication of two signals in the time domain therefore corresponds to convolution of their FTs in the frequency
domain.

4.3.1.12 Convolution
g1 (t) ∗ g2 (t) ⇌ G1 (f )G2 (f ) (4.91)

The convolution of two signals or functions in the time domain corresponds to the multiplication of their FTs
in the frequency domain. Multiplication being a simpler operation than convolution, this property is exploited
extensively in systems analysis. For example, we saw in Section 3.6.1 that the output y(t) of a linear time invariant
(LTI) system is the convolution of the input signal x(t) with the system’s impulse response h(t). This property
indicates that, in analysing an LTI system to determine its output, we may sidestep the convolution operation by
adopting the following alternative analysis steps

1. Determine the FT of x(t): x(t) ⇌ X(f )

2. Determine the FT of h(t): h(t) ⇌ H(f )
3. Multiply the two FTs: X(f )H(f ) = Y (f )
4. The desired output y(t) is the IFT of this product: y(t) = F−1 [Y (f )].

These steps might seem long-winded. However, FTs are extensively tabulated and therefore the above steps are
often faster than undertaking the convolution operation. We return to this method of analysis later in this chapter.
The convolution property may also be applied as follows to show that convolving a signal g(t) with the impulse
function 𝛿(t) leaves the signal unchanged. Since F[𝛿(t)] = 1, it follows by the convolution property that

g(t) ∗ 𝛿(t) ⇌ G(f ) × 1 = G(f )

which states that the FT of g(t) ∗ 𝛿(t) is G(f ) and hence that
Telegram: @Persian_IEEEComSoc

g(t) ∗ 𝛿(t) = g(t) (4.92)

4.3.1.13 Areas
∞
g(0) = G(f )df (a)
∫−∞
∞
G(0) = g(t)dt (b) (4.93)
∫−∞
The total area under the spectrum of a signal gives the value of the signal at time t = 0, whereas the total area
under the waveform of the signal gives the value of the spectrum of the signal at f = 0.
 4.3 Fourier Transform 261

Given: t
g(t) = A rect τ ⇌ G(f) = Aτ sinc (fτ)

(a) A
t
–τ/2 τ/2
Desired:

g1(t) ⇌ G1(ƒ)? g2(t) ⇌ G2(ƒ)? g3(t) = A trian(t/τ) ⇌ G3(f)?

A A

τ τ/2
t t t
–τ/2 –τ/2 τ/2

–A –A

(b) (c) (d)

Figure 4.31 Worked Example 4.8.

4.3.1.14 Energy
∞ ∞
|g(t)|2 dt = |G(f )|2 df (4.94)
∫−∞ ∫−∞
This is the so-called Parseval’s theorem (also known as Rayleigh’s energy theorem) which allows the same integration
process to be used to calculate signal energy in either domain. Each integral must be finite, implying that g(t) is
an energy signal.

Worked Example 4.8 Application of Fourier Transform Properties

Given that

A rect(t∕𝜏) ⇌ A𝜏 sinc(f 𝜏)

we wish to apply some of the above FT properties to quickly determine the FT of each of the pulses g1 (t), g2 (t), and
g3 (t) shown in Figure 4.31b–d without the need to start from first principles.
The first pulse g1 (t) is the result of delaying the centred rectangular pulse g(t) (shown again in Figure 4.31a) by
𝜏/2 and multiplying by a scale factor − 1. Thus
Telegram: @Persian_IEEEComSoc

( )
t − 𝜏∕2
g1 (t) = −A rect
𝜏
We take the FT of both sides of the above equation, applying the linearity and time shifting properties of the FT,
to obtain the FT of g1 (t) as

G1 (f ) = −G(f )e−j2𝜋f 𝜏∕2

= −A𝜏 sinc(f 𝜏)e−j𝜋f 𝜏
= e−j𝜋 A𝜏 sinc(f 𝜏)e−j𝜋f 𝜏 , (since − 1 = e−j𝜋 )
= A𝜏 sinc(f 𝜏)e−j𝜋(1+f 𝜏)
 262 4 Frequency Domain Analysis of Signals and Systems

The second pulse g2 (t) in Figure 4.31c is a bipolar pulse obtained as follows from two centred rectangular pulses
g 1/2 (t) of amplitude A and duration 𝜏/2: (i) advance one pulse by 𝜏/4; (ii) delay the other g 1/2 (t) by 𝜏/4 and scale
by −1; (iii) add the two pulses of (i) and (ii). That is
( ) ( )
𝜏 𝜏
g2 (t) = g1∕2 t + − g1∕2 t −
( 4 ) 4 ( )
t + 𝜏∕4 t − 𝜏∕4
= A rect − A rect
𝜏∕2 𝜏∕2
( ) ( )
2(t + 𝜏∕4) 2(t − 𝜏∕4)
= A rect − A rect
𝜏 𝜏

When compared to the rectangular pulse Arect(t/𝜏) whose FT we are given, the two terms on the right-hand
side are the same rectangular pulse compressed by a factor of 2 and, respectively, advanced and delayed by 𝜏/4. We
therefore take the FT of both sides of the above equation, employing the FT properties of time scaling (Eq. (4.83))
and time shifting (Eq. (4.80)), to obtain the FT of g2 (t) as
( ) ( )
𝜏 𝜏 j2𝜋f 𝜏∕4 𝜏 𝜏 −j2𝜋f 𝜏∕4
G2 (f ) = A sinc f e − A sinc f e
2 ( )2 2 2 ( )
𝜏 𝜏 𝜏 𝜏
= A sinc f [ej𝜋f 𝜏∕2 − e−j𝜋f 𝜏∕2 ] = A sinc f 2j sin(𝜋f 𝜏∕2)
2 2 2 2
( ) 𝜋f 𝜏∕2 ( ) ( )
𝜏 𝜏 A𝜏 2 𝜏 𝜏
= A sinc f 2j sin(𝜋f 𝜏∕2) × = j𝜋 f sinc f sinc f
2 2 𝜋f 𝜏∕2 2 2 2
( )
A𝜏 2 𝜏
= j𝜋 f sinc2 f
2 2
Notice that G2 (f ) is imaginary as expected since g2 (t) is an odd function. The phase of G2 (f ) is ∠G2 (f ) = 90∘ for
f > 0, and ∠G (f ) = −90∘ for f < 0.
2
Moving on to the third pulse (Figure 4.31d), we observe that the given triangular pulse g3 (t) can be expressed in
terms of the above bipolar pulse g2 (t) since

⎧0, t < −𝜏∕2

⎪
dg3 (t) ⎪2A∕𝜏, −𝜏∕2 ≤ t < 0 2
=⎨ = g (t)
dt ⎪−2A∕𝜏, 0 < t ≤ 𝜏∕2 𝜏 2
⎪0, t > 𝜏∕2
⎩

Integrating both sides yields

t
2
g3 (t) = g (t)dt
𝜏 ∫−∞ 2
Telegram: @Persian_IEEEComSoc

Thus, g3 (t) is the integral of g2 (t) along with a scale factor 2/𝜏. And since we know G2 (f ) from the previous
solution, we take the FT of both sides of the above equation, employing the integration property of FT (with
G2 (0) = 0) and the linearity property, to obtain the FT of g3 (t) as
( )
2 G2 (f ) 2 1 A𝜏 2 𝜏
G3 (f ) = = × × j𝜋 f sinc2 f
𝜏 j2𝜋f 𝜏 j2𝜋f 2 2
( )
𝜏 2 𝜏
= A sinc f
2 2
This spectrum is real as expected since g3 (t) is even. In addition, G3 (f ) is positive at all frequencies, so its phase
∠G3 (f ) = 0∘ at all frequencies.
 4.3 Fourier Transform 263

To summarise, we have obtained the FT of the delayed (and polarity-inverted) rectangular pulse g1 (t), the bipolar
rectangular pulse g2 (t), and the triangular pulse g3 (t) = Atrian(t∕𝜏) as follows

G1 (f ) = A𝜏 sinc(f 𝜏)e−j𝜋(1+f 𝜏)
( )
A𝜏 2 𝜏
G2 (f ) = j𝜋 f sinc2 f
2 2
( )
𝜏 𝜏
G3 (f ) = A sinc2 f
2 2
These expressions in general evaluate to complex numbers, and it is important to be able to interpret them to
extract the correct magnitude and angle, which, respectively, provide amplitude and phase spectral values. For
example, the magnitude and angle of G1 (f ) are given by

|G1 (f )| = |A𝜏 sinc(f 𝜏)|

⎧0, A𝜏 sinc(f 𝜏) = 0
⎪
∠G1 (f ) = ⎨−(1 + f 𝜏)𝜋, A𝜏 sinc(f 𝜏) > 0
⎪
⎩−𝜋𝜏f , A𝜏 sinc(f 𝜏) < 0
The spectra of the above three pulses are plotted in Figure 4.32a–c. It would be useful to take a moment to verify
some of the values plotted, especially for ∠G1 (f ), to satisfy yourself that you have the required understanding. The
plot of ∠G3 (f ) = 0∘ is trivial and not shown.

4.3.2 Table of Fourier Transforms

The FT of a signal may be derived from first principles by evaluating the defining FT integral in Eq. (4.74) or by
applying one or more of the FT properties in Eqs. (4.79)–(4.91) to manipulate a known FT to obtain the FT of a
desired related signal. We have dealt in detail with both these methods in the preceding pages and two worked
examples. Yet a third way to obtain the FT of a time domain waveform or the IFT of a frequency domain signal
description is to read this information from a table of FT pairs for the most common standard signals, such as
Table 4.5. If a desired signal is not in the table, it may still be possible to obtain its FT by manipulating a related
entry in the Table according to the FT properties discussed.

Worked Example 4.9 Analysis of Digital Transmission Pulses

In digital transmission, information is conveyed by transmitting energy signals of duration T s . These signals are
referred to as pulses or symbols, and each unique pulse is associated with a unique group of bits. The symbol rate is
Rs = 1/T s , and if each transmitted symbol represents k bits then the bit rate of the transmission system is Rb = kRs .
Telegram: @Persian_IEEEComSoc

The minimum number of unique symbols required to cover all possible combinations of k bits is M = 2k , hence
the system is described as M-ary transmission.
In the case of a modulated transmission system, the symbols are sinusoidal pulses of duration T s and the unique-
ness of each symbol is achieved through a unique state of the carrier signal, corresponding to a unique set of values
of the amplitude, phase, and frequency of the sinusoidal waveform that is constrained to duration T s to form the
transmitted symbol. The constraining of an infinite-duration sinusoid to duration T s is achieved by multiplying
the sinusoid by a suitable windowing pulse of duration T s . It is worth mentioning that in this role the sinusoid is
called a carrier in obvious reference to its function of conveying or carrying information.
In this worked example, we wish to examine the time and frequency domain effects of using the two windowing
pulses, a rectangular pulse w1 (t) and a raised cosine pulse w2 (t), shown in Figure 4.33a and b, and to comment on
their impact on symbol rate and interference considerations.
 264 4 Frequency Domain Analysis of Signals and Systems

Table 4.5 Fourier transform pairs.

Time domain signal Description Fourier transform

1 𝛿(t) Unit impulse, Eq. (2.28) 1

2 1 Unit constant 𝛿(f )

3 𝛿(t − to ) Delayed unit impulse exp(−j2𝜋fto )

4 exp(j2𝜋f c t) Complex exponential 𝛿(f − f c )

5 sgn(t) Signum function, Eq. (2.13) 1/(j𝜋f )

6 1/𝜋t −j sgn(f )
1 1
7 u(t) Unit step, Eq. (2.12) 𝛿(f ) +
2 j2𝜋f
8 rect(t∕𝜏) Rectangular pulse, Eq. (2.16) 𝜏 sinc(f 𝜏)

9 sinc(t∕𝜏) Sinc pulse with nulls at ±𝜏, ±2𝜏, ±3𝜏, … 𝜏 rect(f 𝜏)

( )
𝜏 𝜏
10 trian(t∕𝜏) Triangular pulse, Eq. (2.21) sinc2 f
2 2
1 1
11 cos(2𝜋fc t) 𝛿(f − fc ) + 𝛿(f + fc )
2 2
1 1
12 sin(2𝜋fc t) 𝛿(f − fc ) − 𝛿(f + fc )
2j 2j
[ ( )] ( )
1 1 2𝜋 t
Raised cosine pulse of duration 𝜏 𝜏 𝜏 𝜏
13 + cos t rect sinc(f 𝜏) + sinc(f 𝜏 + 1) + sinc(f 𝜏 − 1)
2 2 𝜏 𝜏 2 4 4
( ) ( ) [ ( ) ( )]
m𝜋 t 𝜏 m m
14 cos t rect Centred sinusoidal pulse with odd sinc f 𝜏 + + sinc f 𝜏 −
𝜏 𝜏 number m of half-cycles in 2 2 2
−𝜏∕2 ≤ t ≤ 𝜏∕2
( ) ( ) [ ( ) ( )]
m𝜋 t 𝜏 m m
15 sin t rect Centred sinusoidal pulse with even j sinc f 𝜏 + − sinc f 𝜏 −
𝜏 𝜏 number m of half-cycles in 2 2 2
−𝜏∕2 ≤ t ≤ 𝜏∕2

16 exp(−𝜋t2 ) Gaussian function exp(−𝜋f 2 )

1
17 exp(−at)u(t), a>0
a + j2𝜋f
2a
18 exp(−a|t|), a>0
a2 + (2𝜋f )2
Telegram: @Persian_IEEEComSoc

1
19 exp(−at)tu(t), a>0
(a + j2𝜋f )2
1 j2𝜋f
20 cos(2𝜋fc t)u(t) Single-sided cosine [𝛿(f − fc ) + 𝛿(f + fc )] +
4 (2𝜋fc )2 − (2𝜋f )2
1 (2𝜋f )2
21 sin(2𝜋fc t)u(t) Single-sided sine [𝛿(f − fc ) − 𝛿(f + fc )] +
4j (2𝜋fc )2 − (2𝜋f )2
a + j2𝜋f
22 exp(−at) cos(2𝜋fc t)u(t) Exponentially decaying single-sided
cosine (2𝜋fc )2 + (a + j2𝜋f )2
2𝜋fc
23 exp(−at) sin(2𝜋fc t)u(t) Exponentially decaying single-sided
sine (2𝜋fc )2 + (a + j2𝜋f )2
 4.3 Fourier Transform 265

|G1(f)|
Aτ

(a) 0 f
–5/τ –4/τ –3/τ –2/τ –1/τ 0 1/τ 2/τ 3/τ 4/τ 5/τ
∠G1(f), deg
180

0 f

–180
–5/τ –4/τ –3/τ –2/τ –1/τ 0 1/τ 2/τ 3/τ 4/τ 5/τ

|G2(f)|
0.725Aτ2

(b)
0 f
∠G2(f), deg
90

0 f

–90
–4/τ –2/τ 0 2/τ 4/τ
|G3(f)|
Aτ 2

(c)
∠G3(f) = 0˚

0 f
–4/τ –2/τ 0 2/τ 4/τ

Figure 4.32 (a): Worked Example 4.8: Amplitude and phase spectra of delayed and polarity-inverted rectangular pulse g1 (t)
of duration 𝜏 (shown in Figure 4.31b) (b) and (c): Worked Example 4.8; (b) Amplitude and phase spectra of bipolar
rectangular pulse g2 (t) of duration 𝜏; (c) Amplitude spectrum of triangular pulse g3 (t) of duration 𝜏.
Telegram: @Persian_IEEEComSoc

We will assume a carrier that completes 20 cycles in one symbol interval, so the basic expression for the sinu-
soidal carrier is g(t) = cos(2𝜋f c t), with f c = 20Rs , where Rs = 1/T s is the symbol rate (in symbols per second, called
a baud). Multiplying g(t), which is periodic and of infinite duration, by the given window functions w1 (t) and
w2 (t) produces the respective sinusoidal pulses g1 (t) and g2 (t), of duration T s , shown in Figure 4.33c and d. We
see that g2 (t) tapers to zero at symbol boundaries, whereas g1 (t) does not. Since the state of the carrier will change
from one symbol interval to the next (depending on the identity of data bits in each interval), jump discontinuities
will occur in g1 (t) when its phase or amplitude undergoes a sudden change at a symbol boundary. For example,
a change in phase from 0∘ to 180∘ will cause a jump discontinuity equal to double the carrier amplitude. Such
discontinuities give rise to higher-frequency components in the transmitted waveform, which causes interference
into adjacent channels or a reduction in bit rate or in the number of users when trying to mitigate the problem.
 266 4 Frequency Domain Analysis of Signals and Systems

w1(t) w2(t)
1 1

(a) (b)

0 t 0 t
–Ts/2 0 Ts/2 –Ts/2 0 Ts/2

g1(t) g2(t)
1 1

(c) 0 t (d) 0 t

–1 –1
–Ts/2 0 Ts/2 –Ts/2 0 Ts/2

Figure 4.33 Worked Example 4.9: Windowing pulses. (a) Rectangular pulse; (b) Raised cosine pulse; (c) Rectangular
windowed sinusoidal pulse; (d) Raised cosine windowed sinusoidal pulse.

In contrast, the tapering of g2 (t) eliminates discontinuities and ensures a smooth transition of signal level through
symbol boundaries, whatever the state of the carrier in each symbol interval. This reduces high-frequency content
in the transmitted waveform and hence minimises interference into adjacent channels.
Let us now turn to the frequency domain, with the help of the Fourier transform (FT), to provide a more quan-
titative analysis of the above time domain observations. We have the following FT pairs

w1 (t) ⇌ W1 (f ); g1 (t) ⇌ G1 (f )
w2 (t) ⇌ W2 (f ); g2 (t) ⇌ G2 (f )

Our interest is in G1 (f ) and G2 (f ), but since g1 (t) and g2 (t) are, respectively, the result of multiplying w1 (t) and
w2 (t) by cos(2𝜋f c t), it follows from Eq. (4.82) that the spectra of G1 (f ) and G2 (f ) will be the respective translations
of W 1 (f ) and W 2 (f ) through ±f c . All the spectrum information about G1 (f ) and G2 (f ) is therefore contained in
W 1 (f ) and W 2 (f ), except for spectral location, so it will be enough for us to examine W 1 (f ) and W 2 (f ) instead.
The FT of a rectangular pulse is derived in Worked Example 4.7, from which we have
Telegram: @Persian_IEEEComSoc

W1 (f ) = Ts sinc(fTs )

The raised cosine pulse w2 (t) in Figure 4.33b is given by

[ ( )] ( )
1 1 2𝜋 t
w2 (t) = + cos t rect
2 2 Ts Ts
( ) ( ) ( )
1 t 1 t 1
= rect + rect cos 2𝜋 • •t
2 Ts 2 Ts Ts

Taking the FT of both sides and (for the second term on the right-hand side) invoking the FT property in
Eq. (4.82), which specifies the frequency domain effect of multiplying by a sinusoid – of frequency 1/T s in this
 4.3 Fourier Transform 267

case, we obtain the following result quoted in Entry 13 of Table 4.5

[( ) ] [( ) ]
T T 1 T 1
W2 (f ) = s sinc(fTs ) + s sinc f + Ts + s sinc f − Ts
2 4 Ts 4 Ts
T T T
= s sinc(fTs ) + s sinc(fTs + 1) + s sinc(fTs − 1) (4.95)
2 4 4
We calculate |W1 (f )| and |W2 (f )| using the above formulas at f = −25/T s to f = 25/T s in steps of 1/(100T s )
to give a smooth plot. In order to better display the small values and variations which in a linear plot would be
indistinguishable from zero, we will use logarithmic units and plot the normalised amplitude spectra shown in
Figure 4.34 and defined as the following quantities
( ) ( )
|W1 (f )| |W2 (f )|
Z1 = 20 log10 ; Z2 = 20 log10
max(|W1 (f )|) max(|W2 (f )|)

In the plot, the y axis is terminated at −50 dB to zoom in better on the region of interest. Note that the right-hand
side of the amplitude spectra |G1 (f )| and |G2 (f )| will be identical to Figure 4.34 except for the addition of f c
to the frequency values along the x axis. We see that g1 (t) contains strong and persistent sidelobes (and hence
high-frequency components), with its 25th sidelobe still above −40 dB (relative to peak). In contrast, the sidelobes
of g2 (t) decay very rapidly so that by the fourth sidelobe the amplitude level of frequency components is already
below −50 dB.
To understand the impact that this will have on digital transmission based on these two pulses, let B denote
transmission bandwidth (which means that the occupied frequency range is f c − B/2 → f c + B/2) and let it be
required to keep interference into adjacent channels below −30 dB. To satisfy this spec, B/2 must extend beyond

0
B
–10

–20
Z1, dB

Threshold interference
–30 level = –30 dB

–40

–50 →f
–20Rs –10Rs 0 10Rs 20Rs
0
Telegram: @Persian_IEEEComSoc

–10
B
–20
Z2, dB

Threshold interference
–30 level = –30 dB

–40

–50 →f
–25Rs –20Rs –10Rs 0 10Rs 20Rs 25Rs

Figure 4.34 Worked Example 4.9: Normalised amplitude spectra Z 1 and Z 2 of rectangular and raised cosine windows
respectively. B indicates bandwidth after translation to location f c ≫ B.
 268 4 Frequency Domain Analysis of Signals and Systems

0.5Ts

0.4Ts
Ts
sinc( f Ts)
2 Ts T
W2(f ) = sinc( f Ts) + s sinc( fTs + 1)
0.3Ts 4 4
Ts
+ sinc( f Ts – 1)
4
0.2Ts
Ts Ts
sinc(f Ts + 1) sinc( f Ts – 1)
4 4
0.1Ts

0 →f

–0.1Ts
–6Rs –5Rs –4Rs –3Rs –2Rs –Rs 0 Rs 2Rs 3Rs 4Rs 5Rs 6Rs

Figure 4.35 Spectrum W 2 (f ) of a raised cosine pulse of duration T s is the sum of three component sinc pulses.

f c until the start of the sidelobe that first dips below −30 dB. Reading from Figure 4.34, we have
{
B 10Rs , for g1 (t)
=
2 2Rs , for g2 (t)
This means that transmission bandwidth B would be 20 times the symbol rate when using g1 (t), compared to
four times the symbol rate if using g2 (t). A larger required transmission bandwidth per user inevitably leads to a
reduction in the number of users that can be simultaneously supported. If, on the other hand, allocated bandwidth
B is fixed (as is typically the case), it means that we would be limited to operating at a symbol rate Rs ≤ B/20 when
using g1 (t) and Rs ≤ B/4 with g2 (t). Therefore, using the raised cosine window would allow a fivefold increase in
symbol rate (and hence bit rate) when compared to the rectangular window.
The required solution is complete, but to give a little insight into the reason for the rapid decay of sidelobes in the
raised cosine pulse, we show in Figure 4.35 a linear plot of W 2 (f ) in Eq. (4.95) along with its three component sinc
pulses. We see that the polarity of the sidelobes of the two lower-amplitude (and frequency-shifted) sinc pulses and
the polarity of the larger sinc pulse are opposite at every point, and this produces a rapid cancellation of sidelobes
when they are summed to give W 2 (f ). Note, however, that, on the downside, there is a broadening of the main
Telegram: @Persian_IEEEComSoc

lobe of W 2 (f ) when compared to the lobe width of the larger sinc pulse. This is as a result of the main lobes of the
two frequency-shifted sinc pulses extending beyond the main lobe of the larger sinc pulse on either side.

4.3.3 Fourier Transform of Periodic Signals

We derived the FT by extending the Fourier series to apply to nonperiodic signals, and we did so by following the
evolution of a signal from periodic gT (t) to nonperiodic g(t) as illustrated in Figure 4.28. It turns out that taking
those steps in reverse will allow us to extend the FT to be applicable to periodic signals, thus closing the loop that
connects the FT and the Fourier series.
Every periodic signal gT (t) of period T is the result of repeating a pulse g(t) of duration T over and over along
the time axis at regular intervals 0, ±T, ±2 T, ±3 T, …, where the pulse g(t) constitutes the fundamental shape of
 4.3 Fourier Transform 269

the periodic signal. That is, g(t) is the waveform of gT (t) in the interval (−T/2, T/2), so we may write
{
gT (t), −T∕2 ≤ t ≤ T∕2
g(t) = (4.96)
0, elsewhere

We assume that the pulse g(t) is an energy signal. It is therefore guaranteed to have an FT G(f ), the magnitude
of which is shown in the bottom row of Figure 4.28 for the specific case of a triangular pulse. Other fundamental
shapes of an arbitrary periodic signal will understandably have a different spectral shape to what is shown in
Figure 4.28. We see in Figure 4.28 that the spectrum of gT (t), which we will denote as GT (f ), is a weighted sampling
of G(f ) at a regular frequency spacing f o = 1/T. The sample at frequency f = nf o , n = 0, ±1, ±2, ±3, …, is a spectral
line that represents a complex exponential of frequency nf o and coefficient Cn = f o G(nf o ), by definition of G(f ) in
Eq. (4.72). We represent this spectral line as

fo G(nf o )𝛿(f − nf o )

which is an impulse of weight f o G(nf o ) located at f = nf o . The FT of the periodic signal is a collection or sum of
these impulses for all n. Thus
∑
∞
GT (f ) = fo G(nf o )𝛿(f − nf o ), fo = 1∕T (4.97)
n=−∞

This is an important result. It indicates that, whereas the FT G(f ) of a nonperiodic signal g(t) is continuous, the FT
GT (f ) of a periodic signal gT (t) is discrete, having spectral lines or impulses of weight f o G(nf o ) at discrete frequency
points nf o . Each spectral line represents a complex exponential having frequency nf o , amplitude |f o G(nf o )|, and
phase equal to the angle of f o G(nf o ), which is simply ∠G(nf o ) since f o is positive.
For interested readers, an alternative and mathematically rigorous derivation of Eq. (4.97) may be obtained as
follows. Let us express gT (t) as a complex exponential Fourier series according to Eq. (4.23)
∑
∞
gT (t) = Cn ej2𝜋nf o t , fo = 1∕T
n=−∞

The coefficient Cn in the above equation is related to the FT G(f ) of the fundamental shape of gT (t) according to
the definition of G(f ) in Eq. (4.72) as

Cn = fo G(f )|f =nf o = fo G(nf o )

Making this substitution for Cn in the previous equation yields

∑
∞
gT (t) = fo G(nf o )ej2𝜋nf o t (4.98)
Telegram: @Persian_IEEEComSoc

n=−∞

We notice that the right-hand side is a sum of complex exponentials each of which is Fourier transformable
according to entry 4 in Table 4.5

ej2𝜋nf o t ⇌ 𝛿(f − nf o )

Therefore, denoting the FT of gT (t) as GT (f ) and taking the FT of both sides of Eq. (4.98) yields
∑
∞
GT (f ) = fo G(nf o )𝛿(f − nf o ), fo = 1∕T
n=−∞

This is the earlier heuristically derived Eq. (4.97), and is simply an alternative way of expressing the double-sided
discrete spectrum of a periodic signal.
 270 4 Frequency Domain Analysis of Signals and Systems

4.4 Discrete Fourier Transform

Our discussion has so far been focused on continuous-time (CT) signals, starting with the Fourier series for peri-
odic CT signals and adapting this to obtain the FT for nonperiodic CT signals. The Fourier series may also be
adapted as discussed below to obtain a discrete Fourier transform (DFT), which is applicable to discrete-time (DT)
signals.
Consider a DT signal represented by g[n] = {g(0), g(1), g(2), · · · , g(N − 1)}. This is a sequence of N samples
obtained by sampling one cycle (of duration T) of a CT signal g(t) at regular intervals T s , known as the sampling
interval. The sampling rate F s is therefore
1
Fs = ; ⇒ Fs Ts = 1 (4.99)
Ts
The sampling is carried out by dividing interval T of g(t) into N equal sub-intervals of size T s . One sample is then
taken at the start of each subinterval, yielding the samples g(0), g(1), g(2), …, g(N − 1). The samples may, however,
also be taken at the midpoint of each subinterval, provided there is consistency so that all samples are uniformly
spaced in time. Recall that g(t) is constituted of complex exponentials and Eq. (4.24) gives the coefficient of the kth
complex exponential in g(t) as
T∕2
1
Ck = g(t)e−j2𝜋kf o t dt
T ∫−T∕2
This kth complex exponential has frequency kf o = k/T, amplitude equal to the magnitude |Ck | of Ck (which
is in general complex-valued), and phase equal to the angle ∠Ck of Ck . As a result of the time variable in g(t)
becoming discrete in g[n], the right-hand side of the above equation for Ck will change as follows to describe the
complex exponentials in g[n]: t becomes nT s , the impulse of weight g(t)dt becomes the sample g(nT s ) ≡ g(n), and
the integral becomes a summation from n = 0 to N − 1. Thus

1 ∑
N−1
Ck = g(n)e−j2𝜋kf o nT s
T n=0
Substituting T s = T/N and f o T = 1 and defining the DFT of the data sequence g[n], denoted G(k), as the coef-
ficient per unit frequency interval of the kth of the complex exponentials that constitute the data sequence, i.e.
G(k) = Ck /f o = Ck T, we obtain
∑
N−1
2𝜋
G(k) = g(n)e−j N kn , k = 0, 1, 2, · · · , N − 1 (4.100)
n=0

Equation (4.100) defines the DFT of g[n]. The spacing of spectral lines, usually referred to as frequency resolution,
in the spectrum G[k] is
Telegram: @Persian_IEEEComSoc

1 1 F
Δf = fo = = = s (4.101)
T NT s N
where T is the total duration of g[n]; and the kth sample G(k) of the spectrum is located at frequency f k (in Hz)
given by
k F
fk = kf o = =k s (4.102)
T N
It is implicitly assumed that the sequence g[n] repeats outside the interval T within which the N samples {g(0),
g(1), g(2), …, g(N − 1)} were taken. This means that

g(n) = g(n ± N) (4.103)

 4.4 Discrete Fourier Transform 271

Thus, g[n] is a periodic DT signal having period N samples, and is said to complete one cycle (or 2𝜋 rad) in a
span of N samples, so that its fundamental angular frequency (in rad/sample) is given by
2𝜋
Ωo = (4.104)
N
The angular frequency Ωk of the kth sample G(k) of the spectrum is
2𝜋
Ωk = kΩo = k , (4.105)
N
and this is a factor in the exponent of the DFT expression.
Equation (4.100) states that a data sequence g[n] = {g(0), g(1), g(2), · · · , g(N − 1)}, of length N, has a DFT
G[k] = {G(0), G(1), G(2), · · · , G(N − 1)}, also of length N. The kth element G(k) of the DFT sequence is obtained
as a weighted sum of g[n] using corresponding weights W[n] = {1, e−jΩk , e−j2Ωk , · · · , e−j(N−1)Ωk }. The weights are
all of unit magnitude but of increasing angle −nΩk , so the essence of the weighting is simply to rotate each data
sample g(n), treated as a phasor, through nΩk radians in the clockwise direction prior to summing to obtain G(k).
In the case of G(0), all weights have unit magnitude and zero angle (since Ωk = kΩo = 0 × Ωo = 0), so G(0) is just
the (unweighted) sum of the entire data sequence.
To derive an expression for the inverse discrete Fourier transform (IDFT) which allows the exact data sequence
g[n] to be recovered from the transform sequence G[k], recall the complex exponential Fourier series of g(t)
∑
∞
g(t) = Ck ej2𝜋kf o t
k=−∞

and adapt this series to a data sequence by making the following changes: replace t with nT s ; note that g(t)dt ≡
g(n), so that g(t) ≡ g(n)/dt; replace dt with T s ; and recall that (by definition) G(k) = Ck T, so that Ck = G(k)/T, k = 0,
1, 2, …, N − 1. Introducing these changes into the above equation, and noting again that T s = T/N and f o T = 1,
we obtain
g(n) ∑ G(k) j2𝜋kf o nT s 1 ∑
N−1 N−1
= e = G(k)ej2𝜋kf o nT∕N
Ts k=0
T T k=0

Ts ∑ 1 ∑
N−1 N−1
2𝜋 2𝜋
g(n) = G(k)ej N kn = G(k)ej N kn
T k=0 N k=0
That is
1 ∑
N−1
2𝜋
g(n) = G(k)ej N kn , n = 0, 1, 2, … , N − 1 (4.106)
N k=0
Equation (4.106) is the IDFT of G(k). It states that the original data sample g(n) is the weighted aver-
age of the transform sequence G[k] = {G(0), G(1), G(2), … , G(N − 1)} using corresponding weights W[k] =
Telegram: @Persian_IEEEComSoc

{1, ejΩn , ej2Ωn , … , ej(N−1)Ωn }, where Ωn = n(2𝜋∕N). The weighting simply rotates each transform sample G(k)
through kΩn radians counterclockwise prior to averaging (i.e. summing the rotated samples and dividing by
N) to obtain g(n). In the case of g(0), where Ωn = 0, there is no rotation prior to averaging, so g(0) is just the
(unweighted) average of the entire transform sequence G[k].
To summarise, a discrete signal may be represented in two ways: either in the time domain by g[n], which shows
the signal’s waveform features, or in the frequency domain by G[k], showing the signal’s spectral characteristics.
g[n] and G[k] are said to form a DFT pair, a relationship denoted as
g[n] ⇌ G[k]
Note our careful choice of index notation, using n to denote a sample count in the time domain and k for a count
in the frequency domain. Given g[n], then G[k] may be computed; and given G[k], then an exact reconstruction
 272 4 Frequency Domain Analysis of Signals and Systems

of g[n] may be obtained using the pair of equations

∑
N−1
2𝜋
G(k) = g(n)e−j N kn , k = 0, 1, 2, … , N − 1 (DFT)
n=0

1 ∑
N−1
2𝜋
g(n) = G(k)ej N kn , n = 0, 1, 2, … , N − 1 (IDFT) (4.107)
N k=0

We should point out that some literature will describe the DFT derived above as discrete-time Fourier series
(DTFS), with the word series emphasising the discrete nature of the frequency domain representation G[k]. Such
literature will also define the discrete-time Fourier transform (DTFT), an adaptation of the FT equations (Eqs. (4.74)
and (4.75)) for a nonperiodic CT signal g(t), namely
∞
G(f ) = g(t)e−j2𝜋ft dt (FT)
∫−∞
∞
g(t) = G(f )ej2𝜋ft df (IFT) (4.108)
∫−∞
for application to a nonperiodic DT signal g[n], which is the result of sampling g(t) at regular intervals T s or
sampling rate F s = 1/T s . For completeness, we derive the DTFT equation by making the following changes in the
above FT equation: g(t)dt → g(n); t → nT s = n/F s ; integration → summation over all n. This leads to
∑
∞
G(f ) = g(n)e−jn2𝜋f ∕Fs (DTFT) (4.109)
n=−∞

The DTFT may be expressed in terms of an angle parameter Ω in radians defined as the normalised frequency f
(i.e. f divided by F s ) multiplied by 2𝜋
f
Ω = 2𝜋 (4.110)
Fs
So, f = F s corresponds to Ω = 2𝜋, f = F s /2 corresponds to Ω = 𝜋, and DC corresponds to Ω = 0. In terms of Ω,
the DTFT equation becomes
∑
∞
G(Ω) = g(n)e−jnΩ (DTFT) (4.111)
n=−∞

However, to avoid any confusion we will stick to an exclusive use of f (in Hz) for frequency domain variable and
t (in seconds) for time domain variable. Equation (4.111) is stated simply for completeness and to help you relate
our discussion with other conventions that you may encounter in the literature. Returning therefore to Eq. (4.109)
and replacing f with f + F s we find
Telegram: @Persian_IEEEComSoc

( ) ( )
∑
∞
−jn2𝜋
f +Fs ∑
∞
−jn2𝜋
f
G(f + Fs ) = g(n)e Fs = g(n)e Fs e−jn2𝜋
n=−∞ n=−∞
( )
∑∞
−jn2𝜋
f
= g(n)e Fs

n=−∞
= G(f )

This means that the DTFT G(f ) is periodic with period F s . So, one cycle of G(f ) lies in the frequency range
(−F s /2, F s /2) and the rest of G(f ) are just replications of this cycle along the frequency axis at integer multiples of
F s . This agrees with our discovery in Figure 4.20c about the frequency domain effect of instantaneous sampling
at rate F s .
 4.4 Discrete Fourier Transform 273

To obtain the inverse discrete-time Fourier transform (IDTFT) expression, we make the following changes to
the IFT expression in Eq. (4.108): dt → T s ; g(t) → g(n)/dt = g(n)/T s = g(n)F s ; t → nT s = n/F s ; integration over
(−∞, ∞) → integration over one cycle (−F s /2, F s /2) of G(f ). This gives
s F ∕2
1
g(n) = G(f )ejn2𝜋f ∕Fs df (IDTFT) (4.112)
Fs ∫−Fs ∕2
In terms of Ω in Eq. (4.110), we use df = dΩ•Fs ∕2𝜋; f = Fs ∕2 → Ω = 𝜋 to obtain
𝜋
1
g(n) = G(Ω)ejnΩ dΩ (IDTFT) (4.113)
2𝜋 ∫−𝜋
As earlier stated, we will stick to the use of the frequency variable f . For comparison, let us collect in one place
the DFT (also called DTFS), IDFT, DTFT, and IDTFT expressions that we have derived above
∑
N−1
2𝜋
G(k) = g(n)e−j N kn , k = 0, 1, 2, … , N − 1 (DFT)
n=0

1 ∑
N−1
2𝜋
g(n) = G(k)ej N kn , n = 0, 1, 2, … , N − 1 (IDFT)
N k=0
∑
∞
G(f ) = g(n)e−jn2𝜋f ∕Fs (DTFT)
n=−∞
∑∞
f
G(Ω) = g(n)e−jnΩ , Ω = 2𝜋 (DTFT)
n=−∞
Fs
s F ∕2
1
g(n) = G(f )ejn2𝜋f ∕Fs df (IDTFT)
Fs ∫−Fs ∕2
𝜋
1
g(n) = G(Ω)ejnΩ dΩ (IDTFT) (4.114)
2𝜋 ∫−𝜋
Now consider the above DTFT expression. If the nonperiodic CT signal g(t), from which the samples g(n) are
taken, is of duration T and occupies the interval t = (0, T) and it is sampled at interval T s to produce the data
sequence g[n] = {g(0), g(T s ), …, g([N − 1]T s )} ≡ {g(0), g(1), …, g(N − 1)}, the values of g(n) in the DTFT expression
will be zero outside the range n = (0, N − 1) so that the DTFT expression reduces to
∑
N−1
G(f ) = g(n)e−jn2𝜋f ∕Fs (DTFT)
n=0

If we take samples of the DTFT at frequency intervals f o = 1/T = 1/NT s = F s /N, the kth sample will be at fre-
quency f = kF s /N, having a value, denoted G(k), that is obtained by substituting f = kF s /N into the above equation
Telegram: @Persian_IEEEComSoc

to obtain
( )
F ∑
N−1 kF
−jn2𝜋 Ns × F1 ∑
N−1
2𝜋
G k s ≡ G(k) = g(n)e s = g(n)e−j N kn ≡ DFT
N n=0 n=0

The DFT is therefore a sequence of equally spaced samples of the DTFT, the samples being taken at frequency
intervals (called frequency resolutions) equal to the reciprocal of the duration T of the DT signal represented by
the DTFT. The DFT is therefore an appropriate tool for frequency domain representation of both periodic and
nonperiodic data sequences, and we will have no need for DTFT or its inverse in this book.
If the data sequence is periodic (of period T and fundamental frequency f o = 1/T) then the DFT G[k] provides the
magnitude (per unit frequency) and phase of the (discrete set of) harmonic complex exponentials (at frequencies
0, f o , 2f o , …, (N/2)f o ) that constitute the periodic data sequence. We will see later that the rest of the elements of
 274 4 Frequency Domain Analysis of Signals and Systems

g(1)
g(0) g[n] = {g (0), g (1), g (2), g (3)};
g(2) 1 1
g(3) N = 4; fo = =
T 4Ts
nTs
Ts Ts Ts Ts
T

g(1)

g(0) {g[n], 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};

g(2) 1 1 fo
N = 16; fo′ = = =
g(3) T′ 16Ts 4
nTs
Ts Ts Ts Ts
T′

Figure 4.36 Appending zeros to improve the frequency resolution of the DFT of a data sequence g[n].

the sequence of G[k], for k = N/2 + 1, N/2 + 2, …, N − 1 are respective complex conjugates of the elements at
N/2 − 1, N/2 − 2, …, 1 and correspond to frequencies kf o .
If, on the other hand, the data sequence is nonperiodic (of duration T) then the DFT G[k] provides the magnitude
(per unit frequency) and phase of samples of the continuum of harmonic complex exponentials that constitute the
data sequence. In other words, a nonperiodic data sequence g[n] of duration T has a continuous spectrum given by
the DTFT, and the DFT provides samples of this spectrum at resolution f o = 1/T. This resolution can be improved
(i.e. f o made smaller) by increasing T. But since the data sequence g[n] and its original duration T and sampling
rate F s (and hence number of samples N) are usually fixed, one way to change the resolution to a new value fo ′ is
by appending zeros to g[n], as illustrated in Figure 4.36, where the DFT of the zero-padded data has a resolution
that is one-quarter that of the DFT of g[n].

Worked Example 4.10 DFT of Simple Sequences

We wish to calculate the DFT of the following simple sequences

(a) A unit impulse 𝛿[n] = {1, 0, 0, 0, 0, 0, 0, 0}

(b) Data sequence g[n] = {1, 0, 0, 1}
Telegram: @Persian_IEEEComSoc

(a) Number of samples N = 8, so angular frequency Ωo = 2𝜋/N = 𝜋/4, and Ωk = k𝜋/4. The kth transform sample
G(k) is the weighted sum of 𝛿[n] using weights W[n] = {1, e−jk𝜋∕4 , e−jk𝜋∕2 , e−j3k𝜋∕4 , e−jk𝜋 , e−j5k𝜋∕4 , e−j3k𝜋∕2 ,
e−j7k𝜋∕4 }
Therefore
∑
N−1
G(k) = 𝛿(n)W(n) = 1 × 1 + 0 × e−jk𝜋∕4 + 0 × e−jk𝜋∕2 + · · · + 0 × e−j7k𝜋∕4
n=0
=1

and hence G[k] = {1, 1, 1, 1, 1, 1, 1, 1}.

 4.4 Discrete Fourier Transform 275

(b) In this case, number of samples N = 4, so Ωo = 2𝜋/N = 𝜋/2 and Ωk = k𝜋/2. The kth transform sample G(k) is
the weighted sum of g[n] = {1, 0, 0, 1} using weights W[n] = {1, e−jk𝜋∕2 , e−jk𝜋 , e−j3k𝜋∕2 }. We therefore obtain
G(0) = sum[{1, 0, 0, 1} × {1, e−jk𝜋∕2 , e−jk𝜋 , e−j3k𝜋∕2 }]; k=0
= sum[{1, 0, 0, 1} × {1, 1, 1, 1}] = 1 + 0 + 0 + 1 = 2
G(1) = sum[{1, 0, 0, 1} × {1, e−jk𝜋∕2 , e−jk𝜋 , e−j3k𝜋∕2 }]; k=1
−j3𝜋∕2
=1+e = 1 + cos(−3𝜋∕2) + j sin(−3𝜋∕2)
√
= 1 + 0 + j = 2∠45∘

G(2) = sum[{1, 0, 0, 1} × {1, e−jk𝜋∕2 , e−jk𝜋 , e−j3k𝜋∕2 }]; k=2

−j3𝜋
=1+e = 1 + cos(−3𝜋) + j sin(−3𝜋)
=1−1+j×0=0
G(3) = sum[{1, 0, 0, 1} × {1, e−jk𝜋∕2 , e−jk𝜋 , e−j3k𝜋∕2 }]; k=3
−j9𝜋∕2
=1+e = 1 + cos(−9𝜋∕2) + j sin(−9𝜋∕2)
√
= 1 + 0 − j = 2∠ − 45∘
√ √
That is, G[k] = {2, 1 + j, 0, 1 − j} = {2, 2∠45∘ , 0, 2∠ − 45∘ }.

4.4.1 Properties of the Discrete Fourier Transform

Properties of the DFT closely mirror those of the FT presented in Section 4.3.1, so we will not repeat them here.
For example, the linearity property of DFT states that
if g1 [n] ⇌ G1 [k]
and g2 [n] ⇌ G2 [k]
then a1 g1 [n] + a2 g2 [n] ⇌ a1 G1 [k] + a2 G2 [k]

and the property of convolution in the time domain states that

g1 [n] ∗ g2 [n] ⇌ G1 [k]G2 [k]

Parseval’s theorem, earlier stated in Eq. (4.94) for CT signals, takes the following form for a DT signal g[n] of
length N
∑
N−1
1 ∑
N−1
g2 (n) = |G(k)|2 (4.115)
Telegram: @Persian_IEEEComSoc

n=0
N k=0

There are, however, two features and properties that are specific to DFT.

4.4.1.1 Periodicity

G(k) = G(k + N) (4.116)

This means that the DFT is periodic with period N. The frequency resolution of the spectrum G[k] is f o = 1/T,
so this period corresponds to a frequency interval
N N 1
Nf o = = = = Fs
T NT s Ts
 276 4 Frequency Domain Analysis of Signals and Systems

which is as expected from sampling theorem, since sampling replicates the spectrum of the original CT signal at
intervals of the sampling frequency F s . The periodicity property of Eq. (4.116) may be readily proved by substituting
k + N for k on both sides of the DFT expression in Eq. (4.114)

∑
N−1
2𝜋 ∑
N−1
2𝜋 2𝜋
G(k + N) = g(n)e−j N (k+N)n = g(n)e−j N kn e−j N Nn
n=0 n=0
∑
N−1
2𝜋
= g(n)e−j N kn e−j2𝜋n
n=0
∑
N−1
2𝜋
= g(n)e−j N kn (since e−j2𝜋n = 1)
n=0
= G(k)

4.4.1.2 Symmetry

G(N∕2 − m) = G∗ (N∕2 + m) for real g[n] (4.117)

The DFT G[k] = {G(0), G(1), G(2), …, G(N − 1)} of a real data sequence is a double-sided spectrum, with its
amplitude spectrum having even symmetry about k = N/2, and its phase spectrum having odd symmetry about
the same point. This means that the highest-frequency component in g[n] is

N F
fmax = f = s (4.118)
2 o 2

This is as expected in order to obey the sampling theorem and therefore avoid what is known as alias distortion.
Substituting m = 0 in Eq. (4.117) gives

G(N∕2) = G∗ (N∕2)

which means that the value G(N/2) of the transform sequence at the highest-frequency component is always a
real number if the data sequence g[n] is real. Alternatively, substituting k = N/2 in the DFT expression yields

( ) ∑
N−1
∑
N−1
N 2𝜋 N
•n
g(n)e−j N
•
G = 2 = g(n)e−jn𝜋
2 n=0 n=0
∑
N−1
Telegram: @Persian_IEEEComSoc

= g(n) cos(n𝜋)
n=0

which is real if g[n] is real.

The symmetry property may be readily proved by substituting N/2 + m for k on both sides of the DFT expression
in Eq. (4.114), which leads to

( ) N−1
∑ ( )
∑
N−1
N −j 2𝜋 N +m n 2𝜋
G +m = g(n)e N 2 = g(n)e−jn𝜋 e−j N mn
2 n=0 n=0
∑
N−1
2𝜋
= g(n) cos(n𝜋)e−j N mn
n=0
 4.4 Discrete Fourier Transform 277

Now substituting N/2 − m for k on both sides, we find that

( ) N−1
∑ ( )
∑
N−1
N −j 2𝜋 N −m n 2𝜋
G −m = g(n)e N 2 = g(n)e−jn𝜋 ej N mn
2 n=0 n=0
∑
N−1
2𝜋
= g(n) cos(n𝜋)ej N mn
n=0
= G∗ (N∕2 + m)
Note that the last line holds because the expression for G(N/2 − m) differs from that of G(N/2 + m) only in the
sign of j, which indicates a complex conjugate relationship.

4.4.2 Fast Fourier Transform

The DFT algorithm, as implemented by direct evaluation of the DFT formula in Eq. (4.114), is computationally
inefficient because it requires N complex multiplications and N − 1 complex additions to obtain each transform
sample, making a total of N 2 multiplications and N(N − 1) additions to compute the DFT. An N-point DFT with
N = 1024, for example, would therefore require over a million complex multiplications and over a million complex
additions to complete, which would be very time-consuming. A more efficient algorithm for computing the DFT
and its inverse can be developed by observing that many of the calculations in the DFT expression produce the same
result and therefore do not need to be done afresh each time. Denoting the factor e−j2𝜋∕N in the DFT expression as
W N , we can highlight some of the inherent computational simplifications as follows
WN = e−j2𝜋∕N

WN0 = 1

WN2 = WN∕2

WNN = 1
N∕2
WN = −1
3N∕4
WN =j
N∕4
WN = −j
(k+N∕2)
WN = −WNk
Telegram: @Persian_IEEEComSoc

WN(k+lN)(n+mN) = WNkn ; l, m = 0, ±1, ±2, · · · (4.119)

It is a straightforward matter to derive each of the above relations. For example
WNN = (e−j2𝜋∕N )N = e−j2𝜋N∕N = e−j2𝜋 = cos(−2𝜋) + j sin(−2𝜋) = 1 + 0 = 1
and
WN2 = (e−j2𝜋∕N )2 = e−j4𝜋∕N = e−j2𝜋∕(N∕2) ≡ WN∕2
Fast Fourier transform (FFT) is the name given to a range of algorithms that exploit these simplifications to
significantly speed up DFT computation [1–3]. The most commonly used method is the Cooley–Tukey FFT
algorithm [1]. The radix-2k Cooley–Tukey algorithm (k = 1, 2, 3, 4) is applicable to an N-point DFT when N is
an integer power of 2k . For example, the radix-16 algorithm is applicable to an N-point DFT for N = 256, 4096,
 278 4 Frequency Domain Analysis of Signals and Systems

65 536, … since these are all integer powers of 16. If the original data sequence does not satisfy this requirement,
trailing zeros are added to the sequence to make it of length 2kL , where L is an integer. The DFT is decomposed
into 2k smaller DFTs each of size N/2k , and this decomposition is repeated successively until after L stages we
reach N single-point DFTs where each transform is simply equal to the data sample. This is therefore a divide
and conquer algorithm. If the decomposition is accomplished by going sequentially through the data sequence
and taking one sample at a time to form the sequence for each of the smaller DFTs in turn, the technique is
described as decimation-in-time (DIT) FFT. For example, the radix-2 DIT FFT algorithm decomposes the DFT
of g[n] into two DFT expressions, one involving the even-numbered samples of g[n] and the other involving the
odd-numbered samples. If, on the other hand, the data sequence is partitioned into sequential blocks that are
allocated to each of the smaller DFTs, the technique is known as decimation-in-frequency (DIF) FFT. We discuss
below the radix-2 DIF FFT algorithm.
The DIF FFT algorithm is developed as follows by splitting the data sequence g[n] into two halves: n = 0 → N/2–1
and n = N/2 → N − 1
∑
N−1
∑
(N∕2)−1
∑
N−1
G(k) = g(n)WNkn = g(n)WNkn + g(n)WNkn
n=0 n=0 n=N∕2

Making a change of index m ≡ n − N/2 in the second summation, its summation limits become m = 0 → (N/2) − 1;
k(m+N∕2)
g(n) becomes g(m + N/2) and WNkn becomes WN . We then return to using index n in this summation simply
by replacing m with n. Thus

∑
(N∕2)−1
∑
(N∕2)−1
k(n+N∕2)
G(k) = g(n)WNkn + g(n + N∕2)WN
n=0 n=0

∑
(N∕2)−1
kN∕2 N∕2
= [g(n) + (−1)k g(n + N∕2)]WNkn , since WN = (WN )k = (−1)k
n=0

Separating the even points of the transform {G(0), G(2), …, G(N − 2)} at which (−1)k = 1 from the odd points
{G(1), G(3), …, G(N − 1)} at which (−1)k = −1 leads to
(N∕2)−1 [ ( )]
∑ N
G(2m) = g(n) + g n + WN2mn
n=0
2
∑
(N∕2)−1
= mn
ga (n)WN∕2 , since WN2mn = (WN2 )mn = (WN∕2 )mn
n=0
N
≡ Ga (m), m = 0, 1, · · · , −1
2
And
Telegram: @Persian_IEEEComSoc

(N∕2)−1 [ ( )]
∑ N
G(2m + 1) = g(n) − g n + WN(2m+1)n
n=0
2
(N∕2)−1 [ ( )]
∑ N
= g(n) − g n + WNn WN2mn
n=0
2
(N∕2)−1 [( ( )) ]
∑ N
= g(n) − g n + WNn WN∕2
mn

n=0
2
∑
(N∕2)−1
N
= mn
gb (n)WN∕2 ≡ Gb (m), m = 0, 1, · · · , −1
n=0
2
 4.4 Discrete Fourier Transform 279

where
Ga [m] is the (N∕2)-point DFT of the sequence {ga (n)};
ga (n) = g(n) + g(n + N∕2); n = 0, 1, … , N∕2 − 1
Gb [m] is the (N∕2)-point DFT of the sequence {gb (n)}; and
gb (n) = [g(n) − g(n + N∕2)]WNn
n+N∕2
= g(n)WNn + g(n + N∕2)WN ; n = 0, 1, … , N∕2 − 1

What we have achieved above is that an N-point DFT has been reduced to two (N/2)-point DFTs. This process
is repeated until after L = log2 N stages we reach trivial N single-point DFTs where each transform is simply equal
to the data sample. The following two worked examples for N = 4 and N = 8 should serve to clarify this algorithm
further.

Worked Example 4.11 DIF FFT for N = 4

We wish to use the DIF FFT algorithm to compute the DFT G[k] of data sequence g[n] of length N = 4 samples,
where
g[n] = {g(0), g(1), g(2), g(3)} ⇌ G[k] = {G(0), G(1), G(2), G(3)}
Since N = 4, the algorithm is implemented in L = log2 4 = 2 stages as follows:
STAGE 1, N = 4: we have one 4-point DFT, namely G[k] to compute, hence N = 4. The 4-point DFT is reduced to
two 2-point DFTs denoted Ga [m] for even points of G[k], and Gb [m] for odd points. Since G[k] = {G(0), G(1), G(2),
G(3)}, its even points are Ga [m] = {G(0), G(2)} and its odd points are Gb [m] = {G(1), G(3)}.
Ga [m] is the DFT of the data sequence ga [n] obtained from the original sequence g[n] as ga [n] = {g(n) + g(n +
N/2)}, n = 0 to N/2 – 1, which is n = 0, 1. Therefore, ga [n] = {g(0) + g(2), g(1) + g(3)}.
Next, Gb [m] is the DFT of the data sequence gb [n] obtained from the original sequence g[n] as gb [n] = {g(n)WNn +
n+N∕2
g(n + N∕2)WN }; N = 4; n = 0, 1. Thus, gb [n] = {g(0)W40 + g(2)W42 , g(1)W41 + g(3)W43 }, which, with
W40 = 1, W41 = −j, W42 = −1, and W43 = j, reduces to gb [n] = {g(0) − g(2), j(g(3) − g(1))}.
This completes stage 1, which we may summarise as follows
ga [n] = {g(0) + g(2), g(1) + g(3)} ⇌ Ga [m] = {G(0), G(2)}
gb [n] = {g(0) − g(2), j(g(3) − g(1))} ⇌ Gb [m] = {G(1), G(3)} (4.120)

STAGE 2, N = 2: we now have two 2-point DFTs, Ga [m] and Gb [m] to compute, hence N = 2. Each DFT is reduced
to two single-point DFTs. Reaching single-point DFTs indicates the end of the process. We use the subscript nota-
tions aa for the even point of Ga [m], ba for the odd point of Ga [m], ab for the even point of Gb [m], and bb for the
odd point of Gb [m].
Telegram: @Persian_IEEEComSoc

From Eq. (4.120), we see that the desired first transform point G(0) is the even point of Ga [m], which is denoted
Gaa [m], and this is the DFT of gaa [n] obtained from ga [n] as gaa [n] = {ga (n) + ga (n + N/2)}, n = 0 to N/2 – 1, which
is n = 0, since N = 2. Therefore, gaa [n] = {ga (0) + ga (1)} = g(0) + g(2) + g(1) + g(3).
Again from Eq. (4.120), the desired second transform point G(1) is the even point of Gb [m], which is denoted
Gab [m], and this is the DFT of gab [n] obtained from gb [n] as gab [n] = {gb (n) + gb (n + N/2)}, n = 0.
Therefore, gab [n] = {gb (0) + gb (1)} = g(0) − g(2) + j(g(3) – g(1)).
From Eq. (4.120), the third transform point G(2) is the odd point of Ga [m], which is denoted Gba [m], and this is
the DFT of gba [n] obtained from ga [n] as
n+N∕2
gba [n] = {ga (n)WNn + ga (n + N∕2)WN }; N = 2; n=0
= ga (0)W20 + ga (1)W21 = ga (0) − ga (1)
 280 4 Frequency Domain Analysis of Signals and Systems

Substituting the values of ga [n] from Eq. (4.120) gives

gba [n] = ga (0) − ga (1) = g(0) + g(2) − g(1) − g(3)
Finally, from Eq. (4.120), the fourth transform point G(3) is the odd point of Gb [m], which is denoted Gbb [m],
n+N∕2
and this is the DFT of gbb [n] obtained from gb [n] as gbb [n] = {gb (n)WNn + gb (n + N∕2)WN } = gb (0) − gb (1).
Substituting the values of gb [n] from Eq. (4.120) yields gbb [n] = g(0) − g(2) − j(g(3) − g(1)).
This completes our work. To summarise, the DFT of the 4-point data sequence g[n] = {g(0), g(1), g(2), g(3)}
is the 4-point transform sequence G[k] = {G(0), G(1), G(2), G(3)}, where G(0) ≡ Gaa [m] is the DFT of the single
data point gaa [n] = g(0) + g(2) + g(1) + g(3), so G(0) = g(0) + g(2) + g(1) + g(3). Similar comments apply to the other
computations presented above. Thus
G(0) = gaa [n] = g(0) + g(1) + g(2) + g(3)
G(1) = gab [n] = g(0) − g(2) + j(g(3) − g(1))
G(2) = gba [n] = g(0) + g(2) − g(1) − g(3)
G(3) = gba [n] = g(0) − g(2) − j(g(3) − g(1))
If the data sequence g[n] = {1, 0, 0, 1} as in the previous worked example then we obtain G[k] =
{G(0), G(1), G(2), G(3)} = {2, 1 + j, 0, 1 − j}, which agrees with the result obtained earlier.
You are right to feel that the direct computation of the DFT in Worked Example 4.10 was more straightforward.
The FFT algorithm is designed for computer code implementation (as will become clearer shortly) and it achieves
a huge saving in computation time when N is large.

Worked Example 4.12 DIF FFT for N = 8

We wish to use the DIF FFT algorithm to compute the DFT G[k] of data sequence g[n] of length N = 4 samples,
where
g[n] = {g(0), g(1), g(2), g(3), g(4), g(5), g(6), g(7)}
G[k] = {G(0), G(1), G(2), G(3), G(4), G(5), G(6), G(7)}
A more detailed explanation of the algorithm steps is given in the last worked example, which should be studied
first before attempting this one. Since N = 8, the algorithm is implemented in L = log2 8 = 3 stages as follows:
STAGE 1, N = 8: the even points of G[k] are Ga [m] = {G(0), G(2), G(4), G(6)} and this is the DFT of ga [n] =
{g(n) + g(n + 4)}; n = 0, 1, 2, 3. Thus
ga [n] = {g(0) + g(4), g(1) + g(5), g(2) + g(6), g(3) + g(7)}
⇌ Ga [m] = {G(0), G(2), G(4), G(6)}
The odd points of G[k] are Gb [m] = {G(1), G(3), G(5), G(7)} which is the DFT of
Telegram: @Persian_IEEEComSoc

gb [n] = {g(n)W8n + g(n + 4)W8n+4 }; n = 0, 1, 2, 3

≡ {gb (0), gb (1), gb (2), gb (3)}; gb (0) = g(0)W80 + g(4)W84 ; etc.
.
STAGE 2, N = 4: decompose each of the 4-point DFTs, Ga [m] and Gb [m], into the following 2-point DFTs:
Even points of Ga [m]
gaa [n] = {ga (n) + ga (n + 2)} ⇌ Gaa [p] = {G(0), G(4)}; n = 0, 1
Odd points of Ga [m]
gba [n] = {ga (n)W4n + ga (n + 2)W4n+2 } ⇌ Gba [p] = {G(2), G(6)}; n = 0, 1
 4.4 Discrete Fourier Transform 281

Even points of Gb [m]

gab [n] = {gb (n) + gb (n + 2)} ⇌ Gab [p] = {G(1), G(5)}; n = 0, 1
Odd points of Gb [m]
gbb [n] = {gb (n)W4n + gb (n + 2)W4n+2 } ⇌ Gbb [p] = {G(3), G(7)}; n = 0, 1
STAGE 3, N = 2: take the even point and odd point of each of the above 2-point DFTs to decompose them into the
following single-point DFTs
gaaa [n] = {gaa (n) + gaa (n + 1)} ⇌ Gaaa [q] = G(0); n=0

gbaa [n] = {gaa (n)W2n + gaa (n + 1)W2n+1 } ⇌ Gbaa [q] = G(4); n=0

gaba [n] = {gba (n) + gba (n + 1)} ⇌ Gaba [q] = G(2); n=0

gbba [n] = {gba (n)W2n + gba (n + 1)W2n+1 } ⇌ Gbba [q] = G(6); n=0

gaab [n] = {gab (n) + gab (n + 1)} ⇌ Gaab [q] = G(1); n=0

gbab [n] = {gab (n)W2n + gab (n + 1)W2n+1 } ⇌ Gbab [q] = G(5); n=0

gabb [n] = {gbb (n) + gbb (n + 1)} ⇌ Gabb [q] = G(3); n=0

gbbb [n] = {gbb (n)W2n + gbb (n + 1)W2n+1 } ⇌ Gbbb [q] = G(7); n=0
We may then work backwards from this final stage up to the first stage to express each transform point as a
linear combination of samples of the data sequence g[n]. To demonstrate this for G(1): the stage 3 result is used to
express G(1) in terms of gab [n], then the stage 2 result is used to replace gab [n] with gb [n], and finally the stage 1
result is used to replace gb [n] with g[n]. Thus
G(1) = gab (n) + gab (n + 1); n=0

= gab (0) + gab (1)

= gb (0) + gb (2) + gb (1) + gb (3)

= g(0)W80 + g(4)W84 + g(2)W82 + g(6)W86

+ g(1)W81 + g(5)W85 + g(3)W83 + g(7)W87

1
= g(0) − g(4) − jg(2) + jg(6) + √ (1 − j)g(1)
2
Telegram: @Persian_IEEEComSoc

1 1 1
− √ (1 − j)g(5) − √ (1 + j)g(3) + √ (1 + j)g(7)
2 2 2
The FFT algorithm may be conveniently and elegantly presented in the form of a signal flow graph, as shown
in Figure 4.37 for the above 8-point DIF FFT. A signal flow graph is an interconnection of nodes and branches
having the following characteristics:
● Direction of signal flow along a branch is indicated by an arrow.
● The input of a node is the sum of signals in all branches entering that node.
● The signal in a branch exiting a node equals the input of that node.
● A branch modifies a signal flowing through it by the branch transmittance, which may be a multiplication by 1
(i.e. no change) if the branch has no label or coefficient, or a multiplication by the branch coefficient or label.
 282 4 Frequency Domain Analysis of Signals and Systems

Output
location:
ga(0) gaa(0) gaaa(0)
g(0) + + + G(0) 000
W40 W20
W80
ga(1) gaa(1) gbaa(0)
g(1) + + + G(4) 001
W41 W21
W81
ga(2) gba(0) gaba(0)
g(2) + W42 + + G(2) 010
W20
W82
ga(3) gba(1) gbba(0)
g(3) + + + G(6) 011
W43 W21
W83
gb(0) gab(0) gaab(0)
g(4) W84 + + + G(1) 100
W20
W40
gb(1) gab(1) gbab(0)
g(5) W85 + + + G(5) 101
W41 W21
gb(2) gbb(0) gabb(0)
g(6) + W42 + + G(3) 110
W86 W20
gb(3) gbb(1) gbbb(0)
g(7) + + + G(7) 111
W78 W43 W21
Twiddle factor Butterfly

Figure 4.37 Signal flow graph for an 8-point DIF FFT.

The signal flow graph of the FFT algorithm consists of a repetitive structure, called a butterfly, with two inputs
of the form x(n) and x(n + N/2) and two outputs y1 and y2 . One such butterfly is shaded in the bottom right side
in Figure 4.37. The downward going diagonal and bottom branches of each butterfly has a coefficient known as
m
a twiddle factor of the form WM , where M = N at the first stage and decreases progressively by a factor of two
through the stages, and m ≤ M − 1. There are N/2 butterflies per stage and log2 N stages. Each butterfly performs
computations of the form
Telegram: @Persian_IEEEComSoc

y1 = x(n) + x(n + N∕2)

n+N∕2
y2 = x(n)WNn + x(n + N∕2)WN = [x(n) − x(n + N∕2)]WNn

This means that there are one complex addition, one complex subtraction, and one complex multiplication per
butterfly, or a total of (N/2)log2 N complex multiplications and Nlog2 N complex additions and subtractions in
the FFT algorithm. This FFT algorithm is therefore computationally much more efficient than the direct DFT
computation algorithm, with a saving of a factor of 2 N/log2 N on complex multiplications and (N − 1)/log2 N on
complex additions and subtractions. The output ports of the signal flow graph have been numbered in binary on the
right-hand side of Figure 4.37 to show that the algorithm produces the transform sequence stored in bit-reversed
order. If the algorithm output is stored in an N-element MATLAB vector Gk, where N is an integer power of 2 (as
 4.4 Discrete Fourier Transform 283

G[k] Complex G*[k] Ng*[n] Complex Ng[n] g[n]

FFT 1/N
conjugate conjugate

IFFT

Figure 4.38 Inverse FFT algorithm.

required) then the following MATLAB code will restore the right order so that G(0) is in Gk(1), G(1) is in Gk(2),
G(2) is in Gk(3), and so on

n = (0:N-1)'; Gk = Gk(1+bi2de(de2bi(n,log2(N)), 'left-msb'));

We may devise a simple way to compute inverse DFT using the computationally efficient FFT algorithm devel-
oped above if we exploit the similarity between the DFT and IDFT expressions repeated below for convenience.
∑
N−1
2𝜋
G(k) = g(n)e−j N kn , DFT
n=0

1 ∑
N−1
2𝜋
g(n) = G(k)ej N kn , IDFT
N k=0

Taking the complex conjugate of both sides of the IDFT expression yields
∑
N−1
2𝜋
Ng∗ (n) = G∗ (k)e−j N kn (4.121)
k=0

The right-hand side of this Eq. (4.121) is similar in form to the DFT expression implemented by the FFT algo-
rithm. It indicates that if we take the complex conjugate of G[k] before inputting it into the FFT algorithm then
the output of the algorithm will be N times the complex conjugate of the data sequence g[n]. A block diagram of
this inverse FFT algorithm is shown in Figure 4.38.

4.4.3 Practical Issues in DFT Implementation

The DFT sequence G[k] = {G(0), G(1), …, G(N − 1)}, computed from N samples of a nonperiodic signal g(t) taken
within an interval T to produce a data sequence g[n], gives the spectrum of g[n], being the magnitude (per unit
frequency) and phase of the complex exponential components of the signal rect(t/T)g(t) at respective frequen-
cies {0, Δf , 2Δf , …, (N/2)Δf }, where Δf = 1/T = 1/(NT s ) = F s /N is the frequency resolution, T s is the sampling
interval, and F s = 1/T s is the sampling frequency. The spectrum is folded over at frequency f = (N/2)Δf Hz, as
Telegram: @Persian_IEEEComSoc

earlier discussed. There are several practical issues which should be taken into consideration when evaluating and
interpreting the DFT.

4.4.3.1 Aliasing
Alias distortion is a phenomenon whereby a high-frequency component in g(t) is reproduced within the spectrum
G[k] of the sampled signal g[n] at a false lower frequency. To avoid this distortion, the signal g(t) must be passed
through an anti-alias filter to limit its maximum frequency component or bandwidth (whichever is lower) to f max
prior to sampling. Note that a bandpass signal will contain frequency components much higher than the signal
bandwidth, in which case f max here refers to bandwidth, whereas in a lowpass or baseband signal, the bandwidth
and maximum frequency component are equal. Once f max has been set in this way by filtering, the sampling rate
is chosen as F s ≥ 2f max to obey the sampling theorem and therefore avoid alias distortion.
 284 4 Frequency Domain Analysis of Signals and Systems

4.4.3.2 Frequency Resolution

The spectrum G[k] is a sequence that represents a discrete selection – spaced apart in frequency by the frequency
resolution Δf – of the continuum of complex exponentials that constitute g(t). It is therefore important to make Δf
small in order not to skip over or miss a significant frequency component in g(t) and the resulting data sequence
g[n]. With sampling rate F s already fixed as above by the sampling theorem and the span T of the data sequence also
fixed, the frequency resolution Δf = 1/T = F s /N may be reduced by using a higher value of N in the DFT computa-
tion. This is achieved by appending N 1 trailing zeros to g[n] = {g(0), g(1), …, g(N − 1)} such that N + N 1 = 2m , and
frequency points of G[k] form a faithfully representative selection from the continuum of frequency components
of g(t).
The frequency resolution issue is illustrated in Figure 4.39 in which the DFT of a sampled signal of duration
T = 5 s is computed at two different resolutions Δf . The signal g(t) is the sum of two sinusoids at frequencies 1 Hz
and 1.3 Hz with respective amplitudes 1 and 0.5. The data sequence g[n] was obtained by taking N = 512 samples
of g(t) within the interval T = 5 s, which corresponds to a sampling rate of F s = 102.4 Hz that more than satisfies
the sampling theorem, so there is no alias distortion issue. The middle plot shows the magnitude spectrum |G(k)|
of g[n] computed at a resolution of Δf = 0.2 Hz, using the unpadded data sequence g[n]. The spectrum is shown in
logarithmic units of dB relative to the maximum amplitude. We see that the frequency component of 1 Hz has been
captured in |G(k)| and is depicted correctly at a relative amplitude of 0 dB, but the other frequency component of
1.3 Hz has been skipped over and missed in |G(k)| due to the poor resolution employed. In the bottom plot, the
frequency resolution of the DFT computation was improved to Δf = 0.1 Hz by padding g[n] with 512 trailing
zeros, so that Δf = F s /N = 102.4/1024 = 0.1 Hz. We see that both frequency components of 1 Hz and 1.3 Hz are

1.5
g(t)rect(t/T)

–1.5
T = 5s
0
–10 Δf = 1/T = 0.2 Hz
|G(k)|, dB

–20
–30
–40
–10 –5 0 5 10
Telegram: @Persian_IEEEComSoc

k→
0
–10 Δf = 0.1 Hz
|G(k)|, dB

–20
–30
–40
–20 –10 0 10 20
k→

Figure 4.39 DFT analysis of 5s segment of g(t) at two different resolutions Δf using a rectangular window. Signal g(t)
comprises 2 sinusoids at 1Hz and 1.3 Hz having relative amplitudes 0 dB and −6 dB, respectively.
 4.4 Discrete Fourier Transform 285

now captured in the spectrum. However, the relative amplitude of the 1.3 Hz component is −5.2 dB (rather than
its true −6 dB value). This is because of spectral leakage, which is further discussed below.

4.4.3.3 Spectral Leakage

Each spectral line of g(t) is replaced by a scaled version of the amplitude spectrum W(f ) of the windowing function
w(t). The sidelobes of W(f ) will somewhat corrupt adjacent spectral components, a problem described as spectral
leakage. The default windowing function w(t) = rect(t/T) has significant spectral sidelobes in W(f ), and this is
the window employed in Figure 4.39. We see in the bottom plot of that figure that the frequency component at
1.3 Hz, which is present in g(t) at a relative amplitude of 20 log(0.5) = −6 dB, appears in the spectrum at a relative
amplitude of −5.2 dB. This enhancement of its amplitude by 0.8 dB is due to interference from a strong spectral
sidelobe of the 1 Hz component.
Spectral leakage may be minimised by employing a windowing function that tapers at the ends and thus has
smaller and more rapidly decaying sidelobes in its spectrum than those of the spectrum of a rectangular window.
This makes spectral leakage into adjacent spectral lines negligible if these spectral lines are separated by more
than the width of the main lobe of the windowing function.

4.4.3.4 Spectral Smearing

When the main lobes of adjacent spectral components overlap, there will be a smearing of their spectral lobes and
hence a blurring of the distinction between the spectral components involved. Windows with suppressed sidelobes
will have wider main lobes and therefore their reduction of spectral leakage is achieved at the expense of a higher
level of spectral smearing. The default rectangular window corresponds to the sequence w(n) = 1, n = 0, 1, 2, …,
N − 1, where NT s = T is the duration of the signal and F s = 1/T s is the sampling rate. Several tapered windows
have been defined, the most common of which include:
● Hamming window
( )
25 21 2𝜋n
w(n) = − cos ; n = 0, 1, 2, · · · , N − 1 (4.122)
46 46 N −1

● Raised cosine or Hanning window

[ ( )] ( )
1 2𝜋n n𝜋
w(n) = 1 − cos = sin2 ; n = 0, 1, 2, · · · , N − 1 (4.123)
2 N −1 N −1

● Blackman–Harris window
( ) ( ) ( )
2𝜋n 4𝜋n 6𝜋n
w(n) = a0 − a1 cos + a2 cos − a3 cos
N −1 N −1 N −1
a0 = 0.35875; a1 = 0.48829; a2 = 0.14128; a3 = 0.01168;
Telegram: @Persian_IEEEComSoc

n = 0, 1, 2, · · · , N − 1 (4.124)

● Kaiser–Bessel window
( √ )
( )2
2n
I0 𝜋𝛼 1− N−1
−1
w(n) = ; n = 0, 1, 2, · · · , N − 1
I0 (𝜋𝛼)
1
2𝜋 ∑∞
(x∕2)2m ∑ (x∕2)2m
32
I0 (x) = exp(x cos 𝜙)d𝜙 = ≈ (4.125)
2𝜋 ∫0 m=0
(m!)2 m=0
(m!)2
 286 4 Frequency Domain Analysis of Signals and Systems

Rectangular window Blackman-Harris window

1

0
T T

0
1/T
–20
Amplitude, dB →

–40
–60
–80
–100
–120
0 Frequency → 0 Frequency →

Figure 4.40 Rectangular and Blackman–Harris windows and their amplitude spectra.

where I 0 is the zeroth order modified Bessel function of the first kind, and 𝛼 is a variable parameter that decides
the trade-off between main lobe width and sidelobe level. Typically, 𝛼 = 3.
Figures 4.40 and 4.41 show plots of the time domain functions and amplitude spectra of some of the above win-
dows. Notice, for example, how the Blackman–Harris window has negligible sidelobes, with levels below −90 dB,
but its main lobe is four times as wide as the main lobe of the rectangular window. The rectangular window, on
the other hand, has the narrowest main lobe of all the windows but has the highest sidelobe levels, up to −13.3 dB
for the first sidelobe.
Figure 4.42a–c show the results of a high-resolution DFT analysis (at Δf = 1/32 Hz) of a 0.5 s segment of a signal
g(t) that consists of two sinusoids of relative amplitudes 0 dB and − 5 dB and frequencies as specified in each plot.
In (a) the signal was multiplied by a raised cosine window to extract the 0.5 s segment and N = 512 samples were
taken to produce the data sequence g[n], which was then padded with 32 256 zeros before the FFT algorithm was
used to compute the amplitude spectrum |G(k)| shown. A similar procedure was followed for plots (b) and (c), but
with a rectangular window used in (b). We see in (a) that the two spectral components have been clearly detected
by the DFT analysis and their lobes are distinct and at the correct amplitudes. There are no discernible effects of
spectral leakage or spectral smearing. In (b), involving two spectral components at 10 and 14 Hz, the rectangular
window avoids spectral smearing (due to its narrow main lobe), but there is discernible spectral leakage effect
Telegram: @Persian_IEEEComSoc

as the peak of the 14 Hz lobe is at −4 dB (an increase of 1 dB) and has been shifted slightly away from its true
14 Hz location. In (c), the raised cosine window, with its wide main lobe, creates a significant overlap between the
main lobes of the 10 and 14 Hz spectral components. This produces significant spectral smearing, which blurs the
identity of the 14 Hz component.
Spectral smearing between the 10 and 14 Hz components was avoided in Figure 4.42b by using a rectangular
window. However, if the spectral components are sufficiently close then even the rectangular windowed data will
also experience spectral smearing. Padding the windowed data with more zeros will increase frequency resolution
but is ineffective in reducing spectral smearing. One way to reduce spectral smearing (in order to be able to detect
more closely spaced spectral components) is to increase the window duration T so that its main spectral lobe
becomes proportionately narrower. This solution requires observing the original signal g(t) and data sequence
g[n] over a longer duration T. Figure 4.42d demonstrates the effectiveness of a longer observation interval T in
 4.4 Discrete Fourier Transform 287

Hanning (Raised-Cosine) window Hamming window

1

0
T T
0
–20
Amplitude, dB →

–40
–60
–80
–100
–120
–6/T –4/T –2/T 0 2/T 4/T 6/T –6/T –4/T –2/T 0 2/T 4/T 6/T
Frequency → Frequency →

Figure 4.41 Raised-cosine and Hamming windows and their amplitude spectra.

0
No spectral leakage;
|G(k)|, dB

No spectral smearing
(a)

–40 →f, Hz
–20 –10 0 10 20
0
Spectral leakage;
No spectral smearing
|G(k)|, dB

(b)

–25 →f, Hz
–14 –10 0 10 14
0
Spectral smearing
|G(k)|, dB
Telegram: @Persian_IEEEComSoc

(c)

–40 →f, Hz
0
No spectral leakage;
|G(k)|, dB

No spectral smearing
(d)

–40 →f, Hz
–14 –10 0 10 14

Figure 4.42 High resolution DFT analysis of 0.5 s segment, (in (a) – (c)), and 1 s segment, (in (d)), of a signal g(t) that
contains two frequencies f 1 = 10 Hz and f 2 at relative amplitudes 0 dB and −5 dB: (a) Raised-cosine window; f 2 = 20 Hz;
(b) Rectangular window; f 2 = 14 Hz; (c) Raised-cosine window; f 2 = 14 Hz; (d) Raised-cosine window; f 2 = 14 Hz.
 288 4 Frequency Domain Analysis of Signals and Systems

combating spectral smearing. The observation interval used for DFT analysis was doubled to T = 1 s in (d). As a
result, the width of the main lobe of the raised cosine window was halved. Using this window on the data produced
a spectrum free of both spectral leakage and spectral smearing between the 10 and 14 Hz components, as seen in
Figure 4.42d.
The width W of the main spectral lobe is different for each window, but may in general be expressed as the
reciprocal of window duration T in the form
M
W= (4.126)
T
where the value of M depends on window type. For example, from Figure 4.34 and Worked Example 4.9, M = 2
(the lowest possible) for a rectangular window and M = 4 for a raised cosine window. For two spectral components
at frequencies f 2 and f 1 to be resolvable (i.e. spectral smearing between them is avoided) in a DFT analysis, their
frequency difference must be no smaller than W. That is
M M
f2 − f1 = ; ⇒ T= (4.127)
T f2 − f1
Thus, for example, if we wish to resolve the spectral components in a signal segment down to a frequency differ-
ence of 1 Hz then the segment must be at least four seconds long (assuming a raised cosine window is employed).
Of course, you may reduce the required segment length to two seconds by using a rectangular window, but spectral
leakage may then cause a significant distortion in the results obtained.

4.4.3.5 Spectral Density and Its Variance

We now have all the tools we need to look at the distribution of the energy or power of a signal amongst its various
frequency components. This information is important in revealing any hidden repetitive patterns within the signal,
identifying the location of strong or dominant spectral components, guiding decision making in the allocation of
system resources for signal coding across the various frequency bands of the signal, and providing a means to
compute the amount of energy or power in all or selected sub-bands of the signal.
It was established in Eq. (3.148) that the autocorrelation function Rg (𝜏) of signal g(t) is the convolution (in the
time domain) of g(t) with its time-reversed version. That is

Rg (𝜏) = g(𝜏) ∗ g(−𝜏)

We now know from Eq. (4.91) that if we take the FT of both sides of the above equation the right-hand side of
the result will be the multiplication of the FT of g(t), denoted G(f ), with the FT of g(−t), which we know from
Eq. (4.84) to be G(−f ). Thus

F[Rg (𝜏)] = G(f )G(−f )

Telegram: @Persian_IEEEComSoc

If g(t) is a real signal then G(−f ) equals the complex conjugate of G(f ), according to Eq. (4.86), and this allows
us to write

F[Rg (𝜏)] = G(f )G∗ (f )

= |G(f )|2 (4.128)

From the discussion surrounding Eq. (4.74), if g(t) is in volts (V) then |G(f )| is in V/Hz and therefore the squared
magnitude of the FT on the right-hand side of the above equation is a quantity in V2 /Hz2 , which (noting that
Hz ≡ s−1 ) may be manipulated as follows
V2 V 2 •s W •s J
2
= = = ≡ joules∕hertz
Hz Hz Hz Hz
 4.4 Discrete Fourier Transform 289

The squared magnitude of the FT of g(t) is therefore a quantity in units of energy per unit frequency, called the
energy spectral density (ESD) of g(t) and denoted Ψg (f ). We conclude from Eq. (4.128) that the FT of the autocor-
relation function of an energy signal is the ESD of the signal. That is
Rg (𝜏) ⇌ |G(f )|2 = Ψg (f ) ≡ ESD in J∕Hz (4.129)
The energy in an infinitesimally small frequency band (f , f + df ) is Ψg (f )df , and the total energy E of the signal
is obtained by summing these contributions over the entire frequency axis −∞ < f < ∞ so that
∞ ∞
E= Ψg (f )df = |G(f )|2 df
∫−∞ ∫−∞
∞
=2 |G(f )|2 df (even symmetry) (4.130)
∫0
Since signal bandwidth is the range of significant positive frequencies in the signal, it is important to note that
Energy per unit bandwidth = 2 × ESD (4.131)
Equating the above frequency domain expression for energy with the time domain formula given in Eq. (3.101)
for the energy of CT signals leads to Parseval’s theorem stated in Eq. (4.94) and before that in Eq. (3.102).
We may extend the concept of spectral density to power signals g(t), including ergodic (and hence stationary)
random processes, by applying the above analysis to a segment of g(t) of duration 𝕋 , spanning the interval −𝕋 ∕2 ≤
t ≤ 𝕋 ∕2. The FT of this segment of g(t) is denoted G𝕋 (f ) so that its energy per unit frequency is |G𝕋 (f )|2 as earlier
established. Since power is the average rate of energy, or energy per unit of time, the power per unit frequency is
|G𝕋 (f )|2 ∕𝕋 , a quantity which has units of (J/Hz)/s or W/Hz. The interval of analysis is then extended to cover the
entire signal g(t) by letting 𝕋 → ∞. The power spectral density (PSD) of g(t), denoted Sg (f ), is therefore given by
|G𝕋 (f )|2
Sg (f ) = lim (4.132)
𝕋 →∞ 𝕋
If the power signal is periodic, with period T, then it is represented by the complex exponential Fourier series
of Eq. (4.23), which shows that the signal contains complex exponentials of magnitude |Cn |, given by Eq. (4.24),
and power |Cn |2 at frequencies nf o for n = 0, ±1, ±2, ±3, … The frequency spacing is f o , and the power per unit
frequency is a function of nf o given by
|Cn |2
Sg (nf o ) = = |Cn |2 T (4.133)
fo
Seeing in Eq. (3.114) that the definition of autocorrelation function for a power signal has the same form as
that for an energy signal except for the factor lim 𝕋1 , it follows by the same steps as for energy signals that the
𝕋 →∞
autocorrelation function of a power signal and its PSD form a FT pair, so we may write
Telegram: @Persian_IEEEComSoc

Rg (𝜏) ⇌ Sg (f ) ≡ PSD in W∕Hz (4.134)

The power P of g(t) is obtained by integrating its PSD over all frequencies. Thus
∞ ∞
E= Sg (f )df = 2 Sg (f )df (4.135)
∫−∞ ∫0
And this also means that
Power per unit bandwidth = 2 × PSD (4.136)
It is important to always remember the relationship stated Eq. (4.136). For example, white noise has a constant
power per unit bandwidth denoted N o , so the PSD of white noise is N o /2, but the total white noise power in a
bandwidth B is P = N o B.
 290 4 Frequency Domain Analysis of Signals and Systems

0 N/4 N/2 3N/4 N

(a) Three half-overlapping segments: N = 2m + 1

0 N/8 N/4 3N/8 N/2 5N/8 3N/4 7N/8 N

One data segment

of length M = 2m

(b) Seven half-overlapping segments: N = 2m + 2

Figure 4.43 Partitioning of data of length N into half-overlapping segments. Both the data length and the segment length
are integer powers of 2: (a) three half-overlapping segments; and (b) seven half-overlapping segments.

A plot of spectral density as a function of frequency is often called a periodogram. When applied to nondeter-
ministic signals, such as a random process or data contaminated by noise, the periodogram will exhibit spectral
variance because the exact values of the transform sequence G[k] computed on the data sequence g[n] will some-
what vary from one sample function (i.e. observation over duration 𝕋 ) of the random process to the next or due
to random changes in the values of the contaminating noise samples. One way to reduce this variance and come
closer to a true estimate of the spectral density of such signals is to compute the periodogram over each of sev-
eral half-overlapping segments of the data and then to average the computed periodograms. Each data segment is
multiplied by an appropriate window function, as earlier discussed, before its DFT is computed. Figure 4.43 shows
the partitioning scheme for three and seven half-overlapping segments. If both the original data length N and the
segment length M are required to be integer powers of two then there are strict constraints on the values of N and
M and the number S of half-overlapping data segments. For example, a 1024-length data sequence can be parti-
tioned into three half-overlapping segments each of length 512. A 2048-length data sequence can be partitioned
Telegram: @Persian_IEEEComSoc

into seven half-overlapping 512-length segments. And so on. In general, the possible values are

S = 2k − 1; k = 2, 3, 4, · · ·
M=2 ; m
m ≡ positive integer ≥ 5
m+k−1
N=2 (4.137)

If the samples of the N-length data sequence are numbered n = 0, 1, 2, …, N − 1, and the segments are numbered
i = 0, 1, 2,…, S − 1, then the ith segment will span the data samples

N N
n= i → n = k (i + 2) − 1; i = 0, 1, 2, … , S − 1 (4.138)
2k 2
 4.5 Laplace and z-transforms 291

–5
Magnitude, dB

–10

–15

–20
→ f, Hz
0 50 100 150 200 250

Figure 4.44 Periodogram of noise contaminated double-sinusoid signal.

Often, we will know the data length N and wish to determine the segment length M and number of segments
S. In that case, we work backwards in Eq. (4.137) and write
k + m = 1 + log2 N (4.139)
We then choose k to give the required number of segments as S = 2k − 1. This choice of k leaves us with only
one possible value for m = 1 + log2 N − k. For example, if N = 4096 = 212 , then k + m = 13; and if we choose k = 3
(for seven half-overlapping segments) then m = 10, which gives M = 210 = 1024. That is, a 4096-length data may
be partitioned into seven half-overlapping 1024-length segments. Other segment lengths are possible by choosing
a different value for k.
Figure 4.44 shows the periodogram (in dB relative to maximum level) of a noise contaminated data sequence
that contains two sinusoids at frequencies 100 and 200 Hz with relative amplitudes 0 dB and − 5 dB. The analysis
was carried out as above over an 8 s segment of the data, using a raised cosine window and F s = 512 Hz, N = 4096,
S = 7, M = 512.

4.5 Laplace and z-transforms

In the previous three sections, we discuss Fourier analysis in detail with an emphasis on its engineering appli-
cations and the ability not only to formulate or derive the mathematical expressions but to graphically illustrate,
interpret, and apply the results. This foundation is enough for our needs of communication systems analysis and
design in this book. However, for the sake of completeness and to see how they fit in with Fourier techniques,
there are two related transforms that we must briefly introduce.
Telegram: @Persian_IEEEComSoc

4.5.1 Laplace Transform

The Laplace transform (LT) is defined for an arbitrary function or signal g(t) as
∞
G(s) = g(t)e−st dt, s = 𝜎 + j𝜔 (4.140)
∫−∞
The parameter s is a complex variable with real part 𝜎 and imaginary part 𝜔 ≡ 2𝜋f that takes on values in the
so-called s-plane shown in Figure 4.45a. Explicitly using these real and imaginary parts in the above definition
allows us to express the LT in the form
∞ ∞
G(s)|s=𝜎+j2𝜋f = [g(t)e−𝜎t ]e−j𝜔t dt ≡ [g(t)e−𝜎t ]e−j2𝜋ft dt (4.141)
∫−∞ ∫−∞
 292 4 Frequency Domain Analysis of Signals and Systems

jω Im (z)
f = Fs/4

f=0
f = Fs/2

n
n

gio
gio

σ Re(z)
re
re

e
le

bl
ab

sta
St

Un

Unit circle

(a) s-plane (b) z-plane

Figure 4.45 s- and z-planes.

Thus, the LT of g(t) is the Fourier transform (FT) of g(t)e−𝜎t . So, provided g(t) is a practical signal (i.e. it is finite
and does not have an infinite number of discontinuities) then its LT will converge if the FT of g(t)e−𝜎t exists, and
this means if g(t)e−𝜎t is absolutely integrable as follows
∞
|g(t)e−𝜎t |dt < ∞ (4.142)
∫−∞
The LT therefore exists for some causal signals g(t)u(t) that do not have a FT. In such cases, the region of conver-
gence (ROC) of the LT is the region 𝜎 ≥ 𝜎 1 in the s-plane where 𝜎 has a large enough value to make g(t)e−𝜎t decay
sufficiently rapidly with time to satisfy Eq. (4.142).
Setting 𝜎 = 0 in Eq. (4.141), we find that the LT of g(t) becomes identical with the FT of g(t). So, provided g(t) is
absolutely integrable, we may write

G(f ) = G(s)|𝜎=0 (4.143)

The LT is therefore a generalisation of the FT, both being applicable to CT signals and systems. If the ROC
of the LT of a signal includes the imaginary axis in the s-plane and the LT is evaluated exclusively along this
imaginary axis, the result of this computation is the FT of the signal, which provides complete information about
the magnitudes and phases of the frequency components of the signal, as previously studied in detail.
The existence of the LT for a wide range of signals, and its various features, such as transforming convolution
into multiplication and linear differential equations into algebraic equations, make the LT a very effective tool
for the analysis and design of electrical, control, and other CT systems. The LTs of standard functions have been
widely tabulated and simple methods such as partial fraction expansion may be used for the inverse operation
of determining the corresponding t-domain signal for a given s-domain function without the need to evaluate
the inverse LT integral. We should mention that Eq. (4.140) defines the so-called bilateral Laplace transform. If the
Telegram: @Persian_IEEEComSoc

signal is causal, or if we arbitrarily choose the time of interest to start at t = 0, then the unilateral Laplace transform
is enough, and is defined for a function g(t) as
∞
G(s) = g(t)e−st dt, s = 𝜎 + j𝜔 (4.144)
∫0

4.5.2 z-transform
The z-transform (ZT) G(z) of a data sequence g[n] is defined as
∑
∞
G(z) = g(n)z−n (4.145)
n=−∞
 4.5 Laplace and z-transforms 293

where z is a complex variable that takes on values in the so-called z-plane as shown in Figure 4.45b. Recalling the
DTFT expression in Eq. (4.114)
∑
∞
G(Ω) = g(n)e−jnΩ
n=−∞

where
f
Ω = 2𝜋 (4.146)
Fs
we see that the ZT is indeed a generalisation of the DTFT. More specifically, when the ZT is evaluated at z = e−jΩ
it reduces to the DTFT. That is
G(Ω) = G(z)|z=e−jΩ (4.147)
Since ejΩ = 1∠Ω, which is the unit circle (centred at the origin) in the z-plane, it follows that evaluating the
ZT of g[n] exclusively along this unit circle yields the DTFT of g[n]. Points along this circle correspond to all the
frequency components of g[n], according to Eq. (4.146), which is repeated below with f as subject
Ω
f = F (4.148)
2𝜋 s
So, the DC component (i.e. f = 0) is at Ω = 0 and the highest-frequency component of g[n], which is of value
f = F s /2 in view of the sampling theorem, is at Ω = 𝜋 radian. To further emphasise: a nonperiodic data sequence
g[n] obtained by sampling at rate F s is composed of a continuum of complex exponentials – complex conjugate
pairs of which form a sinusoid of the same frequency. If we know the ZT G(z) of this data sequence then we can
obtain the magnitude (per unit frequency) and phase of these complex exponentials at any desired frequency as
the magnitude and angle of the complex number
G(z)|z=1∠2𝜋f ∕Fs
For example, if F s = 500 Hz, the maximum frequency component of the data sequence is 250 Hz, and the mag-
nitude and phase of the 125 Hz component of the data sequence is provided by the value of the ZT at angle 90∘
counterclockwise along the unit circle. Note that the maximum value of Ω is 𝜋 radian, and that the bottom half of
the unit circle in Figure 4.45b is for values of Ω from 0 rad to −𝜋 rad clockwise (i.e. not Ω = 𝜋 to 2𝜋), and correspond
to the negative frequencies from 0 to −F s /2 which form the complex conjugates for the components represented
by corresponding points along the upper half of the unit circle.
Returning to Eq. (4.145) and expressing z in its polar form z = r∠Ω = rejΩ , where r is the magnitude of z (not
necessarily 1) and Ω is its angle, we obtain
∑
∞
G(z) = [g(n)r −n ]e−jnΩ , z = rejΩ (4.149)
Telegram: @Persian_IEEEComSoc

n=−∞

This indicates that the ZT of g[n] is the DTFT of g[n]r −[n] . Thus, the ZT may exist for certain data sequences that
do not have a FT, and this existence will be in a region of the z-plane (outside the unit circle r = 1) where r is large
enough to make the sequence g[n]r −[n] decay sufficiently rapidly to zero as n increases.
It is often the case that the data sequence g[n] is zero for n < 0, so that the one-sided or unilateral z-transform is
applicable, defined as
∑
∞
G(z) = g(n)z−n (4.150)
n=0

The relationship between the ZT and the FT should now be very clear from the above discussions. The ZT is
to DT signals and systems what the LT is to CT signals and systems. It facilitates easy solutions for difference
 294 4 Frequency Domain Analysis of Signals and Systems

equations (the discrete equivalent of differential equations) that describe, for example, digital filters and is a very
effective tool for the analysis and design of DSP systems. A relationship also exists between the ZT and the LT
which may be shown by mapping the s-plane into the z-plane through the substitution
z = esT s = e(𝜎+j𝜔)Ts = e𝜎Ts ej𝜔Ts
= e𝜎Ts ∠𝜔Ts = e𝜎Ts ∠2𝜋f ∕Fs (4.151)
The following may be deduced from Eq. (4.151):
When 𝜎 = 0 (in the s-plane), z = ej𝜔Ts = 1∠2𝜋f ∕Fs in the z-plane. Thus, the y axis in the s-plane corresponds to
the unit circle in the z-plane. This is as expected since the FT is computed along the y axis (i.e. the j𝜔 axis) in the
s-plane, but along the unit circle in the z-plane.
The angle of z is 2𝜋f /F s , so an increase of frequency f from f = 0 to f = F s /2 completes a half-cycle counterclock-
wise around the top half of the unit circle, whereas a decrease in f from 0 to −F s /2 completes a half-cycle clockwise
around the bottom half of the unit circle. This relationship between frequency and angular displacement along
the unit circle was earlier noted when discussing the link between the DTFT and the ZT. Repeated cycles around
this circle will merely repeat the same values, consistent with the F s -periodic nature of the spectrum of sampled
signals and systems.
When 𝜎 < 0 (which corresponds to the left half of the s-plane), the magnitude of z is |z| = e𝜎Ts < 1, since the
sampling interval T s is a positive quantity. But |z| < 1 corresponds to points inside the unit circle. Thus, the entire
left half of the s-plane is mapped into the unit circle in the z-plane.
A stable causal system must have all its poles in the left half of the s-plane, so if a causal system is to be stable
then the ZT of its impulse response must have all poles inside the unit circle and none on or outside the circle.
Equation (4.150) indicates that the ZT is a power series in z−1 . When the ZT of a signal is expressed as a
power series then the identity of the signal is obvious as simply the coefficients of the series. For example, if
G(z) = 8 + 2z−1 + z−2 , then g[n] = {8, 2, 1, 0, 0, …} at respective instants n = {0, 1, 2, …}. It therefore follows that
multiplication by z−k has the effect of introducing a delay of k sampling intervals.
The ROC of the ZT is the range of values of z where the power series of Eq. (4.150), i.e. g(0)+ g(1)z−1 + g(2)z−2 +· · ·,
converges so that G(z) is finite. Values of z at which G(z) is infinite are known as the poles of G(z), whereas values
of z at which G(z) = 0 are known as the zeros of G(z). For a causal finite-duration signal such as an energy signal,
the summation in Eq. (4.150) yields a finite value except at z = 0, so the ROC is everywhere in the z-plane except
z = 0. But for causal infinite-duration sequences, there may be poles at locations other than zero. In that case,
the ROC is defined as everywhere outside a circle centred at the origin and having a radius equal to the pole of
largest magnitude. That is, draw a line from the origin to the pole that is furthest from the origin, and the ROC is
everywhere outside the circle centred at the origin and having that line as its radius.
The inverse z-transform (IZT), denoted Z −1 , is an operation that yields the DT sequence g[n] given its ZT G(z).
That is
g[n] = Z −1 [G(z)] (4.152)
Telegram: @Persian_IEEEComSoc

The most common form for expressing the ZT is as a ratio of two polynomials in z−1 . In this form, the inverse ZT
may be obtained using several methods such as power series expansion or partial fraction expansion. A discussion
of these methods, though straightforward, is beyond the scope of this book. The interested reader is referred to
any one of the many good textbooks on signals and systems and DSP, such as [4 and 5].

Worked Example 4.13 z-transform

We wish to derive the ZT (in closed-form) of the following sequences:

(a) The unit impulse 𝛿[n].

(b) The unit step sequence u[n].
 4.6 Inverse Relationship Between Time and Frequency Domains 295

(a) Recalling that 𝛿[n] = 0 everywhere except at n = 0 where it equals 1, the ZT of 𝛿[n] follows straightforwardly
from the definition of ZT as
∑∞
Z[𝛿(n)] = 𝛿(n)z−n
n=0
= 𝛿(0)z0 + 𝛿(1)z−1 + 𝛿(2)z−2 + · · ·
= 1 × 1 + 0 × z−1 + 0 × z−2 + · · ·
=1
This result is independent of z, so the transform’s ROC is everywhere in the z-plane.
(b) Applying the ZT definition gives
∑
∞
∑
∞
Z[u(n)] = u(n)z−n = 1•z−n
n=0 n=0
= 1 + z−1 + z −2 −3
+z +···
≡ S∞
The sum denoted S∞ is a geometric series having first term 1 and constant ratio z−1 . Writing down the equation
for the sum of the first N + 1 terms of this series and also a second equation obtained by multiplying both sides
of the first equation by z−1
∑
N
SN = z−n = 1 + z−1 + z−2 + z−3 + · · · + z−N
n=0
z−1 SN = z−1 + z−2 + z−3 + · · · + z−N + z−(N+1)
Now subtracting the two equations
SN (1 − z−1 ) = 1 − z−(N+1)
Thus
1 − z−(N+1)
SN =
1 − z−1
1 − 1∕z(N+1)
S∞ = lim SN = lim
N→∞ N→∞ 1 − z−1
1
= , |z| > 1
1 − z−1
Multiplying top and bottom of the right-hand side by z yields the required transform
Telegram: @Persian_IEEEComSoc

z
Z[u(n)] = , |z| > 1
z−1

4.6 Inverse Relationship Between Time and Frequency Domains

We have learnt that signals may be fully specified either in the time domain or in the frequency domain. The former
gives the instantaneous values of the signal as a function of time, e.g. g(t), whereas the latter specifies the spectrum
of the signal. In general the spectrum is given by the FT G(f ) of the signal, but in the special case of a periodic
signal having period T, the spectrum is specified as the amplitude An and phase 𝜙n of each sinewave (of frequency
nf o ) that adds to make up the signal, where f o = 1/T. Once we specify g(t) then G(f ) can be determined by Fourier
 296 4 Frequency Domain Analysis of Signals and Systems

analysis. If G(f ) is specified then g(t) follows from Fourier synthesis. Thus, the time and frequency domains provide
alternative methods of describing the same signal. There are important general relationships between the features
of a signal waveform g(t) and the features of its spectrum G(f ).
The shorter the time duration of a signal, the broader its spectrum. Observe in Figure 4.29 that the null band-
width Bn (considered as the range of significant positive frequencies) of the rectangular pulse of duration 𝜏 is
1
Bn = (4.153)
𝜏
Thus, as demonstrated in Figure 4.46, if we expand the duration of a signal then its spectrum contracts by the
same factor and vice versa, in such a way that the product of signal duration and signal bandwidth, called the
time-bandwidth product, remains constant. The value of this constant depends on pulse shape. In the special case
of a rectangular pulse, Eq. (4.153) and Figure 4.46 show that the constant is 1, which is the lowest possible for
all pulse shape types, but this comes at the expense of stronger spectral sidelobes (beyond the main lobe). It can
be seen from the bottom three plots in Figure 4.46 that the triangular pulse has a time-bandwidth product equal
to 2. Note that we have used a null bandwidth definition for this discussion. The choice of a different bandwidth
definition will change the value of the constant but not the inverse relationship.
A signal cannot be both strictly band-limited and strictly duration-limited. If the spectrum of a signal is precisely
zero outside a finite frequency band then the time domain waveform g(t) will have infinite duration, although g(t)
may tend to zero as t → ∞. Conversely, if g(t) is precisely zero outside a finite time duration then the spectrum G(f )

g1(t) |G1(f)|
τ = 1 ms Bn = 1 kHz
⇌
t f
g2(t)
|G2(f)|

τ = 0.5 ms ⇌ Bn = 2 kHz
t f
g3(t)
|G3(f)|
τ = 0.25 ms ⇌ Bn = 4 kHz
t f
g4(t) |G4(f)|

τ = 1 ms
⇌ Bn = 2 kHz
Telegram: @Persian_IEEEComSoc

t f
g5(t) |G5(f)|

τ = 0.5 ms
⇌ Bn = 4 kHz
t f
g6(t) |G6(f)|
τ = 0.25 ms
⇌ Bn = 8 kHz
t f

Figure 4.46 Null bandwidths of rectangular and triangular pulses of various durations.
 4.7 Frequency Domain Characterisation of LTI Systems 297

g1(t) |G1(f)|

t f
τ
g2(t) |G2(f)|

t f
τ

Figure 4.47 Comparing the effect of sudden transitions in signals. There are more significant higher frequency components
in g1 (t) than in g2 (t).

will carry on and on, although |G(f )| will tend to zero as f → ∞. For example, a (duration-limited) rectangular pulse
has a sinc spectrum, which carries on and on, and a (strictly band-limited) rectangular spectrum belongs to a sinc
pulse.
Sudden changes in a signal produce high-frequency components in its spectrum. Figure 4.47 shows the spectra
of two pulses of the same duration, but of different shapes. One pulse is rectangular and has sharp transitions
between zero and maximum value, whereas the other pulse is cosine shaped with no sharp transitions. It can be
observed that the amplitudes of the higher-frequency components decay less rapidly in the rectangular pulse.
A signal that is periodic in the time domain will be discrete in the frequency domain with the spectrum con-
sisting of spectral lines spaced apart by the reciprocal of the signal’s time domain period. Each spectral line in
a single-sided spectrum represents a harmonic sinusoidal component of the signal. A signal that is nonperiodic
in the time domain will be continuous in the frequency domain with the spectrum containing a continuum of
spectral lines. If a signal is continuous in the time domain then it will be nonperiodic in the frequency domain.
Apart from the special cases of an impulse function and white noise, the amplitude spectrum of a CT signal will
eventually tend to zero as frequency increases and will certainly not have a periodic pattern. Finally, if a signal is
discrete in the time domain then it will be periodic in the frequency domain. A DT signal is usually the result of
taking regular data samples at interval T s or sampling rate F s = 1/T s . The spectrum of such a signal will have a
repetitive (i.e. periodic) pattern at a regular frequency spacing F s .

4.7 Frequency Domain Characterisation of LTI Systems

Telegram: @Persian_IEEEComSoc

4.7.1 Transfer Function

The frequency response or transfer function H(f ) of an LTI system is defined as the ratio between system output
and input when the input is a sinusoidal signal of frequency f . It indicates how the system alters the amplitude
and phase of each sinusoid passed through it and is in general therefore a complex quantity which we may express
in the form

H(f ) ≡ |H(f )|∠𝜙H (f ) = |H(f )|ej𝜙H (f ) (4.154)

where |H(f )| and 𝜙H (f ) are, respectively, the magnitude and phase of H(f ). When a sinusoid of frequency f k is
passed through a system with transfer function H(f ) the sinusoid will have its amplitude multiplied by a factor of
 298 4 Frequency Domain Analysis of Signals and Systems

|H(f k )| and its phase increased by 𝜙H (f k ) radians, so we may write

R
cos(2𝜋fk t + 𝜙k ) −−−−→ |H(fk )| cos[2𝜋fk t + 𝜙k + 𝜙H (fk )] (4.155)
|H(f )| therefore specifies gain as a function of frequency and is called the amplitude response or gain response of
the system.
The phase of H(f ) is given by
[ ]
−1 Im{H(f )}
𝜙H (f ) = tan (4.156)
Re{H(f )}
where Im{x} and Re{x} give the imaginary and real parts of x respectively. 𝜙H (f ) specifies phase shift as a function
of frequency and is called the phase response of the system.
In Section 3.6 we learnt how to determine the output of an LTI system using a time domain characterisation
of the system in terms of its impulse response h(t). Given the transfer function H(f ) of an LTI system, which is a
frequency domain characterisation of the system, we can also obtain a general expression for the response of the
system to an arbitrary input signal by taking the following steps, illustrated in Figure 4.48.
A single sinusoidal input of frequency f k yields a sinusoidal response at the same frequency with amplitude gain
|H(f k )| and phase shift 𝜙H (f k ), whatever the value of f k .
In view of the principle of superposition stated in Eq. (3.126), when the input is a group of sinusoids of frequencies
f k , k = 1, 2, 3, …, N, the output will be the result of the system changing each sinusoidal signal in amplitude by a
factor of |H(f k )| and in phase by a shift 𝜙H (f k ), and then adding these modified sinusoids together. Recall that this
principle of superposition implies that what the system does to any one of the sinusoidal signals is not affected in
any way by the presence of other signals.
But an arbitrary signal x(t) is in fact a sum of sinusoids expressed as
∞
x(t) = X(f )ej2𝜋ft df (4.157)
∫−∞
where X(f ) is the FT of x(t). The integrand [X(f )df ]exp(j2𝜋ft) at the pair of frequency locations ±(f → f + df ) of
infinitesimally small width df represents a sinusoid of frequency f and amplitude 2|X(f )|df . The right-hand side

cos(2πfkt + ϕk) ∣H(fk)∣ cos[2πfkt + ϕk + ϕH(fk)]

LTI System
(a)
H(f)
f
fk

N N
Σ Ak cos(2π fkt + ϕk) Σ Ak∣H(fk)∣cos[2π fkt + ϕk + ϕH(fk)]
k=1 LTI System k=1
(b)
f H(f)
f1 f2 f3 fN
Telegram: @Persian_IEEEComSoc

|X(f)| ∞ ∞
x(t) = ∫ X( f)ej2πftdf y(t) = ∫ X( f)H( f)ej2πftdf
–∞ LTI System –∞
(c)
H(f)
f

LTI System
(d) X(f) Y(f) = X(f)H(f)
H(f)

Figure 4.48 Frequency domain approach in LTI system analysis.

 4.7 Frequency Domain Characterisation of LTI Systems 299

and hence x(t) is therefore a continuous sum of sinusoids of all frequencies in the range f = (0, ∞). So, when we
present x(t) to our system the output y(t) will be the result of modifying each input sinusoid of frequency f by the
complex factor H(f ) which includes both a gain factor and a phase shift, and then adding them all together in the
same continuous (i.e. integral) fashion. That is
∞
y(t) = X(f )H(f )ej2𝜋ft df (4.158)
∫−∞
We see that Eqs. (4.157) and (4.158) have the same format. So since X(f ) in Eq. (4.157) is the FT of x(t), it follows
that X(f )H(f ) in Eq. (4.158) must be the FT of y(t), denoted Y (f ). Thus

Y (f ) = H(f )X(f ) (4.159)

This important result states that the output spectrum Y (f ) for a transmission through an LTI system is the
product of the input spectrum and the system’s transfer function or frequency response H(f ).
Therefore, we now have two approaches to the analysis of an LTI system. In the time domain approach of
Figure 3.24 we convolve the input signal x(t) with the system’s impulse response h(t) to give the output signal
y(t). We can then obtain the output spectrum Y (f ) if required by taking the FT of y(t). In the frequency domain
approach of Figure 4.48, we multiply the input signal’s spectrum X(f ) by the system’s transfer function H(f ) to
obtain the output signal’s spectrum Y (f ). If desired, we may then take the IFT of Y (f ) to obtain the output signal’s
waveform y(t). But how are h(t) and H(f ) related? To answer this question we invoke the convolution property of
the FT stated in Eq. (4.91). Taking the FT of both sides of the time domain relation

y(t) = x(t) ∗ h(t)

we obtain
F[y(t)] = F[x(t) ∗ h(t)] = F[x(t)]F[h(t)]
Y (f ) = X(f )F[h(t)]

Equating the right-hand side of this equation with the right-hand side Eq. (4.159) yields

F[h(t)] = H(f )

The impulse response h(t) and frequency response H(f ) therefore form a FT pair

h(t) ⇌ H(f ) (4.160)

Equation (4.160) completes the picture of interrelationship between the two analysis approaches. H(f ) can be
determined theoretically by analysing the system’s circuit diagram or the channel’s signal propagation mecha-
nisms. It may also be obtained experimentally by measuring the system’s gain and phase shift for a sinusoidal
input of frequency stepped through the range of interest. The impulse response h(t) of the system is then obtained
Telegram: @Persian_IEEEComSoc

by taking the IFT of H(f ). The next worked example presents the circuit analysis method.

Worked Example 4.14 Analysis of a Simple Lowpass Filter

In Chapter 3, Worked Example 3.9, we analyse an LPF using a time domain approach. In this example, we will
follow an alternative frequency domain approach, which is the preferred method for most engineers.
Determine the transfer function H(f ) and impulse response h(t) of the simple RC LPF shown in Figure 4.49.
For R = 1 kΩ and C = 79.58 nF, determine

(a) The output spectrum when the input signal is an impulse function 𝛿(t).
(b) The output voltage v2 (t) for an input voltage v1 (t) = 10 cos(2000𝜋t + 30∘ ) V.
(c) The 3 dB bandwidth of the filter.
 300 4 Frequency Domain Analysis of Signals and Systems

R Z1 Z1 = R;
Input C Output V1(f) Z2 V2(f) Z2 = –j/2π fC

System
(a) (b)

Figure 4.49 RC low-pass filter: (a) Circuit diagram; (b) Equivalent circuit for a sinusoidal input signal of amplitude V 1 and
frequency f .

The equivalent RC circuit for a sinusoidal input signal of frequency f is shown in Figure 4.49b. The input voltage
V 1 (f ) is divided between the resistance R and the capacitance C according to the ratio of their impedances Z 1 and
Z 2 , respectively. The voltage drop across C is the required output voltage. Thus
V2 (f ) Z2 −j∕2𝜋fC
H(f ) ≡ = =
V1 (f ) Z1 + Z2 R − j∕2𝜋fC
Multiplying the top and bottom of the last term by j2𝜋fC yields
1 1
H(f ) = = √
1 + j2𝜋fRC 1 + (2𝜋fRC)2 ∠tan−1 (2𝜋fRC)
1
= √ ∠ − tan−1 (2𝜋fRC)
1 + 4𝜋 2 f 2 R2 C2
≡ |H(f )|∠𝜙H (f )
where
1
|H(f )| = √ ≡ Gain response
1 + 4𝜋 2 f 2 R2 C2
𝜙H (f ) = −tan−1 (2𝜋fRC) ≡ Phase response (4.161)
The impulse response h(t) is the IFT of H(f )
[ ]
−1 −1 1
h(t) = F [H(f )] = F
1 + j2𝜋fRC
[ ( )]
1 1
= F−1
RC 1∕RC + j2𝜋f
[ ]
1 −1 1
= F
RC 1∕RC + j2𝜋f
Looking in the list of FT pairs in Table 4.5, we find in entry 17 that
1
e−at u(t) ⇌
Telegram: @Persian_IEEEComSoc

a + j2𝜋f
Therefore
[ ]
−1 1
F = e−t∕RC u(t)
1∕RC + j2𝜋f
And hence
1 −t∕RC
h(t) = e u(t)
RC
(a) Applying Eq. (4.159) and noting from Table 4.5 that the FT of a unit impulse is 1, we obtain the output spectrum
when the input is 𝛿(t) as
Y (f ) = H(f )F[𝛿(t)] = H(f )
 4.7 Frequency Domain Characterisation of LTI Systems 301

This confirms Eq. (4.160) and is also a general result: whenever the input is a unit impulse, the spectrum of
the output signal gives the transfer function of the system. Substituting the values of R and C in the amplitude
and phase expressions for H(f ) yields the required amplitude and phase spectra
1 1
|H(f )| = √ = √
1 + 4𝜋 2 f 2 (79.58 × 10−9 )2 (1000)2 1 + (f ∕2000)2
𝜙H (f ) = −tan−1 [2𝜋fCR] = −tan−1 (5 × 10−4 f )

(b) v1 (t) is a sinusoid of frequency f = 1000 Hz. The output v2 (t) will be a sinusoid of the same frequency but with
amplitude reduced by the factor |H(f )| and phase increased by 𝜙H (f ). Thus

v2 (t) = 10|H(f )| cos[2000𝜋t + 30∘ + 𝜙H (f )]|f =1000

10
= √ cos[2000𝜋t + 30∘ + tan−1 (−5 × 10−4 × 103 )]
1 + (1000∕2000) 2

= 8.94 cos[2000𝜋t + 30∘ − 26.6∘ ]

= 8.94cos(2000𝜋t + 3.4∘ )

√
(c) The 3 dB bandwidth of the filter is the frequency f 2 at which the gain response |H(f )|max is 1∕ 2 of its peak
value. We see from Eq. (4.161) that |H(f )|max = 1 at f = 0, so we may write
1 |H(f )|max 1
|H(f2 )| = √ ≡ √ = √
( )2 2 2
f2
1+ 2×103

Squaring both sides reduces the equation to

( )2
f2
1+ =2
2 × 103
and hence

f2 = 2 kHz.

(d) The impulse response h(t), gain response |H(f )| and phase response 𝜙H (f ) of this LPF are plotted in Figure 4.50.
You may wish to compare what we have done here with Worked Example 3.9 to decide which method you
prefer and why.
Telegram: @Persian_IEEEComSoc

4.7.2 Output Spectral Density of LTI Systems

It is useful to determine how the ESD of an energy signal and the PSD of a power signal are affected by transmission
through an LTI system. We know from Eq. (4.159) that the output spectrum Y (f ) of an LTI system of transfer
function H(f ) is related to the input spectrum X(f ) by

Y (f ) = X(f )H(f )

Squaring this equation to obtain

|Y (f )|2 = |X(f )|2 |H(f )|2

 302 4 Frequency Domain Analysis of Signals and Systems

20 log | H(f) |, dB
0
–3
–6
1 ↑ h(t)
–9
RC
–12

0.6 –15 → f, kHz

RC –12 –6 0 2 4 6 8 10 12
⇌ ϕH (f), deg
90
0.2 60
RC
t 30
0 RC 2RC 3RC 4RC 5RC 0 → f, kHz
–30
–60
–90
–12 –8 –4 0 2 4 6 8 10 12

Figure 4.50 Impulse response and transfer function of RC low-pass filter.

and noting, from Eq. (4.129), that the square of the magnitude spectrum of a signal is the ESD of the signal, we
obtain the relationship
Ψy (f ) = Ψx (f )|H(f )|2 (4.162)
where Ψy (f ) and Ψx (f ) are, respectively, the ESD of the output and input signals y(t) and x(t). Since PSD is simply
the average rate of ESD, a similar relationship between output PSD Sy (f ) and input PSD Sx (f ) will hold for power
signals. Thus
Sy (f ) = Sx (f )|H(f )|2 (4.163)
Therefore, the effect of the system is to modify – in other words colour – the signal’s spectral density by the factor
|H(f )|2 , which is the square of the system’s gain response.
For an important example, consider white noise as the input signal. We know that this has a constant
PSD = N o /2. In view of Eq. (4.163), the noise at the system output will be coloured noise of PSD
No
|H(f )|2
Scn (f ) = (4.164)
2
Summing the above contribution over the entire frequency axis yields the total noise power at the output of the
system
Telegram: @Persian_IEEEComSoc

No ∞ ∞
Pn = |H(f )|2 df = No |H(f )|2 df (4.165)
2 ∫−∞ ∫0

4.7.3 Signal and System Bandwidths

We have up to now used the term bandwidth at various points in the book but without a formal definition. The word
bandwidth has in recent times become commonplace in society and is very often misused, even in technical circles.
You hear of computer bandwidth and smartphone bandwidth (sic), etc. Download and upload speeds (in bits per
second) of your broadband connection are perfectly correct societal usages. However, in telecommunications,
bandwidth is not measured in bits per second and there is no direct one-to-one relation between bandwidth and
bit rate. How much bit rate is achieved within a given transmission bandwidth very much depends on several link
 4.7 Frequency Domain Characterisation of LTI Systems 303

design factors, such as modulation and coding schemes, ratio between signal power and noise plus interference,
level of acceptable bit error rate, etc. We will explore the interplay of these parameters throughout the book, but
for now it is time for a precise definition of bandwidth.
The Fourier series expression for periodic signals – see Eq. (4.2) – would suggest that such signals contain an
infinite number of harmonic frequencies. Similarly, the Fourier integral expression of Eq. (4.75) suggests includ-
ing contributions from frequency components up to infinity in the continuous summation for the signal g(t). In
realisable signals, however, the amplitude An of the nth harmonic sinusoid becomes very small for large n, and G(f )
becomes insignificant for large f , so these high-frequency components may be neglected with negligible effect on
the signal g(t). The only exception will be if g(t) is an impulse function, in which case G(f ) = 1.
The bandwidth of a signal is the range of positive frequencies of the sinusoidal signal components making a
significant contribution to the signal’s power or energy. This is the width in hertz (Hz) of the significant SSAS or
periodogram of the signal. The bandwidth of a communication channel (i.e. transmission medium) or electronic
device or (in general) system is the range of positive frequencies of the sinusoidal signals that may be passed
through the system from its input to its output without significant distortion. A lowpass or baseband signal contains
significant positive frequencies down to or near DC (f = 0) and its bandwidth is usually taken as the maximum
significant frequency component even if there is no DC component. A bandpass signal, on the other hand, has
its significant positive frequencies centred at a high-frequency fc ≫ 0, and its bandwidth is always given by the
range from the lowest to the highest significant positive frequency components. It is important to always bear this
distinction in mind. For example, a sinusoidal message signal of frequency f m is treated as a baseband signal and
its bandwidth is therefore B = f m , but a sinusoidal carrier signal of high frequency f c is regarded as a bandpass
signal, hence bandwidth B = 0. A similar explanation is applicable to lowpass and bandpass systems, where the
former passes frequencies near DC and the latter passes frequencies centred around fc ≫ 0.
A precise statement of what is significant in the above discussion leads to various bandwidth definitions, includ-
ing (i) subjective bandwidth, (ii) null bandwidth, (iii) X dB bandwidth, (iv) fractional power containment band-
width, and (v) noise equivalent bandwidth.

4.7.3.1 Subjective Bandwidth

The significant frequency range is chosen to be just enough for a subjectively acceptable signal quality. This
bandwidth definition is employed exclusively for audio signals and the luminance and chrominance signals of tele-
vision. For example, speech signals typically contain frequencies in the range 50 Hz to 7 kHz. The low-frequency
components (50–200 Hz) enhance speaker recognition and naturalness, whereas the high-frequency components
(3.5–7 kHz) enhance intelligibility, such as being able to distinguish between the sounds of ‘s’ and ‘f’. Good sub-
jective speech quality is, however, obtained in telephone systems with the baseband spectrum limited to the range
300–3400 Hz and this subjective bandwidth has been adopted as the ITU-T standard for telephony. And in tele-
vision, because the eye is not very sensitive to changes in colour, the fine details of colour changes are usually
Telegram: @Persian_IEEEComSoc

omitted to allow the use of a smaller bandwidth for the chrominance signals (represented as two colour-difference
components, each of typical subjective bandwidth 1.3 MHz).
Examples of subjective bandwidths that have been standardised by international agreement for the purpose of
maximising system capacity and radio spectrum exploitation include, analogue telephone speech = 3400 Hz, AM
broadcast signal = 10 kHz, FM broadcast signal = 200 kHz, and analogue television (TV) broadcast signal = 6 or
8 MHz (depending on standard and including luminance and chrominance signals). Note that analogue telephone
and TV signals are nowadays digitised prior to transmission using digital technology.

4.7.3.2 Null Bandwidth

The significant frequency range extends up to the first null in the amplitude spectrum of the signal. Clearly,
this bandwidth definition applies only to signals (mostly pulses) whose spectra have well-defined nulls. If the
 304 4 Frequency Domain Analysis of Signals and Systems

|G(f)|

(a)

f
f1 0 f2
Null bandwidth

|G(f)|

(b)

f
–fc 0 f1 fc f2
Null bandwidth

Figure 4.51 Null-bandwidth: (a) Lowpass signal; (b) Bandpass signal.

double-sided spectrum of the signal has a main lobe bounded by one null (zero amplitude spectral point) at a
positive frequency f 2 and another null at a lower frequency f 1 , then the bandwidth of the signal is given by
{
f2 , f1 < 0 (lowpass signal)
Null bandwidth = (4.166)
f2 − f1 , f1 > 0 (bandpass signal)

Figure 4.51 illustrates each of the two cases in Eq. (4.166).

4.7.3.3 3 dB Bandwidth
The spectra of many signals and the gain response of most systems do not have well-defined nulls. In such cases
the significant frequency range extends up to the point at which the amplitude spectrum of the signal or the gain
response of the system is down by X dB from its peak value. This bandwidth definition is mainly applied to filters
or systems. The most common specification is for X = 3, known as the 3 dB bandwidth or half-power bandwidth
Telegram: @Persian_IEEEComSoc

since a 3 dB drop in amplitude corresponds to a reduction in signal power by half. Values of X = 6 and X = 60 are
sometimes used to give the shape factor of a filter. Figure 4.52 shows the gain response of a fifth order Butterworth
LPF with 3 dB bandwidth = 10 kHz and 60 dB bandwidth = 40 kHz. In general, if the gain response |H(f )| of a
filter or system has peak value |H(f )|max , then the cut-off frequency f X for determining the X dB bandwidth of the
system is obtained by solving the equation

|H(fX )| = |H(f )|max × 10−X∕20 (4.167)

Note that this was the approach followed in Worked Example 4.14c to determine the 3 dB bandwidth of the RC
LPF. If, however, a plot (such as Figure 4.52) of the normalised gain response is available then it is a straightforward
matter to read the bandwidth (for any value of X) from such a graph. Figure 4.53 shows various examples of 3 dB
bandwidth.
 4.7 Frequency Domain Characterisation of LTI Systems 305

–3

–20
Normalised Gain, dB

–40

–60

–80

–100
–100 –80 –60 –40 –20 0 20 40 60 80 100
Frequency, kHz

Figure 4.52 Gain response of a fifth order Butterworth lowpass filter.

|H(f)|
1
|H(f)|
1/ 2
(a) (b)
Bandwidth
bandwidth
f f
f1 0 f2 f1 0 f2

|H(f)|

(c)
Bandwidth
f
–fc 0 f1 fc f2
|H(f)|
1
1/ 2 bandwidth
(d)
Telegram: @Persian_IEEEComSoc

f
–fc fc

Figure 4.53 3 dB bandwidths: (a) strictly bandlimited lowpass system; (b) non-bandlimited lowpass system; (c) strictly
bandlimited bandpass system; and (d) non-bandlimited bandpass system.

A filter is designed to have a non-flat frequency response, in order to pass a specified band of frequencies with
little or no attenuation while heavily attenuating (or ideally blocking) all frequencies outside the passband. In
most cases, it is desired that the filter’s gain is constant within the passband. There is, however, a class of filters,
known as equalisers, that are designed not only to exclude signals in the stopband but also to compensate for
the distortion effects introduced into the passband by, for example, the transmission medium. An equaliser has
 306 4 Frequency Domain Analysis of Signals and Systems

a non-flat frequency response, which when combined with the response of the transmission medium gives the
desired overall flat response. Examples of filters include the lowpass filter (LPF), which passes only frequencies
below a cut-off frequency F 1 . A highpass filter (HPF) passes only frequencies above a cut-off frequency F 2 . A
bandpass filter (BPF) passes only those frequencies in the range from F 1 to F 2 . A bandstop filter (BSF) passes all
frequencies except those in the range from F 1 to F 2 . The normalised frequency response of an ideal LPF is shown
in Figure 4.54a. The gain of the filter drops from 1 to zero at the cut-off frequency F 1 . Such a brick wall filter is
not feasible in real time. A realisable filter, shown in Figure 4.54b, requires a finite
√ frequency interval in which
to make the transition from passband, where the minimum normalised gain is 1∕ 2 = 0.707, to stopband, where
the maximum gain is 𝛼 min . The interval from F 1 to the start F s of the stopband is known as the transition band.

4.7.3.4 Fractional Power Containment Bandwidth

The significant frequency range is the minimum range containing no less than a specified fraction of the total
signal power. For example, the 95% fractional power containment bandwidth of a baseband signal is the range of
frequencies, starting from DC, that contains at least 95% of the total signal power. In the case of a bandpass signal,
this range is centred on the centre frequency. Fractional power containment bandwidth definition is used mainly
for specifying transmission bandwidths. For example, in frequency modulation (FM), Carson’s bandwidth is the
width of the band of frequencies (centred on the carrier frequency) which contains at least 98% of the total power
in the FM signal. The Federal Communications Commission (FCC) specifies occupied bandwidth as the frequency
range containing 99% of signal power, with exactly 0.5% of the power above and 0.5% below the occupied band.
The p percent (p%) power containment bandwidth Bp of a periodic signal is obtained by first calculating the
total power Pt of the signal in the time domain (from its waveform structure) and then totalling the cumulative
power PN in the frequency components starting from DC – see Eq. (4.35) – until a harmonic np is first reached that

|H(f)|

(a)
Passband
Stopband
f
–F1 0 F1
Telegram: @Persian_IEEEComSoc

1 |H(f)|

1/√2
(b) Transition band
Passband Stopband
αmin
f
–F1 F1 Fs

Figure 4.54 Frequency response of (a) Ideal (also called brick wall) lowpass filter (b) Realisable lowpass filter.
 4.7 Frequency Domain Characterisation of LTI Systems 307

satisfies the first line of Eq. (4.168) below. Bp is then given by the second line.
( np )
1 1∑ 2 p
2
A0 + A ≥
Pt 2 n=1 n 100
Bp = nf o , fo ≡ fundamental frequency (4.168)

The above computation may be carried out by trial and error (using a computer code that returns the cumu-
lative power up to any given harmonic), or it may be conveniently laid out in tabular form as done in Worked
Example 4.15.
For nonperiodic signals, Bp may be obtained by integrating the signal’s single-sided periodogram from f = 0 up
to the lowest positive frequency at which the cumulative energy or power reaches p% of the total. For example,
an energy signal g(t) of energy E (computed from its waveform structure) is guaranteed to have a FT G(f ). So, Bp
is obtained by evaluating the following equation (using numerical integration if |G(f )|2 cannot be integrated to
obtain a closed form expression)
Bp
2 p
|G(f )|2 df = (4.169)
E ∫0 100

Worked Example 4.15 Further Analysis of Rectangular Pulse Train

We wish to analyse an RPT (such as Figure 3.19a) having amplitude A = 10 V, duty cycle d = 1/5, and period
T = 10 μs to determine the following quantities:

(a) Its 95% fractional power containment bandwidth B95 , determined to the nearest whole percentage.
(b) Percentage of power contained in the main lobe.
(c) Percentage of power contained in the first and second sidelobes.
The total power of the RPT is obtained from Eq. (3.105) as

Pt = A2rms = A2 d = 20 W

The Fourier series of the RPT is derived in Worked Example 4.1 and is as stated in Eq. (4.17), with DC com-
ponent A0 = Ad = 2 V, amplitude of nth harmonic An = 2Ad|sinc(nd)| = 4|sinc(nd)|, and fundamental frequency
f o = 1/T = 100 kHz. A tabular layout of the computations is given in Table 4.6. Column 1 is a list of the harmonic
numbers from n = 0 to 16. Column 2 lists the amplitude of each harmonic, with the corresponding power in col-
umn 3, calculated as A20 for DC power, and A2n ∕2 for the power in each harmonic n = 1 to 16. Colum 4 presents a
running total of cumulative power starting from DC. Finally, column 5 converts column 4 to percentage power by
dividing each column 4 entry by total power Pt and multiplying by 100.

(a) Looking down the fifth column of Table 4.6, we see that the percentage of power first reaches or exceeds 95%
Telegram: @Persian_IEEEComSoc

(to the nearest whole integer) at harmonic n = 8. Thus the 95% fractional power containment bandwidth of
this pulse train is B95 = 8f o = 800 kHz.
(b) It can be seen from Table 4.6 that the first null of the spectrum occurs at harmonic n = 5, so this is the end of
the main lobe. Column 5 indicates that at this point the cumulative power is 90.2878% of total power. Thus,
the percentage of power in the main lobe is 90.29%.
(c) The first sidelobe is the frequency range from the first null at n = 5 to the second null at n = 10. We see from
Table 4.6 that, over this range, power increases from 90.2878% to 94.9946%. Thus, percentage of power in the
first sidelobe is 94.9946 − 90.2878 = 4.71%. Similarly, the second sidelobe is the range from n = 10 to n = 15.
Over this range, power increases from 94.9946 to 96.6413%. The percentage of power in the second sidelobe is
therefore 96.6413 − 94.9946 = 1.65%.
 308 4 Frequency Domain Analysis of Signals and Systems

Table 4.6 Worked Example 4.15. Harmonics of rectangular pulse train

and their power contributions.

Cumulative % of total power,

n An , V Pn , W power, r∑ = 100(P∑ /Pt )

0 2 4 4 20.0000
1 3.7420 7.0011 11.0011 55.0056
2 3.0273 4.5823 15.5834 77.9171
3 2.0182 2.0366 17.6200 88.1000
4 0.9355 0.4376 18.0576 90.2878
5 0 0 18.0576 90.2878
6 0.6237 0.1945 18.2520 91.2602
7 0.8649 0.3741 18.6261 93.1305
8 0.7568 0.2864 18.9125 94.5625
9 0.4158 0.0864 18.9989 94.9946
10 0 0 18.9989 94.9946
11 0.3402 0.0579 19.0568 95.2839
12 0.5046 0.1273 19.1841 95.9204
13 0.4657 0.1085 19.2925 96.4627
14 0.2673 0.0357 19.3283 96.6413
15 0 0 19.3283 96.6413
16 0.2339 0.0273 19.3556 96.7780

The harmonic at which cumulative power first reaches 95% (rounded to the
nearest integer) is highlighted.

4.7.3.5 Noise Equivalent Bandwidth

We are often interested in the noise power at the output of a system, for example to determine signal-to-noise
ratio (SNR), which gives a measure of transmission quality. Usually, the system, through its frequency-dependent
gain response |H(f )|, will pass more noise at some frequencies than at others. But as far as output noise power is
concerned this system is equivalent to a system with a rectangular-shaped gain response that equally admits all
the white noise falling within its finite passband. As illustrated in Figure 4.55 this noise-equivalent system has a
bandwidth B and a constant gain response K (within its passband) equal to the maximum value |H(f )|max of the
gain response of the original system.
Telegram: @Persian_IEEEComSoc

We may formally define the noise equivalent bandwidth of a system as

the bandwidth of a noiseless ideal brick wall filter, of constant gain response equal to the maximum gain
response of the actual system, which passes the same amount of noise power through to its output as does
a noiseless version of the actual system when both have white noise of equal PSD as input.

From Eq. (4.165) and as shown in Figure 4.55, the noise power at the output of the equivalent system is
Pne = K 2 No B, (Noise power at system output) (4.170)
For equivalence this noise power must be equal to the noise power Pna at the output of the actual sys-
tem – obtained earlier in Eq. (4.165) and repeated in Figure 4.55. Equating the expressions for Pne and Pna yields
 4.7 Frequency Domain Characterisation of LTI Systems 309

Actual gain response Noise-equivalent gain response

|H(f)| |He(f)|
|H(f)|max

K = |H(f)|max
≡

f f
–B B

No N N No
Sx(f) = Actual Sya(f) = o ∣H(f)∣2 Sx(f) = o Equivalent Sye(f) = 2 ∣He(f)∣
2
2 2 ≡ 2
System H(f) System He(f)

∞
No ∞ ∞
No B 2
Pna = ∫ Sya(f)df = ∫ ∣H(f)∣2 df Pne = ∫ Sye(f)df = ∫ K df
–∞ 2 –∞ –∞ 2 –B
∞
=
= No ∫ ∣H(f)∣2 df = K2NoB
0

Figure 4.55 Noise equivalent bandwidth B.

an expression for B as
∞
∫0 |H(f )|2 df
B= (4.171)
|H(f )|2max
Equation (4.171) defines the noise-equivalent bandwidth of the system. This is a very useful concept, which
allows us to work with the more convenient ideal filter, as far as noise is concerned. In other words, we replace
the actual system of frequency-dependent gain response |H(f )| with an ideal brick wall filter of bandwidth B,
given by Eq. (4.171). Once we know the noise-equivalent bandwidth of a system, it is a matter of straightforward
multiplication to obtain noise power using
Pn = No B, (noise power referred to system input) (4.172)
for power referred to input, or Eq. (4.170) for output power. We will adopt this approach in all system noise cal-
culations, and it will be taken for granted that B (except where otherwise indicated) refers to noise-equivalent
bandwidth. If, however, this bandwidth is not known then the 3 dB bandwidth of the system may be used in its
place. This substitution underestimates noise power but the error will be small if the filter response has steep sides
with a small transition width.
Telegram: @Persian_IEEEComSoc

Worked Example 4.16 Noise Equivalent Bandwidth of Raised Cosine Filter

The raised cosine filter is universally employed in digital transmission systems to reduce inter-symbol interference.
For transmission at symbol rate Rs (baud), the normalised gain response of the filter is
⎧
⎪1, [ |f | ≤ f1
( )]
⎪1 |f | − f1
|H(f )| = ⎨ 1 + cos 𝜋 , f1 ≤ |f | ≤ f2
⎪2 f2 − f1
⎪0, |f | ≥ f2
⎩
f1 = (1 − 𝛼)Rs ∕2; f2 = (1 + 𝛼)Rs ∕2; 0≤𝛼≤1 (4.173)
where 𝛼 is known as the roll-off factor of the filter.
 310 4 Frequency Domain Analysis of Signals and Systems

|H(f)|
1
α=0
α = 0.2
α = 0.5

α=1

0.5

0 f
–Rs –Rs/2 0 Rs/2 Rs

Figure 4.56 Gain response of raised cosine filter of various values of roll-off factor 𝛼.

We wish to determine the null bandwidth Bnull and noise equivalent bandwidth B of the filter in terms of 𝛼
and Rs .
Putting f = f 2 in Eq. (4.173) yields |H(f2 )| = 0.5(1 + cos 𝜋) = 0. Thus, the filter reaches its first and only null at
f = f2 = (1 + 𝛼)Rs ∕2, so its null bandwidth is
{ R
(1 + 𝛼) 2s , Baseband (f centred at 0)
Bnull = (4.174)
(1 + 𝛼)Rs , Bandpass (f centred at fc ≫ Rs )
This null bandwidth has a minimum baseband value Bnull = Rs /2 when 𝛼 = 0 (which corresponds to an ideal
Nyquist channel – unrealisable in real time), and a maximum value Bnull = Rs when 𝛼 = 1 (which corresponds to
a full-cosine roll-off filter). The gain response of the filter is shown in Figure 4.56 for various values of 𝛼. With a
gain of zero at all frequencies above Bnull , the raised cosine filter has the effect of limiting the spectrum of signals
passed through it to an occupied bandwidth equal to Bnull . The (occupied) transmission bandwidth Bocc of a radio
transmission link that employs a raised cosine filter of roll-off factor 𝛼 and operates at symbol rate Rs baud is
therefore given by
Bocc = Rs (1 + 𝛼) (4.175)
Telegram: @Persian_IEEEComSoc

To compute the noise equivalent bandwidth B of the raised cosine filter, we use Eq. (4.173) in (4.171) with
|H(f )|max = 1, employing the substitution 𝜃 = 𝜋(f − f1 )∕(f2 − f1 ) to evaluate the resulting integrals
∞ f1 f2 [ ( )]2 ∞
1 f − f1
B= |H(f )| df =
2 2
1 df + 1 + cos 𝜋 df + 0 ⋅ df
∫0 ∫0 4 ∫f1 f2 − f1 ∫f2
f2 [ ( ) ( )]
1 3 f − f1 1 f − f1
= f1 + + 2 cos 𝜋 + cos 2𝜋 df
4 ∫f1 2 f2 − f1 2 f2 − f1
𝜋 𝜋
3 1f −f 1f −f 1
= f1 + (f2 − f1 ) + 2 1 2 cos 𝜃d𝜃 + 2 1 cos 2𝜃d𝜃
8 4 𝜋 ∫0 4 𝜋 ∫0 2
3 3 5
= f1 + (f2 − f1 ) + 0 + 0 = f2 + f1
8 8 8
3 R 5 R R
= (1 + 𝛼) s + (1 − 𝛼) s = s (1 − 𝛼∕4)
8 2 8 2 2
 4.7 Frequency Domain Characterisation of LTI Systems 311

x(t)
x(t) y(t) = Kx(t – to) y(t)
H(f)
t t
to

Figure 4.57 Distortionless transmission system.

Thus, the noise equivalent bandwidth of a raised cosine filter of roll-off factor 𝛼 is
{ R
(1 − 𝛼∕4) 2s , Baseband (f centred at 0)
B= (4.176)
(1 − 𝛼∕4)Rs , Bandpass (f centred at fc ≫ Rs )
Equation (4.175) and (4.176) are two very important results that will serve us well in the rest of the book.

4.7.4 Distortionless Transmission

When a signal is transmitted through a medium, it is desired that the received signal be an exact copy of the
transmitted signal except for some propagation delay, which causes the signal to be received later than it was
transmitted. This ideal performance is known as distortionless transmission. We have observed that a transmission
medium can be characterised by its transfer function, which completely specifies how the medium modifies the
amplitudes and phases of frequency components in an information signal. We are interested in determining the
specification of a transfer function that would achieve distortionless transmission.
Figure 4.57 illustrates the input and output waveforms of a distortionless transmission channel or system having
transfer function H(f ), which we seek to determine. The channel output y(t) is a scaled and delayed version of the
input x(t) according to
y(t) = Kx(t − to ) (4.177)
where K is a positive real constant. If y(t) has FT Y (f ) and x(t) has FT X(f ), then recalling the time shifting property
of FT given by Eq. (4.80) and taking the FT of both sides of this equation yields
Y (f ) = KX(f ) exp(−j2𝜋fto )
= X(f )K exp(−j2𝜋fto )
≡ X(f )H(f )
where we have used the fact stated in Eq. (4.159) that the output spectrum Y (f ) is the product of the input spec-
trum X(f ) and the system’s transfer function H(f ). Therefore, to achieve distortionless transmission, the required
channel transfer function specification is
Telegram: @Persian_IEEEComSoc

H(f ) = K exp(−j2𝜋fto )
= K∠ − 2𝜋to f
≡ |H(f )|∠𝜙H (f ) (4.178)
We see that two conditions must be satisfied, as illustrated in Figure 4.58. First, the gain response of the channel
must be constant and, second, the phase response of the channel must be linear
|H(f )| = K, Constant gain response
𝜙H (f ) = −2𝜋to f , Linear phase response (4.179)
The transfer function of practical transmission channels will in general not satisfy the above conditions without
additional filtering. Over the frequency band of interest, any departure of the gain response of the channel from a
constant value K gives rise to attenuation distortion. Similarly, any departure of the phase response from a linear
 312 4 Frequency Domain Analysis of Signals and Systems

|H(f)|
K

(a)

f
ϕH(f)

(b)
f

Slope = –2π × Delay

Figure 4.58 (a) Gain response, and (b) phase response of a distortionless transmission system.

graph gives rise to phase distortion, also called delay distortion. A parameter known as the group delay 𝜏 g of the
channel is related to the slope of the channel’s phase response by
1 d𝜙H (f )
𝜏g (f ) = − (4.180)
2𝜋 df
Substituting the expression for 𝜙H (f ) from Eq. (4.179) into the above definition, we see that a channel with no
phase distortion and hence a constant-slope phase response will have a constant group delay 𝜏g (f ) = to . For all
other channels, the phase response is nonlinear, and this means that group delay will be a function of frequency,
varying over the frequency band of interest. A frequency domain measure of the phase distortion incurred in
transmission through a distorting channel is usually given by differential delay 𝜏 d , which is the difference between
the maximum and minimum values of group delay within the frequency band of interest.

𝜏d = 𝜏g (f )|max − 𝜏g (f )|min (4.181)

This frequency domain measure of phase distortion is not to be confused with the time domain measures of
delay, namely average delay and rms delay spread, which are discussed in Section 3.2.1 in connection with multi-
path propagation. Distortionless transmission through a channel may be approximated over a desired frequency
range by using a filter known as an equaliser at the output of the channel to compensate for the amplitude and
phase distortions caused by the channel. The arrangement is as shown in Figure 4.59. Since the overall system is
distortionless, we may write
Hc (f )He (f ) = K exp(−j2𝜋fto )
Telegram: @Persian_IEEEComSoc

|Hc (f )|∠𝜙Hc (f ) × |He (f )|∠𝜙He (f ) = K∠ − 2𝜋to f

|Hc (f )||He (f )|∠(𝜙Hc (f ) + 𝜙He (f )) = K∠ − 2𝜋to f

Distortionless channel H(f) = Hc(f)He(f)

Input signal Distorting Channel Equaliser Output signal

Hc(f) He(f)

Figure 4.59 Channel equalisation.

 4.7 Frequency Domain Characterisation of LTI Systems 313

Equating the magnitude and phase terms on both sides yields

|Hc (f )||He (f )| = K
𝜙Hc (f ) + 𝜙He (f ) = −2𝜋to f

Thus
K
|He (f )| =
|Hc (f )|
𝜙He (f ) = −[2𝜋to f + 𝜙Hc (f )] (4.182)
Eq. (4.182) stipulates that
● The gain response of the equaliser should be the reciprocal of the gain response of the distorting channel. This
provides gain (or attenuation) equalisation.
● The sum of the phase response of the equaliser and that of the distorting channel should be a linear function of
frequency. This gives phase (or delay) equalisation.
Attenuation equalisation alone may be enough in some applications such as speech transmission. For example,
the attenuation (in dB) of audio telephone lines increases as the square root of frequency. The gain response of the
transmission medium in this case is
√
|Hc (f )| = K1 exp(−a f )
An equaliser with gain response
√
|He (f )| = K2 exp(+a f )
will adequately compensate for the attenuation distortion of the medium. The overall gain response of the combi-
nation of transmission medium and equaliser in tandem is flat and independent of frequency, since
√ √
|Hc (f )||He (f )| = K1 exp(−a f )K2 exp(+a f )
= K1 K2 = K, a constant

4.7.5 Attenuation and Delay Distortions

An important cause of attenuation and delay distortions is multipath propagation, which arises when the signal
arrives at the receiver having travelled over more than one path with different delays. Such a situation arises,
for example, when a radio signal is received both by direct transmission and by reflection from an obstacle. If we
consider two paths that differ in propagation time by Δ𝜏, this translates to a phase difference of 2𝜋f Δ𝜏 between two
signals x1 (t) and x2 (t) of frequency f arriving over the two paths. If the phase difference is an integer multiple of 2𝜋
Telegram: @Persian_IEEEComSoc

then x1 (t) and x2 (t) add constructively to give a received signal x(t) that has an enhanced amplitude. However, if the
phase difference is an odd integer multiple of 𝜋 then the two components add destructively, and the received signal
is severely attenuated. If x1 (t) and x2 (t) have equal amplitude then x(t) is zero under this situation. In practice, the
direct signal (termed the primary signal) has a larger amplitude than the reflected signal (termed the secondary
signal). The received signal amplitude therefore varies from a nonzero minimum under destructive interference
to a maximum value under constructive interference. For values of phase difference other than integer multiples
of 𝜋, the amplitude and phase of x(t) are determined according to the method of sinusoidal signal addition studied
in Section 2.7.3. There are three important consequences:
Because the phase difference between the two paths is a function of frequency, the amplitude and phase of the
received (resultant) signal depends on frequency. Some frequencies are severely attenuated, whereas some are
enhanced. This results in attenuation and phase distortion.
 314 4 Frequency Domain Analysis of Signals and Systems

The propagation delay difference between the two paths depends on the location of the receiver. This gives rise
to fast fading in mobile communication systems where there may be dynamic relative motion involving receiver,
transmitter, and multipath sources.
In digital communications, multipath propagation over two or more differently delayed paths gives rise to pulse
broadening or dispersion. One transmitted narrow pulse becomes a sequence of two or more narrow pulses at the
receiver and this is received as one broadened pulse. Broader pulses place a limit on the pulse rate (also known as
the symbol rate) that can be used without the overlap of adjacent pulses.
An equaliser may be used, as earlier discussed, to compensate for amplitude and phase distortions over a desired
frequency range. However, in a dynamic mobile communication environment, the gain and phase responses of
the channel will vary with time. The equaliser will therefore need to be implemented as an adaptive filter with
filter coefficients that are regularly optimised to satisfy Eq. (4.182) in the face of changing channel conditions.

4.7.6 Nonlinear Distortions

Our discussion of attenuation distortion was concerned with the modification of the shape of a signal’s amplitude
spectrum by a transmission medium. The amplitudes of existing frequency components were modified, but no
new frequency components were created. When the transmission medium or communication system is nonlinear
then the output signal is subject to a nonlinear distortion, which is characterised by the following:
The output signal is no longer directly proportional to the input signal.
The output signal contains frequency components that are not present in the input signal.
A common example of nonlinear distortion is an amplifier that is overdriven into saturation or operated in its
nonlinear region. Nonlinear distortion may also be as a result of clipping when the input signal exceeds the input
range of a system, such as a quantiser.
Analysis of nonlinear distortion can be very complex for an information signal, which contains a band of fre-
quencies, and for a transmission system having a nonlinearity that may require a high-order polynomial for
accurate representation. However, to demonstrate how the new frequency components are created, let us take
a simple signal x(t) consisting of only two sinusoids of zero phase at angular frequencies 𝜔1 and 𝜔2
x(t) = A1 cos(𝜔1 t) + A2 cos(𝜔2 t) (4.183)
Let this signal be transmitted through a nonlinear medium whose output y(t) is a nonlinear function of the
input, represented by a third-order polynomial
y(t) = a1 x(t) + a2 x2 (t) + a3 x3 (t) (4.184)
Equation (4.184) specifies what is known as the transfer characteristic of the nonlinear medium. You may verify
that using Eq. (4.183) in Eq. (4.184) gives the output of this nonlinear medium as
1
y(t) = a2 (A21 + A22 )
Telegram: @Persian_IEEEComSoc

2[ ]
1
+ a1 A1 + a3 (3A31 + 6A22 A1 ) cos 𝜔1 t
[ 4 ]
1
+ a1 A2 + a3 (3A32 + 6A21 A2 ) cos 𝜔2 t
4
1 1
+ a2 A21 cos 2𝜔1 t + a2 A22 cos 2𝜔2 t
2 2
1 1
+ a3 A1 cos 3𝜔1 t + a3 A32 cos 3𝜔2 t
3
4 4
+ a2 A1 A2 cos(𝜔1 + 𝜔2 )t + a2 A1 A2 cos(𝜔1 − 𝜔2 )t
3 3
+ a3 A21 A2 cos(2𝜔1 + 𝜔2 )t + a3 A22 A1 cos(2𝜔2 + 𝜔1 )t
4 4
3 3
+ a3 A1 A2 cos(2𝜔1 − 𝜔2 )t + a3 A22 A1 cos(2𝜔2 − 𝜔1 )t
2
(4.185)
4 4
 4.7 Frequency Domain Characterisation of LTI Systems 315

Let us take a moment to examine this equation. Note first that the nonlinear distortion is caused by the
coefficients a2 , a3 , … If these coefficients were identically zero, the transmission medium would be linear and y(t)
would be an exact replica of x(t), except for a gain factor of a1 , which would include a phase shift if a1 is complex.
Now observe the following distortions:
There is a DC component (i.e. f = 0), which was not present in the input. This happens whenever any of the
even coefficients a2 , a4 , … in Eq. (4.184) is nonzero.
The output amplitude of a frequency component present in the input signal no longer depends exclusively on
the system gain, but also on the amplitude of the other frequency component(s).
For each input frequency component 𝜔k , there appear new frequency components in the output signal at m𝜔k ,
m = 2, 3, …, N, where N is the order of the nonlinearity (N = 3 in the Eq. (4.184) illustration). Since these new
frequencies are harmonics of the input frequency, this type of distortion is termed harmonic distortion.
For any two input frequencies 𝜔1 and 𝜔2 , there appear new components at m𝜔1 ± n𝜔2 , |m| + |n| = 2, 3, …, N.
These are the sum and difference of the harmonic frequencies. This type of distortion is termed intermodulation
distortion. The frequency component at m𝜔1 ± n𝜔2 is called an intermodulation product (IMP) of order |m| + |n|.
The power in an IMP decreases with its order.
Some of the above new frequencies may fall in adjacent channel bands in frequency division multiplex (FDM) or
multicarrier satellite transponder systems and appear as unwanted interference. Increasing signal power to boost
the SNR, for example by increasing A1 and A2 in Eq. (4.183), also increases the harmonic and intermodulation
products. The practical way to minimise this type of distortion is to ensure that amplifiers and other system com-
ponents operate in their linear region. Figure 4.60 demonstrates the transmission of a signal x(t) comprising two
frequencies f 1 and f 2 through a third-order nonlinear system with coefficients a1 = 1, a2 = 0.6, and a3 = 0.12. The
harmonic distortion and intermodulation distortion components can be seen in the output spectrum. Notice that
the second-order IMPs f 2 ± f 1 have higher amplitudes than the third-order IMPs f 2 ± 2f 1 and 2f 2 ± f 1 . This is as
expected since IMP power decreases with its order.
Nonlinearity is, however, not always a nuisance. It finds extensive application in telecommunication for modu-
lation, demodulation, frequency up-conversion and down-conversion, etc. For example, if a message signal (con-
taining a band of frequencies f 1 ) is added to a sinusoidal carrier signal of frequency f2 ≫ f1 and the sum signal is
then passed through a nonlinear device, the band of second-order IMPs centred around f 2 would constitute an
amplitude modulated (AM) signal. Figure 4.60 is the special case in which the message signal is a single sinusoid of

f1 = fo f2 = 7fo
Input spectrum

nfo
Telegram: @Persian_IEEEComSoc

x(t)

Nonlinear System
y = x + 0.6x2 + 0.12x3
y(t)
f1 f2
Output spectrum

f2 – f1 f2 + f1
f=0

2f1 2f2
3f1 f2 – 2f1 f2 + 2f1 2f2 – f1 2f2 + f1 3f2
nfo

Figure 4.60 Input and output spectra of signal transmission through a nonlinear system.
 316 4 Frequency Domain Analysis of Signals and Systems

frequency f 1 . The output spectrum contains frequency components f 2 − f 1 , f 2 and f 2 + f 1 which are, respectively,
the lower side frequency, carrier, and upper side frequency of the AM signal. In this application of nonlinearity, the
other ‘unwanted’ frequency components in the output would be excluded using a bandpass filter. We have more
to say on this in Chapter 7.

4.8 Summary

The subject of frequency domain analysis of signals and systems is foundational to the study of communication
engineering. Although calculus is an indispensable tool for anyone wishing to acquire complete technical mas-
tery of this important subject, this chapter followed an engineering-first approach that made use of maths only
where necessary and prioritised clarity of understanding as well as engineering problem solving skill over pure
mathematical rigour.
You should now have a sound understanding of the various transform techniques and their interrelationships,
including the sinusoidal and complex exponential forms of the Fourier series, the Fourier transform (FT), the
discrete time Fourier series (DTFS), discrete time Fourier transform (DTFT), discrete Fourier transform (DFT), the
fast Fourier transform (FFT), the Laplace transform (LT), and the z-transform (ZT). Our treatment emphasised
how to derive from first principles the frequency domain representation of a wide range of signals, including
energy signals, power signals, periodic and nonperiodic signals, and discrete- (DT) and continuous-time (CT)
signals. We also learnt how, given the transform of signal A (from standard tables or otherwise), we may derive the
spectrum of a related signal B by exploiting relevant transform properties to modify the given transform according
to the time domain relationships between signals A and B. The most versatile result that we derived in this regard
was the Fourier series coefficients for a trapezoidal pulse train, which can be applied through straightforward
modifications to obtain the Fourier series of a wide range of standard signals, including unipolar and bipolar
rectangular, triangular, sawtooth, and ramp waveforms.
One of the novel undertakings of the chapter was our use of Fourier analysis to gain valuable insights into
the features and operational constraints of various practical telecom signals and systems. In this way, we discov-
ered aperture distortion in sample-and-hold signals, the frequency domain effect of flat-top sampling, and indeed
the sampling theorem, the spectrum and bandwidth requirements of BASK, the effectiveness of tapered window
pulses, such as the raised cosine pulse, in reducing adjacent channel interference and boosting digital transmis-
sion system capacity, etc. We examined a range of practical issues in DFT implementation and discussed ways to
avoid alias distortion, improve frequency resolution, minimise spectral leakage, prevent spectral smearing, and
reduce the variance of computed periodograms for nondeterministic signals. We discussed the inverse relation-
ship between time and frequency domains, an early indication of what will be a recurring theme throughout this
book, that communication system design is more a game of trade-offs than of free lunches. In this case, narrower
Telegram: @Persian_IEEEComSoc

pulses would allow us to operate at a higher symbol rate, but this advantage or improvement comes at the price of
a proportionately wider bandwidth.
Fourier analysis underpins two important concepts that play a pivotal role in telecommunications, namely:

● The spectrum of a signal, which specifies the relative amplitude and phase of each member of the collection of
sinusoids that constitute the signal.
● The frequency response of a communication system, which specifies how the system will alter the amplitude and
phase of a sinusoidal input signal at a given frequency when that sinusoid is passed through the system.

The signal spectrum gives a complete specification of the signal from which important signal parameters such as
bandwidth, power, and signal duration can be computed. The frequency response of a linear time-invariant system
gives a full description of the system, allowing us to determine the output signal spectrum corresponding to an
 Questions 317

arbitrary input signal, by virtue of the principle of superposition. Other important transmission parameters can
then be obtained, such as system gain, output signal power, and attenuation and phase distortions.
For a nonlinear system or device, the superposition principle does not hold. In this case, we may employ the
device’s transfer characteristic, which specifies the output signal as a polynomial function of the input. By doing
this, we find that the effect of nonlinearity is to introduce new frequency components (known as harmonic and
intermodulation products) into the output signal, which were not present in the input signal. Nonlinearity is unde-
sirable in transmission media and repeaters but is exploited in transmitters and receivers to perform important
signal processing tasks, such as frequency translation, modulation, and demodulation.
At this point, it is worth summarising the major issues in communication system design, which we address in
this book.
● The message signal must be transformed into a transmitted signal that requires a minimum transmission band-
width and experiences minimum attenuation and distortion in the transmission medium. Furthermore, those
transformation techniques are preferred in which the information contained in the transmitted signal is insen-
sitive to small distortions in the spectrum of the transmitted signal.
● The system should approach as closely as possible to a distortionless transmission. This may require the use
of equalisers to shape the overall frequency response of the system and the implementation of measures to
minimise noise and interference.
● The original information should be recovered at the intended destination from a received signal that is in general
a weak, noise-corrupted, and somewhat distorted version of the transmitted signal.
● We must be able to evaluate the performance of a communication system and to fully appreciate the significance
of various design parameters and the trade-offs involved.
In this chapter, we have introduced important concepts and tools, which will be relied upon throughout the
book to develop the above systems design and analysis skills. We delve deeper into this task in the next chapter
with a study of the three main transmission media in modern communication systems.

References

1 Cooley, J.W. and Tukey, J.W. (1965). An algorithm for machine computation of complex Fourier series. Mathe-
matics of Computation 19: 297–301.
2 Oran Brigham, E. (1988). Fast Fourier Transform and Is Applications, Pearson. ISBN: 978-0133075052.
3 S Bouguezel, M. Ahmad and M Swamy (2006), “An alternate approach for developing higher radix FFT algo-
rithms”, APCCAS 2006–2006 IEEE Asia Pacific Conference on Circuits and Systems, DOI: https://doi.org/10
.1109/APCCAS.2006.342373
4 Haykin, S. and Van Veen, B. (2003). Signals and Systems, 2e. Wiley. ISBN: 978-0471378518.
Telegram: @Persian_IEEEComSoc

5 Ifeachor, E. and Jervis, B. (2001). Digital Signal Processing: A Practical Approach. Prentice: Hall. ISBN:
978-0201596199.

Questions

4.1 .(a) Determine the fundamental frequency and first three harmonics of the periodic voltage waveforms
shown in Figure Q4.1.
(b) Does g2 (t) have the same amplitude spectrum as g3 (t)? If not, giving reasons for your answer but without
deriving any Fourier series, which of the two waveforms will have stronger higher frequency compo-
nents?
 318 4 Frequency Domain Analysis of Signals and Systems

g1(t), volts Figure Q4.1 Question 4.1.

1

(a) 0 t, μs
5

–1

g2(t), volts
5

(b)
t, ms
3
g3(t), volts
5

(c)
t, ms
3

4.2 Given the voltage waveform

g(t) = 10 − 20 cos(100𝜋t) + 10 cos(200𝜋t + 𝜋∕3) − 5 cos(500𝜋t − 𝜋∕2) volts

(a) Sketch the single-sided amplitude and phase spectra of g(t).

(b) Express g(t) as a sum of complex exponentials and hence sketch its double-sided amplitude and phase
spectra.
(c) Determine the total power in the amplitude spectrum of (a) and (b) and hence comment on the equiv-
alence or not of the two frequency domain representations.

4.3 Given that the Fourier series of a centred unipolar rectangular pulse train (RPT) (e.g. Figure 4.4a) of ampli-
tude A, waveform period T, and duty cycle d is
∑
∞
g(t) = Ad + 2Ad sinc(nd) cos(2𝜋nf o t); fo = 1∕T
Telegram: @Persian_IEEEComSoc

n=1

(a) Determine the Fourier series of the bipolar waveform g1 (t) in Figure Q4.3.
g1(t), V
10
t
T/2 T/3

T/6
–30
T

Figure Q4.3 Question 4.3.

 Questions 319

Figure Q4.4 Question 4.4. g1(t)

A
(a)
t
–T/2 0 T/2

g2(t)
A
(b)
t
–T/4 T/4
T

(b) For T = 100 μs, sketch fully labelled double-sided amplitude and phase spectra of g1 (t) up to the ninth
harmonic.

4.4 Figure Q4.4a shows a triangular waveform g1 (t) of amplitude A and period T.
(a) Obtain an expression for the normalised power of the waveform in terms of A and T.
(b) Starting from entry 10 of Table 4.5 for the Fourier transform of a triangular pulse, show that the
double-sided Fourier series of g1 (t) is given by

A ∑
∞
g1 (t) = sinc2 (n∕2) cos(2𝜋nf o t), fo = 1∕T
2 n=−∞

(c) Hence, sketch the single-sided amplitude spectrum of g1 (t) up to the sixth harmonic.
(d) Determine the null bandwidth of g1 (t) and the fraction of total power contained therein.
(e) The triangular waveform g2 (t) shown in Figure Q4.4b is identical to g1 (t) in every respect, except that
g2 (t) leads g1 (t) by one-quarter of a period. Sketch the single-sided phase spectrum (SSPS) of g2 (t) up to
the seventh harmonic.

4.5 Given that the Fourier series of a centred unipolar triangular pulse train (e.g. g2 (t) in Figure 4.4c) of ampli-
tude A, waveform period T and duty cycle d is

∑
∞
g2 (t) = Ad∕2 + Ad sinc2 (nd∕2) cos(2𝜋nf o t), fo = 1∕T
n=1
Telegram: @Persian_IEEEComSoc

(a) Determine the Fourier series of the bipolar triangular waveform g3 (t) in Figure Q4.5.
(b) For T = 1 ms, sketch a fully labelled single-sided amplitude spectrum of g3 (t) up to the sixth harmonic.

g3(t)
A

t
–T/2 T/2

–A

Figure Q4.5 Question 4.5.

 320 4 Frequency Domain Analysis of Signals and Systems

g(t)
d = τ/T

t
–T/2 –τ/2 0 τ/2 T/2

Figure Q4.6 Question 4.6.

4.6 Derive from first principles the Fourier series of the raised cosine pulse train g(t) of amplitude A and duty
cycle d shown in Figure Q4.6. One cycle of the pulse train is defined by
[ ( )]
⎧ A 1 + cos 2𝜋 t , − 𝜏 ≤ t ≤ 𝜏
⎪2 𝜏 2 2
gT (t) = ⎨
⎪0, Otherwise
⎩

4.7 Figure Q4.7 shows three pulse trains g1 (t), g2 (t), and g3 (t) that differ only in their pulse shape. g1 (t) is
rectangular, g2 (t) is triangular, and g3 (t) is raised cosine as defined in Question 4.6.
(a) Determine the 99% fractional power containment bandwidth of each pulse train, rounding percentage
values to the nearest integer.
(b) Determine the percentage of power in the main lobe of the spectrum of each pulse train.
(c) Determine the percentage of power in each of the first three sidelobes of the spectrum of each pulse
train.
(d) Based on the above results, discuss the effects of pulse shape on the spectral content of signals.

g1(t), V
100
d = τ/T = 1/3; T = 50 μs

g2(t), V
100
Telegram: @Persian_IEEEComSoc

g3(t), V
100

t
τ
T

Figure Q4.7 Question 4.7.

 Questions 321

g1(t)
d = τ/T

A
t
τ
T
g2(t)
d = τ/T
A T
t
–τ/2 τ/2
g3(t)
τ = τr + τc; d = τ/T
A T
t
–τr τc
g4(t)
d = τ/T
A

τr τ t
T

Figure Q4.8 Question 4.8.

. Starting from the results in Eq. (4.68) for the coefficients of the Fourier series of a trapezoidal pulse train,
4.8 (a)
derive the Fourier series of each of the waveforms g1 (t), g2 (t), g3 (t), and g4 (t) shown in Figure Q4.8.
(b) Validate each of the Fourier series derived above by carrying out a Fourier synthesis of each waveform
using its series up to the 15th harmonic, selecting your own waveform parameter values (i.e. ampli-
tude, period, and duty cycles). Note that a MATLAB code as in Worked Example 4.3c will help ease the
synthesis effort.

4.9 The triangular pulse train g1 (t) and ramp pulse train g2 (t) shown in Figure Q4.9 have the same amplitude,
pulse duration, and waveform period. Calculate the following parameters for each waveform.
(a) Null bandwidth.
(b) Ninety-nine percent fractional power containment bandwidth.

g1(t)
Telegram: @Persian_IEEEComSoc

d = τ/T = ½; T = 10 μs
10 V T
t
–τ/2 τ/2
g2(t)
d = τ/T = ½; T = 10 μs
10 V
T
t
–τ/2 τ/2

Figure Q4.9 Question 4.9.

 322 4 Frequency Domain Analysis of Signals and Systems

(c) Percentage of power in the main spectral lobe.

(d) Percentage of power in each of the first three spectral sidelobes.
How do the parameters compare between the two signals? Give reasons for any differences.
4.10 Figure Q4.10 shows a periodic staircase waveform g(t).
(a) Use Eq. (4.20) to calculate the values of the DC component A0 and amplitudes A1 , A2 , and A3 and phases
𝜙1 , 𝜙2 , and 𝜙3 of the first three harmonics of g(t) at respective frequencies f o , 2f o , and 3f o .
(b) Check the correctness of your results by plotting the synthesised waveform
gs (t) = A0 + A1 cos(2𝜋fo t + 𝜙1 ) + A2 cos(4𝜋fo t + 𝜙2 ) + A3 cos(6𝜋fo t + 𝜙3 )

4.11 Given that a periodic signal g(t) of period T is represented by the Fourier series
∑
∞
∑
∞
g(t) = A0 + an cos(2𝜋nf o t) + bn sin(2𝜋nf o t), fo = 1∕T
n=1 n=1

and that the normalised power P of a periodic signal g(t) of period T is

T∕2
1
P= g2 (t)dt
T ∫−T∕2
show that
1∑ 2
∞
P = A20 + A , where A2n = a2n + b2n
2 n=1 n

4.12 The Fourier series of a sinusoidal pulse train (Figure 4.22) that completes m half-cycles per pulse interval is
given by Eq. (4.56). A binary amplitude shift keying (BASK) system operates in on–off keying (OOK) fashion
at a carrier frequency of 500 kHz and bit rate 50 kb/s. Employ this equation to determine the amplitude
spectrum of this BASK signal for the following bit sequences:
(a) 100100100…
(b) 110110110…
(c) 110011001100…
Compare the bandwidth and frequency content of each spectrum with the spectrum for the fastest-changing
sequence 101010… discussed in Figure 4.24.

4.13 Figure Q4.13 shows the waveform of a binary phase shift keying (BPSK) signal g(t) for the fastest-changing
bit sequence 101010… The carrier (of frequency f c ) completes an integer number M of cycles in each bit
interval and is transmitted with phase −90∘ for binary 1 and +90∘ for binary 0.
(a) Derive the Fourier series of g(t).
Telegram: @Persian_IEEEComSoc

(b) Use this Fourier series to calculate and sketch the amplitude spectrum of a BPSK signal for the
fastest-changing sequence 101010… for a system that operates at a bit rate of 50 kb/s with carrier
frequency f c = 500 kHz.

g(t), V
20

10
3 11 13
t, ms
0 5 7 8
–10

Figure Q4.10 Question 4.10.

 Questions 323

g(t)
A

–A
T/2 T/2

Figure Q4.13 Question 4.13. BPSK waveform for fastest changing bit sequence 101010…

(c) Compare the spectrum and bandwidth of the above signal with those of Figure 4.24 for the same bit
sequence transmitted at the same bit rate and carrier frequency but using binary amplitude shift keying.

4.14 Figure Q4.14 shows the waveform of a binary frequency shift keying (BFSK) signal g(t) for the
fastest-changing bit sequence 101010… The carrier frequency is set to f 2 (which completes M cycles in a
single bit interval) for binary 1 and to f 1 (which completes Q cycles in a single bit interval) for binary 0,
where M > Q and both are positive integers.
(a) Derive the Fourier series (FS) of g(t).
(b) Use this FS to calculate and sketch the amplitude spectrum of a BFSK signal for the fastest-changing
sequence 101010… for a system that operates at a bit rate of 50 kb/s with carrier frequencies
f2 = 500 kHz for binary 1 and f1 = 200 kHz for binary 0.
(c) Compare the spectrum and bandwidth of the above signal with those of Figure 4.24 for the same bit
sequence transmitted at the same bit rate and carrier frequency but using BASK.

4.15 The triangular pulse g(t) shown in Figure Q4.15a has its Fourier transform (FT) listed in row 10 of Table 4.5
as
( )
𝜏 𝜏
G(f ) = A sinc2 f
2 2

g(t)
A

–A
T/2 T/2
Telegram: @Persian_IEEEComSoc

Figure Q4.14 Question 4.14. BFSK waveform for fastest changing bit sequence 101010…
g(t) = Atrian(t/τ) g1(t)
A A

τ
(a) t (b) t
–τ/2 τ/2 –τ 0

–A

Figure Q4.15 Question 4.15.

 324 4 Frequency Domain Analysis of Signals and Systems

Starting from this result and applying relevant FT properties, determine the FT of the bipolar triangular
pulse g1 (t) shown in Figure Q4.15b.

4.16 .(a) Starting from first principles, derive an expression for the FT of the ramp pulse shown in Figure Q4.16a
in terms of its amplitude A and duration 𝜏.
(b) Hence, using the time shifting property of FT, find the FT of the delayed ramp pulse g1 (t) in
Figure Q4.16b.
(c) Hence, using the time reversal and time shifting properties of FT, find the FT of the ramp pulse g2 (t) in
Figure Q4.16c.
(d) Hence, using the above results and the FT of a rectangular pulse listed in row 8 of Table 4.5, find the
FT of the trapezoidal pulse g3 (t) in Figure Q4.16d, which has rising edge duration 𝜏 r , constant level
duration 𝜏 c and falling edge duration 𝜏 f .
(e) Show that, under the right conditions, the result in (d) will reduce to the FT of a triangular pulse as
stated in Q4.15.

4.17 Eq. (4.68) gives the Fourier coefficients of the trapezoidal pulse train (Figure 4.26), which has period T. In
the limit T → ∞, this pulse train reduces to a trapezoidal pulse as shown in Figure Q4.16d. Starting from
Eq. (4.68) and using the relationship between the Fourier transform G(f ) and the Fourier series coefficient
Cn given in Eq. (4.72), obtain an expression for the FT of a trapezoidal pulse.

4.18 By direct computation using its defining formula, find the DFT sequence G[k] of the following data
sequences:
(a) g[n] = {0, 1, 1, 0}
(b) g[n] = {2, 0, 1, 1}.

4.19 .(a) By direct computation using its defining formula, find the inverse DFT of each of the transform
sequences G[k] obtained in Question 4.18. How does each inverse DFT compare with the original data
sequence?
(b) Use each of the above DFT pairs to verify Parseval’s theorem stated in Eq. (4.115) for discrete-time
(DT) signals. Do this by calculating the energy of each sequence in both time and frequency domains
as suggested by the theorem and checking if both results are equal.

g(t) g1(t)

(a) A (b) A
Telegram: @Persian_IEEEComSoc

t t
0 τ 0 τc/2
τf
g2(t) g3(t)

(c) A (d) A
t t
–τc/2 0 –τc/2 0 τc/2
τr τr τf

Figure Q4.16 Question 4.16.

 Questions 325

4.20 .(a) Sketch a fully labelled signal flow graph for a four-point decimation-in-frequency (DIF) fast Fourier
transform (FFT) algorithm.
(b) Using this signal flow graph, find the transform sequence of each of the data sequences g[n] = {0, 1, 1, 0}
and g[n] = {2, 0, 1, 1}. How do your results here compare with those obtained in Question 4.18 using a
direct DFT evaluation?

4.21 Derive a closed-form expression for the z-transform (ZT) of each of the following sequences, where u[n]
denotes the unit sequence:
(a) nu[n]
(b) e-𝛼n u[n]
(c) 𝛼 n u[n].

4.22 Consider the transmission of a signal through the simple LTI system shown in Figure Q4.22.
(a) Obtain the transfer function H(f ) of the system.
(b) Using Table 4.5 or otherwise, obtain the impulse response h(t) of the system.
For R = 200 Ω and C = 39.79 nF
(c) Determine the output voltage v2 (t) for the following input voltages
(i) v1 (t) = 10 cos(1000 𝜋t)
(ii) v1 (t) = 10 cos(8 × 104 𝜋t)
Comment on your results by discussing the action of the circuit.
(d) Obtain and sketch the amplitude and phase spectra of the output when the input signal is a centred
RPT (see Figure 4.4a) of amplitude A = 10 V, pulse duration 𝜏 = 20 μs, and waveform period T = 100 μs.
Discuss the manner in which the pulse train has been distorted by the system.

4.23 Figure Q4.23 shows a parallel LC circuit, usually referred to as a tank circuit.
(a) Obtain an expression for the transfer function H(f ) of the circuit and hence plot the amplitude and
phase response of the filter as a function of frequency for C = 0.5 nF, L = 240 μH, and R = 15 Ω.
(b) Show that the circuit has maximum gain at the frequency (known as the resonant frequency)
1
fr = √
2𝜋 LC
(c) Show that the circuit has 3 dB bandwidth
R
B3dB =
2𝜋L
Figure Q4.22 Question 4.22.
Telegram: @Persian_IEEEComSoc

C
ʋ1(t) R ʋ2(t)

Figure Q4.23 Question 4.23.

C L

Input ʋ1(t) R ʋ2(t) Output

 326 4 Frequency Domain Analysis of Signals and Systems

(d) A filter’s quality factor may be defined as the ratio Q = fr ∕B3dB . It gives a measure of the selectivity of
the filter, i.e. how narrow the filter’s passband is compared to its centre frequency. What is the Q of this
filter for the given component values?
(e) The shape factor of a filter is defined as the ratio between its 60 dB bandwidth and 6 dB bandwidth. It
gives a measure of how steeply the filter’s attenuation increases beyond the passband. An ideal band-
pass filter has a shape factor of 1. Determine the shape factor of this filter.

4.24 A system has the impulse response h(t) = 10𝛿(t)

(a) Determine the response of this system to an input x(t) = 50cos(800𝜋t + 45∘ )
(b) Is the system distortionless? Explain.
(c) Is there a limit on the symbol rate which this system can support?

4.25 Determine the response of a system with impulse response h(t) = 20u(t) to a pulse input
x(t) = 50cos(800𝜋t)rect(t/0.01 − 1/2). Discuss the extent of pulse dispersion in the system and the
implication on the symbol rate that can be employed without intersymbol interference.

4.26 Determine the noise equivalent bandwidth of a Gaussian filter with frequency response H(f ) =
exp(−4a𝜋 2 f 2 − j2𝜋fto ), where a and to are constants. How does this compare with the 3 dB bandwidth of
the filter?

4.27 The gain response of a Butterworth lowpass filter (LPF) of order n is given by
1
|H(f )| = √
1 + (f ∕f1 )2n
(a) Determine the noise equivalent bandwidth B of the filter.
(b) For filter order n = 1 to 5, determine how much error in dB would be incurred if the 3 dB bandwidth of
this filter were used for noise power calculation in place of B. Comment on the trend of the error as n
increases.

4.28 In Worked Example 3.6 the autocorrelation function of a sinusoidal signal g(t) = Am cos(2𝜋fm t + 𝜙m ) was
derived as
A2
Rg (𝜏) = m cos(2𝜋fm 𝜏)
2
Based on this result, show that the power spectral density (PSD) of the sinusoidal signal g(t) is
A2m A2
Sg (f ) = 𝛿(f − fm ) + m 𝛿(f + fm )
4 4
Telegram: @Persian_IEEEComSoc

and hence that the power obtained as the total area under the PSD curve is P = A2m ∕2, in agreement with
the power calculated in the time domain.

4.29 Using a frequency domain approach, show that the response y(t) of a system with impulse response
h(t) = 5rect(t/6 − 0.5) to an input x(t) = rect(t/12) is given by
[ ( ) ( )]
t t−6
y(t) = 30 trian + trian
12 12
How easy is this method compared to the graphical time-domain approach employed to solve the same
problem in Question 3.15a?
[NOTE: You may wish to refer to sections 2.6.3 and 2.6.5 for definitions of the rect() and trian() pulses used
above].
 327

Transmission Media

Our greatest hindrance is not so much what we don’t have as what we don’t use.

In this Chapter
✓ Overview of closed and open transmission media.
✓ Signal attenuation and impairments in various transmission media including copper lines, optical fibre,
and radio.
✓ Transmission line theory: a detailed discussion of wave attenuation and reflection on metallic transmission
lines and various techniques of line termination and impedance matching.
✓ Scattering parameters of two-port networks.
✓ Optical fibre: a historical review and brief discussion of fibre types, extrinsic and intrinsic losses, dispersion,
and optical fibre link design.
✓ Radio: a nonmathematical review of Maxwell’s equations and a discussion of various radio wave propa-
gation modes, effects, and mechanisms, and path loss calculations on terrestrial and earth–space radio
paths.
✓ Worked examples to demonstrate the interpretation and application of concepts and to deepen your insight
and hone your skills in engineering problem solving.
✓ End-of-chapter questions to test your understanding and (in some cases) extend your knowledge of the
material covered.

5.1 Introduction
Telegram: @Persian_IEEEComSoc

The transmission medium provides a link between the transmitter and the receiver in a communication system.
One important note on terminology should be made here. We will sometimes use the term communication channel
to mean transmission medium but will use the word channel on its own to mean the bandwidth or other system
resource devoted to one user in a multiple-user communication system. For example, in Chapter 13 we refer to
a 30-channel TDM, which means a system that can simultaneously convey 30 independent user signals in one
transmission medium using time division multiplexing (TDM).
There are two broad classifications of transmission media, namely closed and open media. Closed media enable
communication between a transmitter and a specific physically connected receiver. They include twisted wire
pair (also called paired cable), coaxial cable, optical fibre, and metallic waveguide. Open media provide broad-
cast (point-to-area communication) and mobility capabilities in telecommunication, which are not possible with
Communication Engineering Principles, Second Edition. Ifiok Otung.
© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 328 5 Transmission Media

closed media. It includes all forms of electromagnetic wave (e.g. radio and infrared) propagation not only in the
earth’s atmosphere, but also in empty space and seawater.
A metallic waveguide is preferred to coaxial cables above about 3 GHz for very short links, for example to link
an antenna to its transmitter or receiver. It is a metallic tube of rectangular or circular cross-section. An electro-
magnetic wave is launched into the waveguide using a wire loop or probe. The wave travels down the guide by
repeated reflections between the walls of the guide. Ideally, energy is totally reflected at the walls and there are
no losses. In practice, although the guide walls are polished to enhance reflection, some absorption takes place
leading to losses, which, however, are small compared to cable losses at these frequencies. Our discussion of closed
transmission media will be limited to metallic copper lines and optical fibre.
We begin the discussion of each transmission medium with a nonmathematical treatment followed by a deeper
analysis of the relevant wave propagation phenomena and concepts that equip us to understand and quantify the
limitations and impairments of each medium. Our goal is twofold: to develop the tools necessary for determining
signal attenuation, and hence the received signal strength at various points in each transmission medium, and to
apply our understanding of the characteristics of each transmission medium in the design of a communication
system, including particularly its signal power budget.

5.2 Metallic Line Systems

Figure 5.1 shows various metallic line examples, including unscreened twisted pair (UTP), screened twisted
pair (STP), coaxial line, and microstrip line. The conducting metal is usually copper, and the insulation or
dielectric material includes polystyrene (with relative permittivity 𝜀r = 2.5), polyethylene (𝜀r = 2.3), and Teflon
(𝜀r = 2.1). Metallic lines dominated the signal transmission landscape until the advent of commercial optical fibre
transmission in the 1980s culminating in TAT8 (8th transatlantic cable system), the world’s first transoceanic
undersea optical fibre system in 1988. The use of metallic lines for power transmission and short distance
telecommunication links, for example to connect an outdoor antenna to an indoor unit or a broadband/landline
hub to a street cabinet device, or for transmission within buildings will continue into the foreseeable future.
However, long-distance metallic line communication links are now obsolete. Even the use of copper cables in
the link between local exchange and customer premises (known as local loop or last mile) of the public switched
telephone network (PSTN) is not expected to continue much beyond 2025 due to the rapid deployment of optical
fibre to connect between street cabinet or customer premises and the local exchange. Many of the applications of
metallic lines in communication systems summarised below are therefore largely historical and discussed for the
sake of completeness.
Metallic lines can be used in a balanced (also called differential) mode or in an unbalanced (also called
single-ended) mode. In the unbalanced mode, one wire carries the signal while the other serves as the ground
connection. In the balanced mode, both wires carry the signal v(t) divided such that one wire carries + 1/2v(t) while
Telegram: @Persian_IEEEComSoc

the other carries − 1/2v(t). There is symmetry about ground. The receiving end reads the signal as the difference
between the voltages on the two wires which therefore cancels out common-mode noise. Two-wire lines use
balanced modes and incorporate twisting, as shown in the top row of Figure 5.1 to ensure that both wires pick
up roughly the same amount of noise and interference. Coaxial lines use the unbalanced mode, with the inner
conductor carrying the signal and the outer conductor grounded.

5.2.1 Wire Pairs

Wire pairs are made from copper in several standardised diameters, e.g. 0.32, 0.4, 0.5, 0.63, 0.9, and 1.1 mm. The
conductors are insulated with plastic, which also protects against humidity effects. A pair of insulated conduc-
tors may be twisted together, and the pair contained in plastic insulation to provide the go and return paths for
 5.2 Metallic Line Systems 329

d Insulation
Unscreened twisted
pair (UTP): s Wire pair

Insulation

Screened twisted Wire pair

pair (STP):
Braided copper screen
Rubber covering

Coaxial line: Air dielectric Polyethylene

discs
μ
L= ln (b/a) H/m
2π
a
C = 2πε/ln (b/a) F/m
Zo = 60 ln (b/a) εr b
Insulation Inner conductor
Outer conductor

Microstrip line: Dielectric

𝑤 Conductors
377 εr
Zo ≃
([𝑤/h] + 2) h

Figure 5.1 Examples of metallic lines.

Wire pair
Outer plastic cover
Conductors
Conductor insulation
Wire pair
(a) (b)

Figure 5.2 Wire pairs: (a) single pair; (b) star-quad arrangement of two wire pairs.

one circuit connection, as in Figure 5.2a. Four insulated conductors may be grouped together in a star-quad
arrangement with diagonally opposite conductors forming a pair, as shown in Figure 5.2b. This arrangement
reduces capacitance imbalance between pairs and increases packing density in multipair cable cores, when com-
pared to the twisted-pair arrangement of Figure 5.2a. Often, e.g. in a local telephone link, many wire pairs are
Telegram: @Persian_IEEEComSoc

required between two points. This is realised by making a cable with tens or even hundreds of insulated wire
pairs carried in the cable core. A sheath surrounds the cable core to give mechanical protection. If the sheath is
made from a combination of plastic and aluminium foil then it also shields the wires from electrical and magnetic
interference. The wire pairs are colour coded for pair identification.
Limiting factors in the use of wire pairs include:
● Crosstalk between pairs.
● Interference noise from power lines, radio transmission, and lightning strikes.
● Severe bandwidth limitation.
The attenuation in decibels (dB) of wire pairs increases as the square root of frequency, as shown in Figure 5.3
for a 0.63 mm diameter copper wire pair. Wire pairs make a good transmission medium for low-bandwidth
 330 5 Transmission Media

50

45

40

35
Attenuation, dB/km

30

25

20

15

10

Frequency
1kHz 10kHz 100kHz 1MHz 10MHz

Figure 5.3 Attenuation of unloaded 0.63 mm copper wire pair at 20 ∘ C.

applications. Underground copper cable cores were a massive global investment designed for the analogue
transmission technologies of the twentieth century. They began to be adapted for high-bandwidth digital trans-
mission applications well into the first decade of the twenty-first century but are now being gradually superseded
by the increasingly ubiquitous superfast fibre.
Some of the applications of copper wire pairs include:

● A wire pair is used to carry a single voice channel in telephone circuits, where the frequency band of interest is
300–3400 Hz. It is important to maintain a constant attenuation in this band to avoid distortion. This is achieved
through inductive loading, a technique invented by Oliver Heaviside in 1885 but which took several years to be
appreciated and used. Here, extra inductance is inserted in the wire pair at a regular spacing. For example, a
loading of 88 mH per 1.83 km is used in 0.63 mm audio cables to obtain a constant attenuation of ∼0.47 dB/km
up to 3.5 kHz, which is down from about 0.9 dB/km at 3.5 kHz for an unloaded wire pair. Thus, loading reduces
the attenuation of the wire pair at voice frequencies. However, the bandwidth of a loaded cable is much more
restricted, with attenuation increasing very rapidly beyond 3.5 kHz.
● Extensive growth in the use of digital communication made it necessary to explore ways of transmitting digi-
tised speech signals over these audio cables. This could not be done if the cable bandwidth were restricted to
3.5 kHz by inductive loading that was designed to support analogue speech transmission. A solution was found
Telegram: @Persian_IEEEComSoc

by removing the loading coils (called de-loading the cables) and installing regenerative repeaters at the old load-
ing coil sites. The de-loaded cables were used to carry 24-channel and 30-channel time division multiplexed
signals operating at bit rates 1.5 and 2 Mbit/s, respectively.
● Wire pairs were also specified for carrying 12–120 analogue voice channels in frequency division multiplexed
systems, necessitating operation at bandwidths up to 550 kHz. The wires were of conductor diameter 1.2 mm
and had an attenuation of about 2 dB/km at 120 kHz. The same type of wire pair was also used for digital com-
munication systems operating at bit rates of 6 and 34 Mbit/s.
● With careful installation, wire pairs have been successfully deployed within buildings for some local area net-
work (LAN) connections. UTP have been used for data rates up to 10 Mbit/s (e.g. 10Base-T Ethernet), with cable
lengths up to 100 m. STP can support data rates up to 100 Mbit/s (e.g. 100Base-T Ethernet), with cable lengths
of a few hundred metres.
 5.2 Metallic Line Systems 331

Downstream transmission
(≤ 24 Mb/s)
Local loop ADSL Data User
(To/from NSP) Splitter modem equipment

Upstream transmission POTS

(≤ 3.3 Mb/s)
Telephone
handset Subscriber equipment

Figure 5.4 ADSL system components.

● Wire pairs are still widely used (as at the end of 2019) to provide fixed-line broadband connection between
communication equipment at a customer site and the local exchange (called central office in North America)
of a network service provider (NSP). A range of digital subscriber line (DSL) technologies are used, which began
to be deployed in the late 1990s. The abbreviation xDSL is often used, where x is a placeholder for specifying
different DSL standards. The latest asymmetric digital subscriber line (ADSL2+) standard provides higher down-
load transmission bit rate or downstream rate (up to 24 Mb/s) from the NSP to the subscriber, and lower upload
bit rate or upstream rate (up to 3.3 Mb/s) from the subscriber to the NSP. ADSL allows simultaneous transmis-
sion of three independent services on one wire pair of maximum length ranging from 3.5 to 5.5 km (depending
on transmission bit rate): (i) analogue voice – called plain old telephone service (POTS), (ii) downstream data,
and (iii) upstream data. Figure 5.4 shows an FDM-based ADSL system, which requires a splitter and an ADSL
modem to place the three services into separate frequency bands in the wire pair – POTS in the lower band
from 0 to 4 kHz, upstream data in the next band up to 140 kHz, and downstream data in the final band up to
2.2 MHz. Other xDSL standards include SDSL (single-line DSL), which provides equal, and therefore symmetric,
transmission rates in both directions using a single wire pair of maximum length 3 km. It is suitable for certain
high-bandwidth applications such as video conferencing, which require identical upstream and downstream
transmission speeds. The latest standard of Very high-speed DSL (VDSL2-V+) uses up to 12 MHz of the wire pair
bandwidth to support downstream rates up to 300 Mb/s and upstream rates up to 100 Mb/s, the top data rates
being achieved only over short distances not exceeding 300 m.
● Another use of wire pairs in fixed-broadband connection is in a part-copper, part-fibre arrangement known
as fibre-to-the-cabinet (FTTC). A wire pair connects customer equipment to a street cabinet from where the
subscriber is connected via optical fibre to a digital subscriber line access multiplexer (DSLAM) located in the
local exchange. Because the signal is carried over copper for a very short distance (typically <∼300 m), a 12 MHz
bandwidth of the wire pair transmission medium can be exploited using the VDSL2-V+ standard to achieve data
rates up to 300 Mb/s as earlier stated.
Telegram: @Persian_IEEEComSoc

An important clarification is in order here. The xDSL technologies discussed above must not be confused with
an ISDN (integrated services digital network), which is a time-division-multiplex-based service that allows voice,
data, and video signals to be carried simultaneously on one transmission link such as a wire pair. The following
differences are worth noting, although it must be pointed out that the ISDN is a twentieth-century technology that
is rapidly being replaced by IP (Internet Protocol) networks.
● ISDN converts voice to a digital signal before transmission, whereas xDSL transmits voice in its original analogue
format, i.e. POTS.
● ISDN is a switched service that requires the two communicating ends to support the same ISDN standard if
transmission at optimum speed is to be realised. xDSL, on the other hand, is a point-to-point access technology.
In simple terms, there is a permanent wire pair connection between your equipment and your NSP, which allows
a faster flow of information. At the NSP premises, a splitter separates the voice (i.e. POTS) and data (e.g. video)
 332 5 Transmission Media

components of your transmission, connecting the former to the PSTN and the latter to a high-speed digital line.
Importantly, the addressee at the other end does not require an xDSL connection to receive your voice or video
message.
● ISDN requires external electrical power supply for operation, whereas xDSL carries power on the same wire
pair. In the event of a power failure, xDSL POTS service is sustained, although the data service is interrupted.
● ISDN is currently being phased out, whereas xDSL deployment is ongoing.

5.2.2 Coaxial Cable

A coaxial cable (or coax for short) consists of a plastic-covered, tube-shaped outer conductor (sometimes of braided
design) with a second conductor fixed into the centre of the tube and separated from it by air and/or plastic insula-
tion. See the third row of Figure 5.1. The ITU-T has standardised three coaxial cable sizes, specified by the diameter
of the inner conductor followed by the inner diameter of the outer conductor. Normal core coaxial cable has dimen-
sions of 2.6/9.5 mm. It has enough bandwidth to support digital transmission at bit rates up to 565 Mbit/s, or (in
the past) to carry up to 10 800 analogue speech channels in FDM systems. Small core coax has dimensions of
1.2/4.4 mm. It can be used for digital transmission at bit rates up to 140 Mbit/s. The micro coax (of dimensions
0.7/2.9 mm) is recommended for transmission at bit rates up to 34 Mbit/s. In normal core coax, the inner conduc-
tor is kept centred using polyethylene discs spaced at 30 mm. This arrangement also protects the outer conductor
from being squashed.
The outer conductor of a coax is usually earthed. As a result, it neither interferes with nor is interfered with by
nearby cables but provides screening against interference for the inner conductor. This greatly reduces radiation
loss, interference, and crosstalk, compared to twisted pair cables. The attenuation of coaxial cables increases with
frequency but is lower than the attenuation encountered in twisted pairs. See Figure 5.5. At 10 MHz the attenua-
tions of the standard coaxial cables are 7.5, 16.4, and 28.5 dB/km, in order of decreasing conductor size, whereas
the attenuation of the 0.63 mm twisted pair (Figure 5.3) is 49 dB/km.
Coaxial cable is widely used (for example by Virgin Media in the UK) to provide fixed-line broadband connection
in a part-coax, part-fibre arrangement to achieve download data rates up to 516 Mb/s and upload data rates up to
35 Mb/s. In this arrangement, a coaxial cable connects user equipment to street cabinet equipment from where

40

35
micro
30
Attenuation, dB/km

25
Telegram: @Persian_IEEEComSoc

20 small-core

15

10
normal-core
5

Frequency, MHz
0.2 1 10 20

Figure 5.5 Attenuation of coaxial cables.

 5.2 Metallic Line Systems 333

connection to the local exchange is via optical fibre. Multichannel cable TV (CATV), comprising audio, data, and
television signals, is also still distributed to home subscribers via normal-core coax. Coaxial cables are also used
for data terminal equipment interconnection within buildings. They can operate at higher bit rates with better
immunity to noise and crosstalk than twisted wire pairs.
In addition to these digital transmission systems, coaxial cable was also once popular for the following analogue
transmission applications:
● Coaxial cables were used in the interexchange network where the number of voice channels is high. The link
consisted of a cable core containing 2, 4, 6, 8, 12, 18, or 40 coaxial cables.
● Normal-core cable was used to carry FDM telephony at bandwidths of 4 MHz (960 voice channels), 12 MHz
(2700 voice channels), 18 MHz (3600 voice channels), and 60 MHz (10 800 voice channels). Repeaters had to
be more closely spaced in the higher bandwidth systems due to increased attenuation at higher frequencies.
The recommended repeater spacing ranged from 1.5 km for the 10 800-channel FDM system to 9.1 km for the
960-channel system.
● Small core coax was used for FDM telephony transmission at bandwidths of 4, 12, and 18 MHz, but with
closer repeater spacing (necessitated by higher attenuation) compared to normal core coax. For example, the
960-channel 4 MHz FDM system is carried in the small core coax with a repeater spacing of 4 km.
● Thicker coaxial cables (e.g. 9.3/37.3 mm) with a solid dielectric insulation were used for submarine lines.
The development of optical fibre led to a gradual decline (starting in the 1980s) and eventual discontinua-
tion (by the end of the 1990s) of the deployment of coax for long-distance transmission. All new installations
of long-distance signal transmission line systems are optical fibre.

5.2.3 Attenuation in Metallic Lines

In metallic lines, some of the signal energy is converted to heat dissipated in the conducting material of nonzero
resistance R (ohm, Ω), and in the insulating material of nonzero conductance G (siemens, S). In a perfect conductor
R = 0, and in a perfect insulator G = 0, so that there is no attenuation. But in practice both R and G have small
values, which results in signal energy loss at a rate (in J/s or watt) given by
{
2 R,
Irms Conductor
P=
2
Vrms G, Insulation
𝜌l 𝜎A
R= ; G= (5.1)
A l
Here, I rms is the root-mean-square (rms) current flowing in the conductor of resistivity 𝜌 (ohm-m), cross-
sectional area A, and length l (m); V rms is the rms voltage across the insulation separating the conductor pair, and
𝜎 is the conductivity in siemens per metre (S/m) of the insulating material.
Telegram: @Persian_IEEEComSoc

High-frequency current tends to flow on the outer skin of a wire conductor rather than being uniformly dis-
tributed over the entire cross-section. This phenomenon is known as skin effect. It reduces the cross-sectional area
A used by the high-frequency components of the current flow, which, from Eq. (5.1), makes these high-frequency
components experience a higher line resistance R and hence a larger attenuation.
Every metallic line has a series conductance L due to the magnetic field associated with current-carrying conduc-
tors, and a parallel capacitance C associated with a separation of positive and negative charges whenever there is
a potential difference between two conductors, which creates an electric field. These two fields store some of the
signal energy in transit, the amount stored increasing with signal frequency. Thus, the signal energy reaching the
intended destination is reduced.
Metallic lines are also subject to radiation loss, which increases at higher signal frequencies, as some of the signal
energy is carried away in the electromagnetic wave that is inevitably generated by the time-varying current in the
 334 5 Transmission Media

conductor pair. In this respect the metallic line behaves as a poorly designed antenna. Its time-varying current
induces a time-varying magnetic field in its immediate vicinity, which then induces a time-varying electric field
in its surrounding, which also induces a time-varying magnetic field, and so on. The result is a pair of electric and
magnetic fields travelling outwards in space at the speed of light that are known as electromagnetic waves.
Finally, the propagation delay of a metallic line is in general frequency-dependent, a phenomenon known as dis-
persion. This causes pulse broadening in digital communications as the pulse energy carried in different frequency
components reach the receiver at slightly different instants. As the receiver employs an observation interval equal
to the transmitted pulse duration, this dispersion will cause the receiver to find a pulse of reduced energy. There
is, however, a more serious problem associated with dispersion, known as intersymbol interference (ISI), whereby
trailing values of earlier transmitted pulses add into the present pulse interval leading to some degree of corruption
of the pulse detection process. ISI places a limitation on bit rate.
The total attenuation in metallic lines therefore increases rapidly with frequency and depends on conductor
size, material (usually copper), and temperature. For example, at 20 ∘ C and 10 MHz, a twisted wire pair of diam-
eter 0.63 mm attenuates the signal by 49 dB/km. A normal-core coaxial cable of inner/outer conductor diameters
2.6/9.5 mm, on the other hand, attenuates the signal by 7.5 dB/km at the same frequency and temperature.
The overview of metallic line systems and outline of attenuation on such lines which we have presented above
may be adequate for the needs of some readers. However, to gain a deeper insight into signal propagation, atten-
uation, and other adverse effects on metallic transmission lines and to hone our problem solving skills and learn
how to design the system to mitigate the identified problems, we must draw upon the tool of mathematics as we
delve into transmission line theory.

5.3 Transmission Line Theory

The transmission line theory presented in this section is in general required when the time it takes a signal to
travel along a transmission line from input to output is significant compared to the time it takes the signal to
complete one oscillation. This will be the case on long transmission lines and at high signal frequencies (or small
wavelengths 𝜆). As a rule of thumb, we may ignore transmission line theory and treat the line (of length l) as an
ordinary circuit with lumped values of resistance, inductance, and capacitance whenever the following condition
is satisfied
𝜆 > 20𝜋l (5.2)
A transmission line is a four-terminal device for conveying energy or information-bearing signals from one
point to another. The signal enters through two terminals at the line’s input and leaves through two terminals at
the output. A metallic line is characterised by the following four primary line constants:
Telegram: @Persian_IEEEComSoc

● Resistance R of both conductors of the line per unit length in ohm/m (Ω/m).
● Inductance L of both conductors of the line per unit length in henry/m (H/m).
● Leakage conductance G between the line conductors per unit length in siemens/m (S/m). Note that the unit of
siemens was formally called mho (which is ohm written backwards).
● Leakage capacitance C between the line conductors per unit length in farad/m (F/m).
With these four parameters constant with time, the transmission line can be treated as a linear time invariant
system, which allows us to focus on what the system does to a complex exponential input signal of frequency
f and (hence) angular frequency Ω = 2𝜋f . The response of the system to some arbitrary input at any time may
then be determined from our knowledge of the principle of superposition and the fact that the arbitrary input
signal will simply be a collection of complex exponentials. We cover this strategy in exhaustive detail in previous
chapters.
 5.3 Transmission Line Theory 335

A brief digression is in order to briefly explain the concepts of impedance and admittance in circuit analysis.
The impedance Z of a circuit element (resistor, capacitor, or inductor) is the ratio between the voltage v across the
element and the current i flowing through the element. Admittance Y is the reciprocal of impedance and hence
the ratio between the current and the voltage of the element.
For a resistor of resistance R, Ohm’s law states that v = iR, so that v/i = R, and therefore impedance
Z=R (5.3)
For an insulator of leakage conductance G, Ohm’s law again states that i = vG, so that i/v = G, and therefore
admittance
Y =G (5.4)
For a capacitor of capacitance C, the voltage v(t) = Q/C, where Q is the charge accumulated on the capacitor
plates due to the flow of current until time t. Assuming a complex exponential current i(t) = I max ej𝜔t , we have
t
1 1
v(t) = Q= I e j𝜔𝜏 d𝜏
C C ∫−∞ max
I e j𝜔𝜏 ||
t
Imax e j𝜔t
= max | =
j𝜔C ||−∞ j𝜔C
i(t)
=
j𝜔C
Thus, the impedance Z and admittance Y of the capacitor are
v(t) 1
Z= = = −j∕2𝜋fC
i(t) j𝜔C
Y = j𝜔C (5.5)
The presence of the factor j in the expression for the admittance of a capacitor is essential to capture the fact
that the current i through the capacitor leads the voltage V across the capacitor by 90∘ .
Finally, for an inductor of inductance L
di(t) d
v(t) = L = L (Imax e j𝜔t )
dt dt
= j𝜔LI max e j𝜔t
= j𝜔Li(t)
Thus, the impedance of an inductor is
v(t)
Z= = j𝜔L (5.6)
i(t)
Telegram: @Persian_IEEEComSoc

This indicates that an inductor of inductance L has impedance 𝜔L at angular frequency 𝜔 and that the voltage
across the inductor leads the current through it in phase by 90∘ .
Returning to the transmission line problem at hand, it follows that, with the primary line constants earlier stated,
a sinusoidal current of angular frequency 𝜔 (being a pair of conjugate complex exponentials at the same frequency)
will see impedance Z per unit length (accounting for voltage drop along the line) and admittance Y per unit length
(accounting for leakage current through the insulation) given by
Z = R + j𝜔L
Y = G + j𝜔C (5.7)
An infinitesimally short section of the transmission line of length denoted dx will therefore have impedance Zdx
and admittance Ydx, as shown in Figure 5.6. It is important to recognise the coordinate system that we will use
 336 5 Transmission Media

i + di i

Zdx

ʋ + dʋ Ydx ʋ

+x direction
dx
x + dx x

Figure 5.6 Transmission line section of length dx. This section length is hugely exaggerated below for illustration.

throughout in our analysis. The transmission line lies along the x axis, which increases from right to left starting
with the load termination at x = 0 (not shown in Figure 5.6). The change in voltage from v + dv at the section
input to v at the output is because of a voltage drop dv across the impedance Zdx, which has current i + di flowing
through it. Also, the change in current from i + di at the section input to i at the output is because of a leakage
current di flowing through admittance Ydx, which has a voltage drop v across it. Applying Ohm’s law to these two
situations allows us to write
dv = (i + di) Zdx = iZdx (since i ≫ di)
di = vYdx
Restating the first and second lines above
dv
= iZ (i)
dx
di
= vY (ii) (5.8)
dx
Taking the derivative of (i) with respect to (wrt) x and using (ii) to substitute for the resulting di/dx yields
d2 v
= ZYv (5.9)
dx2
Similarly, taking the derivative of (ii) wrt x and using (i) to substitute for dv/dx yields
d2 i
= ZYi (5.10)
dx2
Equation (5.9) is a second-order differential equation, the solution of which gives the voltage v on the line. The
form of this equation, whereby the second derivative of v wrt x equals a constant times v, suggests that the variation
of v with x is exponential and of the form v = K 1 exp(𝛾x), where K 1 and 𝛾 are independent of x. We have
Telegram: @Persian_IEEEComSoc

dv d2 v
v = K1 e𝛾x ; = 𝛾K1 e𝛾x ;2
= 𝛾 2 K1 e𝛾x
dx dx
Substituting the above expressions for v and d2 v/dx2 into Eq. (5.9) yields
𝛾 2 K1 e𝛾x = ZYK 1 e𝛾x
√
⇒ 𝛾 2 = ZY ; ⇒ 𝛾= ZY
Thus, v = K 1 exp(𝛾x) is a solution, provided
√
𝛾 = ZY ≡ 𝛼 + j𝛽 (5.11)
 5.3 Transmission Line Theory 337

where 𝛼 is the real part of√𝛾 and 𝛽 its imaginary part. Following the same steps, we find that v = K 2 exp(−𝛾x) is also
a solution, provided 𝛾 = ZY . Therefore, the general solution for the voltage on the line is
v = K1 e𝛾x + K2 e−𝛾x (5.12)
Now that we have a solution for v, we make use of Eq. (5.8) (i) to obtain a solution for current i on the line as
follows
√ √
1 dv 1 d
i= = (K1 e ZY x + K2 e− ZY x )
Z dx Z dx
1 √ √
= ( ZY K1 e𝛾x − ZY K2 e−𝛾x )
Z
Thus
K K
i = √ 1 e𝛾x − √ 2 e−𝛾x (5.13)
Z∕Y Z∕Y
Equations (5.12) and (5.13) explicitly show the variation of v and i with distance x along the transmission
line. The variation of v and i with time is implicit in K 1 and K 2 , which are functions of time in the form of
complex exponentials of angular frequency 𝜔 since the input signal is a complex exponential. So, substituting
K1 = A1 e j𝜔t , K2 = A2 e j𝜔t in the two equations, we obtain the final expressions for voltage and current on the trans-
mission line as follows
v = A1 e j𝜔t e𝛾x + A2 e j𝜔t e−𝛾x = A1 e j𝜔t e(𝛼+j𝛽)x + A2 e j𝜔t e−(𝛼+j𝛽)x

= A1 e𝛼x e j(𝜔t+𝛽x) + A2 e−𝛼x e j(𝜔t−𝛽x) ≡ v1 + v2 (5.14)

A A
i = √ 1 e𝛼x e j(𝜔t+𝛽x) − √ 2 e−𝛼x e j(𝜔t−𝛽x) ≡ i1 − i2 (5.15)
Z∕Y Z∕Y
Equations (5.14) and (5.15) are hugely significant results that encapsulate all the features of signal transmission
on metallic lines, as elaborated in the subsections below.

5.3.1 Incident and Reflected Waves

The above solution stipulates that the voltage v is not just a signal varying with time, but it is a function of both time
t and distance x along the line. In other words, the voltage v (and similarly for the current i) is a ‘disturbance’ or
wave that propagates down the line while also varying with time at any given spot. The solution further stipulates
that there are two voltage waves on the line, namely
v1 = A1 e𝛼x e j(𝜔t+𝛽x)
Telegram: @Persian_IEEEComSoc

v2 = A2 e−𝛼x e j(𝜔t−𝛽x) (5.16)

Since v1 = A1 e𝛼x e j(𝜔t+𝛽x)
≡ A1 e𝛼x ∠(𝜔t e𝛼x
+ 𝛽x), i.e. v1 has magnitude A1 and phase 𝜔t + 𝛽x radian, it means that
to track a particular phase of v1 one must decrease x as time t increases so that phase 𝜔t + 𝛽x remains constant.
That is, the location x of a phase of v1 (such as the crest for which 𝜔t + 𝛽x = 0) decreases with time. v1 is therefore
a wave travelling in the −x direction towards the load at x = 0 and is therefore referred to as the incident wave.
The amplitude of v1 equals A1 at x = 0, t = 0. This amplitude increases exponentially with +x by the factor e𝛼x .
Another way of putting this is that the amplitude of the incident wave decreases exponentially as it travels along
the line towards the load. By a similar consideration, we see that v2 is a wave travelling in the +x direction away
from the load and is therefore referred to as the reflected wave. The amplitude of v2 equals A2 at x = 0, t = 0, and
 338 5 Transmission Media

then decreases exponentially with x by the factor e−𝛼x . The line also carries incident and reflected current waves
i1 and i2 , respectively, given by
A
i1 = √ 1 e𝛼x e j(𝜔t+𝛽x)
Z∕Y
A2 −𝛼x j(𝜔t−𝛽x)
i2 = √ e e (5.17)
Z∕Y

5.3.2 Secondary Line Constants

√
Since impedance and admittance are in general complex quantities, the parameter 𝛾 = ZY given in Eq. (5.11) is
also a complex quantity, known as the propagation constant of the line, having a real part 𝛼 and an imaginary part
𝛽. From Eq. (5.16), the amplitudes V 1 (x + l) and V 1 (x) of the incident wave at distances x + l and x away from the
load are
V1 (x + l) = A1 e𝛼(x+l) ; V1 (x) = A1 e𝛼x
The attenuation in dB imposed by the transmission line on the signal over length l is
( ) ( )
V1 (x + l) A1 e𝛼(x+l)
20log10 = 20 log10 = 20 log10 (e𝛼l )
V1 (x) A1 e𝛼x
= 20𝛼l log10 e
= 8.686𝛼l (dB)
Thus
Line attenuation = 8.686𝛼 dB per unit length (5.18)
Note that this attenuation is equivalent to 𝛼 neper per unit length – see Eq. (2.64). The parameter 𝛼 is called the
attenuation constant of the line and has units of neper per metre (Np/m). A loss-free line has 𝛼 = 0. Recalling that
the phase of v1 is 𝜔t + 𝛽x radian, it follows that, observing the wave at some time instant t (i.e. by keeping t fixed,
which is akin to taking a snapshot of the wave), we will find that the phase of the wave changes by 𝛽 radian per
unit length. The parameter 𝛽 is therefore known as the phase constant (sometimes called the wavenumber) of the
line. The distance over which the phase of the wave changes by 2𝜋 radian is known as the wavelength of the wave,
so 𝛽𝜆 = 2𝜋, and we may write
2𝜋
𝛽= (5.19)
𝜆
On a loss-free line, the incident and reflected wave expressions in Eq. (5.16) may be simplified by substituting
𝛼 = 0, expanding the complex exponential using Euler’s formula (ej𝜃 = cos𝜃 + jsin𝜃), and retaining only the real
Telegram: @Persian_IEEEComSoc

part to obtain
v1 = A1 cos(𝜔t + 𝛽x)
v2 = A2 cos(𝜔t − 𝛽x) (5.20)
Figure 5.7 shows snapshots of these two waves taken at time instants t = 0, T/4, and T/2, where T = 1/f = 2𝜋/𝜔
is the wave period. Putting a tracker (♣) on the leftmost crest (defined as 𝜔t + 𝛽x = 6𝜋) of the incident wave at
( will be at a)location given by 𝛽x = 6𝜋 − 𝜔t or x = (6𝜋 − 𝜔t)/𝛽. Thus
t = 0, we see that at some later time t this crest
6𝜋 2𝜋
At t = 0, crest is located at x = = 3𝜆 since =𝜆
𝛽( ) 𝛽
T 1
At t = T/4, crest is located at x = 6𝜋 − 𝜔 = 2.75𝜆 (since 𝜔T = 2𝜋)
4 𝛽
 5.3 Transmission Line Theory 339

ʋ2 = A2 cos(ωt – kx) ʋ1 = A1 cos(ωt + kx)

t=0
x

3λ 5λ/2 2λ 3λ/2 λ λ/2 0

β = 2π/λ
t = T/4 ω = 2π/T

3λ 5λ/2 2λ 3λ/2 λ λ/2 0

t = T/2
x

3λ 5λ/2 2λ 3λ/2 λ λ/2 0

Figure 5.7 Snapshots of incident and reflected waves at t = 0, T/4, and T/2.

( )
T 1
At t = T/2, crest is located at x = 6𝜋 − 𝜔 = 2.5𝜆
2 𝛽
So, a constant-phase point of the incident wave travels in the −x direction on the line at a speed known as the
phase velocity and given by
3𝜆 − 2.75𝜆 𝜆 𝜔
vp = = = f𝜆 = (5.21)
T∕4 T 𝛽
Similarly, putting a tracker (⧫) on the third crest of the reflected wave (defined as 𝜔t − 𝛽x = 2𝜋) at t = 0, we see
that at some later time t this crest is at a location given by (2𝜋 + 𝜔t)∕𝛽. Thus
2𝜋
At t = 0, tracked crest is at x = =𝜆
𝛽( )
T 1
At t = T/4, tracked crest is at x = 2𝜋 + 𝜔 = 1.25𝜆
( 4 𝛽 )
Telegram: @Persian_IEEEComSoc

T 1
And at t = T/2, tracked crest is at x = 2𝜋 + 𝜔 = 1.5𝜆
2 𝛽
So, a constant-phase point of the reflected wave travels to the left in the +x direction on the line at the same
phase velocity vp given in Eq. (5.21), which is applicable to incident and reflected voltage and current waves on
both lossy and loss-free lines.
The constants discussed above are usually referred to as secondary line constants to distinguish them from the
primary line constants (R, L, C, and G) from which they may be calculated using approximate formulas that are
applicable at specified frequency ranges. In what follows we use the well-known approximation
n(n − 1) 2 n(n − 1)(n − 2) 3
(1 + a)n = 1 + na + a + a +···
2 3×2
≈ 1 + na, a≪1 (5.22)
 340 5 Transmission Media

At radio frequencies when 𝜔L ≫ R and 𝜔C ≫ G, propagation constant 𝛾 becomes

√ √ √
𝛾 = ZY = (R + j𝜔L)(G + j𝜔C) = j𝜔L(R∕j𝜔L + 1)j𝜔C(G∕j𝜔C + 1)
√ √ ( R
)(
G
)
= j𝜔 LC(1 + R∕j𝜔L)1∕2 (1 + G∕j𝜔C)1∕2 ≃ j𝜔 LC 1 + 1+
2j𝜔L 2j𝜔C
( ) √ √
√ R G 1 R G R C G L √
≃ j𝜔 LC 1 + + − ≃ + + j𝜔 LC
2j𝜔L 2j𝜔C 4 𝜔L 𝜔C 2 L 2 C
≡ 𝛼 + j𝛽

Therefore, at radio frequencies when 𝜔L ≫ R and 𝜔C ≫ G, attenuation constant and phase constant are given by
√ √
R C G L
𝛼= +
2 L 2 C
√
𝛽 = 𝜔 LC (5.23)

At audio frequencies when R ≫ 𝜔L but G ≪ 𝜔C, the propagation constant becomes

√ √
𝛾 = (R + j𝜔L)(G + j𝜔C) ≃ (R)(j𝜔C)
√ √ √ √
≃ j𝜔CR = 𝜔CR(j)1∕2 = 𝜔CR(e j𝜋∕2 )1∕2 = 𝜔CRe j𝜋∕4
( )
√ √ 1 1
∘
= ( 𝜔CR) × 1∠45 = 𝜔CR √ + j √
2 2
√ √
= 𝜔CR∕2 + j 𝜔CR∕2 ≡ 𝛼 + j𝛽

Thus, at audio frequencies, when R ≫ 𝜔L and G ≪ 𝜔C, the attenuation constant 𝛼 and phase constant 𝛽 are
given approximately by
√
𝛼 = 𝛽 = 𝜔CR∕2 (5.24)

Using the approximate formulas for 𝛽 in Eqs. (5.23) and (5.24), we obtain approximate formulas for phase veloc-
ity at radio and audio frequencies as
𝜔
vp =
𝛽
⎧ 𝜔 1
⎪ √ = √ , Radio frequencies
⎪ 𝜔 LC LC
Telegram: @Persian_IEEEComSoc

≈⎨ √ (5.25)
⎪√ 𝜔 =
2𝜔
, Audio frequencies
⎪ 𝜔CR∕2 RC
⎩

5.3.3 Characteristic Impedance

The characteristic impedance Z o of a transmission line is the ratio between voltage and current for a single wave
travelling in one direction. Equivalently, characteristic impedance is the input resistance of an infinite length of
the line. It is also the ratio of electric to magnetic field strengths of a single wave and is analogous to the wave
impedance of free space, but has a different value determined by the primary line constants. We will shortly find
that when a line is terminated by its characteristic impedance, there is no reflection from the end of the line, so
 5.3 Transmission Line Theory 341

this is an important parameter in the design of efficient transmission line systems. From Eqs. (5.16) and (5.17) the
ratio between incident voltage v1 and incident current i1 is
v1 A1 e𝛼x e j(𝜔t+𝛽x)
Zo = =
i1 A1 𝛼x j(𝜔t+𝛽x)
√ e e
Z∕Y
√ √
Z R + j𝜔L
= =
Y G + j𝜔C
√
L(R∕L + j𝜔)
= (5.26)
C(G∕C + j𝜔)
At radio frequency (RF) (where 𝜔 ≫ R∕L, 𝜔 ≫ G∕C) or for a loss-free line (where R = G = 0) or under Heaviside
condition (G/C = R/L), the last expression reduces to
√
L
Zo = (5.27)
C
which is independent of frequency and is also a real quantity. At audio frequencies when R ≫ 𝜔L but G ≪ 𝜔C,
we have
√
Zo = R∕j𝜔C (5.28)
In an actual line, G∕C ≪ R∕L so it was common practice in the past to add loading coils at regular intervals
along transmission lines in order to make L large enough to satisfy the Heaviside condition
G R
= (5.29)
C L
which, as stated above, is required to ensure Z o is real and independent of frequency. This coil loading practice has
long been discontinued for reasons discussed in the overview of wire pairs. In the few remaining installations of
metallic lines for long-distance (> ∼ 2 km) signal transmission nowadays, digital repeaters have replaced loading
coils.

Worked Example 5.1 Primary and Secondary Line Constants

A coaxial cable, as shown in the third row of Figure 5.1, consists of copper-conducting material and air dielectric
medium. The diameter of the inner conductor is a = 0.7 mm and the inner diameter of the outer conductor is
b = 2.9 mm. Resistivity of the air medium is in excess of 109 Ω m, so leakage conductance G ≈ 0. A matched load
is connected to the receiving end of 100 m of this cable. The sending end is connected to a 10 MHz sinusoidal
signal generator and adjusted to give 10 V rms measured across its input terminals. Given the following primary
parameters of the relevant media
Telegram: @Persian_IEEEComSoc

Resistivity of copper material 𝜌 = 1.72 × 10−8 Ω m

Relative permeability of copper 𝜇 r = 1
Relative permittivity of air 𝜀r = 1
Permittivity of free space 𝜀o = 8.8541878128 × 10−12 F/m
Permeability of free space 𝜇 o = 1.25663706212 × 10−6 H/m

(a) Determine the primary line constants R, L, and C of the cable.

(b) Determine the secondary line constants 𝛼 and 𝛽.
(c) Calculate the magnitude of the receiving end current and receiving end voltage.
(d) Calculate the efficiency of the transmission line.
 342 5 Transmission Media

(a) Primary line constant R is resistance per unit length. Eq. (5.1) gives
𝜌l 1.72 × 10−8 × 1 1.72 × 10−8 × 1
R= = =
A 𝜋(a∕2)2
𝜋(0.7 × 10−3 ∕2)2
= 0.045 Ω∕m
Formulas for inductance L and capacitance C per unit length are given in row 3 of Figure 5.1
( ) ( ) ( )
𝜇 b 𝜇𝜇 b 1 × 1.25663706212 × 10−6 2.9
L= ln = r o ln = ln
2𝜋 a 2𝜋 a 2𝜋 0.7
= 284 nH∕m
2𝜋𝜀 2𝜋𝜀r 𝜀o 2𝜋 × 1 × 8.8541878128 × 10−12
C= = =
ln(b∕a) ln(b∕a) ln(2.9∕0.7)
= 39.14 pF∕m
(b) Operating frequency f = 10 MHz, so we first check to see whether the RF condition (𝜔L ≫ R, 𝜔C ≫ G) is
satisfied
𝜔L = 2𝜋 × 10 × 106 × 284 × 10−9 = 17.84 ≫ R = 0.045
𝜔C = 2𝜋 × 10 × 106 × 39.14 × 10−12 = 0.00246 ≫ G = 0
RF condition is therefore satisfied, so we employ the expressions for 𝛼 and 𝛽 in Eq. (5.23) that are applicable
at RF to obtain
√ √ √
R C G L 0.045 3.914 × 10−11
𝛼= + = +0
2 L 2 C 2 2.84 × 10−7
= 2.622 × 10−4 Np∕m
√ √
𝛽 = 𝜔 LC = 2𝜋 × 10 × 106 2.84 × 10−7 × 3.914 × 10−11
= 0.21 rad∕m = 12 deg /m
(c) Characteristic impedance of the line (under RF conditions) is given by
√ √
L 2.84 × 10−7
Zo = = = 85 Ω
C 3.914 × 10−11
(d) Sending end voltage V s = 10 V rms. Since the line is terminated with a matched load, the signal source sees an
input impedance equal to Z o . Thus, sending end current I s = V s /Z o = 117.3 mA. As the voltage and current
waves travel down the line through length l = 100 m to the receiving end, they are attenuated by 8.686𝛼l
dB = 0.2278 dB. Thus, receiving end voltage V o and receiving end current I o are given by
Vo = Vs × 10−0.2278∕20 = 10 × 10−0.2278∕20 = 9.74 V rms
Telegram: @Persian_IEEEComSoc

Io = Is × 10−0.2278∕20 = 117.3 × 10−0.2278∕20 = 114.3 mA rms

(e) Transmission line efficiency 𝜂 is given by the ratio between output power and input power. Thus
Vo I o 9.74 × 0.1143
𝜂= × 100% = × 100% = 94.89%
Vs Is 10 × 0.1173

5.3.4 Reflection and Transmission Coefficients

A transmission line is typically employed to convey a signal from a source to a receiving end or load that is char-
acterised by an impedance Z L . The signal will take some time to propagate as a wave along the line from its source
 5.3 Transmission Line Theory 343

i2
i = i1 – i 2
i1
ZL ʋ = ʋ1+ ʋ2

+x direction Zx
x=0

Figure 5.8 Line terminated by load impedance Z L at x = 0.

to the load, which could be a transmitting antenna, a low noise amplifier, or other electrical system. We wish to
examine what happens to the signal when it arrives at the load. Ideally, the signal should be delivered in its entirety
to the load, as any reflection represents not only a signal loss to the load but also a potentially damaging return
signal towards the source.
Figure 5.8 shows a transmission line terminated by a load impedance Z L at x = 0. Recall that distance along the
line is measured leftward from the load. The impedance measured between the line pair at a distance x from the
load is known as the line impedance at x, denoted Z x . This impedance is the ratio between the resultant voltage and
resultant current at the location and will in general vary with distance, not necessarily being equal to characteristic
impedance Z o since there are two voltage waves v1 and v2 and two current waves i1 and i2 on the line. An incident
current i1 flows along the line towards the load and a reflected current i2 flows along the same line away from the
load, giving a resultant current i = i1 − i2 at any point. The incident voltage v1 and reflected voltage v2 combine to
give a resultant voltage v = v1 + v2 .
The voltage reflection coefficient 𝜌v provides a measure of how much reflection takes place at the load and is
defined as the ratio between reflected voltage v2 and incident voltage v1 at the load. At the load (x = 0), Eqs. (5.16),
(5.17), and (5.26) give
v1 = A1 e𝛼x e j(𝜔t+𝛽x) |x=0 = A1 e j𝜔t
And similarly
A A A2 j𝜔t
v2 = A2 e j𝜔t ; i1 = √ 1 e j𝜔t = 1 e j𝜔t ; i2 = e
Z∕Y Zo Zo
so that
v2 || A
𝜌v = | = 2
v1 |x=0 A1
A2 = 𝜌v A1 = |𝜌v |A1 ∠𝜃v (5.30)
where 𝜃 v is the angle of 𝜌v which is in general a complex quantity. 𝜃 v is the phase shift that is added to v2 purely
Telegram: @Persian_IEEEComSoc

by the reflection process. A reflection-induced phase shift will occur whenever the characteristic impedance Z o
and load Z L are not both a pure resistance. One other situation where a reflection-induced phase shift will occur
is that if both Z o and Z L are a pure resistance with Z L < Z o , then phase shift 𝜃 v = 180∘ . Since Z L = v/i = (v1 + v2 )/
(i1 − i2 ), and (by definition) Z o = v1 /i1 = v2 /i2 , which gives i1 = v1 /Z o and i2 = v2 /Z o , it follows that
v1 + v2 || A1 e j𝜔t + A2 e j𝜔t
ZL = | =
i1 − i2 |x=0 (A1 ∕Zo )e j𝜔t − (A2 ∕Zo )e j𝜔t
A1 + A2 1 + A2 ∕A1
= = Z
A1 ∕Zo − A2 ∕Zo 1 − A2 ∕A1 o
1 + 𝜌v
= Z
1 − 𝜌v o
 344 5 Transmission Media

Rearranging to make 𝜌v the subject of the equation yields

ZL − Zo
𝜌v = ≡ |𝜌v |∠𝜃v (5.31)
ZL + Zo
where we have emphasised the fact that the reflection coefficient 𝜌v is in general complex, having magnitude |𝜌v |
and angle 𝜃 v , which amounts to a reflection-induced phase shift.
The voltage transmission coefficient 𝜏 v is defined as the ratio between the voltage v that is delivered to the load
and the incident voltage v1 . Thus
v v + v2 v
𝜏v = = 1 = 1 + 2 = 1 + 𝜌v
v1 v1 v1
2ZL
= (5.32)
ZL + Zo
The voltage vL delivered to a load, which could be another transmission line, may be expressed in terms of
incident voltage v1 and voltage transmission coefficient 𝜏 v as

vL = 𝜏v v1 (5.33)

Notice in Eq. (5.32) that if Z L = Z o , then 𝜏 v = 1 and the voltage delivered to the load equals the incident voltage.
The current reflection coefficient 𝜌i is the ratio of reflected current i2 to incident current i1 . With direction of
current flow as shown in Figure 5.8, it follows that
i2 v ∕Z v
𝜌i = − =− 2 o =− 2
i1 v1 ∕Zo v1
= −𝜌v (5.34)

The current transmission coefficient 𝜏 i is the ratio between the current iL delivered to the load and the incident
current i1 . Thus
iL i −i i
𝜏i = = 1 2 =1− 2
i1 i1 i1
= 1 + 𝜌i = 1 − 𝜌v (5.35)

The primary coefficient is therefore the voltage reflection coefficient 𝜌v . Once 𝜌v is known, all the other coeffi-
cients may be calculated as follows

Voltage transmission coefficient: 𝜏v = 1 + 𝜌v

Current reflection coefficient: 𝜌i = −𝜌v

Telegram: @Persian_IEEEComSoc

Current transmission coefficient: 𝜏i = 1 − 𝜌v (5.36)

The extent of reflections on a transmission line may also be quantified in terms of signal powers. A parameter
known as return loss (RL) is defined as the ratio of incident power P1 to reflected power P2 expressed in dB. Thus
( ) ( ) ( ) ( )
P1 A21 | A1 | | A2 |
RL = 10 log10 = 10 log10 = 20 log | | = −20 log | |
10 |A | 10 |A |
P2 A22 | 2| | 1|

= −20 log10 (|𝜌v |) dB (5.37)

Notice that in an open- or short-circuited line, |𝜌v | = 1, so that RL = 0, whereas in a matched line |𝜌v | = 0 and
RL = ∞. The design goal is therefore usually a large return loss.
 5.3 Transmission Line Theory 345

Mismatch loss (ML) is defined as the ratio (in dB) between incident power P1 and power PL delivered to the load.
Thus
( ) ( )
P1 P1
ML = 10 log10 = 10 log10
P P1 − P2
( L )
1
= 10 log10 = −10 log10 (1 − P2 ∕P1 )
1 − P2 ∕P1
= −10 log10 (1 − |𝜌v |2 ) dB (5.38)

We see that an open- or short-circuited line leads to ML = ∞, since |𝜌v | = 1; and a matched line, for which
|𝜌v | = 0, has ML = 0. Thus, the aim in design will be to achieve a negligibly small mismatch loss.
Finally, the power transfer efficiency of the transmission line is defined as the percentage ratio between power
delivered to the load and incident power. Thus
PL
Power transfer efficiency = × 100%
P1
P − P2
= 100 1
P1
= 100(1 − |𝜌v |2 ) (5.39)

5.3.5 Standing Waves

The voltage signal v(x, t) observed on the transmission line at time t and distance x from the load is the resultant
or sum of the incident wave v1 and reflected wave v2 . There will be points along the line where v1 and v2 are exactly
in phase and their resultant has maximum amplitude Amax . There are also points along the line where v1 and v2
are exactly out of phase so that their resultant has minimum amplitude Amin . The rest of the points along the line
will have a range of amplitudes between Amin and Amax . What we have here is therefore a situation where two
oppositely propagating waves create a pattern of ‘disturbance’ at each location along the line. Once enough time
has elapsed for the two waves to be established throughout the line, every point along the line is actively ‘disturbed’
but each point by a different amount, ranging from a peak-to-peak voltage variation of 2Amin to a peak-to-peak
variation of 2Amax . This pattern of disturbance is called a standing wave pattern.
For simplicity, let us assume a loss-free so that

v1 = A1 cos(𝜔t + 𝛽x); v2 = A2 cos(𝜔t − 𝛽x + 𝜃v )

where A1 and A2 are positive and real and any reflection-induced phase shift in v2 has been transferred into the
angle as 𝜃 v . At point x, v1 and v2 are just two sinusoidal signals of the same (angular) frequency 𝜔 but different
amplitudes A1 and A2 and different phases 𝛽x and −𝛽x + 𝜃 v . We therefore use the method of sinusoidal addi-
Telegram: @Persian_IEEEComSoc

tion discussed in Section 2.7.3 to obtain the resultant signal v(x, t) = A cos(𝜔t + 𝜙(x)). Making use of relevant
trigonometric entries to simplify the expression for resultant amplitude A, the in-phase component AI , quadrature
component AQ , resultant amplitude A, and phase 𝜙(x) of v(x, t) are

AI = A1 cos(𝛽x) + A2 cos(−𝛽x + 𝜃v )
AQ = A1 sin(𝛽x) + A2 sin(−𝛽x + 𝜃v )
√
A = A2I + A2Q
√
= A21 + A22 + 2A1 A2 cos(2𝛽x − 𝜃v )
𝜙(x) = tan−1 (AQ ∕AI )
 346 5 Transmission Media

There is a little more subtlety to the phase 𝜙(x) than indicated by the above formula. See Eq. (2.52) for details.
Our interest here is, however, only in the amplitude A. Using Eq. (5.30) to make the substitution A2 = |𝜌v |A1 in
the above expression for A, we obtain that the sinusoidal voltage signal at a distance x from the load has amplitude
√
A = A21 + A22 + 2A1 A2 cos(2𝛽x − 𝜃v )
√
= A1 1 + |𝜌v |2 + 2|𝜌v | cos(2𝛽x − 𝜃v ) (5.40)
This equation indicates that the variation of resultant voltage amplitude with distance along the line is a func-
tion of the voltage reflection coefficient 𝜌v . If 𝜌v = 0 then A = A1 at every point on the line. A line terminated with
a matched load Z L = Z o has no signal reflection and hence no standing wave pattern. In all other situations of
an unmatched line, 𝜌v will be nonzero, and the resultant amplitude will vary with distance. Since, in Eq. (5.40),
cos(2𝛽x − 𝜃 v ) ranges in value from −1 to +1, it means that, at locations x where cos(2𝛽x − 𝜃 v ) = 1, the resultant
voltage will have maximum amplitude Amax , whereas locations x at which cos(2𝛽x − 𝜃 v ) = −1 will have mini-
mum amplitude Amin . A point of maximum amplitude is often referred to as an antinode and a point of minimum
amplitude a node. Locations of nodes and antinodes are fixed along the line; they do not drift with the waves; they
just stand in their various locations, hence the term standing waves. Denoting antinode locations as xa and node
locations as xd , and recalling that cosΘ = 1 for Θ = 0, 2𝜋, 4𝜋, …, 2n𝜋, and that cosΘ = −1 for Θ = 𝜋, 3𝜋, 5𝜋, …,
(2n+1)𝜋, we solve cos(2𝛽xa − 𝜃 v ) = 1 and cos(2𝛽xd − 𝜃 v ) = −1 to obtain the following results (using Eq. (5.19) to
aid simplification)
( )
2n𝜋 + 𝜃v 𝜃 𝜆
xa = = n+ v
2𝛽 2𝜋 2
( )
(2n + 1)𝜋 + 𝜃v 1 𝜃v 𝜆
xd = = n+ +
2𝛽 2 2𝜋 2
n = 0, 1, 2, 3, · · · (5.41)
Therefore, antinodes are spaced apart by half a wavelength along the line. Nodes are also spaced apart by 𝜆/2,
and the distance between a node and an adjacent antinode is 𝜆/4. If 𝜃 v = 0 (i.e. no phase shift upon reflection) then
the first antinode is located at the load and the first node is located one-quarter of a wavelength from the load.
The maximum and minimum resultant amplitudes are obtained by substituting cos(2𝛽x − 𝜃 v ) = ±1 into
Eq. (5.40)
√ √
Amax = A1 1 + |𝜌v |2 + 2|𝜌v | = A1 (1 + |𝜌v |)2
= A1 (1 + |𝜌v |)
√ √
Amin = A1 1 + |𝜌v |2 − 2|𝜌v | = A1 (1 − |𝜌v |)2
= A1 (1 − |𝜌v |) (5.42)
Telegram: @Persian_IEEEComSoc

The ratio Amax /Amin is called the voltage standing wave ratio (VSWR). Thus
1 + |𝜌v |
VSWR = (5.43)
1 − |𝜌v |
VSWR is a real and positive parameter that gives an important indication of how well an electrical load such
as an antenna is matched to a transmission line that feeds it. Under perfect matching, 𝜌v = 0 and VSWR has its
smallest value of 1.0. Since impedance is a function of frequency, such perfect matching will only be achieved at one
frequency, say f c . The level of mismatch and hence VSWR and the amount of lost power (i.e. power reflected away
 5.3 Transmission Line Theory 347

Zo = 100 Ω; ZL = 100 Ω; VSWR = 1

2
Matched line:
1
ZL = Zo
0
Zo = 100 Ω; ZL = ∞; VSWR = ∞
Resultant amplitude A (volt) →

2
Open line:
1
AN N AN N AN ZL = ∞
0
Zo = 100 Ω; ZL = 0 Ω; VSWR = ∞
2
Short-circuited line:
1
ZL = 0
N AN N AN N
0
2
Zo = 100 Ω; ZL = 200 Ω; VSWR = 2
Resistive load:
1 RL > Zo
AN N AN N AN
0
1 0.5 0
← Distance from load (wavelengths)

Figure 5.9 Standing wave pattern for various line terminations and VSWR. Locations of nodes (N) and antinodes (AN) are
identified on each plot.

from the load) will then increase as signal frequency departs from f c up or down. So, the operational bandwidth
of an antenna is typically specified as the range of frequencies over which VSWR remains below a certain level.
It is straightforward to manipulate Eq. (5.43) to express the magnitude of reflection coefficient in terms of
VSWR as
VSWR − 1
|𝜌v | = (5.44)
VSWR + 1
Figure 5.9 shows the standing wave pattern on a transmission line for various line terminations and VSWR
values. Plots of the resultant amplitude of the voltage signal at various points on the line starting from the load
and covering up to one wavelength from the load are shown. For example, an open line has ZL = ∞ so Eqs. (5.31)
and (5.43) yield
ZL − Zo Z 1 + |𝜌v | 1 + 1
𝜌v = = L = 1∠0∘ ; VSWR = = =∞
ZL + Zo ZL 1 − |𝜌v | 1 − 1
So, with 𝜃 v = 0, it follows from Eq. (5.41) that in an open-circuit the first antinode (for which n = 0) is located
at distance xa = 0 from the load (and thus at the load), and the first node is located at one-quarter of a wavelength
Telegram: @Persian_IEEEComSoc

from the load. Notice also how the standing wave ripple reduces from being twice the incident wave amplitude
when VSWR = ∞ (for open-circuit and short-circuit conditions) to zero when VSWR = 1.0 (for a matched line
condition).

5.3.6 Line Impedance and Admittance

The line impedance Z x at a distance x from load Z L (looking towards the load as indicated in Figure 5.10) on a
transmission line of characteristic impedance Z o is the ratio between the voltage v measured across the line pair at
x and the current i flowing in the line at the same point. Using the expressions for v and i in Eqs. (5.12) and (5.13)
 348 5 Transmission Media

Zo cable ZL
Zx

Figure 5.10 Line impedance Z x .

√
with K1 = A1 e j𝜔t , K2 = A2 e j𝜔t , Zo = Z∕Y , A2 = 𝜌v A1 as earlier established leads to
v e𝛾x
K1 + K2 e−𝛾x
Zx = =
i K1 𝛾x K
√ e − √ 2 e−𝛾x
Z∕Y Z∕Y
j𝜔t 𝛾x
A1 e e + A2 e ej𝜔t −𝛾x
=
A1 e j𝜔t 𝛾x A2 e j𝜔t −𝛾x
e − e
Zo Zo
e𝛾x + 𝜌v e−𝛾x
= Zo 𝛾x
e − 𝜌v e−𝛾x
Using Eq. (5.31) for 𝜌v
⎡ 𝛾x ZL − Zo −𝛾x ⎤
⎢e + e ⎥ [ ]
ZL + Zo ZL e𝛾x + Zo e𝛾x + ZL e−𝛾x − Zo e−𝛾x
Zx = ⎢ ⎥Z = Z
⎢ 𝛾x ZL − Zo −𝛾x ⎥ o ZL e𝛾x + Zo e𝛾x − ZL e−𝛾x + Zo e−𝛾x o
⎢e − e ⎥
⎣ ZL + Zo ⎦
[ ] ⎡ Z + Z e𝛾x − e−𝛾x ⎤
ZL (e𝛾x + e−𝛾x ) + Zo (e𝛾x − e−𝛾x ) ⎢ L o 𝛾x
e + e−𝛾x ⎥
=
Zo (e𝛾x + e−𝛾x ) + ZL (e𝛾x − e−𝛾x )
Zo = ⎢ 𝛾x −𝛾x ⎥ Zo
⎢ Zo + ZL e − e ⎥
⎣ e𝛾x + e−𝛾x ⎦
The hyperbolic functions sinh(x), cosh(x), and tanh(x) are defined as follows
ex − e−x
sinh(x) = = −j sin(jx)
2
ex + e−x
cosh(x) = = cos(jx)
2
sinh(x) ex − e−x
tanh(x) = = x = −j tan(jx) (5.45)
cosh(x) e + e−x
Telegram: @Persian_IEEEComSoc

Introducing the hyperbolic tangent allows us to express Z x compactly in terms of load impedance and the line’s
characteristic impedance and propagation constant as
[ ]
ZL + Zo tanh(𝛾x)
Zx = Z (5.46)
Zo + ZL tanh(𝛾x) o
On a loss-free line (𝛼 = 0), propagation constant 𝛾 = 𝛼 + j𝛽 = j𝛽, so
e j𝛽x − e−j𝛽x 2j sin(𝛽x)
tanh(𝛾x) = = = j tan(𝛽x)
e j𝛽x + e−j𝛽x 2 cos(𝛽x)
and Eq. (5.46) reduces to
[ ]
ZL + jZ o tan(𝛽x)
Zx = Z (5.47)
Zo + jZ L tan(𝛽x) o
 5.3 Transmission Line Theory 349

There are three special cases of this equation that find extensive applications in impedance matching and RF
system design. First, if the line is short-circuited (i.e. Z L = 0), then the line impedance Z x reduces at every point
to a pure reactance, and hence the line admittance Y x = 1/Z x is a pure susceptance. These are given by

Zx = jZ o tan(𝛽x) ⎫
⎪ Short-circuit stub
Yx = −jY o cot(𝛽x)⎬ , (5.48)
(Yo = 1∕Zo ; yx = Yx ∕Yo )
yx = −j cot(𝛽x) ⎪ ⎭
Second, if the line is open-circuited (i.e. Z L = ∞), the line impedance also reduces to a pure reactance (and the
line admittance to a pure susceptance) given by

Zx = −jZ o cot(𝛽x)⎫
⎪
Yx = jY o tan(𝛽x) ⎬ , Open-circuit stub (5.49)
yx = j tan(𝛽x) ⎪
⎭
These behaviours provide a means of creating pure inductors and capacitors at higher frequencies by using
a short piece of short- or open-circuited transmission line known as a resonant line section or tuning stub. A
short-circuit stub is usually preferred because of fringing effects in open-circuit stubs.
Third, when x = 𝜆/4, the term 𝛽x equals 𝜋/2. This means that the second term in the denominator and numerator
of Eq. (5.47) dominates since tan(𝛽x) → ∞ at this point. Therefore, the input impedance of a loss-free quarter-wave
line of characteristic impedance Z o terminated with load Z L is obtained as
[ ] [ ]
ZL + jZ o tan(𝛽𝜆∕4) ZL + jZ o tan(𝜋∕2)
Z𝜆∕4 = Zo = Z
Zo + jZ L tan(𝛽𝜆∕4) Zo + jZ L tan(𝜋∕2) o
[ ]
jZ o tan(𝜋∕2) Z
= Zo = o Zo
jZ L tan(𝜋∕2) ZL
= Zo2 ∕ZL (5.50)
Before leaving this topic, it is helpful to introduce admittance parameters, namely characteristic admittance
Y o = 1/Z o , load admittance Y L = 1/Z L , and line admittance Y x = 1/Z x . Also, we divide impedances by Z o and admit-
tances by Y o to obtain normalised parameters (denoted using lowercase letters) such as normalised characteristic
impedance zo = Z o /Z o = 1, normalised characteristic admittance yo = Y o /Y o = 1, normalised load admittance
yL = Y L /Y o = Z o /Z L , and normalised line admittance yx = Y x /Y o = Z o /Z x . We may then manipulate Eq. (5.47) to
obtain an expression for normalised line admittance yx at distance x from normalised load yL as
{ }
Zo Zo + jZ L tan(𝛽x) Z + jZ L tan(𝛽x)
yx = = Zo = o
Zx Zo [ZL + jZ o tan(𝛽x)] ZL + jZ o tan(𝛽x)
Z ∕Z + j tan(𝛽x)
= o L
1 + jZo ∕ZL tan(𝛽x)
Telegram: @Persian_IEEEComSoc

y + j tan(𝛽x)
= L (5.51)
1 + jyL tan(𝛽x)

Worked Example 5.2 Transmission Line Analysis 1

A loss-free 100 Ω transmission line is terminated in a load 80 + j60 Ω. We wish to analyse various aspects of the
performance of the line.
(a) Determine the voltage reflection coefficient.
(b) Determine the voltage standing wave ratio.
(c) Determine impedance at a distance 0.3𝜆 from load.
(d) Determine impedance at a distance 𝜆/4 from load.
 350 5 Transmission Media

(e) Discuss the variation of impedance with distance from load and identify all salient features along with a com-
parison with the line’s standing wave pattern.
A 100 Ω sinusoidal signal generator of frequency 25 MHz which on open circuit gives 50 V rms is now connected
to the input of a 2 m length of this line terminated by the above load.
(f) Sketch a phasor diagram showing the incident, reflected and resultant rms voltages at the load.
(g) Calculate the average power consumed by the load.
(h) Determine the maximum and minimum rms voltages on the line.
(i) Determine the impedance presented to the generator at the line input.
(a) Voltage reflection coefficient
Z − Zo 80 + j60 − 100 −20 + j60 63.246∠108.435∘
𝜌v = L = = =
ZL + Zo 80 + j60 + 100 180 + j60 189.737∠18.435∘
1
= ∠90∘
3
(b) Voltage standing wave ratio
1 + |𝜌v | 1 + 1∕3 4∕3
VSWR = = = =2
1 − |𝜌v | 1 − 1∕3 2∕3
(c) At distance x = 0.3𝜆 from load, 𝛽x = 2𝜋/𝜆 × 0.3𝜆 = 0.6𝜋. Impedance Z x at this point is obtained using Eq. (5.47)
since line is loss-free
80 + j60 + j100 × tan(0.6𝜋)
Zx |x=0.3𝜆 ≡ Z0.3𝜆 = × 100
100 + j(80 + j60) × tan(0.6𝜋)
80 − j247.77 260.364∠−72.11∘
= × 100 = × 100
284.661 − j246.215 376.369∠−40.858∘
= 69.18∠−31.25∘ Ω
= 59.14 − j35.89 Ω
(d) Impedance at a distance 𝜆/4 from load is determined using Eq. (5.50)
Zo2 1002 104 (80 − j60) 104 (80 − j60)
Z𝜆∕4 = = = =
ZL 80 + j60 (80 + j60)(80 − j60) 802 + 602
= 80 − j60 Ω
= 100∠−36.87∘ Ω
(e) A full discussion requires plots of resultant amplitude and line impedance versus distance from load as shown
in Figure 5.11. We note the following features and trends, which apply in general to all transmission lines.
Maximum impedance Z max occurs at antinodes and is purely resistive with a value given by
Telegram: @Persian_IEEEComSoc

Zmax = Zo × VSWR (5.52)

(i) Minimum impedance Z min occurs at nodes and is also purely resistive with a value given by
Zmin = Zo ∕VSWR (5.53)
(ii) Looking at line sections between adjacent antinode and node, the imaginary part of impedance, called
reactance, alternates between positive and negative as one goes from one section to the next. The line is
inductive when reactance is positive and is capacitive when reactance is negative. The reactance is zero at
nodes and antinodes.
(iii) The real part of impedance, called resistance, is always positive, having a minimum value of Z o /VSWR at
nodes and a maximum value of Z o × VSWR at antinodes.
 5.3 Transmission Line Theory 351

1.5A1

Resultant Amplitude
A1

0.5A1
2λ 7λ/4 3λ/2 5λ/4 λ 3λ/4 λ/2 λ/4 0
200
nce
Im peda 150

Impedance (Ω)
100

50
Resistance
x 0

Reactan –50
ce
–100
2λ 7λ/4 3λ/2 5λ/4 λ 3λ/4 λ/2 λ/4 0
← Distance from load

Figure 5.11 Worked Example 5.2e: Standing wave pattern and line impedance on transmission line with Z o = 100 Ω,
Z L = 80 + j60 Ω, and VSWR = 2; Reflection-induced phase shift = 90∘ .

(iv) The standing wave pattern and line impedance are periodic along the line, having a period of half a wave-
length (𝜆/2). This of course assumes a loss-free line. If there are losses then the standing wave pattern would
be damped down as the signal propagates down the line due to an exponentially decaying amplitude.
(f) At the instant of connection, the signal is not yet reflected; therefore, the source sees impedance Z o . The
scenario is as illustrated in Figure 5.12a. The rms voltage V i applied to the line, which travels down the loss-free
line as the incident voltage, is obtained by voltage division between Rs and Z o as
Zo 100
Vi = V = × 50 = 25 V rms
Rs + Zo s 100 + 100
A phasor diagram depicts the relative phases and amplitudes of sinusoidal signals of the same frequency that
are summed to obtain a resultant signal. One of the signals must be chosen as reference and its phase arbitrarily
set to zero. On transmission lines, the reference is usually the incident voltage. Thus, denoting the incident
rms voltage, reflected rms voltage, and resultant rms voltage at load as V i , V r , and V L respectively, we have
Telegram: @Persian_IEEEComSoc

V = 25∠0∘ V rms
i
1
Vr = 𝜌v Vi = ∠90∘ × 25∠0∘ = 8.33∠90∘ V rms
3
VL = Vi + Vr = 25∠0∘ + 8.33∠90∘
= 25 + j8.33 = 26.35∠18.4∘ V rms
Note that V L is the voltage delivered to the load, which is also given by Eq. (5.33) in terms of the voltage
transmission coefficient and incident voltage as
( )
1
VL = 𝜏v Vi = (1 + 𝜌v )Vi = 1 + ∠90∘ × 25∠0∘ = 26.35∠18.4∘ V rms
3
The phasor diagram depicting V i , V r , and V L is shown in Figure 5.12b.
 352 5 Transmission Media

Rs = 100 Ω
(a) Vs = 50 V rms Vi Zo = 100 Ω

rms
°V
18.4
26.35∠
VL = Vr = 8.33∠90° V rms
(b)
18.4°
Vi = 25∠0° V rms

Figure 5.12 Worked Example 5.2: (a) Line scenario at instant of connection; (b) Phasor diagram.

(g) You may wish to refer to Section 3.5.2 for a discussion of active or average power consumed in a load based on
Figure 3.17 if you are new to the concepts of reactive and active powers. With ZL = 80 + j60 = 100∠36.87∘ ≡
|ZL |∠ZL , average power consumed by the load is
Vl2 rms 26.352
PL = cos(∠ZL ) = cos(36.87∘ )
|ZL | 100
= 5.56 W
(h) Maximum and minimum rms voltages Vrmsmax and Vrmsmin , respectively, occur at an antinode and a node and
are given by
Vrmsmax = Vi rms (1 + |𝜌v |) = 25(1 + 1∕3) = 33.33 V rms
Vrmsmin = Vi rms (1 − |𝜌v |) = 25(1 − 1∕3) = 16.67 V rms
(i) Since no information is provided, we assume a velocity ratio of 1 on the line, which means that signals prop-
agate at the speed of light. The wavelength of the signal is therefore 𝜆 = v/f = 3 × 108 /25 × 106 = 12 m. Thus, a
2 m length of the line is 𝜆/6 in terms of wavelength and what is required is to use Eq. (5.47) to calculate line
impedance Z𝜆/6
[ ] [ ]
ZL + jZ o tan(𝛽𝜆∕6) ZL + jZ o tan(𝜋∕3)
Z𝜆∕6 = Zo = Z
Zo + jZ L tan(𝛽𝜆∕6) Zo + jZ L tan(𝜋∕3) o
80 + j60 + j100 × tan(𝜋∕3) 80 + j233.21
= × 100 = × 100
100 + j(80 + j60) × tan(𝜋∕3) −3.923 + j138.56
Telegram: @Persian_IEEEComSoc

= 166.53 − j62.45 Ω
= 177.86∠−20.56∘ Ω

Worked Example 5.3 Transmission Line Analysis 2

Measurements at 100 MHz show that a loss-free 75 Ω transmission line with air dielectric has a voltage maximum
25 V, and a voltage minimum 12.5 V (located 45 cm from the load). Determine.
(a) Operating wavelength.
(b) Voltage reflection coefficient.
(c) Load impedance.
 5.3 Transmission Line Theory 353

(d) Maximum line impedance.

(e) Return loss.
(f) Mismatch loss.
(g) Power transfer efficiency.
(h) Shortest distance between points of voltage minimum and maximum.
√
(a) Air dielectric has relative permittivity 𝜀r = 1, which gives propagation speed on the line v = c∕ 𝜀r = 3 ×
108 m∕s, and hence operating wavelength
v 3 × 108
𝜆= = =3m
f 100 × 106
(b) To determine the voltage reflection coefficient 𝜌v , we first calculate VSWR using the Amax and Amin information
provided then the magnitude of 𝜌v using VSWR, and finally the angle of 𝜌v using the information given on the
location of the voltage minimum
A 25 VSWR − 1 2 − 1 1
VSWR = max = = 2. Thus, |𝜌v | = = =
Amin 12.5 VSWR + 1 2 + 1 3
From Eq. (5.41), voltage minima occur at
( )
1 𝜃 𝜆
xd = n + + v
2 2𝜋 2
and the first such minimum corresponds to n = 0 and occurs at location
( )
1 𝜃v 𝜆
xdmin = + = 45 cm (given)
2 2𝜋 2
Thus, phase angle of 𝜌v is
𝜃v = 4𝜋(xmin ∕𝜆 − 1∕4) = 720∘ (45∕300 − 0.25) = −72∘
Hence
1
∠−72∘ = 0.103 − j0.317
𝜌v = |𝜌v |∠𝜃v =
3
(c) Load impedance is determined from Z o and 𝜌v as follows using the equation just before Eq. (5.31)
1 + 𝜌v 1.103 − j0.317
ZL = Z = × 75
1 − 𝜌v o 0.897 + j0.317
82.73 − j23.78 86.07∠−16.04∘
= =
0.897 + j0.317 0.9514∠19.46∘
∘
= 90.47∠−35.5 Ω = 73.66 − j52.54 Ω
(d) Maximum line impedance Z max = Z o × VSWR = 75 × 2 = 150 Ω.
(e) Return loss, RL = −20 log10 (|𝜌v |) = −20 log10 (1∕3) = 9.54 dB
Telegram: @Persian_IEEEComSoc

(f) Mismatch loss, ML = −10 log10 (1 − |𝜌v |2 ) = −10 log10 (8∕9) = 0.51 dB
(g) Power transfer efficiency
𝜂 = (1 − |𝜌v |2 ) × 100% = (8∕9) × 100% = 88.9%
(h) Shortest distance between points of voltage minimum and maximum is the separation between a node and
an adjacent antinode, which is 𝜆/4 = 75 cm.

5.3.7 Line Termination and Impedance Matching

We have seen that, whenever a transmission line of characteristic impedance Z o is terminated in a load Z L (as
shown in Figure 5.10), a fraction 𝜌v = (Z L − Z o )/(Z L + Z o ) of the voltage signal is reflected back. The load Z L
 354 5 Transmission Media

could be another cable. For example, TV aerials use 75 Ω coaxial cable, whereas LANs use 50 Ω coax. Reflec-
tions represent a waste of some of the power intended for the load. Reflections can also give rise to unpleasant
echo (due to multiple reflections from both ends of the line), distortion, instability at the source, or damage to
sensitive system components. In extreme cases of high-power connection to a transmit-antenna, reflections could
even lead to electrical breakdown at points of voltage maximum on the line. The following three special cases of
line termination have consequences that are readily analysed.
Matched circuit, Z L = Z o : the line is said to be terminated with a matched load (i.e. a load matched to the
characteristic impedance of the line). In that case 𝜌v = 0, 𝜌i = 0 and there is no reflection and hence no standing
waves along the line. The incident voltage is delivered in its entirety to the matched load. Assuming a loss-free line,
Eq. (5.47) shows that the line impedance Z x at every point along the line equals the characteristic impedance Z o
of the line, and Eq. (5.42) shows that wave amplitude is the same at every point of the line and equals the incident
amplitude A1 .
Open circuit, Z L = ∞: in an open circuit, 𝜌v = 1, 𝜌i = −1. On switching on the source, a current pulse i1 travels
down the line, reaching the load after a time l/vp whence a reflected current pulse i2 = i1 is produced travelling
in the opposite direction. The resultant current at every point on the line reached by i2 is i = i1 − i2 = 0. After
a transient time 2l/vp this reflected current i2 reaches the source end so that current is zero everywhere on the
line. Considering the steady-state open-circuit condition along the line, Eq. (5.42) indicates that maximum volt-
age amplitude Amax = 2A1 and minimum voltage amplitude Amin = 0 at antinodes and nodes, respectively. With
reflection-induced phase shift 𝜃 v = 0, Eq. (5.41) gives antinode locations at x = 0, 𝜆/2, 𝜆, …; and node locations at
𝜆/4, 3𝜆/4, … Since line impedance under open-circuit condition follows from Eq. (5.47) as
[ ] [ ]
ZL + jZ o tan(𝛽x) ZL
Zx |ZL =∞ ≡ Zoc = Z = Z
Zo + jZ L tan(𝛽x) o jZ L tan(𝛽x) o
Zo
= (5.54)
j tan(𝛽x)
it follows that at antinodes (x = 0, 𝜆/2, 𝜆, …) where voltage is maximum, impedance is infinite (since tan𝛽x = 0)
and therefore current is zero. At nodes (x = 𝜆/4, 3𝜆/4, …), we have tan(𝛽x) = tan((2𝜋∕𝜆) × (𝜆∕4)) = tan(𝜋∕2) → ∞,
etc., which means that impedance → 0; but voltage is zero, so current is also zero. Therefore steady-state current
is zero in an open-circuit line.
Short circuit, Z L = 0: in a short circuit, 𝜌v = −1, 𝜌i = 1. On switching on the source, a current pulse i1 travels
down the line, reaching the load after a time l/vp whence a reflected current pulse i2 = −i1 is produced travelling
in the opposite direction. The resultant current at every point on the line reached by i2 is i = i1 − i2 = 2i1 , and this
grows indefinitely after time 2l/vp as further reflections add to the current. Line impedance under short-circuit
condition follows from Eq. (5.47) as
[ ]
jZ o tan(𝛽x)
Zx |ZL =0 ≡ Zsc = Zo
Zo
Telegram: @Persian_IEEEComSoc

= jZ o tan(𝛽x) (5.55)

Thus, from Eqs. (5.54) and (5.55), the product of open-circuit line impedance and short-circuit line impedance
equals the square of the line’s characteristic impedance

Zoc Zsc = Zo2 (5.56)

To avoid reflections due to a mismatch, proper termination must be used with Z L = Z o . A range of methods
exist to match two cables of different characteristic impedances, or to match a transmission line of characteristic
impedance Z o with an electrical load or system having input impedance Z L ≠ Z o . We briefly introduce the methods
of quarter-wave transformer, attenuation pad, and single-stub matching, and present a detailed design example on
the last.
 5.3 Transmission Line Theory 355

λ/4

(a) R

Zo
Zo1

Zin = Z2o1/R = Zo

λ/4 d1

(b) ZL

Zo
Zo1 Zo2

Zin = Z2o1/R = Zo R

Figure 5.13 Quarter-wave transformer: (a) Matching resistor R to Z o ; (b) Matching impedance Z L to Z o .

Quarter-wave transformer: Eq. (5.50) indicates that the impedance looking into the input terminals of an
arrangement or subsystem consisting of a quarter-wavelength long transmission line of characteristic impedance
Z o1 that is terminated by a resistive load R (as illustrated in Figure 5.13a) is
2
Zo1
Zin = (5.57)
R
It means that when this arrangement is connected to a transmission line of characteristic impedance Z o , there
will be impedance matching if
2
Zo1
Zo = Zin =
R
or
√
Zo1 = Zo R (5.58)
This relationship reveals one way to match a resistive load R to a transmission line (of characteristic impedance)
Z o : we connect R to the line through a 𝜆/4-length of a cable of characteristic impedance Z o1 selected to satisfy
Telegram: @Persian_IEEEComSoc

Eq. (5.58). The desired value of Z o1 is achieved using the physical dimensions a and b of the 𝜆/4-length interface
cable and the relative permittivity 𝜀r of the cable’s dielectric medium, as defined in Figure 5.1. Once the correct
values of a, b and 𝜀r have been set, Eq. (5.58) will be satisfied at all frequencies since Z o and R are practically
frequency independent over a wide range of frequencies. However, a perfect match is only achieved while the
interface cable is of length exactly 𝜆/4, and for a physical cable length l this will be the case at one frequency f c
given by
c
fc = √ (5.59)
4l 𝜀r
Here, 𝜀r is the relative permittivity of the dielectric medium between the inner and outer conductors of the
interface cable and c = 3 × 108 m/s is the speed of light. For example, for 𝜀r = 1, a cable length l = 150 cm is a
 356 5 Transmission Media

quarter wavelength only at f c = 50 MHz. When one departs from this frequency, up or down, the specification
will no longer be satisfied, and a mismatch will develop that leads to VSWR > 1, which increases as the difference
between signal frequency and f c increases.
The quarter-wave transformer described above transforms any real impedance R to another real impedance
Z o = Z in given by Eq. (5.57). Since a line terminated by a general impedance Z L is purely resistive at its nodes and
antinodes (see Figure 5.11), it is possible to apply the quarter-wave transformer method to transform a non-real
load impedance Z L to a real impedance Z o in two stages as shown in Figure 5.13b. First connect Z L to length d1 of
a cable of characteristic impedance Z o2 . The impedance seen looking into the input of this Z o2 -cable has a purely
resistive value R determined by whether d1 is a nodal or anti-nodal distance according to the relations
{
Zo2 × VSWR, d1 = (𝜃v ∕𝜋)𝜆∕4
R=
Zo2 ∕VSWR, d1 = (1 + 𝜃v ∕𝜋)𝜆∕4
ZL − Zo2 1 + |𝜌v |
≡ |𝜌v |∠𝜃v ; VSWR = (5.60)
ZL + Zo2 1 − |𝜌v |
With Z L transformed into R in this way, connecting this Z L -terminated Z o2 -cable of length d1 to the quarter-wave
transformer as discussed above achieves the desired match to Z o .
Attenuation pads: an attenuation pad may be inserted to match one cable of, say, 75 Ω to another cable or
device of a different impedance such as 50 Ω. Figure 5.14 shows three different arrangements. The minimum loss
pad uses two resistors whereas the T-network and 𝜋-network pads make use of three resistors. Besides impedance
matching, attenuation pads also serve an important purpose of introducing a controlled amount of attenuation into
the line, which, for example, helps to prevent sustained multiple reflections in the event of a mismatch at both ends
or unwanted discontinuities in the line, especially at joints. The correct design values have been inserted in the
minimum loss pad to match a 75 Ω cable to a 50 Ω cable and vice versa. To see that there is impedance matching in
both directions, consider that the 75 Ω cable sees a 43.3 Ω resistance connected in series with a parallel connection

Minimum loss pad

43.3 Ω
75 Ω 75 Ω Cable 86.6 Ω 50 Ω Cable 50 Ω

T-Network

R1 R3
Telegram: @Persian_IEEEComSoc

Ro1 Ro1 Cable R2 Ro2 Cable Ro2

π-Network

R2
Ro1 Ro1 Cable R1 R3 Ro2 Cable Ro2

Figure 5.14 Attenuation pads.

 5.3 Transmission Line Theory 357

of 86.6 Ω and 50 Ω resistances. Thus, the resistance seen by the 75 Ω cable looking into the minimum loss pad
input is
86.6 × 50
43.3 + = 43.3 + 31.7 = 75 Ω
86.6 + 50
So, the minimum loss pad with the indicated resistance values does successfully match a 75 Ω cable to a 50 Ω
cable. Now consider the matching achieved for signal flowing in the opposite direction. Looking back into the
minimum loss pad, the 50 Ω cable also sees a 50 Ω impedance since it sees a parallel connection of 86.6 Ω with a
series connection of 43.3 Ω and 75 Ω, which is equivalent to
86.6 × (43.3 + 75)
= 50 Ω
86.6 + 43.3 + 75
Single-stub matching: the stub tuner technique matches a load Z L to a transmission line of characteristic
impedance Z o by connecting length d2 of a short-circuited Z o -line (called a stub) in parallel at distance d1 from
the load, as shown in Figure 5.15. The stub is connected to the line at SS′ . The design task is to calculate d2 and
d1 . Recalling that the line admittance of a stub is given by Eq. (5.48), the idea behind single-stub matching is that
d1 is the shortest distance from the load at which the real part of line admittance Y x equals line conductance Y o
(= 1/Z o ), i.e. Y x = Y o + jB, and d2 is the length of the stub with input admittance –jB. Since the total admittance
of two circuit elements that are connected in parallel is the sum of their individual admittances, summing the
admittances of the stub and the line at SS′ gives a total admittance Y o (and hence total impedance Z o ) at SS′ ,
which achieves the required matching of Z L to the line. Expressing in terms of normalised admittance, yx = 1 + jb
and d2 is the length of the stub with normalised input admittance –jb. In the next worked example, we show that,
to match a load of normalised admittance yL = a − jh, the distance d1 is the smallest positive value in the set
{ ( √ ) }
1 −h ± a[h2 + (a − 1)2 ] n
d1 = tan −1
+ 𝜆, n = 0, 1, 2, · · · (5.61)
2𝜋 h2 + a(a − 1) 2
This value of d1 is then employed to calculate b as
h tan2 𝛽d1 + (1 − h2 − a2 ) tan 𝛽d1 − h
b= (5.62)
(1 + h tan 𝛽d1 )2 + a2 tan2 𝛽d1
And finally, the design is completed by obtaining distance d2 as the smallest positive value in the set
{ −1 }
tan (1∕b) n
d2 = + 𝜆, n = 0, 1, 2, · · · (5.63)
2𝜋 2
The quickest way to obtain the values of d1 and d2 needed for single-stub matching is to implement the above
three equations in a short computer code using a range of software platforms, such as MATLAB. However, the
Smith Chart (briefly introduced later) may also be used to carry out a manual design.

Zo d1
Telegram: @Persian_IEEEComSoc

Zo-line YL = 1/ZL

Sʹ

d2 Stub

Figure 5.15 Single-stub matching.

 358 5 Transmission Media

Worked Example 5.4 Single-stub Matching

We wish to match a load Z L = 30 + j25 Ω to a loss-free 50 Ω transmission line using the stub tuning technique.
The design solution requires us to calculate the minimum distances d1 and d2 in Figure 5.16a so that there is no
reflection at point S on the line.
Figure 5.16b shows the normalised admittances that we wish to achieve. The solution is in two steps. First, we
use Eq. (5.51) to determine the minimum distance d1 at which the real part of the normalised line admittance
is 1, i.e. yx = 1 + jb, where jb is the normalised susceptance of the line at that point. Second, we use Eq. (5.48) to
calculate length d2 of the stub that has normalised input admittance equal to −jb. When this stub is connected in
parallel to the line at SS′ , the total normalised admittance yx = 1 (which means that Y x = Y o and Z x = Z o , i.e. total
impedance at SS′ equals Z o ), so the required matching of Z L to the Z o line has been accomplished.
Setting yx in Eq. (5.51) equal to 1 + jb at x = d1 , we have
yL + j tan 𝛽d1
yd1 = = 1 + jb
1 + jyL tan 𝛽d1
Given Z o = 50 and Z L = 30 + j25, normalised load yL (which we will denote as a − jh for convenience) in the
above equation is
1 Z 50 50(30 − j25)
yL = = o = =
zL ZL 30 + j25 (30 + j25)(30 − j25)
1500 − j1250 1500 1250 60 50
= 2 2
= −j = −j
30 + 25 1525 1525 61 61
≡ a − jh; a = 60∕61, h = 50∕61

d1
S

(a) Zo = Ro = 50 Ω ZL = 30 + j25 Ω
Sʹ

d2 Stub

1 + jb
Telegram: @Persian_IEEEComSoc

y = 1.0 d1

(b) yL = 1/zL

Sʹ

–jb

d2
Stub

Figure 5.16 Worked Example 5.4.

 5.3 Transmission Line Theory 359

Thus
a − jh + j tan 𝛽d1 a + j(tan 𝛽d1 − h)
1 + jb = =
1 + j(a − jh) tan 𝛽d1 1 + h tan 𝛽d1 + ja tan 𝛽d1
a + atan2 𝛽d1 htan2 𝛽d1 + (1 − h2 − a2 ) tan 𝛽d1 − h
= + j
(1 + h tan 𝛽d1 )2 + a2 tan2 𝛽d1 (1 + h tan 𝛽d1 )2 + a2 tan2 𝛽d1
The real part being equal to 1 and imaginary part equal to b means that
a + atan2 𝛽d1 = (1 + h tan 𝛽d1 )2 + a2 tan2 𝛽d1
⇒ (h2 + a2 − a)tan2 𝛽d1 + 2h tan 𝛽d1 + (1 − a) = 0 (5.64)
and
h tan2 𝛽d1 + (1 − h2 − a2 ) tan 𝛽d1 − h
b= (5.65)
(1 + h tan 𝛽d1 )2 + a2 tan2 𝛽d1
Equation (5.64) is a quadratic equation in tan 𝛽d1 , and therefore yields
√
−h ± a[h2 + (a − 1)2 ]
tan 𝛽d1 =
h2 + a2 − a
= −0.01, −2.49 (Putting a = 60∕61, h = 50∕61)
Taking the inverse tangent of both sides, recalling that 𝛽 = 2𝜋/𝜆 and that tan(𝜃) = tan(𝜃 + n𝜋), n = 0, 1, 2, …,
we obtain
𝜆
d1 = {tan−1 (−0.01, −2.49) + n𝜋} , n = 0, 1, 2, · · ·
2𝜋
𝜆
= {(−0.01, −1.189) + n𝜋}
2𝜋
{ } { }
−0.01 n −1.189 n
= + 𝜆 or + 𝜆
2𝜋 2 2𝜋 2
= {−0.002, 0.4984, 0.9984, · · ·}𝜆 or {−0.1892, 0.3108, 0.8108, · · ·}𝜆
= 0.3108 𝜆(choosing the smallest positive result)
We are now able to determine b in Eq. (5.65) by using this value of d1 (and hence 𝛽d1 = 2𝜋/𝜆 × 0.3108𝜆 = 1.9527)
along with h = 50/61 and a = 60/61 to obtain b = 0.82664. This means that the normalised admittance at point SS′
on the line is 1 + jb. An admittance of value −jb must therefore be connected in parallel at this point in order to
make total normalised admittance equal to 1 + jb − jb = 1, which establishes a match with the line. Since we know
from Eq. (5.48) that a stub of length x has admittance −jcot(𝛽x) = −j/tan(𝛽x), we determine stub length d2 as the
value of x at which −j/tan(𝛽x) = −jb or tan(𝛽x) = 1/b. Thus
𝛽d2 = tan−1 (1∕b) + n𝜋 = tan−1 (1∕0.82664) + n𝜋 = 0.88002 + n𝜋
𝜆
d2 = (0.88002 + n𝜋)∕𝛽 = (0.88002 + n𝜋) , n = 0, 1, 2, · · ·
Telegram: @Persian_IEEEComSoc

2𝜋
{ }
0.88002 n
= + 𝜆 = {0.1401, 0.6401, 1.1401, · · ·}𝜆
2𝜋 2
= 0.1401𝜆, (Choosing smallest positive value)
In conclusion, the load Z L = 30 + j25 Ω may be matched to a 50 Ω line by connecting in parallel to the line a
short-circuit stub of length 0.1401𝜆 at a distance 0.3108𝜆 from the load.

5.3.8 Scattering Parameters

Scattering parameters or S-parameters are used for characterising microwave networks, and their definitions are
based on 50 Ω termination of network ports. They are readily measured in the laboratory using network analysers.
Figure 5.17 shows a two-port network with incident and reflected voltages propagating in opposite directions in
each port of the network. The voltages are identified using two subscripts, which are either digit 1 or 2. The first
 360 5 Transmission Media

1 2

Zg ʋ11 ʋ21
Two-port ZL
ʋg ʋ12 Network ʋ22

1ʹ 2ʹ
Z1 Z2

Figure 5.17 S-parameters.

subscript identifies the port (using 1 for input port and 2 for output port), whereas the second subscript identifies
the wave (using 1 for incident wave and 2 for reflected wave). It is important to note that incidence and reflection
are relative to the ports and not the load or source. Thus, v21 is the wave incident on port 2, whereas v12 is the wave
reflected from port 1.
The reflected voltages v12 and v22 are given in terms of the incident voltages v11 and v21 through the S-parameters

v12 = s11 v11 + s12 v21

v22 = s21 v11 + s22 v21 (5.66)

This indicates that the reflected voltage v12 observed at port 1 comes from two contributions, namely a reflection
of the voltage v11 that is incident at port 1 and a transmission through the network of the voltage v21 that is incident
on port 2. Similarly, the reflected voltage v22 observed at port 2 is the result of reflection of v21 at port 2 and trans-
mission of v11 through the network from port 1 to port 2. If s12 = 0 then the signal v21 incident on port 2 does not
make any contribution to the signal v12 propagating away from port 1. This means that signals propagate through
the network in only one direction from port 1 to port 2. The network is therefore said to be unilateral if s12 = 0.
Since v21 is the signal coming back (i.e. reflected) from load Z L connected at port 2, it follows that under condi-
tions of matched load at port 2 (i.e. Z L = Z 2 ) the signal v21 = 0 and we obtain from Eq. (5.66)
v12 ||
s11 =
v11 ||v21 =0
Matched load at port 2 (5.67)
v |
s21 = 22 ||
v11 |v21 =0

Furthermore, if Z g = Z 1 (i.e. matched load at port 1) and the generator vg is turned off so that signals in the
network originate from v21 only (being reflected at port 2 to produce v22 and transmitted through the network to
produce v12 ), there will be no reflection at the ‘load’ Z g and hence v11 = 0, so we have from Eq. (5.66)
Telegram: @Persian_IEEEComSoc

v22 ||
s22 =
v21 ||v11 =0
Matched load at port 1 (5.68)
v12 ||
s12 =
v21 ||v11 =0

These S-parameters therefore have the following physical meaning and applications. Parameter s11 is the voltage
reflection coefficient 𝜌v1 of the input port. Parameter s22 is the voltage reflection coefficient 𝜌v2 of the output port.
|s21 |2 is the forward insertion power gain of the two-port network, usually expressed in dB. And |s12 |2 is the reverse
insertion power gain of the two-port network, also called the reverse power leakage. It is important to note that these
parameters are always defined (and measured) under a matched condition at one of the two ports. So, v11 is the
available voltage from a matched generator and v22 is the voltage delivered to a matched load.
 5.3 Transmission Line Theory 361

Worked Example 5.5 S-parameters

We wish to calculate the S-parameters of the T-network shown in Figure 5.18a and hence to determine the return
loss of the network at its input and the insertion loss of the network.
S-parameter definitions are based on 50 Ω source and 50 Ω terminations, so these must be adhered to as we seek
to calculate each parameter for the given network. First, s11 is the voltage reflection coefficient 𝜌v1 of the input

1 2 1 2
25 Ω 25 Ω 25 Ω 25 Ω

50 Ω 50 Ω 50 Ω

Z1

1ʹ 2ʹ 1ʹ 2ʹ
(a) (b)

1 2
25 Ω 25 Ω

50 Ω 50 Ω

Z2

1ʹ 2ʹ
(c)
1 2
25 Ω 25 Ω

50 Ω
(d) ʋa 50 Ω ʋ22 50 Ω

2ʋ11

1ʹ 2ʹ
Telegram: @Persian_IEEEComSoc

1 2
25 Ω 25 Ω

50 Ω
50 Ω ʋ12 ʋa 50 Ω
(e)
2ʋ21

1ʹ 2ʹ

Figure 5.18 Worked Example 5.5: (a) T-network; (b) Configuration for calculating s11 ; (c) Configuration for calculating s22 ;
(d) Configuration for calculating s21 ; (e) Configuration for calculating s12 .
 362 5 Transmission Media

port, so to determine s11 , we terminate output port 2 with a (matched) 50 Ω load, as shown in Figure 5.18b, and
under that configuration we determine the impedance Z 1 seen looking into port 1. Z 1 is a series connection of
25 Ω to a parallel connection between 50 Ω and a series connection of 25 Ω and 50 Ω. We adopt the convention of
using ‘+’ to represent a series connection and ‘∥’ to represent a parallel connection. Thus
50 × (25 + 50)
Z1 = 25 + (50 ∥ (25 + 50)) = 25 +
50 + (25 + 50)
= 25 + 30 = 55 Ω
Z − Zo 55 − 50 1
s11 ≡ 𝜌v1 = 1 = = ∠0∘
Z1 + Zo 55 + 50 21
Next to s22 , which is the voltage reflection coefficient of the output port 2. So, to determine s22 we terminate
input port 1 with 50 Ω, as shown in Figure 5.18c, and (under this configuration we) determine the impedance Z 2
seen looking into port 2. Thus
Z2 = 25 + (50 ∥ (25 + 50)) = 25 + 30 = 55 Ω
Z − Zo 55 − 50 1
s22 ≡ 𝜌v2 = 2 = = ∠0∘
Z2 + Zo 55 + 50 21
We notice that s11 = s22 , but this is only due to the symmetry of this network and is in no way a general relation-
ship for all networks.
Parameter s21 is the ratio between voltage v22 at port 2 and voltage v11 at port 1 when a matched load is connected
at port 2 and a matched generator is connected at port 1, as shown in Figure 5.18d. Note that the generator voltage
is 2v11 in order for it to deliver an incident voltage equal to v11 when a matched load is connected to it at port 1.
We calculate the required ratio v22 /v11 in two stages. First, va is the voltage across resistance 50 ∥ (25 + 50) when
voltage 2v11 is divided between resistance 50 + 25 and resistance 50 ∥ (25 + 50). Second, v22 is the voltage across
resistance 50 when voltage va is divided between resistance 25 and resistance 50. Thus
50 ∥ 75 30 4
va = 2v = 2v = v
(50 ∥ 75) + (50 + 25) 11 30 + 75 11 7 11
50 2 2 4 8
v22 = v = v = × v = v
25 + 50 a 3 a 3 7 11 21 11
v22 8
s21 = = ∠0∘
v11 21
Parameter s12 is defined in a similar way to s21 but with the roles of ports 1 and 2 reversed, as shown in
Figure 5.18e. It follows, as earlier explained, that
50 ∥ 75 30 4
va = 2v = 2v = v
(50 ∥ 75) + (50 + 25) 21 30 + 75 21 7 21
50 2 2 4 8
Telegram: @Persian_IEEEComSoc

v12 = v = v = × v = v
25 + 50 a 3 a 3 7 21 21 21
v 8
s12 = 12 = ∠0∘
v21 21
We observe again that s12 = s21 just for this network but not as a rule for all other networks.
The return loss (RL) of a transmission line was defined in Eq. (5.37), and is applicable to a network but with
S-parameter s11 replacing 𝜌v . Thus
RL = −20 log10 (|s11 |) = −20 log10 (1∕21) = 26.4 dB
 5.3 Transmission Line Theory 363

Note that the higher the RL value, the better is the coupling of a signal from source into the network with
minimal reflection.
Insertion loss (IL) is the reciprocal of the forward insertion power gain |s21 |2 of the network. Thus
IL = 1∕|s21 |2
= 10 log10 (1∕|s21 |2 ) dB
= −20 log10 (|s21 |) = −20 log10 (8∕21)
= 8.4 dB
Note that the lower the IL value, the better is the transfer of a signal through the network with minimal dissi-
pation.

5.3.9 Smith Chart

The Smith chart of Figure 5.19 is a graphical tool that was once very popular for solving problems in transmission
lines and impedance matching. Nowadays, such problems are more quickly solved using widely available software
tools, but the Smith chart is still widely used in radio frequency (RF) circuit analysis software and measurement
instruments such as network analysers as a display screen to present various RF parameters in a format that is
more instructive than a simple tabular layout. We only briefly introduce features of the Smith chart here. The
interested reader is referred to [1] for examples of a range of problem solving using the Smith chart.
The Smith chart is based on normalised impedances, with the centre (labelled O in Figure 5.19) representing
the characteristic impedance zo = 1.0 of the line. The centre line (labelled BD) is the real axis. Circles of constant
normalised resistance traverse the real axis and have values from 0.0 on the left (B) to ∞ on the right (D). The
resistance circle that passes through O has a value of 1.0. Curves of constant normalised reactance radiate out-
ward away from BD, starting from 0.0 along BD to ∞ for the most curved away from BD. The reactance curves
above BD are positive (or inductive reactance), whereas the ones below are negative or capacitive reactance. The
circumference of the chart consists of several circular scales. The innermost circular scale gives the angle of trans-
mission coefficient in degrees from 0∘ at D going counterclockwise to 90∘ near B for positive angles, and from 0∘ at
D going clockwise to −90∘ near B for negative angles. The next circular scale gives the angle of reflection coefficient
in degrees from 0∘ at D to 180∘ at B, going counterclockwise for positive angles and clockwise for negative angles.
The next circular scale gives the distance moved in wavelengths along the transmission line towards the load from
0.0 at B going counterclockwise to 0.25 at D and continuing to 0.5 back at B. The outermost circular scale gives
the distance in wavelengths towards the generator from 0.0 at B going clockwise to 0.25 at D and continuing to 0.5
back at C. Finally, the six bottom scales give various reflection and transmission coefficients and losses (in dB).
The Smith chart is a very versatile tool for manually solving transmission line and impedance matching prob-
lems. For example, given a line of characteristic impedance Z o = 50 Ω that is terminated by a load Z L = 30 + j25 Ω,
Telegram: @Persian_IEEEComSoc

we mark the normalised load impedance zL = (30 + j25)/50 = 0.6 + j0.5 on the Smith chart as shown in Figure 5.20.
Note that this mark is at the point of intersection between the normalised resistance circle 0.6 and the normalised
reactance curve 0.5. A circle of radius OzL centred at O as shown in Figure 5.20 is a constant VSWR circle that
passes through all possible impedance values on the transmission line.
VSWR, Z max , Z min : moving from point zL clockwise along this circle is equivalent to moving from the load
along the entire transmission line towards the source or generator. Since the maximum impedance on the line
is Z o × VSWR (≡ VSWR normalised) and is entirely resistive, it follows that the intersection of this circle on the
right-hand side of the real axis of the chart gives VSWR. Thus, VSWR = 2.24. This circle intersects the left-hand
 364 5 Transmission Media

B D
O

Figure 5.19 Smith chart.

side of the real axis of the chart at 0.44, which gives the normalised minimum impedance on the line. So, the line
has Z max = Z o × 2.24 = 112 Ω and Z min = Z o × 0.44 = 22 Ω.
Telegram: @Persian_IEEEComSoc

Reflection coefficient: drawing a straight line OzL and extending it to intersect the circumference of the chart
at F, the line OF intersects the circular scale for angle of reflection at 𝜃 v = 111.5∘ . Using a ruler or compass to
transfer the radius OzL to read from the centre of the (second from) bottom scale towards the left of the scale gives
a reading for |𝜌v | of 0.383. Thus, the reflection coefficient is 0.383∠111.5∘ . Note that the bottom scales are shown
in Figure 5.19 but not in Figure 5.20.
Line impedance: we may use the Smith chart to read the impedance at any point on the line by moving from F
clockwise along the outermost circular scale (towards the generator) through the required distance in wavelengths.
For example, to read the impedance at a distance 0.3𝜆 from the load, we move from F (= 0.0952) through 0.3 to get
to point G (= 0.0952 + 0.3 = 0.3952). To locate point G on the transmission line, we draw line OG which intersects
the constant VSWR circle (representing the transmission line) at zx = 0.64 − j0.56. This is the normalised line
impedance at a distance 0.3𝜆 from the load, so line impedance Z x = zx × 50 = 32 − j28 = 42.5∠−41.2∘ .
 5.4 Optical Fibre 365

zL

zL = 0.6 + j0.5
zx
zx x = 0.3λ
= 0.64 – j0.56
θʋ = 111.5o
VSWR = 2.24
zmax = 2.24
zmin = 0.44
VSWR – 1
ρʋ = = 0.383
VSWR + 1

Figure 5.20 Example of Smith chart use.

5.4 Optical Fibre

Due to its cost, rigid structure, and bulk, a metallic waveguide is only suitable for very short links (e.g. to connect
between an outdoor antenna and an indoor unit) at frequencies above about 2 GHz. Optical fibre, however, is
a much cheaper and more suitable form of waveguide, which may be used for long-distance links. Signals are
carried by light waves in the fibre. Optical fibre is nowadays the main transmission medium for long-distance
high-capacity terrestrial communication, including intercontinental links (e.g. transoceanic cables), interex-
change networks, metropolitan area networks (MANs), backhaul links for satellite, and terrestrial wireless
systems, fixed broadband links for homes and businesses through fibre to the cabinet (FTTC), and fibre to the
Telegram: @Persian_IEEEComSoc

premises (FTTP), etc.

Most of the key developments that firmly established optical fibre as an indispensable transmission medium in
the global telecommunications network infrastructure took place in the second half of the twentieth century and
included:

● 1954: microwave amplification by stimulated emission of radiation (Maser) and light amplification by stimu-
lated emission of radiation (laser) are demonstrated by Charles H. Townes (1915–2015, USA) in 1954 and 1958,
respectively, building on a theory first proposed in 1917 by Albert Einstein.
● 1966: in a landmark paper in IEE Proceedings [2], Charles Kao and George Hockham (UK) propose glass fibre
for optical signal transmission. They are convinced that the prohibitive glass attenuation (∼1000 dB/km) at the
time was due to absorption by impurities in the glass, and that losses less than 20 dB/km were attainable in
 366 5 Transmission Media

Figure 5.21 Optical fibre structure.

Coating

Cladding

Core

better purified glass. In 1972, within six short years, Corning Glass Works (USA) produce germanium-doped
fibre core having a loss of just 4 dB/km.
● 1974: John MacChesney and Paul O’Connor (Bell Labs, USA) develop the modified chemical vapour deposition
(MCVD) process for the manufacture of ultra-pure glass, a method that remains the standard for mass-producing
low-loss optical fibre cables.
● 1975: the first commercial continuous-wave semiconductor laser operating at room temperatures is developed
at Laser Diode Labs (USA).
● 1987: erbium doped fibre amplifier (EDFA), able to directly boost light signals, is developed by David Payne
(UK) [3]. This leads to an all-optical system by 1991 which supports 100 times more data than a system that uses
electronic amplification where light is converted to electrical and back again.
● 1988: the first transoceanic optical fibre cable, TAT-8, of length 5580 km is laid across the Atlantic Ocean to
connect the USA with the UK and France. TAT-8 had a capacity of 280 Mb/s and was the eighth transatlantic
cable but the first to use fibre, the previous ones having used copper coaxial cable.
● 1996: the Trans-Pacific Cable 5 (TPC-5), an all-optic fibre cable and the first to use optical amplifiers, is laid in
a loop across the Pacific Ocean from San Luis Obispo in California to Guam in Hawaii and Miyazaki in Japan
and back to the Oregon coast. TPC-5 has a total cable length of 22 500 km and a transmission capacity of 5 Gb/s.
An optical fibre is a dielectric waveguide about 125 μm in diameter made from high-purity silica glass.
Figure 5.21 shows the structure of optical fibre. It consists of a glass core made of nearly pure silicon dioxide
(SiO2 ) or silica surrounded by a glass cladding, which has a slightly lower refractive index than the core and is
instrumental in confining the transmitted information-bearing light signal entirely within the core. A plastic
sheath or coating made from durable resin covers the cladding to protect both the core and the cladding from
moisture and mechanical damage. A group of fibres is usually bundled together in a larger protective jacket
known as a cable, which may hold in excess of a thousand individual fibres. The fibres were originally used in
pairs, one for each transmission direction, but today the technique of wavelength division multiplexing (WDM)
enables a single fibre to support bidirectional communication using separate light wavelengths, e.g. 1310 nm
downstream and 1550 nm upstream. Light waves propagate within the inner core over long distances. The
dimension of an optical fibre is usually stated as a pair of numbers in micrometres (μm), called microns, which
gives the diameters of the core and cladding. For example, a 50/125 fibre has a core diameter of 50 μm and a
Telegram: @Persian_IEEEComSoc

cladding diameter of 125 μm.

Optical fibre offers several significant advantages as a transmission medium.
● Optical fibre conveys an optical carrier of very high frequency, around 200 000 GHz. Assuming a transmission
bandwidth of only 1% of the supported carrier frequency, we see that optical fibre offers an enormous potential
bandwidth of about 2000 GHz. This is practically a limitless bandwidth when one considers the requirements
of foreseeable broadband communication services. However, the realisable bandwidth in practical optical com-
munication systems will be constrained by the limitations of the electronic interface devices.
● Intensive research into optical fibre over the last 50 years has led to the development of optical fibre with attenua-
tion reduced from a figure of about 20 dB/km in 1970 to around 0.2 dB/km and 0.5 dB/km today within transmis-
sion windows at 1550 nm and 1310 nm, respectively. This is in sharp contrast to an attenuation figure of around
7.5 dB/km at 10 MHz for the least lossy (standardised landline) coaxial cable. The much lower attenuation
 5.4 Optical Fibre 367

offered by optical fibre and the fact that this attenuation is independent of frequency (within the said transmis-
sion windows) mean that very high data rates can be supported using repeaters that are widely spaced at around
80–100 km. This spacing is an order of magnitude better than the largest repeater spacing used in normal-core
coax to support a transmission bandwidth of only 4 MHz. This yields a significant reduction in initial and main-
tenance costs of the communication system. In fact, some long-haul communication systems and MANs may
be implemented using optical fibre without intermediate repeaters.
● The fibre carries an optical signal, which is immune to electrical interference. Optical fibre can therefore be
deployed in electrically noisy environments without the need for the expensive measures that must be adopted
for wire pairs and coax.
● The problem of crosstalk is eliminated. The optical signal in one fibre is confined entirely within that fibre. It
does not cause any direct or indirect effects on adjacent fibres in the same or other cable core.
● The fibre (an electrical insulator) presents infinite resistance to the transmitter and receiver that it connects.
Voltage and current levels at the transmitter and receiver are therefore kept apart. The electrical isolation (of
DC bias levels) of different parts of a communication system is desirable to simplify design. Optical fibre provides
this isolation naturally, and no further measures like transformer or capacitor coupling need to be taken.
● Fibre is made from a cheap and abundant raw material, and advances in the manufacturing process have reached
a stage where the finished fibre is now much cheaper than copper cable.
● Fibres are small in dimension (about the thickness of a human hair), low-weight, and flexible. They take up
much less space than copper cables and can therefore be easily accommodated in existing cable conduits.
Optical fibre does have several disadvantages, but these are entirely manageable and are far outweighed by the
above advantages.
● Special and costly connectors are required to join two fibres and to couple a fibre to an optical source. Losses
are always incurred even in the best splices or connectors. A good splice introduces a loss <0.1 dB, whereas a
connector is considered good if its loss is below 0.5 dB. The interconnection of copper cables is easier, cheaper,
and less lossy. Furthermore, because of its small dimension, great care is required to avoid subjecting fibre to
mechanical stress.
● Electrical power needed, for example, to drive electronic components at repeaters cannot be sent over the fibre
since it is not a conductor. In a copper line local telephone loop, the electric current needed to operate the
subscriber terminal is sent from the local exchange along the same wire pair that carries the message signal.
With fibre, an alternative means of supplying power to the home subscriber unit must be used.
● It is more difficult to locate faults on a fibre than it is on a metallic conductor. However, pulse reflectometers have
been developed that can locate a fibre break to within 0.1% of a repeater section, which could be an uncertainty
of up to 90 m (for systems that use 90 km repeater spacing).
● Optical fibre, like all other closed transmission media, cannot directly provide mobility and broadcast capabili-
ties, although it is an indispensable part of the backhaul network for interconnecting base stations and gateways
Telegram: @Persian_IEEEComSoc

in terrestrial wireless and satellite communication systems.

5.4.1 Optical Fibre Types

There are basically two types of fibre used as a transmission medium for optical communication, namely the
single-mode fibre and multimode fibre, where mode refers to the number of paths taken by the light in propagating
within the core. The primary distinction between the two fibre types is in the diameter of the core, which is typically
8–10 μm for the single-mode and 50–62.5 μm for the multimode fibre. The surface diameter of the cladding in both
fibres is about 125 μm. Colouring of the outside jacket is often used to indicate the type of fibre, with the jacket
of single-mode fibre coloured yellow and that of multimode fibre coloured orange. Multimode fibre is further
classified as step index or graded index according to the profile of the refractive index of the core.
Figure 5.22 illustrates the dimensions and refractive index profiles of the fibre types briefly discussed below.
 368 5 Transmission Media
Refractive index

n2 n2 n2
n1 n1 n1

d d d

62.5 μm 8 μm 50 μm
125 μm 125 μm 125 μm

(a) (b) (c)

Figure 5.22 Optical fibre types: (a) Multimode step index; (b) Single-mode; (c) Multimode graded index.

5.4.1.1 Single-mode Step Index

This has a very small core with a diameter of 8–10 μm and a uniform refractive index n2 . The refractive index
n1 of the cladding is smaller than that of the core by about 0.4% and there is a step change in refractive index
at the core/cladding boundary. Rays are launched into the core from an optical source, but only the central ray
can propagate in the core along a single path. The problem of multipath propagation encountered in multimode
fibre propagation and its consequent bandwidth constraint do not arise. This type of fibre therefore has the largest
bandwidths, limited only by material dispersion arising from the variation of the speed of light with wavelength.
However, its tiny core diameter makes splicing and coupling to an optical source extremely difficult. An optical
source generating coherent light of very narrow wavelength spread is required for the proper operation of the
single-mode fibre. This is provided by an injection laser diode (ILD) such as the Fabry-Pérot laser diode. The
single-mode fibre is suitable for long-distance high-speed (i.e. high data rate) links.

5.4.1.2 Multimode Step Index Fibre

This has a core of diameter 50–62.5 μm, which has a uniform refractive index n2 . There is a step drop in refrac-
tive index by about 1% at the core/cladding boundary. The cladding has a uniform refractive index n1 , which is
Telegram: @Persian_IEEEComSoc

about 0.99n2 . The larger core diameter allows splicing (joining two pieces) and coupling to an optical source to be
done with only a small loss penalty. Light is launched into the core from an optical source and the rays propagate
within the core by total internal reflection at the core/cladding interface. The light source is usually an infrared
light-emitting diode (LED) or an injection laser diode (ILD) and the wavelengths of practical operation are nomi-
nally 850, 1310, and 1550 nm. Bandwidth is limited by modal dispersion (also called multipath dispersion), which
arises because rays launched at small angles of incidence into the core travel along a shorter path and hence
arrive earlier at the receiver than rays with larger incidence angles which follow a more inclined and hence longer
multiple-straight-line path. The effect of multipath dispersion is to broaden narrow pulses and therefore limit the
bit rate that can be used without adjacent pulses overlapping in time. Multimode fibre is therefore only suitable for
short-distance lower data rate links such as in LANs. In fact, the multimode step index fibre is rarely used today
having been largely replaced by multimode fibres with a graded index core profile.
 5.4 Optical Fibre 369

5.4.1.3 Multimode Graded Index

This fibre type was devised to take advantage of the convenience afforded by the larger core of the multimode fibre,
while at the same time alleviating the multipath propagation problem. The core is of diameter 50–62.5 μm, but its
refractive index is not uniform. Rather it decreases according to a parabolic law, from a maximum n2 at the core axis
to a minimum n1 (≈ 0.99n2 ) at the core/cladding interface. By careful selection of the refractive index profile, the
wave can be made to propagate within the core by continuous refraction along serpentine paths at nearly the same
axial speed irrespective of angle of incidence. Rays that travel along longer paths do so in the lower index region,
where they have greater speed. On the other hand, those rays that travel through shorter distances have smaller
average speeds since they spend most of their time near the fibre axis where the refractive index is maximum. All
rays, irrespective of their path, will therefore arrive at the other end of the fibre at roughly the same time, and
this reduces modal dispersion. This feature can increase the bandwidth of this type of fibre by up to an order of
magnitude (i.e. a factor of 10) above that obtainable with multimode step index fibres. A graded index fibre also
gives the practical benefit of a larger core diameter and is the dominant type of multimode fibre in use today.

5.4.2 Coupling of Light into Fibre

The process of inserting light from a source into the core of an optical fibre is known as coupling. Figure 5.23 shows
a cone of acceptance within which light from the source must lie in order to be confined within the fibre core of
refractive index n2 . The half-angle 𝜃 a of this cone of acceptance is determined as follows. In order to be confined
entirely within the core (and not be transmitted via refraction into the cladding of refractive index n1 < n2 ), the
light ray that is inserted into the core must be incident on the core/cladding interface at an angle no lower than
the critical angle 𝜃 c . When light propagates from a medium of higher refractive index n2 into a medium of lower
refractive index n1 , the angle of incidence 𝜃 i and angle of refraction 𝜃 r are related according to Snell’s law as
n2 sin𝜃 i = n1 sin𝜃 r . The critical angle 𝜃 c is the angle of incidence at which the angle of refraction is 90∘ . Thus
n2 sin 𝜃c = n1 sin 90∘
⇒ 𝜃c = sin−1 (n1 ∕n2 ), n2 > n1 (5.69)
Applying Snell’s law of refraction to the interface between the input medium (usually air of refractive index
na ≈ 1) and the core of refractive index n2 , and noting that (for total internal reflection to occur in the core) the
maximum angle of refraction in the core is 𝜃 2 = 90 − 𝜃 c , we obtain
na sin 𝜃a = n2 sin 𝜃2
n2 n n
sin 𝜃a = sin 𝜃2 = 2 sin(90 − 𝜃c ) = 2 cos 𝜃c
na na na
This angle 𝜃 a is the maximum angle of acceptance. Any ray that is incident on the air/core interface at an angle
larger than 𝜃 a will have an angle of refraction larger than 90 − 𝜃 c and hence will be incident on the core/cladding
Telegram: @Persian_IEEEComSoc

interface at an angle less than the critical angle and will therefore be partly refracted through into the cladding

Figure 5.23 Critical angle, cone of acceptance and numerical Input medium (na)
aperture. Cladding (n1)
θ2
θc
θa θ2 Core (n2)

Cone of
acceptance
 370 5 Transmission Media

and partly reflected back into the core. In other words, the total internal reflection necessary to confine the light
entirely within the core will only take place for rays within the cone of acceptance shown in Figure 5.23 and
√na , core n2 , and cladding n1 . Since cos 𝜃c + sin 𝜃c = 1 and
2 2
specified below in terms of the refractive indices of air
sin 𝜃c = n1 ∕n2 (by Eq. (5.69)), it follows that cos 𝜃c = 1 − (n1 ∕n2 ) so the expression for sin𝜃 a simplifies to
2

n √ 1
√
sin 𝜃a = 2 1 − (n1 ∕n2 )2 = n22 − n21
na na
sin𝜃 a gives a measure of the maximum angle of acceptance and is called the numerical aperture (NA). Thus
√
1
Numerical aperture (NA) = n22 − n21 (5.70)
na
It should be noted, however, that even for light falling in the cone of acceptance a small fraction will be rejected
due to reflection at the input medium/core interface. Reflectivity or reflectance gives an indication of the fraction
of light from a source that is reflected at the source/fibre interface so that it fails to be coupled into the fibre core.
It is given by the expression
( )
n2 − na 2
Reflectivity = (5.71)
n2 + na
where n2 and na are, respectively, the refractive index of the core and the input medium (usually air). Once the
light has been coupled, the number of modes that can propagate in the core of the fibre depends on core diameter
D, NA, and wavelength 𝜆, and may be estimated from the normalised frequency parameter 𝜐
NA
𝜐 = D𝜋 ⋅ (5.72)
𝜆
If 𝜐 ≤ 2.4 then only a single mode can be supported, but if 𝜐 > 2.4 then two or more modes exist, in which case
if 𝜐 ≫ 1 then the actual number of modes M can be estimated using
{
0.5𝜐2 , Step index
M= (5.73)
0.25𝜐 ,
2 Graded index

Worked Example 5.6 A fibre has core and cladding refractive indices n2 = 1.5 and n1 = 1.485, respectively. If
light of wavelength 1550 nm is coupled into the fibre from an air input medium of refractive index 1.0, determine:
(a) The NA of the coupling.
(b) The apex angle of the cone of acceptance
(c) The reflectance of the coupling.
(d) The number of propagation modes in a 5/125 fibre.
(e) The number of propagation modes in a 62.5/125 graded index fibre.
Telegram: @Persian_IEEEComSoc

(a) Numerical aperture (NA)

√ √
1 1
NA = n22 − n21 = 1.52 − 1.4852
na 1
= 0.2116
(b) The apex angle of the cone of acceptance is twice the angle 𝜃 a shown in Figure 5.23. Thus, apex
(c) angle = 2𝜃 a = 2sin−1 (NA) = 2sin−1 (0.2116) = 24.4∘ .
( )
n2 − na 2 ( 1.5 − 1 )2 ( 0.5 )2
Reflectance = = =
n2 + na 1.5 + 1 2.5
= 1∕25
 5.4 Optical Fibre 371

(d) Using the numerical aperture obtained in (a) and the given signal wavelength and fibre dimension
(D = 5 μm = 5000 nm), the normalised frequency parameter is
NA 0.2116
𝜐 = D𝜋 ⋅ = 5000𝜋 ⋅
𝜆 1550
= 2.144
Since 𝜐 ≤ 2.4, there is only one mode of propagation.
(e) With D = 62.5 μm = 62 500 nm and the other parameters remaining the same as in (d), the normalised fre-
quency parameter is a factor (62 500/5000) larger than the previous value, so 𝜐 = 26.81. This is much larger
than 1, so the number of modes M is
M = 0.25𝜐2 = 0.25 × 26.812
= 180

5.4.3 Attenuation in Optical Fibre

Attenuation in optical fibre may be broadly classified into two categories, namely intrinsic and extrinsic attenua-
tion. Intrinsic attenuation arises from the interaction of light with the material of the fibre resulting in scattering
and absorption of the light. Absorption loss is the result of a small portion of the signal energy being dissipated
as heat by impurities in the fibre. The main contributors are traces of transition metal ions and hydroxyl ions.
The glass material itself will also absorb and dissipate a small amount of the signal energy as heat. Scattering loss
occur due to the scattering of signal energy at points of refractive index irregularity in the fibre. The irregularities
arise from the amorphous structure of silica material and are of a scale less than the wavelength of the light sig-
nal. On such a scale the resulting scattering is known as Rayleigh scattering, which causes the signal power to be
attenuated by 8.686𝛼l dB over a fibre length l, where 𝛼 and 8.686𝛼 are the attenuation coefficient in Np/km and
dB/km, respectively, and decrease as the fourth power of wavelength. The combination of absorption and Rayleigh
scattering gives an intrinsic fibre loss that has a minimum value of around 0.2 dB/km at a wavelength of 1550 nm.
Extrinsic attenuation arises from external factors not related to the fibre material and includes bending loss,
splicing loss, connector loss, and numerical aperture loss. These extrinsic losses may make up for the bulk of the
total attenuation on the link. Considerable bending loss will occur if the fibre is bent at some point into an arc of
a few millimetres in radius, which allows some of the light energy to escape into the fibre cladding. The escape
at sharp bends is by refraction in multimode fibre, and by radiation in single-mode fibre. Losses are also incurred
at joints between two fibre lengths. For a permanent joint by means of a fusion splice the loss is small, typically
<0.1 dB. But for a non-permanent butt joint, accomplished by holding the two fibre lengths in very close proximity,
the loss can range from about 0.3 dB for perfect alignment to a few dB if there is a significant misalignment. Finally,
coupling losses occur at the interface between the fibre and optical source/detector, as well as in a fibre coupler,
Telegram: @Persian_IEEEComSoc

which couples the signal from one input fibre to several output fibres, or from several input fibres to one or more
outputs.

5.4.3.1 Intrinsic Fibre Loss

Figure 5.24 shows the intrinsic loss (labelled total loss) in dB/km in a typical optical fibre with impurity level below
0.2 ppm (parts per million). The plot spans the wavelength or spectral bands given in Table 5.1, which have been
defined by the ITU for optical communication using single-mode fibre. The four contributory factors to intrinsic
fibre loss are (i) Rayleigh scattering which decreases as the fourth power of wavelength; (ii) nonlinear effects; (iii)
absorption by the intrinsic material of the glass, mainly the absorption of infrared light by silica which increases
dramatically beyond 1600 nm; and (iv) absorption by minute quantities of impurities in the glass, particularly
hydroxyl (OH− ) ion which has minor absorption peaks at wavelengths 720, 880, 950, 1130, and 1240 nm and a
 372 5 Transmission Media

4.0

3.0 Total loss

ea )
np H–
k
tio O
2.0

rp l (
so xy
ab dro
Hy
1.0
1310 nm
Loss (dB/km)

Window
(O-band)
0.5
0.4
Rayleigh 1550 nm
0.3 scattering window
Infrared
0.2 absorption

0.1
800 1000 1200 1400 1600 1800
Wavelength (nm)

Figure 5.24 Fibre loss as a function of wavelength. The two optical transmission windows at 1310 nm and 1550 nm are
shown.

Table 5.1 Optical transmission bands defined by ITU.

Band letter Band wavelength

Band description designation range

Original O 1260–1360 nm
Extended E 1360–1460 nm
Short wavelength S 1460–1530 nm
Conventional C 1530–1565 nm
Long wavelength L 1565–1625 nm
Ultralong wavelength U 1625–1675 nm
Telegram: @Persian_IEEEComSoc

major peak at 1380 nm, and metallic ions (Fe2+ , Cu2+ , Cr3+ , etc.) with minor peaks at wavelengths 625, 725, 850,
and 1100 nm. Loss due to absorption by impurities can be minimised by refining the glass mixture to an impuri-
ties level below 1 part per billion (ppb) using improved manufacturing techniques. However, Rayleigh scattering
and infrared absorption by silica combine to set an irreducible theoretical limit for fibre loss. Rayleigh scattering
dominates at wavelengths below 1500 nm, whereas infrared absorption is negligible below 1500 nm but increases
rapidly above 1600 nm, and the two components combine to give a minimum intrinsic fibre loss at 1550 nm. For
this reason, the 1500 nm window has been the prime spectral band for optical communication, described as the C
(i.e. conventional) band in ITU designation.
 5.4 Optical Fibre 373

5.4.3.1.1 Rayleigh Scattering

Scattering is a process in which the energy of an electromagnetic wave is dispersed in direction by mechanisms not
involving specular reflection or refraction. As a result, only a fraction of the wave energy continues in the direction
of incidence. The scattering objects, or scatterers, may be of a range of sizes compared to the wavelength of the
incident electromagnetic wave, leading to Rayleigh scatter when scatterer sizes are much smaller than wavelength,
Mie scatter when scatterer sizes are comparable to wavelength (i.e. ∼> 0.1𝜆), or nonselective scatter (i.e. wavelength
independent scatter) when scatterer sizes are much larger than wavelength. Rayleigh scattering within optical fibre
is therefore the result of light interaction with molecules, particles, and refractive index discontinuities within the
fibre whose sizes and dimensions are much smaller than the wavelength of the incident light. As illustrated in
Figure 5.25, a portion of the incident light will be scattered away from the forward direction as back scatter (towards
the source) and as side scatter (in other non-backward and non-forward directions) so that the amount of light
propagating towards the intended destination is slightly reduced. This reduction in signal strength is conveniently
expressed as specific attenuation, also called attenuation coefficient in dB/km, which for Rayleigh scattering in
optical fibre may be estimated from
( )
850 4
𝛼(𝜆) = 1.7 dB∕km (5.74)
𝜆
where 𝜆 is wavelength in nm. So, Rayleigh scattering accounts for around 1.7 dB/km loss at 850 nm wavelength
but only around 0.14 dB/km at 1600 nm.
It is worth mentioning that it is the phenomenon of scattering of light from the sun that is responsible for
many of our observations of colour in nature. For example, Rayleigh scattering of light is inversely proportional to
the fourth power of wavelength and is produced by molecules in the propagation medium whose sizes are much
smaller than the wavelength of light. Shorter wavelengths (such as blue light) are therefore more Rayleigh scattered
than longer wavelengths (such as red light) and this explains why the sky appears blue on a clear day when there
are no particulates in the sky. The sizes of water droplets in a cloud are much larger than the wavelength of light

Side
scatter
Forward
Incident em wave scatter
Scattering
medium
Back Transmitted em wave
scatter (Weaker than incident wave)
Telegram: @Persian_IEEEComSoc

Incident em wave Absorptive

medium Transmitted em wave
(weaker than incident wave)

Heat

Figure 5.25 Scattering and absorption of electromagnetic (em) wave in a medium.

 374 5 Transmission Media

and this gives rise to nonselective scatter whereby all wavelengths in the light from the sun are equally scattered,
which explains why clouds appear white.

5.4.3.1.2 Nonlinear Effects

At high transmitted light intensity, the fibre medium will exhibit nonlinear behaviour that further contributes to
signal attenuation and introduces nonlinear distortion such as the creation of new wavelength components in the
output signal. The fundamental causes of a nonlinear system response in the fibre transmission medium include
acoustic molecular vibrations and the Kerr effect.
If variations in the electric field component of the transmitted light are strong enough, acoustic molecular vibra-
tions and hence refractive index fluctuations may be generated in the fibre which give rise to scattering. Stimulated
Brillouin scattering (SBS) describes the situation where the refractive index fluctuations produce a backward prop-
agating wave called Stokes wave in response to the transmitted wave, which in this context is referred to as the
pump wave. Another consequence of the molecular vibrations is known as stimulated Raman scattering (SRS),
whereby scattered waves are produced in both forward and backward directions and some energy is transferred
from shorter to longer wavelengths. SRS impairs multichannel transmission due to its relative attenuation of
shorter wavelengths but may be beneficially exploited to achieve optical amplification.
The Kerr effect describes a phenomenon in which the refractive index of a material is changed by an applied
electric field. All materials exhibit this characteristic but some more strongly than others. Therefore, if the trans-
mitted light intensity is sufficiently strong, the electric field component of the light will produce slight variations in
the refractive index of the fibre material. This effect manifests itself in various ways in an optical communication
signal. Since different intensities of light will therefore propagate at slightly different speeds, intensity modulation
of transmission at a given wavelength will be converted into phase variation of that wavelength, an effect described
as self-phase modulation (SPM). The variation of light intensity in one wavelength varies the refractive index seen
by light in a different channel (or wavelength) and gives rise to phase variation in that other channel, an effect
known as cross-phase modulation (CPM). Finally, in view of its nonlinear refractive index in the presence of a
strong light intensity, the fibre medium has a nonlinear transfer characteristic. Therefore, if two or more wave-
lengths are simultaneously transmitted, new wavelength components will be generated in the medium. This effect
is like the intermodulation distortion discussed in Section 4.7.6 but in optical systems it is known as four-wave mix-
ing (FWM). The new wavelengths will cause interference in WDM optical systems where independent signals are
simultaneously transmitted in the same fibre using several regularly spaced wavelengths.

5.4.3.1.3 Absorption
This is a process by which energy in an electromagnetic wave is converted to heat in a medium through which the
wave is transmitted. See Figure 5.25. As a result, the strength of the wave is proportionately reduced with increasing
length of propagation through the medium. It should be noted that the attenuation of a voltage wave on a metallic
transmission line (discussed in Section 5.3) is also an absorption process by which energy in the voltage wave is
Telegram: @Persian_IEEEComSoc

converted to heat in the resistance of the metallic material. The reduction in signal strength caused by absorption
is usually also quantified as specific attenuation in dB/km.
The amount of light absorption in optical fibre depends on the wavelength of the light and comes from three
main contributing factors:
● The presence of impurities in the glass, namely hydroxyl (OH− ) and transition metal ions. These ions have
absorption peaks at the various optical band wavelengths identified earlier. Absorption by these impurities may
be reduced to negligible levels in ultra-pure fibre having impurity concentrations of less than 1 ppb.
● Intrinsic light absorption property of silica glass. The absorption of light by silica is negligible below 1500 nm
and increases rapidly above 1600 nm. It is this factor that is responsible for the sharp rise in total intrinsic fibre
loss above 1600 nm, as shown in Figure 5.24.
● The presence of defects in the atomic structure of the glass material of the fibre.
 5.4 Optical Fibre 375

5.4.3.2 Extrinsic Fibre Loss

5.4.3.2.1 Bending Loss
A bend in the fibre may cause the light within the fibre core to be incident on the core/cladding interface at an angle
that is less than the critical angle of the medium, which breaches the condition for total internal reflection within
the core and leads to a portion of the light escaping from the core. The light propagating within the core thereby
suffers some loss or attenuation. This situation is illustrated in Figure 5.26 for both macrobend and microbend.
Microbend losses are due to small wrinkles in the fibre, which are not visible to the naked eye, caused by tem-
perature variations during manufacture, extreme temperature variations in installed cable, external forces that
deform the cable jacket, or small-scale fibre discontinuities. Macrobend losses are due to bends that are visible to
the naked eye caused, for example, by laying the fibre around a sharp corner or wrongly installing tie wraps. A
minimum bend radius of 5–10 times the outer diameter of the fibre is recommended to avoid excessive macrobend
loss. Longer wavelengths are more impacted by macrobending than shorter wavelengths are.

5.4.3.2.2 Splicing Loss

A long-distance optical fibre link will typically consist of multiple fibre segments that are joined together through
splicing. There are two methods of splicing, namely fusion splicing and mechanical splicing. In fusion splicing, the
two fibre ends are heated using an electric arc and melted and fused together to create a homogeneous, nonre-
flective, and reliable joint of very low loss, typically between 0.01 and 0.10 dB for a single-mode fibre. Mechanical
splicing creates a fast and easy joining of two fibre lengths using adhesives and gels. It is typically employed in
emergency link restoration situations and introduces a loss of around 0.05–0.2 dB for a single-mode fibre.

5.4.3.2.3 Connector Loss

When two fibre segments are joined together using connectors, small losses might be introduced by the following
defects, leading to losses totalling typically between 0.1 and 0.25 dB.

● If there is a small air gap between the connected ends of the two fibre lengths as a result of a faulty mechani-
cal connector or wrong fitting, a fraction of the light energy in the incoming fibre segment will be lost due to
reflection at the glass/air interface created by the air gap. Equation (5.71) gives the reflectivity or fraction of
lost power in such situations. In Worked Example 5.6, we calculate a typical reflectivity value of 1/25, which
means that only a fraction 24/25 of the power would be delivered from one fibre to the next, representing a loss
of −10log10 (24/25) = 0.18 dB due to reflection at the air gap.
● Any misalignment of the two fibre cores at the connector will cause some of the light in the source fibre seg-
ment not to be passed into the core of the second fibre segment. The amount of loss depends on the degree of
misalignment.
● If the core diameter Dr of the receiving fibre is smaller than the core diameter Ds of the source fibre, there will
Telegram: @Persian_IEEEComSoc

be a loss of −20log10 (Dr /Ds ) dB. This loss is avoidable in both directions of transmission by ensuring that only
fibres of the same dimensions are connected.

Escaped light
Escaped light

Fibre core
(a) Macrobend (b) Microbend
Fibre core

Figure 5.26 Bending loss.

 376 5 Transmission Media

● Contamination of the connectors (e.g. by finger oil or dirt particles) will lead to losses due to absorption and
the scattering of light by the contaminants. It is therefore important that connectors are handled with care and
examined using an inspection scope and properly cleaned (if necessary) prior to use.

5.4.3.2.4 Numerical Aperture Loss

When two devices are coupled and the numerical aperture (NA) of the receiving device, denoted NAr , is smaller
than the NA of the source device NAs , then only a fraction of the energy from the source device will be successfully
transferred into the receiving device. The loss incurred is called NA loss and is given by
( )
NAr
NA loss = −20 log10 dB (5.75)
NAs
Light-emitting diodes (LEDs) have NA ∼ 1, whereas laser diodes (LDs) have NA ∼ 0, so this loss may be ignored
in laser sources.
The attenuation A of an optical fibre link of length l is given by the sum of all intrinsic and extrinsic losses as
A = 𝛼l + 𝛼s ns + 𝛼c nc dB (5.76)
where 𝛼 is the attenuation coefficient (dB/km) of the fibre used in the link, 𝛼 s is the mean splice loss, ns is the
number of splices in the link, 𝛼 c is the mean loss of line connectors, and nc is the number of line connectors used
in the link.

5.4.4 Dispersion in Optical Fibre

Information is conveyed in optical fibre as pulses of light transmitted in sequential symbol intervals. Ideally, the
width of each pulse should be unchanged by propagation in the medium to avoid the problem of intersymbol
interference (ISI), which we identify in Section 5.2.3 under metallic lines. In practice, however, the different com-
ponents or portions of a light pulse will experience slightly different propagation delays in the fibre and therefore
will arrive at slightly different times at the receiving end and combine to produce a broader pulse than what was
transmitted. The broadening of an electromagnetic pulse in a transmission medium is known as dispersion. In
this case, each light pulse will be broader at exit from the fibre than at entry, as illustrated in Figure 5.27. The
components of the pulse may be the various wavelengths that constitute the light signal or the range of angles of
incidence in the coupled light beam or the two orthogonal polarisations of the light. If each of these components
has a different path length or propagation time through the fibre then dispersion will occur. There are two types
of dispersion in optical fibre, namely intermodal and intramodal dispersion.

5.4.4.1 Intermodal Dispersion

Also called multimode or multipath or modal dispersion, this occurs only in multimode fibre as a result of the
Telegram: @Persian_IEEEComSoc

dependence of path length through the fibre on angle of incidence, as illustrated in Figure 5.22a and c. A multi-
mode step index fibre has a uniform core refractive index, so all incident rays will travel at the same speed through
the fibre core. The outer longer-path rays will therefore take longer to reach the receiver than the axial rays, thereby
causing a broadening of the received light pulse. A measure of the amount of dispersion is given by the pulse spread

Optical fibre medium

Input pulse Output pulse

Figure 5.27 Pulse dispersion.

 5.4 Optical Fibre 377

𝜏 ps , which is the difference in propagation time between a pulse or ray propagating in lowest-order mode along
the core axis and another ray propagating by total internal reflection in highest-order mode along the longest path
through the core. Modal dispersion is greatly reduced if the multimodal fibre core has a parabolic refractive index
profile such that longer ray paths through the core have proportionately higher pulse propagation speeds. It is
through this mechanism that the modal dispersion of multimode graded index fibre is in the range 𝜏 ps = 0.3 to
1.0 ns/km, whereas that of multimode step index fibre is 𝜏 ps = 50 ns/km.

5.4.4.2 Intramodal Dispersion

Also referred to as material dispersion, intramodal dispersion occurs in single-mode fibre and includes chromatic
dispersion (CD) and polarisation mode dispersion (PMD). CD arises because each of the wavelength components of
the light pulse will see a different fibre core refractive index, and therefore will travel at a slightly different speed.
CD depends on the spectral width of the light, and on the type of single-mode fibre. It is usually specified in units
of picoseconds per nanometre per kilometre. A CD of 1 ps/(nm km) means that a pulse of spectral width 1 nm will
spread out by 1 ps for each kilometre travelled along the fibre. CD may be ignored in low-speed transmissions but
must be considered in high-speed systems operating at 10 Gb/s or higher. Measures to combat CD include: (i) use
of special fibres such as the dispersion shifted and nonzero dispersion shifted fibres stipulated by the ITU [4, 5],
(ii) regeneration of the light pulses at regular intervals along the optical fibre link, and (iii) use of light sources and
transmitters with narrow spectral width.
The light pulse that is transmitted through a fibre core is usually unpolarised light that may be resolved into two
pulses having mutually orthogonal polarisations, each describing the orientation of the electric field component
of the light. Ideally, these two pulses travel at the same speed through the fibre so that, barring other sources
of dispersion, the receiver sees a single undistorted and unpolarised pulse. However, random imperfections in
practical fibres will disrupt the circular symmetry of the core structure, and this causes the refractive index of the
fibre material to be dependent on light polarisation. Therefore, two orthogonal polarisation modes will travel at
slightly different speeds through the fibre medium. An unpolarised light pulse sent by a transmitter will arrive
at the receiver as two slightly time-displaced polarised pulses that combine to produce a single broadened pulse.
The delay between the two polarisation modes over the length of the link is referred to as differential group delay
(DGD) and the resulting distortion is known as polarisation mode dispersion (PMD), measured in picoseconds.
The impact of PMD is, however, insignificant on links less than about 1600 km in length or 40 Gb/s in data rate.
The main effect of dispersion in optical communication systems is to place an upper limit on link symbol rate,
i.e. the rate Rs at which light pulses can be transmitted on the link. We may estimate this limitation by considering
Figure 5.28 in which pulses of width 𝜏 are sent at regular intervals T s . The symbol rate is therefore Rs = 1/T s and
the duty cycle of the transmitted waveform is d = 𝜏/T s . If there is to be no interference due to pulse spreading then
the maximum allowed dispersion 𝜏 ps is as shown in Figure 5.28b such that each broadened pulse just stops short
of the sampling instant of the next symbol interval. Thus
Telegram: @Persian_IEEEComSoc

𝜏ps = Ts − 𝜏∕2 = Ts − dT s ∕2 = Ts (1 − d∕2)

This relationship allows us to express maximum symbol rate in terms of medium dispersion as
1 1 − d∕2 2 − d
Rs = = = (5.77)
Ts 𝜏ps 2𝜏ps

Figure 5.28 illustrates a general case involving return-to-zero (RZ) pulses with waveform duty cycle d < 1. In the
special case of full width pulses, called non-return-to-zero (NRZ) pulses, pulse width equals symbol interval, so
that d = 1 and the above equation reduces to
1
Rs = (5.78)
2𝜏ps
 378 5 Transmission Media

τ
Duty cycle d = τ/Ts
(a) Transmitted
pulse sequence τ/2
t
0 Ts 2Ts → symbol intervals

τps
(b) Maximum Broadened pulse
allowed spread τps
t
0 Ts 2Ts → sampling instants

Figure 5.28 Analysis of limitation placed on symbol rate by pulse broadening.

Modal dispersion 𝜏 ps in a multimode step index fibre may be estimated from the path difference between the
outermost ray and the axial ray divided by the speed of light in the fibre. This leads to
n1 Δ
𝜏ps ≈ l
c
n − n1
Δ= 2 (5.79)
n2
where, n2 = refractive index of fibre core; n1 = refractive index of fibre cladding; c = speed of light, l = link length,
and Δ ≡ normalised core-cladding refractive index difference. Note the modal dispersion in a graded index fibre
will be significantly lower than the value given by Eq. (5.79), which only applies to a step index fibre.

Worked Example 5.7 Dispersion and Symbol Rate

Information is transmitted on a 1 km long multimode step index optical fibre using RZ pulses of negligible duty
cycle. The fibre has core and cladding refractive indices of 1.5 and 1.485, respectively.
(a) What is the maximum allowable symbol rate to avoid errors due to modal dispersion?
(b) If the above link is implemented using a graded index fibre with dispersion 0.4 ns/km, what is the new maxi-
mum symbol rate?
(c) If the link length in (b) is to be increased to 10 km without using intervening repeaters, what is the maximum
bit rate possible on this longer link?
(a) We first calculate modal dispersion as
𝜏ps n Δ n − n1 0.015
= 2 = 2 = = 50 ns∕km
Telegram: @Persian_IEEEComSoc

l c c 3 × 105 km∕s
We then determine maximum symbol rate, noting that link length l = 1 km, and that transmission is by RZ
pulses
2−d 2−0 1 1 109
Rs = = = = = = 20 MBd
2𝜏ps 2𝜏ps 𝜏ps 50 ns 50
(b) Total dispersion is now only 0.4 ns over the 1 km link. Using this value in the above calculation in place of
50 ns yields the new maximum symbol rate

1 109
Rs = = = 2.5 GBd
0.4 ns 0.4
 5.4 Optical Fibre 379

(c) Total dispersion on the 10 km link will be 0.4 ns/km × 10 km = 4 ns. Therefore, the maximum allowed symbol
rate on this longer link will be
1 109
Rs = = = 250 MBd
4 ns 4
Notice that as the repeater-less distance increases, the symbol rate decreases proportionately so that the product
of symbol rate and link length is a constant called bandwidth-distance product or bandwidth-length product. The
optical link in (b) thus has a bandwidth-length product of 2.5 GHz⋅km.

Worked Example 5.8 Signal Power Budget on Optical Fibre Link

An optical fibre communication system has the following specifications:
● Transmit power = 1 mW.
● Receiver sensitivity = −42 dBm.
● Maximum length of continuous fibre section s = 2 km.
● Splicing loss 𝛼 = 0.5 dB per splice.
● Coupling loss = 1 dB at each end.
Determine the maximum length of this optical fibre link that can be operated without a repeater.
Let l denote the maximum link length in km that we seek. It is more convenient to set out the solution compactly
in tabular form, as shown in Table 5.2. Notice that all losses and the fade margin are entered as negative values so
that they reduce the received signal power when algebraically summed in the final row. Received power is given
in the final row of the table as
Pr = −(0.45l + 1.5) dBm
Since receiver sensitivity (i.e. minimum received signal power needed by the receiver for reliable operation) is
specified as −42 dBm, we set the above expression for Pr to this value and solve for link length l
Pr = −(0.45l + 1.5) = −42
42 − 1.5
⇒l= = 90 km
0.45

Table 5.2 Worked Example 5.8: Signal power budget on optical link.

Quantity Computation Value

Pt , Transmit Given (1 mW) = 10 log10 (1) dBm 0 dBm

output power
Telegram: @Persian_IEEEComSoc

Total fibre loss Fibre loss per km multiplied by total −0.2 l dB

link length
Total coupling Sum of coupling loss at either end −2 dB
loss (1 + 1 dB)
Total splicing (Number of sections − 1)𝛼 = (l/s − 1)𝛼 −0.5(l/2–1) dB
loss
M, fade margin This is zero at maximum link length −0 dB
since no minimum value was specified
or required
Pr , received Algebraic sum of above 5 quantities −(0.45 l + 1.5) dBm
signal power
 380 5 Transmission Media

5.5 Radio

Radio is a small portion of the wider electromagnetic spectrum tabulated in Table 5.3, the final column of which
identifies the radio bands, optical bands, and ionising radiation bands. The radio spectrum has traditionally been
divided into bands from ELF (extremely low frequency) to EHF (extra high frequency), with each band spanning
a factor of 10 in frequency (and wavelength). The bands from UHF to EHF are collectively referred to as the
microwave band, whereas the EHF band on its own is commonly called the millimetre wave band in recognition of
the fact that wavelengths in this band have values ranging from 1 to 10 mm. Much of the microwave radio spectrum
has been further divided into smaller sub-bands identified by letter designations, as listed in Table 5.4.
An electromagnetic wave consists of a changing electric field E, which generates a changing magnetic field H
in the surrounding region, which in turn generates a changing electric field in the surrounding region, and so on.
The coupled electric and magnetic fields therefore travel out in space. The speed of propagation in vacuum can
be shown to be c = 299 792 458 m/s (often approximated as 3 × 108 m/s), which is the speed of light. The fields E
and H are vector quantities having both magnitude and direction at every time instant and at every point in space
covered by the wave. The direction or orientation of the E field defines the polarisation of the wave. Figure 5.29
shows a snapshot of a vertically polarised electromagnetic wave that propagates in the +z direction. Monitoring
the E or H field strength at a fixed point in space we find that it varies sinusoidally with time, completing one
cycle of values in a time T (called the wave period), which means that the wave completes 1/T cycles per second, a

Table 5.3 The electromagnetic spectrum.

Frequency Band Frequency Wavelength

ELF (Extremely low frequency) < 3 kHz > 100 km
VLF (Very low frequency) 3 – 30 kHz 10 –100 km
LF (Low frequency) 30 – 300 kHz 1 –10 km
MF (Medium frequency) 300 kHz – 3MHz 100 m –1 km
Millimetre wave

Bands
Radio
Microwave

HF (High frequency) 3 – 30 MHz 10 m – 100 m

VHF (Very high frequency) 30 – 300 MHz 1 m – 10 m
UHF (Ultra high frequency) 300 MHz – 3 GHz 10 cm – 1 m
SHF (Super high frequency) 3 – 30 GHz 1 cm – 10 cm
EHF (Extra high frequency) 30 – 300 GHz 1 mm – 1 cm
Sub-millimetric 300 GHz – 3 THz 100 µm –1 mm
Optical
Bands

Far-infrared 3 – 30 THz 10 µm – 100 m

Near-infrared 30 – 430 THz 698 nm – 10,000 nm
Visible light 430 – 860 THz 349 nm – 698 nm
Telegram: @Persian_IEEEComSoc

Radiation

Ultraviolet 860 THz – 30 PHz 10 nm – 349 nm

Ionizing

X-ray 30 PHz – 300 EHz 1 pm – 10,000 pm

Gamma ray > 100 EHz < 3 pm
NOTES: Standard SI prefixes:
deca(D) = 101; hecto(h) = 102; kilo(k) = 103; mega(M) = 106; giga(G) = 109; tera(T) = 1012; peta(P) = 1015; exa(E) = 1018; zetta(Z) = 1021; yotta(Y) = 1024;
deci(d) = 10–1; centi(c) = 10–2; milli(m) = 10–3; micro(µ) = 10–6; nano(n) = 10–9; pico(p) = 10–12; femto(f) = 10–15; atto(a) = 10–18; zepto(z) = 10–21; yocto(y) = 10–24
 5.5 Radio 381

Table 5.4 Sub-bands of the microwave radio

spectrum.

Letter designation Frequency range

of band of band (GHz)

L 1–2
S 2–4
C 4–8
X 8–12
Ku 12–18
K 18–26.5
Ka 26.5–40
Q 30–50
U 40–60
V 50–75
E 60–90
W 75–110
F 90–140
D 110–170

x
W
av
ele
ng
th,
λ

E field

E = Em cos(ωt – kz)x̂
H = Hm cos(ωt – kz)ŷ

y
H field
Telegram: @Persian_IEEEComSoc

Figure 5.29 Snapshot of a vertically polarised electromagnetic wave propagating in the +z direction.
 382 5 Transmission Media

quantity called the wave frequency f (in hertz, Hz). Since one cycle is 2𝜋 radian, it follows that the wave completes
2𝜋f radian per second, a quantity called the angular frequency of the wave and denoted 𝜔 (in rad/s).
Taking a snapshot of the wave at a given time instant (as in Figure 5.29), we find that the fields also vary sinu-
soidally with distance z, completing one cycle in a distance 𝜆, called the wavelength of the wave. Again, since one
cycle is 2𝜋 radian, it means that the wave completes 2𝜋 radian per 𝜆 metres, or 2𝜋/𝜆 radian per meter, a quantity
called the wavenumber or phase constant of the wave and denoted 𝛽 (in rad/m). Combining the two sinusoidal
dependencies (on time t and distance z) gives an expression for the electric field value at any time t and distance z
as E = Em cos(𝜔t ± 𝛽z). The crest (or peak) of this field is located wherever the argument 𝜔t ± 𝛽z of the cosine func-
tion in this expression equals zero. Therefore, to track (i.e. move in step with) this crest and hence determine the
speed of propagation or phase velocity v of the wave, the value of z must change as time t increases in such a way that
𝜔t ± 𝛽z = 0 at all times. In the case of a negative sign, i.e. (𝜔t − 𝛽z), z must increase as t increases to keep 𝜔t − 𝛽z = 0,
whereas in the case of a positive sign, i.e. (𝜔t + 𝛽z), z must decrease as t increases. Thus, Em cos(𝜔t − 𝛽z) represents
a wave moving in the +z direction and Em cos(𝜔t + 𝛽z) is a wave moving in the −z direction. The phase velocity is
the ratio z/t in the equation 𝜔t − 𝛽z = 0. Thus, v = z/t = 𝜔/𝛽. Since 𝜔 = 2𝜋f and 𝛽 = 2𝜋/𝜆, it means that v = 𝜆f = 𝜆/T
and 𝜆 = vT, from which we see that wavelength 𝜆 is the distance travelled by the wave in a time of one period T.
The above parameters of a sinusoidally varying electromagnetic wave are related as follows
Period ≡ T (s)
Frequency: f = 1∕T (Hz)
Angular frequency: 𝜔 = 2𝜋f (rad∕s)
Wavelength ≡ 𝜆 (m)
Wavenumber: 𝛽 = 2𝜋∕𝜆 (rad∕m)
Phase velocity: v = 𝜔∕𝛽 = 𝜆f (m∕s) (5.80)
For radio waves in air
Speed of light (≈ 3 × 108 m∕s)
Radio wavelength (m) = (5.81)
Radio frequency (Hz)
Radio waves have been increasingly exploited for information transmission for over a hundred years and this
trend is expected to continue into the foreseeable future. The information signal to be transmitted is first employed
to vary a parameter (namely amplitude, frequency, phase, or combinations of these) of a high-frequency sinusoidal
current signal in a process known as carrier modulation. The time-varying current signal is then fed into a suitably
designed conducting structure or antenna, and this sets up a similarly time-varying magnetic field in the surround-
ing region, which in turn sets up a time-varying electric field, and so on. The antenna thereby radiates electromag-
netic waves. Based on Faraday’s law of electromagnetic induction, we know that these waves can induce a similarly
time-varying current signal in a distant receive-antenna. In this way the information signal may be recovered at
the receiver, having travelled from the transmitter at the speed of light. For radio communications application, an
Telegram: @Persian_IEEEComSoc

appropriate frequency band is usually chosen that satisfies the required coverage and propagation characteristics.

5.5.1 Maxwell’s Equations

Maxwell’s equations are a collection of four physical laws that successfully predict the existence of electromagnetic
(EM) waves travelling at the speed of light. For simplicity, we will not discuss these equations in detail, but it is
useful to be at least familiar with their word descriptions given below.
● The first of Maxwell’s equations is Gauss’s law, which states: the total electric flux emanating from a closed
surface S is equal to the total charge enclosed within S. If the charge enclosed is zero, the electric flux lines are
continuous.
● The second of Maxwell’s equations is Gauss’s law for magnetism, which states: the net magnetic flux out of a
closed surface is always zero. Thus, there is no magnetic equivalent of electric charge.
 5.5 Radio 383

● The third of Maxwell’s equations is Faraday’s law of electromagnetic induction, which stipulates that a time-
varying magnetic field produces a time-varying electric field and formally states: the induced electromotive
force (emf) along a closed path equals the rate of decrease of the magnetic flux through the area enclosed by the
path.
● The fourth of Maxwell’s equations is Ampere’s modified law, which stipulates that a time-varying electric field
produces a time-varying magnetic field and formally states: the line integral of the magnetic field intensity along
a closed path equals the conduction and displacement current through the area enclosed by the path.

The above laws lead to wave equations in E and H which may be solved in a general propagation medium
having primary parameters 𝜎, 𝜀, and 𝜇, being respectively the conductivity, electric permittivity, and magnetic
permeability of the medium. We find in the solution that the E and H waves travel at the same speed, so we say
that they are coupled together and constitute what is named electromagnetic wave – ‘electro’ for E and ‘magnetic’
for H. The solution yields expressions for important secondary parameters of the medium in terms of 𝜎, 𝜀, and 𝜇.
These secondary parameters include those already introduced in Eq. (5.80) (i.e. wavenumber 𝛽, wavelength 𝜆, and
phase velocity v) as well as impedance Z, refractive index n, and attenuation constant 𝛼.
The impedance Z at a given point in a medium is defined as the ratio between the amplitude Em of the electric
field and the amplitude H m of the magnetic field observed at the same point. Since electric field is in volt per unit
length and magnetic field is in ampere per unit length, note that this definition is consistent with the definition (in
Section 5.3) of impedance on transmission lines as the ratio between voltage and current. The index of refraction
or refractive index n of a medium is defined as the ratio between the speed c of electromagnetic wave in vacuum
and the speed v of the same wave in the medium. That is
Em
Z≡
Hm
Speed of EM wave in vacuum
n≡ (5.82)
Speed of same wave in medium
The attenuation constant 𝛼 is the rate (in neper per unit distance) at which the amplitudes of the E and H waves
decay with distance travelled in the medium. Thus, if the EM wave propagates in the +z direction and its electric
and magnetic components have amplitudes Em and H m at z = 0, then their amplitudes after travelling distance
z = l in the medium will be

E(l) = Em e−𝛼l ; H(l) = Hm e−𝛼l (5.83)

This means that if the attenuation constant 𝛼 is specified in Np/m, then the wave will be attenuated by
8.686𝛼 dB/m.
The expressions for the six secondary parameters are tabulated in Table 5.5, including their exact forms in all
media and in a perfect insulator (𝜎 = 0) and their approximations in good dielectric and good conductor media.
Telegram: @Persian_IEEEComSoc

The electric permittivity 𝜀 of a medium is usually specified through the relative permittivity or dielectric constant
𝜀r of the medium, from which 𝜀 = 𝜀r 𝜀0 , where 𝜀0 = 8.8541878128 × 10−12 farad per metre (F/m) is the electric
permittivity of free space or vacuum. Similarly, the relative permeability 𝜇r of a medium is usually specified, from
which the magnetic permeability 𝜇 of the medium is 𝜇 = 𝜇 r 𝜇 0 , where 𝜇 0 = 1.25663706212 × 10−6 henry per metre
(H/m) is the magnetic permeability of vacuum. Table 5.5 gives very useful results. For example, impedance and
wave speed in free space (where 𝜎 = 0, 𝜀 = 𝜀0 , 𝜇 = 𝜇 0 ) are given by
√ √
𝜇0 1.25663706212 × 10−6
Z= = = 376.73 Ω
𝜀0 8.8541878128 × 10−12
1
v= √ = 299, 792, 458 m∕s ≡ c (5.84)
𝜇 0 𝜀0
 384 5 Transmission Media

Table 5.5 Secondary parameters for electromagnetic wave in various media.

Secondary Exact expression Perfect insulator Good dielectric Good conductor

parameter (all media) (𝝈 = 0) (𝝈/𝛀𝜺)2 ≪ 1 (𝝈/𝛀𝜺)2 ≫ 1
√ [√ ]
√ ( )2 √
√ 𝜇𝜀 𝜎 √ √ 𝜔𝜎𝜇
Wave number, 𝛽 𝜔√ 1+ +1 = 𝜔 𝜇𝜀 ≈ 𝜔 𝜇𝜀 ≈
2 𝜔𝜀 2
√ [√ ]
√ ( )2 √ √
√ 𝜇𝜀 𝜎 𝜎 𝜇 𝜔𝜎𝜇
Attenuation constant, 𝛼 𝜔√ 1+ −1 0 ≈ ≈
2 𝜔𝜀 2 𝜀 2
√ √ √ √
j𝜔𝜇 𝜇 𝜇 𝜔𝜇
Wave impedance, Z = ≈ ≈ (1 + j)
𝜎 + j𝜔𝜀 𝜀 𝜀 2𝜎
√
2𝜋 2𝜋 2
Wavelength, 𝜆 2𝜋/𝛽 = √ ≈ √ ≈ 2𝜋
𝜔 𝜇𝜀 𝜔 𝜇𝜀 𝜔𝜎𝜇
√ √ √
Phase velocity, v 𝜔/𝛽 = 1∕ 𝜇𝜀 ≈ 1∕ 𝜇𝜀 ≈ 2𝜔∕𝜎𝜇
√ √ √
Refractive index, n c/v = 𝜇r 𝜀r ≈ 𝜇r 𝜀r ≈ 𝜎𝜇r ∕2𝜔𝜀0

And the refractive index of a dielectric medium (where 𝜎 ≈ 0) is

√
n = 𝜇 r 𝜀r (5.85)

The solution of Maxwell’s equations in free space further reveals that the electric and magnetic fields are mutu-
ally perpendicular to each other and to the direction of propagation, the orientations of the three vectors being
related according to the right-hand rule: let the four fingers of the right hand be opened out together and the thumb
outstretched perpendicular to them then if the fingers are curled to fold the electric field vector onto the magnetic
field vector the thumb points in the direction of propagation. Thus, a complete expression of each field for a plane
electromagnetic wave propagating in the +z direction in a loss-free medium (𝛼 = 0) is
E = Em cos(𝜔t − 𝛽z)̂
x
H = Hm cos(𝜔t − 𝛽z)̂
y (5.86)

where ̂x, ̂y are unit vectors in the +x and +y directions, respectively, E is in units of volt per meter (V/m), H
in ampere per metre (A/m), 𝜔 is angular frequency (rad/s), 𝛽 is wavenumber (rad/m), and Em and H m are the
respective amplitudes of the fields.
Telegram: @Persian_IEEEComSoc

5.5.2 Radio Wave Propagation Modes

How a radio wave makes its journey from transmitter to receiver, i.e. its mode of propagation, very much depends
on the wave frequency. The modes of radio wave propagation in the atmosphere are depicted in Figure 5.30,
which shows how a radio wave travels from a transmitter to a receiver in five different ways, namely ground wave,
sky wave, line-of-sight (LOS), ionospheric scatter, and tropospheric scatter. Before discussing these propagation
modes, it is useful first to review the vertical structure of the atmosphere.
Figure 5.31 shows the division of the earth’s atmosphere into four shells. The troposphere is the lowest layer of
the atmosphere and contains approximately 75% of total gaseous mass and 99% of water vapour and aerosols. The
average depth of the troposphere varies with latitude, reducing from about 18 km in the tropics to as low as 6 km
 5.5 Radio 385

Ionosphere
Sky wave
Troposphere Ionospheric scatter
Tropospheric scatter
Rx
Tx Line-of-sight (LOS)
Tx ≡ transmitter
Earth Rx ≡ receiver
Ground wave

Figure 5.30 Different radio wave propagation modes in the atmosphere.

km Ionosphere or Thermosphere
6 00 (tenuous; plasma; neutral molecules)
– m
00 0k
r ~2 ~12 km
e < 0 2000 km
lay er ~9
F- -lay er < Mesosphere
E lay (Negative temp. gradient)
D -

Stratosphere 80 km
Mesopause, –90° C (ozone; positive temp. gradient)
30 km
Troposphere temp. ≡ temperature
Stratopause, 0° C
16 km
Earth
Tropopause, –55° C
(temp. inversion layer)

Figure 5.31 Vertical structure of the earth’s atmosphere (not to scale).

at the poles in winter. It has a negative temperature gradient of about −6.5 ∘ C/km and terminates in most places
with a temperature inversion layer called the tropopause, which effectively limits convection.
Above the tropopause lies the stratosphere, a region that contains most of the atmospheric ozone and extends
to about 30 km. Temperature increases with height to a maximum of about 0 ∘ C at the stratopause. The heating of
the stratopause is caused by absorption of the sun’s ultraviolet radiation by the ozone.
The mesosphere is a region of negative temperature gradient that lies above the stratopause. It extends to about
Telegram: @Persian_IEEEComSoc

80 km where temperatures reach a minimum of about −90 ∘ C.

The ionosphere or thermosphere is a region of plasma (electrons and positive ions) and large quantities of neutral
molecules extending from about 80 km to 2000 km above the earth’s surface. In this region, the sun’s radiation
is strong enough to cause the ionisation of gas molecules. Since at this height the atmosphere is rare, the rate
of recombination (of electron and ion back to neutral molecule) is very low. Thus, electrons, ions, and neutral
molecules coexist. The amount of ionisation is usually given in terms of the number of free electrons per cubic
metre.
The term homosphere is sometimes used to refer to the first 80 km of the atmosphere where atmospheric gases
are well mixed, and the term heterosphere is used to refer to the rest of the atmosphere beyond this height where
the gases tend to stratify according to weight. About 99.9998% of the atmosphere is contained in the homosphere,
of which about 75% lies in the comparatively small tropospheric volume. The ionosphere is therefore very tenuous.
 386 5 Transmission Media

Ionospheric effects (below about 3 GHz):

Faraday rotation
on
Direction of arrival variati Disp
ersio
Scint n
illatio lay
n p de ift
Grou Absorption y sh
quenc
loss r fre
ple
p ath Dop
ce
Tropospheric effects: spa
ee
iton Fr
ra c on
Ref atio
n rpti
n u n s a bso
tte tio
ase
ou tion
a in a l a risa G t t e nua
R po og
a
De &f
u d ion
Clo n t i llat
Sci

Figure 5.32 Radio wave propagation effects in the troposphere and ionosphere.

Practically all radio wave impairments in the atmosphere occur in the troposphere (due to interactions of the
radio wave with gases, hydrometeors, i.e. airborne water and ice particles, and aerosols), and in the ionosphere
(due to interactions of the radio wave with free electrons and ions). The only exception is free space path loss,
which does not require the presence of matter in the transmission medium and is in any case, strictly speaking,
not an impairment. Figure 5.32 shows a catalogue of various tropospheric and ionospheric propagation effects on
a radio link between an earth station and a satellite.

5.5.2.1 Ground Wave

Ground-wave propagation is the dominant mode of propagation below about 2 MHz. The wave follows the con-
tour of the earth by diffraction. This is the propagation mode used in amplitude modulation (AM) broadcasting.
Reliable intercontinental communication can be achieved by ground-wave communication using high transmit
power (>1 MW). The minimum usable frequency is determined by the condition that for efficient radiation the
antenna should be longer than one-tenth of the signal wavelength. For example, a 1 kHz wave has a wavelength
Telegram: @Persian_IEEEComSoc

of 300 km – from Eq. (5.81) – and therefore would require an antenna at least 30 km long to efficiently radiate it.
Energy in the surface waves is expended in setting up displacement and conduction currents in the earth. The
resulting wave attenuation increases with frequency and depends on the salinity (salt content) of the earth’s sur-
face. This attenuation is what sets the maximum usable frequency.

5.5.2.2 Sky Wave

In the sky-wave propagation mode, the atmosphere acts like a gigantic waveguide that is bounded below by the
earth’s surface and above by a layer of the ionosphere. The wave travels by repeated reflections from these two lay-
ers. It should be noted, however, that describing this phenomenon as a reflection from an ionospheric layer is an
oversimplification for convenience. What happens in practice is that the radio wave undergoes continuous refrac-
tion as it travels upwards through a decreasing refractive index profile until it reaches a layer in the ionosphere
 5.5 Radio 387

at which the condition for total internal reflection is satisfied and the wave is thereby reflected towards the earth.
Sky wave is the dominant mode of propagation for frequencies between about 2 and 30 MHz. It is the sky wave
mode of propagation that contributes to the multipath fading experienced by AM broadcasts at night when the
transmitted signal reaches a reception point via both ground wave and sky wave (after reflection from the D-layer).

5.5.2.3 Line-of-sight (LOS)

This is the dominant mode of propagation for frequencies above about 30 MHz. The wave propagates directly from
transmitter to receiver without relying on any intervening matter. The signal path must be above the radio horizon,
which differs from the geographical horizon due to atmospheric refraction. If the two antennas are located on the
earth’s surface then the maximum distance d between the transmit-antenna of height ht and the receive-antenna
of height hr for LOS to still be achieved (i.e. for the receiver to still ‘see’ the transmitter) is given by
√ √ √
d = 2Re ( ht + hr ) (5.87)

Here, Re = 8.5 × 106 m is the equivalent radius of the earth, which is 4/3 times the true mean earth radius of
6378.137 km. Using an equivalent earth radius gives correct results with the radio path treated as a straight line
from the transmitter to the receiver, whereas radio waves are bent due to refraction. We see from Eq. (5.87) that
to increase d we must increase the transmitter and/or receiver height. This partly explains why the antennas
used for radio relay links are usually located on raised towers. LOS is also the mode of propagation employed in
(unobstructed) mobile communications and satellite communications, although Eq. (5.87) is clearly not applicable
to the latter.

5.5.2.4 Satellite Communications

In satellite communications, one end of the link is a transmitting and/or receiving earth station and the other end
is an earth-orbiting spacecraft or satellite equipped with signal reception and transmission facilities. If the satellite
is placed in an eastward orbit directly above the equator at an altitude of 35 786 km then it circles this orbit in
synchrony with the earth’s rotation and therefore remains above the same spot on the earth. This special orbit is
described as geostationary, and its use allows signal transmission to and reception from the satellite by an earth
station antenna that is pointed in a fixed direction (towards the satellite). Tracking is not required for continuous
communication, which is a significant reduction in earth station complexity and costs.
A unique advantage offered by satellite communication is broad-area coverage. One geostationary satellite can be
‘seen’ from about 34% of the earth’s entire surface at elevation angles not below 10∘ , allowing reliable communica-
tions to take place between any two points within this large coverage area. In its simplest design, the satellite acts as
a repeater located in the sky. It receives a weak signal from an earth station transmitter, boosts the signal strength,
and re-transmits it back to earth on a different frequency. The original signal can then be received by one or more
earth stations. In this way, a live television broadcast of an important event can be provided to many countries using
one satellite (and of course many receiving earth stations). A relay of just three geostationary satellites allows the
Telegram: @Persian_IEEEComSoc

entire world to be covered, except for the polar regions, which cannot be seen from a geostationary orbit (GEO).
The main drawback of the GEO is that the large distance between the earth station and the satellite gives rise
to large propagation delays (≥120 ms one way) and extremely weak received signal strengths. These problems
are alleviated by using low earth orbits (LEOs), at altitudes from about 500 to 2000 km, and medium earth orbits
(MEOs), at altitudes in theory from above 2000 km to below GEO height but in practice around 19 000 to 24 000 km.
However, tracking is required for communication using one LEO or MEO satellite. Furthermore, communication
from a given spot on the earth is limited to a short duration each day when the satellite is visible. The use of a
constellation of many such satellites, equipped with intersatellite links (to facilitate seamless handover from one
satellite that is disappearing below the horizon to another visible satellite), allows continuous communication
using nontracking portable units. The Iridium satellite constellation is a good operational example of this type of
satellite communication system design.
 388 5 Transmission Media

5.5.2.5 Mobile Communications

A mobile communication link consists of a mobile unit, which is capable of large-scale motion, at one end and a
fixed base station at the other end, both equipped with radio transmission and reception facilities. An important
difference between the mobile link and a fixed satellite or radio relay link is the increased significance of multipath
propagation. The transmitted radio signal reaches the other end of the link through various paths due to reflections
from a plane earth’s surface, nearby buildings and/or terrain features and diffraction around obstacles. In fact,
the LOS path is often blocked altogether at some locations of the mobile unit. As the mobile unit moves, the
received signal strength will therefore vary rapidly in a complicated manner since the multipath components add
destructively in some locations and constructively in others.

5.5.2.6 Ionospheric Scatter

Radio waves are scattered by refractive index irregularities in the lower ionosphere. This mode occurs for fre-
quencies between about 30 and 60 MHz and can provide communication (beyond LOS) up to distances of several
thousand kilometres.

5.5.2.7 Tropospheric Scatter

Radio waves at frequencies of about 40 MHz to 4 GHz are scattered by refractive index irregularities in the tro-
posphere. This scattering can provide communication beyond the visual horizon to distances of several hundred
kilometres. Reliable communication with annual availability up to 99.9% is readily achieved over distances up
to 650 km on a single hop, or thousands of kilometres using several hops in tandem. Tropospheric scatter (or tro-
poscatter) communication has today been largely replaced by satellite communications, but it remains an attractive
method to the military because it requires transmitter–receiver alignment which makes interception of the signal
by an unauthorised entity more difficult. It has been used to provide multichannel FDM telephony (with a capacity
of 12–240 voice channels) or wideband data communication over long distances spanning large bodies of water
or other inhospitable terrain. Tropospheric scatter can, however, be a source of interference in radio relay and
satellite communication systems when two independent systems become coupled as a result of their radio paths
traversing a common scattering volume, which causes the signal of one system to be scattered into the antenna of
the other.

5.5.3 Radio Wave Propagation Effects

The presence of matter in the propagation medium and variations or discontinuities in the primary parameters
of the medium (namely permittivity, permeability, and conductivity) will affect radio waves in various ways, lead-
ing to the propagation mechanisms of reflection, refraction, diffraction, and scattering and various propagation
effects, as enumerated in Figure 5.32. It should be noted that scattering is both a mechanism (i.e. a means) of
propagation and an effect that is imposed on radio waves As the former, it enables beneficial applications such
Telegram: @Persian_IEEEComSoc

as troposcatter communication but also causes unwanted impairments such as interference. As the latter, it gives
rise to attenuation of the radio signal. We will here, however, consider only the attenuation effect of scattering
along with absorption as we briefly summarise various ionospheric and tropospheric effects on radio waves. The
propagation mechanisms of reflection, refraction, and diffraction are discussed in Sections 5.5.4–5.5.6.

5.5.3.1 Ionospheric Effects

Table 5.6 summarises the frequency dependence and maximum values of various ionospheric effects at 1, 3, and
10 GHz. The values listed are for total electron content (TEC) = 1018 electrons/m2 and path elevation = 30∘ , and
are meant to convey a rough idea of the scale of the various effects. In addition, we deduce from this table that
ionospheric effects decrease rapidly with frequency and the ionosphere may therefore be regarded as transparent
to radio waves above about 3 GHz.
 5.5 Radio 389

Table 5.6 Frequency dependence and maximum values of various ionospheric effects (assuming TEC = 1018 el/m2 , and
path elevation angle = 30∘ ).

Frequency
Effect dependence 1 GHz 3 GHz 10 GHz

Faraday rotation 1/f 2 108∘ 12∘ 1.1∘

2
Group delay 1/f 0.25 μs 0.028 μs 0.0025 μs
Refraction 1/f 2
<0.01∘ <0.001∘ <0.0001∘
Variation in direction of arrival (rms) 1/f 2 0.0033∘ 0.00037∘ 0.000033∘
Absorption (auroral and/or polar cap) 1/f 2 0.05 dB 0.0056 dB 0.0005 dB
Absorption (mid-latitude) 1/f 2 <10−2 dB <10−3 dB <10−4 dB
Dispersion 1/f 3 400 ps/MHz 14.8 ps/MHz 0.4 ps/MHz
Scintillation 1/f 1.5 Highly location dependent See Rec. ITU-R P.531 [6]

Most ionospheric effects depend on the TEC of the ionosphere, a parameter defined as the number of electrons
N T in a column of the ionosphere of cross-sectional area 1 m2 having the earth–space path as its axis.

NT = ne (s)ds (5.88)
∫s
where ne is the number of electrons per cubic metre (el/m3 ) and s is the propagation path.
Electron concentration ne in the ionosphere varies diurnally, seasonally, and with the 11-year sunspot cycle
as well as with altitude or ionospheric layers. For example, in the D-layer ne ≈ 109 el/m3 in the daytime but is
negligible at night; and in the F2-layer ne may be up to 5 × 1012 el/m3 peak in the daytime and ≈ 4 × 1011 to 5 × 1011
el/m3 at night. TEC can range from 1016 to 1018 electrons/m2 , peaking during sunlit hours. In practice, Eq. (5.88)
is difficult to evaluate to obtain TEC due to complicated variations in ne along the propagation path. However, a
good estimate of TEC can be obtained by multiplying the peak value of ne at the time of interest by an equivalent
slab thickness value of 300 km.
Faraday rotation is the rotation of the plane of polarisation of a linearly polarised radio wave propagating
through a magnetised dielectric medium. It occurs in the ionosphere due to the presence of free electrons and
the earth’s magnetic field and was discovered in 1845 by Michael Faraday (1791–1867). The presence of free
electrons in the ionosphere causes a slight frequency-dependent reduction in the phase velocity of radio waves
when compared to their speed in free space. This introduces not only a group delay but also dispersion or pulse
broadening as the higher-frequency components within a pulse of finite bandwidth will propagate slightly faster
than their lower-frequency counterparts. Variability of group delay (due to TEC fluctuation) impairs radio nav-
Telegram: @Persian_IEEEComSoc

igation or ranging systems which rely on an accurate knowledge of propagation time for reliable performance.
Ionospheric scintillation is the variation in radio signal intensity due to rapid fluctuations of signal amplitude,
phase, and direction of arrival caused by small-scale irregularities in ionisation density. Ionospheric scintillation
has a geographical, diurnal, seasonal, and solar activity dependence, with intense scintillation zones located near
the equator and at high latitudes. Ionospheric absorption is caused by the interaction of radio waves with free
electrons through induced vibrations of electrons which then transfer some of the radio energy to neighbouring
molecules upon collision with them. In this way a portion of the signal power is dissipated as heat. Ionospheric
absorption is mostly significant in the D-layer (∼ ≤ 90 km altitude) of the ionosphere where the concentration of
molecules is enough to produce a high collision rate. It is an important factor in HF propagation but (as can be seen
in Table 5.6) is negligible at radio frequencies of 1 GHz and above. Ionospheric refraction is caused by variation of
ionospheric refractive index with height as a result of which the direction of propagation of a radio wave will be
 390 5 Transmission Media

slightly changed as the wave passes through layers of different refractive indices. This effect is significant at HF
and is responsible for the sky wave mode of propagation but (as Table 5.6 shows) is negligible at 1 GHz above.

5.5.3.2 Tropospheric Attenuation

Scattering and absorption by gaseous molecules, aerosols, particulates, and hydrometeors in the troposphere cause
attenuation of the radio wave to an extent that is primarily dependent on wave frequency. In addition to attenuation
of the wanted radio signal, an absorbing medium contributes to an increase in system noise since, under thermal
equilibrium, it incoherently re-radiates the energy that it absorbs from a radio wave, and this re-radiation is picked
up as noise in a receive-antenna. A scattering medium, on the other hand, in addition to signal attenuation, will
contribute to co-channel interference if the scattered wave is picked up by other systems operating at the same
frequency. The main components of tropospheric attenuation include gaseous absorption and attenuation by dust,
fog, cloud, and rain.

5.5.3.2.1 Gaseous Absorption

A general background absorption of radio waves in the troposphere occurs at all frequencies due to energy being
expended by the electromagnetic wave in every cycle to align and realign the dipoles of water vapour molecules
and the magnetic moments of oxygen molecules. This absorption increases with frequency. In addition, there is a
spike in absorption at certain frequencies due to resonant absorption. The resonance frequencies for water vapour
are 22.3, 183.3, and 323.8 GHz; and for oxygen there is a collection of closely spaced peaks around 60 GHz and
another at 118.74 GHz. Above 100 GHz, nitrogen also makes a pressure-induced contribution to the absorption of
radio waves in the atmosphere. These gases are ever present, so allowance must always be made in a link power
budget for signal loss due to gaseous absorption in all weather. ITU-R recommendation P.676 [7] gives a proce-
dure for calculating gaseous attenuation using input parameters of wave frequency, local atmospheric pressure,
temperature and water vapour density, or relative humidity.
Figures 5.33 and 5.34 feature four curves each of atmospheric gaseous attenuation over the frequency range
1–125 GHz obtained using the ITU-R gaseous absorption model [7] at atmospheric pressure 1013.25 hectopascal
(hPa ≡ millibar, or mb) and temperatures 15 and 30 ∘ C and at two relative humidity values as specified on the plots.

A,B
16
10 C,D
Specific Attenuation, γ (dB/km)

D
1 C
B
A
Telegram: @Persian_IEEEComSoc

0.1
A: t = 15°C; RH = 58.5%; (WVD = 7.5 g/m3)
B: t = 15°C; RH = 90%; (WVD = 11.5 g/m3)
C: t = 30°C; RH = 50%; (WVD = 15.2 g/m3)
D: t = 30°C; RH = 90%; (WVD = 27.3 g/m3)
0.01

0.004
1 10 20 30 40 50 60 70 80 90 100 110 120
Frequency (GHz)

Figure 5.33 Specific attenuation due to atmospheric gases for atmospheric pressure = 1013.25 hPa. Temperature, relative
humidity (RH), and water vapour density (WVD) are as identified on each plot.
 5.5 Radio 391

A,B
160
100 C,D
Zenith Attenuation, Az (dB)

10

D
C
1 B
A
A: t = 15°C; RH = 58.54%; (WVD = 7.5 g/m3)
B: t = 15°C; RH = 90%; (WVD = 11.53 g/m3)
C: t = 30°C; RH = 50%; (WVD = 15.16 g/m3)
0.1 D: t = 30°C; RH = 90%; (WVD = 27.29 g/m3)

0.02
1 10 20 30 40 50 60 70 80 90 100 110 120
Frequency (GHz)

Figure 5.34 Zenith attenuation from sea level due to atmospheric gases. (Atmospheric pressure = 1013.25 hPa.
Temperature, relative humidity (RH), and water vapour density (WVD) are as identified on each plot).

A brief digression to explain the concept of relative humidity is in order here. The amount of water vapour,
expressed as water vapour density 𝜌 in g/m3 , that can be absorbed in the air depends on air temperature t (∘ C).
Relative humidity RH (%) is a convenient way to indicate to what extent air has approached its limit 𝜌max (t) of water
vapour content. RH = 0% means the air is completely dry, whereas RH = 100% means the air is fully saturated with
water vapour. Given one parameter (𝜌 or RH) the other may be determined from the relation
𝜌(t + 273.15) ( )
−17.502t
RH(%) = exp (5.89)
13.2449 t + 240.97
Putting RH = 100 in this equation allows us to estimate the maximum amount of water vapour that air (of
temperature t ∘ C) can absorb as
( )
1324.49 17.502t
𝜌max (t) = exp g∕m3 (5.90)
t + 273.15 t + 240.97
Back to Figures 5.33 and 5.34, curve D is typical of a warm and humid tropical climate, whereas curve A suits a
temperate region. Figure 5.33 gives the specific attenuation (dB/km) due to atmospheric gases along a horizontal
path. The total gaseous attenuation Aghor in dB on a terrestrial radio link that has a horizontal or slightly inclined
path of length l may be estimated by multiplying the specific attenuation 𝛾 read from Figure 5.33 at the link
Telegram: @Persian_IEEEComSoc

frequency by the path length in km. Figure 5.34 gives the zenith attenuation from sea level due to atmospheric
gases. The total gaseous attenuation Agslant on an earth–space path (i.e. slant path for satellite communication or
high-altitude platform system) of path elevation 𝜃 ≥ 5∘ may be determined by dividing the reading Az of Figure 5.34
by sin𝜃. That is
Aghor = l𝛾
Az
Agslant = (5.91)
sin 𝜃
5.5.3.2.2 Cloud and Fog Attenuation
Other significant sources of tropospheric attenuation are, unlike gaseous attenuation, not ever-present and include
fog, cloud, and rain attenuation. Cloud and fog attenuations are due mainly to absorption by hydrosols, which are
suspended droplets of liquid water of diameter <0.1 mm. Knowledge of the liquid water content, temperature,
 392 5 Transmission Media

elevation angle and signal frequency is sufficient to obtain a good estimate of attenuation in flog and cloud based
on the Rayleigh approximation for attenuation cross-section, which applies to drops that are small compared to
wavelength 𝜆 and is therefore valid up to around 200 GHz (where 𝜆 = 1.5 mm). Attenuation by fog is negligible
below 100 GHz since path length through fog is usually very short for earth–space paths, and fog has low liquid
water density, typically from about 0.05 g/m3 for medium fog having visibility around 300 m to 0.5 g/m3 for thick
fog of visibility around 50 m. For example, fog attenuation on a 20 GHz terrestrial link when passing through a
1 km path of thick fog is only about 0.12 dB.
Liquid water concentration in clouds varies widely according to cloud type, from around 0.05 to 2 g/m3 or a
little more. Ice particles are present in some high-level clouds such as cirrus but do not make any significant
contribution to signal attenuation. Cloud attenuation is obviously not an issue for terrestrial radio systems but
will become significant on earth–space paths at frequencies above 20 GHz. The ITU-R recommendation P.840 [8]
provides a procedure for estimating radio wave attenuation by fog and cloud.

5.5.3.2.3 Rain Attenuation

Although rain is usually present for less than 10% of the time in an average year (depending on climatic region),
rain attenuation poses the biggest obstacle to the design of high-availability satellite communication links operat-
ing at frequencies between 20 and 50 GHz. The specific attenuation 𝛾 (dB/km) in a rain volume having rainfall rate
R (mm/h) measured on the ground using averaging intervals of one minute is given by an empirical power–law
relationship involving coefficients a and b which are functions of frequency and polarisation
𝛾 = aRb dB∕km (5.92)
Values of the coefficients a and b are given in ITU-R recommendation P.838 [9]. Based on this model, Figure 5.35
shows curves of specific attenuation due to rain at various rainfall rates from 5 to 50 mm/h, frequencies from 1
to 1000 GHz and for both horizontal and vertical polarisations. We see that rain attenuation is negligible below
4 GHz, except in violent rain (R ≥ 50 mm/h). Furthermore, rain attenuation increases rapidly with frequency from
around 5 to 50 GHz, then increases less rapidly with frequency up to 100 GHz, saturates beyond 100 GHz, and
begins to decrease slightly with frequency above about 200 GHz depending on rain rate.
Once specific attenuation 𝛾 is known (from Eq. (5.92) or Figure 5.35), the total rain attenuation Arain (dB) in a
rain event may then be obtained by multiplying 𝛾 by an effective rainy path length leff (R) so that
Arain = 𝛾leff (R) (5.93)
There are challenges, however, in determining leff (R) along a path, which may be horizontal, slant, or zenith:
● What is the rain height? This is the height of the melting layer in the atmosphere beyond which atmospheric
temperature is below 0 ∘ C so water cannot exist in liquid form as raindrops. Rain height varies with geographical
location, mainly according to latitude.
Telegram: @Persian_IEEEComSoc

● What is the horizontal extent of the rain cell? In general, the higher the rain rate, the shorter its horizontal
extent. Figure 5.36 illustrates two scenarios, one in which the rain event is widespread so that, below the rain
height, the entire slant path lies within the rain volume and the other in which a portion of the slant path below
the rain height is dry.
● By what factor do we modify the physical dimensions of the rain cell to take account of the variability of rainfall
rate with distance along the path?
● How do we include the impact of rain interception aloft? Typically, rainfall rate is measured by a rain gauge at a
single location (usually an earth station) and it is entirely possible that the signal might traverse a rain cell along
its path while there is no rain at the earth station location.
The answer to each of the above questions, if it exists at all, will vary from one rain event to another. This pre-
cludes the possibility of reliably relating an arbitrary instantaneous rain rate recorded at a single location to total
 5.5 Radio 393

20 E
10 D
B C
A
Specific Attenuation, γ (dB/km)

0.1
A: Rain rate = 5 mm/h
B: Rain rate = 10 mm/h
C: Rain rate = 15 mm/h
10–2 D: Rain rate = 20 mm/h
E: Rain rate = 50 mm/h
: Horizontal polarisation
: Vertical polarisation
10–3

10–4
1 10 100 1000
Frequency (GHz)

Figure 5.35 Specific attenuation due to rain at various rainfall rates.

Rain height

Widespread rain (Lower rain rates) Convective rain (Higher rain rates)

Figure 5.36 Horizontal extent of two types of rain events.

rain attenuation along a path. We must therefore adopt a statistical approach to model leff (R), which allows good
estimates of the annual statistics of rain attenuation based on the annual statistics of rainfall rate at the geograph-
ical location of the radio system. ITU-R recommendations P.618 [10] and P.530 [11] provide such procedures for
LOS earth–space and terrestrial paths, respectively. These furnish an estimate of the rain attenuation exceeded for
p% of the time in an average year (0.001 ≤ p ≤ 5%) at any given locality based on the rain rate exceeded for 0.01% of
Telegram: @Persian_IEEEComSoc

the time at the location. In theory, by including the rain attenuation exceeded for p% of the time in the link power
budget, it is possible to design the link to achieve an availability of (100 − p)%. However, depending on climatic
region, the amount of rain attenuation may be excessive at high frequencies and small p, and this poses a signif-
icant obstacle to the design of radio communication systems operating at frequencies above 20 GHz and having
high availability ≥99.99% of an average year.

5.5.3.3 Tropospheric Scintillation

This is a nonabsorptive fluctuation of the amplitude of a radio wave when it passes through regions in the tropo-
sphere exhibiting turbulent mixing of eddies or bubbles of different refractive indices. The impact of tropospheric
scintillation is generally minor except on very low-elevation earth–space paths located especially in tropical cli-
mates and operating at frequencies above 10 GHz. However, scintillation may be enough to hamper the operation
 394 5 Transmission Media

of antenna tracking systems and rain fade mitigation protocols. A model for predicting annual scintillation fade
exceedance statistics is included in ITU-R P.618 [10].

5.5.3.4 Depolarisation
This is a modification of the polarisation state of a transmitted wave. It is usually caused by hydrometeors (rain-
drops or ice particles) in the troposphere and by reflection (as discussed later). As the size of raindrops increases,
their shape changes from perfectly spherical for small droplets to oblate spheroidal for large drops, and this gives
rise to differential attenuation of horizontally polarised and vertically polarised radio waves by rain. This phe-
nomenon is evident in Figure 5.35, where the specific attenuation for horizontal polarisation exceeds that of
vertical polarisation especially at higher rain rates (due to an increased content of larger drops in heavy rain).
If Et denotes the electric field of the transmitted radio wave that enters the rain medium and Er the electric field
of the received radio wave that emerges from the medium, as illustrated in Figure 5.37, we may resolve Et into its
horizontal and vertical components EtH and EtV as shown. The medium attenuates EtH , reducing it by some factor
to ErH at exit. However, due to differential attenuation, EtV is reduced by a smaller factor to ErV . The received wave
Er is the phasor sum of ErH and ErV , as shown in Figure 5.36. Since the ratio ErV /ErH is not the same as EtV /EtH ,
the orientation or polarisation angle of Er is different from that of Et . We may resolve Er into two components Erc
and Erx , respectively parallel and perpendicular to Et . Erc is called the co-polar component since it has the same
polarisation as the transmitted wave, whereas Erx is called the cross-polar component.
The effect of rain on radio waves is therefore both a co-polar attenuation Et /Erc , which is represented by the
specific attenuation curves in Figure 5.35, and depolarisation, which results in the generation of a cross-polar
component. Ice crystals, on the other hand, cause depolarisation through the mechanism of differential phase shift
rather than differential attenuation. Therefore, unlike the case with rain, depolarisation by ice is not accompanied
by co-polar attenuation. Depolarisation is a source of impairment in dual polarisation systems where orthogonal
polarisations of the same wave frequency are simultaneously utilised to double capacity. A cross-polar component
generated in one channel will be received as interference in an orthogonally polarised channel operating at the
same frequency.
Depolarisation is quantified by a parameter known as cross-polarisation discrimination and defined as the ratio
between the received co-polar voltage and the received cross-polar voltage when a single polarisation is transmitted
( )
Erc
XPD = 20 log10 dB (5.94)
Erx
It is desired for this XPD to be very large, ideally infinite, indicating that the cross-polar signal is negligible com-
pared to the co-polar signal. Real antennas have residual XPD values ranging from 20 to 35 dB, which indicates that
the antenna does radiate a small cross-polar component in addition to the desired polarisation. ITU-R P.618 [10]
provides a procedure for estimating the annual statistics of depolarisation on an earth–space path.
Telegram: @Persian_IEEEComSoc

Erx ≡ Cross-polar component Et

Erc ≡ Co-polar component
rx

Er
E

Et Er EtV
Polarisation change
ErV
E rc
EtH
Rain medium
ErH EtH EtV
>
ErH ErV

Figure 5.37 Effect of differential attenuation on wave polarisation.

 5.5 Radio 395

5.5.3.5 Tropospheric Refraction

The refractive index of air is only marginally larger than unity, with a value n ≈ 1 + 315 × 10−6 . It is therefore more
convenient to express tropospheric refractive index in terms of a parameter N known as refractivity and defined in
terms of n as
N = (n − 1) × 106
n = 1 + N × 10−6 (5.95)
Tropospheric refractivity depends on atmospheric pressure P (mb), water vapour pressure e (mb), and temper-
ature T (kelvin) according to
P e e
N = 77.6 − 5.6 + 3.75 × 105 2 (5.96)
T T T
and it decreases roughly exponentially with height h above sea level as
( )
h
N(h) = Ns exp − ; Ns ≈ 315; H ≈ 7.35 km (5.97)
H
If the troposphere is stratified into layers having a decreasing refractivity profile as given by Eq. (5.97) then
a radio wave that is launched in a nonzenith direction will have the upper portion of its wave front travelling
slightly faster than its lower portion since the upper portion sees a medium of slightly lower refractive index. As a
result, the wave will be tilted towards the earth at every layer and will therefore follow a curved path through the
atmosphere. The wave is said to be continuously refracted. A standard atmosphere is one that has the exponential
refractivity profile of Eq. (5.97) and a refractivity gradient G in the first 1 km above sea level of value
dN
G= = −39 (5.98)
dh
In the real atmosphere, waves follow a curved path over a real earth of average radius R ≈ 6378 km, but it is
easier to work with an equivalent scenario in which waves travel in straight lines over an equivalent earth radius
Re . It may be shown that
1 1 dn
= +
Re R dh
1 dN 1
= + 10−6 = + 10−6 G (5.99)
R dh R
The ratio Re /R is called the k-factor, which may be expressed in terms of refractivity gradient G by manipulating
the above equation as follows
1
Re =
1∕R + 10−6 G
Re 1∕R 106 ∕R
≡k= = 6
Telegram: @Persian_IEEEComSoc

R −6
1∕R + 10 G 10 ∕R + G
Substituting R = 6378 km yields
157
k= (5.100)
157 + G
In a standard atmosphere where G = −39, the k-factor is k = 4/3, which gives an equivalent earth radius
Re = 4/3 × 6378 ≈ 8500 km. A standard atmospheric profile with refractivity gradient G ≈ −39 is not always present.
A significant departure from the standard condition gives rise to various types of anomalous propagation. Examples
of these are outlined below based on the values of G and k and the maximum LOS range d between a transmitter
and a receiver of respective heights ht and hr derived in Figure 5.38 as
√ √ √
d = 2kR( ht + hr ) (5.101)
 396 5 Transmission Media

hr
dr dr (Re + ht)2 = dt2 + Re2; (Re + hr)2 = dr2 + Re2
+
dt
=
d Thus
dt Re
dt = (Re + ht)2 – Re2 = 2ht Re + ht2
Re
≈ 2ht Re, for ht ≪ Re
Re Earth dr ≈ 2hrRe, for hr ≪ Re
ht
centre
d = dt + dr = 2Re ( ht + hr )
= 2kR ( ht + hr )

Figure 5.38 Geometry and derivation of maximum LOS range.

Note that beyond this separation d, the LOS path is blocked by the bulge of a spherical earth, and that the
physical radius R of the earth has been replaced by the equivalent radius Re = kR to allow us to use the straight-line
propagation path shown. For convenience of comparison with a standard atmosphere, let us set ht = hr = 10 m, so
Eq. (5.101) gives d = 26 km in a standard atmosphere.
● If G is positive, we have what is known as a sub-refraction condition. Atmospheric refractive index increases
with height and rays are bent away from the earth’s surface. This reduces the maximum separation between
terrestrial transmitter and receiver, the wave being bent upwards beyond the height of the receive-antenna if it
is not brought closer to the transmitter. For example, if G = 157 then k = 1/2 and Eq. (5.101) gives d = 16 km for
ht = hr = 10 m, compared to 26 km in a standard atmosphere. The receiver appears to be blocked from the view of
the transmitter by a more pronounced earth curvature, and this impact may be accounted for as diffraction loss.
Sub-refractive conditions may be created by a strong decrease in temperature with height or a strong increase
in humidity with height, as may happen if cool moist air from the sea blows over a hot dry land.
● If G ≤ −100 we have super-refraction. Atmospheric refractive index decreases much more rapidly with height
than in a standard atmosphere and rays are bent more rapidly towards the earth’s surface. This increases terres-
trial propagation range since radio waves that might have propagated at a height beyond that of a distant receiver
are effectively brought closer by the increased refraction. For example, if G = −100 then k = 2.75 and Eq. (5.101)
gives d = 37.5 km for ht = hr = 10 m, compared to 26 km in a standard atmosphere. Super-refraction may, how-
ever, cause a radar beam to dip earthward below its path under a standard atmosphere and thereby miss its
target or exaggerate the actual height of a target. Super-refraction may also give rise to multipath fading if the
Telegram: @Persian_IEEEComSoc

anomalous propagation combines with ground reflections to create multiple signal paths between transmitter
and receiver.
● If G ≤ −157 we have ducting propagation. Within a layer of the atmosphere where G = −157, the k-factor is
infinite, and the wave will be bent towards the earth at just the right gradient to follow the curvature of the
earth. Note that in this case, with k = ∞, Eq. (5.101) gives d = ∞. The wave appears to be trapped within a
waveguide or duct, which may be surface-based or elevated depending on the height of the atmospheric layer
or region where the condition on G is satisfied. Ducting propagation will also occur if G < −157 in a layer or
duct (which means that the wave is bent towards the ground more sharply than the curvature of the earth)
so that the wave is seemingly confined to and guided within the duct by repeated reflections. For example, if
the layer is surface based, the wave is refracted back to earth by the layer then reflected into the layer by the
earth then refracted back to earth by the layer, and so on. A sky wave might become trapped in this way within
 5.5 Radio 397

an elevated duct and therefore pass over its intended destination leading to a blackout. Surface-based ducting
propagation might combine with ground reflections to cause severe multipath fading. By extending the wave
beyond its intended coverage range ducting propagation may cause interference to other systems.
● If G = 0, then k = 1 and there is no refraction. In this case, Eq. (5.101) gives d = 22.6 km for ht = hr = 10 m, which
is less than the LOS propagation range in a standard atmosphere.

5.5.4 Reflection and Refraction

When an electromagnetic wave is incident on an interface between two media that are characterised by different
primary parameters, the wave will be partly reflected in the first medium and partly transmitted into the second
medium. We are interested in the directions and strength of the reflected and refracted waves, and in the attenua-
tion of the wave by these mechanisms. Figure 5.39 shows a wave that is incident on a plane boundary or interface
between two media 1 and 2, characterised by parameters ni , 𝜇 i , 𝜀i , 𝜎 i , and Z i , i = 1, 2, being respectively the
refractive index, magnetic permeability, electric permittivity, conductivity, and impedance of each medium. The
following terminologies are employed in Figure 5.39 and the discussion below. The normal is a line perpendicular
to the interface at the point of incidence. A ray is the propagation vector of a wave and is a line perpendicular to
the wave front and pointing in the direction of propagation of the wave. The scattering plane or plane of incidence
is the plane formed by the incident ray and the normal. The angle of incidence 𝜃 1 is the angle between the incident
ray and the normal. The angle of reflection 𝜃 r is the angle between the reflected ray and the normal. And the angle
of refraction 𝜃 2 is the angle between the refracted or transmitted ray and the normal. The incident and reflected
rays are defined to be in medium 1, whereas the refracted or transmitted ray is in medium 2.
The directions of the reflected and refracted rays 𝜃 r and 𝜃 2 are, respectively, given by Snell’s law of reflection and
Snell’s law of refraction
𝜃1 = 𝜃r Snell’s law of reflection
n1 sin 𝜃1 = n2 sin 𝜃2 Snell’s law of refraction (5.102)
The strength of the reflected and refracted rays is given by Fresnel coefficients, which depend on the orientation
of the electric field vector relative to the scattering plane. If the E field is parallel to the scattering plane, the
Re
ay

fle
tr

Normal
cte
en
cid

d
ra
In

Medium 1: Wave fronts

θr
n1, μ1, ε1, σ1, Z1 θ1
Telegram: @Persian_IEEEComSoc

Plane boundary

θ2
Medium 2:
n2, μ2, ε2, σ2, Z2

ray
itted)
ra nsm
ed (t
ract
Ref

Figure 5.39 Incident, reflected, and refracted electromagnetic waves.

 398 5 Transmission Media

coefficients 𝜌r∥ and 𝜌t∥ , respectively, give the amplitude of the reflected wave and amplitude of the transmitted
(i.e. refracted) wave relative to the amplitude of the incident wave. If the E field is perpendicular to the scattering
plane then the applicable coefficients are denoted as 𝜌r⟂ and 𝜌t⟂ . The value of each coefficient depends on the
impedances of the two media and on the angles of incidence and refraction according to
Z1 cos 𝜃1 − Z2 cos 𝜃2 Reflection coefficient
𝜌r∥ = ;
Z1 cos 𝜃1 + Z2 cos 𝜃2 (E-field parallel)
Z2 cos 𝜃1 − Z1 cos 𝜃2 Reflection coefficient
𝜌r⟂ = ;
Z2 cos 𝜃1 + Z1 cos 𝜃2 (E-field perpendicular)
2Z2 cos 𝜃1 Transmission coefficient
𝜌t∥ = ;
Z1 cos 𝜃1 + Z2 cos 𝜃2 (E-field parallel)

2Z2 cos 𝜃1 Transmission coefficient

𝜌t⟂ = ; (5.103)
Z2 cos 𝜃1 + Z1 cos 𝜃2 (E-field perpendicular)
If the media are dielectric and nonmagnetic, the latter meaning that relative permeabilities 𝜇r1 = 𝜇 r2 = 1, the
expressions in Table 5.5 for impedance and refractive index lead to
√ √ √ √
𝜇r1 𝜇0 𝜇0 𝜇r2 𝜇0 𝜇0
Z1 = = ; Z2 = =
𝜀r1 𝜀0 𝜀r1 𝜀0 𝜀r2 𝜀0 𝜀r2 𝜀0
√ √ √ √
n1 = 𝜇r1 𝜀r1 = 𝜀r1 ; n2 = 𝜇r2 𝜀r2 = 𝜀r2
√
Z1 𝜀r2 n
= √ = 2
Z2 𝜀r1 n1
Thus, the ratio of impedances equals the inverse ratio of refractive indices and Eq. (5.103) may be expressed in
terms of media refractive indices as follows (by dividing the top and bottom of each equation by Z 1 and replacing
Z 2 /Z 1 with n1 /n2 wherever this impedance ratio occurs)
n2 cos 𝜃1 − n1 cos 𝜃2 ⎫
𝜌r∥ = ⎪
n2 cos 𝜃1 + n1 cos 𝜃2 ⎪
n cos 𝜃1 − n2 cos 𝜃2 ⎪
𝜌r⟂ = 1 ⎪
n1 cos 𝜃1 + n2 cos 𝜃2 ⎪ Nonmagnetic,
2n1 cos 𝜃1 ⎬ (5.104)
𝜌t∥ = ⎪ dielectric media
n2 cos 𝜃1 + n1 cos 𝜃2 ⎪
2n1 cos 𝜃1 ⎪
𝜌t⟂ = ⎪
n1 cos 𝜃1 + n2 cos 𝜃2 ⎪
⎭
Telegram: @Persian_IEEEComSoc

At a certain angle of incidence 𝜃 1 ≡ 𝜃 B , called the Brewster angle, the numerator of the equation on the first line
above will be zero so that 𝜌r∥ = 0. This means that when an EM wave having a parallel E field is incident at the
Brewster angle on an interface between two media it will be entirely transmitted through into the second medium
without experiencing any reflection. It can be shown (Worked Example 5.9) that, for an EM wave propagating from
medium 1 of refractive index n1 into medium 2 of refractive index n2 , the Brewster angle (i.e. angle of incidence
at which 𝜌r∥ = 0) is given by
( )
n2
𝜃B = tan−1 (5.105)
n1
You may look at the second line of Eq. (5.104) and wonder whether it is also possible for its numerator to be
zero. We show in Worked Example 5.9 that this numerator cannot be zero. This means that 𝜌r⟂ ≠ 0 and therefore
 5.5 Radio 399

(a) Incident E field (b) Reflected E field

ρr‖Ei cosϕ
Ei Scattering plane
ϕʹ ρr⊥Ei sinϕ
Eisinϕ Er

ϕ
Scattering plane
Eicosϕ

Unpolarised wave Polarised wave

(c)
θB
Medium 1

Medium 2

Figure 5.40 Effect of reflection on EM wave polarisation: (a) Incident E field; (b) Reflected E field (to scale for the case
𝜀r1 = 1; 𝜀r2 = 2; 𝜃 1 = 26∘ ); (c) Unpolarised wave incident at Brewster angle 𝜃 B .

a perpendicular E field that is incident on an interface between two media will always undergo some reflection
regardless of its angle of incidence.
An incident E field that is neither perpendicular nor parallel to the scattering plane must first be resolved into
parallel and perpendicular components before Eq. (5.104) is employed. For example, if the incident E field has
amplitude Ei and its orientation makes angle 𝜙 with the scattering plane then (from Figure 5.40a) the amplitudes
of its parallel and perpendicular components will, respectively, be

Ei∥ = Ei cos 𝜙; Ei⟂ = Ei sin 𝜙

and the reflected E field (Figure 5.40b) will have parallel and perpendicular components with respective ampli-
tudes

Er∥ = 𝜌r∥ Ei cos 𝜙; Er⟂ = 𝜌r⟂ Ei sin 𝜙

The amplitude and orientation (relative to the scattering plane) of the reflected E field are therefore
√ √
Er = Er∥2 2
+ Er⟂ = Ei 𝜌2r∥ cos2 𝜙 + 𝜌2r⟂ sin2 𝜙
Telegram: @Persian_IEEEComSoc

( ) ( )
𝜌r⟂ Ei sin 𝜙 𝜌r⟂
𝜙′ = −tan−1 = −tan−1 tan 𝜙 (5.106)
𝜌r∥ Ei cos 𝜙 𝜌r∥
Using the scattering plane as reference, the orientation 𝜙 defines the polarisation of the incident E field and
𝜙 the polarisation of the reflected E field. Eq. (5.106) therefore gives us the amplitude and polarisation of the
′

reflected wave in terms of the amplitude and polarisation of the incident wave and the reflection coefficients of
Eq. (5.104). There is an interesting special case in Eq. (5.106) that we must highlight. If the angle of incidence equals
the Brewster angle then 𝜌r∥ = 0 and the argument of the inverse tangent function in Eq. (5.106) becomes infinite,
so 𝜙′ = −90∘ whatever the polarisation 𝜙 ≠ 0 of the incident wave. An unpolarised incident wave (i.e. one having
random values of 𝜙) will therefore be converted upon reflection to a polarised wave (i.e. one having polarisation
along a fixed line, in this case perpendicular to the scattering plane) when the wave is incident on an interface
 400 5 Transmission Media

ρr‖
1

0.8
εr =
100
0.6 εr = 10
0.4

0.2 εr = 2

–0.2
–0.4

–0.6

–0.8
–1
0 10 20 30 40 50 60 70 80 90
Incidence angle (deg.) Brewster angles

Figure 5.41 Reflection coefficient for parallel E field incident from free space onto a good dielectric medium.

at its Brewster angle. This scenario is illustrated in Figure 5.40c and happens because at the Brewster angle the
parallel E field in the unpolarised incident wave is entirely transmitted into medium 2 without any reflection, so
only the perpendicular E field component is partly reflected.
Figures 5.41–5.43 show plots of the coefficients calculated using Eq. (5.103) for an EM wave incident from free
space (𝜇 r1 = 1, 𝜀r1 = 1) onto a good dielectric medium of various relative permittivity values. Figures 5.41 and 5.42
give the reflection coefficient when the incident E field is, respectively, parallel and perpendicular, and Figure 5.42
gives the transmission coefficients for both parallel and perpendicular E fields. The Brewster angle of each dielec-
tric medium is indicated in Figure 5.41. It is important to note that the Brewster angle given by Eq. (5.105) is only
applicable to a nonmagnetic dielectric medium with negligible conductivity 𝜎. If this condition is not satisfied, and
in particular if 𝜎 ≠ 0, then the exact expressions for Fresnel coefficients given by Eq. (5.103) must be used with the
impedance of each medium calculated using its exact formula in column 2 of Table 5.5. This impedance as well as
the reflection and transmission coefficients in Eq. (5.103) will in general be complex. In particular, if Z 1 is real (e.g.
if medium 1 is air) whereas Z 2 is complex then the numerator of the expression for 𝜌r∥ in Eq. (5.103), and hence
𝜌r∥ , cannot be zero. This therefore means that the medium has no Brewster angle as defined above. However, the
Telegram: @Persian_IEEEComSoc

angle of incidence at which |𝜌r∥ | attains its minimum value is called the pseudo-Brewster angle. This situation is
shown in Figure 5.44, which gives plots of the magnitude |𝜌r∥ | and angle ∠𝜌r∥ of the reflection coefficient when a
parallel EM wave at frequencies 1 GHz and 20 GHz is incident from free space onto wet ground and seawater. Note
that the conductivity and permittivity of a medium are in general frequency dependent, hence the specification
of the frequency of the wave in Figure 5.44. The pseudo-Brewster angle of each medium is identified in the figure
and ranges from 72∘ to 85∘ .
In the above discussions, we deliberately focused solely on the computation of the E field component of the
EM wave. Once the reflected and transmitted E field components of the EM wave are known, the corresponding
magnetic field components are obtained by dividing by the impedance of the medium, since H = E/Z according
to Eq. (5.82).
 5.5 Radio 401

ρr⊥
–0.1

–0.2
εr = 2
–0.3

–0.4

–0.5

–0.6 εr =
10
–0.7

–0.8 εr = 100

–0.9

–1
0 10 20 30 40 50 60 70 80 90
Incidence angle (deg.)

Figure 5.42 Reflection coefficient for perpendicular E field incident from free space onto a good dielectric medium.

ρt
0.9
E field parallel
0.8 E field perpendicular
εr = 2
0.7

0.6

0.5

0.4

0.3 εr =
10
εr = 100
0.2
Telegram: @Persian_IEEEComSoc

0.1

0
0 10 20 30 40 50 60 70 80 90
Incidence angle (deg.)

Figure 5.43 Transmission coefficient for E field incident from free space onto a good dielectric medium.
 402 5 Transmission Media

1
B
0.8
C
0.6 A
ρr‖
D
0.4

0.2
Pseudo-Brewster angle
0
0
A
–30
C B
–60
∠ρr| |, deg

A: Wet ground, 1 GHz: εr = 30; σ = 0.15 S/m D

–90 B: Sea water, 1 GHz: εr = 70; σ = 5 S/m
–120 C: Sea water, 20 GHz: εr = 30; σ = 35 S/m
–150 D: Wet ground, 20 GHz: εr = 7; σ = 6.5 S/m

–180
0 10 20 30 40 50 60 70 80 90
Incidence angle (deg.)

Figure 5.44 Magnitude and phase of reflection coefficient for parallel E field incident from free space onto the earth’s
surface.

Worked Example 5.9 Brewster Angle

An EM wave is incident on an interface between a dielectric medium of refractive index n1 and a second dielectric
medium of refractive index n2 , where n1 ≠ n2 . The two media are nonmagnetic.
(a) In terms of n1 and n2 , determine the angle of incidence 𝜃 B at which an EM wave having its E field vector
parallel to the scattering plane will experience no reflection at the interface and therefore be transmitted in its
entirety from medium 1 into medium 2.
(b) Show that an EM wave with the E field vector perpendicular to the scattering plane will always experience
some reflection at the interface regardless of its angle of incidence.
(a) What this question requires is for us to determine the angle of incidence at which the reflection coefficient
𝜌r∥ of an EM wave with a parallel E field vector is zero. Since the two media are dielectric and nonmagnetic,
the applicable Fresnel equations are Eq. (5.104). We need to determine the value of the angle of incidence 𝜃 1
(which we will denote as 𝜃 B ) at which the numerator of the expression for 𝜌r∥ in Eq. (5.104) equals zero. Thus
Telegram: @Persian_IEEEComSoc

n2 cos 𝜃B − n1 cos 𝜃2 = 0
cos 𝜃2 = (n2 ∕n1 ) cos 𝜃B
cos2 𝜃2 = (n2 ∕n1 )2 cos2 𝜃B (i)
We can eliminate 𝜃 2 from equation (i) by invoking Snell’s law of refraction (Eq. (5.102)), recalling that angle
of incidence 𝜃 1 ≡ 𝜃 B . This leads to
n1 sin 𝜃B = n2 sin 𝜃2
sin 𝜃2 = (n1 ∕n2 ) sin 𝜃B
sin2 𝜃2 = (n1 ∕n2 )2 sin2 𝜃B (ii)
 5.5 Radio 403

Adding together equations (i) and (ii)

(n1 ∕n2 )2 sin2 𝜃B + (n2 ∕n1 )2 cos2 𝜃B = sin2 𝜃2 + cos2 𝜃2

Trigonometric identities sin2 𝜃2 + cos2 𝜃2 = 1 and sin2 𝜃B + cos2 𝜃B = 1 allows us to replace the right-hand side
of the above equation with sin2 𝜃B + cos2 𝜃B . Thus

(n1 ∕n2 )2 sin2 𝜃B + (n2 ∕n1 )2 cos2 𝜃B = sin2 𝜃B + cos2 𝜃B

Dividing through by cos2 𝜃 B , substituting tan2 𝜃 B for sin2 𝜃 B /cos2 𝜃 B , and simplifying the resulting equation leads
to the desired expression for 𝜃 B as a function of n1 and n2
(n1 ∕n2 )2 tan2 𝜃B + (n2 ∕n1 )2 = tan2 𝜃B + 1

tan2 𝜃B [(n1 ∕n2 )2 − 1] = 1 − (n2 ∕n1 )2

1 − (n2 ∕n1 )2 1 − (n2 ∕n1 )2
tan2 𝜃B = = = (n2 ∕n1 )2
(n1 ∕n2 ) − 1 (n1 ∕n2 )2 (1 − (n2 ∕n1 )2 )
2

tan 𝜃B = n2 ∕n1

Therefore, we have deduced Eq. (5.105), which specifies the Brewster angle as

𝜃B = tan−1 (n2 ∕n1 )

(b) Here we need to show that the reflection coefficient 𝜌r⟂ of an EM wave with a perpendicular E field vector
cannot be zero. Again, the applicable Fresnel equations are Eq. (5.104). For 𝜌r⟂ to be zero, the numerator of
its expression in Eq. (5.104) would have to be equal to zero. That is

n1 cos 𝜃1 = n2 cos 𝜃2 (i)

But Snell’s law of refraction (which we know to be true) stipulates that

n1 sin 𝜃1 = n2 sin 𝜃2 (ii)

Dividing (ii) by (i) yields tan𝜃 1 = tan𝜃 2 or 𝜃 1 = 𝜃 2 , which (when substituted in equation (ii)) leads to n1 = n2 .
This contradicts the fact that n1 ≠ n2 , the two media being different in refractive index. We therefore conclude
that equation (i) cannot be true if n1 ≠ n2 and therefore that 𝜌r⟂ ≠ 0.

Worked Example 5.10 Wave Propagation

A linearly polarised plane wave at 5 GHz propagates in seawater along the positive z direction with an electric
field amplitude Eo = 2.4 V/m at z = 0. The electric field is parallel to the plane of incidence. Given that the primary
Telegram: @Persian_IEEEComSoc

parameters of the seawater medium at 5 GHz are 𝜎 = 10 S/m, 𝜀r = 65, 𝜇 r = 1, calculate:

(a) Impedance seen by the wave.

(b) Magnetic field amplitude H o at z = 0.
(c) Power flux density of the wave at z = 0.
(d) Attenuation constant 𝛼 and skin depth 𝛿 in the seawater.
(e) The attenuation (in dB) of the wave over 5 cm in the seawater.
(f) The amplitude and phase of the reflected E field if this wave is incident at z = 0 at an angle of incidence 𝜃 1 = 30∘
from free space onto a body of seawater having a horizontally flat and smooth surface.
Expressions for the required secondary parameters of the transmission medium are given in Table 5.5. To choose
the correct expression in this table we calculate (𝜎/𝜔𝜀)2 for the medium to determine whether it is a good dielectric,
 404 5 Transmission Media

good conductor, or just a general medium

( )2 ( )2 ( )2
𝜎 𝜎 10
= =
𝜔𝜀 2𝜋f 𝜀r 𝜀o 2𝜋 × 5 × 109 × 65 × 8.8541878128 × 10−12
= 0.3059
This value is neither much less than nor much greater than 1, so seawater is a general medium at 5 GHz and
we use the exact expressions in column 2 of Table 5.5.
(a) Impedance
( )1∕2 ( )1∕2
j𝜔𝜇 j2𝜋f 𝜇r 𝜇o
Z= =
𝜎 + j𝜔𝜀 𝜎 + j2𝜋f 𝜀r 𝜀o
( )1∕2
j2𝜋 × 5 × 109 × 1 × 1.25663706212 × 10−6
=
10 + j2𝜋 × 5 × 109 × 65 × 8.8541878128 × 10−12
( )1∕2 ( )1∕2
j39478.4176258 39478.4176258∠90∘
= =
10 + j18.0805634 20.6617224∠61.0539∘
198.69176537∠45∘
=
4.545516737∠30.5269∘
= 43.7∠14.5∘ Ω
(b) Magnetic field amplitude H o at z = 0
Eo 2.4
Ho = =
Z 43.71159∠14.5∘
= 54.91 mA∕m, with H field lagging E field by 14.5∘
(c) Power flux density Pfd of a wave is the power per unit area of the wave front and is analogous to power in a
load on a transmission line but with voltage replaced by an electric field and current replaced by a magnetic
field. More precisely, Pfd is the average value of the Poynting vector S, which is the vector product of E and
H. We will not pursue this slightly more mathematical line concerning S any further here because, although
straightforward, it requires some familiarity with vector analysis. Pfd is the product of the rms value of the
electric field, the rms value of the magnetic field, and the power factor, which is the cosine of the angle of
impedance, denoted ∠Z
Pfd = Erms ⋅ Hrms ⋅ cos(∠Z)
E H
= √o ⋅ √o ⋅ cos(∠Z)
2 2
Telegram: @Persian_IEEEComSoc

1
= (2.4)(0.054905346) cos(14.4731∘ )
2
= 63.80 mW∕m2
(d) In a lossy medium (𝜎 ≠ 0), the field amplitude decays exponentially with distance as E(z) = Eo e-𝛼z , where 𝛼 is
the attenuation constant of the medium and Eo is the amplitude of the E field at z = 0. Skin depth 𝛿 is defined
as the distance it takes for the field amplitude to drop to e−1 (≡ 36.8%) of its (initial) value at z = 0. Thus
E(𝛿) = Eo e−𝛼𝛿 ≜ Eo e−1
⇒ 𝛼𝛿 = 1
⇒ 𝛿 = 1∕𝛼
 5.5 Radio 405

Skin depth is therefore the reciprocal of attenuation constant 𝛼. We calculate 𝛼 using its expression in Table 5.5
and then take its inverse to obtain 𝛿
√ [√ ]
√ ( )2 ( )2
√ 𝜇𝜀 𝜎 𝜎
𝛼=𝜔 √ 1+ −1 ; = 0.3059 (earlier calculated)
2 𝜔𝜀 𝜔𝜀
√
9 1.256637 × 10−6 × 5.755222 × 10−10
= 10𝜋 × 10 (1.14275877 − 1)
2
= 225.72 Np∕m
1
𝛿 = = 4.43 × 10−3 m = 4.43 mm
𝛼
(e) Attenuation in dB over a distance l in a medium with attenuation constant 𝛼 is 8.686𝛼l. With l = 5 cm, we
obtain
Attenuation = 8.686𝛼l
= 8.686 × 225.72 × 5 × 10−2
= 98 dB
This is a huge attenuation. The wave’s power is reduced by a factor of more than six billion in just 5 cm. It
indicates that a 5 GHz EM wave is hardly able to penetrate and propagate in seawater. A much lower radio
frequency must be used for undersea communication.
(f) To determine the reflected E field, we note that medium 1 is free space with n1 = 1, Z 1 = 376.73 (from
Eq. (5.84)), and angle of incidence 𝜃 1 = 30∘ . Medium 2 is seawater with Z 2 = 43.71159∠14.5∘ (as earlier calcu-
lated). Also, the E field is parallel to the plane of incidence, so the applicable reflection coefficient expression
is that of 𝜌r∥ in Eq. (5.103). First, we must obtain the refractive index n2 of medium 2 using its expression in
Table 5.5 and then apply Snell’s law of refraction to calculate the angle of refraction 𝜃 2
√ [√ ]
√ ( )2
√ 𝜇𝜀 𝜎
c
n2 = = c = c
k √ 1+ +1
v w 2 𝜔𝜀
√
= 299792458 7.748457166185 × 10−16
= 8.345;
( ) ( )
n1 1
𝜃2 = sin −1
sin 𝜃1 = sin−1 sin 30∘
n2 8.345
= 3.435 ∘

The reflection coefficient now follows as

Z cos 𝜃1 − Z2 cos 𝜃2 376.73 cos 30∘ − 43.71159∠14.5∘ × cos 3.435∘
Telegram: @Persian_IEEEComSoc

𝜌r∥ = 1 =
Z1 cos 𝜃1 + Z2 cos 𝜃2 376.73 cos 30∘ + 43.71159∠14.5∘ × cos 3.435∘
326.25775 − (42.24838 + j10.905) 284.21865∠−2.19888∘
= =
326.25775 + (42.24838 + j10.905) 368.667444∠1.695∘
= 0.771∠−3.894∘
Reflected wave amplitude Ero is the product of this reflection coefficient and the incident wave amplitude Eo .
Therefore
Ero = |𝜌r∥ |Eo = 0.771 × 2.4
= 1.85 V∕m
The reflected wave is shifted in phase by −3.89∘ relative to the incident wave.
 406 5 Transmission Media

5.5.5 Rough Surface Scattering

When the boundary between two media is rough to the extent that there are significant height differentials between
close points of incidence, two situations arise: First, Snell’s law of reflection is satisfied over a broad range of angles
of reflection. Second, the reflected waves experience a broad range of phase shifts due to small differences in their
individual path lengths. As a result, reflection is no longer in a single direction with a single phase shift – a mecha-
nism described as specular reflection, which produces a coherent field. Rather, the wave is reflected through a broad
range of directions and phase shifts, giving rise to an incoherent field. This is a propagation mechanism known as
rough surface scattering or diffuse reflection, the latter description being perhaps more appropriate to avoid any con-
fusion with the fundamentally unrelated phenomenon of wave scattering in random media. Figure 5.45 illustrates
a range of surface roughness and the effect on an incident wave.
At normal incidence (𝜃 i = 0), a height differential Δh creates a path length difference 2Δh between waves
reflected at the top and bottom of the rough surface and hence a phase shift between the two waves given by

Δ𝜙 = Wave number × Path difference

2𝜋 4𝜋Δh
= × 2Δh =
𝜆 𝜆
Whether a given surface may be regarded as rough or smooth for a wave incident at some arbitrary angle 𝜃 i
depends on the size of the range of phase shifts in the reflected waves. A useful rule of thumb is to treat the surface
as smooth if Δ𝜙 is no more than about 𝜋/8 radian. To derive a general criterion for judging surface roughness, let
us consider Figure 5.46, which shows a beam incident with angle of incidence 𝜃 i on a rough surface having height
differential or roughness depth Δh. We have isolated the two rays at the extremes of the beam. The range of phase
shift in the reflected waves will be the phase difference between these two rays. Ray 1 is reflected from the top
of the surface at point D, whereas ray 2 is reflected from the bottom of the surface at point C. The relationships
among the various sides and angles in Figure 5.46 are given in the figure. The two rays have traversed the same
distance up to the wave front AB and they traverse the same path beyond point D. Therefore, the path difference
between ray 2 and ray 1 is

Δz = BC + CD − AD
2Δh 2Δh cos 𝜓 2Δh 2Δhcos2 𝜓
= − = −
sin 𝜓 tan 𝜓 sin 𝜓 sin 𝜓

Smooth Rough Very rough

Telegram: @Persian_IEEEComSoc

Figure 5.45 Effect of surface roughness on incident wave.

Ra ψ = 90 – θi
y1 A
Ra BC = CD = Δh/sin ψ
y2
BE = ED = Δh/tan ψ
θi AD = BD cos ψ
E ψ ψ
ψ = (BE + ED) cos ψ
B θi ψ D
Δh = 2Δh cos ψ/tan ψ
ψ ψ
C

Figure 5.46 Reflection of an incident wave from the top and bottom of a rough surface with height differential Δh.
 5.5 Radio 407

2Δh 2Δh
= (1 − cos2 𝜓) = sin2 𝜓
sin 𝜓 sin 𝜓
= 2Δh sin 𝜓 = 2Δh sin(90 − 𝜃i )
= 2Δh cos 𝜃i

This corresponds to a phase difference

2𝜋 2𝜋
Δ𝜙 = ⋅ Δz = ⋅ 2Δh cos 𝜃i
𝜆 𝜆
4𝜋Δh cos 𝜃i
=
𝜆
For the surface to be considered smooth, this phase difference must be less than 𝜋/2 (Rayleigh criterion) or less
than 𝜋/8 (Fraunhofer criterion). Substituting, Δ𝜙 = 𝜋/2, 𝜋/8 and making Δh the subject of the equation gives us
the following criterion for maximum roughness depth Δh. A reflecting surface is considered smooth to a radio
wave of wavelength 𝜆 and angle of incidence 𝜃 i if its height differential or roughness depth Δh satisfies

⎧ 𝜆 , Rayleigh criterion
⎪ 8 cos 𝜃i
Δh < ⎨ (5.107)
⎪ 𝜆
, Fraunhofer criterion
⎩ 32 cos 𝜃i

Worked Example 5.11 Surface Roughness

A body of water is subject to disturbances causing surface height variations of up to 20 cm. If a radio wave is
incident at angle 45∘ on this water surface, use the Rayleigh criterion to assess whether the surface is smooth or
rough at the following radio frequencies:

(a) (i) 3 GHz, (ii) 300 MHz, (iii) 300 kHz.

(b) Repeat the above assessment at near-grazing incidence 5∘ above the water surface and comment on the results.
This is a straightforward application of the first roughness criterion in Eq. (5.107). For smoothness the height
differential Δh = 20 cm of the water surface must be less than 𝜆/8cos𝜃 i . Converting the given radio frequencies to
wavelength yields 𝜆 = 0.1, 1, and 1000 m. Values of the criterion at each wavelength and angle of incidence are as
follows (noting in the case of (b) that 𝜃 i = 90–5∘ = 85∘ )

⎧1.77 cm, at 3 GHz

𝜆 ⎪
(a) 𝜃i = 45∘
8 cos 𝜃i ⎨
= 17.68 cm, at 300 MHz
⎪
⎩176.8 m, at 300 kHz
Telegram: @Persian_IEEEComSoc

⎧14.3 cm, at 3 GHz

⎪
= ⎨143 cm, at 300 MHz (b) 𝜃i = 85∘
⎪
⎩1.43 km, at 300 kHz
In the first case (𝜃 i = 45∘ ), Δh is less than the criterion only at 300 kHz; but in the second case (𝜃 i = 85∘ ),
Δh is less than the criterion at 300 kHz and at 300 MHz. We conclude that in the first case (𝜃 i = 45∘ ) the water
surface is rough at both 300 MHz and 3 GHz but smooth at 300 kHz; and in the second case (𝜃 i = 85∘ ) the water
surface is rough only at 3 GHz. In general, the nearer we approach to grazing incidence the smoother will surfaces
appear to incident radio waves. Second, lower frequency (and hence higher wavelength) EM waves are more able
to ‘smooth over’ the roughness of surfaces: a surface that is rough to a higher frequency (thereby reflecting it in a
diffuse manner) may be smooth to a lower frequency and therefore reflect it in specular manner.
 408 5 Transmission Media

5.5.6 Diffraction
Diffraction is a mechanism by which EM waves propagate into and therefore illuminate a shadow region behind
an obstruction. It is the apparent bending of a radio wave around the edge of an obstacle without the involvement
of reflection or refraction. This is illustrated in Figure 5.47. There is no other interacting matter in the vicinity
of the wave, and the absorbing screen is completely impervious to the incident wave, so any wave detected in
the shaded shadow region got there only by diffraction. The mechanism of diffraction is explained by Huygens’
principle, which states that any point on a wave front may be regarded as a secondary source of spherical wavelets.
The sum of all such wavelets yields the position of the wave front at any later time. Thus, a wave field might reach
a shadow region (which is blocked from direct view of the original source) through illumination by secondary
wavelets emanating from field points that have direct LOS of the shadow region. The field strength in the shadow
region is very much reduced and is usually expressed in terms of a diffraction loss that increases as one goes deeper
into the shadow region.
We will discuss only the simple scenario called knife-edge diffraction, which is due to a single perfectly absorbing
and infinitely thin screen. Practical situations involving buildings and hills that cannot be approximated by a
knife edge will require computer software routines implementing the uniform theory of diffraction to obtain more
accurate results.

5.5.6.1 Diffraction Configuration and Terms

To introduce the parameters involved in specifying diffraction loss, consider Figure 5.48, which features a trans-
mitter T and receiver R separated by distance d1 + d2 , with an absorbing screen located at distance d1 from T. The
absorbing screen has excess height h above the LOS path joining T and R. Note that h is the height of the top edge
of the absorbing screen measured upwards from the LOS path. If this top edge is below the LOS path then h is

Figure 5.47 Diffraction.

Shadow region:
illuminated through
diffraction
Incident
EM wave

Absorbing screen
Telegram: @Persian_IEEEComSoc

D Figure 5.48 Diffraction parameters.

B α ≡ diffraction angle
βT βR
h
θT θR
T R
A
d1 d2
 5.5 Radio 409

negative. Also, diffraction only occurs over the top edge of the absorbing screen. For the purpose of diffraction, the
bottom edge of the absorbing screen is of infinite depth.
The path difference Δd between a wave that travels directly from T to R via TAR and a wave that travels indirectly
via TBR is
√ √
Δd = d21 + h2 + d22 + h2 − (d1 + d2 )
Transmitter and receiver separations are usually much larger than the height of the obstruction, so let us impose
the reasonable conditions
h∕d1 ≪ 1
h∕d2 ≪ 1 (5.108)
A compact form for Δd may then be derived by using the well-known approximation (1 + a)n ≈ 1 + na, a ≪ 1
to obtain
Δd = d1 (1 + h2 ∕d21 )1∕2 + d2 (1 + h2 ∕d22 )1∕2 − (d1 + d2 )
≈ d1 (1 + h2 ∕2d21 ) + d2 (1 + h2 ∕2d22 ) − (d1 + d2 )
≈ h2 ∕2d1 + h2 ∕2d2
( )
h2 d1 + d2
≈
2 d1 d2
This corresponds to a phase difference 𝜙 between the two paths given by
( )
2𝜋 2𝜋 h2 d1 + d2
𝜙= Δd =
𝜆 𝜆 2 d1 d2
( ) √ 2
⎛ 2(d1 + d2 ) ⎞⎟
𝜋 2h2 d1 + d2 𝜋⎜
= = h
2 𝜆 d1 d2 2⎜ 𝜆d1 d2 ⎟
⎝ ⎠
𝜋 2
≡ v (5.109)
2
where
√
2(d1 + d2 )
v=h (5.110)
𝜆d1 d2
is known as the Fresnel–Kirchhoff’s diffraction parameter.
Another useful parameter is the diffraction angle 𝛼 indicated in Figure 5.48, which is the angular displacement
from the shadow boundary TBD to the observation point or receiver R. We may obtain an expression for 𝛼 in terms
of h, d1 , and d2 by noting from the geometry of Figure 5.48 that
Telegram: @Persian_IEEEComSoc

𝛼 = 180 − (𝛽T + 𝛽R )
𝜃T = 90 − 𝛽T ; 𝜃R = 90 − 𝛽R ; ⇒ 𝜃T + 𝜃R = 180 − (𝛽T + 𝛽R ) = 𝛼
and
tan 𝜃T = h∕d1 ; tan 𝜃R = h∕d2
Since tan 𝜃 ≈ 𝜃 for |𝜃| ≪ 1, it follows that under the assumed conditions h∕d1 ≪ 1 and h∕d2 ≪ 1, we can
replace tan 𝜃T with 𝜃T and tan 𝜃R with 𝜃R to obtain
𝛼 = 𝜃T + 𝜃R
h h h(d1 + d2 )
≈ + =
d1 d2 d1 d2
 410 5 Transmission Media

Thus, the excess height h and the Fresnel–Kirchhoff’s diffraction parameter v (Eq. (5.110)) may be expressed in
terms of the diffraction angle 𝛼 as
d 1 d2
h≈𝛼
d 1 + d2
√
2d1 d2
v≈𝛼 (5.111)
𝜆(d1 + d2 )

5.5.6.2 Fresnel Zones

When the path difference Δd between the LOS path TR and the shortest diffraction path TBR is half a wavelength
√
√
𝜆/2, then the phase difference is 𝜙 = (2𝜋/𝜆) × 𝜆/2 = 𝜋 and the diffraction parameter is v = 2𝜙∕𝜋 = 2. Under
this condition, a wave arriving at R via TBR directly cancels the direct wave. Varying the location d1 of the obstacle,
the locus of points B at which this condition is satisfied forms an ellipsoidal volume called the first Fresnel zone
(FZ), having the LOS path TR as its axis. Subsequent FZs (i.e. second, third, … and in general nth ) have annular
cross-sections and outer boundaries defined by the locus of points B that yield 𝜙 = n𝜋. These all have axis TR.
Recalling the definition of diffraction parameter v in Eq. (5.109), it follows that when 𝜙 = n𝜋, the diffraction
parameter v is given by
𝜋
𝜙 = n𝜋 ≡ v2
2
which means that the nth FZ corresponds to the diffraction parameter
√
v = 2n (5.112)
Substituting this value of v in Eq. (5.110) yields the height hn of the nth FZ as
√
n𝜆d1 d2
hn = , n = 1, 2, 3, · · · (5.113)
d 1 + d2
Figure 5.49 shows the front view of various FZs up to the fourth and their cross-sections at halfway between
transmitter T and receiver R. In the diagram, the first FZ is an ellipsoid with axis along LOS and outer bound-
ary defined by the locus of points for which the distance a1 + b1 = d1 + d2 + 𝜆/2. The second FZ is an annular
cross-section volume with axis along the LOS, inner boundary equal to the outer boundary of the first FZ, and

Cross-section at halfway
Front view between T and R
4th FZ 4th FZ
3rd FZ 3rd FZ
2nd FZ 2nd FZ
b2
Telegram: @Persian_IEEEComSoc

a2
a1 1st FZ b1

LOS 1st FZ

T R
d1 d2

Figure 5.49 Fresnel zones (FZ).

 5.5 Radio 411

outer boundary defined by the locus of points for which the distance a2 + b2 = d1 + d2 + 2 × 𝜆/2; and so on for the
other FZs.

5.5.6.3 Knife-edge Diffraction Loss

In the configuration of Figure 5.48, a vertical absorbing screen is placed between transmitter T and receiver R,
at a distance d1 from T and d2 from R. The vertical absorbing screen is of infinite extent except in the vertical
direction where it stops with knife-edge thinness at height h above the LOS path joining T to R. In the absence of
this absorbing screen, i.e. in free space, the EM transmission from T has an electric field value Eo at the receiver
R. In the presence of the screen as described, the electric field E at R relative to Eo is determined using Huygens’
principle, and is given by
1+j ∞ ( )
E 𝜋
= exp −j t2 dt
Eo 2 ∫v 2
1 + j [1 (
1
)]
= − C(v) − j − S(v) (5.114)
2 2 2
The terms C(v) and S(v) are, respectively, the Fresnel cosine integral and Fresnel sine integral given by
v ( )
𝜋 2
C(v) = cos t dt
∫0 2
v ( )
𝜋 2
S(v) = sin t dt (5.115)
∫0 2
Note, for example, that C(v) is the area under the function cos(𝜋t2 /2) between the limits t = 0 and t = v.
Figure 5.50 shows plots of the two integrals evaluated for various values of v from −10 to +10. It is interesting to
observe how both integrals converge in an oscillatory fashion towards ±0.5 as v → ±∞.

0.8

0.5
C(ʋ)

S(ʋ)

0
Telegram: @Persian_IEEEComSoc

–0.5

–0.8
–10 –8 –6 –4 –2 0 2 4 6 8 10
ʋ→

Figure 5.50 Fresnel cosine integral C(v) and Fresnel sine integral S(v).
 412 5 Transmission Media

Diffraction loss Ld is the factor by which the electric field amplitude at R has been reduced from the value Eo
that it would have had at R in free space. Expressed in dB
Ld = −20 log10 (|E∕Eo |) dB (5.116)
With the Fresnel integrals C(v) and S(v) of Eq. (5.115) evaluated at various values of the diffraction parameter
v and plotted as shown in Figure 5.50, we may obtain diffraction loss Ld (v) at any value of v by substituting these
results for C(v) and S(v) into Eqs. (5.114) and (5.116). In the design of a terrestrial radio link (e.g. radio relay system)
between a transmitter at location T and a receiver at location R, it is very important to fully understand the impact
that the presence of an obstruction inside the FZs formed between T and R will have on diffraction loss. To aid this
understanding it is useful to revisit the relationships among diffraction parameter v, excess height h, diffraction
angle 𝛼, and the respective distances d1 and d2 from obstruction to transmitter and receiver.
Let d1 = d′1 𝜆, d2 = d′2 𝜆, d′1 = pd′2 , where d′1 and d′2 are the respective distances from T and R to the obstruc-
tion in wavelength units, and p is the factor by which d1 exceeds d2 . For example, p = 1 means obstruction is
halfway between T and R. Making these substitutions in Eq. (5.113) gives height hn of the nth FZ as
√ √ √
n𝜆 ⋅ pd′2 𝜆 ⋅ d′2 𝜆 npd′2 nd′1
hn = = 𝜆 = 𝜆 , n = 1, 2, 3, · · ·
pd′2 𝜆 + d′2 𝜆 p+1 p+1
which we express in wavelength units as
√
hn nd′1
′
hn = = , n = 1, 2, 3, · · · (5.117)
𝜆 p+1
Making the same substitutions into Eqs. (5.111) and (5.110), we obtain expressions for diffraction angle 𝛼 and
height h′ (in wavelength units) of obstruction above LOS in terms of diffraction parameter v as
√
h′ = h∕𝜆 = v d′1 ∕2(p + 1)
√
𝛼 = v (p + 1)∕2d′1 (5.118)

It is important to emphasise that the underlying assumption for the validity of Eqs. (5.117) and (5.118) is that
the height of the obstruction is much smaller than distances from the obstruction to the transmitter and receiver,
√
i.e. h′ ≪ d′1 , d′2 . Also, recall that the top of the nth FZ corresponds to the diffraction parameter value v = 2n as
earlier stated in Eq. (5.112). √
If the
√ obstruction√ is halfway between T and R then p = 1 and the top of the first FZ has v = 2 and height
h′1 = d′1 ∕2 = d′2 ∕2. Since this first FZ has the LOS as its axis, it follows that its bottom point is at height h′ =
√ √
− d′1 ∕2 = − d′2 ∕2, the negative sign indicating that it is below LOS. The diameter of the nth FZ measured from
its top to its bottom in number of wavelengths is
Telegram: @Persian_IEEEComSoc

√
nd′1
D′n = 2h′n = 2 , n = 1, 2, 3, · · · (5.119)
p+1
Thus, the diameter of the FZ depends not only on the distance of the obstruction from the transmitter but also
on its relative separation from the receiver. For example, if the √
obstruction is equidistant from the transmitter and
the receiver then p = 1 and the diameter of the first FZ is D′1 = 2d′1 , but if the obstruction is closer to the receiver
√ √
by a factor of p = 3 then D′1 = d′1 , whereas if it is closer to the transmitter by a factor p = 1/3 then D′1 = 3d′1 .
 5.5 Radio 413

−2
0
FZ7−
FZ6−
FZ5−
FZ4−
FZ3−

FZ2−

FZ1−

FZ1+

FZ2+
FZ3+
FZ4+
FZ5+
FZ6+
FZ7+

FZ8+
5
Diffraction loss, Ld(ʋ) (dB)

10

15

20

25
−4 −3 −2 −1 0 1 2 3 4
Diffraction parameter, ʋ

Figure 5.51 Knife edge diffraction loss.

We are now ready to examine the diffraction loss Ld (v) plotted in Figure 5.51. Using Eq. (5.112), the top (+) and
bottom (−) of the first eight FZs have been indicated on the plot at
√
v = ± 2n, n = 1, 2, 3, 4, 5, 6, 7, 8
= ±1.414, ±2, ±2.45, ±2.828, ±3.162, ±3.464, ±3.742, ±4
Three special cases merit attention:
√
● At v = 2, Ld (v) = 16.3 dB. To link this value of√ v to the height of the obstruction above or below LOS, we sub-
√
stitute v = 2 into Eq. (5.118) and obtain h = d′1 ∕(p + 1), which, from Eq. (5.117) with n = 1, is the excess
′

height of the first FZ. Therefore, this is a case where the entire first FZ is obstructed, as illustrated in Figure 5.52.
Geometrical optics would suggest that there is zero field (i.e. infinite loss) at the receiver R, which is in a shadow
region, but we see that through the mechanism of diffraction there is some reception at R with an extra (diffrac-
tion) loss of 16.3 dB on top of free space path loss (which is discussed in the next section).
Telegram: @Persian_IEEEComSoc

● At v = 0, Ld (v) = 6 dB. From Eq. (5.118), this value of v corresponds to excess height h′ = 0, which means that the
top edge of the obstruction is flush with the LOS path, and hence that the bottom half of the first FZ is obstructed
as illustrated in Figure 5.53. This scenario introduces an extra (diffraction) loss of 6 dB on top of free space path
loss.
● At v = −0.778, Ld (v) = 0 dB. From Eq. (5.118), this value of v corresponds to a downward gap (which we denote
as r c in Figure 5.54) between the LOS and the top edge of the obstruction given by
√
rc = 0.778 d′1 ∕2(p + 1)
 414 5 Transmission Media

First Fresnel zone

T R

Figure 5.52 Obstruction of just the entire first Fresnel zone leads to a loss of 16.3 dB.

First Fresnel zone

T R

Figure 5.53 Obstruction of just the bottom half of the first Fresnel zone produces a 6 dB loss.

resn el
First F
D1ʹ/2
zone

LOS
rc
D1ʹ/2
rb rb = 0.225D1ʹ

T R
Telegram: @Persian_IEEEComSoc

Figure 5.54 Keeping the shaded portion of the first Fresnel zone clear of obstruction eliminates diffraction loss.

Using Eq. (5.119) for the diameter D′1 of the first FZ

√ √
′
d 1 d′1 D′
D′1 = 2 ; ⇒ = √1
(p + 1) 2(p + 1) 2 2

allows us to express r c in terms of D′1 as

0.778D′1
rc = √ = 0.2751D′1 (5.120)
2 2
 5.5 Radio 415

We illustrate in Figure 5.54 that the obstruction penetrates into and blocks the bottom portion (denoted r b ) of
the first FZ, where
rc + rb = D′1 ∕2
rb = D′1 ∕2 − rc = D′1 ∕2 − 0.2751D′1
= D′1 (0.5 − 0.2751)
= 0.225D′1
Therefore, when only the bottom 22.5% of the first FZ is obstructed, diffraction loss Ld (v) = 0 dB and path loss
is the same as in free space. Link designers therefore aim to ensure a minimum clearance that keeps the shaded
region in Figure 5.54 free of any obstruction or structure. This is usually achieved through careful choice of a link
route and by raising the heights of the antennas to achieve the required clearance since some of the intervening
structures are typically permanent fixtures (such as buildings, trees, or hills) over which the link designer has little
control.
Good estimates of diffraction loss may be obtained using the following functions that have been fitted to the
non-oscillatory regions of Ld (v) for v > −0.8
⎧
⎪−20 log(0.5 − 0.62v), −0.8 < v ≤ 0
⎪−20 log(0.5e−0.95v ), 0≤v<1
Ld (v) ≈ ⎨ √ (5.121)
2
⎪−20 log[0.4 − 0.1184 − (0.38 − 0.1v) ], 1 ≤ v < 2.4
⎪−20 log(0.225∕v), v ≥ 2.4
⎩

Worked Example 5.12 Diffraction Considerations in Link Design

A point-to-point microwave radio link at 0.3 GHz is to be set up between two sites separated by 20 km. There is a
roof structure of height 25 m halfway along the direct path between the two antennas, and this can be approximated
as a perfectly absorbing thin knife edge.
(a) Assuming a flat earth between the two locations and ignoring all other terrain effects, determine the minimum
antenna height needed to ensure adequate clearance for a free space path loss condition between the two
locations.
(b) If the antennas have height 24 m, determine diffraction loss on the link.
First, we need the wavelength at 0.3 GHz
c 3 × 108
𝜆= = = 1m
f 0.3 × 109
(a) For free space path loss condition, the scenario is as earlier discussed in Figure 5.54. We assume equal heights
ht and hr for both antennas. With a flat earth, it means that the height of LOS equals antenna height, and this
Telegram: @Persian_IEEEComSoc

equals the clearance r c plus obstruction (i.e. building) height hobs . We employ Eq. (5.120) for r c and Eq. (5.119)
for the diameter of the first FZ (remembering to convert from wavelength units and noting that the obstruction
is equidistant from transmitter and receiver, so p = 1 and d1 = 10 km)
ht = rc + hobs = 0.275D1 + 25
√
d′1
= 0.275 × 2𝜆 + 25
p+1
√
d1
= 0.55𝜆 + 25, (since p = 1; d′1 = d1 ∕𝜆)
2𝜆
√
d1 𝜆
= 0.55 + 25
2
 416 5 Transmission Media

h=1m

25 m
24 m

Building
T R
d1 + d2 = 20 km

Figure 5.55 Worked Example 5.12.

√
10000 × 1
= 0.55 + 25, (d1 = 10, 000 m; 𝜆 = 1 m)
2
= 63.89 m
(b) The scenario is as illustrated in Figure 5.55, which shows that the obstruction has excess height h = 1 m.
Converting this to diffraction parameter v using Eq. (5.117) yields
h∕𝜆 h∕𝜆 2h
v= √ = √ = √
′
d1 ∕2(p + 1) d 1 ∕4𝜆 d1 𝜆
2×1
= √
10000 × 1
= 0.02
The second expression in Eq. (5.121) is applicable and yields diffraction loss
Ld = −20 log(0.5e−0.95v ) = −20 log(0.5e−0.95×0.02 )
= 6.2 dB

5.5.7 Path Loss

In addition to signal attenuation due to the various radio wave propagation effects and mechanisms discussed
above, path loss must be included in the determination of received signal power in every radio system where the
Telegram: @Persian_IEEEComSoc

wave fronts are not confined within an artificial or natural waveguide. Below, we identify two different scenarios
and analyse the resulting path loss in each case. The first is a free space propagation scenario where the presence of
matter (including the earth’s surface) may be ignored. The second is the so-called plane earth propagation scenario
where the transmit- and receive-antennas are both sufficiently nondirectional and close to the surface of the earth
that the influence of a plane earth surface (by way of contribution through reflection) must be included in the
determination of path loss. We also briefly examine the complications introduced into path loss levels or signal
strength in a terrestrial cellular radio system.

5.5.7.1 Free Space Path Loss

During LOS radio wave propagation in free space, such as on terrestrial radio relay links or on earth–space slant
paths (both links having adequate clearance to avoid multipath or diffraction, as earlier discussed), the signal
 5.5 Radio 417

strength grows weaker with distance. This reduction in signal strength is described as free space path loss (FSPL)
and occurs because the (unguided) radio energy is spread over the surface of an imaginary sphere of radius equal
to the distance d from the transmitter. And as d increases, the signal power per unit area or power flux density
decreases in proportion to d2 . FSPL is not a propagation impairment but rather a fundamental consequence of
EM wave propagation that occurs even in a vacuum and must be included in the determination of received signal
power in every unguided radio wave transmission. We wish to derive an expression for the ratio between trans-
mitted power Pt and received power Pr of a radio wave propagating in free space. This expression being a ratio of
powers, we will work with normalised powers (where medium impedance Z is set to unity) since, provided we are
consistent at both ends, the impedances cancel out and their value is immaterial.
Consider an isotropic antenna that radiates a (sinusoidal) radio wave of amplitude At equally in all directions.
The (normalised) power of the transmitted wave is Pt = A2t ∕2 and at a distance d from the transmitter this power
will be uniformly distributed over the surface of an imaginary sphere of area 4𝜋d2 centred on the transmitter so that
a receiver at this distance d will see a power flux density Pfd = Pt /4𝜋d2 and therefore will collect power Pr = Pfd Ae ,
where Ae is the effective aperture or collection area of the receive-antenna, which for an isotropic antenna receiving
a radio signal of wavelength 𝜆 is given by [12], Ae = 𝜆2 /4𝜋. The received power is therefore
Pt 𝜆2
Pr = Pfd Ae = 2
⋅
4𝜋d 4𝜋
Pt Pt
= 2
≡ (5.122)
(4𝜋d∕𝜆) L s
We see that the transmitted power is reduced by a factor of (4𝜋d/𝜆)2 by the time it reaches the receiver. This
factor is the free space path loss Ls , which is the ratio between transmitted and received radio powers when both
antennas are isotropic and in free space. By substituting 𝜆 = c/f , we may express Ls more conveniently in dB in
terms of frequency f in GHz and distance d in km as follows
( )2 ( )
P 4𝜋d 4𝜋d
Ls = t = 10 log10 = 20 log10 dB
Pr 𝜆 𝜆
( )
f
= 20 log10 4𝜋dkm × 1000 ×
c
( )
fGHz × 109
= 20 log10 4𝜋dkm × 1000 ×
3 × 108
= 20 log10 (4𝜋 × 104 ∕3) + 20 log10 (dkm ) + 20 log10 (fGHz )
This gives the important formula for FSPL
Ls = 92.44 + 20 log10 (fGHz ) + 20 log10 (dkm ) dB (5.123)
Note that the added constant 92.44 only has this value when the units of frequency and distance are, respectively,
GHz and km, as explicitly indicated. The received power Pr is proportional to 1/d2 (i.e. it follows an inverse-square
Telegram: @Persian_IEEEComSoc

law dependence on distance). Notice also that the distance-dependent term in the logarithmic expression for Ls
contributes dB values of 0, 20, 40, … at dkm = 1, 10, 100, …, km, respectively, which indicates that Ls increases
by 20 dB for every factor of 10 increase in distance. This is usually expressed as 20 dB/decade. Alternatively, the
contribution is 0, 6, 12, 18, …, dB at 1, 2, 4, 8, …, km, respectively, which is a loss of 6 dB per factor of 2 increase
in distance or 6 dB/octave.
Expressing Eq. (5.122) in terms of transmitted and received amplitudes At and Ar gives
Pr A2 ∕2 1
= r2 =
Pt At ∕2 (4𝜋d∕𝜆)2
( )2 ( )2
Ar 1
=
At 4𝜋d∕𝜆
 418 5 Transmission Media

Hence
𝜆
Ar = At ⋅ (5.124)
4𝜋d
which shows that the signal amplitude is inversely proportional to distance as expected if power follows an inverse
square law dependence on distance.
In practice, directional (i.e. nonisotropic) transmit- and receive-antennas of respective gains Gt and Gr are
used. Fed with power Pt (from a high-power amplifier), an isotropic antenna radiates Pt equally in all directions,
whereas a directional antenna acts preferentially and significantly concentrates its radiation in some directions
while severely depleting it in others. More specifically, a directional transmit-antenna of gain Gt radiates power
Pt Gt in its direction of maximum radiation (called boresight) along which the receiver is assumed to be located.
The product of amplifier output power Pt and transmit-antenna gain Gt is known as the effective isotropically
radiated power (EIRP). A receive-antenna of gain Gr increases its power collection capacity by a factor Gr when
compared to an isotropic antenna since its effective aperture increases from Ae = 𝜆2 /4𝜋 for an isotropic antenna
to Ae = Gr 𝜆2 /4𝜋. With these directional antennas, the received power Pr is therefore increased by the factor Gt Gr
so that Eq. (5.122) yields
( )
Pr 𝜆 2 Gt Gr
= Gt G r = (5.125)
Pt 4𝜋d Ls
This is the so-called Friis equation, which relates received power Pr and transmitted power Pt in free space
propagation. In logarithmic units this reads
Pr = Pt + Gt + Gr − Ls
= EIRP + Gr − Ls (dBW) (5.126)
where Gt and Gr are transmit- and receive-antenna gains in dB, Ls is FSPL in dB given by Eq. (5.123), Pt is the
transmitted power in dBW, and EIRP = Pt + Gt is the effective isotropically radiated power in dBW. EIRP is the
amount of input power that an isotropic antenna would require in order to emit the same power as radiated along
its boresight by a directional antenna of gain Gt supplied with input power Pt . Note that the dBm unit is sometimes
used in place of dBW, and the two units are related by adding 30 to a dBW value to convert it to a dBm value. For
example, 20 dBW = 50 dBm.

5.5.7.2 Plane Earth Propagation Path Loss

The free space propagation scenario analysed above will not apply when transmission is between two nondirec-
tional antennas near the earth. The simplest possible scenario that takes account of the earth’s influence on the
received signal is illustrated in Figure 5.56a. A transmit-antenna of height ht is separated from the receive-antenna
of height hr by distance d, which is much larger than ht and hr but small enough for us to ignore the curvature of
the earth and therefore treat the ground between the two antennas as a flat surface, hence the term plane earth.
Telegram: @Persian_IEEEComSoc

The receiver receives both a direct wave (called the primary signal) that has travelled through distance dp = AC and
a ground reflected wave (called the secondary signal) that has travelled through a longer distance ds = AG + GC.
The received signal is therefore the sum of these two signals, and we determine its amplitude below using the
method of the addition of two sinusoidal signals discussed in Section 2.7.3. First, we calculate the path difference
Δd between the primary and secondary paths; then we convert this to a phase difference 𝜙 between the primary
and secondary signals, being careful to include the phase shift effect of reflection on the secondary signal; and
finally, we determine the resultant received amplitude and compare it to the transmitted amplitude to obtain an
expression for the plane earth propagation path loss.
 5.5 Radio 419

A
Primary signal
Transmit- C
Seco dp Receive-
antenna ht ndar
y sig hr antenna
nal
(a)
G
Image ds
Tx
d

A
ht – hr dp
C
ht F
hr d θ θr hr
(b) d1 i d2
G D
ht ds
θi = θr implies ht /d1 = hr /d2 = hr /(d – d1)
Hence, d1 = dht /(ht + hr)
B tan θi = d1/ht =
θi = tan–1[d/(ht + hr)]

Figure 5.56 (a) Scenario for plane earth propagation path loss computation; (b) Geometry for analysis.

From the geometry of Figure 5.56b, AG = BG (by similar triangles), so the secondary and primary path lengths
are given by
√
ds = AG + GC = BGC = (FC)2 + (FB)2
√
= d2 + (ht + hr )2
√
dp = (FC)2 + (FA)2
√
= d2 + (ht − hr )2
which, using the approximation (1 + a)n ≈ 1 + na, a ≪ 1, leads to a path difference Δd and corresponding phase
difference Δ𝜙 given by
√ √
Δd = ds − dp = d2 + (ht + hr )2 − d2 + (ht − hr )2
Telegram: @Persian_IEEEComSoc

( [ ] )1∕2 ( [ ] )1∕2
ht + hr 2 ht − hr 2
=d 1+ −d 1+
d d
( [ ]2 ) ( [ ]2 )
1 ht + hr 1 ht − hr
≈d 1+ −d 1+ , d ≫ ht , hr
2 d 2 d
{ } { }
2 2 2 2
d ht + hr + 2ht hr − (ht + hr − 2ht hr ) d 4ht hr
≈ ≈
2 d2 2 d2
 420 5 Transmission Media

2ht hr
≈
d
2𝜋 4𝜋ht hr
Δ𝜙 = Δd ⋅ ≈
𝜆 d𝜆
The phase difference 𝜙 between the primary and secondary signals is the sum of Δ𝜙 and the phase shift 𝜃 added
to the secondary signal due to reflection. Thus
4𝜋ht hr
𝜙=𝜃+ (5.127)
𝜆d
Denoting the amplitude of the transmitted signal as At , the amplitude Ap of the primary signal (after propagating
through distance dp ) and the amplitude As of the secondary signal (after propagating through distance ds and being
subjected to a reflection with coefficient 𝜌) are given by Eq. (5.124) as
𝜆 𝜆
Ap = At ≈ At
4𝜋dp 4𝜋d
𝜆 𝜆
As = |𝜌|At ≈ |𝜌|At (5.128)
4𝜋ds 4𝜋d
Therefore, we have a primary signal with amplitude Ap and phase zero (since this signal is the reference), and
a secondary signal of amplitude As and phase 𝜙, so that the resultant (received signal) amplitude is
√
Ar = (Ap + As cos 𝜙)2 + (As sin 𝜙)2
√
= A2p + A2s + 2Ap As cos 𝜙 (5.129)

At large d, reflection in Figure 5.56a occurs at near-grazing incidence. This means that the angle of incidence is
nearly 90∘ , and we see from Figures 5.41 and 5.42 that (irrespective of wave polarisation) reflection coefficient 𝜌 ≈
−1 ≡ |𝜌|∠𝜃 = 1∠180∘ . To see this, note the derivation given in Figure 5.56b, where we set the angle of incidence
equal to the angle of reflection, and this leads to the angle of incidence being given by

𝜃i = tan−1 [d∕(ht + hr )] (5.130)

If, for example, ht = 10 m and hr = 2 m, then a separation d of 1 km and 3 km between transmitter and receiver
gives respective angles of incidence 𝜃 i = 89.3∘ and 89.8∘ , which for wet ground with 𝜀r = 30, 𝜎 = 0.12 S/m at
900 MHz, gives respective reflection coefficients 𝜌 = 0.9955∠179.90∘ and 0.9987∠179.99∘ for a perpendicular E
field and 𝜌 = 0.8725∠179.69∘ and 0.9618∠179.91∘ for a parallel E field.
Thus, Eqs. (5.127)–(5.129) simplify as follows to yield Ar in terms of At
√
Ar = A2p + A2s + 2Ap As cos 𝜙
√
Telegram: @Persian_IEEEComSoc

= (At 𝜆∕4𝜋d)2 + (At 𝜆∕4𝜋d)2 + 2(At 𝜆∕4𝜋d)2 cos(𝜋 + 4𝜋ht hr ∕𝜆d)

√
𝜆
= At 2 − 2 cos(4𝜋ht hr ∕𝜆d), (since cos(𝜋 + 𝜑) = − cos 𝜑)
4𝜋d
√
𝜆
= At 4sin2 (2𝜋ht hr ∕𝜆d), (since 1 − cos(2𝜑) = 2sin2 𝜑)
4𝜋d
𝜆
= At 2 sin(2𝜋ht hr ∕𝜆d)
4𝜋d
𝜆
≈ At 2 × 2𝜋ht hr ∕𝜆d, (since sin 𝜑 ≈ 𝜑, 𝜑 ≪ 1)
4𝜋d
hh
≈ At t 2 r
d
 5.5 Radio 421

Noting that the ratio of powers equals the square of the ratio of amplitudes, and including the increase in received
power due to transmit- and receive-antenna gains Gt and Gr , we obtain
Pr h2 h2r
= Gt Gr ⋅ t 4 (5.131)
Pt d
In logarithmic units, the received signal power is given by
Pr = Pt + Gt + Gr + 20 log(ht ) + 20 log(hr ) − 40 log(d)
≡ EIRP + Gr − Lp (5.132)
where
Lp (dB) = 10n log(d) − 20 log(ht ) − 20 log(hr ), n=4 (5.133)
is referred to as the plane earth path loss in analogy with Ls in Eq. (5.123), which represents FSPL. Plane earth
path loss is the ratio between the transmitted and received radio powers when both antennas are isotropic and
wave propagation is influenced only by a plane earth. In this scenario, the received power at large distances from
the transmitter decreases rapidly with distance as 1/d4 . In other words, the path loss exponent is n = 4, which
means that path loss (as a ratio) increases as the fourth power of distance, so that there is an extra 12 dB loss
every time the distance is doubled. This represents a more rapid decay of signal strength with distance than in
free space propagation where n = 2. Plane earth propagation path loss increases with distance by 40 dB/decade or
12 dB/octave, whereas FSPL increases with distance by 20 dB/decade or 6 dB/octave.

5.5.7.3 Terrestrial Cellular Radio Path Loss

Mobile communication with smartphones and portable devices in population centres around the world is typically
delivered in a terrestrial microcellular wireless (i.e. radio) environment where transmitter–receiver separation dis-
tances are small and there are other sources of a secondary signal (i.e. multipath) in addition to a flat earth. The
transmitter could be a base station serving a small geographical area or cell and the receiver could be an active
smartphone within the cell, or vice versa. Considering small distances d and only the direct signal and one sec-
ondary signal due to reflection from a flat earth, Figure 5.57 shows the exact path loss calculated without using
any of the approximations applied at large distances to derive Eq. (5.133). The link and environment parameters
used to obtain Figure 5.57 include a wave frequency of 900 MHz, transmit- and receive-antenna heights of 10 and
2 m, respectively, and wet ground of conductivity 𝜎 = 0.12 S/m and relative permittivity 𝜀r = 30 as the intervening
flat earth surface between the transmitter and receiver. We see that at distances close to the sources of the pri-
mary and secondary signals, there is a rapid fluctuation of path loss and hence of signal strength with distance, a
phenomenon called fast fading. That Figure 5.57 shows that these fluctuations are stronger in an EM wave whose
E field polarisation is perpendicular to the scattering plane than in one with a parallel E field is simply a mani-
festation of a stronger reflection of the former by wet ground at the angles of incidence involved in the range of
Telegram: @Persian_IEEEComSoc

distances d plotted. A stronger secondary signal produces deeper fades (i.e. higher path loss) at those locations
where it directly cancels the primary signal and larger enhancements (and hence lower path loss) at locations
where it adds in phase with the primary signal. At large d, reflection is at near-grazing incidence where both
polarisations have reflection coefficients ≈ −1 so that their path losses are identical. In this microcellular environ-
ment, a small change in distance can result in a very large change in received signal strength, perhaps by up to
several tens of dB. For example, for the system configuration used in Figure 5.57, a receiver located at 32 m from
the transmitter will experience around 20 dB more attenuation when moved 3.8 m (equal to 11.4 wavelengths)
closer to the transmitter. A mobile receiver will therefore experience time-selective fading as its distance from the
transmitter changes with time.
The phase relationship between the primary and secondary signals (and hence whether they add constructively
or destructively) at a given location depends on frequency. Figure 5.58 shows the path loss at distance 119 m from
 422 5 Transmission Media

100
de
ca
/de
dB
40
90

E⊥
Path loss, dB

80

70
E‖ ade
B/dec
20 d

60

50
10 20 30 40 50 70 100 120 150 200
Transmitter-receiver separation distance, d (m)

Figure 5.57 Plane earth propagation path loss at short distances from the transmitter for EM wave with perpendicular E⟂
and parallel E|| fields. (Frequency = 900 MHz; transmit- and receive-antenna heights = 10 m, 2 m; and Conductivity and
relative permittivity of wet ground: 𝜎 = 0.12 S/m, 𝜀r = 30.)

100

95
Path loss, dB

90

85
Telegram: @Persian_IEEEComSoc

80

75
800 820 840 860 880 900 920 940 960 980 1000
Frequency, MHz

Figure 5.58 Attenuation distortion due to variation of plane earth propagation path loss with frequency at a short distance
d from the transmitter. (d = 119 m; transmit- and receive-antenna heights = 10 m, 2 m; Conductivity and relative permittivity
of wet ground: 𝜎 = 0.12 S/m, 𝜀r = 30; and E field of EM wave perpendicular to scattering plane.)
 5.5 Radio 423

the transmitter of the above system at various frequencies from 800 MHz to 1 GHz. The path loss, and hence chan-
nel gain, is clearly frequency-dependent, so the microcellular propagation channel will exhibit both attenuation
and phase distortions, which will require the use of equalisers for mitigation, as discussed in Section 4.7.4.
Furthermore, in addition to the ground reflection, other fixed and moveable objects in the environment (e.g.
trees, buildings, lamp posts, moving vehicles, etc.) will contribute multiple secondary signals having randomly
changing amplitude and phase to the summation to produce the resultant received signal at any given location and
time. The received signal envelope therefore becomes a random variable which may be statistically characterised
by a Rician distribution (in the presence of direct LOS to the transmitter) or a Rayleigh distribution (if this LOS is
blocked). Details of this characterisation may be found in Section 3.4. The Rayleigh distribution is more likely to
be applicable in a larger cell, or macrocell, where separation distances between transmitter and receiver may be
large and the blocking of direct LOS by a building or other obstruction is highly probable. The Rician distribution,
on the other hand, is more likely to be applicable in a microcell, where distances are small and the presence of a
clear LOS is highly probable. But whether the received signal envelope is Rician or Rayleigh distributed, reception
may now be characterised in terms of a local mean received signal power, or equivalently a mean path loss Lp ,
averaged in a small locality of distance d from the transmitter.
Various approaches have been proposed to estimate mean path loss in this practical scenario. In one approach,
an empirically determined factor (called the clutter factor) 𝛽 is introduced into the plane earth path loss formula
of Eq. (5.133). The clutter factor depends on frequency and the type of environment, but not on the distance from
the transmitter. An example of such formulation is the semi-empirical path loss model by Ibrahim and Parsons
[13], derived from extensive measurements in London at frequencies 168, 445, and 896 MHz, which gives
Lp (dB) = 40 log10 (d) − 20 log10 (ht ) − 20 log10 (hr ) + 𝛽
fMHz
𝛽 = 20 + + 0.18L − 0.34H + K
{ 40
0.094U − 5.9, Highly urbanised areas
K= (5.134)
0, Otherwise
where d ≤ 10 000 m is the distance between base station antenna of height ht and mobile antenna of height
hr ≤ 10 m; L is the land usage factor that specifies the percentage of the 500 m × 500 m test square that is covered
by buildings of whatever height; U is the degree of urbanisation factor which specifies the percentage of building
site area in the test square occupied by buildings having four or more floors; H is the difference between the
average ground heights of the test squares containing the transmitter and receiver; and f MHz is the signal frequency
in MHz.
In another approach to determine mean path loss, free space propagation with a path loss of 20 dB/decade is
assumed up to a break point at a distance d1 from the transmitter, and then a path loss of 10n dB/decade is assumed
beyond d1 up to the receiver location at distance d from the transmitter. The value of the path loss exponent n is
Telegram: @Persian_IEEEComSoc

determined to fit measurements and may be in the range n = 3.5 to 4.5. Since a decade in this context means a
factor of 10, and the number of factors of 10 in the range from 1 to d1 is log10 (d1 ), whereas the number of factors
of 10 in the range from d1 to d is log10 (d/d1 ), it follows that this model gives path loss as
Lp = 92.44 + 20 log10 (fGHz ) + 20 log10 (d1km ) + 10n log10 (d∕d1 )
= 92.44 + 20 log10 (fGHz ) + 10 n log10 (dkm ) − 10(n − 2)log10 (d1km ) (5.135)
where f GHz is signal frequency in GHz and dkm and d1km are respective distances in km from the transmitter to the
break point and the receiver.
 424 5 Transmission Media

5.5.8 Radio Frequency Allocation

In order to avoid interference, and for the efficient exploitation of the radio spectrum, it is necessary for appropriate
frequency bands to be allocated and agreed for different services. Not all frequency bands are suitable for certain
services. For example, the medium frequency band (MF, 300 kHz–3 MHz) cannot be used for cellular mobile com-
munication services to portable devices not just because of insufficient bandwidth but because MF requires bulky
antennas of size in excess of 10 m. We have seen that, depending on the frequency band and hence mode of propa-
gation, radio waves can travel over a range of distances. Without some regulation and control over the frequencies
used for radio communication, there would be both national and international chaos in the utilisation of radio.
On a global scale, frequency allocation and technical standards are recommended by the International Telecom-
munication Union (ITU), which is a specialised agency of the United Nations with administrative headquarters in
Geneva, Switzerland. The ITU is divided into three sectors. The radio communication sector (ITU-R) is responsible
for frequency allocation, the telecommunications standardisation sector (ITU-T) recommends global standards for
telecommunication, and the telecommunication development sector (ITU-D) provides technical assistance, espe-
cially to developing countries. To help it manage the global radio spectrum, the ITU divides the world into three
ITU regions with each region having its own set of frequency allocations. ITU Region 1 comprises Europe, Africa,
Middle East, the former Soviet Union (FSU), and Mongolia. ITU Region 2 comprises North America, includ-
ing Greenland, South America, and the eastern Pacific Islands. ITU Region 3 comprises Asia-Pacific, including
Oceania and non-FSU Asia.
Inaugurated on 1st January 1866 with 16 European countries, the ITU now has 193 member nations covering
every part of the world. The newest member nation is South Sudan, which joined on 3rd October 2011. Member
nations retain autonomy over the telecommunication standards and frequency allocation adopted in their terri-
tories. They are, however, expected to abide by the recommendations of the ITU. It is heartening to note that in a
world of political, cultural, and religious diversity and (at times) confrontation, ITU member nations have always
voluntarily abided by ITU recommendations. The administration of radio frequency assignments in each nation
is usually the responsibility of a specially designated agency, such as the Federal Communications Commission
(FCC) in the USA and Ofcom in the UK.

5.6 Summary

Our discussion of the three main transmission media in modern communication systems, namely metallic lines,
optical fibre, and radio is now complete. Our approach was to first present a nonmathematical discussion of the
characterisation, signal impairments, and applications of each medium. This was then followed by a more in-depth
analysis of wave propagation and signal transmission in each medium to develop the tools needed for calculating
secondary medium parameters that allow the estimation of path loss, signal impairments, and received signal
Telegram: @Persian_IEEEComSoc

strength at various points in the system. For example, knowing the attenuation constant 𝛼 of a medium yields the
attenuation after distance l in the medium as 8.686𝛼l dB.
We found that the solution of voltage and current equations in a metallic transmission line predicts the existence
of incident and reflected waves on the line. Reflection at a load connected to the output terminals of a transmission
line is, in general, undesirable and arises due to a mismatch between load impedance Z L and the characteristic
impedance Z o of the line. The amplitude and phase of the reflected wave relative to the incident wave is given
by the reflection coefficient, whereas the extent of impedance mismatch on the line is characterised by the voltage
standing wave ratio (VSWR), which has a minimum value VSWR = 1 for a matched line (Z L = Z o ) and increases
from unity for various degrees of mismatch up to VSWR = ∞ when Z L = 0 (short circuit) or Z L = ∞ (open circuit).
We examined various methods of impedance matching on transmission lines, including quarter-wave transformer,
attenuation pad, and stub tuning, and we briefly introduced the Smith chart, a popular graphical tool for manually
 References 425

solving transmission line problems and displaying RF parameters. Scattering parameters were also discussed for
specifying the forward and reverse gains and reflection coefficients of two-port microwave networks.
In the 1980s optical fibre began to replace coaxial cables in long-distance transmission links. Since then, the
penetration of optical fibre into the telecommunication network has accelerated in pace and depth to the point
that fibre is now rapidly becoming the transmission medium of choice in the local loop of the PSTN to connect
customer premises equipment to the local exchange. We reviewed the advantages of optical fibre over other closed
transmission media, explored the issues surrounding the coupling of light into fibre cores, and discussed in some
detail the extrinsic and intrinsic losses and dispersion impairments in single-mode and multimode fibres. We were
then able to apply our understanding of the limitations imposed by these impairments in the design of repeaterless
sections of optical fibre links and to determine maximum section length and bit rate.
The future of radio as a transmission medium in telecommunications is assured, and current research efforts
are mainly focused on developing better radio resource management protocols and devising novel transmission
techniques and strategies that deliver higher efficiencies in spectrum and energy utilisation. After a brief and
nonmathematical overview of radio and Maxwell’s equations, we discussed the different modes of radio wave
propagation in the atmosphere with and without the influence of the earth’s surface. We also reviewed different
radio wave propagation effects in the ionosphere and troposphere with an emphasis on the factors that contribute
to their severity or otherwise and references to any existing ITU-R models for their prediction. We also covered a
carefully gauged treatment of various radio wave propagation mechanisms, including reflection, refraction, rough
surface scattering, diffraction, free space propagation, and plane earth multipath, and concluded with a brief com-
ment on the work of the International Telecommunication Union (ITU) in radio spectrum management.
We are now well equipped to calculate received signal power Pr in various transmission media. Our attention
turns in the next chapter to learning how to quantify and estimate noise power Pn , after which we will be able to
determine carrier-to-noise-ratio C/N, a critical parameter in the quality of service (QoS) performance of commu-
nication systems.

References

1 Young, P.H. (1999). Electronic Communication Techniques, 4e. Prentice Hall. ISBN: 0-13-779984-5.
2 Kao, K.C. and Hockham, G.A. (1966). Dielectric-fibre surface waveguides for optical frequencies. Proceedings of
the IEE 113 (7): 1151–1158.
3 Mears, R.J., Reekie, L., Jauncey, I.M., and Payne, D.N. (1987). Low-noise erbium-doped fibre amplifier operat-
ing at 1.54 μm. Electronics Letters 23 (19): 1026–1028. https://doi.org/10.1049/el:19870719.
4 ITU-T Recommendation G.653 (2010). Characteristics of a dispersion-shifted, single-mode optical fibre and
cable. Available at www.itu.int.
5 ITU-T Recommendation G.655 (2009). Characteristics of a nonzero dispersion-shifted, single-mode optical fibre
Telegram: @Persian_IEEEComSoc

and cable. Available at www.itu.int.

6 ITU-R Recommendation P.531 (2016). Ionospheric propagation data and prediction methods required for the
design of satellite services and systems. Available at www.itu.int.
7 ITU-R Recommendation P.676 (2016). Attenuation by atmospheric gases. Available at www.itu.int.
8 ITU-R Recommendation P.840 (2019). Attenuation due to clouds and fog. Available at www.itu.int.
9 ITU-R Recommendation P.838 (2005). Specific attenuation model for rain for use in prediction methods. Avail-
able at www.itu.int.
10 ITU-R Recommendation P.618 (2017). Propagation data and prediction methods required for the design of
Earth–space telecommunication systems. Available at www.itu.int.
11 ITU-R Recommendation P.530 (2017). Propagation data and prediction methods required for the design of
terrestrial line-of-sight systems. Available at www.itu.int.
 426 5 Transmission Media

12 Kraus, J.D. (1988). Antennas, 2e. New York: McGraw-Hill. ISBN: 0-07-100482-3.
13 Parsons, J.D. (2000). The Mobile Radio Propagation Channel, 2e, 88–91. Chichester: Wiley. ISBN: 0-471-98857-X.

Questions
Review Questions

5.1 Discuss the importance and limiting factors of a copper wire pair as a transmission medium in modern
communication systems. List current applications of this transmission medium and the practical measures
taken to minimise the effects of its limitations.

5.2 Discuss the features and applications of the coaxial cable as a transmission medium. Which of these appli-
cations do you expect to continue? Give reasons for your answer.

. Present a review of the structure and features of the three standard types of optical fibre.
5.3 (a)
(b) Discuss the advantages and disadvantages of the optical fibre transmission medium.
(c) Give your opinion on the future role of optical fibre in the local access network, considering emerging
telecommunication services, competing local access technologies, and costs.

5.4 Give a brief review of the five modes of radio wave propagation in the atmosphere. Identify the major appli-
cations and frequency range of each mode and state the factors that set the limits of the usable frequency.

Problems

5.5 Show that the condition stated in Eq. (5.2), regarding when transmission line theory may be ignored, cor-
responds to the condition that the phase of the wave should change by less than 0.1 rad (≈6∘ ) in the time it
takes the wave to travel the length of the line.

5.6 Determine whether transmission line theory is required in the analysis of the following circuits. Assume
phase velocity = 2 × 108 m/s in each case.
(a) A local loop carrying voice frequencies 300–3400 Hz between a fixed telephone handset and a local
exchange located 2 km apart.
(b) A laboratory connection of a 1 MHz sinusoidal signal from a signal generator to an oscilloscope using a
Telegram: @Persian_IEEEComSoc

coaxial cable of length 1 m.

(c) A 50 cm conducting track on a printed circuit board (PCB) carrying a clock signal of frequency 20 MHz.

5.7 A metallic line attenuates a radio frequency signal at the rate 8.686𝛼 dB/km, where 𝛼 (Np/km) is the attenu-
ation constant of the line given by Eq. (5.23). The primary line constants of one coaxial cable at a frequency
of 10 MHz are L = 234 μH/km, C = 93.5 nF/km, R = 77.7 Ω/km, and G = 3450 μS/km.
(a) If a 10 V rms signal at 10 MHz is transmitted on this line, determine the rms value of the signal after
5 km on the line.
(b) The above cable is used for transmission of a wideband FDM signal that contains frequency components
up to 10 MHz. If the signal on the line must be detected before the attenuation of the 10 MHz component
exceeds 30 dB, determine the maximum repeater spacing that can be tolerated.
 Questions 427

5.8 At audio frequencies (such as on local telephone lines), the attenuation constant of a metallic line is given
as a function of frequency by Eq. (5.24). By how much does a 5 km length of the cable in the previous
problem (Question 5.7) attenuate a 1 kHz signal? How does your result compare with the attenuation of
the cable at 10 MHz calculated previously?

5.9 The last two questions refer to expressions for attenuation that apply at radio frequencies (Question 5.7) and
at audio frequencies (Question 5.8). The latter has an explicit frequency dependence, whereas the former
does not. Does this mean that cable loss is not a function of radio frequency? Discuss fully.

5.10 A loss-free 75 Ω transmission line with air dielectric is used at 100 MHz and is terminated in a load
Z L = 45 + j50 Ω. Determine:
(a) Voltage reflection coefficient at the load.
(b) VSWR on the line.
(c) Impedance at a distance 75 cm from the load given that the wave travels at the speed of light
c = 3 × 108 m/s.

5.11 A 50 Ω coaxial cable of length 40 m is terminated by a load of impedance 40 + j30 Ω. A sinusoidal generator
of internal resistance 50 Ω, which on open circuit gives an output of 30 V rms at 6 MHz, is now connected
to the line input.
(a) Determine the voltage delivered to the load.
(b) If the load is removed, leaving an open circuit at the end of the line, determine the line impedance seen
by the generator.

5.12 Measurements at 300 MHz show that a loss-free 50 Ω transmission line with air dielectric has a voltage
maximum 30 V, and a voltage minimum 10 V (located 20 cm from the load). Determine the following line
parameters:
(a) Voltage reflection coefficient.
(b) Load impedance.
(c) Maximum line impedance.
(d) Power transfer efficiency.

5.13 The primary line constants of a coaxial cable at a frequency of 30 MHz are: L = 200 nH/m, C = 80 pF/m,
R = 0.75 Ω/m, and G = 0. A matched load is connected to the receiving end of 100 m of this cable. The
sending end is connected to a 30 MHz sinusoidal signal generator and adjusted to give 25 V rms measured
across its input terminal.
Telegram: @Persian_IEEEComSoc

(a) Determine the attenuation constant and phase constant of the line.
(b) Determine the total attenuation of the line in dB.
(c) Calculate the efficiency of the line.

5.14 The single-stub tuner technique matches a load Z L to a transmission line of characteristic impedance Z o by
connecting a length d2 of a short-circuited Z o -line (called a stub) in parallel at a distance d1 from the load.
If Z L = 100 − j60 Ω and Z o = 75 Ω, determine:
(a) The location of the stub. This is the shortest distance d1 from the load to the stub, expressed in terms of
wavelength.
(b) The shortest length d2 of the short-circuited stub, expressed in terms of wavelength.
 428 5 Transmission Media

5.15 Determine suitable values for resistors R1 , R2 , and R3 to match a 75 Ω cable to a 50 Ω cable using:
(a) A T-network attenuation pad (Figure 5.14).
(b) A 𝜋-network attenuation pad (Figure 5.14).

5.16 Calculate the S-parameters of the T-network shown in Figure Q5.16 and hence determine:
(a) Input VSWR.
(b) Return loss.
(c) Forward insertion power gain.
(d) Reverse insertion power loss.

1 2 Figure Q5.16 T-network.

50 + j25 Ω 50 – j25 Ω

50 Ω

1ʹ 2ʹ

5.17 The S-parameters of a microwave amplifier are specified as: s11 = 0.06∠15∘ , s22 = 0.08∠5∘ , s21 = 100∠180∘ ,
and s12 = 0.01∠180∘ . Determine this amplifier’s:
(a) Input VSWR.
(b) Return loss.
(c) Forward insertion power gain.
(d) Reverse insertion power loss.

5.18 Attenuation in optical fibre due to Rayleigh scattering is given by the expression
n8 p2 𝛽kT F
Lsc = 3.591 × 105 dB∕km
𝜆4
where
● k ≡ Boltzmann’s constant = 1.38 × 10
−23 J/K.
Telegram: @Persian_IEEEComSoc

● 𝜆 ≡ Free space wavelength of the light signal.

● n ≡ Refractive index of the fibre material.

● p ≡ Photoelastic coefficient.

● 𝛽 ≡ Isothermal compressibility at temperature T .

F
● T ≡ Freezing temperature at which the irregularities become frozen into the fibre structure.
F
(a) Determine Lsc for transmission over a 200 km distance in the three absorption windows, of nominal
wavelengths 𝜆 = 820, 1330, and 1550 nm. The fibre material has the following parameters: n = 1.45,
p = 0.286, 𝛽 = 7 × 10−11 m2 /N, and T F = 1400 K.
(b) Consider the transmission at 1550 nm. With Lsc having such a small value, would regenerators be needed
along this link? Give a rough estimate of the total attenuation that would be encountered on this link in
practice.
 Questions 429

5.19 A 1310 nm wavelength light is incident from air medium (refractive index = 1) on optical fibre made with
flint glass core (refractive index = 1.62) and crown glass cladding (refractive index = 1.54). Determine:
(a) The critical angle of the core-cladding interface.
(b) The numerical aperture of the interface between light source and fibre.
(c) The coupling loss in dB due to reflectance at the source-fibre interface.
(d) The beam width or apex angle of the cone of acceptance (i.e. angle 2𝜃 a in Figure 5.23) for light to be
successfully coupled into the fibre core.
(e) The maximum diameter of the core if single-mode propagation is to be achieved.

5.20 A vertically polarised plane wave of electric field amplitude 10 V/m and frequency 100 MHz is incident
from free space at an angle of incidence 𝜃 1 = 75∘ onto a body of seawater having a horizontally flat and
smooth surface. The electric field is parallel to the plane of incidence. The primary parameters of seawater
at this frequency are 𝜎 = 5 S/m, 𝜀r = 70, and 𝜇 r = 1.
(a) What is the impedance of this wave in seawater?
(b) Determine the amplitude and phase of the reflected E field.
(c) Determine the amplitude and phase of the transmitted H field.
(d) Repeat the calculation in (b) if the electric field is perpendicular to the plane of incidence, and comment
on how both results compare.

5.21 Using the Fraunhofer criterion, determine the electromagnetic wave frequency beyond which the following
surfaces will be classed as rough because of diffuse reflections at (i) near-grazing incidence of 𝜃 i = 85∘ and
(ii) normal incidence of 𝜃 i = 0∘ :
(a) A flat football pitch covered with grass of maximum height 1 cm.
(b) Flat ground covered with shrubs of variable height up to a maximum of 1 m.

5.22 A point-to-point microwave radio link at 1 GHz is to be set up between two locations separated by 10 km.
There is a roof structure of height 30 m along the direct path between the transmit- and receive-antennas
both of which have a gain of 30 dB. Assuming a flat earth between the two locations and ignoring all other
terrain effects, and that the roof structure can be approximated as a perfectly absorbing thin knife edge,
determine:
(a) The received signal power if the transmitted power is 50 W, the transmit- and receive-antennas are both
of height 28 m, and the roof structure is halfway between the two locations.
(b) The received signal power if the link is as in (a) except that the roof structure is located (along the path)
at 1 km from the transmitter.
(c) The received signal power if the link is as in (a) except that the roof structure is located (along the path)
at 1 km from the receiver.
Telegram: @Persian_IEEEComSoc

5.23 For the radio link of Question 5.22, determine the minimum transmit- and receive-antenna heights (ht = hr )
required to achieve free space line-of-sight propagation conditions when the 30 m high roof structure, being
the only obstacle along the path from transmitter to receiver, is
(a) Located halfway between the transmitter and the receiver.
(b) Located 1 km from the transmitter.
(c) Located 1 km from the receiver.

5.24 A satellite transponder has an EIRP of 35 dBW and transmits over a 38 800 km slant path to an earth station
of receive-antenna gain Gr = 40 dB. Neglecting all link losses except FSPL, determine the output power of
the earth station receive-antenna for a transmission frequency of:
(a) 20 GHz
 430 5 Transmission Media

(b) 12 GHz
(c) 3820 MHz.

5.25 The gain of a parabolic antenna of diameter D is given by the expression

Gr = 9.943 + 10 log10 (𝜂D2 ) − 20 log10 (𝜆) dB
where 𝜂 is the antenna efficiency and 𝜆, the signal wavelength, is expressed in the same units as D. Assum-
ing 𝜂 = 0.6:
(a) Determine the diameters of the parabolic antennas in Question 5.24(a)–(c).
(b) If the three frequencies in Question 5.24 were received using antennas of the same size, how would the
outputs compare?
(c) Comment on your results in (a) and (b).
Telegram: @Persian_IEEEComSoc
 431

Noise in Communication Systems

It is the wrong strategy to seek to eliminate all noise. Systems are designed to excel in the presence of small
‘enemies’ and so were you.

In this Chapter
✓ A review of the physical sources of random noise in communication systems.
✓ A brief discussion of Gaussian noise, white noise, and narrowband noise.
✓ Quantification of random noise in a single device and in a system that is a cascade connection of multiple
devices, such as a satellite communication earth station.
✓ Evaluation of the impact of noise in various analogue and digital communication systems.
✓ Worked examples to demonstrate the interpretation and application of concepts and to deepen your insight
and hone your skills in engineering problem solving.
✓ End-of-chapter questions to test your understanding and (in some cases) extend your knowledge of the
material covered.

6.1 Introduction
The purpose of a communication system is to convey information from a source point to a destination point, which
may be separated by a few metres or by thousands of kilometres. Using various (modulation or baseband transmis-
sion) techniques (discussed in Chapters 10 and 11), the information is transformed into a suitable electrical signal
of adequate power level (i.e. strength) and transmitted towards the destination. However, as the signal propagates
Telegram: @Persian_IEEEComSoc

in the transmission medium linking source and destination points, its strength is gradually reduced. A discussion
of the causes of this attenuation and their evaluation is the subject of Chapter 5.
The signal reaching the receiver will in general be very weak. For example, in satellite communications the
signal power at the input of a receiving earth station antenna could be just −120 dBW (= 1 pW). There would
be no problem with detecting such small signal levels if we were dealing with an ideal communication system
in which only the wanted signal was present. However, problems arise in practice because of the (unavoidable)
presence of one or more of the following unwanted signals.
● Random signals or noise: these have a variety of sources and are added to the signal in the transmission medium
and within the receiver system itself.
● Interference signals: the type of interference signal and its significance depend largely on the communication
system. A few examples will suffice.
Communication Engineering Principles, Second Edition. Ifiok Otung.
© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 432 6 Noise in Communication Systems

⚬ Crosstalk may occur in which intelligible information is coupled by inductance or capacitance from one com-
munication channel (e.g. wire-pair) to another.
⚬ Two separately located radio systems that intercept a common volume in the atmosphere may become coupled
by scattering, the signal in one link being scattered into the antenna of the other link.
⚬ Two radio systems that transmit on the same frequency using orthogonal polarisation may interfere with each
other as a result of cross-polarisation, whereby a portion of the energy in one polarisation is converted to the
other in an anisotropic medium.
⚬ An interference signal may be produced in one communication channel due to insufficient guard bands in
multiplexed systems, or due to insufficient filtering. For example, hum noise at 50 Hz and its harmonics may
result from insufficient rectification and smoothing of the public power supply. Furthermore, hum noise is in
the air as electromagnetic radiation from power lines and can be picked up along with the desired signal.
⚬ Interference voltages may be induced in aerials and metallic lines by lightning and by sparks (e.g. ignition) at
electrical contacts.
Most interference signals can be eliminated by careful system design, although the cost may sometimes be pro-
hibitive. However, random noise will be always present.
The ability of a receiver system to correctly detect an incoming signal is not determined by the signal strength
alone. A reasonably strong signal can be obscured in a noisy receiver, whereas a much weaker signal may be
accurately recovered using a low-noise system. So, the principal consideration is not just signal power but signal
power in comparison to noise power, which is expressed in the parameter known as signal-to-noise ratio (SNR).
The SNR at a given point in a communication system is defined as the ratio of average received signal power Pr at
that point to the average noise power Pn at the same point. That is
Average Received Signal Power Pr
SNR = ≡
Average Noise Power Pn
= Pr |dBW − Pn |dBW (dB) (6.1)
In this chapter we will learn how to determine noise power and SNR at various points in a communication
system. We will also assess the impact of noise in communication systems and learn how to select modulation
schemes that perform satisfactorily in the presence of noise. A good understanding of the characterisation of ran-
dom signals, as discussed in some detail in Chapter 3, would be helpful.
We begin with a discussion of the physical sources of noise, which includes (where suitable) quantitative esti-
mates of noise level. We build on the foundation of Gaussian noise laid in Chapter 3 to discuss the important
concepts of white and narrowband noise before proceeding to a treatment of system noise calculations, including
noise power, noise temperature, and noise factor in single- and multiple-stage systems. The impact of noise in
various analogue and digital transmission systems is discussed, which equips us to make an informed choice of
modulation technique in each situation.
Telegram: @Persian_IEEEComSoc

6.2 Physical Sources of Random Noise

6.2.1 Thermal or Johnson Noise

Thermal agitation of electrons in conductors produces random voltage fluctuations across the conductor. The
mean square value of thermal noise voltage measured across the terminals of a conductor is given by

v2n = 4kTBR (6.2)

Here, k is Boltzmann’s constant (= 1.38 × 10−23 J/K); R is the resistance of the conductor in ohm, T is the absolute
temperature of the conductor in kelvin (K), and B is the bandwidth of the measurement in Hz. Note that this
 6.2 Physical Sources of Random Noise 433

bandwidth B is the noise equivalent bandwidth and not any of the other definitions of bandwidth as discussed in
Section 4.7.3.

6.2.2 Quantisation Noise

The process of converting an analogue signal to digital inevitably introduces a quantisation error in every digitised
sample. For small quantisation step sizes and at sampling rates that are large enough to avoid aliasing, the errors
are random and have the same effect as thermal noise on the reconstructed analogue signal, hence the name
quantisation noise. In the simple case of uniform quantisation, the mean square value of quantisation noise voltage
is given by
A2p−p
v2q = exp(−1.3863n) (6.3)
12
Here, Ap–p is the peak-to-peak amplitude of the analogue signal required to completely fill the quantiser without
clipping, and n is the number of bits per sample. Note that quantisation noise is entirely under the control of the
system designer through the two parameters Ap-p and n.

6.2.3 Radio or Sky Noise

Under thermal equilibrium, an absorbing medium re-radiates the radio wave energy that it absorbs. This
re-radiation constitutes noise for a receive-antenna. The brightness temperature or effective noise temperature of a
source gives a measure of the power radiated in each frequency band in the direction of a receive-antenna. This
is the physical temperature T of a blackbody that emits the same power P in that frequency band, as given by
Planck’s law
hfB
Pn = (6.4)
exp(hf ∕kT) − 1
where B is the bandwidth (Hz), f is the centre frequency (Hz), and h is Planck’s constant = 6.626 × 10−34 J⋅s. At
radio frequencies, hf /kT ≪ 1, and exp(hf /kT) ≈ 1 + hf /kT, so that
hfB
Pn =
1 + hf ∕kT − 1
= kTB (6.5)
This noise power will be delivered in its entirety only to a device that is matched (as discussed later) to the noise
source. Mismatched devices will absorb only a fraction of this power and reject (i.e. reflect) the rest. Unfortunately,
we cannot resort to using mismatched antennas in order to suppress radio noise power, since this would also cause
a significant portion of the wanted signal to be rejected. Noise power is often quantified using noise temperature
Telegram: @Persian_IEEEComSoc

T, rather than Pn , where

Pn
T= (6.6)
kB
Pn depends strongly on the frequency of operation, amongst other factors, and includes contributions from one
or more of the following:
● Extra-terrestrial noise temperature T xn , including galactic noise, solar noise, noise from the surface of the moon,
and a cosmic background noise of 2.7 K. Galactic noise decreases rapidly with frequency from more than 1000 K
at 100 MHz to a negligible level above about 2 GHz. The sun is a strong source of noise at all radio frequen-
cies, and if it fully illuminates the antenna main beam then its noise temperature will range from 106 K at
100 MHz to about 104 K at frequencies above 10 GHz under quiet sun conditions. Direct viewing of the sun
 434 6 Noise in Communication Systems

by a receive-antenna will therefore cause a weak received signal (especially in satellite communications) to be
swamped by noise, resulting in link outage.
● Atmospheric noise temperature T an , including noise radiated by hydrometeors (i.e. rain, ice, snow, and cloud
water droplets), aerosols, and atmospheric gases. This is given by the expression
[ ]
1
Tan = Tm 1 −
Latm
Latm = 10(A∕10) ; Tm ≈ 1.12Tsurf − 50 (6.7)
where A (dB) is the total attenuation due to absorption in the atmosphere, Latm is the resulting loss factor,
T m ≈ 275 kelvin (K) represents a mean radiating temperature, and T surf ≈ 290 K is the surface temperature (K).
The mean radiating temperature T m is the absorption-weighted average physical temperature of the atmosphere
along a slant or zenith path, the contribution of each layer being reduced by the loss incurred in the atmo-
sphere beneath the layer. In other words, the value of T m is a result referenced to the surface of the earth. It is
worth pointing out that the formula given in Eq. (6.7) for T m in terms of surface temperature T surf is a rough
empirical relation. Slightly improved first-order polynomial fits for T m as a function of T surf , based on extensive
radiosounding observations, have been developed by Lucas-Vegas and Riera [1] for various climatic zones and
frequencies up to 100 GHz. On a satellite communication link, the total attenuation (and hence T an ) depends
on path elevation angle 𝜃, increasing as 𝜃 decreases. On such links, there will always be some amount of atmo-
spheric gaseous absorption even if quite small at some frequencies (see Figure 5.34), so the condition A = 0 dB,
which would lead to Latm = 1 and T an = 0, will never arise. Hence, T an > 0 always.
● Noise from the surface of the earth, having noise temperature given by
Ten = 𝜀𝜙 Tsurf (6.8)
where T surf is the surface temperature as above, and 𝜀𝜙 is the emissivity of the surface – defined as the ratio of the
absorbed power to the incident power at an angle 𝜙 to the horizontal. Land surfaces have higher emissivity than
water surfaces, and hence higher brightness temperature T en . If the earth is considered a blackbody then 𝜀𝜙 = 1,
and T en = T surf ≈ 290 K. However, radiation from the surface of the earth usually enters a directional antenna
(such as those employed in satellite communication earth stations) through its sidelobe, which is typically at
least a factor Lsl = 100 (or 20 dB) below the main lobe gain. Thus, even in the worst case of a blackbody earth,
the contribution of T en to the system noise temperature is reduced by this factor to at most ∼3 K.
By analogy with thermal noise – Eq. (6.2), the mean square sky noise voltage at the terminals of an antenna is
given by

v2s = 4kT a BRr (6.9)

where T r is the radiation resistance of the antenna and T a is the antenna noise temperature. The radiation resistance
Telegram: @Persian_IEEEComSoc

of an antenna is a fictitious resistance that would dissipate as much power as the antenna radiates (in transmit
mode) if the resistor were connected to the same transmission line. Thus, an antenna radiating P watt when draw-
ing the root-mean-square (rms) current I rms has a radiation resistance Rr = P∕Irms 2 . The noise temperature of a

radio system’s receive-antenna is the result of contributions from all the classes of radio noise introduced above
and is given by
Txn T
Ta = + Tan + en (6.10)
Latm Lsl
The first term represents extra-terrestrial radio noise T xn (≈3 K for frequencies above about 2 GHz) reduced by
the atmospheric loss factor Latm . The third term is the radio noise from a warm earth (T en ≈ 290 K) reduced by
the factor Lsl , which is the loss factor (relative to the main lobe) of the antenna sidelobe that points towards the
earth. In the worst case of an omnidirectional antenna, Lsl = 1 which leads to the third term on the right-hand
 6.2 Physical Sources of Random Noise 435

side of Eq. (6.10) contributing about 290 K. The second term T an is negligible below about 1 GHz, but the first term
increases rapidly below 1 GHz, as earlier indicated. For a satellite link under clear-sky conditions, T an may be deter-
mined by reading zenith atmospheric attenuation Az from Figure 5.34, converting this to slant path atmospheric
attenuation A at the link’s elevation angle 𝜃 ≥ 10∘ as A = Az /sin𝜃, and then using this value of A in Eq. (6.7) to obtain
the atmospheric noise temperature T an of the link. For example, Figure 5.34 gives Az = 0.7 dB at 20 GHz for a humid
tropical condition. If path elevation 𝜃 = 50∘ , the slant path attenuation will be A = Az /sin𝜃 = 0.7/sin50∘ = 0.914 dB,
which gives the link antenna noise temperature as
[ ] [ ]
1 1
Tan = Tm 1 − = Tm 1 − 0.9138∕10
Latm 10
≈ 275(1 − 0.81025)
≈ 52.2 K
For nonclear weather, such as during a rain event, the total attenuation, made up of gaseous absorption (in
Figure 5.34) plus rain attenuation (based on Figure 5.35), must be used in Eq. (6.7) to determine T an .
It should be emphasised that this discussion of Eq. (6.10) applies only to satellite downlinks, where the
receive-antenna is located near the earth’s surface and pointed skyward towards a satellite. For transmissions in
the opposite direction, called an uplink, the receive-antenna is on the satellite and is pointed directly towards a
warm earth. Its noise temperature T a is therefore dominated by the third term of Eq. (6.10), with Lsl = 1, so that
T a ≈ 300 K in all conditions.

6.2.4 Shot Noise

Current flow in electronic devices is not continuous but discrete, being the result of the motion of many charged
particles of smallest charge qe = 1.602 × 10−19 coulomb. Random fluctuations in the number of these particles,
including electrons emitted by a cathode in vacuum tubes and minority carriers in semiconductor devices (such
as diodes and transistors), give rise to shot noise. The mean square value of shot noise current is given by
i2s = 2qe IB (6.11)
where I is the average anode or emitter current and qe is the electronic charge, as given above.

6.2.5 Partition Noise

This results from fluctuations in the recombination of minority charge carriers in the base region of bipolar junc-
tion transistor (BJT). In the bygone era of valves, partition noise also occurred in multigrid valves, such as tetrodes,
due to fluctuations in the sharing of current between the electrodes.

6.2.6 Quantum Noise

Telegram: @Persian_IEEEComSoc

Light signals in optical fibre consist of discrete amounts of energy particles known as photons. Fluctuations in
photon flow in fibre give rise to quantum noise, which places a fundamental limit on the sensitivity of OOK
(on–off keying) detectors. An error occurs if, due to fluctuations in arrival rate, no photon is received during a
binary 1 interval. We must therefore ensure that each transmitted light pulse contains a certain minimum average
number of photons so that the required bit error rate (BER) is satisfied despite random fluctuations occasionally
resulting in no photons being received. This consideration leads to a required mean received power given by
[ ]
Rs ln(2 × BER)
Pr = 10log10 − − 70 dBm (6.12)
𝜆
where 𝜆 is the wavelength in nm, Rs is the symbol rate in MBd, and BER is the stipulated bit error rate of the OOK
transmission system.
 436 6 Noise in Communication Systems

6.2.7 Flicker or 1/f Noise

Flicker noise is observed at very low frequencies and is thought to be due to fluctuations in the conductivity of
semiconductor materials as well as the slow drift of an electronic device (e.g. an amplifier) from its operating
point. The power spectral density (PSD) of this noise is inversely proportional to frequency; hence it is also called
1/f noise. Flicker noise is negligible at high frequencies and may be ignored.

Worked Example 6.1 Quantisation Noise and Sky Noise

(a) Determine the rms quantisation noise voltage in a 12-bit uniform quantiser of peak-to-peak voltage 10 V.
(b) A domestic television receive-antenna delivers a sky noise power of −105 dBm to a matched coaxial feeder
in a noise equivalent bandwidth of 8 MHz. Determine the antenna noise temperature.

(a) From Eq. (6.3), with n = 12 and Ap-p = 10, we obtain the mean square quantisation noise voltage, the square
root of which is the desired rms noise voltage. Thus
102
v2q = exp(−1.3863 × 12)
12
= 0.497 square millivolts
√
vqrms = v2q = 0.7 mV
(b) Given, Pn = −105 dBm = 3.1623 × 10−14 W. See the discussion of logarithmic units in Section 2.8 if in any doubt
regarding this conversion from dBm to watt. It follows from Eq. (6.6) that antenna noise temperature is
P 3.1623 × 10−14
Ta = n =
kB (1.38 × 10−23 )(8 × 106 )
= 286.4 K

Worked Example 6.2 Antenna Noise Temperature

An earth station antenna receives a satellite transmission at 20 GHz on a path elevation of 30∘ . Noise from the
earth’s surface enters the antenna via its first sidelobe level at −16 dB. If the earth station is in a mild temperate
climate, determine the antenna noise temperature T a :
(a) During clear-sky conditions.
(b) During rain, when the total atmospheric attenuation (assumed absorptive) is 12 dB.

We will determine T a using Eq. (6.10)

Telegram: @Persian_IEEEComSoc

Ta = Txn ∕Latm + Tan + Ten ∕Lsl

Extra-terrestrial noise T xn ≈ 3 K at 20 GHz. The earth’s brightness temperature T en = εT surf ≈ 290 K. Sidelobe loss
factor Lsl = 10(16/10) = 39.81072.
(a) During clear-sky conditions, gaseous absorption is the only source of attenuation which we determine as
A = Az /sin𝜃, where Az is read from curve A of Figure 5.34 as Az = 0.25 dB. Hence A = 0.25/sin30∘ = 0.5 dB and
Latm = 10(0.5/10) = 1.122. Thus, in clear sky, the atmospheric noise temperature is
Tan = Tm (1 − 10−A∕10 )
≈ 275(1 − 10−0.5∕10 )
≈ 30 K
 6.3 Additive White Gaussian Noise 437

Combining this with the other noise components gives the antenna noise temperature as
Ta = Txn ∕Latm + Tan + Ten ∕Lsl
3 290
= + 30 +
1.122 39.81072
= 40 K
(b) A 12 dB atmospheric attenuation corresponds to an atmospheric loss factor Latm = 15.85. The corresponding
atmospheric noise temperature is
Tan = Tm (1 − 10−A∕10 ) ≈ 275(1 − 10−12∕10 )
≈ 258 K
And this combines with the other noise components to give the antenna noise temperature
Ta = Txn ∕Latm + Tan + Ten ∕Lsl
3 290
= + 258 +
15.85 39.81072
= 265 K
It is important to observe that the antenna noise temperature increases significantly during this rain event that
gives rise to a total atmospheric attenuation of 12 dB. This represents a double degradation in carrier-to-noise
ratio (C/N). First, the carrier signal level is reduced by the attenuation of 12 dB. Second, the noise power
delivered by the antenna increases from 40 K in clear air to 265 K in the rain event. This will lead to an increase
in the overall system noise temperature (discussed later) and hence to a further reduction in C/N.

Worked Example 6.3 Quantum Noise

An optical fibre transmission system uses OOK at 2592 MBd and a light wavelength of 1.55 μm. If a BER of at
most 10−9 is to be satisfied, determine the minimum average received power Pr (in μW) imposed by quantum
noise.

By a straightforward substitution in Eq. (6.12), we obtain

[ ]
2592 ln(2 × 10−9 )
Pr = 10log10 − − 70 dBm
1550
= 10log10 (33.5) − 70 dBm
= −54.75 dBm
= 3.35 nW
Telegram: @Persian_IEEEComSoc

In practice, other noise sources at the receiver, such as thermal and shot noise, will make it necessary to provide
a much higher level of Pr to achieve the specified BER.

6.3 Additive White Gaussian Noise

Each noise type discussed above (except flicker noise) is the result of many statistically independent and random
contributions. For example, thermal noise results from the contribution of many electrons each moving randomly
under thermal agitation about an equilibrium position. There is no general drift of electrons in one direction in the
conductor, this requiring some externally applied voltage. If we observe the conductor over a significant period,
we will find that at some instants the random motions lead to a slight depletion of electrons in the upper half of
 438 6 Noise in Communication Systems

the conductor, which gives a positive noise voltage. At other instants there is a slight surplus of electrons in the
upper half of the conductor, giving a negative noise voltage. The average or mean of the noise voltage samples
measured in the observation interval is zero since there is no reason for either a surplus or depletion of electrons
to be preferred.

6.3.1 Gaussian PDF of Noise

An important theorem, known as the central limit theorem, indicates that under the above circumstances the noise
voltage samples follow a Gaussian distribution with zero mean, as shown in Figure 6.1. The curve is the zero-mean
Gaussian probability density function (PDF)
( )
1 v2n
p(vn ) = √ exp − 2 (6.13)
𝜎 2𝜋 2𝜎

where 𝜎 2 is the variance of the noise voltage vn and, in this case of where the noise voltage has zero mean, the
variance equals the mean square noise voltage v2n or normalised noise power Pn – obtained by averaging the squares
of the noise voltage samples taken over a sufficiently long observation interval. The square root of variance is called
the standard deviation 𝜎, which gives an indication of the spread of vn about its mean. So, we may write

Pn = v2n = 𝜎 2 ≡ Square of standard deviation (6.14)

Please refer to Section 3.4.1 for a discussion of the Gaussian (also called normal) distribution, where the comple-
mentary error function erfc(x) and Q-function Q(x) are also defined. Here, it will suffice to state that the probability
that vn will have a value between z1 and z2 is given by the area under the PDF curve between z1 and z2 as
z2
Pr(z1 ≤ vn ≤ z2 ) = p(vn )dvn (6.15)
∫z1

1 ʋn2
p(ʋn) = exp –
σ 2π 2σ2
0.4/σ

0.35/σ

0.3/σ
Telegram: @Persian_IEEEComSoc

0.25/σ

0.2/σ

0.15/σ
Area = Pr(σ ≤ ʋn ≤ 2σ)
0.1/σ

0.05/σ

0 ʋn
–3σ –2σ –σ 0 σ 2σ 3σ

Figure 6.1 Gaussian probability density function (PDF) of zero mean and standard deviation 𝜎.
 6.3 Additive White Gaussian Noise 439

and that the probability that vn exceeds a certain value z is

∞
Pr(vn ≥ z) = p(vn )dvn
∫z
∞ ( )
1 v2n
= √ exp − 2 dvn
∫z 𝜎 2𝜋 2𝜎
∞
1 2
= √ √ exp(−y )dy
∫
𝜋 z∕𝜎 2
( )
( )
1 z z
= erfc √ =Q (6.16)
2 𝜎 2 𝜎

The following useful relations hold as a result of the symmetry of the Gaussian PDF and the fact that the total
area under a PDF curve is unity.
Pr(vn ≥ −∞) = 1
Pr(vn ≥ 0) = Pr(vn ≤ 0) = 0.5
Pr(vn ≤ z) = 1 − Pr(vn ≥ z)
Pr(vn ≥ −z) = 1 − Pr(vn ≥ z)
Pr(z1 ≤ vn ≤ z2 ) = Pr(vn ≥ z1 ) − Pr(vn ≥ z2 ) (6.17)
The first relation is a statement that the total area under any PDF is 1. The second relation indicates that negative
and positive samples of noise voltage are equally likely. The other relations allow us, for example, to determine
various probabilities, including those involving negative sample values, using tables of Q(x) or erfc(x) that usually
only provide the probability of a positive value x being exceeded.

6.3.2 White Noise

To determine with confidence whether there is any correlation between two samples of noise voltage that are
observed at time instants separated by 𝜏, we would proceed as follows. We measure noise voltage vn (t) over a long
observation interval  → ∞. Then, starting from t = 0, we take the recorded samples in pairs vn (t) and vn (t + 𝜏),
form the product vn (t)vn (t + 𝜏), and obtain the average of these products over the observation interval. This corre-
sponds to the computation

1
Rv (𝜏) = lim vn (t)vn (t + 𝜏)dt (6.18)
 →∞  ∫0

which is called the time-averaged autocorrelation function of the noise voltage. If 𝜏 = 0, then we are pairing each
Telegram: @Persian_IEEEComSoc

sample with itself, and the computation is simply the mean square value of vn (t), otherwise known as the nor-
malised noise power Pn , given by Eq. (6.14). If, on the other hand, 𝜏 ≠ 0, then we are dealing with two different
samples, and the sign of the product will be as shown in Table 6.1.
Recall from the second line of Eq. (6.17) that vn (t) is equally likely to be positive or negative. Let us make the
following assumptions.
● That vn (t + 𝜏) is a random value that does not depend on vn (t).
● That the above condition holds no matter how small 𝜏 is, provided it is not zero.
The first assumption is valid for noise resulting from many statistically independent contributions. The sec-
ond is an idealisation, which implies that the noise generating mechanism has zero recovery time. That is, it can
change instantaneously from one value to any other value within its range. For reasons that will become clear
 440 6 Noise in Communication Systems

Table 6.1 Sign of the product of two samples of noise voltage.

v n (t) v n (t + 𝝉) v n (t)v n (t + 𝝉)

Negative Negative Positive

Negative Positive Negative
Positive Negative Negative
Positive Positive Positive

shortly, we call this type of noise white noise w(t). We must emphasise that such perfect randomness cannot be
attained in real systems, because of inherent inertia. The rate of change of current in real devices is slowed down
by inductance, which is always present even if extremely small. Similarly, the rate of change of voltage is impeded
by capacitance – this also being always present.
Under the conditions outlined above, the product term in Table 6.1 is equally likely to be positive or negative
and has zero mean, so that the integral of Eq. (6.18) is zero when 𝜏 ≠ 0. It follows that for white noise
{
Pn 𝜏 = 0
Rw (𝜏) = (6.19)
0 𝜏≠0
We see that the autocorrelation function Rw (𝜏) of white noise is a zero-width pulse of height Pn . Let the area
under this pulse be N o /2. Clearly, N o is nonzero if and only if Pn is infinite, since as you decrease the width of a
pulse towards zero, you must increase its height towards infinity in order to maintain a nonzero area. Thus, Rw (𝜏)
is an impulse function of weight N o /2. It follows from the first entry of Table 4.5 that Rw (𝜏) has Fourier transform
Sw (f ) = N o /2, which (from Eq. (4.134)) is therefore the PSD of white noise. Thus, as plotted in Figure 6.2
No
Rw (𝜏) =𝛿(𝜏)
2
(6.20)
N
Sw (f ) = o
2
We make the following important observations:

● w(t) contains the same amplitude of all frequencies from f = 0 to ∞. It is therefore called white noise, by analogy
with white light, which contains equal amounts of all wavelengths in the visible spectrum.
● Power per unit bandwidth is twice PSD, as discussed in Eq. (4.136). Therefore, white noise power per unit band-
width is N o and the total noise power within a bandwidth B is
Telegram: @Persian_IEEEComSoc

PnB = No B (6.21)

Figure 6.2c shows a single-sided spectrum of the PSD of white noise.

● From our comments on Eq. (6.19), white noise has infinite average power Pn , and is not realisable in practice.
Nevertheless, the concept of white noise is a simplifying idealisation that satisfactorily models actual noise in
real systems, which are always of finite bandwidth. Within the limited bandwidth of interest, the spectrum of
actual noise looks flat and hence white. And we are not concerned with the inevitable tailing off of the spectrum
of actual noise as f → ∞, since the passband of our system does not extend to those frequencies. As an analogy,
a person running on a football pitch can safely ignore the earth’s curvature and assume a flat earth. Such an
assumption incurs practically no errors because of the limited range (in our case bandwidth) involved. The
flat-earth idealisation would, however, fail where the range is significantly extended to include flying, say, from
Tokyo to New York.
 6.3 Additive White Gaussian Noise 441

No
Rw(τ) = δ(τ)
2
(a)

(b) Sw(f)

No/2
f

Sw(f)

(c)
No

Figure 6.2 White noise: (a) Autocorrelation function; (b) Power spectral density (PSD); (c) Single sided spectrum of PSD.

It is therefore clearly justifiable to assume that the random noise types encountered in communication systems
(except flicker noise) and introduced in Section 6.2 are both Gaussian and white. This noise is further described
as additive since it adds onto the wanted signal and will be present in the medium even when the wanted signal
is zero. This is unlike multiplicative noise, which results from a random variation of channel gain or path loss, as
briefly discussed in deriving the lognormal distribution in Section 3.4.3. Multiplicative noise effect is present only
when the signal is nonzero.
White noise exists only at the input of a communication receiver. As this noise passes through the system it is
inevitably filtered and limited in bandwidth. As discussed in Section 4.7.2, when a signal x(t) having PSD Sx (f ) is
transmitted through a linear time invariant system (generally a filter) of transfer function H(f ), the PSD Sv (f ) of
the output signal v(t) is given by
Sv (f ) = Sx (f )|H(f )|2 (6.22)
We see that the output PSD is modified – in other words coloured – by the square of the filter’s gain response
|H(f )|. If the input is white noise then Sx (f ) = N o /2 and Eq. (6.22) yields output noise PSD as
Telegram: @Persian_IEEEComSoc

No
Sv (f ) = |H(f )|2 (6.23)
2
The output noise power Pn is obtained by integrating Sv (f ) over the entire frequency axis from f = −∞ to f = +∞.
Figure 6.3 shows the transmission of white noise through three systems having gain responses specified as follows
{
1, −B ≤ f ≤ B
|H(f )| = Ideal LPF
0, Otherwise
{
1, fc − B∕2 ≤ |f | ≤ fc + B∕2
|H(f )| = Ideal BPF
0, Otherwise
|H(f )| ≡ |H(f )| Realisable Filter (6.24)
 442 6 Noise in Communication Systems

Input PSD System Output PSD

Sʋ(f)
No
ω(t) Ideal LPF ʋn(t)
(0→B) f
B

Sw(f) Sʋ(f)
No No
ω(t) Ideal BPF ʋn(t)
(fc ± B/2)) f
f
fc – B/2 fc fc + B/2

Sʋ(f)
K2No ∣H(f)∣2
ω(t) ʋn(t) K = ∣H(f)∣max
H(f)
f
B

Figure 6.3 Filtering of white noise.

Single-sided representations of the input PSD and output PSD for each system are shown in Figure 6.3 to illus-
trate the effect of each system on white noise. The noise at the output of each ideal filter has a uniform PSD within
a finite bandwidth and is therefore described as bandlimited white noise. This type of noise has finite power Pn
given by Eq. (6.21), which is simply the area of a rectangle of height N o and base B. The third system in Figure 6.3
is a realisable filter whose gain response is not strictly constant within the system’s passband. The output PSD in
this case is not constant but exactly replicates the variation of the square gain response of the system with fre-
quency. Noise of this kind that has a nonuniform PSD is known as coloured noise. Both bandlimited white noise
and coloured noise are commonly referred to simply as narrowband noise. Note that narrowband noise obtained
in this manner by linear processing of white Gaussian noise is also itself Gaussian, though nonwhite.
Given a system with gain response |H(f )|, the value of B with which we would multiply N o to obtain the same
noise power as the coloured noise power of the system referred to its input is known as the noise equivalent band-
width of the system. An expression for this bandwidth in terms of |H(f )| is derived in Chapter 4 (Figure 4.55)
as
∞
1
B= |H(f )|2 df
K 2 ∫0
Telegram: @Persian_IEEEComSoc

K = |H(f )|max (6.25)

Once we know the noise equivalent bandwidth B of a system, the noise power Pn of the system is obtained simply
(without the need for a complicated integration) as
{
K 2 No B, At system output
Pn = (6.26)
No B, Referred to system input
where K is the maximum (ratio) gain (i.e. maximum value of the gain response) of the system and N o is the noise
power per unit bandwidth at the system’s input. The noise equivalent bandwidth concept replaces a realisable
filter of gain response |H(f )| having maximum value K with an ideal brick wall filter of bandwidth B (given by
Eq. (6.25)) and gain K. This converts coloured noise to bandlimited white noise of the same power, computed
 6.3 Additive White Gaussian Noise 443

more easily and without any error using Eq. (6.26). We will adopt this approach in all system noise calculations,
and it will be taken for granted that B (except where otherwise indicated) refers to noise equivalent bandwidth.
When the noise equivalent bandwidth is not known then the 3 dB bandwidth of the system may be used in its
place. This substitution underestimates noise power but the error is small if the system’s gain response has steep
sides with a small transition width between its passband and stop band.

Worked Example 6.4 Resistor-capacitor (RC) Lowpass Filter

The RC lowpass filter (LPF) shown in Figure 6.4 has gain response

1 1
|H(f )| = √ ≡ √ , a = 2𝜋RC
1+ (2𝜋fRC)2 1 + a2 f 2

Determine:
(a) The noise equivalent bandwidth B of the filter.
(b) The 3 dB bandwidth B3dB of the filter.
(c) The noise power at its output when connected to a matched antenna of noise temperature T a = 80 K, assum-
ing the filter’s components are noiseless.
(d) How much error is incurred in noise power computation by using B3dB in place of B.

(a) We note that |H(f )|max = 1 at f = 0, so K = 1 in Eq. (6.25), which gives noise equivalent bandwidth as
∞
1
B= df
∫0 1 + a2 f 2
The form of the integrand suggests the use of the substitution af = tan𝜃. This transforms the integral as follows

1 + a2 f 2 = 1 + tan2 𝜃 = sec2 𝜃 = 1∕cos2 𝜃

df ( )
d 1 1 1
= tan 𝜃 = ; ⇒ df = d𝜃
d𝜃 d𝜃 a a cos2 𝜃 a cos2 𝜃
Limits f = 0 → ∞ becomes 𝜃 = tan−1 (af ) = 0 → 𝜋∕2

Thus
𝜋∕2 𝜋∕2
d𝜃 d𝜃 𝜋 𝜋
B= cos2 𝜃 = = =
∫0 a cos2 𝜃 ∫0 a 2a 2 × 2𝜋RC
Telegram: @Persian_IEEEComSoc

1
= (Noise equivalent bandwidth of RC LPF)
4RC
Putting R = 300 Ω, C = 132.63 nanofarad (nF) yields B = 6283 Hz.

R R = 300 Ω;
Input C Output C = 132.63 nF

Figure 6.4 RC lowpass filter (LPF) for Worked Example 6.4.

 444 6 Noise in Communication Systems

(b) The 3 dB bandwidth

√ B3dB of a LPF is defined as the frequency f = B3dB at which the gain response of the filter
equals 1∕ 2 of its maximum value. Thus
1 |H(f )|max 1 1
|H(f )||f =B3dB = √ = √ = √ = √
2
1 + a2 B3dB 2 2 1+1

⇒ a2 B23dB = 1
1 1
⇒ B3dB = =
a 2𝜋RC
Putting the values of R and C yields B3dB = 4 kHz.
(c) The antenna delivers noise having power per unit bandwidth N o = kT a to the input of the LPF, where k is
Boltzmann’s constant and T a is the antenna noise temperature. Knowing the noise equivalent bandwidth B
(obtained in (a)) and maximum gain K = 1 of the LPF allows us to calculate the noise power at the output of
the LPF (assumed noiseless) as
Pn = K 2 No B = kT a B
= 1.38 × 10−23 × 80 × 6283 W
= 6.936 × 10−18 W = −141.6 dBm
(d) The ratio between noise equivalent bandwidth B and 3 dB bandwidth B3dB of the RC LPF is
B 1
= × 2𝜋RC
B3dB 4RC
𝜋
= = 10log10 (𝜋∕2) dB
2
= 1.96 dB
Use of the 3 dB bandwidth in place of B for this LPF therefore causes noise power to be underestimated by
1.96 dB.

6.3.3 Canonical and Envelope Representations of Noise

A communication receiver usually consists of several stages of processing, but the simple receiver model shown in
Figure 6.5a will serve our purpose here of noise analysis in bandpass (i.e. modulated) systems. The bandpass filter
(BPF) is designed to pass the wanted signal, which is a modulated carrier of frequency, and to prevent (unwanted)
signals at other frequencies from reaching the demodulator. The quality of the output signal depends on the SNR
at the input of the demodulator. As earlier discussed, we have white noise w(t) at the input of the receiver and
bandlimited white noise vn (t) at the demodulator input. Since the wanted signal is represented in terms of the
carrier frequency f c , we wish to express vn (t) in a similar form, which may be employed, for example, to analyse
Telegram: @Persian_IEEEComSoc

the effect of noise on the system.

Figure 6.5b shows the (continuous) amplitude spectrum V n (f ) of the bandlimited white noise vn (t). Let us
approximate vn (t) using a discrete Fourier series by dividing V n (f ) into infinitesimally small frequency intervals
Δf and placing in each interval a sinusoid of power equal to the noise power N o Δf of that interval. Thus, each
sinusoid has amplitude
√
A = 2No Δf (6.27)

Using f c as reference, the spectral line of the mth sinusoid is shown in Figure 6.5b and has random phase 𝜙m (t)
and frequency
fm = fc − mΔf (6.28)
 6.3 Additive White Gaussian Noise 445

Incoming signals BPF ʋn(t) + ʋs(t) Output signal

(a) + Demodulator
(fc)

White noise
w(t)

|Vn(f)|

(b) Δf

f
fc – B/2 fc fc + B/2
fc – mΔf

|VnI(f)| = |VnQ(f)|

(c)

f
–B/2 0 B/2

Figure 6.5 (a) Simple receiver model; (b) Amplitude spectrum of bandlimited white noise v n (t) at demodulator input; (c)
Amplitude spectrum of in-phase and quadrature components of v n (t).

So, we may express vn (t) as a Fourier series

∑
M∕2
vn (t) = A cos[2𝜋fm t + 𝜙m (t)] (6.29)
m=−M∕2

The summation involves M sinusoids (for m = −M/2, −M/2 + 1, …, −2, −1, 1, 2, …, M/2 − 1, M/2) in the band-
width B, where
B
Δf = (6.30)
M
Note that m ≠ 0. The approximation of Eq. (6.29) becomes exact as M → ∞, and Δf → 0. Substituting Eq. (6.28)
Telegram: @Persian_IEEEComSoc

in (6.29) and expanding yields

∑
vn (t) = A cos[2𝜋(fc − mΔf )t + 𝜙m (t)]
m
[ ]
∑
M∕2
= A cos(−2𝜋mΔft + 𝜙m ) cos(2𝜋fc t)
m=−M∕2
[ ]
∑
M∕2
− A sin(−2𝜋mΔft + 𝜙m ) sin(2𝜋fc t) (6.31)
m=−M∕2

This equation has the following canonical form

vn (t) = vnI (t) cos(2𝜋fc t) − vnQ (t) sin(2𝜋fc t) (6.32)
 446 6 Noise in Communication Systems

where

∑
M∕2
vnI (t) = A cos(−2𝜋mΔft + 𝜙m )
m=−M∕2

∑
M∕2
vnQ (t) = A sin(−2𝜋mΔft + 𝜙m ) (6.33)
m=−M∕2

Thus, we have represented the bandpass noise voltage vn (t) as the sum of two quadrature amplitude modulated
carriers. We call vnI (t) the in-phase component of vn (t) and refer to vnQ (t) as the quadrature component. In this
form, the bandpass noise vn (t) is a modulated carrier that carries two baseband signals vnI (t) and vnQ (t) in quadra-
ture. Figure 6.6a shows an arrangement suggested by Eq. (6.32) for combining vnI (t) and vnQ (t) to produce vn (t).
Given vn (t), we can extract vnI (t) and vnQ (t), as shown in Figure 6.6b. First, vn (t) is multiplied by a cosine and a
sine carrier, respectively, and then each balanced modulator output is passed through an LPF. You may wish to
perform the operations indicated in this block diagram in order to verify the outputs.
The random signals vnI (t) and vnQ (t) have several important properties, a few of which we mention below.

● Both vnI (t) and vnQ (t) have zero mean. You can see that Eq. (6.33) is a Fourier series having no DC (direct current)
component, i.e. zero average value.
● From their Fourier series approximation, we see that vnI (t) and vnQ (t) have the same amplitude spectrum, shown
in Figure 6.5c, which is simply the amplitude spectrum of the bandpass noise vn (t) translated from f c to base-
band. However, the phase spectra of these quadrature components, as the name implies, differ by 90∘ .

ʋnI(t) Figure 6.6 Bandpass noise: (a) construction from quadrature

components; (b) extraction of quadrature components.

LO cos(2πfct)
+
ʋn(t)
(a) 90° Phase
Delay
–
sin(2πfct)
ʋnQ(t)
Telegram: @Persian_IEEEComSoc

ʋnI(t)
LPF

LO 2cos(2πfct)
ʋn(t)
(b) 90° Phase
Advance
–2sin(2πfct)
ʋnQ(t)
LPF
 6.3 Additive White Gaussian Noise 447

● vnI (t) and vnQ (t) have the same power Pn (= variance 𝜎 2 ) as the bandpass noise vn (t). You can see this by noting
that their Fourier series representations contain M sinusoids each of amplitude A. Thus
A2
PnI = PnQ = M
2
2No Δf
= M
2
= No B
= Pn = 𝜎 2 (6.34)

where we have made use of Eqs. (6.27) and (6.30).

● The narrowband noise vn (t) is assumed to be Gaussian with zero-mean, and so are vnI (t) and vnQ (t), these being
obtainable from vn (t) by linear processing, according to Figure 6.6b.

By sinusoidal addition of vnI (t) cos(2𝜋fc t) and −vnQ (t) sin(2𝜋fc t), we can rewrite Eq. (6.32) in the alternative form

vn (t) = r(t) cos[2𝜋fc t + 𝜓(t)] (6.35)

where
√
r(t) = v2nI (t) + v2nQ (t)
[ ]
vnQ (t)
𝜓(t) = tan−1 (6.36)
vnI (t)
The random function r(t) is called the envelope of the bandpass noise vn (t). Equation (6.36) specifies nonlinear
operations on two independent zero-mean Gaussian random signals of variance 𝜎 2 = N o B. The result of each
processing is a new random signal that follows a non-Gaussian distribution. The probability density function
(PDF) of r(t) is given by
( )
⎧ r r2
⎪ exp − 2 , r ≥ 0
p(r) = ⎨ 𝜎 2 2𝜎 (6.37)
⎪0, Otherwise
⎩
This equation defines what is known as a Rayleigh probability density function, which we derive and discuss in
detail in Section 3.4.2. The envelope of vn (t) is said to be Rayleigh-distributed.
The function 𝜓(t) appearing in Eqs. (6.35) and (6.36) is the phase of vn (t). It is uniformly distributed between
−𝜋 and 𝜋 radian. That is, the phase of narrowband noise vn (t) has the uniform PDF given by
⎧ 1
⎪ , −𝜋 ≤ 𝜓 ≤ 𝜋
p(𝜓) = ⎨ 2𝜋 (6.38)
Telegram: @Persian_IEEEComSoc

⎪0, Otherwise
⎩
Thus, bandpass Gaussian noise vn (t) has a Rayleigh distribution of its envelope and a uniform distribution of
its phase. These PDFs are shown in Figure 6.7. Note that the envelope r(t) is always positive and corresponds to
a complex sample value r(t)∠𝜓(t) of bandpass noise. This complex sample has real part vnI (t) and imaginary part
vnQ (t) given by
vnI (t) = r(t) cos[𝜓(t)]
vnQ (t) = r(t) sin[𝜓(t)] (6.39)

each of which is a zero-mean Gaussian random variable of variance 𝜎 2 = N o B and is equally likely to be positive or
negative. Note that because 𝜓(t) has a uniform PDF in the range (−𝜋, 𝜋), it is equally likely (i.e. with probability
 448 6 Noise in Communication Systems

r
p(r) = exp(–r2/2σ2)
σ2
0.6/σ

0.5/σ

0.4/σ
(a)

0.3/σ

0.2/σ

0.1/σ

0 r
0 σ 2σ 3σ 4σ

p(ψ)

1
(b)
2π

ψ
–π π

Figure 6.7 Statistical distributions of bandpass Gaussian noise voltage v n (t) having power 2𝜎 2 . (a) PDF p(r) of envelope r(t)
of v n (t); (b) PDF p(𝜓) of phase 𝜓(t) of v n (t).

0.5) to fall in the range (0, 𝜋), which – from Eq. (6.39) – makes vnQ (t) positive, as in the range (−𝜋, 0), which makes
vnQ (t) negative. A similar comment applies to vnI (t), which is positive when 𝜓(t) is in the range (−𝜋/2, 𝜋/2) and
negative otherwise.

6.4 System Noise Calculations

Let us now consider various ways of specifying the noisiness of communication systems and learn how to arrange
system components to minimise the effects of noise.
Telegram: @Persian_IEEEComSoc

6.4.1 Available Noise Power

A noisy resistor R can be modelled as shown in Figure 6.8. We replace
√ the practical resistor by an ideal noiseless
resistor of the same resistance. A noise generator of rms voltage v2n is connected in series, and accounts for the
noise generated by the practical resistor. With this model, a resistor R is clearly depicted as a noise voltage source.
The maximum-power transfer theorem states that a signal source, of internal resistance R, delivers maximum power
to a load that is matched to the source. Matching means that the load resistance equals R. Thus, the maximum
noise power, also called √the available noise power Pn , is obtained from resistor R under the condition shown in
Figure 6.9. With voltage v2n dropped across total resistance 2R, it follows that the rms noise current delivered to
 6.4 System Noise Calculations 449

Noisy resistor
noiseless resistor

R
R rms noise
ʋn2 voltage

Figure 6.8 Model of a noisy resistor.

Figure 6.9 Condition for maximum available noise power.

in = ʋn2 2R

R
R Load

ʋn2

the load resistance is given by

√
v2n
in = (6.40)
2R
and the (maximum) noise power delivered to the load is therefore

v2n 4kTBR
Pn = i2n R = R= R
4R2 4R2
= kTB
= −198.6 + 10log10 T + 10log10 B dBm (6.41)

Here, we substituted Eq. (6.40) for in and Eq. (6.2) for the mean square noise voltage. Temperature T is in kelvin
(K) and bandwidth B in Hz. Equation (6.41) is an important result, which states that the available thermal noise
power of a resistor depends only on the physical temperature T of the resistor and the bandwidth B of the receiving
system or load. That Pn does not depend on R may at first seem surprising. But note that, although a larger resis-
tance generates more noise according to Eq. (6.2), it also delivers proportionately less noise current to a matched
load, according to Eq. (6.40), by virtue of its increased resistance to current flow. Comparing Eqs. (6.41) and (6.21),
we see that the noise power per unit bandwidth (in W/Hz) is given by
Telegram: @Persian_IEEEComSoc

No = kT (6.42)

Equation (6.41) was derived for the case of thermal noise, where T represents the physical temperature of the
noise source. We now extend this result to apply to all types of random noise encountered in communication
systems by introducing the concept of equivalent noise temperature T e . Needless to restate, the system or load
bandwidth B referred to above and used in Eq. (6.41) is the noise equivalent bandwidth. Note also that Eq. (6.41)
may sometimes be written (especially within a link power budget table) simply as

Pn = k + T + B (6.43)

with the tacit understanding that all variables are in their logarithmic units such as k = −198.6 dBm/K/Hz, T in
dBK, B in dBHz, and hence Pn in dBm.
 450 6 Noise in Communication Systems

6.4.2 Equivalent Noise Temperature

The equivalent noise temperature T e of a system is the absolute temperature at which a noisy resistor has to be
maintained so that when it is connected to the input of a noiseless version of the system the available noise power
observed at the output of the system is exactly the same as that produced by all the noise sources in the actual sys-
tem. Equivalent noise temperature is a very important concept that allows us to quantify the noisiness of devices.
The following step-by-step procedure helps to fully explain the concept.

1. Consider a noisy system (Figure 6.10a) of input resistance Ri and output resistance Ro .
2. Measure the available noise power Pn at the output of the system, as shown in Figure 6.10b.
3. Replace the system with its noiseless version (Figure 6.10c). This gives a reading of zero on the noise power
metre at the output.
4. Connect a noisy resistor Ri at the input (Figure 6.10d) and vary the temperature of Ri until the output noise
power is again Pn . The absolute temperature of Ri at this instant is the equivalent noise temperature T e of the
noisy system.

What this means is that a resistor would have to be at a physical temperature T e in order to generate the same
amount of thermal noise power as the total noise power produced by a device of equivalent noise temperature T e .
Equivalent noise temperature, or noise temperature for short, is always referred to the input of the device. Thus, the
model of a noisy device of (numeric) gain G and noise temperature T e is as shown in Figure 6.11. It is important to
note that the noise power kT e B generated by (i.e. available from) the device according to Eq. (6.41) is at the device
input. Clearly then, the available noise power at the device output is given by

Pn = GkT e B (6.44)

A unit-gain device with a noise temperature T e = 290 K is called a standard noise source. Note that 290 K is a
reference temperature, which is universally taken to represent room temperature, and is denoted by T o . The only
exception is in Japan, where T o = 293 K.

(a) (b)

Noisy Same
System Ro Pn
System

Ri Ro
Telegram: @Persian_IEEEComSoc

(c) (d)

Noiseless Noiseless
version of Ro Pn = 0 Ri version of Ro Pn
same system same system

Vary temperature of Ri

Figure 6.10 Definition of equivalent noise temperature T e .

 6.4 System Noise Calculations 451

1 2 1 Noiseless 2
Noisy
kTeB version
system
of system
1ʹ (G, Te) 2ʹ 1ʹ 2ʹ
(G, 0)

Figure 6.11 Model of a noisy linear system of gain G and equivalent noise temperature T e .

6.4.3 Noise Figure of a Single System

Another important parameter for quantifying the noisiness of a device is its noise factor, which when expressed in
dB is referred to as the noise figure.

● The noise factor F of a system is defined as the ratio of actual noise power output when the input is a standard
noise source (T e ≡ T o = 290 K) to the noise power output that exists under the same conditions if the system is
noiseless. That is
Actual output noise power |
F= | (6.45)
Output noise power if system noiseless ||Input = Standard noise source To
Let us apply this definition to a linear system of gain G and noise factor F. As required by the definition, we let
the input be a source of signal power Sin and noise power N in = kT o B, which produces output signal power Sout
and output noise power N out . The signal-to-noise ratios SNRi at the system input and SNRo at the system output
are given by

SNRi = Sin ∕Nin ; SNRo = Sout ∕Nout

But if the system is noiseless, the output noise power would be GN in , which is simply the input noise power
multiplied by the gain G of the system. It follows from the above definition that the noise factor of this system is
given by
Nout
F=
GN in
Multiplying the right-hand side by Sin /Sin and manipulating leads to
Sin × Nout S × Nout
F= = in (Since Sout = GSin )
Sin × GN in Sout × Nin
S ∕N
= in in
Sout ∕Nout
Telegram: @Persian_IEEEComSoc

We see that this is the ratio between input and output SNR, which gives us the following alternative definition
of noise factor:

● The noise factor F of a system is the ratio between input and output SNR with the system input being a standard
noise source T o . That is
SNRi
F= (6.46)
SNRo
Since noise temperature T e and noise factor F both describe the noisiness of an electrical component or system,
it would be useful to establish how one parameter relates to the other. Figure 6.12 shows a linear system with
a standard noise source at its input terminals 1 − 1′ . To obtain an expression for the noise factor based on the
first definition (Eq. (6.45)), we need the noise power Pna at the output terminals 2 − 2′ of the actual noisy system
 452 6 Noise in Communication Systems

kToB

1 Noisy 2
(a) Rs kTeB
system Ro
ʋs 1ʹ (G, Te) 2ʹ
Standard
noise source Rs Ro Pna = GkTeB + GkToB

kToB

1 Noiseless 2
Rs Version of
(b) Ro
ʋs System Pnnv = GkToB
1ʹ (G, 0) 2ʹ
Standard
noise source Rs Ro

Figure 6.12 Relating noise factor F and equivalent noise temperature T e .

(Figure 6.12a), and the noise power Pnnv at the output of a noiseless version of the same system (Figure 6.12b).
Clearly

Pna = GkT o B + GkT e B

Pnnv = GkT o B

It follows from Eq. (6.45) that

Pna
F=
Pnnv
GkT o B + GkT e B
=
GkT o B
Te
=1+ (6.47)
To

Equation (6.47) relates two important parameters, namely noise factor F and noise temperature T e . Bearing in
mind that both T e and T o (= 290 K) are positive numbers, we see that in practical systems F always exceeds unity. F
gives the factor by which the proportion of the noise content of a signal is increased as the signal is passed through
a system. SNR therefore drops by this factor, hence the equivalence of the two definitions. In an ideal noiseless
system, T e = 0, and F = 1. We may rearrange Eq. (6.47) into the following useful form
Telegram: @Persian_IEEEComSoc

Te = To (F − 1) = 290(F − 1) (6.48)

And for later use, we also note that

Te
= F − 1. (6.49)
To

That is, the ratio between the noise temperature T e of a device and the reference noise temperature T o gives the
noise factor of the device less one.
Returning to Figure 6.12 and applying the second definition of noise factor given in Eq. (6.46), we let Si denote
the signal power at the input terminals 1 − 1′ and note that the input noise power associated with Si comes from
 6.4 System Noise Calculations 453

the standard noise source only. Thus, we may write

Input signal power = Si
Input noise power = kT o B
Output signal power = GSi
Output noise power = GkT o B + GkT e B

It follows from Eq. (6.46) that

SNRi Si ∕kT o B
F= =
SNRo GSi ∕(GkT o B + GkT e B)
= 1 + Te ∕To (6.50)

This is the same result as Eq. (6.47) and shows that the two definitions of noise factor are equivalent. The second
definition in terms of SNR is, however, more informative, allowing us to determine the change in SNR between the
input and output of a linear system of known noise factor. In fact, since F > 1, it means that SNRi always exceeds
SNRo in practical linear systems. See the exception below. In other words, the highest level of SNR is obtained at
the signal source. From that point onwards, SNR decreases each time the signal is processed in an amplifier. This
happens because the amplifier will boost the signal and noise applied at its input by the same factor to produce
the signal and noise at its output, but at the same time the amplifier will also inevitably inject its own internally
generated noise.
It should be noted, however, that improvement in SNR can be realised in a signal-processing device such as
some demodulators in which the input signal bandwidth is significantly larger than the output bandwidth. The
improvement is known as the processing gain and comes about because of the use of a modulation technique that
trades bandwidth for SNR improvement. Examples of the exploitation of this trade-off include spread spectrum
communication and frequency modulation (FM).

Worked Example 6.5 Noise Figure

An amplifier is quoted as having a noise figure of 3.5 dB. Determine its noise factor, noise temperature, and
noise power per unit bandwidth.

The relation between noise figure (dB) and noise factor (numeric) is
Figure∕10)
Noise factor = 10(Noise (6.51)

Thus

F = 10(3.5∕10) = 2.24
Telegram: @Persian_IEEEComSoc

and the noise temperature follows from Eq. (6.48)

Te = 290(2.24 − 1)
= 360 K

Finally, Eq. (6.42) gives the noise power per unit bandwidth as
No = kT e = 1.38 × 10−23 × 360
= 4.97 × 10−21 W∕Hz = 4.97 × 10−18 mW∕Hz
= −173 dBm ⋅ (Hz)−1
 454 6 Noise in Communication Systems

6.4.4 Noise Figure of Cascaded Systems

A communication system such as a receiver usually consists of several stages of noisy devices connected in series.
If there is impedance matching between stages then the series connection between the stages of the system is
described as a cascade connection. In that case, the output of one stage is entirely delivered without any reflection
into the next stage as the input of that next stage. So, if the kth stage of an n-stage (cascade connected) system has
gain Gk and we denote its input and output as xk and yk , respectively, it follows that
x2 = y1 = G1 x1
x3 = y2 = G2 x2 = G2 G1 x1
y3 = G3 x3 = G3 G2 G1 x1
⋮
yn = Gn xn = Gn Gn−1 · · · G3 G2 G1 x1

The overall gain G of a cascade connection of n stages is therefore

Output yn
G= =
Input x1
= Gn Gn−1 · · · G3 G2 G1 (ratio values)
= Gn + Gn−1 + · · · + G3 + G2 + G1 (dB) (6.52)

This process is illustrated in Figure 6.13a for n = 3 stages. It is a seemingly trivial result and states that overall
system gain is the sum of the dB gains of component stages of the system, but it must be emphasised that this is
only the best-case scenario achieved when all stages are matched in their impedances.
What is equally important and much less trivial is for us to quantify the noisiness of the overall system using one
noise temperature T and its corresponding noise factor F. Figure 6.13b shows the problem at hand for n = 3 stages.

x1 y1 x2 y2 x3 y3
(a) G1 G2 G3

Overall system gain, G

kToB G1kToB G1G2kToB G1G2G3kToB

Telegram: @Persian_IEEEComSoc

G1 G1kT1B G2 G1G2kT1B G3 G1G2G3kT1B

Pna
(b) T1 T2 T3
G2kT2B G2G3kT2B
F1 F2 F3
G3kT3B

Overall system (G, T, F)

Figure 6.13 Cascade connection of n = 3 stages. (a) Overall gain G; and (b) overall noise temperature T and noise factor F
of the system.
 6.4 System Noise Calculations 455

Each stage has gain, noise factor, and noise temperature identified by subscripts 1, 2, and 3. The overall system (of
gain G = G1 G2 G3 ) is enclosed in a shaded box, and we wish to determine its equivalent noise temperature T and
noise factor F.
As required by the definition, we connect a standard noise source of noise power kT o B at the input of the overall
system and determine the noise power at its output. First, we obtain the output Pna for the actual system, and
then the output Pnnv when every stage is replaced by its noiseless version. The ratio of the two noise powers is the
required noise factor F of the overall system. You should walk through the cascade connection stage by stage to
verify how Pna is obtained at the final stage. Each stage produces an output noise power that consists of the noise
power at its input multiplied by its gain, plus the noise internally generated in that stage. For noiseless stages, the
only noise output is the input noise power multiplied by the gain of the cascade connection. Thus
Pna = G1 G2 G3 kT o B + G1 G2 G3 kT 1 B + G2 G3 kT 2 B + G3 kT 3 B
Pnnv = G1 G2 G3 kT o B
and it follows that
G G G kT B + G1 G2 G3 kT 1 B + G2 G3 kT 2 B + G3 kT 3 B
F= 1 2 3 o
G1 G2 G3 kT o B
T1 T2 T3
=1+ + +
To G1 To G1 G2 To
F − 1 F3 − 1
= F1 + 2 + (6.53)
G1 G1 G2
Here, we have used Eq. (6.49) in the last step so that T 1 /T o = F 1 − 1, T 2 /T o = F 2 − 1, and T 3 /T o = F3 − 1.
Equation (6.53) gives the noise factor of the overall system in terms of the noise factor and gain of each stage.
We can obtain a similar expression for the noise temperature T of the system by rearranging the second line of the
above equation and making use of the relation between T and F given in Eq. (6.48)
T = To (F − 1)
T T3
= T1 + 2 + (6.54)
G1 G1 G2
It is clear from the above discussion that in the general case of a cascade connection of n noisy devices, the
system noise factor F and noise temperature T are given by
T T3 T4 Tn
T = T1 + 2 + + +···+
G1 G1 G2 G1 G2 G3 G1 G2 G3 · · · Gn−1
F − 1 F3 − 1 F −1 Fn − 1
F = F1 + 2 + + 4 +···+ (6.55)
G1 G1 G2 G1 G2 G3 G1 G2 G3 · · · Gn−1
Let us take a moment to understand the implication of this important result on the design of communication
receivers to minimise their noisiness. The (overall) noise factor of the receiver is dominated by the contribution
Telegram: @Persian_IEEEComSoc

of the first stage, i.e. the first device seen by the incoming signal. The contribution from each subsequent stage
is reduced by the product of the gains of all preceding stages. Therefore, to minimise the noise factor (or noise
temperature) of a receiver we must place a low-noise high-gain amplifier at the front end or first stage of the
receiver, hence the ubiquitous front-end low-noise amplifier (LNA). In fact, if the gain G1 of this amplifier is high
enough then F ≈ F 1 and the receiver is practically immune to the effect of noisy components, such as mixers, that
follow the first stage.

Worked Example 6.6 Arrangement of Receiver Stages

Figure 6.14 shows three different arrangements of the same four stages S1 , S2 , S3 , and S4 of a receiver. We wish
to examine the noise figure and noise temperature of the receiver under each arrangement.
 456 6 Noise in Communication Systems

Input signal Output signal

(a) S1 S2 S3 S4

G1 = –2dB G2 = 30dB G3 = –4dB G4 = 20dB

F1 = 2.0dB F2 = 0.8dB F3 = 6.0dB F4 = 3.5dB

(b) Input signal Output signal

S2 S1 S3 S4

G2 = 30dB G1 = –2dB G3 = –4dB G4 = 20dB

F2 = 0.8dB F1 = 2.0dB F3 = 6.0dB F4 = 3.5dB

Input signal Output signal

(c) S4 S1 S3 S2

G4 = 20dB G1 = –2dB G3 = –4dB G2 = 30dB

F4 = 3.5dB F1 = 2.0dB F3 = 6.0dB F2 = 0.8dB

Figure 6.14 Worked Example 6.6: Arrangement of receiver stages.

The gain and noise figure of the receiver stages are expressed in dB in Figure 6.14. Before applying Eq. (6.55),
we must express these values as ratios

F1 = 2 dB = 1.585; G1 = −2 dB = 0.631
F2 = 0.8 dB = 1.202; G2 = 30 dB = 1000
F3 = 6 dB = 3.981; G3 = −4 dB = 0.398
F4 = 3.5 dB = 2.239; G4 = 20 dB = 100
Telegram: @Persian_IEEEComSoc

(a) For the S1 S2 S3 S4 arrangement we obtain

F2 − 1 F3 − 1 F −1
F = F1 + + + 4
G1 G1 G2 G1 G2 G3
1.202 − 1 3.981 − 1 2.239 − 1
= 1.585 + + +
0.631 0.631 × 1000 0.631 × 1000 × 0.398
= 1.915 = 2.82 dB

And Te = 290(F − 1) = 265 K

 6.4 System Noise Calculations 457

(b) For the S2 S1 S3 S4 arrangement we obtain

F1 − 1 F3 − 1 F −1
F = F2 + + + 4
G2 G2 G1 G2 G1 G3
1.585 − 1 3.981 − 1 2.239 − 1
= 1.202 + + +
1000 1000 × 0.631 1000 × 0.631 × 0.398
= 1.213 = 0.84 dB
And Te = 290(F − 1) = 62 K
(c) Finally, for the S4 S1 S3 S2 arrangement we obtain
F1 − 1 F3 − 1 F −1
F = F4 + + + 2
G4 G4 G1 G4 G1 G3
1.585 − 1 3.981 − 1 1.202 − 1
= 2.239 + + +
100 100 × 0.631 100 × 0.631 × 0.398
= 2.30 = 3.62 dB
And Te = 290(F − 1) = 377 K
The above results show that the noise temperature T e of the receiver depends crucially on the order of arrange-
ment of the stages. Typically, S1 would be a lossy waveguide or coaxial cable connecting an outdoor antenna
to the rest of the receiver block located indoors; S2 would be an LNA, S3 a mixer, and S4 another (noisier)
amplifier.
If the LNA is the very first stage seen by the input signal (from the antenna), achieved by locating it outdoors
with the antenna then we have the S2 S1 S3 S4 order in Figure 6.14b and the lowest possible noise temperature
(62 K in this case). Note how the overall noise figure of the receiver (0.84 dB) is only marginally worse than
that of the LNA (0.8 dB).
If the lossy feed run precedes the LNA, as in the S1 S2 S3 S4 order in Figure 6.14a, we have a higher noise tem-
perature (= 265 K), which corresponds to an increase of 10log10 (265/62) = 6.3 dB in noise power compared to
the previous arrangement.
The situation is worse still with the S4 S1 S3 S2 arrangement shown in Figure 6.14c. Here the LNA is relegated
to the last stage. As a result, T e = 377 K and the noise power is about 8 dB higher than could be achieved using
a better ordering of the same set of receiver stages.
An increase in noise power leads to a reduction in SNR by the same amount. Thus, in the design of communi-
cation receivers a high-gain LNA must be used at the front end, and every effort must be made to keep losses
preceding the LNA to a minimum. Losses and noise in devices that follow a high-gain LNA have a negligible
effect on the noise temperature of the receiver.
Telegram: @Persian_IEEEComSoc

6.4.5 Overall System Noise Temperature

We are now able to determine the noise temperature T sys of the overall receiver system. This allows us to place a
single noise source of temperature T sys at a reference point in the overall system. The system is then replaced by
its noiseless version for all purposes, such as when determining the C/N at the demodulator input. The overall
system noise temperature T sys accounts for all noise in the receiver. This includes:
● Antenna noise, represented by noise temperature T a computed using Eq. (6.10).
● Noise from a lossy feed run (a coaxial cable or waveguide) that connects the antenna to a LNA.
● Receiver noise, represented by noise temperature T e computed using Eq. (6.55). The receiver is usually a cascade
connection of several stages starting from the LNA.
 458 6 Noise in Communication Systems

Reference
Antenna Ta
point

(a)
Tf

IF. Amp.
Mixer
LNA

etc.
Feed Run: Loss = Lf

Receiver, Te

Tsys

kTaB
kTaB/L
From antenna
(b) Feed Run kTf B/L Receiver
From feed (L, Tf) (G, Te)
kTf B

kTeB
Pn = kTsysB

Figure 6.15 Overall noise temperature T sys of receiver system: (a) overall receiver system; (b) noise model.

Figure 6.15 shows the overall receiver system. It is standard practice to set the reference point at the LNA input
as indicated. So, unless otherwise specified, you may safely assume that a given value of T sys refers to this point.
However, provided there is consistency, the same SNR will be obtained using some other point as reference (such
as the antenna output port just before the feed run). To determine T sys , we calculate the total noise power Pn at
the reference point and equate this to kT sys B.
A self-explanatory noise model of the overall system is shown in Figure 6.15b. It is assumed that all stages are
matched so that available noise power is delivered at every point. The noise factor F f of a passive attenuator such
as an antenna feed cable is equal to its numeric loss Lf . Therefore, the noise temperature T f of the feed run is

Tf = To (Ff − 1) = To (Lf − 1) (6.56)

The feed run transmits noise from its input to output reduced by the factor Lf . At the reference point, the total
noise power is therefore
Telegram: @Persian_IEEEComSoc

kT a B kT f B
+ + kT e B ≡ kT sys B
Lf Lf
This gives the desired overall system noise temperature

Tsys = Te + (Ta + Tf )∕Lf

= Te + Ta ∕Lf + To (1 − 1∕Lf ) (6.57)

where we have used Eq. (6.56) for T f .

If there is no loss between the antenna and the receiver, Lf = 1, so that

Tsys = Te + Ta (6.58)
 6.4 System Noise Calculations 459

Ta = 62 K Receiver, Te
Receiving
antenna
To
Lf = 1.58 demodulator
LNA Mixer IF amp.
Feed Run

G1 = 25dB G2 = –4dB G3 = 30dB

F1 = 0.8dB F2 = 6.0dB F3 = 3.5dB

Tsys

Figure 6.16 Earth station receiver.

Equation (6.58) holds, for example, where the LNA is mounted in a waterproof unit just under the antenna
(and referred to as a low-noise block [LNB]) so that the feed length between antenna and LNA is negligible. In this
case, there will be a cable leading from the output of the LNB to the remaining sections of the receiver inside a
building. The noise effect of this cable is included in the computation of T e using Eq. (6.55), with the cable serving
as the second stage of the cascade connection, which therefore has noise temperature and gain values T 2 = T f and
G2 = 1/Lf .

Worked Example 6.7 Overall System Noise Temperature

We wish to calculate the overall system noise temperature T sys of the satellite communication earth station in
Figure 6.16.

Converting the dB values of gain and noise figure in Figure 6.16 to ratios, we have G1 = 316.23, F 1 = 1.202,
G2 = 0.3981, F 2 = 3.981, and F 3 = 2.239. Putting these values into Eq. (6.55) yields the noise temperature T e of the
receiver block (enclosed in the dotted rectangle in Figure 6.16) as
F − 1 F3 − 1
F = F1 + 2 +
G1 G1 G2
3.981 − 1 2.239 − 1
= 1.202 + +
316.23 316.23 × 0.3981
= 1.222
Te = 290(F − 1) = 64 K
The overall system noise temperature then follows from Eq. (6.57)
Telegram: @Persian_IEEEComSoc

Tsys = Te + Ta ∕Lf + 290(1 − 1∕Lf )

= 64 + 62∕1.58 + 290(1 − 1∕1.58)
= 210 K
Note that, by co-locating the LNA with the antenna, the overall system noise temperature of the earth station
could be reduced to just 64 K + 62 K = 126 K, achieving a reduction of 2.2 dB in noise power.

6.4.6 Signal-to-noise Ratio

We now have all the information required to determine the ratio between the power in the received signal at the
input of the demodulator and the noise power at the same point. For a fixed-link line-of-sight transmission system,
 460 6 Noise in Communication Systems

such as in microwave radio relay and satellite communications, the received signal power Pr at the reference point
in the receiver is obtained by extending Eq. (5.126), which includes only free space path loss (FSPL), to incorporate
the following extra losses:

● Attenuation in the atmosphere, Latm .

● All losses Lr in the receiver before the reference point. The reference point will usually be at the LNA input. We
include in Lr , losses due to antenna mis-pointing and polarisation mismatch, antenna feed losses, etc.

Thus, at the reference point

Pr = EIRP − Ls − Latm + Gr − Lr (dBW)

= EIRP − Ls − Latm + G (dBW) (6.59)

where G = Gr − Lr represents the net gain of the receive-antenna. The noise power Pn at this point is given by

Pn = kT sys B = −228.6 + 10log10 (Tsys ) + 10log10 (B) dBW (6.60)

Subtracting Eq. (6.60) from (6.59) yields the SNR of the entire receiver system up to the demodulator input, also
denoted C/N – for carrier-to-noise ratio
( )
G
C∕N = EIRP − Ls − Latm + 228.6 + 10 log − 10 log B
Tsys
( )
G
C∕No = EIRP − Ls − Latm + 228.6 + 10 log (6.61)
Tsys
This is an important result, which should be well understood. C/N o is the ratio between received signal power
and noise power per unit bandwidth. Effective isotropically radiated power (EIRP) is the transmit power (dBW)
plus net transmit-antenna gain (dB). Ls is the FSPL given by Eq. (5.123). G is the net gain of the receive-antenna
(expressed as a ratio). The ratio between G and the system noise temperature T sys is a figure of merit, denoted G/T,
that gives some indication of a receiving system’s ability to handle weak signals. It is usually taken for granted
that all quantities in Eq. (6.61) have been expressed in their correct logarithmic units, so that we may simply
write

C∕N = EIRP + G∕T + 228.6 − Ls − Latm − B

C∕No = EIRP + G∕T + 228.6 − Ls − Latm (6.62)

We emphasise that the SNR calculated as above at the reference point is maintained at every point in the receiver
up to the demodulator input, since all noise has been accounted for in T sys , which allows all receiver stages to be
subsequently treated as noiseless. Furthermore, most transmission systems employ carrier modulation in which
the modulated carrier amplitude and hence the transmitted carrier (TC) power may change from one symbol
Telegram: @Persian_IEEEComSoc

interval to the next. The received signal power Pr in Eq. (6.59), and hence C/N in Eq. (6.62), should therefore be
understood to be averaged over all transmission symbols. It is also conventional to use the term carrier-to-noise
ratio (C/N) to describe the ratio between signal power and noise power at all points in the transmission system
up to the demodulator input, and the term signal-to-noise ratio (SNR) at the demodulator output and beyond after
the signal has been recovered from the carrier.
The C/N of any communication receiver may be determined by using applicable methods to determine total path
attenuation and hence Pr (see Sections 5.2 for metallic lines, 5.4 for optical fibre, and 5.5 for radio) and to calculate
noise power Pn (see Section 6.4) at the reference point of the system. C/N in dB is then obtained by subtracting
Pn in dBm or dBW from Pr expressed in the same logarithmic unit as Pn . These calculations are usually laid out
in tabular form, called the link power budget, to compactly include all contributory loss and gain factors. Worked
Example 6.8 illustrates this approach for a terrestrial mobile radio link.
 6.4 System Noise Calculations 461

Worked Example 6.8 Link Power Budget of Mobile Radio System

A mobile radio transmission system has the following specifications:

● Coverage range = 4 km.
● Free space propagation break point = 100 m.
● Path loss exponent in nonfree-space propagation region, n = 4.
● Carrier frequency = 950 MHz.
● Bandwidth = 25 kHz.
● Transmit-antenna gain = 12 dB.
● Feed losses at transmitter = 3 dB.
● Receive-antenna gain = −4 dB.
● Noise figure of receiver = 5 dB.
If the required minimum C/N is 10 dB, determine the minimum transmit power needed to provide a fade
margin of 16 dB.

The solution is laid out in Table 6.2, all quantities being expressed in appropriate logarithmic units and signs
(positive sign for positive gains and powers, and negative sign for positive loss). The table has two sections, the
signal power budget followed by a noise power budget. The transmit power Pt is unknown and to be determined,
so it is entered as Pt (dBm) in the table. Path loss for this type of link is given by Eq. (5.135), which is used as
shown in the table to compute Lp . The stipulated fade margin of 16 dB is entered as a negative value in the signal
power budget since we are designing the link to withstand this extra loss on top of the computed path loss. System
noise temperature T sys is given by Eq. (6.58), with antenna noise T a entered as 300 K to include radiation from
the surface of the earth picked up in its entirety by the omnidirectional receive-antenna without any sidelobe loss

Table 6.2 Worked Example 6.8: mobile radio link power budget.

Quantity Computation Value

Signal power budget

Pt , Transmit output power To be determined Pt , dBm
Gt , Transmit-antenna gain Given +12 dB
Lft , Transmit feed losses Given −3 dB
Lp , Path loss 92.44 + 20logf c + 10nlogd − 10(n − 2)logd1 −136.1 dB
where d1 = 0.1 km, d = 4 km, f c = 0.95 GHz, n = 4
Telegram: @Persian_IEEEComSoc

M, Fade margin Given −16 dB

Gr , Receive-antenna gain Given +(−4) dB
Pr , Received signal power Algebraic sum of above six quantities Pt − 147.1 dBm
Noise power budget
k, Boltzmann’s constant 1.38 × 10−23 W/K/Hz = 1.38 × 10−20 mW/K/Hz −198.6 dBm/K/Hz
5/10
T sys , System noise temp T a + T e = 300 + T o (F e − 1) = 300 + 290(10 –1) = 927.1 K 29.7 dBK
Bandwidth 25 kHz (Given) 44 dBHz
N, Noise power Algebraic sum of above three quantities −125 dBm
C/N Pr − N = Pt − 147.1 + 125 = Pt − 22.1 10 dB (Given)
 462 6 Noise in Communication Systems

mitigation, plus small extra-terrestrial and atmospheric noise contributions to give T a ≈ 300 K. The given noise
figure F e = 5 dB is converted to noise temperature T e in the usual manner.
The final row of the table gives C/N = Pt − 22.1, which, when equated to 10 dB as specified, yields minimum
transmit power Pt = 32.1 dBm = 1.622 W.

6.5 Noise Effects in Communication Systems

There is a wide range of parameters by which the performance or quality of a communication system may be
described. These parameters may be objective (i.e. quantitative), such as SNR, BER, bit rate, outage probability,
link availability, delay or latency, delay variation, etc. The parameters may be subjective (i.e. qualitative), expressed
through opinion ratings, such as the mean opinion score (MOS) for speech quality on a scale of 1 (for bad) to 5 (for
excellent), the aesthetic quality of a mobile handset, quality of service (QoS) perceived by a customer, etc. Noise
will directly or indirectly exert some influence on most of the objective parameters. For example, the indirect
influence of noise on delay is that noise could lead to detected errors in a received data packet, which necessitates
a retransmission request, which increases packet delay. We are interested here only in briefly discussing the direct
impact of noise on the SNR in analogue communication systems and the BER in digital communication systems.

6.5.1 SNR in Analogue Communication Systems

The SNR at the demodulator output of an analogue communication system is illustrated in Figure 6.17a as
expressed by
SNR = C∕N + Gp (6.63)
where C/N is the carrier-to-noise ratio at the demodulator input, given for example by Eq. (6.62), and Gp is the
processing gain of the demodulator, which is the amount by which the SNR at the demodulator output exceeds
the C/N at its input. The recovered message signal (at demodulator output) is perceived as a clean signal with no
discernible corruption by noise if SNR ≥ 30 dB for voice and SNR ≥ 45 dB for video. Given an analogue modula-
tion scheme, the link must therefore be designed to deliver a level of C/N at the demodulator input which can be
converted into an acceptable SNR by the demodulator through the addition of a processing gain. Analogue mod-
ulation schemes include phase (PM), frequency (FM), and amplitude (AM) modulations. AM schemes (discussed
in the next chapter, including single sideband (SSB) and double sideband (DSB) – suppressed (SC) and transmit-
ted (TC) carriers) have very small processing gains that do not possess meaningful trade-off flexibility. Phase and
frequency modulations, on the other hand, allow bandwidth to be traded for a higher processing gain and hence

Modulated carrier Message signal

Telegram: @Persian_IEEEComSoc

+ Noise Analogue + Noise

(a)
demodulator
C/N SNR = Gp + C/N

Modulated carrier Message bitstream

+ Noise Digital + Bit errors
(b) demodulator/decoder
(ModCod)
C/N BER = f(C/N, ModCod)

Figure 6.17 Noise effect in communication systems.

 6.5 Noise Effects in Communication Systems 463

an improved signal quality. It is shown in Chapter 8 that, provided the C/N at its input is at least C/N ≈ 9.5 dB, an
FM demodulator delivers a processing gain
Gp = 7.78 + 10 log[𝛽 2 (𝛽 + 1)] − 20 log(R) dB
B
𝛽 = c −1 (6.64)
2fm
√
where R is the peak-to-rms ratio of the message signal (e.g. R = 2 for a sinusoidal message signal), 𝛽 is the modu-
lation index of the FM signal, Bc is Carson’s bandwidth of the FM signal (which is a fractional power containment
bandwidth that contains at least 98% of the total power of the FM signal), and f m is the maximum frequency of
the baseband message signal (and hence the message bandwidth). Assuming f m = 15 kHz, Eq. (6.64) shows, for
example, that we may increase Gp and hence SNR by around 13 dB by doubling transmission bandwidth from 60
to 120 kHz. Thus, FM provides a means of sacrificing bandwidth to reconstruct a good-quality message signal from
a low-C/N received signal. In the days of analogue transmission technology which continued into the 1980s, this
feature made FM the enabling modulation technique for satellite communications where link C/N is typically low
due to long path lengths.
Amplitude modulation (AM) is the subject of the next chapter, but before this we may quickly determine the pro-
cessing gain of double sideband suppressed carrier (DSBSC) modulation based on the receiver model in Figure 6.18
and our earlier discussion of narrowband noise in Section 6.3.3. Consider a message signal g(t) of bandwidth B
and power
P = g2 (t) = A2rms
where Arms is the rms value of the signal. DSBSC modulation is achieved by multiplying a sinusoidal carrier signal
of frequency f c ≫ B with this signal to obtain a DSBSC signal
vdsb (t) = 𝛾g(t) cos(2𝜋fc t)
The scale factor 𝛾 is introduced simply for dimensional consistency. We know (from Chapter 4) that the effect of
multiplying g(t) by cos(2𝜋f c t) is to scale the spectrum G(f ) of g(t) by one-half and then shift it from being centred
on f = 0 to being centred on f = ±f c . Since f c ≫ B, there is no overlap between the two replicated spectra, so vdsb (t)
has bandwidth 2B and total power (in the upper and lower sidebands)
Pdsb = 2(𝛾∕2)2 A2rms = 𝛾 2 A2rms ∕2 (6.65)
In Figure 6.18, the BPF has bandwidth 2B centred on f c , just wide enough to pass vdsb (t) to the demodulator input
along with narrowband noise vn (t), the expression of which is given in Eq. (6.32). Since N o is the noise power per
unit bandwidth, the noise power at demodulator input is
Pni = 2No B (6.66)
The ratio between Pdsb and Pni gives C/N at demodulator input as
Telegram: @Persian_IEEEComSoc

Pdsb 𝛾 2 A2rms
C∕N = = (6.67)
Pni 4No B

Coherent DSB demodulator

Incoming signals BPF ʋa(t) ʋb(t) LPF ʋc(t) Message

+ (fc – B → fc + B) a × (0 → B) c
b signal + noise
White noise cos(2π fct)
w(t)

Figure 6.18 AM DSBSC receiver model.

 464 6 Noise in Communication Systems

The signals at points a, b, and c in Figure 6.18 are as follows. At point a

va (t) = vdsb (t) + vn (t)
= 𝛾g(t) cos(2𝜋fc t) + vnI (t) cos(2𝜋fc t) − vnQ (t) sin(2𝜋fc t)
At point b
vb (t) = va (t) × cos(2𝜋fc t)
}
1 1
= 𝛾g(t) + vnI (t) Passed by LPF
2 2

1 ⎫
+ 𝛾g(t) cos(4𝜋fc t) ⎪
2 ⎪
1 ⎪
+ vnI (t) cos(4𝜋fc t) ⎬ Blocked by LPF
2 ⎪
1 ⎪
− vnQ (t) sin(4𝜋fc t)⎪
2 ⎭
At point c (demodulator output)
1 1
vc (t) = 𝛾g(t) + vnI (t)
2 2
where the first term is the scaled message signal of power
1
Pm = 𝛾 2 A2rms (6.68)
4
The second term represents noise in the recovered message signal. We know from Eq. (6.34) that vnI (t) is of the
same power (= 2N o B) as the narrowband noise vn (t), and since a scale factor of one-half on amplitude corresponds
to a scale factor of one-quarter on power, the noise power at demodulator output is
1 1
Pno = × 2No B = No B (6.69)
4 2
The SNR at demodulator output is the ratio between message power Pm in Eq. (6.68) and noise power Pno in
Eq. (6.69). Thus
Pm 𝛾 2 A2rms ∕4 𝛾 2 A2rms
SNR = = = (6.70)
Pno No B∕2 2No B
Finally, DSBSC processing gain is the ratio between this SNR and the C/N in Eq. (6.67). Thus
SNR 𝛾 2 A2rms 𝛾 2 A2rms
Gp = = ∕
C∕N 2No B 4No B
Telegram: @Persian_IEEEComSoc

= 2 = 3 dB (6.71)
By a similar analysis, we may show that the processing gain of SSB AM is Gp = 0 dB. The 3 dB improvement in
SNR achieved by a coherent DSBSC demodulator comes about because there are two message sidebands in vdsb (t),
which are translated down to baseband and added coherently. This doubles the message signal power within
the message bandwidth, whereas noise power is not doubled since its quadrature component is rejected by the
coherent demodulation process. However, the noise power at the input of a DSB demodulator is higher than that
at the input of an SSB demodulator by 3 dB since DSB operates at twice the transmission bandwidth. Thus, for the
same signal power, C/N for DSB transmission will be 3 dB lower than that of SSB so that, with a 3 dB processing
gain in the DSB case, equal-power DSB and SSB transmissions will have the same SNR at their demodulator
outputs. One final point to emphasise is that the processing gains of AM schemes are fixed, and, unlike FM, there
is no scope to trade bandwidth for an improved SNR.
 6.5 Noise Effects in Communication Systems 465

6.5.2 BER in Digital Communication Systems

One measure of the quality or performance of a digital communication system is the BER. This is the number of
bit errors in a recovered message bit stream within a sufficiently large transmission interval divided by the total
number of recovered message bits in the same interval. BER is also referred to as the bit error rate in some literature.
As illustrated in Figure 6.17b, BER depends on three factors, namely the digital modulation scheme, the C/N at
the input of the digital demodulator, and the error control coding scheme. The combination of modulation and
error control coding schemes is often referred to as modcod. The interplay of these three factors along with signal
power and transmission bandwidth in supporting digital transmission at a required message bit rate with errors
not exceeding a specified BER are discussed below.
In digital transmission, one of M unique symbols (each being an energy signal of duration T s ) is transmitted in
each signalling interval of duration T s . M is usually an integer power of 2, and each symbol represents up to log2 M
message bits. The symbol rate Rs and message bit rate Rb are given by

Rb = Rs log2 M

⎧ Bocc , APSK
1 ⎪𝛼 + 1
Rs = =⎨ (6.72)
Ts ⎪ Bocc
⎩ 𝛼 + (M + 1)∕2 , FSK

where 0 ≤ 𝛼 ≤ 1 is the roll-off factor of the raised cosine filter used in the transmission system, Bocc is the transmis-
sion (occupied) bandwidth, APSK refers to amplitude and phase shift keying digital modulation (of which phase
shift keying (PSK) and amplitude shift keying (ASK) are special cases) and FSK is the frequency shift keying modula-
tion. The raised cosine filter is defined in Worked Example 4.16, and the minimum value 𝛼 = 0 corresponds to an
ideal brick wall filter response which is not realisable in real time because it requires future inputs to contribute
to the present output. The ratio Rb /Bocc is the message bit rate achieved per hertz of transmission bandwidth and
is known as the bandwidth efficiency 𝜂 in bits per second per hertz (b/s/Hz). This follows from Eq. (6.72) as

⎧ log M
⎪ 2 , APSK
Rb ⎪ 𝛼+1
𝜂= =⎨ (6.73)
Bocc ⎪ log2 M
, FSK
⎪ 𝛼 + (M + 1)∕2
⎩

Figure 6.19 shows bandwidth efficiencies of M-ary APSK and M-ary FSK modulation schemes (at M = 2–1024)
for transmission using a raised cosine filter of roll-off factor 𝛼 = 0. We note that M-ary APSK uses bandwidth more
Telegram: @Persian_IEEEComSoc

efficiently than M-ary FSK and that its bandwidth efficiency increases with M, whereas the bandwidth efficiency
of M-ary FSK decreases as M increases beyond M = 4.
For a given modulation scheme, BER depends on the C/N at the demodulator input and decreases monotonically
as C/N increases. Noting that received signal power Pr may be expressed as

Energy Energy bit

Pr = = × = E b Rb
second bit second

and that, using Eq. (4.176) for the noise equivalent bandwidth B of a raised cosine filter, noise power Pn is given
by

Pn = No B = No (1 − 𝛼∕4)Rs = No (1 − 𝛼∕4)Rb ∕log2 M

 466 6 Noise in Communication Systems

10

6
M-ary
Bandwidth Efficiency, η = Rb/Bocc

4 APSK

0
0.8

0.6

0.4 M-ary
FSK

0.2

0
2 4 8 16 32 64 128 256 512 1024
M

Figure 6.19 Bandwidth efficiency of M-ary APSK (including PSK and ASK) and M-ary FSK for transmission using raised
cosine filter of roll-off factor 𝛼 = 0.

we may relate C/N to the ratio Eb /N o between average energy per bit Eb and noise power per unit bandwidth N o as
C P E b Rb
= r =
N Pn No (1 − 𝛼∕4)Rb ∕log2 M
Eb log2 M
= × , (M-ary APSK & FSK) (6.74)
No (1 − 𝛼∕4)
Thus, for 𝛼 = 0, C/N is, respectively, 3, 6, and 7.8 dB larger than Eb /N o for M = 4, 16, and 64. At the other end of
roll-off factor 𝛼 = 1, C/N is, respectively, 4.3, 7.3, and 9 dB larger than Eb /N o for M = 4, 16, and 64. The nature of the
dependence between BER and Eb /N o is developed in Chapter 11. For now, it will suffice to consider Figure 6.20,
which is a plot of BER versus Eb /N o (dB) for various modulation schemes and values of M in a transmission
medium that is affected only by additive white Gaussian noise (AWGN). We observe the following:
Telegram: @Persian_IEEEComSoc

● For all modulation schemes and values of M, when Eb /N o (and hence C/N) is increased the BER is guaranteed to
decrease. This is the monotonic relationship earlier referred to. Communication services would normally have
a specified maximum (or threshold) BER for the QoS to be considered acceptable, for example 10−7 for data and
10−4 for voice. Therefore, one way to achieve a specified BER threshold on a transmission link is to increase the
C/N of the link to the level required by the modulation scheme employed.
● For a given type of APSK modulation scheme (including PSK and ASK) and a fixed BER (such as BER = 10−7 ),
the required Eb /N o (and hence C/N) increases as M increases. For example, to achieve BER = 10−7 , the respec-
tive values of Eb /N o (dB) required by 4PSK, 8PSK, 16PSK, and 64PSK are 11.3, 14.8, 19.3, and 29.3, which (from
Eq. (6.74) with 𝛼 = 0) correspond to C/N (dB) values 14.3, 19.5, 25.3, and 37.1. What is happening here is that
M-ary APSK facilitates a trade-off between bandwidth and signal power, which allows bandwidth to be used
more efficiently at the expense of signal power. For example, with 64PSK we can send up to 6 bits/second per
 6.5 Noise Effects in Communication Systems 467

0.1

10–2

10–3

16

16
64
4FS
16F

PS

AS
PS
10–4

K
K
SK

K
6 4 FS

64A
4PS

16A
2AS
BER

PS
K≡
K

10–5

PSK

K
1024F

4AS
QPS

≡O

K
10–6
SK

OK

8PS
10–7
K

10–8

10–9
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32
Eb/No, dB

Figure 6.20 Bit error ratio (BER) of various modulation schemes in an AWGN channel.

hertz of bandwidth, whereas quadriphase shift keying (QPSK) allows only up to 2 bits/second per hertz of trans-
mission bandwidth. Thus, bandwidth efficiency improves by a factor of 3 in switching from QPSK to 64PSK,
and the price paid for this improved bandwidth efficiency is that signal power must be increased by 22.8 dB
(obtained as the difference between required C/N for 64PSK and QPSK).
● In M-ary FSK and at a fixed BER, the required Eb /N o decreases as M increases. For example, to achieve
BER = 10−7 , the respective values of Eb /N o (dB) required by 4FSK, 16FSK, 64FSK, and 1024FSK are 11.5, 8.9,
7.5, and 5.9. M-ary FSK facilitates an inverse trade-off between bandwidth and signal power when compared
to M-ary APSK schemes. If the required bit rate and BER are fixed then, as M increases, M-ary FSK uses more
bandwidth and a lower Eb /N o . This allows signal power to be used more efficiently at the expense of bandwidth.
It is this feature of M-ary FSK that makes it an effective modulation technique in deep space communication
scenarios, where received signal power (and hence Eb /N o ) is typically very small.
● M-ary APSK (for M > 4) requires lower Eb /N o than M-ary PSK, which in turn requires smaller Eb /N o than M-ary
ASK to achieve a given BER and bit rate using the same transmission bandwidth. The order of power efficiency
is thus APSK > PSK > ASK. For example, for a BER of 10−7 , 16ASK requires Eb /N o = 30 dB, whereas 16PSK
Telegram: @Persian_IEEEComSoc

and 16APSK require Eb /N o = 19.3 and 15.7 dB, respectively. M-ary ASK employs M unique sinusoidal signals
of duration T s (referred to as symbols) having the same frequency f c and phase and differing only in amplitude
drawn from the set {0, A, 2A, 3A, …, (M − 1)A}, where A is some constant. The M symbols of M-ary PSK have the
same frequency and amplitude but differ in phase drawn from the set {𝜃, 𝜃 + 2𝜋/M, 𝜃 + 4𝜋/M, …, 2𝜋(M − 1)/M},
where 𝜃 is some angular offset, e.g. 𝜃 = 𝜋/4 rad. As M increases, the average signal power of M-ary ASK increases
if the amplitude spacing is kept constant in order to maintain the same BER. Also, as M increases, the amplitude
(and hence power) of M-ary PSK must be increased in order to maintain the demodulator’s ability to distinguish
between adjacent phases in the presence of AWGN. M-ary APSK allows more flexibility in choosing amplitude
and phase combinations to minimise detection errors in the presence of noise. The 16APSK scheme shown in
Figure 6.20 uses a 4 + 12 arrangement in which four phase states of the carrier are transmitted with the same
amplitude A1 and the remaining 12 phase states are transmitted with a larger amplitude 2.57A1 . The 64APSK
 468 6 Noise in Communication Systems

scheme is based on a 4 + 12 + 20 + 28 arrangement with amplitudes A1 , 2.4A1 , 4.3A1 , and 7A1 of the respective
groups of carrier phase states. These APSK schemes are defined in the DVB-S2X standard [2].
The value of Eb /N o needed to satisfy a specified BER threshold may be reduced, thereby improving the power
efficiency of the transmission system, by introducing error control coding. Prior to modulation, redundant bits are
systematically introduced into the message bit stream at a code rate r (which means that a fraction r of the error
control coded bits are message bits and the remaining fraction (1 − r) are redundant bits). The link may then be
designed for a lower Eb /N o at the demodulator input which produces a higher than acceptable ratio of errors at the
demodulator output. However, when this errored bit stream is fed into the error control decoder at the receiver,
most of the errors are corrected with the aid of the redundant bits and this reduces BER down to an acceptable
level.
The price paid for error control coding is that (for a fixed message bit rate and BER) a larger bandwidth is
required (to convey the inserted redundant bits in addition to the message bits). Since, when error control coding
with code rate r is employed only a fraction r of transmitted bits are message bits, it follows that the bandwidth
efficiency 𝜂 of an error control coded M-ary system will reduce by factor r to
⎧ r log2 M , APSK
⎪ 𝛼+1
𝜂=⎨ (6.75)
⎪ r log2 M , FSK
⎩ 𝛼 + (M + 1)∕2
Nevertheless, error control coding makes it possible to achieve a more efficient trade-off between bandwidth
and signal power than is possible through switching modulation scheme alone. This subtle advantage will become
clearer in the following worked example.

Worked Example 6.9 Bandwidth and Signal Power Trade-off

A digital transmission system operates at a message bit rate Rb = 2048 kb/s in AWGN using a raised cosine filter
of roll-off factor 𝛼 = 0.1. Reliable communication is set at BER = 10−7 .
(a) Determine the transmission bandwidth Bocc and Eb /N o (at demodulator input) required for transmission
using (i) QPSK, (ii) 8PSK, (iii) 16PSK, (iv) 16APSK, (v) 16FSK, and (vi) 1024FSK with no error control
coding.
(b) If an error control codec is introduced having code rate r = 1/2 and capable of correcting 5 random bit errors
per 100 bits in 16APSK modulation, determine the values of Bocc and Eb /N o now required.
(c) Comment on your results.

(a) The bandwidth efficiency of each modulation scheme is given by Eq. (6.75) (with r = 1, 𝛼 = 0.1) from
which transmission bandwidth Bocc = Rb /𝜂. For example, for QPSK, M = 4, 𝜂 = (log2 4)/1.1 = 2/1.1
and Bocc = 2048/𝜂 = 2048 × 1.1/2 = 1126.4 kHz. And for 16FSK, 𝜂 = (log2 16)/(0.1 + 17/2) = 4/8.6 and
Telegram: @Persian_IEEEComSoc

Bocc = 2048 × 8.6/4 = 4403.2 kHz.

The required value of Eb /N o is read from Figure 6.20 for the BER curve of each modulation scheme. For
example, at BER = 10−7 , QPSK requires Eb /N o = 11.3 dB and 16FSK requires Eb /N o = 8.9 dB.
(b) Bandwidth efficiency 𝜂 of 16APSK modulation is given by Eq. (6.75) (with code rate r = 1/2, 𝛼 = 0.1, and M = 16),
so 𝜂 = 2/1.1 and Bocc = Rb /𝜂 = 1126.4 kHz. Since the error control decoder will correct 5 random bit errors
per 100 bits, reliable communication can be achieved if the 16APSK demodulator at the receiver achieves
BER = 5/100 = 0.05 at its output which is delivered to the decoder. The value of Eb /N o at the demodulator
input needed to achieve this BER is read from the BER curve for 16APSK in Figure 6.20 as Eb /N o = 5.3 dB.
(c) The results of Bocc and Eb /N o for each of the above schemes are listed in Table 6.3, where we have introduced
an extra column for the parameter C/N o , which is the ratio between received signal power and noise power
 6.6 Summary 469

Table 6.3 Worked Example 6.9.

Scheme Bocc (kHz) E b /N o (dB) C/N o (dBHz)

QPSK 1 126.4 11.3 74.4

8PSK 750.933 14.8 77.9
16PSK 563.2 19.3 82.4
16FSK 4 403.2 8.9 72
1024FSK 104 980.48 5.9 69
16APSK 563.2 15.7 78.8
Coded 16APSK, r = 1/2 1 126.4 5.3 71.4

per unit bandwidth, obtained as

C P ER E |
= r = b b = b || + Rb ||dB⋅Hz (6.76)
No No No No |dB
Note that, in this equation, Rb is the transmitted bit rate. So, when error control code rate r is introduced, Rb
becomes the message bit rate multiplied by 1/r. For example, the uncoded 16APSK in Table 6.3 has
C∕No = 15.7 + 10 log(2048 × 103 ) = 78.8 dBHz
whereas the rate- 1/2-coded 16APSK has
C∕No = 5.3 + 10 log(2 × 2048 × 103 ) = 71.4 dBHz
We see from these results that there are various combinations of bandwidth and signal power that deliver
reliable communication at a bit rate of 2048 kb/s. Before highlighting two of these combinations, an impor-
tant clarification is necessary to avoid a common pitfall: we cannot compare the signal powers of coded and
uncoded transmission schemes simply by subtracting their respective Eb /N o dB values. This is because Eb is
the energy per transmitted bit and the number of bits in each scheme will be different. The C/N o parameter,
however, does permit direct comparison of signal power since N o is the same for both schemes if they are
implemented in the same transmission medium and receiver. C/N o therefore correctly represents the relative
level of required signal power in each of the schemes of Table 6.3.
We see that, without error control coding, we may save 4.4 dB of power (= 78.8 − 74.4) by switching modulation
scheme from 16APSK to QPSK. The price paid for this power saving is that the required bandwidth doubles
(from 563.2 to 1126.4 kHz). However, if we introduce error control coding (having the error-correction capabil-
ity stipulated in the problem statement), we save 78.8 − 71.4 = 7.4 dB of power while doubling the bandwidth
Telegram: @Persian_IEEEComSoc

requirement when switching from uncoded 16APSK to coded 16APSK of code rate r = 1/2. Notice how the
saving in power is 3 dB more in this second scenario than in the first although both involve a doubling of
bandwidth. The specific levels of benefit will differ from system to system but, in general, error control cod-
ing facilitates a more efficient and economical trading of bandwidth for signal power and vice versa than is
possible through switching modulation schemes alone.

6.6 Summary
Our study of noise in communication systems is now complete. You have acquired a good grounding in the quan-
tification of random noise and assessment of their impact on digital and analogue communication systems. You
 470 6 Noise in Communication Systems

are also reasonably well acquainted with the design parameters that affect SNR in analogue systems and BER in
digital systems.
Some modulation techniques provide more options for SNR improvement beyond the obvious solutions of
increasing received signal power or reducing system noise power. For example, in FM transmission, bandwidth
can be increased to realise a quadratic increase in SNR. However, a bandwidth/SNR trade-off is nonexistent in
amplitude modulation and all its variants, including SSB and DSB. Digital transmission, with the possibility of
multilevel modulation and error control coding, allows a greater flexibility than analogue FM in the exchange
between transmission bandwidth, signal power, BER, and transmission capacity or bit rate.
We are now well equipped to embark on a comprehensive study of the signal processing tasks of communications
systems, starting with amplitude modulation in the next chapter.

References

1 Lucas-Vegas, M.J. and Riera, J.M. (2016). Estimation of the atmospheric mean radiating temperature throughout
the world in a nonscattering atmosphere. IEEE Geoscience and Remote Sensing Letters 13 (2): 167–171. https://
doi.org/10.1109/LGRS.2015.2504409.
2 ETSI TR 102 376-2 (2015). Digital Video Broadcasting (DVB): Implementation guidelines for the second gener-
ation system for Broadcasting, Interactive Services, News Gathering and other broadband satellite applications:
Part 2: S2 Extensions (DVB-S2X): V1.1.1.

Questions

6.1 An amplifier has gain G = 20 dB and noise temperature T e = 250 K.

(a) If the input power is 3 mW, what is the output power? Express your answer in dBm and in mW.
(b) What is the available noise power generated by the amplifier in a bandwidth of 4 kHz referred to its
input?
(c) What is the available noise power generated by the amplifier in a bandwidth of 4 kHz measured at its
output?
(d) What is the noise figure of the amplifier?
(e) If the input is a voice signal of signal-to-noise ratio (SNR) 31.7 dB, what is the output SNR? Does this
satisfy the standard quality threshold of 30 dB for the SNR of voice signals?
(f) If input SNR ratio is 31.7 dB, and input signal power is 3 mW in a bandwidth of 4 kHz, what is the input
noise power within the same bandwidth?
Telegram: @Persian_IEEEComSoc

6.2 You are given three amplifiers, identified A, B, and C, to connect in cascade. Amplifier A has gain 10 dB
and noise figure 5 dB, amplifier B has gain 30 dB and noise figure 1.5 dB, and amplifier C has gain 20 dB and
noise figure 1.5 dB.
(a) What is the overall gain of the cascade connection?
(b) Arrange the three amplifiers in the order that minimises the noise figure of the overall system.
(c) What is the overall noise figure of the system as connected in (b)?
(d) What is the available noise power generated by the system in (b) in a bandwidth of 100 kHz measured
at its output?
 Questions 471

6.3 The arrival of photons at a receiver in an optical fibre communication system is governed by Poisson statis-
tics. This means that if the average number of photons received during a binary 1 interval is 𝜇 then the
probability p(k) of detecting k photons during any binary 1 interval is given by
𝜇k exp(−𝜇)
p(k) = k!
Clearly, in an OOK system, an error occurs if no photon is received during a binary 1 interval.
(a) Assuming that binary 1s and 0s are equally likely, derive Eq. (6.12) for the minimum average received
power Pr to ensure a specified BER. (Note: energy of one photon = hf , where h (= 6.626 × 10−34 Js) is
Planck’s constant, and f is the light signal frequency (Hz).)
(b) Determine Pr required for a bit error rate of 10−9 in a 1310 nm system operating at 296 MBd

6.4 An earth station antenna receives a satellite transmission at 40 GHz on a path elevation of 15∘ located in a
temperate climate. Noise from the Earth’s surface enters the antenna through its first sidelobe at a relative
level of −20 dB. The antenna is connected via a waveguide of loss 2 dB to an indoor receiver unit of noise
temperature Te = 120 K.
(a) Determine the antenna noise temperature T a during a rain event when the total attenuation is 15 dB.
(b) What is the total degradation in carrier-to-noise ratio (C/N) during the above rain event?
(c) What is the noise power delivered by the antenna to a matched waveguide feed in a bandwidth of 6 MHz?

6.5 The thermal noise voltage vn observed in a certain resistor has a standard deviation 𝜎 = 200 mV.
(a) What is the mean of vn ?
(b) What is the probability that vn is equal to its mean?
(c) Can vn have a sample value that exceeds 10 V? Explain.
(d) Determine the probability that vn ≥ −300 mV.
(e) If many samples of the noise voltage are taken, what percentage will lie within a range ±3𝜎 of the mean?

6.6 .(a) Calculate the noise figure and equivalent noise temperature of a cascade connection of an amplifier and
a mixer. The amplifier has a noise figure of 5 dB and a gain of 20 dB, whereas the mixer stage has a noise
figure of 12 dB and a gain of −2 dB.
(b) Calculate the SNR at the output of the mixer in (a) if the SNR at the input of the amplifier is 30 dB.

6.7 Determine the noise equivalent bandwidth of the following filters and compare with their 3 dB bandwidths.
(a) A bandpass filter (BPF) whose gain response is shown in Figure Q6.7a.
(b) The RLC BPF shown in Figure Q6.7b.
Telegram: @Persian_IEEEComSoc

Figure Q6.7a Question 6.7(a). |H(f)|

K

0 f
fc – fs fc f c + fs
 472 6 Noise in Communication Systems

L C Figure Q6.7b Question 6.7(b).

Input R Output

6.8 A satellite receiving earth station has the following parameters:

(i) Antenna noise temperature (in clear weather) = 70 K.
(ii) Antenna gain = 40 dB.
(iii) Total ohmic losses from antenna feed to LNA input = 3 dB.
(iv) LNA gain = 30 dB, and noise figure = 1.0 dB.
(v) Mixer gain = −3.5 dB, and noise figure = 8.0 dB.
(vi) Intermediate frequency (IF) amplifier gain = 25 dB, and noise figure = 4.0 dB.
The satellite transmits at a frequency of 20 GHz with an EIRP of 42 dBW, and the slant path length from the
earth station to the satellite is 40 000 km.
(a) What is the (figure of merit) G/T of the earth station?
(b) What is Eb /N o at a bit rate of 1 Mbps?
(c) Determine the maximum bit rate Rbmax that can be achieved in clear weather using QPSK if the BER
must not exceed 10−4 .
(d) If there is a total atmospheric attenuation of 5 dB during rain and the bit rate is maintained at the value
obtained in (c), determine the BER. Comment on this result and specify the maximum bit rate that can
be supported during this rain event without exceeding the BER limit of 10−4 , if the other transmission
parameters, including satellite EIRP and earth station system noise and gains, are unchanged.
(e) Repeat (c) for a BER of 10−7 .

6.9 (. a) A QPSK signal generated by a transmitter with raised cosine filters having roll-off factor 𝛼 = 0.2 occupies
a 32 MHz bandwidth of a satellite transponder. Determine the symbol rate (in MBd) and bit rate of the
QPSK signal.
(b) If the QPSK demodulator in the above system has an implementation margin of 1 dB, and the overall
C/N ratio in the earth station receiver in clear air conditions is 13 dB, determine the probability of bit
error (i.e. BER) of the baseband digital signal in clear air conditions. How often does a bit error occur?
(Note: the implementation margin of a modem is the amount in dB by which the Eb /N o of its input signal
must exceed the theoretical value given by Figure 6.20 for a required BER to be achieved.)
Telegram: @Persian_IEEEComSoc
 473

Amplitude Modulation

If at first you don’t succeed, check your method.

In this Chapter

✓ Time domain discussion of amplitude modulation (AM) signals.

✓ Spectrum and power of AM signals.
✓ AM modulators and demodulators, including the superheterodyne receiver.
✓ Waveform description, spectrum, modulators, demodulators, and applications for double sideband sup-
pressed carrier AM (DSB). You will particularly benefit from the presentation on balanced modulators.
✓ Single sideband suppressed carrier amplitude modulation (SSB): spectrum, modulators, demodulators, and
applications.
✓ Independent sideband suppressed carrier amplitude modulation (ISB): spectrum, modulators, demodula-
tors, and applications.
✓ Vestigial sideband suppressed carrier amplitude modulation (VSB): spectrum, modulators, demodulators,
and applications.

7.1 Introduction

The roles of modulation in telecommunication are discussed in detail in Chapter 1. Modulation allows us to:
Telegram: @Persian_IEEEComSoc

● Accommodate multiple users on the same communication link.

● Achieve efficient radiation in radio systems.
● Exploit the larger bandwidths available at higher radio frequencies.
● Use a frequency band best suited to the transmission medium.
In Chapter 1 we also define modulation as the process of imposing the variations (or information) in a
lower-frequency electrical signal (called modulating or baseband or message signal) onto a higher-frequency
sinusoidal signal (called the carrier), and note the three parameters of the sinusoidal carrier signal that may be
varied, namely:
● The carrier amplitude, leading to what is termed amplitude modulation (AM).

Communication Engineering Principles, Second Edition. Ifiok Otung.

© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 474 7 Amplitude Modulation

● The carrier frequency, resulting in frequency modulation.

● The carrier phase, giving rise to phase modulation.
This chapter is devoted to AM, whereas frequency and phase modulations are treated in the next chapter. We
will apply the skills acquired in previous chapters to build on that introductory discussion of modulation. First,
there is an introduction of AM, and definition of common terms such as modulation depth and modulation sensi-
tivity based on a simple staircase message signal. Next, we obtain the spectrum of an AM signal since signal spectra
must be known in every communication situation to determine the required transmission bandwidth and a suit-
able transmission medium. To do this, a sinusoidal message signal is used, and it is shown that the AM process
generates three frequency components, namely the lower side frequency (LSF), the carrier, and the upper side fre-
quency (USF). You will then see that for an information signal consisting of a band of sinusoids, a lower sideband
(LSB) replaces the LSF, and an upper sideband (USB) replaces the USF.
We discuss the power content of an AM signal and highlight AM’s demerits in power and spectral usage but
emphasise its great advantage in transmitter and receiver circuit simplicity. Alternative AM schemes that save
power and/or bandwidth at the cost of increased circuit complexity are then introduced, including double side-
band suppressed carrier amplitude modulation (DSB), single sideband amplitude modulation (SSB), independent
sideband amplitude modulation (ISB), vestigial sideband amplitude modulation (VSB), and quadrature amplitude
modulation (QAM). The modulator and demodulator circuits that implement the above AM schemes are discussed
in block diagram format.
The chapter includes a discussion of the applications of AM and its variants, and the factors that dictate these
applications.

7.2 AM Signals: Time Domain Description

In AM, the amplitude of a sinusoidal carrier signal is varied in proportion to the instantaneous value of a message
signal, which is the modulating signal. The change in carrier amplitude per volt (V) of modulating signal is called
the modulation sensitivity k, expressed in V/V. The modulation sensitivity is usually 1 V/V, and this value should
be assumed except where otherwise indicated. The information contained in the variations of the message signal
is perfectly preserved in the variations of the carrier amplitude, and this information is recovered at the receiver
by tracking the carrier amplitude. The main benefit is that, whereas the frequency content of the message sig-
nal is relatively fixed, there is a great deal of freedom in the choice of the frequency of a carrier onto which the
information is impressed, the choice being guided by the four roles of modulation stated earlier. In this way, a
variety of message signals can be transmitted using a variety of transmission media and communication systems.
For example, voice signals with frequency content in the range ∼50–7000 Hz can be transmitted by various means,
such as
Telegram: @Persian_IEEEComSoc

● Via satellite by impressing the voice information onto a carrier of frequency in the superhigh frequency (SHF)
band (3–30 GHz).
● Via optical fibre by modulating an optical carrier (frequency ∼194 THz) with a digitised version of the voice
signal.
● Via AM radio by modulating a carrier of frequency between 540 kHz and 1.64 MHz.

7.2.1 AM Waveform
Consider the simple staircase message signal vm (t) shown in Figure 7.1a, which is used to amplitude modulate the
sinusoidal carrier vc (t) shown in Figure 7.1b. Observe that vc (t) has a constant amplitude V c = 3 V and completes
three cycles in every 1 ms interval. That is, the unmodulated carrier oscillates between the fixed levels ±V c at a
 7.2 AM Signals: Time Domain Description 475

2
1
ʋm(t), volt

t, ms
(a) 0
–1
–2
0 1 2 3 4 5 6
Vpp = 2Vc
Vc = 3
2
ʋc(t), volt

1 t, ms
(b) 0
–1
–2
–Vc = –3
0 1 2 3 4 5 6
5 Vppmax = D – C = 10 V D=5
4 Vppmin = B – A = 2 V
3
2
ʋam(t), volt

1 B=1
(c) 0 t, ms
–1 A = –1
–2
–3
–4
–5 C = –5
0 1 2 3 4 5 6

Figure 7.1 (a) Message signal v m (t); (b) carrier signal v c (t); (c) AM signal v am (t) with modulation sensitivity k = 1 volt/volt.

frequency f c = 3 kHz. AM of vc (t) by the message signal vm (t) is achieved by changing the amplitude of vc (t) from
the constant value V c to the value
Vam = Vc + kvm (t) (7.1)
The result is the AM signal vam (t) shown in Figure 7.1c for modulation sensitivity k = 1 V/V. Note that the AM
signal vam (t) maintains the same rate of three oscillations per ms as the carrier vc (t), but its amplitude V am is now
a function of time depending on the instantaneous value of the message signal vm (t), and has the following values
⎧
⎪3 V, 0 ≤ t < 1 ms
⎪
⎪5 V, 1 ≤ t < 2 ms
⎪
⎪4 V, 2 ≤ t < 3 ms
Vam (t) = Vc + kvm (t) = 3 + vm (t) = ⎨
⎪2 V, 3 ≤ t < 4 ms
⎪
Telegram: @Persian_IEEEComSoc

⎪1 V, 4 ≤ t < 5 ms
⎪
⎪3 V, 5 ≤ t ≤ 6 ms
⎩
V am (t) defines the envelope of the AM waveform, which is indicated in dotted lines in Figure 7.1c. The dotted
lines are not part of the AM waveform – they would not appear in an oscilloscope display of the waveform – but
are very useful as guides for sketching the AM waveform.

7.2.2 Sketching AM Waveforms

The following steps, shown in Figure 7.2, were involved in drawing the AM waveform shown in Figure 7.1c, and
can be followed in sketching any AM waveform.
 476 7 Amplitude Modulation

Vc

(a) t

–Vc

kʋm(t)
Vc

(b) t

–Vc
–kʋm(t)

ʋam(t)
5 kʋm(t)
4
Vc = 3
2
1
(c) 0
–1
–2
–Vc = –3
–4 –kʋm(t)
–5 t, ms
1 2 3 4 5 6

Figure 7.2 Steps in sketching AM waveform.

● Draw two dotted horizontal lines at ±V c , where V c is the carrier amplitude. In the absence of modulation, the
carrier oscillates between these two levels, implying constant amplitude V c .
● Sketch the waveform kvm (t) at the level + V c , and the (inverted) waveform −kvm (t) at the level −V c . The envelope
of the AM waveform is now defined.
● Insert the sinusoidal carrier centred within the envelope defined above, ensuring that its positive and nega-
tive peaks are always stretched or reduced as necessary to touch the envelope, and that its rate of oscillation
is unchanged. Significant points on the voltage and time axis should be clearly labelled as shown. If the car-
rier frequency is so high that it is impractical to sketch individual cycles then it is acceptable to represent
the AM waveform as a set of closely packed vertical lines or shading that completely fills the area within the
Telegram: @Persian_IEEEComSoc

envelope.

7.2.3 Modulation Factor

The peak value or amplitude of a sinusoidal signal does not change, and hence its peak-to-peak value V pp is con-
stant. In Figure 7.1b, we see that for a sinusoidal signal of amplitude V c
Vpp = 2Vc
The peak value of an AM signal is not constant but varies according to the modulating signal from a minimum
of V ppmin to a maximum of V ppmax . A parameter called modulation factor (denoted m) provides a measure of the
 7.2 AM Signals: Time Domain Description 477

extent of carrier amplitude variation caused by the modulating signal. It is defined by the expression
Maximum peak to peak − Minimum peak to peak
m=
Maximum peak to peak + Minimum peak to peak
Vpp max − Vpp min
= (7.2)
Vpp max + Vpp min
Equation (7.2) applies to all AM signals regardless of modulation sensitivity k and type of modulating signal vm (t),
and is particularly convenient to use when the AM waveform has been sketched on paper or displayed on an
oscilloscope. For the AM waveform shown in Figure 7.1c, V ppmax = 10 V and V ppmin = 2 V, and it follows from
Eq. (7.2) that
10 − 2
m= = 0.667
10 + 2
Note carefully that here and in all cases
Vpp min = B − A (7.3)
where B is the lowest level of the top envelope, and A is the highest level of the bottom envelope. In this case B = 1,
A = −1, and V ppmin = 1 − (−1) = 2 V. The maximum peak-to-peak value is given by
Vpp max = D − C (7.4)
where D is the highest level of the top envelope and C is the lowest level of the bottom envelope. In this case,
D = 5 V and C = −5 V so that V ppmax = 5 − (−5) = 10 V.
When expressed in percentage, m is referred to as the modulation index, or percentage modulation, or depth of
modulation. Therefore, in the above example, the modulation index is m = 66.7%. When the modulation factor has
a value between 0 and less than unity (i.e. 0 < m < 1), as in this case, the AM signal is said to be undermodulated.
Figure 7.3a shows an AM waveform obtained using the same message signal and carrier as in Figure 7.1, but
with the modulation sensitivity increased to k = 1.5. The carrier amplitude has been varied until the minimum
peak-to-peak value V ppmin of the AM signal is zero, since B = A = 0. In this case, it follows from Eq. (7.2) that
m = 1. The AM signal is said to be 100% modulated. We show later that the most power-efficient AM transmission
is obtained when operation is as close as possible to 100% modulation.
Let us consider the result of varying the carrier amplitude beyond the 100% modulation limit. This may be done
by sufficiently increasing the modulation sensitivity or by using a sufficiently large modulating signal. An example
of this is shown in Figure 7.3b for the same staircase modulating signal and carrier as before, but with k = 2.
Observe that the lowest level of the top envelope (dotted line) is B = −1; and the highest level of the bottom envelope
(dashed line) is A = 1. Therefore V ppmin (= B − A) is negative. It then follows from Eq. (7.2) that modulation factor
m > 1. In this case
14 − (−2)
Telegram: @Persian_IEEEComSoc

m= = 1.33
14 + (−2)
The AM signal is said to be overmodulated. There is then a portion of the AM waveform at which the amplitude
V am = Vc + kvm (t) is negative. This is equivalent to a phase shift of 180∘ , or phase reversal. There is a carrier phase
reversal at every point where the top envelope crosses the x axis. The top and bottom envelopes cross each other
at these points. You will also observe that the envelope of the AM signal is no longer a replica of the message
signal. This envelope distortion makes it impossible to recover the original message signal from the AM wave-
form envelope. Overmodulation must be avoided by ensuring that the message signal vm (t) satisfies the following
condition
kV m ≤ Vc (7.5)
 478 7 Amplitude Modulation

Vppmax = D – C = 12 V Vppmin = B – A = 0
D=6

4
(a)
2
t, ms
B=A=0

–2

–4

C = –6
0 1 2 3 4 5 6
Vppmax = D – C = 14 V Vppmin = B – A = –2 V
D=7
6
(b) 5 4
3
2
A=1 t, ms
0
B = –1
–2
–3
–4 Phase reversals
–5
–6
C = –7
0 1 2 3 4 5 6

Figure 7.3 AM signal with (a) 100% modulation; (b) Over-modulation.

where V m is a positive voltage equal to the maximum excursion of the message signal below 0 V, V c is the carrier
amplitude, and k is the modulation sensitivity (usually k = 1).

Worked Example 7.1 An audio signal vm (t) = 30 sin(5000𝜋t) V modulates the amplitude of a carrier
vc (t) = 65sin(50000𝜋t) V.
(a) Sketch the AM waveform.
(b) What is the modulation factor?
(c) Determine the modulation sensitivity that would give a modulation index of 80%.
(d) If the message signal amplitude is changed to a new value that is 6 dB below the carrier amplitude, deter-
mine the resulting modulation factor.

(a) We will sketch the AM waveform over two cycles of the audio signal vm (t). The audio signal frequency f m and
the carrier frequency f c must be known in order to determine how many carrier cycles are completed in one
cycle of vm (t)
Telegram: @Persian_IEEEComSoc

vm (t) = 30 sin(5000𝜋t) ≡ Vm sin(2𝜋fm t), ⇒ fm = 2.5 kHz

vc (t) = 65 sin(50000𝜋t) ≡ Vc sin(2𝜋fc t), ⇒ fc = 25 kHz
Thus, vm (t) has the period T m = 1/f m = 0.4 ms and carrier frequency f c = 10f m , which means that the carrier
completes 10 cycles in the time 0.4 ms that it takes the audio signal to complete one cycle. Following the steps
outlined in Section 7.2.2 for sketching AM waveforms, we sketch two cycles of vm (t) at level V c = 65 V, and two
cycles of −vm (t) at level −65 V. This defines the envelope of the AM waveform. We then sketch in the carrier,
stretching its amplitude to always touch the envelope, and ensuring that there are exactly 10 cycles of this
carrier in one cycle of the envelope. This completes the required sketch of the AM waveform vam (t), which is
shown properly labelled in Figure 7.4. Note that we assumed modulation sensitivity k = 1. This is always the
case, except where specifically otherwise indicated, as in (c) below.
 7.2 AM Signals: Time Domain Description 479

95

65

35

–35
Vppmin
–65

Vppmax
–95
0 0.4 0.8 → t (ms)

Figure 7.4 AM waveform in Worked Example 7.1.

(a) From Figure 7.4 and Eq. (7.2)

Vpp max = 2 × 95 = 190 V
Vpp min = 2 × 35 = 70 V
190 − 70
m= = 0.462
190 + 70
(b) The expression for V ppmax and V ppmin taking account of a nonunity modulation sensitivity k is
Vpp max = 2(Vc + kV m ); Vpp min = 2(Vc − kV m )
Using Eq. (7.2)
2(Vc + kV m ) − 2(Vc − kV m ) kV m
m= =
2(Vc + kV m ) + 2(Vc − kV m ) Vc
Therefore
mV c 0.8 × 65
k= = = 1.73
Vm 30
(c) Assuming k = 1, we have from (c) above
Telegram: @Persian_IEEEComSoc

V
m= m
Vc
Given that V m is 6 dB below V c , we have
( )
Vm
20 log10 = −6
Vc
or
Vm
= 10(−6∕20) = 0.5
Vc
Therefore, m = 0.5.
Observe that m is simply the dB value converted to a ratio.
 480 7 Amplitude Modulation

7.3 Spectrum and Power of Amplitude Modulated Signals

By virtue of the Fourier theorem, any message signal vm (t) can be realised as the discrete or continuous sum of sinu-
soidal signals. The spectrum of an AM signal can therefore be obtained by considering a sinusoidal message signal
and extending the result to information signals, which in general consist of a band of frequencies (or sinusoids).

7.3.1 Sinusoidal Modulating Signal

So, consider a carrier signal of amplitude V c and frequency f c
vc (t) = Vc cos(2𝜋fc t) (7.6)
and a sinusoidal modulating signal of amplitude V m and frequency f m
vm (t) = Vm cos(2𝜋fm t) (7.7)
Usually
Vm ≤ Vc ; fm ≪ fc (7.8)
An expression for the AM signal vam (t) is obtained by replacing the constant carrier amplitude V c in Eq. (7.6)
with the expression for the modulated amplitude V am given in Eq. (7.1)
vam (t) = [Vc + kvm (t)] cos(2𝜋fc t)
= [Vc + Vm cos(2𝜋fm t)] cos(2𝜋fc t), for k = 1 (7.9)
where we have set k = 1 (as is usually the case), and substituted the expression for vm (t) from Eq. (7.7). Following
the steps outlined in Section 7.2, the AM waveform given by Eq. (7.9) is sketched in Figure 7.5. It is obvious from
Figure 7.5 that the maximum and minimum peak-to-peak amplitudes of vam (t) are given, respectively, by
Vpp max = (Vc + Vm ) − (−Vc − Vm ) = 2(Vc + Vm )
Vpp min = (Vc − Vm ) − (−Vc + Vm ) = 2(Vc − Vm )

Vc + Vm

Vc

Vc – V m
Telegram: @Persian_IEEEComSoc

Vppmax
Vppmin

t
0

–Vc + Vm

–Vc

–Vc – Vm

Figure 7.5 AM waveform for a sinusoidal message signal of frequency f m and amplitude V m . The plot is for carrier
frequency f c = 100f m , and carrier amplitude V c = 3V m .
 7.3 Spectrum and Power of Amplitude Modulated Signals 481

From Eq. (7.2), we obtain a simple expression for the modulation factor m in the special case of a sinusoidal
message signal and the usual case of unity modulation sensitivity
2(Vc + Vm ) − 2(Vc − Vm )
m=
2(Vc + Vm ) + 2(Vc − Vm )
V
= m (7.10)
Vc
The modulation factor is given by the ratio between the amplitude of a sinusoidal message signal and the ampli-
tude of the carrier signal. The ideal modulation factor m = 1 is obtained when V m = V c , and overmodulation occurs
whenever V m > V c .
Returning to Eq. (7.9) and expanding it using the trigonometric identity
1 1
cos A cos B = cos(A − B) + cos(A + B)
2 2
we obtain
vam (t) = Vc cos(2𝜋fc t)
V
+ m cos[2𝜋(fc − fm )t]
2
Vm
+ cos[2𝜋(fc + fm )t] (7.11)
2
The AM signal therefore contains three frequency components, namely
● The carrier frequency f c with amplitude V c .
● A frequency component f c −f m with amplitude 12 Vm . This frequency component is called the LSF, since it lies
below the carrier frequency.
● A frequency component f c + f m with amplitude 1/2V m . This frequency component is called the USF, since it lies
above the carrier frequency.
Note that the AM signal does not contain any component at the message signal frequency f m . Figure 7.6 shows
the single-sided amplitude spectra of vm (t), vc (t), and vam (t). You will observe by studying this figure along with
Eq. (7.11) that AM translates the message frequency f m of amplitude V m to two side frequencies f c − f m and f c + f m ,
each of amplitude 1/2V m . Let us denote this process as follows
AM
fm |Vm −−−−→ (fc − fm )| 1 Vm + (fc + fm )| 1 Vm (7.12)
fc 2 2

Equation (7.12) states that a frequency component f m of amplitude V m is translated by an AM process (that uses
a carrier of frequency f c ) to two new frequency components at f c − f m and f c + f m each of amplitude 1/2V m . From
Eq. (7.10), this amplitude can be expressed in terms of the carrier amplitude V c and the modulation factor m
Telegram: @Persian_IEEEComSoc

Vm mV c
= (7.13)
2 2
Treating the sinusoidal message signal as a lowpass (i.e. baseband) signal, and the AM signal as a bandpass
signal, it follows (see Section 4.7.3) that they have the following bandwidths
Message signal bandwidth = fm
AM signal bandwidth = (fc + fm ) − (fc − fm ) = 2fm
Thus, the AM bandwidth is twice the message bandwidth. It should be noted that erroneously treating the
sinusoidal message signal vm (t) as a bandpass signal would give it zero bandwidth, since the width of significant
frequency components is the difference between maximum and minimum frequency, which in this case are both
equal to f m , so that B = f m − f m = 0. This would lead to incorrect results.
 482 7 Amplitude Modulation

Amn

Vm
(a)

f
fm

Acn
Vc

(b)

f
fc

Aamn Carrier
Vc

(c) LSF USF

1
Vm
2
f
fc – fm fc fc + fm

Figure 7.6 Single-sided amplitude spectrum of (a) sinusoidal modulating signal; (b) carrier signal; and (c) AM signal.

Worked Example 7.2 For the AM waveform vam (t) obtained in Worked Example 7.1

(a) Determine the frequency components present in the AM waveform and the amplitude of each component.
(b) Sketch the double-sided amplitude spectrum of vam (t).

(a) The sinusoidal message signal vm (t) of amplitude 30 V and frequency f m = 2.5 kHz is translated by the AM
process with carrier frequency f c = 25 kHz in the manner given by Eq. (7.12)
AM
2.5 kHz|30 V −−−−→ (25 − 2.5 kHz)|15 V + (25 + 2.5 kHz)|15 V
25 kHz

The carrier of amplitude 65 V is also a component of vam (t). So, there are three components as follows:
Telegram: @Persian_IEEEComSoc

(i) LSF of frequency 22.5 kHz and amplitude 15 V.

(ii) Carrier of frequency 25 kHz and amplitude 65 V.
(iii) USF of frequency 27.5 kHz and amplitude 15 V.
(b) The double-sided amplitude spectrum showing each of the above components as a positive and negative fre-
quency pair is sketched in Figure 7.7.

7.3.2 Arbitrary Message Signal

In practice, a message signal consists of a band of frequencies from f 1 to f m , where f 1 may be zero and f m is larger
than f 1 and finite. We will represent this band of frequencies symbolically by the trapezoidal spectrum shown in
 7.3 Spectrum and Power of Amplitude Modulated Signals 483

An, volts
32.5

7.5

f, kHz
–27.5 –25 –22.5 22.5 25 27.5

Figure 7.7 Double-sided spectrum of AM waveform in worked example 7.2.

Figure 7.8 Symbolic representation of the amplitude |Vm(f)|

spectrum of a message signal.
V1

Vm

f
f1 fm

|Vm(f)| |Vam(f)|
Vc
V1

Vm AM 0.5V1
fc 0.5Vm
f f
f1 fm fc
fc + f1

fc + fm
fc – f1
fc – fm

Figure 7.9 Translation of the maximum and minimum frequency components of a message signal in AM.

Figure 7.8. If this message signal is used to modulate a carrier of frequency f c , then Eq. (7.12) leads to
AM
f1 |V1 −−−−→ (fc − f1 )| 1 V1 + (fc + f1 )| 1 V1
fc 2 2

AM
fm |Vm −−−−→ (fc − fm )| 1 Vm + (fc + fm )| 1 Vm
fc 2 2

The above frequency translation is shown in Figure 7.9. Each of the remaining frequency components of the
Telegram: @Persian_IEEEComSoc

message signal is translated to a USF in the range f c + f 1 to f c + f m , and an LSF lying between f c − f m and
f c − f 1 . The result is that the message signal spectrum is translated to a USB and an LSB, as shown in Figure 7.10.
The shape of the message spectrum is preserved in the sidebands, although the LSB is mirror-inverted, whereas
the USB is erect. However, the amplitude of each sideband is reduced by a factor of 2 compared to the message
spectrum. The condition f c > f m ensures that the LSB lies entirely along the positive-frequency axis and does not
overlap with the negative-frequency LSB in a double-sided spectrum. We will see that, for the AM waveform to
have an envelope that can be easily detected at the receiver, the carrier frequency must be much larger than the
highest frequency component f m in the message signal. That is
f c ≫ fm (7.14)
 484 7 Amplitude Modulation

|Vm(f)| |Vam(f)|
Vc
V1
AM

Carrier
Vm 0.5V1 LSB USB
fc
0.5Vm
f f
f1 fm fc

fc + f1

fc + fm
fc – f 1
fc – f m
Figure 7.10 Production of lower sideband (LSB) and upper sideband (USB) in AM.

A message signal (also called the baseband signal) is always regarded as a lowpass signal, since it usually contains
positive frequency components near zero. Its bandwidth is therefore equal to the highest significant frequency
component f m . Thus, the message signal whose spectrum is represented by Figure 7.8 has bandwidth f m , rather
than f m − f 1 , which would be the case if the signal were treated as bandpass. The AM signal, on the other hand,
results from the frequency translation of a baseband signal, and is therefore a bandpass signal with a bandwidth
equal to the width of the band of significant positive frequencies. It follows that, for an arbitrary message signal
of bandwidth f m , the AM bandwidth (see Figure 7.10) is given by

(fc + fm ) − (fc − fm ) = 2fm

In general

AM bandwidth = 2 × Message bandwidth (7.15)

Worked Example 7.3

(c) A 1 MHz carrier is amplitude modulated by a music signal that contains frequency components from 20 Hz to
15 kHz. Determine
(d) The frequencies in the AM signal and sketch its double-sided amplitude spectrum.
(e) The transmission bandwidth of the AM signal.
(f) Carrier frequency f c = 1 MHz and the message frequency band is from f 1 = 20 Hz to f m = 15 kHz. The AM
signal therefore contains the following frequencies:
(i) LSB in the frequency range
fc − (f1 → fm ) = 1 MHz − (20 Hz → 15 kHz)
= 985 kHz → 999.98 kHz
Telegram: @Persian_IEEEComSoc

(ii) Carrier frequency at f c = 1 MHz = 1000 kHz

(iii) USB in the frequency range
fc + (f1 → fm ) = 1 MHz + (20 Hz → 15 kHz)
= 1000.02 kHz → 1015 kHz

The double-sided spectrum is shown in Figure 7.11. Note that this plot is a mixture of a discrete spectrum (for
the carrier) with the y axis in volts, and a continuous spectrum (for the sidebands) with the y axis in V/Hz.
(g) The transmission bandwidth BT is given by Eq. (7.15)

BT = 2fm = 2 × 15 kHz = 30 kHz

Note that BT is the width of the positive frequency band in Figure 7.11.
 7.3 Spectrum and Power of Amplitude Modulated Signals 485

1 Vam(f)
V
2 c

f, kHz
–1015 –1000 –985 985 1000 1015
998.98 1000.02
BT

Figure 7.11 AM spectrum in Worked Example 7.3.

7.3.3 Power
The distribution of power in AM between the carrier and sidebands is easier to determine using the spectrum of
a carrier modulated by a sinusoidal message signal. This spectrum is shown in Figure 7.6c, from which we obtain
the following (normalised) power distribution:
Power in carrier
Vc2
Pc = (7.16)
2
Power in USF
(Vm ∕2)2 V2 (mV c )2
PUSF = = m =
2 8 8
m2 Pc
= (7.17)
4
Power in LSF
m2 Pc
PLSF = PUSF = (7.18)
4
Power in side frequencies (SF)

m2 Pc
PSF = PLSF + PUSF = (7.19)
2
Telegram: @Persian_IEEEComSoc

Total power in AM waveform

m2 Pc
Pt = Pc + PSF = Pc +
( ) 2
m2
= Pc 1 + (7.20)
2

Equation (7.19) states that the total power PSF in the side frequencies is a fraction m2 ∕2 of the power in the carrier.
Thus, PSF increases with modulation factor. The maximum total side frequency power PSFmax is 50% of the carrier
power, and this is obtained when m = 1, i.e. at 100% modulation. We may use Eq. (7.20) to express Pc in terms of Pt
Pt 2
Pc = = P (7.21)
1 + m2 ∕2 2 + m2 t
 486 7 Amplitude Modulation

This allows us to determine PSF as a fraction of the total power in the AM waveform as follows
( )
2
PSF = Pt − Pc = Pt − Pt
2 + m2
( )
2
= Pt 1 −
2 + m2
m2
= P (7.22)
2 + m2 t
Therefore the maximum power in side frequencies, obtained at 100% modulation (m = 1), is
1 1
PSFmax = P = P (7.23)
2+1 t 3 t
This corresponds to a minimum carrier power
2
Pcmin = P (7.24)
3 t
Equation (7.24) may also be obtained by putting m = 1 in. Eq. (7.21). It shows that at least two-thirds of the trans-
mitted AM power is contained in the carrier, which carries no information. The sidebands, which contain all the
transmitted information, are only fed with at most one-third of the transmitted power. This is a serious demerit
of AM.
Equations (7.16)–(7.24) are derived assuming a sinusoidal modulating signal, but they are equally applicable
to all AM signals involving arbitrary message signals. The only change is that LSF, USF, and SF are replaced by
LSB, USB, and SB (sideband), respectively. The frequency domain representation of the arbitrary message and AM
signals is presented in Figure 7.8. The above equations are applicable to this general case provided we define the
modulation factor m in terms of the total sideband power PSB and carrier power Pc as follows, based on Eq. (7.19)
√
2PSB
m= (7.25)
Pc

Worked Example 7.4 An AM broadcast station operates at a modulation index of 95%.

(a) Determine what percentage of the total transmitted power is in the sidebands.
(b) If the transmitted power is 40 kW when the modulating signal is switched off, determine the total trans-
mitted power at 95% modulation.

(a) Using Eq. (7.22), the ratio of sideband power PSF to total power Pt is
PSF m2 0.952
= 2
= = 0.311
Pt 2+m 2 + 0.952
Therefore, 31.1% of the total power is in the sidebands.
Telegram: @Persian_IEEEComSoc

(b) The unmodulated carrier power Pc = 40 kW. Using Eq. (7.21), the total transmitted power Pt is given by
( ) ( )
m2 0.952
Pt = Pc 1 + = 40 1 +
2 2
= 40(1 + 0.45) = 58 kW

Worked Example 7.5 The carrier vc (t) = 100 sin(3 × 106 𝜋t) V is amplitude modulated by the signal
vm (t) = 60 sin(80 × 103 𝜋t) + 30 sin(100 × 103 𝜋t) V.

(a) Determine the total power in the AM signal.

(b) Determine the modulation index.
 7.3 Spectrum and Power of Amplitude Modulated Signals 487

An, volt
100

30
15
f, kHz
1450 1500 1540
1460 1550

Figure 7.12 AM spectrum in Worked Example 7.5.

(a) Carrier amplitude V c = 100 V, and carrier frequency f c = 1.5 MHz. The message signal contains two frequencies
f 1 = 40 kHz with amplitude 60 V, and f 2 = 50 kHz with amplitude 30 V. From Eq. (7.12), the AM signal has the
following frequency components, which are shown in the single-sided spectrum of Figure 7.12:
(i) f c −f 2 = 1450 kHz with amplitude 15 V.
(ii) f c −f 1 = 1460 kHz with amplitude 30 V.
(iii) f c = 1500 kHz with amplitude 100 V.
(iv) f c + f 1 = 1540 kHz with amplitude 30 V.
(v) f c + f 2 = 1550 kHz with amplitude 15 V.
The total power Pt in the AM wave is the sum of the powers of these components. Therefore
152 302 1002 302 152
Pt = + + + +
2 2 2 2 2
= 6125 W

(b) The modulation index can be determined from Eq. (7.25). The carrier power Pc , i.e. the power in the frequency
component f c , and the total sideband power PSF are given by
1002
Pc = = 5000 W
2
PSF = Pt − Pc = 6125 − 5000 W = 1125 W

Therefore modulation index is

√ √
2PSB 2250
× 100% = × 100%
Pc 5000
= 67.08%
Telegram: @Persian_IEEEComSoc

You may wish to verify, using the last result of Worked Example 7.5, that in the case of a message signal consisting
of multiple tones (i.e. sinusoids) of amplitudes V 1 , V 2 , V 3 , …, the modulation factor m can be obtained from the
following formula
√
( )2 ( )2 ( )2
V1 V2 V3
m= + + +···
Vc Vc Vc
√
= m1 2 + m2 2 + m3 2 + · · · (7.26)

Here, m1 is the modulation factor due to the carrier of amplitude V c being modulated only by the tone of ampli-
tude V 1 , and so on. See Question 7.5 for a derivation of Eq. (7.26).
 488 7 Amplitude Modulation

7.4 AM Modulators

7.4.1 Generation of AM Signal

AM signals can be generated using two different methods. In one, the carrier is passed through a device that has its
gain varied linearly by the modulating signal and, in the other, both the carrier and modulating signal are passed
through a nonlinear device.

7.4.1.1 Linearly-varied-gain Modulator

The first method is illustrated in Figure 7.13. The carrier signal is transmitted through a device whose gain G is
varied linearly by the modulating signal according to the relation

G = 1 + k1 vm (t) (7.27)

The output of the device is given by

vo (t) = Gvi (t)

= [1 + k1 vm (t)]Vc cos(2𝜋fc t)
= [Vc + kvm (t)] cos(2𝜋fc t), k = k1 Vc
≡ vam (t) (7.28)

The output is therefore an AM signal. This modulator has a modulation sensitivity k = k1 V c , where k1 is a con-
stant that determines the sensitivity of the gain G to the value of the modulating signal. The operational amplifier
(opamp) configuration shown in Figure 7.13b has the gain variation given by Eq. (7.27), and therefore will imple-
ment this method of AM. The variable input resistance Ri is provided by a field-effect transistor (FET), which is
biased (using a fixed DC voltage) to operate at the centre of its linear characteristic. A modulating signal vm (t)
connected to its gate then causes the FET to be more conducting as vm (t) increases. In effect, the source-to-drain

Input, ʋi Output, ʋo = Gʋi

Carrier signal, ʋc(t) Variable Gain
ʋc(t) = Vc cos(2πfct) G = 1 + k1ʋm(t)
(a) AM signal, ʋam(t)

Message signal, ʋm(t)

Rf
Telegram: @Persian_IEEEComSoc

ʋm(t)

–
Ri AM signal
(b)
Rf
+ ʋo(t) = 1+ ʋ (t)
Ri c
Carrier
ʋc(t)

Figure 7.13 (a) AM generation using a variable gain device; (b) Opamp implementation.
 7.4 AM Modulators 489

conductance Gi (= 1/Ri ) of the FET is made to vary linearly with the modulating signal
1
Gi = avm (t) =
Ri
or
1
Ri = (7.29)
avm (t)
Two important characteristics of an opamp, in the configuration shown, are that both the inverting (labelled −)
and noninverting (labelled +) terminals are forced to the same potential, and negligible current flows into either
of these terminals. There are two implications:
● The absence of current flow into the inverting terminal implies that the same current flows through Rf and Ri ,
which are therefore in series. Thus, the voltage vo is shared between Rf and Ri according to the ratio of their
resistance. In particular, the voltage drop across Ri , which is the voltage at the inverting terminal, is given by
Ri
v− = v
Rf + Ri o

● The voltage v− = vc , since the two terminals are at the same potential.
Therefore, the output voltage vo and the carrier voltage vc are related by
[ ]
Ri + Rf Rf
vo = vc = 1 + v (7.30)
Ri Ri c
Substituting Eq. (7.29) for Ri gives the gain of the opamp circuit
v
G = o = 1 + Rf avm (t) = 1 + k1 vm (t)
vc
We see that the amplifier has a gain that varies linearly with modulating signal as specified by Eq. (7.27), with
k1 = Rf a, and therefore vo is the AM signal given by Eq. (7.28).

7.4.1.2 Switching and Square-law Modulators

The second method for generating an AM signal is by adding together the carrier and modulating signals, and then
passing the sum signal through a nonlinear device. Diodes and suitably biased transistors provide the required
nonlinear characteristic. Transistors are, however, preferred because they also provide signal amplification. The
summation can be realised by connecting the carrier and message voltages in series at the input of the nonlinear
device, so that the device input is
vi (t) = vm (t) + vc (t) (7.31)
Telegram: @Persian_IEEEComSoc

7.4.1.2.1 Switching Modulator

The device is called a switching modulator if the principle of operation is by the device being switched on and off (at
the carrier frequency) during the positive half of the carrier signal cycle. This requires that the carrier amplitude
be much greater than the peak value of the message signal, so that the switching action is controlled entirely by
the carrier signal. The output voltage vo (t) equals input voltage vi (t) during one-half of the carrier frequency (when
the device is on) and equals zero during the remaining half-cycle (when the device is off). Therefore the output
is effectively the product of input signal and a periodic square wave vs (t), which has amplitude 1 during half the
carrier cycle (when the device is on), and amplitude 0 during the other half-cycle (when the device is off). Note
then that vs (t) has period T = 1/f c . We may write
vo (t) = vi (t)vs (t) = [vm (t) + vc (t)]vs (t) (7.32)
 490 7 Amplitude Modulation

Recall from Chapter 4 that vs (t) can be written as a Fourier series

vs (t) = Ao + A1 cos(2𝜋fc t) + A3 cos(2𝜋(3fc )t) + A5 cos(2𝜋(5fc )t) + · · ·
Substituting this expression in Eq. (7.32) and assuming for simplicity a sinusoidal message signal, we obtain the
output signal
vo (t) = [Vm cos(2𝜋fm t) + Vc cos(2𝜋fc t)][Ao + A1 cos(2𝜋fc t) + · · ·]
= Ao Vm cos(2𝜋fm t)
+ Ao Vc cos(2𝜋fc t) + A1 Vm cos(2𝜋fm t)A1 cos(2𝜋fc t)
+ A1 Vc cos2 (2𝜋fc t) + · · ·
AV AV
= Ao Vc cos(2𝜋fc t) + 1 m cos 2𝜋(fc − fm )t + 1 m cos 2𝜋(fc + fm )t
2 2
A1 Vc A1 Vc
+ + Ao Vm cos(2𝜋fm t) + cos(4𝜋fc t) + · · · (7.33)
2 2
The output vo (t) therefore contains the three frequency components that constitute an AM signal, namely the
carrier at f c , the LSF at f c − f m , and the USF at f c + f m . However, it also contains other components at DC (i.e.
f = 0), the message frequency f m , even harmonics of the carrier frequency (i.e. 2f c , 4f c , 6f c , …), and side frequencies
around odd harmonics of the carrier frequency (i.e. 3f c ± f m , 5f c ± f m , …). You may verify this by including more
terms in the Fourier series representation of vs (t), bearing in mind that it contains only odd harmonics of f c . To
obtain the required AM signal vam (t), we pass vo (t) through a bandpass filter (BPF) centred on f c and having a
bandwidth equal to twice the message bandwidth. This filter passes the carrier frequency and the two sidebands
and blocks all the other frequency components in vo (t) including f m , the maximum frequency component in the
message signal. The lowest frequency in the LSB is f c − f m , and this frequency must be larger than f m for the
message signal to be excluded by the filter. Thus, the following condition must be satisfied
fc > 2fm (7.34)

7.4.1.2.2 Square-law Modulator

The nonlinear circuit is called a square-law modulator if the device is continuously on, but with a variable resis-
tance that depends on the input voltage. In this case, the output voltage is a nonlinear function of input voltage,
which may be approximated by the quadratic expression
vo (t) = avi (t) + bvi 2 (t) (7.35)
Substituting Eq. (7.31) for vi (t), and again assuming a sinusoidal message signal for simplicity
vo (t) = a[Vm cos(2𝜋fm t) + Vc cos(2𝜋fc t)]
Telegram: @Persian_IEEEComSoc

+ b[Vm cos(2𝜋fm t) + Vc cos(2𝜋fc t)]2 (7.36)

Expanding and simplifying the above equation using relevant trigonometric identities, we obtain
1 1
vo (t) = b(Vc2 + Vm2 ) + aV m cos(2𝜋fm t) + bV 2m cos(4𝜋fm t)
2 2
+ aV c cos(2𝜋fc t) + bV c Vm cos 2𝜋(fc − fm )t
+ bV c Vm cos 2𝜋(fc + fm )t
1
+ bV 2c cos(4𝜋fc t) (7.37)
2
Notice that vo (t) contains the AM signal comprising the carrier f c and sidebands f c ± f m . However, there are
also other components at DC, f m , 2f m , and 2f c . These extra components can be excluded using a BPF as discussed
above for the switching modulator. Because there is a component at twice the message frequency 2f m , which must
 7.4 AM Modulators 491

Message ʋi(t) Nonlinear ʋo(t) AM

Σ device
ʋm(t) + ʋam(t)
+ BPF
Carrier
ʋc(t)

Figure 7.14 Square-law and switching (AM) modulator.

be excluded by the filter while still passing the LSB, it means that if f m is the maximum frequency component of
the message signal then the following condition must be satisfied

fc > 3fm (7.38)

In Eq. (7.35), the nonlinear characteristic was approximated by a polynomial of order N = 2. In practice, a
higher-order polynomial may be required to represent the input–output relationship of the nonlinear device. But
the AM signal can still be obtained as discussed above using a BPF, provided the carrier frequency is sufficiently
higher than the maximum frequency component f m of the message signal. There will in this general case be a
component in vo (t) at Nf m , and therefore the carrier frequency must satisfy the condition

fc > (N + 1)fm (7.39)

Figure 7.14 shows a block diagram of the AM modulator based on one of the two nonlinear principles, i.e.
switching or square law. The BPF is usually realised using an LC tuned circuit, which has a resonance frequency
f c , and a bandwidth that is just large enough to pass the sidebands.

7.4.2 AM Transmitters
The AM transmitter is used to launch electromagnetic waves that carry information in their amplitude variations.
It can be implemented using either low-level or high-level modulation.

7.4.2.1 Low-level Transmitter

Low-level modulation uses a nonlinear device, such as a diode, bipolar junction transistor (BJT), or FET, to gen-
erate a weak AM signal. This signal is then amplified in one or more linear amplifiers to bring it to the required
high-power level before radiation by an antenna. A low-level modulation system is shown in Figure 7.15, where
the amplitude modulator block consists of a nonlinear device followed by a BPF, as earlier discussed in connection
with Figure 7.14.
Telegram: @Persian_IEEEComSoc

Antenna

Low-power AM
Amplitude Linear
Carrier Modulator amplifiers High-power
signal AM
Audio frequency
signal

Figure 7.15 Low-level AM transmitter.

 492 7 Amplitude Modulation

Crystal NL tuned
oscillator RF amplifiers Antenna

High-power
carrier
High-power
NL tuned AM
RF amplifier

High-power
audio
NL ≡ Non-linear
Audio Linear audio
input amplifiers

Figure 7.16 High-level AM transmitter.

The radio frequency (RF) power amplifiers must be linear in order to preserve the AM signal envelope, avoiding
the nonlinear distortions discussed in Section 4.7.6. The main disadvantage of these linear amplifiers is that they
are highly inefficient in converting DC power to RF power. Transmitters for AM broadcasting rarely use low-level
operation. The method was, however, widely used on international high frequency (HF) links for radiotelephony.

7.4.2.2 High-level Transmitter

In high-level modulation, the message signal directly varies the carrier amplitude in the final RF amplifier stage
of the transmitter. Figure 7.16 shows the block diagram of a standard high-level broadcast transmitter. The carrier
is generated by a crystal-controlled oscillator and amplified by several Class C tuned RF amplifiers. Some of these
amplifiers may be operated as frequency multipliers to give the desired carrier frequency f c . The audio input is
linearly amplified, and this may be by a Class A audio amplifier followed by a high-power Class B audio ampli-
fier. This raises the audio signal to a level required to sufficiently modulate the carrier. The high-power audio is
then connected (via transformer coupling) to a final (nonlinear) Class C tuned RF amplifier of passband f c −5 to
f c + 5 kHz, where it amplitude modulates the high-power carrier signal. This gives a high-power AM signal that
is launched by the antenna. High-power transmitters that radiate several kilowatts of power still use thermionic
valves, at least in the last stage of amplification. However, low-power transmitters, such as those used in mobile
units, may be fully implemented using solid-state components.
The main advantage of the high-level transmitter is that it allows the use of the highly efficient, albeit nonlinear,
Class C tuned amplifiers throughout the RF section. Unlike the audio signal, the single-frequency carrier can be
nonlinearly amplified since the tuned section can be readily designed to eliminate the resulting harmonic products.
Telegram: @Persian_IEEEComSoc

The main drawback of a high-level transmitter is that it requires a high-power audio signal, which can only be
achieved by using expensive high-power linear amplifiers. Nevertheless, almost all AM transmitters use high-level
operation due to the above-mentioned overriding advantage.

7.5 AM Demodulators
Demodulation is the process of recovering the message signal from the received modulated carrier. If the message
signal is digital, the demodulation process is more specifically called detection since the receiver detects the range
of the modulated parameter and is not concerned with determining its precise value. However, beware! The most
common usage treats demodulation and detection as synonymous terms.
 7.5 AM Demodulators 493

A simple yet highly efficient circuit for demodulating an AM signal is commonly called an envelope or diode
detector, although a more appropriate name would be envelope or diode demodulator. This is a noncoherent demod-
ulation technique, which does not require a locally generated carrier. Coherent demodulation is also possible,
involving mixing the AM signal with a reference carrier (extracted from the incoming AM signal), but this a much
more complex circuit and therefore is not commonly used.
The input signal at a receiver usually consists of several weak carriers and their associated sidebands. A complete
receiver system must therefore include an arrangement for isolating and amplifying the desired carrier before
demodulation. A very efficient technique for achieving this goal is based on the superheterodyne principle, which
we also discuss in this section.

7.5.1 Diode Demodulator

A simple diode demodulator circuit is shown in Figure 7.17a. The circuit consists of a diode, which has the ideal
characteristic shown in Figure 7.17b, in series with an RC filter. The AM signal vam (t) is fed from a source with
internal resistance Rs .

DC blocking
iD

Cb
ʋD
Rs
(a) C RL ʋo(t) LPF ʋm(t)

ʋam(t)

iD

Vr Vf
(b) Rr = =∞ If Rf =
Ir If

Ir = 0
ʋD
Vr Vf

ʋam(t)
Telegram: @Persian_IEEEComSoc

(c) t

Figure 7.17 Diode demodulator: (a) circuit; (b) ideal diode characteristic; (c) input AM signal; (d) envelope demodulator
output v o (t) (bold curve); (e) smoothed output v m (t).
 494 7 Amplitude Modulation

ʋo(t)

(d)
t

ʋm(t)

(e) t

Figure 7.17 (Continued)

When the diode is reverse-biased, i.e. vD is negative, no current flows. That is, the diode has an infinite resistance
(Rr = ∞) when reverse-biased. Under a forward bias (i.e. vD positive), the diode has a small and constant resistance
Rf equal to the ratio of forward-bias voltage vD to diode current iD . To understand the operation of the demodulator,
let the AM signal vam (t) be as shown in Figure 7.17c. This consists of a 50 kHz carrier that is 80% modulated by a
1 kHz sinusoidal message signal. Thus
vam (t) = Vc [1 + m sin(2𝜋fm t)] sin(2𝜋fc t)
= Vam (t) sin(2𝜋fc t) (7.40)
with m = 0.8, and the envelope V am (t) = V c [1 + m sin(2𝜋f m t)]. During the first positive cycle of vam (t), the diode is
forward-biased (i.e. vD ≥ 0), and has a small resistance Rf . The capacitor C then charges rapidly from 0 V towards
V am (1/4f c ), which is the value of the envelope V am at t = 1/4f c . At the instant t = 1/4f c , the AM signal vam (t) begins
Telegram: @Persian_IEEEComSoc

to drop in value below V am = (1/4f c ), which causes the diode to be reverse-biased, since its positive terminal is now
at a lower potential than its negative terminal. With the diode effectively an open circuit, the capacitor discharges
slowly towards 0 V through the load resistance RL , and this continues for the remaining half of the first positive
cycle and for the entire duration of the first negative cycle of vam (t).
During the second positive cycle of vam (t), the diode becomes forward-biased again at the instant that vam (t)
exceeds the (remaining) capacitor voltage. The capacitor then rapidly charges towards V am (5/4f c ). At the instant
t = 5/4f c , the input vam (t) begins to drop below V am (5/4f c ), the diode becomes reverse-biased, and the capacitor
again begins to discharge slowly towards 0 V.
The process just described is repeated over and over, and this gives rise to the output signal vo (t) shown in
Figure 7.17d. The output signal contains two large frequency components – one at DC and the other at the message
frequency(s) f m , and several small (unwanted) components or ripples at frequencies f = nf c , and nf c ± f m , where
 7.5 AM Demodulators 495

n = 1, 2, 3, … The DC component is often used for automatic gain control. It can be removed by passing vo (t) through
a large DC-blocking capacitor. The ripples are usually ignored, being outside the audible range of frequencies, but
can be easily removed by lowpass filtering. If both DC and ripples are removed, it leaves a smooth message signal
vm (t), as shown in Figure 7.17e.
For the diode demodulator to work properly as described above, several important conditions must be satisfied.
Before stating these conditions it is important to note that, when a capacitor C charges or discharges through a
resistor R from an initial voltage V i (at t = 0) towards a final voltage V f (at t = ∞), the voltage drop vC across the
capacitor changes exponentially with time t according to the expression
( )
t
𝜐C (t) = Vf − (Vf − Vi ) exp − (7.41)
RC
The capacitor is charging if V f > V i and discharging if V f < V i . Figure 7.18 shows the voltage across a charging
capacitor. The maximum change in capacitor voltage from its initial value at t = 0 to its final value at t = ∞ is
V f − V i . After a time t = RC, the voltage has changed to
𝜐c (RC) = Vf − (Vf − Vi ) exp(−1)
= Vf − 0.368(Vf − Vi )
and the amount of voltage change in the time from t = 0 to t = RC is
𝜐c (RC) − 𝜐c (0) = Vf − 0.368(Vf − Vi ) − Vi
= 0.632(Vf − Vi )
This is 63.2% of the maximum change. This time interval is called the time constant of the series RC circuit. That
is, the capacitor voltage undergoes 63.2% of its total change within the initial time of one time constant. The rate of
voltage change slows down continuously with time, so that in fact the capacitor only approaches but never actually
reaches 100% of its maximum possible change, except until t = ∞. The smaller the time constant, the more rapidly
the capacitor charges or discharges, and the larger the time constant, the longer it takes the capacitor to charge or
discharge. We may now state the conditions that must be satisfied in the design of the envelope demodulator:
● The charging time constant must be short enough to allow the capacitor to charge rapidly and track the AM
signal up to its peak value before the onset of the negative-going portion of the oscillation when charging is cut
off by the diode. Since charging is done through resistance R = Rf + Rs , and the AM signal oscillates with period
T c = 1/f c , this requires that
1
(Rf + Rs )C ≪ (7.42)
fc
● The discharging time constant must be long compared to T c so that the capacitor voltage does not follow the AM
signal to fall significantly towards 0 V during its negative-going portion of its oscillation. At the same time, the
discharging time constant must be short enough for the capacitor voltage to be able to track the fastest-changing
Telegram: @Persian_IEEEComSoc

Figure 7.18 Exponential rise in voltage across a capacitor ʋC(t)

C charging through resistance R from initial voltage V i Vf
towards a final voltage V f .
0.63(Vf – Vi)

Vi

t
0 RC 2RC 3RC 4RC
 496 7 Amplitude Modulation

component of the AM signal envelope. Knowing that this component is the maximum frequency f m in the
message signal, and that the capacitor discharges through the load resistor RL , it follows that the condition that
must be satisfied is
1 1
≫ RL C ≫ (7.43)
fm fc
● The third condition is implied in Eq. (7.43) but is so important that it is separately stated here for emphasis. The
carrier frequency f c must be much larger than the maximum frequency component f m of the message signal.
In practice, f c is greater than f m by a factor of about 100 or more.
● In the above discussions, we assumed an ideal diode characteristic (see Figure 7.17b). In practice, a diode has
a nonlinear characteristic for small values of forward-bias voltage vD , followed by a linear characteristic (as
shown in Figure 7.17b) for large values of vD . The effect of the nonlinear characteristic is that diode resistance
Rf becomes a function of bias voltage, and this causes the output vo (t) to be distorted in the region of small carrier
amplitudes. To avoid this distortion the modulated carrier amplitude must always be above the nonlinear region,
which imposes the additional condition that 100% modulation cannot be employed.

Figures 7.17c–e were obtained using the following values kalpana

fc = 50 kHz (Carrier frequency)
fm = 1 kHz (Message signal frequency)
Rs = 50 Ω (AM signal source resistance)
Rf = 20 Ω (Forward-biased diode resistance)
C = 10 nF (Charging∕discharging capacitor)
RL = 10 kΩ (Load resistance)
m = 0.8 (Modulation factor) (7.44)

Note that all four conditions are satisfied by the above selection of values:

● Condition 1: The charging time constant = (Rf + Rs )C = 0.7 μs. This is much less than the carrier period
1/f c = 20 μs as required.
● Condition 2: The discharging time constant RL C = 100 μs. As required, this is much less than the period of the
maximum frequency component in the message signal 1/f m = 1000 μs and much larger (although by a factor of
five only) than the carrier period (20 μs). A choice of carrier frequency f c = 100 kHz would have satisfied this
condition better, while at the same time still satisfying condition 1. However, f c = 50 kHz was chosen in order
to produce clearer illustrative plots for our discussion.
● Condition 3: The carrier frequency f c = 50 kHz is much larger than the maximum frequency component
(f m = 1 kHz) in the message signal as required.
Telegram: @Persian_IEEEComSoc

● Condition 4: The modulation depth is 80%, which is less than 100% as required.

Figure 7.19 shows the envelope demodulator output using the same values in Eq. (7.44), except that the carrier
frequency is changed to f c = 4 kHz. Observe that the fluctuations in the output vo (t) are now more than just small
ripples. You should work out which of the above four conditions have been flouted and explain why the capacitor
discharges significantly towards zero before being recharged.

7.5.2 Coherent Demodulator

Figure 7.20a shows the block diagram of a coherent demodulator. It consists of a multiplier and a lowpass filter
(LPF). This excludes the arrangement for obtaining the carrier frequency, which is discussed later. The multiplier
 7.5 AM Demodulators 497

ʋo(t)

ʋam(t)

0 1/fm 2/fm

Figure 7.19 Input AM waveform v am (t) and envelope demodulator output v o (t) when f c = 4 kHz in Eq. (7.44).

ʋam(t) ʋo(t) ʋm(t)

(a) Multiplier

LPF
Carrier ʋc(t)

ʋam(t) ʋph(t) Voltage Carrier ʋc(t)

Phase
(b) controlled
discriminator
oscillator

ʋc(t)

Figure 7.20 (a) Coherent AM demodulator; (b) phase-locked loop (PLL).

is also called the product modulator, and its implementation is be discussed in Section 7.7. Assuming a sinusoidal
message signal, the output of the multiplier is given by

𝜐0 (t) = 𝜐am (t) × 𝜐c (t) = GV c [1 + m cos(2𝜋fm t)] cos(2𝜋fc t) × cos(2𝜋fc t)

Telegram: @Persian_IEEEComSoc

= GV c cos2 (2𝜋fc t) + mGV c cos(2𝜋fm t)cos2 (2𝜋fc t)

GV c GmV c
= + cos(2𝜋fm t)
2 2
GmV c G V
+ cos[2𝜋(2fc − fm )t] + m c cos[2𝜋(2fc )t]
4 4
GmV c
+ cos[2𝜋(2fc + fm )t]
4
The constant G is a scaling factor that represents the total gain of the transmission link from transmitter output
to demodulator input. The signal vo (t) contains the wanted message signal (at frequency f m ) and other (unwanted)
components at f = 0, 2f c −f m , 2f c , and 2f c + f m . The DC component (f = 0) is removed by a series capacitor,
 498 7 Amplitude Modulation

and the higher-frequency components are filtered out by an LPF that has a bandwidth just enough to pass the
highest-frequency component of the message signal.
The technique of coherent demodulation is very different from that of envelope demodulation. You will recall
that AM translates the message spectrum from baseband (centred at f = 0) to bandpass (centred at −f c and +f c ),
without any further change to the spectrum except for a scaling factor. Coherent demodulation performs a further
translation of the (bandpass) AM spectrum by ±f c . When this is done, the band at −f c is translated to locations
f = −2f c and 0, whereas the band at +f c is translated to locations f = 0 and +2f c . The band of frequencies at f = 0
has the same shape as the original message spectrum. It has twice the magnitude of the bands at ±2f c , being the
superposition of two identical bands originally located at ±f c . This baseband is extracted by an LPF and provides
the message signal vm (t) at the output of Figure 7.20.
The use of coherent demodulation requires that the receiver have a local carrier that is at the same frequency
and phase as the transmitted carrier. Since the received AM signal already contains such a carrier, what is needed
is a circuit that can extract this carrier from the AM signal. One way of achieving this is to use the phase-locked loop
(PLL), a block diagram of which is shown in Figure 7.20b. This consists of a voltage-controlled oscillator (VCO),
which generates a sinusoidal signal that is fed into a phase discriminator along with the incoming AM signal.
The VCO is designed to have an output at the carrier frequency f c when its input is zero. The phase discriminator
includes an LPF (not explicitly shown). It produces an output voltage vph (t) that is proportional to the phase differ-
ence between its two inputs. If vph (t) is fed as input to the VCO then it will be proportional to the phase difference
between the VCO output signal and the carrier component in the AM signal, and will cause the VCO frequency
to change in order to minimise this difference. When vph (t) = 0, the loop is said to be locked and the output of the
VCO is the required carrier.

7.5.3 AM Receivers
A radio receiver performs several important tasks.

● It selects one signal from the many reaching the antenna and rejects all the others. The receiver could be more
easily designed to receive only one particular carrier frequency, but in most cases it is required to be able to select
any one of several carrier signals (e.g. from different broadcast stations) spaced over a wide frequency range.
The receiver must therefore be capable of both tuning (to select any desired signal) and filtering (to exclude the
unwanted signals).
● It provides signal amplification at the RF stage to boost the signal to a level enough to properly operate the
demodulator circuit and at the baseband frequency stage to operate the information sink – a loudspeaker in the
case of audio broadcast.
● It extracts the message signal through the process of demodulation, which we have already discussed for the
case of AM signals.
Telegram: @Persian_IEEEComSoc

7.5.3.1 Tuned Radio Frequency (RF) Receiver

Figure 7.21 shows one simple design, which performs the above tasks. This design was used in the first broadcast
AM receivers. It consists of a tuneable RF amplifier that amplifies and passes only the desired carrier and its
associated sidebands, an envelope demodulator that extracts the audio signal, and an audio amplifier that drives
a loudspeaker. Called the tuned radio frequency (TRF) receiver, this implementation has serious drawbacks.
It is difficult to design the RF amplifier with the required level of selectivity, since the bandwidth needs to be as
small as 10 kHz to separate adjacent AM broadcast carriers. Several stages of tuned circuits are required, and this
poses problems with stray capacitance and inductance, which gives rise to unwanted feedback and possible oscil-
lations. Furthermore, each circuit stage is tuned by a separate variable capacitor, and gang-tuning these stages (i.e.
arranging for all the variable capacitors to be adjusted by a single control knob) is no small mechanical challenge.
 7.5 AM Demodulators 499

Antenna

Tunable RF Audio
Demodulator
amplifier amplifier

Loudspeaker

Figure 7.21 Tuneable radio frequency (TRF) receiver.

The bandwidth of a TRF receiver varies depending on the resonance frequency f c to which the above stages have
been tuned. The Q of a tuned circuit (i.e. a BPF) is defined as the ratio of its centre frequency f c to its bandwidth B.
This ratio remains roughly the same as the capacitance is varied (to change f c ). The bandwidth of a TRF receiver is
therefore B = f c /Q, which changes roughly proportionately with f c . For example, a receiver that has a bandwidth
B = 10 kHz at the centre (1090 kHz) of the medium wave band (540–1640 kHz) will have a narrower bandwidth
B = 4.95 kHz when tuned to a carrier frequency at the bottom end of the band. This is a reduction by a factor of
1090/540. If this receiver is tuned to a carrier at the top end of the band, its bandwidth will be about 15.05 kHz,
representing an increase by a factor of 1640/1090. Thus, at one end we have impaired fidelity due to reduced
bandwidth and at the other end we have increased noise and interference from adjacent channels due to excessive
bandwidth.

7.5.3.2 Superheterodyne Receiver

The superheterodyne receiver (or superhet for short) is an elegant solution that performs all the required receiver
functions without the drawbacks of the TRF receiver discussed above. The desired RF carrier is translated to a fixed
intermediate frequency (IF), which is the same irrespective of the carrier frequency selected. The bulk of required
amplification and selectivity is performed at the IF stage. This IF amplifier can be more easily designed since
it is required to amplify a fixed band of frequencies. All commercial receivers for analogue radio use superhet.
Figure 7.22 shows the block diagram of an AM superhet receiver. The operation of the superhet receiver is as
follows.

Mixer
IF Amplifier
Telegram: @Persian_IEEEComSoc

Tuned and BPF Demodulator

RF amplifier fIF ± B/2 kHz

Local oscillator Audio

Common tuning fLO = fc + fIF amplifier
Loudspeaker
10 kHz, AM radio
B=
200 kHz, FM radio

Figure 7.22 The superheterodyne AM receiver.

 500 7 Amplitude Modulation

7.5.3.2.1 Frequency Translation

To select the desired transmission at carrier frequency f c , the local oscillator (LO) is jointly tuned (using the same
control knob) with the RF section so that when the RF section is tuned to f c the LO frequency is at the same time
set at f c + f IF . The frequency f IF is called IF – being lower than the incoming RF carrier frequency and higher
than the message baseband frequencies. For AM radio with RF carrier in the frequency range 540–1640 kHz,
f IF = 470 kHz (or 455 kHz in North America), and for FM radio with RF carrier in the range 88–108 MHz, f IF is
typically 10.7 MHz. Thus
fLO = fc + fIF (7.45)
The required frequency translation could also be achieved with the LO maintained at frequency f IF below the
incoming carrier rather than above the carrier frequency as specified in Eq. (7.45). However, this is not done in
practice because it would lead to a much larger ratio between the highest oscillator frequency (needed to receive a
transmission at the top end of the band) and the lowest oscillator frequency (needed to receive a transmission at the
bottom end of the band). Take the example of an AM radio receiver that covers the RF carrier range 540–1640 kHz
with f IF = 470 kHz. For f LO above f c
fLO min = fc min + fIF = 540 + 470 = 1010 kHz
fLO max = fc max + fIF = 1640 + 470 = 2110 kHz
This gives a ratio f LOmax /f LO min = 2.1. Whereas for f LO below f c
fLO min = fc min + fIF = 540 − 470 = 70 kHz
fLO max = fc max + fIF = 1640 − 470 = 1170 kHz
This is a ratio f LO max /f LO min = 16.7, which is more difficult to achieve using a variable capacitor in the LO circuit.
The mixer is a nonlinear device that multiplies the modulated carrier at f c with the sinusoidal signal generated
by the LO at frequency f LO . The result is that the message signal, carried by the carrier at f c , is translated to new
frequency locations at the sum and difference frequencies
fLO ± fc = fc + fIF ± fc
= 2fc + fIF and fIF

7.5.3.2.2 IF Amplifier
The IF stage provides tuned amplification with a pass band of 10 kHz (for AM radio) centred at f IF . This amplifier
therefore performs the following functions:
● It rejects the sum-frequency signal at 2f c + f IF since this is outside its pass-band.
● It amplifies the difference-frequency signal at f IF to a level required for the correct operation of the demodulator.
Telegram: @Persian_IEEEComSoc

Note that this signal carries the exact sidebands of the original RF carrier. The mixer merely translated this
carrier along with its sidebands without any distortion.
● It removes any adjacent channels since these also lie outside its pass-band.

7.5.3.2.3 Image Interference

The superhet receiver has the disadvantageous possibility of simultaneous reception of two different transmis-
sions, one at the desired frequency f c and the other at a frequency f i equal to twice the IF above the desired
frequency. The frequency f i is referred to as the image frequency of f c . To see how this happens, note that the
unwanted transmission at a carrier frequency f i = f c + 2f IF would be translated by the mixer to the sum and
 7.6 Merits, Demerits, and Application of AM 501

difference frequencies
fi ± fLO = fc + 2fIF ± fLO
= fc + 2fIF ± (fc + fIF )
= 2fc + 3fIF and fIF
Whereas the sum frequency at 2f c + 3f IF is of course rejected at the IF stage, the difference frequency equals f IF
and is not rejected. The transmission at f i is therefore also received. It interferes with the wanted transmission at
f c . This problem is known as image interference. A practical solution to this undesirable simultaneous reception of
two transmissions is to use selective stages in the RF section that discriminate against the undesired image signal.
Since the image frequency is located far away from the wanted carrier – a separation of 2f IF , the pass-band of the
RF section does not need to be as narrow as in the case of the TRF receiver, which is required to pass only the
wanted carrier and its sidebands.

Worked Example 7.6

(a) Describe how an incoming AM radio transmission at a carrier frequency of 1 MHz would be processed.
Assume an IF f IF = 470 kHz.
(b) What is the image frequency of this transmission?

(a) The action of using a control knob to tune to the transmission at f c = 1000 kHz also sets the LO frequency to
fLO = fc + fIF = 1000 + 470 = 1470 kHz
The RF amplifier is thus adjusted to have a passband centred on f c . It boosts the carrier at f c and its sidebands,
but heavily attenuates other transmissions at frequencies far from f c , including the image frequency at fi = fc +
2fIF = 1940 kHz. The amplified carrier and its sidebands are passed to the mixer, which translates f c (along
with the message signal carried by f c in its sidebands) to two new frequencies, one at f LO + f c = 2470 kHz and
the other at f LO − f c = 470 kHz. The copy of the signal at 2470 kHz is rejected by the IF BPFs, whereas the other
copy at the IF frequency of 470 kHz is amplified and applied to an envelope demodulator, which extracts the
message signal. The message signal is amplified in a linear audio amplifier, which drives a loudspeaker that
converts the message signal to audible sound.
(b) The image frequency f i is given by
fi = fc + 2fIF = 1000 + 2 × 470 = 1940 kHz
To confirm that this is correct, we check that f i is translated by the mixer to a difference frequency equal to
the IF
fi − fLO = 1940 − 1470 = 470 kHz
Telegram: @Persian_IEEEComSoc

7.6 Merits, Demerits, and Application of AM

The most important advantage of AM over other modulation techniques is its circuit simplicity. Specifically
● AM signals are easy to generate using simple circuits such as a switching or square-law modulator.
● AM signals can also be easily demodulated using the simple envelope demodulator discussed in Section 7.5.1,
or a simple square-law demodulator discussed in Question 7.11.
 502 7 Amplitude Modulation

However, AM has the following drawbacks.

● It is wasteful of power. It is shown in Eq. (7.24) that at least two-thirds of the power is in the transmitted car-
rier, which carries no information. Only a maximum of one-third of the transmitted power is used to support
the information signal (in the sidebands). For example, to transmit a total sideband power of 20 kW, one must
transmit a total power above 60 kW, since 100% modulation is never used in practice. In this case, more than
40 kW of power is wasted.
● AM is also wasteful of bandwidth. We see in Eq. (7.15) that AM requires twice the bandwidth of the message
signal. Two sidebands, one a mirror image of the other, are transmitted even though any one of them is enough
to reproduce the message signal at the receiver.

AM is suitable for the transmission of low-bandwidth message signals in telecommunication applications that
have low-cost receiver implementation as a major design consideration. Its main application is therefore in audio
broadcasting where it is economical to have one expensive high-power transmitter and numerous inexpensive
receivers.

7.7 Variants of AM

So far, we have discussed what may be fully described as double sideband transmitted carrier amplitude mod-
ulation (DSB-TC-AM). You may also wish to describe it as double sideband large carrier amplitude modulation
(DSB-LC-AM), to emphasise the presence of a large carrier component in the transmitted signal. Various modi-
fications of this basic AM technique have been devised to save power and/or bandwidth, at the cost of increased
transmitter and receiver complexity.

7.7.1 DSB
The double sideband suppressed carrier amplitude modulation (DSB-SC-AM), usually simply abbreviated DSB, is
generated by directly multiplying the message signal with the carrier signal using a balanced modulator. The
carrier is not transmitted (i.e. the carrier is suppressed), and this saves a lot of power, which can be put into the
two sidebands that carry the message signal.

7.7.1.1 Waveform and Spectrum of DSB

The DSB waveform is given by

𝜐DSB (t) = 𝜐m (t) × 𝜐c (t) = 𝜐m (t)Vc cos(2𝜋fc t) (7.46)

Telegram: @Persian_IEEEComSoc

And for a sinusoidal message signal 𝜐m (t) = V m cos(2𝜋f m t)

𝜐DSB (t) = Vm cos(2𝜋fm t) × Vc cos(2𝜋fc t)

1 1
= Vm Vc cos 2𝜋(fc − fm )t + Vm Vc cos 2𝜋(fc + fm )t (7.47)
2 2
⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟ ⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟
LSF USF

The DSB consists of an LSF of amplitude 0.5V m V c , and a USF of the same amplitude. There is no carrier compo-
nent. Figure 7.23 shows the waveform of a DSB signal for the case f c = 20f m , and V c = 2 V m . Note that the envelope
of the DSB is not the same as the message signal, and therefore a simple envelope demodulator cannot be used
to extract the message signal as was done in basic AM. Compared to basic AM, we see that a simple (and hence
low-cost) receiver circuit has been traded for a saving in transmitted power.
 7.7 Variants of AM 503

ʋm(t)

(a) t

ʋc(t)

(b) t

Envelope
ʋdsb(t)

(c) t

Phase reversals

Figure 7.23 DSB signal waveform: (a) sinusoidal message; (b) carrier; (c) DSB.

Amn Adsbn

(a) DSB
fc

f f
fm fc – fm fc fc + fm

|Vdsb(f)|
|Vm(f)|

LSB USB
DSB
(b)
fc
Telegram: @Persian_IEEEComSoc

f f
f1 fm fc
fc + fm
fc + f1
fc – fm

fc – f1

Figure 7.24 DSB spectrum: (a) Sinusoidal message signal; (b) Arbitrary message signal.

Figure 7.24a shows the spectrum of the DSB signal in Eq. (7.47). For an arbitrary message signal consisting of
a band of sinusoids in the frequency range from f 1 to f m , each component sinusoid is translated to an LSF and
a USF. This results in the LSB and USB shown in Figure 7.24b. DSB therefore has the same bandwidth as AM,
which is twice the message bandwidth.
 504 7 Amplitude Modulation

e a b f
D1
T1 T2 g
D4 .

ʋm(t) ʋs×m(t) BPF ʋDSB(t)

fc ± fm

D2
g′
D3

e′ c d f′
+ –

ʋc(t)

Figure 7.25 Ring or lattice implementation of a balanced modulator.

7.7.1.2 DSB Modulator

A circuit that generates a DSB waveform must be able to prevent the carrier signal from getting to its output while at
the same time passing the side frequencies. This cannot be easily achieved by filtering since the side frequencies
are very close to the carrier. A balanced modulator can generate a DSB waveform, suppressing the carrier and
passing only the side frequencies to the output. It is also called a product modulator.
A realisation of balanced modulation using a diode ring is shown Figure 7.25. The diode ring or lattice mod-
ulator consists of an input transformer T 1 , an output transformer T 2 , and four diodes connected to form a ring
(a–b–c–d–a) in which each diode is traversed in the same direction (from its positive terminal to its negative ter-
minal). The carrier signal vc (t) is connected between the exact centre taps of T 1 and T 2 ; the message signal vm (t) is
connected to the primary terminals e–e′ of T 1 ; and the output signal vs × m (t) is taken from the secondary terminals
f–f′ of T 2 .
It is easy to understand the operation of this popular circuit if we take the discussion in parts. We first look at
the effect of the carrier signal and then at how the message signal is transferred from the input terminals e–e′ to
the output terminals f–f′ .

7.7.1.2.1 Carrier Suppression

Ignore the message signal for the moment and concentrate on the carrier. Let the polarity of the carrier source be
Telegram: @Persian_IEEEComSoc

as shown in Figure 7.25 during the positive cycle of the carrier. The polarity will be reversed during the negative
half of the carrier cycle. Now consider what happens during the positive half of the carrier cycle. There is a positive
voltage at node a, which forward-biases diode D1 so that it is effectively a small resistance RfD1 , and reverse-biases
diode D4 so that it is effectively an open circuit. There is also a positive voltage at node c, which forward-biases
diode D3 so that it is effectively a small resistance RfD3 , and reverse-biases diode D2 so that it is effectively an open
circuit. The circuit (ignoring the message signal) therefore reduces to what is shown in Figure 7.26a. Note that
we cut out diodes D2 and D4 (since they are open circuits) and rotated the remaining circuit clockwise through
90∘ to obtain this figure. Z 1 is the impedance of the lower half of the secondary coil of T 1 ; Z 2 is the impedance
of the upper half of this coil. Similarly, Z 3 is the impedance of the lower half of the primary coil of T 2 , and Z 2 is
the impedance of the upper half of this primary coil. The carrier voltage produces current I t , which divides into I a
that flows through Z 2 –RfD1 –Z 4 , and Ic that flows through Z 1 –RfD3 –Z 3 . If the transformers are perfectly balanced
 7.7 Variants of AM 505

T1 T1
Z1 Z2 Z1 Z2
c a c a
Ic Ia

+ It
–
RfD3 ʋc(t) RfD1 RfD2 ʋc(t) RfD4
– +
It
Ib Id
d Z3 Z4 b b Z3 Z4 d

T2 T2

(a) (b)

Figure 7.26 Currents due to carrier voltage during (a) positive and (b) negative cycles of carrier.

and the diodes are identical then

Z1 = Z2
Z3 = Z4
RfD1 = RfD4
Under this condition, I t divides equally into the two paths and
Ia = Ic
The result is that the carrier causes two equal currents to flow in opposite directions in the primary of T 2 . These
currents induce opposite magnetic fields that cancel out exactly so that no current is induced in the secondary
winding of T 2 . Thus, the carrier has been suppressed or balanced out and does not reach the output terminals f–f′
during the positive half-cycle of the carrier.
Now consider what happens during the negative half-cycle of the carrier. There is a positive voltage at node b,
which forward-biases diode D2 so that it is effectively a small resistance RfD2 , and reverse-biases diode D1 so that
it is effectively an open circuit. There is also a positive voltage at node d, which forward-biases diode D4 so that it
is effectively a small resistance RfD4 , and reverse-biases diode D3 so that it is effectively an open circuit. The circuit
(again ignoring the message signal) therefore reduces to what is shown in Figure 7.26b. To obtain this figure, we
cut out diodes D1 and D3 (since they are open circuits), twist the circuit through 180∘ so that paths b–c and a–d
no longer cross, and rotate the remaining circuit clockwise through 90∘ . By a similar argument as for the positive
cycle, it follows that the carrier is also prevented from reaching the output terminals f–f′ during the negative half of
the cycle. Thus, the function of the carrier is to switch the diodes rapidly (at carrier frequency) providing different
Telegram: @Persian_IEEEComSoc

paths for the message signal to reach the output, as we now discuss, but the carrier itself is prevented from reaching
the output. This is a notable achievement.

7.7.1.2.2 Message Signal Transfer

We have seen that during the positive cycle diodes D1 and D3 are forward biased, whereas D2 and D4 are reverse
biased. Ignoring the carrier, since it has no further effect on the circuit beyond this switching action, we obtain
the circuit of Figure 7.27a for the positive cycle. The message signal is coupled from e–e′ to a–c by transformer T 1
action. It is then applied across b–d, with a small drop across RfD1 and RfD3 (which we ignore). The message signal
is coupled from b–d to the output terminal f–f′ by transformer T 2 action. Thus, during the positive half of a carrier
cycle
𝜐s×m (t) = 𝜐m (t)
 506 7 Amplitude Modulation

e a b f
RfD1
T1 T2

ʋm(t) ʋs × m(t) = ʋm(t)

(a)

RfD3
eʹ c d fʹ

e a b f

T1 T2

(b) ʋm(t) RfD4 ʋs × m(t) = –ʋm(t)

RfD2

eʹ c d fʹ

Figure 7.27 Transfer of message signal v m (t) from input terminals e–e′ to output terminals f–f′ during (a) positive and
(b) negative cycles of carrier.

Now consider what happens during a negative half of the cycle. The effective circuit is as shown in Figure 7.27b.
The message signal is coupled to terminal a–c as before. The paths to terminal b–d have been reversed causing
the message signal to be applied to b–d with a reversed polarity. That is, signal −vm (t) is applied to b–d. The small
drop across RfD2 and RfD4 is again ignored. From b–d this signal is coupled to the output terminal f–f′ . Thus

𝜐s×m (t) = −𝜐m (t)

7.7.1.2.3 DSB Signal

From the foregoing, we see that the signal vs × m (t) is the product of the message signal vm (t) and a bipolar
unit-amplitude square wave vs (t) of period equal to the carrier period. This is demonstrated in Figure 7.28 for
a sinusoidal message signal. The bipolar square wave has amplitude +1 during the positive half of the carrier
cycle and amplitude −1 during the negative half of the cycle. Being a square wave, it contains only odd harmonic
frequencies. Furthermore, since its average value is zero, it does not contain a DC component Ao in its Fourier
Telegram: @Persian_IEEEComSoc

series. So we may write

𝜐s×m (t) = 𝜐m (t) × 𝜐s (t)

= 𝜐m (t){A1 cos[2𝜋fc t] + A3 cos[2𝜋(3fc )t] + · · ·}
= 𝜐m (t)A1 cos[2𝜋fc t] + 𝜐m (t)A3 cos[2𝜋(3fc )t] + · · · (7.48)

The first term is the required DSB signal – being a direct multiplication of the message signal and the carrier.
Subsequent terms are the sidebands of the message signal centred at odd harmonics of the carrier frequency,
namely 3f c , 5f c , …
Therefore, as shown in Figure 7.25, the DSB signal vdsb (t) is obtained by passing vs × m (t) through a BPF of centre
frequency f c and bandwidth 2f m , where f m is the maximum frequency component of the message signal. For this
 7.7 Variants of AM 507

ʋm(t)

ʋs(t) Period T = 1/fc

ʋs × m(t)

Figure 7.28 Output v s × m (t) of a diode ring modulator is the product of the message signal v m (t) and bipolar unit-amplitude
square wave v s (t).

to work, adjacent sidebands in Eq. (7.48) must not overlap. This requires that the lowest lying frequency 3f c − f m
in the LSB located just below 3f c must be higher than the highest frequency f c + f m in the USB located just above
f c . That is

3fc − fm > fc + fm
Or fc > fm (7.49)

Subsequently, a balanced modulator or product modulator will be treated as comprising a diode ring modulator
and an appropriate BPF. This leads to the block diagram of Figure 7.29 in which the product modulator receives
two inputs, namely the message signal and the carrier, and produces the DSB signal as output.

7.7.1.3 DSB Demodulator

The original message signal vm (t) can be extracted from a received DSB signal vdsb (t) by a product modulator
followed by an LPF, as shown in Figure 7.30. However, this requires that the locally generated carrier should
Telegram: @Persian_IEEEComSoc

Message DSB
Product
ʋm(t) Modulator ʋdsb(t) = ʋm(t) × Vccos(2πfct)

Carrier
Vccos(2πfct)

Figure 7.29 The product modulator consists of a ring modulator followed by a bandpass (BPF) filter of centre frequency f c
and bandwidth 2f m .
 508 7 Amplitude Modulation

Demodulated
Incoming DSB signal
Product ʋo(t)
ʋdsb(t) = LPF ʋ′m (t)
modulator
ʋm(t)Vc cos(2π fct + ϕ)

ʋLO(t)
= Vcl cos(2π fct)

Local
oscillator

Figure 7.30 Coherent demodulation of DSB.

have exactly the same frequency and phase as the carrier used at the transmitter including the phase perturba-
tions imposed by the transmission medium. This demodulation scheme is therefore referred to as coherent (or
synchronous) demodulation.

7.7.1.3.1 Effect of Phase Error

To see why phase synchronisation is important, let us assume in Figure 7.30 that variations in the transmission
medium have perturbed the phase of the missing DSB carrier by 𝜙 relative to the locally generated carrier, which
is matched in frequency. The incoming DSB signal vdsb (t) and the locally generated carrier vLO (t) are therefore
given by
𝜐dsb (t) = 𝜐m (t) × Vc cos(2𝜋 fc t + 𝜙)
𝜐LO (t) = Vcl cos(2𝜋 fc t)
The output of the product modulator is therefore
𝜐o (t) = 𝜐DSB (t) × 𝜐LO (t)
= [𝜐m (t)Vc cos(2𝜋 fc t + 𝜙)][Vcl cos(2𝜋 fc t)]
= 𝜐m (t)Vc Vcl cos(2𝜋 fc t) cos(2𝜋 fc t + 𝜙)
= K𝜐m (t) cos(𝜙) + K𝜐m (t) cos(4𝜋 fc t + 𝜙) (7.50)
1
where we have used the trigonometric identity for the product of two cosines and set the constant K = V V . The
2 c cl
last term in Eq. (7.50) is a bandpass signal centred at 2f c , and is rejected by the LPF, which therefore yields the
demodulated signal
𝜐′m (t) = K cos(𝜙) 𝜐m (t) (7.51)
Telegram: @Persian_IEEEComSoc

Thus, the received message signal is proportional to the cosine of the phase error 𝜙. If 𝜙 has a constant value
other than ±90∘ then the message signal is simply scaled by a constant factor and is not distorted. However, in
practice the phase error will vary randomly leading to random variations (i.e. distortions) in the received signal.
The received signal is maximum when 𝜙 = 0 and is zero when 𝜙 = ±90∘ . This means that a DSB signal carried by
a cosine carrier cannot be demodulated by a sine carrier, and vice versa – the so-called quadrature null effect.

7.7.1.3.2 Phase Synchronisation

The Costas loop shown in Figure 7.31 is an arrangement that keeps the locally generated carrier synchronised in
phase and frequency with the missing carrier of the incoming DSB signal. It employs two coherent demodula-
tors. A coherent demodulator consists of a product modulator fed with the incoming DSB and a locally generated
carrier and followed by an LPF. Check that you can identify the two coherent demodulators in Figure 7.31. Both
 7.7 Variants of AM 509

In-phase
Demodulated
Incoming DSB K cos(ϕ)ʋm(t) signal
Product
LPF
ʋm(t)Vc cos(2πfct) modulator
Vcl cos(2πfct + ϕ)
Voltage ʋph(t) Phase
controlled
oscillator discriminator

90°
Phase shift

Vcl sin(2πfct + ϕ)

Product
LPF
modulator K sin(ϕ)ʋm(t)

Quadrature-phase

Figure 7.31 Demodulation of DSB: Costas loop.

demodulators are fed with a carrier generated by the same VCO. However, one of the demodulators, referred to
as the quadrature-phase demodulator, has its carrier phase reduced by 90∘ (to make it a sine carrier), whereas the
other (in-phase) demodulator is fed with the VCO-generated cosine carrier that is ideally in-phase with the miss-
ing carrier of the DSB signal. The demodulated signal is taken at the output of the in-phase demodulator. When
there is no phase error, the quadrature-phase demodulator output is zero – recall the quadrature null effect – and
the in-phase demodulator provides the correct demodulated signal.
Now consider what happens when there is a phase difference or error 𝜙 between the VCO output and the missing
carrier in the DSB signal. Treating this missing carrier as the reference, its initial phase is zero and the incoming
DSB signal is
𝜐dsb (t) = 𝜐m (t)Vc cos(2𝜋 fc t)
The carrier fed to the in-phase and quadrature-phase demodulators is therefore, respectively, V cl cos(2𝜋 f c t + 𝜙)
and V cl sin(2𝜋 f c t + 𝜙). The in-phase demodulator output is then K cos(𝜙)𝜐m (t) as earlier derived, whereas the
quadrature-phase demodulator output is K sin(𝜙)𝜐m (t). Thus, a phase error causes the in-phase demodulator out-
put to drop and the quadrature-phase demodulator output to increase from zero.
Telegram: @Persian_IEEEComSoc

Phase synchronisation, which maintains the value of the phase error 𝜙 around zero, is achieved by the combined
action of the phase discriminator and the VCO. The two demodulator outputs are fed into the phase discriminator,
which produces an output voltage vph (t) proportional to 𝜙 and causes the VCO output frequency to change slightly
in such a way that the phase error is reduced towards zero.

7.7.1.4 DSB Applications

The advent of integrated circuits made it possible for the problem of phase synchronisation in DSB reception to
be overcome using affordable receivers and paved the way for the use of DSB in several areas.
● The now obsoleted analogue television systems NTSC (National Television System Committee) (in North Amer-
ica) and PAL (Phase Alternating Line) (in Europe) employed DSB to transmit colour (or chrominance) informa-
tion. Two colour-difference signals are transmitted on two subcarriers that have the same frequency but differ
 510 7 Amplitude Modulation

in phase by 90∘ . You may wish to view it this way: One signal is carried on an in-phase cosine carrier of fre-
quency f c , and the other on a (quadrature-phase) sine carrier of the same frequency. It can be shown (Question
7.13) by virtue of the quadrature null effect that the two signals will be separated at the receiver without mutual
interference. This modulation strategy may be described in full as double sideband suppressed carrier quadrature
amplitude modulation.
● DSB is also used for transmitting stereo information in FM sound broadcast at very high frequency (VHF). Send-
ing two different audio signals vL (t) and vR (t) termed the left and right channels, respectively, representing, for
example, sound from different directions entering two sufficiently spaced microphones at a live music concert,
greatly enriches the reproduction of the concert’s sound at a receiver. However, this stereo transmission must be
on a single carrier in order not to exceed the bandwidth already allocated to FM. Furthermore, it must be sent in
such a way that nonstereo receivers can give normal mono-aural reproduction. These stringent conditions are
satisfied as follows:
At the FM stereo transmitter (Figure 7.32a)
● Sum signal vL + R (t) and difference signal vL − R (t) are generated by, respectively, summing and subtracting the
two channels.
● vL − R (t) is DSB modulated using a 38 kHz carrier obtained by doubling the frequency of a 19 kHz crystal oscillator.
Let’s denote this DSB signal vdsb (t).
● The signals vL + R (t), vdsb (t), and the 19 kHz oscillator frequency are summed to give a composite signal vm (t). The
spectrum V m (f ) of the composite signal vm (t) is shown in Figure 7.32b. Clearly, there is no mutual interference
between the three signals that are summed, since each lies in a separate frequency band, vL + R (t) in the baseband
from 0 to 15 kHz, vdsb (t) in the passband from 23 to 53 kHz, and of course the pilot carrier at 19 kHz. This is an
application of frequency division multiplexing, which is discussed in detail in Chapter 13.
● The composite signal vm (t) is transmitted as vfm (t) using frequency modulation, the subject of the next chapter.
At the receiver (Figure 7.32c)
1. vm (t) is extracted from vfm (t) by frequency demodulation.
2. An LPF of 15 kHz bandwidth extracts vL + R (t) from vm (t). A nonstereo receiver plays vL + R (t) on a loudspeaker
and that completes its signal processing. A stereo receiver, however, goes further through the next steps (3)–(5).
3. The 19 kHz pilot is extracted using a narrow BPF, and vdsb (t) is extracted using a BPF of bandwidth 30 kHz
centred on 38 kHz.
4. The 19 kHz pilot is doubled in frequency. This provides a phase synchronised 38 kHz carrier that is used to
demodulate vdsb (t) to yield vL−R (t). In this way, the sum and difference signals have been recovered.
5. The left and right channels are obtained by taking the sum and difference of vL + R (t) and vL − R (t). You may wish
to check that this is the case. The two channels can now be played back on separate loudspeakers to give stereo
reproduction.
Telegram: @Persian_IEEEComSoc

7.7.2 SSB
DSB provides a power-saving improvement over basic AM. However, it still requires twice the bandwidth of the
message signal since both the lower and USBs are transmitted. It is obvious in Figure 7.24b that these sidebands
(LSB and USB) are mirror images of each other about the carrier frequency f c . That is, measuring outward from
f c , the LSB contains the same frequencies at the same amplitudes as the USB. They therefore represent the same
information – that contained in the message signal, and it is wasteful of bandwidth to send two copies of the same
information.
Single sideband suppressed carrier amplitude modulation – abbreviated SSB – transmits only one sideband. As
in DSB, the carrier is also not transmitted. Figure 7.33 shows the spectrum of an SSB signal.
 7.7 Variants of AM 511

ʋR(t)
+ ʋL + R(t)
Σ
+

– ʋm(t) ʋfm(t)
ʋL – R(t) DSB ʋdsb(t)
(a) ʋL(t)
Σ modulator
Σ FM
modulator
+ To
antenna
2fc
f×2

fc = 19 kHz
Crystal
oscillator

Vm(f)
Pilot

L+R L–R
(b)
f, kHz
0 15 19 23 53

ʋfm(t) FM ʋm(t) LPF ʋL + R(t) +

From demodulator (0 → 15 kHz) Σ 2ʋL(t)
antenna +
Narrow BPF f ×2
(19 kHz)
(c)
+
BPF ʋdsb(t) DSB Σ 2ʋR(t)
(23 → 53 kHz) demodulator –
ʋL – R(t)

Figure 7.32 FM stereo: (a) Transmitter; (b) Spectrum V m (f ) of composite signal v m (t); (c) Receiver.
Telegram: @Persian_IEEEComSoc

7.7.2.1 Merits and Demerits of SSB

SSB provides numerous benefits.

● The bandwidth of an SSB signal is the same as that of the original message signal, and is therefore half the band-
width that would be required to transmit the same message by DSB or basic AM. Thus, SSB doubles spectrum
utilisation in that it allows twice as many signals to be packed into the same frequency range as could be done
with DSB or AM.
● The passband of an SSB receiver is half that of AM and DSB receivers. As a result, noise power – proportional
to bandwidth – is reduced by a factor of two. This yields a 3 dB improvement in signal-to-noise ratio.
● Power that was spread out over two sidebands (and a carrier in the case of AM) is now concentrated into one side-
band. So, for the same output power, the SSB signal can be received with a higher signal power per unit message
 512 7 Amplitude Modulation

Assbn
LSF

Amn
f
(a) fc – fm fc
SSB
fc OR
f Assbn
fm USF

f
fc fc + fm

|Vssb(f)|

LSB

|Vm(f)|
f
fc – fm fc – f1 fc
SSB
(b)
fc OR
f
f1 fm
|Vssb(f)|

USB

f
fc fc + f1 fc + f m

Figure 7.33 Amplitude spectrum of SSB for (a) sinusoidal message signal and (b) arbitrary message signal.
Telegram: @Persian_IEEEComSoc

bandwidth than AM and DSB. For the same reason, SSB transmission can be received at greater distances than
AM and DSB transmission of the same power.
● The SSB transmitter produces a nonzero power output only when a message signal is present, unlike an AM
transmitter, which continues to radiate a high-power carrier during those time intervals when there is a pause
in the message signal. The SSB (and DSB) transmitter is therefore more efficient.
● SSB transmission is less susceptible to the phenomenon of selective fading than AM and DSB. Under selective
fading, different frequency components of a signal will arrive at the receiver having undergone different amounts
of propagation delay. This may arise in the case of sky wave propagation because these frequencies have been
effectively reflected from different layers of the ionosphere. It can be shown (see Question 7.13) that for AM and
DSB transmission to be correctly received the LSB, USB, and carrier must have the same initial phases. Selective
 7.7 Variants of AM 513

Received AM
signal envelope

Received
AM signal

(a)

Original
message signal

Received
message signal

(b)

Figure 7.34 Effect of selective fading that shifts the phases of the side-frequencies by 60∘ relative to the carrier. In this
plot, the original AM signal is an 80% modulation of a 1 MHz carrier by a 25 kHz sinusoidal message signal: (a) received AM
signal; (b) original and received message signals compared.

fading causes the phases of these three components to be altered by amounts that are not proportional to their
frequency. As a result, they appear at the receiver to have different initial phases, and this causes distortion in
the received signal.
● Figure 7.34a demonstrates an example of the potential effect of selective fading on AM signals. The (undistorted)
AM signal results from 80% modulation of a 1 MHz carrier by a 25 kHz sinusoidal message signal. We assume
Telegram: @Persian_IEEEComSoc

that selective fading causes the side frequencies to be shifted in phase by 60∘ relative to the carrier frequency.
The received AM signal will have the waveform shown in Figure 7.34a. An envelope demodulator would then
produce the output shown in Figure 7.34b, which is compared with the original message signal on the same
plot. It is apparent that signal distortion occurs when carrier and side frequencies have unequal initial phases.
● In exceptional circumstances, complete signal cancellation may occur at certain time instants.
● SSB transmission has only one sideband, which is demodulated using a locally generated carrier. The effect of
selective fading is therefore greatly reduced.
● The SSB technique reduces the effect of amplifier nonlinearity in frequency division multiplex (FDM) systems.
We saw in Chapter 4 that the effect of a nonlinear transmission medium is to generate intermodulation products,
the amplitudes of which increase with signal power. In FDM systems, where many independent signals (or
channels) are transmitted in adjacent frequency bands on the same link, some of these products will fall in
 514 7 Amplitude Modulation

bands occupied by other signals, giving rise to crosstalk. Carrier suppression allows the use of a smaller signal
power in SSB (and to a lesser extent in DSB), and therefore minimises the effect of nonlinearity.

The main disadvantage of SSB is that it requires complex and expensive circuits since a local carrier signal must
be generated that is synchronised in frequency and phase with the missing carrier in the incoming SSB signal.

7.7.2.2 SSB Modulators

SSB can be generated in two different ways, namely filtering (i.e. frequency discrimination) and phase
discrimination.

7.7.2.2.1 Frequency Discrimination Method

The most obvious method of generating an SSB signal is to first generate a DSB signal using a product modulator
and then to filter out one of the sidebands. This filtering method is shown in Figure 7.35. The main difficulty
with this method is that the required selectivity of the BPFs may be very high, making the filters expensive and
uneconomical.
If f 1 is the lowest-frequency component of the message signal then the frequency gap between the two sidebands
that must be separated by the filter is 2f 1 . In telephony with voice baseband standardised at 300–3400 Hz, this gap
is 600 Hz, whereas in TV with f 1 = 0, there is no gap between the sidebands. Obviously, therefore this method
cannot be employed with TV signals. Even for voice telephony, a BPF that has a transition width of only about
600 Hz is uneconomical at RF frequencies. For example, if the carrier frequency is, say, 3 MHz, the filter would
require a steep enough frequency response slope within the transition band so that it goes from pass band at
3 000 300 Hz to stop band at 2 999 700 Hz. The transition width in this case would be only about 0.02% of the centre
frequency.
If, however, a lower-frequency carrier is used, say f c = 100 kHz, the required transition width becomes 0.6% of
the centre frequency and the filter can be realised more cheaply. The generation of a high-frequency carrier SSB is
therefore usually done in two stages. First, a lower-frequency carrier f c1 is employed to generate the SSB. The gap
between the sidebands is only 2f 1 , where f 1 is the lowest-frequency component of the message signal. However,
because f c1 is small, an affordable BPF can be used to separate the sidebands. The output of this filter enters a
second product modulator that operates at a higher carrier frequency f c2 . This again generates two sidebands.
However, the gap between the sidebands is now 2(f c1 − f m ), where f m is the maximum frequency component of
the message signal. Since f c1 is much larger than f m , the gap between the sidebands in this second stage is large
enough for an affordable BPF to separate the two sidebands. The overall result of this two-stage procedure is the
generation of an SSB signal at a carrier frequency f c1 + f c2 .
To transmit the LSB, which occupies the frequency interval from f c2 + f c1 − f m to f c2 + f c1 − f 1 , one selects the
LSB in the first stage and the USB in the second stage. On the other hand, the USB, which occupies the frequency
interval from f c2 + f c1 + f 1 to f c2 + f c1 + f m , is transmitted by selecting the USB in both stages.
Telegram: @Persian_IEEEComSoc

Message
ʋm(t) Product ʋdsb(t) BPF ʋssb(t)
Modulator f c – fm → f c – f1 (LSB)

Carrier
(fc) BPF ʋssb(t)
fc + f1 → fc + fm (USB)

Figure 7.35 SSB generation by filtering method.

 7.7 Variants of AM 515

Antenna
fc2 + fc1 – f1

fm fc2 + fc1 – fm

f1 fc1 fc1 fc2

Message Product BPF and Product BPF and
modulator amplifier modulator power amp. High-power
SSB
Lower-frequency carrier Higher-frequency carrier
fc1 = 100 kHz fc2 = e.g. 10 MHz

Figure 7.36 SSB transmitter.

Figure 7.36 shows a block diagram of an SSB transmitter that uses the method just described to transmit the
LSB. Follow the spectrum sketches in the diagram and you will observe how the LSB is transmitted at carrier
frequency f c2 + f c1 by selecting the LSB in the first DSB modulation, and the USB in the second. Most high-power
SSB transmission in the HF band (3–30 MHz) employs the filter method.

7.7.2.2.2 Phase Discrimination Method

A different scheme for SSB generation is based on phase discrimination. Two signals are summed whose phase
relationship is such that one of the sidebands is eliminated. It is easier to demonstrate how this method works by
using a sinusoidal message signal
𝜐m (t) = Vm cos(2𝜋 fm t) (7.52)
With a carrier of frequency f c , the SSB signal is given by
{
A cos[2𝜋(fc − fm )t], LSB
𝜐ssb (t) = (7.53)
A cos[2𝜋(fc − fm )t], USB
where A represents the amplitude of the SSB signal. Using the trigonometric identities B.3 and B.4 in Appendix
B, the expressions for the SSB signals can be expanded as follows
{
A cos(2𝜋 fc t) cos(2𝜋 fm t) + A sin(2𝜋fc t) sin(2𝜋 fm t), LSB
𝜐ssb (t) = (7.54)
A cos(2𝜋fc t) cos(2𝜋 fm t) − A sin(2𝜋fc t) sin(2𝜋 fm t), USB
Equation (7.54) is very important. It shows that we may obtain a single sideband signal by adding together two
double sideband signals. Consider first the expression for the LSB. The first term is the output of a product mod-
ulator that takes the carrier and the message signal as inputs. The second term is the output of another product
modulator whose inputs are (i) the same carrier delayed in phase by 90∘ (to change it from cosine to sine) and
Telegram: @Persian_IEEEComSoc

(ii) the message signal delayed in phase by 90∘ . The sum of the outputs of these two product modulators yields
the lower sideband SSB signal, whereas the difference yields the USB SSB signal. Eq. (7.54) therefore suggests the
block diagram shown in Figure 7.37 for the generation of SSB.
A transformation that changes the phase of every positive frequency component of a signal by −90∘ and the
phase of every negative frequency component of the signal by +90∘ but does not alter the amplitude of any of
these components is known as the Hilbert transform. Thus, the inputs to the second product modulator are the
Hilbert transform of the carrier signal, and the Hilbert transform of the message signal. It is easy to obtain the
Hilbert transform of the carrier signal – a circuit that changes the phase of this single frequency by exactly −90∘
is readily available. However, an accurate hardware implementation of the Hilbert transform of the message sig-
nal is more difficult because of the wide range of frequency components that must each be shifted in phase by
exactly 90∘ .
 516 7 Amplitude Modulation

Message ʋm(t) Product

modulator

Carrier

+ (USB)
Σ SSB signal
90° 90° – ʋssb(t)
Phase shift Phase shift

Product
modulator

Figure 7.37 SSB generation by phase discrimination: Hartley modulator. USB output is shown. For an LSB output, change
subtraction to addition in the summing device.

The SSB generator based on phase discrimination implemented as shown in Figure 7.37 is known as the Hartley
modulator. It has several advantages over the frequency discrimination or filtering technique of Figure 7.36.

● The SSB signal is generated directly at the required RF frequency without the need for an intermediate
lower-frequency stage.
● Bulky and expensive BPFs have been eliminated.
● It is very easy to switch from a lower sideband to a USB SSB output. The former is obtained by adding the outputs
of the two product modulators, and the latter by subtraction of the two outputs.

The main disadvantage of the Hartley modulator is that it requires the Hilbert transform of the message signal.
This transform changes the phase of each positive frequency component in the message signal by exactly −90∘ . If
the wide-band phase shifting network shifts the phase of any frequency component in the message signal by an
amount not equal to −90∘ , it causes a small amplitude of the unwanted side frequency of this component to appear
at the output. Complete suppression of the unwanted sideband is achieved only with a phase difference of exactly
−90∘ between corresponding frequency components at the inputs of the two product modulators. In practice, it is
easier to achieve this by using two phase-shifting networks. One network shifts the message signal input to one
Telegram: @Persian_IEEEComSoc

modulator by 𝜙1 , and the other shifts the message signal input to the other modulator by 𝜙2 , with

𝜙1 − 𝜙2 = 90∘ (7.55)

7.7.2.3 SSB Demodulator

Just as for DSB reception, coherent demodulation is necessary to extract the original message signal from an incom-
ing SSB signal. Figure 7.38 shows the block diagram of an SSB demodulator. To understand the operation of this
circuit we assume for simplicity a sinusoidal message signal: Eq. (7.52). Then for USB transmission, the SSB signal
at the input of the demodulator is

𝜐ssb (t) = Vssb cos[2𝜋 (fc + fm )t] (7.56)

 7.7 Variants of AM 517

SSB signal ʋo(t) Demodulated signal

Product
LPF
ʋssb(t) modulator ʋ′m (t)

Carrier
(fc)

Figure 7.38 SSB demodulator.

Let us assume that the locally generated carrier vc (t) has a phase error 𝜙 compared to the missing carrier in
vssb (t). The output of the product modulator is then given by
𝜐o (t) = 𝜐ssb (t) × 𝜐c (t)
= Vssb cos[2𝜋(fc + fm )t] × Vc cos(2𝜋 fc t + 𝜙)
= K cos(2𝜋fm t − 𝜙) + K cos[2𝜋(2fc + fm )t + 𝜙] (7.57)
where we have set K = 0.5V ssb V c . The last term in Eq. (7.57) is a high-frequency component at 2f c + f m . This
component is eliminated by passing vo (t) through an LPF. The output of this filter gives the demodulated signal
𝜐′m (t) = K cos(2𝜋 fm t − 𝜙) (7.58)
Comparing this to the original message signal in Eq. (7.52), we see that, apart from a constant gain factor, which
is not a distortion, the demodulated signal differs from the original message signal by a phase distortion equal to the
phase error in the LO. The receiver changes the phase of each frequency component in the original message signal
by a constant amount 𝜙. From our discussion in Section 4.7, this causes a phase distortion, which is unacceptable
in data and video transmission as well as in music, where some harmonic relationships could be destroyed by a
small shift in the demodulated frequency components from their original frequency values. However, it may be
tolerated in speech transmission because the human ear is relatively insensitive to phase distortion.
Older systems minimised the phase error by inserting a low-level pilot carrier into the transmitted SSB signal,
which is then used to periodically lock the LO at the receiver. Modern systems use a crystal-controlled oscillator
along with a frequency synthesiser to generate a local carrier with good frequency stability, e.g. one part in 106 . The
need to generate a carrier of highly stable frequency is the main factor in the complexity and cost of SSB receivers.

7.7.2.4 Applications of SSB

SSB has numerous applications.
● It is employed for two-way radio communication to conserve spectrum and transmitted power requirements.
Telegram: @Persian_IEEEComSoc

Thus, more users can be accommodated in a given bandwidth, and battery life can be prolonged. SSB is used
for marine and military communication, and is very popular with radio amateurs, allowing them to maximise
signal range with a minimum of transmitted signal power.
● SSB is universally employed in the implementation of frequency division multiplexing. It allows twice as many
independent channels to be packed into a given frequency band as can be done with, say, DSB.
● Although not usually identified as such, SSB is the technique of frequency up-conversion in numerous
telecommunication systems. For example, in satellite communications, the message signal modulates a
lower-frequency carrier signal using, for example, phase modulation. The phase-modulated carrier is then
translated or up-converted to the required up-link frequency using what is essentially the filter method of
SSB modulation. At the satellite, the signal is amplified (but not phase demodulated), translated to a different
downlink frequency, and transmitted back towards the earth.
 518 7 Amplitude Modulation

|Vm1(f)|

|Visb(f)|
Message 1 Message 2
f in LSB in USB
f1 fm
ISB
fc
|Vm2(f)|
f
fc

fc – fm

fc – f1

fc + f1

fc + fm
f
f1 fm

Figure 7.39 ISB spectrum V isb (f ) resulting from two independent input signals with spectra V m1 (f ) and V m2 (f ).

● It is important to point out that the ubiquitous nonlinear device, termed a mixer, is realised as the SSB demod-
ulator shown in Figure 7.38. The mixer is found in all receivers based on the superheterodyne principle (see
Figure 7.22) and employed for frequency down-conversion to translate a bandpass RF signal to a lower centre
frequency.

7.7.3 ISB
Independent sideband amplitude modulation, abbreviated ISB, and sometimes called twin sideband suppressed car-
rier amplitude modulation, is a type of SSB. Two different message signals are transmitted on the two sidebands of
a single carrier, with one message carried in the lower sideband and the other carried in the USB.
Figure 7.39 shows the spectrum of an ISB signal that carries two arbitrary and independent message signals on
a single carrier of frequency f c . Each message signal contains a band of sinusoids (or spectrum) in the frequency
range from f 1 to f m . The spectrum of one message signal has been translated to form a lower sideband below the
carrier frequency, whereas the spectrum of the other signal forms the USB. There are therefore two sidebands
around the carrier, but each corresponds to a different message signal.

7.7.3.1 ISB Modulator

Figure 7.40 shows the block diagram of an ISB modulator. It consists of two SSB modulators (filter method) fed
Telegram: @Persian_IEEEComSoc

with the same carrier signal of frequency f c . One modulator generates a lower sideband SSB signal vlsb (t) of one
message signal, whereas the other generates a USB SSB signal vusb (t) of the other message signal. The sum of these
two signals gives the ISB signal. A reduced level of the carrier (termed pilot carrier) may be inserted for use in
coherent demodulation at the receiver.

7.7.3.2 ISB Demodulator

One way to demodulate an ISB signal is shown in block diagram form in Figure 7.41. The incoming ISB signal (cen-
tred on an RF carrier f c ) is translated down to a lower IF f IF = say 100 kHz, where affordable BPFs can be used
to separately extract the two sidebands. The frequency translation is accomplished in a mixer that uses an oscil-
lator frequency f LO = f c + f IF and generates the sum frequency f LO + f c = 2f c + f IF , and the difference frequency
f LO − f c = f IF .
 7.7 Variants of AM 519

LSB

Product
ʋm1(t) BPF
modulator

ʋlsb(t)
+ ISB
Carrier Σ ʋisb(t)
+
ʋusb(t)

Product USB
ʋm2(t) BPF
modulator

Figure 7.40 ISB modulator.

ISB fIF
and

fc 2fc + fIF fIF f1 fm

SSB Demod.
Mixer BPF ʋm1(t)
Incoming fIF
ISB fIF – fm → fIF – f1

SSB Demod.
Local oscillator BPF ʋm2(t)
fLO = fc + fIF fIF
fIF + f1 → fIF + fm

fIF f1 fm
Telegram: @Persian_IEEEComSoc

Figure 7.41 ISB demodulator.

The output of the mixer is fed into two BPFs with pass bands f IF − f m → f IF − f 1 and f IF + f 1 → f IF + f m ,
respectively. The first BPF passes the lower sideband of the frequency translated ISB signal. This sideband contains
one message signal. The other BPF passes the USB that contains the other message signal. Both filters reject the
sum frequency output of the mixer. The output of each BPF is passed through an SSB demodulator, which extracts
the original message signal. The operation of the SSB demodulator is discussed in Section 7.7.2.3 using Figure 7.38.
The only point that needs to be added in this case is that two sinusoidal signals are used at the receiver, namely
f LO (at the mixer) and f IF (at the SSB demodulator). It is important for the sum of these two signals to be matched
in frequency and phase with the missing carrier of the incoming ISB signal. A low-level pilot carrier f c may be
inserted at the transmitter and used at the receiver to maintain this synchronisation.
 520 7 Amplitude Modulation

Bandwidth = 6.8 kHz

Message 1 Message 2
in LSB in USB
(a)

f, kHz
fc – 3.4 fc – 0.3 fc fc + 0.3 fc + 3.4

Bandwidth = 7.1 kHz

Message 1 in LSB Message 2 in LSB

of carrier 1 of carrier 2
(b)

f, kHz
fc – 3.4 fc – 0.3 fc fc + 0.6 fc + 4

Figure 7.42 Bandwidth required to transmit two voice channels of baseband frequency from f 1 = 300 Hz to f m = 3400 Hz
using (a) ISB and (b) SSB at carrier spacing 4 kHz.

7.7.3.3 ISB Merits, Demerit, and Application

ISB offers all the advantages of SSB. In addition, it allows closer packing of sidebands than is possible with SSB. A
more efficient spectrum utilisation can therefore be achieved in transmitting multiple signals.
As an example, let us consider analogue speech communication. The baseband frequency is in the range
f 1 = 300 Hz to f m = 3400 Hz. The DSB bandwidth is therefore 6800 Hz, and with ISB two independent channels
can be transmitted within a bandwidth of 6800 Hz, as shown in Figure 7.42a. The separation between the
two sidebands is twice the minimum frequency component of the baseband speech signal, or 600 Hz. In SSB
modulation, each channel is carried on a separate subcarrier, and two subcarriers are required to transmit two
independent channels. These subcarriers must be allowed enough guard band to avoid mutual interference
between the two channels when realisable filters of nonzero transition width are employed to separate them. A
carrier spacing of 4 kHz (in FDM telephony) would be a realistic minimum. Thus, as shown in Figure 7.42b, the
two channels would occupy a bandwidth of 7.1 kHz in SSB. This is larger than the bandwidth requirement of
ISB by 300 Hz. A spectrum saving of 150 Hz/channel is therefore achieved with ISB. It is important to emphasise
that this saving in bandwidth has resulted simply from the closer spacing of sidebands that is possible with ISB.
Telegram: @Persian_IEEEComSoc

It can be seen in Figure 7.42 that the sideband spacing is 600 Hz with ISB and 900 Hz with SSB. The modulated
spectrum of each of the two signals remains the same (3.1 kHz) for both SSB and ISB.
The main demerit of ISB is that its per-channel circuit requirement for transmission is about the same as that
of SSB, but its receiver circuit is more extensive, and therefore more expensive. The ISB technique would only
be considered in situations where two or more independent signals must be transmitted on a link of very limited
bandwidth. ISB has only a few applications, mostly in military communication.

7.7.4 VSB
Vestigial sideband modulation, abbreviated VSB, was employed mainly in the now obsoleted analogue television
for the transmission of the luminance signal. The baseband TV signal contains frequency components down to
 7.7 Variants of AM 521

DC. This made SSB impractical since there is no frequency separation between the upper and lower sidebands to
allow the use of a realisable filter for separating them. At the same time, the bandwidth requirement of DSB was
excessive. For example, the bandwidth of a luminance signal was 4.2 MHz in the NTSC (CCIR M) TV standard
used in North America, and 5.5 MHz in the PAL standard used in Western Europe. Double sideband transmission
would have required bandwidths of 8.4 MHz and 11 MHz, respectively, for these standards, well above the total RF
bandwidth of 6 and 8 MHz, respectively, which were allocated in the two standards for transmitting one complete
television channel (including audio and colour signals).
VSB provides a compromise in which one almost complete sideband and a portion, or vestige, of the other
sideband are transmitted. The bandwidth requirement of VSB is typically about 1.25 times the message signal
bandwidth. This is larger than that of SSB, but a significant saving on DSB and AM requirements.
Figure 7.43 shows the spectrum of a VSB signal. A representative rectangular message spectrum has been
adopted in order to demonstrate the filtering of sidebands more clearly. The USB is retained in full except for
a small portion that has been filtered off. A vestige of the LSB equal to an inverted mirror image of the missing
USB portion is retained. The width of this vestigial LSB has been denoted Bv . It can be seen, measuring outward
from the carrier frequency, that all components in the LSB from Bv to f m have been eliminated. Compare this with
SSB, where all LSB components would be removed – not just a portion – and with DSB, where all LSB components
would be retained. The bandwidth of a VSB signal is therefore
VSB bandwidth = fm + B𝜐 (7.59)
where f m is the bandwidth of the message signal and Bv is the width of the vestigial sideband. Note that, although
Figure 7.43 shows a linear slope clipping of the sidebands, all that is required is that the vestigial LSB be an inverted
mirror image of the missing portion in the USB. This allows a lot of flexibility in the choice of a suitable filter to
generate the VSB signal.

7.7.4.1 VSB Modulator

The discussion of VSB thus far suggests the simple modulator shown in Figure 7.44a. A product modulator gen-
erates a DSB signal, which is passed through a BPF that has an appropriate frequency response to attenuate one
of the sidebands and reduce it to a vestige, while passing the other sideband with only a small attenuation of a
portion of the frequency band. This filter is known as the VSB filter, and its normalised frequency response (with
maximum gain equal to unity) is shown in Figure 7.44b. It is assumed in our discussion that it is the lower sideband
that is reduced to a vestige.
The frequency response of the VSB filter must be such that it allows the original message signal to be recov-
ered at the receiver using coherent demodulation. A filter with the normalised frequency response H vsb (f ) shown
in Figure 7.44b satisfies this requirement. The response of this filter is asymmetric about the carrier frequency.
However, the filter can have any arbitrary response outside the interval f = f c ± f m , where the DSB signal is
Telegram: @Persian_IEEEComSoc

|Vvsb(f)| carrier
|Vm(f)|

vestige USB
VSB of LSB
fc

f f
f1 = 0 fm fc – Bʋ fc fc + Bʋ fc + fm

Figure 7.43 VSB spectrum for a message signal of bandwidth f m . Bv is the width of the remaining portion (or vestige) of the
LSB.
 522 7 Amplitude Modulation

Message
ʋm(t) ʋdsb(t)
Product VSB ʋʋsb(t)
(a)
modulator filter

Carrier
(fc)

Hʋsb(f)

1.0
Don’t care

Don’t care
(b) 0.5

0 f
fc – fm fc – Bʋ fc fc + Bʋ fc + fm

Figure 7.44 (a) VSB modulator; (b) normalised frequency response of VSB filter.

zero. These ‘don’t care’ regions are shown shaded in Figure 7.44b, and the filter response is continued within
these regions in a completely arbitrary manner. The requirement of asymmetry about f c is satisfied by filters
with a variety of response slopes. A linear-slope filter is shown in Figure 7.44b for simplicity, but this is not
required.
If the width of the vestigial sideband is Bv , the filter response must be zero in the frequency range f c − f m → f c − Bv
in order to eliminate this portion of the LSB. It follows from the asymmetry condition that the response must be
unity in the interval f c + Bv → f c + f m so that this portion of the USB is passed with no attenuation. Thus

⎧1 fc + Bv ≤ f ≤ fc + fm
⎪
⎪0 fc − fm ≤ f ≤ fc − B𝜐
⎪
H𝜐sb (f ) = ⎨0.5 f = fc (7.60)
⎪
⎪Asymmetric fc − B𝜐 ≤ f ≤ fc + B𝜐
⎪Arbitrary Otherwise
⎩
Telegram: @Persian_IEEEComSoc

7.7.4.2 VSB Demodulator

If the normalised frequency response of the VSB filter used in the modulator satisfies Eq. (7.60) then a VSB signal
can be extracted by coherent demodulation, as shown in Figure 7.45. The operation of a coherent demodulator
has been discussed in previous sections and will not be repeated here, except to remind the reader that the sum
frequency output of the product modulator is rejected by the LPF. We therefore concentrate on the difference fre-
quency, which is the output of the LPF. Our concern here will be to demonstrate how the vestigial LSB compensates
perfectly for the attenuated portion of the USB, resulting in the recovery of the original message spectrum.
We need to use the double-sided VSB spectrum shown in Figure 7.46a. The difference frequency output is the
result of the band of positive frequencies in the VSB signal being translated downwards by f c , and the band of
negative frequencies being translated upwards by f c . You will recall from Chapter 4 that the amplitude spectrum
 7.7 Variants of AM 523

VSB signal ʋo(t) Demodulated signal

Product
LPF
ʋʋsb(t) modulator ʋm(t)

Carrier
(fc)

Figure 7.45 Coherent demodulation of VSB.

|Vʋsb(f)|

(a)

–fc – fm –fc – Bʋ fc –fc + Bʋ fc – Bʋ fc fc + Bʋ fc + fm

|Vm(f)| |Vm(f)|

(b)
=

–fm –Bʋ Bʋ fm –fm fm

Figure 7.46 (a) Double-sided spectrum of VSB signal; (b) output of LPF.

of a real signal has even symmetry. Thus, if a positive frequency component is shifted to the left (i.e. downwards
in frequency) then the corresponding negative frequency component is shifted to the right (i.e. upwards in fre-
quency) by the same amount. Figure 7.46b shows the double-sided spectrum of the LPF output. The negative
band of frequencies −f c − f m → −f c + Bv in the VSB signal vvsb (t) has been translated to −f m → Bv , whereas the
corresponding positive band f c − Bv → f c + f m has been translated to −Bv → f m . These bands overlap as shown
and the resultant spectrum, which is the spectrum of the LPF output, is the double-sided equivalent of the original
Telegram: @Persian_IEEEComSoc

message spectrum shown earlier in Figure 7.43. The condition of asymmetry about f c is required so that the sum
of the bands in the region of overlap equals a constant – normalised to unity.
In commercial television receivers, where the cost of the receiver is an important design consideration, the use
of envelope demodulation (as in AM) is preferred to coherent demodulation. To accomplish this the carrier signal
is added to the VSB signal before transmission. A cheap diode demodulator (discussed in Section 7.5.1) can then
be used to recover the message signal. However, the recovered signal is distorted in that frequency components
(from DC to Bv ) that appear in both sidebands are doubled in amplitude compared to components in only one
sideband (from Bv to f m ). To compensate for this the diode demodulator is preceded with a tuned amplifier, the
gain of which varies asymmetrically about the carrier frequency. Figure 7.47a shows a basic block diagram of an
envelope demodulator for a VSB signal, with the frequency response of the VSB compensation amplifier shown
in Figure 7.47b.
 524 7 Amplitude Modulation

VSB signal
(with carrier) Demodulated signal
VSB compensation Envelope
(a)
tuned amplifier demodulator ʋm(t)
ʋʋsb(t)

Gain

carrier
A

(b) A/2

0 f
fc – fm fc – Bʋ fc fc + Bʋ fc + fm

Figure 7.47 (a) Envelope demodulation of VSB; (b) gain response of VSB compensation tuned amplifier.

7.8 Summary

In this chapter, we have studied amplitude modulation and its four variants in detail. The basic scheme,
DSB-TC-AM, is referred to simply as AM. The amplitude of a sinusoidal carrier signal is varied proportionately
with the message signal between a nonnegative minimum value and a maximum value. The result is that the
AM signal consists of the carrier signal and two sidebands. The carrier signal does not carry any information,
but its presence allows a simple envelope demodulation scheme to be employed at the receiver to recover
the message signal. The main advantage of AM is the simplicity of the circuits required for transmission and
reception. AM can be generated using a nonlinear device followed by a suitable filter. It can be demodulated
using an envelope demodulation circuit that consists of a diode, capacitor, and load resistor. However, AM is very
wasteful of power. At least two-thirds of the transmitted power are in the carrier. To put, say, 10 kW of power in
the information-bearing sidebands, one must generate a total power of at least 30 kW. AM is therefore employed
mainly in sound broadcasting, where it is advantageous to have numerous cheap receivers and one expensive
high-power transmitter.
To save power the AM carrier may be suppressed, leading to the first variant of AM known as double sideband
suppressed carrier amplitude modulation, abbreviated simply as DSB. There is, however, a penalty. The simple
Telegram: @Persian_IEEEComSoc

envelope modulator can no longer be used at the receiver because the envelope of the missing carrier does not
correspond with the message signal. Coherent demodulation is required, which introduces increased circuit com-
plexity. The Costas loop was discussed, which allows the locally generated carrier to be synchronised in phase
and frequency with the missing carrier in the incoming DSB signal. Another way of achieving synchronisation is
to insert a low-level pilot carrier at the transmitter, which is extracted at the receiver and scaled in frequency to
generate the required carrier. DSB is used in those radio communication applications involving low-bandwidth
message signals that must be transmitted over a long range with a minimum of power.
It was noted that AM and DSB both transmit two sidebands, one a mirror image of the other, which carry iden-
tically the same information. The bandwidth requirement can be halved, with a further reduction in the required
transmitted power level, if both the carrier and one sideband are removed. This leads to what is known as single
sideband suppressed carrier amplitude modulation, abbreviated simply as SSB. Telecommunication applications
 Questions 525

that favour SSB are those with significant bandwidth constraints, limited transmit power capability, and compa-
rable numbers of transmitters and receivers.
Independent sideband modulation (ISB) places two message signals onto one carrier, with one message in each
of the sidebands. We showed that it makes extra spectrum saving by allowing the two sidebands to be more closely
packed than is possible with SSB. However, it requires more extensive circuitry than SSB, and is therefore less
commonly used.
When a message signal has a combination of a large bandwidth and significant frequency components near DC
then neither DSB nor SSB is suitable. DSB would involve excessive bandwidth, and SSB would require impractical
filters to separate the sidebands. A compromise technique, called vestigial sideband amplitude modulation (VSB),
sends one nearly complete sideband plus a vestige of the other sideband. This is achieved at the transmitter by fil-
tering the DSB signal with a filter whose response is asymmetric about the carrier frequency. The original message
signal can be recovered at the receiver by straightforward coherent demodulation. A cheaper envelope demodula-
tor is used in analogue TV receivers. To do this the carrier must be inserted into the VSB signal at the transmitter,
and the incoming VSB signal must first be passed through a tuned amplifier that compensates for those frequency
components appearing in only one sideband.
This completes our study of amplitude modulation applied to analogue message signals. We shall briefly return
to it in Chapter 11, when we apply it to digital signals. In the next chapter, we take another important step in our
study of communication systems by examining in some detail the techniques of frequency modulation and phase
modulation (jointly referred to as angle modulation) as applied to analogue message signals.

Questions

7.1 A carrier signal 𝜐c (t) = 5 sin(2𝜋 × 106 t) V is modulated in amplitude by the message signal vm (t) shown in
Figure Q7.1.
(a) Sketch the waveform of the AM signal that is produced.
(b) What is the modulation factor?
(c) Determine the modulation sensitivity that would give 100% modulation.

7.2 An oscilloscope display of an AM waveform involving a sinusoidal message signal shows a maximum
peak-to-peak value of 10 V and a minimum peak-to-peak value of 2 V. Calculate
(a) Carrier signal amplitude
(b) Message signal amplitude
(c) Modulation index.
Telegram: @Persian_IEEEComSoc

ʋm(t), volt
4

t, μs
3 9 15 18
–2

–4

Figure Q7.1 Question 7.1

 526 7 Amplitude Modulation

7.3 An AM signal is analysed by a spectrum analyser, which shows that there are the following frequency
components:
● 998 kHz of amplitude 10 V

● 1 MHz of amplitude 50 V

● 1.002 MHz of amplitude 10 V.

(a) Specify the carrier and message signals as functions of time, assuming that the initial phase of each
is zero.
(b) Determine the depth of modulation.
(c) What power would be dissipated by this AM signal in a 50 Ω load?

7.4 An engineering student wishing to improve AM power utilisation generates an AM signal in which the
carrier and each sideband have equal power. By calculating the modulation factor, determine whether a
conventional (noncoherent) AM receiver would be able to demodulate this AM signal.

7.5 A carrier of amplitude V c is modulated by a multitone message signal, which is made up of sinusoids of
amplitudes V 1 , V 2 , V 3 , … Starting from Eq. (7.25), show that the modulation factor is given by
√
m = m12 + m22 + m32 + · · ·

where m1 = V 1 /V c , m2 = V 2 /V c , m3 = V 3 /V c , …

7.6 The carrier vc (t) = 120sin(4 × 106 𝜋t) V is amplitude modulated by the message signal

vm (t) = 80 sin(40 × 103 𝜋t) + 50 sin(80 × 103 𝜋t) + 30 sin(100 × 103 𝜋t) V.

(a) Sketch the AM signal spectrum.

(b) Determine the fraction of power in the sidebands.
(c) Determine the modulation index.

7.7 A 75% modulated AM signal has 1 kW of power in the lower sideband. The carrier component is attenuated
by 4 dB before transmission, but the sideband components are unchanged. Calculate:
(a) The total transmitted power
(b) The new modulation index.

7.8 A 1.2 MHz carrier of amplitude 10 V is amplitude modulated by a message signal containing two frequency
components at 1 and 3 kHz, each having amplitude 5 V. Determine
Telegram: @Persian_IEEEComSoc

(a) The bandwidth of the message signal.

(b) The width of each of the sidebands of the AM signal.
(c) The condition under which the bandwidth of the AM signal is twice the width of each sideband.
(d) The modulation index.
(e) The total power in the AM signal.

7.9 The output voltage vo and input voltage vi of a nonlinear device are related by

vo = vi + 0.02v2i

A series connection of a carrier signal source of amplitude 20 V and frequency 100 kHz, and a message
signal source of amplitude 10 V and frequency 5 kHz provides the input to this device.
 Questions 527

Figure Q7.11 Question 7.11 iD

RL ʋo(t)
ʋam(t)

(a) Sketch the amplitude spectrum of the output signal vo .

(b) Specify the frequency response of a filter required to extract the AM signal component of vo without
distortion.
(c) Determine the modulation index of the AM signal.

7.10 The envelope demodulator in an AM superheterodyne receiver consists of the diode demodulator circuit
shown in Figure 7.17a. Determine suitable values of load resistance RL and capacitance C, assuming a
forward-bias diode resistance Rf = 20 Ω, and IF amplifier output impedance Rs = 50 Ω. Note that carrier
frequency is f IF = 470 kHz and the message signal is audio of frequencies 50 Hz to 5 kHz.

7.11 A square-law demodulator is shown in Figure Q7.11, where the input voltage vam (t) is the incoming AM
signal. For small input voltage levels, the diode current is a nonlinear function of vam (t) so that the output
voltage vo (t) is given approximately by
𝜐o = 𝛼1 𝜐am + 𝛼2 𝜐2am
(a) Evaluate the output voltage vo (t) and show that it contains the original message signal.
(b) Discuss how the message signal may be extracted from vo (t) and the conditions that must be satisfied
to minimise distortion.

7.12 .(a) Show that the block diagram in Figure Q7.12a generates an AM waveform vam (t). This connection con-
sists of a linear adder, a unity gain multiplier, and two signal generators (Sig. Gen.), one generating a
sinusoidal signal of amplitude A1 at frequency f m and the other generating a sinusoidal carrier signal
of amplitude A2 and frequency f c > > f m .
(b) Determine the amplitude and frequency settings on the signal generators in (a) required to produce the
AM waveform sketched in Figure Q7.12b.
(c) Sketch the amplitude spectrum of the AM signal of Figure Q7.12b.
(d) Determine the rms value of the AM signal of Figure Q7.12b.
Telegram: @Persian_IEEEComSoc

7.13 Quadrature amplitude modulation (QAM) is used to transmit two DSB signals within the same bandwidth
that would be occupied by one DSB signal. One message signal vm1 (t) is carried on a carrier of frequency
f c , whereas the other (independent) message signal vm2 (t) is carried on a carrier of the same frequency but
with a 90∘ phase difference. The QAM signal is therefore given by
𝜐qam (t) = 𝜐m1 (t)Vc cos(2𝜋 fc t) + 𝜐m2 (t)Vc sin(2𝜋 fc t)
(a) Draw the block diagram of a QAM modulator that generates vqam (t).
(b) Show that the message signal vm1 (t) can be extracted at the receiver by passing vqam (t) through a coherent
demodulator that uses an oscillator generating a synchronised carrier Vc′ cos(2𝜋 fc t), and that another
coherent demodulator operating with carrier signal Vc′ sin(2𝜋 fc t) extracts vm2 (t).
(c) Draw the block diagram of a QAM demodulator.
 528 7 Amplitude Modulation

ʋm(t) Unity gain Linear ʋam(t)

Sig. Gen.
fm A1 multiplier adder
(a)

Sig. Gen. ʋc(t)

fc A2

1 μs
72 V 20 μs

8V
(b) 0

Figure Q7.12 Question 7.12

7.14 Let a sinusoidal message signal of frequency f m be transmitted on a carrier of frequency f c . Assume that,
due to selective fading, the lower side frequency reaches the receiver with a phase 𝜙1 relative to the carrier,
whereas the upper side frequency has a phase 𝜙2 relative to the carrier.
(a) Obtain an expression for the coherently demodulated signal, given that the transmission is DSB and
the local carrier is perfectly synchronised with the missing carrier in the incoming signal. Discuss the
distortion effects of 𝜙1 and 𝜙2 and specify the condition under which there is complete cancellation so
that the demodulated signal is zero.
(b) Sketch the AM waveform for 𝜙1 = 60∘ and 𝜙2 = 0∘ . How does the selective fading affect the output of an
envelope demodulator?
(c) Use a sketch of the AM waveform to examine the effect on the AM envelope when selective fading
attenuates the carrier signal by 3 dB more than the attenuation of the side frequencies. Assume that the
side frequencies remain in phase with the carrier.
Telegram: @Persian_IEEEComSoc
 529

Frequency and Phase Modulation

A journey across the Atlantic is impossible if you swim, dangerous if you canoe, long if you sail, but safe and
fun if you fly.

In this Chapter
✓ Basic concepts of frequency modulation (FM) and phase modulation (PM): a simplified and lucid approach
to the theory of angle modulation.
✓ FM and PM waveforms: a detailed discussion of all types of angle modulated waveforms. You will be able
to sketch these waveforms and to identify them by visual inspection.
✓ Spectrum and power of FM and PM: you will be able to solve various problems on the spectrum, bandwidth,
and power of FM and PM signals. Narrowband FM and PM are discussed with comparisons to amplitude
modulation (AM).
✓ FM and PM modulators: a detailed discussion of various methods of applying the theory of previous
sections to generate FM and PM signals.
✓ FM and PM demodulators: a discussion of how to track the frequency variations in received angle modu-
lated signals and to convert these frequency variations to voltage variations.
✓ FM transmitters and receivers: a discussion of the building blocks and signal processing in a complete FM
communication system, including the tasks of pre-emphasis and de-emphasis. A simplified treatment of the
trade-off between transmission bandwidth and signal-to-noise power ratio (SNR) is also presented.
✓ Noise effect: a phasor approach in the analysis of noise effect in FM receivers.
✓ Features overview: merits, demerits, and applications of angle modulation.
Telegram: @Persian_IEEEComSoc

8.1 Introduction
Modulation plays very significant roles in telecommunication, as briefly discussed in Chapter 1. AM, treated in
detail in the previous chapter, is the oldest modulation technique and is obtained by varying the amplitude of a
high-frequency sinusoidal carrier in sync with the message signal. One of the problems with AM is that additive
noise in the transmission medium will also impose variations on the amplitude of the modulated carrier in a
manner that is impossible for the receiver to separate from the legitimate variations caused by the message signal.
The received message signal will therefore be corrupted to some extent by noise.

Communication Engineering Principles, Second Edition. Ifiok Otung.

© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 530 8 Frequency and Phase Modulation

An alternative modulation technique that is less susceptible to additive noise was first proposed in 1931. In this
technique, which is given the generic name angle modulation, the amplitude V c of the sinusoidal carrier is kept
constant, but the angle 𝜃 c of the carrier is varied in sync with the message signal. The angle 𝜃 c of a sinusoidal
carrier at any instant of time t depends on the frequency f c and initial phase 𝜙c of the carrier, according to the
relation
𝜃c (t) = 2𝜋fc t + 𝜙c (8.1)
Thus, the angle of the carrier may be varied in one of two ways:
(i) By varying the frequency f c of the carrier, giving rise to what is known as frequency modulation; or
(ii) By varying the initial phase 𝜙c of the carrier, resulting in phase modulation. Some textbooks refer to 𝜙c as the
phase shift.
This chapter is devoted to the study of the related techniques of FM and PM. First, we introduce FM and PM
using a simple staircase signal, and define common terms such as frequency sensitivity, phase sensitivity, frequency
deviation, and percent modulation. Next, we study the angle modulation of a carrier by a sinusoidal message sig-
nal. It is shown that FM and PM are implicitly related in that FM transmission of a message signal is equivalent
to the PM transmission of a suitably lowpass filtered version of the message signal. Narrowband and wideband
angle modulations are discussed in detail, including two of the most common definitions of the bandwidth of FM
and PM.
The modulator and demodulator circuits that implement FM and PM schemes are discussed in block diagram
format. The presentation also includes FM transmitter and receiver systems, and concludes with a discussion of
the applications of angle modulation and its merits and demerits compared to AM.

8.2 Basic Concepts of FM and PM

Angle modulation (FM or PM) represents a message signal using nonlinear variations in the angle of a sinusoidal
carrier signal. The amplitude of the carrier signal is not varied, and therefore any noise-imposed fluctuations in
carrier amplitude are simply ignored at the receiver. This gives FM and PM a greater robustness against noise and
interference than is possible with AM. The information contained in the variations of the message signal is per-
fectly preserved in the variations of the carrier angle, and this information is recovered at the receiver by tracking
the carrier angle. As with AM, there is a great deal of freedom in the choice of the frequency of the unmodulated
carrier signal, allowing information transmission through a variety of transmission media and communication
systems.
The angle 𝜃 c of an unmodulated sinusoidal carrier of frequency f c is given by Eq. (8.1), which shows that 𝜃 c
Telegram: @Persian_IEEEComSoc

increases linearly with time at the constant rate 2𝜋f c rad/s. The amplitude of this carrier remains constant at V c .
Figure 8.1 is a plot of both parameters (the angle and amplitude) of an unmodulated carrier as functions of time.
We see in the previous chapter that the effect of AM is to cause the amplitude to deviate from the constant value V c
in accordance with the instantaneous value of the message signal. The angle of an amplitude-modulated carrier is
not varied by the message signal, so it continues to be a linear function of time. What happens in the case of angle
modulation is that the carrier amplitude remains constant with time, but the carrier angle is caused to deviate from
being a linear function of time by variations of the message signal. The carrier angle has two components, namely
 8.2 Basic Concepts of FM and PM 531

Unmodulated
angle, θc
Amplitude Vc
(volt)
Unmodulated
amplitude
Δθc
Δθc Slope = = 2πfc
Δt
Initial phase ϕ Δt
(rad) c
t
t=0

Figure 8.1 Amplitude and angle of an unmodulated carrier as functions of time.

frequency f c and initial phase 𝜙c , and the manner of the deviation depends on which of these two components is
varied by the message signal.
As an example, Figure 8.2a shows a simple staircase message signal that has the following values
⎧ 0 V, 0 ≤ t < 1 ms
⎪
⎪ 2 V, 1 ≤ t < 2 ms
⎪ 1 V, 2 ≤ t < 3 ms
vm (t) = ⎨ (8.2)
⎪−1 V, 3 ≤ t < 4 ms
⎪−2 V, 4 ≤ t < 5 ms
⎪ 0 V, 5 ≤ t ≤ 6 ms
⎩
Let this signal modulate the angle of the carrier
vc (t) = Vc cos[𝜃c (t)]
= Vc cos(2𝜋fc t + 𝜙c ) (8.3)
In the following discussion we use the values amplitude V c = 2 V, frequency f c = 3 kHz, and initial phase
𝜙c = −𝜋/2 rad to demonstrate the angle modulation process. Of course, any other set of values could have been
selected. The modulation can be performed by varying either the frequency or the initial phase of the carrier, as
discussed below.
Telegram: @Persian_IEEEComSoc

8.2.1 Frequency Modulation Concepts

In FM, the frequency of the carrier at any given time is changed from its unmodulated value f c by an amount that
depends on the value of the message signal at that time. The instantaneous frequency f i of the carrier is given by
fi = fc + kf vm (t) (8.4)
 532 8 Frequency and Phase Modulation

ʋm(t), V
2

1
(a) t (ms)
0 2 4 6
–1

–2
θc(t), rad
71π/2

59π/2

47π/2

(b) 35π/2
Phase modulated angle

23π/2 Unmodulated angle

Frequency modulated angle
11π/2

–π/2
t (ms)
0 1 2 3 4 5 6

Figure 8.2 (a) Staircase message signal; (b) angle of sinusoidal carrier. (Frequency sensitivity k f = 1 kHz/volt; Phase
sensitivity k p = 𝜋/2 rad/V.)

where kf is the change in carrier frequency per volt of modulating signal, called frequency sensitivity and expressed
in units of Hz/V. In our example, with kf = 1 kHz/V and the message and carrier signals specified in Eqs. (8.2) and
(8.3), the instantaneous frequency of the frequency modulated carrier is as follows
⎧
⎪3 kHz, 0 ≤ t < 1 ms
⎪5 kHz, 1 ≤ t < 2 ms
⎪
⎪4 kHz, 2 ≤ t < 3 ms
fi = fc + kf vm (t) = ⎨ (8.5)
⎪2 kHz, 3 ≤ t < 4 ms
Telegram: @Persian_IEEEComSoc

⎪1 kHz, 4 ≤ t < 5 ms
⎪
⎪3 kHz, 5 ≤ t ≤ 6 ms
⎩
In the interval 0 ≤ t < 1 ms the message signal is 0 V, so the carrier frequency is unchanged from its unmodu-
lated value f c = 3 kHz. However, in the interval 1 ≤ t < 2 ms the message signal is 2 V, and the carrier frequency is
changed from f c by an amount kf × vm (t) = (1 kHz/V) × 2 V = 2 kHz. The instantaneous carrier frequency during
this interval is therefore fi = 5 kHz. The values of fi in other time intervals are similarly determined.
Now consider the effect of this variable carrier frequency on the angle 𝜃 c (t) of the carrier. The carrier angle 𝜃 c
starts at the initial value 𝜙c (= −𝜋/2 in this example). During the first 1 ms interval when the modulating signal
vm (t) = 0 V, f i = f c (= 3 kHz in this example), and 𝜃 c increases at the rate of 2𝜋f i rad/s = 6𝜋 rad/ms, reaching
the value 𝜃 c = 11𝜋/2 at t = 1 ms. At this time instant vm (t) increases to 2 V, so the instantaneous carrier frequency
 8.2 Basic Concepts of FM and PM 533

is increased to f i = 5 kHz, as already shown, and the carrier angle then increases at a faster rate 2𝜋f i rad/s or
10𝜋 rad/ms, reaching the value 𝜃 c = 31𝜋/2 at t = 2 ms. During the third 1 ms interval the instantaneous frequency
f i = 4 kHz, and 𝜃 c increases from 31𝜋/2 at the rate 8𝜋 rad/ms to reach 𝜃 c = 47𝜋/2 at t = 3 ms.
Following the above reasoning, the values of carrier angle at all time instants are obtained. This is plotted in
Figure 8.2b. The carrier angle is constantly increasing and the instantaneous value of the carrier changes cyclically
(i.e. oscillates), taking on the same value at carrier angles that differ by integer multiples of 2𝜋. However, unlike
the unmodulated carrier, the rate of increase of carrier angle (≡ 2𝜋f i rad/s) is not constant but changes in sync
with the message signal. The carrier oscillates faster (than f c ) for positive message values, and more slowly for
negative message values. If, as in this example, the message signal has zero average value (i.e. equal amounts of
negative and positive values) then the carrier oscillation (or increase in angle) is slowed down by as much as it is
speeded up at various times. The final value of 𝜃 c is then the same as that of the unmodulated carrier. However,
the modulated carrier has increased from its initial angle value (𝜙c ) to its final angle value using a variable rate of
increase, which is what conveys information. On the other hand, the unmodulated carrier maintains a constant
rate of increase of angle in going between the same endpoints and therefore conveys no information.
Let us now introduce several important terms that are associated with FM. We show later that these terms are
also applicable to PM. It is obvious from Eq. (8.4) that the instantaneous frequency f i increases as the modulating
signal vm (t) increases. The minimum instantaneous frequency f imin occurs at the instant that vm (t) has its min-
imum value V min , and the maximum instantaneous frequency f imax occurs when vm (t) has its maximum value
V max
fimin = fc + kf Vmin
fimax = fc + kf Vmax (8.6)
The difference between f imax and f imin gives the range within which the carrier frequency is varied, and is known
as the frequency swing f p−p
fp−p = fimax − fimin
= kf (Vmax − Vmin )
= kf Vmp−p (8.7)
where V mp−p is the peak-to-peak value of the modulating signal vm (t).
The maximum amount by which the carrier frequency deviates from its unmodulated value f c is known as the
frequency deviation f d . It depends on the frequency sensitivity kf of the modulator, and the maximum absolute
value |vm (t)|max of the modulating signal. Thus
fd = kf |vm (t)|max (8.8)
A maximum allowed frequency deviation is usually set by the relevant telecommunication regulatory body in
order to limit bandwidth utilisation. This maximum allowed frequency deviation is called the rated system devia-
Telegram: @Persian_IEEEComSoc

tion F D
⎧75 kHz, FM radio (88 − 108 MHz band)
⎪
⎪25 kHz, TV sound broadcast
FD = ⎨ (8.9)
⎪ 5 kHz, 2-way mobile radio (25 kHz bandwidth)
⎪2.5 kHz, 2-way mobile radio (12.5 kHz bandwidth)
⎩
The ratio (expressed as a percentage) of actual frequency deviation f d in an FM implementation to the maximum
allowed deviation is known as the percent modulation m
fd
m= × 100% (8.10)
FD
 534 8 Frequency and Phase Modulation

Note that this concept of percent modulation is different from that associated with AM, where there is no regu-
latory upper limit on the deviation of the carrier amplitude from its unmodulated value V c .
The ratio of frequency deviation f d of the carrier to the frequency f m of the modulating signal is known as the
modulation index 𝛽
fd
𝛽= (8.11)
fm
The modulation index has a single value only in those impractical situations where the modulating signal is a
sinusoidal signal – this contains a single frequency f m . For the practical case of an information-bearing message
signal, a band of modulating frequencies from f 1 to f m is involved. The modulation index then ranges in value
from a minimum 𝛽 min to a maximum 𝛽 max , where
fd fd
𝛽min = ; 𝛽max = (8.12)
fm f1
In this case, the more useful design parameter is 𝛽 min , which is also called the deviation ratio D. It is the ratio
between frequency deviation and the maximum frequency component of the message signal. Deviation ratio
replaces modulation index when dealing with nonsinusoidal message signals. It corresponds to a worst-case situ-
ation where the maximum frequency component f m of the message signal has the largest amplitude and therefore
causes the maximum deviation of the carrier from its unmodulated frequency f c . Thus, if an arbitrary message
signal of maximum frequency component f m modulates a carrier and causes a frequency deviation f d , then
fd
D= (8.13)
fm

Worked Example 8.1

A message signal vm (t) = 2sin(30 × 103 𝜋t) volt is used to frequency modulate the carrier vc (t) = 10sin(200 ×
106 𝜋t) volt. The frequency sensitivity of the modulating circuit is kf = 25 kHz/V. Determine
(a) The frequency swing of the carrier.
(b) The modulation index.
(c) The percent modulation.

(a) The message signal has amplitude V m = 2 V, and peak-to-peak value V mp–p = 2V m = 4 V. From Eq. (8.7), the
frequency swing is given by
fp−p = kf Vmp−p
= (25 kHz∕V) × 4 V = 100 kHz
Telegram: @Persian_IEEEComSoc

(b) The maximum absolute value |vm (t)|max of this (sinusoidal) message signal is its amplitude V m = 2 V. Multiply-
ing this by the frequency sensitivity gives the frequency deviation f d = 50 kHz. From the expression for vm (t),
the modulating signal frequency f m = 15 kHz. Thus, modulation index
fd 50
𝛽= = = 3.33
fm 15
(c) Since the carrier frequency is 100 MHz, we may safely assume that the application is FM radio with a rated
system deviation of 75 kHz. The percentage modulation is then
fd 50
m= × 100% = × 100% = 66.7%
FD 75
 8.2 Basic Concepts of FM and PM 535

8.2.2 Phase Modulation Concepts

In PM, the phase of the carrier at any given time is changed from its unmodulated value 𝜙c by an amount that
depends on the value of the message signal at that time. Recall the expression for the unmodulated carrier given
in Eq. (8.3)
vc (t) = Vc cos[𝜃c (t)]
= Vc cos(2𝜋fc t + 𝜙c )
With the variations in phase imposed in sympathy with the message signal, we can now refer to the instantaneous
phase 𝜙i of the carrier given by
𝜙i = 𝜙c + kp vm (t) (8.14)
where kp is the change in carrier phase per volt of modulating signal, called phase sensitivity and expressed in units
of rad/V. The unmodulated carrier phase 𝜙c is usually set to zero without any loss of generality, since the message
signal is carried in the variation of the carrier phase rather than its absolute magnitude. Then
𝜙i = kp vm (t) (8.15)
Let us now examine the variation of the angle of the carrier in Eq. (8.3) when it is phase modulated by the stair-
case message signal vm (t) in Eq. (8.2). We assume a phase modulator sensitivity kp = 𝜋∕2 rad/V. The unmodulated
phase of the carrier is 𝜙c = −𝜋/2, and Eq. (8.14) gives the instantaneous phase of the phase modulated carrier
⎧ −𝜋∕2 rad, 0 ≤ t < 1 ms
⎪
⎪ 𝜋∕2 rad, 1 ≤ t < 2 ms
⎪ 0 rad, 2 ≤ t < 3 ms
𝜙i = 𝜙c + kp vm (t) = ⎨ (8.16)
⎪ −𝜋 rad, 3 ≤ t < 4 ms
⎪−3𝜋∕2 rad, 4 ≤ t < 5 ms
⎪ −𝜋∕2 rad, 5 ≤ t ≤ 6 ms
⎩
It would pay to check that you agree with the above values. The carrier angle 𝜃 c is given at any time t by the sum
of two terms
𝜃c (t) = 2𝜋fc t + 𝜙i (8.17)
The first term is a component that increases at the constant rate 2𝜋f c rad/s or 6𝜋 rad/ms – since the frequency
f c (= 3 kHz) is not altered in PM. It has the values 0, 6𝜋, 12𝜋, 18𝜋, 24𝜋, 30𝜋, and 36𝜋, at t = 0, 1, 2, 3, 4, 5, and
6 ms, respectively. The second term is the instantaneous phase, which varies according to the modulating signal,
and has the values given in Eq. (8.16). The sum of these two terms gives the phase modulated angle 𝜃 c (t) plotted
against time in Figure 8.2b. Observe that in those intervals where the modulating signal is constant, the phase
Telegram: @Persian_IEEEComSoc

modulated carrier angle increases at the same constant rate (= 2𝜋f c ) as the unmodulated carrier angle. Then at
the instant that vm (t) changes by, say, ΔV volt, the phase modulated angle undergoes a step change equal to kp ΔV.
Herein lies an important difference between FM and PM. In PM, the carrier angle deviates from the unmodulated
rate of change (= 2𝜋f c ) only when the modulating signal is changing, whereas in FM the carrier angle changes at
a different rate than 2𝜋f c whenever the message signal is nonzero.
It should be observed that PM does indirectly produce FM. For example, PM may cause the angle of a carrier
to jump from a lower to a higher value, which is equivalent to the carrier oscillating at a faster rate, i.e. with a
higher frequency. Similarly, a drop in carrier angle is equivalent to a reduction in oscillation frequency. FM and
PM are therefore very closely related. In Figure 8.2b, the phase modulated angle deviates by discrete jumps from
the constant rate of increase that is followed by unmodulated angles. This is because the modulating signal is
staircase and changes in steps. If a smoothly varying (analogue) signal phase modulates a carrier, the deviations of
 536 8 Frequency and Phase Modulation

the rate of change of carrier angle from 2𝜋f c will also be smooth, and FM and PM will differ only in the value and
time of occurrence of the carrier frequency variations. We explore the relationship between FM and PM further
in the next section.
We could define the terms phase swing, phase deviation, and so on, for PM in analogy with the corresponding
definitions for FM. But these terms are rarely used, since it is conventional to treat PM in terms of frequency
variations, making the FM terms applicable to PM. However, the term phase deviation, denoted 𝜙d , merits some
discussion. It is defined as the maximum amount by which the carrier phase – or the carrier angle – deviates from
its unmodulated value. Thus

𝜙d = kp |vm (t)|max (8.18)

Remember that a sinusoidal signal completes one full cycle of oscillation when its angle advances through
2𝜋 rad. Any further advance in angle merely causes the carrier to repeat its cycle. Using the standard convention
of specifying positive angles as anticlockwise rotation from the +x axis direction and negative angles as clockwise
rotation from the same direction, the range of angles that covers a complete cycle is −𝜋 to +𝜋. See Figure 8.3a. Any
phase change by an amount Φ that is outside this range can be shown to be equivalent to some value 𝜙 within this
range, where

𝜙 = Φ + 2n𝜋, n = · · · − 3, −2, −1, 1, 2, 3, · · · (8.19)

That is, 𝜙 is obtained by adding to or subtracting from Φ an integer number of 2𝜋’s in order to place it within
the interval −𝜋 to +𝜋. For example, Figure 8.3b shows that the phase change Φ = 3𝜋/2 rad is equivalent to

𝜙 = Φ − 2𝜋 = 3𝜋∕2 − 2𝜋 = −𝜋∕2 rad

And the phase change Φ = −7𝜋/2 rad is equivalent to

𝜙 = Φ + 4𝜋 = −7𝜋∕2 + 4𝜋 = 𝜋∕2 rad

π/2

Positive angles 0 → π

(a) π or –π 0 x direction

Negative angles 0 → –π

–π/2
Telegram: @Persian_IEEEComSoc

π/2

–7π/2
3π/2
(b) π or –π 0

–π/2

Figure 8.3 (a) Positive and negative angles; (b) examples of equivalent angles: 3𝜋/2 is equivalent to −𝜋/2, and −7𝜋/2 is
equivalent to 𝜋/2.
 8.2 Basic Concepts of FM and PM 537

It therefore follows that the maximum possible value of phase deviation 𝜙d is 𝜋 rad. A value of 𝜙d having absolute
value in excess of 𝜋 rad is equivalent to a smaller value in the interval −𝜋 to +𝜋, according to Eq. (8.19), and
would lead to an error at a PM demodulator that tracks the carrier phase. A large message value (that gives a
phase deviation >𝜋 rad) would be mistaken for a smaller message value that would cause an equivalent phase
deviation between −𝜋 and 𝜋. Note that this problem only applies in direct wideband phase demodulation, which
is never used in analogue communication systems. We define phase modulation factor m as the ratio of actual
phase deviation in a PM implementation to the maximum possible deviation
𝜙
m= d (8.20)
𝜋
Worked Example 8.2

A message signal vm (t) = 5sin(30 × 103 𝜋t) volt is used to phase modulate the carrier vc (t) = 10sin(200 × 106 𝜋t)
volt, causing a phase deviation of 2.5 rad. Determine
(a) The phase sensitivity of the modulator
(b) The phase modulation factor.

(a) The maximum absolute value |vm (t)|max of the message signal is its amplitude V m = 5 V. From Eq. (8.18), phase
sensitivity is
𝜙d 2.5 rad
kp = = = 0.5 rad∕V
|vm (t)|max 5V
(b) Phase modulation factor
𝜙 2.5
m= d = = 0.796
𝜋 𝜋

8.2.3 Relationship Between FM and PM

Let us now obtain a quantitative indication of the relationship between frequency and PMs. We first explore the
type of carrier frequency variation that occurs in PM and then consider phase variations in FM signals.

8.2.3.1 Frequency Variations in PM

Any message signal vm (t) can be approximated by a staircase signal vst (t), as shown in Figure 8.4a. Time is divided
into small intervals Δt, and the curve of vm (t) in each interval is approximated by two line segments. One segment is
horizontal and has a length Δt. The other is vertical and of length Δvm , which gives the change undergone by vm (t)
during the time interval Δt. In the limit when Δt is infinitesimally small the approximation becomes exact – i.e.
Telegram: @Persian_IEEEComSoc

vst (t) and vm (t) become identical.

Consider the result of phase modulating a carrier of frequency f c by the signal vst (t). The variation of the carrier
angle 𝜃 c during one time interval Δt is shown in Figure 8.4b. vst (t) has a constant value until just before the end
of the interval, so the carrier angle increases at the unmodulated rate of 2𝜋f c from point A to point B. When vst (t)
makes a step change of Δvm at the end of the interval, this causes the carrier phase, and hence its angle, to increase
by kp Δvm from point B to point C. The result is that the carrier angle has increased from 𝜃 1 to 𝜃 2 in a time Δt. The
average rate of change of carrier angle is therefore
𝜃2 − 𝜃1 2𝜋fc Δt + kp Δvm
2𝜋fa = =
Δt Δt
Δvm
= 2𝜋fc + kp
Δt
 538 8 Frequency and Phase Modulation

ʋm(t) ʋst(t)

(a)
Δʋm

Δt
t

θc(t)

θ2 C
kpΔʋm
(b) B
2πfcΔt
θ1
A
t
Δt

Figure 8.4 (a) Staircase approximation of an arbitrary message signal; (b) increase in carrier angle during time interval Δt.

where f a is the average carrier frequency during the interval Δt. By making the time interval Δt infinitesimally
small, vst (t) becomes the message signal vm (t), the ratio of Δvm to Δt becomes the derivative or slope of the message
signal, and f a becomes the instantaneous carrier frequency f i . Thus
1 dvm (t)
f i = fc + k (8.21)
2𝜋 p dt
Equation (8.21) is a remarkable result that reveals how PM varies the carrier frequency. Compare it with Eq. (8.4),
repeated below for convenience, which gives the instantaneous frequency of this carrier when frequency modu-
lated by the same message signal vm (t) in a modulator of frequency sensitivity kf

fi = fc + kf vm (t)

PM and FM are therefore related as follows:

Telegram: @Persian_IEEEComSoc

(i) Both PM and FM vary the carrier frequency – but not the carrier amplitude of course. FM’s frequency varia-
tion is achieved directly, as expressed in Eq. (8.4). However, PM’s frequency variation is indirect and occurs
because a (direct) change in the carrier phase causes the carrier’s angle to change at a different rate. In other
words, it alters the carrier’s angular frequency, and hence frequency.
(ii) PM varies the carrier frequency from its unmodulated value f c only at those instants when the message signal
is changing. The carrier frequency is unchanged whenever the message signal has a constant value, say V 1 .
During this time, the phase of the carrier is simply held at a level kp V 1 above its unmodulated value 𝜙c .
FM causes the carrier frequency to deviate from its unmodulated value whenever the message signal has a
nonzero value, irrespective of whether that value is constant or changing.
(iii) In PM, the maximum instantaneous carrier frequency f imax occurs at the instant where the message signal
is changing most rapidly, i.e. at the point where the derivative of the message signal is at a maximum. An
 8.2 Basic Concepts of FM and PM 539

FM carrier has its maximum instantaneous frequency at the instant that the message signal is at a maximum
value. This feature provides a sure way of distinguishing between FM and PM waveforms when the message
signal is a smoothly varying analogue signal. This is discussed further in Section 8.3.2.
(iv) Frequency deviation f d in PM depends on both the amplitude and frequency of the message signal, whereas
in FM it depends only on the message signal amplitude. To see that this is the case, assume a sinusoidal
message signal vm (t) = V m sin(2𝜋f m t) in Eqs. (8.4) and (8.21). It follows that
{
fc + kf Vm sin(2𝜋fm t), FM
fi = (8.22)
fc + kp fm Vm cos(2𝜋fm t), PM
{
fc + kf Vm , FM
fimax =
fc + kp fm Vm , PM
The frequency deviation is therefore
{
kf Vm , FM
fd = fimax − fc = (8.23)
kp fm Vm , PM
which agrees with the previous statement. Thus, given two frequency components of the same amplitude in
a modulating signal, the higher frequency component produces a larger carrier frequency deviation than the
deviation caused by the lower frequency component if PM is used, whereas in FM both frequency components
produce the same frequency deviation. For this reason, FM has a more efficient bandwidth utilisation than
PM in analogue communication.
(v) PM can be obtained using an FM modulator. A block diagram of the arrangement is shown in Figure 8.5. The
operation of the FM modulator results in the signal vo (t) frequency-modulating a carrier of frequency f c . This
produces an FM signal with instantaneous frequency f i given by Eq. (8.4) as

fi = fc + kf vo (t)

Since vo (t) is the output of the differentiator circuit whose input is the message signal vm (t), we have
dvm (t)
vo (t) =
dt
Substituting this identity for vo (t) in the expression for f i , and using a modulator of frequency sensitivity
kf = kp /2𝜋, we have the following result for the instantaneous frequency at the output of the FM modulator
kp dvm (t)
fi = fc +
2𝜋 dt
You will recognise this as Eq. (8.21) – the instantaneous frequency of a PM signal. That is, the message signal
Telegram: @Persian_IEEEComSoc

has phase modulated the carrier, and this was achieved by passing the message signal through a differen-
tiator and then using the differentiator output to frequency modulate a carrier. The differentiator essentially
performs highpass filtering that boosts high frequency components of vm (t) relative to the lower frequency
components. When this filtered signal vo (t) is fed into an FM modulator (whose frequency deviation, you will

ʋo(t)
Message ʋm(t)
Differentiator FM
PM
d modulator
signal signal
dt kf = kp/2π

Figure 8.5 PM generation using FM modulator.

 540 8 Frequency and Phase Modulation

recall, is proportional to modulating signal amplitude), the (boosted) higher frequency components will pro-
duce a larger frequency deviation than the deviation produced by lower frequency components of originally
similar amplitude. This is a PM characteristic, and therefore the overall result is that the carrier has been
phase modulated by vm (t).

8.2.3.2 Phase Variations in FM

We have explored in detail the relationship between FM and PM by looking at the frequency variations in both
signals. An alternative approach that concentrates on phase variations is also very illuminating. We already know
that PM varies the phase of a carrier signal. Let us show that FM also varies the carrier phase, albeit indirectly,
and explore the relationships between the phase variations in both techniques.
Figure 8.6a shows an arbitrary message signal vm (t) approximated by a staircase signal. We want to determine
the instantaneous phase 𝜙i of the carrier signal vc (t) = Vc cos(2𝜋fc t + 𝜙i ) when it is frequency modulated by vm (t).
The angle of the modulated carrier at any time is given by Eq. (8.17), where 𝜙i is the instantaneous carrier phase
that we are after. We already know from Eq. (8.14) that PM gives 𝜙i = 𝜙c + kp vm (t). To determine 𝜙i in the case
of FM, consider how the modulated carrier angle increases in the time from 0 to t. We have divided this duration
into N infinitesimal intervals, numbered from 0 to N−1, each of width Δt. The message signal, approximated by
the staircase signal vst (t), has a constant value during each interval Δt, which leads to a constant instantaneous
frequency of the carrier during that interval. Let us denote the instantaneous frequency of the nth infinitesimal

ʋm(NΔt)

ʋm(nΔt)

ʋm
(a) ʋst
ʋm(2Δt)

ʋm(Δt)

ʋm(0)
t = NΔt
Δt
t
Interval 0 1 2 ………. n ………. N–1

θc(t) etc.
Telegram: @Persian_IEEEComSoc

2πf2Δt
(b) fn = fc + kfʋm(nΔt)
2πf1Δt

2πf0Δt

ϕc
t

Figure 8.6 (a) Message signal v m (t) and staircase approximation v st (t); (b) increase in angle of carrier, of unmodulated
frequency f c , when modulated in frequency by v st (t).
 8.2 Basic Concepts of FM and PM 541

interval as f n , so that the carrier frequency is f 0 during the interval numbered 0, f 1 during the interval numbered
1, and so on. The carrier angle increases by 2𝜋f n Δt inside the nth interval, so that the angle of the carrier after time
t = NΔt is
𝜃c (t) = 𝜙c + 2𝜋f0 Δt + 2𝜋f1 Δt + 2𝜋f2 Δt + · · ·
∑
N−1
= 𝜙c + 2𝜋 fn Δt (8.24)
n=0

Observe in Figure 8.6b that the modulating signal has a constant value vm (nΔt) during the nth interval. Thus,
Eq. (8.4) gives the instantaneous frequency f n in this interval as
fn = fc + kf vm (nΔt)
Substituting this expression in Eq. (8.24) yields
∑
N−1
𝜃c (t) = 𝜙c + 2𝜋 (fc + kf vm (nΔt))Δt
n=0
∑
N−1
= 𝜙c + 2𝜋fc t + 2𝜋kf vm (nΔt)Δt (8.25)
n=0
In the limit Δt → 0, the staircase approximation becomes exact, and the above summation becomes an integration
from 0 to t, so that
t
𝜃c (t) = 2𝜋fc t + 𝜙c + 2𝜋kf vm (t)dt (8.26)
∫0
It follows from Eq. (8.17) that the instantaneous phase of the frequency modulated carrier is
t
𝜙i = 𝜙c + 2𝜋kf vm (t)dt (8.27)
∫0
Now compare Eq. (8.27) with Eq. (8.14), repeated below for convenience, which gives the instantaneous phase of
the same carrier when it is phase modulated by the same message signal vm (t)
𝜙i = 𝜙c + kp vm (t)
Therefore, the relationships between FM and PM stated in terms of carrier phase variation are as follows:
(i) Both PM and FM vary the carrier phase from its unmodulated value 𝜙c , which is usually set to zero. Phase
variation in FM is, however, indirect and results from the fact that a change in the carrier frequency causes
the carrier angle to rise to a level that is different from its unmodulated value, and this difference in angle
can be accounted as a phase change.
(ii) PM changes the carrier phase anytime that the message signal is nonzero, whereas FM changes the carrier
phase at a given instant only if the average of all previous values of the signal is nonzero.
Telegram: @Persian_IEEEComSoc

(iii) The maximum instantaneous carrier phase occurs in PM at the instant that the modulating signal has its
maximum value, whereas in FM it occurs at the instant that the average of all previous values of the signal is
at a maximum.
(iv) Phase deviation 𝜙d in FM depends on both the amplitude and frequency of the message signal, increasing
with amplitude but decreasing with frequency. In PM, on the other hand, phase deviation depends only on
the amplitude of the message signal. It can be shown that this is the case by substituting a sinusoidal message
signal vm (t) = V m cos(2𝜋f m t) in Eqs. (8.14) and (8.27). This yields the result
⎧
⎪𝜙c + kp Vm cos(2𝜋fm t), PM
𝜙i = ⎨ Vm
⎪𝜙c + kf f sin(2𝜋fm t), FM
⎩ m
 542 8 Frequency and Phase Modulation

Integrator PM
Message ʋm(t) t
ʋo(t) FM
Modulator
signal ∫ ʋm(t)dt signal
0
kp = 2πkf

Figure 8.7 FM generation using PM modulator.

and hence
{
kp Vm , PM
𝜙d = Vm (8.28)
kf fm
, FM

Higher frequency components therefore produce smaller phase deviations in FM.

(v) FM can be generated from a PM modulator using the arrangement shown in Figure 8.7. The PM modulator
causes the signal vo (t) to phase modulate a carrier. This produces a PM signal with instantaneous phase
𝜙i = 𝜙c + kp vo (t)
However
t
vo (t) = vm (t)dt
∫0
So that
t t
𝜙i = 𝜙c + kp vm (t)dt = 𝜙c + 2𝜋kf vm (t)dt
∫0 ∫0
where we have used a modulator of phase sensitivity kp = 2𝜋kf . You will no doubt recognise this as
Eq. (8.27) – the instantaneous phase of an FM signal. That is, the overall result of the arrangement in
Figure 8.7 is that the message signal vm (t) has frequency modulated the carrier. FM has been achieved by
passing the message signal through an integrator and then using the integrator output to phase modulate
a carrier. In more practical terms, the integrator performs lowpass filtering that attenuates high-frequency
components of vm (t) relative to the lower-frequency components. When this filtered signal vo (t) is fed into
a PM modulator (whose phase deviation, you will recall, is proportional to modulating signal amplitude),
the (attenuated) higher-frequency components will produce a smaller phase deviation than the deviation
produced by lower-frequency components of (originally) similar amplitude. We showed above that this is a
feature of FM, and therefore the overall result is that the carrier has been frequency modulated by vm (t).

Worked Example 8.3

Telegram: @Persian_IEEEComSoc

Determine
(a) The phase deviation that arises in the FM implementation in Worked Example 8.1.
(b) The frequency deviation experienced by the PM carrier in Worked Example 8.2.

(a) What is required here is the maximum amount 𝜙d by which the phase of the frequency modulated carrier
deviates from the phase 𝜙c of the unmodulated carrier. By Eq. (8.28)
kf Vm fd
𝜙d = = =𝛽 (8.29)
fm fm
This is an important result, which shows that the phase deviation (in rad) of an FM carrier equals its modula-
tion index. Therefore, from Worked Example 8.1, 𝜙d = 3.33 rad.
 8.3 FM and PM Waveforms 543

Note that in this example 𝜙d exceeds 𝜋 rad. There will, however, be no error at the receiver, as discussed earlier
for PM, because an FM demodulator will be used, which only responds to the rate of change of carrier phase
and not the phase magnitude or amount of phase change.
(b) We wish to determine the maximum amount f d by which the frequency of the phase modulated carrier deviates
from the frequency f c of the unmodulated carrier. By Eq. (8.23)
fd = kp Vm fm = 𝜙d fm = 2.5 × 15 kHz = 37.5 kHz
We see that the frequency deviation of a PM carrier is given by the product of the carrier phase deviation and
the message signal frequency.

8.3 FM and PM Waveforms

We studied AM waveforms in the previous chapter and saw that it was very easy to sketch them: you simply draw
a sinusoid with the same number of cycles per second as the carrier but with its amplitude varied to match the
message signal waveform. FM and PM waveforms are not as easy to sketch by hand except for the simple cases
involving staircase message signals. However, armed with the basic concepts of angle modulation discussed above
you can easily distinguish between oscilloscope displays of FM and PM waveforms. You can also readily plot the
FM and PM waveforms for an arbitrary message signal with the aid of a computer.

8.3.1 Sketching Simple Waveforms

When the message signal has a staircase shape then the FM and PM waveforms can be easily sketched, following
the procedure discussed below through a specific example. It should be emphasised that the following approach is
only valid for staircase message waveforms. However, this condition is not as restrictive as may first appear. Many
message signals, including all digital signals (binary, ternary, quaternary, etc.) belong in this category.
Consider, then, the simple staircase message signal vm (t) and carrier vc (t) = V c cos(2𝜋f c t + 𝜙c ) introduced in
Eqs. (8.2) and (8.3). The carrier signal has amplitude V c = 2 V, frequency f c = 3 kHz, and initial phase 𝜙c = −𝜋/2 rad.
We wish to sketch the FM and PM waveforms obtained when vm (t) modulates the frequency and phase of vc (t),
respectively. We assume frequency sensitivity kf = 1 kHz/V and phase sensitivity kp = 𝜋/2 rad/V. Figure 8.8 shows
the waveforms involved. The message signal is sketched at the top in (a). The unmodulated carrier is sketched in
(b) over the duration of the message signal from t = 0 to t = 6 ms. This is easily done by noting that a frequency of
3 kHz means that the sinusoid completes three cycles in each 1 ms interval, and that an initial phase of −𝜋/2 rad
simply converts the cosine sinusoid to a sine.
The FM signal is sketched in (c). To do this, we determine, using Eq. (8.4), the instantaneous frequency f i of
the modulated carrier within each interval over which the message signal has a constant value. The simplification
Telegram: @Persian_IEEEComSoc

made possible by the staircase nature of the message signal is that we can treat the FM signal as having constant
amplitude and phase over all intervals and having a single value of instantaneous frequency in each interval.
During the first 1 ms, vm (t) = 0 and f i = f c = 3 kHz. So, we sketch three cycles of a cosine sinusoid of amplitude 2 V,
and phase −𝜋/2 – the amplitude and phase remain unchanged throughout. In the next 1 ms interval, vm (t) = 2 V
and
fi = fc + kf × 2 V = 3 kHz + 1 kHz∕V × 2 V = 5 kHz
Therefore, we sketch five cycles of the sinusoid. Proceeding in this manner the sketch shown in (c) is completed.
Eq. (8.5) gives a complete list of values of the instantaneous frequency.
The staircase nature of the message signal again greatly simplifies the procedure for sketching the PM waveform
shown in Figure 8.8d. We treat not only the amplitude but also the frequency of the carrier as constant throughout.
 544 8 Frequency and Phase Modulation

2 ʋm(t), V
1
(a) t
–1
–2
ʋc(t), V
2
1
(b) t
–1
–2
ʋfm(t), V
2
1
(c) t
–1
–2
ʋpm(t), V
2
1
(d) t
–1
–2
0 1 2 3 4 5 6 → t (ms)

Figure 8.8 (a) Staircase modulating signal; (b) carrier; (c) frequency modulated carrier; (d) phase modulated carrier.

So we sketch in each 1 ms interval three cycles (because f c = 3 kHz) of a sinusoid of amplitude 2 V, but the phase
𝜙i in each interval is determined by the message signal value during that interval, according to Eq. (8.14). During
the first 1 ms, vm (t) = 0 and 𝜙i = 𝜙c = −𝜋/2 rad. So, we sketch three cycles of a cosine sinusoid of amplitude 2 V
and phase −𝜋/2 rad. In the next 1 ms interval, vm (t) = 2 V and
𝜙i = 𝜙c + kp × 2 V = −𝜋∕2 + 𝜋∕2 rad∕V × 2 V = 𝜋∕2 rad
So we sketch the (cosine) sinusoid with a phase 𝜋/2 rad. Proceeding in this manner we complete the sketch
shown in (d) using the complete list of values of 𝜙i given in Eq. (8.16). You may wish to review much of Section
2.7 if you have any difficulty with sketching sinusoids of different phases.

8.3.2 General Waveform

Let an arbitrary message signal vm (t) modulate the angle of the sinusoidal carrier, given by Eq. (8.3), of amplitude
Telegram: @Persian_IEEEComSoc

V c , frequency f c , and initial phase 𝜙c

vc (t) = Vc cos(2𝜋fc t + 𝜙c )
Based on the concepts discussed earlier, the resulting angle modulated signal is
v(t) = Vc cos(2𝜋fc t + 𝜙i ) (8.30)
where 𝜙i is the instantaneous phase given by Eqs. (8.14) and (8.27). Specifically
⎧ t
⎪𝜙c + 2𝜋kf v (t)dt, FM
𝜙i = ⎨ ∫0 m (8.31)
⎪𝜙c + kp vm (t), PM
⎩
 8.3 FM and PM Waveforms 545

Substituting Eq. (8.31) in Eq. (8.30) yields the following general expressions for the waveforms of a frequency
modulated signal vfm (t) and a phase modulated signal vpm (t)
[ t ]
vfm (t) = Vc cos 2𝜋fc t + 𝜙c + 2𝜋kf vm (t)dt
∫0
vpm (t) = Vc cos[2𝜋fc t + 𝜙c + kp vm (t)] (8.32)

Eq. (8.32) gives a complete specification of FM and PM signals in the time domain. It shows that they have the
same constant amplitude V c as the unmodulated carrier, but a variable rate of completing each cycle, caused by the
dependence of the instantaneous phase on the modulating signal. The unmodulated carrier, on the other hand,
completes each cycle at a constant rate f c (cycles per second).
Given a specification of the carrier and message signals, a reliable hand sketch of vfm (t) and vpm (t) is generally
not possible. However, an accurate waveform can be displayed by using Eq. (8.32) to calculate the values of vfm (t)
and vpm (t) at a sufficient number of time instants, and plotting these points on a graph. We will examine the result
of this procedure applied to four different message signals, namely bipolar, sinusoidal, triangular, and arbitrary.
The same carrier signal is used in the first three examples, with amplitude V c = 2 V, frequency f c = 10 kHz, and
initial phase 𝜙c = 0 rad. The bipolar signal has a staircase waveform. It is used here to demonstrate that Eq. (8.32)
is applicable to all types of message signals, including the simple staircase waveforms discussed earlier. One of our
aims is to show that, although FM and PM both involve carrier frequency variation, the difference between their
waveforms can be spotted by visual inspection.
Figure 8.9 shows the plots for a bipolar message signal vm (t) of duration 2 ms and amplitude 1 V; using modu-
lations of frequency sensitivity kf = 5 kHz/V and phase sensitivity kp = −𝜋/2 rad/V. A plot of vm (t) computed at
a large number N of time instants ranging from t = 0 to 2 ms is shown in (a). The phase modulated signal vpm (t)
is easily calculated at each time instant using Eq. (8.32), and this is plotted in (c) with the unmodulated carrier
plotted in (b) for comparison. Calculating the frequency modulated signal vfm (t) is a little more involved. Specifi-
cally, to determine the value of vfm (t) at the time instant t = 𝜏, one numerically integrates vm (t) from t = 0 to t = 𝜏,
obtaining a value of, say, V(𝜏), and then determines vfm (𝜏) as

vfm (𝜏) = Vc cos[2𝜋fc 𝜏 + 𝜙c + 2𝜋kf V(𝜏)]

The computation is done for all the N time instants in the range from 0 to 2 ms, where N must be large to
make the numerical integration accurate. The signal vfm (t) computed as described here is plotted in Figure 8.9d.
A detailed discussion of numerical integration is outside our scope. But if you have access to MATLAB, you can
perform the entire computation for vfm (t) using the following single line of code

vfm = Vc*cos(2*pi*fc*t + phic + 2*pi*kf*cumtrapz(t, vm)) (8.33)

To use Eq. (8.33), you must first create a vector of time instants t and a vector of corresponding message values
Telegram: @Persian_IEEEComSoc

vm, and assign values to the following variables: carrier amplitude Vc, carrier frequency fc, carrier initial phase
phic, and frequency sensitivity kf. A typical code that would precede Eq. (8.33) could be

N = 4096; t = linspace(0, 2, N)′ ; %N time instants in ms

vm = [ones(N∕2, 1); -ones(N∕2, 1)]; %Values of bipolar signal
Vc = 2; fc = 10; %Carrier amplitude (volt) and freq (kHz)
phic = 0; kf = 5; %Initial phase (rad) and freq sensitivity (kHz∕V) (8.34)

Observe in Figure 8.9 that the unmodulated carrier completes 10 cycles per ms, as expected of a 10 kHz sinusoidal
signal. The difference between the waveforms of the phase modulated signal vpm (t) and the frequency modulated
signal vfm (t) is obvious. The frequency of vpm (t) is the same (f c = 10 kHz) in each interval where vm (t) is constant,
 546 8 Frequency and Phase Modulation

ʋm(t), V
1

(a) t

–1
ʋc(t), V
2

(b) t

–2
ʋpm(t), V
2

(c) t

–2
ʋfm(t), V
2

(d) t

–2
0 1 2 → t (ms)

Figure 8.9 Angle modulation by bipolar message signal. Waveforms of (a) message signal; (b) sinusoidal carrier; (c) PM
signal; (d) FM signal.

whereas the frequency of vfm (t) is different from f c wherever vm (t) is nonzero, which in this case is at all time
instants.
Following the procedure discussed above, angle modulated waveforms are plotted in Figures 8.10–8.12 for differ-
ent message signals. The carrier is the same as in Figure 8.9b, so its plot is omitted. Figure 8.10 shows a sinusoidal
message signal of amplitude V m = 2 V and frequency f m = 1 kHz in (a). The PM waveform is shown in (b), for
phase sensitivity kp = 4 rad/V, which according to Eq. (8.23) gives a frequency deviation of 8 kHz. The FM wave-
form shown in Figure 8.10c was calculated with frequency sensitivity kf = 4 kHz/V, which also yields a frequency
deviation of 8 kHz. These values of kf and kp were chosen specifically to achieve a large and equal frequency devi-
ation in both FM and PM. This gives a pronounced variation in instantaneous frequency that aids our discussion
and allows us to demonstrate that the two waveforms have distinguishing features even when their frequency
deviations are equal. However, the value of kp here is larger than would be used in practice, since it leads to a
phase deviation 𝜙d = 8 rad, and hence a PM factor m larger than unity (m = 𝜙d /𝜋 = 2.55). You may wish to verify
Telegram: @Persian_IEEEComSoc

that, when FM and PM signals are obtained using the same carrier and the same message signal (of frequency f m ),
the condition for their frequency deviations to be equal is given by

kf = kp fm (8.35)

It is interesting to observe how the instantaneous frequency of both the PM and FM waveforms in Figure 8.10
changes smoothly, because the modulating signal is non-staircase. The question then arises of how to identify
which waveform is PM and which is FM. The identification can be done very straightforwardly using any one of
several tests, all based on the discussion in Section 8.2.3.1. The simplest of these tests are as follows:

(i) If the instantaneous frequency f i of the modulated waveform is maximum at the same instants in which
the message signal is maximum then it is FM. Otherwise, it is PM. In Figure 8.10, the message signal is
 8.3 FM and PM Waveforms 547

ʋm(t), V
2

(a) t

–2
ʋpm(t), V
2

(b) t

–2
ʋfm(t), V
2

(c) t

–2
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 → t (ms)

Figure 8.10 Angle modulation by sinusoidal message signal. Waveforms of (a) message signal; (b) PM signal; (c) FM signal.

maximum at t = 0, 1, and 2 ms. The waveform in (b) is not oscillating most rapidly at these instants, making it
PM, whereas the waveform in (c) has a maximum oscillation rate at these instants, making it definitely FM.
(ii) If f i is minimum wherever the modulating signal vm (t) is minimum then the waveform is FM; otherwise, it
is PM. Looking again at Figure 8.10 we see that vm (t) is minimum at t = 0.5 and 1.5 ms. The waveform in
(c) has the lowest rate of oscillation at these instants, making it FM, whereas waveform (b) does not, and is
therefore PM.
(iii) If f i is maximum wherever vm (t) is increasing most rapidly then the waveform is PM; otherwise, it is FM.
Note that in Figure 8.10 waveform (b) has the largest oscillation rate at the time instants t = 0.75 and 1.75 ms,
when the modulating signal is increasing most rapidly.
(iv) Finally, if f i is minimum wherever vm (t) is decreasing most rapidly then the waveform is PM; otherwise, it
is FM. Again, note that waveform (b) in Figure 8.10 has a minimum oscillation rate at t = 0.25 and 1.25 ms
when vm (t) has the largest negative slope.
Telegram: @Persian_IEEEComSoc

Another interesting observation in Figure 8.10 is that both modulated waveforms complete the same number of
cycles as the unmodulated carrier in any 1 ms interval. This is not a coincidence. As a rule, the average frequency
of an FM signal is equal to the unmodulated carrier frequency f c in any time interval over which the modulating
signal has a mean value of zero. A PM signal, on the other hand, has an average frequency equal to f c in any interval
in which the modulating signal has an average rate of change (i.e. mean slope) equal to zero.
Figure 8.11 shows the results for a triangular modulating signal. It is recommended that you verify that each of
the waveform identification tests discussed above is valid in this case. Can you explain why the PM waveform has
the same number of cycles as the unmodulated carrier over the interval t = 0 to 1 ms, whereas the FM waveform
has a larger number of cycles?
Figure 8.12 shows the modulated waveforms for an arbitrary and non-staircase modulating signal obtained with
modulation parameters kf = 8 kHz/V, kp = 𝜋/2 rad/V, f c = 20 kHz, and 𝜙c = 0. You may wish to verify that the
 548 8 Frequency and Phase Modulation

ʋm(t), V
2

(a) 1

0 t
ʋpm(t), V
2

(b) t

–2
ʋfm(t), V
2

(c) t

–2
0 0.25 0.5 0.75 1→ t (ms)

Figure 8.11 Angle modulation by triangular message signal. Waveforms of (a) message signal, (b) PM signal, (c) FM signal.

ʋm(t), V
2

(a) t

–2
ʋpm(t), V
2

(b) t
Telegram: @Persian_IEEEComSoc

–2
ʋfm(t), V
2

(c) t

–2
0 0.25 0.5 0.75 1 → t (ms)

Figure 8.12 Angle modulation by an arbitrary non-staircase message signal. Waveforms of (a) message signal, (b) PM
signal, and (c) FM signal.
 8.4 Spectrum and Power of FM and PM 549

waveform identification tests are again valid. Neither the PM nor FM waveform has the same number of cycles
(= 20) as the unmodulated carrier over the time interval from 0 to 1 ms. Can you explain why?
The carrier frequency in the above modulation examples was selected to be low enough to allow clear illustra-
tions of the cycles, but high enough to be sufficiently larger than the highest significant frequency component f m
of the message signal. The latter condition is important to avoid distortion, as explained in the next section.

8.4 Spectrum and Power of FM and PM

We are now able to discuss the important issues of spectrum and power in angle modulated signals, which leads
naturally to bandwidth considerations. To simplify the discussion, let us assume a sinusoidal modulating signal
vm (t) = V m cos(2𝜋f m t), and a carrier of initial phase 𝜙c = 0. Substituting these in Eq. (8.32) gives
[ ]
kf Vm
vfm (t) = Vc cos 2𝜋fc t + sin(2𝜋fm t)
fm
vpm (t) = Vc cos[2𝜋fc t + kp Vm cos(2𝜋fm t)]
But, from Eq.(8.29), the term kf V m /f m is the FM modulation index 𝛽; and, from Eq. (8.28), the term kp V m is the
PM phase deviation 𝜙d . Therefore
vfm (t) = Vc cos[2𝜋fc t + 𝛽 sin(2𝜋fm t)]
vpm (t) = Vc cos[2𝜋fc t + 𝜙d cos(2𝜋fm t)] (8.36)
The time domain relationships between FM and PM have been discussed at length already. This and the striking
similarity of the two expressions in Eq. (8.36) suggest that FM and PM will also be strongly related in the frequency
domain. Let us explore this further by first assuming that the modulation index 𝛽 and phase deviation 𝜙d are small,
which leads us to what is known as narrowband angle modulation. When the restriction on the size of 𝛽 and 𝜙d is
removed, we have what is termed wideband angle modulation.

8.4.1 Narrowband FM and PM

8.4.1.1 Frequency Components
Let us expand the right-hand side of Eq. (8.36) using the compound angle identity of Eq. B.3 (Appendix B)
vfm (t) = Vc cos(2𝜋fc t) cos[𝛽 sin(2𝜋fm t)]
− Vc sin(2𝜋fc t) sin[𝛽 sin(2𝜋fm t)]
vpm (t) = Vc cos(2𝜋fc t) cos[𝜙d cos(2𝜋fm t)]
Telegram: @Persian_IEEEComSoc

− Vc sin(2𝜋fc t) sin[𝜙d cos(2𝜋fm t)] (8.37)

For sufficiently small angles 𝜃 (in rad), we may use the approximation

cos(𝜃) ≈ 1 ⎫
⎪
sin(𝜃) ≈ 𝜃 ⎬, 𝜃→0 (8.38)
⎪
tan(𝜃) ≈ 𝜃 ⎭
Table 8.1 shows the error involved in this approximation for several small angles from 𝜃 = 0 to 𝜃 = 0.4 rad. The
error increases as 𝜃 increases, and the cut-off point for using the approximations in Eq. (8.38) is usually taken as
 550 8 Frequency and Phase Modulation

Table 8.1 Errors in the approximations cos(𝜃) ≈ 1, sin(𝜃) ≈ 𝜃, and tan(𝜃) ≈ 𝜃, for small values of 𝜃 (rad).

𝜃 (rad) 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4

cos𝜃 1 0.999 0.995 0.989 0.980 0.969 0.955 0.939 0.921
Approximation (x) 1 1 1 1 1 1 1 1 1
100(x−cos 𝜃)
% error = cos 𝜃
0 0.125 0.502 1.136 2.034 3.209 4.675 6.454 8.570
sin𝜃 0 0.050 0.010 0.149 0.199 0.247 0.296 0.343 0.389
Approximation (y) 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
100(y−sin 𝜃)
% error = sin 𝜃
0 0.042 0.167 0.376 0.670 1.049 1.516 2.071 2.717
tan𝜃 0 0.050 0.100 0.151 0.203 0.255 0.309 0.365 0.423
Approximation (z) 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
| 𝜃) |
% Error = | 100(z−tan | 0 0.083 0.334 0.751 1.337 2.092 3.018 4.117 5.391
| tan 𝜃 |

𝜃 = 0.25 rad. At this point the approximation cos(𝜃) ≈ 1 introduces an error of 3.209%, sin(𝜃) ≈ 𝜃 produces a 1.049%
error, and tan(𝜃) ≈ 𝜃 causes an error of 2.092%. It follows that in Eq. (8.37) when the conditions
𝛽 ≤ 0.25, FM
𝜙d ≤ 0.25, PM (8.39)
are satisfied we may use the substitutions
cos[𝛽 sin(2𝜋fm t)] ≈ 1
sin[𝛽 sin(2𝜋fm t)] ≈ 𝛽 sin(2𝜋fm t)
cos[𝜙d cos(2𝜋fm t)] ≈ 1
sin[𝜙d cos(2𝜋fm t)] ≈ 𝜙d cos(2𝜋fm t) (8.40)
to obtain
vfm (t) ≈ Vc cos(2𝜋fc t) − 𝛽Vc sin(2𝜋fc t) sin(2𝜋fm t)
≡ vnbfm (t) (8.41)

vpm (t) ≈ Vc cos(2𝜋fc t) − 𝜙d Vc sin(2𝜋fc t) cos(2𝜋fm t)

≡ vnbpm (t) (8.42)
Using the trigonometric identities of Appendix B to expand the product terms on the right-hand side of
Eqs. (8.41) and (8.42), replacing the sine terms in the resulting expression for vnbpm (t) with cosine according to
Telegram: @Persian_IEEEComSoc

the identity sin(𝜃) = cos(𝜃 − 𝜋/2), and then absorbing any negative sign that precedes a cosine term by using the
identity −cos(𝜃) = cos(𝜃 + 𝜋), we finally obtain
𝛽Vc
vnbfm (t) = Vc cos(2𝜋fc t) + cos[2𝜋(fc − fm )t + 𝜋]
2
𝛽Vc
+ cos[2𝜋(fc + fm )t] (8.43)
2
𝜙 V
vnbpm (t) = Vc cos(2𝜋fc t) + d c cos[2𝜋(fc − fm )t + 𝜋∕2]
2
𝜙d Vc
+ cos[2𝜋(fc + fm )t + 𝜋∕2] (8.44)
2
 8.4 Spectrum and Power of FM and PM 551

The above signals are referred to as narrowband frequency modulation (NBFM) and narrowband phase modu-
lation (NBPM), and result from angle modulation using a small modulation index 𝛽 (for FM) and a small phase
deviation 𝜙d (for PM) as specified by Eq. (8.39). Under this condition, and for a sinusoidal modulating signal, the
resulting angle modulated signal has only three frequency components, namely the carrier at f c , a lower side fre-
quency (LSF) at f c − f m , and an upper side frequency (USF) at f c + f m . You will recall from the previous chapter
that an AM signal vam (t) consists precisely of these same components
vam (t) = Vc cos(2𝜋fc t)
mV c mV c
+ cos[2𝜋(fc − fm )t] + cos[2𝜋(fc + fm )t] (8.45)
2 2
It is worthwhile to take a moment to examine Eqs. (8.43–8.45) in order to have a thorough understanding of
narrowband angle modulation and its relationship with AM.

8.4.1.2 Comparing AM, NBFM, and NBPM

8.4.1.2.1 Amplitude Spectra
Figure 8.13 shows the amplitude spectrum of AM, NBFM, and NBPM. We see that they have the same bandwidth
BAM = BNBFM = BNBPM = 2fm (8.46)
where f m is the frequency of the message signal. The amplitude of side frequencies in each spectrum is
⎧mV c ∕2, AM
⎪
VSF = ⎨𝛽Vc ∕2, NBFM (8.47)
⎪
⎩𝜙d Vc ∕2, NBPM
The goal in AM design is to make modulation factor m close to but not exceed unity. However, the maximum
value of modulation index 𝛽 (for NBFM) and phase deviation 𝜙d (for NBPM) is 0.25. Therefore, the maximum
amplitude of each side frequency is
⎧Vc ∕2, AM
⎪
VSFmax = ⎨Vc ∕8, NBFM (8.48)
⎪
⎩Vc ∕8, NBPM

8.4.1.2.2 Waveforms
The waveforms of AM, NBFM, and NBPM signals obtained with parameter values m = 𝛽 = 𝜙d = 0.25, V c = 1 V,
f c = 40 kHz, and f m = 2 kHz are shown in Figure 8.14. It is interesting to note how very different the AM waveform
vam (t) is from the angle modulated waveforms even though all three waveforms contain the same frequency com-
ponents of the same amplitude. The amplitude of vam (t) varies markedly between (1 ± m)V c , whereas the angle
Telegram: @Persian_IEEEComSoc

modulated waveforms exhibit only a slight amplitude variation. This peculiar behaviour is due to differences in the
phase relationships of the frequency components in each signal, which can be readily understood using phasors.

8.4.1.2.3 Phase Considerations

Equations (8.43)–(8.45) reveal that at the initial time (t = 0) AM has both the LSF and USF in phase with the carrier.
NBFM has its USF in phase with the carrier, but its LSF is 180∘ out of phase. NBPM, on the other hand, has both its
LSF and USF leading the carrier by 90∘ . This information is summarised in the phase spectra of Figure 8.15, which
gives the initial phases (i.e. angle at t = 0) of all frequency components in the modulated signals. The implication
of these phase relationships on the instantaneous amplitude of each signal is very significant. The best way to
illustrate this is by the method of phasors introduced in Chapter 2 (Section 2.7.2).
 552 8 Frequency and Phase Modulation

Aam
Carrier
Vc
m≤1
(a) LSF USF
mVc /2

f
fc – fm fc f c + fm

A nbfm
Vc Carrier
β ≤ 0.25

(b)

βVc /2 LSF USF

f
fc – fm fc f c + fm

A nbpm
Vc Carrier
ϕd ≤ 0.25
(c)

ϕdVc /2 LSF USF

f
fc – fm fc fc + fm

Figure 8.13 Single-sided amplitude spectrum of (a) AM, (b) NBFM, and (c) NBPM.

Figure 8.16 uses the phasor addition of the carrier, LSF, and USF to obtain the AM, NBFM, and NBPM signals in
both amplitude and phase at various time instants. The amplitudes of the carrier, LSF, and USF are, respectively,
denoted V c , V l , and V u . At t = 0, the LSF and USF of the AM signal both have zero phase, so their phasors point
Telegram: @Persian_IEEEComSoc

in the 0∘ direction. In the NBFM signal, the phase of the LSF is 𝜋 rad, and that of the USF is 0 rad, so their phasors
point in the 180∘ and 0∘ directions, respectively. Similarly, the LSF and USF phasors of the NBPM signal both point
in the 90∘ direction at t = 0. The phasor addition in each diagram involves drawing the phasors of the carrier, LSF,
and USF in turn, one phasor starting from the endpoint of the previous one. The resultant signal is the phasor
obtained by joining the start point of the first drawn phasor to the endpoint of the last drawn phasor. Note that
phasors that would lie on top of each other in Figure 8.16 have been displaced to keep each one distinct and
therefore visible in the diagram.
To better understand Figure 8.16, you may wish to consider a useful analogy of three cars – L (for LSF), C (for
carrier), and U (for USF) – starting at the same time (t = 0) and travelling anticlockwise on a circular road at
different speeds. Speed is specified in units of cycles per second, i.e. how many rounds of the circular road a car
 8.4 Spectrum and Power of FM and PM 553

1.25Vc
Vc

AM 0 t

–Vc
–1.25Vc
Vc

NBFM 0 t

–Vc
Vc

NBPM 0 t

–Vc

Figure 8.14 1 ms duration of AM, NBFM and NBPM waveforms. Parameters: m = 𝛽 = 𝜙d = 0.25; f c = 40 kHz; f m = 2 kHz;
V c = 1 V.

completes in 1 s. Assume that car L travels at speed f c − f m , car C at speed f c , and car U at speed f c + f m . Let car
C be chosen as a reference point from which the other two cars are observed. The result will be that C appears
stationary, L appears to be travelling clockwise at speed f m , and U appears to be travelling anticlockwise at the same
speed f m . That is, L is falling behind at the same speed that U is edging ahead. One final point in our analogy is
to imagine that there are three different starting-position scenarios, with car C always starting (at t = 0) from the
starting point, called the 0 rad point. In what we may call the AM scenario, all three cars start together from the
starting point. In the NBFM scenario, car L starts from a half-cycle ahead – at the 𝜋 rad point; and in the NBPM
case, both cars L and U start at the 𝜋/2 rad point, which is one-quarter cycle ahead of the starting point.
Returning then to the three frequency components – LSF, carrier, and USF – of the modulated signals, we choose
the carrier (of frequency f c ) as reference, which allows the carrier to be represented as a zero-phase sinusoid always.
This means that its phasor will always point horizontally to the right. On the other hand, the phasor of the USF
Telegram: @Persian_IEEEComSoc

(of frequency f c + f m ) will change its direction with time as though it were rotating anticlockwise at the rate of f m
cycles per second. The phasor of the LSF (of frequency f c − f m ) will also change its direction with time, only that
the change occurs as though the LSF were rotating clockwise at the rate of f m cycles per second.
It takes the USF a time T m = 1/f m to complete one cycle or 2𝜋 rad – relative to the carrier of course. An arbitrary
angle 𝜃 is therefore traversed in a time
𝜃 𝜃
t= T = (8.49)
2𝜋 m 2𝜋fm
Figure 8.16 shows the phasor diagrams and resultant amplitudes for the three modulated waveforms at intervals
of 1/4T m starting at t = 0. It follows from Eq. (8.49) that the USF advances (anticlockwise) by 𝜋/2 rad relative to
 554 8 Frequency and Phase Modulation

ϕam
180°

(a) 90°

0° f
fc – fm fc fc + fm

ϕnbfm
180°

(b)
90°

0° f
fc – fm fc fc + fm

ϕnbpm
180°

(c)
90°

0° f
fc – fm fc fc + fm

Figure 8.15 Single-sided phase spectra of (a) AM, (b) NBFM, and (c) NBPM.

the carrier during this interval, whereas the LSF retreats (clockwise) by the same angle. By studying Figure 8.16
carefully, you will see why AM has a significant amplitude variation, whereas both NBFM and NBPM have a much
smaller amplitude variation.
In AM, the initial phases of the LSF and USF relative to the carrier are such that they add constructively in
phase with the carrier at t = nT m , and in opposition to the carrier at t = (n + 1/2)T m , for n = 0, 1, 2, …, giving the
AM signal a maximum amplitude V am max and minimum amplitude V am min , respectively, where
Telegram: @Persian_IEEEComSoc

mV c mV c
Vam max = Vc + Vl + Vu = Vc + +
2 2
= Vc (1 + m)
Vam min = Vc − (Vl + Vu )
= Vc (1 − m) (8.50)
Furthermore, the AM signal is always in phase with the carrier because the initial phases of the LSF and USF
are such that their quadrature components are always equal and opposite.
 8.4 Spectrum and Power of FM and PM 555

(a) AM (b) NBFM (c) NBPM

t Vl = Vu = mVc/2 Vl = Vu = βVc/2 Vl = Vu = ϕdVc/2

Vl
V Vu
Vc Vl Vu
0 Vc Vu θ Vc Vl
ʋam = Vc(1 + m)∠0°
ʋnbfm = Vc ∠0° ( )
ʋnbfm = V∠ θ = Vc 1 +ϕd 2 ∠tan–1ϕd
Vu
Vc V Vu
Vl Vu
Tm
θ Vc Vl Vc Vl
4
ʋam = Vc ∠0°
(
ʋnbfm = V∠θ = Vc 1 + β 2
)∠tan –1β ʋnbpm = Vc ∠0°
Vl Vu Vu θ Vc
Vu
Tm Vc V
Vc Vl Vl
2
ʋam = Vc(1 – m)∠0° ʋnbfm = Vc ∠0° ( )
ʋnbpm = V∠–θ = Vc 1 + ϕd 2 ∠–tan–1ϕ
d

θ Vc Vl
Vc Vu
Vu Vl
3T m V
Vl Vc Vu
4
ʋam = Vc ∠0°
(
ʋnbfm = V∠–θ = Vc 1 + β 2
)∠–tan –1β ʋnbpm = Vc ∠0°
Vl
V Vu
Vc Vl Vu
Tm Vc Vu θ Vc Vl
ʋam = Vc(1 + m)∠0°
ʋnbfm = Vc ∠0° ( )
ʋnbfm = V∠θ = Vc 1 + ϕd 2 ∠tan–1ϕ
d

Figure 8.16 Phasor diagrams at intervals of T m /4 for (a) AM, (b) NBFM and (c) NBPM signals. Parameters: Message signal
frequency f m (= 1/T m ); carrier amplitude V c ; AM modulation factor m; FM modulation index 𝛽; and PM phase deviation 𝜙d .

In NBFM and NBPM, on the other hand, whenever one side frequency is in phase with the carrier, the other is
exactly in opposition. The minimum amplitude of the resultant signal is therefore equal to the carrier amplitude
V c . The maximum amplitude occurs when the LSF and USF add constructively at 90∘ to the carrier. Therefore
Telegram: @Persian_IEEEComSoc

Vnbfm min = Vnbpm min = Vc

√
( ) √
𝛽Vc 𝛽Vc 2
Vnbfm max = Vc 2 + + = Vc 1 + 𝛽 2
2 2
√
Vnbpm max = Vc 1 + 𝜙2d (8.51)

Note in Figure 8.16 that the NBFM and NBPM waveforms do not attain maximum or minimum together, but at
different time instants separated by 1/4T m .
 556 8 Frequency and Phase Modulation

We see that the maximum peak-to-peak amplitude variation in NBFM or NBPM is only 3% of the carrier ampli-
tude, when the modulation index or phase deviation is at the maximum value of 0.25 allowed by the narrowband
criterion of Eq. (8.39). This should be compared with AM, which has a peak-to-peak amplitude variation given
by (200 × m)% of the carrier. This is 50% at m = 0.25, and 200% when m has its maximum allowed value m = 1, at
which point the modulated carrier amplitude varies between 0 and 2V c . Thus, the effect of the above differences
in phase relationships is to cause the amplitude variation in an AM waveform to be at least 16 times greater than
in NBFM and NBPM even when the three signals have identical amplitude spectra.

8.4.1.3 Amplitude Variations in NBFM and NBPM

General expressions for the envelope of the three modulated waveforms can be obtained through phasor addition.
Considering Figure 8.17, we see that the resultant phasor representing each modulated waveform at a time t when
each side frequency has rotated through angle 𝜃 (relative to the carrier) is given by
Vam = (Vc + 2x)∠0 = Vc [1 + m cos 𝜃] ∠0
[√ ] [ √ ]
Vnbfm = Vc + (2y) ∠𝜙fm = Vc 1 + 𝛽 sin 𝜃 ∠tan−1 (𝛽 sin 𝜃)
2 2 2 2

[√ ] [ √ ]
Vnbpm = Vc2 + (2z)2 ∠𝜙pm = Vc 1 + 𝜙2d cos2 𝜃 ∠tan−1 (𝜙d cos 𝜃)

Using Eq. (8.49) to express 𝜃 as a function of time, and employing the approximation 𝜃 ≈ tan−1 (𝜃) suggested in
Eq. (8.38), we obtain

F
fm US
V
(a) carrier θ x x mVc
θ x = V cos θ = cos θ
Vc 2

fm
LS
F

Vam

m
m
V nbf y V nbp z
F

F
US

LS

LS U
F fm fm SF fm z
y V fm
ϕnbpm θθ V
Telegram: @Persian_IEEEComSoc

ϕnbfm θ θ
(b) (c)
carrier Vc carrier

βVc ϕdVc
y = V sin θ = sin θ y = V cos θ = cos θ
2 2
ϕnbpm = tan (2y/Vc) = tan–1 (β sin θ)
–1 ϕnbpm = tan (2z/Vc) = tan–1 (ϕd cos θ)
–1

Figure 8.17 LSF, carrier and USF phasors at t = 1/(8f m ) for (a) AM, (b) NBFM and (c) NBPM. The carrier remains fixed (as a
reference) while the LSF and USF phasors rotate at the same speed but in opposite directions as indicated.
 8.4 Spectrum and Power of FM and PM 557

[ ]
Vam = Vc 1 + m cos(2𝜋fm t) ∠0
[ √ ]
Vnbfm = Vc 1 + 𝛽 2 sin2 (2𝜋fm t) ∠𝛽 sin(2𝜋fm t)
[ √ ]
Vnbpm = Vc 1 + 𝜙2d cos2 (2𝜋fm t) ∠𝜙d cos(2𝜋fm t) (8.52)

Equation (8.52) gives the AM, NBFM, and NBPM signals in phasor form. It states, for example, that vnbfm (t)
√
has amplitude Vc 1 + 𝛽 2 sin2 (2𝜋fm t), and phase 𝛽 sin(2𝜋fm t) relative to the carrier signal used as reference in
Figure 8.17. We may therefore write
vam (t) = Vc [1 + m cos(2𝜋fm t)] cos(2𝜋fc t)
[ √ ]
vnbfm (t) = Vc 1 + 𝛽 sin (2𝜋fm t) cos[2𝜋fc t + 𝛽 sin(2𝜋fm t)]
2 2

[ √ ]
vnbpm (t) = Vc 1 + 𝜙d cos (2𝜋fm t) cos[2𝜋fc t + 𝜙d cos(2𝜋fm t)]
2 2 (8.53)

Equation (8.53) reveals that in AM only the carrier amplitude is varied by the modulating signal, with up to 200%
peak-to-peak variation possible. NBFM and NBPM involve up to a maximum of about 3% peak-to-peak amplitude
variation, as well as angle modulation in which the instantaneous phase 𝜙i of the carrier is a function of the
modulating signal given by
{
𝛽 sin(2𝜋fm t), NBFM
𝜙i = (8.54)
𝜙d cos(2𝜋fm t), NBPM
You may wish to verify that Eqs. (8.43–8.45) are equivalent to Eq. (8.53) by evaluating them at selected time
instants. However, Eq. (8.53) explicitly gives the envelope and phase of the modulated waveforms at any given
instant. The AM signal conveys information in its envelope, whereas NBFM and NBPM convey information in the
variation of their angles beyond the linear increase of unmodulated carriers. Note that the variation in NBFM and
NBPM amplitudes is due to the error introduced by the approximations of Eq. (8.40).

8.4.2 Wideband FM and PM

When the modulation index 𝛽 (in FM) and phase deviation 𝜙d (in PM) exceed about 0.25, the narrowband angle
modulation approximations are no longer valid. To determine the bandwidth requirements of a carrier that is angle
modulated by a sinusoidal message signal, we must know the frequency, amplitude, and phase of each sinusoidal
component contained in the angle modulated signals vfm (t) and vpm (t) in Eq. (8.36). It turns out by Fourier analysis
(see Question 8.9) that vfm (t) and vpm (t) can be written as a sum of sinusoids as follows
Telegram: @Persian_IEEEComSoc

vfm (t) = Vc Jo (𝛽) cos(2𝜋fc t)

+ Vc J1 (𝛽){cos[2𝜋(fc + fm )t] + cos[2𝜋(fc − fm )t + 𝜋]}
+ Vc J2 (𝛽){cos[2𝜋(fc + 2fm )t] + cos[2𝜋(fc − 2fm )t]}
+ Vc J3 (𝛽){cos[2𝜋(fc + 3fm )t] + cos[2𝜋(fc − 3fm )t + 𝜋]}
+ Vc J4 (𝛽){cos[2𝜋(fc + 4fm )t] + cos[2𝜋(fc − 4fm )t]}
+ Vc J5 (𝛽){cos[2𝜋(fc + 5fm )t] + cos[2𝜋(fc − 5fm )t + 𝜋]}
+··· (8.55)
 558 8 Frequency and Phase Modulation

vpm (t) = Vc Jo (𝜙d ) cos(2𝜋fc t)

+ Vc J1 (𝜙d ){cos[2𝜋(fc + fm )t + 𝜋∕2] + cos[2𝜋(fc − fm )t + 𝜋∕2]}
+ Vc J2 (𝜙d ){cos[2𝜋(fc + 2fm )t + 𝜋] + cos[2𝜋(fc − 2fm )t + 𝜋]}
+ Vc J3 (𝜙d ){cos[2𝜋(fc + 3fm )t − 𝜋∕2] + cos[2𝜋(fc − 3fm )t − 𝜋∕2]}
+ Vc J4 (𝜙d ){cos[2𝜋(fc + 4fm )t] + cos[2𝜋(fc − 4fm )t]}
+ Vc J5 (𝜙d ){cos[2𝜋(fc + 5fm )t + 𝜋∕2] + cos[2𝜋(fc − 5fm )t + 𝜋∕2]}
+··· (8.56)
Equations (8.55) and (8.56) are applicable to both narrowband (𝛽 ≤ 0.25, 𝜙d ≤ 0.25) and wideband angle modu-
lation (𝛽 > 0.25, 𝜙d > 0.25). They provide a complete insight into the spectrum of FM and PM signals involving a
sinusoidal modulating signal. It is worthwhile to examine in some detail their implications.
The expressions for vfm (t) and vpm (t) reveal that FM and PM have the same amplitude spectrum, with phase
deviation 𝜙d playing the same role in PM that modulation index 𝛽 does in FM. If PM and FM are implemented
with 𝜙d = 𝛽, and with the same values of carrier amplitude V c and modulating frequency f m , then their amplitude
spectra are identical.
However, PM differs from FM in its phase spectrum, which accounts for the waveform differences discussed
earlier. Ignore for the moment the sign of the amplitude of each frequency component – this will introduce a
phase shift of 𝜋 rad if negative. We see from Eq. (8.55) that in FM, the USFs, namely f c + f m , f c + 2f m , … and the
even-numbered LSFs, e.g. f c − 2f m , all have zero initial phase, and only the odd-numbered LSFs, differ in phase by
𝜋 rad. In PM, Eq. (8.56) shows that each USF and the corresponding LSF (e.g. f c + 2f m and f c − 2f m ) have the same
phase, which increases in steps of 𝜋/2 rad starting from zero phase at the carrier. That is, the phase of the nth pair
of side frequencies is n𝜋/2. For example, the phase of the fourth pair of side frequencies is 4 × 𝜋/2 = 2𝜋 ≡ 0 rad,
and the phase of the third pair of side frequencies is 3 × 𝜋/2 = 3𝜋/2 ≡ −𝜋/2 rad.
In view of the above similarities, the following discussion is focussed entirely on FM. The results on spectrum,
bandwidth, and power are applicable to PM, with phase deviation 𝜙d replacing modulation index 𝛽. Recall, how-
ever, that, to avoid distortion in phase demodulators that track absolute phase, 𝜙d is limited to values not exceeding
𝜋 rad, whereas 𝛽 has no theoretical restriction on its values, except that transmission bandwidth requirement
increases as 𝛽 is increased. Phase spectrum is not discussed further, but this can be easily obtained where desired,
by following the above comments.

8.4.2.1 Spectrum
It follows from Eq. (8.55) that an FM signal vfm (t) contains the carrier frequency f c and an infinite set of side
frequencies, which occur in pairs on either side of f c . The spectral components are spaced apart by the modulating
signal frequency f m , and the nth pair of side frequencies consists of the nth LSF at f c − nf m and the nth USF at
Telegram: @Persian_IEEEComSoc

f c + nf m , both of which have the same amplitude. Figure 8.18 gives an example of an FM spectrum, which should
be compared with the spectrum in AM. AM contains only the carrier and the first pair of side frequencies, whereas
FM contains an infinite set of side frequencies in addition to the carrier.
So, does FM have infinite bandwidth? It will become obvious in the following discussion that the amplitude of
the side frequencies f c ± nf m at sufficiently high n is negligible, so that FM bandwidth is indeed finite. In fact,
an FM signal can be realised with little distortion if only the first few side frequencies are retained. For example,
Figure 8.19 shows the synthesis of an FM signal, of modulation index 𝛽 = 3, using only the carrier and the first
few side frequencies. Observe that the distortion is significant when only the first two pairs of side frequencies are
included, but negligible when up to the first five pairs are retained. The number of side frequencies required to
attain negligible distortion in the synthesised waveform increases as 𝛽 is increased.
 8.4 Spectrum and Power of FM and PM 559

A fm 2nd LSF 2nd USF

1st LSF 1st USF

3rd LSF 3rd USF
carrier

4th LSF 4th USF

f
fc – 4fm fc – 3fm fc – 2fm fc – fm fc fc + fm fc + 2fm fc + 3fm fc + 4fm

Figure 8.18 Example of single-sided amplitude spectrum of FM signal.

(a)

FM signal Synthesised signal

(b)

Figure 8.19 Synthesis of FM waveform using carrier plus (a) first two pairs of side frequencies, and (b) first five pairs of side
frequencies. FM waveform parameters: 𝛽 = 3.0, f c = 10 kHz, f m = 1 kHz.
Telegram: @Persian_IEEEComSoc

The amplitude of the nth pair of side frequencies is given by the unmodulated carrier amplitude V c scaled by
the factor J n (𝛽), which is the nth order Bessel function of the first kind evaluated with argument 𝛽. That is, an FM
signal generated with modulation index 𝛽 will contain several frequency components, namely
(i) The unmodulated carrier frequency f c , which may be viewed as the 0th (zeroth) side frequency (f c ± 0f m ), of
amplitude V c J 0 (𝛽).
(ii) The first pair of side frequencies (f c + f m and f c − f m ) of amplitude V c J 1 (𝛽).
(iii) The second pair of side frequencies (f c + 2f m and f c − 2f m ) of amplitude V c J 2 (𝛽).
(iv) And so on.
 560 8 Frequency and Phase Modulation

1
J0(β)

J1(β)
0.5 J2(β)
J3(β)
J4(β) J (β)
5 J6(β) J (β) J (β)
7 8 J9(β) J10(β)

0 β

–0.4
0 5 10 15

β = 2.4048 β = 5.5201 β = 8.6537 β = 11.7915 β = 14.9309

Figure 8.20 Graph of Bessel functions Jn (𝛽) versus 𝛽 for various values of n.

It is therefore important to understand the characteristic of the Bessel function. Figure 8.20 shows a plot of
J n (𝛽) as a function of 𝛽 for various integer values of n. To understand this diagram, assume a normalised carrier
of unit amplitude (V c = 1). Then the curve labelled J 0 (𝛽) gives the amplitude of the carrier component (0th side
frequency) in the FM signal as a function of the modulation index 𝛽. If you follow this curve through the plot, you
will observe that J 0 (𝛽) = 0 at modulation index 𝛽 = 2.4048, 5.5201, 8.6537, 11.7915, 14.9309, etc. At these values
of modulation index the FM spectrum does not contain a component at the unmodulated carrier frequency f c .
Energy in the unmodulated carrier has been entirely distributed to the side frequencies. We will have more to say
about this later. Note also that J 0 (𝛽) is negative for certain values of modulation index, e.g. for 𝛽 between 2.4048
and 5.5201. The effect of a negative value of J n (𝛽) is that the phase of the nth pair of side frequencies is advanced by
𝜋 rad from that indicated in Eq. (8.55). Therefore, when dealing with the amplitude spectrum, only the absolute
value of J n (𝛽) is considered, and its sign is ignored since that only affects the phase spectrum.
Considering the other J n (𝛽) curves in Figure 8.20, a trend can be observed that, as n increases, the curves remain
near zero until a larger value of 𝛽 is reached. For example, the value of J 4 (𝛽) is negligible until 𝛽 has increased to
about 1.1, whereas J 10 (𝛽) is negligible up to 𝛽 ≈ 5.5. This is an important characteristic of J n (𝛽), which signifies that
the higher-side frequencies (n large) are insignificant in the FM spectrum when the modulation index is low. As
Telegram: @Persian_IEEEComSoc

modulation index increases, more side frequencies become significant. In other words, FM bandwidth increases
with modulation index 𝛽. In fact, Figure 8.20 shows that when 𝛽 ≤ 0.25 then only J 0 (𝛽) and J 1 (𝛽) have significant
values. This means that only the carrier and the first pair of side frequencies f c ± f m are noticeably present in the
spectrum. You should recognise this as the special case of NBFM discussed earlier.
Values of J n (𝛽) can be more easily read from tables such as Table 8.2, which is given accurate to four decimal
places, although an accuracy of two decimal places is sufficient for practical calculations.

8.4.2.2 Power
Equation (8.30) gives the general expression for an FM signal, and shows that it is a sinusoidal signal of constant
amplitude V c but variable or modulated instantaneous phase. Bearing in mind that the power in a sinusoidal signal
 Table 8.2 Bessel function to four decimal places.

𝜷 J0 (𝜷) J1 (𝜷) J2 (𝜷) J3 (𝜷) J4 (𝜷) J5 (𝜷) J6 (𝜷) J7 (𝜷) J8 (𝜷) J9 (𝜷) J10 (𝜷) J11 (𝜷) J12 (𝜷) J13 (𝜷) J14 (𝜷) J15 (𝜷) J16 (𝜷)

0 1.0000 — — — — — — — — — — — — — — — —
0.10 0.9975 0.0499 0.0012 — — — — — — — — — — — — — —
0.20 0.9900 0.0995 0.0050 0.0002 — — — — — — — — — — — — —
0.25 0.9844 0.1240 0.0078 0.0003 — — — — — — — — — — — — —
0.50 0.9385 0.2423 0.0306 0.0026 0.0002 — — — — — — — — — — — —
1.0 0.7652 0.4401 0.1149 0.0196 0.0025 0.0002 — — — — — — — — — — —
1.5 0.5118 0.5579 0.2321 0.0610 0.0118 0.0018 0.0002 — — — — — — — — — —
2.0 0.2239 0.5767 0.3528 0.1289 0.0340 0.0070 0.0012 0.0002 — — — — — — — — —
2.4048 — 0.5192 0.4318 0.1990 0.0647 0.0164 0.0034 0.0006 0.0001 — — — — — — — —
3.0 −0.2601 0.3391 0.4861 0.3091 0.1320 0.0430 0.0114 0.0025 0.0005 0.0001 — — — — — — —
4.0 −0.3971 −0.0660 0.3641 0.4302 0.2811 0.1321 0.0491 0.0152 0.0040 0.0009 0.0002 — — — — — —
5.0 −0.1776 −0.3276 0.0466 0.3648 0.3912 0.2611 0.1310 0.0534 0.0184 0.0055 0.0015 0.0004 0.0001 — — — —
5.5201 — −0.3403 −0.1233 0.2509 0.3960 0.3230 0.1891 0.0881 0.0344 0.0116 0.0035 0.0009 0.0002 — — — —
6.0 0.1506 −0.2767 −0.2429 0.1148 0.3576 0.3621 0.2458 0.1296 0.0565 0.0212 0.0070 0.0020 0.0005 0.0001 — — —
7.0 0.3001 −0.0047 −0.3014 −0.1676 0.1578 0.3479 0.3392 0.2336 0.1280 0.0589 0.0235 0.0083 0.0027 0.0008 0.0002 0.0001 —
8.0 0.1717 0.2346 −0.1130 −0.2911 −0.1054 0.1858 0.3376 0.3206 0.2235 0.1263 0.0608 0.0256 0.0096 0.0033 0.0010 0.0003 0.0001
9.0 −0.0903 0.2453 0.1448 −0.1809 −0.2655 −0.0550 0.2043 0.3275 0.3051 0.2149 0.1247 0.0622 0.0274 0.0108 0.0039 0.0013 0.0004
10 −0.2459 0.0435 0.2546 0.0584 −0.2196 −0.2341 −0.0145 0.2167 0.3179 0.2919 0.2075 0.1231 0.0634 0.0290 0.0120 0.0045 0.0016
11 −0.1712 −0.1768 0.1390 0.2273 −0.0150 −0.2383 −0.2016 0.0184 0.2250 0.3089 0.2804 0.2010 0.1216 0.0643 0.0304 0.0130 0.0051
12 0.0477 −0.2234 −0.0849 0.1951 0.1825 −0.0735 −0.2437 −0.1703 0.0451 0.2304 0.3005 0.2704 0.1953 0.1201 0.0650 0.0316 0.0140
13 0.2069 −0.0703 −0.2177 0.0033 0.2193 0.1316 −0.1180 −0.2406 −0.1410 0.0670 0.2338 0.2927 0.2615 0.1901 0.1188 0.0656 0.0327
14 0.1711 0.1334 −0.1520 −0.1768 0.0762 0.2204 0.0812 −0.1508 −0.2320 −0.1143 0.0850 0.2357 0.2855 0.2536 0.1855 0.1174 0.0661
15 −0.0142 0.2051 0.0416 −0.1940 −0.1192 0.1305 0.2061 0.0345 −0.1740 −0.2200 −0.0901 0.1000 0.2367 0.2787 0.2464 0.1813 0.1162
16 −0.1749 0.0904 0.1862 −0.0438 −0.2026 −0.0575 0.1667 0.1825 −0.0070 −0.1895 −0.2062 −0.0682 0.1124 0.2368 0.2724 0.2399 0.1775
17 −0.1699 −0.0977 0.1584 0.1349 −0.1107 −0.1870 0.0007 0.1875 0.1537 −0.0429 −0.1991 −0.1914 −0.0486 0.1228 0.2364 0.2666 0.2340
18 −0.0134 −0.1880 −0.0075 0.1863 0.0696 −0.1554 −0.1560 0.0514 0.1959 0.1228 −0.0732 −0.2041 −0.1762 −0.0309 0.1316 0.2356 0.2611
19 0.1466 −0.1057 −0.1578 0.0725 0.1806 0.0036 −0.1788 −0.1165 0.0929 0.1947 0.0916 −0.0984 −0.2055 −0.1612 −0.0151 0.1389 0.2345
20 0.1670 0.0668 −0.1603 −0.0989 0.1307 0.1512 −0.0551 −0.1842 −0.0739 0.1251 0.1865 0.0614 −0.1190 −0.2041 −0.1464 −0.0008 0.1452
21 0.0366 0.1711 −0.0203 −0.1750 −0.0297 0.1637 0.1076 −0.1022 −0.1757 −0.0318 0.1485 0.1732 0.0329 −0.1356 −0.2008 −0.1321 0.0120
22 −0.1207 0.1172 0.1313 −0.0933 −0.1568 0.0363 0.1733 0.0582 −0.1362 −0.1573 0.0075 0.1641 0.1566 0.0067 −0.1487 −0.1959 −0.1185
Telegram: @Persian_IEEEComSoc

23 −0.1624 −0.0395 0.1590 0.0672 −0.1415 −0.1164 0.0909 0.1638 0.0088 −0.1576 −0.1322 0.0427 0.1730 0.1379 −0.0172 −0.1588 −0.1899
24 −0.0562 −0.1540 0.0434 0.1613 −0.0031 −0.1623 −0.0645 0.1300 0.1404 −0.0364 −0.1677 −0.1033 0.0730 0.1763 0.1180 −0.0386 −0.1663
25 0.0963 −0.1254 −0.1063 0.1083 0.1323 −0.0660 −0.1587 −0.0102 0.1530 0.1081 −0.0752 −0.1682 −0.0729 0.0983 0.1751 0.0978 −0.0577
(Continued)
 Table 8.2 (Continued)

𝜷 J17 (𝜷) J18 (𝜷) J19 (𝜷) J20 (𝜷) J21 (𝜷) J22 (𝜷) J23 (𝜷) J24 (𝜷) J25 (𝜷) J26 (𝜷) J27 (𝜷) J28 (𝜷) J29 (𝜷) J30 (𝜷)

0 — — — — — — — — — — — — — —
0.10 — — — — — — — — — — — — — —
0.20 — — — — — — — — — — — — — —
0.25 — — — — — — — — — — — — — —
0.50 — — — — — — — — — — — — — —
1.0 — — — — — — — — — — — — — —
1.5 — — — — — — — — — — — — — —
2.0 — — — — — — — — — — — — — —
2.4048 — — — — — — — — — — — — — —
3.0 — — — — — — — — — — — — — —
4.0 — — — — — — — — — — — — — —
5.0 — — — — — — — — — — — — — —
5.5201 — — — — — — — — — — — — — —
6.0 — — — — — — — — — — — — — —
7.0 — — — — — — — — — — — — — —
8.0 — — — — — — — — — — — — — —
9.0 0.0001 — — — — — — — — — — — — —
10 0.0005 0.0002 — — — — — — — — — — — —
11 0.0019 0.0006 0.0002 0.0001 — — — — — — — — — —
12 0.0057 0.0022 0.0008 0.0003 0.0001 — — — — — — — — —
13 0.0149 0.0063 0.0025 0.0009 0.0003 0.0001 — — — — — — — —
14 0.0337 0.0158 0.0068 0.0028 0.0010 0.0004 0.0001 — — — — — — —
15 0.0665 0.0346 0.0166 0.0074 0.0031 0.0012 0.0004 0.0002 0.0001 — — — — —
16 0.1150 0.0668 0.0354 0.0173 0.0079 0.0034 0.0013 0.0005 0.0002 0.0001 — — — —
17 0.1739 0.1138 0.0671 0.0362 0.0180 0.0084 0.0037 0.0015 0.0006 0.0002 0.0001 — — —
18 0.2286 0.1706 0.1127 0.0673 0.0369 0.0187 0.0089 0.0039 0.0017 0.0007 0.0003 0.0001 — —
19 0.2559 0.2235 0.1676 0.1116 0.0675 0.0375 0.0193 0.0093 0.0042 0.0018 0.0007 0.0003 0.0001 —
20 0.2331 0.2511 0.2189 0.1647 0.1106 0.0676 0.0380 0.0199 0.0098 0.0045 0.0020 0.0008 0.0003 0.0001
21 0.1505 0.2316 0.2465 0.2145 0.1621 0.1097 0.0677 0.0386 0.0205 0.0102 0.0048 0.0021 0.0009 0.0004
22 0.0236 0.1549 0.2299 0.2422 0.2105 0.1596 0.1087 0.0677 0.0391 0.0210 0.0106 0.0051 0.0023 0.0010
Telegram: @Persian_IEEEComSoc

23 −0.1055 0.0340 0.1587 0.2282 0.2381 0.2067 0.1573 0.1078 0.0678 0.0395 0.0215 0.0110 0.0054 0.0025
24 −0.1831 −0.0931 0.0435 0.1619 0.2264 0.2343 0.2031 0.1550 0.1070 0.0678 0.0399 0.0220 0.0114 0.0056
25 −0.1717 −0.1758 −0.0814 0.0520 0.1646 0.2246 0.2306 0.1998 0.1529 0.1061 0.0678 0.0403 0.0225 0.0118
 8.4 Spectrum and Power of FM and PM 563

depends only on its amplitude, and is independent of frequency and phase, it follows that power in an FM signal
equals the power in the unmodulated carrier. Working with normalised power, we may write
1 2
Pfm = V (8.57)
2 c
Total power in an FM signal can also be obtained by summing the power in all spectral components in the FM
spectrum. Thus
1
Pfm = Vc2 J02 (𝛽) + Vc2 J12 (𝛽) + Vc2 J22 (𝛽) + Vc2 J32 (𝛽) + · · ·
2 [ ]
1
= Vc2 J02 (𝛽) + J12 (𝛽) + J22 (𝛽) + J32 (𝛽) + · · · (8.58)
2
Obviously, the power in the side frequencies is obtained only at the expense of the carrier power. Equating the two
expressions for Pfm in Eqs. (8.57) and (8.58), we obtain the mathematical identity
∑
n=∞
J02 (𝛽) + 2 Jn2 (𝛽) = 1 (8.59)
n=1

It follows that the fraction of power r N contained in the carrier frequency and the first N pairs of side frequencies
is given by
∑
n=N
rN = J02 (𝛽) + 2 Jn2 (𝛽) (8.60)
n=1

8.4.2.3 Bandwidth
Two different definitions of FM bandwidth are in common use, namely Carson’s bandwidth and 1% bandwidth.
Carson’s bandwidth is an empirical definition, which gives the width of the band of frequencies centred at f c that
contains at least 98% of the power in the FM signal. The 1% bandwidth is defined as the separation between the
pair of side frequencies of order nmax beyond which there is no spectral component with amplitude up to 1% of the
unmodulated carrier amplitude V c .
In Figure 8.21, we employ Eq. (8.60) to show, at various integer values of modulation index 𝛽, the fraction of
power r N carried by spectral components that include f c and N pairs of side frequencies. It can be observed that
with N = 𝛽 + 1 the minimum fraction of power is r N = 0.9844. With N = 𝛽, r N = 0.9590 minimum (at 𝛽 = 6). With
N = 𝛽 + 2, r N = 0.9905 (at 𝛽 = 361, a point not reached on the plot). Thus, N = 𝛽 + 1 is the minimum number of
pairs of side frequencies that, along with f c , account for at least 98% of the total FM power. Therefore, Carson’s
bandwidth BC of an FM signal is given by the frequency interval from the (𝛽 + 1)th LSF to the (𝛽 + 1)th USF. Since
the spacing of the spectral components is f m , it follows that
BC = 2(𝛽 + 1)fm
Telegram: @Persian_IEEEComSoc

= 2(fd + fm ) (8.61)
where we have used the definition of modulation index given in Eq. (8.11). Note that, in the limit 𝛽 < < 1, Eq. (8.61)
yields BC = 2f m . Thus, the definition of Carson’s bandwidth is consistent with our earlier discussion of NBFM,
which also has a bandwidth of 2f m .
The 1% bandwidth BP is given by the width of the frequency interval (centred on the carrier frequency) from the
(nmax )th LSF to the (nmax )th USF beyond which the amplitude of every spectral component is less than 1% of the
unmodulated carrier amplitude V c . Thus
BP = 2nmax fm (8.62)
The value of nmax is usually read from a table of Bessel functions such as Table 8.2 by looking along the row
specified by the modulation index 𝛽 for the furthest column (or side frequency n) with the smallest value ≥0.01.
 564 8 Frequency and Phase Modulation

100

N=β+2
99
N=β+1

98
rN(%)

N=β
97

96

95.5
0 10 20 30 40 50 60 70 80 90 100
Modulation index, β

Figure 8.21 Fraction r N (expressed as a percentage) of power in FM signal carried by spectral components at the
unmodulated carrier frequency f c and N pairs of side frequencies, as a function of integer values of modulation index 𝛽.

For example, if 𝛽 = 5.0 we look in Table 8.2 along the row 𝛽 = 5.0 and find that the highest column with the
smallest entry ≥0.01 is the J 8 (𝛽) column. Therefore, nmax = 8 in this case and BP = 16f m , whereas (for comparison)
BC = 12f m .
At practical values of modulation index 𝛽 ≤ 100, BP accounts for at least 99.97% of the total FM power, compared
to a minimum of 98.44% in Carson’s bandwidth BC . Figure 8.22 shows the normalised (i.e. f m = 1 Hz) Carson’s and
1% bandwidths as functions of modulation index 𝛽. It can be observed that BP exceeds BC for 𝛽 > 0.5. However, as
𝛽 → ∞, the two bandwidths become equal (at 𝛽 ≈ 78 474), and eventually BC exceeds BP , since the carrier power is
increasingly distributed in smaller fractions to a larger number of side frequencies.
One disadvantage of the 1% bandwidth definition is that, unlike Carson’s bandwidth, it is not equal to the NBFM
bandwidth of 2f m in the limit 𝛽 → 0. Rather, BP = 0 for 𝛽 ≤ 0.02.
The above discussion of FM bandwidth assumes a sinusoidal modulating signal. However, the bandwidth defi-
nitions in Eq. (8.61) and Eq. (8.62) can be applied to the more general case in which the modulating signal consists
of frequencies up to a maximum f m . To do this, the modulation index 𝛽 is replaced by the deviation ratio D defined
in Eq. (8.13).
Telegram: @Persian_IEEEComSoc

Worked Example 8.4

An audio signal that contains frequencies in the range of 30 Hz to 15 kHz is used to frequency modulate a
100 MHz carrier causing a frequency deviation of 45 kHz. Determine the Carson’s and 1% transmission band-
widths.

This is an example of a nonsinusoidal modulating signal, in which deviation ratio D plays the role that modula-
tion index 𝛽 plays when the modulating signal is sinusoidal. The maximum frequency component of the message
signal is f m = 15 kHz and the frequency deviation f d = 45 kHz, giving

D = fd ∕fm = 3.0
 8.4 Spectrum and Power of FM and PM 565

215
200

180
160
Normalised bandwidth

140
dth
120 d wi
ban dth
1% d wi
100
’s ban
80 rs on
Ca
60
40
20
0
0 10 20 30 40 50 60 70 80 90 100
Modulation index, β

Figure 8.22 Normalised bandwidth (f m = 1).

Thus, Carson’s bandwidth

BC = 2(D + 1)fm = 2(4)(15) = 120 kHz
To determine the 1% bandwidth BP , we look in Table 8.2 along row 𝛽 = 3.0 (since D = 3.0), where we find that
all entries beyond J 6 (𝛽) are less than 0.01. Thus, nmax = 6, and
BP = 2nmax fm = 2(6)(15) = 180 kHz
Observe that BP > BC as discussed earlier.

Worked Example 8.5

A sinusoidal signal of frequency 15 kHz modulates the frequency of a 10 V 100 MHz carrier, causing a frequency
deviation of 75 kHz.
(a) Sketch the amplitude spectrum of the FM signal, including all spectral components of amplitude larger
than 1% of the amplitude of the unmodulated carrier frequency.
Telegram: @Persian_IEEEComSoc

(b) Determine the fraction of the total power contained in the frequency band 99.93–100.07 MHz.

(a) Modulation index 𝛽 = fd ∕fm = 75∕15 = 5.0, f c = 100 MHz, and V c = 10 V.

Using row 𝛽 = 5.0 in Table 8.2 we obtain the amplitudes of the frequency components in the FM spectrum as
follows.
Carrier frequency f c = 100 MHz, of amplitude V c |J 0 | = 10(0.1776) = 1.78 V.
First pair of side frequencies f c ± f m = 99 985 and 100 015 kHz, of amplitude V c |J 1 | = 10(0.3276) = 3.28 V.
Second pair of side frequencies f c ± 2f m = 99 970 and 100 030 kHz, of amplitude V c |J 2 | = 10(0.0466) = 0.47 V.
Third pair of side frequencies f c ± 3f m = 99 955 and 100 045 kHz, of amplitude V c |J 3 | = 10(0.3648) = 3.65 V.
Fourth pair of side frequencies f c ± 4f m = 99 940 and 100 060 kHz, of amplitude V c |J 4 | = 10(0.3912) = 3.91 V.
Fifth pair of side frequencies f c ± 5f m = 99 925 and 100 075 kHz, of amplitude V c |J 5 | = 10(0.2611) = 2.61 V.
 566 8 Frequency and Phase Modulation

Afm, volt 3.91 3.91

3.65
4th LSF 4th USF
3.28

2.61

carrier
1.78
1.31

0.53 0.47 0.47

0.18 f, MHz
99.88 99.91 99.94 99.97 100 100.03 100.06 100.09 100.12

Band 99.93 to 100.07 MHz

Figure 8.23 Single-sided amplitude spectrum of FM signal in Worked Example 8.5.

Sixth pair of side frequencies f c ± 6f m = 99 910 and 100 090 kHz, of amplitude V c |J 6 | = 10(0.1310) = 1.31 V.
Seventh pair of side frequencies f c ± 7f m = 99 895 and 100 105 kHz, of amplitude V c |J 7 | = 10(0.0534) = 0.53 V.
Eighth pair of side frequencies f c ± 8f m = 99 880 and 100 120 kHz, of amplitude V c |J 8 | = 10(0.0184) = 0.18 V.
The ninth and higher side frequencies are ignored since they correspond to table entries <0.01 and there-
fore have amplitudes of less than 1% of V c . The required amplitude spectrum simply provides a graphical
presentation of the above results. See Figure 8.23.
(b) The specified band encompasses the carrier and the first four pairs of side frequencies, as indicated in
Figure 8.23. Therefore, the fraction of power in this band is obtained using Eq. (8.60) as follows
r4 = 0.17762 + 2(0.32762 + 0.04662 + 0.36482 + 0.39122 )
= 0.8228
This means that of the total (normalised) power in the FM signal (Pc = Vc2 ∕2 = 50 W), 82.3% or 41.14 W is in
the carrier and the first four pairs of side frequencies – a bandwidth of 120 kHz. For comparison, you may wish
to verify that in this case the Carson’s bandwidth BC = 180 kHz contains 99.34% of total power, whereas the
1% bandwidth BP = 240 kHz contains 99.98% of total power.

Worked Example 8.6

A 5 kHz tone of amplitude 5 V serves as the modulating signal for two modulator circuits, one an FM modulator
Telegram: @Persian_IEEEComSoc

of sensitivity 15 kHz/V and the other a PM modulator of sensitivity 0.6 rad/V.

(a) Determine the Carson’s bandwidth of each angle modulated signal.
(b) Repeat (a) with tone frequency increased to 15 kHz.

(a) Given f m = 5 kHz, V m = 5 V, kf = 15 kHz/V, and kp = 0.6 rad/V.

FM modulation index, 𝛽 = kf Vm ∕fm = 15
PM phase deviation, 𝜙d = kp Vm = 3
It follows that
FM Carson’s bandwidth, BCFM = 2(𝛽 + 1)fm = 160 kHz
PM Carson’s bandwidth, BCPM = 2(𝜙d + 1)fm = 40 kHz
 8.5 FM and PM Modulators 567

(b) With f m increased to 15 kHz, the FM modulation index reduces to 𝛽 = kf Vm ∕fm = 15 × 5∕15 = 5, but the phase
deviation does not depend on frequency and so remains at 𝜙d = kp V m = 3 rad. Hence

FM Carson’s bandwidth, BCFM = 2(𝛽 + 1)fm = 180 kHz

PM Carson’s bandwidth, BCPM = 2(𝜙d + 1)fm = 120 kHz

8.4.2.4 FM or PM?
It is noteworthy in Worked Example 8.6 that the PM bandwidth increases from 40 to 120 kHz in direct proportion
to the increase in the message signal frequency, compared to a less dramatic increase in FM bandwidth from
160 to 180 kHz. The reason for this is that frequency deviation in PM is proportional to both the amplitude and
the frequency of the modulating signal, whereas in FM it depends only on the modulating signal amplitude. See
Eq. (8.23). PM therefore requires a wider transmission bandwidth for higher-frequency components of a message
signal. Because receiver noise is directly proportional to transmission bandwidth, the noise performance of PM
will therefore be inferior to that of FM. For this reason, and because phase deviation in PM is limited to 𝜙d ≤ 𝜋 rad,
allowing less scope to trade bandwidth for noise performance improvement, FM is preferred to PM for analogue
communication. However, in digital communication, the staircase nature of the modulating signals gives PM an
important advantage over FM in bandwidth efficiency. Because there is no carrier frequency deviation in PM when
the modulating signal is constant, the bandwidth requirement of PM for digital transmission is less than that of
FM. PM is therefore widely used in digital communication. This is the subject of Chapter 11.

8.5 FM and PM Modulators

A detailed discussion of narrowband angle modulation is presented in Section 8.4.1. We will now show how NBPM
may be generated using the product modulator that was introduced in the previous chapter. Armed with an NBPM
modulator, it is an easy step to obtain an NBFM modulator based on the inherent relationship between FM and PM.
Wideband FM and PM may be obtained in two ways. In the indirect method, frequency multiplication is applied
to a narrowband angle modulated signal. The direct method of wideband angle modulation uses the modulating
signal to vary the output frequency of a voltage-controlled oscillator (VCO).

8.5.1 Narrowband Modulators

Consider Eq. (8.42), which gives the expression for an NBPM signal that is obtained when a carrier signal
vc (t) = V c cos(2𝜋f c t) is weakly phase modulated by the message signal v′m (t) = Vm cos(2𝜋fm t), such that the phase
deviation 𝜙d ≤ 0.25. Let us rewrite this equation as follows
Telegram: @Persian_IEEEComSoc

vnbpm (t) = Vc cos(2𝜋fc t) − 𝜙d cos(2𝜋fm t)Vc sin(2𝜋fc t)

= vc (t) − vm (t)v′c (t) (8.63)
The modulating signal vm (t) has maximum absolute value
|vm (t)|max = 𝜙d ≤ 0.25 (8.64)
and is obtained by attenuating the original message signal v′m (t) of maximum absolute value V m by the factor
V m /𝜙d . From earlier discussions, you will recognise the importance of Eq. (8.64). It is the condition that must be
satisfied to ensure that vnbpm (t) is an NBPM signal.
The form of Eq. (8.63) suggests that a narrowband phase modulated signal vnbpm (t) may be generated by the
circuit shown in block diagram form in Figure 8.24. A stable crystal oscillator generates the carrier signal vc (t),
 568 8 Frequency and Phase Modulation

Message ʋm(t)
Product
modulator

Vc sin(2π fct) – NBPM signal

Σ ʋnbpm(t)
90° +
Phase shift

ʋc(t) = Vc cos(2π fct)

Crystal
controlled
oscillator

Figure 8.24 Narrowband phase modulator.

Message ʋm(t) Narrowband NBFM signal

ʋo(t)
(a) signal Integrator phase ʋnbfm(t)
modulator

–
R
(b)
ʋm(t) +
ʋo(t)

Figure 8.25 (a) Narrowband frequency modulator; (b) integrator circuit.

and hence provides the first term on the right-hand side of Eq. (8.63). The oscillator output is also fed through a
phase-shifting network, which produces the signal v′c (t) = Vc sin(2𝜋fc t) at its output by imposing a phase delay of
Telegram: @Persian_IEEEComSoc

90∘ on the carrier signal at its input. The second term in the expression for vnbpm (t) requires the use of a product
modulator (see Chapter 7) that has the signals vm (t) and v′c (t) at its inputs. A summing device subtracts the output
of the product modulator from the carrier signal to give the required NBPM signal.
We show in our discussion of the relationships between FM and PM that if a message signal is first integrated
before being applied to phase modulate a carrier then the result is an FM signal. You may wish to review
Section 8.2.3.2 at this point. Therefore, the block diagram of a narrowband frequency modulator is as shown in
Figure 8.25a. The message signal is integrated and fed through the narrowband phase modulator discussed above.
The basic requirement is for the integrator to perform a specific type of lowpass filtering in which the gain is
inversely proportional to frequency. We saw that frequency deviation in PM is directly proportional to both the
frequency and amplitude of the modulating signal. Therefore, if the integrator output phase modulates a carrier,
the frequency dependence will cancel out, and we have an angle modulated signal whose frequency deviation
 8.5 FM and PM Modulators 569

depends only on amplitude. We learnt earlier that this is an FM signal. Thus, the output of the narrowband phase
modulator is the desired NBFM signal.
A typical integrator circuit is the operational amplifier (opamp) circuit shown in Figure 8.25b. The gain of the
circuit is proportional to the impedance of the element in the feedback path. So, placing in the feedback path
an element whose impedance is inversely proportional to signal frequency leads to the lowpass filtering action
described above. A capacitor is such an element. It blocks DC (i.e. it has infinite impedance at zero frequency) and
it passes AC with impedance that reduces as frequency increases.
To show more quantitatively how the circuit of Figure 8.25b works in conjunction with the NBPM modulator to
generate an NBFM signal, let us assume a sinusoidal message signal vm (t) = V m cos(2𝜋f m t). Then the impedance
of the capacitor in Figure 8.25b is
1
Z=
j2𝜋fm C
The noninverting terminal of the opamp is grounded, so the inverting terminal is forced to the same zero poten-
tial, which means that the input voltage vm (t) is dropped entirely across resistor R, and the output voltage vo (t)
is dropped entirely across impedance Z of capacitor C. Furthermore, the same current flows through R and C
because, by a property of opamps, negligible current flows into the input terminals marked (−) and (+). And since
current flow through C is in the opposite direction to the polarity of vo (t), as indicated on the circuit, it follows that
Vm cos(2𝜋fm t) v (t)
=− o
R Z
= −vo (t)[j2𝜋fm C]
Thus
Vm cos(2𝜋fm t) Vm sin(2𝜋fm t)
vo (t) = j =
2𝜋fm CR 2𝜋fm CR
kf Vm
= sin(2𝜋fm t) = 𝛽 sin(2𝜋fm t) (8.65)
fm
where kf is the frequency sensitivity given by
1
kf = (8.66)
2𝜋RC
and we have used the fact (discussed in Section 2.7.2) that the factor j represents a phase shift of 90∘ . Component
values and the amplitude of vm (t) are chosen to ensure that the modulation index 𝛽 ≤ 0.25.
When the integrator output vo (t), given by Eq. (8.65), is fed as input into the NBPM modulator shown in
Figure 8.24, you should verify that the output signal will be
vnbfm (t) = vc (t) − vo (t)v′c (t)
Telegram: @Persian_IEEEComSoc

= Vc cos(2𝜋fc t) − 𝛽Vc sin(2𝜋fc t) sin(2𝜋fm t) (8.67)

This is exactly the same expression as Eq. (8.41), derived for NBFM. The opamp circuit of Figure 8.25b therefore
works as an integrator in conjunction with an NBPM modulator to generate an NBFM signal.

8.5.2 Indirect Wideband Modulators

Indirect wideband modulators generate wideband angle modulation by frequency multiplication of a narrow-
band angle modulated signal. The indirect wideband modulator was first proposed by Edwin Howard Armstrong
(1890–1954) in 1936. For this reason, it is also called the Armstrong modulator. The following discussion assumes
FM. However, it is equally applicable to PM by substituting PM and 𝜙d for FM and 𝛽, respectively, wherever they
occur.
 570 8 Frequency and Phase Modulation

ʋm(t) ʋnbfm(t) Frequency ʋ′wbfm(t) ʋwbfm(t) Wideband

Message Down-
NBFM multiplier
signal fc1, β nfc1, nβ converter fc, nβ FM
(×n)
fLO = nfc1 + fc

LO

Figure 8.26 Armstrong wideband frequency modulator: basic elements.

Figure 8.26 shows the block diagram of a basic Armstrong modulator. It consists of an NBFM modulator, which
generates an NBFM signal vnbfm (t) of carrier frequency f c1 and modulation index 𝛽 ≤ 0.25. A completely general
expression for vnbfm (t) is given by Eq. (8.30)
vnbfm (t) = Vc cos(2𝜋fc1 t + 𝜙i )
where 𝜙i is the instantaneous phase. To increase the modulation index to the level required for wideband frequency
modulation (WBFM), this NBFM signal is fed into a frequency multiplier, which multiplies the angle of vnbfm (t)
by a factor of n. The output of the frequency multiplier is therefore
v′wbfm (t) = Vc cos(2𝜋nf c1 t + n𝜙i )
Since the instantaneous phase is proportional to modulation index, it follows that v′wbfm (t) is a WBFM signal
of (unmodulated) carrier frequency nf c1 , and modulation index n𝛽. With the new modulation index (n𝛽) at the
desired wideband value, the frequency nf c1 is usually much higher than desired. A downconverter is therefore
used to down convert to the desired carrier frequency f c . The downconverter consists of a mixer followed by a
bandpass filter (BPF) of bandwidth equal to the bandwidth of v′wbfm (t) and centre frequency chosen to pass only
the difference-frequency output of the mixer. A crystal-controlled local oscillator (LO) supplies the downconverter
with a sinusoid of frequency
fLO = fc + nf c1
The output of the frequency multiplier is the desired WBFM signal vwbfm (t) of unmodulated carrier frequency f c
and modulation index n𝛽.
The frequency multiplier in Figure 8.26 is a nonlinear device of order n followed by a BPF of centre frequency
nf c1 and bandwidth equal to n times the bandwidth of vnbfm (t). You may recall from Section 4.7.6 that when a
sinusoidal signal of angle 𝜃(t) = 2𝜋fm t + 𝜙m is passed through a nonlinear device of order n new sinusoids are
created in the output, including one with angle 𝜃n (t) = 2𝜋nf m t + n𝜙m , which of course has frequency nf m . Thus, if
a signal containing sinusoids of frequencies in the range f 1 to f m is passed through this nonlinear device, the output
will contain, among others, frequencies nf 1 to nf m . A BPF of passband nf 1 to nf m , hence bandwidth n(f m − f 1 ),
Telegram: @Persian_IEEEComSoc

may be used to extract this band of frequencies, and in this way frequency multiplication has been accomplished.
In practice, the frequency multiplier block consists of several cascaded stages to achieve the required frequency
multiplication factor. For example, multiplication by the factor 81 may be achieved using four multiply-by-three
stages. Worked Example 8.7 discusses the design of an Armstrong wideband frequency modulator involving two
stages of frequency multiplication.
It is important to note the difference between frequency down conversion and frequency multiplication. The
former process merely shifts a signal to a lower centre frequency. It does not change the modulation index or the
bandwidth of the shifted signal. The latter process alters all three parameters, namely centre frequency, modulation
index, and bandwidth by the factor n.
 8.5 FM and PM Modulators 571

Downconverter
NBFM (Frequency changer) WBFM
Message Narrowband First Second
frequency frequency Power
frequency multiplier multiplier amplifier
modulator (× n1) (× n2)
fLO

fc1, fc2 = n1fc1, fc3 = fLO – fc2 fc = n2 fc3

LO
β1 β2 = n1β1 β3 = β2 β = n2β3

Figure 8.27 Armstrong wideband FM transmitter employing two stages of frequency multiplication.

Worked Example 8.7

Figure 8.27 shows the block diagram of an Armstrong wideband FM transmitter that employs two stages of
frequency multiplication. The message is a sinusoid of frequency f m = 15 kHz and the NBFM signal has car-
rier frequency f c1 = 100 kHz and frequency deviation f d1 = 750 Hz. The required transmission from the power
amplifier stage is a WBFM signal having carrier frequency f c = 120 MHz and frequency deviation f d = 75 kHz.
The first frequency multiplication factor is n1 = 5.
Determine:
(a) The required second frequency multiplication factor n2 .
(b) The required LO frequency f LO .
(c) The (Carson’s) bandwidth of the FM signal at the output of each processing block in the transmitter.

Based on the action of frequency multiplication and down conversion discussed above, the carrier frequency
and modulation index at each stage of the transmitter are shown on the block diagram (Figure 8.27). For example,
at the output of the NBFM, the carrier frequency is f c1 and the modulation index is 𝛽 1 ; at the output of the first
frequency multiplier, the carrier frequency is f c2 = n1 f c1 and the modulation index is 𝛽 2 = n1 𝛽 1 ; and so on.
(a) The modulation index 𝛽 1 of the NBFM is 𝛽 1 = f d1 /f m = 0.75/15 = 0.05 and the modulation index 𝛽 of the
WBFM is 𝛽 = f d /f m = 75/15 = 5.
Walking backwards from the final stage of the transmitter to the first stage and replacing the modulation
index at each stage with the expression for modulation index from the previous stage, we obtain the overall
modulation index 𝛽 as
Telegram: @Persian_IEEEComSoc

𝛽 = n2 𝛽3 = n2 𝛽2 = n2 n1 𝛽1
Thus, with n1 = 5 (given), the second frequency multiplication factor n2 is
𝛽 5
n2 = = = 20
n1 𝛽1 5 × 0.05
(b) To determine LO frequency f LO , we start from the final stage of the transmitter and substitute backwards for
carrier frequency. This leads to
fc = n2 fc3 = n2 (fLO − fc2 ) = n2 (fLO − n1 fc1 )
 572 8 Frequency and Phase Modulation

and hence
fc
fLO = + n1 fc1
n2
120 MHz
= + 5 × 0.1 MHz
20
= 6.5 MHz
(c) Let B1 , B2 , B3 , and B4 , respectively, denote the bandwidth of the FM signal at the output of the NBFM, output
of the first frequency multiplier, output of the downconverter, and output of the second frequency multiplier.
It follows from Eq. (8.61) for Carson’s bandwidth that
B1 = 2(𝛽1 + 1)fm = 2(0.05 + 1)15 = 31.5 kHz
B2 = 2(𝛽2 + 1)fm = 2(n1 𝛽1 + 1)15 = 2(5 × 0.05 + 1)15 = 37.5 kHz
B3 = B2 = 37.5 kHz
B4 = 2(𝛽 + 1)fm = 2(5 + 1)15 = 180 kHz
B1 is the bandwidth of the NBFM signal and B4 the bandwidth of the final WBFM signal.

8.5.3 Direct Wideband Modulators

Oscillators are widely used in electronics to generate periodic waveforms for applications as a carrier or a control
signal that directs various repetitive system actions. The basic output of the oscillator is usually a nonsinusoidal
waveform (e.g. sawtooth or rectangular) from which a sinusoidal signal may be obtained, especially for application
as a carrier signal, by extracting the fundamental frequency or other harmonic frequency using a BPF. This is
shown in Figure 8.28a.
FM generation by the direct method works by a modulating signal directly varying the frequency of the oscillator
output signal. This yields the desired FM signal. The basic block diagram is shown in Figure 8.28b. How the
oscillator frequency is varied depends on the type of oscillator, and this is briefly discussed below. To obtain PM,
the signal is first passed through a differentiator circuit before entering the FM modulator. The basis of this method
of obtaining PM is discussed in Section 8.2.3 and shown in Figure 8.5. The differentiator is a special type of highpass
filter with gain that is directly proportional to signal frequency.
Two types of oscillators are in common use for FM generation. The simplest is the inductor (L)-capacitor (C)
oscillator (LCO), which has the capacitance of its tank circuit varied somehow by the modulating signal. This has
the effect of varying the resonant frequency of the oscillator. The second type is the voltage-controlled oscillator
(VCO) in which the modulating signal varies the voltage limits between which a timing capacitor charges and
discharges. In this way the period, and hence fundamental frequency, of the capacitor voltage is varied to yield the
Telegram: @Persian_IEEEComSoc

desired FM signal.
An important disadvantage of the direct method of angle modulation is that the carrier signal is not generated
by a stable crystal oscillator and will therefore tend to drift in frequency due to, for example, the effect of tem-
perature changes on component values. An arrangement known as automatic frequency control (AFC) must be
put in place to ensure frequency stability. A common AFC method is shown in Figure 8.29. The mixer is fed with
the direct modulator output signal vfm (t) – of average frequency f c – and a reference signal vref (t) of frequency f c
generated using a stable crystal oscillator. The difference-frequency signal output of the mixer serves as the input
to a frequency discriminator, the output of which is proportional to the frequency of the signal at its input. The
discriminator output ve (t) is zero when the modulator carrier frequency is at precisely the right value f c . Any car-
rier frequency error leads to a nonzero voltage of the correct sign, which is applied to adjust the direct modulator
in order to eliminate the deviation of the modulator carrier frequency from its allotted value.
 8.5 FM and PM Modulators 573

Carrier signal generator

rectangular
or sawtooth sinusoidal
Oscillator BPF
(a) fo
fo, 2fo, 3fo, …

Direct FM modulator

Oscillator FM signal
(b) BPF
fc fc + kfʋm(t)

ʋm(t)

Modulating
signal

Figure 8.28 (a) Carrier signal source; (b) direct FM modulator.

Modulating ʋm(t) Direct FM ʋfm(t) Frequency-stabilised

signal modulator FM signal

ʋe(t)

Frequency ʋref(t) Crystal

Mixer
discriminator oscillator

Figure 8.29 Automatic frequency control (AFC) of a direct FM modulator.

8.5.3.1 LCO Modulator

In the LCO modulator, the capacitance of the tank circuit is varied by the modulating signal. Figure 8.30a shows a
tank circuit with a parallel arrangement of inductance and capacitance. The total capacitance Ci at any instant is
Telegram: @Persian_IEEEComSoc

contributed by two capacitors, one with a fixed capacitance Co and the other with a capacitance Cv that is varied
by the modulating voltage vm (t). The resonant (or oscillation) frequency of the LC oscillator circuit, that is the
frequency of the oscillator’s output signal, at any instant is given by
1
fi = 1
(8.68)
2𝜋(LCi ) 2

Let the total capacitance of the tank circuit in the absence of a modulating signal be Cc . Then we may write

Ci = Cc + ΔC
= Cc − kvm (t) (8.69)
 574 8 Frequency and Phase Modulation

(a) Cʋ Co L

Ci = Cʋ + Co
C
Z

Cʋ = ʋm(t) OR ʋm(t)
+ + R R≪ 1
VB VB 2πfC
_ _

(b) (c)

Figure 8.30 (a) Tank circuit of LCO; and variable-capacitance C v obtained using either (b) varactor or (c) transistor.

where k is a constant known as the capacitance sensitivity, which gives the increase in capacitance per voltage
decrease in modulating signal. ΔC is the change in capacitance due to the modulating signal. It is assumed that
operation is restricted to the region of linear change of variable capacitance with modulating signal.
The oscillator’s output frequency in the absence of a modulating signal gives the unmodulated carrier frequency
1
fc = 1
(8.70)
2𝜋(LCc ) 2
Substituting Eq. (8.69) into Eq. (8.68) leads to the following expression for the instantaneous frequency f i of the
oscillator’s output signal.
1
fi = 1
2𝜋[L(Cc + ΔC)] 2
( )− 1
1 ΔC 2
= ( ) 1
= f c 1 +
1 Cc
2𝜋(LCc ) 2 1 + ΔC
2

Cc
( )
1 ΔC kf
≈ fc 1 − = fc + c vm (t)
2 Cc 2Cc
where we use the approximation
(1 + x)n ≈ 1 + nx, for |x| ≪ 1 (8.71)
Telegram: @Persian_IEEEComSoc

Equation (8.71) is applicable in this case, with x = ΔC∕Cc and n = − 1/2, the change in capacitance ΔC being a
small fraction of the unmodulated capacitance value Cc . The error in the above approximation is less than 0.4%
for x ≤ 0.1. We see that f i has the form
fi = fc + kf vm (t) (8.72)
where
kf c k
kf = = √ (8.73)
2Cc
4𝜋 LC3c
The LCO output is therefore the desired FM signal, with instantaneous frequency given by Eq. (8.72), and fre-
quency sensitivity kf given by Eq. (8.73). It is obvious that the variable capacitance must be operated in its linear
 8.5 FM and PM Modulators 575

region; otherwise, k, and hence kf , becomes a function of the modulating signal, leading to distortion when the sig-
nal is demodulated at the receiver. This requirement, and the need to keep ΔC/Cc < < 1 in order for approximation
(8.71) – which leads to Eq. (8.72) – to be valid, means that the frequency deviation may not be sufficient for WBFM
operation. In such situations, it is a straightforward matter to obtain WBFM with the required frequency deviation
by passing the oscillator output (FM) signal through a frequency multiplier and downconverter as discussed in
Section 8.5.2 and Question 8.14.
Figure 8.30b shows two popular arrangements for obtaining the voltage-variable capacitance Cv . One arrange-
ment uses a special variable-capacitance diode known as a varactor, whereas the other employs a transistor that
is set up in such a way that it has a reactive output impedance, which is varied by a modulating signal at the
input. The transistor is equivalent to a capacitor, with capacitance set by the bias voltage V B plus modulating
signal.

8.5.3.1.1 Variable Capacitance Diode

All pn diodes exhibit junction capacitance under reverse-bias. A varactor is a pn diode that is specially constructed
to increase the sensitivity of its junction capacitance to reverse-bias voltage and extend the linear region of the
capacitance variation. The nominal capacitance of a varactor diode may be in the range from about 2 to 400 pF
depending on type. In all cases, the capacitance decreases as the reverse bias voltage increases. A DC bias voltage
V B sets the varactor capacitance at the value

Cvo = Cc − Co

needed to fix the unmodulated carrier frequency as given by Eq. (8.70). The modulating signal vm (t) causes the
total reverse bias voltage to vary about V B , which changes the varactor capacitance and hence the instantaneous
capacitance Ci of the oscillator tank circuit according to Eq. (8.69). An LCO modulator that uses a varactor diode
in its tank circuit is sometimes referred to as a varactor diode modulator.
Ultra high frequency (UHF) two-way radios generate NBFM by using a varactor diode modulator followed by
frequency multiplication of ×36 (typical) to maintain linearity.

8.5.3.1.2 Variable Reactance Transistor

Figure 8.30c shows the basic transistor connection needed to provide a voltage-variable capacitance Cv seen look-
ing into the collector of the transistor. For simplicity, the usual transistor biasing components that provide the
indicated DC bias voltage V B and those that perform decoupling functions have been omitted. The impedance Z
looking into the collector terminal is given by
1 1
Z= ≡ (8.74)
j2𝜋fgm RC j2𝜋fCv
where gm is the transconductance of the transistor. The transistor therefore appears as a capacitor of equivalent
Telegram: @Persian_IEEEComSoc

capacitance

Cv = gm RC (8.75)

A modulating signal applied to the base of the transistor varies gm and hence the capacitance Cv . An LCO mod-
ulator that obtains its variable capacitance in this way is sometimes referred to as a transistor reactance modulator.

8.5.3.2 VCO Modulator

A VCO modulator generates an FM signal that has a highly linear relationship between frequency and modulating
signal, which is applied as the control voltage. Practical VCO modulators are available in integrated circuit (IC)
form, the most popular of which is the 566 VCO used for frequencies below 1 MHz. Figure 8.31 shows the block
diagram and external connections to a 566 VCO IC.
 576 8 Frequency and Phase Modulation

Vcc
R1 R
0.001 μF
6 8

Cc VB Current Schmitt Buffer Rectangular

source/sink trigger amp. 3 waveform output
5

R2
ʋm(t) Buffer Triangular
amp. 4 waveform output

7 1
C

Figure 8.31 The 566 VCO.

The control terminal (5) is biased to V B using the supply voltage V cc and the voltage divider formed with resistors
R1 and R2 . This bias voltage, along with the timing resistor R, determines the constant charging (discharge) current
I c that the timing capacitor C receives (sends) from (to) the current source (sink)
Vcc − VB
Ic = (8.76)
R
The repetitive charging and discharging of capacitor C makes available at terminal (4) a periodic triangular wave-
form of fundamental frequency
2(Vcc − VB )
fc = (8.77)
RCV cc
This voltage is fed into a Schmitt trigger to generate a rectangular waveform of the same fundamental fre-
quency, available at terminal 3. The Schmitt trigger output also serves to switch the operating mode of the current
source/sink each time the trigger output changes level.
When a modulating signal is applied to terminal 5 via a coupling capacitor Cc , it varies the effective bias voltage
V B , and hence the output frequency, according to Eq. (8.77). Connecting either terminal 3 or 4 to a BPF of centre
frequency f c then produces an FM signal based on a sinusoidal carrier signal. The bandwidth of the BPF is required
to be just enough to pass the fundamental frequency f c , and all significant side frequencies associated with f c , but
not any of the side frequencies associated with 2f c , 3f c , etc.
Telegram: @Persian_IEEEComSoc

The 566 has had a long history as an instrumentation VCO and is also used in FM telemetry.

8.6 FM and PM Demodulators

The practical implementation of angle modulation is discussed in Section 8.5.2. Let us now turn our attention to the
reverse process of angle demodulation. Given a PM or FM signal, how do we recover the original message signal
from the angle variations of the carrier? We first discuss frequency demodulation systems, and then show how
they may be extended to perform phase demodulation. There are two broad classes of frequency demodulators,
namely the direct demodulator, which uses frequency discriminators or zero crossing detectors, and the indirect
demodulator, which employs a phase-locked loop (PLL), which is discussed in Chapter 7.
 8.6 FM and PM Demodulators 577

8.6.1 Direct Demodulator

FM represents a message signal using a sinusoid of frequency proportional to the message signal value. A direct
demodulator somehow detects the instantaneous frequency of the incoming FM signal and produces an output
voltage of proportionate value. There are two ways of achieving this.

8.6.1.1 Filter-based Demodulator

The message signal can be recovered at a receiver by passing the incoming FM signal through a filter with a linear
frequency response, i.e. with a gain that is proportional to frequency. Thus, the filter takes in a constant amplitude
FM signal and produces an output signal whose amplitude varies in sync with the instantaneous frequency of the
FM signal, and hence with the original message signal. Thus, this filter changes the envelope of the FM signal
from a constant value V c to a value that varies exactly as the original message signal – a sort of AM/FM hybrid
signal. An envelope demodulator (discussed in Chapter 7) will ignore the frequency variations and recover the
original message signal. The filter just described is known as a frequency discriminator, and the combination of a
frequency discriminator followed by an envelope demodulator gives direct FM demodulation.
Note that the frequency response of the discriminator need only be linear in the frequency band that contains
all possible instantaneous frequency values of the FM signal. This is the band f c ± f d . Furthermore, there may be
amplitude variations in the incoming FM signal because of noise and distortion in the transmission medium, or
due to the small amplitude variations that are inherent in NBFM generation. To prevent these unwanted varia-
tions getting into the recovered message signal, the incoming FM signal is first passed through a circuit known as
a limiter, which clips the FM signal to eliminate fluctuations in its envelope. To exclude the new frequency compo-
nents associated with the clipping action (see Section 4.7.6), the limiter usually includes a tuned tank circuit that
passes only the wanted frequency components around f c . Furthermore, a transistor circuit is usually employed,
which provides signal amplification as well, all carrier amplitudes above a certain minimum or threshold level
being boosted to the same output level. The output of the limiter is then the desired constant-amplitude FM sig-
nal. A block diagram of the direct FM demodulator is shown in Figure 8.32, which also shows the waveform at
each stage of the demodulation process.

8.6.1.2 Digital Demodulator

Direct FM demodulation can also be accomplished using a digital technique. The number of zero crossings in the
FM signal within fixed time intervals of duration T is gated into a binary counter, which feeds a digital-to-analogue
converter (DAC). The DAC output gives a direct measure of the instantaneous frequency of the FM signal, and
hence the value of the original message signal during the interval. The interval T over which the number of
positive-going zero crossings is counted should satisfy the condition
1 1
≪T≪ (8.78)
fc fm
Telegram: @Persian_IEEEComSoc

where f m is the maximum frequency component of the message signal. The first condition T > > 1/f c ensures that
there are zero crossings within the interval T, whereas the second condition T < < 1/f m avoids excessive averaging
of the message signal.

8.6.2 Indirect Demodulator

The indirect method of FM demodulation uses a PLL to match the frequency of a VCO to the frequency varia-
tions in the FM signal. When the VCO correctly tracks the FM signal in frequency then it follows that its control
voltage corresponds exactly to the original message signal. Figure 8.33a shows a block diagram of the PLL (FM)
demodulator.
 578 8 Frequency and Phase Modulation

(a) ʋ'fm(t) ʋfm(t) Frequency ʋafm(t) Envelope ʋm(t)

Limiter
discriminator demodulator

ʋ'fm(t)

(b)

ʋfm(t)

(c)

ʋafm(t)

(d)

ʋm(t)

(e)

Figure 8.32 Direct FM demodulator using frequency discriminator: (a) Block diagram; (b)–(e) Waveforms.

The PLL consists of a phase discriminator with a VCO in its feedback path, which is arranged such that the
output vv (t) of the VCO tends towards the same frequency as the FM signal input to the phase discriminator.
The phase discriminator consists of a multiplier or product modulator followed by a lowpass filter (LPF). The
LPF rejects the sum-frequency output of the multiplier but passes the difference-frequency output, denoted vp (t),
which serves as the control voltage of the VCO. The VCO is set so that when its input or control voltage is zero its
output vv (t) has the same frequency as the unmodulated carrier, but with a phase difference of 𝜋/2 rad. Thus, if
the (normalised) unmodulated carrier is
Telegram: @Persian_IEEEComSoc

vc (t) = cos(2𝜋fc t)

then, from Eq. (8.32), the VCO output at any time is given by

vv (t) = cos(2𝜋fc t + 𝜋∕2 + 𝜙v ) (8.79)

where
t
𝜙v = 2𝜋kv vp (t)dt (8.80)
∫0

kv is the frequency sensitivity of the VCO and 𝜙v = 0 when the VCO is free running.
 8.6 FM and PM Demodulators 579

Phase discriminator

FM ʋfm(t) Product ʋp(t) Message

LPF
(a) signal modulator signal

ʋʋ(t)

Voltage
controlled
oscillator

ʋm(t)
ʋm(2Δt)
(b) ʋm(Δt)

ʋm(0) t
0 Δt 2Δt

Figure 8.33 (a) Indirect frequency demodulator using phase-locked loop (PLL); (b) Staircase approximation of original
message signal v m (t).

8.6.2.1 PLL Demodulation Process

For a simplified view of how the PLL demodulates the input FM signal vfm (t), let us assume a staircase approxima-
tion of the original message signal vm (t) as shown in Figure 8.33b and reckon time from the instant when the VCO
is free-running and vfm has frequency f c . That is, the original message signal is zero at this instant (t = 0), so we
may write vm (0) = 0. Furthermore, let us assume that the VCO output only changes at the discrete time instants
t = Δt, 2Δt, …
The phase discriminator output vp (t) at t = 0 is the difference-frequency product of vc (t) and vv (t) with 𝜙v = 0
vp (0) = cos(𝜋∕2) = 0
Now consider the situation at t = Δt. The message signal made a step increase to the value vm (Δt) at time t = 0,
and therefore just before this time (t = Δt) the phase of vfm (t) has gained on the phase of the VCO output vv (t),
which is still free-running, by a small amount 𝜙fm1 given by
Telegram: @Persian_IEEEComSoc

𝜙fm1 = [2𝜋kf vm (Δt)]Δt (8.81)

At the instant t = Δt, the VCO re-establishes tracking by making a step change of 𝜙v1 = 𝜙fm1 in its phase. From
Eq. (8.80), this requires a control voltage of average value vpa1 during the interval t = 0 to Δt given by the relation
𝜙v1 = 2𝜋kv vpa1 Δt (8.82)
Equating the right-hand side of Eqs. (8.81) and (8.82) yields the result
kf
vpa1 = vm (Δt) (8.83)
kv
By the next time instant t = 2Δt, the FM signal phase has increased to
𝜙fm2 = 𝜙fm1 + [2𝜋kf vm (2Δt)]Δt
 580 8 Frequency and Phase Modulation

and the VCO again makes a step change in the phase of vv (t) to the new value
𝜙v2 = 𝜙v1 + 2𝜋kv vpa2 Δt
where vpa2 is the average value of vp (t) in the interval t = Δt to 2Δt. Equating the expressions for 𝜙v2 and 𝜙fm2 , and
noting that 𝜙v1 = 𝜙fm1 , we obtain
kf
vpa2 = vm (2Δt)
kv
It follows by a similar argument that the average value of vp (t) in the nth time interval is related to the original
message signal by
kf
vpa,n = vm (nΔt), n = 1, 2, 3, · · · (8.84)
kv
In the limit Δt → 0, tracking becomes continuous, and the average values vpa1 , vpa2 , …, become the instantaneous
values of vp (t), and we may write
kf
vp (t) = vm (t) (8.85)
kv
The Eq. (8.85) shows that, when the PLL is in the tracking or phase-locked state, its output gives the original mes-
sage signal vm (t), with a scaling factor kf /kv . FM demodulation has therefore been achieved. The LPF is designed
with a bandwidth that is just enough to pass the original message signal, e.g. up to 15 kHz for audio signals.

8.6.2.2 PLL States

The PLL can operate in three different states. It is in the free-running state when the VCO control voltage is zero, or
when the input and VCO frequencies are too far apart. In this state the VCO output vv (t) oscillates at a frequency
determined by an external timing capacitor (in the usual IC implementation of PLLs). The tracking or locked state
is discussed in Section 8.6.2.1. Once locked to an input signal vfm (t), the PLL can vary the frequency of vv (t) to keep
step with the frequency variations of vfm (t) over a wide but finite band, referred to as the lock range. The PLL is said
to be in the capture state when it is in the process of continuously changing the frequency of vv (t) until it is equal to
the frequency of the input signal vfm (t), i.e. until lock is achieved and the tracking state ensues. However, lock will
not be achieved if the initial separation between the input frequency and the VCO frequency is too large. The band
of input signal frequencies over which the PLL will capture the signal is known as the capture range. This is usually
much narrower than the lock range. Both the capture and lock ranges are centred at the free-running frequency.

8.6.2.3 PLL Features

The PLL is an extremely versatile device with numerous applications in electronics. It provides excellent FM
Telegram: @Persian_IEEEComSoc

demodulation with several advantages over discriminator-based direct demodulation. For example, it is not sensi-
tive to amplitude variations in the FM signal, so a limiting circuit is not required. Since its circuit does not include
inductors, the PLL is more easily implemented in IC form. It also obviates the need for complicated coil adjust-
ments. The PLL is also very linear, giving accurate recovery of the original message signal. Furthermore, the PLL’s
feature of capturing or passing only signals that are within a small band makes it very effective at rejecting noise
and interference and gives it a superior signal-to-noise ratio.

8.6.3 Phase Demodulator

A phase demodulator can be easily derived from the frequency demodulators discussed above, based on the the-
oretical relationships between PM and FM. To see how this may be achieved, consider Eq. (8.21), which gives the
 8.6 FM and PM Demodulators 581

d ʋ (t)
ʋm(t)
PM ʋpm(t) Frequency dt m Message
Integrator
signal demodulator signal

Figure 8.34 Phase demodulator.

instantaneous frequency f i of a PM signal. Note that the frequency variation in a PM signal is proportional to the
derivative of the message signal. We were able to use the above FM demodulator schemes because frequency vari-
ation in FM is proportional to the message signal. See Eq. (8.4). Thus, if a PM signal is the input to any of the above
FM demodulators, the output signal will be the derivative of the original message signal. The original message
signal can then be obtained by following the FM demodulator with an integrator. A PM demodulator is therefore
as shown in Figure 8.34.

8.6.4 Frequency Discriminators

Frequency discriminators can be implemented using various circuits. The main requirements are that the fre-
quency response H(f ) of the circuit should be linear in the frequency band of interest and should change rapidly
with frequency to give acceptable demodulation sensitivity. Thus, any circuit whose amplitude response can be
represented by the following linear function of frequency f will function properly as a discriminator
|H(f )| = k1 + k2 f (8.86)
Here k1 is a constant, which may be zero, but k2 is a nonzero constant that determines the sensitivity with which
frequency variation in the FM signal at the discriminator input is converted to amplitude variation in the AM/FM
hybrid signal at the output. A few examples of discriminators are discussed in the following sections.

8.6.4.1 Differentiators
We want to show that a differentiator satisfies Eq. (8.86), which means that it can be used as a frequency discrim-
inator. Let the input to the differentiator in Figure 8.35 be a sinusoidal signal of frequency f
x(t) = cos(2𝜋ft)
Then the output signal is given by
dx
y= = −2𝜋f sin(2𝜋ft)
dt
= 2𝜋f cos(2𝜋ft + 𝜋∕2)
≡ |H(f )| cos[2𝜋ft + 𝜙H (f )]
Telegram: @Persian_IEEEComSoc

Note that, as discussed in Section 4.7, the effect of the differentiator is to change the amplitude of the input
sinusoid by a factor |H(f )| equal to its amplitude response, and to shift the phase of the input sinusoid by an
amount 𝜙H (f ) equal to its phase response. It follows that the frequency response of a differentiator is given by
H(f ) ≡ |H(f )| exp[j𝜙H (f )]
= 2𝜋f exp(j𝜋∕2)
= 2𝜋f [cos(𝜋∕2) + j sin(𝜋∕2)]
= j2𝜋f (8.87)
which satisfies Eq. (8.86) with k1 = 0, and k2 = j2𝜋. Thus, FM demodulation can be carried out by differentiating
the FM signal followed by envelope demodulation. Question 8.15 addresses this further. Eq. (8.87) shows that, in
 582 8 Frequency and Phase Modulation

x(t) dx
Differentiator y=
(a) dt
Input H(f)
Output

|H(f)| x(t)

Slope k2 = 2π x(t)
x(t) – x(t – t)
Slope y(t) =
τ
x(t – τ)
f t
t–τ t
(b) (c)

Figure 8.35 Differentiator: (a) block diagram; (b) amplitude response; (c) approximation.

general, if the transfer function of a circuit in a specified frequency range can be approximated by
H(f ) = jKf
where K is a constant then that circuit acts as a differentiator of gain K/2𝜋 in that frequency band. Note that the
frequency response of a differentiator (see Figure 8.35b) is linear over the entire frequency axis, not just within a
limited band of interest. We will consider two examples of differentiator circuits.

8.6.4.1.1 Delay-line Differentiator

The output of a differentiator is the derivative of its input signal, which, according to Figure 8.35c, may be approx-
imated by
x(t) − x(t − 𝜏)
y(t) = (8.88)
𝜏
where 𝜏 is the time delay between adjacent sample values, which must be small enough to satisfy the condition
1
𝜏≪
2𝜋fc
Eq. (8.88) suggests a system known as a delay-line differentiator that uses a delay line and a summing device to
obtain the difference between adjacent input signal samples, followed by a gain factor of 1/𝜏. Since a delay of 𝜏
(on a signal having frequency f ) is equivalent to a phase shift of 2𝜋f𝜏, the demodulator realised using a delay-line
differentiator followed by an envelope demodulator is referred to as a phase shift demodulator. This is shown in
Telegram: @Persian_IEEEComSoc

Figure 8.36.

Delay-line differentiator

ʋfm(t) + ʋm(t)
FM
signal Σ 1/τ
Envelope
demodulator
Message
signal
–

Delay
τ ʋfm(t – τ)

Figure 8.36 Phase shift demodulator.

 8.6 FM and PM Demodulators 583

8.6.4.1.2 Resistor-capacitor (RC) Differentiator

The simple highpass RC filter circuit shown in Figure 8.37a has the frequency response
R
H(f ) =
R + 1∕(j2𝜋fC)
j2𝜋fRC
=
1 + j2𝜋fRC
≈ j2𝜋fRC, for 2𝜋fRC ≪ 1 (8.89)
It therefore acts as a differentiator of gain RC at low frequencies (f < < 1/2𝜋RC), which is the linear portion
of the frequency response shown (shaded) in Figure 8.37b. The RC filter will work as a discriminator for FM
demodulation if the frequency band f c ± f d falls in this linear region. Assuming f c ∼ 10 MHz and R = 1 kΩ, this
requires a capacitance C of less than about 3.0 pF, which is a small value indeed, comparable to stray capacitances
in circuits. An LC BPF is more suitable for linear frequency response approximation at the usual high values of f c ,
an intermediate frequency (IF) value of 10.7 MHz being common in FM broadcast receivers.

8.6.4.2 Tuned Circuits

A tuned (RLC) circuit can also be used as a discriminator if it is detuned such that the unmodulated carrier fre-
quency falls in the centre of the linear portion of the circuit’s gain response, as shown in Figure 8.38. The main
limitation of this simple circuit is that the linear portion of the gain response is very narrow and not enough

Gain = ʋo/ʋi

Nonlinear portion

C
ʋi ʋo
R Linear portion

Frequency
(a) (b)

Figure 8.37 (a) Simple RC highpass filter; (b) frequency response.

Output voltage
Linear portion
Telegram: @Persian_IEEEComSoc

ΔV

Δf

Input frequency
fc

Figure 8.38 Frequency response of tuned circuit showing linear portion.

 584 8 Frequency and Phase Modulation

Balanced discriminator Envelope demodulator

C2
(a) FM L2 Message
C L
in out
L1
C1

Output voltage

fc
(b)
f1 f2 Frequency

Extended linear
region

Figure 8.39 Balanced discriminator: (a) circuit diagram; (b) frequency response.

for WBFM, which has a large frequency deviation. The linear portion can be extended by using two back-to-back
BPFs tuned to two different resonant frequencies spaced equally below and above the carrier frequency. This gives
what is referred to as a balanced discriminator circuit, an example of which is shown in Figure 8.39a along with
the resulting frequency response in (b). In the circuit, L and C are tuned to frequency f c , L1 and C1 are tuned to
frequency f 1 , and L2 and C2 to frequency f 2 . The remaining part of the circuit provides envelope demodulation.
Telegram: @Persian_IEEEComSoc

8.7 FM Transmitter and Receiver

8.7.1 Transmitter
The basic elements of a broadcast FM transmitter are shown in Figure 8.40. The output of the WBFM modulator,
the operation of which is described in connection with Figure 8.26, is amplified into a high-power FM signal
and coupled to an antenna for radiation. There is, however, a new processing block, called pre-emphasis, which
operates on the message signal before it is applied to angle modulate the carrier. The role of pre-emphasis is to
improve the noise performance of the FM communication system, as briefly discussed below.
 8.7 FM Transmitter and Receiver 585

Antenna

Audio Pre- Wideband RF power High-power

input emphasis FM modulator amplifier FM

fc1 fLO

Crystal
oscillators

Figure 8.40 FM transmitter.

Figure 8.41 Worst-case phase error 𝜙e due to noise voltage V n .

V fmn
ϕe Vn
Vfm

The effect of additive noise is to alter both the amplitude and phase of the transmitted FM signal, as illustrated
in Figure 8.41. A signal of amplitude V fm is transmitted, but because of additive noise of random amplitude V n ,
the demodulator receives a signal V fmn , which is the sum of V fm and V n . The noise voltage V n can have any
phase relative to the FM signal V fm but is shown in the phasor diagram with a phase of 90∘ (relative to V fm ),
where the phase error 𝜙e is maximum for a given noise amplitude. Since an FM signal carries information in its
angle, the amplitude error in V fmn is of no effect and all amplitude variations are removed by a limiting circuit
prior to demodulation. However, the phase error 𝜙e will somehow corrupt the received message signal since the
demodulator has no way of knowing that the additional phase shift comes from noise.
The maximum phase error 𝜙e increases with noise amplitude up to a value of 𝜋/4 rad at V n = Vfm . Using
Eq. (8.28), we see that this phase error will lead to an error f dn in the frequency deviation, which translates to an
error V nr in the received message signal given by
fdn = kf Vnr = 𝜙e fm (8.90)
where f m is the frequency of the message signal. We make two important observations:
(i) The effect of the phase error increases with message signal frequency. That is, the voltage error due to noise
is larger for the higher-frequency components of the message signal. The problem is further compounded by
Telegram: @Persian_IEEEComSoc

the fact that the higher-frequency components of the message signal are usually of small amplitudes. As a
result, the ratio s/n between signal and noise voltages decreases significantly with frequency. To reduce this
problem, the amplitude of the high-frequency components of the message signal can be artificially boosted
prior to modulation. This process is known as pre-emphasis, and the circuit that is used is called a pre-emphasis
filter. This introduces a known distortion into the message signal, which is readily removed at the receiver after
demodulation by applying a filter whose transfer function is the exact inverse of the pre-emphasis filter. This
second operation, performed at the receiver, is known as de-emphasis.
(ii) The effect of noise, as indicated by the error voltage V nr in the demodulated signal, can be reduced arbitrarily
by increasing frequency sensitivity kf . Since modulation index 𝛽 = kf V m /f m , this amounts to increasing modu-
lation index, and hence transmission bandwidth. Thus, FM provides an effective means of trading bandwidth
for reduced noise degradation, or improved noise performance.
 586 8 Frequency and Phase Modulation

8.7.2 SNR and Bandwidth Trade-off

We can obtain a rough calculation of the worst-case improvement in noise performance afforded by FM as follows.
From Figure 8.41, the ratio between signal voltage and noise voltage at the input of the demodulator is
(s∕n)i = Vfm ∕Vn
If V n is small compared to Vfm , i.e. (s∕n)i ≫ 1, then the phase error can be approximated by
Vn 1
tan 𝜙e = =
Vfm (s∕n)i
≈ 𝜙e (8.91)
Note that 𝜙e is the largest possible phase error due to noise of amplitude V n and leads to the worst-case or
smallest ratio between signal voltage and noise voltage at the output of the demodulator. Let us denote this ratio
as (s∕n)o . It is clearly given by the ratio between the frequency deviation f d due to the message signal and the
frequency deviation f dn due to noise, since the demodulator translates frequency variation to voltage variation.
Thus
f f 𝛽
(s∕n)o = d = d =
fdn 𝜙e fm 𝜙e
= 𝛽(s∕n)i (8.92)
where we have used Eq. (8.91). Since the square of voltage gives normalised power, we square both sides
of Eq. (8.92) to obtain the relationship between the output signal-to-noise power ratio SNRo and the input
signal-to-noise power ratio SNRi
SNRo = 𝛽 2 (SNRi ) (8.93)
Equation (8.93) is an important result, which shows that FM gives an improvement in output signal-to-noise
power ratio that increases as the square of modulation index 𝛽. However, this is the worst-case scenario: the actual
improvement is usually better than stated in Eq. (8.93). When account is taken of the random variation of the phase
of the noise voltage V n relative to Vfm , then the effective phase error is less than 𝜙e . The exact relationship between
SNRo and SNRi – for large SNRi – is derived in Section 8.8.

8.7.3 Pre-emphasis and De-emphasis

Simple RC filters are usually employed for pre-emphasis and de-emphasis. Figure 8.42 shows the pre-emphasis
and de-emphasis circuits used for standard FM broadcasting. The frequency response of the pre-emphasis filter is
given by
Telegram: @Persian_IEEEComSoc

R2 1 + jf ∕f1
Hp (f ) = R1
=K
R2 + 1 + jf ∕f2
1+j2𝜋fR1 C
R2 1 1
K= ; f1 = =
R1 + R2 2𝜋𝜏1 2𝜋R1 C
1 1 R 1 R2
f2 = = ; R= (8.94)
2𝜋𝜏2 2𝜋RC R1 + R2
Similarly, the frequency response of the de-emphasis filter is
1
Hd (f ) = (8.95)
1 + jf ∕f1
with f 1 given as above. The two frequency responses are plotted in Figure 8.43.
 8.7 FM Transmitter and Receiver 587

Figure 8.42 (a) Pre-emphasis filter; (b) de-emphasis filter. C

(a) Input R1 R2 Output

R1
(b) Input C Output

log10∣Hp(f)∣

(a)

log10(f)

log10∣Hd(f)∣

(b)

log10(f)
f1 fm

Figure 8.43 Frequency response of (a) pre-emphasis filter and (b) de-emphasis filter.
Telegram: @Persian_IEEEComSoc

In FM audio broadcasting, the RC time constant 𝜏 1 = 75 μs, so that the first corner frequency f 1 = 2120 Hz. The
second corner frequency f 2 is chosen to be well above the maximum frequency component f m (= 15 kHz) of the
audio signal. Observe that the linearly increasing response of the pre-emphasis circuit between f 1 and f m means
that the circuit behaves as a differentiator at these frequencies. Refer to Eq. (8.87). It follows from Figure 8.5 that,
when the pre-emphasised audio signal frequency modulates the carrier, the result will be FM for frequencies up
to about 2120 Hz, and PM for higher frequencies. Thus, FM with pre-emphasis is a combination of FM (at lower
message frequencies) and PM (at the higher message frequencies).
We have stated earlier that the de-emphasis circuit is used at the receiver after the frequency demodulator to
undo the effect of pre-emphasis. However, an illuminating alternative view of the role of de-emphasis can be
obtained by observing that the frequency response of the de-emphasis circuit is approximately uniform up to f 1
 588 8 Frequency and Phase Modulation

Tuneable RF IF
amplifier amplifier FM De-
Mixer
fc fIF ± 100 kHz demod. emphasis

Local
oscillator Audio
Common tuning fLO = fc + fIF amplifier

Loudspeaker

Figure 8.44 FM superheterodyne receiver.

and linearly decreasing beyond this point. That is, this circuit acts as an integrator for frequencies above f 1 . Thus, in
the light of Figure 8.34, we have FM demodulation for frequencies below f 1 , and PM demodulation for frequencies
above f 1 , and the original message signal is accurately recovered from the received hybrid FM/PM signal. This is
remarkable.

8.7.4 Receiver
Figure 8.44 shows a block diagram of a complete FM receiver system. It is based on the superheterodyne principle
discussed in detail in Chapter 7 for AM reception. See Figure 7.22. Note the following differences in Figure 8.44.
The IF is f IF = 10.7 MHz for audio broadcasting, and the bandwidth of the IF amplifier is 200 kHz to accommodate
the received FM signal. The demodulator is usually by the direct method of Figure 8.32, consisting of limiter,
discriminator, and envelope demodulator, but it may also be by the indirect method of Figure 8.33 involving a
PLL. Note that, as discussed above, de-emphasis is carried out after FM demodulation.
You may wish at this point to review the discussion of the operation of an FM stereo receiver given in Chapter
7 (Section 7.7.1.4 and Figure 7.32).

8.8 Noise Effect in FM

One of the main advantages of FM over all the AM techniques discussed in Chapter 7 is that FM minimises the
degrading effect of additive white Gaussian noise on the recovered message signal in a way that can be further
enhanced by increasing the modulation index (and hence transmission bandwidth). FM conveys information
exclusively through the phase of the carrier, so any noise-induced amplitude variations in the FM signal will be
of no consequence. Noise will introduce carrier phase errors which will be translated into errors in the recov-
Telegram: @Persian_IEEEComSoc

ered message; however, these errors are small, and the message degradation is negligible, if carrier amplitude (or
power) is much greater than noise voltage (or power).
To gain a quantitative and more complete understanding of the effect of noise in FM, consider Figure 8.45, which
shows the phasor addition of noise voltage r to a transmitted FM signal of amplitude Ac to produce the received
FM signal of amplitude Acn . The addition is illustrated for various scenarios of the relative amplitudes of carrier
and noise, namely (i) Ac much larger than r, (ii) Ac just larger than r, and (iii) Ac less than r. The noise voltage r
has random phase 𝜓 that is uniformly distributed between −𝜋 and 𝜋 rad, so its phasor when added to Ac will start
at point D (the endpoint of the transmitted FM signal phasor) and will terminate anywhere on the shaded circle
of radius r centred on D. The transmitted FM signal phasor has phase 𝜙 (which contains the transmitted message
signal and information), but the received FM signal phasor has phase 𝜙 + 𝜀, which includes a phase error 𝜀 due to
the addition of the noise voltage r. We see that when r is much smaller than Ac (as in Figure 8.45a) the maximum
 8.8 Noise Effect in FM 589

B r
D

Acn B r
D
D r
Ac Ac
Acn ε B ε
ϕ
max Ac Acn
O
εmax ϕ
ϕ O
O
(a) Ac >> r: εmax small (b) Ac ≈ r: εmax large (c) Ac < r: εmax = 180°
B

r ψ–ϕ C
ψ
D E
A cn
(d) Analysis of scenario Ac >> r
Ac
ε
θ
ϕ
O

Figure 8.45 Phasor addition of noise voltage r to FM signal of amplitude Ac . Maximum phase error 𝜀max in the received FM
signal of amplitude Acn depends on the magnitude of Ac relative to r.

phase error 𝜀max is small, but when r is close in magnitude to Ac (as in Figure 8.45b) then 𝜀max is large, and when r is
larger than Ac (as in Figure 8.45c) then 𝜀max = 180∘ . In order for the transmitted message to be reliably recovered by
the FM demodulator, it is necessary that 𝜀max is small; otherwise, Acn will undergo large random changes in phase
which will be converted into large signal excursions at the demodulator output, perceived as loud crackling noise
in an audio message signal. The analysis that follows is therefore based on Figure 8.45d, which assumes Ac > > r.
In Figure 8.45d, the transmitted FM signal phasor Ac , noise voltage r, and received FM signal phasor Acn have
respective phases 𝜙, 𝜓, and 𝜃 = 𝜙 + 𝜀, where 𝜀 is the phase error due to noise. Consider triangle OBC. Let us resolve
the noise voltage phasor r into in-phase and quadrature components vnI (t) and vnQ (t) relative to the carrier phasor
Ac so that in Figure 8.45d length DC ≡ vnI (t) and length BC ≡ vnQ (t). Note that we could have resolved r relative to
the 0∘ direction so that length DE ≡ vnI (t) and length BE ≡ vnQ (t), but the statistical distributions of vnI (t) and vnQ (t)
would be unchanged in either case. This simply exploits the fact that the distribution of sin(𝜓 − 𝜙) is unaffected
by the value of 𝜙, allowing sin(𝜓 − 𝜙) to be replaced by sin𝜓. Thus
vnQ (t)
Telegram: @Persian_IEEEComSoc

BC
tan 𝜀 = =
OC Ac + vnI (t)
For Ac > > r, 𝜀 → 0 and tan𝜀 ≈ 𝜀, so that
vnQ (t)
𝜃(t) = 𝜙(t) + 𝜀 ≈ 𝜙 +
Ac + vnI (t)
vnQ (t)
≈ 𝜙(t) + (since Ac ≫ vnI (t)) (8.96)
Ac
Now consider the FM receiver model of Figure 8.46. The BPF blocks all signals outside the range f c ± B/2 and
passes the noise corrupted FM signal
va (t) = vfm (t) + vn (t) ≡ Acn cos[2𝜋fc t + 𝜃(t)]
 590 8 Frequency and Phase Modulation

Figure 8.46 FM receiver model.

Incoming signals
w(t)

White noise
+

BPF
(fc – B/2 → fc + B/2)

ʋa(t) C/N

Limiter

ʋb(t)
FM demodulator

Discriminator

ʋd(t)

LPF
(–fm → fm)

Message signal + noise SNR

where
vfm (t) = Ac cos[2𝜋fc t + 𝜙(t)]
vn (t) = r(t) cos[2𝜋fc t + 𝜓(t)] = vnI (t) cos(2𝜋fc t) − vnQ (t) sin(2𝜋fc t)

are, respectively, the wanted FM signal and noise voltage. This representation of noise voltage in a bandpass system
is derived in Section 6.3.3. With noise power equal to N o B, and carrier power = A2c ∕2, the carrier-to-noise ratio
(C/N) at the FM demodulator input is
A2c
C∕N = (8.97)
2No B
Note that this is the signal-to-noise ratio (SNRi ) at the input of the demodulator, but since the signal is still on the
Telegram: @Persian_IEEEComSoc

carrier signal at this point in the transmission system, the description of this quantity as carrier to noise ratio (C/N)
is preferred. The discriminator produces an output that is proportional to its input frequency (which is the rate of
change of angle divided by 2𝜋). Normalising the constant of proportionality to unity, the discriminator output is
given by
[ ]
1 d𝜃(t) 1 d vnQ (t)
vd (t) = = 𝜙(t) +
2𝜋 dt 2𝜋 dt Ac
1 dv nQ (t)
= kf vm (t) +
2𝜋Ac dt
𝛽f
≡ m vm (t) + vnd (t)
Am
 8.8 Noise Effect in FM 591

ʋnQ(t) ʋnd(t)
H(f) = j2πf/(2πAc)
SnQ(f) Snd(f) = SnQ(f)|H(f)|2

Snd(f)
SnQ(f)

No
f f
–B/2 B/2 –B/2 B/2

Figure 8.47 Processing of v nQ (t), having baseband equivalent PSD S nQ (f ) shown, by discriminator to produce v nd (t) having
PSD S nd (f ), also shown.

where d𝜙(t)∕dt was obtained as 2𝜋kf vm (t) by taking the derivative of Eq. (8.27). The first term of vd (t) is the message
signal which is fully passed by the LPF to produce signal power Ps at demodulator output given by

Ps = 𝛽 2 fm2 A2rms ∕A2m = (𝛽fm ∕R)2 (8.98)

where R = Am ∕Arms is the peak-to-rms ratio of the message signal. The second term of vd (t) is noise whose power is
more readily computed in the frequency domain, noting that only a portion of this noise in the frequency range −f m
to +f m is passed by the LPF. The time domain operation of (1∕2𝜋Ac )d∕dt on vnQ (t) to produce vnd (t) is illustrated in
the frequency domain in Figure 8.47 based on the Fourier transform property (Eq. (4.88)) in which differentiation
in the time domain corresponds to multiplication by j2𝜋f in the frequency domain. Thus, this operation is equiv-
alent to passing vnQ (t), having a flat baseband equivalent power spectral density (PSD) SnQ (f ) of height N o in the
frequency range −B/2 to B/2 plotted in Figure 8.47, through a linear system of transfer function H(f ) = j2𝜋f /(2𝜋Ac )
to produce an output vnd (t), which has PSD Snd (f ) given by Eq. (4.163) as

Snd (f ) = SnQ (f )|H(f )|2

f2
= No , −B∕2 ≤ f ≤ B∕2 (8.99)
A2c
Thus, since f m < B/2, it follows that noise power at the demodulator output is
fm fm
No f 2
Pn = Snd (f )df = df
∫−fm ∫−fm A2c
2No fm3
= (8.100)
3A2c
Telegram: @Persian_IEEEComSoc

The ratio between Eqs. (8.98) and (8.100) gives a signal-to-noise ratio (SNR) at demodulator output as
Ps 1.5𝛽 2 A2c
(SNR)o = = (8.101)
Pn No R2 fm
The processing gain Gp of an analogue demodulator was introduced in Section 6.5.1 as the amount in dB by
which the demodulator raises the SNR at its output above the C/N at its input. Thus, Eqs. (8.97) and (8.101) yield
(SNR)o 1.5𝛽 2 A2c 2No B
Gp = = ×
C∕N No R2 fm A2c
2
3𝛽 B
= 2 (8.102)
R fm
 592 8 Frequency and Phase Modulation

Setting B equal to Carson’s bandwidth 2(𝛽 + 1)f m in the above equation yields

Gp = 6𝛽 2 (𝛽 + 1)∕R2
= 7.78 + 10 log10 (𝛽 2 (𝛽 + 1)) − 20 log10 R dB
√
= 4.77 + 10 log(𝛽 2 (𝛽 + 1)) dB (for sinusoid where R = 2) (8.103)

We see that, provided C/N is above about 9.5 dB (to justify our assumption that carrier amplitude is much larger
than noise voltage), FM provides a mechanism for raising signal quality (i.e. increasing processing gain and hence
output SNR) by using a higher modulation index and hence transmission bandwidth. Also, we see in Figure 8.47
and Eq. (8.99) that the noise vnd (t) at the discriminator output is coloured noise in that its PSD Snd (f ) is not flat but
increases as the square of frequency f . Noise power can therefore be reduced, and hence SNR increased, by using
an LPF of transfer function H d (f ) to attenuate the high-frequency components of vnd (t). To avoid distorting the
message signal by this action, we boost the high-frequency components of the message signal at the transmitter
using a highpass filter (HPF) of transfer function H p (f ) = 1/H d (f ). The filtering operation at the transmitter is
known as pre-emphasis and the inverse operation at the receiver is known as de-emphasis, as discussed in Section
8.7.3. This combined technique of pre-emphasis and de-emphasis is often simply referred to as de-emphasis. It
delivers an increase in SNR of 5–10 dB for audio transmission and around 9 dB for television (TV) and is therefore
always implemented in FM systems.
The processing gain of an FM-TV demodulator is given by

Gp = 4.77 + 10 log10 (𝛽 2 (𝛽 + 1)) + P + Q dB (8.104)

In the above equation, the first two terms give Gp for a sinusoidal message signal, P ≈ 9 dB gives the improvement
due to de-emphasis, and Q ≈ 8 dB is the subjective improvement factor, √which arises from a combination of two
factors. First, the peak-to-rms ratio R of a video signal is less than the 2 value applicable to a sinusoid. Second,
the effect on a display screen of the coloured noise at the FM demodulator output is less annoying than the effect
of white noise, so we may reduce coloured noise power by several dB to reflect our higher subjective tolerance of
its impact.
The processing gain of an FM system carrying a single voice channel is given by

Gp = 4.77 + 10 log10 (𝛽 2 (𝛽 + 1)) + P dB

P ≈ 7 dB (8.105)

Worked Example 8.8

Determine the processing gain Gp of an FM demodulator for the following FM transmissions with modulation
index 𝛽 = 5. Assume a de-emphasis improvement factor P = 7 dB where applicable.
Telegram: @Persian_IEEEComSoc

(a) Sinusoidal message signal.

(b) Message signal having a uniform probability density function (PDF).
(c) Voice signal having peak-to-rms ratio 20log10 R = 9 dB.

The applicable relationship for the processing gain Gp in all three cases is

Gp = P + 7.78 + 10 log10 (𝛽 2 (𝛽 + 1)) − 20 log10 R

= P + 7.78 + 10 log10 (52 (5 + 1)) − 20 log10 R
= P + 29.54 − 20 log10 R dB
 8.8 Noise Effect in FM 593

√
(a) A sinusoidal message signal Am cos(2𝜋fm t) has peak value Ap = Am and rms value Arms = Am ∕ 2, and hence
peak-to-rms ratio
Ap Am √
R= = √ = 2
Arms Am ∕ 2
De-emphasis improvement is not possible for a message signal that contains only one frequency component.
Thus, P = 0 for a sinusoidal message and hence
√
Gp = P + 29.54 − 20 log10 R = 0 + 29.54 − 20 log10 ( 2)
= 29.54 − 10 log10 2
= 26.53 dB
(b) A signal X having a uniform PDF and taking on continuous values x in the range −Ap to Ap has PDF pX (x)
and mean square value A2rms obtained from Eqs. (3.44) and (3.21), respectively, as
Ap
1
pX (x) = ; A2rms = x2 pX (x)dx
2Ap ∫−Ap
Thus
pA
1
A2rms = x2 dx = A2p ∕3
2Ap ∫−Ap
√ √
Therefore, this signal has peak value Ap and rms value Arms = Ap ∕ 3, and hence peak-to-rms ratio R = 3.
With de-emphasis improvement factor P = 7 dB, the FM demodulator processing gain for this signal is thus
√
Gp = P + 29.54 − 20 log10 R = 7 + 29.54 − 20 log10 ( 3)
= 36.54 − 10 log10 3
= 31.77 dB
(c) For the voice signal, we insert the given value of R along with P = 7 dB into the above relation for Gp to obtain
Gp = P + 29.54 − 20 log10 R = 7 + 29.54 − 9
= 27.54 dB

Worked Example 8.9

A TV signal with a baseband video bandwidth of 4.2 MHz is transmitted by FM. Determine the transmission
bandwidth that must be used to ensure that an incoming signal with C/N = 9.5 dB at the receiver’s demodulator
Telegram: @Persian_IEEEComSoc

input is recovered to meet the minimum SNR quality threshold of 45 dB for video at the demodulator output.
Assume de-emphasis and subjective improvement factors P = 9 dB, Q = 8 dB, respectively.
Comment on the recovered signal quality during a rain event in which the C/N at the receiving station drops
to 8 dB.

Since SNR = Gp + C∕N, and we are given SNR = 45, C/N = 9.5, it means that we need to achieve a processing
gain Gp = 45–9.5 = 35.5 dB. It follows from Eq. (8.104) that
10 log10 (𝛽 2 (𝛽 + 1)) = Gp − 4.77 − (P + Q) = 35.5 − 4.77 − 17 = 13.73
𝛽 2 (𝛽 + 1) = 1013.73∕10 = 23.6048
𝛽 = 2.571
 594 8 Frequency and Phase Modulation

The required (Carson’s) transmission bandwidth is therefore

B = 2(𝛽 + 1)fm = 2(2.571 + 1) × 4.2 MHz
= 30 MHz
If C/N of the received signal at demodulator input drops to 8 dB, the message signal will be irrecoverable by
the demodulator since this C/N is below the required minimum threshold of C/N = 9.5 dB for FM demodulation.
Such a rain event will therefore drive the communication link into outage.

8.9 Overview of FM and PM Features

Let us conclude our study of angle modulation with a discussion of its merits and demerits when compared to
amplitude modulation (studied in Chapter 7). The major applications of FM and PM are also presented.

8.9.1 Merits
(i) WBFM gives a significant improvement in the signal-to-noise power ratio (SNR) at the output of the receiver.
The FM demodulator has a processing gain that increases with modulation index. A further improvement
in SNR can be obtained using pre-emphasis (to boost the amplitude of high-frequency components of the
modulating signal) and de-emphasis (to remove the pre-emphasis distortion from the demodulated signal).
(ii) Angle modulation is resistant to propagation-induced selective fading, since amplitude variations are unim-
portant and are removed at the receiver using a limiting circuit.
(iii) Angle modulation is very effective in rejecting interference in the same manner that it minimises the effect
of noise. The receiver locks onto the wanted signal and suppresses the interfering signal, provided it is not
nearer in strength to the wanted signal than the capture ratio. A capture ratio of, say, 5 dB means that the
receiver suppresses any signal that is weaker than the wanted signal (to which it is tuned) by 5 dB or more. A
small capture ratio is desirable.
(iv) Angle modulation allows the use of more efficient transmitters. The FM signal is generated with low-level
modulation. Highly efficient nonlinear Class C amplifiers are then employed to produce a high-power RF
signal for radiation. These amplifiers can be optimally operated since the angle modulated signal is of fixed
amplitude. There is no need for the high-power audio amplifiers used in high-level AM transmitters, or the
inefficient linear RF amplifiers required to preserve the information-bearing RF envelope in low-level AM
transmitters.
(v) Angle modulation can handle a greater dynamic range of modulating signal than AM without distortion by
using a large enough frequency sensitivity to translate all message signal variations to a proportionate and
significant carrier frequency variation. The penalty is an increase in transmission bandwidth.
Telegram: @Persian_IEEEComSoc

8.9.2 Demerits
(i) The most significant disadvantage of angle modulation is that it requires a transmission bandwidth that is
much larger than the message signal bandwidth, depending of course on the modulation index. For example,
in FM audio broadcasting, a 15 kHz audio signal requires a bandwidth of about 200 kHz.
(ii) The interference rejection advantage of angle modulation can be detrimental in, for example, mobile receivers
near the edge of a service area, where the wanted signal may be captured by an unwanted signal or noise
voltage. If the two signals are of comparable amplitude, the receiver locks intermittently onto one signal or
the other – a problem known as the capture effect.
(iii) Angle modulation generally requires more complex and expensive circuits than AM. However, with advances
in IC technology this is no longer a very significant demerit.
 Questions 595

8.9.3 Applications
(i) Audio broadcasting within the VHF band at frequencies from 88 to 108 MHz. The audio signal occupies
the frequency band 50 Hz to 15 kHz, and the allowed transmission bandwidth is 200 kHz with a rated sys-
tem deviation F D = 75 kHz. In the United States, noncommercial stations are assigned carrier frequencies
in increments of 200 KHz from 88.1 to 91.9 MHz, and commercial stations from 92.1 to 107.9 MHz. The
implementation of FM stereo is discussed in Chapter 7 under applications of double sideband (DSB) (Section
7.7.1.4).
(ii) Transmission of accompanying sound in analogue television broadcast using VHF frequencies 54–88 MHz
and 174–216 MHz. An audio signal of baseband 50 Hz to 15 kHz frequency modulates a carrier with a rated
system deviation F D = 25 kHz. This application is becoming globally obsolete because of the switchover from
analogue TV broadcast to digital around the world, the UK, for example, completing that exercise back in
October 2012.
(iii) Two-way mobile radio systems transmitting audio signals of bandwidth 5 kHz, with rated system deviations
F D = 2.5 and 5 kHz for channel bandwidths 12.5 and 25 kHz, respectively. Various frequency bands have been
assigned to different services, such as amateur bands at 144–148 MHz and 420–450 MHz, and public service
bands at 108–174 MHz.
(iv) Multichannel telephony systems. Long-distance telephone traffic is carried on analogue point-to-point ter-
restrial and satellite links using FM. Several voice channels are stacked in frequency to form a composite
frequency division multiplex (FDM) signal, which is used to frequency modulate a carrier at an IF of, say,
70 MHz. The resulting FM signal is then up converted to the right UHF – for terrestrial microwave links or
super high frequency (SHF) – for satellite links. This is an analogue transmission technique which is now
largely obsolete, having been gradually replaced starting in the 1980s by digital telephony.
(v) PM and various hybrids are used in modulated digital communication systems, such as the public switched
telephone networks and terrestrial wireless and satellite communication systems.

8.10 Summary
This now completes our study of analogue signal modulation techniques started in the previous chapter. In this
chapter, we have covered in detail the principles of phase and FMs and emphasised throughout our discussions
the relationships between these two angle modulation techniques. Various angle modulation and demodulation
circuits were discussed. FM may be generated using a PM modulator that is preceded by a suitable LPF or integrator
and demodulated using a PM demodulator followed by a suitable HPF or differentiator. Similarly, PM may be
generated by an FM modulator preceded by a differentiator and demodulated using an FM demodulator followed
by an integrator. However, PM is not used for transmitting analogue message signals due to its poor bandwidth
efficiency when compared to FM. The use of PM for digital signal transmission is addressed in Chapter 11.
Telegram: @Persian_IEEEComSoc

In the next chapter, we embark on a study of the issues involved in the sampling of analogue signals – an essential
step in the process of analogue-to-digital conversion.

Questions
8.1 The staircase waveform shown in Figure Q8.1 modulates the angle of the carrier signal vc (t) = V c cos(1000𝜋t).
Make a sketch of the angle 𝜃 c (t) of the modulated carrier in degrees as a function of time in ms, over the
entire duration of the modulating signal, for the following cases.
(a) Frequency modulation with a frequency sensitivity kf = 0.2 kHz/V.
(b) Phase modulation with phase sensitivity kp = 𝜋/4 rad/V.
 596 8 Frequency and Phase Modulation

ʋm(t), volt Figure Q8.1 Question 8.1

4

2
3
0 t, ms
0 1 2 4
–2

ʋm(t), volt Figure Q8.2 Question 8.2

2

0 t, ms
0 1 2

–2

8.2 Repeat Question 8.1 for the case of the triangular modulating signal shown in Figure Q8.2. Compare the
final angle (at t = 2 ms) of the unmodulated carrier with the final angle of the modulated carrier and com-
ment on your result.

8.3 The message signal in Figure Q8.1 frequency modulates the carrier vc (t) = 10sin(4000𝜋t) volts using a circuit
of frequency sensitivity 500 Hz/V. Sketch the FM signal waveform. Determine the frequency deviation and
frequency swing of the modulated carrier.

8.4 Sketch the resulting PM waveform when the message signal in Figure Q8.1 phase modulates the carrier
vc (t) = 5sin(4000𝜋t) volts using a circuit of phase sensitivity 45∘ /V.

8.5 A message signal vm (t) = 2sin(10 × 103 𝜋t) volt is used to frequency modulate the carrier vc (t) = 10sin(180 ×
106 𝜋t) volt, giving a percent modulation of 60%. Determine
(a) The frequency deviation of the modulated carrier.
(b) The frequency sensitivity of the modulator circuit.
(c) The frequency swing of the modulated carrier.
(d) The modulation index.
Telegram: @Persian_IEEEComSoc

(e) The phase deviation of the modulated carrier.

8.6 A message signal vm (t) = 5sin(20 × 103 𝜋t) volt phase modulates the carrier vc (t) = 10sin(106 𝜋t) volt, giving
a phase modulation factor m = 1. Determine
(a) The phase sensitivity of the modulating circuit.
(b) The frequency deviation of the modulated carrier.

8.7 MATLAB Exercises: the triangular waveform shown in Figure Q8.2 modulates the carrier signal
vc (t) = 5sin(20 × 103 𝜋t) volts. Following the discussion and equations in Section 8.3.2, make an accurate
plot of the modulated carrier waveform for each of the following cases:
(a) Frequency modulation with sensitivity 2 kHz/V.
(b) Phase modulation with sensitivity 𝜋/4 rad/V.
 Questions 597

8.8 The display of an NBFM signal on a spectrum analyser shows three frequency components at 18, 20, and
22 kHz, with respective amplitudes 1, 10, and 1 V.
(a) What is the modulation index?
(b) Determine the minimum and maximum amplitudes of the modulated carrier and hence the percentage
variation in carrier amplitude.
(c) Compare the results obtained in (b) with the case of an AM signal that has an identical amplitude
spectrum.
(d) Draw a phasor diagram – similar to Figure 8.16, but with all sides and angles calculated – of the NBFM
signal at each of the time instants t = 0, 62.5, 125, 250, and 437.5 μs.
(e) Based on the results in (d), or otherwise, make a sketch of the NBFM waveform, as would be displayed
on an oscilloscope. Your sketch must be to scale and with clearly labelled axes.

8.9 Obtain Eqs. (8.55) and (8.56) for the Fourier series of tone modulated FM and PM signals. To do this,
you may wish to expand Eq. (8.36) using the relevant trigonometric identity and then apply the following
relations

cos[𝛽 sin(𝜔m t)] = J0 (𝛽) + 2J2 (𝛽) cos(2𝜔m t)

+2J4 (𝛽) cos(4𝜔m t) + · · ·
sin[𝛽 sin(𝜔m t)] = 2J1 (𝛽) sin(𝜔m t) + 2J3 (𝛽) sin(3𝜔m t)
+2J5 (𝛽) sin(5𝜔m t) + · · ·

8.10 A message signal of baseband frequencies 300 Hz → 5 kHz is used to frequency modulate a 60 MHz carrier,
giving a frequency deviation of 25 kHz. Determine the Carson’s and 1% transmission bandwidths.

8.11 When the carrier signal vc (t) = 10sin(106 𝜋t) volts is frequency modulated by the message signal
vm (t) = 2sin(104 𝜋t) volts, the carrier frequency varies within ±4% of its unmodulated value.
(a) What is the frequency sensitivity of the modulator?
(b) What is the modulation index?
(c) Determine the Carson’s and 1% transmission bandwidths.
(d) Sketch the amplitude spectrum of the FM signal. Include all spectral components with an amplitude
larger than 1% of the unmodulated carrier amplitude.
(e) Determine the percentage of the total power contained in the frequency band 473 → 526 kHz.

8.12 Determine the percentage of total power contained within the Carson’s and 1% bandwidths of a tone mod-
ulated FM signal with the following values of modulation index:
Telegram: @Persian_IEEEComSoc

(a) 𝛽 = 0.2
(b) 𝛽 = 2
(c) 𝛽 = 20.

8.13 The circuit of Figure 8.25 is employed to generate an NBFM signal of modulation index 𝛽 = 0.2. The message
signal is a 2 kHz sinusoid. If the required frequency sensitivity is kf = 5 kHz/V, give a specification of suitable
component values R and C and the message signal amplitude V m .

8.14 Figure Q8.14 is the block diagram of an Armstrong modulator involving two stages of frequency multiplica-
tion. The message signal vm (t) contains frequencies in the range 50 Hz to 15 kHz. The WBFM output signal
has a carrier frequency f c = 96 MHz and a minimum frequency deviation f d = 75 kHz. The NBFM modu-
lator uses a carrier frequency f c1 = 100 kHz, with a modulation index 𝛽 1 = 0.2. Determine the frequency
 598 8 Frequency and Phase Modulation

ʋm(t) fc1, β1 Frequency n f , n β fc2, n1β1 Frequency fc, β

1 c1 1 1 Down
NBFM multiplier multiplier
Message signal (×n1) converter (×n2) Wideband FM

fc1 fLO

Figure Q8.14 Question 8.14

multiplication ratios n1 and n2 , which will allow an oscillator frequency f LO = 8.46 MHz to be used in the
downconverter, with f LO > n1 f c1 .

8.15 By taking the derivative of the general expression for an FM signal – Eq. (8.32) – show that frequency
demodulation can be obtained using a circuit that consists of a differentiator followed by an envelope
demodulator. Specify the modification or extra processing block required to use this circuit for phase
demodulation.
Telegram: @Persian_IEEEComSoc
 599

Sampling

Those who write off education as too expensive may be paying more for ignorance.

In this Chapter
✓ Sampling theorem: how to sample a bandlimited analogue signal to ensure distortion-free reconstruction.
Both lowpass and bandpass signals are discussed.
✓ Aliasing: the penalty of undersampling.
✓ Anti-alias filter: considerations in the design of a lowpass filter (LPF) to limit the bandwidth of an analogue
signal prior to sampling.
✓ Non-instantaneous sampling: the use of a finite-width pulse train as switching signal for flat-top sampling
and the resulting aperture effect.

9.1 Introduction

This chapter lays an important foundation for the introduction of digital modulation techniques. The subject of
sampling is introduced as a nondestructive elimination of the redundancy inherent in analogue signal represen-
tations. The sampling theorem is presented and terminologies such as Nyquist frequency, Nyquist interval, and
aliasing are discussed.
Using sinusoidal signals, we demonstrate how sampling at an adequate rate duplicates (without distortion)
the baseband spectrum of the sampled signal. The penalty of undersampling is also illustrated in the time and
frequency domains, and the measures usually taken to minimise alias distortion are discussed with an example
Telegram: @Persian_IEEEComSoc

from telephone speech.

Instantaneous sampling is discussed, and it is shown that, although this method is highly desirable, it is impos-
sible to realise in practice. Natural and flat-top sampling using a pulse train that has a nonzero duty cycle are dis-
cussed. The distortion due to aperture effect is presented along with practical measures to minimise this distortion.

9.2 Sampling Theorem

Analogue signals are continuous both in time and in amplitude, and they require exclusive use of a communi-
cation system resource for the entire duration of transmission. However, the values of an analogue signal at two
sufficiently close time instants are usually related in some way and there is therefore inherent redundancy in such
Communication Engineering Principles, Second Edition. Ifiok Otung.
© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 600 9 Sampling

signals. For example, the values of a DC signal at all time instants are related as equal, so that only one value is
required to recover the entire signal. A signal that is a linear function of time can be fully reconstructed from its
values taken at only two time instants. The advantage of storing and transmitting only selected values of one signal
is that the system resource can be allocated to other signals during the unused intervals.
The process of taking only a few values of an analogue signal g(t) at regular time intervals is referred to as sam-
pling. Thus, sampling converts the continuous-value continuous-time signal g(t) to a continuous-value discrete-time
signal g(nT s ), where n = 0, 1, 2, 3, …, and T s is the sampling interval. The surprise is that you can perfectly recon-
struct the original signal g(t) from the samples, provided you follow a simple rule, known as the sampling theorem,
which may be stated as follows.

A band-limited lowpass signal that has no frequency components above fm (Hz) may be perfectly recon-
structed, using an LPF, from its samples taken at regular intervals at the rate Fs ≥ 2fm (samples per second,
or Hz).

There are three important points to note about the sampling theorem.

● The analogue signal must have a finite bandwidth f m . A signal of infinite bandwidth cannot be sampled without
distortion. In practice, it is necessary to artificially limit the bandwidth of the signal using a suitable LPF.
● At least two samples must be taken during each period T m of the highest-frequency sinusoid in the signal. That
is, the sampling interval T s must be less than or equal to half T m , which is the same as stated above that the
sampling rate F s must be at least twice f m .
● When the above two conditions are satisfied then the original signal g(t) can be recovered from these samples,
with absolutely no degradation, by passing the sampled sequence g[nT s ] = {g(0), g(T s ), g(2T s ), g(3T s ), …} through
an LPF. The filter used in this way is referred to as a reconstruction filter.

The minimum sampling rate F smin specified by the sampling theorem, equal to twice the bandwidth of the
analogue signal, is called the Nyquist rate or Nyquist frequency. The reciprocal of the Nyquist rate is referred to as
the Nyquist sampling interval, which represents the maximum sampling interval T smax allowed by the sampling
theorem.

Nyquist rate = Fsmin = 2fm

1 1
Nyquist interval = Tsmax = = (9.1)
Nyquist rate 2fm

9.3 Proof of Sampling Theorem

Telegram: @Persian_IEEEComSoc

Rather than attempt a formal mathematical proof of the sampling theorem, we will follow an intuitive discussion
that highlights the conditions for distortion-free sampling stated in the theorem. The sampling of an arbitrary
analogue signal g(t) can be obtained as shown in Figure 9.1a. The switching signal 𝛿 Ts (t) is an impulse train of
period T s , which causes the electronic switch to close and open instantaneously at intervals of T s . The result is a
sampled signal g𝛿 (t) that is equal to the analogue signal g(t) at the instants t = nT s , n = 0, 1, 2, 3, …, when the
switch is closed and is zero everywhere else. This type of sampling is therefore referred to as instantaneous or ideal
sampling. The waveforms g(t), 𝛿 Ts (t), and g𝛿 (t) are shown in Figure 9.1b–d. We see that the sampled signal may be
expressed as the product of the continuous-time analogue signal g(t) and a switching signal 𝛿 Ts (t)

g𝛿 (t) = g(t) × 𝛿Ts (t) (9.2)

 9.3 Proof of Sampling Theorem 601

Electronic switch
Analogue g(t)
gδ(t) Sampled
(a) input
output

δT (t) = Σ δ(t – nTs)

s n
Switching
signal ≡ Impulse train of period Ts = 1/Fs
g(t)

(b)
t

δT (t) = Σ δ(t – nTs)

s n

(c)
t
Ts
gδ(t)

(d)
t
Ts

Figure 9.1 Instantaneous sampling of analogue signal: (a) switch; (b) analogue signal; (c) switching impulse train;
(d) instantaneously sampled signal.

Note that the switching signal 𝛿 Ts (t) is a periodic rectangular pulse of infinitesimally small duty cycle d, and of
period T s . It can therefore be expressed in terms of the Fourier series
( )
𝛿Ts (t) = Ao + A1 cos[2𝜋Fs t] + A2 cos[2𝜋 2Fs t]
( )
+ A3 cos[2𝜋(3Fs )t] + A4 cos[2𝜋 4Fs t] + · · · (9.3)

where F s = 1/T s is the sampling frequency. From the discussion of the amplitude spectrum of a rectangular pulse
train in Section 4.2 (see, for example, Worked Example 4.1), we recall that the amplitudes of the harmonic fre-
quency components of 𝛿 Ts (t) are given by

Ao = Ad
An = 2Ad sinc (nd) , n = 1, 2, 3, · · · (9.4)
Telegram: @Persian_IEEEComSoc

where A is the amplitude of the pulse train, and the first null in the spectrum occurs at frequency f = F s /d. Thus,
as d → 0, the sinc envelope flattens out and the first null occurs at f → ∞, so that the amplitude An of the harmonic
frequency components become

Ao = Ad
An = 2Ad, n = 1, 2, 3, · · · (9.5)

Normalising the factor Ad to unity, and substituting Eq. (9.5) in (9.3), we obtain the normalised Fourier series
of an impulse train as

𝛿Ts (t) = 1 + 2 cos[2𝜋Fs t] + 2 cos[2𝜋(2Fs )t]

+ 2 cos[2𝜋(3Fs )t] + 2 cos[2𝜋(4Fs )t] + · · · (9.6)
 602 9 Sampling

Substituting in Eq. (9.2) yields an alternative expression for the instantaneously sampled signal g𝛿 (t)

g𝛿 (t) = g(t){1 + 2 cos[2𝜋Fs t] + 2 cos[2𝜋(2Fs )t] + · · ·}

= g(t) + 2g(t) cos[2𝜋Fs t] + 2g(t) cos[2𝜋(2Fs )t] + · · ·
∑
∞
= g(t) + 2 g(t) cos(2𝜋nF s t) (9.7)
n=1

Equation (9.7) is an important result, which shows that the instantaneously sampled signal g𝛿 (t) is the sum of
the original signal g(t) and the product of 2 g(t) and an infinite array of sinusoids of frequencies F s , 2F s , 3F s , …
Recall that the double-sided spectrum of 2g(t) cos(2𝜋nF s t) is merely the spectrum G(f ) of g(t) shifted without
modification from the location f = 0 along the frequency axis to the locations f = ±nF s .

9.3.1 Lowpass Signals

Let the original signal g(t) have the spectrum G(f ) shown in Figure 9.2a. The shape of G(f ) shown is merely
representative and is unimportant. What matters is that g(t) is bandlimited with bandwidth f m . Therefore, the
double-sided spectrum of g𝛿 (t) is as shown in Figure 9.2c. The normalised spectrum of the switching impulse
train is also shown in Figure 9.2b.

|G(f)|

(a)

f
–fm fm

ΔT (f)
s

….. …..

(b)

f
–3Fs –2Fs –Fs 0 Fs 2Fs 3Fs
Telegram: @Persian_IEEEComSoc

Gδ(f)
….. …..
(c)

f
–3Fs –2Fs –Fs –fm fm Fs 2Fs 3Fs
0
Fs – fm

Figure 9.2 Instantaneous sampling of analogue signal. Spectra of the following signals: (a) analogue signal; (b) switching
impulse train; (c) sampled signal.
 9.3 Proof of Sampling Theorem 603

Note that (instantaneous) sampling does not distort the spectrum G(f ) of a bandlimited signal g(t). Rather, it
replicates G(f ) at intervals of the sampling frequency F s . We see from Figure 9.2c that the lowest frequency com-
ponent in the first band replicated at F s is F s – f m , whereas the highest-frequency component in the baseband is f m .
So, if F s − f m ≥ f m (i.e. F s ≥ 2f m ) then the first replicated band does not overlap into the baseband, and neither does
any of the other replicated bands overlap with another. Thus, any one of these bands, and hence the original signal
g(t), can be recovered without distortion by employing a realisable filter. An LPF with a cut-off frequency f m will
recover g(t) by passing the baseband (or zeroth band) in G𝛿 (f ), corresponding to the first term in the right-hand
side of Eq. (9.7), and blocking all other replicated bands. We see therefore that a bandlimited signal of bandwidth
f m can be recovered without distortion from its samples taken at a rate F s ≥ 2f m . This is a statement of the sampling
theorem.

9.3.2 Bandpass Signals

A lowpass signal has been assumed so far. However, the sampling theorem is also applicable to a bandpass sig-
nal gbp (t) of bandwidth B that contains positive frequencies centred on f c > > 0, as shown in Figure 9.3a. The
highest-frequency component of gbp (t) is f m and the lowest frequency component is f L . According to the sam-
pling theorem, the spectrum |Gbp𝛿 (f )| of the sampled signal results from replications (at regular frequency spacing
F s ) of the spectrum |Gbp (f )|of gbp (t). Figure 9.3a shows that the spectrum of gbp (t) lies in the positive frequency
band (fL , fm ) and corresponding negative frequency band (−fm , −fL ). The nth replicated band within Gbp𝛿 (f ) is at
locations

Bandn = (nF s − fm , nF s − fL ), (nF s + fL , nF s + fm )

n = ±1, ±2, ±3, · · · (9.8)

where positive n specifies the positive band and negative n specifies the corresponding negative band.
If we apply the sampling theorem as earlier stipulated for lowpass signals then the original bandpass signal can
obviously be recovered from samples of gbp (t) taken at the rate F s ≥ 2f m . The resulting spectrum of the sampled
signal is shown in Figure 9.3b for F s = 2f m . Only the first replicated band is shown in Figure 9.3b, where the positive
bands (F s − f m , F s − f L ) and (F s + f L , F s + f m ) have been labelled 1 and 1′ , respectively, and their corresponding
negative bands are labelled −1 and −1′ , respectively. Band 1 at (F s − f m , F s − f L ) is the lowest replicated positive
band. Substituting F s = 2f m , we see that this band is at location (2f m − f m , 2f m − f L ) = (f m , f m + B), which just
avoids any overlap into the original spectrum Gbp (f ) at (f L , f m ) shown in Figure 9.3a. A bandpass filter (BPF) of
bandwidth B centred at f c (having gain response as shown in dotted outline in Figure 9.3b) will therefore recover
Gbp (f ) and hence the bandpass signal gbp (t) from the sampled sequence. It is worth noting that here the required
reconstruction filter is a BPF, whereas the filter employed to reconstruct baseband signals is lowpass, as discussed
in the previous section.
Telegram: @Persian_IEEEComSoc

The biggest drawback of sampling a bandpass signal at the rate F s = 2f m discussed above is that this rate can
be prohibitively high since f c is often quite large, sometimes many orders of magnitude larger than the signal
bandwidth B. Observing that large swathes of Gbp (f ) are empty, with occupancy only within the two bands (f L , f m )
and (−f m , −f L ), distortion-free reconstruction ought to be possible at a much lower sampling rate if we can find
one that simply ensures that none of the replicated bands of Eq. (9.8) overlaps into any of these two original bands.
This is indeed the case for a sampling rate F s in the range
⌊ ⌋
2fm 2fL f
≤ Fs ≤ , 1≤p≤ m (9.9)
p p−1 B
where p is an integer and ⌊x⌋ denotes the integer part of a positive real number x. Let us elaborate on two special
cases of Eq. (9.9) to shed more light on its interpretation:
 604 9 Sampling

|Gbp(f)|
B = fm – fL

(a)
f
–fc 0 fc
–fm –fL fL fm
(b) Fs = 2fm |Gbpδ(f)| BPF gain response

–1′ –1 1 1 1 1′
f
–Fs –fc 0 fc Fs

(c) Fs = 2B |Gbpδ(f)|

–4′ –3′ –2′ –1′ –4 –3 1 –2 2 –1 3 4 1′ 2′ 3′ 4′

f
–4Fs –3Fs –fc –Fs 0 Fs fc 3Fs 4Fs
–2Fs 2Fs
(d) Fs = 2.8B |Gbpδ(f)| Realisable filter gain response

–3 –2 –4′ –1 –3′ –2′ 1 –1′ 2 3 1′ 4 2′ 3′

f
–4Fs –3Fs –2Fs –fc –Fs 0 Fs fc 2Fs 3Fs 4Fs

Figure 9.3 Instantaneous sampling of bandpass signal gbp (t) of bandwidth B centred on f c and located at (f L , f m ), where
f L = 3B: (a) representative spectrum of gbp (t); (b)–(d) spectra of sampled signal using various sampling rates F s .

● If f m = B, meaning that the signal is a baseband (i.e. lowpass) signal, which by definition has bandwidth B equal
to its highest-frequency component f m , then p = 1 and the applicable range of sampling rates is

2fm 2fL
≤ Fs ≤ ; ⇒ 2fm ≤ Fs ≤ ∞;
1 1−1
⇒ Fs ≥ 2fm
Telegram: @Persian_IEEEComSoc

which is simply the sampling theorem stated earlier for baseband signals and agrees with a minimum sampling
rate F smin = 2f m as expected.
● If f L = qB, where q is an integer, it means that the positive frequency range f L to f m of gbp (t) starts at a location
which is an integer multiple of bandwidth B. Note that q = 0 corresponds to a start at the origin, which is the
baseband case discussed above, so here we consider only the bandpass cases where q ≥ 1. Substituting f L = qB
and f m = qB + B = (q + 1)B into Eq. (9.9) yields the applicable range of sampling rates as

2(q + 1)B 2qB

≤ Fs ≤ , 1≤p≤q+1
p p−1

Let us take a moment to examine the possibilities for F s . First, when q = 1, there are only two possible values
for p, namely p = 1 and p = 2, which respectively yield the following ranges for F s
 9.3 Proof of Sampling Theorem 605

4B 2B
q = 1, p = 1 ∶ ≤ Fs ≤ ; ⇒ Fs ≥ 4B
1 1−1
4B 2B
q = 1, p = 2 ∶ ≤ Fs ≤ ; ⇒ Fs = 2B
2 2−1
Thus when q = 1 (which means that f L = B and f m = 2B) then alias-free sampling is possible at the specific rate
F s = 2B or at any rate greater than or equal to 4B. The latter option (F s ≥ 4B) corresponds to sampling at a rate that is
at least twice the maximum frequency component of the bandpass signal. This is explored earlier and is illustrated
in Figure 9.3b. So it is the first option (F s = 2B) that delivers a saving in sampling rate by taking advantage of the
unoccupied frequency region f < f L to allow sampling to replicate the band (f L , f m ) at a regular spacing F s < 2f m
without any overlap into ±(f L , f m ).
Next, examining the situation for q = 2 (when p can take on values 1, 2, and 3), q = 3 (when p can be 1, 2, 3, 4)
and so on, we find that all cases of integer band positioning (i.e. f L = qB, q = 0, 1, 2, 3, …) allow alias-free sampling
(i.e. no overlap between image and original bands) at a minimum rate F s = 2B. In addition, there are windows of
other nonminimum rates F s > 2B that support alias-free sampling. For example, for a bandpass signal in which the
lowest-frequency component f L is three times its bandwidth (i.e. q = 3 and f m = 4B), there are four increasingly
narrower windows from which F s can be selected for alias-free sampling, namely
Window 1 ∶ Fs ≥ 8B
Window 2 ∶ 4B ≤ Fs ≤ 6B
8
Window 3 ∶ B ≤ Fs ≤ 3B
3
Window 4 ∶ Fs = 2B
In general, for all values of q, the smallest value of p (= 1) gives the sampling rates window F s ≥ 2f m , the largest
value of p (= q + 1) gives the minimum sampling rate F s = 2B, and the values of p in between gives a total of q − 1
windows of allowed sampling rates between 2B and 2f m . Figure 9.3c and d illustrate alias-free sampling at a rate
of F s = 2B and F s = 2.8B, respectively, for a bandpass signal in which f L = 3B. The pair of image bands replicated
at nF s is labelled n and n’ in the figures. Notice in Figure 9.3c how the first three replicated bands fall below f L ,
whereas the fourth (and higher) replicated bands fall above f m , and none of them overlaps into the original band
(f L , f m ). This original band can therefore be extracted using a BPF to reconstruct the original signal gbp (t) without
any distortion. A nonminimum rate (such as F s = 2.8B in this example) allows a realisable BPF to be used as
illustrated in Figure 9.3d.
It is important to recognise the fact that allowed sampling rates for a bandpass signal fall in disjoint windows.
This means that one cannot increase the sampling rate willy-nilly above an allowed minimum value in an attempt,
for example, to insert or increase a gap (called guard band) between the original and the replicated bands to per-
mit the use of realisable reconstruction filters. The following general guidelines should be borne in mind when
sampling a bandpass signal of bandwidth B and frequency band (f L , f m ).
Telegram: @Persian_IEEEComSoc

● You cannot sample at any rate less than 2B, otherwise there will be alias distortion.
● You can sample at any rate ≥ 2f m without alias distortion.
● You can use the theoretical minimum rate F s = 2B only if the passband is located at an integer number of
bandwidths from the origin, i.e. if f m = (q + 1)B, where q is a positive integer.
● There are some rates between 2B and 2f m which will allow alias-free sampling, but these rates must satisfy
Eq. (9.9). Any rate not satisfying this equation will result in an overlap of a replicated band into the original
band thereby causing alias distortion.
● To provide a transition band Δf on either side of (f L , f m ) for a realisable reconstruction filter, apply Eq. (9.9) to the
augmented band (f L − Δf /2, f m + Δf /2) and determine the minimum allowed sampling rate for this augmented
band. When the true signal of band (f L , f m ) is sampled at the minimum rate so determined, there will be a
guaranteed guard band ≥ Δf on either side of (f L , f m ) separating it from adjacent image bands.
 606 9 Sampling

The method of bandpass sampling described above is known as uniform bandpass sampling. Another method
of bandpass sampling, known as quadrature sampling, allows the use of any rate F s ≥ 2B. For a more detailed
discussion of bandpass sampling supported by worked examples, the reader is referred to Chapter 4 of [1].
In the light of Figure 9.3, it should be emphasised that the required minimum sampling rate or Nyquist fre-
quency F smin is twice the bandwidth of the analogue signal, and not necessarily twice the maximum frequency
component of the signal. In lowpass signals, bandwidth and maximum frequency component are equal, and F smin
may be correctly expressed as twice the maximum frequency component. However, the bandwidth of bandpass
signals is typically much less than the maximum frequency component of the signal, and F smin in this case must
be expressed as twice the bandwidth.
In the rest of this chapter it will be assumed that the analogue signal g(t) is a lowpass signal. The discussion that
follows may be applied to a bandpass signal gbp (t) of centre frequency f c if the signal is first transformed into a
lowpass signal by a frequency translation of f c .

9.3.3 Sampling at Nyquist Rate

Figure 9.4 shows the result of sampling an analogue signal g(t) of bandwidth f m at the Nyquist rate 2f m . Note that
the replicated bands in the spectrum of the sampled signal touch but do not overlap. In order to recover the original
signal an LPF is required that passes the baseband spectrum containing frequencies f = 0 to f m , and completely
rejects the replicated bands at frequencies f = f m to ∞. The filter must have the ideal brickwall frequency response
of zero transition width, as shown by the dotted line in Figure 9.4c. Such a response is unrealisable in real time.

|G(f)|

(a)

f
–fm fm
|ΔT (f)|
s

(b)

f
Telegram: @Persian_IEEEComSoc

–3Fs –2Fs –Fs 0 Fs 2Fs 3Fs

|Gδ(f)| Brickwall filter
response

(c)

f
–3Fs –2Fs –Fs 0 Fs 2Fs 3Fs

Figure 9.4 Sampling at Nyquist rate F s = 2f m . Spectra of the following signals: (a) analogue signal g(t); (b) switching
impulse train; (c) sampled signal g𝛿 (t).
 9.4 Aliasing 607

Realisable reconstruction Undistorted

filter response |Gδ(f)| spectrum 1st replicated band

Fs – f m > f m
⇒ Fs > 2fm

f
–Fs –Fs + fm –fm 0 fm Fs – fm Fs

Figure 9.5 Sampling a rate of F s larger than Nyquist rate allows use of a realisable reconstruction filter.

Therefore, in practice a sampling rate higher than the Nyquist frequency is employed to allow the use of a
realisable reconstruction filter having a finite transition width. The frequency response of the reconstruction filter
is shown in dotted lines in Figure 9.5. It has the following specifications
Pass band: 0 ≤ f ≤ fm
Transition band: fm ≤ f ≤ Fs − fm
Stop band: Fs − fm ≤ f < ∞ (9.10)
Let us now consider what happens if the sampling theorem is flouted by reducing the sampling frequency below
the Nyquist rate.

9.4 Aliasing
An analogue signal g(t) sampled at less than the Nyquist rate is said to be undersampled. The replicated bands
in the spectrum G𝛿 (f ) of the sampled signal overlap, as shown in Figure 9.6, giving a resultant spectrum that is
no longer an exact replica of the original spectrum G(f ). The baseband spectrum in G𝛿 (f ) – the shaded region of
Figure 9.6 – is clearly distorted. The original signal g(t) can no longer be recovered from the sampled signal g𝛿 (t)
even with an ideal LPF.
This distortion resulting from undersampling is known as alias distortion because every frequency component
f h in the original signal that is higher than half the sampling frequency F s appears in the sampled signal at a false

Distorted
resultant Overlapping
|Gδ(f)|
Telegram: @Persian_IEEEComSoc

spectrum replicated
bands
Fs – fm < fm
⇒ Fs < 2fm

f
–3Fs –2Fs –Fs –fm 0 fm Fs 2Fs 3Fs
Fs – fm

Figure 9.6 Aliasing distortion due to undersampling a rate of F s less than Nyquist rate.
 608 9 Sampling

or alias frequency
fa = |Fs − fh | (9.11)
To understand how the alias frequency is produced, let us consider the sampling of a sinusoidal signal g(t) of
frequency f m = 4 kHz. Figure 9.7a shows samples of the sinusoid taken at a rate F s = 12 kHz, which satisfies the
sampling theorem. You may observe in Figure 9.7b,c that this sequence of samples, denoted g𝛿 (t), not only fits the
original sinusoid f m but also exactly fits and replicates an infinite set of sinusoids at frequencies
Fs ± fm , 2Fs ± fm , 3Fs ± fm , ···
This means that g𝛿 (t) contains the set of frequency components
fm , Fs − fm , Fs + fm , 2Fs − fm , 2Fs + fm , 3Fs − fm , 3Fs + fm , · · ·
Thus, when g𝛿 (t) is passed through a lowpass reconstruction filter as specified by Eq. (9.10), the lowest-frequency
component f m is extracted, thereby recovering the original sinusoid without distortion. Figure 9.8 gives a plot
of the spectrum of g𝛿 (t) showing the above frequency components and the response of a reconstruction filter that
would extract the original sinusoid from the sampled sequence. It is worth pointing out that for this arrangement to
work without any distortion the reconstruction LPF must pass only frequency f m and block all the other frequency
components in the above set, starting from F s − f m . This requires that F s − f m must be no lower than f m . That is
Fs − fm ≥ fm ; ⇒ Fs ≥ 2fm (9.12)
which is a statement of the sampling theorem given earlier.
Let us consider what happens when the sinusoid is sampled as shown in Figure 9.9 a rate of F s = 6 kHz, which
is less than the Nyquist rate (= 8 kHz). Now the sequence of samples is so widely spaced that it also exactly fits

Sinusoid g(t) at fm = 4 kHz

Sampled sinusoid gδ(t)
at Fs = 12 kHz
0.5
(a) t (ms)

Sinusoid at Fs + fm = 16 kHz Sinusoid at Fs – fm = 8 kHz

0.5
(b) t (ms)
Telegram: @Persian_IEEEComSoc

Sinusoid at 2Fs + fm = 28 kHz Sinusoid at 2Fs – fm = 20 kHz

0.5
(c) t (ms)

Figure 9.7 The samples g𝛿 (t) of a sinusoidal signal g(t) of frequency f m taken a rate of F s larger than Nyquist rate will fit
(i.e. replicate) an infinite set of sinusoids f m , F s ± f m , 2F s ± f m , 3F s ± f m , . . . .
 9.4 Aliasing 609

|G(f)|

(a)

f
–fm fm

1st replicated 2nd replicated

LPF recovers original sinusoid fm |Gδ(f)| band band

(b) … …

f
–2Fs –Fs –fm fm Fs – fm F Fs + fm 2Fs
s
Baseband 2Fs – fm 2Fs + fm

Figure 9.8 Amplitude spectra of (a) sinusoid g(t) of frequency f m and (b) the sinusoid sampled a rate of F s .

fm = 4 kHz Fs – fm = 2 kHz ≡ fa

0.5
t (ms)

Figure 9.9 Sampling a sinusoid of frequency f m = 4 kHz a rate of F s = 6 kHz (less than Nyquist rate). The samples also fit
(i.e. replicate) an infinite set of sinusoids having frequencies nF s ± f m , n = 1, 2, 3, … In this case, the lowest replicated
Telegram: @Persian_IEEEComSoc

frequency F s − f m = 2 kHz is lower than f m and therefore cannot be blocked by an LPF designed to pass f m , so it will
constitute an alias frequency, denoted f a .

a lower-frequency sinusoid f a = F s − f m = 2 kHz, which is shown in Figure 9.9. The sequence will of course also
fit higher-frequency sinusoids at F s + f m and nF s ± f m , n = 2, 3, 4, …, which are not shown in Figure 9.9 to avoid
clutter. Thus, in this case of undersampling, the sampled signal contains the frequency components
fa , fm , Fs + fm , 2Fs − fm , 2Fs + fm , 3Fs − fm , 3Fs + fm , · · ·
The lowpass reconstruction filter of Eq. (9.10) will therefore pass two sinusoids, namely the original sinusoid f m ,
and an alias sinusoid f a , which causes distortion. This can be seen in Figure 9.10b, which shows the spectrum of the
undersampled sinusoid. For clarity, the replicated bands have been sketched using different line patterns – solid
 610 9 Sampling

|G(f)|

(a)

f
–fm fm

|Gδ(f)|

F
LP

Alias, fa
(b) … …

f
–2Fs – fm –2Fs –2Fs + fm –Fs –fm 0 fm Fs 2Fs – fm 2Fs 2Fs + fm
–Fs – fm –Fs + fm Fs – fm Fs + fm

Figure 9.10 Amplitude spectrum |G𝛿 (f )| of a sinusoid of frequency f m that is sampled a rate of F s < Nyquist rate. Spectra of
(a) the sinusoid and (b) the undersampled sinusoid.

for the baseband, dotted for the first replicated band F s ± f m , and dashed for the second replicated band 2F s ± f m .
Note that the alias sinusoid of frequency f a arises from the first replicated band overlapping into the baseband.
The overlapping of replicated bands occurs because they have been duplicated too close to each other along the
frequency axis at intervals of (a small value of) F s .
It must be emphasised that only frequency components higher than half the sampling frequency produce an
alias. When the sampling frequency is chosen to be at least twice the maximum frequency component of the
(lowpass) analogue signal as required by the sampling theorem then no such high-frequency components exist in
the signal, and aliasing does not occur. Figure 9.10 shows aliasing caused by the overlap of only the first replicated
band into the baseband of the sampled signal’s spectrum. In general, depending on how small F s is relative to
f m , aliasing may be caused by the overlap of one or more of the replicated bands into the baseband. That is, if
kF s − f m < f m (i.e. kF s < 2f m ), it means that the kth replicated band overlaps into the baseband (0, f m ), where f m
is the maximum frequency component of the lowpass signal. As a result of this overlap, a frequency component f
in the lowpass signal in the range

kF s − fm ≤ f ≤ fm , k = 1, 2, 3, 4, · · · (9.13)

will present itself within the sampled signal sequence at an alias frequency

fa,k = |kF s − f |
Telegram: @Persian_IEEEComSoc

(9.14)

so that a kth band of alias frequencies is produced in the range

Banda,k = |kF s − fm | → fm (9.15)

inside the signal reconstructed (by the LPF) from the sampled sequence.
Figure 9.11 shows the result of sampling a 5 kHz sinusoid using sampling rate F s = 3 kHz, which clearly violates
the sampling theorem. The message is a sinusoid, so it contains only one frequency component f which is also the
maximum frequency f m . Substituting F s = 3, f m = 5 into the condition stated in Eq. (9.13), we see that it is satisfied
for k = 1, 2, and 3, since, for k = 1, 2, and 3, kF s − f m , respectively, = −2, 1, and 4 kHz, each of which is less than
f m . Thus, the first three replicated bands overlap into the baseband and produce the respective alias components
 9.4 Aliasing 611

(a) –fm fm
f, kHz
–5 0 5

Alias components at 1, 2, 4 kHz

(b) –fm fm
–3Fs – fm

–2Fs – fm

–3Fs + fm

2Fs + fm
–Fs – fm

2Fs – fm
–2Fs+fm

3Fs + fm
3Fs – fm

Fs + f m
–Fs + fm
Fs – fm

–10 –5 5 10 f, kHz
–3Fs –2Fs –Fs 0 Fs 2Fs 3Fs

Figure 9.11 Spectra of (a) Sinusoid of frequency f m = 5 kHz and (b) its sampled sequence using sampling rate F s = 3 kHz.

START k=1

k=k+1

Yes
kFs < 2fm? fa,k = kFs – fm

No
Alias frequencies fa,1, fa,2, ...

No (more) alias END

Figure 9.12 Steps for extracting all alias frequency components in a sampled sequence obtained by sampling an analogue
signal of maximum frequency f m at a sampling rate F s .

labelled in Figure 9.11b and obtained using Eq. (9.14) as

fa,1 = |Fs − fm | = |3 − 5| = 2 kHz
Telegram: @Persian_IEEEComSoc

fa,2 = |2Fs − fm | = |6 − 5| = 1 kHz

fa,3 = |3Fs − fm | = |9 − 5| = 4 kHz
Figure 9.12 provides steps in the form of a flowchart which may be followed to systematically check for and
extract all alias components in a sampled sequence. Starting with k = 1, one tests if F s < 2f m . If the sampling
theorem was obeyed in the sampling then the test outcome will be negative (i.e. ‘No’) and the process terminates
with no alias components. However, if F s < 2f m , then the test outcome is positive (i.e. ‘Yes’), indicating the presence
of an alias component (or band) located at frequency f a,k = |kF s − f m | for k = 1. The value of k is incremented by
1 and the steps are repeated using k = 2. This continues until a value of k is reached at which the test yields a
negative outcome, indicating that there are no further alias components, and the process terminates.
Aliasing is further explored in the following worked example.
 612 9 Sampling

Worked Example 9.1

An information-bearing signal g(t) has the spectrum shown in Figure 9.13a.

(a) What is the minimum sampling frequency F smin to avoid aliasing?
(b) Sketch the spectrum of the sampled signal (in the interval ± 28 kHz) when sampled at F smin . Why would
F smin not be used in practice?
(c) Sketch the spectrum of the sampled signal over the interval ± 16 kHz when g(t) is sampled at a rate
F s = 6 kHz and determine the band of alias frequencies.

(a) Minimum sampling frequency F smin required to avoid aliasing is twice the maximum frequency component
f m of the analogue signal. From Figure 9.13a, f m = 4 kHz, so that

Fsmin = 2fm = 8 kHz

(b) Figure 9.13b shows a sketch of the spectrum of the information signal when sampled at the above rate. This
minimum sampling rate is not used in practice because it would necessitate the use of an ideal brickwall
reconstruction filter to recover the original signal g(t) at the receiver. A higher sampling rate is used, which
creates a frequency gap between the replicated spectra. This gap is needed by a realisable filter to make a
transition from passband to stopband.
(c) The spectrum of the sampled signal with F s = 6 kHz is shown in Figure 9.13c. Only the first replicated band
overlaps into the baseband to create a band of alias frequencies given by Eq. (9.15), with k = 1, as

Banda,1 = |Fs − fm | → fm = 2 → 4 kHz

|G(f)|

(a)

f, kHz
–4 4

fm = 4 kHz; Fs = 8 kHz |Gδ(f)|

Telegram: @Persian_IEEEComSoc

(b)
f, kHz
–24 –16 –8 –4 0 4 8 12 16 20 24 28

fm = 4 kHz; Fs = 6 kHz |Gδ(f)| Distorted spectrum

(c)
f, kHz
–16 –12 –6 –4 0 2 4 6 8 10 12 14 16

Figure 9.13 Worked Example 9.1.

 9.5 Anti-alias Filter 613

A lowpass reconstruction filter will recover the spectrum in the shaded region of Figure 9.13c (from −4 to 4 kHz).
It is this alias band that is responsible for the distortion in this recovered spectrum, altering its shape from a linear
slope to a flat level between 2 and 4 kHz.
Two steps are usually employed to minimise alias distortion.

● Prior to sampling, the signal bandwidth is limited to a small value f m that gives acceptable quality. The LPF
employed for this purpose is therefore called an anti-alias filter.
● The filtered signal is sampled at a rate F s > 2f m . For example, in standard telephone speech transmission using
pulse code modulation (PCM), f m = 3.4 kHz and F s = 8 kHz.

9.5 Anti-alias Filter

As stated above, the anti-alias circuit is an LPF used to remove nonessential or insignificant high-frequency compo-
nents in the message signal in order, ultimately, to reduce the required transmission bandwidth. The application
and desired fidelity determine the extent to which high-frequency components are discarded. For example, in
high-fidelity compact disc audio, all audible frequencies must be faithfully recorded. This means that frequency
components up to f m = 20 kHz are retained, and a sampling rate F s > 2f m is employed. Usually F s = 44.1 kHz.
However, in telephone speech the requirements for fidelity are much less stringent and the bandwidth of the
transmission medium is very limited. Although the low frequencies (50–200 Hz) in speech signals enhance speaker
recognition and naturalness, and the high frequencies (3.5–7 kHz) enhance intelligibility, good subjective speech
quality is still possible in telephone systems with the baseband frequency limited to 300–3400 Hz. This frequency
range has been adopted as the standard telephone speech baseband. In television signals, the eye is largely insen-
sitive to the high-frequency components of the colour signal, so these frequencies may be suppressed without a
noticeable degradation in colour quality.
An anti-alias filter is required to pass the baseband signal frequencies up to f m , and ideally to present infi-
nite attenuation to higher frequencies. In this way there is no overlapping of replicated bands when the filtered
signal is sampled at a rate of F s > 2f m . In practice, we can achieve neither infinite attenuation nor brickwall
(zero-transition-width) performance but can only design a filter that has a specified minimum attenuation in the
stopband and has a small but non-negligible transition width from passband to stopband.
Let us examine the issues involved in the specification of an anti-alias filter using the worst-case situation shown
in Figure 9.14 where a baseband information signal gi (t) has a uniform spectrum and is to be limited in bandwidth
to f m before sampling. The spectrum G(f ) of the filtered signal is shown in Figure 9.14a along with the spectrum
G𝛿 (f ) of the signal obtained by sampling g(t) at a rate of F s . Figure 9.14b shows a more detailed view of G𝛿 (f ).
Observe that frequency components beyond F sb in the first replicated band cross over into the message baseband
(0 → f m ) and will cause aliasing. Thus, we must have the following specifications for the anti-alias filter.
Telegram: @Persian_IEEEComSoc

● The stopband of the filter must start at frequency F sb given by

Fsb = Fs − fm (9.16)

● In most cases, sampling is followed by quantisation to restrict the signal to a discrete set of values, as discussed
in Chapter 10. The process of quantisation inevitably introduces error, and it is therefore enough to ensure
that aliasing error in the passband is maintained at a level just below the inevitable quantisation noise. So the
minimum stopband attenuation Amin must be as indicated in Figure 9.14b

Amin = SQNR + 3 dB (9.17)

where SQNR is the signal-to-quantisation-noise-power ratio.

 614 9 Sampling

Input signal Filtered signal Sampled signal

gi(t) Anti-alias g(t) Sampler gδ(t)
(a) LPF Fs

|Gi(f)| |G(f)| |Gδ(f)|

f f f
–Fs Fs

|Gδ(f)|, dB
0
–3

(b)
SQNR Amin

f
–Fs 0 fm Fs

Fsb

Figure 9.14 Specification of an anti-alias filter.

Let the anti-alias filter be an nth order lowpass Butterworth filter. The attenuation A of this filter as a function
of frequency f is given by
[ ( )2n ]
f
A = 10 log10 1 + dB (9.18)
fm

where f m is the maximum passband frequency (i.e. the 3 dB point, which is the frequency at which the gain of the
filter is down by 3 dB from its maximum value). Since A = Amin at f = F sb , we may write

Amin = 10 log10 [1 + (Fsb ∕fm )2n ]

and solve for F sb to obtain

Telegram: @Persian_IEEEComSoc

Fsb = fm [10(Amin ∕10) − 1]1∕2n

Finally, using Eq. (9.16), we obtain an expression for the required sampling frequency F s
{ }
Fs = fm 1 + [10(Amin ∕10) − 1]1∕2n (9.19)

The above equation is an important result, which gives the minimum sampling frequency F s required to main-
tain alias frequencies at a level at least Amin (dB) below the desired baseband frequency components.
 9.6 Non-instantaneous Sampling 615

Worked Example 9.2

The analogue input signal to a 12-bit uniform quantisation PCM system has a uniform spectrum that is required
to be limited to the band 0 → 4 KHz using a sixth order Butterworth anti-alias filter. The input signal fully loads
the quantiser. In designing the filter, the goal is to maintain aliasing error in the passband at a level just below
quantisation noise.
(a) Determine the minimum stopband attenuation Amin of the anti-alias filter.
(b) Calculate the minimum sampling frequency F s .
(c) How does Fs compare with the standard sampling rate of 8 kHz used for speech?

(a) Amin is given by Eq. (9.17). It is shown in Chapter 10 that

SQNR = 1.76 + 6.02k (9.20)

where k (= 12 in this case) is the number of bits needed to identify each of the discrete signal levels in the
quantiser. Thus
Amin = 1.76 + 6.02 × 12 + 3 = 77 dB
(b) Substituting the values f m = 4 kHz, n = 6, and Amin = 77 dB in Eq. (9.19) gives the desired minimum sampling
frequency
{ }
Fs = 4 1 + [10(77∕10) − 1]1∕12 = 21.53 kHz
(c) The required sampling frequency, F s = 21.53 kHz, is rather high compared to a Nyquist rate of 8 kHz. To
see why, note that a worst-case scenario was assumed in which the input analogue signal has a uniform
amplitude spectrum that needs to be limited to the band 0 → f m using a realisable filter. The anti-alias fil-
ter requires enough transition width to reduce the signal’s spectral components within the stopband by the
required amount Amin . In practice, for example speech signals, the amplitude spectrum Gi (f ) is not uniform,
and the frequency components above f m are of very small amplitudes, requiring only a little extra attenuation
by a filter to bring them down to the level of the quantisation noise voltage. Even for signals of uniform spec-
trum, the sampling frequency required to maintain aliasing error below quantisation noise can be reduced by
using a higher-order (i.e. a steeper roll-off) filter. For example, if in the above problem we use a Butterworth
filter of order n = 12 then F s would be only 12.37 kHz. Note, however, that in many cases filters with a steeper
roll-off introduce more phase distortion. Question 9.6 deals with the design of an anti-alias filter for a signal
of nonuniform spectrum.
Telegram: @Persian_IEEEComSoc

9.6 Non-instantaneous Sampling

We have so far discussed instantaneous sampling, which is not feasible in practice for the following reasons:
● Sampling pulses must have infinite amplitude in order to carry enough signal energy. This can be seen by exam-
ining Eq. (9.4) and noting that the pulse amplitude A must be infinite if the factor Ad, and hence the spectral
components of the impulse train, are to be nonzero as d → 0.
 616 9 Sampling

● A system must have infinite bandwidth in order to sustain instantaneously sampled pulses without introducing
some distortion.
● Instantaneous switching is required, which is not attainable using physically realisable electronic devices with
their inherent capacitive and inductive elements that tend to slow down the rates of change of voltage and
current, respectively.

Let us therefore consider the practical implementation of sampling using pulses of finite width, known as
non-instantaneous sampling. We will discuss its effect on the spectrum of the sampled signal, and the distor-
tion on the reconstructed signal. There are two types of non-instantaneous sampling, namely natural and flat-top
sampling.

9.6.1 Natural Sampling

If we replace the impulse train switching signal 𝛿 Ts (t) in Figure 9.1a with the rectangular pulse train rectTs (t/𝜏) of
nonzero duty cycle d = 𝜏/T s shown in Figure 9.15b, we obtain the sampled signal gΠn (t) shown in Figure 9.15c.
Note how the top of each pulse follows the variation of the original analogue signal g(t). This scheme, in which an
input analogue signal is simply switched through to the output by a rectangular pulse train, is known as natural
sampling. The electronic switch in this case is realisable using, for example, CMOS hardware. The output or sam-
pled signal follows the variations of the analogue input during the fraction d of time that the switch is closed and
is zero during the fraction (1 − d) of time that the switch is open. The question now is whether the original signal
g(t) can be recovered from this non-instantaneously sampled signal. This question is explored in the frequency
domain below, but first, for the sake of completeness, let us draw on the work of Chapter 2 to give a mathematical

g(t)
(a)

rectT (t/τ)
s

(b) τ
d = τ/Ts

t
Telegram: @Persian_IEEEComSoc

Ts

gпn(t)
(c)

Figure 9.15 Natural sampling: (a) analogue signal; (b) switching rectangular pulse train; (c) sampled signal.
 9.6 Non-instantaneous Sampling 617

definition of the rectangular pulse train of Figure 9.15b as

( ) ∑ ( )
t t − nT s
rectTs = rect , 𝜏 = dT s
𝜏 n
𝜏
{
( ) 1, −𝜏∕2 ≤ t ≤ 𝜏∕2
t
rect = (9.21)
𝜏 0, Otherwise
Let us assume that g(t) is of bandwidth f m and that it has the amplitude spectrum shown in Figure 9.16a. The
specific shape of the spectrum is irrelevant; however, we have chosen a rectangular shape because it makes spectral
distortion more easily observable. The same arguments leading to Eq. (9.2) and Eq. (9.4) apply. However, in this
case d > 0 so that the simplification of Eq. (9.5) does not apply. Using Eq. (4.17) in Chapter 4 (with f o ≡ F s = 1/T s ),
the normalised (Ad = 1) Fourier series of the rectangular pulse train rectTs (t/𝜏) (of duty cycle d = 𝜏/T s and ampli-
tude A) and that of the naturally sampled signal gΠn (t) are given by
( ) ∑∞
t
rectTs =1+2 sinc(nd) cos(2𝜋nF s t) (9.22)
𝜏 n=1
( )
t
gΠn (t) = g(t) × rectTs
𝜏
∑∞
= g(t) + 2 g(t) sinc(nd) cos(2𝜋nF s t) (9.23)
n=1

|G(f)|

(a)

f
–fm fm

RT (f)
s

(b)

–5Fs 5Fs
f
–4Fs –3Fs –2Fs –Fs –fm 0 fm Fs 2Fs 3Fs 4Fs
Telegram: @Persian_IEEEComSoc

GΠn(f)

(c)
d = ¼; Fs = 2.5fm

–5Fs 5Fs
f
–4Fs –3Fs –2Fs –Fs –fm 0 fm Fs 2Fs 3Fs 4Fs

Figure 9.16 Natural sampling with rectangular pulse train of duty cycle 1/4. Spectrum of (a) analogue signal, (b) switching
rectangular pulse train, and (c) sampled signal.
 618 9 Sampling

We see that the naturally sampled signal gΠn (t) is the sum of the original signal g(t) and the product of 2 g(t) and
an infinite array of sinusoids of frequencies nF s , n = 1, 2, 3, …, each product scaled by the factor sinc(nd). This is
an interesting result, which shows that the only difference between natural sampling and instantaneous sampling
is that the nth replicated band in the sampled spectrum is reduced in size (but not distorted) by the factor sinc(nd).
The spectrum of gΠn (t) is shown in Figure 9.16c for d = 1/4.
You will recall from Section 2.6.8 that sinc(nd) = 0 whenever nd is an integer (±1, ±2, ±3, …). This means that
the following replicated bands will be scaled by a zero factor and therefore will be missing from the spectrum of
gΠn (t)
1 2 3 4
n= , , , , ···
d d d d
In Figure 9.16 with d = 1/4, the missing replicated bands are the 4th, 8th, 12th, … The original signal g(t) can be
recovered from gΠn (t) using an LPF as discussed earlier, provided the sampling frequency F s ≥ 2 f m , as specified
by the sampling theorem.
Natural sampling is rarely used in practice because it places a severe limitation on the maximum frequency
f m that can be accurately digitised after sampling. It can be shown that if the quantiser (that follows the natural
sampler) has a conversion time 𝜏, and the desired conversion accuracy is half the quantiser step size then the
highest frequency that can be digitised is
1
fm = (9.24)
2 𝜋𝜏
k

where k is the number of bits per sample of the quantiser. Note that there is a limitation on f m whenever
the conversion time 𝜏 is nonzero, as is usually the case. For example, with 𝜏 = 0.1 μs and k = 12, we have
f m = 777 Hz.

9.6.2 Flat-top Sampling

If the top of each pulse is maintained constant or flat with a height equal to the instantaneous value of the analogue
signal at the beginning of the pulse then we have what is known as flat-top sampling. This is the usual method of
sampling implemented by an integrated circuit referred to as sample-and-hold, the operation of which is shown
in Figure 9.17 in terms of a block diagram in (a), circuit implementation in (b), and illustrative input and output
waveforms in (c). In Figure 9.17b, switch S1 is normally open and S2 is normally closed. A rectangular pulse train
rectTs (t/𝜏) of switching period T s , pulse duration 𝜏, and duty cycle d = 𝜏/T s is used as the switching signal to actuate
the two switches. S1 is turned on momentarily by the positive-going edge of the pulse, while S2 is turned off at
the same instant and held off for the entire duration 𝜏 of the pulse. The capacitor C charges rapidly (through the
negligible resistance path of ‘S1 on’) to the value of g(t) at the instant that S1 is turned on and holds that voltage for
the entire pulse duration 𝜏 (during which S2 is off) since it has negligible discharge through the high-resistance
Telegram: @Persian_IEEEComSoc

path of ‘S2 off’. At the end of the pulse, S2 returns to its normally closed state and C discharges rapidly through
S2 to zero and remains there until the next pulse causes S1 to turn momentarily on thereby charging C to the
voltage level of g(t) at this next sampling instant. This carries on repeatedly so that the capacitor voltage gives a
flat-top-sampled version of g(t). The input signal g(t) is usually connected to S1 through an operational amplifier
(opamp, not shown) and the output gΠ (t) is taken from across the holding capacitor C through another opamp
(also not shown).
Figure 9.17c shows an illustrative analogue waveform and the resulting flat-top-sampled output for d = 1/4.
Again, we must ask whether the original signal g(t) can be extracted from this flat-top-sampled signal gΠ (t). To
examine this, we note that the instantaneously sampled signal g𝛿 (t) in Figure 9.1d is a train of impulse or Dirac
delta functions each weighted by g(nT s ), the value of the analogue signal at the sampling instant. Figure 9.18
shows that the flat-top-sampled signal gΠ (t) consists of a rectangular pulse rect(t/𝜏) replicated at the locations of
 9.6 Non-instantaneous Sampling 619

(a) Analogue g(t) Sample & gΠ(t) Sampled

signal Hold signal
rectTs(t/τ)
Switching
signal
g(t) S1 gΠ(t)
Analogue Sampled
signal signal
(b) rectT (t/τ)
Switching s S2 C
signal

g(t)
gΠ(t)

τ
(c) d = τ/Ts = 1/4

t
Ts

Figure 9.17 Sample-and-hold operation. (a) Block diagram; (b) circuit implementation; (c) analogue and sampled
waveforms.

gΠ(t)

=
rect(t/τ)
Telegram: @Persian_IEEEComSoc

t
τ

*
gδ(t)

Figure 9.18 Flat-top sampling as a convolution between instantaneous sampling and a rectangular pulse.
 620 9 Sampling

the impulse functions that constitute g𝛿 (t). The height of each replicated pulse is equal to the weight of the impulse
at that location. This can be recognised as a convolution process. That is, gΠ (t) is obtained by convolving g𝛿 (t) with
rect(t/𝜏), which is written as follows
gΠ (t) = g𝛿 (t) ∗ rect(t∕𝜏) (9.25)
Noting that convolution in the time domain translates into multiplication in the frequency domain, it follows
that the spectrum GΠ (f ) of the flat-top-sampled signal is given by the product of the spectrum G𝛿 (f ) of the instan-
taneously sampled signal g𝛿 (t) and the spectrum RTs (f ) of the rectangular pulse
GΠ (f ) = G𝛿 (f )RTs (f ) (9.26)
With the spectrum G𝛿 (f ) being a replication of G(f ) at intervals F s along the frequency axis, and RTs (f ) being a sinc
envelope (Figure 4.29b), it follows that GΠ (f ) is as shown in Figure 9.19b.
We see that the spectrum of gΠ (t) is distorted by the sinc envelope of RTs (f ) – the spectrum of the finite-width
rectangular pulse. This distortion is called the aperture effect and is like the distortion observed in television and
facsimile arising from a finite scanning aperture size. Note, however, that the distortion to the baseband spectrum
is very small, depending on the duty cycle of the sampling pulse. An LPF can therefore be used to recover the
spectrum G(f ) and hence the original signal g(t) from the flat-top-sampled signal, with compensation made for
aperture effect, as discussed after the worked example. Again, we require that F s ≥ 2f m .

|G(f)|

(a)

f
–fm fm

|GΠ(f)|
Telegram: @Persian_IEEEComSoc

(b)
Fs = 3fm

f
–4Fs –3Fs –2Fs –Fs –fm 0 fm Fs 2Fs 3Fs 4Fs

Figure 9.19 Flat-top sampling using rectangular pulse train of duty cycle 1/4 as switching signal. Spectrum of (a) analogue
signal and (b) sampled signal.
 9.6 Non-instantaneous Sampling 621

Worked Example 9.3

Obtain expressions for the following spectra in terms of the spectrum G(f ) of the original analogue signal g(t):
(a) G𝛿 (f ), the spectrum of the instantaneously sampled signal.
(b) GΠn (f ), the spectrum resulting from natural sampling.
(c) GΠ (f ), the spectrum of the flat-top-sampled signal.

(a) G𝛿 (f ) is obtained by taking the Fourier transform of both sides of Eq. (9.7). Before doing this, we first
re-introduce in the right-hand side of (9.7) the factor Ad that was normalised to unity, and note that
Ad = A𝜏∕Ts = 1∕Ts
since A𝜏 is the area under the impulse function and we are dealing with a unit impulse train. Therefore
[ ]
2 ∑
∞
1
G𝛿 (f ) = G(f ) + F g(t) cos(2𝜋nF s t)
Ts Ts n=1

1 ∑
∞
= G(f − nF s ) (9.27)
Ts n=−∞
where we have used the fact that the spectrum of 2 g(t)cos(2𝜋nF s t) is G(f ± nFs), which means the spectrum
G(f ) shifted to the locations −nFs and +nFs along the frequency axis. Eq. (9.27) states that G𝛿 (f ) is given by
exact duplications (except for a scaling factor 1/T s ) of G(f ) at intervals of F s along the frequency axis. This
spectrum is shown in Figure 9.2c for a representative G(f ).
(b) Let us denormalise the right-hand side of Eq. (9.23) by re-introducing the factor Ad, and note that usually the
rectangular pulse train is of unit amplitude, so that A = 1. Taking the Fourier transform of Eq. (9.23) after this
change yields
[ ∞ ]
∑
GΠn (f ) = dG(f ) + F d 2g(t)sinc(nd) cos(2𝜋nF s t)
n=1
∑
∞
=d sinc(nd)G(f − nF s ) (9.28)
n=−∞

Equation (9.28) states that GΠn (f ) is obtained by replicating G(f ) without distortion at intervals of F s along the
frequency axis; however, the duplicates located at ±nF s are scaled down by the factor dsinc(nd), where d is the
duty cycle of the rectangular pulse train employed in sampling. The spectrum GΠn (f ) is shown in Figure 9.16c
for a representative G(f ).
(c) GΠ (f ) is given by Eq. (9.26), with G𝛿 (f ) given by Eq. (9.27) and the spectrum RTs (f ) of a rectangular pulse given
by entry 8 of Table 4.5 as RTs (f ) = 𝜏sinc(f𝜏). Thus
Telegram: @Persian_IEEEComSoc

1 ∑
∞
GΠ (f ) = 𝜏 sinc(f 𝜏) × G(f − nF s )
Ts n=−∞
∑
∞
= d sinc(df ∕Fs ) G(f − nF s ) (9.29)
n=−∞

where we have substituted d for the factor 𝜏/T s and d/F s for 𝜏 in the argument of the sinc function. Eq. (9.29)
states that GΠ (f ) is obtained by duplicating G(f ) at intervals of F s along the frequency axis, and modifying
the frequency component f within each duplicate by the factor dsinc(df/F s ). A plot of GΠ (f ) is shown in
Figure 9.19b for a representative G(f ).
 622 9 Sampling

9.6.3 Aperture Effect

We observe above that neither instantaneous nor natural sampling is suitable for practical applications, although
they provide samples from which the original signal can be recovered without distortion. However, the more
feasible technique of flat-top sampling introduces a distortion of the baseband spectrum known as the aperture
effect, which increases with duty cycle d, as shown in Figure 9.20 for d = 0.1, 0.5, and 1, and F s = 3f m .
A measure of aperture effect is given by the attenuation Aa (in dB) of a frequency f in the baseband of GΠ (f )
relative to the attenuation at f = 0. Ignoring the factor d in Eq. (9.29) since it represents a constant scaling factor,
we see that Aa is given by
Aa = −20 log10 [sinc(df ∕Fs )], 0 ≤ f ≤ Fs ∕2 (9.30)
where the negative sign is required to make Aa positive. The indicated frequency range applies since the maximum
frequency component f m of the original signal is at most half the sampling frequency. The maximum attenuation
due to the aperture effect occurs at the maximum frequency component f m . In the worst case, f m = F s /2, giving
Aamax = −20 log10 [sinc(d∕2)] (9.31)
Equation (9.31) yields a maximum distortion of 0.04 dB, 0.9 dB, and 3.9 dB at duty cycles d = 0.1, 0.5, and 1.0,
respectively. There is, therefore, negligible distortion for duty cycles d ≤ 0.1, which is evident in Figure 9.20.

(a) d = 0.1

f
–Fs –fm 0 fm Fs

(b) d = 0.5

f
–Fs –fm 0 fm Fs
Telegram: @Persian_IEEEComSoc

(c) d = 1

f
–Fs –fm 0 fm Fs

Figure 9.20 Spectrum |GΠ (f )| of flat-top-sampled signal at various values of duty cycle d of sampling pulses. Based on a
representative rectangular shaped spectrum |G(f )| of the original analogue signal g(t).
 9.7 Summary 623

There are three different ways to minimise the aperture effect.

● By using sampling pulses of small duty cycle d. As noted above, the distortion is negligible when d is less than
about 10%.
● By oversampling with F s > > f m . If this condition is satisfied then the term f/F s in Eq. (9.30) is much less than
unity for all frequency components of the original signal, which leads to very small values of attenuation Aa .
For example, if F s = 20f m , then maximum aperture distortion in the worst-case scenario (d = 1) is
[ ( )]|
df |
Aamax = Aa |d=1, f =fm = −20 log10 sinc |
Fs ||d=1, f =f
[ ( )] m

fm
= −20 log10 sinc = −20 log10 [sinc(0.05)]
20fm
= 0.04 dB
However, there is a penalty of an increased bandwidth requirement as the sampling frequency is increased.
● By using a compensated reconstruction LPF. At large values of duty cycle and a moderate sampling frequency
F s satisfying the sampling theorem, the amplitude distortion due to aperture effect is significant and must be
compensated for. Since the functional form of the distortion is known, it is a straightforward matter to replace
the ordinary LPF having a uniform passband response with a lowpass equaliser whose (normalised) response
in the passband (f = 0 → f m ) is given by
1
|He (f )| = (9.32)
sinc(df ∕Fs )
In practice, the distortion can be treated as a linear effect, allowing the use of an equaliser whose gain increases
linearly with frequency in the range 0 to f m .

9.7 Summary
In this chapter, we have studied in some detail the three types of sampling, namely instantaneous, natural, and
flat-top. It was noted that the first two do not introduce any distortion, provided the sampling frequency is larger
than twice the analogue signal bandwidth. However, instantaneous sampling is not realisable using physical
circuits, and natural sampling places a limitation on the maximum frequency that can be digitised when the con-
version time is nonzero. The more practical technique of flat-top sampling introduces a distortion known as the
aperture effect, which is, however, not a serious drawback since it can be readily minimised.
The problem of aliasing, which arises when a signal is sampled at a rate of less than twice its bandwidth, was
examined in detail, both in the time and in the frequency domains. Several measures for minimising alias distor-
Telegram: @Persian_IEEEComSoc

tion were also discussed, including the specifications of an anti-alias filter.

The samples of an analogue signal have a continuum of values, which may be transmitted by analogue means
using the so-called pulse amplitude modulation (PAM), pulse duration modulation (PDM), or pulse position
modulation (PPM). These techniques are discussed in Chapter 1, where their demerits are highlighted. Signal
processing that goes beyond sampling is required in order to exploit the numerous advantages of digital commu-
nications.
In the next chapter, we will study the processes of quantisation and encoding, which complete the transforma-
tion of a signal from analogue to digital. Our focus will be on the techniques of PCM and its variants.
 624 9 Sampling

Questions

9.1 Determine the Nyquist rate and Nyquist sampling interval for the following signals:
(a) 5cos(200𝜋t) volts
(b) 20 – sin2 (104 𝜋t) volts
(c) 10rect(2 × 103 t) volts
(d) 20rect(104 t)cos(106 𝜋t)
(e) 5sin(105 𝜋t)trian(500 t)
(f) 10sinc(400 t)sin(2 × 106 𝜋t).
(Note: the rectangular (rect) and triangular (trian) pulse functions are defined in Section 2.6.5.)

9.2 A sinusoidal voltage signal v(t) = 20sin(2𝜋 × 104 t) volts is sampled using an impulse train 𝛿 Ts (t) of period
T s = 40 μs. Sketch the waveform and corresponding double-sided amplitude spectrum of the following sig-
nals:
(a) v(t)
(b) 𝛿 Ts (t)
(c) Sampled signal v𝛿 (t).
(Note: the sketched spectrum of v𝛿 (t) should extend over three replicated bands.)

9.3 Repeat Question 9.2 with a sampling interval Ts = 66 23 𝜇s. Can v(t) be recovered from v𝛿 (t) in this case?
Explain.

9.4 Figure Q9.4 shows the single-sided spectrum of a signal g(t). Sketch the spectrum of the instantaneously
sampled signal g𝛿 (t) over the frequency range ± 4F s , for the following selections of sampling frequency F s :
(a) F s = Nyquist rate (i.e. F smin )
(b) F s = 2F smin
(c) Fs = 23 Fsmin .
(d) Determine the band of alias frequencies in (c).

|G(f)| Figure Q9.4 Question 9.4.

A

f (kHz)
0 6
Telegram: @Persian_IEEEComSoc

9.5 An AM signal g(t) lies in the frequency band 40–50 kHz. Assuming a triangular spectrum for g(t), sketch the
spectrum of the instantaneously sampled signal g𝛿 (t) obtained by sampling g(t) at three times the Nyquist
rate. Your sketch should extend over three replicated bands. How would g(t) be recovered from g𝛿 (t)?

9.6 Let us assume that the spectrum G(f ) of a speech signal can be approximated as shown in Figure Q9.6,
where the spectrum is constant up to 500 Hz and then decreases linearly to −52 dB at 7 kHz. The signal is
to be sampled at F s = 8 kHz and digitised in a uniform analogue-to-digital conversion (ADC) using k = 8
bits/sample.
(a) Determine the order n of an anti-alias Butterworth filter of cut-off frequency f m = 3.4 kHz that is required
to maintain aliasing error in the passband at a level just below quantisation noise.
(b) Repeat (a) for a signal that has a uniform spectrum and compare the two results.
 Reference 625

|G(f)|, dB
0

–52 f (kHz)
0.5 7.0

Figure Q9.6 Question 9.6

9.7 A sinusoidal voltage signal v(t) = 20sin(2𝜋 × 104 t) volts is sampled using a rectangular pulse train rectTs (t/𝜏)
of period T s = 40 μs and duty cycle d = 0.5 as the switching signal. Sketch the waveform and corresponding
double-sided amplitude spectrum of the following signals: (i) v(t), (ii) rectTs (t/𝜏), and (iii) the sampled signal
vΠ (t), assuming
(a) Natural sampling
(b) Flat-top sampling.
(Note: the spectrum of the sampled signal should extend over three replicated bands.)

9.8 Starting with a sinusoidal message signal of frequency f m , show that if the quantiser has a conversion time
𝜏, and the desired conversion accuracy is half the quantiser step size then the highest frequency that can be
digitised using natural sampling is given by Eq. (9.24).

9.9 Speech signal of baseband frequencies 300 Hz to 3400 Hz is sampled in a sample-and-hold circuit that uses
a rectangular pulse train of duty cycle d and sampling frequency F s . Determine the maximum distortion (in
dB) due to aperture effect for the following cases:
(a) F s = 6.8 kHz, d = 0.8
(b) F s = 8 kHz, d = 0.8
(c) F s = 40 kHz, d = 0.8
(d) F s = 8 kHz, d = 0.1.
Comment on the trends in your results.

Reference
Telegram: @Persian_IEEEComSoc

1 Otung, I. (2014). Digital Communications: Principles & Systems. London: Institution of Engineering and Technol-
ogy (IET). ISBN: 978-1849196116.
 627

10

Digital Baseband Coding

Have a goal! Erect it before the start of play.

Without goalposts, even the most exciting football game becomes boring and aimless.

In this Chapter
✓ Types of quantisation: midrise, mid-step, rounding, and truncation.
✓ Uniform quantisation: all the design parameters, trade-offs, and limitations.
✓ Nonuniform quantisation: a detailed discussion of the quantisation and encoding processes in 𝜇-law and
A-law pulse code modulation (PCM) and improvements in signal-to-quantisation-noise ratio (SQNR).
✓ Differential pulse code modulation (DPCM) and low bit rate (LBR) speech coding: introduction to a wide
range of techniques to digitally represent analogue signals, especially speech with as few bits as possible.
✓ Line coding: how bit streams are electrically represented in communication systems.

10.1 Introduction

The four steps involved in converting analogue signals to digital are introduced in Chapter 1 and the first two steps
of lowpass filtering and sampling are discussed in detail in Chapter 9. This chapter focuses on the remaining steps
involving quantisation and encoding. Quantisation converts a sampled analogue signal g(nT s ) to digital form by
approximating each sample to the nearest of a set of discrete values. The result is a discrete-value discrete-time
signal gq (nT s ), which can be conveyed accurately in the presence of channel noise that is less than half the spacing
Telegram: @Persian_IEEEComSoc

of quantisation levels.
However, further robustness to noise can be achieved if the N quantisation levels are numbered from 0 to N − 1,
and each level is expressed as a binary number consisting of k binary digits (or bits), where k = log2 N. This encoding
process in which gq (nT s ) is converted to a string of binary 0’s and 1’s has been traditionally called pulse code
modulation (PCM). Note, however, that the use of the word modulation in this context is inappropriate, in view of
our discussions in Chapters 7 and 8. The resulting bit stream is electrically represented as voltage values by using a
suitable line code, e.g. +12 V for binary 0 and −12 V for binary 1. Binary coding gives maximum robustness against
noise and is easy to regenerate. The concern at the receiver is not with the exact voltage level, but with whether
the received voltage level falls in the range that represents a binary 0 or 1. Thus, the noise level must be large in
order to cause any error. The technique of line coding (first introduced in Chapter 1) is discussed further in this
chapter.
Communication Engineering Principles, Second Edition. Ifiok Otung.
© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 628 10 Digital Baseband Coding

In discussing PCM, we concentrate first on uniform quantisation, which is also called linear ADC
(analogue-to-digital conversion), in order to discuss in quantitative terms the problem of quantisation noise
and various design considerations, including the noise-bandwidth trade-off in PCM. The demerits of uniform
quantisation that make them unsuitable for bandwidth-limited applications are highlighted. Attention then shifts
to the more practical nonuniform quantisation techniques. Encoding and decoding procedures are explained in
detail for the standard A-law and 𝜇-law PCM.
Various modifications to PCM, mainly aimed at improved spectrum efficiency, are discussed. These include dif-
ferential PCM, with delta modulation (DM) and adaptive differential pulse code modulation (ADPCM) introduced
as special cases. LBR coding techniques for speech are then also briefly introduced.

10.2 Concept and Classes of Quantisation

The sampling process converts an analogue signal g(t) into a discrete sequence of sample values g(nT s ), n = 0, 1,
2, 3, …, spaced at the sampling interval T s . As these samples are obtained from the instantaneous values of an
analogue (continuous value) signal, they can take on any value in a range from the smallest to the largest sample
value. The sampled signal g(nT s ), although discrete in time, is still continuous in value and is thus not a digital
signal. To convert g(t) into a digital signal, the next step after sampling must be to constrain the sequence of values
to a finite set of possible values. This process is known as quantisation.
We saw earlier that sampling a signal at the correct rate does not introduce any errors. That is, sampling is a
reversible process. The original signal can be accurately recovered from the samples. By contrast, quantisation
introduces irrecoverable errors, known as quantisation noise. We can make this noise as small as we wish, but at
a price! Quantisation may be classified as:
● Uniform or nonuniform
● Midrise or mid-step (also called mid-tread)
● Rounding or truncation.
In uniform quantisation the quantisation levels to which sample values are approximated are equally spaced
using a constant step size over the entire input range of the quantiser. Nonuniform quantisation, on the other
hand, employs unequal spacing of quantisation levels, with step sizes that are smaller near zero and progressively
larger away from the origin. We will find in Section 10.4 that these step sizes are usually designed to increase
exponentially away from the origin. Uniform and nonuniform quantisation are illustrated in Figure 10.1a for N = 8
intervals. The quantiser has input range 2C, from −C to +C, which is divided into N equal intervals. The dashed
line bisecting each interval is the quantiser output for all inputs in that interval, i.e. it is the discrete level to which
all sample values falling in the interval are approximated. Each output level of the quantiser has been assigned a
positive integer quantisation index L(x) by counting the output levels from 0 to N − 1, going away from the origin
Telegram: @Persian_IEEEComSoc

in either direction towards ±C. The sign of the quantiser input x is usually identified by the most significant bit
(MSB) in the encoding process, as discussed later. This allows each output level to be represented by a unique
binary codeword.
Midrise quantisation has the value zero as a boundary between two quantisation intervals so that zero is a point
of transition where you ‘rise’ from one quantisation level to another. The transfer characteristic (i.e. graph of output
versus input) of an N-level midrise quantiser therefore has exactly N/2 output levels on either side (i.e. left and
right) of the y axis. In mid-step quantisation on the other hand, there is no transition between quantisation levels
at zero; and zero is one of the quantisation levels or steps or treads but not a boundary between two quantisation
levels. The mid-step quantiser therefore has N/2 − 1 output levels below zero and N/2 − 1 output levels above zero
in addition to the zero-output level, making a total of N − 1 output levels. The description ‘mid’ is a reference to the
fact that zero is the midpoint of the quantiser input range. One beneficial feature of mid-step (also referred to as
 10.2 Concept and Classes of Quantisation 629

Input level Input level

C
C
3
3
2
1 2
1
0 0
0 0
(i) 0 (ii) 1
0
2
1
2 3
3
–C
–C
Quantisation index L(x) Quantisation index L(x)

(a)

Midrise Mid-step

Output Output
C C
7C/8 6C/7
5C/8 4C/7
3C/8
2C/7
Rounding

C/8 Input Input

0 0
–C/8
–2C/7
–3C/8
–5C/8 –4C/7

–7C/8 –6C/7
–C –C
–C – 6C – 4C – 2C 0 2C 4C 6C C –C – 5C – 3C – C 0 C 3C 5C
C
8 8 8 8 8 8 7 7 7 7 7 7

Output Output
C C
6C/8 6C/8
4C/8 4C/8
Truncation

2C/8 2C/8
Input Input
0 0
–2C/8 –2C/8
–4C/8 –4C/8
Telegram: @Persian_IEEEComSoc

–6C/8 –6C/8
–C –C
6C 4C 2C 0 2C 4C 6C 6C 4C 2C 0 2C 4C 6C
–C – – – C –C – – – C
8 8 8 8 8 8 8 8 8 8 8 8

(b)

Figure 10.1 (a) Classes of quantisation: (i) uniform, (ii) nonuniform; (b) Classes of quantisation: Left column = Midrise;
Right column = Mid-step or Mid-tread; Top row = Rounding; Bottom row = Truncation.
 630 10 Digital Baseband Coding

mid-tread) quantisation is that it removes noise chatter on idle lines since all small voltage fluctuations about zero
are set to zero at the quantiser output. Such noise on an idle line will cause a midrise quantiser output to fluctuate
between the two output levels nearest to zero. Mid-step quantisation is employed in 𝜇-law PCM, whereas A-law
PCM uses midrise quantisation. A-law and 𝜇-law PCM are, however, based on nonuniform quantisation, which is
discussed in detail in Section 10.4.
In rounding quantisation, the quantiser output is generated by approximating all input values within each quan-
tisation interval to the midpoint of the interval, whereas in truncation quantisation all input values within each
interval are approximated to the boundary point of the interval. Thus, the maximum quantisation error in a
rounding quantiser is half the quantisation step size, whereas that of a truncating quantiser is the full step size.
Each of these four classes of quantisation can be implemented in the uniform or nonuniform class. Figure 10.1b
shows their transfer characteristics for the uniform class of implementation. The top left graph is for a midrise
rounding quantiser, the top right is for a mid-step rounding quantiser, bottom left is for a midrise truncating quan-
tiser and bottom right is for a mid-step truncating quantiser. Notice that the mid-step truncating quantiser has
an enlarged dead zone (i.e. input range quantised to zero) equal to two step sizes; the mid-step rounding and
midrise truncating quantisers have a dead zone equal to one step size, but the midrise rounding quantiser has no
dead zone. Also, all the quantisers have a step size of 2C/N except the mid-step rounding quantiser which has a
step size of 2C/(N − 1). The mid-step rounding quantiser uses a larger step size than the midrise rounding quan-
tiser and therefore incurs slightly larger approximation errors or quantisation noise. The difference in step size is
2C∕(N − 1) − 2C∕N = 2C∕[N(N − 1)], which is negligibly small for large N, e.g. N = 256, and typically low values
of C.
Each of the above quantisers, when implemented as a uniform quantiser having a constant step size Δ, maps an
input sample of value x to an output quantisation index L(x), which is a positive integer given by
⌊ ⌋
⎧ |x| + 1 , Mid-step rounding
⎪ Δ 2
⎪
⎪⌊|x|∕Δ⌋, Mid-step truncation
L(x) = ⎨ (10.1)
⎪⌊|x|∕Δ⌋, Midrise rounding
⎪
⎪
⎩⌊|x|∕Δ⌋, Midrise truncation
where |x| denotes the magnitude or absolute value of x and the notation ⌊⌋ is the floor operator. That is, ⌊z⌋ means
the largest integer less than or equal to the real number z. For example, ⌊2.01⌋ = 2; ⌊8.99⌋ = 8; ⌊−3.2⌋ = −4.
An N-level quantiser requires a k-bit source encoder (where N is an integer power of 2 and k = log2 N) to convert
the quantised output into a binary codeword by coding the above index L(x) using k − 1 bits along with an extra bit
for the sign of x. This sign of x is specified below using the signum function sgn(x), which is +1 for x positive (or
zero) and is −1 for negative x. A suitable uniform or linear decoder will read or extract L(x) and sgn(x) from this
codeword and then convert them into a reconstructed sample y(x) given by
Telegram: @Persian_IEEEComSoc

⎧Δ ⋅ L(x), Mid-step rounding

⎪
⎪Δ ⋅ L(x), Mid-step truncation
⎪ ( )
y(x) = sgn(x) × ⎨ 1 (10.2)
Δ ⋅ L(x) + , Midrise rounding
⎪ 2
⎪ ( )
⎪Δ ⋅ L(x) + 1 − sgn(x) , Midrise truncation
⎩ 2
There are various source encoding standards for constructing binary codewords to represent the quantised out-
puts. These include A-law and 𝜇-law PCM (discussed in Section 10.4), fixed-point number representation using
signed magnitude, one’s complement and two’s complement formats (introduced in Worked Example 10.1) and
floating-point number representation (not discussed).
 10.2 Concept and Classes of Quantisation 631

Worked Example 10.1

This is an important worked example in which we introduce three formats for fixed-point binary represen-
tation of quantiser outputs and compare the quantisation noise of midrise rounding and midrise truncating
quantisers.
An analogue signal x(t) = 12sin(200 𝜋t + 30∘ ) V is sampled at five times the minimum rate stipulated by the
sampling theorem starting at time t = 0.1 ms. The sampled signal is then quantised using:
(a) Uniform midrise rounding quantiser
(b) Uniform midrise truncating quantiser
that has a quantiser input range of ±12 V. The quantised samples are coded using a 6-bit linear encoder. Deter-
mine the following for each quantiser:
(i)Binary representation of the first 10 samples in signed magnitude format.
(ii)Binary representation of the first 10 samples in one’s complement format.
(iii)Binary representation of the first 10 samples in two’s complement format.
(iv) The mean square quantisation error (MSQE) of the midrise rounding quantiser based on the first
10 samples.
(v) The MSQE of the midrise truncating quantiser based on the first 10 samples.
(vi) Compare the MSQE of the two quantisers and comment on your result.

We will show detailed calculations for only two samples and then present the rest of the results for the first 10
samples in two tables, one for each quantiser.
The input signal x(t) is a sinusoid of frequency f m = 100 Hz. As discussed in Chapter 9, the minimum sam-
pling rate is 200 Hz, so at five times this minimum, the sampling rate is F s = 1 kHz, giving a sampling interval
T s = 1/F s = 1 ms. Sampling starts at 0.1 ms and is spaced T s apart, so the sampling instants, numbered from n = 0
to n = 9, are t(0) = 0.1 ms, t(1) = 1.1 ms, t(2) = 2.1 ms, t(3) = 3.1 ms, …, t(9) = 9.1 ms. The samples at these respective
sampling instants are denoted x(0), x(1), …, x(9) and obtained as

x(0) = 12 sin(200π × 0.1 × 10−3 + π∕6) = 6.6407 V

x(1) = 12 sin(200π × 1.1 × 10−3 + π∕6) = 11.2474 V
⋮
x(9) = 12 sin(200π × 9.1 × 10−3 + π∕6) = −0.5025 V

Given that the quantiser has input range ±C = ±12 V and that the number of bits/sample is k = 6, the number
Telegram: @Persian_IEEEComSoc

of quantisation levels is N = 2k = 64 and the step size Δ of the (specified midrise) quantisers is

Δ = 2C∕N = 2 × 12∕64 = 3∕8 V

Using Eq. (10.1), we obtain the output quantisation index, denoted Lr (x) for the rounding quantiser and Lt (x)
for the truncating quantiser, for each input sample x(n) as
⌊ ⌋ ⌊ ⌋
|x(0)| 6.6407
Lr (x(0)) = = = ⌊17.7085⌋ = 17
Δ 3∕8
⌊ ⌋ ⌊ ⌋
|x(1)| 11.2474
Lr (x(1)) = = = ⌊29.9930⌋ = 29
Δ 3∕8
⋮
 632 10 Digital Baseband Coding

⌊ ⌋ ⌊ ⌋
|x(9)| 0.5025
Lr (x(9)) = = = ⌊1.34⌋ = 1
Δ 3∕8
Lt (x(0)) = 17; Lt (x(1)) = 29; · · ·; Lt (x(9)) = 1

Now to define the three formats for binary representation B(z) of a number z using k bits: in signed magnitude,
k − 1 bits are used to represent the magnitude of z as a binary number. An MSB is then prepended (i.e. inserted
in front), set to 0 if z is positive and 1 if z is negative. In one’s complement, if z is positive then B(z) is as for signed
magnitude. But if z is negative then B(z) is obtained by flipping (i.e. negating) every bit of the signed magnitude
representation of |z|. Two’s complement format is as above (i.e. the same as signed magnitude) when z is positive.
But when z is negative, 1 is added to the one’s complement representation of z to obtain its two’s complement
representation.
For example, with k = 6 and k − 1 = 5, the 5-bit binary number for Lr (x(0)) = 17 is 10001 (since 17 = 16 + 1).
Because x(0) is positive, we prepend MSB = 0 to this 5-bit number to obtain the 6-bit signed magnitude codeword
010001, which is also the one’s complement and two’s complement representation. Similarly, the 5-bit binary
number for Lr (x(9)) = 1 is 00001. Because x(9) is negative, we prepend MSB = 1 to this 5-bit number to obtain
the 6-bit signed magnitude codeword 100001. The one’s complement codeword must convey the information that
x(9) is negative and this is done by flipping every bit of 000001 (which is the 6-bit signed magnitude representation
of |Lr (x(9))|) to obtain 111110. The two’s complement codeword is obtained by adding 1 to the one’s complement
codeword (since in this case the number is negative), giving 111110 + 1 = 111111. The complete set of results is
listed in Table 10.1 for the midrise rounding quantiser. Codewords for the midrise truncating quantiser outputs
are identical to the list in Table 10.1.
To compute MSQE, we must first determine the quantisation error eq (n) in each quantised sample. This is given
by

eq (n) = x(n) − y(x(n)) (10.3)

where x(n) is the input sample and y(x(n)) is the quantised sample given by Eq. (10.2) for all the uniform
quantisers.

Table 10.1 Worked Example 10.1: Results for midrise rounding quantiser.

Codewords

n t(n), ms x(n), V L(x(n)) sgn(x(n)) Signed magnitude One’s complement Two’s complement y(x(n)), V eqr (n), mV
Telegram: @Persian_IEEEComSoc

0 0.1 6.6407 17 1 010001 010001 010001 6.5625 78.20

1 1.1 11.2474 29 1 011101 011101 011101 11.0625 184.88
2 2.1 11.5580 30 1 011110 011110 011110 11.4375 120.45
3 3.1 7.4538 19 1 010011 010011 010011 7.3125 141.27
4 4.1 0.5025 1 1 000001 000001 000001 0.5625 −59.99
5 5.1 −6.6407 17 −1 110001 101110 101111 −6.5625 −78.20
6 6.1 −11.2474 29 −1 111101 100010 100011 −11.0625 −184.88
7 7.1 −11.5580 30 −1 111110 100001 100010 −11.4375 −120.45
8 8.1 −7.4538 19 −1 110011 101100 101101 −7.3125 −141.27
9 9.1 −0.5025 1 −1 100001 111110 111111 −0.5625 59.99
 10.2 Concept and Classes of Quantisation 633

Table 10.2 Worked Example 10.1: Results for midrise truncating quantiser.

y(x(n)), V 6.375 10.875 11.25 7.125 0.375 −6.75 −11.25 −11.625 −7.5 −0.75
eqt (n), mV 265.70 372.38 307.95 328.77 127.51 109.30 2.62 67.05 46.23 247.49

The quantisation error eqr (0) of the first sample of the midrise rounding quantiser is
[ ]
1
eqr (0) = x(0) − sgn(x(0)) × Δ × L(x(0)) +
[ ] 2
3 1
= 6.6407 − (+1) × × 17 +
8 2
= 78.20 mV

The quantisation error eqt (9) of the 10th sample of the midrise truncating quantiser is
[ ]
1 − sgn(x(9))
eqt (9) = x(9) − sgn(x(9)) × Δ × L(x(9)) +
2
[ ]
3 1 − (−1) 3
= −0.5025 − (−1) × × 1 + = −0.5025 +
8 2 4
= 247.49 mV

And so on for all the results listed in Tables 10.1 and 10.2.
The MSQE is obtained by averaging as follows over the quantisation errors in the 10 outputs of each quantiser

1 ∑ 2
9
MSQE = e (n) (10.4)
10 n=0 q

Using the values listed in Table 10.1 the MSQE of this midrise rounding quantiser operating on this input
signal is

1 ∑ 2
9
MSQEr = e (n)
10 n=0 qr
1 [ ]
= 0.07822 + 0.184.882 + · · · + 0.059992
10
= 15.6725 mW

Similarly, from the values in Table 10.2 we obtain the MSQE of the midrise truncating quantiser operating on
Telegram: @Persian_IEEEComSoc

the same input signal as

1 ∑ 2
9
MSQEt = e (n)
10 n=0 qt
1 [ ]
= 0.26572 + 0.372382 + · · · + 0.247492
10
= 50.8288 mW

MSQE gives a measure of quantisation noise power. Comparing the MSQE of both quantisers, we see that the
quantisation noise power of the truncating quantiser exceeds that of the rounding quantiser by

10log10 (MSQEt ∕MSQEr ) = 5.11 dB

 634 10 Digital Baseband Coding

In general, the quantisation noise power of truncating quantisers is larger than that of rounding quantisers. It
is shown in the next section and in Section 10.5.2 that
⎧ Δ2 , Rounding Quantiser
⎪
MSQE = ⎨ 122 (10.5)
⎪ Δ , Truncating Quantiser
⎩ 3
So, the exact ratio between the two quantisation noise powers is 10log10 (4) = 6 dB, rather than the 5.11 dB
obtained above. Using a much larger number of samples than just the 10 in this worked example yields a closer
estimate to the exact theoretical result. For example, choosing a sampling rate of 100 kHz (to collect 1000 samples
of the signal over one cycle) yields a ratio of 5.83 dB between the two quantisation noise powers.We now focus
our discussion on uniform quantisation based exclusively on the midrise rounding implementation in order to
introduce key design parameters.

10.3 Uniform Quantisation

The simplest method of quantisation is to divide the full range of input signal values, from −C to +C, into N
equal intervals, as shown in Figure 10.2. Note that there are N/2 intervals (indexed L = 0, 1, 2, …, N/2 − 1) in the
positive range and N/2 intervals in the negative range (also indexed L = 0, 1, 2, …, N/2 − 1) and zero is an interval
boundary, so this is a midrise quantisation scheme. Furthermore, all input samples falling within a given interval
are approximated or quantised to the midpoint of that interval, making this a rounding quantiser. Figure 10.2 thus
represents a midrise rounding uniform quantisation scheme.

Input, x
C
yN/2 – 1 = (N – 1)Δ/2
C–Δ

yj = (2j + 1)Δ/2

Δ
y1 = 3Δ/2 s
Δ eq
Δ y0 = Δ/2 Δ/2
0
–y0
Telegram: @Persian_IEEEComSoc

–Δ 0
–y1
Quantisation index,
L = 0, 1, 2, …, N/2 – 1
–yj

–yN/2 – 1
–C
Quantised output, yL

Figure 10.2 Midrise rounding uniform quantisation.

 10.3 Uniform Quantisation 635

Table 10.3 Choosing N = 8 intervals uses all 3-bit binary codewords.

Quantisation
interval 0 1 2 3 4 5 6 7
Binary 000 001 010 011 100 101 110 111
codeword
← Not used if N = 5 →

The number of intervals is usually chosen to be an integer power of 2

N = 2k ; ⇒ k = log2 N (10.6)

This choice allows the k-bit binary codes to be fully utilised at the coding stage of the ADC process to represent
the N intervals. For example, if we choose N = 5 then we require k = 3 bits to represent these five intervals. See
Table 10.3, where, for convenience, the intervals have been unidirectionally indexed from 0 to N − 1 (instead of
the usual bidirectional indexing from 0 to N/2 − 1 for the positive range and 0 to N/2 − 1 for the negative range).
Notice how the codes 101, 110, and 111 are surplus to requirement if N = 5.
It is shown below that quantisation noise is reduced as N increases. In this example therefore, we can reduce
quantisation noise at no extra cost (i.e. without increasing the number of bits k required to represent each quan-
tisation interval) simply by increasing N from 5 to 23 = 8.

10.3.1 Quantisation Noise

In a uniform or linear quantisation, all intervals are of an equal size Δ, called the quantiser step size. And in a
rounding quantiser, the quantised value of each interval is set to the average of all input samples that fall in that
interval. If the samples in each interval are equally likely to be located anywhere within the interval, this average
and hence quantised value will be the midpoint of the interval. Each sample is therefore approximated to the
midpoint of the interval in which it falls. For example, any sample of value in the range 0 to Δ (which is the zeroth
positive quantisation interval with index L = 0) is quantised to y0 = Δ/2; any sample in the range 0 to −Δ (which
is the zeroth negative quantisation interval) is quantised to −y0 ; any sample in the topmost interval (C, C−Δ), of
index L = N/2 − 1, is quantised to (N − 1)Δ/2; and in general any sample in the jth positive quantisation interval
is quantised to yj = (2j + 1)Δ/2. Therefore, for each sample of value s in the jth interval, quantisation produces an
error given by

eq = |s − yj | (10.7)
Telegram: @Persian_IEEEComSoc

This error has a maximum value eq max when s is either by the top or by the bottom of the interval, i.e. s = yj ± Δ/2.
Thus

eq max = |yj ± Δ∕2 − yj | = Δ∕2 (10.8)

The maximum possible error incurred in each quantisation is therefore half the quantiser step size.
As discussed in Section 3.5.2, the mean square value of a signal gives its (normalised) power. Therefore, to
determine quantisation noise power, we need to calculate the MSQE across all the samples of the signal. However,
if the input signal lies entirely within the quantiser range, i.e. it does not overload the quantiser, the error statistics
are the same across all the uniform quantiser intervals spanned by the input. Therefore, we may focus on the
bottom positive interval (shown blown out on the right-hand side of Figure 10.2) and sum over the entire interval
(0 → Δ) the product of e2q = (s − Δ∕2)2 and the probability ds/Δ that the sample s lies in an infinitesimal interval
 636 10 Digital Baseband Coding

ds around s. This gives

Δ Δ Δ
1 1
MSQE = e2q ds∕Δ = (s − Δ∕2)2 ds = (s2 − sΔ + Δ2 ∕4)ds
∫0 Δ ∫0 Δ ∫0
[( ) Δ] ( )
1 s3 s2 Δ sΔ2 || 1 Δ3 Δ3 Δ3
= − + | = − +
Δ 3 2 4 ||0 Δ 3 2 4
4Δ 2 6Δ 2 3Δ 2
= − +
12 12 12
Therefore
Δ2
MSQE = (10.9)
12
We see that the MSQE depends only on the step size Δ. This error will appear as noise associated with the signal.
The main advantage of this quantisation process is that if we can somehow convey the information about each
quantised level without error then we will introduce no further degradation to the signal. We can therefore make
this quantisation noise as small as we wish by sufficiently reducing Δ. However, there is a price to pay for this
improved noise performance. From Figure 10.2 and Eq. (10.6) it follows that
2C 2C
Δ= = k (10.10)
N 2
Here, k is the number of bits required to represent each quantised level, i.e. the number of bits per sample. Thus,
Δ and hence quantisation noise can be reduced by increasing the number of bits per sample. This increases the bit
rate and hence the bandwidth required for transmission.

10.3.2 Dynamic Range of a Quantiser

A sinusoidal input signal of amplitude V max = C will fully load the quantiser in Figure 10.2, since it has values that
cover the entire range from −C to +C. Sinusoids of a larger amplitude would cause clipping and distortion. On the
other hand, the variations of a sinusoidal input signal of amplitude V min = Δ/2 are confined to a single interval
of the quantiser. In other words, variations in this sinusoid will go undetected at the quantiser output. The ratio
of the largest amplitude V max of a sinusoid that avoids clipping to the largest amplitude V min of a sinusoid whose
variations go undetected is called the dynamic range of the quantiser
Vmax C
Dynamic Range = = = 2k
Vmin Δ∕2
= 6.02k dB (10.11)
Telegram: @Persian_IEEEComSoc

The dynamic range therefore depends on the number of bits per sample. It increases by 6 dB for each extra bit
available for representing each sample.

10.3.3 Signal-to-quantisation-noise Ratio (SQNR)

An important parameter for assessing the performance of the quantiser is the ratio of signal power to quantisa-
tion noise power, called the signal-to-quantisation-noise ratio (SQNR). Let us define a parameter R known as the
peak-to-rms ratio of the signal to be quantised. For a signal of rms = 𝜎, and peak value V p

Vp Vp2
R= ; Signal Power = 𝜎 2 =
𝜎 R2
 10.3 Uniform Quantisation 637

2
Signal Power Vp ∕R
2 12 × 22k Vp 2
SQNR = = 2 =
MSQE Δ ∕12 4C2 R2
2
3Vp
= 2 2 22k (10.12)
C R
where we have made use of the expression for Δ in Eq. (10.10). Expressing Eq. (10.12) in dB
[ ]
3Vp2 2k
SQNR = 10log10 2
C 2 R2
= 10log10 (3) + 10log10 (22k ) + 10log10 (R−2 ) + 10log10 (Vp 2 ∕C2 )
= 4.77 + 6.02k − 20log10 (R) + 20log10 (Vp ∕C) dB (10.13)

If the signal fully loads the quantiser, V p = C, and the SQNR improves to

SQNR = 4.77 + 6.02 k − 20log(R) (10.14)

Worked Example 10.2

Determine the SQNR as a function of number of bits/sample for each of the following signals:
(a) Sinusoidal signal.
(b) Signal with a uniform probability density function (PDF).
(c) Speech signal.

(a) If a sinusoidal signal g(t) fully loads the quantiser then its amplitude V p = C, and we may write g(t) = Csin(𝜔t).
The signal g(t) has a period T = 2𝜋/𝜔, and a mean square value
T T
1 1
𝜎2 = g2 (t)dt = C2 sin2 (𝜔t)dt
T ∫0 T ∫0
T
C2
= (1 − cos 2𝜔t)dt
2T ∫0
C2
=
2
Peak value C √
R= = √ = 2
𝜎 C∕ 2
It follows from Eq. (10.14) that
√
Telegram: @Persian_IEEEComSoc

SQNR = 4.77 + 6.02k − 20 log( 2)

= 1.76 + 6.02k dB (10.15)

(b) The samples of this signal can take on any value between a minimum −C and a maximum +C with equal
probability. It is assumed that the signal fully loads the quantiser. The probability that a sample of the signal
lies between s − ds/2 and s + ds/2 is given by the shaded area pds in Figure 10.3. Since each sample must lie
somewhere between −C and + C, we must have p × 2C = 1, or p = 1/2C. The mean square value 𝜎 2 of the signal
is obtained by summing (over the entire signal range −C to +C) the product of the square of the sample value
s and the probability pds that a sample lies within the infinitesimal interval centred on s
+C +C
1 C2
𝜎2 = s2 pds = s2 ds =
∫−C ∫
2C −C 3
 638 10 Digital Baseband Coding

PDF
p

ds
Sample value
–C s +C

Figure 10.3 Worked Example 10.2: uniform probability density function (PDF).

Thus
Peak C √
R= = √ = 3.
𝜎 C∕ 3
Eq. (10.14) then yields
SQNR = 6.02 k (10.16)
(c) Measurements show that speech signals have on average 20log(R) = 9 dB. Thus, if the speech signal fully loads
the quantiser (i.e. peak value V p = C) then it follows from Eq. (10.14) that

SQNR = 6.02 k − 4.23 dB (10.17)

Worked Example 10.3

A speech signal is to be transmitted by PCM with an output SQNR of 55 dB.

(a) What is the minimum number of bits per sample that must be used to achieve this performance, if the
speech signal fully loads the quantiser?
(b) If the quantiser is only half-loaded by the speech signal, what is the resulting output SQNR for the same
number of bits/sample as above?

(a) The required bits/sample is obtained by rewriting Eq. (10.17) to make k the subject
SQNR + 4.23 55 + 4.23
k= = = 9.84.
Telegram: @Persian_IEEEComSoc

6.02 6.02
The smallest integer larger than or equal to the above result gives the minimum number of bits/sample: k = 10.
(b) The full expression for SQNR in Eq. (10.13) must be used in this case, with 20log(R) = 9 and V p /C = 12
( )
1
SQNR = 4.77 + 6.02k − 9 + 20 log
2
= 4.77 + 6.02 × 10 − 9 − 6.02
= 50 dB
Note how the SQNR degrades when the quantiser is underloaded by a small input signal. Overloading, on the
other hand, leads to clipping. Optimum performance is obtained by scaling the signal prior to quantisation to
ensure that it just fully loads the quantiser.
 10.3 Uniform Quantisation 639

Codewords of quantised outputs

011

010 Quantised output

001 Input analogue signal

000

100

101

110
111
Error Ts = 1/Fs

Figure 10.4 Eight-level uniform quantisation.

A demonstration of a combined process of sampling and uniform quantisation of a sinusoidal signal is shown in
Figure 10.4. There are eight quantiser output levels, requiring k = 3 bits to represent each output level. The input
signal is scaled prior to sampling in order to fully load the quantiser. The bottom plot of Figure 10.4 shows the
quantisation error, which is the difference between the quantised output and the analogue input signal. At each
sampling instant the value of the analogue signal sample is approximated to the midpoint of the quantisation
interval in which it lies. This quantised value is then held until the next sampling instant. The result is a staircase
output signal.

10.3.4 Design Considerations

Several important PCM system design considerations may be deduced from the relationships between the design
parameters in Eqs. (10.12) and (10.13).
● SQNR increases exponentially with the number of bits per sample k, which is itself directly proportional to the
bandwidth B required to transmit the PCM signal. Thus, PCM provides an exponential increase of SQNR with
bandwidth. This is a better bandwidth for noise improvement trade-off than offered by frequency modulation
(FM), where signal-to-noise ratio (SNR) increases roughly as the square of bandwidth. That is, if the SNR is
Telegram: @Persian_IEEEComSoc

SNR1 at bandwidth B1 , and the bandwidth is increased by a factor n to nB1 , then the SNR increases to n2 × SNR1
in FM, but more dramatically to SNRn1 in PCM.
● More simply put, you make a gain of 6.02 dB per extra bit used for coding each sample in PCM, but you gen-
erate more bits per second as a result, and therefore require a larger transmission bandwidth. The number of
bits/sample required for a desired SQNR can be read from Figure 10.5 for the three signals discussed in Worked
Example 10.2.
● SQNR decreases as the square of the quantiser range 2C needed to accommodate the input signals without
clipping. An improvement in SQNR can therefore be realised by reducing the range of input signal values. Some
differential quantisers achieve such gains by quantising the difference between adjacent samples, rather than
the samples themselves. If the sampling rate is sufficiently high, adjacent samples are strongly correlated and
the difference between them is very small, resulting in a reduced range of quantiser input values.
 640 10 Digital Baseband Coding

75
70
65
60
SQNR (dB)

55
DF ech
50 mP Spe
u soid ifor
Sin Un
45
40
35
30
25
5 6 7 8 9 10 11 12
Number of bits/sample, k

Figure 10.5 SQNR of a uniform quantiser as a function of number of bits/sample for various signal types.

● A large segment of the quantisation error signal resembles a sawtooth waveform with a fundamental frequency
that increases with the sampling frequency F s . Thus, oversampling an analogue signal (i.e. choosing a much
higher value of F s than required by the sampling theorem) will have the effect of spreading out the quantisation
noise power over a wider frequency band. As a result, only a significantly reduced fraction of the noise lies
within the signal band at the reconstruction filter.
● When an input signal underloads the quantiser, SQNR decreases by 20log(r) dB, where r is the ratio between
the quantiser range 2C and the peak-to-peak value of the input signal. More simply put, a signal that is at r dB
below the level that fully loads the quantiser will have an SQNR that is r dB worse than the values obtained from
Figure 10.5. In speech communication, for example, this would mean a good SQNR for the loudest speakers and
a significant degradation for soft speakers.

10.3.5 Demerits of Uniform Quantisation

We have observed that the quantisation error associated with each quantised sample can range from 0 to a maxi-
mum of Δ/2, where Δ is the quantiser step size. This error must be kept small compared to the sample value. To
faithfully quantise small samples, one must therefore choose a very small step size, which requires the number of
quantisation intervals and hence number of bits per sample k to be large. For example, speech signals are charac-
Telegram: @Persian_IEEEComSoc

terised by a nonuniform PDF with a preponderance of low values. To maintain fidelity, these low values must be
faithfully transmitted as they mostly represent the consonants that carry intelligibility. The typical dynamic range
of a speech signal is 60 dB, which means a ratio of highest to lowest sample magnitude given by
( )
VH
60 = 20log10
VL
Thus
VH
= 10(60∕20) = 1000
VL
That is, if the peak value allowed when digitising a speech signal is V H = 1 V, then the weakest passage may be
as low as V L = 1 mV. A step size Δ < V L is required to faithfully quantise the smallest samples. Choosing Δ = V L
 10.4 Nonuniform Quantisation 641

results in 1000 intervals for the positive samples, and another 1000 for the negative samples, or 2000 intervals in
total. The number of bits required to code up to 2000 levels is given by

k = ⌈log2 (2000)⌉ = 11 bits

where, ⌈x⌉ denotes the smallest integer larger than or equal to x.

With a sampling frequency F s = 8 kHz, the bit rate that must be transmitted is 8000 samples × 11 bits/sample =
88 kb/s. Such a high bit rate translates to an unacceptably large transmission bandwidth requirement.
Another problem with uniform quantisation may be observed by noting that the quantisation error magnitude
is constant across all intervals. The SQNR is therefore much smaller at the lower intervals (corresponding to small
signal values) than nearer the top since

VL2 ∕e2q ≪ VH2 ∕e2q

In telephony, it is desirable to have a constant SQNR over a wide range of input signal values so that the service
quality is maintained at the same level for both quiet and loud talkers.
The above problems of large bit rate and nonconstant SQNR can be alleviated by using a nonuniform quantiser
in which the step size is a function of input signal value. Large input samples are coarsely quantised using larger
step sizes, whereas the smaller input samples are more finely quantised using smaller step sizes.

10.4 Nonuniform Quantisation

Nonuniform quantisation may be achieved through either of the schemes shown in Figure 10.6. In Figure 10.6a,
the analogue signal is first compressed before being quantised in a uniform quantiser. At the receiver, the decoded
PCM signal is expanded in a way that reverses the effect of the compression process. The combined process of
compression at the transmitter and expansion at the receiver is called companding. Note that companding does
not introduce any distortion, and quantisation error remains the only source of distortion in the ADC process.

Analogue Uniform PCM

Transmitter Compressor
input ADC output

(a)

PCM Uniform Analogue

Receiver Expander
input DAC output
Telegram: @Persian_IEEEComSoc

n bits/sample m bits/sample

Analogue Fine-grain Digital PCM

Transmitter
input uniform ADC translator output
n>m
(b)
n bits/sample
PCM Digital Fine-grain Analogue
Receiver input translator uniform DAC output

Figure 10.6 Nonuniform quantisation schemes.

 642 10 Digital Baseband Coding

The scheme of Figure 10.6b first quantises the input signal using a fine-grain uniform quantiser of, say,
n = 13 bits/sample, corresponding to 213 = 8192 levels. A digital translator is then used to reduce the number of
transmitted bits/sample to, say, m = 8, corresponding to 28 = 256 levels. The reduction is achieved in compressor
fashion by mapping an increasing number of fine-grain levels to the same output level as you go from the low to
high range of signal values. For example, the translation may progress from 2-to-1 near 0, to 128-to-1 near the
maximum signal value. A 128-to-1 translation means that 128 fine-grain levels are equated to 1 output level,
usually the midpoint of the 128 levels. At the receiver, a reverse translation is performed that converts from m to
n bits/sample. The overall effect of this scheme is again finer quantisation of small signal samples and coarser
quantisation of the larger samples.
The quantisation scheme of Figure 10.6b is in practice the preferred approach, since it can be implemented
using low-cost digital signal processors (DSPs). However, to understand the required compressor characteristic,
we first discuss the system of Figure 10.6a and then present the implementation of Figure 10.6b as a piece-wise
linear approximation of the nonlinear compression function in the system of Figure 10.6a.

10.4.1 Compressor Characteristic

Figure 10.7 shows the input–output characteristic of the compressor. The input is normalised to the range −1 to
+1. The output (y axis) is divided into N uniform intervals (illustrated in Figure 10.7 for N = 8), corresponding to
uniform quantisation with step size Δ = 2/N. Note that the uniform intervals of the compressor output y corre-
spond to differently sized intervals of the compressor input signal x, which are small near the origin and increase
steadily for larger magnitudes of x. This arrangement achieves finer quantisation of the smaller samples of x and
coarser quantisation of the larger samples. In the example shown in Figure 10.7, all input samples in the second
positive interval are quantised to output y1 with maximum error Δ1 /2, whereas the input samples in the fourth
positive interval are quantised to level y3 with a bigger maximum error of Δ3 /2.
Recall that one of the goals of nonuniform quantisation is to achieve a constant SQNR throughout the entire
range of input signal x. Let us obtain the SQNR for a signal that is compressed and uniformly quantised as illus-
trated in Figure 10.7. Our aim is to determine the shape of the compression curve that gives an SQNR that is
independent of x.

y
+1
y3

y2 dy = 2/N
dx = Δj
Telegram: @Persian_IEEEComSoc

y1
Δ1 Δ2 Δ3
y0 x
–1
Δ0 +1
–y0

–y1
Input Output
Compressor
x y
–y2

–y3
1

Figure 10.7 Compressor characteristic.

 10.4 Nonuniform Quantisation 643

When the step size varies from interval to interval as in Figure 10.7, the overall MSQE of the quantiser is obtained
by averaging Eq. (10.9) over all the intervals as follows
∑ Δ2j
MSQE = Pj (10.18)
j
12

where Δj is the step size of the jth interval and Pj is the probability that a sample of the input signal will have a
value lying within the range of the jth interval. Assuming sufficiently small intervals, the slope of the compressor
curve is constant within the space of one interval, and (from Figure 10.7) is given for interval j by
dy 2∕N
=
dx Δj
It follows that the jth step size is given by
2 dx
Δj = (10.19)
N dy
The probability of the input signal value falling in the jth interval (of width dx) is
Pj = pX (x)dx (10.20)
where pX (x) is the PDF of the input signal x. In the limit of a large number of quantisation intervals, the summation
for MSQE in Eq. (10.18) becomes an integration operation over the normalised input signal range −1 to +1. Thus
( )2
1 ∑ 2 1 ∑ dx
MSQE = PΔ = p (x)dx
12 j j j 3N 2 All intervals X dy
+1 ( ) 2
1 dx
= p (x) dx (10.21)
3N 2 ∫−1 X
dy
where we have used Eqs. (10.19) and (10.20) for Δj and Pj .
But the mean square value operation – see Eq. (3.21) – gives input signal power as
+1
Signal Power = x2 pX (x)dx (10.22)
∫−1
The desired SQNR is given by the ratio between Eqs. (10.22) and (10.21)
Signal Power
SQNR =
MSQE
+1
∫−1 x2 pX (x)dx
= 3N 2 ( )2 (10.23)
+1
∫−1 dxdy
pX (x)dx
Telegram: @Persian_IEEEComSoc

The right-hand side of Eq. (10.23) must be independent of x if SQNR is to be independent of x as desired. By
examining the above equation, it is easy to see that this can be achieved by setting
dx
= Kx, (10.24)
dy
where K is a constant.
This leads to the result
+1
∫−1 x2 p(x)dx
SQNR = 3N 2 +1
∫−1 K 2 x2 p(x)dx
3N 2
= (10.25)
K2
 644 10 Digital Baseband Coding

which is independent of input signal level x as desired. Thus, the correct compression characteristic is one that
satisfies Eq. (10.24), or
dy 1
=
dx Kx
Integrating
1
y= ln(x) + D
K
where D is a constant that we choose in order to make (x, y) = (1, 1) a point on the curve, since the normalised
maximum input is compressed to the normalised maximum output. Thus D = 1, and the desired compressor char-
acteristic, is
1
y= ln(x) + 1 (10.26)
K
which has a slope
dy 1
= (10.27)
dx Kx
So, what have we achieved? We now have in Eq. (10.26) the full specification of a compression curve that can
be used to compress the input signal x to give an output y, which when uniformly quantised produces a constant
SQNR across the entire input signal range. The result of these two steps (of compression using the curve specified
in Eq. (10.26) followed by uniform quantisation) is fine quantisation of small input values and coarse quantisation
of larger input values.
However, there is a practical problem with the compression function of Eq. (10.26). The slope of the curve (see
Eq. (10.27)) is infinite at x = 0, implying infinitesimally small quantiser steps as x → 0. To circumvent this problem,
the logarithmic function in Eq. (10.26) is replaced by a linear function in the region x → 0. The ITU-T (Interna-
tional Telecommunication Union – Telecommunication) has standardised two such compressor characteristics,
the A-law in Europe and the 𝜇-law in North America and Japan.

10.4.2 A-law Companding

This companding law follows from Eq. (10.26) by setting the constant

K = 1 + ln(A) (10.28)

where A is a positive constant (usually A = 87.6). This defines the logarithmic portion of the characteristic
1 ln(x) + K
ylog = ln(x) + 1 =
K K
Telegram: @Persian_IEEEComSoc

ln(x) + 1 + ln(A)
=
1 + ln(A)
1 + ln(Ax)
= (10.29)
1 + ln(A)
A linear function ylin is used in the region |x| ≤ 1∕A. This is the region x → 0 referred to above. The linear function

ylin = mx + c (10.30)

where m and c are constants, is determined by satisfying two conditions:

● That ylin passes through the origin, so that x = 0 is compressed to y = 0. This means that in Eq. (10.30), ylin = 0
when x = 0, so that the constant c = 0.
 10.4 Nonuniform Quantisation 645

● That, for continuity, the linear and logarithmic functions have the same value at x = 1∕A. Since

ylin |x=1∕A = m∕A

and
( )
1 + ln A × A1
1
ylog |x=1∕A = =
1 + ln(A) 1 + ln(A)
it follows by equating both expressions that
A
m=
1 + ln(A)
Thus
Ax
ylin = mx + c = (10.31)
1 + ln(A)
To summarise, the A-law compression curve is defined by the following equations

⎧ Ax
⎪ 1 + ln(A) , 0 ≤ x ≤ 1∕A
⎪
⎪
y = ⎨ 1 + ln(Ax) , 1∕A ≤ x ≤ 1 (10.32)
⎪ 1 + ln(A)
⎪
⎪−y(|x|), −1 ≤ x ≤ 0
⎩
The last expression in Eq. (10.32) indicates that the compression curve y has odd symmetry, so that a negative
input value, say, −X, where X is positive, is compressed to give a negative output that has the same magnitude as
the output corresponding to input X.
For later use, the gradient of the A-law compression curve at x = 0 is
[ ]|
dy || d Ax | A ||
= | =
dx ||x=0 dx 1 + ln(A) ||x=0 1 + ln A ||x=0
A
= (10.33)
1 + ln A
and the gradient at x = 1 is
[ ]
dy || d 1 + ln(Ax) || 1∕x ||
| = | =
dx |x=1 dx 1 + ln(A) ||x=1 1 + ln A ||x=1
Telegram: @Persian_IEEEComSoc

1
= (10.34)
1 + ln A

10.4.3 𝝁-law Companding

This companding function is obtained from Eq. (10.26) in two steps:

1. Set the constant

K = ln(1 + 𝜇) (10.35)

where 𝜇 is a positive constant (usually 𝜇 = 255), so that

 646 10 Digital Baseband Coding

ln(x) ln(x) + ln(1 + 𝜇)

y= +1=
ln(1 + 𝜇) ln(1 + 𝜇)
ln(x + 𝜇x)
=
ln(1 + 𝜇)
2. Modify the above result by replacing x + 𝜇x in the numerator with 1 + 𝜇x. This gives the 𝜇-law compression
function
⎧ ln(1 + 𝜇x)
⎪ , x≥0
y = ⎨ ln(1 + 𝜇) (10.36)
⎪−y(|x|), −1 ≤ x ≤ 0
⎩
The modification in the second step is necessary in order to satisfy the requirement for linear compression in the
region x → 0. To see that this is the case, note that ln(1 + z) ≈ z for z → 0, which means that the 𝜇-law compression
curve reduces to the following linear relation near the origin
𝜇x
y ≡ ylin = , x→0 (10.37)
ln(1 + 𝜇)
Furthermore, if 𝜇 ≫ 1 then 1 + 𝜇x ≈ 𝜇x for x → 1. Thus, the compression is logarithmic as required for large input
values
ln(𝜇x)
y ≡ ylog = , x→1 (10.38)
ln(1 + 𝜇)
The 𝜇-law function has derivative
dy 𝜇 1
= ⋅
dx 1 + 𝜇x ln(1 + 𝜇)
and thus, its slope at x = 0 is
dy || 𝜇
= (10.39)
dx ||x=0 ln(1 + 𝜇)
and at x = 1 its slope is
dy || 𝜇
= (10.40)
dx ||x=1 (1 + 𝜇) ln(1 + 𝜇)

Worked Example 10.4

Discuss how the values of A and 𝜇 affect the relative sizes of the quantisation steps in A-law and 𝜇-law com-
Telegram: @Persian_IEEEComSoc

panding.

The maximum step size occurs at x = 1 (normalised), and the minimum at x = 0. Let us denote these step sizes
as Δmax and Δmin , respectively. Applying Eq. (10.19)
2 dx || 2∕N
Δmin = =
N dy ||x=0 (dy∕dx)|x=0
2 dx || 2∕N
Δmax = | =
N dy |x=1 (dy∕dx)|x=1
Thus
dy || 2∕N dy || 2∕N
| = ; | = (10.41)
dx |x=0 Δmin dx |x=1 Δmax
 10.4 Nonuniform Quantisation 647

and the ratio of maximum step size to minimum step size is the ratio between the gradients of the compressor
curve at x = 0 and x = 1
Δmax (dy∕dx)|x=0
=
Δmin (dy∕dx)|x=1
For A-law, we use the expressions for the above derivatives given in Eqs. (10.33) and (10.34) to obtain
Δmax
=A (10.42)
Δmin
For 𝜇-law, the above derivatives are given in Eqs. (10.39) and (10.40), from which we obtain
Δmax
=1+𝜇
Δmin
or
Δmax
𝜇= −1 (10.43)
Δmin
We see therefore that the constant A sets the ratio between the maximum and minimum step size in A-law
compression. If A = 1, then Δmax = Δmin and the step sizes are all equal. This is the special case of uniform quan-
tisation. A significant compression is achieved by choosing A ≫ 1. In the case of 𝜇-law compression, 𝜇 = 0 gives
Δmax = Δmin and corresponds to uniform quantisation. The required compression characteristic, Eq. (10.38), is
obtained only by choosing a large value for the constant 𝜇, usually 𝜇 = 255. Figure 10.8 shows the A-law and 𝜇-law
characteristics for various values of A and 𝜇.

10.4.4 Companding Gain and Penalty

Companding gain Gc is defined as the improvement in SQNR for small input values in a nonuniform quan-
tiser compared with the SQNR of the same signal type when using a uniform quantiser of the same number of

y
1.0
A = ∞; μ = ∞
A-law
μ-law
A = 1000; μ = 2000 A = 87.6; μ = 255
0.5
Compressed output, y →

A = 1; μ → 0 x
0
Telegram: @Persian_IEEEComSoc

–0.5

–1.0
–1.0 –0.5 0 0.5 1.0
Normalised input, x →

Figure 10.8 Worked Example 10.4: A-Law and 𝜇-Law compression characteristics.
 648 10 Digital Baseband Coding

bits/sample. It follows that Gc is the square of the ratio between the step size of a uniform quantiser and the small-
est step size of a nonuniform quantiser of the same bits/sample and input range. Noting from Figure 10.7 that dy
is the step size of a uniform quantiser and dx is the corresponding step size of a nonuniform quantiser, it follows
that
[ ] ( )
Gc = 10log10 (dy∕dx)2 |x=0 = 20log10 (dy∕dx)|x=0 dB (10.44)

Using Eqs. (10.31) and (10.37), we obtain the companding gain for A-law and 𝜇-law companding as follows
[ ]
A
Gc A-law = 20log10 dB
1 + ln(A)
[ ]
𝜇
Gc 𝜇-law = 20log10 dB (10.45)
ln(1 + 𝜇)

Thus, if A = 87.6, A-law nonuniform quantisation gives a gain of 24 dB over a uniform quantisation that uses
the same number of bits/sample. A gain of 33 dB is realised with 𝜇-law for 𝜇 = 255. We will see later that, in
practice, a piecewise linear approximation is adopted, resulting in a slightly lower companding gain of 30 dB for
𝜇-law. Recall that an improvement of 6 dB in SQNR is provided by each extra bit used for coding the quantised
samples. A gain of 24 dB is therefore equivalent to four extra coding bits. This means that a uniform quantiser
would require 12 bits/sample in order to have the same performance as A-law with 8 bits/sample. By using A-law
nonuniform quantisation, we have reduced the required number of bits/sample from 12 to 8, representing a saving
of 33% in bit rate and hence bandwidth. In the case of 𝜇-law that achieves 30 dB of companding gain, the bit rate
reduction is 5 bits – from 13 to 8, a saving in bandwidth of 38.5%. Note that these figures apply only to the bandwidth
required for transmitting information bits. In practice, overhead (noninformation) bits must be added, as discussed
in Chapter 13, leading to a lower saving in bandwidth.
Companding penalty Lc is defined as the ratio between the SQNR of an input signal that fully loads a uniform
quantiser and the SQNR of the same signal type when using a nonuniform quantiser of the same number of
bits/sample. We noted earlier that the SQNR of a uniform (or linear) quantiser decreases with peak input signal
level. SQNR in the nonuniform (or log-) quantiser discussed above does not vary significantly over the entire input
signal range. However, we fail to realise an ideal nonlinear quantisation in which SQNR is strictly constant. This
is because, for practical implementation, we had to replace the logarithmic curve of Eq. (10.26) with a linear curve
for |x| ≤ 1/A in the A-law. In the case of 𝜇-law, an approximate curve was employed that becomes linear as x → 0
and logarithmic as x → 1 (normalised).
Consider a log-quantiser and a linear quantiser employing the same number of bits/sample and let their SQNR
be respectively denoted SQNRlog and SQNRlin . We will work with normalised input values, so the quantiser input
range is ±1 and the peak value of the input signal is V p ≤ 1, with V p = 1 (= 0 dB relative to quantiser limit C) if
the input signal fully loads the quantiser. At the top end of the quantiser input range (i.e. x ≈ 1) the step size of
Telegram: @Persian_IEEEComSoc

the linear quantiser (= 2/N) is smaller than the step size of the log-quantiser (= Δmax ), whereas at the bottom end
(x ≈ 0, or x ≪ 1) the linear quantiser step size (still 2/N) is much larger than the log-quantiser step size (which
in this region is Δmin ). Thus, for small input signals (having peak value V p ≪ 1), SQNRlog > SQNRlin and the
difference SQNRlog − SQNRlin in dB is the companding gain Gc discussed above and given by
( )
2∕N
Gc ≡ (SQNRlog − SQNRlin )|x≈0 = 20log10 dB
Δmin
( )
dy ||
= 20log10 dB
dx ||x=0

where we have used Eq. (10.41) for the last line, which you will recognise to be Eq. (10.44) for the companding
gain of a compressor of transfer characteristic specified by output y as a function of input x.
 10.4 Nonuniform Quantisation 649

On the other hand, for large input signals (with peak value V p ≈ 1), SQNRlin > SQNRlog and their difference in
dB is known as companding penalty Lc given by
( )
Δmax
Lc ≡ (SQNRlin − SQNRlog )|x≈1 = 20log10 dB
2∕N
( )
dy ||
= −20log10 dB
dx ||x=1
where we have again used Eq. (10.41) for the last line. Substituting Eqs. (10.34) and (10.40) for the slopes of the
A-law and 𝜇-law functions at x = 1, we obtain
⎧20log (1 + ln A) dB, A-law
⎪ 10
[( ) ]
Lc = ⎨ 1 (10.46)
⎪20log10 1+ ln(1 + 𝜇) dB, 𝜇-law
⎩ 𝜇
Putting A = 87.6 and 𝜇 = 255 in the above equations yields companding penalty Lc = 14.8 dB for A-law and
Lc = 14.9 dB for 𝜇-law. What this means is that the SQNR improvement (by 24 dB in A-law and 30 dB in 𝜇-law)
achieved by a log-quantiser for small input signals is through sacrificing signal quality at the top end of input
(by ∼15 dB in both A-law and 𝜇-law). This results in a lower but consistent SQNR across the entire input range
which is more satisfying in communication services than the situation that obtains in uniform quantisation where
SQNR is high at the top end and poor at the bottom end. Log-quantisation is truly an ingenious way of imitating
the familiar societal practice of taking a little more from the rich or strong (≡ companding penalty) to benefit the
poor or weak (≡ companding gain).
SQNRlin (in dB) decreases proportionately with peak input signal level V p (in dB), whereas SQNRlog remains
roughly constant as V p decreases until V p ≪ 1 (or more precisely for A-law, V p ≡ V th = 1/A) when log-quantisation
is abandoned and linear quantisation is adopted with a fixed step size Δmin , so that SQNRlog also begins to decrease
in step with V p from this point. Figure 10.9 illustrates this variation of SQNR with V p in a linear quantiser and

SQNR (dB)
SQNRlinmax

Lc

Log-quantiser
SQNRlogmax
Li
ne
ar
qu
an Gc
tis
Telegram: @Persian_IEEEComSoc

er

S Gc

Vp
0 Vth Vlog
Peak input level, Vp (dB relative to quantiser limit C)

Figure 10.9 SQNR versus peak input level for linear and log-quantisers. Lc is the companding penalty and Gc the
companding gain.
 650 10 Digital Baseband Coding

also in a log-quantiser that utilises log-quantisation up to the point V p = V th and linear quantisation below V th .
Both quantisers operate with the same number of quantisation levels N, and therefore the same number of bits
per sample k. Note in this graph that V p decreases from left to right along the x axis so that V log < V th < 0. The
linear and log-quantisers have respective maximum SQNR denoted SQNRlinmax and SQNRlogmax . The difference
between these two values is the companding penalty Lc defined earlier
Lc = SQNRlinmax − SQNRlogmax dB
At V p = V th , the SQNR of the log-quantiser is still SQNRlogmax but the SQNR of the linear quantiser has dropped
to S. The difference between these two SQNR values is the companding gain Gc earlier defined
Gc = SQNRlogmax − S dB
⇒ S = SQNRlogmax − Gc dB
Beyond V th , the log-quantiser’s SQNR decreases linearly in step with V p , reaching the value S at V p = V log . Thus
S = SQNRlogmax + Vlog − Vth dB
Equating the right-hand side of the last two equations yields
Gc = Vth − Vlog dB (10.47)
Equation (10.47) suggests the following equivalent definition of companding gain: to achieve the same SQNR for
small input signals, a linear quantiser requires a higher peak input signal level than a log-quantiser operating with
the same number of bits/sample. The dB difference between the peak input signal level of the linear quantiser
and the peak input signal level of the log-quantiser when their SQNR are equal is the companding gain of the
log-quantiser.

10.4.5 Practical Nonlinear PCM

In practice, nonlinear A-law and 𝜇-law companding is implemented using the piecewise linear approximation
shown in Figure 10.10a,b for A-law and 𝜇-law, respectively. Details of this piecewise companding scheme are
presented in Tables 10.4 and 10.5 for the two laws. Consider first the A-law scheme in Figure 10.10a and Table 10.4.
The input signal is normalised to the range ±C, where C = 4096. This is simply to scale all step sizes into integer
numbers. The magnitude of negative and positive input signal values is processed in the same way, and in the
output code the MSB is set to 1 for positive values and 0 for negative. Figure 10.10a, Table 10.4 and the following
discussion show how the positive input values (0 → 4096) are handled.
The A-law curve is approximated using a series of eight straight-line segments covering the input intervals
(0 → 32), (32 → 64), (64 → 128), (128 → 256), (256 → 512), (512 → 1024), (1024 → 2048), (2048 → 4096). These seg-
ments are numbered from 0 to 7 in the first column of Table 10.4. Each segment is divided into 16 equal intervals,
numbered from 0 to 15 in column four of the table. The first two segments (s = 0, 1) have equal intervals or step
Telegram: @Persian_IEEEComSoc

sizes, normalised to 2, whereas in the remaining six segments (s = 2 to 7) the step size in one segment is double
that of the previous segment. The compression line for segments 0 and 1 is a single straight line that straddles the
origin and covers the input range −64 to +64. Thus, there are 13 unique line segments in all.
The range of input values that constitutes a quantisation interval in each segment is given in the third column
of Table 10.4, which also shows the 128 output levels available to the positive input signal, numbered from 0 to 127
in column five. There are a further 128 output levels, not shown, for negative input signals. This gives 256 output
levels in total, which can therefore be represented using 8 bits.
Observe that segments s = 0 and 1 have the smallest step size Δmin = 2. If the entire input range (±C) were
quantised uniformly using this step size, the number of quantisation levels would be
2C 2 × 4096
N= = = 212
Δmin 2
 10.4 Nonuniform Quantisation 651

128
A-law function
s7
112
Piecewise linear
s6 approximation
96
s5
80
Output, y

s4
64

s3
48
s2
32
s1
16
s0
0
256 512 1024 2048 4096
0, 32, 64, 128 Input, x
(a)
Output, y
128
112
96
80
64
48
32 32 96
224
16
–8160 –4064 –2016 –992 480
Input, x
992 2016 4064 8160
–16
–32
Telegram: @Persian_IEEEComSoc

–48
–64
–80
–96
–112
–128
(b)

Figure 10.10 (a) A-law piecewise companding (A = 87.6). (b) 𝜇–law piecewise companding (𝜇 = 255).
 652 10 Digital Baseband Coding

Table 10.4 A-law coding and decoding specifications.

Segment (s) Step size, (𝚫) Input range, (X) Interval (l) Output (256 levels), (Yq ) Output code, (8-bit) Receiver output, (Xq )

0 2 0–2 0 0 1 000 0000 1

⋮ ⋮ ⋮ ⋮ ⋮
30–32 15 15 1 000 1111 31
1 2 32–34 0 16 1 001 0000 33
⋮ ⋮ ⋮ ⋮ ⋮
62–64 15 31 1 001 1111 63
2 4 64–68 0 32 1 010 0000 66
⋮ ⋮ ⋮ ⋮ ⋮
124–128 15 47 1 010 1111 126
3 8 128–136 0 48 1 011 0000 132
⋮ ⋮ ⋮ ⋮ ⋮
248–256 15 63 1 011 1111 252
4 16 256–272 0 64 1 100 0000 264
⋮ ⋮ ⋮ ⋮ ⋮
496–512 15 79 1 100 1111 504
5 32 512–544 0 80 1 101 0000 528
⋮ ⋮ ⋮ ⋮ ⋮
992–1024 15 95 1 101 1111 1008
6 64 1024–1088 0 96 1 110 0000 1056
⋮ ⋮ ⋮ ⋮ ⋮
1984–2048 15 111 1 110 1111 2016
7 128 2048–2176 0 112 1 111 0000 2112
⋮ ⋮ ⋮ ⋮ ⋮
3968–4096 15 127 1 111 1111 4032

which would require 12 bits to represent. Thus, the small input values have an equivalent of k = 12 bits/sample
linear quantisation, which we may express as
kmax = ⌈log2 (2C∕Δmin )⌉ (10.48)
Telegram: @Persian_IEEEComSoc

Segment 7, on the other hand, has the maximum step size Δmax = 128. If the entire input range were uniformly
quantised at this spacing, the number of output levels would be
2C 2 × 4096
N= = = 64 = 26
Δmax 128
Thus, the large input values are coarsely quantised at a resolution equivalent to k = 6 bits/sample linear quan-
tisation, or
kmin = ⌈log2 (2C∕Δmax )⌉ (10.49)
The binary code b7 b6 b5 b4 b3 b2 b1 b0 for each of the 256 output levels is given in column six of Table 10.4. This
code is determined as follows.
 10.4 Nonuniform Quantisation 653

Table 10.5 𝜇-law coding and decoding specifications.

Segment (s) Step size, (𝚫) Input range, (X) Interval (l) Output (256 levels), (Yq ) Output code, (8-bit) Receiver output, (Xq )

0 2 0–2 0 0 1 000 0000 1

⋮ ⋮ ⋮ ⋮ ⋮
30–32 15 15 1 000 1111 31
1 4 32–36 0 16 1 001 0000 34
⋮ ⋮ ⋮ ⋮ ⋮
92–96 15 31 1 001 1111 94
2 8 96–104 0 32 1 010 0000 100
⋮ ⋮ ⋮ ⋮ ⋮
216–224 15 47 1 010 1111 220
3 16 224–240 0 48 1 011 0000 232
⋮ ⋮ ⋮ ⋮ ⋮
464–480 15 63 1 011 1111 472
4 32 480–512 0 64 1 100 0000 496
⋮ ⋮ ⋮ ⋮ ⋮
960–992 15 79 1 100 1111 976
5 64 992–1056 0 80 1 101 0000 1024
⋮ ⋮ ⋮ ⋮ ⋮
1952–2016 15 95 1 101 1111 1984
6 128 2016–2144 0 96 1 110 0000 2080
⋮ ⋮ ⋮ ⋮ ⋮
3936–4064 15 111 1 110 1111 4000
7 256 4064–4320 0 112 1 111 0000 4192
⋮ ⋮ ⋮ ⋮ ⋮
7904–8160 15 127 1 111 1111 8032

1. The MSB b7 is set to 1 for a positive input value and 0 for a negative input value.
2. b6 b5 b4 is the binary number equivalent of segment s. For example, if s = 6 then b6 b5 b4 = 110; and if s = 3 then
b6 b5 b4 = 011.
Telegram: @Persian_IEEEComSoc

3. b3 b2 b1 b0 is the binary number equivalent of the interval l within a segment. For example, if l = 13 then
b3 b2 b1 b0 = 1101; and if l = 2 then b3 b2 b1 b0 = 0010.

At the receiver, this code is converted to the value shown in the last column of Table 10.4, which is the midpoint
of the interval (column three) in which the original input sample falls.
Before moving on to 𝜇-law coding, it is worth highlighting again that in A-law PCM the largest input samples
are coarsely quantised at a resolution of 6 bits/sample, whereas the smallest input samples are finely quantised at a
resolution of 12 bits/sample. This means that, compared to 8-bit linear ADC, A-law PCM delivers a 4-bit improve-
ment for small inputs at the bottom end, which, since each extra bit yields a 6 dB increase in SQNR, corresponds
to a companding gain of 24 dB, as discussed in Section 10.4.4. Notice, however, that this improvement is achieved
at the expense of a 2-bit shortage for large input signals at the top end, which corresponds to a companding penalty
 654 10 Digital Baseband Coding

of 12 dB. This companding penalty is less than 15 dB, as calculated in Section 10.4.4, because the piecewise linear
approximation to the compression curve leads to a smaller maximum step size than in the exact curve.
𝜇-law piecewise companding (Figure 10.10b and Table 10.5) follows a similar strategy to the A-law, but with
several important differences. Note that the input signal is normalised to the range ±8160 to allow the use of
integer numbers for the step sizes. Of the 16 segments, the two that straddle the origin are co-linear, giving 15
unique segments. Starting from segment 0 (for the positive signal range), the step size of each segment is double
that of the previous one.
Following Eqs. (10.48) and (10.49), we see that the 𝜇-law scheme is equivalent to a fine linear quantisation of
the smallest input values (in segment s = 0) using
⌈ ( )⌉
2 × 8160
kmax = ⌈log2 (2C∕Δmin )⌉ = log2
2
= 13 bits∕sample
and a coarse linear quantisation of the largest input values (in segment s = 7) using
⌈ ( )⌉
2 × 8160
kmin = ⌈log2 (2C∕Δmax )⌉ = log2
256
= 6 bits∕sample
An important summary of the segments and step sizes used in A-law and 𝜇-law PCM is provided in Table 10.6.
In Worked Example 10.5, we show how the information in this table is used for PCM coding and decoding.
Piecewise linear companding may be viewed as a two-step process (Figure 10.6b) of fine uniform quantisation
and digital translation. This view is demonstrated in Figure 10.11 for A-law piecewise companding. Note that the
results of Figure 10.11 and Table 10.4 are the same.
Figure 10.12 provides a summary of the steps involved in converting an analogue sample to a PCM code. In the
coding process, the analogue sample x(n) is first scaled by a factor of F for the input signal x(t) to fully load the
quantiser, which has an input range of −C to +C. In our case, in order to use Table 10.6, the value of C is 4096 for
A-law PCM and 8160 for 𝜇-law. The scaled sample, denoted X(n), is converted in the quantiser to a value X q (n)
equal to the midpoint of the quantisation interval in which X(n) lies. The encoder then converts X q (n) to a binary
code according to the procedure outlined above. It is noteworthy that at every stage of Figure 10.12 the signal is
processed in a reversible manner, except at the quantiser. The original input sample x(n) may be obtained from
X(n), and likewise X q (n) from the binary PCM code b7 b6 b5 b4 b3 b2 b1 b0 . But once the exact sample X(n) has been
converted to the approximate value X q (n), knowledge of X(n) is lost for ever and the incurred quantisation error
is a permanent degradation of the transmitted signal. It is important to keep this error to a minimum.

Table 10.6 Segments and step sizes in A-law and 𝜇-law PCM.

A-law 𝝁-law
Telegram: @Persian_IEEEComSoc

Input range Input range

Segment (s) Step size (𝚫) X s min X s max Step size (𝚫) X s min X s max

0 2 0 32 2 0 32
1 2 32 64 4 32 96
2 4 64 128 8 96 224
3 8 128 256 16 224 480
4 16 256 512 32 480 992
5 32 512 1024 64 992 2016
6 64 1024 2048 128 2016 4064
7 128 2048 4096 256 4064 8160
 10.4 Nonuniform Quantisation 655

Fine-grain Tx digital Rx digital Tx ≡ transmitter;

Analogue A-law PCM Rx ≡ receiver
linear ADC translator translator
Input (k = 13) (13 → 8) (k = 8) (8 → 13) (k = 13)
High resolution (k = 13) Various resolutions Various resolutions
1 0000 0000 0000 2 to 1 1 000 0000 1 0000 0000 0001
32 levels 1 0000 0000 0001 s0 16 levels
1 0000 0001 1111 1 000 1111 1 0000 0001 1111
1 0000 0010 0000 2 to 1 1 001 0000 1 0000 0010 0001
32 levels 1 0000 0010 0001 s1 16 levels
1 0000 0011 1111 1 001 1111 1 0000 0011 1111
1 0000 0100 0000 4 to 1 1 0000 0100 0010
… 1 010 0000
64 levels 1 0000 0100 0011 s2 16 levels
1 0000 0111 1111 1 010 1111 1 0000 0111 1110
1 0000 1000 0000 8 to 1 1 0000 1000 0100
… 1 011 0000
128 levels 1 0000 1000 0111 s3 16 levels
1 0000 1111 1111 1 011 1111 1-to-1 1 0000 1111 1100
1 0001 0000
… 0000 16 to 1 1 100 0000 1 0001 0000 1000
256 levels 1 0001 0000 1111 s4 16 levels
1 0001 1111 1111 1 100 1111 1 0001 1111 1000
1 0010 0000
… 0000 32 to 1 1 101 0000 1 0010 0000 0110
512 levels 1 0010 0001 1111 s5 16 levels
1 0011 1111 1111 1 101 1111 1 0011 1111 0000
1 0100 0000 0000 64 to 1 1 0100 0010 0000
… 1 110 0000
1024 levels 1 0100 0011 1111 s6 16 levels
1 0111 1111 1111 1 110 1111 1 0111 1110 0000
1 1000 0000 0000 128 to 1 1 1000 0100 0000
… 1 111 0000
2048 levels 1 1000 0111 1111 s7 16 levels
1 1111 1111 1111 1 111 1111 1 1111 1100 0000

Figure 10.11 Digital translation view of A-law piecewise companding.

Analogue x(n) X(n) Xq(n) 8-bit PCM code

F Quantiser Encoder
sample
(b7b6b5b4b3b2b1b0 ≡ Ssssllll)

Figure 10.12 Converting analogue sample to PCM code. In the PCM code, S represents a sign bit, sss represents one of 8
segments, and llll represents one of 16 quantisation intervals within each segment.

At the receiver, an incoming PCM code is converted to a quantised value X q (n), which is the midpoint of interval
l within segment s, the values of both l and s being indicated by the code. X q (n) is de-scaled by the factor 1/F to
yield a received signal sample xq (n), which, barring channel-induced errors in the received PCM code, differs from
the original sample x(n) by a quantisation error given by
eq = |xq (n) − x(n)|
Telegram: @Persian_IEEEComSoc

| Xq (n) − X(n) |
| |
=| | (10.50)
| F |
| |
The maximum value of eq , denoted eq max , is half the step size of the segment s in which X(n) lies. It follows from
Tables 10.4 and 10.5 that for A-law PCM
⎧
⎪1∕F, s=0
eq max =⎨ (10.51)
⎪2(s−1) ∕F, s = 1, 2, · · · , 7
⎩
And for 𝜇-law PCM
eq max = 2s ∕F (10.52)
 656 10 Digital Baseband Coding

Since samples of larger magnitude fall in higher segments s, we see that the maximum quantisation error
increases as the sample value increases. This maintains the ratio of sample value to quantisation error and hence
SQNR approximately constant over the entire input signal range.
All the above coding and decoding processes, including the anti-alias filtering and sample-and-hold functions
discussed in Chapter 9, are usually performed by a single large-scale IC unit known as a codec. On the transmit
side the codec takes in an analogue signal and generates a serial bit stream output, being the A-law or 𝜇-law PCM
representation of the analogue signal. On the receive side it takes an incoming PCM serial bit stream and recovers
the original analogue signal, subject of course to quantisation noise and any channel-induced errors in the received
PCM code.

Worked Example 10.5

An analogue signal of values in the range −V p to +V p is to be digitised. Let V p = 5 V and assume that the signal
fully loads the quantiser. Determine
(a) The A-law PCM code for the sample value −4 V
(b) The voltage value to which the A-law code 0 101 1100 is converted at the receiver
(c) The quantisation error incurred in transmitting the sample value 3.3 V using 𝜇-law.

(a) In order to use Table 10.6, we scale the input signal from the range ±V p to the range ±C, where C = 4096 (for
A-law). This requires that every input sample x(n) is multiplied by a factor of
⎧
⎪4096∕Vp , A-law
F=⎨ (10.53)
⎪8160∕Vp , 𝜇-law
⎩
In this case, with V p = 5, F = 819.2. Ignoring for the moment the sign of the input sample x(n) = −4 V, since
that only affects the sign bit in the PCM code, and following Figure 10.12

X(n) = Fx(n) = 819.2 × 4 = 3276.8

Table 10.6 shows that X(n) lies in the segment s = 7. The interval l in which X(n) lies within this segment is
given by the number of steps Δ (= 128 in segment 7) required to reach a value just below or equal to X(n)
starting from the lower limit (X min = 2048) of the segment
⌊ ⌋
X(n) − Xmin
l= (10.54)
Δ
where ⌊y⌋ is the floor operation introduced in Eq. (10.1). Using the above values in Eq. (10.54) gives l = 9.
Telegram: @Persian_IEEEComSoc

Thus, the (negative) sample lies in the interval s = 7; l = 9; and is therefore assigned the 8-bit PCM code
(= Ssssllll)
0 111 1001
where the MSB = 0 for a negative sample. Note that the segment s must be expressed as a 3-bit binary number
sss, and interval l as a 4-bit binary number llll, using leading zeros if necessary. This means, for example, that
s = 0 would be expressed as sss = 000; and l = 2 as llll = 0010
(b) The A-law code 1 101 1100 represents a positive sample X(n) in the interval l = 11002 = 12 and within segment
s = 1012 = 5, where the subscript 2 denotes a binary number equal to the indicated decimal equivalent. The
receiver converts this sample to the mid-point X q (n) of the interval

Xq (n) = Xmin + lΔ + Δ∕2 (10.55)

 10.4 Nonuniform Quantisation 657

where X min is the lower limit and Δ the step size of segment s (= 5 in this case). From Table 10.6
Xmin = 512 and Δ = 32
giving
Xq (n) = 912
De-scaling yields the received sample
xq (n) = Xq (n)∕F = 912∕819.2 = 1.11 V
(c) Proceeding in a similar manner as in (a) for this 𝜇-law case, we obtain the scaling factor F = 1632, and a scaled
sample
X(n) = Fx(n) = (1632)(3.3) = 5385.6
We see from Table 10.6 that X(n) lies in segment s = 7, which has X min = 4064 and Δ = 256. Substituting in
Eq. (10.54) yields the interval l = 5. The mid-point X q (n) of this interval to which X(n) is quantised is given by
Eq. (10.55)
Xq (n) = 4064 + 5 × 256 + 256∕2 = 5472
Thus the quantisation error incurred is, from Eq. (10.50)
| 5472 − 5385.6 |
eq = || | = 52.9 mV
|
| 1632 |
You may prefer to use the following direct formulas that give the sample value xq (n) to which any received
PCM code b7 b6 b5 b4 b3 b2 b1 b0 is decoded
s = 4b6 + 2b5 + b4

l = 8b3 + 4b2 + 2b1 + b0

⎧ s
⎪2 (l + 0.5), A-law, s = 0
(−1)b7 +1 ⎪
xq (n) = × ⎨2s (l + 16.5), A-law, s = 1, 2, · · · , 7 (10.56)
F ⎪
⎪2s+1 (l + 16.5) − 32, 𝜇-law
⎩
where F is the scaling factor given by Eq. (10.53). You should verify that Eq. (10.56) yields the expected result
for the A-law PCM code 1 101 1100 of Worked Example 10.5(b).
Telegram: @Persian_IEEEComSoc

10.4.6 SQNR of Practical Nonlinear PCM

Let us consider the SQNR characteristic of A-law and 𝜇-law PCM as a function of the peak input signal level and
compare this with that of a linear ADC of the same number of bits per sample (k = 8). This important exercise
allows us to examine the extent to which A-law and 𝜇-law PCM achieve the ideal of a constant SQNR independent
of the peak input signal level.
We note earlier that the SQNR of a linear ADC (or uniform quantisation) decreases with the peak value of
the input signal. This follows from Eq. (10.13), the last term of which gives the peak value V p of the input sig-
nal expressed
√ in dB relative to the maximum level C of the quantiser. It is shown in Worked Example 10.2 that
R = 3 for a uniform PDF input signal. Thus, the SQNR of an 8-bit linear ADC decreases exactly as the peak input
signal level (dB) decreases, the maximum SQNR being 48.2 dB for an input signal whose samples are uniformly
distributed in the range −C to C.
 658 10 Digital Baseband Coding

Usually, the input signal is scaled before quantisation in order that the overall peak value of the input signal just
fully loads the quantiser. There will, however, be a large variation in the short-term peak value of the input signal.
For example, the voice level in one conversation may vary between a whisper (with a short-term peak value much
less than C) and a shout when the peak value equals C. The SQNR will be unacceptably low during the ‘whisper’
intervals if 8-bit linear ADC is employed.
To overcome this problem we derived a nonlinear, more specifically logarithmic, PCM scheme that has a con-
stant SQNR at all signal levels, given by Eq. (10.25). Implementing this ideal log-PCM scheme, with N = 2k = 256
and K given by Eq. (10.28) for A-law and Eq. (10.35) for 𝜇-law gives a constant SQNR

( ) ⎧
3N 2 ⎪38.2 dB, for K = 1 + ln(A), A = 87.6
SQNR = 10log10 =⎨
K2 ⎪38.1 dB, for K = ln(1 + 𝜇), 𝜇 = 255
⎩

However, for practical implementation it was necessary to modify the ideal log-PCM in two ways, resulting
in the A-law and 𝜇-law PCM schemes. First of all, small input values were uniformly quantised using a small
step size, which is equivalent to using a linear compression curve in the region where the input value x → 0, as
discussed earlier. Secondly, the step size Δ does not decrease continuously with x as required by the ideal log-PCM.
Rather, Δ only changes in discrete steps and is held constant within specified segments of the input signal, listed in
Table 10.6. This corresponds to a piecewise linear approximation. How is SQNR affected by these modifications?
Let the input signal samples be uniformly distributed in the range −V p to V p . Then, from Worked
Example 10.2(b), the signal power is

Vp2
Signal Power = (10.57)
3

Figure 10.13 shows the first four segments s = 0, 1, 2, 3 of the quantiser, with V p falling in the top segment. Each
segment is of size equal to 16 times its step size. The probability Pj that a sample of the input signal falls in any of
the lower segments j = 0, 1, 2 is simply the ratio between the segment size and V p , whereas the probability P3 that
a sample will fall in the top segment of Figure 10.13 is the ratio between (X3min − V p ) and V p , where X3min is the
lower limit of segment 3. That is

16Δj Vp − X3 min
Pj = ; P3 = ; j = 0, 1, 2
Vp Vp
Telegram: @Persian_IEEEComSoc

Therefore, in the general case, if V p lies in segment s, then it follows from Eq. (10.18) that the MSQE is given by

( ) ( )
∑ Δ2j
1 ∑ 2 16Δj
s−1
Δ2s Vp − Xs min
MSQE = Pj = Δ +
j
12 12 j=0 j Vp 12 Vp
( )
4 ∑ 3 Δs Vp − Xs min
s−1 2
= Δj +
3Vp j=0 12 Vp
 10.4 Nonuniform Quantisation 659

Figure 10.13 First four segments of a non-uniform X3max

quantiser. An input signal with uniformly distributed samples
in the range (−V p to V p ) has its peak value in segment s = 3.

16Δ3

Vp

X2max X3min

16Δ2

X1max X2min
16Δ1
X0max X1min
16Δ0
0
–X0max

–X1max

–X2max

–Vp

–X3max

The ratio between signal power in Eq. (10.57) and this MSQE, expressed in dB, gives the SQNR of the piecewise
linear log-quantiser as
[ ]
Δ2s ∑
s−1
SQNRlog = 30log10 (Vp ) − 10log10 (V − Xs min ) + 4 Δ3j dB (10.58)
4 p j=0

Equation (10.58) applies to both A-law and 𝜇-law PCM. It gives the SQNR of a signal of peak value V p , which
Telegram: @Persian_IEEEComSoc

falls in segment s having lower limit X smin and step size Δs . Note that this result is not affected by normalisation.
For example, we presented the A-law and 𝜇-law schemes (see Table 10.6) based on a quantiser range of ±4096 for
A-law and ±8160 for 𝜇-law. The above expression for SQNR holds even when the quantiser range is arbitrarily
scaled to a different limit (such as ±1), provided all step sizes are also scaled by the same factor to retain the
shape of the compression curve. For comparison, the SQNR of a k-bit linear quantiser of input range ±C with a
uniform-PDF input signal of peak value V p follows from Eq. (10.13) as

SQNRlin = 6.02k + 20log10 (Vp ∕C) dB (10.59)

As an example, let us apply Eq. (10.58) to obtain the SQNR of a signal that fully loads the quantiser (i.e. the
signal has 0 dB peak value relative to quantiser input limit C). For A-law: V p = C = 4096 and lies in segment s = 7
 660 10 Digital Baseband Coding

with step size Δs = 128 and lower limit X smin = 2048. Equation (10.58) yields

SQNRA-law = 30log10 (4096)

[ ]
1282 (4096 − 2048)
− 10 log + 4(23 + 23 + 43 + 83 + 163 + 323 + 643 )
4
= 108.37 − 69.82 = 38.55 dB

For 𝜇-law: V p = C = 8160 and lies in segment s = 7 with step size Δs = 256 and lower limit X smin = 4064.
Equation (10.58) yields

SQNR𝜇-law = 30log10 (8160)

[ ]
2562 (8160 − 4064)
− 10 log + 4(23 + 43 + 83 + 163 + 323 + 643 + 1283 )
4
= 117.35 − 78.85 = 38.50 dB

For the linear quantiser with V p = C and k = 8, Eq. (10.59) yields SQNRlin = 48.16 dB.
As a second example, consider a small input signal of peak value 20 dB below quantiser limit. This means that
20log10 (V p /C) = −20 dB, or V p = C/10 = 409.6 for A-law (which places it in segment s = 4 where Δs = 16 and
X smin = 256); and V p = 816 for 𝜇-law (which places it in segment s = 4 where Δs = 32 and X smin = 480). The SQNR
of this small signal when quantised in each quantiser is therefore
[ 2 ]
16 (409.6 − 256)
SQNRA-law = 30 log(409.6) − 10 log + 4(23 + 23 + 43 + 83 ) = 37.51 dB
4
[ 2 ]
32 (816 − 480) 3 3 3 3
SQNR𝜇-law = 30 log(816) − 10 log + 4(2 + 4 + 8 + 16 ) = 37.15 dB
4
SQNRlin = 48.16 + 20 log(Vp ∕C) = 48.16 − 20 = 28.16 dB

We see that (as a result of the piecewise linear approximation) the SQNR of A-law and 𝜇-law PCM is not quite
constant, but in this case, it is around 1 dB lower for a small input signal whose peak value is 20 dB below quantiser
input limit than for a large signal that fully loads the quantiser. The linear quantiser’s SQNR has, however, dropped
significantly by 20 dB for the small input signal.
The results of Eqs. (10.58) and (10.59) are presented in Figure 10.14 for A-law and 𝜇-law PCM schemes and for
k-bit linear ADC at k = 8, 12, 13, from which we make the following important observations.

● A-law PCM maintains a near constant SQNR of about 38 dB over a wide range of peak input levels. It is
interesting to observe that the SQNR of A-law PCM at a peak input level V p = −36 dB (which corresponds to
V p = X 1max = 64 in the ±4096 normalised range) is 36 dB, a drop of only 2 dB from 38 dB. On the other hand,
Telegram: @Persian_IEEEComSoc

the SQNR of an 8-bit linear ADC at this input level has fallen to an unacceptable value of 12 dB. The input level
X 1max marks the beginning of the linear portion of the A-law compression curve. Equation (10.32) shows that
the A-law curve is linear in the region x ≤ 1/A or −38.9 dB, which corresponds to input values X ≤ 46.76 when
the normalised range ±4096 is used, with A = 87.6. However, in the practical implementation of this curve
using a piecewise linear approximation, the terminal linear portion starts slightly earlier at X = 64, which is
−36.12 dB relative to 4096.
● The SQNR of (8-bit) A-law PCM at input levels below −36 dB is the same as that of a 12-bit linear ADC.
This agrees with our earlier observation that small input values are finely quantised at the equivalent of
12 bits/sample. Inputs to the A-law quantiser at levels below −36 dB effectively see a linear ADC of (nor-
malised) step size Δ = 2, and therefore the SQNR decreases linearly with the peak input level in step with that
of a 12-bit linear ADC.
 10.5 Differential PCM (DPCM) 661

50

Li
ne

Li
ne
ar

ar
A
DC
A-law PCM

A
DC
40

,k

,k
=
12

=
13
30 μ-law PCM
SQNR (dB)

Li
ne
ar
A
DC
,k
20 =
8

GcA-law = 24 dB

10
Gcμ-law = 30 dB

0
0 –10 –20 –30 –40 –50 –60 –70 –80
Peak input level, Vp (dB relative to quantiser limit C)

Figure 10.14 SQNR versus peak input signal level for A-law, 𝜇-law, and linear ADC. The input signal is assumed to have
uniformly distributed samples.

● In the case of a 𝜇-law PCM, small input levels below −48 dB have the same SQNR as a 13-bit linear ADC, which
again confirms our earlier observation that small input samples are quantised in 𝜇-law at the resolution of a
13-bit linear ADC.
● The companding gains of A-law and 𝜇-law PCM schemes are indicated in Figure 10.14 as GcA−law =
24 dB, Gc𝜇−law = 30 dB. As discussed earlier, every 6 dB of companding gain is exchanged for a saving of
1 bit/sample. This is the reason why, at small input levels, 𝜇-law PCM delivers the SQNR of a 13-bits/sample
linear ADC using only 8 bits/sample, a saving of 5 bits/sample. Similarly, A-law PCM achieves the SQNR of a
12-bits/sample linear ADC using only 8 bits/sample, a saving of 4 bits/sample.
● Improvements in the SQNR of A-law and 𝜇-law PCM at low input levels have been achieved at a price. The larger
input levels are more coarsely quantised in log-PCM, compared to linear ADC. The effect of this can be seen
in Figure 10.14, which shows that, for a large input signal that fully loads the quantiser, the SQNR of an 8-bit
linear ADC is better by 10 dB than that of log-PCM. This is the companding penalty earlier discussed. However,
log-PCM gives a subjectively more satisfying result, maintaining a near-constant SQNR over a wide input range.
Telegram: @Persian_IEEEComSoc

10.5 Differential PCM (DPCM)

It was pointed out earlier from Eq. (10.12) that SQNR is inversely proportional to the square of the quantiser range.
This means that an improvement in SQNR can be realised by reducing the range of sample values presented to
the quantiser. Each 6 dB improvement in SQNR can be exchanged for a unit reduction in the number of bits per
sample so that the same SQNR as in a PCM system is achieved using fewer bits/sample and hence a smaller
transmission bandwidth. In fact, the underlying principle in all differential pulse code modulation (DPCM), low bit
rate LBR speech coding, and data compression techniques is to avoid the transmission of redundant information.
The original signal is processed in some way at the transmitter to obtain a signal of reduced information content
 662 10 Digital Baseband Coding

Bandlimited Sampled Prediction

Analogue signal signal error DPCM signal
e(nTs) Quantiser eq(nTs) kfs bits/s
input Lowpass s(t) Sample s(nTs)
Σ Encoder
filter (LPF) and hold +
– (2k levels)

+ +
(a) Transmitter Σ

ŝ(nTs) sq(nTs)
Predictor

Special case: integrator

eq(nTs) sq(nTs) Analogue output

DPCM Reconstruction
Decoder Σ s(t)
signal + + LPF

(b) Receiver fs = 1/Ts Predictor

ŝ(nTs)
Clock
Special case: integrator

Figure 10.15 DPCM system.

that requires a proportionately lower transmission bandwidth. An acceptably close copy of the original signal is
obtained at the receiver by processing the received signal to recover the information.
In DPCM the required quantiser range is reduced by encoding the difference e(nT s ) between the actual signal
sample s(nT s ) and a predicted value ̂s(nT s ) generated by a predictor circuit

e(nT s ) = s(nT s ) − ̂s(nT s ) (10.60)

In a properly designed predictor, the error e(nT s ) is small, allowing the number of quantiser levels 2k and hence
the number of bits/sample k to be significantly reduced. The sampling rate f s may be chosen not just to satisfy
the sampling theorem but also to be above the rate that would be used in ordinary PCM. This maintains a high
correlation between adjacent samples and improves prediction accuracy, thereby keeping e(nT s ) small. However,
k is reduced by a larger factor than f s is increased, so that the bit rate Rb = kf s is lower than that of a PCM system
of the same SQNR.
Figure 10.15 shows a block diagram of a DPCM system. The lowpass filter (LPF) serves to minimise aliasing
by limiting the bandwidth of the input signal before it is sampled at intervals T s in a sample-and-hold circuit. A
summing device produces the difference or error e(nT s ) between the sampled signal s(nT s ) and the output of a
predictor. It is this error signal that is quantised as eq (nT s ) and encoded to produce an output DPCM bit stream. If
the analogue signal s(t) changes too rapidly, the predictor will be unable to track the sequence of samples s(nT s )
and the error signal e(nT s ) will exceed the range expected at the quantiser input, resulting in clipping. This type
of distortion is known as slope overload.
Telegram: @Persian_IEEEComSoc

The variance of the error signal e(nT s ) is much smaller than that of s(nT s ). Thus, e(nT s ) can be more accurately
represented than s(nT s ) using the same number of quantisation levels. This implies improved SQNR for the same
bit rate. Alternatively, e(nT s ) can be represented with the same accuracy as s(nT s ) using fewer quantisation levels,
which implies the same SQNR at a reduced bit rate. Note from the block diagram that the input of the predictor
at the transmitter is the original sample plus a small quantisation error
sq (nT s ) = ̂s(nT s ) + eq (nT s )
= ̂s(nT s ) + e(nT s ) + q(nT s )
= ̂s(nT s ) + s(nT s ) − ̂s(nT s ) + q(nT s )
= s(nT s ) + q(nT s ) (10.61)
 10.5 Differential PCM (DPCM) 663

Predictor

sq(nTs) Delay Delay Delay

Delay
Ts Ts Ts Ts

a1 a2 a3 ap

a3sq(nTs – 3Ts)
a2sq(nTs – 2Ts)

a1sq(nTs – Ts)
Σ apsq(nTs – pTs)

ŝ(nTs)

Figure 10.16 Predictor realised using a tapped-delay-line filter.

where q(nTs) is the quantisation error associated with the nth sample.
Identical predictors are used at both the transmitter and the receiver. At the receiver, the DPCM bit stream is
passed through a decoder to recover the quantised error sequence eq (nT s ). This is added to the output of a local pre-
dictor to give a sequence of samples sq (nT s ), according to the first line of Eq. (10.61), which when passed through a
lowpass reconstruction filter yields the original analogue signal s(t), degraded only slightly by quantisation noise.
The predictor is a tapped-delay-line filter, as shown in Figure 10.16. Using the quantised samples sq (nT s ), rather
than the unquantised samples s(nT s ), as the predictor input is important to avoid the accumulation of quantisation
errors. The predictor provides an estimate or prediction of the nth sample ̂s(nT s ) from a linear combination of p
past values of sq (nT s ). It is therefore referred to as a linear prediction filter of order p
̂s(nT s ) = a1 sq (nT s − Ts ) + a2 sq (nT s − 2Ts ) + · · ·
∑
p
= aj sq (nT s − jT s ) (10.62)
j=1

Note that, taking t = nT s as the current sampling instant then s(nT s ) denotes the current sample of signal s(t),
s(nT s − T s ) denotes the previous sample – at time t = (n − 1)T s , and s(nT s − jT s ) denotes a past sample at time
t = (n − j)T s . The optimum values for the coefficients aj (also called tap gains) depend on the input signal. In
Telegram: @Persian_IEEEComSoc

the simplest case, all the coefficients are equal, i.e. a1 = a2 = … = ap , and the predicted value is a scaled sum or
integration of the last p samples. The DPCM can then be implemented using an integrator connected as indicated
in the shaded block in Figure 10.15.
Except for delta modulators discussed later, DPCM systems are in general more complex than PCM since they
require a prediction filter in addition to all the other components of a PCM system. Furthermore, they are subject
to quantisation noise as in PCM, and to slope overload distortion, which is not present in PCM.
However, DPCM offers an important advantage over PCM in that it requires a lower bit rate than a PCM system
of comparable SQNR. The ITU-T has adopted a 32 kb/s DPCM standard (identified as G.726) for voice telephony.
This corresponds to using k = 4 bits/sample at a sampling rate of f s = 8 kHz. The standard provides for opera-
tions at other bit rates, namely 16, 24, and 40 kb/s (corresponding, respectively, to k = 2, 3, and 5 bits/sample at
the same sampling rate). This allows transmission to be adapted to available channel capacity. Another ITU-T
 664 10 Digital Baseband Coding

standard (G.722) specifies the use of DPCM to transmit wideband audio (of bandwidth 7 kHz) at the same bit rate
(64 kb/s) employed by standard PCM for standard telephony speech (of 3.4 kHz bandwidth). In this case, k = 4 and
f s = 16 kHz. Wideband audio gives a significant improvement in the fidelity of the received sound for applications,
such as audio conferences and loud speaking telephones.
There are two special cases of DPCM worth considering further, one known as delta modulation, in which k = 1,
and the other known as adaptive DPCM, in which the step size is adjusted depending on the difference signal
e(nT s ), in order to minimise noise arising from slope overload distortion and quantisation error.

10.5.1 Adaptive Differential Pulse Code Modulation (ADPCM)

In the ITU-T standards referred to above, adaptive quantisation and adaptive prediction are employed to reduce
the required number of bits per sample from k = 8 for standard PCM to about k = 4. Such a system is referred
to as ADPCM. The speech quality of 32 kb/s ADPCM is about the same as 64 kb/s PCM. Ordinary DPCM (with
k = 4, f s = 8 kHz) delivers a poorer quality because the fixed step size Δ would have to be large to avoid overload
distortion, or with a small step size the prediction error e(nT s ) would frequently exceed the quantiser range. It is
for this reason that ordinary DPCM is not used in practice.
Adaptive quantisation uses a time-varying step size Δ in the quantiser. The value of Δ is changed in proportion to
the variance of the prediction error, which is the input signal of the quantiser. In this way the ratio between signal
power and quantisation noise power in the adaptive quantiser is maintained at an acceptable and approximately
constant value.
Adaptive prediction employs samples of the quantiser output as well as past prediction errors to compute opti-
mum values for the predictor coefficients aj . This leads to a significant improvement in prediction accuracy and
allows fewer bits to be used to code the small prediction errors involved.
ITU standard G.721 specifies a transcoding algorithm in which the coder takes A-law or 𝜇-law PCM bit stream as
input and yields an ADPCM output bit stream. At the receiver, a decoder accepts the ADPCM format and converts
it back to PCM.

10.5.2 Delta Modulation

Figure 10.17 shows the block diagrams of the DM transmitter and receiver. By comparing Figure 10.17 with the
DPCM block diagram in Figure 10.15, we see that DM is a special case of DPCM with several distinctive features.
There are only two quantisation levels (i.e. N = 2) in the quantiser, labelled as binary 0 and 1 and separated by
the step size Δ, as shown in Figure 10.18. With N = 2, we know that k = 1 bit/sample. The difference signal e(nT s )
is approximated to the nearer of the two levels, and at each sampling instant the encoder output is a single bit,
which is a binary 1 if e(nT s ) is positive and a binary 0 if e(nT s ) is negative. Let us examine the quantisation error
and important design parameters of the DM scheme.
Telegram: @Persian_IEEEComSoc

10.5.2.1 Quantisation Error

To address this issue, we assume that there is no overload distortion, so that the difference signal e lies within the
range ±Δ. Let P1 be the probability that the difference signal is positive and P2 the probability that it is negative.
Then
P1 + P2 = 1
Consider first the values of e in the positive range 0 to Δ. These are all quantised to output level Δ (represented
as binary 1) with a quantisation error e − Δ. The maximum quantisation error is Δ, which is twice that of PCM.
Following the same arguments leading to Eq. (10.18), we obtain the MSQE for positive differences
Δ
MSQE1 = (e − Δ)2 P1 de∕Δ
∫0
 10.5 Differential PCM (DPCM) 665

Bandlimited Sampled
Analogue signal signal
input s(t) Sampler s(nTs) e(nTs) Quantiser eq(nTs) DM signal
Σ Encoder
LPF fs = 1/Ts + (two levels) fs bits/s
–
(a)
+ +
Σ

sq(nTs)
Delay
sq(nTs – Ts) Ts

eq(nTs)
=±Δ sq(nTs) Reconstruction s(t) Analogue output
DM signal Decoder Σ
(b) + LPF
+
Delay
sq(nTs – Ts) Ts

Figure 10.17 Delta modulation (DM) system: (a) transmitter; (b) receiver.

Figure 10.18 Quantisation levels in DM. Input, e(nTs)

+Δ 1
de
e

–Δ 0

Output

Replacing e − Δ by 𝜀, then de = d𝜀 and the limits e = (0, Δ) become 𝜀 = (−Δ, 0), so that the integration
simplifies to
( )
P1 0 2 P1 𝜀3 ||0
MSQE1 = 𝜀 d𝜀 =
Δ ∫−Δ Δ 3 ||−Δ
P Δ2
Telegram: @Persian_IEEEComSoc

= 1
3
Similarly, the MSQE for negative differences is MSQE2 = P2 Δ2 ∕3. The total quantisation noise power is therefore
MSQE = MSQE1 + MSQE2
P1 Δ2 P2 Δ2 Δ2
= + = (P + P2 )
3 3 3 1
Δ 2
= (10.63)
3
The scheme illustrated in Figure 10.18 and analysed above performs quantisation by truncation. It is stated in
Eq. (10.5) that the quantisation noise power of a truncating quantiser is four times that of a rounding quantiser, the
latter being the approach used in the linear ADC and PCM systems earlier discussed. See, for example, Eq. (10.9).
 666 10 Digital Baseband Coding

Note, however, that the noise power in a DM system is distributed over a frequency range from 0 to f s , which is
much wider than the receiver bandwidth B. The effective noise power is therefore a fraction B/f s of the MSQE.
Using the minimum allowed step size obtained below in Eq. (10.66) for a sinusoidal signal of peak value V m and
frequency f m , we obtain
Signal power 1 2 2
SQNR = = 2 Vm∕Δ B
Effective noise power 3 fs
2 ( )2
3f V fs
= s m ×
2B 2𝜋fm Vm
3fs3
= (10.64)
8𝜋 2 Bf 2m
It can be seen that the SQNR of a DM system increases as the cube of f s , so that doubling the sampling frequency
yields a 9 dB improvement in SQNR.

10.5.2.2 Prediction Filter

A DM system uses a prediction filter of order p = 1, with tap gain a1 = 1. It follows from Eqs. (10.62) and (10.61)
that the predicted sample is simply the previous sample plus a small quantisation noise
̂s(nT s ) = sq (nT s − Ts )
= s(nT s − Ts ) + q(nT s − Ts ) (10.65)
The predictor is therefore a delay line, with delay equal to the sampling interval T s .

10.5.2.3 Design Parameters

A sampling rate f s , much higher than Nyquist frequency, is employed in DM to reduce the difference e(nT s )
between adjacent samples, thereby avoiding overload distortion and minimising quantisation noise. To see the
factors that influence the choice of f s , consider a sinusoidal input signal
s(t) = Vm sin(2𝜋fm t)
The maximum change in s(t) during one sampling interval T s is given by
| ds(t) | 2𝜋fm Vm
| | × Ts = 2𝜋fm Vm Ts =
| dt |
| |max fs
To avoid overload distortion, this change must not exceed the step size Δ. Thus
2𝜋fm Vm
≤Δ (10.66)
fs
Telegram: @Persian_IEEEComSoc

Equation (10.66) is an important result, which provides the interrelationship that must be satisfied by the
parameters of a DM system, namely sampling frequency, step size, message signal frequency, and message sig-
nal amplitude. It is apparent that, irrespective of the values of the other parameters, Eq. (10.66) can be satisfied by
making f s sufficiently large, the penalty being increased bit rate and hence transmission bandwidth.

10.5.2.4 Merits and Demerits of DM

DM has two important advantages over standard PCM.
● It can be realised using a very simple codec that does not require the type of quantiser and encoder found in PCM.
It is therefore more reliable and more economical. The quantiser is a simple pulse generator (or comparator) that
gives an output +V volts (or binary 1) when its input e(nT s ) is positive and −V volts (or binary 0) when e(nT s )
is negative. The predictor uses only a single tap gain and may be replaced by an integrator. A simple analogue
 10.5 Differential PCM (DPCM) 667

Bandlimited Sampled
Analogue signal signal
input s(t) s(nTs)
LPF Sampler +
Comparator DM signal (binary ±V)
(a) – fs bits/s
Clock sq(nTs – Ts)
fs = 1/Ts
Integrator

DM sq(nTs) s(t)
(b) Reconstruction Analogue
signal Integrator output
LPF

Granular Overload
noise distortion

s(t)
(c) Step size, Δ
sq(nTs)
Ts
t
DM signal 1 1 1 1 0 0 0 0 0 1 0 1 0 1 0 1 1 1 1 1

Figure 10.19 Simplified DM system and waveforms: (a) transmitter; (b) receiver; (c) waveforms.

RC LPF could be used. The output voltage of the integrator rises or falls by one step Δ in response to each pulse
input. This enables the integrator to track the input voltage. The receiver then uses a similar integrator whose
output rises by Δ when its input is a binary 1 and falls by the same amount when its input is binary 0. In this
way both integrators produce the same staircase waveform sq (nT s ), and a lowpass reconstruction filter will easily
smooth out the integrator output at the receiver to yield the original analogue signal. Figure 10.19 shows a block
diagram of the simplified DM codec, and the associated waveforms including the generated DM bit stream for
an arbitrary input signal s(t).
● DM is more robust to transmission errors than PCM, and intelligibility can be maintained at bit error rates as
high as one in a hundred (i.e. 10−2 ). Every transmitted bit has the same weight and the maximum error due to
one bit being received in error is equal to a quantiser step size Δ. In PCM, on the other hand, an error in the
MSB can have a significant effect on the received signal, causing an error ranging in magnitude from Δ to 2k Δ.
This robustness to noise makes DM the preferred technique in certain military applications. The US military
Telegram: @Persian_IEEEComSoc

and NATO have operational DM systems at 16 and 32 kb/s.

DM has, however, the following disadvantages:

● Oversampling with f s > > 2f m is required to allow the use of a small step size Δ while ensuring that the
difference between adjacent samples rarely exceeds Δ. With 1 bit/sample and f s samples per second, the bit
rate of DM is f s bits/s, which may in some cases exceed that of standard PCM and therefore require a larger
bandwidth.
● With a fixed step size, the maximum error per sample is constant at ±Δ for all samples. However, while the
quantisation noise power is fixed, the signal power varies across the range of input levels. Thus, SQNR varies
with input level and may fall to unacceptably small values at low signal levels. The useful dynamic range of
input signals over which the SQNR does not change significantly is therefore severely limited.
 668 10 Digital Baseband Coding

● When there is a roughly constant portion of the input signal s(t), the staircase approximation sq (nT s ) hunts about
this level, as shown in Figure 10.19c. This gives rise to a type of noise known as granular noise, which can only
be minimised by making Δ as small as possible, subject to Eq. (10.66). However, a small Δ makes the system
prone to overload distortion, as discussed below. Granular noise manifests itself in the DM bit stream as a string
of alternating 0’s and 1’s.
● Figure 10.19c also shows an incidence of overload distortion in which the input waveform s(t) changes too
quickly to be tracked by sq (nT s ). To avoid overload distortion, Eq. (10.66) must be satisfied in the choice of
step size and sampling frequency. This requires one of the following options:
– Make f s sufficiently large, so that fs ≥ 2𝜋fm Vm ∕Δ. This increases transmission bandwidth requirements.
– Limit the message signal amplitude to Vm ≤ Δfs ∕2𝜋fm , which limits the allowable dynamic range of the input
signal. Note that the allowable peak value Vm decreases with the frequency f m of the input signal.
– Increase the step size to Δ ≥ 2𝜋fm Vm ∕fs . This, however, increases granular noise.
– Limit the input signal frequencies to fm ≤ Δfs ∕2𝜋Vm . The problem here is that the range of f m depends on the
type of information signal and cannot be significantly reduced without compromising fidelity.

10.5.2.5 Adaptive Delta Modulation (ADM)

We have seen the conflicting requirement of a large step size Δ to combat overload distortion, and a small step size
to minimise granular noise. Adaptive delta modulation (ADM) provides for the control of both types of noise by
varying the step size. This technique belongs in the general class of adaptive delta pulse code modulation (ADPCM).
It is introduced here to show how both overload distortion and granular noise may be reduced.
In basic ADM, the step size is changed in discrete amounts. To start with, the step size Δ is set to a very small
value 𝛿 to minimise granular noise. This (the smallest) step size is maintained for as long as the DM bit stream
consists of a string of alternating 1’s and 0’s, or two successive 1’s, or two successive 0’s. After encountering three
successive 1’s or 0’s the step size is increased to Δ = 2𝛿, and Δ is increased even further to 4𝛿 when four consecutive
1’s or 0’s occur. The larger step sizes allow the integrator output, which is the staircase approximation sq (nT s ), to
catch up more quickly with the sequence of input samples s(nT s ).
ADM has also been implemented with the step size continuously varied in proportion to the input signal. This
variant of ADM is referred to as continuously variable slope delta modulation (CVSDM).

10.5.2.6 Delta Sigma Modulation

It was noted earlier that the amplitude of the higher-frequency components in the input signal must be minimised
to avoid overload distortion. This can be done by first passing the input signal through an LPF whose gain decreases
linearly with frequency. Recall that such an LPF is actually an integrator. At the receiver, a highpass filter (HPF)
having a gain response that increases linearly with frequency must be employed to remove the spectral distortion
imposed at the transmitter. Note that this HPF performs the operation of differentiation. However, a DM receiver
already contains an integrator. The resulting receiver will consist of an integrator followed by a differentiator, and
the two components can therefore be eliminated altogether. This leaves a lowpass reconstruction filter as the only
Telegram: @Persian_IEEEComSoc

component required at the receiver. The scheme described above is shown in Figure 10.20 and is traditionally
known as delta sigma modulation (DSM), but should more appropriately be called sigma delta modulation (SDM).
SDM allows the use of a greatly simplified receiver, and the transmission of signals with allowable peak levels that
are independent of frequency.

10.6 Low Bit Rate Speech Coding

The purpose of LBR speech coding is to achieve a digital representation of speech signals with acceptable fidelity
using as few bits as possible. This is essential in bandwidth- and/or power-limited voice communication ser-
vices such as mobile communications and Internet telephony, and in voice storage services such as voicemail,
 10.6 Low Bit Rate Speech Coding 669

Input s(t) Delta SDM signal

(a) Integrator
signal modulator (binary ±V)

SDM Reconstruction s(t) Analogue

(b)
signal LPF output

Figure 10.20 Sigma delta modulation system (SDM): (a) transmitter; (b) receiver.

store-and-forward messaging, and automatic voice response systems. Transmission bandwidth and storage capac-
ity requirements increase proportionately with bit rate. For example, standard PCM with a bit rate of 64 kb/s
requires a minimum transmission bandwidth of 32 kHz, and a storage capacity of 1.44 MB for a call of duration
of three minutes. If we somehow manage to reduce the bit rate to, say, 6.4 kb/s, then the bandwidth and storage
capacity requirements fall proportionately to 3.2 kHz (minimum) and 144 kB, respectively.
Figure 10.21 shows the trade-offs involved in LBR speech coding. The main characteristics against which the
performance of a speech codec is judged include:

● Quality: in general, the achievable speech quality falls as the bit rate is reduced. How much degradation there
is depends on the type of algorithm used to reduce the bit rate. Obviously, if a bit rate of 32 kb/s were achieved
simply by reducing the number of bits per sample from k = 8 to k = 4 in standard PCM, there would be a drop in
SQNR of 24 dB and hence a significant degradation in quality. However, 32 kb/s ADPCM and even some 16 kb/s
codecs provide speech quality that is practically indistinguishable from that of 64 kb/s PCM.
Speech quality is usually specified using a subjective mean opinion score (MOS), which is obtained by averag-
ing the judgement of many listeners expressed in the form of a score on a scale from 1 to 5. Table 10.7 provides a

Standard PCM 64 kb/s

Computational Complexity

Processing Delay
Speech Quality

Sophistication
Transparency
Bandwidth
Bit Rate
Telegram: @Persian_IEEEComSoc

Vocoder 2.4 kb/s

Figure 10.21 Trade-offs in low bit rate speech coding.

 670 10 Digital Baseband Coding

Table 10.7 Speech quality using mean opinion score (MOS).

MOS Quality description Quality class

1 Bad, unacceptable Synthetic

2 Poor
3 Fair Professional
Communication
4 Good Near transparent, toll quality
5 Excellent

classification of the scores. Note, however, that the MOS recorded for a speech codec can vary significantly with
tests and language. An MOS of < 3.0 corresponds to speech of synthetic quality, which may have high intelligi-
bility (i.e. the speech is understandable) but sounds unnatural and lacks the attributes that allow recognition
of the speaker and their emotion. Communication quality speech of an MOS of between 3.5 and 4.0 contains
perceptible, but not annoying, distortion, while speech with perceptible and slightly annoying distortion could
be classed as professional quality, with an MOS of between 3.0 and 3.5. MOS values in excess of 4.0 indicate
high-quality and natural-sounding speech with practically imperceptible distortion. A-law and 𝜇-law PCM pro-
duce high-quality speech with an MOS of about 4.2. MOS scores of 4.0 and above are regarded as toll quality. A
vocoder (discussed below) produces synthetic speech with MOS of about 2.2.
● Transparency: significant reductions in bit rate are achieved by exploiting signal characteristics that are specific
to speech. Such codecs will cause excessive distortion to nonvoice signals, such as voiceband data modem sig-
nals. It then becomes essential that nonvoice signals are recognised and diverted from the codec for processing
using a different algorithm. In general, a codec’s transparency (its ability to handle both speech and nonspeech
signals) improves with bit rate.
● Computational complexity: the amount of computation required to maintain acceptable quality in LBR coded
speech increases significantly as the bit rate is reduced. Sophisticated DSP hardware must be employed to
minimise processing delay, and this increases both codec cost and operational power consumption. The lat-
ter is of concern especially in battery-operated portable units. However, due to advances in integrated circuit
technology, DSP units capable of increasingly complex operations are now available in the market at very low
costs.
● Robustness to transmission errors: as the original speech signal is represented using fewer and fewer bits, the role
of each transmitted bit becomes more crucial, and therefore the impact of transmission errors is more drastic.
For example, a single bit error in a vocoder could render a 20 ms speech segment unintelligible. By contrast, the
Telegram: @Persian_IEEEComSoc

impact of a single bit error in standard PCM is restricted to one sample and will be virtually imperceptible if
it involves the least significant bit (LSB). Error protection of at least some of the more crucial bits is therefore
required in LBR codecs. It is for this reason that the full-rate GSM (i.e. the Global System for Mobile Communica-
tions, the Pan-European digital cellular mobile system) speech codec operating at 13 kb/s devotes an additional
3 kb/s for error protection.
● Processing delay: most LBR codecs must accumulate a statistically significant block of samples (typically over
10–25 ms) in a storage buffer, from which certain parameters are extracted. To this buffering or frame delay
is added the time it takes to complete the processing of the accumulated samples. The total processing delay
may be up to three times the frame delay, depending on computational complexity. An isolated delay of this
magnitude may contribute to an increase in sidetone but is otherwise imperceptible and perfectly acceptable to
a user. However, if such a codec is connected to the public telephone network, or several of them are present in a
 10.6 Low Bit Rate Speech Coding 671

long-distance link then the total delay may lead to annoying echo, necessitating the use of echo control circuits
(e.g. echo suppressors or cancellers).

In addition to the above characteristics, other important considerations include robustness to background noise
and the effect of tandem operation with other codecs. The latter arises because, in many cases, a speech codec must
operate on the output of coders located in other parts of the network. The distortion resulting from such multiple
coding/decoding must not be excessive. The priority assigned to each of these characteristics is determined by the
application. For example, processing delay is not crucial in applications such as voice messaging and videocon-
ferencing. In the latter, the accompanying video signal usually has a much larger processing delay than that of
speech. Extra delay would normally be inserted in the speech processing to synchronise with the processed video.
The three broad classes of coders, namely waveform coders, vocoders, and hybrid coders are briefly introduced in
the following sections.

10.6.1 Waveform Coders

A waveform coder attempts an accurate digital representation of the speech waveform. If this is done directly in
the time domain, the coder is described as a predictive coder. It is also possible to first divide the speech signal into
contiguous frequency bands using bandpass filters, before coding each sub-band (SB) using a separate waveform
coder. The coder is then said to be a frequency domain coder. The main advantage of such an SB coding approach
is that the number of bits per sample assigned to each SB can be varied according to the information content of
the band. Those SBs with negligible information content or energy can be skipped altogether. The result is that
a speech quality comparable to that of predictive waveform coders can be achieved at a lower bit rate. Waveform
coders are generally the least complicated of the three classes of coders. They introduce the lowest processing
delays, and produce speech of high quality but require high bit rates ≥16 kb/s. The distortions at lower bit rates
are excessive, and other techniques must be adopted in order to reduce the bit rate any further while maintaining
acceptable quality.
The most important examples of predictive waveform coders include DM, PCM, and ADPCM, which have been
discussed earlier. The ITU-T G.726 standard, which effectively superseded the G.721 (32 kb/s ADPCM) standard,
provides ADPCM speech coding at variable rates, namely 40, 32, 24, and 16 kb/s. This allows the flexibility of
sacrificing speech quality in order to maintain transmission under heavy traffic conditions. The ITU-T G.727 pro-
vides for coding at the same variable bit rates as in G.726, but uses a technique known as embedded ADPCM. This
involves 32, 16, 8, or 4 uniform quantisation levels, the levels for coarse quantisation being a subset of levels in
finer quantisation. Using all 32 levels, requiring 5 bits/sample, gives the best speech quality at a bit rate of 40 kb/s.
As traffic load increases, the network can adopt lower bit rates by simply discarding one, two, or three of the less
significant bits. Discarding the LSB, for example, effects coarser quantisation using only 16 of the 32 possible levels,
with the step size doubled.
Telegram: @Persian_IEEEComSoc

The ITU-T G.722 standard for coding wideband (7 kHz) speech at rates of 64, 56, and 48 kb/s is a good example
of an SB waveform coder. The input speech signal, sampled at 16 kHz with a resolution of 14 bits/sample, is split
into two 4 kHz bands. The lower band, which contains most of the information, is sampled at 8 kHz and coded
using embedded ADPCM at 6, 5, or 4 bits/sample. This yields lower-band coding bit rates of 48, 40, or 32 kb/s.
ADPCM at a resolution of 2 bits/sample is used for the upper band, which gives a bit rate of 16 kb/s. Thus, a total
bit rate of 64, 56, or 48 kb/s is achieved.

10.6.2 Vocoders
A vocoder, or vocal tract coder, does not code the speech waveform at all. Rather, it explicitly models the vocal tract
as a filter in order to represent the speech production mechanism. At the receiver, the representation is animated
 672 10 Digital Baseband Coding

to reproduce a close version of the original speech signal. The main source of degradation in reproduced speech
quality is not quantisation error but the use of a model that cannot accurately represent the speech production
mechanism of the vocal tract. For example, each segment of speech (of duration, say, 20 ms) is classified as either
voiced or unvoiced and reproduced at the receiver by exciting the filter with a pulse train or random noise, respec-
tively. In practice, the speech segment will be partially voiced and unvoiced. This ‘black or white’ classification
alone gives a vocoder a synthetic speech quality. Vocoders do give good intelligibility at very LBRs (e.g. 2.4 and
4.15 kb/s) and therefore are widely used in applications that require intelligibility without a strong need for speaker
identity/emotion recognition. However, they involve complex processing and introduce delays above about 20 ms.
Three types of vocoders are briefly discussed below.

10.6.2.1 IMBE
The improved multiband excited (IMBE) vocoder employs filters to break the speech signal into narrow frequency
bands, each spanning about three harmonics. The energy in each band is measured and a voicing decision is made
that identifies each band as voiced or unvoiced. This information (namely the energy and voice flag of each band)
is coded and transmitted. At the receiver, each filter is excited at its identified energy level using a sinusoidal
oscillator for a voiced band and a noise source for an unvoiced band. The synthesised speech signal is obtained as
the sum of the output of all the filters.

10.6.2.2 LPC
The linear predictive coding (LPC) vocoder was developed by the US Department of Defense (DoD) and adopted
as Federal Standard (FS) 1015. Figure 10.22 shows a block diagram of the speech synthesis model. LPC-10 uses
a predictor (i.e. filter) of order p = 10 for voiced speech and p = 4 for unvoiced speech. At the transmitter, each
22.5 ms segment of the speech signal is analysed to obtain the following parameters:

● A voiced/unvoiced flag, which is coded with 1 bit.

● A gain factor G to model the energy of the segment, which is quantised using 5 bits.
● A pitch period in the range 51.3– 400 Hz, which is quantised using 6 bits.
● Ten predictor coefficients a1 to a10 , which are quantised using 5 bits each for a1 , a2 , a3 , and a4 . If it is a voiced
segment then a further 21 bits are used to code six more coefficients, with 4 bits used for each of a5 to a8 , 3 bits

Pitch period

Voiced/unvoiced Filter
switch coefficients

Pulse
generator Gain
Telegram: @Persian_IEEEComSoc

Filter ŝ(n) ŝ(t)

G DAC
(vocal tract) Synthesised
speech

White noise
generator

Figure 10.22 LPC speech synthesis model.

 10.6 Low Bit Rate Speech Coding 673

used for a9 , and 2 bits for a10 . Higher coefficients are not used in the case of an unvoiced segment, and the 21 bits
are devoted to error protection.

One bit, alternating between 1 and 0, is added to each segment for synchronisation at the receiver. Thus, we
have 54 bits per 22.5 ms, or a bit rate of 2400 b/s. The above parameters are used at the receiver to synthesise the
speech signal, as shown in Figure 10.22.

10.6.2.3 MELP
A 2.4 kb/s vocoder was adopted in 1996 to replace FS 1015. It achieves professional speech quality by using mixed
excitation linear prediction (MELP). The excitation consists of a mixture of periodic pulses and white noise, which
varies across the frequency band according to an estimate of the voiced speech component within the bands.

10.6.3 Hybrid Coders

This class of LBR coders is so named because it combines the advantages of waveform coders and vocoders to
achieve good reproduction of speech at LBRs. Hybrid coders use a technique called analysis-by-synthesis. At the
transmitter various combinations of the model parameters are used to synthesise the speech as would be done
at the receiver, and the combination yielding the best perceptual match to the original speech signal is selected.
Hybrid coders are highly complex and introduce processing delays at least of the same order as vocoders. The
distinction amongst various implementations of hybrid coders lies mainly in the excitation signal pattern, the
analysis procedure, and the type of information transmitted to the receiver.

10.6.3.1 APC
Adaptive predictive coding (APC) is used in the toll-quality 16 kb/s Inmarsat-B standard codec. Here segments of
the speech signal are processed to remove voice model information leaving a residual signal, a quantised version of
which is scaled by various trial gains and used as the excitation signal to synthesise the speech segment. The scaled
residual yielding the best perceptual match to the original speech segment is chosen and is encoded along with the
voice model parameters and transmitted. At the receiver, a good copy of the original speech signal is reproduced
using the scaled residual signal as the excitation signal of a filter constituted from the voice model parameters.

10.6.3.2 MPE-LPC
Multipulse excited linear predictive coding (MPE-LPC) employs perceptual weighting to obtain an improved mod-
elling of the excitation. This signal is constituted as a sequence of pulses with amplitude, polarity, and location
determined to minimise the total weighted squared error between synthesised and original speech signals. The
9.6 kb/s Skyphone codec, developed by BT (formerly British Telecom), is based on this algorithm. It was adopted
by Inmarsat and the Airlines Electronic Engineering Committee (AEEC) for aeronautical satellite communica-
Telegram: @Persian_IEEEComSoc

tions. If in MPE the (unequal amplitude) pulses are regularly spaced within the excitation vector, we have what
is termed regular pulse excitation (RPE). Regular pulse spacing shortens the search procedure and hence reduces
processing delay. RPE, combined with long-term prediction (LTP), is the basis of the 13 kb/s speech codec which
was standardised for GSM.

10.6.3.3 CELP
Code excited linear prediction (CELP) allows the use of an MPE with lower bit rates. The excitation vector that gives
the best perceptual match between synthesised and original speech signals is selected from a set or codebook of
vectors. The excitation signal is usually a weighted sum of contributions from two codebooks, one codebook being
adaptive and constantly updated with past excitation vectors and the other consisting of zero-mean unit-variance
random sequences. The saving in bit rate comes about because fewer bits can be used to code the addresses of the
 674 10 Digital Baseband Coding

selected vectors, allowing the excitation signal to be reconstituted at the receiver from local copies of the codebooks.
An example of a CELP-based codec is the ITU-T G.729 conjugate structure algebraic (CSA) CELP codec, which
delivers toll-quality speech at only 8 kb/s.

10.7 Line Codes

Line codes are introduced in Chapter 1 as a means of electrically representing a bit stream in a digital baseband
communication system. In this section we want to familiarise ourselves with simple line codes and to develop an
appreciation of their merits and demerits in view of the desirable characteristics of line codes. You may wish to
refer to Chapter 1 for a discussion of these characteristics. A more detailed treatment of line codes, including their
bit error ratios and spectral analysis, is provided in Chapter 7 of [1]. Figure 10.23 shows waveforms of the line
codes discussed below, under various classes, for a representative bit stream.

10.7.1 NRZ Codes

Non-return-to-zero-level (NRZ-L): there are two types of NRZ-L, namely unipolar and bipolar NRZ-L. In unipolar
NRZ-L, binary 1 is represented by a positive voltage pulse +V lasting for the entire bit interval 𝜏, and a binary
0 is represented by no pulse (0 V) during the bit interval. For this reason, unipolar NRZ-L is also referred to as
on–off signalling. Bipolar NRZ-L represents binary 1 with a positive voltage pulse +V, and binary 0 with a negative
voltage pulse −V, each pulse lasting for the entire bit interval. Note that NRZ codes are so called because the code
level remains constant within a bit interval and does not return to zero.

Bit stream 1 0 1 0 0 0 0 0 1 1 0 0 0 0 1 1 0 0 0 0 0
Unipolar NRZ-L V
(On-off signalling) 0
V
Bipolar NRZ-L
–V
V
NRZ-M
0
V
NRZ-S
0
V
RZ
0
V
AMI 0
–V
V
Biphase-L
(Manchester)
–V
V
Telegram: @Persian_IEEEComSoc

Biphase-M
–V
V
Biphase-S
–V
V
CMI
–V
V
B V
HDB3 0
–V V B V
V
3B4B
–V
Bit stream 1 0 1 0 0 0 0 0 1 1 0 0 0 0 1 1 0 0 0 0 0

Figure 10.23 Common line codes.

 10.7 Line Codes 675

By considering the desirable characteristics of a good line code, you can readily convince yourself that although
NRZ-L has 100% efficiency it nevertheless exhibits numerous unsatisfactory features. It is usually seen as a basic
data format and referred to as uncoded data. It is used only for transmission over a very short distance, such as
within one piece of equipment. A separate clock signal must be used since the NRZ-L, M, and S codes have no
guaranteed clock content. Another of the many undesirable features of this code is that a level inversion during
transmission causes all symbols to be incorrectly received. A differential NRZ code overcomes this problem.
Differential NRZ: bits are coded using voltage transitions, rather than actual voltage levels as in NRZ-L. There
are two types of differential NRZ. In NRZ-mark (NRZ-M), there is a transition at the beginning of a bit interval if
the bit is 1 (called mark in the days of telegraphy), and no transition if the bit is 0. The other type is NRZ-space
(NRZ-S), which codes a binary 0, formerly called space, with a transition at the beginning of the bit interval, and
a binary 1 with no transition. Therefore, denoting the current input bit as x(n), the current output of the coder as
y(n), and the previous coder output as y(n − 1), we may summarise the coding rule of differential NRZ as follows:

Output, y(n)
Input
x(n) NRZ-M NRZ-S

0 y(n − 1) y(n − 1)
1 y(n − 1) y(n − 1)

where the overbar denotes a complement operation (i.e. change of state), so that V = 0 and 0 = V. Note that x(n)
is binary, having two possible values: 0 and 1, and the code y(n) is also binary, with two possible voltage levels: 0
and V. In Figure 10.23 it is assumed that the output is initially high, that is y = V before the first input bit. You
may wish to verify that if y is initially at a low level (0 V) then the resulting code waveform for NRZ-M and NRZ-S
is the complement of the corresponding waveforms shown in Figure 10.23.
A major handicap with NRZ codes is that they have very poor and nonguaranteed timing content. A long run
of the same bit in NRZ-L, 0’s in NRZ-M and 1’s in NRZ-S, gives rise to a transmitted waveform that is void of level
transitions. It is then impossible to extract the clock signal from the received waveform at the receiver. A small
improvement in timing content is provided by the return-to-zero (RZ) code, although at the cost of doubling the
bandwidth requirement compared to NRZ.

10.7.2 RZ Codes
Return-to-zero (RZ): the RZ code represents binary 1 using a voltage pulse of amplitude +V during the first half of
the bit interval followed by 0 V (no pulse) in the remaining half of the bit interval. Binary 0 is represented as no
pulse (0 V) for the entire bit interval.
Telegram: @Persian_IEEEComSoc

Observe that a long run of 0’s in the RZ code will still cause timing problems at the receiver. Furthermore,
the code exhibits another serious handicap in that it has a DC component (average value) that depends on the
fraction of 1’s in the transmitted data. Most links for long-distance data transmission incorporate transformers
and series capacitors, which will effectively block any DC component in a message signal. A long run of 1’s in
RZ code introduces a DC offset that may result in droop and baseline wander due to charging and discharging
capacitors. Another code, the alternate mark inversion (AMI) eliminates the DC offset problem, but at the expense
of increased codec complexity.
AMI: this code is obtained from RZ by reversing the polarity of alternate binary 1’s. As a result, three voltage
levels (−V, 0, and +V) are employed, making this a three-level, or ternary, code.
AMI eliminates droop and base-line wander while keeping the bandwidth requirement the same as in RZ. How-
ever, a train of data 0’s still results in a transmitted waveform without transitions, which causes timing problems
 676 10 Digital Baseband Coding

when the receiver tries to establish bit synchronisation. The problem of a lack of transitions in some bit sequences
is overcome in biphase codes, which contain a transition in each symbol whether it represents a binary 1 or 0.

10.7.3 Biphase Codes

There are three types of biphase codes, namely biphase-L, which is also called Manchester code, biphase-M, and
biphase-S. In biphase-L, a binary 1 is represented by a transition from high (+V) to low (−V) at the middle of the
bit interval, whereas a binary 0 is represented by a transition from low (−V) to high (+V) at the middle of the
bit interval. That is, to represent a binary 1, a positive pulse +V is transmitted for the first half of the bit interval
followed by a negative pulse −V for the remaining half of the bit interval. A symbol of opposite polarity is used to
represent a binary 0.
Manchester code finds application in the ethernet standard IEEE 802.3 for local area networks (LANs). It has
good timing content but suffers the same polarity inversion problem as NRZ-L, a handicap eliminated in the other
biphase code types.
In biphase-M, there is always a transition at the beginning of each bit interval. Binary 1 is represented with
an additional transition at the middle of the bit interval, whereas binary 0 has no additional transition. Biphase-S
similarly always has a transition at the beginning of each bit interval but uses an additional transition at the middle
of the bit interval for a binary 0 and no additional transition for a binary 1.

10.7.4 RLL Codes

Run-length-limited (RLL) codes limit the length of a run of voltage levels void of transition. One type of imple-
mentation is called bipolar with n zeros substituted (BnZS), e.g. B3ZS, B4ZS, B6ZS, and B8ZS. The case of n = 4 (i.e.
B4ZS) is also called high-density bipolar with 3 zero maximum (HDB3).
Coded mark inversion (CMI): this is the simplest type of RLL code. It is a binary code in which binary 1 is
represented by full width alternating polarity pulses, and binary 0 by −V volts for the first half of the bit interval
followed by +V volts for the remaining half. CMI is preferred to HDB3 in high bit rate systems because of its
simplicity and efficiency.
HDB3: this is an AMI code in which the number of successive 0’s transmitted is somehow limited to three. The
fourth of four adjacent 0’s is represented using a violation (V) pulse. That is, it is represented as a 1 that violates the
alternating polarity rule. In this way, a level transition is introduced, and the receiver will recognise the violating
pulse as representing a binary 0. A little thought will show, however, that violation pulses can cause a build-up of
DC offset. To avoid this, whenever necessary, a balancing (B) pulse is used for the first of four adjacent zeros in
order to prevent two successive violation pulses having the same polarity. To identify a B-pulse, note that whenever
a V-pulse occurs after only two zeros, the previous pulse is a B-pulse, and therefore represents a binary 0. These
pulses are labelled in Figure 10.23.
Let us take a moment to understand HDB3 coding. Figure 10.24 shows the same bit stream as before, with the
Telegram: @Persian_IEEEComSoc

bit positions numbered for easy reference. The AMI waveform for this bit stream is repeated in Figure 10.24a. Note
how binary 1’s are represented using pulses of alternating polarity, so that the average value of the waveform is
always approximately zero. Thus, unlike RZ code, there is no DC build-up in the waveform as binary 1’s are coded.
However, there is a string of 0’s from bit 4 to bit 8, and the AMI waveform is void of transition during the entire
interval.
To overcome this problem of poor timing content, HDB3 limits the maximum run of 0’s to three by changing
the fourth zero in a four-zero-string to binary 1 and representing it with a V-pulse to enable the receiver to spot
the substitution. The implementation of this idea is shown in Figure 10.24b. Note how positions 7, 14, and 20 are
coded. These are V-pulses because each of them has the same polarity as the previous pulse, whereas the rule
stipulates an alternating polarity. Now we have solved the problem of poor timing content and can guarantee that
in a four-bit interval there will be at least one transition in the code waveform. However, there is a new problem.
 10.7 Line Codes 677

Bit stream 1 0 1 0 0 0 0 0 1 1 0 0 0 0 1 1 0 0 0 0 0

+V
(a) 0
–V

+V
(b)
V V V
–V

+V
B V
(c)
V B V
–V

Bit position 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

Figure 10.24 Steps leading up to HDB3 code: (a) AMI; (b) AMI with fourth zero in a four-zero-string represented with a
violation (V) pulse; (c) balancing (B) pulse introduced to complete HDB3 coding.

The code waveform contains three positive and six negative pulses, and so contains an undesirable DC component.
Observe that this has arisen because successive V-pulses have the same polarity. To eliminate this build-up of DC
off-set in the waveform, we make the following modification that prevents successive V-pulses from having the
same polarity: before inserting a new V-pulse we check to see if it would have the same polarity as the previous
V-pulse. If so, we change the first zero in the four-zero-string to a 1; represent this 1 with a pulse that obeys the
alternating polarity rule and call it the B-pulse, and then insert the new V-pulse to violate this B-pulse.
Figure 10.24c shows the implementation of the above modification to obtain the desired HDB3 waveform.
Let’s walk through this waveform. The first V-pulse at No. 7 does not have the same polarity as the previous
V-pulse (in this case simply because there is no previous V-pulse), so no modification is necessary. Next, there is a
four-zero-string from No. 11–14, and a V-pulse is therefore required at No. 14. This would have the same polarity
as the previous V-pulse at No. 7. A modification is needed. Therefore, insert a B-pulse at No. 11 and then insert
the V-pulse at No. 14 to violate this B-pulse. After this, there is a four-zero-string from No. 17– 20, meaning that a
V-pulse is needed at No. 20. We see that this V-pulse would have to be positive (in order to violate the pulse at No.
16), and therefore would have the same polarity as the previous V-pulse at No. 14. Therefore, insert a B-pulse as
shown at No. 17 before inserting the V-pulse at No. 20.
Note that at the receiver it is a straightforward matter for the decoder to correctly interpret the code waveform
Telegram: @Persian_IEEEComSoc

as follows:
● Every 0 V (for the entire bit interval) represents binary 0.
● Every V-pulse represents binary 0.
● Every pulse that is followed by a V-pulse after only two bit intervals is a B-pulse and hence represents binary 0.
● Every other pulse represents binary 1.

10.7.5 Block Codes

We have so far discussed line codes that work on 1 bit at a time to select the code symbols. In block codes, on
the other hand, blocks of input bits are coded into code symbols according to rules defined in a coding table.
As shown in Figure 10.25, a ternary block code uses m ternary symbols to represent n bits, and is designated
 678 10 Digital Baseband Coding

Binary m binary
n bits
block code symbols

Ternary m ternary
n bits
block code symbols

Figure 10.25 Block coding.

nBmT. A binary block code uses m binary symbols to represent n bits, and is designated nBmB. Usually, not all
the possible symbol combinations or codewords are used in the block code output. This allows the flexibility of
choosing those codewords with the desired characteristics. For example, balanced codewords containing equal
amounts of positive (+) and negative (−) pulses, and hence no DC offset, are preferred to unbalanced codewords.
Furthermore, codewords with more frequent transitions, such as − + − + − +, are preferred to those with fewer
transitions, such as − − − + + +.
From the definition of code efficiency in Chapter 1, it follows that the efficiency of nBmT and nBmB codes are
given, respectively, by
n
𝜂nBmT = × 100%
mlog2 (3)
n
𝜂nBmB = × 100% (10.67)
m
Equation (10.67) is obtained by noting that, in the nBmT code, n bits of information are carried using m ternary
(i.e. three-level) symbols, each of which has a potential information content of log2 (3) bits. Similarly, in the nBmB
code, m binary symbols (which have a potential information content equal to m bits) are employed to carry only
n bits of information, yielding the efficiency stated above. For a given m, efficiency is maximised in a binary block
code by choosing n = m − 1. With this relation, the coding efficiency increases with m, but at the expense of
increased codec complexity, increased packetisation delay (to assemble up to m − 1 bits before coding) and an
increased likelihood of a long run of like pulses.
Table 10.8 shows the coding table of three simple block codes. The last column gives the disparity, which is
the digital sum of each codeword, a digital sum being a count of the imbalance between negative and positive
pulses in a sequence of symbols. Observe that CMI and Manchester codes can be treated as a 1B2B block code,
which takes one bit at a time and represents it using two binary symbols. For example, the coding table for CMI
shows that binary 0 is represented by a negative pulse followed by a positive pulse. This is denoted − + and was
earlier presented as two half-width pulses. Binary 1 is represented using either two negative or two positive pulses,
denoted − − and + +, respectively, and introduced earlier as a single full-width pulse.
Telegram: @Persian_IEEEComSoc

The waveform of a 3B4B code obtained using the coding table in Table 10.8 is shown at the bottom of Figure 10.23.
To see how the 3B4B code waveform was obtained, note that the input bit stream is taken in blocks of three bits.
See the demarcation of the bit stream listed below the 3B4B waveform. The encoder maintains a running dig-
ital sum (RDS), which is the cumulative sum of the disparity of each transmitted codeword. To code an input
block that has two codeword options, the negative-disparity option is selected if RDS is positive or zero, and the
positive-disparity option is chosen if RDS is negative. Let us assume that initially RDS = 0. Then, with the bit
stream as in Figure 10.23, the first input block is 101, which is represented with the codeword + − + −, according
to Table 10.8. The 3B4B waveform during this interval corresponds to this code. The RDS stays at zero. The next
input block is 000, which (since RDS = 0) is represented using its negative-disparity code option − − + −, accord-
ing to Table 10.8. RDS is updated to −2, by adding the disparity of the new codeword. Again, the portion of the
3B4B waveform in Figure 10.23 during this interval corresponds to the new codeword. The coding continues in
 10.7 Line Codes 679

Table 10.8 Coding table for 3B4B and 1B2B codes.

Output codeword

Type Input Negative 0 Positive Disparity

CMI 0 −+ 0
= 1B2B 1 −− ++ ±2
Manchester 0 −+ 0
= 1B2B 1 +− 0
001 −−++ 0
010 −+−+ 0
011 −++− 0
3B4B 100 +−−+ 0
101 +−+− 0
110 ++−− 0
000 −−+− ++−+ ±2
111 −+−− +−++ ±2

this manner with the RDS being updated after each codeword until the last input block 000, which is represented
with its positive-disparity code option + + − + to raise the RDS from −2 to 0.
Block codes have several advantageous features.

● Good timing content: codewords are selected that have sufficient transitions.
● No baseline wander: DC offset is eliminated by using balanced codewords whenever possible. When an input
block must be represented using unbalanced codewords then two options that balance out each other are pro-
vided for the block.
● Greater efficiency: this means that a lower signalling rate can be used to provide the desired bit rate, which allows
a greater spacing between repeaters without excessive degradation of transmitted symbols.

However, block codes have the demerit of increased codec complexity, which translates to higher costs. They
are typically used on long-distance transmission links where savings in number of repeaters justify the higher
complexity.
Telegram: @Persian_IEEEComSoc

Worked Example 10.6

Determine the code efficiency, signalling rate, and symbol period for the following baseband transmission sys-
tems
(a) 139 264 kb/s transmitted using CMI
(b) 139 264 kb/s transmitted using 6B4T.

(a) CMI is a 1B2B code. Its efficiency therefore follows from Eq. (10.67)

1
𝜂CMI = × 100% = 50%
2
 680 10 Digital Baseband Coding

Signalling rate Rs is the number of symbols transmitted each second. In this case (1B2B), two binary symbols
are used for each bit, and since 139 264 000 bits are transmitted each second, it follows that
Rs = 2 × 139264000
= 278528000 baud
= 278.53 MBd
We have used the word baud for symbols/second, as is common practice.
The symbol period T s is the reciprocal of Rs . Thus
1 1
Ts = = = 3.59 ns
Rs 278528000
(b) 6B4T is a ternary code with n = 6, m = 4. Eq. (10.67) gives
6
𝜂6B4T = × 100% = 94.64%
4log2 (3)
In this case, four ternary symbols are used to carry 6 bits. With 139 264 000 bits transmitted each second, the
number of ternary symbols transmitted per second (i.e. the signalling rate) is

139264000
Rs = ×4
6
= 92.84 MBd
The symbol period T s = 1/Rs = 10.77 ns.
Observe that the more efficient ternary system uses a lower signalling rate to accommodate the same informa-
tion transmission rate as the less efficient CMI. However, CMI has the advantage of simplicity. The decision
on which code is used will be dictated by the priorities of the application.

10.8 Summary
This now completes our study of digital baseband coding. We have acquired a thorough grounding in the prin-
ciples of uniform and nonuniform quantisation and PCM encoding of analogue signals for digital transmission.
The continuous-value sampled signal at the quantiser input may be converted to discrete values at the output
by rounding or by truncation. Both approaches inevitably introduce approximation errors which are perceived as
noise in the signal. The quantisation noise power when truncation is used is 6 dB higher than that of rounding. The
input range ±C of the quantiser may be partitioned into quantisation intervals using either a midrise or mid-step
Telegram: @Persian_IEEEComSoc

approach. Mid-step partitioning creates a dead zone that enables all small fluctuations about zero at the quantiser
input (e.g. due to noise in the absence of a signal) to be mapped to zero at the output. In midrise partitioning, zero
is an interval boundary so that there are N/2 quantisation intervals for the positive input range and N/2 intervals
for the negative, giving a total of N = 2k intervals (where k is the number of bits per sample). Mid-step, on the other
hand, has N − 1 quantisation intervals, comprising N/2 − 1 intervals for the positive input range, N/2 − 1 intervals
for the negative input range, and one interval (= 0 V output) for the dead zone.
Quantisation intervals may be equally spaced across the entire quantiser input range leading to uniform quanti-
sation. The approximation errors in each interval depends exclusively on the size (called step size) of the interval.
If the input signal is small then these errors may become comparable to input sample values leading to a low
ratio between signal power and quantisation noise power and hence an unacceptably poor signal quality at the
quantiser output. One solution to this problem is to use finer quantisation involving sufficiently small step sizes,
 Questions 681

but this requires large k, which leads to high bit rate and ultimately high storage and transmission bandwidth
requirements. A better solution is to use nonuniform quantisation in which small inputs are finely quantised but
large inputs are coarsely quantised. With the goal of achieving a constant SQNR across the entire input range, we
found that the optimum quantiser transfer characteristic is a logarithmic function which delivers step sizes that
increase exponentially away from the origin. Constraints of practical implementation necessitated a piecewise lin-
ear approximation of this logarithmic function, leading to two global standards, namely A-law and 𝜇-law PCM. We
discussed both standards in detail, introduced the concepts of companding gain and penalty, learnt how to carry
out PCM coding and decoding, and examined the extent of variation of SQNR across the quantiser input range in
each standard.
In seeking to achieve an acceptable SQNR when converting an analogue signal to digital, another approach is to
quantise and encode the error between the input sample and a local prediction of the sample rather than coding
the actual sample. Prediction errors can be made very small by reducing the time between adjacent samples (i.e.
increasing the sampling rate) or by using a more sophisticated and adaptive prediction filter. This means that the
quantiser range can be made very small, which reduces the number of intervals required for fine quantisation and
therefore reduces the bit rate of the resulting digital signal. This is a class of differential quantisation and includes
DPCM and DM as special cases which we briefly discussed. If the analogue signal is speech then further reduction
in bit rate (from 64 kb/s for a standard PCM codec through various rates down to less than 1 kb/s for a MELP
vocoder) can be achieved using a wide range of LBR speech coding techniques, which we briefly reviewed. In
general, LBR codecs achieve lower bit rates through lossy compression that sacrifices signal quality and increases
processing delay.
The final section of the chapter was devoted to line coding to explore how the above digitised signals, and indeed
all types of digital data, are electrically represented and transmitted as voltage waveforms in digital baseband
systems. We reviewed a range of basic line codes and then devoted some time to learning, using HDB3 and block
codes, how line codes are designed to satisfy the important requirements of a guaranteed clock content and zero
DC content. The importance of a high line code efficiency in reducing transmission symbol rate was also explored
through a worked example.
Baseband transmission is limited to fixed wire-line channels. To exploit the huge bandwidth of optical fibre in
addition to the broadcast and mobility capabilities of radio channels (albeit with smaller bandwidths), we must
modulate a suitable carrier with the digital data before transmission. The next chapter therefore deals with digital
modulation, which allows us to apply the principles learnt in Chapters 7 and 8 to digital signals, and to quantify
the impact of additive white Gaussian noise on such digital modulated systems.

Reference

1 Otung, I. (2014). Digital Communications: Principles & Systems. London: Institution of Engineering and Technol-
Telegram: @Persian_IEEEComSoc

ogy (IET). ISBN: 978-1849196116.

Questions

1 The output of a midrise rounding uniform quantiser of input range ±5 V is coded using 8 bits/sample.
Determine:
(a) The maximum quantisation error.
(b) The quantisation error associated with an input sample of value −2.3 V.
(c) The quantisation noise power.
 682 10 Digital Baseband Coding

(d) The dynamic range of the system.

(e) The SQNR when the input is a speech signal that fully loads the quantiser.
(f) The SQNR during a weak passage when the peak value of the input speech signal is only 0.5 V.

2 Determine the SQNR of a linear ADC as a function of number of bits per sample for an input signal that has
a Gaussian PDF with zero mean. Ignore sample values that occur on average less than 0.1% of the time. How
does your result compare with Eq. (10.17), which gives the SQNR for a speech input signal?

3 Determine the values of A and 𝜇 that yield a companding gain of 48 dB in A-law and 𝜇-law PCM, respectively.
Why aren’t such (or even larger) values of A and 𝜇 used in practical PCM systems to realise more companding
gain and hence greater savings in bandwidth?

4 Produce a diagram like Figure 10.11 for a digital translation view of 𝜇-law PCM.

5 An analogue input signal of values in the range ±2 V fully loads the quantiser of a 𝜇-law PCM system. Deter-
mine:
(a) The quantisation error incurred in transmitting the sample value −1.13 V.
(b) The PCM code for the sample value −1.9 V.
(c) The voltage value to which the code 10011110 is converted at the receiver.
(d) The maximum quantisation error in the recovered sample in (c).
(e) The minimum and maximum quantisation errors of the whole process.

6 Determine the maximum SQNR in the following nonlinear PCM systems, and comment on the trend of your
results.
(a) A-law with A = 1 and k = 8 bits/sample.
(b) A-law with A = 100 and k = 8 bits/sample.
(c) A-law with A = 1000 and k = 8 bits/sample.
(d) A-law with A = 100 and k = 6 bits/sample.
(e) A-law with A = 100 and k = 10 bits/sample.

7 The message signal vm (t) = 5cos(2000𝜋t) is coded using DM and a sampling frequency of 10 kHz. Determine:
(a) The minimum step size to avoid overload distortion.
(b) The quantisation noise power when the minimum step size is used.
(c) The SQNR. How does this compare with the maximum SQNR realisable using a linear ADC of the same
bit rate?
Telegram: @Persian_IEEEComSoc

8 Sketch the HDB3 and 3B4B waveforms for the following bit stream:
111011000100000110000111

9 Determine the code efficiency, signalling rate, and symbol period of a baseband transmission operating at
140 Mb/s and employing the 4B3T line code.
 683

11

Digital Modulated Transmission

Is your goal big enough to excite you, rewarding enough to motivate you, challenging enough to dare you, and
precise enough to direct you?

In this Chapter
✓ Why digital modulated transmission?
✓ Two views of digital modulated transmission: (i) frequency, phase, or amplitude modulation of a sinusoidal
carrier with a digital signal; (ii) coding of binary data using sinusoidal (i.e. bandpass) pulses.
✓ Signal space: a simple yet powerful tool for representing and analysing both baseband and modulated
digital systems.
✓ Noise effects and bit error ratio. You will be able to evaluate the impact of noise on all types of binary
modulated and baseband systems. You will also have the foundation needed later to extend the analysis
to multilevel digital systems.
✓ Binary modulation and coherent detection: how ASK, PSK, and FSK signals are generated at a transmitter
and detected at a receiver. You will also learn the effect of frequency spacing on the bandwidth and bit
error ratio of FSK.
✓ Noncoherent detection: this avoids the complexity of phase synchronisation in coherent detectors but
suffers from an inferior noise performance.
✓ M-ary transmission: a detailed presentation of the generation, detection, and bit error ratios of multilevel
ASK, PSK, FSK, and hybrid systems.
✓ Design considerations: a lucid discussion that gives you an excellent insight into the interplay of various
design parameters, namely bandwidth, signal power, bit rate, and bit error ratio.
Telegram: @Persian_IEEEComSoc

11.1 Introduction
Digital baseband coding, discussed at length in Chapter 10, conveys information using symbols (or pulses) that
contain significant frequency components down to, or near, DC. This technique is impracticable in several impor-
tant situations:
● When it is required to confine the transmitted frequencies in a digital system within a passband centred at a
frequency f c > > 0. This situation arises in the exploitation of radio and optical fibre transmission media whose
usable passband is naturally limited to frequencies well above DC. Radio is the only type of medium that allows

Communication Engineering Principles, Second Edition. Ifiok Otung.

© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 684 11 Digital Modulated Transmission

broadcast and mobility capabilities, whereas optical fibre is preferred to coaxial cable for high-capacity fixed
telecommunication links. These two media, radio and optical fibre, are involved at some stage in most modern
transmissions.
● When the available bandwidth of the transmission medium is insufficient to convey the baseband pulses at
the desired symbol rate without significant distortion. An important example here is the global wire-line tele-
phone network, which was optimised in the early days for the transmission of analogue voice signals, containing
frequencies between 300 and 3400 Hz. However, with digital transmission becoming the preferred method of
communication, a means had to be devised to transmit digital signals over these voice-optimised lines in order
to exploit the huge financial investment that they represent. This limited-bandwidth situation also arises in
radio where separate frequency bands must be allocated to different users to allow simultaneous transmission
by many users on the same link.
The means of utilising the above media for digital communication is by digital modulated transmission in which
one or more parameters of a sinusoidal carrier are varied by the information-bearing digital signal. The basic
techniques of digital modulated transmission, namely amplitude shift keying (ASK), frequency shift keying (FSK),
and phase shift keying (PSK), are presented in Chapter 1, which is worthwhile reviewing at this point.
An important clarification of terminology is in order here. Throughout this chapter, we will use ASK, FSK,
and PSK to refer to their binary implementation. Multilevel transmission will be explicitly identified with terms
such as quadriphase shift keying (QPSK), 4-ASK, 4-FSK, M-ary, etc. Furthermore, ASK is treated as on–off keying
(OOK) in which one of the two amplitude levels of the transmitted sinusoid is zero. There is nothing to be gained
from making both levels nonzero, except a poorer noise performance (i.e. higher bit error ratio [BER]) for a given
transmitted power.
We may approach the subject of digital modulated transmission in two ways.
● The more obvious approach is to treat the technique as that of sinusoidal carrier modulation involving digital
signals. The theories of amplitude and angle modulations presented in Chapters 7 and 8 can then be applied,
but with the modulating signal vm (t) being digital. Thus, ASK, for example, is obtained using an AM modulator
with modulation factor m = 1; and the (digital) message signal is recovered at the receiver by AM demodulation
using an envelope detector. The circuit components in this case are very similar to those of analogue modulation,
except that the recovered message signal will be applied to a regenerator to obtain a pure digital signal free from
accumulated noise effects.
● A less obvious but greatly simplifying approach is to treat the technique as an adaptation of baseband trans-
mission to bandpass channels. The basis of this approach is that the modulated sinusoidal carrier can only
take on a discrete number of states in line with the discrete values of the digital modulating signal. Thus, we
simply treat the modulated carrier transmitted during each signalling interval as a ‘bandpass’ pulse or symbol.
Under this approach, the process of modulation simplifies to symbol generation. More importantly, the receiver
does not need to reproduce the transmitted waveform but merely determine which symbol was transmitted
Telegram: @Persian_IEEEComSoc

during each interval. Demodulation therefore simplifies to symbol detection, and subsequent digital baseband
signal regeneration. Binary ASK, FSK, and PSK each involves the transmission of two distinct symbols – shown
in Figure 11.1, whereas, in general, M-ary modulated transmission involves M distinct bandpass symbols, as
shown in Figure 11.2 for M = 4. To understand why these sinusoidal symbols are bandpass, we may consider the
case of binary ASK shown in Figure 11.1. Here, a symbol g(t) is transmitted for the bit duration T s if the bit is a
1, and no symbol is transmitted if the bit is a 0. We see from Figure 11.3 that the symbol g(t) can be written as
the product of a rectangular pulse of duration T s and a sinusoidal function (of limitless duration). That is
g(t) = V rect(t∕Ts ) cos(2𝜋fc t) (11.1)
Recall (e.g. Figure 4.29) that a rectangular pulse rect(t/T s ) has a sinc spectrum of null bandwidth 1/T s , and that
the effect of multiplication by cos(2𝜋f c t) is to shift this spectrum from its baseband (centred at f = 0) to a passband
 11.1 Introduction 685

Bit stream → 1 0 1 1 1 0 0 1
V

ASK 0

–V
V

FSK 0

–V
V

PSK 0

–V
Bit duration

Figure 11.1 Waveforms of ASK, FSK, and PSK.

Bit stream → 1 0 0 1 1 1 0 0
V
2V/3
V/3
4-ASK 0
–V/3
–2V/3
–V
V

4-FSK 0

–V
Telegram: @Persian_IEEEComSoc

4-PSK 0

–V
Phase = –90˚ Phase = 90˚ Phase = 0˚ Phase = 180˚
Symbol duration

Figure 11.2 Waveforms of 4-ASK, 4-FSK, and 4-PSK.

 686 11 Digital Modulated Transmission

g(t)
V

rect(t/Ts)

Ts

|G(f)|
VTs/2

f
–fc 0 fc fc + 1/Ts

Figure 11.3 Bandpass symbol g(t) and its amplitude spectrum |G(f )|.

centred at ±f c . Therefore, the spectrum G(f ) of g(t) is as shown in Figure 11.3, which is clearly a bandpass spectrum,
of bandwidth 2/T s , having significant (positive) frequencies centred around f c
VT s
G(f ) = {sinc[(f − fc )Ts ] + sinc[(f + fc )Ts ]} (11.2)
2
Our presentation in this chapter follows the second approach. This will allow us to apply the important con-
cepts related to matched filtering (Chapter 12). The matched filter can be realised using a correlation receiver or
a coherent detector. The main task at the receiver is to determine (in the presence of noise) which symbol was
sent during each interval of duration T s . If a matched filter is employed then what matters is not the symbol shape
but the symbol energy E compared to noise power. In the case of sinusoidal symbols used in digital modulated
transmission, the symbol energy is given by

Es = Symbol Power × Symbol Duration

= A2c Ts ∕2 (11.3)
Telegram: @Persian_IEEEComSoc

where Ac is the sinusoidal symbol amplitude, and we have assumed (as is usual) a unit load resistance in the
computation of power.
Each modulated carrier state or symbol can therefore be fully represented by a point on a signal-space or con-
stellation diagram, the distance of the point from the origin being the square root of the energy of the symbol.
In what follows, we first discuss important general concepts that are applicable to both baseband and modu-
lated digital systems. These include signal orthogonality, signal space, digital transmission model, noise effects, and
bit error ratio. Working carefully through these sections will give you a sound understanding of vital principles
and equip you to easily deal with the special cases of digital modulated transmission discussed in the remaining
sections of the chapter. These special cases are briefly discussed under the headings of binary modulation (Section
11.7), coherent binary detection (Section 11.8), noncoherent binary detection (Section 11.9), and M-ary transmission
 11.2 Orthogonality of Energy Signals 687

(Section 11.10). The chapter ends with a brief comparison of various digital modulation techniques in terms of the
important system design parameters, namely bit rate, transmission bandwidth, signal power, and BER.

11.2 Orthogonality of Energy Signals

The concepts of signal correlation and orthogonality are discussed at length for various types of random and deter-
ministic signals. Since in digital transmission information is conveyed using a finite set of M distinct symbols g0 (t),
g1 (t), …, gM − 1 (t), where M = 2 for binary, M > 2 for multilevel, or M-ary transmissions, and each symbol gk (t) is
an energy signal of duration T s , we wish in this section to apply these concepts to energy signals.
Given two signals g1 (t) and g2 (t), if their product g1 (t)g2 (t) has zero area over an interval of duration, say, T s ,
whereas the areas of g12 (t) and g22 (t) are both nonzero over the same interval then the two signals are said to be
orthogonal over the interval T s . The principle of orthogonality finds extensive applications in communication
systems, as we will learn in this chapter. For example, if a binary transmission system sends bit 1 using signal g1 (t)
and bit 0 using g2 (t) then we can determine at the receiver which bit was sent during each interval as follows.
Multiply the incoming signal by g1 (t) and compute the area of the resulting product. Note that this computation of
area is the mathematical operation of integration which is very easily carried out in real systems using a suitably
designed lowpass filter (LPF). If the area is zero – in practice smaller than a set threshold – we conclude that the
incoming signal is g2 (t) and hence that bit 0 was sent. But if this area is significant (i.e. larger than the set threshold)
then it is concluded that the incoming signal is g1 (t) and hence that bit 1 was sent.
Orthogonality can also exist right across a set of signals, with every pair in the set being orthogonal. Energy
signals g1 (t), g2 (t), g3 (t), …, gN (t), each of duration T s , are said to be orthogonal with respect to each other if
{
Ts
0, k≠m
gk (t)gm (t)dt = (11.4)
∫0 E , k=m k

where Ek , the energy of gk (t), is nonzero and positive. When two energy signals g1 (t) and g2 (t) are orthogonal then
their energies add independently. That is, the energy E of the sum signal g(t) = g1 (t) + g2 (t) is given by the sum of
the energies E1 and E2 of g1 (t) and g2 (t), respectively. You can see that this is the case by observing that
Ts
E= g2 (t)dt
∫0
Ts
= [g1 (t) + g2 (t)]2 dt
∫0
Ts Ts Ts
= g12 (t)dt + g22 (t)dt + 2 g1 (t)g2 (t)dt
∫0 ∫0 ∫0
= E1 + E2 + 0
Telegram: @Persian_IEEEComSoc

If in Eq. (11.4) Ek = 1 for k = 1, 2, 3, …, N, then the waveforms g1 (t), g2 (t), g3 (t), …, gN (t) are said to be orthonor-
mal. Thus, orthonormal signals are unit-energy orthogonal signals.
Orthogonality can also be defined for periodic power signals. Power signals g1 (t), g2 (t), g3 (t), …, of period T are
said to be orthogonal with respect to each other if
{
1
T∕2
0, k≠m
gk (t)gm (t)dt = (11.5)
T ∫−T∕2 Pk , k = m
If the power signal is nonperiodic then Eq. (11.5) is applied in the limit T → ∞.
The following examples of orthogonal signal sets are of special interest in digital communications. It is easy to
show the orthogonality of each of the sets given below simply by evaluating Eq. (11.4) for any two functions gk (t)
 688 11 Digital Modulated Transmission

g–2(t)

A–2
t
τ
g–1(t)

A–1
t
τ
g0(t)

A0

τ t
g1(t)

A1

τ t
g2(t)

A2

τ t

Figure 11.4 Orthogonal set of rectangular pulses gk (t) = Ak rect([t−k𝜏]/𝜏), k = … −2, −1, 0, 1, 2, . . . .

and gm (t) in the set to show that the integration yields zero in all cases except when k = m. A demonstration of
this verification is given for the final set in Eq. (11.10) below:

● Harmonically related sinusoidal signals

( )
k•
gk (t) = Ak cos 2𝜋 • t + 𝜙k
Ts
k = 1, 2, 3, · · · (11.6)

form an orthogonal set over the interval 0 ≤ t ≤ T s .

● Nonoverlapping pulses
Telegram: @Persian_IEEEComSoc

( )
t − (k + 𝛾)𝜏
gk (t) = Ak rect
𝜏
k = · · · , −2, −1, 0, 1, 2, · · · ;
0≤𝛾<1 (11.7)

form an orthogonal set. Figure 11.4 is an example of this set with 𝛾 = 0 and each pulse of duration 𝜏. To convert
gk (t) into an orthonormal set, we assign each pulse an amplitude Ak that gives it unit energy

Ek = A2k 𝜏 = 1
√
⇒ Ak = 1∕𝜏 (11.8)
 11.3 Signal Space 689

√
● The cosine and sine pulses 𝛼 0 (t) and 𝛼 1 (t) given below, of amplitude 2∕Ts , having an integer number n of
cycles within their duration T s , are orthonormal
( )
√ n
𝛼0 (t) = 2∕Ts cos 2𝜋 • •t rect(t∕Ts )
T
( s )
√ n
𝛼1 (t) = − 2∕Ts sin 2𝜋 • •t rect(t∕Ts ) (11.9)
Ts
● The sinusoidal pulses given below, which complete an integer number (k + n) of half-cycles within their duration
T s , form an orthonormal set
( ) ( )
√ k+n • t − Ts ∕2
𝛼k (t) = 2∕Ts cos 2𝜋
′ • Rs t rect
2 Ts
k = 0, 1, 2, 3, · · · ; Rs = 1∕Ts (11.10)

Note that the sinusoidal signals that are curtailed in duration to T s to produce the above pulses 𝛼0′ (t), 𝛼1′ (t),
· · · have respective frequencies f 0 , f 0 + Rs /2, f 0 + Rs , …, where f 0 = nRs /2, n ≥ 1, and Rs = 1/T s . To verify
𝛼2′ (t),
that the above set is orthonormal, consider any two members of the set
√
𝛼k′ (t) = 2Rs cos[𝜋(k + n)Rs t]
√
𝛼m ′
(t) = 2Rs cos[𝜋(m + n)Rs t] 0 ≤ t ≤ 1∕Rs

Evaluating the integral of their product over one symbol interval yields
1∕Rs 1∕Rs
𝛼k′ (t)𝛼m
′
(t)dt = 2Rs cos(𝜋(k + n)Rs t) cos(𝜋(m + n)Rs t)dt
∫0 ∫0
[ 1∕Rs 1∕Rs ]
= Rs cos 𝜋Rs (k + m + 2n)tdt + cos 𝜋Rs (k − m)tdt
∫ ∫0
[ 0 ]
sin 𝜋(k + m + 2n) sin 𝜋(k − m)
= Rs +
𝜋Rs (k + m + 2n) 𝜋Rs (k − m)
= sinc(k + m + 2n) + sinc(k − m)
{
1, k = m
= (11.11)
0, Otherwise
where we made use of the trigonometric identity for the product of two cosines in the second line. Since n, k, and
m are integers, so are (k + m + 2n) and (k−m). But when N is an integer, sinc(N) = 1 for N = 0 and sinc(N) = 0 for
N = ±1, ±2, ±3, …, hence the final line in the above equation. This verifies that the entire set is orthonormal.
Telegram: @Persian_IEEEComSoc

11.3 Signal Space

The M transmitted symbols g0 (t), g1 (t), …, gM-1 (t) are rectangular pulses in the baseband systems discussed in
Chapter 10 but are required to be sinusoidal symbols in bandpass transmission media. This leads to digital mod-
ulated transmission – the subject of this chapter. Each symbol gk (t) is an energy signal, being of finite duration T s
and of finite amplitude Ac , and hence of finite energy Ek , given by Eq. (11.3) for a sinusoidal symbol.
We wish to adopt a geometric representation of the set of symbols {gk (t)}, which is an excellent tool for visualising
the corresponding transmitted states {Sk } of a system and their energies and closeness to each other, and for eval-
uating the impact of additive white Gaussian noise (AWGN) on the system. We may express each of the M energy
 690 11 Digital Modulated Transmission

signals as a linear combination of N orthonormal basis functions 𝛼 0 (t), 𝛼 1 (t), …, 𝛼 N-1 (t), where N ≤ M and 0 ≤ t ≤ T s
gk (t) = sk0 𝛼0 (t) + sk1 𝛼1 (t) + · · · + sk,N−1 𝛼N−1 (t)
∑
N−1
= skj 𝛼j (t),
j=0

k = 0, 1, · · · , M − 1 (11.12)
The set of functions {𝛼 k (t)} satisfy the condition
{
Ts
0, k ≠ j
𝛼k (t)𝛼j (t)dt = (11.13)
∫0 1, k = j
It follows that the energy of gk (t) is given by
Ts
Ek = gk2 (t)dt
∫0
Ts
= [sk0 𝛼0 (t) + sk1 𝛼1 (t) + · · · + sk,N−1 𝛼N−1 (t)]2 dt
∫0
Ts Ts Ts
= s2k0 𝛼02 (t)dt + s2k1 𝛼12 (t)dt + · · · + s2k,N−1 𝛼N−1
2
(t)dt
∫0 ∫0 ∫0
∑
N−1
∑
m−1 Ts
+2 skm skj 𝛼j (t)𝛼m (t)dt
m=1 j=0
∫0

∑
N−1
= s2k0 + s2k1 + · · · + s2k,N−1 = s2kj (11.14)
j=0

Therefore, by analogy with Pythagoras’s theorem, we represent the kth transmitted state Sk as a point in
N-dimensional Euclidean space, which consists of N mutually perpendicular axes 𝛼 0 , 𝛼 1 , …, 𝛼 N-1 , and is called
the signal space. Signal spaces with N > 3 cannot be visualised or sketched in real-life space, which is limited to
three dimensions, but they remain an important mathematical concept. Distances in this space represent the
square root of energy. In particular, the distance of a point from the origin gives the square root of the energy of
the transmitted state that the point represents. Figure 11.5 shows signal-space examples with N = 1, M = 2 in (a);
N = 2, M = 8 in (b); and N = 3, M = 4 in (c).

11.3.1 Interpretation of Signal-space Diagrams

Let us take a moment to understand what signal-space or constellation diagrams tell us about the transmission
Telegram: @Persian_IEEEComSoc

systems that they represent.

● Number of transmitted states: information (in the form of a string of 0s and 1s, or bit stream) is conveyed using
2, 8, and 4 distinct symbols or transmitted signal states in (a), (b), and (c), respectively. The assignment of states
may be as follows. In Figure 11.5a, state S0 represents bit 0, and S1 represents bit 1. In Figure 11.5b, each state
represents a block of three bits; for example, S0 represents 000, S1 represents 001, …, and S7 represents 111.
Similarly, each state in Figure 11.5c represents a block of two bits, with S0 representing 00, S1 representing 01,
and so on. Generally, the number of states M in a signal-space diagram gives the order of the modulation and
each state in the diagram represents log2 M bits.
● Type of modulation scheme: the distribution and location of states in the signal-space diagram reveal the type of
modulation scheme employed in the transmission system. If all M states lie on a circle, the modulation scheme
is PSK. If all M states lie along a unipolar axis, the modulation scheme is ASK. If the M states are distributed
 11.3 Signal Space 691

S0 S1
(a) α0
–3 –2 –1 0 1 2 3

α1
α′2
S0 3 S2
+1 S1 S3
2
E0 1
S1
S2 S7 α′0
(b) α0 (c) 1 2 3
–1 +1 S3
1
S0 2

S6 –1 S4 S5 α′1
3

Figure 11.5 Signalspace diagrams: (a) N = 1, M = 2; (b) N = 2, M = 8; (c) N = 3, M = 4.

in one plane but not all on one circle, the scheme is amplitude and phase shift keying (APSK), sometimes called
quadrature amplitude modulation (QAM). If the M states are located at the same distance from the origin, with
one state on each mutually orthogonal axis, and each axis representing a sinusoidal basis function having a
unique frequency then the scheme is FSK.
● Energy of each transmitted state: the energy of each state equals the square of the distance of that state from the
origin. Using arbitrary energy units for the moment, we have in Figure 11.5a
E0 = 0
E1 = 3 2 = 9
In Figure 11.5b
E0 = 1 2 + 1 2 = 2
E1 = 12 = 1, and so on.
And in Figure 11.5c
E1 = 2 2 = 4
E3 = 32 + 32 + 22 = 22, and so on.
Telegram: @Persian_IEEEComSoc

In general, the energy Ek of state k (representing symbol Sk ) is given by

Ek = D2k (11.15)
where Dk is the distance of state k from the origin.
● Average energy per symbol Es : the average energy per symbol is an important parameter which is obtained by
dividing the total energy in the constellation by the number of symbols

1 ∑ 1 ∑ 2
M−1 M−1
Es = Ek = D (11.16)
M k=0 M k=0 k

You may wish to verify that the constellation of Figure 11.5b has Es = 1.5.
 692 11 Digital Modulated Transmission

● Average energy per bit Eb : another important parameter which we obtain directly from a signal-space diagram
is the average energy per bit. This is the average energy per symbol divided by the number of bits per symbol,
which is log2 M. Thus
Es 1 ∑
M−1
Eb = = D2 (11.17)
log2 M M log2 M k=0 k

It follows that
⎧ (M − 1)(2M − 1)d2
⎪ , M-ary ASK
Eb = ⎨ 6 log2 M
⎪d2 ∕log2 M, M-ary PSK, M-ary FSK
⎩
M = 2m ; m = 1, 2, 3, · · · (11.18)
where, in M-ary ASK, d is the distance between adjacent states, whereas, in M-ary PSK and M-ary FSK, d is the
distance of each state from the origin.
● Signal power P: noting that power
Energy
P=
Second
Energy Symbol Energy Bit
≡ × ≡ ×
Symbol Second Bit Second

it follows that
P = Es Rs = Es ∕Ts
= Eb Rb = Eb ∕Tb (11.19)
where Rs = symbol rate, T s = symbol duration, Rb = bit rate, and T b = bit duration. Thus, if we know the symbol
rate of the transmission system then we may obtain the transmitted signal power using Eq. (11.19) along with the
average energy per symbol determined from the signal-space diagram.
● Amplitude of each symbol: since, in a digital modulated system, the symbols are sinusoidal, we may determine
the amplitude Ak of the kth symbol from its energy Ek as follows
Ek = D2k = Pk Ts (i.e. Power × Duration)
A2k
= Ts
2
Telegram: @Persian_IEEEComSoc

which yields
√ √
2 √ 2Rb
Ak = Dk = Dk 2Rs = Dk (11.20)
Ts log2 M
● Transmitted symbols: the symbol gk (t) transmitted for each state Sk follows from Eq. (11.12), with the coefficients
equal to the distance moved along (or parallel to) each axis in order to get to point Sk starting from the origin.
In general
gk (t) = sk0 𝛼0 (t) + sk1 𝛼1 (t) + · · · + sk,N−1 𝛼N−1 (t)
 11.3 Signal Space 693

where skj is the component of state Sk along the kth axis in signal space, with the kth axis representing the basis
function 𝛼 k (t). Thus, if the signal-space diagrams in Figure 11.5 are based on the sinusoidal basis functions in
Eq. (11.9) and (11.10), it follows that in Figure 11.5a
g0 (t) = s00 𝛼0 (t) = 0
g1 (t) = s10 𝛼0 (t) = 3𝛼0 (t)
( )
√ n•
= 3 2∕Ts cos 2𝜋 • t rect(t∕Ts )
Ts
In Figure 11.5b
g0 (t) = s00 𝛼0 (t) + s01 𝛼1 (t) = −𝛼0 (t) + 𝛼1 (t)
[ ( ) ( )]
√ n √ n
=− 2∕Ts cos 2𝜋 • •t + 2∕Ts sin 2𝜋 • •t rect(t∕Ts )
Ts Ts
g1 (t) = s10 𝛼0 (t) + s11 𝛼1 (t)
( )
√ n•
= 𝜙1 (t) = 2∕Ts cos 2𝜋 • t rect(t∕Ts )
Ts
And in Figure 11.5c
( ) ( )
√ n• t − Ts ∕2
g1 (t) = 2𝛼0′ (t) = 2 2∕Ts cos 2𝜋 • t rect
Ts Ts
g3 (t) = 3𝛼0′ (t) + 3𝛼1′ (t) + 2𝛼2′ (t)
[ ( ) ( ) [ ]] ( )
√ n • n 2 3n • t − Ts ∕2
= 3 2∕Ts cos 2𝜋 • t + cos 2𝜋 • •t + cos 2𝜋 • t rect
2Ts Ts 3 2Ts Ts
Signal-space diagrams are hugely important and very widely used in digital transmission system modelling and
analysis. However, a signal-space diagram does not provide any information about symbol duration T s or the form
of the basis functions (whether baseband or bandpass) or the carrier frequency f c (if bandpass). This information
must be obtained from some other specification of the system. Baseband systems use rectangular (or shaped) basis
functions, whereas modulated systems use sinusoidal basis functions.

11.3.2 Complex Notation for 2D Signal Space

The signal space of M-ary PSK and M-ary APSK (e.g. Figure 11.5b) is two-dimensional (2D), consisting of two basis
functions, 𝛼 0 (t) and 𝛼 1 (t), which are, respectively, a cosine pulse and negative sine pulse, as defined in Eq. (11.9).
M-ary ASK is a special case having only one dimension√ with just 𝛼 0 (t) as its basis function. Since −sin𝜃 leads
cos𝜃 by 90∘ just as the imaginary number j = −1 (introduced in Eq. (2.42)) leads the real number +1 by 90∘ ,
Telegram: @Persian_IEEEComSoc

it is both mathematically consistent and convenient to adopt a complex notation for representing states in such
signal spaces, namely ASK, PSK, and APSK. With this notation, 𝛼 0 (t) is treated as the reference axis oriented in
the positive real or 0∘ direction and is referred to as the in-phase axis, whereas 𝛼 1 (t) is treated as the imaginary
axis oriented in the 90∘ direction, referred to as the quadrature axis. Thus, the location of any state Sk in signal
space, which is at a distance Dk from the origin and at a phase 𝜃 k (≡ angle measured counterclockwise from the
0∘ direction), is given by the complex number
Sk = xk + jyk
where
xk = Dk cos 𝜃k ; yk = Dk sin 𝜃k
k = 0, 1, 2, · · · , M − 1 (11.21)
 694 11 Digital Modulated Transmission

The components xk and yk are distances in signal space and therefore are in unit of square root of energy or
1
(J) /2 . Following the discussion leading
√ to Eq. (11.20), these components may be scaled to amplitudes having a unit
of volt by multiplying by the factor 2∕Ts , where T s is the duration of each transmitted bandpass symbol. This
leads to
√
AkI = xk 2∕Ts ;
√
AkQ = yk 2∕Ts (11.22)
AkI and AkQ are, respectively, the in-phase and quadrature components of the transmitted bandpass pulse gk (t)
corresponding to state Sk . This pulse is sinusoidal and has amplitude Ak given by
√
√ 2(xk2 + y2k )
Ak = A2kI + A2kQ = (11.23)
Ts
Since all states in M-ary ASK lie along the 𝛼 0 (i.e. real) axis, there is no quadrature component in ASK and 𝜃 k = 0
for all k.
We recall the definition in Section 3.5.6 of the correlation coefficient 𝜌 of two energy signals g0 (t) and g1 (t) of
duration T s and respective energies E0 and E1
T T
∫0 s g0 (t)g1 (t)dt 2 ∫0 s g0 (t)g1 (t)dt
𝜌= = (11.24)
Average Energy E0 + E1
and note from this definition and the definition of orthogonality in Eq. (11.4) that the correlation coefficient of two
orthogonal energy signals is zero. Furthermore, in this 2D signal space, the correlation coefficient 𝜌km between
two states Sk and Sm having respective components (xk , yk ) and (xm , ym ) simplifies to
2(xk xm + yk ym )
𝜌km = (11.25)
(xk2 + y2k ) + (xm
2
+ y2m )

11.3.3 Signal-space Worked Examples

The following worked examples provide a more detailed discussion and further insight into the signal-space con-
cept.

Worked Example 11.1 Baseband System

A baseband transmission system conveys information using the symbols g0 (t), g1 (t), and g2 (t) shown in
Figure 11.6a.
(a) Determine a suitable set of two basis functions 𝛽 0 (t) and 𝛽 1 (t).
Telegram: @Persian_IEEEComSoc

(b) Sketch the signal-space or constellation diagram for this system.

(a) The form of the transmitted symbols suggests that a suitable set of basis functions would be two half-width
rectangular pulses 𝛽 0 (t) and 𝛽 1 (t), as shown in Figure 11.6b, one occupying the interval 0 ≤ t ≤ T s /2, the other
the interval T s /2 ≤ t ≤ T s , and both having amplitude V 𝛽 that gives unit energy. Their formal mathematical
expressions are
( )
t − Ts ∕4
𝛽0 (t) = V𝛽 rect
T ∕2
( s )
t − 3Ts ∕4
𝛽1 (t) = V𝛽 rect
Ts ∕2
 11.3 Signal Space 695

g0(t) g1(t) g2(t)

V V V

Ts
(a) t t t
Ts Ts

–V –V –V

β0(t) β1(t)
Vβ Vβ

(b) t t
Ts /2 Ts /2 Ts

β(t) = β0(t) + β1(t)

Vβ

(c)
t
Ts
β1

S0 V T /2 S1
s

(d) β0
–V T /2 V T /2
s s

S2 –V T /2
s

Figure 11.6 Worked Example 11.1: (a) transmitted symbols; (b) orthonormal basis functions; (c) sum pulse; (d) signal space
diagram.

A general method for selecting basis functions will be given shortly. Since 𝛽 0 (t) and 𝛽 1 (t) are nonoverlapping,
their product (and hence its integral) is zero, whereas the area under the square of each pulse is nonzero.
Therefore, by the definition of orthogonality given in Eq. (11.4), the two pulses are orthogonal. Note that, as
expected of orthogonal signals, the energy of the sum of the two pulses equals the sum of the energy of each
pulse. The sum pulse 𝛽(t) is shown in Figure 11.6c and has energy
E𝛽 = V𝛽2 Ts
Telegram: @Persian_IEEEComSoc

whereas the individual pulses 𝛽 0 (t) and 𝛽 1 (t) have energies

E𝛽0 = E𝛽1 = V𝛽2 Ts ∕2 (11.26)
which shows that
E𝛽 = E𝛽0 + E𝛽1
and confirms that 𝛽 0 (t) and 𝛽 1 (t) are orthogonal. In general, pulses that occupy nonoverlapping time intervals
are orthogonal as stated in Eq. (11.7) for rectangular pulses. However, this rule holds regardless of pulse shape
(provided of course the pulses are not identically zero). Basis functions in baseband systems therefore consist of
fractional-width rectangular pulses, one for each interval between transitions in the voltage level of transmitted
symbols. In this example, we have two separate time instants at which voltage level transitions occur, one at
 696 11 Digital Modulated Transmission

t = T s /2 in g0 (t) only, and the other at t = T s in all three transmitted symbols. Thus, we require two basis
functions, as shown in Figure 11.6b. One important exception to this rule is when there are only two opposite
polarity (or antipodal) symbols g0 (t) and g1 (t), as in Manchester line code. In this case only one basis function
𝛽 0 (t) is required, which is the symbol g0 (t) appropriately scaled to unit energy. The two symbols are then given
by g0 (t) = K𝛽 0 (t) and g1 (t) = −K𝛽 0 (t), where K is a constant.Finally, we complete the specification of the basis
functions by determining the height V 𝛽 that gives them unit energy. Thus, from Eq. (11.26)
√
V𝛽 = 2∕Ts (11.27)
(b) The transmitted symbols have amplitude V, arrived at by dividing the basis functions by V 𝛽 and then multi-
plying by V. So we may express the transmitted symbols in terms of 𝛽 0 (t) and 𝛽 1 (t) as
V V
g0 (t) = − 𝛽 (t) + 𝛽 (t)
V𝛼 0 V𝜙 1
√
= V Ts ∕2[−𝛽0 (t) + 𝛽1 (t)]
√
g1 (t) = V Ts ∕2[𝛽0 (t) + 𝛽1 (t)]
√
g2 (t) = −V Ts ∕2[𝛽0 (t) + 𝛽1 (t)] (11.28)
The signal-space diagram is therefore as shown in Figure 11.6d. Note that the transmitted symbols have equal
energy
E0 = E1 = E2 = V 2 Ts (11.29)
If you have studied Chapter 10, you might have observed by now that this is a baseband system that employs
coded mark inversion (CMI) line code in which bit 0 is conveyed by g0 (t) and bit 1 is conveyed alternately by
g1 (t) and g2 (t).

Worked Example 11.2 Modulated System

In the following problems, a sinusoidal pulse is defined as a signal of duration T s that has a sinusoidal variation
with an integer number of cycles within the interval 0 ≤ t ≤ T s , and is zero elsewhere.
(a) Show that the sinusoidal pulses g0 (t) = V 0 cos(2𝜋f c t) and g1 (t) = −V 1 sin(2𝜋f c t) are orthogonal, where
f c = n/T s and n = 1, 2, 3, …
(b) Show that the set of sinusoidal pulses cos(2𝜋f s t), cos(2𝜋2f s t), …, cos(2𝜋nf s t), …, where f s = 1/T s , are mutu-
ally orthogonal.
(c) Sketch the signal-space diagrams of 4-ASK, 4-PSK, and 4-FSK.
Telegram: @Persian_IEEEComSoc

(a) From Eq. (11.3), the energy of each pulse is given by

E0 = V02 Ts ∕2
E1 = V12 Ts ∕2 (11.30)
The sum pulse g(t) = g0 (t) + g1 (t) is also of duration T s and is given in the interval 0 ≤ t ≤ T s by
g(t) = V0 cos(2𝜋fc t) − V1 sin(2𝜋fc t)
= V0 cos(2𝜋fc t) + V1 cos(2𝜋fc t + 𝜋∕2)
√
= V02 + V12 cos(2𝜋fc t + 𝜑)
≡ V cos(2𝜋fc t + 𝜑)
 11.3 Signal Space 697

where 𝜑 = arctan(V1 ∕V0 ). Note that we applied the technique of sinusoidal signal addition learnt in Chapter 2.
The energy of g(t) is
Ts T
E = V2 = (V02 + V12 ) s
2 2
= V02 Ts ∕2 + V12 Ts ∕2
= E0 + E1
We therefore conclude that g0 (t) and g1 (t) are orthogonal since their energies add independently. An alternative
and straightforward way of solving this problem would be to show that the integral of the product signal
g0 (t)g1 (t) is zero, whereas the integrals of g02 (t) and g12 (t) are nonzero. Also, note from Eq. (11.30) that g0 (t) and
g1 (t) will have unit energy, and hence become a set of two orthonormal basis functions 𝜙0 (t) and 𝜙1 (t) if we
set their amplitudes to
√
V0 = V1 = 2∕Ts (11.31)
The orthonormal basis functions of ASK, PSK, and APSK modulated systems are always sinusoidal signals of
duration T s , with frequency f c equal to an integer multiple of 1/T s and amplitude given by Eq. (11.31).
(b) There is no general rule for adding sinusoids of different frequencies, so in this case we apply Eq. (11.4) to prove
orthogonality. Consider two different functions gm (t) = cos(2𝜋mf s t) and gn (t) = cos(2𝜋nf s t), in the given set.
We have
Ts Ts
gm (t)gn (t)dt = cos(2𝜋mf s t) cos(2𝜋nf s t)dt
∫0 ∫0
Ts
1
cos[2𝜋(m + n)fs t]dt
2 ∫0
= (fs = 1∕Ts )
Ts
1
+ cos[2𝜋(m − n)fs t]dt
2 ∫0
=0
where we expanded the first integrand on the right-hand side using the trigonometric identity for the product
of cosines. The resulting integrals evaluate to zero since each is the area under a sinusoidal function over a
time interval T s in which the sinusoid completes an integer number of cycles. Over this interval, there is the
same amount of positive and negative area, giving a total area (i.e. integral) equal to zero.Now for m = n, we
have
Ts Ts
gm (t)gm (t)dt =cos2 (2𝜋mf s t)dt
∫0 ∫0
T
= s
Telegram: @Persian_IEEEComSoc

2
where we evaluated the integral by noting that it is the energy of a unit-amplitude sinusoidal pulse, which
allows us to apply Eq. (11.3) with V = 1. From the foregoing, it follows that
{
Ts
0, m≠n
cos(2𝜋mf s t) cos(2𝜋nf s t)dt = (11.32)
∫0 T ∕2, m = n s

where f s = 1/T s . Therefore, the set of sinusoidal pulses with frequencies at integer multiples of 1/T s are mutu-
ally orthogonal over the interval 0 ≤ t ≤ T s .
The set of orthonormal basis functions in FSK modulation consists of these sinusoidal pulses with amplitude
given by (11.31). M-ary FSK requires M basis functions giving rise to an M-dimensional signal space. On the
contrary, M-ary ASK is one-dimensional (1D), requiring only one basis function, whereas M-ary PSK is 2D,
 698 11 Digital Modulated Transmission

requiring two basis functions, albeit of the same frequency. Hybrid modulation techniques, which combine
ASK and PSK, are also 2D and are realised using a linear combination of the same two basis functions, as
in PSK.

4-ASK: this has a 1D signal space with four states: S0 , S1 , S2 , and S3 . Only one basis function is involved
{√
2∕Ts cos(2𝜋nf s t), 0 ≤ t ≤ Ts
𝛼0 (t) = (11.33)
0, elsewhere
where n is an integer and f s = 1/T s . Transmitted symbols differ only in amplitude, which, if equally spaced in the
range from 0 to V, leads to the following symbols
g0 (t) = 0
√
V Ts
g1 (t) = 𝛼 (t)
3 2 0
√
2V Ts
g3 (t) = 𝛼 (t)
3 2 0
√
Ts
g2 (t) = V 𝛼 (t) (11.34)
2 0
4-PSK: the signal space is 2D with the four states represented using a linear combination of two basis functions
{√
2∕Ts cos(2𝜋nf s t), 0 ≤ t ≤ Ts
𝛼0 (t) =
0, elsewhere
{ √
− 2∕Ts sin(2𝜋nf s t), 0 ≤ t ≤ Ts
𝛼1 (t) =
0, elsewhere
fs = 1∕Ts (11.35)

Transmitted symbols have the same amplitude V and frequency nf s , differing only in phase. One implementa-
tion that places states S0 , S1 , S2 , and S3 (representing di-bits 00, 01, 10, and 11, respectively) at respective phases
−135∘ , 135∘ , −45∘ , and 45∘ is
√
g0 (t) = −V Ts ∕4[𝛼0 (t) + 𝛼1 (t)]
√
g1 (t) = V Ts ∕4[−𝛼0 (t) + 𝛼1 (t)]
√
g2 (t) = V Ts ∕4[𝛼0 (t) − 𝛼1 (t)]
√
g3 (t) = V Ts ∕4[𝛼0 (t) + 𝛼1 (t)] (11.36)

where symbol gk (t) corresponds to state Sk , k = 0, 1, 2, 3.

Telegram: @Persian_IEEEComSoc

4-FSK: four basis functions are needed, one for each of the four states S0 , S1 , S2 , and S3
{√
2∕Ts cos[𝜋(n + mk)Rs t], 0 ≤ t ≤ Ts
𝛼k (t) =
′
0, elsewhere
k = 0, 1, 2, 3, · · · ; Rs = 1∕Ts (11.37)

where n, m ≥ 1 are integers and m determines the spacing of the frequencies of the transmitted sinusoidal symbols
gk (t), which are of equal amplitude V, but different frequencies. The smallest spacing is Rs /2 when m = 1. The
transmitted symbols are therefore
√
Ts ′
gk (t) = V 𝛼 (t), k = 0, 1, 2, 3 (11.38)
2 k
 11.4 Digital Transmission Model 699

S0 S1 S3 S2
(a) α0
0 V Ts 2V Ts Ts
V
3 2 3 2 2

α1

S1 V Ts 4 S3

(b) α0
–V Ts 4 V Ts 4

S0 –V Ts 4 S2

Figure 11.7 Worked Example 11.2: signal space diagrams of (a) 4-ASK; (b) 4-PSK.

Figure 11.7 shows the signal-space diagrams of the 4-ASK and 4-PSK discussed above. The orientation of the
𝛼 1 and 𝛼 0 axes in Figure 11.7b is consistent with our definition of 𝛼 1 (t) in Eq. (11.35) as a negative sine pulse and
𝛼 0 (t) as a cosine pulse, and the fact that the cosine function leads the sine function by 90∘ and therefore lags the
negative sine function by 90∘ . We have omitted the signal-space diagram of 4-FSK because it is hardly insightful
and could be confusing to sketch four mutually perpendicular axes.

11.4 Digital Transmission Model

The discussion so far leads us to adopt the simplified model for digital transmission shown in Figure 11.8. It
consists of a symbol generator capable of generating M distinct transmitted symbols {gk (t), k = 0, 1, …, M − 1}
each of duration T s ; a transmission medium that accounts for noise w(t), including contributions from the
receiver; and a symbol detector that determines which of the M symbols is most likely to have been sent given the
Telegram: @Persian_IEEEComSoc

(noise-corrupted) received signal r(t).

The transmission system is fully described by a set of N orthonormal basis functions 𝛼 0 (t), 𝛼 1 (t), …, 𝛼 N-1 (t), and
its signal-space diagram. The signal-space diagram provides a lucid summary of the adopted coding procedure that
maps each distinct input block of log2 M bits to a distinct point in N-dimensional space. There are M such input
blocks, and therefore M (message) points S0 , S1 , SM-1 in the signal-space diagram, which corresponds respectively
to the M transmitted symbols g0 (t), g1 (t), …, gM-1 (t). Point Sk is identified by a vector sk called the signal vector with
N elements sk0 , sk1 , …, sk,N-1 , which are the components of the point along the N mutually perpendicular axis. In
a commonly used arrangement, known as Gray coding, adjacent message points represent blocks of bits that differ
in only one bit position. Figure 11.5b is an example of Gray coding. Unless otherwise stated, we employ a notation
in which Sk represents a block of bits whose decimal equivalent is k. For example, with M = 8, S0 represents 000,
S5 represents 101, etc.
 700 11 Digital Modulated Transmission

Transmission
Transmitter medium Receiver

Input Symbol g k (t) r(t) Symbol Output

(a)
bitstream
Σ bitstream
generator + detector
+
w(t)
Noise

sk0 ×

α0(t)
Signal
vector Transmitted
sk1 × symbol
N–1
sk
(b) Σ gk(t) = Σ skjαj(t)
j=0
α1(t)

Sk,N–1
×

αN–1(t)

Ts
× ∫0 sk0

α0(t)
Signal
Received vector
Ts
symbol × ∫0 sk1
sk
(c)
g k (t)
α1(t)

Ts
× ∫0 sk,N–1
Telegram: @Persian_IEEEComSoc

αN–1(t)

matched filter
Received
Impulse
symbol
(d) response yo(Ts) = skj
gk(t) = αj(Ts – t) yo(t)
sample at
t = Ts

Figure 11.8 (a) Digital transmission model; (b) symbol generation; (c) symbol detection; (d) matched filter equivalent of jth
correlator.
 11.5 Noise Effects 701

We may summarise the processes performed at the transmit end of the simplified model in Figure 11.8 as follows:
● During each symbol interval, a block of log2 M input bits is read and mapped to point Sk according to the agreed
coding procedure. Equivalently, we say that signal vector sk is generated.
● Using the basis functions (which characterise the transmission system), and the signal vector sk , the transmitted
symbol gk (t) is generated according to the block diagram shown in Figure 11.8b. Note that this diagram is an
implementation of Eq. (11.12).
Ignoring noise effects for the moment, the receiver has the task of extracting the signal vector sk from the symbol
gk (t) received during each symbol interval. Once sk has been extracted, the corresponding bit block is obtained by
mapping from message point Sk to a block of log2 M bits according to the coding procedure used at the transmitter.
To see how this symbol detection may be carried out, consider the result of the following integration of the product
of the received symbol gk (t) and the jth basis function 𝛼 j (t). This is a correlation operation performed in a correlation
receiver when it is fed with inputs gk (t) and 𝛼 j (t). The correlation receiver is discussed further in Chapter 12.
Ts
gk (t)𝛼j (t)dt
∫0
Ts
= [sk0 𝛼0 (t) + sk1 𝛼1 (t) + · · · + skj 𝛼j (t) + · · · + sk,N−1 𝛼N−1 (t)]𝛼j (t)dt
∫0
Ts Ts Ts
= skj 𝛼j2 (t)dt + sk0 𝛼0 (t)𝛼j (t)dt + · · · + sk,j−1 𝛼j−1 (t)𝛼j (t)dt
∫0 ∫0 ∫0
Ts Ts
+sk,j+1 𝛼j+1 (t)𝛼j (t)dt + · · · + sk,N−1 𝛼N−1 (t)𝛼j (t)dt
∫0 ∫0
= skj (11.39)
In the above, we obtain the second line by expanding gk (t) according to Eq. (11.12), and the last line by invoking
the orthonormality property of the basis functions – Eq. (11.13). We see that the operation yields the jth element
skj of the desired vector sk . Thus, to determine the entire vector sk , we feed the received symbol gk (t) as a common
input to a bank of N correlators each of which is supplied with its own basis function. The N outputs of this
arrangement, which is shown in Figure 11.7c, are the N elements of the vector sk . We find in Chapter 12 that
the correlator supplied with basis function 𝛼 j (t) is equivalent to a matched filter that gives optimum detection (in
the presence of white noise) of a symbol of the same waveform as 𝛼 j (t). Thus, each correlator in the bank may be
replaced by the matched filter equivalent shown in Figure 11.8d.

11.5 Noise Effects

Telegram: @Persian_IEEEComSoc

In practice, the transmitted symbol is corrupted by noise before it reaches the detection point at the receiver, and
the input to the bank of correlators discussed above is the signal
r(t) = gk (t) + w(t) (11.40)
In most practical situations it is adequate to assume that w(t) is AWGN. The output of the jth correlator is
Ts Ts
r(t)𝛼j (t)dt = [gk (t) + w(t)]𝛼j (t)dt
∫0 ∫0
Ts Ts
= gk (t)𝛼j (t)dt + w(t)𝛼j (t)dt
∫0 ∫0
= skj + wj (11.41)
 702 11 Digital Modulated Transmission

Comparing Eqs. (11.39) and (11.41), we see that the effect of noise is to shift the output of the jth correlator by a
random amount
Ts
wj = w(t)𝛼j (t)dt (11.42)
∫0
In other words, rather than the desired vector sk (which corresponds to a precise message point Sk ), we now
have at the output of Figure 11.8c an output vector

r = sk + w (11.43)

where w is a random vector with components w0 , w1 , …, wN-1 , given by Eq. (11.42) with j = 0, 1, …, N-1, respec-
tively. The received vector r corresponds to a received signal point R in signal space. This point is displaced from the
message point Sk by a distance |w|. The displacement can be in any direction with equal likelihood, but smaller
displacements are more likely than large ones, in line with the probability density function (pdf) of a Gaussian
random variable.
√ The picture is as shown in Figure 11.9 for two adjacent message points S1 and S2 separated by a
distance E. The level of shading at a point gives an indication of the likelihood of the received signal point R to
be around that point.
Two important tasks must now be performed at the receiver in order to recover an output bit stream from the
received signal r(t). First, a bank of correlators (as in Figure 11.8c) is used to extract the received vector r. Next,
given vector r, or equivalently point R, which in general does not coincide exactly with any of the message points
{Sk , k = 0, 1, …, M-1}, the receiver has to make a decision on which message point is most likely to have been
transmitted. On the assumption that all M message points are transmitted with equal probability, the maximum
likelihood rule leads to a decision in favour of the message point that is closest to the received signal point. The
decision boundary for message points S1 and S2 (in Figure 11.9) is the perpendicular bisector of the line joining S1
and S2 . Therefore, a symbol error will occur whenever noise effects shift the received point R across this decision
boundary. √
Clearly, the likelihood of such an error increases as the spacing E between signal points is reduced. Because of
the random nature of noise, we can only talk of the probability Pe that a symbol error will occur but cannot predict
with certainty the interval when this error will occur. For example, a probability of symbol error Pe = 0.01 means
that on average one symbol in a hundred will be incorrectly received. Note that this does not imply that there
will always be one error in every 100 transmitted symbols. In fact, there may well be some periods of time during
which a thousand or more symbols are received without a single error, and others in which there are two or more
errors in 100 symbols. What this statement means is that if we observe the transmission over a sufficiently long
time then we will find that the ratio of the number of symbols in error to the total number of symbols transmitted
is Pe = 0.01. This probability is therefore also referred to as symbol error ratio, which we determine below for a
binary transmission system.
Telegram: @Persian_IEEEComSoc

Decision Figure 11.9 Effect of noise.

boundary

R
*
w

S0 S1

E
 11.6 Symbol and Bit Error Ratios 703

11.6 Symbol and Bit Error Ratios

To obtain a more quantitative expression of √the effect of noise, consider a binary system with two message points
S0 and S1 that are separated by distance E in signal space as shown in Figure 11.10a. We wish to derive an
expression for the probability of symbol error in this binary transmission system when incoming symbols are
coherently detected in the presence of AWGN using the correlation receiver discussed in Figure 11.8c. If the system
transmits at a symbol rate Rs = 1/T s , where T s equals symbol duration then its minimum baseband noise equivalent
bandwidth is
Rs 1
B= =
2 2Ts

and, at the detection point, the noise power Pn (which equals the variance 𝜎 2 of the Gaussian noise vn (t) since it
has zero mean) is

Pn = 𝜎 2 = No B = No ∕2Ts (11.44)

where N o is the noise power per unit bandwidth.

A sample of this noise has in-phase and quadrature components vnI and vnQ (as discussed in Section 6.3) with
respective energies EnI and EnQ over one symbol duration given by

EnI = v2nI Ts ; EnQ = v2nQ Ts

As a result of the addition of this noise, a transmitted state S0 will be received at point S0′ having been displaced
through distance a along the in-phase axis 𝛼 and distance b along the quadrature axis 𝛽, as shown in Figure 11.10a.

Figure 11.10 (a) Two-dimensional signal √ space with two β

states S 0 and S 1 separated by distance E; (b) shaded area
gives the probability that S 0 is received in error. S′o d= E 2

So b S1
a
α
–d +d

Decision boundary
Telegram: @Persian_IEEEComSoc

(a)

p(ʋnI) =
1
exp
⎛ ʋnI2 ⎞
σ 2π ⎝ 2σ 2 ⎠

1
z= E Ts
2

ʋnI
z
(b)
 704 11 Digital Modulated Transmission

Since distance in signal space is the square root of energy, it follows that
√ √
a = EnI = vnI Ts
√ √
b = EnQ = vnQ Ts (11.45)

There will be symbol error if S0′ lies to the right of the decision boundary, which will be the case if a > d, i.e. if
√ √
√ 1 E
vnI Ts > E∕2; ⇒ vnI > ≡z (11.46)
2 Ts
Since vnI is a Gaussian random variable of zero-mean and variance 𝜎 2 given by Eq. (11.44), it follows that the
probability Pe0 of symbol error given that S0 is sent is the shaded area of Figure 11.10b, which we evaluate in Eq.
(6.16) as
Pr(vnI > z) = Pr[S1 ∣ S0 ] ≡ Pe0
( )
( )
1 z z
= erfc √ =Q (11.47)
2 𝜎 2 𝜎
This equation involves the complementary error function erfc(x) and Q-function Q(x) discussed in Section 3.4.1
and tabulated in Appendix C. From the symmetry of the problem in Figure 11.10a, the probability Pe1 of an error
occurring when S1 is sent is the same as Pe0 . A transmission channel that satisfies this condition, Pe1 = Pe0 , is
referred to as a binary symmetric channel. Therefore the probability Pe of an error occurring in the detection of
any symbol is given by Eq. (11.47), which when we substitute the expressions for z and 𝜎 (given in Eq. (11.46) and
(11.44)) yields
( √ ) (√ )
1 1 E E
Pe = erfc =Q (11.48)
2 2 No 2No
To reiterate, Eq. (11.48) gives the symbol error ratio (SER) in a binary transmission system where (i) bandlimited
white Gaussian noise of power per unit√bandwidth N o is the only source of degradation and (ii) the two transmit-
ted states are separated by a distance E in signal space. We show in Section 11.6.1 that Eq. (11.48) is directly
applicable to the following digital transmission systems:
● Unipolar baseband (UBB) systems, e.g. all variants of the non-return-to-zero (NRZ) line code in Figure 10.23
(except bipolar) and the RZ code.
● Bipolar baseband (BBB) systems, e.g. bipolar NRZ and Manchester codes.
● All binary modulated systems, including ASK, PSK, and FSK.
We will also show with a specific example of 4-PSK how Eq. (11.48) may be applied to obtain the BER of M-ary
systems. In the above basic analysis, the transmitted states were located along a single axis in a signal space. In
Section 11.6.2 we extend the analysis to obtain the BER of all binary transmission systems in terms of the received
Telegram: @Persian_IEEEComSoc

average energy per bit Eb . All our results are expressed exclusively in terms of erfc. If required, these may be
converted into equivalent Q-function expressions using the relation
√
erfc(x) = 2Q( 2 x) (11.49)
The following two characteristics of the complementary error function have implications on the interplay
between the carrier-to-noise ratio (C/N) at the receiver’s detection point and the BER of the transmission system:
● erfc(x) decreases monotonically as x increases.
● The rate of decrease of erfc(x) is higher at large x.
The first characteristic means that we can always improve (i.e. reduce) BER by increasing C/N, whereas the
second implies that the improvement in BER per dB increase in C/N is larger at higher values of C/N.
 11.6 Symbol and Bit Error Ratios 705

11.6.1 Special Cases

Worked Example 11.3

Apply Eq. (11.48) to obtain expressions for the BER of the following systems in terms of the average energy per
bit Eb and the noise power per unit bandwidth N o :

(a) UBB transmission and ASK

(b) BBB transmission and PSK
(c) FSK
(d) 4-PSK.
Symbol error ratio (SER) and BER are identical in binary systems, since each transmitted symbol conveys one
bit. Eq. (11.48) will therefore give the desired expression when E is expressed in terms of Eb . In the following
√ we
sketch the signal-space diagram of each system with their two states S0 and S1 separated by distance E, write
out the implied relationship between E and Eb , and substitute this relationship into Eq. (11.48) to obtain BER.
(a) Unipolar baseband (UBB) transmission and ASK have the 1D signal space shown in Figure 11.11a. Binary ASK
is always implemented as on–off keying (OOK), so state S0 representing bit 0 has zero energy, which means
that there is no signalling during the symbol interval T s . State S1 , representing bit 1, has energy Es . Clearly,
Es = E and therefore Eq. (11.48) yields
( √ )
1 1 Es
BER = erfc
2 2 No

Es is the peak symbol energy, which is contained in state S1 , whereas state S0 contains zero energy. If both
states are equally likely then the average energy per symbol is Esav = Es /2. In fact, Esav is also the average
energy per bit Eb since each symbol conveys one bit. Thus, substituting 2Eb for Es in the above expression
yields the desired formula
(√ )
1 Eb
BER = erfc (UBB and 2-ASK) (11.50)
2 2No

(b) Figure 11.11b shows the signal-space diagram of bipolar

√ baseband (BBB) and (binary) PSK systems. S0 and S1
√
have equal energy Es , but opposite polarity. Clearly, E = 2 Es , and in this case Es is also the average energy
per bit Eb . Thus, Eq. (11.48) becomes
(√ )
1 Eb
BER = erfc (BBB and 2-PSK) (11.51)
2 No
Telegram: @Persian_IEEEComSoc

(c) The signal-space diagram of a (binary) FSK system is 2D, as shown in Figure 11.11c. The two states S0 and S1
have equal energy Es (which is also the average energy per bit Eb ), but are transmitted at different frequencies
represented by the orthogonal basis functions 𝛼0′ (t) and 𝛼1′ (t). These states are shown separated by distance
√
E as required for using Eq. (11.48). Applying Pythagoras’s rule to the diagram, we see that E = 2Es (= 2Eb ).
Eq. (11.48) therefore yields
(√ )
1 Eb
BER = erfc (FSK) (11.52)
2 2No

(d) 4-PSK, also referred to as quadriphase shift keying (QPSK), has a 2D signal-space diagram, which was sketched
in Figure 11.7b, but is repeated in Figure 11.12a using a labelling that is appropriate to the following
√discussion.
There are four states, each with energy Es . By Pythagoras’s rule, the distance between S0 and S1 is 2Es , and so
 706 11 Digital Modulated Transmission

(a)
S0 S1
α0
0 Es
E

(b)
S0 S1
α0
– Es 0 Es
E

α′1

Es S1
2 2 2
( Es ) +( Es ) =( E)
(c) ⇒ 2 Es = E
E

S0
α′0
0 Es

Figure 11.11 Signalspace diagrams for Worked Example 11.3. (a) unipolar baseband systems and ASK (usually OOK);
(b) bipolar baseband and PSK; (c) FSK.
√
is the distance between S0 and S2 . The separation between S0 and S3 is obviously 2 Es . Let us determine
the probability Pe0 that there is an error given that S0 was transmitted. Clearly, an error will occur if S0 is sent
whereas the received point R lies in quadrant 2, 3, or 4 – the shaded region in Figure 11.12b.
In order to apply Eq. (11.48), we take two states at a time as shown in Figure 11.12c. Consider c(i). All points
in the shaded region are nearer to S1 than S0 , and the receiver will therefore decide in favour of S1 whenever
the received point lies in this region. So, an error occurs if S0 is sent but the received state lies in the shaded area. The
probability
√ of this error, denoted Pr(S1 |S0 ) and read ‘probability S1 received given S0 sent’, is given by Eq. (11.48)
√
Telegram: @Persian_IEEEComSoc

with E = 2Es
(√ )
1 Es
Pr(S1 ∣ S0 ) = erfc
2 2No
Similarly, in c(ii) and c(iii)
(√ )
1 Es
Pr(S2 ∣ S0) = erfc
2 2No
(√ )
1 Es
Pr(S3 ∣ S0) = erfc
2 No
 11.6 Symbol and Bit Error Ratios 707

2Es α1
α1
S1 S0 S1 S0

Es
Es
Quadrant 2 Quadrant 1
(a) α0 2Es (b) α0

Es Quadrant 3 Quadrant 4
Es

S3 S2 S3 S2

α1 α1 α1
S1 S0 S0 S0

(c) α0 α0 α0

S2 S3

(i) Pr(S1|S0) (ii) Pr(S2|S0) (iii) Pr(S3|S0)

Figure 11.12 QPSK: (a) signal space diagram; (b) error occurs if S 0 is sent but the received signal lies in shaded region;
(c) taking two states at a time, error occurs if the received signal falls in the shaded region.

Observe that the shaded area of Figure 11.12b is given by the sum of the shaded areas in Figure 11.12c(i) and (ii)
less half the shaded area in c(iii) – to correct for quadrant 3 being included twice in the summation. Therefore,
the probability Pe0 that the received state lies in the shaded region of Figure 11.12b is given by
1
Pe0 = Pr(S1 ∣ So ) + Pr(S2 ∣ So ) − Pr(S3 ∣ So )
2
(√ ) (√ )
Es 1 Es
= erfc − erfc
2No 4 No
(√ )
Es
≃ erfc
2No
Telegram: @Persian_IEEEComSoc

where the approximation holds because the second term in the previous line is negligible compared to the first term
for practical values of Es /N o . For example, the ratio between the two terms is 25 at Es /N o = 5 dB, increasing rapidly
to 809 at Es /N o = 10 dB. An important implication of this observation is that when Es /N o is large then Pr(S3 |S0 )
is small compared to Pr(S1 |S0 ) and Pr(S2 |S0 ). In other words, it can be assumed that errors involve the mistaking
of one symbol for its nearest neighbours only. From the symmetry of the signal-space diagram, the probability
of error in any of the other symbols is the same as obtained above for S0 . Thus, the probability of error in any
symbol is
(√ ) (√ )
Es Eb
Pes ≡ SER = erfc = erfc (11.53)
2No No
 708 11 Digital Modulated Transmission

0.1

10–2 2–
AS
K,2
–F
10–3 SK
,U
BB

Q
10–4

PS
K
,2
–P
BER

10–5

SK
,B
BB
10–6

10–7

10–8

10–9
2 4 6 8 10 12 14 16
Eb/No, dB

Figure 11.13 Bit error ratio (BER) versus E b /No of selected digital transmission systems. (BBB = bipolar baseband;
UBB = unipolar baseband.)

since each symbol conveys two bits, so that Es = 2Eb . Finally, to obtain BER, we observe that, in M-ary transmission
with Gray coding, neighbouring states differ in only one bit position. An error in one symbol, which represents
log2 M bits, gives rise to one bit error. Thus
SER
BER = (11.54)
log2 M
In this case, with M = 4 and SER given by Eq. (11.53), we obtain
(√ )
1 Eb
BER = erfc (QPSK) (11.55)
2 No

From Worked Example 11.3, we summarise the following important results, which are also plotted
in Figure 11.13 with Eb /N o expressed in dB. An important word of caution is in order here: before using
Telegram: @Persian_IEEEComSoc

any of the formulas for BER presented in this chapter, the quantity Eb /N o must first be computed as the (non-dB)
ratio between Eb in joules (≡ watt-second) and N o in watt/hertz (≡ watt-second).
⎧1 (√ )
Eb
⎪ 2 erfc , (2-ASK, 2-FSK & UBB)
BER = ⎨ 1 (√ 2No) (11.56)
Eb
⎪ 2 erfc , (2-PSK, QPSK & BBB)
⎩ No

11.6.2 Arbitrary Binary Transmission

Equation (11.48) may be expressed in a very useful form that gives the BER of any type of binary transmission
system directly in terms of the average energy per bit Eb . To do so, consider the general signal-space diagram
 11.6 Symbol and Bit Error Ratios 709

Figure 11.14 Signalspace diagram of an arbitrary binary α1

transmission system.
S0

E s01 – s11
E0
s01

S1
s11
E1
α0
s00 s10 – s00
s10

shown in Figure 11.14. This diagram applies to all binary transmission systems with an appropriate choice of
coefficients
√ (s00 , s01 , s10 , s11 ) and basis functions 𝛼 0 (t) and 𝛼 1 (t). We also know that the BER is given by Eq. (11.48)
with E the distance between states S0 and S1 , and that this system employs two symbols
g0 (t) = s00 𝛼0 (t) + s01 𝛼1 (t), Binary 0
g1 (t) = s10 𝛼0 (t) + s11 𝛼1 (t), Binary 1

The respective energies of the symbols are

E0 = s200 + s201
E1 = s210 + s211 (11.57)

The average energy per bit Eb is

E0 + E1
Eb = (11.58)
2
The correlation coefficient of the two symbols follows from Eq. (11.24)
Ts
1
𝜌= g0 (t)g1 (t)dt
Eb ∫ 0
s T
1
= [s00 𝛼0 (t) + s01 𝛼1 (t)][s10 𝛼0 (t) + s11 𝛼1 (t)]dt
∫
Eb 0
{ Ts Ts }
1
= s00 s10 𝛼0 (t)dt + s01 s11
2
𝛼1 (t)dt
2
∫0 ∫0
Telegram: @Persian_IEEEComSoc

Eb
s s + s01 s11
= 00 10 (11.59)
Eb
In the above, we obtain the third line by ignoring the integrals involving the product 𝛼 0 (t)𝛼 1 (t) since they evaluate
to zero (in view of the orthogonality property), and the last line by noting that 𝛼 0 (t) and 𝛼 1 (t) are unit-energy basis
functions.
Finally, applying Pythagoras’s rule in Figure 11.14 allows us to express the energy E in terms of Eb and 𝜌 as
follows
√
( E)2 = (s01 − s11 )2 + (s10 − s00 )2
= (s200 + s201 ) + (s210 + s211 ) − 2(s00 s10 + s01 s11 )
 710 11 Digital Modulated Transmission

Replacing each term on the right-hand side with its equivalent from Eqs. (11.57–11.59) yields the following impor-
tant relation
E = E0 + E1 − 2𝜌Eb
= 2Eb − 2𝜌Eb
= 2Eb (1 − 𝜌) (11.60)
Substituting this relation in Eq. (11.48) gives the BER of any binary transmission system (assumed to have binary
symmetry)
√
⎛ E (1 − 𝜌) ⎞
1
BER = erfc ⎜ b ⎟ (11.61)
2 ⎜ 2No ⎟
⎝ ⎠
It is worth emphasising that Eq. (11.61) applies to all binary symmetric transmission systems, whether modu-
lated or baseband. A few special cases will help to demonstrate the utility of this important equation.
● Identical symbols: if g0 (t) = g1 (t), then 𝜌 = 1, and Eq. (11.61) gives BER = 0.5erfc(0) = 0.5. It would be ridiculous
to use the same symbol to convey both binary 1 and 0. The BER is the same as would be obtained by basing each
decision entirely on the result of flipping a fair coin. The receiver does not gain any information from detecting
the incoming symbols and should not even bother.
● PSK and BBB: two antipodal symbols are used. With g0 (t) = −g1 (t), we obtain 𝜌 = −1. Equation (11.61) then
reduces to Eq. (11.56).
● FSK: two orthogonal symbols are used, giving 𝜌 = 0. Equation (11.61) then reduces to (11.56).
● ASK and UBB: two symbols are used that differ only in their amplitudes A0 and A1 , which are of course positive
numbers. You may wish to verify that in this case 𝜌 ≥ 0. Specifically
2A0 A1
𝜌= (11.62)
A20 + A21

● We see from Eq. (11.61) that, for a given Eb , the lowest BER is obtained when A0 = 0, giving 𝜌 = 0. For all other
values of A0 , the correlation coefficient 𝜌 has a positive value between 0 and unity. This reduces the argument
of the complementary error function and leads to a larger BER. Setting A0 = 0 gives what is known as on–off
keying (OOK). It is therefore clear that OOK gives ASK its best (i.e. lowest) possible BER. Assigning nonzero
values to both A0 and A1 always results in a higher BER compared to an OOK of the same energy per bit. Note
that setting A0 and A1 to the same nonzero value yields 𝜌 = 1, and BER = 0.5. This, and the case of A0 = A1 = 0,
corresponds to the identical-symbol system discussed above.
From the foregoing we have a very clear picture of the BER performance of the three types of binary modulation.
Telegram: @Persian_IEEEComSoc

We see that PSK gives a lower BER than either ASK or FSK for a given received signal power, which is measured in
terms of the average energy per bit Eb . To achieve the same error ratios in the three systems, twice as much symbol
energy (i.e. 3 dB increase) is required in ASK and FSK. The increase is more worthwhile at higher values of Eb /N o .
For example, when Eb /N o is 12 dB then a 3 dB increase in Eb improves the BER of ASK and FSK dramatically
from 3.4 × 10−5 to 9 × 10−9 . On the other hand, at Eb /N o = 2 dB, a similar increase in Eb only yields a modest
improvement in BER, from 0.1 to 0.04.
An important clarification is necessary about the BER of QPSK when compared to binary phase shift keying
(BPSK). For the same symbol rate and hence bandwidth, QPSK allows transmission at twice the bit rate of BPSK.
However, Figure 11.13 shows that QPSK and BPSK have the same BER at the same Eb /N o and this could be
erroneously interpreted to mean that QPSK has somehow achieved a doubling of bandwidth efficiency (over BPSK)
at the same signal quality without requiring additional signal power. But Eb is energy per bit and not a direct
 11.6 Symbol and Bit Error Ratios 711

measure of signal power. If N o is equal in both systems then a direct comparison of their signal power is given by
the difference between the quantity C/N o (expressed in dB) in each system. This ratio between carrier power and
N o may be expressed in terms of Eb /N o as
C E
= b Rb (11.63)
No No
where we have made use of Eq. (11.19) for signal power in terms of Eb and bit rate Rb . Therefore, to have the same
Eb /N o as BPSK (as in Figure 11.13), the QPSK signal power must be a factor of two (i.e. 3 dB) higher than that of
BPSK since its bit rate is twice that of BPSK.

Worked Example 11.4

A binary PSK system transmits at 140 Mbit/s. The noise power per unit bandwidth at the detection point of the
receiver is 5 × 10−21 W/Hz and the received signal power is −82 dBm at the same point.
(a) Determine the BER.
(b) Show how BER may be improved to 1 × 10−8 if the modulation technique and noise level remain
unchanged.

(a) To determine BER we need the average energy per bit Eb , which equals the product of received power and bit
duration. The received power P (in watts) is
P = [10(−82∕10) ] × 10−3 = 6.31 × 10−12 W = 6.31 pW
The duration of one bit is given by
1 1
Ts = = = 7.143 × 10−9 s
Bit Rate 140 × 106
Therefore, energy per bit is
Eb = PT s = 4.507 × 10−20 J
With N o = 5 × 10−21 , the ratio Eb /N o = 9.0137 or 9.55 dB. There are now several options for obtaining the BER.
From the PSK curve in Figure 11.13, we see that at Eb /N o = 9.55 dB, the BER is 1.1 × 10−5 . Alternatively, using
Eq. (11.56), with Eb /N o = 9.0137 we obtain
√
BER = 0.5erfc( 9.0137) = 0.5erfc(3)
Now we may read the value erfc(3) = 2.2 × 10−5 from the tables in Appendix C, or directly calculate it using
the formula provided in Eq. (3.39). Whichever way, the result is BER = 1.1 × 10−5 .
(b) Figure 11.13 shows that in binary PSK a BER of 1 × 10−8 requires Eb /N o = 12 dB. Therefore, with N o
Telegram: @Persian_IEEEComSoc

unchanged, we must increase Eb by 2.45 dB (i.e. 12–9.55) or a factor of 1.758 to achieve this lower BER. Since
Eb = PT s , it means that Eb may be increased either by increasing the transmitted power (and hence received
power P) or increasing the symbol duration T s . Note that increasing T s is equivalent to reducing bit rate by
the same factor. Therefore, we may maintain the bit rate at 140 Mbit/s but raise the transmitted power to
give a received power level of 6.31 × 1.758 = 11.1 pW. Alternatively, we maintain the previous power level
but reduce the bit rate to 140/1.758 = 79.64 Mbit/s. The transmitted power level is often restricted in order
to minimise interference to other systems, or to reduce radiation hazards to users of handheld transmitters,
or to prolong battery life in portable systems. However, with Eb determined by both transmitted power level
and bit rate, appropriate values of these two parameters may often be found to match the noisiness of the
transmission system and achieve a desired BER.
 712 11 Digital Modulated Transmission

11.7 Binary Modulation

In binary modulation one symbol is transmitted to represent each bit in the information-bearing bit stream. There
are therefore two distinct symbols, namely g0 (t) representing binary 0 and g1 (t) representing binary 1. Binary
modulation is the simplest special case of our discussion in previous sections with
Number of states in Signal Space M = 2
Symbol duration Ts = Bit duration Tb
Symbol Rate Rs = Bit Rate Rb
Symbol Energy Es = Bit Energy Eb (11.64)
Binary modulation has already been discussed to some extent in this chapter, especially in the worked examples.
This section is devoted to a brief discussion of the generation and bandwidth of the three types of binary modulated
signals, namely ASK, FSK, and PSK. Each generator in the following discussion involves a product modulator,
which is represented as a multiplier. A detailed description of the operation of a product modulator is presented
in Section 7.7.1.2.

11.7.1 ASK
ASK signal g(t) may be generated using the circuit shown in block diagram form in Figure 11.15a. The bit stream
is first represented as a unipolar non-return-to-zero (UNRZ) waveform. At the output of the UNRZ coder, binary
1 is represented by a pulse of height +1 (normalised) and duration spanning the entire bit interval T b , and binary
0 by the absence of a pulse in the bit interval. The pulse shape is shown as rectangular but may be shaped (using,
for example, the root raised cosine filter (defined in Eq. (4.173) and further discussed in Chapter 12)) in order to
reduce the frequency components outside the main lobe of Figure 11.3. The UNRZ waveform and a sinusoidal
carrier signal Ac cos(2𝜋f c t) (which is a constant multiple K of the basis function 𝛼 0 (t) of the system) are applied to
a product modulator. The resulting output is the desired ASK signal. This consists of a sinusoidal pulse during the
bit interval for a binary 1 and no pulse for a binary 0
{
g1 (t), Binary 1
vask (t) =
0, Binary 0
{
Ac cos(2𝜋fc t), 0 ≤ t ≤ Tb
g1 (t) =
0, elsewhere
fc = n∕Tb = nRb , n = 1, 2, 3, · · · (11.65)
The frequency f c of the sinusoidal symbol is an integer multiple of the bit rate Rb , and the amplitude Ac has a
Telegram: @Persian_IEEEComSoc

value that gives the required average energy per bit Eb . If binary 1 and 0 are equally likely in the input bit stream,
it follows from Eq. (11.3) that
( 2 )
1 Ac A2 T
Eb = Tb + 0 = c b
2 2 4
√
Or, Ac = 2 Eb ∕Tb (11.66)
To examine the spectrum of ASK, let us rewrite vask (t) in the form
Ac
vask (t) = (1 ± m) cos(2𝜋fc t), m=1 (11.67)
2
 11.7 Binary Modulation 713

Ac
1 Tb 0
Bit stream 0
1 0 1 … Unipolar –Ac
ASK
(a) × signal
NRZ coder ʋm(t) ʋASK(t)
Kα0(t) = Ac cos(2πfct)

|Vunrz(f)| |Vask(f)|
Lower
sideband Upper
(b) AM
sideband
f f
0 B fc
Bask = 2B

Sask(f) A2c
δ( f – fc)
A2c 4
4Rb
(c) Rb = 1/Tb

0 f
fc – 3Rb fc – 2Rb fc – Rb fc fc + Rb fc + 2Rb fc + 3Rb

Figure 11.15 ASK: (a) modulator; (b) single-sided amplitude spectrum |Vask (f )|, based on a triangular shape for the
spectrum |Vunrz (f )| of a unipolar NRZ waveform; (c) single-sided power spectral density (PSD), assuming a random bit stream.

where the positive sign holds during an interval of binary 1 and the negative sign during binary 0. In this form, we
see that ASK is a double sideband transmitted carrier amplitude modulation signal, with modulation factor m = 1.
A little thought and, if need be, reference to Chapter 7 will show that in this view, the modulating signal vm (t) is
a bipolar NRZ waveform of value Ac /2 during binary 1, and −Ac /2 during binary 0, and the unmodulated carrier
amplitude is Ac /2. The spectrum of the ASK signal then consists of an upper sideband, a lower sideband (LSB), and
an impulse (of weight Ac /2) at the carrier frequency, as shown in Figure 11.15b for a symbolic spectrum of vm (t).
The bandwidth of ASK is therefore twice the bandwidth of the baseband bipolar NRZ waveform. We have assumed
that the input bit stream is completely random, 1’s and 0’s occurring with equal likelihood. Under this condition
the power spectral density SB (f ) of the bipolar NRZ waveform equals the square of the amplitude spectrum of the
Telegram: @Persian_IEEEComSoc

constituent pulse. Thus, assuming a rectangular pulse shape

SB (f ) = sinc2 (fTb ), Normalised (11.68)

Figure 11.15c shows a plot of SB (f ), which decreases rapidly as the inverse square of frequency, and has a null
bandwidth equal to the bit rate Rb (= 1/T b ). Setting the baseband signal bandwidth equal to this null bandwidth,
it follows that the bandwidth of ASK is given by

Bask = 2Rb (11.69)

 714 11 Digital Modulated Transmission

11.7.2 PSK
A block diagram for the generation of PSK is shown in Figure 11.16. The bit stream is first coded as a bipolar
non-return-to-zero (BNRZ) waveform. At the output of the BNRZ coder, binary 1 is represented by a pulse of height
+1 and duration spanning the entire bit interval T b , and binary 0 is represented by a similar pulse but of opposite
polarity. Pulse shaping may be included prior to modulation by following the coder with a suitable filter or making
the filter an integral part of the coder. The BNRZ waveform is applied to a product modulator along with a sinu-
soidal signal K𝛼 0 (t), a constant multiple of the basis function of the system. The resulting output is the desired PSK
signal. This consists of two distinct sinusoidal pulses of duration T b that have the same frequency f c and amplitude
Ac but differ in phase by 180∘ . The generation process shown in Figure 11.16a leads to a sinusoidal pulse with 0∘
phase during intervals of binary 1, and an opposite polarity pulse (i.e. 180∘ phase) during intervals of binary 0.
That is
{
g1 (t), Binary 1
vpsk (t) =
−g1 (t), Binary 0
{
Ac cos(2𝜋fc t), 0 ≤ t ≤ Tb
g1 (t) =
0, elsewhere
fc = n∕Tb = nRb , n = 1, 2, 3, · · · (11.70)

1 Ac
Tb
0
Bit stream –1 –Ac
101… Bipolar PSK
(a)
NRZ coder ʋm(t)
× ʋpsk(t) signal

Kα0(t) = Ac cos (2πfct)

|Vbnrz(f)| |Vpsk(f)|
Lower Upper
sideband sideband
(b) DSB
f f
0 B fc
Bpsk = 2B
A2c Spsk(f)
Telegram: @Persian_IEEEComSoc

Rb
Rb = 1/Tb
(c)

f
fc – 3Rb fc – 2Rb fc – Rb fc fc + Rb fc + 2Rb fc + 3Rb

Figure 11.16 PSK: (a) modulator; (b) amplitude spectrum |Vpsk (f )|, based on a triangular shape for the spectrum |Vbnrz (f )| of
a BNRZ waveform; (c) single-sided PSD, assuming a random bit stream.
 11.7 Binary Modulation 715

Both transmitted symbols have the same energy. In this case the energy per bit Eb is given by
Eb = A2c Tb ∕2
√
Or, Ac = 2Eb ∕Tb (11.71)
We may obtain the spectrum of PSK by noting that Figure 11.16a represents a double sideband suppressed carrier
amplitude modulation. PSK will therefore have a spectrum like that of ASK, except that there is no impulse at the
frequency point f c , which reflects the absence or suppression of the carrier. Figure 11.16b and c show, respectively,
a representative amplitude spectrum of PSK and the power spectral density of a PSK signal when the input bit
stream is completely random. Clearly, PSK has the same bandwidth as ASK, which is twice the bandwidth of the
baseband bipolar waveform
Bpsk = 2Rb (11.72)

11.7.3 FSK
11.7.3.1 Generation
In FSK, two orthogonal sinusoidal symbols are employed, one of frequency f 1 to represent binary 1 and the other
of frequency f 0 to represent binary 0. FSK can therefore be generated by combining (i.e. interleaving) two ASK
signals as shown in Figure 11.17a. The bit stream is first represented using a UNRZ waveform. The output of
the UNRZ coder is fed directly into the top product modulator, which is also supplied with a sinusoidal signal of

1 Ac
Bit stream Tb
101… –Ac
Unipolar 0
NRZ coder × ASKf1 Ac
+
–Ac
(a) Inverter
Ac cos(2π f1t) Σ FSK signal
+
1 ASKf0

0
× Ac

–Ac
Ac cos(2π f0t)

A2c A2c
Sfsk(f) δ( f – f0) δ( f – f1)
A2c 4 4
Telegram: @Persian_IEEEComSoc

4Rb

(b)
Rb = 1/Tb

f
f0 – Rb f0 f0 + Rb f 1 – Rb f1 f1 + Rb

Figure 11.17 FSK: (a) modulator; (b) single-sided PSD, assuming a random bit stream.
 716 11 Digital Modulated Transmission

frequency f 1 . The output of the top modulator is therefore an ASK signal that has a sinusoidal pulse of frequency
f 1 during intervals of binary 1, and no pulse during intervals of binary 0. The UNRZ coder output is also fed into
the lower product modulator but is first passed through an inverter. The lower modulator is supplied with another
sinusoidal signal of frequency f 0 . The inverter produces a UNRZ waveform, which has a value +V during intervals
of binary 0, and a value 0 for binary 1. As a result, the output of the lower modulator is another ASK signal, but
one which contains a sinusoidal pulse of frequency f 0 during intervals of binary 0 and no pulse during intervals
of binary 1. It is easy to see that by combining the outputs of the two modulators in a summing device, we obtain
a signal that contains a sinusoidal pulse of frequency f 1 for binary 1, and another sinusoidal pulse of frequency f 0
for binary 0. This is the desired FSK signal, and we may write
{
g1 (t), Binary 1
vfsk (t) =
g0 (t), Binary 0
{
Ac cos(2𝜋f1 t), 0 ≤ t ≤ Tb
g1 (t) =
0, elsewhere
{
Ac cos(2𝜋f0 t), 0 ≤ t ≤ Tb
g0 (t) =
0, elsewhere
f1 = n1 ∕Tb ; f0 = n0 ∕Tb ; n1 ≠ n0 = 1, 2, 3, · · · (11.73)
It is important that the two transmitted symbols g1 (t) and g0 (t) are orthogonal. This requires that the sinusoidal
signals supplied to the pair of modulators in Figure 11.17a should always have the same phase. There is an implicit
assumption of this phase synchronisation in Figure 11.17a, where the sinusoidal signals both have the same phase
(0∘ ). Phase synchronisation coupled with the use of sinusoids whose frequencies are integer multiples of the bit rate
ensures that there is phase continuity between symbols. The FSK is then described as continuous phase frequency
shift keying (CPFSK). Note that both symbols have the same energy. The average energy per bit is the same as in
PSK and is given by Eq. (11.71).

11.7.3.2 Spectrum
The view of FSK as two interleaved ASK signals leads logically to the power spectral density of FSK shown in
Figure 11.17b. Each constituent ASK power spectrum has an impulse of weight A2c ∕4 at its respective carrier fre-
quency, as explained earlier. It follows from this power spectrum that the bandwidth of FSK is given by
Bfsk = (f1 − f0 ) + 2Rb (11.74)
The bandwidth increases as the frequency spacing f 1 −f 0 between the two orthogonal symbols g1 (t) and g0 (t). In
Eq. (11.73), f 1 and f 0 are expressed as integer multiples of 1/T b . This means that they are selected from the set of
orthogonal sinusoidal pulses discussed in Worked Example 11.2b. The minimum frequency spacing in this set is
Telegram: @Persian_IEEEComSoc

f 1 − f 0 = 1/T b , which is obtained by setting n1 = n0 + 1 in Eq. (11.73).

11.7.3.3 Frequency Spacing and MSK

You may wonder whether it is possible to have two orthogonal sinusoidal pulses that are more closely spaced than
1/T b . We can gain excellent insight into this issue by considering two unit-energy pulses g1 (t) and g0 (t) with a
frequency separation Δf
{√
2∕Tb cos(2𝜋f0 t), 0 ≤ t ≤ Tb
g0 (t) =
0, elsewhere
{√ [ ]
2∕Tb cos 2𝜋(f0 + Δf )t , 0 ≤ t ≤ Tb
g1 (t) =
0, elsewhere
 11.7 Binary Modulation 717

From Eq. (11.24), the correlation coefficient of the two pulses is given by
Tb
2
𝜌= cos(2𝜋f0 t) cos[2𝜋(f0 + Δf )t]dt
Tb ∫0
b T b T
1 1
= cos(2𝜋Δft)dt + cos[2𝜋(2f0 + Δf )t]dt
Tb 0∫ Tb 0∫
sin(2𝜋ΔfTb ) sin[2𝜋(2f0 + Δf )Tb ]
= +
2𝜋Tb Δf 2𝜋Tb (2f0 + Δf )
= sinc(2ΔfTb ) + sinc[2(2f0 + Δf )Tb ] (11.75)

Don’t worry much about the derivation of this equation but concentrate rather on the simplicity of its graphical
presentation in Figure 11.18. There are several important observations based on this graph.

● The correlation coefficient 𝜌 = 0 at integer multiples of bit rate Rb (= 1/T b ). That is, the two sinusoidal pulses
of duration T b are orthogonal when their frequencies differ by an integer multiple of Rb . This means that both
pulses complete an integer number of cycles in each bit interval T b , a finding that agrees with the result obtained
in Worked Example 11.2b.
● The correlation coefficient is in fact zero at all integer multiples of half the bit rate, e.g. 1 × 0.5Rb , 2 × 0.5Rb ,
3 × 0.5Rb , etc. We see that two sinusoidal pulses are orthogonal when they differ in frequency by only half the
bit rate. This is the smallest frequency separation at which two sinusoidal pulses can be orthogonal and means
that both pulses complete an integer number of half-cycles in the interval T b . Below this minimum frequency
spacing of Rb /2, the pulses become increasingly positively correlated. An FSK scheme that uses this minimum
frequency separation as well as having continuous phase is given the special name minimum shift keying (MSK).

0.75
Correlation coefficient, ρ

0.5

0.25
Telegram: @Persian_IEEEComSoc

–0.24
0 0.5Rb Rb 1.5Rb 2Rb 2.5Rb 3Rb
0.718Rb
Frequency separation, Δf

Figure 11.18 Correlation coefficient of two sinusoidal pulses of frequencies f 0 and f 0 + Δf , as a function of their frequency
separation Δf . Here f 0 = 3Rb .
 718 11 Digital Modulated Transmission

It follows from Eq. (11.74) that the bandwidth of MSK is

Bmsk = 2.5Rb (11.76)

● 𝜌 has a minimum value of −0.24 at Δf = 0.718Rb . In view of Eq. (11.61), this frequency separation gives FSK the
largest possible immunity to noise. To appreciate this, recall that PSK has the lowest BER for a given Eb , because
it uses symbols that have a correlation coefficient 𝜌 = −1. And it follows from Eq. (11.61) that the ‘effective’ Eb
is increased by 100%. In the case of FSK that employs two sinusoidal pulses separated in frequency by 0.718Rb ,
the effective Eb is increased by about 25% compared to that of orthogonal-symbol FSK. It is for this reason that
FSK modems operate with a frequency separation that lies in the first region of negative correlation, between
Rb /2 and Rb . A commonly used separation is two-thirds the bit rate.
● The specific value of minimum correlation coefficient 𝜌min and the frequency spacing Δf at which it occurs
depends on the integer multiple of Rb that f 0 assumes. Figure 11.18 is plotted with f 0 = 3Rb and the values of
𝜌min = −0.24 at Δf = 0.718Rb quoted above are for f 0 = 3Rb . However, 𝜌min will always occur at Δf between
Rb /2 and Rb . That is, the above observations hold for other values of f 0 with only a minor variation in the mini-
mum value of 𝜌. At one extreme when f 0 → ∞, the second term of Eq. (11.75) becomes negligible, and we have
𝜌min = −0.217 at Δf = 0.715Rb . At the other extreme, when f 0 = Rb , we have 𝜌min = −0.275 at Δf = 0.721Rb .

11.7.4 Minimum Transmission Bandwidth

It is important to bear in mind the assumptions involved in obtaining the bandwidths of ASK, PSK, FSK, and MSK
given by Eqs. (11.69), (11.72), (11.74), and (11.76):
● A completely random bit stream with equally likely 1’s and 0’s.
● A baseband waveform representation of the bit stream that is either rectangular or shaped using a realisable
raised-cosine filter with roll-off factor 𝛼 = 1.
These assumptions lead to a baseband waveform (which serves as the modulating signal) that has bandwidth
B = Rb , and hence the above results for Bask , Bpsk , and Bfsk .
The minimum transmission bandwidth requirement of these binary modulation systems may be specified using
two alternative arguments that lead to the same result. First of all, if the rectangular pulses of the baseband wave-
form are shaped using an ideal Nyquist filter then (as is discussed further in Chapter 12) the baseband waveform
will have a bandwidth B = Rb /2. Secondly, and for now, we can use a simpler argument in which we consider the
fastest-changing bit sequence 101010… of the message signal. The corresponding baseband waveform is a peri-
odic signal of period 2T b , and fundamental frequency f 0 = 1/2T b = Rb /2. Figure 11.19 shows how this sequence
is clearly identifiable from a waveform that contains only the fundamental frequency Rb /2. The minimum band-
width required to convey this sequence so that it is detectible at the destination is therefore one that is large
Telegram: @Persian_IEEEComSoc

1 0 1 0 1 0
1 R
Frequency f0 = = b
2Tb 2

Tb = 1/Rb

Figure 11.19 Bipolar waveform and fundamental frequency sinusoid for the most rapidly changing bit sequence.
 11.8 Coherent Binary Detection 719

enough to pass this fundamental frequency. All other sequences will change more slowly and hence have a lower
fundamental frequency. So, the minimum bandwidth of the modulating baseband waveform is Rb /2.
Correspondingly, the minimum transmission bandwidth of a binary modulated system is given by
Baskmin = Bpskmin = Rb
Bfskmin = (f1 − f0 ) + Rb = Δf + Rb
Bmskmin = 1.5Rb (11.77)

11.8 Coherent Binary Detection

Coherent demodulation is discussed at length in Chapter 7 as a technique for receiving analogue signals transmit-
ted by amplitude modulation of a sinusoidal carrier. The matched filter, for the detection of rectangular or shaped
pulses in digital baseband systems and sinusoidal pulses in digital modulated systems, is briefly introduced in
Chapter 1 and further discussed in Chapter 12. And in Section 11.4 and Figure 11.8c we briefly discuss the cor-
relation receiver, which consists of a product modulator followed by an integrator. All three processors, namely
coherent demodulator, correlation receiver, and matched filter, are in fact equivalent. Note, however, that coherent
demodulators used in analogue systems differ in a subtle way from those in digital systems. In the former, we want
the output signal to be a faithful reproduction of the original modulating (i.e. message) signal. In the latter, the
main aim is that the output has a form that gives the most reliable indication whether a matched pulse is present
at a given decision instant. It is then more appropriate to use the term coherent demodulation when referring to
analogue receivers and coherent detection in reference to digital receivers.
With this clarification in mind, we may now briefly discuss the coherent detection of ASK, PSK, and FSK signals.
We present the receiver as a matched filter of impulse response h(t) followed by a decision device. The output y(t)
of the matched filter is given by the convolution of the received input signal r(t) and h(t). See Section 3.6.2 for
a detailed discussion of the computation of this convolution. For simplicity and clarity, we will concentrate on a
graphical presentation of the output signal for various input signals. This will serve to clarify the criteria used by
the decision device to decide in favour of a binary 1 or 0 in each interval of duration T b .
Equation (11.61) gives the probability of error or BER in each of the coherent detectors discussed below. You
may wish to refer to the important discussion accompanying that equation. The BER may also be conveniently
read from the graph of Figure 11.13.

11.8.1 ASK Detector

Figure 11.20 shows the block diagram of a coherent ASK receiver. The filter is matched to cos(2𝜋f c t). The output
of the filter for each of the two symbols g1 (t) and g0 (t) transmitted for binary 1 and 0, respectively, is shown in
Telegram: @Persian_IEEEComSoc

Figure 11.21. It can be seen that the output y(t), taken at t = T b and denoted y(T b ), equals 2Eb for binary 1 and 0 for
binary 0, where Eb is related to the amplitude Ac of the sinusoidal pulse by Eq. (11.66). Figure 7.21 represents an
ideal situation of noiseless transmission in which y(T b ) is exactly 2Eb for binary 1 and is 0 for binary 0. In practice,

ASK signal Matched filter y(Tb) To decision Binary 1 if > Eb

r(t) h(t) y(t) device
Binary 0 if < Eb
Matched to sample at
cos(2πfct) t = Tb Eb = Ac2Tb/4

Figure 11.20 Coherent ASK receiver.

 720 11 Digital Modulated Transmission

Noise-free input Output

g1(t)
Ac 2Eb y1(t)

Binary 1 t t

–Ac

g0(t) y0(t)
Binary 0
0 t 0 t
0 0 Tb 2Tb

Figure 11.21 Matched filter output for noise-free inputs in ASK detector.

Noisy input Output

r1(t) y1(t)
2Eb

Eb

Binary 1 t 0 t

–Eb

–2Eb
r0(t)
y0(t)
Eb

Binary 0 t 0 t

–Eb
0 Tb 0 Tb 2Tb
Telegram: @Persian_IEEEComSoc

Figure 11.22 Matched filter output for noisy inputs in ASK detector.

the symbols are corrupted by noise, as shown in Figure 11.22. The inputs r 1 (t) and r 0 (t) in this diagram result from
adding Gaussian noise w(t) of variance 5Eb to g1 (t) and g0 (t). Note that the exact shape of the output y1 (t) and y0 (t)
will depend on the sequence of values in w(t), which is in general not identical for two noise functions – even if
both have the same variance. To demonstrate this, Figure 11.23 shows the outputs y1 (t) and y0 (t) in six different
observation intervals with the same noise variance (= 10Eb ).
It is interesting to observe the similarity between the output y1 (t) in Figures 11.21 and 11.22. The matched filter
effectively pulls the sinusoidal pulse g1 (t) (to which it is matched) out of noise. However, in the presence of noise,
y1 (T b ) ≠ 2Eb , and y0 (T b ) ≠ 0. The decision threshold is therefore set halfway between 2Eb and 0, and binary 1 is
chosen if y(T b ) > Eb , and binary 0 if y(T b ) < Eb . In the rare event of y(T b ) being exactly equal to Eb , a random choice
 11.8 Coherent Binary Detection 721

Observation Output for binary 0 Output for binary 1

↓
2Eb
Eb
1
–Eb
–2Eb
2Eb
Eb
2
–Eb
–2Eb
2Eb
Eb
3
–Eb
–2Eb
2Eb
Eb
4
–Eb
–2Eb
2Eb
Eb
5 Error
–Eb
–2Eb
2Eb
Eb
6
–Eb
–2Eb
0 Tb 2Tb 0 Tb 2Tb

Figure 11.23 Matched filter output in six different observation intervals with Gaussian noise of the same variance.

is made between 1 and 0. Therefore, in the fifth observation interval of Figure 11.23, binary 1 would be erroneously
detected as binary 0.

11.8.2 PSK Detector

A coherent PSK detector is shown in Figure 11.24a. The filter is matched to the pulse g1 (t) = Ac cos(2𝜋f c t) that
represents binary 1. Recall that binary 0 is represented using a pulse of opposite polarity, g0 (t) = −g1 (t). The output
of the filter with g1 (t) and g0 (t) as input is shown in Figure 11.24b. The output is sampled at t = T b . It can be seen
that y(T b ) = Eb for binary 1, and −Eb for binary 0, where Eb is related to the pulse amplitude Ac by Eq. (11.71).
The effect of noise is to shift y(T b ) from +Eb during the interval of binary 1, and from −Eb during binary 0. See
the discussion of Figures 11.22 and 11.23. The decision threshold is set halfway between +Eb and −Eb , so that the
decision device chooses binary 1 if y(T b ) > 0, and binary 0 if y(T b ) < 0. In the rare event of y(T b ) = 0, the receiver
makes a random guess of 1 or 0.
Telegram: @Persian_IEEEComSoc

11.8.3 FSK Detector

Figure 11.25a shows the block diagram of a coherent FSK detector. Two matched filters are employed. The upper
filter has an impulse response hU (t) which is matched to the sinusoidal pulse g1 (t) = cos(2𝜋f 1 t) that represents
binary 1, whereas the lower filter’s impulse response hL (t) is matched to the sinusoidal pulse g0 (t) = cos(2𝜋f 0 t)
that represents binary 0. Figure 11.25b shows the output of the two filters, denoted yU (t) and yL (t), respectively,
when the received input signal r(t) (under noise-free conditions) is either g1 (t) or g0 (t). The outputs of both filters
are sampled at t = T b and subtracted in a summing device to give a value y(T b ) that is fed into the decision device.
It can be seen that when the input is the binary 1 pulse g1 (t), then yU (T b ) = Eb and yL (T b ) = 0, so that y(T b ) = Eb .
On the other hand, when the input is the binary 0 pulse g0 (t), then yU (T b ) = 0 and yL (T b ) = Eb , giving y(T b ) = −Eb .
 722 11 Digital Modulated Transmission

PSK signal y(Tb) Binary 1 if y(Tb) > Eb

(a) Matched filter To decision
r(t) h(t) y(t) device
Binary 0 if y(Tb) < Eb
Matched to sample at
cos(2πfct) t = Tb

(b)
Input Output
y1(t)
g1(t)
Eb
Ac

Binary 1 t t

–Ac
g0(t) y0(t)
Ac

Binary 0 t t

–Ac –Eb
0 Tb 0 Tb 2Tb

Figure 11.24 Coherent PSK detector: (a) receiver; (b) input and output of matched filter in noise-free conditions.

With y(T b ) obtained in this manner, the situation in the decision device is like that of PSK detection. Therefore,
the decision threshold is set halfway between +Eb and −Eb , and the decision device chooses binary 1 if y(T b ) > 0,
and binary 0 if y(T b ) < 0, and a random guess of 1 or 0 if y(T b ) = 0.
The values of y(T b ) quoted above apply only when the symbols g1 (t) and g0 (t) are orthogonal. Specifically,
Figure 11.25b is based on the frequency values f 0 = 3Rb and f 1 = f 0 + Rb , so that Δf = f 1 −f 0 = Rb . If the frequency
spacing Δf = f 1 −f 0 has a value between 0.5Rb and Rb then, as discussed in Section 11.7.3.3, the two symbols are
negatively correlated. In this case, y(T b ) greater than +Eb for binary 1 and less than −Eb for binary 0. The decision
threshold is still at the halfway point at y(T b ) = 0 but there is increased immunity to noise due to the wider gap
between the nominal values of y(T b ) corresponding to binary 1 and binary 0. If, on the other hand, Δf is less than
0.5Rb , or between Rb and 1.5Rb , etc. then the two symbols are positively correlated and y(T b ) is less than +Eb for
Telegram: @Persian_IEEEComSoc

binary 1 and greater than −Eb for binary 0. The gap between the nominal outputs for binary 1 and binary 0 is
therefore narrower, and this increases the susceptibility of the system to noise.
To further demonstrate the importance of frequency spacing, we show in Figure 11.26 the difference signal
y(t) = yU (t) − yL (t) of the two matched filters for three different values of frequency spacing Δf = Rb , 0.718Rb
and 0.25Rb . With Δf = Rb , we have the orthogonal-pulses scenario plotted earlier in Figure 11.25, and you can
see that y(T b ) = ±Eb for binary 1 and binary 0, respectively, giving a gap of 2Eb between these nominal values of
the difference signal. The case Δf = 0.718Rb corresponds to the most negative correlation possible between the
two pulses, and we have an increased gap of 2.49Eb . Finally, the pulses separated in frequency by Δf = 0.25Rb
are positively correlated, giving a reduced gap of 0.68Eb . In the extreme case of Δf = 0, both matched filters are
identical, and so are the transmitted symbols g1 (t) and g0 (t). The difference signal y(t) is then always zero and
the decision device chooses between 1 and 0 each time by a random guess. This is the identical-symbol system
 11.9 Noncoherent Binary Detection 723

Matched filter yU(Tb)

hU(t) yU(t)
FSK signal Matched to sample at + To decision device Binary 1 if y(Tb) > 0
cos(2πf1t) t = Tb Σ
r(t) y(Tb)
– Binary 0 if y(Tb) < 0
Matched filter y (T
L b )
hL(t) yL(t)
Matched to sample at
cos(2πf0t) t = Tb
(a)

Input yU(t) Output

Eb ↑
g1(t)
Ac ↑ 0 →t

–Eb
Binary 1 0 →t yL(t)
Eb ↑

–Ac 0 →t

–Eb
yU(t)
Eb ↑
g0(t)
Ac ↑
0 →t

–Eb
Binary 0 0 →t yL(t)
Eb ↑

–Ac 0 →t

–Eb
0 Tb 0 Tb 2Tb
(b)

Figure 11.25 Coherent FSK: (a) block diagram of detector; (b) response of the matched filters to transmitted symbols g0 (t)
and g1 (t).
Telegram: @Persian_IEEEComSoc

discussed earlier in connection with Eq. (11.61). You can see that the probability of a wrong guess is 0.5, which
agrees with the BER obtained earlier for such a system.

11.9 Noncoherent Binary Detection

Coherent detection gives a high immunity to noise but poses two major challenges in its implementation.

● We must have a complete knowledge of the phase of the incoming pulses in order to have a perfectly matched
filter. Put another way, the locally generated pulse used in the correlation receiver must match the incoming
 724 11 Digital Modulated Transmission

y(t)
Eb Gap = 2Eb
Binary 1
Orthogonal
pulses: 0 t
Δf = Rb
Binary 0
–Eb
y(t)
Eb Gap = 2.49Eb
Negatively Binary 1
correlated
pulses: 0 t
Δf = 0.718Rb
Binary 0
–Eb
y(t)
Eb
Positively Binary 1
correlated
pulses: 0 t
Δf = 0.25Rb Binary 0
–Eb Gap = 0.68Eb
0 Tb 2Tb

Figure 11.26 Difference signal y(t) = y U (t) − y L (t) in a coherent FSK detector using orthogonal, negatively-correlated, and
positively-correlated pulses.

pulse exactly in phase. In practice, variations in the transmission medium will cause the incoming pulse to arrive
with a variable phase. This gives rise to a nonzero and variable phase difference between the incoming pulse
and a local pulse generated with fixed initial phase. This phase error will significantly degrade the detection
process and increase the probability of error. Figure 11.27 shows the effect of phase errors of 45∘ and 120∘ in a
coherent ASK detector. The filter is matched to the expected symbol for binary 0. Note that, in the former, the
output of the filter at t = T b is y(T b ) = 1.414Eb , rather than the value 2Eb that is obtained in the absence of phase
error. Since the decision threshold is at y(T b ) = Eb , it means that the noise margin has been lowered by 58.6%
because of this phase error. In the second case, the phase error leads to y(T b ) = −Eb which causes outright error
since this output falls below the decision threshold and the decision device would therefore decide in favour
of binary 0.
● We must sample the output of the matched filter at the correct instants t = T b . By examining, for example,
Figure 11.24b you can see that the filter output drops very rapidly away from the sampling instant t = T b . There is
therefore little tolerance for timing error, the effect of which is to reduce the noise margin or (beyond a small frac-
Telegram: @Persian_IEEEComSoc

tion) give rise to outright error. This problem is less crucial in baseband transmission systems where a matched
filter output decreases more slowly away from the sampling instant, as discussed further in Chapter 12.
We can minimise the first problem by extracting the desired pulse frequency from the incoming signal and using
this to provide the locally generated pulse. ASK and FSK contain impulses at the pulse frequency, which can
therefore be extracted using a bandpass filter (BPF) or, more accurately, a phase-locked loop. See Section 7.5.2 for
a discussion of this carrier extraction process.
A PSK signal does not, however, contain an impulse at the carrier frequency f c but this may be obtained by a
full-wave rectification of the PSK signal to create a component at 2f c , extracting this component using a BPF and
dividing by 2 to obtain the desired pulse frequency. This process is elaborated in Figure 11.28. There is a phase
uncertainty of 180∘ in the generated carrier, depending on the phase of the division. This makes it impossible to
 11.9 Noncoherent Binary Detection 725

y(t)
2Eb↑
Phase error = 45° y(Tb) = 1.414Eb
Eb

(a) 0

–Eb

–2Eb →t

2Eb ↑y(t)
Phase error = 120°
Eb

(b) 0

–Eb
y(Tb) = –Eb
–2Eb →t
0 Tb 2Tb

Figure 11.27 Effect of phase error in coherent ASK. The filter is matched to Ac cos(2𝜋f c t). Graphs show output, i.e.
difference signal y(t), when the input is (a) Ac cos(2𝜋f c t + 45∘ ); (b) Ac cos(2𝜋f c t + 120∘ ).

or

Period = 1/2fc;
1/fc Fundamental Frequency = 2fc Frequency = fc
frequency = 2fc
Full-wave BPF ÷2 Carrier
PSK rectifier (2fc) signal

Figure 11.28 Generating a carrier signal from an incoming PSK signal.

be certain that the receiver is matched to g1 (t) – the symbol for binary 1, and not to g0 (t) – the symbol for binary
0, which would cause the output bit stream to be an inverted copy of the transmitted bit stream. To resolve this
phase ambiguity of the locally generated carrier, a known training sequence of bits, or preamble, is first transmitted
to the receiver. By comparing the receiver output to the expected output, the carrier phase is correctly set.
The arrangement discussed above achieves phase synchronisation at the cost of increased complexity of the
Telegram: @Persian_IEEEComSoc

coherent binary detector. The effect of phase errors can be completely eradicated if we ignore phase information
in the incoming binary modulated signal. The receiver is then described as a noncoherent detector. Obviously,
this method is not applicable to PSK since it is the phase that conveys information. But the need for generating a
phase-synchronised carrier at the receiver can also be eliminated in PSK by using a variant technique known as
differential phase shift keying (DPSK). These noncoherent receivers are briefly discussed in the following sections.

11.9.1 Noncoherent ASK Detector

Figure 11.29 shows the outputs of a matched filter in a coherent ASK receiver with inputs of phase error 0∘ , 90∘ ,
and 180∘ . Note that all three outputs have a common envelope, which is shown in dotted lines. It is true in general
 726 11 Digital Modulated Transmission

y(t)
2Eb ↑
0° phase error
180° phase error
Common envelope 90° phase error

Eb

–Eb

–2Eb →t
0 Tb 2Tb

Figure 11.29 Output of a matched filter in a coherent ASK receiver for inputs with various phase errors.

Binary 1 if y(Tb) > Eb

ASK signal Matched filter Envelope y(Tb) To decision
r(t) h(t) detector y(t) device
Binary 0 if y(Tb) < Eb
Matched to sample at
Ac cos(2πfct) t = Tb Eb = Ac2Tb/4

Figure 11.30 Noncoherent ASK detector.

that the envelope of the output of a matched filter is independent of the phase of the input signal. Therefore, we
may follow a matched filter with an envelope detector. When the output of the envelope detector is sampled at
t = T b , it will have a value y(T b ) = 2Eb for binary 1 irrespective of the phase error in the incoming sinusoidal pulse
g1 (t). For binary 0 the sample value will of course be y(T b ) = 0. For a discussion of the operation and design of
envelope detectors (also known as diode demodulators), see Section 7.5.1.
The receiver just described is a noncoherent ASK detector and is shown in Figure 11.30. Note that the matched
filter is just a (special response) BPF centred at f c and is frequently indicated as such in some literature. Nonco-
herent detection has two main advantages over coherent detection.
Telegram: @Persian_IEEEComSoc

● The output is independent of phase error, making it unnecessary to provide phase synchronisation at the
receiver. This greatly simplifies receiver design.
● The envelope does not decrease rapidly away from the sampling instant T b . Therefore, the system is much more
tolerant of timing errors than is a coherent detector.
The main drawback of a noncoherent receiver is that it is more susceptible to noise. By ignoring phase information,
the receiver inadvertently admits contributions from noise of all phases. A coherent receiver, on the other hand,
is not affected by noise components that are 90∘ out of phase with the incoming signal to which the receiver is
matched. Eqs. (11.56) and (11.61) therefore do not apply to noncoherent reception. We will not derive this here,
but assuming that (i) Eb > > N o and (ii) the bandwidth of the BPF (shown as a matched filter in Figure 11.30) is
the minimum required to pass the ASK signal – see Eq. (11.77) – then the probability of error of a noncoherent
 11.9 Noncoherent Binary Detection 727

ASK receiver is given by the expression

(√ ) ( )
1 Eb 1 E
BER = erfc + exp − b (11.78)
4 2No 2 2No
where Eb is the average energy per bit given by Eq. (11.66) and N o is the noise power per unit bandwidth. Note
that the BER expression in Eq. (11.78) is dominated by the second term on the right-hand side.

11.9.2 Noncoherent FSK Detector

From the foregoing discussion, a noncoherent FSK detector is obtained by inserting an envelope detector after each
of the matched filters in a coherent FSK detector. Figure 11.31 is a block diagram of the resulting noncoherent FSK
detector. In this diagram, it is helpful to view the upper matched filter as a BPF that passes the pulse of frequency
f 1 while completely rejecting the other pulse of frequency f 0 . The lower matched filter similarly serves as a BPF
that passes f 0 and rejects f 1 . To help you understand the operation of this receiver, the waveforms at various points
of the circuit are plotted in Figure 11.32. The upper half of Figure 11.32 gives the outputs when the incoming pulse
is g1 (t) = Ac cos(2𝜋f 1 t + 𝜑e ), which represents a binary 1, whereas the lower half of the figure gives outputs for
input g0 (t) = Ac cos(2𝜋f 0 t + 𝜑e ), which represents binary 0. The parameter 𝜑e is a phase error, the precise value of
which does not change the output at the decision instant t = T b and only marginally affects the output at other
time instants between 0 and 2T b . Figure 11.32 was produced with 𝜑e set to 90∘ .
The difference signal is the input to the decision device and, regardless of phase error, is y(T b ) = Eb for binary 1
and y(T b ) = −Eb for binary 0. Therefore, the decision threshold is set halfway at y(T b ) = 0. Figure 11.32 assumes
that g1 (t) and g0 (t) are orthogonal pulses. For correlated pulses, the value of y(T b ) changes in the manner discussed
in Section 11.8.3. On the same assumptions as in the previous section, we find that the BER of a noncoherent FSK
receiver is given by
( )
1 Eb
BER = exp − (11.79)
2 2No
In this case, the energy per bit Eb is calculated from the received pulse amplitude Ac and pulse duration T b using
Eq. (11.71). By comparing Eqs. (11.78) and (11.79), you can see that, for the same average Eb /N o ratio, noncoherent
FSK has a slightly better noise performance (in terms of lower BER) than noncoherent ASK. However, the ASK
system would have double the peak energy of FSK since it is only on for half the time (during binary 1) and off
during every interval of binary 0.

11.9.3 DPSK
We have noted that it is impossible to distinguish the two pulses used in PSK by observing the envelope of an incom-
ing PSK signal. Therefore, strictly speaking, incoherent PSK detection does not exist. However, we can obviate the
Telegram: @Persian_IEEEComSoc

Matched filter Envelope yU(Tb)

hU(t) detector yU(t) To decision
FSK signal Matched to sample at + device Binary 1 if y(Tb) > 0
Ac cos(2πf1t) t = Tb Σ
r(t) y(Tb) Binary 0 if y(Tb) < 0
–
Matched filter Envelope yL(Tb)
hL(t) detector yL(t)
Matched to sample at
Ac cos(2πf0t) t = Tb

Figure 11.31 Noncoherent FSK detector.

 728 11 Digital Modulated Transmission

Upper (U) and lower (L) branches Difference signal

Eb↑ yU(t) Eb ↑ y(t) = yU(t) – yL(t)

Eb/2
Binary 1

0
y (t) Eb/2
Eb↑ L

Eb/2

0 0 →t
y (t) y(t) = yU(t) – yL(t)
Eb↑ U 0↑ →t

Eb/2
Binary 0

0
y (t) –Eb/2
Eb↑ L

Eb/2

0 –Eb
0 Tb 2Tb 0 Tb 2Tb

Figure 11.32 Output waveforms in a noncoherent FSK detector.

need for phase synchronisation at the receiver if information (i.e. binary 1 and 0) is coded at the transmitter as
changes in phase, rather than as absolute phase values. This means that we keep the sinusoidal pulse amplitude
and frequency constant at Ac and f c , respectively, but we transmit the pulse with its phase incremented by, say, 180∘
to represent binary 0, and the phase left unchanged to represent binary 1. This technique is known as differential
phase shift keying (DPSK). We may write
{
g1 (t), Binary 1
vdpsk (t) =
g0 (t), Binary 0
{
Ac cos(2𝜋fc t + 𝜙n−1 ), 0 ≤ t ≤ Tb
g1 (t) =
0, elsewhere
{
Ac cos(2𝜋fc t + 𝜙n−1 + 𝜋), 0 ≤ t ≤ Tb
Telegram: @Persian_IEEEComSoc

g0 (t) = (11.80)
0, elsewhere

In the above, f c is an integer multiple of bit rate, and 𝜑n-1 is the phase of the sinusoidal pulse transmitted during
the previous bit interval, which will always be either 0 or 𝜋 radians. Note that 𝜑n-1 = 0 in the first bit interval.
Figure 11.33a shows the block diagram of a DPSK modulator. Comparing this with the block diagram of a PSK
modulator in Figure 11.16a, we see that the only difference is in the type of baseband coder that precedes the
product modulator. A bipolar NRZ-S coder is used in a DPSK modulator, whereas a PSK modulator uses a bipolar
NRZ coder. These line codes are discussed in Section 10.7.1. For a random bit stream consisting of equally likely
1’s and 0’s, DPSK and PSK will therefore have identical power spectra, and hence bandwidth.
The bipolar NRZ-S coder is implemented using an inverted exclusive-OR (XNOR) gate and a one-bit delay device
(e.g. a clocked flip-flop) connected as shown in Figure 11.33b. Note the accompanying truth table, where bn denotes
 11.9 Noncoherent Binary Detection 729

Input bit stream Bipolar ±1 DPSK

(a)
bn NRZ-S coder cn
× ʋdpsk(t) signal

Ac cos(2πfct)

bn
XNOR cn bn cn–1 cn
(b) Gate 0 0 1
cn–1 0 1 0
Delay 1 0 0
Tb 1 1 1

Current pulse x(t) yn Output

DPSK Matched
BPF × LPF bit stream
(BNRZ)
(c) Previous
Delay pulse
Tb

Figure 11.33 DPSK: (a) modulator; (b) NRZ-S coder and truth table of XNOR gate; (c) detector.

current input, cn denotes current output, and cn−1 denotes previous output. What is entered as binary 1 in the truth
table is represented electrically in the circuit as +1 V (normalised) and binary 0 as −1 V. The output of the XNOR
gate is the bipolar NRZ-S waveform cn , which is shown in Figure 11.34 for a selected input bit stream bn . The
easiest way to remember the coding strategy of NRZ-S is that the waveform makes a transition between ±1 at the
beginning of the bit interval if the bit is binary 0, and makes no transition if the bit is binary 1. Note that the cn
waveform shown corresponds to an initially high (+1) state of the XNOR gate output.
When this NRZ-S waveform is multiplied by a carrier Ac cos(2𝜋f c t) in a product modulator, the result is the
DPSK waveform given by Eq. (11.80) and also shown in Figure 11.34.
Detection is performed at the receiver by comparing the phase of the current pulse to that of the previous pulse. A
significant phase difference indicates that the current interval is binary 0, whereas a negligible difference indicates
binary 1. It is assumed that phase variations due to the transmission medium are negligible over the short period of
one bit interval T b . That is, the only significant change in phase from one bit interval to the next is due entirely to
the action of the DPSK modulator. To make this phase comparison, the current and previous pulses are multiplied
Telegram: @Persian_IEEEComSoc

together, which yields the signal x(t) shown in Figure 11.34. Note that in the very first bit interval the previous
pulse used has zero-phase. It is easy to see that when the two pulses are antipodal then their product is negative,
whereas when the pulses have the same phase then the product is positive. Passing x(t) through an LPF gives an
output voltage yn , which is positive for binary 1 and negative for binary 0, as shown in Figure 11.34. You will no
doubt recognise yn as a bipolar NRZ representation of the transmitted bit stream bn . Thus, the DPSK signal has
been successfully demodulated.
A block diagram of a DPSK detector that operates as described above is shown in Figure 11.33c. For optimum
detection, the BPF is matched to the sinusoidal pulse Ac cos(2𝜋f c t) of duration T b . The BER of this optimum DPSK
detector is given by the expression
( )
1 E
BER = exp − b (11.81)
2 No
 730 11 Digital Modulated Transmission

Transmitted bitstream (bn)

cn 1 1 0 1 0 0 0
+1
NRZ-S
Transmitter

waveform
–1
ʋdpsk(t)
+Ac
DPSK
waveform
–Ac
x(t)
Receiver

Recovered bitstream
yn 1 1 0 1 0 0 0
+1
BNRZ
waveform
–1
0 Tb 2Tb 3Tb 4Tb 5Tb 6Tb 7Tb

Figure 11.34 Waveforms in DPSK modulator and detector.

You can see that DPSK – our ‘noncoherent PSK’ – has a better noise performance than either noncoherent ASK or
noncoherent FSK but is inferior in this respect to coherent PSK. However, the main advantage of DPSK compared
to PSK is the simplicity of its receiver circuit, which does not require phase synchronisation with the transmitter.

11.10 M-ary Transmission

It is only in a limited number of situations, e.g. optical fibre communication which employs (binary) ASK, that the
available channel bandwidth is enough to allow the use of binary modulated transmission to achieve the required
bit rate. In communication systems involving radio and copper-wired transmission media, bandwidth efficiency is
an important design consideration which makes M-ary transmission (M > 2) necessary.

11.10.1 Bandwidth Efficiency

Telegram: @Persian_IEEEComSoc

Bandwidth efficiency 𝜂 is defined as the ratio Rb /Bocc between message bit rate Rb and transmission (or occupied)
bandwidth Bocc and is expressed in bits per second per hertz (b/s/Hz). In Section 6.5.2, we discuss at length the
bandwidth efficiency of M-ary transmission systems that employ raised cosine filtering and plotted 𝜂 versus M
in Figure 6.19. To summarise these results, M-ary ASK, PSK, APSK, and FSK systems use M unique symbols to
convey log2 M bits per symbol so that bit rate Rb and symbol rate Rs are related by

Rb = Rs log2 M (11.82)

All digital transmission systems employ a raised cosine filter of roll-off factor 𝛼 (0 ≤ 𝛼 ≤ 1), introduced in Section
4.7.3.5 (Worked Example 4.16) and discussed further in Chapter 12. In M-ary ASK, PSK, APSK systems, each
transmitted symbol is a sinusoidal pulse of the same (carrier) frequency f c and duration T s = 1/Rs , leading to
 11.10 M-ary Transmission 731

Bb = (1 + α)Rs/2
Bocc
Δf = Rs/2

f
f0 – Bb f0 f0 + Bb f1 – Bb f1 f1 + Bb fM–1 – Bb fM–1
fM–1 + Bb

Figure 11.35 Spectrum of M-ary FSK signal employing orthogonal sinusoidal pulses of duration T s = 1/Rs and frequencies
f 0 , f 1 , f 2 , …, f M−1 having minimum spacing Δf = Rs /2.

occupied bandwidth

Bocc = Rs (1 + 𝛼) M-ary ASK, PSK and APSK (11.83)

In the case of M-ary FSK, the M symbols are orthogonal sinusoidal pulses of duration T s = 1/Rs at respective
frequencies f 0 , f 1 , f2 , …, f M-1 . The amplitude spectrum of an M-ary FSK signal is therefore as shown in Figure 11.35.
Using the minimum frequency spacing Δf = Rs /2 required for the pulses to be mutually orthogonal, occupied
bandwidth is therefore

Bocc = (M − 1)Δf + 2Bb

( )
M+1
= Rs +𝛼 , M-ary FSK (11.84)
2
Most digital transmission systems also incorporate error control coding in which redundant bits are systematically
inserted to aid error detection and/or correction at the receiver. Thus, only a fraction r (0 < r ≤ 1) of the transmitted
bits are message bits, r being referred to as the code rate. Error control coding reduces the message bit rate Rb from
the value given by Eq. (11.82) to

Rb = rRs log2 M (11.85)

so that, in view of Eq. (11.83) and (11.84), the bandwidth efficiency of modulated M-ary transmission is given by
Rb rR log M
𝜂= = s 2
Bocc Bocc

⎧ rlog2 M , ASK, PSK, APSK

⎪ 1+𝛼
=⎨ (11.86)
rlog2 M
⎪ , FSK
⎩ 𝛼 + (M + 1)∕2
Telegram: @Persian_IEEEComSoc

where, 𝛼 ≡ roll-off factor of raised cosine filter. 𝛼 = 0 for an ideal Nyquist channel or brickwall filter, which is not
realisable in real-time, so 𝛼 typically exceeds 0.05. r ≡ code rate. r = 1 for a system operating without error control
coding.
Maximum bandwidth efficiency 𝜂 max is obtained when 𝛼 = 0 and r = 1, leading to

⎧log M, ASK, PSK, APSK

⎪ 2
𝜂max = ⎨ 2log M (11.87)
2
⎪ , FSK
⎩ M+1
It is this maximum bandwidth efficiency that is plotted in Figure 6.19 from which we see that beyond M = 4
the maximum bandwidth efficiency of M-ary FSK decreases steadily with M and is, for example, 𝜂 max = 0.8 at
 732 11 Digital Modulated Transmission

M = 4 and 𝜂 max = 0.02 at M = 1024. On the other hand, the bandwidth efficiency of M-ary ASK, PSK, and APSK
increases steadily with M from 𝜂 max = 1 at M = 2, reaching 𝜂 max = 10 at M = 1024. ASK, PSK, and APSK systems
therefore have a significantly superior spectral efficiency compared to FSK. However, as we will see shortly, what
FSK lacks in spectral efficiency it makes up for in noise immunity. M-ary FSK is therefore the preferred modulation
technique in applications, such as deep space communication, that involve very weak received signals, making
noise immunity a prime design consideration.

11.10.2 M-ary ASK

The general principles of M-ary modulation and detection are presented in Section 11.4. We now apply those ideas
to M-ary ASK.

11.10.2.1 M-ary ASK Modulator

Figure 11.36a shows a block diagram of an M-ary ASK modulator. A serial-to-parallel converter, implemented
using a shift register, converts the serial input bit stream to a parallel block Bin consisting of log2 M bits. A Gray
code converter (GCC) then maps each of the M distinct states Bin to a unique state Bout with the same number
of bits. The binary sequences Bin and Bout have respective decimal values k (in column 3 of Table 11.1) and i (in
column 4).
Table 11.1 also shows the mapping rule for M = 16. Column 2 is the 4-bit output of the GCC and is the binary
equivalent of the decimal numbers or state index i from 0 to 15 (column 4). Notice the pattern in this binary
number sequence. Going down column 2 and considering the first (i.e. rightmost) bit (or LSB) position only, we
see that it consists of the two-element repeating pattern 01, which we will call the fundamental pattern. Similarly,
the second bit position consists of the 4-element repeating pattern 0011, obtained by repeating each element of
the fundamental pattern. The third bit position consists of the pattern 00001111 obtained by using each element
of the fundamental pattern four times. In this way, the kth bit position of a binary number sequence (representing
the decimal numbers 0, 1, 2, 3, …) is generated from a pattern obtained by using each element of the fundamental
pattern 2k−1 times.
There is also a pattern in the Gray code Bin of column 1. The fundamental pattern in this case is, however,
0110. Thus, the first bit position of the Gray code consists of the sequence 0110… From the above discussion, we
therefore expect the second bit position to consist of the sequence 00111100…, the third bit position to consist
of the sequence 0000111111110000…, and so on. The importance of Gray coding lies in the fact that adjacent
codewords (Bin in column 1) differ in only one bit position. That is, the ith codeword differs from the (i−1)th and
(i + 1)th codewords by only one bit, for i = 1, 2, 3, …, M − 2, where Bin is the Gray code of the decimal number
or state index i. For example, Table 11.1 shows that the Gray code for i = 2 is 0011, rather than the usual binary

Input bit Bin Bout

Telegram: @Persian_IEEEComSoc

Serial to log2M-bit
stream Gray code M-ary ASK
(a) parallel to M-level ×
converter cn signal
converter converter
cos(2πfct)

M-aryASK Output
signal y(Ts) Parallel bit stream
Matched filter Gray
(b) ADC to serial
h(t) y(t) coder
converter
Matched to sample at
A cos(2πf t) t = Ts
c c

Figure 11.36 M-ary ASK: (a) modulator; (b) detector.

 11.10 M-ary Transmission 733

Table 11.1 Step-by-step mapping of a 4-bit input sequence Bin into a transmitted symbol gk (t) in 16-ASK modulator.

Decimal values
Bin Normalised Transmitted
(Gray code) Bout Bin ⇒ k Bout ⇒ i DAC output, cn symbol, gk (t)

0000 0000 0 0 0 g0 (t)

0001 0001 1 1 1/15 g1 (t)
0011 0010 3 2 2/15 g3 (t)
0010 0011 2 3 3/15 g2 (t)
0110 0100 6 4 4/15 g6 (t)
0111 0101 7 5 5/15 g7 (t)
0101 0110 5 6 6/15 g5 (t)
0100 0111 4 7 7/15 g4 (t)
1100 1000 12 8 8/15 g12 (t)
1101 1001 13 9 9/15 g13 (t)
1111 1010 15 10 10/15 g15 (t)
1110 1011 14 11 11/15 g14 (t)
1010 1100 10 12 12/15 g10 (t)
1011 1101 11 13 13/15 g11 (t)
1001 1110 9 14 14/15 g9 (t)
1000 1111 8 15 1 g8 (t)

equivalent 0010. The latter requires a change in two bit positions from the adjacent code word 0001, which violates
the Gray code rule.
The GCC in Figure 11.36a is a logic circuit that maps column 1 in Table 11.1 to column 2. So, for example, the
4-bit sequence 1010 is converted to 1100. The output of the converter is then processed in a log2 M-bit to M-level
converter to yield the normalised output shown in column 5 of Table 11.1. This output consists of M distinct and
uniformly spaced levels from 0 to 1 and is multiplied by a sinusoidal carrier of frequency f c to produce the desired
M-ary ASK signal. Make sure you can see that the overall result of the whole process is that adjacent states i in the
signal-space diagram of this M-ary ASK represent a block of log2 M input bits that differ in only one bit position.
This means that Eq. (11.54) applies in relating BER and SER. Refer to Figure 11.7a for the signal-space diagram of
4-ASK.
Numbering the states with index i, starting from i = 0 for the state at the origin to i = M − 1 for the state furthest
Telegram: @Persian_IEEEComSoc

from the origin, the M-ary ASK signal consists of one of the following M sinusoidal pulses in each symbol interval
of duration T s = T b log2 M
( )
iAc t − Ts ∕2
gi (t) ≡ gk (t) =
′
cos(2𝜋fc t)rect
M−1 Ts
i = 0, 1, 2, · · · , M − 1; k = GCC[i] (11.88)
where GCC[] denotes Gray code conversion and the rectangular function rect() serves the simple purpose of con-
straining the sinusoidal function to a pulse in the interval (0, T s ). For example, from Table 11.1, GCC[10] = 15,
so (in this example where M = 16) the symbol g15 (t) has amplitude 10Ac /15 and corresponds to state number 10
(counting from the origin). Notice that the pulses differ only in amplitude, which is determined by the output of
the M-level converter in each log2 M-bit interval.
 734 11 Digital Modulated Transmission

11.10.2.2 M-ary ASK Detector

An M-ary ASK detector operates on the principle shown in block diagram form in Figure 11.36b. The incoming
signal is passed through a BPF that is matched to the sinusoidal pulse cos(2𝜋f c t) of duration T s . The output of the
filter is sampled at t = T s and fed into an analogue-to-digital conversion (ADC), which quantises the sample y(T s )
to the nearest of M levels and provides a binary number representation of the level at its output. The M levels are
Esmax 2Esmax 3Esmax
, , , · · · , Esmax
0,
M−1 M−1 M−1
where Esmax is the maximum pulse energy given by
A2c Ts
Esmax = (11.89)
2
Note that the output of the ADC is given by column 2 in Table 11.1. To recover the original bit stream, the ADC
output is fed into a Gray coder, which maps from column 2 to column 1.

11.10.2.3 BER of M-ary ASK

The main handicap of M-ary ASK is its very poor BER caused by the fact that it uses strongly correlated sinusoidal
pulses. You will have the opportunity in Question 11.16 to show that the correlation coefficient of adjacent states
with indices i and i + 1 in Eq. (11.88) is given by
2i(i + 1)
𝜌i,i+1 = ; i = 0, 1, 2, · · · , M − 2 (11.90)
2i(i + 1) + 1
This result is shown in Figure 11.37 for M = 8. It shows that only the zeroth state is orthogonal to the other states.
The rest of the pairs of adjacent states are strongly correlated, with 𝜌 increasing rapidly towards unity as i increases.
For example, the correlation coefficient between the sixth and seventh states (i.e. i = 6 in Eq. (11.90)) is 0.99. In
view of Eq. (11.61), the consequence of such high-correlation coefficients is a high BER, which we now determine.
Figure 11.38 shows the signal-space diagram of an M-ary ASK with message states Si corresponding to the trans-
mitted symbols gi′ (t) given by Eq. (11.88). Consider the probability of error Pesi in state Si . In Worked Example 11.3

0.9

0.8
Correlation coefficient, ρi,i + 1

0.7

0.6

0.5
Telegram: @Persian_IEEEComSoc

0.4
0.3

0.2
0.1

0
0 1 2 3 4 5 6
State index, i

Figure 11.37 Correlation coefficient of adjacent states (i and i + 1) in 8-ASK.

 11.10 M-ary Transmission 735

S0 S1 S2 ……. Si – 1 Si Si + 1 ……. SM – 1
α0

Figure 11.38 Signal space diagram of M-ary ASK.

we justify the fact that Pesi is dominated by the event of Si being mistaken either for Si−1 or for Si+1 . Denoting the
probability that Si is mistaken for Si−1 as Pesi− , and the probability that it is mistaken for Si+1 as Pesi+ , we may write

Pesi = Pesi− + Pesi+ (11.91)

Pesi+ follows from Eq. (11.61) with Eb replaced by Esi+ – the average energy of Si and Si+1 , and 𝜌 given by Eq. (11.90).
Using the pulse amplitudes given by Eq. (11.88) and the expression for pulse energy given by Eq. (11.3) we obtain
[ ]
1 A2c i2 Ts A2c (i + 1)2 Ts
Esi+ = +
2 (M − 1)2 2 (M − 1)2 2
2
Ac Ts
= [2i(i + 1) + 1] (11.92)
4(M − 1)2

Furthermore
2i(i + 1)
1−𝜌=1−
2i(i + 1) + 1
1
=
2i(i + 1) + 1

So that Eq. (11.61) yields

√
⎛ E (1 − 𝜌) ⎞
1 ⎜ si+ ⎟
Pesi+ = erfc
2 ⎜ 2No ⎟
⎝ ⎠
√
⎛ A2c Ts ⎞
1
= erfc ⎜ ⎟≡P (11.93)
2 ⎜ 8(M − 1)2 No ⎟ esa
⎝ ⎠

We see that the result is independent of i, and therefore is the probability (denoted Pesa ) of mistaking any sym-
bol for an adjacent symbol. It follows from Eq. (11.91) that the probability of error in Si is Pesi = 2Pesa , and this
Telegram: @Persian_IEEEComSoc

applies for all i = 1, 2, 3, …, M − 2. Symbols S0 and SM-1 , however, have only one immediate neighbour, and hence
Pes0 = PesM−1 = Pesa . The desired probability of symbol error Pes in the M-ary ASK detector is obtained by averaging
these errors over all symbols. Thus
[ ]
1 ∑
M−2
Pes = Pesa + 2Pesa + Pesa
M i=1
2(M − 1)
= Pesa
M
√
⎛ A2c Ts ⎞
(M − 1)
= erfc ⎜ ⎟ (11.94)
M ⎜ 8(M − 1)2 No ⎟
⎝ ⎠
 736 11 Digital Modulated Transmission

It is more useful to express this equation in terms of the average energy per bit Eb in the M-ary system. Now the
average energy per symbol is
M−1 [ ]
1 ∑ A2c i2 Ts
Es =
M i=0 (M − 1)2 2
A2c Ts ∑
M−1
= 2
i2
2M(M − 1) i=0
A2c Ts (M − 1)M(2M − 1)
=
2M(M − 1)2 6
A2c Ts (2M − 1)
=
12(M − 1)
We obtained the third line by using the standard expression for the sum of squares
1
12 + 22 + 32 + · · · + n2 =n(n + 1)(2n + 1) (11.95)
6
Since each symbol represents log2 M bits, the average energy per bit is therefore
Es
Eb =
log2 M
A2c Ts (2M − 1)
= (11.96)
12(M − 1)log2 M
We can now use this relationship to eliminate A2c Ts from Eq. (11.94). Thus
√
⎛ 3Eb log2 M ⎞
(M − 1)
Pes = erfc ⎜ ⎟
M ⎜ 2No (2M − 1)(M − 1) ⎟
⎝ ⎠
Finally, with Gray coding, Eq. (11.54) applies, and we obtain the BER of M-ary ASK as
√
⎛ 3Eb log2 M ⎞
(M − 1)
BER = erfc ⎜ ⎟ (11.97)
Mlog2 M ⎜ 2No (2M − 1)(M − 1) ⎟
⎝ ⎠
This is a remarkable equation. It gives the BER of M-ary ASK explicitly in terms of M and our now familiar Eb /N o .
Note that when M = 2 this equation reduces nicely to Eq. (11.56) for (binary) ASK, as expected. Figure 11.39 shows
a plot of BER against Eb /N o for various values of M. We see that the BER increases rapidly with M. For example,
at Eb /N o = 14 dB, the BER is 2.7 × 10−7 for M = 2 but increases dramatically to 2.8 × 10−3 for M = 4. To put it in
another way, we need to increase transmitted signal power very significantly in order to obtain the same BER
in multilevel ASK as in binary ASK. For example, a BER of 1 × 10−7 is achieved in binary ASK and 32-ASK at
Telegram: @Persian_IEEEComSoc

respective Eb /N o values of 14.3 and 35.2 dB. Using Eq. (11.63) and the fact that the bit rate of 32-ASK is five times
that of binary ASK when using the same bandwidth, it follows that for both systems to operate at the same BER
of 1 × 10−7 the transmitted signal power in a 32-ASK system must be raised above that of binary ASK by
ΔP = 35.2 − 14.3 + 10log10 (5)
= 27.89 dB
This represents a very large signal power increase by a factor of 615. This finding is in line with Shannon’s informa-
tion capacity theorem (Chapter 12), which stipulates that signal power must be increased if one wishes to transmit
at a higher bit rate using the same bandwidth and maintaining the same level of acceptably low BER. Therefore,
despite the excellent bandwidth efficiency of M-ary ASK, its extremely poor power efficiency makes multilevel
ASK unsuitable for most applications. Binary ASK is, however, widely used, especially on optical fibre links. Note
 11.10 M-ary Transmission 737

0.1

256
10–2
128
64
32
10–3

16
Bit error ratio (BER)

10–4
8
10–5
4
10–6
M=2
10–7

10–8

10–9
0 5 10 15 20 25 30 35 40
Eb/No (dB)

Figure 11.39 BER of coherent M-ary ASK, for M = 2, 4, …, 256.

that Eq. (11.97) is the optimum noise performance attainable with a coherent detector. A poorer noise performance
will be obtained if an envelope (noncoherent) detector is used at the receiver.

11.10.3 M-ary PSK

An M-ary PSK signal is made up of one of the following M sinusoidal pulses in each symbol interval of duration
T s . The pulses have the same amplitude and frequency, differing only in phase.
( )
t − Ts ∕2
gi (t) ≡ gk (t) = Ac cos(2𝜋fc t + 2𝜋i∕M + 𝜃0 )rect
′
Ts
i = 0, 1, 2, · · · , M − 1; fc = n∕Ts ; k = GCC[i] (11.98)
The constant 𝜃 0 is an angular offset, which specifies the phase of state i = 0 but has no effect on the BER per-
formance of the PSK system, so it can be set to any value, including zero. The signal-space diagram of M-ary PSK
consists of message points uniformly spaced at an angular spacing 2𝜋/M along a circle that is centred at the origin.
Telegram: @Persian_IEEEComSoc

An example is shown in Figure 11.40 for 8-PSK. Because M-ary PSK has a circular constellation, the numbering
or index i of the states may start at any angle and may proceed either clockwise or counterclockwise. Furthermore
(and this applies to ASK and APSK as well), the bits of each state may be written left to right (i.e. MSB first) or
right to left (i.e. MSB last). In Figure 11.40, 𝜃 0 = 45∘ , numbering starts at angle 𝜃 0 and proceeds counterclockwise,
and the bits are written MSB first. For example, the state at angle −135∘ is state number i = 4. Eq. (11.98) gives the
phase of this state as 2𝜋 × 4/8 + 𝜋/4 rad = 225∘ (≡ −135∘ ). Converting i = 4 to Gray code k (using Table 11.1) yields
k = 6, so this state must represent the bits 110, as shown in Figure 11.40.
A special feature of the Gray code arrangement of states in M-ary PSK constellation is that, starting at the
all-binary-zero state (e.g. 000 in 8-PSK), the Gray code sequence progresses in one direction through one semicircle
until all states with MSB = 0 are exhausted, and then the sequence is repeated in the other semicircle by starting
at the state adjacent to the all-zero-state and flipping only the MSB and progressing in the opposite direction. The
 738 11 Digital Modulated Transmission

Q (≡ α1)
001

011 000

45°
010
I (≡ α0)
100

110 101

111

Figure 11.40 Signal space diagram of 8-PSK.

terminal states of both semicircles are adjacent to each other in the circular constellation and are guaranteed to
differ in only one bit position, provided the number of states M is an integer power of 2 (as is usually the case). In
Figure 11.40, for example, the Gray code sequence starts at 000 going counterclockwise for one-half of the circle
and then starts at 100 going clockwise for the remaining half of the circle.
Binary PSK (for which M = 2) has already been discussed at length, and the case M = 4 (QPSK) was introduced in
Worked Examples 11.2 and 11.3. We will now briefly consider the generation and detection of M-ary PSK, treating
QPSK (with M = 4) first and separately, before presenting a general discussion of BER in M-ary PSK systems.

11.10.3.1 QPSK Modulator and Detector

Figure 11.41a shows the block diagram of a QPSK modulator. Consider first the modulator. The serial-to-parallel
converter is a 2-bit shift register that takes in two serial bits b1 b0 during one symbol interval T s = 2T b and makes
b1 available to the upper modulator (labelled I (in-phase) channel) and b0 to the lower modulator (labelled Q
Telegram: @Persian_IEEEComSoc

(quadrature) channel). The sinusoidal carrier supplied to each of the two product modulators comes from a com-
mon source, but the carrier is passed through a 90∘ phase shifting network before being fed to the lower modulator.
The carrier signal of the in-phase channel is therefore cos(2𝜋f c t), whereas the carrier signal of the quadrature
channel is −sin(2𝜋f c t), which leads the cosine carrier by 90∘ . Bits b1 and b0 are represented in the circuit as bipo-
lar voltages of normalised value +1 V (for binary 1) and − 1 V (for binary 0). The QPSK signal is the sum of the I
and Q channel outputs. Thus, when the input is b1 b0 = 00 the (normalised) QPSK pulse is obtained as follows by
straightforward phasor addition

g00 (t) = − cos(2𝜋fc t) + sin(2𝜋fc t)

√
= 2 cos(2𝜋fc t − 135o ) (11.99)
 11.10 M-ary Transmission 739

α1
01 11

α0

00 10

+1
Ts

–1
I channel 180° 180° 0° 0°

cos(2πfct)
Serial-to-parallel

Input Bits
converter

00011110 … QPSK
BNRZ
coder

–sin(2πfct)

+1 Q channel –135° 135° 45° –45°

+1
Ts
–1 Tb
–1
–90° 90° 90° –90°
(a)

BPSK BPSK
QPSK with sine carrier with cosine carrier
–sin (≡ α1) –sin
01 11 1
Es 2
Es 0 1
cos = + cos
(≡ α0) Es 2

0 Es = A2cTs /2
00 10

(b)

yU(Ts) Decision device:

Telegram: @Persian_IEEEComSoc

Matched filter binary 1 if yU(Ts) > 0;

hU(t) yU(t) binary 0 if yU(Ts) < 0
QPSK signal Matched to sample at Parallel
Output
cos(2πfct) t = Ts to serial
r(t) bit stream
converter
yL(Ts) Decision device:
Matched filter
binary 1 if yL(Ts) > 0;
hL(t) yL(t) binary 0 if yL(Ts) < 0
Matched to sample at
–sin(2πf t) t = Ts
c

(c)

Figure 11.41 (a) QPSK modulator and waveforms; (b) QPSK constellation as the sum of two BPSK constellations.
 740 11 Digital Modulated Transmission

Similarly, for the remaining 2-bit inputs 01, 10, and 11, we obtain
√
g01 (t) = − cos(2𝜋fc t) − sin(2𝜋fc t) = 2 cos(2𝜋fc t + 135o )
√
g10 (t) = cos(2𝜋fc t) + sin(2𝜋fc t) = 2 cos(2𝜋fc t − 45o )
√
g11 (t) = cos(2𝜋fc t) − sin(2𝜋fc t) = 2 cos(2𝜋fc t + 45o ) (11.100)
The signal-space diagram of the generated QPSK signal and the signal waveforms at every stage of the QPSK mod-
ulation process are shown in Figure 11.41a. Comparing with Eq. (11.98), we see that this modulator has angular
offset 𝜃 0 = −3𝜋/4 rad, and the states are numbered clockwise from this point, with the bits written left to right.
From the block diagram, we also see that QPSK consists of the simultaneous transmission of two BPSK signals,
one on a cosine carrier and the other on an orthogonal sine carrier of the same frequency. This means that the
QPSK signal-space diagram can be realised by combining two BPSK constellations, as shown in Figure 11.41b.
QPSK detection is accomplished using two matched filters, as shown in Figure 11.41c. The task is to detect the
two bits b1 b0 transmitted during each symbol interval T s . The upper filter is matched to the cosine pulse and detects
b1 , whereas the lower filter is matched to the sine pulse and detects b0 . It is clear from Eqs. (11.99) and (11.100)
that the cosine component in the transmitted pulse is positive for b1 = 1 and negative for b1 = 0. Similarly, the sine
component is positive for b0 = 1 and negative for b0 = 0. Thus, yU (T s ), the output of the upper filter sampled at
t = T s , equals +Es /2 for b1 = 1, and −Es /2 for b1 = 0. In the same way, yL (T s ) = +Es /2 for b0 = 1, and yL (T s ) = −Es /2
for b0 = 0. So, the sampled output of each filter is passed through a decision device with a threshold of zero, which
gives a binary 1 output when its input is positive and a binary 0 output for a negative input. You may wish to review
our detailed discussion of matched filter operation in Section 11.8 if you are in any doubt. The parallel-to-serial
converter is a 2-bit shift register that takes the bits generated by the two decision devices in each interval T s and
clocks them out serially, b1 first followed by b0 .

11.10.3.2 M-ary PSK Modulator and Detector

M-ary PSK modulation (for M > 4) may be achieved using the block diagram shown in Figure 11.42a. The system
involves two branches of modulation that are combined to yield the M-ary PSK signal. During each symbol interval,
the serial-to-parallel converter takes k (= log2 M) bits from the input bit stream and presents this block of bits to
both branches. Each branch uses a nonbinary multilevel bipolar NRZ coder (described in some literature as a
digital-to-analogue converter (DAC)) to map each group of k bits to a discrete output level, denoted AI in the upper
branch and AQ in the lower branch. The assignment of an output level to a bit group in each branch is done in
such a way as to lead to a generated M-ary PSK signal having a Gray code arrangement of states, all lying on
one circle centred at the origin. The multilevel pulse sequence AI of the upper branch is multiplied by a cosine
carrier of frequency f c . This branch is therefore referred to as the in-phase channel. In the lower branch, the pulse
sequence AQ is multiplied by a negative sine carrier of the same frequency f c . This lower branch is therefore called
the quadrature channel since the negative sine carrier leads the cosine carrier by 90∘ in phase. Adding the outputs
Telegram: @Persian_IEEEComSoc

of both product modulators yields an M-ary PSK pulse in each symbol interval (0, T s ) as
( )
t − Ts ∕2
gpsk (t) = [AI cos(2𝜋fc t) − AQ sin(2𝜋fc t)]rect (11.101)
Ts
which (by straightforward phasor addition) has amplitude A and phase 𝜑 given by
√
A = A2I + A2Q ≡ Constant for all bit groups
𝜙 = tan−1 (AQ ∕AI ) ≡ Unique to each bit group (11.102)
As discussed in Section 11.3.2, AI and AQ are the in-phase and quadrature components of each bandpass symbol.
The real and imaginary
√ parts x and y of the state representing each symbol in signal space are obtained by multi-
plying AI and AQ by Ts ∕2 to obtain x and y, respectively. Table 11.2 shows the mapping of input bits to discrete
 11.10 M-ary Transmission 741

Multilevel AI
bipolar
NRZ coder
Input bit
stream Serial to cos(2πfct) M-ary PSK
Gray-coding
(a) parallel
converter –sin(2πfct)

Multilevel AQ
bipolar
NRZ coder
k bits
(k = log2M)

Ts
gpsk(t) xI(t) AI
∫
0

Input M-ary
PSK signal 2cos(2πfct) ADC Output bit stream
(b) decision
–2sin(2πfct) device
Ts
gpsk(t) xQ(t) AQ
∫
0

Multiply Integrate

Figure 11.42 M-ary PSK system: (a) modulator; (b) detector.

Table 11.2 Required mapping of input bits to discrete output levels AI and AQ (normalised) in the
multilevel bipolar NRZ coders of Figure 11.42 to produce the 8-PSK constellation of Figure 11.40.

Input bits AI AQ 8-PSK phase Components in signal space

Real part (x) Imag. part (y)
√ √ √ √
000 1∕ 2 1∕ 2 45∘ Ts ∕2 Ts ∕2
√
001 0 1 90∘ 0 Ts ∕2
√
010 −1 0 180∘ − Ts ∕2 0
√ √ √ √
011 −1∕ 2 1∕ 2 135∘ − Ts ∕2 Ts ∕2
√
100 1 0 0∘ Ts ∕2 0
Telegram: @Persian_IEEEComSoc

√ √ √ √
101 1∕ 2 −1∕ 2 −45∘ Ts ∕2 − Ts ∕2
√ √ √ √
110 −1∕ 2 −1∕ 2 −135∘ − Ts ∕2 − Ts ∕2
√
111 0 −1 −90∘ 0 − Ts ∕2

output levels in Figure 11.42 that is required to produce the 8-PSK constellation discussed in Figure 11.40. The
signal-space components of each state are also shown in the last two columns as x and y.
In general, if we set the M-ary PSK constellation angular offset 𝜃 0 of Eq. (11.98) to

𝜃0 = 𝜋∕M rad (11.103)

 742 11 Digital Modulated Transmission

then the output levels of the bipolar NRZ coders required for state i, counting from i = 0 at the all-binary-zero state
and going counterclockwise to cover the entire upper semicircle of the constellation (with terminal state index
i = M/2–1), are
( )
𝜋
AI (i) = AI (M − 1 − i) = cos (1 + 2i)
M
( )
𝜋
AQ (i) = −AQ (M − 1 − i) = sin (1 + 2i)
M
i = 0, 1, 2, · · · , M∕2 − 1 (11.104)

The assignment of bits to the above states follows a Gray code sequence as earlier discussed, starting at the right-
most point of the upper semicircle with 00…0 and going counterclockwise to cover the upper semicircle, and then
from the rightmost point of the lower semicircle with 10…0 and going clockwise to cover the lower semicircle.
For example, this scheme produces the 8-PSK constellation shown in Figure 11.43 along with tabulated in-phase
and quadrature components AI and AQ for each state.
To devise a process for coherent detection of the above M-ary PSK signal at the receiver, let us multiply the PSK
signal gpsk (t) in Eq. (11.101) by 2cos(2𝜋f c t) and use trigonometric identities to simplify this product signal, denoted
xI (t)

xI (t) = gpsk (t) × 2 cos cos(2𝜋fc t)

( )
t − Ts ∕2
= [AI cos(2𝜋fc t) − AQ sin(2𝜋fc t)]rect × 2 cos cos(2𝜋fc t)
Ts
( )
2 t − Ts ∕2
= [2AI cos (2𝜋fc t) − 2AQ sin(2𝜋fc t) cos(2𝜋fc t)]rect
T
( ) s
t − Ts ∕2
= [AI + AI cos(4𝜋fc t) − AQ sin(4𝜋fc t)]rect (11.105)
Ts

α1 ≡ Q
011 001
Input AQ 8-PSK
AI
Bits Phase
000 0.9239 0.3827 22.5°
010 000
Telegram: @Persian_IEEEComSoc

001 0.3827 0.9239 67.5°

22.5° 010 –0.9239 0.3827 157.5°

α0 ≡ I
011 –0.3827 0.9239 112.5°
100 0.9239 –0.3827 –22.5°
100
110 101 0.3827 –0.9239 –67.5°
110 –0.9239 –0.3827 –157.5°
111 –0.3827 –0.9239 –112.5°
111 101

Figure 11.43 8-PSK constellation with angular offset 𝜃 0 = 22.5∘ and its table of in-phase and quadrature components.
 11.10 M-ary Transmission 743

Next, noting that the sinusoidal pulse completes an integer number n of cycles within a symbol duration T s , so
that f c = n/T s , we integrate xI (t) over one symbol duration to obtain an output yI (T s ) as

Ts
yI (Ts ) = [AI + AI cos(4𝜋fc t) − AQ sin(4𝜋fc t)]dt
∫0
[ ]T
AI sin(4𝜋fc t) AQ cos(4𝜋fc t) || s
= AI t + + |
4𝜋fc 4𝜋fc |
|0
1
= AI Ts + [A sin(4𝜋n) + AQ (cos(4𝜋n) − cos(0))]
4𝜋fc I
= AI Ts ≡ AI (Normalised) (11.106)

Thus, the process of multiplying the incoming M-ary PSK signal by 2cos(2𝜋f c t) before integrating over the symbol
duration yields an estimate of the in-phase component AI of the signal. An equivalent view of this process is that
when the product signal xI (t) in Eq. (11.105) is passed through an LPF then the components at frequency 2f c are
blocked and only the DC component AI is passed. This view is entirely consistent with the fact that an integrator
is of course an LPF. Similarly, if we multiply gpsk (t) by −2sin(2𝜋f c t) to obtain xQ (t) and then integrate xQ (t) over
one symbol duration to obtain an output yQ (T s ), we find that

yQ (Ts ) = AQ (Normalised) (11.107)

In this way, the pair of values (AI , AQ ) used at the transmitter is recovered from the incoming M-ary PSK signal for
each symbol interval. This pair may then be fed into a decision device (or special ADC) which maps or converts
the pair into a group of k bits according to the bits-assignment rules followed at the transmitter. The decisions take
account of the effect of noise and are based on the magnitude of the ratio |AQ ∕AI | and the signs of AI and AQ . For
example, in Figure 11.43, the sectors of the 8-PSK states have been alternately shaded to show the angular range
(or decision boundaries) which the detector uses for each state. The angular range 0–45∘ belongs to state 000, the
range 45∘ –90∘ belongs to state 001, the range 90∘ –135∘ belongs to state 011, and so on. Therefore, for this 8-PSK
constellation, decisions will be made as follows:

● If 0 < |AQ ∕AI | < 1 and AI , AQ > 0, then output bits = 000.
● If |AQ ∕AI | > 1 and AI , AQ > 0, then output bits = 001.
● If |AQ ∕AI | > 1 and AI < 0, AQ > 0, then output bits = 011.
● And so on.

In the unlikely event of the above conditions falling exactly on a decision boundary, one of the two sectors is chosen
Telegram: @Persian_IEEEComSoc

purely at random. Figure 11.42b shows a block diagram of the M-ary PSK coherent detection process described
above.

11.10.3.3 BER of M-ary PSK

The task of determining the BER of M-ary PSK is simplified by the symmetry in the signal-space diagram, e.g.
see Figure 11.43. Every transmitted state has the same probability of error, which is therefore the probability of
symbol error Pes of the system. We will continue with our justified assumption that symbol errors arise from the
mistaking of a message state for its immediate neighbour.
In Figure 11.44, the probability Pesi+ of mistaking a message state Si for its immediate counterclockwise neigh-
bour Si+1 is given by Eq. (11.61), with Eb replaced by Es (= A2c Ts ∕2), and 𝜌 the correlation coefficient of the pulses
 744 11 Digital Modulated Transmission

–sin
Si + 1

ϕ Si ϕ = 2π/M
cos
ϕ

Si – 1

Figure 11.44 Computation of BER in M-ary PSK.

gi′ (t) and gi+1

′
(t) in Eq. (11.98). By definition, see Eq. (11.24)
T
∫0 s gi (t)gi+1 (t)dt
𝜌=
Es
T
A2c ∫0 s cos[2𝜋fc t + 2𝜋i∕M + 𝜃0 ] cos[2𝜋fc t + 2𝜋(i + 1)∕M + 𝜃0 ]dt
=
A2c Ts ∕2
{ Ts Ts }
1
= cos(2𝜋∕M)dt + cos[4𝜋fc t + 2𝜋(2i + 1)∕M + 2𝜃0 ]dt
Ts ∫0 ∫0
= cos(2𝜋∕M) ≡ cos 𝜙 (11.108)

Note that the second integral in the third line is zero, being the integration of a sinusoidal function of frequency
2f c (≡ 2n/T s ) over an interval T s that is an integer number of its period. Eq. (11.108) is an interesting result which
confirms what we already know in the following special cases.

(i) Two states separated by 𝜙 = 90∘ are orthogonal: (𝜌 = cos90∘ = 0), e.g. QPSK.
(ii) Two states separated by 𝜙 = 180∘ are antipodal: (𝜌 = cos180∘ = −1), e.g. BPSK. Note, therefore, that the two
message states in a BPSK system do not have to lie on (opposite sides of) the cosine axis. All that is required
is that they differ in phase by 180∘ .
(iii) Two (equal energy) states separated by 𝜙 = 0∘ are identical: (𝜌 = cos0∘ = 1).

Armed with this result, we return to the problem at hand and obtain
Telegram: @Persian_IEEEComSoc

√
⎛ E (1 − cos 𝜙) ⎞
1 ⎜ s ⎟
Pesi+ = erfc
2 ⎜ 2No ⎟
⎝ ⎠
√
⎛ E (1 − cos 𝜙)log M ⎞
1
= erfc ⎜ b 2 ⎟
2 ⎜ 2No ⎟
⎝ ⎠
where Eb is the energy per bit. Pesi+ obtained above would be the probability of symbol error in an M-ary system
with M = 2. You may wish to verify that this result agrees with Eq. (11.56). However, for M > 2, there is also an
immediate clockwise neighbour Si−1 for which Si can be mistaken with equal probability. Thus, the probability of
symbol error Pes = 2Pesi+ , and BER = Pes /log2 M. It follows that the BER of M-ary PSK (with coherent detection
 11.10 M-ary Transmission 745

0.1 ASK
AS : M = 25
K: 6
M
10–2 =1
28

10–3
256
Bit error ratio (BER)

ASK
10–4
PSK

128

M=:
8 16 32 64
:M

10–5

64
ASK:
= 2,

ASK:
4

ASK: M
10–6

M=3
M=1
ASK: M
ASK: M

2
10–7

6
=8
=4
10–8
=2

10–9
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
Eb/No (dB)

Figure 11.45 BER of coherent M-ary PSK (solid curve) with comparison to M-ary ASK (dashed curve) for M = 2, 4, …, 256.

and M > 2) is given by

√
⎛ E [1 − cos(2𝜋∕M)]log M ⎞
1
BER = erfc ⎜ b 2 ⎟ (11.109)
log2 M ⎜ 2No ⎟
⎝ ⎠
As expected, when M = 4 this equation reduces to Eq. (11.56) for the BER of a QPSK system. It should be empha-
sised that to apply Eq. (11.109) to the binary case (M = 2) a factor of one-half is required to account for the absence
of a second immediate neighbour in the signal-space diagram. Figure 11.45 gives a plot of Eq. (11.109) for various
values of M. The BER of M-ary ASK is also shown in dashed lines for comparison. We may deduce the following
observations from this plot:
● Signal power P (= Eb Rb ) in a multilevel PSK system (M > 2) must be increased in order to maintain the error
ratio obtainable with BPSK (M = 2). If noise power per unit bandwidth (N o ) is the same for all M, then the
required increase in signal power when going from M 1 -ary PSK to M 2 -ary PSK, where M 2 > M 1 and both are
Telegram: @Persian_IEEEComSoc

integer powers of 2, is
( )
Eb | Eb | log2 M2
ΔP = | − | + 10log10 dB (11.110)
No ||M2 -PSK No ||M1 -PSK log2 M1

where the first two terms on the right-hand side are the values of Eb /N o in dB required to achieve a specified BER
in each system (as read from Figure 11.45). For example, we read from Figure 11.45 that to achieve BER = 10−7 ,
8-PSK requires Eb /N o = 14.75 dB and 64-PSK requires Eb /N o = 29.35 dB. Therefore, to switch from 8-PSK to 64-PSK
while maintaining the same signal transmission quality of BER = 10−7 and using the same bandwidth, Eq. (11.110)
stipulates that signal power must be increased by
ΔP = 29.35 − 14.75 + 10log10 (6∕3) = 17.6 dB
 746 11 Digital Modulated Transmission

You may recall (see Figure 6.19) that the bandwidth efficiency of M-ary PSK increases logarithmically with M,
so this observation is simply highlighting the trading of signal power for improved bandwidth efficiency.

● The noise performance of M-ary PSK is clearly superior to that of M-ary ASK. In fact, 8-PSK has about the same
BER as a binary ASK (i.e. OOK) system that uses the same average energy per bit. To further illustrate, achieving
the same BER = 10−7 in a 4-ASK system as in a 4-PSK system requires the signal power of the 4-ASK system to be
increased by a factor of 6.9 (≡ 8.4 dB) above that of the 4-PSK system. Since M-ary ASK and M-ary PSK require
the same bandwidth, there is no reason to choose multilevel ASK, except for the simplicity of its modulation
and detection circuits.

The complexity of M-ary PSK detection can be greatly simplified if the modulator represents message state i by a
shift 𝛼i in the carrier phase of the previous interval, rather than by an absolute phase value. The receiver then
performs detection by comparing the phase of the sinusoidal pulse received in the current interval to that of
the previous interval. This implementation is known as M-ary DPSK, the binary case of which is discussed in
Section 11.9.3. M-ary DPSK, however, has the disadvantage of an inferior noise performance compared to the
coherent M-ary PSK discussed in this section.

11.10.4 M-ary FSK

An M-ary FSK signal consists of one of the following M orthogonal sinusoidal pulses in each symbol interval
of duration T s . The pulses differ only in frequency, which are spaced at half the symbol rate to achieve mutual
orthogonality of the transmitted symbols with a minimum required bandwidth.
( )
t − Ts ∕2
gi (t) = Ac cos[2𝜋(f0 + iΔf )t]rect
Ts
i = 0, 1, 2, · · · , M − 1; Δf = 1∕2Ts = Rs ∕2 (11.111)

The frequency f 0 is chosen to place the transmission in the desired or allocated frequency band.

11.10.4.1 M-ary FSK Modulator and Detector

A simple arrangement for generating M-ary FSK is shown in Figure 11.44a. During each symbol interval (of dura-
tion T s ) the log2 M-bit to M-level converter takes log2 M bits from the input bit stream and converts them to a
normalised output equal to their decimal equivalent, 0, 1, 2, …, M-1. The output of the M-level converter drives a
voltage-controlled oscillator (VCO). If the normalised frequency sensitivity of the VCO is Δf and its free-running
frequency is f 0 , then its output is the M-ary FSK signal given by Eq. (11.111).
Detection of M-ary FSK requires M matched filters or, equivalently, correlation receivers connected as shown in
Telegram: @Persian_IEEEComSoc

Figure 11.46b. Only one orthogonal pulse is present in the received signal during each symbol interval. Therefore,
during each interval only one of the M branches (the one matched to the transmitted pulse during that interval)
will have a significant output whereas the output of the other branches will be negligible. So, the outputs of all M
branches after each integration cycle are fed into a decision device, which generates a normalised output equal to
the index number (0, 1, 2, …, M − 1) of the input port with the maximum input. A binary encoder then produces
the corresponding group of log2 M bits, which may be clocked out in a serial fashion using a parallel-to-serial
converter.

11.10.4.2 BER of M-ary FSK

We can obtain an upper bound, referred to as the union bound, for the BER of M-ary FSK in a straightforward
manner if we make three important observations.
 11.10 M-ary Transmission 747

Input bit stream Serial to log2M-bit

M-ary FSK
(a) parallel to M-level VCO
signal
converter converter

Ts y0(Ts)
∫
0

cos[2πf0t]
(b)
Ts y1(Ts) Decision Output
∫ device: Parallel bit stream
M-ary FSK 0 Binary
chooses to serial
encoder
gfsk(t) cos[2π(f0 + Δf)t] maximum converter
input

Ts
∫ yM–1(Ts)
0

cos[2π(f0 + (M – 1)Δf)t]

Figure 11.46 M-ary FSK: (a) modulator; (b) coherent detector.

f0 + Rs/2

S1

2d d= Es
2d d
d S0
f0
d
2d
S2

f0 + R s

Figure 11.47 Distances in M-ary FSK signal space, M = 3.

Telegram: @Persian_IEEEComSoc

● M-ary FSK has an M-dimensional signal space with all message points at the same distance from each
other. Thus, every message point has M-1 adjacent (or nearest) neighbours. This observation is illustrated in
Figure 11.47 for M = 3. In fact, the distance between any pair of message points in the signal space of M-ary
√
FSK (for all M) is 2Es , where Es is the energy per symbol.
● All message points are mutually orthogonal, which yields 𝜌 = 0 in Eq. (11.61).
● Gray coding is not applicable. By averaging the number of bit errors incurred when a message state is mistaken
for each of its M − 1 neighbours we find that

M∕2
BER = × Symbol Error Rate (11.112)
M−1
 748 11 Digital Modulated Transmission

With these observations, the probability Pes1 of mistaking a message point for one other message point follows
from Eq. (11.61) with Eb replaced by Es (the energy per symbol), and 𝜌 = 0. Thus
(√ )
1 Es
Pes1 = erfc
2 2No
√
⎛ E log M ⎞
1
= erfc ⎜ b 2 ⎟ (11.113)
2 ⎜ 2No ⎟
⎝ ⎠
where Eb is the energy per bit. The maximum probability of symbol error Pesmax is the sum of the probabilities (all
equal to Pes1 ) of mistaking a message state for each of its M − 1 neighbours. That is

Pesmax = (M − 1)Pes1 (11.114)

This is an upper bound on Pes , because summing the individual probabilities implicitly assumes independence of
each of the events. That is, it assumes that when an error occurs the received state is nearer to only one other state
than to the transmitted state. However, there will be some situations in which a received state is nearer to two or
more other states than the transmitted state. In this case, summing the probabilities needlessly increases the overall
probability of error by including regions of intersection more than once. It follows from Eqs. (11.112–11.114) that
the upper bound on BER in an M-ary FSK system (with coherent detection) is given by
√
⎛ Eb log2 M ⎞
M
BER ≤ erfc ⎜ ⎟ (11.115)
4 ⎜ 2No ⎟
⎝ ⎠
For M = 2 the bound becomes an equality and Eq. (11.115) reduces to the result obtained earlier for binary FSK.

11.10.4.3 Noise-bandwidth Trade-off in M-ary FSK

Equation (11.115) is plotted in Figure 11.48 for selected values of M. Note that, in contrast to M-ary ASK and M-ary
PSK, the BER performance improves as M increases. It is very important that you understand why this is the case.
M-ary ASK and M-ary PSK transmit different amplitudes and phases of the same carrier frequency. In so doing
bandwidth is conserved, but the transmitted sinusoidal pulses become increasingly positively correlated as M
is increased. As a result, it becomes more and more difficult to distinguish the pulses at the receiver and the
system becomes increasingly vulnerable to noise. In M-ary FSK, on the other hand, mutually orthogonal pulses are
transmitted at all values of M, making them perfectly distinguishable at the receiver. However, this is achieved at
the expense of bandwidth since each of the M pulses must have a different frequency, with a separation equal to half
the symbol rate or 1/2T s . Increasing M increases T s and allows closer spacing of the frequencies. But the spacing
Telegram: @Persian_IEEEComSoc

reduces proportionately to log2 M, whereas the number of frequencies required increases with M. So, overall the
bandwidth increases roughly proportionately to M/log2 M. As M is increased, the symbol interval T s (= T b log2 M)
increases. Since the receiver performs detection by integrating over an interval T s , the contribution of random
noise is significantly reduced as the integration interval increases, leading to improved BER performance.
You can see how an increase in T s reduces noise effects by noting that if random noise is observed over a suffi-
ciently long time then the amount of positive and negative samples will be equal, giving a sum of zero. The action
of integration is equivalent to summing these samples and including a scaling factor. Of course, the effect of noise
is also reduced in M-ary ASK and M-ary PSK as M and hence T s increases, but the benefit to the receiver of this
reduction is more than cancelled out by the increased correlation of the transmitted pulses. Therefore, M-ary FSK
gives us the ability to trade bandwidth for an improved noise performance in a way that is not possible with M-ary
PSK and M-ary ASK.
 11.10 M-ary Transmission 749

0.1

10–2

10–3
M=2
Bit error ratio (BER)

10–4
4

10–5

16
10–6
64
10–7
1024
256
10–8

10–9
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Eb/No (dB)

Figure 11.48 Upper bound of BER of M-ary FSK for M = 2, 4, 16, 64, 256, 1024.

11.10.5 M-ary APSK

M-ary PSK constrains all M message states to a circle centred about the origin in signal space. BER performance
improvement can be achieved if the message states are spaced more freely and widely in the 2D signal space. This
requires combining two quadrature (sine and cosine) carriers that are modulated both in amplitude and phase.
This is therefore a hybrid modulation technique known as M-ary amplitude and phase shift keying (APSK), which in
some literature is also described as quadrature amplitude modulation (QAM). There are different implementations
of APSK, leading to square, circular, and star constellations, as shown in Figure 11.49 for M = 16. The star APSK
for M = 16 has a minimum phase difference of 90∘ between message points of the same energy and is therefore
thought to perform better than square APSK in transmission media with a predominance of phase distortion.
Our discussion focuses on square APSK and its BER performance, with 16-APSK modulation and detection pre-
sented in detail. The same principles can be applied to other types of APSK. In general, APSK involves sinusoidal
pulses of duration T s and of the form
( )
Telegram: @Persian_IEEEComSoc

t − Ts ∕2
gi (t) = Ai cos(2𝜋fc t + 𝜙i )rect (11.116)
Ts
The amplitude Ai and phase 𝜙i take on discrete sets of values, which depend on the implementation.

11.10.5.1 16-APSK
An illustration of the combination of two quadrature carriers to form M-ary APSK is shown in Figure 11.50 for
M = 16. A sine carrier (of frequency f c and duration T s ) codes 2 bits using two phases (0 and 180∘ ) and two (nor-
malised) amplitudes 1 and 3. An orthogonal cosine carrier also codes 2 bits in a similar manner. The two pulses
are added, with the result that 4 bits are conveyed in each symbol interval using a sinusoidal pulse (of frequency
f c and duration T s ) that can take on three different amplitudes (indicated by the dotted circles) and a number of
phase angles. Thus, whereas all transmitted pulses in 16-PSK have the same energy Es , in this square 16-APSK
 750 11 Digital Modulated Transmission

–sin
–sin

cos cos

Square APSK Circular APSK

–sin

cos

Star APSK

Figure 11.49 Various APSK constellations.

four symbols are transmitted with minimum energy E0 , eight symbols with energy E1 = 5E0 , and four with peak
energy E2 = 9E0 . The average energy per symbol in 16-APSK is therefore
4E0 + 8(5E0 ) + 4(9E0 )
Es =
16
= 5E0 = 5A2c Ts (11.117)
where Ac is the amplitude of each quadrature carrier.
Figure 11.51a shows the block diagram of a 16-APSK modulator. The 2-bit GCC performs the following conver-
Telegram: @Persian_IEEEComSoc

sions
00 → −3
01 → −1
11 → +1
10 → +3 (11.118)
During each symbol interval T s = 4T b , the serial-to-parallel converter takes 4 bits b3 b2 b1 b0 of the input bit stream
and presents b1 b0 to the lower GCC and b2 b3 to the upper GCC. Note the flipping of bit order in the latter. The
output of the upper GCC multiplies a carrier Ac sin(2𝜋fc t) to give the quadrature component of the 16-APSK sig-
nal, whereas the output of the lower GCC multiplies a carrier Ac cos(2𝜋fc t) to give the in-phase component. Both
 11.10 M-ary Transmission 751

–sin

00

10
origin
origin + cos
00 01 11 10
11
16-APSK
–sin

01

0000 0001 0011 0010

2d
1000 1001 1011 1010
E0 = 2d2 E1
d E0
E1 = 10d2 = 5E0 cos
E2 = 18d2 = 9E0 d

1100 1101 1111 1110

E2
2d

0100 0101 0111 0110

2d d d 2d

Figure 11.50 Square M-ary APSK constellation, M = 16.

carriers are derived from a common source and have the same amplitude Ac and frequency f c but differ in phase
by 90∘ . The desired 16-APSK signal is obtained by summing the in-phase and quadrature channels.
As an example, assume an input b3 b2 b1 b0 = 0111. Following the Gray code conversion in Eq. (11.118), the
in-phase channel output is

yi11 (t) = Ac cos(2𝜋fc t)

Telegram: @Persian_IEEEComSoc

And the quadrature channel output is

yq01 (t) = 3Ac sin(2𝜋fc t)

Note that the swapping of bits at the input to the upper GCC causes b3 b2 to be received as b2 b3 , and hence the
above result for b3 b2 = 01. The resulting 16-APSK pulse is therefore
g0111 (t) = Ac cos(2𝜋fc t) + 3Ac sin(2𝜋fc t)
√
= 10A cos(2𝜋f t − 71.6∘ )
c c (11.119)

Note that this pulse (of duration T s ) has energy 5A2c Ts = E1 , and phase −71.6∘ , which is in agreement with the
location of 0111 on the 16-APSK constellation diagram of Figure 11.50. By proceeding in the same manner, you
 752 11 Digital Modulated Transmission

(a) 2-bit
Gray code ± 1, ±3
×
to 4-level Quadrature
converter channel
b3 Ac sin(2πfct) +
Input bit stream b2
Σ 16-APSK
b1
Ac cos(2πfct) +
b0
Serial-to-parallel 2-bit In-phase
converter Gray code ± 1, ±3 channel
to 4-level ×
converter

(b)
4-level to
Matched filter yU(Ts) 4-level 2-bit
hU(t) yU(t) quantiser Gray code
converter b3
Output
Matched to sample at
16-APSK signal ± E0/2, b2 bit stream
Ac sin(2πfct) t = Ts
± 3E0/2 b1
4-level to b0
Matched filter yL(Ts) 4-level 2-bit
hL(t) quantiser Gray code Parallel-
yL(t) to-serial
converter
Matched to sample at converter
A cos(2πf t) t = Ts
c c

Figure 11.51 Square 16-APSK: (a) modulator; (b) detector.

may verify that the block diagram of Figure 11.51a does generate the array of message points in Figure 11.50. It
is noteworthy that all adjacent points on the 16-APSK constellation differ in only one bit position, even for points
(like 1111) with up to four immediate neighbours. This is important for BER minimisation, as discussed earlier,
and results from the use of a Gray code in each channel.
A 16-APSK detector is shown in Figure 11.51b. The 4-level quantiser approximates each matched filter output
to the nearest of 4 levels −3E0 /2, −E0 /2, +E0 /2, and + 3E0 /2. A 4-level to 2-bit GCC then generates the bits in each
channel according to the conversions
−3E0 ∕2 → 00
−E0 ∕2 → 01
E0 ∕2 → 11
Telegram: @Persian_IEEEComSoc

3E0 ∕2 → 10 (11.120)
The 2 bits from each channel are combined as shown in a parallel-to-serial converter to produce a serial output
bit stream.

11.10.5.2 BER of Square M-ary APSK

A square M-ary APSK constellation is one in which the signal states form a regular 2D grid with equal spacing in
both horizontal and vertical directions centred at the origin in signal space. An example is shown in Figure 11.50
for M = 16. The system conveys k = log2 M bits per symbol interval using two orthogonal and hence independent
channels. If M is an even integer power of 2 (i.e. M = 4, 16, 64, 256, 1024, …) then the bits are shared
√ equally among
the two channels, with the in-phase and quadrature channels each conveying k∕2 = log2 M bits per symbol
 11.10 M-ary Transmission 753

interval. But if M is an odd integer power of 2√ (i.e. M = 8, 32, 128, 512, …) then the two channels will be unequally
loaded
√ with one conveying (k + 1)∕2 = log2 2M bits per symbol interval and the other conveying (k − 1)∕2 =
log2 M∕2 bits per symbol interval. It may be shown (see page 382–385 of [1]) that the BER of such M-ary APSK
is given by the expression
[√ ] √
⎛ 3log2 M Eb ⎞
2 M−u ⎜ ⎟
BER = √ erfc
log2 M M ⎜ 2(aM − 1) No ⎟
⎝ ⎠
where,
a = u = 1, for M an even integer power of 2
√
3 2
a = 1.25; u = , for M an odd integer power of 2 (11.121)
4
Note that when M = 4 this equation reduces to Eq. (11.56) for the BER of a QPSK system. This is to be expected
since 4-APSK and QPSK are of course identical. Figure 11.52 shows the BER of square M-ary APSK versus Eb /N o
(expressed in dB) for various values of M. For comparison, the BER of the corresponding M-ary PSK is also shown
in dashed lines. We can see that M-ary APSK provides a significant improvement in BER and allows a lower value
of Eb /N o to be used for a given error performance. For example, there is a saving of 9.8 dB in the signal power
required by 64-APSK for a BER of 10−7 compared to a 64-PSK system of the same BER. However, unlike M-ary
PSK, the performance of M-ary APSK is sensitive to channel nonlinearity. The superior noise performance evident
in Figure 11.52 assumes that there are no amplitude distortions in the transmission system. We also see that the
noise performance of 8-APSK is only marginally better than that of 8-PSK (and this is because errors in 8-APSK
are dominated in one of the orthogonal channels that carries 2 bits per symbol), so, given the sensitivity of 8-APSK
to channel nonlinearity, 8-PSK will be preferred to 8-APSK in all applications. Furthermore, Figure 11.52 shows

APSK PSK
0.1

10–2

10–3
Bit error ratio (BER)

10–4

10–5
Telegram: @Persian_IEEEComSoc

10–6

10–7

10–8

10–9
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
Eb/No (dB)

Figure 11.52 BER of square M-ary APSK, with comparison to M-ary PSK.
 754 11 Digital Modulated Transmission

the noise performance of 64-APSK as being very close to that of 16-PSK. However, in view of Eq. (11.110) we know
that an extra signal power of 2.1 dB is required to increase bit rate by a factor of 3/2 (using the same bandwidth
and maintaining the same low BER) when one switches from 16-PSK to 64-APSK.

11.11 Design Parameters

A designer of a digital transmission system has the following important parameters at their disposal.

● Transmission bandwidth: this is a scarce resource that must be used judiciously to maximise the number of users
or services. All digital transmission systems employ raised cosine filters (introduced in Worked Example 4.16)
for which the transmission bandwidth (also called occupied bandwidth) Bocc in Hz when transmitting at symbol
rate Rs in baud (i.e. symbols per second) is
⎧
⎪Rs (1 + 𝛼), M-ary ASK, PSK, APSK
Bocc = ⎨ (M + 1 ) (11.122)
R
⎪ s + 𝛼 , M-ary FSK
⎩ 2

and the noise equivalent bandwidth B (discussed in Section 4.7.3.5) is

B = Rs (1 − 𝛼∕4) (11.123)

where 𝛼 is the roll-off factor of the raised cosine filter. Eq. (11.123) applies to all M-ary modulation schemes (includ-
ing FSK) since they employ a bank of correlation receivers, each matched to one symbol of duration T s = 1/Rs .

● Signal power: this is an expensive resource that has a direct bearing on component sizes, battery life, radiation
safety, and potential interference to other systems. The signal power Ps at the reference point of the receiver
input may be determined in various transmission media, as discussed extensively in Chapter 5.
● Noise power: the presence of noise places a fundamental limit on the ability of the receiver circuit to distinguish
between each of the unique symbols used by the transmitter to convey different bit groups. Every instance of
symbol misidentification produces one or more errors in the recovered bit stream. The noise power Pn at the
reference point of the receiver input is given by the product of the noise power per unit bandwidth N o and the
noise equivalent bandwidth B of Eq. (11.123):

Pn = No B = kT sys B (11.124)

where k = 1.38 × 10−23 J/K is Boltzmann’s constant and T sys is the noise temperature of the receiver, as discussed
in Chapter 6.
Telegram: @Persian_IEEEComSoc

● Carrier-to-noise ratio (C/N): the performance quality of a transmission system is determined not just by signal
power but by the ratio between signal power Ps and noise power Pn (called the carrier-to-noise ratio and denoted
C/N) at the reference point of the receiver input. Another parameter that is widely used in digital transmission
systems to compare signal and noise powers is the ratio Eb /N o between the average energy per bit Eb and the
noise power per unit bandwidth N o at the receiver input. Using Eq. (11.123) and (11.19), the two parameters are
related as follows
C ER E Rs log2 M
= b b = b ×
N No B No Rs (1 − 𝛼∕4)
Eb log2 M
= × (11.125)
No (1 − 𝛼∕4)
 11.11 Design Parameters 755

Here, Rb is the coded bit rate which includes any redundant bits inserted for the purpose of error control. The
system designer must think beyond simply increasing the transmitted signal power and explore ways of reducing
overall signal loss as well as the noisiness of the transmission medium and the receiver.
● Bit rate: some services, such as interactive digital television, require a minimum transmission bit rate for proper
operation. The time of transmitting data over a communication network decreases proportionately with bit rate,
and this will have implications on service cost if charged according to service duration, as in plain old telephone
service (POTS). Bit rate Rb is directly related to symbol rate Rs through the order M of the M-ary modulation
scheme and is only indirectly related to transmission bandwidth Bocc through Rs (as stated in Eq. (11.122))

Rb = Rs log2 M (11.126)

● Bit error ratio: most services specify a maximum BER that can be tolerated, for example 10−4 for voice and 10−7
for data. The transmission link must therefore be designed to deliver at the reference point of the receiver input
a signal that has a C/N or Eb /N o which satisfies the required BER specification upon demodulation and error
control decoding.
● Modcod: the combination of modulation and error control coding schemes is often referred to as modcod. All
the M-ary modulation schemes discussed in this chapter enable log2 M bits to be conveyed by each transmitted
symbol. In the schemes of M-ary ASK, PSK, and APSK the transmitted symbols all have the same frequency, so
bandwidth is used more efficiently by a factor of log2 M compared to binary transmission in which M = 2, but
signal power must be increased to maintain the same BER. In the case of M-ary FSK, the symbols have different
orthogonal frequencies, so bandwidth is used less efficiently as M increases, but a lower signal power is required
to maintain the same BER. In all cases, redundant bits may be systematically introduced prior to modulation
to enable error detection/correction at the receiver which allows the use of a lower signal power to maintain
the same BER. When error control coding is used in this way, the ratio between message bits and message plus
redundant bits is known as the code rate r and the dB reduction in the required value of Eb /N o is known as
coding gain Gc . The threshold value of Eb /N o required to not exceed a specified maximum BER on the link is
then given by

(Eb ∕No )threshold = (Eb ∕No )theoretical − Gc + Lmil (11.127)

where (Eb /N o )theoretical is the theoretical value of Eb /N o at the specified BER read from graphs such as
Figure 11.45, 11.48 and 11.52 for the modulation scheme, and Lmil is the modem implementation loss, which
indicates that a practical modem will not be as efficient as suggested by the theoretical BER curves and will
therefore require an extra signal power given by Lmil in dB.
Telegram: @Persian_IEEEComSoc

Let us bring together in one place the expressions that relate the above design parameters in each of the digital
modulated systems that we have studied.
Message bit rate Rb (which includes all bits except those inserted for the sole purpose of error control at code
rate r), transmission or occupied bandwidth Bocc , modulation order M, bandwidth efficiency 𝜂, and raised cosine
filter’s roll-off factor 𝛼 are related as follows

⎧ rlog2 M , ASK, PSK, APSK

Rb ⎪ 1+𝛼
𝜂= =⎨ (11.128)
Bocc rlog2 M
⎪ , FSK
⎩ 𝛼 + (M + 1)∕2
 756 11 Digital Modulated Transmission

BER and Eb /N o are related as follows in the various transmission systems

⎧ √
⎛ ⎞
⎪ (M − 1) ⎜
3Eb log2 M
⎟,
⎪ Mlog M erfc ASK
⎜ 2No (2M − 1)(M − 1) ⎟
⎪ 2
⎝ ⎠
⎪ √
⎪ ⎛ ⎞
⎪ 1 erfc ⎜ Eb [1 − cos(2𝜋∕M)]log2 M ⎟ , PSK
⎪ log2 M ⎜ 2No ⎟ (M>2)
⎪ ⎝ ⎠
BER = ⎨ [√ ] √ (11.129)
⎪ 2 ⎛ 3log2 M Eb ⎞
M−u
⎪ √ erfc ⎜ ⎟, APSK
⎪ log2 M M ⎜ 2(aM − 1) No ⎟
⎪ ⎝ ⎠
⎪ √
⎪ M ⎛ E log M ⎞
⎪≤ erfc ⎜ ⎟,
b 2
FSK
⎪ 4 ⎜ 2No ⎟
⎩ ⎝ ⎠
√
where a = u = 1 for M an even integer power of 2, and a = 1.25, u = 3 2∕4 for M an odd integer power of 2. Note
that the APSK constellation to which the above BER expression applies is a square APSK having states arranged
in a uniform 2D grid. Also, you will recall that the BER for BPSK (M = 2) is the same as for QPSK (M = 4) and is
given as half the above PSK expression with M = 2. Furthermore, for binary FSK the BER is exactly equal to the
upper bound given above.

Worked Example 11.5

An on-board-processing (OBP) geostationary orbit (GEO) satellite system provides broadband communication
to 20 VSAT terminals. Transmission on the inbound link from each VSAT to the satellite is at 12 Mb/s using
8-PSK modulation. The satellite employs 16-APSK modulation to transmit on the outbound link at 240 Mb/s
to the VSATs, and this transmission consists of packets addressed to each of the 20 VSATs. Each VSAT receives
the 240 Mb/s bit stream from the satellite and processes this to extract the information addressed to it. All links
employ raised cosine filters with roll-off factor 𝛼 = 0.2. Determine:
(a) The noise equivalent bandwidth of the VSAT receiver.
(b) The occupied bandwidth of the transmission from VSAT to satellite.
(c) The BER at the satellite receiver if the 8-PSK demodulator on the satellite has a 1 dB implementation loss
and the inbound link from VSAT to satellite has C/N = 20 dB.
(d) The fade margin of the inbound link in (c) if threshold BER (prior to error correction in the onboard decoder)
Telegram: @Persian_IEEEComSoc

is set at 10−3 .
(e) The BER at the VSAT receiver if the 16-APSK demodulator in the VSAT receiver has a 1.5 dB implementation
loss and the outbound link from satellite to VSAT has C/N = 22 dB.
(f) The fade margin of the outbound link in (e) if threshold BER (prior to error correction in the VSAT decoder)
is set at 3 × 10−4 .

(a) Each VSAT receiver (on downlink) is required to receive at bit rate Rb = 240 Mb/s transmitted via 16-APSK by
the satellite. Thus, received symbol rate Rs = Rb /log2 M = 240/log2 16 = 240/4 = 60 MBd. Using Eq. (11.123),
the VSAT receiver’s noise equivalent bandwidth is

B = Rs (1 − 𝛼∕4) = 60(1 − 0.2∕4) = 57 MHz

 11.12 Summary 757

(b) Transmission from VSAT to satellite is at Rb = 12 Mb/s using 8-PSK. Thus, symbol rate Rs = Rb /log2
M = 12/log2 8 = 12/3 = 4 MBd, and occupied bandwidth is
Bocc = Rs (1 + 𝛼) = 4(1 + 0.2) = 4.8 MHz
(c) A practical modem with implementation loss Lmil = 1 dB has the same BER at C/N = 20 dB as a theoretically
ideal modem has at C/N = 20−Lmil = 19 dB. Since the BER curve for 8-PSK in Figure 11.45 is plotted as a
function of Eb /N o , we use Eq. (11.125) to convert C/N to Eb /N o , obtaining
( ) ( )
Eb C log2 M log2 8
= − 10log10 = 19 − 10log10 = 19 − 5 = 14 dB
No N 1 − 𝛼∕4 1 − 0.2∕4
Reading the 8-PSK BER curve of Figure 11.45 at Eb /N o = 14 dB yields the BER in the satellite receiver as
BER = 8.8 × 10−7 .
From the BER vs Eb /N o curve of Figure 11.45 for 8-PSK, a threshold BER of 10−3 corresponds to a threshold
Eb /N o of 10 dB. Converting to C/N
( )
C Eb log2 M
= + 10log10 = 10 + 5 = 15 dB
N No 1 − 𝛼∕4
This is the C/N level required to achieve the threshold BER of 10−3 in a theoretical 8-PSK modem. A practical
modem will need something higher by Lmil = 1 dB. Thus, the threshold C/N is 15 + 1 dB = 16 dB. Fade margin
is the amount by which actual link C/N exceeds the threshold value. Thus, the fade margin of the inbound
link is 20–16 = 4 dB.
(d) This practical modem with modem implementation loss Lmil = 1.5 dB will have the same BER at 22 dB as a
theoretical modem has at C/N = 22–1.5 = 20.5 dB. We will assume that the 16-APSK scheme has a square
constellation so that the BER curves of Figure 11.52 are applicable. Converting C/N to Eb /N o yields
( ) ( )
Eb C log2 M log2 16
= − 10log10 = 20.5 − 10log10
No N 1 − 𝛼∕4 1 − 0.2∕4
= 20.5 − 6.24 = 14.26 dB
Reading the 16-APSK BER curve of Figure 11.52 at Eb /N o = 14.26 dB yields the BER in the VSAT receiver as
BER = 1.5 × 10−6 .
(e) From the 16-APSK BER curve of Figure 11.52, Eb /N o = 11.5 dB at BER = 3 × 10−4 . The corresponding C/N is
C/N = Eb /N o + 6.24 = 18.04 dB. The practical C/N is 1.5 dB higher due to modem implementation loss. Thus,
threshold C/N = 19.54 dB, and since the link’s actual C/N is 22 dB, the fade margin is 2.46 dB.

11.12 Summary
Telegram: @Persian_IEEEComSoc

We have now come to the end of our study of digital modulated transmission, which featured an in-depth instruc-
tion in the analysis, design, and operation of the major digital modulation schemes. It must be emphasised that
our discussion was not exhaustive. However, the solid foundation and thorough understanding which you have
now acquired in the principles should give you plenty of confidence and all the tools required for dealing with the
many applications and variant techniques in this rapidly developing field.
Binary and M-ary ASK offer the advantages of bandwidth efficiency and simple modulation and demodulation
circuits. However, this class of techniques suffer from a poor BER and require, comparatively, the largest signal
power for an acceptable BER.
Binary and M-ary PSK have the same high bandwidth efficiency as the corresponding ASK system as well as the
added advantage of good (i.e. low) BERs. However, they require complex modulation and demodulation circuits
 758 11 Digital Modulated Transmission

and, beyond M = 4, they are significantly inferior to M-ary APSK in BER performance when the transmission
system does not introduce amplitude distortions.
Binary and M-ary FSK have the poorest bandwidth efficiency of all the digital modulation techniques, with
a peak at M = 4. The BER performance is the same as in ASK for M = 2, but M-ary FSK allows a unique and
subtle exchange between bandwidth, signal power, and BER, as has been discussed. The circuit complexity of FSK
systems lies somewhere between that of ASK and PSK.
M-ary APSK, with its less restricted distribution of signal states in a 2D signal space, provides good efficiency
in signal power consumption. It has the same bandwidth efficiency as the corresponding ASK and PSK systems
but a better BER than both systems. It, however, has the drawbacks of circuit complexity (comparable to PSK) and
susceptibility to channel nonlinearity.
Historically, early low-speed modems used for data transmission over the public switched telephone network
(PSTN) employed FSK. Many International Telecommunication Union (ITU) standards were specified. For
example, the V.21 specified a full duplex modem operating at 300 b/s. Sinusoids at frequencies f 0 = 980 Hz and
f 1 = 1180 Hz were used in the forward direction, and f 0 = 1650 Hz and f 1 = 1850 Hz in the return direction. The
V.23 modem provided a half-duplex operation at 1.2 kb/s using frequencies f 0 = 1300 Hz and f 2 = 2100 Hz. There
was provision in this standard for a 75 b/s back channel that used tones at 390 and 450 Hz. FSK was also employed
for teletype transmission via HF and VHF radio. PSK was used in voice-band full-duplex synchronous modems
for data transmission over the PSTN. For example, QPSK was used in the ITU-T V.22 and V.26 modems, which,
respectively, operated at carrier frequencies of 1.2 kHz and 1.8 kHz, and bit rates of 1.2 kb/s and 2.4 kb/s. The
V.29 and V.32 modems used 16-APSK to achieve a bit rate of 9.6 kb/s. The V.32 can also operate with a 32-state
signal space (or 32-APSK), where extra bits are included for forward error correction, using a technique known
as trellis coding. This allows the same BER to be achieved with 4 dB less signal power than in 16-APSK. Such a
saving in signal power at the same BER and transmission bandwidth is referred to as a coding gain. There is of
course necessarily a higher circuit complexity. The V.33 modem operated at a carrier frequency of 1.8 kHz and
delivered a maximum bit rate of 14.4 kb/s using 128-APSK, and 12 kb/s using 64-APSK. Both included trellis
coding. The V.34 modem delivered various bit rates up to a maximum of 33.6 kb/s using trellis coding and subsets
of a 1664-APSK constellation. The V.90 modem also employed APSK, with a constellation in excess of 1024 states,
to achieve a bit rate of 56 kb/s.
Modern applications of the digital modulation principles discussed in this chapter are much more varied and
sophisticated. Binary ASK, implemented as OOK and in many cases combined with wavelength division multi-
plexing, is widely employed for transmission in optical fibre links. High-order M-ary FSK is employed in deep
space communications where its unique features can be exploited to enhance the reliable detection of extremely
weak signals. BPSK, QPSK, and 8-PSK are employed in radio frequency identification (RFID), wireless personal
Telegram: @Persian_IEEEComSoc

area networks (such as Bluetooth and ZigBee), wireless local area networks (such as Wi-Fi), various generations
of terrestrial mobile communication systems and satellite communications. High-order M-ary APSK (for M ≥ 16)
is increasingly widely employed in terrestrial and satellite communication systems to support an adaptive mod-
ulation strategy which strives always to use the most bandwidth-efficient modulation and coding combination
permitted by the prevailing transmission channel conditions.

Reference

1 Otung, I. (2014). Digital Communications: Principles & Systems. London: Institution of Engineering and Technol-
ogy (IET). ISBN: 978-1849196116.
 Questions 759

Questions

1 .a) Determine the duration, energy, centre frequency, and bandwidth of the bandpass pulse
v(t) = 20rect(2 × 103 t) sin(4𝜋 × 104 t) V
b) Sketch the waveform of the above pulse.

2 Figure Q11.2 shows the orthonormal basis functions and constellation

diagram of a baseband transmission system.
(a) Determine the amplitude V 𝜙 of the basis functions.
(b) Sketch a clearly labelled waveform of each transmitted symbol.
(c) Calculate the energy of each transmitted symbol.

3 A transmission system conveys information using the symbols shown in Figure Q11.3.
(a) Determine and sketch the orthonormal basis functions of the system.
(b) Calculate the energy of each of the transmitted symbols.
(c) Express each symbol as a linear combination of the basis functions.
(d) Sketch the constellation diagram of the transmission system.

4 Sketch the constellation diagram of a transmission system that employs the symbols g0 (t) to g7 (t) shown in
Figure Q11.4. Calculate the energy of each symbol.

5 Determine and sketch the orthonormal basis function(s) of a transmission system that uses the following
pulses: g2 (t) and g5 (t) in Figure Q11.4. g0 (t) and g1 (t) in Figure Q11.3. g3 (t) and g8 (t) in Figure Q11.3.

6 A digital modulated system transmits the following symbols

⎧ [ ]
𝜋
⎪10(1 + j) cos 8000𝜋t + 4 (2i + 1) , 0 ≤ t ≤ 1 ms
gk (t) = ⎨
⎪ 0, elsewhere
⎩
where i = 0, 1, 2, 3; j = 0, 1; k = i + 4j

α0(t) α1
Vϕ
2 S3
Telegram: @Persian_IEEEComSoc

t, μs S1 S2
100 α0
α1(t) –2 2
Vϕ

–2 S0
t, μs
100 200

Figure Q11.2 Question 11.2.

 760 11 Digital Modulated Transmission

.
(a) What is the dimension N of the system’s signal space?
(b) Sketch the constellation diagram. What are the values of symbol energies used?
(c) Determine the system’s orthonormal basis functions.
(d) Express each transmitted symbol as a linear combination of the orthonormal basis functions.
(e) What is the name of the modulation technique employed?

7 Calculate the correlation coefficient of the following energy signals:

(a) g4 (t) and g5 (t) in Figure Q11.4
(b) g1 (t) and g2 (t) in Figure Q11.4
(c) g4 (t) and g6 (t) in Figure Q11.4
(d) g1 (t) and g4 (t) in Figure Q11.3
(e) g3 (t) and g8 (t) in Figure Q11.3.

8 A binary ASK system uses two sinusoidal pulses of duration T s and amplitudes Ao and A1 = 𝛼Ao , where
Ao > 0 and 𝛼 ≥ 0. Determine:
(a) An expression for the average energy per bit Eb .
(b) An expression for the BER of the system in terms of 𝛼 and Eb /N o only, where N o is the noise power per
unit bandwidth.
(c) The value of 𝛼 that yields the lowest BER for a given Eb .

g0(t), V g3(t), V g6(t), V

5 5 5

6 6 2 4 6
t, μs t, μs t, μs

–5 –5 –5

g1(t), V g4(t), V g7(t), V

5 5 5

2
t, μs t, μs t, μs
2 6 6 4 6
Telegram: @Persian_IEEEComSoc

–5 –5 –5

g2(t), V g5(t), V g8(t), V

5 5 5

4
t, μs t, μs t, μs
4 6

–5 –5 –5

Figure Q11.3 Question 11.3.

 Questions 761

9 A binary PSK system uses two sinusoidal pulses of phases 𝜃 o and 𝜃 1 = 𝜃 o + 𝜑. Determine:
(a) An expression for the BER of the system in terms of 𝜑 and Eb /N o , where N o is the noise power per unit
bandwidth and Eb is the energy per bit.
(b) The value of 𝜑 that yields the lowest BER for a given Eb /No .
(c) Comment on the significance of 𝜃 o and 𝜑 on BER.

10 A binary modulated system transmits at 8448 kbit/s. The noise power per unit bandwidth at the receiver
input is 10−19 W/Hz and each received sinusoidal pulse has amplitude Ac = 5.2 μV. What are the transmission
bandwidth and BER when the following schemes are used?
(a) Coherent PSK.
(b) Coherent FSK.
(c) Coherent OOK, if 1’s and 0’s are equally likely.

11 A binary baseband system transmits the following pulses, shown in Figure Q11.4:
(a) g5 (t) and g1 (t) for binary 0 and binary 1, respectively.
(b) g5 (t) and g2 (t) for binary 0 and binary 1, respectively.
The pulses experience an attenuation of 140 dB up to the detection point where the noise power per unit
bandwidth is 10−17 W/Hz. Calculate the bit rate and BER of each transmission. Comment on the difference
in BER.

12 A data transmission system operates at a bit rate of 4 kb/s using an 11 kHz tone to convey binary 1 and an
8 kHz tone for binary 0. The noise power per unit bandwidth at the receiver is 83.33 nW/Hz. The transmitted
tones are of amplitude 1 V and they suffer a net loss of 20 dB along the path leading to the detection point.
Determine:

g0(t), V g3(t), V g6(t), V

10 10 10

t, μs t, μs t, μs
200 100 100 200

g1(t), V g4(t), V g7(t), V

200 100 100 200
t, μs t, μs t, μs
Telegram: @Persian_IEEEComSoc

–10 –10 –10

g2(t), V g5(t), V
10 10

200
t, μs t, μs
200

–10 –10

Figure Q11.4 Question 11.4.

 762 11 Digital Modulated Transmission

(a) The transmission bandwidth

(b) The bit error ratio (BER).

13 The noise power per unit bandwidth at the receiver of a 140 Mb/s data transmission system is 10−19 W/Hz.
Determine the minimum received average signal power in dBm that is required to achieve a maximum BER
of 10−7 using the following modulation techniques:
(a) Coherent BASK
(b) QPSK
(c) DPSK
(d) Noncoherent BFSK.

14 A transmission system is to have a maximum BER of 10−4 . The average received signal power is −60 dBm and
the noise power per unit bandwidth is 4.2 × 10−18 W/Hz. Determine the maximum bit rate that is possible
with the following modulation schemes:
(a) Coherent BASK
(b) BPSK
(c) QPSK
(d) DPSK.

15 Derive an expression for the BER of the following systems when there is a phase error 𝜑 in the incoming
signal:
(a) BPSK
(b) Coherent BASK.
Hence, determine the extra signal power required to make up for a phase error of 10∘ in each system and
prevent an increase in BER.

16 By making use of Eq. (11.62), show that the correlation coefficient of adjacent pulses gi (t) and gi + 1 (t) in
Eq. (11.88) is as given in Eq. (11.90).

17 Repeat Question 11.13 for the following modulation schemes:

(a) 16-FSK
(b) 16-PSK
(c) 16-ASK
(d) 16-APSK.

A transmission system is to have a maximum BER of 10−7 . The average received signal power is −60 dBm and
Telegram: @Persian_IEEEComSoc

18
the noise power per unit bandwidth is 4.2 × 10−18 W/Hz. Determine the maximum bit rate that is possible
with the following modulation schemes:
(a) 64-ASK
(b) 64-PSK
(c) 64-FSK
(d) 64-APSK.
 763

12

Pulse Shaping and Detection

Live optimistically and hope for the best, but design deliberately in expectation of the worst. If a bad scenario
can arise, assume in your design that it will.

In this Chapter
✓ Pulse shaping to eliminate intersymbol interference (ISI): a clear discussion of various anti-ISI filtering
techniques, including Nyquist, raised cosine, root raised cosine, and duobinary.
✓ Information capacity law: a nonmathematical introduction to the Shannon–Hartley law followed by a
detailed analysis of its implications.
✓ Digital receiver: a brief discussion of the core functions of a digital communication receiver, including a
detailed treatment of the matched filter for optimum pulse detection in the presence of additive white
Gaussian noise.
✓ Worked examples: a mix of graphical, heuristic, and mathematical approaches to further develop your
understanding and hone your problem-solving skills.

12.1 Introduction

So far, we have mostly represented transmitted symbols as rectangular pulses in baseband systems (Chapter 10)
and as sinusoidal pulses (being the product of a sinusoidal signal and a rectangular window function) in
bandpass systems (Chapter 11). However, the rectangular function, because of its sharp transitions, contains
Telegram: @Persian_IEEEComSoc

high-frequency components, which are attenuated in practical finite-bandwidth transmission systems. This
causes the transmitted pulse to spread out beyond its symbol interval, so that it contributes significant energy
into one or more adjacent symbol intervals. This phenomenon is known as intersymbol interference (ISI) and, if
not addressed, will contribute to symbol errors, as the receiver cannot discriminate between energy in the current
symbol and leftover energy from previous symbols.
In Worked Example 4.9 of Chapter 4, we explore the limiting effect of ISI on symbol rate and transmission system
capacity. It would be helpful to review this worked example before working through the rest of this chapter, which
is devoted to (i) exploring various techniques of filtering and pulse shaping to minimise ISI; (ii) gaining a deeper
insight into the interplay among various system parameters, particularly the constraints placed on bit rate and
system capacity by ISI, bandwidth, signal power, and channel noise; and (iii) understanding the specifications of
a matched filter to optimise signal detection in the presence of noise. We will carefully quantify the challenges at
Communication Engineering Principles, Second Edition. Ifiok Otung.
© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 764 12 Pulse Shaping and Detection

y(t), when B = Rs/4

y(t), when B = Rs/2

x(t)

x(t) y(t)
y(t), when B = Rs

Ideal LPF
(0 → B)
Past Current Future
intervals interval, Ts intervals
y(t), when B = 4Rs

Figure 12.1 Transmission of rectangular pulse x(t) of duration T s through ideal lowpass LTI channel of bandwidth B.
Symbol rate Rs = 1/T s ; output = y(t).

hand and then explore practical measures that can be taken at the transmitter to minimise ISI and at the receiver
to enhance the correct detection of incoming symbols in the presence of noise.
Figure 12.1 illustrates the transmission of a rectangular pulse x(t) (also called symbol) of duration T s through an
ideal lowpass linear time invariant (LTI) channel (i.e. filter – note that every channel or system is a filter of some
sort). At symbol duration T s , the number of symbols transmitted each second, or symbol rate (denoted Rs ), is
1
Rs = (symbols per second ≡ baud) (12.1)
Ts
The output y(t) of the channel is shown for various values of channel bandwidth from B = Rs /4 to B = 4Rs . We
see the occurrence of varying degrees of ISI, which may contribute to symbol detection errors at the receiver due
to the identity of the current symbol being blurred by contributions from previous symbols.
From Figure 12.1, one obvious solution to ISI would be to increase channel bandwidth B, since the trend in
Telegram: @Persian_IEEEComSoc

the plots (for B = Rs /4 to B = 4Rs ) shows that pulse spreading reduces as B is increased. This behaviour is to be
expected from the inverse relationship between time and frequency discussed in Section 4.6: the channel narrows
the bandwidth of the input pulse x(t) to B at the output, so the duration of the output pulse y(t) broadens beyond
T s in response. The amount of pulse broadening decreases as bandwidth narrowing lessens (i.e. as B becomes
larger), and vice versa. However, attempting to solve the ISI problem by increasing transmission bandwidth is
a very expensive approach, which may even be unfeasible if it requires bandwidth to be increased beyond what
is physically possible in the transmission medium. In Section 12.2 we explore effective and economic solutions
through filtering that controls pulse broadening to ensure that the received pulse has zero value at the decision
instants of adjacent intervals.
Whatever the anti-ISI filtering measure employed, the channel bandwidth must be at least wide enough to pass
the fundamental frequency f 0 of the fastest-changing sequence of transmitted pulses, as discussed in Figure 11.19
 12.2 Anti-ISI Filtering 765

in connection with the fastest changing bit sequence 101010… If the channel has enough bandwidth to pass
the fundamental sinusoidal waveform then sampling this waveform at the midpoint of each symbol interval
enables the receiver to completely identify the pulse sequence. However, if the channel bandwidth is smaller
than f 0 , the fundamental sinusoidal waveform will be blocked, making it impossible for the receiver to detect this
fastest-changing pulse sequence. All other pulse sequences will change less slowly and will therefore have a lower
fundamental frequency and hence require less bandwidth, so a bandwidth that passes f 0 is adequate for detecting
all possible pulse sequences having pulse duration T s .
For baseband transmission this means that the minimum passband must be from DC to f 0 , whereas for bandpass
transmission obtained by multiplying the baseband message signal by a sinusoidal carrier signal of frequency f c ,
it means that minimum passband must be from f c − f 0 to f c + f 0 . Since (from Figure 11.19) f 0 = 1/2T s = Rs /2, it
follows that a baseband system can support transmission at a symbol rate that is at most twice the bandwidth of
the system. That is
{
2B, Baseband system
Rsmax = (12.2)
B, Bandpass system
Furthermore, in order to convey bits, there needs to be M unique symbols each of which is used to represent
a unique combination of k = log2 M bits, the assignment of which is agreed between transmitter and receiver. In
general, the transmission system is described as M-ary, but in the special cases M = 2, 3, and 4 it is described as
binary, ternary, and quaternary, respectively. Beyond M = 3, the number of unique symbols M is always chosen
to be an integer power of 2 to ensure full utilisation of all possible bit combinations. The task of the receiver is
to identify each received symbol and then to locally generate the group of k bits represented by that symbol. One
or more bit errors occur whenever there is symbol misidentification. Each transmit symbol may be described as
possessing one of M distinct states that differ in amplitude, phase, frequency, or a combination of these three
parameters. Since each received symbol delivers log2 M bits, the maximum bit rate follows from Eq. (12.2) as

Rbmax = Rsmax log2 M = 2 Blog2 M (12.3)

Can we therefore indefinitely increase the bit rate through a channel of bandwidth B simply by increasing M? In
Section 12.3 we look at the constraint placed by noise on M and hence on the maximum bit rate that is possible for
error-free transmission using bandwidth B. We then develop in Section 12.4 the specification of a filter, known as
a matched filter, that enables the receiver to optimise its ability to correctly detect a known symbol in the presence
of noise.

12.2 Anti-ISI Filtering

Telegram: @Persian_IEEEComSoc

If we disregard the approach of attempting to provide enough transmission bandwidth to prevent any pulse spread-
ing then an outline of the required shape of the spread pulse p(t) at the receiver’s decision point is as shown in
Figure 12.2 for the nth pulse. That is, at the decision point in the receiver where incoming pulses are sampled at
a regular interval T s , the waveform of p(t) should pass through the points marked with an asterisk on the dia-
gram, having a maximum absolute value at the decision instant of its own interval and zero value at the decision
instants of adjacent intervals. Provided the receiver follows the correct timing, we don’t care much about the val-
ues of p(t) at nonsampling time instants. To explore this concept of controlled pulse spreading in more depth,
consider Figure 12.3, which shows a baseband transmission system in which a baseband signal x(t) produced by
a symbol generator (in this case a line coder) at point (a) in a transmitter is decided upon at point (b) in a receiver.
The baseband signal consists of a sequence of pulses spaced T s seconds apart, which means that the symbol rate
is Rs = 1/T s . The worst-case situation for pulse spreading occurs when the pulses have very narrow widths so that
 766 12 Pulse Shaping and Detection

p+(t – nTs)
+1 *

(a) 0 * * * * t

–1
p–(t – nTs)
+1

(b) 0 * * * * t

–1 *
(n – 2)Ts nTs (n + 2)Ts Sampling instants
(n – 1)Ts (n + 1)Ts

Figure 12.2 Outline of pulse shape to avoid ISI: (a) positive symbol; (b) negative symbol.

Data in (a) Baseband (b)

Line Decision Data out
transmission
1101… coder x(t) system y(t) y(nTs) device

Sample at t = nTs

x(t)
2Ts
(a) t
0 Ts 3Ts
Ts

y(t)
Telegram: @Persian_IEEEComSoc

(b) t

Sampling instants at decision point at receiver

Figure 12.3 Zero-ISI pulse spreading: The sequence of narrow pulses at point (a) in the transmitter is spread in the
transmission system to become a sequence of sinc pulses at point (b) in the receiver.
 12.2 Anti-ISI Filtering 767

the nth pulse can be approximated by an impulse of weight bn which carries the identity of the nth bit. For example,
bn = 1 for binary 1 and bn = −1 for binary 0. Note that the discussion that follows is equally applicable to M-ary
transmission, in which case bn takes on values drawn from a discrete set of M amplitudes. Thus
∑
x(t) = bn 𝛿(t − nT s )
n

Due to the finite bandwidth of the transmission system from symbol generator output at point (a) to decision
point (b) at the receiver, the baseband signal y(t) arriving at point (b) will be a sequence of spread pulses. Normal-
ising the transmission path to unit gain and zero delay, we may write
∑
y(t) = bn h(t − nT s )
n

Since h(t) is the output pulse in response to an input impulse 𝛿(t), it represents the impulse response of the entire
transmission path from point (a) to point (b).
At the detection point, y(t) is sampled at a regular interval T s in step with the transmitter and the sample value
y(nT s ) is passed to a decision device which compares it to one threshold level (in the case of binary transmission) or
multiple levels (for M > 2) to reach a decision as to which bit(s) were transmitted in that interval. Thus, to avoid ISI
it is necessary and sufficient for the nth pulse h(t − nT s ) at the detection point to have (normalised) unit value at its
own sampling instant t = nT s , and a value of zero at all other sampling instants …, (n − 2)T s , (n − 1)T s , (n + 1)T s ,
(n + 2)T s , … This means that we require that at the sampling instant t = mT s the contribution bn h(mT s − nT s )
from h(t − nT s ) to the sample output y(mT s ) should be zero in all cases except m = n when the contribution equals
bn . Stated mathematically
{
1 for m = n
h(mT s − nT s ) = (12.4)
0 otherwise

So long as there is correct timing at the receiver, the values and hence waveform of h(t − nT s ) between sampling
instants are immaterial. Equation (12.4) is the Nyquist criterion for zero ISI which allows a number of solutions,
as discussed in the following subsections.

12.2.1 Nyquist Filtering

A sinc impulse response

h(t) = sinc(t∕Ts ) (12.5)

satisfies Eq. (12.4) since

{
Telegram: @Persian_IEEEComSoc

( )
mT s − nT s 1 for m = n
sinc = sinc(m − n) = (12.6)
Ts 0 otherwise

This is illustrated in Figure 12.3 for a binary impulse sequence in (a) corresponding to a bit sequence 1101… We
see that among the received spread pulses arriving at (b) only the current pulse has a nonzero value at the current
sampling instant. This means that if the impulse response of the transmission path from symbol output point at
transmitter to symbol sampling point at receiver is as given by Eq. (12.5) then ISI-free operation is guaranteed in a
properly synchronised receiver. From row 9 of Table 4.5, the Fourier transform (FT) of Eq. (12.5) yields the transfer
function of an ISI-free channel as
( )
1 f
H(f ) = Ts rect (fTs ) = rect (12.7)
Rs Rs
 768 12 Pulse Shaping and Detection

h(t) = sinc(t/Ts)
1

(a)

t
–5Ts –4Ts –3Ts –2Ts –Ts 0 Ts 2Ts 3Ts 4Ts 5Ts

H(f) = Tsrect(f/Rs)

(b) Ts

f
–Rs/2 Rs/2

Figure 12.4 (a) Impulse response h(t) and (b) transfer function H(f ) of ideal Nyquist channel for zero-ISI.

The impulse response h(t) and transfer function H(f ) are plotted in Figure 12.4, which makes clear that the
channel specified by H(f ) is an ideal lowpass filter (LPF) of bandwidth
Rs
B= (12.8)
2
known as the Nyquist bandwidth for ISI-free transmission at symbol rate Rs . Such a channel is called the ideal
Nyquist channel and allows transmission at a symbol rate, called the Nyquist rate, which is twice the available
bandwidth. This is the very best that we can do without incurring ISI. That is

Rsmax = 2B

which you may recognise as Eq. (12.2) in the introduction. The passband gain of 1/Rs is not significant to the
scheme of ISI elimination and may be normalised to unit gain. It is important to emphasise that Eq. (12.7) does
Telegram: @Persian_IEEEComSoc

not necessarily prescribe a single filter for the transmission system; rather, it lays down the filtering characteristic
which must be satisfied by the entire transmission path from symbol output point at the transmitter to the sampling
point at the receiver. Typically this path will consist of several filters, in which case Eq. (12.7) specifies what the
product of their transfer functions must be for ISI-free operation.
There are, however, two practical problems with this solution to ISI, namely:

● It requires an overall filtering action that produces constant gain up to frequency f = Rs /2, and infinite attenu-
ation beyond this cut-off frequency. Such a sharp transition cannot be achieved in real-time because it requires
noncausality (see Section 2.10.3) whereby future inputs contribute to current output. Thus, our ideal scenario
of ISI-free transmission at a symbol rate that is twice the channel bandwidth is practically unrealisable in real
time.
 12.2 Anti-ISI Filtering 769

● The envelope of the sinc pulse sinc(t/T s ) decays very slowly with t as 1/|t| which leaves significant residual
energy in adjacent intervals, a contribution that is only avoided if we sample at precisely the right timings t = 0,
T s , 2T s , 3T s , … The Nyquist channel therefore imposes a stringent requirement on timing accuracy. Any timing
error leads to significant contribution to each sampled output from previous and future (yes, future!) sinc pulses
at each mistimed sampling instant. That the future affects the present only serves to highlight that a sinc pulse
is not of this world (i.e. it is not possible to implement it in real time)!

12.2.2 Raised Cosine Filtering

We may address the problem of residual energy highlighted above by introducing a factor which forces a more
rapid decay of the impulse response envelope. Now consider the set of impulse response functions
[ ]
cos(𝜋𝛼t∕Ts )
h(t) = sinc(t∕Ts ) (12.9)
1 − 4(𝛼t∕Ts )2
where 0 ≤ 𝛼 ≤ 1 is a dimensionless parameter. These all satisfy the Nyquist criterion for zero ISI stated in Eq. (12.4)
since
( )
mT s − nT s
cos 𝜋𝛼 ( )
Ts mT s − nT s
h(mT s − nT s ) = ( ) sinc
mT s − nT s 2 Ts
1 − 4𝛼 2
Ts
⎧
cos[(m − n)𝜋𝛼] ⎪1, m=n
= sinc(m − n) = ⎨
1 − 4𝛼 2 (m − n)2 ⎪0 otherwise
⎩
This can be readily seen in Figure 12.5a, where the impulse responses, plotted for three values of the parameter
𝛼 = 0, 0.5, 1, have zero crossings at all nonzero integer multiples of T s . Notice that, when 𝛼 = 0, Eq. (12.9) reduces
to the sinc impulse response of Eq. (12.5) and the rate of decay of the tails of the function is left unchanged. At
the other end of the scale, 𝛼 = 1 gives the fastest rate of decay of the tails of the impulse response function so that
it contains negligible energy outside its main lobe and therefore contributes negligible ISI at slightly mistimed
sampling instants. By now we have come to expect that this gradual narrowing of the duration of the impulse
function as 𝛼 is increased from 0 to 1 must come at the price of bandwidth broadening.
Taking the FT of Eq. (12.9) yields the corresponding transfer function H(f ) plotted in Figure 12.5b. The analytic
form of this transfer function is
Telegram: @Persian_IEEEComSoc

⎧1, |f | ≤ f1
⎪ [ ( )]
1 ⎪1 |f | − f1
H(f ) = × ⎨ 1 + cos 𝜋 , f1 ≤ |f | ≤ f2
Rs ⎪ 2 f2 − f1
⎪0, |f | ≥ f2
⎩
f1 = (1 − 𝛼)Rs ∕2; f2 = (1 + 𝛼)Rs ∕2; 0≤𝛼≤1 (12.10)

For 𝛼 > 0, this spectrum represents an LPF having a constant gain portion up to f = f 1 , followed by a portion
in which the gain decreases gradually until it reaches zero at f = f 2 . The decrease in gain with frequency in the
interval f 1 ≤ f ≤ f 2 follows a cosine function to which the number 1 is added to limit the result to nonnegative
values. This filter is therefore called a raised cosine filter. Its null bandwidth (≡ occupied bandwidth) is obtained
 770 12 Pulse Shaping and Detection

h(t)
1

(a) α=0
α=1

0 t

α = 0.5
–0.4
–4Ts –3Ts –2Ts –Ts 0 Ts 2Ts 3Ts 4Ts
H(f)
1/Rs α=0

α = 0.5
(b)
α=1

0 f
–Rs –Rs/2 0 Rs/2 Rs

Figure 12.5 (a) Impulse response h(t) and (b) transfer function H(f ) of raised cosine filter for zero-ISI, at various values of
roll-off factor 𝛼.

in Worked Example 4.16 as

⎧ Rs
⎪(1 + 𝛼) , Baseband
Bocc =⎨ 2 (12.11)
⎪(1 + 𝛼)Rs , Bandpass
⎩
and it can be seen that at 𝛼 = 1 the (baseband) raised cosine filter has occupied bandwidth Bocc = Rs which is
double the bandwidth of the unrealisable Nyquist channel, so that as expected the reduction in the tails of the
filter’s impulse response (to make the filter realisable and to improve its tolerance to timing error) is achieved at
the expense of an increase in transmission bandwidth.
Other noteworthy features of the raised cosine filter include:
● The parameter 𝛼 is the roll-off factor which controls the size of the frequency interval f 2 − f1 = 𝛼Rs over which
Telegram: @Persian_IEEEComSoc

the filter transitions from maximum normalised unit gain in the passband to zero gain in the stopband. 𝛼 = 0
corresponds to the special case of the Nyquist filter having no transition band at all, whereas 𝛼 = 1 gives the
full-cosine roll-off filter having the most gradual roll-off or transition that starts with unit normalised gain at
f = 0 and reaches zero gain at f = Rs .
● The raised cosine filter gain response exhibits antisymmetry about the vertical at the Nyquist bandwidth Rs /2.
This means that starting at f = Rs /2 the gain response increases by the same amount when one moves to the left
along the frequency axis by, say, Δf as it decreases when one moves by Δf to the right.
● Strictly speaking, the raised cosine filter is unrealisable because it exhibits zero gain or infinite attenuation at
f = f 2 and beyond. However, its gradual roll-off makes the raised cosine filter characteristic easier to approximate
than the ideal Nyquist filter using a realisable tapped delay line (also known as finite-duration impulse response
(FIR)) filter. The required number of taps and hence filter complexity increase as 𝛼 decreases.
 12.2 Anti-ISI Filtering 771

● Putting 𝛼 = 1 in Eq. (12.9) and simplifying yields the impulse response of the full-cosine roll-off filter as
sin(2𝜋t∕Ts )
h(t) = (12.12)
2𝜋t∕Ts [1 − 4(t∕Ts )2 ]

This has zero crossings at 2𝜋t∕Ts = n𝜋, n = · · · , ±3, ±2, 2, 3, · · · or

Ts
t=n , n = · · · , ±3, ±2, 2, 3, · · ·
2
which means that, beyond the main lobe, this pulse has double the zero-crossing rate of the sinc pulse, a feature
which is very useful in extracting timing information on T s for synchronisation.

12.2.3 Square Root Raised Cosine Filtering

A practical arrangement for zero-ISI transmission is as shown in Figure 12.6a. We assume that equalisation has
been employed to eliminate channel distortion as illustrated in Figure 4.59 and discussed in Section 4.7.4. The
channel is therefore shown as distortionless. Figure 12.6a is applicable to both baseband and bandpass transmis-
sion systems. In a bandpass system the line coder and line decoder blocks are replaced by a modulator and a
demodulator, respectively, as shown. The transmit and receive filters serve fundamentally different roles, the for-
mer being required for shaping the pulses generated by the line coder or modulator in order to reduce ISI, whereas
the latter helps to reduce the impact of noise on symbol detection.
Assuming (as before) for convenience that the line coder produces a weighted impulse in each signalling interval,
the output of the transmit filter in each interval is the impulse response hx (t) also proportionately weighted. This
pulse passes through the distortionless channel and arrives along with additive white Gaussian noise (AWGN)
w(t) at the input of the receive filter. We show in Section 12.4 that the optimum detection of hx (t) in the presence
of w(t) is achieved when the receive filter is a matched filter having an impulse response that is a time-reversed

(a)
Data Data
Transmit Distortion- Receive y(t)
in Line Line out
filter less filter
coder hx(t) ⇔ Hx(f) channel hy(t) ⇔ Hy(f) decoder
t = nTs
or Noise, w(t) or
Modulator Demodulator
(BP system) (BP system)
Telegram: @Persian_IEEEComSoc

(b)
Distortionless
Data Data
in y(t) out
Line RRC Hc(f) He(f) RRC Line
coder filter filter decoder
t = nTs
or Noise, w(t) or
Modulator Demodulator
(BP system) (BP system)

Figure 12.6 Arrangement for zero-ISI transmission in baseband or bandpass (BP) systems.
 772 12 Pulse Shaping and Detection

and delayed version of hx (t). That is

hy (t) = hx (Ts − t) (12.13)
Taking the FT of both sides, noting (from Eq. (4.86)) that time-reversing a real signal has the effect of
complex-conjugating its FT and (from Eq. (4.80)) that a delay of T s in the time domain corresponds to a factor
exp(−j2𝜋fT s ) in the frequency domain, we obtain the gain response of the receive filter as
|Hy (f )| = |Hx∗ (f ) exp(−j2𝜋fTs )| = |Hx∗ (f )||exp(−j2𝜋fTs )|
= |Hx (f )| (12.14)
Normalising the transmission path to unit gain and zero delay and ignoring w(t) – since our focus in this section
is solely on ISI, the output pulse y(t) at the sampling point of the receiver in response to an impulse 𝛿(t) generated
by the line coder is
y(t) = 𝛿(t) ∗ hx (t) ∗ hy (t)
= hx (t) ∗ hx (Ts − t) ≡ h(t) (12.15)
where we have used the fact – see Eq. (4.92) – that convolving a signal with 𝛿(t) leaves the signal unchanged, and
that for zero ISI, y(t) must be the raised cosine filter impulse response h(t) given in Eq. (12.9) with transfer function
H(f ) given in Eq. (12.10). Taking the FT of Eq. (12.15) and retaining the magnitude yields
|H(f )| = |F[hx (t) ∗ hx (Ts − t)]|
= |Hx (f )Hx∗ (f ) exp(−j2𝜋fTs )|
= |Hx (f )|2 (12.16)
Thus, in view of Eq. (12.14), we may state that
√
|Hx (f )| = |Hy (f )| = |H(f )|
√
≡ Raised cosine filter gain response
≡ |HRRC (f )| (12.17)
That is, the requirements of zero ISI and optimum detection in the presence of white noise may be jointly satisfied
by using a pair of identical square root raised cosine (RRC) filters, one at the transmitter to shape the output pulses
of the line coder and the other as the matched filter at the receiver. Figure 12.6b shows the transmission block
diagram based on this remarkable result. Using Eq. (12.10) in Eq. (12.17) yields the transfer function H RRC (f ) of
the RRC filter as
⎧1, |f | ≤ f1
⎪ ( )
⎪ 𝜋 |f | − f1
Telegram: @Persian_IEEEComSoc

1
HRRC (f ) = √ × ⎨cos , f1 ≤ |f | ≤ f2
Rs ⎪ 2 f2 − f1
⎪0, |f | ≥ f2
⎩
f1 = (1 − 𝛼)Rs ∕2; f2 = (1 + 𝛼)Rs ∕2; 0≤𝛼≤1 (12.18)
The inverse FT of this expression gives the impulse response hRRC (t) as follows
√ [ ( ) ( )]
1∕ Ts t 4𝛼 t
hRRC (t) = (1 − 𝛼)sinc (1 − 𝛼) + cos 𝜋(1 + 𝛼) (12.19)
1 − (4𝛼t∕Ts )2 Ts 𝜋 Ts
The RRC filter impulse response and transfer function are plotted in Figure 12.7, and, like the raised cosine type,
may be closely approximated using a tapped delay line. Notice that, except for the case 𝛼 = 0, the zero crossings of
 12.2 Anti-ISI Filtering 773

hRRC(t)
1.3 Rs
α=1

(a)
α=0
α = 0.5
0 t

–1.3 Rs
–3Ts –2Ts –Ts 0 Ts 2Ts 3Ts
HRRC(f)
1/ Rs
α=0

α = 0.5
(b) α=1

0 f
–Rs –Rs/2 0 Rs/2 Rs

Figure 12.7 (a) Impulse response hRRC (t) and (b) transfer function HRRC (f ) of root raised cosine filter at various values of
roll-off factor 𝛼.

the RRC impulse response are not at integer multiples of T s . Therefore, a single RRC filter cannot eliminate ISI,
although it has the same bandwidth as its raised cosine counterpart. RRC filters must be used in pairs, as shown
in the block diagram of Figure 12.6b so that after passing through both filters a transmitted pulse will arrive at the
sampling point of the receiver having been filtered by the equivalent of one raised cosine filter causing it to have
zero crossings at nonzero integer multiples of the sampling interval T s , which therefore averts ISI.
A few words are in order on how the channel equaliser H e (f ) shown in Figure 12.6b might influence the speci-
fication of the transmit and receive filters. Channel attenuation tends to increase with frequency, so the equaliser
would need to have a gain that increases with frequency in order to compensate. If that is the case then the noise
reaching the receive filter, having passed through the equaliser, will no longer be white but will be ‘coloured’,
having an amplitude spectrum that increases with frequency. Under this condition, a matched filter is one that
attenuates the higher-frequency components more drastically in order to reduce noise, maximise SNR, and opti-
Telegram: @Persian_IEEEComSoc

mise symbol detection. The RRC receive filter must therefore be modified accordingly. However, this will also
attenuate the desired pulse energy at these frequencies, so the RRC transmit filter must be modified to propor-
tionately boost the high-frequency components of the pulse in preparation for their increased attenuation in the
receive filter. In this way the combination of transmit and receive filters still yields a raised cosine filter and hence
ISI-free pulses at the decision point, and there is also optimum performance in the presence of noise, but the
transmit and receive filters are no longer identical RRC filters, as specified in Eq. (12.18).
It is worth noting that the well-known scheme of pre-emphasis and de-emphasis employed in analogue fre-
quency modulation (FM) transmission (Chapter 8) is based on the above principle. Noise at the output of the
frequency discriminator is coloured, increasing as the square of frequency. So, an LPF (called a de-emphasis filter)
is placed at the output of the FM demodulator to attenuate high-frequency components more drastically in order
to reduce noise. And in preparation for this ‘controlled distortion’ of the message signal, a highpass filter (called
 774 12 Pulse Shaping and Detection

a pre-emphasis filter) is used at the transmit end to proportionately boost the high-frequency components of the
message.

12.2.4 Duobinary Signalling

Note from the outset that this is an entirely different technique that should not be confused with the raised cosine
filtering technique discussed above. A raised cosine filter averts ISI by operating on each pulse individually and
independently to shape it into having zero-crossings at nonzero integer multiples of sampling interval at the detec-
tion point in the receiver. The duobinary signalling technique, on the other hand, averts ISI at the receiver by
combining two or more adjacent pulses at the transmitter, thereby introducing a controlled amount of ISI. To bor-
row from medicine, the technique cures the ISI disease by vaccinating each transmitted pulse with a controlled
amount of ISI.
The technique of zero-ISI transmission discussed here is variously described in the literature as partial response
signalling or correlative coding, and was first introduced by Adam Lender in 1963 [1] as duobinary signalling,
the prefix duo- indicating a doubling in bit rate (or operation at half the bandwidth) when compared to binary
transmission using a raised cosine filter of roll-off 𝛼 = 1. Our discussion will assume all filtering operation to be
based at the transmitter, but actual systems share filtering between the transmitter and the receiver [2]. In this
case, a procedure like our presentation on the square RRC filter is followed to derive filter specifications for the
transmitter and receiver necessary to achieve the required overall filter characteristic. Different extensions and
modifications to the basic duobinary signalling scheme are possible, all based on the same principle of introducing
a controlled ISI into the transmitted pulses in order to gain various advantages [3]. We will discuss in detail the
basic duobinary implementation which leads to the cosine filter, introduce a variant called modified duobinary that
leads to the sine filter, comment briefly on polybinary signalling, and discuss the trade-off involved in this class of
zero-ISI techniques.

12.2.4.1 Cosine Filter

Consider the arrangement shown in Figure 12.8a. We wish to determine the impulse response h(t) of the filter
comprising the highlighted portion of the block diagram. The filtering process involves summing the pulses in the
current and previous signalling intervals and then passing this combined pulse through an ideal LPF. But don’t
worry about the ideal LPF because we don’t intend to build it. What we will build is the overall highlighted filter
having impulse response h(t) and transfer function H(f ) to be determined by making the input an impulse 𝛿(t) as
shown.
The input to the ideal LPF is
x(t) = 𝛿(t) + 𝛿(t − Ts )
Taking the FT of this equation gives the signal spectrum at the input of the ideal LPF as
Telegram: @Persian_IEEEComSoc

X(f ) = 1 + exp(−j2𝜋fTs )
Multiplying this by the transfer function of the ideal LPF (which is T s rect(fT s ) – see Eq. (12.7) and
Figure 12.4b) – yields the spectrum at the output of the ideal LPF as
H(f ) = [1 + exp(−j2𝜋fTs )]Ts rect(fTs )
= Ts rect(fTs ) + Ts rect(fTs ) exp(−j2𝜋fTs ) (12.20)
Noting that H(f ) consists of one rectangular function and another that is scaled by the factor exp(−j2𝜋fT s ) which
corresponds to a delay T s in the time domain, we see that the inverse FT, which gives h(t), will comprise an impulse
function and a delayed impulse function. That is
( ) ( )
t t − Ts
h(t) = sinc + sinc (12.21)
Ts Ts
 12.2 Anti-ISI Filtering 775

cosine filter

δ(t) x(t) Ideal h(t) ⇌ H(ƒ)

LPF
(a)
Delay δ(t – Ts)
Ts

4 h(t) |H(f)|
2Ts
π

t
0 f
0
–4Ts –3Ts –2Ts –Ts 0 Ts 2Ts 3Ts 4Ts –Rs –Rs/2 0 Rs/2 Rs
(b) (c)

Figure 12.8 The cosine filter: (a) block diagram; (b) impulse response; (c) transfer function.

We may manipulate Eq. (12.20) into the following more familiar form for the gain response of the filter

|H(f )| = |[ej𝜋fTs + e−j𝜋fTs ]e−j𝜋fTs Ts rect(fTs )|

= |2Ts cos(𝜋fTs )rect(fTs )e−j𝜋fTs |

= 2Ts cos(𝜋fTs )rect(fTs )

⎧2 Rs R
⎪ cos(𝜋f ∕Rs ), − ≤f ≤ s
= ⎨ Rs 2 2 (12.22)
⎪0, Otherwise
⎩
These impulse response and gain response functions are plotted in Figure 12.8b,c. The gain of the filter varies
according to a cosine function and therefore the filter is known as a cosine filter. It is remarkable that it has exactly
Telegram: @Persian_IEEEComSoc

the Nyquist bandwidth of Rs /2, with zero gain above this cut-off frequency. However, unlike the Nyquist filter, the
cosine filter does not have a sharp transition from passband to stop band so it may be more readily synthesised
using a tapped delay line to achieve a very high but noninfinite attenuation beyond f = Rs /2. It can also be seen that,
whereas the raised cosine filter impulse response (see Figure 12.5a) is nonzero only at one sampling point nT s = 0,
the cosine filter impulse response has nonzero value h(t) = 1 at two sampling points t = nT s for n = 0, 1. This shows
that energy in one input pulse has been spread over two signalling intervals, which is indicative of the controlled
ISI introduced. Furthermore, unlike the Nyquist filter, the cosine filter impulse response decays rapidly beyond
its main lobe, making the transmission system more tolerant to slight errors in the setting of sampling instants.
We will discuss the use of a cosine filter to facilitate zero-ISI transmission based on the block diagram shown
in Figure 12.9. The precoder section in the transmitter, which performs modulo-2 summing of the current bit and
the previous output to produce the current output, is essential in every application of the cosine filter in order to
 776 12 Pulse Shaping and Detection

Precoder Line coder Decoder

Data
in Channel
Binary
Cosine yn |yn| Data
mn XOR bipolar || – out
dn bnδ(t) filter y(t)
coder t = nTs m̂n
dn–1 +
Noise +1
Delay
Ts

Figure 12.9 Block diagram of zero-ISI transmission system using a cosine filter. The pre-coder section is essential and must
have the structure shown, but the line coder and corresponding decoder can be of a design other than the ones shown here.

prevent error propagation at the receiver whereby an error in one bit corrupts future decoding decisions. Cosine
filters can be used with any M-ary symbol generator, but here we will employ a binary bipolar line coder, which
produces an impulse of weight bn = +1 for binary 1 and weight bn = −1 for binary 0. To understand the operation
of the system we will examine how the message sequence mn = 100101 is handled, where m1 = 1, m2 = 0, m3 = 0,
…, are the message bits in the first, second, third signalling intervals, etc. The processing of this bit sequence by
the transmission system is shown in the table of Figure 12.10, where the first column is the signalling interval n;
the second column is the message sequence mn ; the third column is the previous output dn−1 of the precoder; the
fourth column is the current output dn of the precoder, which is also the line coder input; the fifth column is the
line coder output and cosine filter input bn ; y(t) is the waveform at the output of the cosine filter (plotted in the
graph of Figure 12.10), which is sampled at t = nT s to obtain the sample yn listed in the seventh column; and the
last column is the recovered message sequence m ̂ n . For comparison, we also show the waveform z(t) that would
reach the sampling point if the message sequence was transmitted through a raised cosine filter of roll-factor 𝛼 = 1
instead, first by applying it directly to the line coder – a raised cosine filter does not require precoding – and then

n mn dn–1 dn bn zn yn m̂n
0 0 (start-up) 0 (pre-set) 0 –1
1 1 0 1 1 1 0 1
2 0 1 1 1 –1 2 0
3 0 1 1 1 –1 2 0
4 1 1 0 –1 1 0 1
5 0 0 0 –1 –1 –2 0
6 1 0 1 1 1 0 1

2
y(t)
z(t)
Telegram: @Persian_IEEEComSoc

t
0

–1

–2
Sampling instants n = 0 1 2 3 4 5 6

Figure 12.10 Operation of block diagram of Figure 12.9 for input data = 100101. The waveform reaching the sampling
point is y(t) for a cosine filter and z(t) for a raised cosine filter.
 12.2 Anti-ISI Filtering 777

passing the line coder output through the raised cosine filter. This waveform z(t) is sampled at t = nT s to obtain
the sequence zn listed in the sixth column of the table. The decision device that converts zn into the recovered
message sequence (not shown in Figure 12.10) is a comparator that decides in favour of binary 1 if zn > 0 and in
favour of binary 0 otherwise.
Still on Figure 12.10, at start-up (signalling interval n = 0), there is no previous output bit d−1 , so we define a
start-up bit m0 = 0, which is not part of the message and combine this with d−1 = 0 to obtain d0 = 0. The rest of the
precoding then proceeds with message bits mn combined with the previous output (column 3), which is obtained
by translating from column 4, as shown by the arrows in the table. The line coder converts dn to normalised
impulses −𝛿(t) for dn = 0, and 𝛿(t) for dn = 1. Thus, the impulses have (normalised) weights ±1 given by bn in the
Table. This sequence of impulses is processed by the cosine filter to produce the waveform y(t) at the sampling
point. Note that the channel does add noise to y(t), which has been ignored in Figure 12.10 since we are here
primarily concerned with ISI. Sampling y(t) at sampling instants nT s , n = 1, 2, …, 6, yields yn , which has one of
three possible values: −2, 0, +2. The correct decoding decision rule is that if yn = ±2 then m ̂ n = binary 0, but if
yn = 0 then m̂ n = binary 1. In practice, noise will make yn vary somewhat from these precise values, so the decoder
is implemented as a rectifier followed by a comparator that compares |yn | to a (normalised) threshold level of +1
and outputs binary 0 if |yn | > 1, and binary 1 otherwise. In this way the original message sequence is correctly
recovered.

12.2.4.2 Signal Power Trade-off

We found earlier that we could avert ISI by using a realisable raised cosine filter of roll-off factor 𝛼 > 0. The price
we paid then was in using a larger bandwidth B = (1 + 𝛼)Rs /2 than the minimum that is theoretically possible,
namely the Nyquist bandwidth Rs /2. However, by using a cosine filter we can operate at Nyquist bandwidth using
a realisable filter free from ISI-induced error. There is always a price or trade-off in engineering, so what is the
cosine filter trade-off? To understand this trade-off, look again at Figure 12.10 and notice that the raised cosine
filter converts a binary input ±1 from a line coder into output z(t) from which the sampled sequence zn is obtained,
having two possible values ±1 (normalised). In the presence of noise these two values will of course map into two
intervals separated by a threshold level chosen to be equal to the mean level (which would be zero in this case,
assuming that binary 1 and binary 0 are equally likely in the message sequence mn ). The cosine filter, on the other
hand, converts the same binary input into output waveform y(t) which when sampled yields a ternary sequence yn
having values drawn from the alphabet {−2, 0, 2}. In other words, the raised cosine filtering technique maintains
binary signalling (or whatever M-ary signalling is delivered to it by the line coder used), whereas the cosine filtering
technique transforms binary signalling into ternary, or more generally M-ary signalling into (2M − 1)-ary. This
happens because a cosine filter introduces a controlled ISI between the current and previous pulses, so that if
these had binary levels ±1 the possible outcomes will be adding two positive pulses, or two negative pulses, or two
opposite polarity pulses, which leads to the output levels +2, −2, or 0, respectively. Precoding allows each level to
be linked to a unique identity of the bit in the current interval, thus making instantaneous decoding possible as
Telegram: @Persian_IEEEComSoc

illustrated in Figure 12.10.

It was important to explain the transformation of M-ary signalling into (2M − 1)-ary signalling by a cosine filter
before stating the consequence, which is that M-ary signalling is more robust to channel noise than (2M − 1)-ary
signalling at the same signal power and frequency. You may recall from the discussion of M-ary modulation in
Chapter 11 that the states of M-ary amplitude shift keying (ASK), phase shift keying (PSK), and amplitude and
phase shift keying (APSK) are closer together in signal space as M increases. This means, for example, that channel
noise added to a pulse sent with amplitude in the midpoint of one interval can more readily cause the pulse to be
received with amplitude in an adjacent interval, giving rise to bit errors. To maintain the same bit error ratio (BER)
as M increases, we must increase signal amplitudes and hence signal power in order to preserve a constant spacing
between states. These comments are also applicable to PSK where the states have equal amplitude but different
phases and can be treated as being arranged on a circle centred at the origin, with radius equal to normalised signal
 778 12 Pulse Shaping and Detection

amplitude. As the number of states on the circle increases, the only way to maintain a constant distance between
the states is to increase the radius of the circle and hence signal amplitude and power.
Cosine filtering therefore eliminates ISI while operating at Nyquist bandwidth but requires more signal power
in order to ensure that its ternary transmission (assuming a binary line coder) has the same BER as a zero-ISI
binary transmission using a raised cosine filter at larger bandwidth. So this is further evidence of a fact emphasised
throughout this book that engineering just does not do free lunches: in building a realisable ISI-free system using a
raised cosine filter, the price we pay is more bandwidth than the minimum, but in building the system at the mini-
mum bandwidth based on a cosine filter, the price we pay is more signal power (for the same transmission quality).

12.2.4.3 Sine Filter

Consider a modification to the basic scheme discussed above whereby in Figure 12.8 we increase the delay to 2T s
and subtract (rather than add) this delayed element, as shown in Figure 12.11a. The result is a sine filter with
impulse response
( ) ( )
t t − 2Ts
h(t) = sinc − sinc (12.23)
Ts Ts
and gain response, derived using a similar manipulation as for the cosine filter, given by
⎧2 Rs R
⎪ |sin(2𝜋f ∕Rs )|, − ≤f ≤ s
|H(f )| = ⎨ Rs 2 2 (12.24)
⎪0, Otherwise
⎩
These are shown in Figure 12.11b and c. We see that the sine filter has a spectral null at DC and its output pulse
is therefore well suited to capacitor- or transformer-coupled channels that block DC. This sine filter is referred to

sine filter

δ(t) + x(t) Ideal h(t) ⇌ H(ƒ)

– LPF
(a)
Delay δ(t – 2Ts)
2Ts

h(t) |H(f)|
2Ts
1
Telegram: @Persian_IEEEComSoc

t
0

–1 f
0
–5Ts –4Ts –3Ts –2Ts –Ts 0 Ts 2Ts 3Ts 4Ts 5Ts –Rs –Rs/2 0 Rs/2 Rs
(b) (c)

Figure 12.11 The sine filter: (a) block diagram; (b) impulse response; (c) transfer function.
 12.2 Anti-ISI Filtering 779

in the literature as a modified duobinary filter. Notice in Figure 12.11 that the impulse response of the sine filter
has nonzero value h(t) = ±1 at two sampling points t = nT s for n = 0, 2, which indicates that, in this variant of the
scheme, controlled ISI is introduced between each pulse and its next but one subsequent neighbour. The impulse
response of the sine filter also decays rapidly beyond its twin main lobe which minimises ISI in the event of small
timing errors at the receiver.

12.2.4.4 Polybinary Signalling

Another modification to the basic scheme discussed here leads to so-called polybinary signalling in which the
delay element in Figure 12.8 is replaced by a tapped delay line so that controlled ISI is introduced amongst multiple
adjacent pulses resulting in multilevel signalling and hence further improvement in bandwidth efficiency without
ISI, but at the price of a larger signal power requirement. See reference [4] for a discussion of the application of
polybinary signalling in optical transmission systems.

Worked Example 12.1

(a) Determine the occupied bandwidth of a 6B4T baseband system that transmits at 139264 kb/s using pulses
with a full-cosine roll-off characteristic.
(b) If the bandwidth requirement is to be reduced to 60 MHz, calculate the roll-off factor of the raised-cosine
filter.
(c) Suggest how the same bit rate may be transmitted using less than the bandwidth of the ideal Nyquist chan-
nel.

(a) We determine in Worked Example 10.6 that the symbol rate for this baseband system is Rs = 92.84 MBd. Pulses
with a full-cosine roll-off characteristic are produced by a raised cosine filter with roll-off factor 𝛼 = 1, which,
from Eq. (12.11) for a baseband system has bandwidth
Rs R
Bocc = (1 + 𝛼) = (1 + 1) s = Rs = 92.84 MHz
2 2
(b) Making 𝛼 the subject of Eq. (12.11) and putting Bocc = 60 MHz and Rs = 92.84 MBd, we obtain
2Bocc 2 × 60
𝛼= −1= −1
Rs 92.84
= 0.293

(c) If an ideal Nyquist channel (𝛼 = 0) is used then the bandwidth can be reduced to Rs /2 = 46.42 MHz. This is
Telegram: @Persian_IEEEComSoc

the minimum bandwidth required to transmit at Rs = 92.84 MBd. The only way to reduce the bandwidth any
further without incurring significant impairment due to ISI is by reducing the symbol rate, and this would
require changing the coding scheme if the same bit rate Rb is to be maintained. In the 6B4T system, each
ternary symbol carries 6/4 = 1.5 bits. We must adopt a scheme in which each symbol carries k > 1.5 bits. That
is, we represent a block of k bits using one code symbol. Clearly, we require M = 2k unique symbols to cover all
possible k-bit input blocks. Following the naming convention for block codes (Chapter 10), this coding scheme
may be identified as kB1M, where M denotes M-ary (just as T denotes ternary in, e.g. 6B4T). In general, using
M unique symbols (sometimes called levels) allows us to represent up to k = log2 (M) bits per symbol, which
gives a symbol rate
Bit Rate Rb
Rs = ≡ (12.25)
k k
 780 12 Pulse Shaping and Detection

For example, if M = 16, we have k = 4, and a symbol rate Rs = Rb ∕4 = 139264000∕4 = 34.82 MBd can be used in
this example. With this coding scheme, we can (ideally) transmit at 139264 kb/s using a transmission bandwidth
of only 17.41 MHz. In theory, we can reduce transmission bandwidth indefinitely by correspondingly increasing k.
So why, you must be asking, are kB1M block codes (called M-ary modulation in modulated systems) not used
in baseband systems? There are two main reasons. Codec complexity increases dramatically, and the unique code
symbols become so close in identity that it is difficult to correctly detect them at the receiver in the presence of
noise. As a result, symbol error becomes more frequent and excessive signal power may be required to bring errors
down to an acceptable level. The impact is even more severe because each symbol error potentially affects not just
one bit but up to k bits. See Section 11.11 for an in-depth discussion of these trade-offs in the context of M-ary
modulated systems.

12.3 Information Capacity Law

Equation (12.3) suggests that we can increase bit rate indefinitely simply by increasing M. For example, a baseband
channel of bandwidth B = 3.1 kHz can be made to support a bit rate of 62 kb/s by choosing M = 1024 symbols, so
that Eq. (12.3) yields Rbmax = 2 × 3.1 × log2 1024 = 62 kb∕s. The bit rate could be doubled to 124 kb/s in the same
channel by increasing M to 1 048 576. You are right to wonder whether there is no restriction on the maximum
possible value of M.
Consider Figure 12.12 for the case where the M symbols differ only in amplitude between 0 and A. In
Figure 12.12, the receiver associates any symbol having amplitude within a shaded interval with the k bits
assigned to that interval. As M increases, these amplitude intervals become smaller. For a fixed value of A, the
symbols would remain distinguishable as M → ∞ only if the channel were noiseless and the receiver had the
capability to distinguish between infinitesimally close levels. We may use the following intuitive argument to
determine the limiting effect of noise on bit rate. With a noise power Pn at the receiver, the spacing of symbol
√
levels cannot be less than the root-mean-square (rms) noise voltage Pn . Otherwise, adjacent symbols would be
indistinguishable due to noise. The received signal (of power Ps ) and noise (of power Pn ) are uncorrelated, so
√
their powers add to give a range of symbol levels equal to Pn + Ps . With a noise-imposed minimum separation
√
Pn between levels, the maximum number of distinguishable symbol levels is therefore
√
Pn + Ps √
M= √ = 1 + Ps ∕Pn
Pn
which gives the maximum number of bits per symbol kmax as
√
kmax = log2 ( 1 + Ps ∕Pn )
Telegram: @Persian_IEEEComSoc

1
= log2 (1 + Ps ∕Pn ) (12.26)
2
Substituting in Eq. (12.3) yields the maximum rate at which information may be transmitted through a channel
without error, i.e. with the transmitted information-bearing symbols remaining distinguishable at the receiver so
that error-free operation is possible. This maximum rate is known as the information capacity Rbmax of the system.
Thus

Rbmax = 2Bkmax
= Blog2 (1 + Ps ∕Pn ) ≡ Blog2 (1 + C∕N) b∕s (12.27)

Equation (12.27) is the celebrated information capacity law, referred to as the Shannon–Hartley law in recognition
of the work of Claude Shannon [5] building on the early work of Hartley [6]. A rigorous and complete derivation
 12.3 Information Capacity Law 781

......
M=2 M=4 M=8 M = 16 M = 32
A 00000
0000 00001
000 00011
0001 00010
00 00110
0011 00111
001 00101
0010
00100
0 01100
0110 01101
011 01111
0111
01110
01 01010
0101 01011
010
Amplitude

01001
0100 01000
11000
1100 11001
110 11011
1101 11010
11 11110
1111 11111
111 11101
1110 11100
1 10100
1010 10101
101 10111
1011 10110
10 10010
1001 10011
100 10001
1000 10000
0

Figure 12.12 M distinct states represent log2 M bits per state as shown for amplitude states. Bits allocated to adjacent
states differ in only one bit position, an arrangement called Gray coding.

of this equation is presented in Claude Shannon’s seminal paper [5]. It should by now be clear that when trans-
mitting through a bandlimited noisy channel ISI places a limit on symbol rate as stated in Eq. (12.2), whereas
noise precludes an unlimited bit rate, which would otherwise have been possible by indefinitely increasing M as
per Eq. (12.3). It is the latter that places a limit on bit rate when signal power is finite, as stated in Eq. (12.27),
which lays down the rule that governs how bandwidth and signal power may be exchanged in the design of a
transmission system affected by noise. Shannon’s channel coding theorem states as follows:

There exists a coding scheme which can be used to transmit with a vanishingly small probability of error
Telegram: @Persian_IEEEComSoc

over a channel of capacity kmax at a rate not greater than kmax bits per symbol. For rates greater than kmax it
is not possible by any encoding method to have an arbitrarily small probability of error.

Thus, reliable communication through a channel cannot be carried out at a bit rate higher than stipulated by
Eq. (12.27). In practice, communication is deemed reliable if the probability of error or BER satisfies a specified
threshold, such as BER ≤ 10−4 for voice communication and BER ≤ 10−7 for data.
The Shannon–Hartley law indicates that as bandwidth B is increased the signal power required for reliable
communication decreases. Letting Eb denote the average signal energy per bit and N o the noise power per unit
bandwidth then (see Eq. (11.19) in Section 11.3.1 and Eq. (11.124) in Section 11.11)
C ER E R
= b bmax ≡ b • bmax
N No B No B
 782 12 Pulse Shaping and Detection

where we have assumed the best-case (ideal) filtering situation in which occupied bandwidth and noise-equivalent
bandwidth are equal and denoted B. The ratio between bit rate and occupied bandwidth is known as the band-
width efficiency of the system in b/s/Hz, denoted 𝜂 (see Section 11.10.1). Substituting this relation in Eq. (12.27)
allows us to state that the Shannon–Hartley law specifies a relationship between Eb /N o and the highest achievable
bandwidth efficiency 𝜂 of an error-free transmission system as
( )
Rbmax Eb
≡ 𝜂 = log2 1 + 𝜂 b∕s∕Hz
B No
Eb 2𝜂 − 1
⇒ = (12.28)
No 𝜂
.
With Rb /B defined as bandwidth efficiency, the dimensionless ratio Eb /N o is usually interpreted as giving a
measure of the power efficiency of the transmission. The information capacity law therefore shows us how to
trade between bandwidth efficiency and power efficiency. The latter is an important consideration, for example in
portable transceivers where battery life must be prolonged. The Shannon–Hartley law, in the form of Eq. (12.28),
is plotted in Figure 12.13, which shows bandwidth efficiency (b/s/Hz) against Eb /N o in dB. The shaded region of
the graph lying above the curve of 𝜂 versus Eb /N o is the region of unattainable information capacity or bandwidth
efficiency.
We devote the rest of this section to discussing the many implications of the Shannon–Hartley law of Eq. (12.27).
Equation (12.27) indicates that, by increasing bandwidth, bit rate Rb can be proportionately increased without
degrading noise performance (i.e. without increasing BER) or raising carrier-to-noise ratio C/N. Thus, if the cur-
rent operating point is at bit rate Rb , noise equivalent bandwidth B, carrier-to-noise ratio C/N, and bit error ratio

32
24
able
f u n attain ficiency
16 no ef
Regio bandwidth
12 y o r
it
capac r
8 ion fo
o f o perat systems
Bandwidth Efficiency η, bits/s/Hz

6 n n
Regio ransmissio
i c a l t
4 p rac t
3
2

1
Telegram: @Persian_IEEEComSoc

0.5

0.25

1/8

1/16
0 10 20 30 40 50 60 70 80
Eb/No, dB

Figure 12.13 Specification of Shannon’s information capacity theorem as bandwidth efficiency in b/s/Hz versus E b /No in dB.
 12.3 Information Capacity Law 783

BER, denoted (Rb , B, C/N, BER), and the bit rate is to be increased by a factor of n at the same C/N and BER then
bandwidth B must be increased by the same factor n so that the new operating point is (nRb , nB, C/N, BER). In
practice, this change is facilitated by staying with the same modulation or line coding scheme which has a fixed
number of bits k per symbol, but increasing the bandwidth by a factor of n in order to allow symbol rate Rs to be
increased by the same factor without incurring ISI. In this way, bit rate Rb = kRs is increased by the required factor.
It is important, however, to note that to keep C/N constant when B increases by a factor of n, signal power Ps will
also have to increase by the same factor n since noise power N ≡ Pn = N o B. Thus, the original operating point is
(Rb , B, Ps /N o B, BER), whereas the new one, with a factor of n increase in bit rate at an unchanged carrier-to-noise
ratio, is (nRb , nB, nPs /nN o B, BER).
Equation (12.27) also indicates that bit rate can be increased while keeping bandwidth and BER fixed, but this
requires an exponential increase in C/N. For example, with B = 4 kHz and C/N = 63, we obtain Rbmax = 24 kb/s. To
double the bit rate to 48 kb/s at the same bandwidth, we must increase the carrier-to-noise ratio to (C/N)2 = 4095.
Note that (C/N)2 = [C/N + 1]2 −1. In general, to increase bit rate by a factor of n without increasing bandwidth
or BER, we must move from operating point (Rb , B, C/N, BER) to a new point (nRb , B, [C/N + 1]n −1, BER). This
change is carried out in practice by increasing k (the number of bits per symbol) by a factor of n while maintaining
the same symbol rate Rs (since bandwidth is fixed). This requires switching to a higher M-ary APSK modulation
scheme which necessitates an exponential increase in power if transmitted symbols are to continue to be detected
at the same BER as in the lower M-ary scheme. Here, we use APSK to include ASK and PSK but specifically exclude
frequency shift keying (FSK).
The law does not place any lower limit on the bandwidth required to support any given bit rate. For example,
Eq. (12.27) indicates that it is possible to reliably communicate at Rb = 100 kb/s over a 10 Hz bandwidth. The
required C/N is obtained by manipulating the equation to obtain

100 × 103 = 10log2 (1 + C∕N)

3 3
⇒ C∕N = 210×10 − 1 ≈ 10log10 (210×10 ) dB
≈ 10 × 10 × 103 log10 (2) = 30,103 dB

And if the bandwidth is further reduced to just 1 Hz then reliable communication at 100 kb/s over this 1 Hz
bandwidth is still possible, provided C/N ≥ 301 030 dB, and so on without limit. The practical constraint is obvious.
If noise power is −210 dBW/Hz (typical) then received signal power would need to be in excess of 300 820 dBW or
1030 076 MW to allow reliable communication at 100 kb/s using a 1 Hz bandwidth. You would need to install a sun
transmitter to get that kind of power, but it would vaporise the entire planet on being turned on! Therefore, while
in theory there is no lower limit on required bandwidth, in practice there is an indirect limit due to constraints on
achievable or safe or authorised levels of signal power.
In general, increasing channel bandwidth allows us to reduce the signal power needed for reliable communi-
Telegram: @Persian_IEEEComSoc

cation. However, the law does place a lower limit on signal power below which reliable communication is not
possible even at infinite transmission bandwidth. This limit is given in terms of Eb /N o and is called the Shannon
limit. Since 𝜂 → 0 when B → ∞, this limiting value is given by the value of Eb /N o as 𝜂 → 0. You can do this either
graphically or analytically. For the analytic approach, Eq. (12.28) yields
Eb || 2𝜂 − 1
| = lim
No |B→∞ 𝜂→0 𝜂
2𝜂 ln 2 ||
=
1 ||𝜂=0
= ln 2 = 0.69315
= −1.6 dB (Shannon limit) (12.29)
 784 12 Pulse Shaping and Detection

In the above, the second line is obtained by – applying L’Hôpital’s rule – taking the derivatives (with respect to
𝜂) of the numerator and denominator of the expression under the limit operator. The channel capacity Rblim at this
Shannon limit is obtained by evaluating Eq. (12.27) in the limit B → ∞
[ ( )]
P
Rb lim = lim Blog2 1 + s
B→∞ No B
[ ( )]
P
= lim Bloge 1 + s log2 e
B→∞ No B
[ ]
P
= B s log2 e
No B
P
= 1.442695 s b∕s (12.30)
No
A little examination of Figure 12.13 reveals that the benefit gained from a trade-off between bandwidth and
signal power depends on the operating point of the transmission system. At low bandwidth efficiency 𝜂, which
corresponds to a scenario of plentiful bandwidth and low or scarce signal power, a small increase in Eb /N o yields
the benefit of a large increase in 𝜂. But at high bandwidth efficiency, which corresponds to a scenario of scarce
bandwidth and plentiful signal power, a small decrease in 𝜂 (which means a little extra bandwidth) produces the
benefit of a large reduction in Eb /N o . What this means is that the factor by which bandwidth must be increased to
compensate for a given amount of reduction in signal energy per bit depends on the operating point. For example,
if Eb /N o is reduced by 1.5 dB, the required increase in bandwidth is a factor of 16 at Eb /N o = 0 dB, but a factor of
only ≈ 1.02 at Eb /N o = 60 dB. This also means that the amount of saving in energy per bit achieved by increasing
bandwidth depends on the operating point. For example, doubling bandwidth at 𝜂 = 1/8 (by reducing 𝜂 from 1/8
to 1/16) delivers a reduction in Eb /N o of only 0.1 dB, whereas doubling bandwidth at 𝜂 = 32 (by reducing 𝜂 from
32 to 16) allows a reduction in Eb /N o by 45.2 dB.
Finally, we must point out that Shannon’s channel coding theorem merely tells us that it is possible to have
error-free transmission at the maximum bit rate given by the information capacity law of Eq. (12.27) but does
not show us how to design such a system. In practice, all digital transmission systems – both modulated and
baseband – fall short of achieving the specified maximum bit rate for a given bandwidth and C/N. Equation (12.27)
is nevertheless a useful benchmark against which the performance of practical systems can be measured.

Worked Example 12.2

In this worked example, we want to compare three practical transmission systems to the Shannon–Hartley
benchmark. You may wish to also read the related Worked Example 6.9.
A digital transmission system operates at a message bit rate Rb = 2048 kb/s in AWGN using a raised cosine
filter of roll-off factor 𝛼 = 0.1. Reliable communication is set at BER = 10−7 . Determine how each of the following
implementations of this system compares with the Shannon–Hartley benchmark of Eq. (12.28) and comment
Telegram: @Persian_IEEEComSoc

on your results.
(a) Using quadriphase shift keying (QPSK) modulation with a modem having modem implementation loss
Lmil = 0.5 dB and no error control coding.
(b) Using 16-APSK modulation with a modem having modem implementation loss Lmil = 1 dB and no error
control coding.
(c) Using 16-FSK modulation with a modem having modem implementation loss Lmil = 1 dB and no error
control coding.
(d) Using 16-APSK modulation with a modem having modem implementation loss Lmil = 1 dB along with error
control coding of code rate r = 1/2 and coding gain Gc = 10.5 dB.
 12.3 Information Capacity Law 785

(a) For QPSK modulation, M = 4, so symbol rate

Rs = Rb ∕log2 M = 2048∕2 = 1024 kBd
Occupied bandwidth
Bocc = (1 + 𝛼)Rs = (1 + 0.1)Rs = 1126.4 kHz
And bandwidth efficiency
𝜂 = Rb ∕Bocc = 2048∕1126.4 = 1.8182
Using Eq. (12.28), the minimum required value of Eb /N o for operating at this bandwidth efficiency is deter-
mined as
2𝜂 − 1 21.8182 − 1
(Eb ∕No )min = = = 1.3895 = 1.43 dB
𝜂 1.8182
We determine the Eb /N o value required by our QPSK system, denoted (Eb /N o )QPSK , by reading the BER ver-
sus Eb /N o curve of Figure 11.45 for QPSK at BER = 10−7 (which gives Eb /N o = 11.3 dB) and adding the
extra value of 0.5 dB required due to modem implementation loss. Thus, (Eb /N o )QPSK = 11.8 dB. Since the
Shannon–Hartley law specifies a minimum Eb /N o of 1.43 dB for this bandwidth efficiency, we conclude that
our QPSK is less power efficient than the benchmark specification by
ΔP = (Eb ∕No )QPSK − (Eb ∕No )min = 11.8 − 1.43 = 10.4 dB
(b) For 16-APSK modulation, M = 16, so symbol rate
Rs = Rb ∕log2 M = 2048∕4 = 512 kBd
Occupied bandwidth
Bocc = (1 + 𝛼)Rs = (1 + 0.1)Rs = 563.2 kHz
And bandwidth efficiency
𝜂 = Rb ∕Bocc = 2048∕563.2 = 3.6364
Equation (12.28) yields the minimum required value of Eb /N o as
2𝜂 − 1 23.6364 − 1
(Eb ∕No )min = = = 3.1447 = 4.98 dB
𝜂 3.6364
We determine the Eb /N o value required by our 16-APSK system, denoted (Eb /N o )16APSK , by reading the BER
versus Eb /N o curve of Figure 11.52 for 16-APSK at BER = 10−7 (which gives Eb /N o = 15.2 dB) and adding the
extra 1 dB required due to modem implementation loss. Thus, (Eb /N o )16APSK = 16.2 dB. We conclude that our
Telegram: @Persian_IEEEComSoc

16-APSK system is less power efficient than the benchmark specification by

ΔP = (Eb ∕No )16APSK − (Eb ∕No )min = 16.2 − 4.98 = 11.2 dB
(c) For 16-FSK modulation, M = 16, so symbol rate
Rs = Rb ∕log2 M = 2048∕4 = 512 kBd
Occupied bandwidth is determined using Eq. (11.84) as
( )
M+1
Bocc = + 𝛼 Rs = (8.5 + 0.1) × 512 = 4403.2 kHz
2
And bandwidth efficiency
𝜂 = Rb ∕Bocc = 2048∕4403.2 = 0.4651
 786 12 Pulse Shaping and Detection

Equation (12.28) yields the minimum required value of Eb /N o as

2𝜂 − 1 20.4651 − 1
(Eb ∕No )min = = = 0.818 = −0.87 dB
𝜂 0.4651
We determine the Eb /N o value required by our 16-FSK system, denoted (Eb /N o )16FSK , by reading the BER
versus Eb /N o curve of Figure 11.48 for 16-FSK at BER = 10−7 (which gives Eb /N o = 8.9 dB) and adding the
extra 1 dB due to modem implementation loss to obtain (Eb /N o )16FSK = 9.9 dB. We conclude that our 16-FSK
system is less power efficient than the benchmark specification by
ΔP = (Eb ∕No )16FSK − (Eb ∕No )min = 9.9 − (−0.87) = 10.8 dB
(d) For 16-APSK modulation with error control coding of code rate r = 1/2, the bit rate is increased by the inserted
redundant bits to Rbc = Rb × 1/r = 4096 kb/s, so the symbol rate is now
Rs = Rbc ∕log2 M = 4096∕4 = 1024 kBd
Occupied bandwidth
Bocc = (1 + 𝛼)Rs = (1 + 0.1) × 1024 = 1126.4 kHz
And bandwidth efficiency
𝜂 = Rb ∕Bocc = 2048∕1126.4 = 1.8182
Equation (12.28) yields the minimum required value of Eb /N o as
2𝜂 − 1 21.8182 − 1
(Eb ∕No )min = = = 1.3895 = 1.43 dB
𝜂 1.8182
We determine the Eb /N o required by this error control coded 16-APSK system, denoted (Eb /N o )c16APSK , by
reading the BER versus Eb /N o curve of Figure 11.52 for 16-APSK at BER = 10−7 (which gives Eb /N o = 15.2 dB),
adding the extra 1 dB due to modem implementation loss and subtracting the coding gain of 10.5 dB. Thus
(Eb ∕No )c16APSK = (Eb ∕No )theoretical + Lmil − Gc
= 15.2 + 1 − 10.5 = 5.7 dB
We conclude that our error control coded 16-APSK system is less power efficient than the benchmark specifi-
cation by
ΔP = (Eb ∕No )c16APSK − (Eb ∕No )min = 5.7 − 1.43 = 4.27 dB
Examining the above results, we note that the use of error control coding enabled us to approach closer to the
benchmark level of power efficiency than is possible with modulation without error control coding. To com-
pare the bandwidth and power trade-off involved in each implementation, we will use the uncoded 16-APSK
system as a reference and will need the value of C/N o for each system, which is obtained from the equation
Telegram: @Persian_IEEEComSoc

(see Eq. (11.63))

C∕No = (Eb ∕No )theoretical + 10log10 (Rb ) + Lmil − Gc (12.31)
Substituting relevant values, noting that Gc = 0 for the first three systems and Rb in the fourth system is the
coded bit rate (= 4096 kb/s), we obtain
(C∕No )QPSK = 74.9 dB; (C∕No )16APSK = 79.3 dB;
(C∕No )16FSK = 73.0 dB; (C∕No )c16APSK = 71.8 dB
Thus, switching from (uncoded) 16-APSK to (uncoded) QPSK enabled a signal power saving of 79.3−74.9 =
4.4 dB at the price of a factor of 2 increase in bandwidth (from 563.2 to 1126.4 kHz), whereas switching from
(uncoded) 16-APSK to the coded 16-APSK having the above specified capability (r = 1/2; Gc = 10.5 dB) enabled
 12.4 The Digital Receiver 787

a larger saving in signal power of 79.3−71.8 = 7.5 dB at the same price of a factor of 2 increase in bandwidth.
Switching from (uncoded) 16-APSK to (uncoded) 16-FSK enabled a power saving of 79.3−73 = 6.3 dB but at
the price of a factor of 7.8 increase in bandwidth (from 563.2 to 4403.2 kHz). Error control coding in general
facilitates a more efficient exchange between bandwidth and signal power than is possible by only switching
modulation schemes.

12.4 The Digital Receiver

We establish in Section 12.2 that at the decision point in the receiver the received pulse, although spread by the
transmission medium and of a longer duration than it had when generated at the transmitter, must have the right
waveform or shape (such as a since pulse) to avoid ISI. The focus of Section 12.2 is on mitigating the impact of pulse
spreading caused by a finite system bandwidth. In addition to spreading, the transmission medium or channel does
distort the transmitted pulse in both amplitude and phase in a manner fully described by the channel transfer
function H c (f ). Furthermore, noise is also added to the transmitted pulse in the transmission medium and the
front end of the receiver. This section briefly considers the measures of equalisation, matched filtering, and clock
extraction used at the receiver to optimise the detection of a transmitted pulse sequence in the presence of channel
distortion and noise.

12.4.1 Adaptive Equalisation

In nearly every case, the channel effect is undesirable and, as discussed in Section 4.7.4, a filter known as an
equaliser is needed at the receiver to remove the channel distortion. That is, we must have
Hc (f )He (f ) = 1 (normalised) (12.32)
where H e (f ) is the transfer function of the equaliser as shown in the block diagram of Figure 12.6b. Note
that Eq. (12.32) specifically excludes propagation delay 𝜏, which can be accounted for by inserting the factor
exp(−j2𝜋f𝜏) in the right hand side.
In most cases, as in the public switched telephone network (PSTN), the channel characteristic and hence H c (f )
is time-varying in a nondeterministic way. The equalisation then has to be adaptive, with H e (f ) constantly and
automatically adjusting itself to satisfy Eq. (12.32). One way of achieving this is by using a tapped-delay-line filter
(such as in Figure 10.16). In this case, however, the tap gains are adjustable to minimise the mean squared error
between expected and received pulse shapes averaged over several sampling instants. This can be done with the
help of a training sequence of bits transmitted prior to the information bits. The optimum values computed for
the tap gains are then maintained for the duration of the call on the assumption that the channel does not change
Telegram: @Persian_IEEEComSoc

significantly during this (short) period.

12.4.2 Matched Filter

Let us now consider the effect of noise in order to correctly specify the transfer function H y (f ) of the receiver filter
indicated in the block diagram of Figure 12.6. The noise w(t) indicated in this block diagram includes contributions
from both the channel and the front end of the receiver. We will assume that this noise is AWGN. As discussed in
detail in Chapter 6, the noise is described as additive because it is present in the same amount regardless of signal
value. It is white because it contains all frequency components at the same amplitude (i.e. it has a uniform ampli-
tude spectrum, or equivalently, a uniform power spectral density), just as white light contains equal amounts of all
colours. Finally, it is described as Gaussian because it has a normal (i.e. Gaussian) probability density function.
 788 12 Pulse Shaping and Detection

g(t) ⇌ G(f) x(t) Matched y(t) ⇌ Y(f) y(Ts) = go(Ts) + ñ(Ts)

(a) filter
h(t) ⇌ H(ƒ) go(t) + ñ(t)
t = Ts
w(t)
h(t – τ) ≡ g(τ+ Ts – t)
g(t) g(–t) g(Ts – t) ≡ h(t) h(–τ)

(b)
t t t τ τ
Ts –Ts Ts –Ts Ts–t t
Ts
Time reversal Delay by Ts Time reversal; Delay by t
switch to τ-axis

Correlator

gi(t) Ts go(Ts)
(c)
∫0
g(t)

Figure 12.14 Optimum detection of a signal g(t) in the presence of additive white Gaussian noise (AWGN): (a) the matched
filter; (b) graphical illustration of relationship between input pulse g(t) and matched filter impulse response h(t);
(c) correlation implementation of matched filter.

The receive filter must be designed to minimise the effect of noise. This is accomplished by maximising the
ratio between signal power and noise power at the receiver output at the decision instant T s . The situation is as
illustrated in Figure 12.14a in which a pulse g(t) has AWGN w(t) added to it. Our task is to specify a matched filter
which will produce an output y(t) that has a maximum SNR. This ensures that the sample of y(t) taken at t = T s will
contain the smallest possible amount of noise perturbation ñ(T s ). We derive the matched filter’s transfer function
H(f ) and impulse response h(t) required to optimise the detection of the pulse in the presence of AWGN and then
discuss a practical implementation approach using correlation processing.

12.4.2.1 Specification of a Matched Filter

The transfer function H(f ) of the matched filter in Figure 12.14a may be obtained using a heuristic approach
involving three increasingly prescriptive observations. Once H(f ) has been specified, the impulse response h(t) is
obtained by taking the inverse FT.
1. The bandwidth of the filter must be just enough to pass the incoming pulse g(t). If filter bandwidth is too wide,
Telegram: @Persian_IEEEComSoc

noise power is unnecessarily admitted, and if it is too narrow then some pulse energy is blocked. Thus, the
filter transfer function H(f ) must span the same frequency band as the FT G(f ) of the pulse g(t). How should
the spectral shape of |H(f )| compare to |G(f )|?
2. The gain response |H(f )| of the filter should not necessarily be flat within its passband. Rather, it should be
such that the filter attenuates the white noise significantly at those frequencies where G(f ) is small – since these
frequencies contribute little to the signal energy. And the filter should boost those frequencies at which G(f )
is large in order to maximise the output signal energy. Therefore, the filter should be tailored to the incoming
pulse, with a gain response that is small where G(f ) is small and large where G(f ) is large. In other words, the
gain response of the filter should be identical in shape to the amplitude spectrum of the pulse. That is
|H(f )| = K|G(f )| (12.33)
where K is some constant.
 12.4 The Digital Receiver 789

3. To complete the specification of the filter, its phase response is required. The filter output y(t) may be written
as
̃(t)
y(t) = go (t) + n (12.34)
where go (t) is the signal component and ñ(t) is coloured noise – coloured because after white noise passes
through a filter its amplitude spectrum is no longer flat but is stronger at some frequencies than at others. The
maximum instantaneous output signal power occurs at the sampling instant t = T s if every frequency component
(i.e. cosine function) in go (t) is delayed by the same amount T s and has zero initial phase so that
go (t) = A1 cos[2𝜋f1 (t − Ts )] + A2 cos[2𝜋f2 (t − Ts )] + A3 cos[2𝜋f3 (t − Ts )] + · · · (12.35)
which results in the maximum possible signal sample at t = T s given by
go (Ts ) = A1 + A2 + A3 + · · ·
where A1 , A2 , A3 , …, are the amplitudes of the sinusoidal components of go (t) of respective frequencies f 1 , f 2 ,
f 3 , … Note that, in practice, these frequencies will be infinitesimally spaced, giving rise to a continuous spectrum
Go (f ). Rewriting Eq. (12.35) in the form
go (t) = A1 cos(2𝜋f1 t − 2𝜋f1 Ts )
+ A2 cos(2𝜋f2 t − 2𝜋f2 Ts )
+ A3 cos(2𝜋f3 t − 2𝜋f3 Ts )
+···
makes it clear that the phase spectrum of go (t) is
𝜙o (f ) = −2𝜋fTs
From the discussion in Chapter 4, this output phase response is the sum of the input signal’s phase spectrum
𝜙g (f ) and the filter’s phase response 𝜙H (f ). That is, 𝜙o (f ) = 𝜙g (f ) + 𝜙H (f ), which means that 𝜙H (f ) = 𝜙o (f ) − 𝜙g (f ),
and hence
𝜙H (f ) = −2𝜋fTs − 𝜙g (f ) (12.36)
Equation (12.33) gives the required filter gain response and Eq. (12.36) gives its phase response. What remains is
for us to combine these two into a single expression for the filter’s transfer function H(f ) and then take the inverse
FT of H(f ) to obtain the filter’s impulse response h(t)
H(f ) = K|G(f )| exp[j𝜙H (f )]
= K|G(f )| exp[−j𝜙g (f )] exp(−j2𝜋fTs )
Telegram: @Persian_IEEEComSoc

= KG∗ (f ) exp(−j2𝜋fTs ) (12.37)

where * denotes complex conjugation, and we have used the fact that
G(f ) = |G(f )| exp[j𝜙g (f )]; G∗ (f ) = |G(f )| exp[−j𝜙g (f )]
Noting (from Eqs. (4.84) and (4.86)) that complex conjugation of G(f ) corresponds to a time reversal of the
real signal g(t) to give g(−t); and (from Eq. (4.80)) that multiplying G*(f ) by the exponential term exp(−j2𝜋fT s )
corresponds to delaying g(−t) by T s , we see that the inverse FT of Eq. (12.37) yields
h(t) = Kg(Ts − t) (12.38)
This is an important result which states that if the detection filter is to be matched to an AWGN-corrupted pulse
g(t) of duration T s – matched in the sense of maximising the output signal power when compared to the noise
 790 12 Pulse Shaping and Detection

power at the decision instant t = T s – then the filter must be designed with an impulse response that is (except
for a nonzero positive scale factor K) a time-reversed replica of the pulse, delayed by the sampling interval T s . See
Figure 12.14b for a graphical illustration of the process of converting g(t) into g(T s −t) for any given value of T s . It
may also be useful to review the material on time shifting and time reversal in Sections 3.2.1 and 3.2.2 for more
in-depth information. Note that here and elsewhere in this chapter the term delay is used with reference to time t
and amounts to a rightward translation in the +t axis direction, which corresponds to an advance with reference
to −t and the −t axis direction.

12.4.2.2 Matched Filter by Correlation

It is informative to consider how the matched filter relates to the correlation receiver employed in Chapter 11
for coherent detection of modulated pulses. A matched filter designed to detect a known pulse g(t) will have the
impulse response h(t) given by Eq. (12.38). As discussed in Section 3.6, the output go (t) of this filter in response to
an input pulse gi (t) is obtained by convolving gi (t) with the impulse response of the filter. Thus
go (t) = gi (t) ∗ h(t)
∞
= gi (𝜏)h(t − 𝜏)d𝜏
∫−∞
Ts
= gi (𝜏)g(Ts − t + 𝜏)d𝜏 (12.39)
∫0
where the second line is merely the definition of the convolution integral, and in the last line we make use of the
fact that the input pulse is of duration T s to narrow the range of integration, and we substitute h(t) ≡ g(T s − t),
ignoring the scale factor K or normalising it to unity. Refer to Figure 12.14b for a graphical aid in seeing how
h(t − 𝜏) becomes g(T s − t + 𝜏). You will notice the series of steps that transform the known pulse g(t) for which the
matched filter is designed, shown as the leftmost waveform of Figure 12.14b, to h(t − 𝜏), the rightmost waveform.
By comparing these two waveforms you will be able to see that the rightmost is simply the first plotted on a 𝜏 axis
and made to start earlier (i.e. advanced) by T s − t, hence it is g(𝜏 + T s − t). Sampling the matched filter output at
t = T s simply means replacing t by T s on both sides of Eq. (12.39). This yields the result
Ts
go (Ts ) = gi (𝜏)g(Ts − Ts + 𝜏)d𝜏
∫0
Ts
= gi (𝜏)g(𝜏)d𝜏 (12.40)
∫0
The right-hand side of this equation shows that to obtain go (T s ) – which is the matched filter’s output sam-
pled at t = T s – we may use an arrangement that multiplies the incoming pulse gi (t) by the known pulse g(t)
and then integrates the product over the pulse interval 0 → T s . This process is known as the correlation of gi (t)
with g(t). A block diagram of this implementation of the matched filter is shown in Figure 12.14c. Note therefore
Telegram: @Persian_IEEEComSoc

that the coherent demodulator (Figure 7.30), correlation receiver (Figures 11.8c and 12.14c), and matched filter
(Figure 12.14a) in fact perform equivalent operations, as stated in Chapter 11. This equivalence is further empha-
sised in Figure 12.15. When comparing Figure 12.15a,c, recall that an integrator is just a special LPF – with a gain
response that decreases linearly with frequency.
In Eqs. (12.39) and (12.40), time t is measured from the start of each pulse interval, hence the range 0 to T s even
though we are in the nth pulse interval, where n can take on any positive integer value. The correlator performs
an integrate-and-dump operation whereby an accumulator is reset to zero at the start of each pulse interval, and
the product of gi (t) and g(t) is then accumulated for a period T s at the end of which the accumulator contains the
result go (T s ) for that interval. This result is passed to a decision device and the accumulator is immediately reset
to zero ready to repeat the same computation in the next pulse interval. It is worth emphasising that the various
descriptions of a demodulator as coherent, a carrier or pulse as known, and a filter as matched to a known pulse
 12.4 The Digital Receiver 791

Figure 12.15 Equivalence between (a) correlation receiver; Correlation receiver

(b) matched filter; (c) coherent demodulator.
Received
pulse Ts
(a) ∫0
gʹ(t) go(Ts)

g(t) (known pulse)

Matched filter
Received
pulse h(t)
(b) = g(Ts – t)
gʹ(t) go(t) go(Ts)
sample at
t = Ts

Coherent demodulator
Modulated Message
signal signal
(c) LPF

Known carrier

imply accurate knowledge of the phase and frequency of the incoming carrier or pulse. This information is usually
obtained by clock extraction as discussed after the following worked examples.

12.4.2.3 Matched Filter Worked Examples

Worked Example 12.3 Matched Filter Impulse Response: Graphical Approach

We wish to sketch the impulse response of a matched filter for receiving the pulse g(t) shown in Figure 12.3a,
where the pulse duration T s = 5 μs. Further discussion of the operations of time delay and time reversal is
available in Sections 3.2.1 and 3.2.2.

The required impulse response is given by Eq. (12.38) with K = 1. This can be obtained in two steps. First,
the pulse g(t) is time-reversed to give g(−t). Then, g(−t) is delayed by T s to give g(T s − t), which is the required
Telegram: @Persian_IEEEComSoc

impulse response. The waveforms g(−t) and g(T s − t) are sketched in Figure 12.3b,c. It is important to understand
how these two waveforms are obtained. Observe that the waveform of g(−t) may be obtained simply by flipping
g(t) horizontally about t = 0, and that the waveform g(T s − t) results from delaying g(−t) by a time T s . In this case,
since g(−t) ‘starts’ at the time t = −5 μs, it follows that g(T s − t) must ‘start’ at a time T s (= 5 μs) later, which is
therefore the time t = 0.
Table 12.1 provides verification of the above procedures. Noting that g(ti ) is the value of the pulse g(t) at t = ti , it
follows by definition of g(t) in Figure 12.16a that g(−10) = 0, g(−5) = 0, g(1) = 1, g(4) = 0.5, g(10) = 0, and so on,
where t is in μs. Table 12.1 gives values of the waveforms g(t), g(−t), and g(T s − t) at various values of t. For example,
at t = 4, g(t) = g(4) = 0.5 (by definition); g(−t) = g(−4) = 0 (by definition); and g(T s − t) = g(5−4) = g(1) = 1 (by
definition). Plotting the entries of this table leads to Figure 12.16, with column 3 plotted against column 1 to give
Figure 12.16b, and column 4 plotted against column 1 to give Figure 12.16c.
 792 12 Pulse Shaping and Detection

Table 12.1 Worked Example 12.3. Entries plotted in Figure 12.16. T s = 5 μs.

t (𝛍s) g(t) g(−t) hr (t) = g(T s − t)

−10 0 = g(10) = 0 = g(5 − (−10)) = g(15) = 0

−5 0 = g(5) = 0 = g(5 − (−5)) = g(10) = 0
−4 0 = g(4) = 0.5 = g(5 − (−4)) = g(9) = 0
−3 0 = g(3) = 1 = g(5 − (−3)) = g(8) = 0
−2 0 = g(2) = 1 = g(5 − (−2)) = g(7) = 0
−1 0 = g(1) = 1 = g(5 − (−1)) = g(6) = 0
0 0 = g(0) = 0 = g(5 − 0) = g(5) = 0
1 1 = g(−1) = 0 = g(5 − 1) = g(4) = 0.5
2 1 = g(−2) = 0 = g(5 − 2) = g(3) = 1
3 1 = g(−3) = 0 = g(5 − 3) = g(2) = 1
4 0.5 = g(−4) = 0 = g(5 − 4) = g(1) = 1
5 0 = g(−5) = 0 = g(5 − 5) = g(0) = 0
10 0 = g(−10) = 0 = g(5 − 10) = g(−5) = 0

g(t)
1.0

(a) 0.5

t, μs
–5 –4 –3 –2 –1 0 1 2 3 4 5

g(–t)
1.0

(b) 0.5

t, μs
Telegram: @Persian_IEEEComSoc

–5 –4 –3 –2 –1 0 1 2 3 4 5

h(t) = g(Ts – t)
Ts = 5 μs
1.0

(c) 0.5

t, μs
–5 –4 –3 –2 –1 0 1 2 3 4 5

Figure 12.16 Worked Example 12.3.

 12.4 The Digital Receiver 793

Worked Example 12.4 Matched Filter Output: Graphical Approach

We wish to determine the output pulse go (t) that is obtained at the decision point of the receiver when the trans-
mitted pulse g(t) in the previous example (Figure 12.16a) is detected using a matched filter of impulse response
h(t). See a more detailed discussion of the convolution operation in Section 3.6.2.

What is needed here is the response go (t) of a (matched) filter of impulse response h(t) to an input signal g(t).
We know from Section 3.6 that the output signal go (t) is given by
go (t) = g(t) ⋆ h(t)
∞
= g(𝜏)h(t − 𝜏)d𝜏 (12.41)
∫−∞
Equation (12.41) defines the convolution integral, which states that go (t) is given at each time instant t by the total
area under the function g(𝜏)h(t − 𝜏), which is the product of the input waveform and a time-reversed and delayed
(by t) version of the impulse response.
For convenience, the input pulse g(𝜏) and the impulse response h(𝜏) of the matched filter (obtained in the pre-
vious worked example) are sketched again in Figure 12.17a,b. When both g(t) and h(t) are of finite duration as in
this case then it is easier, and indeed very illuminating, to evaluate the convolution integral graphically as follows.
(i) Obtain the waveform h(t−𝜏) using the procedure described in the previous worked example. In Figure 12.17c,
a few examples of h(t − 𝜏) are shown for t = −2, 0, 2, 5, 7, and 10 μs.
(ii) Multiply together the waveforms h(t − 𝜏) and g(𝜏) to obtain the integrand g(𝜏)h(t − 𝜏) in Eq. (12.41). Note that
this integrand is identically zero for those values of t that lead to a h(t − 𝜏), which does not overlap g(𝜏). It can
be seen in Figure 12.17 that this happens for t ≤ 0, and t ≥ 10, which means that the output pulse go (t) is zero
in these two regions of time. Example curves of g(𝜏)h(t−𝜏) are shown in Figure 12.17d, for t = 2, 5, and 7 μs.
(iii) The value of the output pulse go (t) at a time t is the area under the curve of g(𝜏)h(t−𝜏). For example, it can
be seen in Figure 12.17d that the area under the curve of g(𝜏)h(7−𝜏) is 1.5, which means that go (t) = 1.5
at t = 7 μs.
(iv) Repeat the above steps for different values of t to obtain the output go (t) sketched in Figure 12.18.
Note that the matched filter has distorted the transmitted pulse g(t) in such a way that the maximum value of the
output pulse go (t) occurs at the decision instant t = T s . It can be seen in Figure 12.17c that h(T s − 𝜏) = g(𝜏). Since
the pulse g(t) is a real signal, it follows from Eq. (12.41) that
∞ ∞
go (Ts ) = g(𝜏)h(Ts − 𝜏)d𝜏 = g(𝜏)g(𝜏)d𝜏
∫−∞ ∫−∞
∞
= |g(𝜏)|2 d𝜏
∫−∞
Telegram: @Persian_IEEEComSoc

≡ E ≡ Energy of signal g(t) (12.42)

Thus, if a known pulse g(t) is presented to the input of a filter matched to it then when the output go (t) of the
matched filter is sampled at time t = T s , that sample go (T s ) gives an estimate of the energy of the transmitted pulse
g(t), assuming of course that the gain of the matched filter is normalised to unity and the effect of the transmission
channel has been equalised according to Eq. (12.32).

Worked Example 12.5 Irrelevance of Pulse Shapes

We wish to show that the signal-to-noise ratio (SNR)o at the output of a matched filter depends only on the
ratio between input pulse energy E and noise power density, and not on the particular shape of the pulse.
 794 12 Pulse Shaping and Detection

g(τ)
1.0

(a) 0.5

τ, μs
–5 –4 –3 –2 –1 0 1 2 3 4 5
h(τ)
1.0

0.5
(b)
τ, μs
–5 –4 –3 –2 –1 0 1 2 3 4 5

1.0
h(–2 – τ) h ( – τ) h (2 – τ) h (5 – τ) h (7 – τ) h (10 – τ)

(c) 0.5

τ, μs
–7 –5 –4 –3 –2 –1 0 1 2 3 4 5 7

g(τ)h(t – τ)
1.0
g(τ)h(5 – τ)

(d) 0.5 g(τ)h(7 – τ)

g(τ)h(2 – τ)
τ , μs
–5 –4 –3 –2 –1 0 1 2 3 4 5

Figure 12.17 Worked Example 12.4.

It is clear from the last worked example, and more specifically Eq. (12.42), that the signal at the output of a
matched filter has a maximum value E at the decision instant t = T s , where E is the transmitted pulse energy.
Since instantaneous power is defined as the square of the absolute value of the signal at a given instant, it follows
that the instantaneous power of the output signal go (t) at the decision instant is

Ps = [go (Ts )]2 = E2

Telegram: @Persian_IEEEComSoc

(12.43)

From Eq. (4.164) in our discussion of output spectral density of LTI systems (Section 4.7.2), the output noise
power spectral density of the matched filter is
No
So (f ) = |G(f )|2 (12.44)
2
where N o /2 is the power spectral density of white noise w(t) at the input of the matched filter, G(f ) is the spectrum
of the transmitted pulse, and we have used Eq. (12.33) for the filter’s gain response with the constant K set to 1.
Integrating So (f ) over the entire frequency axis yields output noise power as
No ∞ N
Pn = |G(f )|2 df = o E (12.45)
∫
2 −∞ 2
 12.4 The Digital Receiver 795

go(t) Decision instant

3 ≡ go(Ts)

t, μs
0
0 1 2 3 4 5 ≡ Ts 6 7 8 9 10

Figure 12.18 Worked Example 12.4: output go (t) of matched filter.

where we obtained the final term by applying Rayleigh’s energy theorem (also known as Parseval’s theorem) given
in Eq. (4.94).
We obtain SNRo as the ratio between Ps in Eq. (12.43) and Pn in Eq. (12.45)
Ps E2
SNRo = =
Pn ENo ∕2
2E
= (12.46)
No
Equation (12.46) is the desired result. It is interesting that the shape or waveform of the transmitted pulse g(t) does
not feature in the achievable signal-to-noise ratio. All that matters is the pulse energy, which may be increased
to improve SNRo by increasing the amplitude and/or duration of the pulse. The latter option, however, would
reduce symbol rate. In summary then, provided a matched filter is used at the receiver, all pulses of the same energy
are equally detected in the presence of white noise irrespective of the pulse shapes. We must therefore emphasise
that pulse shaping (studied in Section 12.2) is required for ISI minimisation and has no bearing whatsoever on the
impact of white noise.
Telegram: @Persian_IEEEComSoc

Worked Example 12.6 Matched Filter Output: Mathematical Approach

The outputs of various matched filters are presented in Chapter 11. See, for example, Figure 11.24. We now
wish to show how some of these outputs were determined. We will derive an expression for the output y(t) of a
matched filter for detecting a binary PSK pulse g(t), which is made up of a sinusoidal carrier that completes n
cycles in its interval of duration T b from t = 0 to t = T b , where n is a positive integer.

The expression for the binary PSK pulse is

( ) ( )
n t − Tb ∕2
g(t) = Ac cos 2𝜋 t rect (12.47)
Tb Tb
 796 12 Pulse Shaping and Detection

You may wish to refer to discussions of the rectangular function rect() in Section 2.6.3 and sinusoidal function
in Section 2.7, if in any doubt. In line with Eq. (12.38), setting K = 1, we replace t wherever it occurs in Eq. (12.47)
with T b − t to obtain the impulse response h(t) of the matched filter as
( ) ( )
n (Tb − t) − Tb ∕2
h(t) = Ac cos 2𝜋 (Tb − t) rect (12.48)
Tb Tb
Next, we replace t wherever it occurs in Eq. (12.48) with t − 𝜏 to obtain
( ) ( )
n (𝜏 + Tb − t) − Tb ∕2
h(t − 𝜏) = Ac cos 2𝜋 (𝜏 + Tb − t) rect
Tb Tb
( ) ( )
n 𝜏 − (t − Tb ∕2)
= Ac cos 2𝜋 (𝜏 + Tb − t) rect (12.49)
Tb Tb
Equation (12.41) specifies that the desired output y(t) is obtained by integrating the product signal g(𝜏)h(t − 𝜏)
in the limits −∞ to ∞. The integration only needs to be carried out over the interval where this product signal is
nonzero, i.e. in the region where the pulses g(𝜏) and h(t − 𝜏) overlap. Since

( ) ⎧
t ⎪1, −Tb ∕2 ≤ t ≤ Tb ∕2
rect =⎨
Tb ⎪0, Otherwise
⎩
it follows that
( )
⎧ n
⎪Ac cos 2𝜋 T 𝜏 , 0 ≤ 𝜏 ≤ Tb
g(𝜏) = ⎨ b
⎪0, Otherwise
⎩
and
( ) ( )
⎧ n Tb T Tb
A
⎪ c cos 2𝜋 (𝜏 + Tb − t) , − 2
≤𝜏− t− b ≤ 2
h(t − 𝜏) = ⎨ Tb 2
⎪0, Otherwise
⎩
( )
⎧ n
⎪ Ac cos 2𝜋 (𝜏 + Tb − t) , t − Tb ≤ 𝜏 ≤ t
=⎨ Tb
⎪0, Otherwise
⎩
Note that the right end of the interval of h(t − 𝜏) is 𝜏 = t and the left end of the interval of g(𝜏) is 𝜏 = 0. Clearly
there is no overlap if the right end of h(t − 𝜏) is below the left end of g(𝜏). This means that the output y(t) = 0
Telegram: @Persian_IEEEComSoc

for t ≤ 0. Similarly, there is no overlap if the left end of h(t − 𝜏), which is 𝜏 = t − Tb, exceeds the right end of g(𝜏),
which is t = T b . So, y(t) = 0 for t−T b ≥ T b , i.e. for t ≥ 2 Tb. Combining these two regions, it means that y(t) is zero
outside the interval 0 ≤ t ≤ 2T b .
Next, we note that when T b ≤ t ≤ 2T b the region of overlap is from the left end of h(t − 𝜏) to the right end of g(𝜏),
whereas when 0 ≤ t ≤ T b , the region of overlap is from the left end of g(𝜏) to the right end of h(t − 𝜏). The output
signal is therefore given by the integrations
⎧ { ( ) ( )}
⎪∫ t Ac cos 2𝜋 n 𝜏 •Ac cos 2𝜋 n (𝜏 + Tb − t) d𝜏, 0 ≤ t ≤ Tb
⎪ 0 Tb Tb
y(t) = ⎨ { ( ) ( )}
⎪∫ Tb n n
Ac cos 2𝜋 𝜏 •Ac cos 2𝜋 (𝜏 + Tb − t) d𝜏, Tb ≤ t ≤ 2Tb
⎪ t−Tb Tb Tb
⎩
 12.4 The Digital Receiver 797

The integrand is the product of two sinusoidal functions of 𝜏. For the first interval, applying the trigonometric
identity for the product of two cosines
t{ ( ) ( )}
A2c n n
y(t) = cos 2𝜋 (Tb − t) + cos 2𝜋 (2𝜏 + Tb − t) d𝜏
2 ∫0 Tb Tb
{ ( )|𝜏=t ( )|𝜏=t }
A2c n | Tb n |
= 𝜏 cos 2𝜋 (Tb − t) | + sin 2𝜋 (2𝜏 + Tb − t) |
2 Tb | 4𝜋n T |
|𝜏=0 b |𝜏=0
( ( ) [ ( ) ( )])
A2c 2𝜋nt Tb 2𝜋nt 2𝜋nt
= t cos 2𝜋n − + sin 2𝜋n + − sin 2𝜋n −
2 Tb 4𝜋n Tb Tb
2 ( )
A T
= c t cos(2𝜋nt∕Tb ) + b [sin(2𝜋nt∕Tb ) − sin(−2𝜋nt∕Tb )]
2 4𝜋n
( )
A2c Tb
= t cos(2𝜋nt∕Tb ) + sin(2𝜋nt∕Tb )
2 2𝜋n
Similarly, for the second interval
{ ( ) ( )}
A2 Tb n n
y(t) = c cos 2𝜋 (Tb − t) + cos 2𝜋 (2𝜏 + Tb − t) d𝜏
2 ∫t−Tb Tb Tb
( )
A2c Tb
= (2Tb − t) cos(2𝜋nt∕Tb ) − sin(2𝜋nt∕Tb )
2 2𝜋n
To summarise
⎧( T
)
⎪ t cos(2𝜋nt∕Tb ) + b sin(2𝜋nt∕Tb ) , 0 ≤ t ≤ Tb
A2 ⎪ 2𝜋n
y(t) = c ⎨( ) (12.50)
2 ⎪ T
(2Tb − t) cos(2𝜋nt∕Tb ) − b sin(2𝜋nt∕Tb ) , Tb ≤ t ≤ 2Tb
⎪ 2𝜋n
⎩
The symmetry of the two expressions for y(t) in Eq. (12.50) allows them to be merged into the following single
expression for y(t)
[ ( ) ( )]
⎧ A2c 2𝜋n Tb 2𝜋n
⎪ (Tb − |t − Tb |) cos t − sin t , 0 ≤ t ≤ 2Tb
y(t) = ⎨ 2 Tb 2𝜋n Tb
⎪0, Otherwise
⎩
[ ( ) ( )] ( )
A2 2𝜋n T 2𝜋n t − Tb
= c (Tb − |t − Tb |) cos t − b sin t rect (12.51)
2 Tb 2𝜋n Tb 2Tb
Telegram: @Persian_IEEEComSoc

This result is plotted in Figure 12.19 for n = 3. We see that the matched filter output y(t) has a maximum value
y(Tb ) = A2c Tb ∕2 at the decision instant t = T b . Note that this maximum value is the energy Eb of the binary pulse,
which is a sinusoidal signal of amplitude Ac and duration T b .

12.4.3 Clock Extraction

Decision instants at the receiver must be accurately spaced at intervals of the transmitted symbol duration T s .
This allows the matched filter output to be sampled at the optimum instants, for negligible ISI and maximum
output SNR. Small short-term deviations from the optimum timing instants are known as timing jitter. If this is
unchecked, especially in long-distance high-data-rate systems with many intermediate repeaters, it may accumu-
late sufficiently so that the timing error exceeds half the symbol duration, causing the decision instant to be set
 798 12 Pulse Shaping and Detection

y(t)
Eb
Eb = A2c Tb/2

0 t

–Eb
0 Tb 2Tb

Figure 12.19 Worked Example 12.6: output of filter matched to BPSK pulse of amplitude Ac and duration T b .

within the wrong symbol interval, entirely missing out one or more intervals. This problem is known as symbol
slip. It causes subsequent symbols to be in error until there is a realignment.
Clock or timing extraction is a process that seeks to derive from the incoming symbol stream a sinusoidal signal
of the correct phase and of a frequency equal to the symbol rate (Rs = 1/T s ). This sinusoid may then be passed
through a comparator – a zero-crossing detector – to give a square wave clock signal of period T s . The incoming
symbol stream is then decoded by arranging for the matched filter output to be sampled at every rising (or falling)
edge of the clock signal.
The need for the transmitted symbol stream to contain frequent voltage transitions (e.g. between ±V volts for
binary coding) is emphasised in our discussion of line coding in Chapter 10. When this is the case, the symbol
stream may contain a significant component at the sampling frequency f s (= Rs ), which can be directly filtered
out using a narrow bandpass filter tuned to f s . However, some symbol patterns may only contain a fraction or
multiple of the desired frequency component. Therefore, in general, the incoming symbol stream is passed through
a suitable nonlinear device, e.g. a square-law device, a full-wave rectifier, etc. From our discussion of nonlinear
distortion (Section 4.7.6), the output of such a device will contain the desired frequency component f s , which may
then be filtered out. Figure 12.20 shows one possible arrangement for clock extraction. A phase-locked loop (PLL),
Telegram: @Persian_IEEEComSoc

discussed in Sections 7.5.2 and 8.6.2 may be used in place of the narrow band filter to improve the phase match
between the clock signal used at the transmitter and that extracted at the receiver.

Received Narrow band

symbol stream Nonlinear BPF or PLL Threshold Clock signal
device fs comparator
fs /2, etc. fs, etc. fs only Ts

Figure 12.20 Clock extraction.

 12.5 Summary 799

(a)
t

1 1 0 1 0 0 Bit stream

(b) ´ Threshold level

Best decision instant

Figure 12.21 (a) Incoming distorted NRZ waveform; (b) corresponding eye diagram.

12.4.4 Eye Diagrams

An indication of the likelihood of decision error at the receiver due to the corruption of the incoming symbol
stream by undesirable noise and filtering can be readily displayed using eye diagrams. Figure 12.21a shows a
corrupted bipolar non-return-to-zero (NRZ) symbol stream, with adjacent symbol elements identified by different
line patterns. If all the incoming symbol elements are superimposed in one symbol interval, the result is the plot
shown in Figure 12.21b, which is called an eye diagram because it resembles the human eye. The eye diagram
of an actual transmission can easily be displayed on an oscilloscope. The symbols in successive intervals will be
automatically superimposed on the screen when the oscilloscope is triggered using the receiver’s clock signal.
Useful performance information provided by the eye diagram include:
● The width of the eye opening gives the timing error that can be tolerated in the sampling instants at the receiver.
The best sampling instant is at the centre of the eye opening.
● The slope of the opening gives an indication of the sensitivity of the transmission system to timing error.
● The height of the eye opening gives the noise margin of the system.
It is therefore obvious that the larger the eye opening the lower will be the symbol error ratio of the system.
Figure 12.22 demonstrates the impact of noise and timing error on the eye diagram of a binary system that uses
Telegram: @Persian_IEEEComSoc

raised-cosine-filtered pulses. A narrowing of the eye opening by these effects clearly indicates an increased proba-
bility of error. The eye diagram is indeed a very useful diagnostic tool for checking for the presence of timing error,
noise, and pulse distortion in a digital transmission system.

12.5 Summary

The design of a reliable communication system throws up many challenges which may be skilfully navigated
through a sound understanding of the interplay amongst key design parameters and the trade-offs involved, as
well as a good grounding in the tools and techniques needed to optimally exploit such trade-offs. One of the most
 800 12 Pulse Shaping and Detection

Perfect Timing error only

× ×

Noise only Noise and timing error

× ×

Figure 12.22 Effects of noise and timing error on eye diagram.

significant equations in information theory is the Shannon–Hartley law, which lays down the rule governing how
bandwidth and signal power may be exchanged in the design of a reliable transmission system affected by noise.
We presented a simple intuitive argument leading to the law and then paid a great deal of attention to evaluating
its implications. All practical systems fall short of the combined bandwidth and power efficiencies stipulated by
this law. We demonstrated through a worked example that the use of error control coding facilitates a more effi-
cient exchange between bandwidth and signal power than is possible by only switching modulation schemes and
therefore enables a closer approach to the benchmark laid down by the law.
We examined the techniques available for dealing with the challenges of inter-symbol interference (ISI) caused
by the finiteness of the transmission system’s bandwidth and the challenges of noise arising from the transmission
medium and the receiver’s front end. Various ISI-mitigating filtering techniques were discussed, and their merits
and drawbacks compared in terms of occupied bandwidth, sensitivity to timing error, signal power requirement,
and complexity or realisability in real-time. Of these, the most widely used is the raised cosine filter which permits
Telegram: @Persian_IEEEComSoc

an excellent trade-off between occupied bandwidth and complexity through its roll-off factor parameter.
Using a heuristic and nonmathematical approach, we derived a specification for the transfer function of a
matched filter which provides the best detection (in terms of maximising the output signal-to-noise ratio) of known
pulses in the presence of additive white Gaussian noise (AWGN). It turns out that the impulse response of this fil-
ter is simply a time-reversed version of the pulse delayed by the pulse duration. We showed that, quite remarkably,
both requirements of mitigating ISI and optimising symbol detection in the presence of noise can be met by using
a pair of square root raised cosine (RRC) filters, one at the transmitter and the other at the receiver.
In addition to noise and ISI, the communication channel will introduce channel distortion and the receiver may
experience small timing errors in the clock it uses to set precise decision instants for pulse detection. We briefly
discussed the techniques of equalisation and clock extraction and presented the eye diagram as a useful diagnostic
tool for investigating the effects of ISI, timing error, and noise on received pulses.
 Questions 801

In the next chapter, we undertake a step-by-step and comprehensive study of multiplexing strategies for
multi-user communication systems, which will include information on various international telecommunication
standards.

References

1 Lender, A. (1963). The duobinary technique for high-speed data transmission. Transactions of the American
Institute of Electrical Engineers, Part I: Communications and Electronics 82 (2): 214–218.
2 Newcombe, E.A. and Pasupathy, S. (1980). Effects of filtering allocation on the performance of a modified duobi-
nary system. IEEE Transactions on Communications, COM-28 (5): 749–752.
3 Kabal, P. and Pasupathy, S. (1975). Partial response signaling. IEEE Transactions on Communications, COM-23
(9): 921–934.
4 Walklin, S. and Conradi, J. (1999). Multilevel signalling for increasing the reach of 10 Gb/s lightwave systems.
Journal of Lightwave Technology 17 (11): 2235–2248.
5 Shannon, C.E. (1948). A mathematical theory of communication. The Bell System Technical Journal 27 379–423,
623–656.
6 Hartley, R.V.L. (1928). Transmission of information. The Bell System Technical Journal 7: 535–563.

Questions

1 A wideband audio signal of baseband frequencies 50 Hz to 7 kHz is processed in a 10-bit linear

analogue-to-digital conversion (ADC) at a sampling rate of 1.5 times the Nyquist rate. The resulting
bit stream is conveyed in a noiseless channel using:
(a) Binary (M = 2) signalling; and
(b) Quaternary (M = 4) signalling.
In each case, determine (i) the minimum required transmission bandwidth and (ii) the transmission band-
width when the channel has a raised cosine response with 𝛼 = 0.5

2 Determine the minimum SNR and hence signal power required for error-free transmission of the signal in
Question 12.1 over an AWGN channel of noise power per unit bandwidth 10−15 W/Hz and bandwidth
(a) 10 kHz
(b) 20 kHz
(c) 200 kHz.
Telegram: @Persian_IEEEComSoc

Comment on the trend of your results.

3 Sketch the impulse response of a matched filter for detecting each of the pulses shown in Figure Q12.3.

4 Determine and sketch the matched filter output for each of the pulses in Question 12.3. What is the maximum
value of each output pulse?

5 Repeat Questions 12.3 and 12.4 for a triangular pulse of amplitude A and duration T s .

6 Assuming zero modem implementation loss, a Nyquist channel (i.e. raised cosine filter of roll-off factor 𝛼 = 0)
and reliable communication when BER = 1 × 10−8 , determine the limit placed by the Shannon–Hartley law
 802 12 Pulse Shaping and Detection

g1(t), V g2(t), V g3(t), V

5 5 5

4 4 6
t, μs t, μs
4 2

–5 –5 –5

(a) (b) (c)

Figure Q12.3 Question 12.3.

on the maximum possible error control coding gain when the modulation scheme is QPSK and the code rate
is
(a) r = 9/10
(b) r = 1/2
(c) r = 1/8.
(Note: you may find the graphs of BER versus Eb /N o in Chapter 11 useful.)

7 a. ) Repeat Question 12.6 for a 256-FSK modulation scheme.

b) Comment on the trends of your results in Questions 12.6 and 12.7(a).

8 Making the same assumptions as in Question 12.6, determine the limit placed by the Shannon–Hartley law
on the maximum BER in the bit stream from a demodulator output to the input of a realisable error control
decoder if reliable communication is to be achieved for the pairs of modulation scheme and code rate listed
below. Comment on the trend of your results.
(a) 16-APSK, r = 9/10
(b) 16-APSK, r = 1/8
(c) 256-APSK, r = 9/10
(d) 256-APSK, r = 1/8.

9 A digital transmission system operates at a message bit rate Rb = 139 264 kb/s in AWGN using a raised cosine
filter of roll-off factor 𝛼 = 0.25. Reliable communication is set at BER = 10−6 . Determine how each of the
following implementations of this system compares with the Shannon–Hartley benchmark of Eq. (12.28)
and comment on your results. Assume that all modems have a modem implementation loss Lmil of 1 dB and
Telegram: @Persian_IEEEComSoc

determine BER versus Eb /N o values using Figures 11.39, 11.48, and 11.52 as appropriate.
(a) Binary ASK modulation and no error control coding.
(b) 64-APSK modulation and no error control coding.
(c) 1024-FSK modulation and no error control coding.
(d) 64-APSK modulation with error control coding using a codec of code rate 4/5 that can correct on average
seven bit errors in every 100 bits.

10 You are given that a 1024-APSK modem has a BER of 0.1 at Eb /N o = 11.6 dB. In view of the Shannon–Hartley
information capacity law, assess the possibility of realising a reliable transmission system that uses a
1024-APSK modem with modem implementation loss 1 dB in conjunction with a codec of code rate 9/10
that can correct on average one bit error in every 10 bits.
 Questions 803

11 The pulse g(t) in Eq. (12.47) is a sinusoidal signal that is constrained to duration T b through multiplication
by a rectangular window function. Using a raised cosine window function (instead of a rectangular window
function) produces the sinusoidal pulse
( )[ ( )] ( )
n 1 1 2𝜋 t
grc (t) = Ac cos 2𝜋 t + cos t rect
Ts 2 2 Ts Ts
which is a pulse that completes an even number n of cycles within its duration T s in the range −T s /2 to T s /2.
(a) Derive an expression for the output yrc (t) of a matched filter that receives grc (t).
(b) Make a graphical sketch of yrc (t) and discuss how it compares with the case of a rectangular-windowed
pulse examined in Worked Example 12.6.
Telegram: @Persian_IEEEComSoc
 805

13

Multiplexing Strategies

No, an enemy did not build our prisons and set our limits in life. We did all that ourselves.

In this Chapter
✓ A nonmathematical introduction of four classes of techniques for simultaneously accommodating multiple
users in a communication system.
✓ Frequency division multiplexing (FDM): you will see that FDM is indispensable to radio communication
services and learn various standardised hierarchical implementations of FDM telephony.
✓ Time division multiplexing (TDM): a step-by-step and detailed discussion of plesiochronous and syn-
chronous digital hierarchies and an introduction to ATM (asynchronous transfer mode).
✓ Code division multiplexing (CDM): a discussion of spread spectrum techniques, including a detailed
step-by-step graphical description of signal processing in CDM. You will learn the simplicity of this
free-for-all sharing strategy.
✓ Space division multiplexing (SDM): this indispensable strategy for global and cellular mobile telecoms is
discussed in the introductory section.
✓ Multiple access: a brief treatment of frequency division multiple access (FDMA), time division multiple
access (TDMA), and code division multiple access (CDMA) in the final section.

13.1 Introduction
The discussion in previous chapters concentrates mainly on the processing of a telecommunication signal emanat-
Telegram: @Persian_IEEEComSoc

ing from a single source. There are several reasons why a communication system must be able to simultaneously
handle signals from multiple and independent sources without mutual interference.
● To satisfy the communication needs of a larger number of people. Modern lifestyle has become very dependent
on telecommunication so that at any given time in an average city there will be a large number of people needing
to make a phone call, send a text message, access the Internet, hold a teleconference, etc. If the communication
system could only handle one signal at a time, and each user occupied the system continuously for an average
duration of three minutes then only 480 users per day could be serviced, assuming inconvenient times (such as
2.00 a.m.) are not rejected. If such a communication system served a city of one million people then at this rate
it would take nearly six years for every person to have just one three-minute access. Clearly, you couldn’t rely
on such a system to call an ambulance in a health emergency. By the time it reached your turn on the service
queue, you would either have fully recovered or been dead and buried.
Communication Engineering Principles, Second Edition. Ifiok Otung.
© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 806 13 Multiplexing Strategies

● To reduce the cost of the service to each user. This important consideration can be demonstrated by assuming a
satellite communication system built exclusively for telephony at a total cost of £300 m, which includes design,
construction, launching, and maintenance costs over a projected satellite lifetime of 10 years. Allowing a 16%
profit margin, the operator must earn (by charging users of the service) a total sum of £348 m during a period of
10 years or 5.22 million minutes. Excluding system idle time of eight hours per day – you would not normally
like to make or receive a phone call during sleeping hours – leaves us with 3.48 million income-yielding minutes
over which to recover £348 m. It is easy to see that if the system could only handle one call at a time then the
charge for each call would have to be £100 per minute. However, if we could somehow design the system to
handle up to 24 000 simultaneous calls then, assuming on average 20 000 users every minute, the operator’s
required earning could be spread out over this number of users, bringing down the charge per user to a mere
half a pence per minute.
● To allow the coexistence of a multiplicity of telecommunication services in each geographical area or city. Audio
broadcast, television broadcast, and mobile communication, to name but a few radio services, must operate
simultaneously and independently without mutual interference.
● To improve the exploitation of the available bandwidth of a transmission medium. For example, if a coaxial
cable of bandwidth 10 MHz is used to carry one voice signal (of bandwidth 4 kHz), only 0.04% of the cable
capacity is being utilised. As the communication distance and hence link cost increases it becomes more and
more important to dramatically increase the utilisation of the cable capacity by somehow packing many voice
signals onto the cable medium.
● To allow the use of identical radio systems for the provision of localised broadcast and communication services
in different geographical regions. For example, frequency modulation (FM) radio broadcast can be provided in
two different cities using the same carrier frequency of, say, 98.5 MHz.
To realise the above benefits, there are four multiplexing strategies that may be used separately, but frequently
in combination, to simultaneously accommodate multiple users and services in a common transmission medium.
Figure 13.1 provides an illustration of these resource-sharing techniques for N users. Three axes are used, namely
frequency, which represents the available bandwidth of the transmission medium; time, which represents the
instants of usage of the medium; and space, which represents the physical location of the medium.
● In time division multiplexing (TDM) the duration of usage of the transmission medium is divided into time slots
each of which is allocated to a single user. Thus, each of the N signals has exclusive use of the entire transmission
medium during the time slot allocated to it. TDM is briefly introduced in Sections 1.3.1.2 and 1.5.3.2, which you
may wish to review at this point. A useful analogy to TDM is the sharing of the use of a lecture room by four
different groups of students, each needing the room for a total period of one hour. We may draw up a schedule
dividing the room usage into one-hour time slots so that each group occupies the entire room once in turn; or we
may allocate 20-minute time slots so that it takes three slots for each group to complete their business, etc. There
is, however, an important difference between this analogy and the way TDM is implemented in communication
Telegram: @Persian_IEEEComSoc

systems. Here there is a noticeable sense of queuing and waiting for one’s turn, whereas in real TDM the time
slots are extremely short and are used to convey samples of each signal taken at regular and sufficiently short
intervals. Thus, users of the TDM system are totally oblivious of the time-sharing roster and the receiver can
reconstruct (without distortion) each of the original signals from their samples. See Chapter 9 if in any doubt
about sampling.
● Frequency division multiplexing (FDM) gives each user exclusive use of a separate frequency band (often referred
to as a channel) for all time. Ideally then, with an average channel bandwidth Bc , and total available bandwidth
Bt in the transmission medium, the maximum number of users that can be accommodated is
N = Bt ∕Bc (13.1)
 13.1 Introduction 807

Frequency
Frequency (b) SDM
(a) TDM

N Slot
2
3
1 2 3
…. Zone Time
Time N

Space
Space
Frequency Frequency
(c) FDM (d) CDM

Band
N

Time
All Users
3 Time
2
1

Space Space

f2
f2 f7 f3
f7 f3 f1
f1 f6 f4
f6 f4 f5 f2
f5 f2 f7 f3
f2 f7 f3 f1
f7 f3 f1 f6 f4
f1 f6 f4 f5
f6 f4 f5 f2
f5 f2 f7 f3
Telegram: @Persian_IEEEComSoc

f7 f3 f1
D f1 f6 f4
f6 f4 f5
f5

(e)
Figure 13.1 Multiplexing strategies: (a) time division multiplexing; (b) space division multiplexing; (c) frequency division
multiplexing; (d) code division multiplexing; (e) SDM in cellular telephony.
 808 13 Multiplexing Strategies

● Space division multiplexing (SDM) allocates the same frequency band or all the available bandwidth to more
than one user for all time, but user signals of the same frequency are confined to physically separate regions or
zones. In closed transmission media it means that each user has exclusive use of a separate line pair, whereas in
open media it requires that the radiated strength of each signal be negligible outside the signal’s geographical
region. In our lecture room analogy, we may apply SDM by allowing all four groups simultaneous use of the
room, but with each group seated sufficiently far apart at different corners of the room. As long as the students
follow a simple SDM rule of speaking softly then all groups can coexist with little mutual disturbance.
An important area of application of SDM is in cellular mobile communications where the same frequency bands
are reused many times. In this way a limited radio spectrum allocation is very efficiently utilised to meet a huge
demand in a given serving area, such as a city. For example, in the North American advanced mobile phone system
(AMPS) only 25 MHz in the ultra high frequency (UHF) band was available to one operator in a serving area. Of
this, 12.5 MHz was for transmission in the forward direction from base station to mobile, and a further 12.5 MHz
for transmission in the reverse direction. With 30 kHz per channel and making provision for control channels it
follows that only about 400 users could be accommodated simultaneously in the available bandwidth. This was
grossly inadequate to meet the demand for mobile services. The use of SDM dramatically increased capacity,
enabling the operator to handle tens of thousands of simultaneous calls.
A typical SDM or frequency reuse plan is shown in Figure 13.1e. The serving area is divided into small zones called
cells, each of which has one base station for communication with mobile units. A group of cells (enclosed in bold
lines in the diagram) across which the entire bandwidth allocation is used up is called a cluster. Figure 13.1e
shows a cluster size of 7, but it can also be 3, 4, 9, 12, or multiples of these. The available channels are shared
amongst the cells in each cluster. We identify the sets of channels as f 1 , f 2 , f 3 , etc. A mobile unit wanting to
make a call is assigned an available channel from the set allocated to its cell. Notice how the frequencies are
reused in cells separated by a distance D, meaning, for example, that calls can be made at the same time in
each of the shaded cells using exactly the same set of frequency bands. Obviously, radiated power in each cell
must be limited to minimise co-channel interference, i.e. interference between cells that use the same frequency.
The choice of cell diameter and cluster size is influenced by many factors, such as required capacity, acceptable
carrier-to-interference ratio, etc. A smaller cell size allows a particular frequency band to be reused more times
in the serving area, thus increasing capacity, but handover (the process of a mobile unit’s transmission being
changed from one channel to another as the mobile crosses a cell boundary) occurs more frequently.
● Code division multiplexing (CDM) is a kind of free-for-all sharing strategy in which multiple users transmit in the
same frequency band at the same time and in the same physical medium. The secret is that each user is assigned
a unique pseudorandom code sequence with which their signal is spread over a wide bandwidth giving it a
noise-like appearance. A target receiver equipped with exactly the same code sequence is able to extract the
wanted signal from the received mix of signals, and to effectively block out the unwanted signals from other
users. Returning to our lecture room analogy, the four groups of students may simultaneously share the entire
room in CDM fashion, with one group speaking, say, in German, another in Swahili, another in Igbo, and the
Telegram: @Persian_IEEEComSoc

remaining in Chinese. So long as the students understand only the language of their group then secure and
effective communication can take place, with only a slight inconvenience of background noise in each group.
It should be noted that these multiplexing strategies are rarely used in isolation. In fact, by taking a broad inter-
pretation of FDM we see that it is inherent in all radio communication systems to allow a multiplicity of services
in a given locality. Similarly, to allow the reuse of the same radio band in different regions (or localities in some
cases) of the world, SDM is inherent in nearly all radio systems, except, for example, international broadcasting at
high frequency (HF). Thus, if TDM is used on a satellite link at, say, 6 GHz then we could describe the system as
employing SDM/FDM/TDM. However, we will henceforth define multiplexing more restrictively in terms of how
 13.2 Frequency Division Multiplexing 809

multiple signals are combined for transmission on a common link. Therefore, the satellite system in this example
is regarded simply as a TDM system.
In the remaining sections of this chapter we consider FDM, TDM and CDM in detail, and study the implemen-
tation of various standard FDM and TDM hierarchies.

13.2 Frequency Division Multiplexing

13.2.1 General Concepts
Frequency division multiplexing stacks a number of narrowband signals in nonoverlapping frequency bands and
transmits them simultaneously in a common transmission medium. The required frequency translation of each
baseband signal can be achieved using any of the modulation techniques studied in Chapters 7 and 8, such as FM,
phase modulation, and amplitude modulation (AM) or its variants, namely double sideband (DSB) and single
sideband (SSB). The use of SSB modulation minimises bandwidth and power requirements and hence the cost
of high-capacity long-distance transmission systems. However, for low-capacity short-haul systems (e.g. in rural
telephony) the use of AM with its simple modulation and demodulation circuits may yield a more cost-effective
FDM implementation. The following discussion is confined to the use of SSB, a much more prevalent choice for
which international standards exist in FDM telephony. It is a straightforward matter to extend the principle to
other modulation techniques. A detailed discussion of SSB modulation and demodulation is presented in Section
7.7.2. For convenience, Figure 13.2 shows a block diagram of an SSB modulator. Note that it contains pre- and
post-modulation filters of the indicated passbands. The roles of these filters will become clear in the following
discussion.
Figure 13.3a shows an FDM multiplexer that combines N independent input signals v1 (t), v2 (t), …, vN (t) to
give a composite (FDM) signal vfdm (t). Each input signal occupies the baseband spectrum with frequencies in the
range f a to f b . Clearly, combining these signals in their baseband form would result in an unacceptable state of
complete mutual interference. To avoid this, each signal vi (t), i = 1, 2, …, N, is first passed through a lower-sideband
SSB modulator supplied with a unique carrier f ci . The resulting modulation converts vi (t) to the signal vssbi (t),
which occupies an exclusive frequency band or channel f ci − f b → f ci − f a . This frequency translation process is
illustrated in Figure 13.3b for i = 1, 2. A symbolic baseband spectral shape – a triangle by convention – is used for
the illustration. Note that the spectrum of vssbi (t) is inverted relative to that of vi (t) because it is the lower sideband
that is taken by the post-modulation filter from the product of vi (t) and the carrier. An erect spectrum of vssbi (t) in
the frequency band f ci + f a → f ci + f b would be obtained if the upper sideband (USB) were taken. Following this
modulation, the SSB signals are added together in a summing device to yield the desired FDM signal

vfdm (t) = vssb1 (t) + vssb2 (t) + · · · + vssbN (t) (13.2)

Telegram: @Persian_IEEEComSoc

ʋ(t) Post-modulation ʋssb(t)

Pre-modulation
filter (fa → fb) × filter
Message (fc – fb → fc – fa) SSB
signal signal
Carrier
(fc)

Figure 13.2 (Lower sideband) SSB modulator.

 810 13 Multiplexing Strategies

ʋ1 (t) SSB Modulator ʋssb 1 (t)

(f c1 )

(a) ʋ2 (t) SSB Modulator ʋssb 2 (t)

(f c2 ) ʋfdm (t)
Σ
FDM signal

ʋN (t) SSB Modulator ʋssbN (t)

(f cN )

N message signals N modulated carriers

|V1(f)| |Vssb1(f)|

SSB
fc1 fc1
f f
0 fa fb 0 fc1 – fb fc1 – fa
(b)
|V2(f)| |Vssb2(f)|

SSB
fc2 fc2
f f
0 f
a fb 0 fc2 – fb fc2 – fa

|Vfdm(f)|

(c)
…...

f
0 fc1 fc2 f cN–1 fcN

G G G
B fdm

Figure 13.3 FDM: (a) multiplexer; (b) frequency translation effect of SSB modulation; (c) spectrum of FDM signal.
Telegram: @Persian_IEEEComSoc

The remarkable outcome of this multiplexing process is that vfdm (t) is a composite signal containing N inde-
pendent signals each of which can be extracted at the receiver without mutual interference. Figure 13.3c shows
the spectrum of vfdm (t), denoted V fdm (f ). An arrangement for de-multiplexing vfdm (t) is shown in Figure 13.4. The
FDM signal is connected to a bank of N bandpass filters. Clearly, the filter with passband f ci − f b → f ci − f a passes
the ith component vssbi (t) in Eq. (13.2) and blocks all others. The signal vssbi (t) so extracted is demodulated in an
SSB demodulator which is supplied with a carrier of frequency f ci . This yields the original signal vi (t). In this way,
all the multiplexed signals are successfully recovered.
There are a number of conditions that must be satisfied for the above implementation of FDM to be free of
interference in any of the channels.
 13.2 Frequency Division Multiplexing 811

Bandpass filter ʋssb1(t) SSB demodulator ʋ1(t)

(fc1 – fb → fc1 – fa) (fc1)

Bandpass filter ʋssb2(t) SSB demodulator ʋ2(t)

ʋfdm(t) (fc2 – fb → fc2 – fa) (fc2)

Bandpass filter ʋssbN(t) SSB demodulator ʋN(t)

(fcN – fb → fcN – fa) (fcN)

Figure 13.4 FDM demultiplexer.

● Each of the N input signals to the multiplexer must be bandlimited, with frequency components in the range
f a → f b , where
0 < fa < fb < ∞. (13.3)
If this condition is not satisfied and f b is infinite then the signals cannot be confined within exclusive bands.
On the other hand, if f a = 0 then SSB cannot be used for frequency translation since it becomes impossible
to separate the sidebands using a realisable filter. To satisfy the condition of Eq. (13.3), a pre-modulation filter
is employed to remove all nonessential frequency components below f a and above f b in each input signal. In
speech telephony, for example, f a = 300 Hz and f b = 3400 Hz. Video signals contain essential components down
to DC. This means that f a = 0. It is for this reason that SSB cannot be used for obtaining the FDM of television
signals.
● The carrier frequencies f c1 , f c2 , …, f cN used by the bank of SSB modulators in the multiplexer must be sufficiently
spaced to allow a frequency gap, called a guard band (GB), between adjacent spectra of the SSB signals that
constitute the FDM signal. Without such a gap, a nonrealisable brickwall bandpass filter would be required at
the receiver to extract each of the SSB signals. In Figure 13.3c a guard band G is shown. Thus, the bandwidth of
each signal, and hence the spacing of the carrier frequencies, is given by
B = fb − fa + G. (13.4)
With the bank of carrier frequencies starting at f c1 for the lowest channel, it follows that the value of the ith
carrier is
fci = fc1 + B(i − 1) (13.5)
and the bandwidth of the composite FDM signal is
Telegram: @Persian_IEEEComSoc

Bfdm = NB
= N(fb − fa + G). (13.6)
A GB of 900 Hz is used in speech telephony, with f b and f a as given above, so that the channel bandwidth
B = 4 kHz.
● Amplifiers in the transmission system must be operated in the linear region of their transfer (i.e. input to output)
characteristic. Any nonlinearity causes harmonic and intermodulation products to be generated in one chan-
nel that may fall in the frequency interval of some other channel, giving rise to noise. See Section 4.7.6 for a
discussion of this effect.
 812 13 Multiplexing Strategies

● The post-modulation filter must suppress the USB; otherwise, any remnants will interfere with the next higher
channel, causing unintelligible crosstalk – since the interfering USB is inverted relative to the wanted lower
sideband (LSB) of the next channel. In practice, perfect elimination of the USB is not possible, and it is sufficient
for the post-modulation filter to reduce the USB by 60 dB or more relative to the LSB.

13.2.2 Demerits of Flat-level FDM

We have so far presented a flat-level FDM in which N signals are frequency translated in one step using N carriers
uniformly spaced in frequency from f c1 to f c1 + B(N − 1). This approach has several serious drawbacks when used
in the implementation of high-capacity systems where N may be very large, e.g. up to 10 800 for FDM telephony.
First, consider the design of the post-modulation filter, which is required to suppress one of the sidebands at the
output of the product modulator. Figure 13.5 shows a piecewise linear approximation of the response of this filter
(for the ith channel, i = 1, 2, …, N), indicating the passband for the LSB, the transition width, and the stopband to
block the USB. A standard measure of the selectivity of a filter is its quality factor Q, which is defined by
Filter centre frequency fr
Q= =
Filter bandwidth fb − fa
fci − 12 (fa + fb )
=
fb − fa
fc1 + (i − 1)B − 12 (fa + fb )
= (13.7)
fb − fa
A related measure of filter performance is the required steepness in the roll-off of the filter’s response in order to
block the unwanted sideband while passing the wanted sideband. This measure is best given by the ratio between
the centre frequency of the transition band and the transition width, which we will call the slope factor ℤ
Transition centre frequency f
ℤ= = ci
Transition width 2fa
f + (i − 1)B
= c1 (13.8)
2fa
Note that a brickwall filter (an unrealisable ideal device) has zero transition width and hence ℤ = ∞. Eqs. (13.7)
and (13.8) show that both Q and ℤ increase with i, the channel number. For example, consider the post-modulation
filters required for the 10th and 10 000th channels of a high-capacity FDM telephony system, with f c1 = 64 kHz,
Telegram: @Persian_IEEEComSoc

Post-modulation
filter response
LSB USB

f
fr fci
Stopband Passband Stopband
2fa 2fa
fb – fa
Transition width Transition width

Figure 13.5 Post-modulation filter response: a piecewise linear approximation.

 13.2 Frequency Division Multiplexing 813

f a = 0.3 kHz, f b = 3.4 kHz, and B = 4 kHz. Substituting these values in the above equations yields
Q = 32, ℤ = 167; for i = 10
Q = 12 922, ℤ = 66 767; for i = 10 000
Note that the channel i = 10 000 requires a filter with very high values of quality and slope factors, which is both
expensive and difficult to achieve. So we see that in a flat-level FDM it is difficult to realise post-modulation filters
for the higher-frequency channels. The same argument holds for the bandpass filters required in the demultiplexer
at the receiver. The other problems posed by flat-level FDM implementation include the following.
● Provision would have to be made for generating N different carrier frequencies at the transmitter, and the same
number at the receiver. Considering that the carrier frequencies are required to be highly stable (to guaran-
tee distortionless transmission of especially nonvoice signals), it is obvious that such a system would be very
complex for large N.
● The required Q and ℤ factors of the filter in each channel depend on the channel number, according to Eqs. (13.7)
and (13.8). So, no two filters are identical, leading to N different filter designs. Building and maintaining a system
with, say, 10 800 unique filters is, to say the least, very cumbersome.
● The structure of each FDM system depends on the number of channels. Standardisation is therefore lacking.
Standardisation makes it easier and cheaper to set up systems of various capacities by using a small set of stan-
dardised equipment obtainable from various manufacturers.
● The summing device at the multiplexer is fed by N different sources, whereas at the receiver the incoming FDM
signal is connected to N different bandpass circuits. As N increases, the problem of loading becomes significant,
necessitating, for example, a much higher signal level to drive the bandpass filters.
● We require N different pairs of wires to carry the signals to the single multiplexing point. This can be very
expensive (and unsightly if carried on overhead poles) for providing telephone services to, say, N different homes.
Preferably, we would like to perform the multiplexing in stages and use a single wire pair to carry signals from
a small cluster of homes.
To overcome the above problems FDM is implemented in a hierarchy. Hierarchical arrangements were standard-
ised for FDM telephony, which we discuss in detail in Section 3.2.4. The way nonvoice signals were accommodated
within these FDM telephony hierarchy plans are also be briefly addressed in that section. However, we must first
make an important clarification about the future of FDM technology.

13.2.3 Future of FDM Technology

We should point out straightaway that the hierarchical FDM telephony discussed in the next section is an analogue
technology that has no place in our twenty-first-century digital telecommunication network. The future of FDM
in general is, however, assured. For as long as radio communication endures (and this is guaranteed), there will
Telegram: @Persian_IEEEComSoc

always be the need to allocate different frequency bands to different services and users (as in two-way radio systems
and mobile cellular telephony). These are all instances of FDM. Also, wavelength division multiplexing (WDM) is
an application of FDM in optical fibre communication that has a very promising future. FDM will also continue to
be an important multiple access technique in satellite communications, for example. Here, a satellite transponder
is partitioned into frequency bands, which are assigned to users who (almost certainly) use digital transmission
techniques within their allotted band.
The deployment of (analogue) FDM telephony reached its peak in the mid-1970s. Since then, developments in
digital transmission techniques with all their advantages led to a rapid digitalisation of the telephone network and
a replacement of FDM telephony with TDM. The telephone network in most countries is now practically 100%
digital. There was a period of transition from analogue to digital transmission, which lasted for a few years because
of the huge investment that had been made in analogue transmission technology.
 814 13 Multiplexing Strategies

The International Telecommunication Union (ITU) specified some transition equipment, namely transmulti-
plexers (TMUXs), FDM codecs, and transition modems, that allowed the interconnection of digital and analogue
systems during this transition period. The TMUX transformed an FDM telephony signal to TDM telephony in one
direction of transmission and performed the opposite conversion in the other direction. For example, a 60-channel
TMUX transformed a supergroup (SG) signal (discussed below) to two 2048 kb/s TDM signals, and vice versa,
whereas a 24-channel TMUX converted between two group signals (see below) and one 1544 kb/s TDM signal. An
FDM codec was used for digitising an FDM signal before transmission over a digital link. At the other end of the
link, the codec converted the incoming bit stream back into the original FDM signal. The ITU recommended two
types of transition modems for high-speed data transfer over an analogue link, namely the data-in-voice (DIV)
modem and the data-over-voice (DOV) modem. A suitable carrier was modulated by the digital signal in both
modems. The DIV modem displaced several FDM channel assemblies, whereas the DOV modem placed the signal
above the frequency band occupied by the voice signals and so did not replace them.
Therefore, although the future of FDM technology in general is assured, the material on FDM telephony hierar-
chy presented in the next section is a (now obsolete) twentieth-century analogue technology and may be skipped
in a tight curriculum.

13.2.4 FDM Hierarchies

There were three different hierarchical implementations of FDM telephony in the world, namely the UK System,
the European System, and the Bell System (used in North America). These schemes are identical at the first two
levels of the hierarchy and differ only at the higher levels. It is important to bear this in mind as we initially
discuss these first two levels, which achieve the multiplexing of 60 voice channels into an FDM signal known as
the supergroup (SG) signal. The three FDM standards are realised by multiplexing SG signals in different ways,
which we discuss in the relevant sections below.
Figure 13.6 shows how a SG signal is generated using two stages or levels of multiplexing. A notation f ci,j has
been used to identify the carrier frequencies in the SSB modulators. It denotes the carrier frequency used in the
ith level of the FDM hierarchy to translate the jth signal combined at that level. In the first multiplexing level, 12
voice signals each of which contains frequency components from f a = 0.3 kHz to f b = 3.4 kHz are frequency trans-
lated and summed to give an FDM signal known as a group signal. The 12 carriers used are spaced at 4 kHz with
frequencies 64, 68, 72, …, 108 kHz. This carrier spacing allows a 900 Hz GB between the translated voice signals
and gives a nominal voice channel bandwidth of 4 kHz. Since it is the LSB that is selected, it follows that the first
voice signal is translated by the 64 kHz carrier to the band 60 → 64 kHz. Note that this band includes the 900 Hz
GB, made up of two gaps of 600 Hz and 300 Hz on either side of the voice spectrum. The second voice signal is
translated by the 68 kHz carrier to the band 64 → 68 kHz, and so on, until the 12th voice signal, which is trans-
lated by the 108 kHz carrier to the band 104 → 108 kHz. Thus, the group signal, which comprises 12 independent
and noninterfering voice signals, lies in the frequency band 60 → 108 kHz, and has a bandwidth of 48 kHz. The
Telegram: @Persian_IEEEComSoc

spectrum of a group signal is shown in Figure 13.7a.

A standard existed that allowed a 33% increase in system capacity with the same transmission bandwidth by
packing 16 voice channels into the 48 kHz group signal. To do this, each voice signal is restricted to the frequencies
0.25 → 3.05 kHz, and the frequency translation is performed in a manner that allows a GB of 0.2 kHz between each
of the translated voice bands. The voice channel bandwidth is therefore 3 kHz. The use of a 16-channel group signal
(i.e. 3 kHz voice channels) was restricted mostly to submarine cable installations where equipment cost was very
high. We will henceforth concentrate on the 4 kHz voice channel, which was prevalent but will consider further
details of the 3 kHz channel in Question 13.1.
In the second level, five group signals are frequency translated and summed to give a SG signal. The carriers
used to effect these translations are spaced apart by 48 kHz and have frequencies 420, 468, 516, 564, and 612 kHz.
Clearly, the first group signal (of frequencies f a = 60 kHz to f b = 108 kHz) is translated by the carrier of frequency
 13.2 Frequency Division Multiplexing 815

12 voice signals
(0.3 → 3.4 kHz)

ʋ 1 (t) SSB modulator

(fc1,1 = 64 kHz)

ʋ 2 (t) SSB modulator

(fc1,2 = 68 kHz)

ʋ 3 (t) 5 group signals

SSB modulator
(fc1,3 = 72 kHz) (60 → 108 kHz)

ʋ 4 (t) SSB modulator

(fc1,4 = 76 kHz) SSB modulator
(fc2,1 = 420 kHz)
ʋ 5 (t) SSB modulator
(fc1,5 = 80 kHz) SSB modulator
ʋ 6 (t) (fc2,2 = 468 kHz)
SSB modulator Supergroup signal
(fc1,6 = 84 kHz) (312 → 552 kHz)
Σ Σ
SSB modulator
ʋ 7 (t) (fc2,3 = 516 kHz)
SSB modulator
(fc1,7 = 88 kHz) SSB modulator
ʋ 8 (t) (fc2,4 = 564 kHz)
SSB modulator
(fc1,8 = 92 kHz) SSB modulator
(fc2,5 = 612 kHz)
ʋ 9 (t) SSB modulator
(fc1,9 = 96 kHz)
ʋ 10 (t) SSB modulator
(fc1,10 = 100 kHz)
ʋ 11 (t) SSB modulator
(fc1,11 = 104 kHz)
ʋ 12 (t) SSB modulator
(fc1,12 = 108 kHz)
Telegram: @Persian_IEEEComSoc

FDM Level 1 FDM Level 2

(Group) (Supergroup)
Figure 13.6 First two levels of FDM hierarchy: generation of supergroup (SG) signal.

f c2,1 = 420 kHz to the band f c2,1 − f b → f c2,1 − f a , which is 312 → 360 kHz. The other group signals are similarly
translated, and the fifth group signal is translated to the band 504 → 552 kHz. Thus, the SG signal occupies the
frequency band 312 → 552 kHz, has a bandwidth of 240 kHz, and contains 60 voice channels. The spectrum of
this SG signal is shown in Figure 13.7b. Note particularly that the spectrum is erect, having undergone double
inversion in the two multiplexing stages.
 816 13 Multiplexing Strategies

(a) 4 kHz voice

channel

f, kHz
60 64 68 72 76 80 84 88 92 96 100 104 108
48 kHz bandwidth

(b) 48 kHz group

channel

f, kHz
312 360 408 456 504 552
240 kHz bandwidth

Figure 13.7 (a) Spectrum of group signal; (b) spectrum of SG signal.

The advantages of this two-level hierarchical multiplexing are immediately obvious. The most stringent filter
performance required is for the 12th voice channel at the first multiplexing level and has a quality factor of 34.
Using flat-level FDM to combine 60 voice signals would require a filter with Q = 96 for the 60th channel. Secondly,
standardised equipment can be used for the hierarchical implementation. A group signal is generated using a
channel translating equipment (CTE) shown in Figure 13.8a, and a SG signal by a group translating equipment
(GTE) in Figure 13.8b. This means that a 60-channel FDM system can be set up very quickly using only five CTEs
and one GTE connected, as shown in Figure 13.8c.
To build systems of higher capacity, we must go to higher levels in the FDM hierarchy, and this is where the
adopted standards differed. The ITU recommended two procedures. The European system corresponded to ITU
Procedure 1, the UK system to ITU Procedure 2, whereas the Bell system used in North America did not conform
to either of the two recommendations.

13.2.4.1 UK System
Figure 13.9 shows a self-explanatory block diagram of the supergroup translating equipment (STE) in the UK sys-
tem. Fifteen SG signals vsg1 (t), vsg2 (t), …, vsg15 (t) are multiplexed to give one hypergroup (HG) signal vhg (t). Clearly,
Telegram: @Persian_IEEEComSoc

vhg (t) contains 60 × 15 = 900 independent and noninterfering voice channels. Examining Figure 13.9, we can make
the following observations on the operation of the STE in this system.

● The first SG signal vsg1 (t) is connected directly to the summing point without any frequency translation. It will
therefore have an erect spectrum in the range 312 → 552 kHz within the spectrum of the HG signal vhg (t). The
remaining SGs, namely vsgi (t), i = 2, 3, …, 15, all have inverted spectra at the output since they are frequency
translated using a carrier of frequency f c3,i . Thus, vsgi (t) occupies the frequency range f c3,i − 552 → f c3,i − 312 kHz
within the spectrum of vhg (t).
● The carriers used for the frequency translation of vsg2 (t), vsg3 (t), …vsg15 (t) have frequencies spaced apart by
248 kHz and starting from 1116 kHz to 4340 kHz. Since the SGs have a bandwidth of 240 kHz, it follows that
the spectrum of the composite HG signal includes a GB of 8 kHz between each of the component spectra, except
between the first and second component spectra, which are separated by 12 kHz. You should be able to see
 13.2 Frequency Division Multiplexing 817

Channel
12 voice Translating One group signal
(a)
signals Equipment (60 → 108 kHz)
(CTE)

Group
5 group Translating One SG signal
(b)
signals Equipment (312 → 552 kHz)
(GTE)

Voice signal inputs

ʋ1 (t)
CTE
ʋ12 (t)

ʋ13 (t)
CTE
ʋ24 (t)

ʋ25 (t)
60-channel
(c) CTE GTE
FDM signal
ʋ36 (t)

ʋ37 (t)
CTE
ʋ48 (t)

ʋ49 (t)
CTE
ʋ60 (t)
Telegram: @Persian_IEEEComSoc

Figure 13.8 (a) CTE; (b) GTE; (c) 60-channel FDM.

that vsg2 (t) is translated to the band 564 → 804 kHz, whereas vsg1 (t) of frequency range 312 → 552 kHz is directly
added, hence the separation of 12 kHz (i.e. 564–552) between the two bands.
● Following the above observations, the spectrum of the HG signal can be easily sketched, as was done in
Figure 13.7a for a group signal. This is left as an exercise in Question 13.2, which you may wish to tackle
at this point. Note that the last SG signal vsg15 (t) is translated using a 4340 kHz carrier from its baseband at
312 → 552 kHz to the band 3788 → 4028. Thus, reckoning from the location of vsg1 (t), we see that the HG signal
occupies the band 312 → 4028 kHz. It therefore carries 900 voice signals in a bandwidth of 3716 kHz.
 818 13 Multiplexing Strategies

SG signals HG signal
(312 → 552 kHz) (312 → 4028 kHz)
ʋsg1 (t)

ʋsg2 (t) SSB modulator

(fc3,2 = 1116 kHz)

ʋsg3 (t) SSB modulator

(fc3,3 = 1364 kHz)

ʋsg4 (t) SSB modulator

(fc3,4 = 1612 kHz)

ʋsg5 (t) SSB modulator

(fc3,5 = 1860 kHz)

ʋsg6 (t) SSB modulator

(fc3,6 = 2108 kHz)

ʋsg7 (t) SSB modulator

(fc3,7 = 2356 kHz)

ʋsg8 (t) SSB modulator

(fc3,8 = 2604 kHz) Σ ʋhg(t)

ʋsg9 (t) SSB modulator

(fc3,9 = 2852 kHz)

ʋsg10 (t) SSB modulator

(fc3,10 = 3100 kHz)

ʋsg11 (t) SSB modulator

(fc3,11 = 3348 kHz)

ʋsg12 (t) SSB modulator

(fc3,12 = 3596 kHz)

ʋsg13 (t) SSB modulator

(fc3,13 = 3844 kHz)

ʋsg14 (t) SSB modulator

(fc3,14 = 4092 kHz)

ʋsg15 (t) SSB modulator

Telegram: @Persian_IEEEComSoc

(fc3,15 = 4340 kHz)

SG translating equipment (STE)

Figure 13.9 UK System: generation of hypergroup (HG) signal.

The HG signal is used in a fourth level of the FDM hierarchy as a building block to assemble more voice channels
depending on the required capacity. A few examples are given below.

● Multiplexing two HG signals, as shown in Figure 13.10a, to obtain an 1800-channel FDM signal with frequencies
in the range 312 → 8120 kHz, and a bandwidth of 7.808 MHz. This FDM signal is used to frequency-modulate a
suitable high-frequency carrier and transmitted by radio.
 13.2 Frequency Division Multiplexing 819

HG1
1800-channel FDM signal
(a) Σ (312 → 8120 kHz)
SSB modulator
HG2 (fc4,2 = 8432 kHz)

HG1

SSB modulator 2700-channel FDM signal

(b) HG2 (fc4,2 = 8432 kHz) Σ (312 → 12336 kHz)

SSB modulator
HG3 (fc4,3 = 12648 kHz)

Hypergroup signals
(312 → 4028 kHz)
SSB modulator
HG1 (fc4,1 = 8432 kHz)

SSB modulator
HG2 (fc4,2 = 12648 kHz)

SSB modulator SSB modulator

HG3 (fc = 12648 kHz) (fc4,3 = 25520 kHz)

SSB modulator SSB modulator

HG4 (fc = 12648 kHz) (fc4,4 = 30360 kHz) 10800-channel
FDM signal
SSB modulator SSB modulator
HG5 (4404 → 59580 kHz)
(fc = 12648 kHz) (fc4,5 = 35200 kHz)

SSB modulator SSB modulator

HG6 (fc = 12648 kHz) (fc4,6 = 39600 kHz)

HG7
SSB modulator
(fc = 12648 kHz)
SSB modulator
(fc4,7 = 44000 kHz) Σ
SSB modulator SSB modulator
HG8 (fc = 12648 kHz) (fc4,8 = 48400 kHz)
Telegram: @Persian_IEEEComSoc

SSB modulator SSB modulator

HG9 (fc = 12648 kHz) (fc4,9 = 55000 kHz)

SSB modulator SSB modulator

HG10 (fc = 12648 kHz) (fc4,10 = 59400 kHz)

SSB modulator SSB modulator

HG12 (fc = 12648 kHz) (fc4,11 = 63800 kHz)

SSB modulator SSB modulator

HG12 (fc = 12648 kHz) (fc4,12 = 68200 kHz)

Figure 13.10 Examples of UK system high capacity FDM built using HG signals as building blocks: (a) 1800-channel
system; (b) 2700-channel system; (c) 10 800-channel system.
 820 13 Multiplexing Strategies

● Multiplexing three HG signals, as shown in Figure 13.10b, to obtain a 2700-channel FDM signal with frequencies
in the range 312 → 12 336 kHz, and a bandwidth of 12.024 MHz. This signal could be conveyed as is on a coaxial
cable system or by radio using FM.
● A 3600-channel FDM signal with frequencies in the range 312 → 16 612 kHz and a bandwidth of 16.3 MHz,
which is obtained by multiplexing four HG signals. It was suitable for transmission on 18-MHz coaxial cable
systems.
● A 10 800-channel FDM signal occupying the frequency band 4404 → 59 580 kHz and resulting from the
multiplexing of 12 HG signals. This was recommended for transmission on 60 MHz coaxial cable systems.
Figure 13.10c shows in a self-explanatory manner how the signal was assembled. You will have an opportunity
in Question 13.3 to determine the most stringent filter performance (in terms of Q and ℤ factors) required in
the entire 10 800-channel hierarchical FDM system, and to compare this with the case of flat-level FDM.

13.2.4.2 European System

In the European system, a two-stage multiplexing procedure was employed to assemble (starting from SG signals)
a 900-channel FDM signal, called supermastergroup (SMG). Figure 13.11 shows a block diagram of an SMG gener-
ator. In the first stage, which corresponds to level 3 of the overall FDM hierarchy, five SG signals are translated and
combined to give one mastergroup (MG) signal occupying the band 812 → 2044 kHz. The translation uses carriers
at frequencies 1364, 1612, 1860, 2108, and 2356 kHz. Thus, an MG signal has bandwidth 1232 kHz, contains 300
voice channels and includes a GB of 8 kHz between its component translated SG signals.
In the second stage, which corresponds to level 4 in the overall FDM hierarchy, three MG signals are translated
and combined into an SMG signal occupying the band 8516 → 12 388 kHz. The translation uses carriers at frequen-
cies 10 560, 11 880, and 13 200 kHz. Thus, an SMG signal has bandwidth 3872 kHz, contains 900 voice channels
and includes a GB of 88 kHz between the spectra of the translated MG signals that form it.
Higher-capacity systems were built by multiplexing various combinations of the MG and SMG signals at level
5 of the overall FDM hierarchy. For example, four MGs were combined to give a 1200-channel signal in the
band 312 → 5564 kHz. Two SMGs were combined to give an 1800-channel signal in the band 316 → 8204 kHz. A
2700-channel system of baseband 316 → 12 388 kHz was realised by combining three SMG signals, a 3600-channel

5 SG signals 3 MG signals SMG signal

(312 → 552 kHz) (812 → 2044 kHz) (8516 → 12388 kHz)

SSB modulator MG1 SSB modulator

SG1 (fc3,1 = 1364 kHz) (fc4,1 = 10560 kHz)

SSB modulator
SG2 (fc3,2 = 1612 kHz)
Telegram: @Persian_IEEEComSoc

MG2
Σ Σ
SSB modulator SSB modulator
SG3 (fc3,3 = 1860 kHz) (fc4,2 = 11880 kHz) SMG

SSB modulator
SG4 (fc3,4 = 2108 kHz)

SSB modulator MG3 SSB modulator

SG5 (fc3,5 = 2356 kHz) (fc4,3 = 13200 kHz)

SG translating equipment MG translating equipment

(STE) (MTE)

Figure 13.11 European System. Generation of mastergroup (MG) and supermastergroup (SMG) signals.
 13.2 Frequency Division Multiplexing 821

10 SG signals UMG signal

(312 → 552 kHz) (564 → 3084 kHz)

SSB modulator
SG1 (f c3,1 = 1116 kHz)

SSB modulator
SG2 (f c3,2 = 1364 kHz)

SSB modulator
SG3 (f c3,3 = 1612 kHz)

SSB modulator
SG4 (f c3,4 = 1860 kHz)

SG5
SSB modulator
(f c3,5 = 2108 kHz) Σ UMG

SSB modulator
SG6 (f c3,6 = 2356 kHz)

SSB modulator
SG7 (f c3,7 = 2652 kHz)

SSB modulator
SG8 (f c3,8 = 2900 kHz)

SSB modulator
SG9 (f c3,9 = 3148 kHz)

SSB modulator
SG10 (f c3,10 = 3396 kHz)

SG translating equipment (STE)

Figure 13.12 Bell System. Generation of Bell U600 MG (UMG) signal.

system of baseband 316 → 17 004 kHz was realised by combining four SMG signals, and a 10 800-channel system
of baseband 4332 → 59 684 kHz was realised by combining 12 SMG signals.
Telegram: @Persian_IEEEComSoc

13.2.4.3 Bell System

Figure 13.12 shows the implementation of level 3 of the overall FDM hierarchy in the Bell system. Ten SG signals
are translated and combined into one MG signal occupying the band 564 → 3084 kHz and referred to as the U600
mastergroup (UMG) signal. The carrier frequencies used for the translation are as indicated in the diagram, and
are spaced at 248 kHz, except at the seventh carrier, where the frequency increment is 296 kHz. The UMG signal
therefore has a bandwidth of 2520 kHz, contains 600 voice channels, and includes a GB of 8 kHz between the
constituent supergroups, except between the sixth and seventh, where the gap is 56 kHz.
Systems of various capacities were built using mainly the UMG. For example, six MG signals may be multi-
plexed to form what is referred to as a jumbogroup (JG), which contains 3600 voice channels and occupies the band
564 → 17 548 kHz. To form the JG, one UMG is connected directly to the summing point without frequency trans-
lation, whereas the other five UMGs are added after being frequency translated using carriers at frequencies 6336,
 822 13 Multiplexing Strategies

9088, 11 968, 14 976, and 18 112 kHz. A 10800-channel system in the band 3124 → 51 532 kHz could be realised by
multiplexing 18 UMGs or three JGs.

13.2.4.4 Nonvoice Signals

Only telephony speech signals have been considered thus far. However, the following nonvoice signals were also
transmitted in the hierarchical FDM systems discussed above.

13.2.4.4.1 Pilot Tones

The translating equipment at the transmitter inserts a reference pilot sinusoid into the composite signal at its
output. This signal can be monitored at various points in the transmission system for fault location – a missing
pilot indicates a malfunctioning of the relevant translating equipment. The pilot signal is also used at repeater
stations and receivers for automatic gain control to compensate for variations in the attenuation along the cable
path due especially to seasonal temperature changes. A CTE adds a group reference pilot at 84.080 kHz, a GTE adds
a supergroup reference pilot at 411.920 kHz, an STE adds a hypergroup or mastergroup reference pilot at 1552 kHz,
and an MTE (mastergroup translating equipment) adds an SMG reference pilot at 11092 kHz. Note that the group
and SG reference pilots lie in the 0.9 kHz gaps between translated voice bands, whereas the 1552 and 11 092 kHz
pilots are in the GBs between translated SG and HG/MG assemblies, respectively.
Cable transmission systems also included regulation pilots with frequencies (shown in Table 13.1) at the top end
of the passband of the cable, where the unwanted variation in path attenuation is at a maximum. The gain of each
repeater along the transmission system is adjusted to maintain a constant level of the regulation pilot signal. In
addition, the frequency comparison pilots (FCPs) shown in Table 13.1 are included at the bottom end of the cable
passband, where phase error is at a minimum. The FCP was employed at frequency translation points to maintain
the frequency stability of the master oscillator from which all carrier frequencies were derived.

13.2.4.4.2 Data
The individual 4 kHz voice channels were extensively used to carry data signals. First, the bit stream of the data
signal is used to modulate a voice-frequency carrier in a modem. The technique of digital modulation is covered
in Chapter 11. Early modems (e.g. Bell 202 standard) achieved bit rates of 1.2 kb/s using frequency shift keying
(FSK). Bit rates up to 56 kb/s were achieved (e.g. in ITU V.90 standard) using amplitude and phase shift keying
(APSK) modulation formats. To achieve higher bit rates, an entire group channel of bandwidth 48 kHz or SG
channel of bandwidth 240 kHz was used to transmit data, which were of course again carried using a suitable
carrier frequency. Data transmission is particularly sensitive to phase distortion and it was necessary to employ
an adaptive equaliser at the receiver (see Chapter 12) to compensate for group delay distortion.

Table 13.1 Regulation and frequency comparison pilots of various cable

transmission systems.
Telegram: @Persian_IEEEComSoc

Cable System
Bandwidth No. of Cable Repeater Ref. Pilot
(MHz) channels size (mm) spacing (kHz) FCP (kHz)

1.3 300 1.2/2.4 8 km 1 364 60

4 960 1.2/2.4 4 km 4 092 60
12 2 700 1.2/2.4 2 km 12 435 308
18 3 600 2.6/9.5 4 km 18 480 564
60 10 800 2.6/9.5 1.5 km 61 160 564
 13.2 Frequency Division Multiplexing 823

13.2.4.4.3 Wideband Audio

The bands of several adjacent voice channels were used for transmitting a single wideband audio or sound pro-
gramme signal. The three standards specified by the ITU included:
● 2 voice channels for 50 Hz → 6.4 kHz audio
● 3 voice channels for 50 → 10 kHz audio
● 6 voice channels for 30 → 15 kHz audio.
Note that the lower-frequency limit of the above audio signals is lower than the value of 300 Hz allowed for voice
signals. This gives rise to a more stringent filter requirement, increasing ℤ by a factor of about 6 and 10 for the
50 Hz and 30 Hz limits, respectively. An entire group channel was also used for the transmission of stereophonic
sound.

13.2.4.4.4 Television
Analogue television signals could also be carried in high-capacity FDM systems. Because of the significant
low-frequency content, which makes frequency translation by SSB impracticable, and the large video signal
bandwidth (∼ 6.0 MHz), a modulation technique known as vestigial sideband (VSB) was employed to place the
television signal in the desired frequency band. VSB is discussed at length in Chapter 7. One television signal (in
the upper band) and up to 1200 voice channels (in the lower band) could be accommodated in a 12 MHz coaxial
cable system. In the 18 and 60 MHz coaxial cable systems, one television signal could be carried in two adjacent
HG or SMG bands. Thus, the 18 MHz system could carry a maximum of two television signals, and the 60 MHz
system could carry six. Alternatively, 1800 voice channels + one television signal were simultaneously carried
in the 18 MHz system. And the 60 MHz system could carry 9000 voice channels + one television signal, or 7200
voice channels + two television signals.

13.2.5 Wavelength Division Multiplexing

FDM is used in some high-capacity optical fibre communication systems, where it is referred to as wave-
length division multiplexing (WDM). The reason for this change of nomenclature is that above the radio band
3 kHz → 3000GHz physicists have traditionally identified electromagnetic radiation by its wavelength, rather
than frequency. And since the electromagnetic radiation that can propagate in a fibre transmission medium
with minimum attenuation lies in the infrared band, we have followed physicists to identify the carrier signal by
its wavelength. WDM then results when multiple carrier signals, each carrying an independent bit stream, are
transmitted simultaneously along one fibre.
Figure 13.13a shows a basic implementation of WDM in which N independent bit streams are multiplexed
onto a single fibre transmission medium. Each bit stream is a TDM signal (see Section 13.3), for example the
OC-48 signal, which carries 32 256 voice channels, or the OC-192 carrying 129 024 voice channels. The bit streams
Telegram: @Persian_IEEEComSoc

are represented as non-return-to-zero (NRZ) voltage waveforms, and each modulates (by on–off keying) the opti-
cal emission of a separate laser source of respective wavelengths 𝜆1 , 𝜆2 , …, 𝜆N . The ITU has defined six optical
transmission bands in the infrared region. See Table 5.1 in Chapter 5, where Figure 5.24 also shows the fibre loss
per kilometre in these bands, which is roughly 0.5 dB/km in the original (O) band and 0.2 dB/km in the con-
ventional (C) band. A separation of 2 nm between the wavelengths of the optical emissions (i.e. carrier signals)
from the laser sources would allow up to N = 50 WDM channels in O band (1260–1360 nm) and N = 17 channels
in C band.
Current systems have three different WDM regimes depending on the value of N. Normal WDM is the most basic
in which N = 2 and the two simultaneously transmitted wavelengths are usually 1310 nm and 1550 nm on a single
optical fibre. Coarse WDM (CWDM) describes systems that have N > 2 and a moderate spacing between wave-
lengths. In 2003, the ITU standardised an 18-channel CWDM involving simultaneously transmitted wavelengths
 824 13 Multiplexing Strategies

Bit stream 1 Laser λ1 λ1 Optical Bit stream 1

source 1 detector 1

Optical demultiplexer
Optical multiplexer
Bit stream 2 Laser λ2 λ2 Optical Bit stream 2
source 2 detector 2
Single fibre

Bit stream N Laser λN λN Optical Bit stream N

source N detector N

Transmitter Receiver
(a)

Diffraction
Lens
λ1 grating
Output
fibres λ2 λN

Input fibre
λ1 + λ2 +…+ λN

(b)

Figure 13.13 (a) WDM system; (b) optical demultiplexer.

from 1271 to 1611 nm at a spacing of 20 nm on a single optical fibre. Finally, dense WDM (DWDM) refers to sys-
tems implemented in the optical C band using a very dense spacing of wavelengths. For example, using speed
of light c = 299 792 458 m/s, a 40-channel DWDM in C band (1530–1565 nm) requires optical carrier spacing of
109.55 GHz, which corresponds to a wavelength spacing of 0.875 nm. And an 80-channel DWDM requires an opti-
cal carrier spacing of 54.77 GHz or 0.4375 nm. Raman amplification enables the usable wavelengths to be extended
into the optical L band (1565–1625 nm), which allows the number of channels in DWDM systems to be increased
even further.
The optical multiplexer is a passive coupler, which may be realised by butting all N laser diodes to a
large-diameter fibre or mixing rod of short length. A single fibre butted to the other end of the rod collects the
composite signal, which is a well-diffused mixture of the emissions from all the diodes. Not surprisingly, there
is a significant insertion loss penalty in this simple multiplexer realisation. Demultiplexing is based on the
spatial dispersion of the mixture of wavelengths by a prism or diffraction grating, as shown in Figure 13.13b. The
incoming optical signal consisting of N wavelengths is focussed by a lens onto a diffraction grating, which returns
Telegram: @Persian_IEEEComSoc

the incident beam back to the lens with the different wavelengths separated at different angles. The lens focuses
the separated optical signals onto different output fibres so that each fibre carries one of the originally multiplexed
signals to an optical detector (e.g. a PIN diode or avalanche photodiode), which extracts the corresponding bit
stream.
A 16-channel WDM of OC-48 TDM signals enables a single fibre to carry 16 × 32 256 or 516 096 voice channels.
Applying the same multiplexing strategy to OC-192 TDM signals allows one fibre to handle a massive 2 064 384
voice channels. Note that this is an example of hybrid multiplexing in which independent signals are packed
into a common transmission medium using more than one multiplexing strategy. In this case, the OC-48 or
OC-192 signal is assembled by TDM of a number of digitised voice signals. These TDM signals are then stacked
in different frequency bands of the fibre transmission medium using frequency (all right, wavelength) division
multiplexing.
 13.3 Time Division Multiplexing 825

13.3 Time Division Multiplexing

13.3.1 General Concepts

We see in Chapter 9 that an analogue signal can be perfectly reconstructed from its samples taken at intervals of T s
(= 1/f s ), provided the sampling frequency f s is at least equal to the bandwidth of the analogue signal. Based on this
(sampling) theorem, all bandlimited analogue signals can be digitised (e.g. converted to a pulse code modulation
(PCM) signal), as is discussed in detail in Chapter 10. The signal is sampled at a suitable frequency f s and each
sample is represented using k bits, referred to as a word. The case k = 8 is called a byte or octet. This yields an
information-bearing bit stream having bit rate

Rc = kf s bits∕second (13.9)

Equation (13.9) gives the bit rate of one signal, which in this context is also referred to as a channel or tributary.
We wish to examine how N such tributaries may be combined into one composite bit stream by TDM and the steps
necessary to ensure accurate recovery of each channel at the receiver.
An analogue TDM system is discussed in Chapter 1 (Section 1.5.3.2) using Figures 1.16 and 1.17, which you
may wish to refer to at this point. Figure 1.17 shows a TDM signal (for N = 3) obtained by interleaving samples
from each of the N tributaries. Here we are dealing with a digital system in which the samples have been digitised
and each is represented with k bits. Thus, for correct reconstruction at the receiver, each of the N channels must
have k bits in time slots of duration T s . This time interval over which one word has been taken from each of the N
channels is known as a frame. There are two types of frame organisation.

● Word-interleaved frame: the frame (of duration T s ) is filled by an interleaver, which visits each of the N channel
ports once during the interval T s , and at each visit takes one word (of k bits) from the storage dedicated to that
channel. These bits are clocked out serially to give the TDM signal. The result is the frame structure shown
in Figure 13.14a, and we see that a frame contains kN message bits. In this diagram, Wordj is the k-bit code
bk-1 …b2 b1 b0 of the sample taken from the jth channel during the interval T s , where bk−1 is the most significant
bit of the word, b0 the least significant bit (lsb), etc.
● Bit-interleaved frame: a bit-interleaved frame is formed by taking one bit at a time from each of the N channel
ports visited by the interleaver in a cyclical order (i.e. 0, 1, 2, …, N − 1, 0, 1, 2, …). The bits are clocked out in
a serial fashion to give the output TDM signal. Since each channel requires a word of k bits to be sent in each
interval of T s , the interleaver must visit each port k times during this interval. The structure of the bit-interleaved
frame is therefore as shown in Figure 13.14b, where b0 (j) is the lsb of the sample from the jth channel, etc.
Note that both types of frames (bit- and word-interleaved) are of the same duration T s and contain the same
number of message bits kN. However, bit-interleaving does not require storage at the tributary ports, as does
word-interleaving, to hold each message word until it is read by the interleaver. We will see that TDM is obtained
Telegram: @Persian_IEEEComSoc

at the first level of the plesiochronous digital hierarchy (PDH) by word-interleaving, whereas bit-interleaving is
used at the higher levels.

Synchronisation between transmitter and receiver is crucial to the correct operation of any digital transmission
system. In Section 12.4.3 we discuss clock extraction, which enables the receiver to achieve bit synchronisation
with the transmitter and hence to use precisely correct decision instants for detecting the incoming bit stream.
However, the packaging of bits from N tributaries into frames introduces a further synchronisation requirement,
known as frame alignment or frame synchronisation. This is needed to give the receiver a precise knowledge
of the start of each frame so that the bits in the TDM signal can be correctly distributed to their respective
channels without the need for additional address information. To this end, the multiplexer inserts at regular
intervals a special pattern of bits known as a frame alignment word (FAW). This serves as a marker with which the
demultiplexer is synchronised at the receiver. Two different arrangements of the framing bits are in common use.
 826 13 Multiplexing Strategies

(a) WordN–1 …. Word2 Word1 Word0

Frame = kN bits; duration = Ts

(b) bk–1(N–1) …. bk–1(1) bk–1(0) …. b1(N–1) …. b1(1) b1(0) b0(N–1) …. b0(1) b0(0)

Frame = kN bits; duration = Ts

Figure 13.14 Frame organisation: (a) word-interleaving; (b) bit-interleaving.

● Grouped or bunched FAW: here the FAW occupies a number of consecutive bit positions in each frame.
● Distributed FAW: a distributed FAW consists of several bits spread over one frame, or one bit per frame spread
over several adjacent frames, called a multiframe or superframe.

Grouped FAW is employed in the European E1 TDM system, whereas the T1 system of North America uses a
distributed FAW. There is a chance that a FAW can occur within the message bits leading to wrong alignment,
and that a transmitted FAW can be corrupted by one or more bit errors. To minimise the problems posed by
these two events, alignment is declared only after a correct FAW is detected at the same relative position within
several (say three) consecutive time intervals. This interval is that over which a complete FAW was inserted at the
transmitter, which could be a frame (for a bunched FAW) or a multiframe (for some distributed FAWs). Secondly,
a loss of alignment is declared (and a free search for the FAW thereby initiated) only after a number of (say four)
incorrect FAWs are received in consecutive intervals. Thirdly, the FAW is chosen to be of an intermediate length.
Too long and it is more readily corrupted by noise; too short and it is more frequently imitated in the message
bits. Furthermore, the FAW must be a sequence of bits that cannot be reproduced when a part of the FAW is
concatenated with adjacent message bits (with or without bit errors), or when several FAWs are bit interleaved.
The control of switching and execution of other network management functions require the transmission of
signalling information in addition to the message and FAW bits discussed above. This is accomplished by inserting
a few auxiliary bits in various ways.

● Bit robbing: a signalling bit periodically replaces the lsb of a message word. This is done in every sixth frame in
the T1 system that employs this technique. The resulting degradation is imperceptible for voice messages but is
clearly totally unacceptable for data (e.g. American Standard Code for Information Interchange (ASCII)-coded)
Telegram: @Persian_IEEEComSoc

messages. For this reason, in a TDM system that uses bit-robbed signalling the lsb of message words in all frames
are left unused when carrying data.
● Out-of-word signalling: within the sampling interval T s , the message word from each channel is accompanied
by one signalling bit, which gives a signalling rate of f s bits/second per channel. Alternatively, a time slot of k
bits in every sampling interval is dedicated as a signalling channel whose bits are assigned in turn to each of the
N channels. The signalling rate in this case is therefore kf s /N bits/second per channel. We will see that the E1
system uses this type of signalling.
● Common signalling: one slot of k bits is dedicated in each time interval T s to signalling, which leads to an overall
signalling rate of kf s bits/second. The entire signalling slot is assigned to one message channel at a time according
to need. Some of the bits are, however, used to provide a label that identifies which channel the signalling
belongs to.
 13.3 Time Division Multiplexing 827

From the foregoing discussion we see that the bit rate of an N-channel TDM signal exceeds N times the bit rate of
each tributary because of the insertion of overhead bits for frame alignment and signalling. There are kN message
bits in each frame of duration T s . Thus, f s (= 1/T s ) is the frame rate. If we denote the total number of framing and
signalling bits in each frame by l (for control bits), it follows that the bit rate of the TDM signal is given by
Nk + l
R=
Ts
= Nkf s + lf s
= NRc + lf s (13.10)

where Rc is the tributary bit rate stated earlier in Eq. (13.9). Considering the fraction of message bits in the TDM
signal, we may define the data transmission efficiency as
Number of message bits
𝜂= × 100%
Total number of bits
NRc
= × 100% (13.11)
R
It is important to note the significance of the parameters on the right-hand side of Eq. (13.11). N is the number
of message channels at the input of the nonhierarchical or flat-level TDM multiplexer, Rc is the message bit rate
emanating from each channel, and R is the output bit rate of the multiplexer. Equation (13.11) can be applied to a
TDM signal obtained after several hierarchical levels of multiplexing, with NRc being the total number of message
bits per second in the TDM signal, which includes bits added ahead of the multiplexer to each of the tributary bit
streams for error control.
We have so far discussed in very general terms what is a flat-level TDM. To allow the building of high-capacity
TDM systems using standardised equipment, a hierarchical multiplexing procedure was adopted.

13.3.2 Plesiochronous Digital Hierarchy

The basic building block of the PDH is the 64 kb/s channel, which results from the digitisation of analogue speech
in the following manner. Analogue speech signal is first filtered to limit its frequency content to a maximum value
of 3400 Hz. It is then sampled at the rate f s = 8 kHz, meaning that the sampling interval is T s = 1/f s = 125 μs,
which constitutes a frame. Each sample is quantised and represented using k = 8 bits. The coding scheme follows
a nonuniform quantisation procedure, which is either A-law (in Europe) or 𝜇-law (in North America). Thus, each
voice signal is converted to a bit stream generated at the rate 8000 samples/second × 8 bits/sample = 64 kb/s. This
is the bit rate of each input channel or tributary at the very first level of the multiplexing hierarchy. This brief
summary of sampling and digitisation is adequate for our treatment here, but you should feel free to consult
Telegram: @Persian_IEEEComSoc

Chapters 9 and 10 for a more detailed discussion.

There are three different procedures for hierarchical multiplexing of these 64-kb/s channels, namely the E1
system in Europe, the T1 system in North America, and the (non-ITU standardised) J1 system in Japan.

13.3.2.1 E1 System
The first level of multiplexing combines 30 digitised speech signals, each of bit rate 64 kb/s, to give the Order-1
TDM signal or simply E1. The equipment used for this purpose is known as a primary muldex – a portmanteau
of multiplexer and demultiplexer – a block diagram of which is shown in Figure 13.15 with emphasis on the
multiplexing operation. Note that the PCM codec (for coder and decoder) is the A-law type. The E1 frame is often
described as CEPT PCM-30, where CEPT refers to Conference of European Posts and Telecommunications, and
30 signifies the number of voice channels.
 828 13 Multiplexing Strategies

30 analogue voice Order-1 TDM

signal inputs signal output

A-law PCM 64 kb/s Multiplex

ʋ1 (t) Codec control bits
64 kb/s
A-law PCM 64 kb/s
ʋ2 (t)
Codec
0
1
2

Byte Interleaver
A-law PCM 64 kb/s
ʋ15 (t) 15
Codec 2048 kb/s
16
A-law PCM 17
ʋ16 (t)
Codec 64 kb/s 18

A-law PCM
ʋ17 (t) 31
Codec 64 kb/s
Slot
Nos.
64 kb/s
A-law PCM Signalling
ʋ30 (t)
Codec 64 kb/s control bits

E1 Primary Muldex

Figure 13.15 E1 first-order TDM.

13.3.2.1.1 E1 Frame Structure

A frame of duration T s = 125 μs is divided into 32 time slots, and 8 bits are placed in each slot. These bits are
clocked out serially to give the E1 signal, which therefore has a bit rate
32 × 8 bits
R1 = = 2048 kb∕s (13.12)
125 × 10−6 seconds
Of the 32 slots or channels C0 to C31 in each frame, 30 carry message bits, which will vary from frame to frame
according to the samples of the respective message signals. Two of the channels (C0 and C16) carry overhead bits
for managing the multiplexing operation and signalling. The efficiency of the E1 signal therefore follows from
Eq. (13.11), with N = 30, Rc = 64 kb/s, and R = 2048 kb/s
30 × 64
𝜂1 = × 100% = 93.75%
2048
Telegram: @Persian_IEEEComSoc

Out-of-word signalling is employed with channel C16 providing the signalling needs of two of the message
channels at a time, 4 bits to each channel. It therefore takes 15 adjacent frames to cover the signalling of the 30
message channels. Dedicating channel C16 in the first frame for marking the beginning of this group of frames,
we have what is known as a multiframe that consists of 16 adjacent frames and is of duration 16 × 125 μs = 2 ms.
The complete content of channels C0 and C16 can be seen over an entire multiframe consisting of frames F1 to
F16, as shown in Figure 13.16. In considering this multiframe, we ignore the contents of channels C1 to C15 and
C17 to C31 in each frame since these are message bits. We note the following:
● Signalling channel C16

In the first frame F1, the first 4 bits of the channel (C16) are used to carry a multiframe alignment word
(MAW) = 0000, which marks the beginning of a multiframe and allows correct numbering of the component
 13.3 Time Division Multiplexing 829

Multiframe One frame Alignment channel (CO) only

= 16 frames = 32 channels

Odd frames Even frames

(F1, F3, …, F15) (F2, F4, …, F16)
F1 C0 IB IB
C1
C2 0 1
F2 C3
C4 0 AB
F3 C5 1 NB
C6
F4 C7 FAW 1 NB
Each frame C8
F5 C9 0 NB
C10 1 NB
F6 C11
C12 1 NB
F7 C13
F8 C14
C15
C16
F9 C17
C18 Signalling channel (C16) only
F10 C19
C20
F11 C21
F12 C22 F1 only Other frames Fj; j = 2, 3, …, 16
C23
C24 0 b3
F13 C25
0 b2 Signalling bits for
C26 MAW
F14 C27 b1 message in channel Cj – 1
0
F15 C28
C29 0 b0
F16 C30 b3
C31 XB
AB b2 Signalling bits for
XB b1 message in channel Cj + 15
XB b0

Figure 13.16 E1 level-1 frame format.

frames. Three bits are unassigned extra bits (XB); and one bit is used as an alarm bit (AB) to signal the loss of
multiframe alignment. AB is binary 0 during normal operation.
In all other frames Fj (for j = 2, 3, …, 16), four bits of the channel carry signalling data (e.g. on or off hook, dialling
digits, call progress, etc.) for channel Cj − 1, and the remaining four bits carry the signalling for channel Cj + 15.
What this means is that channel C16 in frame F2 carries the signalling bits for message-carrying channels C1
and C17. In frame F3, the signalling channel C16 carries signalling bits for channels C2 and C18, and so on until
in the last frame F16 of the multiframe it carries signalling bits for channels C15 and C31. In this way, each one
Telegram: @Persian_IEEEComSoc

of the message channels uses four signalling bits in every multiframe of duration 2 ms, which gives a channel
signalling bit rate of
4 bits
Rsignalling = = 2 kb∕s
2 ms
● Alignment channel C0

The first bit of this channel in all frames is an international bit (IB), which is now used to provide a check
sequence for error detection as follows. A 4-bit cyclic redundancy check (CRC-4) code is computed using all
2048 bits in frames F1 to F8, referred to as the first submultiframe (SMF). This code is conveyed in the first bit
(IB) of frames F9, F11, F13, and F15 (i.e. within the next SMF). Another CRC-4 code is computed on the bits
in the second SMF (i.e. frames F9 to F16), and conveyed in the first bit of frames F1, F3, F5, and F7 in the next
 830 13 Multiplexing Strategies

SMF. At the receiver the same computation is repeated, and the result is compared with the received CRC-4
code. Any discrepancy is an indication of error in one or more bits of the relevant SMF.
In the odd frames F1, F3, …, F15, the last seven bits of channel C0 carry a FAW = 0011011.
In the even frames F2, F4, …, F16, the second bit is always set to binary 1 to avoid a chance imitation of the
FAW within these even frames. The third bit is used as an alarm bit (AB), which is set to 1 to indicate the loss of
frame alignment and is 0 during normal operation. The last five bits are designated as national bits (NB), which
are set to 1 when an international boundary is crossed.

13.3.2.1.2 E1 Hierarchy
A digital multiplexing hierarchy, shown in Figure 13.17, is used to build TDM systems of the required capacity,
which allows a better exploitation of the bandwidth available on the transmission medium. Five levels of multi-
plexing are shown.

● In level 1, referred to as the primary level and discussed in detail above, 30 voice channels are multiplexed by
byte-interleaving in a primary muldex to yield an Order-1 TDM or E1 signal of bit rate 2048 kb/s. This rate is
often identified simply as 2 Mb/s.
● Four Order-1 TDM signals are combined in a muldex, more specifically identified as a 2–8 muldex. The out-
put is an Order-2 TDM or E2 signal of bit rate 8448 kb/s (a rate often referred to as 8 Mb/s), which contains
4 × 30 = 120 voice channels. The multiplexing is by bit-interleaving in this and higher levels of the hierarchy.
From Eq. (13.11), the efficiency 𝜂 2 of this Order-2 TDM signal is given by
120 × 64 kb∕s
𝜂2 = = 90.91%
8448 kb∕s
● Four Order-2 TDM signals are combined in an 8–34 muldex. The output is an Order-3 TDM or E3 signal of bit
rate 34 368 kb/s – often abbreviated to 34 Mb/s. This signal contains 4 × 120 = 480 voice channels and has an
efficiency 𝜂 3 = 89.39%.
● Four Order-3 TDM signals are multiplexed in a 34–140 muldex to give an Order-4 TDM or E4 signal of bit rate
139 264 kb/s – often referred to simply as 140 Mb/s, which contains 4 × 480 = 1920 voice channels, and has an
efficiency 𝜂 4 = 88.24%.
● Finally, four Order-4 TDM signals are multiplexed in a 140–565 muldex to give an Order-5 TDM or E5 signal of
bit rate 564 992 kb/s – referred to simply as 565 Mb/s, which contains 4 × 1920 = 7680 voice channels, and has
an efficiency 𝜂 5 = 87.00%. A further multiplexing level is possible that combines four Order-5 TDM signals to
yield a 2.5 Gb/s TDM signal carrying 30 720 voice channels.

We observe that at each level of the hierarchy the bit rate of the output TDM signal is more than the sum of the
bit rates of the input tributaries. The efficiency of the TDM signals therefore decreases monotonically as we go up
Telegram: @Persian_IEEEComSoc

2048 kb/s
(30 voice channels)
8448 kb/s
(120 voice channels)
ʋ1(t) 34368 kb/s
Primary (480 voice channels)
139264 kb/s
Muldex 2–8 (1920 voice channels)
ʋ30(t)
Muldex 8 – 34 564992 kb/s
(7680 voice channels)
30 analogue Four Order-1 Muldex 34 – 140
voice signals TDM signals
Four Order-2 Muldex 140 – 565
TDM signals Four Order-3 Muldex
TDM signals
Four Order-4 Order-5
TDM signals TDM signal

Figure 13.17 CEPT plesiochronous digital hierarchy.

 13.3 Time Division Multiplexing 831

the hierarchy. The reason for this is the insertion of control bits into the TDM frame produced by each muldex in
the hierarchy. It is worthwhile to examine this further.

13.3.2.1.3 Higher-level Frames Structure

The frame format of the primary muldex has already been discussed in detail. The frame structure at the other
levels of the hierarchy is shown in Figure 13.18. We note the following.
● A bunched FAW is inserted at the beginning of each frame, with
FAW2 = FAW3 = 1111010000
FAW4 = 111110100000
This enables the demultiplexer to recognise the start of each frame and therefore to correctly route the bits to
their tributaries.
● The tributary bits are filled by taking one bit at a time in order from each of the four (input) tributaries. That is,
bit-interleaving is used.
● Ji is a justification bit for the ith tributary, i.e. J1 for tributary 1, J2 for tributary 2, and so on. It is either a dummy
bit or a legitimate bit taken from the tributary. The basis of this decision and the purpose of the J bit are explained
below. The role of the J bits is to allow correct multiplexing of four tributaries whose bit rates, although nominally
equal, may drift slightly apart. In fact, this is why the hierarchy is referred to as plesiochronous, which means
nearly synchronous. Each tributary bit stream is written into a buffer under the control of a clock frequency

848-bit E2 Frame 1536-bit E3 Frame 2928-bit E4 Frame

10 -bit FAW2 10 -bit FAW3 12 -bit FAW4

1 Alarm bit 4 Alarm bits

2 Alarm bits
1 National bit 472 Tributary bits

200 Tributary bits 372 Tributary bits C 1 C 2 C 3 C 4 (4 bits)

484 Tributary bits

C 1 C 2 C 3 C 4 (4 bits) C 1 C 2 C 3 C 4 (4 bits)
C 1 C 2 C 3 C 4 (4 bits)
208 Tributary bits 380 Tributary bits 484 Tributary bits

C 1 C 2 C 3 C 4 (4 bits) C 1 C 2 C 3 C 4 (4 bits)
C 1 C 2 C 3 C 4 (4 bits)
484 Tributary bits
208 Tributary bits 380 Tributary bits
Telegram: @Persian_IEEEComSoc

C 1 C 2 C 3 C 4 (4 bits)

C 1 C 2 C 3 C 4 (4 bits) C 1 C 2 C 3 C 4 (4 bits) 484 Tributary bits

C 1 C 2 C 3 C 4 (4 bits)
J 1 J 2 J 3 J 4 (4 bits) J 1 J 2 J 3 J 4 (4 bits)
J 1 J 2 J 3 J 4 (4 bits)
204 Tributary bits 376 Tributary bits 480 Tributary bits

FAW2 = FAW3 = 1111010000; FAW4 = 111110100000

Figure 13.18 CEPT higher level frame formats. Nominal bit rates are 8448 kb/s, 34368 kb/s, and 139 264 kb/s for E2, E3,
and E4, respectively.
 832 13 Multiplexing Strategies

extracted from the bit stream. The buffer is then read by the interleaver under the control of a common clock
of slightly higher frequency. Occasionally, to prevent buffer i (for tributary i) from emptying, a dummy bit Ji is
given to the interleaver rather than a bit being read from the buffer. This is known as positive justification or bit
stuffing. Thus, Ji will be either a legitimate bit from the ith tributary or a dummy bit that must be discarded at
the demultiplexer. The demultiplexer must therefore have a way of knowing which one is the case.
● Ci is a control bit that is set to 1 to indicate that Ji is a dummy bit. Ci = 0 thus indicates that Ji is a legitimate
bit from the ith tributary. To protect this important control bit from error, it is sent more than once at different
locations within the frame. The demultiplexer decides on the value of Ci based on majority voting. For example,
in the third-order multiplex frame, Ci is taken to be a 1 if up to two of its three repeated transmissions are 1’s.
Note that a wrong decision about Ci and hence about whether Ji is a dummy bit would lead to a very serious
problem of bit slip in subsequent bit intervals of the frame.

13.3.2.2 T1 and J1 Systems

The T1 system (in North America) and J1 system (in Japan) both have identical first-order or primary multiplex-
ing. Twenty-four analogue voice signals are each digitised to 64 kb/s using 𝜇-law PCM and multiplexed by byte
interleaving to yield a TDM signal, which is referred to as DS1. The DS1 frame, of duration T s = 125 μs, contains
24 time slots of 8 bits per slot, plus one extra bit used for framing. Thus, the bit rate R1 and efficiency 𝜂 1 of the DS1
signal are given by
(24 × 8) + 1 bits
R1 = = 1544 kb∕s
125 × 10−6 seconds
24 × 64 kb∕s
𝜂1 = × 100% = 99.48%
1544 kb∕s

13.3.2.2.1 T1 Hierarchy
Figure 13.19a shows the North American PDH, which features four levels of multiplexing that are used to build
systems of the required capacity. The first level of multiplexing generates the DS1 signal referred to above and
discussed further shortly. Subsequent levels of multiplexing are based on bit-interleaving of the input tributaries,
with extra bits inserted for distributed frame alignment, justification and justification control, and other services,
such as alarm. The second level of multiplexing combines four DS1 signals into a 96-channel 6312 kb/s DS2 signal
of efficiency 97.34%. At the third level, seven DS2 signals are multiplexed into a 672-channel 44 736 kb/s DS3 signal
of efficiency 96.14%. There are three options at the fourth level of multiplexing. In one procedure, six DS3 signals
are multiplexed into a 4032-channel 274 176 kb/s DS4 signal of efficiency 94.12%. Another standard involves the
multiplexing of three DS3 signals into a 2016-channel 139 264 kb/s DS4 signal of efficiency 92.65%. Yet another
procedure (not standardised by ITU) combines 12 DS3 signals into an 8064-channel 564 992 kb/s DS4 signal of
efficiency 91.35%. Observe that the last two procedures yield signals of the same bit rates as the Order-4 and Order-5
TDM signals in the CEPT hierarchy, but the DS4 signals have a higher efficiency by about 4.4%.
Telegram: @Persian_IEEEComSoc

13.3.2.2.2 T1 Frame Structure

Let us consider the constitution of the DS1 frame in more detail. Figure 13.19b shows what is referred to as a
superframe, which consists of 12 adjacent DS1 frames numbered F1 to F12. We note that each frame (of duration
125 μs) contains 193 bits, which are assigned as follows:

● The first bit of each frame in the superframe is used to provide a distributed 12-bit FAW = 100011011100.
● The remaining 192 bits are message bits taken 8 bits at a time from 24 input channels numbered C0 to C23.
● Every 6 frames – the 6th and 12th frames of the superframe, the bit interval of the lsb of each channel is used
to send a signalling bit, which we have identified as A-bit for the 6th frame, and B-bit for the 12th frame. The
distortion is imperceptible for voice signals but totally unacceptable for data. Two different approaches may be
 13.3 Time Division Multiplexing 833

(a) 1544 kb/s

(24 voice channels)
6312 kb/s
(96 voice channels)
ʋ1(t) 44736 kb/s
Primary DS1 (672 voice channels)
Muldex 1.5 – 6 DS2
ʋ24(t)
Muldex
274 176 kb/s
24 analogue 4 DS1 6 – 45 DS3 (4032 voice channels)
voice signals signals Muldex
45 – 274
DS4
7 DS2 signals
Muldex

6 DS3 signals
139 264 kb/s
or (2016 voice channels)
DS3 45 – 140 DS4
Muldex
3 DS3 signals

or
564 992 kb/s
DS3
(8064 voice channels)
45 – 565
DS4
Muldex

12 DS3 signals

(b)
F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12
1 framing bit 1 0 0 0 1 1 0 1 1 1 0 0

8 bits CO CO CO CO CO CO CO CO CO CO CO CO
A B

8 bits C1 C1 C1 C1 C1 C1 C1 C1 C1 C1 C1 C1
A B

A B

8 bits C23 C23 C23 C23 C23 C23 C23 C23 C23 C23 C23 C23
A B
Telegram: @Persian_IEEEComSoc

1 signalling bit

Figure 13.19 (a) North American PDH; (b) DS1 (or T1) superframe; (c): T2 (DS2) frame structure; (d) 4760-bit T3 (DS3) frame
structure.

adopted to get around this problem when transmitting data. (i) The 8th bit is not used at all in all channels of
every frame. This restricts each channel to 7 bits per frame at 8000 frames per second, which gives a channel
capacity of only 56 kb/s. Efficiency of the output TDM signal drops significantly from 99.48 to 87.05%. (ii) The
24th channel (C23) is devoted as a common signalling channel, called the D-channel. In this case efficiency
equals 95.34%. This was the technique adopted for the North American primary rate integrated services digital
network (PRI), termed 23B + D service, in which there were 23 bearer channels each of bit rate 64 kb/s, and one
 834 13 Multiplexing Strategies

(c) Figure 13.19 (Continued)

M1 48 tributary bits
C1 48 tributary bits
F0 48 tributary bits 1 st sub-frame
(294 bits)
C1 48 tributary bits
C1 48 tributary bits
F1 J 1 T 2 T 3 T 4 44 Trib. bits
M2 48 tributary bits
C2 48 tributary bits
F0 48 tributary bits 2 nd sub-frame
(294 bits)
C2 48 tributary bits
C2 48 tributary bits
1176 -bit frame

F1 T 1 J 2 T 3 T 4 44 Trib. bits
M3 48 tributary bits
C3 48 tributary bits
F0 48 tributary bits 3 rd sub-frame
(294 bits)
C3 48 tributary bits
C3 48 tributary bits
F1 T 1 T 2 J 3 T 4 44 Trib. bits
A 48 tributary bits
C4 48 tributary bits
F0 48 tributary bits 4 th sub-frame
(294 bits)
C4 48 tributary bits
C4 48 tributary bits
F1 T 1 T 2 T 3 J 4 44 Trib. bits

Tk indicates a bit from tributary k = 1, 2, 3, 4;

Trib. bits refer to bits from all tributaries.

64 kb/s data channel. The corresponding European PRI was 30B + D, providing 30 bearer channels and one
data channel. These integrated services digital network (ISDN) services were carried over wire pairs and offered
bit rates of up to 2.048 Mb/s (in multiples of 64 kb/s). They were termed narrowband to distinguish them from
broadband integrated services digital network (B-ISDN), which was provided over optical fibre and offered data
rates in excess of 45 Mb/s, up to 9.95 Gb/s.
● There exists a different signalling procedure for the T1 system in which 24 frames are grouped into what is
known as an extended superframe (ESF). The first bit of each member-frame, formerly dedicated to framing
Telegram: @Persian_IEEEComSoc

only, is then used to perform various control functions. These 24 bits are assigned as follows. Six bits provide
a distributed FAW = 001001, six bits are used for CRC error checking, and the remaining 12 bits are used to
provide a management channel known as the facilities data link (FDL). However, signalling is still performed
by bit-robbing the lsb of all message channels in every sixth frame.

13.3.2.2.3 DS2 Frame Structure

The structure of the 6312 kb/s DS2 or T2 signal generated at the second level of the North American PDH is shown
in Figure 13.19c. One DS2 frame contains 1176 bits made up as shown in the diagram, including:
● A distributed FAW M1 M2 M3 = 011.
● Alarm bit A which is set to bit 1 during normal operation.
● Repeating pattern F0 F1 = 01 which helps to identify the control bit time slots.
 13.3 Time Division Multiplexing 835

(d)
X 84 tributary bits P 84 tributary bits
F1 84 tributary bits F1 84 tributary bits
C1 84 tributary bits C4 84 tributary bits

Sub-frame 4
Sub-frame 1

F0 84 tributary bits F0 84 tributary bits

C1 84 tributary bits C4 84 tributary bits
F0 84 tributary bits F0 84 tributary bits
C1 84 tributary bits C4 84 tributary bits
77
F 1 J 1 T 2 T 3 T 4 T 5 T 6 T 7 Trib. 77
F 1 T 1 T 2 T 3 J 4 T 5 T 6 T 7 Trib.
bits bits
X 84 tributary bits M0 84 tributary bits
F1 84 tributary bits F1 84 tributary bits
C2 84 tributary bits C5 84 tributary bits
Sub-frame 5
Sub-frame 2

F0 84 tributary bits F0 84 tributary bits

C2 84 tributary bits C5 84 tributary bits
F0 84 tributary bits F0 84 tributary bits
C2 84 tributary bits C5 84 tributary bits
F T J T T T T T 77 F T T T T J T T 77
1 1 2 3 4 5 6 7 Trib. bits 1 1 2 3 4 5 6 7 Trib. bits
P 84 tributary bits M1 84 tributary bits
F1 84 tributary bits F1 84 tributary bits
C3 84 tributary bits C6 84 tributary bits
Sub-frame 6
Sub-frame 3

F0 84 tributary bits F0 84 tributary bits

C3 84 tributary bits C6 84 tributary bits
F0 84 tributary bits F0 84 tributary bits
C3 84 tributary bits C6 84 tributary bits
77
F 1 T 1 T 2 J 3 T 4 T 5 T 6 T 7 Trib. 77
F 1 T 1 T 6 T 3 T 4 T 5 J6 T 7 Trib.
bits bits

M0 84 tributary bits
F1 84 tributary bits
C7 84 tributary bits
Sub-frame 7

F0 84 tributary bits
C7 84 tributary bits
F0 84 tributary bits
C7 84 tributary bits
77
F 1 T 1 T 2 T 3 T 4 T 5 T 6 J7 Trib. bits
Telegram: @Persian_IEEEComSoc

Figure 13.19 (Continued)

● Justification bits J1 , J2 , J3 , J4 , which work as earlier described for the European CEPT PDH.
● Control bits C1 , C2 , C3 , C4 , which work as earlier described for the European CEPT PDH. Ck = 0 indicates that
Jk is a legitimate bit from input tributary k; otherwise, (for Ck = 1) it indicates that Jk is a dummy (stuffing) bit.
Ck is so vital that it is sent three times simply to protect it from error.
● All other bits are tributary bits (as indicated in the diagram) formed by interleaving the four input tributaries.

13.3.2.2.4 DS3 Frame Structure

Figure 13.19d shows the structure of the 44 736 kb/s DS3 or T3 signal generated at the third level of the North
American PDH. One DS3 frame contains 4760 bits and each frame is divided into seven 680-bit subframes as
 836 13 Multiplexing Strategies

1544 kb/s
(24 voice channels) 6312 kb/s
(96 voice channels)
ʋ1 (t)
Primary DS1 32064 kb/s
(480 voice channels)
Muldex 1.5 – 6 DS2 97728 kb/s
ʋ24 (t)
Muldex (1440 voice channels)
6 – 32 J3
24 analogue 4 DS1 Muldex 32 – 98
voice signals signals J4
Muldex
5 DS2 signals Three J3
signals

Figure 13.20 Japanese PDH.

shown. There are seven input tributaries and the C bits (C1 to C7 ) and J bits (J1 to J7 ) work as discussed above for
the DS2 structure. Bits F1 F0 F0 F1 = 1001 serve as a distributed FAW for each subframe. Bits M0 M1 M0 = 010 serve
as a distributed multi-subframe alignment word for the entire DS3 frame. Bit X is permanently set to 1 or may be
used for low-speed signalling. Bit P (sent in two separate locations within the frame) is the modulo-2 sum of the
4704 tributary and J bits in the previous frame. It serves as a parity check for error control.

13.3.2.2.5 J1 Hierarchy
The Japanese PDH is shown in Figure 13.20. The first two levels of multiplexing are identical with the North
American hierarchy. Beyond this, at the third multiplexing, five DS2 signals are combined to give a 480-channel
32 064 kb/s TDM signal of efficiency 95.81%. We call this signal J3. At the fourth level, three J3 signals are multi-
plexed to obtain a 1440-channel 97 728 kb/s J4 signal of efficiency 94.30%.

Worked Example 13.1

We wish to determine the following parameters for an M1–2 muldex that generates a DS2 signal from four DS1
inputs:
(a) Nominal stuffing rate of a DS2 signal.
(b) Maximum stuffing rate of a DS2 signal.
(c) Allowable range of bit rate variation in DS1 signals if they are to be successfully multiplexed into a DS2
signal.

(a) Figure 13.21a shows the nominal condition of the muldex that produces the DS2 signal. Each of the four
(input) tributaries has its nominal DS1 bit rate of 1544 kb/s and the output DS2 signal has its nominal bit
rate of 6312 kb/s. Looking at the DS2 frame structure given in Figure 13.19c, we see that 24 overhead bits
Telegram: @Persian_IEEEComSoc

(counting all nontributary bits except J, i.e. counting the M, A, F, and C bits) are inserted per 1176 bits. In
addition, some stuffing bits are inserted (through some of the J bits) under nominal condition. This info is
shown in Figure 13.21a.
It means that for every 6 312 000 bits flowing out of the muldex (per second) under nominal condition,
1 544 000 × 4 are from the four tributaries, a fraction 24/1176 of the 6 312 000 bits are overhead bits, and the
rest are stuffing bits. The stuffing bit rate, denoted Snom , is therefore determined by equating the number of
all muldex input bits each second with the number of output bits. That is
24
1544000 × 4 + × 6312000 + Snom = 6312000
1176
⇒ Snom = 7184 b∕s = 1796 b∕s∕tributary
where we divide by 4 to obtain the number of stuffing bits per tributary.
 13.3 Time Division Multiplexing 837

1544000 b/s
DS1
1544000 b/s
(a) DS1 1.5 – 6 6312000 b/s
1544000 b/s DS2
DS1 Muldex
1544000 b/s
DS1

Stuffing bits
Overhead bits
(24 per 1176-bit frame)

R 1min
DS1
R 1min
DS1 1.5 – 6 6312000 b/s
(b) R 1min Muldex DS2
DS1
R 1min
DS1

Overhead bits J bits

(24 per 1176-bit frame) (4 per frame)

R 1max
DS1
R 1max
DS1 1.5 – 6 6312000 b/s
(c) R 1max DS2
DS1 Muldex
R 1max
DS1

Overhead bits
(24 per 1176-bit frame)

Figure 13.21 Conditions of muldex that produces DS2 signal: (a) nominal; (b) minimum workable, which corresponds to
maximum stuffing rate; (c) maximum workable, which corresponds to no stuffing bits required
Telegram: @Persian_IEEEComSoc

(b) The maximum stuffing rate occurs when all J bits in the frame are stuff bits. This happens when the input trib-
utaries are at their lowest workable rate R1min so that stuffing bits must be inserted in all J positions to complete
an outgoing DS2 frame. This scenario is illustrated in Figure 13.21b including the fact (from Figure 13.19c)
that there are four J bits per 1176-bit frame. The maximum stuffing rate (per tributary) is therefore
1
Smax = × 6312000 = 5367 b∕s∕tributary
1176
(c) At the highest workable input rate R1max , the tributary bits are coming in so fast that every J position in the
frame must be used to convey a tributary bit. That is, every J bit is a tributary bit and there is no stuffing
bit. This scenario is illustrated in Figure 13.21c. The allowable range of DS1 bit rate variation is from R1min
in Figure 13.21b to R1max in (c). We determine these values by simply equating muldex input with output
 838 13 Multiplexing Strategies

(per second) in each scenario. Thus

28
4 × R1min + × 6312000 = 6312000
1176
⇒ R1min = 1540.429 kb∕s
And
24
4 × R1max + × 6312000 = 6312000
1176
⇒ R1max = 1545.795 kb∕s
That is, the allowable range of DS1 bit rate variation is from 1540.429 kb/s to 1545.795 kb/s.

13.3.2.3 PDH Problems

● It is clear from the above discussions that the plesiochronous digital hierarchies used in North America (T1),
Japan (J1), and the rest of the world (E1) are not compatible. The interconnection of PDH systems originating
from different parts of the world thus requires proprietary conversion equipment, which maps 8-bit time slots
(at the 1.5 and 2 Mb/s levels) from one hierarchy into slots in the other hierarchy, a process known as timeslot
interchange. Note that when converting between E1 and T1 or J1 systems that carry voice traffic then it is also
necessary to re-code each 8-bit word between A-law and 𝜇-law PCM.
● PDH was designed to cater for the basic 64 kb/s telephony rate. It is therefore unsuitable for the transmission of
video traffic, which (even with the use of compression) requires much higher bit rates. Inherent limitations in
PDH make it impossible to realise a well-managed system having the required video rates simply by concatenat-
ing several 64 kb/s channels. Furthermore, multiple stages of multiplexing are needed to realise higher-capacity
systems for data and voice traffic. There is a small problem here in that control bits must be inserted at each
stage to manage the multiplexing process. This leads to lower efficiencies at the higher levels of the hierar-
chy. For example, the T1 system has an efficiency of 99.48% at the first multiplexing stage that produces the
1544 kb/s 24-channel signal, but an efficiency of only 92.65% for the 139 264 kb/s 2016-channel signal generated
at the fourth multiplexing stage.
● Beyond the first level of the hierarchy, PDH involves the multiplexing of plesiochronous tributaries, which
requires occasional insertion of dummy bits into the TDM signal, as earlier explained. This leads to two
problems, one minor and the other major. When the dummy bits are removed at the receiver, gaps are left
in the extracted bit stream, which causes jitter that must be smoothed using a process that involves extra
buffering. More seriously, the presence of these dummy bits and framing bits makes it impossible to drop and
insert a lower rate tributary without completely demultiplexing (i.e. unpacking) the TDM signal down to its
component tributaries at the desired rate. Thus, a multiplexing mountain is required at the drop and insert
point, an example of which is shown in Figure 13.22 for the case where a 2 Mb/s bit stream is extracted from
a 140 Mb/s system. PDH-based networks are therefore expensive, inflexible, and cumbersome to build with
Telegram: @Persian_IEEEComSoc

provision for cross-connect points. Such points are needed in modern networks to allow lower-rate tributaries to
be dropped and inserted at intermediate points, channels to be provided for private networks, and subnetworks
to be interconnected to provide alternative paths through a larger network as a backup against the failure of a
particular link.

13.3.3 Synchronous Digital Hierarchy

The synchronous digital hierarchy (SDH) is a multiplexing technique designed to address the above PDH short-
comings and to operate in synchronism with the digital switches used at network nodes. It allows individual
tributaries (down to 64 kb/s) to be readily accessed, and it very conveniently accommodates the standardised PDH
signals presented above, as well as being well suited for carrying ATM payloads (discussed later). Moreover, SDH
 13.3 Time Division Multiplexing 839

34 Mb/s streams

34 – 140

34 – 140
Muldex

Muldex
140 Mb/s In 140 Mb/s Out

8 Mb/s streams

Muldex

Muldex
8 – 34

8 – 34
2 Mb/s streams

Muldex

Muldex
2–8

2–8
2 Mb/s 2 Mb/s
Extracted Inserted

Muldexes performing demultiplexing

Muldexes performing multiplexing

Figure 13.22 PDH multiplex mountain required to access one of the 2 Mb/s channels within a 140 Mb/s signal.

makes ample provision of channel capacity to meet all the requirements of advanced network management and
maintenance for the foreseeable future.

13.3.3.1 SDH Rates

You will recall that the E1 frame (in Figure 13.16) contains 32 bytes in a 125 μs duration, which gives it a bit rate of
2048 kb/s. The basic SDH frame is called synchronous transport module at level 1 (STM-1) and contains 2430 bytes.
It has the same duration T s = 125 μs, which means that a total of 8000 STM-1 frames are transmitted each second,
i.e. frame rate = 8 kHz. Thus, the basic SDH bit rate is
2430 × 8
R1 = = 155.52 Mb∕s
125 × 10−6
The STM-1 frame can therefore accommodate all the American and European PDH multiplex signals up to the
fourth level of the plesiochronous hierarchy, namely 1.5, 2, 6, 8, 34, 45, and 140 Mb/s.
Higher SDH rates are obtained by multiplexing, through byte-interleaving, a number (N) of STM-1 frames to
give what is referred to as the STM-N frame. The ITU has standardised those rates in which N is a power of 4.
For example, the STM-4 and STM-16 frames have duration T s = 125 μs, contain 2430 × 4 and 2430 × 16 bytes,
respectively, and hence have bit rates
2430 × 4 × 8
R4 = = 622.08 Mb∕s
125 × 10−6
2430 × 16 × 8
Telegram: @Persian_IEEEComSoc

R16 = = 2488.32 Mb∕s

125 × 10−6
Figure 13.23 shows the assembly of higher-capacity transport modules up to STM-64. It can be seen that the
output bit rate at each level is the sum of the bit rates of the input tributaries. This is because of the synchronous
operation of these inputs, which makes bit-stuffing unnecessary. Note that although we have depicted a hierarchi-
cal assembly in Figure 13.23, an STM-N frame, whatever the value of N, can also be obtained by byte-multiplexing
N STM-1 frames in a 1-N SDH muldex.

13.3.3.2 SDH Frame Structure

Figure 13.24 shows the structure of the STM-1 and STM-N frames. It is sufficient that we consider only the STM-1
frame structure in detail, since the structure of the STM-N frame follows straightforwardly from a byte-by-byte
interleaving of N STM-1 frames.
 840 13 Multiplexing Strategies

155.52 Mb/s 622.08 Mb/s 2488.32 Mb/s 9953.28 Mb/s

STM-1
SDH 1 – 4 STM-4
Muldex STM-16
SDH 4 – 16
Muldex STM-64
SDH 16 – 64
Muldex
4 STM-1
signals
4 STM-4
signals 4 STM-16
signals

Figure 13.23 Higher-capacity synchronous transport modules.

We have shown the 2430 bytes of the STM-1 frame arranged in nine rows of 270 bytes each. However, it must be
emphasised that the frame is transmitted serially 1 bit at a time starting from row 1, then row 2, and so on to row
9. The MSB of each byte is transmitted first. One STM-1 frame is sent in an interval of 125 μs, followed by the next
frame in the next 125 μs interval, and so on. Note that there are 2430 (= 270 × 9) cells in our rectangular-matrix
representation of the STM-1 frame. Each cell corresponds to 8 bits transmitted in 125 μs, which represents a 64 kb/s
channel capacity. Similarly, each column represents a channel capacity of 64 × 9 = 576 kb/s. Clearly then, one cell
of the STM-1 frame can carry one PCM voice signal. Three columns can carry one DS1 signal (of bit rate 1544 kb/s),
with some bits to spare. Four columns can carry one E1 signal (of bit rate 2048 kb/s), etc. We will have more to say
on this when considering how the STM-1 frame is assembled.
The STM-1 frame is divided into two parts. The first part is the frame header and consists of a 9-byte pointer
field and a 72-byte section overhead (SOH). The frame header covers the first nine columns of the frame, which
corresponds to a channel capacity of 5.184 Mb/s. It is used for carrying control bits, such as frame alignment,
error monitoring, multiplex and network management, etc. The remaining part of the frame is the payload, which
consists of 261 columns or a channel capacity of 150.336 Mb/s. This area is used for carrying a variety of signals
and is therefore referred to as a virtual container (VC). More specifically, it is identified as VC-4, to distinguish it
from smaller-sized virtual containers, since it is large enough to contain the 140 Mb/s PDH signal at the fourth
level of the plesiochronous hierarchy. In general, the payload area provides virtual containers of various sizes
identified as VC-j, which is large enough to accommodate the PDH signal at the jth level of the plesiochronous
hierarchy, but too small for the signal at the next higher level. At lower levels j < 4, a second digit is appended to
the identification to distinguish between the American (1) and European (2) signals. Thus, VC-11 (pronounced
veecee-one-one) is a virtual container adequate for the American DS1 signal (of bit rate 1544 kb/s). Similarly, VC-12
is for the European E1 signal (of bit rate 2048 kb/s), VC-21 is for the American DS2 signal (of bit rate 6312 kb/s),
Telegram: @Persian_IEEEComSoc

VC-22 is for the European E2 signal (of bit rate 8448 kb/s), etc. VC-1 and VC-2 are described as lower-order virtual
containers, whereas VC-3 and VC-4 are higher order.
VCs include bits for a path overhead (POH), which is added at the point that the tributary signal is incorporated
into the SDH system and is used to manage the transmission of the signal and ensure its integrity. The process of
adding a POH is known as mapping. A VC without its POH is known simply as a container (C), which is therefore
the maximum information payload available to a user in the VC. The entire first column (9 bytes) of a VC-4 is used
for the POH. Thus, a C-4 has 260 × 9 bytes or a capacity of 149.76 Mb/s, which is the maximum information rate
in a VC-4 – more than enough for the 140 Mb/s PDH signal. As a reminder we may write
VC = C + POH (13.13)
 13.3 Time Division Multiplexing 841

STM-1 frame: 270 columns of 9 bytes each

(a)
Frame header Payload
(9 Columns) (261 Columns)
A1 A 1 A1 A2 A2 A2 C 1 Row 1
RSOH

B1 M1 M2 E 1 M3 F1 Row 2
D1 M4 M5 D2 M6 D3 Row 3
AU pointers Row 4
B2 B2 B2 K1 K2 Row 5
D4 D5 D6 Row 6
MSOH

D7 D8 D9 Row 7
D 10 D 11 D 12 Row 8
Z1 Z1 Z1 Z2 Z2 Z2 E 2 Row 9
Col. 1
Col. 2
Col. 3
Col. 4
Col. 5
Col. 6
Col. 7
Col. 8
Col. 9
Col. 10

Col. 270
(b) STM-N SOH STM-N Payload
9 × N Columns 261 × N Columns
Row 1
Row 2
Row 3
Row 4
Row 5
Row 6
Row 7
Row 8
Row 9
STM-N frame: 270 × N columns of 9 bytes each

Figure 13.24 SDH frame structure: (a) STM-1; (b) STM-N, for N = 4, 16, 64, …

When a tributary signal is inserted into the SDH system, we say that it has been incorporated in a container.
This process requires single-bit or asynchronous justification if the tributary and SDH clocks are not locked in
frequency. Justification is discussed in Section 13.3.2 in connection with the multiplexing of nearly synchronous
tributaries in PDH. The capacity of a container is always larger than that required by the tributary signal for which
it is defined. So, as part of the mapping process, the spare byte positions in the container are filled with a defined
filler pattern of stuffing bits to synchronise the tributary signal with the payload capacity. The POH and stuffing
bits are removed at the drop point in the network where the tributary is demultiplexed.
Before leaving this issue, it is worth pointing out that the maximum efficiency of an SDH system can be obtained
Telegram: @Persian_IEEEComSoc

as follows
Max no.of information cells in frame
𝜂max = × 100%
Total no.of cells in frame
260 × 9
= × 100% = 96.30% (13.14)
2430
A VC need not start at the first byte of the frame payload. Typically, it begins in one frame and ends in the next.
The starting point of a VC (i.e. the location of its first byte) in an STM frame is indicated by a pointer, which keeps
track of the phase offset between the two. There is therefore no need for delay-causing re-timing buffers at network
nodes, which may be controlled by slightly different clock rates. A higher-order VC and its pointer constitute an
 842 13 Multiplexing Strategies

administrative unit (AU). The pointer is called an AU pointer. It is 3 bytes long and located in the header part of
the STM frame.
The scenario just described applies to the simple case where no intervening multiplexing is required because the
input tributary is large enough (e.g. the 140 Mb/s signal) to use up the available information capacity of the STM-1
frame. In this case a VC-4 is formed, followed by the addition of a pointer to give an AU-4. Finally, an SOH is
added to the AU-4 to complete the STM-1 frame. A more general case involving the multiplexing (always through
byte-interleaving) of several lower-rate signals to fill the STM-1 frame is considered under frame multiplexing.

13.3.3.2.1 Frame Header

Returning to Figure 13.24a we see that the 81-byte frame header consists of 27 bytes for a regenerator section
overhead (RSOH), 9 bytes for AU pointers, and 45 bytes for a multiplex section overhead (MSOH). The RSOH is
interpreted at regenerators along the transmission path, whereas the MSOH remains intact and is interpreted
only at multiplexing points.
● RSOH: The regenerative SOH consists of the following elements:
⚬ A 48-bit FAW A1A1A1A2A2A2, composed using two bytes A1 and A2 only. This marks the start of the STM-1
frame.
⚬ A label C1 that identifies the order of appearance of this frame in the STM-N frame, N = 4, 16, …
⚬ A byte B1 used for error detection by bit interleaved parity (BIP). In general, a BIP-n code provides error
protection for a specified section of a transmitted bit stream. The jth bit of the code gives an even parity check
on the jth bit in all blocks of n bits in the section. For example, you should verify that the BIP-3 code for
protecting the bit stream
100010110111011000101
is given by
BIP-3 code = 0 0 1.
⚬ Medium-specific bytes M1 to M6 , the use of which depends on the type of transmission medium – radio,
coaxial cable, or optical fibre.
⚬ A 64 kb/s voice channel E1 , for use by system maintenance personnel. This communication channel is usually
referred to as engineer’s orderwire (EOW), often also called engineering order wire.
⚬ A 64 kb/s user channel F1 .
⚬ Data communication channels (DCCs) D1 , D2 , and D3 , providing a 192 kb/s capacity for network management.
● AU pointers: the AU pointer field consists of 9 bytes, three of which indicate the exact location of one VC in the
STM-1 frame. The frame may either carry one VC-4 in which case only 3 AU pointer bytes are used, or it may
carry up to three VC-3’s.
● MSOH: the assignment of the multiplex SOH are as follows:
⚬ Byte B2 is used for error monitoring.
⚬ Bytes K1 and K2 are used for automatic protection switching (APS) in line transmission.
Telegram: @Persian_IEEEComSoc

⚬ Bytes D4 to D12 provide a 576 kb/s data communication channel for network management.
⚬ Bytes Z1 and Z2 are reserved for future functions.
⚬ Byte E2 provides a 64 kb/s EOW.

13.3.3.2.2 Path Overhead (POH)

We indicate earlier that the frame payload may consist of only one virtual container VC-4, or several smaller-sized
VCs. An important part of a VC is its POH, which is used to support and maintain the transport of the VC between
its entry and exit path terminations. The POH of a VC-4 as well as that of a VC-3 is 9 bytes long and occupies the
entire first column of the VC. Figure 13.25 shows the composition of this 9-byte POH. On the other hand, VC-1
 13.3 Time Division Multiplexing 843

VC
J1
B3
C2
G1
F2 Container
H4
Z3
Z4
Z5
POH

J 1 = Path trace. a unique identifier for verifying the VC-n connection

B 3 = BIP-8 code. computed over all bits of the previous container
and used for monitoring error.
C 2 = Signal label. shows the composition of the VC.
G 1 = Path status. indicates received signal status
F 2 = Path user channel. for network operator communication between path equipment
H 4 = Multiframe indicator. gives multiframe indication or cell start for ATM
Z3 , Z 4 , Z 5 = Three bytes reserved for use by national network operator

Figure 13.25 Composition of path overhead (POH) for VC-3 and VC-4.

and VC-2 each has a shorter POH that is only one byte long. The bits b7 b6 b5 b4 b3 b2 b1 b0 of this 1-byte POH are
assigned as follows:
● b7 b6 = BIP-2 for error monitoring.
● b5 = Far end block error (FEBE) to indicate receipt of a BIP error.
● b4 = Unused.
● b3 b2 b1 = Signal label (L1, L2, L3) to indicate type of VC payload.
● b0 = remote alarm to indicate receiving failure.

13.3.3.2.3 Frame Assembly

Telegram: @Persian_IEEEComSoc

The process of constituting an STM-1 frame starting from a C-4, which may carry, for example, the 140 Mb/s
PDH signal, is described earlier and is illustrated in Figure 13.26. The STM-1 frame may also be constituted using
lower-rate tributaries in smaller-sized containers. An illustration of this procedure for C-12, which carries one
2.048 Mb/s E1 signal, is shown in Figure 13.27a. The result of this multiplexing procedure is that the STM-1 frame
has been employed to convey 63 E1 signals, each of which can be easily extracted at a network node using a simple
add/drop muldex (ADM) without having to unpack the entire frame. For this to be possible, the starting point of
every lower-order VC must be indicated by a pointer known as a tributary unit (TU) pointer. The process of adding
a pointer to a VC (whether higher order or lower order) is known as aligning. Other multiplexing possibilities are
shown in the ITU-defined basic SDH multiplexing structure of Figure 13.27b. Note that Figures 13.26 and 13.27a
were obtained by following two different routes in Figure 13.27b. Taking a moment to identify these routes will
 844 13 Multiplexing Strategies

C-4 Figure 13.26 Assembly of STM-1 frame from C-4.

e.g.
E4 + fixed stuff

Add POH

POH
VC-4 E4 + fixed stuff

Add
AU Pointer
POH

AU-4 x E4 + fixed stuff

Add
SOH

RSOH
POH

x E4 + fixed stuff
MSOH
STM-1

help you understand how to interpret Figure 13.27b. The following discussion explains the new SDH terms that
appear in these figures.
A lower-order VC together with its TU pointer constitute what is known as a tributary unit, which occupies a
certain number of columns in the STM-1 payload area. For example, it is shown in Figure 13.28 that TU-11 has 3
columns, TU-12 has 4, and TU-21 and TU-22 both have 12 columns. An assembly of identical-rate TUs, obtained
using byte-interleaving, is known as a tributary unit group (TUG). A TUG-2 consists of one TU-2, or three TU-12’s,
or four TU-11’s. And a TUG-3 consists of one TU-3, or seven TUG-2’s. Similarly, an assembly of identical-rate
administrative units is known as an administrative unit group (AUG). Only two realisations of an AUG have been
defined, namely one AU-4 or three AU-3’s. Finally, adding an SOH to an AUG yields an STM-1 frame, and N of
these frames may be multiplexed to obtain the higher-capacity transport module STM-N.

13.3.3.3 SONET
The synchronous optical network (SONET) transmission standard was developed in 1988 by the T1X1 commit-
tee of the American National Standards Institute (ANSI). It is the forerunner of SDH. Both SONET and SDH are
based on the same principles, the most noticeable differences between them being in terminology and the stan-
Telegram: @Persian_IEEEComSoc

dardised transmission rates. The basic SONET frame is called the synchronous transport system level 1 (STS-1) or
optical carrier level 1 (OC-1). This frame has duration 125 μs and contains 810 bytes, which corresponds to a rate
of 51.84 Mb/s. The structure of the frame may be represented similarly to Figure 13.24 as a rectangular matrix
of 90 columns of 9 bytes each. The first 3 columns (equivalent to 1.728 Mb/s) serve as the header, whereas the
remaining 87 columns (or 50.112 Mb/s) are the payload. Thus the STS-1 frame can contain one DS3 signal (of bit
rate 44.736 Mb/s) or 28 DS1 signals or 28 × 24 = 672 voice channels. Based on the SDH considerations discussed
earlier, individual DS1 signals in the STS-1 frame can be extracted without having to disassemble the entire frame.
Higher-capacity frames, called STS-N or OC-N, are obtained by multiplexing N basic frames. In particular, N = 3
gives the STS-3 frame, which has exactly the same capacity (155.52 Mb/s) as the basic SDH frame, namely STM-1.
The other standardised SONET and SDH rates that are identical include OC-12 and STM-4 with a line rate of
 13.3 Time Division Multiplexing 845

(a) (b)
C-12
C-4 C-3 C-2 C-12 C-11

Add POH
VC-3 VC-2 VC-12 VC-11
VC-12

Add
TU Pointer
TU-3 TU-2 TU-12 TU-11
TU-12
×3
×1
Multiplex (×3) ×4
(Byte interleave) ×1 TUG-2
TUG-2 ×7

Multiplex (×7)
(Byte interleave) TUG-3

VC-3 ×7
×3
Add
AU Pointer VC-4 VC-3

AU-3

Multiplex (×3) AU-4 AU-3

(Byte interleave)

AUG ×1
×3
Add AUG
SOH Pointer processing
×N
Multiplexing
STM-N
STM-1 Aligning

Mapping

Figure 13.27 (a) Assembly of STM-1 frame from C-12; (b) basic SDH multiplexing structure.
Telegram: @Persian_IEEEComSoc

TU-11 TU-12 TU-21 & TU-22

Figure 13.28 Tributary units. Each cell represents 64 kb/s, i.e. 8 bits per frame duration T s = 125 μs.
 846 13 Multiplexing Strategies

622.08 Mb/s, OC-48, and STM-16 with a line rate of 2488.32 Mb/s, and OC-192 and STM-64 with a line rate of
9953.28 Mb/s.
International transmission is based on SDH, with the required conversions performed at North American gate-
ways. SONET has, however, been around longer than SDH and is currently more widely implemented in North
America than SDH is in the rest of the world. As the name suggests, optical fibre is the transmission medium
for which SONET was designed. The PDH-based T1 carrier systems in North America were gradually replaced by
SONET technology. Offices in a metropolitan area can be linked together in an optical fibre ring network that runs
the OC-48 system carrying 48 × 672 = 32 256 channels. An add/drop muldex at each office allows desired channels
to be efficiently extracted and inserted.

13.3.4 ATM
You may have observed that PDH and SDH techniques are optimised for voice transmission, their basic frame
duration of 125 μs being the sampling interval of a voice signal. Nonvoice traffic of diverse bit rates cannot be
simultaneously accommodated in a flexible and efficient manner. ATM is a flexible transmission scheme that
efficiently accomplishes the following:

● Accommodation of multiple users by statistical TDM, as demonstrated in Figure 13.29. Time slot allocation is
not regular as in the (nonstatistical) TDM techniques discussed hitherto. Rather, each user is allocated time slots
(and hence bandwidth) as required by their bit rate. A time slot contains a group of bits known as a cell or packet,
which consists of user data plus identification and control bits called a header. If in any time slot there are no
bits to send then an idle cell is inserted to maintain a constant bit rate (CBR) in the transmission medium.
● Provision of multiple services, such as transmission of voice, text, data, image, video, and high definition televi-
sion, and connections to LAN, including LAN and WAN interconnections.
● Support of multiple transmission speeds or bit rates (ranging from 2 to 622 Mb/s) according to the requirements
of each service.

ATM is more efficient than PDH and SDH because it dynamically and optimally allocates available network
resources (e.g. bandwidth) via cell relay switching. It was the transfer mode or protocol adopted for B-ISDN, which
supported all types of interactive point-to-point and distributive point-to-multipoint communication services.
These include voice and video telephony, videoconferencing, high-speed data connection, email messaging, infor-
mation retrieval, multimedia communication, video-on-demand, pay-per-view TV, digital audio broadcast, digital

1 1 0 1…
Voice
Codec
Telegram: @Persian_IEEEComSoc

Idle cell

010011101011…
Video Statistical
Codec Multiplexer
Transmission medium

1 01 … = Cell header
Data
Source

Figure 13.29 Statistical multiplexing.

 13.3 Time Division Multiplexing 847

5 bytes 48 bytes

Header Payload

Figure 13.30 ATM cell.

TV broadcast, and high-definition television (HDTV). In what follows, we briefly discuss the features, structure,
and network components and interfaces of ATM.
ATM breaks the information bit stream, whatever their origin (voice, video, text, etc.) into small packets of fixed
length. A header is attached to each data packet to enable correct routing of the packets and reassembling of the bit
stream at the desired destination. The fixed-length combination of service (or other) data and header is known as
an ATM cell, which is shown in Figure 13.30. It is 53 bytes long, with a 48-byte payload that carries service data, and
a 5-byte header that carries identification, control, and routing information. The maximum transmission efficiency
of ATM is therefore
48
𝜂ATM = × 100% = 90.57%
53
The size of the cell is a compromise between the conflicting requirements of high transmission efficiency and
low transmission delay and delay variation. To see this, imagine that the header is maintained at 5 bytes and the
cell size is increased to 50 000 bytes. The efficiency would increase to 99.99%, but so would the delay if two sources
A and B attempted to send data simultaneously, and the cell from, say, source A had (inevitably) to wait temporarily
in a buffer for the cell from B to go first. The waiting time is a switching delay given by the cell duration – in this
simple case of waiting for only one cell
Cell Size (in bits)
𝜏d = (13.15)
Line Speed (in bits∕second)
Thus, at a typical line speed of 2 Mb/s and the above cell size, we have 𝜏 d = 200 ms. It is important to see the
implication of this result. A received signal would have to be assembled at a destination from cells some of which
were not buffered at all, and some of which were buffered for 200 ms or even longer in the event of a queue at the
switch. This amounts to a variation in propagation time or cell delay variation of at least 200 ms, which is unac-
ceptable for delay-sensitive traffic such as voice and video. On top of this, there is also another cell-size-dependent
delay known as packetisation delay 𝜏 p . This occurs at the source of real-time signals and is the time it takes to
accumulate enough bits to fill one cell.
Cell Payload Size (in bits)
𝜏p =
Telegram: @Persian_IEEEComSoc

(13.16)
Source Bit Rate (in bits∕second)
Thus, for a voice signal (of source bit rate 64 kb/s) and a 50 000-byte cell (with a 5-byte header as above) then
𝜏 p = 6.25 s. Some samples of the signal would be more than six seconds old before even beginning the journey
from transmitter to receiver, and this is clearly unacceptable in interactive communication.
At the other extreme, if we make the cell size very small, say 6 bytes then the efficiency is only 16.67%, but the
packetisation delay and cell delay variation are also drastically reduced, with 𝜏 p = 125 μs and 𝜏 d = 24 μs, which are
practically imperceptible.

13.3.4.1 ATM Layered Architecture

The functions performed in an ATM system can be organised hierarchically into layers with clearly defined inter-
faces. See Figure 13.31.
 848 13 Multiplexing Strategies

Figure 13.31 ATM layered architecture.

Higher Layers

ATM Adaptation Layer (AAL)

ATM Layer

Physical Layer

13.3.4.1.1 Physical Layer

The physical layer is divided into two sublayers, the physical medium (PM) sublayer, and the transmission conver-
gence (TC) sublayer. The PM sublayer defines (i) the electrical/optical interface; (ii) the line code, i.e. the voltage
waveforms used for representing binary 1’s and 0’s; (iii) the insertion and extraction of bit timing information; and
(iv) the transmission medium, e.g. optical fibre, coaxial cable, or wire pair.
The TC sublayer performs the following cell functions:
● Transmission frame (e.g. STM-1) generation at the transmitter and recovery at destination.
● Transmission frame adaptation: this is the process of adapting the flow of cells to match the capacity and struc-
ture of the transmission frame.
● Generation of header error correction (HEC) at sending node, and verification at destination. Any cell arriving
at a network node with an error in the header that cannot be corrected is discarded. Note that the ATM network
does not perform error monitoring on bits in the payload.
● Cell delineation: this is a process of identifying the cell boundaries in a stream of bits arriving at the destination.
If byte boundaries are known – this will be the case if ATM cells are transported using SDH frames – then
the process is as follows. In the HUNT state, the receiver (using a 5-byte-long window) takes the first 5 bytes,
performs HEC calculation on the first 4 and compares the result to the 5th. If they match then those 5 bytes are
probably the header, and the receiver skips the next 48 (supposedly payload) bytes and repeats the process on
the 5 bytes that follow. If a match is obtained several times (typically 6) in a row then it is concluded that the
cell boundaries have been found, and the receiver enters the SYNCH state. However, if the specified number of
consecutive matches is not obtained, the receiver simply slides the window before starting all over in the HUNT
state. While in the SYNCH state, cell delineation is assumed lost, and the HUNT state is initiated, if the HEC
calculation fails a certain number of times (typically seven) in a row.
● Cell rate decoupling: a constant cell rate is maintained by inserting idle cells (as necessary) for transmission, and
removing them at the destination. Idle cells are indicated by the following standardised bit pattern in the cell
header, listed from the first to the fifth byte:

00000000 00000000 00000000 00000001 01010010

Telegram: @Persian_IEEEComSoc

All (48) bytes of the idle cell payload are filled with the bit pattern 01101010.

13.3.4.1.2 ATM Layer

The ATM layer provides the functionality of a basic ATM network and controls the transport of cells through the
network of ATM switches. Specific functions performed include the following:
● Generation and extraction of cell header, excluding HEC calculation.
● Multiplexing and demultiplexing of cells. Services are allocated bandwidth on demand by assigning to them
only the number of cells they require. This is clearly a more efficient utilisation of transmission system resource
 13.3 Time Division Multiplexing 849

than in nonstatistical TDM (e.g. PDH), which allocates a fixed time slot and hence system bandwidth to each
service.
● Translation of values of virtual path identifier (VPI) and virtual channel identifier (VCI) at switches and
cross-connect nodes. See later.
● Generic flow control (GFC). This controls the rate at which user equipment submits cells to the ATM network.

13.3.4.1.3 ATM Adaptation Layer

The ATM adaptation layer (AAL) defines how the higher layer information bits are bundled into the ATM cell
payload. It is divided into two sublayers, namely the segmentation and re-assembly (SAR) sublayer and the con-
vergence sublayer (CS). The functions of these sublayers depend on the class of service, of which ITU has defined
four. Generally, the CS divides the higher-level information into suitable sizes, whereas the SAR sublayer segments
them into 48-byte chunks, and at the destination re-assembles the payloads into data units for the higher layers.
The ITU-defined classes of service are as follows:
● Class A service has a CBR, is connection-oriented (CO), and requires end-to-end timing relation (TR). Examples
of a class A service include voice and CBR video. The applicable protocol is identified as AAL1. One byte of the
48-byte payload is used as a header, which performs various functions, such as cell loss detection. If a cell is lost
the receiver inserts a substitute cell to maintain the TR between transmitter and receiver. Numbering of the cells
by the SAR sublayer enables the detection of cell loss.
● Class B service, like Class A is both CO and TR, but unlike Class A has a variable bit rate (VBR). A VBR arises in
compressed video and audio where the compression ratio and hence bit rate at any time depends on the detailed
content of the signal segment.
● Class C service, like Class B is both CO and VBR, but unlike Class B does not require timing relation, and so is
described as a timing-relation-not-required (TRN) service. An example of this service is CO data transfer.
● Class D service, like Class C is both VBR and TRN, but unlike any of the other services is connectionless (CL).
An example is CL data transfer.
Two different AAL protocols have been defined for data transfer (Classes C and D).
● In AAL3/4 the first 2bytes of the 48-byte payload are used as a header, which gives a message identifier (MID)
that allows the multiplexing of several packets onto a single virtual channel. The last two bytes provide a trailer
for a CRC to monitor bit errors. Thus, only 44 bytes of the 53-byte ATM cell carry data bits, giving an efficiency
of only 83%.
● AAL5 is for those data transfer services that do not require shared media support and protection against
mis-sequencing, e.g. point-to-point ATM links. A block of bits is formed in the convergence sublayer consisting
of the following: (i) data payload ranging from 0 to 65 536 bytes, (ii) a padding ranging from 0 to 47 bytes to
make the block an integer number of 48-byte segments, (iii) a length field, and (iv) a CRC-32 field for error
Telegram: @Persian_IEEEComSoc

detection. This block is then broken into 48-byte cells in the SAR sublayer and sent sequentially. A payload type
identifier (PTI) bit in the ATM cell header is set to 1 to indicate the last cell. Thus, AAL5 makes (almost) the
entire 48-byte ATM cell payload available for data. This yields an efficiency value approaching the maximum
90.57%.

13.3.4.1.4 Higher Layers

There are three types of higher layer information, usually identified as the user plane, the control plane, and the
management plane. The user plane involves all types of user application information: voice, video, etc. The con-
trol plane deals with control information for setting up or clearing calls, and for providing switched services.
 850 13 Multiplexing Strategies

The management plane provides network management information for monitoring and configuring network ele-
ments, and for communication between network management staff.

13.3.4.2 ATM Network Components

Figure 13.32 shows a network reference model for ATM made up of four types of equipment and three standard
interfaces.

● The customer equipment (CEQ) or B-ISDN terminal equipment (B-TE) communicates across the network, serving
as a source and sink for the video, audio, and data bit streams carried by ATM. These streams are referred to
as virtual channels (VCs). The interface between a CEQ and the network is known as the user network interface
(UNI), and is standardised to allow interoperability of equipment and network from different manufacturers.
● The ATM multiplexer enables a number of VCs from different UNI ports to be carried over a single transmission
line. In ATM parlance we say that the virtual channels have been bundled into a container, called a virtual path
(VP) just as several letters are bundled into a postal sack in the postal system for easier transportation to a depot
or sorting office. Note, however, that a VP is not synonymous with the physical link, and there may be several
VPs on one link, just as there may be several postal sacks in one van.
● The ATM cross-connect routes a VP from an input port to an output port according to a routing table, leaving the
contents of each VP (i.e. their VCs) undisturbed. In this respect a cross-connect is analogous to a postal depot
where sacks may be moved unopened from one van to another.
● An ATM switch is the most complicated equipment of the ATM network, able not only to cross-connect VPs but
also to sort and switch their VC contents. This is like a postal sorting office where some sacks are opened and
the letters are re-sorted into new sacks that contain letters with a narrower range of destinations. Other sacks
may be switched, i.e. loaded onto a designated van, with all their contents intact.
● The network node interface (NNI) is the interface between network nodes or subnetworks, whereas the internet-
work interface (INI) is the interface between two ATM networks. INI includes features for security, control and
administration of connections between networks belonging to different operators.

CEQ
multiplexer

ATM
ATM

CEQ cross-
connect

CEQ

ATM ATM
CEQ
Telegram: @Persian_IEEEComSoc

switch switch

INI
UNI NNI

2nd ATM
network

Figure 13.32 ATM network and interfaces.

 13.3 Time Division Multiplexing 851

13.3.4.3 ATM Cell Header

The structure of the ATM cell header is shown in Figure 13.33. The header consists of 5 bytes or 40 bits in all with
the following designations.

● 28 bits (at NNI) or 24 (at UNI) are VPI and VCI fields used for routing.
● At UNI, the first 4 bits provide a GFC field, which is used to control cell transmission between the CEQ and the
network. The GFC field is only of local significance and is usually set to the uncontrolled access mode with a
value of 0000 where it has no effect on the CEQ. Any other value in this field will correspond to the controlled
access mode, where the rate of transmission from the CEQ is expected to be modified in some (yet to be specified)
manner.
● The PTI field has 3 bits b4 b3 b2 . Bit b4 is set to 0 to indicate that the cell is carrying user information. A mainte-
nance/operation information cell is identified with b4 = 1. Bit b3 is a congestion experience bit, which is set to 1
if the cell passes a point of network congestion, to allow a (yet unspecified) reaction. Bit b2 is carried transpar-
ently by the network and is currently used by AAL5 (as explained earlier) to indicate the last cell in a block of
bits.
● One bit serves as the cell loss priority (CLP) field. When set (i.e. CLP = 1), it indicates that the cell is of lower
priority and should be discarded (if need be) before cells with CLP = 0.
● The HEC field has 8 bits, which are used in one of two modes to provide error protection for the cell header.
This is especially important to prevent an error in the VPI/VCI values causing a cell to be delivered to the wrong
address. In the correction mode, 1-bit errors can be corrected. The detection mode, on the other hand, only
allows errors to be detected. The corrupted cell is then simply discarded. Using the correction mode may be
appropriate in an optical fibre transmission medium where errors are rare and isolated. The detection mode is,
however, preferred in copper transmission media where error bursts are not uncommon. This avoids the risk of
a multiple-bit error being mistaken for a single-bit error and erroneously ‘corrected’. The VPI/VCI values change
at each network node, necessitating a recalculation of the HEC field.

bit
8 7 6 5 4 3 2 1
byte
GFC VPI 1
VPI VCI 2
(a)
VCI 3
VCI PTI CLP 4
HEC 5
Telegram: @Persian_IEEEComSoc

bit
8 7 6 5 4 3 2 1
byte
VPI 1
VPI VCI 2
(b)
VCI 3
VCI PTI CLP 4
HEC 5

Figure 13.33 Structure of ATM cell header at: (a) UNI; (b) NNI.
 852 13 Multiplexing Strategies

13.3.4.4 ATM Features Summary

In concluding our brief discussion of ATM let us summarise some of the most important features of this transmis-
sion technique.
● ATM was the transmission technique adopted for B-ISDN. It did not handle voice signals as well as PDH and
SDH systems and was not as efficient for data transmission as the packet switching protocols (e.g. the X.25 whose
efficiency approached 99.93%), but it was an excellent compromise for handling all types of services.
● Bits to be transmitted are packaged in fixed-length cells of 53 bytes, which include a 5-byte header. The cells are
transported at regular intervals, with idle periods carrying idle (i.e. unassigned) cells.
● Cell sequence integrity is maintained. There is a nonzero cell delay variation and occasional loss of cells, but
cells are delivered to their destinations in the right order.
● ATM provides a connection establishment contract whereby the bit rate and quality of service (QOS) are spec-
ified by the end-user device at call set-up and guaranteed by the network for the duration of the call. QOS
parameters include maximum permissible delay, delay variation, and cell loss ratio.
● Network overload control is implemented. In the event of network congestion, new connections are prevented,
and available capacity is allocated to delay sensitive services, namely voice and video.
● ATM functionality follows a layered architecture with a physical layer that is nearly equivalent to the physical
layer of the open systems interconnection (OSI) model. This is followed by an ATM layer, which implements
a basic ATM network. Finally, there is an ATM adaptation layer (AAL), which interfaces various user, control,
and management information to ATM.

13.3.4.5 ATM Versus IP

You may want to revisit Section 1.5.4.2 to find a detailed context for the analogies that we will employ in this brief
section. If we liken the communication system resource to a community cake, ATM allows many to eat as much as
they need, although it means that some may have to do without when the cake is finished. IP (Internet protocol),
on the other hand, ensures that everyone has a smaller piece of the cake, although it means that the piece may
sometimes be too small to satisfy some appetites. Furthermore, as a servant or messenger, ATM is a slow-start
sprinter who will obediently follow the route the master mapped out in advance and will faithfully report back if
things go wrong. IP, on the other hand, is an instant-start jogger who simply grabs the address from the master and
attempts to make its own way through thick and thin but will simply walk away without saying a word if it fails.
So, which of these two servants would you prefer to charge with the important responsibility of taking your urgent
message from A to B? And which of these two community-cake-sharing models would you prefer to live under?
In the 1990s the technical world hailed ATM as the switching technology we had all been waiting for. With
much hype and fanfare, it was adopted as the transmission technique of B-ISDN, which was at the time seen
as the network of the future. Meanwhile, IP continued to quietly grow with the Internet. Many began to fall in
love with IP’s egalitarianism and flexibility and to concede that IP’s weaknesses, especially its propensity to be
Telegram: @Persian_IEEEComSoc

irresponsible, could be fully mitigated by always pairing it with a smart and diligently responsible supervisor such
as TCP (transmission control protocol). So gradually, many masters who had hitherto shunned IP began to come
around. First it was voice, and we coined the phrase voice over Internet protocol (VoIP) in celebration of the deal.
Then it was TV, and we were stunned and gave it the less celebrated name TVoIP. However, because data had
always been comfortable using IP, we had no need to prefix that service. It was simply just IP. Eventually, and
ultimately to no one’s surprise, everything came around. Yes, it became everything over IP, but with no need for a
fancy name. The competition between ATM and IP had been decisively won by IP. ‘So, what happened to ATM?’
you ask. Well, it has been relegated to being an in-house servant where it can be generous in sharing the cake with
a small number of diners and where it can also be an excellent messenger on an internal route from A to B that
rarely needs a detour.
 13.4 Code Division Multiplexing 853

IP has become the switching technology of our twenty-first-century broadband networks, including both the
Internet and mobile communication networks since 4G. We will, however, resist the urge to delve any further
into IP so that we do not stray too far into networking, which, although extremely exciting, is beyond the scope of
this book.

13.4 Code Division Multiplexing

CDM is based on spread spectrum modulation, a technique that was developed in the 1940s for military communi-
cations. The message signal, of (unspread) bandwidth Bm , is spread in a pseudorandom manner over a bandwidth
Bc > > Bm . The bandwidth ratio
Bc
G= (13.17)
Bm
represents a processing gain, which accounts for an increase in the signal-to-noise ratio at the output of a spread
spectrum receiver. Transmitting a signal by spread spectrum modulation yields important benefits.
● The signal is immune to intentional interference, called jamming. A high-power jamming signal is necessarily
narrowband and will fail to drown the information signal since only a small fraction of the signal energy is
corrupted. More accurately, the process of spread spectrum demodulation at the receiver involves the use of a
pseudorandom code, which de-spreads the wanted signal back into a narrow band Bm . Interestingly, the effect
of this process on the jamming signal is to spread it over a wide band Bc . In this way, the jamming signal energy
is rendered insignificant within the narrow band occupied by the recovered wanted signal.
● By a similar consideration, spread spectrum signals are immune to frequency-selective fading arising from mul-
tipath propagation.
● An unauthorised receiver cannot recover the information signal from the transmitted spread spectrum signal.
Simply put, you must have knowledge of the carrier frequency in order to tune into a transmission. And if, as in
an equivalent view of spread spectrum, the carrier frequency is not fixed but changes pseudorandomly then the
oscillator frequency at the receiver must change exactly in step for demodulation to be possible. Only authorised
receivers will know precisely the pseudorandom sequence of carrier frequencies used at the transmitter.
● Spread spectrum signals have a noise-like appearance to other (unauthorised) receivers. Thus, multiple user
transmissions can simultaneously occupy the same frequency band with guaranteed message privacy, provided
each user’s signal has been spread using a unique pseudorandom code, also referred to as pseudonoise (PN)
sequence. This is CDM, which is finding increased nonmilitary applications in satellite and mobile cellular
communications. Clearly, as the number of users increases a point is reached where the ‘background noise’ at
each receiver becomes excessive leading to unacceptable bit error ratios (BERs).
Telegram: @Persian_IEEEComSoc

13.4.1 Types of Spread Spectrum Modulation

There are various types of spread spectrum (SS) modulation depending on the method employed to spread the
message signal over a wider bandwidth.
● Time-hopping (TH): the message signal is transmitted in bursts during pseudorandomly selected time slots.
Figure 13.34a shows the block diagram of a TH transmitter. Let Rm denote the bit rate of the encoded message
signal, giving a bit interval T m = 1/Rm , and a message bandwidth Bm = Rm . Each time interval T > > T m is
divided into L equal time slots, and one of these slots is selected pseudorandomly (by opening the gate for this
duration) for transmission. To keep up with the message rate, we must take from the buffer an average of Rm
 854 13 Multiplexing Strategies

(a)
Message
bit stream THSS signal
PSK
Coder Buffer Gate
ʋm(t) ʋth(t) modulator ʋthss(t)

RF
PN Code
carrier
generator

(b) Message
THSS signal bit stream
PSK
Gate Buffer Decoder
ʋthss(t) detector

RF
PN code carrier
generator

Figure 13.34 Time-hopping spread spectrum (THSS): (a) transmitter; (b) receiver.

bits per second, or Rm T bits in each interval T, which must all be sent during the one time slot (of duration T/L)
when the gate is open. Thus, the burst bit rate is
Rm T
Rs = = LRm
T∕L
With PSK modulation, the transmission bandwidth is Bc = LRm , which gives processing gain G = L.
A TH receiver is shown in Figure 13.34b. The gate must be opened in precise synchronism with the transmitter,
which requires that (i) the gate is controlled by the same PN code used at the transmitter and (ii) both codes are
in phase. This synchronisation is very stringent and becomes more difficult to achieve as L increases. Note that
the role of the buffer at the receiver is to play out the demodulated bursty bit stream at the uniform rate of the
coded message signal.
● Frequency-hopping (FH): the message signal is conveyed on a carrier, which hops pseudorandomly from one
frequency to another, making Rh hops per second. Figure 13.35a shows a block diagram of a frequency-hopping
spread spectrum (FHSS) transmitter. A coded message bit stream first FSK modulates a carrier signal, which
is then multiplied in a mixer by a digital frequency synthesiser output, and the sum frequency is selected. The
output frequency f o of the synthesiser is controlled by a PN sequence taken k bits at a time. Noting that an
Telegram: @Persian_IEEEComSoc

all-zero combination does not occur in a PN sequence, we see that there are L = 2k − 1 different values over
which f o hops. The FSK modulator generates symbols at a rate Rs – one symbol per bit for binary FSK, or per
log2 M bits for M-ary FSK. If the hop rate Rh is an integer multiple of the symbol rate Rs , several frequency hops
occur during each symbol interval. This type of FHSS is known as fast-frequency hopping. If, however, Rh ≤ Rs ,
then one or more symbols are transmitted on each hop, and we have slow-frequency hopping.
At the receiver (Figure 13.35b), exactly the same pseudorandom sequence of frequencies f o is generated and used
in a mixer to remove the frequency hopping imposed on the FSK signal. It is extremely difficult for frequency
synthesisers to maintain phase coherence between hops, which means that a noncoherent FSK demodulator
must be used at the receiver. The main advantages of FHSS are that synchronisation requirements are less strin-
gent, and larger spread spectrum bandwidths can be more easily achieved to realise higher processing gains
G ≈ 2k − 1.
 13.4 Code Division Multiplexing 855

Message
bit stream FHSS signal
FSK
Coder Mixer
ʋm(t) modulator ʋfsk(t) ʋfhss(t)

(a)
Frequency
RF
synthesiser
carrier

PN code
generator

Message
FHSS signal Noncoherent bit stream
Mixer FSK Decoder
ʋfhss(t) ʋfsk(t) demodulator ʋm(t)

Frequency RF
synthesiser
(b) carrier

PN code
generator

Figure 13.35 Frequency-hopping spread spectrum (FHSS): (a) transmitter; (b) receiver.

● Direct sequence (DS): the coded message signal, of bit duration T m , is multiplied by a PN bit stream of much
shorter bit duration T c , referred to as chip duration. This pseudorandomises the message bit stream and spreads
its (null) bandwidth from Bm = 1/T m to 1/T c , which yields a processing gain

G = Tm ∕Tc (13.18)

This highly spread product signal is then used to modulate a carrier by BPSK, QPSK, or M-ary APSK. Direct
Telegram: @Persian_IEEEComSoc

sequence spread spectrum (DSSS) is the type of spread spectrum modulation employed in CDM-based mobile cel-
lular communication (e.g. the old standard IS-95), and our discussion of CDM will be restricted to this method.
One disadvantage of DSSS (compared to FHSS) is that the processing gain that can be achieved is limited by
current device technology as T m decreases (in high information rate systems), since the required low values of
T c become difficult to implement. Timing requirements in DSSS are also more stringent than in FHSS, but less
than in time-hopping spread spectrum (THSS).
● Hybrid methods: hybrid SS techniques are possible that combine TH, FH, and DS. The most common
hybrid technique is DS/FH, which combines the large processing gain possible in FH with the advantage
of coherent detection in DS. Each frequency hop carries a DS spread spectrum signal and is coherently
detected, but the signals from different hops have to be incoherently combined because of their lack of phase
coherence.
 856 13 Multiplexing Strategies

13.4.2 CDM Transmitter

Figure 13.36a shows the block diagram of a CDM transmitter based on DSSS modulation. The waveforms asso-
ciated with this transmitter are shown in Figure 13.36b. Unit-amplitude bipolar waveforms are assumed for con-
venience. The coded message waveform vm (t) has the indicated bit duration T m , whereas the PN waveform vpn (t)
has a chip duration T c . Note that the waveforms correspond to the case T m = 15T c . More than one user (say N)
can be accommodated, with each assigned a unique PN code vpn1 (t), vpn2 (t), …, vpnN (t), or a unique time shift in a
common PN code vpn (t − 𝜏 1 ), vpn (t − 𝜏 2 ), …, vpn (t − 𝜏 N ).
The PN code generator is in general a linear feedback shift register. Figure 13.37a shows the circuit connection
that produces the PN sequence vpn (t) used in Figure 13.36b. The shift register consists of four flip-flops (FF1 to
FF4), which are controlled by a common clock. Clock pulses occur at intervals of T c , and at each clock pulse the
input state of each flip-flop is shifted to its output. The outputs of FF1 and FF4 are added in an EX-OR gate and
fed back as input to the shift register. This gate performs a modulo-2 addition defined as follows
0 ⊕ 0=0
0 ⊕ 1=1
1 ⊕ 0=1
1 ⊕ 1=0 (13.19)

Message
bit stream DSSS signal
Bipolar ʋm (t) ʋmpn (t) PSK
NRZ coder
× modulator ʋdsss (t)

ʋpn (t)

RF
PN code carrier
generator

(a)

Input bits → 1 0 0 1
+1
ʋm(t) t
Tm
–1 Tc
+1
Telegram: @Persian_IEEEComSoc

ʋpn(t) t
–1
+1
ʋmpn(t) t
–1
+1
ʋdsss(t) t
–1
(b)

Figure 13.36 Direct sequence spread spectrum: (a) transmitter; (b) waveforms.
 13.4 Code Division Multiplexing 857

XOR
(a) Gate

Shift register

1 1 1 0 0 [4, 1] PN
FF1 FF2 FF3 FF4 sequence

Clock
(Tc)

X-OR
(b) X-OR
X-OR

Shift register

FF1 FF2 FF3 FF4 FF5

[5, 4, 3, 2] PN
sequence
Clock
(Tc)

1 1 0 1 0
0 1
1 0
0 0
0 0
1
0
(c) 0 [5, 4, 3, 2] PN sequence 1
1
1 0
1 1
0 1
0 1 0 1 1 1
0

Figure 13.37 Maximum-length PN sequence: (a) generator of the [4, 1] code listed in Table 13.2; (b) [5, 4, 3, 2] code
generator; (c) [5, 4, 3, 2] code.
Telegram: @Persian_IEEEComSoc

Because the feedback taps are located at the outputs of the fourth and first flip-flops, we have what is known
as a [4, 1] code generator. In general, a linear feedback shift register that consists of m flip-flops and has feedback
taps at the outputs of flip-flops m, i, j, … is identified as [m, i, j, …]. The serial PN code generated is of course the
sequence of states of the mth flip-flop. As an example, Figure 13.37b shows the connection of a [5, 4, 3, 2] PN code
generator, which gives the cyclic pseudorandom sequence shown in (c). The following discussion clarifies how
this sequence is obtained.
Let us assume that the [4, 1] PN code generator in Figure 13.37a has the indicated initial register state (FF1, FF2,
FF3, FF4) = (1, 1, 0, 0). This is the state before the first clock pulse occurs at time t = 0. The initial feedback input
is therefore FF1 ⊕ FF4 = 1.
Table 13.2 lists the sequence of flip-flop outputs. After the first clock pulse at t = 0, the initial feedback state is
shifted to become the FF1 output, the initial FF1 output becomes FF2 output, etc. Thus, the register state just after
 858 13 Multiplexing Strategies

Table 13.2 Sequence of flip-flop outputs in [4, 1] PN code generator.

Flip-flop Output
Input to shift FF4
Time (t) register (Feedback) FF1 FF2 FF3 (PN sequence)

<0 1 1 1 0 0
0 1 1 1 1 0
Tc 0 1 1 1 1
2T c 1 0 1 1 1
3T c 0 1 0 1 1
4T c 1 0 1 0 1
5T c 1 1 0 1 0
6T c 0 1 1 0 1
7T c 0 0 1 1 0
8T c 1 0 0 1 1
9T c 0 1 0 0 1
10T c 0 0 1 0 0
11T c 0 0 0 1 0
12T c 1 0 0 0 1
13T c 1 1 0 0 0
14T c 1 1 1 0 0
15T c 1 1 1 1 0

t = 0 is (1, 1, 1, 0), and the feedback state is 1 ⊕ 0 = 1. This gives the entry 1, 1, 1, 1, 0 in row t = 0 of the table. You
may wish to carry on in this way and verify the remaining entries of Table 13.2, and then skip to Question 13.3 for
more practice. Note in Table 13.2 that the register goes through all possible 24 states, except the all-zero state (0, 0,
0, 0), before starting all over again at t = 15T c . In general, a PN sequence (generated by a linear feedback register
of m flip-flops) that has the maximum period

L = (2m − 1) Clock cycles (13.20)

is called a maximum-length sequence, or simply an m-sequence. The all-zero state is forbidden and in fact cannot
be entered except from an all-zero initial state, which would then cause the register to remain permanently in this
Telegram: @Persian_IEEEComSoc

state, and the PN sequence to be a train of 0’s. Note further that the periodic code sequence generated by a linear
feedback shift register is fixed entirely by the number of flip-flops m and the feedback tap locations. The initial
state of the register merely determines the starting point of the cycle.

13.4.3 CDM Receiver

A CDM receiver is shown in Figure 13.38a. There are two stages of processing. First, the spread signal vmpn (t) is
extracted by coherent PSK detection. The PSK detector consists of a product modulator followed by a lowpass
filter of bandwidth equal to that of vmpn (t). Further details of PSK detection are given in Chapter 11 but are not
required for this discussion. The second stage of processing involves a correlation receiver (discussed in Chapter
12). This multiplies vmpn (t) by a locally generated code, which must be identical to and exactly in step with the
 13.4 Code Division Multiplexing 859

Correlation receiver
DSSS
signal ʋmpn(t) Binary 1 if > 0
PSK ʋo(t) Tm Vo(Tm) To decision
× ∫
detector device (DD)
0 Binary 0 if < 0

Local RF
carrier PN code
generator

(a)

ʋmpn(t) = ʋm(t) × ʋpn(t)

ʋo(t) = ʋmpn(t) × ʋpn(t)

ʋod(t) = ʋmpn(t) × ʋpn(t–Tc)

ʋo2(t) = ʋmpn(t) × ʋpn2(t)

(b)

Figure 13.38 Direct sequence spread spectrum: (a) receiver; (b) waveforms of indicated signals. (Note: v mpn (t) corresponds
to message signal segment v m (t) = 101 spread using v pn (t) = Code [6, 1]; and v o (t), v od (t), and v o2 (t) are the results of
de-spreading using locally generated codes, where v pn2 (t) = Code [6, 5, 2, 1].)

code vpn (t) used at the transmitter. The multiplication yields a de-spread signal vo (t), which is then integrated in
Telegram: @Persian_IEEEComSoc

regular intervals of one message bit duration T m . A decision device compares the integration result V o (T m ) in each
interval to a zero threshold. The (message) bit interval is declared to contain a binary 1 if V o (T m ) exceeds zero. A
decision in favour of binary 0 is taken if V o (T m ) is less than zero, and a random guess of 1 or 0 is made if V o (T m )
is exactly equal to zero.
An illustration of the operation of the receiver is given in Figure 13.38b. The waveform vmpn (t) corresponds to a
message bit stream segment vm (t) ≡ 101 that was spread at the transmitter using vpn (t) = Code [6, 1]. Multiplying
vmpn (t) with a perfectly synchronised Code [6, 1] yields vo (t). Clearly, this process has somehow extracted the
original waveform vm (t) from a signal vmpn (t) that is noise-like in appearance. You can see that this is the case by
noting that vpn (t) = ±1, so that

v2pn (t) = 1 (13.21)

 860 13 Multiplexing Strategies

Hence
vo (t) = vmpn (t)vpn (t)
= [vm (t)vpn (t)]vpn (t)
= vm (t)[v2pn (t)]
= vm (t) (13.22)
The importance of synchronisation is illustrated in the waveform vod (t), which is the result of using the right
code [6, 1] but with a misalignment of one chip duration T c . In addition, we illustrate in vo2 (t) the effect of using
a wrong code vpn2 (t) = Code [6, 5, 2, 1]. Proceeding as in Eq. (13.22), we write
vod (t) = vm (t)[vpn (t)vpn (t − Tc )]
vo2 (t) = vm (t)[vpn (t)vpn2 (t)] (13.23)
You can see that in these two cases we have failed to de-spread vmpn (t) since the term in brackets is not a constant
but just another PN code (i.e. a random sequence of ±1). The input to the integrator is therefore a randomised
version of the original signal vm (t), the spreading signal being the term in brackets. It means that vm (t) remains
hidden, and the integrator sees only noise-like signals vod (t) and vo2 (t). By examining these two waveforms you can
see that the decision device will make random guesses of 1 or 0, since the average of these waveforms in intervals
of T m is approximately zero. Note that the process of integration is equivalent to averaging except for a scaling
factor.
Figure 13.39 illustrates the code misalignment problem more clearly. Here the output V o (T m ) of the correlation
receiver is plotted against misalignment 𝜏. With perfect synchronisation between transmitter and receiver codes,
𝜏 = 0 and V o (T m ) = Em for binary 1 and −Em for binary 0, where Em is the energy per message bit. We see that
the noise margin (i.e. difference between the output levels of the correlation receiver for binary 1 and 0 in the
absence of noise) is 2Em . As 𝜏 increases, the noise margin decreases steadily causing increased BER, and reaching
zero – with V o (T m ) = 0 and BER = 0.5 – at
L
𝜏= T (13.24)
L+1 c
Here L = 2m − 1 is the length of the PN code and m is the length of the linear feedback register that generates the
code. For a misalignment of T c or larger we have
{
−Em ∕L, Binary 1
Vo (Tm ) = (13.25)
+Em ∕L, Binary 0
That is, the noise margin is −2Em /L, which is negative and implies that in a noiseless receiver a binary 1 would
always be mistaken for binary 0, and vice versa. However, a practical receiver will always be subject to noise and,
Telegram: @Persian_IEEEComSoc

because L is large, the value V o (T m ) = ±Em /L will be negligible compared to noise. Thus, the input to the decision
device will simply fluctuate randomly about 0 according to the variations of noise. Under this scenario, the output
of the decision device will be a random sequence of bits 1 and 0 so that BER = 50%, which is what you get in the long
run from random guesses in a binary sample space. It is important to note that these comments are only applicable
to a misalignment in the range T c ≤ 𝜏 ≤ (L − 1)T c . Beyond (L − 1)T c , the two codes will begin to approach perfect
alignment at 𝜏 = LT c due to their cyclic sequence.
We have demonstrated above that a message signal can only be recovered from a spread spectrum signal in a
receiver equipped with a synchronised correct PN code. We now demonstrate with the aid of Figure 13.40 that a
receiver will also correctly extract its desired signal from a multitude of spread spectrum signals. We show two
message waveforms vm1 (t) ≡ 101 and vm2 (t) ≡ 001, which have been spread using different PN codes vpn1 (t) and
 13.4 Code Division Multiplexing 861

Vo(Tm)(τ)
+Em

Bi
na
ry
1

+Em/L L
τ= T
L+1 c
τ

–Em/L

0
ary
Bin

–Em
0 Tc 2Tc

Figure 13.39 Output V o (T m ) of correlation receiver as a function of PN code misalignment 𝜏.

ʋm 1 (t) = 1 0 1

t
(a)

ʋm 2 (t) = 0 0 1

ʋcdm (t) = ʋm 1 (t)ʋpn 1 (t) + ʋm 2 (t) ʋpn 2 (t)

(b) t

ʋo1 (t) = ʋcdm (t)ʋpn 1 (t)

Telegram: @Persian_IEEEComSoc

(c) ʋo2 (t) = ʋcdm (t)ʋpn 2 (t)

Tm

Figure 13.40 Two-channel CDM example showing (a) user data waveforms v m1 (t) and v m2 (t), (b) received CDM waveform
v cdm (t), and (c) waveforms v o1 (t) and v o2 (t) after each channel at the receiver multiplies v cdm (t) by its allotted PN code.
 862 13 Multiplexing Strategies

vpn2 (t) assigned to users 1 and 2, respectively. Thus, the signal at the receiver of each user (at the output of the PSK
detector) is a composite signal given by
vcdm (t) = vm1 (t)vpn1 (t) + vm2 (t)vpn2 (t) (13.26)
Receiver 1 multiplies vcdm (t) by its unique code vpn1 (t), whereas receiver 2 multiplies by vpn2 (t) to obtain

v01 (t) = vm1 (t) + vm2 (t)vpn2 (t)vpn1 (t)

v02 (t) = vm2 (t) + vm1 (t)vpn1 (t)vpn2 (t) (13.27)
These waveforms are also plotted in Figure 13.40 from which we see that integrating each one over every message
bit interval will lead to correct decisions regarding the transmitted bit streams. That is, each receiver successfully
extracts only the bit stream intended for it from the mix of bit streams contained in the CDM signal vcdm (t). This
is a remarkable result. In general, the signal at the input of a receiver in a CDM system with N simultaneous
transmissions is given by
vcdm (t) = vm1 (t)vpn1 (t) + vm2 (t)vpn2 (t) + · · · + vmN (t)vpnN (t) (13.28)
The de-spread signal of, say, receiver 1 is therefore
vo1 (t) = vcdm (t)vpn1 (t)
= vm1 (t) + vm2 (t)vpn2 (t)vpn1 (t) + · · · + vmN (t)vpnN (t)vpn1 (t)
= vm1 (t) + Noise (13.29)
Similar results apply to every one of the N receivers. The situation is therefore reduced to the familiar problem
of digital signal detection in noise. As the number of users N goes up, the noise term in Eq. (13.29) becomes larger,
and the probability of error, or BER, increases. So, there is a limit on the number of users in order to guarantee
a specified BER. To quantify this limitation, consider an N-user CDM system in which the same signal power Ps
reaches the receiver from each user. The unwanted accompanying N − 1 CDM signals constitute noise in addition
to the receiver noise power Pn , but typically Ps > > Pn . The carrier-to-noise ratio (C/N)i of the wanted signal at the
receiver input is thus
( ) ( )
Ps 1
(C∕N)i = 10log10 ≃ 10log10 dB
Pn + (N − 1)Ps (N − 1)
De-spreading by the receiver adds a processing gain G given earlier in Eq. (13.17), so that the carrier-to-noise ratio
C/N of the de-spread signal (which is the signal from which a demodulator will attempt to recover the original bit
stream) increases to
C∕N = (C∕N)i + 10log10 (G)
( ) ( )
Telegram: @Persian_IEEEComSoc

G G
= 10log10 ≃ 10log10 dB (13.30)
(N − 1) N
We see therefore that G must be much larger than the number of users N if this C/N is to meet the threshold
needed by the demodulator to achieve a specified BER. We may rearrange the above equation to obtain number
of users N in terms of G and C/N
G
N= (13.31)
10(C∕N)∕10
For example, assuming that the transmission system includes error control coding which allows the BER at demod-
ulator output to be as high as 10−2 (because it is followed by a codec that reduces BER from this high value to an
acceptable 10−7 ) then we can determine minimum C/N needed by a QPSK modem as follows. We read from Figure
11.45 the value of Eb /N o needed to achieve BER 10−2 . This gives Eb /N o = 4.32. We then convert this value to C/N,
 13.4 Code Division Multiplexing 863

assuming an ideal modem (with no implementation loss) and an ideal Nyquist filter (i.e. a raised cosine filter with
roll-off factor 𝛼 = 0) and recalling that QPSK is M-ary PSK with M = 4. Thus
C∕N = 4.32 + 10log10 (log2 M) = 4.32 + 10log10 (2)
= 7.33 dB
So, in this case, Eq. (13.31) gives N = G/5.41. This means that to accommodate 100 users we would need G = 541,
which means that we require a 541-fold increase in bandwidth (from message bandwidth to the bandwidth of the
transmitted CDM signal), and this could be a significant barrier. Alternatively, this means that chip duration T c
must be (1/541)th the message bit duration T m and this may be difficult to achieve with available technology if
message bit rate is high.

13.4.4 Crucial Features of CDM

Let us conclude our discussion of CDM by emphasising those features that are crucial to the smooth operation of
this multiplexing strategy.

13.4.4.1 Synchronisation
The PN code generated at the receiver must be identical to and synchronised with the spreading code used at
the transmitter. There is usually no problem with the two codes being identical – unless of course the receiver is
unauthorised – so we concentrate on the synchronisation requirement. Let the transmitter code be vpn (t) and the
receiver code vpn (t − 𝜏) – with a misalignment 𝜏. It follows from the receiver block diagram that
Tm
Vo (Tm ) = vmpn (t)vpn (t − 𝜏)dt
∫0
Tm
= vm (t)vpn (t)vpn (t − 𝜏)dt
∫0
Em Tm
=± vpn (t)vpn (t − 𝜏)dt
Tm ∫0
= ±Em Rp (𝜏) (13.32)
In the above we have used the fact that vm (t) is a (normalised) constant ±Em /T m in the integration interval
spanning one message bit interval. The positive sign applies to binary 1 and the negative sign to binary 0. Rp (𝜏) is
the autocorrelation function of a periodic signal – in this case vpn (t) – of period T m and is defined by
Tm
1
Rp (𝜏) = vpn (t)vpn (t − 𝜏)dt (13.33)
Tm ∫0
The autocorrelation function of a signal has several interesting properties, which we discuss in Section 3.5.5.
Equation (13.33) has been evaluated for a normalised unit-amplitude maximum-length PN sequence of length L
Telegram: @Persian_IEEEComSoc

and chip duration T c , and is shown in Figure 13.41. By examining this figure, we see the importance of synchro-
nisation. Equation (13.32) states that the output of the correlation receiver is proportional to Rp (𝜏), which from
Figure 13.41 is clearly maximum at 𝜏 = 0 and decreases rapidly to −1/L at 𝜏 = T c . You may wish to look back at
Figure 13.39 and note that it is actually a plot of Eq. (13.32).
In practice, synchronisation is accomplished at the receiver in two stages. First is the acquisition stage, also
known as coarse synchronisation, which is performed at the start of signal reception, or after loss of synchronisa-
tion, by sliding the timing of the locally generated PN code until a peak output is obtained. To do this the PN code
first modulates a carrier, as was done in the transmitter. The resulting signal is then correlated with the incoming
spread spectrum signal, and the code alignment is shifted until maximum correlation is achieved. Next follows
the tracking stage or fine synchronisation in which a phase-locked loop is used to keep the locally generated PN
code in step with the transmitter code.
 864 13 Multiplexing Strategies

Rp(τ)
1

1/L
τ

–LTc –Tc Tc LTc

Figure 13.41 Autocorrelation of a maximum-length PN sequence of length L and chip duration T c .

13.4.4.2 Cross-correlation of PN Codes

Returning to Eq. (13.29), which deals with the detection of a spread spectrum signal in a multi-user environment,
we see that the correlation output of receiver 1 may be written as
Tm
Vo1 (Tm ) = vcdm (t)vpn1 (t)dt
∫0
Tm
= [vm1 (t)vpn1 (t) + vm2 (t)vpn2 (t − 𝜏2 ) + · · · + vmN (t)vpnN (t − 𝜏N )]vpn1 (t)dt
∫0
Em2 Tm E Tm
= ±Em1 ± vpn1 (t)vpn2 (t − 𝜏2 )dt ± · · · ± mN vpn1 (t)vpnN (t − 𝜏N )dt
Tm ∫0 Tm ∫0
= ±Em1 ± Em2 R12 (𝜏2 ) ± · · · ± EmN R1N (𝜏N ) (13.34)
Here, 𝜏 k (for k = 2, 3, …, N) is the misalignment between receiver 1 and the PN code of the kth user transmission.
And R1k (𝜏) is the cross-correlation function of the PN sequences vpn1 (t) and vpnk (t), defined by
Tm
1
R1k (𝜏) = vpn1 (t)vpnk (t − 𝜏)dt (13.35)
Tm ∫0
Equation (13.34) shows that for there to be no interference between users in a CDM system, the cross-correlation
of any two PN codes in the system must be zero. Equivalently, we say that the PN sequences should be mutually
orthogonal. This requirement is impossible to meet in practice, and we can only search for classes of PN codes that
give acceptably small (i.e. good) cross-correlation. Figure 13.42 shows the cross-correlation of maximum-length
PN sequences [5, 2] versus [5, 4, 3, 2], and [7, 1] versus [7, 6, 5, 4, 2, 1]. A class of codes known as Gold sequences
gives better cross-correlation properties. A Gold sequence results from an EX-OR combination of two carefully
Telegram: @Persian_IEEEComSoc

selected m-sequences. In general (and this is apparent from Figure 13.42), the larger the sequence length L, the
smaller the cross-correlation, which leads to reduced mutual interference. However, processing delay (for example
during coarse synchronisation) increases with L.

13.4.4.3 Power Control

By examining Eq. (13.34) we see that our failure to find a class of PN codes with zero cross-correlation leads
to stringent power control requirements in order to minimise mutual interference in a multi-user CDM system.
To be more specific, consider a receiver such as a base station in CDM-based cellular telephony or a satellite
transponder that employs a CDM-based multiple access. This receiver is equipped with N correlators, one for
each of the N user transmissions. Let V oj (T m ) denote the output of correlator j and Emj the energy per message bit
reaching the receiver from the jth user. It is clear that the unwanted contribution to V o1 (T m ) from the other user
 13.4 Code Division Multiplexing 865

ʋpn1(t) = Code [5,2]; ʋpn2(t) = Code [5,4,3,2]

0.2
0.1
0 τ
R12(τ)

–0.1
–0.2

0 5Tc 10Tc 15Tc 20Tc 25Tc 30Tc

ʋpn1(t) = Code [7,1]; ʋpn2(t) = Code [7,6,5,4,2,1]

0.1

0.05
R12(τ)

0 τ

–0.05
–0.1

0 20Tc 40Tc 60Tc 80Tc 100Tc 120Tc

Figure 13.42 Cross-correlation R12 (𝜏) of maximum-length PN sequences.

transmissions will depend on Em2 , Em3 , …, EmN according to Eq. (13.34). Similarly, the unwanted contribution to
V o2 (Tm) depends on Em1 , Em3 , …, EmN . And so on. You can therefore see that the condition for minimum mutual
interference is that

Em1 = Em2 = Em3 = · · · = EmN (13.36)

That is, the transmission from each of the N users must reach the receiver at the same power level. As the radio
link from each user to the receiver is typically of a different length and subject to different propagation conditions,
we must implement some form of power control in order to achieve Eq. (13.36).
One way of implementing power control is by each user terminal monitoring the level of a pilot signal from
the base station and adjusting its transmitted power level accordingly. A low pilot level indicates high path loss
between terminal and base station, perhaps because the two are far apart, and causes the terminal to increase its
transmitted power level. A high pilot level, on the other hand, indicates low path loss, perhaps due to increased
proximity between user terminal and base station and causes the terminal to reduce its transmitted power. This
Telegram: @Persian_IEEEComSoc

technique is known as open loop power control. It assumes identical propagation conditions in the two directions of
transmission between terminal and base station, which will often not be the case for a mobile terminal or if the pilot
signal is at a different frequency from the user transmission. For example, if the monitored pilot signal undergoes
a frequency-selective fade, the user terminal could overestimate the path loss experienced by its transmissions to
the base station, which would cause it to increase its transmitted power excessively. There could also be situations
when the user terminal grossly underestimates the attenuation on its transmissions to the base station because it
is in a deep fade, whereas the pilot signal is not.
Closed loop power control solves the above problem but requires a higher operational overhead. Here, the base
station monitors the transmission from each terminal and regularly issues a command that causes the terminal
to increase or decrease its transmitted power. For example, a one-bit command could be issued every 1.25 ms.
A binary 1 indicates that the power transmitted by the terminal is too high, the terminal responding by decreasing
 866 13 Multiplexing Strategies

its power by 1 dB. A binary 0 indicates that the power reaching the base station from the terminal is too low, and
in response the terminal increases its radiated power by 1 dB. The power transmitted by a base station is also
controlled based on power measurement reports received from user terminals, which indicate the signal strength
reaching each terminal and the number of detected bit errors.

13.4.4.4 Processing Gain

It is illuminating to examine PN codes in the frequency domain in order to make important observations on their
signal processing roles. The autocorrelation function Rp (𝜏) of an m-sequence vpn (t) was sketched in Figure 13.41.
The Fourier transform of Rp (𝜏) gives the power spectral density Sp (f ) of vpn (t), which furnishes complete infor-
mation on the frequency content of the PN sequence. We obtain Sp (f ) readily from the waveform of Rp (𝜏) by
noting that Rp (𝜏) is a centred triangular pulse train of period T = LT c , pulse width 2T c , and hence duty cycle
d = 2T c /LT c = 2/L. The pulse train has amplitude A = 1 + 1/L and has been reduced (shifted vertically down) by a
constant level 1/L. From Chapter 4 (see, for example, Eq. (4.19)), we know that this waveform has the following
Fourier series
Ad 1 ∑∞
1
Rp (𝜏) = − + Ad sinc2 (nd∕2) cos(2𝜋nf o 𝜏); fo =
2 L n=1
T
Substituting the above values, we obtain

2(L + 1) ∑
∞
1 1
Rp (𝜏) = + sinc2 (n∕L) cos(2𝜋nf o 𝜏); fo =
L2 L2 n=1
LT c

This indicates that Rp (𝜏) has DC component of amplitude Ao = 1/L2 and contains harmonics spaced apart by
frequency f o = 1/LT c , with the nth harmonic (of frequency nf o ) having amplitude An given by
2(L + 1)
An = sinc2 (n∕L)
L2
This, being the spectrum of the autocorrelation function Rp (𝜏) of the PN sequence, is the PSD of the sequence and
is shown in Figure 13.43 for L = 15. This spectrum provides valuable information.
A PN sequence vpn (t) contains sinusoidal components of frequencies up to a (null) bandwidth equal to the
reciprocal of the chip duration T c . You may recall from Chapter 7 that the effect of multiplying a baseband signal by
a sinusoid of (carrier) frequency f c is to shift the baseband spectrum to be centred at f c . Furthermore, the frequency
translation may be removed without distortion to the baseband spectrum simply by performing the multiplication
a second time using a carrier of the same frequency and phase. Thus, multiplying a message bit stream vm (t) by a
PN sequence will duplicate (in other words spread) the spectrum of vm (t) at intervals of 1/LT c over a bandwidth
of 1/T c . A little thought will show that the duplicated spectra are diminished in amplitude in proportion to T c ,
and that they overlap each other. The composite spread spectrum is therefore roughly uniform (somewhat like
Telegram: @Persian_IEEEComSoc

that of white noise) and bears little resemblance to the baseband spectrum, which means that the signal has been
successfully hidden. Overlapping of the duplicated spectra is important; otherwise, the situation reduces to the
straightforward process of sampling where the spectra are distinguishable and the ‘spreading’ can be removed by
lowpass filtering. We ensure overlapping by using a spreading sequence that has a dense spectrum – implying a
small value of 1/LT c . For a given chip duration T c , this requires that we make the sequence length L very large.
A second multiplication by vpn (t) at the receiver has the ‘magical’ effect of reconstructing the message signal
spectrum, because the spectra originally translated to f = k/LT c , k = 1, 2, 3, …, L are simply thrown back to f = 0,
and these all add to give the original baseband spectrum. This is provided the ‘carrier frequencies’ k/LT c have the
same phases at transmitter and receiver, which is another way of saying that the two PN codes must be synchro-
nised. You will recall that a time shift of 𝜏 on vpn (t) has the effect of altering the phases of its frequency components
by 2𝜋f𝜏. Yes, the multiplication also creates new spectra at 2 k/LT c , but these are filtered out.
 13.5 Multiple Access 867

2(L + 1) PSD
L2

Square sinc
L = 15
envelope

1
L2 f
0 1 1/Tc 2/Tc
LTc

Figure 13.43 PSD of PN sequence of length L = 15.

Herein lies the spread spectrum processing gain. For interference signals entering the receiver along with the
wanted signal, this is their first multiplication with this code, which therefore spreads them over a bandwidth
1/T c . However, for the wanted signal it is the second multiplication, and this returns its frequency components
back to the (null) bandwidth 1/T m , where T m is the message bit interval. Clearly then, the C/N at the output of
the de-spreader is larger than the C/N at the input by the amount
( )
Tm
Processing Gain = 10log10 , dB
T
[ c ]
Spread (null) bandwidth (Hz)
= 10log10 , dB (13.37)
Message bit rate (b∕s)
For example, if a spread bandwidth of 1.23 MHz is employed to transmit at a message bit rate of 9.6 kb/s, Eq. (13.37)
gives a processing gain of 21.1 dB. For a given message bit rate, processing gain can be increased to realise improved
performance in the presence of noise and interference by using a larger spread bandwidth 1/T c . But allocated radio
bandwidths are limited, and device technology places a limit on how small we can make the chip duration T c .
Telegram: @Persian_IEEEComSoc

13.5 Multiple Access

Each of the main multiplexing strategies discussed in this chapter may be utilised as a multiple access technique.
It is important to note the distinction between multiplexing and multiple access. Multiplexing is the process of
combining multiple user signals for simultaneous transmission as a composite signal on one transmission link,
whereas multiple access is concerned with how one communication resource such as a satellite transponder or a
terrestrial base station is shared by various transmitting stations, each operating on a separate transmission link.
Although the concept is more generally applicable, we will limit our brief discussion to a satellite communication
application of multiple access. There are three main types of multiple access, namely frequency division multiple
access (FDMA), time division multiple access (TDMA), and code division multiple access (CDMA).
 868 13 Multiplexing Strategies

13.5.1 FDMA
In FDMA, each transmitting earth station (ES) is allocated a separate frequency band in the satellite transponder.
Figure 13.44 illustrates the sharing of a 36 MHz transponder among three earth stations (ESs). Each ES is allocated
an exclusive 10 MHz bandwidth, which is used for simultaneous and continuous transmission as required. Note
that a composite FDM signal exists on the downlink, whereas the uplink has separate links, each operating on
a separate carrier frequency. To allow the use of realisable filters when extracting a desired channel from the
downlink FDM signal, FDMA always includes a guard band (GB) between allocated adjacent sub-bands. In the
illustration, a GB of 2 MHz is used. The signal transmitted by each ES may contain a single user signal or it may
be a multiplex of several user signals. The former is known as single channel per carrier (SCPC) and the latter as
multiple channel per carrier (MCPC).
FDMA is very simple and cheap to implement using well-established filter technology, but it is prone to inter-
modulation distortion when the transponder amplifier is operated in its nonlinear region. To reduce this distortion,
a lineariser is often used to pre-distort the incoming signal at the transponder in such a way as to make up for the
subsequent amplification distortion. Additionally, input power is reduced from the level that would saturate the
amplifier – an action known as back-off – by the amount (in dB) necessary to ensure operation in a linear region
of the transfer characteristic of the combined lineariser/amplifier system.
FDMA capacity may be readily determined in terms of the number of ESs N that can share a transponder band-
width Bxp with each station needing bandwidth Bes and a GB Bg maintained between allocated sub-bands
⌊ ⌋
Bxp
N= (13.38)
Bes + Bg
The bandwidth requirement Bes of each transmission depends on the required bit rate and the modulation
scheme employed. These relationships are summarised in Section 11.11. Power issues must be considered when

ES 1 ES 2 ES 3
(10 MHz) (10 MHz) (10 MHz)

GB = 2 MHz GB
36 MHz Transponder Bandwidth
Telegram: @Persian_IEEEComSoc

Uplink Downlink
f

Simultaneous
transmission
by each earth f
station ES1 ES2 ES3

Figure 13.44 FDMA concept.

 13.5 Multiple Access 869

assessing capacity because, while there may be enough bandwidth to accommodate N stations, there may be insuf-
ficient power to support that number at the required C/N. When considering power in FDMA, it must always be
remembered that a receiving ES shares the advertised transponder power (i.e. satellite transmit effective isotrop-
ically radiated power (EIRP)) only in proportion to the ES’s allocated bandwidth (as a fraction of transponder
bandwidth).

13.5.2 TDMA
The concept of TDMA is illustrated in Figure 13.45. ESs take turns in making use of the entire transponder
bandwidth by transmitting bursts of RF signals within centrally allocated and nonoverlapping time slots. If N
transmitting ESs share the transponder then over an interval known as a frame duration T f , each station is allo-
cated one time slot for its burst transmission. Therefore, on the uplink, there are separate transmission links
operating at the same frequency but at different time intervals, and on the downlink there is the TDMA frame
carrying the signal of each of the N stations in separate time slots. A frame duration may range from 125 μs up to
20 ms, but a value of 2 ms is common. A large frame duration yields high frame efficiency but increases delay and
complexity. It is always necessary to include a guard time between each time slot to ensure that bursts from sta-
tions using adjacent time slots will not overlap in their arrival time at the satellite even if there is a slight variation
in the arrival of each burst. Such small variations in arrival time may be caused by timing error at the ES, changes
in the distance between an ES and the satellite, and changes in propagation condition.
TDMA is not affected by intermodulation distortion since only one carrier is present in the transponder at any
given time. This is an advantage because it allows amplifiers to be operated at their maximum capacity, which
saves on bulk and weight (since one no longer must use a larger amplifier at a reduced output due to back-off).
However, each ES must transmit at a high symbol rate (to fill the transponder bandwidth) and must do so using

Satellite

Uplink Downlink
Tr iffer

e
tim
an en
d
sm t e
iss art
ion h s
bu tati
rst on
Telegram: @Persian_IEEEComSoc

sf s
rom

TDMA Frame (exists only on satellite downlink):

Guard time
Guard time

Guard time
Preamble

Preamble

Preamble

Preamble
(ES N)
(ES 1)

(ES 2)

(ES 1)

Info bits Info bits Info bits Info bits

(ES 1) (ES 2) (ES N) (ES 1)

Frame duration, Tf

Figure 13.45 TDMA concept.

 870 13 Multiplexing Strategies

enough signal power to provide an acceptable C/N, taking into consideration the proportionate increase in noise
power with bandwidth as N o B. For this reason, TDMA is not as suitable as FDMA for transmitting narrowband
signals from small ESs.
TDMA capacity, in terms of the number of stations N that can share a transponder of bandwidth Bxp using
a TDMA frame of duration T f and guard time T g between time slots and number of preamble bits nbp in each
station’s transmit burst, may be determined in several ways. One way is to start by specifying the required bit rate
Res of each station, the M-ary modulation scheme (APSK or PSK) and the raised cosine filter roll-off factor 𝛼. This
sets the burst bit rate as
Bxp
Rburst = log M (13.39)
(1 + 𝛼) 2
The duration of the time slot T es needed by each ES within the TDMA frame of duration T f is
Res Tf
Tes = (13.40)
Rburst
We must also allow a total guard time NT g within the frame as well as total time NT p for all stations to send their
preamble bits, where
npb
Tp = (13.41)
Rburst
Since NT es + NT g + NT p = T f , it follows that the TDMA system capacity is
⌊ ⌋
Tf
N= (13.42)
Tg + Tes + Tp
where T f and T g are in the system specification, and T es and T p are calculated using the preceding equations.

Worked Example 13.2

A TDMA system has the following specifications:

● Satellite transponder bandwidth, Bxp = 72 MHz.
● TDMA frame duration, T f = 2 ms.
● Guard time between time slots, T g = 1 μs.
● Number of preamble bits in each ES burst, nbp = 148 bits.
Determine the number of stations that can be accommodated if each station operates at a bit rate
Res = 20 Mb/s (which includes redundancy due to error control coding) using 16-APSK and a raised cosine
filter of roll-off 𝛼 = 0.05.
Determine also the TDMA frame efficiency.
Telegram: @Persian_IEEEComSoc

The supported burst bit rate is

Bxp 72
Rburst = log M = × 4 = 274.286 Mb∕s
(1 + 𝛼) 2 1.05
Required time slot per station per frame is
Res Tf 20 × 2
Tes = = = 0.1458 ms
Rburst 274.286
Required time for preamble per station per frame is
npb 148
Tp = = = 0.5396 𝜇s
Rburst 274.286
 13.5 Multiple Access 871

The number of stations is therefore

⌊ ⌋ ⌊ ⌋
Tf 2
N= =
Tg + Tes + Tp 1 × 10−3 + 0.1458 + 0.5396 × 10−3
= ⌊13.57⌋
= 13
Frame efficiency is the fraction of the frame duration that is devoted to carrying bits from the N stations. Since,
in this case, in every 2 ms we have 13 stations using 0.1458 ms each to transmit their data, frame efficiency is
NT es 13 × 0.1458
𝜂= × 100% = × 100%
Tf 2
= 94.79%
This TDMA efficiency is quite high. In practical systems, the guard time will be higher than 1 μs and the number
of preamble bits will be larger. This will reduce frame efficiency and hence system capacity.

13.5.3 CDMA
In CDMA, each transmitting ES multiplies its signal by a unique orthogonal spreading code prior to transmis-
sion. These signals reach the satellite transponder at the same time and in the same frequency band. Following
frequency down conversion and power amplification, a composite CDM signal is transmitted on the satellite
downlink, containing N differently spread signals from the sharing ESs. The theoretical details of the CDM signal
generation and detection are as discussed under CDM. Figure 13.46 shows a CDMA system consisting of N trans-
mit ESs. Details of the receiver operations on the incoming downlink signal from the satellite in order to recover

Uplink Downlink

Transmit earth station 1

m1(t) Filtering, Receive earth station 1
Amp., etc N
Σ mk(t)pk (t)cos(2πfct)
p1(t) cos(2πfct) k=1

PN code LNA,
generator Mixer, etc
Telegram: @Persian_IEEEComSoc

p1(t)

Transmit earth station N m1(t)cos(2πfct) + nI(t)

mN(t) Filtering
2cos(2πfct)
& Amp.
m1(t) + m1 (t)cos(4π fct)
pN(t) cos(2πfct) + 2n1 (t)cos(2πfct)
Tm
∫
PN code 0 m1(t) + n′I(t)
generator
m1(t)

Figure 13.46 CDMA system operation.

 872 13 Multiplexing Strategies

the wanted message signal from any of the N stations are also given in the diagram, using Earth Station 1 as an
example.
CDMA has several benefits, such as helping to reduce interference from co-channel systems, since the unwanted
carrier signals will be spread and therefore mostly rejected by the receiver. It also serves to reduce multipath effect
since the reflected waves will be spread by the receiver if their delay exceeds a chip duration. Also, unlike TDMA,
coordination between ESs sharing one transponder is not required, although synchronisation of the spreading
sequences at transmitter and receiver is essential, as discussed in Section 13.4. Theoretically, CDMA facilitates
100% frequency reuse between beams in a multiple spot beam satellite system. However, CDMA’s main drawback
is its low efficiency, expressed in terms of the total achievable data rate of a CDMA system utilising an entire
transponder when compared to the data rate of a single carrier that fully occupies the same transponder. We did
earlier discuss the constraint on CDMA capacity and derived Eq. (13.31) for number of users in terms of processing
gain and C/N. CDMA requires a large contiguous bandwidth in order to achieve the high processing gain required,
and this may not be available on a satellite transponder. Furthermore, strict power control must be maintained,
as also earlier discussed.

13.5.4 Hybrid Schemes

The multiple access techniques discussed above are often used in combination. For example, a satellite transpon-
der bandwidth may be divided into sub-bands and each sub-band is shared by multiple small ESs using TDMA.
This combination of FDMA and TDMA is known as multifrequency TDMA (MF-TDMA). An example is shown in
Figure 13.47, where 12 stations share one transponder which has been partitioned into three sub-bands denoted
f 1 , f 2 , and f 3 . Four stations share each sub-band using a TDMA frame with four allocated time slots. Note therefore
that the system resource allocated in FDMA is a frequency band, and in TDMA it is a time slot, but in MF-TDMA

f1 f2 f3 Frequency
TDMA frame 3 TDMA frame 2 TDMA frame 1
Telegram: @Persian_IEEEComSoc

Time
Transponder Bandwidth

Figure 13.47 Example of 12 stations sharing one transponder using MF-TDMA.

 13.6 Summary 873

Inbound narrowband SCPC channels Outbound wideband TDM channel

Transponder Bandwidth
TDMA slots →

Inbound MF-TDMA channels Outbound wideband TDM channel

Sub-band 1 Sub-band 2
Transponder Bandwidth

Inbound CDMA channels Outbound wideband TDM channel

Transponder Bandwidth

Figure 13.48 Hybrid multiple access schemes for a star VSAT network.

the allocation is both a time slot and a frequency band. This allocation is usually dynamic and according to need
and may change in a time interval of a few frames.
Figure 13.48 shows three hybrid multiple access arrangements for a VSAT (very small aperture terminal) star
network implementation. In all three scenarios, half of the transponder bandwidth is used to carry a wideband
TDM signal on the outbound link (from the hub station to the VSAT terminals) and the other half is used for the
inbound link supporting transmissions from multiple VSAT terminals to the hub. These VSATs share their half of
the transponder bandwidth using three different schemes: SCPC FDMA is used in the top scenario, MF-TDMA is
used in the middle, and CDMA is used in the bottom. Note therefore that the bottom scenario is a combination of
CDMA and FDMA, the top scenario is pure FDMA but with unequal sub-band partitioning, whereas the middle
scenario is a combination of MF-TDMA and FDMA.

13.6 Summary

This now completes our study of multiplexing strategies. We started by giving several compelling reasons for mul-
Telegram: @Persian_IEEEComSoc

tiplexing in modern telecommunications and then presented a nonmathematical discussion of the four strategies
of SDM, FDM, TDM, and CDM. This was followed by a more detailed discussion of the last three techniques.
FDM is truly indispensable to radio communications and has a very long list of applications. It allows the exis-
tence of several audio and TV broadcast houses in one locality, and the simultaneous provision of a large variety of
communication services. Capacity enhancement in cellular telephony and satellite communication relies heavily
on FDM. Closed media applications of FDM include wavelength division multiplexing (WDM) in optical fibre,
which allows literally millions of toll-quality digital voice signals to be transmitted in one fibre. FDM telephony
allowing up to 10 800 analogue voice signals to be transmitted in one coaxial cable was popular up till the 1980s.
FDM implementation is very straightforward. A frequency band is allocated to a user, and the user’s signal is
applied to modulate a suitable carrier thereby translating the signal into its allocated band. The type of modula-
tion technique depends on the communication system. SSB was used in FDM telephony, on–off keying (OOK) is
 874 13 Multiplexing Strategies

used in optical fibre, FM was used in first generation cellular telephony and analogue satellite communication, and
M-ary APSK is used in modern satellite and terrestrial communications, etc. Each of these modulation techniques
is covered in previous chapters. Our discussion of FDM included its application in telephony where a complete
set of standards was specified for hierarchical implementation.
TDM fits in very well with the techniques of digital switching in modern networks and is ideally suited for trans-
mitting digital signals or bit streams from multiple sources. It allows the advantages of digital communications to
be extended to a larger number of simultaneous users in a common transmission medium than would be possible
with FDM. We presented a detailed discussion of (nonstatistical) TDM optimised for digital transmission of voice,
including the plesiochronous and synchronous digital hierarchies. To show how the requirements of broadband
integrated services are satisfied, we presented the statistical TDM technique of ATM, which satisfactorily multi-
plexes all types of digital signals, including voice, data, and video. However, we noted by way of analogies that IP
has beaten ATM to become the preferred transmission technology of the twenty-first century.
We also discussed various spread spectrum modulation techniques, including time-hopping, frequency-hopping,
and direct sequence. In studying CDM, we demonstrated the importance of several factors, including code
synchronisation and cross-correlation, power control, processing gain, and the length of the code sequence.
We concluded the chapter with a brief discussion of the application of the above multiplexing strategies for multi-
ple access in satellite communication systems. The suitability of a multiple access technique for a given application
or its superiority to other techniques is still open to debate. We briefly presented the merits and drawbacks of each
technique and provided formulas for system capacity calculations. You should therefore now be better equipped
to make an informed choice.

Questions

13.1 Higher line utilisation may be realised by multiplexing 16 voice signals into one 48 kHz group signal. The
following procedure has been specified by the ITU for doing this. Each voice signal is limited to frequen-
cies of 0.25–3.05 kHz. Frequency translation of each voice signal to an exclusive passband is accomplished
for the odd-numbered channels using eight LSB-modulated carriers at frequencies (kHz) 63.15, 69.15,
75.15, … The even-numbered channels are translated using eight USB modulated carriers at frequencies
(kHz) 62.85, 68.85, 74.85, …
(a) Draw detailed block diagrams of the multiplexer and demultiplexer for this 16-channel FDM system.
(b) Sketch a clearly labelled spectrum (similar to Figure 13.7a) of the 16-channel group signal.
(c) Determine the nominal bandwidth and GB of each voice channel.
(d) Determine the Q and ℤ factors of the most stringent filter used in this FDM system.
(e) Compare your result in (d) to that of a standard 12-channel group signal.
Telegram: @Persian_IEEEComSoc

13.2 (. a) Sketch a clearly labelled spectrum of the HG signal in the UK FDM system.
(b) Determine the Q factor of the most stringent filter in the STE of Figure 13.9.

13.3 Determine the Q and ℤ factors of the most stringent filter in a 10 800-channel UK hierarchical FDM
system. How does this compare with a flat-level assembly of the same number of voice channels using
subcarriers of 4 kHz spacing starting at 1 MHz?

13.4 Determine the number of CTE, GTE, and STE required to set up a 600-channel Bell FDM system. Draw a
block diagram showing the connection of these devices from the second multiplexing stage.
 Questions 875

13.5 Frequency gaps or GBs are necessary in FDM signals to allow the use of realisable filters and the trans-
mission of control tones in the gaps. The efficiency of an FDM signal is the percentage of total bandwidth
that contains voice frequencies. Determine the efficiency of a 10 800-channel FDM signal in each of the
three hierarchical standards, namely UK, Europe, and Bell, discussed in the chapter.

13.6 Determine the number of signals multiplexed and the GBs involved when a group signal (60–108 kHz) is
built, according to ITU standards, exclusively from each of the following types of wideband audio signals:
(a) 50–6400 Hz
(b) 50–10 000 Hz
(c) 30–15 000 Hz.
If each baseband audio signal is translated by LSB modulation so that half of the GB is on either side of
the translated spectrum, determine the set of carrier frequencies required in (a)–(c).

13.7 Examine the frame structures of the E1 and T1 signals shown in Figures 13.16 and 13.19b and calculate
the rate of each of the following types of bits for the indicated frame:
(a) Framing bits in E1 and T1.
(b) Signalling bits in E1 and T1.
(c) Signalling bits per channel in E1 and T1.
(d) CRC-4 error checking using IB bit in E1.
(e) Message bits in T1.

13.8 Justification bits are used in PDH frames to accommodate slight variations in the rates of input tributaries.
For example, the E1 signal may vary slightly from its nominal 2048 kb/s rate without hindering the oper-
ation of the 2–8 muldex. Examine the frame structures given in Figure 13.18 and determine the allowable
range of bit rate variation in the following CEPT PDH signals:
(a) E1
(b) E2
(c) E3.

13.9 We showed that the maximum efficiency of SDH is 𝜂 max = 96.30%. Considering the conveyed PDH signals
to be the message bits, determine the actual efficiency of the following STM-1 frames:
(a) STM-1 assembled as shown in Figure 13.26.
(b) STM-1 assembled as shown in Figure 13.27a.
(c) Why is actual efficiency significantly lower than 𝜂 max ?
(d) Give reasons for the discrepancy between your answers in (a) and (b).
(e) In what situation would the assembly procedure of Figure 13.27a be preferred to that of Figure 13.26?
Telegram: @Persian_IEEEComSoc

13.10 Determine the rates of the following types of bits in the STM-1 frame of Figure 13.27a:
(a) Filler pattern bits
(b) POH bits
(c) TU pointer bits
(d) AU pointer bits
(e) SOH bits.

13.11 Repeat all of Question 13.10, except (c), for the STM-1 frame of Figure 13.26
 876 13 Multiplexing Strategies

13.12 The fixed-length ATM cell is a compromise between the requirements of high efficiency in data transmis-
sion and low delay in voice and video transmission. Determine the following:
(a) The ATM packetisation delay for 64 kb/s voice signals.
(b) The ATM cell duration in a transmission medium operating at (i) 2 Mb/s and (ii) 140 Mb/s.
(c) The efficiency of an AAL1 ATM cell.
Comment on your results in (b) and the impact of line rate on cell delay variation.

13.13 Using a table like Table 13.2, show that the PN sequence of the code generator of Figure 13.37b is as given
in Figure 13.37c. You may assume any initial register state, except of course all-zero.

13.14 Draw the block diagram of a [4, 2] PN sequence generator. Determine the PN sequence. Is this an
m-sequence? Repeat your analysis for a [4, 3] code, and determine whether it is an m-sequence.

13.15 Determine the processing gain of a CDM system that uses BPSK modulation and an m-sequence spreading
code generated by a linear feedback shift register of length 12. Note that the spreading code is periodic,
with period equal to the bit interval of the coded message signal.

13.16 13.16 A TDMA system has the following specifications:

(a) Satellite transponder bandwidth, Bxp = 54 MHz.
(b) TDMA frame duration, Tf = 5 ms.
(c) Guard time between time slots, Tg = 5 μs.
(d) Preamble bits in each ES burst, nbp = 280 bits.
This system is to be used to provide broadband connection for 25 rural bank branches and the satellite
link will use 64-APSK and raised cosine filter of roll-off factor 𝛼 = 0.1. Determine the maximum data rate
(including any redundant bits for error control coding) of each connection.
Telegram: @Persian_IEEEComSoc
 877

Appendix A

Character Codes

Table A.1 International Morse Code

Character Morse Code Character Morse Code

A •— 2 ••———
B —••• 3 •••——
C —•—• 4 ••••—
D —•• 5 •••••
E • 6 —••••
F ••—• 7 ——•••
G ——• 8 ———••
H •••• 9 ————•
I •• : (colon) ———•••
J •——— , (comma) ——••——
K —•— ; (semicolon) —•—•—•
L •—•• ? ••——••
M —— . (period) •—•—•—
N —• ’ (apostrophe) •————•
O ——— " •—••—•
P •—•— / —••—•
Q ——•— - (hyphen) —••••—
Telegram: @Persian_IEEEComSoc

R •—• = —•••—
S ••• ) or ( —•——•—
T — Attention —•—•—
U ••— Break —•••—•—
V •••— End of Message •—•—•
W •—— Error ••••••••

Communication Engineering Principles, Second Edition. Ifiok Otung.

© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 878 Appendix A Character Codes

Table A.1 (Continued)

Character Morse Code Character Morse Code

X —••— Go ahead —•—

Y —•—— OK •—•
Z ——•• SOS •••———•••
0 ————— End of Contact •••—•—
1 •———— Wait •—•••

Table A.2 International Telegraph Alphabet No. 2 (ITA-2). A modification of the

Baudot code.

Character ITA-2 Character ITA-2

Code Code
Letters Figures Letters Figures

A - 11000 Q 1 11101
B ? 10011 R 4 01010
C : 01110 S ’ (US1 = Bell) 10100
2 1
D WRU (US = $) 10010 T 5 00001
E 3 10000 U 7 11100
F ! (UD3 ) 10110 V = (US1 = ;) 01111
G & (UD3 ) 01011 W 2 11001
1 3
H £ (US = #) (UD ) 00101 X / 10111
I 8 01100 Y 6 10101
1 1
J Bell (US = ’) 11010 Z + (US = ") 10001
K ( 11110 Letter shift 11111
L ) 01001 Figure shift 11011
M 00111 Space 00100
N , 00110 Carriage Return 00010
O 9 00011 Line Feed 01000
P 0 01101 Blank (Null) 00000
Telegram: @Persian_IEEEComSoc

1
Where US standard differs from ITA-2, this is indicated in brackets as (US = …);
2
WRU ≡ Who are you?
3
Three codes were left undefined (UD) to allow for national variants.
 Table A.3 EBCDIC code.

Second hexadecimal (hex) digit (0 → F ≡ b8 b7 b6 b5 = 0000 → 1111)

1st hex digit 0 1 2 3 4 5 6 7 8 9 A B C D E F
↓ 0000 0001 0010 0011 0100 0101 0110 0111 1000 1001 1010 1011 1100 1101 1110 1111
0 NUL DLE DS SP & - { } \ 0
1 SOH DC1 SOS RSP / a j ∼ A J NSP 1
2 STX DC2 FS SYN b k s B K S 2
3 ETX DC3 WUS IR c l t C L T 3
4 SEL RES BYP PP d m u D M U 4
5 HT NL LF TRN e n v E N V 5
6 RNL BS ETB NBS f o w F O W 6
7 DEL POC ESC BOT g p x G P X 7
8 GE CAN SA SBS h q y H Q Y 8
9 SPS EM SPE IT ‘ i r z I R Z 9
A RPT UBS SM RFF ¢ ! ^ : [
B VT CU1 CSP CU3 . $ , # ]
C FF IFS MFA DC4 < *
% @
D CR IGS ENQ NAK ( ) _ '
E SO IRS ACK + ; > =
F SI IUS BEL SUB | ¬ ? " ±
Telegram: @Persian_IEEEComSoc
 880 Appendix A Character Codes

Table A.4 ASCII code (ISO-646-IRV).

b7 b6 b5 → hex→
0 1 2 3 4 5 6 7
b4 b3 b2 b1 bin→
000 001 010 011 100 101 110 111
↓hex ↓bin
0 16 32 48 64 80 96 112
0 0000 NUL DLE SP 0 @ P ` p
1 17 33 49 65 81 97 113
1 0001 SOH DC1 ! 1 A Q a q
2 18 34 50 66 82 98 114
2 0010 STX DC2 " 2 B R b r
3 19 35 51 67 83 99 115
3 0011 ETX DC3 # 3 C S c s
4 20 36 52 68 84 100 116
4 0100 EOT DC4 $ 4 D T d t
5 21 37 53 69 85 101 117
5 0101 ENQ NAK % 5 E U e u
6 22 38 54 70 86 102 118
6 0110 ACK SYN & 6 F V f v
7 23 39′ 55 71 87 103 119
7 0111 BEL ETB 7 G W g w
8 24 40 56 72 88 104 120
8 1000 BS CAN ( 8 H X h x
9 25 41 57 73 89 105 121
9 1001 HT EM ) 9 I Y i y
10 26 42 58 74 90 106 122
A 1010 LF SUB * : J Z j z
11 27 43 59 75 91 107 123
B 1011 VT ESC + ; K [ k {
C 1100 12
FF 28
FS 44
, 60
< 76
L 92
\ 108
l 124
|
13 29 45 61 77 93 109 125
D 1101 CR GS - = M ] m }
E 1110 14
SO 30
RS 46
. 62
> 78
N 94
^ 110
n 126
∼
15 31 47 63 79 95 111 127
F 1111 SI US / ? O _ o DEL

The number at the top left corner of each cell is the decimal value of the ASCII code for the character or control signal.
ACK, acknowledge; BEL, bell or alarm; BS, backspace; CAN, cancel; CR, carriage return; DC1 … 4, device control 1 … 4; DEL,
delete; DLE, data link escape; EM, end of medium; ENQ, enquiry; EOT, end of transmission; ESC, escape; ETB, end of
transmission block; ETX, end of text; FF, form feed; FS, file separator; GS, group separator; HT, horizontal tab; LF, line feed;
NAK, negative acknowledge; NUL, null or all zeros; RS, record separator; SI, shift in; SO, shift out; SOH, start of heading; SP,
space; STX, start of text; SUB, substitute; SYN, synchronous idle; US, unit separator; VT, vertical tab.
Telegram: @Persian_IEEEComSoc
 Appendix A Character Codes 881

Table A.5 ISO-8859-1 or Latin-1 alphabet for Western Europe. Only columns A → F
(hex) are shown. Columns 0 → 7 are identical to Table A.4, and columns 8 and 9 define
so-called C1 control characters. (Note: (i) NBSP ≡ non-breaking space; (ii) ̈ is the
diaeresis character used, for example, in naïve.)

hex → A B C D E F
bin → 1010 1011 1100 1101 1110 1111
↓hex ↓bin
0 0000 A0
NBSP B0
∘ C0
À D0
Ð E0
à F0
ð
A1 B1 C1 D1 E1 F1
1 0001 ¡ ± Á Ñ á ñ
A2 B2 2 C2 D2 E2 F2
2 0010 ¢ Â Ò â ò
A3 B3 3 C3 D3 E3 F3
3 0011 £ Ã Ó ã ó
A4 B4 ′ C4 D4 E4 F4
4 0100 ⚫⦻ Ä Ô ä ô
A5 B5 C5 D5 E5 F5
5 0101 =
Y μ Å Õ å õ
A6 B6 C6 D6 E6 F6
6 0110 ¶ Æ Ö æ ö
--

7 0111 A7
§ B7
⋅ C7
Ç D7
× E7
ç F7
÷
8 1000 A8
⋅⋅ B8
, C8
È D8
Ø E8
è F8
ø
A9 B9 1 C9 D9 E9 F9
9 1001 © É Ù é ù
AA BA CA DA EA FA
A 1010 a o Ê Ú ê ú
AB BB CB DB EB FB
B 1011 « » Ë Û ë û
AC BC 1 CC DC EC FC
C 1100 ¬ /4 Ì Ü ì ü
AD BD 1 CD DD ED FD
D 1101 - /2 Í Ý í ý
AE BE 3 CE DE EE FE
E 1110 ® /4 Î þ î þ
AF BF ? CF DF EF FF
F 1111 – Ï ß ï ÿ
Telegram: @Persian_IEEEComSoc
 883

Appendix B

Trigonometric Identities

Let us start from the following compound angle relations, which can be readily obtained from the solution of
triangles
sin(A + B) = sin A cos B + cos A sin B (B.1)

sin(A − B) = sin A cos B − cos A sin B (B.2)

cos(A + B) = cos A cos B − sin A sin B (B.3)

cos(A − B) = cos A cos B + sin A sin B (B.4)

We obtain identities for the multiplication of sinusoids as follows:
(i) Subtract Eqs. (B.3) from Eq. (B.4) to obtain Eq. (B.5).
(ii) Add Eqs. (B.3) and (B.4) to obtain Eq.(B.6).
(iii) Add Eqs. (B.1) and (B.2) to obtain Eq. (B.7).
Thus
1
sin A sin B = [cos(A − B) − cos(A + B)] (B.5)
2
1
cos A cos B = [cos(A + B) + cos(A − B)] (B.6)
2
1
sin A cos B = [sin(A + B) + sin(A − B)] (B.7)
2
Noting that Eqs. (B.1–B.7) hold for all values of A and B, we can obtain other very useful identities as follows:
Telegram: @Persian_IEEEComSoc

(i) Replace B with A in Eqs. (B.1), (B.3), (B.5), (B.6), and (B.4) to obtain Eqs. (B.8) to (B.12), respectively.
(ii) Substitute B = 180∘ and B = 90∘ , in Eqs. (B.1) to (B.4) to obtain Eqs. (B.13) to (B.16), respectively.
(iii) Finally, substitute A = 0∘ in Eqs. (B.2) and (B.4) to obtain Eqs. (B.17) and (B.18), respectively.
Thus
sin 2A = 2 sin A cos A (B.8)

cos 2A = cos2 A − sin2 A (B.9)

1
sin2 A = (1 − cos 2A) (B.10)
2

Communication Engineering Principles, Second Edition. Ifiok Otung.

© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 884 Appendix B Trigonometric Identities

1
cos2 A = (1 + cos 2A) (B.11)
2
sin2 A + cos2 A = 1 (B.12)

sin(A ± 180∘ ) = − sin A (B.13)

cos(A ± 180∘ ) = − cos A (B.14)

sin(A ± 90∘ ) = ± cos A (B.15)

cos(A ± 90∘ ) = ∓ sin A (B.16)

sin(−B) = − sin B (B.17)

cos(−B) = cos B (B.18)

Making repeated use of Eqs. (B.11) and (B.6) yields the following useful identities for integer powers of a cosine
function
1
cos3 A = (cos 3A + 3 cos A)
4
1
cos4 A = (cos 4A + 4 cos 2A + 3)
8
1
cos5 A = (cos 5A + 5 cos 3A + 10 cos A)
16
1
cos6 A = (cos 6A + 6 cos 4A + 15 cos 2A + 10)
32
1
cos7 A = (cos 7A + 7 cos 5A + 21 cos 3A + 35 cos A)
64
etc., etc. (B.19)
Similar identities can be obtained for a sine function. For example, for odd powers
1
sin3 A = (− sin 3A + 3 sin A)
4
1
sin5 A = (sin 5A − 5 sin 3A + 10 sin A)
16
1
sin7 A = (− sin 7A + 7 sin 5A − 21 sin 3A + 35 sin A) (B.20)
64
Telegram: @Persian_IEEEComSoc
 885

Appendix C

Tables and Constants

C.1 Constants

Acceleration due to gravity g 9.80665 m/s2

Avogadro’s number N 6.023 × 1026 /(kg⋅mol)
Base of natural logarithm e 2.7182818285
Bohr magneton 𝛽 9.27 × 10−24 Am2
Boltzmann’s constant k 1.38 × 10−23 J/K
Characteristic impedance of free space Zo 120𝜋 Ω
Electron charge qe 1.602 × 10−19 C
Electron rest mass me 9.109 × 10−31 kg
Faraday constant F 9.65 × 107 C/(kg⋅mol)
Permeability of free space 𝜇o 1.25663706212 × 10−6 H/m
Permittivity of free space 𝜀o 8.8541878128 × 10−12 F/m
Pi 𝜋 3.1415926536
Planck’s constant h 6.626× 10−34 J⋅s
Proton mass mp 1.672 × 10−27 kg
Speed of light in vacuum c 299 792 458 m/s
Stefan Boltzmann’s constant 𝜎 5.67 × 10−8 W/(m2 K4 )
Universal constant of gravitation G 6.67 × 10−11 Nm2 /kg2
Universal gas constant Ro 8314 J/(kg⋅mol⋅K)
Telegram: @Persian_IEEEComSoc

Communication Engineering Principles, Second Edition. Ifiok Otung.

© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 886 Appendix C Tables and Constants

C.2 SI Units

Quantity Unit Symbol Equivalent

2
Area square metre m —
Electric current ampere A —
Length metre m —
Luminous intensity candela cd —
Units
Basic

Mass kilogram kg —
Plane angle radian rad —
Temperature kelvin K —
Time second s —
Volume cubic metre m3 —
Capacitance farad F A⋅s/V
Charge coulomb C A⋅s
Electric field strength — — V/m
Electric flux density — — C/m2
Derived Units

Electric potential volt V W/A

Energy, work, or heat joule J N⋅m
Force newton N kg⋅m/s2
Frequency hertz Hz s−1
Inductance henry H V⋅s/A = N⋅m/A2
Magnetic field strength — — A/m
Magnetic flux weber Wb V⋅s = N⋅m/A
Magnetic flux density tesla T Wb/m2
Power watt W J/s
Pressure or stress pascal Pa N/m2
Resistance ohm Ω V/A
Telegram: @Persian_IEEEComSoc
 C.3 Complementary Error Function erfc(x) and Q function Q(x) 887

C.3 Complementary Error Function erfc(x) and Q function Q(x)

Telegram: @Persian_IEEEComSoc
 888 Appendix C Tables and Constants
Telegram: @Persian_IEEEComSoc
 C.3 Complementary Error Function erfc(x) and Q function Q(x) 889
Telegram: @Persian_IEEEComSoc
 891

Index

a ADPCM See Adaptive differential pulse code

AAL See ATM (Asynchronous transfer mode), modulation
adaptation layer ADSL See Asymmetric digital subscriber line
Absorption Advanced mobile phone system See AMPS
atmospheric 434–435 Advanced Research Project Agency Network 15
cloud 391–392 AFC 572
gaseous 390–391 A-law PCM 630, 653–658
hydroxyl 372 practical implementation of 658, 681
infrared 372 signal to quantisation noise ratio of 657–661
ionospheric 386, 389 Alias frequency 608–611
medium 373 Aliasing
optical fibre 371, 374 distortion 607
filter design 613–615
AC 42, 137, 166, 207, 208, 569
steps to minimise 612–613
Acoustic signal 2, 61
Alignment channel 829
Acquisition 863
Alternate mark inversion See AMI
Active power 159, 165, 352
Alternating current See AC
Adaptive delta modulation (ADM) 668
AM See Amplitude modulation
Adaptive differential pulse code modulation (ADPCM)
American National Standards Institute 9, 844
664
American Standard Code for Information Interchange
Adaptive equalisation 787
8, 64, 826
Adaptive prediction 664, 681
AMI 42, 199, 674–677
Telegram: @Persian_IEEEComSoc

Adaptive predictive coding (APC) 673 Ampere’s modified law 383

Adaptive quantisation 664 Amplifier See operational amplifier
ADC See Analogue to digital conversion Amplitude modulation 473 See also ASK
Additive white Gaussian noise 437–448 See also demodulators 492
AWGN features and applications 426
ADM See Adaptive delta modulation modulation factor 476–478
Administrative unit 842, 844 modulation sensitivity 474–475, 477
group 844 modulators 488
pointer 842 power 486
Admittance 335, 347, 357 receivers 498
Communication Engineering Principles, Second Edition. Ifiok Otung.
© 2021 John Wiley & Sons Ltd. Published 2021 by John Wiley & Sons Ltd.
Companion Website: www.wiley.com/go/otung
 892 Index

Amplitude modulation (contd.) ARPANET See Advanced Research Project Agency

spectrum 498 Network
transmitters 491 ARQ 33
variants 502 ASCII See American Standard Code for Information
waveforms 475 Interchange
Amplitude response 298, 581, 582 ASK
Amplitude shift keying See ASK binary modulation 46, 712
Amplitude spectrum 204, 224 coherent binary detection 719
AMPS 808 M-ary ASK 732–737
Analogue communication system non-coherent binary detection 725–727
baseband 35 Asymmetric digital subscriber line 331
compared to digital 33 Asynchronous data transmission 10, 622
modulated 44, 473, 529 ATM (Asynchronous transfer mode) 34, 805
noise effects 462, 588 adaptation layer 849, 852
Analogue signal 33 cell header 851
Analogue to digital conversion 30, 41, 627 features 852
Ancillary signal 65 versus IP 852–853
Angle 86–87, 529 layered architecture 847
Angle modulation 530–531 network components 850
basic concepts 530 Atmosphere
waveforms 475, 543 noise temperature 434
Angular frequency 87 vertical structure 384–386
Anisotropic medium 432 Attenuation
Anomalous propagation 395–396 distortion 312–314
ANSI See American National Standards Institute in metallic lines 332–334, 338
Antenna in optical fibre 371–376
gain 104–105, 418, 421, 430, 461 in radio 390, 416, 418, 421
height 415, 418–421 Attenuation coefficient 371, 373, 376
noise 434–437 Attenuation constant 338, 340, 376, 383–384
as transducer 58, 60 Attenuation pad 356–357
Anti-alias filter 613–615 AU See Administrative unit
Antipodal 696, 710, 729, 747 Audio
APC See Adaptive predictive coding input device 22
Telegram: @Persian_IEEEComSoc

APD See Avalanche photodiode output device 24

Aperiodic signal See nonperiodic signal wideband signal 63
Aperture distortion loss (ADL) 237–242, 623 AUG See Administrative unit, group
Aperture effect 622–623 Autocorrelation function 138, 172–173
APS See Automatic protection switching Automatic frequency control See AFC
APSK 749 Automatic gain control 495, 822
Arbitrary binary transmission system 709 Automatic protection switching 842
Armstrong Automatic repeat request See ARQ
Edwin Howard 19 Avalanche photodiode 60, 824
wideband frequency modulator 570 AWGN See also noise sources
 Index 893

characterisation 437–448 Beat frequency 97–98

effects on Bell
AM 462–464 Alexander Graham 13
digital communication systems 465–470, FDM system 821
701–711 laboratories 16, 19
DSB 462–464 BER
FM 588–594 of arbitrary binary transmission 708–710
SSB 462–464 of binary ASK, FSK and PSK 708
of M-ary APSK 752–754
b of M-ary ASK 734–737
Baird, John Logie 19 of M-ary FSK 746–749
Balanced discriminator 584 of M-ary PSK 743–746
Balanced modulator See product modulator trade-offs 465–470, 754–756
Balancing pulse 676–677 Bessel function
Bandlimited white noise 442–445 graph and discussion 559–560
Bandpass filter 305–306 table 561–562
Bandpass signal sampling 603 Binary amplitude shift keying See ASK
Bandstop filter 306 Binary block code 678
Bandwidth Binary coded decimal 8–9
Carson’s 563 Binary frequency shift keying See FSK
3 dB 304 Binary modulation 712
definitions 302–308, 564 Binary phase shift keying See PSK
efficiency 730–732 Binary symmetric channel 704
fractional power containment 306 BIP See Bit interleaved parity
noise equivalent 308 Biphase code 676
noise trade-off 586, 748 Bipolar baseband system 704–708
null 304 Bipolar junction transistor See Transistor
1% 563 Bipolar non-return-to-zero See NRZ code
subjective 303 Bipolar with n Zeros Substituted See BnZS code
Barcode reader 23 B-ISDN 834, 846, 850–852
Baseband communication system terminal equipment 850
analogue 33–35 Bit duration 176, 199
digital 41, 674 Bit energy See Energy
Telegram: @Persian_IEEEComSoc

discrete 36–39 Bit error rate, bit error ratio See BER
Baseband signal 35, 303, 602 Bit interleaved frame 825
Baseline wander 42 Bit interleaved parity 842
Basis function 412–417 Bit-interleaving 825, 830–832
BASK See ASK Bit rate 465, 755
Baud 8 Bit robbing 826, 834
Baudot Bit stuffing 832, 839
code 7–8 BJT See Transistor
Jean-Maurice-Émile 7 Blackman-Harris window 285–286
BBC See British Broadcasting Corporation Block code 677–681
 894 Index

BNRZ See NRZ code Cascade connection

BnZS code 676 definition of 183, 454
Bonfires 3 noise temperature of 455
Bound 746 Cascade connection, noise figure of 454–457
BPSK See PSK Cathode ray tube 26
B-pulse See Balancing pulse CATV 59, 333
2B1Q line code 43 Causal, Causality 112, 183
Brewster angle 398–400, 402–403 CBR 846, 849
Brickwall filter 606, 731, 812 CCD 14
Brightness temperature 433–434 CCDF See Complementary cumulative distribution
British Broadcasting Corporation 16 function
Broad-area coverage 19, 387 CCITT 8, 14 See also ITU
Broadband ISDN See B-ISDN CCTV 35
BSF See Bandstop filter CD See Compact disk
B-TE See B-ISDN, terminal equipment CDM See Code division multiplexing
Burst bit rate 845, 870 CDMA (Code division multiple access) 871
Butterworth filter 614–615, 624 Cell 50, 807, 851
Byte-interleaving 830, 839, 842, 844 Cell delay variation 847, 852, 876
B3ZS 676 Cell delineation 848
B4ZS 676 Cell header 851
B6ZS 676 Cell loss priority 851
B8ZS 676 Cell rate decoupling 848
Cell relay switching 846
c Cellular telephony 813, 864, 873–874
Canonical representation of noise 444 CELP 673
Capacitance sensitivity 574 Central limit theorem 139, 143, 149, 438
Capacitor 22, 42, 110 Central office 331
charging/discharging 496 CEPT PCM-30 (E1) system 827
Capture CEPT PDH hierarchy 827–832
effect 594 CEQ 850–851
range 580 Channel bandwidth See Bandwidth
ratio 594 Channel coding 30
Carrier Channel frequency response 297–301
Telegram: @Persian_IEEEComSoc

amplitude 46, 473 Channel impulse response See Impulse response

demodulation 32 Channel model 449, 451, 590
frequency 20, 46, 529 Channel noise See Noise sources
frequency extraction 497–498, 725, 797–799 Channel transfer function 297, 311, 787
modulation 30–31, 35, 45, 382 Channel translating equipment See CTE
to noise ratio 425, 437, 462, 465, 754 Character codes 7–13, 30, 877–881
phase 46, 529 Characteristic impedance 340, 347, 353
phase synchronisation 508–509 Charge coupled device See CCD
suppression 504, 514 Chip duration 855–856
Carson’s bandwidth 306, 463, 563 Chromatic dispersion 377
 Index 895

Chrominance signals 63, 303 Coding gain 755, 758

Circuit Coherent binary amplitude shift keying See ASK
balanced discriminator 584 Coherent binary frequency shift keying See FSK
balanced modulator 446, 504 Coherent binary phase shift keying See PSK
CCTV 35 Coherent demodulation 498, 516, 518, 521
de-emphasis 587 Coherent detection 719, 723
diode demodulator 493 Coherent M-ary amplitude and phase shift keying See
integrator 568 APSK
operational amplifier 488, 569 Coherent M-ary amplitude shift keying See ASK
pre-emphasis 587 Coherent M-ary frequency shift keying See FSK
switching 15, 48–50 Coherent M-ary phase shift keying See PSK
telegraph 16 Coherent minimum shift keying See MSK
telephone 17 Coherent quadriphase shift keying (QPSK) See QPSK
Circular APSK constellation 750 Colour display 26
CL See Connectionless (CL), packet switching Coloured noise 442, 592
Clarke, Arthur C. 19 Colour printing 27
Clipping distortion 314 Common signalling 826, 833
Clock Communication channel See Channel
content 42, 675 Communication system
extraction 797 block diagram 21
recovery 44 design issues 754
Closed circuit television See CCTV types 32
Closed loop power control 865 Commutator 39
CLP See Cell loss priority Compact disk 29, 613
CMI 676 Companding See also A-law PCM; 𝜇-law PCM
Coaxial cable 329, 332–333 definition 641
Co-channel interference 390, 808 gain 647–650
Codebook 30, 673 penalty 647–650
Codec 44, 656, 667, 669, 814 Complementary cumulative distribution function
Code division multiplexing 136, 140
basic introduction 853 Complementary error function See also Q-function
benefits 853 definition 141
PN codes 864 table 886
Telegram: @Persian_IEEEComSoc

power control 864–865 Compressor characteristic 642

processing gain 866 Computable code 10
receiver 583 Cone of acceptance 369–370
synchronisation 863 Conjugate structure algebraic code excited linear
transmitter 856 prediction (CSA-CELP) 673–674
Coded mark inversion See CMI Connectionless (CL), packet switching 52, 849
Code efficiency 43, 678 Connection oriented (CO), packet switching 50–51,
Code excited linear prediction See CELP 849
Code radix 43 Consonant sound 61
Codeword 8–13, 30 Constant bit rate See CBR
 896 Index

Constellation diagram 690, 694 d

Continuously variable slope delta modulation DAC See Digital to analogue conversion
(CVSDM) 668 Data communication See Data transmission
Continuous phase frequency shift keying 716 Data communication channel 842
Continuous refraction 369, 386 Data compaction 30
Continuous spectrum 274, 484, 789 Data compression 30, 34, 42, 64
Continuous-time signal 65–67 Data encryption 33
Continuous-time system 109, 181 Data input devices 23
Convergence sub-layer 849 Data-in-voice 814
Conversion time 618, 623 Data-over-voice 814
Convolution 186, 190, 196, 260, 790–797 Data terminal equipment 23, 64, 333
Cooke, William Fothergill 5 Data transmission 13, 33
Correlated pulses 724, 727 3 dB bandwidth 304–305
Correlation coefficient 176–180 6 dB bandwidth 326
of specific systems 60 dB bandwidth 304, 326
binary transmission symbols 710 DC 62, 100, 108
FSK symbols 717 DCC See Data communication channel
M-ary ASK symbols 734 Decibel 100
Decision threshold 42, 720–724
M-ary PSK symbols 744
Decommutator 39, 41
Correlation receiver 719, 723, 790
De-emphasis 30, 586
Correlator 701–702, 790, 864
de Forest, Lee 13
Cosine filter 774–777
Delay
Cosine rule 96
differential 312
Costas loop 508, 524
distortion (See Phase distortion)
Coupling 369–371, 576
excess 130
loss 371
group 129, 312, 377, 389
Covariance 176
line differentiator 582
CPFSK See Continuous phase frequency shift keying
line predictor 663
CRC 529, 829–830, 849
mean 130
Cross-correlation 864–865 packetisation 50, 678, 847
Cross-polarisation 394, 432 processing 670–671, 673
Crosstalk 41, 43, 103, 332–333, 367, 432 propagation 3, 5, 311, 314
Telegram: @Persian_IEEEComSoc

CRT See Cathode ray tube rms 130

Crystal-controlled oscillator 492, 517 switching 847
CSA-CELP See Conjugate structure algebraic code Delta modulation (DM) 664
excited linear prediction design parameters 666
CTE 508, 514 merits and demerits 666
Customer equipment See CEQ prediction filter 666
CVSDM See Continuously variable slope delta quantisation error 664
modulation waveform 667
Cycle 84 Delta sigma modulation (DSM) 668
Cyclic redundancy check See CRC Demodulator See also Receiver
 Index 897

AM 497 Digital subscriber line See DSL

DSB 507 Digital to analogue conversion 32–33, 64
FM 463, 577, 581 Digital translator 642
ISB 518 Digital transmission model 699
PM 537, 581 Digital versatile disc See DVD
QAM 527 Diode demodulator
SSB 561 circuit and operation 493–495
VSB 522 conditions 495–496
De-multiplexer Diode detector See Diode demodulator
analogue TDM 40, 825 Diode ring modulator 504–507
FDM 813 Dirac delta function 78, 618
optical (WDM) 824 Direct current See DC
Dense wavelength division multiplexing (DWDM) Direct sequence spread spectrum
20, 824 processing gain 855
Depolarisation 394 receiver 858–863
Design considerations 639, 666, 754 transmitter 856
Detector, ASK 719 Discrete baseband communication system 36, 38
DPSK 729 Discrete-time signal 65–67
FSK 721 Discrete-time system 109, 181
PSK 721 Dispersion 130, 314, 334, 368, 376–379, 386, 389,
QAM 47, 474, 527, 691 824
Deterministic signal 68 Display devices
Deviation ratio 534 See also Modulation index CRT 26
Dibit 43 flat panel 26
Differential group delay 377 printers 26–27
Differential NRZ code 675 Distortion
Differential phase shift 394 aliasing 283
Differential phase shift keying See DPSK aperture effect 622
Differential pulse code modulation See DPCM attenuation 311, 314
Differentiator 581 delay 312, 822
Diffraction 408 harmonic 315
Diffuse reflection 406–407 intermodulation 315, 374
Digital baseband receiver 55 nonlinear 314–316
Telegram: @Persian_IEEEComSoc

Digital communication system See also Design overload 664, 668

considerations phase 47, 111, 204
baseband 41 quantisation 34
compared to analogue 33 Distortionless transmission 311
modulated 44 DIV See Data-in-voice
noise effects 701 DM See Delta modulation
Digital modulation techniques, comparison Doolittle, Thomas 17
755–756 Doppler
Digital signal 65–67 Christian Andreas 97
Digital signal processing See DSP frequency shift 97, 386
 898 Index

Double sideband suppressed carrier amplitude Earth-space slant paths 416

modulation (DSB) 474, 502, 524 EBCDIC 8, 9, 879
applications 509 Efficiency
demodulator 507 antenna 430
effect of phase error 508, 725 bandwidth 730
modulation 515 code 43
phase synchronisation 508–509 data transmission 827
spectrum 503 EHF See Extra high frequency
waveform 503 Electromagnetic waves 32, 59–60, 84, 334, 380
Double sided spectrum 227–230 Encryption 30, 33
DOV See Data-over-voice Energy
DPCM 661–662 of bit 692
DPSK of pulse 691
BER 460 Rayleigh’s theorem 261
detection 458 of signal 69, 167
generation 458 of sinusoidal symbol 686
introduction 457 Energy spectral density (ESD) 288–291, 301–302
M-ary 746 Engineer’s orderwire 842
Droop 42–43, 675 Envelope detector 684, 726–727 See also Diode
DS See Direct sequence demodulator
DSB See Double sideband suppressed carrier Envelope distortion 477
amplitude modulation Envelope representation of noise 444
DSL 17, 33, 43, 331–332 EOW See Engineer’s orderwire
DSM See Delta sigma modulation Equalisation See Equaliser
DSP 204, 294, 642, 670 Equaliser 44, 120, 305, 312–314, 623, 773, 787
DTE See Data terminal equipment Equivalent noise temperature 450
Duality 259 Ergodicity 138
Ducting propagation 396 Error burst 851
Duobinary signalling 774 Error correction 33–34
Duplex Error function See Complementary error function
full duplex 31–32, 758 ESF See Extended superframe
half duplex 31, 758 Euclidean space 690
Duty cycle 71 Even function 69
Telegram: @Persian_IEEEComSoc

DVD 29 Excess delay See Delay

Dynamic range 25, 34, 61 Exclusive-OR See XNOR gate
Exponential distribution 157
Extended binary coded decimal interchange code See
e EBCDIC
Earth Extended superframe 834
radiated noise 434 Extra high frequency 45, 380
radius 387 Extrinsic loss 20, 375
station receiver 459, 472 Eye diagram 799–800
 Index 899

f Fleming, John 13
Facsimile See Fax Flicker noise 436
Faraday FM See Frequency modulation
constant 885 FM stereo transmission 510
law of electromagnetic induction 22, 382–383 FM transmitter 584
Michael 389 Forward error correction 33, 758
rotation 386, 389 Fourier
Far end block error 843 series 205
Farnsworth, Philo 19 synthesis 205–206, 214
Fast fading 149, 314, 421 theorem 205–206
Fast Fourier transform 204, 277 transform 253
FAW See Frame alignment word transform table 264
Fax 14–15, 64 Four-wave mixing (FWM) 374
FCC See Federal Communications Commission Frame alignment word
FDMA (Frequency division multiple access) 868 bunched or grouped 520, 523–524, 526, 535
FDM signals 875 See also Frequency division distributed 520, 528, 529
multiplexing Frame organisation 825
group 815 Free space propagation 416, 418, 421
hypergroup 816, 818 Frequency comparison pilot 822
jumbogroup 821 Frequency deviation 530
mastergroup 820, 822 Frequency discriminator 581
pilot tones 822 balanced discriminator 584
supergroup 814–818 delay line differentiator 582
supermastergroup 820 RC differentiator 583
FDX See Duplex Frequency division multiplexing (FDM) 809
FEBE See Far end block error See also FDM signals
FEC See Forward error correction flat-level FDM 812–813
Federal Communications Commission 306, 424 hierarchical FDM 814
Fessenden, Reginald 18 standards
FET See Transistor Bell 821
FFT See Fast Fourier transform European 820
Fibre See Optical fibre UK 816
Field effect transistor See Transistor Frequency modulation 529 See also FSK
Telegram: @Persian_IEEEComSoc

Filter applications 595

anti-alias 613 bandwidth 558, 560, 563
ideal (brickwall) 606 comparisons 540, 543, 551
matched 787–797 demodulators 581
raised cosine 769 merits and demerits 530, 594
root raised cosine 771 modulators 567
Flags 3 narrowband 549
Flat-level FDM 812–813, 816, 820 noise performance 567, 584–585
Flat-top sampling 235, 240, 618 power 563
 900 Index

Frequency modulation (contd.) Gibbs’ phenomenon 206

waveforms 546 Global System for Mobile communication 670
wideband 557, 567 Glottis 60–61
Frequency multiplication 567, 569 Gold sequence 864
Frequency resolution 284 Granular noise 668
Frequency response Gray code converter 732
of human ear 25 Gray coding 699, 708, 732, 781
of system 297 Ground wave 386–387
of tuned circuit 583 Group delay 129, 312, 389
Frequency reuse 808, 872 Group signal See FDM signals
Frequency sensitivity 530 Group translating equipment See GTE
Frequency shift keying See FSK GSM See Global System for Mobile communication
Frequency swing 533 GTE 816
Frequency translation 500 Guard band 432, 605
Frequency variations in PM 537
Fresnel coefficients 397–398 h
Fresnel zones 410 Half duplex See Duplex
FSK See also Bandwidth Hamming window 285, 287
BER 708, 746 Hanning window 285
coherent detection 719, 723 Harmonic
frequency spacing 716 distortion 315
generation 715 frequency 68
M-ary generation and detection 738 Hartley modulator 516
noise-bandwidth trade-off 748 Haykin, Simon, xxiii
noncoherent detection 727 HDB3 674, 676
spectrum 716 HDSL See High-speed digital subscriber line
FTTC (Fibre-to-the-cabinet) 331, 365 HDTV See High-definition television
FTTP (Fibre-to-the-premises) 365 HDX See Duplex
Full cosine roll-off characteristic 770 Header error correction 848
Full duplex See Duplex Heaviside
Functions 71, 886 condition 341
Fundamental frequency 67–68 Oliver 330
Future of FDM 813–814 HEC See Header error correction
Telegram: @Persian_IEEEComSoc

Heliogram 4
g Heliography 4
Gain response See Amplitude response Hertz
Gaseous absorption 390 cycles per second 84
Gaussian distribution 139, 438 Heinrich Rudolf 18
Gaussian noise 437 See also AWGN Heterosphere 385
Generic flow control 849 HF See High frequency
Geometric representation 689 High-definition television 26
Geostationary orbit (GEO) 19, 387, 756 High density bipolar with 3 zero maximum See
GFC See Generic flow control HDB3
 Index 901

High frequency 380 Inter-arrival time 159, 161

High-level AM transmitter 492 Interference
Highpass filter 306, 572, 593, 773 acoustic 2
Hilbert transform 515 co-channel 390, 808
Hockham, George 365 image 500–501
Hollerith code 8–9 intentional 853
Hollerith, Herman 8 ISI (See ISI)
Homosphere 385 types 431–432
HPF See High pass filter Intermediate frequency (IF) amplifier 500
Huygens’ principle 408, 411 Intermodal dispersion 376
Hybrid coders 673 Intermodulation product 315, 811, 868
Hybrid modulation 698 International Business Machines Corporation 9
Hybrid multiplexing 824 International Organization for Standardization 10
Hypergroup signal See FDM signals International Telecommunication Union See ITU
Internet 12, 15, 33
i Inter-network interface 850
IBM See International Business Machines Corporation Inter-symbol interference See ISI
Ideal filter See Brickwall filter Intramodal dispersion 377
Ideal Nyquist channel 768 Intrinsic loss 20, 371
Idle cell 846 Inverse Fourier transform 254
ILD See Injection laser diode Ionosphere 46, 385
Image interference 500–501 Ionospheric absorption 389
IMBE See Improved multiband excited vocoder Ionospheric effects 389
IMP See Intermodulation product Ionospheric scatter 385, 388
Impedance 165, 300, 335, 347, 569 IP (Internet protocol) 15, 50, 52, 331, 852
matching 183, 353, 363, 454 ISB See Independent sideband amplitude modulation
Improved multiband excited vocoder (IMBE) 672 ISDN See Integrated services digital network
Impulse function 78 ISI 63, 64, 381–393, 401, 572
Impulse response 79, 181–186, 299 ISO See International Organization for Standardization
Incident wave 337, 397 8859 code 10–11
Independent sideband amplitude modulation 518 ITU
Inductor-capacitor oscillator (LCO) 574 organisation 424
Information capacity 780 referenced standards
Telegram: @Persian_IEEEComSoc

Information sink 24 G.653 425

Information source 21 G.655 425
INI See Inter-network interface G.721 664
Injection laser diode 59, 368 G.722 663
Instantaneous frequency 221 G.726 663
Instantaneous phase 535 G.727 671
Instantaneous sampling 79, 235, 600–606 G.729 674
Integrate-and-dump filter 790 P.530 393
Integrated services digital network 331, 833–834 P.531 389
Integrator 568 P.618 393, 394
 902 Index

ITU (contd.) Line of sight 103, 387

P.676 390 Line spectrum 227, 229
P.838 392 Line termination 353
P.840 392 Liquid crystal display 27
P.1057 198 Local access network 426
Local area network 33, 330, 676, 758
j Local loop 16, 17, 20, 43, 328, 331, 425
Jitter 49, 51, 797 Local mean power See power
Johnson noise 432 Lock range 580
Jumbogroup signal See FDM signals Lodge, Oliver 12
Justification 831–832, 835, 841 Logarithmic units 99
Lognormal distribution 148–149
k Log PCM 658, 661
Kaiser–Bessel window 285 LOS See Line of sight
Kao, Charles K. 19, 365 Loudspeaker 24
Kerr effect 374 Low bit rate speech coding 30, 34, 42, 668–674
Knife-edge diffraction 408, 411 Low earth orbit (LEO) 19, 387
Lower sideband 474, 484
l Low frequency 35, 45, 380
LAN See Local area network Low-level AM transmitter 491
Laplace transform (LT) 204, 291–292 Lowpass filter 190, 260, 299–301, 306, 326, 443–444
Laser diode 31, 36 Lowpass signal See baseband signal
Lattice modulator 504 LPC See Linear predictive coding
LBR See Low bit rate LPF See Lowpass filter
LCD See Liquid crystal display LSB See Lower sideband
LCO See Inductor-capacitor oscillator LSF See Side frequency
LC oscillator modulator 573 LTI See Linear time-invariant system
LED 27, 59–60, 368 Luminance signal 63, 520–521
LEO See Low earth orbit (LEO)
LF See Low frequency m
Light emitting diode See LED Magnetic
Limiter 577 disk 28
Linear ADC See Linear PCM reader 23
Telegram: @Persian_IEEEComSoc

Linearity 113, 258 tape 28

Linearly-varied-gain modulator 488 MAN See Metropolitan area network
Linear PCM 634–641 Manchester code 676, 678, 704
Linear predictive coding (LPC) 672 Marconi, Guglielmo 18
Linear time-invariant system 79, 113, 116, 180 Mark (binary 1) 8, 14, 18, 42, 675
Line code Markov process 159
classes 674 M-ary modulation See also Multilevel modulation
desirable characteristics 42–44 kB1M block code 780
Line decoding 44 M-ary APSK 749
Line impedance 343, 347, 364 M-ary ASK 732
 Index 903

M-ary DPSK 746 Mixed excitation linear prediction vocoder (MELP)

M-ary FSK 746 673
M-ary PSK 737 Mobile communication 388, 421–423, 808
Masking effect 3, 25 Modem 33, 331, 472, 758
Mastergroup signal See FDM signals Modified duobinary filter 779
Mastergroup translating equipment See MTE Modulation
Matched filter 787 definition 44
derivation 788 roles 44–46
worked examples 791 types 46–47
Material dispersion 368, 377 Modulation depth 496
MATLAB code 219, 221, 283 Modulation factor 476
Maurer, Robert 20 Modulation index 463, 477, 534
Maximum-length sequence 857–858, 863–866 Modulation sensitivity 474, 581
Maxwell Modulator See also Transmitter
equations 18, 382–384 AM 488
James 18 APSK 749
MCPC (multiple channel per carrier) 868 ASK 712
Mean opinion score 462, 669 DPSK 727
Mean square quantisation error 631 DSB 504
Mean square value 163, 168 FM 530
Medium earth orbit (MEO) 19, 387 FSK 715
Medium frequency 380 ISB 518
MELP See Mixed excitation linear prediction vocoder Narrowband FM and PM 567
MEO See Medium earth orbit PSK 714
Mesosphere 385 QAM 47, 527
Message identifier 849 SSB 510
Metallic line systems 328 VSB 521
Metropolitan area network 20, 365 Morse
Microphone code 7
carbon 22 Samuel 6
condenser 22 MOS See Mean opinion score
dynamic 22 MPE-LPC See Multipulse excited linear predictive
electret 22 coding
Telegram: @Persian_IEEEComSoc

electromagnetic 22 MSB (Most significant bit) 10, 54, 628, 632, 650, 737,
piezoelectric 22 840
variable-resistance 22 m-sequence See Maximum-length sequence
Microwave radio relay 19, 460 MSK 717
MID See Message identifier MSOH See Multiplex section overhead
MIDI See Musical instruments’ digital interface MSQE See Mean square quantisation error
Midrise quantisation 628, 630, 634 MTE 822
Midstep quantisation 628–634 𝜇-law PCM 661
Mie scatter See Scattering practical implementation 653–654
Minimum shift keying 717 See also MSK SQNR 657–661
 904 Index

Muldex canonical representation 444

Add/drop 843, 846 envelope representation 444
E1 hierarchy 830 noise equivalent bandwidth 308–311
J1 hierarchy 836 Narrowband phase modulation (NBPM) 551 See also
NA PDH hierarchy 832 Narrowband frequency modulation
primary 827 National Television System Committee 509
SDH 839 Natural sampling 616–620
Multiframe 520, 523 Neper 101–102
Multiframe alignment word 523–524 Network node interface 850
Multilevel modulation Network service provider (NSP) 331
bandwidth efficiency 730 NNI See Network node interface
design considerations 465, 754 Noble, Daniel 19
Multimode graded index fibre 369, 377 Noise-bandwidth trade-off 628, 748
Multimode step index fibre 368, 376 Noise calculations in systems 594–607
Multipath propagation Noise figure (noise factor) 451, 454
in optical fibre 368, 376 Noise immunity 34, 732
in radio 130, 143, 149, 153–156, 313–314 Noise sources 432–437 See also AWGN
Multiple access 31, 867 Noise weighting 127
Multiplexer Nondeterministic signals 290 See also Random
analogue TDM 39 signals
ATM 850 Nonlinear PCM See Log PCM
DSLAM 331 Nonperiodic signal 67, 204, 253
FDM 810 Non-return-to-zero 377, 674, 712 See also NRZ code
optical (WDM) 824 Nonuniform PCM See Log PCM
Multiplexing Nonuniform quantisation 641 See also Quantisation
role 805 Normal distribution See Gaussian distribution
statistical 846 Normalised average power See Power
types 806–809 Normalised bandwidth 565
Multiplex section overhead 842 NRZ code 674–675
Multipulse excited linear predictive coding (MPE-LPC) NSP See Network service provider
673 NTSC See National Television System Committee
Music 62 Null bandwidth 304
Musical instruments’ digital interface 22 Numerical aperture (NA) 369–371
Telegram: @Persian_IEEEComSoc

loss 376
n Nyquist
Narrowband frequency modulation (NBFM) channel 310
amplitude spectrum 552 filtering 767
amplitude variations 556 interval 599
derivation 549–551 rate 600
phase spectrum 554
phasor approach 555 o
waveforms 553 OC See Optical carrier
Narrowband noise 442 Octet 825
 Index 905

Odd function 69 Uniform 637–638

OLED (organic light-emitting diode) 27 PDH See Plesiochronous digital hierarchy
On-off keying See OOK PDM See Pulse duration modulation
On-off signalling 674. Percent modulation 530, 533–534
OOK 14, 46 See also ASK Periodic signal 67–69, 205
Opamp See Operational amplifier Periodogram 290, 303
Open loop power control 865 Phase constant 338, 340, 382
Open systems interconnection 852 Phase demodulator 580
Operational amplifier 488, 569, 618 Phase deviation 536–537
Optical carrier 20, 46, 366 Phase difference 89
Optical character reader 23 Phase discrimination 514–516
Optical disk 29 Phase discriminator 498
Optical fibre Phase distortion 311–312
advantages 366–367 Phase error 508–509, 517, 585–589, 724–727, 762
attenuation 371 Phase locked loop (PLL) 497–498, 546, 576, 579, 724,
disadvantages 367 798, 863
types 367 Phase modulation (PM) 529 See also PSK
Orthogonal 178, 377, 687 basic concepts 535
Orthonormal 687 See also Basis function comparison with AM 551
Oscilloscope 89, 127, 543 comparison with FM 537, 551, 567
OSI See Open systems interconnection demodulators 576
Out-of-word signalling 826, 828 frequency variations 537
Overhead bits 49–51, 827–828 modulators 567
Overload distortion 663–664, 666–668 narrowband 549
Overmodulation 477 power 549
waveforms 543
wideband 557
p Phase response 189, 203, 298
Packet switching 50–52 Phase sensitivity 530, 535
PAM See Pulse amplitude modulation Phase shift demodulator 582
Partial fraction expansion 292, 294 Phase shift keying See PSK
Partition noise 435 Phase spectrum 204, 224
Path loss 416, 418, 421 Phase synchronisation 508–509, 725
Telegram: @Persian_IEEEComSoc

Path overhead 840, 842 Phase variations in FM 540

Payload type identifier 849 Phasors 86, 93, 98
PCM See Pulse code modulation Photodetector 60
PDF 135 Piecewise companding 650, 654
Exponential 157–162 Piecewise linear functions 214
Gaussian 139–142, 438 Pilot carrier 510, 517–519, 524
Lognormal 148–153 PIN diode 60
Poisson 157–162 Plain old telephone service See POTS
Rayleigh 143–148, 447 Plane earth propagation 418
Rician 153–157 Plasma-panel display 54
 906 Index

Plesiochronous digital hierarchy (PDH) 827 Pseudorandom sequence (code) 808, 853–857
problems 838 PSK See also Bandwidth; BER; DPSK
standards binary generation 715
CEPT (E1) system 827 coherent binary detection 719
Japanese (J1) system 836 introduction and waveforms 46–47
North American (T1) system 832 M-ary signal 737
PLL See Phase locked loop Psophometer 103
PM See Phase modulation PSTN See Public switched telephone network
PN See Pseudonoise PTI See Payload type identifier
POH See Path overhead Public switched telephone network 328, 595, 758
Poisson departure process 161 Pulse amplitude modulation 36, 623
Poisson distribution 157 features 38
Polarisation 377, 380, 394, 399, 420–1, 432 system block diagram 37–38
Polarisation mode dispersion 377 waveform 37
Polybinary signalling 779 Pulse code modulation
Post-modulation filter 809, 812–813 basic discussion 16, 41, 613
POTS 331–332, 755 detailed discussion 627
Power 69, 163–167 differential 661–668
local mean 148–149 linear 628–641
logarithmic measures 102 log 641–661
Power efficiency 467–468, 736, 782 Pulse duration modulation 36, 623
Power factor 165, 404 features 38–39
Power spectral density 288–291, 301–302, 440–442, system block diagram 38
591–592, 866–867 waveform 37
Poynting vector 404 Pulse position modulation 36, 623
PPM See Pulse position modulation features 38–39
Preamble See Training sequence system block diagram 38
Predictor, tapped-delay-line filter 663 waveform 37
Pre-emphasis 30, 584–586 Pulse width modulation See Pulse duration
Primary line constants 334 modulation
Primary medium parameters 383, 388, 397 PWM See Pulse duration modulation
Primary muldex See Muldex Pythagoras’ rule 94, 690, 705, 709
Primary rate ISDN (PRI) 833
Telegram: @Persian_IEEEComSoc

Printers 27 q
Probability density function See PDF QAM 47, 527, 717, 749 See also APSK
Probability distribution function See PDF analogue 527
Probability of symbol error 703, 743 See also Symbol Q-function See also Complementary error function
error rate definition 140
Product modulator 497, 504–507 table 886
Propagation constant 338, 340, 348 QOS See Quality of Service
PSD See power spectral density QPSK
Pseudo-Brewster angle 400 BER 705–708
Pseudonoise 853 modulator and detector 738
 Index 907

Quadrature amplitude modulation See QAM Real-time transport protocol (RTP) 52

Quadrature channels 751–752 Receiver 31 See also Demodulator
Quadrature component 143 AM 498
Quadrature null effect 508–510 analogue TDM 825
Quadriphase shift keying See QPSK CDM 858
Quality factor 326, 812, 816 delta sigma modulation 668
Quality of Service 52, 462 digital baseband 44
Quantisation 41, 66, 433, 436, 613, 615, 627–666 digital transmission model 699–700
Quantiser step size 635, 668 DM 665, 667–668
Quantum noise 435 DPCM 662
Quarter-wave transformer matching 355 FM 588
Quasiperiodic signal 67–68 FM stereo 510
Quaternary code 43 matched filter 787
Quaternary phase shift keying See QPSK PAM 38
spread spectrum 854, 855
r superheterodyne AM 499
Radar 32, 58 superheterodyne FM 588
Radian 84 tuned radio frequency (TRF) 498
Radiation resistance 45 WDM 813, 823
Radio 18–19, 380–424 Reconstruction filter 38, 600, 603, 623
attenuation 390 Rectangular pulse 75
frequency allocation 424 Rectangular window 266, 268, 284–288
noise 433 Reeves, Alec 16
path loss 416, 418, 421 Reflectance 370
propagation effects 384 Reflected wave 337, 397
propagation modes 384 Reflection 342, 397, 406
spectrum 380 coefficients 342–345
Rain attenuation 136, 392–393 Refraction 395, 397
Raised cosine filter 309–310 Refractivity 395
Raised cosine window 266–268, 285–288, 291 Regenerator section overhead 842
Ramp pulse 76 Regular pulse excitation (RPE) 673
Random access memory (RAM) 29 Return-to-zero See RZ code
Random processes 134, 138 Rician distribution 153, 155, 423
Telegram: @Persian_IEEEComSoc

Random signal parameters 135 RLL See Run length limited code
Random signals 68, 134 Roll-off factor 309, 310, 465, 731
Rated system deviation 533 Root-mean-square value 168
Rayleigh Root raised cosine filter 771
distribution 143, 423 Rough surface scattering See Scattering
energy theorem 167 Rounding quantisation 66, 628–634, 665
scattering 20, 371–373 RSOH See Regenerator section overhead
RC differentiator 583 Run length limited code 676
RDS See Running digital sum Running digital sum 678
Reactive power 143, 157, 165–166 RZ code 675–676
 908 Index

s Shape factor 304, 326

Sample and hold circuit 38, 41 Shot noise 435
Sampling methods Sideband 483–484
flat-top 235, 618 Sideband power 486–487, 502
instantaneous 601, 603 Side frequency
natural 616 lower side frequency (LSF) 480–482, 552
Sampling theorem 36, 599–600 upper side frequency (USF) 480–482, 552
applied to band pass signal 603 Sigma delta modulation See Delta sigma modulation
applied to low pass signal 602 Signal characterisation 162
SAR See Segmentation and re-assembly sublayer Signalling 65, 822
Satellite communication 2, 7, 19, 387–388 Signalling channel 826, 833
Sawtooth pulse 77 Signal space 689 See also Constellation diagram
Scanner 23, 60 Signal to noise ratio 432, 459–460 See also Additive
Scattering white Gaussian noise
coupling 388, 432 Signal to quantisation noise ratio 636
ionospheric 385, 388 of delta modulation 664–666
Mie 373 of A-law PCM 661
nonselective 373 of log PCM 643
parameters 359 of 𝜇-law PCM 661
plane 397–399 of uniform PCM 637
Rayleigh 20, 371–373 Signum function 74
rough surface 406 Simplex 32–33
stimulated Brillouin (SBS) 374 Sinc function 79
stimulated Raman (SRS) 374 Sine filter 778
tropospheric 388 Sine rule 96
Scintillation 389, 393 Single mode step index fibre 368
SCPC (single channel per carrier) 868, 873 Single sideband suppressed carrier amplitude
Screened twisted pair 328 modulation 510
SDH See Synchronous digital hierarchy Single-sided spectrum 228–230
SDM See Space division multiplexing Single-stub matching 357
SDSL See DSL Sinusoidal message signal See sinusoidal modulating
Secondary line constants 338 signal
Section overhead 840, 842 Sinusoidal modulating signal 480, 486, 549, 551, 558
Telegram: @Persian_IEEEComSoc

Segmentation and re-assembly sublayer 849 Sinusoidal signal 81

Selective fading Sinusoidal symbol 179, 684, 686, 689
fast fading 149, 314, 421 SI prefixes 380
frequency-selective 31, 130, 512–513, 594, 853 Skin depth 403–405
time-selective 421 Sky noise 433
Semaphore 4 Sky wave 386
Sensors 23 Slope factor 812–813
Shannon-Hartley law 780 Slope overload distortion See overload distortion
Shannon limit 783–784 Smartphone 15, 16, 27, 53, 65, 421
Shannon’s channel coding theorem 781 SMF See Submultiframe
 Index 909

Smith chart 363 Square APSK constellation 750, 751

Smoke 3 Square law demodulator 501
SNR See Signal to noise ratio Square law modulator 490, 501
SOH See Section overhead Square law noise-bandwidth trade-off 586
SONET See Synchronous optical network SSB See Single sideband suppressed carrier amplitude
Sound pressure level 3, 24 modulation
Space (binary 0) 8, 14, 18, 42, 675 SSB transmitter 512, 515
Space division multiplexing 808 Standard atmosphere 395–397
S-parameters 359–363 Standard distribution functions 128, 139
Specific attenuation 373–374, 390–394 Standing waves 345–347
Spectral leakage 285–288 Star APSK constellation 750
Spectral smearing 285–288 Stationarity 138
Spectrum STE 816
AM 484 Step size See Quantiser step size
ASK 713 Stereo 510
bandlimited white noise 445 STM See Synchronous transport module
DSB 503 Stokes wave 374
electromagnetic 380 Storage devices 28
FDM signals 810, 816, 819–821 STP See Screened twisted pair
FM & PM (narrowband) 552 Stratopause 385
FM & PM (wideband) 558–560 Stratosphere 385
FM stereo signal 511 Strict-sense stationary 138
FSK 715 Strowger, Almond 13
ISB 518 STS See Synchronous transport system
PSK 714 Sub-band 381, 671
raised cosine filter 310 Submultiframe 829
sampled bandpass signal 604 Sub-refraction 396
sampled lowpass signal 602 Superframe 826, 832
speech signal 62 Supergroup signal See FDM signals
SSB 512 Supergroup translating equipment See STE
types 224–230, 297 Superheterodyne (superhet) receiver 499, 527
VSB 521 Super high frequency 45, 136, 595
Specular reflection 373, 406–407 Supermastergroup signal See FDM signals
Telegram: @Persian_IEEEComSoc

Speech signal Superposition principle 317

characteristics 61 Super-refraction 396
coding 668–674 Switching impulse train 602
production 60 Switching modulator 489–490
Speech synthesis model 672 Symbol error rate 747
SPL See Sound pressure level Symbol rate 47, 754–757
Splicing 20, 368 Symbol slip 798
Spread spectrum modulation 31, 853 Synchronisation
Sputnik I 19 bit 825
SQNR See Signal to quantisation noise ratio carrier phase 508
 910 Index

Synchronisation (contd.) maximum power transfer 448

frame 825 Rayleigh’s energy 167, 261
Synchronous demodulation See Coherent sampling 36, 599
demodulation Thermal noise 139, 143, 432, 437
Synchronous digital hierarchy 838 Thermionic emission 26
Synchronous optical network 844 Thermosphere 385
Synchronous transport module 840 3D printer 27
Synchronous transport system 844 Timbre 62–63
System response 181 Time bandwidth product 296
Time constant 495
t Time division multiplexing
Tapped-delay-line filter 663 analogue 39
TAT See Transatlantic analogy 806
TC See Transmission convergence digital 825
TCP (Transmission control protocol) 52, 852 Time reversal 127, 130–132
TDM See Time division multiplexing Time scaling 132–134, 258
TDMA (Time division multiple access) 869 Time shifting 127, 211, 258
Telecommunication Timing content 43, 675–676
developments TMUX See Transmultiplexer
character codes 7 Townes, Charles H. 365
services 13 Trade-off
transmission media 16 bandwidth and power 782–784
international union (See ITU) in low bit rate speech coding 669
modern 5 in various modulation schemes 466–467
nonelectrical 2 Training sequence 725, 787
signals 58 Transatlantic
system design 141, 482 coaxial cable 17
verbal 2 optical fibre 20
visual 3 Transducers 58
Telegram 13–14 Transfer characteristic 127, 314, 628–630, 648
Telegraphy 5–7, 17–18, 53 Transfer function See channel transfer function
Television Transistor 489, 575, 577
CCTV 35 Transistor reactance modulator 575
Telegram: @Persian_IEEEComSoc

modulation 509, 520–522 Transition band 306, 605, 607, 770, 812
signal components 303 Transition modem 814
transmission in FDM system 823 Transmission coefficients 342–345
Telex 14 Transmission convergence 848
Telstar I 19 Transmission line theory 334
Ternary code 43, 675 Transmission medium 16, 327
Theorem Pythagoras’ 94 Transmitter See also Modulator
central limit 139 AM 491–492
Fourier 205–206 analogue TDM 39
information capacity 736 digital transmission model 699
 Index 911

FM 584 User network interface 850

FM stereo 510–511 USF See Upper side frequency
spread spectrum 855 UTP See Unscreened twisted pair
SSB 515
Transmultiplexer 814 v
Trapezoidal pulse 77 Vail, Alfred 6
Trapezoidal pulse train 248 Varactor 574–575
TRF See Tuned radio frequency Varactor diode modulator 575
Triangular pulse 77 Variable bit rate 671
Tributary unit 843–844 Variable capacitance diode 575
Tributary unit group 844 Variable reactance transistor 575
Tropopause 385 VBR See Variable bit rate
Troposphere 384–386 VC See also Virtual channel; Virtual container
Tropospheric refraction 395 VCI
Tropospheric scatter 388 See Virtual channel identifier 849
Tropospheric scintillation 393 VCO See Voltage controlled oscillator
Truncation 30, 628–634, 665 566 VCO 575–576
TU See Tributary unit VCR See Video cassette recorder
TUG See Tributary unit group Velum 60
Tuned radio frequency receiver 498 Verbal non-electrical telecommunication 2–3
TVoIP 52, 852 Very high frequency 20, 45, 366, 510
Tweeter 24 Very high-speed DSL 331
Twisted wire pair See Wire pair Very-large-scale integration circuit 17, 33, 34
Very low frequency 380
u Vestigial sideband amplitude modulation 474, 525
UBB See Unipolar baseband coherent demodulation 523
UHF See Ultra high frequency envelope demodulator 523–524
modulator 521
Ultra high frequency 380
spectrum 521
U600 mastergroup 821
VF See Voice frequency
UMG See U600 mastergroup
VHF See Very high frequency
Uncoded data 675
Video
UNI See User network interface
camera 15, 23
Unicode 11–13
Telegram: @Persian_IEEEComSoc

input device 23
Uniform quantisation 628, 634
signal 23, 35
Union bound 746
Video cassette recorder 23
Unipolar baseband system 706
Violation pulse 676
Unipolar non-return-to-zero See NRZ code
Virtual channel 850
Unit step function 74
Virtual channel identifier 849
UNRZ See NRZ code
Virtual circuit switching 52
Unscreened twisted pair 328
Virtual container 840–845
Upper side frequency 474
Virtual path identifier 849
USB See Sideband
Visual non-electrical telecommunication 3–5
User datagram protocol (UDP) 52
 912 Index

VLF See Very low frequency White noise 439–441

VLSI See Very-large-scale integration circuit Wideband audio 42, 63, 823
Vocal cords 60–61 Wideband FM 557, 567
Vocal tract coder (vocoder) 671 Wideband frequency modulator
Vocoders 671–673 direct 572–576
Voice frequency 822 indirect 570–572
VoIP 52, 852 Wideband PM 557
Voltage controlled oscillator 498, 567, 572 Wide-sense stationary (WSS) 138
Voltage standing wave ratio 346 Wireless communication 15, 18, 31, 65 See also Radio
Vowel sound 60 Wire pair 16–17, 328
VPI See Virtual path identifier applications 330
V-pulse See Violation pulse attenuation 330
VSB See Vestigial sideband amplitude modulation limiting factors 329
VSB filter 521–522 Woofer 24
VSWR See Voltage standing wave ratio Word-interleaving 825

w x
Walkie-talkie 33 X dB bandwidth 303–304
Waveform coders 671, 673 xDSL 331–332
Wavelength 18, 45, 88, 149 XNOR gate 728–729
Wavelength division multiplexing 20, 366, 813, 823
Wavenumber 338, 382–384. z
WDM See Wavelength division multiplexing Zero-level reference point (ZRP) 107–108
Wheatstone, Charles 5 Z-transform 204, 292
Telegram: @Persian_IEEEComSoc