ECE 453

Wireless Communication Systems

1st IF Filter 2nd IF Filter

Mixer Mixer
Preselector
Filter
RF Amplifier fIF 1 IF Amplifier fIF 2 IF Amplifier

fLO1 fLO2

Course Notes

Spring 2017

Steven J. Franke
Department of Electrical and Computer Engineering
University of Illinois at Urbana-Champaign

c 2017 by Steven J. Franke. All rights reserved.

Permission is hereby given to reproduce and distribute copies of these notes for nonprofit educational
purposes, provided that the notes are left intact and that the author and copyright notice are included on
each copy. Comments and suggestions for improvement are welcome and may be sent to [email protected].
2
Contents

1 Communication Signals and Systems 13

1.1 The Radio Frequency Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2 Some Characteristics of Radio Propagation . . . . . . . . . . . . . . . . . . . 14
1.2.1 ELF through VLF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2.2 LF and MF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.2.3 HF (Shortwave) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.2.4 VHF (FM Broadcast, Television Channels 2-13) . . . . . . . . . . . . 17
1.2.5 UHF (Television Channels 14-51, cellular telephones, WiFi) . . . . . . 18
1.3 Power Transfer via Free Space Propagation Links . . . . . . . . . . . . . . . . 18
1.3.1 Link example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.4 Review of Fourier Transforms and Spectra . . . . . . . . . . . . . . . . . . . . 21
1.5 Message signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.6 Linear Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.6.1 Modulation Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.6.2 Double-sideband suppressed-carrier (DSB-SC) Modulation and De-
modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.6.3 DSB with carrier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.6.3.1 DSB with full carrier . . . . . . . . . . . . . . . . . . . . . . 34
1.6.4 Power eciency for DSB signals . . . . . . . . . . . . . . . . . . . . . 37
1.6.5 Single-Sideband (SSB) . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
1.6.6 SSB Demodulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
1.6.7 Quadrature Multiplexing . . . . . . . . . . . . . . . . . . . . . . . . . 42
1.6.8 Vestigial Sideband Modulation - VSB . . . . . . . . . . . . . . . . . . 44
1.7 Angle Modulation (Nonlinear Modulation) . . . . . . . . . . . . . . . . . . . . 45
1.7.1 Spectrum of Angle-modulated Signals . . . . . . . . . . . . . . . . . . 46
1.7.1.1 Sinusoidal Modulation . . . . . . . . . . . . . . . . . . . . . . 46
1.7.1.2 Example - sinusoidally modulated signal . . . . . . . . . . . 48
1.7.2 Bandwidth of Angle-modulated Signals . . . . . . . . . . . . . . . . . 50
1.7.3 Demodulation of FM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
1.8 Quadrature Modulation/Demodulation . . . . . . . . . . . . . . . . . . . . . 54
1.8.1 Carrier Frequency and Phase Synchronization . . . . . . . . . . . . . . 55
1.9 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
1.10 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

3
4 CONTENTS

2 Receivers 67
2.1 Introduction and Historical Progression . . . . . . . . . . . . . . . . . . . . . 67
2.1.1 Tuned Detector/Demodulator . . . . . . . . . . . . . . . . . . . . . . . 67
2.1.2 Tuned Radio Frequency (TRF) Receiver . . . . . . . . . . . . . . . . . 68
2.1.3 Regenerative Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.1.4 Genesis of the Superheterodyne Receiver . . . . . . . . . . . . . . . . . 71
2.2 Characteristics of Practical Filters . . . . . . . . . . . . . . . . . . . . . . . . 71
2.2.1 Transmission-line and cavity resonator filters . . . . . . . . . . . . . . 73
2.2.2 Filters based on piezoelectric devices - Quartz-Crystal Filter, Ceramic
Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.2.3 SAW filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.2.4 Filter limitations dictate carrier-frequency conversion . . . . . . . . . 74
2.3 The Superheterodyne Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.3.1 Image Frequencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.3.2 Operation of the Mixer/LO Stage - Useful Relationships . . . . . . . . 76
2.3.3 Example - AM Broadcast Receiver. . . . . . . . . . . . . . . . . . . . . 76
2.3.4 Example - FM Broadcast Receiver . . . . . . . . . . . . . . . . . . . . 78
2.3.5 Up-conversion versus Down-conversion . . . . . . . . . . . . . . . . . . 78
2.3.6 Single- versus Double-conversion . . . . . . . . . . . . . . . . . . . . . 78
2.3.7 The 1/2-IF response . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.4 Zero-IF receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
2.5 Software Defined Radio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.7 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

3 Properties of Passive Components 95

3.1 High Frequency Characteristics of Components . . . . . . . . . . . . . . . . . 95
3.1.1 Wire Above a Ground Plane . . . . . . . . . . . . . . . . . . . . . . . 95
3.1.1.1 Resistance of Wires . . . . . . . . . . . . . . . . . . . . . . . 96
3.1.1.2 Inductance of wires . . . . . . . . . . . . . . . . . . . . . . . 97
3.1.2 Resistors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.1.3 Capacitors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.1.4 Inductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.1.4.1 Air Core Inductors . . . . . . . . . . . . . . . . . . . . . . . . 102
3.1.4.2 Toroidal Inductors . . . . . . . . . . . . . . . . . . . . . . . . 104
3.1.4.3 Equivalent circuit model for an inductor . . . . . . . . . . . . 105
3.2 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.3 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

4 RLC Networks, Resonance, and Q 115

4.1 Series RLC Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.1.1 Example - Series RLC circuit as a filter. . . . . . . . . . . . . . . . . . 118
4.2 Parallel RLC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.2.1 Unloaded vs Loaded Q of RLC circuits . . . . . . . . . . . . . . . . . . 119
4.3 More on Q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.4 Series-to-Parallel Transformations . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.4.1 Example - Series to parallel conversion . . . . . . . . . . . . . . . . . . 124
4.4.2 Example - Impedance transformation . . . . . . . . . . . . . . . . . . 124
CONTENTS 5

4.4.3 Single-resonator filters . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

4.5 Application Example - Quadrature demodulator for FM . . . . . . . . . . . . 129
4.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.7 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

5 Oscillators 145
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
5.2 Oscillator Analysis using Loop Gain . . . . . . . . . . . . . . . . . . . . . . . 147
5.3 Oscillator Analysis using Negative Resistance . . . . . . . . . . . . . . . . . . 150
5.4 Example - Common-collector Colpitts Oscillator . . . . . . . . . . . . . . . . 152
5.4.1 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5.4.2 Numerical Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
5.5 Example: Voltage Controlled Oscillator (VCO) . . . . . . . . . . . . . . . . . 163
5.6 Oscillator Phase Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
5.8 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

6 Impedance Matching Networks 183

6.1 Impedance Matching for Maximum Power Transfer . . . . . . . . . . . . . . . 183
6.1.1 Mismatch Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
6.1.1.1 Example - Mismatch Factor and Mismatch Loss . . . . . . . 185
6.1.2 Properties of Lossless Impedance Matching Networks . . . . . . . . . . 185
6.2 Impedance Matching with Lossless L-networks . . . . . . . . . . . . . . . . . 187
6.2.1 Resistive Terminations . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
6.2.2 Q of an L-network . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
6.2.3 Summary: L-network design equations . . . . . . . . . . . . . . . . . . 189
6.2.3.1 Example - Matching resistive source and load with a low-pass
L-network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
6.2.4 Matching Complex Loads with a Lossless L-network . . . . . . . . . . 190
6.2.4.1 Example - Absorption . . . . . . . . . . . . . . . . . . . . . . 191
6.2.4.2 Example - Resonance . . . . . . . . . . . . . . . . . . . . . . 193
6.3 Harmonic Attenuation in Lossless Matching Networks Using Traps . . . . . . 194
6.4 Three-element Matching Networks . . . . . . . . . . . . . . . . . . . . . . . . 196
6.4.1 Design of Pi- and T-networks for Specified Bandwidth (Q) . . . . . . . 196
6.4.2 Matching Two Resistive Terminations with Specified Attenuation and
Phase Shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
6.4.2.1 Pi-network with specified attenuation and phase shift . . . . 199
6.4.2.2 T-network with specified attenuation and phase shift . . . . 200
6.4.3 Design of Lossless Pi- and T- Matching Networks with Specified Phase
Shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
6.4.3.1 Example - Design of a lossless Pi-network with specified
phase shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
6.4.3.2 Example - Matching complex load with a specified current
phase shift. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
6.4.4 Resistive Three-element Matching Networks . . . . . . . . . . . . . . . 206
6.4.4.1 Example - Design a 10 dB Pi-type resistive attenuator . . . . 207
6.4.4.2 Minimum-loss Resistive Matching Networks . . . . . . . . . . 208
6.4.4.3 Summary of Resistive Minimum-loss Network . . . . . . . . . 209
6 CONTENTS

6.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210

6.6 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211

7 Introduction to 2-port Parameters 221

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
7.1.1 Y-parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
7.1.2 Z-parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
7.1.3 Hybrid (h) parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
7.1.4 ABCD-parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
7.1.5 2-port parameters for some common 2-port networks . . . . . . . . . . 225
7.2 Special types of 2-ports and their matrix properties . . . . . . . . . . . . . . . 225
7.2.1 Reciprocal 2-ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
7.2.2 Reciprocal lossless 2-ports . . . . . . . . . . . . . . . . . . . . . . . . . 226
7.3 Parallel, series, cascade connections of 2-port networks . . . . . . . . . . . . . 226
7.3.1 2-ports connected in parallel . . . . . . . . . . . . . . . . . . . . . . . 227
7.3.2 2-ports connected in series . . . . . . . . . . . . . . . . . . . . . . . . . 228
7.3.3 Cascaded 2-ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
7.4 Power Gain Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
7.5 Calculation of Impedance and Gain using the Impedance Matrix . . . . . . . 233
7.6 Applications of 2-port analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 235
7.6.1 Losses in L-networks for impedance matching . . . . . . . . . . . . . . 235
7.6.2 Two-winding Transformers . . . . . . . . . . . . . . . . . . . . . . . . 236
7.6.2.1 Equivalent Circuit Model for Two-winding Transformers . . 236
7.6.2.2 Impedance Transformation with the Two-winding Transformer238
7.6.2.3 Single-tuned Transformer . . . . . . . . . . . . . . . . . . . . 239
7.6.3 Two Magnetically-Coupled Resonators (Doubly-Tuned Transformer) . 241
7.6.4 Analysis of a Small-signal Series/Shunt Feedback Amplifier . . . . . . 243
7.7 Y, Z, h, ABCD relationships . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
7.7.1 Converting to Y-parameters . . . . . . . . . . . . . . . . . . . . . . . . 246
7.7.2 Converting to Z-parameters . . . . . . . . . . . . . . . . . . . . . . . . 247
7.7.3 Converting to h-parameters . . . . . . . . . . . . . . . . . . . . . . . . 247
7.7.4 Converting to ABCD-parameters . . . . . . . . . . . . . . . . . . . . . 247
7.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
7.8.1 Z parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
7.8.2 Y parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
7.9 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
7.10 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249

8 2-port Scattering (S) Parameters 255

8.1 Introduction and Definition of S-parameters . . . . . . . . . . . . . . . . . . . 255
8.2 Interpretation of S-parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 257
8.2.1 Example - Computing S-parameters for a given circuit model . . . . . 260
8.2.2 Summary of 2-port S-parameters: . . . . . . . . . . . . . . . . . . . . . 261
8.2.3 Special relationships for reciprocal, and lossless 2-ports . . . . . . . . . 261
8.2.3.1 Reciprocal 2-ports . . . . . . . . . . . . . . . . . . . . . . . . 261
8.2.3.2 Lossless 2-ports . . . . . . . . . . . . . . . . . . . . . . . . . 261
8.2.3.3 Reciprocal and lossless 2-ports . . . . . . . . . . . . . . . . . 262
8.3 Applications of Scattering Parameters . . . . . . . . . . . . . . . . . . . . . . 262
CONTENTS 7

8.3.1 Derivation of Input and Output Reflection Coecients and Voltage

and Current Gains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
8.3.2 Stability of 2-ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
8.3.3 Example - 2-port stability analysis. . . . . . . . . . . . . . . . . . . . . 269
8.3.4 Derivation of a Criterion for Unconditional Stability of 2-ports . . . . 271
8.3.5 Terminations for Simultaneous Conjugate Match . . . . . . . . . . . . 275
8.3.6 Power Gains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
8.3.7 Example - Mismatch factor in terms of S and L . . . . . . . . . . . 278
8.3.8 Example - Power transfer to an antenna through a lossy transmission
line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
8.4 Summary of useful S-parameter formulas . . . . . . . . . . . . . . . . . . . . . 280
8.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
8.6 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283

9 Filter Design 293

9.1 Butterworth, Chebyshev, Bessel-Thompson Filters . . . . . . . . . . . . . . . 295
9.1.1 Butterworth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
9.1.2 Chebyshev . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
9.1.3 Bessel-Thompson . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
9.2 Example: Synthesis of 4th order Butterworth filter . . . . . . . . . . . . . . . 303
9.2.1 Filter synthesis based on the S21 function. . . . . . . . . . . . . . . . . 305
9.2.2 Filter synthesis based on the input impedance function . . . . . . . . 307
9.2.3 Component values for lowpass prototype Butterworth filters . . . . . . 309
9.3 Example - 3rd order/0.1 dB ripple Chebyshev lowpass . . . . . . . . . . . . . 310
9.3.1 Component values for odd-order lowpass prototype Chebyshev filters
with 0.1 dB ripple . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
9.4 Frequency and Impedance scaling . . . . . . . . . . . . . . . . . . . . . . . . . 313
9.5 Bandpass Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
9.5.1 Example - Bandpass filter based on 4th order Butterworth lowpass
prototype . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
9.6 Coupled resonator filters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
9.6.1 Example - Coupled resonator filter with 2 parallel LC resonators based
on Butterworth lowpass prototype . . . . . . . . . . . . . . . . . . . . 319
9.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
9.8 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322

10 Noise in 1- and 2-ports 323

10.1 Noise Characterization of 1-ports . . . . . . . . . . . . . . . . . . . . . . . . . 323
10.1.1 Thermal Noise in Resistors . . . . . . . . . . . . . . . . . . . . . . . . 323
10.1.2 Noise Representation of Arbitrary 1-ports . . . . . . . . . . . . . . . . 327
10.1.3 Noise Representation of a Receiving Antenna . . . . . . . . . . . . . . 327
10.1.3.1 The eective antenna temperature of an inecient antenna 330
10.2 Noise Characterization of Linear 2-ports . . . . . . . . . . . . . . . . . . . . . 332
10.2.1 Eective Input Noise Temperature . . . . . . . . . . . . . . . . . . . . 332
10.2.2 Noise Factor (F) and Noise Figure (NF) . . . . . . . . . . . . . . . . . 333
10.2.3 Eective Input Temperature of a Passive Attenuator . . . . . . . . . . 335
10.2.4 Noise Temperature of Cascaded 2-ports . . . . . . . . . . . . . . . . . 336
10.2.4.1 Example - Noise temperature of cascaded amplifiers. . . . . . 337
8 CONTENTS

10.2.4.2 Example - Noise temperature of attenuator-amplifier cascade. 337

10.3 Sensitivity of a Receiving System . . . . . . . . . . . . . . . . . . . . . . . . . 339
10.3.1 Equivalent Noise Bandwidth . . . . . . . . . . . . . . . . . . . . . . . 339
10.3.2 Noise Floor, or Minimum Discernible Signal (MDS) . . . . . . . . . . 340
10.3.2.1 Example - MDS for a receiving system . . . . . . . . . . . . 341
10.3.2.2 Example - MDS for TV receiving system . . . . . . . . . . . 342
10.3.2.3 Example - TV system MDS with a lossy cable . . . . . . . . 343
10.3.2.4 Example - Calculating preamp NF required for a specified
MDS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
10.4 Measurement of Noise Temperature . . . . . . . . . . . . . . . . . . . . . . . . 345
10.4.1 Practical Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . 346
10.5 A model for the dependence of Te on ZS . . . . . . . . . . . . . . . . . . . . . 347
10.5.1 The relationship between Te and input noise voltage and current. . . . 348
10.5.2 Low frequency approximation and op-amp example . . . . . . . . . . . 351
10.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
10.7 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353

11 Mixers 357
11.1 Mixers Based on Gradual Nonlinearities . . . . . . . . . . . . . . . . . . . . . 357
11.1.1 Single-ended BJT Mixer . . . . . . . . . . . . . . . . . . . . . . . . . . 357
11.1.2 Single-ended FET Mixers . . . . . . . . . . . . . . . . . . . . . . . . . 358
11.1.3 Balanced Mixers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
11.2 Mixers Based on Switches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
11.3 Conversion Loss in Mixers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
11.4 Spurious Responses in Receivers - Spur Charts . . . . . . . . . . . . . . . . . 368
11.4.1 Crossovers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
11.4.2 Example - AM Broadcast band radio . . . . . . . . . . . . . . . . . . . 372
11.4.2.1 Radio tuned to receive a signal at 910 kHz: . . . . . . . . . . 372
11.4.2.2 Strong signal at 1000 kHz . . . . . . . . . . . . . . . . . . . . 374
11.5 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376

12 Nonlinear Eects in 2-ports 377

12.1 Power series model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
12.1.1 Specific Examples - BJT and FET nonlinearities . . . . . . . . . . . . 379
12.2 Single-tone Input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
12.2.1 Gain Compression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
12.3 Two-tone Input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
12.3.1 Desensitization and Blocking . . . . . . . . . . . . . . . . . . . . . . . 385
12.3.2 Cross modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
12.3.3 More than two tones and nonlinear terms with order higher than 3 . . 386
12.4 Quantitative Characterization of IM Distortion . . . . . . . . . . . . . . . . . 386
12.4.1 Example - Calculating IMR . . . . . . . . . . . . . . . . . . . . . . . . 388
(i)
12.4.2 Example - Calculating IIP3 (PI ). . . . . . . . . . . . . . . . . . . . . 389
12.5 Dynamic Range of a Receiving System . . . . . . . . . . . . . . . . . . . . . . 390
12.6 Intercept Point of a Cascade . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
12.6.1 The eect of adding a preamp to a receiver . . . . . . . . . . . . . . . 392
12.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
12.8 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
CONTENTS 9

13 Phase-locked Loops (PLLs) 399

13.1 PLL Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
13.1.1 PLL Transfer functions . . . . . . . . . . . . . . . . . . . . . . . . . . 400
13.1.2 Loop Gain and Notation . . . . . . . . . . . . . . . . . . . . . . . . . . 400
13.1.3 Order and Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
13.1.4 Loop Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
13.1.5 Steady-state Error Analysis . . . . . . . . . . . . . . . . . . . . . . . 404
13.2 Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
13.2.1 Examples of Stability Analysis . . . . . . . . . . . . . . . . . . . . . . 409
13.3 Transient Response of PLLs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 410
13.3.1 Summary of Second-order Loops . . . . . . . . . . . . . . . . . . . . . 414
13.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
13.4.1 Demodulation of an FM signal . . . . . . . . . . . . . . . . . . . . . . 415
13.4.2 PLL Response to AM . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
13.4.3 Carrier Recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
13.5 Frequency Synthesis with PLLs . . . . . . . . . . . . . . . . . . . . . . . . . . 418
13.5.1 Noise and Spurious Signals . . . . . . . . . . . . . . . . . . . . . . . . 419
13.5.2 Phase Detectors - Digital . . . . . . . . . . . . . . . . . . . . . . . . . 422
13.5.2.1 Exclusive-OR Phase Detector . . . . . . . . . . . . . . . . . . 422
13.5.2.2 Hold-in Range of PLL, !H . . . . . . . . . . . . . . . . . . . 423
13.5.2.3 Set-reset (SR) flip-flop . . . . . . . . . . . . . . . . . . . . . . 424
13.5.2.4 Quad-D Phase-Frequency Detector . . . . . . . . . . . . . . . 425
13.5.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
13.5.4 Pre-scalers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
13.5.5 Dual Modulus Dividers . . . . . . . . . . . . . . . . . . . . . . . . . . 430
13.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
13.7 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434

A Circuit Models for BJT and FET 435

A.1 Hybrid-pi equivalent circuit for bipolar junction transistor (BJT) . . . . . . . 435
A.2 Hybrid-pi equivalent circuit for field eect transistor (FET) . . . . . . . . . . 437
A.3 Large-signal transconductance of a BJT with sinusoidal Vbe . . . . . . . . . . 437

B Three-winding Transformer 441

B.1 Conjugate Ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
B.2 Hybrid Transformer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
B.3 Applications of the Hybrid Transformer . . . . . . . . . . . . . . . . . . . . . 446
B.3.1 Power Splitters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
B.3.1.1 180-degree splitter . . . . . . . . . . . . . . . . . . . . . . . . 446
B.3.1.2 In-phase splitter . . . . . . . . . . . . . . . . . . . . . . . . . 447
B.3.2 Sum or Dierence Combiners using a Hybrid Transformer . . . . . . . 448
B.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449

C Appendix: Useful Constants and Trigonometric Identities 451

10 CONTENTS
List of Tables

1.1 Nomenclature for Frequency Bands. . . . . . . . . . . . . . . . . . . . . . . . 14

1.2 Nomenclature for IEEE Radar Bands. . . . . . . . . . . . . . . . . . . . . . . 14
1.3 Frequency allocations for common radio services. . . . . . . . . . . . . . . . . 15
1.4 Television channel frequency assignments. . . . . . . . . . . . . . . . . . . . . 16
1.5 Zeros of Bessel functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

2.1 Generic Configurations for a Single-conversion Superheterodyne Receiver . . . 77

3.1 Conductivities for common metals. . . . . . . . . . . . . . . . . . . . . . . . . 96

5.1 Constraints for Overall Phase Angle of Alo to be Zero . . . . . . . . . . . . . 149

5.2 Parameters used for oscillator simulations. . . . . . . . . . . . . . . . . . . . 157
5.3 Comparison between predicted and simulated |Vbe | and |Ve |/|Vbe | ratio. . . . 162

9.1 Chebyshev polynomials for n=1 through 5. . . . . . . . . . . . . . . . . . . . 297

9.2 Bessel polynomials for orders n=1 through n=4. . . . . . . . . . . . . . . . . 302
9.3 Component values for lowpass prototype Butterworth filters. . . . . . . . . . 309
9.4 Component values for odd order lowpass prototype Chebyshev filters with 0.1
dB ripple and equal source and load terminations. These prototype values
produce a filter with attenuation equal to the passband ripple (-0.1 dB) at
! = 1 s 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312

10.1 Conversion between NF and eective input temperature. . . . . . . . . . . . . 335

10.2 Some typical Noise Figures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335

11.1 Conversion loss for ideal double-balanced diode-ring mixer. . . . . . . . . . . 367

13.1 Comparison of passive lag-lead and active loop filters. . . . . . . . . . . . . . 405

C.1 Some useful constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451

C.2 Some trigonometric identities. . . . . . . . . . . . . . . . . . . . . . . . . . . . 452

11
12 LIST OF TABLES
Chapter 1

Communication Signals and

Systems

This chapter provides a brief overview of the basic nomenclature used to describe various
parts of the radio frequency (RF) spectrum, followed by an overview of the dominant prop-
agation modes that operate in various frequency ranges. Then, various methods that can
be used to encode a message signal onto an RF carrier will be discussed.

1.1 The Radio Frequency Spectrum

In a communications system a message signal (e.g., analog video, analog audio, or a sequence
of analog waveforms representing digital information) is encoded onto a carrier signal, which
is transmitted and propagated to a receiver via one or more propagation modes. The process
of encoding information onto a carrier is called modulation and the inverse process is called
demodulation. In most cases the choice of carrier frequency is determined by the type of
signal coverage that is desired, e.g., for world-wide coverage from a ground-based transmit-
ter, one might choose a carrier frequency between 2 and 30 MHz (depending on the time of
day and various other factors). In this frequency range signals can be propagated over long
distances via the ionosphere sky-wave mode. On the other hand, for coverage of a partic-
ular geographic area from a satellite, the carrier frequency would need to be greater than
30 MHz (in order to penetrate the ionized regions of the upper atmosphere), and perhaps
as high as 14 GHz. Of course, other considerations are also involved in the choice of carrier
frequency. Because of the need to avoid interference between the various broadcast and
communications services, frequency allocations for dierent purposes are strictly regulated.
Within the United States the regulatory body is the Federal Communications Commission
(FCC). The FCC conforms to international agreements established by the International
Telecommunications Union (ITU). Periodically the frequency allocation scheme is revised
in accordance with the needs of the various radio services and also to accommodate new
technological developments. These revisions are agreed on by World Administrative Radio
Conferences (WARCs) under the auspices of the ITU.
The basic principles, techniques and analysis methods that will be discussed herein are
applicable over a broad range of frequencies. The terminology that is used to describe the
various frequency bands is summarized in Table 1.1.

13
14 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

Table 1.1: Nomenclature for Frequency Bands.

Freq. Range Wavelength Designation

30-3000 Hz 10 .1 megameters ELF (Extremely Low Freq.)
3-30 kHz 100 10 kilometers VLF (Very Low Freq.)
30-300 kHZ 10 1 kilometers LF (Low Freq.)
300-3000 kHz 1000 100 meters MF (Medium Freq.)
3-30 MHz 100 10 meters HF (High Freq.)
30-300 MHz 10 1 meters VHF (Very High Freq.)
300-3000 MHz 1 0.1 meters UHF (Ultra High Freq.)
3-30 GHz 10 1 centimeters SHF (Super High Freq.)
30-300 GHz 1 0.1 centimeters EHF (Extremely High Freq.)
300-3000 GHz 1 .1 millimeters THF (0.3-3.0 Terahertz)
3000-30000 GHz .1 .01 millimeters (3.0-30.0 Terahertz)

Another nomenclature that is often employed, especially when referring to radar and
satellite systems, is the so-called radar frequency band designations as defined by the IEEE
and summarized in Table 1.2. The frequency allocations for common radio services are
summarized in Table 1.3.

Table 1.2: Nomenclature for IEEE Radar Bands.

Frequency Range Designation
0.23-1 GHz P Band
1-2 GHz L Band
2-4 GHz S Band
4-8 GHz C Band
8-12.5 GHz X Band
12.5-18 GHz Ku Band
18-26.5 GHz K Band
26.5-40 GHz Ka Band
40-75 GHz V Band
75-110 GHz W Band

Television broadcasts in the United States are allowed in several frequency bands within
the range 54-698 MHz. Each assigned channel is 6.0 MHz wide. The frequency assignments
and the associated channel numbers are summarized in Table 1.4.

1.2 Some Characteristics of Radio Propagation

Below approximately 10-20 MHz the ionosphere looks like a conducting plate, or mirror, at
a height of 80-400 km (see Figure 1.1). Signals with frequencies less than approximately 20
MHz are reflected back to earth when incident on the ionosphere from below. Signals with
frequencies well above 20 MHz will pass right through the ionosphere. The upper frequency
limit where signals will begin to penetrate the ionosphere depends on a number of factors,
including the angle at which the signal is incident on the ionosphere, time of day, season,
1.2. SOME CHARACTERISTICS OF RADIO PROPAGATION 15

Table 1.3: Frequency allocations for common radio services.

AM Broadcast band: 540-1700 kHz 10 kHz chan. BW

FM broadcast band: 88.0-108.0 MHz 200 kHz chan. BW
Television See Table 1.4.
700 MHz cellular phone 698-806 MHz
800 MHz cellular phone 824-849 MHz (mobile to base)
869-894 MHz (base to mobile)
Global Positioning System (GPS) 1575.42 MHz (L1)
1227.60 MHz (L2)
1176.45 MHz (L5)
PCS band phone 1850-1910 MHz (mobile to base)
1930-1990 MHz (base to mobile)
WiFi, 802.11b/g 2400-2483.5 MHz
Bluetooth 2400-2483.5 MHz
WiFi, 802.11a 5150-5850 MHz 54 Mbps LAN

the amount of ionizing solar radiation, etc.

VHF

Ionosphere
h=400 km
HF

h=80 km
Earth
h=0 km

Figure 1.1: HF signals are refracted back toward the earth in the lower regions of the
ionosphere. Depending on the frequency of the signal, passage through the ionosphere can
result in eects ranging from virtually complete absorption of the signal to nearly lossless
refraction of the signal. When absorption is low multiple earth-ionosphere-earth hops
allow the signal to travel great distances. At HF, propagation completely around the world
via many such hops.

1.2.1 ELF through VLF

In this frequency range wavelengths are comparable to the separation between the earths
surface and the ionosphere; therefore the earth-ionosphere cavity looks like a waveguide.
Because signals are confined to the earth-ionosphere waveguide, very long propagation dis-
tances are possible with relatively small signal attenuation. Frequencies at the low end of
this range (ELF) are useful for communication with submarines, primarily due to the large
skin depth at these frequencies, and hence relatively deep penetration of these signals in
salt water.
16 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

Table 1.4: Television channel frequency assignments.

Channel Frequency Channel Frequency

Number MHz Number MHz
2 54-60 27 548-554
3 60-66 28 554-560
4 66-72 29 560-566
5 76-82 30 566-572
6 82-88 31 572-578
7 174-180 32 578-584
8 180-186 33 584-590
9 186-192 34 590-596
10 192-198 35 596-602
11 198-204 36 602-608
12 204-210 37 608-614
13 210-216 38 614-620
39 620-626
14 470-476 40 626-632
15 476-482 41 632-638
16 482-488 42 638-644
17 488-494 43 644-650
18 494-500 44 650-656
19 500-506 45 656-662
20 506-512 46 662-668
21 512-518 47 668-674
22 518-524 48 674-680
23 524-530 49 680-686
24 530-536 50 686-692
25 536-542 51 692-698
26 542-548
1.2. SOME CHARACTERISTICS OF RADIO PROPAGATION 17

1.2.2 LF and MF
A frequency in the LF range (60 kHz) is employed for the time-signal broadcasts used by
many of the clocks and wristwatches that feature automatic setting and atomic-standard
accuracy. In the continental US, these signals originate in Colorado, from radio station
WWVB. The AM broadcast band (540-1700 kHz) is in the MF range. During daytime
the ionospherically reflected MF signals are strongly attenuated by absorption in the lower
levels of the ionosphere. Therefore daytime propagation is primarily via the ground wave
which is a term used to describe the component of the signal that travels along the earths
surface. Propagation somewhat beyond the line of sight is possible in daytime because the
ground wave follows the earths curvature. At night there is little ionospheric absorption and
the sky wave can travel great distances via multiple hops between the reflecting ionosphere
and the ground.

1.2.3 HF (Shortwave)
HF is similar to MF, but absorption during daytime is much less severe. Local reception
is via ground wave and very long distance reception is via sky wave. For long distances,
multiple sky wave paths are almost always present, resulting in signal fading as the multiple
paths interfere constructively and destructively.

1.2.4 VHF (FM Broadcast, Television Channels 2-13)

Under most conditions the ionosphere has no eect on the signal, i.e. signals pass through
the ionosphere. Propagation is usually line-of-sight via a direct path and, possibly, one
or more reflected paths due to the ground and conducting objects located between the
transmitter and receiver. Occasionally, relatively small patches of high electron density
form in the E-region of the ionosphere (100-150 km in altitude) with densities suciently
high so as to cause VHF signals to be reflected from the patches. This phenomenon is called
sporadic E. Figure 1.2 shows how these patches can reflect signals in the lower VHF range
and cause long-distance reception (and possible interference).

h=100 km

Chicago New York

Earth
h=0 km

Figure 1.2: VHF signals can be reflected by thin and localized ionized layers in the E-region
of the ionosphere at approximately 100 km altitude. While the occurrence of such layers is
quite common in the summer and winter, the exact location and time of occurrence cannot
be predicted, hence the name Sporadic-E layers.
18 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

1.2.5 UHF (Television Channels 14-51, cellular telephones, WiFi)

Similar to VHF but without significant sporadic-E propagation. At both VHF and UHF
over-the-horizon propagation can occur due to refraction and tropospheric scatter. The
term tropospheric scatter refers to weak scatter from turbulence-induced irregularities in
the refractive index of the air within the troposphere. The troposphere includes the lower
10 km of the atmosphere.
Figure 1.3 illustrates refraction - slight ray bending due to altitude dependence of atmo-
spheric humidity and temperature. Figure 1.4 illustrates tropospheric-scatter (troposcat-
ter) - scattering of energy o of atmospheric turbulence. Refraction and tropo-scatter are
omni-present, but the amount of refractive bending and the strength of tropospheric scatter
are variable and they depend on meteorological conditions in the troposphere between the
transmitter and receiver sites.

Urbana Chicago

Earth

Figure 1.3: Waves propagating nearly horizontally in the troposphere tend to be refracted
such that the propagation path follows the curvature of the earth. The refraction occurs
because temperature and humidity naturally decrease with altitude, causing the refractive
index to decreases with height from a value slightly larger than 1 (e.g. n'1.0003) at the
earths surface to 1 at very high altitudes.

Urbana Chicago

Earth

Figure 1.4: Turbulence in the atmosphere causes slight variations in refractive index which
scatter radio signals in all directions.

1.3 Power Transfer via Free Space Propagation Links

Consider a free space propagation channel consisting of a transmitting antenna and a re-
ceiving antenna separated by distance R. The signal of interest has free space wavelength
denoted by . The following discussion assumes that the distance between the antennas is
large enough so that a signal radiated by either antenna can be approximated as having a
planar constant phase surface in the vicinity of the other antenna. If the largest dimension
of either antenna is denoted by D, the planar phase front approximation will be appropriate
as long as R 2D2 / .
If power Pt is delivered to the transmitting antenna, some of that power will be dissipated
as heat in the antenna (Pd ) and the rest of the power will be radiated into space (Prad ) so
1.3. POWER TRANSFER VIA FREE SPACE PROPAGATION LINKS 19

Pt = Prad + Pd . Denote the transmitting antenna power gain in the direction of the receiver
by Gt . This gain is defined to be the ratio of the actual power density at the receiver location
to the power density that would have resulted if the total power Pt = Prad + Pd had been
radiated isotropically (uniformly) in all directions. This gain is therefore the gain relative
to a hypothetical lossless, isotropic radiator and it includes the eects of power loss due to
dissipation in the antenna and any concentration or dimunition of power density due to the
coherent reinforcement or cancellation of fields radiated by dierent dierential elements
of the antenna structure. When expressed in decibels, the antenna gain is given the units
dBi to remind us that the reference is the isotropic radiator. It is important to remember
that Gt depends on direction, i.e. it is a function of two angles (, ). The incident power
density Wi at the receiving antenna can now be written as:
Pt G t
Wi = . (1.1)
4R2
Assume that the receiving antennas polarization is matched to the polarization of the
incident signal and define an eective capture area Ar for the receiving antenna so that the
power available from the receiving antennas terminals, denoted by Pavr , can be written as
the product of the incident power density and the capture area:
Pt G t
Pavr = Wi Ar = Ar . (1.2)
4R2
This is the basic communications link formula.
If the communications link is reciprocal (which will be true if the propagation medium is
free space), then the roles of the transmitter and receiver can be reversed without changing
the power transfer between the antennas. In other words, if the same total power Pt is
delivered to the receiving antenna, the power available from the transmitting antenna will
be Pavr , i.e.:
Pt G r
Pavr = At (1.3)
4R2
where Gr is the gain of the receiving antenna in the direction to the transmitter and At is
the eective capture area of the transmitting antenna. It follows that

G t Ar = G r At (1.4)

or
Gt /At = Gr /Ar (1.5)
This result was obtained without making any specific assumptions about the transmitting
and receiving antennas, so the ratio G/A must be a universal quantity, independent of
antenna details. This is a consequence of reciprocity. Generally speaking, propagation links
will be reciprocal as long as the signal does not propagate through a medium (such as the
ionosphere) that changes (e.g. rotates) the polarization of the electromagnetic fields.
Since the ratio G/A is a universal quantity, we can evaluate it once using results for any
particular antenna, and the result must hold for any other antenna. This allows us to pick
a simple antenna that allows for an easy calculation of G and A. The simple cases that are
usually considered are those of a Hertzian dipole or a large uniformly illuminated aperture.
In any case, the ratio G/A is found to be (e.g. see [4]):
G 4
= 2 (1.6)
A
20 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

so the relationship between capture area and antenna gain is

2
G
A= (1.7)
4
This relationship allows us to write the equation that governs power transfer in a radio link
in terms of gains or capture area only, i.e.

Gt G r 2
Pavr = Pt (1.8)
16 2 R2
or
At Ar
Pavr = Pt 2 R2
(1.9)

The last version is known as the Friis transmission formula [4]. Friis liked this form of
the link equation because it does not contain any numerical coecients, a feature that he
claimed would make it easier to remember.

1.3.1 Link example

The Friis transmission formula can be combined with knowledge of the operating system
temperature (defined and discussed in Chapter 10) and required SNR to predict the perfor-
mance of a communications link. The signal to noise ratio in an antenna-receiver system can
be written as the ratio of available signal and noise powers at the antenna terminals, after
referring all receiver noise to the antenna by assigning an operating system temperature
Top = TA + Te to the antenna terminals. Here, TA is the eective antenna temperature and
Te is the eective input noise temperature of the receiver:

Pavr Gt Gr 2 1
SN R = = Pt
kTop Bn 16 2 R2 kTop Bn
If the minimum SN R required for communications is denoted by SN Rmin , then the maxi-
mum distance between the antennas is:
s
Pt G t G r 2
Rmax =
SN Rmin (16 2 kTop Bn )

For example, suppose that detection and/or demodulation of a particular signal requires
an SNR of 2 (3 dB). Well assume that the receiver filters define a noise bandwidth Bn equal
to 10 kHz, that the transmitter power is (Pt ) is 1 Watt, and the receiver and transmitting
antennas are both lossless dipole antennas with length of one half wavelength, with the
dipoles oriented in space so as to maximize the received signal power. The maximum gain
of the lossless (100% ecient) half-wave dipole antenna is Gt = Gr = 1.64, or 10 log(1.64) =
2.14 dBi.
Suppose that the link operates at a frequency of 300 MHz ( = 1 m) and that the
operating system temperature Top = 1000 K (typical for an operating frequency in the VHF
part of the spectrum). The maximum distance between the the two antennas in this case
is 2.4 107 m, or 24, 000 km. This tells us that the power required to communicate over a
distance of 24,000 km is comparable to that required to power a small flashlight!
Notice that Rmax so that increasing (decreasing) the frequency will decrease (in-
crease) Rmax . At the same time, the size of both the transmitting and receiving antennas
1.4. REVIEW OF FOURIER TRANSFORMS AND SPECTRA 21

would decrease (increase) since they are assumed to be half-wavelength dipoles in this exam-
ple. It is worthwhile to think about what would happen if, instead of keeping the antenna
length fixed in terms of wavelengths, the absolute size of the transmitting and receiving an-
tennas was held constant while the frequency is varied. This is left as a thought experiment
to be carried out by the reader.

1.4 Review of Fourier Transforms and Spectra

Let f (t) be a real-valued time-signal. The Fourier transform of f (t) will be denoted by F (!)
where Z 1
F (!) = f (t)e j!t dt (1.10)
1

The time-signal can be recovered from its Fourier transform using

Z 1
1
f (t) = F (!)e+j!t d! (1.11)
2 1

The reader is assumed to be familiar with the general properties of Fourier transforms
and transform pairs, e.g., the scaling and shifting properties. Time signals of interest are
assumed to be absolutely integrable or to be periodic functions. In either case, the Fourier
transform and its inverse will exist.
The magnitude of the Fourier transform, |F (!)| is called the amplitude spectrum. For
signals with finite energy, the square of the magnitude, |F (!)|2 is called the energy spectrum
since, by Parsevals theorem, the total energy, E, in the signal is
Z 1 Z 1
1
E= |f (t)|2 dt = |F (!)|2 d!. (1.12)
1 2 1

When the time-domain signal is real, the negative frequency part of the Fourier transform
is related to the positive frequency part through

F ( !) = F (!) (1.13)

where the denotes complex conjugate. Thus if the positive part of the frequency spectrum
is known, the negative frequency part can be obtained, since the amplitude of the spectrum
is always symmetric about the frequency origin (|F ( !)| = |F (!)|) and the phase of the
spectrum is anti-symmetric (arg[F ( !)] = arg[F (!)]). For real signals, it is sometimes
convenient to plot only the positive frequency part of the amplitude or energy spectrum.

1.5 Message signals

We will call the signal that is to be transmitted over the communications channel the message
signal , denoted by m(t). In an analog communications system the message signal could be
the audio signal from a microphone. For example, Figure 1.5 (top) shows 0.8 seconds of
audio from a radio talk show. Figure 1.5 (bottom) shows the average energy spectrum
calculated using an audio sample of duration 15 seconds. This audio sample contained male
and female speakers. Notice that the energy is contained within the frequency range 200-
3500 Hz. We shall use the symbol W to refer to the bandwidth of the message signal. The
22 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

1 m(t)

0.5

-0.5

-1
0.4 0.6 0.8 1.0 1.2
t (seconds)
0 2

10 10 log10 (2 <|M(f
E
)| >
)
20
30
40
50
60
70
80
0 1000 2000 3000 4000 5000
f (Hz)

Figure 1.5: Top: message signal consisting of 0.8 seconds of audio from a radio talk show.
Bottom: normalized average energy spectrum (in dB) of a 15-second segment of audio
from the same talk show. The average energy spectrum was calculated by segmenting the
original time series, sampled at 44.1 ksamples/s, into 512 point segments. The discrete
Fourier transform (DFT) was calculated for each segment, and the squared magnitude of
each DFT was averaged.
1.5. MESSAGE SIGNALS 23

message signal bandwidth is defined such that the frequency interval [ W, W ] contains all
(or, essentially all) of the energy in the signal. For the signal shown in Figure 1.5, W is
approximately 3.5 kHz.
In a digital communication system the message signal will typically be formed from a
superposition of pulses with amplitudes chosen to represent a sequence of information bits.
In this case m(t) can be written
X
m(t) = an p(t nT ) (1.14)
n

where p(t) is a pulse function chosen to control the shape of the amplitude spectrum of
m(t), {an } represents the sequence of information bits and is a sequence of numbers chosen
from a finite set of possible values, and T 1 is the signaling rate the rate at which pulses
are transmitted. In the case of binary data, the pulse amplitudes would typically be chosen
from the antipodal set { 1, 1}, so that an = 1. In this case each information bit results
in one transmitted pulse so that 1 bit of information is transmitted each T seconds. More
generally, an could take on any of M discrete values, in which case we have an M ary
data sequence and log2 M bits are transmitted each T seconds. In general, the number of
information bits transmitted per second will be the signaling rate (T 1 ) times the number
of bits represented by each pulse that is transmitted (log2 M ).
Pulse-shapes that are commonly used to generate the message signal include the rectan-
gular pulse (Figure 1.6a):
1
1 , 2 Tt < 12
p(t) = ,
0 , elsewhere

with Fourier transform

sin f T
P (f ) = .
f T

The half-cosine pulse (Figure 1.6b):

cos t
T , 1
2 Tt < 12
p(t) = ,
0 , elsewhere

with Fourier transform

sin T ( 12 + f ) sin T ( 12 f )
P (f ) = +
T ( 12 + f ) T ( 12 f )

has a Fourier spectrum with a wider central lobe than that of the rectangular pulse, but
the sidelobes are significantly smaller than those of the rectangular pulse. The rectangular
and half-cosine pulses have infinite bandwidth, however the half-cosine pulse has a more
compact spectrum because the amplitude of the half-cosine pulses spectrum falls o much
more rapidly with increasing frequency.
A very useful family of pulses with finite bandwidth is obtained by defining the pulse such
that its Fourier transform transitions from a constant (flat) central region to zero through a
smooth transition region having a raised cosine shape. These pulses are called raised-cosine
24 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

rectangular pulse rectangular pulse spectrum

1.0 1.0

p(t) |P (f )|
0.5
0.0

0.0
5 4 3 2 1 0 1 2 3 4 5 5 4 3 2 1 0 1 2 3 4 5
(a) t/T fT

half-cosine pulse half-cosine pulse spectrum

1.0 1.0

p(t) |P (f )|
0.5
0.0

0.0
5 4 3 2 1 0 1 2 3 4 5 5 4 3 2 1 0 1 2 3 4 5
(b) t/T fT

= 0.1, raised cosine pulse = 0.1, raised cosine spectrum

1.0 1.0

p(t) |P (f )|
0.5
0.0

0.0
5 4 3 2 1 0 1 2 3 4 5 2 1 0 1 2
(c) t/T fT

= 0.5, raised cosine pulse = 0.5, raised cosine spectrum

1.0 1.0

p(t) |P (f )|
0.5
0.0

0.0
5 4 3 2 1 0 1 2 3 4 5 2 1 0 1 2
(d) t/T fT

= 0.9, raised cosine pulse = 0.9, raised cosine spectrum

1.0 1.0

p(t) |P (f )|
0.5
0.0

0.0
5 4 3 2 1 0 1 2 3 4 5 2 1 0 1 2
(e) t/T fT

Figure 1.6: Pulse shapes suitable for use in digital communications systems. The time-
domain pulse waveform is shown on the left and the magnitude of the Fourier Transform of
the pulse is shown to the right. (a) rectangular pulse, (b) half-cosine pulse, (c)-(e) pulses
with raised-cosine spectrum with excess bandwidths of (c) 10% ( = 0.1), (d) 50% ( = 0.5)
and (e) 90% ( = 0.9).
1.5. MESSAGE SIGNALS 25

pulses (Figure 1.6(c-e)). The Fourier transform of the raised-cosine pulse is

8
>
> T |f T | 12 (1 )
>
>
<
T 1 1
P (f ) = 2 {1 + cos[ (|f T | 2 (1 )]} 2 (1 ) < |f T | 12 (1 + ) , (1.15)
>
>
>
>
:
0 |f T | > 12 (1 + )

where 0 1. The dimensionless parameter controls the width of the raised-cosine

transition region. The time-domain pulse shape of the raised-cosine pulse is

sin(t/T ) cos( t/T )

p(t) = . (1.16)
t/T 1 (2 t/T )2

The bandwidth of these pulses is

1
W = (1 + ). (1.17)
2T
The dimensionless parameter is called the fractional excess bandwidth and it is often
expressed as a percentage. The minimum possible bandwidth results when = 0 which
results in W = 2T1
. In this case, the spectrum becomes rectangular and the pulse shape is
a sinc function with relatively large sidelobes that extend over many signaling intervals on
either side of the center of the pulse. For 0 < 1, the amplitude spectrum exhibits a
gradual transition to zero. As increases, the sidelobes are increasingly damped causing the
time-domain pulses sidelobes to have significant amplitudes over fewer signaling intervals.
It is important to point out that the spectrum of the chosen pulse will determine the
shape of the spectrum of the message signal m(t) and hence the bandwidth occupied by the
message signal. Let us assume that the summation in equation 1.14 is finite and consists of
N terms. The resulting m(t) has finite energy. The Fourier transform of m(t) is
N
X N
X
M (f ) = F[m(t)] = an F[p(t nT )] = P (f ) an e j2f nT
. (1.18)
n=1 n=1

The energy spectrum is

N
X
|M (f )|2 = |P (f )|2 | an e j2nf T 2
| . (1.19)
n=1

For a specific data sequence {an } the second term on the right-hand side will vary in an
irregular manner as a function of frequency. Nevertheless, the second term has a well-defined
average value. If an = 1 the average value is equal to the length of the data sequence, N .
Therefore, if the energy spectra of many message signals resulting from dierent random
data sequences of length N are averaged, the average of the second term will be close to N .
Using brackets <> to denote an average over a large number of message signals,

< |M (f )|2 >' N |P (f )|2 , (1.20)

which means that the energy spectrum of the pulse determines the average energy spectrum
of the message signal. This is illustrated in Figure 1.7 which shows a portion of a message
26 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

m(t)
2.0

1.0

0.0

1.0

2.0
0 40 80 120 160 200
t/T
2
2 <|M(f
E
)| >

3.0

2.0

1.0

0.0
0.00 0.25 0.50 0.75 1.00
fT

Figure 1.7: Top: A message signal produced using a random data sequence and raised-
cosine pulses with = 0.5. Bottom: Normalized average energy spectrum of m(t) computed
numerically using the following procedure: a message signal of length 32768T was generated
using a random sequence of bits. The long message signal was divided into short segments of
length approximately 100T . The energy spectrum of each short segment was computed, and
all spectra were averaged. The average spectrum was normalized so that the area under the
spectrum is 1.0. Note that the shape of the energy spectrum exhibits the raised-cosine shape
of the pulses. Since = 0.5, the bandwidth of the spectrum is W T = 12 (1 + 0.5) = 0.75.
1.6. LINEAR MODULATION 27

signal generated using raised-cosine pulses with = 0.5 and the average energy spectrum
calculated by dividing the long message signal into shorter segments and averaging the
energy spectra of each segment. Note that the shape of the average energy spectrum is
determined by the raised-cosine spectrum of the individual pulses.
The raised-cosine pulses are Nyquist pulses, having the property that they are zero at
non-zero multiples of the signaling interval T . Because of this property, such pulses can be
superposed as in equation 1.14 without causing any inter-pulse (or intersymbol) interference
at times corresponding to integer multiples of the signaling interval (t = nT ) because only
the pulse centered at t = nT contributes to m(t) at those times. Refer ahead to the upper
plot in Figure 1.8 to see a sample message signal generated from a binary pulse amplitude
sequence (an = 1) and raised-cosine pulses with = 0.1. The binary message can be read
o of the upper plot by taking samples at integer values of t/T ; in this case, the message
was the sequence
{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}.
In the following sections the basic types of analog modulation will be discussed. They
fall into two general categories: (1) Linear and (2) Nonlinear (or angle) modulation schemes.
Each category can be subdivided into special cases:
1. Linear modulation:
(a) DSB-SC - Double Sideband-Suppressed Carrier
(b) DSB with carrier
(c) SSB - Single Sideband
(d) VSB - Vestigial Sideband
2. Angle modulation (nonlinear modulation):
(a) FM - Frequency Modulation
(b) PM - Phase Modulation

1.6 Linear Modulation

Throughout all of our discussions we will assume that the message signal, m(t), is a real
function of time and that the time-average value of m(t) is zero (m(t) has no DC component).
The linear modulation schemes can be represented by
s(t) = A(t) cos(!c t + ) (1.21)
where s(t) is called the modulated carrier signal, A(t) = Am(t) + B, !c is the carrier
frequency, and is a constant phase angle. The magnitude of the instantaneous amplitude,
|A(t)| is called the envelope of s(t).
The various types of linear modulation come about from dierent choices for B (e.g.
B = 0 or B > max |m(t)|) or, in the case of single-sideband (SSB), from superposing two
or more linearly-modulated signals, e.g., A1 (t) cos(!c t + 1 ) + A2 (t) cos(!c t + 2 ) where
A1 (t) and A2 (t) are are obtained by passing m(t) through dierent filters. The linear
modulation schemes are linear in the sense that if m1 (t) ! s1 (t) and m2 (t) ! s2 (t), then
m1 (t) + m2 (t) ! s1 (t) + s2 (t)1 .
1 Strictly speaking, this linearity property applies only when B = 0.
28 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

1.6.1 Modulation Theorem

Before considering the various linear modulation schemes it will be useful to derive a fun-
damental relationship between the Fourier transforms of s(t) and A(t). Denote the Fourier
transforms of s(t) and A(t) by S(!) and A(!), respectively; i.e.,
Z 1
S(!) = s(t)e j!t dt (1.22)
1
Z 1
j!t
A(!) = A(t)e dt. (1.23)
1

Substituting Equation 1.21 into Equation 1.22 we obtain:

Z 1
j!t
S(!) = A(t) cos(!c t + )e dt. (1.24)
1

Using the identity cos() = 1

2 [ej + e j
], Equation 1.24 becomes
Z 1
1
S(!) = A(t) [ej(!c t+) + e j(!c t+) ]e j!t dt (1.25)
1 2
Z Z
1 j 1 j(! !c )t 1 j 1 j(!+!c )t
= e A(t)e dt + e A(t)e dt
2 1 2 1

which leads to the final result:

1 j 1 j
S(!) = e A(! !c ) + e A(! + !c ). (1.26)
2 2
This important relationship between the spectrum of A(t) and the spectrum of the modu-
lated carrier s(t) is known as the modulation theorem. It states that the spectrum of s(t)
can be obtained by superposing two copies of the spectrum of A(t) that have been displaced
by +!c and !c on the frequency axis.

1.6.2 Double-sideband suppressed-carrier (DSB-SC) Modulation and

Demodulation
DSB-SC represents the simplest possible mapping between the message signal and A(t).
Here we take A(t) = Am(t) where A is a constant, i.e.,

s(t) = Am(t) cos(!c t + ) (1.27)

This corresponds to multiplying the carrier by the message signal as illustrated in Figure
1.8, where the message signal is shown in the upper plot and the modulated carrier is shown
below. For this example, the message signal is generated from a binary pulse amplitude
sequence {an } using raised-cosine pulses with = 0.1. The binary message can be read o
of the upper plot by taking samples at integer values of t/T ; in this case, the message was
the sequence

{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}.
1.6. LINEAR MODULATION 29

2.0
1.0
m(t) 0.0
1.0
2.0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
t/T
3.0
2.0
1.0
s(t) 0.0
1.0
2.0
3.0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
t/T

Figure 1.8: Top: message signal representing 20 information bits generated using raised-
cosine pulses with = 0.1. Bottom: DSB-SC signal, s(t), normalized such that < s2 (t) >=
1.

The DSB-SC signal, s(t), shown in the lower panel has been normalized to have a mean-
square value of 1.0. Thus, if this signal represents the voltage developed across a 1 resistor,
the average power dissipated in the resistor would be 1 Watt.
Using the modulation theorem, it is a simple matter to relate the Fourier spectra of m(t)
and s(t). Suppose that the two-sided spectrum of m(t) is as shown in Figure 1.9 where
m(t) is assumed to be band-limited to W Hz. For simplicity, we plot only the magnitude
of the spectrum. Then, if the carrier frequency is larger than W (fc > W ), the spectrum

|M (f )|

|M |max

W W f

Figure 1.9: Two-sided spectrum of m(t).

of s(t) will look like Figure 1.10. If fc < W , the picture would be dierent, since the two
components of the spectrum would overlap in a region centered on f = 0. To avoid such
overlap the carrier frequency is chosen so that fc > W . It should be apparent that the DSB
signal bandwidth is 2W (Hz) and therefore takes up twice as much of the spectrum as the
original message signal.
Schematically DSB modulation can be represented as in Figure 1.11 where the modulated
30 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

|S(f )|

1
2 |M |max

fc W fc + W fc W fc + W f
fc fc

2W 2W

Figure 1.10: Two-sided spectrum of s(t).

signal is shown being fed to an antenna for transmission.

s(t) = Am(t) cos !ct

m(t) A

cos !ct

Figure 1.11: DSB modulation. The triangle represents a linear amplifier which scales the
signal by the constant gain, A.

At the receiver, DSB can be demodulated using a process called coherent demodulation
(or synchronous demodulation), which is illustrated in Figure 1.12. In this Figure, we have
accounted for the fact that propagation between a transmitter and receiver will cause the
carrier component of the signal to be phase-shifted with respect to the signal that was
actually transmitted. We have ignored the fact that the received signals amplitude will be
reduced and that the message signal will be delayed.

The synchronous demodulation process requires the receiver to have a carrier recovery
system which somehow recovers the phase-shifted carrier, cos(!c t + ), from the DSB signal.
Assuming that the carrier recovery system does its job perfectly, then it is easy to understand
how the coherent demodulation process works. The signal after the multiplier, s0 (t), is the
1.6. LINEAR MODULATION 31

s(t) = Am(t) cos(!ct + )

LPF
0
s (t)
sout(t)

Carrier
cos(!ct + )
Recovery

Figure 1.12: Coherent detection/demodulation.

product of s(t) and the recovered carrier:

s0 (t) = s(t) cos(!c t + ) (1.28)

= Am(t) cos2 (!c t + )

1
= Am(t) [1 + cos(2!c t + 2)]
2
1 1
= Am(t) + Am(t) cos(2!c t + 2)
2 2
The spectrum of s0 (t) is easily found using the modulation theorem and is sketched in
Figure 1.13.

|S 0(f )|
Lowpass filter response
1
2 A|M |max
1
4 A|M |max

W f
2fc
2W

Figure 1.13: Spectrum of s0 (t). The ideal lowpass filters cuto frequency can be anywhere
in the range W < fcutof f < 2fc W .

The purpose of the lowpass filter is to reject the term 12 Am (t) cos(2!c t + 2) which
corresponds to the part of the spectrum centered on 2fc . The output from the lowpass filter
is
1
sout (t) = A m(t) (1.29)
2
32 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

Note that for the purpose of this discussion we assumed that fc > W , so that the spectra
of the two terms in s0 (t) (after the multiplier) did not overlap. If the terms overlap, then
it is not possible to separate the undesired component from the desired one with a lowpass
filter and the demodulation scheme will not work. We also assumed that the receivers
carrier-recovery circuit was able to exactly recover the frequency and phase of the received
carrier signal. Since these quantities must be estimated from a noisy received signal, we
should consider the possibility that the carriers frequency and phase might be imperfectly
recovered, as in Figure 1.14. In this case

s(t) = Am(t) cos(!ct + )

LPF
0
s (t)
sout(t)

Carrier
cos((!c + !)t + + )
Recovery

Figure 1.14: Recovered carrier slightly o-frequency and out-of-phase.

s0 (t) = Am(t) cos(!c t + ) cos((!c + !)t + + ) (1.30)

1
= Am(t)[cos( !t + ) + cos((2!c + !)t + 2 + )]
2
After the lowpass filter
1
sout (t) =
Am(t) cos( !t + ) (1.31)
2
Comparing Equation 1.31 with Equation 1.29 we see that the eect of the frequency and
phase error is to multiply the desired signal by cos( !t + ). This is unsatisfactory and
would result in a distorted output. As an example, suppose that the message signal is a
pure tone, i.e., m(t) = cos(!m t). Then the output from a demodulator with frequency and
phase error would be
1
sout (t) = A cos(!m t) cos( !t + ) (1.32)
2
1
= A[cos((!m + !)t + ) + cos((!m !)t )]
4
i.e., the output would consist of two tones separated in frequency by twice the frequency
error. Clearly, a speech waveform that consists of the superposition of many tones would
be seriously distorted unless the frequency error is extremely small. Thus it is important
that ! = 0, i.e., it is necessary for the receivers oscillator frequency to be exactly the same
as that of the transmitters carrier oscillator. Suppose ! = 0, then:
1
sout (t) = Am(t) cos (1.33)
2
1.6. LINEAR MODULATION 33

This is acceptable if is constant and 6= /2. It is not satisfactory if varies with

time or if is close to /2. The conclusion is that it is necessary for the receiver and
transmitter oscillators to be frequency-synchronized and phase-locked to within a constant
phase oset that is less than /2. Carrier synchronization can be relatively easy to achieve
if some synchronization information is sent by the transmitter.
An example of a familiar system that employs DSB-SC and provides separate carrier
synchronization information is the FM stereo multiplex message signal where the dierence
between the left and right channel audio signals (L(t) R(t)) DSB-SC modulates a 38 kHz
subcarrier. This DSB-SC is summed with the monophonic signal (the sum of the left and
right channels) along with a 19 kHz pilot tone derived by dividing the 38 kHz subcarrier
frequency by two. The pilot tone is used by the receiver to reconstruct a 38 kHz carrier
with the proper phase for demodulating the stereo information.
The analog color television signal also includes a component provided explicitly for the
purpose of helping the receiver to achieve proper phase synchronization for coherent demod-
ulation. In this case, coherent demodulation is necessary to extract the color information
from the received video signal.
It is possible to design a carrier synchronization system that recovers the carrier from
the received signal on its own without the aid of any extra synchronization information. It
turns out that the missing carrier can be regenerated if the signal is first passed through
a nonlinear device. For example, if the signal is passed through a square-law device, then
the output will contain a component at twice the carrier frequency. This component can
be extracted and its frequency divided by two to give the necessary carrier reference. This
operation can be carried out using a phase-locked loop (PLL) in a circuit called a squaring
loop or using a so-called Costas loop. (See Chapter 13.)

1.6.3 DSB with carrier

One way to provide the carrier synchronization information that is required to demodulate a
DSB signal is to include an unmodulated carrier component in the transmitted signal. When
a carrier component is included the the modulation is called DSB w/carrier. The dierence
between DSB-SC and DSB w/carrier is simply the addition of a carrier component to the
spectrum in the latter case. The signal is represented mathematically by:

s(t) = (Am(t) + B) cos(!c t) (1.34)

The signal can be written as the superposition of a DSB-SC signal and a carrier component:

s(t) = Am(t) cos(!c t) + B cos(!c t) (1.35)

The spectrum of s(t) is shown in Figure 1.15. The bandwidth occupied by DSB w/carrier
is the same as DSB-SC and is 2W, where W is the highest frequency in the message signal,
m(t). The coherent demodulator shown in Figure 1.12 can be used to demodulate DSB
w/carrier. The presence of the carrier simplifies the implementation of the carrier recovery
module. When a carrier component is present, the output of a coherent demodulator will
contain a DC oset that is proportional to the strength of the received carrier component.
The magnitude of the DC term can be used as a relative indication of received signal
strength. It can also be used to control the gain of amplifiers preceeding the demodulator in
order to keep the amplitude of the signal at the demodulator input relatively constant. Such
a feedback system for controlling gain is called an automatic gain control (AGC) system.
34 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

|S(f )|

f
fc
2W

Figure 1.15: Spectrum of s(t) for DSB w/carrier.

1.6.3.1 DSB with full carrier

There is an important distinction between the cases where Am(t) + B is allowed to become
negative and where Am(t) + B is constrained to always stay positive. If Am(t) + B is always
> 0 the envelope of the modulated signal (defined by a line that connects positive peaks of
s(t)) reproduces the message signal, m(t). When Am(t) + B > 0 for all t the signal is called
DSB with full carrier. In this case it is possible to use a particularly simple demodulation
circuit to recover m(t) at the receiver. This circuit is known as an envelope detector and in
its simplest form it consists of a half- or full-wave rectifier followed by a lowpass filter. Figure
1.16 (top) shows the same message signal that was given in Figure 1.8 and the bottom plot
shows the corresponding DSB signal with full carrier. The signal in the bottom panel was
produced from m(t) in the upper plot by first forming the signal

m(t)
s0 (t) = ( + 1) cos !c t.
max(|m(t)|)

For this example the carrier frequency was chosen to be 8 times the signaling rate, i.e.
m(t)
!c
fc = 2 = T8 . Dividing m(t) by max(|m(t)|) ensures that the envelope ( max(|m(t)|) + 1) > 0
for all t. The signal s (t) was then normalized to have a mean-square of 1.0. The normalized
0

signal, s(t), is shown in the bottom plot of Figure 1.16.

Figure 1.17 illustrates how the envelope detector demodulates a full-carrier DSB signal.
The upper plot shows the result of passing s(t) from Figure 1.16 through a full-wave rectifier,
which produces |s(t)|. The lower plot shows the result of applying a low-pass filter to |s(t)|.
Compare the lower plot of Figure 1.17 with the original message signal, m(t), as shown
in the upper plot of Figure 1.16 to verify that, aside from a constant oset and a scaling
factor, the rectified and low-pass filtered signal is the same as m(t).2 It should be clear that
if the envelope detector is used in a situation where Am(t) + B becomes < 0 (DSB with
partial carrier), then the recovered envelope will only follow m(t) during those times where
Am(t) + B > 0. In such cases an envelope detector will recover a distorted version of m(t).
The circuit of Figure 1.18 is commonly used to approximate the operation of an ideal
rectifier/low-pass filter envelope detector. Assuming an ideal diode, the capacitor voltage
will charge to the peak of the input signal s(t) on a positive excursion of s(t), and then
decay exponentially with time constant = RC thereafter until the input voltage rises and
2 The recovered signal is also slightly delayed in time due to the group-delay of the lowpass filter.
1.6. LINEAR MODULATION 35

2.0
1.0
m(t) 0.0
1.0
2.0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
t/T
DSB, full carrier
3.0
2.0
1.0
s(t) 0.0
1.0
2.0
3.0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
t/T

Figure 1.16: Top: message signal representing 20 information bits generated using raised-
cosine pulses with = 0.1. Bottom: DSB with full-carrier signal, s(t), normalized such that
< s2 (t) >= 1. Compare the bottom plot with the corresponding plot in Figure 1.8.

full-wave rectified s(t)

2.0

1.0

0.0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
t/T
after low-pass filter
2.0

1.0

0.0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
t/T

Figure 1.17: Upper plot shows |s(t)|, which is the result of passing s(t) through a full-wave
rectifier. The lower plot was obtained by passing the signal shown in the upper plot through
a low-pass filter with cuto frequency larger than the bandwidth of m(t) (W) and smaller
than fc W . The transient at the beginning of the lower plot is the start-up transient of
the digital lowpass filter used to produce the simulated signal.
36 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

+
s(t) C R Vo ' A(1 + m(t))
-

Figure 1.18: Practical envelope detector circuit.

exceeds the capacitor voltage, at which time the capacitor voltage again follows the input
signal to its peak. The resulting output voltage is an approximation to the signal envelope,
i.e. Vo ' Am(t) + B. In order for the output to be a faithful reproduction of m(t), it is
necessary for the cut-o frequency of the lowpass RC filter to be larger than the bandwidth
of m(t) but smaller than the carrier frequency, i.e., W 2RC 1
fc W .
It is also possible to use a square-law device instead of the rectifier, i.e., the rectifier is
replaced with a square-law device as shown in Figure 1.19. Here the squaring operation is
drawn to emphasize the similarity to a coherent detector as shown in Figure 1.12. Recall
that the coherent demodulator can provide perfect recovery of the message signal. The
square-law detector can be thought of as an approximation to the coherent detector, where
the input signal is used as the local carrier reference.

LPF
0
s(t) = A(1 + m(t)) cos(!c t + ) s (t)
sout(t)

Figure 1.19: A square-law detector.

After the squarer the signal is

s0 (t) = A2 (1 + 2m(t) + m2 (t)) cos2 (!c t + ) (1.36)

A2
= (1 + 2m(t) + m2 (t)) [1 + cos(2!c t + 2)]
2
The cos 2 !c t term is rejected by the lowpass filter. What remains is

1 2
sout (t) = A (1 + 2m(t) + m2 (t)) (1.37)
2
The dc term can be removed with a coupling capacitor, leaving

1
A m(t) + m2 (t)
2
(1.38)
2

The first term is the desired output and the second term is an unwanted distortion term
which will be small compared to the first term when |m(t)| 1.
1.6. LINEAR MODULATION 37

The circuit shown in Figure 1.18 can function as an envelope detector or square-law
detector, depending on the magnitude of the input signal. If the peak input signal voltage is
relatively small (i.e. less than approx. 26 mV peak), then the diodes exponential current-
voltage characteristic can be expanded in Taylor series, in which case the leading nonlinear
term will be a square-law term. For large input signals, with peak values exceeding the diode
turn-on voltage, the circuit behaves more like an envelope detector based on a rectifier and
low-pass filter.

1.6.4 Power eciency for DSB signals

Let us assume that the DSB signal (with carrier) is applied across a 1 resistor and define
the time-average power in the signal to be the average of the instantaneous power over a
time interval that is long compared to all time scales of the message signal:
Z T /2
1
Pavg = s2 (t)dt.
T T /2

For a DSB signal

Z T /2
1
Pavg = (Am(t) + B)2 cos2 (!c t)dt.
T T /2

Using brackets (<>) to denote the averaging operation, we have

Pavg = < (Am(t) + B)2 cos2 (!c t) >

= < 12 (A2 m2 (t) + 2ABm(t) + B 2 )(1 + cos(2!c t)) >

1 2
' 2 (A < m2 (t) > +B 2 )

where we have used the fact that the long-time average value of the terms m(t), m2 (t) cos(2!c t)
and m(t) cos(2!c t) are negligibly small. The term 12 A2 < m2 (t) > is the contribution to
average power from the message-signal modulation - this power is contained in the upper
and lower sidebands of the DSB signal. The term 12 B 2 is the carriers contribution to the
average power. Define the modulation power eciency of a DSB-with-carrier signal to be the
fraction of the total transmitted power that resides in the information-bearing sidebands.

Psidebands A2 < m2 >

= 2 (1.39)
Pavg A < m2 > +B 2

In the above equations < m2 > represents the mean square value of the modulating signal,
m(t). Let us assume that m(t) varies symmetrically around zero. To ensure that Am(t) +
B > 0 a full-carrier signal will have B A max |m(t)|. The largest power eciency will
occur when B is chosen to be as small as possible so let us assume that B = A max |m(t)|.
The power eciency in this case is
<m2 >
Psidebands max |m(t)|2
<m2 >
Pavg max |m(t)|2 + 1
38 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

2
Since max |m(t)|2 1, we conclude that the power eciency of a full-carrier signal can be
<m >

no larger than 50%. For example, suppose that m(t) is a waveform consisting of contiguous
rectangular pulses with amplitudes equal to 1. This corresponds to the digital modulation
scheme known as on-o keying; when m(t) = 1, then s(t) = 0 and the transmitter is
<m2 >
o. In this case, max |m(t)|2 = 1 and the modulation eciency is 50%. Now suppose
that m(t) is something more like a typical analog message signal, i.e. suppose that m(t)
<m2 >
is a single sinusoidal tone, e.g., m(t) = cos(!m t). In this case max |m(t)|2 = 0.5 and the
modulation eciency is 33%. A realistic value for analog voice or music signals might be
<m2 >
max |m(t)|2 < 0.25, which results in modulation eciency smaller than 20%.
In summary, in DSB with full carrier more than 50% of the average power is used to
send the carrier which carries no message essential information it is sent only to simplify
the demodulation process. With realistic signals, as much as 80% of the transmitted power
might be used to send the carrier. The remaining power is split between the two information-
bearing sidebands. This seems like a significant waste of power. However, the advantage
gained by using this scheme is that the demodulator in the receiver can be a very simple
circuit consisting of as few as 3 passive components. If we attempt to make the scheme more
ecient by reducing the carrier power such that the modulation index becomes greater than
1, then it is necessary to use coherent demodulation (as described for DSB-SC) to recover
m(t) and demodulator complexity rises dramatically.

1.6.5 Single-Sideband (SSB)

As we have seen in sections 1.6.1 and 1.6.3, DSB signals (with or without carrier) occupy
a bandwidth B = 2W , twice the bandwidth of the message signal, m(t). Since all of the
information contained in m(t) is in one sideband, the other sideband is redundant. This
should be intuitively clear if we recall that the lower sideband in DSB results from translation
of the negative-frequency part of the spectrum of m(t) up into the positive frequency range.
We also know that the negative-frequency part of the spectrum of m(t) is related to the
positive-frequency part through M ( !) = M (!). Thus, all of the information is contained
in one sideband, and DSB modulation schemes do not make the most ecient use of the
radio frequency spectrum. It is more ecient to transmit only one sideband; either the
upper sideband (USB) or lower sideband (LSB). Since only one sideband is included in the
SSB signal, the bandwidth of a SSB signal is approximately W, which is the same as that of
the message signal. Thus, SSB is spectrally ecient, since the bandwidth of the transmitted
signal is the same as the bandwidth of the message signal.
Two methods for generating a SSB signal are described here.

The filter method:

The filter method is conceptually the simplest. First, the DSB signal is generated and then
the unwanted sideband is filtered out, as in Figure 1.20.
This approach places severe constraints on the SSB filter, and it is necessary for M (f )
to have a low frequency gap so that the sideband filter can reject the unwanted sideband
without significant attenuation of the desired sideband. The situation is illustrated in Figure
1.21. The two-sided spectrum of the message signal, m(t), is shown on the left, and the
one-sided spectrum of the DSB signal is shown on the right. The dotted line shows the
frequency response of a realistic upper-sideband filter. The gap in the spectrum of m(t) is
1.6. LINEAR MODULATION 39

sDSB (t) Sideband

m(t) sSSB (t)
Filter

cos !ct

Figure 1.20: Filter method for generating an SSB signal.

|M (f )|
|SDSB (f )| H(f ) for USB Filter

f f
W W fc W fc fc + W

Figure 1.21: Left: two-sided amplitude spectrum of the message signal, m(t). Right:
One-sided amplitude spectrum of sDSB (t) (solid line) and frequency response of an upper-
sideband filter (dotted line).

required, because the sideband filter will have a transition region of finite width as shown in
Figure 1.21. If the spectrum of m(t) did not have a gap, i.e. if the spectrum extended down
to DC with little or no attenuation, then the finite width of the sideband filters transition
region would result in unwanted attenuation of the desired sideband and/or leakage of the
undesired sideband past the filter.
The primary advantages of the filtering method of SSB generation are simplicity and
excellent rejection of the unwanted sideband. The disadvantages are the need for a low
frequency gap in M (f ) and the need for highly selective unwanted sideband rejection filters.
Filters that are suitable for rejecting the unwanted sideband are generally fairly expensive.

Phasing method:

A single sideband signal can be generated using a special type of filter (Hilbert transform
filter) and a quadrature multiplexer/modulator. Given a message signal m(t), define the
Hilbert Transform of m(t), denoted by m(t),
to be the signal that is obtained by delaying
each frequency component of m(t) by 1/4 of a cycle, or /2 radians. A filter that provides
this delay function is called a Hilbert Transform filter. The Fourier transform of m(t)
will
have the same magnitude as the Fourier transform of m(t), but the phase will dier by /2
(for positive frequencies) and +/2 for negative frequencies. Thus, m(t)
can be written in
terms of M (!) as follows:
1
m(t)
=F [ j sgn(!) M (!)]
40 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

m(t)
cos !ct

90 0
phase +
-
shift 90 sin !ct
s(t) = m(t) cos !c t m(t)
cos !ct
sin !ct

m(t)

Figure 1.22: Phasing method for generation of an SSB signal.

where sgn(!) is the signum function, defined by

8
< 1, ! > 0
sgn(!) = 0, ! = 0
:
1, ! < 0
The single-sideband modulator forms the modulated carrier signal
s(t) = m(t) cos !c t m(t)
sin !c t,
as illustrated in Figure 1.22. The upper sign corresponds to a lower-sideband (LSB) signal,
and the bottom sign corresponds to an upper-sideband (USB) signal. To prove this, we can
calculate the Fourier transform of s(t). It is necessary to remember the following facts:
F[cos(!c t)] = [ (! !c ) + (! + !c )]
F[sin(!c t)] = j[ (! + !c ) (! !c )]

1
F[f (t)g(t)] =
F (!) G(!)
2
where denotes convolution. Now we can calculate the Fourier transform of s(t):
S(!) = F[s(t)]

1
= 2 [M (!) [ (! !c ] + (! + !c )] jsgn(!)M (!) j[ (! + !c ) (! !c )]]

1
= 2 [M (! !c ) + M (! + !c ) [sgn(! + !c )M (! + !c ) sgn(! !c )M (! !c )]]

= M (! !c ) 12 [1 sgn(! !c )] + M (! + !c ) 12 [1 sgn(! + !c )]
Consider the upper sign for now, and examine the first term, which represents the positive-
frequency part of S(!). The term
8
1 < 1, ! < !c
1
[1 sgn(! !c )] = , ! = !c
2 : 2
0, ! > !c
1.6. LINEAR MODULATION 41

is equal to zero above the carrier frequency, and hence removes the upper sideband of
M (! !c ). If the lower sign is chosen, then the lower sideband will be multiplied by zero.
Hence, when the upper sign is chosen, the part of the spectrum above the carrier frequency
will be zeroed out, producing LSB. If the lower sign is chosen, the part of the spectrum
below the carrier frequency will be zeroed out, producing USB. Exact cancellation of the
unwanted sideband does not occur in practice, because of imperfections in the filters that
produce the /2 phase shift and because of amplitude imbalance between the two terms.

1.6.6 SSB Demodulation

Single sideband signals can be demodulated using coherent demodulation. The input signal
is either a USB or LSB signal:
s(t) = A[m(t) cos !c t m(t)
sin !c t] (1.40)
With perfect carrier recovery, the output of the coherent demodulator is
1
sout (t) = LPF[s(t) cos(!c t)] = Am(t).
2
If the recovered carrier has a phase error, i.e., if the recovered carrier is cos(!c t + ) then
it is easy to show that
1 1
sout (t) = LPF[s(t) cos(!c t + )] = Am(t) cos Am(t)
sin ,
2 2
where LP F [] represents the lowpass filter operation. The first term is the desired signal
which has amplitude 12 A cos , and the second term represents undesired crosstalk from
the Hilbert transform signal m(t).
If m(t) is an analog voice signal, the crosstalk contribution
from m(t)
would not sound significantly dierent from the desired signal, m(t). Intelligibility
of voice does not depend critically on the phase relationship between the composite frequency
components. For digital data transmission, however, the crosstalk term represents distortion
and will degrade the performance of the system. It is necessary for the phase error to
be nearly zero in order for the desired signal to have an amplitude that is much larger than
the crosstalk term. In a digital data transmission system some form of automatic carrier
synchronization would be necessary.
When the message signal is an analog voice waveform, there is an important distinction
between coherent demodulation as applied to demodulation of DSB and SSB. As we learned
earlier, when demodulating DSB it is necessary to have frequency perfect synchronization
in order to recover an intelligible signal. When demodulating an analog SSB voice signal,
however, it turns out that some frequency error can be tolerated. The eect of a frequency
error, !, is to change the pitch of the received signal. For example, suppose that the
modulating signal is a tone, i.e., m(t) = cos !m t. Then sU SB (t) = A[cos(!m t) cos(!c t)
sin(!m t) sin(!c t)]. Lets assume that at the receiver this signal is multiplied by cos[(!c +
!)t]. Then, after the multiplier we have the following:
s0 (t) = A{cos(!m t) cos(!c t) sin(!m t) sin(!c t)} cos[(!c + !)t] (1.41)

= A cos[(!c + !m )t] cos[(!c + !)t]

1
= A{cos[(2!c + !m + !)t] + cos[(!m !)t]}
2
42 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

We assume that both !m and ! are !c so that the lowpass filter will reject the first
term. The output of the filter is then
A
sout (t) = cos[(!m !)t] (1.42)
2
Note that the eect of the frequency error is to cause the demodulated tone to be shifted
in frequency by an amount equal to the frequency error. Contrast this to the DSB case
where a frequency error resulted in two tones at the output for a single input tone. In
SSB communication receivers used for analog voice reception the local oscillator can be a
free-running oscillator (often called the beat frequency oscillator, or BFO). In practice the
operator of the receiver would tune the BFO until the recovered speech signal is intelligible.
The tuning is fairly critical, but frequency errors of several 10s of Hz can be tolerated. The
audible eect is to shift the pitch of the recovered speech up or down by the frequency error,
f.
Notice that coherent detection can be used to demodulate all of the forms of linear
modulation that have been discussed so far - DSB-SC, DSB with carrier, and SSB. Only in
the latter case is it permissible for the receiver oscillator to have a frequency error, and this
applies only when the message signal is an analog voice signal. For DSB-SC and DSB with
carrier it is necessary to have perfect frequency synchronization between the transmitter
and receiver.
While SSB makes the most ecient use of the radio-frequency spectrum, this type of
modulation places relatively severe demands on the power amplifier used to amplify the
signal for transmission, because the amplifier needs to be capable of delivering peak power
that is significantly larger than the average power contained in the signal. This is illustrated
in homework problem 10. In many cases, a better scheme for obtaining the same bandwidth
eciency as SSB with more reasonable peak to average power requirements is the quadrature
multiplexing scheme discussed in the following section.

1.6.7 Quadrature Multiplexing

Quadrature multiplexing is not really a new modulation scheme, rather it is a method for
modulating two baseband signals, each having bandwidth W , onto one carrier such that the
total occupied bandwidth is 2W , i.e. the bandwidth occupied by the modulated carrier is
equal to the total bandwidth occupied by the original message signals. This doubles the
bandwidth eciency of single-channel DSB, wherein a baseband signal with bandwidth W is
associated with an occupied bandwidth of 2W . Of course, we already have encountered one
modulation scheme (SSB) which produces a modulated carrier with the same bandwidth as
the original message signal. In fact, using SSB, it is possible to transmit the upper sideband
of one signal, m1 (t), and the lower sideband of another signal, m2 (t), thereby transmitting
two signals using one carrier, and using the same total bandwidth as that occupied by the
original baseband signals. This is sometimes done and the technique is called independent
sideband or ISB modulation. Quadrature multiplexing oers a simpler way to accomplish
the same thing while avoiding the complexities (filters and phasing networks) involved with
generating the SSB signal, and that also results in a signal with more favorable peak to
average power characteristics, The idea behind quadrature multiplexing it is to use two
carriers with the same frequency but which dier in phase by 90 degrees. If the two message
signals are m1 (t) and m2 (t), then the modulated carrier signal resulting from quadrature
multiplexing is:
s(t) = m1 (t) cos !c t + m2 (t) sin !c t (1.43)
1.6. LINEAR MODULATION 43

Figure 1.23 is a schematic of a quadrature multiplexer/modulator that implements this

scheme. The quadrature-multiplexed signal simply consists of two DSB-SC components,

m1(t)
cos !ct

0
s(t) = m1(t) cos !c t + m2(t) sin !c t
90
cos !ct
sin !ct
m2(t)

Figure 1.23: Quadrature multiplexer/modulator.

and the spectra of the two components will overlap, as in Figure 1.24. Because the spectra

|S(f )|

f
fc
2W

Figure 1.24: Two DSB-SC components with overlapping spectra.

overlap it may look as if it would be impossible to recover the individual signals, m1 (t)
and m2 (t), however it can be done if the signals are coherently demodulated as shown
in Figure 1.25, which shows a quadrature de-multiplexer. This system consists of two
coherent demodulators, one (the upper branch) operates with an in-phase local oscillator,
and produces an output, I(t), that is called the in-phase signal component. The lower branch
operates with a quadrature local oscillator, and produces an output, Q(t), that is called the
quadrature signal component. Provided that the local oscillator is properly synchronized
with the carrier of the incoming signal, the in-phase output will be proportional to m1 (t) and
the quadrature output is proportional to m2 (t). The demonstration that this demodulation
scheme will work is left as an exercise.
An application of quadrature multiplexing is the analog color television signal. The color,
or chrominance, information is contained in two signals, I(t) and Q(t), and is transmitted
by quadrature multiplexing these signals onto a subcarrier with a frequency of approximately
3.58 MHz. In order to detect the color information at the receiver, it is necessary to have
44 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

I(t) = 12 m1(t)
cos !ct

s(t) = m1(t) cos !c t + m2(t) sin !c t 0

90
cos !ct
sin !ct
Q(t) = 21 m2(t)

Figure 1.25: De-multiplexing of quadrature-multiplexed signals.

synchronization information for the receiver oscillator. This information is provided in the
television signal by adding a short burst of the transmitters 3.58 MHz signal into the
composite video signal at the end of each horizontal scan line. The signal is known as
the color burst. Although with this scheme the carrier information is present for only a
small fraction of the time, it is still possible to use its information to lock an oscillator in
the receiver to the transmitters frequency and phase. This is usually done by employing a
phase-locked loop which generates a continuous carrier and uses the periodic synchronization
information to adjust the frequency and phase of the carrier.

1.6.8 Vestigial Sideband Modulation - VSB

Recall that SSB is the most bandwidth-ecient modulation, but it is dicult to get good
low-frequency response in a SSB system because of the obstacles involved in the filtering
or phasing methods of SSB generation. Low-frequency response is necessary in an analog
television system so that video images with large areas of the same color and brightness can
be transmitted. In a digital communication system, low frequency response is necessary in
order to transmit pulses with non-zero mean. The hardware requirements are considerably
relaxed if we allow a part (vestige) of the unneeded sideband to be transmitted. If, in
addition, a carrier component is transmitted, demodulation is simplified significantly. This
vestigial sideband scheme is used for all commercial television transmissions, wherein the
upper sideband and a vestige of the lower sideband are transmitted. Vestigial sideband is
used for transmission of the video portion of analog television signals using the National
Television Systems Committee (NTSC) system (see Figure 1.26a) and for transmittion of
digital television (DTV) signals using the Advanced Television Systems Committee (ATSC)
standard (Figure 1.26b). The ATSC modulation is called 8VSB.
ATSC message signals are constructed using pulses whose Fourier transform is the
square-root of the raised-cosine pulses discussed earlier. In the receiver, the signal is fil-
tered using a square-root raised-cosine response so that the received pulses will have a
raised-cosine spectrum. A pulse can take on 8 possible amplitudes, so each pulse repre-
1.7. ANGLE MODULATION (NONLINEAR MODULATION) 45

fr /2 = 5.381119 MHz
1
p1
2 Carrier

309.441 kHz 309.441 kHz

6.0 MHz

Figure 1.26: The spectrum of a television signal as transmitted for digital television (DTV)
using the ATSC (Advanced Television Systems Committee) 8VSB system.

sents 3 bits of information. The fractional excess bandwidth of the pulses is = 0.0575
and the signaling rate is fr = T 1 = 10.762238 Mpulses/sec. Each pulse represents 3 bits
of information so channel bit-rate is 3 10.762 = 32.286 Mbits/s. Redundant bits (cod-
ing) and overhead reduce the eective data rate to 19.3 Mbits/s. The signaling interval is
T = 1/10.762 106 ' 93 ns. The one-sided bandwidth of the pulses is (1 + )fr /2 = 5.6906
MHz. A DSB signal generated from a message signal constructed from these pulses would
have bandwidth (1 + )fr = 11.381 MHz which exceeds the 6 MHz width of a single TV
channel. As shown in Figure 1.26b, most of the lower sideband is removed to form the
ATSC 8VSB signal spectrum. This is accomplished with a root raised-cosine rollo that
begins at approx. 309 kHz above the carrier frequency and falls to zero 309 kHz below the
carrier frequency. Hence, the bandwidth occupied by the VSB signal is equal to the width
of the upper sideband (5.6912 MHz) plus the width of the vestige of the lower sideband
(309.4 kHz). The 8VSB signal is usually described as having an excess bandwidth of 11.5%
(although the message signal pulses have 5.75% excess bandwidth) because the vestige of
the lower sideband doubles the excess bandwidth.

1.7 Angle Modulation (Nonlinear Modulation)

We turn now the the other class of modulation schemes known as angle modulation, or
sometimes nonlinear modulation. In this case the transmitted signal has the form

s(t) = Ac cos(!c t + (t)) (1.44)

where Ac is a constant. Thus the peak amplitude of the modulated signal is constant, but
the phase angle is varied in response to the modulating signal. One important advantage
of angle-modulation systems is their inherent insensitivity to amplitude fluctuations that
may be present on the the received signal due to fading or noise. Since the angle-modulated
signal has a constant amplitude envelope, the received signal can be passed through a hard
limiter to remove amplitude fluctuations due to noise and fading; the modulation will not be
distorted by this operation. The constant amplitude feature of these signals is also desirable
because the peak to average power ratio is equal to one, allowing for ecient amplification
46 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

of such signals to high power levels. Angle modulation is nonlinear in the sense that if m1 (t)
angle modulates a carrier to produce s1 (t) (i.e. m1 ! s1 ) and m2 ! s2 , then m1 + m2 does
not result in the modulated signal s1 + s2 . As a result, the spectrum of an angle-modulated
signal is not related to the spectrum of the modulating signal in a straightforward way.
There are two basic types of angle modulation used for analog signals. These are distin-
guished by the way in which m(t) is mapped onto the angle (t).
1. Phase Modulation (PM):
The phase angle is a scaled version of m(t), i.e.,

(t) = kp m(t) (1.45)

where kp is called the phase deviation constant.

2. Frequency Modulation (FM):
The derivative of the phase angle is proportional to m(t), i.e.,

d
= kf m(t) (1.46)
dt
where kf is called the frequency deviation constant.
Note that in either case (PM or FM) the instantaneous frequency of the signal is given by

d d
!inst = (!c t + (t)) = !c + (1.47)
dt dt
In the case of FM we have

!inst = !c + kf m(t) = !c + !(t) (1.48)

Thus, frequency modulation is generated simply by varying the frequency of the carrier
signal in direct response to m(t). For a general m(t), then, the mathematical forms of
phase- and frequency-modulated signals are as given in Equations 1.49 and 1.50:

Phase modulated : s(t) = Ac cos(!c t + kp m(t)) (1.49)

Z t
Frequency modulated : s(t) = Ac cos(!c t + kf m(t0 )dt0 ) (1.50)
1

1.7.1 Spectrum of Angle-modulated Signals

For nonlinear modulation there is no simple modulation theorem that provides a rela-
tionship between M (f ) and S(f ). Well consider a simple case to gain some insight, and
then well give a general rule of thumb for estimating the bandwidth of an angle-modulated
signal.

1.7.1.1 Sinusoidal Modulation

Suppose that the message signal is a sine wave, i.e.,

m(t) = Am sin !m t (1.51)

1.7. ANGLE MODULATION (NONLINEAR MODULATION) 47

Then

sP M (t) = Ac cos(!c t + Am kp sin !m t) (1.52)

Am k f
sF M (t) = Ac cos(!c t cos !m t) (1.53)
!m
This example illustrates that except for a phase shift the PM and FM signals have the
same functional form for the sinusoidal modulation case. For the purpose of deriving the
spectrum of this type of signal it is sucient to consider a function of the form

s(t) = Ac cos(!c t + sin !m t) (1.54)

where is called the modulation index, which is simply the maximum phase deviation for
either FM or PM. The modulation index can be related to the phase, or frequency, deviation
constants through

= Am k p (PM) (1.55)

Am k f
= (FM) (1.56)
!m
Another interpretation for the modulation index comes from noting that the maximum
instantaneous frequency deviation of the phase- or frequency-modulated signal is !max =
!m , so
!max fmax
= = (1.57)
!m fm
Now the spectrum of s(t) can be derived. The Fourier transform of s(t) can be written as
follows:
Z 1
S(!) = s(t)e j!t dt (1.58)
1
Z 1
j!t
= Ac e cos(!c t + sin !m t)dt (1.59)
1

This integral can be found in an integral table or table of Fourier transforms and the result
is
1
X
S(!) = Ac Jn ( ) (! !c n!m ) + Jn ( ) (! + !c + n!m ) (1.60)
n= 1

This result shows that the signal s(t) has a spectrum which consists of a set of impulses
that are located at the discrete frequencies ! = (!c n!m ). The spectrum consists of
an infinite number of sidebands that are separated from the carrier frequency by integer
multiples of the frequency of the modulating tone, !m . The modulation index is just a
constant that describes the peak phase deviation of the signal. The function Jn (x) is called
the Bessel function of first kind of order n. These functions are tabulated, and for a given
value of n, tables of Jn (x) can be found in many mathematics reference books. The Bessel
functions of negative order (n < 0) are related to Bessel functions of positive order through

J n (x) = ( 1)n Jn (x) (1.61)

48 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

The terms Jn ( ) that appear in the expression for the spectrum can be thought of as
coecients that determine the strength of each of the impulses. Plots of the Bessel functions
for n = 0, 1, 2, 3, 4, and 5 are shown in Figure 1.27.
Taking the inverse transform of the S(!) gives an alternative form of the time-signal
that may yield some insight. Note that each pair of delta functions at frequency !n will
yield a term of the form 2 cos !n t when the inverse transform is applied. Then
Z 1 1
X
1
s(t) = S(!)ej!t d! = Ac Jn ( ) cos(!c t + n!m t) (1.62)
2 1 n= 1

Thus, the angle-modulated signal (with sinusoidal modulation) can be represented as an

infinite superposition of cosinusoidal components having frequencies |!c + n!m |.

1.

0.8

0.6

0.4

0.2

2. 4. 6. 8. 10.
-0.2

-0.4

Figure 1.27: Bessel functions of order n for n=0,1,2,3,4,5.

1.7.1.2 Example - sinusoidally modulated signal

Consider an angle-modulated signal of the form:

s(t) = Ac cos(220t + sin(22t)) (1.63)

The carrier frequency for this signal is 20 Hz and the frequency of the modulating tone
is 2 Hz. The peak phase deviation associated with the signal ( ) is 1 radian, and the
peak frequency
P1 deviation is 2 Hz. According to Equation 1.62 this signal can be written as
s(t) = Ac n= 1 Jn (1) cos(2(20 + 2n)t). The spectrum of this signal is easily determined
by inspection of Equation 1.63. Each cosinusoidal component will contribute a pair of
delta functions located at f = (20 + 2n) with amplitude given by the value of Jn (1). The
resulting line spectrum is shown in Figure 1.28. Note that for this example the amplitudes
1.7. ANGLE MODULATION (NONLINEAR MODULATION) 49

of the sidebands are smaller than that of the carrier. Figure 1.28 should be compared with
Figure 1.27. The amplitude of each line in the spectrum can be determined by inspecting
the values of Jn ( ) at = 1 shown in Figure 1.27 by the dashed line. The value of J0 (1)
is the amplitude of the carrier component; the value of J1 (1) gives the amplitude of the
first sidebands, etc. If the peak phase deviation of an angle-modulated signal happens to

0.8
0.6
0.4
0.2

10. 20. 30. 40.

-0.2
-0.4
-0.6
-0.8

Figure 1.28: Line spectrum of angle-modulated signal with 20 Hz carrier and 2 Hz modu-
lating tone.

correspond to a zero of one of the Bessel functions, then the corresponding component will
not be present in the spectrum. A few of the zeros of the Bessel functions J0 (x) and J1 (x)
are given in Table 1.5.

Table 1.5: Zeros of Bessel functions.

J0 (x) has zeros at: x= 2.4048 5.5201 8.6537 11.7915
J1 (x) has zeros at: x= 0.0000 3.8317 7.0156 10.1735

If, for example, the peak phase deviation is 5.5201 radians, the Fourier spectrum of the
(sinusoidally modulated) angle-modulated signal will not have a carrier component. This
phenomenon is useful for precise adjustment of the frequency deviation of FM transmitters.
In practice the peak frequency deviation ( fmax ) allowed for an FM signal is fixed at some
value which depends on the type of signal that is being generated. For example, a commercial
FM broadcast signal is limited to a peak frequency deviation of 75 kHz. With sinusoidal
modulation the peak phase deviation ( ) of such a signal will depend on the frequency of
the modulating tone through
fmax
= (1.64)
fm
Suppose an engineer wants to adjust the peak frequency deviation of a transmitter to some
value, fmax . A precise adjustment can easily be made by using sinusoidal modulation and
50 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

adjusting the frequency of the modulating tone to a value fm such that, when fmax is set
properly, the peak phase deviation will take on a value corresponding to a zero of J0 (x).
The spectrum of the transmitted signal is then monitored on a spectrum analyzer, and the
frequency deviation is adjusted until the carrier component vanishes. Similar adjustments
could be made by using any of the sidebands instead of the carrier component.

1.7.2 Bandwidth of Angle-modulated Signals

Figures 1.29 through 1.32 show the magnitude of the Fourier Transform of a carrier that
is angle modulated by a sinusoidal tone with = 1, 2.4, 7, and 20. The spectrum actually
consists of delta functions. The plots show the location and the relative amplitudes (absolute
value) of the delta functions. Notice that in all cases the most significant sidebands are
contained within the interval fc ( + 1)fm . Outside of this interval the amplitude of the
sidebands is relatively small and decreases rapidly with increasing separation from the carrier
frequency. So a good approximation for the bandwidth of a carrier that is angle-modulated
by a sinusoidal message signal is

BW ' 2( + 1)fm .

Since
fmax
= ,
fm
the bandwidth can be written as

BW ' 2( fmax + fm ). (1.65)

This is known as Carsons rule. It can be shown that the bandwidth given by equation 1.65
will contain at least 98% of the power associated with the angle modulated signal.
When the message signal is non-sinusoidal (as in almost all practical situations) and
has bandwidth W , a reasonable approximation to the bandwidth results if Carsons rule is
modified by replacing the tone frequency by the bandwidth of the message signal. Thus, for
a non-sinusoidal message signal

BW ' 2( fmax + W ).
When the modulating signal has finite bandwidth W can usually be taken to be the highest
frequency that is contained in the signal. If the modulating signal has infinite bandwidth
then W is chosen so that it contains most of the power in the signal.
Notice that when the peak frequency deviation is much smaller than the highest fre-
quency in m(t), i.e., if fmax W , then Carsons Rule reduces to

BW ' 2W (1.66)

When this approximation is valid, the modulation is called narrowband angle modulation,
and the bandwidth is essentially the same as for AM or DSB. On the other hand, if fmax
W , the modulation is called wideband angle modulation and Carsons rule becomes

BW ' 2 fmax (1.67)

which states that the bandwidth is approximately twice the peak frequency deviation.
1.7. ANGLE MODULATION (NONLINEAR MODULATION) 51

= 1.0
1.0
0.8
0.6
|S(f )|
0.4
0.2
0.0
15 10 5 0 5 10 15
f fc
fm

Figure 1.29: Spectrum of a carrier angle-modulated by a single tone with modulation index
= 1. Note that the frequency axis is the normalized frequency dierence from the carrier
frequency, i.e. ffmfc .

= 2.4
1.0
0.8
0.6
|S(f )|
0.4
0.2
0.0
15 10 5 0 5 10 15
f fc
fm

Figure 1.30: Spectrum of a carrier angle-modulated by a single tone with modulation index
= 2.4.
52 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

= 7.0
1.0
0.8
0.6
|S(f )|
0.4
0.2
0.0
15 10 5 0 5 10 15
f fc
fm

Figure 1.31: Spectrum of a carrier angle-modulated by a single tone with modulation index
= 7.

= 20.0
1.0
0.8
0.6
|S(f )|
0.4
0.2
0.0
25 20 15 10 5 0 5 10 15 20 25
f fc
fm

Figure 1.32: Spectrum of a carrier angle-modulated by a single tone with modulation index
= 20.
1.7. ANGLE MODULATION (NONLINEAR MODULATION) 53

1.7.3 Demodulation of FM
Most schemes for demodulating FM make use of the concept illustrated in the idealized
demodulator shown in Figure 1.33. The input signal is of the form:

s(t) s1(t) Envelope s2(t) sout(t)

d
dt Detector
DC Block

Figure 1.33: Idealized demodulator for FM signals.

Z t
s(t) = Ac cos(!c t + kf m(t0 )dt0 ) (1.68)
1

After dierentiation,
Z t
d
s1 (t) = s(t) = Ac [!c + kf m(t)] sin(!c t + kf m(t0 )dt0 ) (1.69)
dt 1

If !c + kf m(t) 0, this looks very much like the AM signal. Recall:

SAM (t) = Ac (1 + m(t)) cos(!c t + ) (1.70)

After the envelope detector, approximately,

s2 (t) = Ac (!c + kf m(t)) (1.71)

After dc blocking,
sout (t) = Ac kf m(t) (1.72)
The dierentiation operation can be approximated using a delay/dierence scheme as
shown in Figure 1.34. For the dierence approximation to be a good approximation to the

s(t) s1(t) Envelope s2(t) sout(t)

Detector
DC Block

Delay,

Figure 1.34: Approximation to the idealized FM demodulator.

derivative, the delay must be small compared to the period of the carrier oscillation, i.e.
1/fc . This approach to demodulating FM yields a very small output signal, because
the envelope of the signal after the dierentiator is proportional to !c + kf m(t). The second
54 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

term is simply the instantaneous frequency deviation, so if the maximum deviation from the
carrier frequency is fmax , the ratio of peak to minimum envelope amplitude is
fmax
1+ fc
fmax
.
1 fc

Consider FM Broadcast, where fmax = 75 kHz and fc 100 MHz. The maximum to min-
imum envelope amplitude ratio is then 1.0015. Thus, the output from the envelope detector
will be very small. For this reason, the dierentiator/envelope detector is practical only for
signals with relatively large deviations. When some distortion can be tolerated, much larger
recovered signal can be obtained by replacing the dierentiator with a single-resonator band-
pass filter whose center frequency is deliberately oset from the carrier frequency. The idea
is to place the carrier frequency on the skirts of the bandpass response, so that frequency
deviations are translated into envelope variation. By employing a filter with fairly steep
skirts (e.g. an LC filter with high Q), relatively large envelope variation can be obtained,
even for small frequency deviations. This approach to demodulating FM is called slope
detection.
An even better method for demodulating FM can be implemented by converting the
instantaneous frequency deviation to relative phase deviations, and then using a phase-
detector to convert the phase deviation to a voltage. A simple LC resonator can be used
to convert the frequency deviation of the incoming signal into (relative) phase deviation. A
circuit that accomplishes this is described in detail in Section 4.5.

1.8 Quadrature Modulation/Demodulation

In section 1.6.7 we pointed out that quadrature multiplexing can be used to send two
message signals, m1 (t) and m2 (t), on one carrier, thereby doubling the bandwidth eciency
relative to single-channel DSB modulation. This property makes the quadrature multiplexer
essentially a universal modulator, since it can be used to produce any type of modulation by
properly choosing the two message signals. Consider a carrier signal that is both amplitude
and angle modulated,
s(t) = A(t) cos(!c t + (t)). (1.73)
Straightforward application of the identity cos(a+b) = cos a cos b sin a sin b allows equation
1.73 to be written as

s(t) = A(t) cos (t) cos !c t A(t) sin (t) sin !c t

or
s(t) = m1 (t) cos !c t + m2 (t) sin !c t
with
m1 (t) = A(t) cos (t) (1.74)
m2 (t) = A(t) sin (t). (1.75)
Therefore, to create a signal that is amplitude and/or angle modulated using a quadra-
ture modulator, it is necessary to create the appropriate m1 (t) and m2 (t) signals according
to 1.74 and 1.75, and apply these signals to a quadrature multiplexer, as shown in Figure
1.23.
1.8. QUADRATURE MODULATION/DEMODULATION 55

Similarly, a quadrature de-multiplexer, combined with appropriate signal processing can

serve as a universal demodulator whether or not a modulated signal was produced using
a quadrature modulator. Refer to Figure 1.25, and note that when the input signal is an
amplitude and/or angle modulated signal, the components m1 (t) and m2 (t) are given by
equations 1.74 and 1.75. Thus, the output from the in-phase channel will be I(t) = 12 m1 (t) =
2 A(t) cos (t), and the output from the quadrature channel will be Q(t) =
1 1
2 A(t) sin (t).
The I(t) and Q(t) outputs can be provided as inputs to a signal processor, which extracts
the envelope, A(t), and/or the phase modulation, (t). The envelope is obtained by noting
that p
A(t) = 2 I(t)2 + Q(t)2 . (1.76)
The phase modulation is obtained from
1 Q(t)
(t) = tan ( ). (1.77)
I(t)
If the input signal is frequency modulated, then the demodulator must produce a signal
that is proportional to ddt . Dierentiating equation 1.77 gives the instantaneous frequency
of the signal in terms of the I and Q outputs of the quadrature demultiplexer:
d 1 dQ dI
!(t) = = 2 (I Q ). (1.78)
dt I + Q2 dt dt
If the signal envelope is constant, i.e. if the signal is purely angle modulated, then the
term I 2 + Q2 will be constant. This will be the case if the signal has been passed through a
limiter before application to the quadrature demodulator. In that case, it is only necessary to
compute the quantity I dQ dt dt , which will be proportional to the instantaneous frequency
Q dI
deviation of the input signal, where the deviation is measured with respect to the local carrier
reference.
In many applications, frequency demodulation is performed after the signal has been
digitized. A discrete-time version of equation 1.78 can be written in terms of sampled-data
sequences {In } and {Qn } by replacing the time-derivatives with first-dierence approxima-
tions, e.g. dQ
dt !
1
t (Qn Qn 1 ). The result simplifies nicely and is easily implemented in
a digital signal processor:
(In Qn 1 Qn In 1 )
!n = . (1.79)
t(In2 + Q2n )
It can be shown (see homework problem 14) that if the quadrature demodulator is to be
used only for implementing envelope or instantaneous frequency recovery (as in equations
1.76 and 1.78), then it is not necessary for the local oscillator to be synchronized with the
carrier of the incoming signal. In the case of envelope recovery, frequency or phase osets
have no eect, provided that the bandwidth of the LPFs is large enough to accomodate
any frequency shifts in the I and Q branches. In the case of FM demodulation according
to equation 1.78, frequency error in the local oscillator causes a DC oset in the output of
the demodulator. The DC oset can be used as a tuning aid, or as an error signal in an
automatic frequency control (AFC) loop, which would tune the local oscillator to minimize
the frequency error.

1.8.1 Carrier Frequency and Phase Synchronization

The I and Q outputs from a quadrature demodulator can be used to form a control signal
that corrects the instantaneous frequency of the local oscillator, for the purpose of syn-
56 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

chronizing the local oscillator to that of the incoming signals carrier. First, suppose that
the local oscillator is perfectly synchronized with the incoming carrer, and that the input
signal is a simple DSB signal so that s(t) = m(t) cos !c t. In this case, the I branch of the
quadrature demodulator output contains the desired message signal, I(t) = 12 m(t), and the
Q branch output is zero. Notice that the product of the I and Q outputs is zero in this case.
If the local oscillator is not synchronized to the incoming signals carrier, then, in general,
the local oscillator has both frequency and phase errors, i.e. the local oscillator is given by

cos((!c + !)t + ).

The quadrature local oscillator will then be

sin((!c + !)t + ).

If the input signal is DSB, then s(t) = m(t) cos(!c t). The output from the I channel is:
1
I(t) = m(t) cos( !t + ).
2
The output from the Q channel is:
1
Q(t) = m(t) sin( !t + ).
2
The product of the I and Q channel outputs is
1 2
I(t)Q(t) = m (t) sin(2 !t + 2 ).
8
Notice that the IQ product is non-zero in the presence of frequency and/or phase errors.
It turns out that if this IQ product is used to control the instantaneous frequency of the
local oscillator, the resulting closed-loop feedback system will quickly drive the IQ product
to a small value, and will reach an equilibrium where the frequency error ! = 0, and the
phase error is small. Such a feedback control system is called a Costas Loop, and is
shown in Figure 1.35, where the IQ product is formed to generate a control voltage, Vc (t).
The local oscillator must be implemented as a Voltage Controlled Oscillator, so that the
voltage Vc controls the instantaneous frequency of the oscillator.
In Figure 1.35, the feedback loop is opened by a switch in the control voltage path.
It can be shown that, if the initial frequency error is small enough, when the switch is
closed, the frequency error will be driven to zero, so that the control voltage becomes
Vc (t) = 18 m2 (t) sin(2 ). The DC component of this control signal is Vc = 18 m2 (t) sin 2 .
The loop adjusts itself so that this DC voltage is just large enough to tune the local VCO
to the carrier frequency of the incoming signal. The adjustment is achieved by allowing a
finite, static phase error, in the loop. The phase error will be just enough to provide the
DC control voltage necessary to tune the VCO to the frequency of the incoming signal. If
the VCO output frequency is a sensitive function of the control voltage, the required static
phase oset will be very small, and the associated phase error will also be small. Practical
implementations of the Costas Loop usually include a lowpass filter between Vc and the VCO
to smooth the control voltage and control the dynamic response of the loop (the response
to transient changes in the incoming signals phase).
1.9. REFERENCES 57

I(t)

cos((!c + !t) + )
s(t) = m(t) cos !c t 0
90 Vc(t)

Q(t)

Figure 1.35: Costas Loop for carrier synchronization. The IQ product is used to control the
instantaneous frequency of the local oscillator. When the switch is closed, the loop drives
the IQ product to a small value.

1.9 References
1. Jordan, Edward C. and Keith G. Balmain, Electromagnetic Waves and Radiating
Systems, Prentice Hall, 1968.
2. Proakis, J. and M. Salehi, Communication Systems Engineering, 2nd ed., Prentice-
Hall, 2002.
3. Haykin, Simon, Communication Systems, 3rd ed., John Wiley and Sons, Inc., 1994.
4. Friis, Harald T., A Note on a Simple Transmission Formula, Proc. IRE, pp. 254-256,
May, 1946.
58 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

1.10 Homework Problems

1. One signal applied to the input of an ideal multiplier is

s1 (t) = cos(1000 t) + cos(1800 t) (1.80)

The signal applied to the other input is

s2 (t) = cos(7000 t) + cos(20, 000 t) (1.81)

The output from the multiplier is applied to an ideal filter that passes only frequencies
between 3.5 kHz and 10.0 kHz. List all frequencies that are present at the output of
the filter.
2. An AM signal with full carrier can be represented as follows:

s(t) = A(1 + m(t)) cos(!c t), |m(t)| < 1 for all t.

An envelope detector can be used to demodulate full-carrier AM. An ideal envelope

detector is realized with a rectifier and a lowpass filter, as shown in Figure 1.36. The

Full-wave s0 (t)
s(t) sout(t)
Rectifier
B

Figure 1.36: An ideal envelope detector, consisting of ideal full-wave rectifier and lowpass
filter.

output of the full-wave rectifier can be written as s0 (t) = |s(t)| = s(t)p(t) where p(t)
is a square wave with the following properties:

p(t) is an even function,

max{p(t)} = 1, min{p(t)} = 1,
p(t) has period T = 2/!c .
(a) Express p(t) in a Fourier series.
(b) Sketch the amplitude spectrum of s0 (t). Assume that m(t) has the amplitude
spectrum shown in Figure 1.37. The bandwidth of m(t) is assumed to be

|M (f )|

W W f

Figure 1.37: Amplitude spectrum of m(t).

small compared to the carrier frequency, i.e. W fc .

1.10. HOMEWORK PROBLEMS 59

(c) Specify the minimum and maximum possible cuto frequency, B, for the
low-pass filter that will allow m(t) to be recovered at the output of the filter.

3. The message signal m(t) = A sin (!m t) modulates a carrier signal given by cos(!c t)
where we assume !m !c .

(a) If the modulation is lower sideband (LSB), show that the modulated signal can
be represented by

A
f (t) =
[sin(!m t) cos(!c t) cos(!m t) sin(!c t)] (1.82)
2
(Hint: Write down an expression for the DSB signal and manipulate it so that
you have two terms one for the upper sideband and one for the lower sideband.
Then subtract the upper sideband component.)
(b) Suppose that we attempt to demodulate this signal using a square-law detector.
The output of a square-law detector is the square of the input signal. Assume that
the detector is followed by a low-pass filter that removes all frequency components
with frequencies larger than !c . Find the output of the square-law detector/low-
pass filter. Can the modulating signal (A sin(!m t)) be recovered?
(c) Consider the same problem as in 3b, except assume that the carrier is first rein-
serted, i.e., we add a term K cos(!c t) to f(t) to give:

A
f 0 (t) = [sin(!m t) cos(!c t) cos(!m t) sin(!c t)] + K cos(!c t) (1.83)
2
Assume that this signal is applied to the square-law detector/low-pass filter. Can
the modulating signal be recovered? How will the output of the demodulator
depend on the amplitude of the reinjected carrier? Note: In practice the oscillator
that provides the reinjected carrier for SSB demodulation is called the beat-
frequency oscillator (BFO).

4. Find the instantaneous frequency as a function of time for the following signal:

s(t) = 10 cos[2(108 t 104 t2 ) 4]

5. Consider an angle-modulated signal of the form

s(t) = A cos(!c t + 5 cos !m t + 15 cos 3 !m t) (1.84)

where !c = (2) 107 s 1

, and !m = (4) 103 s 1
.

(a) What is the peak frequency deviation of this angle-modulated signal? Give your
result in kHz. Be careful, you need to find the absolute maximum value of a
function with several sub-maxima.
(b) Estimate the bandwidth of this signal using Carsons rule. You will need to
determine the bandwidth of the baseband signal (W) for this calculation. Choose
the smallest bandwidth that contains all of the power in the baseband signal. Give
your result in kHz.
60 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

(c) Suppose that the given s(t) is the result of frequency modulation by a baseband
signal denoted by m(t). Specify the function m(t). (Note: You may assume that
the frequency deviation constant, kf , is equal to 1.0.)

6. Consider a signal m(t) with Fourier transform M(!) where

Z 1 Z 1
1
M (!) = m(t)e j!t dt and m(t) = M (!)ej!t dw (1.85)
1 2 1

(a) Consider a particular M(!) which is sketched in Figure 1.38. You may assume
that M(!) is a real function for this problem. Sketch the Fourier transform

2 1 1 2

Figure 1.38: M(!)

(spectrum) of the DSB signals

i. m(t) cos(t)
ii. m(t) cos(6t)
(b) Suppose we attempt to recover m(t) from the DSB signal using coherent demod-
ulation. Can m(t) be recovered in both cases? Why or why not?

7. An angle-modulated signal can be represented by

s(t) = A cos[!c t + (t)] (1.86)

Consider the instantaneous phase deviation (t) of an angle-modulated signal as shown

in Figure 1.39.

(t)

T/2

Figure 1.39:
1.10. HOMEWORK PROBLEMS 61

(a) Suppose that this signal is the result of frequency-modulating the carrier signal.
Sketch the message signal m(t). You may assume that kf = 1.
(b) Suppose that T = 1 ms. What is the peak value of the instantaneous frequency
deviation of this signal?
(c) Estimate the bandwidth of the angle-modulated signal using Carsons Rule. Ex-
plain any assumptions or approximations that you make.

8. An FM transmitter has carrier frequency fc = 100 MHz. The peak frequency deviation
of the transmitter is unknown, but FCC regulations state that the deviation must be
no greater than 75 kHz. To measure the actual deviation of the transmitter, the
engineer modulates the carrier with a single tone (m(t) = cos !m t). The frequency of
the tone is constrained to be in the range 20 Hz to 20 kHz to stay within the audio
bandwidth of the modulator. The frequency of the tone is adjusted while the spectrum
of the modulated signal is monitored. When the tone frequency is decreased from 20
kHz, it is found that the carrier component of the spectrum vanishes at fm = 10.40
kHz, the first sidebands vanish at fm = 6.52 kHz, the carrier vanishes at fm = 4.53
kHz, the first sidebands vanish at fm = 3.56 kHz, etc.

(a) What is the actual peak deviation of the transmitter? Express your result in
kHz.
(b) Next, the engineer sets the frequency of the modulating tone to fm 0
, where 12.5
kHz < fm < 20 kHz, and increases the frequency deviation of the transmitter. As
0

the deviation is increased (from the original setting found in 8a), the amplitude
of the carrier component of the spectrum is observed to go to zero twice. When
the amplitude of the carrier component reaches zero for the second time, the
engineer knows that the deviation has been set to 75 kHz. What is fm 0
in kHz?
(c) If the engineer had stopped at the first zero, what would the deviation be?
(Hint: See Table 1.5 for a list of the first few zeros of J0 (x) and J1 (x).)

9. A tone (cos 2fm t) frequency modulates a carrier. The tone frequency, fm , is varied
while the frequency deviation, fmax is held constant. Consider the Fourier amplitude
spectrum of the modulated signal. Suppose the carrier component of the spectrum
(i.e., the delta function at the carrier frequency) is observed to vanish when the tone
frequency, fm , is equal to 2.350 kHz and at 3.684 kHz.

(a) What is the peak frequency deviation of the signal? Express your result in kHz.
(b) Give two more tone frequencies that would make the carrier component vanish.
The zeros of the first two Bessel functions are given below:
J0 (x) has zeros at x = 2.4048, 5.5201, 8.6537, 11.7915
J1 (x) has zeros at x = 0.000, 3.8317, 7.0156, 10.1735

10. When designing or specifying a transmitter it is necessary to consider the peak enve-
lope power (PEP) as well as the average power of signals that are to be delivered to
the antenna for transmission. The envelope power of a modulated carrier signal s(t)
is a time-varying quantity defined to be the average of the instantaneous power over
one cycle of the RF carrier, i.e. the envelope power (assuming 1 impedance) is:
Z
1 t+c /2 0 2 0 2
Penv (t) = s(t ) dt , c =
c t c /2 !c
62 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

It is usually a good approximation to assume that the envelope of the signal is eec-
tively constant over the duration of a carrier cycle, so that the envelope function may
be pulled out of the integral. The peak envelope power (PEP) is the peak value of
Penv (t).
The average power is defined as the average of the instantaneous power over a time
interval that is long compared to all time scales of the message signal:
Z
1 T /2
Pavg = lim s(t0 )2 dt0
T !1 T T /2

Consider the message signal given below:

N
X
4 sin(n!m t)
m(t) =
n=1,3,5...
n

This series is the truncated Fourier series for a square wave.

(a) Write down the series expression for the Hilbert transform, m(t).
b The Hilbert
transform of m(t) (m(t))
b is obtained by delaying every frequency component of
m(t) by 90 . Note, if m(t) = sin(!t) then m(t)
= cos(!t).
(b) Plot the waveforms m(t) and m(t)
for N=1,5,9,11,15.
(c) If the band-limited square wave is transmitted using SSB, the p envelope (ampli-
tude) of the resulting modulated carrier signal will be A(t) = m(t)2 + m(t) 2.
The envelope power of the modulated carrier signal will be proportional to A2 (t).
For each N, plot A2 (t) and estimate the peak to average ratio numerically.

Note: The purpose of this problem is to show you why we would not con-
sider using single-sideband (SSB) for a digital communication system that uses
pulse amplitude modulation with square pulses, or any pulses with very sharp
edges. You should find that m(t) exhibits large values wherever m(t) has a
large derivative. The peak values of m(t)
increase as the derivatives of m(t) in-
crease (with increasing N, in this problem). Since the transmitted SSB signal
is m(t) cos !c t m(t)
sin !c t, the large amplitude excursions exhibited by m(t)
require the transmitter to be able to produce large instantaneous output powers.
Thus, a transmitter designed for 10 kW output power might have to be capable of
delivering 50 kW for short periods of time. Notice that the peak-to-average power
ratio is 1.0 for a straight DSB signal, m(t) cos !c t, with a message signal consist-
ing of square pulses. Thus, by going to SSB, we reduce the occupied bandwidth
by a factor of two - but the price is increased peak-to-average power ratio. Since
the bandlimited pulses (small N) do not have sharp edges, the Hilbert transform
does not exhibit large peaks. Thus a shaped pulse with rounded edges (such as
a raised-cosine pulse) does not exhibit the large amplitude excursions, and hence
the peak-to-average power ratio is not very large. Digital television uses vestigial
sideband (almost SSB) with raised-cosine
P pulses and multi-level pulse-amplitude
modulation (PAM), i.e. m(t) = n an p(t nT ) where an can take on any of
4 discrete positive or negative values. The HDTV signal has a peak to average
power ratio of about 4, or 6 dB.

11. Estimate the bandwidth of these signals:

1.10. HOMEWORK PROBLEMS 63

(a) s(t) = 10 cos[2108 t 2 cos(2104 t + /4)]

(b) s(t) = 50 cos[2108 t] cos[2104 t + /4]

12. Consider the following angle modulated signal:

s(t) = 50 cos(!c t + 5 sin(21000t))

(a) Determine the peak phase deviation in radians.

(b) Determine the instantaneous frequency deviation in Hz.
(c) Determine the peak frequency deviation in Hz.
(d) Use Carsons rule to calculate the bandwidth of s(t). Express your result in kHz.
(e) Suppose s(t) is passed through a bandpass filter that passes all frequencies within
3.5kHz of the carrier frequency fc without attenuation and rejects all other fre-
quency components (i.e. the bandpass filter has a rectangular transfer function
that is centered on fc with a bandwidth of 7 kHz). Express the power at the
output of the filter as a percentage of the power in s(t). You may want to use
Mathematica, Matlab, or similar program to obtain the numerical values of Jn ( )
that you will need for this part.

13. A tone m(t) = cos(2fm t) frequency modulates a carrier. The tone frequency, fm , is
initially set to 400 Hz and is slowly increased. As the tone frequency is increased the
frequency spectrum of the modulated signal is monitored. The sidebands at fc fm
are observed to vanish when fm is equal to 491.5 Hz and again at 712.7 Hz. What is
the peak frequency deviation of the signal?

14. The signal at the input to a quadrature demodulator is

Z t
s(t) = A cos(!c t + kf m(t0 )dt0 ).
1

The quadrature demodulator produces two output signals, I(t) and Q(t), by multi-
plying s(t) with in-phase and quadrature local oscillator signals, respectively. Each
multiplier is followed by a low-pass filter that rejects double-frequency terms. A signal
processor takes the outputs of the quadrature demodulator and produces the output
signal sout (t) = I(t) dQ(t)
dt Q(t) dI(t)
dt . In general, the local oscillator signals employed
within the quadrature demodulator have frequency and phase error, i.e. the in-phase
and quadrature local oscillator signals are cos((!c + !)t+ ) and sin((!c + !)t+ ),
respectively.

(a) If ! = 0 and = 0, show how sout (t) depends on the message signal m(t); i.e.,
derive an expression for sout (t) that is in terms of m(t).
(b) Now allow the frequency and phase errors to be finite. How do frequency and
phase errors aect sout (t)?

15. Two methods for generating SSB (and VSB) signals were introduced in this chapter
the filter method and the phasing method. There is a third method, called the
Weaver method which is illustrated in the Figure. The bandwidth of the message
64 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS

H(!)

a 1 b c
!
W W
cos(!1 t) cos(!2 t)
cos(!1 t) cos(!2 t)
0 0
m(t) + s(t)
90 90
H(!)
sin(!1 t) sin(!2 t)
1
!
a0 b0 c0
W W

signal is assumed to be W Hz (2W s 1

). In the frequency domain, the signal at
point a can be written as
1
Sa (!) = [M (! !1 ) + M (! + !1 )].
2
At point b, the signal becomes
1
Sb (!) = Sa (!)H(!) = [M (! !1 ) + M (! + !1 )]H(!).
2
And at point c
1
Sc (!) = 2 Sb (!) [ (! !2 ) + (! + !2 )]

1
= 4 [M (! !2 !1 ) + M (! !2 + !1 )]H(! !2 )+

1
4 [M (! + !2 !1 ) + M (! + !2 + !1 )]H(! + !2 ).

(a) Find the corresponding expressions for the frequency-domain signals at points a,
b, and c.
(b) If the upper (+) sign is chosen for the combiner, determine the frequency-domain
signal at the output of the combiner. Simplify as much as possible - some terms
should cancel. If !2 !1 , specify !1 in terms of W if the output is to be a
SSB signal. Determine whether the modulator yields an upper sideband (USB)
or lower sideband (LSB) signal. Hint: the term H(! !2 ) is 1 for !2 W
! !2 + W and is zero elsewhere.
(c) Now assume that the lower (-) sign is chosen. Determine !1 such that the mod-
ulator yields a SSB signal. Is the output USB or LSB?
(d) If the message signal has bandwidth W = 5.0 MHz, and you want to generate
a USB signal with carrier frequency at 1250 MHz and upper cuto frequency at
1255 MHz, specify the correct sign (+ or -) to use in the combiner and the values
of f1 = !1 /(2) and f2 = !2 /(2). Give your results in MHz.
1.10. HOMEWORK PROBLEMS 65

(e) For the same message signal as in part (d), specify the sign to use in the combiner
and f1 and f2 if it is desired to generate LSB with the carrier frequency at 1250
MHz and the lower cuto frequency at 1245 MHz.
P
16. DSB-SC is used to transmit a message signal, m(t) = an p(t nT ), where T = 1 s
and p(t) is a raised-cosine pulse with = 0.5. Calculate the bandwidth occupied by
the transmitted signal.
17. Consider the following signal:

s(t) = 10 sin[2(109 )t + 2 cos(3105 t)].

(a) Find the instantaneous frequency (in Hz) as a function of time for this signal.
(b) What is the peak instantaneous frequency deviation of s(t)? Express your result
in kHz.
(c) Estimate the bandwidth of s(t). Express your result in kHz.
(d) The fractional bandwidth of s(t) is defined to be the signal bandwidth divided
by the carrier frequency. What is the fractional bandwidth of s(t)?
(e) Suppose that s(t) is the result of frequency modulation by a message signal
denoted by m(t). Specify the function m(t). You may assume that the frequency
deviation constant, kf , is equal to 1.0.
(f) In general, the Fourier spectrum of s(t) will contain delta functions with compo-
nents at the carrier frequency, fc , and at frequencies separated from the carrier
frequency by multiples of some frequency fm . What is fm ? Give your answer in
kHz.
(g) Refer to Figure 1.27 to answer this question. In the Fourier amplitude spectrum
of s(t), (|S(f )|), denote the strength of the delta function at fc by a0 , and the
strength of the delta functions at fc nfm by an . Sort the list {a0 , a1 , a2 , a3 , a4 }
in order of decreasing amplitude.
66 CHAPTER 1. COMMUNICATION SIGNALS AND SYSTEMS
Chapter 2

Receivers

2.1 Introduction and Historical Progression

The history of radio communication begins at the end of the 19th century, when Italian
inventor Guglielmo Marconi developed the first practical radio communication system using
spark-gap radio transmitters and a nonlinear circuit element, called a coherer, as a detec-
tor at the receiving end of the link. His most famous accomplishment was demonstration
of transatlantic wireless communications, in 1902, when the Morse-code letter S (dot dot
dot) was transmitted from England, and received in Newfoundland. Wireless telegraphy
transmission using spark-gap transmitters used the broadband damped oscillations gener-
ated when the DC current in an LC circuit is suddenly interrupted. Spark-gap telegraphy
was the primary wireless communications technology in use throughout World War I. Mean-
while, beginning in 1906, the technology required for the next phase of radio communication
development was put into place when Lee De Forest patented the audion, a device created
by inserting a third electrode into the vacuum diode, creating a triode vacuum tube. This
3-terminal device could be used to amplify signals, and could be made to oscillate, thus
serving as a very stable source of continuous-wave (CW) high-frequency signals. By 1915,
the American Telegraph and Telephone Company (AT&T) had developed a high-power
amplitude-modulated (AM) radio transmitter based on the triode, and in that year, for the
first time in history, voice transmissions from Arlingon, VA were heard in Paris, France,
and in Honolulu, HI. In 1920, the first commercial AM radio stations began operating. For
much more information on the fascinating early history of radio communication, the book
The Science of Radio, by Paul J. Nahin, is highly recommended.
Our purpose in this chapter is, first, to give a brief outline of the historical development
of radio receivers, beginning with the simple, passive, tuned detector, and second, to present
a fairly detailed discussion of the modern superheterodyne receiver.

2.1.1 Tuned Detector/Demodulator

The purpose of a receiver is to select, amplify, and demodulate a modulated-carrier signal,
while ignoring the many other signals which will be picked up by the antenna, but are
not of interest. The earliest receivers employed a single tuned circuit as a filter in front of
an envelope detector implemented with a crystal detector, a point-contact type of diode.
Voice transmissions were amplitude modulated with full-carrier, so an envelope detector was

67
68 CHAPTER 2. RECEIVERS

sucient to demodulate the signal. The block diagram of such a tuned-detector receiver is
shown in Figure 2.1. The tuned-detector is a passive circuit, and the audio power available

si (t) = A(1 + m(t)) cos(!ct + )+noise

Demodulator so (t) = A(1 + m(t))+noise

Figure 2.1: Top: block diagram of a tuned-detector receiver. Bottom: typical implementa-
tion. This type of receiver is also known as a crystal set, because early implementations
of the diode junction were implemented with the point-contact junction between a wire and
crystal such as galena (lead sulfide) or carborundum (silicon carbide).

from the detector output must be provided by the energy collected by the antenna. Large
antennas are necessary to collect enough signal, even from relatively close transmitters, to
provide an output that is audible even with sensitive headphones. Selection of the desired
signal, and rejection of unwanted signals, is the responsibility of the resonant circuit, which
acts as a filter, and which must be tuned to pass the desired carrier frequency of interest.
It is easy to improve the ability of a tuned-detector to receive weak signals by adding an
amplifier to the system, but this does not solve the other fundamental problem with this
receiver, which is the fundamental lower limit to the bandwidth that can be achieved with
simple filters based on inductor/capacitor resonators. This limitation will be discussed in
more detail in section 2.2. For the time being, it suces to say that an LC filter, whose center
frequency must be adjustable, is limited to a bandwidth no smaller than approximately 1% of
its center frequency. While tunable LC filters with smaller bandwidths can be designed, the
attenuation (loss) of such filters soars as the design bandwidth is decreased below about 1%.
In the early days of radio, carrier frequencies were low, so fractional bandwidths of signals
were relatively high, and the 1% lower limit on bandwidth was not a serious problem. As
shorter wavelengths came into use, the fractional bandwidth occupied by signals became
smaller than 1%, and some means of obtaining narrower bandwidth was needed.

2.1.2 Tuned Radio Frequency (TRF) Receiver

One very popular receiver architecture that was employed in the 1920s for AM broadcast
receivers was sort of a brute-force approach to obtaining high gain and narrow bandwidth.
The tuned-radio-frequency (TRF) receiver utilized a cascade of several tuned amplifiers,
2.1. INTRODUCTION AND HISTORICAL PROGRESSION 69

each having a bandpass response, as shown in Figure 2.2, where each tuned amplifier is
represented by separate filter and amplifier blocks. If the filter/amplifier stages are identical,
and if the amplifiers provide an eective buer between the filters, then the overall transfer
function of the cascade is equal to the transfer function of a single filter/amplifier stage
raised to the nth power, where n is the number of stages. This architecture provides high
gain and narrower bandwidth than can be obtained with a single filter. The TRF receivers
were fairly dicult to tune properly, as each of the tuned circuits had to be tuned to the
carrier frequency of interest. If only one of the filters was mis-tuned, the overall gain of
the receiver would drop to a very low value. Another problem with the TRF receiver
was the tendency for the system to oscillate. This problem is a manifestation of the too
much gain in one box syndrome, which occurs whenever reverse isolation (the attenuation
between the output and the input) of an amplifier chain does not exceed the forward gain
by a comfortable margin. Leakage from the output back to the input provides a feedback
path, which can lead to oscillation. In early receivers, the internal feedback within the triode
vacuum tubes used as amplifiers provided enough reverse coupling to cause oscillation unless
the forward gain was kept small. A technique called neutralization was developed to cancel
the internal feedback of the active devices, and the TRF receiver employing neutralization
became known as the neutrodyne receiver. The triode-based neutrodyne was followed by a
second generation TRF receiver based on the tetrode vacuum tube, which had almost no
internal feedback and did not require neutralization. The tetrode-based TRF receiver was
available through the early 1930s.

Demodulator

fc fc fc

Figure 2.2: Tuned-radio-frequency (TRF) receiver architecture.

2.1.3 Regenerative Receiver

The TRF receiver became a practical commercial device only when the cost, and availability,
of triode vacuum tubes made these devices available to the general public. Even before the
dawn of AM broadcasting, and the subsequent popularity of the TRF receiver, a particularly
ingenious concept was developed by Edwin Howard Armstrong, in 1912, which allowed him
to develop a receiver with high gain, and narrow bandwidth, using only a single triode.
Armstrong, an undergraduate at Columbia University at the time, used positive feedback
to increase the gain, and narrow the bandwidth of a single-stage tuned amplifier. A block
diagram of the system is shown in Figure 2.3. The transfer function of the system between
the antenna terminals and the demodulator input is
so (!) GF (!)
= . (2.1)
si (!) 1 AGF (!)
Without the positive feedback, the transfer function would be equal to the numerator of
equation 2.1. The gain of the filter/amplifier would be GF (!o ), and the fractional bandwidth
70 CHAPTER 2. RECEIVERS

F (!)

so (!)
si (!) G Demodulator

fc
so (!) GF (!)
si (!) = 1 AGF (!)

Figure 2.3: A regenerative receiver employing a regenerative amplifier in front of a demod-

ulator.

of the system would equal to that of the filter. With feedback, the gain at the center
frequency is GF (!o )/(1 AGF (!o )), which can be made arbitrarily large if AGF (!o ) is
allowed to approach 1 (from below!). The positive feedback that is built into the regenerative
amplifier is precisely what is necessary to build an oscillator. Note that if the quantity
AGF (!o ) is allowed to become equal to one, the transfer function becomes infinite, which
may be interpreted as a condition where finite output signal occurs with zero input, in
other words, the system oscillates and becomes a radio-frequency signal source, rather than
an amplifier. Armstrongs regenerative receiver operates the feedback loop just below the
threshold of oscillation, where the denominator of equation 2.1 is very small. Hence, the
gain of the system is very large. It turns out that if the filter is a simple LC resonator, then
the gain-bandwidth product of the system is independent of the value of AGF (!o ). Hence,
when the gain is very large, the bandwidth of the system can be very small. This provides a
means to obtain extremely high gain and extremely small fractional bandwidths, using only
a single active device.
The drawback to the regenerative receiver is the fact that the desired operating point is
where AGF (!o ) is close to one. If something happens to cause AGF (!o ) to become equal
to, or greater than, one, then the system becomes unstable, and oscillates. When the system
is oscillating, it no longer acts as an extremely high gain and narrow bandwidth amplifier
for small signals delivered by the antenna. In the oscillating mode, the regenerative receiver
essentially becomes a self-oscillating mixer, which does not provide any narrow bandwidth
filtering action. The diculty associated with adjusting the filter and feedback to achieve
high gain and narrow bandwidth made this receiver relatively dicult to use. It is also
inherently very sensitive to changes in component values and circuit layout, so that it is
rarely used in modern systems. Nevertheless, in the 1920s, there was an intense competition
between Westinghouse, who had acquired the rights to Armstrongs regenerative patent, and
a group of radio manufacturers that owned the rights to the neutrodyne patent.
An interesting, and even more exotic, variant of the regenerative receiver concept can be
found in the superregenerative receiver. Keeping in mind that the regenerative receiver is
most sensitive (in the sense that it has the highest gain) when it is operated at the threshold
between stability and instability (oscillation), the superregenerative receiver is designed so
that the quantity AGF (!o ) is actually larger than 1. Thus, oscillation starts and builds
2.2. CHARACTERISTICS OF PRACTICAL FILTERS 71

up immediately after power is applied to the system. Superregenerative receivers have a

built-in circuit that senses when oscillation starts (this is easily detected by changes in the
bias point of the active device) and, when oscillation is detected, the gain of the system is
momentarily reduced, so that the oscillation is quenched. After the oscillation is quenched,
the gain is restored, and oscillation builds up again. This cycle is repeated with a period
that is short compared to the shortest time scale associated with the modulation of the
signal that is being received. Thus, the system operates at extremely high gain for short
periods of time (while oscillation is building up) during each cycle, eectively providing
samples of the desired signal at a sampling rate that is adjusted to be high enough to allow
the detector to recover the modulation. Superregenerative receivers can still be found today
in devices such as garage door openers and radio controlled toys, where the primary design
requirement is absolute minimum cost.

2.1.4 Genesis of the Superheterodyne Receiver

Although the regenerative and TRF receivers dominated the consumer radio market into the
early 1930s, their replacement, and the architecture that is dominant today, was patented
by Edwin Armstrong in 1917, and is called the superheterodyne receiver. After his stint
at Columbia, Armstrong became a member of the U.S. Army Signal Corps during World
War I. Involved with eorts to find a way to detect enemy airplanes from a distance, he
knew that it might be possible to detect the electromagnetic emissions from the spark plugs
in the engine. The problem was that the emissions were strongest at the (then) unusually
high frequencies above a few MHz. Triodes of the day had very little gain at such high
frequencies, so Armstrong hit on the idea of employing the heterodyne principle to shift the
high-frequency signals to a lower frequency, where they could be more eciently amplified
and filtered. Once he decided to incorporate the heterodyne concept into his receiver, it
became possible to heterodyne any signal of interest, regardless of its frequency, to a fixed
intermediate frequency (IF). Selective filtering, and high-gain amplification could be done
at the fixed IF.
In contrast to the TRF and regenerative receivers, the new heterodyne-based receiver
did not require a tunable, high-gain, narrow-bandwidth, filter/amplifier. Instead, the high
gain, narrow bandwidth filter/amplifier only had to operate at the fixed IF which was a
relatively low frequency. This made it much easier to optimize the IF filters and amplifiers
for high gain and narrow bandwidth. A further advantage was obtained because converting
the desired radio frequency (RF) signal to a lower IF does not change the absolute band-
width of the signal. For example, a signal that occupies a bandwidth of 10 kHz at a carrier
frequency of 10 MHz occupies a fractional bandwidth of 0.01/10=0.001, or 0.1%. When
that same signal is heterodyned to an IF of, say 400 kHz, it retains its 10 kHz bandwidth,
so its fractional bandwidth becomes 10/400=0.025, or 2.5%, a comfortably large fractional
bandwidth. Receivers based on Armstrongs superheterodyne concept were originally pro-
duced by Radio Corporation of America (RCA) who, as a result, dominated the commercial
radio market by 1930. Before proceeding with a detailed overview of the superheterodyne
receiver, we shall first discuss the properties of practical filters.

2.2 Characteristics of Practical Filters

Losses in the components used to implement filter networks set a lower limit on the band-
width that can be achieved in a bandpass filter. For a bandpass filter with center frequency
72 CHAPTER 2. RECEIVERS

fo , a common measure of the bandwidth of the filter is the frequency spacing between the
3dB points on the filter transfer function, denoted by f 3dB in Figure 2.4. The dimen-

F (!)
Fo
p
Fo / 2

fo

f 3dB

Figure 2.4: Definition of -3dB bandwidth.

sionless fractional bandwidth, f 3dB /fo , of a filter implemented with a single LC resonant
circuit is 1/Q , where Q is the so-called loaded Q of the resonant circuit. In many cases,
the resonant circuit Q is limited by the Q of the inductor. Capacitor losses are usually
small compared to the losses in inductors. For miniature inductors, Qs may be limited to
values in the range 10-100, whereas inductors implemented using spirals of metallization
in integrated circuits typically have very low Q, often <10. The fractional bandwidth of
a filter implemented using an LC resonator with Q of 100 is 0.01, which means that the
minimum bandwidth of an LC filter is about 1% of the center frequency. If physically large
inductors can be tolerated, it is possible to achieve Qs as large as several hundred, even
approaching 1000, and fractional bandwidths significantly smaller than 0.01. Such high-
Q inductors are rarely practical for use in modern systems where small size, light weight,
and low cost are primary considerations. Generally speaking, however, LC filters with frac-
tional bandwidths smaller than around 0.01 tend to become impractical, even when they
can be realized, as mechanical and thermal stability of the components becomes an issue at
such small bandwidths, and such filters are easily de-tuned by changes in temperature, or
mechanical stresses.
Filters implemented using a single resonant circuit (resonator) do not have a very well-
defined passband. The transfer function is not constant within the -3dB bandwidth of the
filter, and it decays slowly outside of the passband. Thus, the bandwidth of the filter at
the -30dB points is much wider than the bandwidth between the -3dB points. A measure
of how closely a bandpass filters passband approaches that of an ideal, rectangular filter, is
the so-called shape factor of the filter response. Shape factor is defined as the ratio of the
bandwidth at some large attenuation (say 30dB) and the bandwidth at the 3dB attenuation
points. As a filters transfer function approaches the ideal, rectangular shape, its shape
factor approaches 1.0. A single-resonator LC filter is far from this ideal - for example, such
a filter with a Q of 50 has a fractional bandwidth of 1/50, or 0.02, and a shape factor (using
the -30 dB and -3 dB bandwidths) of 31.6. Thus, the -30 dB bandwidth is more than 30
times as wide as the -3 dB bandwidth. Obviously, if a single resonator filter response is
adjusted so that the signal of interest fills the -3dB bandwidth, an adjacent channel will not
2.2. CHARACTERISTICS OF PRACTICAL FILTERS 73

be attenuated by very much. Thus, the single resonator filter is not appropriate for providing
so-called adjacent-channel rejection. The shape of a filters transfer function can be made
to approach the idealized rectangular shape by using more (sometimes, many more) than a
single resonator. For example, using only two resonators, a filter with fractional bandwidth
of 0.02 and shape factor of 10.1 is easily obtained this is more than a 3:1 improvement
over the shape factor of a single-resonator filter. A comparison between the attenuation of
a single-resonator filter and a 2-resonator filter, each having fractional bandwidth of 0.02,
is shown in Figure 2.5.

Attenuation in dB
single resonator, Q=50 10 2 critically coupled
8 resonators, Q=62
6

0.97 0.98 0.99 1.01 1.02 1.03

Normalized Frequency (f /fc )

Figure 2.5: Attenuation vs normalized frequency for single and double resonator filters with
the same fractional bandwidth. The single resonator filter has a shape factor of approxi-
mately 32, whereas the 2-resonator filter has a shape factor of 10.

2.2.1 Transmission-line and cavity resonator filters

For microwave applications it becomes feasible to utilize resonators based on distributed
elements such as transmission lines, or waveguide cavities. Generally speaking, in their
optimum implementations these resonators allow higher Qs than lumped LC resonators,
and provide smaller minimum fractional bandwidths than provided by lumped LC filters.
They are not easily tunable, however, restricting their application to fixed center frequencies
and bandwidths. Filters based on distributed elements are most suitable for use as RF front-
end filters in microwave receivers.

2.2.2 Filters based on piezoelectric devices - Quartz-Crystal Filter,

Ceramic Filter
When very small fractional bandwidths and shape factors approaching 1.0 are required,
it is necessary to turn to electro-mechanical resonators implemented using piezoelectric
materials such as quartz crystals or ceramics, or electro-acoustic resonators implemented
using surface-acoustic-wave (SAW) devices. These devices have Q-factors several orders of
magnitude larger than those available from LC resonators, and they have much smaller
thermal coecients, allowing extremely narrowband filters to be realized. For example, a
quartz crystal may have a Q of 100,000, allowing fractional bandwidths as small as 10 5 to be
implemented. At a center frequency of 10 MHz, a fractional bandwidth of 10 5 corresponds
to -3 dB bandwidth of 100 Hz. Unfortunately, the electro-mechanical resonators are useful
74 CHAPTER 2. RECEIVERS

only at relatively low center frequencies (typically, < 30 MHz). Quartz-crystal and ceramic
filters are fixed-frequency devices, and cannot be tuned over any appreciable range.

2.2.3 SAW filters

At frequencies in the range 30 MHz2.5 GHz, filters based on surface-acoustic wave (SAW)
devices provide a convenient means to realize filters with fractional bandwidths as small
as 0.1% ( f /fo = 0.001) with excellent shape factors in a small package. These filters
eectively fit many resonators within a small volume by converting an electrical signal into
an acoustic signal, where the wavelength is 5 orders of magnitude smaller because of the
slow speed of sound relative to the speed of electromagnetic wave propagation. By filtering
the acoustic signal, it is possible to build filters which, eectively, contain 10s or hundreds
of resonators within an extremely small package. This makes it possible to realize filters
with excellent shape factors. SAW filters are fabricated for a fixed center frequency, and are
not tunable.

2.2.4 Filter limitations dictate carrier-frequency conversion

Consider the fractional bandwidth required to make a single PCS-band CDMA (code-
division-multiple-access) cellphone channel fill the -3dB bandwidth of a filter. Why fill
the bandwidth of the filter with the signal of interest? Generally speaking, when a signal
is accompanied by noise and interference, maximum signal-to-noise ratio will occur at the
output of a filter when the filter shape is matched to the spectrum of the signal. The band-
width of CDMA channel is 1.25 MHz, and the center frequency is approximately 1900 MHz.
A fractional bandwidth of 1.25/1900 .00066 is required, with shape factor as close to 1 as
possible, to select one channel while providing rejection of adjacent channels.
It is not possible to implement anything close to this small fractional bandwidth using
practical filters based on any type of conventional electronic resonator (LC, transmission line,
cavity resonators, etc). Quartz crystal filters can achieve the required fractional bandwidth,
but not at a center frequency above 100 MHz or so. The solution is to move the carrier
frequency of the desired channel to a lower frequency, without changing the bandwidth of
the signal or distorting the modulation in any way. If the carrier frequency is lowered to, say
45 MHz, then the fractional bandwidth of the signal becomes 1.25/45 0.028. This center
frequency and fractional bandwidth are well within the range where SAW filters provide
excellent performance. The concept of using frequency conversion to downconvert high
frequency signal to a lower frequency where filtering and amplification is easier to achieve
is embodied in the superheterodyne receiver.

2.3 The Superheterodyne Receiver

A block diagram of a single-conversion receiver is shown in Figure 2.6. The superheterodyne
receiver operates by converting the input frequency of interest (fc ) to a fixed intermediate
frequency (IF). This frequency conversion is performed by the mixer and local oscillator
(LO). The intermediate frequency may be higher or lower than the carrier frequency of
interest, but it is important to note that the IF is fixed (constant). Conversion to a fixed
frequency takes advantage of the fact that it is much easier to realize narrow-band filters
and stable, high-gain amplifiers if the frequency of operation doesnt change.
2.3. THE SUPERHETERODYNE RECEIVER 75

Radio Frequency (RF) Intermediate Frequency (IF)

section section
Preselector
filter RF amplifier IF filter IF amplifier
Mixer
Demodulator

fc fIF

Local Oscillator (LO)

fLO

Figure 2.6: Superheterodyne receiver

A perfect multiplier can be used to model the operation of the mixer for preliminary
discussions of the basic operation of the receiver, in which case the output of the mixer will
consist of two signals with carrier frequencies equal to the sum and dierence of the input
and LO frequencies. Consider a modulated input signal with carrier frequency fc which
may be both angle- and amplitude-modulated in general:

s(t) = A(t) cos[!c t + (t)] (2.2)

The local oscillator signal will be an unmodulated carrier with frequency fLO , i.e.,
cos(!LO t). The output signal from the mixer will be

sout (t) = A(t) cos[!c t + (t)] cos(!LO t) (2.3)

1
= {A(t) cos [(!c !LO )t + (t)] + A(t) cos [(!c + !LO )t + (t)]}
2
Note that the output from the mixer/LO consists of two signals with carrier frequencies
fc fLO and fc + fLO . Note also that the amplitude and angle modulation that was present
on the input signal has been transferred to the two output signals without any distortion.
The IF filter picks out one of the two signals and rejects the other. If a range of
input frequencies is to be covered, then the LO will be tunable. Since the IF frequency
is fixed, the LO frequency is adjusted, in practice, in order to make fIF = fc + fLO or
fIF = |fc fLO |. The choice of whether the sum or dierence frequency is picked out
by the IF filter is determined in the design stage. If fIF < fc the configuration is called
down-conversion, since the carrier frequency has been converted down to fIF . If fIF > fc ,
the receiver is said to employ up-conversion.

2.3.1 Image Frequencies

Just as there are two output signals from the mixer for each input frequency, there are two
input frequencies to the mixer that will give an output at the IF frequency. In practice
only one of these will be the desired frequency, and the other, undesired, frequency is called
76 CHAPTER 2. RECEIVERS

the image frequency. If an undesired signal happens to have a carrier frequency that is the
same as the image frequency, then that signal would be mixed into the IF filters passband
along with the desired signal. In the superhet (see Figure 2.6) the primary function of the
preselector filter is to pass the desired signal and to reject signals at the image frequency.
In a well designed receiver the image frequency is separated from the carrier frequency by
a relatively large interval. Thus the preselector filter does not need to have a very narrow
bandwidth. This is desirable, since the preselector filter often needs to be tunable. It is
dicult to build a narrowband filter that has a tunable center frequency.
The preselector filter does not provide adjacent channel rejection, i.e., it does not reject
signals that have carrier frequencies immediately adjacent to that of the desired signal.
This is the function of the intermediate frequency (IF) filter which will determine the signal
bandwidth of the receiver and, therefore, provide the rejection of adjacent channels. The
bandwidth of the IF filter will normally be chosen to be just wide enough to accept the
spectrum of the desired signal. Of course, interfering signals may still occur within the
spectrum of the desired signal. In this case a notch filter can sometimes be used in the
IF path to reject a very narrow slice of the signal spectrum which contains the interfering
signal without significantly aecting the desired signal. The adjacent channel selectivity of
a receiver is a function of the steepness at which the IF filter response rolls o outside of
the passband.

2.3.2 Operation of the Mixer/LO Stage - Useful Relationships

For given IF and input frequencies, there are two possible LO frequencies that will cause
the input frequency to be converted to the IF frequency:

fLO = fIF + fC or fLO = |fIF fC | (2.4)

For given IF and LO frequencies, there are two possible carrier frequencies that will give
an output at the IF frequency:

fc1 = fLO + fIF and fc2 = |fLO fIF | (2.5)

One of these will be the desired carrier frequency, fc , and the other will be an undesired
image frequency, fIM . The choice that was made for the LO frequency will determine which
of these equations gives the image and/or desired frequency. The primary purpose of the
preselector is to reject the undesired image frequency.
There are four possible generic configurations for a single-conversion superheterodyne
receiver. Define fcmin and fcmax to be the lower and upper limits, respectively, of the
input frequency range that the receiver is to cover. In practice the IF frequency will either
be smaller than fcmin or larger than fcmax . These two cases can be further subdivided,
because there are two possible choices for the LO tuning range for each choice of IF. The
image frequency and the separation between the desired and image frequencies (|fIM fc |)
are summarized in Table 2.1 for each of the four cases.

2.3.3 Example - AM Broadcast Receiver.

A typical AM broadcast receiver covers the frequency range 540 - 1700 kHz. AM broadcast
stations in the U.S. are assigned frequencies that are integer multiples of 10 kHz, therefore
the adjacent channel separation is 10 kHz. The IF frequency is very often chosen to be 455
2.3. THE SUPERHETERODYNE RECEIVER 77

Table 2.1: Generic Configurations for a Single-conversion Superheterodyne Receiver

(1) Up - conversion: fIF > fcmax > fcmin

(1a) fLO = fIF + fc fIM = fc + 2fIF |fIM fc | = 2fIF
(1b) fLO = fIF fc fIM = 2fIF fc |fIM fc | = 2fLO

(2) Down - conversion: fIF < fcmin < fcmax

(2a) fLO = fIF + fc fIM = fc + 2fIF |fIM fc | = 2fIF
(2b) fLO = fc fIF fIM = |fc 2fIF | |fIM fc | = 2fIF if fc > 2fIF
|fIM fc | = 2fLO if fc < 2fIF

kHz. This is an example of down-conversion. From (2a) and (2b), there are two choices
for the local oscillator tuning range:
1. 995 kHz to 2155 kHz
2. 85 kHz to 1245 kHz
The tuning ratio (R = fLOmax /fLOmin ) is an important factor in determining the cost
and stability of an oscillator. It is easier to build a stable oscillator if the tuning ratio is
small. Note that the tuning ratio for choices (1) and (2) are 2.16 and 14.6, respectively.
For this reason the first choice given above is almost always used. In fact, in most (but not
all) cases the higher of the two possible LO frequencies will be used (i.e., (1a) or (2a) from
Table 2.1), because this leads to the smaller tuning ratio. In this particular example the
second choice has another disadvantage, since it would require the local oscillator to tune
through the IF frequency, i.e., to receive a signal at 910 kHz; the second choice requires the
local oscillator to be tuned to 455 kHz. It is very likely that some of the local oscillator
signal would leak into the IF stage of the receiver. The result would be interference with
the desired signal, as well as possible overload of the IF and succeeding stages.
Assuming the LO tunes from 995 to 2155 kHz, then the image frequency will always be
(from (2a)) given by fIM = fc + 2fIF . For example, if we wish to tune the receiver to WILL
at 580 kHz, the preselector filter would be tuned to 580 kHz and the local oscillator would
be tuned to 1035 kHz. The bandwidth of the preselector needs to be narrow enough to
reject any incoming signals at the image frequency fIM = 1490 kHz. The IF filter needs
to be 10 kHz wide in order to provide adjacent channel selectivity.
As the radio is tuned across the band, it is necessary to tune both the LO and the
preselector simultaneously. This is why the tuning knob on mechanically tuned AM radios
is connected to a two-section, or ganged, variable capacitor. One section tunes the LO,
and the other tunes the preselector filter. One or more adjustments is usually provided to
insure that the two sections track each other.
Note carefully that for carrier frequencies less than 1245 kHz, the local oscillator fre-
quency will be less than 1700 kHz and therefore inside the AM broadcast band. It is
sometimes possible to hear this LO signal on another nearby AM radio. This phenomenon
is undesirable and is caused by the local oscillator signal being fed back through the mixer
and out the antenna, or by direct radiation from the LO circuitry. Commercial and military
receivers are carefully designed to minimize such radiation. This is done by using balanced
mixers that have very low direct output-to-input coupling coecients and by carefully shield-
ing the LO circuitry to avoid direct radiation. If the LO frequency is well removed from the
carrier frequency, a notch filter can be put before the mixer to reduce the LO signal that is
78 CHAPTER 2. RECEIVERS

coupled to the antenna.

The configuration of a standard FM broadcast receiver is described in the next example.

2.3.4 Example - FM Broadcast Receiver

The FM broadcast band covers 88 - 108 MHz. The channels are separated by 200 kHz and
are assigned to odd multiples of 100 kHz. In almost all cases the IF frequency is chosen to
be 10.7 MHz. As in the previous example high-side LO is often used, i.e., the LO tunes
from 98.7 to 118.7 MHz.
Notice that the IF used for FM broadcast receivers is substantially higher than that
used for AM receivers. This choice is motivated by the fact that the image frequency is
separated from the desired carrier frequency by 2fIF . Generally speaking, in order to make
the preselector filter relatively easy to build and tune, the IF frequency must be raised as
the carrier frequency increases.

2.3.5 Up-conversion versus Down-conversion

The AM and FM broadcast receivers considered above both use down-conversion and the
higher of the two possible LO frequencies. The image frequency was 2fIF above the desired
carrier frequency. In the AM receiver case the width of the carrier frequency band of interest
is larger than 2fIF . This means that the image frequency can fall within the band of interest,
thus necessitating the use of a tunable preselector filter that has a bandpass characteristic.
In receivers designed to cover very wide frequency bands the use of a tunable preselector
becomes impractical, because it is dicult to build filters whose center frequency must be
tuned over a wide frequency range. In this case it is advantageous to use up-conversion,
because with up-conversion the image frequency will always be larger than fcmax . Thus,
the image frequency will always fall outside (above) the band of interest, and a preselector
filter with a lowpass characteristic is sucient. A more practical choice, however, would be
a fix-tuned bandpass filter that passes the entire carrier frequency band of interest. The
bandpass filter is usually the best option, since it minimizes the possibility of interference
and receiver overload from strong signals outside of the frequency range of interest. The
up-conversion scheme can be a significant advantage, especially in a receiver that is designed
to cover a wide frequency range, since only the LO needs to be tuned.

2.3.6 Single- versus Double-conversion

There is a lower limit to the fractional bandwidth ( f /fo ) that can be realized with practi-
cal filters. This sets an upper limit on the intermediate frequency for a given signal spectrum
bandwidth. However, because the separation of the image frequency from the desired fre-
quency is usually twice the intermediate frequency, a large intermediate frequency is desired.
A large separation is desirable because it makes rejection of the image easier. In some cases
these conflicting considerations make it impossible to select an intermediate frequency that
is low enough to achieve adequate adjacent channel selectivity but large enough to make it
possible to reject the image frequency with a simple preselector filter. Then it becomes nec-
essary to use a double-conversion scheme (Figure 2.7) where the carrier frequency of interest
is first converted to a relatively high IF (allowing rejection of the image frequency with a
simple preselector filter) and then converted again to a lower IF, in order to obtain the
2.3. THE SUPERHETERODYNE RECEIVER 79

Preselector
filter RF amplifier First IF filter IF amplifier 2nd IF filter IF amplifier
Mixer Mixer

to
demod
fc fIF 1 fIF 2

fLO1 fLO2

Figure 2.7: Double-conversion superheterodyne receiver. The first conversion (fc ! fIF 1 )
can be either up- or down-conversion. The second conversion (fIF 1 ! fIF 2 ) is always a
downconversion.

required bandwidth and adjacent channel selectivity. In practice, the first conversion can be
either an up- or down-conversion, but the second conversion is always a down-conversion.
In certain cases there are advantages to using more than two conversions, i.e., triple-
conversion, although problems with spurious responses multiply rapidly as the number of
frequency conversions increases. The image frequency causes one type of spurious response
that is commonly observed in inexpensive AM and FM broadcast receivers where a certain
radio station is received when the receiver is tuned to a frequency other than the true carrier
frequency of the signal. There are several other origins for spurious responses involving both
external signals and signals generated within the receiver. These will be discussed in some
detail in Chapter 12. An important step in the design procedure is to identify and evaluate
the impact of spurious responses on the overall operation of the receiver.
In a multiple-conversion receiver the bandwidth of the last IF stage sets the overall
bandwidth of the receiver. In a double-conversion receiver the bandwidth of the first IF
filter would be chosen to reject secondary image frequencies within the first IF bandwidth.
A secondary image would be an undesired frequency within the passband of the first IF filter
that could be mixed into the second IF filters passband. The second IF filters bandwidth
would usually be just wide enough to pass the entire spectrum of the desired signal.

2.3.7 The 1/2-IF response

Nonlinearity in the RF, LO, and mixer sections of the receiver causes mixing products
involving harmonics of the RF and LO signals to be present at the mixer output. This
causes spurious responses, i.e. the receiver responds to input frequencies other than the
desired frequency. One example of this type of spurious response is the so-called 1/2-IF
response. The 1/2-IF response results from the second harmonic of an input signal mixing
with the second harmonic of the LO signal to produce an output at fIF . The mechanism is
illustrated here for the case of high-LO and down-conversion. Suppose that the receiver is
tuned to receive a desired signal with carrier frequency fC . The LO will be tuned to

fLO = fC + fIF (2.6)

Suppose a signal with frequency fS > fC is also present at the receiver input. If the
second harmonic of this signal mixes with the second harmonic of the LO, the mixer output
will have a component at |2fS 2fLO |. This signal will interfere with the desired signal if
80 CHAPTER 2. RECEIVERS

|2fS 2fLO | = fIF , or if

fS = fC + fIF fIF /2.
Thus a spurious response occurs when fS = fC + (3/2)fIF or fS = fC + fIF /2. Notice that
both of these responses are closer to the desired signal than is the image frequency. The
closest spurious response is separated from the desired signal by 1/2 of the IF frequency.
The 1/2-IF response plays a role in the choice of fIF . Suppose that the signals of
interest are located within a carrier frequency range denoted by R. It is desirable to use
a fix-tuned preselector that passes the entire carrier frequency band of interest. In order
for the preselector to reject potential spurious responses it is necessary for all significant
spurious responses to fall outside of the carrier frequency range of interest. If only the
image response is considered, then this constraint is satisfied if fIF is chosen such that
2fIF > R, or fIF > R/2. When the 1/2-IF response is considered, then spurious responses
will fall outside of the desired range if fIF /2 > R, or fIF > 2R. Notice that when the
1/2-IF response is considered the minimum value of fIF is 4 times as large as when only
the image response is considered.

2.4 Zero-IF receiver

In many applications, it is highly desirable to be able to implement an entire receiver in an
integrated circuit, without the need for bulky and expensive IF filters, which require that
signals be routed o-chip to the filter. For such applications the zero-IF (ZIF) (also called
direct-conversion or homodyne) approach has gained popularity. There are various ways
of explaining the evolution of this concept, e.g. it is sometimes described as the limit in
which the intermediate frequency of a superhet approaches zero. Perhaps it is more accurate
to say that a ZIF receiver amounts to using a quadrature demodulator as the receiver. A
ZIF receiver is shown in Figure 2.8. The dierence between this and a basic quadrature

I(t)
cos !ct

0
90
cos !ct
fc
sin !ct
Q(t)

Figure 2.8: Direct-conversion receiver. Depending on the application, the local carrier
oscillator may need to be derived from a carrier recovery circuit.

demodulator is the addition of an RF filter and low-noise amplifier (LNA) in front of the
demodulator, and addition of baseband amplifiers after the lowpass filters in each arm of
the demodulator. In a ZIF receiver, the filters responsible for selecting a desired channel,
2.5. SOFTWARE DEFINED RADIO 81

and rejecting undesired channels, are the lowpass filters at the output of each multiplier
(mixer). This is attractive, since active lowpass filters can be applied here. A very flexible
receiver architecture results if the I and Q outputs of the ZIF demodulator are sampled with
an analog-to-digital converter so that digital signal processing techniques can be applied to
implement very good channel selection filters and to perform the signal processing required
for demodulation. This relaxes the requirement on the analog LPFs to the simpler task of
providing the anti-aliasing function.
The ZIF concept looks deceptively simple. In practice, a number of issues combine to
make it a challenge to obtain good performance using this architecture. In a ZIF receiver,
most of the gain will be provided by the baseband amplifiers after the LPFs. Hence, desired
signals at the output of the multipliers are generally much smaller than they would be if
the quadrature demodulator was used after the IF stage of a conventional superhet receiver,
where signal levels are relatively high by virtue of the high IF gain. The small desired signals
are easily corrupted by noise and DC osets contributed by the mixers, local oscillator,
and the DC-coupled baseband amplifiers. The local oscillator contributes to the DC oset
problem because self-mixing of the local oscillator with itself results in a DC component. In
addition, the self-mixing brings the phase noise sidebands of the local oscillator down into
the baseband spectrum, especially in the region near DC. The downconverted LO phase
noise adds to 1/f noise (flicker noise) contributed by the mixer and amplifiers, and can
significantly degrade the signal to noise ratio for signal components located near the carrier
frequency, which are converted to very low frequencies near DC after the mixers.
Development of ZIF receivers is an active research area, with eorts focused on mini-
mizing mixer and LNA noise, development of extremely linear LNAs, and development of
low phase noise oscillators.
Some of the advantages of the zero-IF configuration can be obtained while avoiding the
problems associated with DC osets and 1/f noise near 0 Hz by downconverting the signal
to a low-IF instead of all the way to 0 Hz. Figure 2.9 compares the architectures of a
conventional superhet receiver, the low-IF, and the zero-IF configurations.

2.5 Software Defined Radio

The term Software Defined Radio (SDR) has become popular to describe transmitter and
receiver systems that utilize digital signal processing techniques to minimize the amount of
specialized analog hardware required for a specific application, and to allow reconfiguring
the communications system with changes in software alone. This is especially important
in view of the large number of standards in place, and under development, for the various
communications applications; cellular telephony is a good example - as of today, there are at
least 5 popular standards for modulation format in use, implemented in two widely separated
frequency bands.
Considering the receiver portion of a SDR-based transceiver, the ideal SDR would
employ an A/D converter to sample the voltage across the antenna terminals, and then
perform all filtering and demodulation in software. This is not practical at the present time,
mainly due to the limitations on A/D converter dynamic range (number of useful bits of
resolution) and sampling speed. Practical implementations of SDRs, in order of increasing
amount of software control and decreasing amount of hardware required, include:

1. sample the I and Q outputs of a quadrature demodulator, allowing fine-tuning, carrier

synchronization, demodulation, to be carried out entirely in software. The analog
82 CHAPTER 2. RECEIVERS

LPF
Superheterodyne
A/D
BW/2

0
DSP-Demodulator
90
fIF LPF

A/D
BW/2
(a)
Digital Signal Processor

LPF
Low IF

Image-reject mixer BW/2

0
A/D Demodulator
90
fIF LPF

BW/2

(b)
LPF
Zero-IF
A/D
BW/2

0
DSP-Demodulator
90
LPF

A/D
BW/2
(c)

Figure 2.9: Comparison of (a) superheterodyne, (b) low-IF superhet, and (c) zero-IF receiver
configurations with analog-to-digital (A/D) conversion and digital signal processing stages.
In (a) and (b) the IF filter passbands have been drawn to illustrate the fact that fIF is
typically large compared to the signal bandwidth in the case of a standard superheterodyne
receiver, whereas fIF is relatively small (low) in the case of the low-IF receiver. The rela-
tively low IF in the latter case makes it possible to digitize the IF output with high fidelity,
allowing the quadrature downconversion and demodulation stages to be implemented in a
DSP. The need for only one A/D converter is an advantage for the low-IF superhet receiver.
On the other hand, the preselector in a low-IF receiver will not be able to provide much
image rejection because the image response will be located close to the desired frequency.
An image-reject mixer can be used to provide additional image rejection (see homework
problem 18). The advantage of having only a single A/D converter comes at the cost of the
additional complexity of the image-reject mixer.
2.5. SOFTWARE DEFINED RADIO 83

frontend must implement some filtering, amplification, quadrature LO generation,

and multiplication/mixing of the RF and LO signals.
2. when very high speed A/D converters are available, a wide-bandwidth slice of the
filtered/amplified antenna output, or a wideband IF output, can be sampled at high
speed and digitally downconverted to produce sampled I and Q components with
narrower bandwidth and lower sampling frequency using a high speed dedicated signal
processor. Several specialized digital downconverter chips are available to perform this
function. Typically, the digital downverter samples at high rate and then applies a
rudimentary digital filter to the samples. The main purpose of the digital filter is to
place nulls at frequencies that will be folded into the desired slice of the spectrum after
decimation is performed. After the filter, the high-speed samples are decimated to a
rate that is at least twice the bandwidth required for subsequent processing. The lower-
rate samples are then processed by a conventional programmable DSP or computer.
The analog frontend provides filtering, amplification, and possible downconversion to
an IF; generation of the quadrature LO, mixing to I/Q channels, and demodulation is
performed in the digital downconverter and in software.

For those interested in learning more about SDR techniques through experimentation, there
are two excellent open-source programs available for *NIX platforms:
1. Linrad (www.nitehawk.com/sm5bsz/linuxdsp/linroot.htm) is a software receiver for
*NIX plaftorms that uses a stereo sound card to sample the I and Q outputs of a
quadrature demodulator. The program has AM/FM/SSB demodulators, and provides
a realtime waterfall display. This program is intended for demodulation of weak,
narrowband signals. A great deal of work has been put into implementing state-of-
the-art noise blanking, making this the program of choice for amateur radio operators
experimenting with communications using signals bounced o of the moon; called
earth-moon-earth, or eme, communications.

2. GnuRadio is the GNU software radio (www.gnu.org/software/gnuradio/index.html).

According to their web site: GNU Radio is a collection of software that when combined
with minimal hardware, allows the construction of radios where the actual waveforms
transmitted and received are defined by software. What this means is that it turns
the digital modulation schemes used in todays high performance wireless devices into
software problems.. GnuRadio software runs on *NIX, and can use various input
devices, including sound cards and high-speed A/D cards. Many state-of-the-art DSP
techniques applicable to SDR are available as modules. Receiving systems can be
easily put together and reconfigured using a Python interface.
84 CHAPTER 2. RECEIVERS

2.6 References
1. Communications Receivers: Principles and Design, Ulrich L. Rohde, Jerry Whitaker,
and T. T. N. Bucher, 2nd Edition, McGraw Hill, New York, 1997.

2. Single-Sideband Systems and Circuits, Edited by William E. Sabin and Edgar O.

Schoenike, McGraw Hill, New York, 1987.
3. The Science of Radio, Paul J. Nahim, 2nd Edition, Springer Verlag, New York, 2001.
4. The Design of CMOS Radio-Frequency Integrated Circuits, Thomas H. Lee, Cambridge
University Press, 1998.
5. RF Microelectronics, Behzad Razavi, Prentice Hall, 1998.
2.7. HOMEWORK PROBLEMS 85

2.7 Homework Problems

1. The transfer function for a regenerative amplifier is found to be

GF (!)
T (!) = (2.7)
1 AGF (!)
where A and G are assumed to be positive real constants with AG < 1. Suppose that
the filter frequency response function is approximated by a Gaussian function, i.e.,

F (!) = exp[ (! !c )2 /B 2 ] (2.8)

This filter has a 3 dB bandwidth W = 1.177 B.

(a) Find an expression for the 3 dB bandwidth of the regenerative amplifiers transfer
function, T(!).
(b) Evaluate the 3 dB bandwidth for the cases AG = 0.9 and AG = 0.99.
2. The transfer function for a regenerative amplifier is:
GF (!)
T (!) =
1 AGF (!)
where A and G are positive, real constants with AG < 1. Suppose that the filter is
implemented with a series RLC circuit, in which case the frequency response function,
F (!), will have the form:
1
F (!) = !o ,
1 + jQ( !!o ! )

where the factor Q is a dimensionless constant.

(a) Find an expression for the -3 dB bandwidth of F (!). Note that |F (!)| will be a
bandpass function with a peak at ! = !o . The -3 dB bandwidth is defined as the
separation between the two frequenciesp !1 , and !2 which satisfy !1 > 0, !2 > 0
and |F (!1 )| = |F (!2 )| = |F (!o )|/ 2.
(b) Find an expression for the -3 dB bandwidth of T (!). Discuss how the magnitude
of the loop gain (AG) aects the peak gain of T (!) and the bandwidth of T (!).
(c) Refer to your results from part b, and comment on how the gain-bandwidth
product associated with T (!) depends on the feedback gain parameter, A.

3. Consider a bandpass filter consisting of a number, N , of cascaded filters which are

isolated from each other by buer amplifiers. Assume that each filter has the transfer
function given by F (!) in problem 2. The overall transfer function of the system will
be F (!)N . Derive an expression for the -3 dB bandwidth of the cascade of N filters.
Express the bandwidth as a product of two terms: (i) the bandwidth of a single filter
(this was found in problem 2, part a), and (ii) a bandwidth reduction factor which
depends only on N. Calculate the bandwidth reduction factor for N = 2, 3, 4, 5, 6.
4. Suppose we want to design a superhet receiver to cover the frequency range 1 - 30 MHz.
Two possible intermediate frequencies (IFs) for the receiver are:
86 CHAPTER 2. RECEIVERS

(i) 500 kHz

(ii) 60 MHz

For these IFs:

(a) Find the possible local oscillator frequencies (fLO ) and image frequencies (fIM )
for reception of signals with carrier frequencies of 1, 5, 15, and 25 MHz. Make a
table.
(b) What are the LO tuning ratios required to cover the entire 1-30 MHz range for
each possible choice of LO frequency?
(c) From the standpoint of the easiest and least expensive preselector and LO design,
what is the optimum configuration? Assume that the required IF bandwidth is
easy to obtain, whatever the choice of the LO frequency. For the optimum con-
figuration, describe the required preselector frequency response. Is it necessary
to tune the preselector?

5. Consider a single conversion superhet receiver that uses low-LO and downconversion.

(a) If the signals of interest have carrier frequencies in the range 140-170 MHz and
fIF = 21.4 MHz, find the tuning range for the local oscillator. Specify the
minimum and maximum local oscillator frequencies.
(b) If the receiver is tuned to receive a desired signal with carrier frequency fc = 150
MHz, find the image frequency.
(c) Assume that the preselector will have an ideal rectangular frequency response
function. Does the preselector for this receiver have to be tunable? (Yes or no).
Explain your answer. Do not consider the 1/2-IF response for this question.
(d) When the receiver is tuned to receive a desired signal with carrier frequency
fc = 150 MHz, denote the frequency of a signal that could cause interference if
the second harmonic of the interferer mixes with the second harmonic of the LO
by fi . Find all possible values of fi .

6. Consider the double-conversion receiver shown in Figure 2.10.

Preselector
filter RF amplifier First IF filter IF amplifier 2nd IF filter IF amplifier
Mixer Mixer

to
demod
fc fIF 1 fIF 2

fLO1 fLO2

Figure 2.10: Double-conversion receiver

Note that a double-conversion receiver requires two IF filters. Assume that the receiver
is to cover the 2-to-30 MHz range. The center frequency of the first IF filter is 50 MHz
and that of the second is 500 kHz.
2.7. HOMEWORK PROBLEMS 87

(a) Specify the possible frequencies (fLO1 and fLO2 ) for a receiver covering 2 to 30
MHz. You must specify a range of frequencies for fLO1 , and a single frequency
for fLO2 . Be sure to list both possibilities for each.
(b) Assume the first IF filter is a bandpass filter with ideal (rectangular) frequency
response. What is the maximum bandwidth of the filter if secondary images are
to be avoided? A secondary image is an undesired frequency within the first IF
filters passband that could be mixed into the second IF filters passband. Assume
that the second IF filter has a rectangular response with bandwidth 10 kHz.

7. We will design a receiver that can receive lower sideband signals. A block diagram
of the receiver is shown in Figure 2.11. Note that the SSB signal is demodulated
using a mixer and beat-frequency oscillator (BFO). The BFO and second mixer can
be thought of as a second frequency conversion stage where the second IF is at 0 Hz
(DC). The purpose of this stage is to translate the signal spectrum down to base-band,
i.e., to translate the carrier frequency to DC.

Preselector
filter RF amplifier First IF filter IF amplifier Low-pass filter
Mixer BW1 Mixer

Audio out

fc fIF 1
3 kHz
9 MHz
fLO1 Beat Frequency
Oscillator (BFO)

Figure 2.11: Receiver to receive lower sidebands

Suppose the input signal is an LSB signal with carrier frequency fc =14.3 MHz and
bandwidth = 3 kHz as shown in Figure 2.12. You may assume that the IF filter has

LSB

fc 3 kHz fc

Figure 2.12: LSB signal

an ideal rectangular bandpass characteristic with center frequency of 9.0 MHz and
bandwidth BW1 = 3 kHz. Thus, the IF filter bandwidth is just wide enough to pass
the LSB signals spectrum.

(a) Give two choices for the frequency of the first local oscillator, fLO1 . Be careful:
make sure that your choice of local oscillator frequency will cause the entire
spectrum of the desired signal to fall within the IF filters passband.
88 CHAPTER 2. RECEIVERS

(b) For each choice given in part 7a, sketch the spectrum of the signal as it would
appear just after the IF filter.
(c) For each choice given in part 7a, give the band of frequencies that correspond to
the image response of the receiver.
(d) For each choice given in part 7a, give the frequency of the beat-frequency oscillator
such that the signal is properly demodulated.

8. Consider the triple-conversion receiver shown in Figure 2.13. Suppose that fLO1 =
1300 MHz, fLO2 = 1410 MHz, and fLO3 = 179.3 MHz. Note that the only filter is the
final IF filter which has an ideal rectangular frequency response function centered at
fIF =10.7 MHz:

fIF

fLO1 fLO2 fLO3

Figure 2.13: Triple-conversion receiver

(a) Find and list all input frequencies that will give an output from the final IF
filter. You may ignore the finite bandwidth of the IF filter, i.e., list only the
input frequencies that will give an output at exactly 10.7 MHz.
(b) How many input frequencies would be on your list if there were 4 conversions
instead of 3?

9. Consider the design of the receiver portion of a portable cellular telephone. The
cellular telephone receives frequency-modulated that are transmitted from a base-
station. The received signals have a bandwidth of 30 kHz and a carrier frequency, fc ,
somewhere within the frequency range 869-894 MHz. While receiving, the telephone
must simultaneously transmit frequency-modulated signals back to the base-station
at a carrier frequency fc -45 MHz, i.e., the telephone transmits at a frequency 45 MHz
below the frequency at which it receives.
Sketch a block diagram of a double-conversion superhet receiver that could be used for
receiving cellular phone signals. Your design should have a tunable first local oscillator
that will serve a dual purpose. This oscillator will also serve as the carrier oscillator
for the signals transmitted by the telephone. Assume that the second IF filter has a
center frequency of 5.5 MHz.
Specify the following and state the reasoning that led to your choice.

(a) The center frequency of the first IF filter.

2.7. HOMEWORK PROBLEMS 89

(b) The maximum bandwidth of the first IF filter. Assume that the filter response is
symmetric around the center frequency. Indicate what considerations led to your
choice for the maximum bandwidth.
(c) The bandwidth of the second IF filter.
(d) Sketch an appropriate transfer function for the preselector.
(e) Specify the tuning range for the first LO, i.e., specify the minimum and maximum
first LO frequencies.
(f) Specify the image frequency when the phone is set up to receive signals at fc =875
MHz.

10. Design a double-conversion receiver (refer to Fig. 2.10) that will receive frequencies in
the range 144-148 MHz. The signal bandwidth is 10 kHz. The first IF frequency is to
be 10.7 MHz. The second IF frequency is 455 kHz. You also know that the receiver is
going to be operated in a region where very strong local stations exist in the frequency
range 122-125 MHz.
Specify the following:

(a) The bandwidths of the first and second IF filters, BW1 and BW2 . Explain your
choices.
(b) The frequency (or frequency range) for each local oscillator, fLO1 and fLO2 .
Explain your choices. Tuning ratio is not a significant consideration in this
receiver, so do not invoke smaller tuning ratio as a reason for preferring one
choice for fLO1 over another one.
(c) The preselector filter would normally be a bandpass filter in this application. Is
it necessary to use a tunable (variable) preselector filter? Why or why not?

11. Consider a single-conversion superhet receiver with fIF = 260 MHz. The receiver
must be able to tune to Sirius and XM satellite radio signals. These signals have
carrier frequency 2332.5 MHz and bandwidth 25 MHz. The receiver must present
the entire 25 MHz bandwidth to a downstream signal processor that is responsible for
demultiplexing the many audio channels that are present on the received signal. Thus,
the IF filter will have center frequency 260 MHz, and bandwidth 25 MHz.

(a) Specify two possible frequencies for the local oscillator.

(b) For each answer in part (a.), specify the image frequency.
(c) Suppose that the receiver is implemented using low-LO. Give the frequency
of a signal that would cause interference as a result of the 1/2 IF response
mechanism.

12. Consider the double-conversion superhet receiver in Figure 2.14:

This receiver is to be designed to cover two dierent carrier-frequency bands (144-194
MHz and 420-470 MHz) with a single tunable first local oscillator and first IF Filter.
To change the carrier-frequency band, only the preselector filter will be switched as
shown in Figure 2.14. You may assume that the signals of interest have a bandwidth
of 15 kHz.
90 CHAPTER 2. RECEIVERS

Preselector 1
144-194 MHz

First IF filter 2nd IF filter

BW1 BW2

to
demod
fIF 1 fIF 2

Preselector 2 fLO1 fLO2

420-470 MHz

Figure 2.14: Dual band superhet receiver

(a) Specify the first intermediate frequency (fIF 1 ). There are two possibilities - find
both of them.
(b) For each possible IF found in part (a), specify the 50 MHz-wide range of frequen-
cies covered by the first local oscillator (fLO1 ).
(c) Specify the smallest and the largest bandwidth (BW1 ) that could be used for the
first IF filter. Explain what factors determined your choices. You may assume
that the second IF filter has an ideal rectangular frequency response function
with fIF 2 = 10.7 MHz and BW2 = 15 kHz.

13. The double conversion receiver shown in Figure 2.15 uses a special system to generate
the first local oscillator signal. This is called the Wadley Loop drift cancelling system.
It is used to make the receiver tuning insensitive to relatively small changes in the
second local oscillator frequency that arise because of frequency drift. This makes it
possible to use a relatively inexpensive oscillator for the second LO.

f RF = 200 MHz
First IF Filter Second IF Filter
Mixer Mixer
BW BW2
1
Preselector
Filter

f IF1 = 790 MHz f IF2 = 70 MHz

Local Oscillator
Filter
fLO1 =590 MHz

f LO f = 860 MHz
LO2

Figure 2.15: Wadley Loop drift canceling system

2.7. HOMEWORK PROBLEMS 91

Suppose the input signal of interest has carrier frequency fRF =200 MHz. The first IF
filter is centered at fIF 1 =790 MHz and the second IF filter is centered at fIF 2 =70 MHz.
The local oscillator filter is centered on 590 MHz. The RF signal has a bandwidth
of 50 kHz. Assume that the IF filters have ideal, symmetrical rectangular passbands
with bandwidth BW1 and BW2 as shown in Figure 2.15.

(a) Specify the bandwidth of the second IF filter, BW2 . State the reason for your
choice.
(b) Give the maximum allowable bandwidth of the first IF filter (BW1 ) so that
secondary images will be rejected.
(c) Determine fLO0
such that an input signal with carrier frequency of 200 MHz will
be centered in the passband of both the first and second IF filters when fLO2 =860
MHz and such that the frequency of the desired signal at the output of the second
mixer is independent of the exact value of fLO2 . Hint: Write fLO2 as fLO2 =860
+ and find the choice for fLO0
that makes the frequency after the second mixer
independent of . There is only one correct choice for fLO0
.
(d) If you have done part 13c correctly, then the frequency of the desired signal after
the second mixer will not change if fLO2 drifts away from its nominal value of 860
MHz. This does not mean that fLO2 can take on any value, however. Suppose
that represents the frequency drift, i.e., suppose that fLO2 =860 + . What
factor(s) limit the allowable size of ?

14. Consider a single conversion superhet receiver with fIF = 70 MHz. The receiver must
be able to tune to carrier frequencies in the range 1930 MHz fRF 1975 MHz.

(a) Specify the two possible frequency ranges for the first local oscillator.
(b) For each answer in part (a.), specify the image frequency when the receiver is
tuned to receive a desired signal with carrier frequency fRF = 1940MHz.
(c) In one or two sentences, explain how the 1/2 IF response of a receiver comes
about.
(d) Suppose that the receiver described in part (a.) is implemented using high-LO.
Give the frequency of a signal that would cause interference as a result of the
1/2 IF response mechanism when the receiver is tuned to receive a desired signal
with carrier frequency fRF = 1935 MHz.

15. You are designing a superhet receiver for a PCS-band cellular phone handset to receive
carrier frequencies in the range 1930 MHz fRF 1975 MHz. Assume that the
receiver will employ high-LO.

(a) Suppose that the preselector has a fixed rectangular bandpass response that
passes 1930-1975 MHz. Determine the smallest value for the IF that will allow
the preselector to reject signal(s) at the image frequency.
(b) Suppose that the preselector is the same as described in part a. Determine the
smallest value for the IF that will allow the preselector to reject signal(s) whose
second harmonic could mix with the second harmonic of the LO to produce an
output at the IF.
92 CHAPTER 2. RECEIVERS

16. Consider a double conversion superhet receiver with first IF and last IF denoted by
fIF 1 and fIF 2 , and bandwidth of the first and last IF filters denoted by f1 and f2 ,
respectively. You may assume that both IF filters have ideal rectangular response
functions that are symmetrical about the center frequencies and that fIF 2 fIF 1 .
(a) Give an expression for the maximum bandwidth of the first IF filter, f1,max ,
such that secondary images will be rejected. You do not need to consider the
half-IF response.
(b) Suppose, for the purpose of this problem, that it is impractical to build or acquire
IF filters with fractional bandwidth f /fIF smaller than 2%. For a receiver using
a last IF of 455kHz, specify the largest practical value for the first IF. For this
calculation, you may assume that the bandwidth of the second IF filter ( f2 ) is
small enough to ignore.
17. Consider a single conversion superhet receiver that uses downconversion to an IF
denoted by fIF and high-LO. We showed that when such a receiver is tuned to
receive a desired signal with carrier frequency fc , an undesired signal with frequency
fi = fc + fIF
2 could cause interference if the second harmonic of the undesired signal
mixes with the second harmonic of the LO. This is called the half-IF response because
the potential interferer is oset from the desired carrier frequency by fIF2 . Find the
oset from the desired carrier frequency, fc , for a signal at frequency fi that could
cause interference when the third harmonic (3fi ) mixes with the third harmonic of
the local oscillator (3fLO ). There are two answers. Find them both and express your
results in terms of fIF .
18. (Image-reject mixer) When a multiplier is used to perform frequency conversion it is
necessary to include a pre-selector filter in front of the multiplier to prevent signals at
the image frequency from being converted to the intermediate frequency (IF) at the
mixer output. The filter becomes hard to realize when the IF is small, since the image
frequency is very close to the desired frequency. When the IF is small, a better method
for rejecting images is the phasing method, whereby the response from the image is
cancelled at the mixer output. One implementation of a phasing-type image-reject
mixer is shown in Figure 2.16.

cos(!LO t)

0 +
RF - IF
90
cos(!LO t)
sin(!LO t)
90
phase
b c
delay

Figure 2.16: Image-reject mixer

2.7. HOMEWORK PROBLEMS 93

(a) For the RF input signal A cos !RF t, find the time-domain IF output signal if the
+ sign is used in the signal combiner. You will need to consider two cases: (i)
!LO > !RF and (ii) !LO < !RF . Be careful - if you have a signal of the form
sin[(!1 !2 )t] and you want to delay it by 90 degrees, then the result will be
cos[(!1 !2 )t] where the upper sign applies if !1 > !2 and the lower sign
applies if !1 < !2 .
(b) Repeat part a for the case where the sign is used in the signal combiner.
(c) Use your results from parts (a) and (b) to answer this part. Suppose this mixer
is to be used to convert a desired signal at frequency !c (i.e. RF input signal is
A cos !c t) to an IF denoted by !IF using high LO (!LO = !c + !IF ). The image
frequency is !IM = !c + 2!IF . Which sign should be used in the combiner in
order for the system to produce the desired frequency conversion?
(d) For the sign that you chose in part (c), write the IF output signal when the
input signals frequency is !IM = !c + 2!IF , i.e. when the RF input signal is
A cos !IM t. You should find that this input signal produces a finite output from
the mixer. Why, then, is this called an image-reject mixer?

19. We have seen several systems that use cancellation to eliminate unwanted signal com-
ponents. In practice, perfect cancellation is dicult to achieve because of inevitable
amplitude and phase dierences between the two signals that are being combined. In
this problem, we analyze the impact of amplitude and phase imbalance on the sup-
pression of unwanted terms. Consider the superposition of two signals which dier
slightly in amplitude and phase, i.e.

s (t) = A cos(!t) (A + A) cos(!t + ).

(a) Consider phase errors only ( A = 0) and find the maximum phase imbalance
( ), in degrees, if the dierence signal is to be suppressed by at least 10, 30,
and 60 dB relative to the sum signal. The dierence signal suppression, in dB,
is S = 10 log( PP+ ), where P represents the time-average power in the sum and
dierence signals.
(b) Consider amplitude errors only and repeat part (a). The amplitude imbalance,
expressed in dB, is I = 20 log( A+A A ). Determine the maximum amplitude im-
balance if the dierence signal is to be suppressed by at least 10, 30, and 60 dB
relative to the sum signal.
94 CHAPTER 2. RECEIVERS
Chapter 3

Properties of Passive Components

3.1 High Frequency Characteristics of Components

Passive electronic components such as resistors, capacitors, and inductors all exhibit some
amount of dielectric and/or ohmic loss, as well as energy storage in electric and magnetic
fields within and surrounding the component. Accurate models for the impedance of real
components must, therefore, include resistance, capacitance, and inductance. In addition,
the skin eect, and inductance and capacitance associated with the conductors that connect
the component to the rest of a circuit cannot be neglected. More generally, at suciently
high frequencies, the concept of a lumped element must be replaced with that of a distributed
circuit. An accurate equivalent circuit model that includes all of these eects may be
significantly more complex than the low-frequency circuit model that includes only the
basic circuit element. It is important to understand the nature of these eects, so that they
can be accounted for when designing practical circuits. Some of the most useful equivalent
circuit models for common components will be discussed in this section.

3.1.1 Wire Above a Ground Plane

l
d
h

Figure 3.1: Wire above a ground plane.

Figure 3.1 shows a cylindrical wire parallel to a ground plane. A similar geometry may
be used to model a trace on a circuit board if the cylindrical conductor is replaced with a
flat conducting strip.

95
96 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS

3.1.1.1 Resistance of Wires

Current tends to flow near the external surface of a conductor at high frequencies. The
time-varying magnetic field associated with an initially uniform current density induces
an electric field within the conductor which, in turn, drives a current against the initial
current, forcing the net current density deep within the conductor to zero. As a result, the
current density directed along the axis of a conducting wire is largest at the surface of the
conductor, and falls to small values inside the conductor. Most of the current flows within
the cylindrical shell within one skin depth from the surface. The skin depth is denoted by
, where
1
=p ,
f
f is frequency, in Hz, and , are the permeability and conductivity of the conductor. For
non-magnetic conductors the permeability will be equal to the permeability of free space,
= o = 4 10 7 H/m. Some values of conductivity are given in Table 3.1.

Material S/m
Aluminum 3.5107
Copper 5.8107
Brass 1.5107
Gold 4.1107
Silver 6.1107

Table 3.1: Conductivities for common metals.

The skin depth in copper at 100 kHz, 10 MHz, and 1 GHz is approximately 0.2 mm,
0.02 mm, and 0.002 mm, respectively. A typical wire used for connections and component
leads is American Wire Gauge (AWG) 22 gauge wire, which has diameter d = 25.3 mils, or
0.643 mm. Hence, the skin depth in copper is a small fraction of the wire diameter at 10
MHz and higher frequencies.
At low frequencies, where the wire diameter is comparable to, or smaller than, the skin
depth, then the DC resistance will be a good approximation to the wire resistance. The DC
resistance is calculated using the total cross sectional area of the wire, A = d2 /4, i.e.

l 4l
RDC = = . (3.1)
A d2
At higher frequencies, where the skin depth is small compared to the wire diameter, an
accurate approximation for the resistance of the wire is obtained by assuming that the
current flows with uniform density within one skin-depth, , from the wire surface, and with
zero density elsewhere. The AC resistance is then obtained by replacing A in equation (3.1)
with the area contained within one skin depth of the surface, i.e. d . The AC resistance
can then be written as
(d/2)2 d
R = RDC = RDC , d (3.2)
d 4
where RDC is the DC resistance of the wire. Since the skin depth decreases with increasing
frequency, the resistance will increase as f 1/2 . Obviously, the wire resistance can be much
greater than the DC resistance, especially at high frequencies. For example, consider AWG
3.1. HIGH FREQUENCY CHARACTERISTICS OF COMPONENTS 97

22 wire, which has DC resistance 0.0053 /cm. At 10 MHz the resistance is 0.04 /cm,
and at 1 GHz the resistance is 0.4 /cm.
The approximation given in equation (3.2) is based on the assumption that the wire is
far from any other conductors, so that the time-varying magnetic field within the wire is
only due to the current within the wire. Suppose that two parallel wires, carrying the same
current, are brought into close proximity. In this case, the magnetic field within each wire
will include a contribution from the current in the neighboring wire, i.e. the wires will be
inductively coupled. In this case, the current density within the wire will not be uniformly
distributed around the outer shell defined by the skin-depth. The electric field induced
within the wires will cause the current density to be decreased on the side of the wire that
is closest to the neighboring wire. This proximity eect reduces the eective cross-sectional
area even further, and increases the AC resistance of the wire. The proximity eect can be
very important in inductors where multiple wires are in close proximity.

3.1.1.2 Inductance of wires

The inductance, per unit length, of a wire with length l, diameter d, and distance h from a
ground plane is (when d l and h l):

0 1 2h
L= cosh . (3.3)
2 d
If the wire axis is located at least one wire diameter above the ground plane (h/d > 1), then
the following approximation is useful

0 4h
L' ln . (3.4)
2 d
For values of h/d in the range 1 to 100, equation (3.4) predicts that L ranges from 2.8
nH/cm to 12 nH/cm. Notice that the inductance per unit length is relatively insensitive to
the exact value of h/d.
A useful number to remember is that AWG 22 copper wire placed directly on the top
surface of a printed circuit board with dielectric thickness of .062 inches =1.57 mm (a
standard thickness), and with a ground plane on the bottom of the board, will have an
inductance of approximately 4.8 nH/cm. At 100 MHz the inductive reactance of such a wire,
XL = !L, will be approximately 3 /cm, and at 1 GHz the figure is 30 /cm. For short
wires, the resistance is often small enough to be ignored, however the inductance of the wire
is often significant. In a circuit where impedances are relatively low, the series impedance
of even a short connecting lead may have a significant impact on circuit performance. This
leads to a fundamental rule of RF circuit design - at high frequencies it is important to
keep the length of interconnecting wires and circuit-board traces short in order to minimize
lead inductance. When components are separated by significant distances, interconnections
must be treated as distributed circuit elements, and transmission line models are used to
model the conductors that interconnect components.

3.1.2 Resistors
Several types of resistors are used in RF circuits, including wire-wound, carbon composition,
thick film, and thin film units. Wire-wound resistors consist of a length of lossy wire that is
coiled up to fit into a small package. This type of resistor is seldom used at RF because they
98 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS

have relatively large inductance. Carbon composition resistors consist of a lossy dielectric
material sandwiched between two conducting electrodes. Thick or thin-film resistors consist
of a film of conducting material deposited on an insulating substrate. The film is in contact
with two conductive electrodes to provide a means for connection to external circuitry.
Film resistors are available in cylindrical packages with attached connecting leads and also
as surface mount devices (SMDs). Thick or thin film resistors in a surface mount package (
aka, chip resistors) are the most common type for RF applications.
Over a fairly wide frequency range, resistors can be modeled using the equivalent circuit
shown in Figure 3.2.

L R

Figure 3.2: High frequency equivalent circuit for a resistor without external connecting
leads.

The inductor in the model shown in Figure 3.2 represents the inductance associated with
the current path through the resistor, and the capacitor represents the capacitance between
the two electrodes used to connect the resistor to external circuitry. The unavoidable induc-
tance and capacitance associated with the resistors (and other components) are sometimes
termed parasitic inductance and capacitance. Both the inductance and capacitance of a re-
sistor depend on the geometry and dimensions of the resistor. The inductance is primarily
determined by the length of the current path within the element, and the capacitance is
determined by the size and separation of the contact electrodes, as well as the dielectric
permittivity of the material between the electrodes. Generally, the inductance and capaci-
tance associated with a miniature surface mount resistor package are on the order of 1 nH
and 1 pF, respectively. Significantly higher inductance would be associated with a part in
an axial package with wire leads. Even with very short leads, such a package would have a
typical inductance value on the order of 10 nH. If the component has external connecting
leads attached to the package, then the inductance of the wire leads and the capacitance
between the leads may need to be added to the model, as shown in Figure 3.3 .

Lw /2 L R Lw /2

Cw

Figure 3.3: Model for a resistor with connecting leads. The total inductance of the leads is
Lw and the capacitance between the leads is Cw .

For many purposes, the simpler model of Figure 3.2 can be used even for resistors with
3.1. HIGH FREQUENCY CHARACTERISTICS OF COMPONENTS 99

external connecting leads if the inductance is taken to be the sum of the resistor inductance
and the lead inductance, and the shunt capacitance is taken to be the sum of the package
capacitance and the capacitance between the leads. The following discussion will be based
on the simpler model shown in Figure 3.2.
When the resistance is small ( 100 ), and the frequency is not too high, the series
LR branch of the model has a much lower impedance than the capacitance that shunts it.
The capacitive reactance can then be neglected, leading to the simplified series RL model
in Figure 3.4.

L R

Figure 3.4: Equivalent circuit for a small resistance

For example, consider a 50 resistor with lead inductance L = 10 nH, shunt capacitance
C = 1 pF. Figure 3.5 shows the magnitude of the impedance versus frequency up through
1 GHz calculated using the full model. In this case the series inductance acts to increase
the impedance at high frequencies.

140.0
120.0
100.0
|Z|, 80.0
60.0
40.0
20.0
0.0
0.0 200.0 400.0 600.0 800.0 1000.0
Frequency (MHz)

Figure 3.5: Impedance versus frequency of a 50 resistor. The dotted line shows the
impedance of an ideal resistor.

When the resistance is large ( 100 ) and the frequency is not too high the inductive
reactance will be small compared to R and can be neglected; Figure 3.6 shows the equivalent
circuit in this case.
For example, consider a 10 k resistor with the same inductance and shunt capacitance
as before (L=10 nH, C=1 pF). Figure 3.7 shows the impedance versus frequency calculated
using the full model. Here the shunt capacitance is dominant and tends to short out the
resistance, resulting in a dramatic reduction in the impedance at high frequencies.
For moderate resistance values the parasitic elements tend to have a smaller eect on
the impedance. Figure 3.8 shows the magnitude of the impedance for a 200 resistor with
L = 10 nH, C = 1 pF. In this case neither of the parasitic reactances is significant compared
to 200 and the impedance variation is relatively small up through 1000 MHz.
100 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS

Figure 3.6: Equivalent circuit for a resistor with high resistance.

12000.0
10000.0
8000.0
|Z|, 6000.0
4000.0
2000.0
0.0
0.0 200.0 400.0 600.0 800.0 1000.0
Frequency (MHz)

Figure 3.7: Impedance versus frequency of a 10 k resistor. The dotted line shows the
impedance for an ideal resistor.

350.0
300.0
250.0
200.0
|Z|,
150.0
100.0
50.0
0.0
0.0 200.0 400.0 600.0 800.0 1000.0
Frequency (MHz)

Figure 3.8: Impedance versus frequency of a 200 resistor

3.1. HIGH FREQUENCY CHARACTERISTICS OF COMPONENTS 101

Notice that the impedance of a resistor can be much larger or smaller than the DC resis-
tance, depending on the resistance value and frequency. Until some intuition is developed,
it is a good idea to measure the impedance of any resistor that is being considered for use
in an RF circuit.

3.1.3 Capacitors
Capacitors are constructed by separating two conducting electrodes by an insulating medium
such as air, or a low-loss dielectric material. Loss in the dielectric is modeled as a resistance
(Rp ) in parallel with the intrinsic capacitance. The inductance associated with the current
path through the electrodes and any connected leads appears in series as shown in Figure
3.9. A resistance in series with the inductor (Rs ) models losses in the electrodes and leads.

L Rs Rp

Figure 3.9: Capacitor model including lead inductance (L), dielectric loss resistance (Rp ),
and conductor loss (Rs ).

The model can be transformed into a series RLC model using a parallel to series trans-
formation, as shown in Figure 3.10. where it is assumed that the dielectric loss resistance
is large compared to the capacitive reactance (Rp 1/(!C)). The sum of the ohmic resis-
tance and the transformed dielectric loss resistance is termed the equivalent series resistance
(ESR), denoted here by Resr , i.e.
Resr = Rs + 1/(Rp ! 2 C 2 ).
The ESR of a capacitor is dominated by dielectric loss at suciently low frequencies, whereas
the resistance of the electrodes and leads will dominate at high frequencies.

1
L Rs Rp (!C)2 C

Figure 3.10: Equivalent series RLC circuit of a capacitor, valid where Rp 1/(!C), derived
from the model shown in Figure 3.9.

The Q of a capacitor at any frequency is the ratio of the reactance and the ESR
|X|
Q= , (3.5)
Resr
where the reactance X = !L 1/(!C). The dissipation factor , d, is the inverse of the Q,
i.e
1 Resr
d= = . (3.6)
Q |X|
102 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS

The dissipation factor is also called the loss tangent because it is the complement of the
tangent of the phase angle associated with the capacitor impedance.

Resr
tan = = !CResr
|X|

At frequencies well below the series resonant frequency of a capacitor, the reactance reduces
to X ' 1/(!C) and Q, d, and tan , can be written as follows:

1
Q=
!CReq

d = tan = !CReq .
p
All capacitors will have a series resonant frequency, fs = 1/(2 LC). Above this frequency,
the inductive reactance dominates, and the net reactance is positive. Hence, capacitors are
inductive at frequencies above fs ! For a given package type the inductance will be roughly
independent of the capacitance value, hence the series resonant frequency will be lower for
larger values of capacitance.
As an example, consider a 0.01 F capacitor with total lead length of 1 cm. We know,
from previous discussion, that a typical value of the lead inductance is 10 nH/cm, which gives
a total inductance of 10 nH and a series resonant frequency of 15.9 MHz The magnitude of
the impedance and the reactance versus frequency are shown in Figure 3.11. The impedance
goes to zero at the series resonant frequency in this lossless model. A real capacitors
impedance will fall to a minimum value equal to Resr at the series resonant frequency, fs .
Circuit designers make explicit use of the fact that capacitor impedance is smallest at and
near the series resonant frequency. When a capacitor is used as a DC bias circuit decoupling
element or as an interstage DC-blocking coupling element, it is sometimes possible to choose
the capacitor so that the intended frequency of operation falls near the series resonant
frequency of the capacitor, where the impedance is smallest.

3.1.4 Inductors
3.1.4.1 Air Core Inductors
An approximate formula for the inductance of a close-wound single-layer coil with nonmag-
netic core (e.g., air) is
(rN )2
L= (3.7)
9r + 10l
where
L = inductance in H

N = number of turns

r = radius of coil (inches)

l = length of coil (inches)

Formula 3.7 is accurate to within 1% if l > 0.8r, i.e., if the coil is not too short.
3.1. HIGH FREQUENCY CHARACTERISTICS OF COMPONENTS 103

100.0

10.0
|Z|,
1.0

0.1
1.0 10.0 100.0
(a) Frequency (MHz)

4.0

2.0

X, 0.0
10.0 20.0 30.0 40.0 50.0
2.0

4.0
(b) Frequency (MHz)

Figure 3.11: (a) Log-magnitude of impedance and (b) reactance for a 0.01 F capacitor with
series inductance of 10 nH. The dotted line shows the result for an ideal 0.01 F capacitor.
104 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS

3.1.4.2 Toroidal Inductors

Toroidal inductors are formed by winding a coil of wire around a donut-shaped (toroidal)
core. An important advantage of this type of inductor is its self-shielding property. Typ-
ically, the core material will have high relative magnetic permeability (r 1), and the
magnetic field will be essentially confined to the core. This means that the coil inductance
will be unaected by its physical orientation and also that negligible mutual coupling will
exist between toroidal coils in close proximity. Circuit design can be greatly simplified if
mutual coupling eects can be neglected.
The core material for a toroidal inductor is usually a magnetic material, that is, a material
with magnetic permeability larger than the permeability of free space. Inductor cores for RF
applications are either manufactured from iron powder or ferrite material. The powdered
iron cores consist of small iron particles suspended in an insulating compound. The mixture
is compressed and baked at high temperature to produce a rigid structure. Dierent mixtures
exhibit dierent magnetic permeabilities. Ferrites are ceramic materials with iron oxide
as the dominant constituent. The iron oxide is mixed with nickel, manganese, zinc, or
magnesium. Again, dierent mixtures exhibit dierent magnetic permeabilities. Generally,
ferrite cores oer higher magnetic permeability than iron-powder cores, which results in
fewer turns required to realize a given inductance value. Iron-powder cores generally have
higher saturation flux densities. Saturation flux density refers to the largest flux density
for which the linear relationship B = H holds. A general rule of thumb for high power
RF applications is that the power handling capability of a ferrite core is limited by flux
saturation, while the limiting factor for iron powder is temperature rise.
The important parameters of the core material are its relative magnetic permeability,
cross-sectional area and diameter, and volume resistivity (which determines core loss). Other
considerations for high power applications are the saturation flux density, which determines
the largest magnetic field that the core can support, and the temperature rise resulting
from power dissipation. For small signal applications these need not be considered, but the
temperature stability of the core may need to be considered.
An approximate formula for the inductance of a toroidal winding having cross sectional
area A and eective length lef f is:

A 2
L = N (3.8)
lef f

= A0L N 2

where

= r o (3.9)

A
A0L = (3.10)
lef f
The numerical value of the parameter A0L is the inductance for a single-turn winding.
The core manufacturer will provide the A0L values for each type of core. The inductance of
an N-turn winding is found from equation (3.8). Often, core manufacturers will characterize
their cores by specifying the inductance of a winding with some particular number of turns.
3.1. HIGH FREQUENCY CHARACTERISTICS OF COMPONENTS 105

For example, one manufacturer specifies the inductance per 100 turns for iron powder cores.
If this constant is denoted by AL , it is related to A0L in equation (3.10) by

AL = A0L (100)2 . (3.11)

On data sheets, the units of AL would typically be given as as H/(100 turns).

3.1.4.3 Equivalent circuit model for an inductor

A reasonably accurate equivalent circuit for an inductor consisting of a coil of wire wound
as a solenoid, around a torus, or in a plane includes a series resistance to model the ohmic
loss in the wire and a shunt capacitance to model the distributed capacitance between the
turns of the coil. The inductance of the wire results in voltage dierences between the
dierent parts of the coil. This voltage dierence sets up an electric field in the air and in
any dielectric material near the coil. The eect of this stored electric energy can be modeled
with a capacitance shunted across the terminals of the coil. This eective capacitance is
called the distributed capacity of the coil. The distributed capacity of a coil is dependent on
the geometry of the coil and the number of turns. In general, very closely spaced windings
will have a larger distributed capacity. The equivalent circuit for an inductor is shown in
Figure 3.12. If the inductor is wound on a core with finite conductivity (ferrite or iron-
powder are often used), then it may be necessary to account for core losses by augmenting
the model shown in Figure 3.12 with a resistor in parallel with the capacitor.

L R

Figure 3.12: Equivalent circuit for an inductor

A lower-bound on the coil resistance may be estimated from the AC resistance of an

isolated wire having length equal to the total length of wire in the coil, i.e.
ndD
R = rDC (3.12)
4
where rDC is the DC resistance per unit length of the wire, d is the wire diameter, D is the
coil diameter, n is the number of turns in the coil, and is the skin depth in the wire. A
more accurate estimate of the coil resistance can be obtained by accounting for the proximity
eect. The importance of the proximity eect will depend on the geometry of the coil. The
proximity eect may cause the actual resistance to be significantly larger than the estimate
given in equation 3.12.
The impedance and reactance versus frequency are shown in Figures 3.13 and 3.14 for
a 10 H inductor with (constant) series resistance ofp15 and distributed capacity of 20
pF. Note the parallel resonant frequency fp ' 1/(2 LC) at which the impedance of the
inductor has a relatively large value.
Circuit designers often make use of the fact that an inductor has a very large impedance
near the parallel resonant frequency, and will deliberately use the device at or just below
106 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS

30000.0
25000.0
20000.0
|Z|, 15000.0
10000.0
5000.0
0.0
5.0 10.0 15.0 20.0 25.0
Frequency (MHz)

Figure 3.13: Impedance versus frequency for a 10 H inductor with series resistance of 15
and distributed capacity of 20 pF. The dotted line shows the impedance for an ideal 10
H inductor.

15000.0

10000.0

5000.0

X, 0.0
5.0 10.0 15.0 20.0 25.0
5000.0

10000.0

15000.0 Frequency (MHz)

Figure 3.14: Reactance versus frequency for a 10 H inductor with series resistance of 15
and distributed capacity of 20 pF. The dotted line shows the reactance for an ideal 10 H
inductor.
3.1. HIGH FREQUENCY CHARACTERISTICS OF COMPONENTS 107

this frequency. For example, in many cases an inductor is used to provide a DC bias signal
to a circuit, but it is necessary to isolate the bias supply from the circuit at the operating
frequency. In this case, an inductor with parallel resonant frequency near the operating
frequency may be employed to provide a low impedance to DC and large impedance to
RF signals. In such an application the inductor element is referred to as an RF choke. It
should be noted that the parallel resonant frequency of an inductor depends critically on
its construction. It is dicult to accurately estimate the parallel resonant frequency of an
inductor. In practice it is usually necessary to measure the parallel resonant frequency as
well as the series resistance.
When considering a particular inductor for use in a circuit, the designer needs to be aware
of the parallel resonant frequency as well as the Quality Factor, or Q, of the inductor. The
Q of an inductor is defined to be the ratio of inductive reactance and resistance associated
with the component, i.e.,
|Xs |
Q= (3.13)
Rs
where the impedance of the inductor is Z = Rs + jXs . The higher the Q, the better
the inductor approximates an ideal component. The Q is an important parameter if the
inductor is to be used in a resonant circuit, filter, or matching network.

50.0
45.0
40.0
35.0
30.0
QL 25.0
20.0
15.0
10.0
5.0
0.0
0.0 5.0 10.0 15.0 20.0 25.0
Frequency (MHz)

Figure 3.15: The quality factor (QL = |XS |/RS ) for the inductor whose impedance charac-
teristics are plotted in Figures 3.13 and 3.14. The component is inductive only at frequencies
below its self-resonant frequency (approx. 11.2 MHz in this case), so the Q is not plotted
at frequencies above the self-resonant frequency. The optimum frequency of operation for
an inductor is at or near the peak of the Q vs. frequency curve.

Finally, it should be noted that substantial mutual inductance may exist between induc-
tors that are located in close proximity. This coupling can cause serious problems if it has
not been accounted for in the design of the system, especially when it occurs between the
input and output circuits of an amplifier, since unwanted oscillation may result. Mutual
coupling eects can be minimized by using self-shielding construction (e.g., toroidal induc-
tors) or, in the case of air-wound coils, by orienting the axes of the coils so that they are
perpendicular. A model and theoretical analysis for coupled inductors will be presented in
a later chapter.
108 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS

3.2 References
1. Bowick, Chris and Howard W. Sams, RF Circuit Design, Indianapolis, Indiana, 1982.
2. DeMaw, M. F., Ferromagnetic Core Design and Application Handbook, Prentice Hall,
Inc, 1981.
3. Ludwig, Reinhold and Pavel Bretchko, RF Circuit Design - Theory and Applications,
Prentice Hall, 2000.
4. Terman, Frederick Emmons, Radio Engineers Handbook, McGraw Hill, 1943.
3.3. HOMEWORK PROBLEMS 109

3.3 Homework Problems

1. You are given a black box with two terminals. Suppose that you know that the
box contains a passive circuit that is constructed from 3 elements: a resistor (R),
lossless capacitor (C), and lossless inductor (L). Your task is to figure out how the
elements are connected, and what their values are. You make some measurements of
the impedance of the box Z(f ) = R(f ) + jX(f ). The results of the measurements
are shown in Figure 3.16. Sketch the circuit that is inside of the box. Estimate the

1200
1100
1000
900
800
700
R, 600
500
400
300
200
100
0
2 4 6 8 10 12 14 16 18 20 22 24
Frequency (MHz)
1200

1000

800

600

X, 400

200

200

400
2 4 6 8 10 12 14 16 18 20 22 24
Frequency (MHz)

Figure 3.16: Measured resistance (top) and reactance (bottom).

110 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS

element values. You may assume that the elements (R, L, C) are ideal, i.e., assume
that distributed capacitance across the inductor and the lead inductance of the resistor
and capacitor can be neglected.

2. Same as problem 1, except the measured data is shown in Figure 3.17. Sketch the

400
375
350
325
300
275
250
225
R, 200
175
150
125
100
75
50
25
0
2 4 6 8 10 12 14 16 18 20 22 24
Frequency (MHz)
250
225
200
175
150
125
100
75
50
25
X, 0
25
50
75
100
125
150
175
200
225
250
2 4 6 8 10 12 14 16 18 20 22 24
Frequency (MHz)

Figure 3.17: Measured resistance (top) and reactance (bottom).

circuit that is inside of the box. Estimate the element values. You may assume that
the elements (R, L, C) are ideal, i.e., assume that distributed capacitance across the
inductor and the lead inductance of the resistor and capacitor can be neglected.
3.3. HOMEWORK PROBLEMS 111

3. Same as problem 1, except the measured data is shown in Figure 3.18. Sketch the

250
225
200
175
150
R, 125
100
75
50
25
0
2 4 6 8 10 12 14 16 18 20 22 24
Frequency (MHz)
200

150

100

50

X, 0

50

100

150

200
2 4 6 8 10 12 14 16 18 20 22 24
Frequency (MHz)

Figure 3.18: Measured resistance (top) and reactance (bottom).

circuit that is inside of the box. Estimate the element values. You may assume that
the elements (R, L, C) are ideal, i.e., assume that distributed capacitance across the
inductor and the lead inductance of the resistor and capacitor can be neglected.
4. The circuit shown in Figure 3.19 is used as a model for a realistic resistor or inductor.
It can also be used to model a realistic parallel resonant circuit.

(a) Find an expression for the impedance of this circuit.

112 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS

L R

Figure 3.19: Model for realistic inductor and for parallel resonant circuit

(b) The frequency at which the impedance is purely resistive is called the resonant
frequency. Under what conditions will there be a frequency (> 0) at which
the impedance is purely resistive? Show that if this condition is satisfied, and if
CR2 /L 1, then the resonant frequency is given approximately by:
1 CR2
!o = p (1 ) (3.14)
LC 2L
Note that this frequency is the parallel resonant frequency of the circuit and that
the eect of the resistance, R, is to make the parallel resonant frequency slightly
smaller than that of the lossless (R=0) case.
(c) Continuing from part 4b, assume that the term CR2 /L can be ignored and find
the magnitude of the impedance at the parallel resonant frequency. How does it
depend on R?
(d) For parts 4d and 4e you should start with the exact expression for the impedance.
Find an approximate expression for the impedance valid when ! pLC 1
and
L
R2 C 1. Draw a simplified equivalent circuit that is valid under these conditions.
(e) Find an approximate expression for the impedance valid when ! p1
LC
and
L
R2 C 1. Draw a simplified equivalent circuit that is valid under these conditions.

5. Define the eective capacitance, Ce , of a realistic capacitor having finite lead in-
ductance to be the capacitance of the ideal capacitor that has the same reactance as
the real capacitor. Ce will depend on frequency.

(a) Find an expression for the eective capacitance of a realistic capacitor. Express
your result in terms of the capacitance, C, and the resonant frequency, fo of the
real capacitor.
(b) What is the eective capacitance of a 300 pF capacitor with 20 nH of lead induc-
tance at 10 MHz and at 40 MHz?

6. Figure 3.12 shows the equivalent circuit for an inductor.

(a) Derive an exact expression for the self-resonant frequency of the inductor. The
self resonant frequency is defined to be the frequency (> 0) where the impedance
of the inductor is purely resistive. Express your result in terms of R, L, and C.
When is 2p1LC a good approximation to fp ?
(b) Consider a 500 nH inductor with series resistance R = 1 and shunt capacitance
C = 1 pF. Find the self-resonant frequency of the inductor, fp . Express your
result in MHz.
3.3. HOMEWORK PROBLEMS 113

(c) For the inductor specified in part (b), find the impedance of the inductor at fp .
(d) For the inductor specified in part (b), find the eective inductance, Le , of the
inductor at 0.75 fp . Le is defined to be XL (!)/!, where XL (!) is the reactance
of the inductor at frequency !. Express your result in nH.
114 CHAPTER 3. PROPERTIES OF PASSIVE COMPONENTS
Chapter 4

RLC Networks, Resonance, and Q

4.1 Series RLC Network

Consider the series RLC circuit in a filter configuration where the output voltage is taken
across the resistor, as shown in Figure 4.1. The voltage transfer function is

L C
+

Vin R Vout
-

Figure 4.1: Series RLC circuit as a filter.

Vout (s) R
H(s) = = 1 (4.1)
Vin (s) R + sL + sC

We will consider sinusoidal excitation under steady-state conditions, in which case we are
interested in the frequency response, H(j!):
R
H(j!) = 1 (4.2)
R + j!L 1 ! 2 LC
p
When ! = 1/ LC, the phase shift of the transfer function is zero; this is called the resonant
frequency, !o , of the network and is the frequency at which the inductive and capacitive
reactances are exactly equal in magnitude and, consequently, cancel each other:
1
!o = p (4.3)
LC
The transfer function depends on R, L, and C, but only two parameters are necessary to
specify the characteristics of the function. Define another quantity Qs where:
r
!o L 1 L
Qs = = (4.4)
R R C

115
116 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

The frequency response function can be rewritten in terms of only !o and Qs :

1
H(j!) = (4.5)
! !o
1 + jQs !o !

The parameter Qs is referred to as the series resonant circuit Q. In a subsequent section

it will be shown that the inverse of this quantity tells us what fraction of the total energy
stored in the RLC circuit is dissipated in one complete cycle of the resonant frequency.
The magnitude and phase of the voltage transfer function (Equation 4.5) are plotted as
a function of !/!o in Figure 4.2(a) and (b) for Q=2 and Q=10. The same data is plotted
with a logarithmic frequency axis in Figure 4.3(a) and (b).

1.0
Q=2
Q=10
0.8

0.6
|H(j!)|
0.4

0.2

0.0
1 3 5 7 9
(a) !/!o
90 Q=2
Q=10
70
50
30
10
arg[H(j!)]
10
1 3 5 7 9
30
50
70
90
(b) !/!o

Figure 4.2: (a) Magnitude and (b) phase of the voltage frequency response for Q=2 (solid
line) and Q=10 (dotted line).
4.1. SERIES RLC NETWORK 117

1.0
Q=2
Q=10

|H(j!)| 0.5

0.0
0.1 1.0 10.0
(a) !/!o
90 Q=2
Q=10
70
50
30
10
arg[H(j!)]
10
0.1 1.0 10.0
30
50
70
90
(b) !/!o

Figure 4.3: (a) Magnitude and (b) phase of the voltage frequency response for Q=2 (solid)
and Q=10 (dotted) vs. !/!o with a logarithmic frequency axis.
118 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

The RLC filter has a bandpass characteristic. The separation between the half-power
(-3 dB) frequencies is often used to specify the bandwidth of a filter. The 3 dB bandwidth
of the filter can be found by determining the dierence between the frequencies where
|H(j!)| = 0.707. Denote the lower and upper -3 dB frequencies by !1 and !2 as shown in
Figure 4.4. Then ! = (!2 !1 ) is referred to as the 3 dB bandwidth of the filter. It is
left as an exercise to show that
!o
! = (!2 !1 ) = (4.6)
Qs

The bandwidth of a series RLC filter is inversely proportional to the Qs of the circuit.

1.0

0.707

1 0 2

Figure 4.4: 3dB bandwidth of filter

4.1.1 Example - Series RLC circuit as a filter.

Use a series RLC circuit to couple a voltage source with negligible source resistance to a 50
load as shown in Figure 4.5. The circuit should have a center frequency of 5 MHz and a
3 dB bandwidth of 100 kHz. The bandwidth and center frequency determine Qs :

L C

50

Figure 4.5: Circuit with 5 MHz center frequency and a 3 dB bandwidth of 100 kHz

fo 5 MHz !o L 2(5 106 )L

Qs = = = 50 = = (4.7)
f 100 kHz R 50
so
50 50
L= = 79.6 H
2(5 106 )

1
C= = 12.7 pF
!o2 L
4.2. PARALLEL RLC 119

Iin C L Rp Vout
-

Figure 4.6: Parallel RLC as a filter

4.2 Parallel RLC

A parallel RLC circuit being driven by an ideal current source is shown in Figure 4.6. In
this application the input current and output voltage are related by an impedance function,
i.e.,

Vout (s)
= Z(s) (impedance) (4.8)
Iin (s)
1
1 1
Z(s) = + + sC
Rp sL
For sinusoidal steady-state excitation
Rp
Z(j!) = (4.9)
! !o
1 + jQp !o !

where

1
!o = p (4.10)
LC
Rp
Qp = (4.11)
!0 L
r
C
= Rp
L
This transfer function has exactly the same form as that of the series RLC circuit except
for the scaling factor, Rp . Note, however, that the Q is defined dierently for the parallel
RLC. As before, the 3dB bandwidth is
!o
!= (4.12)
Qp

4.2.1 Unloaded vs Loaded Q of RLC circuits

If the source has a non-negligible impedance as shown in Figure 4.7 then
Vout Rp kRS
= !o (4.13)
Iin 1 + jQ0p ( !!o ! )
120 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

Iin RS C L Rp Vout
-

Figure 4.7: Parallel RLC filter driven by a source with finite source impedance.

where
0 Rp kRS
Qp = (4.14)
!o L
Compared to the case with infinite source impedance, the finite source impedance causes the
Q to be reduced and, hence, the bandwidth to be increased. It is common practice to call
the Q of the resonant circuit alone (either series or parallel RLC) the unloaded Q, and the
Q of the composite circuit, which includes the source resistance and any other resistances
that are external to the LC resonator, the loaded Q. The loaded Q is always smaller than
the unloaded Q.
In the case of a series resonant circuit with resistance R driven by a source with finite
impedance RS , the loaded Q becomes
!o L
Q0s = .
RS + R
Again, the loaded Q is smaller than the unloaded Q.

4.3 More on Q
The previously defined Qs (Qs , Qp ) describe the frequency selectivity of the simplest type
of resonant RLC networks. The term Q has a more general interpretation which can be
applied to any type of system that contains energy storage elements and dissipation. The
general definition of Q for a system is
Maximum instantaneous stored energy
Q = 2 (4.15)
Energy dissipated per cycle
Maximum instantaneous stored energy
= 2f
Time average power dissipated
This definition can be applied to resonant and nonresonant circuits. If this energy-
based definition is applied to resonant second-order RLC circuits the result is compatible
with the (Qs , Qp ) defined in the previous section. In principle, the definition could be
applied to higher-order circuits, e.g, circuits with more than one inductor and capacitor,
however this is usually not very useful. On the other hand, higher order circuits are often
constructed by combining second-order circuits. In such cases, it is useful to characterize
each of the constituent second-order resonant circuits by a Q. This is common practice
in filter-design, where multiple LC resonators are coupled to form a more complex filter.
The energy definition is also commonly applied to characterize lossy inductors or capacitors
which, by themselves, are non-resonant.
4.3. MORE ON Q 121

We shall first show that the energy definition is consistent with the resonant-circuit
Qs that has already been defined for series RLC circuits. Consider the series RLC with
sinusoidal excitation as shown in Figure 4.8.

Iin L C
+

Vin R Vout
-

Figure 4.8: Series RLC circuit for calculation of Q

At resonance,
Vout = Vin (4.16)
Iin = Vin /R (4.17)
The fact that the voltage across the capacitor is 90 degrees out of phase with the current
through the inductor means that the current maximizes at the time when the capacitor
voltage is zero. At that instant in time, all of the stored energy resides in the inductor and
the magnitude of the current phasor (which is the peak current magnitude) can be used to
calculate the total stored energy. The stored energy is:
2
1 2 1 |Vin |
Emax = L |Iin | = L 2 (4.18)
2 2 R
Alternatively, at the time instant when the capacitor voltage is maximum then the current
in the system is zero and all of the stored energy resides in the capacitor. The magnitude of
the capacitor voltage phasor can then be used to calculate the stored energy at that time:
1 1 Iin 2 1
Emax = C|Vcap |2 = C| | = L|Iin |2 (4.19)
2 2 !o C 2
The stored energy comes out the same either way. Actually, it can be shown that the
total stored energy in this driven resonant RLC circuit is a constant, so that the maximum
instantaneous stored energy is equal to the energy stored at any instant of time. The
time-averaged power delivered to (and dissipated in) the network is:

1
Pavg = Re[Vin Iin ] (4.20)
2
1 2
= |Vin | /R
2
Then using the energy definition of Q:
Emax !o L
Qs = 2fo = (4.21)
Pavg R

This result is the same as the definition for the series resonant Q that was given earlier.
122 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

Iin L

Vin R

Figure 4.9: Nonresonant RL circuit.

It is also common to apply the definition to a nonresonant circuit, e.g., consider an RL

circuit as in Figure 4.9.

Vin
Iin = (4.22)
R + j!L
In this case, the inductor is the only energy storage element. Because the current through the
inductor varies sinusoidally, the stored energy will oscillate between zero and the maximum
value given by:

1 2
Emax = L |Iin | (4.23)
2
1 2 1
= L |Vin | 2
2 rs + ! 2 L 2

The time-averaged power delivered is

1
Pavg = Re[Vin Iin ] (4.24)
2

1 2 1
= |Vin | Re
2 rs j!L
1 2 rs
= |Vin | 2
2 rs + ! 2 L 2

Therefore
Emax !L
QL = 2f = (4.25)
Pavg rs

In this case ! can be any frequency. This type of Q will be called the component Q. In this
example, the RL circuit could represent a model for a lossy inductor. The component Q of
an inductor evaluated at some frequency ! can be thought of as the resonant Qs or Qp that
results if a lossless capacitor is added to form a resonant circuit at the frequency !.
The concept of component Q is often used to describe the properties of arbitrary circuit
elements at a particular frequency. For example, if an arbitrary circuit element can be
represented by a series impedance Z = Rs + jXs at some frequency, then applying the
definition of Q to that component yields Q = |Xs |/Rs as illustrated in Figure 4.10(a). For
a parallel representation of a circuit branch the component Q is as shown in Figure 4.10(b).
4.4. SERIES-TO-PARALLEL TRANSFORMATIONS 123

RS
|XS | RP
Q= RS RP jXP Q= |XP |
jXS

(a) (b)

Figure 4.10: Definition of Q applied to a branch represented in (a) series form and (b)
parallel form.

4.4 Series-to-Parallel Transformations

Any circuit element has both a series and a parallel representation. Since the energy storage
and dissipation properties of the element do not depend on how we represent it, the Q is
the same for either representation. The component Q concept is useful for series-to-parallel
impedance transformations. For example, suppose the series impedance representation for
a circuit element is known at a particular frequency as shown in Figure 4.11(a).

RS
|XS |
Q= RS Z = RS + jXS RP jXP
jXS

(a) (b)

Figure 4.11: (a) Series impedance representation for a circuit at a particular frequency
and (b) equivalent parallel representation having the same impedance as the original series
branch.

The circuit Q is given by

|Xs |
Q= (4.26)
Rs
The equivalent parallel representation for the circuit element is shown in Figure 4.11(b). The
equivalent parallel resistance and reactance are easily found after equating the impedances
of the two models:

Rp = Rs (1 + Q2 ) (4.27)

1
Xp = Xs 1+ 2
Q

Notice that the Q of the equivalent parallel circuit is Rp /|Xp | which is equal to Q. Thus,
the Q of the equivalent parallel representation is the same as that of the original series
124 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

representation (and vice versa). A useful simplification results if Q 1, in which case

Rp ' Rs Q 2 (4.28)

Xp ' Xs

Clearly, a complex impedance that is represented in parallel form, i.e. as Zp = Rp ||jXp

can be transformed to series form, Zs = Rs + jXs by defining the Q of the parallel repre-
sentation (Q = Rp /|Xp |) and then calculating Rs and Xs from

Rp
Rs = (4.29)
1 + Q2

Xp
Xs = (4.30)
1 + Q12

4.4.1 Example - Series to parallel conversion

A 1 H inductor has a component Q of 100 at 10 MHz. Find a parallel representation for
the inductor. At 10 MHz, !L = 2(107 ) (10 6 ) = 62.8 . So QL = 100 = !L rs = rs .
62.8

0.628

j62.8

Figure 4.12: Representation of 1 H inductor with a component Q of 100 at 10 MHz

Therefore, rs = 0.628 . Since QL 1,

Xp ' Xs = 62.8 (4.31)

Rp ' Q2 rs = 104 (.628) = 6.28 k

The equivalent network at 10 MHz is shown in Figure 4.13. It is important to remember

that this equivalent circuit is only valid at 10 MHz.
The following example illustrates how a reactance in series with a resistor can form the
basis of an impedance-transforming network that transforms the resistance into a dierent
value.

4.4.2 Example - Impedance transformation

Design a lossless network to transform 50 to 300 using a capacitor and an inductor.
We can start by putting an inductive reactance in series with the 50 resistor, as shown
on the left in Figure 4.14. The equivalent parallel representation, on the right in Figure
4.4. SERIES-TO-PARALLEL TRANSFORMATIONS 125

6.28 k j62.8

Figure 4.13: Equivalent network at 10 MHz.

jXL
XL
Q= 50 Rp = 50(1 + Q2 )
50

1
Xp = XL (1 + Q2 )

Figure 4.14: Inductive reactance in series with a 50 resistor.

4.14, shows that the parallel resistance is larger than 50 by the factor 1 + Q2 . Setting
Rp = 300 = 50(1 + Q2 ) yields the Q of the circuit (Q = 2.24) and therefore the value of the
series reactance, XL = 2.24(50) = 112 . So far, it is known that the series circuit shown on
the left in Figure 4.15 is equivalent to the parallel circuit shown on the right. Now a shunt

j112 300

50

(a) (b) j112(1 + 51 ) = j134.4

Figure 4.15: (a) Series circuit and (b) equivalent parallel circuit.

capacitor can be added to cancel the inductive reactance, as shown in Figure 4.16. Going
back to the original, series, representation yields the final solution for the lossless network
that transforms a 50 resistance into a 300 resistance as in Figure 4.17. This is an
example of an L-section matching network which will be discussed in more detail in chapter
6. It provides an impedance match between 50 and 300 only at the design frequency.
The circuit topology of this matching network is of the lowpass type; frequencies much
126 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

300 300

j134.4 j134.4

Figure 4.16: Shunt capacitor added to cancel inductive reactance.

j112

300 j134.4 50

Figure 4.17: Lossless network to transform a 50 resistance into a 300 resistance.

lower than the design frequency will be transferred to the load, while frequencies much
higher than the design frequency will not. Another solution to this matching problem exists
and has a highpass topology as seen in Figure 4.18. Recall that the series resistance is

j112

300 j134.4 50

Figure 4.18: Highpass circuit topology.

stepped up to a larger parallel resistance by the factor 1 + Q2 , which is independent of the

sign of the series reactance. Thus, the highpass solution could be derived by starting by
adding a capacitive reactance in series with the 50 resistor.
4.4. SERIES-TO-PARALLEL TRANSFORMATIONS 127

4.4.3 Single-resonator filters

The impedance-transforming properties of a reactance in parallel or series with a resistance
can be exploited to design simple RLC filters for a desired bandwidth when the source and
load resistances and inductor (or capacitor) value can not be adjusted to produce the desired
Qs,p . For example, suppose that the source and load resistances are each 50 and a single
inductor is available with reactance 100 at the desired filter center frequency. In this
case, a simple series or parallel RLC bandpass filter configuration will have Qs,p determined
by the source and load resistance and the inductor value, and the bandwidth cannot be
specified independently. If the inductor is used to implement a series LC filter as shown in
Figure 4.19(a), the total series resistance is 100 and the resulting filter has Qs = 1. In

50 j100

j100
50
(a)

50 j100 jXC
+
Vs 50 Vo

(b) jX jX

50 jX
2 |X|2
1+ 50 2 1+
|X| j100 502

jX jXC 50
|X|2 2
1+ 1+ 50 2
502 |X|
(c)

50 j100 j70
+
Vs 50 Vo

(d) j16.7 j16.7

Figure 4.19: (a) With 50 source and load resistances and inductor reactance 100 , Qs = 1,
and the bandwidth cannot be specified independently. (b) Shunt capacitors with reactance
X (X < 0) have been added to transform the source and load resistances to a smaller value
as shown in (c). The values given in (d) set the Qs of the equivalent series RLC circuit
(shown in (c)) to Qs = 10.

Figure 4.19(b), shunt capacitors with impedance jX (X < 0) have been added to transform
the source and load resistance to smaller values, as shown in Figure 4.19(c). The network
shown in Figure 4.19(c) is similar to a series RLC filter and will have a similar frequency
128 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

response function. The resonant Q of this network can be approximated by Qs = 100/(2R0 ),

where R0 = 50/(1 + (50/|X|)2 ). Hence, X can be chosen to set Qs to any desired value.
Suppose that the desired fractional bandwidth of the filter is f /fo = 0.1 (10%). Then the
required loaded Qs = 0.1
1
= 10 and the total series resistance should be set to 10 , i.e.

2(50)
10 = 502
,
1 + |X| 2

which has the solution |X| = 50/3 = 16.7 . Therefore, the impedance of the shunt coupling
capacitors should be set to j16.7 . The shunt capacitors are transformed to series reac-
tances of j 1+(16.7/50)
16.7
2 = j15 . At the resonant frequency, !o , the total series capacitive
reactance must be equal to 100 in order to resonate with the inductor, so the capacitive
reactance XC = 70 . The final design is shown in Figure 4.19(d).
If the desired center frequency is fo = 500 MHz, then the values of the inductor, shunt
capacitors, and series capacitor are L = 31.8 nH, Cshunt = 19.1 pF, and C = 4.55 pF,
respectively. The frequency response of the network with these values is shown by the solid
line in Figure 4.20. For comparison, the dashed line shows the frequency response of a
series RLC filter (i.e., the topology shown in 4.19(a)) with Qs = 10. The series RLC filter
would need an inductor and capacitor each having reactance 1000 to produce a filter with
Qs = 10. Notice that the two curves correspond closely near the peak of the response,
illustrating the validity of approximating the network shown in Figure 4.19(c) as a simple
series RLC network near the resonant frequency of the network.

0.5

0.4

0.3

| VVos |
0.2

0.1

0 200 400 600 800 1000

Frequency (MHz)

Figure 4.20: Solid line shows the frequency response of the filter shown in Figure 4.19(d),
designed for a center frequency of 500 MHz. The dashed line is a plot of the function
0.5
| 1+jQs (!/! o !o /!)
|. The dashed line is the frequency response for a series RLC filter, as
shown in Figure 4.19(a), but with the inductor and capacitor impedances equal to j1000
at fo = 500 MHz. Close correspondence between the solid and dashed lines near the resonant
frequency illustrates the validity of approximating the network shown in Figure 4.19(b)-(d)
as a series RLC network near fo .
4.5. APPLICATION EXAMPLE - QUADRATURE DEMODULATOR FOR FM 129

4.5 Application Example - Quadrature demodulator for

FM
A common application of the parallel RLC resonant circuit is found in the so-called quadra-
ture demodulator for frequency modulated signals. This type of detector is implemented
in a number of receiver-on-a-chip integrated circuits. The block diagram of a quadrature
demodulator is shown in Figure 4.21. The parts within the dashed box are implemented
on-chip. Typically, the multiplier is implemented with a Gilbert cell and the capacitance of
C 0 is a relatively small value, which is easily implemented in an integrated circuit. The res-
onant RLC network is implemented o-chip, and is often sold under the name quadrature
coil.

Low-pass filter

vin vout
C
v
2fo
-1/2
L C R o=(LC)
Qp=oRC

Figure 4.21: Quadrature demodulator for frequency modulated signals. The Qp of the
parallel resonant circuit determines the sensitivity and the bandwidth of the detector.

The principle of operation behind the quadrature demodulator is as follows. The fre-
quency modulated input signal is typically provided by the output of the last IF stage of
a superhet receiver. Usually, the signal will have passed through a limiter, so that the
amplitude is constant. The input voltage can be written as:
vin (t) = A cos[(!o + !)t] (4.32)
where ! represents the instantaneous frequency deviation from the carrier frequency, !o .
The deviation will be a function of time, in general, but since time variations of ! are very
slow compared to the period of !o , we will employ a quasi-static analysis and treat ! as
a (slowly varying) constant. In most applications, the carrier frequency will be equal to the
last IF. The RLC circuit is tuned to resonance at !o . Thus, at !o the impedance of the RLC
circuit is equal to R. If the on-chip capacitance, C 0 , is small so that !o1C 0 R, then the
voltage v 0 will be shifted in phase by /2 relative to vin . Thus v 0 and vin are in quadrature
when the input frequency is equal to !o . When the frequency of the input signal deviates
from !o , the RLC circuit looks capacitive for positive deviations, and inductive for negative
deviations. This upsets the nominal quadrature relationship between vin and v 0 and, for
small frequency deviations, causes the phase dierence between vin and v 0 to vary around
the nominal value of /2 in proportion to the instantaneous frequency deviation, !. The
multiplier/low-pass filter functions as a phase-detector and provides an output voltage that
is proportional to the o-quadrature phase dierence between vin and v 0 . Therefore, for
small frequency deviations, the output voltage is proportional to the frequency deviation.
130 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

For analytical analysis of the quadrature demodulator circuit, it is convenient to assume

that the input impedance of the multiplier circuit is high, so that loading of the RLC circuit
can be neglected. (Alternatively, one can assume that the finite input resistance of the
multiplier has been lumped into R.) Then, using upper-case symbols to denote phasors, the
voltage V 0 can be written in terms of the phasor input voltage, Vin as:

Zp
V0 = 1 Vin (4.33)
Zp + j!C 0

where
R
Zp = !o (4.34)
1 + jQp ( !!o ! )

So
R
V0 = 1 !o Vin (4.35)
R+ j!C 0 (1 + jQp ( !!o ! ))

If C 0 is small enough so that R 1

!C 0 then

j!RC 0
V0 ' !o Vin (4.36)
1 + jQp ( !!o ! )

The transfer function can be re-written in terms of magnitude and phase as

V 0 = B(!)ej(!) Vin

where
!RC 0
B(!) = q (4.37)
!o 2
1 + Q2p ( !!o ! )

and
1 ! !o
(!) = /2 tan [Qp ( )] (4.38)
!o !
Thus, the time-domain voltage, v 0 (t) is given by (with ! = !o + !):

v 0 (t) = AB(!) cos[!t + (!)] (4.39)

The multiplier forms the product vin v 0 , so the signal at the output of the multiplier is:

vin (t)v 0 (t) = A2 B(!) cos[!t] cos[!t + (!)] (4.40)

Taking the nominal phase-shift, /2, in the expression for (!) into account, we can write
! !o
vin (t)v 0 (t) = A2 B(!) cos[!t] sin[!t tan 1
[Qp ( )] (4.41)
!o !

The sine-cosine product yields a double frequency term, and a dierence frequency term.
The lowpass filter removes the double frequency term, so the output of the system can be
written:
1 ! !o
vo = LPF[vin (t)v 0 (t)] = A2 B(!) sin[tan 1 (Qp ( ))] (4.42)
2 !o !
4.5. APPLICATION EXAMPLE - QUADRATURE DEMODULATOR FOR FM 131

Employing the trigonometric identity sin[tan 1 x] = p1+x

x
2
the output voltage can be writ-
ten as
!o
1 Qp ( !!o ! )
vo = A2 B(!) q . (4.43)
2 1 + Q2 ( ! !o 2
)
p !o !

Inserting the expression for B(!), we find:

!o
1 2 Qp ( !!o ! )
vo = A !RC 0 !o 2 . (4.44)
2 1 + Q2p ( !!o ! )

Now, with ! = !o + !

! !o !o + ! !o 2 ! !2
= = + . (4.45)
!o ! !o !o + ! !o + ! !o (!o + !)

For small frequency deviations such that ! !o :

! !o 2 !
' . (4.46)
!o ! !o

So, for small frequency deviations we have

1 2 Qp 2!o! 1 2 C 0 Qp !o
2 !
vo ' A !o RC 0 2 = A Q p 2 . (4.47)
2 1 + Q2p 4 !!
2
2 C 1 + Q2p 4 !!
2
o o

This expression has the form:

x
vo
1 + x2
where x = 2Qp f /fo . This function is plotted in Figure 4.22.
Some quadrature detector implementations use a multiplier with limiting inputs, i.e.
one or both of the inputs is driven into saturation so that the multiplier output is insensitive
to changes in the amplitude of the input signals. If the v 0 input is saturated, then the output
becomes independent of the amplitude of v 0 . In this case, equation 4.40 must be modified
by replacing the term A2 B(!) with a constant. In this case, the output can be written:
!o
Qp ( !!o ! )
vo = K q . (4.48)
!o 2
1 + Q2p ( !!o ! )

Using the same approximations that were used before, we obtain:

Qp 2!o!
vo ' K q 2
(4.49)
1 + Q2p 4 !!
2
o

which has the form:

x
vo p
1 + x2
with x = 2Qp f /fo . This function is plotted in Figure 4.23.
132 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

1.0
0.8
0.6
0.4
0.2
x
1+x2 0.0
0.2
0.4
0.6
0.8
1.0
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
2Q f
x = fpo

Figure 4.22: Detector output as a function of frequency deviation. This curve is referred to
as the demodulators S-curve. Notice that the region where the output is approximately
proportional to the frequency deviation is limited |2Qp f /fo | 1. The dotted line shows
an ideal linear response with the same slope as the actual response at x = 0.

1.0
0.8
0.6
0.4
0.2
p x
1+x2 0.0
0.2
0.4
0.6
0.8
1.0
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
2Q f
x = fpo

Figure 4.23: Output voltage verses normalized frequency deviation for a quadrature demod-
ulator employing a saturated input for v 0 . The dotted line shows an ideal linear response
with the same slope as the actual response at x = 0.
4.6. REFERENCES 133

In either case (ideal multiplier, or multiplier with saturated inputs), if the frequency
deviation is small enough, i.e. if 2Qp !/!o 1, then the output voltage is proportional
to the instantaneous frequency deviation:
Qp
vo f, f fo and 2Qp f fo (4.50)
fo
The coecient in front of f determines the sensitivity of the demodulator. It is apparent
that to obtain larger output voltage for a given deviation one would like to choose a higher
Qp for the resonant RLC circuit. On the other hand, the maximum Qp that can be employed
is limited by the requirement that 2Qp fmax /fo 1 (where fmax is the peak frequency
deviation of the modulated input signal). Recall that the -3 dB bandwidth of an RLC filter
can be written in terms of the Q, i.e. BW = fo /Qp , so the constraint on Qp can be rewritten
in terms of a constraint on the bandwidth of the RLC circuit: BW 2 fmax . In other
words, the -3 dB bandwidth of the RLC circuit should be large compared to the deviation
of the input signal in order for the demodulator to have a linear response.
A useful practical constraint for design purposes is to require that the demodulators
response characteristic should not deviate from perfect linearity by more than 1% over
the whole range of deviation, i.e. for fmax < f < fmax . If the demodulator is
implemented with an ideal multiplier, this requirement leads to x < 0.100. The largest Qp
that will satisfy this constraint is:
fo
Qp = 0.1 (4.51)
2 fmax
If a multiplier with saturated inputs is used to implement the detector, then the departure
from linearity will be held to the 1% limit if x < 0.142. The largest Qp that satisfies this
constraint is:
fo
Qp = 0.142 (4.52)
2 fmax
The extent to which the demodulators characteristic departs from linearity will determine
the amount of distortion in the demodulated signals waveform. For some applications,
such as demodulation of frequency-shift-keyed (FSK) data signals, low distortion is not
particularly important. In such cases a larger deviation from linearity can be tolerated in
return for larger output voltage from the detector.

4.6 References
1. Terman, Frederick Emmons, Radio Engineers Handbook, McGraw Hill, 1943.
134 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

4.7 Homework Problems

1. The circuit shown in Figure 4.24 is usually a good model for a parallel LC circuit
implemented with real, lossy components. The element values are C = 800 pF, L =
15 H, r = 1 , R = 10 k. Use series to parallel transformations to transform this
circuit into an equivalent parallel RLC circuit and find the approximate resonant
frequency and Qp of the equivalent circuit.

L
R C
r

Figure 4.24: Resonant circuit formed from lossy capacitor (modeled by R and C) and a
lossy inductor (modeled by r and L).

2. Two series RLC circuits are connected in series to form a new resonant circuit. Denote
the elements of the individual resonant circuits by (R1 , L1 , C1 ) and (R2 , L2 , C2 ),
respectively. The series resonant frequency of each of the two circuits is the same and
is denoted by !o . The Qs of the two circuits are Q1 and Q2 .

(a) What is the resonant frequency of the series combination of the two circuits?
(b) Find an expression for the Q of the overall circuit. Express your result in terms
of Q1 , Q2 , R1 and R2 only.
(c) Is the statement below TRUE or FALSE? You must justify your answer to receive
full credit.

min(Q1 , Q2 ) Q max(Q1 , Q2 ) (4.53)

3. One property of a parallel resonant circuit is that the current in the reactive compo-
nents can be much larger than the applied current. This must be considered when
choosing the current ratings of components such as capacitors and inductors (espe-
cially in high power devices such as transmitters). Show that the peak current through
the inductor or capacitor in a parallel resonant circuit is Qp Ii at resonance, where Ii
is the current supplied to the entire resonant circuit

4. One property of a series resonant circuit is that the voltage across the reactive com-
ponents can be much larger than the applied voltage. This must be considered when
choosing the voltage ratings of components such as capacitors and inductors (espe-
cially in high power devices such as transmitters). Show that the peak voltage across
the capacitor in a series resonant circuit is Qs Vi at resonance where Vi is the voltage
across the entire resonant circuit. Note that the peak voltage across the inductor will
be the same.
4.7. HOMEWORK PROBLEMS 135

5. Consider the circuit shown in Figure 4.25. The current source has constant amplitude
and frequency fc , and it drives a bandpass filter consisting of a lossy inductor in par-
allel with a variable capacitor and a resistor. You may assume that any capacitance
associated with the inductor has been incorporated into the variable capacitor indi-
cated in the schematic. The variable capacitor C can be set to any value in the range
36 365 pF, and r = 10 , R = 100 k. Suppose that the frequency of the current

L
Io cos 2fct C R Vout
r

Figure 4.25: LC filter with variable center frequency.

source can be adjusted to any frequency in the range 540-1700 kHz. For a given value
of the source frequency, fc , the variable capacitor will be tuned to maximize the out-
put voltage. Approximate the filter as a parallel RLC circuit to answer the following
questions:

(a) Specify a single value of L that would allow the variable capacitor to tune the
filter (i.e. maximize the output voltage) to any frequency in the range 540-1700
kHz. Give your result in H.
(b) With the value of L that was determined in part (a), determine the approximate
3dB bandwidth of the filter when fc = 540 kHz and C is adjusted to maximize
the output voltage.
(c) Same as part (b) but for fc = 1700 kHz.

6. The circuit in Figure 4.26 is a model for the operation of a ferrite loopstick antenna
that is commonly employed in AM broadcast band radios. The antenna is tuned to
resonance by a capacitor C which is usually adjustable to allow the circuit to cover
the entire broadcast band. This circuit also performs the function of the preselector.
The voltage source Vs represents the emf induced in the coil as a result of an incident
electromagnetic wave with frequency !. The resistance r represents the losses in the
coil and R represents the input impedance of the following stage.

(a) Find an exact expression in terms of r, R, L, and C for the frequency where the
output voltage is a maximum (the resonance frequency).
(b) Suppose that Vs = (1 mV ) cos[ 2 fc t ], L = 100H, R = 100k , and r = 6 .
Find the range of values that C must cover in order for the circuit to tune the AM
broadcast band (540-1700 kHz). Note: For this purpose you need an expression
for the value of C that maximizes the voltage response at a given frequency. An
approximate analysis is acceptable, but be sure to carefully state and justify your
assumptions.
(c) Now suppose that the source frequency, fs , is swept from 540 to 1700 kHz. The
zero-to-peak value of the source voltage is held constant at 1 mV as the frequency
136 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

L
r C R Vout
Vs

Figure 4.26: Ferrite loopstick antenna model.

is swept. Also suppose that the circuit is tuned to follow the frequency of the
source so that the output voltage is always maximized. Plot the output voltage
as a function of frequency.

7. Consider a parallel resonant circuit that is used as the preselector of an AM broadcast

band receiver. The inductor used in the resonant circuit is near the peak of its Q-
versus-frequency curve in the band of frequencies from 550-1600 kHz, and hence
the QL of the inductor can be assumed to be constant over the band. Assume that
QL = 100 and that the capacitor is lossless.

(a) What will the 3 dB bandwidth of the preselector be when the preselector is tuned
to 550 kHz? How about 1600 kHz?
(b) Now consider a dierent situation where the inductor is chosen such that the skin
eect and distributed capacitance can be neglected for the frequencies of interest,
i.e., the inductor is operated at frequencies well below the peak of its Q-versus-
frequency curve. Suppose the inductor has QL = 50 at 550 kHz. What will the
3 dB bandwidth of the preselector be at 550 and 1600 kHz?

8. In a series RLC circuit that is resonant at 1150 kHz it is found that when the frequency
diers from resonance by 15 kHz, the current drops to 0.53 of the current at resonance,
for the same applied voltage. Determine the QS of the circuit.
9. Consider the circuit shown in Figure 4.27 where RS = RL = 50 , L = 48 H, C = 10
pF:

RS
+
Vs C L RL Vo
-

Figure 4.27: Series resonant circuit

(a) Find the resonant frequency where maximum power would be delivered to the 50
load resistor. Express your result in MHz.
4.7. HOMEWORK PROBLEMS 137

(b) Suppose that the time-averaged power delivered to the 50 load resistor at
resonance is 1000 W. Find the peak value of the voltage across the capacitor.
(c) Approximately how far o of resonance would the source frequency need to be
moved in order to cause the average power delivered to the load resistor to drop
by 6 dB? Express your result in kHz.

Note: For small frequency shifts, !, around the resonant frequency, !o , the following
approximation is useful:

!o + ! !o !
2 (4.54)
!o !o + ! !o

10. Consider the circuit shown in Figure 4.28 and its Thevenin equivalent circuit.

C L Rth

Vs R r Vth

Figure 4.28: Series resonant circuit with Thevenin equivalent circuit.

(a) If Vs = 2 V, R = 1000 , C = 1000 pF, L = 10 H, r = 1 , find the (non-zero)

frequency, !o , where the indicated Thevenin equivalent circuit would be valid. In
other words, find the frequency where the Thevenin impedance is purely resistive.
Clearly state any assumptions or approximations that you make.
(b) Find Rth and Vth at the frequency found in part (a). For Vth , specify magnitude
and phase (in degrees).

11. Shown in Figure 4.29 are two dierent ways of coupling a voltage source to a load
resistor using a resonant circuit for the purpose of providing a bandpass output voltage
response. The values are: Vs = 10V (peak), Rs = 500 , RL = 500 , L = 1 H, C =
500 pF.
Compute the following for each circuit:

(a) The output voltage at resonance.

(b) The 3dB bandwidth of the transfer function Vo /Vs . Express your result in MHz.
(c) The upper and lower 3 dB frequencies, i.e., give the actual frequencies (in MHz).
The dierence between these frequencies will be the 3dB bandwidth that you
found in part 11b.

12. Consider the capacitive transformer drawn in Figure 4.30.

Under certain conditions this circuit will approximately transform the resistance R to
a new resistance n2 R in parallel with an eective capacitance, Cp , where n = 1 + C
C2 .
1
138 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

RS
+
Vs C L RL Vo
-
(a)

RS
+
Vs C L RL Vo
-
(b)

Figure 4.29: Voltage source coupled to load resistor with (a) parallel LC circuit and (b)
series resonant circuit to provide a bandpass voltage transfer function Vo /Vs .

C2
Z
C1 R

Figure 4.30: Capacitive transformer

4.7. HOMEWORK PROBLEMS 139

(a) Under what condition(s) will this occur?

(b) Under this condition, give an expression for Cp .
13. The circuit shown in Figure 4.31 is a capacitive transformer with resonating inductance
L. Suppose that

R = 50
C1 = 3183 pF
C2 = 3183 pF

C2
Z L
C1 R

Figure 4.31: Resonant capacitive transformer

Use parallel-series and series-parallel transformations, with appropriate approxima-

tions, and find:
(a) The inductance, L, required to resonate the circuit at 10 MHz. At resonance, the
impedance Z will be purely real.
(b) The input impedance, Z at 10 MHz.
(c) The Q of the circuit can be approximated by the Qp of the equivalent parallel
RLC circuit (at 10 MHz). Find the approximate Q.
To save time, note that the reactances of C1 and C2 have magnitude 5 at 10 MHz.
14. Consider in Figure 4.32 the small signal model for a tuned-output amplifier:

+ +
Vin L C R Vout
- gm Vin -

Figure 4.32: Small signal model for tuned-output amplifier.

Suppose that N of these stages are cascaded. Show that the 3dB bandwidth of the
cascade will be given by
140 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

p
BW = BW 1 21/N 1 (Hz) (4.55)
where BW 1 is the 3 dB bandwidth of a single stage
f0
BW 1 = (4.56)
Q
1
=
2RC

15. Consider the circuit shown in Figure 4.33. Define the following quantities:

C0

Vin L C R Vout

Figure 4.33: Phase shifter based on parallel RLC circuit.

1
!o = p and Qp = !o RC (4.57)
LC
(a) Find an expression for the phase shift between Vout and Vin , i.e., find an expres-
sion for the phase angle of the transfer function T (!) = Vout (!)/Vin (!).
(b) Show that if the reactance of the capacitor C 0 is much larger than R, then the
expression for the phase shift reduces to
1 ! !o
= /2 tan [Qp ( )] (4.58)
!o !

(c) Show that the phase shift is approximately a linear function of frequency for
frequencies near the resonant frequency of the tuned circuit.

16. A 100 pF capacitor and 2.53 H inductor are used to form a series resonant circuit
at 10 MHz. The capacitor is lossy and has a component Q of 100 at 10 MHz. The
inductor is also lossy and has a component Q of 80 at 10 MHz.

(a) Find the Qs of the resonant circuit formed by the series combination of the lossy
capacitor and inductor.
(b) Find the impedance of the series resonant circuit at 10 MHz.
(c) Suppose that it is desired to couple a 10 source to a 10 load through a
bandpass filter formed either by:
i. a series resonant circuit or
ii. a parallel resonant circuit.
4.7. HOMEWORK PROBLEMS 141

In either case, the resonant circuit would be formed by the lossy L and C con-
sidered in parts 16a and 16b. Sketch a schematic showing how the source should
be coupled to the load if the smallest possible bandwidth is desired.
(d) Estimate the bandwidth (in MHz) of the filter that you sketched in part 16c.

17. A series LC circuit consisting of an ideal lossless inductor L and ideal lossless capacitor
C is used as a bandpass filter to couple a source with source impedance RS = 50
to a load RL = 50 . The resonant frequency of the filter is fo = 100 MHz, and the
values are selected such that the -3 dB bandwidth of the filter is 10 MHz.

(a) Sketch the system, including the source, the LC filter, and the load and indicate
the values for L and C in nH and pF, respectively.
(b) Now, replace the inductor and capacitor with lossy components that have the
same inductance and capacitance as the original components but with component
Qs that are equal to 40 at the resonant frequency, i.e. QC = QL = 40 at 100
MHz. Calculate the -3 dB bandwidth of the filter implemented using the lossy
components.

18. The circuit shown below is used to transform a load resistance, RL , to a new value,
Rin , at the resonant frequency of the network. Away from resonance, the circuit
provides a bandpass filter response.

jXL1
Zin jXC
jXL2 RL

(a) The load resistance RL = 50 . At the resonant frequency assume that XL1 =
XL2 = 5 . Determine the value of XC required to resonate the circuit (i.e. to
make the input impedance Zin purely resistive).
(b) For the value of XC found in part a. determine Zin .
(c) The Q of the resonant circuit can be approximated by Qp of the equivalent parallel
RLC circuit obtained by applying parallel<=>series transformations. Find the
approximate Q of the network.

19. A 50 source is coupled to a 50 load through a bandpass filter consisting of a series

LC circuit which is resonant at 20 MHz.

(a) Sketch the system, including the source, the filter, and the load.
(b) Design the filter (i.e., find L and C) so that the -3 dB bandwidth of the filter is
5 MHz.
142 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

(c) Now suppose the inductor is lossy and has finite component Q, denoted by QL .
(You may assume that the inductance is the same as determined in part b.) What
is the minimum acceptable QL if the attenuation caused by the lossy L at 20 MHz
is to be smaller than 2 dB?

20. Design an LC circuit consisting of an ideal lossless inductor L and ideal lossless ca-
pacitor C to be used as a bandpass filter to couple a source with source impedance
RS = 100 to a load RL = 100 . Assume that a limited range of inductance values
is available such that 10 nH L 100 nH.

(a) Determine the topology of the LC filter and the values for L and C if the filter
is to have center frequency f0 = 100 MHz and the smallest possible bandwidth.
Sketch the system, including the source, the LC filter, and the load. Specify the
values for L and C in nH and pF, respectively.
(b) Repeat part a for the center frequency f0 = 1 GHz.

21. An LC circuit consisting of an ideal lossless inductor L and ideal lossless capacitor C
is to be used as a bandpass filter to couple a source with source impedance RS to a
load RL . Assume that RS = RL = R. Denote the resonant frequency of the filter by
!o , and denote the reactance of the inductor and capacitor at !o by Xo .

(a) If Xo > R, sketch the system (source, load, and LC filter) that will provide the
smallest possible bandwidth.
(b) Consider the system that you sketched in part a., and denote the loaded Q of
the system by Qo . The -3 dB bandwidth of the system is then BWo = Q !o
o
. Now,
suppose that the system is modified by replacing the lossless inductor with a
lossy inductor. Denote the component Q of the lossy inductor at !o by QL . The
inductance of the lossy inductor is the same as the inductance of the original
lossless component. Denote the bandwidth of the modified circuit by BW 0 . Find
an expression for BW 0 that involves only BWo , Qo , and QL .
(c) If Qo = 10 and QL = 40, by what percentage is the bandwidth of the filter
increased because of the loss in the inductor?
(d) Let PL denote the power delivered to the load at !o with the lossless inductor
0
in the system, and let PL denote the power delivered to the load at !o with the
0
lossy inductor in the system. Find an expression for the power ratio, PL /PL .
Express your result only in terms of Qo and QL .
(e) If Qo = 10 and QL = 40, determine the power loss, in dB, of the filter at the
resonant frequency.

22. An LC circuit consisting of an ideal lossless inductor L and ideal lossless capacitor C
is to be used as a bandpass filter to couple a source with source impedance RS to a
load RL . Assume that RS = RL = R. Denote the resonant frequency of the filter by
!o , and denote the reactance of the inductor and capacitor at !o by Xo .

(a) If Xo < R, sketch the system (source, load, and LC filter) that will provide the
smallest possible bandwidth.
4.7. HOMEWORK PROBLEMS 143

(b) Consider the system sketched in part a., and denote the loaded Q of the system by
Qo . Now, suppose that the system is modified by replacing the lossless inductor
with a lossy inductor and the lossless capacitor with a lossy capacitor. Denote
the component Qs of the lossy inductor and capacitor at !o by QL and QC ,
respectively. The inductance (capacitance) of the lossy inductor (capacitor) are
the same as the inductance and capacitance of the original lossless components.
Denote the new loaded Q of the system by Q0o . Find an expression for Q0o that
involves only Qo , QL , and QC . You may assume that QL 1 and QC 1.

23. Consider the single-resonator bandpass filter shown in Figure 4.34.

100 jXo jXo

Vs jXL jXC 100

Figure 4.34: Single-resonator bandpass filter.

(a) Suppose that the inductor has reactance XL = 100 at the center frequency of
the filter. Find the reactances of the coupling capacitors, Xo = 1/(!o Co ), and
the reactance of the resonator capacitor, XC = 1/(!o C), such that the filter has
a fractional -3 dB bandwidth of 5%. Hint: use series to parallel transformations
to convert the circuit to a parallel RLC circuit.
(b) Find L, C, Co that will place the center frequency of the filter at fo = 500 MHz.
(c) Using the values that you found in part b, plot the voltage frequency response of
the filter. Feel free to use your favorite circuit simulation software.
(d) Suppose that you build the filter using an inductor that has QL = 50 at fo .
Assume that the capacitors are lossless. Estimate the filter loss at fo . Express
your result in dB.

24. The system shown in Figure 4.35 consists of a resistive source and load coupled with a
bandpass filter. Design the filter to have fractional bandwidth ff3dB
o
= 0.04. The filter

100 j50 jXC

Vs jXo jXo 100

Figure 4.35: Series-resonator bandpass filter.

uses a single inductor with reactance j50 at fo . The reactances of the capacitors at
fo are denoted by Xo and XC .
144 CHAPTER 4. RLC NETWORKS, RESONANCE, AND Q

(a) Find numerical values for Xo and XC . Note that XC < 0 and Xo < 0.
(b) Find the component values for the inductor and capacitors (L, CC , Co ) that will
place the center frequency of the filter at fo = 50 MHz.
(c) Using the values that you found in part b, plot the voltage frequency response of
the filter. Feel free to use your favorite circuit simulation software.
(d) Suppose that you build the filter using an inductor that has QL = 50 at fo .
Assume that the capacitors are lossless. Estimate the filter loss at fo . Express
your result in dB.

25. Consider a simple single-resonator bandpass filter consisting of a series LC branch in

series with the source and load, or a parallel LC branch in parallel with the source
and load. Recall that the frequency response function of such a filter can be written
in terms of the loaded Q and the center frequency, !o , as follows:
1
F (!) = !o .
1 + jQ( !!o ! )

(a) Find an expression for the bandwidth, f A , at attenuation level A, where A is

the attenuation expressed in dB. Express your result in terms of the power ratio
= 10A/10 , the loaded Q of the filter, and the center frequency fo .
(b) Use your result from part (a) to anwer this question: A single-resonator LC
bandpass filter has center frequency fo = 10 MHz and loaded Qs,p = 10. Find
the -20 dB bandwidth, f 20 dB , of the filter.
(c) For the filter described in part b, find the lower -20 dB frequency of the filter
frequency response. Express your result in MHz.
Chapter 5

Oscillators

5.1 Introduction
This chapter will describe the methods used to analyze and design oscillators for applicatons
in transmitters and receivers. We will employ small-signal linear analysis techniques to
determine whether or not a particular circuit will oscillate, and a simplified nonlinear model
will be used to estimate the amplitude of the oscillations in a circuit based on a BJT.
Numerical simulations will be used to illustrate the characteristics of realistic oscillator
circuits, and the results will be compared to the analytical predictions.
For small signal analysis it is convenient to think of oscillators as unstable feedback
systems. An unstable system is one in which an initially small excitation or disturbance
produces an output that grows in time due to constructive, or positive, feedback. A block
diagram of a simple feedback system is shown in Figure 5.1.

Vi A Vo

Figure 5.1: A simple feedback system

The voltage transfer function for this system is

Vo A(j!)
= (5.1)
Vi 1 A(j!)B(j!)

where the quantity Alo (!) = A(j!)B(j!) is called the open loop gain of the system
sometimes shortened to loop gain. The subscript o is used to indicate that the loop gain is
computed assuming small-signal operation of the active devices. Note carefully that the loop
gain is the gain obtained by opening the feedback loop and taking the output at the point

145
146 CHAPTER 5. OSCILLATORS

Vf B

Vi A Vo

Figure 5.2: Loop gain at summing junction

where the loop was opened. For example, if the loop is opened at the summing junction,
then the loop gain is Vf /Vi as shown in Figure 5.2.
Suppose that the loop gain is equal to 1 at some frequency !o , i.e., Alo (!o ) = 1. Then
the voltage transfer function in Figure 5.1 is singular (infinite), which can be interpreted
as finite output for zero input. In other words, the circuit is a potential source of radio
frequency energy at the frequency where the loop gain is 1, even in the absence of any input
excitation Vi .
The condition for steady-state oscillation (Alo (!o ) = 1) is intuitively satisfying - when
the loop gain is 1, a sinusoidal excitation presented to the input of the circuit traverses the
feedback loop and appears back at the input with the same amplitude and phase that it
started with. This re-circulation of the disturbance proceeds indefinitely, with the circuit
oscillating in a steady state. In practice the small-signal loop gain is set to a value
somewhat larger than 1. This means that the disturbance is amplified after each pass
through the loop, and the output grows as the disturbance passes repeatedly through the
loop. In most radio frequency oscillators the loop gain is equal to 1 (or some real number
larger than 1) at one particular frequency. At other frequencies the amplitude may be less
than or greater than 1, but the phase angle is non-zero. This means that only one frequency
component can travel around the loop with no phase shift. Only this frequency component
will be amplified and grow to become a steady-state oscillation.
We shall see later in this chapter that even when the condition for oscillation is satisfied
at only one frequency, it is possible to have a non-sinusoidal output. However, this is due
to nonlinear eects in the amplifying devices. As noted above, practical oscillator circuits
are designed so that the small-signal loop gain is larger than one at the desired frequency
of oscillation, !o . This ensures that even thermal-noise-level signals will provide enough
excitation to cause oscillation to start and grow at !o . Some mechanism must be built into
the circuit to limit the amplitude of the oscillation to a finite value. The loop gain must
be decreased as the oscillation amplitude grows so that the loop gain can eventually settle
down to the steady-state value of 1.0. In so-called self-limiting oscillators, the amplitude
of the oscillation eventually becomes large enough to begin to saturate the active device,
eectively reducing the gain of the device, and also reducing the loop gain. Oscillation
amplitude stabilizes at the amplitude where the gain of the active device is reduced just
enough to set the loop gain to 1.0. Since self-limiting oscillators rely on driving the active
device to levels where nonlinearity is important, they often produce non-sinusoidal outputs
with significant harmonic content.
In some cases, it is desired to limit the amplitude of the oscillation at a level smaller than
that required to saturate the active device. One mechanism for doing this was implemented
5.2. OSCILLATOR ANALYSIS USING LOOP GAIN 147

by Hewlett and Packard in their first commercial product, an audio oscillator. They used
a small light bulb as a resistor to set the voltage gain of the amplifier in the oscillator.
As oscillation amplitude grows, the light bulb filament heats up, and the resistance of the
filament increases, reducing the gain of the amplifier. The thermal time constant of the
filament was much longer than the period of the oscillation, so that the filament acts as a
linear resistor, with resistance proportional to the oscillation amplitude. This mechanism
allowed Hewlett and Packard to create an audio oscillator with a nearly perfect sinewave
output. The oscillator exhibited an extremely low level of harmonic distortion in the output
because the amplitude of oscillation was limited at a level well within the linear range of
the amplifier.
In any case, some initial disturbance is necessary to start the oscillations. As long as the
small-signal loop gain is set equal to a value greater than 1 at the potential frequency of
oscillation, the thermal noise voltage that is always present in electrical circuits and/or the
turn-on transient which results when power is applied to the circuit will provide the initial
disturbance that starts the oscillations.

5.2 Oscillator Analysis using Loop Gain

We will now proceed to analyze one type of oscillator circuit using the loop gain approach.
The analysis proceeds as follows. First the feedback loop must be identified and the loop
gain computed. Then the condition for oscillation is applied to the loop gain. Oscillation
occurs if there is some frequency where the loop gain has magnitude 1 and phase angle equal
to zero. The combination of these two constraints is called the Barkhausen Criterion for
oscillation. The two conditions can be written as
arg[Alo (!o )] = 0 (5.2)

|Alo (!)|!=!o = 1 (5.3)

In practice we are usually able to apply condition (5.2) to solve for the potential frequency
of oscillation, !o . Then applying condition (5.3) will determine how much gain the amplifier
must have in order to make the loop gain equal to 1 at !o . The gain value determined by
equation (5.3) is the value required to support steady-state oscillation (oscillation amplitude
neither growing nor decaying). For practical designs, the amplifier gain is set to a value
somewhat higher than the value determined by equation (5.3) to ensure that oscillations
will reliably start and grow when the circuit is powered up.
Circuits with the topology shown in Figure 5.3a are commonly employed as oscillators.
The active device could be a BJT or an FET. This circuit can be analyzed as a feedback
loop. The circuit is redrawn in Figure 5.3b to explicitly show that the feedback from output
to input is through the element Z3 . The loop gain is easily computed for circuits of the type
represented by Figure 5.3. For small-signal analysis we can model the transistor as shown
in Figure 5.4. Note that this is a simplified version of the hybrid-pi model for the transistor
(see Appendix A). The passive elements of the model (e.g., r , C , C , etc.) can be lumped
into the external impedances Z1 , Z2 , Z3 . To find the open loop gain of this circuit, we
break the loop shown in Figure 5.3b at a convenient point and terminate that point in
the impedance that it sees when the loop is closed. The termination is required so that
the open-loop circuit operates under the same conditions as when the loop is closed. This
procedure will give the value of the loop gain that is appropriate for closed loop conditions.
148 CHAPTER 5. OSCILLATORS

Z3

Z3 Z2
Z1
Z2
Z1
(a) (b)

Figure 5.3: (a) Topology of one class of oscillator circuits. (b) Same as (a), redrawn to show
the feedback path from output to input through Z3 .

B C
+
vbe gm vbe
-

Figure 5.4: Simplified hybrid-pi model.

For example, suppose that the loop is broken to the right of Z3 in Figure 5.3b. Since the
output of Z3 normally looks into Z1 when the loop is closed, we terminate the loop with Z1
as shown in Figure 5.5. Then the loop gain is computed by exciting the circuit at the input

Z3
+ +
VIN Z1 gmVIN Z2 Z1 VOU T
- -

Figure 5.5: Feedback loop terminated with Z1

to the opened loop and computing the output across Z1 :

VOU T g m Z1 Z2
Alo = = (5.4)
VIN Z1 + Z2 + Z3
Some useful insights can be gained if we assume for the moment that Z1 and Z2 are purely
reactive, i.e., Z1 = jX1 , Z2 = jX2 . We allow Z3 to have a non-zero (positive) real part.
Then

g m X1 X2
Alo = (5.5)
Z3 + j(X1 + X2 )
5.2. OSCILLATOR ANALYSIS USING LOOP GAIN 149

For oscillation to occur, the phase angle of Alo must be zero at some frequency. Now there
are four possibilities for the signs of X1 and X2 . Define Zs = Z3 + j(X1 + X2 ) = |Zs |ejZs .
Then for each possible choice of sign on X1 and X2 , there is a requirement on the phase
angle of Zs (denoted by Zs ) in order for the overall phase angle of Alo to be zero. These
constraints are summarized in Table 5.1.

Table 5.1: Constraints for Overall Phase Angle of Alo to be Zero

(1) X1 > 0 (inductor), X2 > 0 (inductor) ) Zs =0

(2) X1 < 0 (capacitor), X2 < 0 (capacitor) ) Zs =0
(3) X1 > 0 (inductor), X2 < 0 (capacitor) ) Zs =
(4) X1 < 0 (capacitor), X2 > 0 (inductor) ) Zs =

Refer to Table 5.1 and note that cases 3 and 4 are impossible, since Zs = implies that
Zs = Z3 + j(X1 + X2 ) has a negative real part. If Z3 is a passive element, this is not
possible. The remaining two possibilities yield two circuit configurations for oscillators of
the type under consideration. They are shown in Figure 5.6. Notice that in each case Z3 is
allowed to have a positive real part which is represented by R.

R R
L C
C2 L2
C1 L1
(a) (b)

Figure 5.6: (a) X1 < 0, X2 < 0 ) Colpitts, (b) X1 > 0, X2 > 0 ) Hartley

The frequency of oscillation is easily found by finding the frequency where the phase
angle of Zs = Z3 + j(X1 + X2 ) is equal to zero. This will cause the phase angle of Alo to
be zero as well. The results are:
Colpitts !o = q 1
C1 C2
LC
1 +C2
(5.6)
Hartley !o = p 1
C(L1 +L2 )

These frequencies of oscillation are simply the resonant frequencies of the networks that
result when the transistor is removed from the circuits shown in Figure 5.6. In fact, if Z3
was lossless, then Z1 , Z2 and Z3 would constitute a lossless resonant circuit. A lossless
circuit will oscillate indefinitely at the resonant frequency once it is initially excited. When
Z3 is lossy, then the oscillations cannot be maintained unless the transistor is added to make
up for the losses.
So far we have considered only half of the Barkhausen criterion (arg[Alo ] = 0). The
results yielded the potential frequency of oscillation. For oscillations to occur, the gain of
the transistor must be large enough to make Alo equal to 1 at !o . The results of applying
150 CHAPTER 5. OSCILLATORS

this constraint are summarized below:

gm
Colpitts: |Alo |!=!o = =1 (5.7)
R!o2 C1 C2
R(C1 + C2 )
or, gm =
L

gm !o2 L1 L2
Hartley: |Alo |!=!o = =1 (5.8)
R
RC(L1 + L2 )
or, gm =
L1 L2

The values of gm obtained in equations 5.7 and 5.8 are the values necessary for the circuit
to support steady-state oscillations. We will denote these values by gm,ss . In practical
applications the transistor is biased to set the transconductance to a value somewhat larger,
e.g. a factor of 2 to 5 larger, than gm,ss . Setting gm > gm,ss causes the loop gain at !o
to be larger than 1 by the factor gm /gm,ss . One reason for doing this is to ensure that
oscillations start reliably even if component values change slightly. Setting Alo |!=!o > 1
means that the oscillations will not be maintained at a steady state; rather, they will grow
in amplitude. Growth will proceed until the active device is no longer operating in the
small-signal mode. As the oscillation grows, eventually the amplitude of the oscillation
will be limited by nonlinear eects. To first order, the onset of nonlinear operation coincides
with a reduction in the gain of the active device. This is called gain saturation, and the
eect can be modeled as a decrease in the transconductance, gm , and hence a decrease in
the loop gain. As the amplitude of the oscillation grows, the transconductance is decreased
to the point where the magnitude of the loop gain is 1. At this point steady-state oscillation
will be maintained.
Note that previous considerations did not specify which of the transistor terminals was
at RF ground. Thus they apply without modification to common base, common emitter or
common collector circuits. Also, it was assumed that Z1 and Z2 were purely reactive. If
this is not the case, one must go back to the general expression for loop gain:

g m Z1 Z2
Alo = (5.9)
Z1 + Z2 + Z3

Using this expression it is possible to find the frequency of oscillation and the minimum gm
required for oscillation.

5.3 Oscillator Analysis using Negative Resistance

The requirements for oscillation in a circuit can be derived using the so-called negative
resistance concept. Consider the lossless LC circuit shown in Figure 5.7. The Kirchos
loop equation for this circuit is I(jXC + jXL ) = 0. Hence, finite current I is allowed if
XL + XC = 0, which occurs at the resonant frequency, !o = pLC 1
.
Consider generalizing this idea to a circuit consisting of arbitrary impedances ZA and
ZB , as shown in Figure 5.8. The Kirchos loop equation is I(ZA + ZB ) = 0. Hence, finite
5.3. OSCILLATOR ANALYSIS USING NEGATIVE RESISTANCE 151

1
jXC = j!C I jXL = j!L

Figure 5.7: A lossless LC circuit supports finite current, I, if XL + XC = 0.

ZA ZB
I

Figure 5.8: Finite current, I, is supported if Z1 + Z2 = 0.

current I is allowed if ZA + ZB = 0. Writing ZA = RA + jXA and ZB = RB + jXB , it is

apparent that the conditions for finite current are:

XA + XB = 0 (5.10)

RA + RB = 0 (5.11)
Equation 5.10 is called the resonance condition, and says that the reactances of ZA and ZB
must sum to zero. Equation 5.11 specifies that the real parts of the impedances ZA and ZB
must sum to zero. If one of the impedances has a positive real part, then the other impedance
must have a negative real part, i.e. must exhibit negative resistance. This idea can be used
to analyze and design oscillator circuits. The conditions represented by equations 5.10 and
5.11 are analogous to the two conditions for oscillation given by equations 5.2 and 5.3.
In practice the negative resistance concept is applied by breaking a candidate oscillator
circuit into two parts, which are associated with the impedances ZA and ZB . Most oscillators
consist of a single active device, so when the circuit is divided into two parts, one part will
contain only passive components. This part will exhibit an impedance with positive real
part. In order for oscillations to occur in the circuit, the part containing the active device
must have an impedance with a negative real part.
It is instructive to analyze the oscillator circuit shown in Figure 5.3 using this approach.
Suppose that the circuit is divided into two parts, by defining ZA = Z1 . The impedance
ZB is the impedance seen looking into the port defined by the base and emitter terminals
of the transistor, with Z1 removed from the circuit. If the transistor is modeled using the
simplified hybrid-pi model (Figure 5.4), the impedance between the base and emitter of the
transistor is easily shown to be
Z2 + Z3
ZB = .
1 + g m Z2
Enforcing ZA + ZB = 0 yields
Z2 + Z3
Z1 + = 0.
1 + g m Z2
152 CHAPTER 5. OSCILLATORS

1+gm Z2
Multiplying both sides of this equation by Z1 +Z2 +z3 yields the result

g m Z1 Z2
= 1,
Z1 + Z2 + Z3
which is the same as setting the loop gain, given in equation 5.4, to 1. Thus, the negative
resistance analysis yields the same result as the loop gain analysis.

5.4 Example - Common-collector Colpitts Oscillator

5.4.1 Analysis
To illustrate some of the properties of oscillators, we will consider a realistic circuit in detail.
Practical circuits for Colpitts-type oscillators implemented with the common-collector and
common-base topologies are shown in Figure 5.9. The common-emitter variant is not shown.
We will analyze the common-collector configuration shown in Figure 5.9a. The small-signal

Vcc

R1

C1
L R2
C2 Re

(a)

Vcc

L
R1
C2

R2
Re C1

(b)

Figure 5.9: (a) Common-collector Colpitts oscillator circuit, (b) Common-base Colpitts
oscillator circuit.

equivalent circuit is shown in Figure 5.10, where the transistor has been replaced with its
hybrid-pi model, and the finite Q of the inductor is modeled with a series resistance, r.
The small-signal equivalent circuit can be further simplified by combining some elements as
shown in Figure 5.11 , where
5.4. EXAMPLE - COMMON-COLLECTOR COLPITTS OSCILLATOR 153

+
C1 C vbe r Co
L - gm vbe
R1||R2
r C2 Re

Figure 5.10: Small-signal equivalent circuit for common-collector Colpitts. The resistance,
r, represents the series resistance of the inductor.

+
Z1 vbe
-
Z3
Z2 gm vbe

Zin = Z1 + Z2 + gm Z1Z2

Figure 5.11: Simplified small-signal equivalent circuit of common-collector Colpitts oscilla-

tor.
154 CHAPTER 5. OSCILLATORS

1
Z1 = ||r (5.12)
j!(C1 + C )
1
Z2 = ||Re
j!(C2 + Co )
1
Z3 = R1 ||R2 ||(r + j!L)||
j!C

In practice it is useful to choose C1 C , C2 Co . This ensures that the exter-

nal components swamp the internal capacitances of the transistor, thereby minimizing the
circuits dependence on variations in the internal transistor capacitances. Such variations
could be caused by dierences between individual transistors, or by variations in junction
capacitances as a function of the voltage across the junctions. It is also useful to choose C1
and C2 to be large enough so that 1/(!C1 ) r , 1/!C2 Re . This causes Z1 and Z2 to be
dominated by the capacitances external to the transistor, thereby minimizing dependence
on r (which depends on bias current and transistor ) and losses in r and Re . In view
of these considerations, in the following analysis we shall make the following replacements:
C1 + C ! C1 and C2 + Co ! C2 . In other words, C1 should be interpreted as the to-
tal capacitance across the base-emitter junction, including the internal capacitance of the
transistor and any external capacitance. A similar interpretation holds for C2 .
It is now useful to make some approximations in order to simplify the analysis. We shall
assume that the impedances of the capacitances C1 and C2 are small compared to r and
Re , respectively. In other words, define Q1 = !C1 r and Q2 = !C2 Re ; we assume that
Q1 1 and Q2 1. Then Z1 and Z2 can be transformed using a high-Q parallel to series
transformation, i.e.
r 1 1 1
Z1 ' + = 2 2 + , Q1 1 (5.13)
Q21 j!C1 ! C 1 r j!C1

Re 1 1 1
Z2 ' + = 2 2 + , Q2 1. (5.14)
Q22 j!C2 ! C 2 Re j!C2
We shall also assume that the impedance of the Z3 branch is well approximated by the
impedance of the inductor, i.e.
Z3 ' r + j!L. (5.15)
If this is not the case, then r and L should be interpreted as the eective series resistance,
and inductance, of Z3 .
The circuit has been split into two parts by the dashed line in Figure 5.11. Denote the
impedance looking into the part of the circuit to the right of the dashed line by Zin . Then

Zin = Z1 + Z2 + gm Z1 Z2 .

The condition for steady-state oscillation is Zin + Z3 = 0, or,

Z1 + Z2 + Z3 + gm Z1 Z2 = 0. (5.16)

The requirements for steady-state oscillation are obtained using equations 5.13-5.15 in 5.16.
The real part of this equation is
1 1 1 1
+ 2 2 + r + gm ( 4 2 2 ) = 0. (5.17)
! 2 C12 r ! C 2 Re ! C 1 C 2 r R e ! 2 C1 C2
5.4. EXAMPLE - COMMON-COLLECTOR COLPITTS OSCILLATOR 155

It is now necessary to remember that r depends on gm , i.e. r = /gm , so equation 5.17

is really quadratic in gm , and the solution is somewhat messy. However, at suciently high
frequencies, the term involving ! 4 may be neglected. More precisely, the term involving
! 4 may be neglected provided that ! 2 C1 C2 r Re 1, or if Q1 Q2 1. Since we have
already assumed that Q1 1 and Q2 1, neglect of this term is justified within the
context of our original assumptions. Hence, neglecting the ! 4 term, and using r = /gm
in equation 5.17, the net resistance becomes
gm 1 gm
+ +r = 0. (5.18)
! 2 C12 ! 2 C22 Re ! 2 C1 C2
This equation can be solved for the transconductance needed to set the net resistance to
zero at any particular frequency. This transconductance will be denoted gm.ss because
it is the transconductance needed to support steady-state oscillations. The steady-state
transconductance can be written as follows:
C1
! 2 C1 C2 r + C2 R e
gm,ss = C2
. (5.19)
1 C1

Remember that this result was derived under the assumption that Q1 1, and Q2 1. The
first assumption requires that gm !C1 , and should be checked after gm,ss is calculated
from equation 5.19, however in most practical applications this inequality will be satisfied.
If Re is allowed to approach infinity (so that Z2 becomes a pure reactance), and if
C1 , then gm,ss reduces to gm,ss = ! C1 C2 r, the same as the result given in equation
C2 2

5.7, which was derived by assuming that Z1 and Z2 were pure reactances. In the next
section, it will be shown that it is desirable to make C1 > C2 . Thus, in practical cases with
1, the second term in the denominator of 5.19 can be neglected, in which case
C1 1
gm,ss ' ! 2 C1 C2 r + . (5.20)
C 2 Re
It is worth noting that equation 5.20 can be written in terms of the inductor QL . In
terms of the inductor reactance, XL , and the inductor QL (both are assumed to be known,
or specified, at the frequency of oscillation) r = XL /QL . As long as Q1 1 and Q2 1, the
frequency of oscillation will be the frequency where the inductor reactance resonates with
the series combination of C1 and C2 , hence XL = !o L ' C1 C2 . Thus, r = XL /QL '
1
!o C
1 +C2
(C1 + C2 )/(!o C1 C2 QL ) - if this substitution is made in equation 5.20, the result is:

!o (C1 + C2 ) C1 1
gm,ss ' + . (5.21)
QL C 2 Re
In practical circuits additional reactances or resistances may be placed in parallel with
the inductor. To implement a tunable oscillator it is common practice to use a variable
capacitor in series or in parallel with the inductor. Since power must be extracted from the
oscillator to drive other stages, a resistance representing an external load may be placed in
shunt with the inductor if the output is taken across the inductor. The existing analysis is
easily modified to handle such cases. In the original model, the resistance, r, represented
the series resistance associated with the inductor. If the additional components are in
parallel with the inductor, it is simply necessary to determine the series representation of
the resulting Z3 branch (at !o ) and use the real part of this impedance in place of r in
156 CHAPTER 5. OSCILLATORS

equation 5.20. Alternatively, evaluate the component Q of the Z3 branch at !o , and use
this value in place of QL in equation 5.21. In some circuits the output is taken from the
emitter - in this case, the external load resistance will appear in parallel with Re , and the
value used for Re can be changed to include the contribution from the load.
As mentioned previously, the oscillator circuit will be designed such that the small-signal
loop gain is larger than 1. The loop gain is equal to the ratio gm /gm,ss . As the oscillation
amplitude builds up, the base-emitter voltage swing will increase. Transistor operation
moves out of the small-signal linear regime, nonlinear eects become important, and the
eective transconductance is reduced. Oscillation amplitude will grow until the eective
large-signal transconductance is reduced to gm,ss , which is the value required to sustain
steady-state oscillations. The loop gain is the ratio of the initial transconductance to gm,ss
(Alo =gm /gm,ss ), and is an important parameter that determines how deeply the transistor
must be driven into the nonlinear regime before the steady-state condition is reached. In
the next section we will see that under certain realistic assumptions, the ratio gm /gm,ss
determines the steady-state amplitude of the base-emitter voltage swing. This means that
it is possible to predict, at least approximately, the amplitude of the voltage swing in the
oscillating circuit.

5.4.2 Numerical Simulation

In order to test the results derived in the preceeding section, a common collector Colpitts
circuit was simulated using the Agilent ADS simulation program. The frequency of oscilla-
tion was chosen to be fo = 50 MHz. The circuit was simulated using a 2N5179 transistor
model. To compare the simulated results with analytical predictions, the parameters of the
hybrid-pi model are needed. The transistor model was probed using simulations to deter-
mine that = 62, C ' 10 pF, C ' 1 pF. The capacitance Co was assumed to be zero. A
supply voltage of 12.0V was used. The DC blocking capacitor (in series with the inductor)
was given a value of 0.01 F for the simulation. The values chosen for the inductor and bias
resistances are:
L = 500 nH
QL = 50
R1 = 50 k
R2 = 50 k
RE = 5 k
The bias network is designed to give a quiescent collector current of approximately 1
mA. Thus r ' 1.55 k and gm ' 40 mS. The inductor has QL = 50, however the bias
resistors R1 and R2 appear in parallel across the inductor. This lowers the eective Q of
the inductor to QL =38, which is the value that was used in the analytical calculations. 1
The eect of C , which also appears across the inductor, is small and was ignored. To start
the oscillations, the circuit was artificially excited by pulsing the supply voltage from 12.0
V to 12.1 V and back to 12.0 V within a period of 10 ns. This is necessary in a numerical
simulation in order to provide some initial disturbance that can then grow into a steady-
state oscillation. In practice, omnipresent thermal noise (or the step excitation provided by
connecting the supply voltage) would play this role.
1A common base configuration would be better in this respect, as the bias resistors do not load the
inductor in that circuit.
5.4. EXAMPLE - COMMON-COLLECTOR COLPITTS OSCILLATOR 157

Six dierent {C1 , C2 } pairs were considered (see Table 5.2). For each case, the transcon-

Table 5.2: Parameters used for oscillator simulations.

Case C1 C2 gm,ss gm /gm,ss

1 2500 pF 20.4 pF 45.3 mS 0.88
2 1000 pF 20.6 pF 18.1 mS 2.2
3 500 pF 21.1 pF 9.42 mS 4.4
4 200 pF 22.5 pF 3.77 mS 11.0
5 100 pF 25.4 pF 1.83 mS 21.8
6 22.5 pF 200 pF 2.23 mS 17.9

ductance required to sustain steady-state oscillation (gm,ss ), and the small-signal loop gain
(gm /gm,ss ) are given. The values of C1 and C2 were chosen to satisfy 2(50 MHz) =
q 1
L CC11+C
C2
2
, so that the oscillation frequency is fixed at approximately 50 MHz in all cases.
Notice that if a {C1 , C2 } pair is found that resonates with L at the desired !o , the values
can be reversed without significantly changing the resonant frequency. Compare cases 4 and
6 for an example. As we shall see, however, the case with C1 > C2 will usually give better
performance.
Simulation results are summarized in Figures 5.12-5.17. In each Figure, the upper left
plot shows the voltage across the inductor for an interval of 1 s after the initial transient.
The other plots show the emitter current, base-emitter voltage, and emitter voltage for
the period 0.9-1.0 s after the initial transient. For the cases where oscillation occurs, the
expanded plots show the current and voltage waveforms when the circuit is undergoing
steady-state oscillation.
The loop gain (gm /gm,ss , column 4 in Table 5.2) is smaller than 1 for Case 1. Therefore
the initial disturbance excites damped oscillations at a frequency approximately equal to fo .
This case illustrates the fact that sustained oscillations cannot develop if the small-signal
loop gain is smaller than 1.
In cases 2 through 6 the loop gain is larger than 1, and the initial transient excites
growing oscillations, as shown in the plot of the voltage across the inductor (upper left plot
in each Figure). As the loop gain increases from 2.2 to 21.8 in cases 2 through 5 the initial
transient builds up to steady-state condition faster, and the amplitude of the steady-state
voltage swing increases. Notice that increases in the base-emitter voltage swing result in the
collector current (lower, left) waveform exhibiting increasingly narrow and increasingly large
current spikes. For the larger loop gains, the transistor is essentially cut o for most of the
oscillation cycle. The transistor injects a short current pulse into the resonator once each
cycle, near the positive peak of the voltage swing. Thus, for most of the oscillation period,
the circuit is un-excited and oscillation is maintained by the flywheel eect of the high-Q
resonant circuit composed of L, C1 , and C2 . The emitter current flows into an impedance
that has large magnitude at !o , and very small magnitude at harmonics of !o - hence, the
emitter voltage is nearly sinusoidal. In cases 4 and 5 where the loop gain is largest, and the
current spikes are narrowest, some distortion of the emitter voltage waveform can be seen
near the voltage peaks. In general, larger loop gains are associated with more distortion in
the output waveform.
Case 6 was chosen to illustrate what happens when C1 < C2 . Notice that the loop gain
158 CHAPTER 5. OSCILLATORS

Figure 5.12: Case 1 - loop gain is 0.88, which is less than one, so oscillation does not start.

Figure 5.13: Case 2 - gm /gm,ss = 2.2.

5.4. EXAMPLE - COMMON-COLLECTOR COLPITTS OSCILLATOR 159

Figure 5.14: Case 3 - gm /gm,ss = 4.4.

Figure 5.15: Case 4 - gm /gm,ss = 11.0.

160 CHAPTER 5. OSCILLATORS

Figure 5.16: Case 5 - gm /gm,ss = 21.8.

Figure 5.17: Case 6 - gm /gm,ss = 17.9.

5.4. EXAMPLE - COMMON-COLLECTOR COLPITTS OSCILLATOR 161

in case 6 is not very dierent from the loop gain in case 5 (17.9 vs. 21.8), yet the voltage
swing across the inductor and the emitter voltage swing are quite small. Furthermore, the
emitter voltage is clearly non-sinusoidal, indicative of relatively high harmonic content. This
reflects the fact that C1 is not large enough to swamp r , and the resulting dissipation in r
results in a lower overall Q for the LC tank circuit than in case 4. As a result, the harmonics
contained in the spiky emitter current waveform are more prominent in the emitter voltage
waveform. The output from the oscillator would be taken from the top of the inductor, or
from the emitter - so comparison of cases 4 and 6 shows that choosing C1 > C2 (as in case
4) yields a larger output swing, and less distortion in the output.
Cases 4 and 6, which have comparable loop gains but very dierent amplitudes of oscilla-
tion, motivate the need to understand the factors which govern the amplitude of oscillations.
Consider the large-signal equivalent circuit for the oscillator circuit in Figure 5.18.

+
Z1 Vbe
-
Z3
+
Ve Z2 Gm Vbe
-

Figure 5.18: A model for the Colpitts oscillator when oscillating in steady-state. The small-
signal transconductance has been replaced by a large signal transconductance, Gm , which
is a function of the base-emitter voltage swing, |Vbe |.

The small-signal transconductance has been replaced by an eective (large-signal) transcon-

ductance (Gm ) in Figure 5.18. If the circuit is undergoing steady-state oscillations, the large
signal transconductance will be equal to the transconductance required to sustain oscilla-
tions, i.e. Gm = gm,ss . In Appendix A, it is shown that when a BJT is biased for constant
DC emitter current, and when the base-emitter voltage swing is sinusoidal at frequency !o ,
a large-signal transconductance can be defined, which relates the amplitude of the emitter
current component at !o to the amplitude of Vbe . The large signal transconductance can be
written as
2 I1 (x)
Gm (x) = gm , (5.22)
x I0 (x)
where gm is the small-signal transconductance, and x = |Vbe |/(25 mV) is the normalized,
peak base-emitter voltage amplitude. Figure 5.19 shows how the ratio Gm /gm decreases
as |Vbe | increases. In steady-state, |Vbe | settles at the value required to set the large signal
transconductance equal to the minimum value required to sustain oscillations, i.e. in steady
state Gm (|Vbe |/(25 mV)) = gm,ss . For example, consider case 2, where the small-signal loop
gain is gm /gm,ss = 2.2. To reach the steady state, |Vbe | must increase until Gm /gm =
1/2.2 = 0.45. From Figure 5.19 it is apparent that this occurs when |Vbe |/(25 mV) is
approximately equal to 3.8. A more exact value can be obtained using equation 5.22;
numerically solving the equation 0.45 = x2 I1 (x)/Io (x) yields x ' 3.75. The predicted steady-
state amplitude of |Vbe | is then (3.75)(25 mV) = 94 mV, which compares favorably with what
is observed in the simulations. Table 5.3 compares the base-emitter swings observed in the
simulations with values predicted in this way. In cases 2-6, where oscillation occurs, the
162 CHAPTER 5. OSCILLATORS

0.8

0.6
Gm (x)
gm 0.4

0.2

2 4 6 8 10
x

Figure 5.19: Ratio of large signal to small-signal transconductance. The parameter x =

|Vbe |
25 mV .

Table 5.3: Comparison between predicted and simulated |Vbe | and |Ve |/|Vbe | ratio.

Case gm /gm,ss |Vbe | pred. |Vbe | sim. |Ve | sim. C1 /C2 |Ve |/|Vbe |
1 0.88 - - - 122.5 -
2 2.2 94 mV 71 mV 3.27 V 48.5 46.1
3 4.4 207 mV 185 mV 4.30 V 23.7 23.2
4 11.0 537 mV 554 mV 4.70 V 8.9 8.5
5 21.8 1.08 V 1.29 V 5.40 V 3.9 4.2
6 17.9 882 mV 483 mV * 0.11 *
5.5. EXAMPLE: VOLTAGE CONTROLLED OSCILLATOR (VCO) 163

prediction yields a reasonable approximation to the simulated values.

Once |Vbe | is known, or has been predicted, the amplitude of the emitter voltage, |Ve |,
can be estimated. A node equation at the junction between the current source, Z1 , and Z2
in Figure 5.22 yields
Ve Z2
= + Z2 gm,ss (5.23)
Vbe Z1
C1
= + Z2 gm,ss . (5.24)
C2
The second term will be small compared to the first in practical cases. This can be verified
for the cases considered here by inserting numerical values, or by noting that the simulation
results show that Ve and Vbe are nearly in-phase. Hence, the ratio Ve /Vbe is real - confirming
that the second term is negligible in all cases considered here. Thus, the emitter voltage
swing is determined by the capacitive transformer consisting of C1 and C2 and can be
written as:
C1
|Ve | = |Vbe | . (5.25)
C2
Table 5.3 provides data showing how the ratio |Ve |/|Vbe | determined from simulation com-
pares to the C1 /C2 ratio. In cases 2-5 the agreement is excellent. In case 6, the emitter
voltage waveform was non-sinusoidal, and the amplitude of the fundamental component is
not easily extracted from the time-domain waveform, so the ratio was not calculated. These
results show that making the ratio C1 /C2 > 1 will cause the emitter (output) voltage to
be larger than the base-emitter voltage swing. When C1 /C2 > 1, increasing C1 /C2 yields
smaller loop gains, so oscillators with large C1 /C2 tend to have more sinusoidal output
waveforms because the base-emitter swing is relatively small, and the emitter current is
more sinusoidal.

5.5 Example: Voltage Controlled Oscillator (VCO)

Figure 5.20 is an example of an oscillator circuit used in a commercial television receiver.
The oscillator is part of the VHF tuner that performs the function of converting the VHF
channels to the IF frequency near 45 MHz, this local oscillator is voltage-tuned using a
varactor diode. Figure 5.20 also illustrates how dierent inductances can be switched in and
out of the circuit using diode switches. Although this circuit looks more complicated than
the one we have been studying, it is essentially the same, except for the fact that the base
of the transistor is at RF ground (common-base configuration).
A simplified schematic of Figure 5.20 is shown in Figure 5.21.
Based on earlier analysis, it should be clear the the approximate frequency of oscillation
for this circuit will be the frequency where C1 , C2 and the L-Cv combination are resonant.
We also know that the frequency of oscillation will be such that the L-Cv combination looks
inductive.
Thus !o is the solution to
1 j!o L j!o1Cv
+ =0 (5.26)
j!o CS j!o L + j!o1Cv
where
C1 C2
CS = (5.27)
C1 + C2
164 CHAPTER 5. OSCILLATORS

Vout
2.2 pF
C2

10 pF 2k +12 V
Tuning C1
Diode, Cv
10 k 5 F
L1 L2
680 5k

33 k

Band-switch Tuning
Voltage

Figure 5.20: Example: VHF oscillator for TV tuner.

Cv

C2

C1

Figure 5.21: Simplified schematic of VHF oscillator circuit.

5.6. OSCILLATOR PHASE NOISE 165

Therefore
1
!o = p (5.28)
L(CS + Cv )
Equation 5.28 can be used to find the range of values over which Cv must vary in order
to cover a certain range of oscillation frequencies.
Voltage controlled oscillators (VCOs) find extensive application in phase-locked loops
(PLLs) which in turn are used for a majority of frequency synthesizer applications. For
PLL or synthesizer design, an important parameter is the so-called VCO gain constant,
which is simply the incremental change in oscillation frequency for an incremental change
in varactor bias voltage. Typically, the varactor would be an abrupt transition type, so
that
C (1)
Cv = p
VD
where VD is the varactor bias voltage and C (1) is the junction capacitance at 1 volt of bias.
The gain constant is

@!o
Ko = (5.29)
@VD !o = oscillation frequency

@!o @Cv
=
@Cv @VD !o = oscillation frequency

Generally, Ko will be a nonlinear function of VD . This implies that the gain constant
will dier at various varactor bias settings, a factor which must be considered in the design
of a PLL.

5.6 Oscillator Phase Noise

The most significant departure from ideal behavior in oscillators is the presence of phase
noise, represented by a random variation in the oscillators phase angle, i.e.

V (t) = Vo cos[2fo t + (t)] (5.30)

where (t) is a random noise process that has various physical processes. Amplitude noise
may also be present, but it is easy to remove this noise component by passing the signal
through a limiter. Figure 5.22 shows measurements of the output spectrum of a synthesized
signal generator operating at 10.240 MHz. The width of the narrow central spike is deter-
mined by the resolution bandwidth of the spectrum analyzer which was set to 100 Hz for this
measurement. Residual phase modulation phase noise on the signal contributes the
broad bandwidth pedestal underneath the spike. Phase noise spectral density is specified as
a function of frequency separation from the carrier in terms of the noise power level within
a 1 Hz bandwidth. The noise power level is given in dB referenced to the carrier power
level. For example, in Figure 5.22 the triangular marker has been placed 1000 Hz above the
carrier frequency, and the display indicates that the noise pedestal is at 74.2 dBc, where
dBc means relative to the carrier. Assuming that the noise spectral density is constant
within the resolution bandwidth the measured noise power level in 100 Hz bandwidth can
166 CHAPTER 5. OSCILLATORS

Figure 5.22: Spectrum analyzer display showing phase noise on the output signal from
an HP E4432B synthesized signal generator. The signal generators output frequency was
10.240 MHz and the output level was 0 dBm. The triangular marker is oset 1000 Hz
from the carrier frequency, and the phase noise at this oset is 94.2 dBc/Hz (see text for
explanation).
5.6. OSCILLATOR PHASE NOISE 167

be scaled to 1 Hz bandwidth by dividing by 100 (or by subtracting 10 log10 (100) = 20 dB).

Thus, the phase noise level at 1000 Hz oset would be reported as 74.2 20 = 94.2
dBc/Hz.
Assuming that the random phase process is dierentiable, the instantaneous frequency
of the noisy oscillator can be written as
1 d (t)
finst = fo + . (5.31)
2 dt
The instantaneous frequency deviation is therefore
1 d (t)
f (t) = . (5.32)
2 dt
The fractional deviation in instantaneous frequency can be written:
f (t) 1 d (t)
y(t) = = . (5.33)
fo 2fo dt
Measures of phase (or frequency) stability may be defined in the frequency domain in
terms of a power spectral density, or in the time domain in terms of a phase, or frequency,
or fractional-frequency variance defined over some time-interval.
If the random phase process has spectral density S (f ), then the corresponding spectral
density for fractional frequency fluctuations is:
f2
Sy (f ) = S (f ). (5.34)
fo2
The spectral density Sy (f ) will be invariant to multiplication or division of the frequency
using ideal frequency multipliers or dividers because frequency multiplication or division
scales the reference frequency, fo , and the frequency deviation, f , by the same factor.
In the time-domain, oscillator stability is usually characterized by the expected value
of the sample variance of fractional frequency fluctuations. Define the average fractional
frequency deviation over an interval , beginning at time tk by
R t +
y k = 1 tkk y(t)dt
(5.35)
(tk + ) (tk )
= 2fo .

When N measurements of y k are available, each taken over interval and with starting
times separated by interval T , the sample variance of the measurements is:
N
X N
2 1 1 X
y (N, T, ) = (y n y k )2 (5.36)
N 1 n=1 N
k=1

The expected value of the quantity y2 is known as the Allan variance of the oscillator.
For most purposes, oscillator stability is specified in terms of the Allan variance for N=2
and T = , i.e. the quantity that is specified most often is the variance of the dierence
between 2 successive measurements with no dead time between the measurement intervals.
In this case, the Allan variance reduces to:

2 < (y 2 y 1 )2 >
< y ( ) >= (5.37)
2
168 CHAPTER 5. OSCILLATORS

Thus, the Allan variance is the variance of the dierence between two successive measure-
ments of the fractional frequency deviation, measured over a time interval . The square-root
of the Allan variance is called the Allan Deviation, and is commonly supplied on oscillator
data sheets where it is often plotted verses the measurement interval .
5.7. REFERENCES 169

5.7 References
1. Clarke, Kenneth K. and Donald T. Hess, Communication Circuits: Analysis and De-
sign, Addison-Wesley, 1978.

2. Frerking, Marvin E., Crystal Oscillator Design and Temperature Compensation, Van
Nostrand Reinhold, New York, 1978.
3. Krauss, H. L., C. W. Bostian, and F. H. Raab, Solid State Radio Engineering, John
Wiley & Sons, New York, 1980.

4. Matthys, Robert J., Crystal Oscillator Circuits, John Wiley & Sons, New York, 1984.
5. Parzen, Benjamin, Design of Crystal and other Harmonic Oscillators, John Wiley &
Sons, New York, 1983.
6. Smith, Jack, Modern Communications Circuits, Second Edition, McGraw Hill, 1998.
170 CHAPTER 5. OSCILLATORS

5.8 Homework Problems

1. Consider the voltage amplifier in Figure 5.23, with
Vcc =12 V R1 = 10 k R2 = 30 k Re = 1 k
RC = 1 RL = 1 k L = 2 H C = 50 pF
Capacitors that are not labeled are assumed to be short circuits over the frequency
range of interest. The transistors is large enough so that the bias point does not
explicitly depend on its value. You may neglect the transistor parameters rx , r , C ,
ro , and Co in your analysis for parts 1b, 1c, and 1d.

Vcc
L C RC
R2
vo
vi
RL
R1
Re

Figure 5.23: Common-emitter amplifier with tuned output.

(a) Find the quiescent collector current, ICQ . Express your result in mA.
(b) Find the resonant frequency of the amplifier. The voltage gain will be largest at
this frequency. Express your result in MHz. Do not make any 2 errors!
(c) Find the voltage gain at resonance.
(d) Find the 3 dB bandwidth of the amplifier. Express your result in MHz.

2. The hybrid-pi parameters for a transistor are:

0.025
r = (5.38)
ICQ
C = 25 pF
C = 8 pF
ro = 100 k
= 100
(a) The transistor, with parameters given above, is used in a common emitter ampli-
fier, as shown in Figure 5.24. Find an expression for the voltage gain of amplifier.
Express your result in terms of RC , RL , C , and gm . You may assume that ro
is much larger than RC kRL and can be neglected. Assume that rx and Co are
small and may also be neglected.
(b) If the quiescent collector current is 1mA, RC = RL = 5k , and all coupling and
bypass capacitors are assumed to be perfect short circuits, sketch the magnitude
and phase of the voltage gain as a function of frequency. What are the magnitude
and phase of the transfer function at 15 MHz?
5.8. HOMEWORK PROBLEMS 171

Vcc
RC
R2
vo
vi
RL
R1
Re

Figure 5.24: Common emitter amplifier

(c) Suppose that we attempt to measure the voltage gain of this amplifier using a
10X scope probe. The shunt capacitance of this type of probe is 12pF. Compute
the gain (at 15 MHz) that would be measured. Compare to the result from 2b.

3. Consider the voltage amplifier shown in Figure 5.23. Capacitors that are not labeled
are assumed to be short circuits over the frequency range of interest. The other circuit
parameters are
VCC =12 V R1 = 5 k R2 = 10 k RE = 3.3 k
RC = 5 k RL = 1 k L = 20 H C = 90 pF
Use the simplified hybrid-pi model shown in Figure 5.25 for the transistor:

B C
+
vbe C r gm vbe
-

Figure 5.25: Simplified hybrid-pi model

(a) Find the resonant frequency of the amplifier. The voltage gain (vo /vi ) will be
largest at this frequency. Express your result in MHz. Do not make any 2
errors!
(b) Find the 3 dB bandwidth of the amplifier. Express your result in MHz.
(c) Find the quiescent collector current, ICQ . State any approximations.
(d) Find the voltage gain (vo /vi ) at resonance.

4. A quartz crystal resonator has the equivalent circuit shown in Figure 5.26. Suppose
it is known that a particular crystal has Q=50,000, series resonant frequency of 5
MHz, and parallel resonant frequency of 5.005 MHz. It is also known that the parallel
resonant frequency shifts by 1 kHz, if a 3 pF capacitor is placed in parallel with the
172 CHAPTER 5. OSCILLATORS

Co

r L C

Figure 5.26: Equivalent circuit for a quartz crystal resonator.

crystal. Find the values of the equivalent circuit elements for the crystal. You may
use whatever approximations are appropriate for a low-loss crystal.
5. Consider the AC equivalent circuit for an oscillator circuit in Figure 5.27:

R
C2
C1

Figure 5.27: AC equivalent circuit for oscillator circuit

The loop gain for this circuit can be shown to be

gm !2 C11 C2
Alo = 1 1 j!RL
(5.39)
j!C1 + j!C2 + R+j!L

It has been assumed that the transistor immittances could be ignored in deriving the
above equation. Find an expression for the frequency of oscillation, !o . To simplify
interpretation of your result, define C 0 = C1 C2 /(C1 + C2 ) and express your result in
terms of R, L, and C 0 .
6. Imagine that we could produce an active device, as shown in Figure 5.28, that has an
impedance Z = -600 . Assume that the device has this impedance at all frequencies.
Suppose our idealized circuit is connected to an external R, L, C network as shown in
the Figure and with R = 120 , L = 4.0H, and C = 52pF.
(a) Will the circuit oscillate?
(b) If so, find the frequency at which the circuit will oscillate. If not, find the fre-
quency at which the circuit would oscillate if the negative resistance of the active
source was decreased enough to cause oscillations to start.
7. We want to make an oscillator by cascading a number of identical stages as shown
in Figure 5.29. Assume that the transistors can be represented by the simplified
equivalent circuit in Figure 5.30 and that the capacitances C have finite values.
5.8. HOMEWORK PROBLEMS 173

C R

Z= 600

Figure 5.28: Idealized circuit

C R C R C R

1 2 N

Figure 5.29: An oscillator consisting of N identical sections in cascade.

B C
+
vbe gm vbe
-

Figure 5.30: Simplified equivalent circuit

174 CHAPTER 5. OSCILLATORS

(a) Find the minimum number of stages (N) that are necessary in order for oscillation
to occur.
(b) Assuming that we use the minimum number of stages, find an expression for the
frequency of oscillation.
(c) Find the minimum value of gm which will ensure that oscillation occurs.

8. Consider the circuit in Figure 5.31.

Use the simplified equivalent circuit for the transistor given in problem 7.

L2 R L1

Figure 5.31: Hartley oscillator.

(a) Find an expression for the frequency of oscillation.

(b) Find the minimum value of gm required to make the circuit oscillate.

9. Consider the oscillator circuit in Figure 5.32. The amplifier has infinite input impedance

C1 R1

1
+
3
-
R2 C2 2

Figure 5.32: Wein Bridge oscillator circuit.

and voltage gain A. The equivalent circuit for the amplifier is shown in Figure 5.33.
The numeric labels on the terminals correspond to the labels in Figure 5.32.

(a) Find an expression for the frequency of oscillation (if it occurs).

(b) Find an expression for the value of the voltage gain (A) required to sustain
steady-state oscillations.
5.8. HOMEWORK PROBLEMS 175

1 3
+
v + Av
-
-
2

Figure 5.33: Equivalent circuit for amplifier

R
L
C2
C1

Figure 5.34: Colpitts oscillator

10. Consider the Colpitts oscillator in Figure 5.34 where R represents the series resistance
of the inductor.
The loop gain is given by Equation 5.40, if the transistor immittances are ignored:

gm /! 2 C1 C2
Alo = 1 (5.40)
R + j!L + j! ( C11 + 1
C2 )

The frequency of oscillation is given by

1
!o = q (5.41)
L CC11+C
C2
2

A measure of the relative stability of the oscillator is the slope of the loop-gain phase
characteristic evaluated at the oscillation frequency, i.e.,

d
|!=!o (5.42)
d!
where is the phase of the loop gain.
d 2Q
(a) Show that d! |!=!o = !o where Q is the inductor Q evaluated at the frequency
of oscillation.
(b) Suppose that an oscillator is built at 5MHz with an inductor having a Q = 50.
Estimate the shift in oscillation frequency, if the overall phase of the loop gain is
perturbed by 5 degrees due to temperature drift of the capacitors.
(c) Now assume that the inductor is replaced with a crystal having an eective
Q = 20, 000. Estimate the frequency shift.
176 CHAPTER 5. OSCILLATORS

C R1 L R2

Figure 5.35: Two-stage oscillator circuit.

11. Consider the oscillator circuit in Figure 5.35.

Assume that the transistors are identical and can be represented by the simplified
equivalent circuit in Figure 5.36.

B C
+
vbe gm vbe
-

Figure 5.36: Simplified equivalent circuit.

(a) Find an expression for the (non-zero) frequency of oscillation (if it occurs).
(b) Find the minimum value of gm required for oscillation to occur.

12. Consider the oscillator circuit in Figure 5.37 where a voltage amplifier with 50
input and output impedance and open-circuit voltage gain A is used with a lossy
section of transmission line having characteristic impedance Zo = 50 , length L, and
attenuation and phase shift per unit length denoted by and , respectively. Since
the line is terminated in its characteristic impedance, the voltage at the output of the
line will be related to the voltage at the input of the line by the following equation:
(+j )L
Vout = Vin e (5.43)

where
2f
= (5.44)
vp

and vp is the phase velocity for propagation on the line.

(a) Find an expression for the potential frequencies of oscillation. Assume that the
voltage gain, A, is real (zero phase shift) and that the input and/or output of
the voltage amplifier is AC-coupled with a coupling capacitor (not shown) that
will prevent the circuit from oscillating at zero frequency. The capacitor may be
assumed to have no eect on the circuit performance at finite frequencies.
5.8. HOMEWORK PROBLEMS 177

50
+ +
Lossy T-line
vi 50 Avi Zo , ,
- -

Figure 5.37: Oscillator circuit.

(b) Suppose that the T-line loss per unit length, , is proportional to frequency, i.e.,
= Kf . Define the minimum and maximum values for the voltage gain, A, such
that the circuit will oscillate only at the smallest (non-zero) frequency found in
part 12a.

13. Suppose a capacitor is placed in parallel with a quartz crystal.

(a) Describe how the series and parallel resonant frequencies of the crystal and ca-
pacitor combination will dier from those of the crystal.
(b) Suppose the crystal is used in an oscillator where it operates in the parallel
resonant mode, i.e., the frequency of oscillation is a frequency where the crystal
impedance is inductive. How will the frequency of oscillation change if a capacitor
is added in parallel with the crystal? Explain your reasoning and any underlying
assumptions.

14. The crystal oscillator circuit shown in Figure 5.38 is called a series mode oscillator,
because it will oscillate very close to the series resonant frequency of the crystal.
Here the crystal grounds the base of the transistor at its series resonant frequency.

Vcc

L
R1
C2
RL
Zx R2
Re C1

Figure 5.38: Series mode xtal oscillator derived from common-base Colpitts configuration.

The circuit will also oscillate if the crystal is replaced by an AC short circuit (e.g.,
a bypass capacitor). Use the negative resistance approach to study this oscillator.
You may assume that the transistor can be modeled using the equivalent circuit in
178 CHAPTER 5. OSCILLATORS

B C
+
vbe r gm vbe
-

Figure 5.39: Simplified hybrid-pi model.

Figure 5.39. You can also assume that R1 and R2 can be neglected and that Re is
much larger than the reactance of C2 , so that it can be neglected as well. The coupling
capacitors can be taken to be AC short circuits.

(a) First, remove the crystal and look in to the rest of the circuit. Solve for the
input impedance and show that it is given by

Z1 (ZL + Z2 ) + gm r Z1 Z2
Zin = r + (5.45)
Z1 + Z2 + ZL
where
1
Z1 = (5.46)
j! C1
1
Z2 =
j! C2
j!LRL
ZL =
j!L + RL

(b) Solve for the frequency at which the circuit will oscillate (!o ), if an AC short is
connected from the base of the transistor to ground (instead of the crystal). You
can assume that RL !L at the frequency of oscillation.
(c) What condition must be satisfied in order to guarantee that oscillations will start?

15. Consider the oscillator circuit in Figure 5.40. The active device is modeled as a

1
+ 3
-
2
R2 L C

Figure 5.40: Oscillator circuit.

transconductance amplifier with input resistance R1 as shown in Figure 5.41. (Note

5.8. HOMEWORK PROBLEMS 179

1 3
+
v R1 gm v
-
2

Figure 5.41: Transconductance amplifier with input resistance R1

the important dierence between this equivalent circuit and the simplified hybrid-pi
model!)

(a) Find an expression for the potential frequency of oscillation, !o .

(b) Find the threshold value, gmmin , which the transconductance must exceed in
order for oscillation to occur.

16. Consider the oscillator circuit shown in Figure 5.42. In this problem, Vcc = 3 V, Rb =
33 k, Re = 100 k, C1 = 10 pF, C2 = 22 pF. Unlabeled capacitors are either coupling
or bypass elements and have negligibly small impedance at the frequencies of interest.
You may assume that the load resistor is disconnected from the circuit for parts (a)-(d).

Vcc
not connected for (a)-(d)
Rb

C1
RL L
C2 Re

Zin

Figure 5.42: Load resistor is not connected for parts (a)-(d).

(a) Find an approximate value for the quiescent collector current, ICQ . You may
assume that = 100.
(b) Find the transconductance, gm , for the transistor in this circuit.
(c) Assume that the reactances of C1 and C2 are small enough such that r , Re ,
and Rb can be ignored for small-signal analysis. In this case, if the inductor is
removed the impedance Zin looking in to the circuit at the point shown in the
Figure will be
gm 1
Zin = 2
+ C C2
.
! C1 C2 j! C11+C 2

Ignore the load resistance, RL , (i.e. assume that the load is disconnected) and
calculate the inductance, L, required to set the potential frequency of oscillation
in this circuit to 50 MHz.
180 CHAPTER 5. OSCILLATORS

(d) Suppose a lossy inductor is available with inductance equal to the value that you
calculated in part (c). Calculate the minimum inductor Q (i.e., QL ) required for
oscillations to be sustained in this circuit.
(e) Now suppose that the actual inductor QL is 50. A load is placed on the oscillator
by connecting RL to the inductor through a coupling capacitor as shown by the
dotted line in the figure. Calculate the smallest value of load resistance for which
the circuit will still oscillate.

17. Design a tunable oscillator that covers the frequency range 10 20 MHz using the
Colpitts configuration shown in Figure 5.43 . The coil inductance L = 2H and

R
L
C2
C1

Figure 5.43: Colpitts oscillator with bias circuitry omitted.

the resistance R = 4. For simplicity, assume that the immittances associated with
the transistor and the bias circuitry can be ignored. Furthermore, assume that the
oscillator will be tuned with a single variable capacitor. For this problem, assume that
C2 is the variable capacitor and is adjustable within the limits [C2,min , C2,max ]. As
discussed in the supplementary notes, it is good practice to ensure that C1 C2 to
allow large collector voltage swing with relatively small base-emitter voltage swing.
This allows the transistor to operate without being driven deep into saturation. For
your design, use C1 = 10C2,max in order to satisfy this constraint. Design your circuit
so that the minimum value of the loop gain over the range 10-20 MHz is 2. For full
credit you must specify the value of C1 , the minimum and maximum values for the
variable capacitor (C2min and C2max ), the transconductance, gm , of the transistor,
and the quiescent collector current ICQ .

18. A crystal has packaging capacitance Co = 5 pF and series and parallel resonant fre-
quencies fs = 10 MHz and fp = 10.002 MHz.

(a) Determine the inductance, L1 , and capacitance, C1 , of the elements in the mo-
tional arm of the equivalent circuit for the crystal.
(b) The crystal is used in a parallel-resonance oscillator in which the crystal resonates
with an external load capacitance of 30 pF at the frequency of oscillation. The
frequency of oscillation can be written as fo = 10.000 MHz + f . Specify f to
an accuracy of +/- 5 Hz.

19. Consider the crystal oscillator shown in Figure 5.44. Without the network consisting
of the components labeled L and C, this circuit would oscillate near the fundamental
(lowest) resonant frequency of the crystal, denoted by fo , where fs < fo < fp , and fs ,
5.8. HOMEWORK PROBLEMS 181

C1 C

C2 L

Figure 5.44: Crystal oscillator circuit for operation on 3rd overtone.

fp are the lowest series and parallel resonant frequencies of the crystal. The purpose
of the series LC network is to make it impossible for the circuit to oscillate at the
fundamental frequency, and to allow the circuit to oscillate at the 3rd overtone, 3fo .
If fo = 50 MHz, C1 = 10 pF, C2 = 30 pF, C = 500 pF, find the range of values for
L that will prevent the circuit from oscillating at fo , and allow the circuit to oscillate
at 3fo . For this problem, consider only the fact that the combination of C2 , C, and L
must be capacitive in order for the circuit to oscillate at a frequency where the crystal
is inductive. You do not need to ensure that the magnitude of the loop gain will be
sucient to actually cause oscillations to start and grow.
20. The circuit shown in Figure 5.45 is often used as an overtone XTAL oscillator. In this

C L
C1

Figure 5.45:

problem, we consider how to design the circuit so that it oscillates at the 5th overtone
as a parallel-mode oscillator. Note that bias resistors and coupling/bypass capacitors
are not shown and may be ignored. You may also ignore the internal immittances of
the transistor.

(a) Consider only the XTAL in this part. Denote the series and parallel resonant
frequencies associated with the 5th overtone by fs,5 and fp,5 , respectively. It
is known that fs,5 = 49.990 MHz and fp,5 = 50.030 MHz and the packaging
capacitance, Co , is 5 pF. Find the external load capacitance, CL , required to
form a parallel resonant circuit with the XTAL at exactly 50.000 MHz.
(b) Write down a constraint that must be satisfied by the resonant frequency of the
parallel LC branch in the circuit to allow the circuit to oscillate at the fifth
overtone, and not at the fundamental or third overtone. Denote the desired
frequency of oscillation by fo . The XTALs fundamental resonances are near
fo /5 and the third overtone resonances near 3fo /5.
182 CHAPTER 5. OSCILLATORS

(c) Write an equation in terms of CL , C1 , L, and C which must be satisfied (along

with the constraint derived in part b.) to set the lowest potential frequency of
oscillation to exactly 50.000 MHz ( fo = 50.000 MHz). You do not need to solve
the equation.

21. Consider the Colpitts oscillator shown in Figure 5.46.

C1
L CV
C2

Figure 5.46: Voltage-controlled oscillator (VCO).

(a) Find an expression for the frequency of oscillation, !o . You may assume that the
loop gain of the circuit is large enough to cause oscillation to start. You may
neglect the transistor immittances, so that your result will be in terms of only
the four parameters L, C1 , C2 , CV .
(b) Denote the minimum and maximum tuning-diode capacitances by CV,min and
C
CV,max and the capacitor tuning ratio by rC = CV,max V,min
. Denote the minimum
and maximum frequencies of oscillation by !o,max and !o,min and the oscillator
!
tuning ratio by ro = !o,max
o,min
. Finally, denote the series combination of C1 and C2
by C 0 , i.e. C 0 = C1 +C2 .
C1 C2
Find an expression for ro in terms of rC , C 0 , and CV,max
only.
(c) Find a numerical value for the tuning ratio ro if C 0 = 2CV,max and rC = 4.
Chapter 6

Impedance Matching Networks

6.1 Impedance Matching for Maximum Power Transfer

In this section we review the motivation for impedance matching and introduce important
concepts which will be used in later chapters. Let us first illustrate the basic principles of
impedance matching for maximum power transfer. In Figure 6.1 a source with impedance
ZS = RS + j XS is connected to a load ZL = RL + j XL . The peak voltage of the source
(assumed to be sinusoidal) is VS :

ZS +

VS ZL VL
-

Figure 6.1: A voltage source driving an arbitrary load

The time-averaged real power delivered to the load can be written as

1
PL = Re{VL IL } (6.1)
2
where VL and IL are the voltage across and current through the load impedance, respectively.
Note that the factor 12 is present because VL and IL are peak phasors,i.e., phasors whose
magnitude is the peak value of the sinusoidal time function. The current through the load
is
VL
IL = (6.2)
ZL
so the time-averaged real power is

1 1
PL = |VL |2 Re{ } (6.3)
2 ZL
1 RL
= |VL |2
2 |ZL |2

183
184 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

This can be written in terms of the source voltage VS as follows:

1 RL
PL = |VS |2 (6.4)
2 |ZL + ZS |2
1 RL
= |VS |2
2 (RL + RS ) + (XL + XS )2
2

By studying Equation 6.4 we can determine what conditions are required to maximize
the power delivered to the load. First we will assume that the source voltage and source
impedance are set at the outset, i.e., they are not under our control. This will usually be
the case. The problem is then to choose RL and XL to maximize PL . We will restrict the
solution to values of RL that are greater than or equal to 0 because values of RL that are
< 0 would require that the load contain an active device, whereas we are considering only
passive load terminations. The solution to this simple exercise is

ZL = ZS (6.5)

This result states that the time-average real power delivered to a load is maximized when the
load impedance is equal to the complex conjugate of the source impedance. Using Equation
6.5 in Equation 6.4, we obtain an expression for the maximum power that the source can
deliver to an external passive load. This power is referred to as the power available from
the source, Pavs :

Pavs PL |ZL =ZS (6.6)

|VS |2
=
8RS

The concept of available power is often used in the radio frequency literature. For example,
the output power level of RF signal generators is usually specified by stating the available
power in dBm, i.e., decibels referred to 1 mW. A power level of 6 dBm indicates a power
6 dB higher than 1 mW, or approximately 4 mW. It is important to realize that when a
signal generator is configured for a given output power level, that power level is the power
available from the generator, Pavs , which is the power that the generator will deliver to
a conjugately matched load. Signal generators commonly have source impedances of 50
(sometimes 75 ), i.e., the generator will deliver the rated power only to a 50 load. With
a dierent load impedance the power delivered to the load will be less than the rated value.

6.1.1 Mismatch Factor

The degree to which the actual power delivered to an arbitrary load is smaller than the
available power can be quantified in terms of a mismatch factor , M F , a quantity that
depends on the degree of impedance mismatch between the source and load. For an arbitrary
load impedance ZL the mismatch factor is defined as the ratio of actual delivered power to
available power:
6.1. IMPEDANCE MATCHING FOR MAXIMUM POWER TRANSFER 185

PL
MF =
Pavs

4RS RL
= (6.7)
(RS + RL )2 + (XS + XL )2

4RS RL
=
|ZS + ZL |2
Mismatch factor is a real number and 0 M F 1. For given source and load
impedances, the mismatch factor tells us what fraction of the available power will be deliv-
ered to the load. Since the mismatch factor is a ratio of two power levels, it makes sense to
express the quantity in decibels. The mismatch loss, M L, in decibels is defined as

ML = 10 log M F. (6.8)

6.1.1.1 Example - Mismatch Factor and Mismatch Loss

Suppose a 50 signal generator has available power of 1 mW. The generator is to drive a
load impedance of 250 + j100 . What is the power delivered to the load?
The problem could be solved by computing the power delivered to the load using Equa-
tion 6.1. Another approach is to compute the mismatch factor from Equation 6.7. This gives
a mismatch factor M F = 0.5 , so PL = Pavs M F = (1 mW)0.5 = 0.5 mW. Alternatively, we
can express the powers in dBm and use the mismatch loss in dB. The mismatch loss M L =
10 log(0.5) = 3 dB. The power available from the source is 1 mW, or 10 log 11 mW
mW = 0 dBm.
The power delivered to the load is PL = Pavs M L = 0 dBm 3 dB = 3 dBm. Note that
a power level of 3 dBm is equal to 0.5 mW.

6.1.2 Properties of Lossless Impedance Matching Networks

We have illustrated how impedance matching (or mis-matching) influences the transfer of
power from a source to a load. In many applications it is desirable to maximize the transfer
of power from the source to the load. This can be achieved by using a lossless 2-port
network inserted between the source and the load. The purpose of the matching network is
to transform the load impedance, ZL , into ZS at the input terminals, thereby permitting
the source to deliver all of its available power to the network. If a matching network is
lossless, then all of the power that is delivered to the network must be delivered to the load.
This simple concept has implications that may not be obvious at first glance.
Consider a source and load connected through a lossless matching network as shown in
Figure 6.2. As already noted, the matching network transforms the load impedance ZL into
ZS at the input terminals of the network. Since the source looks into the conjugate of its
impedance, it delivers all of its available power Pavs to the network. Because no power is
dissipated in a lossless network, all of this power is delivered to the load. Now the power
available at the output of the lossless matching network must be the same as the power
available from the source, and we have already stated that all of this available power is
delivered to the load. Therefore the load must be conjugately matched to the output of
the matching network, which means that the impedance at the output of the network (as
seen by the load) is ZL . Thus a lossless matching network has the property of providing a
simultaneous conjugate match at the input and output ports.
186 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

Pavs Pavs

ZS
Lossless
VS Matching ZL PL = Pavs
Network

ZS ZL

Figure 6.2: Source and load coupled by a lossless matching network

Another way to look at this is to note that the original source plus the matching network
can be viewed as a new source with available power Pavs and source impedance ZL . Adding
a lossless network to the output of a source does not change the available power from the
source; it only changes the source impedance. Thus, we can think of the matching network
as (i) a network that transforms the load impedance into ZS or (ii) a network that transforms
the source impedance into ZL without changing the available power of the original source.
The network can be designed either way.
Most lumped-element matching networks are versions of ladder networks as shown in
Figure 6.3. The ladder-type matching network in Figure 6.3 is assumed to be lossless; the

Pavs Pavs

ZS
VS ZL PL = Pavs

ZS ZL

Z Z

Figure 6.3: Ladder-type matching network

blank boxes represent purely reactive circuit elements. Now suppose that the network is
broken at the dashed line. The part of the network to the left of the line can be thought
of as a source. It will have the same available power as that of the original source, Pavs , and
some source impedance, Z. The part of the network to the right of the dashed line is a load
for this source and must receive all of the available power (since that power is ultimately
transferred to ZL ). Thus, a conjugate match must exist at the junction defined by the
dashed line. Clearly, the dashed line could have been drawn anywhere within the lossless
network and the same argument would hold. The conclusion is: A circuit consisting of
a source connected to a load through a lossless matching network can be broken
at any point between the source and the load and a conjugate match must exist
between the two sides of the circuit.
6.2. IMPEDANCE MATCHING WITH LOSSLESS L-NETWORKS 187

6.2 Impedance Matching with Lossless L-networks

6.2.1 Resistive Terminations
Figure 6.4 shows two resistances to be matched with a lossless L-network. The goal is to
transform R1 to R2 at one frequency. The unknown reactances Xs and Xp are easily found
with the use of a parallel-to-series transformation as in Figure 6.5.

jXs
ZIN = R2 jXp R1

Figure 6.4: L-network which transforms R1 into R2

X R2
p 1
j R2 +X 2
1 p

jXs
R1 Xp2
ZIN = R2 R12 +Xp2

Figure 6.5: Parallel-to-series transformation

To make the input impedance equal to R2 , we choose

Xp R12
Xs = (6.9)
R12 + Xp2

R1 Xp2
R2 =
R12 + Xp2

Solving for Xp and Xs gives

r
R2
Xp = R1 (6.10)
R1 R2
q
Xs = R2 R 1 R22 (6.11)

The solutions yield real values for Xp and Xs (and hence purely reactive L-network com-
ponents) only if R1 > R2 . This leads to a rule for using a lossless L-network to match two
resistances:

The shunt arm of the L-network is connected across the larger of the two resistances.
188 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

Also note that there are two possible solutions for the resistive matching problem corre-
sponding to the upper and lower signs in Equation 6.10 and Equation 6.11. These solutions
are referred to as low-pass and high-pass solutions. The circuit configurations for the
two solutions are in shown Figures 6.6 and 6.7.

R2 R1 R 1 > R2

Figure 6.6: High-pass L-network

R2 R1 R 1 > R2

Figure 6.7: Low-pass L-network

The high-pass and low-pass designations refer to the behavior of the network transfer
function at frequencies away from the design frequency.

6.2.2 Q of an L-network
Referring to Figure 6.5, it is apparent that the transformed L-network looks like a series-
resonant circuit. At the design frequency the two reactances have equal magnitudes and
opposite signs, so that the net series reactance is zero. Treating this circuit in the same
manner as a resonant RLC network, the Q of the network is the magnitude of one of the
series reactances divided by the series resistance:

Xp R12 /(R12 + Xp2 )

Q = 2)
(6.12)
R1 Xp2 /(R12 + Xp
(6.13)
R1
= (6.14)
Xp
(6.15)
r
R1
= 1 (6.16)
R2
This result shows that the Q of the L-network is determined by the terminating resistances
R1 and R2 , and, therefore, cannot be chosen independently. This suggests that the band-
width of the matching network depends only on the ratio R1 /R2 .
6.2. IMPEDANCE MATCHING WITH LOSSLESS L-NETWORKS 189

A word of caution is in order. We have defined the Q of the L-network by analogy with
the series RLC network. This analogy works well only when R1 |Xp | or, equivalently,
when Q 1. In such cases the Q can be used to predict the 3 dB bandwidth of the networks
voltage transfer function. For moderate or small values of Q the expression
r
R1 R1
Q= = 1
|Xp | R2
is still valid, but a simple relationship between Q and bandwidth does not exist, since
the series reactance that results from the parallel-to-series transformation has a dierent
frequency dependence from that of a simple inductor or capacitor. Thus an L-network
does not behave exactly like a series RLC circuit. Only when the Q is very large will the
equivalent series reactance behave approximately like a capacitor or inductor.

6.2.3 Summary: L-network design equations

The design of an L-network can be summarized as follows. If the terminating resistances
are denoted by Rbig and Rsmall (where Rbig > Rsmall ), then the design equations can be
written in terms of the network Q,
r
Rbig
Q= 1, (6.17)
Rsmall
as
Xp = Rbig /Q Xs = Rsmall Q. (6.18)
When the upper signs are chosen, the resulting network is of highpass type. The lower signs
give a lowpass network. The L-network will be oriented such that the parallel arm is in
shunt with Rbig .

6.2.3.1 Example - Matching resistive source and load with a low-pass L-network
Match a 100 source to a 25 load with a lossless L-network having a low-pass topology.
Since the series arm of the L-network connects to the smaller of the two resistances, the L-

100

25

Figure 6.8: Example for L-net matching

network will be oriented as shown in Figure 6.9. A capacitor and inductor have been chosen
for the shunt and series elements, respectively, because the problem statement specified a
low-pass network.
q q p
Rbig 100
Q = Rsmall 1 = 25 1 = 3

Rbig 100
Xp = XC = Q = p
3
= 57.7
p
Xs = XL = QRsmall = 3 25 = 43.3
190 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

Figure 6.9: L-network orientation

The completed network is shown in Figure 6.10. It is interesting to compare the power

100 j43.3

25

j57.7

Figure 6.10: Completed lossless L-network with low-pass topology

delivered to the load with and without the matching network in place. Suppose the peak
voltage of the source is 1 Volt. Then the power delivered to the load without the matching
network would be
2
1 |Vout |2 1 1
Pout = = = 0.8 mW (6.19)
2 25 50 5
With the matching network the source sees 100 ; therefore, the power delivered to the
matching network is
1 (1/2)2 1
Pin = = = 1.25 mW (6.20)
2 100 800
Since the matching network is lossless, all of this power will be delivered to the load. Thus,
with the matching network, Pout = 1.25 mW. The improvement gained is

10 log(1.25/.8) ' 2 dB (6.21)

6.2.4 Matching Complex Loads with a Lossless L-network

So far, only purely resistive source and load terminations have been considered. When
complex source and loads are involved, there are two basic conceptual approaches that can
be used:

1. Absorption - absorb the source or load reactance into the matching network.

2. Resonance - series or parallel resonate the source or load reactance at the frequency
of interest.

These approaches will be illustrated by example in the following sections.

6.2. IMPEDANCE MATCHING WITH LOSSLESS L-NETWORKS 191

6.2.4.1 Example - Absorption

Absorption will be illustrated with an example. Suppose that it is necessary to match the
source and load shown in Figure 6.11 at 100 MHz with a lossless L-network.

100 j126
Lossless
Matching 750
Network
j250

Figure 6.11: L-network with complex source and load

Absorption is applied by lumping the source and load reactances into the series and
parallel reactances of the matching network, as shown in Figure 6.12. The lumped reactances

jXs0

100 j126
jXs
jXp 750

j250

jXp0

Figure 6.12: Absorbing the complex part of the load impedance into the network

Xs0 and Xp0 can be found by using the design equations for matching between two resistive
terminations:
p
Q = 750/100 1 = 2.55

Xs0 = 100 (2.55) = 255 (6.22)

Xp0 = 750/2.55 = 294.1

The lumped reactances can be written in terms of the reactances associated with the source
and load, and the L-network reactances:

Xs0 = Xs + 126, (6.23)

250 (Xp )
Xp0 = . (6.24)
250 + Xp
192 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

Thus, values of Xs and Xp can be obtained:

129
Xs = Xs0 126 = 255 126 =
381
(6.25)
250Xp0 1667.2
Xp = 250+Xp0 =
135.1

The two solutions found so far are shown in Figure 6.13.

j129 j381
0.205H 4.18pF

j1667.2 j135.1
2.65H 0.215H

Figure 6.13: Two L-network solutions.

The element values (in H and pF) were obtained from the calculated reactances using
the design frequency of 100 MHz. It should be noted that the two solutions found so far
are not the only possibilities for this particular source and load. This becomes apparent if
series-to-parallel transformations are applied to the source and load. After transformation,
the source and load look like Figure 6.14.

j225

258.8 j205.4 75

Figure 6.14: Source and load after application of series-to-parallel transformation

For simplicity, the Norton equivalent current source has been omitted from the source
representation. Note that the smaller of the two resistances is now on the load side and
hence it is possible to match the source and load using L-networks oriented as in Figure
6.15.

jXs
jXp

Figure 6.15: L-network reoriented for smaller resistance on load side

6.2. IMPEDANCE MATCHING WITH LOSSLESS L-NETWORKS 193

The values of Xs and Xp could be found using absorption. Alternatively, the reso-
nance concept can be used. For illustration, the resonance concept will be used to find the
remaining L-network solutions.

6.2.4.2 Example - Resonance

Continuing with the example from the previous section, the resonance concept will be em-
ployed to find two more L-network solutions for the source and load shown in Figure 6.11.
After transforming the source from series to parallel form, and the load from parallel to
series form, the source and load are represented as in Figure 6.14. To apply the resonance
concept, the source and load are augmented with reactances that resonate with the source
and load reactances as shown in Figure 6.16.

j225 j225

258.8 75

j205.4 j205.4

Figure 6.16: Transformed source and load augmented with resonating reactances

After resonating the source and load in this manner, the new source and load impedances
are purely real (resistive) at the design frequency. An L-network is then designed to match
these two resistances, i.e.,
q
Q = 258.76 1 = 1.57
75

Xs0 = 75 (1.57) = 117.8 (6.26)

Xp0 = 258.76 /1.57 = 164.8

Now the circuit can be drawn as shown in Figure 6.17. To complete the design, the resonating

j225 j225
j117.8
258.8 j164.8 75

j205.4 j205.4

Figure 6.17: L-net with resonating reactances

reactances must be incorporated into the matching network. For example, when the upper
signs are chosen, the net parallel reactance will be j205.4|| j164.8 = j91.4. The
net series reactance will be j117.8 + j225 = j342.8. The final solutions are shown in
Figure 6.18.
194 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

j342.8 j107.2

j91.4 j833.7

Figure 6.18: Resonating reactances incorporated into matching network

Thus, we have found four possible solutions that can be obtained using a lossless L-
network.
The example considered here allowed 4 possible L-network solutions because transform-
ing the source and load caused Rbig and Rsmall to swap positions. This will not always be
the case. Thus, for some source and load combinations, there will be only two L-network
solutions, and in other cases there will be four solutions.

6.3 Harmonic Attenuation in Lossless Matching Networks

Using Traps
In certain applications it is necessary to provide a match at some frequency !o and to
attenuate one or more other frequencies. Typically the undesired frequencies are harmonics
of !o , i.e., 2!o , 3!o , etc.; however, they could also be subharmonic or even unrelated to !o .
We will concentrate on the case where the undesired frequencies are harmonics. General-
ization to other cases is straightforward. The basic idea is to first design a lossless matching
network to provide a match at the desired frequency (!o ). This matching network might
consist of an L-, T-, or Pi-network. Next, one or more series elements of the network are
replaced with parallel L-C-networks. Alternatively, or in addition, the shunt elements of
the network are replaced with series L-C-networks. The replacement (series or parallel L-C)
elements are designed to have the same reactance as the original elements at !o ; however,
they are designed to be resonant at the undesired harmonic frequency. As an example,
consider Figure 6.19 which shows a matching network with a match at fo = 107 /2 Hz.

jXL = j111.8

50 300

jXC = j134.2

Figure 6.19: L = 11.18 H, C = 745.2 pF

Suppose it is desired to trap the second harmonic 2fo in the shunt arm. The shunt
arm would be replaced with a series L-C that has reactance 134.2 at fo and is series
6.3. HARMONIC ATTENUATION IN LOSSLESS MATCHING NETWORKS USING TRAPS195

resonant (looks like a short circuit) at 2fo . The design equations are
p
2 !o = 1/ LC (6.27)
1
!o L = 134.2 (6.28)
!o C
The solution is
L = 4.47 H (6.29)
C = 558.9 pF
The new network looks like Figure 6.20.
!0

11.18H

558.9pF
2!0
300
4.47H

Figure 6.20: Network with trapped second harmonic

The series trap acts to shunt the component at frequency 2fo to ground.
The third (or any other) harmonic could be trapped in the series arm. The inductor
would be replaced with a parallel LC whose element values are found from
1
3!o = p (6.30)
LC
1 1
+ !o C = (6.31)
!o L 111.8
The solution is

C = 111.8 pF (6.32)
L = 9.94 H
Figure 6.21 shows the network with both traps installed. The network will look like an open
circuit to the 3fo component.
It should be noted that this approach will work only when applied to capacitive shunt
elements and inductive series elements if frequencies higher than !o are to be trapped. Thus,
the original network must have a low-pass topology. The student should convince himself
or herself that such is the case. On the other hand, if frequencies smaller than !o are to
be trapped, then the approach can be employed only with inductive shunt elements and
capacitive series elements.
The trapping approach is also useful for reducing the feed-through of nonharmonic fre-
quency components. If the frequencies of the undesired components are very close to the
desired frequency, it may not be possible to build an eective trap, because the Q of the
series or parallel L-C circuits will be too high to be realized.
196 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

!0
111.8pF

3!0
558.9pF
9.94H 300
2!0 4.47H

Figure 6.21: Network with second and third harmonics trapped

6.4 Three-element Matching Networks

The L-network does not give the designer freedom to choose the Q (bandwidth) or phase
shift of the matching network. The addition of a third matching element makes it possible
to design for a match and a specified phase shift or Q. The following section describes a
procedure for designing 3 element matching networks with specified Q. Then, a general
procedure allowing for specified attenuation and phase shift will be discussed.

6.4.1 Design of Pi- and T-networks for Specified Bandwidth (Q)

First consider the Pi-network with a low-pass topology. In addition, we assume that ab-
sorption has been used so that the shunt reactance of the source or load impedance is
included in the Pi-network. The problem is then reduced to matching between two resistive
terminations as shown in Figure 6.22 (where it is assumed that R2 > R1 ).

jXL

R1 jXC1 jXC2 R2

Figure 6.22: Matching two resistive terminations where R2 > R1

The Pi-network can be thought of as two back-to-back L-networks that act to match
both R1 and R2 to a virtual resistance Rv as shown in Figure 6.23.
Because the series arms of both L-networks are connected to Rv , it is clear that Rv is
smaller than R1 and R2 . Define the Qs of the two L-networks to be Q1 and Q2 where
r r
R1 R2
Q1 = 1 , and Q2 = 1 (6.33)
Rv Rv
We are assuming that R2 > R1 , and therefore Q2 will be larger than Q1 . For most practical
purposes the Q of the Pi-network can be approximated by Q2 . This is especially true if
R2 R1 . If R2 is only slightly larger than R1 , then the overall Q of the network will be
somewhat larger than Q2 . The design procedure follows:
6.4. THREE-ELEMENT MATCHING NETWORKS 197

jXL1 jXL2

R1 jXC1 jXC2 R2

Rv Rv

Figure 6.23: The Pi-network as 2 back-to-back L-networks

1. Determine the required Q of the matching network by considering the required band-
width, BW , and center frequency, fo (Q = fo /BW ). This Q is taken to be equal
to Q2 , and thus the virtual resistance, Rv , is determined. Note that Rv must be
smaller than R1 , and therefore the Pi-network can only be used to obtain
a larger Q than would have been provided by the simpler L-network. Also
note that the relationship BW = fo /Q is only exactly true for a simple parallel or se-
ries RLC circuit. Thus, the actual bandwidth of your circuit may be dierent from the
specified design value. If a particular design requires that the bandwidth be precisely
determined, it is a good idea to simulate the performance of the matching network
using a computer-aided design program in order to verify that the performance will
be satisfactory.

2. Once Rv is found, the values of XC1 , XL1 , XC2 , and XL2 can be calculated using the
previously derived formulas for L-network matching.

The design procedure can be summarized by Equations 6.34, where Q2 is determined by the
desired bandwidth. You should verify that these equations result from steps (1) and (2):

R2
XC2 = (6.34)
Q2
s
R1 R2
XC1 = R2
(6.35)
(Q22 + 1) R1
(6.36)
(6.37)
q
R1 2
R2 Q 2 + R2 R2 (Q2 + 1) 1
XL = (6.38)
Q22 +1

The Pi-network is most useful for matching when the values of R1 and R2 are not too small.
If R1 and R2 are small, the virtual resistance will be even smaller, and the capacitor values
will turn out to be impractically large. If either terminating resistance is significantly less
than 50 , the T-network will usually be a more practical choice. One possible T-network
is the band-pass case shown in Figure 6.24.
As before, we can think of this network as two back-to-back L-networks as shown in
Figure 6.25.
198 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

jXC1 jXL2

R1 jXL3 R2

Figure 6.24: Band-pass T-network

jXC1 jXL2

R1 jXL0 jXC0 R2

Rv Rv

Figure 6.25: Band-pass T-network as 2 back-to-back L-networks

The Qs of the two L-networks are

r r
Rv Rv
Q1 = 1 and Q2 = 1 (6.39)
R1 R2

Note that since Rv > R1 and Rv > R2 , this network will have a larger Q than a single
L-network that matches R1 to R2 . The overall Q of the network is set by Q1 since we
assume R2 > R1 . The design formulas are

XC1 = R1 Q 1 (6.40)
r
R1 2
XL2 = R2 (Q + 1) 1 (6.41)
R2 1
(6.42)
R (Q2 + 1)
XL3 = q1 1 (6.43)
R1 2
Q1 R2 (Q1 + 1) 1

Note carefully that the two elements XL0 and XC0 that appear in the back-to-back L-networks
are combined into a single element, XL3 , in the T-network. In practice this is not necessary
and, in fact, may not be desirable. Whether or not the elements are combined makes no
dierence at the design frequency, but it may make a significant dierence at frequencies
well removed from the design frequency. The dierent possibilities are best examined using
a CAD program.
You should be able to derive similar formulas for the other possible topologies, e.g.,
band-pass Pi, low-pass T, etc.
6.4. THREE-ELEMENT MATCHING NETWORKS 199

6.4.2 Matching Two Resistive Terminations with Specified Atten-

uation and Phase Shift
In this section we will consider a more general type of matching network than was considered
in section 6.2. Specifically, we allow the matching element impedances to be complex in the
initial development of our solution. This will allow for a solutions with specified attenua-
tion and phase shift. Then we will specialize to the cases where the elements are purely
reactive (lossless networks) and purely resistive (lossy networks or attenuators). Note that
the terminating impedances are assumed to be real (resistive).

6.4.2.1 Pi-network with specified attenuation and phase shift

Referring to Figure 6.26 we make the following assumptions:

Y1 , Y2 are assumed to be real.

YA , YB , YC may be complex.

IIN IOUT
YC
+ +
Y1 VIN YA YB Y2 VOUT
- -

Y1 Y2

Figure 6.26: Matching two resistive terminations with a Pi-network

The conditions required for a match can be summarized as follows:

1. Must see Y1 looking in from the left when terminated with Y2 on the right. Thus,

(Y2 + YB )YC
Y1 = YA + (6.44)
Y2 + YB + YC

2. Must see Y2 looking in from the right when terminated with Y1 on the left. Thus,

(Y1 + YA )YC
Y2 = YB + (6.45)
Y1 + YA + YC

So far, we have 2 equations but 3 unknowns. The third equation can be used to specify either
the Q of the network (and therefore, the bandwidth) or the phase shift and attenuation.
For now, we consider the latter. The third equation can be written in the form:
r
+j VIN IIN
e = (6.46)
VOU T IOU T
Equation 6.46 can be interpreted as follows. Since Y1 and Y2 are assumed to be real,
the input voltage and current will be in phase and so will the output voltage and current.
200 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

Denoting the phase of the input voltage (current) by IN and the phase of the output voltage
(current) by OU T

s
+j |VIN | |IIN |
e = ej(IN OU T )
(6.47)
|VOU T | |IOU T |

r
PIN j(IN OU T )
= e
POU T

Thus
= IN OU T (6.48)

and
1 PIN
= ln (6.49)
2 POU T
Therefore, is the phase shift of the network and is the power attenuation expressed in
nepers.
Equations 6.44, 6.45 and 6.46 must be solved for the unknowns YA , YB and YC . Note that
Equation 6.46 can be written in terms of the unknowns as shown below. Define = + j ,
then
r s
VIN IIN VIN IIN /VIN
e = = (6.50)
VOU T IOU T VOU T IOU T /VOU T
(6.51)
r
VIN Y1
= (6.52)
VOU T Y2
(6.53)
r
Y1
e = (1 + YC 1 (Y2 + YB )) (6.54)
Y2

Equations 6.44, 6.45 and 6.46 can now be solved. The result is
p
Y1 Y2
YC = (6.55)
sinh
Y2
YB = YC (6.56)
tanh
Y1
YA = YC (6.57)
tanh

6.4.2.2 T-network with specified attenuation and phase shift

Similar considerations apply to using a T-network to match two resistive terminations as

shown in Figure 6.27.
6.4. THREE-ELEMENT MATCHING NETWORKS 201

Z1 IIN IOUT
ZA ZB
+ +
VIN ZC Z2 VOUT
- -

Z1 Z2

Figure 6.27: Matching two resistive terminations with a T-network

The equations that ensure a simultaneous conjugate match at both ports are

ZC (ZB + Z2 )
Z1 = ZA + (6.58)
ZC + ZB + Z2
ZC (ZA + Z1 )
Z2 = ZB + (6.59)
ZC + ZA + Z1

The third equation is the same as before,

r
VIN IIN
e+j = e = (6.60)
VOU T IOU T

which can be written r

IIN Z1
e = (6.61)
IOU T Z2
The solutions for ZA , ZB and ZC are

p
Z1 Z2
ZC = (6.62)
sinh
Z2
ZB = ZC (6.63)
tanh
Z1
ZA = ZC (6.64)
tanh
The interpretation of = + j is the same for the T-network as it was for the Pi-network,
that is,

= IN OU T ) voltage (current) phase shift in radians

(6.65)
= 1
2 ln PIN
POU T ) power attenuation in nepers

Remember that these discussions have assumed that Y1 and Y2 (or Z1 and Z2 ) are real
(resistive). Complex loads are handled by incorporating the reactive part of the termination
into the matching network using either resonance or absorption as discussed in section 6.2.
202 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

6.4.3 Design of Lossless Pi- and T- Matching Networks with Spec-

ified Phase Shift
The solutions found so far allow the designer to specify both phase shift and attenuation of
the network. In practice one is usually concerned with one of the special cases: (i) YA , YB ,
YC or ZA , ZB , ZC are purely reactive this corresponds to the lossless matching network;
(ii) YA , YB , YC orZA , ZB , ZC are purely resistive this corresponds to a lossy network.
Networks of type (ii) are often used to provide specified amounts of attenuation and/or
isolation between circuits. In this section we will consider the lossless matching networks.
When YA , YB , YC or ZA , ZB , ZC are purely reactive, then the network is lossless and
= 0. Hence = j and the design equations reduce to
Pi:

p
Y1 Y2
YC = (6.66)
j sin
Y2
YB = YC (6.67)
j tan
Y1
YA = YC (6.68)
j tan

T:

p
Z1 Z2
ZC = (6.69)
j sin
Z1
ZA = ZC (6.70)
j tan
Z2
ZB = ZC (6.71)
j tan

A word of caution on interpreting when source and/or load admittances are complex.
We have seen that when the source and load are resistive, can be interpreted as the voltage
or current phase shift, since voltage and current are in phase at both the input and output.
If the source and/or load are complex, then the voltage and current are not in phase. In
this situation has a dierent interpretation for the Pi- and T-networks. This can be seen
by considering Figure 6.28.

R1 IIN IOUT
jX1 jXA jXB jX2
+ +
VIN jXC VOUT R2
- -

Figure 6.28: T-network with complex source and load

6.4. THREE-ELEMENT MATCHING NETWORKS 203

The network would be designed by incorporating X1 and X2 into the matching network
as shown in Figure 6.29.

j(X1 + XA) j(XB + X2)

R1 IIN IOUT

+ +
0 0 R2
VIN jXC VOUT
- -

Figure 6.29: Network designed with X1 and X2 incorporated into the matching network

The currents IIN and IOU T in Figure 6.29 are the same as those shown in Figure 6.28,
but the voltages are not. In fact, VIN
0
and VOU
0
T are related to the VIN and VOU T of Figure
6.28 by

0 R2
VOU T = VOU T (6.72)
R2 + j X 2

0 R1 jX1
VIN = VIN (6.73)
R1

Now, is the phase shift between IIN and IOU T or, equivalently, VIN 0
and VOU0
T . It is
not the phase shift between VIN and VOU T , however. The voltage phase shift (VIN VOU T )
can be determined as follows:

VIN V0 R1 jX1 R2
= 0IN (6.74)
VOU T VOU T R2 + jX2 R1
Define

1 X1
Z1 = tan (6.75)
R1

1 X2
Z2 = tan (6.76)
R2
Then 0 0
VIN VOU T = VIN VOU T Z1 Z2
(6.77)
= Z1 Z2
Thus for the lossless T-network gives the current phase shift, and the voltage phase shift
can be found from Equation 6.77.
Similar considerations lead to the conclusion that gives the voltage phase shift for the
lossless Pi-network. In this case Equation 6.78 yields the current phase shift:

IIN IOU T = Y1 Y2 (6.78)

204 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

6.4.3.1 Example - Design of a lossless Pi-network with specified phase shift

Design a lossless Pi-network to match a load of 300 to a 50 source with VOU T leading
VIN by 45 at ! = 107 rad/s as in Figure 6.30. We want VOU T VIN = 45 . Hence

VOUT

45

VIN

Figure 6.30: Phasor diagram showing VOU T leading VIN by 45 .

= 45
Y1 = 1
50 = 20 mS

Y2 = 1
300 = 3.3 mS
p
20(3.33)
YC = j sin( /4) = +j 11.54 mS (6.79)

3.33 mS
YB = j tan( /4) j1154 mS = j 8.21 mS

20 mS
YA = j tan( /4) j1154 mS = +j 8.46 mS
The final solution is shown in Figure 6.31. The network will have a band-pass transfer

1154pF

50 846pF 12.2H 300

Figure 6.31: Pi-network solution

function characteristic, as the series capacitance and shunt inductor guarantee that the
network transfer function will have zero response at DC, whereas the shunt capacitor will
provide high frequency attenuation which increases at 6 dB per octave for frequencies well
above the design frequency.
Complex source or load impedances can be handled by incorporating the load reactances
into the network, as illustrated in the following example.

6.4.3.2 Example - Matching complex load with a specified current phase shift.
At ! = 107 rad/s design a lossless matching network to match a load impedance of 150 - j
75 to a generator having an internal impedance of 50 . The output current is to be in
6.4. THREE-ELEMENT MATCHING NETWORKS 205

phase with the input current as in Figure 6.32. Start by trying to find a T-network solution:

50 j75

IIN IOUT
150

Figure 6.32: Example

= IIN IOU T = 0 (6.80)

sin = 0
tan = 0

Thus, ZA , ZB , ZC ) 1. This illustrates that 0 current phase shift cannot be obtained

with a T-network!
Let us consider a Pi-network. For the Pi-network, is the voltage phase shift. As noted
earlier, can be written in terms of the current phase shift and the phase angles of the
terminating impedances:

= IIN IOU T Z1 Z2 = 0 0 0 Z2 = Z2

75 1
Z2 = tan = 26.6 (6.81)
150
Thus = +26.6 . To proceed, the load is transformed so that absorption can be used as in
Figure 6.33. Then

j75

150 187.5

j375

Figure 6.33: Transform load

p
(1/50)(1/187.5)
YC = j sin 26.6 = j 23.07 mS
(6.82)
1/50
YA = j tan 26.6 + j 23.07 mS = j 16.9 mS

1 1/187.5
+ YB = + j 23.07 mS
j 375 j tan 26.6
YB = j 9.73 mS (6.83)
The final result is shown in Figure 6.34.
206 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

4.34H

5.92H 973pF

Figure 6.34: Final Result.

This example illustrated how to properly account for phase shifts in complex load
impedance. A similar approach would be necessary if the source impedance was complex.
We now turn from the purely reactive 3-element matching networks to a discussion of
purely resistive networks.

6.4.4 Resistive Three-element Matching Networks

The purely resistive matching networks are also known as attenuators, attenuating pads,
or just pads. Here we assume YA , YB , YC (or ZA , ZB , and ZC ) are real (resistances). The
resistive matching network can be employed to provide a broadband match between two
resistive terminations. Of course, some power loss must be accepted. This type of network is
used to build attenuators with specified attenuation. In addition, as we will see, the resistive
network can be used to ensure that a particular device sees a well-defined impedance, even
when the input of the following device has a variable or unknown impedance.
In this application YA , YB , YC are real (resistive). The terminating impedances are also
assumed to be resistive. There will be no phase shift and hence

= (real) nepers (6.84)

As noted previously
PIN
2 = ln (6.85)
POU T

Thus, determines the attenuation of the network. It is useful to relate to the attenuation
in dB:

PIN
Attenuation in dB = 10 log (6.86)
POU T

= 20 log e

= 8.686

If a network that provides 10 dB of attenuation is required, would be 10/8.686 = 1.1513

nepers.
6.4. THREE-ELEMENT MATCHING NETWORKS 207

6.4.4.1 Example - Design a 10 dB Pi-type resistive attenuator

Design a 10 dB Pi-type attenuator for use in a system with 75 impedance (R1 = R2 =
75 ) as in Figure 6.35.

= 10/8.686 = 1.1513

Y1 = Y2 = 1
75 = 13.33 mS
p
Y1 Y2 13.33 mS
YC = sinh = 1.423 = 9.370 mS

13.33 mS
YA = YB = tanh YC = 6.922 mS

ZA = ZB = 144.5 ' 145

ZC = 106.7 ' 107

107

75 145 145 75

Figure 6.35: 10 dB Pi-type attenuator with 75 impedance

The resistive matching networks/attenuators have the useful characteristic of isolating

the impedance found at the input from the impedance that terminates the output. This can
be seen by computing the input impedance for the extreme cases of shorted and open-circuit
output terminations in the example considered above:

shorted output : ZIN = 61.6

open output : ZIN = 92.0

Clearly, for any resistive termination, the input impedance will be in the range 61.6 to 92.0
. This property can be used to advantage when the input impedance of a particular stage
is not well known or is subject to variation. If this stage follows a stage that requires a
certain load impedance in order to operate correctly, an attenuator can sometimes be an
eective isolation stage. For example, for proper operation of certain types of frequency
mixers, it is necessary that all of the frequency components at the output of the mixer see
a well-defined impedance (usually 50 ). Typically the frequencies at the output of a mixer
may span a very broad range, while the following stage is often a narrow-band IF amplifier.
The IF amplifier would often have a well defined input impedance only within its passband.
A resistive matching network is sometimes employed between the mixer and the IF amplifier
in order to ensure proper termination of the mixer output port.
It is important to note that the attenuation of a resistive matching network will only
be equal to the design value if the network is operated between the impedances for which
208 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

it was designed. It is important to remember this, since the resistive networks are often
found in applications where the terminating impedances are substantially dierent from the
correct values. The example cited in the previous paragraph is one such case.

6.4.4.2 Minimum-loss Resistive Matching Networks

In some cases, especially when a broadband impedance match is required, one would like to
design a resistive network that has the smallest possible attenuation (). The problem is
then to find the solution that yields the smallest subject to the constraint that all resistive
elements have values greater than or equal to zero. This constraint ensures that the solution
corresponds to a passive network.
Consider the T-network in Figure 6.36.

RA RB

R1 RC R2

Figure 6.36: T-network

p
R1 R2
RC = (6.87)
sinh
p
R2 R1 R2
RB = (6.88)
tanh sinh
p
R1 R1 R2
RA = (6.89)
tanh sinh
The goal is to find the smallest that gives values for RA , RB , RC which are all 0.
Note from Equation 6.87 that p
R1 R2
sinh = (6.90)
RC
Since sinh increases monotonically as a function of , as shown in Figure 6.37, minimizing
is equivalent to minimizing sinh . Clearly then, we should use the largest value of RC

Figure 6.37: sinh as a function of

6.4. THREE-ELEMENT MATCHING NETWORKS 209

for which RB 0 and RA 0 (see Equation 6.90). Using

p
cosh = 1 + sinh2 (6.91)

p
= 1 + R1 R2 /RC 2 (6.92)

we find from Equation 6.88 and Equation 6.89

s
RC 2
RB = R2 +1 RC (6.93)
R 1 R2

s
RC 2
RA = R1 +1 RC (6.94)
R 1 R2

For the moment, assume R2 > R1 . Then, as RC is increased in order to decrease sinh(),
RA will decrease and will become equal to zero when (see Equation 6.94)
R1
RC = p (6.95)
1 R1 /R2
Be aware that RB will always be greater than zero if R2 > R1 . So, if R2 > R1 , the minimum
attenuation is obtained when RC is given by Equation 6.95. The corresponding value for
(denoted by min ) is obtained from Equation 6.92 and is
p
min = cosh 1 R2 /R1 (6.96)

The value for RB is obtained by using Equation 6.95 in Equation 6.93 to yield
p
RB = R2 1 R1 /R2 (6.97)

Notice that since RA = 0, the minimum-loss resistive network reduces to an L-network with
the series arm (RB ) connected to the larger of the two terminating resistances.
The preceding discussion was based on the T-network, although the final result was
found to be an L-network. The same result would have been obtained if the starting point
had been the Pi-network.

6.4.4.3 Summary of Resistive Minimum-loss Network

The resistive minimum-loss network reduces to an L-network with series arm connected to
the larger of the terminating resistances as in Figure 6.38.
p
Loss in dB = 8.686 cosh 1 Rbig /Rsmall

p Rsmall
Rshunt = (6.98)
1 Rsmall /Rbig

p
Rseries = Rbig 1 Rsmall /Rbig
210 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

Rseries

Rbig Rshunt Rsmall

Figure 6.38: Resistive minimum-loss network

6.5 References
1. Smith, Jack, Modern Communication Circuits, McGraw-Hill, 1986.

2. Terman, Frederick Emmons, Radio Engineers Handbook, McGraw Hill, 1943.

6.6. HOMEWORK PROBLEMS 211

6.6 Homework Problems

1. You are given a black box with two output terminals (a 1-port"). You play around
with the box for awhile and make the following observations:

(a) The output voltage from the box is sinusoidal.

(b) The peak magnitude of the open circuit voltage at the output of the box is
found to be 5 V.
(c) You connect a 50 resistor across the terminals and find that the peak magnitude
of the voltage across the resistor is 2.795 V.
(d) You short the output of the box and find the peak magnitude of the short
circuit current is 100 mA.

Find the power available from the source. Express your result in dBm.

2. Consider a source with impedance ZS = RS + jXS and suppose that the source drives
a load that is purely resistive. Denote the load resistance by RL .

(a) What load resistance should be used if the goal is to maximize the power delivered
to the load?
(b) Find an expression for the power delivered to the load resistance found in part 2a.
Express your result in terms of the source parameters only, i.e., Pavs , RS and
XS .

3. Find the lossless network having the minimum number of elements to match a source
impedance of 20-j60 to a load impedance having an equivalent circuit of 100 of
resistance shunted by 50 of capacitive reactance. Draw the network and label all
parts.

4. Consider the design of a lossless L-network to match the source and load shown in Fig-
ure 6.39. All resistances and reactances are in ohms, and the current source magnitude
is the peak value.

850 j100 j75

Lossless
1 mA Matching 50
Network

Figure 6.39: Complex source and load to be matched with a lossless L-network.

(a) Find the power available from the source. Express your result in dBm.
(b) How much power would be delivered to the load if a lossless matching network
was not used, i.e., if the load is connected directly to the source? Express your
result in dBm.
212 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

(c) There are four possible solutions for the matching network if an L-network is
used. Find all four. Sketch all solutions and indicate whether the elements are
inductors or capacitors.
(d) Verify two of your designs by plotting the path from the load to the source on a
Smith Chart.

5. Consider a source with impedance ZS = RS + jXS and a load that is purely resistive.
Denote the load resistance by RL . Suppose that RS < RL . Under what condition(s)
are there more than 2 lossless L-networks that conjugately match the source to the
load? Derive a simple inequality that can be checked to determine whether there are
2 or 4 solutions.

6. The source and load shown in Figure 6.40 can be matched with an L-network that
consists of two capacitors. All resistances and reactances are given in ohms.

jX1 j40

1 mA 75 j100 jX2 20

Figure 6.40: Source and load matched with 2-capacitor L-network.

Find X1 and X2 that will cause all of the available source power to be delivered to
the load. Both X1 and X2 must be <0.

7. Design a matching network to match the source and load shown in Figure 6.41. Use a
T-network such that Iout lags Iin by 60 at ! = 107 . What is the voltage phase shift?

50 j50 Iin Iout

+ +
Vin Vout 200

j100

Figure 6.41: Source and load for lossless T-network.

8. The circuit in Figure 6.42 is matched at ! = 108 rad s 1 . Redesign for a match with
the second harmonic attenuated in the shunt arm connected to the 100 resistor and
the third harmonic attenuated in the shunt arm connected to the 400 resistor.

9. Consider the design of a band-pass T-type matching network with specified Q. This
can be approached by thinking of the T-net as two back-to-back L-networks. One of
the L-networks has a high-pass topology and the other has low-pass topology.
6.6. HOMEWORK PROBLEMS 213

100 1 H

400

273.2 pF 143.3 pF

Figure 6.42: Circuit matched at ! = 108 s 1

.

Find a band-pass T-network that will match a 50 load to a 10 source impedance.

Find the solution with a series inductor connected to the 10 termination and a
capacitor connected to the 50 termination. Choose the overall Q of the network
to be approximately equal to 10, i.e., choose the virtual resistance, Rv , such that the
L-net with the highest Q has Q=10. Draw the T-network and label all components.
Note: You should combine the two shunt elements into one single element for this
problem.

10. Consider in Figure 6.43 the design of a T-network that transforms a 50 load
impedance to 5 :

jX1 jX2
5 jX3 50

Figure 6.43: T-network transforming 50 load impedance to 5 .

Design a T-network with a Q of 6. Find the solution with X1 > 0 and X2 < 0. Specify
X1 , X2 and X3 to the nearest ohm.

11. Consider the design of a T-network that transforms a 50 load impedance to 10 .

Suppose that the reactance of the shunt element of the T-network is fixed at +j200
as shown in Figure 6.44:

jX1 jX2
10 50
j200

Figure 6.44: T-network transforming 50 load impedance to 10 .

Find two sets of values (X1 , X2 ) which will give a match.

214 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

12. Design a T-network (as shown in Figure 6.45) that does an impedance inversion, i.e.,
that gives the following mapping between ZL and ZIN where R0 is a positive real
constant:
R2
ZIN = 0 (6.99)
ZL

jX1 jX2
Zin jX3 ZL

Figure 6.45: T-network providing impedance inversion.

Find expressions for X1 , X2 and X3 . You may assume that X1 > 0.

13. Consider the design of the matching network in Figure 6.46. Find the reactance Xs
and the turns ratio of the ideal transformer, n. The ideal transformer is characterized
by the relations:

V2 = nV1 and I2 = I1 /n (6.100)

I1 I2
50 1:n
jXs
+ +
V1 V2 1000

Ideal Transformer j2000

Figure 6.46: Matching network to find Xs and N

14. Suppose that you build an L-type matching network using a variable inductor and
capacitor as shown in Figure 6.47. This network is to be used to match the impedance
of a load, ZL , to the output of a generator that has a 50 impedance. At some
frequency you know that the inductors reactance can be adjusted to any value in the
range from +j10 to +j100 . At the same frequency the capacitors reactance can
be adjusted to any value in the range from -j10 to -j100 . On a Smith Chart show
the region corresponding to all possible values of load impedance, ZL , that can be
matched to 50 using this network. Normalize your Smith Chart to 50 . Shade or
color the region of the chart that corresponds to the matchable load impedances.
15. Consider the network shown in Figure 6.48. The resistor, capacitor, and inductor are
variable with 25 R 50 , 25 XL 50 , and 10 XC 40 . On a
6.6. HOMEWORK PROBLEMS 215

50 L

C ZL

Figure 6.47: L-network with variable inductor and capacitor.

Smith Chart show the region corresponding to all possible values of input admittance,
Yin . Normalize your Smith Chart to 50 .

Yin R

jXL jXC

Figure 6.48: Network with variable resistor, capacitor and inductor.

16. Design a resistive T-type attenuator that results in 10 dB attenuation when used with
50 source and load impedances.

(a) Consider all possible resistive loads and find upper and lower bounds on the input
resistance of the attenuator.
(b) Now design a resistive Pi-type attenuator for the same parameters as the T-type
attenuator.
(c) Repeat part 16a for the Pi-type attenuator.
(d) Suppose the Pi-type attenuator is used with a 600 load. Find the attenuation
in dB where attenuation is defined to be 10 log (Pin /Pout ).

17. An 8 dB attenuator that is designed to match a 75 source to a 300 load is used

between a 75 source and a 10 load. The source has available power of 10 dBm.
Find the power delivered to the 10 load. Express your result in dBm.
18. Consider a resistive attenuator/matching-network that is designed to work with 50
source and load resistances, as in Figure 6.49. When connected to 50 terminations,
the attenuator is designed to give an attenuation of 10 dB, i.e., Pin /Pout = 10, and also
to provide a 50 match at both input and output. Now, suppose that the attenuator
is used with an arbitrary load resistance, RL (the source impedance is still 50 ).
Denote the power available from the 50 source by Pavs and find an expression for
the power delivered to the load resistor, RL . Your result should be in terms of Pavs
and RL only (and numerical constants).
19. Design a minimum-loss resistive matching network that will match a 300 source to
a 50 load. What is the attenuation in dB?
216 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

50

M.N./ Pout
50 Pin = 0.1
Atten.

Pin Pout

Figure 6.49: Resistive attenuator/matching-network.

20. Suppose that you have a lossless matching network of unknown type that transforms 50
(resistive) to 450 (resistive) and you need to couple a 50 source to a resistive load,
RL , which may take on any positive, real, value. You have the choice of connecting
the 50 source directly to RL , or using the matching network between the source and
RL (with the 50 side of the matching network connected to the source). For what
values of RL would more power be delivered to RL with the matching network in the
system?

21. Design a lossless L-network that will match a 5 source to a 200 load. Use an
L-network with high-pass topology.

(a) Draw the network and indicate the reactances of the series and shunt elements.
(b) Now, suppose that you build the network that you designed using a lossy inductor
that has QL =32. Because the network contains a lossy element, the power that
is actually delivered to the 200 resistor will be smaller than the power available
from the source. Find the loss of the network. The loss is defined as the dierence
(in decibels) between Pavs and the power that is actually delivered to the 200
load.

22. Design the lossless L-network having a lowpass topology and with the series arm
connected to the source, as shown in Figure 6.50.

Figure 6.50: Lossless L-network with lowpass topology

(a) Specify XC and XL .

(b) Suppose the matching network you just designed is used and the imaginary part of
the load impedance increases by 10%, i.e., suppose the load impedance changes to
6.6. HOMEWORK PROBLEMS 217

10 + j 330 . Determine the change in the power delivered to the load. Express
your result in dB and be sure to correctly indicate the sign of the change.

23. A 50 source has Pavs = 4 mW. It is necessary to couple the source to a load with
impedance ZL = 2 + j20 .

(a) Find the power delivered to the load when the load is connected directly to the
source. Express your answer in dBm.
(b) There are 4 lossless LC L-networks that will match this source and load. Find
the solution that has a capacitive series arm and the shunt arm connected across
the load. You will receive full credit only if you specify the L-network having the
specified topology.

24. Consider the source and load shown in the Figure.

j8

1A 4

50 j50

Figure 6.51: Source and load for problem 24.

(a) Find the power available from the source. Express your answer in dBm.
(b) Match this source and load using a lossless L-network. Any valid solution will be
accepted.
(c) Now suppose that you can only use a single lossless inductor or capacitor, either in
series or in shunt, to couple the source to the load. If the goal is to maximize the
power delivered to the load under this constraint, find the best possible solution.

25. A source is coupled to a resistive load, RL = 1000 , through an ideal transformer.

The specified source current is the peak value. The ideal transformer is described by
the relations: V2 = N V1 and I2 = I1 /N .

I1 I2
1:N
+ +
2 mA V1 V2 1000

- -
100 j50

(a) Find the power available from the source. Express your result in dBm.
(b) Assuming that the turns ratio, N , is a positive real number find the value of N
that maximizes the power delivered to the load.
218 CHAPTER 6. IMPEDANCE MATCHING NETWORKS

(c) Assuming that N is set equal to the optimum value found in part (b.), determine
the power delivered to the load. Express your result in dBm.
(d) Suppose the transformer is removed from the system, i.e. suppose that the 1000
load is connected directly to the source. Find the power that would be delivered
to the load. Express your result in dBm.

26. Design the lossless L-network with a low-pass topology having the series arm connected
to the source, as shown below.

(a) Specify XL and XC .

(b) Determine the power that would be delivered to the (20 + j100) load if the
source was connected directly to the load. Express your result in dBm.
(c) Suppose the L-network designed in part a. is replaced with a new network consist-
ing of a single capacitor. Sketch the single-capacitor network that will maximize
the power delivered to the load. Specify the reactance of the capacitor and the
power that would be delivered to the load (in dBm).

27. Consider the problem of coupling a 200 source with Pavs = 0 dBm to the load
ZL = 10 + j100 .

(a) Design a passive, lossless L-network that matches the source to the load and
consists of 2 capacitors (no inductors may be used). Find the solution with the
shunt element of the L-network connected across the load. Sketch the system,
including the source, the matching network, and the load, and label each capacitor
with its impedance.
(b) An ideal, lossless transformer with turns ratio N:1 is used between the given
source and load. The turns ratio is a positive real number, and N>1. Find
N that results in the largest possible power delivered to the load. Include a
sketch showing the source, the transformer, and the load. Be sure to indicate the
correct orientation of the transformer (i.e. specify which side of the transformer
corresponds to the larger number of turns N).
(c) Find the power delivered to the load if the transformer specified in part b. is
used between the source and load. Express your result in dBm.

28. An ideal transformer can be used to reduce the mismatch loss (increase the mismatch
factor) between an arbitrary source impedance ZS = RS + jXS (RS > 0) and a
pure resistance RL as shown in the Figure. Suppose that the turns ratio, N , of the
transformer can be adjusted to any positive real number (N > 0).
6.6. HOMEWORK PROBLEMS 219

RS jXS N :1

Vs RL

IdealTransformer

Figure 6.52: Ideal transformer matching complex source to real load.

(a) For given ZS and RL , there is an optimum turns ratio, Nopt , that maximizes the
power delivered to the load. Find an expression for Nopt .
(b) Assuming that the optimum turns ratio is used, the resulting mismatch factor
(M F = PPavs
out
) depends only on the phase angle of the source impedance, i.e. if
ZS = |ZS |e , the mismatch factor can be written in terms of . Find such an
j

expression for the mismatch factor.

220 CHAPTER 6. IMPEDANCE MATCHING NETWORKS
Chapter 7

Introduction to 2-port Parameters

7.1 Introduction
An n-port network has 2n terminals that are grouped in pairs to form n ports. The ith port
has a voltage, Vi , between its terminals and a current Ii flowing in to one of the terminals
and out of the other terminal. In this chapter we will be concerned with 2-port networks
under sinusoidal steady-state excitation. Figure 7.1 summarizes the voltage and current
conventions that will be used. Voltages and currents are assumed to be phasor quantities.

I1 I2

+ +
V1 2-port V2
- -

I10 I20

Figure 7.1: Voltage and current conventions

Notice that a port has the property that the value of the current flowing in to the port
through the upper terminal is always equal to the current flowing out of that port through
the lower terminal, i.e. refer to Figure 7.1 and note that this property means that I1 = I10
and I2 = I20 . If two terminals are chosen whose currents do not satisfy this property, then
they cannot be said to comprise a single port - instead, they must be considered to be
terminals associated with two dierent ports.
The discussion in this chapter will concentrate on characterizing the properties of linear,
time-invariant (LTI) 2-ports. For our purposes, any 2-port comprised of linear network
elements (i.e. including linear resistors, linear inductors, linear capacitors, linear transform-
ers, linear controlled sources, linear transmission lines, and independent sources) is a linear
network. Time-invariance simply means that the parameters of the network elements do
not change with time.
Only four complex numbers are needed to completely characterize the properties of a
LTI 2-port network at one frequency. The four complex parameters can be interpreted

221
222 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

in terms of input and output impedances (or admittances) and forward and reverse gains
when the 2-port is terminated with specific source and load impedances. The standard,
or reference, source and load impedances are constants for a given parameter set. The
properties of the 2-port when it is terminated with the reference impedances can be used to
predict the properties of the 2-port in a system with arbitrary terminating impedances. In
particular, given the four complex numbers that comprise a specific 2-port parameter set,
it is possible to determine, for any combination of source and load terminations, the input
and output impedances, voltage and current gains, and power gain. It is also possible to
determine whether or not t he 2-port can oscillate with passive source and load terminations
and, if so, the region of the impedance plane (or Smith Chart) containing the source and
load terminations that support oscillation can be identified.
There are various ways to define the four complex numbers that describe a 2-port. The
dierent possibilities arise from considering two of the four variables (V1 , I1 , V2 , I2 ) to be
dependent on the other two (independent) variables. There are six possible choices for the
pair of dependent variables, leading to six possible parameter sets defined in terms of voltage
and current variables. The four most commonly used parameter sets are the Y, Z, h, and
ABCD parameters. In the next Chapter we will consider two more parameter sets (S and
T parameters) that are defined using variables that are linear combinations of the voltage
and current at a particular port. In any case, once one set of parameters is known it can be
translated into any of the other parameter sets without making additional measurements.

7.1.1 Y-parameters
Consider the Y-parameters, or admittance parameters:
I1 = Y11 V1 + Y12 V2
(7.1)
I2 = Y21 V1 + Y22 V2
The voltages are the independent variables and the currents are dependent variables. Mea-
surement of the Y-parameters involves forcing one of the independent variables to be zero,
i.e., if V2 is forced to zero by placing a zero impedance load across the output terminals,
then:
I1
Y11 = Input admittance with output shorted
V1 V2 =0

I2
Y21 = Forward transfer admittance, output shorted (7.2)
V1 V2 =0

Likewise, if V1 is forced to zero by shorting the input, Y22 and Y12 can be measured using:
I2
Y22 = Output admittance with input shorted
V2 V1 =0

I1
Y12 = Reverse transfer admittance, input shorted (7.3)
V2 V1 =0

All of the Y-parameters have units of admittance. Y-parameters are determined experi-
mentally by measuring input and transfer admittances with one port driven and the other
port terminated in a short circuit. Therefore the standard, or reference, termination for the
Y-parameter set is a short circuit.
7.1. INTRODUCTION 223

7.1.2 Z-parameters
Z-parameters or impedance parameters were used in chapter 5 to characterize the proper-
ties of transformers. Z-parameters are defined such that the currents are the independent
variables and the voltages are dependent, i.e.,

V1 = Z11 I1 + Z12 I2
(7.4)
V2 = Z21 I1 + Z22 I2

The Z-parameters can be interpreted as follows:

V1
Z11 = Input impedance, output open circuited
I1 I2 =0

V2
Z21 = Forward transfer impedance, output open circuited
I1 I2 =0
(7.5)
V2
Z22 = Output impedance, input open circuited
I2 I1 =0

V1
Z12 = Reverse transfer impedance, input open circuited
I2 I1 =0

Z-parameters have units of impedance, and measurement of Z-parameters involves deter-

mining input, output, and transfer impedances with one of the ports driven and the other
open-circuited. Thus, the reference impedance for the Z-parameters is an open circuit.

7.1.3 Hybrid (h) parameters

V1 = h11 I1 + h12 V2
(7.6)
I2 = h21 I1 + h22 V2
Hybrid parameters or h-parameters use both a short circuit and an open circuit as
reference terminations. The reference termination is an open circuit for the input port and
a short circuit for the output port. This is easily recognized by considering h11 and h22 :
V1
h11 = Input impedance with output short circuited (7.7)
I1 V2 =0

I2
h22 = Output admittance with input open circuited (7.8)
V2 I1 =0

Note that the parameters h11 and h22 have dierent units (impedance and admittance,
respectively). This is the origin of the hybrid designation.
A common characteristic of the Y-, Z-, and h-parameter sets is that the independent
variables are forced to zero when the 2-port is terminated in a standard, or reference,
impedance. For example, the independent variables in the Y-parameter representation are
the voltages at the input and output ports. To force one of the independent parameters to
zero, it is necessary to terminate one of the ports with a short circuit. The independent
variables in the Z-parameter representation are the input and output currents. Since input
224 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

or output currents are forced to zero when a port is terminated with an open circuit, the
reference impedance for the Z-parameter set is said to be an open circuit. The reference
impedance for the h-parameter set is an open circuit for the input port and a short circuit
for the output port. Thus the choice of independent variables determines the reference
impedance for the parameter set.

7.1.4 ABCD-parameters

ABCD-parameters are defined such that the port 1 variables depend on the port 2 variables:

V1 = AV2 BI2
(7.9)
I1 = CV2 DI2

The ABCD-parameters can be interpreted or calculated as follows:

1
A = V2
V1 Inverse open circuit forward voltage gain.
I2 =0

1
B = I2
V1 Neg. inverse short circuit forward transadmittance.
V2 =0
(7.10)

C = V2
I1
1
Inverse open circuit foward transimpedance.
I2 =0

1
D = I2
I1 Neg. inverse short circuit forward current gain.
V2 =0

The ABCD parameter representation diers from the Y-, Z-, and h-parameter sets in
that the independent variables are both associated with the output port. Thus, it is not
possible to define a fixed reference impedance termination for the input and output ports
that will force the independent variables to zero. Inspection of equations 7.10 shows that
the output port is terminated with an open for calculation of A and C and with a short
circuit for calculation of B and D. In any case, this parameter set is very useful because
the ABCD matrix for a cascade of 2-ports is the matrix product of the individual ABCD
matrices, as shown in the next section.
 7.2. SPECIAL TYPES OF 2-PORTS AND THEIR MATRIX PROPERTIES 225

7.1.5 2-port parameters for some common 2-port networks

2-port [Z] [Y] [ABCD]
Z
1 1 1 Z
NA 1
Z 1 1 0 1

1 1 1 0
Z Z NA 1
1 1 Z 1

n:1

n 0
NA NA 1
0 n
IdealTransformer
unit=1in,cornersize=absolute,dimen=middle(0,-0.6375)(0.9,0.03)linewidth=0.8ptlinewidth=0.8ptnoCurrentP

arrowsize=1.1pt 4,arrowlength=1.64,arrowinset=0linewidth=0.4pt(0.03,0)0.03
1 1 1 1 (0.06,0)(0.81,0)
(0.84,0)0.03 (0.06,-0.45)0.03 (0.09,-0.45)(0.84,-0.45)
tanh l sinh l 1 (0.87,-0.45)0.03
tanh l sinh l cosh l Zo sinh l
(0.09,-0.525)(0.84,-
Zo 1 1 1 1 1
Zo sinh l cosh l
0.525) [arrowsize=0.05in 0,arrowlength=2,arrowinset=0]-(0.19,-0.525)(0.09,-0.525)
sinh l tanh l sinh l tanh l Zo [ar-
rowsize=0.05in 0,arrowlength=2,arrowinset=0]-(0.74,-0.525)(0.84,-0.525) (0.405,-0.6375)l
(0.405,-0.225)Zo ,
+ 1 1
Z1 0 Z1 0 1 Z2 1
V Z1 gm V Z2 1 gm 1 1
g m Z1 Z2 Z2 gm Z2 Z1 Z2 Z1

7.2 Special types of 2-ports and their matrix properties

7.2.1 Reciprocal 2-ports
A 2-port network that does not contain any sources and is constructed only from any
combination of resistors, inductors, capacitors, transformers and transmission lines will obey
the principle of reciprocity, which states that the current (voltage) response at one port due
to a voltage (current) excitation at another port is independent of which port is excited and
which port the response is measured at. Figure 7.2 illustrates this principle for the case

Reciprocal + + Reciprocal
I1 V2 V2 I1
2-port 2-port
(a) (b)

Figure 7.2: If the 2-port is reciprocal, voltage response at port 2 due to current I1 at port
1 is the same as voltage response at port 1 due current I1 at port 2.

of the voltage response due to a current excitation. Since the voltage response at port i
due to a current excitation at port j is Yij , reciprocal 2-ports satisfy Y12 = Y21 . Note that
226 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

Y11 is not necessarily equal to Y22 in a reciprocal network. If the network is reciprocal and
symmetric, then Y12 = Y21 and Y11 = Y22 .
Figure 7.3 illustrates the reciprocity principle for the case of the current response due to

+ +
Reciprocal Reciprocal
V1 I2 I2 V1
2-port 2-port
(a) (b)

Figure 7.3: If the 2-port is reciprocal, current response at port 2 due to voltage V1 at port
1 is the same as current response at port 1 due to voltage V1 at port 2.

a voltage excitation. A reciprocal 2-port will have Z12 = Z21 . A 2-port that is reciprocal
and symmetric will have Z12 = Z21 and Z11 = Z22 .
It is left as an exercise to show that the ABCD matrix of a reciprocal network will satisfy
AD BC = 1. In other words, the determinant of the ABCD matrix is unity for a reciprocal
2-port.
The lossless and lossy matching networks discussed in Chapter 6 are all examples of
reciprocal networks. Since two of the network parameters are the same, a reciprocal network
can be fully characterized (at each frequency) by 3 complex network parameters.

7.2.2 Reciprocal lossless 2-ports

A lossless 2-port contains no dissipative elements, and hence the time-averaged real power
absorbed by the network is zero. A 2-port network constructed only of lossless transmission
lines, transformers, and lossless Ls and Cs will be reciprocal and lossless. The constraint
that the power absorbed by the network is zero can be written as
1 1
<(V1 I1 ) + <(V2 I2 ) = 0, (7.11)
2 2
where <( ) takes the real part of its argument. Using the Z parameters, equation 7.11 can
be written as
<(Z11 |I1 |2 + Z12 I1 I2 + Z21 I1 I2 + Z22 |I22 |2 ) = 0.
If the network is reciprocal, then Z12 = Z21 so the constraint is
R11 |I1 |2 + 2R21 <(I1 I2 ) + R22 |I2 |2 = 0, (7.12)
where the Rij = <(Zij ). Since equation 7.12 must be satisfied for any set of excitations,
(I1 , I2 ), we conclude that Rij = 0 for a lossless, reciprocal 2-port. In other words, the Zij are
purely imaginary. Similar reasoning can be used to show that the Yij are purely imaginary
as well.

7.3 Parallel, series, cascade connections of 2-port net-

works
Any 2-port network can be viewed as being constructed from simpler 2-port networks whose
ports are connected in various ways. Analysis of complex 2-port networks can be greatly
7.3. PARALLEL, SERIES, CASCADE CONNECTIONS OF 2-PORT NETWORKS 227

simplified if the network is decomposed into a number of simpler interconnected 2-port

networks whose 2-port parameters are easily found. In the following sections, the parallel,
series, and cascade connections are considered. Other possibilities, such as parallel-series
(input ports connected in parallel and output ports in series) and series-parallel are left as
exercises.
In all of the following discussions related to interconnection of 2-ports, it is assumed that
the interconnection does not upset the basic property that says that the current flowing into
one terminal associated with any port must flow out of the other terminal, i.e. referring
back to Figure 7.1 we assume that the relationships I1 = I10 and I2 = I20 are valid for each
of the interconnected 2-ports.

7.3.1 2-ports connected in parallel

Figure 7.4 shows a 2-port created by interconnecting two 2-ports with their input ports and
output ports connected in parallel. The Y-parameters of the constituent 2-ports can be
combined to obtain the Y-parameters of their parallel combination, as shown below.

I1a I2a
I1 I2
+ Y11a Y12a +
V1a Y21a Y22a V2a
+ - - +

V1 I1b I2b V2

- -
+
Y11b Y12b +
V1b Y21b Y22b V2b
- -

Figure 7.4: Two 2-ports interconnected such that the input and output ports are in parallel.

The parallel connection results in the following relationships:

V1 = V1a = V1b , V2 = V2a = V2b (7.13)

I1 = I1a + I1b , I2 = I2a + I2b . (7.14)

In vector notation, e.g.

V1 I1 Y11 Y12
V= ,I= ,Y= ,
V2 I2 Y21 Y22
equations 7.13 and 7.14 can be written as:

V = Va = Vb (7.15)

I = Ia + Ib . (7.16)
228 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

The voltage and current vectors associated with the individual 2-ports must satisfy

Ia = Y a V a (7.17)

Ib = Y b V b . (7.18)
Equations 7.15, 7.17, and 7.18 can be used in equation 7.16 to show that

I = {Ya + Yb }V. (7.19)

Thus, the Y-parameter matrix for the parallel combination of 2-ports is the sum of the
constituent Y-parameter matrices.

7.3.2 2-ports connected in series

Figure 7.5 shows a 2-port created by interconnecting two 2-ports with their input ports
and output ports connected in series. The Z-parameters of the constituent 2-ports can be
combined to obtain the Z-parameters of their series combination, as shown in Figure 7.5.

I1 I1a I2a I2

a a
+ + Z11 Z12 + +
V1a a a
Z21 Z22 V2a
- -

V1 I1b I2b V2

+ b b +
Z11 Z12
V1b b
Z21 b
Z22 V2b
- - - -

Figure 7.5: Two 2-ports interconnected such that the input and output ports are in series.

The series connection results in the relationships:

I1 = I1a = I1b , I2 = I2a = I2b

V1 = V1a + V1b , V2 = V2a + V2b .

In vector notation the voltage and current vectors must satisfy

I = Ia = Ib (7.20)

V = Va + Vb . (7.21)
The voltage and current vectors for the individual 2-ports must satisfy:

V a = Z a Ia (7.22)
7.4. POWER GAIN DEFINITIONS 229

V b = Z b Ib (7.23)

Equations 7.20, 7.22, and 7.23 can be used in equation 7.22 to show:

V = {Za + Zb }I. (7.24)

Thus, the Z-parameter matrix for the series combination of 2-ports is the sum of the
constituent Z-parameter matrices.

7.3.3 Cascaded 2-ports

Figure 7.6 shows two 2-ports connected in cascade, i.e. the output of the first 2-port drives
the input of the second 2-port.

I1a I2a I1b I2b

+
Aa B a + +
Ab B b +
V1a C a Da V2a V1b V2b
- - - C b Db -

Figure 7.6: Two 2-ports in cascade.

The cascade connection results in the following relationships:

V2a = V1b , I2a = I1b (7.25)

V1 V2
Define the input vector IN = , output vector OUT = , and chain parameter
I1 I2
A B
matrix ABCD = . Then the relationships given in equation 7.25 can be written
C D
as:
OUTa = INb (7.26)

The following equations are satisfied by the individual 2-ports:

INa = ABCDa OUTa (7.27)

INb = ABCDb OUTb (7.28)

Using equation 7.28 in equation 7.26, and then using the result in equation 7.27 we find

INa = ABCDa ABCDb OUTb . (7.29)

Thus, the input and output vectors of the overall 2-port are related by the matrix product
of the constituent ABCD matrices.
230 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

ZS Pin Pout

2-port ZL

Pavs Zin Pavo Zout

Figure 7.7: A system consisting of a source, 2-port, and a load. The source and load
impedances, ZS and ZL , may be complex. Pavs is the power available from the source and
Pavo is the power available from the output of the 2-port. Pin is the power delivered to the
input of the 2-port and Pout is the power delivered to the load.

7.4 Power Gain Definitions

One very useful application for the 2-port parameters defined earlier is the calculation of
the input and output impedances and the power gains of a 2-port when terminated with
arbitrary source and load terminations, ZS and ZL . Consider a system consisting of a 2-port
driven by a source with impedance ZS and terminated with a load having impedance ZL as
shown in Figure 7.7.
In general, the input impedance of the 2-port will depend on the 2-ports network pa-
rameters and the load impedance, ZL . The output impedance of the 2-port will depend on
the network parameters and the source impedance, ZS .
The input impedance may not be equal to the conjugate of the source impedance, so
the power that is delivered to the input of the 2-port, Pin , may be only a fraction of the
power that is available from the source. Likewise, the output port can be considered to be a
source with impedance equal to Zout and with available power Pavo . In general, the output
impedance of the 2-port will not be equal to the conjugate of the load impedance, so the
power that is delivered to the load, Pout , will be some fraction of the power that is available
from the 2-port. The relationships between the various powers can be expressed in terms of
several dierent power gains, which are defined next.
1. Operating Power Gain (or just power gain)
Power delivered to load Pout
G = (7.30)
Power delivered to 2 port Pin
The operating power gain will depend only on the load impedance, ZL , and on the
2-port network parameters.
2. Transducer Power Gain
Power delivered to load Pout
GT = = (7.31)
Power available from source Pavs
The transducer gain will depend on both the source and load impedances, ZS and ZL ,
as well as the 2-port network parameters. GT is a measure of the advantage gained by
using the 2-port instead of a lossless matching network. Consider Figure 7.8, which
shows the system with the original 2-port replaced with a lossless matching network.
7.4. POWER GAIN DEFINITIONS 231

ZS Pin = Pavs Pout = Pavs

Lossless
Matching ZL
Network
Pavs Pavo
Zin = ZS Zout = ZL

Figure 7.8: Transducer power gain

Since the source sees ZS , it will deliver its maximum power, Pavs , to the matching
network, so Pin = Pavs . Because the matching network is lossless, power delivered to
the load will also be Pavs , so Pout = Pavs . This means that the transducer gain of a
lossless matching network is always equal to 1.
The transducer gain, GT , calculated for a particular set of source and load terminations
measures the performance of the 2-port in that system relative to that of a lossless
matching network. If GT > 1, then the power delivered to the load is greater than
Pavs , which is better than could ever be achieved by using a lossless matching network.
On the other hand, if GT < 1, more power would be delivered to the load if the 2-port
was replaced with a lossless matching network.
Note that GT is generally a more useful quantity than is G for system design calcu-
lations. To see why this is so, note that it is possible to have a situation where the
operating power gain of an amplifier in a particular system is G = 10 but the trans-
ducer gain in the same system is GT = .5. Here the power gain G looks good since
Pin = 10 obviously, this 2-port must be contain an active device but GT says
Pout

wed be better o (i.e. more power would be delivered to the load) if the 2-port was
replaced with a passive, lossless matching network which, by definition, has GT = 1.
To reconcile the dierence between the two gains, take a look at Equation 7.32, which
shows that the ratio GT /G is equal to the fraction of the Pavs that is delivered to the
input of the 2-port, i.e. the input mismatch factor:
Pout
GT Pavs Pin
= Pout
= (7.32)
G Pin
Pavs
.5
=
10

= 0.05

The problem with this amplifier is the input mismatch just 5 percent of the available
power from the source is actually delivered to the 2-port! The situation could be
improved by adding a lossless matching network between the source and the amplifier
which would cause all of the available power from the source to be delivered to the
input of the amplifier (Pin = Pavs ). The addition of the lossless matching network
would change the source impedance seen by the amplifier, but this will not change the
232 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

operating power gain because G does not depend on ZS . So G would still be equal to
10 and hence Pout = 10Pin = 10Pavs .
The operating power gain, G, only accounts for what happens to the power after it
gets in to the 2-port. It doesnt provide any information about what fraction of the
available power actually gets in.
3. Available Power Gain
Power available from the output of 2 port Pavo
GA = (7.33)
Power available from the source Pavs

GA will depend only on the source impedance and the 2-port network parameters. It
describes how much power is potentially available from the output of the 2-port.
It may be helpful to summarize how the dierent power gains are related. First, the trans-
ducer gain will always be less than or equal to the operating power gain since
Pout
GT Pavs Pin
= Pout
= 1 (7.34)
G Pin
Pavs

Thus, GT /G is equal to the impedance mismatch factor at the input port, which is equal
to one when Pin = Pavs , or when the source is conjugately matched to the input of the
2-port. The ratio given in equation 7.34 can be used to determine how much the power
delivered to the 2-port could be increased if a lossless matching network is added between
the source and the 2-port.
The transducer gain will always be less than or equal to the available gain since
Pout
GT Pavs Pout
= Pavo
= 1 (7.35)
GA Pavs
Pavo

So GT /GA is equal to the impedance mismatch factor at the output port, which is equal
to one when the load is conjugately matched to the output of the 2-port. The ratio given
in Equation 7.35 can be used to determine how much the power delivered to the load could
be increased if a lossless matching network was added between the 2-port and the load.
The operating gain can be greater than or less than the available gain since
Pavo Pin
GA Pavs Pavs
= Pout
= Pout
(7.36)
G Pin Pavo

Depending on whether the input mismatch factor (numerator) or the output mismatch
factor (denominator) is the smaller, the ratio GA /G can be either > 1 or < 1. If the input
is match is better than the output match, then the ratio in Equation 7.36 will be > 1. On
the other hand, if the output match is better than the input match, then the ratio will be
< 1.
Finally, it should be noted that only the operating gain and the available gains can be
cascaded. That is, when several amplifiers are connected in cascade, the overall operating
or available gain of the cascade is simply the product of the operating or available gains of
the individual amplifiers. This is not true for the transducer gain.
When cascading gains, it must be kept in mind that the source impedance that terminates
the input of the second amplifier is the output impedance of the first amplifier, etc. Consider
7.5. CALCULATION OF IMPEDANCE AND GAIN USING THE IMPEDANCE MATRIX233

the problem of cascading a number of identical amplifiers. Suppose the overall available
gain is the quantity of interest. The available gain of the first amplifier depends on the
source termination, ZS . The available gain of the second amplifier depends on the output
impedance of the first amplifier which, in turn, depends on the source termination. The
available gain of the third amplifier depends on the output impedance of the second amplifier,
etc. So the available gain of a cascade of amplifiers depends on the 2-port network parameters
of the individual amplifiers as well as the source termination of the first amplifier. Similarly,
if the operating power gain of the cascade is desired, it will be found to depend only on
the load termination of the last amplifier and the 2-port parameters of all of the individual
amplifiers.

7.5 Calculation of Impedance and Gain using the Impedance

Matrix
The concepts defined in the previous section will be applied in this section where the input
and output impedances and transducer gain are derived for a 2-port represented by its
Z-parameters.
Consider a system consisting of a 2-port embedded between a source with impedance ZS
and peak open-circuit voltage VS and a load with impedance ZL (Figure 7.9). The power
available from the source Pavs is defined to be the power that the source would deliver to a
conjugately matched load and is
|VS |2
Pavs = (7.37)
8RS
The voltage and current at input and output ports are denoted by (V1 , I1 ) and (V2 , I2 ),
respectively, with current chosen to flow into the two port as shown in Figure 7.9.

ZS I1 I2
+ +
VS V1 [Z] V2 ZL
- -

Figure 7.9: A system consisting of source, 2-port, and a load.

The voltage and currents at the input and output of the 2-port are related by impedance
matrix elements, Zij , i.e.
V1 = Z11 I1 + Z12 I2 (7.38)

V2 = Z21 I1 + Z22 I2 (7.39)

The power delivered to the load impedance is related to Pavs by the transducer power gain
GT
Pout 1
Re[V2 I ]
GT = 2 |V |2 2 (7.40)
Pavs S
8RS
234 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

The load constraint I2 = V2 /ZL can be employed in equation 7.40 to write

V 2 2 RS RL
GT = 4| | (7.41)
VS |ZL |2

where RS = Re[ZS ] and RL = Re[ZL ]. Equation 7.41 can be written in terms of the
magnitude of the voltage gain, |V2 /V1 |, and the voltage division that occurs at the input of
the 2-port, |V1 /VS |, i.e.:
V 2 V 1 R S RL
GT = 4| |2 | |2 (7.42)
V1 VS |ZL |2
The voltage division can be written in terms of the input impedance of the 2-port and the
source impedance
V1 ZIN
= (7.43)
VS ZIN + ZS
and the input and output impedances of the 2-port can be obtained from the load constraint
and equations 7.38 and 7.39. The input impedance is:

V1 Z12 Z21
ZIN = = Z11 (7.44)
I1 ZL + Z22
and the output impedance is:

V2 Z12 Z21
ZOU T = |V =0 = Z22 (7.45)
I2 S ZS + Z11

The voltage gain is obtained from 7.38 and 7.39 using the load constraint I2 = V2 /ZL :

V2 Z21 ZL
AV = = (7.46)
V1 Z11 ZL + Z11 Z22 Z12 Z21
Now, equation 7.42 can be rewritten using 7.46 and 7.44 as:
Z12 Z21
Z21 ZL Z11 ZL +Z22 RS RL
GT = 4| |2 | Z12 Z21
|2
Z11 ZL + Z11 Z22 Z12 Z21 Z11 ZL +Z22 + ZS
|ZL |2

which simplifies to the final result:

|Z21 |2 RL RS
GT = 4 (7.47)
|(Z11 + ZS )(Z22 + ZL ) Z12 Z21 |2

Since the available source power Pavs is a constant determined by the capabilities of the
source, the output power will be proportional to GT . The largest possible output power
with passive source and load terminations results when ZS and ZL are chosen to maximize
GT subject to Re[ZS ]>0 and Re[ZL ]>0. These optimum terminations are denoted by ZM S
and ZM L and it can be shown that these terminations result in a simultaneous conjugate
match at the input and output ports, i.e. ZM S and ZM L satisfy the following equations:

ZIN = ZM S
(7.48)

ZOU T = ZM L
7.6. APPLICATIONS OF 2-PORT ANALYSIS 235

7.6 Applications of 2-port analysis

7.6.1 Losses in L-networks for impedance matching
Consider an L-network designed to match two resistive terminations, RS and RL , with
RS < RL . The design of such networks has been considered previously. Suppose that such
a network has been designed and that the series and shunt elements available for use in the
network are lossy i.e., suppose that the series and shunt elements have component Qs at
the design frequency, !o , denoted by Qs and Qp , respectively. The loss in the series element
can be represented by a series resistance, rs , and the loss in the parallel element can be
represented by a shunt resistance, rp . The circuit model is shown below in Figure 7.10. The
series and shunt reactances are assumed to be those necessary to match the source to the
load in the absence of any losses in the network. Thus, the load impedance seen by the
resistive 2-port will be equal to ZL = RS jXS .

|XS | rp
Qs = rs
Qp = |XP |

RS
jXS
rs
rp jXP RL

Z = RS + jXS Z = RS jXS

Figure 7.10: Circuit model for an L-network with lossy components. The loss resistances
can be lumped into a 2-port which is contained within the dashed box.

As shown in the Figure, the loss resistances can be lumped into a 2-port which is con-
tained within the dashed box. The Z-parameter matrix for the 2-port is easily shown to
be: 2 3 2 3
Z11 Z12 rs + r p rp
[Z] = 4 5=4 5 (7.49)
Z21 Z22 rp rp
Notice that the 2-port defined here is reciprocal and hence Z12 = Z21 . The loss caused
by the presence of the resistances rs and rp can be determined by calculating the transducer
gain of the 2-port in the system shown in Figure 7.10. The source impedance seen by the
resistive 2-port is ZS = RS + jXS and the load impedance is ZL = RS jXS . If the
Z-parameters and the terminating impedances are inserted into equation 7.47, the result is:

4rp2 RS2
GT = (7.50)
|(rs + rp + RS + jXS )(rp + RS jXS ) rp2 |2

After some algebraic manipulation, the transducer gain can be rewritten as:
1
GT = RS rs rs
(7.51)
|1 + 2rp (1 + Q2 ) + 2RS + 2rp (1 jQ)|2
236 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

where the fact that |XS | = RS Q has been used to eliminate XS from the expression. Next,
one can make use of the following relationships to write GT in terms of the L-network Q
(Q) and the component Qs of the lossy series and shunt elements:
RS Q
2rp = 2Qp (1+Q2 )

rs Q
2RS = 2Qs (7.52)

rs Q2
2rp = 2Qs Qp (1+Q2 )

After using 7.52 in 7.51, GT can be written as:

1
GT = Q Q (1 jQ)Q2
(7.53)
|1 + 2
2Qp + 2Qs + 2Qs Qp (1+Q2 ) |

This result can be checked by allowing the losses to approach zero (Qs ! 1, Qp ! 1) in
which case GT ! 1, as expected.
In practical applications, the component Qs (Qs and Qp ) will be larger than the L-
network Q (Qs > Q and Qp > Q) in which case the last term in the denominator can be
neglected compared to the second and third terms. Thus, a good approximation for the
transducer gain is:
1
GT ' Q Q 2
(7.54)
(1 + 2Qp + 2Q s
)

This result illustrates the fact that component Qs must be much larger than the L-network
Q if component losses are to be reasonably small. For example, suppose that it is necessary
to keep the matching network loss below 1 dB this means that GT 10( 1/10) =
0.794. Assuming that the component Qs are equal (Qs = Qp ), then the constraint GT
0.794 leads to Qs,p 8.2Q, which means that the L-network must be implemented with
components having component Qs at least 8.2 times as large as the L-network Q. Obviously,
this will become impractical when the L-network Q is too large. Large L-network Qs are
associated with large resistance transformation ratios1 so that it becomes more dicult to
implement an L-network with low losses as the resistance transformation ratio increases.

7.6.2 Two-winding Transformers

7.6.2.1 Equivalent Circuit Model for Two-winding Transformers
Transformers are often utilized in both narrow-band and wide-band RF applications. They
can be employed for impedance transformation, phase inversion, and dc isolation. Trans-
formers are also utilized as resonant circuit elements.
Consider the two-winding transformer shown in Figure 7.11.
Ignoring losses, the equations that describe this device are

V1 = j!L1 I1 + j!M I2 (7.55)

V2 = j!M I1 + j!L2 I2 (7.56)
q
1 Recall Rbig
that the Q of an L-network is Q = Rsmall
1.
7.6. APPLICATIONS OF 2-PORT ANALYSIS 237

I1 I2
M
+ +
V1 L1 L2 V2
- -

Figure 7.11: 2-winding transformer.

where L1 and L2 are the self inductances of the transformer windings and M is the mutual
inductance. The dot convention is such that if current flows into the dotted terminals, the
magnetic fluxes linking the two coils will reinforce each other. With this convention, M will
be a positive number. The 2-winding transformer is completely described by its open-circuit
impedance matrix,
2 3
j!L1 j!M
[Z] = 4 5.
j!M j!L2

Since the circuit operation of the transformer is completely described by this impedance
matrix, any network having the same defining equations can be substituted for the trans-
former. One useful equivalent circuit is shown in Figure 7.12.

(1 k 2)L1
n:1

k 2 L1

Ideal Transformer

Figure 7.12: Equivalent circuit for a two-winding transformer

In this model the real transformer has been replaced with an ideal transformer and
equivalent inductances. The equivalent circuit parameters are k (coupling coecient), n
(turns ratio), and L1 (self inductance of winding 1. This equivalent circuit is not unique.
For example, it is possible to derive an equivalent circuit that has the self inductance of
winding 2 as a parameter. The terminal relations for the ideal transformer are summarized
by Figure 7.13 and equations 7.57 and 7.58.

IA n : 1 IB
+ +

VA VB
- -

Figure 7.13: Terminal relations for the ideal transformer

238 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

VA
VB = (7.57)
n
IB = nIA (7.58)
It is left as an exercise to demonstrate that the equivalent circuit shown in Figure 7.12 has the
same impedance matrix as the actual transformer, provided that the following relationships
hold:
M
k = p (7.59)
L1 L2
r
L1
n = k (7.60)
L2
The k parameter is called the coecient of coupling and n is the eective turns ratio for
the transformer.
A useful approximation results when the coecient of coupling approaches 1. This will
be the situation when both windings are wound on a high permeability magnetic core. Here
the equivalent circuit reduces to the circuit shown in Figure 7.14.
n:1

L1

Ideal Transformer

Figure 7.14: Equivalentpcircuit when k approaches 1. In this case, the eective turns ratio
is, approximately, n = L1 /L2 .

7.6.2.2 Impedance Transformation with the Two-winding Transformer

Consider the situation where a tightly-coupled transformer is to be used as an impedance
transformer as in Figure 7.15.
n:1

Zin L1 ZL

Ideal Transformer

Figure 7.15: Tightly-coupled transformer used as impedance transformer

The Z-parameter matrix for a two-winding transformer is

2 3 2 2 3
j!L1 j!M j!L1 j!L1 kn
[Z] = 4 5=4 5.
j!M j!L2 k2 k2
j!L1 n j!L1 n2
7.6. APPLICATIONS OF 2-PORT ANALYSIS 239

This is the exact impedance matrix. When the transformer is tightly coupled, k ' 1, and
the impedance matrix can be approximated by
2 3
j!L1 j!L1 /n
[Z] ' 4 5.
2
j!L1 /n j!L1 /n
The impedance seen looking in to a tightly coupled transformer that is terminated in load
impedance ZL is now easily computed:
Z12 Z21 (j!L1 /n)2 j!L1 ZL n2
ZIN = Z11 = j!L1 2
= (7.61)
Z22 + ZL j!L1 /n + ZL j!L1 + ZL n2
The tightly-coupled transformer will behave essentially like an ideal transformer if !L1
|ZL n2 | or, since n2 = L1 /L2 , if !L2 |ZL |. In this case

ZIN n2 ZL . (7.62)

1:n

Zin L1 ZL

Ideal Transformer

Figure 7.16: Transformer reversed

Similarly, if the transformer is reversed, as in Figure 7.16 then

1 s L1 Z L
ZIN = (7.63)
n2 s L 1 + Z L
And if
!L1 ZL (7.64)
then
1
ZIN ' ZL (7.65)
n2
These examples lead to a rule that is applicable when designing tightly-coupled impedance
transformers: The inductive reactance of the impedance transformer winding
should be significantly larger than the impedance to which it is connected. Typi-
cally, the inductive reactance is chosen to be at least four times larger than the impedance at
the lowest frequency of intended operation. When this constraint is satisfied, the tightly cou-
pled transformer can provide an impedance transformation that is essentially independent
of frequency over a wide bandwidth.

7.6.2.3 Single-tuned Transformer

In Section 7.6.2.2 the impedance transforming properties of the tightly-coupled transformer
were examined, and a working rule was given for designing a transformer which will be-
have essentially like an ideal transformer. This is most useful when designing transformers
240 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

that must operate over a relatively wide frequency range. If narrow-band operation is de-
sired, then the inductive reactance of the transformer windings can be resonated, and ideal
transformer operation will result at the resonant frequency. For example, suppose that a
capacitance is used to resonate the primary inductance of the transformer as in Figure 7.17.

Zin C RL

Figure 7.17: Actual circuit

n:1

Zin C L1 RL

Ideal Transformer

Figure 7.18: Single-tuned transformer using a tightly coupled transformer and a resonating
capacitance.

If the capacitor resonates with L1 at the frequency of interest, then at that frequency, the
circuit will behave like an ideal transformer, and the input impedance will be Zin = n2 RL .
More generally, the primary circuit can be modeled by reflecting the load resistance through
the transformer as shown in Figure7.19. The result is a parallel resonant circuit with

Zin C L1 n 2 RL

Figure 7.19: Load resistance reflected through transformer

n2 R L RL
Q= =
!o L1 !o L2
and
1
!o = p .
L1 C
The input impedance at the resonant frequency is

ZIN = n2 RL . (7.66)
7.6. APPLICATIONS OF 2-PORT ANALYSIS 241

The bandwidth of the single-tuned transformer circuit is the same as the bandwidth of
the equivalent parallel RLC circuit shown in Figure 7.19, i.e.,

!o ! 2 L1 1
BW = = 2o = 2 .
Q n RL n CRL
The bandwidth can be controlled by adjusting the self inductance of the primary winding,
L1 , and the turns ratio, n. For a given L1 , the resonating capacitance, C, must be chosen to
resonate with L1 at !o . This provides a useful method for designing impedance matching
networks with specified bandwidth.

7.6.3 Two Magnetically-Coupled Resonators (Doubly-Tuned Trans-

former)
The Z matrix for two coupled coils can be written in terms of the self-inductances, L1 and
L2 and mutual-inductance M :

j!L1 j!M
[Z] = (7.67)
j!M j!L2

In this section, well employ the Z matrix to investigate the properties of a filter consisting
of two magnetically coupled resonators. The system is shown in Figure 7.20.

R C r r C
M

L L R

k=M/L

Figure 7.20: Two magnetically-coupled series resonant circuits in a system with source and
load impedances equal to R.

We assume that the two coils are identical (L1 = L2 = L), and that resistive losses can
be modeled by a resistance, r, in series with each of the coils. In addition, a capacitance,
C, can be added in series with each coil to form two magnetically-coupled series resonant
circuits. It is easy to verify that adding the elements r and C in series with each coil changes
the diagonal elements of the Z matrix by adding the impedances of these elements to the
self-impedance of the coils and does not aect the o-diagonal (coupling) elements of the Z
matrix. Thus, the impedance matrix of the coupled resonator system is given by:
1
r + j!L + j!C j!M
[Z] = 1 (7.68)
j!M r + j!L + j!C

In a system with source and load impedance denoted by R, the transducer gain of the
coupled-resonator filter is obtained by using 7.68 and ZS = R, ZL = R in Equation 7.47:

4! 2 M 2 R2
GT = 1 (7.69)
|(R + r + j!L + j!C )2 + ! 2 M 2 |2
242 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

Define the coupling factor :

M
k= , (7.70)
L
the resonant frequency of either resonator in isolation:
1
!o = p , (7.71)
LC
and the loaded Q of the resonator (accounting for the loss resistance and termination resis-
tance):
!o L
Q0 = . (7.72)
R+r
Equations 7.70 - 7.72 can be used in Equation 7.69 to write the transducer gain as:
2
R2 4k 2 Q02 !
!o2
GT = (7.73)
2 !
(R + r) |(1 + jQ0 ( ! ! 2 2 02 ! 2 2
! )) + k Q ! 2 |
o
o o

This is a complicated-looking result, so to simplify interpretation it is useful to examine the

transducer gain at the resonant frequency !o :

R2 4k 2 Q02
GT |!=!o = (7.74)
(R + r) (1 + k 2 Q02 )2
2

In the absence of resonator losses (r ! 0) the first term is equal to one. The second term
will be equal to one also if
1
k= 0 (7.75)
Q
Thus (in the absence of resonator losses), if the coupling factor and/or loaded Q are adjusted
so that equation 7.75 is satisfied, we expect that the coupled resonator filter will transmit
signals unattenuated at the resonant frequency, !o . In this case, the resonators are said to
be critically coupled.
Figure 7.21 shows the transducer gain for a system employing resonators with loaded
Q equal to 5. Curves are plotted for 5 dierent coupling factors, two of which are smaller
than the critical coupling value of 1/5=0.2 and two of which are larger than the critical
value. Coupled resonators with k smaller than the critical coupling value are said to be
undercoupled, whereas resonators with k larger than the critical value are overcoupled.
A characteristic of the overcoupled response is a double-peaked response with separation
between the peaks that depends on the degree of overcoupling. Undercoupling, on the other
hand, yields a single-peaked response with progressively larger attenuation as the coupling
factor is decreased below the critical value. Notice that the critically coupled response
yields a bandpass filter with relatively flat passband response. An obvious disadvantage of
this type of filter is the fact that the optimum coupling factor depends on the source and
load impedances. Thus, a filter that is operated with dierent terminations than it was
designed for may end up either under- or over-coupled and the shape of the filter response
may degrade significantly under such conditions.
7.6. APPLICATIONS OF 2-PORT ANALYSIS 243

Coupled lossless resonators, loaded Q=5

1
k = 0.40 overcoupled
k = 0.30 overcoupled
0.8
k = 0.20 critical coupling
k = 0.15 undercoupled
0.6 k = 0.10 undercoupled
GT
0.4

0.2

0
0.6 0.8 1 1.2 1.4 1.6 1.8 2
!/!o

Figure 7.21: Transducer power gain for a coupled resonator filter using resonators with
loaded Q equal to 5. Critical coupling will occur when k = 1/Q0 = 1/5 = 0.2.

7.6.4 Analysis of a Small-signal Series/Shunt Feedback Amplifier

In this section we will calculate the input and output impedance and voltage gain of the
amplifier shown in Figure 7.22. The amplifier is redrawn in Figure 7.23 to show how the
circuit can be described as a combination of 2-ports connected in series, in cascade, and in
parallel. The hybrid- 2-port is in series with the 2-port containing the single shunt resistor
Re . The output of the resulting 2-port is cascaded with the input of a 2-port consisting
of an ideal transformer with 1:1 turns ratio. The ideal transformer is configured so that it
produces a 4:1 impedance transformation. The cascade is in parallel with a 2-port consisting
of the single series resistor Rf .
Analysis proceeds as follows:
calculate the Z parameters of the hybrid- 2-port and the 2-port containing the shunt
resistor, Re . Add these two Z matrices to obtain the Z matrix of the series combina-
tion.
Convert the Z matrix of the series combination to an ABCD matrix. Denote this ma-
trix by [ABCD]1 . Determine the ABCD matrix of the 4 : 1 transformer and denote
this matrix by [ABCD]2 . The ABCD matrix of the cascade is then [ABCD]1 [ABCD]2 .
Note that the order of the terms in this matrix product is important and must reflect
the order in which the 2-ports are cascaded.
Convert ABCD of the cascade to a Y matrix and sum this matrix with the Y matrix
for the 2-port consisting of the single series resistor, Rf .. The result is the Y matrix
of the overall circuit.
244 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

Calculate Yin , Yout , and Av in terms of the Y parameters of the overall circuit.

+ + +

- -

- -

Figure 7.22: Small signal model of amplifier with series and shunt feedback and 4:1 output
transformer.

Rf

+
Vbe r ro
- gm Vbe
RS

Re RL

Figure 7.23: Feedback amplifier drawn as a set of interconnected 2-ports.

The Z matrix for the hybrid- 2-port is:

2 3 2 3
Z11 Z12 r 0
4 5=4 5
Z21 Z22 g m r ro ro

The Z matrix for the shunt resistor is:

2 3 2 3
Z11 Z12 Re Re
4 5=4 5
Z21 Z22 Re Re
7.6. APPLICATIONS OF 2-PORT ANALYSIS 245

The Z matrix of the series combination of the hybrid- 2-port and the shunt resistor is:
2 3 2 3
Z11 Z12 r + R e Re
4 5=4 5
Z21 Z22 g m r ro + Re ro + Re
Convert the Z matrix to an ABCD matrix:
2 3 2 3
A B (Re + r ) (Re + r )(Re + ro ) Re (Re g m ro r )
4 5= 1 4 5
g m r o r + R e
C D 1 ro + Re
At this point, we will let ro ! 1 since the small signal model shown in Figure 7.22
does not contain this parameter. Why didnt we do this in the first place? Because the
Z matrix of the hybrid- model wouldnt have existed! After letting ro ! 1 the ABCD
matrix reduces to:
2 3 2 3
A B 0 R e + r + g m R e r
4 5= 1 4 5
g m r
C D 0 1
The ABCD matrix of the 4:1 transformer implemented using the ideal transformer with
1:1 turns ratio is: 2 3 2 3
A B 2 0
4 5=4 5
1
C D 0 2
The product of the two ABCD matrices is:
2 3 2 3
A B 0 R e + r + g m R e r
4 5= 1 4 5
2gm r
C D 0 1
Convert the ABCD matrix to a Y matrix:
2 3 2 1 3
Y11 Y12 2gm r 0
4 5= 2g r
m 4 5
R e + r + g m R e r
Y21 Y22 1 0
The Y matrix of the 2-port consisting of the series resistor Rf is:
2 3 2 3
Y11 Y12 1 1
4 5= 1 4 5
Rf
Y21 Y22 1 1
The sum of the 2 Y matrices is the Y matrix for the overall circuit:
2 3 2 1 1 1
3
Y11 Y12 Re +r +gm Re r + Rf Rf
4 5=6 4
7
5.
2gm r 1 1
Y21 Y22 Re +r +gm Re r Rf Rf

The input impedance of a 2-port can be written in terms of the Y parameters of the
2-port and the load admittance, YL , as:
Y12 Y21
YIN = Y11 .
Y22 + YL
246 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

Insert the Y parameters of the overall 2-port and use YL = 1

RL :

1 1
1
Rf ( Re +r2g m r
+gm Re r
1
Rf )
YIN = + 1 1 .
R e + r + g m R e r Rf Rf + RL

The input impedance is then found from ZIN = YIN1 . After some simplification, the input
impedance can be written as:

(RL + Rf )(Re + r + gm Re r )
ZIN = .
Re + r + RL + Rf + gm r (Re + 2RL )

Similarly, the output admittance depends on the Y parameters and the source admittance,
YS :
Y12 Y21
YOU T = Y22 .
Y11 + YS
Inserting the Y parameters of the overall 2-port, use YS = RS ,
1
and ZOU T = YOU1T to show:

(RS + Rf )(Re + r + gm Re r ) + RS Rf
ZOU T = .
RS + Re + r + gm r (Re + 2RS )

The voltage gain of a 2-port depends on the Y parameters and the load admittance YL = RL :
1

Y21
Av = .
Y22 + YL
Insert the Y parameters and simplify to show:

R L Rf 1 RL
Av = 2gm 1 + .
RL + Rf 1 + Re (gm + r )
R L + Rf

7.7 Y, Z, h, ABCD relationships

Relationships between the parameter sets described in this Chapter are given in the following
sections. The determinants of the parameter matrices are defined as follows:

DY = Y11 Y22 Y21 Y12 (7.76)

DZ = Z11 Z22 Z21 Z12 (7.77)

Dh = h11 h22 h21 h12 (7.78)
DABCD = AD BC (7.79)

7.7.1 Converting to Y-parameters

2 Z22 Z12 3 2 1 h12 3 2 D DABCD
3
DZ DZ h11 h11 B B
Y11 Y12 6 7 6 7 6 7
=4 5=4 5=4 5
Y21 Y22 Z21 Z11 h21 Dh 1 A
DZ DZ h11 h11 B B
7.8. SUMMARY 247

7.7.2 Converting to Z-parameters

2 Y22 Y12 3 2 Dh h12 3 2 A DABCD
3
DY DY h22 h22
Z11 Z12 6 7 6 7 6 C C 7
=4 5=4 5=4 5
Z21 Z22 Y21 Y11 h21 1 1 D
DY DY h22 h22 C C

7.7.3 Converting to h-parameters

2 DZ Z12 3 2 1 Y12 3 2 B DABCD
3
Z22 Z22 Y11 Y11 D D
h11 h12 6 7 6 7 6 7
=4 5=4 5=4 5
h21 h22 Z21 1 Y21 DY 1 C
Z22 Z22 Y11 Y11 D D

7.7.4 Converting to ABCD-parameters

2 Z11 DZ 3 2 Y22 1 3 2 Dh h11 3
Y21 Y21 h21 h21
A B 6 Z21 Z21
7 6 7 6 7
=4 5=4 5=4 5
C D 1 Z22 DY Y11 h22 1
Z21 Z21 Y21 Y21 h21 h21

7.8 Summary
7.8.1 Z parameters
Z12 Z21
ZIN = Z11 (7.80)
ZL + Z22

Z12 Z21
ZOU T = Z22 (7.81)
ZS + Z11
Z21 ZL
AV = (7.82)
Z11 ZL + Z11 Z22 Z12 Z21
Z21
AI = (7.83)
Z22 + ZL

7.8.2 Y parameters
Y12 Y21
YIN = Y11 (7.84)
YL + Y22
Y12 Y21
YOU T = Y22 (7.85)
YS + Y11
Y21
AV = (7.86)
Y22 + YL
Y21 YL
AI = (7.87)
Y11 YL + Y11 Y22 Y12 Y21
248 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

7.9 References
1. Terman, Frederick Emmons, Radio Engineers Handbook, McGraw-Hill Book Com-
pany, 1943.

2. Krauss, Herbert L., Charles W. Bostian, and Frederick H. Raab, Solid State Radio
Engineering, John Wiley and Sons, 1980.
3. Ludwig, Reinhold, and Pavel Bretchko, RF Circuit Design - Theory and Applications,
Prentice-Hall, Inc., New Jersey, 2000.

4. Balabanian, Norman and Theodore A. Bickart, Electrical Network Theory, John Wiley
& Sons, 1969.
7.10. HOMEWORK PROBLEMS 249

7.10 Homework Problems

1. Suppose the Y-parameters of a 2-port are known.

(a) Derive an expression for the input admittance of the 2-port when it is terminated
with load admittance YL .
(b) Derive an expression for the voltage gain Av = V2 /V1 . Your result will depend
on some of the Y-parameters and YL .
(c) Using your results for parts 1a and 1b find an expression for the operating power
gain of the 2-port, G. The operating power gain is defined as

Pout
G= (7.88)
Pin
where Pout is the power delivered to the load and Pin is the power delivered to
the 2-port. Your result should be in terms of the Y-parameters of the 2-port and
YL .

2. Suppose the ABCD-parameters of a 2-port are known.

(a) Derive an expression for the input impedance of the 2-port when it is terminated
with load impedance ZL .
(b) Derive an expression for the output impedance of the 2-port when it is terminated
with source impedance ZS .
(c) Derive an expression for the voltage gain Av = V2 /V1 . Your result will depend
on some of the ABCD-parameters and ZL .
(d) Find an expression for the transducer power gain, GT , of the 2-port when it is
driven with a source having impedance ZS and terminated with load impedance
ZL . Your result will depend on the ABCD-parameters as well as ZS and ZL .
Simplify the expression as much as possible.

3. Find the Z parameters for the 2-port in Figure 7.24.

Figure 7.24: 2-port.

4. Find the Z-parameters for the T-network in Figure 7.25. Port 1 is on the left.
5. Find the ABCD parameters for the 2-port shown in Figure 7.24.
6. Consider the unilateral hybrid-pi model shown in Figure 7.26. Find an expression for
the available power gain of this 2-port. Your result should be expressed only in terms
of Z1 , Z2 , gm and the impedance of the source Zs (not shown). Start from the basic
definition of available power gain.
250 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

Z1 Z2
Z3

Figure 7.25: T-network.

+
V Z1 gm V Z2
-

Figure 7.26: Unilateral hybrid-pi model.

7. The small signal equivalent circuit model for an amplifier is shown in Figure 7.27.
The amplifier is designed to operate at a center frequency of 30 MHz. Find the 3 dB

14.1pF 1k 28.2pF
+
Vi RL = 3k
-

2H gmVi 1H

Figure 7.27: Small signal equivalent circuit model for an amplifier.

bandwidth of the operating power gain. Express your result in MHz. Start from
the basic definition of operating power gain.
8. A lossy network has been designed to provide 10 dB of attenuation (i.e. operating
power gain G = 10 dB) when the network is terminated with a 50 load. It is
also known that the input impedance of the network is 300 when the output of the
network is terminated with a 50 load. Find the transducer gain for this 2 port in a
system with ZS = ZL = 50 .
9. The 2-port shown in Figure 7.28 is characterized by its Z-parameters, where Z11 =
Z12 = Z21 = Z22 = R. Find Zin . Express your result in terms of R only. Hint: The
input impedance of a 2-port terminated with load impedance ZL is Zin = Z11 ZZ2212+Z
Z21
L
.

10. This exercise will provide some practice in building up a more complex 2-port by
combining simple 2-ports in series and in parallel. It is also intended to provide
some insight into why we utilize the broadband transformer in the amplifier that is
constructed in laboratory 3. Consider the small-signal equivalent circuit shown in the
7.10. HOMEWORK PROBLEMS 251

Z11 Z12
Z21 Z22
R
Zin

Figure 7.28: Arbitrary 2-port in series with a 2-port consisting of a shunt resistor (a series-
feedback configuration).

Figure. The resistances RS and RL are the source and load resistances, respectively,
and are external to the amplifier. This is a simple transistor amplifier with series (Re )
and shunt (Rf ) feedback. It is similar to the amplifier that you will design and build
in Lab 3, however it does not include the 1:1 transformer between the load and the
transistors collector.

+ +

(a) Find the Y-parameter matrix for the feedback amplifier. Do not include RS and
RL in the amplifier - they are the external source and load terminations.
(b) Find expressions for the input impedance (Zin ), output impedance (Zout ), and
voltage gain (Av ) of the amplifier when terminated with RS and RL .
(c) Now, follow the analysis given in section 7.6.4 to determine expressions for the
feedback resistance Rf , and power gain G under the assumption that the param-
eters are adjusted to provide a simultaneous (approximate) impedance match at
the input and output ports when both ports are terminated with resistance R.
Make a table similar to Table 3.1 in the Lab notes. Table 3.1 lists, for each value
of Re , the power gain and the value of Rf required to produce the impedance
match assuming that RS = RL = R = 50 , ICQ = 10 mA, and = 100. Table
3.1 was produced using equations 3.12 and 3.14 in the Lab notes. Compare the
results for the amplifier analyzed here and the one analyzed in the Lab notes.
What advantage (if any) does the addition of the transformer provide?

11. A 2-port is empirically found to have the following properties. (i) With the output port
252 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS

unterminated the input impedance ZIN = 100 and the voltage gain AV = VV21 = 0.5.
(ii) With the output port terminated in a short circuit, the input impedance ZIN =
75 and the current gain AI = II21 = 0.5. Find ZIN and AV when the output port
is terminated with ZL = 50 .
12. A resistor is added in parallel with the input of an existing 2-port, as shown in the Fig-
ure. Determine the overall ABCD matrix of the system. Your result will be expressed

A B
R C D

in terms of the ABCD parameters of the original 2-port (denoted by A, B, C, D) and

the resistance R.
13. An impedance inverter has the property that the input impedance of the network
is equal to the inverse of the load impedance, i.e. ZIN = ZL 1 . Determine what
properties an ABCD matrix must have to produce the impedance inverter function.
14. Consider the T-network shown in Figure 7.29. The ABCD parameters of the T-

Z1 Z2
Z3

Figure 7.29: T-network to be used as an impedance inverter.

network are:

A B 1 Z1 + Z3 Z1 Z2 + Z2 Z3 + Z1 Z3
=
C D Z3 1 Z2 + Z3

Determine values of the T-network impedances (Z1 , Z2 , Z3 ) that will cause the T-
network to be an impedance inverter.
15. A linear, time-invariant, 2-port is empirically determined to have the following prop-
erties: (i) with the output port unterminated, the input impedance ZIN = 500
and the voltage gain AV = 4. (ii) With the output port terminated in a short cir-
cuit, ZIN = 500 . (iii) With the input port unterminated, the output impedance
ZOU T = 200 . Find the voltage gain, AV , when the output port is terminated with
ZL = 50 .
16. An admittance inverter has the property that the input admittance of the 2-port is
equal to the inverse of the load admittance, i.e. YIN = YL 1 .

(a) Determine the constraints that the Y-parameters of an admittance inverter must
satisfy.
7.10. HOMEWORK PROBLEMS 253

Y2
Y1 Y1

Figure 7.30: Pi-network admittance inverter.

(b) Consider the Pi-network shown in Figure 7.30. Determine the values of the Pi-
network admittances (Y1 , Y2 ) that will cause the Pi-network to be an admittance
inverter.

17. The h-parameters, hij , relate the voltages and currents at the input and output of a
2-port according to:
V1 = h11 I1 + h12 V2
I2 = h21 I1 + h22 V2

(a) Find an expression for the input impedance, ZIN , of a 2-port when the output
port is terminated with load impedance ZL . Express your result in terms of the
h-parameters, hij , and ZL .
(b) Consider two 2-ports characterized by h-parameter matrices denoted by H1 and
H2 , respectively. Make a sketch that shows how to connect the two 2-ports
together to make a new 2-port which will have h-matrix H = H1 + H2 . You may
assume that the current flowing into the upper terminal of every port is equal to
the current flowing out of the lower terminal of the port.
254 CHAPTER 7. INTRODUCTION TO 2-PORT PARAMETERS
Chapter 8

2-port Scattering (S) Parameters

8.1 Introduction and Definition of S-parameters

Recall from Chapter 7 that each parameter set discussed therein has a reference impedance
associated with each port. In the cases discussed so far, the reference impedance is either
a short circuit or an open circuit. The four complex numbers that make up the various
parameter sets include two parameters that characterize the input and output impedance
or admittance of the 2-port when each port is terminated in its reference impedance and
two additional parameters that describe the forward and reverse transfer characteristics
of the 2-port when terminated with the reference impedance. In each case, the reference
impedance is simply the termination that forces one of the independent parameters (e.g.
voltage or current) to zero. For example, the reference impedance for the Y-parameters is a
short circuit because terminating a port with a short circuit forces the voltage at that port
to zero.
At high frequencies it is dicult to implement terminations that accurately represent
short- or open-circuits over a broad frequency range. Thus, it is desirable to define a
parameter set that is based on a finite reference impedance that is easily realized in practice.
In most applications, a reference impedance that is purely resistive and having a moderate
value is relatively easy to implement. In order to retain the feature that terminating a
port with the reference impedance will force the independent variable corresponding to that
port to zero, any parameter set the employs a finite reference impedance must be based on
independent variables that are linear combinations of voltage and current rather than just
voltage or current. For example, consider the fact that terminating the output port of a
2-port with some resistance R forces V2 /I2 = R, by Ohms Law (the minus sign arises
because positive current is defined to flow in to the port). Thus, when the output port is
terminated with R, we have V2 + RI2 = 0. Any constant times the quantity (V2 + RI2 ) could
be chosen as the independent variable in a parameter set based on a reference impedance
R.
Scattering parameters (or S-parameters) can be viewed as a parameter set that is based
on a finite reference impedance. In most cases the reference impedance is either 50 or
75 and is the same for both ports, although in some cases it is advantageous to allow
the reference impedance to be dierent at port 1 and port 2. In addition to being easier
to implement than short- or open-circuits, at HF, VHF, and microwave frequencies 2-ports
are much less likely to oscillate (become unstable) when terminated in a finite resistance or

255
256 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

conductance.
Define Zo (= Ro ) to be the reference termination for the S-parameter set. Throughout
our discussion we will assume that the reference termination is purely resistive and that the
same reference termination is used at both ports. Both of these assumptions can be relaxed
in a more advanced treatment of scattering parameters. Figure 8.1 shows the voltage and
current conventions that we will use for the S-parameter definitions:

I1 I2
+ +

V1 2-port V2
- -
I1 I2

Figure 8.1: Voltage and current conventions for S-parameters.

The S-parameter set is defined in terms of variables (a1 , b1 ) (a2 , b2 ) which are simply
linear combinations of the voltage and current at a particular port, i.e.,

V1 + Z o I 1
a1 = p (8.1)
2 Zo
V2 + Z o I 2
a2 = p (8.2)
2 Zo
V1 Z I
b1 = p o 1 (8.3)
2 Zo
V2 Z I
b2 = p o 2 (8.4)
2 Zo

p
a 1 + b1 = V1 / Z o (8.5)

p
a 2 + b2 = V2 / Z o (8.6)

p
a1 b1 = Zo I1 (8.7)

p
a2 b2 = Zo I2 (8.8)

The independent variables are a1 and a2 and the dependent variables are b1 and b2 . Re-
member, the reference impedance for a parameter set is that impedance which will force
one of the independent variables to zero when it is used to terminate the 2-port. The choice
of a finite reference impedance, Zo , forces us to choose the independent variables to be
proportional to Vi + Zo Ii , as defined in Equations 8.1 and 8.2, so that they will be forced
to zero when the 2-port is terminated in the reference impedance, Zo .
8.2. INTERPRETATION OF S-PARAMETERS 257

The variables defined in Equations 8.1-8.4 have been normalized with respect to the
square root of the reference impedance, Zo . With this normalization (and assuming that
the voltages and currents are rms values) the time-averaged power delivered to a given port
(denoted by i) can be written as P = Re[Vi Ii ] = |ai |2 |bi |2 . Thus, the variables have units
of [Watts]1/2 .
Given the definitions in Equations 8.1-8.8, the scattering parameters are defined as fol-
lows:

b1 = S11 a1 + S12 a2 (8.9)

b2 = S21 a1 + S22 a2 (8.10)

Thus
b1
S11 = |a =0 (8.11)
a1 2
b1
S12 = |a =0 (8.12)
a2 1
b2
S21 = |a =0 (8.13)
a1 2
b2
S22 = |a =0 (8.14)
a2 1

8.2 Interpretation of S-parameters

Consider S11 :
b1
S11 = |a =0 (8.15)
a1 2
Setting a2 = 0 is equivalent to setting

V2 = Zo I2

V2
= Zo (8.16)
I2

That is, to set a2 = 0, we terminate port 2 with the reference impedance, Zo . Thus

b1 V1 Z o I 1
S11 = |a2 =0 = |output port terminated in Zo (8.17)
a1 V1 + Z o I 1

V1 /I1 Zo
= |output port terminated in Zo
V1 /I1 + Zo
or,
Zin Zo
S11 = |output port terminated in Zo (8.18)
Zin + Zo
where Zin is the input impedance of the 2-port when the output is terminated in Zo .
258 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

Recall from transmission line theory that S11 has the same form as the reflection coe-
cient, , that relates the incident and reflected voltage waves on a transmission line

V reflected voltage wave

= = (8.19)
V + incident voltage wave

where the total voltage on the line, V, is given by

V =V + V+ (8.20)

Comparison of Equations 8.19 and 8.20 with

b1
S11 = (8.21)
a1
and p
V1 = (a1 + b1 ) Zo (8.22)
leads to the following correspondence between the variables a1 and b1 and the voltage waves
on a transmission line:
1
ai = p (Voltage wave incident on port i) (8.23)
Zo

1
bi = p (Voltage wave reflected from port i) (8.24)
Zo
This is an interesting observation, but where is the transmission line? Although there
has been no mention of transmission lines in the system so far, it is helpful to imagine
that the source and load are connected to the 2-port through sections of transmission line
with characteristic impedance Zo . Well assume that the sections of transmission line have
infinitesimal length, so that they do not aect the electrical characteristics of the system,
as in Figure 8.2.

ZS I1 I2
+ +
a1 2-port a2
Vs V1 V2 ZL
b1 - -
b2

Zo I1 I2 Zo

Figure 8.2: Source and load connected to 2-port through transmission line with impedance
Zo

Employing this conceptual model, the a and b variables may be interpreted as repre-
senting the (normalized) voltage waves that would exist on the sections of transmission line.
When working with S parameters, the model can be used to visualize the a and b variables
as incident and reflected (normalized) voltage waves, although in practice the terminations
may be connected directly to the 2-port without any intervening transmission lines. (Actu-
ally, in practice you will find that it is virtually impossible to get signals into and out of a
8.2. INTERPRETATION OF S-PARAMETERS 259

2-port without employing transmission lines between the source and the input port and the
output port and the load.)
The model shown in Figure 8.2 makes it clear that setting ZL = Zo will make a2 = 0,
since there will be no reflection from the load termination. Notice that setting Zs = Zo
would not make a1 = 0 in the circuit shown in Figure 8.2, because the source would cause
a non-zero incident wave (a1 ) to exist on the input line.
Now consider S21 :
b2
S21 = |a2 =0 (8.25)
a1
As discussed earlier, to set a2 = 0, terminate port 2 with Zo . To aid in obtaining an
intuitive feel for the significance of S21 , it is helpful to consider the special circuit shown in
Figure 8.3 where the source impedance is assumed to be equal to the reference impedance.

Zo I1 I2
+ +
a1
Vs
b1 V1 2-port V2
b2
Zo
- -

I1 I2

Figure 8.3: A 2-port embedded in a system with ZS = ZL = Zo .

Note that the circuit has been drawn without showing any intervening lengths of trans-
mission line between the 2-port and the terminations. The as and bs can be thought of
as existing at the node where the the 2-port is connected to the terminations. To solve
for S21 , first relate b2 to the output voltage using Equation 8.4 and the auxiliary relation
V2 /I2 = Zo :
1
b2 = p (V2 Zo I2 ) (8.26)
2 Zo
p
= V2 / Z o
Then a1 can be related to the open circuit voltage of the source
1
a1 = p (V1 + Zo I1 ) (8.27)
2 Zo
using
I1 = (Vs V1 )/Zo (8.28)
Thus
1
a1 = p (V1 + Vs V1 ) (8.29)
2 Zo
p
= Vs /2 Zo
Finally,
b2 V2
S21 = |a =0 = |Z =Z (8.30)
a1 2 Vs /2 L o
260 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

So for the circuit shown in Figure 8.3, S21 is the voltage across the load divided by 1/2 of
the open circuit source voltage. Notice that if the 2-port is removed and the load connected
directly to the source, then the voltage across the load would be Vs /2. In other words, S21 is
the insertion voltage gain in a system where the source and load impedances are both Zo .
Remember that this result is not the definition of S21 ; it is a special result that was derived
from the definition for the case when both the source and load impedances are equal to Zo .
This result is often useful when it is necessary to derive an expression for S21 starting from
the circuit model of a 2-port. S12 has a similar interpretation - it is the reverse insertion
voltage gain, i.e. the insertion gain that would be measured if the two port was inserted
into the system backwards, i.e. with the output port connected to the source and input port
connected to the load. The insertion gain interpretion is useful to remember when trying to
get an intuitive feel for published or measured values of S21 or S12 .

8.2.1 Example - Computing S-parameters for a given circuit model

The 2-port in Figure 8.4 consists of a series impedance. To find S11 , terminate the 2-port

ZS

Figure 8.4: 2-port consisting of a series impedance.

in Zo and find Zin as in Figure 8.5.

ZS
Zin Zo

Figure 8.5: Setup for finding S11 .

Zin = Zs + Zo (8.31)

Zin Zo Zs
S11 = = (8.32)
Zin + Zo Zs + 2Zo
Because the circuit is not changed if the two ports are reversed, S22 = S11 .
To find S21 , refer to Figure 8.6.
Zo
V2 = Vs (8.33)
2Zo + Zs
V2 Zo
=
Vs 2Zo + Zs
V2 2Zo
S21 = = (8.34)
Vs /2 2Zo + Zs
8.2. INTERPRETATION OF S-PARAMETERS 261

Zo
ZS
+
Vs V2 Zo

Figure 8.6: Setup for finding S21 . Embed the 2-port in a Zo system, and then S21 = 2V2 /Vs .

Again, because the circuit is not changed if port 1 and port 2 are reversed, S12 = S21 .
The procedure illustrated in the preceeding example can be applied to any 2-port, and
provides a general method for calculating the S-parameters when a circuit model is available.

8.2.2 Summary of 2-port S-parameters:

S11 = Zin Zo
Zin +Zo |ZL =Zo Input reflection coecient with output terminated in Zo

S22 = Zout Zo
Zout +Zo |ZS =Zo Output reflection coecient with input terminated in Zo

S21 = Forward insertion voltage gain when source and load impedances are Zo

S12 = Reverse insertion voltage gain when source and load impedances are Zo

8.2.3 Special relationships for reciprocal, and lossless 2-ports

8.2.3.1 Reciprocal 2-ports
As discussed in Chapter 7, any 2-port that contains no sources, and is constructed only of
resistors, capacitors, inductors, transformers, and transmission lines will satisfy reciprocity.
The [S] matrix of a reciprocal 2-port will be symmetric, i.e. S12 = S21 .

8.2.3.2 Lossless 2-ports

A lossless 2-port contains no dissipative elements, and hence the time-averaged real power
absorbed by the network is zero. A 2-port network constructed only of lossless transmission
lines, transformers, and lossless Ls and Cs will be reciprocal and lossless. As noted in
Chapter 7, the constraint that the power absorbed by the network is zero can be written as

<(V1 I1 ) + <(V2 I2 ) = 0. (8.35)

In terms of the variables used to define the scattering parameters, this constraint becomes:

|a1 |2 |b1 |2 + |a2 |2 |b2 |2 = 0. (8.36)

Use the definitions of the S-parameters to write the bi in terms of the ai , then equation 8.36
can be written as

|a1 |2 + |a2 |2 = (|S11 |2 + |S21 |2 )|a1 |2 + (|S12 |2 + |S22 |2 )|a2 |2 + 2<{(S11 S12 + S21 S22 )a1 a2 }.
262 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

Since this constraint must be true for any possible excitation, it must hold for a1 = 0 and
for a2 = 0. Setting a1 = 0 we find that

|S12 |2 + |S22 |2 = 1. (8.37)

Setting a2 = 0 we find that

|S11 |2 + |S21 |2 = 1. (8.38)
Hence, for arbitrary a1 and a2 we must also have

S11 S12 + S21 S22 = 0. (8.39)

These three relationships say that the dot product of a column of [S] with the conjugate of
the same column is unity, whereas the dot product of a column of [S] with the conjugate of
the other column is zero. Another way to write these relationships is

[S][S] = [I],

where [ ] denotes the conjugate transpose and [I] is the identity matrix.

8.2.3.3 Reciprocal and lossless 2-ports

If a 2-port is reciprocal and lossless, then S12 = S21 and the dot products of the rows of [S]
will satisfy the same constraints as do the columns, in which case the [S] matrix is unitary,
i.e.
[S][S] = [S] [S] = [I].

8.3 Applications of Scattering Parameters

Consider in Figure 8.7 a 2-port with arbitrary source and load terminations (ZS , ZL ).

ZS

Vs [S]Zo ZL

S in out L

Figure 8.7: 2-port with arbitrary source and load terminations

The reference impedance is assumed to be Zo . Define in and out to be the input and
output reflection coecients, i.e.,

Zin Zo
in = (8.40)
Zin + Zo
Zout Zo
out = (8.41)
Zout + Zo
8.3. APPLICATIONS OF SCATTERING PARAMETERS 263

The source reflection coecient, S, is defined as

ZS Zo
S = (8.42)
ZS + Zo
and the load reflection coecient, L, is defined as
ZL Zo
L = (8.43)
ZL + Zo

8.3.1 Derivation of Input and Output Reflection Coecients and

Voltage and Current Gains
Consider the input reflection coecient, in = b1 /a1 . We expect in to be a function of the
four S-parameters as well as the load reflection coecient, L . We start by noting that the
load termination imposes the following constraint:
a2
= L (8.44)
b2
This form of the load constraint may be derived by starting with the voltage/current form:
a2
V2 a 2 + b2 +1
ZL = = Zo = Zo ab22 . (8.45)
I2 a 2 b2 b2 1

Solve for a2
b2 :
a2 ZL Zo
= = L. (8.46)
b2 ZL + Zo
Use Equation 8.44 and Equation 8.10 to write
S21 L
a2 = a1 (8.47)
1 S22 L

Inserting Equation 8.47 into Equation 8.9 yields

b1 S12 S21 L
in = = S11 + (8.48)
a1 1 S22 L

Similarly, it can be shown that the output reflection coecient is given by

b2 S12 S21 S
out = = S22 + (8.49)
a2 1 S11 S

Inspection of Equation 8.48 shows that the input reflection coecient reduces to S11 if
the load reflection coecient is 0 ( L = 0). This is consistent with the discussion in the
previous section regarding the interpretation of S11 . Similarly, out = S22 if S = 0. In
general, however, the input and output reflection coecients depend on the way the 2-port
is terminated. In the special case of a 2-port with S12 = 0, the input and output reflection
coecients are independent of the terminations. Such a 2-port is said to be unilateral, since
the device does not exhibit any reverse transmission. In other words, if a unilateral 2-port
(S12 = 0) is excited at port 2, no response will result at port 1. Although it is usually not
possible to construct a 2-port that is perfectly unilateral, in some cases the S12 coecient
264 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

is small enough that the 2-port can be considered to be approximately unilateral. The case
where S21 = 0 is also unilateral but is usually not useful, since this 2-port would not give a
response at the output when excited at the input.
The voltage gain is:
a2
V2 a 2 + b2 +1
AV = = = ab12 b1 . (8.50)
V1 a 1 + b1 b + b 2 2

Use Equation 8.44 in Equation 8.10:

a1 1
= (1 L S22 ). (8.51)
b2 S21
Then use 8.44 and 8.51 in Equation 8.9 to show that

b1 S11
= (1 L S22 ) + S12 L. (8.52)
b2 S21
Finally, use 8.44, 8.51, and 8.52 in 8.50 to obtain the final result:

S21 (1 + L )
AV = . (8.53)
(1 + S11 )(1 L S22 ) + S12 S21 L

Similarly, the current gain is:

a2
I2 a2 b2 b2 1
AI = = = a1 b1
. (8.54)
I1 a1 b1 b2 b2

Use 8.44, 8.51, and 8.52 in 8.54 to obtain the final result:

S21 ( L 1)
AI = . (8.55)
(1 S11 )(1 L S22 ) S12 S21 L

8.3.2 Stability of 2-ports

Before detailed design calculations based on a particular 2-port are made, it is usually
necessary to investigate whether the 2-port is potentially unstable. Of course, stability is
only a concern when the 2-port contains active elements such as a transistor or a negative
resistance device. The questions that need to be answered are:

1. Is the 2-port unconditionally stable? That is, is there any combination of passive
source and load terminations ( L , S ) for which the 2-port will oscillate? If not, the
2-port is said to be unconditionally stable.

2. If the 2-port is not unconditionally stable, i.e., if it is potentially unstable, then we

would like to be able to determine which source and load terminations make it unstable.
If the 2-port is to be used as an amplifier, we would avoid the unstable terminations.
If the goal is to design an oscillator, the designer would deliberately choose source and
load terminations to cause oscillation.

The stability question can be studied using the negative resistance concept. Considering
Figure 8.8, suppose that one of the impedances represents the input or output impedance
of the 2-port.
8.3. APPLICATIONS OF SCATTERING PARAMETERS 265

Z1 Z2

Figure 8.8: Z1 or Z2 represents input or output impedance of 2-port

Remember that the circuit will oscillate when the two impedances are connected if Re
[Z1 + Z2 ] 0 and Im [Z1 + Z2 ] = 0. Thus, to make an oscillator, one of the Zi s must
have a negative real part.
For a particular 2-port, if there is some load impedance that makes the real part of
Zin 0, or if there is some source impedance that makes the real part of Zout 0, then
the 2-port is potentially unstable because it is possible to choose a passive source or load
termination that will make the system oscillate.
An impedance with a negative resistive part corresponds to a reflection coecient with
a magnitude greater than 1. For example, consider Zin and write Zin = Rin + j Xin , then
Zin Zo
in = (Zo = Ro (real)) (8.56)
Zin + Zo
(Rin Ro ) + jXin
=
(Rin + Ro ) + jXin
1/2
(Rin Ro )2 + Xin2
| in | = 2 (8.57)
(Rin + Ro )2 + Xin
Examination of Equation 8.57 leads to the following observations:
If Rin > 0 ) | in | <1 (8.58)
If Rin 0 ) | in | 1 (8.59)
To determine whether a 2-port is potentially unstable, we check to see if either | in | or
| can be larger than or equal to unity.
out |
First consider | in |:
S12 S21 L
| in | = |S11 + | (8.60)
1 S22 L

S11 L (S11 S22 S12 S21 )

= | |
1 S22 L

Define the determinant of the S-parameter matrix, D, as follows:

D = S11 S22 S12 S21 (8.61)
Then Equation 8.60 becomes
S11 LD
| in |
| =| (8.62)
1 S22 L
We can now set | in | = 1 and solve for the corresponding locus of points in the L plane,
i.e., we can solve for the values of L that make | in | = 1. Setting | in | = 1:
|S11 L D| = |1 S22 L| (8.63)
266 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

The absolute value signs can be eliminated by squaring both sides of Equation 8.63:

(S11 L D)(S11 LD ) = (1 S22 L )(1 S22 L) (8.64)

Expanding both sides of the equation and collecting terms yields

2 2 Re( L (S22 DS11 )) |S11 |2 1
| L| + + (8.65)
|D|2 |S22 |2 |D|2 |S22 |2

In deriving Equation 8.65, use has been made of the following identity

z + z = 2Re(z) (8.66)

where the operator Re() extracts the real part of its argument. Also note that

z z = 2Im(z) (8.67)

where the operator Im() extracts the imaginary part of its argument. Now, it is convenient
to rewrite Equation 8.65 in terms of the real and imaginary parts of L .
Let L = UL + jVL and substitute Equation 8.65. After some fairly extensive algebraic
manipulation, Equation 8.65 can be written in the form

(UL UCL )2 + (VL VCL )2 = rL 2 (8.68)

where

Re(DS11 S22 )
UCL = 2
(8.69)
|D| |S22 |2

Im(S22 DS11 )
VCL = 2 2
(8.70)
|D| |S22 |
|S12 S21 |
rL = (8.71)
||S22 |2 |D|2 |

Equation 8.68 indicates that the locus of points in the L plane that correspond to
input reflection coecients with unit magnitude is a circle with center at the complex point
CL = UCL + jVCL and radius rL . It is convenient to write the coordinates of the center of
the stability circle as a complex number, i.e. define CL = UCL + jVCL . Then

S22 D S11
CL = (8.72)
|S22 |2 |D|2

The circle, when plotted on the L plane, is referred to as the L -plane stability circle.
Consider the stability circle shown in Figure 8.9.
Figure 8.9 contains two circles. The circle that is centered on the origin is a unit circle
(circle with radius = 1) and represents the outer boundary of the Smith Chart. Points
within that circle correspond to load reflection coecients ( L ) with magnitudes less than 1
or, equivalently, load impedances with a positive real part. The other circle is the stability
circle, and values of L which lie on that circle will map to input reflection coecients
with magnitude equal to 1 or, equivalently, input impedances that are purely reactive. The
stability circle represents the boundary between the region of the L plane that maps to
8.3. APPLICATIONS OF SCATTERING PARAMETERS 267

Stability
L Plane VL Circle
rL
CL

UL

Unit
Circle

Figure 8.9: L -plane stability circle

input reflection coecients with magnitude less than one (| in | < 1) and the region that
maps to (| in | > 1). The region of the L plane that maps to (| in | > 1) is called the
unstable region of the L plane. The two regions (stable and unstable) correspond to
the regions inside and outside the stability circle. To decide whether the unstable region
corresponds to the region inside or outside of the stability circle, it is sucient to examine
the value of S11 for the 2-port under consideration. Note carefully that the origin of the L
plane (the point L = 0) maps to S11 , i.e., recall

S12 S21 L
in = S11 + (8.73)
1 S22 L

so that when L = 0, we have

in = S11 (8.74)

Now consider Figure 8.9 and suppose that it is known that |S11 | < 1. The origin in the
L plane is outside the stability circle and this point maps to S11 . This leads us to conclude:
(i) the region of the L plane that lies outside the stability circle maps to | in | < 1 and (ii)
the region inside the stability circle maps to | in | > 1. The region inside the stability circle
would therefore be referred to as the unstable region of the L plane.
Thus far we have treated only the so-called L -plane stability circle. A complete charac-
terization of the 2-ports stability requires that the S -plane stability circle also be studied.
The S -plane stability circle corresponds to the values of S which map to output reflection
coecients ( out ) with magnitude equal to one. The S -plane stability circle is described
by

(US UCS )2 + (VS VCS )2 = rS 2 (8.75)

268 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

where

CS = UCS + jVCS (8.76)

S11 D S22
= 2
(8.77)
|S11 | |D|2
|S12 S21 |
rS = (8.78)
||S11 |2 |D|2 |
A decision regarding whether the unstable region in the S -plane lies inside or outside
of the S -plane stability circle is made by examining the magnitude of S22 and noting that
the point S = 0 (the origin of the S -plane) maps to S22 .
It is now appropriate to define the concept of unconditional stability. An unconditionally
stable 2-port has the property that no choice of passive source and load terminations will
make the 2-port oscillate. In other words, an unconditionally stable 2-port has | in | < 1 and
| out | < 1 for any choice of passive source and load terminations. The restriction to passive
sources and loads means that we limit our attention to sources with | S | 1 and loads with
| L | 1, i.e., those regions of the S - and L -planes that lie inside and on the unit circle
centered on the origin. A 2-port is unconditionally stable if the unstable regions of the S -
and L -planes lie completely outside of the unit circles. Once the center coordinates and
the radii of the stability circles have been computed, the circles can be plotted and, utilizing
the known magnitudes of S11 and S22 , we can determine whether the unstable regions lie
outside of the unit circles.
As an alternative to plotting the stability circles, it is possible to derive a relatively
simple algebraic criterion (or set of criteria) that must be satisfied in order for a 2-port to
be unconditionally stable. The derivation of one such criterion is given in section 8.3.4. A
summary of the various sets of criteria that have been derived is given here. In each case,
if the criterion (or set of criteria) is satisfied, then the 2-port is unconditionally stable:

K > 1, |S12 S21 | < 1 |S11 |2 (8.79)

2
K > 1, |S12 S21 | < 1 |S22 | (8.80)
K > 1, B1 > 0 (8.81)
K > 1, B2 > 0 (8.82)
K > 1, |D| < 1 (8.83)

1 |S11 |2
ES = D| + |S S | > 1 (8.84)
|S22 S11 12 21

1 |S22 |2
0ES = D| + |S S | > 1 (8.85)
|S11 S22 12 21

where

1 |S11 |2 |S22 |2 + |D|2

K= >1 (8.86)
2 |S12 | |S21 |

D = S11 S22 S12 S21 (8.87)

8.3. APPLICATIONS OF SCATTERING PARAMETERS 269

B1 = 1 + |S11 |2 |D|2 |S22 |2 (8.88)

2 2 2
B2 = 1 + |S22 | |D| |S11 | (8.89)
Any one set of criteria given by 8.79-8.85 is a necessary and sucient set of criteria for
a 2-port to be unconditionally stable. If any one of the sets of stability criteria are satisfied,
then no choice of passive source and load terminations will make the 2-port oscillate. This
statement applies only at the frequency where the S-parameters were measured. It is possible
for a 2-port to be unconditionally stable in a certain frequency band but potentially unstable
in some other frequency band. In practice, stability should be checked at many frequencies
in the bandwidth within which the 2-port has appreciable gain. This requires that the S-
parameters be measured at many frequencies. Checking one of the sets of conditions given
in equations 8.79-8.85 is easier than plotting the stability circles, especially when one wishes
to quickly determine the stability status over a wide frequency range; however if the 2-port
turns out to be potentially unstable, then the stability circles provide detailed information
on which terminations make the 2-port unstable.
Finally, note that the parameter K defined in equation 8.86 and used in criteria 8.79
through 8.83 is called the Rollet Stability Factor. While the condition K>1, by itself, does
not guarantee that a 2-port is unconditionally stable, it turns out that K plays an important
role in determining whether or not a 2-port can be simultaneously matched with passive
terminations at both the input and output ports.

8.3.3 Example - 2-port stability analysis.

A 2-port has S-parameters (Zo = 50 )

S11 = 0.4\ 20 (8.90)

S12 = 0.1\40 (8.91)

S21 = 7.5\150 (8.92)

S22 = 0.6\ 30 (8.93)

The Rollett Stability factor for this 2-port is K = 0.853 < 1, so the 2-port is potentially
unstable. The coordinates of the stability circles are

UCS = 0.163 VCS = 0.589 rS = 1.17

(8.94)
UCL = 0.401 VCL = 0.913 rL = 1.70

The stability circles for this 2-port are plotted in Figures 8.10 and 8.11. The stable region
of the S - and L -planes is shaded. For the S -plane stability circle, the stable region was
identified by noting that the origin of the S -plane ( S = 0) maps to an output reflection
coecient out = S22 . Since the origin of the S -plane is in the interior of the stability circle
and |S22 | < 1, we conclude that the interior of the stability circle maps to output reflection
coecients | out | < 1; hence, the interior of the stability circle is the stable region. Similar
reasoning was used to determine that the interior of the L -plane stability circle was stable.
To verify that the stable and unstable regions have been correctly identified, consider the
270 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

S Plane VS

Unit
Circle

US
CS

rS

Stability
Circle

Figure 8.10: S plane stability circle. The stable region, corresponding to values of | S|
that map to | out | < 1, is shaded.

L Plane VL

Unit
Circle
A
B

UL

CL
rL
Stability
Circle

Figure 8.11: L plane stability circle. The stable region, corresponding to values of L that
map to | in | < 1, is shaded. Point A, at L = 0.5 + j0.7 = 0.86\54 , lies in the unstable
region. If the 2-port is terminated with this load reflection coecient, the input reflection
coecient will have magnitude greater than one, corresponding to an input impedance with
negative real part (<(Zin ) < 0).
8.3. APPLICATIONS OF SCATTERING PARAMETERS 271

two points labeled A and B in the L -plane plot. The point A is in the unstable region
and represents a load reflection coecient of L = 0.5 + j0.7 = 0.86\54 . Using this value
in Equation 8.48 yields in = 1.3\ 76.4 which represents an input impedance with a
negative real part, as expected. Point B, which is inside the stable region, represents a
load reflection coecient L = 0.4 + j0.4 = .57\45 . The input reflection coecient is
in = 0.70\ 83 .
The following example is identical to the previous one, except that the magnitude of S12
has been reduced from 0.1 to 0.01.
2-port stability analysis. A 2-port has S-parameters (Zo = 50 ):

S11 = 0.4\ 20 (8.95)

S12 = 0.01\40 (8.96)

S21 = 7.5\150 (8.97)

S22 = 0.6\ 30 (8.98)

The stability factor, K, is 3.74 (> 1), and we also have

1 |S11 |2 > |S12 S21 | (8.99)

1 |S22 |2 > |S12 S21 | (8.100)

so the 2-port is unconditionally stable. (Note, only one of 8.99 and 8.100 needs to be
checked.) Stability circles for this example are shown in Figures 8.12 and 8.13. As in the
first example, we have made use of the fact that the origin of the S - and L -planes map
to output and input reflection coecients, respectively, with magnitudes less than 1. Hence
the origin of each plane is in the stable region. Since the origin is outside the stability circle,
the region outside the circle is the stable region and the region inside is the unstable region.
Since the unstable regions do not include any passive source or load terminations, the 2-port
is said to be unconditionally stable.
In the following section, a simple test is derived that can predict whether or not a
particular 2-port is unconditionally stable.

8.3.4 Derivation of a Criterion for Unconditional Stability of 2-

ports
All of the essential information necessary to investigate the stability of a 2-port (at a par-
ticular frequency) is contained in the 4 parameters that define the stability circles in the
S and L planes. In this section, we derive a criterion that can be used to quickly check
to see if a 2-port is unconditionally stable. As already pointed out, numerous sets of nec-
essary and sucient criteria have been derived that can be used to determine whether or
not a 2-port is unconditionally stable.1 The criterion derived in this section is particularly
interesting because it involves only a single inequality and because the numerical value of
1 Lombardi, G. and B. Neri, Criteria for the evaluation of unconditional stability of microwave linear
two-ports: a critical review and new proof, IEEE Trans. MTT, Vol. 47, No. 6, June, 1999, p746.
272 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

Unstable
Region
S Plane VS CS
rS
Unit
Circle

US

Figure 8.12: S plane stability circle. The unstable region is shaded.

L Plane VL Unstable
Region

UL

Unit
Circle

Figure 8.13: L plane stability circle. The unstable region is shaded.

8.3. APPLICATIONS OF SCATTERING PARAMETERS 273

the stability parameter has the useful property that its magnitude represents the smallest
distance between the origin of the reflection coecient plane and the edge of the unstable
region.
In order for a 2-port to be unconditionally stable, it is necessary for the unstable region
in the L and S planes to lie outside of the unit circle. Consider the L plane first, and
assume that |S11 | < 1 (this is a necessary condition for unconditional stability, because if
|S11 | 1 a negative, or zero, real part of ZIN is obtained by setting ZL = Zo . In this case,
the 2-port can oscillate with a passive source termination and hence is potentially unstable.)
There are two cases to consider.
1. First, suppose that the stability circle does not enclose the origin of the L plane
(i.e. |CL | > rL ). Then the inside of the stability circle is the unstable region. If the
2-port is unconditionally stable it is necessary for the minimum distance between the
origin and the edge of the stability circle to be larger than 1, i.e. it is necessary that
|CL | rL > 1. Inserting the equations for |CL | and rL we find that this amounts to:

|S22 D S11 | |S12 S21 |
>1 (8.101)
||S22 |2 |D|2 | ||S22 |2 |D|2 |
or
|S22 D S11 | |S12 S21 |
>1 (8.102)
||S22 |2 |D|2 |
2. Now consider the other possibility, namely that the stability circle encloses the origin,
i.e. that |CL | < rL . In this case, the necessary condition |S11 | < 1 implies that the
exterior of the stability circle is the unstable region. In order to have the unstable
region lie completely outside of the unit circle, we require that the stability circle
completely encloses the unit circle centered at L = 0, i.e.:

rL |CL | > 1

or
|S12 S21 | |S22 DS11 |
>1 (8.103)
||S22 |2 |D|2 |
Note carefully that equation 8.102 guarantees that the unstable region is outside of the unit
circle provided that |CL | > rL whereas equation 8.103 guarantees that the unstable region
is outside of the unit circle when |CL | < rL . Now consider what the statement |CL | > rL
(or |CL | < rL ) implies in terms of the S parameters. Consider |CL | > rL :

|S22 D S11 | |S12 S21 |
>
||S22 |2 |D|2 | ||S22 |2 |D|2 |
or

|S22 D S11 | > |S12 S21 |
Squaring both sides (note that squaring both sides does not lose any generality, because
both sides are always positive) yields

(S22 D S11 )(S22
DS11 ) > |S12 S21 |2

and expanding the LHS yields

|S22 |2 S22 DS11 S22 D S11 + |D|2 |S11 |2 > |S12 S21 |2
274 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

Expand the second and third terms on the LHS using D = S11 S22 S12 S21 :

|S22 |2 S11 S22 (S11 S22 S12 S21 ) S11 S22 (S11 S22 S12 S21 ) + |D|2 |S11 |2 > |S12 S21 |2

and group terms on the LHS:

|S22 |2 2|S11 |2 |S22 |2 + S11 S22 S12 S21 + S11 S22 S12 S21 + |D|2 |S11 |2 > |S12 S21 |2 (8.104)

Now, noting that

|D|2 = |S11 |2 |S22 |2 S11 S22 S12 S21 S11 S22 S12 S21 + |S12 |2 |S21 |2 (8.105)

Equation 8.105 can be used to eliminate the complex terms (3rd and 4th term on LHS) in
equation 8.104. The result is:

|S22 |2 |S11 |2 |S22 |2 + |S12 |2 |S21 |2 |D|2 + |D|2 |S11 |2 > |S12 |2 |S21 |2

or, after grouping terms, we determine that |CL | > rL amounts to:

(|S22 |2 |D|2 )(1 |S11 |2 ) > 0 (8.106)

Since we have already assumed that |S11 | < 1, the second term on the LHS is always positive
so the combined requirements that the stability circle not enclose the origin ( |CL | > rL )
and that the inside of the stability circle represent the unstable region (|S11 | < 1) means
that:
|S22 |2 |D|2 > 0 (8.107)
Similarly, the requirements that the stability circle encloses the origin (|CL | < rL ) and that
the inside of the stability circle represents the unstable region (|S11 | < 1) amounts to:

|S22 |2 |D|2 < 0 (8.108)

We now know that criterion 8.79 applies only when |S22 |2 |D|2 > 0 and criterion 8.80
applies when |S22 |2 |D|2 < 0. Note that criteria 8.79 and 8.80 dier only in the sign of the
numerator on the LHS, and that the denominator is the magnitude of |S22 |2 |D|2 . Thus,
these criteria can be combined into into a single inequality which is valid for either case:

|S22 D S11 | |S12 S21 |
>1 (8.109)
|S22 |2 |D|2

There is an apparent singularity in equation 8.109 when |S22 |2 |D|2 = 0. This corresponds
to the situation where the radius of the stability circle approaches infinity, i.e. when the
stability circle degenerates into a straight line. The singularity can be removed by noting
that:

|S22 S11 D|2 |S12 S21 |2
|S22 |2 |D|2 =
1 |S11 |2

(|S22 S11 D| |S12 S21 |)(|S22 S11 D| + |S12 S21 |)
= (8.110)
1 |S11 |2
Use equation 8.110 in equation 8.109 to write:

1 |S11 |2
ES = >1 (8.111)
|S22 D S11 | + |S12 S21 |
8.3. APPLICATIONS OF SCATTERING PARAMETERS 275

The parameter ES is referred to as the Edwards-Sinsky stability criterion,2 and has a

useful geometric interpretation: ES is the minimum distance between the origin of the L
plane and the unstable region. If the parameter ES is negative, it means that the unstable
region includes the origin of the L plane.
If the identical analysis is carried out for the S plane stability circles, we derive the
dual constraint:
0 1 |S22 |2
ES = >1 (8.112)
|S11 D S22 | + |S12 S21 |
0
The geometric interpretation of ES is the same as that of ES , but applied to the S plane.
It turns out that either criterion 8.111 or criterion 8.112 is a necessary and sucient
test for unconditional stability of a 2-port, even though each one was derived separately by
considering only one of the L plane or S plane stability circles. To show that this is true
0
we can prove that if ES 1 then ES 1 and the converse are both true statements.
Suppose ES 1 and choose a passive load termination, call it Lu , that causes | in | to be
1 , i.e. | in ( Lu )| 1. The negative resistance criterion for steady-state oscillation will be
satisfied at the input port if the source reflection coecient is chosen to be Su = in ( Lu ) 1
because this choice causes Su in ( Lu ) = 1. (As an exercise, you may wish to verify that
the condition Z1 + Z2 = 0 is equivalent to 1 2 = 1.) Note that Su is passive, since
| Su | = 1/| in ( Lu )| 1. Now, it is left as an exercise to show that out ( Su ) = Lu1 .
Thus out ( Su ) Lu = 1, which means that the oscillation condition is also satisfied at the
output port. The converse can be proven by simply changing subscripts in the preceeding
argument. It follows that if ES > 1 then 0ES > 1 and vice versa. Hence we can determine
whether or not a 2-port is unconditionally stable by checking either criterion 8.111 or 8.112.

8.3.5 Terminations for Simultaneous Conjugate Match

Whenever K > 1 it is possible to find a combination of passive source and load termina-
tions for which both the input and output of the 2-port are conjugately matched. For a
given available source power, a simultaneous conjugate match condition leads to the highest
possible power delivered to the load. The particular source and load reflection coecients
which result in a simultaneous conjugate match at both ports are denoted by ms and ml .
Equations for ms and ml can be derived by solving Equations 8.113 and 8.114, which
enforce the conjugate match relationship at the input and output of the 2-port, i.e.,
S12 S21 ml
in = ms = S11 + (8.113)
1 S22 ml

S12 S21 ms
out = ml = S22 + (8.114)
1 S11 ms

The solution to Equations 8.113 and 8.114 is given by

p
B1 B12 4|C1 |2
ms = 2C1

B1 = 1 + |S11 |2 |D|2 |S22 |2 (8.115)

C1 = S11 DS22
2 Edwards,M. L. and J. H. Sinsky, A new criterion for linear 2-port stability using a single geometrically
derived parameter, IEEE Trans. MTT, Vol. 40., No. 12, December 1992, p2303.
276 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS
p
B2 B22 4|C2 |2
ml = 2C2

B2 = 1 + |S22 |2 |D|2 |S11 |2 (8.116)

C2 = S22 DS11

The proper choice of sign in Equations 8.115 and 8.116 is determined by the requirement
that | ms | < 1 and | ml | < 1. In other words, the correct solution in each case is the
one with a magnitude less than 1. Equations 8.115 and 8.116 will yield valid solutions
whenever K > 1. Since this is only a necessary condition for stability, it is possible to find
a simultaneous conjugate match solution for some potentially unstable 2-ports (those with
K > 1).
Suppose ms and ml are known for a particular 2-port (with K > 1). Then the proce-
dure for designing an amplifier which maximizes the power delivered to the load consists of
designing matching networks that transform the actual source and load reflection coecients
into ms and ml , as in Figure 8.14.

ZS
Lossless [S]Zo Lossless
Vs Matching Matching ZL
Network K>1 Network

S S ms ms ml ml L L

Figure 8.14: An amplifier that is simultaneously matched to source and load terminations.
A conjugate match exists at all ports.

If a 2-port does not satisfy the conditions for unconditional stability, then it is said
to be potentially unstable. Useful and stable amplifiers can be designed with potentially
unstable 2-ports, however it is necessary to carefully choose source and load terminations
to avoid the unstable regions of the S - and L - planes. If K > 1, then one such choice of
terminations is S = ms and L = ml . If K 1, then it will not be possible to achieve
a simultaneous conjugate match, and the terminations will have to be chosen according to
a carefully measured tradeo between transducer gain and relative stability. A 2-port with
K 1 can be modified using resistive loading to raise the stability factor so that the loaded
2-port can be simultaneously conjugate-matched.
Of course, potentially unstable 2-ports can also be used to design oscillators. For exam-
ple, a basic oscillator design could be accomplished by choosing a S in the unstable region,
i.e., a S that maps to | out | > 1. Using this source termination, the output impedance of
the 2-port will have a negative real part. The load termination is then chosen such that the
negative resistance criterion for oscillation is satisfied at the output port. As shown earlier,
this will automatically cause the negative resistance criterion to be satisfied at the input
port.
8.3. APPLICATIONS OF SCATTERING PARAMETERS 277

8.3.6 Power Gains

1. Operating Power Gain (or just power gain)

|S21 |2 (1 | L |2 )
G Pout
= (1 |S11 |2 )+| 2 2 |D|2 ) 2Re(
(8.117)
Pin L | (|S22 | L C2 )

with
D = S11 S22 S12 S21

C2 = S22 DS11
Note that G depends only on the load termination, L.

2. Transducer Power Gain

|S21 |2 (1 | S |2 )(1 | L |2 )
GT Pout
= |(1 S11 S )(1 S22 L ) S12 S21 L 2
(8.118)
Pavs S|

Notice that GT depends on both terminations, S and L.

3. Available Power Gain

|S21 |2 (1 | S |2 )
GA Pavo
= (1 |S22 |2 )+| 2 2 |D|2 ) 2Re(
(8.119)
Pavs S | (|S11 | S C1 )

with
D = S11 S22 S12 S21

C1 = S11 DS22

Note that GA depends solely on the source impedance. It tells us how much power is
potentially available from the output of the 2-port.

4. Unilateral Transducer Power Gain

If the internal feedback within the device is small (S12 0), then the device can be
considered unilateral. This approximation yields a particularly simple form for the
transducer gain:
GT U = GT |S12 =0
2 2
(8.120)
GT U = |11 S|11S |S |2 |S21 |2 |11 S|22L |L |2

5. Maximum Available Gain

A 2-port that is conjugately matched at both ports (as in Figure 8.14) will have
G = GA = GT = GA,max where:

S21 p
GA,max = [K K 2 1] (8.121)
S12

The upper sign applies when B1 < 0 and the lower sign applies when B1 > 0. The
maximum available gain could also be calculated by using ml in the formula for G,
or ms in the formula for GA , or ( ms , ml ) in the formula for GT .
278 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

6. Maximum Stable Gain

A 2-port with K < 1 cannot be simultaneously conjugate-matched at both ports
using passive terminations. The stability factor of such a 2-port can be increased
using resistive loading at the input and/or output. If the resistive loading is chosen to
raise the stability factor of the loaded 2-port to a value slightly larger than 1, then the
loaded 2-port can be simultaneously conjugate-matched at both ports and will have
GA,max ' | SS21
12
|. Hence, for a 2-port with K < 1 we define the maximum stable gain
to be:
S21
GM S = (8.122)
S12

8.3.7 Example - Mismatch factor in terms of S and L

Consider a load with reflection coecient L connected directly to a source with reflection
coecient S . We can obtain an expression for the mismatch factor, M F , by considering the
load to be coupled to the source through a 2-port consisting of zero-length wires connecting
the input and output terminals. Such a 2-port has the following S-parameter matrix:

0 1
[S]Zo = (8.123)
1 0
Using this S-matrix in the transducer gain expression (equation 8.118) yields the expression
for the mismatch factor:
(1 | S |2 )(1 | L |2 )
MF = 2
(8.124)
|1 L S|

8.3.8 Example - Power transfer to an antenna through a lossy

transmission line
A lossy transmission line can be modeled as a 2-port with the following S-parameter matrix:

0 e l
[S]Zo = l (8.125)
e 0
where the reference impedance for the S-parameter matrix, Zo , is assumed to be equal to
the characteristic impedance of the transmission line. The length of the transmission line is
denoted by l and the complex propagation constant on the line is = + j .
Suppose that a lossy transmission line (characteristic impedance Zo ) is used to connect a
source with impedance Zo to an antenna with impedance ZL (and corresponding reflection
coecient L ). Consider three cases:
1. The transmitter is connected to the input of the transmission line and the antenna is
connected to the output of the line. No matching networks are used. In this case the
power delivered to the load is:
2l 2
PL = Pavs GT = Pavs e (1 | L| ) (8.126)

2. The transmitter is coupled to the input of the line through a lossless matching network
whereas the antenna is connected directly to the output of the lossy line. In this case
the power delivered to the load is
2
2l 1 | L|
PL = Pavs G = Pavs e 4l | 2
(8.127)
1 e L|
8.3. APPLICATIONS OF SCATTERING PARAMETERS 279

3. The transmitter is connected directly to the input of the line, and a lossless matching
network is used in between the output of the lossy line and the antenna. In this case
the power delivered to the load is
2l
PL = Pavs GA = Pavs e (8.128)

Notice that the power delivered to the antenna will be largest in case 3, i.e. when the
antenna impedance is transformed to Zo with a matching network located at the antenna3 .
In the limit ! 0, the line becomes lossless and cases 2 and 3 converge to the same
result (PL = Pavs ). This is consistent with the fact that a passive lossless 2-port that is
conjugately matched at one port is automatically matched at the other port. Therefore, in
the lossless line case a single matching network at either the source or the load end of the
line is sucient to provide optimum power transfer to the load.

3 The factor e 2l is the loss when the load impedance is equal to the characteristic impedance of the

line. This is specified on datasheets for transmission lines in units of dB per unit length, i.e. the datasheet
might specify the quantity 10 log(e2l ) in dB for l = 1 m or l = 100 feet or some other standard length.
280 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

8.4 Summary of useful S-parameter formulas

I1 I2
+ +
a1 a2
V1 2-port V
b1 b2 2
- -
I1 I2
Relationship between impedance ZX (X can be in, out, S, L) and reflection coecient X :
ZX Zo 1+ X
X = ZX = Zo
ZX + Zo 1 X
Input reflection coecient with arbitrary ZL :
S12 S21 L
in = S11 +
1 S22 L
Output reflection coecient with arbitrary ZS :
S12 S21 S
out = S22 +
1 S11 S

Voltage gain with arbitrary ZL (voltage gain does not depend on ZS ):

V2 S21 (1 + L )
AV = =
V1 (1 + S11 )(1 S22 L ) + S12 S21 L

Current gain with arbitrary ZL (current gain does not depend on ZS ):

I2 S21 ( L 1)
AI = =
I1 (1 S11 )(1 S22 L) S12 S21 L

Definitions of some commonly used quantities:

D = S11 S22 S12 S21

1 |S11 |2 |S22 |2 +|D|2

K = 2|S12 S21 |

B1 = 1 + |S11 |2 |D|2 |S22 |2

C1 = S11
DS22

B2 = 1 + |S22 |2 |D|2 |S11 |2

C2 = S22
DS11
Operating Power Gain:
Pout |S21 |2 1 | L|
2
G =
Pin (1 |S11 |2 ) + | L|
2 (|S
22 |
2 |D|2 ) 2Re( L C2 )

Available Power Gain:

Pavo |S21 |2 1 | S|
2
GA =
Pavs (1 |S22 |2 ) +| S |2 (|S 11 |2 |D|2 ) 2Re( S C1 )

Transducer Power Gain:

Pout |S21 |2 1 | S|
2 1 | L|
2
GT =
Pavs |(1 S11 S )(1 S22 L) S12 S21 L S|
2

A 2-port is unconditionally stable if:

K > 1 and 1 |S11 |2 > |S12 S21 |
8.4. SUMMARY OF USEFUL S-PARAMETER FORMULAS 281

or, if:
K > 1 and 1 |S22 |2 > |S12 S21 |
or, if:
K > 1 and B1 > 0
or, if:
K > 1 and B2 > 0
or, if:
K > 1 and |D| < 1
or, if:
1 |S11 |2
ES = D| + |S S |
>1
|S22 S11 12 21
or, if:
1 |S22 |2
0ES = D| + |S S |
>1
|S11 S22 12 21
Stability circles in the L plane:

S22 D S11 |S12 S21 |
Center is at : CL = 2
radius : rL =
|S22 | |D|2 ||S22 |2 |D|2 |
Stability circles in the S plane:

S11 D S22 |S12 S21 |
Center is at : CS = 2
radius : rS =
|S11 | |D|2 ||S11 |2 |D|2 |
Source and load reflection coecient for simultaneous conjugate match (K must be > 1) - in each case,
choose the sign that results in a reflection coecient with magnitude less than 1:
q q
B1 B12 4|C1 |2 B2 B22 4|C2 |2
ms = ml =
2C1 2C2
Maximum available power gain (only defined for K > 1):
S21 p
GA,max = [K K 2 1]
S12
GA,max is defined only for 2-ports that can be conjugately matched at both ports (K > 1). For uncondi-
tionally stable 2-ports, B1 > 0, use the lower (negative) sign. For potentially unstable 2-ports, B1 0, use
the upper (positive) sign.
282 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

8.5 References
1. Carson, Ralph S., High Frequency Amplifiers, John Wiley & Sons, New York, 1975.
2. Gonzalez, Guillermo, Microwave Transistor Amplifiers: Analysis and Design, Prentice-
Hall, New Jersey, 1984.
3. Liao, Samuel Y., Microwave Circuit Analysis and Amplifier Design, Prentice-Hall,
New Jersey, 1987.
4. Pozar, David M., Microwave Engineering, Addison-Wesley Publishing Company, 1990.

5. Vendelin, George D., Design of Amplifiers and Oscillators by the S-parameter Method,
John Wiley & Sons, New York, 1982.
6. Vendelin, George D., Anthony M. Pavio, Ulrich L. Rohde, Microwave Circuit Design
Using Linear and Nonlinear Techniques, John Wiley & Sons, 1990.
8.6. HOMEWORK PROBLEMS 283

8.6 Homework Problems

1. Find the S-parameters for the T-network in Figure 8.15. Port 1 is on the left. Denote
the reference impedance by Zo .

Z1 Z2
Z3

Figure 8.15: T-network

2. Find expressions for the S-parameters of the 2-port in Figure 8.16. Denote the refer-
ence impedance by Zo .

+
V Z1 gm V Z2
-

Figure 8.16: Unilateral hybrid-Pi model.

3. Consider an ideal transformer as shown in Figure 8.17. This 2-port is characterized

1:N
I1 I2
+ +

V1 V2
- -

Figure 8.17: Ideal transformer

by the following relations:

V2 = N V1 (8.129)
I2 = I1 /N

Find expressions for the S-parameters of this 2-port. Denote the reference impedance
by Zo .
4. Suppose that two 2-ports are cascaded. Show that the S21 parameter for the overall
network is given by
284 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

(1) (2)
S21 S21
S21 = (2) (1)
(8.130)
1 S11 S22
(1) (2)
where Sij and Sij refer to the S-parameters of the first and second 2-ports, respec-
tively.
5. A 2-port has the following S-parameters (Z0 = 50 ):
S11 = 0.5\ 96 (8.131)
S12 = 0.3\50
S21 = 5.2\45
S22 = 0.4\ 120
(a) What is the input impedance when the 2-port is terminated with a short circuit,
i.e., ZL = 0?
(b) Is this 2-port unconditionally stable? (Check one of the necessary and sucient
sets of criteria.)
(c) Find the coordinates of the center and the radius of the stability circles in the S
and L planes. Sketch the stability circles and shade the regions that correspond
to those values of S and L that make | out | > 1 or | in | > 1.

6. Suppose that S12 in Problem 5 is changed to S12 = 0.05\50 . Find the coordinates
of the center and the radius of the stability circles in the S and L planes. Sketch
the stability circles and shade the regions that correspond to those values of S and
L that make | out | > 1 or | in | > 1. Is the 2-port unconditionally stable?

7. The stability circles in the output ( L ) plane for cases (a) and (b) are shown in Figure
8.18. Suppose that in both cases it is known that |S11 | < 1. For each case indicate

L Plane L Plane
Stability Circle
Stability Circle

Unit circle (| L| = 1) Unit circle (| L| = 1)

Figure 8.18: Case (a), left and case (b), right.

what region of the L plane corresponds to the stable region. The stable region of
the L plane corresponds to those values of L that will cause the input reflection
coecient to have a magnitude less than 1.
8.6. HOMEWORK PROBLEMS 285

8. For a particular 2-port with |S11 | < 1 and |S22 | < 1, the center and radius of the
stability circles in the S and L planes are known to be

CL = 0.269 j1.86 (8.132)

rL = 3.01

CS = 2.23 + j4.76
rS = 4.19

(a) Make two sketches, one each for the S and L planes, and show the stability
circles. Shade the unstable region.
(b) Is the 2-port unconditionally stable?

9. Consider the system in Figure 8.19 where RS = 300 and RL = 100 . Suppose that
the 2-port is known to be unilateral and that S22 = 0.8 (Zo = 50 ). The transducer
gain of the 2-port is GT = 12 dB. Find the available gain of the 2-port in this system.

RS
+ +
VS Vi 2-port Vo RL
- -

Figure 8.19: System with resistive source and load.

10. Consider the system in Figure 8.19 where RS = RL = 300 . The voltage gain,
Av = VVoi , in this system is found to be Av = 6\45 . In addition, it is known that the
2-port is unilateral, and that S11 = 0.2 (Zo = 50 ).

(a) Find the operating power gain. Express your result in dB.
(b) Find the transducer powergain. Express your result in dB.

11. A 2-port has the following S-parameters (Z0 = 50 ):

S11 = 0.6\ 100 (8.133)

S12 = 0.05\45
S21 = 3.50\60
S22 = 0.1\ 30

(a) Is this 2-port unconditionally stable?

(b) Can this 2-port be conjugately matched simultaneously at the input and output?
If so, find the source and load impedances (Zms and Zml ) that will make this
condition occur.
286 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

(c) If this 2-port is used in a system with the source and load impedances found in
part 11b, find the operating, transducer, and available gains (G, GT , and GA ).
Express your results in dB.

12. The 2-port from Problem 11 is used in a system with open circuit source voltage
VS = 3V, source impedance RS = 100, and load impedance 500 (see Figure 8.19).

(a) Compute the power available from the source, Pavs . Express your result in dBm.
(b) Find G, GT , and GA for the 2-port when used in this system. Express your
results in dB
(c) Find the power delivered to the 2-port and the power delivered to the load.
Express your result in dBm.
(d) How much power will be delivered to the load if a lossless matching network
is added between the output of the 2-port and the 500 load? The matching
network would be designed to transform 500 into Zout where Zout is the output
impedance of the 2-port with the 100 source termination. Express your result
in dBm.

13. Consider the 2-port from Problem 11. Suppose the reference impedance, Zo , is changed
to 500. Denote the new S-parameters for this reference impedance by S11 0
, S21
0
, S12
0

and S22 .
0

(a) Find S11

0
.
(b) Find S21
0
. Hint: The voltage gain Av = Vo /Vi in a system with arbitrary load
and source impedance is given by the following formula:

S21 (1 + L )
Av = (8.134)
(1 S22 L )(1 + IN )

where IN is the input reflection coecient and Vo , Vi are the voltages measured
across the output and input terminals of the 2-port, respectively.

14. A 2-port has 50 S-parameters:

S11 = 0.3 (8.135)

S12 = 0.01
S21 = 10.0
S22 = 0.1

The Rollett Stability Factor for this 2-port is K = 4.525, and the 2-port is uncondi-
tionally stable.

(a) The 2-port is used with a source having impedance Zs = 300 and a load
ZL = 50 . The power available from the source is -3 dBm. Find the power
delivered to the load. Express your result in dBm.
(b) Now suppose a lossless matching network is used between the source and the
2-port. Find the power delivered to the load. Express your result in dBm.
8.6. HOMEWORK PROBLEMS 287

(c) Suppose a lossless matching network is used between the 2-port and the load (no
matching network at the input). Find the power delivered to the load. Express
your result in dBm.
(d) Suppose both input and output matching networks are used so that both ports
are conjugately matched. Find the power delivered to the load. Express your
result in dBm.

15. Show that the maximum available gain GA,max for a unilateral 2-port is

|S21 |2
GA,max = (8.136)
(1 |S11 |2 )(1 |S22 |2 )

Hint: When the input and output of the 2-port are conjugately matched, G = GT =
GA,max .

16. Consider a system with source and load reflection coecients S and L as shown in
Figure 8.20. Define the insertion gain

ZS
VS 2-port ZL

S L

Figure 8.20: System with source and load reflection coecients S and L

Pdelivered to load with network

GI = (8.137)
Pdelivered to load without network

where Pdelivered to load without network is the power delivered to the load when the source
is connected directly to the load, and Pdelivered to load with network is the power delivered
to the load when the 2-port is used between the source and the load.

(a) Find an expression for the insertion gain of an arbitrary 2-port.

(b) Find an expression for the insertion gain of a lossless matching network.

17. The T-parameters (sometimes called the chain scattering parameters) are defined
in terms of the same variables as the S-parameters. They are useful when cascading
2-ports, because the T-parameter matrix for a cascade is simply the product of the
T-matrices for the individual 2-ports. (Can you show that this is true?) The definition
of the T-parameters follows from choosing a1 and b1 to be the dependent variables,
i.e.,

a1 = T11 b2 + T12 a2 (8.138)

b1 = T21 b2 + T22 a2
288 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

where the as and bs are defined as

Vi + Z o I i
ai = p (8.139)
2 Zo
Vi Z o I i
bi = p . (8.140)
2 Zo
Suppose the S-parameters for a 2-port are known. Find the T-parameters in terms of
the S-parameters.
18. Consider a cascade of two identical 2-ports. The cascade is used in a system with
source and load as shown in Figure 8.21:

50

[S] [S] 1000

Pavs = 10dBm

Figure 8.21: Cascade of two identical 2-ports

The available source power is -10dBm and the 50 S-parameters of the 2-ports are

S11 = 0.35 (8.141)

S12 = 0.1
S21 = 3.0
S22 = 0.50

The 2-ports are unconditionally stable. Answer the following questions. For parts 18a
- 18g, express all results in dB or dBm, whichever is appropriate.
(a) The power delivered to the first 2-port.
(b) The power delivered to the second 2-port.
(c) The power delivered to the load.
(d) The transducer gain for the cascaded 2-ports.
(e) The available gain of the cascaded 2-ports.
(f) The operating power gain of the cascaded 2-ports.
(g) Suppose three lossless matching networks are used between the source and the
first 2-port, the first and second 2-ports, and the second 2-port and the load.
Find the power delivered to the load.
(h) Find S11 (Zo = 50) for the cascaded 2-ports. The answer is NOT 0.35!
19. Consider the system shown in Figure 8.22.
The 2-port has the following S-parameters (Zo = 50 ):
8.6. HOMEWORK PROBLEMS 289

50 Amplifier

[S] Lossless
Zo = 50
Pavs = 0dBm
Z

Figure 8.22: System with adjustable, lossless matching network between the 2-port and the
load.

S11 = 0.2, S21 = 5.0

S12 = 0.1, S22 = 0.5
(a) Find the impedance Z that will maximize the power delivered to the load, ZL .
(b) Find the power that will be delivered to the load, assuming that the matching
network has been designed to present the Z found in part 19a to the output of
the 2-port. Express your result in dBm.
(c) Find the impedance Z that will maximize the power delivered to the input of the
2-port.
(d) Find the power that will be delivered to the load, assuming that the matching
network has been designed to present the Z found in part 19c to the output of
the 2-port. Express your result in dBm.

20. In this problem, we prove that a particular set of passive terminations that cause the
negative resistance criterion for oscillation to be satisfied at one port will automatically
cause the conditions for oscillation to be satisfied at the other port if the 2 port is
bilateral (i.e. if S12 6= 0 and S21 6= 0).

(a) Show that the negative resistance criterion for steady-state oscillation (Z1 + Z2 =
0) can be written in terms of the corresponding reflection coecients as 1 2 = 1.
(b) Now, suppose that L plane stability circle analysis shows that at least part of
unstable region lies within the unit circle in the L plane. We choose a load
reflection coecient within the unstable region and also inside of the unit circle.
Denote this load termination by Lu . Find the reflection coecient of the passive
source termination, Su , that will cause the conditions for steady state oscillation
to be satisfied at the input port. Write your answer in terms of the S parameters
and Lu .
(c) Now, show that the conditions for oscillation are satisfied at the output port
when the 2-port is terminated with Lu and Su .

21. A 2-port has S parameters ( Zo = 50 ): S11 = 0.2, S12 = 0.0, S21 = 2.0, S22 = 0.8

(a) Suppose that the 2-port is used in a system with source impedance ZS = 50
and load impedance ZL = 50. The power available from the source Pavs = 0
dBm. Find the power that will be delivered to the load. Express your answer in
dBm.
290 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

(b) Calculate the input impedance of the 2-port when it is used in the system de-
scribed in part a.
(c) This 2 port is unconditionally stable. Find the power delivered to the load when
lossless matching networks are used at the input and the output of the 2 port so
that the 2 port is simultaneously matched at both ports. Express your answer in
dBm.
(d) Find the power delivered to the load if 2 of these 2 ports are cascaded and the
cascade is used in between the source and load specified in part a. Express your
answer in dBm.

22. A 2 port has the following Z parameters (all given in ):

Z11 = 20 Z22 = 300 Z12 = 0.0 Z21 = 1000

Find S11 and S21 (Zo = 50 ). Hint: the input impedance and voltage gain of a 2
port can be written in terms of Z parameters and the load impedance as follows:
Z12 Z21
ZIN = Z11
ZL + Z22
V2 Z21 ZL
Av = = .
V1 Z11 ZL + Z11 Z22 Z12 Z21
23. In a particular system it is found that the operating, transducer, and available gains
of a 2-port are G = 10 dB, GT = 8 dB, GA = 14 dB. The 2-port is unilateral and is
unconditionally stable. The source impedance ZS = 100 and the power available
from the source is 3 dBm.

(a) Find the power delivered to the load. Express your result in Watts (NOT in
dBm! I want to see that you know the relationship between power in Watts and
dBm.)
(b) Find the power delivered to the 2-port by the source. Express your result in
dBm.
(c) Find the power that would be delivered to the load if a single lossless matching
network is used between the source and the 2-port. Express your result in dBm.
(d) Find the power that would be delivered to the load if a single lossless matching
network is used between the 2-port and the load. Express your result in dBm.
(e) Find the power that would be delivered to the load if lossless matching networks
are used at both the input and the output of the 2-port. Express your result in
dBm.
(f) Suppose it is known that the 2-port has S22 = 0 (Zo = 100 ) and calculate
the power that would be delivered to the load if a cascade of 2 of these identical
2-ports is used between the source and the load. Express your result in dBm.

24. In a system with Pavs = 0 dBm, ZS = 100 and unknown, but constant, ZL , it
is empirically determined that the operating, transducer, and available gains of a
particular 2-port are G = 12 dB, GT = 8 dB, GA = 14 dB when this single 2-
port is used to couple the source to the load. The 2-port is known to be unilateral,
unconditionally stable, and to have S22 = 0 (Zo = 100 ).
8.6. HOMEWORK PROBLEMS 291

(a) Calculate the power that would be delivered to the load if a cascade of 2 of these
identical 2-ports is used between the source and load described above. Express
your result in dBm.
(b) Suppose that a lossless matching network is added between the load and the
output of the second 2-port in the system described in part a. Determine the
power delivered to the load. Express your result in dBm.

25. Many amplifiers are designed to be simultaneously conjugate-matched at both ports

when used between 50 source and load impedances. Suppose that you have obtained
such an amplifier, and that the specifications for the amplifier state that the power
gain of the unit is 20 dB in a 50 system. Furthermore, the specifications state that
the reverse isolation of the amplifier is 30 dB in a 50 system. Reverse isolation is
the power attenuation of the amplifier when the amplifier is driven at the output port
and the load is connected to the input port. (Reverse isolation of 30 dB means that
the reverse power gain of the amplifier is -30 dB.)

(a) What are S11 and S22 (Zo = 50 ) for the amplifier?
(b) Suppose that the amplifier is used between a 50 source and a 1000 Ohm load.
What is the transducer gain of the amplifier in this system? Express your result
in dB.
(c) What is the largest input reflection coecient magnitude, | IN |, that could ever
be seen at the input of the amplifier, assuming that the amplifier is terminated
with a passive load?
(d) Suppose that the amplifier is used between a 200 source and a 200 load.
You do not have enough information to determine the exact transducer gain in
this system, however you do have enough information to determine upper and
lower limits for what the transducer gain could be. Specify the range of possible
transducer gains (in dB).

26. Consider a 2-port consisting of a passive, lossless ladder network that matches a 50
source to the load ZL = 10 j200 . Answer the following questions. Note carefully
that you can answer all parts of this question without actually designing the matching
network.

(a) Find |S21 | (Zo = 50 ) for the 2-port.

(b) Find |S11 | (Zo = 50 ) for the 2-port.
(c) Find S22 (Zo = 50 ) for the 2-port. Find the magnitude and phase angle.

27. A unilateral amplifier is simultaneously conjugate-matched at both ports and has a

transducer gain of 12 dB when used in a system with ZS = ZL = 300 .

(a) What is the transducer gain (in dB) when this amplifier is used in a system with
ZS = ZL = 50 ?
(b) What is the available gain (in dB) when this amplifier is used in a system with
ZS = 50 ?

28. Consider the unilateral hybrid-pi model shown in Figure 8.23.

292 CHAPTER 8. 2-PORT SCATTERING (S) PARAMETERS

+
V Z1 gm V Z2
-

Figure 8.23:

(a) What constraints must be satisfied by Z1 and Z2 if the 2-port is to be uncondi-

tionally stable?
(b) Find an expression for the operating power gain of the unilateral hybrid-pi model.
Express your result in terms of gm , Z1 , Z2 , and the load impedance ZL . It is not
necessary to find S-parameters (or any other 2-port parameter set) to work this
problem. Start with the definition of operating power gain.
(c) Find an expression for the maximum available gain GA,max for the unilateral
hybrid-pi model. Express your result in terms of gm , Z1 , and Z2 .

29. Many amplifiers are designed to be simultaneously conjugate-matched at both ports

when used between 50 source and load impedances. Suppose that you have obtained
such an amplifier, and that the specifications for the amplifier state that the power
gain of the unit is 20 dB in a 50 system. Furthermore, the specifications state that
the reverse isolation of the amplifier is 23 dB in a 50 system. Reverse isolation is
the power attenuation of the amplifier when the amplifier is driven at the output port
and the load is connected to the input port. (Reverse isolation of 23 dB means that
the reverse power gain of the amplifier is -23 dB.)

(a) What are S11 and S22 (Zo = 50 ) for the amplifier?
(b) Is this amplifier unconditionally stable? Justify your answer.
(c) Suppose that the amplifier is used between a 400 source and a 50 load. The
power available from the source is Pavs = 10 dBm. Find the power delivered
to the load. Express your result in dBm.
(d) Suppose that the system of part b. is modified by adding a lossless impedance
matching network between the output of the 2-port and the 50 load. Find the
power delivered to the load. Express your result in dBm.
(e) Suppose that two of these amplifiers are cascaded, and used between a 400
source with Pavs = 10 dBm and a 50 load. Find the power delivered to the
load. Express your result in dBm.
Chapter 9

Filter Design

This chapter discusses the implementation of filter networks with passive, lossless compo-
nents. In particular, we will concentrate on the design of filters based on ladder networks
comprised of lossless inductors and capacitors. A discussion of general properties of lossless
filters in terms of the scattering parameters of the filter network is followed by a descrip-
tion of some useful functions which approximate the ideal rectangular lowpass response and
are often used as target filter response functions when designing practical filters. Then we
will look at how to design lowpass filters with a prescribed frequency dependence for the
transducer power gain function, GT (!). Finally, we will discuss two methods for trans-
forming an existing lowpass filter design into a bandpass filter. The first method is based
on a straightforward replacement of the series inductors and shunt capacitors in a lowpass
filter with series and parallel resonators, respectively. The second approach involves replac-
ing all of the lowpass filter elements with resonators of the same type which results in a
coupled-resonator filter.
We assume that the filter network is passive, linear, and lossless, is driven with a source
having real impedance Zo , and is terminated with real load impedance Zo , as shown in
Figure 9.1. Since S = L = 0 in this system, the transducer, operating, and available

Figure 9.1: A passive, linear, lossless 2 port in a system with source and load impedances
equal to Zo .

power gains can then be written as follows:

Pout
GT = = |S21 |2 (9.1)
Pavs

293
294 CHAPTER 9. FILTER DESIGN

Pout |S21 |2
G= =1= (9.2)
Pin 1 |S11 |2

Pavo |S21 |2
GA = =1= (9.3)
Pavs 1 |S22 |2
In equations 9.2 and 9.3 we have used the fact that the filter network is lossless, which
means that the operating and available power gains will be equal to one. From equations
9.2 and 9.3:
|S21 |2 = 1 |S11 |2 = 1 |S22 |2 . (9.4)

Figure 9.2: The transducer power gain, |S21 |2 , and input/output reflection coecients for
a second order Butterworth filter implemented using a passive lossless network.

Equation 9.4 shows that the transducer gain frequency response function is related to the
squared magnitude of the input and output reflection coecients of the lossless two-port.
Figure 9.2 shows the transducer gain and and squared magnitude of the input coecient
reflection for a lowpass filter. Notice that the filters attenuation at high frequencies happens
because the input reflection coecient approaches 1, i.e., the attenuation at high frequencies
occurs because the source is unable to deliver much power to the network and not because of
any signal absorption within the network itself. The relationship between transducer power
gain and the squared magnitude of the input reflection coecient means that the design of
lossless filters can be accomplished by designing a network to realize the target transducer
power gain (|S21 |2 ) function or by designing the network to realize the corresponding input
or output power reflection coecient function (|S11 |2 or |S22 |2 ).
Recall that S11 is related to the input impedance of the network through

Zin Zo
S11 = |Z =Z . (9.5)
Zin + Zo L o
9.1. BUTTERWORTH, CHEBYSHEV, BESSEL-THOMPSON FILTERS 295

Notice that a filter with prescribed transfer gain function specified by |S21 (j!)|2 can be
realized by synthesizing the appropriate transducer voltage gain function, S21 (j!), or equiv-
alently, by synthesizing an input impedance function, Zin (j!), that produces |S11 (j!)|2 =
1 |S21 (j!)|2 . Both approaches will be illustrated through an example, but first we shall
examine some common functions that are used for the target transducer power gain function.

9.1 Butterworth, Chebyshev, Bessel-Thompson Filters

9.1.1 Butterworth
The Butterworth response function with cuto frequency !c has the following form:
1
|S21 (j!)|2 = (9.6)
1 + ( !!c )2n

The -3 dB frequency occurs at ! = !c and the shape of the function is defined by the
parameter, n, which is always an integer. The parameter n is called the order of the filter
because, as we shall see, it is the highest power in the denominator polynomial of the
transducer voltage transfer function, S21 (j!). The Butterworth transducer gain functions
for !c = 1 and orders n = 1 through n = 6 are plotted in Figure 9.3. The tranducer gain
falls o more rapidly for larger filter order n.
The Butterworth filter response approaches the ideal rectangular lowpass filter response
function as the order, n, increases. The Butterworth response is called maximally flat
because all derivatives up through order 2n 1 are equal to zero at ! = 0 and as ! ! 1.
The Butterworth approximation to the ideal rectangular response function is used when
flatness of the transfer function magnitude within the passband is the highest priority for a
particular application.

9.1.2 Chebyshev
The Chebyshev response function has the following form:
1
|S21 (j!)|2 = (9.7)
1 + 2 Cn2 ( !!c )

where is a constant called the ripple parameter, and Cn (!) is a Chebyshev polynomial
defined by:
n n(n 3) n 4 n(n 4)(n 5) n 6
Cn (!) = 2n 1
[! n 1!22 !
n 2
+ 2!24 ! 3!24 ! +
(9.8)
n(n 5)(n 6)(n 7) n 8 n(n 6)(n 7)(n 8)(n 9) n 10
4!28 ! 5!210 ! ...]
The summation is stopped when the exponents of ! are no longer positive. The Cheby-
shev polynomials for n=1 through 5 are given in Table 9.1.
Two of the Chebyshev polynomials (n=4 and n=5) are plotted in Figure 9.4.
The Chebyshev response function is plotted for !c = 1, 2 = 0.0233 and n=1 through
n=4 in Figure 9.5, for 2 = 0.259 and n=1 through n=4 in Figure 9.6, and for 2 = 0.585
and n=1 through n=4 in Figure 9.7.
As the figures show, the ripple parameter, , controls the amount of amplitude ripple
within the filters passband. In general, for a particular filter order, n, a larger value for
296 CHAPTER 9. FILTER DESIGN

Butterworth
2.0

0.0

GT
2.0
(dB)

4.0

6.0
0.1 1.0 10.0
!/!c

Butterworth
10.0
5.0
0.0
5.0
10.0
15.0
GT
20.0
(dB)
25.0
30.0
35.0
40.0
45.0
50.0
0.1 1.0 10.0 100.0
!/!c

Figure 9.3: Butterworth Response function for n=1, 2, 3, 4. 5, and 6. The parameters are
the same for both figures, but the lower figure shows a wider range of frequencies and gains.
The higher filter orders correspond to faster descent into the stopband.
9.1. BUTTERWORTH, CHEBYSHEV, BESSEL-THOMPSON FILTERS 297

n Cn (x)
1 x
2 2x2 1
3 4x3 3x
4 8x 4
8x2 + 1
5 16x 5
20x3 + 5x
6 32x 6
48x4 + 18x2 1
7 64x 7
112x5 + 56x3 7x

Table 9.1: Chebyshev polynomials for n=1 through 5.

4.0

3.0

Cn2 (x) 2.0

1.0

0.0
0.0 0.5 1.0 1.5
x

Figure 9.4: Cn2 (x) for n = 4 and n = 5. Within the passband (x<1) the functions oscillate
between 0 and 1. The number of extrema within the passband is equal to the filter order n.
Notice that the even order polynomial (n = 4) is equal to one at x = 0. This corresponds
to finite attenuation at ! = 0 and will not be realizable using a lossless network with equal
source and load resistances.
298 CHAPTER 9. FILTER DESIGN

Chebyshev 0.1 dB ripple

2.0

0.0

GT
2.0
(dB)

4.0

6.0
0.1 1.0 10.0
!/!c

Chebyshev 0.1 dB ripple

10.0

0.0

10.0
GT
20.0
(dB)

30.0

40.0

50.0
0.1 1.0 10.0 100.0
!/!c

Figure 9.5: Chebyshev response; 2 = 0.0233 (0.1 dB ripple), n=1, 2, 3 and 4. The lower
figure shows a wider range of frequencies and gains.
9.1. BUTTERWORTH, CHEBYSHEV, BESSEL-THOMPSON FILTERS 299

Chebyshev 1.0 dB ripple

2.0

0.0

GT
2.0
(dB)

4.0

6.0
0.1 1.0 10.0
!/!c

Chebyshev 1.0 dB ripple

10.0

0.0

10.0
GT
20.0
(dB)

30.0

40.0

50.0
0.1 1.0 10.0 100.0
!/!c

Figure 9.6: Chebyshev response; 2 = 0.259 (1.0 dB ripple), n=1, 2, 3 and 4. The lower
figure shows a wider range of frequencies and gains.
300 CHAPTER 9. FILTER DESIGN

Chebyshev 2.0 dB ripple

2.0

0.0

GT
2.0
(dB)

4.0

6.0
0.1 1.0 10.0
!/!c

Chebyshev 2.0 dB ripple

10.0

0.0

10.0
GT
20.0
(dB)

30.0

40.0

50.0
0.1 1.0 10.0 100.0
!/!c

Figure 9.7: Chebyshev response; 2 = 0.585 (2.0 dB ripple), n=1, 2, 3 and 4. The lower
figure shows a wider range of frequencies and gains.
9.1. BUTTERWORTH, CHEBYSHEV, BESSEL-THOMPSON FILTERS 301

the ripple parameter results in a quicker roll-o into the stopband at the expense of larger
ripple within the passband. The Chebyshev polynomials Cn (!) oscillate between -1 and
1 for |!| < 1, so the transducer power gain oscillates between 1 and 1+ 1
2 . The ratio of

the maximum and minimum responses within the passband is therefore 1 + 2 . The ripple
amplitude expressed in dB is 10 log(1 + 2 ). In practice, the ripple parameter is usually
specified by giving the ripple amplitude in dB. Thus, a Chebyshev filter with 0.5 dB ripple
corresponds to 2 = 100.5/10 1, or = .349, approximately.
The cuto frequency defined in equation 9.7 is not the -3 dB frequency. Instead, it is
the frequency where the response function crosses the level corresponding to the bottom of
the passband ripple on its descent into the stopband. This can be seen clearly in Figures
9.5 through 9.7.

Figure 9.8: Comparison of third order Butterworth and Chebyshev filters with 0.1 dB and
2.0 dB ripple. The 2.0 dB ripple Chebyshev filter exhibits the fastest descent into the stop
band. The 0.1 dB ripple Chebyshev filter has a slower descent into the stop band, but it is
still noticeably faster than that of the Butterworth filter.

As the ripple parameter is decreased so that the passband ripple approaches 0 dB,
the Chebyshev response approaches a maximally-flat response characteristic and becomes
identical to the Butterworth response, provided that the cuto frequency is suitably scaled.
Figure 9.8 shows the n=3 response functions for Butterworth and for Chebyshev with 0.1 dB
and 2.0 dB ripple, respectively. In this plot, the frequency axis has been scaled so that the -3
dB frequency of all three response functions occurs at ! = 1.0. The stopband attenuation is
smallest for the Butterworth response and is largest for the Chebyshev response with 2.0 dB
ripple. This figure illustrates that, for a given filter order n, larger stopband attenuation can
be achieved if larger ripple can be tolerated within the passband. Notice also that the 0.1 dB
ripple Chebyshev filter exhibits a noticeable improvement in stopband attenuation compared
to the Butterworth response, so a significant improvement in stopband attenuation can be
302 CHAPTER 9. FILTER DESIGN

gained by tolerating a relatively small amount of passband ripple.

Finally, notice that the even order Chebyshev tranducer gain functions are not equal to
1 at ! = 0. Instead, the even order transducer gain function is equal to [1 + 2 ] 1 at ! = 0,
i.e. the filter has finite attenuation at ! = 0. It is not possible to realize this type of transfer
function with a lossless lowpass ladder network and equal source and load resistances. Since
the lowpass network filter network reduces to a direct connection between the source and
load terminations at ! = 0, in order to have finite attenuation at ! = 0 it is necessary to
have dierent source and load terminations, such that the mismatch loss between the source
and load terminations is equal to [1 + 2 ] 1 . If the source and load impedances are the same,
as assumed here, it is only possible to realize the odd order Chebyshev response functions
with a lowpass ladder network.

9.1.3 Bessel-Thompson
The Bessel-Thompson response function is desirable in some applications because it results
in a group delay function that is maximally flat in the same sense that the Butterworth
response provides maximally flat amplitude response. This results in nearly distortionless
transmission of pulse-type waveforms. Bessel filters are commonly employed in digital com-
munications and radar systems where it is necessary to employ a filter that will not smear
pulses out over long times.
The family of Bessel transfer functions can be indexed by an integer parameter, n, and
for each n the response function takes the form:
Bn (0)
S21 (s) = (9.9)
Bn (s)
where Bn (x) are the Bessel polynomials, which satisfy

Bn (x) = (2n 1)Bn 1 (x) + x2 Bn 2 (x)

with
B0 (x) = 1, B1 (x) = x + 1.
Bessel polynomials for n=1 through n=5 are tabulated in Table 9.2.

n Bn (x)
1 x+1
2 x2 + 3x + 3
3 x3 + 6x2 + 15x + 15
4 x + 10x3 + 45x2 + 105x + 105
4

5 x + 15x4 + 105x3 + 420x2 + 945x + 945

5

6 x + 21x5 + 210x4 + 1260x3 + 4725x2 + 10395x + 10395

6

7 x + 28x + 378x5 + 3150x4 + 17325x3 + 62370x2 + 62370x2 + 135135x + 135135

7 6

Table 9.2: Bessel polynomials for orders n=1 through n=4.

The Bessel response for order n can be written in the form:

1
S21 (s) = (9.10)
1 + a1 s + a2 s2 + a3 s3 + an sn
9.2. EXAMPLE: SYNTHESIS OF 4TH ORDER BUTTERWORTH FILTER 303

The group delay of a filter is defined as the negative slope of the filters phase response.
Thus, writing the transfer function S21 (j!) = A(!)ej (!) , the group delay, Tg (!), is defined
as:
d (!)
Tg (!) = (9.11)
d!
For a transfer function of the form given in equation 9.10, it can be shown that the group
delay at ! = 0 is equal to the value of the coecient a1 . If the Bessel polynomials given in
Table 9.2 are used in equation 9.9 and the resulting equation is manipulated so that it is in
the form of equation 9.10, the coecient a1 will be equal to one, corresponding to a group
delay of 1 second. To scale a Bessel filter for a specified delay, it is necessary to scale the
coecients a1 through an by multiplying the i0 th coecient by i , where is the desired
delay. Thus, a third-order Bessel filter with delay of 1 s would have coecients:
6
a1 = 10

6 2 6
a2 = (10 )
15

6 3 1
a3 = (10 )
15
The Bessel response for n=1 through n=4 is plotted in Figure 9.9. The coecients of
the Bessel polynomial have been scaled dierently for each plot in order to make the -3 dB
frequencies equal to 1 Hz in all cases. The Bessel response exhibits a gradual descent into
the stopband and, for a given filter order and -3 dB frequency, results in less attenuation in
the stopband than the Bessel or Chebyshev filters.
The group delay for the filters with n=1 through n=4 is shown in Figure 9.10. Notice
that the shape of the Bessel filters group delay curve is the same as the Butterworth filters
maximally flat gain response. The same (scaled) coecients that were used to produce
Figure 9.9 were used to produce Figure 9.10. Notice that when all of the filters are scaled
to have the same -3 dB frequency, as is the case here, the delay increases as the filter order
is increased.

9.2 Example: Synthesis of 4th order Butterworth filter

Suppose that it is necessary to design a passive LC ladder network that realizes the fourth
order Butterworth lowpass response function, i.e., the network transducer gain function
must be
1
|S21 (!)|2 = (9.12)
1 + !8
Notice that the cuto frequency of the filter has been set to 1 rad/s. Later we will discuss
how to scale the design to an arbitrary cuto frequency.
So far, only the magnitude of the transfer function has been specified. To design an actual
circuit that realizes the desired transfer function, we must determine the complex voltage
transfer function, S21 (j!). Alternatively, we could determine the required complex input
reflection coecient, S11 (j!), from which we can solve for the complex input impedance
function that the network must realize.
304 CHAPTER 9. FILTER DESIGN

Figure 9.9: Bessel response for n=1 through n=4. The higher filter orders correspond to
faster descent into the stopband.The coecients of the Bessel polynomial given in Table 9.2
have been scaled for each curve so that the -3 dB frequency occurs at 1 Hz.

Figure 9.10: Group delay for the Bessel filters with n=1 through n=4.
9.2. EXAMPLE: SYNTHESIS OF 4TH ORDER BUTTERWORTH FILTER 305

9.2.1 Filter synthesis based on the S21 function.

We must first determine the form of a realizable S21 function that will provide the desired
transducer gain response. The may be accomplished by noting that equation 9.12 can be
re-written as follows:
1
S21 (j!)S21 (j!) = S21 (j!)S21 ( j!) =
1 + !8
or, with ! = sj :
1 1
S21 (s)S21 ( s) = = (9.13)
1 + ( sj )8 1 + s8
The right hand side of equation 9.13 has 8 poles. To determine S21 (s), the right hand side
of equation 9.13 must be factored into a product of two terms such that the second term
can be obtained by replacing s with s in the first term. This amounts to deciding which
group of 4 poles must be assigned to S21 (s). It will be necessary to ensure that the poles
assigned to S21 (s) are in the left half of the s-plane. The poles must be in the left-half plane
because it is not possible to realize an S21 function that has poles in the right-half plane
using a passive network. The pole locations are given by the roots of the following equation:

s8 = 1 (9.14)

s8 = ej(2n+1) (9.15)

The solutions to equation 9.14 (or, equivalently, equation 9.15) are sk = ej(2n+1) 8 with
n = 0, 1, 2, 7. Thus, the poles are equally spaced around the unit circle centered on the
origin of the s-plane. The 4 poles that must be assigned to S21 (s) are those in the left-half
plane. The pole locations can be written as follows:

sk = ej(2n+1) 8

(2n + 1) (2n + 1)
= cos( ) + j sin( )
8 8
with the left-half plane pole locations obtained when n = 2, 3, 4, 5. S21 (s) can now be written
as follows:
1
S21 (s) = (9.16)
(s s2 )(s s3 )(s s4 )(s s5 )

1
S21 (s) = 5 7 9 11
(s ej 8 )(s ej 8 )(s ej 8 )(s ej 8 )
The denominator can be expanded to yield:
1
S21 (s) = . (9.17)
s4 + 2.61313s3 + 3.41421s2 + 2.61313s + 1
The ladder networks that can realize the transfer function with 4 poles will have 4
branches with one energy storage element per branch. The two possibilities are the lowpass
306 CHAPTER 9. FILTER DESIGN

Zo L1 L3

Vs C2 C4 Zo Vo

Figure 9.11:

Zo L2 L4

Vs C1 C3 Zo Vo

Figure 9.12:

networks shown in Figures 9.11 and 9.12. Note that in the limit as ! approaches 0, these
networks will have the property that S21 ! 1. Thus, only the upper sign in the numerator
of equations 9.16 and 9.17 is relevant.
Consider Figure 9.11. If the termination impedances are 1 , the transducer voltage
gain for this network is easily calculated. (Assume that the voltage across the load is 1 V
and work backwords toward the source to find VS . Then, S21 = 2/VS .) The result is:

1
S21 =
s4 ( L1 C22L3 C4 ) + s3 ( L1 C2 L3 +C
2
2 L 3 C4
) + s2 ( L3 C4 +L1 C4 +L
+ s( L3 +L1 +C
2 2
1 C2 +C2 L3
2 +C4
)+1 )
(9.18)
Comparing the coecients in the denominator of 9.17 with those in the denominator of 9.18
yields four equations for the four unknown parameters:

L1 C2 L3 C4
= 1 (9.19)
2
L1 C2 L3 + C2 L3 C4
= 2.61313 (9.20)
2
L3 C 4 + L1 C 4 + L1 C 2 + C 2 L3
= 3.41421 (9.21)
2
L1 + L3 + C2 + C4
= 2.61313 (9.22)
2

Equations 9.19 through 9.22 can now be solved for the unknown parameters. The solution
9.2. EXAMPLE: SYNTHESIS OF 4TH ORDER BUTTERWORTH FILTER 307

is:

L1 = 0.7654 H (9.23)
C2 = 1.8478 F
L3 = 1.8478 H
C4 = 0.7654 F

These values will give a Butterworth response with cuto frequency !c = 1 rad/s if the
terminating impedances are 1 . This filter is referred to as a lowpass prototype filter
and can be used as the basis for a filter with arbitrary cuto frequency and termination
impedance.

9.2.2 Filter synthesis based on the input impedance function

From 9.4 we have
|S11 (j!)|2 = 1 |S21 (j!)|2
or, with s = j!:
1 s8
S11 (s)S11 ( s) = 1 s 8 = .
1 + (j ) 1 + s8
As in the previous section we must factor the RHS to isolate a term having poles only in
the left-half plane, because an S11 function that can be realized with passive network will
have poles only in the left-half plane. Proceeding exactly as before, we determine that

s4
S11 (s) = (9.24)
(s s2 )(s s3 )(s s4 )(s s5 )

s4
S11 (s) = 5 7 9 11
(s ej 8 )(s ej 8 )(s ej 8 )(s ej 8 )

Expanding the denominator, as before:

s4
S11 (s) = . (9.25)
s4 + 2.61313s3 + 3.41421s2 + 2.61313s + 1
The input impedance function, assuming 1 source and load impedance, is

1 + S11 (s)
zin (s) =
1 S11 (s)

Taking the upper (plus) sign in the numerator of 9.25

2s4 + 2.61313s3 + 3.41421s2 + 2.61313s + 1

zin (s) = .
2.61313s3 + 3.41421s2 + 2.61313s + 1
Using long division, the input impedance can be written in continued fraction form
1
zin (s) = 0.7654s + 1 . (9.26)
1.8478s + 1
1.8478s+ .7654s+1
308 CHAPTER 9. FILTER DESIGN

Next, notice that the input impedance of the network shown in Figure 9.11 can be written
in continued fraction form as:
1
zin (s) = sL1 + 1 . (9.27)
sC2 + sL3 + sC 1+1
4

Equating the coecients in equations 9.27 and 9.26 yields the results already given in
equation 9.23.
If the lower (minus) sign is selected in the numerator of equation 9.25 the expression for
zin (s) will be the inverse of that given in equation 9.26. The input admittance for that case
would be equal to the right hand side of equation 9.26 and the continued fraction expansion
representation of the input admittance will correspond to the right hand side of equation
9.26. The ladder network shown in Figure 9.12 has an input admittance function of the
same form. Equating coecients as before yields the normalized component values for the
ladder network of the form shown in Figure 9.12:

C1 = 0.7654 F (9.28)
L2 = 1.8478 H
C3 = 1.8478 F
L4 = 0.7654 H

Notice that the same list of normalized component values applies to both of the networks
shown in Figures 9.11 and 9.12, but the values of the inductances in one case correspond to
the values of the capacitors in the other case, and vice versa. Therefore, when tabulating the
normalized component values for filters of a given type and order it is sucient to give one
list of component values. The elements in the list are interpreted as alternating inductance
and capacitance values, starting with the left side of the network, if the ladder network has
a series inductor on the left side of the network (as in Figure 9.11). When applied to a
ladder network with a shunt capacitor on the left side of the network (as in Figure 9.12) the
elements in the list are interpreted as alternating capacitance and inductance values.
The element values for lowpass prototype filters of any type and order can be derived
by applying the procedures illustrated in this example. The essential information that is
required to derive the component values is the locations of the left-half plane poles associated
with the Transducer power gain function. A table of component values for lowpass prototype
Butterworth filters is provided in section 9.2.3.
9.2.3 Component values for lowpass prototype Butterworth filters

n g1 g2 g3 g4 g5 g6 g7 g8 g9 g10
1 2.0
2 1.41421 1.41421
3 1.0 2.0 1.0
4 0.765367 1.84776 1.84774 0.765367
5 0.618034 1.61803 2.0 1.61803 0.618034
6 0.517638 1.41421 1.93185 1.93185 1.41421 0.517638
7 0.445042 1.24698 1.80194 2.0 1.80194 1.24698 0.445042
8 0.390181 1.11114 1.66294 1.96157 1.96157 1.66294 1.11114 0.390181
9 0.347296 1.0 1.53209 1.87939 2.0 1.87939 1.53209 1.0 0.347296
10 0.312869 0.907981 1.41421 1.78201 1.97538 1.97538 1.78201 1.41421 0.907981 0.312869

Table 9.3: Component values for lowpass prototype Butterworth filters.

9.2. EXAMPLE: SYNTHESIS OF 4TH ORDER BUTTERWORTH FILTER
309
310 CHAPTER 9. FILTER DESIGN

9.3 Example - 3rd order/0.1 dB ripple Chebyshev low-

pass
The transducer power gain function for a Chebyshev filter has the form
1
|S21 (j!)|2 = .
1 + 2 Cn2 (!)

Then, with ! = s/j:

1
|S21 (s)|2 = .
1+ 2 Cn2 ( sj )

2 Cn2 ( sj )
|S11 (s)|2 = 1 |S21 (s)|2 = . (9.29)
1 + 2 Cn2 ( sj )
Note that the -3 dB corner frequency of the Chebyshev response function does not occur
at ! = 1. Instead, ! = 1 corresponds to the frequency where the response has an attenuation
equal to the specified ripple. The response descends into the stopband at ! > 1. After the
lowpass prototype filter has been designed, the component values can be scaled to yield a
new prototype filter that has -3 dB frequency equal to ! = 1.
A third order Chebyshev filter with 0.1 dB ripple is obtained by using2 = 10(0.1/10) 1 =
0.023293 and C32 (x) = (4x3 3x)2 in equation 9.29, which yields:

(s3 + 0.75s)2
|S11 (s)|2 =
s6 + 1.5s4 + 0.5625s2 2.68321
The left-half plane poles associated with this function are located at:

s1 = 0.969406

s2 = 0.484703 j1.20616

s3 = 0.484703 + j1.20616

Thus, we have

(s3 + 0.75s) (s3 + 0.75s)

S11 (s) = = 3
(s s1 )(s s2 )(s s3 ) s + 1.93881s2 + 2.62949s + 1.63805

If the upper sign is chosen, then:

1 + S11 (s) 2s3 + 1.93881s2 + 3.37949s + 1.63805

zin (s) = = .
1 S11 (s) +1.93881s2 + 1.87848s + 1.63805

Using long division:

1
zin (s) = 1.0316s + 1 .
1.1474s + 1.0316s+1
Thus, the element values for the lowpass prototype filter are:

g1 = 1.0316 g2 = 1.1474 g3 = 1.0316.

9.3. EXAMPLE - 3RD ORDER/0.1 DB RIPPLE CHEBYSHEV LOWPASS 311

Since the network has 3 elements, it will be either a T network or a PI network. If the
T-network is employed, g1 and g3 represent the values of the series inductors (in Henries)
and g2 represents the value of the shunt capacitor (in Farads). If the PI-network is employed
g1 and g3 are the shunt capacitor values (in Farads) and g2 is the series inductor value (in
Henries).
It may be desirable to scale the Chebyshev prototype filters such that the -3 dB frequency
is located at ! = 1 s 1 . The third order prototype filter that has been designed so far has
attenuation equal to 0.1 dB at ! = 1 s 1 . The -3 dB frequency is located at ! = 1.3795 s 1 .
The filter can be scaled so that the -3 dB frequency is at ! = 1 by multiplying the element
values by 1.3795. The new prototype values are:

g1 = 1.4994 g2 = 1.6678 g3 = 1.4994

9.3.1 Component values for odd-order lowpass prototype Chebyshev filters with 0.1 dB ripple
312

n g1 g2 g3 g4 g5 g6 g7 g8 g9
1 0.305241
3 1.03156 1.1474 1.03156
5 1.14681 1.37121 1.975 1.37121 1.14681
7 1.18118 1.42281 2.09667 1.5734 2.09667 1.42281 1.18118
9 1.19567 1.4426 2.13455 1.61672 2.20537 1.61672 2.13455 1.4426 1.19567

Table 9.4: Component values for odd order lowpass prototype Chebyshev filters with 0.1 dB ripple and equal source and load
terminations. These prototype values produce a filter with attenuation equal to the passband ripple (-0.1 dB) at ! = 1 s 1 .
CHAPTER 9. FILTER DESIGN
9.4. FREQUENCY AND IMPEDANCE SCALING 313

9.4 Frequency and Impedance scaling

So far we have synthesized lowpass prototype filters that have provide some desired response
function having a corner frequency of ! = 1 s 1 when terminated with source and load
resistance of 1 . To scale a lowpass prototype filter to a new frequency, it is necessary to
scale the component values so that they have the same reactance at the new cuto frequency
as the prototype values have at ! = !c = 1 rad/s. The new inductor and capacitor values
can therefore be obtained from the calculated values by dividing the calculated values by
the desired cuto frequency, in rad/s.
To scale the filter to a new impedance value, it is necessary to scale the component
impedances so they maintain the same magnitude relative to the terminating impedance.
For example, the fourth order Butterworth lowpass protoype has L1 = 0.7654 H, so the
reactance of L1 is equal to 0.7654 at the cuto frequency, or 0.7654 times the terminating
impedance. Thus, for a terminating impedance of 50 the inductor should have a reactance
of 0.7654(50) = 38.27 at !c . So inductor values have to be multiplied by the desired
terminating impedance, Zo . The capacitor C2 has a reactance of 1/1.8478 = 0.5412 at the
cuto frequency. To scale for a terminating impedance of 50 the capacitor should have a
reactance of 0.5412(50) = 27.06 at the cuto frequency. Thus, the calculated capacitor
value should be divided by the new terminating impedance.
Both frequency and impedance scaling can be performed in one step as follows. Denote
the new scaled element values by primed quantities and the lowpass prototype values by
unprimed quantities. Then the new inductor and capacitor values are given in terms of the
desired termination impedance, Zo , and the desired cuto frequency, !c , by:

LZo
L0 = (9.30)
!c
C
C0 = . (9.31)
!c Zo

9.5 Bandpass Transformation

Bandpass filters can be realized by transforming a lowpass prototype filter. Well discuss
one simple transformation. The idea behind this transformation is to replace each branch of
the lowpass prototype with a new branch consisting of two elements. The series inductors
are replaced with series LC circuits, and the shunt capacitors are replaced with parallel
LC circuits. Recall that the series elements of the lowpass filter (inductors) look like short
circuits at the center of the filter response (at ! = 0), and the shunt elements of the lowpass
filter look like open circuits at the center of the filter response. We can transfer these
characteristics to any other center frequency, !o , by making sure that the series and parallel
branches of the new filter are resonant at !o . Likewise, at the edge of the original filters
passband, each branch had a certain impedance (relative to the terminating impedance). If
we choose the elements of the new branches such that the impedance level is the same at
the desired edges of the passband, then the shape of the bandpass filters transfer function
should look like a shifted version of the lowpass filters transfer function.
The lowpass-to-bandpass filter transformation can be defined in terms of the center
frequency and fractional bandwidth of the bandpass filter. Denoting the desired upper and
lower cuto frequencies of the bandpass filter by fl and fu and the center frequency by fo ,
314 CHAPTER 9. FILTER DESIGN

the absolute bandwidth is:

BW = fu fl (9.32)
and the fractional bandwidth is:
BW
bw = (9.33)
fo
The element values for the parallel LC shunt elements in the bandpass filter are then
written in terms of the original normalized lowpass prototype filter shunt capacitor value,
Clp as:
Clp
Cbpshunt = (9.34)
bw
1
Lbpshunt = (9.35)
Cbpshunt
The element values for the series LC series elements in the bandpass filter are given in
terms of the original normalized lowpass prototype series inductor value, Llp as:
Llp
Lbpseries = (9.36)
bw
1
Cbpseries = (9.37)
Lbpseries
These normalized bandpass filter element values are scaled to the desired center frequency
and impedance value by using equations 9.30 and 9.31 with the cuto frequency, !c , replaced
with the desired center frequency, !o .

9.5.1 Example - Bandpass filter based on 4th order Butterworth

lowpass prototype
Suppose it is necessary to design a bandpass filter with center frequency of 70 MHz, 3 dB
bandwidth of 20 MHz. The filter will operate with source and load impedances of 50. The
fractional bandwidth is then:
20
bw = = .2857 (9.38)
70
Using the 4th order lowpass Butterworth filter prototype derived in section 9.2, and
transforming that filter, the resulting bandpass filter will have the topology shown in Figure
9.13.
The element values for this filter can be obtained by scaling the the lowpass prototype
values for the 4th order Butterworth filter derived in section 9.2. The results are:

L1 = 0.305 H C1 = 17.0 pF
L2 = 17.6 nH C2 = 294 pF
L3 = 0.735 H C3 = 7.03 pF
L4 = 42.4 nH C4 = 126 pF

The transducer power gain of this filter is shown in Figure 9.14 on linear and logarithmic
scales. The lower plot shows that bandpass filters derived from lowpass prototypes using
9.6. COUPLED RESONATOR FILTERS. 315

Figure 9.13: Bandpass filter topology derived from the prototype lowpass filter shown in
Figure 9.11.

the transformation described in this section do not exhibit arithmetic symmetry, i.e., the
response falls o more slowly above the passband than it does below the passband. The
response would appear to be symmetric if plotted on a logarithmic frequency axis, as shown
in Figure 9.15.
In practice, the very large stopband attenuation shown in Figure 9.15 would not be
attained with a real filter. The analysis has ignored parasitic reactances associated with
real components (e.g., lead inductance and capacitance between turns in an inductor). In
addition, unmodeled coupling between elements and the input and output ports of the filter
will inevitably occur. These eects generally limit the attainable stopband attenuation to
numbers in the range 50-70 dB.
Finally, it should be noted that this method of transforming a lowpass prototype into
a bandpass filter will result in realizable component values only when the target fractional
bandwidth is larger than 10-15%. In general, as the target fractional bandwidth becomes
smaller, the ratio between the largest and smallest required component values will increase.
Notice that in this 20% bandwidth design, the ratio between the largest and smallest in-
ductor and between the largest and smallest capacitor is 42. As the target bandwidth is
decreased, the ratios increase until a point where it is no longer practical or possible to
realize the needed component values with reasonably small losses.

9.6 Coupled resonator filters.

The lowpass to bandpass transformation discussed in the previous section results in a net-
work containing multiple resonators, each of which has dierent L and C values. In addition,
it was pointed out that the required element values span a very wide range when the frac-
tional bandwidth of the filter is small. If a small fractional bandwidth (e.g. smaller than
10%) is needed it is generally better to use a filter topology that is based on identical cou-
pled resonators. For example, one such topology is shown in Figure 9.16 where 4 parallel
LC resonators are capacitively top-coupled to each other and to the source and load. This
filter topology has the same number of resonators as the one derived in the previous section,
however it provides the designer with the freedom to independently choose the properties
of the resonators. This extra freedom makes it possible to create a filter in which all of the
resonator inductors (or capacitors) are identical, greatly simplifying implementation of the
filter.
The normalized lowpass prototype element values, gi , provide the essential information
necessary to design a bandpass filter based on coupled resonators. For detailed information
316 CHAPTER 9. FILTER DESIGN

Figure 9.14: Transducer power gain (|S21 |2 ) for the 70 MHz bandpass filter design based on
a fourth order Butterworth lowpass prototype. The upper and lower plots show the power
gain on linear, and logarithmic scales, respectively.
9.6. COUPLED RESONATOR FILTERS. 317

Figure 9.15: Same as Figure 9.14 except the frequency axis is plotted on a logarithmic scale
to illustrate the symmetry of the response with respect to the logarithm of frequency.

Zo C01 C12 C23 C34 C45

Zo

C1 L1 C2 L2 C3 L3 C4 L4

Figure 9.16: A coupled resonator filter based on capacitive coupling between parallel LC
resonators.
318 CHAPTER 9. FILTER DESIGN

on the design of coupled resonator filters see references [2,3,4]. Here we will simply de-
scribe the procedure for the design of a filter based on parallel LC resonators with common
resonator inductance and capacitive coupling. The design proceeds as follows -

1. Determine the required corner frequencies, !1 and !2 , for the prototype filter. The
bandwidth of the filter will be BW = !2 !1 and the center frequency will be
p
!o ' !1 !2 .

2. Choose a common resonator inductance, Lr , to be used for all resonators. The ca-
pacitance required to resonate this inductance at the desired center frequency, !o , is
Cr = (!o2 Lr ) 1 .

3. Calculate parameters Ji,i+1 as follows:

s
BW Cr
J0,1 =
Zo g 1

s
1
Ji,i+1 |i=1 to n 1 = BW Cr
gi gi+1

s
BW Cr
Jn,n+1 =
Zo g n

4. Calculate the coupling capacitor values:

J0,1
C0,1 = p
!o 1 (Zo J0,1 )2

Ji,i+1
Ci,i+1 |i=1 to n 1 =
!o

Jn,n+1
Cn,n+1 = p .
!o 1 (Zo Jn,n+1 )2

5. Calculate the shunt capacitance across each resonator inductor:

C0,1
C1 = Cr C1,2
1 + (!o Zo C0,1 )2

Ci |i=2 to n 1 = Cr Ci 1,i Ci,i+1

Cn,n+1
Cn = Cr Cn 1,n
1 + (!o Zo Cn,n+1 )2
9.6. COUPLED RESONATOR FILTERS. 319

9.6.1 Example - Coupled resonator filter with 2 parallel LC res-

onators based on Butterworth lowpass prototype
In this example we summarize the design of a bandpass filter based on a second order
(n = 2) Butterworth lowpass prototype. The filter is to have center frequency 50 MHz and
bandwidth 2.5 MHz (5% fractional bandwidth). It is to be used in a system with source
and load impedance Zo = 50 .
To obtain a passband response that is approximately symmetric around 50 MHz, we
choose the upper and lower corner frequencies as follows:

!1 = 2(50 2.5/2) 106 = 3.0631 108 s 1

!2 = 2(50 + 2.5/2) 106 = 3.2201 108 s 1

.
Then the target center frequency will be the geometric mean of the two corner frequencies:
p
!o = !1 !2 = 3.1406 108 s 1

The target bandwidth is

BW = 2 2.5 106 = 1.5708 107 s 1

.

The filter will contain two parallel resonators as shown in Figure 9.17. We shall choose

Zo C01 C12 C23

Zo

C1 L1 C2 L2

Figure 9.17: Filter topology for the coupled resonator filter design example. The design
procedure that we have described assumes that the resonator inductances are equal (L1 =
L2 = Lr ).

a resonator inductance Lr = L1 = L2 = 0.2 H. In practice, the inductance value is

typically chosen by selecting the value that results in the highest possible inductor QL
(and hence highest possible resonator Q) at the intended center frequency, !o . Since we
ignored component losses in the design, a real filters performance will approximate the ideal
predicted performance only when the component Q of each filter component is significantly
larger than the filters Q = fo /BW. The freedom to independently choose the inductance
value to optimize resonator Q makes the coupled resonator filter very attractive for filters
with narrow bandwidth (high Q).
The total capacitance required to resonate the chosen resonator inductance is Cr =
1/(!o2 Lr ) = 50.69 pF. The lowpasspprototype component values for a second order Butter-
worth filter (n = 2) are g1 = g2 = 2.
The Ji,i+1 can now be calculated. The values are:

J01 = 0.0033557 J12 = 0.0005631 J23 = 0.0033557.

320 CHAPTER 9. FILTER DESIGN

The coupling capacitances are:

C01 = 10.84 pF C12 = 1.793 pF C23 = 10.84 pF.

The resonator capacitances are:

C1 = C2 = 38.06 pF.

The transducer power gain of the resulting filter is plotted in Figure 9.18 on both linear
and logarithmic scales. The asymmetry that is evident in the bottom plot is a result of the
capacitive top-coupling. At high frequencies the network reduces to an all-capacitor network
with roughly constant insertion loss.

9.7 References
1. Carson, Ralph S., Radio Concepts: Analog, John-Wiley and Sons, New York, 1990.
2. Zverev, Anatol I., Handbook of Filter Synthesis, John-Wiley and Sons, New York,
1967.
3. Rhea, Randall W., HF Filter Design and Computer Simulation, Noble Publishing,
Atlanta, 1994.
4. Matthaei, George L., Leo Young, E. M. T. Jones, Microwave Filters, Impedance-
Matching Networks, and Coupling Structures, Artech House, Massachusetts, 1980.
9.7. REFERENCES 321

1.0

0.8

GT 0.5

0.2

0.0
45.0 46.0 47.0 48.0 49.0 50.0 51.0 52.0 53.0 54.0 55.0
Frequency (MHz)

0.0
5.0
10.0
15.0
20.0
GT (dB) 25.0
30.0
35.0
40.0
45.0
50.0
10.0 100.0
Frequency (MHz)

Figure 9.18: Transducer power gain (|S21 |2 ) for the 50 MHz bandpass filter based on a
second order Butterworth lowpass prototype and implemented with top-coupled parallel LC
resonators. The upper and lower plots show the power gain on linear and logarithmic scales,
respectively.
322 CHAPTER 9. FILTER DESIGN

9.8 Homework Problems

1. A figure of merit that is used to characterize filters is the shape factor, defined to
be the ratio of the transducer power gain bandwidth at the -60 dB and -6 dB points,
respectively. The shape factor will be larger than 1, and will approach 1 as the filter
response function approaches an ideal rectangular response. Find an expression for
the shape factor for Butterworth bandpass filters as a function of the filter order, n.
Give numerical values for the shape factor for n=3 and n=6.
2. Find and sketch the left-half plane pole locations for a fifth-order Butterworth filter.

3. Design a second-order Butterworth lowpass filter with -3 dB frequency, 50 MHz and

terminating impedances Z0 = 50 . Draw the filter and give component values. Use
ADS to plot the transducer power gain (|S21 |2 ). Plot the transducer power gain in
dB against frequency (Use a logarithmic axis for frequency.) from at least 0.5 MHz to
500 MHz.

4. Transform the filter designed in 9-3 to a bandpass filter with center frequency 50 MHz
and 3 dB bandwidth 10 MHz. Draw the filter and give component values. Use ADS to
plot the transducer power gain. Plot the transducer power gain in dB against frequency
on a logarithmic frequency axis from 0.5 MHz to 500 MHz. Make another plot covering
a smaller frequency range around 50 MHz to show the detailed characteristics of the
transfer function within the passband.

5. A Butterworth lowpass filter using the smallest possible number of components (i.e.,
smallest possible filter order n) is to be designed to filter the output of a source that
produces output at a fundamental frequency fo = 15 MHz and at harmonics of that
frequency mf0 (m = 2,3...). The attenuation at fo should be no more than 0.5 dB and
the attenuation at 2fo should be at least 30 dB. Specify the -3 dB cuto frequency of
the filter fc (in MHz) and the filter order n (an integer) that will achieve these design
goals. If your parameters are used, what would the attenuation be at the second
harmonic frequency 2fo ?
Chapter 10

Noise in 1- and 2-ports

10.1 Noise Characterization of 1-ports

There are various physical phenomena that can produce noise in electronic circuits, in-
cluding:

Thermal noise - a result of random motion of charge carriers in a conductor. This type
of noise will be present in any dissipative electronic circuit element (e.g. resistors or
the dissipative parts of transistors and diodes), even in the absence of any externally
applied bias voltage or current. Thermal noise has a white spectrum, which means
that the power available in a frequency bandwidth df is independent of the center
frequency of the band.

Shot noise - is associated with current flow in semiconductor devices and vacuum
tubes. It will be present in these active devices when an average (DC) current flows.
In such devices, current flow can be modeled as a series of independent current pulses
occurring at random and a DC bias current will be associated with a time-varying
shot-noise current whose intensity spectrum is proportional to the DC current. The
shot noise spectrum is white at suciently low frequencies.

Flicker noise or 1/f noise any process with a noise power spectral density that
varies as 1/f for low frequencies. If present, flicker noise will always dominate at
suciently low frequencies.

10.1.1 Thermal Noise in Resistors

A noise voltage, as shown in Figure 10.1, is always present at the terminals of a resistor
because the charge carriers within the device undergo Brownian motion. The random-walk
of individual charges constitutes a random current, which produces a corresponding thermal
noise voltage across the terminals of the device. This noise was identified and carefully
measured by J. B. Johnson while working at Bell Telephone Laboratories1 . A model for
the noise measured by Johnson was published at the same time by H. Nyquist, who worked
out an expression for the rms thermal noise voltage across any conductor. The result was
1 Thermal Agitation of Electricity in Conductors, J. B. Johnson, Physical Review, Vol. 32, p. 97, July,
1928.

323
324 CHAPTER 10. NOISE IN 1- AND 2-PORTS

derived using basic thermodynamics, and a fundamental result from statistical mechanics,
and is known as Nyquists theorem.2

en (t)
+
R en (t)
-
t

Figure 10.1: Noise voltage across the terminals of a resistor as it would be displayed on an
instrument havingp a lowpass frequency response with bandwidth, B. The rms noise voltage
is proportional to B and the typical time between zero crossings is proportional to B 1 .

The noise voltage waveform is properly modeled as a zero-mean, wide-sense-stationary,

random process with Gaussian probability density function. The spectral density of the
noise voltage due to thermal fluctuations is given by the Nyquist noise formula:
hf
Sn (f ) = 4R (Volts2 /Hz) (10.1)
ehf /kT 1
where
f = frequency, Hz
h = 6.62 10 34
J sec, Planck0 s constant
k = 1.38 10 23
J/K, Boltzman0 s constant (10.2)
R = resistance,
T = physical temperature of the resistor, K
When f kT /h so that the exponential term in the denominator of equation 10.1 is well
approximated by the first two terms in a Taylor series expansion, Nyquists formula reduces
to:
Sn (f ) = 4kT R (Volts2 /Hz) (10.3)
At standard temperature, T = 290 K, kT /h ' 6 1012 Hz = 6000 GHz, which lies in the
infrared. So, at standard temperature the low-frequency approximation will be valid for all
RF and microwave applications. The spectral densities given in equations 10.1 and 10.3
represent the mean-square noise across the resistor within a bandwidth of 1 Hz, centered on
any frequency f . In practice, the insrument used to sense the noise voltage will have a finite
bandwidth that diers from 1 Hz. Suppose the instrument responds only to frequencies
between the upper and lower frequencies denoted by fl and fh , respectively. Then the
mean-square noise voltage measured by the instrument would be:
Z fh
< e2n >= Sn (f )df , (10.4)
fl

or, denoting the bandwidth by B = fh fl :

< e2n >= 4kT BR (Volts2 ) (10.5)

2 Thermal Agitation of Electric Charge in Conductors, H. Nyquist, Physical Review, Vol. 32, p. 110,
July 1928.
10.1. NOISE CHARACTERIZATION OF 1-PORTS 325

For example, suppose we wish to calculate the mean-square noise voltage produced
across a 100 k resistor in a bandwidth of 1 MHz at standard temperature (T = 290 K).
At standard temperature:
20
4kT = 1.6 (10 ) Joules (10.6)
< e2n > = 4kT BR
20
= 1.6 (10 )(106 )(100)(103 )
9
= 1.6 (10 ) (Volts)2

Thus the rms noise voltage is

p
erms = < e2n > = 40 (10 6 ) V = 40 V. (10.7)
p
The rms noise voltage across the resistor scales as B, so the noise voltage across the resistor
in a 100 MHz bandwidth would be ten times as large, or 0.4 mV.
Equation 10.5 can be generalized to describe the open circuit noise voltage across a
complex impedance Z that is realized by lossless elements and resistances all having the
same physical temperature. Suppose the impedance is Z(f ):

Z(f ) = R(f ) + jX(f ), (10.8)

then the noise variance across that impedance within bandwidth B will be
Z
< e2n >= 4kT R(f ) df. (10.9)
B

For example, we can calculate the total mean-square noise voltage across the parallel RC
circuit shown in Figure 10.2. Here, we will make use of equation 10.9, which provides

R (noisy) C

Figure 10.2: Parallel RC circuit with a noisy resistor.

a prescription for calculating the noise voltage between two terminals associated with a
network consisting of an arbitrary configuration of dissipative and lossless elements. The
only requirement is that all dissipative elements are at the same physical temperature. The
impedance of the parallel RC combination is:

R(1 j!RC)
Z(!) = (10.10)
1 + (!RC)2

According to equation 10.9 the mean-square noise voltage depends only on the real part of
the impedance:
R
<[Z] = R(f ) = .
1 + (2f RC)2
326 CHAPTER 10. NOISE IN 1- AND 2-PORTS

Using equation 10.9, the total mean-square noise voltage across the resistor is:
Z 1
2 R kT
< en >= 4kT 2
df = .
0 1 + (2f RC) C

Note that the result depends only on the capacitance, C, and is independent of the resistance!
How is this reconciled with the fact that the mean-square voltage within any dierential
bandwidth, df , is proportional to the resistance? The answer is that the bandwidth of the
RC filter is inversely proportional to R; so for fixed capacitance, larger resistances have
larger mean-squre voltages but proportionally smaller bandwidths, hence the total noise
power, integrated over all frequencies, is independent of the resistance.
A Thevenin equivalent circuit can be used to model a noisy resistor, as shown in Figure
10.3a. The Norton equivalent circuit shown in Figure 10.3b is an equivalent alternative

R (noiseless)

en in g = 1/R (noiseless)

2 2
(a) < en >= 4kT BR (b) < in >= 4kT Bg

Figure 10.3: (a) Thevenin equivalent circuit. (b) Norton equivalent circuit.

representation.
The available noise power from a noisy resistor is defined to be the power delivered to a
2
matched load, as in Figure 10.4. The instantaneous power delivered to the load is (enR/2) ,

R
+
en R en /2
-

Figure 10.4: Circuit for calculation of noise power available from a noisy resistor.

<e2n >
so the average available noise power is 4R . Using Equation 10.5, the average available
noise power within a bandwidth B is:

< e2n > 4kT BR

=
4R 4R

= kT B (W) (10.11)

This result says that the noise power available from a resistor (or network of resistors all at
the same physical temperature) is independent of the resistance. For later use, we will also
10.1. NOISE CHARACTERIZATION OF 1-PORTS 327

define the available noise power per unit bandwidth:

Na = kT (W/Hz) (10.12)

It is useful to know that kTo = 4 10 21 Watts, or approximately 174 dBm. So the noise
power available from any resistor at standard temperature is 174 dBm per Hz.
Next, the resistor noise model is generalized to allow it to represent arbitrary one-ports,
such as an antenna.

10.1.2 Noise Representation of Arbitrary 1-ports

If a 1-port device contains sources of noise in addition to the thermal noise it is said to exhibit
excess noise. In this case it is still possible to model the device by a Thevenin or Norton
equivalent circuit, and to use the Nyquist noise equation to calculate mean-square noise
voltage, but it is necessary to define an eective temperature (dierent than the physical
temperature) to account for the extra noise. When this model is used, the formula for the
voltage is
< e2n >= 4kTn BR (10.13)
and the noise current is
< i2n >= 4kTn Bg (10.14)
where R and g=1/R represent the actual resistance and conductance, but the temperature
Tn is called the equivalent noise temperature of the device.

10.1.3 Noise Representation of a Receiving Antenna

Antennas are 1-ports (like resistors) and the noise power available from an antenna can be
represented by an equivalent noise temperature. The real part of the antenna impedance
can be written as the sum of radiation and ohmic resistances

Re[Zant ] = Rrad + Rloss (10.15)

where Rrad accounts for power radiated and Rloss accounts for power that is dissipated
in the antenna and its nearby environment. An antenna that transmits most of the power
delivered to it when used for transmission is called an ecient antenna. An ecient antenna
has:
Rloss Rrad (10.16)
The noise voltage across an antennas terminals comes from two sources:
1. Noise received from external sources.
2. Thermal noise generated in Rloss .
The noise voltage across the antenna terminals can be represented as if it was generated by
an actual resistor, R = Rrad + Rloss , at some temperature TA which is called the eective
antenna temperature or equivalent noise temperature of the antenna. This is essentially the
same as Tn , which was defined in section 10.1.2; but the subscript A is used to make it clear
that the eective temperature is associated with an antenna. For an ecient antenna, TA
primarily represents noise received from external sources and has nothing to do with the
physical temperature of the antenna. Instead, TA will depend on the intensity of external
328 CHAPTER 10. NOISE IN 1- AND 2-PORTS

noise signals picked up by the antenna at the particular frequency of interest. In general,
TA is a parameter that must be measured, or known from previous experience with the
particular antenna and location at which the antenna is to be used.
For example, suppose that an ecient antenna with radiation resistance of 200 has an
rms noise voltage of 0.1 V at its terminals when measured in a bandwidth of 104 Hz. The
equivalent noise temperature of the antenna can be calculated using:
p p
erms = < e2n > = 0.1 V = 4kTA BR.

With R = 200 and B = 104 :

(0.1 V)2
TA = = 90.6 K
4kBR
Figure 10.5 shows the equivalent antenna circuit (for noise analysis).

200 At 90.6 K physical temperature

Figure 10.5: Equivalent circuit for analysis of antenna noise.

Typical antenna noise temperatures are shown in Figure 10.6. Noise radiated by sources
within the earth-ionosphere cavity is the dominant source of noise below approximately
20 MHz, and is largest at night (corresponding to the curve labeled Max.) when low
ionospheric absorption allows noise to be propagated in from long distances. During the
daytime, ionospheric absorption limits the amount of noise received from distant sources
and the antenna temperature is significantly lower than at nighttime. Above 20 Mhz, the
figure assumes that the antenna beam is directed away from the earth, so that dominant
noise sources are extraterrestrial noise sources.
If we ignore radio noise generated by human activities, the primary source of natural
noise between 20 MHz (roughly) and 1 GHz is radio noise emitted by sources in our galaxy
and, depending on the antenna pattern and the pointing direction of the antenna relative
to the sun, perhaps some noise from the sun. Antenna temperatures for antennas that
look at the sky vary with time of day because the plane that contains most of the radio
sources in our galaxy will drift through the antenna beam as the earth rotates. Antenna
temperature due to the galactic sources also varies approximately as f 2 , so that maximum
daily antenna temperatures might range from several thousand K at 100 MHz to a few tens
of K at 1 GHz. At frequencies between 1-5 GHz or so, TA reaches its smallest values. In fact,
the contribution from sources within our galaxy falls to negligible values in this frequency
range, and the dominant noise is contributed by a background noise that is uniform in all
directions and corresponds to an equivalent blackbody temperature (and hence, antenna
temperature) of approximately 3 K. This is the so-called microwave background radiation
which was discovered by Bell Laboratories scientists Arno Penzias and Robert Wilson in
1965. This discovery confirmed theoretical predictions based on the Big Bang model of an
expanding universe. Penzias and Wilson were awarded the Nobel Prize in physics in 1978,
in recognition of the significance of their discovery. Their original paper had the unassuming
title A Measurement of Excess Antenna Temperature at 4080 Mc/s.
10.1. NOISE CHARACTERIZATION OF 1-PORTS 329

1e + 15

1e + 14

1e + 13

1e + 12

1e + 11

1e + 10
Max.
1e + 09

TA 1e + 08
Atmospheric noise
1e + 07

1e + 06

100000
Min. Cosmic noise
10000
Daily max.
1000

100

10
Daily min.

1
0.1 1 10 100 1000
Frequency - MHz

Figure 10.6: Typical antenna temperatures as a function of frequency. These curves are
adapted from Figure 11-44 in the book Electromagnetic Waves and Radiating Systems, by
Jordan and Balmain, Prentice Hall, 1968.
330 CHAPTER 10. NOISE IN 1- AND 2-PORTS

At frequencies above 5 GHz, antenna temperatures begin to rise again and can approach
several hundred K at certain frequencies, due to thermal emission from water vapor and
oxygen (O2 ) in earths atmosphere.
It can be shown that when an antenna beam looks at an object that emits thermal (black
body) radiation, the antenna temperature will be equal to the temperature of the object,
provided that the antenna beam is filled by the object. If the antenna beam views many
objects, each having dierent temperatures, then the resulting antenna temperature will be
a weighted average of the individual object temperatures, where the weighting depends on
the angular size of each object, and the angular response function of the antenna.
When an antenna beam is pointed at the earth, as in Figure 10.7a, the antenna tem-
perature will be approximately equal to the equivalent black body temperature of the earth
or, roughly, 290 K. This has implications for satellite-to-ground communications systems.
For example, when a ground station with a narrow antenna beam looks at a geostationary
satellite (as shown in Figure 10.7b ), the earth station antenna beam sees mostly the sky.

(a) (b)

Figure 10.7: (a) When a satellite antenna beam looks at the earth, the beam sees a warm
body, emitting thermal radiation with an eective temperature on the order of 290 K. (b)
When a ground station antenna looks at a satellite, the beam sees mainly cold space.

Hence, a ground station antenna operated at a frequency between roughly 1-5 GHz will have
a significantly smaller antenna temperature than a satellite antenna looking back at earth.
Since the ground station antenna looks at the cold sky, it picks up very little external
noise and the noise generated within the ground station receiver is usually a limiting factor
in determining the receiver sensitivity. On the other hand, the satellite antenna looks at
warm earth, and is relatively noisy and commonly contributes more noise than the receiver
itself. This means that it is worthwhile to use an extremely low-noise receiver for the ground
station, but not for the satellite.
For a point-to-point terrestrial communications link with the antenna radiation pattern
directed horizontally over the surface of the earth, half of the beam will see cold sky, and
the other half will see earths surface, resulting in an antenna temperature somewhere in
between the sky temperature and 290 K. The antenna beam may also see non-thermal
sources of noise located on earths surface, which raise the eective temperature of the earth
above 290 K.

10.1.3.1 The eective antenna temperature of an inecient antenna

Consider an antenna with feed-point resistance

Re[Zant ] = Rrad + Rloss , (10.17)

10.1. NOISE CHARACTERIZATION OF 1-PORTS 331

where Rrad , Rloss represent radiation and loss components of the feedpoint resistance, re-
spectively. It is assumed that the presence or absence of loss does not aect the current
distribution that is responsible for the radiated fields. Hence if the antenna and all compo-
nents in the vicinity of the antenna are changed to lossless materials (i.e. lossy conductors
replaced with perfect conductors, lossy dielectrics replaced with perfect dielectrics) then
Rloss ! 0, and Rrad is unaected by this change.
Denote the antenna temperature of the (hypothetical) lossless antenna by TA . The noise
power per unit bandwidth (p.u.b.) available from the lossless antenna would be kTA . For
simplicity, assume that all lossy materials are at physical temperature Tloss . The noise
power p.u.b. available from the loss resistance only will be kTloss .
An equivalent circuit for the lossy antenna is obtained by replacing noisy p resistances
R
p rad and R loss with Thevenin noise voltage sources having rms voltages 4kT A Rrad and
4kTloss Rloss in series with noiseless resistances Rrad and Rloss , respectively.
p The open cir-
cuit noise voltage across the antenna terminals will have rms value 4k(TA Rrad + Tloss Rloss ),
and the noise power available p.u.b will be
k(Rrad TA + Rloss Tloss )
Na = = k(TA + (1 )Tloss ) (10.18)
Rrad + Rloss
where
Rrad
= (10.19)
Rrad + Rloss
is the antenna eciency. Therefore, the eective antenna temperature is

Tef f = TA + (1 )Tloss . (10.20)

Equation 10.20 says that as the antenna eciency decreases ( ! 0) the eective antenna
temperature aproaches the temperature of the lossy components of the antenna.
Now consider the eect of antenna losses on signal to noise ratio at the antenna output.
In the lossy case, the available signal power is reduced by the factor relative to a lossless
antenna. The noise power p.u.b. available from a lossless antenna is kTA , whereas in the
lossy case it is kTef f which can be written as kTA [1 + ((1 )/)Tloss /TA ]. Therefore the
presence of loss resistance degrades the signal-to-noise power ratio at the antenna output
terminals by the factor
SN Rlossy 1
= . (10.21)
SN Rlossless 1 + TTloss
1
A

This is an interesting result, because it shows that SNR degradation due to antenna losses
is determined by antenna eciency as well as the ratio Tloss /TA . When Tloss /TA 1, then
even a small decrease in antenna eciency can significantly degrade the SNR. Consider a
satellite ground station where TA may be only 30 K. With Tloss = 290 K, and antenna
eciency = 0.9 (90%), the SNR is degraded by the factor 0.48, or more than 3 dB. On
the other hand, if Tloss /TA 1, then even an inecient antenna can provide essentially the
same SNR as a lossless antenna. For example, the small antennas used for AM broadcast
radios may have eciencies of 1% ( = 0.01) or less, so that the factor (1 )/ ' 100 or so.
The antenna temperature, TA is typically very large at AM broadcast frequencies, so that
Tloss /TA 1. In this situation, the product [(1 )/][Tloss /TA ] can be 1 even when
itself is small, in which case the SNR available from a small inecient antenna is essentially
the same as that available from a much larger antenna with high eciency (provided that
the radiation patterns of the small and large antennas are similar).
332 CHAPTER 10. NOISE IN 1- AND 2-PORTS

10.2 Noise Characterization of Linear 2-ports

10.2.1 Eective Input Noise Temperature
The noise added by a 2-port (e.g., amplifier, filter, mixer) can be characterized by an eective
input temperature for a given source impedance at a given frequency.
Consider in Figure 10.8 a 2-port with a hypothetical noiseless input termination, Zs ,
where Navo = available output noise power (the noise power delivered to a matched load).

Noisy
Ts = 0 ZS Navo
2-port

Figure 10.8: 2-port with a hypothetical noiseless input termination, ZS .

Figure 10.9 shows an equivalent circuit where the noise has been removed from the
2-port and assigned to the input termination by raising the temperature of the input ter-
mination to a value denoted by Te .

Noiseless
Ts = Te ZS Navo
2-port

Figure 10.9: Equivalent circuit for a 2-port with a hypothetical noiseless input termination,
ZS .

Te is the temperature which, when assigned to the input termination, produces the same
available output noise power as that of the actual 2-port. In general, Te is a function of Zs
and frequency.
Te only characterizes the noise generated within the 2-port, not the input termination.
The operating noise temperature of a system (Top ) is used to characterize the total amount of
output noise due to the combined contributions from the 2-port and the input termination,
i.e., in a practical system the source is also noisy and can be assigned a noise temperature,
Ts , as in Figure 10.10. An equivalent circuit for noise analysis can be obtained by adding

Noisy
Ts > 0 ZS Navo
2-port

Figure 10.10: Realistic system with a noisy source.

the eective input temperature of the 2-port to the source temperature as in Figure 10.11.
10.2. NOISE CHARACTERIZATION OF LINEAR 2-PORTS 333

Noiseless
Top = Ts + Te ZS Navo
2-port

Figure 10.11: Eective input temperature of the 2-port is added to the source temperature.

Ts could represent the antenna noise or the noise from the source impedance of a signal
generator. Note carefully that by adding the noise temperatures of the source and the
2-port, we have eectively added the noise powers, since noise power is proportional to
temperature. The implicit assumption is that the input noise voltage due to the source and
due to the 2-port are statistically independent, and hence, uncorrelated. In that case the
noise powers due to the two sources add.

10.2.2 Noise Factor (F) and Noise Figure (NF)

Noise Factor (F) or Noise Figure (NF) are alternatives to the eective input temperature
for characterizing the noise performance of a 2-port. The noise factor (F) can be defined in
two equivalent ways. The first definition is based on the signal-to-noise ratio degradation
between the input and output of a noisy 2-port, assuming that the noise power is calculated
within the same bandwidth at both ports.
Consider Figure 10.12, which can be used as a basis for deriving an expression for the
Noise Factor. The Noise Factor is defined under the assumption that the noise power per

To
ZS
Savs Noisy Savo = GASavs
Navs 2-port Navo = GANavs + N2port

Available Gain, GA

Figure 10.12: System for defining F.

unit bandwidth available from the source is constant at the value kTo , where To is the
standard temperature 290 K:

INPUT SNR
F (10.22)
OUTPUT SNR Ts =To =290K
334 CHAPTER 10. NOISE IN 1- AND 2-PORTS

where
To = 290 K standard temperature

Savs = signal power available from the source

Navs = kTo noise power available from the source

GA = Savo
Savs available gain of the 2 port

N2port = kTe GA noise power (1 Hz BW) available at output due to 2

port only
(10.23)
It is important to note that F does not necessarily describe the SNR degradation in a
real system, because the definition is based on the assumption that the input termination
has eective temperature To = 290 K.
The second definition of F can be derived from Equation 10.22 with reference to Figure
10.12:
Savs /Navs Savs /Navs Navo GA Navs + N2port
F = = = (10.24)
Savo /Navo GA Savs /Navo GA Navs GA Navs
The final result can be thought of as an alternative definition of Noise Factor that does not
make reference to any particular signal. Here F is simply the ratio of the actual noise power
available at the output of the 2-port (with input termination at standard temperature) to
the noise power that would be available if the 2-port was noiseless.
The noise factor can be expressed in terms of the eective input temperature of the
2-port:
N2port N2port /GA kTe
F = 1+ =1+ =1+ (10.25)
GA Navs Navs kTo
so
Te
F =1+ , To = 290 K
To
Since To is a standard temperature (= 290 K), specification of eective input temperature
(Te ) is equivalent to specifying F. For a noiseless 2-port, Te = 0, which means that F = 1,
i.e., there is no SNR degradation.
Noise figure (NF) is simply the noise factor expressed in dB, i.e.,
NF = 10 log(F ) (10.26)

Te
= 10 log 1 +
To
The relationship between NF and Te is summarized in Table 10.1, and NF for some
typical receiving systems is summarized in Table 10.2.
Noise Factor (and therefore Te and NF) is a function of source impedance, which means
that to achieve the best noise performance, the 2-port must be presented with some optimum
source impedance, Zopt , which may not be the impedance that gives an impedance match
at the input of the 2-port in fact, it usually is not. This means that the best NF for an
amplifier doesnt always go hand-in-hand with the highest gain. See section 10.5 for more
details.
10.2. NOISE CHARACTERIZATION OF LINEAR 2-PORTS 335

NF (dB) Te (K)
0.25 17.2
0.50 35.4
1.0 75.1
2.0 169.6
5.0 627.1
10.0 2610

Table 10.1: Conversion between NF and eective input temperature.

System NF (dB)
HF Comm. Receiver 6-10
VHF Comm. Receiver 1-4
Satellite Ground Station <1
Radio Astronomy (L-Band) <0.5

Table 10.2: Some typical Noise Figures.

10.2.3 Eective Input Temperature of a Passive Attenuator

In general, it is dicult to accurately predict the eective input temperature of a 2-port
from basic physical considerations and it is necessary to measure Te . For the special case
of a passive lossy 2-port it is possible to accurately calculate the NF. (A passive lossless
2-port will have Te = 0 since no noise is generated in lossless components.) An example of
a passive lossy 2-port is a long piece of lossy coaxial cable.
Consider a passive 2-port network with available loss L ( G1A > 1). For the moment,
assume that the source termination is also passive and that the 2-port and source termination
are in thermal equilibrium at physical temperature Tatt . Since the network is passive and
at physical temperature Tatt , the noise power per unit bandwidth available at the output
port will be Navo = kTatt . This available output noise power can be thought of as the
combination of the source contribution and a contribution from the 2-port. Referring the
2-port contribution to the input, representing it by an eective input temperature Te , and
equating the available output noise power to the sum of the source and 2-port contributions
yields:
Tatt + Te
kTatt = kTatt GA + kTe GA = k (10.27)
L
Solving for Te yields:
Te = Tatt (L 1) (10.28)
This result is also valid when the source has an arbitrary noise temperature.
The noise factor of a passive attenuating network is:
Te Tatt
F =1+ =1+ (L 1) (10.29)
To To
If the physical temperature of the attenuator is To (Tatt = To ),

F = L ) N F = 10 log L (10.30)
336 CHAPTER 10. NOISE IN 1- AND 2-PORTS

Thus, a room temperature attenuator with 10 log L = 3 dB has NF = 3 dB.

As noted in the previous section, since the available gain (and therefore, L) depends on
the source impedance presented to the attenuator, Te will depend on the source impedance.

10.2.4 Noise Temperature of Cascaded 2-ports

Consider a system consisting of a cascade of two 2-ports, as shown in Figure 10.13. Assuming

Te1 Te2
Noiseless ZS GA1 GA2

Figure 10.13: GA1 , GA2 are available power gains; Te1 , Te2 are eective input temperatures.

a noiseless source termination, the noise power (per unit bandwidth) available at the output
of the first 2-port is
Navo1 = kTe1 GA1 .
The noise power available at the output of the second 2-port has two terms - the first is the
amplified noise from the output of the first 2-port, and the second is the contribution from
the second 2-port:
Navo2 = Navo1 GA2 + kTe2 GA2
= k(Te1 + Te2 /GA1 )GA1 GA2
= kTe GA12
where
Te2
Te = Te1 + (10.31)
GA1
is the eective input temperature of the cascade, and GA12 = GA1 GA2 is the available gain
of the cascade.
The Noise Factor of the cascade is:
Te Te1 Te2 1
F =1+ =1+ + (10.32)
To To To GA1
1
F = F1 + (F2 1)
GA1
This is easily generalized to cascades of more than 2 2-ports. For a cascade of n 2-ports,
the eective input temperature and Noise Factor of the cascade are given by:

Te2 Te3 Ten

Te = Te1 + + + ... (10.33)
GA1 GA1 GA2 GA1 GA2 . . . GAn 1

(F2 1) (F3 1) (Fn 1)

F = 1 + (F1 1) + + + ... (10.34)
GA1 GA1 GA2 GA1 GA2 . . . Gn 1

Equations 10.31 through 10.34 for the cascaded noise temperature and noise factor are
known as Friis formulas.
10.2. NOISE CHARACTERIZATION OF LINEAR 2-PORTS 337

10.2.4.1 Example - Noise temperature of cascaded amplifiers.

Find the eective input temperature, noise factor, and noise figure for the system shown in
Figure 10.14.

ZS 1 2

Te1 = 100 K Te2 = 1000 K

GA1 = 10 GA2 = 10

Figure 10.14: Cascaded 2-ports.

1000
Te12 = 100 + = 200 K (10.35)
10
200
F12 = 1+ = 1.69
290

N F12 = 10 log 1.69 = 2.3 dB

What if the order of cascade is reversed? Assuming that the available gains, and eective
input temperatures do not change, then
100
Te21 = 1000 + = 1010 K (10.36)
10

F21 = 4.48

N F21 = 6.5 dB

This example illustrates that the order of cascade is important and also raises a question:
for minimum cascaded noise temperature, should the 2-port with the lowest Te (F) always
come first in the cascade? The answer is no. The order of cascade that gives the best
noise figure depends on the relative gains of the 2-ports, as well as their eective input
temperatures. Interested readers may want to derive a formula for a figure of merit for a
2- port that depends on both the gain and eective input temperature. The figure of merit
should be such that when cascading two 2-ports, the cascade has the lowest eective input
temperature when the 2-port with the smaller figure of merit is used first. (The resulting
figure of merit is sometimes called the Noise Measure of a 2-port. See homework problem
4.)

10.2.4.2 Example - Noise temperature of attenuator-amplifier cascade.

Find the eective input temperature and NF of the cascade in Figure 10.15. First, find Te
for the attenuator. A 3 dB attenuator has loss L = 103/10 ' 2, thus

Te1 = To (L 1) = 290(2 1) = 290 (10.37)

338 CHAPTER 10. NOISE IN 1- AND 2-PORTS

Passive
Amplifier
Attenuator

Tatt = To Te = 100 K
L = 3 dB GA2 = 15 dB

Figure 10.15: Attenuator-amplifier cascade.

The gain of the attenuator is

1 1
GA1 = = (10.38)
L 2
Te2
. . . Te = Te1 +
GA1
100
= 290 + = 290 + 200 = 490 K
1/2
Te 490
F = 1+ =1 + = 2.69
To 290

NF = 4.3 dB (10.39)
The NF of the amplifier was 1.3 dB, so adding the attenuator significantly increased the
NF. In fact, the NF was increased by exactly 3 dB (the loss of the attenuator). The increase
would not have been equal to the loss of the attenuator if the physical temperature of the
attenuator was dierent from 290 K.
This example shows how the presence of loss in front of a low-noise amplifier can seriously
degrade the overall noise temperature of the system. The example can also be worked using
the noise factor. Recall that for an attenuator at standard temperature To = 290 K
F1 = L = 2 (10.40)
and
Te2 100
F2 = 1+ =1 + = 1.34 (10.41)
To 290
1 1
GA1 = =
L 2
(F2 1) .34
F = F1 + =2 + = 2.68
GA1 .5
This is the same (to within roundo error) as the answer obtained using noise temperatures.
The preceeding example illustrates a special result that is valid for room temperature
attenuation ahead of a 2-port with N F = N F2 , as shown in Figure 10.16.

N F = L(dB) + N F2 (10.42)
10.3. SENSITIVITY OF A RECEIVING SYSTEM 339

Passive
Amplifier
Attenuator

Tatt = To Noise Figure=N F2 dB

Loss=L dB

Figure 10.16: An attenuator in front of a 2-port.

i.e., the loss of the room temperature attenuator (in dB) adds directly to the NF of the
second stage.

10.3 Sensitivity of a Receiving System

Sensitivity is usually specified in terms of a noise floor or, equivalently, the minimum input
signal level required to give a specified SNR at the input to the demodulator. This signal
level is called the minimum discernible signal, or MDS, and is specified in terms of the input
power level, usually in dBm. Calculation of the SNR requires knowledge of the total noise
power available from the output of the system. To facilitate calculation of the noise power,
it is convenient to use the concept of equivalent noise bandwidth which is defined next.

10.3.1 Equivalent Noise Bandwidth

The available power gain of a real system may look like Figure 10.17 when plotted as a
function of frequency. Usually the IF stages of a receiver will set the overall bandwidth, so
the shape of this curve will reflect the shape of the IF filters response function.

Ga(f )

Gao

Figure 10.17: Overall available power gain of a receiver plotted as a function of frequency.
340 CHAPTER 10. NOISE IN 1- AND 2-PORTS

The available noise power at the output of the receiver will be

Z 1
Navo = k Top Ga (f ) df (10.43)
0
Z 1
Ga (f )
= k Top Gao df
0 Gao

Equation 10.43 has the form kTBG if the equivalent noise bandwidth, Bn is defined as
follows:
Z 1
Ga (f )
Bn = df (10.44)
0 Gao

Then Equation 10.43 becomes

Nout = k Top Gao Bn (10.45)

In essence, by defining the equivalent noise bandwidth, the actual gain versus frequency
curve is replaced with an idealized, rectangular response function as seen in Figure 10.18,
where the area under the equivalent rectangular response is equal to the area under the
actual response function.

Ga(f )
Idealized Response
Gao

Actual Response

f
Bn

Figure 10.18: Idealized rectangular response function.

In general, the noise bandwidth of a system will be larger than the 3 dB bandwidth.
For example, it can be shown that for circuits whose frequency responses are determined by
simple series or parallel RLC circuits, the 3 dB bandwidth and Bn are related by:

Bn = B3dB ' 1.6B3dB
2

10.3.2 Noise Floor, or Minimum Discernible Signal (MDS)

As mentioned earlier, the noise floor or MDS is defined to be the input signal level required
to give some specified output SNR. The following example illustrates how the noise floor is
calculated.
10.3. SENSITIVITY OF A RECEIVING SYSTEM 341

10.3.2.1 Example - MDS for a receiving system

Find the noise floor (MDS) for the receiving system in Figure 10.19. Assume that the
minimum required output signal-to-noise ratio that is required for detection of the signal is
0 dB (SN Rout,min = 1). The eective input temperature of the receiver is found as follows:

Ts = 290 K

50 Savo
Vi Receiver
Navo

NF = 8 dB
Bn = 2.1 kHz

Figure 10.19: Receiving system for deriving noise floor (MDS).

NF = 8 dB (10.46)

Te
F = 10N F/10 = 6.31 = 1 +
290

Te = (F 1) 290 = 1540 K
Calculate the operating noise temperature of the system:
Top = 290K + 1540 K = 1830 K (10.47)
Available output noise power:
Navo = k Top Bn Ga (10.48)
Available output signal power:
Savo = Savs Ga (10.49)
Output SNR:
Savo
SN Rout = (10.50)
Navo
Savs Ga
=
k Top Bn Ga
Savs
SN Rout =
k Top Bn
We require SN Rout,min = 1, so the corresponding minimum available input signal power is:
Savs,min = k T opBn (10.51)

23
= (1.38)(10 )(1830)(2.1)(103 )

17
= 5.30(10 ) Watts
342 CHAPTER 10. NOISE IN 1- AND 2-PORTS

In dBm (dB referenced to a milliwatt)

17
5.3 x 10
M DS Savs,min = 10 log (10.52)
10 3

M DS = 132.8 dBm.

Thus, the noise floor or MDS of the receiver is -132.8 dBm. Sometimes the open circuit
antenna voltage is given instead of the input power. Recall that Savs is defined to be the
available signal power from the source. The source (antenna) impedance is 50 , so
2
Vs,min
Savs,min = (10.53)
8Rs
2
Vs,min
=
8 (50)

Vs,min = 0.146 V

This is the peak value of the open circuit antenna voltage. The rms value is

Vs,min(rms) = 0.103 V (10.54)

Note that the output SNR used in the definition of noise floor should be interpreted as
the SNR at the input to the demodulator stage. Nothing has been said about the particular
type of demodulator that is used in the system. In order to specify the SNR that is required
for detection of a signal, it is necessary to have some information about what SNR is required
by the particular demodulator that will be used.

10.3.2.2 Example - MDS for TV receiving system

In Figure 10.20 a 300 antenna is connected to a TV receiver with 300 input impedance.
The eective temperature of the antenna is 1000 K. The noise figure of the receiver is 4 dB.
The eective noise bandwidth is 5 MHz. Find the input signal level (in dBm) required to

TA = 1000 K

Output of last IF stage,

Receiver
to demodulator

N F = 4 dB
Bn = 5 MHz

Figure 10.20: Television receiving system.

give a 15 dB signal-to-noise ratio at the output of the receiver. What is the corresponding
open circuit antenna voltage?
10.3. SENSITIVITY OF A RECEIVING SYSTEM 343

The desired output SNR is: SN Rout ) 15dB ) 1015/10 = 31.62

Savs
SN Rout = (10.55)
k Top Bn

Savs = k Top Bn (SN Rout )

Top = Ts + Te

Find Te from NF:

Te
F = 10N F/10 = 1 + (10.56)
To

F = 2.51 ) Te = 438.5 K

Top = 1000 + 438.5 = 1438.5 K

23
Savs = 1.38 10 (1438.5)(5 106 )(31.62)

12
= 3.14 10 W = 3.14 10 9
mW

= 85 dBm.

Thus, -85 dBm of input signal power is required for 15 dB SNR at the output of the receiver.
The next example calculates the degradation in sensitivity resulting from addition of a
lossy cable between the antenna and the receiver.

10.3.2.3 Example - TV system MDS with a lossy cable

Repeat the calculation for the case where the antenna is connected to the receiver through
a long transmission line having 10 dB loss as in Figure 10.21. Calculate the eective input

TA = 1000 K

Output of last IF stage,

Receiver
to demodulator
Lossy transmission
line; 10 dB loss, N F = 4 dB
physical temp. To Bn = 5 MHz

Figure 10.21: Antenna-cable-receiver.

temperature of the lossy transmission line. First convert loss from dB to power ratio:
L = 1010/10 = 10. Then Te for lossy transmission line is Te = To (L 1) = 2610 K. Eective
344 CHAPTER 10. NOISE IN 1- AND 2-PORTS

input temperature of the transmission line-receiver cascade:

Te2 438.5
Te = Te1 + = 2610 + (10.57)
G1 1/10

= 6995 K (NF = 14dB)

Top = 1000 + 6995 = 7995 K

Savi = k Top Bn (SN Rout )

11
= 1.74 10 W = 1.74 10 8
mW

= 77.6 dBm

The input signal level must be 7.4 dB higher to get the same SNR as in the first example.
The next example illustrates how one can calculate the NF required of a preamp, in
order to set the system MDS to some specified value.

10.3.2.4 Example - Calculating preamp NF required for a specified MDS.

In Figure 10.22 a preamplifier has been added in front of the lossy cable, in an attempt to
make up for the loss in the cable. The preamp has available gain 10 dB, which counteracts
the 10 dB cable loss. The question is, what NF (or Te ) must the preamp have if the overall
system is to have the same MDS as the receiver alone? The NF of this system must be 4

Preamp Receiver

Lossy transmission
GA = 10 dB N F = 4 dB
line; 10 dB loss,
N F =? Bn = 5 MHz
physical temp. To

Figure 10.22: Antenna-preamp-cable-receiver.

dB if it is to be equivalent to the receiver alone.

Preamp Gain = 10 dB ) GA1 = 10

1
Coax Gain = 10 dB ) GA2 = , Te2 = 2610 K
10

Receiver Te3 = 438.5 K

10.4. MEASUREMENT OF NOISE TEMPERATURE 345

We want the overall Te = 438.5 K,

Te2 Te3
Te = Te1 + + (10.58)
G1 G1 G2
Te2 Te3
Te1 = Te
G1 G1 G2

= 438.5 2610/10 438.5/(10)(1/10)

Te1 = 261 K < 0

Te1 came out negative which means that it cant be done! Even a noiseless preamp would
give an overall Te = 699.5 K ) N F = 5.33 dB! It turns out that the gain of the preamp
would have to be at least 12 dB (for a noiseless preamp) and even higher for a real, noisy,
preamp in order to achieve a 4dB NF for the system. This example illustrates that a preamp
must more than just make up for the loss of an attenuator that follows it if the noise figure
of the system is to be preserved.
The following section describes the principles behind practical measurements of noise
temperature.

10.4 Measurement of Noise Temperature

The basic approach used for noise temperature measurements is based on a comparison of
the output noise powers obtained for two dierent eective source temperatures. The ratio
of the two resulting output noise powers can be used to determine Te and, hence, F and
NF. Refer to Figure 10.23.

TH ZS
NH
Te =?
NC
TC ZS

Figure 10.23: Noise temperature measurement with hot and cold loads.

TH = equivalent noise temperature of the hot load (10.59)

TC = equivalent noise temperature of the cold load

NH , NC ) available output noise powers corresponding to TH , TC , respectively, where

NH = k(Te + TH ) Bn GA (10.60)
NC = k(Te + TC ) Bn GA
346 CHAPTER 10. NOISE IN 1- AND 2-PORTS

The ratio NH /NC is called the Y factor and is the parameter that would be measured in
practice:
NH k(Te + TH ) Bn GA
Y = = (10.61)
NC k(Te + TC ) Bn GA
Te + TH
Y =
Te + TC
Since TH and TC are known, it is possible to solve for Te :
T H Y TC
Te = (10.62)
Y 1
Note that measurement of the eective input temperature of a 2-port is possible without
making accurate absolute power measurements. Only the ratio (relative magnitudes) of the
output powers obtained with hot and cold input terminations is needed.
In commercial noise figure meters, the cold load is usually a 50 resistor at room
temperature. A noise diode is turned on for the hot load. The diode generates excess noise
and, for the purposes of the measurement, acts like a resistor at some temperature 290K.
The excess noise ratio is specified for noise diodes is defined by
TH 290
EN R = 10 log (10.63)
290
A typical noise diode ENR is in the neighborhood of 5 dB.
When measuring small NFs the ENR must be known very accurately if precise mea-
surements are necessary. Also, it is important that the source impedance of the noise diode
doesnt change between the o (cold) and on (hot) states, because the eective input tem-
perature and available gain of a 2-port usually depend on the source impedance.

10.4.1 Practical Considerations

In practice the NF measurement set-up looks like Figure 10.24. What is actually measured

Te =?
G =?
Noise Device
Backend NH
Diode Under
Teb , Bn NC
TH , TC Test

Noise Diode
control signal

Figure 10.24: Noise Figure measurement setup.

is the eective temperature (or noise figure) of the cascade:

Teb
Tec = Te + (10.64)
G
10.5. A MODEL FOR THE DEPENDENCE OF TE ON ZS 347

To determine Te , it is necessary to know Teb and G. These parameters can be determined

if the Y-factor for the backend alone is known, as in Figure 10.25.

Noise NH,0
Backend
Diode
Teb , Bn NC,0
TH , TC

Noise Diode
control signal

Figure 10.25: Measuring the eective input temperature of the backend.

To find the Y-factor of the backend, remove the device under test and measure the ratio
of NH,0 and NC,0 where

NH,0 = k(Teb + TH ) Bn Gb (10.65)

NC,0 = k(Teb + TC ) Bn Gb

The Y-factor of the backend

NH,0 Teb + TH
Y0 = = (10.66)
NC,0 Teb + TC
is used to determine the eective input temperature of the backend, Teb . The gain of the
device under test can be determined by noticing that with the DUT in the circuit we measure

NH = k (Tec + TH ) B G Gb (10.67)
NC = k (Tec + TC ) B G Gb

To find G,
NH Tec + TH
= G (10.68)
NH,0 Teb + TH
Since Tec , Teb and TH are known (or calculated), G can be determined. The power ratio
NC /NC,0 could also be used to find G. In practice the measurement of the backend noise
temperature (i.e., measurement of NH,0 , NC,0 ) is done as part of the calibration procedure
and then stored. The gain that is computed using the NF meter is averaged over the eective
noise bandwidth of the backend and is therefore usually dierent from the gain measured
at one frequency using a signal generator.

10.5 A model for the dependence of Te on ZS

The noise generated within a noisy 2-port can be modeled with the equivalent circuit shown
in Figure 10.26 where all noise sources have been removed from the 2-port and replaced
with equivalent noise voltage and noise current generators at the input of the 2-port. The
voltage source accounts for all of the output noise in the case where the input of the 2-port
is terminated with a short circuit and the current source accounts for all of the output noise
when the input of the 2-port is left open-circuited. In general, the current and voltage
348 CHAPTER 10. NOISE IN 1- AND 2-PORTS

-
en +

Noiseless
in
2-port

Figure 10.26: A model for noisy 2-ports.

sources must be assumed to be correlated, i.e. < en in >6= 0, to account for the fact that a
particular noise generation mechanism may produce output noise with both open and short
circuit input terminations.
Up until now, when we have referred to noise voltages and currents, we have implicitly
assumed that en and in were functions of time, i.e. that en and/or in represented the
actual time-domain voltage or current waveforms that would be viewed on a instrument
with some bandwidth, B. For the following discussion, it will be convenient to represent
the bandlimited noise voltage and current with a complex envelope or, in other words, as
noise phasors. For the discussion in this section, let us assume that en (t) and in (t) are the
voltage and current waveforms in a 1 Hz bandwidth centered on some frequency, f . We can
write
en (t) = <(En (t)ej!t )
and
in (t) = <(In (t)ej!t )
where En (t) and In (t) are complex baseband representations of the bandlimited noise signals
en (t) and in (t). The complex signals En and In are analogous to voltage and current phasors
that we are familiar with. The dierence is that a purely sinusoidal signal, at some frequency,
can be represented by a constant, complex, phasor whose magnitude and phase completely
describe the signal. Similarly, a noise signal that has been bandlimited to a bandwidth 1
Hz, centered on some frequency, can be represented by a slowly varying3 complex phasor.
We can do circuit analysis using noise phasors in the same way that we do circuit analysis
using standard signal phasors. The dierence is that with noise phasors, the quantity of
interest will involve the mean-square value of the noise phasor, since its mean value is always
zero, and is not particularly useful. For example, given the noise phasor En , suppose that
it is necessary to calculate the mean-square value of the corresponding time-domain noise
waveformwe may do this using the following identity:

< |En |2 >

< e2n >=< (<(En ej!t ))2 >=
2

10.5.1 The relationship between Te and input noise voltage and

current.
Consider a noisy 2-port terminated with a noiseless source impedance, ZS , as shown in
Figure 10.27.
3 The time scale for significant variation will be on the order of B 1, or 1 second for B = 1 Hz.
10.5. A MODEL FOR THE DEPENDENCE OF TE ON ZS 349

-
En +

Noiseless
ZS In
2-port

Figure 10.27: Noisy 2-port terminated with noiseless input termination.

For the purpose of deriving an expression for the eective input temperature of the 2-
port, the noise generators can be associated with the source, as shown in Figure 10.28. Note
that we have now switched to representing the noise voltage and current with their complex
noise phasors. The eective input temperature can be found by calculating the noise power
available from the source and then equating the result to kTe .

-
En +

ZS In

Figure 10.28: Equivalent input noise generators lumped into the source for the purpose of
deriving an expression for the eective input temperature of the 2-port.

The available noise power from the source shown in Figure 10.28 is
< |En + In ZS |2 > 1
kTe = = {< |En |2 > +2<{< En In > ZS }+ < |In |2 > |ZS |2 },
8RS 8RS
(10.69)
where the operator <{} extracts the real part of its argument. Define the complex correlation
coecient, :
< En In >
= r +j i = p
< |En |2 >< |In |2 >
and the noise voltage and current variances, 2
e, i:
2

2 < |En |2 > < |In |2 >

e =< |en |2 >= , 2
i =< |in |2 >=
2 2
Then 10.69 can be written
1 2 2 2
kTe = { e + 2<{ ZS } e i + i |ZS | }. (10.70)
4RS
Expand equation 10.70 using ZS = RS + jXS and = r +j i and complete the squares
to show that:
2 2
i e 2 e 2 e
Te = {(RS + r ) + (XS i ) + 2 (1 | |2 )}. (10.71)
4kRS i i i
350 CHAPTER 10. NOISE IN 1- AND 2-PORTS

Equation 10.71 shows how the eective input temperature of a 2-port depends on the source
impedance, ZS . The particular source impedance that minimizes Te is called the optimum
source impedance for minimum noise, Zopt . The corresponding minimum value of Te is
denoted by Te,min . It is not hard to show that
q
e e
Zopt = Ropt + jXopt = 1 2
i +j i (10.72)
i i

q
e i
Te,min = { 1 2
i + r }. (10.73)
2k
When written in terms of Zopt and Te,min , the eective input temperature is:

|ZS Zopt |2
Te = Te,min + To Gn (10.74)
RS

where the parameter Gn is called the noise conductance and is given by

2
i
Gn = . (10.75)
4kTo

For a particular 2-port, the parameters Te,min , Zopt , Gn are constants that completely
characterize the noise performance of the 2-port. The set of 4 parameters {Te,min , Zopt =
Ropt + jXopt , Gn } are called the noise parameters of the 2-port. Equation 10.74 shows that
the eective input temperature of a 2-port increases quadratically as the source impedance
moves away from the optimum value Zopt . The sensitivity of Te to changes in the source
impedance is determined by the magnitude of the noise conductance parameter, Gn .
Equation 10.74 can be written in terms of reflection coecients instead of impedances:
2
| s opt |
Te = Te,min + 4Gn Zo To (1 | 2
s | )|1 opt |
2

(10.76)
2
| opt |
= Te,min + 4R
Zo To (1 |
n s
s|
2 )|1+
opt |
2

where
ZS Zo
S = = actual source reflection coecient
ZS + Zo

Zopt Zo
opt = = optimum source reflection coecient
Zopt + Zo
2
e
Rn = = noise resistance
4kTo
In this case, the noise parameters consist of {Rn , opt , Te,min } or {Gn , opt , Te,min }.
Contours of constant noise temperature are circles in the source reflection coecient
plane. In some cases manufacturers will provide plots of constant noise temperature (or
NF) on the device data sheet. Comparison with constant available power gain curves makes
it possible to select a source impedance that provides the best compromise between gain
and NF.
10.5. A MODEL FOR THE DEPENDENCE OF TE ON ZS 351

10.5.2 Low frequency approximation and op-amp example

At lower frequencies, where op-amps are commonly employed, the noise voltage-current
correlation coecient is often ignored by assuming that it is zero, i.e. = 0, in which case:
e
Zopt = Ropt = (10.77)
i

e i
Te,min = (10.78)
2k
When the correlation coecient is zero, the eective input temperature reduces to:
2 2
i RS
Te = 4k + 4kRS
e

(10.79)
1 RS Ropt
= 2 Te,min [ Ropt + RS ]

Notice that the voltage noise dominates when RS is small (when RS < Ropt ) and the
current noise dominates when RS is large (when RS > Ropt ). Manufacturers will provide
measurements of i and e on the device data sheet. The rms noise voltage, e , is estimated
from noise measurements taken with a very small source impedance and the rms noise current
is estimated from noise measurements taken with a large value of RS . As an example, the
following data was taken from the data sheet for a low-noise op-amp. At 1 kHz, the rms
noise voltage and current are given as:
p
e = 2.3 nV/ Hz
p
i = 160 fA/ Hz

The optimum source resistance for minimum noise and the associated minimum eective
input temperature predicted by equations 10.77 and 10.78 are:

Ropt ' 14.4 k

Te,min ' 13.3 K.

The minimum noise figure is therefore
13.3
NFmin = 10 log(1 + ) ' 0.2 dB.
290
It is interesting to calculate the noise temperature of this device when operated with a lower
source impedance, e.g. for RS = 600 :
1 0.6 14.4
Te = 13.3[ + ] ' 160 K.
2 14.4 0.6
In this case, operating with a source impedance of 600 results in a noise temperature that
is more than 10 times as large as the noise temperature with the optimum source impedance.
352 CHAPTER 10. NOISE IN 1- AND 2-PORTS

10.6 References
1. Krauss, H. L., C. W. Bostian, and F. H. Raab, Solid State Radio Engineering, John
Wiley & Sons, New York, 1980.

2. Smith, Jack, Modern Communications Circuits, McGraw Hill, 1986.

3. van der Ziel, Aldert, Noise in Measurements, John Wiley & Sons, New York, 1976.
4. van der Ziel, Aldert, Noise in Solid State Devices and Circuits, John Wiley & Sons,
New York, 1986.

5. Vendelin, George D., Anthony M. Pavio, Ulrich L. Rohde, Microwave Circuit Design
Using Linear and Nonlinear Techniques, John Wiley & Sons, 1990.
10.7. HOMEWORK PROBLEMS 353

10.7 Homework Problems

1. The input of a 2-port is terminated with a 50 resistor that is at room temperature
(290K), and it is found that the noise power delivered to a 50 resistor at the output
of the 2-port is 10 15 W. When the input resistor is then cooled down to 77 K (the
temperature of liquid nitrogen), the output noise power drops by 2.5 dB.

(a) Find the eective input temperature of the 2-port. You may assume that the
eective input temperature is constant over the range of frequencies where the
2-port has appreciable gain.
(b) How accurately must the change in output noise powers be measured (in dB), if
the eective input temperature of the 2-port is to be measured to within plus or
minus 1K?
(c) Suppose that your measurement accuracy is plus or minus 0.1 dB. What is the
uncertainty in the eective input temperature measurement?

2. A receiver has noise bandwidth Bn = 5.0 kHz. The receiver has a built-in meter that
gives a reading proportional to the total power at the output of the last IF stage
(just before the detector). Note that the meter will indicate the total (signal + noise)
power.
Suppose that you connect a signal generator with eective source temperature Ts =
290K to the input of the receiver. The meter reading is found to increase by 2 dB
when the available signal power from the signal generator is raised from 0 (no output)
to -130 dBm. You may assume that the noise power available from the generator is
constant, and independent of the available signal power.
What is the eective input temperature of the receiver?

3. You are given two amplifiers with the following characteristics:

#1 G1 = 4 dB (10.80)
Te1 = 175 K

#2 G2 = 16 dB (10.81)
Te2 = 200 K

where G1 and G2 are the available gains of the amplifiers and Te1 , Te2 are the eective
input temperatures. In what order should they be cascaded so that the cascade has
the minimum possible noise figure? For the optimum configuration, give the eective
input temperature, noise factor, and noise figure.

4. Consider a cascade of identical amplifiers with available gain G and input eective
input temperature Te .

(a) The noise measure of an amplifier is, by definition, the eective input temperature
of an infinite number of the amplifiers in cascade. Find an expression for the noise
measure. Give your final result in closed form, i.e., do not leave your result in
terms of an infinite series.
354 CHAPTER 10. NOISE IN 1- AND 2-PORTS

(b) Can the noise measure concept be applied to a 2-port with available gain that is
less than 1?
(c) Suppose one is confronted with a situation like that in problem 3. That is, you are
given two amplifiers with eective input temperatures Te1 and Te2 and available
gains G1 and G2 , respectively. Show that in order to achieve the lowest possible
noise temperature, they should always be cascaded such that the amplifier with
the lowest noise measure is in front.

5. Suppose that it is necessary to decide how to cascade N 2-ports with eective input
temperatures Ti and available gains Gi (> 1), so that the cascade has the minimum
possible eective input temperature. Explain how noise measure can be used to decide
how to cascade the 2-ports, and why it is more ecient to use the noise measure rather
than simply evaluating the cascaded noise temperature of all of the possible cascades.

6. The following situation is often encountered in practice: An antenna is located some

distance from the receiver. A decision must be made whether to place the preampli-
fier at the antenna (before the lossy cable) or at the receiver (after the lossy cable).
Suppose the distance between antenna and receiver is 100 meters. The connection will
be made with a 100 meter length of coaxial cable having an attenuation of .05 dB per
meter. The physical temperature of the cable will be 290K. A preamplifier is obtained
that has an available gain of 16 dB and a NF of 0.50 dB.

(a) Find the eective input temperature of the cable followed by the preamp.
(b) Find the eective input temperature of the preamp followed by the cable.

7. The front end of a particular receiver consists of the following stages:

Preamp:

Ga = 10 dB (10.82)
NF = 5 dB

Mixer/LO-IF:

Ga = 60 dB (10.83)
NF = 11 dB

(a) Suppose the preamp has an input impedance of 50 and is connected to a 50

antenna with eective temperature 100K. If the equivalent noise bandwidth of
the system is 5 kHz, find the minimum input signal level (in dBm) required to
give a 15 dB SNR at the output of the system.
(b) Find the rms open-circuit antenna voltage (in microvolts) that will give the re-
quired SNR at the output of the system.
(c) Now assume that the preamp noise figure is only 1 dB and repeat part 7a.

8. Consider a receiving system designed to detect signals from a deep-space probe. Sup-
pose that the probe emits a signal with bandwidth of 100 Hz and that the receiving
antenna collects enough energy to make a total signal power of -150 dBm available
10.7. HOMEWORK PROBLEMS 355

from the antenna. The eective antenna temperature is 30K. A preamplifier with eec-
tive input temperature of 20K and available gain of 10 dB is mounted at the antenna
terminals. The output of the preamplifier is connected to the input of the receiver
through a cable with 6 dB of loss. (The physical temperature of the cable is 290K.)
Assume that all ports are conjugately matched, and find the maximum eective input
temperature that the receiver can have, if the signal-to-noise ratio at the detector is
to be at least 3 dB. Assume that the eective noise bandwidth of the receiver is 100
Hz.
9. A spectrum analyzer is a super-heterodyne receiver with a swept LO which displays
the eective input power delivered to its input terminals. The analyzer computes the
eective input power by measuring the total (signal+noise) power delivered by the last
IF stage and dividing this power by the total power gain of all the preceding stages.
Suppose that you want to measure the eective input temperature of a 2-port using a
spectrum analyzer and you know that the available gain of the 2-port is 25 dB. You
may assume that the input and output of the 2-port are conjugately matched when
terminated in 50 , and that the input impedance of the spectrum analyzer is 50 .
With a room temperature (290K) 50 load connected directly to the input of the
spectrum analyzer, the analyzer indicates that the eective input noise power is -168
dBm in a bandwidth of 1 Hz centered on the frequency of interest. You now connect
the 50 load to the input of the 2-port and connect the output of the 2-port to
the spectrum analyzer. With this configuration you find that the equivalent input
noise power in a 1 Hz bandwidth measured by the spectrum analyzer is -140 dBm.
Remember that the number displayed by the spectrum analyzer is the equivalent input
power at the analyzers input port and not at the input of the 2-port under test.
(a) What is the eective input temperature of the spectrum analyzer?
(b) Find the eective input temperature of the 2-port.
10. The signal distribution network for cable television systems consists of cascaded sec-
tions of lossy transmission line and amplifiers. Suppose that the lossy cable and am-
plifiers are both matched to 75. Each lossy transmission line has loss 0.05 dB/meter,
length 500 meters and physical temperature 290K. Each amplifier has available
gain 25dB and eective input temperature Te = 100K. Note that the gain of the am-
plifier is just enough to make up for the loss in the cable, i.e., the cascade of a lossy
t-line and an amplifier results in a new 2-port with 0dB gain.
Now suppose that it is necesssary to distribute a television signal from a main distri-
bution station to a remote site. The site is 1.5km from the main station, so it will be
necessary to use 3 sections of t-line and 3 amplifiers.
(a) Suppose that the t-lines and amplifiers are cascaded in the order:

t line ) amplifier ) t line ) amplifier ) t line ) amplifier

Find the eective input temperature of the cascade of all of the amplifiers and
cables.
(b) Suppose that the t-lines and amplifiers are cascaded in the order:

amplifier ) t line ) amplifier ) t line ) amplifier ) t line

356 CHAPTER 10. NOISE IN 1- AND 2-PORTS

Find the eective input temperature of the cascade of all of the amplifiers and
cables.
(c) Suppose that the source at the main station has equivalent source temperature
TS = 290K. Assuming that the available signal power and system bandwidth is
the same in both cases, specify the improvement (in dB) that would result in the
output SNR if cascade (10a) were changed to cascade (10b).
11. Find the equivalent noise bandwidth, Bn , in terms of the -3 dB bandwidth, B3dB , for
a system with available gain function equivalent to a lowpass Butterworth response
with order n. In other words, assume that the available gain function has the following
form:
1
GA (f ) = f
1 + ( B3dB )2n
Give results for n = 1, 2, 3, 4.
12. Derive equations 10.76 starting with equation 10.74.
Chapter 11

Mixers

Mixers may be classified according to whether they are based on active or passive devices.
Another distinction, that can apply to either passive or active mixers, is whether mixing
occurs as a result of a soft nonlinearity such as the current-voltage relationship in a diode or
transistor, or whether mixing results from a hard nonlinearity such as from a switch. Most
mixers in use today are of the switching type, whereby diodes or transistors are used to
switch the connection between the RF input and the IF output at a rate that is controlled
by the local oscillator. We will give a brief overview of several representative mixer circuits
in this chapter.

11.1 Mixers Based on Gradual Nonlinearities

11.1.1 Single-ended BJT Mixer
The single-ended BJT mixer makes use of the nonlinear relationship between the base-
emitter voltage and the collector current. If the RF and LO voltages are applied to the base
of the BJT so that the base-emitter voltage vi (t) has both RF and LO signal components
then the collector current will contain terms that are proportional to all powers of the total
input voltage:
iC = iDC + k1 vi (t) + k2 vi2 (t) + k3 vi3 (t) + . . . (11.1)
In general, this produces all possible mixing products, including:

fRF , fLO , |fRF fLO |, |2fRF fLO |, |2fLO fRF |, . . . (11.2)

The output voltage can be developed by loading the collector with a parallel resonant circuit
so that a significant output voltage is developed by only one of the frequency components
present in the collector current waveform. The collector circuit is usually tuned to resonate
at either fRF + fLO or |fRF fLO |. Figure 11.1 is an example of a single-ended BJT mixer.
In a practical implementation of the mixer shown in Figure 11.1, a matching network
would be used between the RF source and the RF input in order to couple the maximum
amount of RF signal to the base of the transistor. The matching network would be designed
to present a high impedance to the LO signal. Similarly, a matching network would be used
between the LO source and the LO input, and this network would be designed to present a
high impedance to the RF source. In this way, the LO source is prevented from loading the

357
358 CHAPTER 11. MIXERS

Vcc
n:1 vIF

CLO
vLO

vRF
CRF

Figure 11.1: BJT mixer

RF source, and vice versa. A simpler compromise is to supply the LO voltage through a
small-value coupling capacitor (CLO ) such that the LO source does not significantly disturb
the matching for the RF signal. For this circuit the local oscillator drive level required is on
the order of 100mV. A BJT mixer of this type has conversion gain ' gm RL where RL is the
load resistance seen by the collector at fIF . This circuit is attractive because of its simplicity,
and it has been used in inexpensive mass-produced receivers for consumer applications. It
is not a high performance mixer, however. In general, a single-ended BJT mixer will have a
higher noise figure than a mixer that employs an FET, or a properly designed passive mixer
based on a diode bridge. In addition, BJT mixers are subject to severe intermodulation
distortion. To avoid this problem, vRF must be kept smaller than 10 mV. In general,
BJT mixers of this type have poor large signal characteristics.

11.1.2 Single-ended FET Mixers

If a FET is operated in its constant current region, then
vgs 2
iD = IDSS (1 ) (11.3)
Vp

The circuit is similar to the one shown in Figure 11.1 for the BJT mixer, but the FET has
two advantages over the BJT. These are:
a much lower third-order IMD, since there is no cubic term in the iD versus vgs
relationship
Much higher RF input voltages are usable, i.e., up to at least 100 mV.
A disadvantage of the FET mixer is a somewhat smaller conversion gain than the BJT
circuit.
A major drawback of the single-ended BJT and FET mixers is the presence of a local
oscillator component as well as an RF component at the output of the mixer. Although the
LO and RF signals will be filtered by the tuned output circuit, some of the relatively large
LO signal will inevitably leak through the mixer to the IF stages. A large LO component
11.1. MIXERS BASED ON GRADUAL NONLINEARITIES 359

at the input to the IF stages is obviously undesirable, since it can cause overload at the
IF stages. It can also create undesirable eects such as gain compression and possible
generation of intermodulation products in the IF stages.

11.1.3 Balanced Mixers

A balanced configuration can be used to eectively remove the LO and/or RF signals from
the output of the mixer. One type of balanced mixer often used in integrated circuits is the
dierential-pair multiplier as shown in Figure 11.2.
The multiplier is based on a dierential amplifier. For a single transistor biased in the

iC1 iC2

+
vi
-

IEE

Figure 11.2: A dierential-pair amplifier. The transistors are assumed to be biased in the
active region. The bias network is not shown.

active mode

iC = IS evBE /VT . (11.4)

For the emitter-coupled pair with identical transistors (IS1 = IS2 , VT 1 = VT 2 )
iC1
= evi /VT (11.5)
iC2
where
vi = vBE1 vBE2
Since the emitters are connected to a constant current source,
iE1 + iE2 = IEE (11.6)
or
1
(iC1 + iC2 ) = IEE (11.7)

From Equation 11.5
iC1 = iC2 evi /VT (11.8)
so using Equation 11.7
IEE
iC1 = (11.9)
1 + e vi /VT
IEE
iC2 = (11.10)
1 + evi /VT
360 CHAPTER 11. MIXERS

If the output is taken between the collectors of the two transistors as in Figure 11.4, then
the output voltage will be proportional to the dierence of the emitter currents:
evi /2VT e vi /2VT
iC1 iC2 = IEE (11.11)
evi /2VT + e vi /2VT

vi
iC = IEE tanh
2VT
A plot of the current dierence, iC = iC1 iC2 , is shown in Figure 11.3. If the input

Vi
-VT VT

Figure 11.3: Current dierence

signal levels are small, i.e., if vi

2VT 1, then the hyperbolic tangent can be approximated as
follows:
vi vi
tanh ' (11.12)
2VT 2VT
Thus
vi
iC ' IEE (11.13)
2VT
Note that IEE multiplies the input signal in equation 11.13. If a second signal, ii2 , is
added to IEE then the output will contain a term that is proportional to vi ii2 , i.e. let
IEE ! IEE + ii2 , the the output is

iC ' (IEE + ii2 )vi (11.14)
2VT
Figure 11.4 shows a balanced mixer based on the dierential amplifier. The output voltage
is vo = ic R. The tail current for the dierential pair is IEE + ii2 , where ii2 represents the
time-varying current that results from the input voltage signal vi2 .
According to equation 11.14, the multiplier based on the dierential pair yields an output
that is proportional to (K + vi )vi2 . Hence the output contains the desired product term
plus the input signal vi2 . Suppose that a balanced multiplier of the type shown in Figure
11.4 is used with the RF signal driving the dierential-amplifier input (vi ! vRF ) and the
LO signal driving the tail-current source (vi2 ! vLO ). In that case the output will be
proportional to vRF (K + vLO ) = KvRF + vRF vLO . Notice that the RF signal is present
in the output, along with the desired product signal. On the other hand, if vi ! vLO and
vi2 ! vRF , then the output is proportional to vLO (K + vLO ) = KvLO + vRF , and the LO
signal accompanies the desired product signal.
The unwanted term can be eliminated as followssuppose that we construct two iden-
tical multipliers, and drive the dierential inputs of each amplifier with vi . Drive the tail-
current input of amplifier 1 with vi2 and the tail-current amplifier of multiplier 2 with vi2 .
11.2. MIXERS BASED ON SWITCHES 361

VCC

R R
iC1 iC2
vo
vi1

IEE + ii2
vi2

VEE

Figure 11.4: Dierential-pair multiplier circuit.

The output of multiplier 1 is then proportional to Kvi2 + vi vi2 , whereas the output of multi-
plier 2 is proportional to Kvi2 vi vi2 . The dierence between the two outputs will contain
only a term proportional to vi vi2 because the unwanted term will be canceled out. Multi-
pliers based on this principle are called Gilbert cell multipliers, and are named after Barrie
Gilbert who described a practical, and precise, multiplier circuit based on the technique.1
Notice that in order to obtain an output that contained only the desired product term, and
no RF or LO term, it is necessary to use a double-balanced configuration, where both the
LO and RF inputs are applied dierentially.

11.2 Mixers Based on Switches

Most mixers used today are based on switches. The basic idea can be illustrated using a
very simple circuit as in Figure 11.5. Suppose that the switch is closed when vLO > 0 and
open when vLO < 0. In addition, let vRF (t) = VRF cos !RF t and vLO (t) = VLO cos !LO t,
then
vo (t) = VRF cos !RF t p(t) (11.15)
where p(t) is a switching function. The switching function for this circuit is shown in Figure
11.6.
The switching function can be expanded in a Fourier series:

X1
1 sin n/2
p(t) = + cos n !LO t (11.16)
2 n=1
n/2
1 A Precise Four-Quadrant Multiplier with Subnanosecond Response, by Barrie Gilbert, IEEE Journal
of Solid-State Circuits, Vol. SC-3, No. 4, December 1968, p 265.
362 CHAPTER 11. MIXERS

Controlled by vLO

vRF (t) RL vo(t)

Figure 11.5: Switching principle

p(t)
1

t
2/!LO

Figure 11.6: Switching function.

Using the Fourier series respresentation for p(t) the output voltage can be written as
( 1
)
1 X sin n/2
vo (t) = VRF cos !RF t + cos n !LO t (11.17)
2 n=1
n/2

Notice that the coecient sinn/2

n/2
is equal to 0 when n is an even number. Therefore, the
output signal has components at the following frequencies:

fRF , |fRF n fLO | n = 1, 3, 5, . . . (11.18)

A filter can be used to select the desired component which is usually |fRF fLO |. It is also
possible to select |fRF n 3fLO | or higher order terms - this is called harmonic mixing.
There are many ways to implement a switch that is controlled by the local oscillator
signal. Diodes are often used as switches. If a relatively large LO voltage is impressed on
a diode, the diode will be forward biased on positive excursions of the LO signal, and will
be in a low-impedance state for superposed, small RF signals. On negative excursions of
the LO voltage the diode is driven into a high impedance state and behaves like an open
switch. Transistors can also be used as switches, with the LO controlling the bias current
and determining whether the device is biased into a low-impedance state, or is cuto.
The dierential multiplier circuit described in the previous section is often used to im-
plement a switching mixer. Refer back to Figure 11.4 and suppose that vi ! vLO . Allow
the amplitude of vLO to be large compared to VT ; then the tanh v2VLO T
term will be driven
to +1 on positive excursions of vLO and to 1 on the negative excursions. In this mode
of operation, the transistors are driven as switches, as shown in Figure 11.7. The output
voltage for the switch configuration shown is vo = (IDC + iRF )R. When vLO changes
sign, the states of the switches reverse, and the output voltage is vo = +(IDC + iRF )R. The
output can be written as
vo (t) = (IDC + iRF ) R p(t)
11.2. MIXERS BASED ON SWITCHES 363

VCC

R R
i1 + - i2
vo

Closed when Open when

vLO > 0 vLO > 0

IDC + iRF

Figure 11.7: Equivalent circuit for dierential pair when a large LO signal is applied to the
dierential amplifier.

where p(t) is the symmetrical switching function shown in Figure 11.8.

p(t)

t
1

2/!LO

Figure 11.8: Symmetrical switching function

This switching function has the following Fourier series:

X1
sin n/2
p(t) = 2 cos n!LO t. (11.19)
n=1
n/2

Writing iRF (t) = IRF cos !RF t:

1
( )
X sin n/2
vo (t) = (IDC + IRF cos !RF t) 2 cos n!LO t (11.20)
n=1
n/2

The frequency components at the output will be

nfLO , |fRF nfLO | n = 1, 3, 5, . . . (11.21)
Two dierential switching pairs can be combined in a double-balanced Gilbert cell ar-
rangement to cancel the unwanted LO component. This is illustrated in Figure 11.9. Sup-
364 CHAPTER 11. MIXERS

VCC

i1 i2
R + vo - R

IDC + iRF IDC iRF

Figure 11.9: Two dierential switching pairs combined in a double-balanced configuration.

pose that the switches are in the state shown when vLO > 0 and that all switches change
state when vLO < 0. Then, the currents are i1 = IDC + p(t)iRF and i2 = IDC p(t)iRF .
The output voltage is vo = R(i2 i1 ) = 2RiRF (t)p(t), hence the output contains only
components at the frequencies |fRF nfLO |. Mixers of this type are often implemented in
integrated circuits.
The diode-ring double-balanced mixer shown in Figure 11.10a, b is another circuit that
produces an output signal proportional to the product of the RF input signal and a switch-
ing function, i.e. vRF (t)p(t). This mixer is noted for its ability to tolerate large RF signals
with relatively small nonlinear distortion, and for its relatively low noise figure. Its opera-

1
a T2
T1 T2 d vLO
vRF vLO 1 4
4 a b
2 2 3
c
3
b vIF
vIF T1

vRF
(a) (b)

Figure 11.10: Four-diode double-balanced mixer. Figures (a) and (b) are the same circuit.
Figure (b) emphasizes the fact that the mixer is based on a diode ring topology.

tion is easily understood once the operation of the 3-winding transformers (voltage baluns)
is understood. For simplicity, lets assume that the transformers are ideal 3-winding trans-
11.2. MIXERS BASED ON SWITCHES 365

formers. If the voltage across the windings is denoted by VA , VB , and VC , with polarity
indicated by the dots, and the current in each winding is denoted by iA , iB , and iC , with
positive current defined to flow into the dot, then an ideal 3-winding transformer satisfies

V A = VB = V B , iA + iB + iC = 0.

Physically, these relationships arise because when three windings with the same number of
turns are tightly coupled, e.g. by winding all three on a high-permability ferrite toroid,
then the magnetic flux through all 3 windings will be the same and the emf induced in each
winding will be the same. Since the ideal transformer is a lossless device, power conservation
requires that the sum of the current flowing out of (or into) the dots of all windings must
be zero. In the context of the circuit described here, if current i flows into the dot of the
primary (driven) winding, the sum of the currents flowing out of the dots of the secondary
windings must be i.
We can now describe the operation of the 4-diode ring mixer. The three-winding trans-
formers are assumed to be ideal, hence the voltage impressed across the primary will be
reflected across both secondary windings, with polarity indicated by the dots. Refering to
Figure 11.10b, when vLO is positive, the voltage at the top of T 2s secondary is positive
and the voltage at the bottom of T 2s secondary is negative. Therefore diodes 1 and 2 are
on and in a low-impedance state as far as the small RF signal is concerned. Diodes 3 and
4 will be o, or in a high-impedance state, so the right-hand side winding of transformer
T 1s secondary is eectively disconnected from the system. Refer to Figure 11.11a where the
forward biased diodes have been replaced with an incremental resistance, rd , and the reverse
biased diodes have been removed, as they are assumed to be eectively open circuits. The
symmetry of the circuit ensures that the voltage vRF across the left-hand winding of T 1s
secondary will drive equal currents through diodes 1 and 2. These currents flow to ground
through the windings of T 20 s secondary. Half of the RF current flows into the dot on the
top of T 2s secondary and the other half flows into the bottom of T 2s secondary. These
currents produce opposing magnetic fluxes and no voltage is developed across the windings
of T 2 secondary. Therefore, the RF voltage at nodes d and c is zero, and nodes c and d are
virtual grounds for the RF signal. The RF voltage applied by the left-hand winding of T 1s
secondary appears across the series combination of the load resistance RL and the parallel
combination of diodes 1 and 2 (rd /2). The voltage across the load is vIF = vRF RLR +rd /2 .
L

When the polarity of vLO reverses so that the voltage at the dots of T 2s secondary is nega-
tive, the situation reverses, i.e. diodes 1 and 2 are o and diodes 3 and 4 are on. Nodes
c and d are still virtual grounds for the RF signal, so the voltage across the load resistor is
vIF = vRF RLR +rd /2 . If the forward resistance of the diodes is small compared to the load
L

resistor (rd RL ), then the output from the double-balanced 4-diode ring mixer can be
written as
vIF (t) = vRF (t)p(t)
where p(t) is the symmetrical switching function shown in Figure 11.8. Since the switching
function has no DC component, there is no direct leakage of the RF signal into the IF
output. Also, by virtue of the balance in the circuit, the voltage at nodes a and b due to the
LO signal is always zero so there is no leakage of the LO signal back into T 1, i.e. LO-RF
isolation is high.
For diode-based switching-type mixers to work as described above, it is important that
the local oscillator signal (vLO ) be large, so that the diodes will be biased to a low impedance
state, and so that it can be assumed that vLO controls the state of the diodes at all times.
366 CHAPTER 11. MIXERS

virtual ground
for RF signal

d T2
1 vLO > 0
a rd
2
c
RL
v +
+ RF-
vIF = vRF RLRL
+r /2 d
T1

(a) + vRF

virtual ground
for RF signal

d T2
3 vLO < 0
rd b

4
c
RL
v +
+ RF -
vIF = vRF RLR+rL /2
d
T1

(b) + vRF

Figure 11.11: 4-diode ring double-balanced mixer equivalent circuit when (a) vLO > 0 and
(b) vLO < 0.
11.3. CONVERSION LOSS IN MIXERS 367

Thus passive mixers require relatively high LO drive levels. Typically, the peak value of the
LO signal must be at least 0.7V in order to turn on the diodes completely. Assuming that
the input impedance seen by the LO source is 50 , the required input level is approximately
7 dBm. The harmonic mixing components, |mfRF nfLO |, can cause spurious responses
in a receiver in addition to the usual image response. These responses must be carefully
considered when designing a receiver. This will be addressed in section 11.4.

11.3 Conversion Loss in Mixers

Conversion loss is defined to be
Pin
LC = 10 log10
Pout
where Pin is the power delivered in to the mixer at the RF frequency, and Pout is IF output
power delived to the load at the desired output frequency (usually |fRF fLO |). Lets
take the double-balanced diode-ring mixer as an example. Assuming that the diode forward
resistance is small compared to the load resistance, the output of the mixer is
( 1
)
X sin n
2
vIF (t) = VRF cos !RF t 2 n cos n!LO t .
n=1 2

If a filter selects just one output component, then the peak amplitude of the output voltage
will be
sin n
2
VIF = VRF n .
2

In an ideal 4-diode mixer, the RF input is always looking through the transformer primary
to one of the secondary windings and through the switches to the load. If the switches are
lossless, then, the RF input impedance will be equal to the load impedance. Hence, the RF
input voltage and the IF output voltage are developed across the same impedance, so the
ratio of Pin and Pout is the same as the ratio (VRF /VIF )2 . The conversion loss, for any n,
is therefore
n
LC = 20 log10 .
2 sin n
2

The conversion loss is tabulated in Table 11.1. When the desired term is |fRF fLO |, the
conversion loss for the ideal switching mixer is 3.9 dB. In practice, losses in the transformer
and diodes usually raise this value to 5-6 dB.

n Ouput Frequency Loss (dB)

1 |fRF fLO | 3.9
3 |fRF 3fLO | 13.5
5 |fRF 5fLO | 17.9

Table 11.1: Conversion loss for ideal double-balanced diode-ring mixer.

368 CHAPTER 11. MIXERS

11.4 Spurious Responses in Receivers - Spur Charts

The idealized models for double-balanced switching mixers predict that the output will
contain components at frequencies given by |fRF nfLO |, with n an odd integer. When
more realistic models for the switches are considered, it is found that harmonics of the RF
signal will be generated within the mixer, and that even-order harmonics of the LO will
be present. A fairly general model for the nonlinear properties of a mixer consists of a
nonlinear 2-port with non-zero coecients k1 , k2 , k3 , ... in front of an ideal multiplier which
is driven by a local oscillator signal containing the fundamental frequency component !LO
in addition to harmonics of the local oscillator signal, 2!LO , 3!LO , .... etc.

P1 m
m=1 km vRF (t)

Nonlinear
vRF (t) |mfRF nfLO |
2-port

P1
n=0 an cos[n!LO t]

Figure 11.12: A model for nonlinearities in a mixer. The multiplier shown in the figure is an
ideal multiplier. The output of the mixer will include terms resulting from mixing between
all harmonics of the RF and LO signals. Note that the n = 0 (DC) component is included
in the output of the LO in order to model the direct leakage of harmonics of VRF through
the mixer to the IF port.

When evaluating a particular receiver design, it is highly desirable to consider the possible
spurious responses which will occur because of the non-ideal characteristics of the mixer.
The goal of this section is to determine the frequencies of all possible input signals which
could be mixed to the IF.
The receiver will be designed such that the desired signal frequency, fD , is related to the
IF, fIF , by one of the following relationships:

fLO = fD + fIF (11.22)

or
fLO = fIF fD (11.23)
or
fLO = fD fIF (11.24)
Equation 11.22 corresponds to the choice of high LO for either the up- or down-
conversion receivers. Equation 11.23 corresponds to low LO for the up-conversion receiver,
and equation 11.24 corresponds to low LO for the down-conversion receiver.
A spurious signal that mixes to fIF satisfies the following equation:

|mfS nfLO | = fIF (11.25)

where we assume that m, n are integers; m > 0, n 0. Notice that by allowing n = 0 we

allow for possible IF responses due to direct leakage of the RF signal (or its harmonics) into
11.4. SPURIOUS RESPONSES IN RECEIVERS - SPUR CHARTS 369

the IF passband. Taking account of the absolute value sign, equation 11.25 can be expanded
into 3 equations:
mfS + nfLO = fIF

mfS nfLO = fIF (11.26)

mfS nfLO = fIF

Now, depending on the receiver configuration of interest, one of equations 11.22, 11.23, or
11.24 may be used in equations 11.26 to eliminate fLO and to solve for the frequency of
possible spurious signals as a function of the receiver tuning, represented by the desired
signal frequency fD . Notice that in this context, fD may be interpreted as the receivers
dial setting since it is the number that the receiver display will indicate. In some literature
fD is referred to as the tuned frequency, since it is the intended frequency that the receiver
is tuned to receive.
Suppose that the case of interest is high LO. Then, using equation 11.22 in 11.26 we
find:
mfS = (1 n)fIF nfD

mfS = (1 + n)fIF + nfD (11.27)

mfS = (n 1)fIF + nfD

Since the right-hand side of the first and third equations in 11.27 are the same except for a
factor of -1, and we are interested in positive spurious frequencies, the first equation is not
necessary and the relevant equations are:

mfS = (n 1)fIF + nfD

(11.28)
mfS = (n + 1)fIF + nfD

It is convenient to normalize all frequencies by the IF, i.e. define:

fS fD
S , and D
fIF fIF

then we obtain the equations used to generate a so-called universal spur chart for any
receiver employing high LO:
S = nm1 + m n
D
(11.29)
S = n+1 m + n
m D
For each value of m and n these equations define two lines, each of which determines the
normalized spur frequency, S, for a particular normalized tuning frequency D. The case
m = 1, n = 1 gives the lines S = D and S = D + 2 which correspond to the desired
frequency and the regular image, respectively. These responses were discussed already in
Chapter 3. The other lines correspond to spurious responses that result from the non-ideal
nature of the mixer.
The normalized spur frequency is plotted in Figure 11.13 for all possibilities up through
fifth order where the order of a spur is defined to be the value of n + m. The m, n values
for each curve are plotted on each line. The part of the plot where 0 < D < 1 corresponds
370 CHAPTER 11. MIXERS

Figure 11.13: Universal spur chart for high LO. The numbers on each line (m, n) indicate
the multiple of the RF and LO frequencies, respectively that are summed or dierenced to
produce an output at the IF. Only responses up through 5th order (m + n 5) are shown.
Normalized tuning frequencies greater than 1.0 correspond to downconversion and tuning
frequencies less than 1.0 correspond to upconversion.
11.4. SPURIOUS RESPONSES IN RECEIVERS - SPUR CHARTS 371

to tuning frequencies that are smaller than the IF, (D = fD /fIF < 1) and, hence, to
up-conversion. The part of the plot where D > 1 corresponds to down-conversion.
Universal spur charts may also be produced for up-conversion with low LO and down-
conversion with low-LO. It turns out that the following normalized equations can be used
for both of these cases. When D < 1 the result corresponds to up-conversion/low-LO and
with D > 1 the result corresponds to down-conversion/low-LO:

S = | 1mn + n
m D|
(11.30)
S = | 1+n
m
n
m D|

The universal spur chart for low-LO is shown in Figure 11.14.

Figure 11.14: Universal spur chart for low LO. Only responses up through 5th order are
shown. The numbers on each line (m, n) indicate the multiple of the RF and LO frequencies,
respectively that are summed or dierenced to produce an output at the IF. Normalized
tuning frequencies above 1.0 correspond to downconversion, and tuning frequencies below
1.0 correspond to upconversion.
372 CHAPTER 11. MIXERS

11.4.1 Crossovers
Notice that both of the universal spur charts exhibit crossovers, i.e. certain lines cross the
m=1, n=1 line which corresponds to S = D. A crossover at some tuned frequency D means
that a signal with frequency S = D will interfere with itself! For example notice that a
crossover occurs in Figure 11.13 when S = D = 3. In this case, the line that crosses the
m = 1, n = 1 line at S = D = 3 has m = 3, n = 2. This means that the 3 harmonic of the
desired signal at frequency S = 3 will mix with the second harmonic of the local oscillator
to give an output at the IF. A numerical example will be provided in the next section. For
now, it is sucient to point out that a crossover frequency represents a potential dead
zone in a receivers tuning range, since it may be impossible to receive strong signals at
this frequency.

11.4.2 Example - AM Broadcast band radio

Figure 11.15 shows a spur chart for a broadcast band AM radio employing high LO. For an
IF of 455 kHz, the normalized lower and upper frequencies of the AM broadcast band are
[540/455, 1700/455]=[1.187,3.736]. These limits are denoted in the figure by solid vertical
and horizontal lines.

11.4.2.1 Radio tuned to receive a signal at 910 kHz:

Suppose the radio is tuned to receive a desired signal with carrier frequency 910 kHz
(D=910/455=2.0). Consider possible interference from signals within the AM band re-
sulting from spurious responses up through 5th order (order = n+m). Examine the vertical
dotted line drawn at the normalized tuned frequency D=2 and notice that for spur frequen-
cies within the AM band there are 6 intersections with the dotted line (not counting the
intersection with the desired n=m=1 line). This means there are 6 possible frequencies that
could cause interference. If spurious response orders higher than 5 are considered, there
would be an even larger number of potential spurious responses.
Suppose it is necessary to determine the exact frequencies of each of the potential spurs.
It is not possible to obtain very accurate results using the chart, but the chart can be used
to provide guidance for determining the numerical values of the spur frequencies. Notice
that (m,n) for the 6 spurs are: (3,1), (3,2), (2,1), (3,2), (2,2), (2,2). Also notice that the
(2,1) spur is a crossover spur. To determine the exact frequencies recall that spurs satisfy:

|mfS nfLO | = fIF

which corresponds to the three equations

mfS + nfLO = fIF (11.31)

mfS nfLO = fIF (11.32)

mfS + nfLO = fIF (11.33)
Since the AM radio uses high LO, equation 11.31 can never be satisfied and is not relevant
here. The other two equations can be re-arranged to yield:

fIF + nfLO
fS = (11.34)
m
11.4. SPURIOUS RESPONSES IN RECEIVERS - SPUR CHARTS 373

Figure 11.15: A portion of the universal spur chart for high-LO relevant to an AM Broadcast-
band radio with 455 kHz IF and high LO.
374 CHAPTER 11. MIXERS

nfLO fIF
fS = (11.35)
m
Now, use the fact that when the radio is tuned to 910 kHz, fLO = 910 + 455 = 1365 kHz
to find the exact frequency of each spur, i.e.:
455 + 1365
(3, 1) ) fS = = 606.67 kHz
3
455 + 2(1365)
(3, 2) ) fS = = 1061.67 kHz
3
2(1365) 455
(3, 2) ) fS = = 758.33 kHz
3
455 + 2(1365)
(2, 2) ) fS = = 1592.50 kHz
2
2(1365) 455
(2, 2) ) fS = = 1137.5 kHz
2
455 + 1365
(2, 1) ) fS = = 910 kHz
2
Notice that the (2,1) spur is a crossover spur, i.e. the spur frequency is equal to the desired
frequency (910 kHz in this case). The interference results from the second harmonic of the
910 kHz signal (at 1820 kHz) mixing with the LO at 1365 kHz to produce an output from
the mixer at 455 kHz. This self-interference would be expected to be most noticeable when
the desired signal at 910 kHz is a very strong signal. It has the somewhat peculiar property
that the stronger the desired signal, the more intense the interference will be, since the
second harmonic will be generated most eciently when the mixer is driven hard.

11.4.2.2 Strong signal at 1000 kHz

Next, suppose that a very strong signal is transmitting at 1000 kHz. As the receiver is
tuned across the band, numerous spurious responses may occur. In this case, the strong
signal at 1000 kHz (S=1000/455=2.1978) is represented by a horizontal dotted line. Notice
that there are 4 possible tuned frequencies where spurs from the signal at 1000 kHz would
be potentially observed. (The intersection between the horizontal dotted line and the (1,1)
line is not counted, since that represents the case where the receiver is tuned to receive 1000
kHz.) The 4 spurious responses correspond to (2,2), (3,2), (2,1), (3,2). Equations 11.32
and 11.33 can be used along with the fact that fD = fLO fIF to solve for the tuning dial
setting at which the spurious responses would occur:
m 1
fD = fS fIF ( + 1) (11.36)
n n
and
m 1
fD = fS + fIF ( 1). (11.37)
n n
Using fS = 1000 kHz and fIF = 455 kHz :
1
(2, 2) ) fD = 1000 + 455( ) = 772.5 kHz
2
3 3
(3, 2) ) fD = 1000 455( ) = 817.5 kHz
2 2
11.4. SPURIOUS RESPONSES IN RECEIVERS - SPUR CHARTS 375

3 1
(3, 2) ) fD = 1000 + 455( ) = 1272.5 kHz
2 2
(2, 1) ) fD = 2(1000) 455(2) = 1090 kHz
Thus, spurious responses would potentially be observed at the following dial settings: 772.5
kHz, 817.5 kHz, 1090 kHz, and 1272.5 kHz.
376 CHAPTER 11. MIXERS

11.5 Homework Problems

1. Consider the mixer circuit shown in Figure 11.16:

RS
+
+
vLO (t)
RL vo(t)
+
vRF (t)

Figure 11.16: Mixer circuit

(a) Assume that vLO controls the state of the diodes at all times and that the diodes
are ideal. Derive an expression for the output voltage. Express your result in
terms of a switching function p(t), and give the Fourier cosine series for your
switching function.
(b) What frequency components will appear at the output of this mixer?

2. Consider the mixer circuit shown in Figure 11.17. The transformer is an ideal 3-
winding transformer, which has the property that the voltages across all 3 windings are
equal, with polarity corresponding to the dots shown next to each winding. The time-
varying RF and LO voltages are vRF (t) = VRF cos !RF t, and vLO (t) = VLO cos !LO t.
Assume that vLO (t) controls the state of the diodes at all times, and that the diodes
are ideal switches.

+ vLO (t)
+ -
vRF (t)
+
-
RL vIF (t)
-

Figure 11.17: Mixer circuit

(a) Derive an expression for the IF voltage, vIF (t). Express your result in terms of
a switching function p(t), and give the Fourier cosine series for your switching
function.
(b) List the frequencies of all components that will appear at the output of this mixer.
Chapter 12

Nonlinear Eects in 2-ports

Consider a time-invariant 2-port (Figure 12.1).

Nonlinear
vi(t) vo(t)
2-port

Figure 12.1: Nonlinear 2-port.

For suciently small input signals, the 2-port can often be modeled as a linear 2-port
i.e. Z 1
vo (t) = h( )vi (t )d (12.1)
1

where h( ) is the impulse response of the system. When the bandwidth of vi (t) is much
smaller than the bandwidth of the 2-port, then the impulse response of the 2-port can be
approximated as a delayed delta function, i.e. h(t) ' k (t g ) where k is the small-signal
voltage gain and g is the group-delay of the 2-port. In this case the output waveform will
be a scaled and delayed version of the input signal, i.e.

vo (t) ' kvi (t g ).

Likewise, if vi (t) = s1 (t) + s2 (t), then

vo (t) = ks1 (t g ) + ks2 (t g ) (12.2)

i.e., superposition applies.

In general, for input signals that are not small, the input-output relationship must be
considered to be nonlinear. To illustrate some of the important characteristics of nonlinear
2-ports while keeping the analysis relatively simple, we will ignore the possibility of energy
storage elements (such as inductances and capacitances) within the nonlinear 2-port. This
means we can ignore frequency-dependent phase shifts and, therefore, delays. Such a 2-port
will be frequency independent and is said to be memoryless because lack of energy storage
means that the output can depend only on the present value of the input, and not on past

377
378 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS

values. In many cases, this assumption can be justified by lumping energy storage elements
into external, linear, networks. If the 2-port is time-invariant and memoryless, then the
output signal will be related to the input signal by an nonlinear transfer characteristic, i.e.
vo = f [vi ], where f [ ] is a single-valued nonlinear function. Typically, the nonlinear input-
output characteristic is approximated by a Taylor series in the vicinity of an operating
point, and analysis proceeds based on the truncated Taylor series expansion. For input
signals that are not too large, only a few terms of the Taylor series are sucient. The
Taylor series approach can be viewed as a special case of the more general Volterra series
analysis which is applicable to nonlinear 2-ports with memory.
Denote the nonlinear input-output relationship for a memoryless 2-port as

vo (t) = f [vi (t)], (12.3)

where f [ ] is a single-valued, smooth (dierentiable), nonlinear function. One consequence

of the nonlinear input-output relationship for a memoryless 2-port is that superposition
cannot be applied to determine the output in response to multiple inputs. In this case, if
vi (t) = s1 (t) + s2 (t), then

vo (t) = f [s1 (t) + s2 (t)] (12.4)

6= f [s1 (t)] + f [s2 (t)] (12.5)

i.e., superposition does not apply. As we shall see, when the input signal contains only
one, or a few, sinusoidal components, the output signal may contain many more frequency
components, generated because of the nonlinearity of the 2-port. When generated within
receivers these distortion products can cause interference. In transmitters, nonlinearity in
the power amplifier can cause significant broadening of the signal spectrum, possibly causing
the output spectrum to bleed over into adjacent channels.

12.1 Power series model

For moderately large signals, a memoryless nonlinear input-output characteristic can be
expanded in a Taylor series about some operating, or quiescent, point. Denote the total
input and output signals by xT and yT , respectively so that yT = f [xT ]. The input signal
can be decomposed into quiescent, and time-varying components, i.e.

xT = xQ + x (12.6)

where xQ is the DC component of the input signal and x is a zero-mean, time-varying

component. Similarly, denote the total output by yT where

yT = yQ + y. (12.7)

The first term, yQ , is defined to be the output when x = 0. Then

yT = f [xT ] (12.8)

df 1 d2 f 1 n dn f
= f [xQ ] + x + x2 + ... + x + ...
dx x=xQ 2 dx2 x=xQ n! dxn x=xQ
12.1. POWER SERIES MODEL 379

The part of the output that results from the time-varying part of the input is, therefore, of
the form
y = k1 x + k2 x2 + k3 x3 + . . . (12.9)
where
di f
ki =
dxi x=xq

So-called small-signal models retain only the first term in equation 12.9, which represents a
linear relationship between x and y. This term is dominant when x is small enough so that
higher order terms are negligible. The higher order terms represent nonlinear distortion.
Also, notice that even when x has zero mean, y may not have zero mean since terms which
are even powers of the input signal will have non-zero means; for example, if the input is a
cosine function x = cos !t, the term x2 = cos2 !t = 12 (1 + cos 2!t) is not zero-mean. Hence,
the constant part of the output may contain terms contributed by the zero-mean input
signal, x. This means that the DC bias point of an amplifier will shift when the amplifier is
driven with a time-varying input signal.
To obtain a general understanding of nonlinear eects in 2-ports (without considering
particular circuit configurations), we will assume that the input-output relationship can be
accurately modeled by a power-series expansion

y = k1 x + k2 x2 + k3 x3 + . . . (12.10)

where x represents the input excitation and y is the output quantity. Typically, x and y
will represent input and output voltages or currents. For signals that are not too large
it is sucient to truncate the expansion to terms of third order and lower. Thus for the
subsequent discussion well assume an input-output relationship consisting of a three-term
power series:
y ' k1 x + k2 x2 + k3 x3 (12.11)

12.1.1 Specific Examples - BJT and FET nonlinearities

As an example, consider the input-output relationship for the bipolar junction transistor
(BJT) in Figure 12.2.

C
IB IC
B
+
Vbe- IE
E

Figure 12.2: Bipolar junction transistor (BJT).

IC = IS eVbe /VT (12.12)

and VT = kT
q ' 25 mV at room temperature (290 K). Now suppose that the base-emitter
voltage consists of a DC component (VDC ) and a time-varying signal component (vbe ), i.e.,

Vbe = VDC + vbe (12.13)

380 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS

Then
IC = IS eVDC /VT evbe /VT (12.14)
If vbe /VT < 1, we can expand e vbe /VT
:
" 2 3 #
vbe 1 vbe 1 vbe
IC = IS eVDC /VT 1+ + + + ... (12.15)
VT 2 VT 6 VT

The first term represents the DC component of collector current when vbe = 0. Denote this
quiescent current by ICQ , i.e.:
ICQ = IS eVDC /VT .
Denote the fluctuating component of the collector current by ic , then a three-term power-
series approximation for the nonlinear relationship between iC and vbe is:
ICQ ICQ 2 ICQ 3
iC = vbe + v + v
VT 2VT2 be 6VT3 be
Hence, for the BJT:
ICQ
k1 = VT = gm

ICQ
k2 = 2VT2 (12.16)

ICQ
k3 = 6VT3

The first term is the familiar small-signal approximation; the time-varying part of the col-
lector current is equal to the transconductance, gm , times the time-varying part of the
base-emitter voltage. In this chapter we will be concerned with the implications of the
higher order terms in the input-output characteristic.
It is interesting to compare the BJT input-output characteristic with that for an ideal
field eect transistor (FET) as in Figure 12.3.

D
ID
G
+
VGS -
S

Figure 12.3: Field-eect transistor (FET).

2
VGS
ID = IDSS 1 (12.17)
VP

2 1 2
= IDSS 1 VGS + 2 VGS (12.18)
VP VP
There are no third-order or higher terms in this idealized square-law FET characteristic. In
a more realistic model for the FET such terms would be present.
12.2. SINGLE-TONE INPUT 381

12.2 Single-tone Input

Consider a nonlinear amplifier modeled with a three term power-series:

vo = k1 vi + k2 vi2 + k3 vi3

Suppose the input signal consists of a single tone

vi (t) = a1 cos !1 t (12.19)

Then
vo (t) = k1 a1 cos !1 t + k2 a21 cos2 !1 t + k3 a31 A cos3 !1 t (12.20)
Using trigonometric identities
1
cos2 !t = (1 + cos 2!t) (12.21)
2
1
cos3 !t = (3 cos !t + cos 3!t)
4
The output is written
1
vo (t) = 2
2 k2 a1 + (k1 a1 + 34 k3 a31 ) cos !1 t+
(12.22)
1 2
2 k2 a1 cos 2!1 t + 14 k3 a31 cos 3!1 t

and the input and output spectra are illustrated in Figure 12.4. Since the amplitude of the
second and third harmonics is proportional to a21 and a31 , respectively, this analysis predicts
that for every 1 dB increase of the input signal, the second and third harmonics will increase
by 2 dB and 3 dB, respectively.

Input Spectrum Output Spectrum

f1 f1 2f1 3f1

Figure 12.4: Input and output spectra for single input signal.

12.2.1 Gain Compression

Now consider the output component at the fundamental frequency:

vo0 (t) = (k1 a1 + 34 k3 a31 ) cos !1 t

(12.23)
3 k3 2
= k1 (1 + 4 k1 a1 )a1 cos !1 t
382 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS

The eective voltage gain of the amplifier is

3 k3
Av = k1 (1 + a21 ) (12.24)
4 k1
The term in parenthesis arises because the amplifier exhibits a third-order nonlinearity and
its magnitude depends on the input signal amplitude. Depending on the sign of k3 , as
the input signal amplitude increases the eective gain of the amplifier will either increase
(expansive nonlinearity) or decrease (compressive nonlinearity). If k1 and k3 have the same
sign, then the gain will increase with increasing input amplitude, and if k1 and k3 have
opposite signs, the gain decreases with increasing input amplitude. Our analysis of the BJT
in section 12.1.1 showed that k1 and k3 have the same sign, and hence the nonlinearity is
expansive. The analysis in section 12.1.1 is based on the assumption that the DC component
of Vbe is held constant. In practical amplifier circuits, the transistor will be biased such that
the DC component of the collector current is held more-or-less constant (constant-current
bias), and the DC component of Vbe is allowed to adjust in order to maintain constant
DC collector current. Analysis based on this constraint will show that such an amplifier
exhibits compressive nonlinearity. A detailed analysis of this type is carried out in Appendix
A, where the large signal transconductance is derived and shown to decrease with increasing
base-emitter voltage swing. For now, we shall simply state that in most cases k1 and k3 will
have opposite signs, so that the nonlinearity is compressive. For compressive nonlinearity,
the eective gain can be written as

3 2 k3
k1 1 a1
4 k1
The reduction in gain with increasing input signal amplitude caused by compressive
nonlinearity is called gain compression. The plot in Figure 12.5 shows the output power in
the fundamental component as a function of the input power for an amplifier that exhibits
gain compression.
The gain compression of a 2-port is often characterized by the input power level that
causes the gain to be decreased by 1dB. This parameter is denoted by P1dB in Figure 12.5.

12.3 Two-tone Input

Now suppose that the input signal includes two tones:
vi (t) = a1 cos !1 t + a2 cos !2 t (12.25)
Then

vo (t) = k1 [a1 cos !1 t + a2 cos !2 t] (12.26)

+ k2 [a1 cos !1 t + a2 cos !2 t]2

+ k3 [a1 cos !1 t + a2 cos !2 t]3

Using trigonometric identities the output can be re-written as
vo (t) = first order terms + second order terms + third order terms
12.3. TWO-TONE INPUT 383

Linear
Pout (dBm)

1 dB
Nonlinear

Pin (dBm)
P1dB

Figure 12.5: Output power in fundamental component as a function of input power.

where

first order terms = k1 [a1 cos !1 t + a2 cos !2 t]

second order terms = k2 [ 21 (a21 + a22 ) + 12 a21 cos 2!1 t + 12 a22 cos 2!2 t

+a1 a2 cos(!1 + !2 )t + a1 a2 cos(!1 !2 )t]

third order terms = k3 [( 34 a31 + 32 a1 a22 ) cos !1 t + ( 34 a32 + 32 a21 a2 ) cos !2 t

+ 14 a31 cos 3!1 t + 14 a32 cos 3!2 t

+ 34 a1 a22 (cos(2!2 !1 )t + cos(2!2 + !1 )t)

+ 34 a21 a2 (cos(2!1 !2 )t + cos(2!1 + !2 )t)]

The input and output spectra are illustrated in Figure 12.6.

In narrow-band systems most of these frequency components will be removed by filters
downstream from the nonlinear 2-port, but the so-called in-band third-order terms (2f2 f1 ,
2f1 f2 ) cannot be ignored since they will always have frequencies close to that of f1 and
f2 . In wide-band systems, all of the terms can potentially be significant at the output. In
receivers that use a low-IF, the second-order term at the dierence frequency f2 f1 may be
384 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS

Input Spectrum

f1 f2

Output Spectrum

2f1 f2 2f2 f1 f1 + f2 2f1 + f2 2f2 + f1

f2 f1 f1 f2 2f1 2f2 3f1 3f2

second first order second third

order and third order order order

Figure 12.6: Input (top) and output (bottom) spectra for two input signals.
12.3. TWO-TONE INPUT 385

close to the IF, and is of special concern. For now we will consider only the in-band terms:

3 3 3
vo0 (t) = k1 a1 + k3 a + a1 a22 cos !1 t (12.27)
4 1 2
3
+ k3 a1 a22 cos(2!2 !1 )t
4
3
+ k3 a21 a2 cos(2!1 !2 )t
4

3 3 3 2
+ k1 a2 + k3 a + a a2 cos !2 t
4 2 2 1

There are a number of phenomena that can be illustrated using this result.

12.3.1 Desensitization and Blocking

Suppose a1 cos !1 t is a relatively weak desired signal and a2 cos !2 t is a strong signal at a
nearby frequency that is not of interest. To model this situation we assume that a2 a1 .
Taking into account that a2 a1 , and assuming that the amplifier exhibits compressive
nonlinearity, the amplitude of the desired cos !1 t term is, approximately:

3
k1 a1 + k3 a1 a22 (12.28)
2

k3 3 2
= k1 a1 1 a (12.29)
k1 2 2

Notice that the amplitude of the desired term will be reduced as the amplitude a2 increases.
This means that the presence of a2 cos !2 t causes gain compression for the desired signal,
even though the desired signal may be too weak to cause gain compression by itself. Thus,
an undesired strong signal can desensitize a receiver to the presence of weak signals by
compressing the gain. This phenomenon is called densensitization, or blocking.

12.3.2 Cross modulation

Another undesirable eect can result if both signals are modulated. For example, suppose
that the signals are amplitude modulated, i.e., a1 ) a1 (1 + m1 (t)) and a2 ) a2 (1 + m2 (t))
and that a2 a1 . Then the envelope of the desired signal will be given by

3
= k1 a1 (1 + m1 (t)) + k3 a1 (1 + m1 (t))a22 (1 + m2 (t))2 (12.30)
2

Notice that the signal at frequency !1 now contains amplitude modulation from signal at
frequency !2 . The modulation from the undesired strong signal at !2 has been transferred
to the desired signal at !1 . This phenomenon is called cross modulation. If the signals
are angle modulated, the same mechanism will cause the angle modulation on signal 1 to
contain a term from the angle modulation on signal 2.
386 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS

12.3.3 More than two tones and nonlinear terms with order higher
than 3
So far we have considered only two input signals. If there are more than two input tones,
many more intermodulation products will be generated. In general, for M input tones
({fi }, i = 1, 2, . . . M ) an nth order non-linearity will generate all positive frequencies
given by the magnitude of the sum of n terms of the form fi , i 2 {1, 2, . . . M }. For
example, if there are 3 input signals at frequencies f1 , f2 , f3 the third-order nonlinearity
will produce third-order products at | fi fj fk |, where i, j, k 2 {1, 2, 3}. When
the three input frequencies are closely spaced, the frequencies resulting when two of the
three terms have the same sign will be close to the input frequencies and, hence, are called
in-band intermodulation (IM) products. The in-band third order IM products when there
are three input tones are listed below:

2f1 f2 , 2f2 f1

2f3 f1 , 2f1 f3

2f2 f3 , 2f3 f2

f1 + f2 f3 , f1 + f3 f2 , f2 + f3 f1

f1 , f2 , f3

With three input tones, a 4th order nonlinearity will generate frequencies of the form
| fi fj fk fl | where i, j, k, l 2 {1, 2, 3}, etc. Note that the even-order nonlinearities
will not produce in-band products. Only the odd-order nonlinearities are responsible for in-
band IM products. For example, with two input tones, a 5th order nonlinearity would
produce output components at |f1 + f1 + f1 f2 f2 | = 3f1 2f2 , which is an in-band IM
product.

12.4 Quantitative Characterization of IM Distortion

When multiple input signals are present, nonlinearity causes complex eects. For quanti-
tative comparisons between 2-ports, an idealized situation, or special case, is often used to
provide a common basis for comparisons. The usual conditions for this test are to apply two
input signals with equal amplitudes; this is called a two-tone intermodulation distortion
(IMD) test. The input signals can be written as

vi (t) = a(cos !1 t + cos !2 t) (12.31)

It is assumed that the amplitude of the two signals is small enough so that gain compression
doesnt occur. Then the in-band products at the output of the 2-port are

vo0 (t) = k1 a(cos !1 t + cos !2 t) (12.32)

3
+ k3 a3 (cos(2!2 !1 )t + cos(2!1 !2 )t)
4
12.4. QUANTITATIVE CHARACTERIZATION OF IM DISTORTION 387

Note that the coecient of the fundamental terms is simply k1 which is the result of
ignoring gain compression. Next define the input power of each tone, Pin . For simplicity,
we shall assume that the input impedance of the 2-port is resistive, and equal to the value
Rin . Then
1
Pin = a2 Rin1 (12.33)
2
Assuming that the load impedance is resistive, and denoted by RL , the output power of
each of the desired (fundamental) components is then
1 2 2 1
Pd = k a RL
2 1
The output power of the in-band third-order intermodulation products is
1 2 3 2 6 1
PIM = k ( ) a RL
2 3 4
Notice that Pd will be proportional to Pin , whereas PIM is proportional to Pin3
. Figure 12.7
shows Pd and PIM plotted against Pin on a log-log plot (dBm versus dBm). At the lower
input power levels, PIM increases with a slope of 3 compared to Pd which has a slope of
1. At higher input power, both components will eventually begin to saturate, as shown by
the deviations of the curves from linearity. It should be noted that our simple 3-term power
series model does not predict saturation of the IM products. It is necessary to include higher
order terms to predict this behavior.

(dBm)
(o)
PI

Pd PIM

(i) Pin (dBm)

PI

Figure 12.7: Output power (per tone) in the desired (Pd ) and the IM (PIM ) components
(i)
verses input power per tone. The third order intercept power, PI , is determined by extrap-
olation from measurements at a small input power, where gain compression is negligible.

The fact that PIM increases faster than Pd means that if gain compression did not
eventually occur, the two output curves would ultimately intersect, as shown in Figure 12.7.
388 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS

The input power level at which the linearly extrapolated curves intersect is called the input
(i)
two-tone third-order intercept level and is denoted by PI , or IIP3. The corresponding
(o)
output power level is denoted by PI , or OIP3.
It is important to note that the intercept point is fictitious, since we ignored gain
compression. In actuality, both the PIM and Pd curves would saturate at some finite value.
This is ignored when computing the third-order intercept level. Since we know the slopes
of the ideal PIM and Pd curves, it is a simple matter to measure PIM and Pd at a relatively
low input power level where gain compression is not important. Then the curves can be
extrapolated to the (fictitious) intercept point.
The intercept level is commonly used as a figure of merit for comparing the relative
quality of amplifiers or systems. If it is known, then a system designer can determine
PIM /Pd for any two-tone input level where gain compression can be ignored. Two-ports are
often operated at levels where gain compression is not important but where intermodulation
distortion is still of concern.
Equation 12.34 summarizes the relationship between the input power level and the so-
called intermodulation ratio, or IMR
!2
PIM Pin
IM R = = (i)
(12.34)
Pd PI

(i)
Equation 12.34 can be used to determine PIM /Pd given Pin , if PI is known. This is
useful for system design applications. On the other hand, given measurements of Pd , PIM
(i)
for a known Pin , Equation 12.34 can also be used to find PI .

12.4.1 Example - Calculating IMR

An amplifier has a two-tone third-order input intercept (IIP3) of 20 dBm. What is the
intermodulation ratio (in dB) for a two-tone input level of 0 dBm?
The intermodulation ratio is calculated from:
2
PIM Pin
IM R = = (i)2 (12.35)
Pd PI

or, in dB

(i)
IM R(dB) = 2 Pin (dBm) 2 PI (dBm) (12.36)

= 2 (0) 2 (20)

= 40 dB (12.37)

This means that the IM products (in-band) are 40 dB below the desired signals at the
output, i.e., the output spectrum would look like Figure 12.8.
The next example shows how laboratory measurements of intermodulation products can
be used to calculates the intercept point of a 2-port.
12.4. QUANTITATIVE CHARACTERIZATION OF IM DISTORTION 389

Desired tones

40 dB

In-band IM products

Figure 12.8: Output spectrum showing desired and IM products.

+5 dBm

35 dBm

Figure 12.9: Output spectrum with 2 input signals, Pin = 10 dBm.

(i)
12.4.2 Example - Calculating IIP3 (PI ).
Figure 12.9 shows the spectrum measured at the output of a 2-port when the input consists
of 2 signals with Pin = 10 dBm. The power gain and input intercept of the 2-port
can be found if it is known that gain compression can be neglected when interpreting the
measurements. This can be verified by increasing the power of both input tones by 1 dB, and
verifying that the desired tones at the output also increase by 1 dB, and that the in-band
3rd order intermodulation products each increase by 3 dB. If so, then
G = +5 dBm ( 10 dBm) (12.38)
= 15 dB
The input intercept can be found from
(i)
PIM Pd = 2Pin 2PI (12.39)
where all quantities are expressed in dBm. Solving for the input intercept:
(i) 1
PI = Pin + (Pd PIM ) (12.40)
2
1
= 10 + (5 ( 35))
2

= +10 dBm
390 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS

When performing a measurement like the one illustrated in this example, a typical setup will
use a passive signal combiner to combine the outputs from two signal generators in order
to construct the two-tone signal. It is important to take care to ensure that the combiner
provides significant isolation between the two signal generators, so that intermodulation
products are not generated within the signal generator output stages. It is also important
to verify that intermodulation products generated within the spectrum analyzer can be
neglected.

12.5 Dynamic Range of a Receiving System

The nonlinear eects that have been described in this chapter will determine the largest input
signal level that a receiving system can handle without serious degradation of performance.
The noise figure of the receiver, on the other hand, determines the smallest usable input
signal level. Together the nonlinear eects and the noise floor determine the dynamic range
of a receiver. Consider the receiving system in Figure 12.10.

Ts Receiver

GA, Te ,
( to demodulator
Bn, PI i)

Figure 12.10: Receiving system.

The dynamic range (DR) is defined as follows:

maximum useable input signal power level

DR = (12.41)
minimum useable input signal power level

and is usually expressed in dB, i.e.,

Pmax
DR = 10 log (12.42)
Pmin

The numerator Pmax is the largest usable input signal power level. There are various
definitions for Pmax that depend on what nonlinear eect is being considered by the designer
and what level of degradation is considered unsatisfactory. A commonly employed definition
for the dynamic range is the spurious free dynamic range where the powers Pmax and Pmin
are defined as follows:

Pmin ) input power for specified SN Ro,min at system output (same as MDS)
Pmax ) two-tone input power (per tone) at which SNR of in-band
3rd-order IM products is equal to SN Ro,min at system output.

The minimum usable input signal power is equal to the MDS, which has been defined
previously:
Pmin M DS = k(TS + Te )Bn SN Ro,min .
12.6. INTERCEPT POINT OF A CASCADE 391

The maximum usable input signal power, Pmax , is defined based on the two-tone input
test concept. Suppose that the input consists of two equal amplitude tones. Then Pmax
is the input level that would cause the in-band third-order products to be detectable at
the output of the receiver. In this context detectable means that the signal-to-noise ratio
of the third-order products would be equal to the minimum SNR required for detection,
SN Ro,min .
Assume that Pmin has already been computed. The output signal power that results
from an input power Pmin is just GA Pmin . This output power represents the detection
threshold. Now we ask what two-tone input power is required to cause the in-band third
order products to have an output power equal to the detection threshold. The required
input power is, by definition, Pmax . When the input power, per tone, is Pmax , the output
power in the intermodulation products is GA Pmin . Using equation 12.34:
GA Pmin Pmax
= ( (i) )2 .
Pd P I

Use the fact that Pd = GA Pmax :

Pmin Pmax
= ( (i) )2 .
Pmax P I

Hence 2/3
1/3 (i)
Pmax = Pmin PI (12.43)

Thus, the dynamic range is

2/3
1/3 (i)
Pmin PI
DR = (12.44)
Pmin
!2/3
(i)
PI
=
Pmin

or, in dB
2 (i)
DR = [P Pmin ] (12.45)
3 I
(i)
where PI and Pmin are expressed in dBm. Recall, Pmin is just the noise-floor of the
receiving system and that it depends on the signal-to-noise ratio that is required for the
particular type of signal and demodulator used in the system.

12.6 Intercept Point of a Cascade

Suppose that an amplifier with known input intercept level is cascaded with another amplifier
as in Figure 12.11.
To simplify the analysis, note that the signal level to stage 2 is higher than to stage 1, if
stage 1 has gain. So the nonlinear distortion in amplifier 2 is likely to be more important.
We will ignore distortion generated in the first stage, which is equivalent to assuming that
(i)
stage 1 is a linear preamplifier (equivalent to letting PI1 ! 1).
392 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS

Pin Pin2

GA1, GA2,
(i) (i)
PI1 PI2

Figure 12.11: Cascaded amplifiers

Then, for the second stage alone:

PIM Pin2
= ( (i) )2 (12.46)
Pd PI2
(i)
For the cascade, denote PI as the input intercept (as yet unknown), then
PIM Pin
= ( (i) )2 (12.47)
Pd PI
Rewriting Equation 12.46 using Pin2 = GA1 Pin
PIM (GA1 Pin )2
= 2 (12.48)
Pd PI2
2
Pin
=
(PI2 /GA1 )2
Comparing Equations 12.48 and 12.47 we conclude
(i)
(i) PI2
PI = (12.49)
GA1
The addition of stage 1 reduces the intercept point by a factor equal to the gain of the first
(i)
stage. For example, if PI2 = +6 dBm, GA1 = 10 dB, then the cascade has input intercept
PI = 4 dBm. In the next section, we will show that adding gain in front of an existing
receiver will reduce the system DR.

12.6.1 The eect of adding a preamp to a receiver

Suppose that a receiver has eective input temperature Ter , noise bandwidth Bn , and input
(i)
intercept PI . The dynamic range of the receiver is then
(i)
PI
DRreceiver = [ ]2/3 (12.50)
k(Ts + Ter )Bn SN Ro,min
Suppose that a preamp with eective input temperature Tep and available gain GA is added
before the receiver. Assuming that distortion generated in the preamplifier is negligible, the
new dynamic range is
(i)
PI /GA
DRreceiver with preamp =[ ]2/3 (12.51)
k(Ts + Tep + Ter /GA )Bn SN Ro,min
12.6. INTERCEPT POINT OF A CASCADE 393

Consider the ratioDRwith preamp /DRreceiver :

DRreceiver with preamp Ts + Ter

=[ ]2/3 . (12.52)
DRreceiver GA (Ts + Tep ) + Ter

Now, we ask whether this ratio can exceed unity, i.e. whether adding a preamp can increase
the dynamic range of a receiving system. This requires

Ts + Ter > GA (Ts + Tep ) + Ter (12.53)

Subtracting Ts from both sides of Equation 12.53 results in the following inequality which
must be satisfied, if the addition of a preamp is to increase the DR of a receiver:

Tep < Ts (1 GA )/GA (12.54)

This inequality can never be satisfied if the preamp has available gain > 1, since the RHS
of Equation 12.54 will be negative. The conclusion is that adding a preamp to an existing
receiver will always decrease the DR of the receiver (although it may decrease the noise floor
and therefore improve the sensitivity).
If the available gain of the added stage is less than 1, then the inequality can be satisfied.
For example, suppose that the added stage is a passive attenuator. Then Tep must be
replaced with Tatt ( G1A 1) and the inequality is satisfied if

Tatt < Ts .

Therefore, adding an attenuator in front of an existing receiver will increase the dynamic
range of the system if the physical temperature of the attenuator is smaller than the source
temperature (usually the antenna temperature). Adding an attenuator in front of an existing
receiver will decrease the sensitivity (increase the MDS) of the receiver, however.
The intercept point of a receiver is usually set by the input intercept of the first mixer.
The reason for this is as follows intermodulation distortion is most likely to be generated in
the stages ahead of the relatively narrow band IF amplifiers, i.e., the preamplifier and mixer
stages. Since the input signal level at the mixer is generally larger than for any preceeding
stage, the intercept point of the mixer will usually determine the largest usable input signal
level. After the mixer, the signals are filtered by the IF filters, and generation of further
intermodulation components in the IF amplifiers is less likely since, presumably, the desired
signal component has been picked out from among other components. So the mixer is
the most likely source of intermodulation distortion products (in a well designed receiver).
Generation of intermodulation products in the mixer will be minimized if the gain in front
of the mixer is as small as possible.
This leads us to an important principle for receiver design: The smallest possible amount
of gain (preamp gain) should be used in front of the mixer. The minimum preamp gain will
be determined by the preamp noise temperature and/or the antenna temperature together
with the required MDS. Increasing the preamplifier gain above that required to set the
noise floor to an acceptable level will only increase the input signal levels at the mixer input
and degrade the large-signal handling capability of the receiver without providing useful
improvement in sensitivity.
394 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS

12.7 References
1. Carson, Ralph S., Radio Concepts: Analog, John Wiley & Sons, New York, 1990.
2. Clarke, Kenneth K. and Donald T. Hess, Communication Circuits: Analysis and De-
sign, Addison-Wesley, 1978.
3. Maas, Stephen A., Nonlinear Microwave Circuits, Artech House, 1988.
4. Smith, Jack, Modern Communications Circuits, McGraw Hill, 1986.
12.8. HOMEWORK PROBLEMS 395

12.8 Homework Problems

1. A nonlinear amplifier has an input-output voltage characteristic

vo (t) = k1 vi (t) + k2 vi2 (t) (12.55)

Suppose that three signals with equal amplitudes are input to the amplifier. The
signal frequencies are

f1 = 0.6 M Hz (12.56)
f2 = 1.3 M Hz
f3 = 1.5 M Hz

List the frequencies of all components that will appear at the output of the amplifier.

2. Consider an AM broadcast band receiver that is designed to cover the frequency range
540 to 1700 kHz. The single-conversion superhet receiver uses an IF of 455 kHz and
High LO. The receiver uses an RF amplifier that has the following nonlinear input-
output characteristic:
2 3
vo = k1 vRF + k2 vRF + k3 vRF (12.57)
Suppose two signals at frequencies f1 = 720 kHz and f2 = 780 kHz are input to the
receiver. Assume that both signals pass through the preselector. You may assume
that the receiver stages that follow the RF amplifier are linear and that the mixer can
be modeled as an ideal multiplier. List the frequencies of all possible output signals
from the RF amplifier that could be detected with the receiver. Also, for each signal
that can be detected, list the receiver tuning dial setting at which the signal will be
detected.

3. A nonlinear 2-ports input-output characteristic can be modeled as

vo = 12vi vi3

where vi and vo represent the instantaneous values of the time-varying component of

the input and output voltage, respectively.

(a) Suppose that this 2-port is used in a system with source and load impedance of
50 . Assume that the 2-port is conjugately matched at the input and output
in this system. Find the input power level that causes 1dB of gain compression,
P1dB . Express your result in dBm.
(b) Suppose that two signals are input to the 2-port, i.e., that vi = a1 cos !1 t +
a2 cos !2 t. List the frequencies of all components that will appear at the output
of the 2-port.
(c) Now assume that the two signals in part 3b have equal amplitudes, i.e., a1 =
a2 = a. Find the maximum input signal power (for each signal) that will cause
the in-band third-order components to be down 30 dB with respect to the desired
signals at the output of the 2-port. You may assume that gain compression can
be ignored at this input power level. Express your result in dBm.
396 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS

3 dBm

30 dBm

Figure 12.12: Spectrum analyzer display.

4. Two signals at closely spaced frequencies f1 and f2 are applied to the input of an
amplifier. The input power for each of the signals is -20 dBm. The display in Figure
12.12 is seen when the output of the amplifier is connected to a spectrum analyzer:
(i)
(a) Find the input intercept PI for this amplifier. Assume that gain compression is
not important.
(b) What is the operating power gain of the amplifier (in dB)?

5. Consider the receiving system in Figure 12.13:

TA = 1000 K

Preamp Receiver
Te = 200 K Bn = 50 kHz
GA = 12 dB N F = 8 dB

Figure 12.13: Receiving system.

(a) Assume that the antenna and 2-ports are matched to 50 . Find the available
signal power level from the antenna (in dBm) that is required to give a 15 dB
SNR at the output of the receiver.
(b) The two-tone third-order input intercept level for the receiver is +6.5 dBm. As-
sume that the preamplifier is linear. Find the spurious-free dynamic range (DR)
of the system if a 15 dB SNR is required at the output for detection. Express
your result in dB.

6. A receiver has eective input temperature Te = 1500K and noise bandwidth Bn =

3 kHz. The antenna temperature is 400K. The two-tone third-order input intercept
for the receiver is -5 dBm. A signal-to-noise ratio of 6 dB is required for detection.

(a) Compute the MDS for the receiver. Give your result in dBm.
(b) Compute the spurious-free dynamic range of the receiver. Give your result in dB.
(c) Suppose that the MDS computed in part 6a is too high for the intended ap-
plication, i.e., suppose that the required MDS is -130 dBm. To achieve this
12.8. HOMEWORK PROBLEMS 397

specification, we decide to add a preamplifier at the input of the receiver. Sup-

pose that the preamplifier will have an eective input temperature Te = 175K.
How much gain should the preamplifier have? Specify the value that will result
in an MDS = -130 dBm.
(d) Use your result from part 6c and compute the spurious-free dynamic range of the
preamp-receiver combination. You may assume that the preamplifier is a linear
2-port.

7. Consider a receiver with eective input temperature Te , two-tone third-order input

(i)
intercept level PI , and equivalent noise bandwidth Bn . Suppose we put a passive,
matched attenuator in front of the receiver. The attenuator has loss L > 1. The
physical temperature of the attenuator is equal to standard temperature, i.e., Tatt =
To . The addition of the attenuator will cause the cascade to have a higher input
intercept level than that of the original receiver. The new input intercept level for the
(i)
cascade will be PI L. Denote the source temperature by Ts . Depending upon the
value of Ts , the addition of the attenuator will either increase or decrease the dynamic
range of the system relative to that of the original receiver. What constraint must
be satisfied by Ts in order for the addition of the attenuator to increase the dynamic
range of the system?
8. An amplifier is found to have the following nonlinear input-output voltage character-
istic:
vo = k1 vi + k2 vi2 (12.58)
where k1 = 15 and k2 = 1. This characteristic does not include a third-order term,
k3 = 0, so the two-tone third-order input intercept power is infinite. Define a two-tone
(i)
second-order input intercept power, PI2 , which is the input power for each of the equal
amplitude tones that causes the output power in the second-order intermodulation
products at frequencies f1 + f2 , |f1 f2 | to be equal to the output power in the
(i)
desired tones. Assume that the input impedance of the amplifier is 50 and find PI2
(in dBm).

9. Two signals at closely spaced frequencies f1 and f2 are applied to the input of a
nonlinear amplifier. Denote the input power of the tones by P1 and P2 , respectively.
Suppose that P1 is decreased by 2 dB. Specify the change in the output power of each
of the frequency components: f1 , 2f2 , 2f2 f1 , 2f1 f2 , 3f1 . You may assume that
P1 and P2 are small enough so that gain compression can be ignored.

10. Suppose that an AM broadcase receiver uses a nonlinear preamplifier described by:

vo = k1 vi + k2 vi2 + k3 vi3 (12.59)

and a 4-diode doubly-balanced switching-type mixer to receive a strong signal with

carrier frequency 1200 kHz. The 4-diode doubly balanced mixer produces terms with
frequencies |fRF nfLO |, n=1,3,5... at its output. The radio uses a 455 kHz IF, the
local oscillator tunes from 995-2155 kHz, and the receivers tuning dial spans 540 -
1700 kHz. Find and list all settings of the receiver tuning dial at which you could
potentially detect a signal from the station at 1200 kHz.
398 CHAPTER 12. NONLINEAR EFFECTS IN 2-PORTS
Chapter 13

Phase-locked Loops (PLLs)

13.1 PLL Fundamentals

Figure13.1 shows a linear model for a PLL. In this idealized linear model, the input is the
phase of the reference signal and the output is the phase of the VCO. Both the reference and
VCO signals are assumed to be cosinusoidal signals with constance amplitude (for now) and
slowly changing phase. When the loop is locked and in a steady state, the VCO (output)

Voltage Controlled
Phase Detector Loop Filter Oscillator
r Va Vc o
PD F (s) V CO

Figure 13.1: Linear model for a PLL.

frequency and reference frequency are equal, i.e.,

Vr (t) = Vr cos (!r t + r ) (13.1)

Vo (t) = Vo cos (!r t + o ). (13.2)

The phase detector (PD) produces an output voltage that is proportional to the phase error,
e , which is defined to be the dierence between the phase of the reference signal and the
VCO signal:

Va (t) = Kd (r (t) o (t)) (13.3)

= Kd e (t)

or in the s-domain:

Va (s) = Kd (r (s) o (s)) (13.4)

= Kd e (s)
Kd = PD gain constant (V/radian)

399
400 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

The PD output voltage is applied to the loop filter to produce the VCO control voltage, Vc

Vc (s) = Va (s)F (s) (13.5)

where F(s) has a low-pass frequency response function.

The instantaneous frequency deviation of the VCO output signal is proportional to the
control voltage:
do
= Ko Vc (t) (13.6)
dt
or

s o (s) = Ko Vc (s) (13.7)

radians
Ko = VCO gain constant,
V sec

13.1.1 PLL Transfer functions

o (s) Ko Kd F (s)
= (13.8)
r (s) s + Ko Kd F (s)
e (s) s
= (13.9)
r (s) s + Ko Kd F (s)
Sometimes the control voltage, Vc , is the desired output signal, e.g., when the loop is used
as a demodulator for FM:
Vc (s) s Kd F (s)
= (13.10)
r (s) s + Ko Kd F (s)

13.1.2 Loop Gain and Notation

The closed-loop transfer function H(s) as defined in Equation 13.8 is

o (s)
H(s) = (13.11)
r (s)
Ko Kd F (s)
=
s + Ko Kd F (s)
Note:
e (s) r (s) o (s)
= (13.12)
r (s) r (s)

= 1 H(s)

Vc (s) s
= H(s)
r (s) Ko

H(s) can be written

Ko Kd F (s)/s
H(s) = (13.13)
1 + Ko Kd F (s)/s
13.1. PLL FUNDAMENTALS 401

The quantity A(s) = Ko Kd F (s)/s is called the open loop gain. Note that the closed loop
transfer function, H(s), can be written in terms of the open loop gain:
A(s)
H(s) = . (13.14)
1 + A(s)

13.1.3 Order and Type

The order of a PLL is defined by the highest power of s in the denominator of the closed
loop transfer function, i.e.,
First order:
K
H(s) = (13.15)
s+a
Second order:
K
H(s) = (13.16)
s2 + as + b
The type of a PLL is defined by the number of poles at the origin for the open loop transfer
function, i.e.,
Type 1:
K
A(s) = (13.17)
s
Type 2:
K
A(s) = (13.18)
s2
All PLLs are at least type 1, because the VCO output phase is proportional to the integral
of the control voltage.

13.1.4 Loop Filters

Commonly employed loop filters fall into four classes ranging from the simplest (no filter at
all) to active filters.
1. No filter:
F (s) = 1 (13.19)
Then
Ko K d
H(s) = (13.20)
s + Ko Kd
Ko Kd /s
=
1 + Ko Kd /s
!L /s
=
1 + !L /s

!L = Ko Kd = loop bandwidth

Notice that the open loop gain A(s) = KosKd has one pole at the origin, so this loop
is type 1. Since the power of s in the denominator of H(s) is 1, it is a first-order loop.
402 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

2. The simplest filter that gives a second-order loop is a single-pole low-pass RC filter,
as shown in Figure 13.2. If the PD output is modeled as an ideal voltage source, Va ,

R
+ +
Va C Vc

Figure 13.2: Simple low-pass loop filter.

and the control input to the VCO has high impedance, then the transfer function of
the RC loop filter is:
1
F (s) =
1 + s
With this loop filter, the closed-loop transfer function for the PLL is

1
H(s) = s2 s
(13.21)
Ko Kd + Ko Kd +1

This function can be written in the canonical form

1
H(s) = (13.22)
1 2
!n2 s + !2n s + 1

where r
Ko Kd
!n = (13.23)

!n
= 2Ko Kd
(13.24)
= p 1
2 Ko K d

The parameter !n is the loop bandwidth and is called the damping factor. To-
gether the loop bandwidth and the damping factor determine how quickly the loop
can respond to changes in the reference signals frequency or phase.

3. The lag-lead filter (Figure 13.3) is often preferred to the simple low-pass RC filter
because it leads to an improved phase margin and therefore a more stable loop transfer
function. The lag-lead loop filters transfer function is

1 + s2
F (s) = ,
1 + s1

where
1 = (R1 + R2 )C

2 = R 2 C
13.1. PLL FUNDAMENTALS 403

R1

+ C +
Va Vc
R2

Figure 13.3: Lag-lead loop filter.

The loop transfer function is

Ko Kd F (s)
H(s) = (13.25)
s + Ko Kd F (s)

Ko Kd (s 2 + 1)/1
H(s) =
s2 + s(1 + Ko Kd 2 )/1 + Ko Kd /1

s(2 !n !n2 /Ko Kd ) + !n2

H(s) =
s2 + 2 !n s + !n2

r
Ko Kd
!n =
1

1/2
1 K o Kd 1
= 2 +
2 1 K o Kd

2 ! n !n
= +
2 2 K o Kd

4. An active filter (often used in integrated circuit PLLs) is shown in Figure 13.4: The
transfer function of this filter is

A (s 2 + 1)
F (s) = ,
s 2 + 1 + (1 + A) s 1

where A is the open loop gain of the op-amp. For suciently large A the transfer
function is well approximated by

s 2 + 1
F (s) ' . (13.26)
s 1
404 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

R2 C
2 = R2 C
R1
1 = R1 C
+

Figure 13.4: Active loop filter employing an op-amp.

Thus, for large A the loop transfer function can be written

2 !n s + !n2
H(s) = (13.27)
s2 + 2 !n s + !n2
r
Ko Kd
!n =
1
1/2
2 Ko Kd
=
2 1
2 ! n
=
2
The high gain (large A) active loop filter and the passive lag-lead filter with large loop
gain (Ko Kd ) both yield open-loop transfer functions of the form

2!n s + !n2
H(s) = (13.28)
s2 + 2!n s + !n2
The high-gain active loop filter approximates a type 2 loop because the open loop gain
has two poles at the origin:
Ko Kd F (s)
A(s) = s
(13.29)
' Ko Kd ss22+1
1

The fact that the denominator of A(s) contains s2 means that there are 2 integrators
in the loop. One is the VCO itself, because the phase of the VCO output signal is
the integral of the control voltage (o = Ko Vsc (s) ). The other integrator is the active
filter. Table 13.1 illustrates the how the parameters of the loop transfer function
depend on the elements in the circuits of the passive lag-lead filter and the active
filter. Figure 13.5 shows how the active filter is implemented in the Motorola MC4044
PLL integrated circuit.

13.1.5 Steady-state Error Analysis

Consider two types of inputs:
13.1. PLL FUNDAMENTALS 405

Passiveqlag-lead ActiveqFilter
!n = Ko1Kd !n = Ko1Kd

2 !n 2 !n
= 2 = 2

1 = (R1 + R2 )C 1 = R1 C

2 = R 2 C 2 = R2 C

Table 13.1: Comparison of passive lag-lead and active loop filters.

5V
RL = 1k
R2 C
Vo
R1

Figure 13.5: Active filter implementation in the Motorola MC4044 chip.

1. Step input in phase:

r (t) = u(t) (13.30)

2. Step input in frequency:

fr (t) = fc + f u(t) (13.31)

or, equivalently:

r (t) = ! t u(t) (13.32)

The error signal e (t) = r (t) o (t) has transfer function

e (s) 1
= . (13.33)
r (s) 1 + Ko Kd F (s) /s

The Laplace transform final value theorem (for stable systems) says:

lim e (t) = lim s e (s) (13.34)

t!1 s!0

s2 r (s)
= lim
s!0 s + Ko Kd F (s)
406 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

If r (t) is a step input, e.g., r (t) = u(t), then

1
r (s) = (13.35)
s
s
lim e (t) = lim
t!1 s!0 s + Ko Kd F (s)

Now F (s) is either a constant (first-order loop) or a low-pass filter that may include poles
at the origin, i.e.,
K
lim F (s) = 6= 0 (13.36)
s!0 sn
Thus

sn+1
lim e (t) = lim =0 (13.37)
t!1 s!0 Ko Kd K

i.e., a stable PLL will track step changes in phase with zero steady-state error. If the
frequency changes suddenly, i.e.,
fr (t) = fc + f u(t) (13.38)
or
r (t) = t ! u(t) (13.39)
then
1
r (s) = ! (13.40)
s2
The steady-state phase error is

!
lim e (t) = lim (13.41)
t!1 s!0 s + Ko Kd F (s)
!
=
Ko Kd F (0)
If F (0) = constant (dc gain of filter = constant), then the steady-state phase error is inversely
proportional to Ko Kd . Since a larger value for Ko Kd leads to a larger loop bandwidth for
all the filters considered, we can conclude that a large loop bandwidth is desirable if the
steady-state phase error is to be minimized.
The frequency error fe (t) = dt d
e (t) will tend toward zero as t ! 1 (because e !
constant). If it is necessary to have zero steady-state phase error in response to a step
frequency input, then
!
lim =0 (13.42)
s!0 Ko Kd F (s)
or
lim F (s) = 1 (13.43)
s!0
This means that F(s) must have a pole at the origin which means that the DC gain of the
filter must approach infinity. In this case the system will be type 2, since
Ko Kd F (s)
A(s) = (13.44)
s
13.2. STABILITY ANALYSIS 407

has two poles at the origin. We can approximate the type 2 loop with a high gain active filter
which can be used to achieve an essentially zero steady-state phase error. The addition of a
pole at the origin can, however, cause stability problems. Stability issues will be discussed
in the following section.

13.2 Stability Analysis

The stability of a loop is determined by A(s) which depends on F (s). A loops stability
can be studied by examining the complex-plane location of the poles of H(s). This requires
knowledge of the analytical form of H(s). Alternatively, stability can be studied without
direct knowledge of the pole locations or even an analytical description of the transfer
function by using Bode plots. A Bode plot (shown in Figures 13.6 and 13.7) consists of a
pair of graphs of the magnitude (in dB) and phase of the open loop gain A(s) plotted on
a logarithmic frequency scale. The Bode criterion for stability is simple and is equivalent
to the reverse of the Barkhausen criterion for oscillation. The Bode criterion states that a
loop is stable if the magnitude of A(s) falls below 1 (0 dB) before the phase-shift of A(s)
reaches 180 .

Gain (dB)

Gain crossover frequency

0 log f
Gain margin

Phase (Deg)

Phase crossover
frequency
0
log f

Phase margin
-180

Figure 13.6: Bode plot of A(s) for a stable PLL

Summary of stability-related terminology:

Phase margin is the dierence between the actual phase and 180 at the frequency
where the magnitude of the open loop transfer function is unity.
408 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

Gain (dB)

Gain crossover frequency

0 log f

Phase (Deg)
Phase crossover
frequency

0
log f

-180

Figure 13.7: Bode plot of A(s) for an unstable PLL

13.2. STABILITY ANALYSIS 409

Gain margin is the number of dB below 0 dB for the gain at the frequency where the
phase of the open loop transfer function is 180 .

The greater the phase margin, the more stable the system and the more phase lag
from parasitic eects can be tolerated. Phase compensation provided by the lag-lead
filter can often be used to stabilize a marginally stable loop. See the examples in the
following section.

13.2.1 Examples of Stability Analysis

Consider a PLL which has Ko Kd = 50 rad/s and which contains a simple RC low-pass
filter with corner frequency !c = 100 rad/s. The open-loop gain is

Ko Kd
A(s) = F (s) (13.45)
s
or
50
A(s) = s
s(1+ 100 )
(13.46)
50
A(j!) = j!
j!(1+ 100 )

The magnitude and phase of A(j!) are plotted in Figure 13.8. The crossover frequency,

Figure 13.8: Bode plot for filter corner frequency = 100 rad/s

where |A(j!)| = 1, is ! = 45.51 s 1 . At this frequency the phase angle of the open-loop
gain is 114.5 . The phase margin is therefore 180 114.5 = 65.6 .
Suppose that the filter corner frequency was 10 rad/s instead of 100 rad/s. The magni-
tude and phase of A(j!) are shown in Figure 13.9. In this case the gain crossover frequency
is 21.3 s 1 and the phase angle at the gain crossover is 154.8 . So the phase margin is
410 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

Figure 13.9: Bode plot for filter corner frequency = 10 rad/s

180-154.8 = 25.2 . The phase margin of 25.2 is rather small, and a relatively small per-
turbation of the phase angle (due to aging, temperature, or other eects) could cause the
loop to become unstable and oscillate. It is therefore desirable to find a way to increase the
phase margin. The phase margin can be improved by employing a lag-lead filter which adds
compensation by adding a zero to the transfer function at 50 rad/s. The open-loop gain
with the lag-lead filter is
50 s/50 + 1
A(s) = (13.47)
s s/10 + 1
Now the gain crossover occurs at 22.4 s 1 and the phase angle at this frequency is -131.8
which corresponds to a phase margin of 48.2 . The eect of adding phase compensation is to
cause the phase curve to turn back up away from -180 at high frequencies, thus increasing
the phase margin.

13.3 Transient Response of PLLs

Consider a first-order loop and suppose that the reference phase undergoes a step change
at t = 0 :
e (t) = u(t)
(13.48)
o (t) = (1 e Ko Kd )
Figure 13.11 shows how the VCO output signal responds to the step change in a first-order
loop. When the loop bandwidth Ko Kd is large, the time constant is small, and the loop
responds quickly to changes in the reference signals phase. Now, suppose that the reference
signals frequency undergoes a sudden step change. In that case:
r (t) = !u(t)
(13.49)
! Ko Kd
o (t) = !t Ko Kd (1 e )
13.3. TRANSIENT RESPONSE OF PLLS 411

Figure 13.10: Bode plot for lag-lead filter.

r(t)

o (t)

Figure 13.11: Time constant = 1

Ko Kd .
412 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

Figure 13.12 shows the eect the response of the VCO output phase compared to the ref-
erence phase after a frequency step. Notice that in steady-state, the VCO output phase

Ko Kd
r(t)

o (t)

Figure 13.12: Reference signal phase and VCO output phase after a step change in the
reference frequency.

lags the reference signal phase by Ko K !

d
. This means the the phase error is non-zero in
steady-state. If the loop gain, Ko Kd , is large then the phase error will be small.
We will now consider the transient response of second order type-2 loops. The transfer
function of such a PLL is:
2!n s + !n2
H(s) = 2 (13.50)
s + 2!n s + !n2
where !n is the loop bandwidth (or natural frequency), and is the damping factor. Both
are important in determining the frequency response and transient response of the loop.
Figure 13.13 shows the frequency response |H(j!)| for dierent damping factors.

Figure 13.13: Frequency response |H(j!)| for various damping factors

The way in which a second-order loop tracks a step change in phase or frequency can be
13.3. TRANSIENT RESPONSE OF PLLS 413

studied by computing the transient behavior of the phase error. Recall from Equation 13.12
that the phase error transfer function is given by:
e (s)
=1 H(s) (13.51)
r (s)
For a step change in the reference phase:
r (t) = u(t) (13.52)
or

r (s) = (13.53)
s
Thus
s
e (s) = (13.54)
s2 + 2!n s + !n2
where s has been normalized such that s = j!/!n .
It is not hard to show that the transient response is of the form:
p p
e (t) = e !n t {cosh[ 2 1!n t] p sinh[ 2 1!n t]} (13.55)
2 1
The transient response is plotted in Figure 13.14 for several dierent values of the damping
factor. Notice that (1) Increasing the damping factor results in faster settling and less
ringing, and (2) the phase error tends to zero for large times. The latter was predicted
earlier in our analysis of Equation 13.41.

Figure 13.14: Transient responses for damping factor values

It is useful to point out that these results can also be interpreted as the transient frequency
response due to a step change in frequency. To illustrate, suppose that the input is a step
change in frequency, i.e.,
!r = !c + ! u(t) (13.56)
414 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

The transfer function relating VCO output frequency to the reference frequency is
!(s) o (s)/s
!r (s) = r (s)/s
(13.57)
o (s)
= r (s)

This means that the plot given in Figure 13.14 can be interpreted as the VCO frequency
response to a step change in the reference frequency.
Now consider what happens to the output phase when the reference frequency undergoes
a step change. Suppose that initially the loop is locked at its center frequency, !c . At t =
0 the reference frequency is increased by an amount !, i.e.,

!r = !c + ! u(t) (13.58)

In terms of the reference phase, this input can be written as

d r (t)
= ! u(t) (13.59)
dt
Thus
!
r (s) = (13.60)
s2
The phase error is therefore
!
e (s) = (13.61)
s2 + 2!n s + !n2
The transient response is
! 1 p
!n t
e (t) = e p sinh[ 2 1!n t] (13.62)
!n 2 1
This function is plotted in Figure 13.15. Notice that because the loop being considered is a
type 2 loop, the phase error tends to zero for large times. For type 1 loops the phase error
would tend toward a constant value.

13.3.1 Summary of Second-order Loops

There are 2 parameters to play with if the loop is second-order:

!n ) natural frequency

) damping factor

Rules of thumb:
Large !n ) small time constant, fast response

Large ) damped response, no ringing

Small ) ringing
Also note that:
13.4. APPLICATIONS 415

Figure 13.15: Transient response for a step change in the reference frequency.

For a second-order loop the phase margin increases with increasing .

The natural frequency, !n , and damping factor, , cannot be specified independently if
the loop filter is a simple low-pass filter. If an active filter or lag-lead filter is employed
then !n and can be specified independently.

13.4 Applications
13.4.1 Demodulation of an FM signal
Suppose an FM signal is applied to the reference input of a PLL. The loop will try to track
the deviations of the input frequency. If the loop is able to follow the deviations of the input
frequency, then the control voltage Vc will be proportional to the instantaneous frequency
deviation of the reference signal. In this application the PLL is called a modulation tracking
loop. The transfer function relating the control voltage Vc (s) to the signal phase r (s) as
given earlier in Equation 13.10 is

Vc (s) s Kd F (s)
= (13.63)
r (s) s + Ko Kd F (s)

The input signal for FM is of the form

Z t
Vr (t) = Vr cos !c t + m(t0 ) dt0 (13.64)
o

i.e., Z t
r (t) = m(t0 ) dt0 (13.65)
o
416 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

so
1
r (s) = M (s) (13.66)
s
Thus
Kd F (s)
Vc (s) = M (s) (13.67)
s + Ko Kd F (s)
1
= H(s) M (s)
Ko

The control voltage response is a scaled and filtered version of m(t).

For frequency demodulation, the detailed shape of |H(j!)| is important. Usually, a
flat frequency response is desired. This means that a Butterworth-type transfer function
is desirable. A second-order loop with = 0.707 provides the maximally flat Butterworth
response, as shown in Figure 13.16.

|H(j )|
= 0.707

/ n 1

Figure 13.16: Butterworth response is obtained when = p1 .

2

The loop bandwidth must be at least as large as the bandwidth of the modulation signal,
m(t). The best results are obtained when the loop bandwidth is substantially larger than
the modulation bandwidth. (See Phaselock Techniques, F.M. Gardner, J. Wiley, 2nd ed.,
1979, Ch. 9.)

13.4.2 PLL Response to AM

Let Vr (t) = Vr m(t) cos (!r t + r ). To simplify the analysis, suppose that the PD is imple-
mented with an ideal multiplier so that the error voltage Va (t) is

Va (t) = Kd m(t) sin e (13.68)

The average value of Va (t) is

< Va (t) >=< m(t) > Kd sin e (13.69)

13.4. APPLICATIONS 417

As long as< m(t) >6= 0, the PD output (after averaging, which is carried out by the low-
pass loop filter) is proportional to sin e . This means that loop will lock and will track the
carrier as long as a carrier is present (< m(t) >6= 0). The averaging time should be long
compared to the time-scale over which m(t) varies.
A PLL can be used to generate a local carrier which can be used for coherent demodu-
lation of AM, as shown in Figure 13.17.

VAM (t) Va Vc
PD F (s) V CO

X LP F recovered modulation

Figure 13.17: Carrier for coherent demodulation of AM

The 90 phase shift is required because the the PD is assumed to give 0 output voltage
when the reference and VCO signals are in quadrature.

13.4.3 Carrier Recovery

The loop cant track the carrier if it isnt there. In DSB signals, the carrier is suppressed,
so there is no carrier component for the PLL to track. There are two equivalent schemes
for recovering the suppressed carrier using a PLL:
1. Squaring loop Assume Vr (t) = m(t) cos (!r t + r ) and < m(t) >= 0 (sup-
pressed carrier). A carrier component can be recovered if the signal is squared as
in Figure 13.18: The squared input signal is VDSB
2
(t) = m2 (t) cos2 (!r t + r ). Since

VDSB (t)
( )2 PD F (s) V CO

X LP F recovered modulation

Figure 13.18: Squaring loop.

< m2 (t) > is non-zero, this signal has a component at twice the carrier frequency,
2!r . The loop locks to this double-frequency component. The VCO output is divided
418 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

by 2 to produce a recovered carrier component at frequency !r which is then used

to coherently demodulate the DSB signal. The recovered signal at the output of the
divider has a 180 phase ambiguity that results from the unknown initial state of the
divider. This means that the sign of the recovered modulation is arbitrary either
+m(t) or m(t) could be recovered. This cause of the ambiguity can be understood
by noting that the frequency divider produces a cosinusoidal signal whose argument is
2 of the argument of the double-frequency signal presented to its input. The double
1

frequency signal is written as cos(2!r t + 2r + 2n) with n any integer. The output of
the divider is cos(!r t + r + n), which can be written as cos(!r t + r ), depending
on whether n is an even or odd integer.

2. Costas loop refer to Figure 13.19. Without the bottom loop, this would be a

I(t) = m(t) cos(e )

X LP F

cos(!r t + o )
m(t) cos(!r t + r )
F (s) X
V CO
90

X LP F
Q(t) = m(t) sin(e )

Figure 13.19: Costas loop

conventional PLL and the error signal would have an average value of zero, because
< m(t) >= 0. With the second, or quadrature loop, the error signal is m2 (t) sin(r
o ) cos(r o ). Since < m2 (t) >6= 0, the error signal is finite and operation is
like the squaring loop. The Costas loop has an important advantage over the squaring
loop, however, because it does not need any circuitry that operates at twice the carrier
frequency. When configured as shown the Costas loop automatically provides in-phase
and quadrature outputs, so the Costas loop acts as a quadrature demodulator.

13.5 Frequency Synthesis with PLLs

The simplest PLL frequency synthesizer is created by adding a frequency divider to the
loop that was considered in Figure 13.1. See Figure 13.20. When the PLL is locked both
of the phase detector input frequencies are equal to the reference frequency, fr , as shown.
This means that the frequency at the divider input, which is the VCO output frequency,
must be N fr . The VCO output frequency is an integral multiple of the reference frequency.
The divisor, N , of the N divider is typically programmable. so the output frequency can
be changed by changing the divisor, N . The range of available output frequencies will be
13.5. FREQUENCY SYNTHESIS WITH PLLS 419

fr Va Vc fo = N fr
PD F (s) V CO

N
fr

Figure 13.20: PLL frequency synthesizer.

limited by the tuning range of the VCO. The reference frequency is typically generated by
a stable crystal oscillator, so when the loop is locked, the VCO output frequency will reflect
the stability of the reference oscillator. If the reference oscillator frequency tolerance is f ,
the VCO output frequency tolerance will be N f .
When the divider is added to the loop, the transfer functions are modified as follows:

o (s) Ko Kd F (s)
= (13.70)
r (s) s + Ko Kd F (s)/N
e (s) s
= (13.71)
r (s) s + Ko Kd F (s)/N

The high gain second-order loop transfer function becomes

o (s)
H(s) = r (s)

N (2!n s+s2 )
= s2 +2!n s+!n2

(13.72)
q
Ko Kd
!n = N 1

2 !n
= 2

The loop parameters !n and are functions of N. As N is varied to tune the synthesizer
to dierent frequencies the resulting variations in !n and can result in significant changes
in the loop dynamics. It is necessary to perform a worst case design analysis, i.e., using
Nmin and Nmax , find !nmin , !nmax , min , max , and make sure that transient response and
rejection of spurious signals will be adequate at all frequencies of interest.

13.5.1 Noise and Spurious Signals

In practice, both the reference signal and the VCO signal will be accompanied by noise
and/or additional spurious signals. For example, to model the eect of an imperfect VCO
signal which contains phase noise, the model shown in Figure 13.21. The noise on the VCO
signals phase is represented by the noise signal N o . The transfer function relating this
420 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

No

r o
PD F (s) V CO +

Figure 13.21: Linear model including VCO phase noise.

noise to the output phase is

o (s) 1
= (13.73)
N o (s) 1 + Ko Kd F (s)/Ns
s
=
s + Ko Kd F (s)/Ns

This is a high-pass frequency response function slow, low frequency changes in the VCO
output phase will be filtered out by the loop and will not reach the output, but fast, high
frequency changes in the VCO output phase will be passed through to the output. This
result can be understood by considering that slow changes in the VCO output phase produce
a slowly varying error voltage which is passed by the loop filter and sent to the VCO control
input. The error signal tunes the VCO to compensate for the original change. On the other
hand a fast, high frequency, change in the VCO phase produces a high frequency error signal
which does not pass through the loop filter. High frequency phase perturbations will not be
filtered out by the loop because the loop cannot follow them. So the high-frequency phase
perturbations appear on the VCO output signal.
The 3 major noise sources in a PLL are:
Nr the phase noise on the reference signal
N d noise or spurious signals at the output of the PD. This includes harmonics of the
reference signal, which are present in the output of most PDs. These harmonics will
modulate the VCO and produce unwanted discrete sidebands in the VCOs frequency
spectrum.
No the intrinsic VCO phase noise, which has already been discussed.
Figure 13.22 shows a block diagram that includes these three noise sources.
The total output noise signal is

Ko Kd F (s) s
N =( Nr + N d) + No (13.74)
s + Ko Kd F (s)/N s + Ko Kd F (s)/N

The loop functions as a low-pass filter for phase noise on the reference signal and for reference
frequency harmonics at the output of the phase detector. The loop functions as a high-pass
filter for VCO phase noise. A typical VCO noise spectrum is shown in Figure 13.23. To
minimize the VCO noise contribution, the loop bandwidth should be as wide as possible,
13.5. FREQUENCY SYNTHESIS WITH PLLS 421

Nr Nd No

r o
+ PD + F (s) V CO +

Figure 13.22: Linear model for PLL with noise sources.

-40

-80 Narrow-band
loop response
Wide-band
-120 loop response

-160

102 103 104 105 106 107

Figure 13.23: Typical VCO noise spectrum

because a large loop BW means the high-pass characteristic that filters the VCO noise will
have smaller values at low frequencies. On the other hand, the loop bandwidth should be
significantly less than the reference frequency in order to reduce spurious modulation of the
VCO due to the reference frequency and its harmonics, which are present in the PD output.
These requirements may conflict if the reference frequency is too small.
The level of the sidebands induced on the VCO output due to reference frequency com-
ponents can be estimated using the following equation:

sideband level Vref Ko

= (13.75)
carrier level 2 !ref

where Vref is the peak voltage value of spurious frequency at the VCO input. Vref is related
to the spurious voltage at the output of the phase detector by

Vref = V F (s)|s=j !ref (13.76)

Suppose that the loop filter is the active filter shown in Figure 13.24 and assume that
the op-amp open loop gain is very high. The loop filter transfer function will be
422 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

R2 C
2 = R2 C
R1
1 = R1 C
+

Figure 13.24: High-gain active loop filter.

s 2 + 1
F (s) ' (13.77)
s 1

2 = R2 C (13.78)

1 = R1 C (13.79)

Usually, !ref is > 2 ,

1
so
2 R2
F (j!ref ) ' = (13.80)
1 R1
or, in terms of loop parameters,
2 N !n
F (j!ref ) ' (13.81)
K o Kd
so
2 N !n
|Vref | ' |V | (13.82)
Ko Kd
If equation 13.82 is inserted into Equation 13.75:
sideband level N !n
'V (13.83)
carrier level !ref Kd
Usually !ref , N and Kd are predetermined by other constraints. Only !n and, to a lesser
extent, , can be adjusted to diminish the reference frequency sidebands. If the VCO is
noisy, it may not be feasible to make !n very small because the VCO phase noise may then
become objectionable. And cannot be much smaller than 0.5 without running into phase
margin problems and excessive ringing. In some cases it may be helpful to add additional
poles to the loop filter to attenuate the reference frequency and its harmonics.

13.5.2 Phase Detectors - Digital

13.5.2.1 Exclusive-OR Phase Detector
Figure 13.25 shows an exclusive-OR phase detector. The two input signals represent the
reference and VCO signals after conversion to logic-level signals. The output of the exclusive-
OR gate is high if, and only if, one of the two inputs is high. In the figure the phases of the
input signals dier by one quarter of a cycle, so e = /2. The output signal consists of a
13.5. FREQUENCY SYNTHESIS WITH PLLS 423

Figure 13.25: Exclusive-OR phase detector

pulse stream with 50% duty cycle and twice the frequency the input signals. If the phase
dierence decreases (increases) from the nominal value of /2, the width of the output pulses
will decrease (increase). When the phase dierence is equal to the input signals are always
dierent, and the output is constantly high, and when the phase dierence is zero, the input
signals are always the same and the output is always zero. The lowpass loop filter averages
the phase detector output waveform over many cycles, so the output of the loop filter will
be approximately equal to the mean of the ex-OR output waveform. Hence, only the mean
of the output waveform is important for determining the loop dynamics. The average PD
output value versus phase shift e is shown in Figure 13.26.

Vm

Vm /2

/2 2 e

Figure 13.26: Average output versus phase shift

The exclusive-OR has an operating range of /2. The center of the operating range is
at r o = e = /2, i.e., the loop locks with the reference and VCO signals in quadrature.
If the phase error ever enters the region between and 2 where the slope of the PD
characteristic is negative, the loop will lose lock because the error voltage drives the VCO
in the wrong direction, tending to increase the error signal rather than decrease it.
Ideally, the output contains no energy at the input frequency, but it has significant com-
ponents at twice the reference frequency at also at multiples thereof. The second harmonic
content is maximum at e = /2 and has a peak-to-peak amplitude 1.27 times the peak-to-
peak value of the phase detector characteristic. If these reference frequency harmonics are
not suciently attenuated by the loop filter they can produce significant frequency mod-
ulation of the VCO and potentially objectionable discrete sidebands on the VCO output
signal.

13.5.2.2 Hold-in Range of PLL, !H

The hold-in range is also called the lock range, tracking range, or synchronization range.
It is the maximum reference frequency range over which the PD stays within its operating
range. Suppose the PD operating range is m radians. The output frequency range is then

Kd Ko F (0)m (13.84)
424 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

or, defining KL = F (0) as the DC gain of loop filter,

! H = K d Ko KL m (13.85)

The loop will track reference frequencies over a range of !H .

13.5.2.3 Set-reset (SR) flip-flop

A set-reset flip-flop can be used as a phase detector (Figure 13.27). The input signals fA and

fA S Q
fB R

Figure 13.27: S-R flip-flop

fB consist of narrow pulses, as in Figure 13.28. Signal B resets the flip-flop, and signal A sets

A t

B t

OUT t

Figure 13.28: Signals fA and fB

it. The average output voltage as a function of phase error is shown in Figure 13.29. The
SR flip-flop has twice the operating range of the exclusive-OR (). The loop locks with
r o = . The output contains a component at the input frequency which has a maximum
value at r o = . The amplitude of this component is 1.27 times the peak-to-peak value
of the output characteristic.
Trade-os between the exclusive-OR(EX-OR) and the S-R flip-flop are:

EX-OR small operating range but no output at fundamental frequency, fr

S-R Larger operating range but output at fundamental frequency

The S-R flip-flop has some inherent problems if the input pulses to the S-R flip-flop
have finite width, there will be flat spots in the output characteristic. If the input signal
goes away, the flip-flop output will remain high (or low). The loop interprets this condition
13.5. FREQUENCY SYNTHESIS WITH PLLS 425

V /2

2 4 e

Figure 13.29: Average output as function of phase error

as a large phase error and changes the VCO frequency. Eventually the VCO will bang up
against its limiting frequency and sit there.

13.5.2.4 Quad-D Phase-Frequency Detector

The quad-D phase-frequency detector (PFD) is a more sophisticated digital phase detector
consisting flip-flops and additional logic. Widely available in integrated circuit form, it
is called a phase-frequency detector because it provides an indication of the sign of the
frequency error when the loop is out-of-lock. The PFD has two output terminals labeled
U and D (for up and down). U and D can be high simultaneously, but only one can
be active (low) at any time. Define the duty ratio du or dD as the fractional time either
terminal is active (low). The U output is only active when the phase error is greater than
0, and the U output is active when the phase error is negative. The phase characteristic
when locked looks like Figure 13.30. So du dD looks like Figure 13.31.
A PFD has unique properties. The active phase error range is 360 which is double
that of the S-R flip-flop and 4 times as large as the XOR. If the loop is unlocked, only
the U or the D output pulls low (active). The active output indicates the direction of the
frequency error. Both outputs are quiescent at the equilibrium tracking point (e = 0),
i.e., if the loop is locked at the center of its tracking range there are no reference frequency
harmonics. In the near vicinity of equilibrium one or the other of the outputs pulls down
with a small duty cycle. Thus the spurious signal is a narrow pulse at the input frequency.
This is desirable, since it is much easier to filter narrow pulses than it is to filter square
waves. There may be some crossover distortion in the PFD characteristic around e = 0. If
an input signal transition is missing, the PFD interprets this as loss of lock. Since the PFD
has memory, the eects propagate for more than one cycle.
The PFD has two outputs (U and D) which must be subtracted and then averaged in
order to generate the VCO control voltage. Since each output can be either high or low, the
dierence between the two outputs has three states, positive, zero, or negative. Typically,
these states are used to control a charge pump a constant current source that can source
or sink a current Io or be in an o state. The maximum source/sink current is typically on
the order of a few mA.
426 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

dU

2 e

dD

2 e

Figure 13.30: Locked phase characteristic

dU dD

2 2 e

Figure 13.31: Quad-D flip-flop phase detector characteristic.

13.5. FREQUENCY SYNTHESIS WITH PLLS 427

13.5.3 Examples
A PLL synthesizer has the following parameters:

fout = 100 MHz (13.86)

fref = 100 kHz
rad
Ko = 1 MHz/V ) 2 x 106
V sec
The PD is an S-R flip-flop with output voltage between 0 and 5 V. Consider what the loop
filter attenuation must be at 100 kHz in order to keep the reference frequency sidebands
more than 40 dB down with respect to the carrier (-40 dBc) at the VCO output:

sideband level
40 dB ) (13.87)
carrier level
40/20
= 10
= 0.01

We need 0.01 V |F (j!ref )| 2!Kref

o
0.01, where V is the 0-to-peak spurious component
voltage out of the PD. In this case

1
V = (1.27) (5) (13.88)
2

= 3.175 V

Then we need

2 !ref 1
|F (j!ref )| 0.01 (13.89)
Ko 3.175

2.105 1
= 0.01
106 3.175

= 6.30 x 10 4

= 64 dB

Suppose the loop filter is a simple RC low-pass filter with F(s) of the form
1
F (s) = s (13.90)
1+ !c

and the loop filter is followed by a DC amplifier, as in Figure 13.32.

So
K o Kd KL
!c = p (13.91)
2N
428 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

KL
o

Figure 13.32: Loop filter followed by DC amplifier.

To get -40 dB sidebands, we need

|F (j!ref )| = 6.3 x 10 4
(13.92)

or, assuming !ref !c

KL !c
= 6.3 x 10 4
(13.93)
!ref
Then, using

Kd = 5V/cycle (13.94)

5 V
=
2 rad

Kd = 2 x 106 rad/V-sec

N = 1000

Combining Equations 13.91 and 13.93

s
Ko Kd !ref 6.3 x 10 4
!c = p = 1.183 103 (13.95)
2N
or
fc = 188 Hz
If the phase margin is to be 45 , what are the required values of !c , KL , and !n ? The open
loop gain is
K o Kd 1
A(s) = F (s) (13.96)
s N
KL
F (s) =
1 + s/!c
so
Ko Kd KL /N
A(s) = (13.97)
s(1 + s/!c )
Ko Kd KL /N
A(j!) =
j!(1 + j!/!c )
13.5. FREQUENCY SYNTHESIS WITH PLLS 429

For a phase margin of 45 , we want

|A(j!)| = 1 when ! ' !c (13.98)

Thus
Ko Kd KL /N
1= p (13.99)
!c 2
From Equation 13.93:
!ref 4
KL = 6.3 10 = 0.34 (13.100)
!c

The natural frequency of the loop (loop bandwidth) is

r
Ko Kd K L ! c
!n = = 1418 (13.101)
N
or
fn = 226 Hz
The damping factor for this loop would be
1 !c
= (13.102)
2 !

= .42

The settling time for the loop will be

7
= 5 ms (13.103)
!n

13.5.4 Pre-scalers
A number of factors may combine to make the simple synthesizer discussed in section 13.5.3
unsuitable. For example: (i) the required channel spacing (fr ) may be too small to allow
for adequate suppression of spurious components that appear at the PD output; (ii) a small
fr may result in a lock-up time which is too long; (iii) programmable dividers may not be
available for operation at very high frequencies. Problem (iii) can be addressed by first
dividing by a fixed-modulus divider (pre-scaler) as shown in Figure 13.33. It is possible

fr fo = M N fr
PD F (s) V CO

N M
fr

Figure 13.33: PLL frequency synthesizer with fixed pre-scaler (M ) and programmable di-
vider (N ).
430 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

to obtain fixed-modulus dividers (pre-scalers) that operate well into the microwave region.
The output of the M pre-scaler will be at a lower frequency, where the fully programmable
N can operate. When the loop is locked
fo
fr = (13.104)
MN
or
fo = N (M fr ) (13.105)
A pre-scaler allows a loop to operate at higher frequencies but the output frequency can
only be changed in increments of M fr . To get better resolution we must decrease fr . This
will tend to make problems (i) and (ii) worse.

13.5.5 Dual Modulus Dividers

A method for obtaining good frequency resolution while operating at high output frequencies
uses a variable modulus pre-scaler. Typically, the variable modulus divider is a high-speed
divider with two choices for the modulus. A synthesizer using the dual modulus scheme is
shown in Figure 13.34 where the high-speed divider can divide by either P or P + Q. The
divide ratio is selected by a signal from the low-speed divider A. Here is a summary of how

fr fo = (AQ + P N )fr
PD F (s) V CO

P
P + Q

N A

Figure 13.34: PLL synthesizer with dual modulus pre-scaler. The frequency at the inputs
of the N and A counters is much smaller than the VCO output frequency, so these
counters do not have to be very fast. The dual modulus pre-scaler does have to be fast,
since it accepts the VCO output signal and must divide it by either P or P + Q. The dual
modulus pre-scaler accepts a binary input from counter A, which instructs it to divide by
P or P + Q. Note that N will always be greater than A.

the dual modulus pre-scaler operates:

1. Load N and A counters with N and A, respectively. The dual modulus pre-scaler
is set to divide by P + Q.

(a) After P + Q pulses, both N and A counters are decremented.

(b) Continue until the A counter reaches zero. This will occur after (P + Q)A cycles
of the VCO output signal, fo .
(c) The A counter then instructs the dual modulus pre-scaler to divide by P . The
N counter contains N A. (We must have N > A for this scheme to work).
13.5. FREQUENCY SYNTHESIS WITH PLLS 431

(d) Continue until the N counter reaches zero. This takes P (N A) cycles of fo .
(e) The N counter outputs 1 pulse. The cycle is complete.
(f) Begin again at step (1).

The total number of fo pulses in one complete divide cycle is

(P + Q)A + P (N A) = AQ + P N (13.106)
i.e., for every AQ + P N pulses in, we get 1 pulse out to the PD. Thus, the three counters
together eectively divide by AQ + P N . When the loop is locked
fo = [AQ + P N ]fr (13.107)
Recall that N must be greater than A for the method to work.
A frequently used divide ratio is P = 10, P + Q = 11. Then
fo = [10N + A] fr (13.108)
Suppose that A can be set to any value from 0 to 9. Then N must be at least 10 to keep
N > A. This means that the lowest output frequency is 100fr , but any integer multiple
larger than 100fr can be achieved. Thus the 10/11 dual modulus pre-scaler can be used
to obtain integer multiples of fr even when the output frequency is too large for a fully
programmable divider to handle.
Consider an example. Suppose that we need a frequency synthesizer to cover the fre-
quency range 100 MHz to 1099 MHz in 1 MHz increments. Since the step size is 1 MHz,
it is necessary to choose fr 1 MHz. Let fr = 1 MHz. Suppose that 1099 MHz is too
fast for a fully programmable divider, however a fast (P = 10, P + Q = 11) dual modulus
pre-scaler is available. Then we use the following parameters:
fr = 1 MHz (13.109)
N = 100
A ) 0 to 99
fo = [1000 + A] fr
Notice that the largest frequency at the input to the N and A counters will be 109.9
MHz.
Now, suppose we need to cover 1000.00 to 1000.99 MHz in 10 kHz increments. Then we
need fr = 10 kHz. The minimum divide ratio required would be
1000
= 105 (13.110)
0.01
fo (min)
=
fr
Since A must run from 0 to 99, Nmin is 100. Parameters that would work for this situation
are:
N = 10000 (13.111)
P = 10
P +Q = 11
432 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

The maximum frequency at the input to the fully programmable dividers would be 0.100099
MHz in this case.
Consider an example using parameters available from a commercially available synthe-
sizer chip. The chip has a 5-bit A counter, so that A can range from 0 to 31. The N counter
is a 13-bit counter and N can be set to any value in the range 3 to 8191. The dual-modulus
pre-scaler uses P = 8, P + Q = 9. The output frequency will be

fo = (8N + A)fr ,

where 0 A 31, 3 N 8191, and N > A. Suppose that it is necessary to design a

synthesizer that covers 230 240 MHz in 100 kHz steps and we wish to select possible values
for N and A in order to tune the synthesizer to a particular output frequency. For example,
to tune the synthesizer to the middle of the range (235 MHz) the eective divide ratio must
be 235/0.1 = 2350 = (8N + A). The N counter can be set to the integer part of 2350/8,
which is N = 293. Then, 8N = 8(293) = 2344, so the A counter must be set to A = 6. To
tune to 239.6 MHz, (N, A) = (299, 4) would work, as would (298, 12) or (297, 24).
13.6. REFERENCES 433

13.6 References
1. Best, Roland E., Phase Locked Loops, McGraw Hill, 1984.
2. Egan, William F., Frequency Synthesis by Phaselock, J. Wiley & Sons, New York,
1981.
3. Gardner, Floyd M., Phaselock Techniques, (2nd ed.), J. Wiley & Sons, New York,
1979.
4. Manassewitsch, Vadim J., Frequency Synthesizers, Theory and Design, J. Wiley &
Sons, New York, 1987.
5. Smith, Jack, Modern Communications Circuits, McGraw Hill, 1986.
434 CHAPTER 13. PHASE-LOCKED LOOPS (PLLS)

13.7 Homework Problems

1. Consider the first-order PLL shown in Figure 13.35. The gain constants for the VCO
and phase detector are Ko = 107 radians/Volt-second, and Kd = 10 Volt/radian,
respectively. The center frequency (free-running frequency) of the VCO is fc = 35.0
MHz. Assume that the loop is locked.

Figure 13.35: First-order PLL frequency synthesizer.

(a) If N = 10 and the reference frequency fr = 1.0 MHz, what is the output frequency,
fo ?
(b) What is the steady-state output voltage from the phase detector?
(c) What is the steady-state phase error e = r o ? Give your result in degrees.
Appendix A

Circuit Models for BJT and FET

A.1 Hybrid-pi equivalent circuit for bipolar junction tran-

sistor (BJT)

C
IB IC
B
+
Vbe- IE
E

Figure A.1: BJT

The following approximate relationship is useful when the base-emitter junction is for-
ward biased:
Vbe
IC = IS exp[ ], (A.1)
VT

where
kT
VT = ' 25mV at room temperature. (A.2)
q

The base current, Ib , is related to the collector current by

IC
Ib = . (A.3)

These relationships can be used to solve for the small-signal input resistance, r (see Fig-

435
436 APPENDIX A. CIRCUIT MODELS FOR BJT AND FET

ure A.2)

@Ib 1
r = [ ] Vbe =Vbeq (A.4)
@Vbe
VT
=
ICQ
.025
r '
ICQ

and the small-signal transconductance

@IC
gm = [ ] V =V (A.5)
@Vbe be beq
ICQ
=
VT

=
r

gm ' 40ICQ

where ICQ is the quiescent collector current.

A useful linear model for the behavior of small high frequency signals superimposed on
the DC bias point is the small-signal hybrid-pi model shown in Figure A.2. Typically ro

rx C
B C

+
V C r ro Co
gm V
E

Figure A.2: Hybrid-pi small-signal model for BJT

(tens to hundreds of k), rx (a few tens of ), and r > ro . Since ro and r are relatively
large resistances, and each is shunted by a capacitance, ro and r can usually be ignored at
high frequencies. On data sheets C is often given as Cob , which is the output capacitance
in the common-base configuration. Typical values for C range from a few tenths of a pF
to a few pF . Data sheets may not give C explicitly, but will indicate the value of fT for a
particular bias current, where
1 gm
fT = , (A.6)
2 C + C
A.2. HYBRID-PI EQUIVALENT CIRCUIT FOR FIELD EFFECT TRANSISTOR (FET)437

and fT is the frequency where the short-circuit current gain has a magnitude of unity. The
value of fT depends on gm and, therefore, on how the transistor is biased.
The -3dB frequency for the short-circuit current gain is denoted by f and is given by

1 1
f = (A.7)
2 r (C + C )

Note that the -3dB frequency and the unity gain frequency are related as follows:

fT
= (A.8)
f

For f < f the short-circuit current gain has magnitude , while for f > f the short-circuit
current gain is approximately fT /f .

A.2 Hybrid-pi equivalent circuit for field eect transistor

(FET)
Figure A.3 looks very similar to the BJT model (Figure A.2). The gate-source resistance,
rgs , is generally large compared to the impedance of Cgs . When this condition holds (i.e.
at suciently high frequencies) rgs can be omitted from the model.

Cdg
G D

vgs+ Cgs rgs rds Cds

gm vgs

Figure A.3: Hybrid-pi equivalent circuit for the FET

For a junction FET (JFET), the transconductance is proportional to the square root
of the drain current, ID . The proportionality constant depends on the saturation drain
current, IDSS , and the pincho voltage, VP , i.e.
2 p
gm = IDSS ID . (A.9)
|VP |

The parameters IDSS and VP are usually available from device data sheets.

A.3 Large-signal transconductance of a BJT with sinu-

soidal Vbe
When a BJT is driven with a sinusoidal signal such that the base-emitter voltage swing ap-
proaches or exceeds a few tens of mV, the collector current waveform becomes non-sinusoidal.
438 APPENDIX A. CIRCUIT MODELS FOR BJT AND FET

It is useful to study the eect of sinusoidal base-emitter voltage swing on the collector cur-
rent waveform. The following approximate relationship between base-emitter voltage Vbe
and collector current IC can be used as a starting point

IC = IS eVbe q/kT (A.10)

where kT /q is approximately 25mV for T=290K, and IS is a constant. Decompose the

base-emitter voltage and collector current into quiescent and time-varying components, i.e.,

IC = IDC + iC (A.11)

Vbe = VDC + vbe (A.12)

where lower-case letters refer to the time-varying component of the quantity. Later, we will
make the assumption that the transistor bias network acts to keep the DC component of
the collector current at a nearly constant value. Suppose that the time-varying component
of the base-emitter voltage is sinusoidal, i.e.,

Vbe = VDC + v1 cos!t (A.13)

and let x = v1 q/kT = v1 /25mV (at room temperature). Then

IC = IS exp[VDC q/kT ] exp[ x cos!t] (A.14)

The term exp[ x cos !t] is a non-sinusoidal periodic function of time and can be expanded
in a Fourier series. The series is
1
X
exp[ x cos !t] = Io (x) + 2 In (x) cos(n!t) (A.15)
n=1

where the coecients In (x) are values of the modified Bessel function of the first kind. Using
this relationship, the collector current waveform can be written as
X
IC = IDCo [Io (x) + 2 In (x) cos(n!t)] (A.16)
n=1

Here IDCo is the DC component of collector current when the time-varying component of
the input signal is equal to zero (v1 = 0). The DC component of the collector current when
the time-varying component of the input signals is not zero is given by

IDC = IDCo Io (x). (A.17)

This function is plotted in Figure A.4, which shows that Io (x) grows very rapidly when the
base-emitter voltage swing exceeds a few tens of mV. This analysis has assumed that the DC
component of the base-emitter voltage is held constant. Practical amplifier and oscillator
circuits will employ either constant current source bias for IDC or negative feedback (by
including a resistor in series with the emitter and ground) to force IDC to be nearly constant.
In such cases the DC component of Vbe (denoted by VDC ) will decrease with increasing v1
such that IDC is more or less independent of v1 . With constant-current bias we can write:

X1
In (x)
IC = IDC [1 + 2 cos(n!t)] (A.18)
I (x)
n=1 o
A.3. LARGE-SIGNAL TRANSCONDUCTANCE OF A BJT WITH SINUSOIDAL VBE 439

700
I0(x) 600
500
400
300
200
100

2 4 6 8 10
x

Figure A.4: The function Io (x) governs how the DC component of the collector current
grows as the amplitude of sinusoidal base-emitter voltage increases (x = v1 q/kT ), assuming
that the DC component of the base-emitter voltage is held constant.

where IDC is treated as a constant. Examination of A.18 shows that the collector current
waveform has components at DC, the fundamental frequency, and harmonics of the funda-
mental frequency. The relative amplitude of the harmonics and fundamental gives an indica-
tion of how sinusoidal the collector current waveform will be. Assuming constant-current
bias, the amplitude of the fundamental component is proportional to 2I1 (x)/I0 (x). This
quantity is plotted in Figure A.5. The amplitude of the fundamental approaches a constant
when the base-emitter voltage swing exceeds 25 mV (i.e. for x > 1). The relative strengths
of the second and third harmonics to the fundamental (I2 (x)/I1 (x) and I3 (x)/I1 (x), respec-
tively) are also shown in Figure A.5. Notice that the harmonic amplitudes grow rapidly

1.75 2I1 (x)

I0 (x)
1.5

1.25 I3 (x)
1 I1 (x)
0.75
I2 (x)
0.5 I1 (x)
0.25

5 10 15 20

Figure A.5: 2I1 (x)/Io (x) is the relative amplitude of the fundamental component of the
collector current. The other curves show the ratio of second and third harmonic amplitudes
to the fundamental amplitude.

when x is greater than 1 (v1 > 25 mV). At large values of x the fundamental amplitude
approaches twice the DC bias current, and the harmonic amplitudes approach that of the
fundamental. Thus it becomes apparent that the base-emitter voltage swing must be kept
as small as possible if the collector current waveform is to be sinusoidal.
This approximate analysis of the nonlinear characteristics of the BJT can also help us to
gain some intuitive feeling for the saturation mechanism that limits the growth of oscillations
in self-limiting oscillators, and for the reduction in apparent gain that results when a BJT
amplifier is driven by a large amplitude input signal. For these purposes we consider only
440 APPENDIX A. CIRCUIT MODELS FOR BJT AND FET

the collector current component at the fundamental frequency, under the assumption that
a resonant network eectively removes the harmonics. Then we can compute a large-signal
transconductance for the transistor. With a sinusoidal input signal the transconductance
is simply the ratio of the amplitudes of the fundamental component of iC and v1 . The
small-signal transconductance can be obtained in the limit as x approaches 0, i.e.,

2 I1 (x) IDC
gm = lim IDC = (A.19)
x!0 v1 I0 (x) kT /q

At room temperature kT /q = 25 mV, so

IDC
gm = 40IDC (A.20)
25 mV
The large signal transconductance is

2 I1 (x) q 2 I1 (x) 2 I1 (x)

Gm (x) = IDC = IDC = gm (A.21)
v1 I0 (x) kT x I0 (x) x I0 (x)

The ratio of the large signal to small-signal transconductance is shown in Figure A.6. This
1

0.8

0.6
Gm (x)
gm 0.4

0.2

2 4 6 8 10
x

Figure A.6: Ratio of large signal to small-signal transconductance

result shows that the large signal transconductance of the transistor decreases from the
small-signal value as the base-emitter voltage swing (x) increases. Thus in an oscillator
application, as the oscillation amplitude grows, the eective transconductance decreases.
The oscillations will continue to grow until the transconductance has been reduced to a value
that causes the (large-signal) loop gain to be equal to 1. The large-signal transconductance
is also important in determining the behavior of stable (non-oscillating) transistor circuits
under large signal conditions. For example, the decrease of transconductance under large
signal excitation is responsible for gain compression in transistor amplifiers when large
signals are applied. These eects will discussed in some detail in Chapter 12.
Appendix B

Three-winding Transformer

The three-winding transformer finds many applications in RF circuits. Examples include

balanced-to-unbalanced conversions (baluns), power splitters and combiners and voltage
adders and subtractors. This chapter will explore some of these applications. Through-
out this discussion it will be assumed that the transformer is ideal in other words, we
assume that the device is lossless, that the windings are perfectly coupled, and that the
self-inductance of each winding is infinite. In practice, perfect coupling can be approxi-
mated by constructing all three windings on a common high-permeability core. Infinite
self-inductance is a good approximation if the inductance of each winding is large compared
to the impedance connected across the winding
The schematic representation for an ideal three-winding transformer is shown in Figure
B.1. Two parallel lines running alongside the three windings remind us that a practical
implementation of the transformer requires all three coils to be wound on a common core
that has large relative magnetic permeability so that all three coils are linked by the same
magnetic flux. By virtue of the perfect coupling between windings, the voltages across
windings b and c can be written in terms of the voltage across winding a as follows:

Nb Nc
Vb = Va , Vc = Va (B.1)
Na Na

Since the transformer is assumed to be lossless, the total power delivered to the device must
be zero, i.e.
<(Va Ia + Vb Ib + Vc Ic ) = 0,
where <() takes the real part of its argument. Using the voltage relationships, the voltages
can be written in terms of the voltage across any one of the windings. Using Va the lossless
condition becomes
Nb Nc
<(Va (Ia + I + I )) = 0.
Na b Na c
The lossless condition must hold for any applied voltage, Va , so the currents must satisfy

Na Ia + Nb Ib + Nc Ic = 0. (B.2)

Equations B.1 and B.2 are called the ideal transformer equations, and they fully characterize
the ideal 3-winding transformer.

441
442 APPENDIX B. THREE-WINDING TRANSFORMER

Ia
+
Va Na

Ib
+
Vb Nb

Ic
+
Vc Nc

Figure B.1: Schematic representation of a three-winding transformer. The polarity of the

windings is indicated by a dot. The dot convention is such that when current flows into the
dot in each winding, the resulting magnetic fluxes within the core add constructively. The
turns ratio is Na : Nb : Nc for windings a, b, and c, respectively.

V1 N :1
+ Z1 1

Ia +
Va 3 Z3
V4 + Ic +
+ Z4 4
Vc V3
+
Vb
V2 Ib
+
kk Z2 2

Figure B.2: The three-winding transformer configured as a 4-port device.

B.1. CONJUGATE PORTS 443

For simplicitys sake, well consider a system with Na : Nb : Nc = N : N : 1, and with

windings a and b connected so that they share a common terminal as shown in Figure B.2.
This configuration results in a 4-port device.
Taking into account the current directions defined in Figure B.2, the ideal transformer
relations can be written

V a = Vb = N Vc (B.3)
N (Ia + Ib ) = Ic (B.4)

The loop equations are

V1 + Z1 Ia + Va + (Ia Ib )Z4 + V4 = 0 (B.5)

V4 + (Ib Ia )Z4 + Vb + Z2 Ib + V2 = 0 (B.6)

Vc + I c Z 3 + V3 = 0. (B.7)

Use B.3 andB.7 in B.5 and B.6 to eliminate Va , Vb , Vc :

Ia (Z1 + Z4 ) + Ib ( Z4 ) + Ic (N Z3 ) = V1 V4 N V3 (B.8)

Ia ( Z4 ) + Ib (Z2 + Z4 ) + Ic (N Z3 ) = V4 V2 N V3 (B.9)

Equations B.4, B.8 and B.9 can be written in matrix form:

2 32 3 2 3
Z1 + Z4 Z4 N Z3 Ia V1 V 4 N V3
4 Z4 Z 2 + Z 4 N Z3 5 4 I b 5 = 4 V4 V 2 N V3 5 (B.10)
N N 1 Ic 0

B.1 Conjugate Ports

We now consider some useful special cases. It is possible to select the termination impedances
so that there is isolation between ports. For example, suppose it is desired to isolate port 4
from port 3 so that there will be no output at port 4 when a signal is applied at port 3. To
proceed, we turn o V1 , V2 , V4 and find the condition for no output at port 4.
The current at port 4 will be Ia Ib . Using Cramers Rule to solve for Ia :

N V3 Z4 N Z3

N V3 Z4 + Z2 N Z3

Da 0 N 1
Ia = =
D D
where D is the system determinant and Da is the determinant of the matrix formed by
replacing the first column with the source column vector (the right-hand side of equation
444 APPENDIX B. THREE-WINDING TRANSFORMER

B.10). Evaluating Da :

Da = N V3 [ (Z4 + Z2 ) N 2 Z3 ] + Z4 [N V3 ] + N Z3 [ N 2 V3 ] (B.11)

= V3 N (2 Z4 + Z2 )

Similarly, Ib is obtained from:

Z1 + Z4 N V3 N Z3

Z4 N V3 N Z3

Db N 0 1
Ib = =
D D

Db = (Z1 + Z4 )[N V3 ] + N V3 [Z4 N 2 Z3 ] + N Z3 [N 2 V3 ] (B.12)

= V3 N (2 Z4 + Z1 )

The current flowing in the impedance Z4 is

Ia Ib = (Da Db ) /D (B.13)
= V3 N (Z2 Z1 ) /D.

The current in Z4 will be zero if Z1 = Z2 . Thus, port 4 is isolated from port 3 if Z1 = Z2 .

It is left as an exercise to show that the isolation works both ways, i.e. that port 3 will be
isolated from port 4 when Z1 = Z2 . When ports 3 and 4 are isolated, we say that ports 3
and 4 are conjugate.
Suppose we wish to isolate port 1 from port 2. To find the necessary constraint on the
terminating impedances, set V1 = V3 = V4 = 0 and solve for Ia due to V2 . Setting Ia = 0
will yield the desired condition:

0 Z4 N Z3

V2 Z4 + Z2 N Z3

Da 0 N 1
Ia = =
D D

= [Z4 V2 + N Z3 ( N V2 )] /D (B.14)

= V2 [Z4 N 2 Z3 ] /D

Thus, port 1 will be isolated from port 2 if

Z4 = N 2 Z3 (B.15)
B.2. HYBRID TRANSFORMER 445

Again, it is left as an exercise to show that condition B.15 also causes port 2 to be isolated
from port 1, i.e. Ib = 0 with V1 6= 0 and V2 = V3 = V4 = 0. That is, the isolation works
both ways. When ports 1 and 2 are isolated, we say that ports 1 and 2 are conjugate.
If ports 3 and 4 are conjugate and ports 1 and 2 are conjugate then the system is called
the biconjugate transformer.

B.2 Hybrid Transformer

Recall that the conditions for a biconjugate system are

Z1 = Z2 (B.16)

1
Z3 = Z4 . (B.17)
N2
The system of equations for the biconjugate transformer is
2 32 3 2 3
Z1 + Z4 Z4 Z4 /N Ia V1 V 4 N V3
4 Z4 Z4 + Z1 Z4 /N 5 4 Ib 5 = 4 V4 V 2 N V3 5 (B.18)
N N 1 Ic 0

In addition to isolation between ports, it is sometimes desired to have all 4 ports matched
for maximum power transfer. This means that a conjugate match exists between each port
and its terminating impedance. Consider the input impedance seen by the source that is
connected to port 1. For a conjugate match at port 1, we require that Zin1 = Z1 . The
impedance Zin1 is equal to the voltage at terminal 1, which is V1 Ia Z1 , divided by the
current flowing into this terminal, which is Ia . Thus, Zin1 = VIa1 Z1 . When solving for Ia
we set V2 , V3 , V4 = 0 in the equations B.28, because only port 1 is driven when calculating
Zin1 . Solving for Ia :
V1 Z4 Z4 /N

0 Z4 + Z1 Z4 /N

0 N 1
Ia =
D
Ia = (2Z4 + Z1 )V1 /D (B.19)
where
D= (2Z4 + Z1 )2 (B.20)
Thus
V1
= 2Z4 + Z1 . (B.21)
Ia
The input impedance is then
Zin1 = 2Z4 . (B.22)
For a conjugate match at port 1 we need

Zin1 = Z1 (B.23)
446 APPENDIX B. THREE-WINDING TRANSFORMER

so
1
Z4 = Z1 (B.24)
2
Notice that the combination of the biconjugate conditions (equations B.17) and equation
B.24 determine the terminations at ports 2, 3, and 4 once the termination at port 1 is
specified. There are no additional degrees of freedom left once biconjugacy and a conjugate
match at one port are enforced. This implies that the input impedances at the other ports
(2, 3, and 4) are determined once the conjugate match at port 1 is enforced. It is not hard to
show that a biconjugate system with one port matched for maximum power transfer will au-
tomatically be conjugate matched at all 4 ports. How wonderful! A biconjugate transformer
that is conjugately matched at all 4 ports is called a hybrid transformer. The terminating
impedances for a hybrid transformer system must satisfy the following conditions:

Z1 = Z (B.25)
Z2 = Z (B.26)
2
Z3 = Z /2N
Z4 = Z /2

If port 1 is terminated with a resistance, R. Then port 2 must also be terminated with R,
port 3 must be terminated with R/(2N 2 ) and port 4 must be terminated with R/2. Notice
that if the turns ratio is chosen to be N = p12 then Z1 = Z2 = Z3 = R and Z4 = R/2. We
will assume that the port terminations are resistive and that N = p12 in the discussions to
follow.

B.3 Applications of the Hybrid Transformer

B.3.1 Power Splitters
B.3.1.1 180-degree splitter
Suppose a signal is applied to port 3 as in Figure B.3.
p
R 1: 2
1
Ia +
3 R
Va
R/2 + Ic +
4 Vc V3
+
Vb
Ib

R 2

Figure B.3: A hybrid transformer driven at port 3.

B.3. APPLICATIONS OF THE HYBRID TRANSFORMER 447

The system of equations for this hybrid transformer is

2 3 1
32 3 2 3
p1 R p1 V3
2R 2R 2 Ia 2
6 1R 3 p1 R 7
4 2 2R 2 5 4 Ib 5 = 4 p1 V3
2
5 (B.27)
1 1 Ic
p
2
p
2
1 0
Since port 4 is conjugate to port 3, we already know that there is no response at port 4.
Hence, the current leaving port 4 must be zero, which implies that Ib = Ia . Lets solve for
Ia :
p1 V3 1 p1 R
2 2R 2

p1 V3 3 p1 R
2 2R 2

0 p1 1
2
Ia = .
D

The solution is p
2
Ia = Ib = V3.
4R
Denoting the voltages at terminals 1 and 2 by V1 and V2 , respectively, we have V1 = Ia R
and V2 = Ib R, or p p
2 2
V1 = V3 , V2 = V3 .
4 4
Hence, the voltages at ports 1 and 2 have equal amplitudes, and are 180 degrees out of
phase. The power delivered to terminations 1 and 2 is
1 |V1 |2 1 |V3 |2
P1 = P 2 = = .
2 R 16 R
Note that P1 and P2 are each exactly one half of the power available from the source
connected to port 3. This configuration of the hybrid transformer is called the 180-degree
power splitter.

B.3.1.2 In-phase splitter

Suppose a signal is applied to port 4 as in Figure B.4. The system of equations for this
hybrid transformer is
2 3 1
32 3 2 3
p1 R
2R 2R 2 Ia V4
6 1R 3 p1 R 7
4 2 2R 2 5 4 I b 5 = 4 V4 5 (B.28)
p1 p1 1 I c 0
2 2

Since port 4 is conjugate to port 3, we already know Ic = 0. Lets solve for Ia :

1 p1 R
V4 2R 2

3 p1 R
V4 2R 2

0 p1 1
2
Ia = .
D
448 APPENDIX B. THREE-WINDING TRANSFORMER

p
R 1: 2
1
Ia +
Va 3
V4 R/2 +
+ 4 Vc Ic R
+
Vb
Ib

R 2

Figure B.4: A hybrid transformer driven at port 4.

The solution is
1
Ia = V4 .
2R
Similarly,
1
Ib = +V4 .
2R
The voltages at terminals 1 and 2 are V1 = Ia R and V2 = Ib R, or

1 1
V1 = V4 , V2 = V4 .
2 2
Hence, the voltages at ports 1 and 2 have equal amplitudes, and are in phase. The power
delivered to terminations 1 and 2 is

1 |V1 |2 1 |V4 |2
P1 = P2 = = .
2 R 8 R
Note that P1 and P2 are each exactly one half of the power available from the source
connected to port 4. This configuration of the hybrid transformer is called the in-phase
power splitter.

B.3.2 Sum or Dierence Combiners using a Hybrid Transformer

The hybrid transformer has the property that the phase shift between three of the four
ports is zero and the phase shift to the remaining port will be 180 . This property can be
used to realize both adding and subtracting signal combiners. For example, consider the
response to applied signals V1 and V2 as shown in Figure B.5. 1 The output at port 3 will
1 Note that since ports 1 and 2 are conjugate, the signal generators connected to ports 1 and 2 are isolated

this means that the signal from generator 1 will not appear at the output terminals of generator 2 and
vice versa. Without this isolation, the signal from generator 1 could mix with the signal from generator 2
in the output stage of either generator to produce unwanted intermodulation products.
B.4. REFERENCES 449

V1 p
R 1: 2
+ 1
Ia +
Va 3
R/2 +
4 Vc Ic R
+
Vb
V2 Ib
+
R 2

Figure B.5: Hybrid transformer used to combine two signals.

be proportional to Ic where
3 1
2R 2R V1

1 3
2R 2R V2

p1 p1 0
Dc 2 2
Ic = = .
D D
The solution is p
2
Ic = (V1 V2 )
4R
The voltage at port 3 is V3 = Ic R, so
p
2
V3 = (V1 V2 ).
4
The outport at port 3 is proportional to the dierence between the applied signals.
The output at port 4 will be proportional to Ia Ib . Using the fact that ports 1 and 2
are conjugate, and noting that generators 1 and 2 each see a matched impedance, R, it is
obvious that Ia = 2R
V1
and Ib = 2R V2
. Hence
R 1
V4 = (Ia Ib ) = (V1 + V2 ).
2 4
The output at port 4 is proportional to the sum of the applied signals.

B.4 References
1. Smith, Jack, Modern Communications Circuits, Second Edition, WCB/McGraw-Hill,
1998.
2. Sartori, Eugene P., Hybrid Transformers, IEEE Transactions on Parts, Materials
and Packaging, Vol. PMP-4, No. 3, September 1968, pp 59-66.
450 APPENDIX B. THREE-WINDING TRANSFORMER
Appendix C

Appendix: Useful Constants and

Trigonometric Identities

speed of light in free-space c = 2.998 108 m s 1

permittivity of free-space 0 = 8.854 10 12 F m 1
permeability of free-space 0 = 4 10 7 H m 1
Boltzmann constant k = 1.381 10 23 J K 1
elementary charge q = 1.602 10 19 C
Planck constant h = 6.626 10 34 J s

Table C.1: Some useful constants

451
452APPENDIX C. APPENDIX: USEFUL CONSTANTS AND TRIGONOMETRIC IDENTITIES

eja e ja
sin a = 2j
ja ja
cos a = e +e 2
sin a
tan a = cos a
sec a = cos1 a
csc a = sin1 a
sin(a b) = sin a cos b cos a sin b
cos(a b) = cos a cos b sin a sin b
tan atan b
tan(a b) = 1tan a tan b
sin a sin b = 2 cos(a b) 12 cos(a + b)
1

sin a cos b = 12 sin(a + b) + 12 sin(a b)

cos a cos b = 12 cos(a + b) + 12 cos(a b)
cos2 a = 12 (1 + cos(2a))
sin2 a = 12 (1 cos(2a))
cos3 a = 34 cos a + 14 cos(3a)
sin3 a = 34 sin a 14 sin(3a)
a a
sinh a = e 2e
a a
cosh a = e +e 2
sinh a ea e a
tanh a = cosh a = ea +e a

Table C.2: Some trigonometric identities.