INSTRUCTOR COPY

ECE350

Peter Driessen

December 13, 2021

 Contents

1 Introduction 21
INSTRUCTOR COPY

1.1 Telecommunications systems overview . . . . . . . . . . . . . . . . . . 21

1.1.1 Telecommunications technology . . . . . . . . . . . . . . . . . 21

1.1.1.1 Terminology . . . . . . . . . . . . . . . . . . . . . . . 22

1.1.1.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . 23

1.1.1.3 Standards organizations . . . . . . . . . . . . . . . . 24

1.1.2 Telecommunication systems . . . . . . . . . . . . . . . . . . . 24

1.1.2.1 Point-to-point system . . . . . . . . . . . . . . . . . 24

1.1.2.2 Message source . . . . . . . . . . . . . . . . . . . . . 25

1.1.2.3 Transducer at sender . . . . . . . . . . . . . . . . . . 25

1.1.2.4 Electronic signals . . . . . . . . . . . . . . . . . . . . 26

1.1.2.5 Channel . . . . . . . . . . . . . . . . . . . . . . . . . 26

1.1.2.6 Transducer at receiver . . . . . . . . . . . . . . . . . 27

1.1.3 Analog communication systems . . . . . . . . . . . . . . . . . 27

1.1.3.1 Carrier waves . . . . . . . . . . . . . . . . . . . . . . 28

1.1.3.2 Modulation . . . . . . . . . . . . . . . . . . . . . . . 30

1.1.3.3 Upconversion and amplification . . . . . . . . . . . . 30

1
 CONTENTS 2

1.1.3.4 System examples . . . . . . . . . . . . . . . . . . . . 30

1.1.3.5 Noise and interference . . . . . . . . . . . . . . . . . 31

1.1.4 Digital communications system . . . . . . . . . . . . . . . . . 31

1.1.4.1 A/D, sampling, quantization . . . . . . . . . . . . . 32

1.1.4.2 Types of message signals . . . . . . . . . . . . . . . . 32

1.1.4.3 Bitstream . . . . . . . . . . . . . . . . . . . . . . . . 35

1.1.4.4 Modulation . . . . . . . . . . . . . . . . . . . . . . . 39

1.1.4.5 Upconversion . . . . . . . . . . . . . . . . . . . . . . 40
INSTRUCTOR COPY

1.1.4.6 Amplification . . . . . . . . . . . . . . . . . . . . . . 40

1.1.4.7 System examples . . . . . . . . . . . . . . . . . . . . 41

1.1.5 Channels, Power and Bandwidth . . . . . . . . . . . . . . . . 41

1.1.5.1 Quality . . . . . . . . . . . . . . . . . . . . . . . . . 42

1.1.5.2 Channels and path loss . . . . . . . . . . . . . . . . . 44

1.1.5.3 Receiver . . . . . . . . . . . . . . . . . . . . . . . . . 46

1.2 Communications networks . . . . . . . . . . . . . . . . . . . . . . . . 47

1.2.1 Nodes and Channels . . . . . . . . . . . . . . . . . . . . . . . 48

1.2.2 Circuit and Packet Switching . . . . . . . . . . . . . . . . . . 48

1.2.3 Queueing and buffers . . . . . . . . . . . . . . . . . . . . . . . 50

1.2.4 Types of networks . . . . . . . . . . . . . . . . . . . . . . . . 50

1.2.5 Dedicated and Shared channels . . . . . . . . . . . . . . . . . 50

1.2.5.1 Dedicated (circuit switched) channels . . . . . . . . . 50

1.2.5.2 Shared (packet switched) channels . . . . . . . . . . 51

1.2.6 Telephone network architecture . . . . . . . . . . . . . . . . . 52

 CONTENTS 3

1.2.7 Internet architecture . . . . . . . . . . . . . . . . . . . . . . . 52

1.2.8 Protocols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

1.2.8.1 TCP/IP - moving data on the Internet . . . . . . . . 55

1.2.8.2 Types of nodes on the Internet . . . . . . . . . . . . 57

1.3 Signals in time and frequency domain . . . . . . . . . . . . . . . . . . 58

1.3.1 Introduction to Signals and Systems . . . . . . . . . . . . . . 58

1.3.1.1 Signals . . . . . . . . . . . . . . . . . . . . . . . . . . 59

1.3.1.2 Systems . . . . . . . . . . . . . . . . . . . . . . . . . 59
INSTRUCTOR COPY

1.3.1.3 System implementation . . . . . . . . . . . . . . . . 60

1.3.2 Examples of Signals and Systems . . . . . . . . . . . . . . . . 60

1.3.3 Sine waves, a fundamental signal . . . . . . . . . . . . . . . . 61

1.3.4 Representing a human voice signal . . . . . . . . . . . . . . . 63

1.3.4.1 Time domain video . . . . . . . . . . . . . . . . . . . 63

1.3.4.2 Frequency-domain video . . . . . . . . . . . . . . . . 64

1.3.4.3 Time-frequency video . . . . . . . . . . . . . . . . . 64

1.4 Fourier transform and Discrete Fourier Transform . . . . . . . . . . . 65

1.4.1 Fourier transform concept and definition . . . . . . . . . . . . 65

1.4.1.1 Complex numbers, complex exponentials and phasors 66

1.4.1.2 Fourier transform interpretation using real signals . . 70

1.4.1.3 Fourier transform interpretation using complex signals 75

1.4.2 Discrete Fourier Transform (DFT) . . . . . . . . . . . . . . . 81

1.4.2.1 Sampling the Fourier transform to obtain DFT . . . 82

1.4.2.2 Interpretation of DFT . . . . . . . . . . . . . . . . . 85

 CONTENTS 4

1.4.2.3 Mirror image . . . . . . . . . . . . . . . . . . . . . . 85

1.4.2.4 Definition of DFT in terms of LTI frequency response 87

1.4.2.5 Calculation of DFT . . . . . . . . . . . . . . . . . . 89

1.4.2.6 Zero-padding . . . . . . . . . . . . . . . . . . . . . . 89

1.4.2.7 Review . . . . . . . . . . . . . . . . . . . . . . . . . 90

1.5 Algorithm to generate a spectrogram . . . . . . . . . . . . . . . . . . 91

1.5.1 Spectrogram examples . . . . . . . . . . . . . . . . . . . . . . 96

1.5.2 The need for complex signals . . . . . . . . . . . . . . . . . . 98

INSTRUCTOR COPY

1.5.2.1 Zooming into the spectrogram . . . . . . . . . . . . . 99

1.5.2.2 Instantaneous amplitude and known phase . . . . . . 100

1.6 Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

1.6.1 Discrete-time messages m(t) as sum of cosine waves . . . . . . 101

1.6.2 Message types, analog and digital . . . . . . . . . . . . . . . . 102

1.6.2.1 Analog message (voice or music) . . . . . . . . . . . 103

1.6.2.2 Digital message, a sequence of bits . . . . . . . . . . 103

1.7 Chapter review questions . . . . . . . . . . . . . . . . . . . . . . . . . 104

2 IQ Signals 106

2.1 Carrier waves, amplitude and phase in the time domain . . . . . . . . 106

2.1.1 Carrier wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

2.1.1.1 real carrier waves . . . . . . . . . . . . . . . . . . . . 108

2.1.1.2 complex carrier waves . . . . . . . . . . . . . . . . . 109

2.1.2 Modulated carrier wave with time-varying amplitude and phase 113

2.2 Complex envelope (complex baseband) . . . . . . . . . . . . . . . . . 118

 CONTENTS 5

2.2.1 Examples of complex baseband signals in the time domain . . 121

2.3 Summary of complex baseband and real passband . . . . . . . . . . . 125

2.4 Complex baseband signals in the frequency domain . . . . . . . . . . 126

2.4.1 Fourier transform concept and definition . . . . . . . . . . . . 127

2.4.2 Fourier transform of complex conjugate . . . . . . . . . . . . . 127

2.4.2.1 Properties of complex conjugate . . . . . . . . . . . . 128

2.4.2.2 Fourier transform of complex conjugate . . . . . . . 129

2.4.2.3 Example complex baseband spectrum and its complex

INSTRUCTOR COPY

conjugate . . . . . . . . . . . . . . . . . . . . . . . . 131

2.5 Real passband signals in the frequency domain . . . . . . . . . . . . . 134

2.5.1 Fourier transform shifting properties . . . . . . . . . . . . . . 135

2.5.2 Fundamental result arising from Fourier transform shifting prop-

erties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

2.5.2.1 Example of shifted complex baseband spectrum and

its complex conjugate . . . . . . . . . . . . . . . . . 137

2.6 Analytic signals in time and frequency domain . . . . . . . . . . . . . 140

2.7 Example complex baseband signals in time and frequency domains . . 141

2.7.1 Complex exponentials at discrete frequencies . . . . . . . . . . 141

2.7.2 Continuous-time spectrum . . . . . . . . . . . . . . . . . . . . 143

2.7.3 Discrete exponentials plus continuous time spectrum . . . . . 143

2.7.4 Discrete-time spectrum . . . . . . . . . . . . . . . . . . . . . . 144

2.7.5 Complex baseband signals containing multiple messages . . . . 144

2.7.6 Complex baseband with zero imaginary component . . . . . . 145

2.7.6.1 General real message m(t) . . . . . . . . . . . . . . . 145

2.7.6.2 Frequency-shifted message . . . . . . . . . . . . . . . 148

 CONTENTS 6

2.7.6.3 Multitone message m(t) . . . . . . . . . . . . . . . . 150

2.7.6.4 Dual tone message m(t) . . . . . . . . . . . . . . . . 150

2.7.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

2.8 Application to sampled digital systems (software radio systems) . . . 155

2.8.1 Discrete Fourier Transform . . . . . . . . . . . . . . . . . . . . 155

2.8.2 Frequency shifting by subsampling . . . . . . . . . . . . . . . 156

2.8.3 Complex IQ sampling . . . . . . . . . . . . . . . . . . . . . . 156

2.9 Transmitter and receiver block diagrams . . . . . . . . . . . . . . . . 157

INSTRUCTOR COPY

2.9.1 Radio transmitter (modulator) . . . . . . . . . . . . . . . . . . 158

2.9.1.1 Complex signal implementation - complex passband

output . . . . . . . . . . . . . . . . . . . . . . . . . . 158

2.9.1.2 Complex signal implementation - real passband output 159

2.9.1.3 Real signal implementation - real passband output . 161

2.9.2 Radio receiver (demodulator) . . . . . . . . . . . . . . . . . . 162

2.9.2.1 Complex signal implementation - complex passband

input . . . . . . . . . . . . . . . . . . . . . . . . . . 163

2.9.2.2 Real signal implementation - real signal input . . . . 164

2.9.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

2.9.4 Summary 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

2.10 Analog IQ receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

2.10.1 Overview and block diagram . . . . . . . . . . . . . . . . . . . 169

2.10.2 USRP daughterboard receiver operation . . . . . . . . . . . . 171

2.10.2.1 Receiver operation in real notation . . . . . . . . . . 172

2.10.2.2 Receiver operation in complex notation . . . . . . . . 173

 CONTENTS 7

2.10.3 Another approach using digital IQ receiver . . . . . . . . . . . 176

2.10.4 Super-heterodyne Receiver . . . . . . . . . . . . . . . . . . . . 177

2.10.4.1 Superheterodyne receiver block diagram . . . . . . . 177

2.10.4.2 Frequency shift up and down . . . . . . . . . . . . . 178

2.10.4.3 Receiver tuning . . . . . . . . . . . . . . . . . . . . . 179

2.10.4.4 Image frequency . . . . . . . . . . . . . . . . . . . . 180

2.10.4.5 Superheterodyne with complex signals . . . . . . . . 181

2.10.4.6 Real signal analysis of IQ mixer . . . . . . . . . . . . 181

INSTRUCTOR COPY

2.11 Radio tuning, selecting a particular signal . . . . . . . . . . . . . . . 182

2.12 IQ imbalance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186

2.13 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188

2.14 Chapter review questions . . . . . . . . . . . . . . . . . . . . . . . . . 189

3 Amplitude modulation (AM) 193

3.1 Amplitude modulation with full carrier . . . . . . . . . . . . . . . . . 194

3.1.1 AM signals in time domain with general message . . . . . . . 196

3.1.1.1 General message in time domain . . . . . . . . . . . 196

3.1.1.2 AM complex baseband s̃(t) . . . . . . . . . . . . . . 196

3.1.1.3 AM analytic signal s+ (t) . . . . . . . . . . . . . . . . 198

3.1.1.4 AM real passband s(t) . . . . . . . . . . . . . . . . . 199

3.1.2 General message spectrum . . . . . . . . . . . . . . . . . . . . 200

3.1.3 AM spectrum for general message . . . . . . . . . . . . . . . . 201

3.1.3.1 AM complex baseband spectrum S̃(f ) . . . . . . . . 202

3.1.3.2 AM analytic signal spectrum S+ (f ) . . . . . . . . . . 203

 CONTENTS 8

3.1.3.3 AM real passband spectrum S(f ) . . . . . . . . . . . 204

3.1.4 AM waveform with dual tone message . . . . . . . . . . . . . 206

3.1.4.1 Dual tone message m(t) . . . . . . . . . . . . . . . . 206

3.1.4.2 AM complex baseband signal with dual tone m(t) . . 207

3.1.4.3 AM analytic signal with dual tone m(t) . . . . . . . 207

3.1.4.4 AM real passband signal with dual tone m(t) . . . . 209

3.1.4.5 Summary of dual-tone AM . . . . . . . . . . . . . . 211

3.1.5 AM waveform with single tone message . . . . . . . . . . . . . 212

INSTRUCTOR COPY

3.1.5.1 AM time domain waveform for single tone message . 213

3.1.5.2 AM spectrum for single tone message . . . . . . . . . 216

3.1.6 Observation interval . . . . . . . . . . . . . . . . . . . . . . . 219

3.1.7 AM with single tone message in only one sideband . . . . . . . 219

3.1.8 Power in carrier wave and sidebands . . . . . . . . . . . . . . 221

3.1.9 Overmodulation . . . . . . . . . . . . . . . . . . . . . . . . . . 223

3.1.10 Digital messages transmitted using AM . . . . . . . . . . . . . 225

3.2 AM transmitters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226

3.2.1 Generating an AM signal digitally in software . . . . . . . . . 226

3.2.2 Analog-only method to generate an AM signal . . . . . . . . . 227

3.3 AM receivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230

3.3.1 Digital software AM receiver in complex notation . . . . . . . 231

3.3.1.1 Received AM signal at receiver input . . . . . . . . . 231

3.3.1.2 Demodulation of complex passband AM signal to ob-

tain message . . . . . . . . . . . . . . . . . . . . . . 231
 CONTENTS 9

3.3.1.3 Demodulation of real passband AM signal to obtain

the message . . . . . . . . . . . . . . . . . . . . . . . 233

3.3.2 Analog AM receiver . . . . . . . . . . . . . . . . . . . . . . . . 234

3.3.2.1 Diode-resistor-capacitor circuit . . . . . . . . . . . . 234

3.3.2.2 Implementing the analog AM receiver algorithm in

software . . . . . . . . . . . . . . . . . . . . . . . . . 236

3.3.3 Tuning in (selecting) one particular AM signal with the USRP 237

3.4 Chapter review questions . . . . . . . . . . . . . . . . . . . . . . . . . 238

INSTRUCTOR COPY

4 Double sideband suppressed carrier (DSB-SC) 241

4.1 Introduction to DSB-SC . . . . . . . . . . . . . . . . . . . . . . . . . 241

4.2 DSB-SC transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . 242

4.2.1 DSB-SC signal mathematics . . . . . . . . . . . . . . . . . . . 242

4.2.1.1 Message signal m(t) . . . . . . . . . . . . . . . . . . 242

4.2.1.2 DSB-SC complex baseband s̃(t) . . . . . . . . . . . . 243

4.2.1.3 DSB-SC analytic signal s+ (t) . . . . . . . . . . . . . 243

4.2.1.4 DSB-SC real passband s(t) . . . . . . . . . . . . . . 244

4.2.2 DSB-SC spectrum for general message . . . . . . . . . . . . . 244

4.2.2.1 Message spectrum M (f ) . . . . . . . . . . . . . . . . 245

4.2.2.2 DSB-SC complex baseband spectrum S̃(f ) . . . . . . 246

4.2.2.3 DSB-SC analytic signal spectrum S+ (f ) . . . . . . . 247

4.2.2.4 DSB-SC real passband spectrum S(f ) . . . . . . . . 248

4.2.3 DSB-SC Modulation starting with real passband . . . . . . . . 249

4.2.3.1 DSB-SC in time domain . . . . . . . . . . . . . . . . 250

4.2.3.2 DSB-SC spectrum . . . . . . . . . . . . . . . . . . . 251

 CONTENTS 10

4.2.3.3 Observation interval . . . . . . . . . . . . . . . . . . 253

4.3 DSB-SC receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254

4.3.1 Digital software DSB receiver in complex notation . . . . . . . 254

4.3.1.1 Received DSB-SC signal at receiver input . . . . . . 254

4.3.1.2 Demodulation of complex passband DSB-SC signal . 255

4.3.1.3 Demodulation of real passband DSB-SC . . . . . . . 256

4.3.2 DSB receiver with frequency and phase offset . . . . . . . . . 258

4.3.2.1 Demodulation of complex passband DSB-SC signal . 258

INSTRUCTOR COPY

4.3.2.2 Demodulation of real passband DSB-SC signal . . . . 258

4.3.2.3 Frequency/phase offset compensation - Costas loop

receiver . . . . . . . . . . . . . . . . . . . . . . . . . 260

4.3.3 Review of I-Q receivers . . . . . . . . . . . . . . . . . . . . . . 260

4.3.3.1 Review of AM and DSB receivers in real notation . . 262

4.4 Chapter review questions . . . . . . . . . . . . . . . . . . . . . . . . . 263

5 Single Side Band modulation (SSB) 265

5.1 Single sideband suppressed carrier overview . . . . . . . . . . . . . . 265

5.1.1 SSB applications . . . . . . . . . . . . . . . . . . . . . . . . . 267

5.1.2 SSB may be generalized to independent sideband (ISB) . . . . 268

5.2 SSB-SC complex baseband signal . . . . . . . . . . . . . . . . . . . . 271

5.2.1 SSB and ISB complex baseband in terms of a general complex

baseband signal . . . . . . . . . . . . . . . . . . . . . . . . . . 271

5.2.2 Analytic messages . . . . . . . . . . . . . . . . . . . . . . . . . 272

5.2.3 Derivation of analytic message from real message . . . . . . . 275

5.2.4 Derivation of SSB-SC complex baseband using Hilbert transforms279

 CONTENTS 11

5.2.4.1 Eliminating negative frequencies in the frequency do-

main . . . . . . . . . . . . . . . . . . . . . . . . . . . 279

5.2.4.2 Eliminating negative frequencies in the time domain 282

5.2.4.3 Definition of Hilbert transform . . . . . . . . . . . . 285

5.2.4.4 Eliminating either negative or positive frequencies . . 286

5.2.5 Hilbert transform definition and properties . . . . . . . . . . . 286

5.2.5.1 Intuitive proof . . . . . . . . . . . . . . . . . . . . . 288

5.2.5.2 Proof in time domain . . . . . . . . . . . . . . . . . 289

INSTRUCTOR COPY

5.2.5.3 Proof in frequency domain . . . . . . . . . . . . . . . 290

5.2.5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . 293

5.2.6 Hilbert transform implementation . . . . . . . . . . . . . . . . 295

5.2.6.1 Hilbert digital filter impulse response . . . . . . . . . 295

5.2.6.2 Hilbert filter implementation in Matlab . . . . . . . . 297

5.2.6.3 Hilbert filter implementation in GNURadio . . . . . 297

5.3 SSB-SC real passband signal . . . . . . . . . . . . . . . . . . . . . . . 297

5.3.1 USB-SC derived from pre-envelope in frequency domain . . . . 297

5.3.1.1 Derivation 1 based on fundamental result . . . . . . 297

5.3.1.2 Derivation 2 . . . . . . . . . . . . . . . . . . . . . . . 299

5.3.1.3 Interpretation . . . . . . . . . . . . . . . . . . . . . . 300

5.3.2 USB-SC derived from pre-envelope in time domain . . . . . . 302

5.3.3 SSB-SC in time and frequency domain . . . . . . . . . . . . . 303

5.3.4 Single tone modulation for SSB . . . . . . . . . . . . . . . . . 305

5.3.4.1 Time domain . . . . . . . . . . . . . . . . . . . . . . 305

5.3.4.2 Frequency domain . . . . . . . . . . . . . . . . . . . 306

 CONTENTS 12

5.4 SSB-SC transmitters . . . . . . . . . . . . . . . . . . . . . . . . . . . 306

5.4.1 SSB-SC transmitter 1 using Hilbert transform (phasing method) 306

5.4.1.1 Software implementation . . . . . . . . . . . . . . . . 306

5.4.1.2 Hardware (analog) implementation . . . . . . . . . . 307

5.4.2 SSB-SC transmitter 2, Weaver modulator, avoids Hilbert trans-

form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307

5.4.3 SSB-SC transmitter 3 using filters . . . . . . . . . . . . . . . . 312

5.5 SSB-SC receivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312

INSTRUCTOR COPY

5.5.1 SSB-SC receiver using message-bandwidth low pass filters . . . 313

5.5.1.1 Method 1 . . . . . . . . . . . . . . . . . . . . . . . . 313

5.5.1.2 Method 2 . . . . . . . . . . . . . . . . . . . . . . . . 313

5.5.1.3 Limitation of Methods 1 and 2 . . . . . . . . . . . . 314

5.5.2 SSB-SC receiver using phasing method with Hilbert transform 315

5.5.2.1 Software implementation with Hilbert transform . . . 315

5.5.2.2 Analog (real) implementation with Hilbert transform 316

5.5.3 SSB-SC receiver using Weaver Demodulator . . . . . . . . . . 317

5.5.3.1 Weaver modulator description and operation . . . . . 317

5.5.3.2 SSB-SC Weaver demodulator operation in complex

notation (software implementation . . . . . . . . . . 319

5.5.3.3 Real input, complex math . . . . . . . . . . . . . . . 320

5.5.3.4 Complex input, complex math . . . . . . . . . . . . . 321

5.5.3.5 SSB-SC Weaver demodulator operation in real nota-

tion (hardware implementation . . . . . . . . . . . . 322

5.6 Chapter review questions . . . . . . . . . . . . . . . . . . . . . . . . . 325

 CONTENTS 13

6 Angle modulation (FM and PM) 331

6.1 Angle modulation Overview . . . . . . . . . . . . . . . . . . . . . . . 331

6.2 Angle modulation complex baseband, analytic signal and real passband
signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333

6.2.1 Phase modulation (PM) with general message . . . . . . . . . 334

6.2.1.1 PM complex baseband s̃(t) . . . . . . . . . . . . . . 334

6.2.1.2 PM analytic signal . . . . . . . . . . . . . . . . . . . 337

6.2.1.3 PM real passband signal . . . . . . . . . . . . . . . . 338

INSTRUCTOR COPY

6.2.1.4 PM complex baseband spectrum . . . . . . . . . . . 339

6.2.2 Frequency modulation (FM) with general message . . . . . . . 339

6.2.2.1 FM complex baseband s̃(t) . . . . . . . . . . . . . . 339

6.2.2.2 FM analytic signal . . . . . . . . . . . . . . . . . . . 343

6.2.2.3 FM real passband signal . . . . . . . . . . . . . . . . 344

6.2.2.4 FM complex baseband spectrum . . . . . . . . . . . 344

6.3 Real passband FM with general message - alternate derivation in time

domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345

6.4 PM and FM with Sinusoidal Message . . . . . . . . . . . . . . . . . . 347

6.4.1 FM with sinusoidal message . . . . . . . . . . . . . . . . . . . 347

6.4.2 Narrowband FM with Sinusoidal Message . . . . . . . . . . . 350

6.4.3 PM with sinusoidal message . . . . . . . . . . . . . . . . . . . 352

6.4.4 Comparison of FM and PM with sinusoidal message . . . . . . 353

6.5 Power Spectrum of an FM Signal – Bessel functions . . . . . . . . . . 353

6.5.1 Bessel function definition . . . . . . . . . . . . . . . . . . . . . 354

6.5.2 PM and FM spectrum with single tone message . . . . . . . . 355

 CONTENTS 14

6.5.2.1 FM complex envelope spectrum . . . . . . . . . . . . 356

6.5.2.2 FM analytic spectrum . . . . . . . . . . . . . . . . . 357

6.5.2.3 FM real passband . . . . . . . . . . . . . . . . . . . 358

6.5.2.4 Observation interval . . . . . . . . . . . . . . . . . . 359

6.5.3 FM spectrum properties . . . . . . . . . . . . . . . . . . . . . 359

6.5.3.1 Wideband FM large modulation index . . . . . . . . 361

6.5.3.2 Narrowband FM, small modulation index . . . . . . 362

6.5.3.3 Effective bandwidth of FM – Carson’s rule . . . . . . 362

INSTRUCTOR COPY

6.5.3.4 FM/PM estimated bandwidth for a general message 363

6.5.4 FM spectrum for dual-tone message . . . . . . . . . . . . . . . 364

6.6 Frequency and phase modulators (transmitters) . . . . . . . . . . . . 365

6.6.1 Complex passband implementation . . . . . . . . . . . . . . . 365

6.6.2 Real passband implementation . . . . . . . . . . . . . . . . . . 365

6.6.2.1 Digital (discrete-time) method . . . . . . . . . . . . 365

6.6.2.2 Analog (continuous-time) method . . . . . . . . . . . 366

6.7 Frequency Demodulation . . . . . . . . . . . . . . . . . . . . . . . . 367

6.7.1 Intuition 1: Differentiation . . . . . . . . . . . . . . . . . . . . 368

6.7.2 Intuition 2: Linear Amplitude vs. Frequency Characteristic . . 369

6.7.3 Intuition 3: Zero Crossing Counter . . . . . . . . . . . . . . . 371

6.7.4 Intuition 4: Phase Locked Loop . . . . . . . . . . . . . . . . . 372

6.8 Digital FM Demodulator . . . . . . . . . . . . . . . . . . . . . . . . . 373

6.8.1 I(t), Q(t) as Real Signals . . . . . . . . . . . . . . . . . . . . 373

6.8.2 I(t) + jQ(t) as a Complex Signal . . . . . . . . . . . . . . . . 374

 CONTENTS 15

6.8.2.1 Method 1 . . . . . . . . . . . . . . . . . . . . . . . . 374

6.8.2.2 Method 2 . . . . . . . . . . . . . . . . . . . . . . . . 375

6.8.3 FM demodulator with frequency offset . . . . . . . . . . . . . 377

6.9 Pre-emphasis and de-emphasis in FM . . . . . . . . . . . . . . . . . . 377

6.9.1 History and motivation for FM pre-emphasis and de-emphasis 377

6.9.2 Pre- and de-emphasis analog circuits . . . . . . . . . . . . . . 379

6.9.3 De-emphasis with digital filter . . . . . . . . . . . . . . . . . . 382

6.9.3.1 Implementation . . . . . . . . . . . . . . . . . . . . . 382

INSTRUCTOR COPY

6.9.3.2 Impulse response . . . . . . . . . . . . . . . . . . . . 383

6.9.3.3 Frequency response . . . . . . . . . . . . . . . . . . . 383

6.9.3.4 Summary of digital de-emphasis filter parameters . . 386

6.10 FM stereo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386

6.11 Chapter review questions . . . . . . . . . . . . . . . . . . . . . . . . . 388

7 Baseband data transmission 396

7.1 Digital messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396

7.2 Data source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397

7.3 Differential encoding and decoding . . . . . . . . . . . . . . . . . . . 400

7.3.1 Avoiding polarity inversion . . . . . . . . . . . . . . . . . . . . 400

7.3.2 Differential encoding . . . . . . . . . . . . . . . . . . . . . . . 400

7.3.3 Differential decoding . . . . . . . . . . . . . . . . . . . . . . . 402

7.4 Data transmission system with pulse shaping . . . . . . . . . . . . . . 403

7.4.1 Representing digital data as square pulses . . . . . . . . . . . 404

7.4.2 Pulse shaping to eliminate the sharp edges of a square pulse . 405
 CONTENTS 16

7.4.3 Representing digital data with impulses . . . . . . . . . . . . . 411

7.4.4 Pulse shaping a sequence of impulses . . . . . . . . . . . . . . 412

7.4.4.1 Square pulse filter . . . . . . . . . . . . . . . . . . . 412

7.4.4.2 Time-domain raised-cosine filter . . . . . . . . . . . . 412

7.4.4.3 Frequency domain raised-cosine filter . . . . . . . . . 416

7.4.4.4 Frequency domain square root raised cosine filter . . 419

7.4.5 Summary of pulse shaping methods . . . . . . . . . . . . . . . 421

7.5 Eye diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422

INSTRUCTOR COPY

7.6 Chapter review questions . . . . . . . . . . . . . . . . . . . . . . . . . 425

8 Phase shift keying (PSK) 428

8.1 PSK introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 428

8.2 BPSK Transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432

8.2.1 Real BPSK signal . . . . . . . . . . . . . . . . . . . . . . . . . 437

8.2.2 Complex BPSK signal . . . . . . . . . . . . . . . . . . . . . . 439

8.2.3 BPSK at radio frequency (RF) . . . . . . . . . . . . . . . . . 441

8.3 BPSK Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443

8.3.1 Real BPSK signal receiver . . . . . . . . . . . . . . . . . . . . 443

8.3.2 Complex BPSK signal receiver . . . . . . . . . . . . . . . . . . 445

8.3.3 BPSK receiver at RF . . . . . . . . . . . . . . . . . . . . . . . 446

8.4 QPSK Transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447

8.4.1 Real QPSK signal . . . . . . . . . . . . . . . . . . . . . . . . . 451

8.4.2 Complex (analytic) QPSK signal . . . . . . . . . . . . . . . . 455

8.4.3 QPSK at radio frequency (RF) . . . . . . . . . . . . . . . . . 457

 CONTENTS 17

8.5 QPSK Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 458

8.5.1 Real QPSK signal . . . . . . . . . . . . . . . . . . . . . . . . . 458

8.5.2 Complex QPSK signal . . . . . . . . . . . . . . . . . . . . . . 462

8.5.3 QPSK signal at radio frequency (RF) . . . . . . . . . . . . . . 463

8.6 QPSK with amplitude, frequency and phase errors in the receiver . . 464

8.6.1 QPSK signal at particular time instant . . . . . . . . . . . . . 464

8.6.2 QPSK receiver local oscillator with frequency error . . . . . . 466

8.6.3 QPSK receiver local oscillator with amplitude and phase errors 470
INSTRUCTOR COPY

8.7 Differential BPSK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473

8.7.1 DBPSK with square pulse shape . . . . . . . . . . . . . . . . . 473

8.7.2 D-BPSK with RC2 time domain pulse shape . . . . . . . . . . 476

8.7.3 DPSK with RRC frequency domain pulse . . . . . . . . . . . . 479

8.8 QPSK variants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479

8.8.1 Offset QPSK . . . . . . . . . . . . . . . . . . . . . . . . . . . 480

8.8.2 π/4 QPSK . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480

8.9 Chapter review questions . . . . . . . . . . . . . . . . . . . . . . . . . 482

9 Frequency shift keying (FSK) 484

9.1 FSK intuition in real passband . . . . . . . . . . . . . . . . . . . . . . 484

9.2 FSK equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486

9.2.1 FSK real passband . . . . . . . . . . . . . . . . . . . . . . . . 486

9.2.2 FSK phase ramps . . . . . . . . . . . . . . . . . . . . . . . . . 487

9.2.3 FSK complex envelope . . . . . . . . . . . . . . . . . . . . . . 490

9.2.4 FSK analytic signal . . . . . . . . . . . . . . . . . . . . . . . . 495

 CONTENTS 18

9.2.5 FSK spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . 495

9.3 FSK variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497

9.3.1 Minimum-shift keying and Offset-QPSK . . . . . . . . . . . . 497

9.3.2 Multitone FSK . . . . . . . . . . . . . . . . . . . . . . . . . . 498

9.3.3 Multi-carrier modulation . . . . . . . . . . . . . . . . . . . . . 499

9.3.4 Unified multicarrier ASK-FSK-PSK-QAM-OFDM modulator . 500

9.3.5 FSK pulse shaping . . . . . . . . . . . . . . . . . . . . . . . . 501

9.4 FSK receivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503

INSTRUCTOR COPY

9.4.1 Non-coherent FSK receiver 1 . . . . . . . . . . . . . . . . . . . 503

9.4.2 Non-coherent FSK receiver 2 . . . . . . . . . . . . . . . . . . . 504

9.4.3 Modification of non-coherent FSK receivers 1 and 2 . . . . . . 505

9.4.4 Coherent FSK receiver . . . . . . . . . . . . . . . . . . . . . . 505

9.5 Chapter review questions . . . . . . . . . . . . . . . . . . . . . . . . . 506

10 Synchronization 509

10.1 Carrier Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . 509

10.2 Symbol synchronization . . . . . . . . . . . . . . . . . . . . . . . . . 509

10.3 Frame Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . 510

11 Link Budget 513

11.1 Link Budget Overview . . . . . . . . . . . . . . . . . . . . . . . . . . 513

11.2 Power obtained over a link . . . . . . . . . . . . . . . . . . . . . . . . 513

11.3 Path loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514

11.3.1 Cable or Fibre . . . . . . . . . . . . . . . . . . . . . . . . . . . 514

11.3.2 Free space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514

 CONTENTS 19

11.3.3 Flat earth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515

11.3.4 Practical considerations . . . . . . . . . . . . . . . . . . . . . 517

11.4 Power required at receiver . . . . . . . . . . . . . . . . . . . . . . . . 518

11.5 Link budget calculation . . . . . . . . . . . . . . . . . . . . . . . . . . 518

11.6 Digital system performance . . . . . . . . . . . . . . . . . . . . . . . 519

11.7 Chapter review questions . . . . . . . . . . . . . . . . . . . . . . . . . 520

A List of symbols 524

INSTRUCTOR COPY

A.1 Signals, messages and filters . . . . . . . . . . . . . . . . . . . . . . . 524

A.2 Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525

A.2.1 Complex baseband, analytic and real passband signals . . . . 525

A.2.2 Signals in terms of amplitude, phase and carrier frequency . . 525

A.2.3 Sampled signals and Discrete Fourier transform (DFT) . . . . 526

A.3 Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 526

A.4 Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 527

A.5 Link budget . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 527

B Useful formulas 529

B.1 Mathematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529

B.2 Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 530

B.2.1 complex baseband . . . . . . . . . . . . . . . . . . . . . . . . . 530

B.2.2 analytic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531

B.2.3 real passband . . . . . . . . . . . . . . . . . . . . . . . . . . . 531

B.2.4 sampled signals . . . . . . . . . . . . . . . . . . . . . . . . . . 531

B.3 Messages: real baseband . . . . . . . . . . . . . . . . . . . . . . . . . 532

 CONTENTS 20

B.3.1 single tone message . . . . . . . . . . . . . . . . . . . . . . . . 532

B.3.2 multitone messages . . . . . . . . . . . . . . . . . . . . . . . . 532

B.3.3 sampled discrete-time messages . . . . . . . . . . . . . . . . . 532

B.3.4 Digital messages . . . . . . . . . . . . . . . . . . . . . . . . . 533

B.4 Modulation schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . 533

B.4.1 AM complex baseband . . . . . . . . . . . . . . . . . . . . . . 533

B.4.2 DSB complex baseband . . . . . . . . . . . . . . . . . . . . . . 533

B.4.3 SSB complex baseband . . . . . . . . . . . . . . . . . . . . . . 534

INSTRUCTOR COPY

B.4.3.1 SSB messages . . . . . . . . . . . . . . . . . . . . . . 534

B.4.3.2 USB complex baseband . . . . . . . . . . . . . . . . 534

B.4.3.3 LSB comlex baseband . . . . . . . . . . . . . . . . . 535

B.4.3.4 ISB complex baseband . . . . . . . . . . . . . . . . . 535

B.4.4 PM complex baseband . . . . . . . . . . . . . . . . . . . . . . 535

B.4.5 FM complex baseband . . . . . . . . . . . . . . . . . . . . . . 535

B.4.6 PSK complex baseband . . . . . . . . . . . . . . . . . . . . . . 536

B.4.7 FSK complex baseband . . . . . . . . . . . . . . . . . . . . . . 536

B.5 Link Budget . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536

C Derivations 538
dφ
C.1 φ(t) − φ(t − 1) ≈ dt
. . . . . . . . . . . . . . . . . . . . . . . . . . . 538

D Applications 539

D.1 Spectrum Analyzers, analog and digital . . . . . . . . . . . . . . . . 539

D.2 Digital filters as applied to pulse shaping . . . . . . . . . . . . . . . . 542

D.3 dB and ratios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 542

 Chapter 1

Introduction
INSTRUCTOR COPY

1.1 Telecommunications systems overview

Learning objectives

In this section we learn

• fundamental ideas about what a telecommunications system is,

• the component parts of a telecommunication system and how they relate to each
other.

1.1.1 Telecommunications technology

The purpose of this chapter is to explain the basic principles, technology and systems
of telecommunications. Telecommunications systems and technology are changing
rapidly in this Internet era, but the fundamental principles remain the same.

Webster’s defines “communications” as “the exchange of thoughts, messages or

information, as by speech, signals, writing or behaviour” as well as ”the technology
employed in transmitting messages”. Since “tele” means distance, telecommunica-
tions, according to Webster’s, is “the science and technology of communications at
a distance by electronic transmission of impulses, as by telegraph, cable, telephone,
radio or television” (or cellphones or computer networks).

21
 CHAPTER 1. INTRODUCTION 22

Thus telecommunications is about how to send a message between locations sep-

arated by some distance. The message (thoughts or information) may be in the form
of speech, music, image, video, text or data. At each location, there will be people
who will see or hear the message, or devices which will read it electronically.

In human communications (without the ‘tele’), messages are expressed (transmit-

ted) in a form which can be detected (received) by human senses (e.g. sight and
hearing). Telecommunications provides a means for these human senses to work at a
distance.

1.1.1.1 Terminology
INSTRUCTOR COPY

There are various types of telecommunications. Telecommunications can be one-

way from sender (transmitter) to receiver (e.g. a radio broadcast), and two-way,
where both parties can send and receive. Telecommunications can be one-to-one
from point A to point B (point-to-point/uni-cast, e.g. telephone call), one-to-a-
few (point-to-multi-point/multi-cast, e.g. a lecture), one-to-many (broadcast, e.g.
radio or television), many-to-one (e.g. electronic voting) and many-to-many (e.g.
teleconference).

Telecommunications can be real-time in which the message is sent immediately at

near the speed of light (e.g. telephone call), near-real-time where there may be some
delay (e.g. text messaging, web browsing), or non-real-time where the delay may be
any amount (e.g. physical hard-copy mail delivered by the post office or a courier).

Smoke signals, shouting and hand-waving are ways of communications at a dis-

tance, but telecommunications, as per Webster’s, is exclusively electronic. Thus a
telecommunications system for people must first convert the message from human-
readable form (speech, music, image, video, text, data) into electronic form then
transmit this electronic message over the distance via some channel (electronic path-
way), and convert it back to human-readable form.

Electronic signals are currents of charged particles (electrons) which are made to
move through a wire by a voltage. By analogy with water in a pipe, the voltage is a
kind of pressure that pushes the electrons like water through a pipe, and the current
is the flow of water. The precise details of how a voltage creates a current is explained
by the principles of physics (theory of electromagnetic fields). The voltage is related
to the current via Ohm’s law. A graph of voltage versus time will look the same as
a graph of current versus time. Both graphs represent an electrical waveform which
is the same as the electronic signal. Thus, in what follows, we use the words voltage
and current interchangeably.
 CHAPTER 1. INTRODUCTION 23

1.1.1.2 Examples

To illustrate these ideas, we first give some examples of telecommunications systems

of different types. Then we will define the concepts and explain them in more detail.

The telegraph was the first electronic communication system. It was two-way and
realtime. The message in text form was converted letter-by-letter into the dots and
dashes of Morse Code. The transmitter was a battery and a switch (morse code key)
which was closed and opened in accordance with the morse code). The channel was
a long piece of copper wire which carried the electronic current from A to B. The
receiver was a speaker which emitted clicks when the direct current (DC) started or
stopped. Later forms of the telegraph transmitted alternating current (AC) which
the speaker converted to beep-beep electronic tones.
INSTRUCTOR COPY

The telephone used a microphone to convert the acoustic waves of speech into
corresponding electronic waves, which were then transmitted over copper wires to the
receiver speaker.

The wireless telegraph was invented when it was discovered that with a sufficiently
strong battery, the clicks or sparks which happen at the instants when the morse key
was closed and opened could travel through free space as Hertzian waves without the
need for wires. In this case, the channel is free space. Later versions transmitted
pure tones (high frequency AC, so called carrier waves) instead of sparks. The carrier
waves were turned on and off in accordance with the Morse Code.

Radio broadcasting started when it was learned how the high frequency AC carrier
waves could be modulated gradually in step with the waves of speech, instead of just
turned on and off. Television broadcasting started when it was learned how to scan
an image to make an electronic signal which could be modulated onto the carrier
wave in a similar way as speech.

Data communication using teletype was an automated form of morse code to send
text and other data at higher speeds than the 30 or so words per minute that could
be reliably received by a human operator. The new automated codes were patterns
of ones and zeros (binary digits or bits) used to represent letters and numbers, and
operated at speeds of 60 words per minute and above. Various codes were used,
nowadays we use mainly the 7-bit ASCII code for plain text. The ones and zeros were
transmitted using modems using two different tones (carrier waves). The teletype
machines were later replaced by computers, and used telephone lines or radio (free
space) as the channel. Early telephone modems used by computers operated at speeds
of 110 baud (bits per second) or about 120 words per minute. Telephone modem
speeds gradually increased to the 56,000 bits per second of today.
 CHAPTER 1. INTRODUCTION 24

The telephone system creates a channel (electronic pathway) between two hand-
sets (for a voice call), or between two telephone modems (for a dialup computer
connection). The channel carries the electronic waveform (speech or modem tones)
from sender to receiver. A long distance telephone call may use many different kinds
of channels in combination, such as copper wires, coaxial cable, fiber optic cable,
microwave radio links, satellite links, and a wireless cellular telephone connection.

The telephone system is a network of channels and nodes. The nodes contain
switches which connect the channels together and thus route a call through the net-
work. The Internet is a different kind of network of nodes and channels, used to
create a connection between two computers or other internet appliances.

These examples were presented as a background to set the stage for the more
detailed descriptions in the next section.
INSTRUCTOR COPY

1.1.1.3 Standards organizations

Telecommunication systems can only work if the network nodes and channels oper-
ate and interwork according to some agreed-upon standard methods and protocols.
Thus a number of national and international standards organizations have evolved,
which are referenced in the sequel. A partial list includes ITU-T, ITU-R (Inter-
national Telecommunications Union), (formerly CCITT, CCIR) for wired and wire-
less telecommunications, respectively, ANSI (American National Standards Institute),
ETSI (European Telecommunications Standard Institute), MPEG (Motion Pictures
Experts Group), IETF (Internet Engineering Task Force).

1.1.2 Telecommunication systems

In what follows, we define different telecommunications systems via a block diagram

showing the components (nodes and channels), and how they are connected together
to make a system.

1.1.2.1 Point-to-point system

Table 1.1 shows a one-way point-to-point telecommunications system consisting of

two nodes and one channel. The system includes a human sender or message source,
a transducer (mic, camera) to convert the message to electronic form, a channel
to transmit the message across a distance, another transducer (speaker, display) to
 CHAPTER 1. INTRODUCTION 25

Communication system component Information format

Message source Sound, light
Transducer (microphone, camera) Electrical waveform
Channel Electrical waveform
Transducer (loudspeaker, display screen) Sound, light
Message sink

Table 1.1: Telecommunications system

convert the message back to a form that can be sensed by humans (sound, light), and
a human receiver or message sink.
INSTRUCTOR COPY

1.1.2.2 Message source

The message source may be in the form of sound (speech, music) or light (still or
moving image). Messages which are data or text will be considered later.

1.1.2.3 Transducer at sender

The transducer at the sender converts the message to an electronic signal.

If the message is expressed as sound (speech,music) which can be sensed by the

ears, the transducer is a microphone. The microphone contains a diaphragm or
membrane which vibrates in step with the sound waves. The vibrations are used to
power a kind of tiny electric generator in the form of a magnet and a moving coil
of wire, which creates an electrical current in step with the vibrations. The precise
details of how vibrations create electrical current are explained by basic principles of
physics (Maxwells theory of electromagnetic fields). The output of the microphone
transducer is a time-varying voltage (AC or alternating current) .

If the message is light (still or moving image) which can be sensed by the eyes,
the transducer is a camera. The camera contains a lens to focus and project the
image onto a light-sensitive surface made of electronic semiconductor material (e.g. a
charged coupled device or CCD) which converts light to electrical signals. Bright spots
in the image will generate a stronger electrical signal at that point on the surface.
The precise details of how light creates electrical current are explained by advanced
principles of physics (Quantum Theory). The CCD may have several million light-
sensitive points, arranged in a grid of rows and columns. To create a time-varying
voltage, the projected image is scanned by reading the current at each point on the
surface in a predefined order, e.g. by reading each row in turn from top to bottom.
 CHAPTER 1. INTRODUCTION 26

Each row may be called a scanning line. A video camera scans about 30 images
(frames) per second.

1.1.2.4 Electronic signals

The transducer (microphone or camera) output is a time-varying voltage which we

call an electronic signal or just signal for short. This signal is a so-called analog signal,
in that the voltage is analogous to the message, i.e. the voltage goes up and down in
step with the message waves.

The signal can be viewed on a piece of electronic test equipment called an oscillo-
scope, which will create the graph of the signal voltage (amplitude) versus time, just
INSTRUCTOR COPY

as shown in.

The number of times the signal goes up and down per second is the frequency in
cycles per second (Hz) of the signal at that point. In some parts of the signal, the
voltage will go up and down slowly, whereas in other parts, it will go up and down
quickly. For most signals, there will be many frequencies at the same time, so the up
and down variations will be a mixture of all these frequencies. In musical terms, each
frequency corresponds to a note or pitch, and most music signals are a mixture of
pitches (chords), along with noise and transients (attacks, sudden changes in level).

The signal can also be viewed on a piece of electronic test equipment called a
spectrum analyzer, which will create a graph of the signal voltage (amplitude) versus
frequency. For audio signals, the frequencies will be the range from 20–20 000 Hz. The
bandwidth of the signal is the difference between the highest and lowest frequency,
and is about 20 000 Hz for audio signals.

The two graphs can be combined in a so-called time-frequency diagram showing

the signal amplitude versus both time and frequency.

1.1.2.5 Channel

The channel is an electrical pathway between the message source and sink. which is
used to carry a voltage from the sender over a distance to the receiver.

The basic types of channels are:

• copper wire (like telephone cord or speaker wire, often called twisted pair)
• coaxial cable (used for cable TV)
 CHAPTER 1. INTRODUCTION 27

• optical fiber

• free space (wireless, radio)

The channel can also be a so-called storage channel, where the message is stored
on some media like tape or disk and then played back later.

For the twisted pair wire channel, the signal is simply carried over a distance via
the wire, in the same way electrical power (110 V AC) is carried over wires.

For the free space radio channel a carrier wave (radio wave) is needed to carry the
signal across free space using electromagnetic (radio) waves. Similarly, for the optical
fiber channel, a carrier wave (light wave) is needed to pass the signal through (inside)
the optical fiber. A carrier wave can also be used on the copper wire channel. Carrier
INSTRUCTOR COPY

waves are explained in more detail in the following section.

1.1.2.6 Transducer at receiver

The transducer at the receiver converts the electronic signal back to the message form
which can be sensed by a human.

If the original message is sound (voice, music), then the transducer is a speaker
(or headphones). A speaker is a microphone in reverse, with a magnet, moving coil
and diaphragm. The electrical signal powers a kind of electric motor which moves
the diaphragm and thus creates sound waves.

If the original message is light (still or moving image), then the transducer is an
electronic display device such as a LCD (liquid crystal display) or LED (light emitting
diode display. The signal is scanned across the display, and causes light to emit from
the display, with stronger signals generating brighter light, and thus recreating the
image. The details of how the displays create light from electronic signals depends
on the type of display. The display can also be a printer, where stronger signals will
deposit more ink.

1.1.3 Analog communication systems

In this section, we describe an analog electronic telecommunications system in more

detail. Consider the one-way point-to-point system of table 1.1 again, but with one
additional block at both the sender and receiver end. The modulator is used to
combine the analog signal with the carrier wave (modulate the carrier) ready for
 CHAPTER 1. INTRODUCTION 28

Application Frequency Wavelength

Subsonic vibration 1 Hz 300 000 000 m
Audio 1 kHz 300 000 m
AM broadcast 1 MHz 300 m
Cellphone, GPS, WiFi 1 GHz 0.3 m
Future 6G wireless 1 THz 0.3 mm
Ultraviolet light 1 PHz 300 nm

Table 1.2: Frequencies and wavelengths for some common applications

transmission over the channel, and the demodulator is used to recover the original
analog signal from the modulated carrier wave We first discuss different types of
carrier waves, and then different types of modulation.
INSTRUCTOR COPY

1.1.3.1 Carrier waves

Carrier waves can be used with wire and cable channels, and must be used with radio
and fiber optic channels. Carrier waves are electromagnetic waves used to carry a
message over a distance. Electromagnetic waves were predicted to exist by Maxwell,
based on the laws of physics (electricity and magnetism), and were experimentally
demonstrated by Hertz. Both radio and light waves are electromagnetic waves, and
can travel through free space.

A typical electromagnetic wave contains both electric and magnetic fields. An

electromagnetic wave with electric and magnetic fields looks like a sine wave, and
moves at the speed of light. If you stand in one spot, you will observe that the
strength of the field goes up and down as the wave train moves past you. The
frequency of the wave is the number of times per second that the field goes up and
down and back up again.

Frequency is measured in units of cycles per second, or Hertz, abbreviated Hz. If

the frequency of the wave is chosen to be a particular value near 1000 billion Hz, then
the human eye will recognize it as light of a particular color. Lower frequency waves
are invisible to the human eye, and are used as radio waves. The wavelength of the
wave is the distance between peaks. The wavelength is determined by the speed of
the train (speed of light, 300,000,000 meters per second) divided by the frequency.
Table 1.2 shows frequencies and wavelengths along with some of their applications.

Carrier waves are produced in free space by first generating an electronic signal
at the desired frequency of the carrier wave (called the carrier frequency) using an
electronic oscillator circuit (a kind of fast electronic pendulum), and then connect-
 CHAPTER 1. INTRODUCTION 29

ing this signal to an antenna which launches the carrier wave into space. The size
of the antenna depends on the wavelength (or frequency) of the carrier wave, and
the whether the antenna is directional (like a flashlight) or omnidirectional (like a
lightbulb). A general rule is that omnidirectional antennas must be at least one quar-
ter wavelength in size to work well. Examples of such antennas are TV rabbit ears,
the stubby cellphone antenna, and the whip on your car or the patch antenna on
your GPS receiver. Directional antennas are typically a wavelength or more in size,
and bigger antennas are more directional. Examples are yagis (TV antennas), dishes
(parabolic reflectors) for satellite, and cellular tower base antennas.

Carrier (electromagetic) waves of different frequencies and wavelengths have dif-

ferent properties. For example, radio waves can travel through walls, but light waves
cannot. Lower frequency waves tend to travel further, and can bend around corners.
Higher frequency waves travel more or less only via line of sight. Thus certain parts
INSTRUCTOR COPY

of the radio spectrum are better suited for certain types of telecommunications.

The radio spectrum is a resource that provides the means to send messages over
a distance. Thus governments have made laws and regulations dividing up this spec-
trum into segments for different kinds of telecommunications, such as broadcast radio,
cellular telephones, TV, and taxi radios. In some countries, parts of the spectrum
have been sold at auction. The radio part of this spectrum is sometimes called the
airwaves. The chart shows the kinds of telecommunications using each part of the
spectrum.

A given slice of the radio spectrum has a bandwidth which is the difference between
the highest and lowest frequencies at the boundaries of the slice. For example, the
total bandwidth for FM broadcast radio is 20 MHz, from 88–108 MHz. This total
bandwidth is divided into 100 channels each 200 kHz wide. The bandwidth for one
FM broadcast radio channel is 200 kHz. The wider the total bandwidth of a slice of
spectrum, the more channels can be used, and thus the higher its value.

The frequencies chosen for a particular type of telecommunications service de-

pends on the desired distance to be covered, the bandwidth needed, and the practical
antenna size.

Carrier waves are used not only on radio (free space) channels, but also on cable
and fiber optic channels. In the case of fiber optics, the carrier frequency is that
of visible light, and the carrier wave is generated by a laser instead of an electronic
oscillator circuit.
 CHAPTER 1. INTRODUCTION 30

1.1.3.2 Modulation

Modulation is the process whereby a carrier wave of a particular frequency is modified

or modulated by the message signal, so that the modulated carrier wave can be used
to carry the message signal over a distance. There are two basic kinds of modulation:
Amplitude Modulation (AM) in which the amplitude of the carrier wave is modulated
(increased and decreased) in step with the message signal and Frequency Modulation
(FM) in which the frequency of the carrier wave is modulated in step with the message
signal. Phase modulation (PM) is similar and related to FM. AM and FM may
also be used at the same time on one carrier. For FM, the amount by which the
carrier frequency is increased and decreased in step with the modulation is called the
frequency deviation. Demodulation is the reverse process in which the modulated
carrier wave is received and processed to recover the original signal. Modulation
INSTRUCTOR COPY

and demodulation are achieved by special electronic circuits inputs: carrier, message,
output: modulated carrier.

For AM, the bandwidth of the modulated carrier is two times the bandwidth of
the message signal. For FM and PM, the bandwidth of the modulated carrier can
be from 2 times the bandwidth of the message signal plus 2 times the frequency
deviation.

1.1.3.3 Upconversion and amplification

For radio systems, it is often preferred for cost and other practical reasons to do the
modulation using a carrier at a low frequency, and then shift or upconvert the carrier
to the desired operating frequency.

1.1.3.4 System examples

Examples of analog communication systems include:

• AM and FM radio broadcast

• Citizen’s band two way radio (uses AM)

• FM two way dispatch radio (taxis, couriers)

• TV broadcast (AM for video, FM for sound)

• older satellite TV systems (4 GHz, big dish)(uses FM)

 CHAPTER 1. INTRODUCTION 31

• analog cellular telephone (uses FM)

• cordless telephone (uses FM)

• vinyl LP records

• analog tape

• landline telephone from handset to central switching office

All of the above examples except the last three use a carrier wave.

1.1.3.5 Noise and interference

INSTRUCTOR COPY

Analog communications systems are still in widespread use. However, they suffer
from sensitivity to noise and interference.

For good entertainment quality audio, the noise power must be at least one million
times weaker than the signal power. The hiss and rumble from a good vinyl LP or
analog tape barely meets this standard. The hiss and noise from FM radio often does
not meet this standard. For acceptable telephone quality audio, the noise power must
be at least one thousand times weaker than the signal power. This is generally true
for wired telephone lines, but often not true for cellphone connections.

For a good snow-free television picture, the noise must be about 100,000 times
less than the signal power.

In the case of analog tape, each copy of a tape will have more noise than the
original, so the quality is degraded with each subsequent copy. The same problem
occurs with some old telephone systems, where analog repeaters were used to send long
distance telephone calls across the continent. Thus virtually all new communication
system designs are digital.

1.1.4 Digital communications system

Digital communications systems convert the message into bits (ones and zeros) before
transmitting it (or storing it). One major advantage of digital is higher quality, which
is preserved when the message is repeated or copied. This is why virtually all music is
now stored on CD instead of vinyl records. Another advantage is that digital messages
can be easily stored on a computer and transmitted over computer networks such as
the Internet. Computer memory and disks store the bits in groups of 8 bits called
 CHAPTER 1. INTRODUCTION 32

bytes. A one megabyte (MB) memory chip stores 8 million bits. An audio CD can
store 650 MB or 5.2 billion bits. A single-sided DVD stores about 4.7 4.7 GB. A third
advantage is that different kinds of messages (e.g. audio, video, text, data) can be
combined and transmitted together, since they are all converted to bits.

table 1.3 shows the one-way point-to-point system of table 1.1 again, but with
two additional blocks at both the sender and receiver end. The analog-to-digital
converter (A/D) block converts the analog electronic signal (time-varying voltage) to
digital form (a sequence of binary numbers). The formatter block is used to modify
and rearrange the sequence of numbers as described later.

analog sampler quantizer formatter modulator channel

signal numbers bits bits waveform
INSTRUCTOR COPY

Table 1.3: Digital transmitter

1.1.4.1 A/D, sampling, quantization

The A/D includes two parts: a sampler which takes samples (measurements) of the
analog signal voltage at regular intervals, and a quantizer which assigns a numerical
value in binary form to each sample measurement. Thus the output of the A/D is a
sequence of binary numbers. The process of converting the analog signal to numbers
is called digitizing the signal or converting the signal to digital form.

1.1.4.2 Types of message signals

The message may be analog (audio, video) that is sampled and quantized to yield a
sequence of binary numbers. The message may also be inherently digital (text, data)
that does not require sampling and quantization.

Speech and audio The number of samples taken per second determines the sam-
pling frequency. For an audio CD, this sampling frequency is 44,100 samples per
second or 44 100 Hz. The numerical values of the samples are expressed in binary
notation using only ones and zeros. The maximum number of bits used to represent
these values determines the precision, or the number of possible values for each sam-
ple. For an audio CD, this is 16 bits, so there are 216 or 65,536 possible values for
each sample. If the signal voltage has a maximum of 1 V and a minimum of −1 V,
then we have 65,536 steps between -1 and +1. The size of one of these steps in volts
 CHAPTER 1. INTRODUCTION 33

is called the resolution or quantization step size. The maximum quantization error is
one-half the quantization step size or 1/65 536 V.

The number of samples per second multiplied by the number of bits per samples
determine the total number of bits per second needed to represent the signal in digital
form. For an audio CD, we have 44,100 samples/sec times 16 bits/sample = 708,000
bits-per-second. For two-channel stereo, we need 1.4 million bits/second (1.4 MB/s).

The sampling frequency must be at least twice the highest frequency in the mes-
sage signal. This is because we need at least two samples to represent one cycle of a
sine wave. For music, the highest frequency of human hearing is around 20 000 Hz, so
the CD sampling rate of 44 100 Hz is considered sufficient. The quantization error re-
sults in so-called quantization noise. With 16 bits, this quantization noise is virtually
inaudible, and thus 16 bits is usually considered sufficient. Audiophiles would like
INSTRUCTOR COPY

more safety margin, so the new DVD audio standard is 96 kHz sampling and 24-bit
resolution.
Exercise 1
How many bits/sec are needed for 2 channels of DVD audio?

Solution
4.608 Mbit/s. For digital telephony and digital cellphones, the sampling fre-
quency is 8 kHz with 8 bit resolution, which is sufficient for acceptable voice
quality (but the music-on-hold sounds bad).

Exercise 2
How many bits/sec are needed for telephony?

Solution
64 kbit/s.

Digitized audio signals can also be synthesized by a computer program which

generates the sequence of numbers directly. Most of the sounds made by a computer
(e.g. beeps, clicks, bongs etc) are synthesized, as are the ringing sounds of cellphones,
and the sounds of electronic keyboard instruments. Speech can be synthesized from
text using text-to-speech software and is frequently used in telephone voice response
systems, and for reading email by telephone.
 CHAPTER 1. INTRODUCTION 34

Image and video Image and video signals can also be digitized (sampled and
quantized, and thus converted to numbers). A printed image is digitized using a
scanner which divides the image into a grid (matrix) of pixels (picture elements), and
scans this matrix row by row (line by line). Each pixel is a sample of the image.
The brightness of each pixel determines the binary numerical value for that pixel. A
digital camera includes both the transducer (which does the sampling and produces
a voltage for each pixel) and the A/D (which assigns a binary number to each pixel).
For color images, there are 3 binary numbers for each pixel, one each for red, green
and blue. The number of bits per pixel depends on the desired precision of colors. For
8 bits/pixel, there can be 256 colors, which is rather limited, but usually considered
sufficient for images on web pages. High quality images use 24 bits/pixel. For still
images, the number of samples depends on the resolution required. The higher the
resolution, the less ’grainy’ is the image. A matrix of 3,000 by 2,000 pixels (6 million
INSTRUCTOR COPY

pixels total) with 24 bits/pixel (144 million bits total or 18 MB) is generally considered
to yield the same resolution or graininess as good 35 mm film.

Video is simply a sequence of images, usually about 30 images (frames) per second
to avoid jerkiness. Standard television uses 525 scanning lines (matrix rows) with
about 700 samples per row.

Exercise 3
Assuming bits/pixel, how many bits per second are needed to represent digital
television?

Solution
176.4 Mbit/s.

Digitized images and video signals can be synthesized directly by a computer pro-
gram which generates the numbers representing the pixels directly, without the need
for a transducer or A/D. All the video (and audio) of computer games is synthesized.

Text and data Text messages are naturally in digital form, since each character is
represented by a byte or 8 bits. The printable characters (letters, numbers, punctu-
ation, symbols) are represented using the so-called ASCII code (American Standard
Code for Information Interchange), so there are 27 or 128 possible characters. There
are also 128 non-printable characters used for tabbing and other control functions.
 CHAPTER 1. INTRODUCTION 35

Exercise 4
How many bits per second are needed to transmit text typed at 120 words per
minute?

Solution
2 words/sec times 5 characters/word is 10 characters per second or 80 bits/sec.

Printed text can be converted to digital form using a scanner or fax machine, just
like an image. For black-and-white fax machines, each pixel is represented by just
one bit (1 or 0, black or white). A fax machine with standard resolution uses 100
scanning lines per inch.
INSTRUCTOR COPY

Exercise 5
How many bits per second are needed to transmit one page of text in one
minute?

Solution
For a standard 8.5 by 11 inch sheet, there will be 850 times 1100 pixels equals
935,000 bits transmitted in 60 seconds, which is 15,583 bits/sec.

Data stored in a computer memory is also naturally in digital form. Data may be
arranged in many different formats depending on the application.

1.1.4.3 Bitstream

At this point, we have shown that an analog message (speech, music, image, video)
is converted to digital form, i.e. a sequence of numbers. For audio, there is one
number per sample, and for video, there is one number per pixel. These numbers
are represented as a stream of bits. Text messages are naturally in digital form. The
digital communication system will transmit these bits over a distance.

table 1.4 shows the digital communications system again for convenience and to
show the location of the formatter in the chain. table 1.5 shows how the formatter
modifies the bit stream in 4 stages before it is transmitted. Each of these 4 stages is
explained below.
 CHAPTER 1. INTRODUCTION 36

analog sampler quantizer formatter modulator channel

signal numbers bits bits waveform

Table 1.4: digital transmitter

A/D segmentation source coding encryption channel coding

(compression) (security) (error protection)

Table 1.5: Formatter

Segmentation Segmentation or blocking divides the bit stream into segments or

blocks (groups) of bits. Audio (music and speech) change about 50 times per second,
INSTRUCTOR COPY

so audio segments are typically 20 ms long (about 1,000 samples for CD audio at
a 48 kHz sampling rate, or 160 samples for speech at an 8 kHz sampling rate). For
image or video, a segment is typicaly an 8x8 block (64 pixels in a square), or a single
scanning line.

Source coding Source coding compresses the segment, i.e. reduces the number of
bits in the segment by eliminating redundancy and irrelevant bits. Compression is
very important, since fewer bits means less time or lower cost to transmit the message,
or less memory or disk space to store the message. There are many different kinds of
source coding specialized for speech, music, image and video. They are called codecs
(coder-decoders). Only some examples will be given.

For speech coding such as used on cellphones, the speech is modelled as either
voiced (vowel) or unvoiced (consonants). So-called linear predictive coding (LPC)
algorithms are used to reduce the bit rate from 64 Kbits/sec (8,000 samples/sec
times 8 bits per sample) down to 4-13 Kbits/sec depending on the particular system.
Speech codec acronyms like CSELP have been carefully tuned to give the best possible
perceived quality at a given bit rate.

Audio coding is used on digital audio broadcast and for storage and exchange of
music. Audio coding takes advantage of the properties of human hearing to remove
signal content which is imperceptible to the human ear. The human auditory system
has so-called masking properties, where a loud sound will hide or mask a softer sound
underneath. The softer sound is irrelevant. The audio codec first transforms the
audio segment into the frequency domain (using a DCT instead of an FFT), and
then eliminates more than 90 percent of the bits by using larger quantization steps
(coarser quantization), but in such a way as to hide the quantization noise underneath
the masking threshold of hearing. These so-called perceptual audio codecs reduce the
 CHAPTER 1. INTRODUCTION 37

bit rate of two-channel stereo audio from 1.4 Mbit/s down to 96–128 kbit/s for near
CD quality. This represents a compression ratio of about 11 to 1, i.e. the number
of bits is reduced by a factor of 11. Near-CD quality codecs include MP3, Dolby
AC-3 and the newer MPEG 4 AAC, Other audio codecs reduce the bit rate down to
16–32 kbit/s, including Windows Media Player and Real Audio G2.

Image coding is used for storage and transmission of digital pictures, primarily on
the web. Image coders take advantage of the properties of human vision to remove
redundant signal content which is imperceptible to the human eye. The human eye
is less sensitive to distortion at high frequencies where the image changes rapidly
over a small distance (e.g. leafy trees) than at low frequencies where the image
changes slowly with distance (e.g. artificial objects). The JPEG image codec follows
a pattern similar to the audio codec. First the 64 pixel values in each 8x8 block
are transformed into the frequency domain. Then the codec eliminates more than 90
INSTRUCTOR COPY

percent of the bits by using larger quantization steps, but in such a way as to hide the
quantization noise in the high frequency parts of the image. The loss of image quality
is almost imperceptible at a compression ration of 10 to 1. An entirely different
approach is used by the GIF codec. It uses scanning lines a segments, and searches
for similarity. If it finds many sequences of numbers (e.g. 12345) that are the same, it
replaces that sequence with a shorter sequence (e.g. 1). This is a lossless compression.
FAX machines according to the T.30 specification, do a similar operation. This is
particularly effective for images which are text pages, since the scanning lines in white
space are all zeros, and are represented by a single bit.

Video coding is used for compression of video, such as digital satellite television.
Video codecs use the same ideas as image codec on each frame (intraframe coding).
Video codecs also use the similarities from one frame to the next (interframe coding)
to eliminate redundancy. For example, if the background does not change from one
frame to the next, then the background is not coded again in the second frame.
Motion estimation is used to predict where the image will change, to further eliminate
redundant bits. Video codec acronyms include H.261 and H.263 and MPEG. MPEG-
1 achieves VHS videotape quality (320x240 pixels) at 1.5 Mbit/s. Other codecs are
used for high definition digital TV (HDTV) at 6 Mbit/s and for video conferencing
at 384 kbit/s.

To summarize, source coding can reduce the number of bits needed to represent a
message by a factor of 10 or more. This concludes the overview of source coding for
speech, audio, image and video.

Encryption For some messages such as financial transactions, it is important that

they be received only by the intended recipient, and that others not be able to eaves-
drop on the sender, or spoof the receiver by sending a false message. Other examples
 CHAPTER 1. INTRODUCTION 38

are some text messages, speech over cellphone, pay-TV video, and some audio (MP3)
or images to be sold. Thus many bit streams need to be scrambled or encrypted,
so that only the intended sender can generate the message, and only the intended
receiver can decode it. Cryptography has been around since ancient times. The main
idea is that the message is scrambled by some algorithm using a secret key. The
scrambling is done by flipping the bits from 1 to 0 or vica versa (or changing the
bytes) in an apparently random way determined by the combination of the algorithm
and the key. The message can be unscrambled (decrypted) only if the receiver knows
both the algorithm and the key. A simple example for a text message is to shift each
letter (character) by one in the alphabet, so that A becomes B, B becomes C etc
and the word TEXT becomes UFYU. In this example, the algorithm is to shift the
characters, and the key is 1.

Modern cryptography such as the DES (Data Encryption Standard) uses much
INSTRUCTOR COPY

more sophisticated algorithms than shifting, and use keys of 40 to 1024 bits. The
longer the key, the longer it would take to try all possibilities. 40 bits keys are
considered inadequate, since there are 240 or about 1 million million 1012 possible
keys, and a modern desktop personal computers circa 2001 which can perform 1,000
million (109 ) calculations per second could try all possible keys in 1000 s.

One major problem with DES is that the sender must transmit the key to the
receiver in some secure way, which may be inconvenient or impossible. Public-key
cryptography uses two different keys, a public one for encryption and a private one
for decryption. Anyone can send an encrypted message by looking up the recipient’s
public key. Only the recipient can decrypt the message using her private key. The
security of the algorithm depends on the fact that it is very difficult to factor large
numbers into primes. Common public key algorithms are PGP (Pretty Good Privacy)
and RSA. Their disadvantage is that they are slow compared to DES. A common
solution is to use PGP or RSA to send the DES key, and then use DES for the rest
of the message.

Public key cryptography can also be used for digital signatures by reversing the
process. The encryption can be done only by the sender with her private key (the
signature) and the decryption can be done by anyone with the public key.

Digital watermarking is another kind of security mechanism which is a data stream

inserted into audio, image or video content. The watermark contains information
relevant to the ownership or authorized use of the content. The watermark can be
added before compression as a very low bit rate signal mixed in with the audio or
video. The watermark can instead be added after compression (source coding), and
may or may not be connected in with the compression algorithm.
 CHAPTER 1. INTRODUCTION 39

Channel coding Most digital messages require some protection against errors be-
fore they are transmitted over a noisy and interference-prone channel. Thus channel
coding is used to add some redundancy which can be used to detect errors, and in
some cases also correct errors. The cost of this redundancy is that these extra bits
must also be transmitted, in addition to the message itself. Channel coding uses error
detecting codes or forward error correcting codes. A simple error detecting code is
a redundant parity check bit added to every byte. The number of bits per byte is
increased from 8 to 9. If the number of 1’s in the byte is odd, the parity bit is set to 1,
otherwise it is set to 0. If there is one error in the byte, then the number of 1’s in the
byte will change, and the parity won’t match, and the error is detected. However, if
there are two errors in the same byte, then the parity will still match, and the error is
not detected. If an error is detected, then that byte is either skipped, or the message
is rejected and must be retransmitted. A more sophisticated but very common error
INSTRUCTOR COPY

detection code is the CRC-16 (cyclic redundancy check) code, which adds two bytes
(16 bits) to each segment of the message. The CRC can detect virtually all errors in
the message, with a probability of undetected error of 1 in 216 or 0.00015%.

A forward error correcting code can actually reconstruct the original message even
if some bits are lost. A simple error correcting code is to repeat each bit 3 times, and
do a majority vote. If there is one error in a group of 3 repeated bits, then this code
works fine, but if there are two errors in the 3 bits, it will fail. A more sophisticated
but very common code is the so-called Reed-Solomon (RS) code, named after it’s
inventors. A typical RS code adds several bytes to each message segment. RS codes
are standard on audio CDs and allow a CD to be played even if scratched. RS codes
are also standard on some digital cellphones to guard against noise and interference.

Summary To summarize, a digitized message is formatted in 4 stages: segmenta-

tion, source coding, encryption and channel coding. The source coding reduces the
number of bits needed by a factor of about 10, whereas the channel coding adds back
a few percent, but the net reduction is still substantial. The digitized compressed
encrypted and error protected message is now ready for transmission.

1.1.4.4 Modulation

The bitstream representing the message must now be modulated onto a carrier for
transmission over a distance. A simple system will treat the bitstream as a waveform
of square waves and modulate it onto the carrier in a similar way as was done for
analog signals. There are three basic kinds of modulation: Amplitude Shift Keying
(ASK) (digital AM) in which the amplitude of the carrier wave is modulated in step
with the message signal and Frequency Shift Keying (FSK) (digital FM) in which the
 CHAPTER 1. INTRODUCTION 40

frequency of the carrier wave is modulated in step with the message signal. Phase
Shift Keying (PSK) (digital PM) in which the phase of the carrier wave is modulated
in step with the message signal. ASK and PSK may also be used at the same time
on one carrier.

For optical communication systems where the carrier is a light frequency, ASK
is commonly used, and the laser is turned on and off in step with the 1’s and 0’s.
For radio and cable systems, most modern modulation systems use a combination
of ASK and PSK. Some modulation schemes are combined with channel coding us-
ing so-called convolutional codes. Some modulation schemes use multiple carriers
on slightly different frequencies, with the message bits divided among the carriers.
Some modulation schemes use multiple carriers on the same frequency, each carrier
connected to a separate antenna. In both cases, the modulation may be combined
with channel coding using redundant bits and redundant carriers.
INSTRUCTOR COPY

For radio and cable systems, modulation is normally performed digitally using
samples of the carrier wave along with the bit stream. Thus modulation may be
viewed as another kind of formatting, and the output of the modulator is yet another
sequence of numbers. However, transmission via the radio or cable channel requires
some kind of analog voltage, since numbers alone cannot be transmitted over the
channel. Thus a digital-to-analog converter (D/A) is used to convert the numbers
(samples) back to analog voltages.

1.1.4.5 Upconversion

For digital radio systems, the modulation is normally done using a carrier at a lower
frequency than the desired carrier frequency for transmission over the channel. Thus
the modulated carrier output of the D/A and thus upconverted to the desired carrier
frequency. The upconversion done using a locally generated carrier wave in a special
analog circuit called a mixer that performs a multiplication of the D/A output and
the local carrier wave.

1.1.4.6 Amplification

The modulated carrier wave will need to be amplified to sufficient power to be carried
over the required distance over the channel. The amount of power required will be
determined later.
 CHAPTER 1. INTRODUCTION 41

1.1.4.7 System examples

Examples of digital communications (transmission) systems.

• web access via dialup modem over telephone line

• web access via cable modem for home internet connection

• digital subscriber loop (DSL) for home internet connection

• digital satellite TV (12 GHz small dish)

• digital audio broadcast

INSTRUCTOR COPY

• CD, DVD, DAT

• undersea fiber optic cable

• intercity fiber optic cable

• TV remote control

• digital cellphone (with short message service and internet access)

• satellite cellphone

• satellite internet access

• office wired local area network

• wireless local area network

• wireless internet access

• wireless cable to home (fixed wireless)(microwave multipoint distribution ser-

vice)

1.1.5 Channels, Power and Bandwidth

In this section, we learn how the bandwidth of the message and the distance affects
the design of a communications system. For example, why is a dish antenna needed
for satellite TV, but only a small whip antenna is needed for a satellite cellphone?
Why do the new digital satellite systems use a small pizza size dish, whereas the older
analog satellite TV uses a big 8 foot dish? Why did the Voyager space probe give
many fewer pictures from Uranus then from Jupiter ? Why is DSL (Digital Subscriber
 CHAPTER 1. INTRODUCTION 42

Loop) internet access only available within a few miles of the telephone central office
? Why does a TV broadcaster need one million watts to reach households 30 miles
in the suburbs, whereas the marine radio weather broadcast needs only 100 W to
cover the same distance ? Why is the range (distance) of the newer 900 MHz cordless
phones so much better than that of the older 49 MHz variety ?

The laws of physics impose fundamental limits to the capability (quality, data
rate) of a communications link. Practical engineering and economics (cost) impose
additional limits.

In this section, we learn that cost is related directly to the bandwidth of the
message and the distance over the channel.

For any communications system designed to send a message over a distance from
INSTRUCTOR COPY

point A to point B linked via some channel (fiber, cable, wire, radio), the available
resources are transmit power of the carrier wave and channel bandwidth, and the
obstacles to be overcome are noise and interference at the receiver. The longer the
distance, the weaker the carrier wave becomes, so more transmit power is needed to
overcome the noise at the receiver. For a given distance, more power and more

bandwidth means more information can be sent and/or higher quality and re-
liability can be achieved. But both power and bandwidth cost money; how much
will depend on the details of the communications system and (for radio) the current
auction price of a slice of radio spectrum. In the next sections, we review qual-
ity, channels (noise and interference), resources (power and bandwidth), and system
design to achieve acceptable quality.

1.1.5.1 Quality

To design a working link from A to B over a given distance with acceptable quality
at the lowest possible cost we need to first specify the customer performance (quality
of service) requirements.

Quality may come in different forms. For example, FM radio with a channel
bandwidth of 200 kHz gives better quality (more fidelity, less noise) than AM radio
with a channel bandwidth of 10 kHz. A high power cellphone with antenna mounted
on the car gives better quality (more reliability, less noise) than a low power handheld
cellphone inside the car. An internet connection with cable or DSL can be used to
send more information than with a dialup telephone line.

The quality and performance requirements will depend on the type of signal
(speech, audio, image, video, text, data). These performance requirements are usu-
 CHAPTER 1. INTRODUCTION 43

Ratio 1 1.5 2 3 4 5 10 15 20 30 40 50 100 1000 10,000

Decibel 0 2 3 5 6 7 10 12 13 15 16 17 20 30 40

Table 1.6: Ratios and decibels

Analog signal Bandwidth Signal-to-Noise ratio

Speech 3 kHz 1,000 to 1 (30 dB)
Audio 20 kHz 1,000,000 to 1 (60 dB)
TV video 6 MHz 100,000 to 1 (50 dB)

Table 1.7: Bandwidth and signal-to-noise ratio for different signal types

ally expressed for digital signals as the data rate at a given error rate, or for analog
signals as the fidelity (bandwidth) and signal-to-noise ratio. Alternately, we may be
INSTRUCTOR COPY

given the power and/or bandwidth that we can afford, and then need to compute
the maximum data rate and distance that can be achieved. Example performance
requirements are shown in the table. Ratios such as the signal-to-noise ratio (S/N)
are often written as decibels (dB), where the ratio in dB is 10 times the logarithm of
the ratio, i.e. x(dB) = 10log(x) dB.

For digital signals, the data rate is related to the bandwidth and signal-to-noise
ratio. The theoretical maximum data rate with perfect quality (no errors) that can
be squeezed through a given bandwidth with a given signal-to-noise ratio was worked
out by Claude Shannon in the 1940’s. The Shannon capacity formula is the E = mc2
of communications theory, so it is the only formula we will present in this chapter.
Shannon’s formula is C = W log2 (1+S/N ) where C is the capacity in bits per second,
W is the bandwidth in Hz, S/N is the signal-to-noise ratio. log2 is the logarithm of
base 2.

The Shannon capacity represents a theoretical upper limit to the data rate as
a function of bandwidth and signal-to-noise ratio. The Shannon capacity formula
shows how to trade off power and bandwidth to achieve a desired capacity in bits per
second. The capacity increases directly (linearly) with bandwidth, but logarithmically
with signal power. If we double the bandwidth, we double the capacity. But if we
double the power (and thus the signal-to-noise ratio), then the capacity increases only
slightly. So if bandwidth is limited then increasing capacity requires a lot of power.
But if bandwidth is cheaply available, and power is expensive, then we would choose
to use less power and more bandwidth.

In theory, if the signal-to-noise ratio S/N is 1 (the signal power equals the noise
power), we can transmit a bit rate C equal to the bandwidth W with no errors.
If S/N is 15, then C equals W times log2 (1 + 15) = 4 times W . In practice, the
Shannon capacity can be approached quite closely. Telephone modems at 56 kbit/s
 CHAPTER 1. INTRODUCTION 44

Digital signal Data rate Max acceptable bit error rate

Speech 13 kbit/s 0.01 (10− 2)
CD quality music 96 kbit/s 0.0001 (10− 4)
Video VHS tape quality 1.5 Mbit/s 0.0001 (10− 4)
Text web browsing 56 kbit/s 10− 9

Table 1.8: Data and error rates for different signal types

are near Shannon capacity, with bandwidth W = 3 kHz and S/N = 10, 000, As a
rough estimate when no details are available for a typical channel with a typical
S/N , a practical rule of thumb is that to achieve acceptable quality (low bit error
rate) the data rate is equal to the bandwidth. The most advanced communications
systems can come very close to Shannon capacity.
INSTRUCTOR COPY

1.1.5.2 Channels and path loss

As a carrier wave travels down a channel (wire, cable, fiber, free space), it becomes
weaker with increasing distance. The ratio of the input power to the output power is
called the path loss or attenuation of the channel. The path loss depends both on the
distance and the type of channel. In order of increasing loss: fiber, cable, wire, free
space. The loss for each channel can be calculated from basic principles of physics,
and depends on both the distance and the carrier frequency. If the channel is a cable
(fiber of coax), then we find the loss from the manufacturer loss specifications, and
this loss increases directly with distance (twice the distance, twice the loss). If the
channel is radio or free space optical, then for free space line of sight propagation
such as from earth to a satellite, the signal spreads out over a sphere in all directions,
resulting in the path loss increasing as the square of the distance (twice the distance,
4 times the loss). Thus a light bulb at 2 m distance appears 4 times weaker than at
1 m distance. In other words, assume we can get enough light to read with a 25 W
light bulb at 1 m distance. If we move the light to a 2 m distance, then we would
need a a 100 W lightbulb to get the same amount of light. If the channel is not line
of sight, such as the radio channel between a cellphone and a base station, then the
path loss increases much more rapidly, as the fourth power of the distance (twice the
distance, 16 times the loss). If a cellphone needs 1 W to reach a base station at 10 km
with acceptable speech quality, then if we move to 20 km, we need 16 W watts for the
same quality.

The power of the carrier wave obtained at the receiver over a link for a given
distance is simply the transmitted power divided by the path loss of the channel.

Channels can change the shape of the signal as well as make it weaker. This
 CHAPTER 1. INTRODUCTION 45

change of shape is called distortion, and can be overcome by a receiver designed to

compensate for it. Channels may also contain interference in the form of unwanted
signals from other users sending their own messages. This interference can be reduced
by some receivers, or may have to be tolerated.

Often it is not practical or cost-effective to build a transmitter of the required

power. In this case, the transmitted power can be effectively increased by using
directional antennas such as dishes, yagis etc. In general, if the area of the antenna
is doubled, then the power can be reduced by half.

The cost of a communication system is directly related to the power and the
bandwidth. To achieve a desired quality of service, the power of the transmitted
carrier wave is determined by multiplying 6 factors
INSTRUCTOR COPY

• bandwidth (typically equal to the data rate)

• path loss (from distance requirement, carrier frequency and laws of physics)

• required signal-to-noise ratio (from service quality, typically about 10)

• thermal noise constant (from laws of physics)

• receiver noise factor (from laws of physics)

• safety margin (to overcome variations in the channel path loss and/or unwanted
interference).

The first 2 factors depend on the message data rate and the distance. The power
(and thus cost) is directly related to the bandwidth. Thus even though technology
is always improving, the laws of physics impose fundamental limits to the capability
(quality, data rate) of a communications link.

We can now answer the questions at the beginning of this section by comparing
bandwidth (data rate) and distance for each case.

Why is a dish antenna needed for satellite TV, but only a small whip antenna
is needed for a satellite cellphone? answer: For satellite TV, the distance is long
(22,300 miles geostationary orbit) and the bandwidth is large (6 MHz) whereas for
the satellite cellphone, the distance is much smaller (800 km low earth orbit) and the
bandwidth is small (3 kHz). Thus the power needed is much larger for satellite TV,
hence the dish.

Why do the new digital satellite systems use a small pizza size dish, whereas the
older analog satellite TV uses a big 8 foot dish? Answer: Analog TV using FM on
 CHAPTER 1. INTRODUCTION 46

satellite needs 30 MHz of bandwidth, whereas digital TV takes advantage of source

coding compression to reduce the bandwidth to 1.5 MHz, a savings of 20-fold in power
or antenna size.

Why did the Voyager space probe give many fewer pictures from Uranus then from
Jupiter ? Answer, since the transmit power did not change, the increase in distance
had to be compensated by a decrease in bandwidth (data rate).

Why is DSL (Digital Subscriber Loop) internet access only available within a few
miles of the telephone central office, whereas plain old telephone service is available
in far-away rural areas ? Answer: DSL has a 2 MHz bandwidth compared to 3 kHz
for voice telephone service. For the same power, DSL can travel less distance.

Why does a TV broadcaster need one million watts to reach households 30 miles
INSTRUCTOR COPY

in the suburbs, whereas the marine radio weather broadcast needs only 100 W to
cover the same distance ? Answer: A TV signal needs 6 MHz bandwidth compared
to 30 kHz for the marine radio signal, so to cover the same distance, TV needs more
power.

Exercise 6
Why is the range (distance) of the new 900 MHz digital cordless phones so
much better than that of the older analog 49 MHz FM variety ?

Solution
Analog FM cordless phones need 30 kHz of bandwidth, whereas the digital
cordless phones use source coding to reduce the bandwidth to 13 kHz. This
factor of 2 reduction in bandwidth explains some of the increased range, but
the new digital phones also use much more power.

1.1.5.3 Receiver

At the receiver, the steps are carried out in reverse to produce the message output.
The receiver steps are shown in the Figures

analog sampler quantizer formatter modulator channel

voltage numbers bits bits waveform

Table 1.9: digital transmitter

 CHAPTER 1. INTRODUCTION 47

A/D segmentation source coding encryption channel coding

(compression) (security) (error protection)

Table 1.10: Formatter

channel demodulator deformatter D/A analog signal

waveform bits bits numbers voltage

Table 1.11: digital receiver

synchronization channel decode decryption source decoding

(segment boundaries) (error correction) (descrambling) (expansion)
INSTRUCTOR COPY

Table 1.12: deformatter

The received modulated carrier wave is downconverted to a more convenient lower

frequency and then demodulated. The demodulator may work in analog or may
include an A/D and work digitally. In either case, the demodulator output is a
bitstream. The deformatter does synchronization to identify the segment boundaries,
and then does channel decoding, decryption and source decoding to produce numbers
representing the message. When data segments are missing, some source decoders
can fill the gap with estimated data, and thus conceal errors. The D/A converts these
numbers to a time-varying analog voltage which activates the receiver transducer to
produce audio or video.

This completes the discussion of the functions of a communications link via a

channel connecting two nodes A and B.

1.2 Communications networks

A communications network is a system that provides a means to communicate not

only between two points which are directly connected via some channel, but also
between multiple points which are connected by multiple channels. The different
kinds of channels (fiber, cable, wire, free space) are sometimes called media.

Examples of communications networks include the public switched telephone net-

work (PSTN), private telephone networks inside offices and homes, public computer
networks such as the Internet and private computer networks (Local Area Networks
or LANs) in offices and homes.
 CHAPTER 1. INTRODUCTION 48

1.2.1 Nodes and Channels

A communications network is made up of nodes and channels.

node A node C node B

channel channel

Table 1.13: communication network with nodes and channels

Nodes are connected to other nodes via channels. The nodes transmit and re-
ceive messages via the channels connecting them. The arrangement of the nodes and
channels can take different forms, such as ring, bus, tree, mesh, star and random.
INSTRUCTOR COPY

The layout of a communications network with nodes and channels is directly

analogous to the layout of a railway network with nodes (cities) and channels (rails),
and the railcars acting as messages. A message from node A destined for node B may
have to pass through other nodes C etc. (multiple hops) before getting to B.

The intermediate nodes C act as switches or routers to route the message traffic
from A to B. All nodes must ensure that there will be no collisions between messages
on the channels, i.e. the nodes must avoid sending two or more messages on the same
channel at the same time.

Nodes may be general or special purpose computers, with a processing unit, mem-
ory, network interface circuits to connect to the channels. Nodes may also include
mass storage (disk drives) and/or input and output devices (buttons, lights, key-
boards, transducers).

The internet is a very large network consisting of many nodes and channels. Each
node has an address associated with it. Sending a special message called a ping
command to a distant node will result in return messages that reveal the addresses
of all the intermediate nodes.

1.2.2 Circuit and Packet Switching

There are two basic approaches to arranging the flow of message traffic from A to B
through the network. A circuit switched network such as the PSTN finds a dedicated
route connecting A and B, and does not permit others to share the route. The
message can flow continously from A to B without interference. Using the railway
analogy, circuit switching is like a continuous train running from A to B, with new
cars connected at A, and removed at B. No other trains going from C to D can use
 CHAPTER 1. INTRODUCTION 49

any of the rails occupied by the A to B train. The dedicated route is kept reserved
for the duration of the connection (phone call) from A to B.

In one variant of circuit switching, the channel may use (say) ten carrier frequen-
cies to support ten parallel circuits from A to B, so that the message uses one carrier
and other users can use the other carriers. This sharing of the channel is called Fre-
quency Division Multiplexing (FDM) on radio and cable channels, and Wavelength
Division Multiplexing (WDM) on fiber channels, where each user gets a dedicated
carrier.

In another variant of circuit switching, the continuous train (message) has a loaded
railcar (segment) only every tenth car, the other nine are empty, and dont need to
be sent at all. In this variant, other trains (users, messages) can use the empty
railcar slots, and the channels can be shared in this organized way. This sharing
INSTRUCTOR COPY

of the channnel is called Time Division Multiplexing (TDM), where each user gets
dedicated time slots at regular intervals. The circuit between A and B is called a
virtual circuit, and may also be called a logical channel (a channel in regular time
slices, as opposed to a dedicated physical channel).

A packet switched network such as the Internet divides the message into segments
called packets, and instructs each packet to find its own route from A to B. Other
packets going from C to D may share part of the route. Using the railway analogy,
packet switching is like dividing the train into individual railcars, and letting each
one choose its own route from A to B. Other railcars going from C to D may share
part of the route. The nodes of the packet switched network must make sure each
packet (railcar) arriving at that node is told which node to go to next, and also must
control access to the channel in order to prevent two packets from using the same
channel (rails) at the same time.

One important distinction: In packet switching, the route from A to B may vary
from one packet to the next, so that each packet must contain source and destination
addresses and control information. This adds some additional overhead bits which do
not contain any message data, but take up time on the network. In circuit switching
the route is prearranged, and no addresses are needed.

The choice between circuit and packet switching depends on the characteristics
of the message traffic and the cost of the nodes and channels. For long continuous
messages, circuit switching is preferred since there is no overhead and each message
segment is guaranteed a channel from A to B. For short bursty messages with gaps
in between, packet switching is preferred since the gaps can be used by others and
channels are not wasted.
 CHAPTER 1. INTRODUCTION 50

1.2.3 Queueing and buffers

In a circuit switched network, when a message segment arrives at a node, it does not
have to queue, it simply carries on to the next channel towards the next node. The
message segment may be delayed only a few bit times while it winds its way through
the synchronization circuits in the node.

However, in a packet switched network, a packet may have to wait (queue) at

a node for an undetermined time until the following channel is clear. Thus a node
contains a so-called buffer memory for temporary storage of packets waiting in the
queue. Queueing results in delay, but prevents the packet from entering a busy
channel, causing a collision and thus getting lost. Queueing happens frequently on
the Internet when channels (links) get congested and busy.
INSTRUCTOR COPY

1.2.4 Types of networks

Local area networks (LANs) are networks within a home or office. The most common
standard is Ethernet which is a star network with a central node called a hub and
twisted pair wire as the channel. Wireless LANs are convenient when it is awkward
to string wires or when mobility is required. Wide area networks (WANs) combine
multiple LANs in geographically separate locations. WANs may use mostly fiber
channels to connect the LANs. The Internet is a system of linked networks and
uses a variety of channels (fiber, cable, wire). Wide area wireless networks include
cellular telephone networks which may transmit both voice and data using free space
channels.

1.2.5 Dedicated and Shared channels

1.2.5.1 Dedicated (circuit switched) channels

On WANs, when a node is connected to several channels, the node has separate
buffers for each incoming channel, and thus can receive, queue (buffer) and process
messages arriving from several channels at the same time. The channel (medium) is
dedicated to the link between two nodes, and nodes have dedicated transmitters and
receivers for each channel connected to it.

A dedicated channel may also have several TDM virtual circuits between nodes,
thus appearing as several parallel channels which can be used independently by differ-
ent messages. Examples of such dedicated channels are T1 carrier (1.544 Mbit/sec)
 CHAPTER 1. INTRODUCTION 51

(which may be divided into 23 channels of 64 Kbit/sec each plus a control channel),
T3 carrier (43 Mbit/sec) using copper twisted pair, and Optical Carrier following the
Synchronous Optical Network (SONET) or Synchronous Digital Hierarchy (SDH) at
various levels such as OC-3 (155 Mbit/sec), OC-12 (622 Mbit/sec) and OC-48 (2.4
Gbit/sec). Other dedicated channels are point-to-point microwave circuits between
points up to 30 km apart. Dedicated channels of different data rates can be combined
with so-called add-drop multiplexors, which take selected time slots from

Dedicated channels may also use packet switching technologies such as Frame
Relay and ATM (Asynchronoous Transfer Mode). Frame relay is a form of fast packet
switching which relays frames and effectively creates a virtual circuit for packets, but
where the time slots for packets may not be perfectly regularly spaced. ATM is a
flexible way of using dedicated channels to support multiple virtual circuits, since
it can provide Constant Bit Rate (CBR) which appears like circuit switching on a
INSTRUCTOR COPY

dedicated channel, or 3 kinds of variable bit rate packet switching. Another example
of a dedicated channel is an ISDN line, which is twisted pair like a telephone line,
but contains three virtual channels: two 64 Kbit/sec circuit switched channels, plus
one 16 Kbit/sec packet switched channel.

1.2.5.2 Shared (packet switched) channels

On LANs such as Ethernet or FDDI (Fiber Distributed Data Interface), the channel
(medium) is shared among more than 2 nodes. In this case, nodes have only a single
receiver, and messages on the shared medium may pass right through the node and
not terminate there. In this case, the messages will collide if they arrive at the node
at the same time. Thus a method is needed to provide multiple access, i.e. access to
the channel for multiple users. This is called medium access control.

If there are relatively few messages, then collisions will be infrequent. In this case,
we may use random access, i.e. just send a messsage whenever we want. Any colliding
messages will not be received and thus will be retransmitted. A better method, used
by Ethernet, is to avoid collisions altogether by using Carrier Sense Multiple Access
with Collision Detection (CSMA/CD). In other words, ’listen before sending’ (sense
the carrier) and ’stop sending if you detect someone else sending’ (detect a collision).
Another method of medium acess control is polling, where each user is polled in turn
and asked to transmit. In hub polling, the polling message or token comes from a
central hub. In token ring, the token is passed around a ring from one node to the
next, and nodes can transmit only when they have the token.
 CHAPTER 1. INTRODUCTION 52

1.2.6 Telephone network architecture

The telephone network is a hierarchical network of nodes (telephones and telephone

switches) and channels (dedicated or circuit-switched). The channels range from
twisted pair used in homes and offices, multiple twisted pairs linking homes and
offices to local switches, and fiber channels between local switches and between cities
using SDH, and undersea cables between continents. Satellite channels are not used
much any more for the telephone network, because of the long delay and echos. The
switches range from small local switches (private branch exchanges) used in offices, to
medium size switches for a neighborhood, to large switches that connect a city to the
long distance channels to other cities. The telephone network also includes a separate
packet network called SS7 (Signaling System 7) which sets up and tears down circuits
(calls), and looks up 800, 877, 888 and 900 numbers to find their location.
INSTRUCTOR COPY

The nodes and channels of the telephone network are owned by a variety of tele-
phone companies. Some of these companies specialize in local service in a given area,
and own the copper (twisted pair) wiring to homes and office, and local switches.
Other companies specialize in long distance service, and own fiber optic channels be-
tween cities and large switches connected with the local telephone companies. Still
others own both local and long distance nodes and channels. The telephone companies
have devised elaborate billing schemes to keep track of the duration of telephone calls
flowing through their part of the network and to bill the company whose subscriber
originated the call.

1.2.7 Internet architecture

The Internet is a loose association of LANs and WANs with millions of nodes (com-
puters) around the world, and all types of channels connecting them. The Internet is
analogous to a transportation network, with main channels (railways, highways) and
smaller channels (local roads), main nodes (cities) and smaller nodes (villages). In the
developed countries, there are so many channels, that even if a main channel drops
out, then the data can be sent via a different route. No one organization owns all the
nodes and channels of the internet. Nodes include end user computers, terminals and
internet appliances, as well as various servers, network routers and switches.

The Internet mostly follows a client-server model or architecture, where one node
called the end user device (client) runs the user application, and another node called
the server supports the application with data files, network access, processing power
and devices. A peer-to-peer architecture where each node has equivalent responsibil-
ities is also used.
 CHAPTER 1. INTRODUCTION 53

The main lines between cities are fiber or satellite, and are known as the Internet
backbone. The backbone is owned by the major telecommunications companies such
as GTE, MCI, Sprint, AT&T, UUNet which are the biggest Internet Service Providers
(ISPs). There are 5 major nodes in the US which connect the backbone to other
regional and local networks. The regional networks may also use fiber and satellite,
and the local lines (channels) may be cable or wire or free space. These smaller
networks are owned by regional and local ISPs. Homes and businesses connect to the
Internet via local ISPs using cable, wire (DSL), or dialup telephone channels. Local
ISP’s pay access charges to regional ISPs who in turn pay for access to the backbone.

It is important to note that these backbone channels may be used simultaneously

by the Internet and the PSTN, with some virtual circuits assigned to the Internet
and other virtual circuits assigned to the PSTN, according to traffic demand.
INSTRUCTOR COPY

The Internet started in the late 1960’s as a research project of the US Department
of Defense Advanced Research Projects Agency (DARPA), and the TCP/IP protocol
(explained below) was completed in 1974. The Internet grew gradually in the 1970’s
and connected more government agencies, universities and large corporations. In
1990, the National Science Foundation took over the management of the Internet
backbone, and in 1995, a consortium of commercial ISPs took over.

Each node in the Internet is assigned a so-called IP address, which is the equivalent
of a telephone number in the PSTN. An IP address is a 4-byte (32 bit) number, usually
written in the form 142.104.114.180, where each of the 4 bytes is a value in the range
0-255 (00000000-11111111).

Domain names such as ve7ab.com are translated to a specific IP address using the
Domain Name Service (DNS). Domain names are often used, because they are easier
to remember, and the corresponding IP address can be changed if the domain name
owner moves to a computer with a different IP address. The Internet has DNS servers
which do the translation, and if a particular server does not know the translation, it
sends messages to other DNS servers to find out.

No one company or institution owns the whole Internet. It is a collaborative

enterprise of many organizations.

One organization ICANN (Internet Corporation for Assigned Names and Num-
bers) is a non-profit corporation which is responsible for allocating IP addresses.
ICANN has delegated domain name registration to other organizations within each
country. These domain names can then be mapped to an IP address.
 CHAPTER 1. INTRODUCTION 54

1.2.8 Protocols

A network needs pre-arranged procedures called protocols for setting up a connection

between A and B, transferring message bits from node to node via the channels,
and guaranteeing correct delivery of the message. These protocols are somewhat
analogous to the protocols used to communicate between people: the right things
must be said at the right time.

Circuit switched networks need protocols such as SS7 for setting up and terminat-
ing a circuit, but not for sending and receiving the message itself. Packet switched
networks need protocols for sending and receiving the message packets and routing
them correctly to their destination. Thus we focus on protocols for packet switched
networks.
INSTRUCTOR COPY

The protocols are divided into layers to provide different services. according to
the Open System Interconnection (OSI) reference model. The functionality of this
model exists in all communications systems, but the implementations may vary in
practice, and some layers may be combined.

The OSI layers are:

• layer 1, physical layer: the physical transmission of waveforms from A to B over

the channel, layer 2,

• link layer: sending packets across the link between two nodes and ensuring
correct reception.

• layer 3, network layer: sending and routing packets from A through the network
to B,

• layer 4, transport layer: ensuring the complete message gets from A to B, i.e
all packets arrive and in the correct order.

• layer 5: session layer (often combined with transport layer).

• layer 6: presentation layer: interpreting the bits as audio, video, text, web
pages, etc. source coding and decoding, presenting decoded bits correctly to
the correct transducer.

• layer 7: application layer: program-to-program communications, such as HTTP

(hypertext transfer protocol) between browser and web server used for viewing
web pages.

This sequence of layers is called a protocol stack.

 CHAPTER 1. INTRODUCTION 55

For WANs with dedicated channels between nodes, layer 1 may be a virtual circuit
provided by SDH or T1. For shorter distances, layer 1 may be a real dedicated circuit
such as a twisted pair wire used for DSL.

For shared channels, LAN protocols such as Ethernet and FDDI provide both
Layers 1 and 2 and part of layer 3. If the medium is shared, then layer 2 includes
a medium access control (MAC) layer,and layer 3 includes call admission control to
prevent entry of new messages if the network is congested. SS7 for the PSTN uses
layers 1,2,3 and 7. The mapping of protocol to OSI layers is not always exacat, but
the functions of the layers are present.

1.2.8.1 TCP/IP - moving data on the Internet

INSTRUCTOR COPY

In what follows, we describe the protocols of the Internet which are called TCP/IP for
Transport Control Protocol/Internet Protocol. for layers 4 and 3 respectively. The
mapping of TCP/IP to the OSI model is not exact. Different kinds of channels (wire,
cable, fiber, wireless) use different protocols for layers 1 and 2. Since the Internet
uses all kinds of channels, layers 1 and 2 are not included in TCP/IP specification.

Layer 1 is the physical network cards and cabling. Layer 2 is the drivers that
access the network.

The message is divided up into segments ranging from 48 to 65,000 bytes. Each
protocol layer needs to add and interpret so-called header bits for source and desti-
nation IP addresses, control, and running the protocol. Each layer adds its header
bits to the message bits to form a packet. At the message source, the message is
encapsulated or enveloped by these headers. At the message receiver, these headers
are peeled off like an onion to reveal the message bits.

A packet containing a complete entire message is called a datagram. If a message

is divided into multiple packets, then it is important to keep them in the correct order
and make sure none are missing.

Each node contains a processor to execute the TCP/IP protocol stack, and buffer
memory to store packets.

The physical Layer 1 provides the physical movement of bits from one node to the
next node via the channel. Layer 1 includes modulation.

The link Layer 2 data link control (DLC), moves packets between two nodes
connected by one channel (single hop), and adds CRC bits. If the channel coding
detects an uncorrectable error, the layer 2 DLC requests a retransmission (repeat) of
 CHAPTER 1. INTRODUCTION 56

that packet.

The network Layer 3 (IP) provides unreliable best effort delivery of individual
packets (datagrams) via multiple nodes and channels (multiple hops) from source to
destination. The key functions are addressing and routing, fragmentation to divide
up and handle large packets, and setting priorities. Layer 3 also measures the lifetime
of a packet to make sure it does not enter a routing loop and circle endlessly in the
network. Each node contains a routing table used by layer 3 to decide where a packet
should go next. These tables are updated frequently.

The transport Layer 4 (TCP) provides the equivalent of a circuit switched con-
nection (called a virtual circuit) over a packet network. Thus layer 4 handles con-
nection setup and termination. Layer 4 provides end-to-end connection-oriented
flow-controlled full-duplex reliable byte stream delivery. End-to-end means A to B.
INSTRUCTOR COPY

Connection-oriented means that if the message is divided into multiple packets, then
they are presented in the correct order and no packets are missing. This is achieved
by using a sequence numbers for each packet. Flow-controlled means that if the net-
work is congested, i.e. a link is busy and the buffer to that link is nearly full, then
packets are temporarily delayed at a node until the links are clear. Full duplex means
simultaneous transmission in both directions, from A to B and from B to A.

Layer 4 also provides multiplexing of multiple logical connections, for example

where there is simultaneous audio, video and text communications between two nodes.
For this purpose, TCP uses port numbers to identify the sending and receiving ap-
plications. Common port numbers are 25 for email and 80 for web. The combination
of IP address and port number is called a socket. Each TCP connection between two
nodes is uniquely identified by the two sockets, one at each end.

An alternate layer 4 called User Datagram Protocol (UDP) is used when all packets
are treated as independent datagrams. This is suitable for short messages such as
mouse clicks over the Internet, and is also used in cases such as real-time streaming
of audio or video, where lost packets dont matter very much. Another alternative is
Real-Time Protocol (RTP).

The presentation Layer 6 provides protocols for identifying the meaning of the
message bits (audio, video, text, etc). source coding (e.g. JPEG, MP3), encryption,
and HTML (Hypertext markup Language) or XML (Extended Markup language) for
web page content directing browsers what to display. HTML and XML are similar to
publishing software and use tags to specify page layout and other display features.

The application Layer 7 provides a variety of protocols including HTTP (Hyper-

Text Transfer Protocol), the underlying protocol used by the World Wide Web for
requesting and retrieving web pages. Other protocols are FTP (file transfer protocol)
 CHAPTER 1. INTRODUCTION 57

for moving data files from one computer to another, and Telnet for remote access
(login) to a computer. Several protocols are used for email, including SMTP (sim-
ple mail transfer protocol), POP (post office protocol) and IMAP (Internet message
access protocol).

1.2.8.2 Types of nodes on the Internet

Nodes on the internet come in various forms. All nodes on the Internet will have
layer 1 cabling and layer 2 drivers. Most nodes will implement the TCP/IP protocol
stack, and an application such as HTTP. Client nodes are end-user devices such as
computers, wireless terminals or net appliance, and run application software. Server
nodes include web servers, servers, other application servers, routers proxy servers,
INSTRUCTOR COPY

caching servers, firewalls. Web servers serve web pages and other content. Secure
servers use the Secure Sockets Layer (SSL) application layer protocol to create a
secure connection between client and server. SSL encrypts each packet, and is used
for confidential user information such as credit card numbers and bank statements.

Routers, proxy servers, caching servers and firewalls carry out one or more of three
main functions: traffic supervision (control of traffic flowing through the network),
bandwidth management (reducing the total traffic as much as possible), caching (stor-
ing information for later retrieval) and encryption.

Routers are connected to several channels, receive incoming packets, read the layer
3 IP address, and simply direct the packets onto the next channel towards its final
destination. Routers have routing tables used to select the correct outgoing channel.
Routers do not interpret the higher layers.

Proxy servers act as intermediaries between clients and web servers for the purpose
of reducing network traffic. The client (web browser) is configured to send all HTTP
requests (or other applications) to the proxy server, which fulfills the request and sends
the response to the client. For a new request never made before, the proxy server
sends the client request to the web server, retrieves the information, and responds
to the client. The proxy server also stores the information in a cache (a large disk
drive) for later reuse. Future requests by clients for the same information are then
fulfilled using the information in the cache, thus avoiding the need for traffic to the
web server. This reduction in traffic can be significant for frequently requested web
sites. Web browsers may also retain their own local cache on the client itself.

Caching servers are transparent to the client, and in general cannot be detected.
The browser sends its request to the web server, but the caching server intercept the
traffic on the standard HTTP port 80 (using a router). If the caching server has the
information, it sends the reply to the client, otherwise it forwards the request to the
 CHAPTER 1. INTRODUCTION 58

web server, and keeps a copy of the information. Caching servers can be scaled for
large networks. Some specialized caching servers are configured to cache rich content
(audio/video) of selected web sites in order to improve the speed of delivery of their
content.

Firewalls provide controlled access from the internet to a corporate or other private
internal TCP/IP network (called an intranet). Firewalls acts as very smart routers
which decide which traffic is allowed to pass. Firewalls intercept packets from the
IP layer, but perform processing and routing at the application layer. Packets from
certain applications may be blocked, and other applications may be allowed to pass.
The firewall may work in conjunction with an authentication server which requires a
client to provide a password. If authenticated, then packets from the client IP address
are authorized to pass through the firewall for a finite time.
INSTRUCTOR COPY

The distinctions between firewalls, caching servers, proxy servers and routers may
be blurred, and their functions of traffic management, caching and encryption com-
bined into one or a few computers.

1.3 Signals in time and frequency domain

Learning objectives

In this section, we learn the mathematics of a general signal or waveform and how it
can be represented in the time and frequency domain. We also learn about the idea of
a system that processes (modifies) the signal to generate another (different) signal

1.3.1 Introduction to Signals and Systems

We begin by developing an intuition of what a signal is in the context of signal

processing and communications, and how we can represent the signal in a useful way.

We also develop an intuition of what is a system that processes signals to generate

other signals that are needed to accomplish some goal or objective.
 CHAPTER 1. INTRODUCTION 59

1.3.1.1 Signals

A signal is some quantity as a function of time that can be represented on a graph

with time on the x-axis and amplitude (positive or negative) on the y-axis.

Signals can arise from natural phenomena such as acoustic waves, electromagnetic
waves, mechanical vibrations or from human sources such as speech or touch/pressure.
These signals are converted to electrical form by a sensor (transducer) of some kind,
such as a microphone, antenna, barometer, touchscreen or seismograph. The sensor
output is an electrical signal comprising voltages and currents in wires that can be
observed by electrical instruments such as voltmeter, ammeter, oscilloscope. Some
signals are intrinsically electrical signals in electrical form without the need of a
transducer, such as the voltages in an electric motor. These electrical signals are
INSTRUCTOR COPY

so-called continuous-time signals that vary continuously with time.

1.3.1.2 Systems

A system processes input signals to generate other output signals. In an continuous-

time system,the continuous-time input signals are processed by electronic compo-
nents such as resistors, capacitors, inductors, transistors, op-amps, diodes to gener-
ate other signals. The system made up of electronic components may be modelled by
continuous-time mathematics such as differential equations and Laplace transforms.

For example, a simple system consisting of only a capacitor may be modelled by

the mathematical operation of differentiation I(t) = CdV (t)/dt. The output signal
I(t) is the derivative of the input signal V (t).

In a discrete-time system, the continuous-time signals are sampled and converted

to discrete-time signals that are essentially a sequence of numbers that can be stored
in computer memory. The sequence of numbers and the sampling time represent
the continuous-time signal. The sampling and conversion is done by an analog-to-
digital converter (ADC). The discrete-time signal may be analyzed using digital signal
processing algorithms using discrete-time mathematics such as difference equations
and Z-transforms. The algorithms are implemented in software that runs on computer
hardware (processor/CPU) to identify the system characteristics and parameters.

For example, consider a simple discrete-time system that is intended to perform

the function of a capacitor, The signal processing algorithm to represent the function
of a capacitor is a difference equation y[n] = x[n] − x[n − 1] that implements discrete-
time differentiation. Thus the system generates the output discrete-time signal that
is the digital derivative of the input signal.
 CHAPTER 1. INTRODUCTION 60

The output signal is converted to continuous-time form via a digital-to-analog con-

verter (DAC) and routed to a display device (computer screen, loudspeaker) and/or
used to control an actuator (e.g. motor).

1.3.1.3 System implementation

Every electrical or electronic device (containing electrical or electronic components

and transducers) is an continuous-time system that processes continuous-time signals.
Many electronic devices are discrete-time systems that include an ADC and/or DAC
and computer hardware to process discrete-time signals.

A continuous-time system may be simulated in software. In this case, the input

INSTRUCTOR COPY

discrete-time signals are generated by software (computer program) that simulates a

natural phenomenon and stored in memory. These discrete-time signals are processed
in a manner that simulates the continuous-time signal processing in the continuous-
time system to generate the output discrete-time signals. In this case, there are no
transducers, ADC or DAC and the system is entirely discrete-time and processes only
discrete-time signals. Such discrete-time systems may be used to model and simulate
continuous-time systems and evaluate their performance without having to build the
continuous-time system.

In this book we will encounter both continuous-time and discrete-time signals.

The mathematical description may be done in continuous time, thus allowing the
use of calculus (differentiation, integration) to describe the continuous-time signal
processing done by the system. A practical implementation of the system will often be
in discrete time, thus requiring a mathematical description of discrete-time (sampled,
digital) signals and discrete-time signal processing using discrete-time mathematics
(difference equations that implement delay, multiplication and addition operations).

In some books and articles, continuous-time systems are also call ed analog sys-
tems, and discrete-time systems are also called digital systems.

However, in this book we refer to analog signals such as voice and music as distinct
from digital signals that are a sequence of bits.

1.3.2 Examples of Signals and Systems

There are many examples of systems that process input signals to generate output
signals. Some systems may be viewed as a combination of several sub-systems, each
with its own input and output signals, where the output of one sub-system is the
 CHAPTER 1. INTRODUCTION 61

input to the next sub-system.

system input output

communications voice message voice message
radar radio wave reflected from target location and velocity of target
electric car driver inputs signals to drive wheels

In this book we focus on communications systems that transmit messages over a

distance.

1.3.3 Sine waves, a fundamental signal

INSTRUCTOR COPY

A fundamental signal is a sine wave as shown in fig. 1.1 plotted in the time domain
as a voltage (or current) versus time.

π π 3π 2π
2 2
−1

Figure 1.1: Sine wave

The sine wave has 3 attributes or properties: amplitude, frequency and phase. A
system may modify one or more of these properties to accomplish some goal.

The sine wave has instantaneous values (amplitudes) that vary with time as shown
in the plot. The sine wave has a peak (maximum) value and a negative peak value. In
continuous-time systems the amplitude is usually a voltage or a current. In discrete-
time systems, the amplitude is a number in computer memory.

The power of the sine wave is the mean square of all the instantaneous values
averaged over one cycle of the sine wave. The power depends on the peak instanta-
neous value only and does not depend on the frequency or the phase. We also define
the root mean square value as a single number representing the amplitude of the sine
wave which is an average value over one cycle and is the square root of the power
(since power in watts is the square of the amplitude in volts assuming an impedance
of one ohm).
 CHAPTER 1. INTRODUCTION 62

Amplitude

f Frequency
INSTRUCTOR COPY

(a) Positive frequency component

Amplitude

−f f Frequency

(b) Positive and negative frequency components

Figure 1.2: Amplitude vs frequency for a sine wave

The sine wave can also be plotted as an amplitude and phase versus frequency
(fig. 1.2a). If we extend the frequency axis to show both negative and positive values
of frequency, we observe two spikes δ−functions) at +f and −f (fig. 1.2b). The
concept of negative frequency will be introduced shortly.

Any general signal cam be represented as the sum of many sine waves, each with
different amplitude, frequency and phase. We know this from the theory of Fourier
series for periodic signals and Fourier transforms for any signal, see also section 1.4.

In this next section, we focus on one particular example of a signal and use a
discrete-time system to analyze its properties.
 CHAPTER 1. INTRODUCTION 63

1.3.4 Representing a human voice signal

To develop intuition we start with a natural phenomenon we all know, the human
voice. A sensor (microphone) is used to convert the human voice to electrical form as
an analog signal and an ADC converts the voice to numbers stored in memory and
displayed as a video on a screen.

This process can be done on a laptop computer using free software for Windows,
OSX and Linux such as Audacity or snkpeek or on an Android or iOS smartphone
using any number of apps with names containing the keyword “spectrum” or “spec-
trogram”. The software or app can be set to display the human voice (or any signal)
either in the time-domain or the frequency-domain.
INSTRUCTOR COPY

A view that combines both time domain and frequency domain in one view is
called a spectrogram.

Figure 1.3: Time vs Frequency domain

1.3.4.1 Time domain video

The time domain video is a 2-D graph showing the instantaneous signal level versus
time. The signal level may be positive or negative. The time axis shows a window
 CHAPTER 1. INTRODUCTION 64

of time depending on the display settings. The time domain video is similar to what
would be viewed on an oscilloscope.

If we whistle a tune into the microphone, the video will show sine wave at the
frequency of the whistle that changes with each note of the tune.

If we say a word that includes both vowels and consonants, then the video will
show a complex waveform that is difficult to interpret.

1.3.4.2 Frequency-domain video

The frequency domain video is a 2-D graph showing the instantaneous signal level
INSTRUCTOR COPY

versus frequency. The signal level is positive or zero. The frequency axis shows a range
of frequencies. For audio, this range is normally 20–20 000 Hz. The frequency-domain
video is similar to what would be viewed on a spectrum analyzer (see appendix D.1).

If we whistle a tune into the microphone, the video will show a spike at the
frequency of the whistle that changes position on the frequency axis with each different
note of the tune.

If we say a word that includes both vowels and consonants, then during vowels, the
video will show high amplitude at frequencies corresponding to the pitch of the voice
(typically 200–1000 Hz) and low amplitude at higher frequencies (1000–20 000 Hz).
During consonants the video will show high amplitude at higher frequencies (1000–
20 000 Hz)

1.3.4.3 Time-frequency video

The time-frequency video is called spectrogram or spectral waterfall or waterfall for

short. Each frame of the video is a 3-D plot which is like a 2-D color-coded terrain
map with the terrain viewed from above. The two dimensions are time and frequency
and the third color-coded dimension is amplitude. The colors are chosen using a
so-called colormap. The terrain map can also be displayed with a 3-D perspective
view.

The colormap defines which colors are used for high, medium and low amplitudes.
An actual geographical terrain map typically uses a “terrain” colormap with green for
low elevation, brown and white for high elevation. In this book we use a “spectrum”
colormap with blue and green for low amplitudes, yellow for medium amplitudes
orange and red for high amplitudes.
 CHAPTER 1. INTRODUCTION 65

The spectrogram is a time-frequency representation of the voice signal, since it

shows how the frequency content changes with time. The spectrogram provides infor-
mation about the signal that is not clear from the time-domain view. The spectrogram
also provides historical information about the frequency content that is not stored in
the frequency domain view.

The process (algorithm) for creating this spectrogram or waterfall requires using
a short-time Discrete Fourier Transform (DFT), most often implemented as a Fast
Fourier Transform (FFT), and is a version of the continuous-time Fourier transform
applied to discrete-time signals.

The concepts of continuous-time Fourier transform and Discrete Fourier transform

(DFT) are reviewed in section 1.4. The essential idea is that the Fourier trnmsform
finds the frequencies of the sine waves in a signal.
INSTRUCTOR COPY

In the section that follows section 1.5, we use the DFT to design an algorithm to
make a spectrogram.

1.4 Fourier transform and Discrete Fourier Trans-

form

In this section, we review the ideas and notation for the Fourier transform for continuous-
time signal and the Discrete Fourier Transform (DFT) for discrete-time signals.

1.4.1 Fourier transform concept and definition

Fourier analysis shows that any complex waveform can be resolved into sinusoidal
waveforms of a fundamental frequency and a number of harmonic frequencies. A
spectrum analyzer effectively performs the Fourier integral:

S(f ) = F{s(t)}
Z ∞
(1.1)
== s(t)e−j2πf t dt
t=−∞

The Fourier transform of s(t) is the integral of the product of s(t) and a complex
exponential e−j2πf t = cos 2πf t − j sin 2πf t. The Fourier transform S(f ) is a complex
function with both amplitude and phase.
 CHAPTER 1. INTRODUCTION 66

In theory, since the integral extends over all time −∞ < t < ∞, we need to
observe s(t) for all time to obtain S(f ). However, in practice, the Fourier transform
is valid when the observation interval (i.e. the time spent observing s(t)) is greater
than the inverse of the lowest frequency component greater than zero in the signal
s(t).

Exercise 7
If the lowest frequency in a human voice signal s(t) is 100 Hz, then what is the
required observation interval (integration limits) to obtain a valid S(f )?

In the following subsection, we elaborate on the meaning of the complex exponen-

tial inside the Fourier integral. Then we will interpret the Fourier integral, first using
INSTRUCTOR COPY

real signals (cosines and sines) and again using complex signals (complex exponen-
tials).

1.4.1.1 Complex numbers, complex exponentials and phasors

√
Complex numbers involving j = −1 are reviewed in https://en.wikipedia.org/
wiki/Complex_number. In brief, complex numbers z are points in the complex plane
with real and imaginary axes and are written z = x + jy = aejθ . Some example
complex numbers are

z x y a θ
1 1 0 1 0
j 0 1 1 π/2
-1 -1 0 1 ±π
-j 0 -1 1 −π/2

A complex function of time s̃(t) = a(t)ejθ(t) assigns a complex number for each
value of time t. For a given fixed value of t = t0 , s̃(t0 ) = a(t0 )ejθ(t0 ) is a complex
number. An arrow drawn from the origin of the complex plane to the point defining
the complex number is called a phasor. In general, as t increases, the phasor s̃(t) will
move in the 2-D complex plane.

The integrand of the Fourier integral includes the term e−j2πf t . This term is
referred to as a complex exponential. It is a special case of a complex function of
time a(t)ejθ(t) , where a(t) = 1 is a constant independent of time and θ(t) = −2πf t is
a linear function of time whose slope depends on the frequency f . For the complex
exponential s̃(t) = e−j2πf t , the phasor will move in the complex 2-D plane, tracing
out a circle with center at the origin. Since the exponential is negative, the circle is
 CHAPTER 1. INTRODUCTION 67

traced out in a clockwise direction from θ = 0 to −π/2, etc. In this case, the phasor is
a rotating phasor that is rotating clockwise about the origin. The speed of rotation is
proportional to f . The phasor rotates one complete cycle of the circle (2π radians) in
1/f seconds. If f = 1 Hz, then the phasor rotates one cycle (2π radians) per second.

Some example values of s̃(t) = e−j2πf t for increasing values of t are tabulated
below.
t 2πf t e−j2πf t
0 0 1
1/4f π/2 −j
1/2f π −1
3/4f 3π/2 +j
INSTRUCTOR COPY

To represent the motion of the phasor as time progresses in a fixed plot that does
not change with time, we need to add the time dimension as a third dimension in
addition to the two dimensions of the complex plane (real and imaginary). Thus we
plot the complex function s̃(t) as a 3-D plot of t, <{s̃(t)}, ={s̃(t)}. A 3-D plot is made
in Matlab using the function plot3(t,Re(s),Im(s)).

If we plot the complex exponential s̃(t) = e−j2πf t , the result is fig. 1.4. Note that
the rotation of the helix is clockwise as time increases since the exponent is negative.
 CHAPTER 1. INTRODUCTION 68

1
=(z(t))

−1
INSTRUCTOR COPY

1
0 0
0.5 1 1.5 −1 <(z(t))
t

Figure 1.4: Helix with constant frequency, amplitude and phase.

Figure 1.4 shows three projections of the helix onto the 3 orthogonal planes in
3-D space:

• the real-time plane, This projection shows the real part of the helix versus time
(blue cosine wave).

• the imaginary-time plane. This projection shows the imaginary part of the helix
versus time (red sine wave).

• the real-imaginary (complex) plane. This projection shows the path traced out
by the arrowhead of a phasor (green circle).

Figure fig. 1.5 shows the same helix from a different perspective with the same
three projections.
 CHAPTER 1. INTRODUCTION 69
INSTRUCTOR COPY

Figure 1.5: Helix with constant frequency, amplitude and phase.

In fig. 1.6 we plot a positive complex exponential shifted in phase s̃1 (t) = ej2πf t ejπ/8
(blue) and a negative complex exponential s̃2 (t) = e−j2πf t (green) and their sum
s̃1 (t) + s̃2 (t) (red).
 CHAPTER 1. INTRODUCTION 70
INSTRUCTOR COPY

Figure 1.6: cosine wave is the sum of two complex exponentials

The positive complex exponential (blue) is a helix rotating in the counterclockwise

direction as time increases, and the negative complex exponential (green) is a helix
rotating in the clockwise direction as time increases. The sum is a sinusoidal wave
with both real and imaginary components. The angle of the sinusoidal wave depends
on the phase shift π/8 in s̃1 (t).

Exercise 8
Find an expression for the sum s̃1 (t) + s̃2 (t).

1.4.1.2 Fourier transform interpretation using real signals

The Fourier transform may be written in terms of cosines and sines as

 CHAPTER 1. INTRODUCTION 71

Z ∞
S(f ) = s(t)e−j2πf t dt
Zt=−∞
∞ Z ∞
(1.2)
= s(t) cos(2πf t)dt − j s(t) sin(2πf t)dt
t=−∞ t=−∞
= S1 (f ) − jS2 (f )

The integral finds the frequency components in s(t) by correlating s(t) with cosine
and sine waves at each frequency f .

For a signal s(t) containing only one frequency f = fc we have

INSTRUCTOR COPY

s(t) = cos 2πfc t (1.3)

and
Z ∞
S1 (f ) = cos(2πfc t) cos(2πf t)dt
t=−∞

To evaluate S1 (f ) for f 6= fc the integrand is the product of two cos waves as

shown in fig. 1.7.

Figure 1.7: product of two cos waves where f 1 6= f 2

The two cosine waves are blue and green. The integrand (red) is periodic with
period equal to f − fc and shows a sinusoid at f + fc . Thus the multiplication of the
 CHAPTER 1. INTRODUCTION 72

two sinusoids at f and fc results in the sum of two sinusoids at the sum and difference
frequencies f − fc and f + fc , as shown by the identity in the Appendix

cos (α − β) + cos (α + β)
cos α cos β =
2
α = 2πfc t
β = 2πf t

From the figure, the red curve is the sum of a low frequency cos wave at fc − f and
a high frequency cos wave as fc + f .

Over one or more cycles of the integrand, there is no DC component, and thus
the integral is zero.
INSTRUCTOR COPY

To evaluate S1 (f = fc ) for f = fc , the integrand is the product of two cos waves

at the same frequency as shown in fig. 1.8.

Figure 1.8: product of two cos waves where f 1 = f 2

The two cosine waves are blue and green with the blue underneath the green. The
integrand (red) is a cosine wave at 2fc plus a DC offset, as shown by the same identity
cited above but with α = β.

Over one or more cycles of the integrand, there is a DC component, and thus the
 CHAPTER 1. INTRODUCTION 73

integral is ∞, and we write

Z ∞
S1 (f = fc ) = cos2 (2πfc t)dt
Zt=−∞
∞
= 0.5[1 + cos(4πfc t)]dt
t=−∞
Z ∞ Z ∞
= 0.5 dt + 0.5 cos(4πfc t)dt
t=−∞ t=−∞
= 0.5∞

The same result applies when f = −fc since cos(−x) = cos x.

In general for any f including both f = ±fc and f 6= ±fc . if s(t) = cos 2πfc t then
INSTRUCTOR COPY

S1 (f ) = 0.5δ(f − fc ) + 0.5δ(f + fc )

where we use the δ− function defined as

δ(f ) = ∞, f = 0, 0, f 6= 0
δ(f − fc ) = ∞, f = fc , 0, f 6= fc

To summarize, the cosine waves are in phase for all time, the product of the two
cosine waves contains DC, thus the integral integrates DC over all time, resulting in
infinity (delta function) at f = fc . The same infinity (delta function) is obtained at
f = −fc . For all other frequencies f 6= fc the cosine waves drift in and out of phase
over time, there is no DC component, and the integral is zero.

Thus for s(t) = cos(2πfc t) we can write:

Z ∞
S1 (f ) = cos(2πfc t) cos(2πf t)dt
t=−∞
= 0.5[δ(f − fc ) + δ(f + fc )]
S(f ) = S1 (f ) − jS2 (f )

We still need to evaluate S2 (f ).

Exercise 9
Show that if s(t) = cos(2πfc t) as above, then S2 (f ) = 0, so that S(f ) = S1 (f )
for this case.
 CHAPTER 1. INTRODUCTION 74

Solution

Z ∞
S2 (f ) = cos(2πfc t) sin(2πf t)dt
t=−∞

We apply the identity in the Appendix

− sin (α − β) + sin (α + β)
cos α sin β =
2
When f 6= fc , the integrand is the product of cos and sin waves at different
frequencies. The integrand is periodic with period equal to f − fc . Over one or
more cycles of the integrand, there is no DC component, and thus the integral
INSTRUCTOR COPY

is zero.
When f = fc ,
Z ∞
S2 (f = fc ) = sin(4πfc t)dt
t=−∞
=0

the integrand is the product of a cos wave and a sin wave at the same frequency
which yields a sin wave at 2fc . The integral is again zero.

This result applies for any cosine waves in s(t), so that if there are multiple cosine
waves in s(t) at frequencies f1 , f2 etc, then S(f ) will contain multiple δ functions at
±f1 , ±f2 , etc. In this way, the Fourier transform finds all the cosine waves in a signal
s(t).

In general, s(t) may contain both cos and sin waves since cos(2πfc t + φ) =
cos φ cos 2πfc t − sin φ sin 2πfc t.

Exercise 10
Repeat the above calculation to find S(f ) for s(t) = cos(2πfc t + φ) where
φ = π/4.

The above analysis also applies to sine waves in s(t).

 CHAPTER 1. INTRODUCTION 75

Exercise 11
Repeat the above calculation to find S(f ) = S1 (f ) − jS2 (f ) for s(t) =
sin(2πfc t).

Solution

Z ∞
S1 (f ) = sin(2πfc t) cos(2πf t)dt
t=−∞
=0
Z ∞
S2 (f ) = sin(2πfc t) sin(2πf t)dt
t=−∞
(1.4)
INSTRUCTOR COPY

= 0.5[δ(f − fc ) − δ(f + fc )]
S(f ) = −jS2 (f )
1
= [δ(f − fc ) − δ(f + fc )]
2j
= 0.5e−jπ/2 [δ(f − fc ) − δ(f + fc )]

Exercise 12
Repeat the above calculation to find S(f ) for a complex signal s̃(t) =
a0 ej2πfc t+φ0 for constants a0 , φ0 Hint: Write ej2πfc t = cos 2πfc t + j sin(2πfc t).
Answer: S̃(f ) = a0 ejφ0 δ(f − fc ).

Solution
See the next section.

Exercise 13
Show that if fc = 0 then s̃(t) = a0 ejφ0 and S̃(f ) = a0 ejφ0 δ(f ).

1.4.1.3 Fourier transform interpretation using complex signals

The Fourier transform definition in terms of complex exponentials (firt seen in eq. (1.1),
and repeated here) is Z ∞
S(f ) = s(t)e−j2πf t dt (1.5)
t=−∞
 CHAPTER 1. INTRODUCTION 76

The integral finds the frequency components in s(t) by correlating s(t) with com-
plex exponentials at each frequency f .

The Fourier transform applies to complex signals s̃(t) just as well as to real signals
s(t).

If we choose s̃(t) to be a complex exponential at a particular frequency f = fc

with a complex scaling factor a0 ejφ0 then s̃(t) = a0 ejφ0 ej2πfc t = a0 ej2πfc t+φ0 and
Z ∞
S̃(f ) = s̃(t)e−j2πf t
Zt=−∞
∞
= a0 ejφ0 ej2πfc t e−j2πf t
t=−∞
Z ∞
INSTRUCTOR COPY

= a0 e jφ0
ej2π(fc −f )t
t=−∞

The integral finds the frequency components in s̃(t) by correlating s̃(t) with com-
plex exponential waves e−j2πf t at each frequency f .

To evaluate S̃(f = fc ) at any frequency f 6= fc that does not match the frequency
of s̃(t) the integrand is a complex exponential at frequency fc − f . When we integrate
(average) over one or more periods of the helix, there is no DC component and thus
this integral is zero.

To evaluate S̃(f = fc ) at a frequency f = fc that matches the frequency of s̃(t)

(but the phase and amplitude do not match) we have
Z ∞
S(f = fc ) = a0 ejφ0 ej2πfc t e−j2πfc t dt
Zt=−∞
∞
= a0 ejφ0 ej2π0t dt
t=−∞
Z ∞
jφ0
= a0 e dt
t=−∞

The integrand is a complex constant (DC) term a0 ejφ0 . Thus the integral is ∞,
and we write

S̃(f ) = a0 ejφ0 ∞, f = fc

Thus for any frequency f including f = fc and any other frequency f 6= fc

S̃(f ) = a0 ejφ0 δ(f − fc )

 CHAPTER 1. INTRODUCTION 77

The Fourier transform interpretation using complex signals is simpler than using
real signals. It is useful to observe that a complex exponential has no DC component
when averaged over one or more cycles of the helix and thus the integral of a complex
exponential over one or more cycles of the helix is zero.

Exercise 14
Evaluate the Fourier integral for s(t) a (real) cosine wave at a particular fre-
quency f = fc .

Solution
If s(t) is a (real) cosine wave with a particular frequency f = fc , then

s(t) = cos 2πfc t = 0.5[ej2πfc t + e−j2πfc t ] (1.6)

INSTRUCTOR COPY

Recall that the cosine wave is the sum of two complex exponentials, one positive
and one negative.
The Fourier transform is
Z ∞
S(f ) = 0.5[ej2πfc t + e−j2πfc t ]e−j2πf t dt
Zt=−∞
∞
= 0.5[ej2π(fc −f )t + e−j2π(fc +f )t ]dt
t=−∞

The integral finds the frequency components in s(t) by correlating s(t) with
complex exponential waves e−j2πf t = cos 2πf t − j sin 2πf t at each frequency f .
To evaluate S(f = fc ) at a frequency f = fc that matches the frequency of s(t)
we have
s(t) = cos 2πfc t (1.7)
and
Z ∞
S(f = fc ) = 0.5[ej2πfc t + e−j2πfc t ]e−j2πfc t dt
Zt=−∞
∞
= 0.5[ej2π0t + 0.5e−j2π2fc t ]dt
Zt=−∞
∞
= [0.5 + 0.5e−j2π2fc t dt
t=−∞

Over one or more cycles of the product, there is a DC component 0.5, and a
complex exponential at 2fc that has no DC component. Thus the integral is
∞, and we write

S(f ) = 0.5∞, f = fc
 CHAPTER 1. INTRODUCTION 78

We obtain the same result for S(f = −fc ).

To evaluate S(f ) at other frequency f 6= ±fc we have from above

Z ∞
S(f 6= fc ) = 0.5[ej2π(fc −f )t + e−j2π(fc +f )t ]dt
t=−∞

the integrand (red) is the sum of two complex exponentials, one at ”high fre-
quency” fc + f and one at ”low frequency” fc − f . There is no DC component
so the integral is zero.
Figures illustrating what the integrand looks like in a 3-D plot are provided in
the solution to the next exercise.
The result is that we write
INSTRUCTOR COPY

S(f ) = 0.5δ(f − fc ) + 0.5δ(f + fc )

Exercise 15
Repeat the calculation of S(f = fc ) and S(f 6= fc ) for s(t) = cos 2πfc t + φ0
and plot the integrand.

Solution
The algebra is very similar to the exercise above and thus omitted but figures
are shown.
We first evaluate S(f = fc ).
The two figures below fig. 1.9 and fig. 1.12 show the input signal s(t) =
cos 2πfc t + φ0 (blue), the complex exponential e−j2πf t (green) with f 6= fc ,
and the product s(t)e−j2πf t (red) which is the integrand of the Fourier trans-
form. The two figures show the same functions from different perspectives.
The integrand s(t)e−j2πfc t (red) is not centered at the origin and thus has a
DC component in both the real and imaginary part (magenta). The figure is
for the case φ0 = π/4.
 CHAPTER 1. INTRODUCTION 79
INSTRUCTOR COPY

Figure 1.9: product of cos wave with exp wave where f 1 = f 2

Figure 1.10: product of cos wave with exp wave where f 1 = f 2

The projections of the integrand (red) on the real and imaginary axis (magenta)
look like the figures in the previous subsection interpreting the FT with real
 CHAPTER 1. INTRODUCTION 80

signals. In particular, both projections are cos waves at 2fc with a DC offset
and thus have a DC component that will integrate to infinity.
We next evaluate S(f 6= fc ).
The two figures below fig. 1.11 and fig. 1.12 show the input signal s(t) =
cos 2πfc t + φ0 (blue), the complex exponential e−j2πf t (green) with f 6= fc , and
the product s(t)e−j2πf t (red) which is the integrand of the Fourier transform.
The integrand (red) may also be written as the sum of two complex exponentials
at f − fc and f + fc . The two figures show the same functions from different
perspectives.
Both the high frequency fc + f and the low frequency fc − f can be observed in
the integrand (red). There is no DC component in the integrand so the Fourier
integral S(f 6= fc ) = 0.
INSTRUCTOR COPY

Figure 1.11: product of cos wave and complex exponential where f 1 6= f 2

 CHAPTER 1. INTRODUCTION 81
INSTRUCTOR COPY

Figure 1.12: product of cos wave and complex exponential where f 1 6= f 2

When f 6= fc , the projections of the integrand (red) on the real and imaginary
axis (magenta) look like the sum of two sinusoids, one at high frequency and one
at low frequency, i.e. the same as the figures in the previous section interpreting
the FT with real signals. The projection of the integrand (red) on the complex
plane (magenta) is traced out by the sum of two phasors rotating at f − fc and
f + fc .
If f = fc then the complex exponential at f − fc becomes a complex constant
and there is only one complex exponential (rotating phasor) at 2fc centered at
a non-zero (DC) complex value.

1.4.2 Discrete Fourier Transform (DFT)

In a discrete-time (sampled) system where s(t) is sampled at times Ts so that sn =

s[n] = s(t = nTs ) the continuous-time Fourier transforms used here become a Discrete
Fourier transform, and the observation interval T0 = N0 Ts where N0 is the number
of samples in one frame/window/segment must contain at least one full period of the
 CHAPTER 1. INTRODUCTION 82

signal.

1.4.2.1 Sampling the Fourier transform to obtain DFT

The DFT may be interpreted is as a sampled (discrete-time) version of the continuous

time Fourier transform by writing
Z
S(f ) = s(t)e−j2πf t dt (1.8)

and making the following substitutions from continuous time to discrete-time

sampling rate fs
INSTRUCTOR COPY

sampling time Ts = 1/fs

time index n
sampling in time s(n) = s(t)|t=nTs = s(t = nTs )
observation time T0
number of samples N0 = T0 /Ts
frequency resolution f0 = fs /N0
frequency index k
sampling in frequency S(k) = S(f )|f =kf0 = S(f = kf0 )

to obtain

S[k] = S(f = kf0 )

X
N 0 −1

= s(t = nTs )e−j2πkf0 nTs

n=0
X
N 0 −1

= s(t = nTs )e−j2πkfs nTs /N0

n=0
X
N 0 −1

= s(n)e−j2πnk/N0
n=0

The DFT is a numerical way to compute a Fourier transform using N0 samples

of the time domain signal s(t) in time steps of Ts over the time span T0 = N0 Ts to
obtain N0 samples of the frequency domain signal spectrum S(f ) in frequency steps
of f0 over the frequency span fs = N0 f0 as illustrated in fig. 1.13.
 CHAPTER 1. INTRODUCTION 83

0 samples N0 samples

... xN0 −2 xN0 −1 1

x0 x1 x2 ts = fs

ts
T0

... 1
X0 X1 X2 XN0 −2 XN0 −1 f0 = T0
INSTRUCTOR COPY

f0
fs

Figure 1.13: Numerical arrays showing the mapping from N0 time samples to N0
frequency samples.

The DFT implicitly assumes that the sampled time signal is periodic with period
T0 = N0 Ts equal to the observation interval (window/frame/segment size). The N0
time samples spaced by sampling time (time resolution) Ts are repeated at intervals
T0 . In other words, the DFT assumes sn = sn±N0 . We already know that the sampled
frequency response is periodic with period fs = N0 f0 (since we go around the unit
circle). The N0 frequency samples spaced by frequency resolution f0 are repeated at
intervals fs .

Exercise 16
Prove that S[k] = S[k ± N0 ]
 CHAPTER 1. INTRODUCTION 84

Solution
Here we define exp(jθ) = ejθ

X
N 0 −1

S[k] = s(n)exp(−j2πnk/N0 )
n=0
X
N 0 −1

S[k + N0 ] = s(n)exp(−j2πn(k + N0 )/N0 )

n=0
X
N 0 −1

= s(n)exp(−j2πnk/N0 )exp(−j2πnN0 /N0 )

n=0
X
N 0 −1

= s(n)exp(−j2πnk/N0 )exp(−j2πn)
INSTRUCTOR COPY

n=0
X
N 0 −1

= s(n)exp(−j2πnk/N0 )
n=0
= Sk

since exp(−j2πn) = 1 for any integer value of n.

If we think of the signal s(t) as the impulse response of a digital filter, then the
DFT is a numerical way to compute a Fourier transform using N0 samples of the time
domain filter impulse response in time steps of Ts over the time span T0 = N0 Ts to
obtain N0 samples of the filter frequency response in frequency steps of f0 over the
frequency span fs = N0 f0 .

Similarly, the inverse DFT is a sampled version of the inverse Fourier transform
Z
s(t) = S(f )exp(+j2πf t)df (1.9)

substituting f = kf0 , t = nTs to obtain

N0 −1
1 X
s[n] = S[k]exp(+j2πkn/N0 ) (1.10)
N0 k=0
 CHAPTER 1. INTRODUCTION 85

1.4.2.2 Interpretation of DFT

In this section, we interpret the DFT formula

S[k] = S(f = kf0 )

X
N 0 −1

= s(t = nTs )e−j2πkf0 nTs

n=0
X
N 0 −1

= s(t = nTs )e−j2πkfs nTs /N0

n=0
X
N 0 −1

= s(n)e−j2πnk/N0
INSTRUCTOR COPY

n=0

The sampled signal sn is multiplied by a sampled complex exponential

en = e−j2πkf0 nTs = cos 2πkf0 nTs − j sin 2πkf0 nTs

= e−j2πnk/N0 = cos 2πnk/N0 − j sin 2πnk/N0

The sampled complex exponentials are cosine and sine waves with time samping index
n and an integer number k of cycles during the observation (frame/segment/block/window)
size N0 = T0 /Ts . In effect, the DFT S[k] is a correlation sum (over index n) of the
input signal with each of the N0 discrete sinusoids at frequencies kf0 , 0 ≤ k ≤ N0 − 1.
When the input signal sn is a sinusoid at a particular frequency k0 f0 , then S[k] will
be 1 when k = k0 and 0 otherwise.

1.4.2.3 Mirror image

The DFT of a set of N0 real samples s[n] is a set of N0 complex samples S[k].

Each complex number S[k] has a real and imaginary part, or an amplitude and
phase, so the DFT takes a block of N0 real numbers sn , n = 1, N0 and transforms
them into N0 complex numbers Sk , k = 1, N0 (fig. 1.14).

• The first set of N0 /2 numbers Sk for 0 ≤ k ≤ N/2 represent the amplitude and
phase at frequencies kf0 from 0 to (N0 /2)f0 = (N0 /2)fs /N0 = fs /2.

• The second set of N0 /2 numbers for N0 /2 ≤ k ≤ N0 represent the amplitude

and phase at frequencies kf0 from fs /2 to fs .
 CHAPTER 1. INTRODUCTION 86

• The second set is the mirror image and complex conjugate of the first set,
reflected about k = N0 /2, f = fs /2.

Time x0 x1 x2 ... xN0 −2 xN0 −1

Mirror image

   
<(X0 ) ... < X N0 −1 < X N0 −1 ... <(X0 )
INSTRUCTOR COPY

2 2

Frequency

   
=(X0 ) ... = X N0 −1 −= X N0 −1 ... −=(X0 )
2 2

0 samples N0
samples N0 samples
2

Figure 1.14: The DFT of a real signal turns N0 real time samples into N20 unique
complex frequency samples. The mirror image (blue) is the complex conjugate of the
first half.

We note that since Sk = Sk±N0 that the second set of complex numbers can be
shifted down by N0 so that the mirror image reflection is reflected about k = 0, f = 0
rather than k = N0 /2, f = fs /2.

This shifting of the second set of N0 /2 numbers by N0 results in negative frequen-

cies. The second set of N0 /2 numbers for −N0 /2 ≤ k ≤ 0 represent the amplitude
and phase at frequencies kf0 from −fs /2 to 0. The second set is the mirror image
of the first set, reflected about k = 0, f = 0. Every frequency kf0 in the range
0 ≤ f ≤ fs /2 with a non-zero amplitude |Sk | has a mirror image with the same am-
plitude |S−k | = |Sk | and opposite phase arg S−k = arg Sk in the range −fs /2 ≤ f ≤ 0.
We say that the amplitude spectrum |S[k]| is symmetrical about f = 0 and the phase
spectrum arg S(k) is antisymmetrical about f = 0¿

After the shift, each complex conjugate pair Sk + S−k represents the amplitude
and phase of a cosine wave at frequency kf0 = kfs /N0 .
 CHAPTER 1. INTRODUCTION 87

If we replace the real samples s[n] with complex samples s̃[n] in the DFT calcula-
tion, then the DFT will yield N0 unique complex frequency domain samples S̃[k]. In
this case the DFT converts N0 complex numbers s̃[n] into N0 complex numbers S̃[k].
There is no mirror image (fig. 1.15).

<(x0 ) <(x1 ) <(x2 ) ... <(xN0 −2 ) <(xN0 −1 )

Time

=(x0 ) =(x1 ) =(x2 ) ... =(xN0 −2 ) =(xN0 −1 )

INSTRUCTOR COPY

<(X0 ) <(X1 ) <(X2 ) ... <(XN0 −2 ) <(XN0 −1 )

Frequency

=(X0 ) =(X1 ) =(X2 ) ... =(XN0 −2 ) =(XN0 −1 )

0 samples N0 samples

Figure 1.15: The DFT of a complex signal turns N0 complex time samples into N0
complex frequency samples. There is not mirror image

Every frequency kf0 in the range 0 ≤ f ≤ fs /2 with a non-zero magnitude |Sk |

will in general have a different magnitude |S−k | 6= |Sk | in the range −fs /2 ≤ f ≤ 0.
We say that the spectrum S[k] is asymmetrical about f = 0.

1.4.2.4 Definition of DFT in terms of LTI frequency response

The frequency response (amplitude and phase) of an linear time-invariant (LTI) dig-
ital system with sampling rate fs is the transfer function evaluated on the unit circle

H(f ) = H(z)|z=exp(j2πf /fs ) , −fs /2 < f < fs /2 (1.11)

 CHAPTER 1. INTRODUCTION 88

For the frequency response of a filter, we use the notation H(f ), where as for the
spectrum of a signal, we use the notation S(f ). A filter has an impulse response h(t)
which may also be considered to be a signal s(t) = h(t), so that the filter frequency
response H(f ) = S(f ) is the same as the spectrum of the impulse response.

To calculate and plot H(f ) using a computer, we need to calculate H(f ) at discrete
frequencies in the range −fs /2 < f < fs /2. We choose to use N0 points to cover a
span of fs , spaced at intervals of fs /N0 = f0 .

We define the Discrete Fourier Transform (DFT) as the frequency response of an

LTI system sampled at N0 samples around the whole unit circle −fs /2 < f < fs /2
with frequency resolution (step size) f0 = fs /N0
INSTRUCTOR COPY

H[k] = H(f )|f =kf0 =kfs /N0 , 0 ≤ k ≤ N0 − 1 (1.12)

= H(z)|z=exp(j2πf /fs ) |f =kfs /N0 (1.13)
= H(z)|z=exp(j2πk/N0 ) (1.14)
X
N 0 −1

= h[n]z −n |z=exp(j2πk/N0 ) (1.15)

n=0
X
N 0 −1

= h[n]exp(−j2πkn/N0 ) (1.16)
n=0

where h[n] is the impulse response of the filter.

Thus the DFT input is the sampled impulse response of the filter, i.e. N0 time
samples h[n] spaced at time intervals Ts = 1/fs . The sampled impulse response may
be written as a vector h.

The DFT output is the sampled frequency response of the filter, i.e. N0 frequency
samples H[k]. spaced at frequency intervals f0 . The sampled frequency response may
be written as a vector H.

Similarly, the inverse DFT (IDFT) is

N0 −1
1 X
h[n] = H[k]exp(+j2πkn/N0 ) (1.17)
N0 k=0

so the IDFT input is the sampled frequency response of the filter, i.e. N0 frequency
samples H[k] and the IDFT output is the sampled impulse response of the filter, i.e.
N0 time samples h[n].
 CHAPTER 1. INTRODUCTION 89

1.4.2.5 Calculation of DFT

To evaluate the expression for the DFT, we take a block of N0 time samples h[n],
start with frequency f = kf0 = 0 with index k = 0, and compute a sum over the time
index n to obtain H[k] for k = 0, and then repeat for each index k for 0 ≤ k ≤ N0 − 1.
We can write the DFT
X
N0 −1

H[k] = h[n](WN0 )kn (1.18)

n=0
(1.19)

where we use the shorthand notation

INSTRUCTOR COPY

WN0 = exp(−j2π/N0 ) (1.20)

For example, for N0 = 4, we can write

H[k] = h[0](WN0 )0·k + h[1](WN0 )1·k + h[2](WN0 )2·k + h[3](WN0 )3·k (1.21)

and evaluate H[k] for each k in the range 0 ≤ k ≤ N0 − 1.

An equivalent way to compute the DFT is write it as a matrix multiplication of

a vector h with N0 elements with a N0 × N0 matrix WDFT to obtain an N0 element
vector H. Thus H = Wh, where the matrix for N0 = 4 is written
 
1 1 1 1
 1 W4 (W4 )2 (W4 )3 
WDFT =   1 (W4 ) (W4 )4 (W4 )6 
2
 (1.22)
1 (W4 )3 (W4 )6 (W4 )9

The IDFT h = WIDFT H, where the matrix

 
1 1 1 1

1  1 W4 (W4 )−2 (W4 )−3 
WIDFT =  (1.23)
4  1 (W4 )−2 (W4 )−4 (W4 )−6 
1 (W4 )−3 (W4 )−6 (W4 )−9

1.4.2.6 Zero-padding

If we require a frequency resolution f0 for a signal sampled at fs , we need N0 = fs /f0

time samples. Thus the sampled time signal will have a duration of N0 Ts = N0 /fs =
1/f0 = T0 . However, if the sampled time signal is of short duration less than T0 , i.e.
has N < N0 samples, then the remaining N0 − N samples are simply set to zero.
 CHAPTER 1. INTRODUCTION 90

1.4.2.7 Review

Recall the discrete Fourier transform (DFT) is defined as a Fourier transform operat-
ing on a sampled periodic signal with period N samples. The time domain signal s(t)
is sampled at a sampling rate fs = 1/Ts to make samples sn = s(t = nTs ), where Ts is
the time resolution. After the discrete Fourier transform is done on a block of N real
time domain samples sn , 0 ≤ n ≤ N − 1 to obtain the frequency domain signal S(f ),
the result is a block of N complex frequency domain samples Sk , 0 ≤ k ≤ N − 1,
where Sk = S(f = kf0 ) is the frequency resolution f0 = fs /N . The DFT is written
as follows:
S(f = kf0 ) = Sk
= ∫ Tt=0
0 =nTs
s(t)e−j2πf t dt|t=nTs ,f =kf0
INSTRUCTOR COPY

X
N −1
= s(nTs )e−j2πkf0 nTs (1.24)
n=0
X
N −1
= sn e−j2πkn/N
n=0

where Ts f0 = 1/N and N is the number of samples in both the time domain and
frequency domain. This calculation may be done in Matlab Sk = f f t(sn ) where the
blocks of N samples sn , Sk are Matlab vectors.

The block of N complex frequency domain samples is symmetrical about N/2. The
DFT takes N unique real time domain samples and produces N/2 unique complex
frequency domain samples or N unique real numbers.

Since ej2πkn/N = ej2πk(n+N )/N , the DFT may be written

N/2−1
X
Sk = sn e−j2πkn/N (1.25)
n=−(N/2

This shows that the spectrum of a real signal is symmetrical about zero frequency.

If we replace the real signal s(t) with a complex signal s̃(t) = a(t)ejφ(t) in the DFT
calculation, then the N unique complex time domain samples s(t = nTs ) = sn = h[n]
will yield N0 unique complex frequency domain samples Sk = Hk. This shows that
the spectrum of a complex signal is asymmetrical about zero frequency.
 CHAPTER 1. INTRODUCTION 91

1.5 Algorithm to generate a spectrogram

In this section, we design an algorithm to generate a spectrogram.

The first (oversimplified) algorithm is below.

1. Segment the discrete time domain signal s(n) = s(nTs ) sampled at a given
rate fs = 1/Ts samples-per-second into frames of a fixed number of samples
N , where N is typically a power of two such as 1024. Each frame is of length
N Ts = N/fs seconds.
2. Take the discrete Fourier transform (S(f ) = DF T {s(t)}) of a frame to yield
S(k), a frame of N complex samples representing the amplitude and phase
INSTRUCTOR COPY

versus discrete frequency, fk = kfs /N and frequency index k. The frequency

resolution will be f0 = fs /N Hz. The DFT calculation is a library function in
many languages including Matlab, C and Python. The DFT theory is reviewed
in section 1.4.
3. Draw one horizontal scanning line at the top of the spectrogram with the color
of each sample chosen to represent the amplitude. For example, blue for low
amplitude and red for high amplitude and all colors in between.
4. Push all other scanning lines in the spectrogram down by one time unit (one
frame time).
5. Repeat from Step 2 for the next segment (frame, window).

In this algorithm, the segments (frames, windows) do not overlap. In other words,
the hop size is equal to the window size. The result is that sharp edges between
segments result in an inaccurate representation of the amplitudes and phases of signals
with frequencies that are not at an exact multiple of the frequency resolution f0 =
fs /N .

A improved algorithm is to work with overlapping segments as in fig. 1.16 and

fig. 1.17 below, using a hop size H equal to half the window size N , i.e. H = N/2.
The window includes a raised-cosine function that emphasizes the samples near the
center of the window and suppresses the samples near the edges of the window. The
N -sample raised-cosine window is written
wn = 0.5(1 − cos(2πn/N )), 0 ≤ n ≤ N − 1
The DFT algorithm in section 1.4 is adapted to include the window shape function
w. This algorithm will yield a more accurate representation of signals that are not at
an exact multiple of the frequency resolution f0 = fs /N .
 CHAPTER 1. INTRODUCTION 92

Figure 1.16: creating spectrogram

INSTRUCTOR COPY
 CHAPTER 1. INTRODUCTION 93
INSTRUCTOR COPY

Figure 1.17: Short-time Fourier transform

We observe that the FFT of the signal comprises N complex numbers, but only
N/2 of the numbers are unique. The other N/2 numbers are the mirror image, as
further explained in section 1.4.2.3.

Each scanning line of the spectrogram color-codes the amplitude |Sk | for 0 ≤ k ≤
N.
 CHAPTER 1. INTRODUCTION 94

Most of the software and apps will display the spectrogram amplitude |Sk | or
power |Sk |2 or power in dB 20logSk at discrete frequencies fk = kfs /N over the
frequency range 0 ≤ fk ≤ fs /2 or 0 ≤ k ≤ N/2 and will omit the mirror image and
also omit the phase.

For each segment (window, frame) we assign a time index m so that the FFT
components Sk (with frequency index k) for the m-th segment are written Sm,k . The
spectrogram is then a two-dimensional plot of color coded values |Sm,k | with time
index m and frequency index k. Thus for each point in the m, k time-frequency grid,
there is a color-coded amplitude.

fig. 1.18 shows the time-frequency grid where the hop size is 8 samples.

The spectrogram function is implemented with selectable parameters hop size and
INSTRUCTOR COPY

window size and type is documented in https://www.mathworks.com/help/signal/

ref/spectrogram.html and example applications are in https://www.mathworks.
com/help/signal/ug/practical-introduction-to-time-frequency-analysis.html
 CHAPTER 1. INTRODUCTION 95
INSTRUCTOR COPY

Figure 1.18: Time-frequency grid

 CHAPTER 1. INTRODUCTION 96

Exercise 17
Write the algorithm for creating a spectrogram waterfall showing the signal
samples of s̃(t) and S̃(f ) = F{s̃(t)} in discrete-time notation where s̃n = s̃(t =
nTs ), Ts = 1/fs , S̃k = S̃(f = kf0 ) . Recall the discrete Fourier transform is
defined as a Fourier transform operating on a sampled periodic signal:
Z T0 =nTs
S̃k = s̃(t)e−j2πf t dt|t=nTs ,f =kf0
t=0
X
N −1
= s̃n e−j2πkf0 nTs
n=0
X
N −1
= s̃n e−j2πnk/N
INSTRUCTOR COPY

n=0

where Ts f0 = 1/N . The values of k range from 0 to N − 1 representing the

evenly spaced frequency bins around the unit circle. When the hop size is half
the window size and the raised-cosine window is used, then we combine the
window function with the DFT/FFT formula (called the Short time Fourier
Transform or STFT), and write for the m-th window (m-th scanning line)
∞
X
S̃m,k = s̃n w(n − m)e−j2πnk/N
n=−∞

The summation limits are changed to ±∞ but the actual sum is finite since
the window has a finite number of non-zero samples within the window size N .

The values of S̃m,k are called the time-frequency coefficients with time index m
and frequency index k. The spectrogram color-codes the magnitude |S̃m,k | of the
time-frequency coefficients.

1.5.1 Spectrogram examples

In this section we show spectrogram examples for radio signals received by an antenna
as opposed to human voice audio signals received by a microphone. The frequency
axis of a spectrogram displaying radio signals is typically centered around some radio
frequency in MHz or GHz, whereas The frequency axis of a spectrogram displaying
audio signals will typically be in the range up to 20 KHz.

Figure 1.19 shows an example snapshot in time of a spectrogram or moving spec-

tral waterfall display. The x axis is frequency, the y axis is time (most recent at the
 CHAPTER 1. INTRODUCTION 97
INSTRUCTOR COPY

Figure 1.20: Waterfall spectrum of morse code

top, time increasing downwards). On the left of the display, the blue line represents
a sine wave at a fixed frequency. The green blocks represent radio signal modulated
with information, and thus occupy a finite non-zero bandwidth. Note that the radio
signals on a given frequency start and stop at different times.

Figure 1.19: Waterfall spectrogram with frequency on x axis, time on y axis, ampli-
tude color coded to display z axis.

Another example of a waterfall spectrogram is shown in fig. 1.20. In this example,

all the signals are morse code, i.e. sine waves switched on and off. The amplitude is
grey scale coded to display the z axis amplitude, a tune-able filter shown in green is
used to select desired signal.
 CHAPTER 1. INTRODUCTION 98

An example of a combination spectrum and spectral waterfall is shown in fig. 1.21.

The spectrum is the output of a digital spectrum analyzer and is valid in real time.
The waterfall shows the history of the spectrum with frequency on the horizontal
axis, time on the vertical axis and amplitude represented by color. In fig. 1.21, there
are two signals present.
INSTRUCTOR COPY

Figure 1.21: Combined spectrum and spectral waterfall

Other spectrogram examples appear throughout the text.

Live real-time spectral waterfalls showing radio signals in real time are available
at these links: http://www.websdr.org/ and http://kiwisdr.com/public/

1.5.2 The need for complex signals

In the spectrogram, the amplitude and phase are calculated and displayed at the
discrete frequencies fk = kfs /N and the discrete times with index m.
 CHAPTER 1. INTRODUCTION 99

1.5.2.1 Zooming into the spectrogram

Assume that for a moment, the frequency of the whistle is held constant at a value
f1 = k1 f0 , so that all the values of Xk will be zero for every k except for k = k1 and
k = −k1 (or k = N − k1 ). The spectrogram will show two parallel lines symmetrical
about zero. 1

Now consider zooming into the spectrogram to look closely at a narrower band of
frequencies. For example, we may want to observe the variations in frequency and/or
amplitude of the whistle that represent a message.

By zooming in, we can see only the whistle at k = k1 and we no longer can see
the mirror image whistle at k = −k1 .
INSTRUCTOR COPY

The spectrogram will show only the color-coded values of the power |Xk |2 at k = k1
to represent the whistle. Recall that Xk is a complex number with amplitude and
phase, i.e.

Xk = ak ejφk (1.26)

so that at k = k1 ,

Xk1 = ak1 ejφk1 (1.27)

|Xk1 |2 = a2k1 (1.28)

To write a mathematical expression for the whistle signal in the time domain, we
take the inverse Fourier transform to obtain

x(n) = F −1 {X(k)}
X
N −1
= Xk ej2πnk/N
k=0

For the whistle signal, k = k1 6= 0 and all other k are zero. In particular, by
zooming in we cannot see the mirror signal at k = −k1 (or k = N − k1 ). We find that
x(n) is a complex signal with both amplitude and phase. The signal is written Thus
1
However, as mentioned above, most of the software and apps will display the spectrogram at k1
and will omit the mirror image at −k1 .
 CHAPTER 1. INTRODUCTION 100

x(n) = F −1 {X(k)}
X
N −1
= X(k)ej2πnk1 /N
k=0

= Xk1 ej2πnk1 /N
= ak1 ejφk1 ej2πnk1 /N
= ak1 ej2πnk1 /N +φk1
= ak1 cos(2πnk1 /N + φk1 ) + jak1 sin(2πnk1 /N + φk1 )

The result is a sampled complex exponential which is the sum of a real cosine
INSTRUCTOR COPY

wave plus an imaginary sine wave.

Thus a signal observed on the waterfall at a particular f1 = k1 f0 must be a

complex signal. The combination of that signal plus its mirror image at −f1 = −k1 f0
is a real signal.

1.5.2.2 Instantaneous amplitude and known phase

Consider a discrete-time (sampled) sine wave as in fig. 1.1. As a real signal this
is represented by a sequence of real numbers which on their own carry only the
information of the amplitude of the samples at the sample times. It is not obvious
how to obtain the amplitude, phase and frequency of this sine wave from the samples,
and it is not possible to obtain the amplitude and phase from a single sample.

Consider the following array of values to match the figure:

x[n] = 0.00, 0.71, 1.00, 0.71, 0.00, -0.71, -1.00, -0.71, 0.00
Take the second sample, a value of 0.71, and you know neither the amplitude of the
sine wave nor the phase at that time. It is an identical number to the fourth sample!
In order to extract the amplitude and phase of the sine wave you must look at multiple
samples so that each has context. However given a complex sequence for the same
sine wave:
x[n] = 0.00+1.00j, 0.71+0.71j, 1.00+0.00j, 0.71-0.71j,
0.00-1.00j, -0.71-0.71j, -1.00+0.00j, -0.71-0.71j, 0.00+1.00j
the second sample tells you that the amplitude is 1 and the phase is π/4.

This section motivates why we need to introduce complex signals in chapter 2.

 CHAPTER 1. INTRODUCTION 101

1.6 Messages

In this section, we write expressions for general messages of signals such as the human
voice.

1.6.1 Discrete-time messages m(t) as sum of cosine waves

In this section we show the details of how a discrete-time message divided into (pos-
sibly overlapping) frames is made up of the sum of cosine waves that will be different
from one frame to the next. We will use the theory of the Discrete Fourier Transform
(DFT) which is reviewed in the section 1.4.
INSTRUCTOR COPY

For any real message m(t) in a discrete-time systems where mk = m(t = k/fs ) is
sampled at rate fs , each N -sample frame has N/2 discrete frequencies in m(t). This is
because for each frame, the FFT output is N complex samples spaced f0 = fs /N apart
from −fs 2 to fs /2. Since the message is real, the complex samples are in complex
conjugate pairs thus making N/2 cosine waves at frequencies nf0 for 0 ≤ k ≤ N/2.
Thus in a discrete-time system, we can write each frame in the time domain as the
IFFT of the FFT coefficients, see section 1.4.2.1
N0 −1
1 X
m[n] = M [k]e+j2πnk/N0 (1.29)
N0 r=0
so the IDFT input is N0 frequency samples M [k] and the IDFT output is N0 time
samples m[k]. We change the summation limits so the FFT coefficients H[r] are
symmetrical about r = 0 and write N0 = N
N/2−1
1 X
m[n] = M [k]e+j2πnk/N (1.30)
N0
k=−N/2

Combining the complex conjugate pairs using ejθ + e−jθ = 2 cos θ and replacing h
with m
N/2−1
1 X
m[n] = |M [k]|2 cos(2πkn/N + arg M [k]) (1.31)
N0 k=0
Recall that since f0 ts = f0 /fs = 1/N , we can write 2πkn/N = 2πkf0 nts so the
expression for the time domain message m[n] for one frame can be seen as a sum
of cosine waves at discrete frequencies f = kf0 with amplitudes M (k) and phases
arg M [k] sampled at t = nts .
N/2−1
1 X
m[n] = |M [k]|2 cos(2πkf0 nts + arg M [k]) (1.32)
N0 k=0
 CHAPTER 1. INTRODUCTION 102

We assume that the values |M [k]| and arg M [k] at each frequency fk are constant
during each window but change from one window to the next. Thus using a window
(frame) index r, we write
N/2−1
1 X
m[r, n] = |M [r, k]|2 cos(2πkf0 nts + arg M [r, k]) (1.33)
N0 k=0

The values of M (r, k) = |M (r, k|ej arg M [r,k] are called the time-frequency coefficients
with time index r and frequency index k. The spectrogram color-codes the magnitude
|M (r, k)| of the time-frequency coefficients.

1.6.2 Message types, analog and digital

INSTRUCTOR COPY

In this section we observe that message may be analog (e.g. voice, music) or digital
(sequence of bits). The message type is not to be confused with a continuous-time or
discrete-time signal.

message/signal analog digital

There are 4 possibilities continuous-time
discrete-time

A real signal transmitted over a real physical channel such as optical fiber, coaxial
cable, or free space will be continuous-time. A signal stored in computer memory will
be discrete-time. The contents of the computer memory may be transmitted over a
real physical channel by first converting the bits into a continuous-time signal.

When we study systems that transmit messages over a distance, we often choose
a simple analog message
m(t) = Am cos(2πfm t)
where fm is the modulation/message frequency and is usually on the order of Hz or
kHz. We can also choose a simple digital message with an 10101010... alternating
bit sequence at a bit rate 2fm . For this digital message, the 1 bits are the positive
half of a cosine wave at frequency fm and the 0 bits are the negative half of the same
cosine wave. This message is called a single frequency or single tone message. In
practice, we wish to transmit a more interesting message that contains more than a
single tone at a single frequency.
 CHAPTER 1. INTRODUCTION 103

1.6.2.1 Analog message (voice or music)

A general analog message (voice or music) would be represented by a sum of cosine

waves with different amplitudes Ak = M [r] and phases ψr at each frequency fr as
shown in section 1.6.1
N/2−1
1 X
m[n] = |M [k]| cos(2πkf0 nts + arg M [k]) (1.34)
N0 k=0

We assume that the values Ak = M [k] and ψk = arg M [k] at each frequency fk are
constant during each window but change from one window to the next. Thus using
a window/frame/block index r we write
INSTRUCTOR COPY

N/2−1
1 X
m[r, n] = |M [r, k]| cos(2πkf0 nts + arg M [r, k]) (1.35)
N0 k=0

In later sections, we will use the special cases of a single or dual tone message where
only one or two values of M [r, k] are non-zero, so that only one or two cosine waves
are present in the message.

1.6.2.2 Digital message, a sequence of bits

An analog message can also represent a digital symbol sequence ak by writing

X
m(t) = ak p(t − kT ) (1.36)
k

where p(t) is a pulse that spans a finite time period, ak may be binary symbol ±1 (to
represent binary 1 or 0) or multilevel (e.g. ±1, ±3 to represent 00, 01, 10, 11) and T
is the symbol time.
 CHAPTER 1. INTRODUCTION 104

We can have two such sequences:

X
mI (t) = ak p(t − kT ) = I(t)
k
X (1.37)
mQ (t) = bk p(t − kT ) = Q(t)
k

Thus we can write the complex baseband signal as

X
I(t) + jQ(t) = (ak + jbk )p(t − kT ) (1.38)
k

where
INSTRUCTOR COPY

ck = ak + jbk = rk ejφk (1.39)

is a complex data symbol. If both ak , bk are binary, then ck represents 2 bits of
information at each time t = kT If both ak , bk are multilevel ±1, ±3, then ck represents
4 bits of information at each time t = kT .

The pulse shape p(t) can have shapes other than rectangular, and can be longer
than T . Other pulse shapes are described in a later chapter.

1.7 Chapter review questions

Q 1.1 Write the algorithm for creating a spectrogram waterfall showing the signal
samples s̃(t) and S̃(f ) = F F T {s̃(t)} in discrete-time notation where
1
s̃n = s̃ (t = nTs ) , Ts = , S̃k = S̃ (f = kf0 ) .
fs
Recall the discrete Fourier transform is defined as a Fourier transform operating on
a sampled periodic signal with period T0 and sample rate fs = T1s and there are N
samples per period
Z T0 =nTs
S̃k = s̃(t)e−j2πf t dt|t=nTs ,f =kf0
t=0
X
N −1
= s̃n e−j2πkf0 nTs
n=0
X
N −1
nk
= s̃n e−j2π N
n=0
 CHAPTER 1. INTRODUCTION 105

Ts f0 1
Ts f0 = = =
T0 fs N
Assume the display is 1024x1024 pixels.

Solution
Step 1 : take block of N = 1024 complex time domain samples

s̃n = s̃(t = nTs ), n = 1, . . . , N

Step 2 : do FFT to obtain a block of frequency domain samples

S̃k = S̃ (f = kf0 ) , k = 1, . . . , N
INSTRUCTOR COPY

Step 3 : draw a row of N pixels in a horizontal line at the top of the screen,
where the color (or brightness) of each pixel is chosen depending on the
magnitude |S̃k |

Step 4 : move this row of N pixels vertically down by 1 pixel

Step 5 : Go to Step 1
 Chapter 2

IQ Signals
INSTRUCTOR COPY

2.1 Carrier waves, amplitude and phase in the time

domain

Learning objectives

In this section, we learn the mathematics of a general communications signal wave-

form comprising a carrier wave whose amplitude and phase is modified (modulated)
in step with a message signal. In this section, we describe how the communications
signal may be written as the sum of two complex-valued waveforms resulting in a
real-valued waveform.

Introduction

Telecommunications is defined by Webster’s dictionary as ”the science and technology

of communications at a distance by electronic transmission”.

A communications signal carries a message over a distance from point A to point

B using a carrier wave.

Radio (or optical) waves are used to carry a message over a distance determined
by the link budget. The radio waves travel through a channel which may be free
space (wireless) or via transmission lines (e.g. waveguide, optical fiber, microstrip).
The radio wave (called a carrier wave) c(t) with a defined frequency is ”modulated”
(modified) by the message signal m(t). In other words the amplitude and/or phase

106
 CHAPTER 2. IQ SIGNALS 107

of the carrier wave is modified so as to include the information stored within the
message.

The modified (modulated) carrier wave is called a communications signal s(t) and
has 3 main attributes: frequency (fc ), amplitude a(t) and phase φ(t). Note that the
amplitude and phase will in general vary with time.

We can write the communications signal1

s(t) = a(t) cos (2πfc t + φ(t)) (2.1)

A communications system transmits a message m(t) by first modulating the car-

rier wave c(t) with the message m(t) to create the communications signal s(t) and
INSTRUCTOR COPY

transmits s(t) as a radio (or optical) wave. s(t) is degraded by noise and interference
on the channel. The system receives the noisy and distorted s(t) and extracts the
message m(t) by a process called demodulation.

We wish to find a mathematical way to describe s(t) and define a relationship

between the amplitude a(t), φ(t) and the message m(t).

The mathematical description of s(t) and in particular the message (information)

contained in the amplitude and phase can take several forms. We choose to describe
s(t) using complex signals based on Euler’s equation ejθ = cosθ + jsinθ.

The reason for this choice is that when viewing a spectrogram (spectral waterfall)
and zooming in on one particular signal at one particular frequency, that signal must
be complex, as explained in a previous section

For this reason, complex signals are used in software-defined radios. Many of the
functions of a communications systems that manipulate the signal, such as frequency-
shifting, filtering, modulation and demodulation, are complex functions (complex
arithmetic) implemented in software.

2.1.1 Carrier wave

We begin by considering the carrier wave, which is a ”pure” sinewave, such as would
be emitted by a laser that can travel down an optical fiber or through free space to
a receiver that absorbs the wave. It contains one and only one frequency and has
1
Another useful reference is
https://www.ni.com/en-ca/innovations/videos/07/i-q-data--plain-and-simple.html
 CHAPTER 2. IQ SIGNALS 108

y
π
1 2
π 1
3π
4 4

0.5

π φ 2π
1 x π π π
4 2
3π 5π 3π 7π 2π
4 8 2 8
−0.5
5π 7π
8 8
π −1
INSTRUCTOR COPY

Figure 2.1: A cosine wave being created by projecting a moving point on a unit circle.

constant amplitude and phase that do not change with time. We will call this pure
sine wave c(t).

2.1.1.1 real carrier waves

We can describe the carrier wave mathematically as c(t) = cos θ(t) = cos 2πfc t or
c(t) = sin θ(t) = sin 2πfc t where fc is the frequency of the carrier wave and θ(t) =
2πfc t.

The carrier wave is a real (not complex) waveform that can be represented as a
real function of time and as a time-varying analog (real) voltage and plotted on an
oscilloscope and connected to a transmission line (e.g. coaxial cable, fiber, waveguide,
microstrip) to carry it from one point to another, or connected to an antenna to radiate
it into space.

A cosine or sine wave is obtained from the projection of a point moving to trace
out a unit circle.

Note that we get the same cosine wave regardless of the direction of rotation, but
we get the opposite polarity sine wave when we reverse the direction of rotation.

If the point completes one full rotation of the circle (360 degrees or 2π radians)
in one second, then the frequency of the sine wave is one cycle per second or 1 Hz or
 CHAPTER 2. IQ SIGNALS 109

2π radians per second.

In general, we can write the instantaneous frequency fi (t) of the signal is related
to the phase θ(t) via

1 dθ(t)
fi (t) = (2.2)
2π dt

When θ(t) = 2πfc t, then

1
fi (t) = 2πfc = fc (2.3)
2π

If the point completes one full rotation of the circle (360 degrees or 2π radians)
INSTRUCTOR COPY

in 1/fc seconds, then the frequency of the sine wave is fc cycles per second or fc Hz
or 2πfc radians per second.

2.1.1.2 complex carrier waves

ejθ +e−jθ
Using Euler’s equation ejθ = cosθ + jsinθ and also cosθ = 2
and assuming
θ = θ(t) = 2πfc t, we can write
c(t) = (ejθ(t) + e−jθ(t) )/2 (2.4)
= (ej2πfc t + e−j2πfc t )/2 (2.5)
(2.6)
Thus c(t) can be represented mathematically as the sum of two complex exponentials,
one positive and one negative. There is a scaling factor of 2 that is not important for
the present discussion.

Euler’s equation introduces the idea of complex numbers (z = ejθ = cos θ + j sin θ)
that can be written in polar or rectangular form. We also introduce complex functions
of time (z(t) = ej2πfc t = cos2πfc t + jsin2πfc t) written in polar or rectangular form.

Complex numbers are used in many contexts including communications signals

and digital signal processing (transfer function of digital filters).

More generally we can write a complex number z as

z = aejθ
= a cos θ + ja sin θ
(2.7)
= x + jy
= <{z} + j={z}
 CHAPTER 2. IQ SIGNALS 110

<(z(t)) |z(t)|

a a

T T T T
2 2
−a −a
=(z(t)) φ(z(t))
π
a 2

T T T T
2 2
−a − π2
INSTRUCTOR COPY

Figure 2.2: Two methods for plotting a complex function of time: Real/Imaginary
(left), Magnitude/Phase (right). The symbol T represents the period of the waveform
where T = f1 .

where a = |z| is the magnitude, θ = arg{z} is the phase, x = <{z} is the real part,
y = ={z} is the imaginary part. We write a complex function of time z(t) as

z(t) = a(t)ejθ(t)
= a(t) cos θ(t) + ja(t) sin θ(t)
(2.8)
= x(t) + jy(t)
= <{z(t)} + j={z(t)}
where a(t) = |z(t)| is the magnitude, θ(t) = arg{z(t)} is the phase, x(t) = <{z(t)} is
the real part, y(t) = ={z(t)} is the imaginary part.

How do we plot a complex function of time? One method is to plot both the
amplitude and phase as two separate 2-D plots, or plot the real part and the imaginary
part as two separate 2-D plots. Both of these methods are illustrated in fig. 2.2 for
the complex function of time ej2πt .

Another method is to visualize the positive complex exponential z(t) = ej2πfc t in

3 dimensions: real axis (real), imaginary axis (imag) and time. A 3-D plot can be
made in Matlab using the command plot3(t, x, y). This 3-D curve is called a helix.
z(t) = ej2πfc t is a complex waveform that cannot be viewed on an 2-D oscilloscope
screen as a real voltage versus time, and cannot be carried over a wire or transmission
line or radiated into space via an antenna. However complex functions of time z(t)
are very useful for representing complex communications signals in software.
 CHAPTER 2. IQ SIGNALS 111
INSTRUCTOR COPY

1
=(z(t))

−1 1
0 0
0.5 1 1.5 −1 <(z(t))
t

Figure 2.3: Helix with constant frequency, amplitude and phase.

 CHAPTER 2. IQ SIGNALS 112

The end view of the helix (projection on the real-imag plane) shows the circle.
The bottom (or top) view (projection on the real-time plane) of the helix shows the
real part of the helix(a cosine wave) and the side view (projection on the imag-time
plane) of the helix shows the imaginary part of the helix (a sine wave). Either the real
or the imaginary parts are real waveforms that can be viewed on an oscilloscope. The
end view is what would be seen on an x-y oscilloscope with the real and imaginary
signals connected to the x and y inputs of the oscilloscope respectively. Since the
cosine and sine wave are 90 degrees out of phase, the x-y plot is a circle.

The helix represents the carrier wave with constant amplitude and phase. In the
next section we see what happens when the amplitude and phase are not constant.

In this figure, the helix arises from a counter-clockwise rotation as time increases.
The positive complex exponential ej2πfc t is defined to have a positive frequency.
INSTRUCTOR COPY

We can visualize the negative complex exponential e−j2πfc t as a helix in the op-
posite direction. The helix arises from a clockwise rotation as time increases. The
negative complex exponential e−j2πfc t is defined to have a negative frequency.
j2πfc t −j2πfc t
The sum of the two complex exponentials c(t) = e +e
2
(helixes in opposite
directions) is a cosine wave on the real-time plane, and zero in the imag-time plane.

We can also write c(t) = [c+ (t)+c− (t)]/2 where c+ (t) = ej2πfc t and c− (t) = e−j2πfc t .
This notation shows explicitly that the real carrier wave c(t) = cos(2πfc t) is made
up of a positive frequency component plus a negative frequency component.

We can also write

c+ (t) = ej2πfc t = cos 2πfc t + j sin 2πfc t (2.9)

c− (t) = e−j2πfc t = cos 2πfc t − j sin 2πfc t (2.10)
c(t) = [c+ (t) + c− (t)]/2 = cos 2πfc t (2.11)

It is mathematically equivalent to write the real carrier wave as

c(t) = <{c+ (t)} = <{ej2πfc t } = cos 2πfc t. (2.12)

c(t) = <{c− (t)} = <{e−j2πfc t } = cos 2πfc t. (2.13)

This method of writing c(t) by taking the real part of a complex signal appears in
many textbooks and articles, but is potentially confusing, since it appears that some-
thing (the imaginary part) is being removed from the positive complex exponential
and then the mystery is what happened to that imaginary part and why we don’t
need it anymore. Instead, it is more clear and transparent to think of adding the
 CHAPTER 2. IQ SIGNALS 113

negative complex exponential to the positive complex exponential (adding the two
helixes) to get the same result, and noticing that the imaginary part is cancelled out
by the addition. Thus the real carrier wave c(t) is obtained by adding the positive
and negative frequency components. Obtaining c(t) = <c+ (t) is a mathematically
convenient shortcut that hides the real meaning of what is going on.

This completes the discussion of the carrier wave.

Another perspective on complex signals is at https://wiki.gnuradio.org/index.

php/IQ_Complex_Tutorial

2.1.2 Modulated carrier wave with time-varying amplitude

INSTRUCTOR COPY

and phase

The communications signal s(t) is a modified (or modulated) carrier wave and has 3
main attributes: frequency, amplitude and phase. We wish to find a mathematical
way to describe s(t).
ejθ +e−jθ
We will again use Euler’s equation ejθ = cos θ + j sin θ and also cosθ = 2
as
a starting point.

Recall the carrier wave with positive and negative frequencies is written c(t) =
ej2πfc t + e−j2πfc t . The carrier wave may also be written in a more general form
a(t)ejθ(t) + a(t)e−jθ(t)
c(t) =
2 (2.14)
= a(t) cos θ(t)
This general form expression for a carrier wave is valid only if a(t) = Ac = constant
and θ(t) = 2πfc t.

To write the communications signal s(t) in which the amplitude, frequency and/or
phase are modified to carry a message, we modify the carrier wave by changing its
amplitude and phase.

We can write the same expression as above for a modulated carrier wave s(t)
a(t)ejθ(t) + a(t)e−jθ(t)
s(t) =
2 (2.15)
= a(t) cos θ(t)
However, for a modulated carrier wave, a(t) represents the amplitude of the signal
after modulation and θ(t) is the phase of the carrier wave. a(t) and θ(t) are the
information-bearing (message) part of the signal.
 CHAPTER 2. IQ SIGNALS 114

The message is contained within a(t) and θ(t) in a manner to be described later.
The message may be analog or digital. Note that if there is no message, then a(t) =
Ac = constant and θ(t) = 2πfc t and s(t) = c(t) is a (pure, unmodulated) carrier
wave.

If there is a message, then s(t) 6= c(t) is a modulated carrier wave for which either
or both of a(t), θ(t) will be different from the above values for an unmodulated carrier
wave.2

In many books and articles the general form of a radio signal (or any communi-
cations signal) is written as the real part of a positive complex exponential

s(t) = Re{a(t)ejθ(t) } = a(t)cosθ(t) (2.16)

INSTRUCTOR COPY

This is mathematically identical to taking the projection of the helix in the figure
above.

This method of writing s(t) by taking the real part of a complex signal appears in
many textbooks and articles, but is potentially confusing, since it appears that some-
thing (the imaginary part) is being removed from the positive complex exponential
and then the mystery is what happened to that imaginary part and why we don’t
need it anymore. Instead, it is more clear and transparent to think of adding the
negative complex exponential to the positive complex exponential (adding the two
helixes) to get the same result, and noticing that the imaginary part is cancelled out
by the addition. Thus the real carrier wave c(t) is obtained by adding the positive
and negative frequency components. Obtaining s(t) = <s+ (t) is a mathematically
convenient shortcut that hides the real meaning of what is going on.

Exercise 18
Write out the algebra showing how s(t) can be obtained by adding the positive
(s+ (t)) and negative (s− (t)) frequency components.

2
Examples of expressions for a(t), θ(t) for a modulated carrier wave will be described in a later
section. If a(t) is non-constant and θ(t) = 2πfc t then s(t) is a form of amplitude modulation. A
simple case is where a(t) is either 1 or 0 depending on whether a 1 or a 0 bit is sent.
 CHAPTER 2. IQ SIGNALS 115

Solution

a(t)ejθ(t) + a(t)e−jθ(t)
s(t) =
2
a(t) cos θ(t) a(t) sin θ(t) a(t) cos θ(t) a(t) sin θ(t) (2.17)
= +j + −j
2 2 2 2
= a(t) cos θ(t)

Note the scaling factor of 2 appears here. This factor may be important in some
circumstances where we need to know the signal amplitude and not so important in
others where the signal amplitude would be changed by gain or attenuation stages in
the system.
INSTRUCTOR COPY

Note that the instantaneous frequency fi (T ) of the signal is related to the phase
θ(t) via

1 dθ(t)
fi (T ) = (2.18)
2π dt

For the positive exponential, the frequency is positive and visa versa.

If dθ = 2π and dt = 1 second, then the frequency is one cycle per second or 1 Hz.
Thus 2π radians in one second is one cycle per second or 1 Hz. One cycle per second
is one cycle of a cos or sin wave per second.

In the special case where there is no modulation applied to the carrier signal
(ie: no message sent), then we say the carrier wave is “unmodulated” and we write
s(t) = c(t) is a carrier wave oscillating at a frequency of fc and scaled by a constant
carrier amplitude coefficient a(t) = Ac .

In this special case, the amplitude a(t) = Ac is a constant, and the phase θ(t)
increases linearly with time such that θ(t) = 2πfc t.

The instantaneous frequency is exactly the carrier frequency (exercise)

1 dθ(t) 1 d2πfc t 1
fi (T ) = = = 2πfc = fc
2π dt 2π dt 2π
as expected.

In this case, we write a(t) = Ac (constant independent of time), θ(t) = 2πfc t

(linear increase of phase with time, 2π radians every 1/fc seconds.
 CHAPTER 2. IQ SIGNALS 116

φ = constant φ(t) = func(t)

1 1
t t
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
−1 −1

θ(t) = 2πf t + φ θ(t) = 2πf t + φ(t)

2πf 2πf
πf πf
INSTRUCTOR COPY

t t
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
sin θ(t) = sin(2πf t + constant) sin θ(t) = sin(2πf t + func(t))

1 1

t t
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
-1 -1

Figure 2.4: The impact of constant φ vs. time-varying φ(t) on communications signal.

Therefore c(t) = Ac cos(2πfc t) = Ac cos(ωt).

As mentioned in the previous section, this signal is called the carrier wave c(t)
and can also be visualized as a phasor with angular frequency ωc = 2πfc and with
period T = 1/fc .

In general, when a message is sent and the carrier wave is modulated, the am-
plitude a(t) of the carrier wave may be time varying, and the phase of the carrier
θ(t) may have a time varying phase component φ(t) that is added to the linear phase
2πfc t. This is illustrated in fig. 2.4 and described mathematically as

s(t) = Re{a(t)ejθ(t) } = a(t) cos θ(t)

θ(t) = 2πfc t + φ(t) (2.19)
s(t) = a(t) cos(2πfc t + φ(t))

The radio signal s(t) is a cosine wave at frequency fc with time-varying amplitude
 CHAPTER 2. IQ SIGNALS 117

a(t) and phase φ(t). It is useful to write the radio signal as the real part of a complex
waveform or (equivalently) as the sum of a positive frequency complex waveform s+ (t)
and a negative frequency complex waveform s− (t).

s(t) = Re{a(t)ejφ(t) ej2πfc t }

a(t)ejφ(t) ej2πfc t a(t)e−jφ(t) e−j2πfc t
= + (2.20)
2 2
s+ (t) + s− (t)
=
2
where
INSTRUCTOR COPY

a(t) j[2πfc t+φ(t)]

s+ (t) = e
2 (2.21)
a(t) a(t)
= cos[2πfc t + φ(t)] + j sin[2πfc t + φ(t)]
2 2

a(t) −j[2πfc t+φ(t)]

s− (t) = e
2 (2.22)
a(t) a(t)
= cos[2πfc t + φ(t)] − j sin[2πfc t + φ(t)]
2 2

When we add the positive and negative frequency signals, the imaginary parts
cancel out to yield the real signal

s+ (t) + s− (t)
s(t) = = a(t) cos(2πfc t + φ(t)) (2.23)
2

In the equations above, the complex exponentials are written in polar form showing
amplitude and phase. The radio signal may also be written

s(t) = Re{s+ (t)} = Re{s− (t)}, (2.24)

where the positive frequency signal s+ (t) = sa (t) is also called the analytic signal.

There is also a negative frequency analytic signal s− (t) but this definition is rarely
used. However, the best way to think is that the real signal

s(t) = [s+ (t) + s− (t)]/2

 CHAPTER 2. IQ SIGNALS 118

is the sum of the positive and negative frequency signal3 , even though it is mathe-
matically equivalent to s(t) = Re{s+ (t)} == Re{s− (t)}. Throughout this book we
will commonly use this equation in the form

2s(t) = s+ (t) + s− (t) (2.25)

2.2 Complex envelope (complex baseband)

We define the complex envelope or complex baseband component of the communi-

cations signal to be s̃(t) = a(t)ejφ(t) = I(t) + jQ(t) The complex conjugate of the
complex envelope is s̃∗ (t) = a(t)e−jφ(t) = I(t) − jQ(t).
INSTRUCTOR COPY

The complex envelope is the information-bearing part of the signal without the
carrier. In other words, the complex baseband signal represents the message (or
several messages), which may be analog or digital. The complex baseband signal may
be viewed in the time domain, frequency domain, and as a time-frequency spectrogram
waterfall.

The communications signal is then written

s(t) = <{s̃(t)ej2πfc t }
= a(t) cos[2πfc t + φ(t)]
s+ (t) + s− (t)
= (2.26)
2
a(t) jφ(t) j2πfc t a(t) −jφ(t) −j2πfc t
= e e + e e
2 2
= 0.5s̃(t)ej2πfc t + 0.5s̃∗ (t)e−j2πfc t
3
The positive frequency representation of a sinusoid is also referred to as the analytic signal
(section 2.6). The analytic signal is obtained by expressing the sinusoid in terms of complex-
exponentials, discarding the negative frequency component, and doubling the positive frequency
component. Thus if

s(t) = a(t) cos θ(t)

= a(t)ejθ(t) /2 + a(t)e−jθ(t) /2

then s+ (t) = a(t)ejθ(t) where θ(t) = 2πfc t + φ(t).

 CHAPTER 2. IQ SIGNALS 119

where

s̃(t) = a(t)ejφ(t) (2.27)

∗ −jφ(t)
s̃ (t) = a(t)e (2.28)
s+ (t) = s̃(t)ej2πfc t (2.29)
s− (t) = s̃∗ (t)e−j2πfc t (2.30)

The real communications signal is written in terms of the complex envelope as

s(t) = 0.5s̃(t)ej2πfc t + 0.5s̃∗ (t)e−j2πfc t

INSTRUCTOR COPY

= Re{s̃(t)ej2πfc t } (2.31)
= Re{s̃∗ (t)e−j2πfc t }

In all 3 forms of s(t) above, the complex envelope is multiplied by a complex

exponential carrier wave at fc .

The multiplication of the complex envelope by an exponential carrier is imple-

mented in a so-called mixer. The mixer is another name for a multiplier 4 . The mixer
in this case is a complex mixer (complex multiplier) that is implemented in software.
Thus a mixer implemented in software is simply a complex multiply operator. We
will see later how to implement mixers in hardware.

To encode a message m(t) on the carrier wave we vary a(t) and/or φ(t) in step
with the message m(t). Thus a(t), φ(t) are specified as a function of the message m(t).
These functions will be specified later when we discuss specific modulation types.

We can write the radio signal in a form where the complex envelope is separated
into its real and imaginary parts s̃(t) = a(t)ejφ(t) = I(t) + jQ(t)

Exercise 19
Write a(t) and φ(t) as a function of I(t), Q(t).

Solution
Using the identity

cos(A + B) = cos A cos B − sin A sin B

4
in audio, a mixer is an adder
 CHAPTER 2. IQ SIGNALS 120

the general radio signal

s(t) = a(t) cos(2πfc t + φ(t)) = <{a(t)ejφ(t) ej2πfc t }

may also be written

s(t) = a(t) cos 2πfc t cos φ(t) − a(t) sin 2πfc t sin φ(t)
(2.32)
= I(t) cos 2πfc t − Q(t) sin 2πfc t

where
I(t) = a(t) cos φ(t) = <{a(t)ejφ(t) }
(2.33)
Q(t) = a(t) sin φ(t) = ={a(t)ejφ(t) }
INSTRUCTOR COPY

are called the “in-phase” and “quadrature” components, respectively, and thus

I(t) + jQ(t) = a(t)ejφ(t) .

Thus we can write

s(t) = Re{a(t)ejφ(t) ej2πfc t }
= Re{[I(t) + jQ(t)][cos 2πfc t + j sin 2πfc t]}
(2.34)
= I(t) cos 2πfc t − Q(t) sin 2πfc t
= a(t) cos(2πfc t + φ(t))
The general radio signal s(t) must be a real signal that we can view on an oscilloscope.

Exercise 20
Instead of writing s(t) as the real part of a complex signal as done above,
write it as the sum of a positive complex exponential and a negative complex
exponential, and show that the final result for s(t) is still the same.

From the above equation, we see that s(t) can be described as either

1. a cosine wave with amplitude a(t) ≥ 0 and phase φ(t), or

2. the sum of a cosine wave with amplitude I(t) and a sine wave with amplitude
Q(t) where I(t), Q(t) can be greater or less than zero.

In both cases, the message is a two-dimensional (complex) signal represented using

either a(t), φ(t) (polar form) or I(t), Q(t) (rectangular form). This is shown in
fig. 2.5.
 CHAPTER 2. IQ SIGNALS 121

I(t), Q(t)
Q (imaginary)
a(t)6 φ(t)


Q(t) = a(t)sin φ(t)

a(t)

φ(t)
I (real)


I(t) = a(t)cos φ(t)
INSTRUCTOR COPY

Figure 2.5: Polar and rectangular coordinates

2.2.1 Examples of complex baseband signals in the time do-

main

In this section we show specific examples of the general expression for a complex
baseband signal s̃(t) = a(t)ejφ(t) = I(t) + jQ(t).

Example 1

I(t), Q(t) are functions of the message signal(s), where the exact function de-
pends on the modulation type. A simple example is to consider the message
signal to be a stereo (2-channel) music signal written as mL (t), mR (t) and choose

mL (t) = I(t), mR (t) = Q(t)

I(t) + jQ(t) = a(t)ejφ(t) = s̃(t) = mL (t) + jmR (t)

is the complex baseband or complex envelope signal that is a function of the
message but not a function of the carrier. We previously also called the complex
baseband signal the complex envelope. These two terms are equivalent. For a
stereo music signal, the complex baseband signal s̃(t) = mL (t) + jmR (t). The
radio signal s(t) is the so-called real passband signal that contains the message
modulated onto the carrier wave at frequency fc .
 CHAPTER 2. IQ SIGNALS 122

Example 2
The complex baseband signal represents the time varying information repre-
sented by a(t), φ(t) which is transported by a carrier wave from one point to
another. The complex baseband signal s̃(t) = I(t) + jQ(t) = a(t)ejφ(t) can also
be represented as a 3D waveform with projection I(t) onto the real axis and
projection Q(t) onto the imaginary axis
For example, if the amplitude a(t) is constant and the phase φ(t) has only
two possible values −π/4 and −3π/4 to represent a 101010... data sequence,
then the complex baseband signal appears like a bipolar square wave, and the
projections I(t), Q(t) are also square waves (fig. 2.6). More precisely, we can
write

a(t) = Ac
INSTRUCTOR COPY

(
−π/4, binary 1
φ(t) =
−3π/4, binary 0

In a future chapter, we will recognize this waveform as Phase Shift Keying

(PSK).

0
=(z(t))

−1
1
0 0.5 0.5
1 1.5 0
<(z(t))
t

Figure 2.6: Complex baseband phase shift keying with constant a(t) and two
values for φ(t).
 CHAPTER 2. IQ SIGNALS 123

Example 3
In another example, if the amplitude is constant and the phase is varied in
a sinusoidal manner φ(t) = β sin(2πfm t) at some frequency fm to represent
a 101010... data sequence (as will be seen in a later chapter), then the 3D
waveform appears as in fig. 2.7. In this example, the helix rotates in one
direction, slows down and then rotates in the opposite direction, following the
sinusoidal variation in phase. The amount of rotation in each direction depends
on β. The projections I(t), Q(t) are periodic patterns which are not easy to
interpret without seeing the 3D helix. In a future chapter, we will recognize
this waveform as Frequency Shift Keying (FSK).
INSTRUCTOR COPY

1
=(z(t))

−1 1
0 0
0.5 1 1.5 −1 <(z(t))
t

Figure 2.7: Complex baseband frequency-shift keying with constant a(t) and
sinusoidal φ(t).
 CHAPTER 2. IQ SIGNALS 124

Example 4

In the special case where I, Q are both constants independent of t then s(t)
is the sum of a cos wave and a sin wave with different amplitudes, which is a
cosine wave with constant amplitude and phase. Figure 2.8 shows the radio
frequency (RF) signal

s(t) =I cos(2πfc t) − Q sin(2πfc t)

=a cos(2πfc t + φ)

where both a and φ are functions of I, Q. The in-phase component,

I cos(2πfc t), and the quadrature component, Q sin(πfc t), are also cosine waves
at frequency fc with constant amplitude and phase.
INSTRUCTOR COPY

I(t) cos(2πf t)

2
1
−1 π 2π 3π 4π
−2
Q(t) sin(2πf t)

2
1
−1 π 2π 3π 4π
−2
s(t) = I(t) cos(2πf t) + Q(t) sin(2πf t)

2
1
−1 π 2π 3π 4π
−2

Figure 2.8: s(t) resulting from time-constant I = 1 and Q = 2.

Exercise 21
Write an expression for both a, φ as a function of I, Q.
 CHAPTER 2. IQ SIGNALS 125

Example 5
Consider the complex baseband signal to be the sum of two complex exponen-
tials

s̃(t) = a1 ejφ1 ej2πf1 t + a2 ejφ2 ej2πf2 t

This example can be generalized to the sum of N complex exponentials

X
N
S̃(f ) = an ejφn ej2πfn t
n=1
INSTRUCTOR COPY

Note that fn can be positive or negative.

This example will be used in the next section where we consider the Fourier
transform of the complex baseband signal.

Exercise 22
Write expressions for I(t), Q(t), a(t), φ(t) for this example with N = 1 and
N = 2.

2.3 Summary of complex baseband and real pass-

band

A radio (or other communications) signal may be made up of a cosine wave at a carrier
frequency fc with time varying amplitude and phase. The signal may be written as
a real passband signal
s(t) = a(t) cos(2πfc t + φ(t)))
where the message is contained in a(t), φ(t). The signal may also be written in
complex baseband form as s̃(t) = I(t) + jQ(t) = a(t)ejφ(t) and viewed in the complex
plane. The complex baseband signal contains the amplitude and phase only (i.e. the
message information) and does not explicitly include the carrier frequency fc . The
real passband signal may be obtained from the complex baseband signal as

s(t) = <{s̃(t)ej2πfc t }
= a(t) cos[2πfc t + φ(t)]
 CHAPTER 2. IQ SIGNALS 126

The real passband signal is equivalent to the sum of the positive and negative
complex passband signals

s+ (t) + s− (t)
s(t) =
2
a(t) jφ(t) j2πfc t a(t) −jφ(t) −j2πfc t
= e e + e e (2.35)
2 2
s+ (t) = s̃(t)ej2πfc t
s− (t) = s̃∗ (t)e−j2πfc t

s+ (t) is also called the analytic signal (section 2.6).

A description of message types (analog and digital) and how they are contained
in a(t), φ(t) is explained in detail a later section. A brief example: consider a digital
INSTRUCTOR COPY

message consisting of alternating 101010 bits. We can choose a modulation scheme

as follows: assign φ(t) = 0 during a 0 bit and φ(t) = π during a 1 bit and keep a(t)
constant for both 1 and 0 bits. We could also make different choices.
Exercise 23
Sketch a graph of the passband signal waveform with this example modulation
scheme.

Exercise 24
Invent another (different) modulation scheme where we assign a(t), φ(t) as a
function of the message, and sketch the passband signal waveform.

Exercise 25
Invent a modulation scheme where we assign I(t), Q(t) as a function of the
message, and sketch the passband signal waveform.

2.4 Complex baseband signals in the frequency do-

main

In communication systems we frequently encounter the concept of frequency shifting,

where a signal s(t) or s̃(t) is shifted in frequency. In particular, we will see how
the complex baseband signal s̃(t) may be shifted both up in frequency and down in
frequency to create the real passband signal s(t).
 CHAPTER 2. IQ SIGNALS 127

This frequency shifting can be seen more clearly by first introducing the Fourier
transform so that we can write the time domain functions s(t) or s̃(t) in the frequency
domain.
s(t) ⇔ S(f )
s̃(t) ⇔ S̃(f )
where S(f ) is the Fourier transform of s(t) as defined in section 1.4. Note that s(t)
is a function of time, and S(f ) is a function of frequency.

From the definition of a (real) communications signal with information contained

in a(t), φ(t),
s+ (t) + s− (t)
s(t) =
INSTRUCTOR COPY

2
a(t) jφ(t) j2πfc t a(t) −jφ(t) −j2πfc t (2.36)
= e e + e e
2 2
= a(t) cos[2πfc t + φ(t)]
we observe that it consists of the complex envelope multiplied by a positive complex
exponential plus the complex conjugate of the complex envelope multiplied by a neg-
ative complex exponential. We now consider communications signals in the frequency
domain.

2.4.1 Fourier transform concept and definition

The Fourier Transform was reviewed in section 1.4 and defined as

Z ∞
S(f ) = s(t)e−j2πf t dt
t=−∞

In theory, the Fourier transform is only valid if we can observe the signal forever,
from t = −∞ to +∞. In practice, we cannot observe forever, so instead we observe
for some time T and the integral limits become −T /2 to +T /2. The time-limited
Fourier transform is valid for practical purposes when the observation interval (i.e.
the time T spent observing s(t)) is greater than the inverse of the lowest frequency
component greater than zero in the signal s(t).

2.4.2 Fourier transform of complex conjugate

In this section, we consider the Fourier transform of the complex conjugate.

 CHAPTER 2. IQ SIGNALS 128

2.4.2.1 Properties of complex conjugate

First we review some properties of the complex conjugate.

We note the following results for the complex conjugate that apply both in the
time and frequency domain.

For any complex function, the magnitude of the complex conjugate is the same
as the magnitude of the original function, and the phase of the complex conjugate is
the negative of the phase of the original function In the time domain

s̃(t) = a(t)ejφ(t)
s̃∗ (t) = a(t)e−jφ(t)
INSTRUCTOR COPY

|s̃(t)| = |s̃∗ (t)|

= a(t)
arg s̃(t) = − arg s̃∗ (t)
= φ(t)

In the frequency domain

S̃(f ) = α(f )ejψ(f )

S̃ ∗ (f ) = α(f )e−jψ(f )
|S̃ ∗ (f )| = |S̃(f )|
= α(f )
arg S̃ ∗ (f ) = − arg S̃(f )
= −ψ(f )

This result also applies if we replace f with −f which will be useful in the next section

|S̃ ∗ (−f )| = |S̃(−f )|

arg S̃ ∗ (−f ) = − arg S̃(−f )

Note that

• α(f ) is NOT the Fourier transform of a(t)

• ψ(f ) is NOT the Fourier transform of φ(t)

For two complex functions r and s

 CHAPTER 2. IQ SIGNALS 129

r∗ ± s∗ = (r ± s)∗
(rs)∗ = r∗ s∗
(r/s)∗ = r∗ /s∗

2.4.2.2 Fourier transform of complex conjugate

The Fourier transform of the complex conjugate is also the complex conjugate but
with a negative sign in frequency. The mathematical proof is below. The interpreta-
tion of the meaning of −f is in the following section.
INSTRUCTOR COPY

s̃(t) ↔ S̃(f )
s̃∗ (t) ↔ S̃ ∗ (−f )

s+ (t) ↔ S+ (f )
s∗+ (t) ↔ S+∗ (−f )

Proof for s+
s+ (t) = ∫ ∞
−∞ S+ (f )e
j2πf t
df
s∗+ (t) = [∫ ∞
−∞ S+ (f )e
j2πf t
df ]∗
= ∫∞ ∗
−∞ S+ (f )e
−j2πf t
df
replace f with − f
= ∫ −∞ ∗
+∞ S+ (−f )e
j2πf t
(−df )
= ∫∞ ∗
−∞ S+ (−f )e
j2πf t
df
−1 ∗
= F {S+ (−f )}

The proof for s̃(t) is identical.

Exercise 26
Prove the above for the specific example of s+ (t) = aejφ ej2πfc t .
 CHAPTER 2. IQ SIGNALS 130

Solution

s+ (t) = aejφ ej2πfc t

S+ (f ) = aejφ δ(f − fc )
s∗+ (t) = ae−jφ e−j2πfc t
= s− (t)
S+∗ (−f ) = ae−jφ δ(−f − fc )
= S− (f )

s∗+ (t) ↔ S+∗ (−f )

ae−jφ e−j2πfc t ↔ ae−jφ δ(−f − fc )
INSTRUCTOR COPY

↔ ae−jφ δ(f + fc )

since δ(+f + fc ) = δ(−f − fc ) = ∞, f = −fc (2.37)

The Fourier transform functions S̃(f ) and S+ (f ) are complex so we consider the
complex conjugate results in both magnitude and phase.

The properties of Fourier transform of complex conjugate proven in this section

s̃(t) ↔ S̃(f )
s̃∗ (t) ↔ S̃ ∗ (−f )

are combined with the properties of complex conjugate from the previous subsection

|S̃ ∗ (f )| = |S̃(f )|
arg S̃ ∗ (f ) = − arg S̃(f )
|S̃ ∗ (−f )| = |S̃(−f )|
arg S̃ ∗ (−f ) = − arg S̃(−f )

to yield the result

|S̃ ∗ (−f )| = |S̃(−f )| = mirror image of |S̃(f )|

arg S̃ ∗ (−f ) = − arg S̃(−f ) = negative mirror image of arg S̃(f )

which is interpreted as
 CHAPTER 2. IQ SIGNALS 131

• the magnitude spectrum of s̃∗ (t) is the mirror image of the magnitude spectrum
of s̃(t)

• the phase spectrum of s̃∗ (t) is the negative mirror image of the phase spectrum
of s̃(t)

The interpretation of the meaning of −f in the above equation as a mirror image is

further explained and illustrated with figures in the following example.

2.4.2.3 Example complex baseband spectrum and its complex conjugate

To visualize and interpret the meaning of this equation we will choose an example
INSTRUCTOR COPY

function for the spectral shape of S̃(f ) and sketch (plot) it. We show both the spectral
shape of the magnitude |S̃(f )| and the shape of the phase arg S̃(f ).

In this example we choose |S̃(f )| to have a continuous part plus two discrete
complex exponentials. The shape of the continuous part is chosen to be a simple
shape with the properties that the original shape, the mirror image, the negative and
the negative mirror image are all unique. Other examples are shown in section 2.7.

We choose the continuous part of |S̃(f )| to be a triangular shape centered around

zero frequency with zero magnitude at f ≤ − B2 , magnitude a at f = 0 and maximum
magnitude 2a at f = B2 and zero amplitude for f > B2 . An equation for this triangular
shape such that the spectrum from − B2 ≤ f ≤ B2 has amplitude α(f ) is
 
2α0 B
α(f ) = f+
B 2

This spectrum is plotted in fig. 2.9 where the bandwidth of S̃(f ) is B. Note that this
example spectrum |S̃(f )| is asymmetrical about f = 0 as it must be for s̃(t) with
non-zero real and imaginary components.
 CHAPTER 2. IQ SIGNALS 132

|S̃(f )| |S̃ ∗ (−f )|

2a 2a
a a
f f
− B2 B
2 − B2 B
2
←− t

←− t
INSTRUCTOR COPY

0 a 2a 3a 0 a 2a 3a

Figure 2.9: Spectra and waterfalls showing the mirroring from S̃(f ) to S̃ ∗ (−f ).

Figure 2.9 shows only the continuous part of the spectrum (the triangle) along
with the corresponding waterfall, assuming that there is no changes with time, and
all frames used the generate the waterfall are the same.

We choose the discrete part of |S̃(f )| to be two complex exponentials, one with
amplitude > 2a at f = −B/4 and another with amplitude > 3a at f = +B/4

The phase ψ(f ) is chosen to be a shape as shown in the figure. The phase of the
discrete complex exponentials is the same as that of the continuous spectrum at the
relevant frequency, so it does not show up as a discrete phase. The shape is chosen
to be a simple shape with the properties that the original shape, the mirror image,
the negative and the negative mirror image are all unique.

Since the phase is normally plotted in the range −π to +π, the phase function
may be written
(
2π
B
(f + B) , −B/2 < f < −B/4
ψ(f ) = 2π
B
(−f ) , −B/4 < f < B/2
 CHAPTER 2. IQ SIGNALS 133

Recall from the previous section

|S̃ ∗ (−f )| = |S̃(−f )| = mirror image of |S̃(f )|

arg S̃ ∗ (−f ) = − arg S̃(−f ) = negative mirror image of arg S̃(f )

The magnitude of the term |S̃ ∗ (−f )| is a mirror image of |S̃(f )| flipped around
f = 0 because of the minus sign in front of f as shown in fig. 2.10.

The phase of S̃ ∗ (−f ) is also the mirror image and also the negative of the phase
of S̃(f ).

|S̃(f )| |S̃ ∗ (−f )|

INSTRUCTOR COPY

3a 3a
2a 2a
a a

− B2 B f − B2 B f
2 2
   
arg S̃(f ) arg S̃ ∗ (−f )

π
π
2

− B2 B ψ − π2 − B2 B ψ
2 2
−π

Figure 2.10: The magnitude and phase of S̃(f ) and it’s mirror, S̃ ∗ (−f )

Since S̃(f ) does not change with time, the waterfall plot is a fat vertical line
extending from − B2 to B2 color-coded to show how the magnitude changes with fre-
quency.

Figure 2.11 shows the spectrum including both continuous and discrete parts. The
corresponding waterfall shows the discrete components as (high magnitude, strong
signal) orange or red lines on top of the (low magnitude, weak signal) blue-green
background of the continuous spectrum.
 CHAPTER 2. IQ SIGNALS 134

|S̃(f )| |S̃ ∗ (−f )|

3a 3a
2a 2a
a a
f f
− B2 B
2 − B2 B
2
←− t

←− t
INSTRUCTOR COPY

0 a 2a 3a 0 a 2a 3a

Figure 2.11: The magnitude plots from fig. 2.10 with waterfall plots showing that
they are continuous in time

2.5 Real passband signals in the frequency domain

In communication systems we frequently encounter the concept of frequency shifting,

where a signal s(t) or s̃(t) is shifted in frequency. In particular, we will see how
the complex baseband signal s̃(t) may be shifted both up in frequency and down in
frequency to create the real passband signal s(t).

This frequency shifting can be seen more clearly by first introducing the Fourier
transform so that we can write the time domain functions s(t) or s̃(t) in the frequency
domain.
s(t) ⇔ S(f )
s̃(t) ⇔ S̃(f )
where S(f ) is the Fourier transform of s(t) as defined in section 1.4. Note that s(t)
is a function of time, and S(f ) is a function of frequency.

From the definition of a (real) communications signal with information contained

 CHAPTER 2. IQ SIGNALS 135

in a(t), φ(t),

s+ (t) + s− (t)
s(t) =
2
s̃(t) j2πfc t s̃∗ (t) −j2πfc t
= e + e
2 2 (2.38)
a(t) jφ(t) j2πfc t a(t) −jφ(t) −j2πfc t
= e e + e e
2 2
= a(t) cos[2πfc t + φ(t)]

we observe that it consists of the complex envelope multiplied by a positive complex

exponential plus the complex conjugate of the complex envelope multiplied by a
negative complex exponential.
INSTRUCTOR COPY

We will use the Fourier transform to interpret this multiplication to mean that
communications signal comprises the complex baseband signal shifted up in frequency
plus the complex conjugate of the complex baseband signal shifted down in frequency.

We wish to consider communications signals in the frequency domain. To do this,

we first review the frequency shifting properties of the Fourier transform.

2.5.1 Fourier transform shifting properties

The Fourier transform has frequency-shifting properties that will be used often through-
out this textbook. The properties are

s(t) ⇔ S(f )
j2πf1 t
s(t)e ⇔ S(f − f1 )
−j2πf1 t
s(t)e ⇔ S(f + f1 )

A review of how the shifting properties arise is given below, remembering that
(
∞ x=0
δ(x) = (2.39)
0 x 6= 0

s(t) ⇔ S(f )
S(f ) = ∫ ∞
t=−∞ s(t)e
−j2πf t
dt
 CHAPTER 2. IQ SIGNALS 136

δ(t) ⇔ 1
1 ⇔ δ(f )

ej2πf1 t ⇔ δ(f − f1 )
e−j2πf1 t ⇔ δ(f + f1 )
Proof: S(f ) = ∫ ∞
t=−∞ s(t)e
−j2πf t
dt
= ∫∞
t=−∞ e
−j2πf1 t −j2πf t
e dt
= ∫∞
t=−∞ e
−j2π(f +f1 )t
dt = δ(f + f1 )

s(t)ej2πf1 t ⇔ S(f ) ⊗ δ(f − f1 ) = S(f − f1 )

s(t)e−j2πf1 t ⇔ S(f ) ⊗ δ(f + f1 ) = S(f + f1 )
INSTRUCTOR COPY

Proof: S(f ) ⊗ H(f ) = ∫ ∞α=−∞ S(α)H(f − α)dα

S(f ) ⊗ δ(f ) = ∫ ∞
α=−∞ S(α)δ(f − α)dα
= S(f )
S(f ) ⊗ δ(f − f1 ) = ∫ ∞
α=−∞ S(α)δ(f − f1 − α)dα
= S(f − f1 )

2.5.2 Fundamental result arising from Fourier transform shift-

ing properties

To apply these properties to the communications signal s(t) and its complex baseband
s̃(t), recall the definitions

s+ (t) = s̃(t)ej2πfc t
s− (t) = s̃∗ (t)e−j2πfc t
2s(t) = s+ (t) + s− (t)
= s̃(t)ej2πfc t + s̃∗ (t)e−j2πfc t

Transforming these definitions into the frequency domain using the shifting prop-
erties shown above
 CHAPTER 2. IQ SIGNALS 137

S+ (f ) = S̃(f − fc )
S− (f ) = S̃ ∗ (−(f + fc ))

2S(f ) = S+ (f ) + S− (f )
(2.40)
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))

Equation (2.40) provides useful insight into the ideas of complex baseband, real
passband, positive and negative frequencies in both time and frequency domain.

In what follows, this equation will be interpreted to mean that any real passband
INSTRUCTOR COPY

signal is the sum of the complex baseband spectrum S̃(f ) shifted by +fc and the
flipped (mirror-image) complex baseband spectrum S̃ ∗ (−f ) shifted by −fc .

This interpretation is further explained in the following section.

2.5.2.1 Example of shifted complex baseband spectrum and its complex

conjugate

In this section, we consider this interpretation in more detail and illustrate with
figures.

Using this example S̃(f ), we now denote

S+ (f ) = S̃(f − fc )
(2.41)
S− (f ) = S̃ ∗ (−(f + fc ))

The term S̃(f − fc )) = S+ (f ) is shifted by +fc because of (f − fc ).

The term S̃ ∗ (−(f + fc )) = S− (f ) is shifted by −fc because of (f + fc ).

The term S̃ ∗ (−(f + fc )) = S− (f ) is a mirror image of S̃(f − fc ) = S+ (f ) around

f = 0. This is illustrated in fig. 2.12.
 CHAPTER 2. IQ SIGNALS 138

2a
a

−(f + fc ) f − fc f

←− t
INSTRUCTOR COPY

0 a 2a

Figure 2.12: Spectra for 2S(f ) showing both S+ (f ) equal to S̃(f ) shifted up so that
it is centered at fc and S− (f ) equal to the mirror image S̃(−f ) shifted down so that
it is centered at −fc . This is an illustration of eq. (2.40).

To summarize, the two figures above illustrate

• S̃(f ) as an asymmetrical spectrum centered around f = 0 and S̃ ∗ (−f ) as the

mirror image of that asymmetrical spectrum centered around f = 0, see fig. 2.9,
• S̃(f − fc ) = S+ (f ) as an asymmetrical spectrum centered around f = fc and
S̃ ∗ (−(f + fc )) = S− (f ) as the mirror image of that asymmetrical spectrum
centered around f = −fc , see fig. 2.12.

Thus we see that the real communications signal is made up of the complex base-
band signal (containing the message information) shifted by the carrier frequency.
The shape of the spectrum of the complex baseband signal S̃(f ) is the same as the
shape of the spectrum of the positive frequency signal S+ (f ), and the shape of the
spectrum of the conjugate of the complex baseband signal S̃ ∗ (−f ) is the same as the
shape of the spectrum of the negative frequency signal S− (f ). This result will be seen
again in section 2.3.3.

The Fourier transform of a real signal s(t) will always be symmetrical around
0 Hz, i.e. the shape of the Fourier transform (the spectrum) in the negative frequency
range will be the mirror image of the Fourier transform (spectrum) in the positive
frequency range. This is because for a real signal, s∗ (t) = s(t) so that the spectrum
 CHAPTER 2. IQ SIGNALS 139

S ∗ (−f ) = S(f ). This means that the mirror image spectrum S ∗ (−f ) is the same as
the original spectrum S(f ). Thus S(f ) must be symmetrical about f = 0.

The Fourier transform of a complex signal such as s+ (t) or s̃(t) will have an
asymmetrical spectrum, i.e the spectrum at negative frequencies will not be the mirror
image of the spectrum at positive frequencies. For s+ (t) the spectrum is zero at
negative frequencies but not zero at positive frequencies. For s̃(t) the spectrum may
be non-zero but different (not a mirror image) at negative and positive frequencies.

Exercise 27
Show how the equations below are interpreted to show that any real passband
signal is the sum of the complex baseband spectrum shifted by +fc and the
flipped (mirror-image) complex baseband spectrum shifted by −fc . Illustrate
INSTRUCTOR COPY

with figures for both the magnitude and the phase of each term.

S+ (f ) = S̃(f − fc )
S− (f ) = S̃ ∗ (−(f + fc ))

2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))

Solution
The term S̃ ∗ (−f ) is a mirror image of S̃(f ) flipped around f = 0 because of
the minus sign in front of f .
The term S̃(f −fc )) is shifted by +fc because of (f −fc ). The term S̃ ∗ (−(f +fc ))
is shifted by −fc because of (f + fc ).
The term S̃ ∗ (−(f + fc )) is a mirror image of S̃(f − fc ) around f = 0. This
may be illustrated by

• sketching S̃(f ) as an asymmetrical spectrum centered around f = 0

and S̃ ∗ (−f ) as the mirror image of that asymmetrical spectrum centered
around f = 0,

• sketching S̃(f − fc ) = S+ (f ) as an asymmetrical spectrum centered

around f = fc and S̃ ∗ (−(f + fc )) = S− (f ) as the mirror image of that
asymmetrical spectrum centered around f = −fc .
 CHAPTER 2. IQ SIGNALS 140

2.6 Analytic signals in time and frequency domain

Recall the definitions

s̃(t) = a(t)ejφ(t)
s̃∗ (t) = a(t)e−jφ(t)
s+ (t) = s̃(t)ej2πfc t
s− (t) = s̃∗ (t)e−j2πfc t
2s(t) = s+ (t) + s− (t)
= s̃(t)ej2πfc t + s̃∗ (t)e−j2πfc t

Transforming these definitions into the frequency domain gives

INSTRUCTOR COPY

2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))

The analytic signal Sa (f ) is the same as S+ (f ) as defined below

Sa (f ) = S+ (f )
= 2u(f )S(f )
= S(f ) + sgn(f )S(f )
(
2S(f ), f > 0
Sa (f ) = (2.42)
0, f <0
where u(f ) = 1 + sgn(f ) is the Heaviside step function
(
1 f >0
u(f ) = (2.43)
0 f <0
and sgn(f ) is the sign function
(
1 f >0
sgn(f ) = (2.44)
−1 f < 0

In section 5.2, we will see how to transform Sa (f ) = S+ (f ) back to the time

domain using the idea of a Hilbert transform. We will thus obtain a time domain
expression for an analytic signal.

Exercise 28
Write the frequency domain definition of the analytic signal for the special case
where a(t) = 1, φ(t) = 0.
 CHAPTER 2. IQ SIGNALS 141

2.7 Example complex baseband signals in time and

frequency domains

2.7.1 Complex exponentials at discrete frequencies

Following section 2.2.1 in section 2.2.1, consider a complex baseband signal that is
the sum of N complex exponentials, each with different amplitudes an , frequencies
fn and phases φn

s̃(t) ↔ S̃(f )
X
N
s̃(t) = an ej(2πfn t+φn )
INSTRUCTOR COPY

n=1
XN
S̃(f ) = an ejφn δ(f − fn )
n=1

s̃∗ (t) ↔ S̃ ∗ (−f )

X
N
∗
s̃ (t) = an e−j(2πfn t+φn )
n=1
XN
S̃ ∗ (−f ) = an e−jφn δ(f − (−fn ))
n=1
XN
= an e−jφn δ(−f − fn )
n=1

If we choose the discrete frequencies fn to be equally spaced between − B2 and B

2
,
and corresponding amplitudes an such that
 
2a B
an = fn +
B 2

(as in fig. 2.13) then the spectrum S̃(f ) is a sampled version of the example triangular-
shaped spectrum in fig. 2.9. The phases φn at each frequency fn are not shown in the
figure and may be arbitrarily chosen in this example,
 CHAPTER 2. IQ SIGNALS 142

|S̃ ∗ (f )| |S̃ ∗ (−f )|

3a 3a
INSTRUCTOR COPY

2a 2a
a a
f f
− B2 B
2 − B2 B
2
←− t

←− t

0 a 2a 3a 0 a 2a 3a

Figure 2.13: Spectra and waterfalls showing the mirroring from S̃(f ) to S̃ ∗ (−f ) for
a signal made up of complex exponentials at discrete frequencies fn .
 CHAPTER 2. IQ SIGNALS 143

For this example S̃(f ) we can write

S+ (f ) = S̃(f − fc )
X
N
= an ejφn δ(f − fn − fc )
n=1
∗
S− (f ) = S̃ (−(f + fc ))
X
N
= an e−jφn δ(−(f + fn + fc ))
n=1
XN
= an e−jφn δ(−f − fn − fc ))
n=1
INSTRUCTOR COPY

X
N
= an e−jφn δ(f + fn + fc ))
n=1

Exercise 29
Write out all the terms in the expression for s(t) and S(f ) for the example in
fig. 2.13 but where we choose the specific case N = 4. Hint: recall eq. (2.40).

Now consider the above complex baseband signal s̃(t) where the amplitude and
phase of each complex exponential varies with time, so that we can write an = an (t)
and φn = φn (t). There are many possible choices for the functions an (t), φn (t).

2.7.2 Continuous-time spectrum

An example continuous-time spectrum is the ”envelope” of the discrete-time spectrum

of the previous example.

If we choose a continuous range of frequencies f such that 0 − B2 ≤ f ≤ B2 , we

choose the same function as in section 2.2.1 in section 2.2.1. This spectrum is plotted
in fig. 2.13)

2.7.3 Discrete exponentials plus continuous time spectrum

The most general (and also most realistic) example spectrum contains both discrete
frequency components (complex exponentials) plus a continuous time spectrum that
 CHAPTER 2. IQ SIGNALS 144

may be interpreted as the sum of many unresolved complex exponentials. An illus-

tration of this is fig. 2.11.

2.7.4 Discrete-time spectrum

A discrete-time spectrum will consist of discrete complex exponentials at up to N

discrete frequencies as shown in section 2.7.1.

In effect, by sampling the continuous-time signal at times t = nTs to make a

discrete-time signal, the continuous-time spectrum is resolved into a discrete spec-
trum. The combined continuous-time plus discrete exponentials spectrum in the ex-
ample above is effectively sampled in the frequency domain at intervals f0 = fs /N =
INSTRUCTOR COPY

1/(ts N ) and becomes the sum of N complex exponentials. Typically the discrete
frequency components in the spectrum of the previous example will have much larger
amplitudes than the sampled continuous-time components.

2.7.5 Complex baseband signals containing multiple messages

The example complex baseband signal comprising the sum of many complex exponen-
tials in the previous section may represent a single message or may represent multiple
messages.

For the case of a single message, consider creating the spectrogram as described
in section 1.5. The signal is divided into overlapping windows with window index
m. Recall that the spectrum at time index m and frequency index k is the FFT
components Xm,k of the message. At each time index m, the complex values of Xm,k
will change in general. The message may be a human voice signal as described in
section 1.3.4. In this example, the message is a real signal, so that the imaginary part
of the complex baseband signal is zero, and the spectrum is symmetric about zero
frequency.

For the case of multiple messages, the k-th message ak ejφk is associated with the
k-th complex exponential at fk . In this case the complex exponentials are called
sub-carriers, where the k-th sub-carrier is the carrier wave for the k-th message and
frequency fk .

What follows are two examples of a complex baseband signal containing multiple
subcarriers.

Consider a radio system receiving WiFi (real passband) signals in the 2.4 GHz
 CHAPTER 2. IQ SIGNALS 145

range. For this case the real passband signal may contain up to 11 communications
signals in a bandwidth B = 72 MHz with center frequencies fn from f1 = 2412 MHz
to f11 = 2462 MHz, spaced by 5 MHz. The center frequency is f6 = 2437 MHz and
can be viewed as the ”carrier” frequency for this group of 11 signals. Each signal
is 12 MHz wide, so the adjacent signals overlap. By shifting the center frequency
to zero, we obtain the complex baseband signal in the range −B/2 = −36 MHz to
−B/2 = 36 MHz that contains multiple subcarriers, each with a different message.

Similarly for a radio system receiving real passband signals in the 88–108 MHz
FM broadcast range, the center frequency 98 MHz is shifted to zero to obtain the
complex baseband signal that contains up to 99 FM broadcast signals in the 20 MHz
bandwidth from −B/2 =−10 MHz to B/2 = 10 MHz. Each FM signal occupies up
to 200 kHz bandwidth, so that adjacent channels do not overlap.
INSTRUCTOR COPY

Exercise 30
Write a formula for the frequencies fn in this FM broadcast example where
1 ≤ n ≤ 99 and find the center frequency. Hint: f1 = 88.1 MHz and f99 =
107.9 MHz.

2.7.6 Complex baseband with zero imaginary component

In section 2.1, section 2.2.1 showed that a complex baseband signal s̃(t) consisting of
two complex exponentials summed was indeed complex with an asymmetrical spec-
trum where the amplitudes of the positive and negative frequencies were different.

In this subsection we consider the special case where s̃(t) = I(t) + jQ(t) with
Q(t) = 0., so that the complex baseband signal s̃(t) = I(t) is real (not complex).

This example is useful when the complex baseband signal is used to represent a
single real message m(t), so that s̃(t) = I(t) = m(t).

2.7.6.1 General real message m(t)

For a real message m(t) we can apply the results obtained previously as follows.

Recalling the Fourier transform pair

s̃(t) ↔ S̃(f )
s̃∗ (t) ↔ S̃ ∗ (−f )
 CHAPTER 2. IQ SIGNALS 146

we substitute s̃(t) = m(t) to be a real message, to obtain

m(t) ↔ M (f )
m∗ (t) ↔ M ∗ (−f )

Since m(t) is real,

m∗ (t) = m(t)
M ∗ (−f ) = M (f ) = |M (f )|ej arg M (f )
M (−f ) = M ∗ (f ) == |M (f )|e−j arg M (f )

For any complex function , the magnitude of the complex conjugate is the same as
the magnitude of the original function, and the phase of the complex conjugate is the
INSTRUCTOR COPY

negative of the phase of the original function

S̃(f ) = α(f )ejψ(f )

S̃ ∗ (f ) = α(f )e−jψ(f )
|S̃ ∗ (f )| = |S̃(f )| = α(f )
arg S̃ ∗ (f ) = − arg S̃(f ) = −ψ(f )

This result also applies if we replace f with −f

|S̃ ∗ (−f )| = |S̃(−f )| = α(−f )

arg S̃ ∗ (−f ) = − arg S̃(−f ) = −ψ(−f )

This result also applies for a real function S̃(f ) = M (f ). so that

|M ∗ (−f )| = |M (−f )|
arg M ∗ (−f ) = − arg M (−f )

Combining these results for M (f ) with the result M ∗ (−f ) = M (f ) we obtain

|M (f )| = |M (−f )|
arg M (f ) = − arg M (−f )

This result means that the magnitude of the mirror image spectrum M ∗ (−f ) about
f = 0 (recall section 2.5.2) is the same as the magnitude of the original spectrum
M (f ). This is illustrated in fig. 2.14 where it is shown that the value of |M (−f )|
at a given f > 0 or −f < 0) is the same as |M (f )| at the same f > 0. Thus the
magnitude |M (f )| must be symmetric about f = 0.
 CHAPTER 2. IQ SIGNALS 147

M+ (f ) M− (f )

4Am 4Am
3Am 3Am
2Am 2Am
Am Am
f f
−B B −B B

Figure 2.14: Positive and negative frequency components of a general real message
m(t).
INSTRUCTOR COPY

The phase of the mirror image spectrum is the opposite sign arg M (−f ) = − arg M (f )
and is thus anti-symmetric about f = 0. Recall that for any real m(t), the phase
arg M (f ) is not necessarily zero.

Exercise 31
Show that if m(t) is real so that M ∗ (−f ) = M (f ) then

|M (−f )| = |M (f )|
arg M (−f ) = − arg M (f )

is true for m(t) = sin 2πfc t.

Solution

m(t) = sin 2πfc t

ej2πfc t − e−j2πfc t
=
2j
1
M (f ) = [δ(f − fc ) − δ(f + fc )]
2j
1
M ∗ (−f ) = [δ(−f − fc ) − δ(−f + fc )] = M (f )
2(−j)
1
|M (f )| = [δ(f − fc ) − δ(f + fc )]
2
1
|M (−f )| = [δ(−f − fc ) − δ(−f + fc )] = |M (f )|
2
 CHAPTER 2. IQ SIGNALS 148

arg M (f ) = −π/2, f = fc
= +π/2, f = −fc
1
M (−f ) = [δ(−f − fc ) − δ(−f + fc )]
2j
arg M (−f ) = −π/2, −f = fc
= +π/2, −f = −fc = − arg M (f )

This symmetry result also applies to the real function δ(f ) so that δ ∗ (−f ) = δ(f ).

To review, for a real baseband signal S̃(f ) = M (f ), we found that for the complex
INSTRUCTOR COPY

baseband spectrum M ∗ (−f ) = M (f ). so that the magnitude of the complex baseband

spectrum is symmetrical about f = 0
S̃(f ) = M (f )
∗
M (−f ) = M (f )
|S̃ ∗ (−f )| = |S̃(f )| = |M ∗ (−f )| = |M (f )|
and the phase of the complex baseband spectrum is anti-symmetric about f = 0
S̃(f ) = M (f )
∗
M (−f ) = M (f )
arg{S̃(f )} = arg M (f )
arg{S̃ ∗ (−f )} = − arg{S̃(f )} = − arg M (f )

2.7.6.2 Frequency-shifted message

For a real message m(t) ↔ M (f ) where M ∗ (−f ) = M (f ) for real m(t) = m∗ (t),

s+ (t) = s̃(t)ej2πfc t
= m(t)ej2πfc t
s− (t) = s̃∗ (t)e−j2πfc t
= m∗ (t)e−j2πfc t
= m(t)e−j2πfc t
2s(t) = s+ (t) + s− (t)
= m(t)ej2πfc t + m(t)e−j2πfc t
 CHAPTER 2. IQ SIGNALS 149

2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))
= M (f − fc ) + M ∗ (−(f + fc ))
= M (f − fc ) + M (f + fc )

In the last two lines, M ∗ (−(f + fc )) = M (f + fc ) because for real m(t) = m∗ (t),
M ∗ (−f ) = M (f ) and thus also M ∗ (−(f + fc )) = M (f + fc ).

We observe that using the previous result about the mirror image of the magnitude
and phase spectrum of real message signals

|M (f )| = |M (−f )|
INSTRUCTOR COPY

arg M (f ) = − arg M (−f )

that we have the following symmetries and asymmetries for a frequency-shifted mes-
sage:

• Since |M (−f )| = |M (f )|, and replacing the variable f with f − fc , we have

|M (−(f − fc ))| = |M (f − fc )|, and replacing the variable f with f + fc , we have
|M (−(f + fc ))| = |M (f + fc )|

• The magnitude |S+ (f )| = |M (f − fc )| is symmetric about f = fc ,

• Thus the mirror image of the magnitude about f − fc which is written as

|M (−(f − fc ))| is the same as the original magnitude |M (f − fc )|.

• The magnitude |S− (f )| = |M (f + fc )| is symmetric about f = −fc .

• Thus the mirror image of the magnitude about f + fc which is written as

|M (−(f + fc ))| is the same as the original magnitude |M (f + fc )|.

• Since arg M (−f )| = − arg M (−f )|, and replacing the variable f with f − fc ,
we have arg M (−(f − fc )) = − arg M (f − fc ), and replacing the variable f with
f + fc , we have arg M (−(f + fc )) = − arg M (f + fc ).

• The phase of S+ (f ) = M (f − fc ) is anti-symmetric about f = fc ,

• Thus the mirror image of the phase arg M (f − fc ) about f − fc which is written
as arg M (−(f − fc )) is the opposite sign from the original phase, i.e. the mirror
image is written − arg M (f − fc ).

• The phase of S− (f ) is anti-symmetric about f = −fc .

 CHAPTER 2. IQ SIGNALS 150

• Thus the mirror image of the phase arg M (f + fc ) about f + fc which is written
as arg M (−(f + fc )) is the opposite sign from the original phase, i.e. the mirror
image is written − arg M (f + fc ).

Exercise 32
Sketch example magnitude and phase spectra that are consistent with the above
observations.

2.7.6.3 Multitone message m(t)

In this example we consider a real message m(t) that is the sum of N complex
INSTRUCTOR COPY

exponentials, where the amplitudes ak , frequencies fk and phases φk must be related

in such a way that the message is real. In particular, for the signal to be real, the
complex exponentials must come in pairs such that Ak ejθ + Ak e−jθ = 2Ak cos θ, where
θ = 2πfk t + φk . Thus we write
m(t) ↔ M (f )
N/2
X
2m(t) = Ak ej(2πfk t+φk )
k=−N/2
N/2
X
= Ak 2 cos(2πfk t + φk )
k=−N/2
N/2
X
M (f ) = Ak ejφk δ(f − fk )
k=−N/2

2.7.6.4 Dual tone message m(t)

In this example, we choose m(t) to be four complex exponentials at frequencies

±f1 , ±f2 that combine to make two cosine waves at f1 , f2 . Thus m(t) is a “two-
tone” message like the DTMF (Dual-tone Multi-frequency) touch-tones heard when
dialing a phone and when responding to voice prompts during a phone call to an
automated service such as telephone banking.

The dual tones used for each key on the telephone keypad are shown in table 2.1
where the choices for f1 are along the vertical axis and the choices for f2 are along
the horizontal axis.
 CHAPTER 2. IQ SIGNALS 151

1209 Hz 1336 Hz 1477 Hz

697 Hz 1 2 3
770 Hz 4 5 6
852 Hz 7 8 9
941 Hz * 0 #

Table 2.1: Dual tones used for telephone keypads

k fk ak
-2 −B/2 Am
-1 −B/4 Am /2
0 0 0
1 B/4 Am /2
2 B/2 Am
INSTRUCTOR COPY

Table 2.2: Example values for a dual tone message with bandwidth B and maximum
amplitude Am .

The example message is of bandwidth B and may be written

m(t) ↔ M (f )
N/2
X
m(t) = Ak ej(2πfk t+φk )
k=−N/2
N/2
X
M (f ) = Ak ejφk δ(f − fk )
k=−N/2

We choose N = 4, f2 = B/2 and

2Am
Ak = (|fk |)
B
φk = 0

where Am is an constant amplitude associated with the message independent of the

frequency index k. Compare this expression to section 2.7.1.

Thus we have the values in table 2.2 and we note that a−2 = a2 = Am and
a−1 = a1 = Am /2 and a0 = 0,

Exercise 33
Write expressions for m(t) and M (f ) for this case and sketch a figure for M (f ).
 CHAPTER 2. IQ SIGNALS 152

Solution

N/2
X
m(t) = Ak ej(2πfk t+φk )
k=−N/2

= a−2 ej(2πf−2 t+φ−2 ) + a−1 ej(2πf−1 t+φ−1 ) + a1 ej(2πf1 t+φ1 ) + a2 ej(2πf2 t+φ2 )
= Am ej(2πf−2 t+φ−2 ) + (Am /2)ej(2πf−1 t+φ−1 ) + (Am /2)ej(2πf1 t+φ1 ) + Am ej(2πf2 t+φ2 )
= 2Am cos(2πf2 t + φ2 ) + Am cos(2πf1 t + φ1 )
= 2Am cos(2π(B/2)t + φ2 ) + Am cos(2π(B/4)t + φ1 )
INSTRUCTOR COPY

Am

−f2 −f1 f1 f2 f

Figure 2.15: Magnitude spectrum |M (f )| for a dual tone message.

As shown in fig. 2.15, the spectrum M (f ) of the dual tone message for the case
φ1 = φ2 = 0 is
N/2
X
M (f ) = Ak ejφk δ(f − fk )
k=−N/2

= a−2 δ(f − f−2 ) + a−1 δ(f − f−1 ) + a1 δ(f − f1 ) + a1 δ(f − f2 )

= Am δ(f + B/2) + (Am /2)δ(f + B/4) + (Am /2)δ(f − B/4) + Am δ(f − B/2)
= Am δ(f + f2 ) + (Am /2)δ(f + f1 ) + (Am /2)δ(f − f1 ) + Am δ(f − f2 )

Exercise 34
Write an expression for S+ (f ) and S− (f ) using the result for a real message
S̃(f ) = M (f )
 CHAPTER 2. IQ SIGNALS 153

Solution

2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))
= M (f − fc ) + M ∗ (−(f + fc ))
= M (f − fc ) + M (f + fc )

since for real m(t) = m∗ (t), M (f ) = M ∗ (−f )

2.7.7 Summary
INSTRUCTOR COPY

The main ideas in this section are:

• the idea of a carrier wave c(t) = cos 2πfc t that is an electromagnetic wave that
can propagate over a distance via a pair of wires, waveguide, optical fiber or
free space.

• a general radio signal s(t) = a(t) cos[2πfc t+φ(t)] can be described by amplitude
a(t), frequency fc and phase φ(t).

• we can use a(t) and/or φ(t) or some combination thereof to represent a message
m(t).

• the complex baseband signal s̃(t) = a(t)ejφ(t) that represents the amplitude and
phase of the signal (i.e. the message) does not include the carrier wave at fc ,

• a complex exponential carrier wave can be c+ (t) = ej2πfc t that contains only
positive frequencies or c− (t) = e−j2πfc t that contains only negative frequencies.
These two complex carrier waves can be plotted as a 3-D helix rotating in
opposite directions.

• the real carrier wave is 2c(t) = c+ (t) + c− (t).

• a real signal s(t) is the sum of a positive frequency signal s+ (t) at carrier fre-
quency +fc and a negative frequency signal s− (t) at carrier frequency −fc .

• the real signal 2s(t) = s+ (t) + s− (t)

• the relationship between s(t) and s̃(t) using the complex exponential carrier
wave ej2πfc t can be described using the equations below.
 CHAPTER 2. IQ SIGNALS 154

In the time domain we show the relationship between the complex baseband signal
s̃(t), the analytic signal s+ (t) and the real signal s(t)

s+ (t) = s̃(t)ej2πfc t
s− (t) = s̃∗ (t)e−j2πfc t
2s(t) = s+ (t) + s− (t)
= s̃(t)ej2πfc t + s̃∗ (t)e−j2πfc t

In the frequency domain we show the relationship between the complex baseband
signal S̃(f ), the analytic signal S+ (f ) and the real signal S(f )

S+ (f ) = S̃(f − fc )
INSTRUCTOR COPY

S− (t) = S̃ ∗ (−(f + fc ))
2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))

Since s̃(t) is a complex functions of time, we write in the time domain

s̃(t) = a(t)ejφ(t) = I(t) + jQ(t)

s̃∗ (t) = a(t)e−jφ(t) = I(t) − jQ(t)
s+ (t) = a(t)ejφ(t) ej2πfc t
s− (t) = a(t)e−jφ(t) e−j2πfc t

Thus

2s(t) = s+ (t) + s− (t)

= a(t)ejφ(t) ej2πfc t + a(t)e−jφ(t) e−j2πfc t
= 2a(t) cos[2πfc t + φ(t)]
s(t) = <{s+ (t)}
= <{s− (t)}
= a(t) cos[2πfc t + φ(t)]

The factor of 2 arises from

ejθ + e−jθ
cos θ =
2
In the frequency domain, S̃(f ) is a complex function of f that may be written in
rectangular or polar form.
 CHAPTER 2. IQ SIGNALS 155

2.8 Application to sampled digital systems (soft-

ware radio systems)

2.8.1 Discrete Fourier Transform

For a software radio system where the signal is digitized (sampled) by an analog-
to-digital converter (ADC) at a sampling rate fs = 1/Ts , the Fourier transform is
done using the Discrete Fourier Transform (DFT) algorithm on a block of N sam-
ples. The DFT is most often implemented using the Fast Fourier transform (FFT)
algorithm.The FFT is reviewed in detail in section 1.4.

Recall the discrete Fourier transform (DFT) is defined as a Fourier transform

INSTRUCTOR COPY

operating on a sampled periodic signal with period N samples.

The real time domain signal s(t) is sampled at a sampling rate fs = 1/Ts to make
samples sn = s(t = nTs ), where Ts is the time resolution. After the discrete Fourier
transform is done on a block of N real time domain samples sn , 0 ≤ n ≤ N − 1 to
obtain the frequency domain signal S(f ), the result is a block of N complex frequency
domain samples Sk , 0 ≤ k ≤ N −1, where Sk = S(f = kf0 ) is the frequency resolution
f0 = fs /N .

The FFT calculation may be done in Matlab Sk = f f t(sn ) where the blocks of N
samples sn , Sk are Matlab vectors.

The block of N complex frequency domain samples is symmetrical about N/2.

The DFT takes N unique real time domain samples and produces N/2 unique com-
plex frequency domain samples or N unique real numbers. For real signals s(t), the
sampling rate must be at least twice the highest frequency in s(t).

Since ej2πkn/N = ej2πk(n+N )/N , we can show that the magnitude spectrum of a
real signal is symmetrical about zero frequency, i.e. the magnitude at a negative
frequency −kf0 = −kfs /N for k > 0 is exactly the same as that of a positive frequency
kf0 = kfs /N .

If we now consider a complex signal s̃(t) (with non-zero imaginary part) in the
DFT calculation, then the N unique complex time domain samples will yield N unique
complex frequency domain samples. This shows that the spectrum of a complex signal
is asymmetrical about zero frequency, i.e. the amplitude and phase at a negative
frequency −kf0 = −kfs /N for k > 0 is different from that that of the corresponding
positive frequency kf0 = kfs /N .

Assume that the bandwidth of the complex signal is B, covering the range −B/2 ≤
 CHAPTER 2. IQ SIGNALS 156

f ≤ B/2. For complex signals, just as for real signals, the sampling rate fs > 2(B/2)
must be at least twice the highest frequency B/2 in s̃(t). Note that since the complex
signal includes both positive and negative frequencies, the minimum sampling rate fs
for a complex signal is equal to the bandwidth B of that complex signal.

2.8.2 Frequency shifting by subsampling

A complex passband signal s+ (t) = s̃(t)ej2πfc t at carrier frequency fc whose complex

baseband s̃(t) has bandwidth B  fc can be sampled at minimum rate fs = B and
does not need to be sampled at minimum rate fs = fc + B. The same result applies
to a real passband signal s(t).
INSTRUCTOR COPY

If s(t) was already sampled at fs > fc + B then the sampled sn = s(t = n/fs ) may
be subsampled or decimated by retaining only one out of N samples, where N = fs /B
and all others samples are discarded.

In either case, the output of the subsampling operation is the sampled complex
baseband signal. Thus, in effect, the passband signal has been frequency-shifted down
to baseband.

This is an example of multi-rate signal processing where signals are sampled at

different rates in different parts of the system.

2.8.3 Complex IQ sampling

The I and Q components (i.e. the complex baseband) of a passband signal s(t) or
s+ (t) may be obtained by multiplying by e−j2πfc t . If we choose the sampling rate
fs = 4fc or ts = 1/(4fc ), then the complex exponential takes on only the values

e−j2πfc t |t=nts = e−j2πfc nts = e−j2πfc n/(4fc ) = e−jnπ/2 = 1, j, −1, −j (2.45)

Thus every second real sample is zero and every second imaginary sample is zero.
The result is that the samples at fs = 4fc alternate between I, Q, -I, -Q and thus the
I and Q samples are interleaved.

This sampling technique is useful in the next section on transmitter and re-
ceiver block diagrams. Further details appear at https://www.dsprelated.com/
showarticle/153.php
 CHAPTER 2. IQ SIGNALS 157

2.9 Transmitter and receiver block diagrams

Learning objectives

In this section, we show

• block diagrams that implement the equations from the previous section, including
• transmitters that use the message signal s̃(t) to modulate a carrier wave to
obtain the signals s+ (t) or s(t), and
• receivers that demodulate the signals s+ (t) or s(t) to recover the message signal
INSTRUCTOR COPY

s̃(t).

Overview

The implementation of the equations in section 2.1 may be done in software or hard-
ware.

In the case of software, the complex signal equations are implemented directly in
some language such as C++ and Python (for GNURadio) or in Matlab, Mathematica
or similar. The signals must be represented by discrete-time samples, sampled at some
sampling rate fs = 1/Ts such that

• s̃(t)|t=nTs = s̃(t = nTs ) = s̃n for a complex baseband signal,

• s+ (t)|t=nTs = s+ (t = nTs ) = s+,n for a complex passband signal,
• s(t)|t=nTs = s(t = nTs ) = sn for a real passband signal.

The samples are stored in digital memory as a vector and manipulated in software.

In the case of hardware, complex signals cannot be implemented directly from

the equations since a complex signal cannot be represented by a real current that
flows through a wire. The complex signal equations must be converted to real signals
(voltage) that can make currents flow on wires.

Real signals can be implemented in either hardware or software.

The equations of section 2.1 are used to implement a communications system

(fig. 2.16) that includes a
 CHAPTER 2. IQ SIGNALS 158

Figure 2.16: IQ transmitter-receiver system

• transmitter (modulator) that modulates the information-bearing complex base-

band signal (message) s̃(t) = I(t)+jQ(t) onto a carrier. The transmitter output
is a communications signal, either real passband s(t) or complex passband s+ (t)
INSTRUCTOR COPY

• channel that may be a real analog thing such as optical fiber, copper wire, trans-
mission line or antenna that radiates the signal into free space (wireless) or a
simulated channel that is implemented in software. A simulated channel may
be complex or real. The channel has a path loss L0 that attenuates the commu-
nications signal as a result of the distance between transmitter and receiver.The
channel will include noise. The channel may also include a time-varying filter
that represents distortion caused by the physical nature of the channel. In this
section, we assume that the channel adds a path loss factor 1/L0 , so that the
channel output r(t) = s(t)/L0 or r+ (t) = s+ (t)/L0 .

• receiver (demodulator) that extracts the message r̃(t) from the received commu-
nications signal r(t) = s(t)/L0 or r+ (t) = s+ (t)/L0 . The objective of the com-
munications system is for the message to be received correctly, i.e. r̃(t) = s̃(t).

In fig. 2.16 and sections 2.9.1 to 2.9.4, we show block diagrams of the transmitter
and receiver.

2.9.1 Radio transmitter (modulator)

2.9.1.1 Complex signal implementation - complex passband output

The radio transmitter (modulator) may be described in complex form with a complex
multiplication of the complex baseband message s̃(t) = I(t) + jQ(t) with the complex
 CHAPTER 2. IQ SIGNALS 159

exponential ej2πfc t = cos 2πfc t + j sin 2πfc t to yield a complex passband signal
s+ (t) = s̃(t)ej2πfc t
= a(t)ejφ(t) ej2πfc t (2.46)
= [I(t) + jQ(t)][cos 2πfc t + j sin 2πfc t]

The complex message I(t), Q(t) is multiplied by the complex carrier wave ej2πfc t =
cos 2πfc t + j sin 2πfc t to create a complex passband signal s+ (t) comprising only
positive frequencies.

In the frequency domain, from eq. (2.46) we have shifted the complex baseband
signal up in frequency by fc
S+ (f ) = S̃(f − fc )
INSTRUCTOR COPY

The complex signal implementation diagram in fig. 2.17 assumes the blocks are
processing sampled digital signals in software.

s+ (t)
I(t) + jQ(t)

Real
ej2πfc t Complex

Figure 2.17: Complex transmitter block diagram which generates s+ (t)

2.9.1.2 Complex signal implementation - real passband output

If the transmitter is intended to transmit the signal via a real channel, then the
real radio signal s(t) may be obtained from the positive frequency complex passband
signal s+ (t) by adding the negative frequencies s− (t) to obtain
2s(t) = s+ (t) + s− (t)
 CHAPTER 2. IQ SIGNALS 160

as recalled from eq. (2.25).

In the frequency domain, recall eq. (2.40) as

2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))
We see from the above that adding the negative frequencies is not simply shifting the
complex baseband signal down in frequency. The complex baseband signal must be
complex conjugated and “flipped” to make a mirror image before shifting it down.

Exercise 35
Expand the block diagram fig. 2.17 to show how to generate the real output
s(t) by adding the negative frequencies.
INSTRUCTOR COPY

Solution
To obtain the real passband signal s(t) from the complex passband signal s+ (t)
requires taking the real part of the signal (equivalent to adding the negative
frequencies) and passing the result through a DAC (digital-to-analog converter)
as shown in fig. 2.18.

D
Re s(t)
A
I(t) + jQ(t)

Real
ej2πfc t Complex

Figure 2.18: Complex transmitter block diagram which generates s(t)

Equivalently, the same result may be obtained mathematically by taking the real
part of this complex multiplication
s(t) = <{s+ (t)}
(2.47)
= I(t) cos 2πfc t − Q(t) sin 2πfc t
The method of writing a real signal s(t) as the real part of a complex signal s+ (t) is
mathematically equivalent to adding the negative frequencies and was introduced in
section 2.1.
 CHAPTER 2. IQ SIGNALS 161

If we need a real signal to transmit over a wire to an antennas, then we need a

digital-to-analog converter (DAC) block after the Re block.

If we are building a computer simulation of the radio transmitter in discrete-time

(sampled) form, and don’t need to convert it to analog and transmit it over a real
channel (fiber, wire), then the Re block and DAC may be omitted, and the output is
the complex passband signal s+ (t).

The computer simulation can implement the complete system (transmitter, chan-
nel and receiver) in complex passband using only positive frequencies. There is no
need to repeat these calculations for the negative frequencies, since the result will be
the same.
INSTRUCTOR COPY

2.9.1.3 Real signal implementation - real passband output

In this subsection we implement the transmitter using only real signals, so that the
complex baseband signal s̃(t) = I(t) + jQ(t) is split into two separate real signals I(t)
and Q(t). The implementation follows eq. (2.47) above.

D
I
A

cos(2πfc t)
s(t)
sin(2πfc t)
−

D
Q
A

Figure 2.19: Real transmitter block diagram which generates s(t) from a complex
baseband input.

Figure 2.19 shows the block diagram for the real transmitter which contains the
following blocks:

1. two real signal inputs I(t) and Q(t),

2. the D/A block is a digital to analog converter,
 CHAPTER 2. IQ SIGNALS 162

3. the mixer is a multiplier,

4. the oscillators output a cos wave to be multiplied by I(t) and outputs a sin
wave to be multiplied by Q(t).

5. the combiner is an adder.

Remember from fig. 2.17 that a single solid line means the signal is real while a
double solid line means that the signal is complex.

With the blocks defined as above, the figure shows a radio transmitter (or mod-
ulator) that produces the radio waveform s(t) = I(t) cos 2πfc t − Q(t) sin 2πfc t from
the message signals I(t), Q(t).
INSTRUCTOR COPY

In fig. 2.19, the blocks process analog signals. However, if the D/As are omitted,
the remaining blocks process digital signals, using software such as GNURadio or
Matlab Simulink.
Exercise 36
In fig. 2.19 replace the cos and sin oscillators with complex exponentials and
show that the output is s(t).

2.9.2 Radio receiver (demodulator)

d We now consider the demodulator/receiver half of fig. 2.16.

If the entire communication system including channel is implemented by computer

simulation, then the complex passband signal s+ (t) is attenuated by the channel
represented by a scaling factor 1/L0 .

If the channel is real (wired or wireless), then the real passband signal s(t) is
transmitted over a distance via some real channel (wired or wireless), attenuated by
the path loss L0 and picked up by a receiver in the form r(t) = s(t)/L0

The receiver’s task is to recover the message signals s̃(t) = I(t) + jQ(t) from the
signal r(t). This can be done using the receivers shown below.
 CHAPTER 2. IQ SIGNALS 163

2.9.2.1 Complex signal implementation - complex passband input

The radio receiver (demodulator) may be described in complex form with a complex
multiplication of the complex passband signal r+ (t) = L−1
0 s+ (t) = a(t)e
jφ(t) j2πfc t
e /L0 .
If we are building a computer simulation of the radio receiver in discrete-time (sam-
pled) form, and don’t need to work with an analog input that was received over a
real channel (fiber, wire), then the ADC may be omitted, and the receiver input is
the complex passband signal s+ (t).

The computer simulation can implement the complete system (transmitter, chan-
nel and receiver) in complex passband using only positive frequencies. There is no
need to repeat these calculations for the negative frequencies, since the result will be
the same.
INSTRUCTOR COPY

We write the received signal in complex passband form

r+ (t) = L−1
0 s+ (t)

= L−1
0 a(t)e
jφ(t) j2πfc t
e (2.48)
−1
= L0 [I(t) + jQ(t)][cos 2πfc t + j sin 2πfc t]

The receiver demodulates the complex radio signal by multiplying r+ (t) by the
complex local oscillator e−j2πfc t = cos 2πfc t − j sin 2πfc t to yield

r+ (t)e−j2πfc t = [L−1
0 a(t)e
jφ(t) j2πfc t −j2πfc t
e ]e
= L−1
0 a(t)e
jφ(t)
(2.49)
= L−1
0 [I(t) + jQ(t)]
= L−1
0 s̃(t)

Thus the receiver output is the original complex baseband signal s̃(t) that contains
the message.

The complex signal implementation diagram in fig. 2.20 assumes the blocks are
processing a sampled complex passband digital signal r+ (t) in software to yield a
sampled complex baseband digital signal s̃(t) = I(t) + jQ(t)

In the frequency domain

 CHAPTER 2. IQ SIGNALS 164

r+ (t)
I(t) + jQ(t)

e−j2πfc t
Figure 2.20: Complex receiver block diagram with complex input r+ (t).
INSTRUCTOR COPY

R+ (f ) = L−1
0 S+ (f )

= S̃(f − fc )
R+ (f + fc ) = L−1
0 S+ (f + fc )

= L−1
0 S̃(f − fc + fc ))

= L−1
0 S̃(f )

2.9.2.2 Real signal implementation - real signal input

Recall the real passband signal

s(t) = <{s+ (t)}

= <{a(t)ejφ(t) ej2πfc t }
= a(t) cos[2πfc t + φ(t)]

which is the real part of the complex passband signal.

The block diagram in fig. 2.21 is the dual of the one in fig. 2.19. In this receiver,
the input signal and all processing is analog, and I(t), Q(t) are digitized by an analog-
to-digital converter (A/D or ADC).

A more complete block diagram in fig. 2.22 adds some practical components to
the receiver including: RX filters needed to filter out undesired signals on nearby
frequencies, a Low Noise Amplifier (LNA) to amplify r(t) that is typically in the
microvolt range to a level in the volt range suitable for ADC, Automatic Gain Control
(AGC) to adjust the gain to compensate for variations in the level r(t) and low pass
filters (LPF) before the ADC.
 CHAPTER 2. IQ SIGNALS 165

A
I
D

cos(2πfc t)
r+ (t)
sin(2πfc t)
INSTRUCTOR COPY

A
Q
D

Figure 2.21: Radio receiver block diagram

Figure 2.22: Practical receiver

 CHAPTER 2. IQ SIGNALS 166

To see how this receiver works, we calculate the signals x(t), y(t) at the two ADC
inputs, and find that they are equal to I(t), Q(t)

Exercise 37
Prove that the ADC inputs are equal to I(t), Q(t). Several trigonometric iden-
tities will be needed from , repeated here for convenience

cos α cos β = [cos(α − β) + cos(α + β)]/2

sin α cos β = [sin(α − β) + sin(α + β)]/2
sin α sin β = [cos(α − β) − cos(α + β)]/2

Note that double frequency terms (cosine waves at 2fc ) arise from the calcula-
tions and are filtered out by the low pass filter (LPF).
INSTRUCTOR COPY

Exercise 38
Consider a receiver similar to fig. 2.21 where the ADC is located immediately
after the r(t) input. Write r(t) = 0.5r+ (t) + 0.5r− (t) and draw a block diagram
of a receiver that processes the positive and negative frequencies separately to
obtain s̃(t) = I(t) + jQ(t). Hint, recall

2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))

2.9.3 Summary

We have created a communication system with message signals I(t), Q(t) that are
modulated onto a carrier wave at frequency fc to create the radio signal

s(t) = I(t) cos 2πfc t − Q(t) sin 2πfc t.

s(t) travels over a distance via a channel with path loss Lo .

The receiver picks up the real passband signal r(t) = s(t)/L0 and recovers the
messages s̃(t) = I(t) + jQ(t).

When the communications system is simulated in software, we can implement it

in complex passband r+ (t) = s+ (t)/L0

Any signal s(t) can be written as a carrier wave at frequency fc with time-varying
 CHAPTER 2. IQ SIGNALS 167

amplitude and phase, i.e.

s(t) = a(t) cos[2πfc t + φ(t)]
= <{a(t)ejφ(t) ej2πfc t }
= <{[I(t) + jQ(t)][cos 2πfc t + j sin 2πfc t]}
= I(t) cos 2πfc t − Q(t) sin 2πfc t
= a(t) cos[2πfc t + φ(t)]
where
I(t) = a(t) cos φ(t) = <{a(t)ejφ(t) }
Q(t) = a(t) sin φ(t) = ={a(t)ejφ(t) }
and
INSTRUCTOR COPY

s̃(t) = I(t) + jQ(t)

= a(t)ejφ(t)
is called the complex envelope of the signal and also called the complex baseband
signal. The complex envelope contains two real waveforms I(t), Q(t) that contain
the information or message.

The real passband signal s(t) is obtained by multiplying the complex envelope
s̃(t) with the complex positive frequency carrier wave
c+ (t) = ej2πfc t = cos 2πfc t + j sin 2πfc t (2.50)
to yield
s+ (t) = s̃(t)ej2πfc t (2.51)
and taking the real part to yield
s(t) = <{s̃(t)ej2πfc t } (2.52)

Exercise 39
Obtain the real passband signal from s̃(t) and c+ (t) by adding the negative fre-
quency carrier wave, c− (t) = e−j2πfc t , without taking the real part of anything.

2.9.4 Summary 2

Here we review the general I-Q receiver configuration that may be implemented for
all types of signals using slightly different notation. Every type of signal consists of
 CHAPTER 2. IQ SIGNALS 168
INSTRUCTOR COPY

Figure 2.23: General IQ Transmitter

a standard configuration coupled with a “map” that is specific to the transmission

type.

s(t) = a(t) cos[2πfc t + ψ(t)]

= a(t)[cos(2πfc t) cos ψ(t) − sin(2πfc t) sin ψ(t)]
= a(t) cos ψ(t)cos2πfc t − a(t) sin ψ(t) sin 2πfc t
= vi (t)cos2πfc t − vq (t) sin 2πfc t
where a(t) is the amplitude and ψ(t) is the phase. The coefficients of the carrier
waves in the transmitter are referred to as vi (t) and vq (t) respectively, where
vi (t) = a(t) cos ψ(t)
vq (t) = a(t) sin ψ(t)
These signals are formed by the map that converts the message m(t) into vi (t), vq (t)
and used to create s(t) at the IQ transmitter.
s(t) = vi (t) cos(2πfc t) − vq (t) sin(2πfc t)
The exact form of the map depends on the particular modulation scheme chosen, as
will be discussed in later chapters.

The IQ Receiver multiplies the incoming signal s(t) by two versions of the carrier
wave functions: cos(2πfc t) and the 90◦ phase-shifted version sin(2πfc t).

Again, the form of the map that converts vi (t), vq (t) to the message m(t) depends
on the particular modulation scheme chosen, as will be discussed in later chapters.
 CHAPTER 2. IQ SIGNALS 169

Figure 2.24: General IQ Receiver

INSTRUCTOR COPY

2.10 Analog IQ receiver

Learning objectives
• To describe the operation of a practical software defined radio.

• To describe the implementation of a receiver that receives real analog signals

and processes them both in the analog and digital domain.

• To describe the connection between the block diagram of a software defined

radio and the mathematical description of communications signals

2.10.1 Overview and block diagram

In this section, we apply the I-Q signal theory above to a particular example:

A block diagram for the Ettus USRP N210 radios used in the ECE 350 labs is
shown in fig. 2.25. For this receiver, there is an analog daughterboard that is chosen
for the particular frequency range of interest (red and green lines) and a digital
motherboard.

The daughterboard receives signals with carrier frequency in the range 50–2200 MHz
and generates analog I-Q outputs that are in the frequency range below 50 MHz and
are sampled at 100 MHz.
 CHAPTER 2. IQ SIGNALS 170
INSTRUCTOR COPY

Figure 2.25: Ettus N210 block diagram (www.ettus.com)

In fig. 2.25:

• The red blocks and connections represent real passband signals

• the green blocks and connections represent the real and imaginary parts of
complex baseband signals

• The blue blocks and connections represent digital processing in the software
defined part of the radio receiver.

Many software defined radios that operate at frequencies above about 100 MHz
have an analog I-Q stage similar to this example ahead of the analog-to-digital con-
verter sampling at 100 MHz. The red oscillator (labelled ”c” in fig. 2.25) operates at
a frequency fLO called the local oscillator frequency, which is general is close to but
not the same as the carrier frequency fc .

In the next section, we describe the operation of the daughterboard (red and green
blocks and connections).
 CHAPTER 2. IQ SIGNALS 171

2.10.2 USRP daughterboard receiver operation

The USRP daughterboard is designed to receive radio frequency (RF) signals at

frequencies fc in the range fLO ± fs /2 MHz, where fLO is the local oscillator (LO)
frequency. In this example, we set fLO to about 100 MHz and fs is the sampling rate
of the USRP main board ADCs, set to 100 MS/s.

The USRP daughterboard operates by generating two local oscillator (LO) signals
at fLO and mixing (multiplying) it with a desired radio frequency (RF) signal at fc
(picked up by the antenna or fed in by a signal generator) to yield a signal at the
difference frequency fb = fc − fLO .

The USRP daughterboard has two local oscillators operating 90◦ out of phase,
INSTRUCTOR COPY

cos 2πfLO t and − sin 2πfLO t and two mixers. Thus there are two receiver outputs
that we call I(t) and Q(t) as shown in fig. 2.26.

I(t)
r(t)

cos(2πfc t)

ϕ 90◦ phase shift

Q(t)

Figure 2.26: USRP daughterboard block diagram

The desired RF signal that we wish to receive is written r(t) = a(t) cos(2πfc t +
φ(t)).

We assume a(t) = 1, φ(t) = 0 for the moment, so the desired RF signal is simply
an unmodulated carrier wave r(t) = cos(2πfc t).
 CHAPTER 2. IQ SIGNALS 172

2.10.2.1 Receiver operation in real notation

In what follows, given this RF signal input we will calculate the two receiver outputs
I(t) and Q(t).

To do this, we will use some trigonometric identities from , repeated here for
convenience.

cos α cos β = [cos(α − β) + cos(α + β)]/2

sin α cos β = [sin(α − β) + sin(α + β)]/2
sin α sin β = [cos(α − β) − cos(α + β)]/2
INSTRUCTOR COPY

One of the local oscillator signals is written cos(2πfLO t). The cosine mixer mul-
tiplies this LO signal with the RF input signal to obtain

cos 2πfc t · cos 2πfLO t

Using one of the trigonometric identities, we can write

cos 2πfLO t · cos 2πfc t = 0.5 cos 2π(fc + fLO )t + 0.5 cos 2π(fc − fLO )t

Thus multiplying two sine (or cosine) waves at frequencies fLO and fc results in
two new sine waves: one at the sum frequency fc + fLO and one at the difference
frequency fc − fLO .

The signal at the sum frequency fc + fLO is filtered out by analog low pass filters
in the USRP daughterboard.

The signal at the difference frequency fc − fLO is written

I(t) = 0.5 cos 2π(fc − fLO )t

= 0.5 cos 2πfb t

I(t) can be sampled by the ADC, provided that fc is close enough to fLO , i.e. the
difference is less than half the sampling rate, |fc − fLO | < fs /2 or fLO − fs /2 < fc <
fLO + fs /2

The difference frequency fb = fc − fLO , where |fb | < fs /2.

 CHAPTER 2. IQ SIGNALS 173

The two USRP daughterboard output that we call I(t) is now I(t) = 0.5 cos 2π(fc −
fLO )t. The second local oscillator signal is written − sin 2πfLO t. The sine mixer mul-
tiplies this LO signal with the RF input signal to obtain

cos 2πfc t · (− sin 2πfLO t).

We apply a trigonometric identity with α = 2πfLO t, β = 2πfc t and write

− sin 2πfLO t · cos 2πfc t = −0.5 sin 2π(fc + fLO )t + 0.5 sin 2π(fc − fLO )t (2.53)

The signal at the difference frequency is written

Q(t) = 0.5 sin 2π(fc − fLO )t

INSTRUCTOR COPY

= 0.5 sin 2πfb t

If the receiver outputs I(t) = cos 2πfb t and Q(t) = sin 2πfb t are displayed on a
x-y scope, then a circle is displayed. If fb < 5 Hz or so, then the dot on the scope can
be seen tracing out the circle. We can write these two receiver output signals I(t)
and Q(t) that have time varying amplitude and phase a(t), φ(t), where in this case
p
a(t) = I 2 (t) + Q2 (t)
=1
Q(t)
φ(t) = arctan
I(t)
sin 2πfb t
= arctan
cos 2πfb t
= arctan(tan 2πfb t)
= 2πfb t
r̃(t) = a(t)ejφ(t)
= ej2πfb t

The last equation shows how I(t) and Q(t) can also be represented as one complex
signal r̃(t) = I(t) + jQ(t) = a(t)ejφ(t) as shown in the next section.

2.10.2.2 Receiver operation in complex notation

We can calculate the USRP daughterboard receiver output using complex signals as
follows.
 CHAPTER 2. IQ SIGNALS 174

Recall that the desired RF signal that we wish to receive is written r(t) =
a(t) cos[2πfc t + φ(t)] where for the moment we set the path loss L0 = 1.

We assume a(t) = 1, φ(t) = 0 for the purpose of this section, so the desired RF
signal is simply an unmodulated carrier wave cos 2πfc t. In complex notation we write
r(t) = <{r+ (t)}
= Re{ej2πfc t }
= cos 2πfc t
We can also write r(t) in terms of the positive and negative frequency components.
r(t) = r+ (t)/2 + r− (t)/2 (2.54)
INSTRUCTOR COPY

In complex notation, we consider only the positive frequency signal input to the
receiver. The complex passband unmodulated RF signal r+ (t) = ej2πfc t is multiplied
by the complex local oscillator
e−j2πfLO t| = cos 2πfLO t − j sin 2πfLO t (2.55)
to yield
r+ (t)e−j2πfLO t = ej2πfc t e−j2πfLO t
= ej2πfb t
= I(t) + jQ(t)
= r̃(t)
where the received complex baseband signal is
r̃(t) = I(t) + jQ(t)
= ej2πfb t

r+ (t)
I(t) + jQ(t)

e−j2πfc t
Figure 2.27: Complex signal receiver block diagram, USRP daughterboard analog IQ
receiver in complex notation, in this diagram fc = fLO
 CHAPTER 2. IQ SIGNALS 175

The diagram in fig. 2.27 using complex signals performs the same function as the
previous diagram above using real signals. Note that the complex signal diagram
does not use the low pass filters.

Why? The reason is that the multiplication of the real signal by a cos or sin oscil-
lator results in both the sum and difference frequencies and the sum frequency must
be filtered out. The multiplication of the complex signal by a complex exponential
results in only the difference frequency and thus no need to filter out anything.

In summary,

r+ (t) = ej2πfc t
I(t) = cos 2πfb t
INSTRUCTOR COPY

Q(t) = sin 2πfb t

r̃(t) = I(t) + jQ(t)
= cos 2πfb t + j sin 2πfb t
= ej2πfb t

This is the same result as above, apart from a factor 0.5 arising from the complex
notation.

If I(t) and Q(t) are displayed on a x-y scope, a circle is displayed. If fb < 5 Hz or
so, then the dot on the scope can be seen tracing out the circle.

Exercise 40
Repeat the above analysis in both real and complex notation when the RF
input signal is a general signal a(t) cos[2πfc t + φ(t)] to find I(t) and Q(t) as a
function of fb , a(t), φ(t).

Exercise 41
Repeat the complex notation analysis for a real input signal

r(t) = cos 2πfc t

Hint: write r(t) as the sum of two complex exponentials. Some filters may be
needed.
 CHAPTER 2. IQ SIGNALS 176

Exercise 42
Consider a communications system where the transmitter generates both s+ (t)
and s− (t) separately and does not add them together to make the real signal
s(t). Instead, the communications system contains two receivers, one for s+ (t)
and one for s− (t). How that the two receivers yield the same identical output.

2.10.3 Another approach using digital IQ receiver

Figure 2.28 shows an alternative receiver structure where the analog stage has only
one cosine oscillator at a frequency fLO close to the desired signals near fc followed
by a low pass filter. In this structure, the single analog oscillator shifts the received
INSTRUCTOR COPY

bandwidth down by fLO so that it is centered at 0 Hz. The low pass filter passes
through the difference term at fc − fLO and filter out the sum term at fc + fLO .

The signals I(t), Q(t) are generated after the analog-to-digital converter using
exactly the same math as shown in sections 2.10.2.1 to 2.10.2.2.

Figure 2.28: RTL-SDR block diagram (https://www.rtl-sdr.com/

about-rtl-sdr/)
 CHAPTER 2. IQ SIGNALS 177

2.10.4 Super-heterodyne Receiver

In this section, we further describe the analog section of fig. 2.28 (highlighted in
orange). in particular the consequences of using a single cosine oscillator instead of
the two oscillators (cosine and sine) in figs. 2.25 and 2.26.

The analog stage applies amplification as the typical amplitude range at the re-
ceiver input from the antenna is on the order of microvolts. The ADC requires peak
voltage input levels on the order of Volts. Thus a gain of 106 or 120 dB is required.
This gain may be achieved using a series of op amps. However, such a large gain
requirement around a single carrier frequency can raise practical issues such as the
occurrence of feedback due to the radiation from the circuit board traces because of
the high frequencies (short wavelengths) needing to be amplified.
INSTRUCTOR COPY

2.10.4.1 Superheterodyne receiver block diagram

A so-called super-heterodyne receiver offers a means of overcoming such issues by

shifting the carrier (usually down in frequency) and effectively splitting the gain
requirements over two frequencies: (1) the original carrier and (2) the so-called in-
termediate frequency (fIF . Thus the gain requirement at each frequency is reduced
to 60 dB. This helps to prevent feedback.

Another advantage of using the superheterodyne design is that it allows the re-
ceiver to shift any carrier frequency to an industry standard. This allows the com-
ponents of the standardized IQ receiver to be mass produced at a very low cost.
Amplifiers that support higher frequencies tend to have a narrower bandwidth as
well so by shifting the carrier we also remove this limitation. The architecture of
the super-heterodyne is shown in fig. 2.29 and is equivalent to the analog section in
fig. 2.28. 5
5
A small but significant difference is that in fig. 2.29 the RF filter is ahead of the RF amplifier,
whereas in fig. 2.28 the RF amplifier is ahead of the RF filter. Locating the RF filter ahead of the RF
amplifier is preferred, since it prevents strong signals outside the filter bandwidth from overloading
the amplifier.
 CHAPTER 2. IQ SIGNALS 178

Figure 2.29: Super-heterodyne Receiver

INSTRUCTOR COPY

2.10.4.2 Frequency shift up and down

The consequence of the decision to use a single cosine oscillator instead of two os-
cillators (cosine and sine) is that the input signal frequency is shifted both up and
down.

The single cosine oscillator contains both positive and negative frequencies
c(t) = cos 2πfLO t
ej2πfLO t + e−j2πfLO t
=
2
The receiver input signal
r(t) = a(t) cos[2πfc t + φ(t)]
= cos 2πfc t
since we assume a(t) = 1 and φ(t) = 0 as in the previous section so that in the
frequency domain
R(f ) = 0.5δ(f − fc ) + 0.5δ(f + fc )

Thus the output of the mixer (multiplier) is

rIF (t) = r(t)c(t)
= 0.5r(t)ej2πfLO t + 0.5r(t)e−j2πfLO t
In the frequency domain
RIF (f ) = 0.5R(f − fLO ) + 0.5R(f + fLO )
 CHAPTER 2. IQ SIGNALS 179

Exercise 43
Find RIF (f ) in terms of δ functions and sketch a figure.

Solution
The mathematical solution is

R(f ) = 0.5δ(f − fc ) + 0.5δ(f + fc )

RIF (f ) = 0.5R(f − fLO ) + 0.5R(f + fLO )
= 0.5[0.5δ(f − fc − fLO ) + 0.5δ(f + fc − fLO )]
+ 0.5[0.5δ(f − fc + fLO ) + 0.5δ(f + fc + fLO )]
INSTRUCTOR COPY

Thus the received signal R(f ) at fc is shifted both up and down by fLO to
fc ± fLO . R(f ) is shifted up by the positive frequency component of c(t) and
is shifted down by the negative frequency component of c(t).
This result can also be seen in the time domain using a trigonometric identity

rIF (t) = r(t)c(t)

= cos 2πfc t cos 2πfLO t
= 0.5 cos 2π(fc + fLO )t + 0.5 cos 2π(fc − fLO )t

However, this method of showing the result hides the action of positive and
negative frequencies in the real signal r(t) and the real cosine oscillator c(t).

2.10.4.3 Receiver tuning

The receiver is tuned to receive different carrier frequencies by changing the frequency
of the so-called Local Oscillator (as distinct from the Remote oscillator in the trans-
mitter some distance away from the receiver). The tuning knob on a radio receiver
controls the frequency of the Local Oscillator. A common intermediate frequency
used for AM receivers is 455 kHz and for FM receivers is 10.7 MHz. The IF frequency
for the RTL receiver is 3.57 MHz which is then sampled at 28.8 MHz.

Refer again to .

Once fIF is selected, we need to adjust the local oscillator frequency so as to select
(tune in) the desired carrier frequency. Thus we adjust

fLO = fc ± fIF
 CHAPTER 2. IQ SIGNALS 180

We may choose either + or −.

With the local oscillator adjusted to either of these two frequencies we get the
following 4 possible output frequencies: fc +(fc ±fIF ) = 2fc ±fIF and fc −(fc ±fIF ) =
∓fIF . We then bandpass filter out the high frequency term 2fc ± fIF and thus
we have successfully shifted the signal at fc to a signal at a desired intermediate
frequency±fIF .

The tuning process is considered in more detail in section 2.11.

2.10.4.4 Image frequency

INSTRUCTOR COPY

An issue with the super-heterodyne receiver which is important to note is the ex-
istence of an “image” frequency (fIM ). If there is a signal at the receiver input at
this frequency fIM and is not filtered out, then it will be shifted to the same inter-
mediate frequency fIF , along with the signal at the desired fc , and will cause direct
interference.

If fLO = fc ± fIF then fIM = fLO ± fIF . It is often best to consider specific
numbers rather than formulas with plus/minus signs. For example, consider fRF =
fc = 1070 kHz and fIF = 455 kHz.

Then we can choose fLO = 1070 − 455 = 615 kHz, so that the incoming signal at
1070 kHz mixes with the LO at 615 kHz to yield an output at the 455 kHz IF. Thus
by setting the LO to 615 kHz we have “tuned in” the signal at 1070 kHz.

The image frequency will be 615 − 455 = 160 kHz, since an incoming (image)
signal at 160 kHz can also mix with the same LO at 615 kHz to yield an output at
the 455 kHz IF.

We can also choose fLO = 1070 + 455 = 1525 kHz, so that the incoming signal at
1070 kHz mixes with the LO at 1525 kHz to yield an output at the 455 kHz IF. Thus
by setting the LO to 1525 kHz we have “tuned in” the signal at 1070 kHz.

In this case, the image frequency will be 1525+455 = 1980 kHz, since an incoming
(image) signal at 1980 kHz can also mix with the same LO at 1525 kHz to yield an
output at the 455 kHz IF.
 CHAPTER 2. IQ SIGNALS 181

2.10.4.5 Superheterodyne with complex signals

The superheterodyne principle can also be applied to complex signals. For example,
in the USRP, the daughterboard is mathematically a complex local oscillator that
downconverts a slice of radio frequencies centered around fLO to a zero frequency IF
complex baseband signal (with asymmetrical spectrum). This signal is sampled by
the main USRP board. In the case of complex signals, the pair of real oscillators
in the daughterboard function mathematically as a single local oscillator which is a
complex exponential, and the frequency of the input signal is shifted up or down
(usually down for a receiver), not up and down.
INSTRUCTOR COPY

Figure 2.30: USRP diagram

2.10.4.6 Real signal analysis of IQ mixer

The complex mixer that multiplies the input by a complex exponential may be im-
plemented with real cosine and sine oscillators, as shown in fig. 2.30.

Exercise: Do the analysis of the daughterboard function with real signal input to
show that the image frequencies generated by the real cosine and sine oscillators are
cancelled out.
 CHAPTER 2. IQ SIGNALS 182

2.11 Radio tuning, selecting a particular signal

The USRP receiver (fig. 2.25) is designed to receive radio frequency (RF) signals at
any frequency fc in the range fLO ± fs /2 MHz, where fLO is the local oscillator (LO)
frequency. In this example, consider the complex signal at the output of the USRP
(that is connected to the computer via Ethernet and processed in software such as
GNURadio).

The GNURadio block diagram software includes a so-called USRP Source block
that controls the analog oscillator frequency shown in red on the USRP daughter
board as well as the NCO (nuerically controlled oscillator) frequency on the USRP
motherboard shown in blue. The value of fLO set in the USRP source block determines
the frequency of both oscillators fred and fblue such that fLO = fred + fblue . When
INSTRUCTOR COPY

selecting the frequency fLO in the USRP source block, the values of fred and fblue are
set automatically by the USRP hardware, so that we don’t need to think about them.

The USRP source block has a complex signal output and has the LO frequency and
sampling rate as parameters. This complex signal can be live from the USRP source
block output or a IQ file in 2 channel WAV format that was recorded previously. Note
that this LO frequency in the GNURadio software fblue is not the same as the LO
frequency fred in the USRP daughterboard mentioned in an earlier section.

In this example, we choose fLO = 102 MHz and fs = 2 MHz, so that the frequency
range on the spectrum or waterfall is 101–103 MHz, which is part of the FM broadcast
band. Selecting a signal (radio station) at a particular frequency fc is called “tuning”
the radio. In this example we want to tune in the radio station at 101.9 MHz.

The USRP source block operates by generating the two local oscillator signals to
make fLO = fred + fblue and mixing (multiplying) it with a desired radio frequency
(RF) carrier wave r+ (t) = ej2πfc t at fc to yield a complex baseband signal r̃(t) =
I(t) + jQ(t) at the difference frequency fb = fc − fLO , where we write

r+ (t)e−j2πfLO t = ej2πfc t e−j2πfLO t

= ej2πfb t
= cos 2πfb t + j sin 2πfb t
= I(t) + jQ(t)

I(t) and Q(t) are processed correctly provided that fc is close enough to fLO , i.e.
the difference is less than half the sampling rate, |fc − fLO | < fs /2 or fLO − fs /2 <
fc < fLO + fs /2. The difference frequency fb = fc − fLO , where |fb | < fs /2.
 CHAPTER 2. IQ SIGNALS 183

fs = 2 MHz

fb = −0.1 MHz

f
fLO − fs fc fLO fLO + fs
2 2
(101 MHz) (101.9 MHz) (102 MHz) (103 MHz)

(a) Passband

fs = 2 MHz

fb
INSTRUCTOR COPY

f
fLO − fs fc fLO fLO + fs
2 2
(−1 MHz) (−0.1 MHz) (0 MHz) (1 MHz)

(b) Baseband

Figure 2.31: Frequency tuning examples (frequency axes not to scale).

The USRP Source block’s function is to shift a 2 MHz wide slice of spectrum from
101–103 MHz centered at fLO = 102 MHz down to −1 to 1 MHz (positive and negative
frequencies centered around 0 Hz). The complex baseband signal r̃1 (t) = ej2πfb t can
represent positive and negative frequencies, since fb can be positive or negative and
|fb | < 1 MHz. The 1 MHz slice of spectrum may contain many different signals at
various frequencies within the 1 MHz range (recall waterfall plots mentioned earlier).
For this case, FM radio stations use carrier frequencies in 0.2 MHz steps. In the
frequency range 101–103 MHz, there are 10 stations at 101.1, 101.3 and 101.5 MHz
etc. up to 102.9 MHz. In this example, we wish to tune in the station at 102.9 MHz.

If the RF carrier wave is turned on and off to transmit information in e.g. Morse
code, then we wish to listen to (or digitally decode) the complex baseband signal
r̃1 (t) = cos 2πfb t + j sin 2πfb t and no other signals.

If fb is outside the audio range we want to listen to, or if fb is not the frequency
expected at the digital decoder input, then we multiply r̃1 (t) by a complex exponential
 CHAPTER 2. IQ SIGNALS 184

e−j2πfd t at frequency fd to obtain another complex baseband signal

r̃2 (t) = ej2πfb t e−j2πfd t

= ej2π(fb −fd )t
= ej2πfE t

at frequency fE = fb − fd , where fE is chosen to be the frequency for listening (or for

the decoder).

In effect, we have shifted the spectrum twice. We first shifted fc by fLO to obtain
fb and then shifted fb by fd to get the exact frequency fE we want to listen to (for
Morse code) or for a digital decoder. For most decoders other than the human ear
listening to Morse codes signals, fE = 0 so that fd = fb .
INSTRUCTOR COPY

This idea of two successive spectrum shifts will be seen again later when we discuss
the Weaver demodulator for single sideband receivers.

List of frequencies used above

• fs : sampling rate of USRP source block

• fLO : local oscillator frequency

• fc : carrier frequency of desired signal at radio frequency near 10 MHz (passband)

• (fLO − fs /2) < fc < (fLO + fs /2): passband frequency range

• fb = fc − fLO : desired signal obtained by converting to baseband

• (−fs /2) < fb < (fs /2): baseband frequency range (centered at 0 Hz)

• Conversion (spectrum shifting) is done by complex multiply:

r̃1 (t) = r+ (t)e−j2πfLO t

= ej2πfc t e−j2πfLO t
= ej2πfb t
= cos 2πfb t + j sin 2πfb t
= I(t) + jQ(t)

• fE : desired frequency for listening or for decoder (could be zero or not zero)

• fd = fb − fE : digital local oscillator in baseband used for tuning into desired

signal
 CHAPTER 2. IQ SIGNALS 185

• fd : is controlled by tuning knob or mouse

• The second spectrum shift is done by complex multiply:
r̃2 (t) = ej2πfb t e−2πfd t
= ej2π(fb −fd )t
= e2πfE t

More generally, if the RF signal contains information encoded in its amplitude

and phase, then the RF signal r+ (t) = a(t)ejφ(t) ej2πfc t is multiplied by the complex
local oscillator e−j2πfLO t| = cos 2πfLO t − j sin 2πfLO t to yield
r̃1 (t) = r+ (t)e−j2πfLO t
INSTRUCTOR COPY

= [a(t)ejφ(t) ej2πfc t ]e−j2πfLO t

= a(t)ejφ(t) ej2πfb t
= I(t) + jQ(t)
where the received complex baseband signal is
r̃1 (t) = I(t) + jQ(t)
= a(t) cos φ(t) cos 2πfb t + ja(t) sin φ(t) sin 2πfb t

If we want to receive the information contained in a(t), φ(t) then we multiply r̃1 (t)
by a complex exponential e−j2πfb t at exactly −fb to obtain
r̃2 (t) = r̃1 (t)e−j2πfb t
= a(t)ejφ(t) e−j2πfb t (2.56)
= a(t)ejφ(t)
centered at fE = 0, followed by a low pass filter to filter out any other signals. We
have shifted the spectrum twice, once by fLO using analog circuits and a second time
by fb using software to receive the desired signal. In this case, the decoder uses fE = 0

Exercise 44
Explain the operation of a transmitter that has a selectable carrier frequency.
Sketch a block diagram and the signal spectrum. The transmitter uses the
same frequency shifting operations as the receiver, but in reverse order.

Exercise 45
For an complex baseband signal s̃(t) = I(t) + jQ(t) = a(t)ejφ(t) , find an ex-
pression for a(t) as a function of I(t), Q(t)
 CHAPTER 2. IQ SIGNALS 186

Exercise 46
Given a USRP daughterboard receiver (see fig. 2.26) with fLO = 142 MHz and
an incoming RF signal fc = 142.17 MHz, find an expression for

1. the two real output signals from the receiver I(t), Q(t) , and

2. the complex output signal I(t) + jQ(t) written in polar form a(t)ejφ(t)

2.12 IQ imbalance

In practice, the two local oscillators of the USRP daughterboard analog IQ receiver are
INSTRUCTOR COPY

such that Q(t) is not exactly the same amplitude as I(t) and not exactly 90 degrees
out of phase with I(t). Thus the cancellation of the image frequencies that was
demonstrated in exercise is not exact, and there is some residual image component.
This is called IQ imbalance.

The practical result is that in the spectrum analyzer or waterfall view, the signals
at positive frequencies will have weak images at the corresponding negative frequen-
cies, and visa versa. These images are not really there, they are an artifact of the IQ
imbalance.

If there is IQ imbalance present, then the amplitude and phase of I(t) is shifted
relative to what it should be by a complex factor (1 + α)ejθ for small values of α, θ .
We write the complex output r̃0 (t) of the imperfect receiver with IQ imbalance as

r̃0 (t) = I 0 (t) + jQ0 (t)

= (1 + α)ejθ I(t) + jQ(t)
I 0 (t) = (1 + α)cosθI(t)
Q0 (t) = Q(t) + (1 + α) sin θI(t)

For small theta, this matches the approximation

<{out} = <{in}(1 + Magnitude),

={out} = ={in} + <{in} × Phase
 CHAPTER 2. IQ SIGNALS 187

where we define

<{out} = I 0 (t),
Re{in} = I(t),
={out} = Q0 (t),
={in} = Q(t),
Magnitude = α,
Phase = θ.

In the frequency domain, IQ imbalance appears in the form of an image, i.e. every
signal at frequency fb will have a mirror image at −fb . For a signal at fb , the desired
complex baseband signal is r̃(t) = s̃(t) = ej2πfb t , but with IQ imbalance, the actual
INSTRUCTOR COPY

observed complex baseband signal is

r̃0 (t) = µej2πfb t + νe−j2πfb t

with complex constants µ, ν , thus explicitly showing the desired signal at fb and
the mirror image at −fb . The complex constants µ, νare functions of α, θ . When
α = θ = 0, ν = 0.

In general, we can write the received complex signal obtained at the output of an
imperfect IQ receiver as
r̃0 (t) = µr̃(t) + ν r̃∗ (t)
where r̃0 (t) is the receiver output, r̃(t) = s̃(t) is the desired complex signal, and r̃∗ (t)
is the image. The image rejection is 20 log |ν/µ| dB.

The mirror image can be nulled out by multiplying Q0 (t) by the inverse of this
1
complex factor (1 + α)ejθ , i.e. 1+α e−jθ = βejψ . The values of the magnitude β and
phase ψ can be found manually (with hardware or software controls) or by an adaptive
algorithm.

Exercise 47
Verify the expressions for the imperfect receiver outputs I 0 (t) and Q0 (t) given
above.

Exercise 48
Verify the approximate expressions for IQ imbalance correction given in Section
4 above.
 CHAPTER 2. IQ SIGNALS 188

2.13 Summary

In chapter 2 we wrote a general radio signal s(t) = a(t) cos[2πfc t + φ(t)] with ampli-
tude a(t), frequency fc and phase π(t). We also wrote the complex baseband signal
s̃(t) = a(t)ejφ(t) that does not include the carrier wave at fc and showed the rela-
tionship between s(t) and s̃(t). We showed that we can use a(t) and/or φ(t) or some
combination thereof to represent a message m(t).

We also wrote the radio signals in the time and frequency domain
s+ (t) = s̃(t)ej2πfc t
s− (t) = s̃∗ (t)e−j2πfc t
2s(t) = s+ (t) + s− (t)
INSTRUCTOR COPY

= s̃(t)ej2πfc t + s̃∗ (t)e−j2πfc t

S+ (f ) = S̃(f − fc )
S− (t) = S̃ ∗ (−(f + fc ))
2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))

We also write complex baseband signals in the form

s̃(t) = a(t)ejφ(t) = I(t) + jQ(t)
s̃∗ (t) = a(t)e−jφ(t) = I(t) − jQ(t)
as a general complex function of time. Thus
s(t) = s+ (t) + s− (t)
a(t) jφ(t) j2πfc t a(t) −jφ(t) −j2πfc t
= e e + e e
2 2
= <{s̃(t)ej2πfc t }
= a(t) cos[2πfc t + φ(t)]
s+ (t) = s̃(t)ej2πfc t
s− (t) = s̃∗ (t)e−j2πfc t

2s(t) = s+ (t) + s− (t)

= a(t)ejφ(t) ej2πfc t + a(t)e−jφ(t) e−j2πfc t
= 2a(t) cos[2πfc t + φ(t)]
 CHAPTER 2. IQ SIGNALS 189

2.14 Chapter review questions

Q 2.1 Consider the radio receiver detailed in section 2.10.2.2. Find an expression for
the I and Q outputs of the receiver, writing the input RF signal as

r+ (t) = a(t)ejφ(t) ej2πfc t

and the local oscillator as

e−j2πfLO t = cos 2πfLO t − j sin 2πfLO t

Assume that fLO 6= fc and fc − fLO = fb .

Do this question the following two ways:

INSTRUCTOR COPY

1. Using complex (polar form) exponential notation only, converting to rectangular

notation at the very end to get the real and imaginary parts of the receiver
output, I(t), Q(t).

2. Computing the real and imaginary parts of the receiver output separately, us-
ing real notation, i.e. multiplying cosines and sines and using trigonometric
identities.

Solution

1.

r+ (t) = a(t)ej2πfc t e−j2πfLO t ejφ(t)

= a(t)ej2π(fc −fLO )t ejφ(t)
= a(t)ej2πfb t ejφ(t)
= a(t) cos [2πfb t + φ(t)] + ja(t) sin [2πfb t + φ(t)]
= I(t) + jQ(t)

where fb = fc − fLO 6= 0.

I(t) = a(t) cos [2πfb t + φ(t)]

Q(t) = a(t) sin [2πfb t + φ(t)]

2.

r(t)e−j2πfc t = {a(t) cos [2πfc t + φ(t)]}{cos 2πfLO t − j sin 2πfLO t}

 CHAPTER 2. IQ SIGNALS 190

I(t) = a(t) cos [2πfc t + φ(t)] cos 2πfLO t

= 0.5a(t) cos [2π (fc − fLO ) t + φ(t)] + term filtered out by LPF
= 0.5a(t) cos [2πfb t + φ(t)]

Q(t) = a(t) cos [2πfc t + φ(t)] sin 2πfLO t

= 0.5a(t) sin [2π (fc − fLO ) t + φ(t)] + term filtered out by LPF
= 0.5a(t) sin [2πfb t + φ(t)]

Q 2.2 Consider the summary in section 2.13 and the notation for the complex enve-
lope (without carrier), the complex signal (with carrier) and the real signal
INSTRUCTOR COPY

s̃(t) = a(t)ejφ(t)
= i(t) + jq(t)
ŝ(t) = s̃(t)ej2πfc t
s(t) = Re{ŝ(t)}
= Re{s̃(t)ej2πfc t }

1. Write the complex signal (with carrier) and the real signal in terms of i(t) and
q(t).

2. Write i(t), q(t) in terms of a(t), φ(t).

Solution

1. Complex signal with carrier

ŝ(t) = s̃(t)ej2πfc t
= a(t)ejφ(t) ej2πfc t
= [i(t) + jq(t)] [cos 2πfc t + j sin 2πfc t]
= i(t) cos 2πfc t − q(t) sin 2πfc t + j [q(t) cos 2πfc t + i(t) sin 2πfc t]

Real signal with carrier

s(t) = Re{ŝ(t)}
= i(t) cos 2πfc t − q(t) sin 2πfc t
 CHAPTER 2. IQ SIGNALS 191

2. Given s̃(t) = a(t)ejφ(t) = i(t) + jq(t)

i(t) = a(t) cos φ(t)

q(t) = a(t) sin φ(t)

Q 2.3 Consider the case where r+ (t) = a(t)ejφ(t) ej2πfc t for fc = 101.9 MHz.

1. Draw a block diagram of the receiver showing the signals r+ (t), r̃1 (t), r̃2 (t)
and the complex oscillators at fLO = 102 MHz, fb and the complex multipliers
(mixers).

2. Draw the spectrum of r̃1 (t), r̃2 (t) given the spectrum of r+ (t) as shown in
INSTRUCTOR COPY

fig. 2.32, where W = 3 kHz.

W = 3 kHz

r+ (t)

f
fc fc + W
(101.9 MHz) (101.903 MHz)

Figure 2.32: Spectrum of r+ (t)

 CHAPTER 2. IQ SIGNALS 192

Solution

r+ (t) r̃1 (t) r̃2 (t)

e−j2πfLO t e−j2πfb t

Figure 2.33: Block diagram of complex receiver

INSTRUCTOR COPY

1.

W = 3 kHz W = 3 kHz

r̃2 (t) r̃1 (t)

f
fE fE + W fb fb + W
(0 Hz) (3 kHz) (100 kHz) (103 kHz)

Figure 2.34: Spectrum of r̃1 (t), r̃2 (t)

2.
 Chapter 3

Amplitude modulation (AM)

INSTRUCTOR COPY

Learning objectives

In this chapter, we consider the class of signals for which the amplitude a(t) varies
with time but the phase φ(t) is kept constant with time. This class of signals is
called Amplitude Modulation (AM) signals. In particular, we use the message m(t)
to modulate the amplitude a(t) and leave the phase φ(t) constant. This method of
modulation is called amplitude modulation (AM).

We study AM signals and their waveforms1 in the time and frequency domains
with messages m(t) that are sine waves (single tone). We also study how to implement
a signal generator for AM signals using software as well as analog hardware.

By the end of this chapter we will have covered:

• Math expression for AM signal with sine wave message in time and frequency
domain.

• Concept of sidebands in AM signal spectrum

• Power in carrier and sidebands of AM signals

• AM spectrum with general analog message

• AM signals with digital message

1
The waveform of a signal s(t) is the shape of a graph of s(t). The spectral shape of S(f ) is the
shape of a graph of S(f ).

193
 CHAPTER 3. AMPLITUDE MODULATION (AM) 194

• Generating AM signals using software

• Generating AM signals using analog hardware

3.1 Amplitude modulation with full carrier

We consider AM signals s(t) where the message m(t) to modulate the amplitude a(t)
is varied in step with the message m(t) and the phase φ(t) is left constant.

AM waves typically use carrier frequencies on the order of MHz, whereas the
message frequency is typically on the order of kHz.
INSTRUCTOR COPY

Amplitude modulation is used for:

• broadcasting (AM radio), using carrier frequencies in the medium wave band
(540–1700 kHz), the long wave band 153–279 kHz (in Europe only) and the short
wave bands (3–30 MHz).

• aircraft navigational aids (190–535 kHz and 108–118 MHz)

• aircraft and air traffic control voice communications (118–137 MHz).

Figure 3.1a shows a section of the AM broadcast band, received from a location
30 km NE of Victoria, British Columbia. The spectrum of each AM radio station is
the carrier wave plus the message spectrum. The carrier waves are spaced at 10 kHz
intervals, so each station (channel) has a 10 KHz bandwidth, plus and minus 5 KHz
from the carrier at the centre of the channel. Figure 3.1b shows almost the entire AM
broadcast band containing many stations. The spectrum clearly shows the carrier
wave peaks and the waterfall shows the strong carriers in orange and red. Some of
the channels contain no signals, since the transmitters on those channels are too far
away.
 CHAPTER 3. AMPLITUDE MODULATION (AM) 195
INSTRUCTOR COPY

(a) 570–800 kHz

(b) 600–1500 kHz

Figure 3.1: AM broadcast band showing carrier waves at 10 KHz intervals

In this section we will write the mathematical expressions for the signals

s̃(t), s+ (t), s− (t), s(t), S̃(f ), S+ (t), S− (f ), S(f )

 CHAPTER 3. AMPLITUDE MODULATION (AM) 196

obtained in chapter 2 for the special case of AM.

3.1.1 AM signals in time domain with general message

3.1.1.1 General message in time domain

3.1.1.2 AM complex baseband s̃(t)

For AM signals, the complex baseband (message) signal s̃(t) = a(t)ejφ(t) , where we
set the amplitude to be a DC value plus the message m(t) scaled by a constant ka ,
all multiplied by an overall scaling constant that establishes the voltage and power
INSTRUCTOR COPY

level of the signal. Thus we write

a(t) = Ac [1 + ka m(t)] (3.1)

Since normally we wish the amplitude a(t) ≥ 0, and m(t) will be both greater and
less than zero at times, we choose |ka m (t)| < 1 or [1 + ka m (t)] > 0.

Note that a(t) is a DC term 1 plus the message m(t) scaled by a gain factor ka ,
all scaled by Ac . We choose ka such that a(t) > 0 for all values of m(t) (fig. 3.2).
 CHAPTER 3. AMPLITUDE MODULATION (AM) 197

m(t)

a(t)

|M (f )| S̃(f )
INSTRUCTOR COPY

f f

−B B −B B

Figure 3.2: The relationship between m(t) and a(t) = Ac [1 + ka m(t)] for an AM
baseband signal where the amplitude increases with frequency.

We also assume that the bandwidth of m(t) is much less than the carrier frequency
fc used.

For AM signals, we often assume φ(t) = φ is a constant or zero. However, for AM

signals φ(t) may also be time-varying but is not used to transmit a message and any
time variations in φ are ignored at the receiver.

For AM we write the complex baseband signal with a time-varying amplitude and
constant phase

s̃(t) = a(t)ejφ(t)
= Ac [1 + ka m(t)]ejφ

for an arbitrary value of φ.

 CHAPTER 3. AMPLITUDE MODULATION (AM) 198

3.1.1.3 AM analytic signal s+ (t)

Recall from section section 2.1 the analytic (positive frequency) signal and the nega-
tive frequency signal are written

s+ (t) = s̃(t)ej2πfc t
s− (t) = s̃∗ (t)e−j2πfc t

Substituting the expression for s̃(t) for AM signals, we obtain

s+ (t) = Ac [1 + ka m(t)]ejφ ej2πfc t

= Ac ejφ ej2πfc t + Ac ka m(t)ejφ ej2πfc t
INSTRUCTOR COPY

s− (t) = Ac [1 + ka m(t)]e−jφ e−j2πfc t

= Ac e−jφ e−j2πfc t + Ac ka m(t)e−jφ e−j2πfc t

The analytic signal is illustrated in fig. 3.3.

2
=(s+ (t))

−2
0 2
0.5
1 0
1.5
t 2 −2
<(s+ (t))

Figure 3.3: s+ (t) for a message that increases in amplitude and frequency as time
passes. The envelope of the helix (and of the projections) is the message from fig. 3.2.
 CHAPTER 3. AMPLITUDE MODULATION (AM) 199

3.1.1.4 AM real passband s(t)

Recall from section section 2.1 the real signal s(t) is obtained by adding the positive
and negative frequency components
2s(t) = s+ (t) + s− (t)
= s̃(t)ej2πfc t + s̃∗ (t)e−j2πfc t

2
INSTRUCTOR COPY

−2

Figure 3.4: s(t) (purple) for an AM signal with a message that increases in amplitude
and frequency as time passes. The envelope of the signal (black) is the message from
fig. 3.2

Substituting the expression for s̃(t) for AM signals, we obtain

2s(t) = Ac [1 + ka m(t)]ejφ ej2πfc t + Ac [1 + ka m(t)]e−jφ e−j2πfc t
This result is sum of two complex exponentials, one positive frequency and one neg-
ative frequency. Using the definition ejθ + e−jθ = 2 cos θ with θ = 2πfc t + φ, the
transmitted AM signal is of the form:
2s(t) = 2a(t) cos(2πfc t + φ)
s(t) = Ac [1 + ka m(t)] cos(2πfc t + φ)
= Ac cos(2πfc t + φ) + Ac ka m(t) cos(2πfc t + φ)
φ(t) = φ = constant
This is illustrated in fig. 3.2. φ may be set to zero for convenience, but could be any
constant value independent of time. In the case of φ = 0,
s(t) = Ac [1 + ka m(t)] cos(2πfc t)
(3.2)
= Ac cos(2πfc t) + Ac ka m(t) cos(2πfc t)
 CHAPTER 3. AMPLITUDE MODULATION (AM) 200

From the mathematics we can see that we have modulated the amplitude a(t) of
the signal s(t) in step with the message m(t), so that s(t) is an AM signal.

For the special case where there is no message to send, m(t) = 0 and s(t) =
Ac cos(2πfc t) which is simply the carrier wave. In practice, m(t) could be an analog
or digital message as described in section 2.9.

3.1.2 General message spectrum

To illustrate the AM signal spectra we adapt the general complex baseband spectrum
of section 2.7.3 for the case of a real (not complex) message signal, see section 2.7.6
INSTRUCTOR COPY

The most general (and also most realistic) example spectrum M (f ) contains both
discrete frequency components (complex exponentials) plus a continuous time spec-
trum that may be viewed as the sum of unresolved complex exponentials. For a real
message, the magnitude spectrum |M (f )| must be symmetric about f = 0. Thus we
use the example message spectrum in fig. 3.5.

M (f )

4Am
3Am
2Am
Am
f
−B B

Figure 3.5: Message spectrum M (f ) for a general message.

The figure shows the magnitude spectrum |M (f )| only, which is symmetric because
m(t) is real. The phase spectrum arg M (f ) is assumed to be a linear phase passing
through the origin, i.e. arg M (f ) = kp f for some scaling constant kp , and is anti-
symmetric because m(t) is real.
 CHAPTER 3. AMPLITUDE MODULATION (AM) 201

Since m(t) is real, m∗ (t) = m(t) so that the spectrum M ∗ (−f ) = M (f ). This
means that the magnitude of the mirror image spectrum M ∗ (−f ) about f = 0 (recall
section 2.5.2) is the same as the original spectrum M (f ). Thus the magnitude |M (f )|
must be symmetric about f = 0. The phase of the mirror image spectrum is the
opposite sign arg M ∗ (−f ) = − arg M (f ) and is thus anti-symmetric about f = 0.

Exercise 49
Confirm this result for m(t) = sin 2πfc t.

Solution

m(t) = sin 2πfc t

INSTRUCTOR COPY

ej2πfc t − e−j2πfc t
=
2j
1
M (f ) = [δ(f − fc ) − δ(f + fc )]
2j
∗
m (t) = sin 2πfc t = m(t)
1
M ∗ (−f ) = [δ(−f − fc ) − δ(−f + fc )]
−2j
1
= [δ(−f + fc ) − δ(−f − fc )]
2j
1
= [δ(f − fc ) − δ(f + fc )]
2j
= M (f )

Observe that the mirror image of δ(f −fc ) about f = 0 is obtained by replacing
f with −f so that we write the mirror image as δ(−f − fc ) = δ(f + fc ). The
phase of the delta function peak at fc is −π/2 and at −fc is +π/2, so the
phases are anti-symmetric about f = 0.a
a
If we had used m(t) = cos 2πfc t instead of sin 2πfc t then the phases would have been zero
for both ±fc and the anti-symmetric behaviour of the phase would not have been revealed.
Examples have to be chosen carefully.

3.1.3 AM spectrum for general message

What does the AM wave look like in the frequency domain?

 CHAPTER 3. AMPLITUDE MODULATION (AM) 202

We know from section 2.1 and section the relationship between the complex base-
band signal S̃(f ), the analytic signal S+ (f ) and the real signal S(f )

2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))

This expression will be the starting point for finding the AM spectrum in this section.

This expression is valid when the observation interval (i.e. the time spent ob-
serving s̃(t), s+ (t), s(t)) is long enough that the Fourier transform operation is
valid. The observation interval must be greater than the inverse of the lowest fre-
quency component greater than zero in the complex baseband signal. In a discrete-
time (sampled) system where the continuous-time Fourier transforms used in this
section become a Short-time Discrete Fourier transform, the observation interval
INSTRUCTOR COPY

(frame/window/segment size) must contain at least one full period of the signal.

3.1.3.1 AM complex baseband spectrum S̃(f )

In the case of AM we write the complex baseband signal. We also write its complex
conjugate that we will need to obtain an expression for the real signal spectrum S(f).

s̃(t) = a(t)ejφ(t)
= Ac [1 + ka m(t)]ejφ
s̃∗ (t) = Ac [1 + ka m∗ (t)]e−jφ

Thus using the Fourier transform properties

1 ↔ δ(f )
m(t) ↔ M (f )
m∗ (t) ↔ M ∗ (−f )

we obtain the complex baseband signal in the frequency domain

S̃(f ) = Ac [δ(f ) + ka M (f )]ejφ

S̃ ∗ (−f ) = Ac [δ ∗ (−f ) + ka M ∗ (−f )]e−jφ

Note that since the magnitude of the message spectrum |M (f )| is symmetrical around
f = 0, the magnitude of the AM complex baseband spectrum |S̃(f )| is also symmetric
about f = 0 as shown in fig. 3.6. Also note that since δ(f ) is real that δ ∗ (−f ) = δ(f ).
 CHAPTER 3. AMPLITUDE MODULATION (AM) 203

|S̃(f )|

4Am
3Am
2Am
Am
f
INSTRUCTOR COPY

−B B

Figure 3.6: Complex baseband spectrum for the general message of fig. 3.5 for an AM
signal.

Since m(t) is real, m∗ (t) = m(t) and M ∗ (−f ) = M (f ). Thus we write AM signals
with a real message m(t),
S̃(f ) = Ac [δ(f ) + ka M (f )]ejφ
S̃ ∗ (−f ) = Ac [δ(f ) + ka M (f )]e−jφ

The magnitude of the AM complex baseband spectrum is symmetrical about f = 0

|S̃ ∗ (−f )| = |S̃(f )|
The phase of the AM complex baseband spectrum is anti-symmetric about f = 0
arg{S̃ ∗ (−f )} = − arg{S̃(f )}

3.1.3.2 AM analytic signal spectrum S+ (f )

Substituting the expression for S̃(f ) and S̃ ∗ (−f ) for AM signals, we obtain the spec-
trum of the analytic (positive frequency) signal for AM.
S+ (f ) = S̃(f − fc )
= Ac [δ(f − fc ) + ka M (f − fc )]ejφ
 CHAPTER 3. AMPLITUDE MODULATION (AM) 204

and also the negative frequency signal for AM

S− (f ) = S̃ ∗ (−(f + fc ))
= Ac [δ(f + fc ) + ka M (f + fc )]e−jφ
Note the symmetry of |S+ (f )| about f = fc as shown in fig. 3.7. Also note the
symmetry of |S− (f )| about f = −fc . The AM analytic signal spectrum S+ (f ) is the
message spectrum M (f ) scaled by ka and shifted up by fc plus a delta function spike
exactly at fc .

|S+ (f )|

4Am
INSTRUCTOR COPY

3Am
2Am
Am
f
fc − B fc fc + B

Figure 3.7: Complex passband spectrum for the general message of fig. 3.5 for an AM
signal.

3.1.3.3 AM real passband spectrum S(f )

The spectrum of the real AM signal is the sum of the positive and negative frequency
components
2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))
= Ac [δ(f − fc ) + ka M (f − fc )]ejφ
+ Ac [δ(f + fc ) + ka M (f + fc )]e−jφ

We note that the magnitude of the spectrum is symmetric about f = 0 and the
phase is anti-symmetric about f = 0
 CHAPTER 3. AMPLITUDE MODULATION (AM) 205

If we choose φ = 0 then

2S(f ) = Ac [δ(f − fc ) + δ(f + fc ) + ka M (f − fc ) + ka M (f + fc )]

S(f ) = (Ac /2)[δ(f − fc ) + δ(f + fc ) + ka M (f − fc ) + ka M (f + fc )]

This result is correct but hides the anti-symmetric phase and omits the fact that
an expression for any spectrum S(f ) (including the spectrum M (f ) of a real signal)
is complex in general.

|2S(f )|
4Am
INSTRUCTOR COPY

3Am
2Am
Am f
−fc fc

Figure 3.8: Real passband spectrum for the general message of fig. 3.5 for an AM
signal.

fig. 3.8 shows the magnitude of the spectrum |S+ (f )| and |S− (f )| and their sum
2S(f ) for AM signals. Note the symmetry of |S(f )| about f = 0.

It is useful at this point to compare the time and frequency domain expressions
for the AM signal with φ = 0

s(t) = Ac cos(2πfc t) + Ac ka m(t) cos(2πfc t) (3.3)

(3.4)

and observe two useful Fourier transform properties:

• the Fourier transform of the cosine term is two δ functions

• the Fourier transform of a function m(t) multiplied by a cosine is the spectrum

of m(t) shifted up and down by the frequency of the cosine term.
 CHAPTER 3. AMPLITUDE MODULATION (AM) 206

3.1.4 AM waveform with dual tone message

In section 2.1, section 2.2.1 showed that a complex baseband signal s̃(t) consisting of
two complex exponentials summed was indeed complex with an asymmetrical spec-
trum where the amplitudes of the positive and negative frequencies were different.

3.1.4.1 Dual tone message m(t)

For AM signals, the example message must be real. Thus we consider a real message
that is the sum of N complex exponentials, where the amplitudes an , frequencies fn
and phases φn must be related in such a way that the message is real.
INSTRUCTOR COPY

In particular, we use the dual-tone message described in section 2.7.6.3 and recall
the solution to section 2.7.6.4 repeated here for convenience.

a−2 = a2 = Am and a−1 = a1 = Am /2 and a0 = 0, so we can write

N/2
X
m(t) = an ej(2πfn t+φn )
n=−N/2

= a−2 ej(2πf−2 t+φ−2 ) + a−1 ej(2πf−1 t+φ−1 ) + a1 ej(2πf1 t+φ1 ) + a2 ej(2πf2 t+φ2 )
= Am ej(2πf−2 t+φ−2 ) + (Am /2)ej(2πf−1 t+φ−1 ) + (Am /2)ej(2πf1 t+φ1 ) + Am ej(2πf2 t+φ2 )
= 2Am cos(2πf2 t + φ2 ) + Am cos(2πf1 t + φ1 )
= 2Am cos(2π(B/2)t + φ2 ) + Am cos(2π(B/4)t + φ1 )

As shown in fig. 2.15, the spectrum M (f ) of the dual tone message for the case
φ1 = φ2 = 0 is
N/2
X
M (f ) = an ejφn δ(f − fn )
n=−N/2

= a−2 δ(f − f−2 ) + a−1 δ(f − f−1 ) + a1 δ(f − f1 ) + a1 δ(f − f2 )

= Am δ(f + B/2) + (Am /2)δ(f + B/4) + (Am /2)δ(f − B/4) + Am δ(f − B/2)
= Am δ(f + f2 ) + (Am /2)δ(f + f1 ) + (Am /2)δ(f − f1 ) + Am δ(f − f2 )
 CHAPTER 3. AMPLITUDE MODULATION (AM) 207

Am

−f2 −f1 f1 f2 f

Figure 2.15 (repeated): Magnitude spectrum |M (f )| for a dual tone message.

INSTRUCTOR COPY

3.1.4.2 AM complex baseband signal with dual tone m(t)

Thus for AM signals with a real message m(t),

s̃(t) = a(t)ejφ(t)
= Ac [1 + ka m(t)]ejφ
= Ac ejφ [1 + 2µ cos 2πf2 t + µ cos 2πf1 t]

where where we have introduced µ = ka Am and refer to it as the modulation index.

S̃(f ) = Ac [δ(f ) + ka M (f )]ejφ

= Ac ejφ [δ(f ) + µδ(f + f2 ) + (µ/2)δ(f + f1 ) + (µ/2)δ(f − f1 ) + µδ(f − f2 )
S̃ ∗ (−f ) = Ac [δ(f ) + ka M (f )]e−jφ

The magnitude of the complex baseband spectrum is symmetrical about f = 0

|S̃ ∗ (−f )| = |S̃(f )|

The phase of the complex baseband spectrum is anti-symmetric about f = 0

arg{S̃ ∗ (−f )} = − arg{S̃(f )}

3.1.4.3 AM analytic signal with dual tone m(t)

For the dual tone m(t) the analytic (positive frequency) signal for AM may be written
 CHAPTER 3. AMPLITUDE MODULATION (AM) 208

s+ (t) = Ac [1 + ka m(t)]ejφ ej2πfc t

= Ac [1 + ka (2Am cos 2πf2 t + Am cos 2πf1 t)]ejφ ej2πfc t
= Ac [1 + 2µ cos 2πf2 t + µ cos 2πf1 t]ejφ ej2πfc t

.
Exercise 50
Write s− (t) for an AM signal.

Solution
INSTRUCTOR COPY

s− (t) = Ac [1 + ka m(t)]e−jφ e−j2πfc t

= Ac [1 + ka (2Am cos 2πf2 t + Am cos 2πf1 t)]e−jφ e−j2πfc t
= Ac [1 + 2µ cos 2πf2 t + µ cos 2πf1 t]e−jφ e−j2πfc t

For the dual tone m(t) we obtain the spectrum of the analytic (positive frequency)
signal S+ (f ) for AM,

S+ (f ) =S̃(f − fc )
=Ac [δ(f − fc ) + ka M (f − fc )]ejφ
=Ac [δ(f − fc ) + ka (Am δ(f + f2 − fc ) + (Am /2)δ(f + f1 − fc )
+ (Am /2)δ(f − f1 − fc ) + Am δ(f − f2 − fc ))]ejφ
=Ac [δ(f − fc ) + µδ(f + f2 − fc ) + (µ/2)δ(f + f1 − fc )
+ (µ/2)δ(f − f1 − fc ) + µδ(f − f2 − fc )]ejφ

as shown in fig. 3.9.

 CHAPTER 3. AMPLITUDE MODULATION (AM) 209

Ac

µAc
µ
2 Ac

fc − f2 fc − f1 fc fc + f1 fc + f2 f
INSTRUCTOR COPY

Figure 3.9: Spectrum S+ (f ) for a dual tone message.

Exercise 51
Write an expression for the negative frequency signal S− (f ).

Solution

S− (f ) =S̃ ∗ (−(f + fc ))
=Ac [δ(f + fc ) + ka M (f + fc )]e−jφ
=Ac [δ(f + fc ) + ka (Am δ(f + f2 + fc ) + (Am /2)δ(f + f1 + fc )
+ (Am /2)δ(f − f1 + fc ) + Am δ(f − f2 + fc ))]e−jφ

Note the symmetry of |S+ (f )| about f = fc and the symmetry of |S− (f )| about
f = −fc . The AM analytic signal spectrum S+ (f ) is the message spectrum
M (f ) scaled by ka and shifted up by fc plus a delta function spike exactly at
fc .

3.1.4.4 AM real passband signal with dual tone m(t)

For the dual tone m(t) the real passband signal for AM may be written
 CHAPTER 3. AMPLITUDE MODULATION (AM) 210

2s(t) = Ac [1 + ka m(t)]ejφ ej2πfc t + Ac [1 + ka m(t)]e−jφ e−j2πfc t

= Ac [1 + ka m(t)]2 cos(2πfc t + φ)
= Ac [1 + ka (2Am cos 2πf2 t + Am cos 2πf1 t)]2 cos(2πfc t + φ)
s(t) = Ac [1 + 2µ cos 2πf2 t + µ cos 2πf1 t)] cos(2πfc t + φ)
= Ac cos(2πfc t + φ) + Ac 2µ cos 2πf2 t cos(2πfc t + φ) + Ac µ cos 2πf1 t cos(2πfc t + φ)

The last line of this expression is intended to show explicitly that s(t) includes
terms that are products of two cosine waves. These products may also be written as
cosine waves at the sum and difference frequencies using the identity cos α cos β =
[cos(α + β) + cos(α − β)]/2 as shown in eq. (3.5).
INSTRUCTOR COPY

s(t) =Ac cos(2πfc t + φ)

+ Ac 2µ cos(2πf2 t) cos(2πfc t + φ)
+ Ac µ cos(2πf1 t) cos(2πfc t + φ)
= Ac cos(2πfc t + φ)
+ Ac µ cos(2π(f2 + fc )t + φ) + Ac µ cos(2π(f2 − fc )t − φ) (3.5)
+ 0.5Ac µ cos(2π(f1 + fc )t + φ) + 0.5Ac µ cos(2π(f1 − fc )t − φ)
= Ac cos(2πfc t + φ)
+ Ac µ cos(2π(fc + f2 )t + φ) + Ac µ cos(2π(fc − f2 )t + φ)
+ 0.5Ac µ cos(2π(fc + f1 )t + φ) + 0.5Ac µ cos(2π(fc − f1 )t + φ)

Thus s(t) contains 5 discrete cosine waves at fc , fc ± f2 , fc ± f1 .

The spectrum of the real AM signal with dual tone m(t) is the sum of the positive
 CHAPTER 3. AMPLITUDE MODULATION (AM) 211

and negative frequency components (fig. 3.10):

2S(f ) = S+ (f ) + S− (f )
=S̃(f − fc ) + S̃ ∗ (−(f + fc ))
=Ac [δ(f − fc ) + ka M (f − fc )]ejφ + Ac [δ(f + fc ) + ka M (f + fc )]e−jφ
=Ac [δ(f − fc ) + ka (Am δ(f + f2 − fc ) + (Am /2)δ(f + f1 − fc )
+ (Am /2)δ(f − f1 − fc ) + Am δ(f − f2 − fc ))]ejφ
+ Ac [δ(f + fc ) + ka (Am δ(f + f2 + fc ) + (Am /2)δ(f + f1 + fc )
+ (Am /2)δ(f − f1 + fc ) + Am δ(f − f2 + fc ))]e−jφ
S(f ) =(Ac /2)[δ(f − fc ) + µδ(f + f2 − fc ) + (µ/2)δ(f + f1 − fc )
+ (µ/2)δ(f − f1 − fc ) + µδ(f − f2 − fc )]ejφ
INSTRUCTOR COPY

+ (Ac /2)[δ(f + fc ) + µδ(f + f2 + fc ) + (µ/2)δ(f + f1 + fc )

+ (µ/2)δ(f − f1 + fc ) + µδ(f − f2 + fc ))]e−jφ

Ac

µAc
µAc
2

−fc fc f

Figure 3.10: Spectrum 2S(f ) for a dual tone message.

There is a total of 10 discrete frequencies in the AM spectrum with dual-tone

modulation. Note that the magnitude of the spectrum is symmetric about f = 0 and
the phase is anti-symmetric about f = 0.

3.1.4.5 Summary of dual-tone AM

Mathematical expressions to represent an AM signal are derived starting with the

results for a general signal in chapter 1 and specializing them for the case of AM. The
 CHAPTER 3. AMPLITUDE MODULATION (AM) 212

starting point is to define the complex baseband signal for AM, and then derive the
analytic signal and real passband signal and the spectrum for these 3 signals.

3.1.5 AM waveform with single tone message

In this section, we derive the results from the previous section using a single tone
at f1 and deleting the second tone at f2 and setting the phase φ = 0 at times for
convenience.

AM with a single tone message is a real thing in practice. It is used for aircraft
navigational aids, socalled non-directional beacons (NDB) on frequencies 190-535
KHz (below the AM broadcast band) and VHF Omni-directionak range (VOR) on
INSTRUCTOR COPY

frequencies 108-117.95 MHz (just above the FM broadcast band. The single tone is
switched on and off in a Morse code pattern.

A waterfall plot of an NDB is shown in figure fig. 3.11. This NDB has callsign
FHR and is located close to Victoria BC, see https://ourairports.com/navaids/
FHR/Friday_Harbor_NDB_US/. The carrier is seen at fc = 284 kHz and the single tone
message is Morse code (on-off-keying) at fc − f1 = 283 kHz and fc + f1 = 285 kHz.

Figure 3.11: NDB-FHR at 284 kHz

 CHAPTER 3. AMPLITUDE MODULATION (AM) 213

3.1.5.1 AM time domain waveform for single tone message

This derivation in this section starts with the expression for a real AM passband
signal with a single tone message, whereas the previous section started with the AM
complex baseband signal with a general message. The results are the same.

For convenience, we repeat here eq. (3.5) for the AM signal with dual tone mod-
ulation derived in the previous section:

s(t) = Ac cos(2πfc t+φ)+Ac 2µ cos(2πf2 t) cos(2πfc t+φ)+Ac µ cos(2πf1 t) cos(2πfc t+φ)

We delete the tone at f2 and relabel f1 = fm and set φ = 0 to obtain the starting
point for this section
INSTRUCTOR COPY

s(t) = Ac cos(2πfc t) + Ac µ cos(2πfm t) cos(2πfc t)

Thus given the AM signal for a general signal s(t) = Ac [1 + ka m(t)] cos(2πfc t)
with a single tone message, m(t) = Am cos(2πfm t), the resulting AM wave is given
by
s(t) = Ac [1 + ka Am cos(2πfm t)] cos(2πfc t) (3.6)

Recall that an analog message (voice or music) is time varying in the practical
case and will normally change from one frame to the next. Recall that frames for
audio are typically 5–23 ms. The AM signal above with m(t) = Am cos(2πfm t) is
valid for one frame in which the message contains only one frequency fm during that
frame. In the next frame, the message could be the same or could be different.

For our initial analysis of s(t), we will assume the message m(t) is of constant
frequency fm for all frames. The AM signal s(t) appears in the time domain to be
the carrier wave with an envelope2 that replicates the shape of the modulating tone.

m(t) = Am cos 2πfm t

a(t) = Ac (1 + ka m(t)]
s(t) = Ac (1 + ka m(t)] cos 2πfc t
= Ac (1 + ka Am cos 2πfm t] cos 2πfc t
= Ac (1 + µ cos 2πfm t] cos 2πfc t
= Ac cos 2πfc t + Ac µ cos 2πfm t cos 2πfc t
µ = ka Am < 1
2
here the envelope is the real part of the complex envelope
 CHAPTER 3. AMPLITUDE MODULATION (AM) 214

We use the symbol µ = Am ka for simplification and call it the modulation index.
The modulation index µ can be given as a percentage where 1 = 100%.

m(t)

Am

−Am
c(t)

Ac
INSTRUCTOR COPY

−Ac
s(t)

Ac (1 + ka Am )

−Ac (1 + ka Am )

Figure 3.12: Components of an AM signal with single tone message message m(t) in
blue, carrier c(t) in red and the resulting real passband AM waveform, s(t).

Using the complex I-Q notation and assuming φ(t) = φ which is constant but not
necessarily zero

s(t) = a(t) cos(2πfc t + φ(t))

= <{a(t)ejφ(t) ej2πfc t }

Here we have used the real operator < on the analytic (complex passband, positive
frequency) signal. Recall in chapter 2 we showed that using the < operator on a
positive frequency signal is equivalent to adding the negative frequency signal

<{s+ (t)} = [s+ (t) + s− (t)]/2

<{ejθ(t) } = [ejθ(t) + ejθ(t) ]/2
= cos θ(t)
 CHAPTER 3. AMPLITUDE MODULATION (AM) 215

To summarize this exploration of AM signals in the time domain we have:

a(t) = Ac [1 + µ cos 2πfm t]

φ(t) = φ

which when describing the real passband AM signal are combined into

s(t) = <{Ac [1 + ka m(t)]ej2πfc t ejφ }

= <{Ac [1 + µ cos 2πfm t]ej2πfc t ejφ }
= Ac [1 + µ cos 2πfm t] cos[2πfc t + φ]

and when describing the complex baseband AM signal are combined into

s̃(t) = a(t)ejφ
INSTRUCTOR COPY

= Ac [1 + µ cos 2πfm t]ejφ

= I(t) + jQ(t)
I(t) = a(t) cos φ
= Ac [1 + µ cos 2πfm t] cos φ
Q(t) = a(t) sin φ
= Ac [1 + µ cos 2πfm t] sin φ

When φ = 0 these expressions simplify further into

s(t) = Ac [1 + µ cos 2πfm t] cos[2πfc t] (3.7)

s̃(t) = Ac [1 + µ cos 2πfm t] (3.8)
= a(t)
= I(t) + jQ(t)

and the individual complex components are

I(t) = Ac [1 + µcos2πfm t]
= a(t)
Q(t) = 0

showing that in this case there is no imaginary component of the complex baseband
signal. We will use the version of s(t) with φ = 0 from eq. (3.7) in subsequent sections.
 CHAPTER 3. AMPLITUDE MODULATION (AM) 216

3.1.5.2 AM spectrum for single tone message

We first consider the special case where the message m(t) = Am cos(2πfm t) with
Am = 1, and take the Fourier transform of the message
F ourier
m(t) −→ M (f )
F ourier 1
Am cos(2πfm t) −→ Am [∂(f − fm ) + ∂(f + fm )]
2
This result is obtained from the Fourier transform properties for a cos wave.

m(t) ↔ M (f )
m(t) = cos(2πfm t)
INSTRUCTOR COPY

= (ej2πfm t + e−j2πfm t )/2

ej2πfm t ↔ δ(f − fm ) positive frequencies
e−j2πfm t ↔ δ(f + fm ) negative frequencies
1
M (f ) = [δ(f − fm ) + δ(f + fm )]
2
Note that the cosine wave contains both positive and negative frequencies arising
from the exponentials rotating clockwise and counterclockwise respectively as time
increases.

The spectrum for any real signal is symmetrical about zero frequency. The spec-
trum of the message M (f ) can be drawn as shown in fig. 3.13, with both positive
and negative frequencies. Note that the delta functions are drawn as vertical lines
with arrows that suggest they go to infinity. The delta function δ(f − fm ) is zero for
all frequencies f 6= fm . For the frequency f = fm , the delta function δ(f − fm ) has
infinite
R∞ height, zero width and area 1, so that we can write the area under the curve
−∞
δ(f − fm ) = 1.

Am

−fm fm f

Figure 3.13: Frequency Spectrum of a Single Tone message

 CHAPTER 3. AMPLITUDE MODULATION (AM) 217

The spectrum of the carrier wave is written

c(t) ↔ C(f )
c(t) = cos(2πfc t)
= (ej2πfc t + e−j2πfc t )/2
ej2πfc t ↔ δ(f − fc ) positive frequencies (3.9)
e−j2πfc t ↔ δ(f + fc ) negative frequencies (3.10)
1
C(f ) = [δ(f − fc ) + δ(f + fc )] (3.11)
2
C(f ) appears as two delta function spikes, one at fc and one at −fc .

To find the spectrum of the AM signal

INSTRUCTOR COPY

s(t) = Ac (1 + µ cos 2πfm t] cos 2πfc t

(3.12)
= Ac cos 2πfc t + Ac µ cos 2πfm t cos 2πfc t
F ourier
we take the Fourier transform s(t) −→ S(f ) and obtain the result shown in fig. 3.14.

We will discuss the result for S(f ) first (illustrated in fig. 3.14), and then do the
mathematics.

Observe that the spectrum of an AM signal contains both positive and negative
frequencies. The AM signal spectrum is the sum of two components: the spectrum
of the carrier wave c(t) at ±fc plus the spectrum of the message (M (f ) shifted both
up and down by fc . There is a very important principle at work here: the frequency
shifting properties covered in section 2.1.

Ac
2

µAc
4

−fc fc f

Figure 3.14: Frequency spectrum of AM signal with single tone message

Notice that in the expression

s(t) = Ac cos 2πfc t + Ac µ cos 2πfm t cos 2πfc t
 CHAPTER 3. AMPLITUDE MODULATION (AM) 218

the message signal m(t) is multiplied by the real carrier wave c(t). Whenever a mes-
sage signal is multiplied by a real carrier wave containing both positive and negative
frequencies, the spectrum of the message signal is shifted both up and down.

This result for the AM signal spectrum can be demonstrated in two different ways.
The first method is to do it algebraically by simplifying s(t) in the time domain
and then converting the resulting collection of individual sinusoidal terms to their
frequency domain representation. Note: The trigonometric identity cos α cos β =
1
2
[cos(α + β) + cos(α − β)] is used to produce line three from line two below:

s(t) = Ac [1 + µ cos(2πfm t)] cos(2πfc t)

= Ac cos(2πfc t) + Ac µ cos(2πfm t) cos(2πfc t)
= Ac cos(2πfc t) + (Ac µ/2) cos(2π[fc + fm ]t) + (Ac µ/2) cos(2π[fc − fm ]t) (3.13)
INSTRUCTOR COPY

The second and third cosine terms represent the socalled sidebands 3 as seen in the
frequency domain. The frequency of the sidebands relative to the carrier frequency
holds the useful information describing the message.

The AM signal spectrum S(f ) may be written by taking the Fourier transform of
each of the 3 terms in the expression for s(t) above to obtain

Ac
S(f ) = [δ(f − fc ) + δ(f + fc )]
2
Ac µ
+ [δ(f − fc − fm ) + δ(f + fc + fm )]
4
Ac µ
+ (δ(f − fc + fm ) + δ(f + fc − fm ))] (3.14)
4

Exercise 52
Obtain S(f ) by writing s(t) for an AM signal as s(t) = [s+ (t)+s− (t)]/2, taking
the Fourier transform of the positive and negative frequency terms separately
and adding them together to get the final result for S(f )

Exercise 53
cM AX −AcM IN
Show that µ = A AcM AX +AcM IN
= Am ka , where AcM AX and AcM IN are re-
spectively the highest and lowest positive amplitude obtained by s(t). Hint:
Ac (1+µ)
AcM AX = Ac (1 + µ); AcM IN = Ac (1 − µ), AAcc max
min
=A c (1−µ)
.

3
Here is yet another term for a kind of band, to go with the terms baseband and passband
 CHAPTER 3. AMPLITUDE MODULATION (AM) 219

3.1.6 Observation interval

Observe that even though we seem to only modify the amplitude of the carrier wave
at fc sidebands are observed at fc + fm and fc − fm in the frequency domain repre-
sentation. How could there be frequencies at fc + fm and fc − fm when all we are
doing is varying the amplitude of the carrier at fc without changing its frequency?

To interpret the time and frequency domain waveforms for AM, it is important
to consider the observation interval.

The reason is that to see the sidebands at fc + fm and fc − fm we must observe

the signal over a sufficiently long time (at least one cycle of the message m(t)).4 If
we observe the signal for a short time over a small fraction of a cycle of m(t), then
INSTRUCTOR COPY

we see only the carrier wave at frequency fc with amplitude a(t) = Ac [1 + µcos2πfm t]
at time t and we don’t see any sidebands.

The second method of showing the result is done using Fourier transform proper-
ties. Note that a Fourier transform is taken over a long time including at least one
cycle of the message m(t), so that we expect to see the sidebands. The result can
also be shown using a Fourier series, where the signal is assumed to be periodic with
a period equal to one cycle of the message.

Thus if the observation interval is much less than the period of the single tone
message, then in the AM time domain waveform s(t), the individual carrier wave
cycles can be seen. The AM spectrum S(f ) in this case would be a a pair of delta
function at ±fc that move up and down in step with the message waveform. No
sidebands at ±fc ± fm would be observed.

If the observation interval is equal to (or greater than) the period of the single
tone message, then the AM time domain waveform s(t) is periodic with period 1/fm
and a Fourier series expansion (or the formula for the product of two cos waves) will
show the presence of the carrier at ±fc and the sidebands at ±fc ± fm in S(f ).

3.1.7 AM with single tone message in only one sideband

In some applications, AM with a single tone message is transmitted with only one
sideband. NDBs in Canada use only one sideband with message frequency f1 =
400 Hz. The waterfall plot for this case is shown in figure fig. 3.15. This NDB has
callsign YCD and is located at the Nanaimo airport, see https://ourairports.com/
4
If we have one complete cycle of m(t) then we have a periodic signal and can take the Fourier
series (or Fourier transform) to observe the sidebands.
 CHAPTER 3. AMPLITUDE MODULATION (AM) 220

navaids/YCD/Nanaimo_NDB_CA/. The carrier is seen at fc = 251 kHz and the single

tone message is Morse code (on-off-keying) at fc ± f1 = 251.4 kHz.
INSTRUCTOR COPY

Figure 3.15: NDB-YCD at 251 kHz

Exercise 54
Find an expression for the complex baseband of an AM signal with a single
tone message in only one sideband.

Solution
For a general message m(t) the solution requires the mathematics in chapter 5.
For a single tone message in only one sideband, the complex baseband is DC
plus a single exponential at the message frequency f1

s̃(t) = Ac [1 + (µ/2)ej2πf1 t ]

If the other sideband is added, then

s̃(t) = Ac [1 + (µ/2)ej2πf1 t + (µ/2)e−j2πf1 t ]

= Ac [1 + µ cos 2πf1 t]

AM with only one sideband is also used for analog television. The message wave-
form is the brightness of each pixel with each scanning line. The message can contain
 CHAPTER 3. AMPLITUDE MODULATION (AM) 221

significant low frequency content, for example, when a series of frames are all-black.
Details are at https://en.wikipedia.org/wiki/Analog_television. An engineer-
ing compromise with analog components is to have double sideband AM at low video
frequencies and single sideband at higher video frequencies. This compromise is called
Vestigial sideband (VSB) since it is AM single sideband but low frequency part (the
vestige) or one sideband is not fully suppressed. VSB is mentioned again in section 5.1.

3.1.8 Power in carrier wave and sidebands

The power of a signal is a quantity of interest and one may wish to calculate the power
of the carrier or sideband components of the signal specifically. Using the time domain
expression for s(t), the power is proportional to the square of the coefficient of each
INSTRUCTOR COPY

cosine term. Using the frequency domain expression S(f ), the power is proportional
to the square of the coefficient of each δ-function term.

The power versus frequency curve is displayed on a spectrum analyzer, typically in

dBm, or decibel relative to one milliwatt at 50 Ω. For this case, P (dBm) = 20 log V +
13 where V is rms voltage, as was shown in Section 1.4.

In what follows we assume 1 Ω rather than 50 Ω for convenience as is typically done

for communication signals mathematics. The power into one ohm [s(t)]2 = |S(f )|2
without the need for scaling factors to account for impedance.

Consider an AM signal with single tone modulation m(t) = Am cos 2πfm t:

s(t) = Ac [1 + µ cos(2πfm t)] cos(2πfc t)
which, is expanded to
s(t) = Ac cos(2πfc t) + (Ac µ/2) cos(2π[fc + fm )]t) + (Ac µ/2) cos(2π[fc − fm ]t) (3.15)
It follows that for the above AM signal the powers of the various spectral components
can be transformed from the amplitudes into

• Carrier power (for both fc and −fc combined):

A2c
Pc =
2
• Upper sideband power (for both fc + fm and −fc + fm combined):
1
PU SB = (Ac µ/2)2
2
A2 µ2
= c
8
 CHAPTER 3. AMPLITUDE MODULATION (AM) 222

• Lower sideband power (for both fc − fm and −fc − fm combined):

1
PLSB = (Ac µ/2)2
2
A 2 µ2
= c
8

Remembering that there are both positive and negative frequency components (fc , −
fc ) this can be decomposed into:

• Power in each carrier at each of ±fc :

P+c =P−c
INSTRUCTOR COPY

A2
= c
4

• Power in each USB at each of ±(fc + fm ):

P+U SB =P−U SB
A2 µ2
= c
16

• Power in each LSB at each of ±(fc − fm ):

P+LSB =P−LSB
A2 µ2
= c
16

The total power in the sidebands can be calculated as a proportion of total power
Ac µ2 2 2 2
PU SB + PLSB 8
+ Ac8µ
=
PU SB + PLSB + Pc A2c µ2 + A2c µ2 + A2c
8 8 2 (3.16)
µ2
=
2 + µ2
Consider that the message is carried in the side bands, and so we want a high pro-
portional of transmission power to contain the message. As the ratio in eq. (3.16)
approaches 1, the system becomes more and more power efficient.

Exercise 55
If the power at fc = 0 dBm, what is the power at fc + fm ?
 CHAPTER 3. AMPLITUDE MODULATION (AM) 223

Solution
Given A2c /2 = 1 mW; (1/8)A2c µ2 = (1 mW)µ2 /4. If µ = 1 then the power at
fc + fm is (1 mW)/4 = 0.25 mW. A drop in power by half is −3 dB, thus one
quarter of the power is −6 dB.

0 dB

−6 dB
INSTRUCTOR COPY

fc − fm fc fc + fm f

Figure 3.16: S+ (f ) with modulation index 1

In the special case where µ = 1 (see fig. 3.16), we have 100% modulation
(minimum envelope value is zero).

3.1.9 Overmodulation

Given the AM signal

s(t) = Ac [1 + µ cos(2πfm t)] cos(2πfc t)

with modulation index µ > 1 (greater than 100%) the real envelope, a(t) = Ac [1 +
µ cos(2πfm t)] becomes less than zero and no longer looks like the message being sent.
This case is called “overmodulation”. The phase of the carrier wave is shifted by 180
degrees when [1 + µ cos(2πfm t)] is less than zero.

In the frequency domain, the sum of the amplitudes (not the power) of the side-
bands exceeds the amplitude of the carrier.

Exercise 56
What happens to the complex envelope s̃(t) when overmodulation occurs?
Write an expression for s̃(t).
 CHAPTER 3. AMPLITUDE MODULATION (AM) 224

Solution

s̃(t) = a(t)ejφ = Ac (1 + µ cos 2πfm t)

= I(t) + jQ(t)
I(t) = Ac (1 + µ cos 2πfm t) may be < 0 when µ > 1
Q(t) = 0
|s̃(t)| = |a(t)ejφ |
= |I(t) + jQ(t)|
p
= I 2 (t) + Q2 (t)
= |Ac [1 + µ cos 2πfm t]|
≥0
INSTRUCTOR COPY

The real envelope |a(t)| is the rectified version of I(t).

µ = 0.5

µAm
−µAm
µ=1

µAm

−µAm
µ=2

µAm

−µAm

Figure 3.17: The effect of the modulation index on an AM signal. The message is the
same for each case but the envelope changes with µ.
 CHAPTER 3. AMPLITUDE MODULATION (AM) 225

Exercise 57
Write an expression for S̃(f ) and S(f ) for the overmodulation case with µ = 2.

µAc µAc
4 4

Ac
2

fc − fm fc fc + fm f fc − fm fc fc + fm f
INSTRUCTOR COPY

(a) AM spectrum with 200% modulation) (b) AM spectrum with overmodulation

Figure 3.18: AM spectra with various modulation levels

3.1.10 Digital messages transmitted using AM

Digital messages can be sent by using a binary modulating signal; the amplitude of
the carrier is multiplied by a high voltage (logic one) and a low voltage (logic zero).
A modulated wave with the capacity to represent a binary bit stream is the result.
This method is known as amplitude shift keying (ASK). The most common case is
where logic 1 is a positive voltage, say +1 volts and a logic 0 is a negative voltage
say -1 volts. In general we can write m(t) = ±Am . Thus we can write the digital AM
signal

s(t) = Ac [1 + ka m(t)] cos(2πfc t)

= Ac [1 ± ka Am ] cos(2πfc t)

If we choose Am such that ka Am = 1, then

s(t) = Ac [1 ± ka Am ] cos 2πfc t

= Ac [1 ± 1] cos 2πfc t
(
2Ac cos 2πfc t, logic 1
=
0, logic 0

depending on whether a logic 1 or a logic 0 was sent. In this case the amplitude is
shifted from 2Ac to 0, and it is called on-off keying (OOK) and can be used for Morse
code transmission.
 CHAPTER 3. AMPLITUDE MODULATION (AM) 226

3.2 AM transmitters

3.2.1 Generating an AM signal digitally in software

Next we look at how to build transmitters and receivers. We have a message that we
wish to send and we need a way to send and receive it.
INSTRUCTOR COPY

Figure 3.19: Conceptual modulation and demodulation of the Signal m(t)

One approach is to use digital means, for example by programming the USRP
using GNU Radio. The USRP includes the DAC and analog IQ mixer, filters and
amplifiers connected to an antenna.

Figure 3.20: Digital transmitter configuration

The transmitted AM signal is of the form:

s(t) = a(t) cos(2πfc t + φ)

(3.17)
= Ac [1 + ka m(t)] cos(2πfc t + φ)
where φ is a constant phase. We can write the AM signal s(t) as the real part of a
complex signal:
s(t) = <[a(t)ejφ ej2πfc t ]
= <[s̃(t)ej2πfc t ]
 CHAPTER 3. AMPLITUDE MODULATION (AM) 227

where the complex envelope:

s̃(t) = a(t)ejφ
= a(t) cos φ + ja(t) sin φ
= i(t) + jq(t)

Usually we choose φ = 0, so that i(t) = a(t) = 1 + ka m(t), q(t) = 0

Given the message m(t), the digital AM transmitter simply implements the equa-
tion

s̃(t) = i(t) = 1 + ka m(t)

s(t) = <[s̃(t)ej2πfc t ]
INSTRUCTOR COPY

3.2.2 Analog-only method to generate an AM signal

In the days before software radio, only analog circuit-theory based methods were avail-
able to generate an AM signal. The mathematics tells us that we want to multiply the
message signal m(t) with the carrier wave c(t), in particular: s(t) = [1 + ka m(t)]c(t).
So we need to build an analog multiplier.

One purely analog method uses a non-linear circuit with output

X
v2 (t) = ai [vi (t)]i = a1 v(t) + a2 v 2 (t) + ...
i
P
Any non-linearity v2 (t) = i ai [vi (t)]i will cause multiplication where i ≥ 2.

This non-linearity could be a diode operating with a small input voltage. The
non-linearity includes a square term where the input is multiplied by itself.
 CHAPTER 3. AMPLITUDE MODULATION (AM) 228

(a)
INSTRUCTOR COPY

(b)

Figure 3.21: Fundamental Analog Transmission Method

As shown by the identity, cos α cos β = 21 [cos(α + β) + cos(α − β)] and also in
the previous section, the result of multiplying two signals is sinusoidal components
located at the sum and the difference of carrier and message frequencies.

We model the non-linearity as the first two terms of the Taylor series expansion
X
v2 (t) = ai [vi (t)]i (3.18)
i=1,2

where the input signal v1 (t) = m(t) + c(t) is the sum (not the product) of the message
 CHAPTER 3. AMPLITUDE MODULATION (AM) 229

and the carrier. Thus we write

v2 (t) = a1 v1 (t) + a2 [v2 (t)]2

v1 (t) = m(t) + c(t)
v1 (t) = m(t) + Ac cos(2πfc t)
∴
v2 (t) = a1 [m(t) + Ac cos(2πfc t)] + a2 [m(t) + Ac cos(2πfc t)]2
v2 (t) = a1 m(t) + a1 Ac cos(2πfc t) + a2 m2 (t) + 2a2 m(t)Ac cos(2πfc t) + a2 A2c cos2 (2πfc t)

Using the trigonometric identity, cos2 (α) = (1/2)[1 + cos(2α)], we arrive at

a2 A2c cos2 (2πfc t) = (1/2)a2 A2c [1 + cos(4πfc t)] (3.19)

INSTRUCTOR COPY

Sorting by frequency we get the following simplified expression:

v2 (t) =a1 m(t) + a2 m2 (t) + (1/2)a2 A2c

+ cos(2πfc t)[a1 Ac + 2a2 Ac m(t)] (3.20)
+ cos(4πfc t)[(1/2)a2 A2c ]

We now apply a band-pass filter centred at fc , to get the desired modulator output
signal  
a2
s(t) = a1 Ac cos 2πfc t[1 + 2 m(t)]
a1

Thus we have created an AM wave s(t) by adding m(t) + c(t) and then using a
non-linearity and band pass filter. For this AM wave µ = 2a2 /a1 .

Exercise 58
Write a frequency domain expression for v2 (t).

Solution

Ac
V2 (f ) =a1 [δ(f − fc ) + δ(f + fc )]
2
+ 2a2 Ac [M (f − fc ) + M (f + fc )]
+ a1 M (f ) + a2 F[m2 (t)]
a2 A2c
+ [δ(f ) + δ(f − 2fc ) + δ(f + 2fc )]
2
 CHAPTER 3. AMPLITUDE MODULATION (AM) 230

Another type of non-linearity is diode switching.

(
v1 (t), c(t) > 0
v2 (t) =
0, c(t) < 0

In this case, the message waveform plus the carrier wave signal is half-wave rectified
and the result is band-pass filtered at the carrier wave frequency.

Exercise 59
Show that the filter output contains an AM wave. Hint:

v1 (t) = c(t) + m(t)

v2 (t) = [c(t) + m(t)]gp (t)
INSTRUCTOR COPY

where gp (t) is a uni-polar square wave at fc .

Solution
Find the Fourier series for the square wave
1 2
v2 (t) = [c(t) + m(t)][ + cos 2πfc t + . . .]
2 π
Ac 4
= [1 + m(t)] cos 2πfc t + . . .
2 πAc

4
which is an AM wave with ka = πAc
plus components at frequencies removed
from fc

3.3 AM receivers

From section 2.1, the first task performed by a general receiver with real passband
input r(t) or complex passband input r+ (t) is to find the complex baseband signal
s̃(t). Once s̃(t) is found, the next task is to find the message m(t) that is contained
in s̃(t). The relationship between s̃(t) and m(t) depends on the modulation type.

Figure 3.22 is a representation of the main principles involved in demodulating a

real passband signal using a digital architecture.
 CHAPTER 3. AMPLITUDE MODULATION (AM) 231

Figure 3.22: Demodulation principles block diagram

3.3.1 Digital software AM receiver in complex notation

3.3.1.1 Received AM signal at receiver input

INSTRUCTOR COPY

The transmitted AM signal is of the form:

s(t) =a(t) cos(2πfc t)

=Ac [1 + ka m(t)] cos(2πfc t)

The received AM signal r(t) = s(t)/L0 will in general have a different phase φr 6= 0
as well as a frequency offset ∆f

r(t) =a(t) cos(2πfc t + 2π∆f t + φr )

=Ac [1 + ka m(t)] cos(2πfc t + φ(t))

where we have assumed L0 = 1 for convenience so that r(t) = s(t) and write φ(t) =
2π∆f t + φr . In practice, L0 ' 10−4 to 10−6 so that the received signal is only a few
microvolts. The preamplifier stage with voltage gain up to 106 is needed to restore
r(t) to a level that is suitable for the ADC (typically a few volts).

3.3.1.2 Demodulation of complex passband AM signal to obtain message

The received AM signal at the receiver input may be written

r(t) = [r+ (t) + r− (t)]/2

r+ (t) = Ac [1 + ka m(t)]ej2πfc t ejφ(t)
= r̃(t)]ej2πfc t
r− (t) = Ac [1 + ka m(t)]e−j2πfc t e−jφ(t)
= r̃∗ (t)]e−j2πfc t
 CHAPTER 3. AMPLITUDE MODULATION (AM) 232

We now apply the receiver in chapter 2 to r+ (t) to obtain the received complex
baseband signal
r̃(t) = Ac [1 + ka m(t)]ej2π∆f t ejφ0
r̃∗ (t) = Ac [1 + ka m(t)]e−j2π∆f t e−jφ0
where
i(t) + jq(t) = r̃(t)

For the case of AM, we know that

s̃(t) = Ac [1 + ka m(t)]ejφ
INSTRUCTOR COPY

To recovers the message m(t) from s̃(t), one method is to recover the magnitude
of the complex envelope
|s̃(t)| = a(t) = Ac [1 + ka m(t)]
and subtract the DC component and scale by 1/(ka Ac ) to obtain m(t).

The value of φ(t) = 2π∆f t + φr does not affect the magnitude of the complex
envelope. Note that φr 6= φ in general, since the channel and/or receiver may shift
the phase as well as the frequency.

To show how this is done in software with the USRP or RTL-SDR and GNU Radio
Companion (GRC), recall that the USRP or RTL-SDR source block has a complex
output s̃(t) with real and imaginary components i(t) and q(t).

In what follows we assume L0 = 1 so that r(t) = s(t). We can write the AM

signal s(t) as the real part of a complex signal:
s(t) =Re[a(t)ejφ(t) ej2πfc t ]
=Re[s̃(t)ej2πfc t ]
where the complex envelope
s̃(t) =a(t)ejφ(t)
=a(t) cos φ(t) + ja(t) sin φ(t)
=i(t) + jq(t)

Thus the USRP source block with frequency set to fc will have outputs:
i(t) = a(t) cos φ(t)
q(t) = a(t) sin φ(t)
 CHAPTER 3. AMPLITUDE MODULATION (AM) 233

To obtain a(t), we take the magnitude of the complex envelope s̃(t). Thus we can
write:

|s̃(t)| =|i(t) + jq(t)|

=|a(t) cos φ(t) + ja(t) sin φ(t)|
=a(t)| cos φ(t) + j sin φ(t)|
q
=a(t) cos2 φ(t) + sin2 φ(t)
=a(t)

This shows that we can recover a (t) = 1 + ka m(t) regardless of the value of φ(t). The
GRC Complex to Magnitude block allows us to obtain the magnitude of the complex
envelope by performing the function a(t) = |i(t) + jq(t)|.
INSTRUCTOR COPY

The IQ receiver local oscillator frequency fLO does not need to match the received
carrier frequency fc exactly. If there is a frequency offset ∆f = fc − fLO , then
φ = 2π∆f t, but as we have just seen, |s̃(t)| = a(t) is not affected by the value of φ
and thus not affected by any time variation in φ caused by a frequency offset ∆f ,
provided that s(t) is within the filter bandwidth of the receiver.

Exercise 60
Write an expression for s(t) and s̃(t) with a frequency offset ∆f .

3.3.1.3 Demodulation of real passband AM signal to obtain the message

For a real passband AM signal s(t), we describe four software receiver designs to
obtain m(t). Each of these receivers includes some kind of filter that will introduce a
small delay in the output. Normally this delay is on the order of many radio frequency
(RF) cycles but only a small fraction of a message frequency cycle.

Receiver 1

This design uses an IQ receiver such as the URSP (section 1.3) or RTL-SDR that
receives s(t) and processes it to generate a complex output with real and imaginary
components i(t) and q(t) and use the complex baseband receiver described above.

Variations of this receiver design work directly with the real passband signal r(t)
using one of the two receiver designs outlined in https://www.mathworks.com/help/
dsp/ug/envelope-detection.html

Receiver 2
 CHAPTER 3. AMPLITUDE MODULATION (AM) 234

This receiver squares and low pass filters the signal s(t) and takes the square root

a(t) = 1 + ka m(t)
s(t) = a(t) cos 2πfc t
s2 (t) = a2 (t) cos2 2πfc t
a2 (t)
= [1 + cos 4πfc t]
2

After low pass filter, scale up by factor of 2 and take square root to obtain a(t).
Subtract the DC component to get the message ka m(t)

Receiver 3
INSTRUCTOR COPY

This receiver design uses the Hilbert transform 5 of s(t) to create an analytic signal

s+ (t) = s(t) + jH{s(t)}

= s(t) + jŝ(t)
= s̃(t)ej2πfc t

and then finds the magnitude of s+ (t)

p
|s+ (t)| = s2 (t) + ŝ2 (t)
= |s̃(t)| · |ej2πfc t |
= |s̃(t)|
= a(t)

which is the same as the magnitude of the complex baseband (complex envelope)
which equals a(t) as shown above. Recall that |z1 z2 | = |z1 | · |z2 | Subtract the DC
component to get the message ka m(t).

3.3.2 Analog AM receiver

3.3.2.1 Diode-resistor-capacitor circuit

Analog demodulation of the received AM wave (shown in eq. (3.21) scaled by the
path loss L0 ) can be achieved via an envelope detector made from analog circuit
components. The envelope represents the modulating message wave. The envelope
detector extracts the message through the use of a rectifying diode and a RC lowpass
5
See the chapter on single sideband systems section 5.2.
 CHAPTER 3. AMPLITUDE MODULATION (AM) 235

filter with cutoff frequency chosen so that the high frequency carrier term at fc is
removed while the message with lower frequency fm plus a DC term remains. A
series capacitor is used to remove the DC.

s(t)
r(t) =
L0
a(t)
= cos(2πfc t + φ) (3.21)
L0
Ac
= [1 + ka m(t)] cos(2πfc t + φ)
L0

Mathematically speaking, the analog envelope detector finds the magnitude of

INSTRUCTOR COPY

the complex envelope of s(t). Figure 3.23 shows the waveform as it exists at different
points in the circuit.

Figure 3.23: AM envelope detector with waveforms shown at each stage.

Exercise 61
Do the mathematics to show that if s(t) is the input to the envelope detector
circuit with rectifier and low pass filter s(t) then the output of the circuit is
the magnitude of the complex envelope of s(t).

Exercise 62
Write expressions for the signal at the input to the diode, at the output of the
diode before the low pass filter, and at the output of the low pass filter.
 CHAPTER 3. AMPLITUDE MODULATION (AM) 236

3.3.2.2 Implementing the analog AM receiver algorithm in software

Each modulation method (AM, DSB-SC, SSB, etc) requires its own unique map that
serves to extract the message m(t) from x(t) and y(t). The maps for receiving m(t)
from these signals for different transmission types are outlined below.
INSTRUCTOR COPY

Figure 3.24: Map to receive AM

Recall for AM

x(t) =vi (t)

=a(t) cos ψ(t)
=Ac [1 + ka m(t)] cos ψ(t)
y(t) =vq (t)
=a(t) sin ψ(t)
=Ac [1 + ka m(t)] sin ψ(t)

The AM receiver map extracts m(t) by first obtaining a(t) = Ac [1 + ka m(t)],

removing the DC component and scaling the signal by 1/ka . a(t) is obtained from
p
a(t) = x2 (t) + y 2 (t))

since
p p
x2 (t) + y 2 (t) = a2 (t)cos2 ψ(t) + a2 (t)sin2 ψ(t)
= a(t)

regardless of the value of ψ(t) = 2π∆f t + φ. Thus an AM receiver will function even
with a frequency error ∆f
 CHAPTER 3. AMPLITUDE MODULATION (AM) 237

3.3.3 Tuning in (selecting) one particular AM signal with the

USRP

The USRP multiplies the real valued radio frequency signal s(t) by e−j2πfc t to generate
i(t) + jq(t). This process is called complex downmixing and is equivalent to the
standard IQ receiver shown in fig. 2.20.

Recall the AM spectrum

Ac Ac ka
S(f ) = [δ(f − fc ) + δ(f + fc )] + [M (f − fc ) + M (f + fc )] (3.22)
2 2
The spectrum S(f ) of the real radio frequency (RF) signal s(t) is symmetric about
zero. The complex downmixing shifts the frequency of S(f ) down by fc .
INSTRUCTOR COPY

s(t) ↔ S(f )
−j2πfc t
e s(t) ↔ S(f + fc )

Exercise 63
Write an expression for S(f + fc ), and show that one of the terms needs to be
filtered out.

Solution
The shifted AM spectrum is
Ac Ac ka
S(f + fc ) = [δ(f ) + δ(f + 2fc )] + [M (f ) + M (f + 2fc )] (3.23)
2 2
After complex downmixing, the resulting signal is complex and the frequency
spectrum S(f + fc ) is no longer symmetric about zero. The terms at −2fc are
filtered out, leaving only the message plus a DC component A2c δ(f )+ Ac2ka M (f ).

The complex signal output from the USRP source block i(t) + jq(t) is bandlimited
to the sampling rate of the USRP source block. The USRP source block output can
be recorded to a file and used again at a later time. This file source will have the
same sampling rate and bandwidth as the USRP sink block used to record it.

With a sampling rate of 256 kHz and complex samples, the bandwidth will be
256 kHz (because the complex signal spectrum is not symmetric and does not have
redundant mirror-image positive and negative frequencies).

The AM radio broadcasting band 535–1605 kHz is divided into 10 kHz channels.
 CHAPTER 3. AMPLITUDE MODULATION (AM) 238

Figure 3.25: Downmixing

The AM aircraft band 108–137 MHz is divided into 25 kHz channels. With a file
source sampled at 256 kHz, there can be as many as 10 different AM aircraft signals.
INSTRUCTOR COPY

The AM aircraft signal with carrier frequency fc = fd (fd is set in the USRP source
block) will appear at 0 Hz after the downconversion (at the USRP source output).
Other signals at carrier frequencies fc ±nf0 kHz will appear at multiples of f0 = 25 Hz
away from 0 Hz.

We need to create a filter to select the one signal we want (the one with carrier
frequency fc that now appears at 0 Hz). A low pass filter with 12.5 kHz cutoff fre-
quency will do the job, since all the other signals are centered at frequencies at least
25 kHz away from 0 Hz.

To “tune in” (receive) one of the other signals, we can shift the spectrum of the
USRP source output by nf0 kHz by multiplying the complex signal i(t) + jq(t) by
e−j2πnf0 t = cos 2πnf0 t − j sin 2πnf0 t, so that the signal that first appeared at nf0 Hz
now appears at 0 Hz.

Exact tuning is not required, since a frequency offset ∆f results in a changing

phase φ(t) = 2π∆f t that does not affect the AM demodulation process.

3.4 Chapter review questions

Q 3.1 For AM (amplitude modulation), we set the amplitude a(t) = Ac [1 + ka m(t)]

and the phase φ(t) = φ to be some constant not equal to zero. The modulated signal
is
s(t) = Ac [1 + ka m(t)] cos [2πfc t + φ]
The amplitude a(t) = Ac [1 + ka m(t)] > 0 is always positive with |ka m(t)| < 1 or
[1 + ka m(t)] > 0 and fc >> W where W is the bandwidth of m(t).
 CHAPTER 3. AMPLITUDE MODULATION (AM) 239

1. Write an expression for s(t), i(t), q(t). Note that i(t), q(t) are related to the
message and are independent of carrier frequency.

2. Now consider a specific message m(t) = Am cos 2πfm t where fm is the message
frequency. It follows that

s(t) = Ac [1 + ka m(t)] cos (2πfc t)

= Ac [1 + µ cos (2πfm t)] cos (2πfc t)

where µ = ka Am < 1 is called the modulation index.

Write s(t) as a sum of 3 cosine waves, using trigonometric identities.

3. Write s(t), s+ (t), s̃(t) for this AM signal with this specific message m(t). See
question 1a above. Recall
INSTRUCTOR COPY

φ(t) = φ
a(t) = Ac [1 + µ cos 2πfm t]

Solution

1.

s(t) = a(t) cos (2πfc t + φ)

= Ac [1 + ka m(t)] cos (2πfc t + φ)
i(t) = a(t) cos φ(t)
= Ac [1 + ka m(t)] cos φ
q(t) = a(t) sin φ(t)
= Ac [1 + ka m(t)] sin φ

2. The problem statement assumes φ = 0.

Ac µ Ac µ
s(t) = Ac cos (2πfc t) + cos [2π (fc + fm ) t] + cos [2π (fc − fm ) t]
2 2
 CHAPTER 3. AMPLITUDE MODULATION (AM) 240

3.

s̃(t) = a(t)ejφ(t)
= i(t) + jq(t)
= Ac [1 + µ cos (2πfm t)] ejφ
s+ (t) = s̃(t)ej2πfc t
= Ac [1 + µ cos (2πfm t)] ej2πfc t ejφ
= AC [1 + µ cos (2πfm t)] ej(2πfc t+φ)
s(t) = Re{s+ (t)}
= Re{s̃(t)ej2πfc t }
= Ac [1 + µ cos (2πfm t)] cos (2πfc t + φ)
INSTRUCTOR COPY
 Chapter 4

Double sideband suppressed

carrier (DSB-SC)
INSTRUCTOR COPY

4.1 Introduction to DSB-SC

Carrier waves require power to transmit but they do not in themselves contain any
information about the message. Sometimes we want to transmit an AM wave without
its carrier wave (this lessens the power requirements for transmission and thus may
be a more economical course of action for certain applications). Filtering out (sup-
pressing) the carrier frequency band leads to a double sideband - suppressed carrier
signal (DSB-SC) which is the subject of this section.

Further filtering out one of either the upper sideband (USB) or lower sideband
(LSB) will produce what is called a single sideband – suppressed carrier signal (SSB-
SC) which is the subject of section 5.2. Some of the advantages and disadvantages of
AM, DSB-SC, and SSB-SC are listed in table 4.1.

241
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 242

Modulation
Advantages Disadvantages
Type
High power requirements due
AM Easily demodulated
to carrier
Carrier location required. Re-
DSB-SC Lower power requirements
dundant SB information
Less bandwidth and power re-
SSB-SC Don’t know where carrier is
quirements

Table 4.1: Amplitude based modulation schemes

4.2 DSB-SC transmitter

INSTRUCTOR COPY

4.2.1 DSB-SC signal mathematics

In this section we will write the mathematical expressions for the signals s̃(t), s+ (t),
˜ ), S+ (t), S− (f ), S(f ) obtained in chapter 2 for the special case of
s− (t), s(t), S(f
DSB-SC.

The expressions for DSB-SC are identical to those for AM except that the DC term
in the complex baseband is omitted (as shown in section 4.2.2.2). However, we must
be careful to recall that for the complex baseband signal s̃(t) = a(t)ejφ(t) = I(t)+jQ(t)
that a(t) may take on only non-negative values a(t) ≥ 0 whereas I(t), Q(t) may take
on any value greater or less than zero. The message m(t) normally has no DC content
on average and thus may take on any value greater or less than zero.

Thus we must not write

a(t) = Ac [m(t)] (4.1)
since a(t) ≥ 0. Thus for DSB-SC, we cannot use the AM expression for a(t) and then
simply remove the DC component from the amplitude.

4.2.1.1 Message signal m(t)

The general message signal m(t) is described in chapter 2. section 2.7.6.1.

 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 243

4.2.1.2 DSB-SC complex baseband s̃(t)

The complex baseband (message) signal s̃(t) = a(t)ejφ(t) = I(t) + jQ(t) is s̃(t) =
Ac [m(t)] for DSB-SC signals.

For DSB-SC signals, we often assume φ(t) = φ is a constant or zero.

For DSB-SC we write the complex baseband signal

s̃(t) = a(t)ejφ(t)
= I(t) + jQ(t)
= Ac m(t)ejφ
INSTRUCTOR COPY

Thus

I(t) = m(t) cos φ

Q(t) = m(t) sin φ

Exercise 64
What is a(t) and φ(t) for DSB-SC with message m(t)?

4.2.1.3 DSB-SC analytic signal s+ (t)

Recall from section section 2.1 the analytic (positive frequency) signal and the nega-
tive frequency signal are written

s+ (t) = s̃(t)ej2πfc t
s− (t) = s̃∗ (t)e−j2πfc t

Substituting the expression for s̃(t) for DSB-SC signals, we obtain

s+ (t) = Ac [m(t)]ejφ ej2πfc t

s− (t) = Ac [m(t)]e−jφ e−j2πfc t
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 244

4.2.1.4 DSB-SC real passband s(t)

Recall from section section 2.1 the real signal s(t) is obtained by adding the positive
and negative frequency components

2s(t) = s+ (t) + s− (t)

= s̃(t)ej2πfc t + s̃∗ (t)e−j2πfc t

Substituting the expression for s̃(t) for DSB-SC signals, we obtain

2s(t) = Ac [m(t)]ejφ ej2πfc t + Ac [m(t)]e−jφ e−j2πfc t

INSTRUCTOR COPY

This result is sum of two complex exponentials, one positive frequency and one neg-
ative frequency. Using the definition ejθ + e−jθ = 2 cos θ with θ = 2πfc t + φ, the
transmitted DSB-SC signal is of the form:
2s(t) = 2a(t) cos(2πfc t + φ)
s(t) = Ac [m(t)] cos(2πfc t + φ)
= Ac m(t) cos(2πfc t + φ)
φ(t) = φ = constant
φ may be set to zero for convenience, but could be any constant value independent
of time. In the case of φ = 0,

s(t) = Ac [m(t)] cos(2πfc t)

= Ac m(t) cos(2πfc t)

From the mathematics we can see that we have multiplied the carrier wave c(t) =
cos(2πfc t) in step with the message m(t), so that s(t) is a DSB-SC signal. Note that
m(t) may be positive or negative, so that at the times when m(t) is negative, the
carrier wave cos 2πfc t is multiplied by a negative number, which results in a π phase
reversal, since − cos 2πfc t = cos(2πfc t + π).

For the special case where there is no message to send, m(t) = 0 and thus s(t) = 0.
m(t) could be an analog or digital message as described in section 2.9.

4.2.2 DSB-SC spectrum for general message

What does the DSB-SC signal look like in the frequency domain?
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 245

We know from section 2.1 the relationship between the complex baseband signal
S̃(f ), the analytic signal S+ (f ) and the real signal S(f )

2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))

This expression will be the starting point for finding the DSB-SC spectrum in this
section.

This expression is valid when the observation interval (i.e. the time spent ob-
serving s̃(t), s+ (t), s(t)) is long enough that the Fourier transform operation is
valid. The observation interval must be greater than the inverse of the lowest fre-
quency component greater than zero in the complex baseband signal. In a discrete-
time (sampled) system where the continuous-time Fourier transforms used in this
INSTRUCTOR COPY

section become a Short-time Discrete Fourier transform, the observation interval

(frame/window/segment size) must contain at least one full period of the signal.

4.2.2.1 Message spectrum M (f )

We use the same example message spectrum M (f ) as shown in fig. 3.5.

M (f )

4Am
3Am
2Am
Am
f
−B B

Figure 3.5 (repeated): Message spectrum M (f ) for a general message.

Since m(t) is real, m∗ (t) = m(t) so that the spectrum M ∗ (−f ) = M (f ). This
means that the magnitude of the mirror image spectrum |M ∗ (−f )| about f = 0 (recall
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 246

section 2.5.2) is the same as the magnitude of the original spectrum |M (f )|. Thus the
magnitude |M (f )| must be symmetric about f = 0. The phase of the mirror image
spectrum is the opposite sign arg M ∗ (−f ) = − arg M (f ) and is thus anti-symmetric
about f = 0.

4.2.2.2 DSB-SC complex baseband spectrum S̃(f )

In the case of DSB-SC

s̃(t) = a(t)ejφ(t)
= I(t) + jQ(t)
= Ac m(t)ejφ
INSTRUCTOR COPY

Thus using the Fourier transform properties

1 ↔ δ(f )
m(t) ↔ M (f )
m∗ (t) ↔ M ∗ (−f )

we obtain

S̃(f ) = Ac M (f )ejφ

We also write the complex conjugate S̃ ∗ (−f ) that we will need later to obtain an
expression for the real signal spectrum S(f ).

S̃ ∗ (−f ) = Ac M ∗ (−f )e−jφ

= Ac M (f )e−jφ

since for a real message m∗ (t) = m(t), we have M ∗ (−f ) = M (f ).

Note that

• S̃(f ) = M (f ) except for a complex scaling factor Ac ejφ

• the message spectrum M (f ) centered around f = 0 and is non-zero for both

positive and negative frequencies as shown in fig. 3.5.
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 247

|S̃(f )|

4Am
3Am
2Am
Am
f
INSTRUCTOR COPY

−B B

Figure 4.1: Complex baseband spectra for a general DSB-SC signal. Note that this
is the same as fig. 3.6 except without the DC component.

The magnitude of the complex baseband spectrum is symmetrical about f = 0

|S̃ ∗ (−f )| = |S̃(f )| = Ac M (f )

The phase of the complex baseband spectrum is anti-symmetric about f = 0

arg{S̃ ∗ (−f )} = − arg{S̃(f )} = − arg M (f ) − φ

4.2.2.3 DSB-SC analytic signal spectrum S+ (f )

Substituting the expression for S̃(f ) and S̃ ∗ (−f ) for DSB-SC signals, we obtain the
spectrum of the analytic (positive frequency) signal for AM.

S+ (f ) = S̃(f − fc )
= Ac M (f − fc )ejφ
S− (t) = S̃ ∗ (−(f + fc ))
= M̃ ∗ (−(f + fc ))
= Ac M (f + fc )e−jφ

since for a real message m∗ (t) = m(t), we have M ∗ (−f ) = M (f ).

 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 248

Note the symmetry of |S+ (f )| about f = fc and the symmetry of |S− (f )| about
f = −fc . The DSB-SC analytic signal spectrum S+ (f ) is the message spectrum M (f )
shifted up by fc (fig. 4.2).

|S+ (f )|

4Am
3Am
2Am
INSTRUCTOR COPY

Am
f
fc − B fc fc + B

Figure 4.2: The analytic signal spectrum S+ (f ) for DSB-SC is identical to that of
AM except without a carrier component at fc .

4.2.2.4 DSB-SC real passband spectrum S(f )

The spectrum of the real DSB-SC signal is the sum of the positive and negative
frequency components

2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))
= Ac M (f − fc )ejφ +Ac M (f + fc )e−jφ

We note there are multiple symmetries because both the signal S(f ) and the
message M (f ) are real.

• the magnitude |S(f )| is symmetric about f = 0 and the phase is anti-symmetric

about f = 0 (fig. 4.3).

• the magnitude |S+ (f )| is symmetric about f = fc and the magnitude |S− (f )|

is symmetric about f = −fc
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 249

|2S(f )|
4Am
3Am
2Am
Am f
−fc fc

Figure 4.3: The real passband spectrum 2S(f ) for DSB-SC is identical to that of AM
INSTRUCTOR COPY

except without carrier components at fc and −fc .

If we choose φ = 0 then
2S(f ) = Ac [M (f − fc ) + M (f + fc )]
S(f ) = (Ac /2)[M (f − fc ) + M (f + fc )]

This result is correct but hides the anti-symmetric phase and omits the fact that
an expression for any spectrum S(f ) (including the spectrum M (f ) of a real signal)
is complex in general.

It is useful at this point to compare the time and frequency domain expressions
for the DSB-SC signal with φ = 0

s(t) = Ac m(t) cos(2πfc t) (4.2)

S(f ) = (Ac /2)[M (f − fc ) + M (f + fc )] (4.3)
and observe a useful Fourier transform property: the Fourier transform of a real
function m(t) multiplied by a cosine is the spectrum of m(t) shifted up and down by
the frequency of the cosine term.

4.2.3 DSB-SC Modulation starting with real passband

In this section, we obtain expressions for DSB-SC signals starting with the real pass-
band signal, whereas the previous section obtained these expressions starting with
the complex baseband.
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 250

4.2.3.1 DSB-SC in time domain

For a double sideband signal we know that the only difference from the AM case is
the lack of carrier term. Thus with a message m(t) = Am cos 2πfm t and a carrier
wave c(t) = Ac cos 2πfc t, the DSB-SC signal is represented by

s(t) = m(t)c(t)
= Ac Am cos(2πfm t) cos(2πfc t) (4.4)
Ac Am
= [cos(2π[fc − fm ]t) + cos(2π[fc + fm ]t)]
2
where we let the constant Ac Am represent any scaling that has taken place by way of
amplification or any hardware effects. We have used the trigonometric identity
INSTRUCTOR COPY

1
cos α cos β = [cos(α − β) + cos(α + β)]
2

For DSB-SC with message m(t) = Am cos 2πfm t, I(t) = Ac cos(2πfm t) and Q(t) =
0. A DSB-SC modulator is simply a multiplier that multiplies the message signal m(t)
with the carrier c(t) as illustrated in fig. 4.4.

Figure 4.4: DSB-SC modulator block diagram

Figure 4.5 shows the waveforms (m(t), c(t), s(t) respectively) for DSB-SC. Note
the phase reversal of the carrier wave when the message signal is less than zero.
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 251
INSTRUCTOR COPY

Figure 4.5: DSB-SC waveforms

DSB may be viewed as an extreme case of AM overmodulation

s(t) = Ac [1 + µ cos(2πfm t)] cos(2πfc t)

= Ac cos(2πfc t) + Ac µ cos(2πfm t) cos(2πfc t)

with µ  1 and Ac  1 and µAc held constant. In this case, the power in the
sidebands is much greater than the power in the carrier. In the limit µ → ∞ and
Ac → 0 and µAc = 1, the overmodulated AM signal becomes a DSB signal with zero
power in the carrier.

4.2.3.2 DSB-SC spectrum

The spectrum M (f ) of the single tone message m(t) = Am cos 2πfm t is shown in
fig. 4.6. M (f ) is symmetrical about zero Hz.
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 252

Figure 4.6: Spectrum of single tone message

The spectrum of a DSB signal with carrier c(t) = Ac cos 2πfc t and message m(t) =
Am cos 2πfm t is two peaks at fc − fm and fc + fm (and no power at fc ) as shown in
fig. 4.7. The DSB signal spectrum will also include negative frequencies with peaks
at −(fc − fm ) and −(fc + fm ) (not shown in fig. 4.7).
INSTRUCTOR COPY

Figure 4.7: DSB spectrum showing both peaks

Notice the general principle at work here: multiplying two cos waves together
yields the sum and difference frequencies, as can be seen from the trigonometric
identity cos α cos β = 21 [cos(α − β) + cos(α + β)].

There is another related principle at work here. Notice that in the expression

s(t) = m(t)c(t) = Ac Am cos 2πfm t cos 2πfc t

the message signal is multiplied by the carrier wave. Whenever a message signal is
multiplied by a carrier wave, the spectrum of the message signal is shifted both up
and down by the carrier frequency.

Observe that because the message spectrum contains both positive and negative
frequencies, the signal spectrum contains components both above and below the (sup-
pressed) carrier frequency.

Figure 4.8 shows the DSB spectrum as the limiting case of an AM signal with a
small but non-zero carrier.
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 253

Figure 4.8: DSB as limiting case of overmodulated AM with weak but non-zero carrier
INSTRUCTOR COPY

4.2.3.3 Observation interval

To interpret the time and frequency domain waveforms for DSB, it is important to
consider the observation interval.

The time domain DSB waveform clearly shows the presence of the carrier wave,
so how can it be suppressed? The reason is that the carrier wave phase shifts 180
degrees at the times when the message waveform m(t) is negative.

If the observation interval is much less than the period of the single tone message,
then in the DSB time domain waveform s(t), the individual carrier wave cycles can
be seen and the phase shifts can be seen. The carrier is definitely present, but its
sign changes from positive to negative. The DSB spectrum S(f ) in this case would
be a a pair of delta function at ±fc that move up and down in step with the message
waveform, and thus goes negative half of the time. No sidebands at ±fc ± fm would
be observed.

If the observation interval is equal to (or greater than) the period of the single
tone message, then the DSB time domain waveform s(t) is periodic with period 1/fm
and a Fourier series expansion (or the formula for the product of two cos waves) will
show the presence of the sidebands at ±fc ± fm in S(f ). The carrier wave is averaged
out to zero because of the phase shifts (sign changes) mentioned above.

Another perspective on DSB is to consider two cos waves at ±fc ± fm and adding
them together to make a DSB signal. The time domain result will look like a cos
wave at ±fc with ”beats” (amplitude variations, positive and negative) at fm .
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 254

Exercise 65
For a DSB signal s(t), write an expression for S(f ).

Exercise 66
For a DSB signal, find the power in each sideband.

Exercise 67
For a DSB-SC signal, find the complex envelope.

Exercise 68
Write the spectrum of a DSB signal with a single tone message using delta
INSTRUCTOR COPY

functions.

Exercise 69
Write the spectrum of a DSB signal with a general message m(t) ↔ M (f ).

4.3 DSB-SC receiver

4.3.1 Digital software DSB receiver in complex notation

4.3.1.1 Received DSB-SC signal at receiver input

The transmitted DSB signal is of the form:

s(t) =a(t) cos(2πfc t)

=Ac m(t) cos(2πfc t)

The received DSB signal r(t) = s(t)/L0 will in general have a different phase
φr 6= 0 as well as a frequency offset ∆f

r(t) =a(t) cos(2πfc t + 2π∆f t + φr )

=Ac [m(t)] cos(2πfc t + φ(t))

where we have assumed L0 = 1 for convenience and write φ(t) = 2π∆f t + φr . As in

the AM case considered previously, L0 ' 10−4 to 10−6 so that the received signal is
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 255

only a few microvolts. The preamplifier stage with voltage gain up to 106 is needed
to restore r(t) to a level that is suitable for the ADC (typically a few volts).

4.3.1.2 Demodulation of complex passband DSB-SC signal

The received DSB signal at the receiver input may be written

r(t) = [r+ (t) + r− (t)]/2

r+ (t) = Ac m(t)ej2πfc t ejφ(t)
= r̃(t)ej2πfc t
r− (t) = Ac m(t)e−j2πfc t e−jφ(t)
INSTRUCTOR COPY

= r̃∗ (t)e−j2πfc t

where

φ(t) = 2π∆f t + φ0

We now apply the receiver in section 2.9.2 (figs. 2.20 and 2.21) to r+ (t).

This receiver multiplies r+ (t) by a complex exponential e−j2πfc t to obtain the

received complex baseband signal

r̃(t) = Ac m(t)ej2π∆f t ejφ0

where

r̃(t) = i(t) + jq(t)

The receiver could also multiply r− (t) by a complex exponential e+j2πfc t to obtain
the complex conjugate of the received complex baseband signal (fig. 4.9).

r̃∗ (t) = Ac m(t)e−j2π∆f t e−jφ0

where

r̃∗ (t) = i(t) − jq(t)

 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 256

r− (t) r̃∗ (t)

ej2πfc t
Figure 4.9: Complex receiver to obtain a complex baseband signal
INSTRUCTOR COPY

For the case of DSB-SC, we know that

s̃(t) = Ac ejφ(t) m(t)

If both ∆f and φ0 are zero, then φ(t) = 2π∆f t + φ0 = 0 and

r̃(t) = s̃(t) = Ac m(t)

is exactly the desired receiver output.

The case where ∆f 6= 0 and/r φ0 6= 0 is considered in a later section.

4.3.1.3 Demodulation of real passband DSB-SC

Recall this time that a DSB wave is comprised of only the two (redundant) message
frequency bands (the carrier has been filtered out before transmission):

s (t) =m (t) c (t)

(4.5)
=m (t) cos (2πfc t)

The modulated double sideband signal has no power at the carrier frequency and
therefore exhibits a periodic phase reversal which makes the simple envelope detector
described previously inadequate for demodulation (the rectifying diode will leave us
with a rectified version of the envelope: received envelope = |m(t)|).

How to properly receive m(t) from s(t) for a DSB-SC signal? One general method
may be understood by observing the result of multiplying the modulated wave s(t)
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 257

by the carrier frequency, c(t) (fig. 4.10). Let r(t) represent the received signal within
the receiver structure.

r(t) =s(t)c(t)
=m(t)c(t)c(t)
=m(t) cos2 (2πfc t)
1
=m(t)[ (1 + cos 4πfc t)]
2
1 1
= m(t) + cos 4πfc t
2 2

r(t) 2 m(t)
INSTRUCTOR COPY

LPF

cos(2πfc t)

Figure 4.10: DSB receiver to demodulate a real passband signal into the message
signal

Thus we have created a signal comprised of a high frequency term (a DSB signal)
and an amplitude scaled version of our message. All that must be done to extract m(t)
from the signal is to apply a lowpass filter to the signal: rLP (t) = 12 m(t) (fig. 4.10).

This method, however effective, requires the party operating the receiver to know
the exact carrier frequency, fc and the phase of the carrier wave. Both instances of
the carrier wave (that used here at the receiver and that which is used initially in the
modulation process at the transmitter) must both have exactly the same phase and
frequency.
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 258

4.3.2 DSB receiver with frequency and phase offset

4.3.2.1 Demodulation of complex passband DSB-SC signal

The received DSB signal at the receiver input may be written

r(t) = [r+ (t) + r− (t)]/2
r+ (t) = Ac m(t)ej2πfc t ejφ(t)
where φ(t) = 2π∆f t + φ0 is the frequency and phase offset.

We apply the receiver in section 2.9.2 to r+ (t) but with a complex exponential at
a local oscillator frequency and phase set to cancel the frequency and phase offset.
The LO signal is written
INSTRUCTOR COPY

cLO (t) = e−j2πfc t e−jφ(t)

= e−j2πfc t e−j2π∆f t e−jφ0
= e−j[2π(fc +∆f )t+φ0 }

to obtain the received complex baseband signal.

r̃(t) = r+ (t)cLO (t)
= Ac m(t)
which is the desired receiver output s̃(t).

The challenge for the receiver is to know the values of ∆f and φ0 . In section 4.3.2.3
we show how to determine these values automatically via a feedback loop.

4.3.2.2 Demodulation of real passband DSB-SC signal

When the frequency and phase are not exactly the same, the error terms (frequency
error ∆f and phase shift φ) may be represented by ψ = φ(t) = 2π∆f t + φ0 . Using
the appropriate trig identities, we now may write the received signal before the low
pass filter
r(t) =m(t) cos(2πfc t) cos[2πfc t + 2π∆f t + φ]
=m(t) cos(2πfc t) cos[2πfc t + ψ]
=m(t) cos(2πfc t)[cos(2πfc t) cos ψ − sin(2πfc t) sin ψ]
=m(t) cos2 (2πfc t) cos ψ − m(t) cos(2πfc t) sin(2πfc t) sin ψ
m(t) m(t)
= [1 + cos(4πfc t)] cos ψ − sin(4πfc t) sin ψ
2 2
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 259

Thus a more general resultant signal is attained when we take into account the
possibility of frequency error and phase shift in the local oscillator that provides the
receiver’s version of c(t).

After lowpass filtering this signal, the receiver output is

m (t)
rLP (t) = cos(ψ) (4.6)
2

So if ψ = 0 then we get the ideal case where cos ψ = 1 and the output of the LPF
is a scaled version of the message. If, however, ψ 6= 0 then we get the above LPF
output where the message is multiplied by the cosine of the error term.

Various values of ψ lead to different effects on the extracted message shown in

INSTRUCTOR COPY

table 4.2.

Specific details of Effect on rLP (t)

ψ
ψ = 2π∆f t + φ rLP (t) = m(t)
2
cos(2π∆f t + φ)
Message will be scaled and will be distorted
∆f = 0 r(t) = k m(t)
2
where k = cos φ is a scaling factor
ψ = φ = constant Message can be recovered
φ = 0, ψ = 2π∆f t r(t) = m(t)
2
cos(2π∆f t)
Message will include distortion
ψ=0 rLP (t) = m(t)
2
Exact message is readily recoverable

Table 4.2: Effect of various ψ values on rLP (t)

The distorted signals can be interpreted by considering the frequency error cosine
term cos 2π∆f t to be a sort of low frequency “carrier”, where typically ∆f = 100 Hz.

Assuming m(t) = cos 2πfm t, with typical fm = 1000 Hz the receiver output

r(t) =m(t) cos 2π∆f t = cos 2πfm t cos 2π∆f t

=0.5 cos 2π(fm + ∆f )t + 0.5 cos 2π(fm − ∆f )t

has outputs with frequencies at fm + ∆f and fm − ∆f , so the message at frequency

fm now appears in two places, at fm − ∆f and ∆f + fm . For our example, with
frequency error 100 Hz, the message at 1000 Hz now appears at 900 and 1100 Hz.

We may extract the true message frequency only if we know the exact carrier
frequency. Typical error ratios compared to carrier frequency ( ∆f
fc
) are as follows:
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 260

a factor of 10−4 for cheap equipment, 10−6 for a good crystal oscillator, 10−11 for a
Rubidium oscillator, and approaching 10−13 with GPS discipline (control).

Later we will see a method whereby a DSB-SC receiver can find the frequency
error and eliminate it.

4.3.2.3 Frequency/phase offset compensation - Costas loop receiver

A frequency and/or phase offset φ(t) = 2π∆f t + φ0 may be compensated for via a
feedback system using a phase-locked loop.

We assume that the LO signal does not know the offset initially and thus is written
INSTRUCTOR COPY

cLO (t) = e−j2πfc t

= e−j2πfc t e−j2π∆f t e−jφ0
= e−j[2π(fc +∆f )t+φ0

so that in this case we obtain the received complex baseband signal

r̃(t) = r+ (t)cLO (t)

= Ac m(t)ej2πfc t ejφ(t) e−j2πfc t
= Ac m(t)ejφ(t)
= Ac m(t)ej[2π∆f t+φ0 ]}
= Ac m(t) cos[2π∆f t + φ0 ] + jAc m(t) sin[2π∆f t + φ0 ]
= i(t) + jq(t)

To obtain the desired receiver output s̃(t) = r̃(t) = Ac m(t), we multiply i(t) by q(t)
to obtain an error signal e(t) = i(t)q(t) that is low pass filtered and used to control
the frequency of the LO.

This is an example of a phase-locked loop feedback system that will be seen again
in later sections on PSK and digital bit synchronizers.

4.3.3 Review of I-Q receivers

Here we review the general I-Q receiver configuration that may be implemented for
all types of signals The receiver for every type (for AM, DSB-SC, SSB) consists of
a standard configuration coupled with a “map” that is specific to the transmission
type.
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 261

s(t) =a(t) cos[2πfc t + ψ(t)]

=a(t)[cos(2πfc t) cos ψ(t) − sin(2πfc t) sin ψ(t)]
=a(t) cos ψ(t)cos2πfc t − a(t) sin ψ(t) sin 2πfc t
=vi (t)cos2πfc t − vq (t) sin 2πfc t

where a(t) is the amplitude and ψ(t) is the phase. The coefficients of the carrier
waves in the transmitter are referred to as vi (t) and vq (t) respectively, where

vi (t) = a(t) cos ψ(t)

vq (t) = a(t) sin ψ(t)
INSTRUCTOR COPY

These signals are formed by the map and used to create s(t) at the IQ Transmitter.

s(t) = vi (t) cos(2πfc t) − vq (t) sin(2πfc t)

Figure 4.11: General IQ Transmitter

The IQ Receiver multiplies the incoming signal s(t) by two versions of the carrier
wave functions: cos(2πfc t) and the 90 degree phase-shifted version sin(2πfc t).

Recall the real DSB-SC receiver uses only one version of the carrier wave and thus
has a single cosine oscillator.

The general I-Q receiver has two oscillators, cos and sin as shown in fig. 4.12.
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 262
INSTRUCTOR COPY

Figure 4.12: General IQ Receiver

4.3.3.1 Review of AM and DSB receivers in real notation

The receiver must determine a best guess of the information stored in the signals vi (t)
and vq (t). The generic IQ receiver architecture produces x(t) and y(t) by multiplying
the received signal s(t) by local carriers cos 2πfc and sin 2πfc t. The algebra describing
the production of these signals assuming no frequency error appears below. We will
use the identities

cos α cos β = [cos(α − β) + cos(α + β)]/2

cos2 α = [1 + cos(2α)]/2
sin α sin β = [cos(α − β) − cos(α + β)]/2
sin α cos β = [sin(α − β) + sin(α + β)]/2

For x(t), before the LPF we have,

s(t) cos 2πfc t =[vi (t) cos(2πfc t) − vq (t) sin(2πfc t)] cos(2πfc t)
=vi (t) cos2 (2πfc t) − vq (t) sin(2πfc t) cos(2πfc t)
vi (t) vq (t)
= [1 + cos(2π[2fc ]t)] − sin(2π[2fc ]t)
2 2
vi (t) vi (t) vq (t)
= + cos(2π[2fc ]t) − sin(2π[2fc ]t)
2 2 2
Recall vi (t) and vq (t) are formed from the message frequency only, so of their frequency
components are at or near fm . The expression vi (t) cos[2π(2fc )t] results in two terms
at the sum and difference frequencies (2fc ± fm ), both of which are near double the
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 263

carrier frequency. The same can be said of vq (t) sin[2π(2fc )t]. Thus after the LPF is
applied to s(t) cos 2πfc t the only remaining term will be x(t) = vi2(t) .

Similarly for y(t), before the LPF we have,

s(t) sin 2πfc t =[vi (t) cos(2πfc t) − vq (t) sin(2πfc t)] sin(2πfc t)
=vi (t) sin(2πfc t) cos(2πfc t) − vq (t) sin2 (2πfc t)
vi (t) vq (t) vq (t)
= sin(2π[2fc ]t) − + cos(2π[2fc ]t)
2 2 2
vq (t)
And after the LPF we end up with y(t) = 2
.

We often write that the IQ receiver output x(t) = vi and y(t) = vq (t), thus
neglecting the scaling factor 21 , since this scaling factor in real hardware will depend
INSTRUCTOR COPY

on the gains of the amplifiers.

The receiver will still work in the case where ψ = 2π∆f t + φ 6= 0 as long as the
oscillator that produces the signals cos 2πfc t and sin2πfc t is corrected such that its
frequency mirrors the error (a voltage controlled oscillator, VCO, outputs at frequency
2π(fc + ∆f )t + φ so that the apparent carrier frequency received is the same as that
which is used in demodulation).

When we include the possibility of frequency error so to represent the most general
case, the incoming signal can be represented by s(t) = vi (t) cos(2π[fc + ∆f ]t + φ) −
vq (t) sin(2π[fc + ∆f ]t + φ).

4.4 Chapter review questions

Q 4.1 Write s̃(t), s+ (t), s(t) for a DSB signal with specific message m(t) = Am cos 2πfm t
where fm is the message frequency. Assume ka Am = µ, assume ka = 1, thus µ = Am .
It is also often valid to assume µ = 1 for DSB.
 CHAPTER 4. DOUBLE SIDEBAND SUPPRESSED CARRIER (DSB-SC) 264

Solution

s̃(t) = a(t)ejφ(t)
= i(t) + jq(t)
= Ac µ cos (2πfm t)
s+ (t) = s̃(t)ej2πfc t
= Ac µ cos (2πfm t) ej2πfc t
= Ac µ cos (2πfm t) ej2πfc t
s(t) = Re{s+ (t)}
= Re{s̃(t)ej2πfc t }
INSTRUCTOR COPY

= Ac µ cos (2πfm t) cos (2πfc t)

Q 4.2 For the DSB signal

Ac Am Ac Am
s(t) = cos 2π (fc + fm ) t + cos 2π (fc − fm ) t
2 2
find S(f ) = F{s(t)}.

Solution

F ourier
m(t) −−−−→ M (f )
F ourier 1
Am cos (2πfm t) −−−−→ Am [δ (f − fm ) + δ (f + fm )]
2
Ac Am F ourier Ac Am
cos (2π (fc + fm ) t) −−−−→ [δ (f − (fc + fm )) + δ (f + fc + fm )]
2 2
Ac Am F ourier Ac Am
cos (2π (fc − fm ) t) −−−−→ [δ (f − (fc − fm )) + δ (f + fc − fm )]
2 2

For positive frequencies only

AC Am
S+ (f ) = [δ (f − (fc + fm )) + δ (f − (fc − fm ))]
2
 Chapter 5

Single Side Band modulation

(SSB)
INSTRUCTOR COPY

5.1 Single sideband suppressed carrier overview

SSB type modulation may be viewed as double sideband with one of the sidebands
removed. If the lower sideband is removed, then the result is an upper sideband
(USB) signal. If the upper sideband is removed, then the result is a lower sideband
(LSB) signal. The S+ (f ) spectra of these AM, DSB, USB, and LSB are shown in
fig. 5.1 for a signal that has a dual tone message.

265
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 266

AM USB

Ac Ac

µAc µAc
µAc µAc
2 2

f f
fc fc
DSB-SC LSB

Ac Ac
INSTRUCTOR COPY

µAc µAc
µAc µAc
2 2

f f
fc fc

Figure 5.1: AM, DSB-SC, USB, and LSB spectra (S+ (f )) for a signal with a dual
tone message.

Ironically, we will see that to remove the sideband, we add an additional term to
the expression for DSB-SC to obtain the expression for SSB-SC.

The term SSB usually means SSB-SC, but there does exist SSB where the carrier
is not suppressed, so that it is like AM with one sideband removed. A closely related
scheme is vestigial sideband (VSB) which is AM with one sideband almost completely
removed but a small amount (a vestige) near zero frequency remaining (fig. 5.2), used
to transmit analog TV video. The reason for retaining the vestige is to simplify the
filtering which allows the carrier and one sideband and a bit of the opposite sideband
close to the carrier to pass through.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 267

S̃(f )

f
−B B

Figure 5.2: VSB signal

INSTRUCTOR COPY

The information content in all variations of AM (AM-DSB-VSB-USB-LSB-ISB)

is in the sidebands. The advantage of SSB is that all the available transmit power is
used for one sideband only, whereas for DSB the available transmit power is divided
between two sidebands and for AM the available power is divided between the carrier
wave as well as the two sidebands. Thus SSB is more power-efficient than DSB or
AM.

5.1.1 SSB applications

SSB is commonly used for long distance over-the-horizon voice communications on the
shortwave (HF) bands 1.8–30 MHz that can propagate worldwide via the ionosphere.
Radio services using SSB include marine mobile, aeronautical mobile and Amateur
radio.

Example SSB (upper sideband, USB) signals near 14.2 MHz are shown in fig. 5.3.
The USB spectrum magnitude tends to be larger at lower frequencies corresponding
to vowels that are in the range 300-1000 Hz. The USB spectrum magnitude is smaller
at higher frequencies corresponding to consonants in the range 1000-3000 Hz. The
magnitude of a voice spectrum tends to go down as the frequency goes up, exactly
the opposite from the example message spectra used in chapter 2.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 268
INSTRUCTOR COPY

Figure 5.3: Upper sideband signals near 14.2 MHz.

SSB may also be used on other frequencies where power efficiency is important
such as Amateur radio low earth orbit satellites at 145 MHz and 435 MHz and also the
QO-100 geostationary satellite at 10.5 GHz, see https://eshail.batc.org.uk/nb/
and fig. 5.4.

Figure 5.4: Upper sideband signals near 10 489.75 MHz.

5.1.2 SSB may be generalized to independent sideband (ISB)

SSB may be generalized to independent sideband (ISB) which is a double sideband

(DSB) signal where the two sidebands are different. In other words, ISB is a com-
bination of a USB and a LSB signal using the same carrier, where the USB and the
LSB are different. An ISB signal carries two different messages, one message using
the LSB and one message using the USB. The ISB spectrum will be asymmetrical
because the sidebands are different as shown in fig. 5.5.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 269

−fc fc f

Figure 5.5: ISB signal carrying two different messages, a tri-tone message in the upper
sideband and a dual tone in the lower sideband.
INSTRUCTOR COPY

Any complex passband signal (that must have an asymmetric spectrum because
it is complex) can be represented as an ISB signal centered at some frequency within
the frequency range of the spectrum of that complex signal (fig. 5.6). Thus an ISB
signal with asymmetrical spectrum can represent any complex baseband signal that
has been shifted to a carrier frequency as per the fundamental result in section 2.5.2.
This result is consistent with the idea that any complex baseband signal can carry
two messages I(t), Q(t) or a(t), φ(t), and the ISB signal can also carry two messages,
one for the USB and one for the LSB.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 270

LSB

f
fc

USB

f
fc
INSTRUCTOR COPY

ISB

f
fc

Figure 5.6: S+ (f ) for an ISB signal atfc and it’s component USB and LSB messages.

ISB may be generalized for the case where each sideband includes a group of several
or many signals within its bandwidth. Each such signal has its own “subcarrier”each
with its own complex baseband message shifted up to the subcarrier frequency.

The asymmetrical spectrum and waterfall plots in section 1.5.1 containing many
signals are examples of such an ISB signal, where the carrier frequency is typically
in the center of the spectrum. The waterfalls represent an analytic signal (since the
negative frequency mirror image is not shown).

In the following sections, we will show

• how an ISB complex baseband signal is related to a general complex baseband

signal
• how to create a general analytic message (complex baseband signal) m+ (t) with
positive frequencies only from the real message m(t) using the so-called Hilbert
transform,
• how to derive the SSB real passband signal from the analytic signal
• how to build SSB transmitters and receivers.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 271

5.2 SSB-SC complex baseband signal

5.2.1 SSB and ISB complex baseband in terms of a general

complex baseband signal

A general complex baseband signal s̃(t) ⇔ S̃(f ) may be split into its positive and
negative frequency components

2s̃(t) = s̃+ (t) + s̃− (t)

2S̃(f ) = S̃+ (f ) + S̃− (f )

as shown in fig. 5.7, where fc = 0.

INSTRUCTOR COPY

LSB

USB

ISB

Figure 5.7: S̃(f ) for an ISB signal and it’s component USB and LSB components,
fc = 0.

The positive frequency component s̃+ (t) ↔ S̃+ (f ) of a complex baseband signal
centered at zero frequency is not to be confused with the positive frequency component
s+ (t) ↔ S+ (f ) of a real passband signal that is centered at a carrier frequency fc ,
see chapter 2. However, both s̃(t) and s+ (t) are analytic signals as presented in
section 2.6.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 272

The general complex baseband signal S̃(f ) may be considered as an ISB signal,
where S̃+ (f ) is the upper sideband (USB) and S̃− (f ) is the lower sideband (LSB).
Each sideband can carry an independent message, so S̃(f ) can carry two messages.

We have seen before that S̃(f ) can carry two real messages (e.g. stereo audio) or
one complex message. The two real messages m1 (t), m2 (t) can be carried by a pair
of attributes of S̃(f ) in 3 different ways as follows:

message 1 m1 (t) = m(t) message 2 m2 (t) = n(t)

m(t) = a(t) = |s̃(t)| n(t) = φ(t) = arg s̃(t)
m(t) = i(t) = <{s̃(t)} n(t) = q(t) = ={s̃(t)}
USB s̃+ (t) LSB s̃− (t)

We define m1 (t) = m(t) and m2 (t) = n(t) to avoid double subscripts in the
INSTRUCTOR COPY

following sections. 1

The question arises how to find a time domain expression for

• the USB s̃+ (t) in terms of the first message m(t)

• the LSB s̃− (t) in terms of the second message n(t)

Note that the messages are real signals containing both positive and negative fre-
quencies. Thus m(t) 6= s̃+ (t) and n(t) 6= s̃− (t).

This expression will have to

• eliminate the negative frequencies from the first message m(t) and
• eliminate the positive frequencies from the second message n(t).

This is exactly what was done in section 2.6 on analytic signals. In the following
section, we review analytic messages, and then we are ready to answer the above
question.

5.2.2 Analytic messages

In this subsection, we consider analytic messages, i.e. the positive frequency com-
ponent of a message m(t). These analytic messages will be useful in the following
subsection.
1
The notation n(t) is often used to represent noise, but in this chapter, n(t) is a real baseband
message.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 273

Consider a complex baseband signal with zero imaginary component as was done
in section 2.7.6. Thus for this special case we write

s̃(t) = I(t) + jQ(t)

I(t) = m(t)
Q(t) = 0
s̃(t) = m(t)

Note in this example, the general complex baseband signal s̃(t) = m(t) is a real signal.

The real message m(t) may be written in terms of its positive and negative fre-
quency components2
INSTRUCTOR COPY

2m(t) = m+ (t) + m− (t)

2M (f ) = M+ (f ) + M− (f )

We define M+ (f ) to be the USB of M (f ) and M− (f ) to be the LSB of M (f ), see

fig. 2.14, taken from section 2.7.6. The USB signal M+ (f ) is an analytic signal as
defined in .

The sum of the USB and the LSB is the real message signal which is a special
case of an ISB complex baseband signal where the USB and LSB arise from the same
message m(t).

M+ (f ) M− (f )

4Am 4Am
3Am 3Am
2Am 2Am
Am Am
f f
−B B −B B

Figure 2.14 (repeated): Positive and negative frequency components of a general real
message m(t).
2
These complex baseband message signals centered around f = 0 are not to be confused with the
complex passband (analytic) signals 2S(f ) = S+ (f ) + S− (f ) centered around f = ±fc as presented
in section 2.1.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 274

This is the same result seen previously in section section 2.1 that any real signal
(with or without carrier wave) including a message waveform has positive and negative
frequency components (sidebands around zero frequency).

At this point, it is useful to review section 2.7.6 on the spectrum of real messages.

Note that M (f ) has two sidebands on either side of DC or zero frequency. Thus
M (f ) contains both positive and negative frequencies. Recall that for any real-valued
m(t), m(t) = m∗ (t) and thus from Fourier transform properties M (f ) = M ∗ (−f )
where ∗ denotes the complex conjugate of the spectrum (same amplitude, negative of
the phase). This result is interpreted to mean that the spectrum M (f ) is symmetrical
about f = 0 and the phase is anti-symmetrical about f = 0.

Since M (f ) is symmetrical about f = 0

INSTRUCTOR COPY

M+∗ (−f ) = M− (f )
m∗+ (t) = m− (t)
M−∗ (−f ) = M+ (f )
m∗− (t) = m+ (t)
M (f ) = M ∗ (−f )
2M (f ) = M+ (f ) + M− (f )
2M ∗ (−f ) = M−∗ (−f ) + M+∗ (−f )

Exercise 70
Verify the above equations for a single tone message m(t) = cos 2πfc t.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 275

Solution

2m(t) = 2 cos 2πfc t

= ej2πfc t + e−j2πfc t
= m+ (t) + m− (t)
m+ (t) = ej2πfc t
m− (t) = e−j2πfc t
m∗+ (t) = e−j2πfc t = m− (t)
m∗− (t) = e−j2πfc t = m+ (t)
M+ (f ) = δ(f − fc )
M− (f ) = δ(f + fc )
INSTRUCTOR COPY

M+∗ (−f ) = δ(−f − fc ) = δ(f + fc ) = M− (f )

M−∗ (−f ) = δ(−f + fc ) = δ(f − fc ) = M+ (f )
δ(−f ) = δ(f )

Similarly, for the second message n(t), we can write a complex baseband signal
with zero imaginary component s̃(t) = n(t).

5.2.3 Derivation of analytic message from real message

We now return to the general complex baseband signal for an ISB signal
2s̃(t) = s̃+ (t) + s̃− (t)
2S̃(f ) = S̃+ (f ) + S̃− (f )

Now that we have completed the review of analytic messages m+ (t), we are ready to
answer the question how to derive a time domain expression for

• the USB s̃+ (t) from the first real message m(t)
• the LSB s̃− (t) from the second real message n(t)

Recognizing that both s̃+ (t) and m+ (t) are analytic signals, we can make the
association
s̃+ (t) = m+ (t)
s̃− (t) = n− (t)
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 276

so that we write the ISB complex baseband signal in terms of the messages m(t), n(t)
as follows

2s̃(t) = s̃+ (t) + s̃− (t)

= m+ (t) + n− (t)
2S̃(f ) = S̃+ (f ) + S̃− (f )
= M+ (f ) + N− (f )

f
INSTRUCTOR COPY

(a) Spectrum of real messages M (f ) (green) and N (f ) (orange).

(b) Spectrum of complex messages M+ (f ) (green) and N− (f ) (orange).

In fig. 5.8a, both M (f ) and N (f ) are different examples of the spectrum of a real
signal.

If m(t) = n(t) then s̃(t) = m(t) = n(t) is real and the sidebands M− (f ) = N− (f )
are mirror images of M+ (f ) = N+ (f ). To illustrate, choose M (f ) = N (f ) to be
either the orange or green curve.

If m(t) 6= n(t) then s̃(t) is complex and the two sidebands M+ (f ) (green) and
N− (f ) (orange) are no longer mirror images of each other as shown in fig. 5.8b.

We can now reframe the question. How do we derive a time domain expression
for

• the USB m+ (t) from the real message m(t)

• the LSB n− (t) from the real message n(t)

 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 277

If we consider a single tone message at frequency fm then the question is easy to

answer. We can obtain m+ (t) from m(t) by inspection as follows:

m(t) = cos 2πfm t

= 0.5[ej2πfm t + e−j2πfm t ]
= 0.5[m+ (t) + m− (t)]

then

m+ (t) = ej2πfm t
= cos 2πfm t + j sin 2πfm t
= m(t) + j m̂(t)
m− (t) = e−j2πfm t
INSTRUCTOR COPY

= cos 2πfm t − j sin 2πfm t

= m(t) − j m̂(t)
m+ (t) + m− (t) = 2cos2πfm t
= 2m(t)

Here we have defined m̂(t) to be m(t) shifted by −π/2 radians (−90 degrees), so
that if m(t) = cos 2πfm t then m̂(t) = cos(2πfm t − π/2) = sin 2πfm t. A more general
definition of m̂(t) in terms of m(t) appears in section 5.2.4.

The interesting and essential result for a single tone (real) message is that to
create the analytic message m+ (t) from m(t) we subtract out (eliminate) the negative
frequencies by adding a second (imaginary) term m̂(t).

This same result applies for a general real message (that is the sum of cosine waves
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 278

as per section 2.7.6.3.

X
m+ (t) = Ak ej2πfk t+φk
k
X X
= Ak cos(2πfk t + φk ) + j Ak sin(2πfk t + φk )
k k
= m(t) + j m̂(t)
X
m− (t) = Ak e−j2πfk t−φk
k
X X
= Ak cos(2πfk t + φk ) − j Ak sin(2πfk t + φk )
k k
= m(t) − j m̂(t)
X
INSTRUCTOR COPY

m+ (t) + m− (t) = 2 Ak cos(2πfk t + φk )

k
= 2m(t)

In this case, we obtain m̂(t) by shifting each cosine wave at frequency fk in m(t)
by -90 degrees to make a sine wave. We define

m̂(t) = H{m(t)}

to be the so-called Hilbert transform.

The question

• how to derive a time domain expression for the analytic message m+ (t) in terms
of m(t)

now becomes the question

• how to find a mathematical way to write the Hilbert transform m̂(t) = H{m(t)}
so that it shifts each cosine wave within m(t) by -90 degrees, regardless the
frequency fm of that cosine wave.

Given the answer to this question, we can obtain the analytic message m+ (t) from
the real message m(t) via

m+ (t) = m(t) + j m̂(t)

= m(t) + jH{m(t)}
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 279

Thus we have eliminated the negative frequencies from m(t) by adding j times the
Hilbert transform of m(t).

In the next subsection, we obtain the answer to this question, first in the frequency
domain and then again in the time domain.

5.2.4 Derivation of SSB-SC complex baseband using Hilbert

transforms

We start the derivation of SSB-SC signal mathematics by considering a general real

message m(t) with spectrum M (f ) and find an expression for
INSTRUCTOR COPY

• eliminating the negative frequencies in the frequency domain

• observing that the frequency domain result shifts each cosine wave (sinusoidal
component) by 90 degrees

• transforming the result into the time domain

• observing that the time domain result shifts each cosine wave (sinusoidal com-
ponent) by 90 degrees.

5.2.4.1 Eliminating negative frequencies in the frequency domain

The question is how to eliminate the negative frequencies from m(t) ↔ M (f ).

The answer is to filter out the negative frequencies using a complex filter with
asymmetrical frequency response. Specifically, the negative frequency sideband is
eliminated by multiplying M (f ) by a filter with a Heaviside step function 2u(f ) =
1 + sgn(f ) response, where
(
1 f >0
sgn(f ) = (5.1)
−1 f < 0
(
1 f >0
u(f ) = (5.2)
0 f <0
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 280

This is exactly the same approach taken in eq. (2.42). The new message with the
negative frequencies eliminated can be written as analytic signal

M+ (f ) = 2u(f )M (f )
= [1 + sgn(f )]M (f )
= M (f ) + sgn(f )M (f )
= 2M (f ) for f > 0 and 0 for f < 0

and has an asymmetrical spectrum with only positive frequencies as shown in fig. 5.9.
INSTRUCTOR COPY
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 281

M (f )

4Am
3Am
2Am
Am
f
−B B

2u(f )
INSTRUCTOR COPY

f
−B B

M+ (f )

4Am
3Am
2Am
Am
f
−B B

Figure 5.9: Elimination of negative frequencies via multiplication with the Heaviside
function.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 282

5.2.4.2 Eliminating negative frequencies in the time domain

The analytic message in the time domain is

m+ (t) = F −1 {M+ (f )}
= F −1 {M (f ) + sgn(f )M (f )}
= F −1 {M (f )} + F −1 {sgn(f )M (f )}
= m(t) + F −1 {sgn(f )} ⊗ F −1 {M (f )}

where F −1 is the inverse Fourier transform operator and ⊗ is the convolution op-
erator, and recalling that multiplication in the frequency domain is convolution in
the time domain. Thus to find m+ (t), we need to find the inverse Fourier transform
INSTRUCTOR COPY

F −1 {sgn(f )}.

Exercise 71
Evaluate F −1 {sgn(f )}. This exercise is intended to provide some intuition
about where the answer j/(πt) comes from.

Solution
By definition
Z ∞
−1 1
F {sgn(f )} = sgn(f )e+j2πf t df
2π ∞
Z 0 Z ∞
+j2πf t
= (−1)e df + (+1)e+j2πf t df
f =∞ f =0

+j2πf t 0
∞
−e  +e+j2πf t 
= +
j2πt f =−∞ j2πt f =0
−1 −e−j2π∞t e+j2π∞t 1
= − + −
j2πt j2πt j2πt j2πt
−1 cos(2π∞t) 1
= + −
j2πt jπt j2πt
We know that the cosine function is bounded between ±1. If we assume that
in the limit, cos ∞ = 0 (its average value), then we obtain
−1 +1
F −1 {sgn(f )} = −
j2πt j2πt
j
=
πt
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 283

However, when dealing with infinities, we must tread carefully. A more rigid
mathematical approach is to use the integration property of the Fourier trans-
form, noting that
d
sgn(f ) = 2δ(f )
df
to obtain the same result
j
F −1 {sgn(f )} =
πt
j
↔ sgn(f )
πt
INSTRUCTOR COPY

t f

−2

j
(a) πt
(b) sgn(f )

Figure 5.10: Fourier transform pair. The vertical axis in (a) is the imaginary axis.

Thus we have answered the question posed earlier: to obtain an expression for
m+ (t) in terms of m(t), use this the following result:

m+ (t) = m(t) + F −1 {sgn(f )} ⊗ F −1 {M (f )}

j
= m(t) + ⊗ m(t)
πt
= m(t) + j m̂(t)
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 284

1
where m̂(t) = πt
⊗ m(t). Also, we write an expression for m̂(t) as follows:

j m̂(t) =F −1 {sgn(f )M (f )}
j m̂(t) ↔ sgn(f )M (f )
m̂(t) ↔ − j sgn(f )M (f )
m̂(t) ↔M̂ (f )
M̂ (f ) = − j sgn(f )M (f )
j
m̂(t) =(−j) ⊗ m(t)
πt
1
= ⊗ m(t)
πt
since 1/j = −j.
INSTRUCTOR COPY

The frequency domain relationship

M̂ (f ) = −j sgn(f )M (f )

with the factor −j in front of sgn(f ) shows that at each frequency, the phase of
M (f ) is shifted by -90 degrees for positive frequencies and +90 degrees for negative
frequencies.

Exercise 72
Verify the Hilbert transform phase shifts in the frequency domain for m(t) =
2 cos 2πfm t.

Solution

m(t) = 2 cos 2πfm t

M (f ) = δ(f − fm ) + δ(f + fm )
M̂ (f ) = −j sgn(f )M (f )
= −jδ(f − fm ) + jδ(f + fm )
δ(f − fm ) − δ(f + fm )
=
j
= sin 2πfm t
= cos(2πfm t − π/2)

We observe that the −j(−90 degrees) phase shift at f = +fm and the +j(+90
degrees) phase shift at f = −fm results in the cosine wave becoming a sine
wave.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 285

The time domain convolution result

1
m̂(t) = ⊗ m(t)
πt
must also shift the phase of a single tone message m(t) = cos 2πfm t by −90 degrees.
Some intuition about how the convolution shifts the phase is given in section 5.2.5.

5.2.4.3 Definition of Hilbert transform

We now define
1
m̂(t) = ⊗ m(t)
INSTRUCTOR COPY

πt
= H{m(t)}
M̂ (f ) = −j sgn(f )M (f )
= H{M (f )}

to be the Hilbert transform of m(t). Note that both the input and the output signals
for the Hilbert transform are time domain signals (unlike the Fourier transform).

The analytic message m+ (t) obtained by

m+ (t) = m(t) + jH{m(t)}

(5.3)
= m(t) + j m̂(t)

and has only positive frequencies as expected

m+ (t) = m(t) + j m̂(t)

M+ (f ) = M (f ) + j M̂ (f )
= M (f ) + j(−j) sgn(f )M (f )
= M (f ) + sgn(f )M (f )
= [1 + sgn(f )]M (f )
= 2u(f )M (f )
= 2M (f ) for f > 0 and 0 for f < 0
= M+ (f )

Thus the output m+ (t) of the Hilbert filter contains only positive frequencies, and
thus is an analytic signal as introduced and defined in section 2.6.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 286

5.2.4.4 Eliminating either negative or positive frequencies

The Hilbert transform may also be used to eliminate positive frequencies and retain
only negative frequencies.

We can obtain either m+ (t) or m− (t) from m(t) by either adding or subtracting
j m̂(t) = jH{m(t)} as shown below.

m+ (t) = a(t)ejφ(t)
= m(t) + jH{m(t)}
= m(t) + j m̂(t)
INSTRUCTOR COPY

m− (t) = a(t)e−jφ(t)
= m(t) − jH{m(t)}
= m(t) − j m̂(t)

Exercise 73
Prove that these results are consistent with thoe time domain results given in
section 5.2.2

m+ (t) = m∗− (t) = m(t) + j m̂(t)

m− (t) = m∗+ (t) = m(t) − j m̂(t)

Solution

m∗− (t) = [m(t) − j m̂(t)]∗ = m(t) + j m̂(t) = m+ (t)

m∗+ (t) = [m(t) + j m̂(t)]∗ = m(t) − j m̂(t) = m− (t)

5.2.5 Hilbert transform definition and properties

From the previous section

m̂(t) =H{m(t)}
=m(t) ⊗ (1/πt)
Z
1 ∞ m(τ )
= dτ
π −∞ t − τ
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 287

m̂(t) is called the Hilbert Transform of m(t) 3 .

The Hilbert transform is a special kind of non-causal filter with impulse response
1/πt. As observed in section 5.2.3, the Hilbert transform shifts each sinusoidal com-
ponent of m(t) by −90◦ .

For m(t) = cos 2πfm t,

m̂(t) = cos(2πfm t − π/2)
= cos(2πfm t cos π/2) + sin 2πfm t sin π/2
= sin 2πfm t
for any and all fm . To see that the cosine wave is shifted to the right (i.e. delayed)
to become a sine wave, observe that when t = 0, m̂(t = 0) = cos(−π/2) = cos 3π/2 =
INSTRUCTOR COPY

sin 0.

m(t)

π π 3π 2π
2 2
−1
1
m̂(t) = πt ⊗ m(t)

π π 3π 2π
2 2
−1
Figure 5.11: The Hilbert transform results in a −π
2
phase shift to the input. Shift the
o
cosine wave to the right by 90 to obtain the sine wave.

A cosine wave at any frequency fm is shifted by 90◦ or delayed by a quarter of

a period (1/4)(1/fm ). The 90◦ phase shift is equivalent to a delay of a a quarter of
a period. Thus for lower frequencies with longer periods, the delay is longer and for
higher frequencies with shorter periods, the delay is shorter. Thus different frequency
components in m(t) (see section 2.7.6.3) are delayed by different amounts.
3
m(t) ⊗ (1/πt) = (1/πt) ⊗ m(t) because convolution is commutative
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 288

In the next 3 subsections, we prove this

• intuitively
• in the frequency domain
• in the time domain

5.2.5.1 Intuitive proof

Figure 5.12 shows an approximation of the Hilbert system (filter) impulse response.
In theory h(t) flips from minus infinity to infinity at t = 0.
INSTRUCTOR COPY

Figure 5.12: Approximate impulse response of Hilbert filter h(t) = 1/(πt).

To see intuitively why convolving a cos wave with the Hilbert filter shifts the phase
by 90 degrees regardless of the frequency of the cos wave, note that

• recall that frequency is defined in radians per second, so that 2π radians in one
second is one cycle per second or one Hz.
• a 2π phase shift of a 1 Hz cos wave corresponds to a delay of 1 second.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 289

• a π/2 phase shift of a 1 Hz cos wave corresponds to a delay of 0.25 seconds.

• the same a π/2 phase shift of a 10 Hz cos wave corresponds to a delay of 0.025
seconds.
• Thus at low frequencies approaching zero Hz the delay grows to infinity and at
high frequencies the delay approaches zero.
• The Hilbert filter looks like a negative differentiator and the negative derivative
of a cos wave is a sin wave.
• further insight can be gained from the time domain mathematical proof shown
below.
INSTRUCTOR COPY

5.2.5.2 Proof in time domain

The proof is to show that

Z ∞
1 cos 2πfm τ
H{cos2πfm t} = cos 2πfm t ⊗ (1/πt) = dτ
π −∞ t−τ
= sin 2πfm t

Let p − t − τ so that τ = t − p and dτ = −dp then

Z
1 −∞ cos 2πfm (t − p)
cos 2πfm t ⊗ (1/πt) = (−dp)
π ∞ p
Using
cos 2πfm (t − p) = cos(2πfm t − 2πfm p)
= cos 2πfm t cos 2πfm p + sin 2πfm t sin 2πfm p
we can write
Z ∞
1 cos 2πfm p
H{cos2πfm t} = cos 2πfm t dp
π −∞ p
Z ∞
1 sin 2πfm p
+ sin 2πfm t dp
π −∞ p
Note that cos 2πf
p
mp
is an odd function of p and thus it’s integral from −∞ to ∞ is
zero. It can be shown that
Z ∞
sin 2πfm p
dp = π
−∞ p
Thus
H{cos 2πfm t} = sin 2πfm t
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 290

5.2.5.3 Proof in frequency domain

The proof is easier in the frequency domain, as was already done in section 5.2.4.3,
and is interpreted in more detail below.

m(t) ↔ M (f )
m̂(t) ↔ M̂ (f )
m̂(t) ↔ −j sgn(f )M (f )
m+ (t) = m(t) + j m̂(t)
M+ (f ) = M (f ) + j M̂ (f )
INSTRUCTOR COPY

= M (f ) + jH(f )M (f )
= M (f ) + j(−j) sgn(f )M (f )
= M (f ) + sgn(f )M (f )
= 2M (f ) for f > 0 and 0 for f < 0
= M+ (f )

The frequency domain proof is interpreted as follows:

• The Hilbert transformer is an LTI system (non-causal filter) with transfer func-
1
tion H(f ) = −j sgn f and impulse response h(t) = πt . The transfer function
shows that every positive frequency is multiplied by −j (and every negative
frequency is multiplied by +j), thus shifting the phase by ±90 degrees.
1
• m̂(t) = H{m(t)} = h(t) ⊗ m(t) = πt
⊗ m(t)

• M̂ (f ) = H{M (f )} = H(f )M (f ) = −j sgn f M (f )

• M (f ) is a sum of complex exponentials with symmetric magnitudes and anti-
symmetric phase about f = 0, as shown in section 5.2.2.
• M̂ (f ) shifts each positive frequency complex exponential in M (f ) by −π/2. The
constant j = ejπ/2 shifts each complex exponential by +π/2, so j M̂ (f ) shifts
the phase by −π/2 + π/2 = 0, and thus j M̂ (f ) = M (f ) for positive frequencies
and M+ (f ) = M (f ) + j M̂ (f ) = 2M (f ) for f > 0.
• M̂ (f ) shifts each negative frequency complex exponential in M (f ) by +π/2.
The constant j = ejπ/2 shifts each complex exponential by +π/2, and thus
j M̂ (f ) shifts each complex exponential by +π/2 + π/2 = π (phase reversal), so
j M̂ (f ) = −M (f ) for negative frequencies and M+ (f ) = M (f ) + j M̂ (f ) = 0 for
f < 0.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 291

Exercise 74
Show how the Hilbert transform phase shifts and cancels negative frequen-
cies for a single tone message m(t) = cos 2πfm t and the corresponding
m+ (t) = m(t) + jH{m(t)}. Do the calculations both in the time domain and
in the frequency domain. Hint: write m(t) in terms of positive and negative
frequencies.

Solution

m(t) = cos 2πfm t

1
= [ej2πfm t + e−j2πfm t ]
2
INSTRUCTOR COPY

m̂(t) = sin 2πfm t

1
= [ej2πfm t − e−j2πfm t ]
2j
= sin 2πfm t
m+ (t) = m(t) + j m̂(t)
1 1
= [ej2πfm t + e−j2πfm t ] + j [ej2πfm t − e−j2πfm t ]
2 2j
j2πfm t
=e
= cos 2πfm t + j sin 2πfm t

ej2πfm t ↔ δ(f − fm )
e−j2πfm t ↔ δ(f + fm )
1
M (f ) = [δ(f − fm ) + δ(f + fm )]
2
M̂ (f ) = −j sgn(f )M (f )
1
= [−j(+1)δ(f − fm ) − j(−1)δ(f − (−fm ))]
2
1
= [δ(f − fm ) − δ(f + fm )]
2j
M+ (f ) = M (f ) + j M̂ (f )
1
= [δ(f − fm ) + δ(f + fm )]
2
1
+ j [δ(f − fm ) − δ(f + fm )]
2j
= δ(f − fm )
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 292

Exercise 75
Write the message m(t) = cos 2πfm t in terms of its positive and negative
frequency components in the time domain.

Solution
In the time domain,

m+ (t) =m(t) + j m̂(t)

= cos 2πfm t + j sin 2πfm t
=ej2πfm t
m− (t) =m(t) − j m̂(t)
INSTRUCTOR COPY

= cos 2πfm t − j sin 2πfm t

=e−j2πfm t
2m(t) = m+ (t) + m− (t)
= ej2πfm t + e−j2πfm t
= 2 cos 2πfm t

Exercise 76
Write a general message m(t) ↔ M (f ) in terms of its positive and negative
frequency components in the frequency domain. Hint: use the time domain
results in the previous exercise as a starting point.

Solution
In the frequency domain,

m(t) ↔ M (f )
m̂(t) ↔ M̂ (f )
m+ (t) = m(t) + j m̂(t)
M+ (f ) = M (f ) + j M̂ (f )
= M (f ) + j(−j) sgn(f )M (f )
= M (f ) + sgn(f )M (f )
= 2M (f ) for f > 0 and 0 for f < 0
= M+ (f )
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 293

m− (t) = m(t) − j m̂(t)

M− (f ) = M (f ) − j M̂ (f )
= M (f ) − j(−j) sgn(f )M (f )
= M (f ) − sgn(f )M (f )
= 2M (f ) for f < 0 and 0 for f > 0
= M− (f )

M+ (f ) + M− (f ) = M (f ) + sgn(f )M (f )
+ M (f ) − sgn(f )M (f )
INSTRUCTOR COPY

= 2M (f )

Exercise 77
Write the frequency domain results of the previous exercise for m(t) =
cos 2πfm t.

5.2.5.4 Summary

To recap: an analytic signal (or pre-envelope) is defined as m+ (t) = m(t) + j m̂(t),

where m̂(t) = H{m(t)} = m(t) ⊗ (1/πt) is the Hilbert transform of m(t). The pre-
envelope contains only positive frequencies. We will use this pre-envelope to create
an SSB-SC signal with only the upper sideband, also called simply an upper sideband
(USB) signal. Note that the pre-envelope m+ (t) is a complex baseband signal that
we will use to create an analytic SSB passband signal s+ (t) or an SSB real passband
signal s(t).

The reasoning above can be repeated to define a pre-envelope with only negative
frequencies m− (t) = m(t) − j m̂(t) which is used to create a lower sideband (LSB)
signal.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 294

Exercise 78
Show that the Hilbert transform of the Hilbert transform of any real function
is the negative of that function, i.e. show that

H{m(t)} = m̂(t)
H{H{m(t)}} = H{m̂(t)}
= −m(t)

This result will be useful when we design SSB-SC receivers.

Solution
In the time domain, the intuitive solution is that since H shifts the phase of
INSTRUCTOR COPY

every frequency component in m(t) by π/2, the double transform H{H} shifts
the phase by π which is the same as taking the negative.
In the frequency domain, we use the same H operator notation that represents
the Hilbert transform

H{M (f )} = M̂ (f )
= −j sgn(f )M (f )
H{H{M (f )}} = H{−j sgn(f )M (f )}
= (−j sgn(f )[−j sgn(f )M (f )]
−M (f )

since sgn(f ) sgn(f ) = 1 for both f > 0 and f < 0 and (−j)(−j) = −1

Exercise 79
Show that the Fourier transform of the Hilbert transform of any real function
m(t) is

F{H{m(t)}} = F{m̂(t)}
= M̂ (f )
= (−j) sgn(f )M (f )

Solution
See section 5.2.4
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 295

5.2.6 Hilbert transform implementation

5.2.6.1 Hilbert digital filter impulse response

The Hilbert transform is a non-causal filter whose impulse response is a sampled

version of the Hilbert transform impulse response shown in fig. 5.12. The Hilbert
filter has impulse response and frequency response

h(t) = (1/(πt) ↔ −j sgn(f ) = H(f )

From the definition of the Hilbert transform

INSTRUCTOR COPY

m̂(t) =H{m(t)}
=m(t) ⊗ (1/πt)
= m(t) ⊗ h(t)
M̂ (f ) = H{M (f )}
= −j sgn(f )M (f )
= H(f )M (f )

We compute the time domain Hilbert filter coefficients h(t) by taking the inverse
Fourier transform of the Hilbert frequency response H(f ).

h(t) = F −1 {H(f )}
= F −1 {−j sgn(f )}
Z ∞
1
= (−j) sgn(f )ej2πf t df
2π ∞
Z 0
= (−j)(−1)ej2πf t df
Z f∞
=−∞

+ (−j)(+1)ej2πf t df
f =0

The ±∞ integral limits for a discrete-time system sampled at fs are replaced with
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 296

±fs /2
Z 0
h(t) = j ej2πf t df
f =−fs /2
Z fs /2
= −j jej2πf t df
f =0
0 f /2
je+j2πf t  −je+j2πf t  s
= +
j2πt f =−fs /2 j2πt f =0
+1 −1 +jπfs t
= [1 − e−jπfs t ] + [e − 1]
2πt 2πt
1
= [1 − cos πfs t]
πt
INSTRUCTOR COPY

Note that by replacing the integral limits ±∞ with ±fs /2 we obtain the same result
1/(πt) with an additional cosine term. When we sample at times t = nts

h(n) = h(t = nts )

fs
= [1 − cos πn]
πn

The discrete-time Hilbert filter impulse response is plotted below

Figure 5.13: Sampled impulse response of Hilbert filter h(t) = 1/(πt).

In this example, for h(t) to be causal, the filter will add a delay of half the filter
length or 15 samples.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 297

5.2.6.2 Hilbert filter implementation in Matlab

From https://www.mathworks.com/help/signal/ref/hilbert.html

x = hilbert(xr) returns the analytic signal, x, from a real data sequence, xr.

Two examples are at https://www.mathworks.com/help/signal/ug/hilbert-transform.

html and https://www.mathworks.com/help/signal/ug/analytic-signal-for-cosine.
html.

5.2.6.3 Hilbert filter implementation in GNURadio

INSTRUCTOR COPY

From https://wiki.gnuradio.org/index.php/Hilbert

The Hilbert transformer block input is a real signal and the output is the corre-
sponding (complex) analytic signal. The real part of the complex output is the input
delayed by half the filter length. The imaginary part of the complex output is the
Hilbert filtered (−90 degree phase shift) version of the real input.

5.3 SSB-SC real passband signal

5.3.1 USB-SC derived from pre-envelope in frequency do-

main

5.3.1.1 Derivation 1 based on fundamental result

In this derivation, we start from the fundamental relationship between any real pass-
band signal s(t) and its complex baseband s̃(t) first introduced in section 2.1

2S(f ) =S+ (f ) + S− (f )
=S̃(f − fc ) + S̃ ∗ (−(f + fc ))

This fundamental relationship applies to any complex baseband signal s̃(t) ↔ S̃(f )
that may be ISB, USB, LSB, DSB, AM or other modulation scheme such as FSK,
ASK, PSK mentioned in chapter 2.

For this derivation, we choose s̃(t) = m+ (t) for USB.

 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 298

Exercise 80
Substitute the complex baseband signal spectrum that arises for a USB signal
in the above equation. Hint: start by defining s̃(t) = m+ (t) = m(t) + j m̂(t)
and find S̃(f ) and continue to evaluate both terms in the expression for 2S(f ).

Solution
An intermediate step to the final result is to recall that if we define

S̃(f ) = M+ (f )

then

S̃ ∗ (−f ) = M+∗ (−f )

INSTRUCTOR COPY

= M− (f )

as shown in section 5.2.2.

The final result is

2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))
= M+ (f − fc ) + M+∗ (−(f + fc ))
= M+ (f − fc ) + M− (f + fc )

which is illustrated in fig. 5.14.

2S(f )
4Am
3Am
2Am
Am f
−fc fc

Figure 5.14: 2S(f ) where the negative frequency component is M− (f + fc ) and

the positive frequency component is M+ (f − fc ).
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 299

5.3.1.2 Derivation 2

The complex passband USB-SC signal may be created in the frequency domain, start-
ing with the pre-envelope m+ (t).

A complex passband USB-SC signal s+ (t) is created by multiplying the pre-

envelope m+ (t) by a complex carrier wave

s+ (t) = m+ (t)ej2πfc t
= [m(t) + j m̂(t)]ej2πfc t
INSTRUCTOR COPY

Recall from the Fourier transform properties that multiplying any signal by a
complex exponential at fc shifts the spectrum of that signal by fc .

Thus we shift the pre-envelope up to the carrier frequency

s+ (t) = [m(t) + j m̂(t)]ej2πfc t
S+ (f ) = M (f − fc ) + sgn(f − fc )M (f − fc )
s+ (t) = m+ (t)ej2πfc t
S+ (f ) = M+ (f − fc )

A real passband USB-SC signal is created in one of two ways:

• Taking the real part of the complex passband USB-SC signal

s(t) = Re{s+ (t)}
= Re{m+ (t)ej2πfc t }
= Re{[m(t) + j m̂(t)]ej2πfc t }
= m(t) cos 2πfc t − m̂(t) sin 2πfc t
or
• adding positive and negative frequency complex passband signals to make a real
signal.
– take the analytic signal or pre-envelope as the message signal, shift it up
to the positive carrier frequency fc (as above), repeated here
s+ (t) = m+ (t)ej2πfc t
S+ (f ) = M+ (f − fc )
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 300

– take the complex conjugate of the analytic signal, shift it down to the
negative carrier frequency −fc
s− (t) = [m(t) − j m̂(t)]e−j2πfc t (5.4)
S− (f ) = M (f + fc ) − sgn(f + fc )M (f + fc ) (5.5)
s− (t) = m− (t)e−j2πfc t (5.6)
S− (f ) = M− (f + fc ) (5.7)

– add the positive and negative frequency parts together

2s(t) = s+ (t) + s− (t)
= m+ (t)ej2πfc t + m− (t)e−j2πfc t
2S(f ) = S+ (f ) + S− (f )
INSTRUCTOR COPY

= M+ (f − fc ) + M− (f + fc )

Exercise 81
Show that adding positive and negative frequency components together

s(t) = [s+ (t) + s− (t)]/2

= m(t) cos 2πfc t − m̂(t) sin 2πfc t
= Re{s+ (t)}

yields the same result as was obtained by taking the real part of the analytic
signal s+ (t).

Exercise 82
Repeat section 5.3.1.2 for LSB-SC to obtain an expression for s(t). Note that
the complex passband LSB-SC signal s+ (t) is created by multiplying the (neg-
ative frequency) pre-envelope by a complex carrier wave

s+ (t) = m− (t)ej2πfc t
= [m(t) − j m̂(t)]ej2πfc t

5.3.1.3 Interpretation

The resulting equation obtained in both derivations above may be interpreted to show
that an USB signal is the sum of the complex baseband spectrum M+ (f ) shifted by
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 301

+fc and the flipped (mirror-image) complex baseband spectrum M− (f ) shifted by

−fc .

The term M− (f ) is a mirror image of M+ (f ) flipped around f = 0. This is

illustrated in fig. 2.14.

M+ (f ) M− (f )

4Am 4Am
3Am 3Am
2Am 2Am
Am Am
f f
INSTRUCTOR COPY

−B B −B B

Figure 2.14 (repeated): Positive and negative frequency components of a general real
message m(t).

The term M+ (f − fc ) is shifted by +fc because of (f − fc ). In other words, the

value of M+ (f − fc ) at f = fc is the same as the value of M+ (f ) at f = 0. Similarly,
the term M− (f + fc ) is shifted by −fc because of (f + fc ). The result was shown in
fig. 5.14

The term M− (f + fc ) is a mirror image of M+ (f − fc ) around f = 0. This may

be illustrated by

• sketching M+ (f ) as an asymmetrical spectrum just above f = 0 and M− (f ) as

an asymmetrical spectrum just below f = 0, as in fig. 2.14.
• sketching M+ (f − fc ) as an asymmetrical spectrum just above f = fc and
M− (f + fc ) as an asymmetrical spectrum just below f = −fc , as in fig. 5.14.

Exercise 83
Repeat the previous exercise for LSB. Substitute the complex baseband signal
spectrum that arises for a LSB signal in the exercise above.
Hint: define s̃(t) = n− (t) = n(t) − j n̂(t).
Hint: Recall that s̃(t) is a general complex baseband signal and can be defined
to be any specific complex baseband signal of your choice, for example m+ (t)
or n− (t) or m(t) or n(t) or some other message. It is not a fixed rule that
s̃(t) = m+ (t).
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 302

Exercise 84
Compare the expressions for 2S(f ) for DSB-SC and SSB-SC and sketch.

Exercise 85
Combine the results for USB and LSB to obtain an expression for an ISB signal
that can transmit two independent messages m(t) and n(t).

5.3.2 USB-SC derived from pre-envelope in time domain

Now consider USB-SC in the time domain.

INSTRUCTOR COPY

For DSB-SC, s(t) = m(t)c(t). For SSB-SC with upper sideband only (USB-SC)
s(t) = Re{[m(t) + j m̂(t)]ej2πfc t }
= Re{[m(t) + j m̂(t)][cos 2πfc t + j sin 2πfc t]}
= m(t) cos 2πfc t − m̂(t) sin 2πfc t

In general, an SSB-SC signal with a general message can be written

s(t) = m(t) cos 2πfc t ∓ m̂(t) sin 2πfc t
where the minus sign is for upper sideband and the plus sign for lower sideband.

If m(t) = cos(2πf t), then m̂(t) = sin(2πf t) where we denote the Hilbert transform
by the hat ∧ symbol, and
s(t) = cos 2πfm t cos 2πfc t ∓ sin 2πfm t sin 2πfc t
(5.8)
= cos 2π(fc ± fm )t
which is the single sideband at fc ± fm (plus for USB, minus for LSB).

Thus for USB, I(t) = m(t) and Q(t) = m̂(t). For a single tone message m(t) =
Am cos2πfm t, I(t) = cos(2πfm t) and Q(t) = sin(2πfm t). The constant Am accounts
for the signal power. We use Am to represent the message power. We could also use
Ac instead of Am since for SSB, the message power is the same as the signal power.

A single sideband USB signal s(t) = m(t) cos 2πfc t − m̂(t) sin 2πfc t can be written
as
s(t) = Re{s̃(t)ej2πfc t }
= Re{a(t)ejφ ej2πfc t }
= Re{[m(t) + j m̂(t)]ej2πfc t }
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 303

Thus the complex envelope of an USB-SC signal a(t)ejφ = m(t) + j m̂(t) is an

analytic signal created from the (real) message m(t).

Exercise 86
Write an SSB signal in the form 2s(t) = s+ (t) + s− (t) without taking the real
part of anything.

5.3.3 SSB-SC in time and frequency domain

In this section we show how to obtain the frequency domain expression for USB
directly from the time domain expression for USB.
INSTRUCTOR COPY

m(t) ⇔ M (f )
s(t) ⇔ S(f )

Proof
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 304

Ac
s(t) = [m(t) cos 2πfc t − m̂(t) sin 2πfc t]
2
M̂ (f ) =H(f )M (f )
= − j sgn(f )M (f )
 
Ac 1
S(f ) = M (f ) ⊗ {δ(f − fc ) + δ(f + fc )}
2 2
 
Ac 1
− (−j sgn(f ))M (f ) ⊗ {δ(f − fc ) − δ(f + fc )}
2 2j
Ac
S(f ) = [M (f ) ⊗ δ(f − fc ) + M (f ) ⊗ δ(f + fc )
4
+ sgn(f )M (f ) ⊗ δ(f − fc )
INSTRUCTOR COPY

− sgn(f )M (f ) ⊗ δ(f + fc )]
Ac
S(f ) = [M (f − fc ) + M (f + fc )
4
+ sgn(f − fc )M (f − fc ) − sgn(f + fc )M (f + fc )]
Ac
S(f ) = M (f − fc ), f − fc > 0, f > fc
2
Ac
= M (f + fc ), f + fc > 0, f < −fc
2
Ac Ac
= M+ (f − fc ) + M− (f + fc )
4 4
If Ac = 2 then

S(f ) = M+ (f − fc )/2 + M− (f + fc )/2

= S+ (f )/2 + S− (f )/2

Exercise 87
Repeat the analysis for LSB to obtain

S(f ) = M− (f − fc )/2 + M+ (f + fc )/2

= S+ (f )/2 + S− (f )/2
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 305

5.3.4 Single tone modulation for SSB

5.3.4.1 Time domain

In this section, we write the expression for an SSB signal s(t) with a single tone
message. The upper signs are for USB, the lower signs are for LSB.

m(t) = Am cos 2πfm t

m̂(t) = Am sin 2πfm t
Ac
s(t) = [m(t) cos 2πfc t ∓ m̂(t) sin 2πfc t]
2
INSTRUCTOR COPY

Ac Am
= [cos 2πfm t cos 2πfc t ∓ sin 2πfm t sin 2πfc t]
2
Ac Am
= cos[2π(fc ± fm )t]
2
using cos(α ± β) = cos α cos β ∓ sin α sin β. Thus s(t) is a single tone at frequency
fc ± fm .

Observe how the opposite sideband is exactly cancelled out by adding the Hilbert
transform term containing m̂(t)

Exercise 88
Find the complex baseband signal s̃(t) for USB and also for LSB.

Exercise 89
Find the positive frequency USB signal s+ (t)

Exercise 90
Write the USB signal with a single tone message in the form

s(t) = [s+ (t) + s− (t)]/2.

Repeat for a LSB signal.

 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 306

5.3.4.2 Frequency domain

Exercise 91
Repeat the time domain analysis in the frequency domain. Point out where
the the cancellation of the opposite sideband occurs in the math.

5.4 SSB-SC transmitters

In this section, we show three methods of building an SSB-SC transmitter.

INSTRUCTOR COPY

5.4.1 SSB-SC transmitter 1 using Hilbert transform (phasing

method)

This transmitter is suitable for implementation in software using a digital Hilbert

transform filter. The phasing method of generating SSB may also be called a Hartley
modulator.

5.4.1.1 Software implementation

To create an SSB-SC signal at fc we follow the SSB-SC equations above in the previous
sections.

1. Take the Hilbert transform m̂(t) = m(t) ⊗ (1/πt) of the message.

2. Create the analytic signal m+ (t) = m(t) + j m̂(t) for USB or m− (t) = m(t) −
j m̂(t) for LSB
3. Upconvert it to the desired carrier frequency by multiplying by ej2πfc t to obtain
s+ (t)
4. Take the real part to obtain s(t) = Re{s+ (t)}.

The net result of these steps is

s(t) = <{[m(t) + j m̂(t)]ej2πfc t }

= m(t) cos 2πfc t − m̂(t) sin 2πfc t
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 307

which is the SSB-SC real passband equation.

In GNU Radio, the upconversion to a radio frequency (RF) wave at fc (steps 3

and 4) is the function of the USRP Sink block. Thus an SSB signal with message
m(t) is generated in GNU Radio by a USRP sink block (standard IQ transmitter)
with inputs i(t) = m(t) and q(t) = ±m̂(t) with + sign for LSB and - sign for USB.

5.4.1.2 Hardware (analog) implementation

The phasing method SSB-SC transmitter may also be implemented in analog hard-
ware using the SSB-SC real passband equation directly.
INSTRUCTOR COPY

A block diagram of the analog implementation contains the analog IQ mixer fol-
lowed by a Hilbert transformer (delay) in the sine branch.

This analog method requires devising a phase shifting circuit that performs the
function of the Hilbert transformer. Over a narrow bandwidth around fc , the phase
shifting circuit is a simple delay of a quarter cycle of fc which is equivalent to a 90
degree phase shift for frequencies close to fc .

5.4.2 SSB-SC transmitter 2, Weaver modulator, avoids Hilbert

transform

This transmitter is suitable for implementation in software.

The Hibert transform can be inconvenient to compute. An SSB-SC signal can be

created without using the Hilbert transform with the so-called Weaver modulator.
The block diagram of this modulator in real form is shown in fig. 5.15. The complex
form is shown in fig. 5.16.

In this section, a USB signal is generated without using a Hilbert transform. The
signs of the exponentials are chosen for USB.

LO1 operating at f1 is chosen to be in the centre of the band of the message m(t),
e.g. f1 = 1500 Hz for a voice bandwidth message 0–3000 Hz (in practice 300–2700 Hz).
We define BW = 3000 Hz so that f1 = BW/2.

Notice that the low pass filter is indicated to pass frequencies from 0 → BW 2
in
both the real (fig. 5.15) and complex (fig. 5.16) block diagrams. Then notice that the
spectra drawn in fig. 5.17 shows the filter passband extending from − BW2
→ BW2
. A
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 308

DC to BW
2

cos(2πf1 t) cos(2πf2 t)
IN OUT
m(t) with s(t) on
sin(2πf1 t) sin(2πf2 t)
bandwidth BW carrier fc

DC to BW
2 Legend
f1 f2
BW BW
LSB 2 fc − 2

U SB − BW
2 fc + BW
2
INSTRUCTOR COPY

Figure 5.15: Real block diagram for Weavers modulator.

IN Re OUT
m(t) with s(t) on
bandwidth BW carrier fc
DC to BW
2

Legend
f1 f2
j2πf1 t j2πf2 t BW BW
e e LSB 2 fc − 2

U SB − BW
2 fc + BW
2

Real
Complex

Figure 5.16: Complex block diagram for Weavers modulator.

low pass filter works over the same negative frequencies as the positive frequencies.
So the filter passing 0 → BW2
will also cover − BW2
→ 0. When purchasing a filter in
the real world it will be specified in positive frequencies.

For USB, LO2 operates at fc + f1 .

In fig. 5.17, we observe how the multiplying the message by the complex exponen-
tials shift the spectrum of the message.

The operations to create the USB signal are as follows:

1. The real message spectrum M (f ) contains positive and negative frequencies.

 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 309

BW

f
BW
2

f
INSTRUCTOR COPY

BW
fc + 2

f
fc

f
−fc fc
0 Hz

Figure 5.17: Stages of a USB Weaver’s modulator

BW
2. The real message m(t) is first shifted down by f1 = 2
thus making it a
complex signal s̃1 (t) with an asymmetrical spectrum.
3. s̃1 (t) is low pass filtered to remove the lower sideband resulting in s̃2 (t) which
is now a complex signal
4. s̃2 (t) is shifted up by f2 = fc + BW
2
resulting in another complex signal related
to s(t) with positive frequencies only.
5. The real part of this signal is selected, or the real and and imaginary parts of
this signal are added together, resulting in a real signal s(t) with positive and
negative frequency components.

The net result of the two frequency conversions (before taking the real part) is that
the zero frequency component of m(t) ↔ M (f ) has been shifted up to the suppressed
carrier frequency fc and the components of m(t) ↔ M (f ) above zero frequency (i.e.
the message content) are shifted to be above fc , i.e. M (f ) → M (f − fc ) with the
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 310

lower sideband (red) filtered out, resulting in S+ (f ) = M+ (f − fc ). Taking the real

part results in S(f ) = 0.5[M+ (f − fc ) + M− (f + fc )].

The operations to create the USB signal are written mathematically as follows:

1. At the input there is the (real) message m(t) with spectrum M (f ) symmetrical
about zero.
2. After the complex multiply (mixer)4 with e−j2πf1 t we have
s̃1 (t) =m(t)e−j2πf1 t
=m(t) cos 2πf1 t − jm(t) sin 2πf1 t

3. After the LPF filter the result, s̃2 (t) is the analytic signal shifted down by f1 .
INSTRUCTOR COPY

This is clear from the frequency diagram. The algebra is an exercise (see below).
s̃2 (t) =0.5[m(t) + j m̂(t)]e−j2πf1 t
=0.5m(t) cos 2πf1 t − j0.5m(t) sin 2πf1 t
+ j0.5m̂(t) cos 2πf1 t + 0.5m̂(t) sin 2πf1 t

4. After the complex multiply (mixer) with f2 we have

s̃3 (t) =s̃2 (t)ej2πf2 t
BW
=s̃2 (t)ej2π(fc + 2
)t

=s̃2 (t)ej2π(fc +f1 )t

=0.5[m(t) + j m̂(t)]e−j2πf1 t ej2π(fc +f1 )t
=0.5[m(t) + j m̂(t)]ej2πfc t
=0.5[m(t) cos 2πfc t − m̂(t) sin 2πfc t]
+ j0.5[m̂(t) cos 2πfc t + m(t) sin 2πfc t]
=s+ (t)/2

5. To obtain the real USB-SC signal s(t) we take the real part of s+ (t), or equiv-
alently we calculate 0.5s− (t) and add it to 0.5s+ (t) to obtain
s(t) = 0.5s+ (t) + 0.5s− (t)

Exercise 92
Mathematically show the low pass filtering s̃1 (t) in step 3 to get s̃2 (t). Hint:
write m(t) = [m+ (t) + m− (t)]/2

4
Mixers in communications are multipliers, whereas mixers in audio are adders
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 311

Exercise 93
Repeat the figure and mathematics for LSB

It is also possible to make a real USB signal by adding the real and imaginary
branch outputs together, adding i + q not i + jq. If i and q are added
s(t) =m(t) cos 2πfc t − m̂(t)sin2πfc t + m̂(t) cos 2πfc t + m(t)sin2πfc t
=m(t)[cos 2πfc t + sin 2πfc t] + m̂(t)[cos 2πfc t − sin 2πfc t]
√ √
=m(t) cos(2πfc t − π/4)/ 2 + m̂(t) cos(2πfc t + π/4)/ 2
√ √
=m(t) cos(2πfc t − π/4)/ 2 + m̂(t) cos(2πfc t − π/4 + π/2)/ 2
√ √
=m(t) cos(2πfc t − π/4)/ 2 − m̂(t) sin(2πfc t − π/4)/ 2
INSTRUCTOR COPY

thus
s(t) =Re{s̃3 (t)e−jπ/4 }
√
=Re{[m(t) + j m̂(t)]ej(2πfc t−π/4) / 2}

using
cos(α + π/4) = cos α cos(π/4) − sin α sin(π/4)
p
= (2)[cos α − sin α]
cos(α − π/4) = cos α cos(π/4) + sin α sin(π/4)
p
= (2)[cos α + sin α]
cos(α + π/2) = − sin α

This completes the discussion of the operations in an SSB-SC transmitter that

creates a real SSB-SC signal s(t) = m(t) cos 2πfc t ∓ m̂(t)sin2πfc t with message m(t),
where the upper sign is for USB and the lower sign is for LSB.

Recall the observations at the beginning of this section that SSB can be generalized
to ISB and that the the message that creates each sideband may contain multiple
signals.

Exercise 94
Write a time domain expression for ISB using two messages m(t), n(t).

Exercise 95
Write an expression for a USB signal whose message contains a group of three
signals, each signal using its own subcarrier.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 312

5.4.3 SSB-SC transmitter 3 using filters

This SSB-SC transmitter (fig. 5.18a) is suitable for implmentation in both software
and hardware, but is most commonly implemented in hardware.

m(t) s(t)

fc to (fc + B)
INSTRUCTOR COPY

cos (2πfc t)
(a) SSB-SC transmitter using only real signals

f
−fc fc

(b) SSB-SC spectra showing bandpass filter

Figure 5.18: SSB-SC transmitter method using filters

The first stage of this SSB-SC transmitter generates a real passband DSB-SC
signal at carrier frequency fc . The second stage is a bandpass filter filter that filters
out either the upper sideband or lower sideband as shown in fig. 5.18b.

5.5 SSB-SC receivers

In this section, we analyse several different SSB-SC receiver designs.

 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 313

5.5.1 SSB-SC receiver using message-bandwidth low pass fil-

ters

We consider two methods of receiving a real passband SSB-SC signal using low pass
filters with cutoff frequency equal to the bandwidth B of the message signal. For a
single tone message, we choose fm = B.

5.5.1.1 Method 1

In Method 1, we multiply the real SSB signal by cos 2πfc t and low pass filter to get
m(t). This method is used in analog receivers. To demodulate SSB-SC, the receiver
INSTRUCTOR COPY

is required to know the exact carrier frequency for proper demodulation.

Exercise 96
show the mathematical steps to obtain m(t) from s(t) using this method.

5.5.1.2 Method 2

In Method 2, we use an IQ receiver. For an SSB-SC signal s(t) = m(t) cos 2πfc t ∓
m̂(t) sin 2πfc t with − for USB and + for LSB. Considering the single tone message
m(t) = Ac cos 2πfm t, for

• LSB we write s(t) = Ac cos(2π[fc − fm ]t),

• USB we write s(t) = Ac cos(2π[fc + fm ]t)).

For USB, the IQ receiver output

x(t) =Ac cos 2π(fc + fm )t cos 2πfc t

Ac
= [cos 2π(2fc + fm )t + cos 2πfm t]
2
y(t) =Ac cos 2π(fc + fm t) sin 2πfc t
Ac
= [sin 2π(2fc + fm t) − sin 2πfm t]
2
After the LPF,
Ac
x(t) = cos 2πfm t
2
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 314
INSTRUCTOR COPY

Figure 5.19: Map to receive SSB-SC

and
Ac
y(t) = sin 2πfm t
2
The LPF output is the message (cosine term) and a phase shifted version (sine term).

Exercise 97
repeat this calculation with a small frequency offset ∆f .

5.5.1.3 Limitation of Methods 1 and 2

A problem with both Method 1 and Method 2 is that if there is another signal present
where the other (opposite) sideband would have been if we were using ISB, DSB or
AM instead of SSB, then that signal will also be demodulated and will interfere with
the desired message.

Exercise 98
show that this is true for the USB signal sU SB (t) = Ac cos(2π[fc + f1 ]t) with
an interfering LSB signal sint (t) = Ac cos(2π[fc − f2 ]t).
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 315

5.5.2 SSB-SC receiver using phasing method with Hilbert

transform

This method of receiving SSB-SC is analogous to the SSB-SC transmitter using the
Hilbert transform. See also https://www.dsprelated.com/showarticle/176.php

5.5.2.1 Software implementation with Hilbert transform

To demodulate the message m(t) from the SSB-SC signal r+ (t), we follow these steps
for the USB case
INSTRUCTOR COPY

1. downconvert r+ (t) = [m(t) + j m̂(t)]ej2πfc t by multiplying by e−j2πfc t to obtain

r̃(t) = m(t) + j m̂(t)
2. Take the Hilbert transform of the imaginary part of r̃(t) to obtain m(t) − jm(t).
3. subtract the imaginary part from the real part of this expression
4. What remains is 2m(t)

The block diagram of this receiver requires a block with a complex signal input
i(t) + jq(t) and output i(t) + j q̂(t).

The block diagram also requires block with complex signal input i(t) + jq(t) and
output i2 (t) + jq2 (t) = i(t) − q(t) + j[i(t) + q(t)], which may be written
    
i2 1 −1 i
= (5.9)
q2 1 1 q
 
√ cos θ − sin θ
This matrix is 2 times the rotation matrix for θ = π/4. The output
sin θ cos θ
√ p i+q
amplitude is 2 i2 + 2q 2 and phase arctan{ i−q }. This block may be viewed as a
√ jπ/4
complex mixer with local oscillator 1 + j = 2e

Exercise 99
Sketch the signal spectrum at each point in the block diagram, assuming that
m(t) has the spectrum shown in section 5.2.4.

Exercise 100
Write the above 4 steps for the LSB case.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 316

Exercise 101
Sketch the signal spectrum at each point in the block diagram, assuming an
ISB signal at the receiver input, one USB with one message and one LSB with
a different message. Show how both the USB and LSB can be demodulated
simultaneously. Hint: consider the receiver input

r+ (t) = [m1 (t) + j m̂1 (t) + m2 (t) − j m̂2 (t)]ej2πfc t (5.10)

and recall

H{m(t)} = m̂(t)
H{H{m(t)}} = −m(t)
INSTRUCTOR COPY

Solution

r̃(t) = [m1 (t) + j m̂1 (t) + m2 (t) − j m̂2 (t)]

r̃qH (t) = [m1 (t) − jm1 (t) + m2 (t) + jm2 (t)]
= m1 (t) + m2 (t) + j[−m1 (t) + m2 (t)]
r̃pm = m1 (t) + m2 (t) − [−m1 (t) + m2 (t)] + j{m1 (t) + m2 (t) + [−m1 (t) + m2 (t)]}
= 2m1 (t) + 2jm2 (t)

The real part of the ISB receiver output is the USB and the imaginary part is
the LSB.

5.5.2.2 Analog (real) implementation with Hilbert transform

Exercise 102
Repeat the analysis above for a real implementation using cos and sin oscillators
and an ISB input signal.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 317

Solution

r(t) = m1 (t) cos 2πfc t − m̂1 (t) sin 2πfc t

+ m2 (t) cos 2πfc t + m̂2 (t) sin 2πfc t
i(t) = r(t) cos 2πfc t
= m1 (t) + m2 (t) + double frequency terms
q(t) = −r(t) sin 2πfc t
= m̂1 (t) − m̂2 (t) + double frequency terms
q̂(t) = −m1 (t) + m2 (t)
i(t) − q̂(t) = m1 (t)
i(t) + q̂(t) = m2 (t)
INSTRUCTOR COPY

From the above, we observe that the two (independent) sidebands are received
without interference between each other.

Exercise 103
Repeat the above analysis for the case m1 (t) = cos 2πf1 t and m2 (t) = cos 2πf2 t

5.5.3 SSB-SC receiver using Weaver Demodulator

5.5.3.1 Weaver modulator description and operation

Another method of receiving SSB-SC uses the Weaver demodulator which is analogous
to the Weaver modulator discussed earlier in section 5.4.2.

The Weaver demodulator uses a low pass filter with cutoff at B/2 (half the message
bandwidth) compared to the full message bandwidth low pass filter B used in the
previous section above.

The Weaver demodulator operates in several steps (fig. 5.20).

The following math is done for the USB case. We assume the message occupies a
bandwidth of 0–3000 Hz. Each of the following 5 stages is shown in order in fig. 5.21.

1. The receiver input signal is real s(t) = m(t) cos 2πfc t− m̂(t) sin 2πfc t or complex
s+ (t) = [m(t) + j m̂(t)]ej2πfc t . The idea is the same in both cases.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 318

IN Re OUT
s(t) on m(t) with
carrier fc bandwidth BW
DC to BW
2

Legend
f1 f2
−j2πf1 t j2πf2 t BW
e e LSB fc − 2 − BW
2
BW BW
U SB fc + 2 2

Real
Complex
INSTRUCTOR COPY

Figure 5.20: Complex block diagram for Weaver’s demodulator

f
−fc fc

f
BW
2

BW

0 Hz

Figure 5.21: Spectra at each stage of a USB receiver

 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 319

2. Downconvert the real USB signal to s̃1 (t) using a complex local oscillator
BW
e−j2πf1 t = e−j2π(fc + 2
)t
= e−j2π(fc +f2 )t
with a frequency offset BW2
= 1.5 kHz relative to fc . We choose BW
2
to be a
frequency in the approximate middle of the message bandwidth, e.g. 1500 Hz
for a 0–3000 Hz voice signal.
3. Low pass filter the result to eliminate adjacent undesired signals. The resulting
signal s̃2 (t) is a complex signal with asymmetric spectrum in the frequency
range − BW 2
→ BW2
or −1.5 kHz → 1.5 kHz.
4. Do a frequency shift of s̃2 (t) with a complex local oscillator e+j2πf2 t where
f2 = BW
2
= 1.5 kHz as defined above, resulting in s̃3 (t) in the frequency range 0
to BW or 0–3000 Hz. s̃3 (t) is an analytic signal with positive frequencies only.
INSTRUCTOR COPY

5. Take the real part of s̃3 (t) to obtain the real signal m(t) with both positive and
negative frequencies that can be listened to or decoded correctly.

The net result of the two frequency shifts and LPF in the Weaver demodulator is
that the suppressed carrier frequency in s(t) is shifted to zero frequency, the message
is recovered and all other unwanted signals are filtered out.

The problem with the interfering signal mentioned in the previous section does
not arise for the Weaver demodulator because the LPF bandwidth is B/2 instead of
B.

The Weaver demodulator also works for Morse code or other digital signals that
are either processed by ear or other decoder that operates on a real (not complex)
(usually audio-bandwidth) signal (such as telephone DTMF tones). Morse code is
SSB where the single tone message switches on and off in Morse code patterns. To
make figures for Morse code, replace the general spectral shape for M (f ) with a single
tone (δ− function) in the middle of the M (f ) frequency range.

The Weaver parameters BW and f1 are changed from the values BW/2 = f1 =
1500 Hz used for SSB voice. For Morse code, the bandwidth B may be set to an
appropriate value (e.g. 50 Hz for Morse code) and the frequency offset f1 is set to a
tone (pitch) that is pleasant to the ear (e.g. 400 Hz).

5.5.3.2 SSB-SC Weaver demodulator operation in complex notation (soft-

ware implementation

This demodulator is suitable for software implementation.

 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 320

We consider two cases for the input signal, real and complex, and for each case,
show the demodulation steps in complex notation. For this we refer to the block
diagram in fig. 5.20.

A real input signal arises when the signal is taken directly from antenna and
preamp and (perhaps) an analog real downconverter stage (cos oscillator only) as
shown in fig. 2.22.

A complex input signal arises when a complex downconversion (e.g. cos and sin
oscillators in an analog mixer with two outputs I and Q) has taken place ahead of 2
ADCs, as done in the USRP receiver (fig. 2.25).

5.5.3.3 Real input, complex math

INSTRUCTOR COPY

The mathematics below is for a real input signal that arises when the signal is taken
directly from antenna and preamp and (perhaps) an analog real downconverter stage
(cos oscillator only).

1. At the input we have the real signal

s(t) =Re{s̃(t)ej2πfc t }
=m(t) cos 2πfc t − m̂(t) sin 2πfc t

where s̃(t) = m(t) + j m̂(t).

2. recall

cos 2πfc t = 0.5ej2πfc t + 0.5e−j2πfc t

− sin 2πfc t = 0.5jej2πfc t − 0.5je−j2πfc t

3. After the first mixer we have

s̃1 (t) =[m(t) cos 2πfc t − m̂(t)sin2πfc t]e−j2πf1 t

=[m(t) cos 2πfc t − m̂(t)sin2πfc t]e−j2π(fc +f2 )t
=0.5[m(t)ej2πfc t + m(t)e−j2πfc t
+ j m̂(t)ej2πfc t − j m̂(t)e−j2πfc t ]e−j2π(fc +f2 )t
=0.5[m(t)e−j2πf2 t + m(t)e−j2π(2fc +f2 )t
+ j m̂(t)e−j2πf2 t − j m̂(t)e−j2π(2fc +f2 )t ].
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 321

4. After the LPF the analytic signal is shifted and the double frequency term is
removed

s̃2 (t) =m(t)e−j2πf2 t + j m̂(t)e−j2πf2 t

=[m(t) + j m̂(t)]e−j2πf2 t

5. After the second mixer we have

s̃3 (t) =s̃2 (t)ej2πf2 t

=m(t) + j m̂(t) = m+ (t) = s̃(t)

6. The last step is to take the real part of the complex baseband signal or add
m− (t) to obtain the message
INSTRUCTOR COPY

m(t) =Re{s̃(t)}
=0.5m+ (t) + 0.5m− (t)
=0.5[m(t) + j m̂(t)] + 0.5[m(t) + j m̂(t)]
=m(t)

Exercise 104
Repeat the figure and mathematics for LSB.

5.5.3.4 Complex input, complex math

The mathematics below is for a complex input that arises when a complex downcon-
version (e.g. cos and sin oscillators in an analog mixer with two outputs I and Q) has
taken place ahead of the USB receiver.

1. The complex input signal is

s+ (t) =s̃(t)ej2πfc t
=m(t) cos 2πfc t − m̂(t) sin 2πfc t
+ j[m(t) sin 2πfc t + m̂(t) cos 2πfc t]

where s̃(t) = m(t) + j m̂(t).

 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 322

2. After the first mixer we have

s̃1 (t) =s̃(t)ej2πfc t e−j2πf1 t

BW
=s̃(t)ej2πfc t e−j2π(fc + 2
)t

=s̃(t)ej2πfc t e−j2π(fc +f2 )t

=s̃(t)e−j2πf2 t .

3. After the LPF eliminates other unwanted signals s̃2 (t) = s̃1 (t).

4. After the second mixer we have

s̃3 (t) =s̃2 (t)ej2πf2 t

=s̃(t)e−j2πf2 t ej2πf2 t
INSTRUCTOR COPY

=s̃(t)
=m(t) + j m̂(t)

5. The last step is to take the real part of the complex baseband signal to obtain
the message
m(t) = Re{s̃(t)}

5.5.3.5 SSB-SC Weaver demodulator operation in real notation (hard-

ware implementation

The mathematics below is for a real input signal that arises when the signal is taken
directly from antenna and preamp and (perhaps) an analog real downconverter stage
(cos oscillator only).

The operations of a Weaver demodulator can also be shown in real notation as

shown in fig. 5.22.

The real version of the Weaver demodulator can be built entirely in analog, or
entirely in digital. It may also be split, with the first cos/sin mixer in analog (as
done in the USRP daugtherboard) and the second mixer digital (implemented in
GNURadio software).

In fig. 5.22, for USB, the first oscillator is f0 + B/2.

The mathematics for the Weaver demodulator in real notation is done in the
example below, assuming that the message is a dual-tone message, i.e. the sum of
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 323

DC to BW
2

cos(2πf1 t) cos(2πf2 t)
IN OUT
s(t) on m(t) with
sin(2πf1 t) sin(2πf2 t)
carrier fc bandwidth BW

DC to BW
2 Legend
f1 f2
BW
LSB fc − 2 − BW
2
BW BW
U SB fc + 2 2

Figure 5.22: Real SSB receiver

INSTRUCTOR COPY

cos waves at 500 Hz and 2000 Hz. This example should give an appreciation of the
value of complex signals to simply both the concepts and the algebra.

Recall
cos α cos β = [cos(α − β) + cos(α + β)]/2
sin α cos β = [sin(α − β) + sin(α + β)]/2
sin α sin β = [cos(α − β) − cos(α + β)]/2
cos α sin β = [− sin(α − β) + sin(α + β)]/2
cos(α + β) = cos α cos β − sin α sin β
sin(α + β) = sin α cos β + cos α sin β

1. The message is
m(t) = cos 2πf2 t + cos 2πf3 t
= cos 2π500t + cos 2π2000t
m̂(t) = sin 2πf2 t + sin 2πf3 t
= sin 2π500t + sin 2π2000t

For a real SSB signal (upper sideband), the input signal is

s(t) =m(t) cos 2πfc t − m̂(t)sin2πfc t
=[cos 2πf2 t + cos 2πf3 t] cos 2πfc t
− [sin 2πf2 t + sin 2πf3 t] sin 2πfc t
= cos 2πf2 t cos 2πfc t − sin 2πf2 t sin 2πfc t
+ cos 2πf3 t cos 2πfc t − sin 2πf3 t sin 2πfc t
= cos 2π(f2 + fc )t + cos 2π(f3 + fc )t
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 324

Writing f0 = fc yields

s(t) = cos 2π(f2 + f0 )t + cos 2π(f3 + f0 )t.

2. The first oscillator is at f0 + B2 for upper sideband. The upper branch after the
first mixer and low pass filter but before the second mixer has B2 = B/2 = f1
and

s(t) cos 2π(f0 + B2 )t =0.5[cos 2π(f2 + f0 )t + cos 2π(f3 + f0 )t] cos 2π(f0 + B2 )t
=0.5[cos 2π(f2 − B2 )t + cos 2π(f3 − B2 )t]
=0.5[cos 2π(500 − 1500)t + cos 2π(2000 − 1500)t]
=0.5[cos 2π(−1000)t + cos 2π(500)t]
INSTRUCTOR COPY

3. The second oscillator is at B2 for upper sideband. The upper branch after the
second mixer is described by

0.5[cos 2π(f2 − B2 )t + cos 2π(f3 − B2 )t] cos 2πB2 t

=0.25[cos 2πf2 t + cos 2π(f2 − 2B2 )t
+ cos 2πf3 t + cos 2π(f3 − 2B2 )t]
=0.25[cos 2π(500)t + cos 2π(500 − 3000)t
+ cos 2π(2000)t + cos 2π(2000 − 3000)t]
=0.25[cos 2π(500)t + cos 2π(−2500)t
+ cos 2π(2000)t + cos 2π(−1000)t]

4. The first oscillator is at f0 + B2 for upper sideband. The lower branch after the
first mixer and low pass filter but before second mixer is descibed by

s(t)[− sin 2π(f0 + B2 )t]

= − [cos 2π(f2 + f0 )t + cos 2π(f3 + f0 )t] sin 2π(f0 + B2 )t
= − 0.5[cos 2π(f2 + f0 )t sin 2π(f0 + B2 )t
− cos 2π(f3 + f0 )t sin 2π(f0 + B2 )t]
=0.5[sin 2π(f2 − B2 )t + sin 2π(f3 − B2 )t]
=0.5[sin 2π(500 − 1500)t + sin 2π(2000 − 1500)t]
=0.5[sin 2π(−1000)t + sin 2π(500)t]

5. The second oscillator is at B2 for upper sideband. The lower branch after second
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 325

mixer is described by

0.5[sin 2π(f2 − B2 )t + sin 2π(f3 − B2 )t][− sin 2πB2 t]

=0.25[− cos 2π(f2 − 2B2 )t + cos 2πf2 t
− cos 2π(f3 − 2B2 )t + cos 2πf3 t]
=0.25[− cos 2π(500 − 3000)t + cos 2π(500)t
− cos 2π(2000 − 3000)t + cos 2π(2000)t]
=0.25[− cos 2π(−2500)t + cos 2π(500)t
− cos 2π(−1000)t + cos 2π(2000)t]

6. The sum of the second mixer outputs is the desired message

INSTRUCTOR COPY

0.5[cos 2πf2 t + cos 2πf3 t] =0.5[cos 2π(500)t + cos 2π(2000)t]

=0.5m(t)

The real analysis of the Weaver demodulator shows the cancellation of terms
needed to get the desired result. The complex analysis is much simpler and there is
no cancellation of terms.

5.6 Chapter review questions

Q 5.1 Recall fig. 5.17 which shows the stages of a USB transmitter as well as the math
that followings describes each stage. Repeat both the figure and the mathematics for
an LSB transmitter.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 326

Solution

BW

f
BW
2

f
INSTRUCTOR COPY

BW
fc − 2

f
fc

f
−fc fc
0 Hz

Figure 5.23: Stages for LSB transmitter

Following fig. 5.23:

1. the first stage is a real symmetrical input spectrum m(t) = 0.5m+ (t) +
0.5m− (t) with positive and negative frequencies. We wish to transmit
only the LSB

m− (t) = m(t) − j m̂(t), negative frequencies only, green, LSB of m(t)

2. after the complex multiply with ej2πf1 t

s̃1 (t) =m(t)ej2πf1 t

=0.5m+ (t)ej2πf1 t + 0.5m− (t)ej2πf1 t

3. after the LPF the result s̃2 (t) is the negative frequency analytic signal
shifted by f1 . This is clear from the frequency diagram (fig. 5.23) where
s̃2 (t) is obtained by filtering out the (red) term 0.5m+ (t)ej2πf1 t in s̃1 (t).
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 327

Omitting the scaling factor 0.5,

s̃2 (t) =m− (t)ej2πf1 t

=[m(t) − j m̂(t)] · [cos 2πf1 t + j sin 2πf1 t]
=m(t) cos(2πf1 t) + jm(t) sin(2πf1 t) − j m̂(t) cos(2πf1 t) + m̂(t) sin(2πf1 t)

4. after the second mixer

s̃3 (t) =s+ (t)

=s̃2 (t)ej2π(fc −f1 )t
=[m(t) − j m̂(t)]ej2πf1 t ej2π(fc −f1 )t
=[m(t) − j m̂(t)]ej2πfc t
INSTRUCTOR COPY

=[m(t) − j m̂(t)] · [cos 2πfc t + j sin 2πfc t]

=m(t) cos(2πfc t) + jm(t) sin(2πfc t) − j m̂(t) cos(2πfc t) + m̂(t) sin(2πfc t)
s(t) =Re{s̃3 (t)}
=m(t) cos(2πfc t) + m̂(t) sin(2πfc t)

which is a real passband LSB signal.

The following is Optional extra information that is not required for a
correct solution.
Check by substituting a single tone message m(t) = cos 2πfm t to obtain

s(t) = cos 2πfm t cos(2πfc t) + sin 2πfm t sin(2πfc t)

= cos 2π(fc − fm )t

and observe that s(t) is LSB with a single tone below fc . Optional:
observe that

Im{s̃3 (t)}
=m(t) sin(2πfc t) − m̂(t) cos(2πfc t)

is also LSB, since with a single tone message

s(t) = cos 2πfm t sin(2πfc t) − sin 2πfm t cos(2πfc t)

= − sin 2π(fc − fm )t = cos[2π(fc − fm )t + π/2]

is the same LSB signal shifted in phase by π/2. Thus we can choose
either the real or imaginary part of s̃3 (t) as the real passband LSB signal.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 328

Q 5.2 Compare the power and efficiency of AM and SSB. One approach is to find the
power in one sideband for each case and compare. You may assume 100% modulation.

Solution
A2c
Assume single tone modulation. AM carrier power is 2
and each sideband
2 2
power is Ac8µ . The total power is with 100% mod so

A2c A 2 µ2 A2 A2 A2
+2 c = c +2 c = 3 c.
2 8 2 8 4
1
For AM, the power in one sideband is 6
of the total power as shown by
INSTRUCTOR COPY

A2c µ2 A2c
8 8 1 A2c 1
A2c
= A2c
= 2
=
34 34 6 Ac 6

For SSB, the power in one sideband is the total power. So, if we use an SSB
receiver to receive the one sideband for both AM and SSB, then SSN is 6 times
more power efficient than AM.

Q 5.3 Consider the SSB-SC Weaver demodulator receiver from section 5.5.3. In
section 5.5.3.1 an analysis is done for a USB signal. Repeat the analysis but for an
LSB signal with a complex input.

Solution
See also the solution to the same problem in Worksheet 5.
Following the block diagram of fig. 5.20:

1.

s̃(t) =m− (t) for LSB

s+ (t) =0.5s̃(t)ej2πfc t

The next equation is extra information that is not required for a correct
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 329

solution: The figure below shows the real LSB input

s(t) =0.5s+ (t) + 0.5s− (t)

=0.5s̃(t)ej2πfc t + 0.5s̃∗ (t)e−j2πfc t
=0.5m− (t)ej2πfc t + 0.5m∗− (t)e−j2πfc t
=0.5[m(t) − j m̂(t)]ej2πfc t + 0.5[m(t) + j m̂(t)]e−j2πfc t
=m(t) cos(2πfc t) + m̂(t) sin(2πfc t)

In this question, we consider only the analytic signal input s+ (t) and thus
omit terms arising from s− (t).
BW
From fig. 5.20: for LSB f1 = fc − 2
and f2 = − BW
2
thus f1 = fc + f2
INSTRUCTOR COPY

2. after the first complex multiply

s̃1 (t) =s+ (t)e−j2πf1 t

=s+ (t)e−j2π(fc +f2 )t
=s̃(t)ej2πfc t e−j2π(fc +f2 )t
=s̃(t)e−j2πf2 t
BW
=s̃(t)e+j2π 2
t

In the above, we have omitted the terms arising from s− (t) because the
problem statement said the receiver input is s+ (t). even though both
s+ (t) and s− (t) are shown in the figure below.

3. After the LPF, the unwanted terms (signals) arising from s− (t) are filtered
out. Since these terms were omitted, s̃2 (t) = s̃1 (t).

4. after the second mixer

s̃3 (t) =s̃2 (t)e+j2πf2 t

BW
=s̃2 (t)e−j2π 2
t

=s̃(t)e−j2πf2 t e+j2πf2 t
=s̃(t)
=m(t) − j m̂(t)
=m− t
m(t) =Re{s̃(t)}

Q 5.4 Draw the spectrum for each point in the complex block diagram of fig. 5.20
for LSB.
 CHAPTER 5. SINGLE SIDE BAND MODULATION (SSB) 330

Solution
See fig. 5.24.

f
INSTRUCTOR COPY

0 Hz

Figure 5.24: Stages for LSB receiver

 Chapter 6

Angle modulation (FM and PM)

INSTRUCTOR COPY

6.1 Angle modulation Overview

In this chapter, we describe angle modulation, which includes both frequency and
phase modulation. In both cases, amplitude is kept constant, and the phase angle of
the carrier wave varies with the message signal.

For angle modulation the complex baseband signal has constant amplitude and
time-varying phase

s̃(t) = a(t)ejφ(t) in general

a(t) = Ac for angle modulation
φ(t) is related to m(t)
s̃(t) = Ac ejφ(t) for angle modulation

The angle φ(t) is time-varying in step with the message m(t) as compared to amplitude
modulation that has constant phase s̃(t) = a(t)ejφ0 and time-varying amplitude a(t)
in step with the message m(t).

For phase modulation (PM), the phase angle

φ(t) = kp m(t)

is linearly related to the message m(t). For frequency modulation (FM), the instan-
taneous frequency (which is 1/(2π times the derivative of the phase angle)

1 dφ(t)
fi (t) = = kf m(t)
2π dt

331
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 332

is linearly related to the message m(t).

FM and PM signals have a constant amplitude or constant envelope, thus enabling

the use of power-efficient nonlinear power amplifiers. FM signals are more resistant
to noise than AM, but at the cost of larger bandwidth, as will be explained in later
sections.

Angle modulation is widely used for

• FM radio broadcasting in the range 88-108 MHz,

• analog TV audio in the range 54-88 MHz (channels 2-6) and 174-216 MHz
(channels 7-13) and UHF TV (470-614 MHz).
INSTRUCTOR COPY

• public safety two-way radios (150-174 and 450-470 MHz).

• family radio service walkie-talkies near 462 MHz.

• digital telemetry and paging systems

As we will see in the section on pre-emphasis and de-emphasis, when the message
is analog (voice or music), both forms of angle modulation (frequency modulation
and phase modulation) are used simultaneously, with FM used for lower message
frequencies and PM for higher message frequencies.

The general case where both amplitude and phase are time-varying at the same
time

s̃(t) = a(t)ejφ(t)

will be considered in Chapter 7.

Figure 6.1 compares AM and FM real passband waveforms for an example analog
message. The figure shows how for AM the carrier amplitude varies with the message,
whereas for FM the carrier frequency varies in step with the message.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 333
INSTRUCTOR COPY

Figure 6.1: AM vs FM modulations

For PM, the phase varies with the message. If we use an example digital message
in the form of a square wave, then the distinction between FM and PM is more
intuitive, as shown later.

6.2 Angle modulation complex baseband, analytic

signal and real passband signal

In this section we will write the mathematical expressions for the signals
˜ ), S+ (t), S− (f ), S(f )
s̃(t), s+ (t), s− (t), s(t), S(f

obtained in chapter 2 for the special case of PM and FM.

 CHAPTER 6. ANGLE MODULATION (FM AND PM) 334

6.2.1 Phase modulation (PM) with general message

6.2.1.1 PM complex baseband s̃(t)

For PM signals, the complex baseband (message) signal s̃(t) = a(t)ejφ(t) , where we
set the amplitude to be a constant DC value Ac that establishes the voltage and
power level of the signal. In PM, the message controls the phase, thus the phase is a
constant plus the message m(t) scaled by a constant kp . Thus we write

φ(t) = kp m(t) + φ0
s̃(t) = a(t)ejφ(t)
= Ac ej[kp m(t)+φ0 ]
INSTRUCTOR COPY

= Ac cos[kp m(t) + φ0 ] + jAc sin[kp m(t) + φ0 ]

fig. 2.6 shows s̃(t) when m(t) is a ±π/2 square wave, kp = 1 and φ0 = −π/4.

0
=(z(t))

−1
1
0 0.5 0.5
1 1.5 0
<(z(t))
t

Figure 2.6 (repeated): Complex baseband phase shift keying with constant a(t) and
two values for φ(t).
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 335

For this case, we write

φ(t) = ±π/2 − π/4

s̃(t) = a(t)ejφ(t)
= Ac e±jπ/2−π/4
= Ac cos[±π/2 − π/4] + jAc sin[±π/2 − π/4]
= Ac cos[π/4] + jAc sin[π/4]
or Ac cos[−3π/4] + jAc sin[−3π/4]
Ac
= √ (1 + j)
2
Ac
or √ (−1 − j)
2
INSTRUCTOR COPY

fig. 2.6 was previously shown as an example complex baseband signal in section 2.2.1.

Figure 6.2: Complex baseband phase shift keying with constant a(t) and cosine wave
for m(t) with φ(t) = kp m(t) and kp small

fig. 6.2 shows another example of = s̃(t) when m(t) = cos 2πfm t and kp = π/8 is
small so that the phase varies by ±π/8 radians around its mean value φ0 = 3π/4.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 336
INSTRUCTOR COPY

Figure 6.3: Complex baseband phase shift keying with constant a(t) and cosine wave
for m(t) with φ(t) = kp m(t) and kp small

fig. 6.3 shows the same example but with kp = 2 so that the phase varies by ±2
radians around its mean value.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 337
INSTRUCTOR COPY

Figure 6.4: Complex baseband phase shift keying with constant a(t) and cosine wave
for m(t) with φ(t) = kp m(t) and kp small

fig. 6.4 show the same example but with kp = 4 so that the phase varies by ±4
radians (a range of more than 2π) around its mean value.

Observe that the projections onto the I and Q planes yield a waveform that is
difficult to interpret without seeing the back-and-forth movement of the helix as time
progresses.

These 3 figures with sinusoidal modulation at a fixed frequency apply to FM and

well as PM.

6.2.1.2 PM analytic signal

The PM analytic (positive frequency, complex passband) signal

 CHAPTER 6. ANGLE MODULATION (FM AND PM) 338

s+ (t) = a(t)ejφ(t) ej2πfc t

= Ac ejφ0 ejkp m(t) ej2πfc t
= Ac ej(2πfc t+kp m(t)+φ0 )
s− (t) = a(t)e−jφ(t) e−j2πfc t
= Ac e−jφ0 e−jkp m(t) e−j2πfc t
= Ac e−j(2πfc t+kp m(t)+φ0 )

Exercise 105
Sketch or make a 3-D Matlab plot of s+ (t) using the parameters
INSTRUCTOR COPY

s+ (t) = a(t)ejφ(t) ej2πfc t

= Ac ejφ0 ej±π/2 ej2πfc t
= Ac ej(2πfc t+±π/2+φ0 )

The carrier wave helix jumps in phase when the message changes.

6.2.1.3 PM real passband signal

s(t) = [s+ (t) + s− (t)]/2

= <{s+ (t)}
= Ac cos(2πfc t + kp m(t) + φ0 )

Exercise 106
Repeat the previous exercise to show s(t)

Solution
The projection of s+ (t) onto the real axis shows the phase jumps in the cosine
wave carrier. Observe how the projections of the helix on the real and imaginary
axes that show the phase shifts vary as φ0 is changed.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 339

6.2.1.4 PM complex baseband spectrum

The complex baseband spectrum for PM

s̃(t) ↔ S̃(f )
s̃(t) = Ac ejφ ejkp m(t)
m(t) ↔ M (f )
ejkp m(t) ↔ unknown
s̃(t) = Ac ejφ ejkp m(t) ↔ S̃(f ) = unknown

is difficult to find in general, since we don’t know the Fourier transform of the ex-
ponential of an arbitrary function. Without an expression for S̃(f ), we cannot find
INSTRUCTOR COPY

S+ (f ) and S(f ).

The workaround for this issue is explained in section 6.5.

6.2.2 Frequency modulation (FM) with general message

6.2.2.1 FM complex baseband s̃(t)

For FM signals, the complex baseband (message) signal s̃(t) = a(t)ejφ(t) , where we set
the amplitude to be a constant DC value Ac that establishes the voltage and power
level of the signal. In FM the message controls the instantaneous frequency, so we
can write

fi (t) = kf m(t)

The instantaneous frequency for a complex baseband signal varies plus or minus its
average value 0.

Recall that frequency is the derivative of phase

1 dφ(t)
fi (t) =
2π Z dt
t
φ(t) = 2π fi (τ )dτ
τ =0

Thus the phase is the integral of the message m(t) scaled by a constant kf , and we
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 340

write

fi (t) = kf m(t)
Z t
φ(t) = 2π fi (τ )dτ
τ =0
Z t
= 2πkf m(τ )dτ
τ =0
jφ(t)
s̃(t) = a(t)e
Rt
= Ac ej2π τ =0 m(τ )dτ
INSTRUCTOR COPY

1
=(z(t))

−1 1
0 0
0.5 1 1.5 −1 <(z(t))
t

Figure 2.7 (repeated): Complex baseband frequency-shift keying with constant a(t)
and sinusoidal φ(t).

fig. 2.7 shows s̃(t) when m(t) is a cosine wave and thus φ(t) is a sine wave. We
observe the helix rotates in one direction and then reverses direction periodically in
step with the message, thus illustrating the positive and negative frequencies fi (t) =
kf m(t) in step with the message m(t).
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 341
INSTRUCTOR COPY

Figure 6.5: FM baseband with a half-cycle of cosine wave modulation

In fig. 6.5 we show s̃(t) when fi (t) = kf m(t) is half a cycle of a cosine wave
Rt
Am cos 2πfm t for 0 ≤ t ≤ 1/fm and thus φ(t) = 2π τ =0 fi (τ )dτ is half a cycle of a
sine wave. We observe the helix rotates in step with m(t) starting at high frequency
when t = 0 and slowing down to zero when t = 1/(2fm ) and speeding up in the
opposite direction until t = 1/fm
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 342
INSTRUCTOR COPY

Figure 6.6: FM complex baseband with square wave modulation

In fig. 6.6
R t we show s̃(t) when m(t) is half a cycle of a square wave and thus
φ(t) = 2πkf τ =0 m(τ )dτ is half a cycle of a triangle wave. We observe the helix rotates
in step with m(t) at constant frequency when t = 0. The phase increases linearly
with time according to the upslope of the triangle wave and thus the derivative of
the phase is constant and thus the helix rotates at a constant frequency. The helix
reverses its direction of rotation abruptly in the opposite direction when the square
wave m(t) jumps from +1 to = 1 and the triangle wave φ(t) starts its downslope.

The projection on the imaginary axis looks like a real passband phase shift keying
(PSK) waveform, but the 3D helix is not an analytic PSK waveform. The projection
hides the presence of negative frequencies, thus reaffirming the need to look at 3-D
signals for a complete understanding.

fig. 6.3 and fig. 6.2 and fig. 6.4 are more examples that apply to FM as well as
PM with sinusoidal modulation since the PM waveform ejβp cos 2πfm t or FM waveform
ejβf sin 2πfm t has the same shape with only a change in starting phase.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 343

6.2.2.2 FM analytic signal

The FM analytic (positive frequency, complex passband) signal

s+ (t) = a(t)ejφ(t) ej2πfc t

= a(t)Ac ejθ(t)
Rt
= Ac ej[2πkf τ =0 m(τ )dτ j2πfc t
e
Rt
= Ac ej[2πfc t+2πkf τ =0 m(τ )dτ ]

s− (t) = a(t)e−jφ(t) e−j2πfc t

Rt
INSTRUCTOR COPY

= Ac e−j[2πkf τ =0 m(τ )dτ −j2πfc t

e
Rt
= Ac e−j[2πfc t+2πkf τ =0 m(τ )dτ ]

fig. 6.7 shows s+ (t) when m(t) is a square wave.

Figure 6.7: FM analytic signal with square wave modulation

 CHAPTER 6. ANGLE MODULATION (FM AND PM) 344

The instantaneous frequency for s+ (t) is

1 dθ
fi (t) = = fc + kf m(t)
2π dt
The helix rotates in one direction, but the rate of rotation (instantaneous frequency)
fi (t) = fc + kf m(t) changes with the message. Note that for an analytic signal, fi (t)
varies plus or minus around its average value fc .

6.2.2.3 FM real passband signal

s(t) = [s+ (t) + s− (t)]/2

= <{s+ (t)}
INSTRUCTOR COPY

Z t
= Ac cos(2πfc t + 2πkf m(τ )dτ )
τ =0

The real passband waveform is obtained from the projection of s+ (t) in fig. 6.7
above, showing the two discrete frequencies corresponding to the two states of the
square wave message. The frequency of s(t) changes in step with the message fi (t) =
fc + kf m(t). Note that for a real passband signal, fi (t) varies plus or minus around
its average value fc .

A real passband FM waveform is also shown in fig. 6.1 for a general message.

6.2.2.4 FM complex baseband spectrum

The complex baseband spectrum for PM

s̃(t) ↔ S̃(f )
Rt
s̃(t) = Ac ej[2π τ =0 m(τ )dτ

m(t) ↔ M (f )
Rt
ej[2π τ =0 m(τ )dτ
↔ unknown
Rt
s̃(t) = Ac ej[2π τ =0 m(τ )dτ
↔ S̃(f ) = unknown
is difficult to find S̃(f ) in general, since we don’t know the Fourier transform of the
complex exponential of an arbitrary function. Without an expression for S̃(f ), we
cannot find S+ (f ) and S(f ).

We avoid the challenge of finding the Fourier transform of the exponential of a

general message m(t) by recognizing that
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 345

• the Fourier transform of the complex exponential of a cosine or sine wave can
be found as shown in section 6.5
• any m(t) can be represented as the sum of cosine waves at different frequencies.
• We recall from section 2.7.6.3 and section 1.6.2 that a discrete-time message
divided into (possibly overlapping) frames is made up of the sum of cosine
waves that will be different from one frame to the next.
• we assume that for the time duration of a frame, the message is the sum of
cosine waves at known frequencies.
• if we choose only one of the cosine waves to make a single tone message, then
the desired Fourier transform can be found.
INSTRUCTOR COPY

• Furthermore, we can find the Fourier transform for a message that is a finite
sum of cosine waves.

Details of this approach are in section 6.5 where we find the spectrum of s̃(t) =
Ac ejβ sin 2πfm t .

6.3 Real passband FM with general message - al-

ternate derivation in time domain

This derivation in this section starts with the expression for a real FM passband
signal, whereas the previous section started with the PM and FM complex baseband
signal with a general message. The results are the same.

Recall the idea of AM being that the instantaneous amplitude a(t) of a carrier
wave is varied linearly with the baseband message signal m(t) around a constant bias
value Ac so that
a(t) = Ac + Ac ka m(t).
To derive the equation for an FM wave, we have three starting points:

1. By analogy, the idea of frequency modulation is that the instantaneous fre-

quency fi (t) of a carrier wave is varied linearly with the baseband message
signal m(t) around a constant value fc . Thus for a general message m(t) the
idea of FM is
fi (t) = fc + kf m(t)
where the constant kf represents the frequency sensitivity of the modulator,
expressed in hertz per volt.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 346

2. Recall that a general signal is written:

s(t) =a(t) cos θ(t)

=a(t)cos[2πfc t + φ(t)]

where θ(t) = 2πfc t + φ(t) has a linear variation at rate fc and a time varying
part φ(t).

3. Recall the instantaneous frequency of a signal is:

1 dθ(t)
fi (t) =
2π dt
1 dφ(t)
=fc +
2π dt
INSTRUCTOR COPY

Combining these three starting points, we can write:

Z t
θ(t) =2π fi (α)dα
0
Z t
=2π [fc + kf m(α)]dα
0
Z t
=2πfc t + 2πkf m(α)dα
0
 Z t 
s(t) =Ac cos 2πfc t + 2πkf m(α)dα
0

The FM signal s(t) can be written in standard IQ format:

s(t) =i(t) cos 2πfc t − q(t) sin 2πfc t

=Re{a(t)ejφ(t) ej2πfc t }
=Re{[I(t) + jQ(t)][cos 2πfc t + j sin 2πfc t]}
=I(t) cos 2πfc t − Q(t) sin 2πfc t
=a(t) cos[2πfc t + φ(t)]
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 347

Thus for FM:

a(t) =Ac
Z t
φ(t) =2πkf m(α)dα
0
Z t
I(t) =Ac cos 2πkf m(α)dα
0
Z t
Q(t) =Ac sin 2πkf m(α)dα
0
jφ(t)
s̃(t) =Ac e
INSTRUCTOR COPY

6.4 PM and FM with Sinusoidal Message

Now consider a sinusoidal modulating wave defined by

m(t) = Am cos(2πfm t)

In the following subsections, we rederive the FM and PM time domain signal with
a sinusoidal (single tone) message.

6.4.1 FM with sinusoidal message

We re-derive the equation for the FM wave with the same three starting points.

1. The idea of FM is that the instantaneous frequency of the resulting FM wave:

fi (t) =fc + kf Am cos (2πfm t)

=fc + ∆f cos (2πfm t)

where ∆f = kf Am . Thus the message causes the instantaneous frequency to

vary above and below the carrier frequency fc , from fc − ∆f to fc + ∆f with
period 1/fm . The quantity ∆f is called the frequency deviation, since the
instantaneous frequency deviates from the carrier by that amount. fig. 6.8 is a
plot of fi (t) versus t.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 348
INSTRUCTOR COPY

Figure 6.8: instantaneous frequency fi (t)

2. A general signal is described by s(t) = <{a(t)ejθ(t) } = a(t) cos θ(t).

1 dθ(t)
3. The instantaneous frequency of a signal is fi (t) = 2π dt
.

Combining these 3 starting points, we can write:

∆f
θ(t) =2πfc t + sin(2πfm t)
fm
=2πfc t + βf sin(2πfm t)
k A
where βf = ffmm = ∆f
fm
is called the modulation index of the FM wave (remember for
AM the modulation index was ka ). Thus the FM wave itself is given in terms of β by

s(t) = Ac cos [2πfc t + βf sin(2πfm t)] .

In fig. 6.8, the vertical axis includes tickmarks for fc + fm , fc 2 + fm , fc + 3fm and
fc + ∆f , where ∆f = 1.5fm . Thus in this example figure, βf = ∆f /fm = 1.5.

If β is small we have narrowband FM (NBFM) and if β is large (compared to one

radian) we have wideband FM (WBFM).
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 349

Thus for a sinusoidal modulating wave m(t) = Am cos(2πfm t) the FM wave can
be written

s(t) =I(t) cos 2πfc t − Q(t) sin 2πfc t

=a(t) cos[2πfc t + φ(t)]

where

a(t) =Ac
φ(t) =βf sin 2πfm t
i(t) =Ac cos(βf sin 2πfm t)
q(t) =Ac sin(βf sin 2πfm t)
INSTRUCTOR COPY

The complex baseband and analytic FM signals with a single tone message are

s̃(t) = a(t)ejφ(t)
= Ac ejβf sin 2πfm t
s+ (t) = a(t)ejφ(t) ej2πfc t
= Ac ej[2πfc t+βf sin 2πfm t]

To summarize, the idea of FM is that the frequency of the FM wave varies in step
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 350

with the message

fi (t) = fc + kf m(t)
= fc + kf Am cos 2πfm t
1 dθ(t)
=
2π Z dt
t
θ(t) = 2π fi (τ )dτ
τ =0
Z t
= 2πfc t + 2πkf Am cos 2πfm τ dτ
τ =0
2πkf Am
= 2πfc t + sin 2πfm t
2πfm
INSTRUCTOR COPY

= 2πfc t + (kf Am /fm ) sin 2πfm t

= 2πfc t + βf sin 2πfm t
= 2πfc t + φ(t)
φ(t) = βf sin 2πfm t
1 dθ
fi (t) =
2π dt
= fc + fm βf cos 2πfm t
= fc + ∆f cos 2πfm t
βf = kp Am /fm
= ∆f /fm
∆f = fm βf
= kf Am

where βf = kf Am /fm is the frequency modulation index of the FM wave. Note

that for FM, the frequency deviation ∆f = kf Am does not change with modulating
frequency. We will see later that for PM , the frequency deviation ∆f = fm βp
increases with modulating frequency.

6.4.2 Narrowband FM with Sinusoidal Message

It may be difficult to implement an FM signal with a large value of ∆f using some

analog circuits such as a stable crystal oscillator Thus we consider the case β << 1,
and refer to this case as narrowband FM (NBFM). We will see in Section 5.5 how to
make an FM signal with any β starting with a NBFM signal.

When β is small compared to one radian, i.e. β << 1, the FM wave may be
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 351

approximated using cos x ≈ 1, sin x ≈ x while x << 1 to obtain

s(t) = Ac cos [2πfc t + βf sin(2πfm t)]

= Ac cos 2πfc t cos[βf sin 2πfm t] − Ac sin 2πfc t sin[βf sin 2πfm t]
' Ac cos(2πfc t) − βf Ac sin(2πfc t) sin(2πfm t)

Thus for β << 1

i(t) =Ac
(6.1)
q(t) =βAc sin 2πfm t
INSTRUCTOR COPY

Figure 6.9: Narrow-band Frequency and Phase Modulator

Exercise 107
find the complex envelope s̃(t) for a NBFM signal. Recall that in general

s̃(t) =a(t)ejφ(t)
s(t) =Re{s̃(t)ej2πfc t }
p
a(t) = i2 (t) + q 2 (t)
q(t)
φ(t) = arctan
i(t)

Exercise 108
Compare the NBFM waveform equation with the AM equation, note the simi-
larity and differences.

s(t) = Ac cos 2πfc t + Ac µ cos 2πfm t cos 2πfc t

 CHAPTER 6. ANGLE MODULATION (FM AND PM) 352

Observe that NBFM signal wave requires essentially the same transmission band-
width (i.e., 2fm ) as an AM wave. Figure 6.9 shows a block diagram for a NBFM
modulator which can be switched to a phase modulator (covered later). When β > 1
we refer to the FM signal as a wideband FM (WBFM) signal.

6.4.3 PM with sinusoidal message

The idea of PM is that the phase of the PM wave varies in step with the message

φ(t) = kp m(t)
= kp Am cos 2πfm t
= βp cos 2πfm t
INSTRUCTOR COPY

θ(t) = 2πfc t + φ(t)

= 2πfc t + βp cos 2πfm t
1 dθ
fi (t) =
2π dt
= fc − fm βp sin 2πfm t
= fc − ∆f sin 2πfm t
βp = kp Am
∆f = fm βp
= kp A m f m

where βp = kp Am is the phase modulation index of the PM wave. Note that for PM,
the frequency deviation ∆f = fm βp increases with modulating frequency.

The PM signals are

s̃(t) = a(t)ejφ(t)
= Ac ejβp cos 2πfm t
s+ (t) = a(t)ejθ(t)
= Ac ej[2πfc t+βp cos 2πfm t]
s(t) = a(t) cos θ(t)
= Ac cos[2πfc t + βp cos 2πfm t]

We will use these expressions to obtain the spectrum of PM and FM signals in

the next section section 6.5.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 353

6.4.4 Comparison of FM and PM with sinusoidal message

FM
s̃(t) = Ac ejφ(t)
s(t) = Ac cos 2πfc t + φ(t)
φ(t) = βf sin 2πfm t
βf = kp Am /fm
= ∆f /fm
∆f = fm βf
= kf Am
fi (t) = fc + ∆f cos 2πfm t)
INSTRUCTOR COPY

PM
s̃(t) = Ac ejφ(t)
s(t) = Ac cos 2πfc t + φ(t)
φ(t) = βp cos 2πfm t
βp = kp Am
∆f = fm βp
= kp Am fm
fi (t) = fc − ∆f sin 2πfm t
Note that for PM, the frequency deviation ∆f = fm βp = kp Am fm increases with
modulating frequency, whereas for FM, ∆f = kf Am does not change with modulating
frequency.

6.5 Power Spectrum of an FM Signal – Bessel func-

tions

Consider the FM signal with a single tone message m(t) = Am cos(2πfm t) so that
s̃(t) =Ac ejβ sin 2πfm t
β = βf = kf Am /fm

For a PM signal
s̃(t) = Ac ejβp cos 2πfm t
βp = kp Am
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 354

We have retained the notation β = βf for the FM modulation index and use βp
for the PM modulation index.

6.5.1 Bessel function definition

In what follows, we will evaluate the FM complex envelope and find that it is given
in term of a special function called a Bessel function. Once we have defined this
function, we continue with finding the FM power spectrum and bandwidth.

The FM complex envelope s̃(t) = a(t)ejφ(t) = Ac ejβ sin 2πfm t is periodic with period
1/fm so we can write the complex envelope as a complex Fourier series with index n.
INSTRUCTOR COPY

∞
X
jβ sin 2πfm t
Ac e = Ac cn ej2πnfm t (6.2)
n=−∞

Similarly, the PM complex envelope s̃(t) = a(t)ejφ(t) = Ac ejβp cos 2πfm t is periodic
with period 1/fm . so we can write the complex envelope as a complex Fourier series
with index n.

Ac ejβp cos 2πfm t = Ac ejβp sin(2πfm t+π/2)

X∞
= Ac cn ej2πnfm t
n=−∞

The Fourier coefficients are given by the integral

Z 1/fm
cn = f m ejβ sin 2πfm t e−j2πnfm t dt
t=0
Z 1/fm
= fm ej(β sin 2πfm t−2πnfm t) dt
t=0

By a change of variable u = 2πfm t for FM or u = 2πfm t + π/2 for PM we can

write
Z 2π
1
cn = ej(β sin u−nu) du
2π u=0
∆
= Jn (β)
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 355

This result applies to both FM and PM, so β may represent either βf or βp .

This integral depends on two parameters n, β, but cannot be evaluated in terms

of elementary functions, so it is given a name Jn (β) called the Bessel function of the
first kind and order n.

The function Jn (β), the Bessel function of the first kind of order n, is plotted
in fig. 6.10 for values of n from 0 to 4. To make this plot, the Matlab function
besselj(n,beta) is used.
INSTRUCTOR COPY

Figure 6.10: First kind Bessel functions for the first 5 orders (n).

Jn (β) looks like a damped cosine wave for n = 0 and damped sine waves for n ≥ 1.
The Bessel function has many properties and identities (just like cos and sin):

Jn (β) =(−1)n J−n (β)

∞
X
Jn2 (β) =1
n=−∞

J0 (β) '1 for small β ≤ 0.5

β
J1 (β) ' for small β < 1
2
Jn (β) '0 for small β < 1, n > 1 or β ≥ 1, n  1

6.5.2 PM and FM spectrum with single tone message

Now that we have introduced the Bessel function, we continue to find the PM and
FM power spectrum and bandwidth.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 356

6.5.2.1 FM complex envelope spectrum

The FM complex envelope can be written

s̃(t) =a(t)ejφ(t)
=Ac ejβ sin 2πfm t
∞ (6.3)
X
=Ac Jn (β)ej2πnfm t
n=−∞

∞
X
S̃(f ) = Ac Jn (β)δ(f − nfm )
n=−∞
INSTRUCTOR COPY

since

Jn (β) =(−1)n J−n (β)

the magnitude spectrum is symmetric about f = 0

|Jn (β)δ(f − nfm )| = |J−n (β)δ(f + nfm )|

|S̃(f )| = |S̃(−f )|

Exercise 109
Is the phase spectrum anti-symmetric about f = 0 or not?

The magnitude of the FM complex baseband spectrum is shown in fig. 6.11.

 CHAPTER 6. ANGLE MODULATION (FM AND PM) 357

(a) Bessel Function value (Jn (β)) defining the FM spec-

trum depends upon β.
INSTRUCTOR COPY

(b) Example of FM spectrum with relatively large β.

Figure 6.11: Spectra showing how the FM/PM spectrum changes with the value of β

For the NBFM case β << 1, the sidebands for n ≥ 2 are small and can be
neglected. Thus NBFM has only one pair of sidebands.

6.5.2.2 FM analytic spectrum

The analytic spectrum

S+ (f ) = S̃(f − fc )
X∞
= Ac Jn (β)δ(f − nfm − fc )
n=−∞
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 358

S− (f ) = S̃(−(f + fc ))
= S̃(−f − fc )
∞
X
= Ac Jn (β)δ(−f − nfm − fc )
n=−∞
X∞
= Ac Jn (β)δ(+f + nfm + fc )
n=−∞
INSTRUCTOR COPY

6.5.2.3 FM real passband

The real passband signal is obtained from the fundamental result

2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))

An alternate approach is to write the FM real passband time domain signal

s(t) =<{a(t)ejφ(t) ej2πfc t }

=<{Ac eβ sin 2πfm t ej2πfc t }
∞ (6.4)
X
=<{Ac [ Jn (β)ej2πnfm t ]ej2πfc t }
n=−∞

and take the Fourier transform.

Exercise 110
Show that with both approaches, the discrete spectrum of the FM wave is
obtained as
∞
Ac X
S(f ) = Jn (β) [δ(f − fc − nfm ) + δ(f + fc + nfm )] (6.5)
2 n=−∞
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 359

6.5.2.4 Observation interval

As we did for the case of AM and DSB-SC, we may ask why it is that for a sinusoidal
message we see discrete sidebands at fc ± nfm , since the FM instantaneous frequency
given by

fi (t) =fc + ∆f cos(2πfm t)

=fc + βf fm cos(2πfm t)

can be any continuous value and is not restricted to a discrete set of values.

The same reasoning applies to PM where the PM instantaneous frequency given

by
INSTRUCTOR COPY

fi (t) =fc − ∆f sin(2πfm t)

=fc − βp fm sin(2πfm t)

If we were to view the frequency domain representation of the FM or PM signal and

the observation (time) interval is much less than the message period 1/fm , the complex
baseband spectrum S̃(f ) consists of one δ− function at the instantaneous frequency
fi (t) that moves around in frequency in a continuous fashion as time progresses.

If the observation interval is greater than 1/fm , then we have averaged the Fourier
transform calculation over at least one full cycle of the frequency variation

−∆f ≤ fi (t) ≤ +∆f

then the complex baseband spectrum is discrete with sidebands spaced at intervals
of fm . The derivation of the FM spectrum above uses the complex Fourier series and
thus assumes an observation interval of one message period 1/fm .

6.5.3 FM spectrum properties

We make the following observations about the FM real passband spectrum as observed
over an interval greater than the message period 1/fm . The same observations apply
to the complex baseband spectrum.

1. the spectrum includes components at frequencies fc ± nfm that are integer

multiples of the message frequency above and below the carrier frequency.

2. The number of significant peaks increases as β increases.

 CHAPTER 6. ANGLE MODULATION (FM AND PM) 360

3. As β increases, more and more peaks become significant, but some peaks become
smaller, all in accordance with the Bessel function curve.

4. Thus the shape (‘envelope’) of the spectrum depends on β in a complicated way.

5. It is not obvious from the spectrum plot, but for any value of β, and thus for
any spectrum shape, the powers in all of the peaks sums to 1. This is because
the average power of an FM wave developed across a 1 ohm resistor is given by
∞
A2c X 2
P = · J (β)
2 n=−∞ n
A2c
=
2
INSTRUCTOR COPY

independent of the values of β, since all the Bessel function power spikes in the
power spectrum add up to 1.

6. For FM with large β, the FM spectrum bandwidth is much greater than the
message bandwidth, as we will see later in section 6.5.3.3 on the Carson’s effec-
tive bandwidth rule.

7. FM signals with large β have many sidebands for a single tone message. Thus
the information in that single tone is represented many times and thus the
many sidebands are redundant. As a result, the FM signal is robust and can be
received even in the presence of interference. A more detailed study can show
that an FM signal can be demodulated provided its signal to noise ratio (S/N)
is above the a threshold. If there are two FM signals on the same frequency at
the same times, then if one of the signals is only a few dB stronger than the
other (the difference in signal strength is greater than the“capture threshold”),
then only the stronger of the two signals will be demodulated.

8. Because FM has many redundant sidebands, FM may be viewed as a ”poor

man’s spread spectrum” where the definition of spread spectrum is a modu-
lation scheme where the signal bandwidth is much greater than the message
bandwidth.

All of above applies to both PM and FM.

The FM/PM spectrum depends upon the modulation index β as mentioned above.

For FM

∆f = βf fm = kf Am
βf = kf Am /fm = ∆f /fm
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 361

For PM

∆f = βp fm = kp Am fm
βp = kp Am = ∆f /fm

Thus

• the FM spectrum magnitude peaks Jn (β) are a function of both kf and fm , in

particular, the ratio kf /fm .

• the PM spectrum magnitude peaks Jn (β) are a function of kp only independent

of fm .
INSTRUCTOR COPY

6.5.3.1 Wideband FM large modulation index

For WBFM, β is large:

∆f
β= = 10
fm
∆f =10fm
fmax =fc + ∆f
fmin = fc − ∆f
∆f fm
β 1

The spectrum for WBFM is shown in fig. 6.12.

Figure 6.12: WBFM spectrum

 CHAPTER 6. ANGLE MODULATION (FM AND PM) 362

6.5.3.2 Narrowband FM, small modulation index

For NBFM, β is small:

∆f
β= = 0.1
fm
∆f =0.1fm
f i (t) =fc + kf m (t)
=fc + ∆f cos(2πfm t)
∆f fm
β 1
INSTRUCTOR COPY

The spectrum for NBFM is shown in fig. 6.13.

Figure 6.13: NBFM spectrum

6.5.3.3 Effective bandwidth of FM – Carson’s rule

For practical purposes, the bandwidth of the FM wave corresponds to the bandwidth
containing 98% of the signal power. The effective bandwidth of the FM signal is
approximately given by Carson’s formula:

B =2(1 + β)fm
=2(1 + kf Am /fm )fm FM
=2(1 + kp Am )fm PM
=2∆f + 2fm (6.6)
=2(∆f + fm )
=2(kf Am + fm ) FM
=2(kp Am fm + fm ) PM
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 363

Carson’s formula is intuitively reasonable: the bandwidth must be at least twice

the frequency deviation since the instantaneous frequency changes from fc − ∆f to
fc + ∆f . The bandwidth must also be twice the modulating frequency.

Observe that

• for FM, B = 2(kf Am + fm ) depends on kf and fm in an additive relationship,

• for PM, B = 2(kf Am + 1)fm depends kp and fm in a multiplicative relationship.

We make a similar observation from a slightly different perspective. Comparing the

bandwidth of PM and FM versus modulating frequency fm
B =2(1 + β)fm
INSTRUCTOR COPY

=2(∆f + fm ) FM
=2(kp Am + 1)fm PM
observe that

• increasing fm increases the bandwidth more for PM than for FM,

• The PM bandwidth increases linearly with fm , the same as for AM,
• for ∆f  fm , the FM bandwidth is mostly determined by ∆f .

6.5.3.4 FM/PM estimated bandwidth for a general message

P
For a general message m(t) = i Ai (t) cos(2πfi t + ψi (t)) containing many cos waves,
we can estimate the FM bandwidth by considering the highest frequency (i.e. the
message bandwidth W ) and highest amplitude A in the message. For this case, we
replace fm → W and Am → A and find
B =2(kf Am + fm )
B =2kf A + 2W
=2(D + 1)W

By comparison, we see that

DW =kf A
kf A
D=
W
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 364

fk A
where the deviation ratio D = W has the same role as the modulation index β =
k f Am
fm
in determining the bandwidth of an FM signal.

The key point is that, unlike AM, the WBFM signal bandwidth may be much
greater than the message bandwidth.

Exercise 111
Estimate the PM bandwidth for a general message.

6.5.4 FM spectrum for dual-tone message

INSTRUCTOR COPY

For a single tone message we wrote the FM complex envelope

s̃(t) =a(t)ejφ(t)
=Ac ejβ sin 2πfm t
∞ (6.7)
X
=Ac Jn (β)ej2πnfm t
n=−∞

For a dual tone message,

s̃(t) =a(t)ejφ(t)
=Ac ej[β1 sin 2πf1 t+β2 sin 2πf2 t]
X ∞ X∞
=Ac Jn (β1 )Jm (β2 )ej2π(nf1 +mf2 )t
n=−∞ m=−∞

This expression contains 4 types of frequency terms:

• A DC component with amplitude Ac J0 (β1 )J0 (β2 )

• a set of sidebands corresponding to ±nf1 with amplitudes Ac Jn (β1 )J0 (β2 )

• a set of sidebands corresponding to ±mf2 with amplitudes Ac J0 (β1 )Jm (β2 )

• a set of cross-modulation terms at frequencies ±nf1 ± mf2 with amplitudes

Ac Jn (β1 )Jm (β2 )
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 365

The cross-modulation terms appear because FM in a non-linear modulation. By

comparison, AM is a linear modulation, so that the cross-modulation terms do not
appear.

The cross modulation terms for FM can be distinguished from the sidebands for
f1 and f2 more clearly by choosing f2 = f1 /(2β). In this case, the sidebands at ±nf1
are each surrounded by more closely spaced sidebands at ±nf1 ± mf2 . However, the
above expression works for any values of f1 and f2 .

6.6 Frequency and phase modulators (transmit-

ters)
INSTRUCTOR COPY

6.6.1 Complex passband implementation

To generate a PM or FM complex passband signal in discrete-time, we simply imple-

ment the relevant equations in software.

For PM,
s+ (t) = s̃(t)ej2πfc t
= a(t)ejφ(t) ej2πfc t
= Ac ejφ0 ejkp m(t) ej2πfc t

For FM,
s+ (t) = a(t)ejφ(t) ej2πfc t
Rt
= Ac ej2πkf τ =0 m(τ )dτ j2πfc t
e

6.6.2 Real passband implementation

6.6.2.1 Digital (discrete-time) method

A digital method of building an FM transmitter simply implements the real passband

FM equation in software.

Another approach is to implement the complex passband equatios (for example

using GNURadio) to generate i and q components, and then the real passband signal
is created from the i and q components by the analog IQ mixer in fig. 2.19.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 366

Exercise 112
Draw a block diagram of an FM transmitter implemented digitally in software.

6.6.2.2 Analog (continuous-time) method

A common analog method is so-called indirect FM modulation. The process starts

with a narrowband FM modulator as shown in fig. 6.9, that implements

s(t) ' Ac cos(2πfc t) − βAc sin(2πfc t) sin(2πfm t) (6.8)

INSTRUCTOR COPY

The next step is to multiply the signal by itself several times, thus increasing both
the carrier frequency and the modulation index, as shown in fig. 6.14.

Figure 6.14: NBFM transmitter

Figure 6.14 is described by

s(t) =Ac cos(2πfc t + β sin 2πfm t)

s2 (t) =Ac 2 cos2 (2πfc t + β sin 2πfm t)
(6.9)
=DC + A2c cos(4πfc t + 2β sin 2πfm t)
s4 (t) =DC + A4c cos(8πfc t + 4β sin 2πfm t)

We observe that both the frequency deviation ∆f and modulation index β =

∆f /fm increase when the signal is multiplied by itself. In practice the multiplication
may be many times, as illustrated in fig. 6.15.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 367
INSTRUCTOR COPY

Figure 6.15: FM modulation using two frequency multiplier chain.

A down-conversion stage may be required if the desired carrier frequency is low

and the desired frequency deviation (∆f ) is large.

6.7 Frequency Demodulation

Frequency demodulation extracts the original message wave from the frequency-
modulated wave. We consider four intuitive approaches that are suitable for analog
or digital implementation.

• Differentiation

• Linear Amplitude vs. Frequency Characteristic

• Zero-crossing counter

• phase-locked loop
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 368

6.7.1 Intuition 1: Differentiation

We know that from the idea of FM

fi (t) =fc + kf m(t)

1 dφ(t)
=fc +
2π dt
so that
1 dφ(t)
m(t) = .
2πkf dt
Since s(t) = a(t)cos[2πfc t + φ(t)] with a(t) = Ac (constant) we see that if we differ-
entiate s(t) we will get a term in ds(t)
dt
that looks like dφ(t)
dt
.
INSTRUCTOR COPY

A demodulator based on this intuition is in the figure below.

s(t) → LIMITER → BPF →

DIFFERENTIATOR (a.k.a DISCRIMINATOR) →
ENVELOPE DETECTOR → DC BLOCK → m(t)

We differentiate s(t) as a first step to extract the message m(t). Assuming Ac is

constant we have
 Z t 
s(t) =Ac cos 2πfc t + 2πkf m(α)dα
0
 Z t 
ds(t)
= − Ac [2πfc + 2πkf m(t)] sin 2πfc t + 2πkf m(t)dt
dt 0

This expression is in the form of an AM signal

s(t) =a(t) cos[2πfc t + φ(t)]

=2πfc Ac [1 + ka m(t)] cos[2πfc t + φ(t)]

with envelope  
kf
|s(t)| = 2πAc fc 1 + m(t)
fc
Rt
where ka = kf /fc and phase 2πkf 0
m(α)dα − π/2.

The resulting AM signal can be demodulated by an envelope detector to obtain

DC plus the message. The envelope detector ignores the phase.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 369

Note that if the FM signal has a(t) 6= Ac is not constant (e.g. due to channel
fading or noise), then differentiation will not work, since for this case
 Z t 
ds(t)
= − a(t) [2πfc + 2πkf m(t)] sin 2πfc t + 2πkf m(t)dt
dt 0
 Z t 
da(t)
+ cos 2πfc t + 2πkf m(α)dα
dt 0

To make a(t) = Ac constant, we can use a limiter (e.g. with back to back diodes)
ahead of the discriminator. The limiter limits (clips) the input signal so that it is of
constant amplitude. The output of the limiter looks like a square wave with changing
frequency, and will contain harmonics at odd multiples of fc . The limiter must be
followed by a bandpass filter at fc to restore the signal.
INSTRUCTOR COPY

 Z t 
s(t) = Ac cos 2πfc t + 2πkf m(α)dα
0

6.7.2 Intuition 2: Linear Amplitude vs. Frequency Charac-

teristic

We know that from the idea of FM fi (t) = fc + kf m(t). If we have a circuit that
has output amplitude that increases linearly with frequency, then the circuit output
amplitude will vary in step with the message.

One such circuit is the ideal differentiator, with a transfer function given by
H(f ) = j2πf .

Proof: If s(t) → S(f ) then ds(t)/dt → j2πf S(f ).

The transfer function acts as a frequency to voltage converter, and is illustrated

in fig. 6.16, centered at the carrier frequency. The slope is such that the transfer
function changes by j2πBT over a bandwidth BT centered at fc
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 370

Figure 6.16: Frequency vs transfer function

The action of an ideal differentiator fig. 6.16 and (fig. 6.17(a)) can be approximated
INSTRUCTOR COPY

by any device whose magnitude transfer function is reasonably linear, within the range
of frequencies of interest.

In fig. 6.17(b) an RL circuit approximation to a differentiator is used followed by

an envelope detector. The RL circuit is a high pass filter.

A bandpass version of this circuit is shown in fig. 6.17(c). These discriminators

are known as slope detectors. A more linear response can be obtained by taking the
difference between two bandpass magnitude responses, as is done by the balanced
discriminator shown in fig. 6.17(d).

In all cases fig. 6.17(a)-(d), a limiter and bandpass filter is required ahead of the
discriminator.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 371
INSTRUCTOR COPY

Figure 6.17: FM detectors

6.7.3 Intuition 3: Zero Crossing Counter

The message information is contained in the time (location) of the zero crossings, and
the amplitude can be ignored, as shown in fig. 6.18.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 372
INSTRUCTOR COPY

Figure 6.18: Zero crossing FM demodulator

6.7.4 Intuition 4: Phase Locked Loop

The circuit in fig. 6.19 uses ideas from the Costas loop receiver used for DSB. The
circuit comprises a phase comparator (multiplier), low pass (loop) filter and voltage
controlled oscillator (equivalent to an FM modulator).

Figure 6.19: Phase-locked loop

With signal input s(t) = cos[2πfc t + θi (t)] and FM modulator (VCO) output
sin[2πfc t + θf (t)],

• the phase comparator (multiplier) output signal e(t) = kc [θi (t) − θf (t)] plus
double frequency terms.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 373

• With high gain in the loop filter, e(t) ≈ 0 and the VCO output frequency is the
same as the input signal frequency.

• Thus the VCO input must be the same as the message.

• The loop filter output voltage is proportional to the instantaneous frequency of

the input, and FM demodulation is achieved.

6.8 Digital FM Demodulator

A digital FM demodulator starts with the I and Q outputs of a general IQ receiver

such as shown in fig. 2.24 or fig. 4.12.
INSTRUCTOR COPY

Recall for an FM signal

 Z t 
s(t) =Ac cos 2πfc t + 2πkf m(α)dα
0
Z t
I(t) =Ac cos 2πkf m(α)dα
0
Z t
Q(t) =Ac sin 2πkf m(α)dα
0

To extract m(t) from I(t), Q(t) we show two methods that can be implemented
in software.

6.8.1 I(t), Q(t) as Real Signals

Formula:
d Q(t)
m(t) = arctan{ }
dt I(t)

Block diagram:

I(t), Q(t) → DIV IDE → ARCT AN → d/dt → m(t)

 CHAPTER 6. ANGLE MODULATION (FM AND PM) 374

Proof :
Rt
d Q(t) d Ac sin 2πkf 0 m(α)dα
arctan{ } = arctan Rt
dt I(t) dt Ac cos 2πkf 0 m(α)dα
Z t
d
= arctan{tan 2πkf m(α)dα}
dt 0
Z t
d
= 2πkf m(α)dα = 2πkf m(t)
dt 0

This method is not good in practice because when I(t) is small, the division by a
small number will cause numerical problems.
INSTRUCTOR COPY

6.8.2 I(t) + jQ(t) as a Complex Signal

6.8.2.1 Method 1

in this method we implement the differentiator needed in an FM demodulator as a

discrete-time differentiation

s(t) =Re{a(t)ejφ(t) ej2πfc t }

=Re{[I(t) + jQ(t)]ej2πfc t }
=Re{s̃(t)ej2πfc t }

Formula:
m(t) = arg[s̃(t − 1)s̃∗ (t)]
where
(t − 1) → z −1
represents a one sample delay.

Block diagram:

s̃(t) → s̃(t − 1), s̃∗ (t) → M U LT IP LY → ARG

A GNU Radio implementation of this is shown in fig. 6.20.

 CHAPTER 6. ANGLE MODULATION (FM AND PM) 375

Figure 6.20: GNU Radio FM receiver flowgraph using the delay and complex conju-
gate demodulation method.

Proof :

arg[s̃(t − 1)s̃∗ (t)] = arg[a(t − 1)ejφ(t−1) a(t)e−jφ(t) ]

dφ
=φ(t − 1) − φ(t) ≈ −
dt
INSTRUCTOR COPY

≈ − 2πkf m(t)

This method of FM demodulation is commonly used in software defined radios.

For a derivation of φ(t) − φ(t − 1) ≈ dφdt
see appendix C. The negative sign will be
discussed after the alternate formula is proven.

Alternate formula:
m(t) = arg[s̃(t)s̃∗ (t − 1)]

Proof :

arg[s̃(t)s̃∗ (t − 1)] = arg[a(t)ejφ(t) a(t − 1)e−jφ(t−1) ]

dφ
=φ(t) − φ(t − 1) ≈
dt
≈2πkf m(t)

The two methods above give the same message output but with different signs
and opposite polarities. In practice this does not matter for a voice signal but the
bits will be flipped for a digital frequency shift keying (FSK) signal, cf. chapter 9. If
the bits are differentially encoded (section 7.3) the different signs won’t matter. The
takeaway is that both methods deliver the message.

6.8.2.2 Method 2

In this section, we show an alterative method of discrete-time differentiation of the

arctan function.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 376

From basic calculus,

d 1 dx(t)
arctan{x(t)} =
dt 1 + x2 dt
q(t)
x(t) =
i(t)
dx dq(t) di(t)
= i(t) − q(t)
dt dt dt

For a discrete-time implementation, the derivatives of i(t) and q(t) are replaced
by discrete differences

Thus the receiver block diagram

INSTRUCTOR COPY

Figure 6.21: FN receiver based on derivative of arctan[q(t)/i(t)]

implements the following equations to obtain the receiver output ∆θ[n]

Inputs to top multiplier:

i[n − 1] · (q[n] − q[n − 2])
Input to bottom multiplier:
(i[n] − i[n − 2]) · q[n − 1]
subtract the outputs before scaling
∆θ[n] = i[n − 1] · (q[n] − q[n − 2]) − (i[n] − i[n − 2]) · q[n − 1]
dq(t) di(t)
' i(t) − q(t)
dt dt

This difference equation represents the chain rule part of the arctan derivative
presented above.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 377

6.8.3 FM demodulator with frequency offset

In the event of a frequency offset, the complex baseband signal will include a rotating
exponential representing the frequency offset
s̃(t) = a(t)ejφ(t) ej(2πfb t+ψ)
In this case,
arg[s̃(t)s̃∗ (t − 1)] = arg[a(t)ejφ(t) ej(2πfb t+ψ) a(t − 1)e−jφ(t−1) e−j(2πfb (t−1)+ψ) ]
=φ(t) − φ(t − 1) + 2πfb t − 2πfb (t − 1)
dφ
≈ + 2πfb
dt
=2π(kf m(t) + fb )
INSTRUCTOR COPY

Thus the frequency offset results in a DC offset added to (or subtracted from) the
message. If the frequency offset is large, then the DC offset may be larger than the
peak value of the message. Since the message is represented digitally, the DC offset
may approach or exceed the maximum (or minimum) value that can be represented
in which case the message may be clipped.

6.9 Pre-emphasis and de-emphasis in FM

6.9.1 History and motivation for FM pre-emphasis and de-

emphasis

In the days when all communications circuits were analog, a frequency modulator that
was easily mass-produced and stable was not developed with the components available
at the time. Thus early FM broadcast was transmitted using phase modulation, often
implemented using a so-called reactance modulator that adds an RC phase shifting
network in parallel with the resonant circuit of the oscillator.

Varactor diodes that became available later on can be used to control high-Q
crystal oscillator circuits that generate a stable carrier frequency and thus made
direct frequency modulation practical.

A detailed study of these analog circuits is outside the scope of this section.

Some useful background is at http://www.repeater-builder.com/tech-info/

fm-theory-discussion.html in particular the first 6 paragraphs starting with ”the
goods”.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 378

An inherent characteristic of a phase modulator is that the frequency deviation

∆f increases linearly with the modulating frequency as shown below

s(t) = Ac cos(2πfc t + φ(t))

= Ac cos(2πfc t + kp m(t))
m(t) = Am cos 2πfm t
s(t) = Ac cos(2πfc t + kp Am cos 2πfm t)
= Ac cos θ(t)
θ(t) = 2πfc t + kp Am cos 2πfm t
1 dθ(t)
fi (t) =
2π dt
1
= (2πfc − 2πfm kp Am sin 2πfm t)
INSTRUCTOR COPY

2π
= fc − kp Am fm sin 2πfm t
= fc + ∆f sin 2πfm t
∆f = −kp Am fm

The output of an FM receiver is fi (t), since the FM receiver detects the instan-
taneous frequency. Thus the FM receiver output is proportional to ∆f (ignoring any
DC offset due to frequency offset). If this FM receiver is used to receive the above
PM signal with ∆f = −kp Am fm , then the higher frequencies will be ”emphasized”.
To preserve the message m(t) and its spectrum M (f ) in its original form without this
emphasis of higher frequencies, a de-emphasis filter is needed to restore the audio to
a flat response. An additional benefit of the de-emphasis filter is that it reduces the
noise which is most often high frequency noise (hiss).

When true frequency modulators became available and were used instead of phase
modulators, a pre-emphasis circuit was needed to replicate the phase modulator char-
acteristic. The pre-emphasis circuit is a high pass filter that creates the same ”em-
phasis” that is created by a phase modulator.

A pre-emphasis (high pass) filter followed by a frequency modulator produces the

same output s(t) (and same spectrum) as a phase modulator.

m(t) -> PM modulator -> s(t)

m(t) -> pre-emph HPF -> FM modulator -> s(t)

The frequency modulator integrates the phase (message waveform). A pre-emphasis

high pass filter is like a differentiator that cancels out the integration, resulting in
phase modulation.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 379

At the receiver, the de-emphasis circuit is a low pass filter which is like an inte-
grator.

A frequency demodulator followed by a de-emphasis circuit produces the same

output m(t) as a phase demodulator.

s(t) -> PM demodulator -> m(t)

s(t) - FM demodulator -> de-emph LPF -> m(t)

Thus it may appear that we can simply use PM for both transmitter and receiver
and avoid pre-emphasis and de-emphasis circuits altogether.

However, for broadcast FM. the pre-emphasis circuit shown in the next section
INSTRUCTOR COPY

became standard practice. A similar circuit with different parameters is used for land
mobile 2-way radio FM. Thus FM continues to be used along with pre-emphasis and
de-emphasis circuits.

6.9.2 Pre- and de-emphasis analog circuits

What follows is another way to look at pre-emphasis and de-emphasis.

All versions of the FM receiver demodulator are essentially a differentiator or high

pass filter whose output will increase linearly with the input frequency. Thus for a
message consisting of two cos waves at f1 and f2 > f1 of equal amplitude that is
phase modulated onto a carrier, then at the FM demodulator output, the wave at f2
will have a larger amplitude than the wave at f1 .

For f2 = 2f1 , the amplitude of f2 will be twice the amplitude at f1 , and the power
will be 4 times higher (6 dB). The noise (with components at all frequencies) will also
have a high pass characteristic with larger amplitude at higher frequencies.

If an FM transmitter is used instead of a PM transmitter, then, to compensate

for this FM receiver effect, the FM transmitter first filters the message with a pre-
emphasis filter to also increase the amplitude of the message at high frequencies before
it is frequency modulated and transmitted. Thus both the message signal and the
noise will have increasing amplitude at higher frequencies. The pre-emphasis filter is
a single pole filter with 6 dB per octave (or 20 dB per decade) slope.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 380
INSTRUCTOR COPY

Figure 6.22: Pre-emphasis circuit and curve

The FM receiver contains a de-emphasis filter to decrease the amplitude at high

frequencies so that the message is reconstructed correctly. For a message with cos
waves at f1 and f2 > f1 of equal amplitude, then at the FM demodulator output
after the de-emphasis filter, the wave at f2 will have the same amplitude as the wave
at f1 .
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 381
INSTRUCTOR COPY

Figure 6.23: De-emphasis circuit and curve

The de-emphasis filter can be implemented with an analog RC circuit as shown

in fig. 6.23.

For Broadcast FM in North America, the RC time constant is set to 75 µs, corre-
1 1
sponding to a cutoff frequency of 2πRC = 2π75·10−6 = 2122 Hz.

The impulse response of the RC circuit is h(t) = e−t/RC so that the signal decays
to 1/e = 36.8% of its initial value in time t = RC For FM land mobile radio used by
public safety, forestry, marine, Amateur radio, the cutoff frequency is set to 300 Hz
which is the lower limit of the voice message bandwidth.

Combining the ideas in the previous section with the ideas in this section, note that
for message frequencies below the cutoff frequency, there is no pre- or de-emphasis,
and thus the transmitter/receiver is FM. Above the cutoff frequency, because of the
pre- or de-emphasis, the FM transmitter/receiver effectively becomes PM.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 382

6.9.3 De-emphasis with digital filter

6.9.3.1 Implementation

The FM de-emphasis filter can be implemented digitally with a single pole IIR filter
with difference equation y[n] − (1 − α)y[n − 1] = αx[n] or yn = (1 − α)yn−1 + αxn
with transfer function derived as follows:

Y (z) − z −1 (1 − α)Y (z) =αX(z)

Y (z)
H(z) =
X(z)
α
=
1 − (1 − α)z −1
INSTRUCTOR COPY

αz
=
z − (1 − α)

This filter has a single pole at 1 − α and a single zero at 0. The impulse response
may be found by inverse z-transform or iteration to find

h[n] = α(1 − α)n u[n]

This filter is implemented in GNU Radio as a single pole IIR filter block with
parameter α, see http://gnuradio.org/doc/sphinx-3.7.0/filter/filter_blk.
html#gnuradio.filter.single_pole_iir_filter_cc.

The Matlab representation is y = filter(b, a, x) for a general filter, where we define

the vectors a = [1 a1 a2 ], b = [b0 b1 b2 ] (see www.mathworks.com/help/matlab/ref/
filter.html) for the filter with difference equation

y[n] + a1 y[n − 1] + a2 y[n − 2] = b0 x[n] + b1 x[n − 1] + b2 x[n − 2]

and transfer function

b0 + b1 z −1 + b2 z −2
H(z) = .
1 + a1 z −1 + a2 z −2

For the filter considered here, y[n] − (1 − α)y[n − 1] = αx[n]. Thus a = [a1 a2 ] =
[1 − (1 − α)] and b = [b0 ] = [α].

The filter characteristics (including frequency response, phase response, phase

delay, group delay, pole-zero plot, impulse response) can be viewed using the Matlab
command fvtool(b,a) (see www.mathworks.com/help/signal/ref/fvtool.html).
For example, if α = 0.1 then set the values in fvtool to be b = 0.1 and a = [1 − 0.9].
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 383

6.9.3.2 Impulse response

The desired impulse response for the FM de-emphasis filter is an exponential decay
with time constant τ = 75 µs

The impulse response of the digital filter is h[n] = α(1 − α)n u[n] and thus is an
exponential decay similar to that of the analog RC filter.

The signal decays to 1/e = 36.8% of its initial value in d samples such that

h[n = d] hd
=
h[n = 0] h0
α(1 − α)d
=
INSTRUCTOR COPY

α
−1
e =(1 − α)d
e−1/d =1 − α

Given the sampling rate of the filter fs and the desired time constant t = RC we
set the number of samples d needed for the filter output to decay to 1/e = 36.8% of
its initial value during the time from t = 0 → RC to be d/fs = RC or d = RCfs .

For example, if we assume a sampling rate of 250 kHz and time constant of 75 µs,
we set

d =RCfs
=75 · 10−6 · 0.25 · 106
=19 samples

The value of α is found via 1−α = e−1/d . In this example 1−α = e−1/19 = 0.9487.

6.9.3.3 Frequency response

The desired frequency response for the FM de-emphasis filter is a low pass filter with
3dB cutoff frequency 2122 Hz as shown in fig. 6.23.

An expression for the frequency response is obtained by choosing z = ej2πf /fs

 CHAPTER 6. ANGLE MODULATION (FM AND PM) 384
INSTRUCTOR COPY

Figure 6.24: Impulse response

around the unit circle in the z-plane, so that

H(f ) =H(z)|z=ej2πf /fs

α
= | j2πf /fs
1 − (1 − α)z −1 z=e
α
=
1 − (1 − α)e−j2πf /fs
1
= −1
α − (α − 1)e−j2πf /fs
−1

Figure 6.25: Magnitude of the frequency response

 CHAPTER 6. ANGLE MODULATION (FM AND PM) 385

The gain at f = 0 of z = 1 is
α
H(z = 1) =
1 − (1 − α)
α
=
α
=1
The gain at f = fs /2 or z = ejπ = −1 is
α
H(z = −1) =
1 + (1 − α)
α
=
2+α
1
=
INSTRUCTOR COPY

1 + 2/α

The gain is reduced by 3 dB at a cutoff frequency fC (notice the difference between

carrier frequency, fc , and cutoff frequency, fC ) such that 1/d = 2πfC /fs or
(1 − α) =e−1/d
=e−2πfC /fs

The gain at fC such that (1 − α) = e−2πfC /fs is

α
H(f = fC ) =
1 − (1 − α)e−j2πf /fs
1 − e−2πfC /fs
=
1 − e−2πfC /fs e−j2πfC /fs

Here fC /fs is normalized to the sampling rate so that 0 ≤ fC /fs ≤ 0.5.

1 1
Recall for the analog filter the cutoff frequency 2πRC = 2π75·10−6 = 2122 Hz. We

want the gain for the digital filter at 2122 Hz to be 3 dB down. In this case, assuming
a sampling rate of 250 kHz, the normalized frequency is ffCs = 250000
2122
= 0.00848.

Substituting these values into the 3 dB cutoff frequency result (1 − α) = e−2πfC /fs ,
we find
e−2πfC /fs =e−2π(0.00848)
=0.9481
=(1 − α)
which is consistent with the value of α found via 1 − α = e−1/d above, and also
consistent with the frequency (magnitude) response plot (fig. 6.25).
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 386

6.9.3.4 Summary of digital de-emphasis filter parameters

We have demonstrated that the time constant and cutoff frequency of a single pole
digital filter are consistent with the desired values for a de-emphasis filter.

Thus a digital de-emphasis filter with time constant 75 µs, cutoff frequency 2122 Hz
and sampling rate fs = 250 kHz may be built using a single pole IIR filter with
parameter α = 1 − 0.948 = 0.052.

6.10 FM stereo
INSTRUCTOR COPY

We have two messages, m1 (t) and m2 (t) each representing left and right parts of the
audio signal mL (t) and mR (t). We have some choices for how we will modulate the
FM wave. We can combine amplitude and frequency modulation:
s (t) =a (t) cos[2πfc t + φ (t)]
m1 (t) =a(t)
m2 (t) =φ(t)

This is not typically a good idea for FM since we are storing information in a(t)
and many FM receiver designs ignore amplitude variations as often a hard limiter is
applied to unify the incoming signal’s amplitude.

Another idea is to use Independent Sideband (ISB) introduced in chapter 5 to

store m1 (t) on the upper sideband m2 (t) on lower sideband using Hilbert Transforms
of each to ensure that they each only occupy a single sideband:
s (t) =sU SB (t) + sLSB (t) + scarrier (t)
s(t) =m1 (t) cos (2πfc t) − m̂1 (t) sin (2πfc t)
+ m2 (t) cos (2πfc t) + m̂2 (t) sin (2πfc t)
+ Ac cos (2πfc t)
s(t) = [Ac + m1 (t) + m2 (t)] cos (2πfc t) − [m̂1 (t) − m̂2 (t)] sin (2πfc t)
i (t) =Ac + m1 (t) + m2 (t)
q (t) =m̂1 (t) − m̂2 (t)

However, ISB has information encoded in both amplitude and phase and thus fails
the test of constant envelope desired for FM and PM.

We now try the idea of sum and difference of two messages:

 CHAPTER 6. ANGLE MODULATION (FM AND PM) 387

• mM (t) = mL (t) + mR (t), a mono signal

• mS (t) = mL (t) − mR (t), a difference signal

INSTRUCTOR COPY

Figure 6.26: Stereo FM transmitter

Therefore we have a new form of message storing all pertinent information for
stereo audio. In this method (fig. 6.26), the total message including both left and
right channels is written

m (t) =mM (t) + [mS (t)] cos (4πfsc t) + cos(2πfsc t)

=mL (t) + mR (t) + cos (2πfsc t) + [mL (t) − mR (t)] cos(4πfsc t)

Notice that the cos(4πfsc t) term is one of the outputs from the cos × cos multiplier
in fig. 6.26. The other is a DC term at 0 Hz which is filtered out before being passed
to the transmitter.

The difference signal mS (t) = mL (t) − mR (t) is modulated onto a subcarrier 2fSC
as a DSB signal, and the subscarrier at fsc = 19 kHz is added. The complete FM
stereo signal may be written
Z t
s (t) = Ac cos[2πfc t + 2πkf m(α)dα]
0
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 388

An FM stereo demodulator obtains mL (t) + mR (t) and also detects the subcarrier
at 19 kHz, doubles it to 2fSC by multiplying the subcarrier by itself, and uses the
doubled subcarrier to demodulate the DSB signal to obtain mL (t) − mR (t). These
two outputs are added and subtracted to yield mL (t) and mR (t).

Exercise 113
Sketch the spectrum of m(t).

Exercise 114
Sketch a block diagram of an FM stereo receiver with input s(t) and 2 outputs
mL (t) and mR (t).
INSTRUCTOR COPY

6.11 Chapter review questions

Q 6.1 (6 Marks) Do a complete derivation of the FM equation

s+ (t) = ej2πfc t ejβ sin(2πfm t)

for a single tone message m(t) = Am cos (2πfm t). Specify your initial assumptions on
which your derivation is based.

Solution
The three starting points for the idea of FM are:

1. the instantaneous frequency of the resulting FM wave is

fi (t) =fc + kf Am cos (2πfm t)

=fc + ∆f cos (2πfm t) .

2. a general signal is
s(t) = a(t)ejθ(t)

3. the instantaneous frequency of a signal is

1 dθ(t)
fi (t) =
2π dt
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 389

Combining these 3 starting points, we can write:

Z t
θ(t) =2π fi (α) dα
0
Z t
=2π [fc + kf m (α)] dα
0
Z t
=2πfc t + 2πkf m (α) dα
0
Z t
=2πfc t + 2πkf Am cos (2πfm α) dα
0
Am
=2πfc t + 2πkf sin (2πfm t)
2πfm
INSTRUCTOR COPY

kf Am
=2πfc t + sin (2πfm t)
fm
∆f
=2πfc t + sin (2πfm t)
fm
=2πfc t + β sin (2πfm t)
k A
where β = ffmm = ∆f fm
is called the modulation index of the FM wave. Thus
the FM wave itself is given in terms of β by

s+ (t) =Ac ejθ(t)

=Ac ej2πfc t+jβ sin(2πfm t)
=Ac ej2πfc t ejβ sin(2πfm t)

Q 6.2 (6 Marks) Consider the first two of the three FM starting points (Course
Notes section 5.2) as a guide for PM. One of the starting points for PM is that the
instantaneous phase is a constant phase plus a constant kp times the message.

Do a complete derivation of the phase modulation (PM) equation in both real

and complex form for a single tone message. Hint: The carrier wave frequency fc is
constant.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 390

Solution

θ(t) =2πfc t + φ(t)

φ(t) =φ0 + kp m(t)
=φ0 + kp Am cos (2πfm t)
s(t) =a(t) cos (θ(t))
=a(t) cos (2πfc t + φ(t))
=a(t) cos (2πfc t + φ0 + kp Am cos (2πfm t))
s+ (t) =a(t)ejθ(t)
=a(t)ej2πfc t+jφ(t)
=Ac ej2πfc t+jφ0 +jkp Am cos(2πfm t)
INSTRUCTOR COPY

=Ac ej2πfc t ejφ0 ejkp Am cos(2πfm t)

note that for both FM and PM there is a cos or sin in the exponential.

Q 6.3 Compare the PM equation with the FM equation. What is the key difference?

Solution
For PM the constant in the exponential kp is independent of fm whereas for
FM the constant in the exponential depends on both kf and fm .

PM

s+ (t) =Ac ej2πfc t+jφ0 +jkp Am cos(2πfm t)

 π
sin α + = cos (α)
2
s+ (t) =Ac ej2πfc t+jφ0 +jkp Am sin(2πfm t+ 2 )
π

FM

s+ (t) =Ac ej2πfc t+jβ sin(2πfm t)

k
f Am

j2πfc t+j sin(2πfm t)
=Ac e fm

Q 6.4 (6 Marks) Find the complex envelope for a NBFM signal.

 CHAPTER 6. ANGLE MODULATION (FM AND PM) 391

Solution
For β << 1

i(t) =Ac
q(t) =βAc sin (2πfm t)

Then the complex envelope is

p
a(t) = i2 (t) + q 2 (t)
q
=Ac 1 + β 2 sin2 (2πfm t)

using (1 + x)n = 1 + nx for x << 1 and n = 0.5,

INSTRUCTOR COPY

 
a(t) = Ac 1 + 0.5β 2 sin2 (2πfm t)
cos(α−β)−cos(α+β)
using sin α sin β = 2
,
 
q(t)
φ(t) = arctan
i(t)
 
βAc sin (2πfm t)
= arctan
Ac
= arctan (β sin (2πfm t))

x3
using arctan (x) = x − 3
for x << 1,

φ(t) = β sin (2πfm t)

Now we combine these with the definition of s̃(t)

s̃(t) =a(t)ejφ(t)
=i(t) + jq(t)
 
=Ac 1 − 0.25β 2 cos (4πfm t) ejβ sin(2πfm t)
 
=Ac 1 − 0.25β 2 cos (4πfm t) {cos (4πfm t) + jAc sin (4πfm t)}

Q 6.5 If an AM receiver is used to receive a NBFM signal, what is the output of the
AM receiver?
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 392

Solution
If an AM receiver is used the receive a NBFM signal (or any signal), the output
of the AM receiver is the magnitude of the complex envelope
 
a(t) = Ac 1 − 0.25β 2 cos (4πfm t)

Q 6.6 (4 Marks) Consider an FM signal with a single tone message m(t) = Am cos 2πf t
with f = 2 kHz transmitted at a carrier frequency of 100 MHz. Assuming the mod-
ulation sensitivity kf = 1 kHz/V, find a value of Am in volts such that the carrier
frequency amplitude is zero. There is more than one possible answer.
INSTRUCTOR COPY

Solution
From Bessel function curves/tables, J0 (β) = 0when β = 2.4, 5.5, . . .. Carrier
frequency is zero when
kf Am
β=
fm
1000Am
2.4 =
2000
Am =4.8
∆f =kf Am
=1000 × 4.8
=4800 Hz

Q 6.7 The following questions are based on the paper

[1] V. Shah, A. Parekh, S. Sawant and N. Bhagat, ”FPGA Implementation of

FM Demodulators Using RTL-SDR,” 2018 Fourth International Conference on
Computing Communication Control and Automation (ICCUBEA), Pune, India,
2018, pp. 1-6, doi: 10.1109/ICCUBEA.2018.8697623.

(4 marks) Write a formula for the simplified FM demodulator in Figure 4 of the

paper.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 393

Solution
Inputs to top multiplier:

i[n − 1] · (q[n] − q[n − 2])

Input to bottom multiplier:

(i[n] − i[n − 2]) · q[n − 1]

Sum output before scaling

∆θ[n] = i[n − 1] · (q[n] − q[n − 2]) − (i[n] − i[n − 2]) · q[n − 1]

INSTRUCTOR COPY

Q 6.8 (4 marks) Compare the formula obtained in Q1 above to the formula for the
demodulator in Figure 3 (given as equation 6) in the paper. How are the equations
different and how are they similar?

Solution
In the figure 4 equation, the scaling is not explicitly defined, whereas equation
6 shows scaling by i[n]2 + q[n]2 . As well, in the figure 4 equation the samples
are delayed by one, and the derivative is expressed as a difference:

dq[n] di[n]
ω = i[n] − q[n]
dt dt
= i[n − 1] · (q[n] − q[n − 2]) − (i[n] − i[n − 2]) · q[n − 1]
= ∆θ[n]

Q 6.9 Consider demodulating a FM signal with single tone message with the demod-
ulator Figure 3 in the paper.

(8 marks) Show mathematically that this demodulator gives the correct message
output. It may be necessary to make some approximations.

Solution
In discrete time t = nT = n. The calculation can be done in continuous time.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 394

An FM signal with a single tone message is

s̃(t) = i(t) + jq(t)

= Ac ejφ(t)
= Ac ejβ sin(2πfm t)
= Ac cos(β sin(2πfm t)) + jAc sin(β sin(2πfm t))

Remembering the chain rule f [g(x)]0 = f 0 [g(x)]g 0 (x), we have

i(t) = Ac cos(β sin(2πfm t))

di(t)
= −β2πfm AAc cos(2πfm t) sin(βsin(2πfm t))
dt
INSTRUCTOR COPY

q(t) = Ac sin(β sin(2πfm t))

dq(t)
= β2πfm Ac cos(2πfm t) cos(β sin(2πfm t))
dt
Plugging these into the non-scaled version of equation 6 (shown in Soln. 2 we
have
dq[n] di[n]
ω =i[n] − q[n]
dt dt
dq[t] di[t]
=i[t] − q[t]
dt dt
=Ac cos(β sin(2πfm t)Ac β2πfm cos(2πfm t) cos(β sin(2πfm t))
+ Ac sin(β sin(2πfm t)Ac β2πfm cos(2πfm t) sin(β sin(2πfm t))
=β2πfm cos(2πfm t)[A2c cos2 (β sin(2πfm t) + A2c sin2 (β sin(2πfm t)]
=β2πfm cos(2πfm t) = m(t)

Q 6.10 (4 marks) Repeat using the demodulator of Figure 4.

Solution
approximate double difference i[n] − i[n − 2] as derivative, and get same as
above

Q 6.11 (2 marks) Comment on possible numerical problems arising from using either
or both of these demodulators.
 CHAPTER 6. ANGLE MODULATION (FM AND PM) 395

Solution
this method avoids division by small numbers
INSTRUCTOR COPY
 Chapter 7

Baseband data transmission

INSTRUCTOR COPY

In this chapter we investigate how to take a digital data sequence of 1’s and 0’s and
turn it into a message waveform (m(t)).

7.1 Digital messages

m(t) is a digital message 11100101 as shown in fig. 7.1.

Figure 7.1: Digital message with data bits 11100101

In fig. 7.1, p(t) is chosen to be a so-called square pulse, where we can write
(
1, 0 < t < T
p(t) =
0, otherwise

396
 CHAPTER 7. BASEBAND DATA TRANSMISSION 397

We send one such pulse for each data bit, a positive pulse p(t) for logic 1 and a
negative pulse for logic 0. In fig. 7.1 the message is shown in the interval t = 0 → 8T .

For the period t = 0 → T , m(t) = A0 p(t) = 1 and the data is logic 1. For the
next period T < t < 2T , m(t) = A1 p(t − T ) = 1, and the data is logic 1.

We can write

m(t) = A0 p(t) + A1 p(t − T ) + A2 p(t − 2T ) + . . .

X
= Ak p(t − kT )
k

where the real constants Ak = ±1 depending on whether a logic 1 or a logic 0 was

sent.
INSTRUCTOR COPY

Converting from logic 0 to Ak = −1 and from logic 1 to Ak = +1 can be done

easily in software like GNU Radio by implementing Ak = 2x − 1 where x = 0 or 1
is the logic bit. Notice that this equation sets logic 1 to Ak = +1 and logic 0 to
Ak = −1.

In fig. 7.1, A0 = 1, A1 = 1, A2 = 1, A3 = −1, A4 = −1, A5 = 1, A6 = −1, A7 =

1. The constants may also be chosen to be Ak = ±1, ±3 . In such a 4 level system,
each pulse represents one of 4 combinations of 2 bits: 00, 01, 11, 10.

7.2 Data source

The data source produces the data symbols Ak = ±1 that may be stored in memory
or generated on the fly. For testing purposes, it is convenient to have a random
data source, with an equal number of logic 1s and logic 0s, or an equal number of
positive and negative pulses. This can be achieved with a so-called linear feedback
shift register (LFSR). Figure fig. 7.2 shows an example with the first 3 iterations
completed.
 CHAPTER 7. BASEBAND DATA TRANSMISSION 398

Seed word: 101011

1 0 1 0 1 1

+ +

1 1 0 1 0 1

+ +

1 1 1 0 1 0
INSTRUCTOR COPY

+ +

Figure 7.2: LFSR with two feedback taps (XOR gates)

At each time step (symbol time) T , the input on the left is determined by the
XOR gate output and all the bits slide along one step and the output is the bit on
the right. If the feedback shift register is of length M and the feedback taps (XOR
gates) are selected correctly, the output sequence will have length 2M − 1 before it
repeats.

GNURadio has a GLSFR block where the length and feedback taps can be selected.
Tables of feedback taps for each length M are available.
 CHAPTER 7. BASEBAND DATA TRANSMISSION 399

Exercise 115
Write the next several iterations of the LFSR shown in fig. 7.3.

Seed word: 111

1 1 1

? ? ?
INSTRUCTOR COPY

Figure 7.3: Simple LFSR with one feedback tap (XOR gate)

Solution
L
The output of the summer goes to the shift register input on the left side
and all bits slide to the right by one position.
0 1 1
1 0 1
0 1 0
0 0 1
1 0 0
1 1 0
1 1 1
The output sequence (left column) 0 1 0 0 1 1 1 is periodic of length 7 =
23 − 1

The output will depend on the position of the feedback taps. Only some choices
of feedback taps will result in the maximal length sequence output.

Exercise 116
What is the output of the LFSR if the seedword is 000
 CHAPTER 7. BASEBAND DATA TRANSMISSION 400

Solution
Zeros forever. For the LFSR to produce a non-zero sequence output, the seed
word must contain at least one 1 bit.

7.3 Differential encoding and decoding

7.3.1 Avoiding polarity inversion

A data receiver may easily invert the polarity of the received sequence. This can
happen for example if the data is represented by phase shifts as will be shown in
INSTRUCTOR COPY

chapter 7 on phase shift keying. It is easy to detect changes in phase but much more
difficult if not impossible to detect absolute phase.

To avoid this problem, the data sequence is differentially encoded. Data are
represented in terms of the changes between successive signal elements (bits) rather
than the signal elements (bits) themselves.

The polarity of the differentially encoded signal can be unintentionally inverted by

the receiver. This inversion will not have any effect on the decoded signal waveform,
as shown in fig. 7.4. The encoder output starts with a 1, but this is arbitrary, it could
also be a 0. The received sequence can be inverted or not inverted. The decoder
output is the same in both cases.

7.3.2 Differential encoding

The arrows in fig. 7.4 show how the differential encoding works. At the beginning,
we have 1 + 1 = 0, then 0 + 0 = 0, then 0 + 1 = 1, etc.
 CHAPTER 7. BASEBAND DATA TRANSMISSION 401
INSTRUCTOR COPY

Figure 7.4: Example of differential encoder and decoder

Exercise 117
Repeat the differential encoder in fig. 7.4, but with the first bit in the differential
encoder set to 0 instead of 1.

In some cases, the decoder output is polarity reversed by convention, so that a

change in the received sequence is decoded as a “0” bit and no change as a “1” bit.

In the differential encoder, input data bits are modulo 2 SUM with the previous
output bits. Modulo 2 SUM is same as XOR notated with ⊕. The differential encoder
equation is
en = dn ⊕ en−1 (7.1)
and the block diagram is shown in fig. 7.5.
 CHAPTER 7. BASEBAND DATA TRANSMISSION 402

Figure 7.5: Differential encoder block diagram

INSTRUCTOR COPY

7.3.3 Differential decoding

The arrows in fig. 7.4 show how the differential decoding works. For the received
sequence, we have 1 + 0 = 1, then 0 + 0 = 0, then 0 + 1 = 1, etc. For the inverted
sequence, 0 + 1 = 0, 1 + 1 = 0, 1 + 0 = 1, etc. In both cases, a change in the received
sequence results in the decoder outputting a “1” bit, and no change results in a “0”
bit.

In the differential decoder, the current input and a delayed version of the same is
fed to the modulo 2 sum. This produces the output bits.

The differential decoder equation is

dn = en ⊕ en + 1 (7.2)

and the block diagram is shown in fig. 7.6.

Figure 7.6: Differential decoder block diagram

 CHAPTER 7. BASEBAND DATA TRANSMISSION 403

Exercise 118
Do a differential decoding of the differential encoder output sequence obtained
in the previous exercise from fig. 7.4 and verify that the decoded sequence is
the same as in the figure.

7.4 Data transmission system with pulse shaping

When transmitting data, it is desirable to minimize the signal bandwidth and also
minimize the bit error rate. To minimize bandwidth, we need to round the edges of
the square pulses in fig. 7.1 or create rounded pulses in another way. To minimize bit
INSTRUCTOR COPY

error rate, we need to make sure that the pulses are shaped so that one pulse does
not interfere with adjacent pulses (no inter-symbol interference).

A data transmission system that sends bits from one place to another contains
transmitter with pulse shaping as shown in fig. 7.7 followed by a receiver with receive
filter and a decision device. The receive filter is ideally matched to the transmit pulse
shape filter, but may be a simple low pass filter.

Figure 7.7: Transmitter with pulse shaping

The received message waveform after filtering by both the transmitter pulse shap-
ing filterPand the receiver low pass or matched filter is a sequence of delayed pulses
m(t) = k Ak p(t − kT ). In general the waveform m(t) will have a peak amplitude
that varies depending on the data sequence Ak .

The decision device carries out two tasks:

1. samples this waveform m(t) at the time t = kT + δ of maximum amplitude for

each symbol, where the value of δ depends on the pulse shape, and

2. decides what is the value of the symbol (±1 for a binary two level system, or
±1, ±3 for a 4 level system).
 CHAPTER 7. BASEBAND DATA TRANSMISSION 404

Pulse shaping can be done in multiple ways. Here we consider four methods:

1. Generate square pulses and filter with low pass filter.

2. Generate impulses and filter with time domain FIR pulse-shaping filter.

3. Generate impulses and filter with frequency domain FIR pulse-shaping filter.

4. Generate impulses and filter with Gaussian pulse-shaping filter (time and fre-
quency domain)

In the following sections, we describe how to:

INSTRUCTOR COPY

• represent data as square pulses in a sampled system

• filter the square pulses in time domain

• represent data sequence with impulses instead of square pulses

• filter impulses with time domain RC filter

• filter impulses with raised cosine filter shape in frequency domain

• filter impulses with root raised cosine filter in frequency domain

appendix D.2 contains a review of digital filters as applied to pulse shaping.

7.4.1 Representing digital data as square pulses

For a sampling frequency fs and symbol time T , there are fs T samples per symbol.
Each square pulse contains fs T samples of identical value.

The data symbols Ak = ±1 are represented by one sample per symbol and come
from the data source at a rate 1/T symbols per second. To generate the square
pulse, each sample of data must be repeated fs T times, so that the square pulse is
represented by fs T samples running at sampling frequency fs to make up a symbol
that takes time T to transmit. We also refer to the symbol time as the symbol length.
A sequence of such symbols will look like the waveform m(t) shown in the top plot
of fig. 7.9 with fs T samples per symbol.
 CHAPTER 7. BASEBAND DATA TRANSMISSION 405

7.4.2 Pulse shaping to eliminate the sharp edges of a square

pulse

The square pulse can be low pass filtered with a cutoff frequency of 1/(2T ) since this is
the highest frequency in the data stream for an alternating data sequence 101010...
(
1, k is odd
Ak = (7.3)
−1, k is even

Ideally the low pass filter has an impulse response such that with a square pulse input
from t = 0 → T (fig. 7.8a), the output is approximately a raised cosine pulse in the
time domain, from t = 0 → 2T (fig. 7.8b).
INSTRUCTOR COPY

1 1

0.5 0.5

T 2T T 2T

(a) Square pulse (b) RC pulse

Figure 7.8: Raised cosine (RC) shape and RC shaped bitstream

We write the raised cosine (RC) pulse p2 (t) = pRC (t) where
(
0.5(1 − cos(πt/T )), 0 < t < 2T
pRC (t) = (7.4)
0, otherwise

Notice that for the RC pulse

(
0, t = 0, 2T
pRC (t) =
1, t = T

This pulse is often called an RC2 pulse since it is truncated to 2 symbol periods.
 CHAPTER 7. BASEBAND DATA TRANSMISSION 406

When a data sequence Ak is sent with these pulses, the waveform m(t) has rounded
edges with minimal overshoot as shown in the bottom plot of fig. 7.9. At times kT ,
only one pulse in m(t) is non-zero and the adjacent pulses are zero and thus there is
no inter-symbol interference.

1T 2T 3T 4T 5T 6T 7T 8T
−1
1
INSTRUCTOR COPY

1T 2T 3T 4T 5T 6T 7T 8T
−1
1

1T 2T 3T 4T 5T 6T 7T 8T
−1

Figure 7.9: Digital waveform with bits 11000101. Top: square pulses representing
digital data msq (t = nTs ), Middle: filtered square pulses showing resultant RC2
pulses, Bottom: resulting smooth waveform m(t = nTs ) (sum of all the RC2 pulses)

The process for transmitting a digital bit stream in this manner is to begin with
the sequence

msq (t) =A0 p(t) + A1 p(t − T ) + A2 p(t − 2T ) + . . .

X
= Ak p(t − kT )
k

where p(t) is a square pulse as described in eq. (7.5) and Ak is either 1 or -1.
(
1, 0 < t < T
p(t) = (7.5)
0, otherwise

This sequence msq (t) is low pass filtered by an analog Bessel filter with an impulse
 CHAPTER 7. BASEBAND DATA TRANSMISSION 407

response h(t) to yield another sequence of pulses

m(t) =A0 (t)p2 (t) + A1 p2 (t − T ) + A2 p2 (t − 2T ) + . . .

X
= Ak p2 (t − kT )
k

where p2 (t) = p(t) ⊗ h(t) has an approximately raised cosine shape from t = 0 → 2T .

In other words, the step response of h(t) to a single square pulse p(t) = 1 from
t = 0 → T will approximate a raised cosine pulse from t = 0 → 2T .

For an alternating sequence as in eq. (7.3),

t
m(t) = cos π
INSTRUCTOR COPY

T
= cos 2πfm t
1
where fm = 2T
.

The step response (integral of the impulse response) of the Bessel filter h(t) approx-
imates half of the raised cosine shape, as shown in fig. 7.10, starting at pRC (t = 0) = 0
to pRC (t = T ) = 1.

Figure 7.10: Step and impulse responses of Bessel, Butterworth and 0.5 dB Chebyshev
filters

A family of Bessel step responses are shown in fig. 7.11a, depending on the number
of poles. The frequency responses of these filters are shown in fig. 7.11b.
 CHAPTER 7. BASEBAND DATA TRANSMISSION 408

(b) Frequency responses

(a) Step responses

INSTRUCTOR COPY

Figure 7.11: Step and frequency responses of Bessel function at varying number of
poles

Pulse shaping of a square pulse using a low pass filter in GRC is shown in fig. 7.12.
The sampling rate is set to fs = 1/Ts = 100 kHz and the symbol rate is set to f0 =
1/T = 1 kHz. The Galois linear feedback shift register (GLFSR) data source generates
one sample for each data symbol Ak = ±1 at a rate 1/T (symbol rate) symbols
per second. The Repeat block interpolates to make samp rate//symbol rate= N =
fs T = 100 samples for each data symbol, so that each square pulse contains 100
samples of identical value. The low pass filter cutoff frequency is set to 1000 Hz and
approximates the Bessel filter.

Options Variable Variable

Title: Square w...lse shaping Id: samp_rate Id: symbol_rate
Output Language: Python Value: 100k Value: 1k
Generate Options: QT GUI

GLFSR Source Throttle

in out
Degree: 10 Sample Rate: 100k
Repeat QT GUI Time Sink
Repeat: Yes out in out in0
Interpolation: 100 Number of Points: 1.024k
Mask: 0
Low Pass Filter Sample Rate: 100k
Seed: 1 in1
Decimation: 1 Autoscale: No
Gain: 1
Sample Rate: 100k
in out
Cutoff Freq: 1k
Transition Width: 200
Window: Hamming
Beta: 1

Figure 7.12: GNU Radio flowgraph showing square pulses being smoothed with a low
pass filter (approximating a Bessel filter)

Since the GNURadio flowgraph is implemented in discrete time, we consider the

 CHAPTER 7. BASEBAND DATA TRANSMISSION 409

signal sample rate fs = 1/Ts , choose N samples per data symbol so that the data
symbol time is T = T0 = N Ts and the data symbol rate is f0 = 1/(N Ts ). This system
is an example of multirate signal processing. For example, the data symbol rate may
be f) = 1 KHz and the signal sampling rate may be fs = 100 KHz, so that N = 100
samples per data symbol.

The digital message is sampled at the data symbol rate

Ak = A(kT )
X
md (t) = A(kT )p(t − kT )
k

In the above equation, there is one sample per data symbol.

INSTRUCTOR COPY

We write the message at the message sampling rate where there are N samples
per data symbol and thus we substitute T = N Ts . We rename md sampled at the
symbol rate to msq sampled at the message rate.

fig. 7.9 illustrates msq (t) in blue, the pulses p2 (t) and the message m(t) in red.
The samples at nTs can be shown by adding N tick marks for each data symbol of
period T .

The math operations to generate fig. 7.9 suitable for implementation in Matlab
or other software are as follows.

X
md (t) = Ak p(t − kT )
k
X
msq (t = nTs ) = A(kN Ts )p(nTs − kN Ts )
k
X
msq (n) = A(kN )p(n − kN )
k

The second line implements the repeat block function to provide N samples of a
square pulse for the nth data symbol. The third line omits the implied sampling time
Ts and represents msq in terms of the sample index n. msq (n) is the blue line in the
first row of fig. 7.9. Here we see multirate signal processing at work, with data symbol
time T = N Ts and waveform sampling time Ts .

The square pulses p(t) pass through the low pass filter with impulse response h(t),
 CHAPTER 7. BASEBAND DATA TRANSMISSION 410

where h(t) is chosen such that p2 (t) = p(t) ⊗ h(t) is RC2. We write

p2 (t = nTS ) = p(nTs ) ⊗ h(nTs )

X
= p(kTS )h(nTs − kTs )
k

The RC2 pulses are shown in the second row of fig. 7.9

The message msq (t) consisting of the sum of delayed square pulses is the input to
the low pass filter h(t). The filter output

m(t = nTs ) = msq (nTs ) ⊗ h(nTs )

INSTRUCTOR COPY

is the red curve in fig. 7.9.

Exercise 119
Find an expression for m(t = nTs ) as a function of the RC2 pulse p2 (t). Hint:
substitute the expression for msq (kTs ) (using summing index q) into the con-
volution with h(nTs ) (using summing index k).

Solution

m(nTs ) = msq (nTs ) ⊗ h(nTs )

X
= msq (kTs )h(nTs − kTs )
k
XX
= A(qnTs )p(kTs − qN Ts )h(nTs − kTs )
k q
X
= A(qN Ts )p2 (nTs − qN Ts )
q
X X
m(n) = A(kN )p2 (n − kN ) = Ak p2 (n − kN )
k k

The 3rd line for m(nTs ) arises from the definition

X
msq (kTs ) = A(qN Ts )p(kTs − qN Ts )
q
 CHAPTER 7. BASEBAND DATA TRANSMISSION 411

The 4th line for m(nTs ) arises from the definition

X
p2 (nTS ) = p(kTS )h(nTs − kTs )
k
X
p2 (nTS − qN Ts ) = p(kTs − qN Ts )h(nTs − kTs )
k

m(nTs ) is the red curve in fig. 7.9. The data symbols Ak occur once per
symbol time T = N Ts , whereas the message waveform m(n) is sampled at rate
fs = 1/Ts .
INSTRUCTOR COPY

7.4.3 Representing digital data with impulses

Instead of a sequence of square pulses, the data bits can be represented by the sequence
of impulses

mδ (t) =A0 δ(t) + A1 δ(t − T ) + A2 δ(t − 2T ) + . . .

X
= Ak δ(t − kT )
k

This sequence of impulses mδ (t) as shown in the top plot of fig. 7.13 is a waveform
with only one sample per symbol, so that we can write mδ,k = mδ (t = kT ) and the
sampling rate is 1/T .

For transmission, this waveform is convolved with a pulse shaping filter with
impulse response p(t) and frequency response P (f ) where the convolution operates
at a sample rate fs . To represent this sequence of impulses mδ,k = mδ (t = kT ) as a
waveform sampled at fs , the impulses must be filtered by an interpolating FIR filter
that yields one sample Ak = ±1 followed by fs T − 1 samples set to zero, so that the
fs T samples per symbol have only one non-zero value. This interpolating FIR filter
will have only one non-zero tap and fs T − 1 zero taps.

This waveform mδ (t) sampled at fs is then in turn convolved with an (regular,

non-interpolating) FIR filter (operating at fs ) with impulse response p(t). The output
of this FIR filter yields the message waveform m(t).

We now investigate methods to pulse shape this sequence of impulses.

 CHAPTER 7. BASEBAND DATA TRANSMISSION 412

7.4.4 Pulse shaping a sequence of impulses

The train of impulses spaced by fs T samples (and sampled at fs ) is convolved with a

pulse shape, p(t). This is done in a FIR filter. The filter coefficients of this FIR filter
will be the pulse shape p(t) sampled at fs .

7.4.4.1 Square pulse filter

p(t) can be a square pulse (eq. (7.5)) and thus will have fs T taps. In this case the
output waveform, m(t) is a sequence of square pulses as shown in the top (blue) plot
of fig. 7.9. To transmit this, more pulse shaping would be required to eliminate its
sharp edges (as we saw in section 7.4.2).
INSTRUCTOR COPY

7.4.4.2 Time-domain raised-cosine filter

As a better option, the FIR coefficients can also represent a longer pulse, for example
the raised-cosine pulse (eq. (7.4)) which will have 2fs T taps.

fig. 7.13 illustrates the process of generating the message waveform m(t) from a
sequence of impulses. Note that the raised cosine pulse extends over 2 symbol periods,
so that adjacent symbol pulses will overlap (middle plot of fig. 7.13). The resulting
message waveform will appear as shown in the bottom plot of fig. 7.13, sampled at
fs T samples per symbol. At times kT , only one pulse in m(t) is non-zero and the
adjacent pulses are zero and thus there is no inter-symbol interference.
 CHAPTER 7. BASEBAND DATA TRANSMISSION 413

1T 2T 3T 4T 5T 6T 7T 8T 9T
−1
1

1T 2T 3T 4T 5T 6T 7T 8T 9T
−1
1
INSTRUCTOR COPY

1T 2T 3T 4T 5T 6T 7T 8T 9T
−1

Figure 7.13: Digital waveform with bits 11000101. Top: impulses representing digital
data mδ (nTs ), Middle: filtered impulses showing resultant RC2 pulses at various
delays, Bottom: resulting waveform m(nTs ) (sum of all the delayed RC2 pulses)

Thus we can generate time domain raised-cosine pulses pRC (t) using either a se-
quence of square pulses or a sequence of impulses. Notice a key difference between
these two methods: the sequence of square pulses was filtered with an analog Bessel
filter while the sequence of impulses was filtered with a raised-cosine filter, both of
which resulted in a raised-cosine shaped waveform.

Pulse shaping of impulses with a pulse shaping filter using GRC is shown in
fig. 7.14. The sample rate fs = 100 kHz and the symbol rate 1/T = 1000 Hz so that
fs T = 100 samples.
 CHAPTER 7. BASEBAND DATA TRANSMISSION 414

Options Variable Variable Import Variable

Title: Impulse pulse shaping Id: samp_rate Id: symbol_rate Import: np Id: RC2_taps
Output Language: Python Value: 100k Value: 1k Value: 0.5*(1-np.cos(np.pi...
Generate Options: QT GUI

GLFSR Source
Degree: 10 Interpolating FIR Filter
Throttle QT GUI Time Sink
Repeat: Yes out in Interpolation: 100 out in out in
Sample Rate: 100k Filter Delay Number of Points: 1.024k
Mask: 0 Taps: 1 out in
Taps: RC2_taps Sample Rate: 100k
Seed: 1 in
Autoscale: No

Figure 7.14: GNU Radio flowgraph for generating a pulse-shaped message waveform
starting with a sequence of impulses mδ (nTs ) filtered by a pulse-shaping filter resulting
in output m(nTs )

The linear feedback shift register (GLFSR) data source generates one sample for
INSTRUCTOR COPY

each data symbol. These samples are filtered by an Interpolating FIR Filter that
yields one sample Ak = ±1 followed by fs T − 1 = 99 samples set to zero, so that the
fs T = 100 samples per symbol have only one non-zero value. This Interpolating
FIR Filter will have only one non-zero tap and fs T − 1 = 99 zero taps. The output
of this filter is a sequence of impulses at a rate 1/T = 1000 Hz and sampling rate
fs = 100 kHz. This output is filtered by a RC2 filter shape contained inside the
RC2 taps variable. Again, this is another example of multirate signal processing.

The RC2 shape is built inside the Variable block using the following line of
Python code 0.5*(1-np.cos(np.pi*np.arange(0,2/symbol rate,1/samp rate)*symbol rate))
The Import block contains the line import numpy as np which allows us to use the
NumPy library.

fig. 7.13 illustrates mδ (t) in blue, the pulses p2 (t) and the message m(t). The
samples at nTs can be shown by adding N tick marks for each data symbol of period
T.

The math operations to generate fig. 7.13 suitable for implementation in Matlab
or other software are as follows
X
mδ (t) = Ak δ(t − kT )
k
X
mδ (t = nTs ) = A(kN Ts )δ(nTs − kN Ts )
k
X
mδ (n) = A(kN )δ(n − kN )
k

The second line implements the Interpolating FIR Filter whose output for a
specific value of k is mδ (nTs ) for a specific value of k is Ak δ(nTs − kN Ts ) which
 CHAPTER 7. BASEBAND DATA TRANSMISSION 415

represents one sample Ak = A(kN Ts ) when n = kN followed by N − 1 zero samples

for other values of n. For all values of k, the δ function results in mδ (nTs ) having
value Ak every N th sample and 0 for all other samples, as shown in the top (blue)
line of fig. 7.13. The last line omits the implied sampling time Ts . Again, we see
multirate signal processing at work, with data symbol time T = N Ts and waveform
sampling time Ts .

The low pass filter is an FIR filter with RC2 shaped impulse response p2 (t). In
practice an IIR low pass filter with cutoff frequency 1/(2T ) will have an approximate
RC2 shaped impulse and may be used in place of the FIR filter. After the sequence
of impulses mδ (nTs ) passes through the low pass filter, the output is the message
waveform
m(nTs ) = mδ (nTs ) ⊗ p2 (nTs )
INSTRUCTOR COPY

Exercise 120
Find an expression for m(nTs ) by evaluating the convolution.

Solution
Following the same ideas as in the previous exercise:

m(nTs ) = mδ (nTs ) ⊗ p2 (nTs )

X
= mδ (kTs )p2 (nTs − kTs )
k
XX
= A(qnTs )δ(kTs − qN Ts )p2 (nTs − kTs )
k q
X
= A(qN Ts )p2 (nTs − qN Ts )
q
X X
m(n) = A(kN )p2 (n − kN ) = Ak p2 (n − kN )
k k

The 4th line arises from the definition

X
p2 (nTs ) = p2 (kTs )δ(nTs − kTs )
k
X
= δ(kTs )p2 (nTs − kTs )
k
X
p2 (nTs − qN Ts ) = δ(kTs − qN Ts )p2 (nTs − kTs )
k

m(nTs ) is the bottom (red) curve in fig. 7.13.

 CHAPTER 7. BASEBAND DATA TRANSMISSION 416

7.4.4.3 Frequency domain raised-cosine filter

As shown in the previous sections, p(t) can be chosen to obtain a particular shape in
the time domain. However, p(t) can also be chosen to have a particular shape in the
frequency domain, i.e. a frequency response P (f ) = H(f ). One example choice is a
raised cosine shape in the frequency domain H(f ) = HRC (f ), as shown in fig. 7.15a
and described by eq. (7.6).

H(f )
T h(t)
1 β=1
β=1
β = 0.5
β = 0.5
β=0
β=0
INSTRUCTOR COPY

t
f −3T−2T −T T 2T 3T
− T1 1
− 2T 1
2T
1
T

(b) Impulse response

(a) Frequency response

Figure 7.15: Raised cosine shape in frequency domain



T,h |f | ≤ 1−β
   i 2T
T πT 1−β 1−β 1+β
H(f ) = 1 + cos |f | − , < |f | ≤ (7.6)
2

β 2T 2T 2T
0, otherwise

The frequency response can be varied with different values of the roll-off factor β
and the symbol time T . The roll-off factor is a measure of the excess bandwidth δf
of the pulse shaping filter, where δf is the bandwidth beyond the Nyquist bandwidth
1/2T . Thus β is defined
δf
β=
0.5T (7.7)
=2T δf

The impulse response h(t) is a sinc shape extending for a time usually truncated
to 6T as shown in fig. 7.15b and described by eq. (7.8).
 CHAPTER 7. BASEBAND DATA TRANSMISSION 417

  πβt


t cos T
h(t) = sinc 4β 2 t2
(7.8)
T 1−
T2

When β = 1,
T 1
H(f ) = [1 + cos(πf T )], |f | ≤ .
2 T
Note that this equation representing a frequency domain pulse has the same form as
the time domain RC2 pulse (eq. (7.4)) but with a sign difference. The sign difference
results from the different limits on the variables t and f , but the RC2 shape is the
same.

A sequence of these sinc-shaped pulses will overlap, as shown in the middle plot
INSTRUCTOR COPY

of fig. 7.16 for a data sequence. It is important to notice that all of the pulses share
zero crossings. This means there is zero intersymbol-interference (zero-ISI) between
the bits.

Because of the sinc shaped pulses, the message waveform will have variable peak
amplitude as shown in the bottom plot of fig. 7.16.

1T 2T 3T 4T 5T 6T 7T 8T 9T 10T 11T 12T 13T

−1

1T 2T 3T 4T 5T 6T 7T 8T 9T 10T 11T 12T 13T

−1

1T 2T 3T 4T 5T 6T 7T 8T 9T 10T 11T 12T 13T

−1

Figure 7.16: Digital waveform with bits 11000101. Top: impulses representing dig-
ital data mδ (nTs ), Middle: filtered impulses showing resultant sinc pulses, Bottom:
resulting waveform (sum of all the sinc pulses) m(nTs )
 CHAPTER 7. BASEBAND DATA TRANSMISSION 418

The math operations to generate fig. 7.16 suitable for implementation in Matlab
or other software are almost identical to those in section 7.4.4.2. The sequence of
impulses (top blue line in fig. 7.16 is as before
X
mδ (t) = Ak δ(t − kT )
k
X
mδ (nTs ) = A(kN Ts )δ(nTs − kN Ts )
k
X
mδ (n) = A(kN )δ(n − kN )
k

where in the last line the sampling time Ts is omitted amd implied. The δ function
results in mδ (n) having value A(kN ) every N th sample and 0 for all other samples,
INSTRUCTOR COPY

These equations are represented by the top (blue) line in fig. 7.16.

For the case of the frequency-domain RC pulse shaping filter, the time domain
pulse p2 (t) is replaced by h(t) ↔ H(f ). After the frequency domain raised cosine
filter with impulse response h(t), the message waveform is the convolution
m(nTs ) = mδ (nTs ) ⊗ h(nTs )
where h(t) is sampled at t = nTs .

Exercise 121
Write an expression for the message m(nTs ) with a frequency domain RC pulse
shaping filter h(nTs )

Solution

m(nTs ) = mδ (nTs ) ⊗ h(nTs )

X
= mδ (kTs )h(nTs − kTs )
k
XX
= A(qnTs )δ(kTs − qN Ts )h(nTs − kTs )
k q
X
= A(qN Ts )h(nTs − qN Ts )
q
X X
m(n) = A(kN )h(n − kN ) = Ak h(n − kN )
k k

These equations are represented by the bottom (red) line in fig. 7.16.
 CHAPTER 7. BASEBAND DATA TRANSMISSION 419

7.4.4.4 Frequency domain square root raised cosine filter

A receiver is optimum if the receiver filter matches the transmit filter. Thus we
often use a Square Root Raised Cosine (RRC) pulse shaping filter HRRC (f ) at the
transmitter and another RRC filter at the receiver, so that the combination of the two
filters yields a Raised Cosine (RC) pulse shape with zero ISI at the receiver output.
2
HRC (f ) = HRRC (f )

The RRC filter impulse response looks similar to the RC impulse response but with
lower sidelobes and does not have zero crossings spaced at multiples of T (fig. 7.17).
The mathematical expression for hRRC (t) = p6 (t) is
INSTRUCTOR COPY

 1 β



 √ 1 − β + 4 , t=0
 βs h
    i
T π



hRRC (t) =
√
2Ts
1 + π2 sin 4β π
+ 1 − π2 cos 4β
π Ts
, t = ± 4β (7.9)

 sin[π Tt (1−β)]+4β Tt cos[π Tt (1+β)]
 1
 √Ts s h s
2
i s
, otherwise
π Tt 1−(4β Tt )
s s

h(t) RC
RRC
1

−3T−2T −T T 2T 3T

Figure 7.17: Impulse responses of raised-cosine and square root raised cosine for
β = 1.

Pulse shaping of data using an RRC filter in GRC is shown in fig. 7.18. The sample
rate fs = 100 kHz and the symbol rate fsym = 1 kHz. The first Root Raised Cosine
Filter block interpolates the bits by 100 (samp rate//symbol rate) to yield 100
samples per symbol time T . The RRC shaped waveform travels through a “chan-
nel” made by the Virtual Source and Virtual Sink blocks before being filtered by
 CHAPTER 7. BASEBAND DATA TRANSMISSION 420

another Root Raised Cosine Filter block. This second RRC also decimates the
signal by 100 (samp rate//symbol rate) to recover the original bitstream.

Options Variable Variable

Title: Impulse ...lse shaping Id: samp_rate Id: symbol_rate
Output Language: Python Value: 100k Value: 1k
Generate Options: QT GUI QT GUI Time Sink
Number of Points: 1.024k
Root Raised Cosine Filter in
Sample Rate: 100k
GLFSR Source Interpolation: 100
Autoscale: No
Degree: 10 Gain: 1
Throttle
Repeat: Yes out in out in Sample Rate: 100k out
Sample Rate: 100k
Mask: 0 Symbol Rate: 1k
Seed: 1 Alpha: 350m
Virtual Sink
Num Taps: 600 in
Stream ID: channel

Root Raised Cosine Filter

Decimation: 100
QT GUI Time Sink
Gain: 1
INSTRUCTOR COPY

Virtual Source Number of Points: 1.024k

out in Sample Rate: 100k out in
Stream ID: channel Sample Rate: 100k
Symbol Rate: 1k
Autoscale: No
Alpha: 350m
Num Taps: 600

Figure 7.18: GNU Radio flowgraph showing root-raised cosine pulse shaping of im-
pulses

The math operations for this implementation of the RRC filter are simpler than
the RC example given in section 7.4.4.3 because the GRC RRC block takes in one
sample per symbol and outputs N samples per symbol. The convolution operations
described in section 7.4.4.3 are internal to the GRC RRC block. The effect of the
multirate signal processing is not explicitly shown in this case, since use use only the
data symbol time T = N Ts and the and waveform sampling time Ts is internal to the
GRC RRC block.

The input data sequence Ak is written at the data sampling rate f0 = 1/T as
X
mδ (t) = Ak δ(t − kT )
k
X
mδ (n) = Ak δ(n − k)
k

where in this case t = nT , not t = nTs .

After the frequency domain RRC filter with impulse response hRRC (t) = p6 (t) that
interpolates by a factor N , the output is at the waveform sampling rate fs = 1/Ts
 CHAPTER 7. BASEBAND DATA TRANSMISSION 421

and we write t = nTs and

X
m(nTs ) = A(kN Ts )p6 (nTs − kN Ts )
k
X
m(n) = Ak p6 (n − kN )
k

The input and output waveforms appear very similar to those in fig. 7.16.

7.4.5 Summary of pulse shaping methods

INSTRUCTOR COPY

Pulse shaping to minimize bandwidth and inter-symbol interference can be done in

multiple ways. Here we considered four methods:

1. Generate square pulses and low pass filter, commonly used in analog systems.

2. Generate impulses and low pass filter with a time domain RC FIR filter. In
this case the RC FIR filter p(t) is strictly time limited with zero value outside
a time range and P (f ) theoretically has sidelobes spread out at all frequencies.
In practice, the sidelobe amplitude versus frequency will be negligible beyond
frequencies greater than a few multiples of 1/T . Thus P (f ) will not spread
significantly in bandwidth beyond a few multiples of 1/T .

3. Generate impulses and filter with a frequency domain RC FIR filter. In this
case P (f ) is strictly band limited with zero value outside a frequency range
and p(t) theoretically has sidelobes spread out over all time. In practice, the
sidelobe amplitude versus time will be negligible beyond times greater than a
few multiples of T . Thus the FIR filter length is chosen so that the sinc-shaped
pulse p(t) does not spread in time beyond a few multiples of T .

4. Generate impulses and filter with a frequency domain RRC FIR filter. The
RRC shape is preferred over the RC shape in the frequency domain so that
the receiver RRC filter can be matched to the transmitter RRC filter, yielding
an overall RC system (transmitter plus receiver) response with zero ISI. The
advantage of the RC (rather than RRC) system response is that the eye diagram
is more open (see the next section) and the bit error rate is minimized.

5. The middle ground between 2 and 3 above is to generate impulses and use a
filter with Gaussian shaped impulse response. A Gaussian pulse shape will be
Gaussian in both time and frequency domains (fig. 7.19). The Gaussian pulse
theoretically spreads over all frequencies and all time, since the Gaussian tails
 CHAPTER 7. BASEBAND DATA TRANSMISSION 422

extend to ±∞. In practice, Gaussian pulses will be both time limited and
bandwidth limited.
INSTRUCTOR COPY

Figure 7.19: Gaussian pulse shaping in both time and frequency domains

7.5 Eye diagrams

The received message waveform after filtering by both the transmitter pulse shaping
filter and the receiver low pass or matched filter is a sequence of pulses that in general
will have a variable peak amplitude. The message waveform may be further varied
by the channel filter (causing waveform distortion), noise and interference (fig. 7.7).

To estimate the probability of a decision error, it is useful to view the message

waveform modulo the symbol time T (m(t) mod T ). This can be achieved in practice
by triggering the scope1 on the zero crossings of the message waveform. The resulting
picture is called an eye diagram.

Ideal eye diagrams are shown in fig. 7.20 for the two cases:

1. time domain RC pulse shape (fig. 7.20a) and

2. the frequency domain RRC pulse shape with sinc-shaped pulses (fig. 7.20b).
1
Many physical scopes have this functionality as well as some blocks in GNU Radio
 CHAPTER 7. BASEBAND DATA TRANSMISSION 423

(a) Time domain RC pulse shape (b) Frequency domain RRC pulse-
shape

Figure 7.20: Eye diagrams for various pulse shapes

INSTRUCTOR COPY

In practice, these eye diagrams are further modified by the channel filter, noise
and interference, resulting in a general eye diagram as shown below.

Figure 7.21: General eye diagram

The eye must be “open” to achieve a low probability of decision error. The decision
device samples this waveform at the time of maximum amplitude for each symbol and
then decides what is the value of the symbol (1 or -1 for a binary two level eye diagram
above, or ±1, ± 3 for a 4 level eye diagram as in fig. 7.22.
 CHAPTER 7. BASEBAND DATA TRANSMISSION 424
INSTRUCTOR COPY

Figure 7.22: Eye diagram for 4 level waveform

An eye diagram with some noise is shown in fig. 7.23.

Figure 7.23: Noisy eye diagram plotted in GNU Radio

An eye diagram may be generated by plotting the message waveform

X
m(t) = Ak p(t − kT ), 0 ≤ k ≤ N
k

(not the individual delayed pulses) over N + 1 symbol periods for a pulse of length
N T and for all possible data patterns of N + 1 symbols.

For example, consider the pulse shape p2 (t) of length 2T . The eye diagram is
obtained by plotting m(t) for all data patterns 000, 001, 010, 011, 100, 101, 110,
111 and overlaying them modulo T to obtain the eye diagram. Note that the data
Ak = ±1 depending of whether the data is 0 or 1.
 CHAPTER 7. BASEBAND DATA TRANSMISSION 425

Matlab has functions eyediagram(x,n) and comm.EyeDiagram to generate an eye

diagram.

The eye diagram in fig. 7.23 was plotted in GNU Radio using fig. 7.24 below.
INSTRUCTOR COPY

Figure 7.24: GNU Radio block to generate eye diagram

This completes the discussion of baseband data transmission, digital messages,

pulse shaping and eye diagrams.

7.6 Chapter review questions

Q 7.1 Differentially encode the message 1110010.

Solution

1 1 1 0 0 1 0

1 0 1 0 0 0 1 1

Q 7.2 Repeat the previous question using a different first bit.

 CHAPTER 7. BASEBAND DATA TRANSMISSION 426

Solution

1 1 1 0 0 1 0

0 1 0 1 1 1 0 0

Q 7.3 Differentially decode your answers to both Q1a and Q1b and show that they
are the same.

Solution
INSTRUCTOR COPY

1 0 1 0 0 0 1 1

1 1 1 0 0 1 0

0 1 0 1 1 1 0 0

1 1 1 0 0 1 0

Q 7.4 The following question asks you to sketch some waveforms. You may find
it helpful to build GNU Radio flowgraphs (like those in chapter 7 and Lab 4 with
either a GLFSR Source block of degree 3 or a Vector Source block with the desired
bit sequence set to repeat. Alternately, write Matlab code following the equations in
chapter 7.

Sketch the baseband waveform for a message 1110010 generated using RC2 time
domain pulses.

Solution
See fig. 7.13.The solution is the same except using a different data sequence.
A few important things to notice: The pulse shape goes from y = 0 → 1 while
the waveform goes from y = −1 → 1. Each 0 bit was turned into a −1 and
when filtered with the pulse shape results in the pulse mirrored about y = 0.
Also see that the impulses and pulse-shaped waveform are not aligned - there
is a delay caused by the filtering. It is the width of the pulse (2T ) divided by
2. To align the impulses and the waveform you can shift the impulses “right”
 CHAPTER 7. BASEBAND DATA TRANSMISSION 427

by T .
The pulse shape here is
(


1 πt
2
1 − cos T
, 0 < t < 2T
h(t) =
0, otherwise

Q 7.5 Sketch the baseband waveform for a message 1110010 generated using RC
frequency domain pulses with β = 1 truncated to 6T .

Solution
See fig. 7.16. The solution is the same except using a different data sequence.
INSTRUCTOR COPY

In this case the delay between impulses and pulse-shaped waveform is 3T . Note
as well that although difficult to resolve at this scale, the pulse shape has zero
crossings at all nT for all integers n 6= 0. One of the advantages of this is
zero-ISI (inter-symbol interference).
The pulse shape here is
 

t cos πβt
T
h(t) = sinc

T 1 − 2βt 2
T

which is in the textbook (here with β = 1 and truncated from −3T → 3T ).

 Chapter 8

Phase shift keying (PSK)

INSTRUCTOR COPY

8.1 PSK introduction

Phase shift keying is a digital form of general phase modulation, where the carrier
phase is modulated in step with the message waveform. The word “keying” is his-
torical and refers to the on-off keying done by a Morse code key acting as a switch
(fig. 8.1).

Figure 8.1: Morse code key

For phase modulation, the complex envelope

s̃(t) =a(t)ejφ
=a(t) cos φ + ja(t) sin φ
=i(t) + jq(t)
has a constant amplitude a(t) = Ac and time varying phase φ(t).

428
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 429

For digital phase modulation, φ(t) is selected from a finite number of discrete
values at the discrete symbol times kT The discrete values of the complex envelope
at the times kT plotted in the 2-D complex plane is called a signal constellation.

In general, φ(t) will assume other values for times t in between the discrete symbol
times kT .

Binary phase shift keying (BPSK) uses two discrete phase values, and Quadrature
phase shift keying (QPSK) uses four discrete phase values. In general, n-PSK uses n
discrete phase values.

In practice, as will be seen later, for BPSK and QPSK with any pulse shape other
than a square pulse, both the amplitude and phase will vary with time.
INSTRUCTOR COPY

PSK can be combined with amplitude shift keying (ASK) to yield amplitude-phase
keying (APK). One form of APK is Quadrature Amplitude Modulation (QAM). One
example is 16QAM that uses 16 discrete amplitude and phase values.

For APK, both a(t) and φ(t) are selected from a finite number of discrete values
at the discrete symbol times kT . In other words, the complex envelope
s̃(t) = a(t)ejφ(t)
takes on only discrete values at times t = kT .

The signal constellation for 16QAM is shown in fig. 8.2. The signal constellation
has 12 discrete phase values and 3 discrete amplitude values. Each point in the signal
constellation represents 4 bits of information. The signal constellation is defined
below in terms of the complex baseband signal
s̃(t) = a(t)ejφ(t) .

Figure 8.2: Signal constellation for 16 QAM

 CHAPTER 8. PHASE SHIFT KEYING (PSK) 430

Recall the 3-D representation of a complex baseband signal

s̃(t) = i(t) + jq(t)
where the 3 dimensions are i(t), q(t), t. The signal constellation is the projection of
the 3-D signal onto the complex plane with the x-axis representing i(t) and the y-axis
representing q(t) and t is perpendicular to the plane.

Just as for the eye diagram, the full signal constellation is seen when the signal is
viewed in continuous (not discrete) time t over all combinations of bits, see fig. 8.18.
The signal constellation diagram is obtained by observing the complex waveform s̃(t)
for a sufficiently long time to include all data patterns while projecting s̃(t) onto the
complex i-q plane.

Recall also that when signals that are written mathematically as if they were
INSTRUCTOR COPY

continuous time signals s̃(t), that in a software implementation, t = nts where fs =

1/ts is a sampling rate sufficiently high to represent the signal bandwidth.

In chapter 7 we wrote a real baseband digital message as

X
m(t) = Ak p(t − kT )
k

where Ak are real numbers representing data.

In this chapter, we extend this idea to a complex baseband digital message s̃(t)
with two separate messages m1 (t), m2 (t)
s̃(t) = m1 (t) + jm2 (t)
X
= Ck p(t − kT )
k

where
Ck = ak ejφk
= Ak + jBk
is now complex, but the pulse shape p(t) is still a real function. When s̃(t) is sampled
at a particular time t = qT + 
X
s̃(qT + ) = Ck p(qT +  − kT )
k
= Cq p()
= Cq
= aq ejφq
= Aq + jBq
In the above equation
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 431

• Ck is written in rectangular and polar forms

•  is an offset that is chosen to be the value of t where p(t) reaches its maximum
value p(t = ) = 1. The RC2 pulse p2 (t) in chapter 7 is a example where  = T .
For a square pulse of length T the optimum sampling time may be any value
0 <  < T.

• The summation over index k disappears because the pulse shape is chosen so
there is no inter-symbol interference, i.e. all other delayed pulses with index
k 6= q are zero at the sampling time t = qT + .

• For further detail, see section 8.6.1

Exercise 122
INSTRUCTOR COPY

Show that for an RC2 pulse p2 (t) the optimum sampling time  = T .

Solution
For an RC2 pulse
 
πt
p2 (t) = 0.5 1 − cos
T
p2 (t =  = T ) = 0.5(1 − cos π) = 1
p2 (t 6= T ) < 1

There are two versions of the signal constellation

• continuous time projection of s̃(t) (look ahead to fig. 8.18)

• discrete time projection sampled at the symbol rate 1/T at discrete times t =
kT +  to yield discrete complex points Ak = ak ejφk = ik + jqk (fig. 8.2).

The discrete time projection may also be viewed as follows. At each sampling
time t = kT + , there is a complex plane corresponding to that time k, and the
signal s̃(kT + ) defines a single point

Ak = ak ejφk
= ik + jqk

in the complex plane. This point is the intersection between the complex plane at
time t = kT +  and the 3D plot of s̃(t) and is the complex number s̃(kT + ). In
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 432

general, a different value of k will define a different point. The signal constellation is
the sum of all such points.

For the example of 16QAM in fig. 8.2, there are 16 possible points Ck = ak ejφk
that can arise at time k. The values of ik and qk for each time k are one of ±1, ±3.
The sum of all these points is the signal constellation.

In the sections that follow, we consider specific choices of the values

Ck = ak ejφk
= ik + jqk

that are used to transmit digital data.

INSTRUCTOR COPY

8.2 BPSK Transmitter

In this section, we show the formulas for a BSPK signal derived from the general
complex signal notation. We also show how BPSK is a special case of DSB-SC.

A general signal

s(t) =<{a(t)ejφ ej2πfc t }

=<{s̃(t)ej2πfc t }
=<{s+ (t)}

where the complex envelope

s̃(t) =a(t)ejφ
=a(t) cos φ + ja(t) sin φ
=i(t) + jq(t)

and i(t), q(t) are obtained from the message signal.

Recall that for real passband DSB-SC, we choose i(t) = m(t) and q(t) = 0 , so
that

s(t) =<{s̃(t)ej2πfc t }
=<{[i(t) + jq(t)]ej2πfc t }
=<{m(t)ej2πfc t }
=m(t) cos 2πfc t
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 433

For DSB-SC, s̃(t) = s̃m (t) = m(t), i.e. the complex envelope is equal to the (real)
message. Here we define s̃m (t) to be the complex baseband signal associated with the
message.

We can choose m(t) = cos 2πfm t to represent a 10 alternating digital message

during one cycle of the cos wave, representing a 1 bit when m(t) = 1 and a 0 bit
when m(t) = −1. Thus fm = 1/2T and T is the symbol time.

From the expression for real passband DSB, we observed in section 4.2.1.4 that
the phase of the carrier changes by π when the message waveform transitions from
positive to negative and vica versa. Thus DSB is a kind of PSK.

We could also choose m(t) to be a square wave with period 2T , see fig. 8.3. The
frequency of the cos (or square) wave is one-half the symbol rate.
INSTRUCTOR COPY

Figure 8.3: Square wave

For binary phase shift keying (BPSK), we also choose i(t) = m(t) and q(t) = 0
and we choose

m(t) =A0 p(t) + A1 p(t − T ) + A2 p(t − 2T ) + . . .

X
= Ak p(t − kT )
k

If p(t) is chosen to be a square pulse and

(
+1, i is even
Ai =
−1, i is odd

then m(t) is a square wave with period 2T .

 CHAPTER 8. PHASE SHIFT KEYING (PSK) 434

The BPSK signal is written

s(t) =Re[m(t)ej2πfc t ]
=m(t) cos 2πfc t
X
= Ak p(t − kT ) cos 2πfc t
k

For BPSK, s̃m (t) = m(t), i.e. the complex envelope is equal to the message. The
real passband BPSK is created in the same way as DSB-SC, i.e. multiply the message
by the carrier.

Exercise 123
INSTRUCTOR COPY

Sketch the BPSK real passband waveform for a digital message 1011010.

Solution
See fig. 8.4.

Figure 8.4: Message, carrier and final BPSK modulated waveform

Exercise 124
Sketch the signal constellation for BPSK.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 435

Solution
In polar form the signal constellation Ak = ak ejφk , so that with Ak = ±1 ,
ak = 1 and φk = nπ are the 2 possible phases of a BPSK signal. Thus the
signal constellation for BPSK consists of two points located at Ak = ±1, one
at +1 and one at -1.

The spectrum of the BPSK signal S̃m (f ) will depend on the data sequence and
pulse shape

i(t) =m(t)
X
= Ak p(t − kT ).
k
INSTRUCTOR COPY

Exercise 125
What is the BPSK spectrum for a RC2 pulse shape and the data sequence is
alternating 1010

Solution
For an alternating sequence
(
+1, k is odd
Ak =
−1, k is even

with time domain RC2 raised cosine pulse shape

p(t) = p2 (t) = [1 − cos(πt/T )]/2

X
m(t) = Ak p(t − kT )
k
= cos πt/T = cos 2πfm t

where fm = 1/2T . Thus the BPSK spectrum is the same as for DSB-SC with
single tone modulation with delta function spikes at fc ± fm = fc ± 1/2T .

In general, the data sequence is random and can be modelled as the output of a
linear feedback shift register. In this case, S̃m (f ) for BPSK is shown in fig. 8.5 with
spectral nulls at multiples of 1/T .
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 436
INSTRUCTOR COPY

Figure 8.5: BPSK spectrum

The spectrum may be written as shown in chapter 2

2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))

where for BPSK S̃(f ) = S̃m (f ).

This equation may be interpreted to show that any real passband signal is the sum
of the complex baseband spectrum S̃(f ) shifted by +fc and the flipped (mirror-image)
complex baseband spectrum S̃ ∗ (−f ) shifted by −fc .

In the following 3 subsections, three versions of BPSK signals are described:

• real baseband s̃m (t) and real passband s(t) generated as sampled signals by
software within GNURadio, with both positive and negative frequencies
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 437

• complex passband s+ (t) generated as sampled signals by software within GNU-

Radio, with positive frequencies only

• any of the above signals fed to a URSP Sink block and transmitted as a real
signal s(t) by the USRP.

8.2.1 Real BPSK signal

To create a real BPSK signal in GRC at intermediate frequency f1 using these equa-
tions, we carry out two steps:
INSTRUCTOR COPY

1. Generate the real baseband signal

s̃m (t) =i(t)

=m(t)
X
= Ak p(t − kT )
k

The notation s̃m (t) = i(t) + jq(t) represents the complex baseband message
which is real in this case with i(t) = m(t) and q(t) = 0.

2. Upconvert it to the desired carrier frequency by multiplying by cos 2πf1 t =

0.5{ej2πf1 t + e−j2πf1 t } to yield s̃(t) = Re[m(t)ej2πf1 t ] = m(t) cos 2πf1 t
In a GNURadio implementation, the real BPSK signal is one input of a float to
complex block, with the second input set to zero (null source), and the resulting
complex signal output is s̃m (t) (with the imaginary part set to zero)

m(t) → ⊗ → s(t)

cos 2πf1 t

The spectrum of s(t) will be centered at ±f1 and has both positive and negative
frequency components.

A GRC flowgraph for real BPSK is shown in fig. 8.6.

 CHAPTER 8. PHASE SHIFT KEYING (PSK) 438
INSTRUCTOR COPY

Figure 8.6: GNU Radio real BPSK modulator

The real BPSK signal can also be generated as follows, recalling s̃m (t) = i(t)+jq(t)
with i(t) = m(t) and q(t) = 0.

m(t) → ⊗ ⇒ s̃1 (t) ⇒ Re → i(t) cos 2πf1 t

e2πf1 t

If we consider only one of the pulses in the message at time k, and consider a
square pulse of length T , then we can write

m(t) =i(t)
=Ak p(t − kT )
s̃m (t) =m(t) cos 2πf1 t

where (
1, 0 ≤ t ≤ T
p(t) =
0, otherwise
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 439

Exercise 126
For a square pulse of length T , write the real passband BPSK signal in terms
of its amplitude and phase

Solution

s(t) =Ak cos(2πf1 t + φk ) for kT ≤ t ≤ (k + 1)T

= cos(2πf1 t + nπ) for kT ≤ t ≤ (k + 1)T

where Ak = 1, φk = nπ. Thus n = 0 when Ak = 1 and n = 1 when Ak = −1.

INSTRUCTOR COPY

8.2.2 Complex BPSK signal

To create a complex (analytic) BPSK signal in software within GNURadio or GRC

at intermediate frequency f1 using these equations, we carry out two steps:

1. Generate the real baseband signal

s̃m (t) =i(t)

=m(t)
X
= Ak p(t − kT )
k

2. Upconvert it to the desired carrier frequency by multiplying by ej2πf1 t . The

result is a complex (analytic) signal.

s̃1 (t) =s+ (t)

=i(t)ej2πf1 t
=i(t) cos 2πf1 t + j[i(t) sin 2πf1 t]
=i1 (t) + jq1 (t)

m(t) → ⊗ → s̃1 (t) = i1 (t) + jq1 (t) = m(t)ej2πf1 t

ej2πf1 t
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 440

Since s̃1 (t) is a sampled complex signal inside GRC that includes a carrier wave at
f1 and contains only positive frequencies, it could also be written as s+ (t). However,
we will use the notation s̃1 (t) since there could be other signals with carrier waves at
other frequencies aside from f1 .

The spectrum of s̃1 (t) will be centered at f1 . Since the signal is analytic, there is
no negative frequency component at −f1 .

A GRC flowgraph is shown in fig. 8.7, where the message is a square wave 1010
data sequence. In this flowgraph, the message is set to be s̃m (t) = m(t) + jm(t) but
could also have been set to m(t) + j0 by using a null source connected to the lower
input of the Float to Complex block.
INSTRUCTOR COPY

Figure 8.7: GNU Radio complex BPSK modulator

Exercise 127
What is the practical result of setting the message s̃m (t) = m(t)+jm(t) instead
of s̃m (t) = m(t) + j0

Solution
√
m(t) + jm(t) = 2m(t)ejπ/4
The constellation is rotated by π/4.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 441

8.2.3 BPSK at radio frequency (RF)

The USRP sink block with complex input s̃1 (t) is used to transmit the BPSK signal
near a radio frequency (RF) fc . The USRP transmitter multiplies s̃1 (t) = i1 (t)+jq1 (t)
by ej2πfc t and takes the real part to generate a real RF signal centered at f1 + fc and
−f1 − fc .

BPSK at RF is generated as follows. First generate the real baseband signal

s̃m (t) =i(t)

=m(t)
X
= Ak p(t − kT )
INSTRUCTOR COPY

Then carry out the steps in the diagram below.

m(t) → ⊗ ⇒ s̃1 (t) = i1 (t) + jq1 (t) ⇒ ⊗ ⇒ Re → m(t) cos 2π(fc + f1 )t

⇑ ⇑

ej2πf1 t ej2πfc t

The USRP sink block does the second complex multiply and taking the real part.

The GRC flowgraph is the same as the complex BPSK above with a USRP sink
block in place of the Virtual Sink.

The real RF signal s(t) is centered at f1 + fc and −f1 − fc . Since the signal is
real, there are both positive and negative frequency components.

Exercise 128
Find an expression for the real BPSK signal at RF
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 442

Solution
In terms of sines and cosines

s(t) =<{s̃1 (t)ej2πfc t }

=<{[i1 (t) + jq1 (t)][cos(2πfc t) + j sin(2πfc t)]}
=i1 (t) cos(2πfc t) − q1 (t) sin(2πfc t)
=i(t) cos(2πf1 t) cos(2πfc t) − i(t) sin(2πf1 t) sin(2πfc t)
=i(t) cos 2π(fc + f1 )t

Alternately, in terms of exponentials

s(t) =<{s̃1 (t)ej2πfc t }

=<{i(t)ej2πf1 t ej2πfc t }
INSTRUCTOR COPY

=<{i(t)ej2π(f1 +fc )t }
=i(t) cos 2π(fc + f1 )t

Exercise 129
To transmit the BPSK signal at fc , the USRP sink block could also have used
the complex input

s̃m (t) =i(t)

=m(t)

with the imaginary part set to zero and without using the exponential at f1 .
In this case the USRP transmitter multiplies s̃m (t) by ej2πfc t and takes the real
part to generate a real RF signal centered at fc and −fc .
Why is this approach to transmitting a BPSK signal not recommended? Hint:
the reason is a practical one.

Solution
s̃m (t) = m(t) may contain DC, and the USRP does not pass DC. The complex
exponential at f1 ensures that there is no DC in s̃m (t).
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 443

8.3 BPSK Receiver

To receive the BPSK signal, the USRP source block is used to downconvert the RF
signal to complex baseband i(t) + jq(t).

We can also simulate the BPSK receiver within GNU Radio without using the
USRP source block. The simulation is done by using the BPSK signal generated at
intermediate frequency f1 .

In the following three subsections, three BPSK receivers are described:

• real passband r(t) received as sampled signals by software within GNURadio,

with both positive and negative frequencies, and not using USRP hardware.
INSTRUCTOR COPY

• complex passband r+ (t) received as sampled signals by software within GNU-

Radio, with positive frequencies only, and not using USRP hardware.

• a real passband signal r(t) received by the USRP and presented as complex
baseband or complex passband signal i(t)+jq(t) at the output of a URSP Source
block.

– In the ideal case with no noise, frequency or phase offsets and if the USRP
was able to pass DC, then the complex baseband will be the original
message s̃m (t) = i(t) = m(t). In practice, the complex signal will be
r+ (t) = m(t)ej2πf1 t at some frequency offset f1 and require further fre-
quency shifting in software.

8.3.1 Real BPSK signal receiver

We first work with the real BPSK signal as a sampled signal within GNURadio
software without using USRP hardware

r(t) =<[m(t)ej2πf1 t ]
=m(t) cos 2πf1 t.

We use a standard IQ receiver set up for f1 with complex output i(t) + jq(t). If
there is no frequency or phase offset, then q(t) = 0.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 444

r(t) →⇒ ⊗ ⇒ s̃1 (t) ⇒ LP F ⇒ i(t) + jq(t)

e−j2πf1 t

The IQ receiver above is implemented with a complex multiplier. Low pass filters
will be required in the complex case, since the input s(t) is real and has both positive
and negative frequencies. The positive frequency f1 in s(t) is downconverted to zero
frequency, and the negative frequency −f1 is downconverted to −2f1 and filtered out.

The IQ receiver can also be implemented with real cosines and sines and low pass
INSTRUCTOR COPY

filters.

With a suitable low pass filter, the received data i(t) + jq(t) at sampling times
t = kT + τ is Ak = ±1 (real), where τ is adjusted to sample in the middle of the
pulse p(t).

τ is a timing offset to sample the data away from the transitions between different
bits. The timing offset τ is sometimes also notated as .

The simplest receiver assumes there is no frequency or phase offset, and can be
implemented with a real (cosine) local oscillator instead of a complex (exponential)
oscillator.

r(t) →⇒ ⊗ ⇒ LP F ⇒ i(t)

cos 2πf1 t

In a GNURadio implementation of this IQ receiver, the real BPSK signal is one

input of a float to complex block, with the second input set to zero (null source),
and the resulting complex signal output s̃(t) (with the imaginary part set to zero) is
connected to one input of a complex multiplier.

The GRC example implementation shown in fig. 8.8 uses a square wave local
oscillator in place of cos 2πf1 t.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 445

Figure 8.8: GNU Radio real BPSK receiver implementation

Exercise 130
Why is it acceptable to use a square wave local oscillator in place of a cosine
INSTRUCTOR COPY

local oscillator in this GRC flowgraph?

Solution
The square wave local oscillator is the sum of harmonics of the fundamental
frequency f1 (consider a Fourier series for the square wave). These harmonics
are filtered out by the low pass filter.

8.3.2 Complex BPSK signal receiver

In this section we receive a complex BPSK analytic signal r+ (t) = r̃1 (t) with only
positive frequencies

r+ (t) = r̃1 (t) =m(t)ej2πf1 t

=m(t) cos 2πf1 t + j[m(t) sin 2πf1 t]
=i1 (t) + jq1 (t)

Again, we use a standard IQ receiver with complex local oscillator. In this case,
we can use the complex IQ receiver that multiplies s̃1 (t) by e−j2πf1 t (note the minus
sign for downconversion). A LPF is not needed. If there is no frequency or phase
offset, then q(t) = 0 and i(t) = m(t).
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 446

s̃1 (t) ⇒ ⊗ ⇒ i(t) + jq(t)

e−j2πf1 t

The GRC flowgraph in fig. 8.9 uses a complex local oscillator.

INSTRUCTOR COPY

Figure 8.9: GNU Radio complex BPSK receiver

8.3.3 BPSK receiver at RF

A BPSK receiver at RF using the USRP would have a GRC flowgraph as fig. 8.9,
except that the virtual source is replaced with a USRP source.

Exercise 131
Why can the virtual source be replaced by a USRP source?

Solution
The output is both cases is an analytic signal.

In the diagram below, the USRP source block carries out the complex downcon-
version with local oscillator at fc to produce the complex baseband signal r̃1 (t) that
is subsequently processed by the GRC flowgraph.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 447

r(t) → ⊗ ⇒ LP F ⇒ r̃1 (t) ⇒ ⊗ ⇒ i(t) + jq(t)

⇑ ⇑

e−j2πfc t e−j2πf1 t

If there is no frequency or phase offset, then q(t) = 0 and i(t) = m(t).

The USRP source block actually does two complex downconversions, coarse and
fine, as shown in fig. 2.30, but these two conversions are combined into one in the
above diagram.
INSTRUCTOR COPY

The Gigabit Ethernet output contains the I and Q samples of r̃1 (t).

8.4 QPSK Transmitter

Recall that we can send separate messages m1 (t), m2 (t) on the I and Q channels, i.e.
i(t) = m1 (t), q(t) = m2 (t).

We can choose m1 (t) and m2 (t) to be separate digital messages. When we use
both the I and Q channels, the modulation is called Quadrature Phase Shift Keying
(QPSK).

To write an expression for QPSK, we can use the same general form as for BPSK
above, except that the constants representing the data are now complex instead of
real.

A digital message can be written as a complex baseband signal

X
s̃m (t) = Ck p(t − kT )
k
=C0 p(t) + C1 p(t − T ) + C2 p(t − 2T ) + . . .

where s̃m (t) = m1 (t) + jm2 (t) is a complex message waveform. The digital data is

Ck =Ak + jBk
=ak ejφk

where Ak = ak cos φk and Bk = ak sin φk .

 CHAPTER 8. PHASE SHIFT KEYING (PSK) 448

If we choose the pulse shape p(t) to be a square pulse of length T , then we can
sample the signal anytime during 0 < t < T . Thus from the expression for s̃m (t), we
see that for the period 0 < t < T ,

s̃m (t) =m1 (t) + jm2 (t)

=A0 + jB0 = C0
=a0 ejφ0

is a complex constant. For the next period T < t < 2T ,

s̃m (t) =A1 + jB1 = C1

=a1 ejφ1
INSTRUCTOR COPY

In general, for the time period kT < t < (k + 1)T ,

s̃m (t) =Ak + jBk = Ck

=ak ejφk

is a fixed amplitude and phase.

If we choose the pulse shape to be RC2 p2 (t) then we obtain the same results for
s̃m (t) but only if we sample at one specific time t = kT + .

For QPSK, we choose Ak = ±1 and Bk = ±1, so that Ck = ±1 ± j. There are 4

possible values of Ck depending on whether a logic 00, 01, 10 or 11 was sent.

The signal constellation for QPSK consists of the 4 points located at Ck = ±1 ± j.

In polar form this is

Ck =Ak + jBk
=ak ejφk

so that

Ak = ±1
Bk = ±1
√
ak = 2
(2n − 1)π
φk =
4
= ±π/4, ±3π/4

are the 4 possible phases of a QPSK signal.

 CHAPTER 8. PHASE SHIFT KEYING (PSK) 449

The waveforms are represented as shown in fig. 8.10 in both complex baseband and
real passband. In this figure, the 4 phases are rotated by π/4 so that φk = (2n−1)π
4
+ π4 .
INSTRUCTOR COPY

Figure 8.10: QPSK waveforms

Exercise 132
Write s̃m (t) = m1 (t) + jm2 (t) in rectangular form and evaluate for the time
period kT < t < (k + 1)T assuming p(t) is a square pulse of length T .
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 450

Solution
In rectangular form, s̃m (t) = m1 (t) + jm2 (t), where
X
m1 (t) = Ak p(t − kT )
k
=A0 p(t) + A1 p(t − T ) + A2 p(t − 2T ) + . . .
m1 (kT < t < (k + 1)T ) =Ak

and
X
m2 (t) = Bk p(t − kT )
k
=B0 p(t) + B1 p(t − T ) + B2 p(t − 2T ) + . . .
INSTRUCTOR COPY

m2 (kT < t < (k + 1)T ) =Bk

so that

s̃m (t) = m1 (t) + jm2 (t)

= (A0 + jB0 )p(t) + (A1 + jB1 )p(t − T ) + (A2 + jB2 )p(t − 2T ) + . . .
= C0 p(t) + C1 p(t − T ) + C2 p(t − 2T ) + . . .

The pulses p(t − kT ) define a finite time interval kT < t < (k + 1)T where it
is non-zero. s̃m (t) is uniquely defined for all times t to be a specific value Ck
that may change for different values of k. When s̃m (t) is plotted in 3D, then
the waveform will look like the 2nd-to-last row in fig. 8.10

Exercise 133
What are the possible values of Ck in fig. 8.10

Solution

Ck = +1, −1, +j, −j (8.1)

(2n − 1)π π
φk = + (8.2)
4 4
nπ
= (8.3)
2
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 451

Exercise 134
Sketch a figure similar to fig. 8.10 for the case

Ck =Ak + jBk
=ak ejφk
(2n − 1)π
φk =
4

Solution
The phase φk is rotated by π/4. A partial solution hint is given in fig. 8.12 in
the next section.
INSTRUCTOR COPY

In the following 3 subsections, three versions of QPSK signals are described:

• real baseband s̃m (t) and real passband s(t) generated as sampled signals by
software within GNURadio, with both positive and negative frequencies

• complex passband (analytic) s+ (t) generated as sampled signals by software

within GNURadio, with positive frequencies only

• any of the above signals fed to a URSP Sink block and transmitted as a real
signal s(t) by the USRP.

8.4.1 Real QPSK signal

To create a real QPSK signal in GRC at intermediate frequency f2 using these equa-
tions, one method is to use real notation and generate the complex baseband signal
s̃m (t) = m1 (t) + jm2 (t) and upconvert it to the desired carrier frequency f2 with a
standard (real) IQ transmitter (fig. 8.11).
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 452

Figure 8.11: Real QPSK transmitter

INSTRUCTOR COPY

Exercise 135
Find an expression for the real passband QPSK signal s(t) for the time period
kT < t < (k + 1)T based on fig. 8.11.

Solution
Recall that for the time period kT < t < (k + 1)T ,

s̃m (t) =m1 (t) + jm2 (t)

=Ak + jBk = Ck
=ak ejφk
m1 (t) =Ak
m2 (t) =Bk

From the figure

s(t) =m1 (t) cos 2πf2 t − m2 (t) sin 2πf2 t

=Ak cos 2πf2 t − Bk sin 2πf2 t
=ak cos(2πf2 t + φk )
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 453

Substituting from the previous section

Ck =Ak + jBk
=ak ejφk
(2n − 1)π
φk =
4  
√ (2n − 1)π
s(t) = 2 cos 2πf2 t +
4

for kT ≤ t ≤ (k + 1)T and where the value of n is chosen to represent the

phase at time sample k. This real signal contains both positive and negative
frequencies.
INSTRUCTOR COPY

The real passband QPSK signal s(t) is shown in fig. 8.12 for specific data values
for 4 symbol periods. Note that for I(t) data logic 1 bit corresponds to Ak = 1 and
data logic 0 bit corresponds to Ak = −1 and similarly for Q(t).

Figure 8.12: QPSK waveform

In fig. 8.12, the waveforms represent I(t), Q(t), s(t) for 4 symbol periods kT ≤ t ≤
(k + 4)T . For the first symbol period kT ≤ t ≤ (k + 1)T

• the top waveform is the first term in s(t)

I(t) = m1 (t) cos 2πf2 t = Ak p(t − kT ) cos 2πf2 t = Ak cos 2πf2 t

• the middle waveform is the second term in s(t)

Q(t) = −m2 (t) sin 2πf2 t = −Bk p(t − kT ) sin 2πf2 t = −Bk sin 2πf2 t
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 454

• the bottom waveform is the complete real QPSK signal s(t).

• Recall that for kT ≤ t ≤ (k + 1)T , p(t − kT ) = 1.

Exercise 136
Write an expression for I(t) and Q(t) that is valid for the 4 symbol periods
kT ≤ t ≤ (k + 4)T shown in fig. 8.12

Solution
Substitute the expression for m1 (t) and m2 (t) obtained in a previous exercise
to obtain
X
INSTRUCTOR COPY

I(t) = Ak p(t − kT ) cos 2πf2 t

k
X
Q(t) = − Bk p(t − kT ) sin 2πf2 t
k

Note that for the components I(t), Q(t), the starting and ending phases are at
multiples of π/2, whereas for the sum of these components (the QPSK signal), the
starting and ending phase of the sine waves are at odd multiples of π/4.

The spectrum of s(t) will be centered at ±f2 and has both positive and negative
frequency components.

A GRC flowgraph of a real QPSK transmitter is shown in fig. 8.13. The result at
the Virtual Sink output is

s(t) = m1 (t) cos 2πf2 t − m2 (t) sin 2πf2 t

 CHAPTER 8. PHASE SHIFT KEYING (PSK) 455
INSTRUCTOR COPY

Figure 8.13: GNU Radio real QPSK transmitter

Exercise 137
Show that the real QPSK signal can also be generated using complex nota-
tion with the following transmitter diagram. Hint: find an expression for the
transmitter output s(t).

m1 (t) + jm2 (t) = s̃m (t) ⇒ ⊗ ⇒ s̃2 (t) ⇒ Re → s(t)

ej2πf2 t

8.4.2 Complex (analytic) QPSK signal

To create an analytic QPSK signal in GRC at intermediate frequency f2 using these

equations, we create the complex baseband message s̃m (t) = m1 (t) + jm2 (t) and
upconvert it to the desired carrier frequency by multiplying by ej2πf2 t . The result is
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 456

another complex signal s+ (t) = s̃2 (t) with positive frequencies only

s̃2 (t) =[m1 (t) + jm2 (t)]ej2πf2 t

=m1 (t) cos 2πf2 t − m2 (t) sin 2πf2 t + j[m2 (t) cos 2πf2 t + m1 (t) sin 2πf2 t]
=i2 (t) + jq2 (t)

The complex QPSK signal is generated as follows

m1 (t) + jm2 (t) = s̃m (t) ⇒ ⊗ ⇒ s̃2 (t)

⇑
INSTRUCTOR COPY

ej2πf2 t

A GRC flowgraph is shown in fig. 8.14. The complex message m1 (t) + jm2 (t) is
multiplied by a complex local oscillator ej2πf2 t .

Figure 8.14: GNU Radio complex QPSK transmitter

We can create a second QPSK signal at a different intermediate frequency f3 in

the same way to yield
s̃3 (t) = i3 (t) + jq3 (t).

We can create a composite signal consisting of several QPSK signals at f2 , f3 etc.

 CHAPTER 8. PHASE SHIFT KEYING (PSK) 457

8.4.3 QPSK at radio frequency (RF)

The upconversion of a QPSK signal to a radio frequency (RF) wave at fc is the

function of the USRP sink block (standard IQ transmitter) with inputs i2 (t) and
q2 (t).

The USRP transmitter accepts a complex baseband signal s̃2 (t) and multiplies it
by ej2πfc t and takes the real part to generate a real RF signal s(t).

s̃m (t) ⇒ ⊗ ⇒ s̃2 (t) ⇒ ⊗ ⇒ Re → s(t)

⇑ ⇑
INSTRUCTOR COPY

ej2πf2 t ej2πfc t

Exercise 138
Find an expression for the transmitter output s(t).

Solution

s̃m (t) = m1 (t) + jm2 (t)

s̃2 (t) = s̃m (t)ej2πf2 t
= i2 (t) + jq2 (t)
s(t) = <{s̃m (t)ej2πf2 t ej2πfc t }
= m1 (t) cos 2π(fc + f2 )t − m2 (t) sin 2π(fc + f2 )t

In GNU Radio, the USRP Sink block does the second complex multiply and takes
the real part. The GRC flowgraph is the same as the complex QPSK (fig. 8.14) with
a USRP Sink block in place of the Virtual Sink.

Similarly, the entire group of QPSK signals s̃2 (t) + s̃3 (t) can be upconverted to
RF by the USRP Sink block.

The USRP Sink block could also have used the complex input
s̃m (t) = m1 (t) + jm2 (t)
to transmit the QPSK signal at fc . In this case the USRP transmitter multiplies
s̃m (t) by ej2πfc t and takes the real part to generate a real RF signal centered at fc
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 458

and −fc . However, this practice is not recommended, since s̃m (t) may contain DC,
and the USRP does not pass DC.

8.5 QPSK Receiver

To receive a QPSK signal in GNU Radio using a USRP, the USRP Source block is
used to downconvert the RF signal to complex baseband i(t) + jq(t). We can also
simulate the QPSK receiver within GNU Radio without using the USRP Source block.

The simulation is done by using the QPSK signal generated at intermediate fre-
quency f2 .
INSTRUCTOR COPY

In the following three subsections, three QPSK receivers are described:

• real passband r(t) received as sampled signals by software within GNURadio,

with both positive and negative frequencies

• complex passband r+ (t) received as sampled signals by software within GNU-

Radio, with positive frequencies only

• a real passband signal r(t) received by the USRP and presented as complex
baseband or complex passband signal i(t)+jq(t) at the output of a URSP Source
block.

– In the ideal case with no noise, frequency or phase offsets and if the USRP
was able to pass DC, then the complex baseband will be the original mes-
sage s̃m (t). In practice, the complex signal will be r+ (t) = s̃m (t)ej2πf1 t at
some frequency offset f1 and require further frequency shifting in software
to recover the message s̃m (t).

8.5.1 Real QPSK signal

The received real passband QPSK signal

r(t) = m1 (t) cos 2πf2 t − m2 (t) sin 2πf2 t

We use a standard real IQ receiver set up for f2 with real outputs i(t) and q(t) that
may be notated as a complex output i(t)+jq(t). This IQ receiver can be implemented
with real cosines and sines and low pass filters.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 459

Figure 8.15: Real QPSK receiver

INSTRUCTOR COPY

With a suitable low pass filter, the received data i(t) at sampling times t = kT + τ
is Ak = ±1 and the received data q(t) is Bk = ±1, where τ is adjusted to sample
in the middle of the pulse p(t). The received data i(t) + jq(t) at sampling times
t = kT + τ is Ck = ±1 ± j.

The IQ receiver can be also implemented in GNURadio with real cosines and sines
and low pass filters as shown in fig. 8.15. A GRC flowgraph of the real QPSK receiver
is shown in fig. 8.16.

Figure 8.16: GNU Radio real QPSK receiver

An XY scope plot will show the signal constellation which is the projection of the
received complex signal i(t) + jq(t) onto the i, q plane. At sampling times t = kT + τ
the constellation shows all possible complex datat values Ck = ±1 ± j as sh9wn in
fig. 8.17.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 460
INSTRUCTOR COPY

Figure 8.17: QPSK demodulated waveform, constellation, NRZ output

If the signal is not sampled and we display the received complex signal i(t) + jq(t)
at all times t, then the signal constellation shows the paths between the 4 points
Ck = ±1 ± j as shown in fig. 8.18. The exact trajectories (shape of the path curves)
will depend on the pulse shaping filter. If we observe the 3-D time evolution of
i(t)+jq(t), then the projections of that 3-D signal onto the i versus t plane or q versus
t plane will look like the pulse shaped baseband waveforms observed in fig. 7.16.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 461
INSTRUCTOR COPY

Figure 8.18: QPSK constellation showing all points (sampled at all t’s).

Exercise 139
The IQ receiver above may also be implemented with a complex multiplier
as shown below. Low pass filters will be required in the complex case, since
the input r(t) is real and thus contains both positive and negative frequencies.
The positive frequency f2 in r(t) is downconverted to zero frequency, and the
negative frequency −f2 is downconverted to −2f2 and filtered out.

r(t) →⇒ ⊗ ⇒ s̃2 (t) ⇒ LP F ⇒ i(t) + jq(t)

e−j2πf2 t

Find an expression for the receiver output at the correct sampling time to
recover the data.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 462

8.5.2 Complex QPSK signal

Here we receive the complex passband (analytic) QPSK signal r+ (t) = r̃2 (t) with
positive frequencies only

r̃2 (t) =[m1 (t) + jm2 (t)]ej2πf2 t

=m1 (t) cos 2πf2 t − m2 (t) sin 2πf2 t + j[m2 (t) cos 2πf2 t + m1 (t) sin 2πf2 t]
=i2 (t) + jq2 (t)

We use a standard IQ receiver. In this case, we use the complex IQ receiver that
multiplies s̃2 (t) by e−j2πf2 t (note the minus sign for downconversion). An LPF is not
needed because s̃2 (t) contains only positive frequencies.
INSTRUCTOR COPY

r̃2 (t) ⇒ ⊗ ⇒ r̃4 (t) = i4 (t) + jq4 (t)

e−j2πf2 t

Exercise 140
Obtain an expression for r̃4 (t).

Solution

r̃4 (t) =i4 (t) + jq4 (t)

=r̃2 (t)e−j2πf2 t
=[i(t) + jq(t)]ej2πf2 t e−j2πf2 t
=i(t) + jq(t)
=m1 (t) + jm2 (t)

Exercise 141
Show that the received data r̃4 (t) at sampling times t = kT + τ is Ck = ±1 ± j.

A GRC flowgraph is in fig. 8.19. The XY scope mode is used to show the signal
constellation. A practical receiver will include a low pass or RRC matched filter.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 463

Figure 8.19: GNU Radio complex QPSK receiver

8.5.3 QPSK signal at radio frequency (RF)

INSTRUCTOR COPY

A BPSK receiver at RF using the USRP would have a GRC flowgraph as in fig. 8.19,
except that the Virtual Source block is replaced with a USRP Source block.

The USRP receiver multiplies the real valued radio frequency QPSK signal r(t)
at carrier frequency (fc + f2 ) by e−j2πfc t to generate r̃2 (t).

In the diagram below, the USRP source block carries out the complex downcon-
version with local oscillator at fc and low pass filters to produce the complex baseband
signal s̃2 (t) that is subsequently processed by the GRC flowgraph.

r(t) → ⊗ ⇒ LP F ⇒ r̃2 (t) ⇒ ⊗ ⇒ i(t) + jq(t) = m1 (t) + jm2 (t)

⇑ ⇑

e−j2πfc t e−j2πf2 t

The USRP Source block has a complex output r̃2 (t). The USRP Source block
actually does two complex downconversions as was shown earlier in the USRP block
diagram (fig. 2.30). In this diagram, the Gigabit Ethernet output contains the I and
Q samples of s̃2 (t).

Exercise 142
Find an expression for the receiver output at the sampling time needed to
obtain the data.

The receiver input r(t) may contain multiple QPSK signals at different carrier
frequencies f2 (t), f3 (t), etc.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 464

The complex signal output from the USRP Source block r̃2 (t) is bandlimited to
the sampling rate of the USRP Source block. The USRP Source block output can be
recorded to a file and used again at a later time. This file source will have the same
sampling rate and bandwidth as the USRP Source block used to record it.

The receiver can receive only those QPSK signals that are within the bandwidth
of the receiver as determined by the sampling rate.

With a sampling rate of 200 kHz and complex samples, the bandwidth will be
200 kHz (because the complex signal spectrum is not symmetric and does not have
redundant mirror-image positive and negative frequencies, as shown in Chapter 1).

We can “tune into” (receive) any of the QPSK signals in this bandwidth by shifting
the spectrum of the USRP Source output using GRC. We shift by f2 Hz by multiplying
INSTRUCTOR COPY

the USRP Source output (complex signal) r̃2 (t) by e−j2πf2 t = cos 2πf2 t − j sin 2πf2 t
to produce i(t) + jq(t), so that the signal that first appeared at f2 Hz now appears
at 0 Hz.

8.6 QPSK with amplitude, frequency and phase

errors in the receiver

In this section, we refer to the received signal as s(t) instead of r(t) as in the previous
section.

8.6.1 QPSK signal at particular time instant

For a received real QPSK signal for a particular time instant t = kT + τ we can write
s(t) = Ak cos 2πfc t − Bk sin 2πfc t
where Ak , Bk = ±1. τ includes both the delay in the pulse itself (half the pulse
length) plus any propagation delay between the transmitter and receiver.

To prove this, we begin with our standard definition of a real pass band signal
s(t) = <{s̃(t)ej2πfc t }
and recall that for QPSK a digital message can be written as a complex envelope
X
s̃m (t) = Ck p(t − kT )
k
=C0 p(t) + C1 p(t − T ) + C2 p(t − 2T ) + . . .
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 465

where

s̃m (t) =i(t) + jq(t)

=m1 (t) + jm2 (t)

is a complex message waveform and

Ck =Ak + jBk
=ak ejφk

is the digital data, where Ak = ak cos φk and Bk = ak sin φk .

Exercise 143
INSTRUCTOR COPY

Evaluate s̃m (t) at sampling times t = mT + τ .

Solution
At sampling time t = mT + τ , the received message waveform
X
s̃m (t) = Ck p(t − kT )
k
X
s̃m (t = mT + τ ) = Ck p((m − k)T + τ )
k
=C0 p(mT + τ ) + C1 p((m − 1)T + τ ) + C2 p((m − 2)T + τ ) + . . .
+Cm−1 p(τ − T ) + Cm p(τ ) + Cm+1 p(τ + T ) + . . .
=Cm

Only one term survives since p(t) 6= 0 only for t ' τ where p(t) is maximum
and the adjacent pulses that are shifted by an integer multiple of T are zero at
that time.

Changing variables m → k we can write at time t = kT + τ

s̃m (t)|t=kT +τ =Ck

=Ak + jBk
=i(t) + jq(t)|t=kT +τ
i(t)|t=kT +τ =i(t = kT + τ )
=Ak
q(t)|t=kT +τ =q(t = kT + τ )
=Bk
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 466

Exercise 144
Given the above results for s̃m (t = kT + ), find an expression for s(t = kT + )
in terms of the data values Ak , Bk .

Solution

s(t)|t=kT +τ =<{s̃m (t)ej2πfc t }

=<{(Ak + jBk )(cos 2πfc t + j sin 2πfc t)}
=Ak cos 2πfc t − Bk sin 2πfc t
=i(t) cos 2πfc t − q(t) sin 2πfc t|t=kT +τ
INSTRUCTOR COPY

Exercise 145
Given the above results for s̃m (t = kT + τ ), find an expression for s(t = kT + τ )
in terms of the 4 possible phases of a QPSK signal.

Solution
In polar form

s̃m (t)|t=kT +τ =Ck

=Ak + jBk
=ak ejφk
√ (2n−1)π
so that with Ak = ±1 and Bk = ±1, we find that ak = 2 and φk = 4
are the 4 possible phases of a QPSK signal. The real passband signal

s(t)|t=kT +τ =Re{s̃m (t)ej2πfc t }

=<{ak ejφk )(ej2πfc t }
=<{ak ej(2πfc t+φk ) }
=ak cos(2πfc t + φk )|t=kT +τ

8.6.2 QPSK receiver local oscillator with frequency error

In this example, we show what happens when the receiver local oscillator

ILO (t) − jQLO (t) = e−j2πfLO t

 CHAPTER 8. PHASE SHIFT KEYING (PSK) 467

runs at fLO = fc + ∆f instead of fc .

We may consider this for the analog IQ local oscillator in the USRP. However,
this is not normally done because the USRP cannot pass the DC that arises from the
complex data symbols. In this section, we consider a frequency error within GNU
Radio where the complex baseband signal is available from the USRP Source block.

The USRP Source block output is the complex baseband (analytic) QPSK signal
that appears on a carrier frequency f2 as shown in (section 8.4.2).

s̃2 (t) =[m1 (t) + jm2 (t)]ej2πf2 t

=m1 (t) cos(2πf2 t) − m2 (t) sin(2πf2 t) + j [m2 (t) cos(2πf2 t) + m1 (t) sin(2πf2 t)]
=i2 (t) + jq2 (t)
INSTRUCTOR COPY

We use the complex IQ receiver that multiplies s̃2 (t) by e−j2πf2 t (note the minus
sign for downconversion). A LPF is not needed. However, in this case we consider
a frequency error (or tuning error) ∆f in the local oscillator , so that we multiply
the analytic signal s̃2 (t) by e−j2π(f2 +∆f )t to produce a signal s̃4 (t) that is centered
at f = −∆f and contains both positive and negative frequencies and thus is NOT
analytic.

s̃2 (t) ⇒ ⊗ ⇒ s̃4 (t) = i4 (t) + jq4 (t)

e−j2π(f2 +∆f )t

Exercise 146
Find an expression for s̃4 (t) at the sampling time needed to recover the data
both withi and without a tuning error.

Solution

s̃4 (t) =i4 (t) + jq4 (t)

=s̃2 (t)e−j2πf2 t
=[i(t) + jq(t)]ej2πf2 t e−j2π(f2 +∆f )t
=[i(t) + jq(t)]e−j2π∆f t
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 468

At the sampling time t = kT + τ

s̃4 (t) =[i(t) + jq(t)]e−j2π∆f t |t=kT +τ

=[Ak + jBk ]e−j2π∆f (kT +τ )
=Ck e−j2π∆f t
=ak ejφk e−j2π∆f t
=ak ej(φk −2π∆f t) |t=kT +τ

For perfect tuning with ∆f = 0, the received data s̃4 (t) at sampling times
t = kT + τ is Ck = ±1 ± j and φk = 2(n−1)π
4
.
With a tuning error ∆f , the transmitted phase φk is received as
INSTRUCTOR COPY

φk − 2π∆f t|t=kT +τ = φk − 2π∆f (kT + τ ).

The phase rotates with time at a rate of 2π∆f radians per second.

Exercise 147
Assuming that the kth symbol is decoded correctly at the sampling time t =
kT + τ , and given a non-zero frequency offset ∆f , how many symbols will be
decoded correctly before a symbol error is made due to the phase rotation.

Solution
If the both the transmitted and received phase is π/4 then a QPSK symbol
error will occur if the received phase drifts (rotates) to π/2 because of the
frequency offset, since then the decision could be either π/4 or 3π/4. Thus the
first QPSK symbol error is made when the phase rotation 2π∆f t is greater
than ±π/4. Thus we solve
2π∆f kT = π/4
for k to obtain
1 f0
k= =
8∆f T 8∆f
where f0 = 1/T is the symbol rate.

Exercise 148
For a QPSK signal at 1,000 symbols per second and a frequency offset of 10
Hz, how many QPSK symbols can be transmitted without errors before the
frequency offset causes errors.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 469

Solution
The QPSK symbols will be correctly decoded over a range of π/2(±π/4). Thus
we solve
2π∆f kT = π/2
for k to obtain
f0 1000
k= = = 25
4∆f 4(10)
An alternate solution is to observe that

• The phase rotates with time at a rate of 2π∆f radians per second.

• A change of 2π radians occurs in 1/∆f seconds.

INSTRUCTOR COPY

• A change of π/2 radians occurs in 1/(4∆f ) seconds.

• The number of bits sent during this time is 1000/(4∆f ) = 1000/(40) = 25

bits.

A GRC flowgraph of the QPSK receiver with tuning error is shown in fig. 8.20.
In this example, the QPSK carrier frequency is set to −16 kHz, and is connected to
a GUI slider to adjust the tuning error.

With a tuning error ∆f , the phase rotates with time at a rate of 2π∆f radians
per second. This rotation can be observed by moving the GRC slider.

Figure 8.20: GNU Radio complex receiver with tuning

In order to receive the correct data symbols Ck = ±1 ± j it is necessary to adjust

the tuning error to zero. This can be done manually with the slider. In a practical
receiver, a frequency offset correction (or phase tracking) circuit is employed, also
called carrier acquisition and carrier tracking.

An example of such a circuit is shown in fig. 8.21. The operation of this circuit
is similar to the operation of a phase-locked loop as described in chapter 4 for DSB
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 470

receivers and chapter 6 for FM receivers.

Figure 8.21: QPSK receiver with frequency offset correction sub-circuit

INSTRUCTOR COPY

8.6.3 QPSK receiver local oscillator with amplitude and phase

errors

The receiver local oscillator can be written ILO (t) − jQLO (t) = e−j2πfLO t only if it is
perfect. In this subsection, we assume it has the correct frequency fLO = fc , but has
amplitude and phase errors, so that we can write

ILO (t) =(1 + δ/2) cos(2πfLO t − θ/2)

QLO (t) =(1 − δ/2) sin(2πfLO t + θ/2)

This type of error is likely to occur in the analog complex downmixer. It will not
occur unintentionally in receiver software like GNU Radio. The effect of these errors
can be modelled and tested in GRC.
Exercise 149
Implement this receiver in GNURadio and include the flowgraph. This problem
can be solved using either real or complex notation, show the signals both ways.

Exercise 150
Find output of I and Q branches of a real QPSK receiver with amplitude and
phase errors.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 471

Solution

s(t) =Ak cos(2πfc t) − Bk sin(2πfc t)

=i(t) cos(2πfc t) − q(t) sin(2πfc t)

at sampling time t = kT + τ we have i(t) = Ak , q(t) = Bk .

ILO (t) =(1 + δ/2) cos(2πfLO t − θ/2)

QLO (t) =(1 − δ/2) sin(2πfLO t + θ/2)
I(t) =s(t)ILO (t)
=[Ak cos(2πfc t) − Bk sin(2πfc t)](1 + δ/2) cos(2πfLO t − θ/2)
=Ak (1 + δ/2) cos(2πfc t) cos(2πfLO t − θ/2)
INSTRUCTOR COPY

− Bk (1 + δ/2) sin(2πfc t) cos(2πfLO t − θ/2)

=0.5Ak (1 + δ/2) cos(θ/2) − 0.5Bk (1 + δ/2) sin(θ/2)
Q(t) =s(t)QLO (t)
=[Ak cos(2πfc t) − Bk sin(2πfc t)](1 − δ/2) sin(2πfLO t + θ/2)
=Ak (1 − δ/2) cos(2πfc t) sin(2πfLO t + θ/2)
− Bk (1 − δ/2) sin(2πfc t) sin(2πfLO t + θ/2)
=0.5Ak (1 − δ/2) sin(θ/2) − 0.5Bk (1 − δ/2) cos(θ/2)

Exercise 151
Plot output of I and Q branches of real QPSK receiver for different values of
the data Ak = ±1, Bk = ±1, assuming that θ = 30◦ , δ = 0.2.

Exercise 152
Find the output of the I and Q branches of a complex QPSK receiver with
amplitude and phase errors.

Solution
In complex notation at sampling time t we have

ILO (t) + jQLO (t) = (1 + δ/2) cos(2πfLO t − θ/2) + j(1 − δ/2) sin(2πfLO t + θ/2)
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 472

s(t) =Re{s̃(t)ej2πfc t }
s̃(t) =Ak + jBk
=ak ejφk
I(t) + jQ(t) =s̃(t)ej2πfc t [ILO (t) − jQLO (t)]
=(Ak + jBk )ej2πfc t [ILO (t) − jQLO (t)]
=(Ak + jBk )ej2πfc t
× [(1 + δ/2) cos(2πfLO t − θ/2) − j(1 − δ/2) sin(2πfLO t + θ/2)]
..
.
I(t) =Ak (1 + δ/2) cos(θ/2) − Bk (1 + δ/2) sin(θ/2)
Q(t) =Ak (1 − δ/2) sin(θ/2) − Bk (1 − δ/2) cos(θ/2)
INSTRUCTOR COPY

Exercise 153
Draw the constellation diagram of the received signals assuming that θ =
30◦ , δ = 0.2 and compare to the ideal constellation with θ = 0◦ , δ = 0.

Solution
Using the analysis from the prior Exercises and substituting numbers θ =
30◦ , δ = 0.2, we arrive at the values shown in table 8.1. When plotted, the result
is that with amplitude and phase errors, the originally square constellation is
distorted and looks like a trapezoid as shown in fig. 8.22.

Q (1.35, 1.1)
1
(−0.78, 0.64)

I
−1 1

(0.78, −0.64)
−1
(−1.35, −1.1)

Figure 8.22: QPSK constellation for θ = 30◦ , δ = 0.2.

 CHAPTER 8. PHASE SHIFT KEYING (PSK) 473

Ak Bk I(t) Q(t)

+1 +1 0.778 -0.636

+1 -1 1.347 1.102

-1 +1 -1.347 -1.102

-1 -1 -0.778 0.636

Table 8.1: Values for QPSK constellation for θ = 30◦ , δ = 0.2.

INSTRUCTOR COPY

8.7 Differential BPSK

In this section we consider differential BPSK, in which the information is encoded in

the phase difference between two adjacent symbols. D-BPSK (also written DPSK) has
the advantage that the absolute phase (i.e. a coherent reference signal) is not required.
In practice, the absolute phase is not generally available unless the transmitter and
receiver are phase-locked via GPS.

Terminology: If absolute phase is required, the receiver is coherent. If only phase

differences are required, the receiver is non-coherent.

8.7.1 DBPSK with square pulse shape

A DBPSK waveform with square pulse shape is shown in fig. 8.23 with one cycle of
the carrier per bit time. This waveform is not used in practical systems because the
sharp transitions will cause the bandwidth to be wider than necessary. In this figure,
the carrier wave is cos 2πfc t + φ where φ = 0 or π. The phases are marked below the
waveform. From the phases φ = [π 0 0 0 0 π 0 0] the transmitted data is 10000100.

Figure 8.23: The phases of this DPSK waveform are φ = [π 0 0 0 0 π 0 0]

 CHAPTER 8. PHASE SHIFT KEYING (PSK) 474

The message bits shown above the waveform in fig. 8.23 are the result of differ-
entially decoding the transmitted data. The message bits above the waveform are
shifted to the right by a half cycle of the carrier wave. To see how the differential
decoding works in this case, shift the message bits a half cycle to left so the bits are
right on top of any change in phase. Then the leftmost 1 bit is a reference followed
by 1 = change, 0 = no change.

For this waveform we have

π0000π00] transmitted phases

10000100 transmitted data bits (not the message bits)

INSTRUCTOR COPY

X1000110 differentially decoded message (X is arbitrary)

10000100 differential encoding of differentially decoded message data

(should yield transmitted data

In fig. 8.24, differential decoding is illustrated with 3 cycles of carrier per symbol
time. The differential decoder can be implemented with a delay and multiply opera-
tion with a delay of one symbol time. In this example we have operated directly on
the signal waveform, we do not need to demodulate first before doing the differential
decoding. The demodulation occurs via the differential decoding which multiplies the
modulated carrier by a delayed version of the modulated carrier. Thus the carrier is
multiplied by itself resulting in DC (the desired data) and double frequency terms
that are removed by a low pass filter.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 475

(a) DPSK demodulator block diagram

INSTRUCTOR COPY

(b) Waveform at each point of DPSK demodulator

Figure 8.24: DBPSK block diagram and waveforms labelled at each point where A
is the transmitted waveform, B is the advanced waveform, C is the output resulting
from a delay and multiply operation, and D is the output of a decision device (slicer)
that decides 1 or 0. Note: the waveforms A and B should be switched so that B is
delayed instead of being advanced.

Exercise 154
Write the mathematics to show that differential decoding of a real passband
DBPSK signal does not require demodulation to complex baseband first. As-
sume that there are an integer number of carrier cycles in a symbol time.

Solution
For the time period kT ≤ t ≤ (K1 )T .
s(t) = Ak cos 2πfc t
s(t − T ) = Ak−1 cos 2πfc t
s(t)s(t − T ) = Ak Ak−1 /2
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 476

which is the differentially decoded data, see fig. 7.6 and noting that multiplying
±1 data symbols is the same as XORing 10 logic levels.

8.7.2 D-BPSK with RC2 time domain pulse shape

In this example, we show DBPSK waveforms with an RC2 time domain pulse shape,
where the pulse spans two symbol periods.

One application of this waveform is low-rate data communication on the shortwave

bands. It is used for live contacts on the amateur radio bands at a data rate of R =
31.25 baud, corresponding to a keyboard speed of approximately 50 wpm. The mode
INSTRUCTOR COPY

is called PSK31 and is further described at http://aintel.bi.ehu.es/psk31.html.

PSK31 standard specifies a specific encoding scheme of the alphabet in addition to
DBPSK/RC2 as described below. 1

The PSK31 mode uses a differential binary encoding where reversal of the carrier
phase indicates the symbol 0 and no change in the carrier phase indicates the symbol
1. This is the opposite convention of that presented in chapter 7, but is equally valid.

Figure 8.25 shows an example of a modulated BPSK31 waveform with data pattern
0100.

Figure 8.25: BPSK31 waveform

For a string of 1 symbols, the modulated waveform is equivalent to just transmit-

ting the carrier. For a string of 0 symbols, the modulated waveform is equivalent to
1
DBPSK with RC2 pulse shaping without a specified encoding scheme may also be used in other
applications, but then it is no longer called PSK31.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 477

transmitting the DSB-SC signal:

s(t) =A cos 2πfc t cos 2πfm t

A A
= cos 2π(fc + fm )t + cos 2π(fc − fm )t
2 2
where fm = R/2 = 15.625 baud. Therefore, the bandwidth of the PSK31 signal is
only 31.25 Hz.

The PSK31 standard specifies a coding scheme in addition to specifying DBPSK

with RC2 pulse shaping. The PSK31 standard uses a variable-length coding to en-
code the alphabet. More common characters use shorter codes while less common
characters use longer codes. Additionally, each code must begin and end with a 1
symbol and may not contain two or more consecutive 0 symbols. A table of codes and
INSTRUCTOR COPY

additional information is at http://aintel.bi.ehu.es/psk31theory.html. In the

PSK31 standard, any string of two or more 0 symbols is ignored for the purposes of
decoding the received message. This variable-length coding scheme has the following
advantages:

1. Transmitting a string of 0 symbols will not result in any decoded message, but
allows the PLL (phase-locked loop) in the receiver to synchronize to the nominal
31.25 Hz symbol rate. Transmitting a string of 0 symbols (known as the idle
signal) before the message text ensures that the receiver synchronizes to the
symbol rate of the transmitted signal.

2. No additional bits are required to synchronize the start and end of characters.
Since all characters start and end with 1 and the string 00 may not appear within
a character, we know that the string 001 indicates the start of a character and
the string 100 indicates the end of a character.

The PSK31 mode does not use any form of error correction, although an intelligent
decoder may try to make a best guess for a string of symbols that does not map to
any character. There is a QPSK variant that provides error correction capabilities.

In fig. 8.25, the message is obtained by seeing where the envelope crosses 0, how-
ever, decoder output is polarity reversed by convention for PSK31, so that a change
in the received sequence is decoded as a 0 bit and no change as a 1 bit.

Exercise 155
Write the PSK31 signal in the standard BPSK format for a real BPSK signal
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 478

Solution

s(t) =Re[m(t)ej2πfc t ]
=m(t) cos 2πfc t
X
= Dk p(t − kT ) cos 2πfc t
k

where Dk = Ak ⊗ Dk−1 is the differentially encoded message Ak and p2 (t) =

[1 − cos(πt/T )]/2 for t in {0, 2T }.

fig. 8.26 shows how the waveform in fig. 8.25 is produced. The explanation is in
the figure caption.
INSTRUCTOR COPY

Figure 8.26: Upper subplot: The pulse shape spanning 2T . Middle subplot: Red
shows impulses Dk corresponding to the differentially encoded message bits. Green
shows decoder output, i.e. the differentially decoded pulses corresponding to the
message bits. Blue shows message waveform m(t). Each red impulse is convolved
with the pulse p(t) to yield an RC pulse spanning two symbol periods. The pulses
overlap resulting in the blue waveform. Note that m(t) has zero crossings one-half
symbol time after a green 1 message bit. Lower subplot: The signal waveform with
carrier. The envelope crosses zero at T2 after a green 1 message bit. Thus the message
can be obtained by observing the zero crossings.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 479

8.7.3 DPSK with RRC frequency domain pulse

The wireless LAN standard 802.11b includes different modes for bit rates of 1, 2, 5.5
and 11 Mbit/s. In all cases the signal looks like an 11 MHz BPSK and QPSK waveform
using square-root raised cosine (RRC) pulse shaping in the frequency domain with
75% excess bandwidth (β = 0.75).

8.8 QPSK variants

In this section, we consider two variants of QPSK: offset QPSK and π/4 QPSK. The
constellation of each variant is shown in fig. 8.27.
INSTRUCTOR COPY

Figure 8.27: QPSK variant constellations

The magnitude of complex envelope will depend on the pulse shape. The contin-
uous time signal constellation shows the values of the magnitude of complex envelope
by inspection.
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 480

8.8.1 Offset QPSK

The offset QPSK signal is the same as the QPSK signal, except that the quadrature
component is delayed by half a symbol period. Thus we modify the complex baseband
signal for offset QPSK.
s̃(t) = i(t) + jq(t − T /2)
where we use the expressions for i(t), q(t) from QPSK, repeated here for convenience
X
i(t) = Ak p(t − kT ) = A0 p(t) + A1 p(t − T ) + A2 p(t − 2T ) + . . .
k
X
q(t) = Bk p(t − kT ) = B0 p(t) + B1 p(t − T ) + B2 p(t − 2T ) + . . .
k
INSTRUCTOR COPY

For QPSK, we choose Ak = ±1 and Bk = ±1, so that Ck = ±1 ± j. The real offset

QPSK signal is
s(t) = i(t) cos 2πf2 t − q(t − T /2) sin 2πf2 t

In offset QPSK, there are no 180 degree transitions so the continuous time signal
constellation traces do not pass through zero with square pulses of length T . However,
there are 90 degree transitions that occur at the half-symbol time T /2.

8.8.2 π/4 QPSK

The π/4 QPSK signal is the same as QPSK except that we code

Ak =Ak−1 cos φk − Bk−1 sin φk

= cos δk
Bk =Ak−1 sin φk − Bk−1 cos φk
= sin δk

We define the phases φk per table 8.2. We use these expressions for Ak , Bk in the
expressions for i(t), q(t) from QPSK and find the real QPSK signal

s(t) = i(t) cos 2πf2 t − q(t) sin 2πf2 t.

 CHAPTER 8. PHASE SHIFT KEYING (PSK) 481
INSTRUCTOR COPY

H
Figure 8.28: π/4 QPSK constellation

Ak Bk φk

1 1 π/4

0 1 3π/4

0 0 -3π/4

1 0 -π/4

Table 8.2: Phases of π/4 QPSK.

We find the phases δk take on 8 possible values nπ/4 for n in {1, ..., 8}. However,
successive values of δk are never nπ/2 apart, as shown in fig. 8.28.

In offset QPSK there are no 180 degree or 90 degree transitions.

 CHAPTER 8. PHASE SHIFT KEYING (PSK) 482

8.9 Chapter review questions

Q 8.1 (3 marks) What is the output of the USRP transmitter when the input to a
GRC URSP Sink block is a real BPSK signal s̃(t) = m(t) cos 2πf1 t?

Solution
The USRP Sink block does the second complex multiply and takes the real part.

s(t) = <{s̃(t)ej2πfc t }
= <{m(t) cos(2πf1 t)ej2πfc t }
n m(t)   j2πfc t o
j2πf1 t −j2πf1 t
INSTRUCTOR COPY

=< e +e e
2
n m(t)  o
=< ej2π(fc +f1 )t + ej2π(fc −f1 )t
2
m(t)
= cos(2π(fc + f1 )t) + cos(2π(fc − f1 )t)
2
Thus there will be two BPSK signals at the output, one at fc + f1 and one at
fc − f1 .

Q 8.2 (4 marks) Show how a real QPSK signal s(t) can be generated starting with
s̃m (t) = m1 (t)+jm2 (t) using components at positive and negative frequencies without
taking the real part of anything.

Solution
Consider that taking the real part of a positive complex signal is mathematically
equivalent to adding the matching negative complex exponential.

s̃2 (t) =s+ (t)

=s̃m (t)ej2πf2 t
=[m1 (t) + jm2 (t)]ej2πf2 t
=m1 (t) cos(2πf2 t) − m2 (t) sin(2πf2 t) + j[m2 (t) cos(2πf2 t) + m1 (t) sin(2πf2 t)]
=i2 (t) + jq2 (t)
 CHAPTER 8. PHASE SHIFT KEYING (PSK) 483

Define a negative frequency QPSK signal

s̃5 (t) =s− (t) = s∗+ (t)

=s̃∗m (t)e−j2πf2 t
=[m1 (t) − jm2 (t)]e−j2πf2 t
=[m1 (t) − jm2 (t)] cos(2πf2 t) − j sin(2πf2 t)]
=m1 (t) cos(2πf2 t) − m2 (t) sin(2πf2 t) − j[m2 (t) cos(2πf2 t) + m1 (t) sin(2πf2 t)]
=i2 (t) − jq2 (t)

Add the two signals

INSTRUCTOR COPY

s̃2 (t) + s̃5 (t) =2i2 (t)

=2m1 (t) cos(2πf2 t) − 2m2 (t) sin(2πf2 t)
=2<{s̃2 (t)}
=2s(t)
=s+ (t) + s− (t)

Q 8.3 (2 marks) Explain qualitatively why the message symbol error rate is increased
due to the amplitude and phase offsets.

Hint: Symbols are correctly decoded provided that the constellation point Ck stays
within its quadrant. Symbol errors are caused by noise that causes the constellation
point to cross the decision boundaries (which are the i and q axes in this case) into
another quadrant. For example, if the transmitted message symbol is Ck = 1 + j, a
symbol error occurs if the received (decoded) symbol is one of the other 3 possible
values of Ck . Each symbol Ck represents two bits of information. Detailed calculations
of symbol error rate are covered in a later course (ECE 450).

Solution
The distance from the constellation point to the decision boundary is reduced
by the amplitude and phase offset as shown in section 8.6.3. Thus the proba-
bility of noise exceeding this distance is increased.
 Chapter 9

Frequency shift keying (FSK)

INSTRUCTOR COPY

Frequency shift keying is a digital form of general frequency modulation, where the
carrier frequency is modulated in step with the message waveform.

9.1 FSK intuition in real passband

One intuitive approach is to write a binary FSK signal as follows

(
A cos(2πf0 t + φ0 ), binary 0
s(t) =
A cos(2πf1 t + φ1 ), binary 1

where kT ≤ t ≤ (k + 1)T . Each frequency component can be written separately and

written as

s0 (t) =A cos(2πf0 t + φ0 ) binary 0

s1 (t) =A cos(2πf1 t + φ1 ) binary 1

for kT ≤ t ≤ (k + 1)T . Then s(t) is either s0 (t) or s1 (t) depending on the time index
k. This is illustrated in fig. 9.1.

484
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 485

Figure 9.1: FSK block diagram and waveform

This FSK signal switches the carrier frequency between f0 and f1 every symbol
time depending on the data 0 or 1. For coherent FSK with continuous phase φ0 =
φ1 = φ, provided that there are an integer number of cycles of f0 or f1 during time
INSTRUCTOR COPY

T.

We choose f0 and f1 to obtain orthogonal FSK where

Z (k+1)T
cos(2πf0 t + φ) cos(2πf1 t + φ)dt = 0
kT

so that the data symbols at f0 and f1 can be separated during the symbol time
≤ t ≤ (k + 1)T without interfering with each other. section 9.4 provides further
detail.
Exercise 156
Prove that this integral is correct provided that
2m − n
f0 =
4T
2m + n
f1 =
4T
such that
n
f1 − f0 =
2T

This integral determines the condition for orthogonality between the two frequen-
cies. When n = 1 we have a special case of FSK called minimum shift keying (MSK)
covered in section 9.3.1. For MSK, the frequency difference is half the symbol rate.
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 486

9.2 FSK equations

9.2.1 FSK real passband

In this section we derive the FM real passband equations, following the same approach
that was used in section 6.3.

1. We start with the FM equations, in particular

fi (t) = fc + kf m(t)
where the message is a square wave, or more precisely,
INSTRUCTOR COPY

X
m(t) = Ak p(t − kT )
k

where Ak = ±1 and (
1, 0 ≤ t ≤ T
p(t) =
0, otherwise

2. The modulation sensitivity kf is chosen so that

fi (t) =fc + kf Ak
(
fc − ∆f, binary 0
=
fc + ∆f, binary 1

where
f0 = fc − ∆f
f1 = fc + ∆f
f1 − f0 = 2∆f

3. The FSK signal is written in standard format

s(t) =a(t) cos(θ(t))
=a(t) cos[2πfc t + φ(t)]
where θ(t) = 2πfc t + φ(t) has a linear variation at rate fc and a time varying
part φ(t).
1 dθ(t)
fi (t) =
2π dt
1 dφ(t)
=fc +
2π dt
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 487

Exercise 157
Combining these 3 starting points, show that
Z tX
φ(t) = 2πkf Ak p(α − kT )dα
0 k

Solution

Z t
θ(t) =2π fi (α)dα
0
Z t
=2π [fc + kf m(α)] dα
INSTRUCTOR COPY

0
Z t
=2πfc t + kf m(α)dα
0
 Z t 
s(t) =Ac cos 2πfc t + 2πkf m(α)dα
0
" Z tX #
=Ac cos 2πfc t + 2πkf Ak p(α − kT )dα
0 k
=Ac cos [2πfc t + φ(t)]

Thus Z tX
φ(t) = 2πkf Ak p(α − kT )dα
0 k

9.2.2 FSK phase ramps

The integral Z tX
φ(t) = 2πkf Ak p(α − kT )dα
0 k

can be defined over intervals kT ≤ t ≤ (k + 1)T for each pulse with data Ak .

For example, if we assume A0 = +1 and a square pulse p(t) = 1 for 0 ≤ t ≤ T ,

 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 488

then we have
Z t
φ(t) =2πkf A0 · 1 · dα
0
=2πkf t
=2π∆f t

Observe that

• the phase is a ramp that starts at φ(t = 0) = 0 and increases linearly to

φ(t = T ) = 2π∆f T .
n
• From the exercise in section 9.1 that for orthogonal FSK, f1 − f0 = 2T
INSTRUCTOR COPY

• From section 9.2.1 f1 − f0 = 2∆f .

n
Thus ∆f = 4T , so that the smallest (n = 1, in the case of MSK) frequency shift
away from the center frequency fc is one quarter the symbol rate, i.e.
f1 =fc + ∆f
=fc + 1/4T
f0 =fc − ∆f
=fc − 1/4T

From the phase ramp result φ(t = T ) = 2π∆f T , we find

n
φ(t = T ) =2π T
4T
nπ
=
2

If we choose the two FSK frequencies to be as close as possible and still orthogonal
(n = 1 for MSK), then the phase ramps from φ(t = 0) = 0 to φ(t = T ) = π/2.

Exercise 158
Continue the example for the next symbol. Assume A1 = −1 for for T ≤ t ≤
2T , and show that for n = 1 that the phase will ramp down from φ(t = T ) =
π/2 to φ(t = 2T ) = 0.

In fig. 9.2, the phase φ(t) and the FSK signal s(t) are shown for an example bit
pattern 1010011110. The phase φ(t) is relative to the carrier wave at fc shown in
yellow.
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 489
INSTRUCTOR COPY

Figure 9.2: Upper subplot: Message symbols. Middle subplot: Phase ramp (integral
of message symbol stream). Lower subplot: FSK signal.

The phase may wander far away from zero when the data pattern has many more
1’s than 0’s or vice versa. In most cases, we choose the data pattern to have the same
number of 1s and 0s on average, so that the phase will stay close to zero on average.

Exercise 159
Consider the s(t) waveform. What are the two frequencies f0 and f1 the center
(yellow) frequency fc and the frequency deviation ∆f in terms of the symbol
time T ?

Solution
• yellow carrier wave fc : one cycle per time T . thus fc = 1/T .

• f1 : 1.25 cycles per time T , thus f1 = 1.25/T

• f0 : 0.75 cycles per time T , thus f0 = 0.75/T

 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 490

• f1 − f0 = 0.5/T = 2∆f , thus ∆f = 0.25/T

Exercise 160
Consider the the phase ramp waveform φ(t), What are the two frequencies
f0 and f1 the center (yellow) frequency fc and the frequency deviation ∆f in
terms of the symbol time T ?

Solution
Observe that the instantaneous frequency of the real passband FSK signal s(t)
is the carrier frequency plus the derivative of the phase ramp. For a phase
ramp with positive slope
INSTRUCTOR COPY

1 dφ(t)
fi (t) =fc +
2π dt
=fc + kf
=fc + ∆f
=f1
dφ(t) π/2
The phase ramp in the first symbol period has slope dt
= T
, thus the
frequency is

1 dφ(t)
fi (t) =fc +
2π dt
1 π/2
=fc +
2π T
1
=fc +
4T
=fc + ∆f
=f1

9.2.3 FSK complex envelope

We want to write FSK in terms of the complex envelope s̃(t) where

s̃(t) =a(t)ejφ(t)
=a(t) cos φ(t) + ja(t) sin φ(t)
=i(t) + jq(t)
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 491

We expect the FSK signal to have constant amplitude a(t) = Ac and time varying
phase φ(t).

To find the complex envelope for FSK, the FM signal s(t) can be written in
standard IQ format:

s(t) =Re{a(t)ejφ(t) ej2πfc t }

=Re{[I(t) + jQ(t)][cos(2πfc t) + j sin(2πfc t)]}
=I(t) cos(2πfc t) − Q(t) sin(2πfc t)
=a(t) cos(2πfc t + φ(t))

Thus for FSK:

INSTRUCTOR COPY

a(t) =Ac
Z tX
φ(t) =2πkf Ak p(α − kT )dα
0 k
Z tX !
I(t) =Ac cos 2πkf Ak p(α − kT )dα
0 k
Z tX !
Q(t) =Ac sin 2πkf Ak p(α − kT )dα
0 k
jφ(t)
s̃(t) =Ac e

The FSK complex baseband signal s̃(t) is illustrated in fig. 6.6 and fig. 2.7. The
positive slope phase ramp appears as a counterclockwise helix and the negative slope
phase ramp appears as a clockwise helix.
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 492

1
=(z(t))

−1
INSTRUCTOR COPY

1
0 0
0.5 1 1.5 −1 <(z(t))
t

Figure 9.3: Complex baseband frequency-shift keying with constant a(t) and sinu-
soidal φ(t).
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 493
INSTRUCTOR COPY

Figure 9.4: Complex baseband frequency-shift keying with constant a(t) and phase
ramps for φ(t).

Exercise 161
Find expressions for s̃(t), i(t), q(t), fi (t) if we assume A0 = +1 and a square
pulse p(t) = 1 for 0 ≤ t ≤ T and 0 otherwise. Repeat for A1 = −1 for
T ≤ t ≤ 2T .

Solution
As shown in section 9.2.2, i we have
Z tX
φ(t) =2πkf Ak p(α − kT )dα
0 k
Z t
=2πkf A0 · 1 · dα
0
=2πkf t
=2π∆f t for 0 ≤ t ≤ T
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 494

s̃(t) =Ac ejφ(t)

I(t) =Ac cos φ(t)
=Ac cos 2π∆f t for 0 ≤ t ≤ T
Q(t) =Ac sin φ(t)
=Ac sin 2π∆f t for 0 ≤ t ≤ T
1 dφ(t)
fi (t) =
2π dt
=kf
=∆f
=f1
INSTRUCTOR COPY

If we assume A0 = −1, then

φ(t) = − 2πkf t
1 dφ(t)
fi (t) =
2π dt
= − kf
= − ∆f
=f0

Thus for FSK complex baseband during time kT ≤ t ≤ (k + 1)T , as shown in

fig. 9.2:

• for data Ak = +1, s̃(t) = ej2π∆f t is a positive frequency exponential at f1 =

+∆f

• for data Ak = +1, the phase ramp φ(t) has positive slope

• for data Ak = −1, s̃(t) = e−j2π∆f t is a negative frequency exponential at f0 =

−∆f

• for data Ak = −1, the phase ramp φ(t) has negative slope
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 495

9.2.4 FSK analytic signal

Exercise 162
Write an expression for an FSK analytic signal s+ (t). Sketch s+ (t) for the data
pattern in fig. 9.2.

9.2.5 FSK spectrum

The spectrum for pulse shaped data signals with a non-periodic random data pattern
will be in the form of a power spectral density G̃(f ) = |S̃(f )|2 that does not include
INSTRUCTOR COPY

phase.

The FSK spectrum may be viewed as the sum of two ASK spectra, one centered
at f0 and the other spectrum centered at f1 , provided that f1 − f0 = 2∆f > 2/T
where T i

The ASK complex baseband power spectral density G̃ASK (f ) depends on the pulse
shape as shown in chapter 7. For a square pulse, the power spectral density will be
as for BPSK fig. 8.5 and is written
 2
sin 2πf T
G̃(f ) = 2P T
2πf T

where P is the average power.

However, the ASK time domain waveform has a DC offset since the data Ak =
0 or 1 with mean DC value 0.5. Thus the ASK complex baseband spectrum will be
the BPSK power spectral density fig. 8.5 plus a δ function at f = 0.

The FSK complex baseband powe spectral density G̃F SK (f ) will consist of two
ASK spectra centered at −∆f and +∆f . The FSK real passband spectrum will have
four ASK spectra centered at ±f0 = ±(fc −∆f ) and ±f1 = ±(fc +∆f ). This result is
an approximation that is valid provided that ∆f > 1/T so that the overlap between
the two ASK spectra is minimal.

Exercise 163
Write expressions for the FSK real passband power spectral density for large
∆f
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 496

Solution

G̃ASK (f ) = G̃BP SK (f ) + 0.5δ(f )

G̃F SK (f ) = G̃ASK (f − ∆f ) + G̃ASK (f + ∆f )
2GF SK (f ) = G̃ASK (f − fc − ∆f ) + G̃ASK (f − fc + ∆f )
+ G̃ASK (−f − fc − ∆f ) + G̃ASK (−f − fc + ∆f )
= G̃ASK (f − f0 ) + G̃ASK (f − f1 )
+ G̃ASK (−f − f0 ) + G̃ASK (−f − f1 )

Exercise 164
INSTRUCTOR COPY

Sketch the FSK power spectral density for ∆f = 2/T , using fig. 8.5 as a starting
point.

For MSK where ∆f = 1/4T the ASK spectra overlap significantly and the resul-
tant power spectral density is found to be
 2
16P T cos 2πf T
G̃(f ) =
π2 1 − 16f 2 T 2

The MSK and BPSK spectrum (power spectral density) are shown in fig. 9.5
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 497
INSTRUCTOR COPY

Figure 9.5: PSK-MSK-spectrum

9.3 FSK variations

9.3.1 Minimum-shift keying and Offset-QPSK

To understand minimum shift keying (MSK) we will consider a real offset QPSK
signal with a symbol rate 1/T :

s(t) = i(t) cos(2πf2 t) − q(t − T /2) sin(2πf2 t)

Assuming the data symbols at the sampling time are

i(t) =Ik
=±1
q(t) =Qk
=±1
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 498

Exercise 165
Consider a variation of this QPSK signal at a slower symbol rate 1/2T and
with a cosine pulse shape over two symbol periods
   
πt πt
s(t) = i(t) cos cos(2πf2 t) − q(t − T ) sin sin(2πf2 t)
2T 2T

From trigonometric identifies, we can rewrite this as

   
1 π
s(t) = cos 2π f − Ik Qk t + (1 − Ik ) , kT ≤ t ≤ (k + 1)T
4T 2

Note that the frequency of s(t) is defined by the produce Ik Qk . The phase is
kept continuous at data transitions by the term (1 − Ik )π/2.
INSTRUCTOR COPY

Show that MSK and offset QPSK with cosine pulse shaping are equivalent with

f1 =fc + ∆f
=fc + 1/4T
f0 =fc − ∆f
=fc − 1/4T

Check this result by plotting waveforms using both the MSK and offset QPSK
formulas for a specific data sequence Ik , Qk .

9.3.2 Multitone FSK

FSK may use four or more frequencies to represent two or more bits per symbol time
T . 4-FSK can represent two bits per symbol 00 01 11 10.

An example of 4-FSK is the FT4 standard described at https://physics.princeton.

edu/pulsar/k1jt/FT4_FT8_QEX.pdf. The FT4 standard specifies the protocol as
well as the 4-FSK modulation scheme. A related standard is FT8 using 8-FSK.

Another example is DTMF section 2.7.6.4. in which two out of four frequencies
(tones) are sent at the same time.
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 499

9.3.3 Multi-carrier modulation

The ideas of FSK and PSK are combined in so-called multi-carrier modulation, also
called Orthogonal Frequency Division Multiplexing (OFDM). The idea is to use mul-
tiple carrier waves, each modulated with a separate data stream using BPSK or QPSK
or QAM.

A complex baseband OFDM signal sn = s(t = nTs ) for a block of N0 time samples
with index 0 ≤ n ≤ N0 − 1 may be generated for a block of N0 complex data values
Ck by using an inverse DFT section 1.4
N0 −1
1 X
sn = Ck e+j2πkn/N0
N0 k=0
INSTRUCTOR COPY

The IDFT takes a block of N0 frequency domain samples Ck and transforms them
into a block of N0 time samples at sampling rate fs = 1/Ts spanning one OFDM
symbol time T0 = N0 Ts .

The IDFT sn for 0 ≤ n ≤ N0 − 1 represents the sum of N0 analytic carrier waves

+j2πkn/N0
e at frequencies k/T0 = kf0 = kfs /N0 , each of which is modulated with a
complex data value Ck for 0 ≤ k ≤ N0 − 1.

The IDFT is repeated for the next block of N0 complex data values (which will
be different) to make the next OFDM symbol of length T0 . Each OFDM symbol of
length T0 contains (represents) N0 complex data symbols Ck .

ODFM is generated in GNUradio using the OFDM block in fig. 9.6

Figure 9.6: OFDM complex baseband transmitter

The OFDM receiver uses a forward DFT after synchronization to obtain the data
values Ck for each ODFM symbol. Details on OFDM data formats and receiver design
are available at https://en.wikipedia.org/wiki/Orthogonal_frequency-division_
multiplexing and many other sources.
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 500

9.3.4 Unified multicarrier ASK-FSK-PSK-QAM-OFDM mod-

ulator

The ideas of multitone FSK and multicarrier modulation can be combined in a unified
system. The diagram in fig. 9.7 may be used to generate any ASK, FSK, PSK, QAM,
and OFDM multicarrier modulation. Only one of the 4 complex oscillators may be
active (ASK, PSK, QAM). All of the 4 complex oscillators may be active one at a
time (4-FSK), two at a time (DTMF) or all at once (OFDM). 2-FSK may be made
using either one oscillator, or two oscillators one at a time. The number of bits per
symbol depends on which modulation type is chosen.

The unified system may be implemented in the time domain using the complex
oscillators in fig. 9.7 or using an inverse DFT that generates the complex oscillators
INSTRUCTOR COPY

as shown in the previous section. The mathematics is the same in either case.

The ”demux map” splits the incoming data stream into a block of N0 = 4 complex
symbols Ck , one for each carrier. Each such block represents one multicarrier symbol
of length T0 .

Figure 9.7: Universal multicarrier ASK-FSK-PSK-QAM system

 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 501

Exercise 166
Use the unified modulator structure to specify systems that transmits 1,2,4,8
or 16 bits per symbol.

Solution
• 1-bit: 2-FSK, BPSK, ASK

• 2-bit: 4 level ASK, QPSK, binary ASk or PSK on two carriers, DTMF,
4-FSK

• 4 or more bits: many combinations of the above

INSTRUCTOR COPY

Exercise 167
Write a sequence of 2N0 = 8 complex data symbols Ck for 0 ≤ k ≤ 7 that will
generate a 2-FSK signal with data sequence 10

Solution
Since 2-FSK uses 2 frequencies, only 2 of the 4 oscillators are used. The input
to the unused oscillators is set to 0. Thus the sequence Ck that generates 2-FSK
is [1 0 0 0 0 1 0 0]

9.3.5 FSK pulse shaping

The operation of FSK with phase ramps is most clearly illustrated by using square
pulses p(t) = 1, 0 ≤ t ≤ T as was done in fig. 9.2. However, the spectrum has
significant sidelobes since the FSK spectrum is based on the BPSK spectrum as
shown in section 9.2.5. The FSK spectrum sidelobes can be significantly reduced by
applying pulse shaping as illustrated in chapter 7.

The waveform for 4-FSK pulse shaping with a Gaussian pulse (called 4-GFSK) is
illustrated in fig. 9.8
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 502

Figure 9.8: FT4 waveform

INSTRUCTOR COPY

A comparison of the spectrum of 2-FSK with square pulse shaping (pink) and 4-
GFSK with Gaussian pulse shaping is shown in fig. 9.9. The 4-GFSK (FT4) symbol
rate is 1/T = 23.4 with tones separated by the baud rate. The 2-FSK (RTTY) symbol
rate is 45.45 with tones separated by 170 Hz.

Figure 9.9: FT4 spectrum

 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 503

9.4 FSK receivers

9.4.1 Non-coherent FSK receiver 1

FSK receivers may be designed using any of the FM demodulation techniques de-
scribed in chapter 6 on page 331 that are based on the idea of differentiation. The
output of the FM demodulator will be the instantaneous frequency fi (t) which is the
desired message (baseband data sequence). The phase ramps are differentiated to
become square pulses
Z tX
φ(t) =2πkf Ak p(α − kT )dα
0 k
INSTRUCTOR COPY

1 dφ(t)
fi (t) =
2πX dt
=kf Ak p(t − kT )
k

The waveform fi (t) may be sampled at times t = kT +  to recover the data bits Ak
in the same way as done in chapter 8.

An example FSK receiver using the FM demodulator

m(t) = arg s̃(t − T )s̃∗ (t)

in section 6.8 is shown in fig. 9.10.

 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 504
INSTRUCTOR COPY

Figure 9.10: FSK receiver

9.4.2 Non-coherent FSK receiver 2

Another type of non-coherent receiver that does not depend on phase (only on phase
differences) consists of two bandpass filters, one at each of f0 and f1 followed by an
envelope (AM) detector and a comparator, as shown in fig. 9.11
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 505
INSTRUCTOR COPY

Figure 9.11: FSK bandpass filter receiver

9.4.3 Modification of non-coherent FSK receivers 1 and 2

A preferred method of processing the waveform is to add all the samples in the symbol
time window kT < t < (k + 1)T and average them. This method is an example of
a matched filter. In the averaging process the signal samples add in voltage and the
noise samples add in power so that the signal=to=noise power ratio is increased.
Specifically, two signal samples added together yield twice the voltage or four times
the power, whereas the average power of two noise samples (random values from
a Gaussian distribution) is twice the average power of one noise sample. Thus the
signal-to-noise power ratio is increased by a factor of 2 by adding two samples together.

9.4.4 Coherent FSK receiver

A coherent FM receiver operates directly on the FSK real passband or complex base-
band waveform without using differentiation. A coherent FM receiver implements the
integral first mentioned in section 9.1 for each frequency
Z (k+1)T
m0 (t) = s(t) cos(2πf0 t + φ)dt
kT
Z (k+1)T
m1 (t) = s(t) cos(2πf1 t + φ)dt
kT
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 506

Since f0 , f1 are chosen so that the FSK is orthogonal, i.e.

Z (k+1)T
cos(2πf0 t + φ) cos(2πf1 t + φ)dt = 0
kT

if s(t) = cos(2πf0 t + φ), then the integral for m1 (t) = 0, i.e. the transmission at f1
will not be detected by the f0 integral.

This coherent FSK receiver is effectively two ASK receivers, one for f0 and one
for f1 , as illustrated in fig. 9.12
INSTRUCTOR COPY

Figure 9.12: FSK coherent receiver

9.5 Chapter review questions

Q 9.1 (4 marks) Sketch a figure similar to fig. 9.2 but for ∆f = 0.5/T . 2 marks
each for phase plot and real passband plot. Label the axes with the correct units and
values.
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 507

Solution
• The phase plot is identical to fig. 9.2 except that the vertical axis on the
phase plot extends from −3π to 3π, so that each phase ramp changes the
phase by ±π during each symbol time T .

• The real passband plots are similar to fig. 9.2:

• the yellow waveform showing the center frequency fc = 1/T with
one cycle per period T is unchanged.

The waveform for f1 = fc + ∆f == 1/T + 0.5/T = 1.5/T will have

1.5 cycles during a symbol time T .
– The waveform for f0 = fc − ∆f == 1/T − 0.5/T = 0.5/T will have
INSTRUCTOR COPY

0.5 cycles during a symbol time T .

Q 9.2 (4 marks) Plot s̃(t) for this case. Hint: start with the ”digital modulation IQ
3-D plotting code” on the course website. Hint 2: adjust the parameters in the first
20 lines of code and consider the code for x7 and/or x7a .
 CHAPTER 9. FREQUENCY SHIFT KEYING (FSK) 508

Solution
INSTRUCTOR COPY

Figure 9.13: FSK complex baseband with ∆f = 0.5/T

The projections on the real-time and imag-time planes are the I and Q com-
ponents of s̃(t = i(t) + jq(t).
 Chapter 10

Synchronization
INSTRUCTOR COPY

10.1 Carrier Synchronization

For modulation schemes where a phase reference is required to correctly demodulate

the signal s(t) to obtain the message m(t), a carrier synchronization algorithm is
used to find the phase of the carrier. One example is the DSB receiver with frequency
offset in section 4.3.2.3 where the phase of the suppressed carrier is found to correctly
demodulate the DSB. A common feature of many carrier synchronization algorithms
is some kind of feedback (phase-locked loop) or feedforward architecture.

Exercise 168
Invent an algorithm that will accomplish carrier synchronization to a real un-
modulated carrier cos 2πfc t

10.2 Symbol synchronization

The received data sequence m(t) must be sampled at the correct time t = kT +  to
corrctly receive the data symbol. The task of a symbol (bit) synchronizer is to find
the optimal delay . This task may be done by simply looking at the eye diagram
and visually observing the time when the eye is open. The challenge is to design
algorithms to perform that task. A common feature of many symbol synchronization
algorithms is some kind of feedback (phase-locked loop) or feedforward architecture.

509
 CHAPTER 10. SYNCHRONIZATION 510

Exercise 169
Invent an algorithm that will accomplish bit synchronization on a real message
waveform with RC2 pulse shaping.

10.3 Frame Synchronization

The idea of frame synchronization is to identify the byte boundaries in a stream of

data

Consider the example stream in the figure below. A known marker sequence
INSTRUCTOR COPY

M = [m1 m2 m3 m4 m5 m6 m7 ] is inserted in the data stream cn , where each of the 7 bits

in the marker sequence is ±1. To find the location of the marker sequence in the data
stream cn , we do cross-correlation of cn with the marker sequenceM to obtain the
sync output

X
7
fn = mk cn+k
k=1

A frame synchronization system is shown in fig. 10.1

 CHAPTER 10. SYNCHRONIZATION 511
INSTRUCTOR COPY

Figure 10.1: Frame synchronization

To accomplish frame synchronization, we look for the maximum value of fn which

is 7 for this example, assuming no errors in the data stream cn .

The diagram look similar to a digital filter (appendix D.2) with filter coefficients
M = [m1 m2 m3 m4 m5 m6 m7 ]. However, in this case the marker sequence is shown
as sliding to the right, and the data sequence is fixed, or equivalently, the marker
sequence fixed and data sliding from right to left. For a digital filter, we would
typically show the filter coefficients as fixed, and the data sliding in from the left to
right, and the cross-correlation becomes a convolution (with cn−k instead of cn+k ).
 CHAPTER 10. SYNCHRONIZATION 512

Exercise 170
Draw a diagram of a frame synchronization system that looks like the dig-
ital filter diagram shown in appendix D.2 using the filter coefficients M =
[m1 m2 m3 m4 m5 m6 m7 ] = [1110010].

If there are some errors in the data stream, then the maximum value of fn may be
less than 7. To allow for such errors, we may set the sync threshold to 6 instead of 7
and declare correct frame sync when fn = 6. However, we want to choose the marker
sequence so that the probability of getting fn = 6 when the marker is not correctly
synchronized (i.e offset by one or more bits) is low.

Exercise 171
INSTRUCTOR COPY

Computer the cross-correlation between the data sequence with embedded

marker in fig. 10.1 and the marker M = [m1 m2 m3 m4 m5 m6 m7 ] = [1110010].
If the sync threshold is set to fn = 6. then is frame synchronization correctly
declared?

A marker sequence of length 7 is too short for a practical system because a random
data sequence has a probability of one in 27 = 128 to match the marker sequence
exactly, and we would get a false frame sync quite often.

In practical wireless systems where signals may be weak, a longer frame sequence
is used to minimize the probability of incorrect frame sync or false frame sync in
noise. For example, the FLEX pager system uses a marker of length 32.
 Chapter 11

Link Budget
INSTRUCTOR COPY

11.1 Link Budget Overview

For any communications system designed to send a message from point A to point B
linked via some channel (fiber, coax, radio), the available resources are transmit power
and channel bandwidth, and the obstacles to be overcome are noise and interference.
Both power and bandwidth cost money; how much will depend on the details of the
communications system.

To design a working link with acceptable quality at the lowest possible cost we
need to first specify the customer performance requirements, and then compute the
necessary power and bandwidth. These performance requirements are usually ex-
pressed for digital signals as the data rate at a given error rate, or for analog signals
as the fidelity (bandwidth) and signal-to-noise ratio.

Another performance requirement is the distance to be covered. Alternately, we

may be given the power and/or bandwidth that we can afford, and then need to
compute the maximum data rate and distance that can be achieved.

11.2 Power obtained over a link

The power obtained over a link for a given distance is given by

Pt Gt Gr
Pr,o = (11.1)
L0
where

513
 CHAPTER 11. LINK BUDGET 514

Pt is the transmit power in watts

Gt is the transmit antenna gain (dimensionless)
Gr is the receiver antenna gain (dimensionless)
L0 is the path loss between points A and B

Usually, the quantities are expressed in dB, so by taking log10 of both sides, we
write
Pr,o = Pt + Gt + Gr − L0 (11.2)
The antenna gains depend on the details of the antenna design. An estimate of the
antenna gain in terms of it’s size is given by
4πAe
G= (11.3)
λ2
INSTRUCTOR COPY

where Ae is the effective antenna area, which is close to the physical area, and λ = c/fc
is √
the wavelength used. The approximate beamwidth of the antenna is given by
λ/ A.

11.3 Path loss

The path loss L0 is a function of the distance d and the carrier frequency fc or
wavelength λ = c/fc and is usually defined to be greater than 1 (0 dB). The function
depends on the particular channel model between point A and point B.

11.3.1 Cable or Fibre

If the channel is a cable (fiber of coax), then we set Gt = Gr = 1, and find the loss
from the manufacturer loss specifications in terms of dB/meter, and multiply by the
path length in meters.

11.3.2 Free space

If the channel is radio, then for free space propagation (e.g. between point A on the
earth and point B in a satellite or aircraft) the signal spreads out over a sphere in all
directions, resulting in  2
4πd
L0 = >1 (11.4)
λ
 CHAPTER 11. LINK BUDGET 515

The path loss increases as the square of the distance. We can show (exercise) that
for d in meters, fc in GHz,
L0 = 32.4 + 20 log fc + 20 log d > 0 (11.5)
where log is to the base 10.

11.3.3 Flat earth

The free space formula also applies for propagation along a flat earth between anten-
nas at heights h1 , h2 , provided that d is less than a breakpoint distance
4πh1 h2
dbrk =
λ
INSTRUCTOR COPY

4πh1 h2 fc
=
c
In this case, the antennas are sufficiently high that the effect of the earth’s surface is
minimal so the propagation path is essentially equivalent to free space.

However, when d > dbrk , the ground-reflected ray is almost the same length as the
direct ray. Since the reflection coefficient is close to -1, the two rays almost cancel,
as shown in fig. 11.1.

Figure 11.1: Propagation path between two antennas

 CHAPTER 11. LINK BUDGET 516

In this case, we find that the path loss increases as the fourth power of the distance
d4
L0 = >1 (11.6)
h21 h22
or in dB
L0 = 40 log d − 20 log h1 − 20 log h2 > 0 (11.7)
independent of fc . We can show (exercise) that both path loss formulas give the same
result when d = dbrk .

On a plot of L0 versus d on a log scale, the slope changes from 2 to 4 at d = dbrk ,

as shown in fig. 11.2.
INSTRUCTOR COPY

Figure 11.2: Path loss versus distance with h1 = 2m, h2 = 20m, f = 5.8 GHz. L1
free space, L3 flat earch, L2 combination

The free space and flat earth path loss formulas can be combined into one formula
 2  2
4πd d
L0 = 1+ (11.8)
λ dbrk
 CHAPTER 11. LINK BUDGET 517

or in dB
L0 = 32.4 + 20 log fc − 20 log dbrk + 40 log d (11.9)
which is equivalent to free space loss for d  dbrk and to flat earth loss for d  dbrk .
(exercise).

11.3.4 Practical considerations

In practice, the path loss depends on the details of the environment, and cannot be
precisely determined by these formulas. Thus, in general, we measure the path loss
over a range of distances, and write
INSTRUCTOR COPY

L0 = Lm + 10n1 log d (11.10)

for d  dbrk , where Lm is the measured path loss at a specified distance (usually 1 m
or 1 km) and n1 ' 2 is the measured slope.

For any value of d, we write

L0 = Lm + 10(n1 − n2 ) log dbrk + 10n2 log d (11.11)

where dbrk is a measured breakpoint distance, and n2 ' 4 is the measured slope for
d > dbrk .

Other propagation models depend on the details of the terrain, and take into
account diffraction around edges, Fresnel zones where two paths differ by λ/2, atmo-
spheric effects, earth curvature K factor, and refractivity gradients.

For example, for a line of sight path to have free space path loss, any obstructions
or obstacles must be outside the ”first Fresnel zone”, i.e. the distance from the
obstacle to the line of sight path
r
λd1 d2
F1 = (11.12)
d1 + d2
where d1 is the distance from the obstacle to one end of the link, and d2 is the distance
from the obstacle to the other end of the link.

Propagation prediction software such as http://radiomobile.pe1mew.nl/ and

https://www.ve2dbe.com/rmonlineinfoeng.html uses a terrain elevation database
to compute the path loss to all points on a map and thus generate a coverage map
showing signal strength versus location.
 CHAPTER 11. LINK BUDGET 518

11.4 Power required at receiver

The power required at the receiver to achieve acceptable performance (also called the
receiver sensitivity) is given by
Pr,n = kT0 (S/N )W F (11.13)
where

k is Boltzmann’s constant (10−22.86 W/Hz)

T0 is the noise temperature (normally 290 K on earth)
F is the (dimensionless) noise figure of the system representing the imperfect am-
plifiers and/or interference
INSTRUCTOR COPY

S is signal power
N is noise power
(S/N ) is signal-to-noise ratio
W is bandwidth

Usually, the quantities are expressed in dB, so by taking log10 of both sides, we
write
Pr,n = k + T0 + (S/N ) + W + F (11.14)
where k = −228.6. Thus Pr,n increases with (S/N ), W and F .

The link budget parameters are sometimes combined as follows (linear not in dB):

Effective Isotropic Radiated Power (EIRP) : EIRP = PT GT

receiver figure of merit : GR /T
N0 : noise power spectral density in Watts/Hz
signal-to-noise power density : S/N0 = PR /(kT ) (also called carrier-to-noise ratio)

where N = N0 W , N0 = kT , F = 1 + T /T0 , where T is the system noise temperature

that combines the noise figure with all other noise effects.

11.5 Link budget calculation

The link will work as long as the power obtained is greater than the power needed,
i.e. as long as the link margin M
Pr,o
M= >1 (11.15)
Pr,n
 CHAPTER 11. LINK BUDGET 519

or in dB
M = Pr,o − Pr,n > 0dB (11.16)

To do a link budget calculation, we are given all link parameters except one, and
then compute that one. For example, given S/N, W, F, Pt , Gt , Gr , d, fc and free space
propagation, we can compute M . In many cases, it is necessary to make reasonable
assumptions about some of these link parameters. We can combine the link budget
equations in various ways. e.g
(S/N ) + M = EIRP + (G/T0 ) − L0 − k − W.

Link budgets are often best calculated using a spreadsheet such as https://
davidhoglund.typepad.com/integra_systems_inc_david/files/NIST_LinkBudgetCalc_
INSTRUCTOR COPY

2_4.xls or http://www.amsatuk.me.uk/iaru/AMSAT-IARU_Link_Model_Rev2.5.5.
xls.

11.6 Digital system performance

For digital systems, the modem performance is usually specified as bit error rate
versus Eb /N0 , where

Eb is energy per bit (J/bit)

N0 is the noise power spectral density (W/Hz)

N0 is the noise power per unit bandwidth. The signal power S in watts is
S = Eb R (11.17)
(units of J/bit × bit/s = J/s = W) and the noise power N in watts
N = N0 W (11.18)
(units of W/Hz × Hz = W). Thus we can write
S Eb R
= (11.19)
N N0 W

The maximum possible data rate Rmax = C is related to the power and bandwidth
via the Shannon capacity formula
S
C = W log2 (1 + ) (11.20)
N
where
 CHAPTER 11. LINK BUDGET 520

C is the capacity in bits per second

W is the bandwidth in Hz
S is the signal power
N is the noise power

The Shannon capacity represents a theoretical upper limit to the data rate as a
function of bandwidth and signal-to-noise ratio. If we use a modulation technique
so that C = W , then we need S/N = 1 (0 dB). However, this requires complex and
expensive circuity. For simple low cost systems with C = W , we need S/N = 10 dB.

11.7 Chapter review questions

INSTRUCTOR COPY

Q 11.1 Derive the dB version of the path loss L0 equation for free space.

Solution
The loss equation for free space is
 2
4πd
L0 = (11.21)
λ

Taking the logarithm of both sides we have

 
c
L0 (dB) = 20 log10 (4π) + 20 log10 (d) − 20 log10 (11.22)
f

where c = 3 × 108 is the speed of light and f = fc × 109 is the carrier frequency
in Hz. Hence eq. (11.22) can be written as
 
3 × 108
L0 (dB) =20 log10 (4π) + 20 log10 (d) − 20 log10
fc × 109
=20 log10 (4π) + 20 log10 (d) − 20 log10 (0.3) + 20 log10 (fc )
 
4π
=20 log10 + 20 log10 (d) + 20 log10 (fc )
0.3
=32.4 + 20 log10 (d) + 20 log10 (fc )

where fc is the carrier frequency in GHz.

Q 11.2 Show that the two formulas for path loss give the same result at the break-
point distance dbrk .
 CHAPTER 11. LINK BUDGET 521

Solution
The loss equation for flat earth is

d4
L0 = (11.23)
h21 h22
4πh1 h2 fc
Substituting dbrk = c
into each of eq. (11.21) and eq. (11.23) yields the
same result:  2
4πfc
L0 = h21 h22
c
where fc is the carrier frequency in Hz.
INSTRUCTOR COPY

Q 11.3 Show that the combined path loss formula simplifies to free space or flat
earth loss for appropriate values of d.

Solution
The combined path loss formula is
 2  2
4πd d
L0 = 1+ (11.24)
λ dbrk
d
When d  dbrk , dbrk ≈ 0. Therefore, eq. (11.24) can be approximated as
eq. (11.21).
d d
When d  dbrk , dbrk ≈ dbrk . Then eq. (11.24) can be approximated as
 2
4πd
L0 = d4 (11.25)
λdbrk
4πh1 h2 fc
Substituting dbrk = c
in eq. (11.25), we have

d4
L0 =
h21 h22

which is the path loss formula for propagation over flat earth.

Q 11.4 Consider a “wireless cable” data link designed to provide high speed internet
access by using radio instead of telephone lines. The required data rate is 1 Gbit/s,
the transmit power is 2 W, the base station antenna is omnidirectional (zero gain),
a patch antenna with 10 dB gain is at the remote station. There is a 2 dB noise
figure and a carrier frequency of 2400 MHz is used. Assume the bit rate is equal to
 CHAPTER 11. LINK BUDGET 522

the bandwidth, and that the signal-to-noise ratio needed for a good quality link (low
error rate) is 10 dB.

Find the maximum range (distance) that can be achieved assuming the path loss
is modelled by free space propagation.

Solution
Assume that bit rate=bandwidth then list parameters given in question

Parameter R W Pt Gt Gr F f S/N
INSTRUCTOR COPY

Linear 1 Gbit/s 1 Gbit/s 2W - - - 2.4 GHz -

Logarithmic 90 dB Hz 90 dB Hz 3 dB W 0 dB 10 dB 2 dB - 10 dB

L0 =32.4 + 20 log10 (d(m)) + 20 log10 (f (GHz))

Pr,o =Pt + Gt + Gr + L0
=3 + 0 + 10 − 32.4 − 20 log10 (d(m)) − 20 log10 (2.4)
= − 27 − 20 log10 (d(m))
Pr,n =k + T0 + W + (S/N ) + F
=10 log10 (10−22.86 ) + 10 log10 (290) + 90 + 10 + 2
= − 228.6 + 24.6 + 90 + 10 + 2
= − 102

for zero link margin M = Pr,o − Pr,n or Pr,o = Pr,n , so we have

−102 = − 27 − 20 log10 (d(m))

d =5623 m

Q 11.5 Continuing from the last question, find the maximum range (distance) that
can be achieved using a more realistic path loss model, i.e. assume propagation along
the ground with antenna heights of 5 m at the transmitter and 1 m at the receiver.
State any reasonable assumptions needed.
 CHAPTER 11. LINK BUDGET 523

Solution

L0 =40 log10 (d(m)) − 20 log10 (h1 (m)) − 20 log10 (h2 (m))

Pr,o =Pt + Gt + Gr + L0
=3 + 0 + 10 + 40 log10 (d(m)) − 20 log10 (1) − 20 log10 (5)
=27 − 40 log10 (d(m))
Pr,n = − 102

for zero link margin M = Pr,o − Pr,n or Pr,o = Pr,n , so we have

−102 =27 − 40 log10 (d(m))

INSTRUCTOR COPY

d =1679 m
 Appendix A

List of symbols
INSTRUCTOR COPY

A.1 Signals, messages and filters

• Signals may be real or complex and may or may not contain a carrier wave.

• Messages are real signals containing information without a carrier wave.

• Filter impulse response is a real time domain signal and filter frequency response
is a complex frequency domain signal.

• Lower case symbols represent time domain signals/messages/filters

• Upper case symbols represent frequency domain signals/messages/filters

524
 APPENDIX A. LIST OF SYMBOLS 525

A.2 Signals

A.2.1 Complex baseband, analytic and real passband signals

s̃(t) S̃(f ) complex baseband signal

s+ (t) S+ (f ) analytic signal

s− (t) S− (f ) analytic signal

s(t) S(f ) real passband signal

INSTRUCTOR COPY

sn Sk discrete time real signal and discrete Fourier transform

A.2.2 Signals in terms of amplitude, phase and carrier fre-

quency

s̃(t) = a(t)ejφ(t) S̃(f ) = |S̃(f )|ej arg S̃(f ) complex baseband

s̃∗ (t) = a(t)e−jφ(t) S̃ ∗ (−f ) = |S̃ ∗ (−f )|e−j arg S̃(−f ) complex baseband

s+ (t) = s̃(t)ej2πfc t S+ (f ) = S̃(f − fc ) analytic signal

s− (t) = s̃∗ (t)e−j2πfc t S− (f ) = S̃ ∗ (−(f + fc )) analytic signal

s(t) = a(t) cos 2πfc t + φ(t) S(f ) = 0.5S+ (f ) + 0.5S− (f ) real passband signal
 APPENDIX A. LIST OF SYMBOLS 526

A.2.3 Sampled signals and Discrete Fourier transform (DFT)

sampling rate fs

sampling time Ts = 1/fs

time index n

sampling in time s(n) = s(t)|t=nTs = s(t = nTs )

observation time T0 (also size/length/time of frame/window/block

INSTRUCTOR COPY

number of samples in frame N0 = T0 /Ts

frequency resolution f0 = fs /N0

frequency index k

sampling in frequency S(k) = S(f )|f =kf0 = S(f = kf0 )

A.3 Messages

m(t) M (f ) real message

n(t) N (f ) real message

m+ (t) M+ (f ) analytic message, positive frequencies only

m− (t) M− (f ) negative frequency analytic message

m∗ (t) = m(t) M ∗ (−f ) = M (f ) complex conjugate

m̂(t) M̂ (f ) Hilbert transform of real message

m[n] M [k] discrete time message and its discrete Fourier transform

m[r, n] M [r, k] discrete time message in rth frame and its DFT
 APPENDIX A. LIST OF SYMBOLS 527

A.4 Filters

h(t) H(f ) impulse and frequency response

P
h[n] H(z) = h[n]z −n discrete time impulse response and z-transform

h[n] H[k] discrete impulse response and Discrete Fourier transform

A.5 Link budget

INSTRUCTOR COPY

Pt transmit power

Gt transmit antenna gain

Gr receiver antenna gain

L0 path loss

Pr,o power obtained at receiver

λ carrier wavelength

fc carrier frequency

d distance between transmitter and receiver

dbrk breakpoint distance

h1 , h2 antenna height
 APPENDIX A. LIST OF SYMBOLS 528

Pr,n power needed at receiver

k Boltzmann constant

T0 noise temperature

S signal power

N noise power

S/N signal-to-noise ratio

W bandwidth
INSTRUCTOR COPY

F noise figure

M link margin

R data rate

C capacity (maximum data rate)

Eb energy per bit

N0 noise power spectral density

 Appendix B

Useful formulas
INSTRUCTOR COPY

B.1 Mathematics

ejθ = cos θ + j sin θ

ejθ + e−jθ ejθ − e−jθ
cos θ = , sin θ =
2 2j

ej(α+β) = cos(α + β) + j sin(α + β)

= ejα ejβ
= (cos α + j sin α) (cos β + j sin β)
= cos α cos β − sin α sin β + j (cos α sin β + sin α cos β)

cos (α + β) = cos α cos β − sin α sin β

sin (α + β) = cos α sin β + sin α cos β
cos (α − β) = cos α cos β + sin α sin β
sin (α − β) = sin α cos β − cos α sin β

529
 APPENDIX B. USEFUL FORMULAS 530

cos (α − β) + cos (α + β)
cos α cos β =
2
sin (α − β) + sin (α + β)
sin α cos β =
2
cos (α − β) − cos (α + β)
sin α sin β =
2
− sin (α − β) + sin (α + β)
cos α sin β =
2

forx << 1

(1 + x)n ' 1 + nx
INSTRUCTOR COPY

sin x ' x

cos x ' 1 − x2 /2

arctan x ' x − x3 /3

B.2 Signals

B.2.1 complex baseband

s̃(t) = a(t)ejφ(t) = I(t) + jQ(t)

s̃∗ (t) = a(t)e−jφ(t) = I(t) − jQ(t)
F{s̃(t)} = S̃(f )
= ∫∞
t=−∞ s̃(t)e
−j2πf t
dt
F{s̃∗ (t)} = S̃ ∗ (−f )
s̃(t) ↔ S̃(f )
s̃∗ (t) ↔ S̃ ∗ (−f )
ej2πfc t ↔ δ(f − fc )
s̃(t)ej2πfc t ↔ S̃(f − fc )
s̃∗ (t)e−j2πfc t ↔ S̃ ∗ (−(f + fc ))
 APPENDIX B. USEFUL FORMULAS 531

B.2.2 analytic

s+ (t) = s̃(t)ej2πfc t
s− (t) = s̃∗ (t)e−j2πfc t
S+ (f ) = S̃(f − fc )
S− (f ) = S̃ ∗ (−(f + fc ))

B.2.3 real passband

INSTRUCTOR COPY

s(t) = Re [s+ (t)] = Re [s− (t)]

2s(t) = s+ (t) + s− (t)
= s̃(t)ej2πfc t + s̃∗ (t)e−j2πfc t
= a(t)ejφ(t) ej2πfc t + a(t)e−jφ(t) e−j2πfc t
= 2a(t) cos[2πfc t + φ(t)]
2S(f ) = S+ (f ) + S− (f )
= S̃(f − fc ) + S̃ ∗ (−(f + fc ))

B.2.4 sampled signals

s̃n = s̃(t = nTs ), Ts = 1/fs

S̃k = S̃(f = kf0 ) = DFT{s̃n }

Z T0 =nTs
S̃k = s̃(t)e−j2πf t dt|t=nTs ,f =kf0
t=0
X
N −1
= s̃n e−j2πkf0 nTs
n=0
X
N −1
= s̃n e−j2πnk/N
n=0

Ts f0 = Ts /T0 = f0 /fs = 1/N

N = T0 /Ts = fs /f0 = T0 fs
 APPENDIX B. USEFUL FORMULAS 532

B.3 Messages: real baseband

m∗ (t) = m(t)
M ∗ (−f ) = M (f )

B.3.1 single tone message

2m(t) = m+ (t) + m− (t)

= Am ej2πfm t ejφ + Am e−j2πfm t e−jφ
INSTRUCTOR COPY

= 2Am cos(2πfm t + φ)
2M (f ) = Am ejφ δ(f − fm ) + Am δ(f + fm )

B.3.2 multitone messages

N/2
X
2m(t) = Ak ej(2πfk t+φk )
k=−N/2
N/2
X
= 2Ak cos(2πfk t + φk )
k=0
N/2
X
2M (f ) = Ak ejφk δ(f − fk )
k=−N/2

B.3.3 sampled discrete-time messages

m[n] = m(t = nTs )

N/2−1
1 X
m[n] = |M [k]| cos(2πkf0 nts + arg M [k])
N k=0
f0 Ts = 1/N
 APPENDIX B. USEFUL FORMULAS 533

B.3.4 Digital messages

X
m(t) = Ak p(t − kT )
k

where Ak are real numbers representing data.

B.4 Modulation schemes

B.4.1 AM complex baseband

INSTRUCTOR COPY

s̃(t) = Ac [1 + ka m(t)]
S̃(f ) = Ac δ(f ) + Ac ka M (f )

B.4.2 DSB complex baseband

s̃(t) = Ac ka m(t)
S̃(f ) = Ac ka M (f )
 APPENDIX B. USEFUL FORMULAS 534

B.4.3 SSB complex baseband

B.4.3.1 SSB messages

X
m+ (t) = Ak ej2πfk t+φk
k
X X
= Ak cos(2πfk t + φk ) + j Ak sin(2πfk t + φk )
k k
= m(t) + j m̂(t)
X
m− (t) = Ak e−j2πfk t−φk
k
X X
INSTRUCTOR COPY

= Ak cos(2πfk t + φk ) − j Ak sin(2πfk t + φk )
k k
= m(t) − j m̂(t)
X
m+ (t) + m− (t) = 2 Ak cos(2πfk t + φk )
k
= 2m(t)

B.4.3.2 USB complex baseband

s̃(t) = m+ (t) = m(t) + j m̂(t)

m̂(t) = H{m(t)} = h(t) ⊗ m(t)
1
h(t) =
πt
1
s̃(t) = m(t) + jH{m(t)} = m(t) + j ⊗ m(t)
πt
S̃(f ) = M+ (f ) = M (f ) + j M̂ (f )
M̂ (f ) = H{M (f )} = H(f )M (f )
H(f ) = −j sgn f
S̃(f ) = M (f ) + jH{M (f )} = M (f ) + jH(f )M (f )
= M (f ) + j(−j) sgn(f )M (f ) = 2M (f ), f > 0, 0, f < 0
 APPENDIX B. USEFUL FORMULAS 535

B.4.3.3 LSB comlex baseband

s̃(t) = m− (t) = m(t) − j m̂(t)

S̃(f ) = M− (f ) = M (f ) − j(−j) sgn(f )M (f ) = 0, f > 0, 2M (f ), f < 0

B.4.3.4 ISB complex baseband

s̃(t) = m+ (t) + n− (t) = m(t) + j m̂(t) + n(t) − j n̂(t)

S̃(f ) = M+ (f ) + N− (f )
INSTRUCTOR COPY

B.4.4 PM complex baseband

φ(t) = kp m(t)
s̃(t) = a(t)ejφ(t)
= Ac ej2πkp m(t)

B.4.5 FM complex baseband

fi (t) = kf m(t)
1 dφ(t)
=
2π Z dt
t
φ(t) = 2π fi (τ )dτ
τ =0
Z t
= 2πkf m(τ )dτ
τ =0
jφ(t)
s̃(t) = a(t)e
Rt
= Ac ej2πkf τ =0 m(τ )dτ
 APPENDIX B. USEFUL FORMULAS 536

B.4.6 PSK complex baseband

s̃(t) = m1 (t) + jm2 (t)

X
= Ck p(t − kT )
k

where

Ck = ak ejφk
= Ak + jBk

are complex numbers representing data.

INSTRUCTOR COPY

B.4.7 FSK complex baseband

s̃(t) =Ac ejφ(t)

Z tX
φ(t) =2πkf Ak p(α − kT )dα
0 k

B.5 Link Budget

Power obtained over a link

Pr,o = Pt + Gt + Gr − L0 (B.1)

Free space path loss for d in meters, fc in GHz,

L0 = 32.4 + 20 log fc + 20 log d > 0 (B.2)

Flat earth path loss for d is meters with antennas at heights h1 , h2 .

L0 = 40 log d − 20 log h1 − 20 log h2 > 0 (B.3)

A combined path loss formula is

L0 = 32.4 + 20 log fc − 20 log dbrk + 40 log d (B.4)

 APPENDIX B. USEFUL FORMULAS 537

where
4πh1 h2
dbrk =
λ
4πh1 h2 fc
=
c
Power required at receiver

Pr,n = −228.6 + T0 + (S/N ) + W + F (B.5)

Link margin
M = Pr,o − Pr,n > 0dB (B.6)
INSTRUCTOR COPY
 Appendix C

Derivations
INSTRUCTOR COPY

Following are some derivations that may help interpret some of the maths scattered
throughout the book. If it is included here it will have been referenced in a relevant
section.

dφ
C.1 φ(t) − φ(t − 1) ≈ dt

dy f (x + h) − f (x)
= lim
dx h→0 h
dy f (x) − f (x + h)
− = lim
dx h→0 h

with f = φ, x = t and h = −1 we have

dφ φ(t) − φ(t − 1)
− ≈
dt −1
dφ
≈φ(t) − φ(t − 1)
dt

538
 Appendix D

Applications
INSTRUCTOR COPY

D.1 Spectrum Analyzers, analog and digital

While an oscilloscope is used to display electrical signals in an amplitude versus time

format, a spectrum analyzer provides an amplitude squared (power) versus frequency
format. fig. D.1 shows the two representations for the same signal.

Figure D.1: Oscilloscope and Spectrum Analyzer Displays

The Fourier analysis introduced in section 1.3 shows that any complex waveform
can be resolved into sinusoidal waveforms of a fundamental frequency and a number
of harmonic frequencies. The spectrum analyzer effectively performs the Fourier
integral:
Z ∞
S(f ) = s(t)e−j2πf t dt
Zt=−∞
∞ Z ∞
= s(t) cos(2πf t)dt − j s(t) sin(2πf t)dt
t=−∞ t=−∞

539
 APPENDIX D. APPLICATIONS 540

The integral finds the frequency components in s(t) by correlating s(t) with cosine and
sine waves at each frequency f . For a particular frequency f = fc , s(t) = cos(2πfc t)
and
Z ∞
S(fc ) = cos(2πfc t) cos(2πfc t)dt
−∞
Z ∞
= 0.5[1 + cos(4πfc t)]dt
−∞

The cosine waves are in phase for all time, the product of the two cosine waves
contains DC, thus the integral integrates DC over all time, resulting in infinity (delta
functions) at f = fc . For all other frequencies f 6= fc the cosine waves drift in and
out of phase over time, there is no DC component, and the integral is zero. Thus for
INSTRUCTOR COPY

s(t) = cos(2πfc t) we can write:

Z ∞
S(f ) = cos(2πfc t) cos(2πf t)dt
−∞
= [δ(f − fc ) + δ(f + fc )]/2

For a digital spectrum analyzer where the signal is digitized (sampled) by an

analog-to-digital converter (ADC), the Fourier transform is done using the Fast
Fourier Transform (FFT) algorithm.

There is a practical upper limit to the sampling rate of an ADC. Thus analog
spectrum analyzers are used for high frequencies. For an analog spectrum analyzer,
frequency scanning is accomplished by electronically tuning a bandpass filter network
across the desired frequency range. In practice, a sweep generator along with a fixed
frequency bandpass filter, rather than tuning the bandpass filter, as shown in fig. D.2
(taken from an Agilent app note).

The amplitudes of all the signals in the bandpass area (at each point during the
scan) are measured to provide the amplitude versus frequency display.

It is important to note that in order to detect two signals which are narrowly
spaced, the frequency span of the bandpass filter must be set appropriately. This
concept is shown in fig. D.3. The frequencies f1 , f2 and f3 in the input spectrum are
summed into a single peak on the displayed waveform due to the excessive frequency
span. Note that the DC reference shown in this figure is a characteristic of the
spectrum analyzer. It is always present, regardless of whether there is a DC bias on
the input signal.

Unlike an oscilloscope, the amplitude scale on a spectrum analyzer can be set to a

logarithmic scale (for power measurements) as well as a linear scale (for voltage mea-
 APPENDIX D. APPLICATIONS 541
INSTRUCTOR COPY

Figure D.2: Block diagram of spectrum analyzer (http://cp.literature.agilent.

com/litweb/pdf/5952-0292.pdf)

Figure D.3: Effect of Span on Spectrum Analyzer Display

 APPENDIX D. APPLICATIONS 542

surements). The logarithmic scale is represented in dBm, or decibels with reference

to a milliwatt.  
Pin
PdBm = 10 log
1 mW

It can then be shown that

 
V2 1
PdBm = 10 log ·
Z 1 mW

where V is the RMS voltage.

If the load is a standard 50 Ω input,

1
INSTRUCTOR COPY

PdBm = 20 log V + 10 log

50 × 10−3
= 20 log V + 13

So the relationship between input power and input voltage is

PdBm = 20 log V + 13

This is a handy formula worth remembering.

On the logarithmic dB scale, the power of signals is always positive. It is the

magnitude of the signals above (in reference to) the noise level that is of interest. The
analog spectrum analyzer shows only the square magnitude (or power) of each Fourier
component. A digital spectrum analyzer can show phase as well as amplitude. The
block diagram of the digital spectrum analyzer is the same as the general I-Q receiver,
followed by an ADC (analog-to-digital converter), FFT processor and display.

D.2 Digital filters as applied to pulse shaping

D.3 dB and ratios

 APPENDIX D. APPLICATIONS 543

Digital  filters  as  applied  to  pulse  shaping  

 

Table  of  Contents  

Digital  filter  review  .................................................................................................................  1  
1.  Discrete  time  systems  .................................................................................................................  1  
2.  Digital  filters  with  Finite  Impulse  Response  (FIR)  ...........................................................  2  
3.  z  transforms  and  transfer  function  ........................................................................................  3  
4.  Feedback  -­‐  Infinite  Impulse  Response  (IIR)  filters  ...........................................................  4  
INSTRUCTOR COPY

5.  Frequency  response  of  digital  filters  .....................................................................................  5  

Representation  of  pulse  shaping  filter  as  a  digital  filter  ............................................  6  
Pulse  shaping  implemented  with  digital  filtering  ........................................................  6  
Digital  messages  with  impulses  in  a  sampled  system  ..........................................................  7  
Digital  message  waveform  with  pulse  shaping  ......................................................................  8  
 

Digital  filter  review    

1.  Discrete  time  systems  
 
A  sampled  system  is  illustrated  below,  where  a  signal  x(t)  is  represented  by  a  
sequence  of  samples  x(n)  
 

Figure D.4: Digital filters

 APPENDIX D. APPLICATIONS 544

dB and ratios, dBx, power, voltage

dB is a ratio of two quantities
Given a system with an input signal of power Pin and output signal with
power Pout then we express the ratio in dB as 10log(Pout/Pin)

If the signals are given as voltages Vin and Vout

then the power ratio in dB is 20log(Vout/Vin)
log here is log to the base 10
INSTRUCTOR COPY

dBx represents a power P (in watts) of some signal relative to some

reference power such as 1 watt or 1 milliwatt.
dBW = 10log P, where P is in units of watts
dBm = 10log P, where P is in units of milliwatts

The table below can be read as (power ratio, dB), or

(absolute power relative to x, dBx)

power ratio dB power ratio dB

0.000001 -60 1,000,000 60
0.00001 -50 100,000 50
0.0001 -40 10,000 40
0.001 -30 1,000 30
0.002 -27 500 27
0.005 -23 200 23
0.01 -20 100 20
0.0125 -19 80 19
0.02 -17 50 17
0.025 -16 40 16
0.033 -15 30 15
0.04 -14 25 14
0.05 -13 20 13
0.07 -12 15 12
0.08 -11 12 11
0.1 -10 10 10
0.125 -9 8 9
0.2 -7 5 7
0.25 -6 4 6
0.33 -5 3 5
0.4 -4 2.50 4
0.5 -3 2.00 3
0.7 -2 1.50 2
0.8 -1 1.25 1
1 0 1.00 0

Figure D.5: dB and ratios

 Digital  filters  as  applied  to  pulse  shaping  
 

Table  of  Contents  

Digital  filter  review  .................................................................................................................  1  
1.  Discrete  time  systems  .................................................................................................................  1  
2.  Digital  filters  with  Finite  Impulse  Response  (FIR)  ...........................................................  2  
3.  z  transforms  and  transfer  function  ........................................................................................  3  
4.  Feedback  -­‐  Infinite  Impulse  Response  (IIR)  filters  ...........................................................  4  
5.  Frequency  response  of  digital  filters  .....................................................................................  5  
Representation  of  pulse  shaping  filter  as  a  digital  filter  ............................................  6  
Pulse  shaping  implemented  with  digital  filtering  ........................................................  6  
Digital  messages  with  impulses  in  a  sampled  system  ..........................................................  7  
INSTRUCTOR COPY

Digital  message  waveform  with  pulse  shaping  ......................................................................  8  

 

Digital  filter  review    

1.  Discrete  time  systems  
 
A  sampled  system  is  illustrated  below,  where  a  signal  x(t)  is  represented  by  a  
sequence  of  samples  x(n)  
 

 
  
A  discrete  time  system  h(n)  has  an  impulse  response  consisting  of  the  sum  of  
weighted  delayed  samples  
 

 
 
INSTRUCTOR COPY

 
 
Samples  of  a  signal  can  be  added  
 

 
 

2.  Digital  filters  with  Finite  Impulse  Response  (FIR)  

 
Samples  can  be  delayed  and  multiplied  by  a  constant  and  added  to  make  a  digital  
filter  
 
  
 
This  example  filter  (system)  is  an  averager  that  takes  the  average  of  3  samples.    It’s  
function  is  to  smooth  a  signal  or  low  pass  filter  a  signal.  
INSTRUCTOR COPY

 
The  above  system  is  an  example  of  an  Finite  Impulse  Response  (FIR)  filter.  
 
A  general  FIR  filter  shown  below  can  be  represented  by  a  difference  equation  which  
is  also  a  convolution  
 

 
 
y[n] = b0 x[n] + b1x[n − 1] + b2 x[n − 2]
= ∑ bk x[n − k] = ∑ b[k]x[n − k]    
k=0 k=0

= b[k]⊗ x[k]
The  impulse  response  of  this  filter  is   bk = b[k]!for!all!k!or!written!as!vector![b0b1b2 ...]    
!
3.  z  transforms  and  transfer  function  
 
For  a  given  input  sequence  x(n)  we  can  use  the  definition  of  the  z-­‐transform  and  
represent  it  as  X(z).  
  
X(z) = ∑ x[n]z −k  
! k=0
 
The  transfer  function  of  the  FIR  system  (filter)  is  obtained  using  the  definition  and  
properties  of  the  z-­‐transform  
 
x[n]↔ X(z)
−k
   
! x[n − k]↔ z X(z)
 
to  obtain  
 
y[n] = b0 x[n] + b1x[n − 1] + b2 x[n − 2] + ...
Y (z) = b0 X (z) + b1z −1 X (z) + b2 z −2 X (z) + ...
INSTRUCTOR COPY

Y (z)    
H (z) = = b0 + b1z −1 + b2 z −2 + ... = ∑ bk z − k
X (z) k=0

H (z) = ∑ bk z − k thus for FIR filter hk = bk

k=0

 
 
From  the  above  we  see  that    
 
y[n]= h[n]⊗ x[n]
   
! Y(z) = H(z)X(z)
 
so  that  convolution  in  the  time  domain  is  multiplication  in  the  z-­‐(frequency)  domain.  
 
Thus  with  a  system  (filter)  h(n)  and  input  x(n),  we  can  compute  the  output  y(n)  
either  by  convolution  or  by  taking  z-­‐transforms,  multiplying  and  taking  the  inverse  
z-­‐transform.  
 
The  convolution  method  is  well  suited  to  computer  calculation,  but  yields  no  insight.  
You  just  get  numbers  y(n)  for    finite  number  of  values  of  n  when  you  stop  
computing.  
The  z-­‐transform  method  yields  a  closed  form  solution  for  the  output  for  all  n  and  
can  yield  insight  as  to  the  behaviour  of  the  output.    
 

4.  Feedback  -­‐  Infinite  Impulse  Response  (IIR)  filters    

 
If  feedback  is  added,  the  filter  is  called  an  Infinite  Impulse  Response  (IIR)  filter    
 
INSTRUCTOR COPY

 
 
The  IIR  filter  has  a  difference  equation    
 
y[n] + a1 y[n − 1] + a2 y[n − 2] + ... = b0 x[n] + b1x[n − 1] + b2 x[n − 2] + ...    
 
or  
 
y[n]+ ∑ ak y[n − k]= ∑ bk x[n − k]    
! k=1 k=0
   
The  transfer  function  is  obtained  using  the  definition  and  properties  of  the  z-­‐
transform  to  obtain  
 

Y (z) b0 + b1z −1 + b2 z −2 + ... ∑ bk z − k

H (z) = = = k=0    
X (z) 1+ a1z + a2 z + ... 1+ ∑ ak z − k
−1 −2

k=1
 
 

5.  Frequency  response  of  digital  filters  

 
The  frequency  response  of  a  filter  is      found  by  evaluating  the  transfer  function  for  
values  of  z  on  the  unit  circle.  
 
H( f ) = H(z)|z=e jw
   
w = 2π fTs = 2π f / f s
!
  
Obtaining  an  expression  for  the  frequency  response  by  analysis  can  be  tedious  or  
intractable.      Thus  we  use  computer  tools  to  plot  the  frequency  response.  
 
The  Matlab  representation  is   y = filter(b, a, x)  for  a  general  filter,  where  we  define  
the  vectors   a = [1 a1 a2 ], b = [1 b1 b2 ] ,  see  
www.mathworks.com/help/matlab/ref/filter.html  
 
The  filter  characteristics  (including  frequency  response,  phase  response,  phase  
delay,  group  delay,  pole-­‐zero  plot,  impulse  response)  can  be  viewed  using  the  
Matlab  command  fvtool(b,a),  see  
www.mathworks.com/help/signal/ref/fvtool.html  
 
INSTRUCTOR COPY

Representation  of  pulse  shaping  filter  as  a  digital  filter  

 
In  this  section  we  illustrate  how  a  pulse  shaping  filter  can  be  represented  as  a  digital  
filter  with  impulse  response  equal  to  the  (sampled)  pulse  shape.  
 
Consider    the  RC2  pulse  with  symbol  time  T  and  a  digital  system  with  sampling  rate  
f = 1/Ts = N /T !or!T = NTs    Thus  there  are  N  samples  per  symbol  time.    For  the  RC2  
! s
pulse  there  will  be  2N  samples  needed  to  represent  the  pulse.  
 
We  can  determine  the  numerical  values  of  the  samples  representing  the  pulse  as  
follows.  
 
p(t ) = p2(t )
= 0.5cos(1− πt /T ),!for!0 ≤ t ≤ 2T
= 0.5cos(1− π kTs / NTs )!for!t = kTs !and!T = NTs    
= 0.5cos(1− π k / N),!for!0 ≤ n ≤ 2N
= p (t = kTs ) = p2(k) = bk = hk
! 2
 
The  impulse  response  of  the  pulse-­‐shaping  filter  is  thus  written  
 
h = b = p2(k) = 0.5cos(1− π k / N),!for!0 ≤ k ≤ 2N  
!k k
 
The  values  of  the  pulse  shaping  FIR  filter  taps   bk  are  given  by  this  equation.  
!

Pulse  shaping  implemented  with  digital  filtering  

 
 Digital  messages  with  impulses  in  a  sampled  system  
 
A  digital  message  can  be  represented  by  a  sequence  of  impulses  
 
mδ (t ) = A0δ (t ) + A1δ (t − T ) + A2δ (t − 2T ) + ... = ∑ Akδ (t − kT )  
k

A
! k !δ (t )
with  the  weights  being  the  data    and  the  pulses      
 
A = A(kT )
! k
This  expression  appears  like  a  convolution  when  we  write      and  
 
m|δ (t) = ∑ A(kT )δ (t − kT )
k  
INSTRUCTOR COPY

 
For  a  digital  implementation  at  sampling  rate   f s = 1/Ts  ,  the  time  becomes  discrete  
!
 
t = nTs  so  that    
!
 
mδ (nTs ) = ∑ A(kT )δ (nTs − kT )
k  
T = NTs !or!T = N / f s !or!N = f sT
!
Assuming  N  samples  per  symbol  or    we  write  the  
convolution  
 
mδ (nTs ) = ∑ A(kNTs )δ (nTs − kNTs )
k  
T
!s
or  omitting  the  sampling  time    as  is  usually  done  for  sampled  systems,  we  write  
 
mδ (n) = ∑ A(kN )δ (n − kN )
k  
 
The  data   Ak = A(k) = A(kNTs )  is  valid  at  times   t = kNTs  and  changes  in  general  for  
! !
each  integer  values  of  k  
 
For  a  particular  value  of  k,  the  delta  function  delayed  by  kN  samples  is  
 
  δ (nTs − kNTs ) = δ ((n − kN)Ts ) = δ (n − kN) = 0 !for!all!n ≠ kN  and  equals  1  for  the  one  
!
sample  where  !n = kN  
 
 Thus  in  the  discrete  time  sampled  system,  the  sum  of  the  weighted  delayed  delta  
functions    
 
mδ (n) = ∑ A(kN )δ (n − kN )
k  
!n ≠ kN
is  zero  for  all      for  all  integer  values  of  k.    The  only  non-­‐zero  samples  are  every  
Nth  sample  n=kN
 

Digital  message  waveform  with  pulse  shaping  

 
 
The  digital  message  waveform  m(t)  is  the  convolution  of  the  message  impulse  
sequence  with  the  pulse  shaping  filter  
INSTRUCTOR COPY

 
m(t ) = mδ (t )⊗ p(t )    
!
 
and  thus  is  written  as  a  sum  of  weighted  and  delayed  pulses  
 
 
 
m(t) =
A0 (t) + A1 p2 (t − T ) + A2 p(t − 2T ) + ... = ∑ Ak p(t − kT )
k  
A
! k !p(t )
with  the  weights  being  the  data    and  the  pulses      
A = A(kT )
! k
This  expression  appears  like  a  convolution  when  we  write      and  
 
m(t) = ∑ A(kT ) p(t − kT )
k  
 
t = nTs   f = 1/Ts = N /T !or!T = NTs
! ! s
In  discrete  time  notation     with  sampling  rate   we  
write  the  convolution  
 
m(n) = mδ (n)⊗ p(n) = ∑ mδ (k)p(n − k)  
! k
 
mδ (k) = ∑ A(qN )δ (k − qN )
Substituting   q  
 
 and  using  the  variable  q  instead  of  k  
 
m(n) = ∑ ∑ A(qN )δ (k − qN ) p(n − k)
k q  
m(n) = ∑ A(kN ) p(n − kN )
k  
 

   r   esult  can  be  seen  another  way.  

This  
For  a  digital  implementation  at  sampling  rate   f s = 1/Ts  ,  the  time  becomes  discrete  
!
 
t = nTs    so  that    
!
 
INSTRUCTOR COPY

m(nTs ) = ∑ A(kT ) p(nTs − kT )

k  
T = NTs
!
Assuming  N  samples  per  symbol  or    we  write  the  convolution  
 
m(nTs ) = ∑ A(kNTs ) p(nTs − kNTs )
k  
 
T
!s
Omitting  the  sampling  time    as  is  usually  done  for  sampled  systems,  we  write  
 
m(n) = ∑ A(kN ) p(n − kN )
k  
 
 
The  data   Ak = A(k) = A(kNTs )  is  valid  at  times   t = kNTs  and  changes  in  general  for  
! !
each  integer  values  of  k  
 
 
For  a  particular  value  of  k    
 
the  pulse  delayed  by  kN  is  
 
  p(nTs − kNTs ) = p((n − kN)Ts ) = p(n − kN) = 0 !for!all!n ≠ {kN ,(k + 2)N}!for!RC2!pulse    
!
 
Thus  in  the  discrete  time  sampled  system,  the  sum  of  the  weighted  delayed  pulses  
 
m(n) = ∑ A(kN ) p(n − kN )
k contains  overlapping  pulses.