Ivan 

B. Djordjevic

Advanced Optical
and Wireless
Communications
Systems
Second Edition
Advanced Optical and Wireless Communications
Systems
Ivan B. Djordjevic

Advanced Optical
and Wireless Communications
Systems

Second Edition
Ivan B. Djordjevic
Department of Electrical
and Computer Engineering
University of Arizona
Tucson, AZ, USA

ISBN 978-3-030-98490-8 ISBN 978-3-030-98491-5 (eBook)

https://doi.org/10.1007/978-3-030-98491-5

# The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2018,
2022
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or
part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and
retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter
developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not
imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and
regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to
be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty,
expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been
made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AGThe registered company
address is: Gewerbestrasse 11, 6330 Cham, Switzerland
To my family
Preface

The purpose of this book is to introduce the reader to most advanced topics of (i) wireless
communications, (ii) free-space optical (FSO) communications, (iii) indoor optical wireless
(infrared, IR) communications, and (iv) fiber-optics communications. So far, these different
types of communication systems have been considered as separate disciplines. However, the
fundamental concepts, such as propagation principles, modulation formats, channel coding,
diversity principles, MIMO signal processing, multicarrier modulation, equalization, adaptive
modulation and coding, detection principles, and software-defined transmission, are common.
The key idea of this book is to consider these different types of communication systems in a
unified fashion. The fundamental concepts, listed above, are described first, followed by
detailed description of each particular system. The book is self-contained and structured to
provide straightforward guidance to readers looking to capture fundamentals and gain theoreti-
cal and practical knowledge about wireless communications, free-space optical
communications, and fiber-optics communications, which can be readily applied in their
research and practical applications.
This book unifies wireless, free-space optical, IR, and fiber-optics communications. Unique
features of the book include the following:

• This book integrates wireless communications, free-space optical communications, indoor

IR (optical wireless) communications, and fiber-optics communications’ technologies.
• Use of the book does not require prior knowledge in communication systems.
• Use of the book does not require any prerequisite material except basic concepts on vector
algebra at undergraduate level.
• This book offers in-depth exposition on propagation effects in different media (free-space,
fiber-optics, atmospheric turbulence channels), channel impairments in these media, noise
sources, key components and modules enabling wireless and optical communications,
channel capacity studies, advanced modulation and multiplexing techniques, advanced
detection and channel compensation techniques, OFDM for wireless and optical
communications, diversity techniques, MIMO techniques, advanced coding and coded
modulation techniques, spread spectrum techniques, CDMA systems, multiuser detection,
ultra-wideband communications, and physical-layer security, to mention few.
• Multiple courses can be offered: advanced wireless communications, advanced fiber-optics
communications, advanced free-space optical communications, advanced indoor IR (wire-
less optical) communications, and advanced wireless and optical communication systems.

This book is intended for very diverse group of readers in communications engineering,
optical engineering, wireless communications, free-space optical communications, optical
wireless communications, mathematics, physics, communication theory, information theory,
devise electronics, photonics, as well as computer science.
The book comprises 11 chapters. Each chapter, except introductory one, contains more than
20 problems to help the reader get a deeper understanding of the corresponding text material. In
the introductory chapter (Chap. 1), we provide historical perspective of both wireless and

vii
viii Preface

optical communications systems, describe their fundamental concepts, identify various noise
sources and channel impairments, and briefly explain how to deal with them, with full details
provided in the chapters that follow. Chapter 2 is devoted to the detailed description of
propagation effects, noise sources, and channel impairments for wireless communications,
free-space optical communications, and fiber-optics communications. In Chap. 3, the basic
components, modules, and subsystems relevant in both optical and wireless communications
are described. Chapter 4 is devoted to both wireless channel and optical channel capacities.
Both memoryless and channels memory are considered. Chapter 5 is related to advanced
modulation and multiplexing techniques suitable for both wireless and optical communication
systems. Chapter 6 is related to advanced detection and channel impairments compensation
techniques applicable to both wireless and optical communication systems. Chapter 7 is
devoted to OFDM fundamentals and applications to wireless and optical communications. In
Chap. 8, various diversity and MIMO techniques are described, capable of improving single
input single output system performance, applicable to both wireless and optical communication
systems. Chapter 9 is related to the advanced channel coding and coded modulation techniques,
relevant to optical and wireless communication systems. Chapter 10 is devoted to spread
spectrum (SS), CDMA, and ultra-wideband (UWB) communication systems. Finally,
Chap. 11 provides detailed description of the physical-layer security concepts applied to
wireless and optical channels. The detailed description of chapters is provided in the
Introduction.
I would like to thank my former students, postdocs, and collaborators for the research we
performed together.
Finally, special thanks are extended to Mary E. James of Springer, USA, for her tremendous
effort in organizing the logistics of the book, in particular promotion and edition, which is
indispensable to make this book happen.

Tucson, AZ, USA Ivan B. Djordjevic

Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Historical Perspective of Optical and Wireless Communication
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Wireless Communication Historical Perspective . . . . . . . . . . . . . 1
1.1.2 Optical Communication Historical Perspective . . . . . . . . . . . . . . 4
1.2 Optical Communications Systems and Networks Fundamentals . . . . . . . . 6
1.3 Wireless Communications Systems Fundamentals . . . . . . . . . . . . . . . . . . 9
1.4 Organization of the Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2 Propagation Effects in Optical and Wireless Communications
Channels, Noise Sources, and Channel Impairments . . . . . . . . . . . . . . . . . . . . 21
2.1 Electromagnetic Field and Wave Equations . . . . . . . . . . . . . . . . . . . . . . 21
2.1.1 Vector Derivatives (grad, div, and curl) in Orthogonal
Curvilinear Coordinate System . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.1.2 Vector Derivatives (grad, div, and curl) in Cylindrical
Polar and Spherical Coordinate Systems . . . . . . . . . . . . . . . . . . 25
2.2 Propagation of Electromagnetic Waves . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2.1 Propagation of Plane Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2.2 Vectorial Nature of the Light, Snell’s Law of Refraction,
Reflection Coefficients, and Total Internal Reflection . . . . . . . . . 28
2.2.2.1 Polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.2.2 Snell’s Law of Refraction, Reflection Coefficients,
and Total Internal Reflection . . . . . . . . . . . . . . . . . . . 33
2.2.2.3 Optical Fibers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.2.4 Polarizing Elements . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.2.3 Electromagnetic Potentials and Electromagnetic Waves . . . . . . . 40
2.2.3.1 Electromagnetic Potentials . . . . . . . . . . . . . . . . . . . . . 40
2.2.3.2 Concept of Antenna . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2.3.3 Spherical Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.2.3.4 Plane Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2.3.5 Hertzian Dipole . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2.3.6 Straight-Wire Antenna . . . . . . . . . . . . . . . . . . . . . . . 47
2.2.3.7 Cylindrical Waves . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.2.3.8 Paraxial Approximation and Gaussian Beam . . . . . . . . 50
2.2.4 Interference, Coherence, and Diffraction in Optics . . . . . . . . . . . 51
2.2.4.1 Interference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.2.4.2 Coherence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.2.4.3 Diffraction Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.2.4.4 Fresnel and Fraunhofer Approximations . . . . . . . . . . . 60

ix
x Contents

2.2.5 Laser Beam Propagation Over the Atmospheric Turbulence

Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.2.5.1 Paraxial Wave Equation . . . . . . . . . . . . . . . . . . . . . . 63
2.2.5.2 Split-Step Beam Propagation Method . . . . . . . . . . . . . 63
2.2.5.3 Paraxial Fresnel and Fraunhofer Diffraction
Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.2.5.4 Rytov Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.2.5.5 Modeling Atmospheric Turbulence Effects . . . . . . . . . 68
2.2.5.6 Kolmogorov Theory of Turbulence and
Corresponding Generalizations . . . . . . . . . . . . . . . . . 69
2.2.5.7 Rytov Method of Small Perturbations . . . . . . . . . . . . . 77
2.2.5.8 Gamma-Gamma Distribution of Irradiance . . . . . . . . . 79
2.3 Propagation Effects in Wireless Communication Channels . . . . . . . . . . . . 85
2.3.1 Path Loss and Shadowing Propagation Effects in Wireless
Communication Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.3.1.1 Free-Space Path Loss and Doppler Shift . . . . . . . . . . . 86
2.3.1.2 Radar Range Equation . . . . . . . . . . . . . . . . . . . . . . . 88
2.3.1.3 Ray Tracing Approximation . . . . . . . . . . . . . . . . . . . 89
2.3.1.4 General Ray Tracing (GRT) Model . . . . . . . . . . . . . . 91
2.3.1.5 Scattering Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.3.1.6 Simplified Path Loss and Multislope Models . . . . . . . 96
2.3.1.7 Empirical Path Loss Models . . . . . . . . . . . . . . . . . . . 97
2.3.1.8 Shadowing Effects . . . . . . . . . . . . . . . . . . . . . . . . . . 98
2.3.1.9 Outage Probability and Cell Coverage Area . . . . . . . . 100
2.3.2 Statistical Multipath Wireless Communication Channel
Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
2.3.2.1 Multipath Channel Models . . . . . . . . . . . . . . . . . . . . 102
2.3.2.2 Narrowband Fading Model . . . . . . . . . . . . . . . . . . . . 105
2.3.2.3 Clark-Jakes Model . . . . . . . . . . . . . . . . . . . . . . . . . . 106
2.3.2.4 Envelope and Power Distributions in Narrowband
Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
2.3.2.5 Average Fade Duration . . . . . . . . . . . . . . . . . . . . . . . 110
2.3.2.6 Discrete-Time (DT) Model . . . . . . . . . . . . . . . . . . . . 112
2.3.2.7 Finite-State Markov Channel (FSMC) Model . . . . . . . 113
2.3.2.8 Wideband Channel Models . . . . . . . . . . . . . . . . . . . . 113
2.3.2.9 Power Delay Profile and Coherence Bandwidth . . . . . 115
2.3.2.10 Channel Coherence Time and Fading Channels
Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
2.4 Signal Propagation in Optical Fibers and Corresponding Channel
Impairments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
2.4.1 Fiber Attenuation and Insertion Losses . . . . . . . . . . . . . . . . . . . 118
2.4.2 Chromatic Dispersion Effects . . . . . . . . . . . . . . . . . . . . . . . . . . 119
2.4.3 Polarization Mode Dispersion (PMD) . . . . . . . . . . . . . . . . . . . . 125
2.4.4 Fiber Nonlinearities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
2.4.4.1 Fiber Nonlinearities Classification . . . . . . . . . . . . . . . 127
2.4.4.2 Effective Length and Effective Cross-Sectional
Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
2.4.4.3 Kerr Effect, Self-Phase Modulation, and Nonlinear
Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
2.4.4.4 Cross-Phase Modulation . . . . . . . . . . . . . . . . . . . . . . 129
2.4.4.5 Four-Wave Mixing . . . . . . . . . . . . . . . . . . . . . . . . . . 130
2.4.4.6 Raman Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
2.4.4.7 Brillouin Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Contents xi

2.4.5 Generalized Nonlinear Schrödinger Equation . . . . . . . . . . . . . . . 133

2.4.5.1 GNLSE for Single Channel Transmission . . . . . . . . . . 133
2.4.5.2 GNLSE for Multichannel Transmission . . . . . . . . . . . 135
2.4.5.3 GNLSE for Polarization Division Multiplexing
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
2.4.5.4 GNLSE for Spatial Division Multiplexing
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
2.5 Noise Sources in Optical Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
2.5.1 Mode Partition Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
2.5.2 Reflection-Induced Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
2.5.3 Relative Intensity Noise (RIN) and Laser Phase Noise . . . . . . . . 138
2.5.4 Modal Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
2.5.5 Thermal Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
2.5.6 Spontaneous Emission Noise and Noise Beating
Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
2.5.7 Quantum Shot Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
2.5.8 Dark Current Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
2.5.9 Crosstalk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
2.6 Indoor Optical Wireless Communication Channels . . . . . . . . . . . . . . . . . 144
2.6.1 Infrared Optical Wireless Communications . . . . . . . . . . . . . . . . 144
2.6.2 Visible Light Communications (VLCs) . . . . . . . . . . . . . . . . . . . 147
2.7 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
2.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
3 Components, Modules, and Subsystems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
3.1 Key Optical Components, Modules, and Subsystems . . . . . . . . . . . . . . . . 157
3.1.1 Optical Communications Basics . . . . . . . . . . . . . . . . . . . . . . . . 157
3.1.2 Optical Transmitters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
3.1.2.1 Lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
3.1.2.2 External Modulators . . . . . . . . . . . . . . . . . . . . . . . . . 164
3.1.2.3 I/Q Modulators and Polar Modulators . . . . . . . . . . . . 165
3.1.3 Optical Receivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
3.1.4 Optical Fibers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
3.1.5 Optical Amplifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
3.1.5.1 Semiconductor Optical Amplifier (SOA) . . . . . . . . . . 173
3.1.5.2 Raman Amplifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
3.1.5.3 Erbium-Doped Fiber Amplifier (EDFA) . . . . . . . . . . . 174
3.1.6 Optical Processing Components and Modules . . . . . . . . . . . . . . 175
3.1.6.1 Optical Couplers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
3.1.6.2 Optical Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
3.1.6.3 WDM Multiplexers and Demultiplexers . . . . . . . . . . . 177
3.1.7 Principles of Coherent Optical Detection . . . . . . . . . . . . . . . . . . 178
3.1.8 Optical Hybrids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
3.1.9 Coherent Optical Balanced Detectors . . . . . . . . . . . . . . . . . . . . . 180
3.2 Modules/Components Relevant to Wireless Communications . . . . . . . . . . 183
3.2.1 DSP Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
3.2.1.1 The Role of System (Transfer) Function and
Relationship Between DT and CT Fourier
Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
3.2.1.2 Discrete-Time System Realizations . . . . . . . . . . . . . . 185
3.2.2 Direct Digital Synthesizer (DDS) . . . . . . . . . . . . . . . . . . . . . . . 187
3.2.3 Multirate DSP and Resampling . . . . . . . . . . . . . . . . . . . . . . . . . 189
xii Contents

3.2.3.1 Downsampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190

3.2.3.2 Upsampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
3.2.3.3 Resampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
3.2.3.4 Noble Identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
3.2.3.5 Cascaded-Integrator-Comb (CIC) Filter-Based
Upsampling and Downsampling . . . . . . . . . . . . . . . . 192
3.2.3.6 Polyphase Decomposition . . . . . . . . . . . . . . . . . . . . . 193
3.2.3.7 Polyphase Filter-Based Resampling . . . . . . . . . . . . . . 193
3.2.4 Antenna Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
3.2.5 Automatic Gain Control (AGC) . . . . . . . . . . . . . . . . . . . . . . . . 200
3.3 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
3.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
4 Capacities of Wireless and Optical Channels . . . . . . . . . . . . . . . . . . . . . . . . . 209
4.1 Mutual Information, Channel Capacity, and Information Capacity
Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
4.1.1 Mutual Information and Information Capacity . . . . . . . . . . . . . . 209
4.1.2 Capacity of Continuous Channels . . . . . . . . . . . . . . . . . . . . . . . 211
4.2 Capacity of Flat-Fading and Frequency-Selective Wireless Fading
Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
4.2.1 Flat-Fading Channel Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . 214
4.2.1.1 Channel Side Information at Receiver . . . . . . . . . . . . 215
4.2.1.2 Full CSI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
4.2.2 Frequency-Selective Fading Channel Capacity . . . . . . . . . . . . . . 221
4.2.2.1 Time-Invariant Channel Capacity . . . . . . . . . . . . . . . . 221
4.2.2.2 Time-Variant Channel Capacity . . . . . . . . . . . . . . . . . 223
4.3 Capacity of Channels with Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
4.3.1 Markov Sources and Their Entropy . . . . . . . . . . . . . . . . . . . . . . 225
4.3.2 McMillan Sources and Their Entropy . . . . . . . . . . . . . . . . . . . . 228
4.3.3 McMillan-Khinchin Model for Channel Capacity Evaluation . . . . 228
4.4 Calculation of Information Capacity by the Forward Recursion
of the BCJR Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
4.5 Information Capacity of Systems with Coherent Optical Detection . . . . . . 232
4.6 Hybrid Free-Space Optical (FSO)-RF Channel Capacity . . . . . . . . . . . . . 235
4.6.1 Hybrid FSO-RF System Model Description . . . . . . . . . . . . . . . . 235
4.6.2 Adaptive Modulation and Coding (AMC) in Hybrid FSO-RF
Communications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
4.7 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
4.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
5 Advanced Modulation and Multiplexing Techniques . . . . . . . . . . . . . . . . . . . 249
5.1 Signal Space Theory in Wireless Communications . . . . . . . . . . . . . . . . . 249
5.1.1 Geometric Representation of Signals . . . . . . . . . . . . . . . . . . . . . 250
5.1.2 Modulators and Demodulators . . . . . . . . . . . . . . . . . . . . . . . . . 252
5.1.3 Frequency-Shift Keying (FSK) . . . . . . . . . . . . . . . . . . . . . . . . . 255
5.1.4 M-ary Pulse Amplitude Modulation (PAM) . . . . . . . . . . . . . . . . 256
5.1.5 Passband Digital Wireless/Optical Transmission . . . . . . . . . . . . . 258
5.2 Multilevel (Two-Dimensional) Modulation Schemes . . . . . . . . . . . . . . . . 259
5.2.1 Two-Dimensional Signal Constellations for Wireless
Communications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
5.2.2 M-ary PSK Transmitters for Optical Communications . . . . . . . . 265
5.2.3 Star-QAM Transmitters for Optical Communications . . . . . . . . . 266
Contents xiii

5.2.4 Square- and Cross-QAM Transmitters for Optical

Transmission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
5.3 Multicarrier Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
5.3.1 Multicarrier Systems with Nonoverlapping Subcarriers . . . . . . . . 271
5.3.2 Multicarrier Systems with Overlapping Subcarriers . . . . . . . . . . . 272
5.3.3 Dealing with Fading Effects on Subcarrier Level . . . . . . . . . . . . 274
5.4 MIMO Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
5.5 Polarization-Division Multiplexing (PDM) and 4-D Signaling . . . . . . . . . 277
5.6 Spatial-Division Multiplexing and Multidimensional Signaling . . . . . . . . 281
5.6.1 SDM in Wireless Communications . . . . . . . . . . . . . . . . . . . . . . 281
5.6.2 SDM and Multidimensional Signaling in Fiber-Optic
Communications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
5.6.3 SDM and Multidimensional Signaling in Free-Space Optical
(FSO) Communications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
5.7 Optimum Signal Constellation Design . . . . . . . . . . . . . . . . . . . . . . . . . . 286
5.7.1 Iterative Polar Modulation (IPM) . . . . . . . . . . . . . . . . . . . . . . . . 288
5.7.2 Signal Constellation Design for Rotationally Symmetric
Optical Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
5.7.3 Energy-Efficient Signal Constellation Design . . . . . . . . . . . . . . . 290
5.7.4 Optimum Signal Constellation Design (OSCD) . . . . . . . . . . . . . 291
5.8 Nonuniform Signaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
5.9 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
5.10 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
6 Advanced Detection Techniques and Compensation of Channel
Impairments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
6.1 Detection and Estimation Theory Fundamentals . . . . . . . . . . . . . . . . . . . 303
6.1.1 Geometric Representation of Received Signals . . . . . . . . . . . . . . 304
6.1.2 Correlation and Matched Filter-Based Receivers
Maximizing the Signal-to-Noise Ratio . . . . . . . . . . . . . . . . . . . . 308
6.1.3 Optimum and LLR Receivers . . . . . . . . . . . . . . . . . . . . . . . . . . 310
6.1.4 Symbol Error Probability Calculation . . . . . . . . . . . . . . . . . . . . 313
6.1.4.1 Calculation of Symbol Error Probability . . . . . . . . . . . 313
6.1.4.2 Calculation of Bit Error Probability of Binary
Signal Constellations . . . . . . . . . . . . . . . . . . . . . . . . 315
6.1.4.3 Calculation of Symbol Error Probability of M-ary
QAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
6.1.4.4 Union-Bound Approximation . . . . . . . . . . . . . . . . . . 317
6.1.5 Estimation Theory Fundamentals . . . . . . . . . . . . . . . . . . . . . . . 319
6.2 Wireless Communication Systems Performance . . . . . . . . . . . . . . . . . . . 321
6.2.1 Outage Probability Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
6.2.2 Average Error Probability Scenario . . . . . . . . . . . . . . . . . . . . . . 321
6.2.3 Combined Outage and Average Error Probability Scenario . . . . . 323
6.2.4 Moment-Generating Function (MGF)-Based Approach
to Average Error Probability Calculation . . . . . . . . . . . . . . . . . . 323
6.2.5 Performance Evaluation in the Presence of Doppler Spread
and Fading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
6.3 Channel Equalization Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
6.3.1 ISI-Free Digital Transmission and Partial-Response Signaling . . . 325
6.3.2 Zero-Forcing Equalizers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
6.3.3 Optimum Linear Equalizer in MMSE Sense . . . . . . . . . . . . . . . . 333
6.3.4 Wiener Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
6.3.5 Adaptive Equalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
xiv Contents

6.3.6 Decision-Feedback Equalization . . . . . . . . . . . . . . . . . . . . . . . . 337

6.3.7 MLSD (MLSE) or Viterbi Equalization . . . . . . . . . . . . . . . . . . . 338
6.3.8 Blind Equalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
6.4 Synchronization Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
6.5 Adaptive Modulation Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
6.5.1 Variable-Power Variable-Rate Modulation Techniques . . . . . . . . 348
6.5.2 Adaptive Coded Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
6.6 Volterra Series-Based Equalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
6.7 Digital Backpropagation in Fiber-Optics Communications . . . . . . . . . . . . 356
6.8 Optical Communication Systems with Coherent Optical Detection . . . . . . 359
6.8.1 Balanced Coherent Optical Detection for 2-D Modulation
Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
6.8.2 Polarization Diversity and Polarization-Division
Demultiplexing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
6.8.3 Homodyne Coherent Optical Detection Based on PLLs . . . . . . . 365
6.8.4 Phase Diversity Receivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
6.8.5 Dominant Coherent Optical Detection Noise Sources . . . . . . . . . 368
6.8.5.1 Laser Phase Noise . . . . . . . . . . . . . . . . . . . . . . . . . . 368
6.8.5.2 Polarization Noise . . . . . . . . . . . . . . . . . . . . . . . . . . 368
6.8.5.3 Transimpedance Amplifier Noise . . . . . . . . . . . . . . . . 370
6.8.5.4 Coherent Optical Detection in the Presence
of ASE Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
6.9 Compensation of Atmospheric Turbulence Effects . . . . . . . . . . . . . . . . . 372
6.9.1 Adaptive Optics Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
6.9.1.1 Zernike Representation of Atmospheric
Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
6.9.1.2 Shack-Hartmann Wavefront Sensor . . . . . . . . . . . . . . 375
6.9.1.3 Wavefront Correctors . . . . . . . . . . . . . . . . . . . . . . . . 376
6.9.2 SLM-Based Backpropagation Method . . . . . . . . . . . . . . . . . . . . 378
6.10 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
6.11 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
7 OFDM for Wireless and Optical Communications . . . . . . . . . . . . . . . . . . . . . 389
7.1 Introduction, OFDM Basics, and Generation of Subcarriers
Using Inverse FFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
7.1.1 Introduction to OFDM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
7.1.2 OFDM Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
7.1.3 Generation of OFDM Signals . . . . . . . . . . . . . . . . . . . . . . . . . . 391
7.2 Guard Time, Cyclic Extension, and Windowing . . . . . . . . . . . . . . . . . . . 392
7.2.1 Guard Time and Cyclic Extension . . . . . . . . . . . . . . . . . . . . . . . 392
7.2.2 Windowing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
7.3 Bandwidth Efficiency of OFDM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
7.4 OFDM Parameter Selection, OFDM Building Blocks, and Parallel
Channel Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
7.4.1 OFDM Parameter Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
7.4.2 OFDM Building Blocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
7.4.3 OFDM Parallel Channel Decomposition and Channel
Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
7.5 CO-OFDM Principles, DFT Windowing, Frequency Synchronization,
Phase Estimation, and Channel Estimation . . . . . . . . . . . . . . . . . . . . . . . 404
7.5.1 Principles of Coherent Optical OFDM (CO-OFDM) . . . . . . . . . . 404
7.5.2 DFT Window Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . 405
7.5.3 Frequency Synchronization in OFDM Systems . . . . . . . . . . . . . 406
Contents xv

7.5.4 Phase Estimation in OFDM Systems . . . . . . . . . . . . . . . . . . . . . 407

7.5.5 Channel Estimation in OFDM Systems . . . . . . . . . . . . . . . . . . . 407
7.5.5.1 Pilot-Aided Channel Estimation . . . . . . . . . . . . . . . . . 408
7.5.5.2 Data-Aided Channel Estimation . . . . . . . . . . . . . . . . . 411
7.6 Differential Detection in OFDM Systems . . . . . . . . . . . . . . . . . . . . . . . . 411
7.7 OFDM Applications in Wireless Communications . . . . . . . . . . . . . . . . . 413
7.7.1 OFDM in Digital Audio Broadcasting (DAB) . . . . . . . . . . . . . . 413
7.7.2 Coded-OFDM in Digital Video Broadcasting (DVB) . . . . . . . . . 417
7.8 OFDM for Wi-Fi, LTE, and WiMAX . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
7.9 OFDM in Ultra-Wideband Communication (UWC) . . . . . . . . . . . . . . . . . 429
7.10 Optical OFDM Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430
7.10.1 Optical OFDM System Types . . . . . . . . . . . . . . . . . . . . . . . . . . 430
7.10.1.1 Coherent Optical OFDM (CO-OFDM) Systems . . . . . 431
7.10.1.2 Direct Detection Optical OFDM (DDO-OFDM) . . . . . 432
7.10.2 High-Speed Spectrally Efficient CO-OFDM Systems . . . . . . . . . 433
7.10.2.1 Polarization-Division Multiplexed OFDM . . . . . . . . . 434
7.10.2.2 OFDM-Based Superchannel Transmission . . . . . . . . . 436
7.10.3 OFDM in Multimode Fiber Links . . . . . . . . . . . . . . . . . . . . . . . 438
7.11 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
7.12 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
8 Diversity and MIMO Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
8.1 Diversity Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
8.1.1 Basic Diversity Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
8.1.2 Receiver Diversity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451
8.1.2.1 Selection Combining . . . . . . . . . . . . . . . . . . . . . . . . . 452
8.1.2.2 Threshold Combining . . . . . . . . . . . . . . . . . . . . . . . . 455
8.1.2.3 Maximum-Ratio Combining . . . . . . . . . . . . . . . . . . . 456
8.1.2.4 Equal-Gain Combining . . . . . . . . . . . . . . . . . . . . . . . 460
8.1.3 Transmitter Diversity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
8.1.3.1 Transmitter Diversity when CSI Is Available on
Transmitter Side . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
8.1.3.2 Transmitter Diversity when CSI Is Not Available:
Alamouti Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
8.2 MIMO Optical and Wireless Techniques . . . . . . . . . . . . . . . . . . . . . . . . 464
8.2.1 MIMO Wireless and Optical Channel Models . . . . . . . . . . . . . . 465
8.2.2 Parallel Decomposition of Optical and Wireless MIMO
Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 468
8.2.3 Space-Time Coding: ML Detection, Rank Determinant,
and Euclidean Distance Criteria . . . . . . . . . . . . . . . . . . . . . . . . . 471
8.2.4 Relevant Classes of Space-Time Codes (STCs) . . . . . . . . . . . . . 476
8.2.4.1 Alamouti Code (Revisited) and Orthogonal Designs . . 476
8.2.4.2 Linear Space-Time Block Codes . . . . . . . . . . . . . . . . 478
8.2.4.3 Space-Time Trellis Codes (STTCs) . . . . . . . . . . . . . . 479
8.2.5 Spatial Division Multiplexing (SDM) . . . . . . . . . . . . . . . . . . . . 480
8.2.5.1 BLAST Encoding Architectures . . . . . . . . . . . . . . . . . 482
8.2.5.2 Multi-group Space-Time Coded Modulation
(MG-STCM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484
8.2.6 Linear and Decision-Feedback MIMO Receivers
for Uncoded Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487
8.2.7 Suboptimum MIMO Receivers for Coded Signals . . . . . . . . . . . 488
8.2.7.1 Linear and Zero-Forcing Receivers (Interfaces) . . . . . . 488
8.2.7.2 Linear MMSE Receiver (Interface) . . . . . . . . . . . . . . 489
xvi Contents

8.2.7.3 Decision-Feedback and BLAST Receivers

(Interfaces) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
8.2.7.4 Spatial Interference Cancellation by Iterative
Receiver (Interface) . . . . . . . . . . . . . . . . . . . . . . . . . 491
8.3 Iterative MIMO Receivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
8.3.1 Factor Graph Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
8.3.2 Factor Graphs for MIMO Channels and Channels
with Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
8.3.3 Sum-Product Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495
8.3.4 Sum-Product Algorithm for Channels with Memory . . . . . . . . . . 496
8.3.5 Iterative MIMO Receivers for Uncoded Signals . . . . . . . . . . . . . 496
8.3.6 Factor Graphs for Linear Block and Trellis Channel Codes . . . . . 498
8.3.7 Iterative MIMO Receivers for Space-Time Coded Signals . . . . . . 501
8.4 Broadband MIMO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 502
8.4.1 MIMO-OFDM Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503
8.4.2 Space-Frequency Block Coding-Based MIMO . . . . . . . . . . . . . . 504
8.5 MIMO Channel Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505
8.5.1 Capacity of Deterministic (Static) MIMO Channels . . . . . . . . . . 505
8.5.2 Capacity of Random MIMO Channels . . . . . . . . . . . . . . . . . . . . 507
8.5.2.1 Ergodic Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . 507
8.5.2.2 Non-ergodic Capacity . . . . . . . . . . . . . . . . . . . . . . . . 509
8.5.2.3 Correlated Fading Channel Capacity . . . . . . . . . . . . . 510
8.5.2.4 MIMO-OFDM Channel Capacity . . . . . . . . . . . . . . . 510
8.6 MIMO Channel Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511
8.6.1 Maximum Likelihood (ML) MIMO Channel Estimation . . . . . . . 511
8.6.2 Least Squares (LS) MIMO Channel Estimation . . . . . . . . . . . . . 512
8.6.3 Linear Minimum Mean Square Error (LMMSE) MIMO
Channel Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513
8.6.4 Selection of Pilot Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514
8.7 Massive MIMO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514
8.7.1 Massive MIMO Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514
8.7.2 Massive MIMO Detection Schemes . . . . . . . . . . . . . . . . . . . . . . 518
8.8 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521
8.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525
9 Advanced Coding and Coded Modulation Techniques . . . . . . . . . . . . . . . . . . 529
9.1 Linear Block Codes Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529
9.2 BCH Codes Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534
9.3 Trellis Description of Linear Block Codes and Viterbi Decoding
Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537
9.4 Convolutional Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541
9.5 RS Codes, Concatenated Codes, and Product Codes . . . . . . . . . . . . . . . . 549
9.6 Coding with Interleaving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 550
9.7 Codes on Graphs Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553
9.8 Turbo Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554
9.9 Turbo-Product Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 561
9.10 LDPC Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563
9.10.1 Binary LDPC Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563
9.10.2 Decoding of Binary LDPC Codes . . . . . . . . . . . . . . . . . . . . . . . 565
9.10.3 FPGA Implementation of Binary LDPC Decoders . . . . . . . . . . . 568
9.10.4 Decoding of Nonbinary LDPC Codes . . . . . . . . . . . . . . . . . . . . 570
9.10.5 Design of LDPC Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571
9.10.5.1 Gallager Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571
Contents xvii

9.10.5.2 Tanner Codes and Generalized LDPC (GLDPC)

Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571
9.10.5.3 MacKay Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 572
9.10.5.4 Large-Girth Quasi-Cyclic (QC) Binary LDPC
Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573
9.10.6 Rate-Adaptive LDPC Coding . . . . . . . . . . . . . . . . . . . . . . . . . . 574
9.10.7 Rate-Adaptive Coding Implementations in FPGA . . . . . . . . . . . . 576
9.10.7.1 Rate-Adaptive LDPC Coding in FPGA . . . . . . . . . . . 576
9.10.7.2 Rate-Adaptive GLDPC Coding in FPGA . . . . . . . . . . 580
9.11 Coded Modulation and Unequal Error Protection . . . . . . . . . . . . . . . . . . 582
9.11.1 Coded Modulation Fundamentals . . . . . . . . . . . . . . . . . . . . . . . 583
9.11.2 Trellis-Coded Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 585
9.11.3 Multilevel Coded Modulation and Unequal Error Protection . . . . 587
9.11.4 Bit-Interleaved Coded Modulation (BICM) . . . . . . . . . . . . . . . . 594
9.11.5 Turbo Trellis-Coded Modulation . . . . . . . . . . . . . . . . . . . . . . . . 594
9.12 Hybrid Multidimensional Coded Modulation Scheme for High-Speed
Optical Transmission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 595
9.12.1 Hybrid Coded Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 596
9.12.2 Multilevel Nonbinary LDPC-Coded Modulation
(ML-NB-LDPC-CM) for High-Speed Optical Transmissions . . . . 599
9.13 Multidimensional Turbo Equalization . . . . . . . . . . . . . . . . . . . . . . . . . . . 602
9.13.1 Nonlinear Channels with Memory . . . . . . . . . . . . . . . . . . . . . . . 602
9.13.2 Nonbinary MAP Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 603
9.13.3 Sliding-Window Multidimensional Turbo Equalization . . . . . . . . 604
9.13.4 Nonlinear Propagation Simulation Study of Turbo
Equalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 607
9.13.5 Experimental Study of Time-Domain 4D-NB-LDPC-CM . . . . . . 608
9.13.6 Quasi-Single-Mode Transmission Over Transoceanic
Distances Using Few-Mode Fibers . . . . . . . . . . . . . . . . . . . . . . 612
9.14 Optimized Signal Constellation Design and Optimized Bit-to-Symbol
Mapping-Based Coded Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 616
9.14.1 Multidimensional Optimized Signal Constellation Design . . . . . . 617
9.14.2 EXIT Chart Analysis of OSCD Mapping Rules . . . . . . . . . . . . . 617
9.14.3 Nonlinear Optimized Signal Constellation Design-Based
Coded Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 621
9.14.4 40 Tb/s Transoceanic Transmission Enabled by 8-ary
OSCD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 622
9.15 Adaptive Coding and Adaptive Coded Modulation . . . . . . . . . . . . . . . . . 625
9.15.1 Adaptive Coded Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 626
9.15.2 Adaptive Nonbinary LDPC-Coded Multidimensional
Modulation for High-Speed Optical Communications . . . . . . . . . 628
9.15.3 Adaptive Hybrid FSO-RF-Coded Modulation . . . . . . . . . . . . . . 629
9.16 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 630
9.17 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 631
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 642
10 Spread Spectrum, CDMA, and Ultra-Wideband Communications . . . . . . . . . 647
10.1 Spread Spectrum Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 648
10.1.1 Spread Spectrum Systems Fundamentals . . . . . . . . . . . . . . . . . . 648
10.1.2 Direct Sequence-Spread Spectrum (DS-SS) Systems . . . . . . . . . . 651
10.1.2.1 DS-SS Transmitters and Receivers . . . . . . . . . . . . . . . 652
10.1.2.2 Acquisition and Tracking . . . . . . . . . . . . . . . . . . . . . 655
10.1.2.3 RAKE Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . 658
10.1.2.4 Spreading Sequences . . . . . . . . . . . . . . . . . . . . . . . . 659
10.1.3 Frequency Hopping-Spread Spectrum (FH-SS) Systems . . . . . . . 663
xviii Contents

10.2 Code-Division Multiple Access (CDMA) Systems . . . . . . . . . . . . . . . . . 668

10.2.1 Signature Waveforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 669
10.2.2 Synchronous and Asynchronous CDMA Models . . . . . . . . . . . . 672
10.2.2.1 Basic CDMA Models . . . . . . . . . . . . . . . . . . . . . . . . 672
10.2.2.2 Discrete-Time CDMA Models . . . . . . . . . . . . . . . . . . 674
10.3 Multiuser Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 679
10.3.1 Conventional Single-User Correlator Receiver . . . . . . . . . . . . . . 679
10.3.2 Optimum Multiuser Detection . . . . . . . . . . . . . . . . . . . . . . . . . . 682
10.3.3 Decorrelating and Linear MMSE Multiuser Detectors . . . . . . . . . 686
10.3.4 Decision-Driven Multiuser Detectors . . . . . . . . . . . . . . . . . . . . . 690
10.4 Optical CDMA Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 693
10.5 Hybrid OFDM-CDMA Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 696
10.6 Ultra-Wideband (UWB) Communications . . . . . . . . . . . . . . . . . . . . . . . . 700
10.7 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 706
10.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 707
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 709
11 Physical-Layer Security for Wireless and Optical Channels . . . . . . . . . . . . . . 713
11.1 Security Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 713
11.2 Information-Theoretic vs. Computational Security . . . . . . . . . . . . . . . . . . 715
11.2.1 Information-Theoretic (Perfect) Security . . . . . . . . . . . . . . . . . . 715
11.2.2 Computational Security . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 716
11.2.3 Information-Theoretic Secrecy Metrics . . . . . . . . . . . . . . . . . . . 716
11.3 Wyner’s Wiretap Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 717
11.4 Broadcast Channel with Confidential Messages and Wireless
Channel Secrecy Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 721
11.4.1 Broadcast Channel with Confidential Messages . . . . . . . . . . . . . 721
11.4.2 Wireless Channel Secrecy Capacity . . . . . . . . . . . . . . . . . . . . . . 723
11.5 Secret-Key Generation (Agreement) Protocols . . . . . . . . . . . . . . . . . . . . 724
11.5.1 Source-Type Secret-Key Generation . . . . . . . . . . . . . . . . . . . . . 726
11.5.2 Channel-Type Secret-Key Generation . . . . . . . . . . . . . . . . . . . . 729
11.6 Coding for Physical-Layer Security Systems . . . . . . . . . . . . . . . . . . . . . . 732
11.6.1 Coding for Weak Secrecy Systems . . . . . . . . . . . . . . . . . . . . . . 732
11.6.1.1 Two-Edge-Type LDPC Coding . . . . . . . . . . . . . . . . . 733
11.6.1.2 Punctured LDPC Coding . . . . . . . . . . . . . . . . . . . . . . 733
11.6.1.3 Polar Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 735
11.6.2 Coding for Strong Secrecy Systems . . . . . . . . . . . . . . . . . . . . . . 736
11.6.2.1 Coset Coding with Dual of LDPC Codes . . . . . . . . . . 737
11.6.2.2 Hash Functions and Extractor-Based Coding . . . . . . . 737
11.6.3 Information Reconciliation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 737
11.7 Privacy Amplification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 739
11.8 Wireless Channels Physical-Layer Security . . . . . . . . . . . . . . . . . . . . . . . 741
11.8.1 PLS for Wireless MIMO Channels . . . . . . . . . . . . . . . . . . . . . . 741
11.8.2 Secret-Key Generation in Wireless Networks . . . . . . . . . . . . . . . 745
11.9 Optical Channels Physical-Layer Security . . . . . . . . . . . . . . . . . . . . . . . . 747
11.9.1 SDM-Fibers-Based Physical-Layer Security . . . . . . . . . . . . . . . . 748
11.9.2 FSO Physical-Layer Security . . . . . . . . . . . . . . . . . . . . . . . . . . 749
11.10 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 754
11.11 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 754
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 756

Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 761
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773
About the Author

Ivan B. Djordjevic (Fellow, IEEE) is Professor of Electrical and

Computer Engineering and Optical Sciences at the University of
Arizona, director of the Optical Communications Systems Labora-
tory (OCSL) and Quantum Communications (QuCom) Lab, and
co-director of the Signal Processing and Coding Lab. He is both
IEEE Fellow and the OSA (Optica) Fellow. He received his PhD
degree from the University of Nis, Yugoslavia, in 1999.
Dr. Djordjevic has authored or co-authored 10 books, more
than 550 journal and conference publications, and he holds
54 US patents. Dr. Djordjevic serves as an area editor/associate
editor/member of editorial board for the following journals: IEEE
TRANSACTIONS ON COMMUNICATIONS, OSA (OPTICA)/IEEE JOURNAL OF
OPTICAL COMMUNICATIONS AND NETWORKING, OPTICAL AND QUANTUM
ELECTRONICS, and FREQUENZ. He was serving as editor/senior edi-
tor/area editor of IEEE COMMUNICATIONS LETTERS from 2012 to
2021. He was serving as editorial board member/associate editor
for IOP JOURNAL OF OPTICS and ELSEVIER PHYSICAL COMMUNICATION
JOURNAL both from 2016 to 2021.
Prior to joining the University of Arizona, Dr. Djordjevic held
appointments at the University of Bristol and the University of the
West of England in UK, Tyco Telecommunications in the USA,
National Technical University of Athens in Greece, and State
Telecommunication Company in Yugoslavia.

xix
 Introduction
1

Abstract
In this introductory chapter, we first describe the historical perspective related to both wireless and optical communication
systems and describe various generations, from early beginning until today. After that we describe the fundamentals of
optical systems and networks, including WDM network architecture, typical optical networking architecture, bit rates for
synchronous/asynchronous optical communication systems, and wavelength regions for fiber-optics communications. In
the same section, we briefly describe various multiplexing schemes such as time-division multiplexing (TDM),
wavelength-division multiplexing (WDM), frequency-division multiplexing (FDM), subcarrier multiplexing (SCM),
code-division multiplexing (CDM), polarization-division multiplexing (PDM), and spatial-division multiplexing (SDM).
We also identify the main noise sources and optical channel impairments, in particular of importance for fiber-optics
communications. After that the key optical devises and modules are identified and briefly described. In section related to
wireless communications fundamentals, we describe a typical wireless communication point-to-point link and identify key
devices/subsystems with brief description of operational principles. The multiaccess communications are then described
with key problems being identified. Further, various effects affecting the received power are described including path loss,
shadowing, and multipath fading. We also introduce various approaches to deal with multipath fading including orthogonal
frequency-division multiplexing (OFDM), diversity, multiple-input multiple-output (MIMO) signal processing, equaliza-
tion, and adaptive modulation and coding. The MIMO operation principle is briefly described next. Finally, the cellular
concept is introduced. In section on organization of the book, the detailed description of chapters is provided. Final section
provides some relevant concluding remarks.

1.1 Historical Perspective of Optical and Wireless Communication Systems

The earliest optical and wireless communications systems consisted of fire or smoke signals, signaling lamps, and semaphore
flags to convey a single piece of information [1–4]. For example, a relatively sophisticated ancient communication system,
along the Great Wall of China, was composed of countless beacon towers [5]. In this ancient communication system, the
number of lanterns or the color of smoke was used as a means to inform the size of an invading enemy, which represents a
crude form of simultaneously digital communication and multilevel signaling. By using the beacon towers, with the guards in
each tower, positioned at regular distances along the Great Wall, a message could be transmitted from one end of the Great
Wall to the other, more than 7300 km, in slightly more than 1 h [5]. Therefore, this ancient communication system has many
similarities with today’s relay or regeneration systems, in which the beacon towers can be considered as relays. Relay or
regeneration systems were further studied by Claude Chappe in 1792 to transmit coded messages optically over distance of
100 km [1]. In next two subsections, we describe the evolution of both wireless communication systems [2, 33–37] and optical
communication systems [1, 3–5, 38–42].

# The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 1

I. B. Djordjevic, Advanced Optical and Wireless Communications Systems, https://doi.org/10.1007/978-3-030-98491-5_1
2 1 Introduction

1.1.1 Wireless Communication Historical Perspective

Electromagnetic wave propagation theory is developed in the 1860s and summarized by Maxwell and demonstrated later by
Hertz. The first radio transmission was demonstrated in 1893 by Tesla, followed by Marconi in 1896. At that time, however,
the systems were of low frequency, high power, huge size, expensive, and mostly mechanical. Interestingly enough, the
telegraphic systems were digital, based on Morse code (1837), and the early steps of electronic age were driven by the need of
wireless communications, rather than wired systems. Unfortunately, with the invention of amplitude-modulated (AM) radio
by Fessenden and de Forest as well as invention of frequency-modulated (FM) radio by Armstrong and de Forest, the era of
analog communication started instead. The development of communication systems was first boosted by the invention of
vacuum tubes in 1906 by de Forest, which can be considered as the start of electronic age. Communication systems with
vacuum tubes were huge, power hungry, and not stable. The real engine of electronic age started with the invention of
transistor by William Shockley during 1948–1951, when the communication hardware get revolutionized for the second time.
Another relevant revolution came in the same time, namely, the information theory was developed by Claude Shannon in
1948, and the golden age of information theory was from 1948 to 1960. Many sophisticated military radio systems were
developed during and after WW2. The third revolution for communication hardware is the invention of integrated circuit
(IC) in 1958 by Jack Kilby (independently by Robert Noyce in 1959), which was developed in the 1960s and enabled the
exponential growth since then.
Another driving force of wireless systems was the invention of mobile phone, with the first trial being in 1915, between
New York and San Francisco. The commercial networks started in 1946, while the takeoff was in the 1980s with the
introduction of the cellular concept. Many wireless communication systems were proposed in the 1990s, with great failures
around 2000. The first generation (1G) of cellular systems was introduced in the 1980s and was based on analog, employing
the frequency-division multiplexing access (FDMA). The most popular 1G cellular systems were [2] advanced mobile phone
service (AMPS) in North America and Australia; total access communication system (TACS) in the UK; Nordic mobile
telephone (NMT) in the Nordic countries, Switzerland, the Netherlands, Eastern Europe, and Russia; and C-450 in West
Germany, Portugal, and South Africa, to mention a few.
The second generation (2G) of cellular systems was introduced in 1991, based on digital communication principles. In
addition to voice, they supported the short message service (SMS) as well as low data rate transmissions (including
multimedia messaging service or MMS). The 2G technologies can be classified into two broad categories, depending on
the multiplexing method: time-division multiple access (TDMA)-based and code-division multiple access (CDMA)-based
standards. The most popular standards in 2G are (i) global system for mobile communications (GSM), which is TDMA-based,
originally used in Europe but later in most of the world except North America; (ii) IS-95, which was CDMA-based, invented
by Qualcomm and later adopted by the Telecommunications Industry Association (TIA); (iii) Japanese digital cellular (JDS or
PDC), which was TDMA-based; and (iv) IS-136 (also known as digital AMPS or D-AMPS), TDMA-based. Regarding the
data rates supported, with (1) general packet radio service (GPRS), there is a theoretical maximum data rate of 50 kbit/s, and
(2) in enhanced data rates for GSM evolution (EDGE), there is a theoretical maximum data rate of 1 Mbit/s.
The first 3G networks were introduced in 1998, and 3G technology is based on a set of standards that comply with the
International Mobile Telecommunications-2000 (IMT-2000) specifications by the International Telecommunication Union
(ITU). To meet the IMT-2000 standards, a 3G system is required to provide peak data rates of at least 200 kb/s. However,
many 3G provide higher data rates than the minimum technical requirements for a 3G service. In particular, 3.5G and 3.75G
releases provide mobile broadband access of several Mb/s to smartphones and laptop mobile modems. The most relevant 3G
standards are (1) Universal Mobile Telecommunications Service (UMTS), standardized by 3GPP, used primarily in Europe,
Japan, and China as well as other regions predominated by GSM 2G system infrastructure, and (2) the CDMA2000 system,
standardized by 3GPP2, used in North America and South Korea, sharing infrastructure with the IS-95 2G standard. The
UMTS can support the data rates up to 2 Mb/s for stationary/walking users and 348 kb/s for moving vehicles. The High-Speed
Packet Access (HSPA) technology specifications, representing an intermediate phase toward 4G, initially supported increased
peak data rates of up to 14 Mb/s in the downlink and 5.76 Mb/s in the uplink. This technology contains two mobile protocols
High-Speed Downlink Packet Access (HSDPA) and High-Speed Uplink Packet Access (HSUPA). The predecessors of 4G
technology are (i) 3GPP Long-Term Evolution (LTE) offering 100 Mbit/s in the downlink and 50 Mbit/s in the uplink over
20 MHz, (ii) mobile WiMAX (IEEE 802.16e-2005) offering up to 128 Mb/s in the downlink and 56 Mb/s in the uplink over
20 MHz of channel bandwidth, (iii) UMB (formerly EV-DO Rev. C), and (iv) fast low-latency access with seamless handoff
OFDM (also known as FLASH-OFDM). The bandwidth as well as location information available to 3G devices gives rise to
some new applications including global positioning system (GPS), location-based services, mobile TV, telemedicine, video
conferencing, and video on demand.
1.1 Historical Perspective of Optical and Wireless Communication Systems 3

The fourth generation (4G) of mobile communications technologies provides a comprehensive and secure all-IP-based
solution, with current/potential applications including mobile web access, IP telephony, gaming services, high-definition
mobile TV, video conferencing, 3D television, and cloud computing. ITU-R specified in Mar. 2008 the set of requirements for
4G standards, named International Mobile Telecommunications-Advanced (IMT-Advanced) specification, setting peak speed
requirements for 4G service at 100 Mbit/s for high-mobility communication links (such as from trains and cars) and 1 Gbit/
s for low-mobility communication links (such as pedestrians and stationary users). IMT-2000 compliant 4G standards are LTE
Advanced and WirelessMAN-Advanced (IEEE 802.16 m). The LTE Advanced is essentially an enhancement to LTE, with
the coordinated multipoint transmission that allows for higher system capacity and helps handling the enhanced data speeds.
The peak download data rate is 1 Gb/s, while the peak upload data rate is 500 Mb/s. On the other hand, the IEEE 802.16 m
represents the evolution of 802.16e offering 1 Gb/s data rate for stationary reception and 100 Mb/s for mobile reception.
Principal 4G technologies can be summarized as follows: (i) multi-antenna and multi-user (MIMO) concept; (ii) frequency-
domain equalization and frequency-domain statistical multiplexing; (iii) turbo codes which are used as channel codes to
reduce the required SNR at Rx side; (iv) channel-dependent scheduling which is used to utilize the time-varying channel;
(v) link adaptation which is used based on adaptive modulation and error-correcting codes; and (vi) relaying which is used
including fixed relay networks (FRNs) and the cooperative relaying concept.
The fifth generation (5G) wireless systems go beyond the 4G IMT-Advanced standards. The 5G standard should fulfill the
following Next Generation Mobile Networks Alliance requirements [6, 7]: (1) supporting tens of thousands of users with data
rates of tens of Mb/s; (2) for metropolitan areas, data rates of 100 Mb/s should be supported; (3) the workers located on the
same floor should get 1 Gb/s simultaneously; (4) for wireless sensor networks, we should be able to establish several hundreds
of thousands of simultaneous connections; (5) significantly higher spectral efficiency compared to 4G; (6) the coverage must
be improved; (7) latency must be reduced significantly compared to LTE; and (8) the signaling efficiency must be improved.
Toward these goals, the FCC recently approved the new spectra for 5G, including the 28 GHz, 24 GHz, 37 GHz, 39 GHz, and
47 GHz bands [8, 9]. The FCC is working on freeing 2.75 GHz for 5G applications in the 26 and 42 GHz bands and has plans
for additional millimeter wave spectrum in the 70/80/90 GHz bands [9]. FCC also considers allocating 600 MHz bandwidth in
2.5 GHz, 3.5 GHz, and 3.7–4.2 GHz bands [9]. Key enabling technologies for 5G include [18] (i) the employment of a
wireless software-defined network (WSDN), (ii) the use of network function virtualization (NFV), (iii) introduction of
massive MIMO, (iv) much higher network density, (v) the utilization of the millimeter wave spectrum, (vi) the breakthroughs
in big data and mobile cloud computing, (vii) high mobility device-to-device connections, (viii) better energy-efficiency
communications, (ix) the introduction of new radio access technologies, and (x) scalable Internet of Things techniques. The
main applications, as specified by the ITU-R, for the enhanced capabilities of 5G are Enhanced Mobile Broadband (eMBB),
Ultra Reliable Low Latency Communications (URLLC), and Massive Machine Type Communications (mMTC). The ITU’s
IMT-2020 standard requires a peak download speed of 20 Gb/s with spectral deficiency up to 30 bits/s/Hz and 10 Gb/s upload
speed with spectral efficiency up to 15 bits/s/Hz [10, 11]. The target values for the user experienced data rates in downlink and
uplink are 100 Mb/s and 50 Mb/s, respectively. The user plane latency minimum requirements are 4 ms for eMBB and 1 ms
for URLLC [10]. Another relevant standard air interface for the 3GPP 5G networks is the 5G NR (New Radio) [12]. The 5G
NR employs two frequency bands: frequency range 1 (FR1) 410 MHz–7.125 GHz and frequency range 2 (FR2)
24.25–52.6 GHz. Early FR1 deployments employ 5G NR software on 4G hardware, which are slightly better than newer
4G systems, and essentially represent 4.5G technology. The 5G-Advanced standard is currently under development by
3GPP [13].
The 6th generation (6G) wireless systems and standards are currently under development. It is expected that the 6G
networks will be more diverse than the 5G networks and these networks will be able to support various applications beyond
current mobile applications including virtual and augmented reality, remote surgery, holographic projection, pervasive
intelligence, the IoT, and instant communications, to mention few [14–17]. The peak data rates in 6G could be up to
10 Tb/s for downlink, with much larger spectral efficiencies compared to 5G. The user plane latencies should go down to
0.1 ms. To achieve these goals, in addition to mm-wave technologies, the THz and free-space optical (FSO) communications
must be exploited.
Regarding the wireless LANs (WLANs), the most relevant is the IEEE 802.11 standard family, representing the set of MAC
and PHY standards implementing WLAN computer communications in 0.9, 2.4, 3.6, 5, and 60 GHz frequency bands [19–24].
The first wireless networking standard from this family was 802.11–1997; however, the 802.11b was the widely accepted one,
which was followed by 802.11a, 802.11 g, 802.11n, and 802.11 ac. Other standards in this family (such as c–f, h, j) represent
the service amendments to the existing ones that that are used to extend the scope, or eventually the amendments represent the
corrections to the previous specification. The IEEE 802.11 g PHY standard is very similar to 801.11a, except that it is related
to the 2.4 GHz unlicensed ISM band (2.4–2.497 GHz).
4 1 Introduction

To increase the aggregated data rates in IEEE 802.11n and 802.11 ac PHY standards, the MIMO-OFDM concept,
explained in Chap. 8, is used. Namely, through MIMO concept, several independent data streams can be simultaneously
transmitted increasing the aggregate data rates. The IEEE 802.11n PHY standard is applicable to both 2.4 and 5 GHz bands
while 802.11 ac to 5 GHz band only. In IEEE 802.11n standard, the available channel bandwidths are 20 MHz and 40 MHz,
while in 802.11 ac (Wi-Fi 5), the available channel bandwidths are 20, 40, 80, and 160 MHz. The highest data rate in
802.11 ac standard, for 160 MHz bandwidth, is 780 Mb/s. Recent 802.11ax (Wi-Fi 6E) standard, also known as a high-
efficiency Wi-Fi, adopted in 2020, is designed to operate in license-exempt bands (1–7.125 GHz), including commonly used
2.4 and 5 GHz as well as the 6 GHz band (5.925–7.125 GHz in the USA) [25]. The main goal of Wi-Fi 6E standard is to
enhance throughput/area for high-density scenarios such as shopping malls, corporate offices, and dense residential areas. The
modulation formats range from BPSK to 1024-QAM, with code rates from 1/2 to 5/6. The data rates range from 135 Mb/s (for
20 MHz channels) to 1.201 Gb/s (for 120 MHz channels).
The WiMAX represents one of the most popular broadband wireless access (BWA) technologies, aiming to provide high-
speed bandwidth access for wireless MANs [21, 26–28]. The WiMAX is based on IEEE 802.16 set of standards, in particular
IEEE 802.16–2004 (fixed-WiMAX) and IEEE 802.16e-2005 (mobile-WiMAX) standards, which provide multiple PHY and
MAC options. The WiMAX represents the last mile wireless broadband access alternative to the cable and DSL services, and
it is a direct competitor to the LTE Advanced standard. The IEEE 802.16e-2005 provides several improvements compared to
the IEEE 802.16–2004 including [26–28] (1) providing support for mobility by enabling the soft and hard handover between
base stations; (2) introducing the concept of scalable OFDM (SOFDM) by scaling the FFT to the channel bandwidth to ensure
that the carrier spacing is constant across different channel bandwidths, ranging from 1.25 MHz through 5 MHz and 10 MHz
all the way to 20 MHz and thus improving the spectral efficiency; (3) introducing the advanced antenna diversity schemes and
hybrid automatic repeat request (HARQ); (4) introducing the adaptive antenna systems (AAS) and MIMO signal processing;
(5) providing the denser sub-channelization and therefore improving the indoor penetration; (6) introducing downlink
sub-channelization, in order to trade the coverage for capacity or vice versa; (7) introducing the LDPC coding to improve
the error correction strength; and (8) introducing an extra QoS class for VoIP applications. In September 2018 the IEEE
802.16–2017 standard was released, relevant for combined fixed and mobile point-to-multipoint BWA systems providing
multiple services [29]. The IEEE 802.16 Working Group ceased their activities on 9 March 2018 [30].

1.1.2 Optical Communication Historical Perspective

Thanks to the success of telegraphy, telephony, and radio communications in the first half of the twentieth century, the optical
communication systems were actually forgotten. However, in the late twentieth century, different communication systems
came to saturation in terms of reach and capacity. For instance, a typical coaxial cable-based transport system operating at
155 Mb/s requires the regeneration at approximately every 1 km, which is costly to operate and maintain. The natural step was
to study the optical communication systems, which can dramatically increase the total capacity. The research in optical
communication was boosted upon demonstration of a laser principle [31]. The first step was to fabricate an appropriate optical
transmission medium. Kao and Hockman [32] proposed to use the optical fiber as the medium, although at the time it had
unacceptable fiber loss. Their argument was that attenuation mostly was coming from impurities, rather than any fundamental
physical effect such as Rayleigh scattering, which could be reduced by improving the fabrication process. Their prediction
was that an optical fiber with attenuation of 20 dB/km should be sufficient for telecom applications, which surprisingly was
developed within 5 years since initial proposal, by researchers from Cornell. This invention opens up opportunities for
development of fiber-optics communication systems. Since then several generations of optical communication systems were
developed. The optical communication evolution path can be summarized as follows.
The first generation appeared in the 1970s, and the operating wavelengths were 0.8 μm–0.9 μm with data rates ranging
from 45 Mb/s to 500 Mb/s. Repeater spacing was 10 km, which was much greater than that for comparable coax systems.
Lower installation and maintenance costs resulted from fewer repeaters.
The second generation, which was focused on a transmission near 1.3 μm to take advantage of the low attenuation (< 1 dB/
km) and low dispersion, was deployed during the early 1980s. Sources and detectors were developed that use InGaAsP
semiconductor. The bit rate of these systems was limited to <100 Mb/s due to dispersion in multimode fibers (MMFs). Single-
mode fiber (SMF) was then incorporated. By 1987 the second-generation systems were operating at 2.5 Gb/s at 1.3 μm with
repeater spacing of 50 km.
The third-generation systems were based on the use of 1.55 μm sources and detectors and dominated in the first half of the
1990s. At this wavelength the attenuation of fused silica fiber is minimal. The deployment of these systems was delayed
1.1 Historical Perspective of Optical and Wireless Communication Systems 5

however due to the relatively large dispersion at this wavelength. Two approaches were proposed to solve the dispersion
problem. The first approach was to develop single-mode lasers, and the second was to develop dispersion-shifted fiber (DSF)
at 1.55 μm. In 1990, 1.55 μm systems operating at 2.5 Gb/s were commercially available and were capable of operating at
10 Gb/s for distances of 100 km [1, 3–5, 38–40]. The best performance was achieved with DSFs in conjunction with single-
mode lasers. A drawback of these systems was the need for electronic regeneration with repeaters typically spaced every
60–70 km. Coherent optical detection methods were investigated in late 1980s and early 1990s to increase receiver sensitivity.
However, this approach was superseded by the development of the optical amplifier.
The 3.5th-generation systems are based on the use of optical amplifiers to increase repeater spacing and wavelength-
division multiplexing (WDM) to increase the aggregate bit rate. Erbium-doped fiber amplifiers (EDFAs) were developed to
amplify signals without electronic regeneration during the 1980s [1, 3–5, 38–40]. In 1991 signals could be transmitted
14,300 km at 5 Gb/s without electronic regeneration [1, 3–5, 38–40]. The first transpacific commercial system went into
operation sending signals over 11,300 km at 5 Gb/s, and other systems are being deployed. System capacity is increased
through use of WDM. Multiple wavelengths can be amplified with the same optical amplifier. In 1996 20
5 Gb/s signals
were transmitted over 9100 km providing a total bit rate of 100 Gb/s and a bandwidth-length (B-L) product of 910 (Tb/s)-km.
In these broadband systems, dispersion becomes an important issue to be addressed.
In the 3.75th-generation systems, the effort is primarily concerned with the fiber dispersion problem. Optical amplifiers
solve the loss problem but increase the dispersion problem since dispersion effects accumulate over multiple amplification
stages. An ultimate solution is based on the novel concept of optical solitons [1, 43, 44]. These are pulses that preserve their
shape during propagation in a loss less fiber by counteracting the effect of dispersion through fiber nonlinearity. Experiments
using stimulated Raman scattering as the nonlinearity to compensate for both loss and dispersion were effective in
transmitting signals over 4000 km [1, 43, 44]. EDFAs were first used to amplify solitons in 1989 [1, 43, 44]. By 1994 a
demonstration of soliton transmission over 9400 km was performed at a bit rate of 70 Gb/s by multiplexing seven 10 Gb/
s channels [1, 43, 44]. In parallel, dispersion compensating fibers (DCFs) were invented to deal with chromatic dispersion, and
various dispersion maps were proposed [1, 43, 44]. The WDM channel count increased to maximum 128, with data rates per
single wavelength ranging from 2.5 Gb/s to 10 Gb/s. The operating wavelength region ranges from 1530 nm to 1560 nm.
Regarding the networking aspects, WDM multiplexers/demultiplexers as well as optical add-drop multiplexers (OADMs)
have been used. Regarding the optical networking protocols, the following protocols have been used: SONET/SDH/OTN, IP,
ATM, ESCON/FICON, and Ethernet. The readers interested learning more about various optical networking protocols are
referred to [45–59].
In the fourth-generation systems, the efforts have been directed toward realizing greater capacity of fiber systems by
multiplexing a large number of wavelengths. These systems are referred to as dense wavelength-division multiplexing
(DWDM) systems, with corresponding channel spacing being either 100 GHz or 50 GHz. Controlling wavelength stability
and the development of wavelength demultiplexing devices are critical to this effort. The maximum DWDM channel count is
160, with the data rates per single wavelength being 10 Gb/s and/or 40 Gb/s, and the corresponding lasers are tunable. In
addition to EDFAs, the Raman amplifiers have been used as well. Regarding the networking aspects, DWDM multiplexers/
demultiplexers, reconfigurable OADMs (ROADMs), and optical cross-connects (OXCs) have been used. Additionally, the
layered control plane is introduced. Regarding the optical networking protocols, the following protocols have been used:
SONET/SDH/OTN, IP, MPLS, and Ethernet.
The fifth-generation systems are related to optical transmission systems with data rates per wavelength ranging from
100 Gb/s to 1 Tb/s, by employing various multilevel modulation and channel coding schemes, polarization multiplexing,
digital signal processing (DSP), and coherent optical detection. The orthogonal frequency-division multiplexing (OFDM)
appears to be an excellent candidate to deal with chromatic dispersion and polarization mode dispersion (PMD). Additionally,
the MIMO signal processing is applied as well. Regarding the optical networking protocols, the following protocols have
been used: OTN, IP, MPLS, and Ethernet. Regarding the standardization activities, ITU-T, IEEE 802.3ba, and OIF have
completed the work on 100 Gb/s Ethernet (100 GbE) [60, 61], while the activities for adoption of 400 Gb/s Ethernet, 1 Tb/
s Ethernet, and beyond are currently underway.
The systems with channel rates beyond 10 Tb/s over SMFs, such as those described in [62], can be placed into the sixth
generation, and these systems are currently in research phase. Finally, the systems based on spatial-division multiplexing
(SDM) [62, 63] (and references therein), employing various SDM fibers including few-mode fiber (FMF), few-core fiber
(FCF), and few-mode-few-core fiber (FMFC), to mention a few, can be placed into the seventh generation.
6 1 Introduction

1.2 Optical Communications Systems and Networks Fundamentals

A generic WDM optical network, which can be used to identify the key optical components, concepts, and system parameters,
is provided in Fig. 1.1. The end-to-end optical transmission involves both electrical and optical signal paths. To perform
conversion from electrical to optical domain, the optical transmitters are used, while to perform conversion in the opposite
direction (optical-to-electrical conversion), the optical receivers are used. The singe-mode fiber (SMF) serves as a foundation
of an optical transmission system because the optical fiber is used as medium to transport the optical signals from the source to
destination. The optical fibers attenuate the signal during transmission, and someone has to use optical amplifiers, such as
EDFAs, Raman amplifiers, or parametric amplifiers, to restore the signal level. Unfortunately, the amplification process is
accompanied with the noise addition. For better exploitation of enormous bandwidth of SMF, the WDM concept is
introduced, which corresponds to the scheme with multiple optical carriers at different wavelengths that are modulated by
using independent electrical bit streams, as shown in Fig. 1.1, and then transmitted over the same SMF. During transmission
of WDM signals, occasionally several wavelengths have to be added/dropped, which is performed by the reconfigurable
optical add-drop multiplexer (ROADM), as shown in Fig. 1.1. The optical networks require the switching of information
among different fibers, which is performed by the optical cross-connect (OXS). To combine several distinct wavelength
channels into composite channel, the wavelength multiplexers are used. On the other hand, to split the composite WDM
channel into distinct wavelength channels, the wavelength demultiplexers is used. To impose the information signal and
perform electro-optical conversion, the optical modulators are used. The optical modulators are commonly used in combina-
tion with semiconductor lasers.
The optical transmission systems can be classified according to different criterions. When bit rate is used as classification
criteria, the optical transmission systems can be classified as low-speed (tens of Mb/s), medium-speed (hundreds Mb/s), high-
speed (Gb/s), and ultra-high-speed (tens of Gb/s). From application perspective point of view, the systems can be either power
budget (loss) limited or bandwidth (transmission speed) limited. If transmission length is used for classification, we can
identify very short reach (hundreds of meters), short reach (from several kilometers to several tens of km), long reach
(hundreds of kilometers), and ultra-long reach (thousands of kilometers) optical transmission systems.
To provide a global picture, we describe a typical optical network shown in Fig. 1.2. We can identify three ellipses
representing the core network, the edge network, and the access network [3, 4]. The long-haul core network interconnects big
cities, major communications hubs, and even different continents by means of submarine transmission systems. The core
networks are often called the wide area networks (WANs) or interchange carrier networks. The edge optical networks are
deployed within smaller geographical areas and are commonly recognized as metropolitan area networks (MANs) or local
exchange carrier networks. The access networks represent peripheral part of optical network and provide the last-mile access
or the bandwidth distribution to the individual end users.
The ultimate goal of an optical signal transmission system is usually defined as achieving desired bit error rate (BER)
performance between two end users or between two intermediate nodes in network reliably and at affordable cost. In order to
achieve so, an optical transmission system needs to be properly designed, which includes the management of key optical
communication systems engineering parameters. These parameters can be related to power, time, and wavelength or be
interrelated. The parameters related only to power are power level, fiber loss, insertion loss, and extinction ratio (the ratio of
powers corresponding to bit “1” and bit “0”). The parameters related only to time are jitter, first-order PMD, and bit/data rate.
The parameters related to wavelength include optical bandwidth and wavelength stability. The parameters, signal

Optical cross-connect (OXS) /

ROADM

Client 1 Transmitter Txs/Rxs Receiver Data 1

Optical
data (Tx) 1 (Rx) 1
amplifier
…
demultiplexer

demultiplexer

(OA)
Wavelength

Wavelength

Client 2 SMF Data 2

Tx 2 Rx 2
data ROADM OA OA
…

…
…

Client N Data N
Tx N Rx N
data

Fig. 1.1 A generic WDM optical network identifying key optical components, concepts, and parameters
1.2 Optical Communications Systems and Networks Fundamentals 7

Fig. 1.2 A typical optical

networking architecture Access network

EN
Edge node Core node
End
(EN) (CN)
node
CN EN
EN
EN
ENEN CN
EN CN
EN
Core network
EN

Edge network

impairments, and additive/multiplicative noise sources related to both power and wavelength are optical amplifier gain,
optical noise (such as amplified spontaneous emission (ASE) noise), different crosstalk effects, four-wave mixing (FWM),
and stimulated Raman scattering (SRS). The parameters related to time and wavelength are laser chirp, second-order PMD,
and chromatic dispersion. The parameters related to power and time are BER, modulation format, polarization-dependent loss
(PDL), and quantum noise. Finally, the channel impairments related to time, power, and wavelength simultaneously are self-
phase modulation (SPM), cross-phase modulation (CPM), and stimulated Brillouin scattering (SBS). Any detection scenario
must include electronic noise, such as thermal noise, which is associated with receiver design. These different parameters,
noise sources, and channel impairments are subject of investigation in Chap. 2.
Different high-speed optical transmission enabling technologies can be related to the usage of novel/better devices, such as
Raman and parametric amplifiers, PMD and chromatic dispersion compensators, and modulators, or be related to the novel
methods, such as advanced modulation formats (various multilevel modulation schemes with both direct and coherent optical
detections and OFDM), forward error correction (FEC), coded modulation, constrained (modulation/line) coding, advanced
detection schemes including maximum likelihood sequence detection/estimation (MLSD/E) and maximum a posteriori
probability (MAP) detection (BCJR algorithm-based equalizers), and various multiplexing schemes including polarization
multiplexing, optical time-division multiplexing (OTDM), subcarrier multiplexing (SCM), code-division multiplexing
(CDM), OFDM, and spatial-division multiplexing. These various enabling technologies will be described in incoming
chapters of the book.
An important concept to be introduced here is related to the so called lightwave path, which can be defined as the trace that
optical signal passes between the source and destination without experiencing any opto-electrical-opto (O-E-O) conversion
[3, 4]. Generally speaking, the lightwave paths may differ in lengths and in the information capacity that is carried along and
can traverse though different portions of an optical network. The lightwave path can be considered as bandwidth wrapper for
lower speed transmission channels, which form virtual circuit services [3, 4]. The time-division multiplexing (TDM)
technique is applied to aggregate the bandwidth of virtual circuits before it is wrapped in the lightwave path. TDM of virtual
circuits can be either fixed (each circuits receives a guaranteed amount of the bandwidth, a bandwidth pipe) or statistical
(in packet switching the data content is divided into data packets, which can be handled independently). The fixed
multiplexing of virtual circuits is defined by SONET/SDH standards. Bit rates of different bandwidth channels, for both
synchronous and asynchronous transmission, are given in Table 1.1 based on [3, 4] (and references therein).
Optical fiber is the key ingredient of an optical transmission system because it has much wider available bandwidth, lower
signal attenuation, and smaller signal distortions compared to any other wired or free-space physical media. The total
bandwidth is approximately 400 nm, or around 50 THz, when related to the wavelength region with fiber attenuation being
below 0.5 dB/km. The usable optical bandwidth is commonly split into several wavelength bands, as summarized in Table 1.2
based on refs. [3, 4, 40]. The bands around the minimum attenuation point, usually referred to as C and L bands, are the most
suitable for high channel count DWDM transmission. The wavelength region around 1300 nm is less favorable for optical
signal transmission because signal attenuation is higher than attenuation in S, C, and L bands. On the other hand, it is quite
suitable for CATV signals, and the course-WDM (CWDM) technique is usually employed in this region.
8 1 Introduction

Table 1.1 Bit rates for different synchronous/asynchronous optical communication systems (both introduced and envisioned)
TDM/synchronous
channels Bit rate Data/asynchronous channels Bit rate
DS-1 1.544 Mb/s 10-BaseT Ethernet 10 Mb/s
E-1 2.048 Mb/s 100-BaseT Ethernet 100 Mb/s
OC-1 51.84 Mb/s FDDI 100 Mb/s
OC-3 , STM-1 155.52 Mb/s ESCON 200 Mb/s
OC-12 , STM-4 602.08 Mb/s Fiber Channel-I 200 Mb/s
Fiber Channel-II 400 Mb/s
Fiber Channel-III 800 Mb/s
Fiber Channel IV 4 Gb/s
OC-48 , STM-16 2.488 Gb/s 1 Gb Ethernet 1 Gb/s
OC-192 , STM-64 9.953 Gb/s 10 Gb Ethernet 10 Gb/s
OC-768 , STM-256 39.813 Gb/s 40 Gb Ethernet 40 Gb/s
100 Gb Ethernet 107–125 Gb/s
400 Gb Ethernet 428–500 Gb/s
1 Tb Ethernet 1.07–1.25 Tb/s
4 Tb/10 Tb Ethernet. 4 Tb/s
10 Tb/s

Table 1.2 The wavelength bands for fiber-optics communications

Wavelength band Descriptor Wavelength range (nm)
O-band Original 1260–1360
E-band Extended 1360–1460
S-band Short 1450–1530
C-band Conventional 1530–1565
L-band Long 1565–1625
U-band Ultra-long 1625–1675

The key optical components, which will be described in next chapter, can be classified as follows: (i) semiconductor light
sources including light-emitting diodes (LEDs) and semiconductor lasers (Fabry-Pérot (FP), distributed feedback (DFB),
distributed Bragg reflector (DBR), vertical cavity surface emitting (VCSEL), tunable lasers (external cavity laser, mutilaser
chip, three-section tunable)); (ii) optical modulators including both direct optical modulators and external modulators (Mach-
Zehnder modulator (MZM), electro-absorption modulator, and I/Q modulator); (iii) optical fibers (MMFs and SMFs);
(iv) optical amplifiers such as semiconductor optical amplifier (SOA), EDFA, Raman amplifiers, and parametric amplifiers;
(v) photodiodes including PIN, avalanche photodiodes (APDs), and metal-semiconductor-metal (MSM) photodetectors; and
(vi) various optical components (optical isolators, optical circulators, optical filters, optical couplers, optical switches, and
optical multiplexers/demultiplexers).
A monochromatic electromagnetic wave, which is commonly used as a signal carrier, can be represented through its
electric field as E(t) ¼ pAcos(ωt + ϕ) (A-amplitude, ω-frequency, ϕ-phase, p-polarization orientation), for which each
parameter can be used to impose the message signal. If the message signal is analog, corresponding modulation formats
are amplitude modulation (AM), frequency modulation (FM), phase modulation (PM), and polarization modulation (PolM).
On the other hand, when the modulating signal is digital, then the carrier signal duration is limited to symbol duration, and
corresponding modulation formats are amplitude shift keying (ASK), frequency shift keying (FSK), phase shift keying (PSK),
and polarization shift keying (PolSK).
In order to better utilize the enormous bandwidth of the optical fiber, we have to transmit simultaneously many channels
over the same bandwidth through multiplexing. The commonly used methods of multiplexing in optical communications are
given below as follows:

• Wavelength-division multiplexing (WDM) is already introduced above.

• Time-division multiplexing (TDM), in which many lower-speed signals are time-interleaved to generate a high-speed
signal. The multiplexing can be performed either in electrical domain, when is known as electrical TDM (ETDM), or in
optical domain, when is known as optical TDM (OTDM).
1.3 Wireless Communications Systems Fundamentals 9

• Frequency-division multiplexing (FDM), in which continuous-wave (CW) modulation is used to translate the spectrum of
the message signal into a specific frequency slot of the passband of optical channel. The optical version of FDM is
commonly referred to as WDM.
• Orthogonal frequency-division multiplexing (OFDM) is a particular version of FDM in which the orthogonality among
subcarriers is obtained by providing that each sub-carrier has exactly an integer number of cycles in the symbol interval.
The number of cycles between adjacent subcarriers differs by exactly one.
• Subcarrier-multiplexing (SCM) is again a particular version of FDM in which different independent data streams are first
microwave multiplexed and then transmitted using the same wavelength carrier.
• Code-division multiplexing (CDM) in which each message signal is identified by a unique signature sequence (“code”),
with signature sequences being orthogonal to each other.
• Spatial-division multiplexing (SDM) in which independent signals excite different spatial modes in FMFs/FCFs.

During the transmission over an optical fiber, the transmitted signal is impaired by various noise sources and channel
impairments. The noise sources can be additive in nature (dark current noise, thermal noise, ASE noise, and crosstalk noise) or
be multiplicative in nature (mode partition noise (MPN), laser intensity noise (RIN), modal noise, quantum shot noise, and
avalanche shot noise). Different channel impairments can be related to fiber attenuation, insertion loss, dispersion effects, or
fiber nonlinearities. Fiber attenuation originates from material absorption, which can be intrinsic (ultraviolet, infrared) or
extrinsic (water vapor, Fe, Cu, Co, Ni, Mn, Cr, various dopants: GeO2, P2O5, B2O3), Rayleigh scattering and waveguide
imperfections (Mie scattering, bending losses, etc.). The dispersion effects can originate from intermodal (multimode)
dispersion (in MMFs), chromatic dispersion (material and waveguide dispersion effects present in SMF), PMD, and
polarization-dependent loss (PDL). The fiber nonlinearities can originate from non-elastic scattering effects (SBS, SRS) or
Kerr nonlinearities (SPM, XPM, FWM). Various noise sources and optical channel impairments are described in Chap. 2.

1.3 Wireless Communications Systems Fundamentals

A typical, generic, point-to-point single-input single-output (SISO) wireless communication system, used in various wireless
applications, is provided in Fig. 1.3. The binary data sequence is first channel encoded [see Fig. 1.3(a)], and the encoder output
is written into interleaver. Typical channel codes include convolutional codes, concatenated convolutional and RS codes,
turbo codes, or LDPC codes [3, 35, 64–66]. For turbo coding s-random interleaver can be used, while for LDPC coding,

Tx antenna
Data modulation section
Binary Encoding & Bit QAM I/Q
DAC RF Tx
input data puncturing interleaver mapper modulator

(a)
Rx
antenna cos(wct) AGC
Pre-select
filter ADC
LPF
QAM Output
Deinterleaver Decoding
de-mapping data
LPF ADC
LNA
VGA
sin(wct)

(b)
Fig. 1.3 A typical, generic, wireless communication system. DAC digital-to-analog converter, LNA low-noise amplifier, LPF low-pass filter, ADC
analog-to-digital converter, AGC automatic gain control, VGA variable-gain amplifier
10 1 Introduction

Fig. 1.4 Illustration of downlink Introduction 15

and uplink channels

Downlink (forward, broadcast, or Uplink (multiaccess, multipoint-to-

point-to-multipoint) channel point channel, or reverse) channel

properly designed block interleaver can be used [67]. For block interleaver, log2(M), where M is signal constellation size,
codewords are written in row-wise fashion into interleaver. Notice that arbitrary two-dimensional (2D) constellation can be
used. Then log2(M) bits are taken into from the interleaver and used to select a point from 2D constellation such as QAM, by
applying appropriate mapping rule. After digital-to-analog conversion (DAC), the corresponding RF signal is upconverted,
amplified, and transmitted by using a transmit antenna.
On receiver side, shown in Fig. 1.3(b), the signal from receive antenna is pre-filtered to select the desired frequency band,
followed by low-noise amplification, down conversion, and analog-to-digital conversion (ADC). The automatic gain control
(AGC) ensures that the signal level at the output of variable gain amplifier (VGA) is constant. After that the QAM de-mapping
takes place, followed by de-interleaver and channel decoder. The purpose of interleaving process is to spread the burst errors
due to fading effects so that the channel code deals only with random errors. When soft-decision decoding is employed, such
as in either turbo or LDPC coding, the QAM de-mapping block is in fact the a posteriori probability (APP) de-mapper, where
symbol log-likelihood ratios (LLRs) are calculated, followed by bit LLRs calculation that gets passed to the soft-decision
decoder, being either turbo decoder or LDPC decoder [3, 35, 64–67].
In wireless communications, multiple users can share the same spectrum, which is commonly referred to as multiaccess
communications. Multiaccess communications suffer from multiaccess interference, which can be both intentional and
unintentional, such as crosstalk or multipath fading. Moreover, the system design, such as in code-division multiple access
(CDMA), can introduce the crosstalk among the users. The performances of multiaccess communication are also dependent
on the fact if the multiaccess communication link is a downlink channel (one transmitter to many receivers scenario) or an
uplink channel (many transmitters to one receiver scenario), as illustrated in Fig. 1.4. The downlink channel is also known as
the forward or broadcast channel as well as point-to-multipoint channel, and all signals are typically synchronous as they
originate from the same transmitter. In contrast, in the uplink channel, also known as multiaccess channel or reverse link as
well as multipoint-to-point channel, the signals are typically asynchronous as they originate from different users at different
locations. In frequency-division multiple access (FDMA), different non-overlapped frequency bands are allocated to different
users. On the other hand, in time-division multiple access (TDMA), the time scale is partitioned into time slots, which are
shared among different users in a round-robin fashion. The TDMA requires precise time synchronism. Therefore, in both
FDMA and TDMA, users are required not to overlap each other in either time domain or frequency domain. However, it is
possible users’ signals to be orthogonal while simultaneously overlapping in both time and frequency domains, and we refer to
such multiaccess scheme as CDMA [68, 69]. To deal with multiuser interference, originating from multiple interfering
signals, we can use the multiuser detection (demodulation) [69], which is also known as interference cancelation or
co-channel interference suppression.
The received power in a wireless channel is affected by attenuation due to following factors:

• Path loss, representing the propagation loss average.

• Shadowing effect, occurring due to slow variations in attenuation mostly caused by scattering obstruction structures. A
signal transmitted through a wireless channel will typically experience random variations due to attenuation from the
objects in the signal path, giving rise to random the variations of the received power at a given distance.
1.3 Wireless Communications Systems Fundamentals 11

• Multipath fading, representing rapid random fluctuations in received power due to multipath propagation and Doppler
frequency shift. In urban/indoor environment, a radio signal transmitted from a fixed source will encounter multiple objects
that produce reflected, diffracted, or scattered copies of transmitted signal, known as multipath signal components.

Because of multipath propagation effect, when a single pulse is transmitted over the multipath channel, the received signal
will appear as a pulse train rather than a single pulse, so that the receiver will experience variations in received power. The
number of multipath components is random, while the i-th multipath component can be characterized by random amplitude
ai(t), random delay τi(t), random Doppler frequency νDi [Hz], and random phase shift ϕi(t) ¼ ωcτi(t)- 2πνDit -ϕ0. This time-
varying nature of a linear wireless channel is described by the time-varying impulse response, denoted as h(τ, t), and here τ is
so called delay spread and t is the time instance. The delay spread is defined as the time difference in signal arriving times of
the last and first arriving (line of sight, LOS) multipath components.
To deal with various fading effects, in particular multipath fading, we will describe in incoming chapters the multicarrier
modulation, in particular orthogonal division multiplexing (OFDM) [2–5, 20, 70, 71], diversity [2, 37], multi-input multi-
output (MIMO) signal processing [2–5, 35, 72–74], adaptive modulation and coding [2–5], and channel equalization [2–4], to
mention a few. The key idea behind the multicarrier modulation is to divide the transmitted bit stream into many lower speed
substreams, which are typically orthogonal to each other, and send these over many different subchannels. The number of
substreams is chosen to insure that each subchannel has a bandwidth smaller than the characteristic channel parameter known
as the coherence bandwidth, so that each subchannel will experience the flat fading. This is an efficient way to deal with
frequency-selective fading effects. The key idea behind the diversity-combining techniques is to use several independent
realizations of that same transmitted signal to improve tolerance to channel impairments. The independent signal copies
transmitted over different paths or degrees of freedom will have a low probability of experiencing the same level of channel
distortion. In wireless communications, the probability that all independent fading paths are in deep fade will be low. The
independent realizations of the same signal will be properly combined to improve the overall reliability of transmission. In
MIMO systems, multiple transmitters and receivers can exploit the knowledge about the channel to determine the rank of
so-called channel matrix H, which is related to the number of independent data streams that can be simultaneously
transmitted. The basic concept of wireless MIMO communication is illustrated in Fig. 1.5, where the number of transmit
antennas is denoted as MTx and the number of receive antennas is denoted as MRx.
The 2D symbol transmitted on the m–th transmit antennas is denoted by xm (m ¼ 1, 2,. . ., MTx), while the received symbol
on the n-th received antenna is denoted as yn. The channel coefficient relating the received symbol on n-th receive antenna and
transmitted symbol on the m-th transmit antenna is denoted as hmn. With MIMO signal processing, we can improve the signal-
to-noise ratio (SNR) of SISO system, improve the symbol error rate slope (vs. SNR), and improve the overall data rate
multiple times. In adaptive modulation techniques, we adapt the transmitter to time-varying wireless channel conditions to
enable reliable and high spectral-efficient transmission. Various wireless communication system’s parameters can be adapted
including the data rate, power, code rate, and error probability. In adaptive coding, we adapt error correction strength (code
rate) depending on time-varying wireless channel conditions. When channel conditions are favorable, the weaker (higher rate)

h11
x1 h12 y1
h1M Tx
h21
h22
x2 y2
h2M Tx
…
…

hM Rx 2 hM Rx 1
…

xM Tx hM Rx M Tx yM Rx

Fig. 1.5 The wireless MIMO communication concept

12 1 Introduction

MTSO
(MSC)

Base
station

Cells

Fig. 1.6 Illustration of the cellular concept

channel code is used. On the other hand, when channel conditions deteriorate, the stronger (lower code rate) channel code is
used. In adaptive modulation and coding (AMC), the channel code and modulation formats are simultaneously adapted.
Before concluding this section, we briefly describe the cellular concept, which is illustrated in Fig. 1.6. A cellular network
consists of large number of mobile users (subscribers) and fixed number of base stations, arranged in particular way to provide
the coverage to the mobile users. The area covered by a particular base station (BS) is called the cell. For better utilization of
resources, the cells are of hexagonal shape. However, in practice the base station is typically arranged irregular, which is
dictated by the configuration of terrane, location of building, and so on. The key idea of cellular concept is to reuse the
channels, to improve the efficiency. The cells denoted by the same color in Fig. 1.6 employ the same frequency in FDMA, the
same time slot in TDMA, or the same signature sequence (code) in CDMA. However, the channel reuse introduces the
interference known as intercell interference.
The spatial separation of the cells that use the same channel, known as the reuse distance, should be as small as possible but
should not exceed the maximum tolerable intercell interference threshold. The different base stations, in a given geographic
area, are connected to the mobile telephone switching office (MTSO), also known as the mobile switching center (MSC). Base
stations and MTSOs coordinate handoff (of a user from one BS to the other) and control functions. Shrinking cell size
increases the capacity, but, on the other hand, it increases the networking burden. The MTSO is connected by the high-speed
link to the public switched telephone network (PSTN), typically by SMF, coaxial cable, or microwave link. The MTSO can
provide the internet access or route the voice calls. The incoming call from a mobile user is connected to the base station,
which is on the other hand connected to the MTSO. The MTSO is further connected to the wired network. The destination
could be an ordinary wire telephone or another mobile user. In another mobile user, the corresponding MTSO will be
identified, and in MTSO, the base station through which destination mobile user is supposed to be connected to will be
identified as well. Finally, the call from one mobile use to another remote mobile user will be established.
Other wireless communication systems include satellite systems; wireless broadband access systems such as WiMAX,
providing fixed and fully mobile internet access; paging systems; radio broadcast systems (AM radio, FM radio, analog TV,
digital audio broadcasting, digital video broadcasting); cordless phone systems; Bluetooth; wireless LAN (Wi-Fi); and ultra-
wideband (UWB) radio, to mention a few, with the most important systems to be describe in incoming chapters.

1.4 Organization of the Book

Each chapter of the book, except introductory one, has over 20 problems to help reader gain deeper knowledge in
corresponding fields. In introduction chapter, we first describe the historical perspective of both wireless and optical
communication systems and describe various generations, from early beginning until today. After that we describe the
fundamentals of optical systems and networks, including WDM network architecture, typical optical networking architecture,
1.4 Organization of the Book 13

bit rates for synchronous/asynchronous optical communication systems, and wavelength regions for fiber-optics
communications. In the same section, we briefly describe various multiplexing schemes such as TDM, WDM, FDM, SCM,
CDM, PDM, and SDM. We also identify the main noise sources and optical channel impairments, in particular for fiber-optics
communications. After that the key optical devises and modules are identified and briefly described. In section related to
wireless communications fundamentals, we describe a typical wireless communication point-to-point link and identify key
devices/subsystems with brief description of operational principles. The multiaccess communications are then described with
key problems being identified. Further, various effects affecting the received power are described including path loss,
shadowing, and multipath fading. We also describe various approaches to deal with multipath fading including OFDM,
diversity, MIMO, equalization, and adaptive modulation and coding. The MIMO operation principle is briefly described next.
Finally, the cellular concept is introduced. In section on organization of the book, the detailed description of chapters is
provided.
Chapter 2 discusses the propagation effects, noise sources, and channel impairments for wireless communications, free-
space optical communications, and fiber-optics communications. The key idea of this chapter is to derive the wave equation
for a given medium from Maxwell’s equation and determine its solutions subject to corresponding boundary conditions. In the
same chapter, we study vectorial nature of the light, derive Snell’s law of refraction from Maxwell’s equations, study total
internal reflection, and explain the light confinement in step-index optical fibers. We also describe how arbitrary polarization
state can be generated by employing tree basic polarizing elements, namely, polarizer, phase shifter (retarder), and rotator. We
also introduce the electromagnetic potentials to simplify the study of generation of electromagnetic waves, followed by
antenna theory basics. After that we discuss interference, coherence, and diffraction effects in optics. Following the detailed
description of laser beam propagation effects over the atmospheric turbulence channels, we describe the simulation models
suitable to study atmospheric turbulence effects. In the section devoted to wireless communication channel effects, we
describe path loss, shadowing, and multipath fading propagation effects. Both ray theory-based and statistical multipath
wireless communication channel models are described. The focus is then moved to signal propagation effects in optical fibers
including fiber attenuation and insertion loss, chromatic dispersion effects, PMD effects, and fiber nonlinearities. Regarding
the fiber nonlinearities, both Kerr nonlinear effects (self-phase modulation, cross-phase modulation) and nonlinear scattering
effects, such as stimulated Raman scattering, are described. The generalized nonlinear Schrödinger equation is then
introduced to study the propagation effects over both single-mode and few-mode fibers, as well as the corresponding split-
step Fourier algorithms to solve it. The following noise sources related to optical communication are further described: mode
partition noise, relative intensity noise, laser phase noise, modal noise, thermal noise, spontaneous emission noise, noise
beating components, quantum shot noise, dark current noise, and crosstalk. Finally, in section devoted to indoor optical
wireless communications, the channel models for both infrared and visible light communications are described.
In Chap. 3, basic components, modules, and subsystems relevant in both optical and wireless communications are
described. The first half of the chapter is related to key components, modules, and subsystems for optical communications.
After describing optical communication basic, focus is changed to optical transmitters. The following classes of lasers are
introduced: Fabry-Perot, distributed feedback (DFB), and distributed Bragg reflector (DBR) lasers. Regarding the external
modulators, both Mach-Zehnder modulator and electro-absorption modulator are described. Regarding the external modulator
for multilevel modulation schemes, phase modulator, I/Q modulator, and polar modulator are described. In subsection on
optical receivers, after the photodetection principle, the p-i-n photodiode, avalanche photodiode (APD), and metal-semicon-
ductor-metal (MSM) photodetectors are briefly described. In the same subsection, both optical receiver high-impedance front
end and transimpedance front end are described. In section on optical fibers, both single-mode fiber (SMF) and multimode
fiber (MMF) are introduced. In subsection on optical amplifiers, after describing optical amplification principles, the following
classes of optical amplifiers are described: semiconductor optical amplifier (SOA), Raman amplifier, and Erbium-doped fiber
amplifier (EDFA). In subsection on optical processing components and modules, the following components/modules are
described: optical couplers, optical filters, and WDM multiplexers/demultiplexers. After that, the principles of coherent
optical detection are introduced, followed by the description of optical hybrids. In subsection on coherent optical balanced
detectors, both homodyne and heterodyne optical balanced detectors for both binary and multilevel (QAM, PSK, APSK)
modulations are described.
In the second half of Chap. 3, basic components and modules of relevance to modern wireless communications are
described. First, the DSP basics, including various discrete-time realizations, are provided. Then the direct digital synthesizer
(DDS), relevant in discrete-time (DT) wireless communications systems, is described. In subsection on multirate DSP and
resampling, various upsampling, downsampling, and resampling schemes/modules are described. In the same subsection, the
polyphase decomposition of the system (transfer) function is introduced and employed in resampling. In subsection on
14 1 Introduction

antenna array antennas, the linear array antennas are described. Finally, in subsection on automatic gain control (AGC), the
AGC schemes operating in passband and baseband are described.
Chapter 4 is devoted to both wireless channel and optical channel capacities. The chapter starts with definitions of mutual
information and channel capacity, followed by the information capacity theorem. We discuss the channel capacity of discrete
memoryless channels, continuous channels, and channels with memory. Regarding the wireless channels, we describe how to
calculate the channel capacity of flat-fading and frequency-selective channels. We also discuss different optimum and
suboptimum strategies to achieve channel capacity including the water-filling method, multiplexed coding and decoding,
channel inversion, and truncated channel inversion. We also study different strategies for channel capacity calculation
depending on what is known about the channel state information. Further, we explain how to model the channel with memory
and describe McMillan-Khinchin model for channel capacity evaluation. We then describe how the forward recursion of the
BCJR algorithm can be used to evaluate the information capacity of channels with memory. The next topic is related to the
evaluation of the information capacity of fiber-optics communication channels with coherent optical detection. Finally, we
describe how to evaluate the capacity of hybrid free-space optical (FSO)-RF channels, in which FSO and RF subsystems
cooperate to compensate for shortcoming of each other.
Chapter 5 is devoted to advanced modulation and multiplexing techniques suitable for both wireless and optical
communication systems. The chapter starts with signal space theory concepts applied to wireless communication systems.
After the geometric representations of signals, we describe various multidimensional modulators and demodulators suitable
for wireless communications applications, in particular the Euclidean distance, correlation, and matched filter-based detectors,
followed by the description of frequency-shift keying (FSK). Both continuous-time (CT) and discrete-time
(DT) implementations for pulse amplitude modulation schemes, suitable for wireless communications, are described as
well. After that the focus is moved to multilevel schemes suitable for both wireless and optical communications applications,
including M-ary PSK, star-QAM, square-QAM, and cross-QAM. Regarding the optical communications, the transmitter for
M-ary PSK, star-QAM, and square/cross-QAM are described in detail. The next topic in the chapter is related to multicarrier
modulation, including description multicarrier systems with both non-overlapping and overlapping subcarriers as well as
introduction of various approaches to deal with fading effects at subcarrier level. The concept of OFDM is also introduced;
however, details are provided in Chap. 7. The MIMO fundamentals are then provided, including the description of key
differences with respect to diversity scheme, as well as the introduction of array, diversity, and multiplexing gains. The
parallel MIMO channel decomposition is briefly described. The details on MIMO signal processing are postponed for
Chap. 8. In section on polarization-division multiplexing (PDM) and four-dimensional (4D) signaling, we describe key
differences between PDM and 4D signaling and describe how both types of schemes can be implemented in both wireless and
optical communications. The focused is the moved to the spatial-division multiplexing (SDM) and multidimensional
signaling. We describe how SDM can be applied in wireless communications first. Then we describe how various degrees
of freedom including amplitude, phase, frequency, polarization states, and spatial modes can be used to convey the
information in optical domain. In the same sections, the SDM concepts for fiber-optics communications are described as
well. The section concludes with SDM and multidimensional signaling concepts applied to free-space optical (FSO)
communications. The next topic is devoted to the signal constellation design, including iterative polar modulation (IPM),
signal constellation design for circular symmetric optical channels, energy-efficient signal constellation design, and optimum
signal constellation design (OSCD). The final section of the chapter is devoted to the nonuniform signaling, in which different
signal constellation points are transmitted with different probabilities.
Chapter 6 is related to advanced detection and channel impairments compensation techniques applicable to both wireless
and optical communication systems. To better understand the principles behind advanced detection and channel impairments
compensation, we start the chapter with fundamentals of detection and estimation theory including optimum receiver design in
symbol error probability sense and symbol error probability derivations. In section on wireless communication systems
performance, we describe different scenarios including outage probability, average error probability, and combined outage-
average error probability scenarios. We also describe the moment generating function-based approach to average error
probability calculation. We also describe how to evaluate performance in the presence of Doppler spread and fading effects. In
section on channel equalization techniques, after short introduction, we first describe how zero-forcing equalizers can be used
to compensate for ISI, chromatic dispersion, and polarization mode dispersion (PMD) in fiber-optics communications; ISI and
multipath fading effects in wireless communications; and ISI and multipath effects in indoor optical wireless communications.
After that we describe the optimum linear equalizer design in the minimum mean-square error sense as well as Wiener filtering
concepts. The most relevant post-compensation techniques, applicable to both wireless and optical communications, are
described next including feedforward equalizer, decision-feedback equalizer, adaptive equalizer, blind equalizers, and
maximum likelihood sequence detector (MLSD) also known as Viterbi equalizer. The turbo equalization technique is
1.4 Organization of the Book 15

postponed for Chap. 9. The next section is devoted to the relevant synchronization techniques. In section on adaptive
modulation techniques, we describe how to adapt the transmitter to time-varying wireless/optical channel conditions to enable
reliable and high spectral-efficient transmission. Various scenarios are described including data rate, power, code rate, error
probability adaptation scenarios, as well as combination of various adaptation strategies. In particular, variable-power
variable-rate modulation techniques and adaptive coded modulation techniques are described in detail. Next, the Volterra
series-based equalization to deal with fiber nonlinearities and nonlinear effects in wireless communications is described. In
section on digital backpropagation, we describe how this method can be applied to deal with fiber nonlinearities, chromatic
dispersion, and PMD in a simultaneous manner. In section on coherent optical detection, we describe various balanced
coherent optical detection schemes for two-dimensional modulation schemes; polarization diversity and polarization
demultiplexing schemes; and homodyne coherent optical detection based on phase-locked loops, phase diversity receivers,
as well as the dominant coherent optical detection sources including laser phase noise, polarization noise, transimpedance
amplifier noise, and amplifier spontaneous emission (ASE) noise. In section on compensation of atmospheric turbulence
effects, we describe various techniques to deal with atmospheric turbulence including adaptive coding, adaptive coded
modulation, diversity approaches, MIMO signal processing, hybrid free-space optical (FSO)-RF communication approach,
adaptive optics, and spatial light modulators (SLM)-based backpropagation method. In particular, linear adaptive optics
techniques are described in detail.
Chapter 7 is devoted to OFDM fundamentals and applications to wireless and optical communications. The chapter starts
with basic of OFDM including the generation of OFDM signal by inverse FFT. After that several basic manipulations on
OFDM symbols, required to deal with multipath effects in wireless communications and dispersion effects in fiber-optics
communications, including guard time and cyclic extension as well as the windowing operation, are described. Next, the
bandwidth efficiency of OFDM is discussed. The next topic in the chapter is devoted to the OFDM system design, description
of basic OFDM blocks, and OFDM-based parallel channel decomposition, including multipath fading radio and dispersion-
dominated channels decomposition. Following the description of coherent optical OFDM (CO-OFDM) principles, the basic
estimation and compensation steps, common to both wireless and optical communication systems, are described including
DFT windowing, frequency synchronization, phase estimation, and channel estimation. Regarding the channel estimation,
both pilot-aided and data-aided channel estimation techniques are described. Following the differential detection in OFDM
description, the focus is on various applications of OFDM in wireless communications including digital audio broadcasting
(DAB), digital video broadcasting (DVB), Wi-Fi, LTE, WiMAX, and ultra-wideband communication (UWB). Next, optical
OFDM applications are discussed starting with description of basic OFDM systems types such as CO-OFDM and direct
detection optical OFDM (DDO-OFDM) systems. Regarding the high-speed spectrally efficient CO-OFDM systems, the
polarization-division multiplexed OFDM systems and OFDM-based superchannel optical transmission systems are described.
Before, the concluding remarks, the application of OFDM in multimode fiber links is described.
Chapter 8 is devoted to various diversity and MIMO techniques capable of improving single-input single-output system
performance. Through MIMO concept it is also possible to improve the aggregate data rate. The diversity and MIMO
techniques described are applicable to wireless MIMO communications as well as free-space optical and fiber-optics
communications with coherent optical detection. Various diversity schemes are described including polarization diversity,
spatial diversity, and frequency diversity schemes. The following receive diversity schemes are described: selection combin-
ing, threshold combining, maximum-ratio combining, and equal-gain combining schemes. Various transmit diversity schemes
are described, depending on the availability of channel state information on transmitter side. The most of the chapter is
devoted to wireless and optical MIMO techniques. After description of various wireless and optical MIMO models, we
describe the parallel decomposition of MIMO channels, followed by space-time coding (STC) principles. The maximum
likelihood (ML) decoding for STC is described together with various design criteria. The following classes of STC are
described: Alamouti code, orthogonal designs, linear space-time block codes, and space-time trellis codes. The corresponding
STC decoding algorithms are described as well. After that we move our attention to spatial-division multiplexing (SDM)
principles and describe various BLAST encoding architectures as well as multi-group space-time coded modulation. The
following classes of linear and feedback MIMO receiver for uncoded signals are subsequently described: zero-forcing, linear
minimum MSE, and decision-feedback receivers. The next topic in the chapter is related to suboptimum MIMO receivers for
coded signals. Within this topic we first describe linear zero-forcing (ZF) and linear MMSE receivers (interfaces) but in
context of STC. Within decision feedback and BLAST receivers, we describe various horizontal/vertical (H/V) BLAST
architecture including ZF and MMSE ones. The topic on suboptimum receivers is concluded with diagonal- (D-)-BLAST and
iterative receiver interfaces. The section on iterative MIMO receivers starts with brief introduction of concept of factor graphs,
following with the description of factor graphs for MIMO channel and channels with memory. We then describe the
sum-product algorithm (SPA) operating on factor graphs, followed by description of SPA for channels with memory. The
16 1 Introduction

following iterative MIMO receivers for uncoded signals are described: ZF, MMSE, ZF H/V-BLAST, and LMMSE H/V-
BLAST receivers. After brief description of factor graphs for linear block and trellis codes, we describe several iterative
MIMO receivers for space-time coded signals. The section on broadband MIMO describes how frequency selectivity can be
used as an additional degree of freedom, followed by description of MIMO-OFDM and space-frequency block-coding
principles. The focus is then moved to MIMO channel capacity calculations for various MIMO channel models including
deterministic, ergodic, and non-ergodic random channels, as well as correlated channel models. The concepts of ergodic and
outage channel capacity are described. The section on MIMO channel capacity ends with MIMO-OFDM channel capacity
description. In MIMO channel estimation section, the following MIMO channel estimation techniques are described: ML,
least squares (LS), and linear minimum MSE MIMO channel estimation techniques. In massive MIMO communications
section, the concepts and various detection schemes are described.
Chapter 9 is related to the advanced channel coding and coded modulation techniques. After linear and BCH codes
fundamentals, we provide the trellis description of linear block codes and describe the corresponding Viterbi decoding
algorithm. After describing the fundamentals of convolutional, RS, concatenated, and product codes, we describe coding with
interleaving as an efficient way to deal with burst of errors and fading effects. Significant space in the chapter is devoted to
codes on graphs, in particular. Turbo, turbo product, and LDPC codes are described together with corresponding decoding
algorithms. Regarding LDPC codes, both binary and nonbinary (NB) LDPC codes are introduced, and corresponding
decoding algorithms as well as their FPGA implementation are described. Additionally, LDPC codes design procedures are
described, followed by rate adaptation. Rate adaptive FGPA implementations of LDPC and generalized LDPC codes are
described as well. The next portion of the chapter is devoted to coded modulation (CM) and unequal error protection (UEP).
After providing the coded modulation fundamentals, we describe trellis-coded modulation (TCM), multilevel coding and
UEP, bit-interleaved-coded modulation (BICM), turbo TCM, and various hybrid multidimensional-coded modulation
schemes suitable for ultra-high-speed optical transmission including multilevel nonbinary LDPC-coded modulation. After
coded modulation sections, the focus of the chapter is on multidimensional turbo equalization. The following topics are
described: nonlinear channels with memory, nonbinary MAP detection, sliding-window multidimensional turbo equalization,
simulation study of multidimensional turbo equalization, and several experimental demonstrations including time-domain
4D-NB-LDPC-CM and quasi-single-mode transmission over transoceanic distances. In section on optimized signal constel-
lation design and optimized bit-to-symbol mapping-based coded modulation, we describe multidimensional optimized signal
constellation design (OSCD), EXIT chart analysis of OSCD mapping rules, nonlinear OSCD-based coded modulation, and
transoceanic multi-Tb/s transmission experiments enabled by OSCD. Finally, in adaptive coding and adaptive coded
modulation section, we describe adaptive coded modulation, adaptive nonbinary LDPC-coded multidimensional modulation
suitable for high-speed optical communications, and adaptive hybrid free-space optical (FSO)-RF coded modulation.
Chapter 10 is devoted to spread spectrum (SS), CDMA, and ultra-wideband (UWB) communication systems. After the
introduction of SS systems, both direct sequence spread spectrum (DS-SS) and frequency-hopping spread spectrum (FH-SS)
systems are described in detail. Regarding the DS-SS systems, after typical transmitter and receiver configurations are
described, we describe in detail the synchronization process including acquisition and tracking stages. To deal with the
multipath fading effects, the use of RAKE receiver is described. The section on DS-SS systems concludes with the spreading
sequences description, in particular pseudo-noise (PN) sequences. In this section, after introduction of the basic randomness
criterions, we introduce maximal-length sequences (m-sequences) and describe their basic properties. In the same section, the
concept of nonlinear PN sequences is introduced as well. Regarding the FH-SS systems, both slow-frequency hopping (SFH)
and fast-frequency hopping (FFH) systems are introduced. We provide the detailed description of FH-SS systems, including
both transmitter and demodulator as well as various time acquisition schemes. In the same section, we describe how to
generate the FH sequences. After the introduction of CDMA systems, we describe various signature waveforms suitable for
use in DS-CDMA systems including Gold, Kasami, and Walsh sequences. We then describe relevant synchronous and
asynchronous CMDA models, with special attention devoted to discrete-time CDMA models. In section on multiuser
detection (MUD), we describe how to deal with interference introduced by various CDMA users. The conventional single-
user detector is used as the reference case. The various MUD schemes are described then, with special attention being paid to
the jointly optimum MUD. The complexity of jointly optimum MUD might be prohibitively high for certain applications, and
for such applications, we provide the description of the decorrelating receiver, which can compensate for multiuser
interference but enhance the noise effect. To solve for this problem, the linear minimum mean-square error (MMSE) receiver
is introduced. The section on MUD concludes with description of the nonlinear MUD schemes employing decisions on bits
from other users (be preliminary or final), with description of successive cancelation and multistage detection schemes. In
section on optical CDMA (OCDMA), we describe both incoherent and coherent optical detection-based OCDMA schemes.
Regarding the incoherent optical detection-based schemes, we first describe how to design various unipolar signature
1.5 Concluding Remarks 17

sequences, known as optical orthogonal codes (OOC), followed by the description of basic incoherent OCDMA schemes
including time-spreading, spectral amplitude coding (SAC), and wavelength hopping (WH) schemes. Related to coherent
optical detection-based OCDMA systems, we describe both pulse laser-based and CW laser-based OCDMA systems. Further,
the hybrid OFDM-CDMA systems are described, taking the advantages of both OFDM and CDMA concepts into account, in
particularly in dealing with frequency-selective fading and narrowband interference while at the same time supporting
multiple users. The following broad classes of hybrid OFDM-CDMA systems are described: multicarrier CDMA
(MC-CDMA), OFDM-CDMA, and OFDM-CDMA-FH systems. In section on UWB communications, we describe the
concepts and requirements as well as transmitter and receiver configurations. Further, various modulation formats, suitable
for UWB communications, are described, categorized into time-domain category (pulse-position nodulation and pulse-
duration modulation) and shape-based category [on-off keying, pulse-amplitude modulation, bi-phase modulation (BPM)
or BPSK, and orthogonal pulse modulation]. Regarding the pulse shapes suitable for orthogonal pulse modulation (OPM), the
modified Hermite polynomials, Legendre polynomials, and orthogonal prolate spheroidal wave functions (OPSWF) are
described. Regarding the UWB channel models suitable for indoor applications, the model due to Saleh and Valenzuela is
described. Both types of UWB systems are described: the impulse radio UWB (IR-UWB), typically DS-CDMA-based, and
multiband OFDM (MB-OFDM).
Chapter 11 is devoted to the physical layer security (PLS) for wireless and optical channels. The chapter starts with
discussion on security issues, followed by the introduction of information-theoretic security and comparison against the
computational security. In the same section, various information-theoretic security measures are introduced, including strong
secrecy and weak secrecy conditions. After that, the Wyner’s wiretap channel model, also known as the degraded wiretap
channel model, is introduced. In the same section, the concept of secrecy capacity is introduced as well as the nested wiretap
coding. Further, the broadcast channel with confidential messages is introduced, and the secrecy capacity definition is
generalized. The focus is then moved to the secret key generation (agreement), the source and channel-type models are
introduced, and corresponding secret key generation protocols are described. The next section is devoted to the coding for the
physical layer security systems, including both coding for weak and strong secrecy systems. Regarding the coding for weak
secrecy systems, the special attention is devoted to two-edge-type LDPC coding, punctured LDPC coding, and polar codes.
Regarding the coding for strong secrecy systems, the focus is on coset coding with dual of LDPC codes and hash functions/
extractor-based coding. The attention is then moved to information reconciliation and privacy amplification. In wireless
channels PLS section, the following topics are covered: MIMO fundamentals, wireless MIMO PLS, and secret key generation
in wireless networks. In section on optical channels PLS, both PLS for spatial-division multiplexing (SDM)-fiber-based
systems and free-space optical (FSO) systems are discussed.

1.5 Concluding Remarks

The historical perspective of both wireless and optical communication systems has been described in Sect. 1.1, together with
various generations, from early beginning until today. After that, in Sect. 1.2, the fundamentals of optical systems and
networks have been described, including WDM network architecture, typical optical networking architecture, bit rates for
synchronous/asynchronous optical communication systems, and wavelength regions for fiber-optics communications. In the
same section, various multiplexing schemes have been described such as TDM, WDM, FDM, SCM, CDM, PDM, and SDM.
The main noise sources and optical channel impairments have been described, in particular for fiber-optics communications.
After that the key optical devises and modules have been identified and briefly described. In section related to wireless
communications fundamentals, Sect. 1.3, we have described a typical wireless communication point-to-point link and
identified key devices/subsystems with brief description of operational principles. The multiaccess communications have
been then described with key problems being identified. Further, various effects affecting the received power have been
described as well including path loss, shadowing, and multipath fading. The various approaches to deal with multipath fading
have been then introduced including OFDM, diversity, MIMO, equalization, and adaptive modulation and coding. The
MIMO operation principles have been briefly described next. Finally, the cellular concept has been introduced. In section on
organization of the book, Sect. 1.4, the detailed description of chapters is provided.
18 1 Introduction

References
1. G.P. Agrawal, Fiber-Optic Communication Systems, 4th edn. (Wiley, Hoboken, NJ, 2010)
2. A. Goldsmith, Wireless Communications (Cambridge University Press, Cambridge, UK/New York, 2005)
3. I. Djordjevic, W. Ryan, B. Vasic, Coding for Optical Channels (Springer, New York, 2010)
4. M. Cvijetic, I.B. Djordjevic, Advanced Optical Communications and Networks (Artech House, Boston/London, 2013)
5. W. Shieh, I. Djordjevic, OFDM for Optical Communications (Elsevier/Academic, Amsterdam/Boston, 2010)
6. A. Osseiran, F. Boccardi, V. Braun, K. Kusume, P. Marsch, M. Maternia, O. Queseth, M. Schellmann, H. Schotten, Scenarios for 5G mobile and
wireless communications: The vision of the METIS project. IEEE Commun. Mag. 52(5), 26–35 (2014)
7. A. Osseiran, J.F. Monserrat, P. Marsch, 5G Mobile and Wireless Communications Technology (Cambridge University Press, Cambridge, UK,
2016)
8. T. Wheeler, Leading Towards Next Generation “5G” Mobile Services (Federal Communications Commission, 2015)
9. America’s 5G Future, FCC Initiatives, available at https://www.fcc.gov/5G
10. Minimum requirements related to technical performance for IMT-2020 radio interface(s), available at: https://www.itu.int/pub/R-REP-M.2410
11. M.2150: Detailed specifications of the terrestrial radio interfaces of International Mobile Telecommunications-2020 (IMT-2020), available at:
https://www.itu.int/rec/R-REC-M.2150/en
12. 5G NR (Rel-15), available at: https://www.3gpp.org/lte-2
13. 3GPP Release 18, available at: https://www.3gpp.org/release18
14. Dohler M et al., Internet of skills, where robotics meets AI, 5G and the Tactile Internet. In Proc. 2017 European Conference on Networks and
Communications (EuCNC), (2017), pp. 1–5
15. W. Saad, M. Bennis, M. Chen, A vision of 6G wireless systems: Applications, trends, technologies, and open research problems. IEEE Netw.
34(3), 134–142 (2020)
16. H. Yang, A. Alphones, Z. Xiong, D. Niyato, J. Zhao, K. Wu, Artificial-intelligence-enabled intelligent 6G networks. IEEE Netw. 34(6),
272–280 (2020)
17. R. Chataut, R. Akl, Massive MIMO systems for 5G and beyond networks-overview, recent trends, challenges, and future research direction.
Sensors (Basel) 20(10), 2753 (2020)
18. I.F. Akyildiz, S. Nie, S.-C. Lin, M. Chandrasekaran, 5G roadmap: 10 key enabling technologies. Comput. Netw. 106, 17–48 (2016)
19. IEEE Standard for Information technology--Telecommunications and information exchange between systems Local and metropolitan area
networks--Specific requirements Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, in IEEE Std
802.11–2012 (Revision of IEEE Std 802.11–2007), pp.1–2793, March 29 2012, https://doi.org/10.1109/IEEESTD.2012.6178212
20. J. Heiskala, J. Terry, OFDM Wireless LANs: A Theoretical and Practical Guide (Sams Publishing, 2002)
21. Hanzo L, Akhtman Y, Wang L, Jiang M, MIMO-OFDM for LTE, WiFi, and WiMAX,. IEEE/Wiley, 2011
22. M.S. Gast, 802.11 Wireless Networks: The Definitive Guide, 2nd edn. (O’Reilly Media, Inc, Sebastopol, CA, 2005)
23. M.S. Gast, 802.11ac: A Survival Guide (O’Reilly Media, Inc, Sebastopol, CA, 2013)
24. M.S. Gast, 802.11an: A Survival Guide (O’Reilly Media, Inc, Sebastopol, CA, 2012)
25. Unlicensed Use of the 6 GHz Band; Expanding Flexible Use in Mid-Band Spectrum Between 3.7 and 24 GHz, available at: https://docs.fcc.gov/
public/attachments/FCC-20-51A1.pdf
26. L. Nuaymi, WiMAX: Technology for Broadband Wireless Access (Wiley, Chichester, UK, 2007)
27. IEEE 802.16–2004, IEEE Standard for Local and Metropolitan Area Networks, Air Interface for Fixed Broadband Wireless Access Systems,
October 2004
28. IEEE 802.16e, IEEE Standard for Local and Metropolitan Area Networks, Air Interface for Fixed Broadband Wireless Access Systems,
Amendment 2: Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands and Corrigendum
1, February 2006
29. IEEE Standard for Air Interface for Broadband Wireless Access Systems. IEEE Std 802.16–2017 (Revision of IEEE Std 802.16–2012),
pp.1–2726, 2 March 2018
30. IEEE 802.16 Working Group on Broadband Wireless Access Standards, see https://grouper.ieee.org/groups/802/16/
31. T.H. Maiman, Stimulated optical radiation in ruby. Nature 187(4736), 493–494 (1960)
32. K.C. Kao, G.A. Hockman, Dielectric-fiber surface waveguides for optical frequencies. Proc. IEEE 113(7), 1151–1158 (1966)
33. D. Tse, P. Viswanath, Fundamentals of Wireless Communication (Cambridge University Press, Cambridge, UK, 2005)
34. S. Benedetto, E. Biglieri, Principles of Digital Transmission with Wireless Applications (Kluwer Academic/Plenum Publishers, New York,
1999)
35. E. Biglieri, Coding for Wireless Channels (Springer, New York, 2005)
36. T.M. Duman, A. Ghrayeb, Coding for MIMO Communication Systems (Wiley, Chichester, UK, 2007)
37. G.L. Stüber, Principles of Mobile Communication, 3rd edn. (Springer, New York, 2012)
38. L. Kazovsky, S. Benedetto, A. Willner, Optical Fiber Communication Systems, vol 1996 (Artech House, Boston, 1996)
39. G. Keiser, Optical Fiber Communications, 3rd edn. (McGraw-Hill, Boston, 2000)
40. R. Ramaswami, K. Sivarajan, Optical Networks: A Practical Perspective, 2nd edn. (Morgan Kaufman, San Francisco, CA, 2002)
41. Z. Ghassemlooy, W. Popoola, S. Rajbhandari, Optical Wireless Communications: System and Channel Modelling with MATLAB (CRC Press,
Boca Raton, FL, 2013)
42. S. Arnon, J.R. Barry, S.R. Karagiannidis, M. Uysal, Advanced Optical Wireless Communication Systems (Cambridge University Press,
New York, 2012)
43. G.P. Agrawal, Nonlinear Fiber Optics, 4th edn. (Elsevier/Academic Press, Amsterdam/Boston, 2007)
44. G.P. Agrawal, Lightwave Technology: Telecommunication Systems (Wiley, Hoboken, NJ, 2005)
45. F.E. Ross, FDDI – a tutorial. IEEE Commun. Mag. 24(5), 10–17 (1986)
46. M.W. Sachs, A. Varma, Fiber channel and related standards. IEEE Commun. Mag. 34(8), 40–49 (1996)
References 19

47. Telcordia Ericsson, Synchronous Optical Network (SONET) Transport Systems: Common Generic Criteria, GR-253-CORE (2009)
48. P. Loshin, F. Kastenholz, Essential Ethernet Standards: RFCs and Protocols Made Practical (Internet Standards) (Wiley, New York, 1999)
49. S.A. Calta, J.A. deVeer, E. Loizides, R.N. Strangwayes, Enterprise systems connection (ESCON) architecture-system overview. IBM J. Res.
Dev. 36(4), 535–551 (1992)
50. S.V. Kartalopoulos, Understanding SONET/SDH and ATM (IEEE Press, Piscataway, NJ, 1999)
51. A. Farrel, I. Bryskin, GMPLS: Architecture and Applications (Morgan Kaufmann Publishers, San Francisco, CA, 2006)
52. P. Yuan, V. Gambiroza, E. Knightly, The IEEE 802.17 media access protocol for high-speed metropolitan-area resilient packet rings. IEEE
Netw. 18(3), 8–15 (2004)
53. ITU-T, Rec. G. 681: Functional characteristics of interoffice and long-haul line systems using optical amplifiers, including optical multiplexing,
ITU-T (10/96), 1996
54. ITU-T, Rec G.704: Synchronous frame structures used at 1544, 6312, 2048, 8448 and 44 736 kbit/s hierarchical levels, ITU-T (10/98), 1998
55. ITU-T, Rec G.7041/Y.1303: Generic framing procedure (GFP), ITU-T (12/01), 2001
56. ITU-T, Rec G.7042/Y.1305: Link capacity adjustment scheme (LCAS) for virtual concatenated signals, ITU-T (11/01), 2001
57. ITU-T, Rec G.694.2: Spectral grids for WDM applications: CWDM wavelength grid, ITU-T (06/02), 2002
58. ITU-T, Rec G.8110/Y.1370.1: Architecture of transport MPLS layer network, ITU-T (11/06), 2006
59. ITU-T, Rec. G.798: Characteristics of Optical Transport Network Hierarchy Functional Blocks”, ITU-T (02/01), 2009
60. S. Gringery, E.B. Basch, T.J. Xia, Technical considerations for supporting bit rates beyond 100 Gb/s. IEEE Commun. Mag. 50(2), 21–30 (2012)
61. Djordjevic IB, Shieh W, Liu X, Nazarathy M (2014), Advanced digital signal processing and coding for multi-Tb/s optical transport. IEEE Sig.
Proc. Mag. 31 (2),. pages 15 and 142. (From the Guest Editors)
62. I.B. Djordjevic, M. Cvijetic, C. Lin, Multidimensional signaling and coding enabling multi-Tb/s optical transport and networking. IEEE Sig.
Proc. Mag. 31, 104–117 (2014)
63. I.B. Djordjevic, On advanced FEC and coded modulation for ultra-high-speed optical transmission. IEEE Commun. Surv. Tutorials 18(3),
1920–1951 (2016)
64. S. Lin, D.J. Costello, Error Control Coding: Fundamentals and Applications, 2nd edn. (Pearson Prentice Hall, Upper Saddle River, NJ, 2004)
65. C. Heegard, S.B. Wicker, Turbo Coding (Springer, New York, 1999)
66. W.E. Ryan, S. Lin, Channel Codes: Classical and Modern (Cambridge University Press, Cambridge, UK/New York, 2009)
67. D. Zou, C. Lin, I.B. Djordjevic, FPGA-based LDPC-coded APSK for optical communication systems. Opt. Express 25(4), 3133–3142 (2017)
68. S. Glisic, B. Vucetic, Spread Spectrum CDMA Systems for Wireless Communications (Artech House, Inc, Boston/London, 1997)
69. S. Verdú, Multiuser Detection (Cambridge University Press, Cambridge, UK/New York, 1998)
70. R. Van Nee, R. Prasad, OFDM Wireless Multimedia Communications (Artech House, Boston/London, 2000)
71. R. Prasad, OFDM for Wireless Communications Systems (Artech House, Boston/London, 2004)
72. E. Biglieri, R. Calderbank, A. Constantinides, A. Goldsmith, A. Paulraj, H.V. Poor, MIMO Wireless Communications (Cambridge University
Press, Cambridge, UK, 2007)
73. J.R. Hampton, Introduction to MIMO Communications (Cambridge University Press, Cambridge, UK, 2014)
74. A. Pulraj, R. Nabar, D. Gore, Introduction to Space-Time Wireless Communications (Cambridge University Press, Cambridge, UK, 2003)
 Propagation Effects in Optical and Wireless
Communications Channels, Noise Sources, 2
and Channel Impairments

Abstract
This chapter discusses the propagation effects, noise sources, and channel impairments for wireless communications, free-
space optical communications, and fiber-optics communications. The key idea of the chapter is to derive the wave equation
for a given medium from Maxwell’s equation and determine its solutions subject to corresponding boundary conditions.
After the introduction of the vector derivatives in orthogonal curvilinear, cylindrical polar, and spherical coordinate
systems, we study vectorial nature of the light, derive Snell’s law of refraction from Maxwell’s equations, study total
internal reflection, and explain the light confinement in step-index optical fibers. We also describe how arbitrary polarization
state can be generated by employing three basic polarizing elements, namely, polarizer, phase shifter (retarder), and rotator.
We then introduce the electromagnetic potentials to simplify the study of generation of electromagnetic waves, followed by
antenna theory basics. We than discuss interference, coherence, and diffraction in optics. After that the detailed description
of laser beam propagation effects over the atmospheric turbulence channels is provided, including the description of
simulation models. In the section on wireless communication channel effects, we study path loss, shadowing, and multipath
fading propagation effects. Both ray theory-based and statistical multipath wireless communication channel models are
described. We then move to signal propagation effects in optical fibers and describe (i) fiber attenuation and insertion loss,
(ii) chromatic dispersion effects, (iii) polarization mode dispersion effects, and (iv) fiber nonlinearities. Regarding the fiber
nonlinearities, both Kerr nonlinear effects (self-phase modulation, cross-phase modulation) and nonlinear scattering effects,
such as stimulated Raman scattering, are described. The generalized nonlinear Schrödinger equation is then described for
both single-mode and few-mode fibers, as well as the corresponding split-step Fourier algorithms to solve it. The following
noise sources related to optical communication are described: mode partition noise, relative intensity noise, laser phase
noise, modal noise, thermal noise, spontaneous emission noise, noise beating components, quantum shot noise, dark current
noise, and crosstalk. In section on indoor optical wireless communications, the channel models for both infrared and visible
light communications are described. The set of problems is provided after the concluding remarks.

2.1 Electromagnetic Field and Wave Equations

Maxwell’s equations are used to describe the change of electric field E and magnetic field H in space and time [1–5]:

∇
E ¼ ∂B=∂t ð2:1Þ
∇
H ¼ J þ ∂D=∂t ð2:2Þ
∇D¼ρ ð2:3Þ
∇B¼0 ð2:4Þ

# The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 21

I. B. Djordjevic, Advanced Optical and Wireless Communications Systems, https://doi.org/10.1007/978-3-030-98491-5_2
22 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

where B denotes the magnetic flux density, D is the electric flux density, and J ¼ σE is the current density. The flux densities
are related to the field vectors by constitutive relations:
D ¼ ε0 E þ P ð2:5Þ

B ¼ μ0 ðH þ M Þ ð2:6Þ
9 1
where P and M denote the induced electric and magnetic densities, ε0 ¼ [4π
910 ] F/m is the permeability in the vacuum,
and μ0 ¼ 4π
107 H/m is the permittivity in the vacuum. Given that there is no electric charge in free-space or fiber media,
the volume density of electric charge ρ ¼ 0, while the conductivity of either free-space or silica is extremely low σ  0; the
Maxwell’s equations can be simplified. Moreover, the silica and free-space are nonmagnetic material, so M ¼ 0. P and E are
mutually connected by:

ð
1

Pðr, t Þ = ε0 χ ðr, t  t 0 ÞEðr, t 0 Þdt 0 , ð2:7Þ

1

where χ denotes the linear susceptibility, which is generally speaking a second-rank tensor. However, this tensor becomes
scalar for an isotropic medium (where P and E are collinear). In free-space P ¼ 0, so that Maxwell’s equations become:

∇
E ¼ μ0 ∂H=∂t ð2:8Þ

∇
H ¼ ε0 ∂E=∂t ð2:9Þ
∇E¼0 ð2:10Þ
∇H ¼0 ð2:11Þ

We are in position now to separate electric and magnetic fields as follows. By taking the curls of Eqs. (2.8) and (2.9), we
obtain:
0 1
∇
ð∇
EÞ ¼ μ0 ∂@∇
H A=∂t ¼ ε0 μ0 ∂ E=∂t 2
2
|fflfflffl{zfflfflffl} ð2:12Þ
ε0 ∂E=∂t
0 1
∇
ð∇
H Þ ¼ ε0 ∂@ |fflffl A 2
ffl} =∂t ¼ ε0 μ0 ∂ H=∂t
∇ffl{zfflffl

E 2
ð2:13Þ
μ0 ∂H=∂t

By employing the following vector identity:

∇
ð∇
Þ ¼ ∇ð∇ Þ  ∇2 ð Þ, ð2:14Þ

and the divergence conditions (2.10), (2.11), we obtain the wave equations:

2 2
1 ∂ E 1 ∂ H
∇2 E ¼ , ∇2 H ¼ 2 , ð2:15Þ
c2 ∂t 2 c ∂t 2

where

c ¼ ðε0 μ0 Þ1=2 ¼ 299792458 m=s ð2:16Þ

is the speed of light. Therefore, the changes of electric and magnetic fields in vacuum propagate with a speed equal to the
speed of light.
2.1 Electromagnetic Field and Wave Equations 23

Maxwell’s equations for electric and magnetic fields in nonconducting isotropic media are the same as those for vacuum,
except that constants ε0 and μ0 for vacuum need to be replaced by the corresponding constant of medium ε and μ. The same
observation is applicable to the wave equations. As a consequence, the speed of propagation v of electromagnetic fields in a
medium is determined by:

v ¼ ðεμÞ1=2 ð2:17Þ

The medium constants ε and μ are related to the vacuum constants ε0 and μ0 as follows:

ε ¼ ε0 εr , μ ¼ μ0 μm , ð2:18Þ

where εr is called dielectric constant or the relative permittivity, while μm is called the relative permeability. The ratio of the
speed of light in vacuum c and the speed of light in medium is known as index of refraction n and can be written as:

 1=2  1=2
εμ ε0 εr μ0 μm
n ¼ c=v ¼ ¼ ¼ ðεr μm Þ1=2 : ð2:19Þ
ε0 μ 0 ε0 μ0

Transparent optical media are typically nonmagnetic (μm ¼ 1) so that the index of refraction is related to the dielectric constant
by n ¼ (εr)1/2. As an illustration, the indices of refraction of the yellow light in the air [at 101.325 kPa (1 atm)] and water are
1.0002926 and 1.33, respectively. The index of refraction in the glass ranges from 1.5 to 1.7, depending on the type of glass.

2.1.1 Vector Derivatives (grad, div, and curl) in Orthogonal Curvilinear Coordinate System

At this point is convenient to introduce the orthogonal curvilinear coordinate system [5, 6]. Any point in orthogonal
curvilinear coordinate system (u, v, w) is determined by the intersection of three warped planes: u ¼ const, v ¼ const, and
w ¼ const. Let u ¼ u(x,y,z), v ¼ v(x,y,z), and w ¼ u(x,y,z) be three independent (uniquely) differentiable functions of Cartesian
coordinates (x,y,z). Then arbitrary point P(x,y,z) in Cartesian coordinates get mapped by these functions into point P(u, v, w) in
orthogonal curvilinear coordinates system u, v, w. The infinitesimal increment in Cartesian coordinates dr ¼ (dx, dy, dz) can be
represented in curvilinear coordinate system as:

∂r ∂r ∂r
dr ¼ du þ dv þ dw ¼ hu dub
u þ hv dvbv þ hw dwb
w, ð2:20Þ
∂u ∂v ∂w

where Lamé coefficients hu, hv, and hw are defined, respectively, as:

  s ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2  2  2   sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2  2  2ffi
 ∂r  ∂x ∂y ∂z ∂r ∂x ∂y ∂z
hu ¼   ¼ þ þ , hv ¼   ¼ þ þ ,
∂u ∂u ∂u ∂u ∂v ∂v ∂v ∂v
cropenup6pt   sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð2:21Þ
 2  2  2ffi
 ∂r  ∂x ∂y ∂z
hw ¼   ¼ þ þ ,
∂w ∂w ∂w ∂w

and ðb
u, bv, w
b Þ are unit-length vectors along coordinate axes in curvilinear coordinate system. The line element can now be
defined as:

pffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
dl ¼ dr  dr ¼ ðhu duÞ2 þ ðhv dvÞ2 þ ðhw dwÞ2 : ð2:22Þ
The volume element in curvilinear coordinate system is simply:

dV ¼ ðb u  f½ðbv  dr Þbv
½ðw

u  dr Þb b  dr Þb
wg ¼ hu hv hw dudvdw: ð2:23Þ
24 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

The surface elements perpendicular to each coordinate curve are simply:

dSu ¼ ðhv dvÞðhw dwÞ ¼ hv hw dvdw, dSv ¼ ðhu duÞðhw dwÞ ¼ hu hw dudw,
cropenup8pt ð2:24Þ
dSw ¼ ðhu duÞðhv dvÞ ¼ hu hv dudv:

Del or nabla operator in orthogonal curvilinear coordinate system is defined as:

1 ∂ 1 ∂ 1 ∂
∇¼ bþ
u bv þ b:
w ð2:25Þ
hu ∂u hv ∂v hu ∂w

The gradient, divergence, and curl in curvilinear coordinates can be determined as follows. The gradient of a scalar function φ
is a vector:

1 ∂φ 1 ∂φ 1 ∂φ
grad φ ¼ ∇φ ¼ bþ
u bv þ b:
w ð2:26Þ
hu ∂u hv ∂v hu ∂w

The divergence of a vector function A is a scalar:

 
1 ∂ðhv hw Au Þ ∂ðhu hw Av Þ ∂ðhu hv Aw Þ
div A ¼ ∇A ¼ þ þ : ð2:27Þ
hu hv hw ∂u ∂v ∂w

The curl of a vector function A ¼ Au b

u þ Avbv þ Aw w
b is a vector:
 
 hu b hvbv b 
 u hw w
1  ∂ ∂ ∂ 
curl A ¼ ∇
A ¼ : ð2:28Þ
hu hv hw  ∂u ∂v ∂w 
 hu Au h v Av hw Aw 

Finally, the Laplacian Δ ¼ ∇2 ¼ ∇  ∇ of a scalar function φ is a scalar:

      
1 ∂ hv hw ∂φ ∂ hu hw ∂φ ∂ hu hv ∂φ
div grad φ ¼ ∇2 φ ¼ þ þ : ð2:29Þ
hu hv hw ∂u hu ∂u ∂v hv ∂v ∂w hw ∂w

Now by using the div and curl Eqs. (2.27) and (2.28), we are in position to solve Maxwell’s equations (2.8, 2.9, 2.10, and
2.11), subject to the appropriate boundary conditions. On the other hand, by using the Laplacian Eq. (2.29), we can solve the
wave equations (2.15).
In Cartesian coordinate system, the generalized coordinates are u ¼ x, v ¼ y, and w ¼ z ðb
u¼b x, bv ¼ by, w
b ¼ bzÞ so that Lamé
coefficients hu, hv, and hw are all equal to 1. The corresponding vector derivatives are:

∂φ ∂φ ∂φ
grad φ ¼ ∇φ ¼ b
xþ by þ bz, ð2:30Þ
∂x ∂y ∂z

∂Ax ∂Ay ∂Az

div A ¼ ∇A ¼ þ þ , ð2:31Þ
∂x ∂y ∂z
 
 b
x by bz 
 
 
∂ ∂ ∂
curl A ¼ ∇
A ¼  
 ∂x ∂y ∂z 
cropenup3pt   ð2:32Þ
A A A 
x y z
     
∂Az ∂Ay ∂Ax ∂Az ∂Ay ∂Ax
¼  b
xþ  by þ  bz,
∂y ∂z ∂z ∂x ∂x ∂y
2.1 Electromagnetic Field and Wave Equations 25

2 2 2
∂ φ ∂ φ ∂ φ
div grad φ ¼ ∇2 φ ¼ þ þ : ð2:33Þ
∂u2 ∂v2 ∂w2

The vector derivatives (grad, div, curl, and Laplacian) in cylindrical polar and spherical coordinate systems are described in
incoming section.

2.1.2 Vector Derivatives (grad, div, and curl) in Cylindrical Polar and Spherical Coordinate Systems

The cylindrical polar coordinates correspond to the polar (ρ,ϕ) coordinates with an additional z-coordinate directed out of the
xy-plane. The cylindrical polar coordinates (ρ,ϕ,z) are related to Cartesian coordinates (x,y,z) as follows:

x ¼ ρ cos ϕ, y ¼ ρ sin ϕ, z ¼ z ð2:34Þ

The Lamé coefficients hρ, hϕ, and hz are determined as follows:

s
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2  2  2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
∂x ∂y ∂z
hρ ¼ þ þ ¼ ð cos ϕÞ2 þ ð sin ϕÞ2 þ 0 ¼ 1
∂ρ ∂ρ ∂ρ
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2  2  2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
cropenup3pt hϕ ¼ ∂x ∂y ∂z ð2:35Þ
þ þ ¼ ðρ sin ϕÞ2 þ ðρ cos ϕÞ2 þ 0 ¼ ρ
∂ϕ ∂ϕ ∂ϕ
s
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2  2  2
∂x ∂y ∂z pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
hz ¼ þ þ ¼ 0þ0þ1¼1
∂z ∂z ∂z

The unit vectors in cylindrical polar coordinate system are related to the unit vectors in Cartesian coordinate system by:

b
ρ ¼ cos ϕb b ¼  sin ϕb
x þ sin ϕby, ϕ x þ cos φby,bz ¼ bz ð2:36Þ

By replacing Lamé coefficients from Eq. (2.35) into Eqs. (2.26, 2.27, 2.28, and 2.29), the operators of space differentiation
become:

∂φ 1 ∂φ b ∂φ
grad φ ¼ ∇φ ¼ b
ρþ ϕþ bz, ð2:37Þ
∂ρ ρ ∂ϕ ∂z


1 ∂ Aρ ρ ∂Aϕ ∂Az
div A ¼ ∇A ¼ þ þ , ð2:38Þ
ρ ∂ρ ∂ϕ ∂z
 
 b b bz 
 ρ ρϕ
 
1 ∂ ,
curl A ¼ ∇
A ¼  ∂ ∂
 ð2:39Þ
ρ  ∂ρ ∂ϕ ∂z 
 
 Aρ ρAϕ Az 
      
1 ∂ ∂φ ∂ 1 ∂φ ∂ ∂φ
div grad φ ¼ ∇ φ ¼
2
ρ þ þ ρ
ρ ∂ρ ∂ρ ∂ϕ ρ ∂ϕ ∂z ∂z
  2 2 ð2:40Þ
1 ∂ ∂φ 1∂ φ ∂ φ
¼ ρ þ 2 2þ 2:
ρ ∂ρ ∂ρ ρ ∂ϕ ∂z
26 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

The spherical coordinates (ρ,θ,ϕ) are related to the Cartesian coordinates (x,y,z) through the following relations:

x ¼ ρ cos ϕ sin θ, y ¼ ρ sin ϕ sin θ, z ¼ ρ cos θ ð2:41Þ

The Lamé coefficients hρ, hθ, and hϕ are determined as follows:

s
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2  2  2
∂x ∂y ∂z
hρ ¼ þ þ ¼1
∂ρ ∂ρ ∂ρ
s
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2  2  2
∂x ∂y ∂z ð2:42Þ
hθ ¼ þ þ ¼ρ
∂θ ∂θ ∂θ
s
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2  2  2
∂x ∂y ∂z
hϕ ¼ þ þ ¼ ρ sin θ
∂ϕ ∂ϕ ∂ϕ
By replacing Lamé coefficients from Eq. (2.42) into Eqs. (2.26, 2.27, 2.28, and 2.29), the operators of space differentiation
become:

∂φ 1 ∂φ b 1 ∂φ b
grad φ ¼ ∇φ ¼ b
ρþ θþ ϕ, ð2:43Þ
∂ρ ρ ∂θ ρ sin θ ∂ϕ


1 ∂ Aρ ρ2 ∂ð sin θAθ Þ ∂Aϕ
div A ¼ ∇A ¼ 2 sin θ þρ þρ , ð2:44Þ
ρ sin θ ∂ρ ∂θ ∂ϕ
 
 b ρb b 
 ρ θ ρ sin θϕ
 
1  ∂ ∂ ∂ 
curl A ¼ ∇
A ¼ 2  , ð2:45Þ
ρ sin θ  ∂ρ ∂θ ∂ϕ 
 
 Aρ ρAθ ρ sin θAϕ 
     2 
1 ∂ ∂φ ∂ ∂φ 1 ∂ φ
div grad φ ¼ ∇2 φ ¼ 2 sin θ ρ2 þ sin θ þ
ρ sin θ ∂ρ ∂ρ ∂θ ∂θ sin θ ∂ϕ2
    2
ð2:46Þ
1 ∂ ∂φ 1 ∂ ∂φ 1 ∂ φ
¼ 2 ρ2 þ 2 sin θ þ 2 :
ρ ∂ρ ∂ρ ρ sin θ ∂θ ∂θ ρ sin 2 θ ∂ϕ2

2.2 Propagation of Electromagnetic Waves

2.2.1 Propagation of Plane Waves

If we resolve the vector wave equations for electric and magnetic fields, Eq. (2.15), into Cartesian coordinates, we will notice
that each component will satisfy the following scalar equation:

2
∂ U ∂ U ∂ U
2
1 ∂ U 
2 2 
∇2 U ¼ þ 2 þ 2 ¼ 2 , U 2 Ex , Ey , Ez , H x , H y , H z ð2:47Þ
∂x 2 ∂y ∂z v ∂t 2

It can easily verified by substitution that the following three-dimensional (3D) plane harmonic wave function satisfies the
Eq. (2.47):
jðωtkrÞ
U ðx, y, z, t Þ ¼ U 0 e , ð2:48Þ

where r is the position vector given by r ¼ xb

x þ yby þ zbz, ω is the carrier angular frequency, and k is the propagation (wave)
vector represented in Cartesian coordinates as k ¼ kxbx þ kyby þ k zbz ( j is the imaginary unit). The magnitude of propagation
2.2 Propagation of Electromagnetic Waves 27

vector is the (angular) wavenumber k ¼ |k| ¼ ω/v ¼ 2π/λ, where λ is the wavelength (the propagation distance over which the
wave function goes through one cycle). The surfaces of constant phase, also known as wavefronts, are determined by:

ωt  k  r ¼ ωt  kx x  ky y  kz z ¼ const, ð2:49Þ

and the corresponding phase velocity is given by:

ω ω
v¼ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð2:50Þ
k
kx þ k2y þ k 2z
2

Clearly, the equiphase surfaces of plane wave travel at phase velocity speed in direction of propagation vector k, which is
normal to the plane wave surfaces.
Let us now observe two harmonic wave traveling slightly different angular frequencies denoted, respectively, as ω  Δω
and ω + Δω. The propagation constants can be slightly different as well: k  Δk and k + Δk. By assuming that these two waves
propagate along z-axis, by employing the superposition principle, we can write:

j½ðωΔωÞtðkΔk Þz j½ðωþΔωÞtðkþΔkÞz jðωtþkzÞ

U ¼ U0e þ U0e ¼ 2 cos ðΔωt  ΔkzÞU 0 e : ð2:51Þ

Therefore, in this case, the phase travels at phase velocity, while the envelope travels at the group velocity vg, defined as:
 
dω dv k dn
vg ¼ ¼vλ ¼v 1 : ð2:52Þ
dk dλ n dk

In media in which signals at different wavelengths travel at different speeds, the dispersion effect occurs.
The action of the time derivative on the harmonic wave, given by Eq. (2.48), is simply multiplication by jω. On the other
hand, the action of del operator ∇ is multiplication with –jk. By taking these two properties into account, the Maxwell’s
equations for plane harmonic waves become:

k
E ¼ ωμH ð2:53Þ

k
H ¼ ωεE ð2:54Þ

kE¼0 ð2:55Þ

kH ¼0 ð2:56Þ

From harmonic wave Maxwell’s equations, we conclude that electric and magnetic fields are perpendicular to one another and
they are simultaneously perpendicular to the direction of propagation. The magnitudes of electric and magnetic fields are
related as:
rffiffiffiffiffi
ω E μ0
H¼ε E ¼ εvE ¼ n , Z 0 ¼ ¼ 120π ffi 377 Ω ð2:57Þ
k
|{z} Z 0 ε0
v

where Z0 is commonly referred to as the impedance of the free space.pffiffiffiffiffiffiffiffi The impedance of the medium, characterized by
dielectric constant ε and magnetic permittivity μ, will be then Z ¼ μ=ε.
The flow of electromagnetic energy per unit area is given by the Poynting vector, defined as the cross product of the electric
and magnetic field vectors:

S ¼ E
H : ð2:58Þ
28 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

For plane harmonic waves, the electric and magnetic fields components have the form given by Eq. (2.48) so that the Poynting
vector becomes:

k
S ¼ E
H  ¼ E0 e jðωtkrÞ

H 0 ejðωtkrÞ ¼ E0
H 0 ¼ E 0 H 0 ¼ Ib
n, ð2:59Þ
|ffl{zffl}|{z}
k
I
bn

where

I ¼ E 0 H 0 ¼ jE 0 j2 =ðZ 0 =nÞ ¼ jE0 j2 =Z

|fflfflffl{zfflfflffl}
Z

is known as irradiance and b

n is a unit vector in the direction of propagation.

2.2.2 Vectorial Nature of the Light, Snell’s Law of Refraction, Reflection Coefficients, and Total
Internal Reflection

2.2.2.1 Polarization
The electric/magnetic field of plane linearly polarized waves is described as follows [1, 7–10]:

Aðr, t Þ ¼ pA0 exp ½ jðωt  k  rÞ, A 2 fE, H g ð2:60Þ

where E (H) denotes electric (magnetic) field, p denotes the polarization orientation, r is the position vector, and k denotes the
wave vector. For the x-polarization waves (p ¼ b x, b
k ¼ kbz), the Eq. (2.60) becomes:

Ex ðz, t Þ = b
xE0x cos ðωt  kzÞ, ð2:61Þ
while for y-polarization (p ¼ by, b
k ¼ kbz), it becomes:

Ey ðz, tÞ ¼ byE0y cos ðωt  kz þ δÞ, ð2:62Þ

where δ is the relative phase difference between the two orthogonal waves. The resultant wave can be obtained by combining
(2.61) and (2.62) as follows:

Eðz, t Þ ¼ Ex ðz, t Þ þ Ey ðz, t Þ ¼ b

xE 0x cos ðωt  kzÞ þ byE 0y cos ðωt  kz þ δÞ: ð2:63Þ

The linearly polarized wave is obtained by setting the phase difference to an integer multiple of 2π, namely, δ ¼ m  2π:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi E 0y
Eðz, t Þ ¼ ðb
xE cos θ þ byE sin θÞ cos ðωt  kzÞ; E ¼ E 20x þ E20y , θ ¼ tan 1 : ð2:64Þ
E 0x

By ignoring the time-dependent term, we can represent the linear polarization as shown in Fig. 2.1(a). On the other hand, if
δ 6¼ m  2π, the elliptical polarization is obtained. From Eqs. (2.61) and (2.62), by eliminating the time-dependent term, we
obtain the following equation of ellipse:
 2  2
Ex Ey Ex Ey
þ 2 cos δ ¼ sin 2 δ, ð2:65Þ
E0x E 0y E 0x E0y
which is shown in Fig. 2.1(b). By setting δ ¼ π2 , 3 π2 , . . ., the equation of ellipse becomes:

 2  2
Ex Ey
þ ¼ 1: ð2:66Þ
E0x E0y
2.2 Propagation of Electromagnetic Waves 29

Fig. 2.1 Polarization forms: (a) linear, (b) elliptic, and (c) circular polarizations

By setting further E 0x ¼ E0y ¼ E 0 ; δ ¼ π2 , 3 π2 , . . ., the equation of ellipse becomes the circle:

E2x þ E2y ¼ 1, ð2:67Þ

and corresponding polarization is known as circular polarization (see Fig. 2.1c). Right circularly polarized wave is obtained
for δ ¼ π/2 + 2mπ:

E = E0 ½b
x cos ðωt  kzÞ  by sin ðωt  kzÞ, ð2:68Þ

Otherwise, for δ ¼ π/2 + 2mπ, the polarization is known as left circularly polarized.
Very often, the Jones vector representation of polarization wave is used:
   
E x ðt Þ E 0x jðωtkzÞ
E ðt Þ ¼ ¼ e , ð2:69Þ
E y ðt Þ E 0y ejδ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
with complex phasor term being typically omitted in practice. If we normalize the Jones vector by dividing by E20x þ E20y ,
we can represent x-polarization state, denoted as |xi, and y-polarization state, denoted as |yi, respectively, by:
   
1 0
jxi ¼ , jyi ¼ : ð2:70Þ
0 1

On the other hand, right- and left-circular polarization states, denoted respectively as |Ri and |Li, can be written as:
   
1 1 1 1
jRi ¼ pffiffiffi , jLi ¼ pffiffiffi : ð2:71Þ
2 j 2 j

In Eqs. (2.70 and 2.71), we used Dirac notation to denote the column vectors [7]. In Dirac notation, with each column vector
(“ket”) | ψi, defined as
" #
ψx
jψ i ¼ , ð2:72Þ
ψy

we associate a row vector (“bra”) hψ| as follows:

h i
hψ j ¼ ψ x ψ y : ð2:73Þ
30 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

The scalar (dot) product of ket | ϕi bra hψ| is defined by “bracket” as follows:

hϕjψ i ¼ ϕx ψ x þ ϕy ψ y ¼ hψjϕi : ð2:74Þ

The normalization condition can be expressed in terms of dot product by:

hψjψ i ¼ 1: ð2:75Þ

Based on (2.70 and 2.71), it is evident that:

hxjxi ¼ hyjyi ¼ 1, hRjRi ¼ hLjLi ¼ 1:

Because the vectors |xi and |yi are orthogonal, their dot product is zero:

hxjyi ¼ 0, ð2:76Þ

and they form the basis. Any polarization state vector | ψi can be written as a linear superposition of basis kets as follows:
!
ψx
jψ i ¼ ¼ ψ x jxi þ ψ y jyi: ð2:77Þ
ψy

Another interesting representation is Stokes vector representation:

2 3
S0 ð t Þ
6 7
6 S1 ð t Þ 7
Sðt Þ ¼ 6
6 S2 ð t Þ
7,
7 ð2:78Þ
4 5
S3 ð t Þ

where the parameter S0 is related to the optical intensity by:

 2
S0 ð t Þ ¼ j E x ð t Þ j 2 þ  E y ð t Þ  : ð2:79Þ

The parameter S1 > 0 is related to the preference for horizontal polarization, and it is defined by:
 2
S1 ð t Þ ¼ j E x ð t Þ j 2   E y ð t Þ  : ð2:80Þ

The parameter S2 > 0 is related to the preference for π/4 SOP:

S2 ðt Þ ¼ Ex ðt ÞEy ðt Þ þ E x ðt ÞE y ðt Þ: ð2:81Þ

Finally, the parameters S3 > 0 is related to the preference for right-circular polarization, and it is defined by:
h i
S3 ðt Þ ¼ j E x ðt ÞEy ðt Þ  E x ðt ÞE y ðt Þ : ð2:82Þ

The Stokes vector for the elliptically polarized light, given by Eq. (2.69), can be expressed as follows:
2 3 2 3
S0 E20x þ E 20y
6 7 6 7
6 S1 7 6 E20x  E 20y 7
S¼6 7¼6
6 S2 7 6 2E E cos δ 7:
7 ð2:83Þ
4 5 4 0x 0y 5
S3 2E E
0x 0y sin δ
2.2 Propagation of Electromagnetic Waves 31

The parameter S0 is related to other Stokes parameters by:

S20 ðt Þ ¼ S21 ðt Þ þ S22 ðt Þ þ S23 ðt Þ: ð2:84Þ

The degree of polarization, denoted as DOP, is defined by:

1=2
S2 þ S22 þ S23
DOP ¼ 1 , 0
DOP
1 ð2:85Þ
S0

For DOP ¼ 1 the light is completely polarized. The natural light is unpolarized as it represents incoherent superposition of
50% of light polarized along x-axis and 50% of light polarized along y-axis. The Stokes vector for unpolarized light is
given by:
2 2 3
3
S0 1
607 607
6 7 6 7
Sunpolarized ¼ 6 7 ¼ S0 6 7 : ð2:86Þ
405 405
0 0

The partially polarized light represents the mixture of the polarized light and unpolarized light so that the corresponding
Stokes vector is given by:
2 3 2 3
S0 S0
607 6S 7
6 7 6 17
S ¼ ð1  DOPÞ6 7 þ DOP6 7: ð2:87Þ
405 4 S2 5
0 S3

The x-polarization (linear horizontal polarization) is defined by E0y ¼ 0, while the y-polarization (linear vertical polarization)
by E0x ¼ 0 so that the corresponding Stokes representations are given, based on Eq. (2.83), by:
2 3 2 3
1 1
617 6 1 7
6 7 6 7
Sx ¼ S0 6 7, Sy ¼ S0 6 7: ð2:88Þ
405 4 0 5
0 0

The right-circular polarization is defined by E0y ¼ E0x ¼ E0 and δ ¼ π/2, while left-circular polarization is defined by
E0y ¼ E0x ¼ E0 and δ ¼ π/2 so that the corresponding Stokes representations are given, based on Eq. (2.83), by:
2 3 2 3
1 1
607 6 0 7
6 7 6 7
SR ¼ S0 6 7, SL ¼ S0 6 7: ð2:89Þ
405 4 0 5
1 1

After the normalization of Stokes coordinates Si with respect to S0, the normalized Stokes parameters are given by:

Si
si ¼ ; i ¼ 0, 1, 2, 3: ð2:90Þ
S0

If the normalized Stokes parameters are used, the polarization state can be represented as a point on a Poincaré sphere, as
shown in Fig. 2.2. The points located at the opposite sides of the line crossing the center represent the orthogonal
polarizations. The normalized Stokes (Cartesian) coordinates are related to the sphere coordinates, as defined in Fig. 2.2, by:
32 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.2 Representation of

polarization state as a point on a
Poincaré sphere

Fig. 2.3 The ellipticity and

azimuth of the polarization ellipse

s1 ¼ cos ð2εÞ cos ð2ηÞ, s2 ¼ cos ð2εÞ sin ð2ηÞ, s3 ¼ sin ð2εÞ: ð2:91Þ

The polarization ellipse is very often represented in terms of the ellipticity and the azimuth, which are illustrated in Fig. 2.3.
The ellipticity tanε is defined by the ratio of half-axes lengths. The corresponding angle is called ellipticity angle and denoted
by ε. The small ellipticity means that polarization ellipse is highly elongated, while for zero ellipticity, the polarization is
linear. For ε ¼ π/4, the polarization is circular. For ε > 0 the polarization is right-elliptical. On the other hand, the azimuth
angle η defines the orientation of the main axis of ellipse with respect to Ex (x-axis). The Stokes parameters are related to the
azimuth (orientation) angle η and ellipticity angle ε as follows:

S1 ¼ S0 cos ð2εÞ cos ð2ηÞ, S2 ¼ S0 cos ð2εÞ sin ð2ηÞ, S3 ¼ S0 sin ð2εÞ: ð2:92Þ

The azimuth angle η and ellipticity angle ε are related to the parameters of the polarization ellipse, given by Eq. (2.65), by

2E0x E0y cos δ 2E 0x E 0y sin δ

tan 2μ ¼ , sin 2ε ¼ : ð2:93Þ
E 20x  E20y E 20x þ E 20y

Let us now observe the polarizer-analyzer ensemble, shown in Fig. 2.4. When an electromagnetic wave passes through the
polarizer, it can be represented as a vector in the xOy plane transversal to the propagation direction, as given by Eq. (2.63),
where the angle θ depends on the filter orientation. By introducing the unit vector b p ¼ ð cos θ, sin θÞ, the Eq. (2.63) can be
rewritten as:

E ¼ E 0b
p cos ðωt  kzÞ: ð2:94Þ

If θ ¼ 0 rad, the light is polarized along x-axis, while for θ ¼ π/2 rad, it is polarized along y-axis. After the analyzer, whose
axis makes an angle ϕ with respect to x-axis, which can be represented by unit vector b n ¼ ð cos ϕ, sin ϕÞ, the output electric
field is given by:
2.2 Propagation of Electromagnetic Waves 33

Fig. 2.4 The polarizer-analyzer

ensemble for study of the photon
polarization

E0 ¼ ðE  b n ¼ E 0 cos ðωt  kzÞðb

nÞb pb
nÞb
n
¼ E0 cos ðωt  kzÞ½ð cos θ, sin θÞ  ð cos ϕ, sin ϕÞb
n
ð2:95Þ
¼ E0 cos ðωt  kzÞ½ cos θ cos ϕ þ sin θ sin ϕb n
¼ E0 cos ðωt  kzÞ cos ðθ  ϕÞb
n:

The intensity of the analyzer output field is given by:

I 0 ¼ jE0 j ¼ jE 0 j2 cos 2 ðθ  ϕÞ ¼ I cos 2 ðθ  ϕÞ,

2
ð2:96Þ
|ffl{zffl}
I

which is commonly referred to as Malus law.

2.2.2.2 Snell’s Law of Refraction, Reflection Coefficients, and Total Internal Reflection
We now study space-time dependence of the incident k, reflected kr, and transmitted kt waves at the plane boundary as
illustrated in Fig. 2.5. The space-time dependencies of electric/magnetic fields of these waves are given by:

A ¼ A0 exp ½ jðωt  k  rÞ, Ar ¼ A0 exp ½ jðωt  kr  rÞ,

ð2:97Þ
At ¼ A0 exp ½ jðωt  kt  rÞ:

At the boundary, since the time terms are already equal, the space terms must be equal as well:

k  r ¼ kr  r ¼ kt  r, ð2:98Þ

indicating that the propagation vectors are coplanar, laying in the same plane commonly referred to as the incident plane, and
that their projections on the boundary (xz) plane are equal:

k sin θ ¼ kr sin θr ¼ kt sin ϕ: ð2:99Þ

Given that the incident and reflected wave propagate in the same medium corresponding wavenumbers are equal k ¼ kr so that
the first equation in (2.99) reduces to:

θ ¼ θr , ð2:100Þ

which is known as the reflection law.

34 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.5 Space dependence of

incident, reflected, and
transmitted waves at a plane
boundary

On the other hand, because the ratio of wavenumbers of incident and transmitted waves can be written as:

k ω=v 1=v c=v n1

¼ ¼ ¼ ¼ , ð2:101Þ
kt ω=vt 1=vt c=vt n2

from Eq. (2.99), we obtain:

n1 sin θ ¼ n2 sin ϕ, ð2:102Þ

which is known as Snell’s law of refraction.

From Maxwell’s curl Equations (2.53) and (2.54), we can determine the spatial dependence of propagation, electric field,
and magnetic field vectors as illustrated in Fig. 2.6, where two cases of interest are observed. In Fig. 2.6(a), the incident
electric field vector is parallel to the boundary plane, that is, perpendicular to the incidence plane. This case is commonly
referred to as the transverse electric or TE polarization. The dot in the circle indicates that the electric field vector is
perpendicular to the plane of incidence with direction towards the reader. In Fig. 2.6(b), the incident magnetic field vector is
parallel to the boundary plane, that is, perpendicular to the incidence plane. This case is commonly referred to as the
transverse magnetic or TM polarization. The cross in the circle indicates that the electric field vector is perpendicular to the
plane of incidence with direction form the reader.
Now we apply well-known boundary conditions from electromagnetics [5, 8, 11] requiring that tangential components of
the electrical and magnetic fields must be continuous as we cross over the interface of two media (the boundary plane). From
Fig. 2.6(a), we conclude that for TE polarization E + Er ¼ Et. On the other hand, from Fig. 2.6(b), we conclude that for TM
polarization HHr ¼ Ht, which is equivalent to kEkEr ¼ ktEt, since H is proportional to kE. The magnetic field tangential
components must satisfy boundary condition for TE polarization, which leads to:

H cos θ þ H r cos θ ¼ H t cos ϕ , kE cos θ þ kE r cos θ ¼ kt E t cos ϕ ð2:104Þ

2.2 Propagation of Electromagnetic Waves 35

Fig. 2.6 Spatial dependence of

wave vectors for: (a) TE
polarization and (b) TM
polarization

In equivalent representation of (2.104), we used the fact that H is proportional to kE. The electric field tangential components
must satisfy boundary condition for TM polarization, which leads to:

E cos θ þ Er cos θ ¼ E t cos ϕ: ð2:105Þ

36 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Based on the boundary conditions’ equations provided above, we can determine the coefficients of reflections for TE and TM
polarizations as follows:
h i h i
Er cos θ  n cos ϕ E cos ϕ  n cos θ n
rs ¼ ¼ , rp ¼ r ¼ ,n ¼ 2: ð2:106Þ
E TE cos θ þ n cos ϕ E TM cos ϕ þ n cos θ n1
In similar fashion, the coefficients of transmission can be determined as:
h i h i
Et 2 cos θ E 2 cos θ
ts ¼ ¼ , tp ¼ t ¼ : ð2:107Þ
E TE cos θ þ n cos ϕ E TM cos ϕ þ n cos θ

By using the Snell’s law, n ¼ sinθ/sinϕ, the reflection and transmission coefficients can be rewritten in the following form:

sin ðθ  ϕÞ tan ðθ  ϕÞ
rs ¼  ,r ¼  : ð2:108Þ
sin ðθ þ ϕÞ p tan ðθ þ ϕÞ

2 cos θ sin ϕ 2 cos θ sin ϕ

ts ¼ ,t ¼ , ð2:109Þ
cos ðθ þ ϕÞ p sin ðθ þ ϕÞ cos ðθ  ϕÞ

and these forms are known as Fresnel’s equations.

The fraction of reflected energy of incident light is known as reflectance and can be calculated as:
 2
Rs ¼ jr s j2 , Rp ¼ r p  : ð2:110Þ

As an illustration for normal incidence, when θ ¼ 0, from Eq. (2.98), we find that:

h i
n1 2
Rs ¼ Rp ¼ : ð2:111Þ
nþ1

So far we considered the case for which n2 > n1 (or n > 1), and this case is known as external reflection. In external
reflection, the incident wave is approaching the boundary (interface) from the medium with smaller refraction index. If we
eliminate the refractive angle ϕ from Eq. (2.106), with the help of Snell’s law, n ¼ sinθ/sinϕ, the reflection coefficients can be
rewritten as:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
cos θ 
n2  sin 2 θ n2  sin 2 θ  n2 cos θ
rs ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi , r p ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð2:112Þ
cos θ þ n2  sin 2 θ n2  sin 2 θ þ n2 cos θ

From Eq. (2.112) it is evident that in external reflection case, the reflection coefficients are always real. However, when the
incident wave is approaching the interface from medium with larger index of refraction, n < 1, for incident angles θ > sin1n,
the reflection coefficients become complex and can be represented as:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
cos θ  j sin 2 θ  n2 n2 cos θ þ j sin 2 θ  n2
rs ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi , r p ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi , θ > tan 1 n: ð2:113Þ
cos θ þ j sin 2 θ  n2 n2 cos θ þ j sin 2 θ  n2

This case is commonly referred to as internal reflection. The angle tan1n is known as the critical angle. The reflectance when
the incident angle is larger than or equal to critical angle is equal to R ¼ 1, indicating that total reflection occurs.

2.2.2.3 Optical Fibers

One of the important applications of total internal reflection is the propagation of light in optical fibers. The optical fiber,
shown in Fig. 2.7, consists of two waveguide layers, the core (of refractive index n1) and the cladding (of refractive index n2),
protected by the jacket (the buffer coating). The majority of the power is concentrated in the core, although some portion can
spread to the cladding. There is a difference in refractive indices between the core and cladding (n1 > n2), which is achieved by
mix of dopants commonly added to the fiber core. The refractive index profile for step-index fiber is shown in Fig. 2.7(a),
2.2 Propagation of Electromagnetic Waves 37

Fig. 2.7 Optical fibers: (a)

refractive index profile for step-
index fiber and (b) the light
confinement in step-index fibers
through the total internal
reflection

while the illustration of light confinement by the total internal reflection is shown in Fig. 2.7(b). Thep ray will ffibe totally
ffiffiffiffiffiffiffiffiffiffiffiffiffiffi
reflected from the core-cladding interface (a guided ray) if the following condition is satisfied: n0 sin θi < n1  n22, where θi
2

is the angle of incidence. The maximump ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi of the vertex of a cone of rays entering the fiber, known as the acceptance
semi-angle
angle, can be determined by α ¼ sin 1 n21  n22. The parameter max(n0sinθi), therefore, defines the light gathering capacity
of an optical fiber, and it is called the numerical aperture (NA):
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffi
NA ¼ n21  n22  n1 2Δ, Δ1 ð2:114Þ

where Δ is the normalized index difference defined as Δ ¼ (n1  n2)/n1. Therefore, from the geometrical optics point of view,
light propagates in optical fiber due to series of total internal reflections that occur at the core-cladding interface. The smallest
angle of incidence ϕ (see Fig. 2.7(b)) for which the total internal reflection occurs is called the critical angle and equals sin1
n2/n1.

2.2.2.4 Polarizing Elements

With the following three polarizing elements polarizer, phase shifter (retarder), and rotator, any elliptical polarization state can
be obtained [9]. The action of polarizing element on polarization vector (state) given by Eq. (2.69) [Eq. (2.72)] can be
described by Jones operator (matrix):

  " #
J xx J xy ψ x,in
jψ i ¼ J jψ in i ¼ jψ in i, jψ in i ¼ ð2:115Þ
J yx J yy ψ y,in

where |ψini is the input polarization state. Linear polarizer changes the amplitude of polarization vector, and it is characterized
by absorption coefficients along x- and y-axis, denoted as px and py, respectively. The action of linear polarizer can be
described by the following Jones matrix:
" #
px 0
J polarizer ¼ ; 0
pi
1, i 2 fx, yg: ð2:116Þ
0 py

Ideal linear horizontal polarizer is specified with px ¼ 1 ( py ¼ 0), while an ideal vertical polarizer with py ¼ 1 ( px ¼ 0).
The wave plates (phase shifters, retarders) are polarizing elements that introduce the phase shift ϕ/2 along x-axis (the fast
axis) and the phase shift ϕ/2 along y-axis (the slow axis), and corresponding Jones matrix is given by:
" #  
ejϕ=2 0 1 0
J wave plate ¼ , J wave plate ¼ : ð2:117Þ
0 ejϕ=2 0 ejϕ
38 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.8 Illustration of the transformation of the photon polarization basis: (a) the rotation of the state vector, (b) the rotation of a coordinate system

Two popular wave plats are quarter-wave plate (QWP), for which ϕ ¼ π/2, and half-wave plate (HWP), for which ϕ ¼ π.
Their Jones matrix representations are as follows:
   
1 0 1 0
J QWP ¼ , J HWP ¼ : ð2:118Þ
0 j 0 1

For instance, if the input state is linear 45 state, denoted as | 45 i ¼ [1 1]T/√2, after passing though QWP, it gets
converted to:
    
1 1 0 1 1 1
J QWP j 45 i ¼ pffiffiffi ¼ pffiffiffi , ð2:119Þ
2 0 j 1 2 ∓j

in other words, left/right-circular polarization state.

To determine the Jones matrix representation of rotator, let us observe the rotation operation as illustrated in Fig. 2.8. In
Fig. 2.8(a), the vector state |ψi is rotated for θ in a clockwise direction, which is equivalent to the rotation of coordinate system
in opposite (counterclockwise) for the same angle (θ), as shown in Fig. 2.8(b).
The state |ψi can be represented in original {|xi,|yi} basis by:

jψ i ¼ jxihxjψ i þ jyihyjψ i, ð2:120Þ

where hx| ψi (hy| ψi) denotes the projection along x-polarization (y-polarization). By multiplying (2.120) by hx0 | and hy0 | from
the left side, we obtain:

hx0 jψ i ¼ hx0 jxihxjψ i þ hx0 jyihyjψ i, hy0 jψ i ¼ hy0 jxihxjψ i þ hy0 jyihyjψ i, ð2:121Þ

or by rewriting the Eq. (2.121) into matrix form, we obtain:


0
  
jψ i ¼ x0 jψ ihy0 jψ iÞ ¼ x0 jxi x0 jyi y0 jxihy0 jyiÞJ ðθÞ hxjψ ihyjψ iÞ: ð2:122Þ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}

From Fig. 2.8(b) it is clear that old basis {|xi,|yi} is related to the new basis {|x0 i,|y0 i} by:

jxi ¼ cos θjx0 i  sin θjy0 i, jyi ¼ sin θjx0 i þ cos θjy0 i: ð2:123Þ

The corresponding projections are:

hx0 jxi ¼ cos θ, hx0 jyi ¼ sin θ, hy0 jxi ¼  sin θ, hy0 jyi ¼ cos θ, ð2:124Þ

so that the state vector in new basis can be expressed as follows:

2.2 Propagation of Electromagnetic Waves 39

jψ 0 i ¼ J rotation ðθÞjψ i, ð2:125Þ

where Jrotation(θ) is the rotation operator, which can be represented in matrix form, based on (2.122) and (2.124), as follows:
 
cos θ sin θ
J rotation ðθÞ ¼ : ð2:126Þ
 sin θ cos θ

In other words, the Eq. (2.114) can be written as:

* !
0
cos θ sin θ
jψ i ¼ x0 jψ ihy0 jψ iÞ ¼ hxjψ ihyjψ iÞjψ i : ð2:127Þ
 sin θ cos θ |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}

The Jones matrix representation of a rotated polarizing element J, polarizer or wave plate, rotated by θ from the x-axis, is
given by:
J ðθÞ ¼ J rotation ðθÞ  J  J rotation ðθÞ: ð2:128Þ

By interpreting TM and TE polarizations as x- and y-polarizations, respectively, the reflection matrix R and transmission
matrix T are given, respectively, by:
   
r p 0 tp 0
R¼ , T¼ : ð2:129Þ
0 rs 0 ts

For the normal incidence case (θ ¼ ϕ ¼ 0), from Eq. (2.106), we conclude that rs ¼ rp ¼ (1  n)/(1 + n), so that the reflection
matrix is given by:
   
1  n 1 0 n1 1 0
R¼ ¼ , n > 1: ð2:130Þ
1þn 0 1 nþ1 0 1

So if the incident polarization state is |Li, the reflected polarization state is:
     
n1 1 0 1 1 n1 1 1 n1
jψ i ¼ RjLi ¼ pffiffiffi ¼ pffiffiffi ¼ jRi: ð2:131Þ
nþ1 0 1 2 j n þ 1 2 j nþ1
|fflfflfflffl{zfflfflfflffl}
jRi

To describe the action of polarizing elements in Stokes coordinates, the Mueller matrix representation can be used:
2 3 2 32 3
S00 M 00 M 01 M 02 M 03 S0
6 S01 7 6 M 10 M 11 M 12 M 13 76 S1 7
6 7¼6 76 7 , S0 ¼ M  S: ð2:132Þ
4 S02 5 4 M 20 M 21 M 22 M 23 54 S2 5
S0 M 30 M 31 M 32 M 33 S3
|fflffl{zffl
3
ffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflffl{zfflffl}
S0 M S

The Mueller operator (matrix) representation for the linear polarizer is given by:
2 3
p2x þ p2y p2x  p2y 0 0
6 2 7
16 px  p2y p2x þ p2y 0 0 7
M polarizer ¼ 6 7, ð2:133Þ
26
4 0 0 2px py 0 7
5
0 0 0 2px py
40 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

where px and py are the absorption coefficients along x- and y-axes, respectively. The Mueller operator (matrix) representation
for the wave plates is given by:
2 3
1 0 0 0
60 7
6 1 0 0 7
M waveplates ¼ 6 7, ð2:134Þ
40 0 cos ϕ  sin ϕ 5
0 0 sin ϕ cos ϕ

where ϕ is the phase shift between x- and y-polarization states. The Mueller matrix representation of a rotator is given by:
2 3
1 0 0 0
60 cos ð2θÞ sin ð2θÞ 07
6 7
M rotator ¼ 6 7, ð2:135Þ
40  sin ð2θÞ cos ð2θÞ 05
0 0 0 1

where θ is the rotation angle. The rotation of a polarizing element, described by Mueller matrix M, can be described by the
following Mueller matrix:

M ðθÞ = M rotator ð2θÞ  M  M rotator ð2θÞ: ð2:136Þ

2.2.3 Electromagnetic Potentials and Electromagnetic Waves

2.2.3.1 Electromagnetic Potentials

The problem of electromagnetic wave propagation over isotropic, linear, and homogenous media can be simplified when
electrical scalar potential φ and magnetic vector potential A are first determined, which are related to electric field vector E
and magnetic field vector and magnetic flux density vector B as follows [5, 8]:

∂A
E ¼ gradφ  and B ¼ curlA: ð2:137Þ
∂t

Given that curl grad φ ¼ 0 and div curl A ¼ 0, while spatial and time differentiation operators commute, Maxwell’s equations
get satisfied. In isotopic media the constitutive relations are given by D ¼ εH and B ¼ μH so that from Maxwell’s equation
(2.2) we obtain:

∂E
curl B ¼ μJ þ εμ : ð2:138Þ
∂t

By substituting Eqs. (2.137) into (2.138), we obtain:

2
! ∂φ ∂ A
grad div A  Δ A ¼ μJ  εμ grad  εμ 2 , ð2:139Þ
∂t ∂t
!
where Δ A denotes the Laplacian vector operator, defined as:

!
Δ A ¼ grad div A  curl curl A, ð2:140Þ

which in Cartesian coordinates becomes simply:

!
Δ A ¼ ΔAxb
x þ ΔAyby þ ΔAzbz:
2.2 Propagation of Electromagnetic Waves 41

Divergence of vector potential can be arbitrarily chosen, and we chose it as follows:

∂φ
div A ¼ εμ , ð2:141Þ
∂t

so that the gradient terms in Eq. (2.139) disappear, which now becomes the vector wave equation:

2
! ∂ A
Δ A  εμ ¼ μJ: ð2:142Þ
∂t 2

The Eq. (2.141) is often called the Lorentz gauge. By substituting the electrical field from Eq. (2.137) into Maxwell’s equation
(2.3), with the help of (2.141), we obtain the scalar wave equation:

2
∂ φ ρ
Δφ  εμ ¼ : ð2:143Þ
∂t 2 ε

The wave equations for the monochromatic state, when ∂/∂t get replaced by jω, are known as Helmholtz equations.

2.2.3.2 Concept of Antenna

We now study the propagation of electromagnetic waves from the source or transmit antenna, as illustrated in Fig. 2.9. The
volume V is composed of homogeneous material of dielectric constant ε and magnetic permittivityμ. The volume density of
charges within this volume V is denoted by ρ ¼ ρ(r0, t), while the density of conduction currents is denoted by J ¼ J(r0, t),
where r0 denotes position vector of a source volume element dV’. We are interested in determining the electric in magnetic
fields at point P, outside of source, and the position vector of this point is denoted by r. In order to do so, we apply the
superposition principle [5, 8], which claims that the electromagnetic field of a complicated system can be determined as a
summation of electromagnetic fields originating from individual parts. By assuming that the distance from point P to the
source element dV0 is much larger than the dimension of the source, the source element dV0 can be interpreted as the point
source. The charge of the source element is then given by dq ¼ ρdV0 , and the corresponding electric scalar potential at point
P is given by:

ρðr0 , t  R=vÞ 0
dφP ¼ dV , R ¼ jr  r0 j, ð2:144Þ
4πεR

where v is the speed of electromagnetic waves, namely, v ¼ (εμ)1/2. By integrating over volume V, we obtain the electric
scalar potential, originating from all charges within the volume V, as follows:
ð
1 ρðr0 , t  R=vÞ 0
φP ðr, t Þ ¼ dV : ð2:145Þ
4πε V R

It is straightforward to show that the electric scalar potential satisfies the scalar wave equation (2.143). By close inspection of
the scalar wave and vector wave equations, we conclude that the i-th component, i2{x,y,z}, of the magnetic vector potential

Fig. 2.9 Illustration of the source of electromagnetic waves (transmit antenna)

42 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

can be obtained from the scalar wave equation by replacing ε with 1/μ and ρ with Ji. Therefore, the i-th component of the
magnetic vector potential can be obtained as follows:
ð
μ J i ðr0 , t  R=vÞ 0
Ai,P ðr, t Þ ¼ dV , i 2 fx, y, zg ð2:146Þ
4π V R

The corresponding vector representation of the magnetic vector potential is given by:
ð
μ J ðr0 , t  R=vÞ 0
AP ðr, t Þ ¼ dV ð2:147Þ
4π V R

The Eqs. (2.145) and (2.147), when the time-dependence is omitted, are known as the Helmholtz integrals. The potentials
given by Eqs. (2.145) and (2.147) in electrodynamics are known as retarded potentials.
Clearly, the retarded potentials for electromagnetic field are generated by time-varying charge density or electric current.
As the electromagnetic field propagates at finite speed, the speed of light in a given medium, denoted as v, the retarded time at
a given point in space r is given by tr ¼ t  |r  r'|/v ¼ t  R/v..
By substitution (2.145) and (2.147) into (2.135), with the help of constitutive relationship H = B/μ, we obtain the following
expressions for electric and magnetic field vectors:
ð ð
1 b 0
∂ρðr0 , t  R=vÞ R 1 Rb
EP ðr, t Þ ¼ dV þ ρðr0 , t  R=vÞ 2 dV 0
4πεv V ∂t R 4πε V R
ð 0
ð2:148Þ
μ ∂J ðr , t  R=vÞ 1 0
 dV ,
4π V ∂t R

and
ð ð
1 b
∂J ðr0 , t  R=vÞ R 1 b
R
H P ðr, t Þ ¼
dV 0 þ Jðr0 , t  R=vÞ
dV 0 , ð2:149Þ
4πv V ∂t R 4π V R2

b ¼ R=R.
where R
When the volume density charge is sinusoidal:
n o
ρðr0 , t  R=vÞ ¼ ρm ðr0 Þ cos ½ωt  kR þ θðr0 Þ ¼ Re ρðr0 Þe jðωtkRÞ
,
0
ð2:150Þ
ρðr0 Þ ¼ ρm ðr0 Þejθðr Þ

the electrical scalar potential in complex form can be represented as:

ð
1 ρðr0 ÞejkR 0
φP ðrÞ ¼ dV : ð2:151Þ
4πε V R
When the current density is sinusoidal:
n o
J ðr0 , t  R=vÞ ¼ J m ðr0 Þ cos ½ωt  kR þ θðr0 Þ ¼ Re Jðr0 Þe jðωtkRÞ
,
0
ð2:152Þ
J ðr0 Þ ¼ Jm ðr0 Þejθðr Þ

the magnetic vector potential in complex form can be represented as:

ð
μ Jðr0 ÞejkR 0
A P ðrÞ ¼ dV ð2:153Þ
4π V R
2.2 Propagation of Electromagnetic Waves 43

2.2.3.3 Spherical Waves

In points located far away from the source, r >> r0 , in the calculation of the electric scalar and magnetic vector potentials, the
following two approximations can be made:

(1) The coarse approximation in amplitude:

1
R1 ¼ jr  r0 j ffi r 1 , ð2:154Þ

(2) The fine approximation in phase:

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi r>>r0 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

kR ¼ kjr  r0 j ¼ k r 2  2rr0 þ r0 2 ffi k r 2  2rr0
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð2:155Þ
2rr0 0
¼ kr 1  2 ffi kr  kbrr ,br ¼ r=r:
r

Using these two approximations, the magnetic vector potential expressions (2.153) become:
ð
μ 0 0
ejkr μ ejkr ejkr
AP ffi Jðr Þe jk^rr dV 0  ¼ N ¼ A0 , ð2:156Þ
4π V r 4π
|ffl{zffl} r r
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
N A0

where the vector N in (2.156) is known as the radiation vector. The radial dependence of the magnetic vector potential is of
the form:
ejkr
Φ¼ : ð2:157Þ
r

Such electromagnetic waves, whose amplitude changes with distance as 1/r, while the phase is a linear function of distance,
are commonly referred to as the spherical electromagnetic waves. At distances sufficiently far away for the source (or transmit
antenna), regardless of the shape or size, the electromagnetic waves are spherical. This space region is called the far-field
region or Fraunhofer region. The propagation function of spherical waves satisfies the following radiation condition:
∂Φ 1 þ jkr
¼ Φ: ð2:158Þ
∂r r

In spherical coordinate system (r,θ,ϕ), because of 1/r dependence of radial component, in the far-field, only angle
components of magnetic vector potential exist and are given by:

ejkr b ejkr b
Aθ ¼ A θ, Aθ ¼ A ϕ: ð2:159Þ
r r

By using the approximations (2.154) and (2.155), the electrical scalar potential (2.151) becomes:

ð
1 ρðr0 ÞejkR 0
φP ðrÞ ¼ dV : ð2:160Þ
4πε V R

Given that the action of the time derivative on the harmonic wave is multiplication by jω, from Eq. (2.141), we obtain:

j
φP ¼ div A, ð2:161Þ
ωεμ
44 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

and by substituting (2.156) into (2.161), the scalar potential becomes:

ð
1 μ 0 1 þ jkr ejkr jkbrr0 0
φP ¼ j Jðr Þbr e dV
ωεμ 4π V r r
|fflfflffl{zfflfflffl}
Z=4πk ð2:162Þ
ð rffiffiffi
jkr
Z 0 1 þ jkr e jkb
rr
0
0 μ
¼ j Jðr Þbr e dV , Z ¼ :
4πk V r r ε

The electric field vector can be now determined as:

ð jkr
ωμ 0 e
e jkbrr dV 0
0
E ¼ grad φ  jωA ¼ j Jðr Þ
4π V r
ð   ð2:163Þ
Z 1 þ jkr 0 0 br ejkr jkbrr0 0
þj Jðr Þ þ Jðr Þb
r ðk r
2 2
 3  jkrÞ e dV
4πk V r2 r2 r

The magnetic field vector can be obtained as:

ð h i
1 1 0 1 þ jkr ejkr jkbrr0 0
H ¼ curl A ¼ Jðr Þ
br e dV : ð2:164Þ
μ 4π V r r

In far-field, the components in (2.163) and (2.164), the components whose amplitudes decay as 1/r, dominate so that the
electric and magnetic field vectors can be written as:
ð nh i o
Zk ejkr
Jðr Þbr br  Jðr Þ e jkbrr dV 0
0 0 0
E¼j ð2:165Þ
4π r V

The magnetic field vector can be obtained as:

ð h i
1 1 0 1 þ jkr ejkr jkbrr0 0
H ¼ curl A ¼ Jðr Þ
br e dV : ð2:166Þ
μ 4π V r r

Similarly to magnetic vector potential, because of 1/r dependence of radial component, in the far-field only angle components
of electric and magnetic fields exist and are given, respectively, by:

E θ ¼ jωAθ , H θ ¼ E θ =Z,
ð2:167Þ
E ϕ ¼ jωAϕ , H ϕ ¼ Eϕ =Z:

It can be straightforwardly shown that the electric and magnetic field vectors are mutually perpendicular and simultaneously
perpendicular to the radial direction:

E  H ¼ 0, E  br ¼ 0, H  br ¼ 0: ð2:168Þ

The Poynting vector has only the radial component, in the far-field, as shown below:

E θ Eθ þ Eϕ E ϕ  
ω2
S ¼ E
H  ¼ Sbr, S ¼ E θ H ϕ þ E ϕ H θ ¼ ¼ 2 Aθ Aθ þ Aϕ Aϕ ð2:169Þ
Z Zr

The flux of the Poynting vector over a sphere of radius r, containing the transmit antenna/source, represents the radiated
power:

ðπ 2ðπ
 
ω2
P¼ Aθ Aθ þ Aϕ Aϕ dθdϕ: ð2:170Þ
Z
θ¼0 ϕ¼0
2.2 Propagation of Electromagnetic Waves 45

2.2.3.4 Plane Waves

When the spherical wave are observed within the small region of dimensions much smaller than the distance from the transmit
antenna/source, the variations in amplitude can be neglected. Such electromagnetic waves, whose amplitudes do not change,
while the phases are linear function of distance, are commonly referred to as the plane waves. The propagation function of the
plane waves is given by:

Φ ¼ ejkr , ð2:171Þ

and satisfies the following radiation condition:

∂Φ
¼ jkΦ: ð2:172Þ
∂r

Strictly speaking, the plane waves do not exist in nature, as the antenna to emit them cannot be manufactured. Namely, such
antenna would be composed of infinitely large plane with the surface current density Js and emission frequencyω. By
expressing the current density of such “antenna” as:

b
J ¼ J S δðxÞb
x, ð2:173Þ

the magnetic vector potential will have only x-component, only z-dependent, so that the wave equation is given by:

d2 A
þ k 2 A ¼ μJ S , ð2:174Þ
dz2

The corresponding solution is given as:

jμJ S jkjzj
A¼ e : ð2:175Þ
2k

The electric scalar potential is, based on Eq. (2.161), φ ¼ 0. The components of electric field are then obtained as follows:

ωμJ S jkjzj
E x ¼ jωA ¼  e , E y ¼ Ez ¼ 0: ð2:176Þ
2k

On the other hand, the components of magnetic field are obtained as:
8
> J
<  S ejkz , z > 0
Hz ¼ 2 , H x ¼ H y ¼ 0: ð2:177Þ
: J S ejkz , z < 0
>
2

Cleary, the nonzero electric and magnetic fields components (2.176), (2.177) have constant complex amplitudes, while the
phases are a linear function of the distance z, indicating that they describe the plane wave. Even though that plane waves
strictly speaking do not exist, they represent a nice approximation in the far-field, when the observation volume is small
compared to the distance, and as such they have been widely used.

2.2.3.5 Hertzian Dipole

Let us observe the current element of momentum Iδlbz , representing the straight-wire antenna with δl ! 0, located in the
origin (of the coordinate system) as illustrated in Fig. 2.10, which is also known as the Hertzian dipole. The magnetic vector
potential has only z-component:

μ ejkr
A ¼ Abz, A ¼ Iδl , ð2:178Þ
4π r
46 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.10 Illustration of the Hertzian dipole

and the spherical components are given by:

Ar ¼ Abzbr = A cos θ, Aθ ¼ Abzb

θ = ‐A sin θ, Aϕ ¼ 0: ð2:179Þ

The components of electric and magnetic fields in the far-field can now be determined as:

Z ejkr
Eθ ¼ jωAθ = j Ikδl sin θ ¼ ZH ϕ ,
4π r ð2:180Þ
E r ¼ 0, Eϕ ¼ ZH θ ¼ 0, H r ¼ 0:

The electric field is angle θ dependent and exhibits emission directivity properties. The minimum radiation is for θ ¼ 0, π,
while the maximum emission is achieved for θ ¼ π/2. The following function:
kδl
F ðθ Þ ¼ sin θ ð2:181Þ
2

can be used to describe the directivity properties of this transmit antenna, and it is commonly referred to as the radiation
pattern. Since the radiation pattern is independent of ϕ, we say that this antenna is omnidirectional. The θ-component of
electric field can now be expressed in terms of radiation function as follows:

ZI ejkr
Eθ ¼ j F ðθÞ: ð2:182Þ
2π r

The very often, the normalized radiation pattern is used:

F ðθ Þ F ðθ Þ
f ðθ Þ ¼ ¼ : ð2:183Þ
max F ðθÞ F ðθ ¼ π=2Þ

The beamwidth of the transmit antenna is defined as the width of the main lobe (the lobe containing maximum power
direction), usually specified by the half power beam width. In other words, the angle encompassed between the points on the
side of the main lobe where the power has fallen to half (3 dB) of its maximum value:

1
pffiffiffi
f ðθÞ
1: ð2:184Þ
2

For Hertzian dipole, the antenna beamwidth is π/2.

2.2 Propagation of Electromagnetic Waves 47

Fig. 2.11 Center-driven straight-wire antenna

2.2.3.6 Straight-Wire Antenna

Let us now consider the straight-wire antenna, driven from the center, extending from z ¼ l to z ¼ +l, as shown in Fig. 2.11.
The current density is given by:

J ¼ δðxÞδðyÞf ðzÞbz, ð2:185Þ

while the corresponding current is integral over cross section of this wire:
ð
I ðzÞ = Jdxdy ¼ f ðzÞbz: ð2:186Þ

Since the antenna is center-drive, the unknown function f(z) must satisfy the following conditions:

f ðlÞ ¼ f ðlÞ ¼ 0: ð2:187Þ

Additionally, the unknown function f(z) must satisfy one-dimensional Helmholtz equation:

d 2 f ðzÞ
þ k2 f ðzÞ ¼ 0: ð2:188Þ
dz2

There exist two solutions for this differential equation, ejkz and ejkz. The corresponding current of upper and lower segments
can be written as:

I 1 ðzÞ ¼ A1 ejkz þ B1 ejkz ,  l

z
0;
ð2:189Þ
I 2 ðzÞ ¼ A2 ejkz þ B2 ejkz , 0
z
l:

By using the continuity condition I1(0) ¼ I2(0) and Eq. (2.187), it can be shown that the overall solution is the standing wave:

I ðzÞ ¼ I 0 sin k ðl  jzjÞ: ð2:190Þ

The magnetic vector potential is given by Eq. (2.156):

48 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

ð ðl
jkr
μ μ e ejkr
Jðr Þe jkbrr dV 0 
0 0
jkz0 cos θ
A¼ I0 e ¼ sin kðl  jzjÞdz0  bz
4πV r 4π l r
ð2:191Þ
μ cos ðklcos θÞ  cos kl ejkr
¼  2I 0 bz:
4π ksin2 θ r

The nonzero components of electric and magnetic field in the far-field are given by:

zb
Aθ ¼Ab θ = A sin θ
z}|{ μ cos ðkl cos θÞ  cos kl ejkr
E θ ¼ jωAθ ¼ jω sin θ
2I
4π 0 k sin 2 θ r
jkr cos ðkl cos θ Þ  cos kl jkr ð2:192Þ
Z e Z e
¼ j I 0 ¼ j I 0 F ðθÞ,
2π r sin θ 2π r
H ϕ ¼ E θ =Z,

where the radiation pattern is given by:

cos ðkl cos θÞ  cos kl

F ðθ Þ ¼ : ð2:193Þ
sin θ

Two illustrative examples of radiation patterns are shown in Fig. 2.12. For kl > π in addition to the main lobe, two side lobes
appear. The radial component of Poynting vector is given by:

 
1 Z I2
Sr ¼ E θ H ϕ þ E ϕ H θ ¼ 2 02 F 2 ðθÞ: ð2:194Þ
2 8π r

An important parameter of the antenna is so-called directivity gain, denoted by gd, defined as follows [8]:

4π 2 Sr, max
gd ¼ ð 4π , ð2:195Þ
Sr ðr, θ, ϕÞr 2 sin θdθdφ
0

Fig. 2.12 Radiation patterns for center-driven straight-wire antenna: (a) kl ¼ π and (b) kl ¼ 7π/6
2.2 Propagation of Electromagnetic Waves 49

where Sr, max is the major lobe maximum of the Poynting vector in the far-field and r is the far-field sphere radius. Clearly,
the directivity gain of an isotropic antenna is gd ¼ 1, while the of directive antennas is larger than 1. The denominator in
(2.195) represents the time-average power radiated by the antenna:
ð 4π
P¼ Sr ðr, θ, ϕÞr 2 sin θdθdφ: ð2:196Þ
0

The directivity gain for the center-driven straight-wire antenna from Fig. 2.11 can be calculated by:
h i2
4π 2 max cos ðkl cos θÞ cos kl
sin θ
θ
gd ¼ ðπ : ð2:197Þ
cos kl2
2π ½ cos ðkl cossinθÞ
θ dθ
0

For the half-wave dipole (kl ¼ π/2), it is straightforward to show that directivity gain is gd ¼ 1.64.

2.2.3.7 Cylindrical Waves

Let us now consider the current element whose current density is given as:

I
J¼ δðr Þz, ð2:198Þ
2rπ

the magnetic vector potential will only have the axial component, namely, A ¼ Abz: The corresponding wave equation is
given by:
 
∂ ∂A μI
r þ k 2 rA ¼  δðr Þ: ð2:199Þ
∂r ∂r 2π

The solutions of this wave equation are Bessel functions of the first and second kind, J0(kr) and Y0(kr). In the absence of any
obstacle, only the incident wave will propagate; therefore, the Hankel functions of the second kind will be used as a solution of
the wave equation [5]:

ð2Þ
H 0 ðkr Þ ¼ J 0 ðkr Þ  jY 0 ðkr Þ: ð2:200Þ

The corresponding solution for the magnetic vector potential becomes:

μI ð2Þ μI ð2Þ
A¼ ðjπ ÞH 0 ðkr Þ ¼ j H 0 ðkr Þ: ð2:201Þ
4π 4

As a generalization, for sufficiently high frequencies, the solid conducting cylinder of sufficient length can be used as a more
realistic transmit antenna. Thanks to the strong skin effect, only the surface current of density JS will exist, and the
corresponding magnetic vector potential can be written as:
þ
μ ð2Þ
A ¼ j J ðr0 ÞH 0 ðkjr  r0 jÞdl0 , ð2:202Þ
4 S
C

where C is the contour defining the cross section of the conducting cylinder. In the far-field, r> > r0 , so that the following
approximation of the Hankel function can be used:
rffiffiffiffiffiffiffi
ð2Þ kr!1 2 jðkrπ=4Þ
H 0 ðkr Þ ! e : ð2:203Þ
πkr
50 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

After substituting the Eq. (2.203) into (2.202), we obtain the following expression for the magnetic vector potential in the
far-field:
rffiffiffiffiffi þ
μ 2 ejkr
JS ðr Þe jkbrr dl0 :
0 0
A ¼ j ejπ=4 pffiffi ð2:204Þ
4 πk r
|ffl{zffl} C
Φ

Clearly, the propagation function Φ has the following form:

ejkr
Φ ¼ pffiffi , ð2:206Þ
r

and satisfies the following radiation condition:

 
∂Φ 1
¼ jþ Φ: ð2:207Þ
∂r 2kr

The electric field will only have axial component Ez and magnetic field only angle Hθ component, given, respectively, as:

E z ¼ jωA, H θ ¼ E=Z: ð2:208Þ

The electromagnetic waves, whose propagation function has the form given by Eq. (2.206), are commonly referred to as the
cylindrical waves.

2.2.3.8 Paraxial Approximation and Gaussian Beam

Before concluding this section, we briefly return back to the spherical wave, whose propagation function is given by (2.157).
Based on Fig. 2.9, we conclude that the position vector magnitude of observed point P(x,y,z) with respect to the origin O(x0,y0,
z0) can be written as:

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
  y  y 2ffi
2 2 2 x  x0 2
r ¼ ðx  x0 Þ þ ðy  y0 Þ þ ðz  z0 Þ ¼ Δz 1 þ þ 0
,
Δz Δz ð2:209Þ
Δz ¼ z  z0 :

In paraxial approximation, when the region of interest is the space close to the optical axis, assuming that the propagation
pffiffiffiffiffiffiffiffiffiffiffi
direction is aligned with the positive z-direction, the following approximation for square root can be used 1 þ u ffi 1 þ u=2,
and from previous equation, we obtain:
     
1 x  x0 2 1 y  y0 2
r ffi Δz 1 þ þ : ð2:210Þ
2 Δz 2 Δz

Corresponding paraxial spherical wave approximation becomes:

xx 2 yy 2 
ejkΔz k
j2Δz ð Δz0 Þ þð Δz0 Þ :
Φffi e ð2:211Þ
Δz

Another relevant electromagnetic wave, in particular in optics, is the Gaussian beam wave, which has a Gaussian
amplitude profile, but the spherical wavefront and corresponding propagation function can be written as:

1 jkx2q2 þyðzÞ2
ΦG ¼ e , ð2:212Þ
qðzÞ

where the complex beam parameter q(z) is defined as:

2.2 Propagation of Electromagnetic Waves 51

πw20
qðzÞ ¼ z  jzR , zR ¼ ð2:213Þ
λ

and zR is the so-called Rayleigh distance (range), the distance at which the waist w is √2 time larger than that at the beam waist
w0 (at z ¼ 0). Another popular representation for q(z) is given in term of reciprocal of q(z) as follows:

1 1 z z 1 1
¼ ¼ þj 2 R 2 ¼    þj "  2 #
qðzÞ z  jzR z2 þ z2R z þ zR z 2
z
z 1þ R zR 1 þ
z z
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} R
RðzÞ
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
w2 ðzÞ=w20
ð2:214Þ
1 1 1
¼ þj
RðzÞ zR w2 ðzÞ=w20
 "  2 #
 2 
zR z
RðzÞ ¼ z 1 þ , w ðzÞ ¼ w0 1 þ
2 2
z zR

where R(z) denotes the radius of the curvature of the beam’s wavefronts at distance z and w(z) denotes the radius at which the
field amplitude falls to 1/e of their axial value, at the plane z along the beam.

2.2.4 Interference, Coherence, and Diffraction in Optics

2.2.4.1 Interference
According to the superposition principle, the electric field vector E at the observed point of space r can be obtained as the
vector sum of individual field vectors Ei originating from different sources:

X
N
E¼ EðiÞ : ð2:215Þ
i¼1

If the density of conduction currents J(r0 ,t) from Fig. 2.9 can be represented as the sum of component conduction currents:

0
X
N
0
Jðr , tÞ ¼ J ðiÞ ðr , tÞ, ð2:216Þ
i¼1

wherein each of the satisfies the vector wave equation:

2
! ∂ AðiÞ
ΔAðiÞ  εμ ¼ μJ ðiÞ , ð2:217Þ
∂t 2

then the overall magnetic vector potential can be represented, based on (2.147), as follows:

P
N 0
ð J ðiÞ ðr , t  R=vÞ XN μ ð J ðiÞ ðr0 , t  R=vÞ
μ i¼1 0
Aðr, tÞ ¼ dV ¼ dV 0 ,
4π V R i¼1 4π
V R ð2:219Þ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
AðiÞ
PN
¼ i¼1 AðiÞ ,

and clearly satisfies the overall vector wave equation (2.142). Let the charge density of the volume Vi in which the i-th
conduction current is located be denoted as ρi. Based on Eq. (2.145), the overall scalar potential can be represented as:
52 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

P
N 0
ð ρðiÞ ðr , t  R=vÞ XN ð 0
1 i¼1 0 1 ρðiÞ ðr , t  R=vÞ 0
φðr, tÞ ¼ dV ¼ dV
4πε V R 4πε V
i¼1 |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
R ð2:220Þ
ffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
φðiÞ ðr, tÞ
PN
¼ i¼1 φðiÞ ðr, tÞ:

By substituting (2.219) and (2.220) into (2.137), we prove the superposition principle given by Eq. (2.215).
As an illustrative example, let us observe two plane linearly polarized waves of the same angular frequency:

jðωtk1 rþϕ1 Þ jðωtk2 rþϕ2 Þ

Eð1Þ ¼ E1 e , Eð2Þ ¼ E2 e : ð2:221Þ

When the phase difference ϕ1  ϕ2 is constant, we say that two sources are mutually coherent. As we have seen in Sect. 2.2.1,
the irradiance function at the given point r is proportional to the magnitude square of the electric field, so that after applying
the superposition principle given by Eq. (2.215) for N ¼ 2, the irradiance function, after ignoring the constant term, is obtained
as follows:

 
I ¼ jEj2 ¼ E  E ¼ Eð1Þ þ Eð2Þ Eð1Þ þ Eð2Þ
 2  2
¼ Eð1Þ  þ Eð2Þ  þ Eð1Þ Eð2Þ þ Eð1Þ Eð2Þ ð2:222Þ
|fflfflffl{zfflfflffl} |fflfflffl{zfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
I1 I2 2E1 E2 cos ðk2 rk1 rþϕ1 ϕ2 Þ

¼ I 1 þ I 2 þ 2E1  E2 cos ðk2  r  k1  r þ ϕ1  ϕ2 Þ

The last term in the bottom line of (2.222) is commonly referred to as the interference term. Since the angle k2  r  k1  r + ϕ1  ϕ2
is the position vector r dependent, the periodic spatial variations in the intensity occur, and these variations are responsible for
the interference fringes when two mutually coherent light beams are combined. On the other hand, when two mutually
incoherent light beams are combined, the interference term is averaged out, and the interference effect does not occur.
Even when two mutually coherent beams are monochromatic and with constant amplitudes, if they propagate different
paths before the interference occur, the instantaneous light flux at a given point r in space fluctuates, and we need to define the
irradiance by using the time-average operation:
D E D
 E
I ¼ jEj2 ¼ hE  E i ¼ Eð1Þ þ Eð2Þ Eð1Þ þ Eð2Þ
D 2 E D  2 E D E
¼ Eð1Þ  þ Eð2Þ  þ Eð1Þ Eð2Þ þ Eð1Þ Eð2Þ
|fflfflfflfflffl{zfflfflfflfflffl} |fflfflfflfflffl{zfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
D E
I1 I2 ð2:223Þ
2 Re Eð1Þ Eð2Þ
 
¼ I 1 þ I 2 þ 2 Re E 1 ðt ÞE 2 ðt þ τÞ ¼ I 1 þ I 2 þ 2 Re Γ12 ðτÞ,
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
Γ12 ðτÞ

2.2.4.2 Coherence
The cross-correlation function:
 
C 12 ðτÞ ¼ Γ12 ðτÞ ¼ E1 ðt ÞE 2 ðt þ τÞ ð2:224Þ

in optics is also known as the mutual coherence function (MCF). On the other hand, the following autocorrelation function:
 
C11 ðτÞ ¼ Γ11 ðτÞ ¼ E1 ðt ÞE1 ðt þ τÞ , ð2:225Þ

in optics is known as the self-coherence function. Clearly, we assumed that the random processes E1(t) and E2(t) are wide-
sense stationary, meaning that the mean values are constant over the time and that autocorrelation functions are solely
2.2 Propagation of Electromagnetic Waves 53

dependent of the time difference at which the process is sampled. The sharp brackets in (2.223, 2.224 and, 2.225) are used to
denote the time-average function:

ð T=2
1
h f i ¼ lim f ðt Þdt: ð2:226Þ
T!1 T T=2

The normalized cross-correlation function in optics is known as the degree of partial coherence:

Γ12 ðτÞ Γ ðτ Þ
γ 12 ðτÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ p12ffiffiffiffiffiffiffiffi : ð2:227Þ
Γ11 ð0ÞΓ22 ð0Þ I1I2
pffiffiffiffiffiffiffiffi
Based on (2.227), the interference term in (2.223) can be expressed as 2 Re Γ12 ðτÞ ¼ 2 I 1 I 2 γ 12 ðτÞ, so that the irradiance
equation can be rewritten as:
pffiffiffiffiffiffiffiffi
I ¼ I 1 þ I 2 þ 2 I 1 I 2 γ 12 ðτÞ: ð2:228Þ

The magnitude of the degree of partial coherence can be used to clarify different types of coherence into three types:
(i) complete coherence when |γ 12| ¼ 1, (ii) complete incoherence when |γ 12| ¼ 0, and (iii) partial coherence, otherwise.
Another relevant parameter in optics is the fringe visibility, defined as:

I max  I min
V ¼ : ð2:229Þ
I max þ I min

and based on irradiance Eq. (2.228), the fringe visibility can be expressed as:
pffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffi
I 1 þ I 2 þ 2 I 1 I 2 jγ 12 j  I 1 þ I 2  2 I 1 I 2 jγ 12 j 2 I 1 I 2 jγ 12 j
V ¼ pffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffi ¼ : ð2:230Þ
I 1 þ I 2 þ 2 I 1 I 2 jγ 12 j þ I 1 þ I 2  2 I 1 I 2 jγ 12 j I1 þ I2

In particular, when I1 ¼ I2 ¼ I:
pffiffiffiffi
2 I 2 jγ 12 j
V ¼ ¼ jγ 12 j, ð2:231Þ
2I

the fringe visibility is equal to the magnitude of the degree of partial coherence. For the complete coherence ( |γ 12| ¼ 1), the
fringe visibility will have the maximum contrast of unity. On the other hand, for the compete incoherence ( |γ 12| ¼ 0), there are
no interference fringes at all as the contrast is zero.
Let us now observe the quasimonochromatic field:

j½ωtþϕðt Þ
E ðt Þ ¼ E 0 e , ð2:232Þ
where the phase is constant for the duration of time interval τc and changes rapidly for τ > τc. For simplicity, let us assume that
the phase changes uniformly every τc seconds. The degree of partial coherence in this case is just the correlation coefficient:
ð
hE ðt ÞE ðt Þi D jωτ j½ϕðtÞϕðtþτÞ E jωτ 1 T j½ϕðtÞϕðtþτÞ
γ ðτ Þ ¼ D E ¼ e e ¼e lim e dt: ð2:233aÞ
T!1 T 0
jE j2

For each time interval of duration τc, the phase difference ϕ(t)  ϕ(t + τ) is nonzero for t between τc  τ and τc so that:
ð τc τ  ð τc τ   
1 jωτ 1 τ
γ ðτ Þ ¼ e e j0 dt þ ejωτ E e jΔϕ dt ¼ ejωτ 1  , τ < τc ð2:233bÞ
τc τc 0 τc
0
|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
¼0
54 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

where we used the statistical average operator E{} to perform the statistical averaging with respect to the random phase Δϕ.
Clearly, we assumed that the process is ergodic so that the statistical and time-averaging operators can be interchangeably
used. Therefore, the fringe visibility is known:
( τ
1 , τ < τc
V ¼ jγ 12 j ¼ τc ð2:234Þ
0, τ τc

Evidently, the fringe visibility drops to zero when the time difference between two observations τ is larger than the
characteristic time interval τc, which is commonly referred to as the coherence time. Equivalently, for the nonzero fringe
visibility, the path difference between two beams must be small than:
lc ¼ cτc , ð2:235Þ

and we refer to the path difference lc as the coherence length.

We are concerned now with the coherence between two fields at different points in space, in other words with spatial
coherence. Let us observe quasimonochromatic point source S. If the two observation points lie in the same direction from the
source, the corresponding spatial coherence is known as the longitudinal spatial coherence of the field. As long as the distance
between these two points is smaller than lc, the coherence between the fields is high. On the other hand, if the two observation
points P1 and P2 are located at the same distance from the source S, but in different directions, then the normalized MCF will
measure the so-called transverse spatial coherence of the field. The partial coherence between the observation points P1 and P2
occurs when the source S is extended source. If the extended source is composed of two mutually incoherent point sources S1
and S2, as illustrated in Fig. 2.13, the corresponding electric fields in observation points P1 and P2 are given as:

E 1 ¼ E P1 S1 þ E P1 S2 , E2 ¼ E P2 S1 þ E P2 S2 : ð2:236Þ

The normalized cross-correlation function is given by:

D h iE
 
E 1 ðt ÞE2 ðt þ τÞ ½E P1 S1 þ E P1 S2  EP2 S1 þ EP2 S2
γ 12 ðτÞ ¼ pffiffiffiffiffiffiffiffi ¼ pffiffiffiffiffiffiffiffi
I1I2 I 1I 2
D E D E D E D E
 
E P1 S1 E P2 S1 E P1 S1 E P2 S2 EP1 S2 EP2 S1 E P1 S2 EP2 S2
¼ pffiffiffiffiffiffiffiffi þ pffiffiffiffiffiffiffiffi þ pffiffiffiffiffiffiffiffi þ pffiffiffiffiffiffiffiffi , ð2:237Þ
I1I2 I 1I 2 I 1I 2 I 1I 2
|fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl}
γ ðτS1 Þ=2 ¼0 ¼0 γ ðτS2 Þ=2
r P2 S1  r P1 S1 r P2 S2  r P1 S2
τS1 ¼ τ þ , τS2 ¼ τ þ
c c

where we used the fact that point sources are mutually incoherent, while γ(τ) is given by Eq. (2.233). The fringe visibility
between the observation points is related to [10]:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
  ffi
1 þ cos ½ωðτS2  τS1 Þ τ S1 τ S2
jγ 12 ðτÞj  1 1 ð2:238Þ
2 τc τc

and it has the periodic spatial dependence. The mutual coherence drops to zero, at either side of the central line, when:
h i
c r P2 S2  r P1 S2 r P2 S1  r P1 S1
ωðτS2  τS1 Þ ¼ 2π  ¼ π: ð2:239Þ
λ c c

Let s denote the distance between point sources, l denote the distance between observation points, and r denote the mean
distance between the sources and observation points.
2.2 Propagation of Electromagnetic Waves 55

Fig. 2.13 Illustration of the lateral coherence from two mutually incoherent sources, assuming r >> s, l

This case is commonly referIllustration of the lateral coherence from two mutually incoherent sources, assuming r> > s, l.
From Fig. 2.13, we see that r P1 S2  r P1 S1 , so that [10]:

r
lt ffi λ: ð2:240Þ
s

The angle separation of the point sources as seen from the observation point P1 is given by tanθs ’ θs ’ s/r so that we can
write:

lt ffi λ=θs : ð2:241Þ

Since lt represents the separation of receiving points for which mutual coherence is relevant, it can be called the transverse
coherence width. When the source is extended circular, we have to modify the previous equation as follows [10]:

lt ffi 1:22λ=θs : ð2:242Þ

2.2.4.3 Diffraction Theory

The divergence theorem [5, 6], also known as Gauss’s theorem or Ostrogradsky’s theorem, represents a relevant theorem in
vector calculus, which relates a flux of vector field through a surface to the behavior of the vector field in the volume inside of
that surface. Let F be a differentiable vector field defined on the neighborhood of the volume V bounded by the closed surface
S. The divergence theorem can be formulated as follows: the flux of the vector field F over the closed surface S is equal to the
divergence of the vector filed F inside the volume V . Mathematical representation of this theorem is given by:
þ ð
FdS ¼ div F dV , dS ¼ b
ndS, ð2:243Þ
S V

where b
n is the outward surface normal. By setting F = U ∇ V  V ∇ U, previous equation can be rewritten:
þ ð
ðU∇V  V∇U Þb ndS ¼ ∇ðU∇V  V∇U Þ dV , ð2:244Þ
|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
S U grad VV grad U V U∇ VV∇ U
2 2
bn bn

and by multiplying the both sides of equation with 1, we obtain:

þ  ð

V gradbn U  U gradbn V dS ¼ V∇2 U  U∇2 V dV , ð2:245Þ

S V

where we used gradbn U to denote the normal component of the gradient at the surface S. When both U and V are the wave
functions, they both satisfy the corresponding wave equations:
56 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

2 2
1 ∂ U 1 ∂ V
∇2 U ¼ , ∇2 V ¼ 2 : ð2:246Þ
v2 ∂t 2 v ∂t 2

By substituting the left sides of the wave equations into Eq. (2.245) and assuming that the wave functions contain harmonic
time dependence of the form exp.( jωt), we conclude that the right side of Eq. (2.245) becomes zero so that the Eq. (2.245)
reduces to:
þ 
V gradbn U  U gradbn V dS ¼ 0: ð2:247Þ
S

Let us now suppose that wave function V is spherical and contains the harmonic time dependence as follows:

jðωtþkr Þ
e
V ¼ V0 : ð2:248Þ
r

We are interested in determining the wave function U at point P, where r ¼ 0. For r ¼ 0, the V becomes infinitely large, and in
integration given by Eq. (2.247), we have to exclude this point from the integration. This can be achieved by subtracting an
integral over an infinitesimal sphere of radius ρ ! 0, with the point P being in its center, as follows:

þ jkr
!
e jkr e
gradbn U  U gradbn r dS
r
S
2 3
ð2:249Þ
þ e jkr
∂ ∂ e jkr
6 7
lim 6 r ∂r U  U ∂ðrÞ r 7 r 2 dΩ ¼ 0,
ρ!0 4 |fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}5
S:r¼ρ jkr
ð jk1r Þe r r¼ρ

where dΩ ¼ sinθdrdθdϕ represents the solid angle in the spherical coordinates (r,θ,ϕ) on the sphere of radius ρ centered at
P and r2 dΩ is the corresponding area. On the interior infinitesimal sphere, enclosing the point P, the normal vector points
inward, so that gradbn ¼ ∂=∂r: In the limit U in the second term becomes UP, while the second integral is simple
Þ
U p ejk0 ρ12 ρ2 dΩ ¼ 4πU P , and from Eq. (2.249), we derive the following solution for UP:
þ 
1 ejkr ejkr
UP ¼  U gradbn  gradbn U dS, ð2:250Þ
4π r r
S

and this equation is commonly referred to as the Kirchhoff integral theorem.

Now we apply this equation to the problem related to the diffraction of light. The diffraction of light represents the
deflection of light on sharp edges, narrow slits, and small apertures. Our problem at hands is to determine the optical
disturbance at the point of interest P, originating from the source S, as illustrated in Fig. 2.14. The position of the point in the
aperture observed from the source is denoted as r0 , while the position of the same point in the aperture from the point P is
denoted as r.
To simplify the problem, we employ the Saint-Venant’s hypothesis claiming that (i) U in the aperture is the same as if there
was no aperture at all and (ii) U ¼ 0 on the screen and at very large distances (outside of the shaded space). The wave function
of the point in the aperture at the position r0 is simply:

jðkr0 ωt Þ
e
U ¼ U0 : ð2:251Þ
r0

By substituting this wave function into Kirchhoff integral, we obtain:

þ  jkr 0 
1 e ejkr0 ejkr ejkr
U P ¼ U 0 ejωt gradbn 0  0 gradbn dS, ð2:252Þ
4π r r r r
S
2.2 Propagation of Electromagnetic Waves 57

Fig. 2.14 Illustration of the of the diffraction problem on the narrow slit

where the integration is performed over the surface of aperture opening. Given that gradbn denotes the normal component of
the gradient at the aperture opening, the corresponding gradients are evaluated as:

0
jkr0
 jkr0 0
ejkr 0 ∂ e 0 jke ejkr
gradbn 0 ¼ cos ∢ðb n, r Þ 0 0 ¼ cos ∡ðb n, r Þ  2 ,
r ∂r r r0 r0
jkr jkr
 jkr jkr
 ð2:253Þ
e ∂e jke e
gradbn ¼ cos ∡ðb
n , rÞ ¼ cos ∡ðb
n , rÞ  2 ,
r ∂r r r r

where the terms decaying as 1/r2 can be neglected when r and r0 are much larger than the wavelength of the source.
After substitution of (2.253), the Kirchhoff integral (2.252) becomes:
þ jkðrþr0 Þ
jk jωt e
UP ¼  U0e ½ cos ∡ðb n, r0 ÞdS,
n, rÞ  cos ∡ðb ð2:254Þ
4π rr 0
S

and this equation is commonly referred to as the Fresnel-Kirchhoff integral formula. By introducing the following
substitution:

0
ejkr
US ¼ U0 , ð2:255Þ
r0

Fresnel-Kirchhoff integral formula can be rewritten as:

þ jðkrωtÞ
jk e 0
UP ¼  US ½cos ∡ðb n, r ÞdS,
n, rÞ  cos ∡ðb ð2:256Þ
4π r
S

which represents the mathematical statement of the Huygens’ principle. The US represents the complex amplitude of the
incident primary spherical wave at the aperture (of surface S). From this primary spherical wave, each element dS of the
aperture gives rise to the secondary spherical wave at point P:

jðkrωt Þ
e
US dS, ð2:257Þ
r
58 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

and the total optical disturbance at point P can be obtained by summing up the secondary spherical waves form each aperture
surface element according to the Eq. (2.256). This summation needs to consider the inclination (obliquity) factor:

I F ðr, r0 Þ ¼ cos ∡ðb n, r0 Þ:

n, rÞ  cos ∡ðb ð2:258Þ

When the aperture is circular, the surface of integration is the spherical cap bounded by the aperture, and the normal vector is
aligned with the positon vector r0 , but in opposite direction so that cos ∡ðb n, r0 Þ ¼ cos π ¼ 1, and the inclination factor
becomes cos ∡ðb n, rÞ þ 1: In the forward direction, cos ∡ðb n, rÞ ¼ cos 0 ¼ 1 so that the inclination factor is equal to the
maximum value of 2. On the other hand, in the backward direction, cos ∡ðb n, rÞ ¼ cos π ¼ 1, and the inclination factor
becomes 0, indicating that no backward progressing wave can be created by the original wavefront. The factor –j indicates that
diffracted waves are phase shifted by π/2 rad with respect to the primary incident spherical wave.
In the far-field, the angular spread due to diffraction is small so that the obliquity factor does not vary much over the
aperture and can be taken out of the integral. Additionally the spherical wave function exp.(jkr0 )/r0 can be considered constant
over the aperture, while the change in amplitude (1/r) can be neglected, so that the Fresnel-Kirchhoff integral formula
reduces to:

jkr0
þ
jk jωt e
UP ¼  U e I ejkr dS, ð2:259Þ
4πr 0 r0 F
S

indicating that the diffracted waves are effectively plane waves. Such a diffraction is commonly referred to as the Fraunhofer
(far-field) diffraction. On the other hand, when either the receiving point or source point is close to the aperture so that the
curvature of the wavefront is relevant, such diffraction is known as the Fresnel (near-field) diffraction. There is no strict
distinction between two cases; however, the following criteria can be used. Let the size of the aperture opening (diameter) be
denoted by D, while the distances of the receiving point P and the source point S are located at distances d and d0 from the
diffracting aperture plane, respectively, as illustrated in Fig. 2.15. The wave is effectively plane over the aperture when the
curvature measure defined below is much smaller than the operating wavelength:
 
1 1 1 2
þ 0 D  λ: ð2:260Þ
2 d d

For d  d0 we are in the Fraunhofer region when d> > D2/λ.

Fig. 2.15 Distinguishing between Fraunhofer and Fresnel diffractions

2.2 Propagation of Electromagnetic Waves 59

The variation in r + r0 from one edge of the aperture to the other, based on Fig. 2.15, can be determined as:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Δðr þ r 0 Þ ¼ d0 þ ða0 þ DÞ2 þ d2 þ ða þ DÞ2  d0 þ a0 2  d2 þ D2
2 2

ð2:261Þ
a0 a 1 1 1
¼ ð 0 þ Þ þ ð þ 0 ÞD2 þ :::
d d 2 d d

Clearly, the second term in Taylor expansion represents the curvature that was used in Eq. (2.260). In terms of the Fresnel
number, the Fraunhofer condition becomes D2/(λd) < <1.
The inclination factor can also be expressed as follows:

0
0 b
nr b
nr
I F ¼ cos ∡ðb
n, rÞ  cos ∡ðb
n, r Þ ¼ 
njjr0 j
nj jrj jb
jb
|{z} ð2:262Þ
¼1
0 0 0 0
¼b
n  br  b
n  br ;br ¼ r=jrj,br ¼ r =jr j

so that the Fresnel-Kirchhoff integral formula can be represented in the following alternative form:

þ jkðrþr0 Þ
jk e 0
UP ¼  U 0 ejωt 0 n  br  b
½b n  br dS, ð2:263Þ
4π rr
S

which is valid when both receiving point P and source point S are a few wavelengths away from the aperture, in so-called
wave zone (kr> > 1).
0
Assuming that the inclination factor is nearly constant over the aperture, and replacing U 0 ejωt ejkr =r 0 with Uinc (r0 ), the
Eq. (2.263) can be rewritten as:
þ
jk ejkr
UP ¼  I F U inc ðr0 Þ dS, ð2:264Þ
4π r
S

and this equation is applicable to the wavefront Uinc generated by an extended source (a collection of point sources) as well.
By representing the aperture by a transmission function τ(r0 ), describing the changes in amplitude and phase, we come up with
more general expression as follows:
þ
jk ejkr
UP ¼  I F U inc ðr0 Þτðr0 Þ dS, ð2:265Þ
4π r
S

As an illustration, let us determine the transmission function of a thin plano-convex lens [12, 13], shown in Fig. 2.16. The
optical path between the ray shown in figure and one aligned along z-axis is kn(t  δ)  kn[t  ρ2/(2R)], where t is the
thickness of the lens and n is the refractive index of lens material.
Let us assume that incident beam is the plane wave Uinc ¼ exp.(jkz) and then transmitted beam is given by, for ray shown in
Fig. 2.16, as follows:
8 2 39
>
>   >
>
< 6 2 7=
6 ρ 2
ρ 7
U trans ¼ exp jk 4z þ þn t : ð2:266Þ
>
> 2R
|{z} 2R 5> >
: ;
δ

The transmission function for the plano-convex lens will be then:

60 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.16 Thin plano-convex lens

ρ2
τðρÞ ¼ U trans =U inc ¼ ejknt ejkðn1Þ2R : ð2:267Þ

For a biconvex lens, with corresponding curvatures being R1 and –R2, the transmission function will be:
   
ρ2
jknðt 1 þt 2 Þ jk ðn1Þ 2 R1 R2
1 1 2
ρ
jknðt 1 þt 2 Þ jk 2f 1 1 1
τ ð ρÞ ¼ e e ¼e e , ¼ ð n  1Þ  , ð2:268Þ
f R1 R2

where f, known as the focal length, is related to the net curvature of the wavefront.

2.2.4.4 Fresnel and Fraunhofer Approximations

We now discuss the Fresnel and Fraunhofer approximations with more details. For this purpose, in Fig. 2.17, we provide the
coordinate systems for source, screen, and observed planes. The point source in source plane is denoted with S, one particular
point in aperture is denoted by Q, and the receiving point is denoted by P. Since we have changed the direction of some
position vectors to facilitate explanations, the Eq. (2.265) is rewritten below to match the notations in Fig. 2.16:
þ jkr jks
jk e 0 e
UP ¼ IF U0 τðr Þ dS: ð2:269Þ
4π r s
S

The position vector of the aperture point Q with respect to the source point S is denoted by r, while r0 denotes the origin in
screen plane with respect to the source point P. The position of the aperture point Q in screen plane is denoted by ρ. These
three vectors are related as:

r ¼ r0 þ ρ, ð2:270Þ

so the length of vector r can be written as:

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 0
2r  ρ ρ2
1=2
r ¼ r  r ¼ ðr0 þ ρÞ  ðr0 þ ρÞ ¼ r 0 2 þ 2r0  ρ þ ρ2 ¼ r 0 ½1 þ 0 2 þ 0 2  : ð2:271Þ
r r

By using the binomial expansion, (1 + x)1/2 ffi 1 + x/2  x2/8 + x3/16  ⋯ and by keeping the first three terms the following
approximation is obtained:

0 0 2 0 2
r  ρ ρ2 ðr  ρÞ 0 0 ρ2 ðr  ρÞ
r ffi r0 þ 0 þ 0 ¼ r þ br  ρ þ 0  : ð2:272Þ
r 2r 2r 0 3 2r 2r 0 3
2.2 Propagation of Electromagnetic Waves 61

Fig. 2.17 Coordinate systems

and geometry for Fresnel and
Fraunhofer approximations

By close inspection of Fig. 2.17, we conclude that projection of q

vector ρ on r0 can be determined as r'  ρ. On the other hand,
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
the projection of vector ρ on the normal to the r0 is given by ρ2  ðr0  ρÞ2 . By applying the Pythagorean theorem, we
establish the following connection between r and r0 :

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 0 1=2

2 2 2 0 2r  ρ ρ2
r ¼ ðr 0 þ r0  ρÞ þ ρ2  ðr0  ρÞ ¼ r 0 þ 2r0  ρ þ ρ2 ¼ r ½1 þ 0 2 þ 0 2  , ð2:273Þ
r r

which is the same expression as given by Eq. (2.271). By applying the following two Fresnel approximations:

1. The coarse approximation in amplitude: 1/r  1/r0 .

2. The fine approximation in phase:

0 2
0 ρ2  ðbr  ρÞ
0
kr ffi kr þ kbr  ρ þ k , ð2:274Þ
2r 0

the spherical wave propagation function becomes:

b
2
jkr jkr0 ρ2 ðr0 ρÞ
e e jkb
r0 ρ jk
ffi e e 2r
0
, ð2:275Þ
r r0
0
where exp( jkr')/r' term represents the spherical wave at the aperture origin, exp ð jkbr  ρÞ corresponds to the plane wave in
direction of br0, and exp f jk½ρ2  ðbr0  ρÞ2 =ð2r Þg represents the lowest order phase approximation of the spherical wavefront.
0

By applying this phase approximation to the spherical wave exp( jks)/s, we obtain:

ðrP ρÞðrP ρÞ

zfflfflfflfflffl}|fflfflfflfflffl{ 0 2
0
jks jks jr  ρj2 ½bs  ðrP  ρÞ
e e jkb
ffi e s0 ðrP ρÞ
exp ½ jk P : ð2:276Þ
s s0 2s0
62 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

By substituting Eqs. (2.275) and (2.276) into (2.269), we obtain the following diffraction integral of a point source in Fresnel
approximation:

0 þ 0 2 0 2

jk
0
e jkðr þs Þ jkbs0 rP jqρ jkf
b
ρ2 ðr ρÞ
þ
b
ðrP ρÞ2 ½s ðrP ρÞ
g
U P ðr, sÞ ¼ I F U 0 e τðrÞe e 2r 0 2s0 dS, ð2:277Þ
4π r 0 s0
S

where the propagation vector q is defined as:

q ¼ kbr0  kbs0 : ð2:278Þ

In (2.277), the term ejkbs0rP represents the plane wave incident at receiving point P, while the term
0 2 0 2
ρ2 ðb
r ρÞ ðrP ρÞ2 ½b
s ðrP ρÞ
exp f jk 2r0 þ 2s0 g represents the wavefront curvature correction factor.
When the wavefront curvature factor can be ignored, in the far-field, the Fraunhofer diffraction integral is obtained:
þ
jqρ
U P ðr, sÞ ¼ ðconstantÞ τðrÞe dS, ð2:279Þ
S

which also represents the two-dimensional (2D) Fourier transform of the aperture transmission function τ(r). Fresnel and
Fraunhofer diffraction integrals can also be used to study the propagation of the light beam in the free-space optical (FSO)
links.
Fraunhofer diffraction integral for a rectangular aperture of dimensions 2a (in ξ-coordinate)
2b (in η-coordinate)
yields to:

0
e jkðr þs Þ jkbs0 rP a Ð b Ð
0
jk jðqx ξþqy ηÞ
U P ðr, sÞ ¼ IF U0 e e dξdη
4π r 0 s0
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} a b
ð2:280Þ
ðconstantÞ

¼ ðconstantÞ2sinc ðqx aÞsinc ðqy bÞ,

where sinc(x) ¼ sin(πx)/(πx). The corresponding intensity is given by:

I ðqÞ ¼ I ð0Þsinc2 ðqx aÞsinc2 qy b , I ð0Þ ¼ 4jðconstantÞj2 : ð2:281Þ

Fraunhofer diffraction integral for a circular aperture of radius a is given by:

0 0
jk e jkðr þs Þ jkbs0 rP 2π Ð a Ð jqρcos ϕ
U P ðr, sÞ ¼ IF U0 e e ρdρdϕ
4π r 0 s0
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} 0 0
ðconstantÞ ð2:282Þ
2J 1 ðqaÞ
¼ ðconstantÞπa2 ,
qa

where J1(x) is the Bessel function of the first kind and the first order [14]. The corresponding intensity, given by:
 2
2J 1 ðqaÞ
2
I ðqÞ ¼ I ð0Þ , I ð0Þ ¼ jðconstantÞj2 πa2 , ð2:283Þ
qa

is commonly referred to as the Airy pattern. For a small diffraction angle θ, the following is valid q ffi kθ. The first zero of J1(x)
is 1.22π, so that qa ffi (2π/λ)θ0a ¼ 1.22π, and we can write:
2.2 Propagation of Electromagnetic Waves 63

λ λ
θ0 ffi 1:22 ¼ 1:22 , d ¼ 2a, ð2:284Þ
2a
|{z} d
d

which is the same as Eq. (2.242).

2.2.5 Laser Beam Propagation Over the Atmospheric Turbulence Channels

2.2.5.1 Paraxial Wave Equation

In this section we are concerned with the laser beam propagation over the random inhomogeneous medium, namely, the
atmospheric turbulence channels. Let us introduce the slow-varying envelope of the electric field vector E(r,t) as follows:

Eðr, t Þ ¼ eEðrÞejωtjkz , ð2:285Þ

where e is the polarization vector. After substitution of (2.285) into the wave equation and upon canceling the common e exp.
(jωt) term, we obtain:

∇2 E ðrÞejkz þ k2 n2 ðrÞE ðrÞejkz ¼ 0, ð2:286Þ

where the refractive index now is a function of the position vector r. By applying the definition of Laplacian and canceling the
common exp(jkz) term, we obtain:
0 1
B 2 2C 2 
B∂ ∂ C ∂ ∂
B 2 þ 2 CE ðrÞ  2jk E ðrÞ þ 2 EðrÞ þ k2 n2 ðrÞ  1 E ðrÞ ¼ 0, ð2:287Þ
@∂x ∂y A ∂z ∂z
|fflfflfflfflfflffl{zfflfflfflfflfflffl}
∇2⊥

2 2
where we use ∇2⊥ ¼ ∂ ∖∂x2 þ ∂ ∖∂y2 to denote the transverse part of the Laplacian. In the paraxial approximation, when
∡ðk,bzÞ  1, the derivative of E is the slow-varying function of z so that the following is valid:
 2   
∂   ∂ 
   
∂z2 E ðrÞ << 2k ∂z EðrÞ, ð2:288Þ

and the third term in (2.287) can be neglected so that the wave equation reduces to:

∂ 
∇2⊥ E ðrÞ  2jk E ðrÞ þ k 2 n2 ðrÞ  1 E ðrÞ ¼ 0, ð2:289Þ
∂z

which is commonly referred to as the paraxial wave equation. By replacing n2(r) with n2 ðrÞ=n20, since in the air n0 ffi 1, we can
δn 2n0
zfflfflfflffl}|fflfflfflffl{ zfflfflfflffl}|fflfflfflffl{
ðn  n0 Þ ðn þ n0 Þ
represent nn2  1 ¼
2

n2
ffi 2δn so that the paraxial wave equation can be rewritten as:
0 0

∂ 
2jk EðrÞ ¼ ∇2⊥ þ 2k 2 δnðrÞ E ðrÞ, ð2:289Þ
∂z

providing that δn < <1.

2.2.5.2 Split-Step Beam Propagation Method

To solve the paraxial wave equation, the finite-difference method can be used [15]; however, the complexity of such approach
is high. The most common method to study the propagation over the atmospheric turbulence channels is the Fourier split-step
64 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

beam propagation method [16–24]. The key idea behind this algorithm is to split the atmospheric turbulence link into sections
of length Δz each and perform integration of the paraxial wave Eq. (2.289) on each section. The parabolic Eq. (2.289) on
section of length Δz can be solved by separation of variables:

∂E ðrÞ 1 2 
¼ ∇⊥ þ 2k 2 δnðrÞ ∂z: ð2:290Þ
E ðrÞ 2jk

By performing the integration over the i-th section, from zi to zi + 1, we obtain:

zð
iþ1 zð
iþ1
dEðrÞ j 2 
dz ¼ ∇⊥ þ 2k 2 δnðrÞ dz: ð2:291Þ
E ðrÞ 2k
zi zi
|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
 ffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
 ffl}
z ð ziþ1
ln ½E ðrÞjziþ1
i j 2
2k ∇⊥ Δzþ2k2
δnðrÞdz
zi

The Eq. (2.291) can also be written as:

  ð ziþ1 
j
Eðx, y, ziþ1 Þ ¼ exp Δz∇2⊥ þ 2k2 δnðrÞdz Eðx, y, zi Þ: ð2:292Þ
2k zi

By introducing the refraction operator R and the diffraction operator D as follows, respectively:
ð ziþ1
j
R ðzi , ziþ1 Þ ¼ jk δnðrÞdz, D ðΔzÞ ¼ Δz∇2⊥ , ð2:293Þ
zi 2k

the Eq. (2.292) can be represented as:

E ðx, y, ziþ1 Þ ¼ exp fD ðΔzÞ þ R ðzi , ziþ1 ÞgE ðx, y, zi Þ: ð2:294Þ

This notation is consistent with that used in fiber-optics communications [2, 4]. The Taylor expansion can be used to represent
the operators in exponential form:

X
1
exp ½D ðΔzÞ þ R ðzi , ziþ1 Þ ¼ ½D ðΔzÞ þ R ðzi , ziþ1 Þn =n! ð2:295Þ
n¼0

Instead of Taylor expansion, the following two methods known as symmetric split-step method (SSSM) and asymmetric split-
step method (ASSM) are commonly used in practice:
(
eD ðΔz=2Þ eR ðzi ,ziþ1 Þ eD ðΔz=2Þ Eðx, y, zi Þ, forSSSM
E ðx, y, ziþ1 Þ ¼ ð2:296Þ
eD ðΔzÞ eR ðzi ,ziþ1 Þ E ðx, y, zi Þ, forASSM

To study the accuracy of these two methods, we employ the Baker-Hausdorff formula from quantum mechanics [7, 25]:

eA eB ¼ eAþBþ2½A,Bþ12½A,½A,B12½B,½A,B::: ,
1 1 1
ð2:297Þ

where [A,B] is the commutator of operators A and B, defined as [A,B] ¼ AB-BA. By applying Baker-Hausdorff formula on
ASSM and SSM, by interpreting A ¼ D and B ¼ R , we obtain:

eDðΔzÞ eR ðzi ,ziþ1 Þ ffi eDðΔzÞþR ðzi ,ziþ1 Þþ2½DðΔzÞ,R ðzi ,ziþ1 Þþ::: ,
1

ð2:298Þ
eDðΔz=2Þ eR ðzi ,ziþ1 Þ eDðΔz=2Þ ffi eDðΔzÞþR ðzi ,ziþ1 Þþ6½R ðzi ,ziþ1 ÞþDðΔz=2Þ½R ðzi ,ziþ1 Þ,DðΔz=2Þþ:::
1
2.2 Propagation of Electromagnetic Waves 65

Fig. 2.18 Illustration of symmetric split-step method. FSO free-space optical

For sufficiently small step Δz, both diffraction and refraction operators are proportional to Δz indicating that the accuracy of
ASSM is in the order of (Δz)2, while the accuracy of SSSM is in the order of (Δz)3. Therefore, the accuracy of SSSM is better.
The SSSM can be illustrated as shown in Fig. 2.18. The algorithm proceeds by first applying the diffraction operator D for one
half step length (Δz/2). Then the refraction operator R is applied for one step length Δz. The linear operator is again applied
for the full step-size Δz. Therefore, the diffraction operator is applied at every half step (Δz /2), while the refraction operator at
every full step (Δz).
The diffraction and refraction operators can also be defined in exponential form [20, 21]. If the refraction operator R and
the diffraction operator D in exponential form are defined as follows:
 ð ziþ1  h i
j
Rðzi , ziþ1 Þ ¼ exp jk δnðrÞdz , DðΔzÞ ¼ exp Δz∇2⊥ , ð2:299Þ
zi 2k

the SSSM can be described as:

Eðx, y, ziþ1 Þ ¼ ½DðΔz=2ÞRðzi , ziþ1 ÞDðΔz=2ÞE ðx, y, zi Þ: ð2:300Þ

2.2.5.3 Paraxial Fresnel and Fraunhofer Diffraction Approximations

Let us now return to the Fig. 2.17 and observe the normal incidence case for which the inclination factor is IF ¼ 2. If the
aperture is so large, that beam can be considered as a plane so that τ(r’) ¼ 1. For this case Eq. (2.269) becomes:
þ jkðrþsÞ þ jkðrþsÞ
2π=λ e j e
UP ¼ j 2U 0 dS ¼ U 0 dS: ð2:301Þ
4π rs λ rs
S S

By applying the Fresnel approximations:

 
ð ρ  r S Þ 2 ð r P  ρÞ 2
1=r  1=r 0 , 1=s  1=s0 , jk ðr þ sÞ ffi jk r 0 þ s0 þ þ , ð2:302Þ
2r 0 2s0

the Eq. (2.301) can be represented as:

h i
þ 0 0 ðρr Þ2 ðr ρÞ2
j ejkðr þs Þ jk 2rS0 þ P2s0
UP ¼ U0 e dS: ð2:303Þ
λ r 0 s0
S

and this equation describes the free-space propagation from a point source to the screen plane and from the screen plane to the
receiving point P. By denoting the terms incident to the screen plane as Uinc, the Fresnel diffraction integral describing the
propagation from the screen plane to the observation plane can be written as:

0 þ
j ejks ðr ρÞ2
jk P2s0
UP ¼ U ð ρ Þe dS: ð2:304Þ
λ s0 inc
S
66 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.19 Illustration of paraxial

approximation to describe free-
space propagation between
observation planes zi and zi + 1.
The ρ denotes the transversal
position

For paraxial approximation, as shown in Fig. 2.19, we can see that the Eq. (2.304) can be represented in the
following form:

þ ðρ ρ Þ2

j ejkðziþ1 zi Þ jk iþ1 i
U ρiþ1 ¼ U ðρi Þe 2ðziþ1 zi Þ dSi , ð2:305Þ
λ ziþ1  zi
Si

where zi + 1  zi is the spacing between the observation planes i and i + 1, while ρ ¼ xbx þ yby is the transversal radius vector.
This equation is widely used in the existing literature [21, 26]. The Fresnel diffraction integral for paraxial approximation in
Cartesian coordinates can be straightforwardly obtained from Eq. (2.305):

ð
1 ð
1

j ejkΔz ðxiþ1 xi Þ2 þðyiþ1 yi Þ2
U xiþ1 , yiþ1 ¼ U ðxi , yi Þejk 2Δz dxi dyi : ð2:306Þ
λ Δz
1 1

The Eq. (2.306) essentially represents the action of the diffraction operator on electrical field envelope in the i-th section of
the ASSM, from observation plane i to the observation plane i + 1. For the SSM, in the i-th section, we would need to apply
Eq. (2.306) twice, once to describe FSO propagation from zi to zi + Δz/2, followed by the application of refraction operator at
zi + Δz, and the second time to describe FSO propagation from zi + Δz/2 to zi + Δz(¼zi + 1). Unfortunately, the form given by
Eq. (2.306) is not suitable for efficient simulation of diffraction effects. If we apply the binomial formula, and take out the
terms independent of integrand variables, we obtain the following form:

ð
1 ð
1

j ejkΔz jkxiþ12Δz
2
þy
iþ1
2 x2 þy2
i i 2π xi xiþ1 þyi yiþ1
U xiþ1 , yiþ1 ¼ e U ðxi , yi Þejk 2Δz ej λ Δz dxi dyi
λ Δz
1 1
2 0 13
ð
1 ð
1
6 B C7
j ejkΔz jkxiþ12Δz Bx y
2 2
þy
iþ1
x2 þy2
i i 6 C7
¼ e U ðxi , yi Þejk 2Δz exp 6j2π B iþ1 xi þ iþ1 yi C7dxi dyi ð2:307Þ
λ Δz 4 λΔz
@|{z} λΔz
|{z} A5
1 1
f xiþ1 f yiþ1

ð
1 ð
1
j ejkΔz jkxiþ12Δz
2 þy 2 x2 þy2
ej2π ð f xiþ1 xi þ f yiþ1 yi Þ
iþ1 i i
¼ e U ðxi , yi Þejk 2Δz dxi dyi
λ Δz
1 1
2.2 Propagation of Electromagnetic Waves 67

where f xiþ1 ¼ xiþ1 =ðλΔzÞ and f yiþ1 ¼ yiþ1 =ðλΔzÞ are spatial frequencies along x- and y-axes, respectively. The Eq. (2.307)


can be interpreted as the two-dimensional (2D) Fourier transform (FT) of U ðxi , yi Þ exp jk x2i þ y2i ∖ð2ΔzÞ . The 2D FT
employing 1D fast Fourier transform (FFT) algorithms can be used for this purpose as explained in [27]. Alternatively,
efficient multidimensional FFT algorithms can be used [28]. We can also interpret the Eq. (2.307) as the convolution
operation:

e jkΔz jk
x2 þy2
i i
Uðxiþ1 , yiþ1 Þ ¼ Uðxi , yi Þ  e 2Δz , ð2:308Þ
jλΔz
N
where we used the operation to denote the spatial-domain convolution. The operator * will be reserved to denote the time-
domain convolution throughout the book. By employing the convolution theorem, claiming that FT of the convolution of two
functions is equal to the product of corresponding FTs, we can rewrite the Eq. (2.308) as follows:

e jkΔz x2 þy2
i i
Uðxiþ1 , yiþ1 Þ ¼ FT 1 fFT½Uðxi , yi ÞFT½ e jk 2Δz g: ð2:309Þ
jλΔz

Given that FT of the second term in (2.308) can be determined in closed form by:

e jkΔz x2 þy2
, f i ¼ ½ f xi f yi  T ,
jk i i jkΔz jπλΔzð f 2xi þ f 2yi Þ
Hðf i Þ ¼ FT½ e 2Δz ¼e e ð2:310Þ
jλΔz

the Eq. (2.309) can be written in the form more suitable for numerical computation:

jkΔz jπλΔzð f 2xi þ f 2yi Þ

Uðxiþ1 , yiþ1 Þ ¼ FT 1 fFT½Uðxi , yi Þe e g, ð2:311Þ

and this method is known in the literature as the angular-spectral propagation method (see [21] and references

therein).
When the distance between two observation planes Δz is sufficiently large, the quadratic term k x2i þ y2i =ð2ΔzÞ in (2.307)
becomes negligible, and paraxial approximation becomes the paraxial Fraunhofer diffraction approximation:
2 0 13
ð
1 ð
1
6 B C7

j ejkΔz jkx2iþ1 þy2iþ1 6 Bx y C7
U xiþ1 , yiþ1 ¼ e 2Δz U ðxi , yi Þ exp 6j2π B iþ1 xi þ iþ1 yi C7dxi dyi
λ Δz 4 λΔz
@|{z} λΔz
|{z} A5
1 1
f xiþ1 f yiþ1 ð2:312Þ
ð
1 ð
1
j ejkΔz jkxiþ12Δz
2 þy 2

U ðxi , yi Þej2π ð f xiþ1 xi þ f yiþ1 yi Þ

iþ1
¼ e dxi dyi :
λ Δz
1 1

2.2.5.4 Rytov Method

Another approach to study the diffraction in free space is to employ the method due to Rytov [29] (see also [20]) as follows. Let
us represent the complex envelope of the electric field in the form of Fourier series [20]:

X
N=2 X
N=2
E mn ðzÞe j½ L ðmxþnyÞ ,
2π
Eðx, y, zÞ ¼ ð2:313:1Þ
m¼N=2þ1 n¼N=2

where the expansion coefficients are determined by:

ðL ðL
E ðx, y, zÞej½ L ðmxþnyÞ dxdy,
1 2π
E mn ðzÞ ¼ 2 2 ð2:313:2Þ
4π L
0 0
68 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

wherein L is the size of the expansion domain and N is the number of terms in each dimension. By substituting the Fourier
series expansion (2.312) into the paraxial wave equation (2.289) (by ignoring the refraction term) and after cancelation
common terms, we obtain:

∂ 4π 2

2jk Emn ¼  2 m2 þ n2 Emn , ð2:314Þ

∂z L

which can be solved in closed form by separation of variables:

j 4π 2 ðm2 þn2 Þz
2
E mn ðzÞ ¼ Emn ð0Þe 2kL : ð2:315Þ

This approach appears to be conceptually simpler and requires the Fourier expansion coefficients to be calculated only once
before transmission takes place at z ¼ 0. On the other hand, we need to compute N2 expansion coefficients.

2.2.5.5 Modeling Atmospheric Turbulence Effects

To solve for the refraction problem in the i-th layer of SSSM, we need to determine the variation of the refractive index δn(r),
and the refraction effect can be described by passing the beam through the phase screen as follows:

zð
iþ1

φi ðρÞ ¼ k δnðρ, zÞdz, ð2:316Þ

zi

and the refraction operator can be determined based on Eq. (2.299) as:
 ð ziþ1 
R ðzi , ziþ1 Þ ¼ exp jk δnðrÞdz ¼ exp ½ jφi ðρÞ: ð2:317Þ
zi

This approach can be used to study the thermal blooming of the high-power laser beams as described in [20]. Here we are
concerned with a fundamental problem that occurs during free-space propagation of the laser beam, even under clear weather
conditions, namely, so-called atmospheric turbulence [4, 20, 21, 30–40]. A commonly used turbulence model assumes that the
variations of the medium can be understood as individual cells of air or eddies of different diameters and refractive indices. In
the context of geometrical optics, these eddies may be observed as lenses that randomly refract the optical wavefront,
generating a distorted intensity profile at the receiver of a communication system. The intensity fluctuation is known as
scintillation and represents one of the most important factors that limit the performance of an atmospheric FSO communica-
tion link. The most widely accepted theory of turbulence is due to Kolmogorov [41–44]. This theory assumes that kinetic
energy from large turbulent eddies (pockets of air, vortices), characterized by the parameter known as the outer scale L0, is
transferred without loss to the eddies of decreasing size down to sizes of a few millimeters characterized by the inner scale
parameter l0. The inner scale represents the cell size at which energy is dissipated by viscosity. The refractive index varies
randomly across the different turbulent eddies and causes phase and amplitude variations to the wavefront. Turbulence can
also cause the random drifts of optical beams – a phenomenon usually referred to as wandering – and can induce the beam
defocusing. Clearly, the longitudinal variation in the refractive index is in the scale of l0, which is typically much smaller than
the longitudinal discretization step Δz, indicating that the Eq. (2.316) cannot be determined in the closed form. Instead, the
statistical methods must be used to properly describe the turbulent fluctuations of the refractive index. The correlation
function of the random phase fluctuation appears to be a good starting point [20]:
ð ð
Rφ ðρÞ ¼ hφðρ0 Þφðρ0 þ ρÞi ¼ k 2
hnðρ0 , zi þ z0 Þnðρ0 þ ρ, zi þ z00 Þidz0 dz00 : ð2:318Þ
Δz Δz

Based on Eq. (2.318), the spectral density of random process φ, denoted as Φφ, given as the FT of the correlation function Rφ,
can be related to the 3D spectral density of refractive index, denoted as Φn. For von Karman spectrum of atmospheric
turbulence, to be described below, it has the following form [20]:
2.2 Propagation of Electromagnetic Waves 69

 
Φφ ðκÞ ¼ FT Rφ ðρÞ ¼ 2πk 2 ΔzΦn ðκ⊥ , κ z ¼ 0, L0 ÞK ðκ, L0 =ΔzÞ, ð2:319Þ

where κ is the angular spatial frequency vector, defined as κ ¼ 2π f xbx þ f yby þ f zbz , and K() is the correction factor, which
is close to 1 when the ratio L0/Δz is sufficiently small. The refraction problem in either SSSM or ASSM can be solved by
generating the random phase screens satisfying the statistics described by Eq. (2.319). In situations when the random phase φ
is a stationary process, we can generate the samples from a Gaussian process and colored them according to Eq. (2.319) by
passing the Gaussian process though a properly designed filter. Before we return to this problem, we provide additional details
on Kolmogorov theory of turbulence [41–43].

2.2.5.6 Kolmogorov Theory of Turbulence and Corresponding Generalizations

To describe the turbulent flow, which represents a nonlinear process, Navier-Stokes equations [46] should be used. Navier-
Stokes equation for incompressible fluid of constant density ρ is given by [46]:

∂u ∇p
þ ðu∇Þu ¼  þ ν∇2 u, ð2:320Þ
∂t ρ

where u is the velocity vector of the fluid pocket, p is the pressure, and ν is the kinematic viscosity equal to 1.51105 m2/s for
the air (and 1.01106 m2/s for water). To solve the Navier-Stokes equations in turbulence regime on eddies level, it would be
quite challenging. Instead the ensemble average properties should be studied as suggested by Kolmogorov in 1941 [41]. The
turbulence can be considered as the collection of eddies of different sizes embedded in each other, changing in a random
fashion. Even under laboratory conditions, it would be quite challenging to reproduce the detailed velocity’s distribution. The
turbulent motions span wide range of eddy sizes r, ranging from macroscale at which the energy is supplied all the way down to
the microscale at which the energy is dissipated by viscosity. The average size of the largest eddies, denoted as L0, is called the
outer scale. On the hand, the average size of smallest eddies, denoted as l0, is called the inner scale. The range for which
l0 < <r < <L0 is known as the inertial range. Finally, the range for which r < l0 is known as the viscous dissipation range. A
parameter that can be used as a control parameter of the flow is known as Reynolds number, denoted as Re, and defined as [41]:

LU
Re ¼ , ð2:321Þ
ν

where L is the characteristic geometric size (length) of the system and U is the characteristic velocity. The Reynolds number,
for a turbulent flow to occur, must be very large so that the kinetic energy of the source, such as wind, is sufficiently high to
generate so-called energy cascade. According to this model, the kinetic energy of the external source excites the largest
possible eddies L0, and the energy from these eddies is gradually passed to smaller eddies, all the way down to the inner scale
l0 when the energy is dissipated by the viscosity. As an illustration, for height of 1 km above ground and a typical wind speed
of 10 m/s, the Reynolds number is huge Re ¼ 6.6108, so that the conditions for turbulence to occur are not that difficult to
satisfy. If the state of turbulence is statistically steady, then the rate of energy transfer, denoted as є, from one scale of eddies to
the next must be the same for all scales so that eddies within the same scale see the same total energy. In this scenario, the
energy rate transferred from the source to the largest eddies (L0) is the same as the energy dissipated at the shortest scale (l0).
The unit of the energy supplied to fluid per unit mass and time is єu ¼ kg m2 s2 / (kg s) ¼ m2 s3. The characteristics of
eddies of scale r depend on angular speed u and energy cascade rate E. Since the unit of speed is uu ¼ m/s, while unit of scale is
ru ¼ m, the only dimensionally acceptable relationship among u, r, and E is:

uðr Þ  E1=3 r 1=3 ð2:322Þ

So the larger the energy transfer rate, the larger the speed of eddies is. The implication of this is that the largest eddies (with
scale L0) contain the bulk of kinetic energy while the smallest eddies (with scale l0) have the lowest speeds. Additionally,
greater energy transfer will generate stronger eddies. Clearly, the largest eddies cannot be larger than the geometrical
dimension of the system, so L0
L. So the outer scale can be estimated as the cubic root of the volume of the system. On
the other hand, the inner scale is set up by viscosity, because the smaller the eddy size, the stronger velocity shear and more
important the viscosity is. The unit of kinetic viscosity is νu ¼ m2/s, as already indicated above. Therefore, the inner scale l0
70 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

must be dependent on kinetic viscosity and the energy transfer rate, so that the only dimensionally acceptable relationship is as
follows:

l0  E1=4 ν3=4 , ð2:323Þ

and often we refer to the inner scale as the Kolmogorov’s length. Given the energy cascade discussion, the span of eddies’
sizes is related to the Reynolds number. Based on Eq. (2.322) U ¼ u(L0) ~ E1/3 L1/3, and from the definition of Reynolds
number, we conclude that the energy supply/dissipation rate can be estimated as:

E  U 3 =L: ð2:324Þ

Now based on Eqs. (2.323 and 2.324) as well as definition of Reynolds number, we conclude that:

L
l0  : ð2:325Þ
Re 3=4

For the example above, we calculate that inner scale is ~0.24 mm and energy dissipation rate is ~1 m2/s3.
Another important characteristics of the turbulent channel is the energy spectrum, and it gives us the distribution of the
energy per mass across the various eddies’ sizes. Since the wave number is related to the wavelength by k ¼ 2π/λ, we conclude
that the wave number is reversely proportional to the scale r. Therefore, the extreme values for the wave number are kmin ~ E 1/
4 3/4
ν and kmax ¼ π/L0. The kinetic energy per mass of fluid, denoted as E, has the unit Eu ¼ kg m2 s2/ kg ¼ m2/s2. The
fraction of eddies with wave numbers between k and k + dk is given by:

dE ¼ EðkÞdk: ð2:326Þ

Since the unit of E(k) is m3/s2, we conclude that only dimensionally correct relationship between E(k), E, and k is:

E ðk Þ ¼ Cε2=3 k 5=3 , 1=L0 << k << 1=l0 ð2:327Þ

and we commonly refer to this equation as Kolmogorov’s 5/3 law or Kolmogorov-Obukhov 5/3 law. It has been found
experimentally that this law is valid in inertial range. The constant C in (2.327) can be determined from:

ð
1
U2
E ðkÞdk ¼ : ð2:328Þ
2
k min

Common way, inspired by Kolmogorov’s theory, to study the velocity fields in a turbulent channel is by means of flow
velocity increments:
2 3 2 3
x1 x
6 7 6 7
δuðrÞ ¼ uðx þ rÞ  uðxÞ, x ¼ 4 x2 5 ¼ 4 y 5: ð2:329Þ
x3 z

Under isotropic assumption, in Kolmogorov’s sense, in other words, when statistical properties are invariant with respect to
rotations and reflections of the original coordinate system, the flow increments (in statistical sense) depend only on |r|, and the
velocity increments emphasize the effects of scale of the order of separation |r|. Based on Eq. (2.322), we conclude that the n-
th moment of the velocity increment is given by:

h½δuðrÞn i ¼ C n εn=3 r n=3 , l0 << r << L0 : ð2:330Þ

2.2 Propagation of Electromagnetic Waves 71

The second moment of the velocity increment is commonly referred to as the structure function:
Du ðr Þ ¼ C 2u r 2=3 , l0 << r << L0 , ð2:331Þ

and the constant C2u ¼ C 2 ε2=3 is known as the velocity structure parameter. For laminar flow, when r < <l0, the dependence in
scale r is quadratic:
4=3 2
Du ðr Þ ¼ C 2u l0 r , 0
r << l0 : ð2:332Þ

Obukhov has shown in [45] that the potential temperature, defined as θ ¼ T + ahh (h is the height and ah is the adiabatic rate
of temperature decrease with height), structure parameter is similar to that of the velocity. The adiabatic rate is negligible for
small heights, close to the ground. The refractive index in point r in the atmosphere can be represented as:
 
b pðrÞ
nðrÞ ¼ hnðrÞi þ δnðrÞ ¼ 1 þ a 1 þ 2 ð2:333Þ
|fflffl{zfflffl} λ T ðrÞ
ffi1 |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
δnðrÞ

where p denotes the pressure and T denotes the absolute temperature (in K). The constants a and b at optical frequencies are
found to be [21, 30]: a ¼ 77.6106 and b ¼ 7.52103. The total differential of the refractive index is given by:
  
b dP dT
dn ¼ a 1 þ 2  2 , ð2:334Þ
λ T T

and since it is directly proportional to dT, we expect the same behavior for the refractive index structure parameter:
(
C 2n r 2=3 , l0 << r << L0
Dn ðr Þ ¼ 4=3
ð2:335Þ
C 2n l0 r 2 , 0
r << l0

where the refractive index structure parameter C2n [m2/3] is given by:

   2
b p
C2n ¼ C 2T a 1þ 2 : ð2:336Þ
λ T

The typical values for the refractive index structure parameter C2n are in the range [10–17] m2/3.
Based on Eq. (2.333), we conclude that the refractive index can be represented as the sum of mean value and fluctuating
part, so that the corresponding covariance function points r0 and r0 + r are given by:

Cn ðr0 , r0 þ rÞ ¼ h½nðr0 þ rÞ  hnðr0 þ rÞi½nðr0 Þ  hnðr0 Þii: ð2:337Þ

Under homogeneity assumption (in Kolmogorov’s sense), meaning that statistical properties are independent on initial
location but dependent only on the separation r between two points, the covariance function is only dependent on r, and by
taking the Eq. (2.333) into account, we can rewrite previous equation as:

Cn ðrÞ ¼ hδnðr0 þ rÞδnðr0 Þi: ð2:338Þ

The correlation function and the spectral density represent the Fourier transform pair, according to Wiener-Khintchine
theorem [48, 49]. Notice that in optics and physics the definitions of Fourier transform (FT) and inversed Fourier transform
(IFT) [2, 30, 33, 47] are reversed compared with corresponding definitions in electrical and computer engineering [48–50]. To
be consistent with reach literature on atmospheric turbulence effects [2, 12, 26, 30, 32–34, 47, 51], in this section we adopt
corresponding definitions used in optics. The spectral density Φn(κ) is related to the covariance function Cn(r) as follows:
72 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

ð ð
1 jκr 3 1
Φ n ð κÞ ¼ C n ðrÞe d r¼ C n ðrÞ cos ðκ  rÞd3 r ð2:339Þ
ð2π Þ3 ð2π Þ3

On the other hand, the covariance function is related to the spectral density by:
ð ð
jκr 3
C n ðrÞ ¼ Φn ðκÞe d κ ¼ Φn ðκÞ cos ðκ  rÞd3 κ ð2:340Þ

Under isotropic assumption (in Kolmogorov’s sense), the covariance function depends only on the distance between two
observation points, but not on the orientation of the line passing through them; we conclude that Cn(r) ¼ Cn(r), r ¼ |r|, and
Φn(κ) ¼ Φn(κ). By using the spherical coordinate system, as defined in Sect. 2.1.2 , and by moving to the spherical coordinates
κ ¼ (κ,θ,ϕ), given that d3κ ¼ κ2sinθ dθdϕdκ and κr ¼ κrcosθ, it is straightforward to show that Eqs. (2.339) and (2.340) can
be simplified to:
ð
1
Φn ðκÞ ¼ 2 Cn ðr Þ sin ðκr Þrdr, ð2:339Þ
2π κ

and
ð
4π
Cn ðrÞ ¼ Φn ðκÞ sin ðκr Þκdκ: ð2:340Þ
r

The refractive index structure function is given by:

D E
Dn ðrÞ ¼ ½δnðr0 þ rÞ  δnðr0 Þ2
ð ð2:341Þ
¼ 2½Cn ð0Þ  C n ðrÞ ¼ 2 Φn ðκÞ½1  cos ðκ  rÞd3 κ:

By using the Eq. (2.340), we can relate the refractive index structure function to the spectral density as follows:
2 3
6 1   7
6 ð sin ðκr Þ 2
ð
1
7
6 4π 7
Dn ðr Þ ¼ 264π Φn ðκ Þ lim κ dκ  Φn ðκÞ sin ðκr Þκdκ 7
6 r!0 κr r 7
4 0 |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} 0 5
lim ½ κ ¼1
κ cos ðκrÞ ð2:342Þ
r!0

ð
1  
sin ðκr Þ 2
¼ 8π Φn ðκ Þ 1  κ dκ:
κr
0

The term [1sin(κr)/(κr)] has the filtering properties and severely attenuations spatial frequencies for which κ < r1; in
other words, the spatial frequencies with periods that are larger than the spacing between observation points.
To avoid for this problem, we can employ the Tatarskii’s approach [51]. By taking the Laplacian of (2.341), we obtain:
ð ð
ΔDn ðrÞ ¼ ∇ð∇Dn ðrÞÞ ¼ ∇f2 Φn ðκÞsin ðκ  rÞκd 3 κg ¼ 2 Φn ðκÞcos ðκ  rÞ|{z}
κκ d3 κ ð2:343Þ
κ2

By inverting the Eq. (2.343), we obtain:

ð ð ð
1 1
cos ðκ  rÞΔDn ðrÞd 3 r ¼ 2 cos ð κ  r Þ Φn ðκÞ cos ðκ  rÞκ2 d3 κd3 r
ð2π Þ3 ð2π Þ3
ð ð 
1 ð2:344Þ
¼2 cos ðκ  rÞΦn ðκÞd r cos ðκ  rÞκ 2 d3 κ ¼ 2k 2 Φn ðκÞ
3
ð2π Þ3
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
C n ðrÞ
2.2 Propagation of Electromagnetic Waves 73

Therefore, from (2.344), we obtain:

ð
1
Φn ðκÞ ¼ cos ðκ  rÞΔDn ðrÞd 3 r: ð2:345Þ
16π 3 κ 2

Since for locally isotropic medium in spherical coordinates the Laplacian is defined as:
 
1 ∂ 2 ∂Dn ðrÞ
ΔDn ðrÞ ¼ r ,
r 2 ∂r ∂r

the previous equation becomes:

ð
1  
1 sin ðκr Þ d 2 d
Φn ðκ Þ ¼ 2 2 r D ðr Þ dr, ð2:346Þ
4π κ κr dr dr n
0

where Dn(r) ¼ Dn(r).

The following covariance function and spectral density represent the Fourier transform pair [51]:
 υ  
a2 r r
C ðr Þ ¼ Kυ , C ð 0Þ ¼ a2
2υ1 ΓðυÞ r 0 r0
ð2:347Þ
Γðυ þ 3=2Þ a2 r 30
ΦðκÞ ¼ pffiffiffi
,
π π ΓðυÞ 1 þ κ 2 r 2 υþ3=2
0

Ð1
where Γ(a) is the gamma function, defined as ΓðaÞ ¼ 0 t a1 et dt. The Kυ(x) denotes the modified Bessel function of the
second kind and order υ (also known as Macdonald function) defined as KÐυ(u) ¼ {π/[2 sin (υπ)]}[Iυ(u)  Iυ(u)], which is
π
related to the modified Bessel function of the first kind by I υ ðuÞ ¼ ð1=π Þ 0 cos ðυϑÞ exp ðu cos ϑÞdϑ: The corresponding
structure function is given by:
  υ  
1 r r
Dðr Þ ¼ 2½C ð0Þ  Cðr Þ ¼ 2a2 1  Kυ : ð2:348Þ
2υ1 ΓðυÞ r 0 r0

By using the following approximation for the modified Bessel function Kυ(u)  {π/[2 sin (υπ)]}{[1/Γ(1  υ)](z/2)υ 
[1/Γ(1 + υ)](z/2)υ} which is valid for r < <r0, we obtain the following approximation for the structure function:
 2υ
Γð1  υÞ r
Dðr Þ ffi 2a2 , r << r 0 : ð2:349Þ
Γð1 þ υÞ 2r 0

Γð1υÞ
By introducing the substitutions C 2 ¼ 2a2 Γð1þυ Þð2r Þ2υ
, μ ¼ 2υ, we can rewrite the Eq. (2.349) as follows:
0

D ðr Þ ffi C 2 r μ , ð2:350Þ

which is consistent with the refractive index structure function (2.335), by setting μ to either 2/3 or 2. Since the approximation
(2.349) is valid for r < <r0, or equivalently κ> > 1/r0, the corresponding approximation for spectral density, based on (2.347),
becomes:

Γðυ þ 3=2Þ a2 r 30 sin ðπμ=2Þ 2 ðμþ3Þ

Φðκ Þ ffi pffiffiffi ¼ Γ ð μ þ 2Þ C κ : ð2:351Þ
π π ΓðυÞ ðκr 0 Þ2υþ3 4π 2
74 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

The refractive index spectral density based on (2.347) and (2.351) can be written as:
8 0:033005391
>
> zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
>
>
< Γð2=3 þ 2Þ sin ðπ=3Þ 2 11=3
Cn κ , 2π=L0 << κ << 2π=l0
Φðκ Þ ¼ 4π 2
>
>
>
> Γð4Þ 2 4=3 5
: C l κ , κ >> 2π=l0 ð2:352Þ
4π 2 n 0
(
0:033005391C 2n κ 11=3 , 2π=L0 << κ << 2π=l0
¼
4=3
0:151981775C 2n l0 κ 5 , κ >> 2π=l0

and this spectral density is commonly referred to as Kolmogorov refractive index spectral density.
To include the effects related to both inner and outer scales of turbulence, the various modification to the Kolmogorov
spectrum have been introduced so far [20, 21, 30, 32, 33, 51, 62]. The modified von Kármán spectral density is given by:

11=6 κ2 =κ2
ΦðκÞ ¼ 0:033005391C 2n κ 2 þ κ20 e m , κ ¼ 2π=L , κ
0 0 m ¼ 5:92=l0 ð2:353Þ

The spectrum is given by Eq. (2.353), for which κ0 ¼ 0 is known as the Tatarskii spectrum [62]. Clearly, by setting l0 ¼ 0, and
L0 ! 1, the Kolmogorov spectral density is obtained. Another relevant spectral density is Andrews spectral density,
representing the generalization of the modified von Kármán spectral density [52] and providing nice agreement with
experimentally verified Hills spectrum:
h i
11=6 κ2 =κ2
ΦðκÞ ¼ 0:033C 2n 1 þ 1:802ðκ=κl Þ  0:254ðκ=κl Þ7=6 κ 2 þ κ20 e l , κ ¼ 3:3=l
l 0 ð2:354Þ

As an illustration, in Fig. 2.20 we plot spectral density for common models of atmospheric turbulence: Kolmogorov,
modified von Kármán, Andrews, and Tatarskii spectra. Clearly, in inertial range we cannot see much difference among
different models. For high spatial frequencies, Tatarskii spectrum behaves the same as modified von Kármán spectrum. On the

Fig. 2.20 Spectral densities for

common models of atmospheric
turbulence
2.2 Propagation of Electromagnetic Waves 75

other hand, Tatarskii spectrum behaves the same as Kolmogorov spectrum at low spatial frequencies. Andrews spectrum
decays faster than modified von Kármán spectrum in high spatial frequencies region.
Now by using the Eq. (2.319), we can establish the connection between spectral densities of atmospheric turbulence-
induced random phase and spectral density of refractive index, which is locally a correct relationship. Since the refractive
index structure parameter is propagation distance dependent, we have to perform the integration over z so that the following
relationship is obtained for Kolmogorov spectral density:

ðz
Δ
11=3
Φφ ðκÞ ¼ 2πk  0:033005391κ
2
C 2n ðzÞdz: ð2:355Þ
0

By taking the zenith-angle ζ into account, the Eq. (2.355) can also be written in more common form as follows:

8 95=3
>
> >
>
> 2 33=5 > >
>
>
< ð
Δ z >
=
2π  0:033005391 11=3 4 5
Φ φ ð κÞ ¼ κ 0:423k sec ðζ Þ Cn ðzÞdz
2 2
0:423
|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} >
> >
>
>
> >
>
0:490258 :|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}>
> 0
; ð2:356Þ
r0
2 33=5
ðz
Δ

, r 0 ¼ 40:423k2 sec ðζ Þ C 2n ðzÞdz5

5=3
¼ 0:490258κ11=3 r 0
0

where r0 is commonly referred to as Fried parameter [53]. Fried defined atmospheric turbulence diameter r0 in a different
context, as the scale length over which the phase errors (due to atmospheric turbulence) in the wavefront are of the order of
1 rad. In other words, the phase standard deviation over an aperture of size (diameter) r0 is approximately 1 rad. The Fried
parameter in Eq. (2.352) is valid for plane waves; the corresponding expression for spherical waves must be modified with a
correction factor z/Δz to obtain [30]:

2 33=5
ðz
Δ
ðspherical wavesÞ
r0 ¼ 40:423k2 sec ðζ Þ C 2n ðzÞðz=ΔzÞ5=3 dz5 : ð2:357Þ
0

By following the similar derivation, as for Kolmogorov spectral density, the phase spectral density for the modified von
Kármán refractive index is given by:

5=3 11=6 κ 2 =κ 2
ΦðmodifiedvonKarmanÞ
φ ðκÞ ¼ 0:490258r 0 ðκ 2 þ κ20 Þ e m : ð2:358Þ

e f x , f y . Therefore,
Let the Fourier transform of the atmospheric turbulence perturbed phase φ(ρ) ¼ φ(x,y) be denoted as φ
it is related to φ(x,y) as follows:

ð
1 ð
1

j2π ð f x xþ f y yÞ
φðx, yÞ ¼ e f x, f y e
φ df x df y : ð2:359Þ
1 1

The total power in the phase screen can be determined by employing the Parseval’s theorem by:

ð
1 ð
1 ð
1 ð
1
2

Ptotal ¼ jφðx, yÞj dxdy ¼ Φφ f x , f y df x df y : ð2:360Þ

1 1 1 1
76 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

By using the trapezoidal rule, we can write Eq. (2.360) as:

ð
1 ð
1
X X

Ptotal ¼ Φφ f x , f y df x df y ffi m
Φ mΔ f x , nΔ f y Δ f x Δ f y ,
n φ
ð2:362Þ
1 1

where Δfx and Δfy are frequency spacing along fx- and fy-axes. On the other hand, by applying the trapezoidal rule on
Eq. (2.359), we obtain:
X X
j2π ð f x xþ f y yÞ
φðx, yÞ ffi m n
e mΔ f x , nΔ f y e
φ Δ f xΔ f y: ð2:363Þ

From Eqs. (2.357), (2.358), we conclude that:

D
2 E

φ
e mΔ f x , nΔ f y  Δ f x Δ f y ¼ Φφ mΔ f x , nΔ f y Δ f x Δ f y : ð2:364Þ

To generate the random phase screen, we can use the Gaussian random generator with zero-mean and unit variance and
multiply the samples with the square root of variance given by Eq. (2.364). On such a way we can generate the samples for 2D
random phase screen. Now we have everything needed to implement the symmetric split-step method. The main drawback of
this method random phase screen generator is the low accuracy of the trapezoidal rule. To improve the accuracy, Zernike
polynomial-based approach can be used [20, 54]. Alternatively, the non-uniform sampling can be employed and sample low
spatial frequencies more precisely [55, 56]. Finally, the subharmonic method can be used [57] (see also [21]). In subharmonics’
method, the low spatial frequency phase screen is a sum of Np phase screens with different accuracy levels [21, 57]:

Np X
X 1 X
1  
j2π ð f x xþ f y yÞ
φLSF ðx, yÞ ffi e mΔ f xp , nΔ f yp e
φ Δ f xp Δ f yp , ð2:365Þ
p¼1 m¼1 n¼1

where the frequency spacing for the p-th frequency grid is set to Δ f xp ¼ Δ f yp ¼ Δf =bp , with Δf being the frequency spacing
used for medium and high spatial frequency ranges, and b
b denotes the size of grid in the p-th frequency grid. As it has been
shown in [21, 57], the subharmonics’ method can significantly improve the accuracy of random phase screen generation.
We turn our attention now back to the correlation function of the random phase correlation, given by Eq. (2.318).
ð ð ð ð
Rφ ðρÞ ¼ k2 hnðρ0 , zi þ z0 Þnðρ0 þ ρ, zi þ z00 Þidz0 dz00 ¼ k 2 Rn ðρ, z00  z0 Þdz0 dz00 : ð2:366Þ
Δz Δz Δz Δz

When Δz is much larger than the correlation length of the refractive index, the autocorrelation function of the refractive index
00 00 00
can be written as Rn(ρ, z  z0) ¼ Rn(ρ, z )Δzδ(z  z0), and after extending the limits of integration and applying the sifting
property, we obtain:

ð
1

Rφ ðρÞ ¼ k Δz 2
Rn ðρ, zÞdz: ð2:367Þ
1

Given that the hδni ¼ 0, we conclude that Cn(ρ, z) ¼ Rn(ρ, z) and consequently Cφ(ρ) ¼ Rφ(ρ). The structure function of the
random phase can now be written as:

ð
1

Dφ ðρÞ ¼ 2 C φ ð0Þ  Cφ ðρÞ ¼ k2 Δz 2½C n ð0, zÞ  C n ðρ, zÞdz
1
8 9 ð2:368Þ
ð
1 >
< >
=
¼ k Δz 2
2½Cn ð0, 0Þ  Cn ðρ, zÞ  2½Cn ð0, 0Þ  Cn ð0, zÞ dz
:|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}>
> ;
1 Dn ðρ, zÞ Dn ð0, zÞ
2.2 Propagation of Electromagnetic Waves 77

Since for Kolmogorov model Dn ðr Þ ¼ C 2n r 2=3 , the Eq. (2.368) becomes:

ð
1
n o

1=3
Dφ ðρÞ ¼ k 2
ΔzC 2n ρ2 þ z2  z2=3 dz ¼ 2:914k 2 ΔzC 2n ρ5=3 : ð2:369Þ
1

Since the refractive index structure parameter is propagation distance dependent, we have to perform the integration over z,
and after taking the zenith-angle into account, we obtain:

ð
1  5=3
ρ
Dφ ðρÞ ¼ 2:914k sec ðζ Þρ2 5=3
C 2n ðzÞdz ¼ 6:889 : ð2:370Þ
r0
0

Another parameter related to atmospheric turbulence, relevant in astronomy, is so-called isoplanatic angle. Namely, in
optical systems in which the system characteristics are not shift-invariant, we say that such systems exhibit the anisoplanatism.
To measure the severity of the anisoplanatism, the starting point would be the Eq. (2.369), which after inclusion of the zenith-
angle can be written as:

Dφ ðρÞ ¼ 2:914k2 Δz sec ðζ ÞC2n ρ5=3 : ð2:371Þ

A stellar separation of θ rad corresponds to the optical path from the stars to the telescopes through the single layer of
atmospheric turbulence at the height Δh above the telescope, and we can write ρ ¼ θ Δh secζ. By substituting this value of ρ
into (2.366), we obtain:

Dφ ðρÞ ¼ 2:914k 2 Δz sec ðζ ÞC2n ðθΔh sec ζ Þ5=3 ¼ 2:914k 2 ΔzðΔhÞ5=3 sec 8=3 ðζ ÞC2n θ5=3 : ð2:372Þ

By integrating over the entire atmosphere, we obtain:

ð1  5=3
θ
Dφ ðρÞ ¼ 2:914k2 sec 8=3 ðζ Þθ5=3 C 2n ðzÞz5=3 dz ¼ ,
0 θ0
 ð1 3=5 ð2:373Þ
θ0 ¼ 2:914k sec ðζ Þ Cn ðzÞz dz
2 8=3 2 5=3
,
0

where θ0 represents the isoplanatic angle. Clearly, the isoplanatic angle is angle for which Dφ(ρ) ¼ 1 rad2.

2.2.5.7 Rytov Method of Small Perturbations

One important method to study the laser beam propagation over atmospheric turbulence channels is the Rytov method of small
perturbations [58–62] (see also [30, 47, 51]), in which the optical field in cylindrical coordinates r ¼ (ρ,L ) (L-the transmission
distance) is represented as:

Uðρ, LÞ ¼ U 0 ðρ, LÞexp ½ψðρ, LÞ, ψðρ, LÞ ¼ ψ 1 ðρ, LÞ þ ψ 2 ðρ, LÞ þ ::: ð2:374Þ

where U0(r) is the vacuum solution, while ψ(r) is used to denote the complex phase perturbation. In (2.369), we use ψ i(r) to
denote the i-th order perturbation. These successive perturbations are used to determine the statistical moments of ψ(r). The
mean value of the optical field is given by hU(ρ, L)i ¼ U0(ρ, L )hexp[ψ(ρ, L )]i. The mutual coherence function (MCF) is
essentially the second moment of the optical field:

Γðρ1 , ρ2 , LÞ ¼ hU ðρ1 , LÞU  ðρ2 , LÞi: ð2:375Þ

The normalized magnitude of the MCF is often called the modulus of the complex coherence factor (or just the coherence
factor):
78 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

j Γ ð ρ 1 , ρ 2 , LÞ j
γ ð ρ 1 , ρ 2 , LÞ ¼ : ð2:376Þ
½Γðρ1 , ρ1 , LÞΓðρ2 , ρ2 , LÞ1=2

The coherence factor over homogeneous and isotropic turbulence is only function of a displacement, ρ ¼ |ρ| ¼ |ρ2  ρ1|, and
for plane wave, we obtain:
8 9
< ðL 1
ð =
γ ðρ, LÞ ¼ exp 4π 2 k2 Φn ðκ, zÞ½1  J 0 ðκρÞdzdκ : ð2:377Þ
: ;
0 0

When Kolmogorov spectral density is used, the coherence factor simplifies to:
8 9
< ðL =
γ ðKolmogorovÞ ðρ, LÞ ¼ exp 1:457k2 ρ5=3 C 2n ðzÞdz : ð2:378Þ
: ;
0

The spatial coherence radius of an optical wave can be defined as the e1 of the coherence factor:

2 33=5
ðL
ρ0 ¼ 41:457k2 C2n ðzÞdz5 : ð2:379Þ
0

The complex phase perturbation ψ(r) can be expressed as follows:

 
U ðρ, LÞ
ψ ðρ, LÞ ¼ log ¼ χ þ jS, ð2:380Þ
U 0 ðρ, LÞ

where χ is the log-amplitude fluctuation and S is the corresponding phase fluctuation. In the weak turbulence regime, it is
reasonable to assume χ and S to be Gaussian random processes, by invoking the central limit theorem. The irradiance I of the
optical field at distance L can be determined as:

I ¼ jU j2 ¼ jU 0 j2 exp ð2χ Þ ¼ I 0 exp ð2χ Þ: ð2:381Þ

|ffl{zffl}
I0

Let the variance of the log-amplitude fluctuation be obtained as:

 
σ 2χ ¼ χ 2  hχ i2 : ð2:382Þ

We are concerned to determine the distribution of new random variable I, related to the old random variable χ by Eq. (2.381).
In general, when the transformation of random variable X has the form Y ¼ f(X), the distribution for Y is given by fY( y) ¼ fX(x)|
dx/dy|. By applying this theory, we obtain the following distribution of the irradiance I:

 0 χ 1 h   i2
 zfflfflfflfflfflfflffl}|fflfflfflfflfflfflffl{ 
      1 log I
I0
 hχ i
 d B1 C ðχhχ iÞ 2 2
dχ   B I C  1  1 1 
f ðI Þ ¼ f ðχ Þ  ¼ f ðχ Þ @ log ¼ pffiffiffiffiffi e ¼ pffiffiffiffiffi e
2
2σ χ 2σ 2χ

dI dI 2 I 0 A σ χ 2π 2I 2σ χ I 2π
 
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} ð2:383Þ
1=ð2I Þ
h   i2
log I 2hχ i
I0
1 
¼ pffiffiffiffiffi e 8σ 2
χ :
2σ χ I 2π
2.2 Propagation of Electromagnetic Waves 79

Clearly, the distribution of irradiance is the log-normal. This distribution is valid only when the first perturbation term in
Rytov approximation dominates, in other words, in so-called weak turbulence regime (to be precisely defined below).
The coherence factor can be used to determine the wave structure function, defined as follows:

Dðρ1 , ρ2 , LÞ ¼ 2 log γ ðρ1 , ρ2 , LÞ ¼ Dχ ðρ1 , ρ2 , LÞ þ DS ðρ1 , ρ2 , LÞ, ð2:384Þ

where we use Dχ and DS to denote the log-amplitude and phase structure parameters, respectively. The phase spectral density
is now related to the phase structure function, based on Eq. (2.346), as follows:

ð
1  
1 sin ðκr Þ d 2 d
ΦS ðκ Þ ¼ 2 2 r DS ðr Þ dr: ð2:385Þ
4π κ κr dr dr
0

Fluctuations in irradiance are governed by the fourth-order moment of the optical field given as follows:

Γ4 ðρ1 , ρ2 , ρ3 , ρ4 , LÞ ¼ hU ðρ1 , LÞU  ðρ2 , LÞU ðρ3 , LÞU  ðρ4 , LÞi: ð2:386Þ

From this equation, the second moment of irradiance hI2(ρ1, L )i is obtained from Eq. (2.383) by setting ρ4 ¼ ρ3 ¼ ρ2 ¼ ρ1, and
it is used to define so-called scintillation index by:
 2  2
I  hI i2 I
σ 2I ¼ ¼  1: ð2:387Þ
hI i2 hI i2

The scintillation index of an unbounded plane wave in a weak turbulence regime (defined as the regime for which the
scintillation index is less than unity) based on a Kolmogorov spectral density is commonly referred to as the Rytov variance
[4, 30]:

σ 2R ¼ 1:23 C 2n k 7=6 L11=6 : ð2:388Þ

Even though it is derived under the first-order perturbation assumption, which is valid in weak turbulence only, the Rytov
variance is commonly used as a measure of optical turbulence strength. The weak fluctuations are associated with σ2R < 1, the
strong with σ2R > 1, and the saturation regime is defined by σ2R ! 1 [4, 30].
From probability theory we know that when X has the normal distribution with mean value μ and variance σ 2, the
distribution of random variable Y ¼ exp.(X) is log-normal with the first two moments given by μ1 ¼ exp.(μ + σ 2/2) and
μ2 ¼ exp.(μ + σ 2). By using this fact, the scintillation index in weak turbulence regime, where the distribution of irradiance is
given by Eq. (2.383), can be related to the log-amplitude variance by:
 2
I e2ð2hxiþ4σχ Þ
2
2
σ 2I ¼ 1¼ 2 =2 2
 1 ¼ e4σ χ  1: ð2:389Þ
hI i2
e 2 hx iþ4σ χ

2.2.5.8 Gamma-Gamma Distribution of Irradiance

We are now concerned with extending the Rytov approximation to strong turbulence. By close inspection on modified von
Kármán spectral density given by Eq. (2.353), we notice that Tatarskii’s term exp.(κ/κm) corresponds to the large-scale filter

11=6
function spatial frequency cutoff κm. On the other hand, the term κ2 þ κ 20 corresponds to the small-scale filter function
with spatial frequency cutoff κ0. By denoting the large-scale and small-scale cutoff spatial frequency by κx and κy,
respectively, we can generalize the Kolmogorov spectrum as follows:

 11=6
Φn ðκ Þ ¼ 0:033005391C2n κ11=3 eκ =κx þ 0:033005391C2n κ2 þ κ2y
2 2
: ð2:390Þ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Φn,X ðκ Þ
Φn,Y ðκ Þ
80 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

The small-scale effects are contributed by eddies of size smaller than small scale size lss defined by Eq. (2.391). On the other
hand, the large-scale effects are contributed by eddies of size larger than Lls, defined as:
 pffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffi 
lss ¼ min ρ0 , L=k , Lls ¼ max L=k, L=ðkρ0 Þ : ð2:391Þ

pffiffiffiffiffiffiffiffi
In (2.391), ρ0 denotes the spatial coherence radius as defined by Eq. (2.379), and L=k is related to the first Fresnel zone
radius and L/(kρ0) to the scattering disk size. The addition operation in spectral density of random phase indicates the product
of random variable X and Y in irradiance I, namely, I ¼ XY. If we assume that large-scale fluctuations, describing the refraction
properties, described by random variable X and the small fluctuations, describing the diffraction properties, described by
random variable Y are mutually independent, in statistical sense, the scintillation index can be written as follows:
  

σ 2I ðρ, LÞ ¼ X 2 Y 2  1 ¼ 1 þ σ 2X 1 þ σ 2Y  1 ¼ σ 2X þ σ 2Y þ σ X σ Y : ð2:392Þ

By expressing the variances of random variables X and Y in terms of log-amplitudes’ variances, we can rewrite the scintillation
index as follows:
   
σ 2I ðρ, LÞ ¼ exp σ 2log X exp σ 2log Y  1: ð2:393Þ

By assuming that both large-scale and small-scale irradiance functions are governed by gamma distributions [63] (see also
[30]), with different parameters:

αðαxÞα1 αx
f X ð xÞ ¼ e ; α > 0, X > 0
ΓðαÞ
ð2:394Þ
βðβyÞβ1 βy
f Y ð yÞ ¼ e ; β > 0, Y > 0
ΓðβÞ

the conditional probability density function (PDF) fI|X(I ) is given by:

βðβI=xÞβ1 βI=x
f IjX ðI Þ ¼ e : ð2:395Þ
ΓðβÞ

By averaging out over random variable X in conditional PDF (2.395), we obtain the following PDF of irradiance I:

ð
1 ð
1
βðβI=xÞβ1 βI=x αðαxÞα1 αx
f ðI Þ ¼ f IjX ðI Þ f X ðxÞdx ¼ e e
Γ ðβ Þ ΓðαÞ
0 0 ð2:396Þ
2ðαβÞ
αþβ
2 αþβ
 pffiffiffiffiffiffiffi
¼ I 2 1 K αβ 2 αβI ,
ΓðαÞΓðβÞ

which is commonly referred to as gamma-gamma distribution [4, 30, 35–39, 63]. Given that the expected value of irradiance
is 1, the second moment of irradiance for gamma-gamma distribution is given by:
 2
I ¼ ð1 þ 1=αÞð1 þ 1=βÞ, ð2:397Þ

so that the scintillation index is given by:

1 1 1
σ 2I ¼ þ þ : ð2:398Þ
α β αβ
2.2 Propagation of Electromagnetic Waves 81
   
By close inspection of Eqs. (2.393) and (2.397), we conclude that 1 þ 1=α ¼ exp σ 2log X and 1 þ 1=β ¼ exp σ 2log Y , and
after rearranging we obtain:
h   i1 h   i1
α ¼ exp σ 2log X  1 ¼ σ 2
X , β ¼ exp σ log Y  1
2
¼ σ 2
Y : ð2:399Þ

Therefore, parameters α and β in gamma-gamma distribution can be related to large-scale and small-scale scintillation
effects, respectively.
To completely specify the gamma-gamma distribution, we need to derive the log-amplitude variances. For this purpose, we
employ the method of small perturbations as described by Tatarskii [51] (see also [30, 47]). We return back to the Helmholtz
equation:

∇2 U þ k2 n2 ðrÞU ¼ 0, ð2:400Þ

and expand the optical field U into a series of terms:

U ¼ U 0 þ U 1 þ U 2 þ :::; jU mþ1 j << jU m j, m ¼ 0, 1, 2, . . . ð2:401Þ

Let us observe the first two terms in equation above, initially U ffi U0 + U1, and after the substitution into (2.400), we the
following two equations:

∇2 U 0 þ k2 U 0 ¼ 0, ∇2 U 1 þ k2 U 1 ¼ 2k 2 δnU 0 , ð2:402Þ

where we obtained these equations by setting to zero all terms of the order δn in the approximation of the square refractive
index Eq. (2.333):

nðrÞ ¼ ½1 þ δnðrÞ2 ffi 1 þ 2δnðrÞ, δnðrÞ << 1 ð2:403Þ

By interpreting the term 2k2δnU0 as the excitation, by employing the Green’s theorem [5], we can determine U1 as the 3-D
convolution of Green’s function (spherical wave) G(r,r0 ) ¼ (4π)1exp(jk|r  r0 |)/|r  r0 | and the excitation 2k2δnU0:

ð 0
1 e jkjrr j 2 0 0
3 0
U 1 ðrÞ ¼ 0 2k δnðr ÞU 0 ðr Þd r : ð2:404Þ
4π V jr  r j

Now by observing only Um and Um + 1 terms, we can establish the identical relationship:

ð 0
1 e jkjrr j 2 0 0
3 0
U mþ1 ðrÞ ¼ 0 2k δnðr ÞU m ðr Þd r ; m ¼ 0, 1, 2, . . . ð2:405Þ
4π V jr  r j

In cylindrical coordinates we can write r ¼ (ρ,z). Let us now apply the coarse approximation in amplitude |r  r0 |  z  z0 and
fine approximation in phase as follows:

0 2 0 4
0 ðρ  ρ Þ ðρ  ρ Þ
kjr  r j ¼ kðz  z0 Þ½1 þ þ þ ::: ð2:406Þ
0
2ðz  z Þ 2
8ðz  z0 Þ4

By substituting (2.405) into (2.404), we obtain:

ð ðρρ Þ
exp ½ jk 2ðzz 
0 2

k2 jkðzz0 Þ
0 Þ2 0 0 0
U mþ1 ðrÞ ffi e δnðr ÞU m ðr Þd3 r ; m ¼ 0, 1, 2, . . . ð2:407Þ
2π V z  z0

By normalizing the perturbations with U0(r), we can rewrite the Eq. (2.406) as follows:
82 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

0 2
ð ðρρ Þ
exp ½ jk 2ðzz  0
U mþ1 ðrÞ k 2
jkðzz0 Þ
0 Þ2 0 U m ðr Þ 3 0
ffi e δnðr Þ d r ; m ¼ 0, 1, 2, . . . ð2:408Þ
U 0 ðrÞ 2π V z  z0 U 0 ðrÞ

For unbounded plane wave, U0 ¼ exp.(jkz0 ) and m ¼ 1, and Eq. (2.408) simplifies to:

0 2
ðplane waveÞ ð exp ½ jk ðρρ Þ

U1 ðrÞ k 2
2ðzz0 Þ
2
0 0
¼ δnðr Þd3 r : ð2:409Þ
U 0 ðzÞ 2π V zz 0

By taking only the first two terms in expansion (2.401), the log-amplitude expression can be written as:
     
U0 þ U1 U jU j<<jU 0 j jU 1 j U1
ψ ¼ χ þ jS ¼ log ¼ log 1 þ 1 ffi log1 ðxÞxx þ jarg : ð2:410Þ
U0 U0 2 =2
jU 0 j U0

Based on Eqs. (2.407) and (2.410), we conclude that:

0 2 0 2
ð cos ½k ðρρ Þ 2  ð sin ½k ðρρ Þ 2 
k2 2ðzz0 Þ 0 k2 0 2ðzz0 Þ 0 0
χffi δnðr Þd3 r , S ¼ δnðr Þd3 r ð2:411Þ
2π V z  z0 2π V z  z0

Under assumption that refractive index variation is only in the plane perpendicular to the propagation direction, we can
represent the refractive index variation as a 2D Fourier-Stieltjes transform [51] (see also [47]):
ð
jκρ
δnðrÞ ¼ e dvðκ, zÞ, ð2:412Þ

where the random amplitude of the refractive index fluctuations has the following correlation function:

0 0 0
hdvðκ, z0 Þdv ðκ, z00 Þi ¼ δðκ  κ ÞF n ðκ , z0  z00 Þd2 κd2 κ ,
Ð 0 ð2:413Þ
F n ðκ0 , z0  z00 Þ ¼ 1 Φn ðκ , κz Þcos ½κz ðz0  z00 Þdκz :
1

By substituting (2.412) in (2.411), and after performing the integration with respect to ρ, we obtain the following solutions
[51] (see also [47]):
ð ð ð ð
jκρ z κ2 ðz  z0 Þ 0 0 jκρ z κ2 ðz  z0 Þ 0
χ¼ e k sin ½ dz dvðκ, z Þ, S ¼ e k cos ½ dz dvðκ, z0 Þ: ð2:414Þ
2k 2k
0 0

The covariance function of χ is defined by:

C χ ðρ, zÞ ¼ hχ ðρ1 þ ρ, zÞχ  ðρ1 , zÞi, ð2:415Þ

and after substitution of (2.414) into (2.415), we obtain:

C χ ðρ, zÞ ¼
ðð ðz ðz  2   02 
jκðρ1 þρÞjκ0 ρ κ ðz  z0 Þ κ ðz  z00 Þ 0 00
k 2
e sin sin dz dz hdvðκ, z0 Þdv ðκ0 , z00 Þi
2k 2k
0 0 ð2:416Þ
ð ðz ðz   2  
jκρ κ 2 ðz  z0 Þ κ ðz  z00 Þ
¼ e sin sin F n ðκ, z0  z00 Þdz0 dz00 d2 κ
2k 2k
0 0
2.2 Propagation of Electromagnetic Waves 83
00 00
For isotropic assumption we have that Fn(κ, z0  z ) ¼ Fn(κ, z0  z ), and after introducing the following substitutions
00 00
z0  z ¼ ξ, z0 + z ¼ 2η (whose Jacobian is 1), we can write the amplitude spectrum as follows [47]:

ð 
ðL Lξ=2  2   2 
κ ξ κ ðL  η Þ
F χ ðκ, 0Þ ¼ k 2
cos  cos F n ðκ, ξÞdξdη ð2:417Þ
2k 2k
0 ξ=2

Ð1
Under assumptions ξ ffi L, κ2ξ/(2k) ffi L, for sufficiently long propagation distance L, we have that 0 F n ðκ, ξÞdξ ¼
πΦn ð0, κÞ ¼ πΦn ðκÞ so that the Eq. (2.417) simplifies to:



F χ ðκ, 0Þ ¼ πkL2 1  sin κ2 L=k = κ2 L=k Φn ðκÞ: ð2:418Þ

For isotropic assumption, it is convenient to move from Cartesian coordinates κ ¼ (κx,κy) to polar coordinates (κ,ϕ), so that the
covariance function at distance z ¼ L becomes [47]:

ð
1 2ðπ ð
1
1
Cχ ðρÞ ¼ 2π ejκρ cos ϕ dϕF χ ðκ, 0Þdκ ¼ 2π J 0 ðkρÞF χ ðκ, 0Þdκ, ð2:419Þ
2π
0 0 0
|fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl}
J 0 ðkρÞ

where J0() is the zero-order Bessel function of the first kind. By substituting (2.418) into (2.419), we obtain the following
expression for covariance function of χ:

ð
1 
sin ðκ2 L=kÞ
C χ ðρÞ ¼ 2π 2 k2 L 1 J 0 ðκρÞΦn ðκÞκdκ, ð2:420Þ
κ2 L=k
0

The variance of χ for plane wave can be determined as:

ð
1 
sin ðκ 2 L=k Þ
σ 2χ ¼ Cχ ð0Þ ¼ 2π k L 2 2
1 Φn ðκ Þκdκ: ð2:421Þ
κ 2 L=k
0

For Tatarskii spectral density, given by [62]:

Φn ðκÞ ¼ 0:033C 2n κ 11=3 exp κ 2 =κ 2m , ð2:422Þ

we obtain the following expression for variance of χ:

6π 2 3 2 7=3
σ 2χ ffi 0:033 L C n κm ¼ 0:214745L3 C 2n κm
7=3
: ð2:423Þ
91

In this derivation, the following series expansions have been used: sin(x)  x  x3/6 and exp(x)  1  x. Inpweak
ffiffiffiffiffiffiffiffi turbulence
regime, the cutoff spatial frequency κm is related to the first Fresnel zone radius by κm ¼ 1:166336 k=L , and after
substitution in (2.423), we obtain:

σ 2χ  0:3075C2n k7=6 L11=6 : ð2:424Þ

84 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Based on Eq. (2.389), we obtain the following expression for scintillation index:

2
 
σ 2I ¼ e4σχ  1 ¼ exp 1:23C 2n k7=6 L11=6  1: ð2:425Þ

In weak turbulence regime, the following series expansion can be used exp(x)  1 + x, so that the Eq. (2.424) simplifies to:

σ 2I  1:23C 2n k7=6 L11=6 ¼ σ 2R , ð2:426Þ

which represents the Rytov variance given by Eq. (2.388).

To extend this theory to all turbulence regimes, we need to use the spatially filtered spectral density as given by Eq. (2.390),
and after substitution into (2.421), we obtain:

ð
1 
sin ðκ 2 L=k Þ
σ 2χ ¼ 2π k L
2 2
1 ½Φn,X ðκÞ þ Φn,Y ðκÞκdκ
κ 2 L=k
0
81
<ð  
sin ðκ2 L=kÞ
0:033C2n κ11=3 eκ =κx κdκ
2 2
¼ 2π k L
2 2
1 ð2:425Þ
: κ L=k
2
0
9
ð
1   11=6 =
sin ðκ2 L=kÞ
þ 1 0:033005391C 2
n κ 2
þ κ 2
y κdκ :
κ2 L=k ;
0

After solving the integrals in (2.421) (for details an interested reader is referred to [30]), we obtain the following solution for
scintillation index, assuming the zero inner scale case [30]:
   
σ 2I ¼ exp 4σ 2χ  1 ¼ exp σ 2log X þ σ 2log Y  1
2 3
6 7
6 7
6 2 2 7 ð2:426Þ
6 0:49σ R 0:51σ R 7
ffi exp 6  þ   7  1:
6 7=6 5=6 7
6 1 þ 1:11σ 12=5 R 1 þ 0:69σ
12=5
R 7
ffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}5
4|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl
σ 2log X σ 2log Y

Based on Eqs. (2.399) and (2.426), we derive the following expressions for parameters in gamma-gamma distribution in terms
of Rytov variance, for zero inner scale:
8 2 3 91 8 2 3 91
>
< >
= >
< >
=
6 0:49σ 2R 7 6 0:51σ 2R
7
α¼ exp 4 7=6 5  1> , β ¼ > exp 4 5=6 5  1> : ð2:427Þ
>
: 12=5 ; : 12=5 ;
1þ 1:11σ R 1 þ 0:69σ R

As an illustration, in Fig. 2.21, we provide the gamma-gamma probability density function with different turbulence
strengths, represented in terms of Rytov standard deviation, for zero inner scale scenario. The case with Rytov standard
deviation σ R ¼ 0.2 (Rytov variance 0.04) corresponds to the weak turbulence regime. The corresponding parameters of
gamma-gamma distribution are α ¼ 51.913 and β ¼ 49.113, while the scintillation index is 0.041. In weak turbulence regime,
gamma-gamma distribution resembles the log-normal distribution. The case with Rytov standard deviation σ R ¼ 1
corresponds to the medium turbulence regime. The corresponding parameters of gamma-gamma distribution are
α ¼ 4.3939 and β ¼ 2.5636, while the scintillation index is 0.70644. Clearly, as the turbulence strength increases, the PDF
skews towards lower values of the irradiance. The case with Rytov standard deviation σ R ¼ 3 (Rytov variance 9) corresponds
to the strong turbulence regime. The corresponding parameters of gamma-gamma distribution are α ¼ 5.485 and β ¼ 1.1156,
while the scintillation index is 1.2421. The distribution function in medium-to-strong turbulence resembles the Rayleigh
2.3 Propagation Effects in Wireless Communication Channels 85

Fig. 2.21 Gamma-gamma

probability density function for
different turbulence strengths

distribution. Finally, the case with Rytov standard deviation σ R ¼ 10 (Rytov variance 100) corresponds to strong-to-saturation
regime. The corresponding parameters of gamma-gamma distribution are α ¼ 14.11 and β ¼ 1.0033, while the scintillation
index is 1.1382. In saturation regime, clearly, the distribution resembles the exponential distribution.

2.3 Propagation Effects in Wireless Communication Channels

In this section we study various signal propagation effects in wireless channels [64–68]. We also describe various wireless
channel models.

2.3.1 Path Loss and Shadowing Propagation Effects in Wireless Communication Channels

The received power in a wireless channel, as illustrated in Fig. 2.22, is affected by attenuation due to following factors:

(i) The path loss, representing the propagation loss average.

(ii) The shadowing effect, occurring due to slow variations in attenuation mostly caused by scattering obstruction structures.
(iii) The multipath fading, representing rapid random fluctuations in received power due to multipath propagation and
Doppler frequency shift.

Transmitted signal (passband) representation in canonical (standard) form:

sðt Þ ¼ sI ðt Þ cos ð2π f c t Þ  sQ ðt Þ sin ð2π f c t Þ, ð2:428:1Þ

where sI(t) and sQ(t) represent in-phase and quadrature signals, while fc is the carrier frequency. The corresponding complex
envelope (low-pass) representation is given by:

sðt Þ ¼ sI ðt Þ þ jsQ ðt Þ: ð2:428:2Þ

86 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

PRx l=vt

PRx/PTx
Multipath
v fading
PTx Shadowing

Path
loss

l=vt
(a) (b)
Fig. 2.22 Time-varying nature of the wireless channel: (a) one particular instance and (b) illustration of path loss, shadowing, and multipath fading
effects against the distance traveled l

The canonical form can be obtained from complex envelope by:

8 9
>
< >
=
j2π f c t
sðt Þ ¼ Re sðt Þ e ¼ sI ðt Þ cos ð2π f c t Þ  sQ ðt Þ sin ð2π f c t Þ, ð2:428:3Þ
>
: |{z} >
;
sI ðt ÞþjsQ ðt Þ

The received signal, when the transmitter and receiver are in close proximity of each other, can be written as:
 j2π f c t

r ðt Þ ¼ Re sðt Þ  hðt Þe þ nðt Þ, ð2:429Þ
where n(t) is the additive white Gaussian noise (AWGN) and h(t) is the channel impulse response.

2.3.1.1 Free-Space Path Loss and Doppler Shift

Path loss definitions in bot linear- and dB-scales are given as:
 
P PTx
PL ¼ Tx , PL ½dB ¼ 10 log 10 : ð2:430:1Þ
PRx PRx

where PTx is the transmit power and PRx is the received power averaged out for variations due to shadowing effect. The path
gain is related to the path loss by:
 
PTx
PG ¼ PL ¼ 10 log 10 : ð2:430:2Þ
PRx

When transmitter or receiver is moving, the received signal will have a Doppler shift associated with it, as illustrated in
Fig. 2.23. The Doppler shift is caused by the fact that transmitter/receiver movement introduces a distance difference dl that
transmitted signal travels between two observations, which can be evaluated by dl ¼ vdt cos ϕ, where ϕ is the angle of arrival
of the transmitted signal relative to the direction of motion. The corresponding phase shift is given by dφ ¼ |{z} k dl ¼
2π=λ
ð2π=λÞvdt cos ϕ: The Doppler shift frequency can be determined as the first derivative of the phase:

1 dφ v
νD ¼ ¼ cos ϕ: ð2:431Þ
2π dt λ
2.3 Propagation Effects in Wireless Communication Channels 87

Fig. 2.23 Illustration of the Doppler effect

Fig. 2.24 Illustration of the illustration of the LOS component and corresponding power law in the far-field. lref: the reference distance

Clearly, the receiver will experience either increase or decrease in frequency depending on the angle of arrival.
Received signal at distance L from transmitter:
 
λ pffiffiffiffiffiffi ej λ L
2π

r ðt Þ ¼ Re GL sðt Þe j2π f c t , ð2:432Þ

4π L
pffiffiffiffiffiffi
where GL is the product of transmit antenna (Tx) and receive (Rx) antenna radiation patterns in the line-of-sight (LOS)
direction, as illustrated in Fig. 2.24. Receive-to-transmit power ratio is given by:
pffiffiffiffiffiffi 2
PRx GL
¼ λ , ð2:433Þ
PTx 4πL

and this equation is commonly referred to as the Friis free-space equation. Received power can be expressed in dB-scale as
follows:

PRx ½dBm ¼ 10 log 10 PRx

ð2:434Þ
¼ PTx ½dBm þ 10 log 10 ðGL Þ þ 20 log 10 ðλÞ  20 log 10 ð4π Þ  20 log 10 ðLÞ:

Free-space path loss in dB-scale:

PTx G λ2
PL ½dB ¼ 10 log 10 ¼ 10 log 10 L 2 : ð2:435Þ
PRx ð4πLÞ
88 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.25 Illustrating the monostatic radar concept

2.3.1.2 Radar Range Equation

At this point, it is convenient to derive the radar range equation, which will be used later on in this section. The basic
operation of the monostatic classical radar, see Fig. 2.25, is to send an electromagnetic (EM) wave towards the target, which
get reflected by the target, and radar return is measured by the receiver.
The range to the target can be determined by R ¼ cΔt/2, where Δt is the transit time (time interval that elapses from the
transmission of the pulse until its return upon reflection). Let GTx denote the gain of transmit antenna and PTx be the power of
transmitter. The power density from uniformly radiating antenna emitting the spherical wave at distance R is given by PTX/
(4πR2). Therefore, the power density at the target will be PTXGTx/(4πR2). The power of reflected signal from the target will be:

PTx GTx
Pt ¼ σ, ð2:436Þ
4πR2

where σ is the radar cross section (RCS) representing the measure of energy that target reflects back towards the radar. The
power density of the reflected signal at receiver side will be:

PTx GTx σ
W Rx ¼ : ð2:437:1Þ
4πR2 4πR2

The power of the reflected signal at receiver side will be:

PTx GTx σARx PTx GTx σARx

PRx ¼ W Rx ARx ¼ ¼ , ð2:437:2Þ
4πR2 4πR2 ð4π Þ2 R4

where ARx is the effective area of receive antenna. This equation is commonly referred to as the radar range equation. If the
medium (air) is not transparent due to presence of clouds and/or fog, we need to introduce the atmospheric attenuation F,
defined as F ¼ exp(αR/2), with α being the attenuation coefficient, as follows:

PTx GTx σARx 4

PRx ¼ F : ð2:437:3Þ
ð4π Þ2 R4

The RCS can be defined more formally by [69, 70]:

jERX j2
σ ¼ lim 4πR2 , ð2:438Þ
R!1 jEt j2

where Et represents the incident electric filed at the target, while ERx denotes the scattered electric field measured at the
receiver side.
2.3 Propagation Effects in Wireless Communication Channels 89

2.3.1.3 Ray Tracing Approximation

In urban/indoor environment, a radio signal transmitted from a fixed source will experience multiple objects that produce
reflected, diffracted, or scattered copies of transmitted signal, known as multipath signal components, as illustrated in
Fig. 2.26.
Ray tracing approximation employs the principles of geometrical optics to describe the wireless channel propagation
effects. Geometry determines received signal from each signal component. This approximation typically includes reflected
rays, scattered rays, and diffracted rays. It requires the knowledge of the site parameters such as the geometry and dielectric
properties of objects. In two-ray model, the received signal consists of two components: the LOS ray and reflected ray, as
shown in Fig. 2.27. The ground reflected ray approximately cancels LOS path above after some characteristic distance, known
as the critical distance lc (to be defined below). The power dependence law is proportional to l2, i.e., at 20 dB/decade, when
the distance between transmit and receive antennas l is larger than the transmitting antenna height hTx but smaller than the
critical distance lc. For distances larger than the critical distance, the power dependence is proportional to l4, i.e., at 40 dB/
decade.
The received signal for the two-ray model (by ignoring surface wave attenuation effect):
   
1 pffiffiffiffiffiffi ejkL pffiffiffiffiffiffiffiffiffi ejkðlþl0 Þ jωc t 2π
r ðt Þ ¼ Re GL sðt Þ þ r refl Grefl sðt  τÞ e ,k ¼ ð2:439Þ
2k L l þ l0 λ

where the frits term corresponds to the LOS ray and the second term to the reflected ray. The delay spread, representing the
difference in arrival times for the LOS and reflected rays, based on Fig. 2.27, is determined by τ ¼ (l + l0  L )/c (c is the speed
of the light in the air). The corresponding phase shift is given by Δϕ ¼ (2π/λ)(l + l0  L ). In Eq. (2.439), we use rrefl to denote
the ground reflection coefficient, which based on Eqs. (2.122), and can be written in the following form:

Fig. 2.26 Illustration of

reflection, diffraction, and
scattering effects
Reflection
Diffraction effects
Scattering

Fig. 2.27 Illustration of the

two-ray channel model. hTx
transmit antenna height, hRx
receive antenna height
90 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

( pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
cos θ  Z εr  sin 2 θ, TE
r refl ¼ , Z¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð2:440Þ
cos θ þ Z εr  sin 2 θ=εr , TM

where εr is the dielectric constant of the round, related to the refractive index by εr ¼ n2. The sign + corresponds to TE
(vertical) polarization while the sign – to the TM (horizontal) polarization state. The gain of LOS ray is determined by
GL ¼ GTx,LGRx,L, while the gain of reflected ray is given by Grefl ¼ GTx,lGRx,l0 .
For narrowband transmitted signals, the delay spread is much smaller than the reciprocal of the signal bandwidth (τ < <1/
Bs), and we can write s(t)  s(t  τ). Under this assumption, the received power is related to the transmit power by:
 2 pffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi 
λ  GL r refl Grefl e jΔϕ 2
PRx ¼ PTx  þ  , Δϕ ¼ k ðl þ l0  LÞ: ð2:441Þ
4π  L l þ l0 

By using the Pythagorean theorem twice (see Fig. 2.27), we determine the LOS and reflected ray paths difference by:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
l þ l0  L ¼ ðhTx þ hRx Þ2 þ D2  ðhTx  hRx Þ2 þ D2 , ð2:442Þ

where D is the horizontal separation of antennas. When D > > hTx + hRx, we can use the following approximation (1 + x)1/
2
 1 + x/2 in (2.442) to obtain:

2hTx hRx
Δϕ  k : ð2:443Þ
D

When Δϕ ¼ π the LOS and reflected ray are out of phase, and the distance for which this happens is known as the critical
distance:

lc ¼ 4hTx hRx =λ: ð2:444Þ

For asymptotically large distance between Ts and Rx antennas for LOS ray L, the following approximations are valid:
l + l’  L  D, θ  π/2, GL  Grefl, and r  1 (TE), so that the received power in linear- and dB-scales can be written as:
 pffiffiffiffiffiffi2  2 pffiffiffiffiffiffi 2
λ GL 4πhTx hRx GL hTx hR
PRx  PTx ¼ PTx ,
4πD λD D2 ð2:445Þ
PRx ½dBm ¼ PTx ½dBm þ 10 log 10 ðGL Þ þ 20 log 10 ðhTx hRx Þ  40 log 10 ðDÞ:

Another relevant model to describe the propagation effects in wireless channels is the dielectric canyon model, also known
as the ten-ray model, developed by Amitay [71], which is illustrated in Fig. 2.28. This model incorporates all rays with one,
two, or three reflections: line-of-sight (LOS), ground-reflected (GR), single-wall (SW) reflected, double-wall (DW) reflected,
triple-wall reflected (TW), wall-ground (WG) reflected, and ground-wall (GW) reflected rays.
The received signal for the ten-ray model can be written as:
( " # )
1 pffiffiffiffiffiffi ejkL X9 pffiffiffiffiffiffi ejkli 2π
jωc t
r ðt Þ ¼ Re GL sðt Þ þ r refl,i Gli sðt  τ i Þ e ,k ¼ ð2:446Þ
2k L i¼1
li λ

DW
… WG LOS Tx
…
Rx SW TW

Fig. 2.28 Illustration of the dielectric canyon model

2.3 Propagation Effects in Wireless Communication Channels 91
pffiffiffiffiffiffi
where rrefl,i is the overall reflection coefficient of the i-th reflected ray, Gli is the product of the transmit and receive radiating
patterns corresponding to the i-th reflected ray, and li is the distance traveled by the i-th reflected ray. The delay spread of the i-
th reflected ray relative to the LOS ray is given by τi ¼ (li  L )/c. Under the narrowband assumption (τ < <1/Bs), s(t)  s(t  τi)
so that the received power in dielectric canyon model can be written as:
 pffiffiffiffiffiffi 2
 2 pffiffiffiffiffiffi X 9
r refl,i Gli e jΔϕi 
1  GL 
PRx ¼ PTx  þ  , Δϕi ¼ kðli  LÞ ð2:447Þ
2k  L i¼1
li 

The power roll-off with distance is proportional to lg, where g 2 [2 to 6].

2.3.1.4 General Ray Tracing (GRT) Model

Another relevant model is so-called general ray tracing (GRT) model [72, 73]. GRT model can be used to predict the electric
field strength and delay spread of any building configuration and antenna placement, providing that building database (height,
location, and dielectric properties) and Tx and Rx locations are accurately specified. The methods of geometric optics, such as
reflection, diffraction, and scattering, as illustrated in Fig. 2.29(a), are used to trace the propagation of different multipath
components. Diffraction occurs on “sharp edges” and is commonly modeled by Fresnel knife-edge diffraction model, as
shown in Fig. 2.29(b).
Here we derive the attenuation due to diffraction from a building or a mountain by modeling the obstruction as the
diffracting knife-edge. Figure 2.29(b) has similarities with the Fig. 2.15. The additional distance traveled of diffracted ray with
respect to the LOS ray can be determined from Fig. 2.29(b), by employing the Pythagorean theorem, as follows:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Δl ¼ DTx þ DRx  ðdTx þ dRx Þ ¼ d 2Tx þ h2m þ d 2Rx þ h2m  dTx  dRx
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
h2m h2
¼ dTx 1 þ 2 þ dRx 1 þ 2m  d Tx  dRx
dTx dRx
! ! ð2:448Þ
ð1þxÞ1=2 1þx=2 1 h2m 1 h2m
’ dTx 1 þ 2 þ d Rx 1 þ 2  d Tx  d Rx
2 dTx 2 d Rx
 
h2 1 1
¼ m þ :
2 d Tx dRx

Fig. 2.29 (a) The methods of

geometric optics are used in GRT
model. (b) The knife-edge
diffraction model
92 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

The corresponding phase shift difference is given by:

  sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi
h2m 1 1 π 2 2 1 1
Δϕ ¼ kΔl ’ k þ ¼ v , v ¼ hm þ ð2:449Þ
2 d Tx d Rx 2 λ d Tx d Rx

where v is so-called Fresnel-Kirchhoff diffraction parameter.

From Sect. 2.2.4 we learned that the wave function at receiving point P can be approximated in the Fresnel regime as:
ð
x2 þy2 1 1 1
ψ P  t ðx, yÞejk 2R dS, ¼ þ , ð2:450Þ
R dTx dRx

where t(x,y) is the aperture transmission function (it can also be called the electric field transfer function), which is 1 above the
knife-edge. The wave function in our case is:

yð2 ð2
w
2
y jπu2
jk 2R
ψP  e dy  e 2 du: ð2:451Þ
y1 w1

By using the Fresnel integral definition:

ðw
jπu2
F ðwÞ ¼ e 2 du ¼ C ðwÞ þ jSðwÞ, ð2:452Þ
0

we can rewrite the Eq. (2.451) as follows:

ð2
w
jπu2
ψP  e 2 du ¼ C ðw2 Þ  C ðw1 Þ þ jfSðw2 Þ  Sðw1 Þg: ð2:453Þ
w1

Therefore, the electric field in the shadow behind the half-plan can be written as:

E0
E diff ¼ ð1 þ jÞF ðvÞ, ð2:454Þ
2

where v is already defined above and E0 is the free-space electric field strength. Alternatively, an approximate method can be
used to describe the received signal diffracted ray as follows:
 
1 pffiffiffiffiffiffi ejkðDTx þDRx Þ jωc t 2π
r ðt Þ ¼ Re αðvÞ Gd sðt  τÞ e ,k ¼ ð2:455Þ
2k DTx þ DRx λ

where α(v) is the knife-edge diffraction path loss relative to the LOS path loss, which can be estimated as [74]:
8
> 20 log 10 ð0:5  0:62vÞ,  0:8
v < 0
>
>

>
> 20 log 10 0:5e0:95v , 0
v < 1
<
 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
αðvÞ ½dB ¼ ð2:456Þ
> 2
> 20 log 10 0:4  0:1184  ð0:38  0:1vÞ , 1
v
2:4
>
>
>
:
20 log 10 ð0:225=vÞ, v > 2:4
2.3 Propagation Effects in Wireless Communication Channels 93

Fig. 2.30 Illustration of the

scattering effect

2.3.1.5 Scattering Effects

The scattering of electromagnetic waves [51, 75, 76] occurs due to the interaction of the incident wave with the obstacle
matter, which results in the remission of the energy of the incident wave into other directions different from the incident wave
direction. The energy of incident wave is not absorbed but rather redistributed to different directions of propagation. The
scattering effects are contributed to the inhomogeneities in a dielectric constant of the obstacle. A simple model to study the
scattering effects is shown in Fig. 2.30. The origin is located inside of the scattering object of volume V. The scattering object
is inhomogeneous in dielectric constant ε ¼ ε(r0 ,t). The position vectors of the transmitter Tx and receiver Rx are denoted by R
and r, respectively. The position of an arbitrary point inside the scattered object is denoted by r0 . The unit vector in direction of
the incident electric field vector is denoted by m b ðr0 Þ, while the unit vector pointing towards the receiver is denoted by b nðr0 Þ.
Let E and H be slow-varying complex electric and magnetic field amplitudes. The corresponding curl and divergence
Maxwell’s equation can be written as:

1 ∂H 1 ∂εE
∇
E ¼ jkH  , ∇
H ¼ jkεE þ , ∇ðεEÞ ¼ 0: ð2:457Þ
c ∂t c ∂t

By taking the curl of ∇
E and substituting ∇
H into such obtained equation, we arrive at:

2
k ∂εE 1 ∂ εE
∇
ð∇
EÞ ¼ k2 εE þ 2j  2 : ð2:457Þ
c ∂t c ∂t 2

By employing the identity ∇
(∇
E) ¼ ∇ (∇E)  ΔE and by rewriting the divergence equation in (2.457) as ∇E ¼
E ∇ε
ε ¼ E∇ð log εÞ, the Eq. (2.457) becomes:

2
k ∂εE 1 ∂ εE
ΔE þ k 2 εE =  ∇ðE∇ð log εÞÞ  2j þ 2 : ð2:458Þ
c ∂t c ∂t 2

Assuming that dielectric constant in the scattering object changes slowly, we can write:

ε ¼ hεi þ δε, ð2:459Þ

where δε is slowly varying perturbation around mean value hεi. For the propagation through atmosphere, we can set the mean
value to be 1, and since |δε| < < hεi, we can use the following approximation log(1 + δε)  δε, so that the Eq. (2.458) becomes:

2
k ∂εE 1 ∂ εE
ΔE þ k2 ð1 þ δεÞE =  ∇ðE∇δεÞ  2j þ 2 : ð2:460Þ
c ∂t c ∂t 2

Given that ε is function of the location in the scattering object, the last two terms in (2.460) can be neglected so that this
equation simplifies to:

ΔE þ k2 E =  k2 δεE  ∇ðE∇δεÞ: ð2:461Þ

94 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Let the wave Ei(r) be incident on the scattering object volume element as illustrated in Fig. 2.30, and this electric field
clearly satisfies the divergence equation div Ei ¼ 0. Since the dielectric constant is slowly varying function of position, we can
use the method of small perturbation, described in Sect. 2.2.5, and represent the electric field vector as:

E ¼ Ei þ Es , ð2:462Þ

where Es is the scattering component of the electric field. After substituting (2.462) into (2.461), by retaining only term linear
in δε, we obtain the following two separate equations for incident field and scattered field:

ΔEs þ k2 Es =  k2 δεEi  ∇ðEi ∇δεÞ, ΔEi þ k2 Ei ¼ 0: ð2:463Þ

By interpreting the term k2δεEi + ∇ (Ei ∇ δε) as the perturbation, the solution of the scattered field component can be found as
the convolution of the perturbation and the Green function G(r, r0) ¼ (1/4π) exp ( jk|r  r0|)/( jk|r  r0|):
ð
1 exp ðjk jr  r0 jÞ 2 
Es ðrÞ ¼ k δεðr0 ÞEi ðr0 Þ þ ∇ðEi ðr0 Þ∇δεðr0 ÞÞ d 3 r0 ð2:464Þ
4π V jr  r0 j
Ð Ð Ð
By employing the Gauss theorem Vu grad v dV + Vv grad u dV ¼ SuvdS, by interpreting u as exp( jk|r  r0|)/( jk|r  r0|)
and v as Ei ∇ δε(r'), we can rewrite the Eq. (2.464) as follows:
ð
1 exp ðjk jr  r0 jÞ 2
E s ðrÞ ¼ k δεðr0 ÞEi ðr0 Þd3 r0
4π V j r  r0 j
ð  
1 exp ðjk jr  r0 jÞ 3 0
 ðEi ðr0 Þ∇δεðr0 ÞÞgradr0 d r ð2:465Þ
4π V j r  r0 j
ð
1 exp ðjk jr  r0 jÞ
þ Ei ðr0 Þ∇δεðr0 Þd 3 S,
4π S j r  r0 j

where S is the surface enclosing the volume V and dS is the vector of magnitude dS directed along the outer normal line to the
surface. The grad term in equation above can be determined as:

exp ð jkjr  r0 jÞ jkexp ð jkjr  r0 jÞjr  r0 j þ exp ð jkjr  r0 jÞ r  r0

gradr0 ½  ¼
jr  r0 j jr  r0 j2 jr  r0 j
|fflfflffl{zfflfflffl}
bnðr, r0 Þ ð2:466Þ
jkexp ð jkjr  r0 jÞjr  r0 j þ exp ð jkjr  r0 jÞ
¼ nðr, r0 Þ:
b
jr  r0 j2

In the far-field, k|r  r0 |> > 1 so that the last term in (2.466) can be neglected as it decays as |r  r0 |3. The surface term in
(2.465) is negligible compared to volume terms for sufficiently large scattering object so that after substitution of (2.466) into
(2.465), we obtain:
ð
1 exp ð jkjr  r0 jÞ 2
Es ðrÞ ffi k δεðr0 ÞEi ðr0 Þd3 r0
4π V jr  r0 j
ð ð2:467Þ
1 exp ð jkjr  r0 jÞ 3 0
þ ðEi ðr0 Þ∇δεðr0 ÞÞjk d r:
4π V jr  r0 j

However, we are still left with another grad term that can be transformed by using Gauss theorem, and after applying div
Ei ¼ 0 and neglect the terms decaying faster than |r  r0 |3, we arrive at the following solution for the scattered electrical field:
ð
k2 exp ð jkjr  r0 jÞ
Es ðrÞ ffi δεðr0 ÞfEi ðr0 Þ  b
nðr, r0 Þðb
nðr, r0 Þ  Ei ðr0 ÞÞgd3 r0 : ð2:468Þ
4π V jr  r0 j
2.3 Propagation Effects in Wireless Communication Channels 95

By using the cross-product identity a
(b
c) ¼ b(a  c)  c(a  b), we can rewrite the Eq. (2.468) in term of cross-
products as follows:
ð
k2 exp ð jkjr  r0 jÞ
Es ðrÞ ffi δεðr0 Þfb
nðr, r0 Þ
½Ei ðr0 Þ
b
nðr, r0 Þgd3 r0 : ð2:469Þ
4π V jr  r0 j

Clearly, the evaluation of scattered electric field requires the knowledge of the spatial variation of dielectric constant, which is
not easy to determine. As suggested by Tatarskii [51], the covariance function of the dielectric constant can be used instead,
defined as:
2 32 3
* +
6 0 76 00 7
Cε ðr , r Þ ¼ 4δεðr Þ  hδεðr Þi54δεðr Þ  hδεðr Þi5 ¼ hδεðr0 Þδεðr00 Þi
0 00 0 00
|fflfflfflffl{zfflfflfflffl} |fflfflfflffl{zfflfflfflffl} ð2:470Þ
¼0 ¼0

¼ Cε ðr0  r00 Þ ¼ C ε ðρÞ,

which is clearly spatially homogeneous. The covariance function Cε is related to the spectral density of the dielectric constant
fluctuations, denoted as Φε(κ), by the following Fourier transform:
ZZZ
jκρ 3
Cε ðρÞ ¼ Φε ðκÞe d ρ: ð2:471Þ

n0 in a solid angle dΩ can be determined as:

The element of scattered energy in direction of b

dW ¼ Si ðr0 Þσ i ðr0 ÞdV 0 dΩ, ð2:472Þ

where Si is the incident wave Poynting vector magnitude and σ i(r) is the effective scattering cross section from a unit volume
b (see Fig. 2.30), determined as [62]:
into a unit solid angle in the direction m

π
b ðrÞ  kb
σ i ðrÞ ¼ k4 Φε ðkm nðrÞÞ sin γ ðrÞ, ð2:473Þ
2

where γ (r) is the angle between Ei(r) and b

nðrÞ. The energy flux density of energy dW scattered in solid angle dΩ is given by
[62]:

dW Si ðr0 Þσ i ðr0 ÞdV 0 dΩ Si ðr0 Þσ i ðr0 ÞdV 0

¼ ¼ , ð2:474Þ
jr  r 0 j2 dΩ jr  r 0 j2 dΩ jr  r 0 j2

where |r  r0|2dΩ is the area of scattered energy distribution. The average Poynting vector at receiver Rx side can be
determined as follows:
ð
Si ðr0 Þσ i ðr0 Þ 3 0
hSðrÞi ¼ nðr0 Þ
b d r: ð2:475Þ
V jr  r0 j2

The total received scattered power can be estimated as PRx  |hS(r)i|2. The scattering theory discussed above is applicable
to both wireless communications and free-space optical communications.
Clearly, the calculation of electrical field strength and power of scattered wave on Rx side is not a trivial task. To simplify
this study, we can proceed as follows. Let the incident wave be a spherical wave Ei(r) ¼ Ei exp ( jk|R  r|)/|R  r|, and after
substitution into (2.468), we obtain:
ð 0 0
k2 e jkjrr j e jkjRr j 0 0
Es ðrÞ ffi b nðr, r0 Þ  E Þgd 3 r0 ,
0 δεðr ÞfEi  nðr, r Þðb
4π 0
V jr  r j jR  r j |fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}i
E i sin γðr0 Þ ð2:476Þ
ð 0
jkjrr j jkjRr j0
k 2
e e
¼ 0 j jR  r0 j
δεðr0 ÞfEi  b
nðr, r0 ÞE i sin γðr0 Þgd3 r0 ,
4π V jr  r

wherein γ ¼ ∢ nb , Ei .
96 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.31 Simplified scattering

model

In the far-field zone, we can use coarse approximation in amplitudes |r  r0|  r, |R  r'|  R and finer approximation in
b so that the previous equation simplifies to:
phases kjr  r0 j  kr  kr0br, kjR  r0 j  kR  kr0 R
ð
k2 e jkðrþRÞ 0b
ejkðr brþr RÞ δεðr0 ÞfEi  b
0
Es ðrÞ ffi nðr, r0 ÞE i sin γðr0 Þgd3 r0 : ð2:477Þ
4π rR V

By taking the power gain antennas definitions GTx and GRx as well the effective (“radar”) cross-sectional area of the
scattering σ object into account, based on radar range Eq. (2.437.2) and Eq. (2.477), the received signal can be written into an
approximate form as follows:
8 9
>
< jk ðrþRÞ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi >
=
1 1 e jωc t 2π
r Rx ðt Þ ffi Re pffiffiffiffiffi sðt  τÞ GTx GRx σ e ,k ¼ ð2:478Þ
>
: 2k 4π rR |fflfflffl{zfflfflffl} >
; λ
Gs

where s(t) is the transmitted signal and τ is delay associated with scattered ray, relative to the LOS ray, given by
τ ¼ (R + r  L )/c. This simplified scattering model is illustrated in Fig. 2.31. This equation is consistent with the radar
technique [76] (see also [64]). It was found in [77] that typical values for effective cross-sectional area in large German cities
range from 4.5 to 55.7 dBm2.
The total received signal in GRT model, obtained as a superposition of a LOS ray, Nrefl reflected rays, Ndiff diffracted rays,
and Nscatt diffusely scattered rays, can be written as:
"
n XN refl pffiffiffiffiffiffi ejkli
1 pffiffiffiffiffiffi ejkL
r ðt Þ ¼ Re crBigg GL sðt Þ þ r refl,i Glk sðt  τ i Þ
2k L i¼1
li
3 9
NP pffiffiffiffiffiffiffiffi
ejkðDTx,m þDRx,m Þ >
>
diff
þ αm ðvÞ Gdm s t  τ j 7 >
> ð2:479Þ
m¼1 D Tx,m þ D Rx,m 7 =
7e jω t 2π
pffiffiffiffiffiffiffiffiffiffiffi 7
c
,k ¼
NP Gsn σ n e jk ðRn þrn Þ 5 >
> λ
þ
scatt
sðt  τn Þ >
>
1=2 Rn r n ;
n¼1 ð4π Þ

2.3.1.6 Simplified Path Loss and Multislope Models

The simplified path loss model, which can be used for general trade-off analysis of various system designs, is suitable for use
when the path loss dominated by reflections. The received power in this model is given by:
 s
lref
PRx ¼ PTx C , l > lref ð2:480Þ
l

where s2 [2 to 8] is the path loss exponent (“the slope”), whose exact value needs to be determined empirically. The referent
distance for antenna far-field lref is typically 1–10 m for indoor applications and 10–100 m for outdoor applications [64], with
exact value to be determined empirically. The corresponding dB-scale form is given by:
2.3 Propagation Effects in Wireless Communication Channels 97

PRx[dBm]
Slope
s1 s2

Slope
s3
s4

log(l1/lref) log(l3/lref) log(l/lref)

log(l2/lref)

Fig. 2.32 Illustration of the multislope model

 
l
PRx ½dBm ¼ PRx ½dBm þ C ½dB  10s log 10 , ð2:481Þ
lref

where C[dB] is the free-space path gain given by C[dB] ¼ 20log10[1/(2klref)]. The two-ray model discussed earlier belongs to
this category.
A better empirical model is a multislope (piecewise linear) model, which is for four slopes illustrated in Fig. 2.32.
A multislope model has multiple segments with the slope of the i-th segment being si (i ¼ 1,2,. . .). For multislope model with
N segments, the N  1 breaking points l1,. . .,lN  1 must be specified. The simplest model from this category is the dual-slope
model, for which the received power in dB-scale is given by:

PTx ðlÞ ½dBm þ C  10s1 log 10 ðl=lref Þ, lref
l
l1
PRx ðlÞ ½dBm ¼ ð2:482Þ
PTx ðlÞ ½dBm þ C  10s1 log 10 ðl1 =lref Þ  10s2 log 10 ðl=lref Þ, l > l1

Two-ray model is in fact a dual-slope model with break point being the critical distance l1 ¼ lc and attenuation slopes being
s1 ¼ 20 dB/decade and s2 ¼ 40 dB/decade. A useful approximation for the dual-slope model, in linear-scale, is given by [78]
(see also [64]):

 s1 "  ðs2 s1 Þq #1=q

P C l l
PRx ¼ Tx , αðlÞ ¼ 1þ , ð2:483Þ
αðlÞ lref l1

where q is the path loss smoothness parameter to be determined empirically.

2.3.1.7 Empirical Path Loss Models

The GRT model is quite difficult to develop in urban areas, in particular in large cities. To solve for this problem, the empirical
path loss models have been developed such as Okumura model [79], Hata model [80], Hata model extension (COST 231)
[81], Hata-Davidson model [82], and Walfisch-Ikegami model [83]. In the empirical models, the path loss at a desired distance
PL(l) is determined by averaging out the local mean attenuation (LMA) measurement results in a given environment (a city,
suburban area, or office building). The purpose of the averaging is to remove the multipath and shadowing components
present in measurements. Okumura model is empirically based on site/frequency-specific model, applicable to frequency
ranges 150–1500 MHz and distances 1–100 km [79]. The Hata model represents the analytical approximation to Okumura
model, and it is valid for distances ranging from 1 km to 20 km. The empirical path loss under Hata model is given by [79, 83]:

PL,Hata ðlÞ ½dB ¼ 69:55 þ 26:16 log 10 ð f c Þ  13:82 log 10 ðhTx Þ  aðhRx Þ
ð2:484Þ
þð44:9  6:55 log 10 ðhTx ÞÞ log 10 ðlÞ  bð f c , environmentÞ,

where hTx (hRx) is the height of transmit (receive) antenna as before, fc [MHz] is the carrier frequency, l [km] is the link length,
and a(hRx) is the receive antenna, carrier frequency, and environment-specific attenuation function:
98 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

( ð1:1log10 ð f c Þ  0:7ÞhRx  ð1:56log10 ð f c Þ  0:8Þ dB,

aðhRx Þ ¼ small to mediumsize cities ð2:485Þ
3:2ðlog10 ð11:75hRx ÞÞ2  4:97 dB, large cities ð f c > 300 MHzÞ

The function b( fc, environment) represents the correction factor to be applied for suburban and rural areas [79, 83]:
(
2½ log 10 ð f c =28Þ2 þ 5:4, suburbanareas
bð f c , environmentÞ ¼ ð2:486Þ
4:78½ log 10 ð f c Þ2  18:33 log 10 ð f c Þ þ C, rural areas

where C ranges from 35.94 in countryside to 40.95 for desert areas [64]. The ITU-R extended the Hata model for distances
ranging from 20 km to 100 km as follows [83]:

PL,Hata‐ITU‐R ðlÞ ½dB ¼ 69:55 þ 26:16 log 10 ð f c Þ  13:82 log 10 ðhTx Þ  aðhRx Þ
ð2:487Þ
þð44:9  6:55 log 10 ðhTx ÞÞð log 10 ðlÞÞg  bð f c , environmentÞ,

where g is the exponent correction factor [83]:

8
>
< 1, l < 20 km
!
g¼ hTx

0:8 ð2:488Þ
:1 þ
> 0:14 þ 0:000187 f c þ 0:00107 log 10 20l , l 20 km
1 þ 7  106 h2Tx

The Hata model was extended by the European Cooperation in the Field of Scientific and Technical Research (EURO-
COST) to 2 GHz [81] by adding the correction factor of +3 dB in Eq. (2.484) for metropolitan centers and restricting the
validity of model to 1.5 GHz < fc < 2GHz, 1 km
l
10 km, 30 m
hTx
200 m, 1 m
hRx
10 m.
The Telecommunications Industry Association (TIA) recommended the modification of the Hata model, now known as
Hata-Davidson model, as follows [83]:

PL,Hata‐Davidson ðlÞ ½dB ¼ PL,Hata ðlÞ ½dB  AðhTx , lÞ  S1 ðlÞ  S2 ðhTx , lÞ  S3 ð f c Þ  S4 ð f c , lÞ, ð2:489Þ

where
8
>
< 0, l < 20 km
  
AðhTx , lÞ ¼ hTx
>
: 0:62137 ð l  20 Þ 0:5 þ 0:15 log 10 , 20 km
l < 300 km
121:92 ð2:490Þ
(
0, l < 64:38 km
S1 ðlÞ ¼
0:174ðl  64:38Þ, 64:38 km
l < 300 km

and

S2 ðhTx , lÞ ¼ 0:00784j log 10 ð9:98=lÞjðhTx  300Þ, hTx > 300 m

S3 ð f c Þ ¼ f c =250 log 10 ð1500= f c Þ, ð2:491Þ
S4 ð f c , lÞ ¼ ½0:112 log 10 ð1500= f c Þðl  64:38Þ, l > 64:38 km

2.3.1.8 Shadowing Effects

A signal transmitted through a wireless channel will typically experience random variations due to attenuation from the
objects in the signal path, giving rise to random the variations of the received power at a given distance, as illustrated in
Fig. 2.33. This effect is commonly referred to as the shadowing effect. Attenuation through the building of total wall depth l is
given by a(l) ¼ exp.( αl), where α is the attenuation coefficient. Attenuation through multiple buildings can be described as:
2.3 Propagation Effects in Wireless Communication Channels 99

Rx Tx

Fig. 2.33 Illustration of the shadowing effect

P
 αi li X
að l t Þ ¼ e i
¼ eαt lt , lt ¼ li , ð2:492Þ
i

where αi is the attenuation coefficient of the i-th building and li is the correspond total wall depth. The corresponding
attenuation in Np-scale is given by:
X
log aðlt Þ ¼  αi li ¼ αt lt ½Np: ð2:493Þ
i

Since different buildings have different dielectric constant and wall depth would be different as well, according to the
central limit theorem, the overall distribution of attenuation in dB-scale is Gaussian:
2  2 3
ψ dB  μψ dB
1 6 7
pðψ dB Þ ¼ pffiffiffiffiffi exp 4 5, ψ dB > 0: ð2:494Þ
2π σ ψ dB 2σ 2ψ dB

where ψ dB ¼ 10log10ψ, ψ ¼ PTRx/PRx, with standard deviation σ 2ψ dB 2 ½4 dB, 13 dB [64, 84, 85]. The mean value is
contributed to the path loss, so that we set μψ dB ¼ 0:
By transformation of random variables, we determine that the distribution of ψ is log-normal:
2  2 3
10 log ψ  μ
10= log 10 6 10 ψ dB 7
pðψ Þ ¼ pffiffiffiffiffi exp 4 5, ψ > 0 ð2:495Þ
2π σ ψ dB ψ 2σ 2
ψ dB

An important parameter in in shadowing is the distance over which the auto-covariance drops down to e1 of its maximum
value, and we refer to this parameter as the decorrelation distance Lc. In other words, from the following equation:
D  E
A ðd Þ ¼ ψ dB ðlÞ  μψ dB ψ dB ðl þ dÞ  μψ dB ¼ σ ψ dB ed=Lc , ð2:496Þ

we can determine the decorrelation distance Lc, which for outdoor systems ranges from 50 to 100 m [64, 86].
The path loss and shadowing effects can be combined as illustrated in Fig. 2.34(a). The very slow variation is associated
with the path loss, while slow fluctuations are associated with the shadowing effect.
The receive-to-transmit power ratio in dB-scale can be determined as:
 
PRx l
ðdBÞ ¼ 10 log 10 C  10s log 10 ref þ ψ dB , ð2:497Þ
PTx l

where the path loss decreases linearly with the log10(l ), s corresponds to the path loss exponent, and ψ dB is a zero-mean
random variable with variance σ 2ψ dB : The corresponding linear-scale model is described as:
100 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.34 (a) Illustration of the

combined path loss and
shadowing effect vs. normalized
distance. (b) Determination of
parameters from the empirical
measurements

 s
PRx l
¼ C ref ψ, ð2:498Þ
PTx l

where transmit-to-receive power ratio ψ is log-normally distributed (2.495). It is possible to determine the parameters of
combined model from empirical measurements as illustrated in Fig. 2.34(b). “Best fit” straight line through measured
attenuation data gives the slope 10s, from which we determine the path loss exponent. The intersection of the line with
y-axis gives us the parameter C [dB]. Form the fluctuation of the measurements points form the straight line, we determine the
variance σ 2ψ dB of the shadow model.

2.3.1.9 Outage Probability and Cell Coverage Area

An important characteristic of the combined path loss and shadowing effects is so-called outage probability. The outage
probability, denoted as Poutage(Pmin,l), under the combined path loss and shadowing, is defined as the probability that the
received power at an observed distance l, PRx(l), falls below the minimum tolerable received power Pmin:

Poutage ðPmin , lÞ ¼ PrðPRx ðlÞ < Pmin Þ

 
Pmin  ðPTx þ 10 log 10 C  10s log 10 ðl=lref ÞÞ ð2:499Þ
¼1Q ,
σ ψ dB

where the Q-function is defined as:

ð
1
1

QðuÞ ¼ PrðX > uÞ ¼ pffiffiffiffiffi exp y2 =2 dy, ð2:500Þ

2π
u

and it is related to the complementary error function by:

  ð
1
1 u 2

QðuÞ ¼ erfc pffiffiffi , erfcðuÞ ¼ pffiffiffi exp y2 dy: ð2:501Þ

2 2 π
u

Another relevant characteristic of the combined path loss and shadowing effect model is so-called cell coverage area,
illustrated in Fig. 2.35. In the figure, the contours of constant received power corresponding to both combined path loss and
average shadowing and combined path loss and random shadowing are shown. Clearly, the contour of constant received
power for combined path loss and average shadowing is the circle of radius R, while the corresponding contour for combined
path loss and random shadowing is an amoeba-like, which changes its shape over the time. The cell coverage area in a cellular
system is defined as the expected percentage of locations within a cell, in which the received power is above the minimum
tolerable received power Pmin. Let PdS denote the probability received power is larger than minimum power in an surface
element dS, that is, PdS ¼ Pr(PRx(r) > Pmin in dS) ¼ Pr(PRx(r)  Pmin > 0 in dS). By introducing the indicator function as
follows I ðxÞ¼ 1, for x > 0, we can write this probability in terms of the indicator function as:
2.3 Propagation Effects in Wireless Communication Channels 101

Fig. 2.35 Illustration of the cell coverage area

PdS ¼ PrðPRx ðr 0 Þ  Pmin > 0 in dS Þ ¼ hI ðPRx ðr 0 Þ  Pmin > 0 in dS Þi

¼ 1  PrðPRx ðr 0 Þ  Pmin < 0 in dS Þ ¼ 1  Poutage ðPmin ðr 0 Þ, lÞ
  ð2:502Þ
P  ðPTx þ 10 log 10 C  10s log 10 ðr 0 =lref ÞÞ
¼ Q min :
σ ψ dB

Now the cell coverage area can be calculated as:

ð ð
* I ½PRx ðr 0 Þ > Pmin in dSdS + hI ½PRx ðrÞ > Pmin in dSidS
C¼ cell area
¼ cell area
πR2 πR2

2ðπ ð
R ð ð ðR
2π
ð 0 0
R PdS r dr dθ ½1  Poutage ðPmin ðr 0 Þ, lÞr 0 dr 0 dθ ð2:503Þ
PdS dS
0 0 0 0 0
¼ ¼
cell area
¼
πR2 πR2 πR2
ð
2π P  ðPTx þ 10log10 C  10slog10 ðr 0 =lref ÞÞ 0 0
¼ 2 R Qð min Þr dr
πR σ ψ dB
0

After introducing the following notation [64]:

Pmin  PRx ðRÞ 10s log 10 ðeÞ

PRx ðRÞ ¼ PTx þ 10 log 10 C  10s log 10 ðR=lref Þ, c1 ¼ , c2 ¼ , ð2:504Þ
σ ψ dB σ ψ dB

we can rewrite Eq. (2.503) in a compact form:

ðR  
2 r0 0 0
C ¼ 2 Q c1 þ c2 log r dr : ð2:505Þ
R R
0

As an illustration, when the average power at distance R is equal to the minimum

tolerable power, the outage probability
will be 0.5 and c1 ¼ 0, so that the cell coverage area becomes C ¼ 0:5 þ exp 2=c22 Qð2=c2 Þ: The large cell coverage area of a
given cell will mean that this cell will introduce the interference to the neighboring cells, and in practice we need to
compromise between the cell coverage area and the interference.
102 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

2.3.2 Statistical Multipath Wireless Communication Channel Models

2.3.2.1 Multipath Channel Models

The term fading is used to describe the fluctuations of received power in time. It is caused by the combined defects of
multipath propagation effect, see Fig. 2.36, and relative motion between transmitter and receiver, which introduces time-
varying variation in received power. Because of multipath propagation effect, when a single pulse is transmitted over the
multipath channel, the received signal will appear as a pulse train rather than a single pulse. In this particular example, the
receiver moves at speed v, and at different time instances, the position of neighboring buildings changes, so that the receiver
will experience variations in received power. This time-varying nature of the multipath channel is illustrated in Fig. 2.37.
Therefore, the number of multipath components is random, while the i-th multipath component can be characterized by
random amplitude ai(t), random delay τi(t), random Doppler frequency νDi [Hz], and random phase shift
ϕi(t) ¼ 2πfcτi(t)  2πνDit  ϕ0. This time-varying nature of a linear wireless channel is described by the time-varying impulse
response, denoted as h(τ,t); here τ is so-called delay spread, and t is the time instance. The delay spread is defined as the time
difference in signal arriving times of the last and first arriving (LOS) multipath components, which is illustrated in Fig. 2.36.
Let s(t) represent the low-pass representation of the passband signal, as discussed in previous section. The corresponding
passband representation of the transmitted signal is determined by sðt Þ ¼ Re fsðt Þe jωc t g ¼ sI ðt Þ cos ðωc t Þ  sQ ðt Þ sin ðωc t Þ,
where ωc is the angular carrier frequency [rad/s], while sI and sQ are in-phase and quadrature components. The signal at
receiver side originating from the i-th multipath component can be written as s(t  τi)ai(t)exp.[jϕi(t)]. The received signal
r(t), originating from all multipath components, can be determined as:
(" N ð t Þ # )
  X  
jωc t jϕi ðt Þ jωc t jωc t
r ðt Þ ¼ Re xðt Þe ¼ Re ai ðt Þsðt  τi ðt ÞÞe e ¼ Re ½hðτ, t Þ  sðt Þe , ð2:506Þ
i¼0

Fig. 2.36 Illustration of the multipath fading (multipath propagation effect and Doppler shift) and delay spread

Fig. 2.37 Illustration of the time-varying nature of multipath channel

2.3 Propagation Effects in Wireless Communication Channels 103

where the time-varying impulse response is given by:

X
N ðt Þ
hðτ, t Þ ¼ ai ðt Þejϕi ðtÞ δðt  τi ðt ÞÞ, ð2:507Þ
i¼0

and we say that channel is time-selective and frequency-selective. The time selectivity indicates that impulse response changes
in time. Given that the output signal can be written in terms of convolution of the input signal and time-varying impulse
response, as given by Eq. (2.506), different frequency (multipath) components will experience different fading, and we say
that such channel is frequency-selective channel. Based on Eq. (2.506), we derive the multipath channel model, shown in
Fig. 2.38(a).
When fcτi(t)> > 1, which is typical in both indoor and outdoor wireless communication systems, a small change in the path
delay τi(t) will lead to a significant phase change 2πfcτi(t)  ϕDi  ϕ0. Since the phase changes rapidly, a uniform phase

Fig. 2.38 (a) The multipath

channel model. (b) The equivalent
multipath channel model
104 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.39 An illustrative example of a nonresolvable path

Fig. 2.40 The modified

multipath channel model
accounting for nonresolvable
cluster components

distribution is commonly used as random phase model. This rapid phase changes in different multipath components are
responsible for either constructive or destructive interference on receiver side; in other words, this effect introduces the
received power fluctuations, that is, the fading effect. The impact of multipath fading on the received signal is strongly
dependent on the relationship of the spread of time delays associated with the LOS and different multipath components with
the inverse of the signal bandwidth 1/Bs. To explain this better, we provide in Fig. 2.38(b) an equivalent multipath channel
model, where the delay spread of the i-th multipath component is defined with respect to the (i  1)-th one by
Δτi(t) ¼ τi(t)  τi1(t). The minimum time separation between the two pulses, such that they can be resolved at the filter
output, of bandwidth B, is Δτ > 1/B. When the channel delay spread is small relative to the inverse of signal bandwidth, then
the multipath components are nonresolvable, and we say that the corresponding channel model is the narrowband channel
model. One illustrative example of nonresolvable path is shown in Fig. 2.39. Cleary, the nonresolvable multipath component
is composed of reflected components from the cluster of objects in close proximity of each other. Namely, the maximum delay
spread among the reflected components in the cluster is smaller than the inverse of the signal bandwidth so that we can write
maxm,n|τn(t)  τm(t)| < <1/Bs.
To account for the nonresolvable cluster component, we need to modify the channel model as shown in Fig. 2.40. In this
modified multipath channel model, the i-th component represents a combined effect from multiple nonresolvable paths, and
corresponding channel coefficient is given by ai ðt Þejϕi ðtÞ , where the magnitude and the phase are averaged out among
different nonresolvable cluster components.
On the other hand, when the maximum delay spread among the reflected components in the cluster is larger than the inverse
of the channel bandwidth, that is, maxm,n|tn(t)  tm(t)| > 1/B, we say that the channel is wideband channel.
2.3 Propagation Effects in Wireless Communication Channels 105

Fig. 2.41 The narrowband

multipath channel model

2.3.2.2 Narrowband Fading Model

In narrowband channel model, the maximum delay spread between any two components is smaller than the inverse of signal
bandwidth, and we can use the following approximation s(t)  s(t  τi(t)), for all i. The received signal is now given by:
(
" N ðt Þ # )
  X
jωc t jϕi ðt Þ jωc t
r ðt Þ ¼ Re xðt Þe ¼ Re sðt Þ ai ðt Þe e , ð2:508Þ
i¼0

and corresponding narrowband channel model is shown in Fig. 2.41. Since the spreading in time is negligible in narrowband
model, there is no channel distortion by intersymbol interference as different pulse from multipath fading stay in the same
pulse duration. In the extreme narrowband case, we can even write that s(t) ffi 1. The received signal in the extreme
narrowband case can be written as:
(" N ð t Þ # )
  X
jωc t jϕi ðt Þ jωc t
r ðt Þ ¼ Re xðt Þe ¼ Re ai ðt Þe e ¼ r I ðt Þ cos ðωc t Þ  r Q ðt Þ sin ðωc t Þ, ð2:509Þ
i¼0

where the in-phase and quadrature components are given as:

N ðt Þ
X N ðt Þ
X
r I ðt Þ ¼ ai ðt Þ cos ϕi ðt Þ, r Q ðt Þ ¼  ai ðt Þ sin ϕi ðt Þ; ϕi ðt Þ ¼ ωc τi ðt Þ  ϕDi  ϕ0 : ð2:510Þ
i¼0 i¼0

If the number of multipath components N(t) is large, and magnitude an(t) and phase ϕn(t) are independent, then rI(t) and
rQ(t) are jointly Gaussian random processes according to the central limit theorem. The autocorrelation function of in-phase
component is given by:
* +
X 
RrI ðt, t þ τÞ ¼ hr I ðt Þr I ðt þ τÞi ¼ ai
2
cos ϕi ð t Þ cos ϕi ð t þ τ Þ
i
|ffl{zffl} |fflfflfflffl
ffl{zfflfflfflfflffl}
2π f c τi 2π f Di tϕ0 2π f c τi 2π f Di ðtþτÞϕ0
2 3
X 61 
 1 
7 ð2:511Þ
¼ a2n 4 cos 2π f Di τ þ cos 4π f c τi  4π f Di t  2ϕ0 5
i
2 2 |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
¼0
 
1 X 2 
 1 X 2  v
¼ ai cos 2π f Di τ ¼ ai cos 2π τ cos θi :
2 i 2 i λ
106 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.42 Clark-Jakes uniform scattering (US) model

Clearly, the autocorrelation function of the in-phase component is not dependent on time origin t, and we can write
RrI ðt, t þ τÞ ¼ RrI ðτÞ, meaning that the process is wide-sense stationary (WSS). It can be shown in similar fashion that the
autocorrelation function of the quadrature component is equal to that of the in-phase component. The cross-correlation
function of the in-phase and quadrature components is given by:
 
  1 X 2  v
RrI ,rQ ðt, t þ τÞ ¼ RrI ,rQ ðτÞ ¼ r I ðt Þr Q ðt þ τÞ ¼  ai sin 2π τ cos θn
2 i λ ð2:512Þ
 
¼  r Q ðt Þr I ðt þ τÞ ¼ RrQ ,rI ðτÞ:

Finally, the autocorrelation function of the passband signal is given as follows:

Rr ðt, t þ τÞ ¼ Rr ðτÞ ¼ hr ðt Þr ðt þ τÞi ¼ RrI ðτÞ cos ð2π f c τÞ þ RrI ,rQ ðτÞ sin ð2π f c τÞ: ð2:513Þ

2.3.2.3 Clark-Jakes Model

To simplify the WSS model further, the Clark [87] proposed so-called uniform scattering (US) model, shown in Fig. 2.42,
which was further developed by Jakes [84]. In this model, which we will refer to as Clark-Jakes model, the channel consists of
many densely packed scattering objects, with the angle of arrival of the i-th multipath
 component being θi ¼ iΔθ, where
Δθ ¼ 2π/N and the corresponding average power per scattering object being a2i ¼ 2PRx =N, where PRx is the total received
power. After substituting the expression for the average power per scattering object into Eq. (2.511), we obtain the following
expression for the autocorrelation function of the in-phase signal:

  N¼2π=Δθ  
PRx X z}|{ PRx X
N
v v
R r I ðτ Þ ¼ cos 2π τ cos ðiΔθÞ ¼ cos 2π τ cos ðiΔθÞ Δθ: ð2:514Þ
N i λ 2π i¼1 λ

In the limit as N ! 1, the summation in (2.514) becomes integral:

2ðπ
 
P v
RrI ðτÞ ¼ Rx cos 2π τ cos θ dθ ¼ PRx J 0 ð2πνD τÞ, ð2:515Þ
2π λ
0
2.3 Propagation Effects in Wireless Communication Channels 107
Ðπ
where J 0 ðxÞ ¼ ð1=π Þ 0 ejx cos θ dθ is the Bessel function of the first kind and zero order, as introduced earlier in this chapter.
The cross-correlation function of the Clark-Jakes US model is clearly zero, that is:
ð  
PRx v
RrI ,rQ ðτÞ ¼ sin 2π τ cos θ dθ ¼ 0: ð2:516Þ
2π λ

Since the first zero of J0(x) is 2.4048, we conclude that the in-phase component decorrelates after νDτ ¼ 2.4048/
2π ffi 0.382. In other words, if we separate two receive antennas by approximately 0.4λ, they can be considered as independent
of each other.
The power spectral densities (PSDs) are obtained by calculating the Fourier transforms of corresponding autocorrelation
functions. The PSD of in-phase and quadrature components is given by:
8
> 2P 1
< Rx qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi , j f j
f D
πfD
PSDrI ð f Þ ¼ PSDrQ ð f Þ ¼ FT fRrI ðτÞg ¼ 1  ð f = f D Þ2 ð2:517Þ
>
:
0, otherwise

Clearly, as f tends to fD, the PSD tends to infinity, but this never occurs in practice. To solve for this problem, the PSD should
be truncated for f/fD ! 1. The PSD of the passband signal is related to the PSD of in-phase component as follows:

1
PSDr ð f Þ ¼ ½PSDrI ð f  f c Þ þ PSDrI ð f þ f c Þ
8 2
>
> PRx 1
< π f qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi , j f f c j
f D ð2:518Þ
D
¼ 1  ðj f f c j= f D Þ2
>
>
:
0, otherwise

The PSDs are suitable to build a simulator based on Clark-Jakes US model, by using zero-mean unit variance Gaussian
random generator. By setting the double-sided PSD of an additive white Gaussian noise (AWGN) by N0/2 ¼ 2PRx/(πfD), we
need to determine the filter of transfer function H( f ) such that the PSD of the filter output is equal to (2.517), in other words:

2PRx 2P 1
jH ð f Þj2 ¼ Rx qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð2:519Þ
πfD πfD
1  ð f = f D Þ2

Clearly, the magnitude of transfer function of such filter should be chosen as:

h i1=4
jH ð f Þj ¼ 1  ð f = f D Þ2 : ð2:520Þ

2.3.2.4 Envelope and Power Distributions in Narrowband Models

Fluctuations in the envelope of the received signal are study of the fluctuations of the power of the received signal, in other
words the study of fading effects. The received signal for narrowband channel model can be represented in either Cartesian or
polar coordinates as follows:

r ðt Þ ¼ r I ðt Þ cos ðωc t Þ  r Q ðt Þ sin ðωc t Þ ¼ ρðt Þ cos ½ωc t þ ψ ðt Þ, ð2:521Þ

where the envelope and phase in polar coordinates are related to the in-phase and quadrature components by:
108 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.43 Transformation of in-phase and quadrature components into envelope and phase coordinates

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  
1 r Q ðt Þ
ρðt Þ ¼ jr ðt Þj ¼ r 2I ðt Þ þ r 2Q ðt Þ, ψ ðt Þ ¼ tan : ð2:522Þ
r I ðt Þ

Since the samples in-phase and quadrature components are independent and Gaussian distributed, the corresponding joint
distribution will be:
" #
1 r 2I þ r 2Q
f RI ,RQ ðr I , r Q Þ ¼ exp  : ð2:523Þ
2πσ 2 2σ 2

Now we can use the method of transformation of random variables, illustrated in Fig. 2.43, to determine the joint
distribution of envelope and phase. After the following substitutions rI ¼ ρ cos ψ, rQ ¼ ρ sin ψ, the joint distribution of
envelope and phase is given as:

f R,Ψ ðρ, ψ Þ ¼ f RI ,RQ ðr I , r Q ÞjJ ðρ, ψ Þj, ð2:524Þ

where the Jacobian is determined by:

 
 ∂r I ∂r I 
 
 ∂ρ ∂ψ 
jJ ðρ, ψ Þj ¼  ¼ρ
 ∂r Q ∂r Q 
 
∂ρ ∂ψ

which after substitution in (2.524), it yields to:

   
ρ ρ2 1 ρ ρ2
f R,Ψ ðρ, ψ Þ ¼ exp  2 ¼  exp  2 : ð2:524Þ
2πσ 2 2σ 2π σ 2
|{z} 2σ
|fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl}

f Ψ ðψ Þ ρ ρ2
pR ðρÞ¼ 2 exp 
σ 2σ 2

Therefore, the distribution of phase is uniform:

ð
1
1
f Ψ ðψ Þ ¼ f R,Ψ ðρ, ψ Þdρ ¼ , 0
ψ
2π, ð2:525Þ
2π
0

while the distribution of envelope is Rayleigh:

2ðπ  
ρ ρ2
f R ð ρÞ ¼ pR,Ψ ðρ, ψ Þdψ ¼ 2 exp  2 : ð2:526Þ
σ 2σ
0
2.3 Propagation Effects in Wireless Communication Channels 109
P  2
Since there is no dominant LOS component, the total power in multipath components is given by PRx ¼ i ai ¼ 2σ ,
2

and we can represent the Rayleigh distribution as follows:

2ρ ρ2
pR ðρÞ ¼  exp ð  Þ, ρ 0: ð2:527Þ
PRx PRx

The corresponding power distribution is exponential, obtained after the substitution p ¼ ρ2(t) ¼ |r(t)|2, as given below:

1 p 1 p
pR2 ðxÞ ¼ exp ð 2 Þ ¼  exp ð  Þ, p 0: ð2:528Þ
2σ 2 2σ PRx PRx

In the presence of a strong LOS component, denoted as scosωct, the received signal can be represented as:

r ðt Þ ¼ s cos ðωc t Þ þ r I ðt Þ cos ðωc t Þ  r Q ðt Þ sin ðωc t Þ ¼ ½s þ r I ðt Þ cos ðωc t Þ  r Q ðt Þ sin ðωc t Þ: ð2:529Þ

The corresponding envelope and phase are given by:

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  
1 r Q ðt Þ
ρðt Þ ¼ jr ðt Þj ¼ ½s þ r I ðt Þ2 þ r 2Q ðt Þ, ψ ðt Þ ¼ tan : ð2:530Þ
s þ r I ðt Þ

By following the similar procedure of transformation of random variable, we found that the distribution of the phase is again
uniform, while the distribution of envelope is Rician:
ð
ρ ρ2 þ s 2 ρs 1 2π
f R ðρÞ ¼ 2 exp ð ÞI 0 ð 2 Þ, I 0 ðxÞ ¼ exp ðxcos ψÞdψ, ð2:531Þ
σ 2σ 2 σ 2π
0

where I0(x) is the modified Bessel function of the first kind and zero order. As an illustration, in Fig. 2.44 we show normalized
Rician distribution for different values of parameter a ¼ ρ/σ. Clearly, for a ¼ 0, Rician distribution reduces down to Rayleigh.
When parameter a is large, the distribution is similar to Gaussian.

Fig. 2.44 Normalized Rayleigh

and Rician distributions
110 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .
P  2
The total power in multipath component different from the LOS component is 2σ 2 ¼ i,i6¼0 ai , while the total power is
given by:

ð
1

PRx ¼ ρ2 f R ðρÞdρ ¼ s2 þ 2σ 2 : ð2:532Þ

0

The Rician distribution can also be expressed in terms of Rician parameter K, defined by K ¼ s2/(2σ 2), as follows:

  sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
2ρðK þ 1Þ ðK þ 1Þρ2 K ðK þ 1 Þ
f R ð ρÞ ¼ exp K  I 0 2ρ , ρ 0: ð2:533Þ
PRx PRx PRx

The Rician parameter can be used as a measure of the fading severity: the small K implies severe fading, a large K implies
mild fading, and when K tends to infinity, there is no fading at all.
More general fading distribution is the Nakagami distribution, given by??:
 
2mm ρ2m1 mρ2
f R ð ρÞ ¼ m exp  , m 0:5 ð2:534Þ
ΓðmÞPRx PRx

The parameter m in Nakagami distribution represents the fading figure, defined as the receive power-to-variance of z2 ratio.
Pm 2
1=2
For integer values of m, fR(ρ) is the PDF of the following random variable Y ¼ i¼1 X i , where Xi are independent
Rayleigh-distributed random variables. The special cases of Nakagami fading are:

• m ¼ 1: Rayleigh fading,
• m ¼ (K + 1)2/(2 K + 1): Rician fading,
• m ¼ 1/2: single-sided Gaussian distribution,
• m- > 1: No fading, the constant fading.

The power distribution for Nakagami fading:

 m  
m pm1 mp
f R2 ðxÞ ¼ exp  : ð2:535Þ
PRx ΓðmÞ PRx

Another relevant fading model is α-μ generalized fading model, with the following distribution [88–91]:
 
αμμ ραμ1 ρα
f R ð ρÞ ¼ exp μ , ð2:536Þ
ραμ ΓðμÞ
b ρα
b

where μ is the inverse of the normalized variance of ρα, that is, μ ¼ 1/Var(ρα)/hραi2, and b ρ is the α-root mean value of ρα:
α 1=α
b
ρ ¼ hρ i : Various fading models such as Rayleigh, Nakagami-m, exponential, Weibull, and one-sided Gaussian
distributions are just spatial cases of this distribution. As an illustration, by setting α ¼ 2 and μ ¼ 1, the Rayleigh distribution
is obtained. On the other hand, by setting α ¼ 2 and μ ¼ 2, the Nakagami-2 distribution is obtained.

2.3.2.5 Average Fade Duration

Based on statistical models above, we can predict the level of fading, but we cannot predict the duration of the fade. This
parameter is very important as it describes how many symbols get affected by the deep fade effect, and we refer to this
parameter as the average fade duration (AFD). The AVD is defined as the average time for which the signal is below a given
target envelope level R. The envelope level R is typically related to the minimum tolerable power Pmin, discussed in previous
section. Alternatively, it can be related to the minimum tolerable signal-to-noise ratio. Let Tobs denote the observation time
2.3 Propagation Effects in Wireless Communication Channels 111

interval. The time intervals for which the signal is below target envelop value R are denoted as ti (i ¼ 1,2,..). The probability
that the envelope is below target value R can be determined as:

1 X
Prðρðt Þ < RÞ ¼ ti : ð2:537Þ
T obs i

In order to calculate the AFD, we need first to determine the level R crossing rate, denoted as LR [number of crossings/s]. The
level crossing rate is defined as the expected rate at which the signal envelope crosses the envelope level R with negative
slope. The AFD can now be defined mathematically as:

LRP
T obs
1
T obs ti
i¼1 Prðρðt Þ < RÞ
tR ¼ ¼ : ð2:538Þ
LR LR

In order to determine the level crossing rate, the first step is to determine the joint distribution of envelope ρ and its
derivative dρ/dt, denoted as f ðρ, ρ_ Þ, ρ_ ¼ dρ=dt: For details of derivation of this joint distribution, an interested reader is
referred to [92–94]. The level crossing rate can now be determined as the expected value of the envelope slope, where
averaging is performed over negative slopes:

ð0
LR ¼ _ ðρ, ρ_ Þdρ:
ρf _ ð2:539Þ
1

Number of crossing in observed interval Tobs is simply NR ¼ LRTobs. Under US assumption, the envelope and envelope slope
are independent, and the distribution of envelope slope is zero-mean Gaussian:

ρ_ 2
1  2
f ðρ_ Þ ¼ pffiffiffiffiffi e ρ_ , ð2:540Þ
2σ

σ ρ_ 2π

where the variance of the slope is given by [93] σ 2ρ_ ¼ ð2πνD Þ2 σ 2 =2:The level crossing rate for Rician distribution would
be then:

ð0
LR ¼ _ ðρ_ Þf ðρÞdρ_
ρf
1
ð0 2
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
1 1  ρ_ 2 2ρ ð K þ 1 Þ K
ðKþ1Þρ2
K ð K þ 1Þ
¼ pffiffiffiffiffi e ρ_ d ρ_ 2 e I 0 2ρ
2σ
PRx
2 σ ρ_ 2π PRx PRx ð2:541Þ
1
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
pffiffiffiffi
σ ρ_ = 2π
  sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ρ ðK þ 1Þρ2 K ð K þ 1Þ
¼ 2π ðK þ 1ÞνD pffiffiffiffiffiffiffi exp K  I 0 2ρ ,
PRx PRx PRx
pffiffiffiffiffiffiffi
which after substitution ρn ¼ ρ= PRx can be rewritten in compact form as follows:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
LR ¼ 2π ðK þ 1ÞνD ρn exp K  ðK þ 1Þρ2n I 0 2ρn K ðK þ 1Þ : ð2:542Þ

For Rayleigh distribution, K ¼ 0, and the level crossing rate is simply:

pffiffiffiffiffi

LR ¼ 2π νD ρn exp ρ2n : ð2:543Þ

112 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

The cumulative distribution function (CDF) for Rayleigh fading is given by:

CDFðρÞ ¼ Prðρ < RÞ ¼ 1  eρ =PRx ¼ 1  eρn ,

2 2
ð2:544Þ

so that the average fade duration becomes:

ðRayleighÞ Prðρðt Þ < RÞ 1  exp ρ2n exp ρ2n  1

tR ¼ ¼ pffiffiffiffiffi
¼ pffiffiffiffiffi : ð2:545Þ
LR 2π νD ρn exp ρ2n 2π νD ρn

On the other hand, the CDF for Rician fading is given by:
ðR  
ðRicianÞ
pffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
CDFðρÞ ¼ Prðρ < RÞ ¼ fR ðρÞdρ ¼ 1  Q 2K , 2ðK þ 1Þρ2n , ð2:546Þ
0

where Q(a,b) is the Marcum Q-function, defined as:

ðb

Qða, bÞ ¼ 1  z exp  z2 þ a2 =2 I 0 ðazÞdz: ð2:547Þ

0

After substitution of (2.546) into (2.538), we obtain the following expression for AFD of Rician fading:
pffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ðRicianÞ Prðρðt Þ < RÞ 1  Q 2K , 2ðK þ 1Þρ2n

tR ¼ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
: ð2:548Þ
LR 2π ðK þ 1ÞνD ρn eKðKþ1Þρn I 0 2ρn K ðK þ 1Þ
2

When AFD is much longer than a bit duration, a long burst of errors will be introduced, indicating that fading is slow.
Therefore, AFD can be used to determine the number of bits affected by a deep fade. Error correction codes developed for
AWGN channels will most probably fail as they need to deal with both random and burst errors. To solve for this problem,
interleaving is typically used so that the burst errors get spread to multiple codewords. When AFD is roughly equal to a bit
duration, a single bit error gets introduced by fading. In this case, a regular error correction code, designed for random errors,
is applicable. Finally, when AFD is much shorter than a bit duration, the fading is averaged out, which indicates that in this
case fading is fast and the Doppler effect may be beneficial in this case.

2.3.2.6 Discrete-Time (DT) Model

When all multipath components are resolvable, the discrete-time (DT) model due to Turin can be used [95]. The N multipath
components are resolvable if for m 6¼ n, the following is |τn(t)  τm(t)|> > 1/B, where B is the transmission bandwidth. This
model is particularly suitable for spread-spectrum systems. The continuous-time output of such a channel is given by:

X
N
xð t Þ ¼ αi ðt Þejϕi ðtÞ sðt  τi Þ: ð2:549Þ
i¼0

By sampling at t ¼ nTs (Ts is the sampling interval), we obtain:

X
N X
N
xðnT s Þ ¼ αi ðnT s Þejϕi ðnT s Þ sðnT s  iT s Þ ¼ αi ðnÞejϕi ðnÞ sðn  iÞ: ð2:550Þ
i¼0
|fflfflffl{zfflfflffl} |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl} i¼0
αi ðnÞ sðniÞ

Based on previous expression, we conclude that DT impulse response is given by:

X
N
hð nÞ ¼ hði, nÞδðn  iÞ, hði, nÞ ¼ αi ðnÞejϕi ðnÞ : ð2:551Þ
i¼0
2.3 Propagation Effects in Wireless Communication Channels 113

Fig. 2.45 Illustration of the discrete-time model

The corresponding DT model is illustrated in Fig. 2.45. The DT model can be further simplified as follows [95]. The
channel use is discretized into MT intervals of duration T each, such that MT is longer than average delay spread. The binary
indicator function is then used in each interval of duration T. Whenever the binary indicator function output is 1, the random
amplitude and phase for a given multipath component are generated from an empirical distribution. This empirical distribution
may belong to the same family of fading distributions but with different parameters.

2.3.2.7 Finite-State Markov Channel (FSMC) Model

Another relevant channel model is the finite-state Markov channel (FSMC) model due to Wang and Moayeri [96], in which the
fading is approximated as a discrete-time Markov process with a time being discretized by a symbol duration. The FSMC
model divides the range of all possible fading powers p into discrete regions as follows RGi ¼ {p| Ai
p < Ai + 1} (i ¼ 1,2. . .),
where Ai and Ai + 1 are boundaries of the i-th region. The stationary probabilities of the regions are determined by:

π i ¼ Prðp 2 RGi Þ ¼ PrðAi

p
Aiþ1 Þ, ð2:552Þ

and can be straightforwardly derived from the cumulative distribution function of the power for each particular fading model.
The transition probabilities between regions are determined by [96]:

Liþ1 T LT
pi,iþ1 ¼ , pi,i1 ¼ i , pi,i ¼ 1  pi,iþ1  pi,i1 , ð2:553Þ
πi πi

where Li is the level Ai crossing rate and T is the symbol duration. Clearly, according to the FSMC model, from the Region
RGi, we can stay in the same region with probability pi,i, move to one region below with probability pi,i-1, or move one region
above the observed region pi,i + 1. Since we can move only to neighboring regions, this model is accurate for sufficiently small
normalized Doppler frequency νDT.

2.3.2.8 Wideband Channel Models

In wideband channel models, the components in a given cluster are resolvable. Let τMDS denote the multipath delay spread
(MDS), defined as τMDS ¼ τmax  τmin (τmax is the maximum delay spread and τmin is the minimum delay spread). If Bs is
transmitted bandwidth, the maximum number of resolvable components is given by τMDSBs. Now since the delay between any
two components in a cluster is much longer than reciprocal of the signal bandwidth, then different multipath components from
a pulse transmitted in a given time slot arrive in neighboring time slots introducing the intersymbol interference. In other
words, the pulse of duration T results in a received pulse of duration T + τMDS. Since the components with each cluster are
resolvable, the received signal can be represented as follows:
114 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.46 The wideband channel model in which the impulse response of the i-th cluster is denoted as hi(τi,t)

("
X X
rðtÞ ¼ Re sðt  τ0 ðtÞÞ ai ðtÞejϕi ðtÞ þ sðt  τ1 ðtÞÞ ai ðtÞejϕi ðtÞ
i2C 0 i2C 1
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
h0 ðτ0 , tÞ h1 ðτ1 , tÞ
) ð2:554Þ
X
jϕi ðtÞ j2π f c t
þ::: þ sðt  τN clussters ðtÞÞ ai ðtÞe e ,
i2C N clusters
|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
hN clusters ðτN clusters , tÞ

where we use Ci to denote the i–th cluster (i ¼ 0,1,. . .,Nclusters), while hi(τi,t) is the impulse response of the i-th cluster. The
overall impulse response of the wideband channel, based on previous equation, is given by:

XðtÞ
N clusters
hðτ, t Þ ¼ hðτi , t Þδðτ  τi Þ, ð2:555Þ
i¼0

and the corresponding wideband channel model is shown in Fig. 2.46.

Based on wideband channel model, we conclude that there are two time variables, the deal spread τ and time instance t.
Corresponding variables in frequency domain are frequency f and Doppler spread ν, respectively. Since the impulse response
is time-varying, it makes sense to define the autocorrelation function of the impulse response as follows:

Rh ðτ1 , τ2 ; t, t þ Δt Þ ¼ hh ðτ1 , t Þhðτ2 , t þ Δt Þi: ð2:556Þ

Under the wide-sense stationary (WSS) assumption, the autocorrelation functions is not a function of the time origin, and
we can write:

Rh ðτ1 , τ2 ; Δt Þ ¼ hh ðτ1 , t Þhðτ2 , t þ Δt Þi: ð2:557Þ

When the WSS assumption is extended by uniform scattering (US), we obtain the following simplified model for the
autocorrelation function, introduced by Bello [97]:

Δ
hh ðτ1 , t Þhðτ2 , t þ Δt Þi ¼ Rh ðτ1 , Δt Þδðτ1  τ2 Þ ¼ Rh ðτ, Δt Þ, ð2:558Þ

which indicates that the channel responses associated with different multipath components are mutually uncorrelated.
2.3 Propagation Effects in Wireless Communication Channels 115

2.3.2.9 Power Delay Profile and Coherence Bandwidth

The WSS + US autocorrelation function of impulse response for which Δt ¼ 0 is commonly referred to as the power delay
profile (PDP), denoted as Rh(τ), and gives the information how the total received power is distributed among different
multipath components. The PDP can be used to determine the average delay spread:

ð
1

τRh ðτÞdτ
0
μτMDS ¼ ð
1 , ð2:559Þ
Rh ðτÞdτ
0

and root-mean-squared (rms) delays spread:

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u1
uð
u τ  μ
2 R ðτÞdτ
u Tm h
u
u0
σ τMDS ðτÞ ¼ u ð : ð2:560Þ
u 1
u
t Rh ðτÞdτ
0

The Fourier transform of h(τ,t) with respect to the delay spread τ, at particular time instance t:

ð
1

H ð f ; tÞ ¼ hðτ; t Þej2πf τ dτ, ð2:561Þ

1

is also time-varying, so that it makes sense to define the autocorrelation function of time-varying transfer function H( f,t) as
follows:

RH ð f 1 , f 2 ; Δt Þ ¼ hH  ð f 1 , t ÞH ð f 2 , t þ Δt Þi
*1 ð ð
1 +
¼ h ðτ1 ; t Þe j2π f 1 τ1 dτ1 hðτ2 ; t þ Δt Þej2π f 2 τ2 dτ2
1 1 ð2:562Þ
ð 1
1 ð
¼ hh ðτ1 ; t Þhðτ2 ; t þ Δt Þie j2π f 1 τ1 j2π f 2 τ2
e dτ1 dτ2 ,
1 1

which under WSS + US assumption simplifies to:

ð
1

RH ð f 1 , f 2 ; Δt Þ ¼ Rh ðτ, Δt Þej2π ð f 2  f 1 Þτ
dτ ¼ RH ðΔf ; Δt Þ, ð2:562Þ
1

and becomes the function on frequency difference Δf. For the fixed observation time (Δt ¼ 0), we conclude that the PDP and
autocorrelation function of transfer function represent a Fourier transform pairs:

ð
1

RH ðΔf Þ ¼ Rh ðτÞej2πΔf τ dτ: ð2:563Þ

1
116 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.47 The relationship

between DPD and autocorrelation
function of frequency response
H( f,t) (for Δt ¼ 0) and the
definition of the coherence
bandwidth Bc

Fig. 2.48 The illustration of frequency-selective fading for one snapshot of the frequency response

The range of frequencies for which RH(Δf ) is nonzero is known as the coherence bandwidth Bc. The definition of
coherence bandwidth and its relationship to the PDP and rms delay spread is illustrated in Fig. 2.47. An example of the
narrowband signal as well as an example of wideband signal is shown to illustrate how the fading affects the signal. Clearly,
for the narrowband signal, in this particular example, different frequency components experience roughly the same fading
amplitude. On the other hand, for the wideband signal, different frequencies experience different fading and refer to this type
of fading as frequency-selective fading. Therefore, the coherence bandwidth can be used to classify the fading into two
categories: flat-fading and frequency-selective fading. When the bandwidth of the signal is much smaller than coherence
bandwidth, Bs < < Bc, the fading is flat-fading. In other words, for the flat-fading, the multipath delay spread is much smaller
than the symbol duration. On the other hand, in frequency selective fading, as illustrated in Fig. 2.48, the bandwidth of the
signal is much larger than the coherence bandwidth, Bs> > Bc. In other words, in frequency-selective fading, the MDS is much
longer than the symbol duration.

2.3.2.10Channel Coherence Time and Fading Channels Classification

The time variations of the channel arising from transmitter and/or receiver motion introduce a Doppler shift in the received
signal. This Doppler effect can be characterized by taking the Fourier transform of RH(Δf;Δt) with respect to Δt:

ð
1

SH ðΔf , νÞ ¼ RH ðΔf , Δt Þej2πνΔt dΔt: ð2:564Þ

1

In order to characterize Doppler effect at a single Doppler frequency, we have to set Δf ¼ 0 and define the Doppler
spectrum SH(ν) ¼ SH(0; ν) as follows:

ð
1

SH ðνÞ ¼ RH ðΔt Þej2πνΔt dΔt: ð2:565Þ

1

The range of time intervals over which the RH(Δt) is approximately nonzero is commonly referred to as the coherence time
Tc, which is illustrated in Fig. 2.49. The time-varying channel decorrelates after Tc seconds. The coherence time can be used to
determine how many symbols get affected by the fading. The maximum ν value for which the Doppler spectrum magnitude is
larger than zero, |SH(ν)| > 0, is known as the Doppler spread of the channel, denoted as BD. Based on coherence time, we can
classify fading channels as slow-fading and fast-fading channels. For slow-fading channels, the symbol duration Ts is much
shorter than the coherence time, Ts < < Tc. In other words, in slow-fading channels, the signal bandwidth is much larger than
2.3 Propagation Effects in Wireless Communication Channels 117

Fig. 2.49 The illustration of

definitions of coherence time Tc
and Doppler spread BD

Bs
Flat in t (slow)
Selective in t (fast)
Selective in f
Selective in f
ISI
Bc

Flat in t (slow) Selective in t (fast)

Flat in f Flat in f

Tc Ts

Fig. 2.50 Classification of wireless channels based on coherence bandwidth Bc and coherence time Tc

the Doppler spread, Bs > > BD. On the other hand, in fast-fading channels, the symbol duration Ts is much longer than the
coherence time, Ts > > Tc. In other words, in slow-fading channels, the signal bandwidth is much smaller than the Doppler
spread, Bs < <BD. The classification of wireless channels based on coherence bandwidth and coherence time is summarized in
Fig. 2.50.
When the symbol duration is much longer than coherence time, we also say that the channel is ergodic, as all possible
channel conditions (states) can take place. On the other hand, when Ts < < Tc we say that the channel is nonergodic.
Finally, the factor TcBc  1/τMDSBD is known as the spread factor [67]. When τ MDSBD < 1 (equivalently BcTc > 1), we say
that the channel is underspread, and such channel can easily be estimated. On the other hand, when τ MDSBD > 1 (equivalently
BcTc < 1), we say that the channel is overspread.
The classification of wireless channels based on coherence bandwidth, coherence time, and spread factor is summarized in
Table 2.1.
Another relevant function in characterizing the wireless channels is the scattering function Sh(τ; ν), which is related to the
autocorrelation function Rh(τ;Δt) by the Fourier transform:

ð
1

Sh ðτ, νÞ ¼ Rh ðτ, Δt Þej2πνΔt dΔt: ð2:566Þ

1

The scattering function describes how the average output power is distributed among multipath delay τ and Doppler
frequency ν. The scattering function Sh(τ; ν) and the autocorrelation function RH(Δf;Δt) are related by:

ð
1 ð
1

SH ðτ, νÞ ¼ RH ðΔf , Δt Þej2πνΔt e j2πτΔf

dΔtdΔf : ð2:567Þ
1 1

Different correlation functions and scattering functions are mutually related as summarized in Fig. 2.51.
118 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Table 2.1 Classification of wireless channels based on coherence bandwidth, coherence time, and spread factor.
Bs  Bc Frequency-flat fading
Bs Bc Frequency-selective channel
Ts  Tc Time-flat (slow) fading
Ts Tc Time-selective (fast) fading
BcTc > 1 Underspread channel
BcTc  1 Overspread channel
Ts  Tc Nonergodic channel
Ts  Tc Ergodic channel

Fig. 2.51 The Fourier transform (FT) relationship among different correlation and scattering functions

2.4 Signal Propagation in Optical Fibers and Corresponding Channel Impairments

In this section, we describe basic signal and noise parameters in fiber-optics communications and major channel impairments
including chromatic dispersion, multimode dispersion, polarization mode dispersion (PMD), and fiber nonlinearities [1–4,
98–104].

2.4.1 Fiber Attenuation and Insertion Losses

Fiber attenuation can be described by the general relation dP/dz ¼ αP, where α is the power attenuation coefficient per unit
length. If Pin is the power launched into the fiber, the power remaining after propagating a length L within the fiber Pout is
Pout ¼ Pinexp(αL). The absorption coefficient varies with wavelength as many of the absorption and scattering mechanisms
vary with λ. For instance, Rayleigh scattering in fiber is due to microscopic variations in the density of glass (density
fluctuation sites < λ) and varies as [1–4]:

αR ¼ C=λ4 , C ¼ 0:7  0:9 ðdB=kmÞ‐μm4 ð2:658Þ

Mie scattering is caused by imperfections (scattering) in the core-cladding interface that are larger than λ.
Intrinsic absorption can be identified as (i) infrared absorption (in SiO2 glass, vibrational modes of Si-O bonds cause an
absorption peak at λ > 7 μm, which has a tail extending to the λ ¼ 1.55 μm range) and (ii) ultraviolet absorption (which is due
to the electronic resonances occurring in ultraviolet region (λ < 0.4 μm)).
Extrinsic absorption results from the presence of impurities: (i) transition metal impurities (Fe, Cu, Co, Ni, Mn, and Cr)
absorb strongly in 0.6–1.6 μm, and the loss level is reduced below 1 dB/km if their amount is kept below 1 part per billion; and
(ii) residual OH- ions cause peaks near λ ¼ 0.95 μm, 1.24 μm, and 1.39 μm
Fiber loss is not the only source of optical signal attenuation along the transmission lines. The fiber splices and fiber
connectors also cause the signal attenuation. The fiber splices can be fused or joined together by some mechanical means, with
typical attenuation being 0.01–0.1 dB per fused splice and slightly above 0.1 dB per mechanical splice.
2.4 Signal Propagation in Optical Fibers and Corresponding Channel Impairments 119

Optical connectors are removable and allow many repeated connections and disconnections, with typical insertion loss for
high-quality SMF not above 0.25 dB. To minimize the connector return loss (a fraction of the optical power reflected back
into the fiber at the connector point), the angled fiber-end surfaces are commonly used.
The number of optical splices and connectors depends on transmission length and must be taken into account, unless the
total attenuation due to fiber joints is distributed and added to the optical fiber attenuation.

2.4.2 Chromatic Dispersion Effects

Dispersion problem can be described as follows. Short optical pulses entering a dispersive channel such as an optical fiber get
spread out into a much broader temporal distribution. Both the intermodal dispersion and chromatic dispersion cause the
distortion in multimode optical fibers, while chromatic dispersion is the only cause of the signal distortion in SMF. The
intermodal dispersion can be specified through the optical fiber bandwidth Bfib that is related to a 1-km-long optical fiber, and
this parameter is specified by the manufactures and commonly measured at wavelength around 1310 nm (chromatic dispersion
is negligible in this region compared to intermodal dispersion) [1, 4, 98]:

Bfib
Bfib,L ¼ , μ ¼ 0:5  1 ð2:569Þ
Lμ

Single-mode optical fibers do, however, introduce another signal impairment known as the chromatic dispersion.
Chromatic dispersion is caused by difference in velocities among different spectral components within the same mode and
has two components: (i) material dispersion and (ii) waveguide dispersion component. The material dispersion is caused by
the fact that the refractive index is a function of wavelength, defined by Sellmeier equation [2, 4, 98]:
" #
XM
Bi λ2
nð λ Þ ¼ 1 þ , ð2:750Þ
i¼1 λ  λi
2 2

with typical Bi and λi parameters for pure silica being:

B1 ¼ 0:6961663 at λ1 ¼ 0:0684043 μm
B2 ¼ 0:4079426 at λ2 ¼ 0:1162414 μm
B3 ¼ 0:8974794 at λ3 ¼ 9:896161 μm

The group index ng can be determined by using the Sellmeier equation and the following definition expression:

ng ¼ n þ ωdn=dω ¼ n  λdn=dλ:

Waveguide dispersion is related to the physical design of the optical fiber. Since the value of Δ is typically small, the
refractive indices of the core cladding are nearly equal, and the light is not strictly confined in the fiber core, and the fiber
modes are said to be weakly guided. For a given mode, say fundamental mode, the portion of light energy that propagates in
the core depends on wavelength, giving rise to the pulse-spreading phenomenon known as waveguide dispersion. By
changing the power distribution across the cross-sectional area, the overall picture related to the chromatic dispersion can
be changed. The distribution of the mode power and the total value of waveguide dispersion can be manipulated by multiple
cladding layers. This is commonly done in optical fibers for special application purposes, such as dispersion compensation.
Different refractive index profiles of single-mode fibers used today are shown in Fig. 2.52. The conventional SMF index
profile is shown in Fig. 2.52(a). The nonzero dispersion-shifted fiber (NZDSF), with refractive index profile shown in
Fig. 2.52(b), is suitable for use in WDM systems. Large effective area NZDSFs, with index profile shown in Fig. 2.52(c), are
suitable to reduce the effect of fiber nonlinearities. Dispersion compensating fiber (DCF), with index profile shown in
Fig. 2.52(d), is suitable to compensate for positive chromatic dispersion accumulated along the transmission line.
Because the source is non-monochromatic, different spectral components within a pulse will travel at different velocities,
inducing the pulse broadening. When the neighboring pulses cross their allocated time slots, the ISI occurs, and the signal bit
120 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.52 Refractive index

profiles for different SMF types:
(a) standard SMF, (b) NZDSF
with reduced dispersion slope, (c)
NZDSF with large effective area,
and (d) DCF

rates that can be effectively transmitted can be severely limited. A specific spectral component, characterized by the angular
optical frequency ω, will arrive at the output end of the fiber of length L after some delay τg, known as the group delay:

L dβ L dβ λ2 dβ
τg ¼ ¼L ¼ ¼ L , ð2:751Þ
vg dω c dk 2πc dλ

where β is the propagation constant introduced earlier and vg ¼ [dβ/dω]1 is the group velocity, the speed at which the energy
of an optical pulse travels. As a result of the difference in time delays, the optical pulse disperse after traveling a certain
distance and the pulse broadening (Δτg) can be characterized by:

dτg dτg
Δτg ¼ Δω ¼ Δλ, ð2:752Þ
dω dλ

where Δω represents the frequency bandwidth of the source and Δλ represents the wavelength bandwidth of the source. By
substituting Eqs. (2.29) into (2.30), we get:
dβ
 
dτg d L dω d2 β d2 β 2πc
Δτg ¼ Δω ¼ Δω ¼ L 2 Δω ¼ L 2  2 Δλ ¼ LDΔλ, ð2:753Þ
dω dω dω dω λ

where D [ps/nm-km] represents the chromatic dispersion parameter defined by:

2πc d2 β 2πc d2 β
D¼ ¼  2 β2 , β2 ¼ ð2:754Þ
λ2 dω2 λ dω2

where β2 denotes the previously introduced group velocity dispersion (GVD) parameter. The chromatic dispersion can be
expressed as the sum of two contributing factors:
 
λ2 dβ
d   
2πc dλ 1 dβ d2 β
D¼ ¼ 2λ þ λ2 2 ¼ DM þ DW ,
dλ 2πc dλ dλ ð2:755Þ
λ d β
2 2
λ dβ
DM ¼  DW ¼ 
2πc dλ2 πc dλ

where DM represents the material dispersion and DW represents the waveguide dispersion.
2.4 Signal Propagation in Optical Fibers and Corresponding Channel Impairments 121

Local birefringence axes

Different fiber sections

Fig. 2.53 A simplified PMD model

The material dispersion arises due to wavelength dependence of the refractive index on the fiber core material, which
causes the wavelength dependence of the group delay. The wavelength dependence of the refractive index n(λ) is well
approximated by the Sellmeier equation, introduced earlier. The material dispersion is related to the slope of ng ¼ n  λdn(λ)/
dλ by DM ¼ (1/c)dng/dλ. For pure silica fiber, the following approximation is valid [2]:

DM  122ð1  λZD =λÞ, 1:25μm < λ < 1:66μm ð2:756Þ

where λZD ¼ 1.276 μm is the zero-dispersion wavelength. We can see that for wavelengths larger than zero-dispersion
wavelength, the material dispersion coefficient is positive, while the GVD is negative, and this regime is known as the
anomalous dispersion region.
The waveguide dispersion occurs because the propagation constant is a function of the fiber parameters (core radius and
difference between refractive indices in fiber core and cladding) and at the same time is a function of wavelength. Therefore,
the propagation paths for a mode due to different boundary matching conditions are slightly different at different wavelengths.
Waveguide dispersion is related to the physical design of the optical fiber. Since the value of Δ is typically small, the refractive
indices of the core cladding are nearly equal, and the light is not strictly confined in the fiber core. For a given mode, say
fundamental mode, the portion of light energy that propagates in the core depends on wavelength: the longer the wavelength,
the more power in the cladding. The effect of waveguide dispersion on pulse spreading can be approximated by assuming that
refractive index of material is independent of wavelength. To make the result independent of fiber configuration, we should
express the group delay in terms of the normalized propagation constant b, defined as:

β  n2 k ðbΔ þ 1Þ, Δ  1 ð2:757Þ

The group delay due to waveguide dispersion can be found as [1]:

   
L dβ L dðVbÞ dðVbÞ 2J 2m ðpaÞ
τW ¼ ’ n2 þ n2 Δ , ¼b 1 ð2:758Þ
c dk c dV dV J mþ1 ðpaÞJ m1 ðpaÞ

where p is the core parameter introduced earlier (p2 ¼ n21 k2  β2). The pulse broadening due to waveguide dispersion can be
determined using Eq. (2.758) by:
 
dτW dτ dV dk n Δ d2 ðVbÞ V k
ΔτW ¼ Δλ ¼ W Δλ ¼ L 2  Δλ ¼ LDW Δλ, ð2:759Þ
dλ dV dk dλ c dV 2 k λ

where the waveguide dispersion parameter DW is defined by:

 2 
n2 Δ 1 d ðVbÞ
DW ’  V , ð2:760Þ
c λ dV 2

and it is negative for the normalized frequencies between 0 and 3.

122 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

The total dispersion can be written as sum of two contributions, D¼DM + DW. The waveguide dispersion (i.e., negative)
shifts the zero-dispersion wavelength to approximately 1.308 μm. D ~ 15–18 ps/(km-nm) near 1.55 μm. This is the low loss
region for fused silica optical fibers. This approach for mutual cancellation was used to produce several types of single-mode
optical fibers that are different in design compared to standard SMF, standardized by ITU-T. In addition to standard SMF,
there are two major fiber types: (i) dispersion-shifted fibers (DSF) (described in ITU-T recommendation G.653) with
dispersion minimum shifted from the 1310 nm wavelength region to the 1550 nm region and (ii) nonzero-dispersion-shifted
fibers (NZDSF) (described in G.655 recommendation), with dispersion minimum shifted from 1310 nm window to anywhere
within C or L bands (commercial examples: TrueWave fiber or LEAF).
Dispersion effects do not disappear completely at zero-dispersion wavelength. Residual dispersion due to higher-order
dispersive effects still exists. Higher-order dispersive effects are governed by the dispersion slope parameter S ¼ dD/dλ.
Parameter S is also known as differential dispersion parameter or second-order dispersion parameter, and it is related to the
GVD parameter and second-order GVD parameter β3 ¼ dβ2/dω by:
   2
dD d 2πc 4πc 2πc
S¼ ¼  2 β 2 ¼ 3 β2 þ β3 : ð2:761Þ
dλ dλ λ λ λ2

The geometrical optics (or ray theory) approach, used earlier to describe the light confinement in step-index fibers through
the total internal reflection, as shown in Fig. 2.7 and text related to it, is valid when the fiber has a core radius a much larger
than the operating wavelength λ. Once the core radius becomes comparable to the operating wavelength, the propagation of
light in step-index fiber is governed by Maxwell’s equations describing the change of electric E and magnetic H fields in space
and time, as discussed at the beginning of the chapter. Since the silica is nonmagnetic material M ¼ 0, so that the induced
electric and magnetic densities, P and E, are mutually connected by Eq. (2.7), which is clearly the convolution integral. By
applying the convolution theorem of the Fourier transform (FT), the Eq. (2.7) becomes:

PðrÞ ¼ ε0 χ ðrÞEðrÞ,
ð2:762Þ
Pðr, ωÞ ¼ FT fPðr, t Þg, Eðr, ωÞ ¼ FT fEðr, t Þg, χ ðr, ωÞ ¼ FT ðχ ðr, t ÞÞ:

The constitutive relation (2.5) now becomes:

2 3
6 7
D = ε0 E þ P ¼ ε0 E þ ε0 χ ðr, ωÞE ¼ ε0 6 þ χ ðr, ωÞ7
41|fflfflfflfflfflffl E,
ffl{zfflfflfflfflfflfflffl}5 ð2:763Þ
εðr, ωÞ

¼ ε0 εðr, ωÞE,

where the dielectric permittivity frequency response εðr, ωÞ is related to the refractive index n(r, ω) by:

εðr, ωÞ ¼ ½nðr, ωÞ þ jαðr, ωÞ2 ;

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð2:764Þ
ω
nðr, ωÞ ¼ 1 þ Re χ ðr, ωÞ, αðr, ωÞ ¼ Imχ ðr, ωÞ;
cnðr, ωÞ

where α(r, ω) is the attenuation coefficient. By applying the curl of Maxwell equation (2.1), and after substitution of (2.2) into
(2.1), we obtain:

2
∇
ð∇
EÞ ¼ μ0 ∂ ∇
H =∂t ¼ μ0 ∂ D=∂t 2 :
|fflfflffl{zfflfflffl} ð2:765Þ
∇
H¼∂D=∂t

Now after applying the FT on (2.765) and by substituting the constitutive relation (2.763) into (2.765), we obtain:

εðr, ωÞω2
∇
ð∇
EÞ ¼ μ0 ε0 εðr, ωÞω2 E ¼ E: ð2:766Þ
|{z} c2
1=c2

By employing the cross-product identity, we can rewrite the previous equation as follows:
2.4 Signal Propagation in Optical Fibers and Corresponding Channel Impairments 123

εðr, ωÞω2
∇
ð∇
EÞ ¼ ∇ð∇EÞ  |ffl{zffl}
∇∇ E ¼ E: ð2:767Þ
c2
Δ

Based on Maxwell equation (2.1), we conclude that ∇E ¼ 0, and since the typical attenuation in modern SMS is below
0.2 dB/km, we can neglect the attenuation in (2.764) so that εðr, ωÞ ffi n2 ðr, ωÞ: The spatial dependence of refractive index can
be neglected, and after substituting εðr, ωÞ ffi n2 ðωÞ and ∇E ¼ 0 into (2.767), we obtain the following wave equation:

n2 ðωÞω2
ΔE þ E = 0: ð2:768Þ
c2

An alternative form of the wave equation is given by:

ΔE þ n2 ðωÞk20 E = 0, ð2:769Þ

where k0 ¼ ω/c ¼ 2π/λ is the free-space wave number. A specific solution of the wave equation, subject to appropriate
boundary conditions, is commonly referred to as an optical mode. The spatial field distributions, E(r,ω) ¼ E0(ρ,ϕ)exp[jβ(ω)
z  jωt] and H(r,ω) ¼ H0(ρ,ϕ) exp[jβ(ω)z  jωt], of a mode do not change as the mode propagates along z-axis except for a
multiplicative factor exp.[jβ(ω)z], with β(ω) being the propagation constant of a mode. The fiber modes may be classified as
guided modes, leaky modes, and radiation modes. Different modes in fiber propagate with different values of propagation
constant, in other words with different speeds, causing the pulse spreading. Therefore, it is desirable to design the fiber which
supports only one mode-fundamental mode. Such a fiber is called single-mode fiber (SMF). If the weak guidance condition
(Δ < <1) is not satisfied, the conventional modes TE, TM, EH, and HE can be found. The cylindrical symmetry of fiber
suggests the use of cylindrical coordinate system (ρ,ϕ,z). The wave equation is to be solved for six components: Eρ,Eϕ,Ez and
Hρ,Hϕ,Hz. We can solve the wave equation for axial components only (Ez and Hz) and use the system of Maxwell’s equations
to express the other components as functions of axial ones. Based on Eq. (2.40), the wave equation for Ez component can be
written as follows:

2
∂ E z 1 ∂Ez 1 ∂ E z
2
n1 , ρ
a
þ þ þ n2 k 20 E z ¼ 0, n ¼ ð2:770Þ
∂ρ2 ρ ∂ρ ρ2 ∂z2 n2 , ρ>a

where n1 is the refractive index of the core of radius a and n2 is the refractive index of the cladding, as shown in Fig. 2.7. The
wave equation for Ez component can easily be solved using the method of separation of variables, leading to the following
overall solution [1–4]:

AJ m ðpρÞejmϕ ejβz , ρ
a
E z ðρ, ϕ, zÞ ¼ jmϕ jβz
ð2:771Þ
BK m ðqρÞe e , ρ>a

where Jm(x) and Km(x) are corresponding Bessel functions of mth order, introduced earlier (A and B are constants to be
determined from the boundary conditions). The similar equation is valid for Hz. The other four components can be expressed
in terms of axial ones by using the Maxwell’s equations, and in the core region, we obtain [2–4, 98]:
   
j ∂Ez ω ∂H z j β ∂Ez ∂H z
Eρ ¼ β þ μ Eϕ ¼  μ ω
p2 ∂ρ 0
ρ ∂ϕ p2 ρ ∂ϕ 0
∂ρ
    ð2:772Þ
j ∂H z ω ∂E z j β ∂H z ∂H z
Hρ ¼ 2 β  ε0 n 2 Hϕ ¼ 2 þ ε 0 n2 ω
p ∂ρ ρ ∂ϕ p ρ ∂ϕ ∂ρ

where p2 ¼ n21 k20  β2. The similar equations can be obtained for cladding by replacing p2 with –q2, where q2 ¼ β2  n22 k 20.
By satisfying the boundary conditions, the homogeneous system of equation is obtained, which has the nontrivial solution
only if corresponding determinant is zero, which leads to the following characteristic (eigenvalue) equation [2–4, 98]:
124 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

  0 
J 0m ðpaÞ K 0 ðqaÞ J m ðpaÞ n2 K 0 ðqaÞ
þ m þ 22 m
paJ m ðpaÞ qaK m ðqaÞ paJ m ðpaÞ n1 qaK m ðqaÞ
" # " # ð2:773Þ
1 1 1 n22 1
¼m 2
þ  þ
ðpaÞ2 ðqaÞ2 ðpaÞ2 n21 ðqaÞ2

where pa and qa are related by the following equation:

ðpaÞ2 þ ðpqÞ2 ¼ V 2 : ð2:774Þ

For given fiber parameters n1, n2, and a and the operating wavelength, we can determine the normalized frequency V. For
fixed V and a given m, there exist multiple solutions n ¼ 1,2,3,.. that lead to different modes-propagating modes. The case
m ¼ 0 corresponds to meridional rays (the rays that periodically intersect the center axis of the fiber), the electric/magnetic
components are independent of ϕ, and the corresponding modes are classified as TE0n (Ez ¼ 0) and TM0n (Hz ¼ 0) modes. The
case m 6¼ 0 corresponds to skew rays, the electric and magnetic field components are functions of ϕ, and the corresponding
modes are classified as HEmn (Hz dominates over Ez) and EHmn (Ez dominates over Hz) modes. Once the mode is identified
(m and n are fixed) from Eq. (2.773) and by using the definition expressions for p and q above, we can easily determine the
propagation constant of the mode βmn.
The propagating light pulse is a composite optical signal containing a number of monochromatic spectral components.
Each spectral component behaves differently in a dispersion medium, such as optical fiber, leading to the light pulse
distortion. Each axial component of the monochromatic electromagnetic wave can be represented by its complex electric
field function [2–4, 98]:

Eðz, t Þ ¼ Ea ðz, t Þ exp ½jβðωÞz exp ½jω0 t : ð2:775Þ

The light pulse distortion, which is observed through the pulse broadening along the fiber, can be evaluated by knowing the
frequency dependence of the propagation constant β ¼ β(ω) at a specific distance z along the fiber. Each spectral component
will experience a phase shift proportional to β(ω)z. The amplitude spectrum observed at point z along the fiber length, in the
frequency domain, is given by:

Ea ðz, ωÞ ¼ E a ð0, ωÞ exp ½jβðωÞz: ð2:776Þ

The behavior of the pulse envelope during the propagation process can be evaluated through the inverse Fourier transform
of previous equation, which is very complicated to calculate, unless additional simplifications are made, for example, by
expressing the propagation constant in terms of a Taylor series:

  
dβ  1 d 2 β  2 1 d3 β 
β ð ωÞ  β ð ωc Þ þ  ð ω  ω Þ þ ðω  ω Þ þ ð ω  ωc Þ 3 þ . . . ð2:777Þ
dω ω¼ωc c
2 dω2 ω¼ωc c
6 dω3 ω¼ωc


where β1 ¼ ðdβ=dωÞjω¼ωc is related to the group velocity vg by β1 ¼ 1/vg, β2 ¼ d 2 β=dω2 ω¼ωc is the group velocity
3

dispersion (GVD) parameter, and β3 ¼ d β=dω3 ω¼ωc is the second-order GVD parameter. Introducing the concept of
slow-varying amplitude A(z,t) as follows [2–4, 98]:

Eðz, t Þ ¼ Ea ðz, t Þ exp ½jβðωÞz exp ðjωc t Þ ¼ Aðz, t Þ exp ð jβc z  jωc t Þ ð2:778Þ

we get:

ð
1   
1 e β2 β3
Aðz, t Þ ¼ Að0, ωÞ exp j β1 Δω þ Δω þ Δω z exp ðjðΔωÞt ÞdðΔωÞ:
2 3
ð2:779Þ
2π 2 6
1
2.4 Signal Propagation in Optical Fibers and Corresponding Channel Impairments 125

By taking partial derivative of A(z,t) with respect to propagation distance z, we obtain:

ð
∂Aðz, tÞ 1 β2 β ~ β β
jðβ1 Δωþ 22 Δω2 þ 63 Δω3 Þz jðΔωÞt
¼ 1 jðβ1 Δω þ Δω2 þ 3 Δω3 ÞAð0, ΔωÞe e dðΔωÞ ð2:780Þ
∂z 2π 2 6
1

Finally, by taking the inverse Fourier transform of (2.780), we derive basic propagation equation describing the pulse
propagation in single-mode optical fibers:

2 3
∂Aðz, t Þ ∂Aðz, t Þ β ∂ Aðz, t Þ β3 ∂ Aðz, t Þ
¼ β1 j 2 þ : ð2:781Þ
∂z ∂t 2 ∂t 2 6 ∂t 3

2.4.3 Polarization Mode Dispersion (PMD)

The polarization unit vector, representing the state of polarization (SOP) of the electric field vector, does not remain constant
in practical optical fibers; rather, it changes in a random fashion along the fiber because of its fluctuating birefringence [105–
108]. Two common birefringence sources are (i) geometric birefringence (related to small departures from perfect cylindrical
symmetry) and (ii) anisotropic stress (produced on the fiber core during manufacturing or cabling of the fiber). The degree of
birefringence is described by the difference in refractive indices of orthogonally polarized modes Bm ¼ |nx  ny| ¼ Δn. The
corresponding difference in propagation constants of two orthogonally polarized modes is Δβ ¼ |βx  βy| ¼ (ω/c)Δn.
Birefringence leads to a periodic power exchange between the two polarization components, described by beat length
LB ¼ 2π/Δβ ¼ λ/Δn. Typically Bm ~ 107, and therefore LB ~ 10 m for λ ~ 1 μm. Linearly polarized light remains linearly
polarized only when it is polarized along one of the principal axes; otherwise, its state of polarization changes along the fiber
length from linear to elliptical, and then back to linear, in a periodic manner over the length LB.
In certain applications it is needed to transmit a signal through a fiber that maintains its SOP. Such a fiber is called a
polarization-maintaining fiber (PMF). (PANDA fiber is a well-known PMF.) Approaches to design PMF having high but
constant birefringence are (i) the shape birefringence (the fiber having an elliptical core) and (ii) stress birefringence (stress-
inducing mechanism is incorporated in fiber). Typical PMF has a birefringence Bm ~ 103 and a beat length Bm ~ 1 mm,
resulting in a polarization crosstalk smaller than 30 dB/km. Unfortunately, the loss of PMFs is high (α ~ 2 dB/km).
The modal group indices and modal group velocities are related by:

c c dβx, y
ngx,y ¼ ¼ , β1x,y ¼ ð2:782Þ
vgx,y 1=β1x,y dω

so that the difference in time arrivals (at the end of fiber of length L ) for two orthogonal polarization modes, known as the
differential group delay (DGD), can be calculated by [2, 4, 98]:
 
L L   

Δτ ¼    ¼ Lβ1x  β1y  ¼ LΔβ1 : ð2:783Þ
vgx vgy

DGD can be quite large in PMF (~ 1 ns/km) due to their large birefringence. Conventional fibers exhibit much smaller
birefringence, but its magnitude and polarization orientation change randomly at scale known as the correlation length lc (with
typical values in the range 10–100 m). The analytical treatment of PMD is quite complex due to statistical nature. A simple
model divides the fiber into a large number of segments (tens to thousands of segments), with both the degree of birefringence
and the orientation of the principal states being constant in each segment but change randomly from section to section, as
shown in Fig. 2.53. Even though the fiber is composed of many segments having different birefringence, there exist principal
states of polarization (PSP). If we launch a pulse light into one PSP, it will propagate to the corresponding output PSP without
spreading due to PMD. Output PSPs are different from input PSPs. Output PSPs are frequency independent up to the first-
order approximation. A pulse launched into the fiber with arbitrary polarization can be represented as a superposition of two
126 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

input PSPs. The fiber is subject to time-varying perturbations (such as temperature and vibrations) so that PSPs and DGD vary
randomly over time.
DGD is Maxwellian distributed random variable with the mean-square DGD value [2, 4, 98]:

D E    
L L
ðΔT Þ2 ¼ 2ðΔβ1 Þ2 l2c exp  þ 1 , ð2:784Þ
lc lc

where the parameters in (2.784) are already introduced above. For long fibers, L> > lc, the root-mean-squared (RMS) DGD
follows:

D E1=2 pffiffiffiffiffiffiffiffiffi pffiffiffi

ðΔT Þ2 ¼ Δβ1 2lc L ¼ Dp L, ð2:785Þ

where Dp[ps/√km] is the PMD parameter. Typical values for PMD parameter are in the range 0.01–10 ps/(km)1/2, in new
fibers Dp < 0.1 ps/(km)1/2. The first-order PMD coefficient, Dp, can be characterized again by Maxwellian PDF [105–108]:
rffiffiffi 2 !

2 Dp D2p
f Dp ¼ exp  2 , ð2:786Þ
π α3 2α

where α is the coefficient with a typical value around 30 ps [105, 106]. The mean value <Dp > determined from
Maxwellian PDF: <Dp > ¼(8/π)1/2α. The overall probability Pr(Dp) that coefficient Dp will be larger than pre-specified
value can be determined by:

D
ðp

CDF Dp ¼ f Dp dDp , ð2:787Þ

0

The value CDF(3 < Dp>) is about 4105, so that the practical expression to characterize the first-order PMD is [98]:
 pffiffiffi
ΔT ¼ 3 Dp L: ð2:788Þ

The second-order PMD occurs due to frequency dependence of both DGD and PSP and can be characterized by coefficient
DP2 [105]:
s
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2  2
1 ∂DP1 DP1 ∂jsj
DP2 ¼ þ , ð2:789Þ
2 ∂ω 2 ∂ω

where the first term describes the frequency dependence of DGD and the second term describes the frequency dependence of
Stokes vector s (describing the position of PSP) [106]. The statistical nature of second-order PMD coefficient in real fiber is
characterized by PDF of DP2 [107]:

2σ 2 DP2 tanh ðσDP2 Þ

f ðDP2 Þ ¼ : ð2:790Þ
π cosh ðσDP2 Þ

The probability that the second-order PMD coefficient is larger than 3 < DP2 > is still not negligible. However, it becomes
negligible for larger values of 5 < DP2 >, so that the total pulse spreading due to the second-order PMD effect can be expressed
as [98]:

ΔτP2 ¼ 5hDP2 iL: ð2:791Þ

2.4 Signal Propagation in Optical Fibers and Corresponding Channel Impairments 127

Fig. 2.54 Fiber nonlinearities

classification Nonlinear effects

Kerrr nonlinearities
nonlinearities Stimulated scattering

Self-phase Cross-phase Stimulated Raman Stimulated Brillouin

Modulation (SPM) Modulation (XPM) Scattering (SRS) Scattering (SBS)

Four-wave mixing
(FWM)

2.4.4 Fiber Nonlinearities

The basic operational principles of optical transmission can be explained assuming that optical fiber medium is linear. The
linearity assumption is valid if the launched power does not exceed several mW in a single channel system. In modern WDM
technology, high-power semiconductor lasers and optical amplifiers are employed, and the influence of fiber nonlinearities
becomes important. Moreover, in some special cases, the fiber nonlinearities may be used to enhance the fiber transmission
capabilities (e.g., soliton transmission).

2.4.4.1 Fiber Nonlinearities Classification

There are two major groups of fiber nonlinearities related either to nonlinear refractive index (Kerr effect) or to nonlinear
optical scattering, which is illustrated in Fig. 2.54.
The Kerr effect occurs due to dependence of index of refraction on light intensity, and fiber nonlinearities belonging to this
category are (i) self-phase modulation (SPM), (ii) cross-phase modulation (XPM), and (iii) four-wave mixing (FWM). The
SPM is related to the single optical channel. The variation of power within the channel causes the changes in refractive index,
which leads to the pulse distortion. In XPM, the refractive index changes due to variations in power not only in observed
channel but also due to variation in powers of other wavelength channels leading to the pulse distortion. In FWM, several
wavelength channels interact to create newly channels. This effect is dependent on both the powers of interacting channel and
chromatic dispersion.
The stimulated scattering effects are caused by parametric interaction between the light and materials. There are two types
of stimulated scattering effects: (i) stimulated Raman scattering (SRS) and stimulated Brillouin scattering (SBS). In the SRS,
the light interacts with material through the vibrations and causes the energy transfer from short wavelength channels to longer
wavelength channels, leading to the interchannel crosstalk. In the SBS, the light interacts with the matter through acoustic
waves, leading to the coupling with backward propagating waves, thus limiting the available power per channel.

2.4.4.2 Effective Length and Effective Cross-Sectional Area

The nonlinear effects introduced above are function of the transmission length and cross-sectional area of the optical fiber. The
nonlinear interaction is stronger for longer fibers and smaller cross-sectional area. On the other hand, the nonlinear interaction
decreases along the transmission line because the signal is getting attenuated as it propagates. Therefore, the strength of the
nonlinear effects may be characterized by introducing the concept of effective length Leff, as an equivalent length so that the
product of launched power P0 and the equivalent length is the same as the area below power evolution curve P(z) ¼ P0exp
(αz):

ðL ðL
P0 Leff ¼ PðzÞdz ¼ P0 eαz dz, ð2:792Þ
0 0
128 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

where α denotes the attenuation coefficient. By solving the previous equation for Leff, we obtain:

1  eαL
Leff ¼  1=α, ð2:793Þ
α

and the approximation is valid for fiber which are at least several tens of kilometers in length. The affective length after
M amplifiers is given by:

1  eαL 1  eαL L
Leff ,total ¼ M¼ , ð2:794Þ
α α l

where L is the total transmission length, l is the amplifier spacing, and M is the number of amplifiers. If the amplifier spacing
increases, the optical amplifier gain increases in proportion to exp.(αl), to compensate for the fiber losses. On the other hand,
the increase in the power enhances the nonlinear effects. Therefore, what matters is the product of launched power P0 and the
total effective length Leff,total as follows: P0Leff,total.
The nonlinear effects are inversely proportional to the area of the fiber core because the concentration of the optical power
per unit cross-sectional area (power density) is higher for smaller cross-sectional area and vice versa. Optical power
distribution across the cross-sectional area is closely related to the overall refractive index, and to characterize it, the concept
of effective cross-sectional area (Aeff) is introduced [3, 4, 98]:
2 32
ðð
6 7
4 rdrdθjEðr, θÞj2 5
r θ
Aeff ¼ ðð , ð2:795Þ
rdrdθjE ðr, θÞj4
r θ

where E(r,θ) is the distribution of electrical field in cross section and (r,θ) denotes the polar coordinates. For the Gaussian
approximation E(r,θ) ¼ E0exp(r2/w2), the effective cross-sectional area is simply Aeff ¼ πw2, where w is the mode radius.
The Gaussian approximation is not applicable to the optical fibers having more complicated index profile, such as DSFs,
NZDFs, and DCFs. In these cases, the effective cross-sectional area can be found by [4, 98]:

Aeff ¼ c f πw2 , ð2:796Þ

where the correction factor cf is smaller than one for DCFs and some NZDSFs and larger than one for NZDSF with larger
effective cross-sectional area.

2.4.4.3 Kerr Effect, Self-Phase Modulation, and Nonlinear Length

For relatively high power density, the index of refraction acts as the function density; the effect is known as Kerr effect:

P
nðPÞ ¼ n0 þ n2 , ð2:797Þ
Aeff

where n2 is the Kerr coefficient (second-order refractive index coefficient), with typical values being 2.2–3.41020 m2/W.
The variation in refractive index due to Kerr effect yields to the propagation constant β variations, because β ¼ 2πn/λ:

2πn2
βðPÞ ¼ β0 þ γP, β0 ¼ 2πn0 =λ, γ ¼ , ð2:798Þ
λAeff

where γ is the nonlinear coefficient, with typical values being 0.9–2.75 W1 km1 at 1550 nm. The propagation constant β
will vary along the duration of optical pulse, because the different points along the pulse will “see” different optical powers
and the frequency chirping is introduced. The propagation constant associated with the leading edge of the pulse will be lower
than that related to the central part of the pulse. The difference in propagation constants will cause the difference in phases
2.4 Signal Propagation in Optical Fibers and Corresponding Channel Impairments 129

associated with different portions of the pulse. The central part of the pulse will acquire phase more rapidly than the leading
and trailing edges. The total nonlinear phase shift after some length L can be found by:

ðL ðL
1  eαL Leff
ΔΦ½P ¼ ½βðP0 Þ  βdz ¼ γPðzÞdz ¼ γP0 ¼ , ð2:799Þ
|{z}|fflfflfflfflffl{zfflfflfflffl
α ffl} LNL
0 0 1=LNL Leff

where LNL ¼ [γP0]1 is the nonlinear length. The frequency variation (chirp) can be found as the first derivative of the
nonlinear phase shift with respect to time:

d ½ΔΦðt Þ 2π Leff n2 dP0 ðt Þ

ΔωSPM ðt Þ ¼ ¼ : ð2:800Þ
dt λ Aeff dt

Generally speaking, the SPM introduces the pulse broadening. However, in some situations, the frequency chirp due to
SPM could be opposite to that due to chromatic dispersion (such as the anomalous dispersion region). In such a case, SPM
helps to reduce the impact of chromatic dispersion. To see this better, let us observe the propagation equation and extend it to
account for the Kerr nonlinear effects as follows:

2 3
∂Aðz, t Þ α ∂Aðz, t Þ β ∂ Aðz, t Þ β3 ∂ Aðz, t Þ
¼  Aðz, t Þ  β1 j 2 þ þ jγ jAðz, t Þj2 Aðz, t Þ, ð2:801Þ
∂z 2 ∂t 2 ∂t 2 6 ∂t 3

where α is the attenuation coefficient and γ is the nonlinear coefficient introduced in (2.798). After the introduction of a new
coordinate system with T ¼ t  β1z, the generalized nonlinear Schrödinger equation (GNLSE) is obtained:

2 3
∂Aðz, T Þ α β ∂ Aðz, t Þ β3 ∂ Aðz, T Þ
¼  Aðz, T Þ  j 2 þ þ jγ jAðz, T Þj2 Aðz, T Þ: ð2:802Þ
∂z 2 2 ∂T 2 6 ∂T 3

Assuming that the fiber loss and the second-order GVD can be neglected, the GNLSE simplifies to:

2
∂Aðz, t Þ ∂Aðz, t Þ β ∂ Aðz, t Þ
¼ β1 j 2 þ jγ jAðz, t Þj2 Aðz, t Þ: ð2:803Þ
∂z ∂t 2 ∂t 2

Substituting τ ¼ (t  β1z)/τ0, ξ ¼ z/LD (LD ¼ τ20/|β2|), and U ¼ A/√P0 (P0 ¼ A20), we get:

∂U sgn ðβ2 Þ ∂ U
2
j  þ N 2 jU j2 U ¼ 0, N 2 ¼ LD =LNL ð2:804Þ
∂ξ 2 ∂τ2

When N < <1, the nonlinear length is much larger than the dispersion length so that the nonlinear effects can be neglected
compared to those of chromatic dispersion. If chromatic dispersion can be neglected (LD- > 1), the NLSE can be solved
analytically [99]:

2
U ðz, τÞ ¼ U ð0, τÞejzjU ð0,τÞj =LNL
: ð2:805Þ

Interestingly, the SPM modulation causes a phase change but no change in the envelope of the pulse. Thus, SPM by itself
leads only to chirping, regardless of the pulse shape. The SPM-induced chirp, however, modifies the pulse-broadening effects
of chromatic dispersion.

2.4.4.4 Cross-Phase Modulation

The XPM is another effect caused by the intensity dependence of the refractive index, but it is related to the multichannel
transmission and occurs during propagation of a composite signal. The nonlinear phase shift of a specific optical channel, say
130 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

the n-th channel, is affected not only by the power of that channel but also by the optical power of the other channels as
follows:

X
N
2π Lef f n2 X
N
ΔΦn ðtÞ ¼ γLef f P0n ðtÞ þ γLef f 2 P0i ðtÞ ¼ ½P0n ðtÞ þ 2 P0i ðtÞ, ð2:806Þ
λ Aef f
i¼1, i6¼n i¼1, i6¼n

where P0n(t) denotes the pulse shape in the n-th channel, other parameters are the same as before, and N denotes the number of
wavelength channels. The nonlinear phase shift is bit-pattern dependent. In the worst-case scenario, assuming that all
wavelength channels carry the same data stream, we can write:

2π Leff n2
ΔΦn ðt Þ ¼ ð2N  1ÞP0n ðt Þ: ð2:807Þ
λ Aeff

In practice, the optical pulses from different optical channels propagate at different speeds (they have different group
velocities). The phase shift given above can occur only during the overlapping time. The overlapping among neighboring
channels is longer than the overlapping of channels spaced apart, and it will produce the most significant impact on the phase
shift. If pulses walk through one another quickly (due to significant chromatic dispersion or when the channels are widely
separated), the described effects on both pulses are diminished because the distortion done by the trailing edge is undone by
the leading edge.

2.4.4.5 Four-Wave Mixing

The FWM is another effect that occurs in optical fibers during the propagation of a composite optical signal, such as the WDM
signal. It gives a rise to the optical signals. The three optical signals with different carrier frequencies fi, fj, and fk (i,j,k ¼ 1,. . .,
M) interact to generate the new optical signal at frequency fijk ¼ fi + fj  fk, providing that the phase matching condition is
satisfied:

βijk ¼ βi þ β j  βk , ð2:808Þ

where βs are corresponding propagation constants. The measure of the phase-matching condition of wavelength channels
involved in the interaction is defined by:

Δβ ¼ βi þ β j  βk  βijk , ð2:809Þ

The FWM is effective only if Δβ approaches zero. Therefore, the phase-matching condition is a requirement for the
momentum conservation. The FWM process can also be considered from quantum-mechanic point of view as the annihilation
of two photons with energies hfi and hfj and generation of two new photons with energies hfk and hfijk. In FWM process the
indices i and j do no need to be necessarily distinct, meaning that only two channels may interact to create the new one; this
case is known as degenerate case.
The power of the newly generated optical frequency is a function of the powers of optical signals involved in the process,
the Kerr coefficient, and the degree of satisfaction of the phase-matching condition, so that we can write [3, 4, 98, 99]:
 
2π f ijk n2 d ijk 2
Pijk  Pi P j Pk L2eff , ð2:810Þ
3cAeff

where dijk is the measure of degeneracy, which takes value 3 in degenerate case or value 6 in non-degenerate case. FWM can
produce significant signal degradation in WDM systems since several newly generated frequencies can coincide with any
specific channel. Namely, the total number of newly generated frequencies N through the FWM process is N ¼ M2(M  1)/2.
Fortunately, some of the newly generated frequencies have negligible impact. However, some of them will have significant
impact especially those coinciding with already existing WDM channels.
The FWM can be reduced either by reducing the power per channel or by preventing the perfect phase-matching by
increasing the chromatic dispersion or increasing the channel spacing. One option would be to use non-uniform channel
2.4 Signal Propagation in Optical Fibers and Corresponding Channel Impairments 131

spacing. On the other hand, the FWM process can be used to perform some useful functions, such as the wavelength
conversion through the process of phase conjugation [2].
Another interpretation of Kerr nonlinearities can be provided as follows. Under isotropy assumption, in the presence of
nonlinearities, induced polarization P(r,t) is along the same direction as the electric field E(r,t), so that corresponding vector
notation can be substituted by the scalar functions and the induced polarization be written in the form:

ðt
Pðr, t Þ ¼ ε0 χ ð1Þ ðt  t 0 ÞE ðr, t 0 Þdt 0 þ ε0 χ ð3Þ E 3 ðr, t Þ ð2:811Þ
1

where the first term corresponds to the linear part and the second term, PNL(r,t), to the nonlinear part. χ(1) denotes the linear
susceptibility, χ(3) denotes the third-order susceptibility, and ε0 is the permittivity of vacuum. (Because of the symmetry of
silica molecules, only odd terms are present in Taylor expansion of P(r,t).) A wavelength division multiplexing (WDM) signal
can be represented as the sum of N monochromatic plane waves of angular frequency ωi (the modulation process is ignored in
this simplified analysis):

X
N
E ðr, t Þ ¼ Ei cos ðωn t  βn zÞ: ð2:812Þ
n¼1

The nonlinear polarization can be written as after substitution of Eq. (2.812) into second term of (2.811) as follows:

X
N X
N X
N

PNL ðz, t Þ ¼ ε0 χ ð3Þ E i cos ðωi t  βi zÞE j cos ω j t  β j z E k cos ðωk t  βk zÞ ð2:813Þ
i¼1 j¼1 k¼1

which can also be rewritten as:

3 X N
3 XX N
PNL ðz, t Þ ¼ ε0 χ ð3Þ E 2i E i cos ðωi t  βi zÞ þ ε0 χ ð3Þ 2Ei E j Ei cos ðωi t  βi zÞ
4 i¼1
4 i¼1 j6¼i

3 X
N X


þ ε0 χ ð3Þ E2i E j cos 2ωi  ω j t  2βi  β j z ð2:814Þ
4 i¼1 j6¼i

6 X XXN


þ ε0 χ ð3Þ E i E j E k cos ωi þ ω j  ωk t  βi þ β j  βk z þ . . .
4 i¼1 j>i k>j

The first term in the first line corresponds to the SPM, the second term in the same line corresponds to the XPM, the term in
the second line corresponds to the degenerate FWM case, and the last term corresponds to the non-degenerate FWM.

2.4.4.6 Raman Scattering

Raman scattering is a nonlinear effect that occurs when a propagating optical signal interacts with glass molecules in the fiber
undergoing a wavelength shift. The result of interaction is a transfer of energy from some photons of the input optical signal to
vibrating silica molecules and creation of new photons with lower energy than the energy of incident photons. The incident
optical signal is commonly referred to as a pump, and the generated signal is called the Stokes signal. The difference in
frequencies between the pump (ωp) and the Stokes signal (ωs) is known as the Raman frequency shift ωR ¼ ωp  ωs. Scattered
photons are not in phase with each other and do not follow the same scattering pattern, meaning that energy transfer from
pump to Stokes photons is not a uniform process. As a result, there exists the frequency band that includes frequencies of all
scattered Stokes photons. Scattered Stokes photons can take any direction that can be either forward or backward with respect
to the direction of the pump, meaning that Raman scattering is isotropic process. If the pump power is lower than a certain
threshold value, the Raman scattering process will have a spontaneous character, characterized by the small number of pump
photons that will be scattered and converted to the Stokes photons. However, when the pump power exceeds the threshold
value, Raman scattering becomes a stimulated process. SRS can be explained as a positive feedback process in the pump
132 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

signal interacts with Stokes signal and creates the beat frequency ωR ¼ ωp  ωs, and the beat frequency then serves as a
stimulator of molecular oscillations, and the process is enhanced or amplified. Assuming that the Stokes signal propagates in
the same direction as the pump signal, the intensity of the pump signal Ip ¼ Pp/Aeff and the intensity of the Stokes signal
Is ¼ Ps/Aeff are related by the following systems of coupled differential equations [3, 4, 98, 99]:
 
dI p ω dI s
¼ gR P I p I s  αp I p , ¼ gR I p I s  αs I s , ð2:815Þ
dz ωS dz

where αp (αs) is the fiber attenuation coefficient of the pump (Stokes) signal and gR is the Raman gain coefficient. Because the
SRS process is not the uniform process, the scattered Stokes photons will occupy a certain frequency band, and the Raman
gain is not constant but the function of frequency. It can roughly be approximated by the Lorentzian spectral profile [3, 4, 98,
99]:

gR ð ΩR Þ
g R ð ωR Þ ¼ , ð2:816Þ
1 þ ðωR  ΩR Þ2 T 2R

where TR is vibration states decay time (in the order 0.1 ps for silica based materials) and ΩR is the Raman frequency shift
corresponding the Raman gain peak. The actual gain profile extends over 40 THz (~320 nm), with a peak around 13.2 THz.
The Raman gain peak gR(ΩR) ¼ gR,max is between 1012 and 1013 m/W for wavelengths above 1300 nm. The gain profile
can also be approximated by a triangle function, which is commonly used in analytical studies of SRS [3, 4, 98, 99]:

gR ðΩR ÞωR
g R ð ωR Þ ¼ : ð2:817Þ
ΩR

The Raman threshold, defined as the incident power at which half of the pump power is converted to the Stokes signal, can
be estimated using the system of differential equations by replacing gR by gR,max. The amplification of Stokes power along
distance L can be determined by:
 
gR, max Ps,0 L
Ps ðLÞ ¼ Ps,0 exp , ð2:818Þ
2Aeff

where Ps,0 ¼ Ps(0). The Raman threshold can be determined by [3, 4, 98, 99]:

16Aeff
PR,threshold ¼ Ps,0  ð2:819Þ
gR, max Leff

and for typical SMF parameters (Aeff ¼ 50 μm2, Leff ¼ 20 km, and gR,max ¼ 71013 m/W), it is around 500 mW.
The SRS can effectively be used for optical signal amplification (Raman amplifiers), as we explained earlier. In WDM
systems, however, the SRS effect can be quite detrimental, because the Raman gain spectrum is very broad, which enables the
energy transfer from the lower to higher wavelength channels. The shortest wavelength channel within the WDM system acts
as a pump for several long wavelength channels while undergoing the most intense depletion. The fraction of power coupled
out of the channel 0 to all the other channels (1, 2, . . ., M  1; M is the number of wavelengths), when the Raman gain shape is
approximated as a triangle, is [3, 4, 98, 99]:

X
N 1
gR, max ΔλPLeff N ðN  1Þ
P0 ¼ P0 ðnÞ ¼ , ð2:820Þ
n¼1
2ΔλC Aeff 2

where Δλ is the channel spacing and Δλc ffi 125 nm. The energy transfer between two channels is bit-pattern dependent and
occurs only if both wavelengths are synchronously loaded with 1 bits, meaning that the energy transfer will be reduced if
dispersion is higher due to the walk-off effect (difference in velocities of different wavelength channels will reduce the time of
overlapping).
2.4 Signal Propagation in Optical Fibers and Corresponding Channel Impairments 133

2.4.4.7 Brillouin Scattering

Brillouin scattering is a physical process that occurs when an optical signal interacts with acoustical phonons, rather than the
glass molecules. During this process, an incident optical signal reflects backward from the grating formed by acoustic
vibrations and downshifts in frequency. The acoustic vibrations originate from the thermal effect if the power of an incident
optical signal is relatively small. If the power of incident light goes up, it increases the material density through the
electrostrictive effect. The change in density enhances acoustic vibrations and forces Brillouin scattering to become
stimulated. The SBS process can also be explained as a positive feedback mechanism, in which the incident light (the
pump) interacts with Stokes signal and creates the beat frequency ωB ¼ ωp  ωs. The parametric interaction between pump,
Stokes signal, and acoustical waves requires both the energy and momentum conservation: the energy is effectively preserved
through the downshift in frequency, while the momentum conservation occurs through the backward direction of the Stokes
signal. Since the SBS can be effectively suppressed by dithering the source (directly modulating the laser with a sinusoid at a
frequency much lower than the receiver low-frequency cutoff), it is not discussed much here.

2.4.5 Generalized Nonlinear Schrödinger Equation

The GNLSE has already been introduced in previous section. Here we describe it in operator form, discus its efficient solving,
and extend it to the multichannel transmission. Additionally, the GNLSE for polarization division multiplexing (PDM) is
described as well. Finally, we describe the GNLSE suitable for study of propagation is spatial division multiplexing (SDM)
systems.

2.4.5.1 GNLSE for Single Channel Transmission

The operator form of GNLSE is given by [3, 4, 98, 109, 110]:

∂Eðz, t Þ  b b  2 3
b ¼  α  j β2 ∂ þ β3 ∂ , N
 
b ¼ jγ E 2 ,
¼ D þ N E, D ð2:821Þ
∂z 2 2 ∂t 2 6 ∂t 3

where E is the signal electric field magnitude, Db and N b denote the linear and nonlinear operators, while α, β2, β3, and γ
represent attenuation coefficient, GVD, second-order GVD, and nonlinear coefficient, respectively. To solve the NSE, the
split-step Fourier method is commonly used [3, 4, 98, 109, 110]. The key idea behind this method is to split the fiber into
sections, each with length Δz, and perform the integration of NSE on each section, which leads to expression:
h   i
E ðz þ Δz, t Þ ¼ exp Db þN
b Δz E ðz, t Þ: ð2:822Þ

In addition, the Taylor expansion can be used to present exponential term from Eq. (2.822) as:

  1 
X n
exp Db þN
b Δz ¼ b þN
D b Δzn =n! ð2:823Þ
n¼0

Instead of Taylor expansion, the following two approximations are commonly used in practice:
8
DΔz=2 b
< eb b
eN Δz eDΔz=2 E ðz, t Þ
E ðz þ Δz, t Þ  ð2:824Þ
: b b
eDΔz eN Δz Eðz, t Þ

where the first method (the upper arm on the right side of above equation) is known as symmetric split-step Fourier method
(SSSFM), while the second (the lower arm on the right side of above equation) is known as asymmetric split-step Fourier
method (ASSFM). Clearly, these methods have similarities with corresponding methods to describe the propagation in
atmospheric turbulence channels. The linear operator in either method corresponds to multiplication in frequency domain,
which is:
134 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

      
b 1 α β2 2 β3 3
exp ΔhD Eðz, t Þ ¼ FT exp  þ j ω þ j ω Δh FT ½Eðz, t Þ , ð2:825Þ
2 2 6

where Δh is the step-size equal to either Δz/2 in SSSFM or Δz in ASSFM, while FT (FT1) denotes Fourier transform (inverse
Fourier transform) operator. On the other side, the nonlinear operator performs the nonlinear phase “rotation” in time domain,
expressed as:
   
b E ðz, t Þ ¼ exp jγΔzjEðz, t Þj2 E ðz, t Þ:
exp ΔzN ð2:826Þ

It is evident that the nonlinear operator depends on electric field magnitude at location z, which is not known and should be
evaluated. It was proposed in [3] to use the trapezoidal rule and express the electric field function as:
0zþΔz 1
  ð  
b
E ðz þ Δz, t Þ ¼ exp DΔz=2 exp @ N ðz0 Þdz0 A exp DΔz=2
b Eðz, t Þ
z ð2:827Þ
     
b ðz þ ΔzÞ þ N
N b ðzÞ
b
 exp DΔz=2 exp b
Δz exp DΔz=2 Eðz, t Þ:
2

The iteration procedure that should be applied in accordance with Eq. (2.827) is illustrated in Fig. 2.55. Since this iteration
procedure can be time-extensive, the following less accurate approximation can be considered instead:
8   h i  
> b
< exp DΔz=2 exp N b ðz þ Δz=2ÞΔz exp DΔz=2b Eðz, t Þ, forSSSFM
Eðz þ Δz, t Þ    h i : ð2:828Þ
>
: exp DΔzb exp Nb ðzÞΔz E ðz, t Þ, forASSFM

Fig. 2.55 Fiber nonlinearities classification. Solving the GNLSE by iterative symmetric split-step Fourier method
2.4 Signal Propagation in Optical Fibers and Corresponding Channel Impairments 135

The complexity of iterative algorithms is determined by complexity of fast Fourier transform, which is proportional to
(NlogN ). To study the accuracy of split-step methods described above, we can use the Baker-Hausdorff formula [7]:

eA eB ¼ eAþBþ2½A,Bþ12½A,½A,B12½B,½A,B24½B,½A,½A,Bþ::: ,
1 1 1 1
ð2:829Þ

where [A,B] is the commutator of two operators A and B, defined as: [A,B] ¼ AB  BA. By applying this formula to SSSFM
and ASSFM, we obtain that:

b b b b 1bb 2
eDΔz eN Δz ffi eDΔzþN Δzþ2½D,N Δz þ::: ,
ð2:830Þ
b b b b b 1 b b bb
eDΔz=2 eN Δz eDΔz=2 ffi eDΔzþN Δzþ6½N þD=2,½N ,D=2Δz þ::: :
3

As we can see, the operator error in ASSFM is in order of (Δz2), while the operator error in SSSFM is in order of (Δz3),
which means that SSSFM offers better accuracy.
The step-size selection is analyzed in several papers [111–113]. The step-size is traditionally chosen as a function of
characteristic length over which the phase distortion reaches maximum tolerable value. For chromatic dispersion (CD) and
Kerr nonlinearities (NL), the characteristic lengths are given as [109, 110]:

LNL  1=γPTx , LCD  2=β2 ω2max , ð2:831Þ

where PTx is the launched power and ωmax is the highest frequency in the spectrum of modulated signal, which is:

2πRs , single‐carriersystems
ωmax  ð2:832Þ
2π ðN sc þ 1ÞΔ f sc , OFDM systems

with Rs being the symbol rate, Nsc being the number of subcarriers, and Δfsc being the subcarrier spacing. If the maximum
phase shifts that can be tolerated due to chromatic dispersion and nonlinearities are denoted by ΔϕCD and ΔϕNL, respectively,
we can determine the step-size as [109, 110]:

Δz ¼ min ðΔϕCD LCD , ΔϕNL LNL Þ: ð2:833Þ

Another approach is to keep the local error Δε below tolerable value by choosing the step-size as [112, 113]:
8n h io1=3
< Δε= γPðzÞ jβ jω2
2 , forSSSFM
2 max
Δz ¼ ð2:834Þ
: 
 1=2
Δε= γPðzÞ jβ2 jω2max , forASSFM

where P(z) is the signal power at distance z.

2.4.5.2 GNLSE for Multichannel Transmission

As for multichannel propagation modeling, we have two options. The first option is to express multichannel WDM signal
through a composite electric field:
X
jl2πΔf
E ðz, t Þ ¼ E ðz, t Þe
l l
, ð2:835Þ

where Δf is the channel spacing. This approach can be called the “total-field propagation model.” The second approach is to
consider individual WDM separately and then solve the set of coupled GNLSEs:
136 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

∂E i ðz, t Þ  b 
b i E i ðz, t Þ,
¼ Di þ N
∂z
! ð2:836Þ
α ∂ β ∂
2
β ∂
3 X 2
b i ¼   β1,i  j
D 2
þ 3 b ¼ jγ jEi j þ
,N 2  Ej  :
2 ∂t 2 ∂t 2 6 ∂t 3 i
j6¼i

2.4.5.3 GNLSE for Polarization Division Multiplexing Systems

The NSE for PDM systems can be described as follows [109, 110]:

∂Eðz, t Þ  b b 
¼ D þ N Eðz, t Þ,
∂z
 ð2:837Þ
2 3

b ¼  α  β1 ∂  j β 2 ∂ þ β 3 ∂ , N
D b ¼ jγ jEj2 I  1 E{ YE Y ,
2 ∂t 2 ∂t 2 6 ∂t 3 3

where E ¼ [Ex Ey]T denotes the Jones vector of electric field; α denotes the attenuation coefficients in x- and y-polarizations;
and β1, β2, and β3 denote the group velocity, GVD, and second-order GVD parameters, respectively, while Y denotes the
Pauli-Y matrix in quantum mechanics notation [7]. The Pauli matrices are defined here as:
     
0 1 0 j 1 0
X¼ ,Y ¼ , Z¼ : ð2:838Þ
1 0 j 0 0 1

The β1 is related to the polarization rotation matrix R as follows:

 
δ
β1 ðθk , εk Þ ¼ Rðθk , εk Þ k Z R{ ðθk , εk Þ,
" 2 #
cos θk cos εk  j sin θk sin εk sin θk cos εk þ j cos θk sin εk ð2:839Þ
R ð θ k , εk Þ ¼ ,
sin θk cos εk þ j cos θk sin εk cos θk cos εk  j sin θk sin εk

where 2θk and 2εk are azimuth and ellipticity angles of k-th fiber section and δk corresponds to differential group delay (DGD)
of the section. The PDM-NSE should be solved through iterative relation:
h  i
b þN
Eðz þ Δz, t Þ ¼ exp Δz D b Eðz, t Þ, ð2:840Þ

where the complexity of propagation simulation is clearly higher than in the single-polarization case.

2.4.5.4 GNLSE for Spatial Division Multiplexing Systems

In the rest of this section, we describe the GNLSE for few-mode fiber (FMF)-based mode-division multiplexing, based on
N modes, we introduced in [114] (see also [115]). The electric field in a group of N/2 degenerate spatial modes is represented
! !
by N-dimensional complex valued vector E. For example, if we consider two modes fiber (LP01 and LP11), then the E can be
expanded as:

!
E ¼ ½ E1 E2 T ¼ ½ E 1x E 1y E2x E 2y T , ð2:841Þ

!
which is constructed by stacking the Jones vectors of N spatial modes one on top of each other. In Eq. (2.841), Ei denotes the i-
th spatial mode. We separate the linear and nonlinear operators as follows:

!  !
dE b þN
b E,
¼ D ð2:842Þ
dz
2.5 Noise Sources in Optical Channels 137

b is the linear part and the N

where the D b is the nonlinear part. The linear part vector form GNLSE can be written as:

b ¼ 1! ! dy ! ∂2 1!∂
3
D α  β1  jβ2 2 þ β3 3 , ð2:843Þ
2 dx ∂t 3! ∂t

! ! ! ! !
where in the ideal case, α is a zero matrix and βi is proportional to the identity. In general cases, the α ¼ α I ( I is identity
!
matrix), assuming that all modes are attenuated to the same level. The β1 can be evaluated by:

!
β1 ¼ U  Α  U{ , ð2:844Þ

where U and Α are random unitary Gaussian matrix [116] and different modes delays matrix, respectively. Both U and Αare
matrices of size 2 N-by-2 N, and U{is the conjugated transpose of U. Α can be further represented as:

! τ !
τ1 þ PMD1 Z ... 02
2
2
A¼ð ⋮ ⋱ ⋮ Þ, ð2:845Þ
! ! τPMDN
02
2 ::: τN þ Z
2

where τPMDi is a two-by-two diagonal matrix with elements equal to differential mode group delay (DMGD):

DPMD1
τPMDi ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi, ð2:846Þ
2 2  Lcorrelation

! ! !
where Lcorrelation is the correlation length of the FMF. Similarly, α, β2, and β3 can be extended to matrix form [115, 117]. The
nonlinear operator N b i can be defined as:

! 2! X
N
b i ¼ jγ½ f self 8 jE 4 ! 2 ! 1 !{ !
N j Iþ f cross j Ek j I  ðEi Y Ei ÞY i ¼ 1, 2, . . . , N, ð2:847Þ
9 i 3 3
k6¼i

!
where Ei denotes the i-th propagation mode, which includes both x- and y-polarization modes. fself and fcross are general self-
 2  2
! !
and cross-mode coupling coefficients [115, 117]. The i-th spatial mode power Ei  can be calculated as Ei  ¼
   
 !2  !2
 Eix  þ  Eiy  . The nonlinearity factor γ in Eq. (2.847) can be obtained as described in previous section by:

2πn2 f ref
γ¼ , ð2:848Þ
cAeff

where n2 is the nonlinear index, fref is the reference frequency, and Aeff is the core area.

2.5 Noise Sources in Optical Channels

The total noise is a stochastic process that has both multiplicative (introduced only when the signal is present) and additive
(always present) components. Multiplicative noise components are mode partition noise (MPN), laser intensity noise (RIN),
modal noise, quantum shot noise, and avalanche shot noise. Additive noise components are dark current noise, thermal noise,
amplified spontaneous emission (ASE) noise, and crosstalk noise.
Originators of the noise in an optical transmission system can be identified as follows. In semiconductor lasers, the laser
intensity noise, laser phase noise, and mode partition noise are present. In optical cable, in fiber and splicing, the modal noise
and reflection-induced noise are present. In optical amplifiers, the spontaneous emission and amplified spontaneous emission
noises are present. At the receiver side, during the photodetection process, the thermal noise and quantum noise are generated.
138 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

The noise components at the receiver side are cumulative. The most relevant optical noise components at the input of optical
pre-amplifier are intensity noise and spontaneous emission noise, which contains the spontaneous emission noise from the
preceding in-line amplifier and accumulated ASE noise from other in-line amplifiers. The optical pre-amplifier will enhance
all optical inputs in proportion to the gain, and it will also generate additional spontaneous emission noise. During the
photodetection process, as already indicated above, the quantum noise and thermal noise will be generated; in addition,
different beating components, such as signal-noise beating and noise-noise beating components, will be created. The most
important noise sources coming from the front-end amplifiers are ASE noise and signal-noise beating component. The front
end amplifier will generate additional thermal noise.

2.5.1 Mode Partition Noise

The mode partition noise (MPN) is the relative fluctuation in power between the main and side modes. The total power can
remain unchanged, but the distribution of power among modes changes. MPN is present (up to certain amount) even in DFBs.
MPN can affect the RIN significantly by enhancing it. It will occur even when the mode suppression ratio (MSR) is around
20 dB. Different modes will travel at different velocities due to dispersion, so that MPN can lead to the ISI.

2.5.2 Reflection-Induced Noise

The reflection-induced noise is related to the appearance of the back-reflected light due to refractive index discontinuities at
optical splices, connectors, and optical fiber ends. The amount of reflected light can be estimated by the reflection coefficient,
introduced earlier by:
 2
na  nb
r ref ¼ , ð2:849Þ
na þ nb

where na and nb denote the refractive index coefficients of materials facing each other.
The strongest reflection occurs at the glass-air interface, rref  [(1.46–1)/(1.46 + 1)]2  3.5% (14.56 dB). It can be
reduced below 0.1% if index-matching oils or gels are used. The considerable amount of back-reflected light can come back
and enter the laser cavity, negatively affecting the laser operation and leading to excess intensity noise. The RIN can be
increased by up to 20 dB if the back-reflected light exceeds 30 dBm. The multiple back and forth reflections between splices
and connectors can be the source of additional intensity noise.

2.5.3 Relative Intensity Noise (RIN) and Laser Phase Noise

The operating characteristics of semiconductor lasers are well described by the set of ordinary differential equations – the rate
equations – which describe the interaction between photons and electrons inside the active region [2–4, 98]:

dP P
¼ GP þ Rsp  þ F P ðt Þ, ð2:850Þ
dt τp

dN I N
¼   GP þ F N ðt Þ ð2:851Þ
dt q τc

and
 
dϕ 1 1
¼ αchirp G  þ F ϕ ðt Þ ð2:852Þ
dt 2 τp
2.5 Noise Sources in Optical Channels 139

In Eq. (2.850) the term GP denotes the increase in photon number (P) due to stimulated emission, the term Rsp denotes the
increase in P due to spontaneous emission and can be calculated by Rsp ¼ nspG (nsp-the spontaneous emission factor, typically
nsp  2), the term –P/τp denotes the decrease in P due to emission through the mirrors and scattering/absorption by free
carriers, and Fp(t) corresponds to the noise process. In Eq. (2.851), the term I/q denotes the increase in electron numbers (N )
due to injection current I, the term –N/τc denotes the decrease in N due to spontaneous emission and nonradiative
recombination, the term –GP denotes the decrease in N due to stimulated emission, and the term Fn(t) corresponds to the
noise process. Finally, in Eq. (2.852), the first term corresponds to dynamic chirp and the second term to the adiabatic chirp,
and Fϕ(t) corresponds to the noise process. The noise terms Fp(t), Fn(t), and Fϕ(t) introduced above are known as Langevin
forces that are commonly modeled as zero-mean Gaussian random processes. The photon lifetime τp is related to the internal
losses (αint) and mirror losses (αmir) by τp ¼ [vg(αint + αmir)]1, with vg being the group velocity. The carrier lifetime τc is
related to the spontaneous recombination time (τspon) and non-radiation recombination time (τnr) by τc ¼ τsponτnr/(τspon + τnr).
The net rate of stimulate emission G is related to the material gain gm by G ¼ Γvggm, where Γ is the confinement factor.
Noise in semiconductor lasers originates from two sources: (i) spontaneous emission (the dominant noise source) and
(ii) electron-hole recombination shot noise. As mentioned above, the noise effects can be modeled by random driving terms in
laser rate equations known as Langevin forces, which can be described as zero-mean Gaussian random processes with
autocorrelation function (Markoffian approximation) [2–4, 98]:

hF i ðtÞF j ðt 0 Þi ¼ 2Dij δðt  t 0 Þ; i, j 2 fP, N, ϕg, ð2:853Þ

with dominating factor being Dpp ¼ RspP and Dϕϕ ¼ Rsp/4P. Fluctuation in intensity can be described by intensity
autocorrelation function [2–4, 98]:

2, P
C PP ðτÞ ¼ hδPðtÞδPðt þ τÞi=P  ¼ hPi, δP ¼ P  P
 ð2:854Þ

The RIN spectrum can be found by FT of the autocorrelation function [2–4, 98]:

ð
1 (
3
jωt 1=P at low P
RINðωÞ ¼ C pp ðτÞe dτ  ð2:855Þ
1=P at high P
1

Finally, the SNR can be estimated by [2–4, 98]:

 1=2
εNL
SNR ¼ ½C PP ð0Þ1=2  P ð2:856Þ
Rsp τp

where the approximation is valid for SNR above 20 dB.

Noise, especially spontaneous emission, causes phase fluctuations in lasers, leading to a nonzero spectral linewidth Δν. In
gas- or solid-state lasers, the linewidth Δν typically ranges from the subhertz to the kilohertz range. In semiconductor lasers,
the linewidth Δν is often much larger, up the megahertz range, because of (i) the small number of photons stored in the small
cavity and (ii) the non-negligible value of the linewidth enhancement factor αchirp. The spectrum of emitted light electric field
E(t) ¼ √(P)exp(jϕ) is related to the field autocorrelation function ΓEE(τ) by:

ð
1

S ð ωÞ ¼ ΓEE ðt Þejðωω0 Þτ dτ,

ð2:857Þ
1
 

ΓEE ðt Þ ¼ hE  ðt ÞE ðt þ τÞi ¼ h exp ð jΔϕðt ÞÞi ¼ exp  Δϕ2 ðτÞ =2 ,

where Δϕ(τ) ¼ ϕ(t + τ)  ϕ(t). By describing the laser phase noise process as Wiener-Lévy process, that is, a zero-mean
Gaussian process with variance 2πΔν|t| (where Δν is the laser linewidth), the spectrum of emitted light is found to Lorentzian:
140 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .


P 1 1
SEE ðνÞ ¼ ½ þ , ð2:858Þ
2πΔν 1 þ ð2ðνþν0 ÞÞ2 1 þ ð2ðνν0 ÞÞ2
Δν Δν

where ν0 is the central frequency and other parameters are already introduced above. The phase fluctuation variance
(neglecting the relaxation oscillation) is [2–4, 98]:

  Rsp  
Δϕ2 ðτÞ ¼ 1 þ α2chirp jτj ¼ 2πΔνjτj: ð2:859Þ
2P

The laser linewidth can, therefore, be evaluated by:

Rsp  
Δν ¼ 1 þ α2chirp , ð2:860Þ
4πP
 
and we conclude that the chirp effect enhances the laser linewidth by factor 1 þ α2chirp .

2.5.4 Modal Noise

Modal noise is related to multimode optical fibers. The optical power is non-uniformly distributed among a number of modes
in multimode fibers, causing the so-called speckle pattern at the receiver end containing brighter and darker sports in
accordance to the mode distribution. If the speckle pattern is stable, the photodiode effectively eliminates it by registering
the total power over the photodiode area. If the speckle pattern changes with time, it will induce the fluctuation in the received
optical power, known as the modal noise. The modal noise is inversely proportional to the laser linewidth, so it is a good idea
to use LEDs in combination with multimode fibers.

2.5.5 Thermal Noise

The photocurrent generated during photodetection process is converted to the voltage through the load resistance. The load
voltage is further amplified by the front-end amplifier stage. Due to random thermal motion of electrons in load resistance, the
already generated photocurrent exhibits additional noise component, also known as Johnson noise. The front-end amplifier
enhances the thermal noise generated in the load resistance, which is described by the amplifier noise figure Fn,el. The thermal
noise PSD in the load resistance RL and corresponding thermal noise power are defined as [2–4, 98]:

4kB Θ  2  4k Θ
Sthermal ¼ , i thermal ¼ B Δf ð2:861Þ
RL RL

where Θ is the absolute temperature, kB is the Boltzmann’s constant, and Δf is the receiver bandwidth.

2.5.6 Spontaneous Emission Noise and Noise Beating Components

The spontaneous emission of light appears during the optical signal amplification, and it is not correlated with signal (it is
additive in nature). The noise introduced by spontaneous emission has a flat frequency characterized by Gaussian probability
density function (PDF), and the (double-sided) PSD in one state of polarization is given by [2–4, 98]:

Ssp ðνÞ ¼ ðG  1ÞF no hν=2, ð2:862Þ

2.5 Noise Sources in Optical Channels 141

where G is the amplifier gain and hν is the photon energy. The optical amplifier noise figure Fno, defined as the ratio of SNRs
at the input and output of the optical amplifier, is related to the spontaneous emission factor nsp ¼ N2/(N2  N1) (for the
two-level system) by [2–4, 98]:
 
1
F no ¼ 2nsp 1  ffi 2nsp 2: ð2:863Þ
G

In the most of practical cases, the noise figure is between 3 and 7 dB. The effective noise figure of the cascade of
K amplifiers with corresponding gains Gi and noise figures Fno,i (i ¼ 1,2,. . .,K ) [2–4, 98] is as follows:

F no,2 F no,3 F no,K

F no ¼ F no,1 þ þ þ ... þ : ð2:864Þ
G1 G1 G2 G1 G2 . . . Gk1

The total power of the spontaneous emission noise, for an amplifier followed by an optical filter of bandwidth Bop, is
determined by:
 2
Psp ¼ 2Esp  ¼ 2Ssp ðνÞBop ¼ ðG  1ÞF no hνBop , ð2:865Þ

where the factor 2 is used to account for both polarizations.

Optical pre-amplifiers are commonly used to improve the receiver sensitivity by pre-amplifying the signal before it reaches
the photodetector, which is illustrated in Fig. 2.1. The gain of the optical amplifier is denoted with G, the optical filter
bandwidth is denoted with Bop, and electrical filter bandwidth is denoted with Bel (Fig. 2.56).
An optical amplifier introduces the spontaneous emission noise in addition to the signal amplification:
pffiffiffiffiffiffi
E ðt Þ ¼ 2P cos ð2π f c t þ θÞ þ nðt Þ, ð2:866Þ

where E(t) denotes the electrical field at the receiver input and n(t) denotes the amplifier spontaneous emission noise, which is
the Gaussian process with PSD given by Eq. (2.862). The photocurrent has the following form:
pffiffiffiffiffiffi
iðt Þ ¼ RPðt Þ ¼ RE 2 ðt Þ ¼ 2RP cos 2 ð2π f c t þ θÞ þ 2R 2Pnðt Þ cos ð2π f c t þ θÞ þ Rn2 ðt Þ: ð2:867Þ

Following the procedure described [2–4, 98], we can determine the variance of different noise and beating components as
follows:

Bðel

σ ¼
2
Si ð f Þdf ¼ σ 2sh þ σ 2sig‐sp þ σ 2sp‐sp , σ 2sh ¼ 2qR GPin þ Ssp Bop Bel ,
ð2:868Þ
Bel

σ 2sig‐sp ¼ 4R2 GPin Ssp Bel , σ 2sp‐sp ¼ R2 S2sp 2Bop  Bel Bel

where we used subscripts “sh” to denote the variance of shot noise, “sp-sp” to denote the variance of spontaneous-
spontaneous beat noise, and “sig-sp” to denote the variance of signal-spontaneous beating noise. Ssp is the PSD of
spontaneous emission noise (Eq. (2.21)), Pin is the average power of incoming signal (see Fig. 2.1), and the components of
the PSD of photocurrent i(t), denoted by Si( f ), are shown in Fig. 2.57.

Optical receiver
p.i.n
G OF EF
Pin P PD OF-optical filter
Bop Bel EF-electrical filter

Fig. 2.56 Optical receiver with the pre-amplifier

142 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.57 Components of PSD of

photocurrent

2.5.7 Quantum Shot Noise

The optical signal arriving at the photodetector contains a number of photons generating the electron-hole pairs through the
photoelastic effect. The electron-hole pairs are effectively separated by the reversed bias voltage. However, not every photon
is going to generate the electron-hole pair contributing to the total photocurrent. The probability of having n electron-hole
pairs during the time interval Δt is governed by the Poisson probability density function (PDF):

N n eN
pð nÞ ¼ , ð2:869Þ
n!

where N is the average number of generated electron-hole pairs. The Poisson PDF approaches Gaussian for large N. The mean
of photocurrent intensity:

qN qN
I ¼ hiðt Þi ¼ ¼ , Δt ¼ T ð2:870Þ
Δt T

For Poisson distribution, the variance equals the mean h(n  N )2i ¼ N so the photocurrent can be calculated by i(t) ¼ qn/T.
The shot noise mean square value can be determined by:
D E
 2 D E q 2
½ n  N  2
 
q2 N qI R 1
i sn ¼ ½iðt Þ  I 2 ¼ ¼ ¼ ¼ 2qIΔf , Δf ¼ b Rb ¼ ð2:871Þ
T2 T2 T 2 T

where Rb is the bit rate. The Eq. (2.13) is derived for NRZ modulation format, but validity is more general. Therefore, the
power spectral density of shot noise is determined by Ssn( f) ¼ 2qI. To determine the APD shot noise mean square value, the p-
i-n shot noise mean square value has to be multiplied by the excess noise factor F(M) to account for the randomness of impact
ionization effect:

 2 APD h i
1
i sn ¼ SAPD ð f ÞΔf ¼ 2qM 2
F ð M ÞIΔf , F ð M Þ ¼ k N M þ ð1  k N Þ 2  , ð2:872Þ
sn M

where M is the average APD multiplication factor and kN is the ratio of impact ionization factors of holes and electrons,
respectively. The often used excess noise factor approximation is the following one [98]:

F ðM Þ ffi M x , x 2 ½0, 1: ð2:873Þ

2.5 Noise Sources in Optical Channels 143

2.5.8 Dark Current Noise

The dark current Id flows through the reversed-biased photodiode even in the absence of incoming light. The dark current
consists of electron-hole pairs created due to thermal effects in p-n junction. The dark current noise power can be calculated as
follows [2–4, 98]:

 2 APD
i dcn ¼ SAPD 2
dcn ð f ÞΔf ¼ 2qhM i F ðM ÞI d Δf : ð2:874Þ

2.5.9 Crosstalk

The crosstalk effects occur in multichannel (WDM) systems and can be classified as (i) interchannel component (the crosstalk
wavelength is sufficiently different from observed channel wavelength) and (ii) intrachannel (the crosstalk signal is at the
same wavelength as observed channel or sufficiently close) component.
Interchannel crosstalk (also known as out-of-band or hetero-wavelength crosstalk) is introduced by either an optical filter
or demultiplexer that selects desired channel and imperfectly rejects the neighboring channels, which is illustrated in Fig. 2.58
(a). Another source of interchannel crosstalk is an optical switch, shown in Fig. 2.58(b), in which the crosstalk arises because
of imperfect isolation among different wavelength ports. The crosstalk signal behaves as noise, and it is a source of linear
(non-coherent) crosstalk, so that the photocurrent at receiver can be written as [2, 4, 99]:
X X
I ¼ Rm Pm þ Rn X mn Pn  I ch þ I X , Rm ¼ ηm q=hνm , IX ¼ Rn X mn Pn ð2:875Þ
n6¼m n6¼m

where Rm is the photodiode responsivity of mth WDM channel with average power Pm, Ich ¼ RmPm is the photocurrent
corresponding to the selected channel, IX is the crosstalk photocurrent, and Xmn is the portion of nth channel power captured by
the optical receiver of the mth optical channel.
The intrachannel crosstalk in transmission systems may occur due to multiple reflections. However, it is much more
important in optical networks. In optical networks, it arises from cascading optical (WDM) demux and mux or from an optical
switch, which is illustrated in Fig. 2.59. Receiver current I(t) ¼ R|Em(t)|2 contains interference or beat terms, in addition to the
desired signal: (i) signal-crosstalk beating terms (like EmEn) and (ii) crosstalk-crosstalk beating terms (like EnEk, k 6¼ m, n 6¼ m).
The total electrical field Em and receiver current I(t), observing only the in-band crosstalk, can be written as [2, 4, 99]:
!
X
Em ðt Þ ¼ Em þ En ejωm t ,
n6¼m
ð2:876Þ
2
X
N pffiffiffiffiffiffiffiffiffiffiffi
I ðt Þ ¼ RjE m ðt Þj  RPm ðt Þ þ 2R Pm Pn cos ½ϕm ðt Þ  ϕn ðt Þ
n6¼m

Fig. 2.58 Illustration of the sources of interchannel crosstalk: (a) demux and (b) the optical switch
144 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.59 Illustration of the sources of intrachannel crosstalk: (a) a demux-mux pair and (b) the optical switch

Bias addition Ambient light

Input OFDM OFDM Output

S/P P/S
data transmitter receiver data
IR link

Direct modulated Direct

LD photodetector

Fig. 2.60 A typical OFDM-based optical wireless communication link with direct modulation and direct detection. LD laser diode

where ϕn(t) denotes the phase of nth channel and R is the photodiode responsivity. Each term acts as an independent random
variable, and the crosstalk effect can be observed as the intensity noise (i.e., Gaussian for large N ):

I ðt Þ ¼ RðPm þ ΔPÞ ðPn << Pm , n 6¼ mÞ, ð2:877Þ

where ΔP is the intensity noise due to intrachannel crosstalk.

2.6 Indoor Optical Wireless Communication Channels

Optical wireless communications (OWCs) [103, 118–124] represent a viable and promising supplemental technology to next-
generation wireless technology (5G). OWCs offer several attractive features, compared to radio, such as [4, 125] (i) huge
available unregulated bandwidth, (ii) systems that do not introduce electromagnetic interference to existing radio systems, (iii)
optical wireless signal that does not penetrate through the walls so that the room represents an optical wireless cell, (iv) the
same equipment that can be reused in other cells without any inter-cell interference, (v) simple design of high-capacity LANs
because coordination between neighboring cells is not needed, and (vi) their inherent applicability in environments when
radio systems are not desired (airplanes, hospitals, etc.).
Majority of OWCs are designed to operate in the near-infrared (IR) region at wavelength of either 850 nm or 1550 nm. The
research in this area is getting momentum, and there are a number of possible point-to-point applications including short-range
links described in the IrDA standard.
In this section, we describe the basic concepts of IR OWCs, different techniques to improve the link performance such as
diversity and adaptive signal processing, and potential use of visible light communications (VLCs) [121–124].

2.6.1 Infrared Optical Wireless Communications

The intensity modulation and direct detection is considered by many authors, and it is only a viable technology for both IR
OWCs and VLCs. One typical OFDM-based WOC link is shown in Fig. 2.60. Because of unguided propagation, the OWC
2.6 Indoor Optical Wireless Communication Channels 145

links faces significant free-space loss. Similarly as wireless indoor links, the most important factor in performance degradation
is the multipath effect. Namely, in addition to the LOS component, several diffuse reflection-based components arrive at
receiver side. Because the direct detection is used, we do not worry about co-phasing of different multipath components.
However, since high-speed signals are transmitted over an IR link, different multipath components arrive in time slots of
neighboring symbols causing the ISI. To deal with ISI effect, different methods can be classified into three broad categories:
(i) equalization, (ii) multicarrier modulation (such as OFDM), and (iii) spread spectrum. Multicarrier modulation can also be
combined with diversity and multi-input multi-output (MIMO) techniques. The dominant noise source in IR links is the
ambient light, which is responsible for the shot noise component. Artificial light sources also introduce harmonics up to
1 MHz in the transmitted signal. When the ambient light is weak, the receiver pre-amplifier is predominant noise source.
Fortunately, most of the ambient noise can be removed by proper optical filtering. A typical receiver for OWC consists of an
optical front end, with an optical concentrator being used to collect the incoming radiation, an optical filter (to reduce the
ambient light), a photodetector, and an electrical front end. The OFDM-based OWC receiver contains an OFDM receiver. By
reducing the number of subcarriers to one, the phase-modulated signals can be transmitted over direct detection links. For
intensity modulation with direct detection (IM/DD), the bias addition circuit should be omitted and OFDM transmitter
replaced by a NRZ/RZ/pulse-position modulator. On receiver side, for IM/DD, OFDM transmitter should be replaced by the
corresponding RF demodulator.
Different link types used in OWC systems, shown in Fig. 2.61, can be classified as [120] point-to-point links (there exists a
LOS component), diffuse links (transmitter emits optical power over a wide solid angle; emitted light gets reflected from
ceiling so that receiver gets illuminated from diffused light), and quasi-diffuse links (the transmitter illuminates the ceiling
with a series of slowly diverging beam sources). The quasi-diffuse links are also known as non-LOS links. The point-to-point
links may contain both LOS component and diffused reflected multipath components (see Fig. 2.61(d)). The channel transfer
function can be written as follows:

H ð f Þ ¼ H LOS þ H diff ð f Þ, ð2:878Þ

where HLOS corresponds to the LOS path amplitude gain and Hdiff( f ) represents the contribution of diffuse reflected multipath
components, which can be approximated as a low-pass filtering function [119]. When the room is homogeneous and isotropic,
the Rician factor, introduced in section on wireless channel modeling, can be used to quantify the ratio between the LOS and
diffuse components as follows K ¼ (HLOS/Hdiff)2.
When the LOS link is reasonably short, and the LED is used as the source and the non-imaging concentrator to collect the
light; we can model the transfer function of LOS as follows [124, 125]:

H LOS ¼ AR ðm þ 1Þ cos m ϕ t s ðψ Þgðψ Þ cos ψ= 2πr 2 , 0

ψ
ψ FOV , ð2:879Þ

where AR is the active area, ϕ is the transmit angle with respect to the maximum radiation direction, while ψ and r are the angle
of irradiance and the distance to illuminated surface, respectively, as shown in Fig. 2.62. In (2.879), ψ FOV denotes the field-of-
view (FOV) angle. Parameter m denotes the Lambert index related to the source radiation semi-angle by m ¼ 1/log2(cosθ1/
2), with θ 1/2 being the IR LED semi-angle at half-power. In (2.879), we use ts(ψ) to denote the transmission of an optical filter
and g(ψ) to denote the ffi of non-imaging concentrator g(ψ) ¼ n /sin ψ FOV, 0
ψ
ψ FOV (n is the refractive index).
gain
pffiffiffiffiffi
2 2

Providing that r >> AR , the corresponding impulse response of the LOS component is given by:

hLOS ðt Þ ¼ AR ðm þ 1Þ cos m ϕ t s ðψ Þgðψ Þ cos ψ= 2πr 2 δðt  r=cÞ, 0

ψ
ψ FOV , ð2:880Þ

where r/c is clearly the propagation delay.

For the point-to-pint links and IM/DD, the Rx signal, after the photodetection takes place, can be determined as the
convolution of transmitted signal s(t) and impulse response of the channel as follows:
ð
rðtÞ ¼ R sðtÞ  ½hLOS ðtÞ þ hdiff ðtÞ þ nðtÞ ¼ R1 sðt 0 Þ½hLOS ðt  t 0 Þ þ hdiff ðt  t 0 Þdt 0 þ nðtÞ, ð2:881Þ
1

where R is the photodiode responsivity, hdiff(t) ¼ FT1{Hdiff( f )} is the diffuse components impulse response, and n(t) is the
noise originating from ambient noise and thermal noise of transimpedance amplifier that typically follows the photodetector.
146 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

Fig. 2.61 Different link types

in OWCs: (a) a point-to-point Output
Input
link, (b) a diffuse link, and (c)
quasi-diffuse link (non-LOS link). data data
(d) In point-to-point links,
received signal may contain both Laser diode Photodiode
LOS component and diffusion or LED
reflected components

(a)

Receiver 1 Transmitter Receiver 2

(b)

Receiver 1 Transmitter Receiver 2

(c)
Opaque surface

Diffuse reflection
component

LOS component
Transmitter Receiver
(d)

The i-th reflected component of a quasi-diffuse link, shown in Fig. 2.63, can be considered as a generalization of the LOS
propagation model from Fig. 2.62. The corresponding impulse response of the i-th reflected component is given by:

AR Aref,i r refl,i r þ r Rx,i

hrefl,i ðtÞ ¼ ðm þ 1Þcosm ϕTx,i cos ðψ Tx,i Þt s ðψ Rx,i Þgðψ Rx,i Þcos ψ Rx,i δðt  Tx,i Þ, ð2:881Þ
2πr 2Tx,i r 2Rx,i c

where rrefl,i is the reflection coefficient, Aref,i is the area of the reflector, and corresponding angles are shown in Fig. 2.63. The
transmission and gain functions are the same as defined above.
2.6 Indoor Optical Wireless Communication Channels 147

Fig. 2.62 Illustrating the LOS propagation model for IR-LED (Lambertian source)

Fig. 2.63 Illustrating the i-th reflected component of the quasi-diffuse link

2.6.2 Visible Light Communications (VLCs)

VLCs represent a relatively new research area [121–123] with operating wavelengths of 380–720 nm by employing white-
light LEDs with available bandwidth of around 20 MHz. With visible-light LEDs (VLEDs), it would be possible to broadcast
broadband signals in various environments such as offices where the lighting is used already. Power line communication
systems or plastic optical fiber systems can be used as feeders. VLCs can also be used in car-to-car communications by
employing LED-based head and tail lights and probably in airplanes to provide high-speed internet. The VLC systems model
is similar to that shown in Fig. 2.60; white-light LED is used instead of laser diode. The VLC channel model can be
represented by:

X exp ðj2πf Δτdiff Þ

Hð f Þ ¼ H LOS , i exp ðj2πf ΔτLOS,i Þ þ H diff , ð2:882Þ
1 þ j f = f cutoff
i |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl}
H diff ð f Þ

where HLOS,i and Hdiff represent the channel gain of i-th LOS component (coming from ith LED) and diffuse signal gain,
respectively. The Δτ LOS,i and Δτdiff represent the corresponding delays of i-th LOS component and diffuse signal
148 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

components. Frequency f cutoff denotes the cutoff frequency of diffuse channel frequency response, and for medium-sized
room, it is about 10 MHz [123]. The LOS gain from ith LED chip is related to the luminous flux:
ð 780 nm
Φ ¼ K max Φe ðλÞV ðλÞdλ, ð2:883Þ
380 nm

where Φe(λ) is the source power distribution, V(λ) is the standard sensitivity function, and Kmax is the maximum visibility
(683 lm/W at 555 nm). The corresponding luminous intensity is the first derivative of the luminous flux with respect to the
solid angle Ω; in other words, I ¼ dΦ/dΩ. The radiation intensity at the desk surface is given by:

I ðϕÞ ¼ I ðϕ ¼ 0Þ cos m ðϕÞ, ð2:884Þ

where ϕ is the transmit angle with respect to the maximum radiation direction and m denotes the Lambert index related to the
source radiation semi-angle by m ¼ 1/log2(cosθ1/2), with θ1/2 being the VLED semi-angle at half-power. Now, the LOS gain
from the i-th white-LED is given by:

H LOS , i ¼ AR ðm þ 1Þ cos m ðϕi Þt s ðψ i Þgðψ i Þ cos ðψ i Þ= 2πr 2i , ð2:885Þ

where AR is the active area, ϕi is the transmit angle with respect to the maximum radiation direction of the i-th light source,
while ψ i and ri are the angle of irradiance and the distance to illuminated surface, respectively. As expected, this expression is
very similar to Eq. (2.879). Figure 2.62 is still applicable here, where Tx is now white LED. In (2.885), we use ts(ψ i) to denote
the transmission of the optical filter and g(ψ i) to denote the gain of non-imaging concentrator g(ψ i) ¼ n2/sin2ψ con, 0
ψ
ψ con
(n is the refractive index and ψ con is the maximum angle of the concentrator). The corresponding impulse response of the i-th
LOS component is given by:

AR ðm þ 1Þ   pffiffiffiffiffiffi
r
hLOS , iðt Þ ¼ cos m ðϕi Þt s ðψ i Þgðψ i Þ cos ðψ i Þδ t  i , r i AR ð2:886Þ
ð2πr i Þ
2 c

The diffuse signal magnitude Hdiff in (2.882) can commonly be modeled as being constant, in average, anywhere in the
room, and it is related to the effective receiver surface AR, the room size Aroom, and average reflectivity rrefl by [123]:

H diff ¼ AR r refl =Aroom ð1  r refl Þ: ð2:887Þ

For the point-to-pint VLC links and IM/DD, the Rx signal, after the photodetection takes place, can be determined as the
convolution of transmitted signal s(t) and impulse response of the VLC channel as follows:
hX i
r ðt Þ ¼ R sðt Þ  i
h LOS,i ð t Þ þ h diff ð t Þ þ nðt Þ
ð
1
hX i ð2:888Þ
¼R sðt 0 Þ i
h LOS,i ð t Þ þ h diff ð t  t 0
Þ dt 0 þ nðt Þ,
1

where R is the photodiode responsivity and hdiff(t) ¼ FT1{Hdiff( f )}.

2.7 Concluding Remarks

This chapter has been devoted to propagation effects, noise sources, and channel impairments in wireless, free-space optical,
and fiber-optics communications systems. The key idea behind the chapter has been to study the propagation effects in
different channel types based on Maxwell’s equations. The chapter has been organized as follows. In Sect. 2.1, basic concepts
from electromagnetic fields and wave equations have been described. In this section, the vector derivatives needed to study the
propagation over different media have been described in generic orthogonal curvilinear coordinate system, for which
Cartesian, cylindrical polar, and spherical coordinate systems have been just special cases. In Sect. 2.2.1, we have studied
2.8 Problems 149

the vectorial nature of the light, derived Snell’s law of refraction from Maxwell’s equations, studied total internal reflection,
and explained the light confinement in step-index optical fibers. Additionally, we have described how arbitrary polarization
state can be generated by employing three basic polarizing elements, namely, polarizer, phase shifter (retarder), and rotator. In
Sect. 2.2.3, we have introduced the electromagnetic potentials to simplify the study of generation of electromagnetic waves,
followed by antenna theory basics. In Sect. 2.2.4, we have discuss the interference, coherence, and diffraction effects in optics,
which have been used later to study the propagation effects in various wireless and optical communication channels. In Sect.
2.2.5, the detailed description of the laser beam propagation effects over the atmospheric turbulence channels has been
provided, including the description of simulation models. In Sect. 2.3, the wireless communication channel propagation
effects have been studied, in particular the path loss, shadowing, and multipath fading propagation effects. Both ray theory-
based and statistical multipath wireless communication channel models have been described. Section 2.4 has been devoted to
the signal propagation effects in optical fibers, namely, (i) fiber attenuation and insertion loss, (ii) chromatic dispersion effects,
(iii) polarization mode dispersion effects, and (iv) fiber nonlinearities. Regarding the fiber nonlinearities, both Kerr nonlinear
effects (self-phase modulation, cross-phase modulation) and nonlinear scattering effects, such as stimulated Raman scattering,
have been described. In Sect. 2.4.5, the generalized nonlinear Schrödinger equation has been described for both single-mode
and few-mode fibers, as well as the corresponding split-step Fourier algorithms to solve for it. In Sect. 2.5, the following noise
sources related to optical communication have been described: mode partition noise, relative intensity noise, laser phase noise,
modal noise, thermal noise, spontaneous emission noise, noise beating components, quantum shot noise, dark current noise,
and crosstalk. Section 2.6 has been devoted to indoor optical wireless communications. In this section, the channel models for
both infrared and visible light communications have been introduced. The set of problems is provided in next section.

2.8 Problems

1. The beam expander is composed of two lenses, the first one of short focal length fs and the second one of long focal length
fl, with the goal to expand the beam.
(a) Sketch the schematic of the beam expander.
(b) How much should these two lenses be separated so that for collimated incident beam the expanded beam to be
collimated as well?
(c) Relate the expansion ratio to the focal lengths.
(d) Design the beam expander with expansion ratio of 4
.
(e) Design the corresponding beam compressor.
2. Let us observe a rectangular aperture of width a ¼ 1 mm and height b ¼ 2 mm that get illuminated by normally incident
plane wave with wavelength 1550 nm.
(a) Determine the distance from the aperture when Fresnel approximation is valid.
(b) Determine the distance from the aperture when Fraunhofer approximation is valid.
3. Let us consider the double-slit aperture composed of two parallel slits of width a each and separated by distance h.
Assuming the double-slit is illuminated by normally incident plane wave.
(a) Derive the irradiance distribution function and plot corresponding Fraunhofer diffraction pattern.
(b) Generalize this case to the multiple slits’ aperture, composed of N parallel slits of width a with separation between
slits being h. Derive the irradiance distribution function and plot corresponding Fraunhofer diffraction patterns in
which number of slits N is used as a parameter.
4. Let a transparent object of transmittance function tobj be placed against the ideal biconvex lens of focal length f. Derive the
diffraction expression assuming the propagation to the back focal plane and normally incident plane wave beam.
5. Let a transparent object of transmittance function tobj be placed in front of, at distance d from, the ideal biconvex lens of
focal length f. Derive the diffraction expression assuming the propagation to the back focal plane and normally incident
plane wave beam. For d ¼ 0 your solution should reduce to that from previous problem.
6. Let a transparent object of transmittance function tobj be placed behind the ideal biconvex lens of focal length f, placed at
distance d-f, wherein d < <f. Derive the diffraction expression assuming the propagation to the back focal plane and
normally incident plane wave beam.
7. In this problem you are expected to study the propagation of Gaussian and LG beams over the atmospheric turbulent
channel, using the split-step algorithm. You are expected to take the diffraction effects into account. To study the FSO
propagation, the following parameters of atmospheric turbulence channel should be considered: the propagation distance
1 km, the outer scale eddies L0 ¼ 20 m, the inner scale l0 ¼ 1 mm, the number of slabs at least 10, operating wavelength
150 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

1550 nm, and the beam waist at z ¼ 0 being 5 mm. The phase power spectral density (PSD) to be used to generate random
phase screens should be a modified version of the Kolmogorov spectrum, which includes both inner and outer scales, also
known as Andrews’ spectrum. Hint: To verify your split-step method, you can use the phase structure function. As an
example, you can use S. S. Chesnokov, V. P. Kandidov, V. I. Shmalhausen, and V. V. Shuvalov, “Numerical/optical
simulation of laser beam propagation through atmospheric turbulence,” DTIC Document (1995).
(a) Provide the phase screen realizations after 1 km, for the following refractive structure parameters Cn2 ¼ 1015, 1014,
and 1013 m-2/3. For Gaussian beam, provide also corresponding magnitude and phase characteristics for the same set
of refractive structure parameters.
(b) This problem is continuation of (a). Determine the Rytov variance and scintillation index for the same set of refractive
structure parameters. Determine the distributions of irradiance and compare them against gamma-gamma distribution.
(c) This problem is continuation of (a). Determine the intensity correlation function (ICF) defined as C-
(ρ1, ρ2, L ) ¼ h(I1  hI1i)(I2  hI2i)i, where I1 and I2 are the captured irradiances of positions ρ1 and ρ2 at distance
L, respectively; the operator hi denotes an ensemble average of the quantity inside the parenthesis.
(d) The field distribution of the LG beam traveling along the z-axis can be expressed in cylindrical coordinates [ρ,ϕ,z] (ρ
denotes the radial distance from propagation axis, ϕ denotes the azimuthal angle, and z denotes the propagation
distance) as: rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  pffiffiffijmj  2  ρ2
2p! 1 ρ 2 2ρ 
um,p ðρ, ϕ, zÞ ¼ Lm
p e w 2 ðzÞ
π ðp þ jmjÞ! wðzÞ wðzÞ w 2 ðzÞ
jkρ2 z

e ð Þe
2 z2 þz2 jð2pþjmjþ1Þ tan 1 zz jmϕ
R eR ,

where w(z) ¼ w0[1 + (z/zR)2]1/2 (w0 is the zero-order Gaussian radius at the waist), zR ¼ πw20 =λ is the Rayleigh range
(with λ being the wavelength), k ¼ 2π/λ is the propagation constant, and Lmp() is the associated Laguerre polynomial,
with p and m representing the radial and angular mode numbers (indices), respectively.
(e) For m = 3 and p = 0, provide the corresponding spatial irradiance and phase characteristics for the same set of
refractive structure parameters as in (a).
8. In this problem we study a two-path wireless channel model with impulse response h(t) ¼ α1δ(τ) + α2δ(τ  0.03 μs).
Under assumption that the reflection coefficient is 1, determine coefficients in impulse response as well as the distance
between the transmitter and receiver. Assume that the carrier frequency is 1 GHz, while the transmitter and receiver are
both 10 m above the ground.
9. In this problem we are concerned with four-ray wireless channel model provided in Fig. P9, composed of line-of-sight
(LOS), reflected, scattered, and diffracted components.
The height of transmit (Tx) antenna measured from the road, as indicated in Fig. P9, is hTx ¼ 75 m, while the receive
(Rx) antenna height is hRx ¼ 1.5 m. The distance between the picks of Tx and Rx antennas is L ¼ 350 m. The distance

Diffracted
component

45o Scaered
component
LOS 30o
30o
Reflected 45o
component

Fig. P9 Illustration of the four-ray model

2.8 Problems 151

from Tx antenna to the diffracting object is d ¼ 150 m, while the distance from diffracting object to Rx antenna is
D ¼ 250 m. The distance from Tx antenna to the scattering object is r ¼ 200 m, while the distance from scattering object
to Rx antenna is R ¼ 300 m. The scattering object radar cross section is σ ¼ 10 dBm2. Assume that dielectric constant of
the road is εr ¼ 1.5 and that reflected ray is TE polarized. The carrier frequency is 1 GHz.
(a) Determine the receiver-to-transmit power ratio for this four-ray problem.
(b) We declare an outage when then received power falls below 30 dBm. Assuming that the transmitted power is
500 mW and the shadowing effect with standard deviation of 3 dB is present, determine the outage probability.
(c) For the same assumptions as in (b), when the cell radius is 500 m, determine the cell coverage area.
10. Let us study a cellular system in which the received signal power is distributed according to the log-normal distribution
with a mean μψ dB dBm and a standard deviation σ ψ dB dBm. Under the assumptions that the acceptable performance is
achieved for received signal power above 8 dBm, determine:
(a) The outage probability for μψ dB ¼ 12 dBm and σ ψ dB ¼ 5 dBm.
(b) The value of μψ dB required, for σ ψ dB ¼ 4.5 dBm, such that the outage probability <1%.
11. Determine the coverage area for a microcellular system with path loss that follows the simplified model with slope s ¼ 3,
lref ¼ 1 m, and C ¼ 1 in the presence of log-normal shadowing with σ ψ dB ¼ 5 dB under the following assumptions: the cell
radius is 90 m, the transmit power is 60 mW, and the minimum received power requirement is 90 dBm.
12. Let the scattering function S(τ, ν) be nonzero over 0
τ
0.1 ms and  0.1
ν
0.1 Hz. Assume that the distribution of
power of the scattering function is approximately uniform over the nonzero range.
(a) Determine the multipath spread and the Doppler spread of the channel.
(b) Suppose that the input to this channel are two sinusoids with the same amplitude separated in time by Δt. Determine
the minimum value of Δf for which the channel response to the first sinusoid is approximately independent of the
channel response to the second sinusoid.
(c) For two sinusoidal inputs to the channel given u1(t) ¼ sin2πft and u2(t) ¼ sin[2πf(t + Δt)], determine the minimum
value of Δt for which the channel response to the first sinusoid u1(t) is approximately independent of the channel
response to the second sinusoid u2(t).
(d) For a typical voice channel with a 3 KHz bandwidth, classify the fading type.
(e) For a cellular channel with a 30 KHz bandwidth, classify the fading type.
13. The purpose of this problem is to develop a Rayleigh fading simulator employing the Clark-Jakes model. The
corresponding program should be easy to follow and well commented. The concepts to implement this simulator are
already provided in Sect. 2.3.2.3.
(a) Develop the Rayleigh fading simulator to generate the signal with Rayleigh fading amplitude in which the average
received envelope is normalized to one and the Doppler frequency νD is used as a parameter.
(b) Plot the received amplitude vs. time over a 5-second interval for the following Doppler frequencies νD2{1, 5, 10,
20, 50, 100 Hz}.
14. In this problem we are interested in approximating the Rayleigh fading simulator developed in previous problem with a
finite state Markov model with sufficient number states and sufficient number of envelope regions. For each Doppler
frequency listed in Problem 10(b), determine the corresponding finite state Markov model and provide the transition
probabilities. By using such developed models, generate fading amplitude vs. time and report the mean-squared error
compared to the simulator.
15. The fabrication of the single-mode fiber (SMF) is imperfect, and we need to associate with it certain fabrication
tolerances. Under the assumptions that the tolerance in the core radius is 3%, while the tolerance in the normalized
refractive index difference Δ is 7% (from their respective nominal values, Δ ¼ 0.005, refractive index of the core
ncore ¼ 1.5), determine the largest nominal value that can specified for the core radius ensuring that the fiber operates as a
single-mode for λ > 1.2 μm in the worst-case scenario of deviations of core radius and Δ.
16. A step-index multimode fiber has the cladding refractive index of 1.45 and a core diameter of 45 μm. Assuming that the
intermodal dispersion of 12 ns/km can be tolerated, determine the acceptance angle. Determine the maximum bit rate for
transmission over a distance of 15 km.
17. For a Gaussian pulse of peak power 10 mW and full-width half-maximum pulse duration of 20 ps, transmitted over the
SMF, determine the frequency chirp due to the SPM.
18. Assuming that the step-index fiber has core radius r, cladding refractive index ncladding, and core refractive index ncore,
derive the corresponding equation for the cutoff wavelength λcutoff. For a fiber of core radius of r ¼ 5 μm and Δ ¼ 0.004,
determine the cutoff wavelength.
152 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

19. Assuming that the fundamental mode can be approximated as a Gaussian beam, determine the confinement factor, defined
as the ratio of the power in core and the total power (both in core and cladding), as a function of core radius and the beam
width.
20. The broadening of the unchirped Gaussian pulse [with an envelope A(0,t) ¼ A0exp(t2/2τ02), A0-the pulse amplitude, τ0-
the pulse half-width at 1/e intensity point] in the presence of GVD and SPM can be evaluated as the ratio of output and
input RMS widths:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
! ffi
u pffiffiffi
σ ðL Þ tu 2Leff Lβ2 4 L 2
eff L β2
2 2
¼ 1þ þ 1 þ pffiffiffi 2 , σ 20 ¼ τ20 =2
σ0 2LNL σ 20 3 3 LNL 8σ 40

where with Leff and LNL are denoted effective and nonlinear lengths.
(a) For which value of L (expressed as a multiple of dispersion length LD) does the launched pulse attain its minimum
width?
(b) Observe a RZ transmission at 10 Gb/s with infinitely large extinction ratio and 50% of duty cycle. The bit one is
represented as Gaussian pulse from (a). Determine the initial pulse width that corresponds to 50% duty cycle. For
such determined initial pulse width and for length determined in (a), find the maximum bit rate that can be supported.
Assume that system operates at wavelength 1550 nm, for which the fiber has the dispersion coefficient D ¼ 17 ps/
(nm-km) and the Kerr nonlinearity coefficient n2 ¼ 31020 m2/W. The fiber attenuation is 0.2 dB/ km, and the
Gaussian pulse peak power is 0 dBm.
21. This problem is related to the link design. We would like to design the system operating either at 1300 nm or 1550 nm,
which supports 2.5 Gb/s data rates using NRZ modulation format. The fiber parameters are as follows. The dispersion at
1550 nm is 16.5 ps/nm-km, while the attenuation coefficient is 0.29 dB/km. The dispersion at 1300 nm is 0 ps/nm-km,
while the attenuation coefficient is 0.5 dB/km.
(a) If (i) a transmitter operating at 1550 nm with spectral width of 1 nm and with launched power of 0 dBm is available
and (ii) receiver with receiver sensitivity of 29 dBm is available, determine the maximum possible length of the link.
(b) If (i) a transmitter operating at 1300 nm with spectral width of 2 nm and with launched power of 3 dBm is available
and (ii) receiver with receiver sensitivity of 29 dBm is available, determine the maximum possible length of the link
in this case.
(c) Let’s assume that transmitter from (b) is available. We are asked to achieve a signal-to-noise ratio (SNR) of 30 dB,
using an APD of responsivity 10 A/W and gain of 10. The excess noise factor is 6 dB, the load resistance is 50 Ω, and
an amplifier noise figure is 4 dB. Assuming the NRZ transmission, determine the maximum transmission length.
22. We would like to design a fiber-optics 160 Gb/s wavelength division multiplexing (WDM) system with M ¼ 4 channels
operating at 40 Gb/s employing the hybrid Raman-EDFA in-line optical amplifiers, as shown in Fig. P22. The total
launched power is restricted by available laser sources and cannot be larger than 9 dBm, and the SMF fiber loss is
α ¼ 0.2 dB/km. SMF dispersion coefficient is 17 ps/nm-km. Dispersion compensating module (DCM) is composed of
DCF with dispersion coefficient of 95 ps/nm-km, while DCF attenuation coefficient is 0.5 dB/km. Notice that DCF
length does not contribute to the total transmission length. Raman amplifier (RA) is used to compensate for SMF fiber
loss, while EDFA is used to compensate for DCM insertion loss. Assuming that double Rayleigh backscattering can be
neglected, the RA noise figure can be calculated by NFRA ¼ (PASE/hvsBop + 1)/(GRAe-αl), where GRA is RA gain (Bop is
the optical filter bandwidth). The ASE noise power of RA can be calculated by PASE ¼ 2 hvsΔvR/[1-exp(hvs/kBT )], kB is
the Boltzmann constant (1.381
1023J/K), vs is the signal frequency, and ΔvR is the separation between signal and
pump frequencies. The pump wavelength is 1450 nm. T is the absolute temperature. Assume that GRA is just enough to
compensate for SMF fiber loss. The EDFA noise figure (NFEDFA) is 6 dB. The equivalent noise figure of both RA and
EDFA can be calculated by NF¼NFRA + (NFEDFA1)/(GRAe-αl) + 1-Γ, where Γ is loss caused by DCM. The predicted
margins to achieve BER ¼ 1012 allocated for chromatic dispersion, extinction ratio, polarization effects, fiber
nonlinearities, component aging, and system margin are 0.5, 0.5, 1.5, 1.5, 2, and 2 dB, respectively. The operating
temperature is 25  C. WDM DEMUX can be modeled as a bank of optical filters, each of them with a bandwidth Bo equal
to 0.3B (B-the bit rate per channel). Assuming NRZ transmission, the electrical filter bandwidth Be ¼ 0.7B, and amplifier
spacing l ¼ 100 km, determine the required optical signal-to-noise ratio (OSNR) to achieve BER of 1012 for the central
channel located at 1552.524 nm. What is the maximum possible transmission distance? Assume that the saturation of
EDFAs due to ASE noise accumulation is negligible.
References 153

Nx
SMF

Dispersion
EDFA
compensating module

Forward
pumping

Fig. P22 hybrid Raman-EDFA concept

23. In this problem you are expected to solve the GNLSE numerically, using Fourier split-step algorithm.

(a) Let us first ignore the attenuation and fiber nonlinearities. Generate the Gaussian pulse of FWHM 25 ps and launch
power 1 mW and transmit it over SMF of length L. If the dispersion coefficient is 16 ps/(nm km) and dispersion slope
is 0.06 ps/(nm2 km), determine and plot the pulse shape after 25 km, 50 km, and 100 km by solving the GNLSE.
Determine the pulse spread. Operating wavelength is 1550 nm.
(b) Let us ignore the attenuation and second-order GVD, and study the soliton transmission. Assume that dispersion
coefficient is 16 ps/(nmkm) and nonlinear coefficient of fiber is γ ¼ 2.5 1/(Wkm). Chose the launch power such that
dispersion length is equal to the nonlinear length. Determine the analytical expression of pulse shape. Plot the pulse
shape after 1000 km and 10,000 km by solving the NLSE. The initial soliton pulse FWHM is 25 ps. Operating
wavelength is again 1550 nm.
(c) Generate now a PRBS sequence of sufficient length to accurately estimate BER down to 106. Represent the symbol
1 pulse by a Gaussian pulse of FWHM of 12.5 ps. Assume the launch power of 0 dBm, the extinction ratio of 20 dB,
and data rate of 40 Gb/s. Transmit this sequence over 6000 km, with dispersion map composed of N sections of SMF
of length 80 km (with dispersion and dispersion slope parameters being 16 ps/(nmkm) and 0.06 ps/(nm2 km),
respectively) and 40 km of DCF. The DCF is chosen to exactly compensate for residual chromatic dispersion, and
dispersion map is slope matched. The nonlinear coefficient of both fibers is γ ¼ 2.5 1/(Wkm). The EDFAs of NF of
5 dB and BW ¼ 3Rb (Rb is the bit rate) are periodically deployed every fiber section to exactly compensate for fiber
loss. The SMF loss is 0.2 dB/km, and DCF loss is 0.5 dB/km. The operating wavelength is 1550 nm, and photodiode
responsivity is 1 A/W. The receiver electronics can be model as Gaussian filter of 3 dB BW ¼0.75 Rb. Determine the
optical SNR (OSNR) at the receiver side as well as the BER. Show the eye diagrams and spectra at transmitter and
receiver sides. Vary the launch power from 6 dBm to 6 dBm in steps of 2 dBm and plot how BER changes as a
function of launch power.

References
1. G. Keiser, Optical Fiber Communications, 3rd edn. (McGraw-Hill, Boston, 2000)
2. G.P. Agrawal, Fiber-Optic Communication Systems, 4th edn. (Wiley, Hoboken, NJ, 2010)
3. G.P. Agrawal, Nonlinear Fiber Optics, 4th edn. (Academic Press, Oxford, UK, 2007)
4. I.B. Djordjevic, W. Ryan, B. Vasic, Coding for Optical Channels (Springer, New York, 2010)
5. D.M. Velickovic, Electromagnetics – Vol. I, 2nd edn. (Faculty of Electronic Engineering, University of Nis, Nis, 1999)
6. K.F. Riley, M.P. Hobson, S.J. Bence, Mathematical Methods for Physics and Engineering: A Comprehensive Guide, 3rd edn. (Cambridge
University Press, Cambridge, New York, 2006)
7. I.B. Djordjevic, Quantum Information Processing and Quantum Error Correction: An Engineering Approach (Elsevier/Academic Press,
Amsterdam, Boston, 2012)
8. C.H. Papas, Theory of Electromagnetic Wave Propagation (Dover, New York, 1988)
9. E. Collet, Polarized Light for Scientists and Engineers (The PolaWave Group, Long Branch, NJ, 2012)
10. G.R. Fowles, Introduction to Modern Optics (Dover, New York, 1989)
11. E. Hecht, A.R. Ganesan, Optics (Pearson Education, Essex, 2014)
12. N. George, Fourier Optics (Institute of Optics/University of Rochester, 2013)
13. J.W. Goodman, Introduction to Fourier Optics, 3rd edn. (Roberts/Greenwood Village, 2005)
14. M. Abramowitz, I. Stegun, Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables (Dover, New York, 1965)
15. J.L. Walsh, P.B. Ulrich, Laser Beam Propagation in the Atmosphere (Springer, Berlin/Heidelberg/New York, 1978)
16. R. Buckley, Diffraction by a random phase-change in screen: a numerical experiment. J. Atm. Terrestrial Phys. 37(12), 1431–1446 (1975)
154 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

17. J.A. Fleck, J.R. Morris, M.D. Feit, Time-dependent propagation of a high-energy laser beam through the atmosphere. Appl. Phys. 10(1),
129–139 (1976)
18. V.P. Kandinov, V.I. Ledenev, Application of the method of statistical testing for investigation of wave beam propagation in statistically random
media. IVUZ Ser. Radiofiz. 24(4), 434–442 (1981)
19. P.A. Konayev, Modification of the splitting method for numerical solution of quasi-optical problems, in Abstracts of reports at the VI
conference on laser beam propagation in the atmosphere (1981), pp. 195–198
20. V.P. Lukin, B.V. Fortes, Adaptive Beaming and Imaging in the Turbulent Atmosphere (SPIE Press, Bellingham, 2002)
21. J.D. Schmidt, Numerical Simulation of Optical Wave Propagation with examples in MATLAB (SPIE Press, Bellingham, 2010)
22. J.M. Jarem, P.P. Banerjee, Computational Methods for Electromagnetic and Optical Systems (Marcel Dekker, New York, 2000)
23. T.-C. Poon, P.P. Banerjee, Contemporary Optical Image Processing with Matlab (Elsevier Science, Oxford, 2001)
24. T.-C. Poon, T. Kim, Engineering Optics with MATLAB (World Scientific Publishing, New Jersey/London/Singapore/Beijing, 2006)
25. J.J. Sakurai, S.F. Tuan, Modern Quantum Mechanics, Revised edn. (Addison-Wesley Publishing, Reading, 1994)
26. M. Nazarathy, J. Shamir, Fourier optics described by operator algebra. J. Opt. Soc. Am. 70, 150–159 (1980)
27. L. Stankovic, Digital Signal Processing (Naucna Knjiga, Belgrade, 1990)
28. Fastest Fourier Transform in the West, FFTW. Available at: http://www.fftw.org/
29. S.M. Rytov, Introduction to Statistical Radiophysics (Nauka, Moscow, 1966)
30. L.C. Andrews, R.L. Philips, Laser Beam Propagation Through Random Media, 2nd edn. (SPIE Press, Bellingham, 2005)
31. H. Willebrand, B.S. Ghuman, Free-Space Optics: Enabling Optical Connectivity in Today’s Networks (Sams Publishing, Indianapolis, 2002)
32. A. Ishimaru, Wave Propagation and Scattering in Random Media (Academic, NY, 1978)
33. L.C. Andrews, R.L. Phillips, C.Y. Hopen, Laser Beam Scintillation with Applications (SPIE Press, 2001)
34. S. Karp, R. Gagliardi, S.E. Moran, L.B. Stotts, Optical Channels (Plenum, New York, 1988)
35. J.A. Anguita, I.B. Djordjevic, M.A. Neifeld, B.V. Vasic, Shannon capacities and error–correction codes for optical atmospheric turbulent
channels. OSA J. Opt. Netw. 4, 586–601 (2005)
36. I.B. Djordjevic, B. Vasic, M.A. Neifeld, LDPC coded OFDM over the atmospheric turbulence channel. Opt. Express 15(10), 6332–6346
(2007)
37. I.B. Djordjevic, J. Anguita, B. Vasic, Error-correction coded orbital-angular-momentum modulation for FSO channels affected by turbulence.
IEEE/OSA J. Lightw. Technol. 30(17), 2846–2852 (2012)
38. Shieh W, Djordjevic I (2009), OFDM for Optical Communications. Elsevier/Academic Press, Oct 2009
39. I.B. Djordjevic, Coding and modulation techniques for optical wireless channels, in Advanced Optical Wireless Communication Systems,
ed. by S. Arnon, J. Barry, G. Karagiannidis, R. Schober, M. Uysal, (Cambridge University Press, Cambridge, 2012)
40. Z. Qu, I.B. Djordjevic, Coded orbital-angular-momentum-based free-space optical transmission, in Wiley Encyclopedia of Electrical and
Electronics Engineering, (Wiley, Chichester, Feb. 2016)
41. A.N. Kolmogorov, Local structure of turbulence in liquid under very large Rayleigh numbers. Dok. Akad. Nauk SSR 30(4), 299–303 (1941)
(in Russian)
42. A.M. Obukhov, Energy dissipation in spectra of a turbulent stream. Dok. Akad. Nauk SSR 32, 22–24 (1943)
43. A.N. Kolmogorov, The local structure of turbulence in incompressible viscous fluid for very large Rayleigh numbers. Proc. R. Soc. Lond. 434,
9–13 (1991) Translated by V. Levin
44. A.M. Obukhov, Concerning the distribution of energy in the spectrum of a turbulent stream. Izvestiia Akadimii Nauk S.S.S.R. Ser. Geogr.
Geofiz. 4–5 (1941)
45. A.M. Obukhov, Structure of the temperature field in turbulent flow. Izvestiia Akademii Nauk S.S.S.R. Ser. Geogr. Geofiz. 2(1), 58–69 (1949)
46. L.D. Landau, E.M. Lifschitz, Fluid Mechanics, 2nd edn. (Pergamon Press, Oxford, 1982)
47. S.F. Clifford, The classical theory of wave propagation in a turbulent medium, in Laser Beam Propagation in the Atmosphere, ed. by J. W.
Strohbehn, (Springer, New York, 1978), pp. 9–43
48. J.G. Proakis, Digital Communications, 4th edn. (McGraw-Hill, Boston, 2001)
49. S. Haykin, Digital Communication Systems (Wiley, Hoboken, 2014)
50. M.H. Hayes, Statistical Digital Signal Processing (Wiley, New York, 1996)
51. V.I. Tatarskii, The Effects of the Turbulent Atmosphere on Wave Propagation (Translated by Israel Program for Scientific Translations in
1971) (U.S. Dept. of Commerce/National Technical Information Service, Springfield, 1967)
52. L.C. Andrews, An analytical model for the refractive index power spectrum and its application to optical scintillation in the atmosphere. J. Mod.
Opt. 39(9), 1849–1853 (1992)
53. D.L. Fried, Statistics of a geometric representation of wavefront distortion. J. Opt. Soc. Am. 55(11), 1427–1431 (1965)
54. R. Noll, Zernike polynomials and atmospheric turbulence. J. Opt. Soc. Am. 66, 207–211 (1976)
55. B.M. Welsh, A Fourier phase screen generator for simulating anisoplanatic geometries and temporal evolution. Proc. SPIE 3125, 327–338
(1997)
56. R.J. Eckert, M.E. Goda, Polar phase screens: a comparative analysis with other methods of random phase screen generation. In. Proc. SPIE
6303, 630301-1–630301-14 (2006)
57. R.G. Lane, A. Glindemann, J.C. Dainty, Simulation of a Kolmogorov phase screen. Waves Random Media 2, 209–224 (1992)
58. S.M. Rytov, Diffraction of light by ultrasound waves. Ivestiya SSSR Ser. Fizicheskaya 2, 223 (1937)
59. A.M. Obukhov, Effect of weak inhomogeneities in the atmosphere on the propagation of light and sound. Izvestiya SSSR Ser. Geofiz. 2,
155 (1953)
60. L.A. Chernov, Correlation of amplitude and phase fluctuations for wave propagation in a medium with random inhomogeneities. Akusticheskii
Zhurnal 1(1), 89 (1955)
61. L.A. Chernov, Wave propagation in a medium with random inhomogeneities. Izdatel’stvo AN SSSR (1958)
62. V.I. Tatarskii, Light propagation in a medium with random refractive index inhomogeneities in the Markov random process approximation.
Sov. Phys. JETP 29(6), 1133–1138 (1969)
References 155

63. M.A. Al-Habash, L.C. Andrews, R.L. Phillips, Mathematical model for the irradiance probability density function of a laser beam propagating
through turbulent media. Opt. Eng. 40, 1554–1562 (2001)
64. A. Goldsmith, Wireless Communications (Cambridge University Press, Cambridge, 2005)
65. D. Tse, P. Viswanath, Fundamentals of Wireless Communication (Cambridge University Press, Cambridge, 2005)
66. S. Benedetto, E. Biglieri, Principles of Digital Transmission with Wireless Applications (Kluwer Academic/Plenum Publishers, New York,
1999)
67. E. Biglieri, Coding for Wireless Channels (Springer, New York, 2005)
68. T.M. Duman, A. Ghrayeb, Coding for MIMO Communication Systems (Wiley, Chichester, 2007)
69. E.F. Knott, J.F. Shaeffer, M.T. Tuley, Radar Cross Section, 2nd edn. (SciTech Publishing, 2004)
70. M. Lanzagorta, Quantum Radar (Morgan and Claypool Publishers, San Rafael, 2012)
71. N. Amitay, Modeling and computer simulation of wave propagation in lineal line-of-sight microcells. IEEE Trans. Vech. Technol. 41(4),
337–342 (1992)
72. K. Schaubach, N.J. Davis, T.S. Rappaport, A ray tracing method for predicting path loss and delay spread in microcellular environments, in
Proceedings of the IEEE Vehiculat Technology Conference, vol 2,pp. 932–935 (Denver, 1992)
73. M.C. Lawton, J.P. McGeehan, The application of GTD and ray launching techniques to channel modeling for cordless radio systems, in
Proceedings of the IEEE Vehiculat Technology Conference, vol 1, pp. 125–130 (Denver, 1992)
74. W.C.Y. Lee, Mobile Communications Engineering (McGraw-Hill, New York, 1982)
75. L. Tsang, J.A. Kong, K.-H. Ding, Scattering of Electromagnetic Waves: Theories and Applications (Wiley, New York, 2000)
76. S.M.I. Seidel, Introduction to Radar Systems, 2nd edn. (McGraw-Hill, Auckland, 1981)
77. S.Y. Seidel, T.S. Rappaport, S. Jain, M.L. Lord, R. Singh, Path loss, scattering and multipath delay statistics in four European cities for digital
cellular and microcellular radiotelephone. IEEE Trans. Veh. Technol. 40(4), 721–730 (1991)
78. P. Harley, Short distance attenuation measurements at 900 MHz and 1.8 GHz using low antenna heights for microcells. IEEE Sel. Area Comm.
7(1), 5–11 (1989)
79. T. Okumura, E. Ohmori, K. Fukuda, Field strength and its variability in VHF and UHF land mobile services. Rev. Electr. Comm. Lab.
16(9–10), 825–873 (1968)
80. M. Hata, Empirical formula for propagation loss in land mobile radio services. IEEE Trans. Veh. Technol. VT-29(3), 317–325 (1980)
81. European Cooperative in the Field of Science and Technical Research EURO-COST 213, Urban transmission loss models for mobile radio in
the 900 and 1800 MHz bands. Rev., Hague (1991)
82. Hata/Davidson, A Report on technology independent methodology for the modeling, simulation and empirical verification of wireless
communications system performance in noise and interference limited systems operating on frequencies between 30 and 1500 MHz. TIA
TR8 Working Group, IEEE Vehicular Technology Society Propagation Committee (1997)
83. J.S. Lee, L.E. Miller, CDMA Systems Engineering Handbook (Artech House, Boston/London, 1998)
84. W.C. Jakes, Microwave Mobile Communications (Wiley, New York, 1974)
85. M. Gudmundson, Correlation model for shadow fading in mobile radio systems. Electr. Lett. 27(23), 2145–2146 (1994)
86. J.A. Weitzen, T.J. Lowe, Measurement of angular and distance correlation properties of log-normal shadowing at 1900 MHz and its application
to design of PCS systems. IEEE Trans. Veh. Technol. 51(2), 265–273 (2002)
87. R.H. Clark, A statistical theory of mobile-radio reception. Bell Syst. Technol. J. 47(6), 957–1000 (1968)
88. M.D.Yacoub, The α  μ distribution: a general fading distribution, in Proceedings of IEEE international symposium on personal, indoor and
mobile radio communications, vol 2 pp. 629–633 (2002)
89. M.D. Yacoub, The αμ distribution: a physical fading model for the Stacy distribution. IEEE Trans. Veh. Technol. 56(1), 27–34 (2007)
90. I.B. Djordjevic, G.T. Djordjevic, On the communication over strong atmospheric turbulence channels by adaptive modulation and coding. Opt.
Express 17(20), 18250–18262 (2009)
91. G.T. Djordjevic, I.B. Djordjevic, G.K. Karagiannidis, Partially coherent EGC reception of uncoded and LDPC-coded signals over generalized
fading channels. Wirel. Pers. Commun. 66(1), 25–39 (2012)
92. S.O. Rice, Mathematical analysis of random noise. Bell Syst. Technol. J. 23(3), 282–333 (1944)
93. S.O. Rice, Statistical properties of a sine wave plus noise. Bell Syst. Technol. J. 27, 109–157 (1948)
94. G.L. Stüber, Principles of Mobile Communication, 3rd edn. (Springer, New York, 2012)
95. G.L. Turin, Introduction to spread spectrum antimultipath techniques and their application to urban radio. Proc. IEEE 68(3), 328–353 (1980)
96. H.S. Wang, N. Moayeri, Finite-state Markov channel-a useful model for radio communication channels. IEEE Trans. Veh. Technol. 44(1),
163–171 (1995)
97. P.A. Bello, Characterization of randomly time-variant linear channel. IEEE Trans. Comm. Syst. 11(4), 360–393 (1963)
98. M. Cvijetic, I.B. Djordjevic, Advanced Optical Communication Systems and Networks (Artech House, Boston/London, 2013)
99. R. Ramaswami, K. Sivarajan, Optical Networks: A Practical Perspective, 2nd edn. (Morgan Kaufman, San Francisco, 2002)
100. L. Kazovsky, S. Benedetto, A. Willner, Optical Fiber Communication Systems (Artech House, Boston, 1996)
101. J.C. Palais, Fiber Optic Communications, 5th edn. (Pearson Prentice Hall, Upper Saddle River, 2005)
102. G.P. Agrawal, Lightwave Technology: Telecommunication Systems (Wiley, Hoboken, NJ, 2005)
103. W. Shieh, I.B. Djordjevic, OFDM for Optical Communications (Elsevier/Academic Press, Amsterdam/Boston, 2009)
104. I. Kaminow, T. Li, A. Willner, Optical Fiber Telecommunications VIB (Elsevier/Academic Press, Oxford, 2013)
105. M. Staif, A. Mecozzi, J. Nagel, Mean square magnitude of all orders PMD and the relation with the bandwidth of principal states. IEEE Photon.
Tehcnol. Lett. 12, 53–55 (2000)
106. H. Kogelnik, L.E. Nelson, R.M. Jobson, Polarization mode dispersion, in Optical Fiber Telecommunications, ed. by I. P. Kaminow, T. Li,
(Academic Press, San Diego, 2002)
107. G.J. Foschini, C.D. Pole, Statistical theory of polarization mode dispersion in single-mode fibers. IEEE J. Lightw. Technol. 9(11), 1439–1456
(1991)
108. P. Ciprut, B. Gisin, N. Gisin, R. Passy, J.P. von der Weid, F. Prieto, C.W. Zimmer, Second-order PMD: impact on analog and digital
transmissions. IEEE J. Lightw. Technol. 16(5), 757–771 (1998)
156 2 Propagation Effects in Optical and Wireless Communications Channels, Noise Sources. . .

109. E. Ip, J.M. Kahn, Compensation of dispersion and nonlinear impairments using digital backpropagation. IEEE J. Lightw. Technol. 26(20),
3416–3425 (2008)
110. E. Ip, J.M. Kahn, Nonlinear Impairment Compensation using Backpropagation, in Optical Fibre, New Developments, ed. by C. Lethien,
(In-Tech, Vienna, 2009)
111. O.V. Sinkin, R. Holzlohner, J. Zweck, C.R. Menyuk, Optimization of the split-step Fourier method in modeling optical-fiber communications
systems. IEEE J. Lightw. Technol. 21(1), 61–68 (2003)
112. A. Rieznik, T. Tolisano, F.A. Callegari, D. Grosz, H. Fragnito, Uncertainty relation for the optimization of optical-fiber transmission systems
simulations. Opt. Express 13, 3822–3834 (2005)
113. Q. Zhang, M.I. Hayee, Symmetrized split-step Fourier scheme to control global simulation accuracy in fiber-optic communication systems.
IEEE J. Lightw. Technol. 26(2), 302–316 (2008)
114. C. Lin, I.B. Djordjevic, D. Zou, Achievable information rates calculation for optical OFDM transmission over few-mode fiber long-haul
transmission systems. Opt. Express 23(13), 16846–16856 (2015)
115. C. Lin, I.B. Djordjevic, M. Cvijetic, D. Zou, Mode-multiplexed multi-Tb/s superchannel transmission with advanced multidimensional
signaling in the presence of fiber nonlinearities. IEEE Trans. Comm. 62(7), 2507–2514 (2014)
116. K.-P. Ho, J.M. Kahn, Statistics of group delays in multimode fiber with strong mode coupling. IEEE J. Lightw. Technol. 29(21), 3119–3128
(2011)
117. S. Mumtaz, R.-J. Essiambre, G.P. Agrawal, Nonlinear propagation in multimode and multicore fibers: generalization of the Manakov
equations. J. Lightw. Technol. 31(3), 398–406 (2012)
118. K.-D. Langer, J. Grubor, Recent developments in optical wireless communications using infrared and visible Light, in Proceedings of the
ICTON 2007, pp. 146–152 (2007)
119. J. Grubor, V. Jungnickel, K. Langer, Rate-adaptive multiple-subcarrrier-based transmission for broadband infrared wireless communication, in
Optical fiber communication conference and exposition and the national fiber optic engineers conference, Technical Digest (CD) (Optical
Society of America, 2006), paper NThG2
120. S. Hranilovic, Wireless Optical Communication Systems (Springer, New York, 2004)
121. Y. Tanaka et al., Indoor visible communication utilizing plural white LEDs as lighting, in Proceedings of the Personal, Indoor & Mobile Radio
Communications (PIMRC) 2001, pp. F81–F85
122. T. Komine, M. Nakagawa, Fundamental analysis for visible-light communication system using LED lightings. IEEE Trans. Consum. Electron.
15(1), 100–107 (2004)
123. J. Grubor, O.C. Gaete Jamett, J.W. Walewski, S. Randel, K.-D. Langer, High-speed wireless indoor communication via visible light, in ITG
Fachbericht 198 (VDE-Verlag, Berlin und Offenbach, 2007), pp. 203–208
124. Z. Ghassemlooy, W. Popoola, S. Rajbhandari, Optical Wireless Communications: System and Channel Modelling with MATLAB (CRC Press,
Boca Raton, 2013)
125. J.M. Kahn, J.R. Barry, Wireless infrared communications. Proc. IEEE 85(2), 265–298 (1997)
 Components, Modules, and Subsystems
3

Abstract
In this chapter, basic components, modules, and subsystems relevant in both optical and wireless communications are
described. In the first half of the chapter, key optical components, modules, and subsystems are described. After describing
optical communication basic, focus is changed to optical transmitters. The following classes of lasers are introduced:
Fabry-Perot, distributed feedback (DFB), and distributed Bragg reflector (DBR) lasers. Regarding the external modulators,
both Mach-Zehnder modulator and electro-absorption modulator are described. Regarding the external modulator for
multilevel modulation schemes, phase modulator, I/Q modulator, and polar modulator are described. In subsection on
optical receivers, after the photodetection principle, the p-i-n photodiode, avalanche photodiode (APD), and metal-
semiconductor-metal (MSM) photodetectors are briefly described. In the same subsection, both optical receiver high-
impedance front end and trans-impedance front end are described. In section on optical fibers, both single-mode fiber
(SMF) and multimode fiber (MMF) are introduced. In subsection on optical amplifiers, after describing optical amplifica-
tion principles, the following classes of optical amplifiers are described: semiconductor optical amplifier (SOA), Raman
amplifier, and erbium-doped fiber amplifier (EDFA). In subsection on optical processing components and modules, the
following components/modules are described: optical couplers, optical filters, and WDM multiplexers/demultiplexers.
After that, the principles of coherent optical detection are introduced, followed by the description of optical hybrids. In
subsection on coherent optical balanced detectors, both homodyne and heterodyne optical balanced detectors for both
binary and multilevel (QAM, PSK, APSK) modulations are described. In the second half of the chapter, basic components
and modules of relevance to modern wireless communications are described. First, the DSP basics, including various
discrete-time realizations, are provided. Then the direct digital synthesizer (DDS), relevant in discrete-time (DT) wireless
communications systems, is described. In subsection on multirate DSP and resampling, various upsampling,
downsampling, and resampling schemes/modules are described. In the same subsection, the polyphase decomposition of
the system (transfer) function is introduced and employed in resampling. In subsection on antenna array antennas, the
linear array antennas are described. Finally, in subsection on automatic gain control (AGC), the AGC schemes operating in
passband and baseband are described. The set of problems is provided after the concluding remarks.

3.1 Key Optical Components, Modules, and Subsystems

This section describes the basic optical components used in an optical transmission system. We start this section with optical
communications basics [1–35]. Components and modules relevant in wireless communications will be covered in Sect. 3.2.

3.1.1 Optical Communications Basics

An exemplary optical network, identifying the key optical components, is shown in Fig. 3.1. The end-to-end optical
transmission involves both electrical and optical signal paths. To perform conversion from electrical to optical domain, the

# The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 157
I. B. Djordjevic, Advanced Optical and Wireless Communications Systems, https://doi.org/10.1007/978-3-030-98491-5_3
158 3 Components, Modules, and Subsystems

Fig. 3.1 An exemplary optical network identifying key optical components

optical transmitters are used, while to perform conversion in opposite direction (optical to electrical conversion), the optical
receivers are used. The optical fibers serve as foundation of an optical transmission system because the optical fiber is used as
medium to transport the optical signals from source to destination. The optical fibers attenuate the signal during transmission,
and someone has to use optical amplifiers, such as erbium-doped fiber amplifiers (EDFAs), Raman amplifiers, or parametric
amplifiers, to restore the signal quality [1–3, 5, 6]. However, the process of amplification is accompanied with the noise
addition. The simplest optical transmission system employs only one wavelength. The wavelength division multiplexing
(WDM) can be considered as an upgrade of the single-wavelength system. WDM corresponds to the scheme in which multiple
optical carriers at different wavelengths are modulated by using independent electrical bit streams, as shown in Fig. 3.1, and
then transmitted over the same fiber. WDM has potential of exploiting the enormous bandwidth offered by the optical fiber.
During transmission of WDM signals, occasionally one or several wavelengths are to be added or dropped, which is
performed by the optical component known as optical add-drop multiplexer (OADM), as illustrated in Fig. 3.1. The optical
networks require the switching of information among different fibers, which is performed in optical cross-connect (OXS). To
combine several distinct wavelength channels into composite channel, the wavelength multiplexers are used. On the other
hand, to split the composite WDM channel into distinct wavelength channels, the wavelength demultiplexers are used. To
impose the information signal, optical modulators are used. The optical modulators are commonly used in combination with
semiconductor lasers.
The typical receiver configuration with direct detection is shown in Fig. 3.2 [1–3]. The main purpose of the optical receiver,
terminating the lightwave path, is to convert the signal coming from single-mode fiber from optical to electrical domain and
process appropriately such obtained electrical signal to recover the data being transmitted. The incoming optical signal may be
pre-amplified by an optical amplifier and further processed by an optical filter to reduce the level of amplified spontaneous
emission (ASE) noise or by wavelength demultiplexer to select a desired wavelength channel. The optical signal is converted
into electrical domain by using a photodetector, followed by electrical post-amplifier. To deal with residual intersymbol
interference (ISI), an equalizer may be used. The main purpose of clock recovery circuit is to provide timing for decision
circuit by extracting the clock from the received signal. The clock recovery circuit is most commonly implemented using the
phase-locked loop (PLL). Finally, the purpose of decision circuit is to provide the binary sequence being transmitted by
comparing the sampled signal to a predetermined threshold. Whenever the received sample is larger than threshold, the
decision circuit decides in favor of bit 1, otherwise in favor of bit 0.
The optical signal generated by semiconductor laser has to be modulated by information signal before being transmitted
over the optical fiber. This can be achieved by directly modulating the bias current of semiconductor laser, which can be done
even at high speed (even up to 40 Gb/s in certain lasers). Unfortunately, this concept although conceptually simple is rarely
used in practice because of the frequency chirp introduced by direct modulation, nonuniform frequency response, and large
current swing needed to provide operation. For transmitters operating at 10 Gb/s and above, instead, the semiconductor laser
3.1 Key Optical Components, Modules, and Subsystems 159

Fig. 3.2 A typical direct detection receiver architecture. O/E, optical to electrical; AGC, automatic gain control

Information-bearing
(modulating) signal

CW
signal Modulated
CW LD External signal
(e.g., DFB) modulator

Fig. 3.3 An illustration of external modulation principle

diode (LD) is commonly biased at constant current to provide continuous wave (CW) output, and external modulators are
used to impose the information signal to be transmitted. The most popular modulators are electro-optic optical modulators,
such as Mach-Zehnder modulators, and electro-absorption modulators. The principle of the external modulator is illustrated in
Fig. 3.3. Through the external modulation process, a certain parameter of the CW signal, used as a signal carrier, is varied in
accordance with the information-bearing signal. For example, a monochromatic electromagnetic wave is commonly used as a
carrier, and its electrical field E(t) can be represented by:

Eðt Þ ¼ pA cos ðωt þ φÞ, ð3:1Þ

where A, ω, and φ are amplitude, frequency, and phase, respectively, while p denotes the polarization orientation. Each of
those parameters can be used to carry information, and the information-bearing signal can be either CW or discrete. If the
information-bearing signal is CW, corresponding modulation formats are amplitude modulation (AM), frequency modulation
(FM), phase modulation (PM), and polarization modulation (PolM). On the other hand, if the information-bearing signal is
digital, the corresponding modulations are amplitude-shift keying (ASK), frequency-shift keying (FSK), phase-shift keying
(PSK), and polarization-shift keying (PolSK).
Optical fibers serve as foundation of an optical transmission system because they transport optical signals from source to
destination. The combination of low-loss and large bandwidth allows high-speed signals to be transmitted over long distances
before the regeneration is needed. A low-loss optical fiber is manufactured from several different materials; the base row
material is pure silica, which is mixed with different dopants in order to adjust the refractive index of optical fiber. The optical
fiber consists of two waveguide layers, the core and the cladding, protected by buffer coating. The majority of the power is
concentrated in the core, although some portion can spread to the cladding. There exists a difference in refractive indices
between the core and cladding, which is achieved by mixing dopants, commonly added to the fiber core. There exist two types
of optical fibers: multimode fiber (MMF) and single-mode fiber (SMF). Multimode optical fibers transfer the light through a
collection of spatial transversal modes. Each mode, defined through a specified combination of electrical and magnetic
components, occupies a different cross section of the optical fiber core and takes a slightly distinguished path along the optical
160 3 Components, Modules, and Subsystems

Fig. 3.4 A typical optical

networking architecture Access network

EN
CN EN
EN CN
EN
EN EN
CN CN
EN CN: Core node
Core network EN EN: Edge node

Edge network End node

fiber. The difference in mode path lengths in multimode optical fibers produces a difference in arrival times at the receiving
point. This phenomenon is known as multimode dispersion (or intermodal dispersion) and causes signal distortion and
imposes the limitations in signal bandwidth. The second type of optical fibers, SMFs, effectively eliminates multimode
dispersion by limiting the number of propagating modes to a fundamental one. SMFs, however, introduce another signal
impairment known as the chromatic dispersion. Chromatic dispersion is caused by the difference in velocities among different
spectral components within the same mode.
The attenuation of signal propagating through optical fiber is low compared to other transmission media, such as copper
cables or free space. Nevertheless, we have to amplify the attenuated signal from time to time, to restore the signal level,
without any conversion into electrical domain. This can be done in optical amplifiers, through the process of stimulated
emission. The main ingredient of an optical amplifier is the optical gain realized through the amplifier pumping (being either
electrical or optical) to achieve the so-called population inversion. The common types of optical amplifiers are semiconductor
optical amplifiers (SOAs), EDFAs, and Raman amplifiers. The amplification process is commonly followed by the noise
process, not related to the signal, which occurs due to spontaneous emission. The amplification process degrades the SNR,
because of amplified spontaneous emission (ASE), added to the signal in every amplifier stage.
Before providing more details about basic building blocks identified in this section, let us give more global picture, by
describing a typical optical network shown in Fig. 3.4. We can identify three ellipses representing the core network, the edge
network, and the access network [1–3]. The long-haul core network interconnects big cities, major communications hubs, and
even different continents by means of submarine transmission systems. The core networks are often called the wide area
networks (WANs) or interchange carrier networks. The edge optical networks are deployed within smaller geographical areas
and are commonly recognized as metropolitan area networks (MANs) or local exchange carrier networks. The access
networks represent peripheral part of optical network and provide the last-mile access or the bandwidth distribution to the
individual end users. The common access networks are local area networks (LANs) and distribution networks. The common
physical network topologies are mesh network (often present in core networks), ring network (in edge networks), and star
networks (commonly used in access networks).
Given this general description of key optical components in the rest of this section, we provide more details about basic
building blocks: optical transmitters are described in Subsect. 3.1.2, optical receivers in Subsect. 3.1.3, optical fibers in
Subsect. 3.1.4, and optical amplifiers in Subsect. 3.1.5, and other optical processing blocks are described, such as WDM
multiplexers/demultiplexers, optical filters, and couplers, in Subsect. 3.1.6. The Subsect. 3.1.7 is devoted to the principles of
coherent optical detection, followed by subsection on optical hybrids (3.1.8). In Subsect. 3.1.9, the balanced coherent optical
detectors are described.

3.1.2 Optical Transmitters

The role of the optical transmitter is to generate the optical signal, impose the information-bearing signal, and launch the
modulated signal into the optical fiber. This section is devoted to different lasers including Fabry-Perot, distributed feedback
(DFB), and distributed Bragg reflector (DBR) lasers as well as to different external modulators including Mach-Zehnder
modulator, electro-absorption modulator, phase modulator, I/Q modulator, and polar modulator.
3.1 Key Optical Components, Modules, and Subsystems 161

Fig. 3.5 Illustrating the interaction of the light with the matter

3.1.2.1 Lasers
The semiconductor light sources are commonly used in state-of-the-art optical communication systems. The light generation
process occurs in certain semiconductor materials due to recombination of electrons and holes in p-n junctions, under direct
biasing. Depending on the nature of recombination process, we can classify different semiconductor light sources as either
light-emitting diodes (LEDs) in which spontaneous recombination dominates or semiconductor lasers in which the stimulated
emission is a dominating mechanism. Namely, there are three basic processes in semiconductor materials, as illustrated in
Fig. 3.5, by which the light interacts with matter: absorption, spontaneous emission, and stimulated emission. In normal
conditions the number of electrons in ground state (with energy E1) N1 is larger than the number of electrons in excited state
(with energy E2) N2, and in the thermal equilibrium, their ratio follows the Boltzmann’s statistics [1–7]:



N2 hν E  E1
¼ exp  ¼ exp  2 , ð3:2Þ
N1 kB T kB T

where hν is a photon energy (h is the Plank’s constant, ν is the optical frequency proportional to the energy difference between
the energy levels E2–E1), kB is the Boltzmann’s constant, and T is the absolute temperature. In the same regime, the
spontaneous emission rate dN2, spon/dt ¼ A21N2 (A21 denotes the spontaneous emission coefficient) and the stimulated
emission rate dN2, stim/dt ¼ B21ρ (ν) N2 (B21 denotes the stimulated emission coefficient and ρ (ν) denotes the spectral density
of electromagnetic energy) are equalized with absorption rate dN1, abs/dt ¼ A12ρ (ν) N1 (A12 denotes the absorption
coefficient):

A21 N 2 þ B21 ρðνÞN 2 ¼ B12 ρðνÞN 1 : ð3:3Þ

In the visible or near-infrared region (hν ~ 1 eV), the spontaneous emission always dominates over stimulated emission in
thermal equilibrium at room temperature (kBT  25 meV) (see Eq. (3.2)). The stimulated emission rate can exceed absorption
rate only when N2 > N1, the condition is referred to as population inversion and can never be realized for systems being in
thermal equilibrium. The population inversion is a prerequisite for laser operation, and in atomic system, it is achieved by
using three- and four-level pumping schemes (an external energy source raises the atomic population from ground to an
excited state). There are three basic components required to sustain stimulated emission and to form useful laser output: the
pump source, the active medium, and the feedback mirrors. The active medium can be solid (such as in semiconductor lasers),
gaseous, or liquid in nature. The pump can be electrical (e.g., semiconductor lasers), optical, or chemical. The purpose of the
pump is to achieve the population inversion. The basic structure of semiconductor laser of Fabry-Perot type is shown in
Fig. 3.6(a), together with equivalent model. The injection (bias) current flows through the p-n junction and stimulates the
recombination of electron and holes, leading to the generation of photons.
For the lasing action to be sustainable, the gain and phase matching condition should be satisfied. In the active medium,
both gain/absorption described by γ(ν) and scattering described by αs are present. The intensity inside the cavity can be
described by the following dependence I(z) ¼ I0exp[(γ(ν)  αs)z]. The lasing is possible when collective gain is larger than
loss after a round trip pass through the cavity:

I ð2LÞ ¼ I 0 R1 R2 exp ½2Lðγ ðνÞ  αs Þ ¼ I 0 , ð3:4Þ

where R1 and R2 are facet reflectivities (see Fig. 3.6), L is the length of active medium, and I0 and I(2 L ) correspond to initial
and round-trip intensities. The gain threshold is obtained by solving the Eq. (3.4) per γ:
162 3 Components, Modules, and Subsystems

Fig. 3.6 Semiconductor lasers: (a) Fabry-Perot semiconductor laser, (b) DFB laser, and (c) DBR laser


1 1
γ th ¼ αs þ ln ¼ αint þ αmir , ð3:5Þ
2L R1 R2

where the internal losses (corresponding to αs) and mirror losses ((1/2 L )ln(1/R1R2)) are denoted by αint and αmir, respectively.
After the round trip, the resultant phase must be equal to the initial phase, leading to the phase matching condition:

exp ½j2βL ¼ 1 ) 2βL ¼ q2π, β ¼ 2πn=λ ð3:6Þ

where β denotes the propagation constant, n is the refractive index of active medium, and λ is the free-space wavelength.
The phase matching condition can be satisfied for many different integers q, representing different longitudinal modes of
frequency νq ¼ q c/(2 nL). The separation between neighboring longitudinal modes is known as the free spectral range:

c c c
Δv ¼ vq  vq1 ¼ q  ðq  1Þ ¼ : ð3:7Þ
2nL 2nL 2nL

Because of the presence of longitudinal modes, the Fabry-Perot laser belongs to the class of multimode lasers. To improve
the coherence of output light and the laser modulation, speed distributed feedback (DFB) lasers, shown in Fig. 3.6(b), are
used. The key idea of this laser is to effectively select one of the longitudinal modes while suppressing the remaining ones.
3.1 Key Optical Components, Modules, and Subsystems 163

This is achieved by introducing the Bragg grating inside of the laser cavity. The wavelength of selected longitudinal mode can
be determined from the Bragg condition:

λ 2n Λ
2Λ ¼ m  ) λB ¼ av , ð3:8Þ
nav m

where Λ is the grating period, nav is the average refractive index of a waveguide mode, and λ/nav is the average wavelength of
the light in the waveguide mode. If the grating element is put outside of active region or instead of the facet mirrors, the
distributed Bragg reflector (DBR) laser is obtained, which is illustrated in Fig. 3.6(c). Both DFB and DBR lasers belong to the
class of single-mode lasers. Different semiconductor lasers shown in Fig. 3.6 are edge-emitting lasers.
Another important semiconductor laser type is vertical-cavity surface-emitting laser (VCSEL), which emits the light
vertical to the active layer plane [1–7]. The VCSELs are usually based on In-GaAs-P layers acting as Bragg reflectors and
providing the positive feedback leading to the stimulated emission.
The spectral curve of the single-mode lasers is a result of transition between discrete energy levels and can often be
represented using the Lorentzian shape [1–7]:

Δν
gð ν Þ ¼ h Δν2 i , ð3:9Þ
2π ðv  v0 Þ2 þ 2

where ν0 is the central optical frequency and Δν represents the laser linewidth [1–7]:
 
nsp G 1 þ α2chirp
Δν ¼ , ð3:10Þ
4πP

where nsp is the spontaneous emission factor, G is the net rate of stimulated emission, P denotes the output power, and αchirp is
the chirp factor (representing the amplitude-phase coupling parameter).
The small-signal frequency response of the semiconductor laser is determined by [1–7]:

Ω2R þ Γ2R
H ð ωÞ ¼ , ð3:11Þ
ðΩR þ ω  jΓR ÞðΩR  ω þ jΓR Þ

where ΓR is the damping factor and ΩR is the relaxation frequency Ω2R  GN Pb =τp ðΓR << ΩR Þ, with GN being the net rate of
stimulated emission, Pb being the output power corresponding to the bias current, and τp being the photon life time related to
the excited energy p
level. The modulation bandwidth (defined as 3 dB bandwidth) is therefore determined by the relaxation
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffi
frequency ω3dB ¼ 1 þ 2ΩR , and for fast semiconductor lasers can be 30 GHz. Unfortunately, the direct modulation of
semiconductor lasers leads to frequency chirp, which can be described by the instantaneous frequency shift from steady-state
frequency ν0 as follows [1–7]:

αchirp d d
δνðt Þ ¼ ln Pðt Þ þ χPðt Þ ¼ C dyn Pðt Þ þ C ad Pðt Þ, ð3:12Þ
4π dt dt

where P(t) is the time variation of the output power, χ is the constant (varying from zero to several tens) related to the material
and design parameters, and αchirp is the chirp factor defined as the ratio between the refractive index n change and gain
G change with respect to the number of carriers N: αchirp ¼ (dn/dN )/(dG/dN ). The first term on the right-hand side of (3.12)
represents dynamic (transient or instantaneous) chirp and the second term the adiabatic (steady-state) frequency chirp. The
random fluctuation in carrier density due to spontaneous emission also leads to the linewidth enhancement proportional to
(1 + α2chirp). To avoid the chirp problem, the external modulation is used, while the semiconductor lasers are biased by a dc
voltage to produce a continuous wave operation.
164 3 Components, Modules, and Subsystems

3.1.2.2 External Modulators

There are two types of external modulators commonly used in practice: Mach-Zehnder modulator (MZM) and electro-
absorption modulator (EAM), whose operational principle is illustrated in Fig. 3.7. The MZM is based on electro-optic effect,
the effect that in certain materials (such as LiNbO3) where the refractive index n changes with respect to the voltage V applied
across electrodes [1–7]:

1 2π
Δn ¼  Γn3 r 33 ðV=de Þ ) Δϕ ¼ ΔnL, ð3:13Þ
2 λ

where Δn denotes the refractive index change, Δϕ is the corresponding phase change, r33 is the electro-optic coefficient
(~30.9 pm/V in LiNbO3), de is the separation of electrodes, L is the electrode length, and λ is the wavelength of the light. The
MZM (see Fig. 3.7(a)) is a planar waveguide structure deposited on the substrate, with two pairs of electrodes: (i) for high-
speed ac voltage representing the modulation data (RF) signal and (ii) for dc bias voltage. Let V1(t) and V2(t) denote the
electrical drive signals on the upper and lower electrodes, respectively. The output electrical field Eout(t) of the second
Y-branch can be related to the input electrical filed Ein by:




1 π π
Eout ðt Þ ¼ exp j V 1 ðt Þ þ exp j V 2 ðt Þ Ein , ð3:14Þ
2 Vπ Vπ

where Vπ is now the differential drive voltage required to introduce the phase shift of π radians between two waveguide arms.
With V1(t) and V2(t), we denoted the electrical drive signals applied to upper and lower electrodes, respectively. We can
implement different modulation formats by using this MZM as a building block, including on-off keying (OOK), binary
phase-shift keying (BPSK), QPSK, differential PSK (DPSK), differential QPSK (DQPSK), and return-to-zero (RZ) OOK. For

Fig. 3.7 External optical Bias

modulators: (a) Mach-Zehnder Modulation voltage
modulator and (b) electro- data (RF) (DC)
absorption modulator

V1(t)
Ein(t) Eout(t)

Modulated
output

V2(t)
(a)
Injection current -V
bias and modulation
to SMF
Bragg grating
p-type p-type

Active region
n-type n-type

+V
bias and modulation
CW laser EA modulator
(b)
3.1 Key Optical Components, Modules, and Subsystems 165

instance, zero-chirp OOK, or BPSK, can be generated by setting V1(t) ¼ V(t)  Vπ/2, V2(t) ¼  V(t) + Vπ /2, so that
Eq. (3.14) becomes:


Eout ðt Þ πV ðt Þ
¼ sin : ð3:15Þ
E in Vπ

On the other hand, if we set V1(t) ¼ V(t)/2, V2(t) ¼  V(t)/2, the Eq. (3.14) becomes:


E out ðt Þ πV ðt Þ
¼ cos , ð3:16Þ
Ein 2V π

and we say that MZM is operated in the push-pull mode. The corresponding power transfer function of MZM is obtained as
magnitude squared of electric field ratio:
 

Eout ðt Þ2
  ¼ cos 2 πV ðt Þ ¼ 1 þ cos ðπV ðt Þ=V π Þ : ð3:17Þ
 Ein  2V π 2

The EAM is a semiconductor-based planar waveguide composed of multiple p-type and n-type layers that form multiple
quantum wells (MQWs). The basic design of EAM is similar to that of semiconductor lasers. The MQW is used to support the
quantum-confined Stark effect (the absorption spectrum being a function of applied field) more effectively. Because of
similarities of EAMs and semiconductor lasers design, it is possible to fabricate them on the same substrate (see Fig. 3.7(b)),
providing that EAM and laser are electrically isolated. Bandgap of quantum wells is larger than photon energy, so that the light
is completely transmitted in the absence of bias, which corresponds to the ON state. When the reverse bias is applied, the input
signal is absorbed, which corresponds to the OFF state. The modulation speed of EAMs is typically comparable to the
modulation speed of MZMs. However, the extinction ratio (the ratio of average powers corresponding to symbol 1 and symbol
0) is lower.
It is evident from Eqs. (3.15) to (3.16) that MZM can be used for both intensity modulation and phase modulation. For
intensity modulation, the MZM is operated at so-called quadrature point (QP), with a DC bias of Vπ/2 and peak-to-peak
modulation voltage lower than Vπ. On the other hand, for BPSK, the MZM is operated at null point (NP), with a DC bias of
Vπ and peak-to-peak modulation voltage lower than 2Vπ. The operating points of MZM for amplitude and phase modulation
are shown in Fig. 3.8. In addition, MZM can be used for conversion of non-return-to-zero (NRZ) digital forms to return-to-
zero (RZ) form. (Otherwise, the NRZ-to-RZ conversion can be performed either in electrical or optical domain.) In optical
domain, we typically use a MZM driven by a properly chosen sinusoidal signal with frequency equal to symbol rate Rs:
 
Vπ π V
V ðt Þ ¼ sin 2πRs t   π, ð3:18Þ
2 2 2

so that RZ pulses of duty cycle 50% are being generated at the output of MZM. Accordingly, we have that:

E out ðt Þ    
π π π
¼ cos sin 2πRs t   : ð3:19Þ
Ein ðt Þ 4 2 4

Additionally, the pulse shape can further be modified by driver amplifiers.

3.1.2.3 I/Q Modulators and Polar Modulators

Another basic building block used to implement more advanced modulators is phase modulator, whose physical properties
have been discussed above. The implementation of phase modulator is illustrated in Fig. 3.9.
The electrical fields at the output and input of phase modulator are related as follows:


Eout ðt Þ πV ðt Þ
¼ exp j ð3:20Þ
E in Vπ
166 3 Components, Modules, and Subsystems

Fig. 3.8 The operating points of

Mach-Zehnder modulator

Fig. 3.9 The operation principle of phase modulator

By using two MZM modules and one PM, which introduces the π/2 phase shift, we can implement the I/Q modulator in a
way shown in Fig. 3.10. It can be shown that output Eout and input Ein electrical fields of I/Q modulator are related as:

Eout ðt Þ 1 1 πV I ðt Þ 1 1 πV Q ðt Þ
¼ cos þ j cos
E in 2 2 Vπ 2 2 Vπ ð3:21Þ
1 1
¼ cos ½ΦI ðt Þ=2 þ j cos ½ΦQ ðt Þ=2, ΦI ðQÞ ðt Þ ¼ πV I ðQÞ ðt Þ=V π :
2 2

where VI and VQ are in-phase and quadrature RF modulating signals. When the amplitudes of VI and VQ are sufficiently small
and in-phase/quadrature signals are chosen as follows V I ðt Þ ¼ 4 π V π I ðt Þ þ V π , V Q ðt Þ ¼ π V π Qðt Þ þ V π ,
4
the MZM
nonlinearities can be neglected and Eq. (3.21) can be rewritten as:
3.1 Key Optical Components, Modules, and Subsystems 167

Fig. 3.10 The operation principle of I/Q modulator

Fig. 3.11 The operation principle of polar modulator

E out ðt Þ
ffi I ðt Þ þ jQðt Þ, ð3:22Þ
E in

where I(t) and Q(t) are in-phase and quadrature components of RF signal. Therefore, different QAM constellations can be
directly applied on this scheme. Equation (3.21) can also be represented in polar coordinates in a form:

E out ðt Þ
¼ Rðt Þ cos ½Ψðt Þ, ð3:23Þ
E in

where the envelope R(t) and phase Ψ (t) are expressed as:
 


1=2
E ðt Þ 1 1 πV I ðt Þ 1 πV Q ðt Þ
Rðt Þ ¼  out  ¼ cos 2 þ cos 2 ,
Ein 2 2 Vπ 2 Vπ
8
9
>
> 2 1 πV Q ðt Þ >> ð3:24Þ
< cos
2 Vπ =
Ψðt Þ ¼ tan 1
 :
>
> πV ðt Þ > >
: cos 2 1 I ;
2 Vπ

The representation given by Eq. (3.23) indicates that instead of I/Q modulator, we can use an equivalent polar modulator
shown in Fig. 3.11. The real part of the output of polar modulator corresponds to Eq. (3.21).

3.1.3 Optical Receivers

The purpose of the optical receiver is to convert the optical signal into electrical domain and to recover the transmitted data.
The typical OOK receiver configuration is already given in Fig. 3.2. We can identify three different stages: front end stage, the
linear channel stage, and data recovery stage. The front end stage is composed of a photodetector and a pre-amplifier. The
most commonly used front end stages are high-impedance front end and trans-impedance front end, both shown in Fig. 3.8.
168 3 Components, Modules, and Subsystems

Fig. 3.12 Optical receiver front end stage schemes: (a) high-impedance front end and (b) trans-impedance front end

High-impedance front end (Fig. 3.12(a)) employs a large value load resistance to reduce the level of thermal noise and has a
good receiver sensitivity. However, the bandwidth of this scheme is low because the RC constant is large. To achieve both the
high receiver sensitivity and large bandwidth, the trans-impedance front end scheme, shown in Fig. 3.12(b), is used. Even
though the load resistance is high, the negative feedback reduces the effective input resistance by a factor of G  1, where G is
the front end amplifier gain. The bandwidth is increased for the same factor compared to high-impedance front end scheme.
The photodiode is an integral part of both front end stage schemes. The key role of the photodiode is to absorb photons in
incoming optical signal and convert them back to the electrical level through the process opposite to one taking place in
semiconductor lasers. The common photodiodes are p-n photodiode, p-i-n photodiode, avalanche photodiode (APD), and
metal-semiconductor-metal (MSM) photodetectors [1–7]. The p-n photodiode is based on a reverse-biased p-n junction. The
thickness of the depletion region is often less than the absorption depth for incident light, and the photons are absorbed outside
of the depletion region, leading to the slow response speed. p-i-n photodiode consists of an intrinsic region sandwiched
between p- and n-type layers, as shown in Fig. 3.13(a). Under the reverse bias, the depletion depth can be made sufficiently
thick to absorb most of the incident photons. Avalanche photodiode, shown in Fig. 3.13(b), is a modified p-i-n photodiode that
is operated at very high reverse bias. Under high-field conditions, photo-generated carriers induce generation of secondary
electron-hole pairs by the process of impact ionization, and this process leads to internal electrical gain. Metal-semiconduc-
tor-metal photodetectors employ interdigitated Schottky barrier contacts on one face of the device and are compatible with
planar processing and optoelectronic integration. Depending on the device design, the device is illuminated through the p- or
n-type contact. In Si, Ge, or GaAs diodes, the substrate is absorbing so that the device has to be illuminated through the top
contact, as shown in Fig. 3.13(a).
On the other hand, in InGaAs or InGaAsP, the substrate is transparent, and the device can be designed to be illuminated
either through the substrate or top contact. In order to increase the depletion region and to minimize the diffusion current
component, an intrinsic layer (i-type) is introduced to the p-i-n photodiode structure. The p-i-n photodiode is reverse biased
and has very high internal impedance, meaning that it acts as a current source generating the photocurrent proportional to the
incoming optical signal power. The equivalent scheme of p-i-n photodiode is shown in Fig. 3.13(c). Typically the internal
series resistance Rs is low, while the internal shunt resistance is high, so that the junction capacitance Cp dominates and can be
determined by:
3.1 Key Optical Components, Modules, and Subsystems 169

Fig. 3.13 The semiconductor photodiodes: (a) p-i-n photodiode and (b) avalanche photodiode. (c) The equivalent p-i-n photodiode model

1=2
A εs N A N D
C p ¼ εs ¼ , ð3:25Þ
w 2ð N A  N D Þ ð V 0  V A Þ

where εs is the semiconductor permittivity, A is the area of the space charge region (SCR), w is the width of SCR, NA and ND
denote dopant (acceptor and donor) densities, V0 is the built-in potential across the junction, and VA is the applied negative
voltage. The photocurrent iph(t) is proportional to the power of incident light P(t), that is, iph(t) ¼ R P(t), where R [A/W] is the
photodiode responsivity. The photodiode responsivity is related to the quantum efficiency η, defined as the ratio of number of
generated electrons and the number of incident photons, by R ¼ ηq/hν, where q is an electron charge and hv is a photon
energy. Using this model, we can determine the 3 dB bandwidth of high-impedance front end scheme as B3 dB ¼ 1/(2πRLCp)
170 3 Components, Modules, and Subsystems

and the 3 dB bandwidth of trans-impedance front end scheme as B3 dB ¼ (G + 1)/(2πRFCp), which is G times higher than
bandwidth of high-impedance front end scheme.

3.1.4 Optical Fibers

Optical fibers serve as foundation of an optical transmission system because they transport optical signals from source to
destination [1–7]. The combination of low-loss and extremely large bandwidth allows high-speed signals to be transmitted
over long distances before the regeneration becomes necessary. A low-loss optical fiber is manufactured from several different
materials; the base row material is pure silica, which is mixed with different dopants in order to adjust the refractive index of
optical fiber. The optical fiber, shown in Fig. 3.14, consists of two waveguide layers, the core (of refractive index n1) and the
cladding (of refractive index n2), protected by the jacket (the buffer coating). The majority of the power is concentrated in the
core, although some portion can spread to the cladding. There is a difference in refractive indices between the core and
cladding (n1 > n2), which is achieved by mix of dopants commonly added to the fiber core. The refractive-index profile for
step-index fiber is shown in Fig. 3.14(c), while the illustration of light confinement by the total internal reflection is shown in
Fig. 3.14(d). The raypwill be ffitotally reflected from the core-cladding interface (a guided ray) if the following condition is
ffiffiffiffiffiffiffiffiffiffiffiffiffiffi
satisfied n0 sin θi < n21  n22 , where θi is the angle of incidence. Max(n0p sinθ i) defines
ffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi the light-gathering capacity of an
pffiffiffiffiffiffi
optical fiber, and it is called the numerical aperture (NA), defined by NA ¼ n21  n22  n1 2Δ, ðΔ  1Þ, where Δ is the
normalized index difference defined as Δ ¼ (n1  n2)/n1. Therefore, from the geometrical optics point of view, light
propagates in optical fiber due to series of total internal reflections that occur at the core-cladding interface. The smallest
angle of incidence ϕ (see Fig. 3.10(d)) for which the total internal reflection occurs is called the critical angle and equals sin1
n2/n1.
There exist two types of optical fibers: MMF (shown in Fig. 3.10(a)) and SMF (shown in Fig. 3.10(b)). Multimode optical
fibers transfer the light through a collection of spatial transversal modes. Each mode, defined through a specified combination
of electrical and magnetic components, occupies a different cross section of the optical fiber core and takes a slightly
distinguished path along the optical fiber. The difference in mode path lengths in multimode optical fibers produces a

Fig. 3.14 Optical fibers: (a) multimode optical fiber, (b) single-mode optical fiber, (c) refractive-index profile for step-index fiber, and (d) the light
confinement in step-index fibers through the total internal reflection
3.1 Key Optical Components, Modules, and Subsystems 171

difference in arrival times at the receiving point. This phenomenon is known as multimode dispersion (or intermodal
dispersion) and causes the signal distortion and imposes the limitations in signal bandwidth. The second type of optical
fibers, single-mode fibers, effectively eliminates multimode dispersion by limiting the number of propagating modes to a
fundamental one. The fundamental mode occupies the central portion of the optical fiber and has an energy maximum at the
axis of the optical fiber core. Its radial distribution can be approximated by Gaussian curve. The number of modes (M) that can
effectively propagate through an optical fiber is determined by the normalized frequency (V parameter or V number): M  V2/
2, when V is large. The normalized frequency is defined by:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2πa
V¼ n21  n22 , ð3:26Þ
λ

where a is the fiber core radius, λ is the carrier wavelength, and n1 and n2 are refractive indices related to the fiber core and the
fiber cladding, respectively.
Each mode propagating through the fiber is characterized by its own propagation constant β. The dependence of the electric
and magnetic fields on axial coordinate z is expressed through the factor exp.(jβz). The propagation constant must satisfy the
following condition:

2πn2 =λ < β < 2πn1 =λ: ð3:27Þ

In order to evaluate the transmission characteristics of the optical fiber, the functional dependence of the mode propagation
constant on the optical signal wavelength has to be known. The normalized propagation constant b is defined for that purpose:

β2  ð2πn2 =λÞ2
b¼ : ð3:28Þ
ð2πn1 =λÞ2  ð2πn2 =λÞ2

The normalized propagation constant is related to the normalized frequency V by [1–11]:

bðV Þ  ð1:1428  0:9960=V Þ2 , 1:5 V 2:5 ð3:29Þ

The multimode dispersion can effectively be eliminated by limiting the number of propagating modes to a fundamental one:
V Vc ¼ 2.405 with Vc being the cutoff frequency. The cutoff frequency is controlled by keeping the core radius small and the
normalized index difference Δ ¼ (n1  n2)/n1 between 0.2% and 0.3%.

3.1.5 Optical Amplifiers

The purpose of an optical amplifier is to restore the signal power level, reduced due to losses during propagation, without any
optical to electrical conversion. The general form of an optical amplifier is given in Fig. 3.15(a). Most optical amplifiers
amplify incident light through the stimulated emission, the same mechanism that is used in lasers but without the feedback

Pump power
Pump
Weak Amplified Fiber
Fiber signal signal Pout
Pin
Optical amplifier medium N2
Gain
N1 medium

Optical signal input Optical signal output z=0 z=L

(a) (b)
Fig. 3.15 (a) Optical amplifier principle, (b) two-level amplifier system model
172 3 Components, Modules, and Subsystems

mechanism. The main ingredient is the optical gain realized through the amplifier pumping (electrical or optical) to achieve
the population inversion. The optical gain, generally speaking, is not only a function of frequency but also a function of local
beam intensity. To illustrate the basic concepts, we consider the case in which the gain medium is modeled as two-level
system, as shown in Fig. 3.15(b). The amplification factor G is defined as the ratio of amplifier output Pout and input Pin
powers G ¼ Pout/Pin. The amplification factor can be determined by knowing the dependence of evolution of power through
the gain media [1–7]:

dP g0
¼ gP, g ð ωÞ ¼ , ð3:30Þ
dz 1 þ ðω  ω0 Þ2 T 22 þ P=Ps

where g is the gain coefficient, g0 is the gain peak value, ω0 is the atomic transition frequency, T2 is the dipole relaxation time
(<1 ps), ω is the optical frequency of incident signal, P is the incident signal power, and PS is the saturation power.
In the unsaturated regime (P < <PS), the differential Eq. (3.30) can be solved by separation of variables to get the following
dependence of power P(z) ¼ P(0)exp.(gz), so that the amplification factor can be obtained by:

GðωÞ ¼ exp ½gðωÞL, ð3:31Þ

and corresponding full width half-maximum (FWHM) bandwidth is determined by:

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ln 2
ΔνA ¼ Δνg , G0 ¼ exp ðg0 LÞ ð3:32Þ
ln ðG0 =2Þ

where Δνg is the FWHM gain coefficient bandwidth.

The gain saturation comes from the power dependence of the gain coefficient (3.30). The coefficient is reduced when the
incident power P becomes comparable to the saturation power PS. Let’s assume that the incident frequency is tuned to the
peak gain (ω ¼ ω0), and then from Eq. (3.30), we obtain:

dP g0 P
¼ : ð3:33Þ
dz 1 þ P=PS

By solving the differential Eq. (3.33) with respect to the boundary conditions, P(0) ¼ Pin and P(L ) ¼ Pout ¼ GPin, we get:

G  1 Pout
G ¼ G0 exp  : ð3:34Þ
G PS

From Eq. (3.34), we can determine another important optical amplifier parameter, the output saturation power as being the
optical power at which the gain G is reduced to G0/2 (3 dB down):

G0 ln 2
Psat
out ¼ P  ð ln 2ÞPs  0:69Ps ðG0 > 20dBÞ: ð3:35Þ
G0  2 s

Three common applications of optical amplifiers are (i) power boosters (of transmitters), (ii) in-line amplifiers, and (iii)
optical pre-amplifiers, which is illustrated in Fig. 3.16.
The booster (power) amplifiers are placed at the optical transmitter side to enhance the transmitted power level or to
compensate for the losses of optical elements between the laser and optical fibers, such as optical coupler, WDM multiplexers,
and external optical modulators. The in-line amplifiers are placed along the transmission link to compensate for the losses
incurred during propagation of optical signal. The optical pre-amplifiers are used to increase the signal level before the
photodetection takes place, improving therefore the receiver sensitivity.
Several types of optical amplifiers have been introduced so far: semiconductor optical amplifiers (SOAs), fiber Raman (and
Brillouin) amplifiers, rare-earth-doped fiber amplifiers (erbium-doped EDFA operating at 1500 nm, praseodymium-doped
PDFA operating at 1300 nm), and parametric amplifiers.
3.1 Key Optical Components, Modules, and Subsystems 173

Fig. 3.16 Possible application of Fiber Fiber Fiber

optical amplifiers: (a) booster
amplifier, (b) in-line amplifiers, Tx … Rx
Tx Rx
and (c) pre-amplifier

Booster (power) amplifier In- line amplifiers

(a) (b)
Fiber

Tx Rx

Pre-amplifier

(c)

Electrical drive
current
R1 R2
Amplified
Weak Optical
Signal Input Optical Signal
Semiconductor cavity

Antireflection
coatings

Fig. 3.17 The semiconductor optical amplifier (SOA) operation principle

3.1.5.1 Semiconductor Optical Amplifier (SOA)

Semiconductor lasers act as amplifiers before reaching the threshold. To prevent the lasing, antireflection (AR) coatings are
used in SOAs, as shown in Fig. 3.17. Even with AR coating, the multiple reflections must be included when considering the
Fabry-Perot (FP) cavity. The gain of FP amplifiers is given by [1–7]:

ð1  R1 Þð1  R2 ÞGðνÞ
GFP ðνÞ ¼  pffiffiffiffiffiffiffiffiffiffi2 pffiffiffiffiffiffiffiffiffiffi , ð3:36Þ
1  GðνÞ R1 R2 þ 4G R1 R2 sin 2 ½π ðν  νm Þ=ΔνL 

where R1 and R2 denote the facet reflectivities, G(ν) is the single-pass amplification factor, νm is the cavity resonance
frequency, and ΔνL is the free spectral range.
The FP amplifier bandwidth can be determined as follows [1–7]:
" pffiffiffiffiffiffiffiffiffiffi #
2ΔνL 1  G R R2
ΔνA ¼ sin 1  pffiffiffiffiffiffiffiffiffiffi11=2 : ð3:37Þ
π 4G R1 R2

3.1.5.2 Raman Amplifiers

A fiber-based Raman amplifier employs the stimulated Raman scattering (SRS) occurring in silica fibers when an intense
pump propagates through it [1–7]. SRS fundamentally differs from stimulated emission: in stimulated emission, an incident
photon stimulates emission of another identical photon; in SRS, the incident pump photon gives up its energy to create another
photon of reduced energy at a lower frequency (inelastic scattering); and the remaining energy is absorbed by the medium in
the form of molecular vibrations (optical phonons). Raman amplifiers must be pumped optically to provide gain, as shown in
Fig. 3.18.
The Raman gain coefficient gR is related to the optical gain g(z) as g(z) ¼ gRIp(z), where Ip is the pump intensity given by
Ip ¼ Pp/ap, with Pp being the pump power and ap being the pump cross-sectional area. Since the cross-sectional area is
different for different types of fibers, the ratio gR/ap is the measure of the Raman gain efficiency. The DCF efficiency can even
174 3 Components, Modules, and Subsystems

Fig. 3.18 The Raman amplifier

operation principle in forward-
pumping configuration

Fig. 3.19 The general EDFA 980 nm or 1480 nm pump Erbium

configuration
doped
Pump laser fiber Amplified
Isolator optical signal

Fiber coupler
Weak incoming
optical signal Narrowband
filter

be eight times better than that of a standard SMF [1–7]. The evolution of the pump Pp and signal Ps powers (in distance z) can
be studied by solving the system of coupled differential equations below [1–7]:

dPs g dPp ωp g R
¼ αs Ps þ R Pp Ps , ¼ αp Pp  P P, ð3:38Þ
dz ap dz ω s ap p s

where as denotes the signal cross-sectional area ωp and ωs denote the pump and signal frequency, respectively, while other
parameters are already introduced above.
In small-signal amplification regime (when the pump depletion can be neglected), the pump power evolution is exponen-
tial, Pp(z) ¼ Pp(0)exp[αpz], so that the Raman amplifier gain is found to be:
 
Ps ð0Þ exp gR Pp ð0ÞLeff =ap  αs L
GA ¼ ¼ exp ðg0 LÞ,
Ps ð0Þ exp ðαs LÞ
ð3:39Þ
Pp ð0Þ Leff gR Pp ð0Þ
g0 ¼ gR  ðαP L >> 1Þ:
ap L ap αp L

The origin of saturation in Raman amplifiers is pump power depletion, which is quite different from that in SOAs.
Saturated amplifier gain Gs can be determined (assuming αp ¼ αs) by [1–7]:

1 þ r0 ωp Ps ð0Þ
GS ¼ , r0 ¼ ð3:40Þ
r 0 þ 1=G1þr
A
0 ωs P p ð 0 Þ

The amplifier gain is reduced down by 3 dB when GAr0  1, the condition that is satisfied when the amplified signal power
becomes comparable to the input pump power Ps(L) ¼ Pp(0). Typically P0 ~ 1 W, and channel powers in a WDM systems are
around 1 mW, meaning that Raman amplifier operates in unsaturated or linear regime.

3.1.5.3 Erbium-Doped Fiber Amplifier (EDFA)

The rare-earth-doped fiber amplifiers are finding increasing importance in optical communication systems. The most
important class is erbium-doped fiber amplifiers (EDFAs) due to their ability to amplify in 1.55 μm wavelength range. The
active medium consists of 10–30 m length of optical fiber highly doped with a rare-earth element, such as erbium (Er),
ytterbium (Yb), neodymium (Nd), or praseodymium (Pr). The host fiber material can be pure silica, a fluoride-based glass, or a
multi-component glass. General EDFA configuration is shown in Fig. 3.19.
The pumping at a suitable wavelength provides gain through population inversion. The gain spectrum depends on the
pumping scheme as well as on the presence of other dopants, such as Ge or Al within the core. The amorphous nature of silica
broadens the energy levels of Er3+ into the bands, as shown in Fig. 3.20.
3.1 Key Optical Components, Modules, and Subsystems 175

980 nm
4I11/2

4I13/2 1530 nm

980 nm
1480 nm
Pump Emission
Bands wavelengths

4I15/2

Fig. 3.20 The EDFA energy diagram

The pumping is primarily done in optical domain with the primary pump wavelengths at 1.48 μm and 0.98 μm. The atoms
pumped to the 4I11/2 0.98 μm decay to the primary emission transition band. The pumping with 1.48 μm is directly to the
upper transition levels of the emission band.
EDFAs can be designed to operate in such a way that the pump and signal travel in opposite directions; this configuration is
commonly referred to as backward pumping. In bidirectional pumping, the amplifier is pumped in both directions simulta-
neously by using two semiconductor lasers located at the both fiber ends.

3.1.6 Optical Processing Components and Modules

Different optical components can be classified into two broad categories depending on whether they can operate without an
external electric power source or not into passive or active. Important active components are lasers, external modulators,
optical amplifiers, photodiodes, optical switches, and wavelength converters. On the other hand, important passive
components are optical couplers, isolators, multiplexers/demultiplexers, and filters. Some components, such as optical filters,
can be either passive or active depending on operational principle. In this section, we will briefly explain some important
optical components not being described in previous subsections.

3.1.6.1 Optical Couplers

The 2
2 optical coupler is a fundamental device that can be implemented using the fiber fusing or be based on graded-index
(GRIN) rods and optical filters, as shown in Fig. 3.21. The fused optical couplers (shown in Fig. 3.21(a)) are obtained when
the cladding of two optical fibers are removed; the cores are brought together and then heated and stretched. The obtained
waveguide structure can exchange energy in the coupling region between the branches. If both inputs are used, 2
2 coupler
is obtained; if only one input is used, 1
2 coupler is obtained. The optical couplers are recognized either as optical taps
(1
2) couplers or directional (2
2) couplers. The power coupler splitting ratio depending on purpose can be different with
typical values being 50%/50%, 10%/90%, 5%/95%, and 1%/99%. Directional coupler parameters (defined when only input
1 is active) are splitting ratio Pout,1 / (Pout,1 + Pout,2), excess loss 10log10[Pin,1/(Pout,1 + Pout,2)], insertion loss 10log10(Pin,i/Pout,
j), and crosstalk 10log10(Pcross/Pin,1).
The operation principle of directional coupler can be explained using coupled mode theory [1, 2, 6, 9, 10] or simple
scattering (propagation) matrix S approach, assuming that coupler is lossless and reciprocal device:

E out,1 E in,1 s11 s12 E in,1 cos ðkLÞ j sin ðkLÞ Ein,1
¼S ¼ ¼ ejβL , ð3:41Þ
E out,2 E in,2 s21 s22 E in,2 j sin ðkLÞ cos ðkLÞ Ein,2
176 3 Components, Modules, and Subsystems

Fig. 3.21 Optical couplers: (a) Ein,1 Pin,1 Pout,1 Eout,1

fiber fusing based and (b) GRIN
rode based
L

Pref
Pcross
Coupling region

Ein,2 Eout,2
Pout,2
(a)
Inputs GRIN rod GRIN rod Output fiber

Optical filter
(b)

where β is propagation constant, k is coupling coefficient, L is the coupling region length, Ein,1 and Ein,2 are corresponding
inputs electrical fields, and Eout,1 and Eout,2 are corresponding output electrical fields. Scattering matrix S elements are denoted
with sij. For example, for 3 dB coupler, we have to select kL ¼ (2 m + 1)π/4 (m is a positive integer) to get:

Eout,1 1 1 j Ein,1
¼ pffiffiffi : ð3:42Þ
Eout,2 2 j 1 Ein,2

The combination of two GRIN rods and an optical filter can effectively be used as an optical coupler, as illustrated in
Fig. 3.21(b). The GRIN rods are used as collimators, to collimate the light from two input ports and deliver to the output port,
while the optical filter is used to select a desired wavelength channel.
The optical couplers can be used to create more complicated optical devices such as MxN optical stars, directional optical
switches, different optical filters, multiplexers, etc.

3.1.6.2 Optical Filters

An optical filter modifies the spectrum of incoming light and can mathematically be described by corresponding transfer
function Hof(ω):

Z1
1 e in ðωÞH of ðωÞejωt dω,
E out ðt Þ ¼ E ð3:43Þ
2π
1

where Ein(t) and Eout(t) denote the input and output electrical field, respectively, and we used ~ to denote the FT as before.
Depending on the operational principle, the optical filters can be classified into two broad categories as diffraction or
interference filters. The important classes of optical filters are tunable optical filters, which are able to dynamically change
the operating frequency to the desired wavelength channel. The basic tunable optical filter types include tunable 2
2
directional couplers, Fabry-Perot (FP) filters, Mach-Zehnder (MZ) interferometer filters, Michelson filters, and acousto-
optical filters. Two basic optical filters, FP filter and MZ interferometer filter, are shown in Fig. 3.22. An FP filter is in fact a
cavity between two high-reflectivity mirrors. It can act as a tunable optical filter if the cavity length is controlled, for example,
by using a piezoelectric transducer. Tunable FP filters can also be made by using liquid crystals, dielectric thin films,
semiconductor waveguides, etc. A transfer function of an FP filter whose mirrors have the same reflectivity R and the cavity
length L can be written as [1–7]:
3.1 Key Optical Components, Modules, and Subsystems 177

Mirrors

Optical fiber Optical fiber

L 3 dB 3 dB
coupler Delay coupler
R R

Piezoelectric transducer

(a) (b)
Fig. 3.22 Basic optical filters: (a) Fabry-Perot filter and (b) Mach-Zehnder interferometer

ð1  RÞejπ
H FP ðωÞ ¼ , τ ¼ 2L=vg ð3:44Þ
1  Re jωτ

where τ is the round-trip time within the cavity and vg is the group velocity. The transfer function of FP filter is periodic with
period being the free spectral range (FSR) Δν ¼ vg/(2 L ). Another FP filter important parameter is the finesse defined as [1–7]:
pffiffiffi
π R
F ¼ Δν =ΔνFP ffi , ð3:45Þ
1R

where ΔνFP is the FP transmission peak width.

Mach-Zehnder interferometer filters, shown in Fig. 3.22(b), can also be used as tunable optical filters. The first coupler
splits the signal into two equal parts, which acquire different phase shifts before they interfere at the second coupler. Several
MZ interferometers can be cascaded to create an optical filter. When cross output of 3 dB coupler is used, the square
magnitude of transfer function is [1–7] |Hc(ω)| ¼ cos2(ωτ/2), so that the transfer function of M-stage MZ filter based on 3 dB
couplers can be written as:

M
jH MZ ðωÞj2 ¼ ∏ cos 2 ðωτm =2Þ, ð3:46Þ
m¼1

where τm is the adjustable delay of mth (m ¼ 1, 2,. . ., M) cascade.

3.1.6.3 WDM Multiplexers and Demultiplexers

Multiplexers and demultiplexers are basic devices of a WDM system. Demultiplexers contain a wavelength-selective element
to separate the channels of a WDM signal. Based on underlying physical principle, different demultiplexer devices can be
classified as diffraction-based demultiplexers (based on a diffraction grating) and interference-based demultiplexers (based on
optical filters and directional couplers).
Diffraction-based demultiplexers are based on Bragg diffraction effect and use an angular dispersive element, such as the
diffraction grating. The incoming composite light signal is reflected from the grating and dispersed spatially into different
wavelength components, as shown in Fig. 3.23(a). Different wavelength components are focused by lenses and sent to
individual optical fibers. The same device can be used as multiplexer by switching the roles of input and output ports. This
device can be implemented using either conventional or GRIN lenses. To simplify design, the concave grating can be used.
The second group of optical multiplexers is based on interference effect and employs the optical couplers and filters to
combine different wavelength channels into a composite WDM signal. The multiplexers employing the interference effect
include thin-film filters multiplexers and the array waveguide grating (AWG) [1–7, 17], which is shown in Fig. 3.23(b). The
AWG is a highly versatile WDM device, because it can be used as a multiplexer, a demultiplexer, a drop-and-insert element,
or even as a wavelength router. It consists of Minput and Moutput slab waveguides and two identical focusing planar star
couplers connected by N uncoupled waveguides with a propagation constant β. The length of adjacent waveguides in the
central region differs by a constant value ΔL, with corresponding phase difference being 2πncΔL/λ, where nc is the refractive
index of arrayed waveguides. Based on the phase-matching condition (see Fig. 3.23(c)) and knowing that the focusing is
178 3 Components, Modules, and Subsystems

Fig. 3.23 Optical multiplexers/demultiplexers: (a) grating-based demultiplexer, (b) array waveguide grating (AWG) demultiplexer, and (c)
illustrating the operating principle of AWG

achieved when the path-length difference ΔL between adjacent array waveguides is an integer multiple of the central design
wavelength λc, that is, ncΔL ¼ mλc, we derive the following expression for the channel spacing [17]:

y ns cd nc
Δν ¼ , ð3:47Þ
L mλ2 ng

where d is the spacing between the grating array waveguides, y is the spacing between the centers of output ports, L is the
separation between center of arrayed waveguides and center of output waveguides, nc is the refractive index of waveguides in
grating array, ns is the refractive index of star coupler, ng the group index, and m is the diffraction order.
The free spectral range can be obtained by [17]:

c
ΔνFSR ¼  , ð3:48Þ
ng ΔL þ d sin Θin,i þ d sin Θout,j

where the diffraction angles from the ith input Θin,i and jth output Θout,j ports (measured from the center of the array) can be
found as Θin,i ¼ iy/L and Θout,j ¼ jy/L, respectively.

3.1.7 Principles of Coherent Optical Detection

In order to exploit the enormous bandwidth potential of the optical fiber different multiplexing techniques (OTDMA, WDMA,
CDMA, SCMA), modulation formats (OOK, ASK, PSK, FSK, PolSK, CPFSK, DPSK, etc.), demodulation schemes (direct
3.1 Key Optical Components, Modules, and Subsystems 179

Fig. 3.24 Block diagrams of (a)

direct detection scheme and (b) np Photodetector Integrator
coherent detection scheme
np Photodetector Integrator nLO
(a) (b)

Input 1

Output 2

Output 1
Control
voltage Phase Input 2
trimmer

Fig. 3.25 Optical hybrid

detection or coherent optical detection) and technologies were developed, and some important aspects are discussed in this
section. The coherent optical detection offers several important advantages compared to direct detection: (i) improved receiver
sensitivity, (ii) better frequency selectivity, (iii) possibility of using constant amplitude modulation formats (FSK, PSK),
(iv) tunable optical receivers similar to RF receivers are possible, and (v) with coherent detection the chromatic dispersion and
PMD can easier be mitigated.
There is local laser, known as local oscillator laser, employed within coherent optical receiver. We can recognize several
detection schemes based on the value of operating frequency of the local laser with the respect to the frequency of incoming
optical signal, and they are (i) homodyne detection, in which these frequencies are identical; (ii) heterodyne detection, in
which the frequency difference is larger than the signal symbol rate, so that all related signal processing upon photodetection
is performed at suitable intermediate frequency (IF); and (iii) intradyne detection, in which the frequency difference is smaller
than the symbol rate but higher than zero. Further on, different coherent detection schemes can be classified into following
categories: (i) synchronous (PSK, FSK, ASK) schemes, (ii) asynchronous (FSK,ASK) schemes, (iii) differential detection
(CPFSK, DPSK) schemes, (iv) phase diversity reception schemes, (v) polarization diversity reception schemes, and
(vi) polarization-division multiplexing schemes. Synchronous detection schemes can further be categorized as residual carrier
or suppressed carrier ones.
In Fig. 3.24 we show the basic difference between direct detection (Fig. 3.24(a)) and coherent detection (Fig. 3.24(b))
schemes. Coherent optical detection, in addition to the photodetector and integrator already in use for direct detection,
employs a local laser whose frequency and polarization is matched to the frequency and polarization of incoming optical
signal. The incoming optical signal and local laser output signal are mixed in optical domain (a mirror is used in this
illustrative example; in practice, an optical hybrid, described in incoming section, is used instead).
For additional details on coherent optical systems, an interested reader is referred to [1, 3, 5, 10, 26, 28].

3.1.8 Optical Hybrids

So far we used an optical mirror to perform optical mixing before photodetection takes place. In practice this operation is
performed by four-port device known as optical hybrid, which is shown in Fig. 3.25.
180 3 Components, Modules, and Subsystems

Electrical fields at output ports E1o and E2o are related to the electrical fields at input ports E1in and E2in as follows:
pffiffiffiffiffiffiffiffiffiffiffi
E 1o ¼ ðE 1i þ E 2i Þ 1  k
pffiffiffi ð3:49Þ
E 2o ¼ ðE 1i þ E 2i exp ðjϕÞÞ k

where k is the power splitting ratio and ϕ is the phase shift introduced by the phase trimmer (see Fig. 3.25). The Eqs. (3.49) can
also be written in terms of scattering (S-) matrix as follows:

" pffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffi #
E o1 E o1 s11 s12 1k 1k
Eo ¼ ¼S ¼ SEi , S¼ ¼ pffiffiffi pffiffiffi ð3:50Þ
E o2 E o2 s21 s22 k e jϕ
k

In expressions (3.49) and (3.50), we assumed that hybrid is lossless device, which leads to s11 ¼ |s11|, s12 ¼ | s12|, s21 ¼ | s21|,
and s22 ¼ j s22jexp(jθ22). Popular hybrids are π hybrid, which S-matrix can be written as (by setting ϕ ¼ π in (3.50)):
" pffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffi #
1k 1k
S¼ pffiffiffi pffiffiffi , ð3:51Þ
k  k

and π/2 hybrid, which S-matrix can be written as:

s11 s12
S¼ : ð3:52Þ
s21 s22 ejπ=2

Well-known π hybrid is 3 dB coupler (k ¼ 1/2). If π/2 hybrid is symmetric |sij| ¼ 1/Loss (8i, j), the phase difference between
the input electrical fields E1i ¼ jE 1i j, E 2i ¼ jE2i je jθi can be chosen on such a way so that total output ports power

2  pffiffiffiffiffiffiffiffiffiffiffiffi
E{0 E0 ¼ P þ P2i þ P1i P2i ð cos θi þ sin θi Þ ð3:53Þ
Loss 1i

is maximized. (We use { to denote Hermitian transposition-simultaneous transposition and complex

pffiffiffi conjugation.) For equal
input powers, the maximum of (3.53) is obtained when θi ¼ π/4, leading to Loss  2 þ 2 . The corresponding loss in
dB-scale is determined as 10log10(Loss/2) ¼ 2.333 dB. For Costas loop and decision-driven loop-based homodyne systems,
there exists optimum k in corresponding S-matrix:
" pffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffi #
1 1k 1k
S ¼ pffiffiffiffiffiffiffiffiffiffi pffiffiffi pffiffiffi : ð3:54Þ
Loss k j k

3.1.9 Coherent Optical Balanced Detectors

To reduce the relative intensity noise (RIN) of transmitting laser and to eliminate the direct-detection and signal-cross-signal
interferences, the most deleterious sources for multi-channel applications, the balanced coherent optical receiver, shown in
Fig. 3.26, is commonly used. For homodyne detection, by ignoring the laser phase noise, the upper and lower photodetectors
output currents can respectively be written as:
0 1
1 B pffiffiffiffiffiffiffiffiffiffiffiffiffi C
i1 ðt Þ ¼ RjE1 j2 ¼ R@ jE s j2 þ jE LO j2 þ 2 PS PLO cos θS A þ n1 ðt Þ
2 |ffl{zffl} |fflffl{zfflffl} ð3:55Þ
PS PLO
1  pffiffiffiffiffiffiffiffiffiffiffiffiffi 
i2 ðt Þ ¼ RjE2 j2 ¼ R PS þ PLO  2 PS PLO cos θS þ n2 ðt Þ
2

where θs is the phase of incoming optical signal and ni(t) (i ¼ 1, 2) is the i-th photodetector shot noise process of PSD
Sni ¼ qRjEi j2 . The balanced receiver output current (see Fig. 3.26) can be written as:
3.1 Key Optical Components, Modules, and Subsystems 181

E1 i1
ES i=i1–i2
π hybrid
ELO i2
E2

Fig. 3.26 Balanced detector

Fig. 3.27 Directional coupler-

based coherent optical detection
schemes for 2D signaling (PSK,
QAM): (a) balanced coherent
optical detector architecture and
(b) single optical balanced
detector based on heterodyne
design. LPF low-pass filter

pffiffiffiffiffiffiffiffiffiffiffiffiffi
iðt Þ ¼ i1 ðt Þ  i2 ðt Þ ¼ 2R PS PLO cos θS þ nðt Þ, nðt Þ ¼ n1 ðt Þ  n2 ðt Þ ð3:56Þ

where n(t) is a zero-mean Gaussian process of PSD Sn ¼ Sn1 þ Sn2 ¼ qRðPS þ PLO Þ ffi qRPLO . For binary BPSK signaling,
we have that θs ¼ π.
The M-ary PSK, M-ary QAM, and M-ary DPSK all achieve the transmission of log2M ¼ m bits per symbol. As an
illustration, in coherent optical detection, the data phasor for M-ary PSK ϕk2{0,2π/M,..,2π(M  1)/M} is sent at each k-th
transmission interval. We should also outline that in direct detection, the modulation is differential since the data phasor
ϕk ¼ ϕk1 + Δϕk is sent instead, where Δϕk 2{0,2π/M,.., 2π(M  1)/M} value is determined by the sequence of log2M input
bits using an appropriate mapping rule. The mapper accepts m bits, c ¼ (c1,c2,..,cm), from the input sequence at time instance
i and determines the corresponding M-ary (M ¼ 2m) constellation point sk ¼ (Ik,Qk) ¼ |Sk|exp(jθk). The coordinates from the
mapper, after the upsampling, are used as the RF inputs of optical I/Q modulator.
Let the input to the coherent detector be given by Sk ¼ |Sk|exp[j(ωTt + ϕPN,T + θk)], where the subscript k is used to denote
the kth symbol being transmitted, ωT denotes the transmit laser frequency (in rad/s), ϕPN,T denotes the laser phase noise of
transmitter laser, and |Sk| and θk denote the amplitude and the phase of the kth symbol for M-ary QAM/PSK (or any other 2D
signal constellation). Let the local laser (oscillator) output be denoted by L ¼ |L|exp.[j(ωLOLt + ϕPN,LOL)], where ωLOL denotes
the local oscillator laser frequency and ϕPN,LOL denotes the laser phase noise of local laser. There exist different versions of
balanced coherent optical detectors for 2D signal constellations. The directional coupler-based balanced coherent optical
detector architecture is shown in Fig. 3.27(a). The scattering matrix of 3 dB directional coupler, based on Eq. (3.42), is given
182 3 Components, Modules, and Subsystems

1 j
by S ¼ p1ffiffi : The electrical fields’ outputs E1 and E2, in the upper branch, are related to input electrical fields Sk
2j 1
and L, in matrix form, as follows:
2 3
1


pffiffiffi ejπ=2 Sk
E1 1 1 j 6 6 2
7
7,
¼ pffiffiffi 4 5 ð3:57Þ
E2 2 j 1 1
pffiffiffi L
2

while the corresponding scalar form representation is given by:

1 1
E 1 ¼ jðSk þ LÞ, E2 ¼ ðSk þ LÞ: ð3:58Þ
2 2

The powers at the output ports 1 and 2 can be represented as:

 
1
P1 ¼ jE 1 j2 ¼ E 1 E 1 ¼ jSk j2 þ jLj2 þ Sk L þ Sk L ,
4
  ð3:59Þ
1
2
P2 ¼ jE 2 j ¼ E 2 E 2 ¼

jSk j2 þ jLj2  Sk L  Sk L :
4

Assuming the unit load resistance in upper branch of the balanced coherent optical detector, the in-phase component is
obtained as (by ignoring the shot and thermal noises):
 
1 
vI,k ffi R jE 1 j2  jE2 j2 ¼ R Sk L þ Sk L
2 2 3
6 7 ð3:60Þ
¼ RjSk jjLj cos 4ðωT  ωLO Þt þ θk þ ϕPN,T  ϕPN,LOL 5
|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
ωIF ϕPN

¼ RjSk jjLj cos ðωIF t þ θk þ ϕPN Þ,

where ωIF ¼ |ωT  ωLOL| denotes the intermediate frequency, while ϕPN ¼ ϕPN, T  ϕPN, LOL denotes the total phase noise
originating from both transmit and local oscillator lasers.
In similar fashion, the electrical fields’ outputs E3 and E4, in the lower branch of Fig. 3.27(a), are related to input electrical
fields Sk and L, in matrix form, as follows:
2 3
1


pffiffiffi Sk
E3 1 1 j 6 2 7
¼ pffiffiffi 6 7, ð3:61Þ
E4 2 j 1 4 p1ffiffiffi L 5
2

while the corresponding scalar form representation is given by:

1 1
E 3 ¼ ðSk þ jLÞ, E4 ¼ ðjSk þ LÞ: ð3:62Þ
2 2

The powers at the output ports 3 and 4 can be represented as:

 
1
P3 ¼ jE3 j2 ¼ E 3 E 3 ¼ jSk j2 þ jLj2  jSk L þ jSk L ,
4
  ð3:63Þ
1
2
P4 ¼ jE4 j ¼ E 4 E 4 ¼

jSk j2 þ jLj2 þ jSk L  jSk L :
4
3.2 Modules/Components Relevant to Wireless Communications 183

Assuming the unit load resistance in the lower branch of the balanced coherent optical detector, the in-phase component is
obtained as (by ignoring the shot and thermal noises):
 
1  
vQ,k ffi R jE 3 j2  jE 4 j2 ¼ R Sk L  Sk L
2 2j 3
6 7 ð3:64Þ
¼ RjSk jjLj sin 4ðωT  ωLO Þt þ θk þ ϕPN,T  ϕPN,LOL 5
|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
ωIF ϕPN
¼ RjSk jjLj sin ðωIF t þ θk þ ϕPN Þ:

To summarize, the balanced coherent optical detector outputs of I- and Q-channel branches at the time instance k can be
written as:

vI,k ffi RjSk jjLj cos ðωIF t þ θk þ ϕPN Þ,

ð3:65Þ
vQ,k ffi RjSk jjLj sin ðωIF t þ θk þ ϕPN Þ,

where R is the photodiode responsivity and ωIF ¼ |ωT  ωLOL| is the intermediate frequency, while ϕPN is the total laser phase
noise originating from both transmit laser and LOL.
The heterodyne receiver can also be implemented based only on single balanced detector, such as lower branch of
Fig. 3.27(a), as shown in Fig. 3.27(b). For ASE noise-dominated scenario, both heterodyne schemes perform comparably.
However, in the shot noise-dominated scenario, the heterodyne design with single optical balanced detector performs better.
In homodyne detection, we need to set the intermediate frequency to zero, ωIF ¼ 0.

3.2 Modules/Components Relevant to Wireless Communications

In this section we describe the modules/components relevant to wireless communications [36–43], in particular to software-
defined radio (SDR). The basic concept related to antennas has already been provided in Chap. 2. Many of the modules
described here are also applicable in software-defined optical transmission (SDOT).
We first describe some basic concepts from DSP relevant in this section [44–49]; for additional details, an interested reader
is referred to the Appendix. Alternatively, the readers can refer to DSP books such as [45, 47].

3.2.1 DSP Basics

We start the description of DSP basic by describing the role of the system (transfer) function as well as establishing
relationship between discrete-time (DT) and continuous-time (CT) Fourier transforms.

3.2.1.1 The Role of System (Transfer) Function and Relationship Between DT and CT Fourier Transforms
The discrete-time (DT) linear time-invariant (LTI) system can be described by linear difference equation with constant
coefficients:

aN yðn  N Þ þ aN1 yðn  ðN  1ÞÞ þ ⋯ þ a0 yðnÞ ¼ bM xðn  M Þ þ ⋯ þ b0 xðnÞ, ð3:66Þ

where input and output to the system are denoted with x(n) and y(n), respectively. By applying z-transform on Eq. (3.66), we
can determine the system function (also known as the transfer function) H(z) as follows:

P
M
bm zm
Y ðzÞ m¼0
H ðzÞ ¼ ¼ N , ð3:67Þ
X ðzÞ P
an zn
n¼0
184 3 Components, Modules, and Subsystems

where X(z) [Y(z)] denotes the z-transform of x(n) [y(n)], namely, X(z) ¼ Z{x(n)} [Y(z) ¼ Z{y(n)}]. The frequency response can
be obtained by substituting z with e jΩ:
 jΩ

H e ¼ H ðzÞjz¼e jΩ : ð3:68Þ

The unit-sample (impulse) response can be obtained as inverse z-transform of the system function: h(n) ¼ Z1{H(z)}. The
output of the system y(n) can be obtained as a linear convolution of the input to the system x(n) and impulse response h(n),
which upon application of the convolution property yields to:
( )
X
1
Y ð z Þ ¼ Z fy ð nÞ g ¼ Z xðk Þhðn  k Þ ¼ X ðzÞH ðzÞ: ð3:69Þ
k¼1

Therefore, the system function has a central role in DT system analysis, as illustrated in Fig. 3.28.
Two important system properties of DT systems are causality and stability. The system is causal if there is no response
before excitation. It can be shown that the system is causal if and only if h(n) ¼ 0 for n < 0. The LTI system is causal if an only
if the region of convergence (ROC) is the exterior of circle of radius r < 1. On the other hand, DT system is bounded input-
bounded output (BIBO) stable if every bounded input sequence x(n) produces a bounded output sequence y(n). It can be
shown that an LTI system is BIBO stable if and only if the ROC of system function contains the unit circle. Finally, a causal
LTI system is BIBO stable if and only if all the poles of the system function are located inside the unit circle.
The relationship between Fourier transforms of a continuous-time (CT) signal xa(t) and a DT signal x(n) obtained by
sampling (every T seconds) is illustrated in Fig. 3.29. The z-transform can be related to DT Fourier transform (DTFT) by
simply substituting z ¼ exp.( jΩ). In other words, the DTFT is the z-transform evaluated on the unit circle. Because of the
periodicity path along the unit circle, the DTFT will be periodic with a period of 2π.

Fig. 3.28 The role of the system

(transfer) function

Fig. 3.29 The relationship

between DT and CT Fourier
transforms. We use the subscript
a to denote the analog signals
3.2 Modules/Components Relevant to Wireless Communications 185

3.2.1.2 Discrete-Time System Realizations

When a0 ¼ 1, we can express y(n) as the function of previous LTI system outputs and current and previous LTI system inputs,
based on Eq. (3.66), as follows:

yðnÞ ¼ a1 yðn  1Þ  ⋯  aN yðn  N Þ þ bM xðn  M Þ þ ⋯ þ b0 xðnÞ, ð3:70Þ

with the corresponding implementation (realization), known as the direct realization, shown in Fig. 3.30. The systems for
which only forward coefficients exist are commonly referred to as the finite impulse response (FIR) systems (filters). On the
other hand, when there exist nonzero feedback coefficients, for M < N, the impulse response will be infinitely long, and such
systems are known as the infinite impulse response (IIR) systems (filters).
By separating the all-zero (AZ) and all-pole (AP) subsystems’ transfer functions as follows:

P
M
!
bm zm X
M
m¼0 1
H ðzÞ ¼ ¼ bm zm  ¼ H AZ ðzÞH AP ðzÞ, ð3:71Þ
P
N
m¼0
P
N
an zn |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} an zn
n¼0 n¼0
H AZ ðzÞ |fflfflfflfflffl{zfflfflfflfflffl}
H AP ðzÞ

the corresponding realization, known as the direct canonic form I, is shown in Fig. 3.31. Since HAZ(z)HAP(z) ¼ HAP(z)HAZ(z),
we can swap the AZ and AP subsystems so that we can use the same delay elements for both subsystems, and corresponding
realization, shown in Fig. 3.32, is known as direct canonic form II.

Fig. 3.30 The direct realization

of DT LTI systems

Fig. 3.31 The direct canonic

form I
186 3 Components, Modules, and Subsystems

Fig. 3.32 The direct canonic form II

Fig. 3.33 The cascade

realization of DT LTI system

By factoring the system function as follows:

K bk0 þ bk1 z1 þ bk2 z2

H ðzÞ ¼ ∏ H k ðzÞ, H k ðzÞ ¼ , ð3:72Þ
k¼1 1 þ ak1 z1 þ ak2 z2

the corresponding realization, known as the cascade realization, is shown in Fig. 3.33. The second-order section Hk(z), from
Eq. (3.72), can be obtained from Fig. 3.32 by setting M ¼ N ¼ 2.
3.2 Modules/Components Relevant to Wireless Communications 187

H1(z)

x(n) H2(z) y(n)

HK(z)

Fig. 3.34 The parallel realization of DT LTI system

By performing the partial fraction expansion, the system function can be represented as:

X
K
bk0 þ bk1 z1 þ bk2 z2
H ðzÞ ¼ H k ðzÞ, H k ðzÞ ¼ , ð3:73Þ
k¼1
1 þ ak1 z1 þ ak2 z2

and the corresponding realization, known as the parallel realization, is shown in Fig. 3.34. Notice that some sections in
parallel realization can be of the first order; in that case, we need to set bk1 ¼ bk2 ¼ 0, ak2 ¼ 0.
Design of IIR filters (subsystems) can be performed by designing first the analog filters and then converting to digital filter
by proper mapping. Regarding the mapping from the s-domain to z-domain, the following two conditions must be satisfied:
(i) imaginary axis in s-plane should map into the unit circle in z-plane, and (ii) the left half-plane of s-plane should map into
inside of unit circle in the z-plane. The bilinear transform, given by:

2 1  z1
s¼ , ð3:74Þ
T 1 þ z1

satisfies both conditions. By using this mapping, the continuous-time frequency [rad/s] ω is mapped to the DT frequency
[rad/sample] Ω by the following nonlinear mapping:
 
ωT
Ω ¼ 2 tan 1 : ð3:75Þ
2

3.2.2 Direct Digital Synthesizer (DDS)

The direct digital synthesizer (DDS) in discrete-time (DT) communication systems has the same role as the voltage control
oscillator (VCO) in continuous-time (CT) communication systems. In VCO, the output frequency is directly proportional to
the input signal x(t) so that the output phase ϕ(t) will be the integral of the input:

Zt
ϕð t Þ ¼ a0 xðt 0 Þdt 0 , ð3:76Þ
1

where a0 is the VCO gain [rad/V]. The VCO output signal will be the following cosinusoidal signal:
0 1
Zt
yðt Þ ¼ cos @ω0 t þ a0 xðt 0 Þdt 0A: ð3:77Þ
1

In similar fashion, the output of the DDS will be:

188 3 Components, Modules, and Subsystems

Fig. 3.35 The block diagram of

direct digital synthesizer (DDS)
with DSP implementation details

Fig. 3.36 The block diagram of

the DDS with finite precision

!
X
n1
yðnT Þ ¼ cos Ω0 n þ A0 xðkT Þ , Ω0 ¼ ω0 T, A0 ¼ a0 T: ð3:78Þ
k¼1

The block diagram of DDS together with corresponding DSP representation is shown in Fig. 3.28. The portion of the DDS
calculating the accumulated phase is known as the phase accumulator.
The DDS shown in Fig. 3.35 does not take the finite precision of the phase accumulator representation. A realistic DDS
with finite precision taken into account is shown in Fig. 3.36, where we use Q() to denote the quantization operation. The
number of quantization bits NQ < N can be determined as NQ ¼ NNt, where Nt is the number of truncated least significant
bits. Let us assume that two’s complement integer representing the phase at the output of accumulator is given by
a0a1⋯aN  1 ¼  a02N  1 + a12N  2 + ⋯ + aN  120. The decimal representation of the accumulated phase would be
θ0θ1⋯θN  1 ¼  θ020 + θ121 + ⋯ + θN  12(N  1). Since the sampling frequency is given by Fs ¼ 1/T, the maximum
frequency, according to the sampling theorem, will be fmax ¼ Fs/2. The minimum frequency is the resolution frequency
determined by fmin ¼ Fs/2N. The number of required bits for the phase accumulator content representation will be then:


f max F =2 f max
¼ s N ¼ 2N1 ) N ¼ log 2 þ1 , ð3:79Þ
f min F s =2 f min

where we use de to denote the smallest integer larger than or equal to the enclosed
l quantity.
h i As man illustration, for
400106
fmax ¼ 400 MHz and resolution of fmin ¼ 0.1 Hz, the number of required bits is N ¼ log 2 0:1 þ 1 ¼ 33, so that the
size of LUT for cos() is too large, 8.58993459210.9 To reduce the size of LUT, one method is to use the truncation,
mentioned above and illustrated in Fig. 3.36. The truncation method, however, will introduce the phase error Δθ(n) in the
represented phase θðnÞ ¼ b θðnÞ þ ΔθðnÞ: The corresponding spectrum of complex exponential will contain the spurious
spectral lines. The complex exponential can be represented by:

^ ^ ΔθðnÞ>0 ^
e jθðnÞ ¼ e j½θðnÞþΔθðnÞ ¼ e jθðnÞ e jΔθðnÞ ffi e jθðnÞ ½1 þ jΔθðnÞ ð3:80Þ
     
ffi cos ^θðnÞ  ΔθðnÞ  sin ^ θðnÞ þ j sin ^
θðnÞ þ ΔθðnÞ  cos ^ θðnÞ :
3.2 Modules/Components Relevant to Wireless Communications 189

Fig. 3.37 The compensation of

the spurious spectral lines by the
phase dithering method

Fig. 3.38 The compensation of the phase error by the feedforward error compensation method

Fig. 3.39 The compensation of

the phase error by the feedback
error compensation method

Based on Eq. (3.80), the first line, we conclude that it is possible to average out the phase error by phase dithering, as
illustrated in Fig. 3.37. The amplitude should be sufficiently small (on the order of 2N), and the samples of the random
generator should follow the probability density function of the phase error, with proper tails included. The end result will be
the compensation of the spurious spectral line, and the uncompensated phase error will appear as the white noise with level
well below that of the desired spectral line.
Another method to compensate the phase error will be the application of the second line in Eq. (3.80), and corresponding
phase error correction circuit is shown in Fig. 3.38. However, this approach requires additional multiplication and addition
operations.
It is also possible to consider the phase error as the noise process and employ the linear predictive approaches to estimate
the quantization error and subtract it from the phase accumulator output, as illustrated in Fig. 3.39.

3.2.3 Multirate DSP and Resampling

We are interested here in DSP modules employing multiple sampling rates, as illustrated in Fig. 3.40. In particular, the most
interesting case is when the ratio of sampling rates fy/fx is a rational number. When this ratio fy/fx ¼ 1/D (D is positive integer)
is smaller than 1, it is commonly referred to as downsampling by D or decimation by D. On the other hand, when the ratio fy/
fx ¼ U (U is positive integer) is larger than 1, the corresponding DSP operation is called upsampling by U or interpolation by
U. The DSP modules performing downsampling and upsampling operations are denoted as shown in Fig. 3.41.
190 3 Components, Modules, and Subsystems

Fig. 3.40 The illustration of DSP module employing two different sampling rates

Fig. 3.41 The notation of DSP modules representing downsampling and upsampling operations

Fig. 3.42 The LPF downsampler DSP module (decimator) performing the downsampling operation

3.2.3.1 Downsampling
Downsampling the sequence x(n) by D decimation by D represents the operation by which the sampling rate is reduced
D times, and every D-th sample from original sequence is retained so that we can write:

xD ðmÞ ¼ xðmDÞ: ð3:81Þ

By reducing the sample rate D times, by downsampling, the spectrum of XD(Ω) ¼ DTFT{xD(m)} will fold at fy/2 ¼ ( fx/2)/
D ¼ fx/(2D), and the aliasing will occur. Because of periodicity of DTFT, the period of XD(jΩ) will be 2π/D, and we can write:

D1  
 jΩ
 1 X jΩk2π
XD e ¼ X e D , ð3:82Þ
D k¼0

indicating that the spectra of aliases get expanded D times. To avoid this aliasing problem, we need first to perform low-pass
filtering with an LPF of cutoff frequency fc ¼ fx/(2D) [Hz], which is equivalent to Fc ¼ 1/(2D) [cycles/sample]. Therefore, the
block diagram of DSP module for downsampling operation is a cascade of the LPF and downsampler, as shown in Fig. 3.42.

3.2.3.2 Upsampling
Upsampling the sequence x(n) by U or interpolation by U represents the operation by which the sampling rate is increased
U times, which is achieved by inserting U1 zeros between neighboring samples in the sequence of samples so that we can
write:
3.2 Modules/Components Relevant to Wireless Communications 191

Fig. 3.43 The upsampler LPF DSP module (interpolator) performing the upsampling operation

Fig. 3.44 DSP module to

perform the resampling by a
rational factor U/D: (a) cascade of
interpolator and decimator and (b)
DSP resampling module in which
two low-pass filters are combined
into one


xðm=U Þ; m ¼ 0,  U,  2U, . . . ,
xU ðmÞ ¼ ð3:83Þ
0, else

Given that the upsampling is the opposite operation of downsampling, we would expect that the spectra of aliases in
oversampling will be compressed U times so that we can write:
 jΩ
  
XU e ¼ X ejUΩ : ð3:84Þ

To get rid of aliases, we need the LPF with the system function:
(  
 jΩy
 C, 0 Ωy  π=U
HU e ¼ ð3:85Þ
0, otherwise

and corresponding DSP module for upsampling operation is a cascade of upsampler and LPF as shown in Fig. 3.43.

3.2.3.3 Resampling
To resample the DT signal by a rational factor U/D, or in other words to perform the rate conversion by a factor U/D, we can
cascade the upsampling and downsampling modules as shown in Fig. 3.44(a). The order of interpolator and decimator is
chosen on such a way so that the two low-pass filters can be combined into one, as shown in Fig. 3.4(b), with the system
function:
(  
 jΩy
 C, 0 Ωy  min ðπ=U, π=DÞ
H e ¼ ð3:86Þ
0, otherwise
192 3 Components, Modules, and Subsystems

Fig. 3.45 The Noble identities: (a) downsampling and (b) upsampling equivalencies (identities)

Fig. 3.46 The employment of Noble identities in CIC-based: (a) decimator and (b) interpolator

3.2.3.4 Noble Identities

Important relationships for upsampling and downsampling, known as Noble identities, are summarized in Fig. 3.45. These
identities indicate that when the transfer function of the subsystem (filter) is polynomial in zN (N is a positive integer), it can be
swapped with corresponding upsampler/downsampler. This swapping enables us to implement the corresponding filters at
lower clock rates.

3.2.3.5 Cascaded-Integrator-Comb (CIC) Filter-Based Upsampling and Downsampling

One possible implementation of LPFs required in decimator and interpolator is the cascaded-integrator-comb (CIC) filter.
This is multiply-free LPF, because impulse response is all-one FIR:

1, 0 n < M
hð nÞ ¼ ð3:87Þ
0, otherwise

so that the corresponding z-transform is given by:

P
1 P
M1 1  zM
H ðzÞ ¼ hðnÞzn ¼ zn ¼
n¼1 n¼0 1  z1
1   ð3:88Þ
¼ 1  zM ¼ H integrator ðzÞH comb ðzÞ,
 z |fflfflfflfflfflffl{zfflfflfflfflfflffl}
1
1|fflfflffl{zfflfflffl}
H comb ðzÞ
H integrator ðzÞ

indicating that CIC filter can be represented as a cascade of an integrator and the comb filter, typically employed to reject
periodical frequencies (or equally spaced frequencies). The CIC filter cascaded with an upsampler/downsampler is known as a
Hogenauer filter. The CIC filter can be used either in decimator or interpolator as the LPF. By using Noble identities, the
implementation of the comb filter in both decimator and interpolator can be simplified as shown in Fig. 3.46. Given that DTFT
of CIC filter has the form:
3.2 Modules/Components Relevant to Wireless Communications 193

  
 jΩ
 1  zM  sin MΩ
H e ¼ H ðzÞjz¼e ¼ ¼  2  ejðM1ÞΩ=2 , ð3:89Þ
1  z1 z¼e
jΩ
jΩ sin Ω2

we conclude that the magnitudes of the side-band will not be sufficiently suppressed so that the multiple integrator stages
maybe used when needed.

3.2.3.6 Polyphase Decomposition

Another relevant implementation of LPFs required in decimator, interpolator, or resampling module is based on the polyphase
decomposition of the system function. Namely, the system function can be represented in the following form:

H ð z Þ ¼ hð 0Þ þ hð1Þz1 þ hð2Þz2 þ ⋯ þ hðN  1ÞzNþ1

þhðN Þz N
þ hðN þ 1Þz N1
þ hðN þ 2Þz N2
þ ⋯ þ hð2N  1Þz2Nþ1
ð3:90Þ
þhð2N Þz 2N
þ hð2N þ 1Þz 2N1
þ hð2N þ 2Þz 2N2
þ ⋯ þ hð3N  1Þz3Nþ1
þ⋯

Let the terms in the first column of Eq. (3.90) define a sub-filter H0(zN):
 
H 0 zN ¼ hð0Þ þ hðN ÞzN þ hð2N Þz2N þ ⋯ ð3:91Þ

Let the terms in the second column of Eq. (3.90), after removing the common factor z1, define the sub-filter H1(zN):
 
H 1 zN ¼ hð1Þ þ hðN þ 1ÞzN þ hð2N þ 1Þz2N þ ⋯ ð3:92Þ

Further, let the terms in the third column of Eq. (3.90), after removing the common factor z2, define the sub-filter H2(zN):
 
H 2 zN ¼ hð2Þ þ hðN þ 2ÞzN þ hð2N þ 2Þz2N þ ⋯ ð3:93Þ

Finally, let the (N  1)-th sub-filter be defined as:

 
H N1 zN ¼ hðN  1Þ þ hð2N  1ÞzN þ hð3N  1Þz2N þ ⋯ ð3:94Þ

We can then decompose the filter with system function (3.90) in terms of transfer functions of sub-filters as follows:
       
H ðzÞ ¼ H 0 zN þ H 1 zN z1 þ H 2 zN z2 þ ⋯ þ H N1 zN zðN1Þ , ð3:95Þ

and this representation is known as the polyphase filter decomposition/representation, and corresponding block scheme is
shown in Fig. 3.47.

3.2.3.7 Polyphase Filter-Based Resampling

The polyphase filter representation can be used in decimation as shown in Fig. 3.48(a), and after application of the
downsampling Noble identity, the polyphase filter can be implemented at D-times lower clock rate as shown in Fig. 3.48(b).
The use of polyphase filter structure in interpolation is shown in Fig. 3.49, both before (a) and after (b) applying the
upsampling Noble identity.
The configuration of decimator can be greatly simplified, when commutator is used, as shown in Fig. 3.50. Unit-delay
elements and downsamplers can be omitted in this case. This configuration is also known as the polyphase filter bank. In
similar fashion, the configuration of the interpolator can also be greatly simplified when the commutator is used as shown in
Fig. 3.51.
Finally, by employing the commutator and the DSP module to perform the resampling by a rational factor U/D shown in
Fig. 3.44(b), we can implement the polyphase filter-based resampling by U/D as shown in Fig. 3.52. In this module, each input
sample generates U output samples in parallel. The commutator strides through the polyphase filter bank and outputs
D samples per cycle. Once the commutator exceeds the U-th polyphase filter bank branch, it rolls over per modulo U, and
194 3 Components, Modules, and Subsystems

Fig. 3.47 The polyphase filter decomposition of the transfer function H(z)

Fig. 3.48 The use of polyphase

filter structure in decimation: (a)
before and (b) after the
downsampling Noble identity

Fig. 3.49 The use of polyphase

filter structure in interpolation: (a)
before and (b) after the
upsampling Noble identity
3.2 Modules/Components Relevant to Wireless Communications 195

Fig. 3.50 The commutator and polyphase filter-based decimator

Fig. 3.51 The commutator and polyphase filter-based interpolator

Fig. 3.52 The commutator and polyphase filter-based resampling by U/D module

the new input sample gets clocked into the polyphase filter bank. When U > D, then multiple outputs get generated for each
input clocked into the module. On the other hand, when U < D, then multiple inputs get clocked into the module between two
neighboring output samples. In hardware implementation, it is possible to use a single filter with coefficients being stored in
corresponding memory.
The polyphase filter bank has different applications including interfacing subsystems with different rates, efficient
implementation of narrowband filters, sub-band source coding, implementing precise phase shifters, digital filter banks,
and quadrature-mirror filter bank, to mention few. In sub-band coding, we break the sequence {x(n)} into different sequences
representing different frequency bands. After that we code the low-frequency bands with larger number of bits because they
are more important in speech processing.
196 3 Components, Modules, and Subsystems

3.2.4 Antenna Arrays

Antenna arrays consist of several antennas or antenna elements properly arranged in the space [50, 51]. Depending on the
space arrangement of antenna elements, antenna arrays can be classified as (i) linear array in which antenna elements are
arranged along a straight line; (ii) circular array in which antenna elements are arranged along a circular ring; (iii) planar
array in which antenna elements are ranged in 2D space over some planar geometry, such as rectangular array; and
(iv) conformal array in which antenna elements are arranged to conform some 3D object. As an illustration, two-element
linear array is shown in Fig. 3.53. The purpose of the antenna array is to provide the angle diversity to deal with fading effects.
The antenna array can also be used in multiuser communications, for instance, to separate different users. At any given
location, the received signal is not only a function of time but also a function of the angle of arrival r(t,θ), t2{1,1}, θ
2[0,2π). The antenna array with radiation pattern R(θ) can be considered as the correlator in the space domain, so that the
signal after antenna array can be represented as:

Z2π
1
r ðt, θÞRðθÞdθ ð3:96Þ
2π
0

Let the two-element array be observed, see Fig. 3.53(a), with elements, denoted as A1 and A2, separated by d, with incident
angle of incoming electromagnetic (EM) waves being θ. Clearly, the EM waves path difference is dcosθ, resulting in path
delay deference of dcosθ/c (c is the speed of light). The incident EM waves can be, therefore, represented as:

Fig. 3.53 Linear antenna arrays:

(a) two-element linear array and
(b) N-element linear array
3.2 Modules/Components Relevant to Wireless Communications 197



d
xA1 ðt Þ ¼ exp ð j2π f c t þ jφÞ, xA2 ðt Þ ¼ exp j2π f c t  cos θ þ jφ , ð3:97Þ
c

where fc is the carrier frequency and φ is the carrier phase shift. Assuming that g and φ are the gain coefficient and phase shift
of antenna element A2 with respect to antenna element A1, the resulting antenna array signal can be represented as:
 d

zðtÞ ¼ xA1 ðtÞ þ ge jϕ xA2 ðtÞ ¼ e j2π f c tþjφ 1 þ ge jϕj2π f c ccos θ
n 2  2 o1=2
¼ e j2π f c tþjφ0 1 þ gcos ðϕ  2π f c dc cos θÞ þ gsin ðϕ  2π f c dc cos θÞ ð3:98Þ
 1=2
¼ e j2π f c tþjφ0 1 þ g2 þ 2gcos ðϕ  2π f c dc cos θÞ ¼ jRðθÞje j2π f c tþjφ0 ,

where φ’ is the equivalent phase shift of the carrier and |R(θ)|2 is the magnitude squared of the radiation pattern given by:


d
jRðθÞj2 ¼ 1 þ g2 þ 2g cos ϕ  2π f c cos θ : ð3:99Þ
c

In particular, when the distance between array elements is the half of wavelength, d ¼ λc/2, the magnitude squared of the
radiation pattern simplifies to:


 λ
jRðθÞj2 d¼λc =2 ¼ 1 þ g2 þ 2g cos ϕ  π f c c cos θ
c ð3:100Þ
λc ¼c= f c
¼ 1 þ g2 þ 2g cos ðϕ  π cos θÞ:

Clearly, by varying the parameters ϕ and g, different radiation patterns can be obtained, as illustrated in Fig. 3.54.
Let us now observe the N-elements linear array, shown in Fig. 3.53(b),with neighboring elementsseparated by distance d,
whose outputs are combined with following respective complex weights g1 e jϕ1 , g2 e jϕ2 , ::, gN e jϕN : From Fig. 3.53(b), we
conclude that the path difference of the n-th ray (n ¼ 1,2,. . .,N ) with respect to the ray incident to the antenna element A1 is
equal to:

Fig. 3.54 The radiation patterns can be varied by changing the parameters g and ϕ
198 3 Components, Modules, and Subsystems

Δd ðAn , A1 Þ ¼ ðn  1Þd cos θ; n ¼ 1, 2, ⋯, N: ð3:101Þ

Based on the first line in Eq. (3.98), the radiation pattern of the N-element linear array will be:

X
N
ðn1Þd cos θ
jϕn j2π f c
RðθÞ ¼ gn e c : ð3:101Þ
n¼1

When d ¼ λc/2, the radiation pattern of N-element linear array reduces to:

X
N
jϕn jπ ðn1Þ cos θ
RðθÞjd¼λc =2 ¼ gn e : ð3:102Þ
n¼1

By varying the number of element in the linear array, the element spacing, amplitudes, and phases of the complex weights, it is
possible to obtain different shapes of the radiation pattern.
For the transmit array antenna of identical elements, the radiation pattern can be found by applying the pattern multiplica-
tion theorem [51], claiming that the array pattern is the product of the array element pattern and the array factor (AF). The AF
is a function dependent only on the geometry of the antenna array and excitation of the elements. Since the array factor is
independent of the antenna type, the isotropic radiators can be employed in the derivation of the AF. The field of an isotropic
radiator located in the origin is given by [51]:

I 0 ejkr
Eθ ¼ , ð3:103Þ
4π r

where r is the distance of the observation point P from the origin, k is the wave number (k ¼ 2π/λ), and I0 is the current
magnitude applied. Let us now observe the transmit N-element linear array shown in Fig. 3.55. The current magnitudes of all
array elements are the same I0, while the phase of the n-th element relative to the first element is denoted as ϕn so that the
current of the n-th element is given by In ¼ I0exp(jϕn). Based on Fig. 3.55, we conclude that the far field for the n-th array
element is given by:

I 0 e jϕn ejkrn I 0 e jϕn ejk½rðn1Þd cos θ

E θ,n ffi ¼ ; n ¼ 1, ⋯, N; ϕ1 ¼ 0: ð3:104Þ
4π r 4π r

The overall N-element array far field can be determined by the superposition principle:

X
N XN
I 0 e jϕn ejk½rðn1Þd cos θ I ejkr X
N
j½ϕn þðn1Þkd cos θ
Eθ ¼ Eθ,n ¼ ¼ 0 e : ð3:105Þ
n¼1 n¼1
4π r 4π r n¼1

One particular version of the linear array is the uniform linear array in which phases follow the arithmetic progression:

ϕ1 ¼ 0, ϕ2 ¼ ϕ, ϕ3 ¼ 2ϕ, ⋯, ϕn ¼ ðn  1Þϕ, ⋯, ϕN ¼ ðN  1Þϕ: ð3:106Þ

After the substitution in Eq. (3.105), we obtain:

I 0 ejkr X X
N N
j½ðn1Þϕþðn1Þkd cos θ I 0 ejkr jðn1Þðϕþkd cos θÞ
Eθ ¼ e ¼  e : ð3:107Þ
4π r n¼1 4π r
|fflfflfflffl{zfflfflfflffl} n¼1
arrayelementpattern |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
AF

PN n
By applying the geometric progression formula n¼1 q ¼ ð1  qn Þ=ð1  qÞ, we obtain the following expression for array
factor:
3.2 Modules/Components Relevant to Wireless Communications 199

Fig. 3.55 Transmit N-element

linear array

 
N ðϕþkd cos θÞ
X
N jN ðϕþkd cos θÞ
e 1 sin 2
jðn1Þðϕþkd cos θÞ jðN1Þϕ
AF ¼ e ¼ jðϕþkd cos θÞ
¼e ϕþkd cos θ : ð3:108Þ
n¼1
e 1 sin 2

The complex exponent term will disappear when the origin is moved to the center of the linear array so that the expression for
AF simplifies to:
 
N ðϕþkd cos θÞ
sin 2
AF ¼ ϕþkd cos θ : ð3:109Þ
sin 2

The maximum of AFmax ¼ N is obtained for ϕ + kd cos θ ¼ 0. There are N-1 lobes, one main, and N-2 side lobes. The main
lobe width is 4π/N, while the minor lobe width is 2π/N.
200 3 Components, Modules, and Subsystems

3.2.5 Automatic Gain Control (AGC)

To simplify detection process, in particular for multilevel schemes, such as M-ary QAM and M-ary APSK, the employment of
the automatic gain control (AGC) is necessary. Namely, for hard decision, it is highly desired to have accurate knowledge of
the average level of the signal, to draw properly the decision boundaries. Similarly, the AGC is required in PLL-based
synchronization.
Regarding the placement of the AGC module, it can be placed either in passband, at intermediate frequency (IF), or in
baseband, after the down-conversion to the baseband is performed. The second case is more attractive for high-speed
communications, including optical communications.
The most common location of the AGC module in passband as well as the configuration of the AGC is provided in
Fig. 3.56.
The received signal y(t) after ADC is resampled at rate 1/T to obtain:

yðnT Þ ¼ I y ðnT Þ cos ðΩc nÞ  Qy ðnT Þ sin ðΩc nÞ þ zðnT Þ, ð3:110Þ

where discrete frequency ΩIF is related to the IF frequency ωIF by ΩIF ¼ ωIFT. In Eq. (3.110), we use (nT) and Qy(nT) to
denote the in-phase and quadrature components of the IF passband signal, while z(nT) represent the sample of the narrowband
noise. The corresponding polar coordinate-based representation is given by:

yðnT Þ ¼ ay ðnT Þ cos ½ΩIF n þ ψ ðnT Þ þ zðnT Þ,

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

1 Qy ðnT Þ ð3:111Þ
ay ðnT Þ ¼ I y ðnT Þ þ Qy ðnT Þ, ψ ðnT Þ ¼ tan
2 2
:
I y ðnT Þ

The error signal e(n) used in AGC is defined as:

eðnÞ ¼ Lref  aðnÞay ðnÞ, ð3:112Þ

where Lref is the reference level. Before the accumulator, the error estimate can be properly scaled by the factor s. The
accumulator stores the previous value of the envelope estimate a(n  1), and the update formula is given by:

Fig. 3.56 The location of the AGC module in the passband (at IF) and the configuration of corresponding AGC module
3.2 Modules/Components Relevant to Wireless Communications 201

Fig. 3.57 The location of the AGC module in the baseband and the configuration of corresponding AGC module


aðnÞ ¼ aðn  1Þ þ s Lref  aðn  1Þay ððn  1ÞT Þ : ð3:113Þ

The AGC can also be placed in the baseband domain as shown in Fig. 3.57.
The envelope after the matched filters is given by:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
e
aðnT Þ ¼ I 2 ðnT Þ þ Q2 ðnT Þ: ð3:114Þ

The corresponding envelope update formula is very similar to Eq. (3.113) and given by:

aðnÞ ¼ aðn  1Þ þ s½Lref  aðn  1Þe

aððn  1ÞT Þ: ð3:115Þ

The key difference of this AGC scheme with respect to the corresponding scheme from Fig. 3.56 is that it operates in baseband
and as such it is possible to implement it at lower rates. Both envelope update expressions are nonlinear and might converge
slow when the signal level is low. To overcome this problem, we can modify the update formula to operate with dB-values of
the signals as follows:

aðnÞ½dB ¼ aðn  1Þ½dB þ sfLref ½dB  aðn  1Þ½dB  e

aððn  1ÞT Þ½dBg: ð3:116Þ

However, the conversion to linear scale is needed before the multiplication stage, after the accumulator. Clearly, the update
formula is now linear.
202 3 Components, Modules, and Subsystems

3.3 Concluding Remarks

In this chapter, basic components, modules, and subsystems relevant in both optical and wireless communications have been
described.
In the first half of the chapter, Sect. 3.1, key optical components, modules, and subsystems have been described. After
describing optical communication basic, Subsect. 3.1.1, focus has been changed to the optical transmitters (Subsect. 3.1.2).
The following classes of lasers have been introduced in Subsect. 3.1.2: Fabry-Perot, DFB, and DBR lasers. Regarding the
external modulators, both Mach-Zehnder modulator and electro-absorption modulator have been described, in the same
subsection. Regarding the external modulators for the multilevel modulation schemes, phase, I/Q modulator, and polar
modulators have been described. In Subsect. 3.1.3, devoted to optical receivers, after the photodetection principle, the p-i-n
photodiode, APD, and metal-semiconductor-metal photodetectors have been briefly described. In the same subsection, both
optical receiver high-impedance front end and trans-impedance front end have been described. In Subsect. 3.1.4, devoted to
optical fibers, both SMF and MMF have been briefly described. In Subsect. 3.1.5, devoted to optical amplifiers, after
describing optical amplification principles, the following classes of optical amplifiers have been described: SOA, Raman
amplifier, and EDFA. In Subsect. 3.1.6, related to optical processing components and modules, the following components/
modules have been described: optical couplers, optical filters, and WDM multiplexers/demultiplexers. After that, the
principles of coherent optical detection have been introduced, in Subsect. 3.1.7, followed by the description of optical hybrids
in Subsect. 3.1.8. In Subsect. 3.1.9, on coherent optical balanced detectors, both homodyne and heterodyne optical balanced
detectors for both binary and multilevel (QAM, PSK, APSK) modulations have been described.
In the second half of the chapter, Sect. 3.2, basic components and modules of relevance to modern wireless
communications have been described. Some of these modules are also relevant in optical communications as well. In Subsect.
3.2.1, the DSP basics have been introduced, including various discrete-time realizations. The direct digital synthesizer,
relevant in discrete-time wireless communication systems, has been described in Subsect. 3.2.2. In Subsect. 3.2.3, related to
the multirate DSP and resampling, various upsampling, downsampling, and resampling schemes/modules have been
described. In the same subsection, the polyphase decomposition of the system (transfer) function has been introduced and
employed in resampling. In Subsect. 3.2.4, devoted to the antenna arrays, the linear array antennas have been described.
Finally, in Subsect. 3.2.5, on automatic gain control, the AGC schemes operating in passband and baseband have been
described. The set of problems is provided in incoming section.

3.4 Problems

1. An electric field is applied along the z-crystallographic axis of a LiNbO3 modulator that operates at 1550 nm. Assuming
that electrodes are separated by 9 μm, r33 ¼ 30.9 pm/V, refractive mode index n ¼ 2.2, and the overlapping between optical
and electrical fields Γ ¼ 0.45, determine the Vπ-voltage for a 3.5-cm-long waveguide.

2. In this problem we study the LiNbO3 (LN) modulator in the symmetric Mach-Zehnder (MZ) configuration implemented
with the help of two 3 dB directional couplers, as shown in Fig. P2, for which kl ¼ (2 m + 1)π/4 (m-an integer).

Fig. P2 LN modulator in symmetric MZ configuration

3.4 Problems 203

Assuming that the voltage V is applied along the arm of length L + ΔL (longer arm). By employing the transfer-matrix
approach, derive corresponding expression for the power at the bar output port.
Hint. The electrical fields of output and input ports of directional coupler are connected by:




Eo1 cos ðklÞ j sin ðklÞ E i1
¼ ejβl
Eo2 j sin ðklÞ cos ðklÞ E i2

3. The binary sequence 00110001110101 is to be transmitted employing RZ pulses with duty cycle of 50%. For CS-RZ,
RZ-DPSK, duobinary, and AMI modulation formats, determine the amplitude and phase at the output of the transmitter.
4. The RZ-DPSK receiver is implemented using Mach-Zehnder delay interferometer (MZDI) and two 3 dB couplers:

Determine the output of a balanced detector, assuming that photodiode output is related to the input electrical field Ein by
Pout ¼ R|Ein|2, where R is the photodiode responsivity. Show that transmitted sequence in Problem 3 (corresponding to
RZ-DPSK) can be recovered using this kind of receiver.
5. This problem is concerned with the InGaAsP laser.
(a) Determine the threshold current for a 300-μm-long InGaAsP laser with an internal loss 30 cm1 and operating in a
single mode at 1550 nm with modal index 3.3 and group index 3.4. Assume that carrier life time is 2 ns and that the gain
varies as G ¼ GN(N  N0), with GN ¼ 8103 s1 and N0 ¼ 210.8 The bias current applied is twice above threshold.
Determine the emitted power from one faucet and the slope efficiency, and sketch the corresponding P-I curve. Assume
the internal efficiency of 95%.
(b) Assuming the laser linewidth of 10 MHz, the spontaneous emission factor nsp ¼ 2, and the average launched power of
2 mW, determine the frequency chirp expressed in Hz. Identify the adiabatic and dynamic terms.
(c) Determine the effective chirp coefficient from (b). Assume that a chirped Gaussian pulse (with initial chirp equal to the
effective chirp coefficient from (b)) is transmitted over DCF with dispersion coefficient of 90 ps/nm-km at 1550 nm.
Assuming the NRZ transmission at 10 Gb/s, determine the transmission distance at which the pulse width is minimum.
What is the maximum possible bit rate that can be supported?
6. Determine the maximum possible transmission distance of a 10Gb/s system operating at 1550 nm and employing the
external modulator. Assume transmission over standard SMF with dispersion coefficient of 16 ps/nm-km. What would be
the maximum transmission distance if the pre-chirping technique is employed? Determine the maximum transmission
distance if the pre-chirping technique is employed together with Gires-Tournois interferometer having a 2-cm-long Fabry-
Perot cavity and mirror reflectivity of 0.95.
7. A Raman amplifier is pumped bidirectionally using two pump laser diodes emitting 0.5 W each. Derive an expression for
the amplifier output power if the input power is known. Determine the output power when 2 μW signal is injected into a
4-km-long amplifier. Assume losses of 0.2 and 0.25 dB/km at the signal (1550 nm) and pump (1450 nm) wavelengths,
respectively. Aeff ¼ 50 μm2, and gR ¼ 6
1014m/W. What is the amplifier gain in dBs? Neglect the saturation effect.
8. Consider an optically pre-amplified pin direct detection receiver in which the receiver thermal noise and shot noise can be
neglected. Assuming the NRZ transmission, the optical filter bandwidth Bo, and electrical filter bandwidth Be, show that the
following relationship between Q-factor and optical signal-to-noise ratio (OSNR) is valid:

qffiffiffiffi
2 BBoe OSNR
Q¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi :
1 þ 1 þ 4OSNR
204 3 Components, Modules, and Subsystems

The OSNR is defined as the ratio between average signal power and total noise power in both polarizations. The amplifier
gain is G, received optical power is P, the spontaneous emission factor is nsp, and photodiode responsivity is R. Consider
now an amplifier chain of N identical optical amplifiers with noise figure (NF) of 7 dB. To reduce the ASE noise, every
amplifier stage has an optical filter of bandwidth Bo ¼ 4Rb, with Rb being the bit-rate set to 10Gb/s. Assuming NRZ
single-channel transmission with the launch power of 0 dBm, determine the maximum number of spans before bit-error
rate reaches 1012. The noise figure of pre-amplifier is 6 dB, and corresponding optical filter is of bandwidth 3Rb. The
transfer function of receiver electronics can be modeled as a low-pass filter of bandwidth Be ¼ 0.7Rb. The operating
wavelength is 1550 nm, and the photodiode responsivity R ¼ 1 A/W.
9. We would like to design an optical 100 Gb/s WDM system with M ¼ 10 channels operating at 10 Gb/s employing the
in-line optical amplifiers of noise figure 5dB. The total launched power is restricted by available laser sources and cannot
be larger than 12 dBm, and the fiber loss is α ¼ 0.2 dB/km. The predicted margins to achieve BER ¼ 109 (or Q ¼ 6)
allocated for chromatic dispersion, extinction ratio, polarization effects, fiber nonlinearities, component aging, and
system margin are 0.5, 0.5, 1.5, 1.5, 2, and 3 dB, respectively. WDM DEMUX can be modeled as a bank of filters,
each of them with a bandwidth Bo equal to 3B (B-the bit rate per channel). Assuming NRZ transmission, the electrical
filter bandwidth Be ¼ 0.65B, and amplifier spacing l ¼ 100 km, determine the required optical signal-to-noise ratio
(OSNR) (expressed in 0.1 nm optical spectral analyzer resolution bandwidth) to achieve BER of 109 for the central
channel located at 1552.524 nm. What is the maximum possible transmission distance? The saturation of optical
amplifiers due to ASE noise accumulation is not negligible.
10. Determine the receiver sensitivity of an APD receiver with a trans-impedance amplifier (TIA) front end of noise figure FA
and feedback resistor RF assuming (i) a finite extinction ratio rex, (ii) the thermal noise, (iii) the shot noise, (iv) the
intensity noise, (v) the reflection noise, and (vi) the mode partition noise. What is the optimum APD multiplication ratio?
What is the bandwidth of this receiver?
11. You are given to design the receiver for 40 Gb/s transmission that employs the Mach-Zehnder modulator with 14 dB of
extinction ratio. Providing that you have available either a p-i-n photodetector or an APD with ionization factor 0.5, both
of quantum efficiency 95%, a load resistance of 10 MΩ, and an electrical amplifier of noise figure 6dB, determine the
receiver sensitivities of both receiver versions required to achieve the BER of 109. Compare the signal-to-noise ratios of
corresponding receivers. If an EDFA of noise figure 5dB is available for how many dBs the receiver sensitivity can be
improved? Assume that the receiver bandwidth is 35 GHz and the bandwidth of optical filter that follows EDFA is
80 GHz.
12. You are given to design the system for 10 Gb/s RZ transmission that employs a very bad Mach-Zehnder modulator with
extinction ratio of only 10 dB. Providing that you have available a p-i-n photodetector of quantum efficiency 0.9, an
EDFA of gain 25 dB and noise figure 5dB to be used as pre-amplifier, and a load resistor RL for high-impedance
configuration, determine the receiver sensitivity required to achieve the BER of 1012. The bandwidth of an optical filter
that follows EDFA is 20 GHz. The junction capacitance is 1 pF, while the input capacitance is negligible. The operating
wavelength is 1550 nm.
13. You are supposed to design a transmission link operating at 160 Gb/s, employing RZ transmission of duty cycle 50%,
over standard SMF with dispersion coefficient of 16 ps/nm-km and attenuation coefficient 0.2 dB/km. The effective cross-
sectional area is 80 μm2, and the second-order refractive index coefficient is n2 ¼ 3.4
108 μm2/W. Gaussian pulses
with peak power of 0 dBm are used for RZ transmission. An ideal external modulator is supposed to be used. Operating
wavelength is 1550 nm.
(a) Determine the maximum possible transmission distance.
(b) What would be the maximum transmission distance if the SPM of fiber is used to compensate for chromatic
dispersion? Assume that 50% of the broadening is allowed.
(c) Determine the maximum transmission distance if the Gires-Tournois interferometer, having a 3-cm-long Fabry-Perot
cavity and mirror reflectivity of 0.98, is used. Again, assume that 50% of the broadening is allowed.
14. An NRZ single-channel system operating at 10 Gb/s employs the Mach-Zehnder modulator (MZM) of extinction ratio of
20 dB, whose driver amplifier for any input 1-bit generates a pulse of the shape:
3.4 Problems 205

8
>
> 1, 0 jt j ð1  αÞT=2
< h   i
1 π 1α
vRF ðt Þ ¼ V 0 1 þ cos jt j  T , ð1  αÞT=2 < jt j ð1 þ αÞT=2 ,
>
> 2 αT 2
:
0, jt j > ð1 þ αÞT=2

where V0[V] is the amplitude of the pulse, T is the bit duration, and α is the so-called roll-off parameter (0 α 1).
Determine (i) the width of the eye opening, (ii) height of the eye opening, (iii) noise margin, (iv) the rise time, and (v) the
decision threshold of the electrical field at MZM output. Sketch the eye diagram and determine the parameters (i)–(v) for
α ¼ 0.5 and α ¼ 1.
15. You are asked to design 50 Gb/s per wavelength link between Tucson and Phoenix by using DPSK with direct detection.
You have available two variable gain EDFAs with maximum gain of 18 dB each and noise figure of 5 dB each. To
compensate for fiber losses of SMF (with attenuation coefficient being 0.2 dB/km at 1550 nm), you can use the available
EDFAs and adjust the gains as needed. Assume that chromatic dispersion can be compensated by using appropriate
equalizer. The responsivity of available PIN photodiode is 1 A/W.
(a) Determine the equivalent noise figure of these two EDFAs in cascade.
(b) Determine the optical SNR at receiver side.
(c) Describe how you would determine the Q-factor.
(d) Describe how you would measure the bit-error rate (BER).
(e) Describe transmitter and receiver configurations for DPSK systems with direct detection.
(f) If the following sequence is to be transmitted 0110101001100, determine the sequence of phases on transmitter side
and show that original sequence can be recovered using DPSK demodulator from problem (e).
16. This problem is related to the coherent detection.
(a) Describe key difference between direct and coherent detections.
(b) Describe the principle of operation of optical hybrid and balanced detector.
(c) If the input to the coherent detector is given by Sk ¼ |Sk|exp.[j(ωTt + ϕPN,T + θk)] where the subscript k is used to
denote the kth symbol being transmitted, ωT denotes the transmitting laser frequency (in rad/s), ϕPN,T denotes the laser
phase noise of transmitter, θk is the phase of the kth symbol for M-ary PSK, and the local laser (oscillator) is denoted
by L ¼ |L|exp.[j(ωLOt + ϕPN,LO)] (ωLO denotes the local laser frequency and ϕPN,LO denotes the laser phase noise of
local laser), determine the output of balanced receiver. Assume that both photodetectors in balanced receiver are
identical with responsivity R. Assume further that corresponding hybrid is used to perform received and local laser
signals mixing. Specify the phases to be used in M-ary PSK.
(d) Describe the difference between polarization control and polarization scrambling.
(e) Describe the operational principle of phase diversity receiver.
17. Consider the reconfigurable M
M wavelength-routing switch shown in Fig. P17.

Fig. P17 Reconfigurable wavelength-routing switch

206 3 Components, Modules, and Subsystems

Each of the input fiber carries WDM signal with N channels, each with average power P, and extinction ratio r ¼ P1/P0
(P1-average power in one, P0-average power in zero-bit). Determine out-of-band crosstalk power penalty with respect to
an ideal system (with no crosstalk and ideal extinction ratio). For the extinction ratio of 10 dB, determine the maximum
tolerable crosstalk level so that the introduced power penalty is 1 dB. Assume that the crosstalk from adjacent channels is
dominant.
18. The analog system of the second order is shown in Fig. P18. Assuming R/L ¼ 8 and 1/(LC) ¼ 25, determine an equivalent
discrete-time system using the direct form realization.

Fig. P18 An analog system of the second order

19. A digital notch filter is required to remove an undesirable 60-Hz hum associated with a power supply in ECG recording
application. The sampling frequency is 500 samples/s. Design a second-order FIR notch filter. Chose the gain such that |H
(ω)| ¼ 1 for ω ¼ 0.
20. Using an all-pass system, determine the causal stable system with the same magnitude response as the system with system
function given by:

1 þ z1
H ðzÞ ¼    :
3z 1  12 z1 ð1  2z1 Þ
1 1
1

21. Design the discrete-time low-pass filter with passband edge frequency 10 GHz, the maximum allowable attenuation in
passband being 3 dB, the stopband edge frequency being 16.5 GHz, and the minimum attenuation in stopband being
15 dB. For filter design, use the impulse invariance method. Determine the parallel realization of this system. Is this
system stable?
22. The system function of a communication channel is given by:
H(z) ¼ (1  0.9e j0.4π z1)(1  0.9ej0.4π z1)(1  1.5e j0.6πz1)(1  1.5ej0.6πz1).Determine the system function of a
causal and stable compensating system so that the cascade interconnection of the two has a flat magnitude response.
Sketch the zero-pole plots of subsystems.
23. A continuous-time (CT) signal xa(t) with bandwidth B and its echo xa(t-τ) arrive simultaneously at a TV receiver. The
received CT signal

sa ðt Þ ¼ xa ðt Þ þ αxa ðt  τÞ, jαj < 1

is processed by the system shown in Fig. P23.

Fig. P23 The CT system under study

Specify the sampling frequency Fs and the filter system function H(z) so that the “ghost” pulse xa(t-τ) is removed from
the received signal.
References 207

24. We would like to design a non-recursive decimator for D ¼ 8 using the following factorization:
h    K1
i5
H ðzÞ ¼ 1 þ z1 1 þ z2 . . . 1 þ z2
Determine the implementation of decimator using filters with system function: Hk(z) ¼ (1 + z1).5 Show that each stage of
the decimator can be implemented more efficiently using the polyphase decomposition.
25. By using the bilinear transform, design a discrete low-pass filter of Butterworth type with passband being from 0 to
10 GHz, the attenuation at passband edge frequency being 3 dB, and minimum attenuation in stopband being 10 dB. The
stopband edge frequency is 15 GHz, and the sampling frequency is 40 GHz.

References
1. W. Shieh, I. Djordjevic, OFDM for Optical Communications (Elsevier/Academic Press, Amsterdam-Boston, Oct 2009)
2. I.B. Djordjevic, W. Ryan, B. Vasic, Coding for Optical Channels (Springer, New York/Dordrecht/Heidelberg/London, 2010)
3. M. Cvijetic, I.B. Djordjevic, Advanced Optical Communication Systems and Networks (Artech House, Norwood, 2013)
4. R. Ramaswami, K. Sivarajan, Optical Networks: A Practical Perspective (Morgan Kaufman, 2nd, 2002)
5. G.P. Agrawal, Fiber-Optic Communication Systems, 3rd edn. (Wiley, Hoboken, 2002)
6. G.P. Agrawal, Lightwave Technology: Components and Devices (Wiley, Hoboken, 2004)
7. G.P. Agrawal, Lightwave Technology: Telecommunication Systems (Wiley, Hoboken, 2005)
8. G.P. Agrawal, Nonlinear Fiber Optics, 4th edn. (Academic Press, Oxford, 2007)
9. G. Keiser, Optical Fiber Communications, 3rd edn. (McGraw-Hill, 2000)
10. L. Kazovsky, S. Benedetto, A. Willner, Optical Fiber Communication Systems (Artech House, Boston, 1996)
11. J.C. Palais, Fiber Optic Communications, 5th edn. (Pearson Prentice Hall, Upper Saddle River, 2005)
12. B. Ramamurthy, Switches, wavelength routers, and wavelength converters, in Optical WDM Networks: Principles and Practice, ed. by K. M.
Sivalingam, S. Subramaniam, (Kluwer Academic Publishers, Boston, 2000)
13. I. B. Djordjevic, ECE 430/530: Optical Communication Systems (Lecture Notes), University of Arizona, 2010/2011
14. R. Kostuk, ECE 430/530: Optical Communication Systems (Lecture Notes), University of Arizona, 2002/2004
15. I. B. Djordjevic, ECE 632: Advanced Optical Communication Systems (Lecture Notes), University of Arizona, 2007
16. A. Yariv, Optical Electronics in Modern Communications (Oxford University Press, 1997)
17. M. Amersfoort, Arrayed Waveguide Grating, Application note A1998003, Concept to volume, 1998
18. M. Staif, A. Mecozzi, J. Nagel, Mean square magnitude of all orders PMD and the relation with the bandwidth of principal states. IEEE Photon.
Tehcnol. Lett. 12, 53–55 (2000)
19. H. Kogelnik, L.E. Nelson, R.M. Jobson, Polarization mode dispersion, in Optical Fiber Telecommunications, ed. by I. P. Kaminow, T. Li,
(Academic Press, San Diego, 2002)
20. G.J. Foschini, C.D. Pole, Statistical theory of polarization mode dispersion in single-mode fibers. IEEE/OSA J. Lightw. Technol. LT-9,
1439–1456 (1991)
21. G. Bendelli et al., Optical performance monitoring techniques, in Proc. ECOC 2000, Munich, vol. 4, pp. 213–216 (2000)
22. I.B. Djordjevic, B. Vasic, An advanced direct detection receiver model. J. Opt. Commun. 25(1), 6–9 (Feb 2004)
23. P.J. Winzer, M. Pfennigbauer, M.M. Strasser, W.R. Leeb, Optimum filter bandwidths for optically preamplified NRZ receivers. J. Lightw.
Technol. 19, 1263–1272 (Sept 2001)
24. R. Hui et al., Technical Report: Advanced Optical Modulation Formats and Their Comparison in Fiber-Optic Systems., University of Kansas
(Jan 2004)
25. V. Wongpaibool, Improvement of fiber optic system performance by synchronous phase modulation and filtering at the transmitter. Virginia
Polytechnic Institute and State University, Ph.D. Dissertation (Jan 2003)
26. G. Jacobsen, Noise in Digital Optical Transmission Systems (Artech House, Boston, 1994)
27. I. B. Djordjevic, Analysis and optimization of coherent optical systems with the phase-locked loop. University of Nis, Serbia, Ph.D. Dissertation
(1999)
28. P.W. Hooijmans, Coherent Optical System Design (Wiley, Chichester, 1994)
29. S.J. Savory, Digital filters for coherent optical receivers. Opt. Express 16, 804–817 (2008)
30. I.B. Djordjevic, L. Xu, T. Wang, PMD compensation in multilevel coded-modulation schemes with coherent detection using BLAST algorithm
and iterative polarization cancellation. Opt. Express 16(19), 14845–14852 (15 Sept 2008)
31. I.B. Djordjevic, L. Xu, T. Wang, PMD compensation in coded-modulation schemes with coherent detection using Alamouti-type polarization-
time coding. Opt. Express 16(18), 14163–14172 (9 Jan 2008)
32. I.B. Djordjevic, M. Arabaci, L. Minkov, Next generation FEC for high-capacity communication in optical transport networks. IEEE/OSA
J. Lightw. Technol. 27(16), 3518–3530 (15 Aug 2009) Invited paper
33. I.B. Djordjevic, On advanced FEC and coded modulation for ultra-high-speed optical transmission. IEEE Commun. Surv. Tutor. 18(3),
1920–1951, third quarter 2016 (19 Aug 2016). https://doi.org/10.1109/COMST.2016.2536726
34. E. Ip, A.P.T. Lau, D.J.F. Barros, J.M. Kahn, Coherent detection in optical fiber systems. Opt. Express 16, 753–791 (2008)
35. K. Kikuchi, Fundamentals of coherent optical fiber communications. J. Lightw. Technol. 34(1), 157–179 (2016)
36. A. Goldsmith, Wireless Communications (Cambridge University Press, Cambridge, 2005)
37. D. Tse, P. Viswanath, Fundamentals of Wireless Communication (Cambridge University Press, Cambridge, 2005)
208 3 Components, Modules, and Subsystems

38. S. Benedetto, E. Biglieri, Principles of Digital Transmission with Wireless Applications (Kluwer Academic/Plenum Publishers, New York,
1999)
39. E. Biglieri, Coding for Wireless Channels (Springer, New York, 2005)
40. T.M. Duman, A. Ghrayeb, Coding for MIMO Communication Systems (Wiley, Chichester, 2007)
41. J.R. Hampton, Introduction to MIMO Communications (Cambridge University Press, Cambridge, 2014)
42. G.L. Stüber, Principles of Mobile Communication, 3rd edn. (Springer, New York, 2012)
43. W.C.Y. Lee, Mobile Communications Engineering (McGraw-Hill, 1982)
44. J.G. Proakis, Digital Communications (McGraw-Hill, Boston, 2001)
45. J.G. Proakis, D.G. Manolakis, Digital Signal Processing: Principles, Algorithms, and Applications, 4th edn. (Prentice-Hall, Upper Saddle River,
2007)
46. S. Haykin, B. Van Veen, Signals and Systems, 2nd edn. (Wiley, 2003)
47. A.V. Oppenheim, R.W. Schafer, Discrete-Time Signal Processing, 3rd edn. (Pearson Higher Ed, 2010)
48. I. B. Djordjevic, ECE 429/529 Digital Signal Processing (Lecture Notes), University of Arizona, 2008
49. I. B. Djordjevic, ECE 340 Engineering Systems Analysis (Lecture Notes), University of Arizona (2012)
50. S. Verdú, Multiuser Detection (Cambridge University Press, New York, 1998)
51. C.A. Balanis, Antenna Theory: Analysis and Design, 4th edn. (Wiley, Hoboken, 2016)
 Capacities of Wireless and Optical Channels
4

Abstract
This chapter is devoted to wireless channel and optical channel capacities. The chapter starts with definitions of mutual
information and channel capacity, followed by the information capacity theorem. We discuss the channel capacity of
discrete memoryless channels, continuous channels, and channels with memory. Regarding the wireless channels, we
describe how to calculate the channel capacity of flat-fading and frequency-selective channels. We also discuss different
optimum and suboptimum strategies to achieve channel capacity including the water-filling method, multiplexed coding
and decoding, channel inversion, and truncated channel inversion. We also study different strategies for channel capacity
calculation depending on what is known about the channel state information. Further, we explain how to model the channel
with memory and describe the McMillan-Khinchin model for channel capacity evaluation. We then describe how the
forward recursion of the BCJR algorithm can be used to evaluate the information capacity of channels with memory. The
next topic is related to the evaluation of the information capacity of fiber-optic communication channels with coherent
optical detection. Finally, we describe how to evaluate the capacity of hybrid free-space optical (FSO)-RF channels, in
which FSO and RF subsystems cooperate to compensate for shortcoming of each other. The set of problems is provided
after the concluding remarks.

4.1 Mutual Information, Channel Capacity, and Information Capacity Theorem

4.1.1 Mutual Information and Information Capacity

Figure 4.1 shows an example of a discrete memoryless channel (DMC), which is characterized by channel (transition)
probabilities. If X ¼ {x0, x1, . . ., xI  1} and Y ¼ {y0, y1, . . ., yJ1} denote the channel input alphabet and the channel output
alphabet, respectively, the channel is completely characterized by the following set of transition probabilities:




p y j jxi ¼ P Y ¼ y j jX ¼ xi , 0  p y j jxi  1, ð4:1Þ

where i2{0, 1, . . ., I  1}, j2{0, 1, . . ., J  1}, while I and J denote the sizes of input and output alphabets, respectively. The
transition probability p(yj|xi) represents the conditional probability that Y ¼ yj for given input X ¼ xi.
One of the most important characteristics of the transmission channel is the information capacity, which is obtained by
maximization of mutual information I(X,Y) over all possible input distributions:

C ¼ max I ðX, Y Þ, I ðX, Y Þ ¼ H ðXÞ  H ðXjY Þ, ð4:2Þ

fpðxi Þg

where H(U) ¼ hlog2P(U)i denotes the entropy of a random variable U. The mutual information can be determined as

# The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 209
I. B. Djordjevic, Advanced Optical and Wireless Communications Systems, https://doi.org/10.1007/978-3-030-98491-5_4
210 4 Capacities of Wireless and Optical Channels

Fig. 4.1 Discrete memoryless p(y1|x1)

channel (DMC) x1 y1
p(y2|x1)
x2 y2
p(y1|xi)
p(yJ|x1) p(y2|xi)

…
xi yj
p(yj|xi)
The input alphabet
The output alphabet

…
p(y1|xI)
p(yJ|xi)
X={x1,…,xI} Y={y1,…,yJ}
xI yJ
p(yJ|xI)

I ðX; Y Þ ¼ H ðX Þ  H ðXjY Þ " #

 
PM 1 PN
P M
 1 ð4:3Þ
¼ pðxi Þ log 2  p yj p xi jy j log 2
 :
i¼1 pðxi Þ j¼1 i¼1 p xi jy j

In the above equation, H(X) represents the uncertainty about the channel input before observing the channel output, also
known as entropy, while H(X|Y) denotes the conditional entropy or the amount of uncertainty remaining about the channel
input after the channel output has been received. (The log function from the above equation relates to the base 2, and it will be
like that throughout this chapter.) Therefore, the mutual information represents the amount of information (per symbol) that is
conveyed by the channel, which represents the uncertainty about the channel input that is resolved by observing the channel
output. The mutual information can be interpreted by means of Venn diagram [1–9] shown in Fig. 4.2(a). The left and right
circles represent the entropy of the channel input and channel output, respectively, while the mutual information is obtained as
intersection area of these two circles. Another interpretation is illustrated in Fig. 4.2(b) [8]. The mutual information, i.e., the
information conveyed by the channel, is obtained as the output information minus information lost in the channel.
Since for M-ary input and M-ary output symmetric channel (MSC), we have that p(yj|xi) ¼ Ps/(M  1) and p(yj|xj) ¼ 1  Ps,
where Ps is the symbol error probability, the channel capacity, in bits/symbol, can be found as
 
Ps
C ¼ log 2 M þ ð1  Ps Þ log 2 ð1  Ps Þ þ Ps log 2 : ð4:4Þ
M1

The channel capacity represents an important bound on data rates achievable by any modulation and coding schemes. It can
also be used in comparison of different coded modulation schemes in terms of their distance to the maximum channel capacity
curve.
Now we have built enough knowledge to formulate a very important theorem, the channel coding theorem [1–9], which
can be formulated as follows. Let a discrete memoryless source with an alphabet S have the entropy H(S) and emit the symbols
every Ts seconds. Let a DMC have capacity C and be used once in Tc seconds. Then, if

H ðSÞ=T s  C=T c ,

there exists a coding scheme for which the source output can be transmitted over the channel and reconstructed with an
arbitrary small probability of error. The parameter H(S)/Ts is related to the average information rate, while the parameter C/Tc
is related to the channel capacity per unit time.
For binary symmetric channel (N ¼ M ¼ 2), the inequality is reduced down to R  C, where R is the code rate. Since the
proof of this theorem can be found in any textbook on information theory, such as [5–8, 14], the proof of this theorem will be
omitted.
4.1 Mutual Information, Channel Capacity, and Information Capacity Theorem 211

Fig. 4.2 Interpretation of the The uncertainty about the

mutual information by using (a) channel input before
Venn diagrams and (b) the H(X,Y)
approach due to Ingels observing the channel output
H(X)

H(X|Y) I(X;Y) H(Y|X)

H(Y)
The amount of uncertainty remaining about
the channel input after the channel output
has been received

Uncertainty about the channel input that is resolved by observing

the channel output [the amount of information (per symbol) conveyed by the channel]
(a)
Information lost
in the channel H(X|Y)
Mutual information
H(X)− H(X|Y)=I(X; Y)

Input Output

information The information

Channel H(Y)
H(X)

Unwanted information
introduced by the channel Mutual
H(Y|X) information
H(Y)− H(Y|X)=I(Y; X)
(b)

4.1.2 Capacity of Continuous Channels

In this section, we will discuss the channel capacity of continuous channels. Let X ¼ [X1, X2, . . ., Xn] denote an n-dimensional
multivariate, with a PDF p1(x1, x2, . . ., xn), representing the channel input. The corresponding differential entropy is defined
by [2, 3, 5, 6].

ð
1 ð
1

hðX 1 , X 2 , . . . , X n Þ ¼  ⋯ p1 ðx1 , x2 , . . . , xn Þlog p1 ðx1 , x2 , . . . , xn Þdx1 dx2 . . . dxn ¼ hlog p1 ðx1 , x2 , . . . , xn Þi,
ð4:5Þ
1 1
|fflfflfflfflffl{zfflfflfflfflffl}
n

where we use hi to denote the expectation operator. In order to simplify explanations, we will use the compact form of
Eq. (4.5), namely, h(X) ¼ h log p1(X)i, which was introduced in [6]. In a similar fashion, the channel output can be
represented as m-dimensional random variable Y ¼ [Y1, Y2, . . ., Ym] with a PDF p2(y1, y2, . . ., ym), while the corresponding
differential entropy is defined by
212 4 Capacities of Wireless and Optical Channels

hðY 1 , Y 2 , . . . , Y m Þ
ð
1 ð
1

¼  ⋯ p2 ðy1 , y2 , . . . , ym Þlog p1 ðy1 , y2 , . . . , ym Þdy1 dy2 . . . dym

ð4:6Þ
1 1
|fflfflfflfflffl{zfflfflfflfflffl}
m
¼ hlog p2 ðy1 , y2 , . . . , ym Þi:

In compact form, the differential entropy of output can be written as h(Y) ¼ h log p1(Y)i.

Example Let an n-dimensional multivariate X ¼ [X1, X2, . . ., Xn] with a PDF p1(x1, x2, . . ., xn) be applied to the nonlinear
channel with the following nonlinear characteristic: Y ¼ g(X), where Y ¼ [Y1, Y2, . . ., Yn] represents the channel output with
PDF p2(y1, y2, . . ., yn). Since the corresponding PDFs are related by the Jacobian symbol as follows, p2 ðy1 , ⋯, yn Þ ¼
 
⋯, X n
p1 ðx1 , ⋯, xn Þ J YX11,, ⋯, Y m , the output entropy can be determined as



X 1 , ⋯, X n
hðY 1 , ⋯, Y m Þ ffi hðX 1 , ⋯, X n Þ 
log J :
Y 1 , ⋯, Y m

To account for the channel distortions and additive noise influence, we can observe the corresponding conditional and
joint PDFs:

Pðy1 < Y 1 < y1 þ dy1 , . . . , ym < Y m < ym þ dym jX 1 ¼ x1 , . . . , X n ¼ xn Þ ¼ pðeyjexÞdey

ð4:7Þ
Pðy1 < Y 1 < y1 þ dy1 , . . . , xn < X n < xn þ dxn Þ ¼ pðex, eyÞdexdey:

The mutual information (also known as information rate) can be written in compact form as follows [6]:

pðX, Y Þ
I ðX; Y Þ ¼ log : ð4:8Þ
pðXÞPðY Þ

Notice that various differential entropies h(X), h(Y), h(Y|X) do not have direct interpretation as far as the information
processed in the channel is concerned, as compared to their discrete counterparts, from the previous subsection. Some
authors, such as Gallager in [7], prefer to define the mutual information directly by Eq. (4.8), without considering the
differential entropies at all. The mutual information, however, has the theoretical meaning and represents the average
information processed in the channel (or the amount of information conveyed by the channel). The mutual information has
the following important properties [5–8]: (i) it is symmetric, I(X;Y ) ¼ I(Y;X); (ii) it is nonnegative; (iii) it is finite; (iv) it is
invariant under linear transformation; (v) it can be expressed in terms of the differential entropy of channel output by I(X;
Y ) ¼ h(Y)  h(Y|X); and (vi) it is related to the channel input differential entropy by I(X;Y ) ¼ h(X)  h(X|Y ).
The information capacity can be obtained by maximization of Eq. (4.8) under all possible input distributions, which is

C ¼ max I ðX; Y Þ: ð4:9Þ

Let us now determine the mutual information of two random vectors, X ¼ [X1, X2, . . ., Xn] and Y ¼ [Y1, Y2, . . ., Ym], which
are normally distributed. Let Z ¼ [X;Y] be the random vector describing the joint behavior. Without loss of generality, we
further assume that X k ¼ 0 8k and Y k ¼ 0 8k. The corresponding PDF for X, Y, and Z are, respectively, given as [6].

1


p1 ð xÞ ¼ n=2 1=2
exp 0:5 A1 x, x ,
ð2π Þ ðdetAÞ
ð ð4:10Þ
 
A ¼ aij , aij ¼ xi x j p1 ðxÞdx,
4.1 Mutual Information, Channel Capacity, and Information Capacity Theorem 213

1


p2 ð y Þ ¼ n=2 1=2
exp 0:5 B1 y, y ,
ð2π Þ ðdetBÞ
ð ð4:11Þ
 
B ¼ bij , bij ¼ yi y j p2 ðyÞdy,

1


p 3 ðzÞ ¼ ðnþmÞ=2 1=2
exp 0:5 C1 z, z ,
ð2π Þ ðdetCÞ
ð ð4:12Þ
 
C ¼ cij , cij ¼ zi z j p3 ðzÞdz,

where (,) denotes the dot product of two vectors. By substitution of Eqs. (4.10, 4.11, and 4.12) into Eq. (4.8), we obtain [6]

1 detAdetB
I ðX; Y Þ ¼ log : ð4:13Þ
2 detC

The mutual information between two Gaussian random vectors can also be expressed in terms of their correlation
coefficients [6]:

1 


I ðX; Y Þ ¼  log 1  ρ21 . . . 1  ρ2l , l ¼ min ðm, nÞ, ð4:14Þ
2

where ρj is the correlation coefficient between Xj and Yj.

In order to obtain the information capacity for additive Gaussian noise, we make the following assumptions: (i) the input X,
output Y, and noise Z are n-dimensional random variables; (ii) X k ¼ 0, X 2k ¼ σ 2xk 8k and Z k ¼ 0, Z 2k ¼ σ 2zk 8k; and (iii) the
noise is additive, Y ¼ X + Z. Since we have that
" #
n1
ezk =2σ zk
2 2
px ðyjxÞ ¼ px ðx þ zjxÞ ¼ ∏ 1=2
¼ pðzÞ, ð4:15Þ
k¼1 ð2π Þ σ zk

the conditional differential entropy can be obtained as

ð
1

H ðYjX Þ ¼ H ðZÞ ¼  pðzÞ log pðzÞdz: ð4:16Þ

1

The mutual information is then

1X
n
I ðX; Y Þ ¼ hðY Þ  hðYjX Þ ¼ hðY Þ  hðZÞ ¼ hðY Þ  log 2πeσ 2zk : ð4:17Þ
2 k¼1

The information capacity, expressed in bits per channel use, is therefore obtained by maximizing h(Y). Because the
distribution maximizing the differential entropy is Gaussian, the information capacity is obtained as
!
1X 1X 1X
n n n σ 2yk
C ðX; Y Þ ¼ log 2πeσ 2yk  log 2πeσ 2zk ¼ log
2 k¼1 2 k¼1 2 k¼1 σ 2zk
!
 ð4:18Þ
1X 1X
n n
σ 2xk þ σ 2zk σ 2x
¼ log ¼ log 1 þ :
2 k¼1 σ 2zk 2 k¼1 σ 2z

For σ 2xk ¼ σ 2x , σ 2zk ¼ σ 2z , we obtain the following expression for information capacity:
214 4 Capacities of Wireless and Optical Channels


n σ2
C ðX; Y Þ ¼ log 1 þ x2 , ð4:19Þ
2 σz

where σ 2x =σ 2z presents the signal-to-noise ratio (SNR). The expression above represents the maximum amount of information
that can be transmitted per symbol.
From a practical point of view, it is important to determine the amount of information conveyed by the channel per second,
which is the information capacity per unit time, also known as the channel capacity. For bandwidth-limited channels and
Nyquist signaling employed, there will be 2 W samples per second (W is the channel bandwidth), and the corresponding
channel capacity becomes


P
C ¼ W log 1 þ ½bits=s, ð4:20Þ
NoW

where P is the average transmitted power and N0/2 is the noise power spectral density (PSD). Equation (4.20) represents the
well-known information capacity theorem, commonly referred to as the Shannon third theorem [17].
Since the Gaussian source has the maximum entropy clearly, it will maximize the mutual information. Therefore, the
equation above can be derived as follows. Let the n-dimensional multivariate X ¼ [X1, . . ., Xn] represent the Gaussian channel
input with samples generated from zero-mean Gaussian distribution with variance σ 2x : Let the n-dimensional multivariate
Y ¼ [Y1, . . ., Yn] represent the Gaussian channel output, with samples spaced 1/2 W apart. The channel is additive with noise
samples generated from zero-mean Gaussian distribution with variance σ 2z : Let the PDFs of input and output be denoted by
p1(x) and p2(y), respectively. Finally, let the joint PDF of input and output of channel be denoted by p(x, y). The maximum
mutual information can be calculated from
ZZ
pðx, yÞ
I ðX; Y Þ ¼ pðx, yÞ log dxdy: ð4:21Þ
pðxÞPðyÞ

By following the similar procedure to that used in Eqs. (4.15, 4.16, 4.17, 4.18, and 4.19), we obtain the following expression
for Gaussian channel capacity:
 
P
C ¼ W log 1 þ , N ¼ N o W, ð4:22Þ
N

where P is the average signal power and N is the average noise power.

4.2 Capacity of Flat-Fading and Frequency-Selective Wireless Fading Channels

In this section, we study both flat-fading and frequency-selective fading channels [10–16] and describe how to calculate the
corresponding channel capacities [17] as well as approaches to achieve these channel capacities.

4.2.1 Flat-Fading Channel Capacity

The typical wireless communication system together with an equivalent channel model is shown in Fig. 4.3. The channel
power gain g[i], where i is the discrete-time instance, is related to the channel coefficient h[i] by g[i] ¼ |h[i]|2 and follows a
given probability density function (PDF) f(g). For instance, as shown in Chap. 2, for Rayleigh fading, the PDF f(g) is
exponential. The channel gain is commonly referred to as channel side information or channel state information (CSI). The
samples of channel gain g[i] change at each time instance i and could be generated from an independent identically distributed
(i.i.d.) process or could be generated from a correlated source. In a block-fading channel, considered in this section, the
channel gain g[i] is constant over some block length T, and once this block length is over, a new realization of channel gain g
[i] is generated based on PDF f(g).
4.2 Capacity of Flat-Fading and Frequency-Selective Wireless Fading Channels 215

Fig. 4.3 Typical wireless Transmit

communication system together antenna Receive
with an equivalent fading channel h[i] antenna
model Power
amplifier LNA
Sequence y[i] Est.
Mapper + I/Q
Encoder Decoder
I/Q modulator demodulator seq.
x[i]
y[i]
Channel g[i]=|h[i]|2
h[i] z[i] estimator

x[i] y[i]=h[i]x[i]+z[i]

Equivalent fading channel model

The channel capacity is strongly dependent on what is known about CSI at the transmitter and/or receiver sides. In this
section, three scenarios of interest will be studied: (i) channel distribution information (CDI), (ii) CSI is available at the
receiver side (CSIR), and (iii) the CSI is available both at the transmitter and receiver sides (full CSI). In CDI scenario, the
PDF of g is known to both transmitter and receiver sides. Determination of the capacity-achieving input distribution and
corresponding capacity of fading channels under CDI scenario for any distribution f(g), in closed form, is still an open
problem, except from specific fading channel models such as i.i.d. Rayleigh fading channels and FSMCs. Other two scenarios,
CSIR and full CSI, are discussed with more details below.

4.2.1.1 Channel Side Information at Receiver

Two channel capacity definitions in CSIR scenario are of interest in the system design: (i) Shannon capacity, also known as
the ergodic capacity, which represents the maximum data rate that can be sent over the channel with symbol error probability
Ps tending to zero; and (ii) capacity with outage, which represents the maximum data rate that can be transmitted over a
channel with some outage probability Poutage, defined as the probability that SNR falls below a given SNR threshold value ρtsh
corresponding to the maximum tolerable symbol error probability Ps. Clearly, in the outage, the transmission cannot be
completed with negligible symbol error probability. The capacity with outage corresponds to the highest data rate that can be
transmitted over the fading channel reliably except when the fading channel is in deep fade. To determine either the average
symbol error probability or outage probability in the presence of fading, the distribution of SNR, f(ρ), is needed. The average
symbol error probability is then defined as
ð1
P s ¼ hP s ðγ Þi ¼ Ps ðρÞf ðρÞdρ, ð4:23Þ
0

where Ps(ρ) is the SNR-dependent expression for symbol error probability, for a given modulation format. The outage
probability is defined as
ð ρtsh
Poutage ¼ Pðρ  ρtsh Þ ¼ f ðρÞdρ, ð4:24Þ
0

where ρtsh is the threshold SNR corresponding to the maximum tolerable symbol error probability.

Shannon (Ergodic) Capacity

Under the assumptions that the channel gain g[i] originates from the flat fading, average power gain is equal to 1, and receiver
knows CSI (CSIR), we study the channel capacity calculation. When fading is fast, with a certain decoding delay requirement,
we transmit the signal over the time duration that contains N coherence time periods Tc, namely, NTc, wherein N > > 1. In this
scenario, since the transmission is sufficiently long and the fading is fast, all possible channel gain realizations g[i] come to the
“picture” and the channel capacity can be calculated by averaging out channel capacities for different realizations. For nth
coherence time period, we assume that channel gain is constant equal to g[n], so the received SNR can be estimated as g½nρ.
The corresponding channel capacity, related to nth gain realization, will be C½n ¼ W log 2 ð1 þ g½nρÞ, where W is the
channel bandwidth. The Shannon (ergodic) channel capacity can be obtained by averaging over N coherence time periods as
follows:
216 4 Capacities of Wireless and Optical Channels

1 X
N
CN ¼ W log 2 ð1 þ g½nρÞ: ð4:25aÞ
N n¼1

Now by letting N ! 1, we obtain

ð
1

C ¼ hW log 2 ð1 þ ρÞi ¼ W log 2 ð1 þ ρÞf ðρÞdρ: ð4:25bÞ

0

By applying Jensen’s inequality [5], the following upper limit is obtained for C:

C ¼ hW log 2 ð1 þ ρÞi  W log 2 ð1 þ hρiÞ ¼ W log 2 ð1 þ ρÞ: ð4:26Þ

Channel Capacity with Outage

Capacity with outage is related to the slowly varying fading channels, where the instantaneous SNR ρ is constant for the
duration of transmission of a large number of symbols, which is also known as a transmission burst, and changes to a new
SNR value at the end of the burst period according to the given fading distribution. When the received SNR ρ is constant for
the duration of the transmission burst, then the maximum possible transmission rate will be W log2(1 + ρ), with symbol error
probability tending to zero. Because the transmitter does not know the exact value of SNR ρ, it will be forced to transmit at the
fixed data rate corresponding to the minimum tolerable SNR ρmin. The maximum data rate corresponding to SNR ρmin will be
C ¼ W log2(1 + ρmin). As indicated earlier, the probability of outage is defined as Poutage ¼ P(ρ < ρmin). The average data rate,
reliably transmitted over many transmission bursts, can be defined as


C outage ¼ 1  Poutage W log 2 ð1 þ ρmin Þ, ð4:27Þ

given that data is reliably transmitted in 1  Poutage transmission bursts. In other words, this is the transmission rate that can be
supported in 100(1  Poutage)% of the channel realizations.

4.2.1.2 Full CSI

In full CSI scenario, both the transmitter and the receiver have CSI available, and the transmitter can adapt its transmission
strategy based on this CSI. Depending on the type of fading channel, fast- or slow-varying, we can define either the Shannon
capacity or capacity with outage.

Shannon Capacity
Let S denote the set of discrete memoryless wireless channels and let p(s) denote the probability that the channel is in state s 2
S. In other words, p(s) is the fraction of the time for which the channel was in state s. The channel capacity, as defined by
Wolfowitz [14], of the time-varying channel, in full CSI scenario (when both the transmitter and the receiver have the CSI
available), is obtained by averaging the channel capacity when the channel is in state s, denoted as Cs, as follows:
X
C¼ s2S
Cs pðsÞ: ð4:28Þ

By letting the cardinality of set S to tend to infinity, |S| ! 1, the summation becomes integration and p(s) becomes the PDF of
ρ, so that we can write

ð
1 ð
1

C¼ C ðρÞf ðρÞdγ ¼ W log 2 ð1 þ ρÞf ðρÞdρ, ð4:29Þ

0 0

which is the same as in the CSIR case. Therefore, the channel capacity does not increase in full CSI case unless some form of
the power adaptation is employed, which is illustrated in Fig. 4.4.
4.2 Capacity of Flat-Fading and Frequency-Selective Wireless Fading Channels 217

Fig. 4.4 The wireless h[i] z[i]

communication system
employing the power adaptation Sequence x[i] y[i] Est.
Mapper + Power
Encoder Decoder
I/Q modulator control, P[i] seq.
Channel
estimator g[i]
g[i]=|h[i]|2

Fig. 4.5 The illustration of the

P( U ) / P
optimum power allocation 1/Utsh
strategy, known as the water-
filling method

P( U ) / P

1/U
U
Utsh

The fading channel capacity, with power adaptation, is an optimization problem:

ð
1 !
PðρÞρ Pg
C¼ max W log 2 1þ f ðρÞdρ, ρ¼ : ð4:30Þ
ð
1 P N 0W
0
PðρÞ: PðρÞf ðρÞdρ ¼ P
0

The optimum solution can be obtained by employing the Lagrangian method, wherein the Lagrangian L is defined as

ð
1
 ð
1
PðρÞρ
L ðPðρÞÞ ¼ W log 2 1 þ f ðρÞdρ  μ PðρÞf ðρÞdρ, ð4:31Þ
P
0 0

which is subject to the following constraint:

ð
1

PðρÞf ðρÞdρ  P: ð4:32Þ

0

By finding the first derivative of Lagrangian with respect to P(ρ) and setting it to zero, we obtain
 
∂L ðPðρÞÞ W= ln 2 ρ
¼  μ f ðρÞ ¼ 0: ð4:33Þ
∂PðρÞ 1 þ ρPðρÞ=P P

By solving for P(ρ) from (4.33), we obtain the optimum power allocation policy, also known as the water-filling method,
which is illustrated in Fig. 4.5, as follows:

PðρÞ 1=ρtsh  1=ρ, ρ  ρtsh
¼ : ð4:34Þ
P 0, ρ < ρtsh

The amount of power allocated on a transmitter side for a given SNR ρ is equal to 1/ρtsh  1/ ρ. In other words, the amount of
allocated power is proportional to the “water” filled between the bottom of the bowl (1/ρ) and the constant water line (1/ρtsh).
218 4 Capacities of Wireless and Optical Channels

The intuition behind the water filling is to take the advantage of realizations when the channel conditions are good. Namely,
when the channel conditions are good and SNR ρ is large, we allocate higher power and transmit at higher data rate over the
channel. On the other hand, when the channel conditions deteriorate and ρ becomes small, we allocate less power and reduce
the data rate. Once the instantaneous channel SNR ρ falls below the threshold (cutoff) value, we do not transmit at all. When
the channel conditions are so bad, the reliable transmission is not possible and the average symbol error rate would be
dominated by such transmissions.
After substituting Eq. (4.34) into (4.30), we obtain the following expression for the channel capacity corresponding to the
water-filling method:

ð
1

ρ
C¼ W log 2 f ðρÞdρ: ð4:35Þ
ρtsh
ρtsh

After substituting Eq. (4.34) into (4.32), we obtain the following equation that needs to be solved numerically to determine
the threshold SNR, ρtsh:

ð
1

1 1
 f ðρÞdρ ¼ 1: ð4:36Þ
ρtsh ρ
ρtsh

Let us now consider the optimum allocation policy in discrete-time domain. The channel model is similar to that from
Fig. 4.3:

y½n ¼ h½nx½n þ z½n, jh½nj2 ¼ g½n, ð4:37Þ

where n is the discrete-time index. The optimization problem can be formulated as

1 X
N
P½njh½nj2
max log2 ð1 þ Þ
P1 , P2 , ⋯, PN N N0
n¼1
ð4:38Þ
1 X
N
P½ng½n
¼ max log2 ð1 þ Þ,
P1 , P2 , ⋯, PN N N0
n¼1

which is subject to the following constraint:

X
N
P½n  P: ð4:39Þ
n¼1

The optimum solution can again be determined by the Lagrangian method:

Popt ½n 1 N þ 1 1þ
¼ð  0 Þ ¼ð  Þ ,
P μP g½nP ρtsh ρ
|{z} |ffl{zffl}
1 1 ð4:40Þ
ρtsh ρ

ðxÞþ ¼ max ð0, xÞ,

and it is clearly the same allocation policy (the water-filling method) as given by Eq. (4.34), which is illustrated in Fig. 4.6.
The threshold SNR equation can be obtained after substituting Eq. (4.40) into (4.39):

N
þ
1 X 1 1
 ¼ 1: ð4:41Þ
N n¼1 ρtsh ρ

As N ! 1, summation becomes integral, and we obtain from Eq. (4.41) the following power allocation policy:
4.2 Capacity of Flat-Fading and Frequency-Selective Wireless Fading Channels 219

Fig. 4.6 The illustration of the P[n’]/P=0

optimum power allocation
strategy in time domain P[n]/P
1/Utsh
P[n]/P

Discrete-time, n

Encoder Utsh Decoder Utsh

Encoder U1 Decoder U1
h[i] z[i]
…

…
x[i] y[i]
…

…
Encoder UN1 Decoder UN1
U[i]-based selection U[i]-based selection

Encoder UL Decoder UL
System encoder System decoder

Fig. 4.7 The illustration of the multiplexed coding and decoding scheme


þ
1 1
 ¼ 1, ð4:42Þ
ρtsh ρ

which is identical to Eq. (4.34). The corresponding channel capacity for the optimum power adaptation strategy in time
domain will be

Popt ðgÞ Pg Popt ðgÞ

C ¼ hWlog2 ð1 þ Þi ¼ hWlog2 ð1 þ ρÞi ,
P N0 P g
|{z}
ρ g ð4:43Þ
Popt ðgÞ þ
1 1
¼ð  Þ :
P ρtsh ρ

The channel capacity-achieving coding scheme, inspired by this method, is the multiplexed coding and decoding scheme,
shown in Fig. 4.7.
The starting point is the quantization of the range of SNRs for different fading effects to a finite set {ρ n | 1  n  L}. For
each ρn, we design an encoder/decoder pair for an AWGN channel with SNR ρn achieving the channel capacity. The output xn
of encoder ρn is transmitted with an average power P(ρn), and the corresponding data rate is Rn ¼ Cn, where Cn is the capacity
of a time-invariant AWGN channel with received SNR P(ρn) ρn/P. These encoder/decoder pairs are selected according to the
CSI, that is, SNR ρn. In other words, when ρ[i] ffi ρn, the corresponding pair of ports get connected through the wireless
channel. Of course, we assume that CSI is perfect. Clearly, this multiplexed encoding and decoding scheme effectively
replaces the time-varying channel with a set of time-invariant channels operating in parallel, wherein the nth channel is only
used when ρ[i] ffi ρn. The average rate transmitted over the channel is determined by averaging, namely, ∑nCnpn, where pn is
the probability of CSI being ρn—in other words, the percentage of time that the channel SNR was equal to ρn. Of course, the
complexity and cost of this scheme are high. To reduce the system complexity and cost, the channel inversion can be used.
220 4 Capacities of Wireless and Optical Channels

Channel Inversion and Zero-Outage Capacity

The channel inversion represents a suboptimal transmitter power adaptation strategy, where the transmitter employs the
available CSI to maintain a constant received power; in other words, this scheme “inverts” the channel fading effects. The
encoder and decoder in this scenario will see a time-invariant AWGN channel. The power loading is reversely proportional to
the SNR:

PðρÞ=P ¼ ρ0 =ρ, ð4:44Þ

so that when the channel conditions are bad, we load higher power such that the receiver sees the constant power. By
substituting Eq. (4.44) into (4.32), we obtain
ð
ðρ0 =ρÞf ðρÞdρ ¼ 1, ð4:45Þ

so that the constant of proportionality (SNR) ρ0 can be determined as

ρ0 ¼ 1=h1=ρi, ð4:46Þ

and the corresponding channel capacity becomes


1
C ¼ W log 2 ð1 þ ρ0 Þ ¼ W log 2 1 þ : ð4:47Þ
h1=ρi

Based on Eq. (4.47), we conclude that the capacity-achieving transmission strategy is to employ fixed-rate encoder and
decoder designed for an AWGN channel with SNR equal to ρ0, regardless of the wireless channel conditions. Given that the
data rate is fixed under all channel conditions, even very bad ones, there is no channel outage when channel conditions are
bad, and the corresponding channel capacity is called zero-outage capacity. The main drawback of this strategy is that the
zero-outage capacity can exhibit a large data rate reduction, compared to the Shannon capacity in extremely bad fading
conditions, when ρ ! 0 and subsequently C ! 0. To solve for this problem, the truncated channel inversion is used.

Truncated Channel Inversion and Outage Capacity

As explained above, the requirement to maintain a constant data rate regardless of the fading state results in inefficacy of this
method, so that the corresponding zero-outage capacity can be significantly smaller than the Shannon capacity in a deep fade.
When the fading state is very bad by not transmitting at all, we can ensure that data rates are higher in the other fading states
and on such a way we significantly improve the channel capacity. The state when we do not transmit is an outage state, so that
the corresponding capacity can be called the outage capacity, which is defined as the maximum data rate that can be kept
constant, when there is no outage, times the probability of not having outage (1  Poutage). Outage capacity is achieved with a
so-called truncated channel inversion policy for power adaptation, in which we only transmit reversely proportional to SNR,
when the SNR is larger than the threshold SNR ρtsh:

ρ0 =ρ, ρ  ρtsh
PðρÞ=P ¼ : ð4:48Þ
0, ρ < ρtsh

The parameter ρ0 can be determined as

ð
1
1
ρ0 ¼ 1=h1=ρiρtsh , h1=ρiρtsh ¼ f ðρÞdρ: ð4:49Þ
ρ
ρtsh

For the outage probability, defined as Poutage ¼ P(ρ < ρtsh), the corresponding channel capacity will be
4.2 Capacity of Flat-Fading and Frequency-Selective Wireless Fading Channels 221

!
1
C ðPout Þ ¼ W log 2 1þ Pðρ  ρtsh Þ: ð4:40Þ
h1=ρiρtsh

The channel capacity can be obtained by the following maximization with respect to ρtsh:
!
1

C ¼ max W log 2 1þ 1  Poutage : ð4:51Þ
ρtsh h1=ρiρtsh

4.2.2 Frequency-Selective Fading Channel Capacity

Two cases of interest are considered here: (i) time-invariant and (ii) time-variant channels.

4.2.2.1 Time-Invariant Channel Capacity

We consider here the time-invariant channel, as shown in Fig. 4.8, and we assume that a total transmit power cannot be larger
than P. In full CSI scenario, the transfer function H( f ) is known to both the transmitter and the receiver. We first study the
block-fading assumption for H( f ) so that the whole frequency band can be divided into subchannels each of bandwidth equal
to W, wherein the transfer function is constant for each block i, namely, H( f) ¼ Hj, which is illustrated in Fig. 4.9.
Therefore, we decomposed the frequency-selective fading channel into a set of the parallel AWGN channels with
corresponding SNRs being |Hi|2Pi/(N0W) on the ith channel, where Pi is the power allocated to the ith channel in this parallel
set of subchannels, subject to the power constraint ∑i Pi  P.
The capacity of this parallel set of channels is the sum of corresponding rates associated with each channel, wherein the
power is optimally allocated over all subchannels:
0 1
X B jH i j 2
PiC X
C ¼ max W log 2 @1 þ N 0 W A ¼ max W log 2 ð1 þ ρi Þ: ð4:52Þ
P |fflfflffl{zfflfflffl} P
Pi : P Pi ρi P: P P i i
i i

The optimum power allocation policy of the ith subchannel is the water filling:

z[i]

Sequence x[i] y[i] Est.

Mapper + Power
Encoder H(f) Decoder
I/Q modulator control seq.
Time-invariant channel Channel
estimator

Fig. 4.8 The illustration of the wireless communication system operating over the time-invariant channel characterized by the transfer function
H( f )

H(f)
H3 H6
H2
W
H5
H1 H4
f

Fig. 4.9 The illustration of the block frequency-selective fading in which the channel is divided into subchannels, each of them having the
bandwidth W
222 4 Capacities of Wireless and Optical Channels

Pi / P
1/Utsh

Pi /P 1/Ui

Fig. 4.10 The illustration of the water filling in the frequency-selective block-fading channels


Pi 1=ρtsh  1=ρi , ρi  ρtsh
¼ , ð4:53Þ
P 0, ρi < ρtsh

and the threshold SNR is obtained as the solution of the following equation:

X
1 1
þ
 ¼ 1: ð4:54Þ
i
ρtsh ρi

The illustration of the water filling in the frequency-selective block-fading channels is provided in Fig. 4.10. The
corresponding channel capacity of the optimum distribution, obtained by substituting Eq. (4.53) into (4.52), is obtained as
follows:

X

ρi
C¼ W log 2 : ð4:55Þ
i:ρi ρtsh
ρtsh

When the transfer function H( f ) is continuous, the capacity for the same power constraint P is similar to the case of the
block-fading channel, but instead of summation, we need to perform the integration so that the corresponding channel
capacity expression is given by
ð

j H ð f Þj 2 P ð f Þ
C¼ ð max W log 1 þ df : ð4:56Þ
2
N0
Pð f Þ: Pð f Þdf  P

The optimum power allocation strategy is, not surprising, again the water filling:

Pð f Þ 1=ρtsh  1=ρð f Þ, ρð f Þ  ρtsh jH ð f Þj2 P
¼ , ρð f Þ ¼ : ð4:57Þ
P 0, ρð f Þ < ρtsh N0W

After substituting Eq. (4.57) into (4.56), we obtain the following channel capacity for the optimum power allocation case:
ð  
ρð f Þ
C¼ W log 2 df : ð4:58Þ
ρtsh
f : ρð f Þρtsh
4.2 Capacity of Flat-Fading and Frequency-Selective Wireless Fading Channels 223

Example Let us now consider the time-invariant frequency-selective block-fading channel with three subchannels of
bandwidth 2 MHz, with frequency response amplitudes being 1, 2, and 3, respectively. For the transmit power 10 mW and
noise power spectral density of N0 ¼ 109 W/Hz, we are interested to determine the corresponding Shannon capacity.
The signal power-to-noise power is given by ρ0 ¼ P/(N0W ) ¼ 5. The subchannels’ SNRs can be expressed as ρi ¼ P|Hi|2/
(N0W ) ¼ ρ0|Hi|2. For H1 ¼ 1, H2 ¼ 2, and H3 ¼ 3, the corresponding subchannel SNRs are ρ1 ¼ ρ0|H1|2 ¼ 5, ρ2 ¼ ρ0|
H2|2 ¼ 20, and ρ3 ¼ ρ0|H3|2 ¼ 45, respectively. Based on Eq. (4.54), the threshold SNR can be determined from the following
equation:

3 1 1 1 1 1 1
¼1þ þ þ ¼1þ þ þ ,
ρtsh ρ1 ρ2 ρ 3 5 20 45

as follows: ρtsh ¼ 2.358 < ρi for very i. The corresponding channel capacity is given by Eq. (4.55) as

X
3 X
3
C¼ W log 2 ð1 þ ρi Þ ¼ W log 2 ð1 þ ρi Þ ¼ 16:846 kb=s:
i¼1 i¼1

4.2.2.2 Time-Variant Channel Capacity

The time-varying frequency-selective fading channel is similar to the time-invariant model, except that H( f) ¼ H( f, i), i.e., the
channel varies over both frequency and time, and the corresponding channel model is given in Fig. 4.11.
We can approximate channel capacity in time-varying frequency-selective fading by taking the channel bandwidth W of
interest and divide it up into subchannels of bandwidth equal to the channel coherence bandwidth Bc, as shown in Fig. 4.12.
Under the assumption that each of the resulting subchannels is independent, time-varying with flat fading within the
subchannel H( f, k) ¼ Hi [k] on the ith subchannel at time instance k, we can obtain the capacity for each of these flat-
fading subchannels based on the average power Pi that we allocate to each subchannel, subject to a total power constraint P.
Since the channels are mutually independent, the total channel capacity is just equal to the sum of capacities on the
individual narrowband flat-fading channels, subject to the total average power constraint, averaged over both time and
frequency domains:

z[i]

Sequence x[i] y[i] Est.

Mapper + Power
Encoder H(f,t) Decoder
I/Q modulator control seq.
Time-variant channel Channel
estimator

Fig. 4.11 The illustration of the wireless communication system operating over the time-varying channel characterized by the time-variant transfer
function H( f,t)

H(f,k) H3[k] Coherence

H2[k] bandwidth
Bc
H5[k]
H4[k]
H1[k]
f

Channel bandwidth W

Fig. 4.12 The illustration of the frequency-selective time-varying fading channels

224 4 Capacities of Wireless and Optical Channels

X
 ð
1

ρi
C¼ P
max C i Pi , C i ¼ Bc log f ðρi Þdρi , ð4:59Þ
f ig
P : Pi P i
2
ρ tsh
i
ρtsh

which requires the knowledge of the distribution functions of SNR for each subchannel, namely, f(ρi).
When we fixed the average power per subchannel, the optimum power allocation policy is the water filling in the time
domain. On the other hand, the optimum power allocation strategy among subchannels is again the water filling, but now in
the frequency domain. Therefore, the optimum power allocation strategy for time-varying channels is the two-dimensional
water filling, over both time and frequency domains. The instantaneous SNR in the ith subchannel at the kth time interval is
given by ρi ½k  ¼ jH i ½k j2 P=ðN 0 Bc Þ:
The Shannon capacity assuming the perfect full CSI will be achieved by optimizing the power allocation in both time
domain, by fixing ρi[k] ¼ ρi, and frequency domain over subchannel indices i as follows:

X ð
Pi ðρi Þ
C¼ ð max 1 Bc log2 ð1 þ  Þf ðρi Þdρi , ð4:60Þ
P P
Pi ðρi Þ: 
Pi ðρi Þf ðρi Þdρi  P i
0
i

which is subject to the following constraint:

Xð
1

Pi ðρi Þf ðρi Þdρi ¼ P: ð4:61Þ

i
0

The optimum power allocation strategy can be determined by the Lagrangian method, and the corresponding solution is given
by

Pi ðρi Þ 1=ρtsh  1=ρi , ρi  ρtsh
¼ : ð4:62Þ
P 0, ρi < ρtsh

The threshold SNR can be obtained as the solution of the following equation:

X ð
1

ð1=ρtsh  1=ρi Þþ f ðρi Þdρi ¼ 1: ð4:63Þ

i
ρtsh

After substituting Eq. (4.62) into (4.60), we obtain the following channel capacity expression for the optimum
two-dimensional adaptation strategy:

X ð
1  
ρi
C¼ Bc log 2 f ðρi Þdρi : ð4:64Þ
i
ρtsh
ρtsh

4.3 Capacity of Channels with Memory

In this section, we will describe Markov and McMillan’s sources with memory [18] and the McMillan-Khinchin channel
model with memory [19] and describe how to determine the entropies of sources with memory and mutual information of
channels with memory [20, 22]. All this serves as a baseline in our later analysis of optical channel capacity, in particular for
fiber-optic communications [23–30].
4.3 Capacity of Channels with Memory 225

4.3.1 Markov Sources and Their Entropy

The finite Markovian chain is a commonly used model to describe both the sources and channels with memory. The
Markovian stochastic process with a finite number of states {S} ¼ {S1, . . ., Sn} is characterized by transition probabilities
π ij of moving from state Si to state Sj (i,j ¼ 1, . . ., n). The Markov chain is the sequence of states with transitions governed by
the following transition matrix:
2 3
π 11 π 12 ⋯ π 1n
  6 6 π 21 π 22 ⋯ π 2n 7
7
Π ¼ π ij ¼ 6 7, ð4:65Þ
4⋯ ⋯ ⋱ ⋮5
π n1 π n2 ⋯ π nn

where ∑j π ij ¼ 1.

Example Let us observe three-state Markov source shown in Fig. 4.13(a). The corresponding transition matrix is given by
2 3
0:6 0:4 0
  6 7
Π ¼ π ij ¼ 4 0 1 0 5:
0:3 0:7 0

As we see, the sum in any row is equal to 1. From state S1, we can move to either state S2 with probability 0.4 or to stay in state
S1 with probability 0.6. Once we enter the state S2, we stay there forever. Such a state is called absorbing, and the
corresponding Markov chain is called absorbing Markov chain. The probability of moving from state S3 to state S2 in two
ð2Þ
steps can be calculated as π 32 ¼ 0:3  0:4 þ 0:7  1 ¼ 0:82. Another way to calculate this probability is to find the second
power of transition matrix and then read out the probability of desired transition:
2 3
h i 0:36 0:64 0
ð2Þ 6 7
Π ¼ π ij ¼ ΠΠ ¼ 4 0
2
1 0 5:
0:18 0:82 0

The probability of reaching all states from initial states after k steps can be determined by

ΠðkÞ ¼ Pð0Þ Πk , ð4:66Þ

where P(0) is the row vector containing the probabilities of initial states.

Fig. 4.13 Two three-state 0.6

Markov chains: (a) irregular
Markov chain and (b) regular
Markov chain
S1 S1 0.75

0.4 0.3 0.25

1
0.7 1
S2 S3 S2 S3

1
(a) (b)
226 4 Capacities of Wireless and Optical Channels

If the transition matrix Π k has only nonzero elements, we say that Markov chain is regular. Accordingly, if the k0th power
of Π does not have any zero entry, any kth power of Π for k > k0 will not have any zero entry either. The so-called ergodic
Markov chains are the most important ones from a communication system point of view. We can say that Markov chain is
ergodic if it is possible to move from any specific state to any other state in a finite number of steps with nonzero probability.
The Markov chain from the example above is non-ergodic. It is also interesting to notice that the transition matrix for this
example has the following limit:
2 3
0 1 0
6 7
T ¼ lim Pk ¼ 4 0 1 0 5:
k!1
0 1 0

Example Let us now observe an example of regular Markov chain, which is shown in Fig. 4.13(b). The transition matrix Π,
its fourth and fifth powers, and matrix Π limit as k ! 1 are given, respectively, as
2 3 2 3
0 0:25 0:75 0:5625 0:0625 0:3750
6 7 6 7
Π¼6
40 0 1 7 4 6
5, Π ¼ 4 0:7500 0 0:2500 7
5,
1 0 0 0:2500 0:1875 0:5625
2 3 2 3
0:3750 0:1406 0:4844 0:4444 0:1112 0:4444
6 7 6 7
Π5 ¼ 6
4 0:2500 0:1875 0:5625 7 lim Πk ¼ 6
5, T ¼ k!1 4 0:4444 0:1112 0:4444 7
5:
0:5625 0:0625 0:3750 0:4444 0:1112 0:4444

We can see that the fourth power has one zero entry, while the fifth power and all higher powers do not have zero entries.
Therefore, this Markov chain is both regular and ergodic one. We can also notice that the stationary transition matrix T has
identical rows.
It is evident from the example above that for regular Markov chain, the transition matrix converges to the stationary
transition matrix T with all rows identical to each other:
2 3
t1 t2 ⋯ tn
6t tn 7
6 1 t2 ⋯ 7
T ¼ lim Πk ¼ 6 7: ð4:67Þ
k!1 4⋯ ⋯ ⋱ ⋮5
t1 t2 ⋯ tn

Additionally, the following is valid:

lim ΠðkÞ ¼ lim Pð0Þ Πk ¼ Pð0Þ T = ½t 1 t 2 ⋯ t n , ð4:68Þ

k!1 k!1

so we can find stationary probabilities of states (or equivalently solve for elements of T) from equations

t 1 ¼ π 11 t 1 þ π 21 t 2 þ ⋯ þ π n1 t n
t 2 ¼ π 12 t 1 þ π 22 t 2 þ ⋯ þ π n2 t n
⋮
ð4:69Þ
t n ¼ π 1n t 1 þ π 2n t 2 þ ⋯ þ π nn t n
X n
t i ¼ 1:
i¼1

For instance, for Markov chain from Fig. 4.13(b), we can write that
4.3 Capacity of Channels with Memory 227

t 1 ¼ t 3 , t 2 ¼ 0:25t 1 , t 3 ¼ 0:75t 1 þ t 2 , t 1 þ t 2 þ t 3 ¼ 1,

while the corresponding solution is given by t1 ¼ t3 ¼ 0.4444, t2 ¼ 0.1112.

The uncertainty of source associated with Markov source {S} ¼ {S1, . . ., Sn} when moving one step ahead from an initial
ð1Þ
state Ai, here denoted as H i , can be expressed as

ð1Þ
X
n
Hi ¼ π ij log π ij : ð4:70Þ
j¼1

If the probability associated with state Si is equal to pi, we can obtain the entropy of Markov source by averaging over
entropies associated with all states. The uncertainty of moving one step ahead becomes

n o Xn X
n X
n
ð1Þ ð1Þ
H ðX Þ ¼ H ð1Þ ¼ E H i ¼ pi H i ¼  pi π ij log π ij : ð4:71Þ
i¼1 i¼1 j¼1

In a similar fashion, the entropy of Markov source for moving k steps ahead from initial states is given by

n o Xn X
n X
n
ðk Þ ðk Þ ðk Þ ðk Þ
H ðkÞ ¼ E H i ¼ pi Hi ¼ pi π ij log π ij : ð4:72Þ
i¼1
|{z} i¼1 j¼1
P
n
ðk Þ ðk Þ
 π ij log π ij
j¼1

It can be shown that for ergodic Markov sources, there is a limit defined as H(1) ¼ lim H ðkÞ =k: In order to prove this, we can
k!1
use the following property, which should be proved as a homework problem:

H ðkþ1Þ ¼ H ðkÞ þ H ð1Þ : ð4:73Þ

By applying this property in an iterative fashion, we obtain that

H ðkÞ ¼ H ðk1Þ þ H ð1Þ ¼ H ðk2Þ þ 2H ð1Þ ¼ ⋯ ¼ kH ð1Þ ¼ kH ðX Þ: ð4:74Þ

From Eq. (4.74), it is evident that

H ðkÞ
lim ¼ H ð1Þ ¼ H ðX Þ: ð4:75Þ
k!1 k

Equation (4.75) can be now used as an alternative definition of entropy of Markov source, which is applicable to arbitrary
stationary source as well.

Example Let us determine the entropy of Markov source shown in Fig. 4.13(b). By using the definition from Eq. (4.71), we
obtain that

P
n P
n
HðXÞ ¼  pi pij log pij
i¼1 j¼1
¼ 0:4444ð0:25log 0:25 þ 0:75log 0:75Þ  0:1111  1log 1
0:4444  1log 1 ¼ 0:6605 bits:
228 4 Capacities of Wireless and Optical Channels

4.3.2 McMillan Sources and Their Entropy

A McMillan’s description [18] of the discrete source with memory is more general than that of Markov chain. In our case, we
will pay attention to stationary sources. Let S represent a finite source alphabet with corresponding letters {s1, s2, . . ., sM} ¼ S.
The source emits one symbol at the time instance tk. The transmitted sequence can be represented as X ¼ {. . ., x1, x0, x1, . . .},
where xi 2 S. Among all members of ensemble {X}, we are interested just in those having specified source symbols at certain
prespecified instances of time. All sequences with these properties create so-called cylinder set.

Example Let specified location be defined by indices 1, 0, 3, and k, with corresponding symbols at these locations equal to
x1 ¼ s2, x0 ¼ s5, x3 ¼ s0, xk ¼ sn  1. The corresponding cylinder set will be given as C1 ¼ {. . ., x2, s2, s5, x1, x2, s0, . . .,
xk ¼ sn  1, . . .}. Since we observe the stationary processes, the statistical properties of cylinder will not change if we shift the
cylinder for one time unit in either direction (either by T or by T1). For instance, the time-shifted C1 cylinder is given by
TC1 ¼ {. . ., x1, s2, s5, x2, x3, s0, . . ., xk + 1 ¼ sn  1, . . .}.
The stationary property for arbitrary cylinder C can be expressed as
 
PfTC g ¼ P T 1 C ¼ PfCg, ð4:76Þ

where P{} denotes the probability measure.

Let us now specify n letters from alphabet S to be sent at positions k + 1, . . ., k + n. This sequence can be denoted as xk + 1,
. . ., xk + n, and there is the total of Mn possible sequences. The entropy of all possible sequences is defined as
X
Hn ¼  pm ðC Þ log pm ðC Þ, ð4:77Þ
C

where pm() is the probability measure. The McMillan’s definition of entropy of stationary discrete source is given by [18]

Hn
H ðX Þ ¼ lim : ð4:78Þ
n!1 n

As we can see, the McMillan’s definition of entropy is consistent with Eq. (4.75), which is applicable to stationary Markovian
sources.

4.3.3 McMillan-Khinchin Model for Channel Capacity Evaluation

Let the input and output alphabets of the channel be finite and denoted by A and B, respectively, while the channel input and
output sequences are denoted by X and Y. The noise behavior for memoryless channels is generally captured by a conditional
probability matrix P{bj|ak} for all bj 2 B and aj 2 A. On the other side, in channels with finite memory (such as the optical
channel), the transition probability is dependent on the transmitted sequences up to the certain prior finite instance of time. For
instance, the transition matrix for channel described by the Markov process has the form P{Yk ¼ b|. . ., X1, X0, X1, . . .,
Xk} ¼ P{Yk ¼ b|Xk}.
Let us consider a member x of input ensemble {X} ¼ {. . ., x2, x1, x0, x1, . . .} and its corresponding channel output y from
ensemble {Y} ¼ {. . ., y2, y1, y0, y1, . . .}. Let X denote all possible input sequences and Y denote all possible output
sequences. By fixing a particular symbol at specific location, we can obtain the cylinder. For instance, cylinder x4,1 is obtained
by fixing symbol a1 to position x4, so it is x4,1 ¼ . . ., x1, x0, x1, x2, x3, a1, x5, . . . The output cylinder y1,2 is obtained by fixing
the output symbol b2 to position 1, namely, y1,2 ¼ . . ., y1, y0, b2, y2, y3, . . . In order to characterize the channel, we have to
determine transition probability P(y1,2|x4,1), which is probability that cylinder y1,2 was received if cylinder x4,1 was
transmitted. Therefore, for all possible input cylinders SA ⊂ X, we have to determine the probability that cylinder SB ⊂ Y
was received if SA was transmitted.
The channel is completely specified by the following: (i) input alphabet A, (ii) output alphabet B, and (iii) transition
probabilities P{SB|SA} ¼ vx for all SA 2 X and SB 2 Y. Accordingly, the channel is specified by the triplet [A, vx, B]. If
transition probabilities are invariant with respect to time shift T, which means that vx(TS) ¼ vx(S), the channel is said to be
stationary. If the distribution of Yk depends only on the statistical properties of sequence . . ., xk-1, xk, we say that the channel is
4.3 Capacity of Channels with Memory 229

without anticipation. If, furthermore, the distribution of Yk depends only on xk-m, . . ., xk, we say that the channel has finite
memory of m units.
The source and channel may be described as a new source [C, ζ], where C is the Cartesian product of input A and output
B alphabets (C ¼ A  B), while ζ is a corresponding probability measure. The joint probability of symbol (x,y) 2 C, where x 2
A and y 2 B, is obtained as the product of marginal and conditional probabilities: P(x\y) ¼ P{x}P{y|x}.
Let us further assume that both source and channel are stationary. Following the description presented in [6, 19], it is useful
to describe the concatenation of a stationary source and a stationary channel as follows:

1. If the source [A, μ] (μ is the probability measure of the source alphabet) and the channel [A, vx, B] are stationary, the product
source [C, ζ] will also be stationary.
2. Each stationary source has an entropy, and therefore [A, μ], [B, η] (η is the probability measure of the output alphabet), and
[C, ζ] each have the finite entropies.
3. These entropies can be determined for all n-term sequences x0, x1, . . ., xn  1 emitted by the source and transmitted over the
channel as follows [19]:

H n ðX Þ fx0 , x1 , . . . , xn1 g
H n ðY Þ fy0 , y1 , . . . , yn1 g
H n ðX, Y Þ fðx0 , y0 Þ, ðx1 , y1 Þ, . . . , ðxn1 , yn1 Þg ð4:79Þ
H n ðYjX Þ fðYjx0 Þ, ðYjx1 Þ, . . . , ðYjxn1 Þg
H n ðXjY Þ fðXjy0 Þ, ðXjy1 Þ, . . . , ðXjyn1 Þg:

It can be shown that the following is valid:

H n ðX, Y Þ ¼ H n ðX Þ þ H n ðYjX Þ
ð4:80Þ
H n ðX, Y Þ ¼ H n ðY Þ þ H n ðXjY Þ:

The equations above can be rewritten in terms of entropies per symbol:

1 1 1
H ðX, Y Þ ¼ H n ðX Þ þ H n ðYjX Þ
n n n n ð4:81Þ
1 1 1
H ðX, Y Þ ¼ H n ðY Þ þ H n ðXjY Þ:
n n n n

For sufficiently long sequences, the following channel entropies exist:

1 1
lim H n ðX, Y Þ ¼ H ðX, Y Þ lim H n ðX Þ ¼ H ðX Þ
n!1 n n!1 n
1 1
lim H n ðY Þ ¼ H ðY Þ lim H n ðXjY Þ ¼ H ðXjY Þ ð4:82Þ
n!1 n n!1 n
1
lim H n ðYjX Þ ¼ H ðYjX Þ:
n!1 n

The mutual information also exists, and it is defined as

I ðX, Y Þ ¼ H ðX Þ þ H ðY Þ  H ðX, Y Þ: ð4:83Þ

The stationary information capacity of the channel is obtained by maximization of mutual information over all possible
information sources:

CðX, Y Þ ¼ max I ðX, Y Þ: ð4:84Þ

230 4 Capacities of Wireless and Optical Channels

The results of analysis in this section will be applied later in the evaluation the information capacity of optical channel with
memory. But, before that, we will briefly describe the adopted model for signal propagation in single-mode optical fibers,
which will also be used in the evaluation of channel capacity.

4.4 Calculation of Information Capacity by the Forward Recursion of the BCJR Algorithm

We will now evaluate the channel capacity of multilevel/multidimensional modulation schemes for an i.i.d. information
source, which is in literature also known as the achievable information rate [2, 3, 20–28]. The approach that will be applied
was initially proposed in [20–22] to evaluate the information rate of binary modulation schemes over linear ISI channels. A
similar method has been also used in [23–27] to study the information capacity of on-off keying channels in the presence of
fiber nonlinearities. Finally, the method has been generalized to multilevel modulations in [28]. The i.i.d. channel capacity
represents a lower bound on channel capacity. Some papers employ the mutual information definition formula to evaluate the
achievable rates [29, 30]; however, this approach does not take the optical channel memory into account.
To calculate the i.i.d. channel capacity, we will model the whole transmission system as a dynamical intersymbol
interference (ISI) channel, in which m previous and next m symbols have influence on the observed symbol, as shown in
Fig. 4.14. Therefore, the model is a particular instance of the McMillan-Khinchin model described in Sect. 4.3.3. The optical
communication system is characterized by the conditional PDF of the output N-dimensional matrix of samples y ¼ (y1, . . ., yn,
. . .), where yi ¼ (yi,1, . . ., yi,D) 2 Y, for given source sequence x ¼ (x1, . . ., xn, . . .), xi 2 X = {0, 1, . . ., M  1}. The set X
represents the set of indices of constellation points in the corresponding M-ary N-dimensional signal constellation diagram,
where N is the number of basis functions used to represent a given signal constellation. (For instance, N equals 2 for M-ary
PSK and M-ary QAM, while it equals 3 for cube constellation.) The Y represents the set of all possible channel outputs, where
yi,j ( j ¼ 1, . . ., N ) corresponds to the jth coordinate of the ith sample of channel output. The state in the trellis from Fig. 4.14 is
defined as sj ¼ (xj-m, xj-m + 1, . . ., xj, xj + 1, . . ., xj + m) ¼ x[j-m, j + m], where xk denotes the index of the symbol from the set
X ¼ {0, 1, . . ., M  1} of possible indices. Every symbol carries m ¼ log2M bits and was constructed using the appropriate
mapping rule (natural, Gray, anti-Gray, etc.). The memory of the state is equal to 2 m + 1, with 2 m being the number of
symbols that influence the observed symbol from both sides. The trellis has M2m + 1 states, where each of them corresponds to

000 s0 0/0 s0 000

1/0
001 s1 s1 001
2/0
002 s2 0/1 3/0 s2 002
003 s3 s3 003
.
s4 010
.
. 1/1
s5 011

100 s16 2/1 s6 012

101 s17 s7 013
102 s18 0/2 s8 020
1/2
2/2 s9 021
.
3/2 s10 022
.
. s11 023

Fig. 4.14 A portion of trellis describing the model of optical channel with memory for 4-ary N-dimensional signal constellations with memory
2m+1¼3
4.4 Calculation of Information Capacity by the Forward Recursion of the BCJR Algorithm 231

a different (2 m + 1)-symbol pattern (configuration). The state index is determined by considering (2 m + 1) symbols as digits
in the numerical system with the base M.
For complete characterization of trellis, conditional PDFs are needed, and they can be evaluated by using one of the
following methods: (i) estimating the histograms [20], (ii) using an instanton approach [24], or (iii) by Edgeworth expansion
[25]. Notice that this channel model is consistent with the McMillan model discussed above. The information rate can be
calculated by Eq. (4.83), rewritten in vector form as follows:

I ðY, X Þ ¼ H ðY Þ  H ðYjX Þ, ð4:85Þ

where H(U) ¼  hlog2P(U)i denotes the entropy of a random (variable) vector U. By using the Shannon-McMillan-Breiman
theorem, which states that [5]

h log 2 PðY Þi ¼  lim n!1 ð1=nÞ log 2 Pðy½1, nÞ, ð4:86Þ

the information rate can be determined by calculating log2(P(y[1,n])) for the sufficiently long source sequence propagated
through the channel. By substituting Eq. (4.86) into (4.85), we obtain the following expression suitable for practical
calculation of information capacity:

1 X X
n n
IðY, XÞ ¼ lim ½ log2 Pðyi jy½1, i  1, x½1, nÞ log2 Pðyi jy½1, i  1Þ: ð4:87Þ
n!1 n
i¼1 i¼1

The first term on the right side of Eq. (4.87) can be straightforwardly calculated from conditional probability density functions
(PDFs) P(y[j  m, j + m]|s). We can use the forward recursion of the multilevel BCJR algorithm, described in detail in Chap. 9,
to calculate log2P(yi|y[1,i  1]). We can now define the forward metric αj(s) ¼ log{p(sj ¼ s,y[1,j])} ( j ¼ 1, 2, . . ., n) and the
branch metric γ j(s0 ,s) ¼ log[p(sj ¼ s,yj,sj1 ¼ s0 )] as

α j ðsÞ ¼ max
½α 0 Þ þ γ ðs0 , sÞ  log M,
0
j1 ðs j 2
s ð4:88Þ
γ j ðs0 , sÞ ¼ log ½pðy j jx½ j  m, j þ mÞ,

where the max* operator is defined by max*(x,y) ¼ log(ex + ey) ¼ max(x,y) + log[1 + exp(|x  y|)].
The ith term, log2P(yi|y[1,i  1]), can now be calculated in an iterative fashion by

log2 Pðyi jy½1, i  1Þ ¼ max αi ðsÞ, ð4:89Þ

s

where the max* operator was applied for all s 2 S (S denotes the set of states in the trellis shown in Fig. 4.14). Information
capacity is defined as

C ¼ max I ðY, XÞ, ð4:90Þ

where the maximization is performed over all possible input distributions. Since the optical channel has memory, it is natural
to assume that optimum input distribution will also be the one with the memory. By considering the stationary input
distributions in the form p(xi|xi  1, xi  2, . . .) ¼ p(xi|xi  1, xi  2, . . ., xi  k), we can determine the transition probabilities
of the corresponding Markov model that would maximize the information rate by nonlinear numerical optimization [31, 32].
This method is applicable to both memoryless channels and for channels with memory. For instance, we calculated the
memoryless information capacities for different signal constellation sizes and two types of QAM constellations (square QAM
and star QAM) by observing a linear channel model, and results are shown in Fig. 4.15. As we can see, the information
capacity can be closely approached even with a uniform information source if constellation size is sufficiently large. It is
interesting to notice that star QAM outperforms the corresponding square QAM for low and medium signal-to-noise ratios
(SNRs), while square QAM outperforms star QAM for high SNRs. The iterative polar quantization (IPQ)-based modulation
format, introduced in [33, 34], also known as iterative polar modulation (IPM), significantly outperforms both square QAM
and star QAM.
232 4 Capacities of Wireless and Optical Channels

Fig. 4.15 Information capacities

Infromation capacity, C [bits/channel use]

11 Shannon Capacity
for linear channel model and
different signal constellation M = 64
sizes. (64-star QAM contains 10 QAM
8 rings with 8 points each, Star-QAM
256-star QAM contains 16 rings 9 IPQ
with 16 points, and 1024-star M = 256
QAM contains 16 rings with 8 QAM
64 points.) SNR is defined as Es/ Star-QAM
N0, where Es is the symbol energy 7 IPQ
and N0 is the power spectral M = 1024
density 6 QAM
Star-QAM
5 IPQ
M = 2048
4 IPQ

3
10 15 20 25 30 35 40
Signal-to-noise ratio, SNR [dB]

Fig. 4.16 The i.i.d. information

Infromation capacity, C [bits/channel use]

capacity per single polarization 2.0

for QPSK of aggregate data rate of
100 Gb/s against the transmission
distance. The dispersion map is
shown in Fig. 4.17
1.8

1.6
QPSK
(disp. managed transmission):
m=1
1.4 m=0
QPSK
(SMF only, backpropagation):
m=1
1.2
m=0

2000 4000 6000 8000 10000

Total transmission distance, L [km]

4.5 Information Capacity of Systems with Coherent Optical Detection

The fiber-optic channel capacity seems to be hot research topics, judging by the number of publications related to this topic
[1–4, 23–30, 35–45]. In this section, we study the information capacity of optical systems with coherent detection by applying
some commonly used transmission scenarios. The i.i.d. information capacity versus the number of fiber spans is shown in
Fig. 4.16. It is obtained by using Monte Carlo simulations assuming the following: (i) dispersion map is the one shown in
Fig. 4.17, with fiber parameters summarized in Table 4.1; (ii) QPSK modulation format of aggregate data rate 100 Gb/s per
4.5 Information Capacity of Systems with Coherent Optical Detection 233

N spans
D- D+ D- D+

Transmitter Receiver

ED FA EDFA EDFA EDFA

Fig. 4.17 Dispersion map under study is composed of N spans of length L ¼ 120 km, consisting of 2 L/3 km of D+ fiber followed by L/3 km of D
fiber, with pre-compensation of 1600 ps/nm and corresponding post-compensation. The fiber parameters are given in Table 4.1

Table 4.1 Fiber parameters

D+ fiber D fiber
Dispersion [ps/(nm km)] 20 40
Dispersion slope [ps/(nm2 km)] 0.06 0.12
Effective cross-sectional area [μm2] 110 50
Nonlinear refractive index [m2/W] 2.61020 2.61020
Attenuation coefficient [dB/km] 0.19 0.25

Fig. 4.18 (a) Dispersion map N spans

composed of SMF sections only
SMF
with receiver-side digital Receiver
backpropagation and (b)
+
transmitter configuration for Transmitter
receiver-side
two-dimensional signal
back-propagation
constellations EDFA

(a)
I- channel

Input m MZM to fiber

data Buffer Mapper DFB
MZM S/2

Q- channel

(b)

single polarization; and (iii) two channel memory settings and transmitter-receiver configurations. As we can see, by using the
LDPC code of sufficient length and large girth, it is possible, in principle, to achieve the total transmission distance of
~8800 km for state memory m ¼ 0 and ~ 9600 km for state memory m ¼ 1. The transmission distance can further be increased
by observing larger memory channel assumptions, which requires higher computational complexity of corresponding turbo
equalizer. On the other hand, we can use the coarse digital backpropagation approach [28] to keep the channel memory
reasonable low and then apply the method for information capacity evaluation described in this section. However, since digital
backpropagation method cannot account for the nonlinear ASE noise-Kerr nonlinearity interaction, we should use the method
described in the previous section in order to account for this effect in the calculation of information capacity. As a reference,
the i.i.d. information capacity when digital backpropagation method is used, for fiber link composed of standard SMF with
dispersion map from Fig. 4.18(a) with EDFAs with a noise figure of 6 dB deployed every 100 km, is also shown in Fig. 4.16.
We can see that digital backpropagation method helps to reduce channel memory, since the improvement for m ¼ 1 over
m ¼ 0 case is small.
In Fig. 4.19, we show the i.i.d. information capacities for three different modulation formats: (i) MPSK, (ii) star QAM
(sQAM), and (iii) iterative polar modulation (IPM), obtained by employing the dispersion map from Fig. 4.17. The chosen
symbol rate is 50 GS/s, while the launch power was set to 0 dBm. For a transmission distance of 5000 km, the
234 4 Capacities of Wireless and Optical Channels

Fig. 4.19 Information capacities

Information capacity, C [bits/channel use]

7
per single polarization for star MPSK: sQAM:
QAM (sQAM), MPSK, and IPM M=4, m=1 M=16
for different constellation sizes.
M=4, m=0 M=32
The dispersion map is shown in 6
Fig. 4.17. EDFA’s NF ¼ 6 dB M=8, m=0 M=64
M=8, m=1 IPM:
M=16 M=64
5 M=128
M=32

1000 2000 3000 4000 5000

Total transmission distance, L [km]

Fig. 4.20 Information capacity

Infomration capacity, C [bits/channel use]

per single polarization against IPM: Ltot=2000 km

launch power P for a total
transmission distance of 5.0 M= 16 Rs =50 GS/s
Ltot ¼ 2000 km. EDFA’s M= 32
NF ¼ 3 dB M= 128 NF= 3 dB

4.5

4.0

3.5

3.0
-6 -4 -2 0 2 4 6 8 10
Launch power, P [dBm]

i.i.d. information capacity is 2.72 bits/symbol (the aggregate rate is 136 Gb/s per wavelength), for 2000 km it is 4.2 bits/
symbol (210 Gb/s), and for 1000 km the i.i.d. information capacity is 5.06 bits/symbol (253 Gb/s per wavelength).
The information capacity as a function of launch power for fixed total transmission distance Ltot ¼ 2000 km and NF ¼ 3 dB,
and dispersion map from Fig. 4.18(a), is shown in Fig. 4.20. As we see, the single polarization channel capacity of Copt ¼ 5.16
bits/s/Hz over transmission distance to 2000 km can be achieved for optimum launch power of Popt ¼ 2 dBm and M ¼ 128,
which is very close to the result reported in [30] where star QAM of size 2048 and optimum dispersion map based on Raman
amplifiers were employed (Figs. 4.19 and 4.20).
Regarding the channel capacity calculation for few-mode-fiber (FMF)-based communication systems, an interested reader
is referred to [35].
4.6 Hybrid Free-Space Optical (FSO)-RF Channel Capacity 235

4.6 Hybrid Free-Space Optical (FSO)-RF Channel Capacity

In this section, we describe the use of a hybrid FSO-RF system with adaptive modulation and coding (AMC) as an efficient
way to deal with strong atmospheric turbulence. For FSO only channels’ performance an interested reader is referred to
[52–54]. Adaptive modulation and coding [2, 46, 47] can enable robust and spectrally efficient transmission over both α-μ
(or generalized gamma) wireless fading channel and FSO channel. To enable high-speed communication over atmospheric
turbulence channels, we describe a scheme in which an LDPC-coded sequence is partially over FSO channel and partially
over wireless channel. The channel state information of both channels is forwarded to the transmitters by an RF feedback
channel. The transmitters then adapt powers and rates so that the total channel capacity is maximized. The AMC concept has
been initially introduced for wireless communications [10, 48].

4.6.1 Hybrid FSO-RF System Model Description

We first describe the hybrid system, wireless channel model, and FSO channel model. The adaptive hybrid FSO-RF
communication system, shown in Fig. 4.21, consists of two parallel FSO and RF channels. The encoded data stream is
partially transmitted over FSO and partially over RF channel. Operating symbol rate of FSO channel is commonly many times
higher than that of RF channel. FSO channel comprises an FSO transmitter, propagation path through the atmosphere, and an
FSO receiver. The optical transmitter includes a semiconductor laser of high launch power, adaptive mapper, and power
control block. To reduce the system cost, the direct modulation of laser diode is used.
The modulated beam is projected toward the distant receiver by using an expanding telescope assembly. Along the
propagation path through the atmosphere, the light beam experiences absorption, scattering, and atmospheric turbulence,
which cause attenuation and random variations in amplitude and phase. At the receiver side, an optical system collects the
incoming light and focuses it onto a detector, which generates an electrical current proportional to the incoming power. The
RF channel comprises adaptive RF mapper, RF power control, RF transmitter (Tx), transmitting antenna, wireless propagation
path, receiver antenna, and RF receiver (Rx).
The RF channel estimates and FSO irradiance estimates are transmitted back to transmitters using the same RF feedback
channel. Because the atmospheric turbulence changes slowly, with correlation time ranging from 10 μs to 10 ms, this is a
plausible scenario for FSO channels with data rates in the order of Gb/s. Notice that erbium-doped fiber amplifiers (EDFAs)
cannot be used at all in this scenario because the fluorescence time is too long (about 10 ms). The semiconductor optical
amplifiers (SOAs) should be used instead, if needed. The data rates and powers in both channels are varied in accordance with
channel conditions. The symbol rates on both channels are kept fixed, while the signal constellation diagram sizes are varied
based on channel conditions. When FSO (RF) channel condition is favorable, larger constellation size is used; when FSO
(RF) channel condition is poor, smaller constellation size is used; and when the FSO (RF) channel signal-to-noise ratio (SNR)
falls below threshold, the signal is not transmitted at all. Both subsystems (FSO and RF) are designed to achieve the same
target bit error probability (Pb). The RF subsystem employs M-ary quadrature amplitude modulation (MQAM), while the FSO
subsystem employs the M-ary pulse amplitude modulation (MPAM). MPAM is selected for the FSO subsystem because
negative amplitude signals cannot be transmitted over FSO channels with direct detection. These two modulation formats are
selected as an illustrative example; the proposed scheme, however, is applicable to arbitrary multilevel modulations. The
optimum variable-power variable-rate policy, maximizing total channel capacity, is described in the next subsection. In the
rest of this section, we describe the wireless and the FSO channel models. Both RF and FSO channels are modeled as block-

RF Channel Extrinsic LLRs

RF mapper Estimate
Buffer RF Tx RF Rx
Input & modulator Output
Expanding
data LDPC telescope
ADC Buffer
P/S
APP LDPC data
S/P
encoder Detector ADC Buffer demapper decoder
Power
Buffer Mapper
control MUX
FSO Channel
Compressing Estimate
RF Power Direct modulated
laser diode telescope
control
RF Feedback
channel

Fig. 4.21 Hybrid FSO-RF system model. S/P serial-to-parallel conversion, LD laser diode, ADC A/D converter, P/S parallel-to-serial converter,
APP a posteriori probability
236 4 Capacities of Wireless and Optical Channels

fading channels, because symbol durations in both channels are much shorter than the corresponding coherence times.
The duration of the block is proportional to the coherence time of the channel, which is from 10 μs to 10 ms for FSO channel
and from 10 ms to 1 s for RF channels. We therefore assume the i.i.d. fading between the blocks.
The signal at the RF receiver antenna (see Fig. 4.21), denoted by sk, can be written as

jγ k
sk ¼ r k e ck þ z k , ð4:91Þ

where rk is the fading envelope for the kth transmission block and γ k is the random phase shift occurred during signal
transmission of the same block of symbols over a fading channel. The fading at antenna is frequency nonselective, it does not
change for the duration of block (slow fading), and it is independent from block to block. The ck is the vector of transmitted
symbols within the kth block. The zk denotes the block of zero-mean circular Gaussian noise samples corresponding to
block k.
The corresponding α-μ PDF is given by [49].


αμμ r αμ1
t r αt
f ðr t Þ ¼ αμ exp μ α , ð4:92Þ
br ΓðμÞ br

where Γ(.) is the gamma function. In Eq. (4.92), μ > 0 denotes the inverse of the normalized variance of r αt , while br is a α-root
mean value, and they are defined, respectively, as

 2   qffiffiffiffiffiffiffiffiffi
 
μ ¼ r αt =Var r αt , br ¼
α
r αt : ð4:93Þ

The α-μ fading model is employed because Rayleigh, Nakagami-m, exponential, Weibull, and one-sided Gaussian distribu-
tion functions are all special cases of this model. For example, by setting α ¼ 2 and μ ¼ 1, we obtain the Rayleigh distribution,
while by setting α ¼ 2 and μ ¼ 2, we obtain Nakagami m ¼ 2 distribution.
The FSO communication channel model is described by

yk ¼ R i k xk þ n k , ð4:94Þ

where xk is the kth transmitted block, ik is the instantaneous intensity gain, yk is the kth received block of symbols, nk is the
vector of additive white Gaussian noise (AWGN) samples with a normal distribution N(0,σ2) representing the trans-
impedance amplifier thermal noise, and R denotes the photodiode responsivity. (Without the loss of generality in the rest
of the chapter, we will set R ¼ 1 A/W.) (All signals in (4.94) are real valued.)
Several probability density functions (PDFs) have been proposed for the intensity variations at the receiver side of an FSO
link [50, 51]. For example, Al-Habash et al. [50, 51] proposed a statistical model that factorizes the irradiance as the product of
two independent random processes each with a gamma PDF. It was shown in [51] that predicted distribution matches very
well the distributions obtained from numerical propagation simulations and experiments, and as such is adopted here. The
PDF of the intensity fluctuation is given by [50].

ðα0 þβ0 Þ=2 pffiffiffiffiffiffiffiffiffiffiffi

2ðα0 β0 Þ ðα0 þβ0 Þ=21
f ðik Þ ¼ 0 0 ik K α0 β0 ð2 α0 β0 ik Þ, ik > 0, ð4:95Þ
Γðα ÞΓðβ Þ

where ik (k  0) is the signal intensity, α0 and β0 are parameters of the PDF, and Kα0  β0 is the modified Bessel function of the
second kind of order α0  β0 . The parameters α0 and β0 are related to the scintillation and in the case of zero inner scale (l0 ¼ 0)
(for plane waves) are given by [50, 51].

1 1
α0 ¼ " # , β0 ¼ " # , ð4:96Þ
0:49σ 2R 0:51σ 2R
exp 7=6 1 exp 5=6 1
ð1þ1:11σ12=5
R Þ ð1þ0:69σ 12=5
R Þ

where σ R2 is the Rytov variance [51].

4.6 Hybrid Free-Space Optical (FSO)-RF Channel Capacity 237

σ 2R ¼ 1:23 C 2n k7=6 L11=6 , ð4:97Þ

where k ¼ 2π/λ (λ is the wavelength), L denotes the propagation distance, and Cn2 is the refractive index structure parameter.
The Rytov variance is an excellent parameter to be used to describe the turbulence strength, because it takes the refractive
structure parameter, the propagation distance, and the operating wavelength into account. Weak fluctuations are associated
with σ2R < 1, the moderate with σ2R
1, the strong with σ2R > 1, and the saturation regime is defined by σ2R ! 1 [51].

4.6.2 Adaptive Modulation and Coding (AMC) in Hybrid FSO-RF Communications

There are many parameters that can be varied at the transmitter side relative to the FSO (RF) channel conditions, including
data rate, power, coding rate, and combinations of different adaptation parameters. The transmitter power adaptation can be
used to compensate for SNR variation due to atmospheric turbulence/fading, with the aim to maintain a desired BER in both
FSO and RF channels. The power adaptation therefore “inverts” the FSO channel scintillation and fading in wireless channel
so that both FSO and RF channels behave similarly as an AWGN channel. 2  The FSOFSO channel, upon channel inversion, appears
to the receiver as standard AWGN channel with SNRFSO ¼ ΓFSO 0 = 1=i t , where Γ 0 ¼ E FSOs =N 0 is the signal-to-noise ratio
in the absence of scintillation, with E s being the symbol energy and N0/2 being the double-side power spectral density of
FSO

AWGN related to variance by σ 2 ¼ N0/2. Notice that this definition [36, 47, 51], commonly used in wireless communications
[10–13], is different from [51] where SNR is defined as Po/σ, where h it i  Po. The  wireless
 fading channel, upon channel
inversion, appears to the receiver as standard AWGN channel with SNRRF ¼ ΓRF 0 = 1=h 2
, where h is the channel coefficient
and Γ0 ¼ E s =N 0 is the signal-to-noise ratio in the absence of fading, with Es being the symbol energy and N0/2 being the
RF RF RF

double-side power spectral density of AWGN.

In the rest of this section, we derive the optimum power adaptation policy that maximizes the total channel capacity [2, 35,
47]. We further derive the rate adaptation policy but assuming that symbol rates in both channels are fixed, and the symbol rate
in RF channel is much smaller than that in FSO channel. As an illustrative example, we assume that MQAM is used in RF
channel, while MPAM is used in FSO channel, and determine the spectral efficiency. MPAM is selected because negative
amplitude signals cannot be transmitted over FSO channels with direct detection. The MQAM is not power efficient for
transmission over FSO channel with direct detection because it requires the addition of DC bias to convert negative
amplitudes to positive ones, and as such is not considered here. Notice that M-ary pulse-position modulation (MPPM) can
also be used for transmission over FSO channel. Because MPPM is highly spectrally inefficient, we restrict our attention to
MPAM instead.
Before we continue with the description of different adaptation policies, we have to derive a target bit error probability
equation Pb for MAPM and MQAM on an AWGN channel. In MPAM, the transmitted signal x takes values from the discrete
set X ¼ {0, d, . . ., (M  1)d} (M  2), where d is the Euclidean distance between two neighboring points. If all signal
constellation points are equiprobable, the average signal energy is given by Es ¼ d2(M  1)(2 M  1)/6, and it is related to the
bit energy Eb by Es ¼ Eblog2M, so that signal-to-noise ratio per bit is defined by SNRb ¼ Eb/N0 ¼ ES/(N0log2M ) ¼ d2(M  1)
(2 M  1)/(6N0log2M ). Following the derivation similar to that reported in [10], the following expression for bit error
probability of FSO channel has been derived in [35, 47]:
0sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1
M1 3ΓMPAM
PMPAM
b ffi erfc@ 0 A, ð4:98Þ
M log 2 M 2ðM  1Þð2M  1Þ

where the symbolÐ SNR ΓMPAM is the symbol SNR in the absence of scintillation and the erfc(z) function is defined by
pffiffiffi 1 0
erfcðzÞ ¼ ð2= π Þ z exp ðu Þdu: Because Eq. (4.98) is not invertible, we derive the following empirical formula which is
2

valid in the regime of medium and high signal-to-noise ratios [35, 47]:
 
1:85ΓMPAM
PMPAM
b ffi 0:2 exp  0
: ð4:99Þ
22:19 log 2 M  1
238 4 Capacities of Wireless and Optical Channels

The corresponding expressions for MQAM are the same as already provided in [10]. In derivations that follow, we will
assume that target bit error probabilities in both channels are the same PMPAM
b ¼ PMQAM
b ¼ Pb . The total spectral efficiency
R as the function of bit error probability Pb can be found as [35, 47].

1


R¼ B log 2 1 þ K FSO ΓFSO þ BRF log 2 1 þ K RF ΓRF
0 , ð4:100Þ
2:19 FSO 0

where KFSO ¼  1.85\ ln (5Pb) and KRF ¼  1.5\ ln (5Pb). Because the total spectral efficiency changes as the channel
conditions in either channel change, the spectral efficiency is a function of FSO channel irradiance ik and RF fading coefficient
h as follows [35, 47]:



1 PFSO ðik Þ PRF ðhÞ
R¼ BFSO log 2 1 þ K FSO ΓFSO ðik Þ þ BRF log 2 1 þ K RF ΓRF ðhÞ , ð4:101Þ
2:19 P P

where ΓFSO ðik Þ ¼ i2k ΓFSO

0 and ΓRF ðhÞ ¼ h2 ΓRF0 . To derive the optimum power adaptation policy, subject to
P (ik) + P (h)  P, we have to define the corresponding Lagrangian, differentiate it with respect to PFSO(ik) and
FSO RF

PRF(h), and set corresponding derivatives to be equal to zero. The optimum power adaptation policy is obtained as the result
of this derivation [35, 47]:
8
< 1  1 , ΓFSO  Γtsh
K FSO P ðik Þ
FSO
¼ Γtsh Γ
FSO
, ΓFSO ¼ ΓFSO
0 ik
2
P :
0, ΓFSO < Γtsh
8 , ð4:102Þ
< 1  1 , ΓRF  Γtsh
K RF PRF ðhÞ
¼ Γtsh Γ
RF
, ΓRF ¼ ΓRF
0 h
2
P :
0, ΓRF < Γtsh

where Γtsh is the threshold SNR, which is common to both channels. With this adaptation policy, more power and higher data
rates are transmitted when the FSO (RF) channel conditions are good, less power and lower data rates are transmitted when
FSO (RF) channel is bad, and nothing is transmitted when the SNR falls below the threshold Γtsh. The optimum threshold Γtsh
can be obtained numerically by solving the following equation [35, 47]:

ð
1
 1ð !
1 1 1 1
 f ð ik Þdik þ b  f ðhÞdh ¼ 1, ð4:103Þ
K FSO Γtsh K FSO ΓFSO
0 ik
2 K RF Γtsh K RF ΓRF
0 h
2
pffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi pffiffiffiffiffiffiffiffiffiffiffiffi
Γtsh =Γ0FSO RF
Γtsh =Γ0

where b ¼ BRF/BFSO, f(it) is the PDF of FSO irradiance ik given by Eq. (4.95), and f(h) is the PDF of RF channel coefficient
h given by Eq. (4.92). The optimum spectral efficiency, defined as data rate R over channel bandwidth B, can be evaluated by
substituting Eq. (4.102) into Eq. (4.101) to obtain [35, 47].

ð
1
 ð
1
2
R 1 ΓFSO
0 ik
2
ΓRF
0 h
¼ log 2 f ðik Þdik þb log 2 f ðhÞdh ½bits=s=Hz: ð4:104Þ
B 2:19 Γtsh Γtsh
pffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi pffiffiffiffiffiffiffiffiffiffiffiffi
Γtsh =Γ0
FSO
Γtsh =Γ0
RF

Although this adaptive-rate adaptive-power scheme provides excellent spectral efficiencies, the optimum threshold
computation in Eq. (4.103) is time extensive. Instead, we can perform the truncated channel inversion with fixed rate. The
truncated channel inversion adaptation can be performed by [35, 47]
4.6 Hybrid Free-Space Optical (FSO)-RF Channel Capacity 239

8
>
<
1
 2  , ΓFSO  Γtsh
PFSO ðik Þ i E
2 FSO
¼ k Γtsh 1=ik
P >
:
0, ΓFSO < itsh
8 , ð4:105Þ
>
< 2 RF 
1
 , Γ RF
 Γ
P ð hÞ
RF tsh
¼ h E Γtsh 1=h
2
P >
:
0, ΓRF < Γtsh

where

 FSO ð
1  RF 1ð
1 f ði k Þ 1 f ðhÞ
¼ di k , ¼ dh: ð4:106Þ
i2k Γtsh pffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi i 2
k h 2
Γtsh p ffiffiffiffiffiffiffiffiffiffiffiffi h2
Γtsh =Γ0
FSO
Γtsh =Γ0
RF

The threshold Γtsh in Eq. (4.106) is obtained by maximizing the spectral efficiency as given below [35, 47]:
8 0 1
<
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 @ 1 A
R ¼ max BFSO log 2 1 þ K FSO Γ0 
FSO
FSO P ik  Γtsh =Γ0
FSO
Γtsh :2:19 1=i2k Γtsh
0 1 9 ð4:107Þ

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi=
1
þBRF log 2 @1 þ K RF ΓRF
0  RF AP h  Γtsh =Γ0
RF
,
1=h2 Γtsh ;

where

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð
1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð
1

P ik  Γtsh =ΓFSO 0 ¼ f ðik Þdik , P h  Γtsh =ΓRF 0 ¼ f ðhÞdh: ð4:108Þ

qffiffiffiffiffiffiffi qffiffiffiffiffi
Γtsh Γtsh
ΓFSO ΓRF
0 0

In Fig. 4.22, we show the spectral efficiencies for FSO system only, which can be achieved using the optimum power and
rate adaptation and MPAM for different target bit error probabilities, and both (a) weak turbulence regime (σ R ¼ 0.2,
α0 ¼ 51.913, β0 ¼ 49.113) and (b) strong turbulence regime (σ R ¼ 2, α0 ¼ 4.3407, β0 ¼ 1.3088). For example, the spectral
efficiency R/B of 2 bits/s/Hz at Pb ¼ 109 is achieved for symbol SNR of 23.3 dB in weak turbulence regime and 26.2 dB in
strong turbulence regime. In the same figure, we report the spectral efficiencies that can be achieved by both channel inversion
(Γtsh ¼ 0) and truncated channel inversion (Γtsh > 0). In the weak turbulence regime (see Fig. 4.24(a)), even simple channel
inversion performs comparable to an optimum adaptive-power adaptive-rate scheme. However, in the strong turbulence
regime (see Fig. 4.24(b)), this scheme faces significant performance degradation. On the other hand, the truncated channel
inversion scheme in strong turbulence regime faces moderate performance degradation, about 3.7 dB at Pb ¼ 109 for a
spectral efficiency of 2 bits/s/Hz. The optimum adaptation policy for FSO channel at Pb ¼ 106 for a spectral efficiency of
4 bits/s/Hz provides a moderate improvement of 3.3 dB in the weak turbulence regime over a nonadaptive scheme, while the
improvement in the strong turbulence regime is significant at 31.7 dB.
In Fig. 4.23, we report the spectral efficiencies for the hybrid FSO-RF system shown in Fig. 4.21, with RF subsystem
fading parameters α ¼ 3, μ ¼ 2 in both weak turbulence regime [Fig. 4.23(a)] and strong turbulence regime [Fig. 4.23(b)]. We
assume that the FSO subsystem symbol rate is ten times larger than the RF subsystem data rate, that is, b ¼ 0.1. For a spectral
efficiency of 2 bits/s/Hz, the hybrid FSO-RF system outperforms the FSO system by 3.39 dB at a BER of 106 and 3.49 dB at
a BER of 109. It is interesting to notice that even truncated channel inversion for the hybrid system outperforms the optimum
adaptation of the FSO system by 0.8 dB at a BER of 109 and a spectral efficiency of 2 bits/s/Hz.
In Fig. 4.24, we report the spectral efficiencies for the hybrid FSO-RF system, with RF subsystem fading parameters α ¼ 2,
μ ¼ 1 (corresponding to Rayleigh fading) in both weak turbulence regime [Fig. 4.24(a)] and strong turbulence regime
[Fig. 4.24(b)]. This case corresponds to the situation where there is no line of site between transmit and receive antennas for
240 4 Capacities of Wireless and Optical Channels

-3
Pb=10 :
Variable-rate variable-power -3
Channel inversion Pb=10 :
Truncated channel inversion Variable-rate variable-power
-6
Pb=10 : Channel inversion
Truncated channel inversion
Variable-rate variable-power -6
Channel inversion Pb=10 :
Truncated channel inversion Variable-rate variable-power
-9
Pb=10 : Channel inversion
Truncated channel inversion
Variable-rate variable-power -9
Channel inversion Pb=10 :
Spectral efficiency, R/B [bits/s/Hz]

4
Variable-rate variable-power

Spectral efficiency, R/B [bits/s/Hz]

Truncated channel inversion
Channel inversion
3 Truncated channel inversion
3

2
=2.0
R
2

1
1

=0.2
R

0
10 15 20 25 10 15 20 25
Signal-to-noise ratio, Es/N0 [dB] Signal-to-noise ratio, Es/N0 [dB]

(a) (b)
Fig. 4.22 Spectral efficiencies of an FSO system against symbol SNR for different target bit probabilities of error: (a) in weak turbulence regime
and (b) in strong turbulence regime

the RF subsystem. We again assume that b ¼ 0.1. For a spectral efficiency of 2 bits/s/Hz, the hybrid FSO-RF system
outperforms the FSO system by 3.01 dB at a BER of 106 and 3.11 dB at a BER of 109. The truncated channel inversion for
the hybrid system performs comparable to the optimum adaptation of the FSO system.
In this section, we described two different adaptive modulation scenarios for both the FSO system with RF feedback and
the hybrid FSO-RF system. In the next section, we describe the adaptive coding based on coded modulation.
By using the trellis-coded modulation (TCM) or cosset codes, we can separate the encoding and modulation process (see
[10, 48] for more details). However, to keep the complexity of this approach reasonably low, the convolutional or block codes
should be simple and short. Those codes are in principle of low rate and weak so that coding gains are moderate. For example,
the adaptive coding scheme based on the TCM proposed in [48] is about 5 dB away from channel capacity. Instead, in this
chapter, we propose to implement adaptive coding based on LDPC-coded modulation. For the FSO system, the input data are
LDPC encoded and written to a buffer. Based on FSO channel irradiance, it, log2M(it) bits are taken at a time from a buffer and
used to select the corresponding point from MPAM signal constellation. For the hybrid FSO-RF system, the LDPC-encoded
sequence is split between FSO and RF subsystems (see Fig. 4.21). To facilitate the implementation, we assume that symbol
rates in both subsystems are fixed, while constellation sizes and emitted powers are determined based on channel conditions in
both channels using the adaptation scenarios described in the previous section. We further assume that the symbol rate in the
RF subsystem is at least ten times lower than that in the FSO subsystem (e.g., 1 giga symbol/s in the FSO subsystem and
100 mega symbol/s in the RF subsystem).
4.6 Hybrid Free-Space Optical (FSO)-RF Channel Capacity 241

-3
-3 Pb=10 :
Pb=10 :
Variable-rate variable-power
Variable-rate variable-power
Truncated channel inversion
Truncated channel inversion -6
-6
Pb=10 :
Pb=10 :
Variable-rate variable-power Variable-rate variable-power
Truncated channel inversion Truncated channel inversion
-9
-9
Pb=10 : 4 Pb=10 :

Spectral efficiency, R/B [bits/s/Hz]

Variable-rate variable-power Variable-rate variable-power
Spectral efficiency, R/B [bits/s/Hz]

4 Truncated channel inversion Truncated channel inversion

3

2
2

1
1
=0.2; =3, =2 R
=2.0; =3, =2
R

4 8 12 16 20 24 28 4 8 12 16 20 24 28
Signal-to-noise ratio, Es/N0 [dB] Signal-to-noise ratio, Es/N0 [dB]

(a) (b)
Fig. 4.23 Spectral efficiencies of a hybrid FSO-RF system with α ¼ 3, μ ¼ 2 fading against symbol SNR for different target bit probabilities of
error: (a) in weak turbulence regime and (b) in strong turbulence regime

In Fig. 4.25(a), we show the R/B performance of the FSO system with adaptive LDPC-coded MPAM for different
adaptation scenarios. Given the fact that the channel capacity of FSO channel under atmospheric turbulence is an open
problem, we show in the same figure an upper bound in the absence of atmospheric turbulence from [54]. The coding gain
over adaptive modulation at Pb ¼ 106 for R/B ¼ 4 bits/s/Hz is 7.2 dB in both (weak and strong) turbulence regimes. Larger
coding gains are expected at lower BERs and for higher spectral efficiencies. Further improvements can be obtained by
increasing the girth of LDPC codes and employing better modulation formats. The increase in code word length to 100,515
does not improve the R/B performance that much as shown in Fig. 4.25(a). It is interesting to notice that by employing
adaptive coding, the communication under saturation regime is possible, as shown in Fig. 4.25(a). Moreover, for a variable-
rate variable-power scheme, there is no degradation in saturation regime compared to strong turbulence regime. Overall
improvement from adaptive modulation and coding for R/B ¼ 4 bits/s/Hz at Pb ¼ 106 over nonadaptive uncoded modulation
ranges from 10.5 dB (3.3 dB from adaptive modulation and 7.2 dB from coding) in the weak turbulence regime to 38.9 dB in
the strong turbulence regime (31.7 dB from adaptive modulation and 7.2 dB from coding). In Fig. 4.25(b), we show the R/B
performance of the hybrid FSO-RF system with adaptive LDPC-coded modulation (MPAM is used in the FSO subsystem and
MQAM in the RF subsystem) for different adaptation scenarios. The symbol rate in the FSO subsystem is set to be ten times
larger than that in the RF subsystem (b ¼ 0.1). For a spectral efficiency of 4 bits/s/Hz at a BER of 106, the improvement of
the hybrid FSO-RF system over the FSO system is 5.25 dB in Rayleigh fading (α ¼ 2, μ ¼ 1), 5.51 dB in Nakagami m ¼ 2
fading (α ¼ 2, μ ¼ 2), and 5.63 dB in α ¼3, μ ¼ 2 fading. For a spectral efficiency of 2 bits/s/Hz at the same BER, the
improvement of the hybrid FSO-RF system over the FSO system is 3.32 dB in Rayleigh fading, 3.72 dB in Nakagami m ¼ 2
fading, and 3.86 dB in α ¼ 3, μ ¼ 2 fading.
242 4 Capacities of Wireless and Optical Channels

-3
Pb=10 : -3
Pb=10 :
Variable-rate variable-power
Variable-rate variable-power
Truncated channel inversion
-6 Truncated channel inversion
Pb=10 : -6
Pb=10 :
Variable-rate variable-power
Truncated channel inversion Variable-rate variable-power
-9 Truncated channel inversion
Pb=10 : -9
4 Pb=10 :

Spectral efficiency, R/B [bits/s/Hz]

Variable-rate variable-power
Spectral efficiency, R/B [bits/s/Hz]

Truncated channel inversion Variable-rate variable-power

4
Truncated channel inversion
3
3

2
2

1 1
=0.2; =2, =1 R
=2.0; =2, =1
R
(Rayleigh) (Rayleigh)
0 0
5 10 15 20 25 5 10 15 20 25
Signal-to-noise ratio, Es/N0 [dB] Signal-to-noise ratio, Es/N0 [dB]

(a) (b)
Fig. 4.24 Spectral efficiencies of a hybrid FSO-RF system with α ¼ 2, μ ¼ 1 (Rayleigh) fading against symbol SNR for different target bit
probabilities of error: (a) in weak turbulence regime and (b) in strong turbulence regime

4.7 Concluding Remarks

This chapter has been devoted to wireless channel and optical channel capacities. The chapter starts with definitions of mutual
information and channel capacity for discrete memoryless channels in Sect. 4.1.1, followed by the capacity of continuous
channels and information capacity theorem in Sect. 4.1.2. Regarding the wireless channel capacity, described in Sect. 4.2, we
describe how to calculate the channel capacity of flat-fading channels in Sect. 4.2.1 and frequency-selective channels in Sect.
4.2.2. We also discuss different optimum and suboptimum strategies to achieve channel capacity including the water-filling
method, multiplexed coding and decoding, channel inversion, and truncated channel inversion. Depending on what is known
about the channel state information (CSI), different strategies are described for channel capacity evaluation. When CSI is
known on the receiver side only (CSI), we describe the Shannon (ergodic) channel capacity and capacity with outage in Sect.
4.2.1.1. On the other hand, when CSI is available on both transmitter and receiver sides (full CSI), we describe the Shannon
capacity, zero-outage capacity, and outage capacity in Sect. 4.2.1.2. Regarding the frequency-selective fading channel
capacity, described in Sect. 4.2.2, we describe how to calculate the channel capacity for both time-invariant channels (Sect.
4.2.2.1) and time-variant channels (Sect. 4.2.2.2). Additionally, we describe the optimum power adaption strategies in the
same sections. For instance, the optimum power adaptation strategy for time-variant fading channels is the two-dimensional
water-filling method in both time and frequency domains. Further, we explain how to model the channel with memory and
describe how to calculate its capacity in Sect. 4.3. The key topics of this section include Markov sources (Sect. 4.3.1),
McMillan sources (4.3.2), together with corresponding entropies, followed by the McMillan-Khinchin model for channel
capacity evaluation (Sect. 4.3.3). We then describe in Sect. 4.4 how to employ the forward recursion of the BCJR algorithm to
evaluate the information capacity. The next topic, described in Sect. 4.5, is related to the evaluation of the information
capacity of fiber-optic communication channels with coherent optical detection. Finally, in Sect. 4.6, we describe how to
evaluate the capacity of hybrid FSO-RF channels, with a hybrid FSO-RF system model being described in Sect. 4.6.1 and
AMC for hybrid FSO-RF communication being described in Sect. 4.6.2. The set of problems is provided in the incoming
section.
4.8 Problems 243

Hybrid FSO-RF system

-6
LDPC(16935,13550)-coded, Pb=10
Variable-rate variable-power:
FSO with RF feedback: =2, =1 (Rayleigh):
=0.2: R
=0.2 R
=100 =2.0
R
R
Variable-rate variable-power =2, =2 (Nakagami m=2):
Truncated channel inversion R
=0.2 R
=100

R
=2.0: =3, =2:
Variable-rate variable-power =0.2
R
=100
R

Channel inversion Truncated channel inversion:

Truncated channel inversion =2, =1 (Rayleigh):
R
=100.0: R
=0.2 R
=100
Variable-rate variable-power =3, =2:
LDPC(16935,13550) =100
R
5

Spectral efficiency, R/B [bits/s/Hz]

Spectral efficiency, R/B [bits/s/Hz]

LDPC(100515,80412)
Channel inversion 5

4
4
Upper FSO bound
3 (without scintillation)
3

2
2

1
-6 1
Pb=10
0
0 5 10 15 20 25 5 10 15 20 25
Signal-to-noise ratio, Es/N0 [dB]
Signal-to-noise ratio, Es/N0 [dB]

(a) (b)
Fig. 4.25 Spectral efficiencies against symbol SNR for adaptive LDPC-coded modulation: (a) FSO with RF feedback only and (b) hybrid FSO-RF
system

4.8 Problems

1. Let a 2  2 binary asymmetric channel matrix be defined as follows:

 
Pðy0 jx0 Þ Pðy1 jx0 Þ
Π¼ ,
Pðy0 jx1 Þ Pðy1 jx1 Þ

where xi (i ¼ 0,1) are channel inputs and yi (i ¼ 0,1) are channel outputs. A digital transmitter sends a binary sequence over
a cascade of two binary asymmetric channels, characterized by corresponding channel matrices Π1 and Π2, respectively.
The known elements of the first channel matrix are P1(0|0) ¼ 0.8 and P1(1|1) ¼ 0.9. On the other hand, the known elements
of the second channel matrix are P2(0|0) ¼ 0.7 and P2(1|1) ¼ 0.9. The a priori probabilities for transmitting 0 and 1 are 0.6
and 0.4, respectively. Determine the overall channel matrix of the cascade of these two binary asymmetric channels.
Determine a posteriori probabilities of receiving 0 and 1 at the output of the channel cascade. Determine the average error
probability at the output of the channel cascade. Determine the mutual information of the channel cascade.
244 4 Capacities of Wireless and Optical Channels

2. The binary erasure channel (BEC) has two inputs and three outputs as described in Fig. P2.
The inputs are labeled as 0 and 1, and the outputs as 0, 1, and e. A fraction f of the incoming bits are erased by the channel.
Find the capacity of the channel.

1f
0 0
f

e
f

1 1
1f

Fig. P2 The binary erasure channel transition probabilities

3. For the channel shown in Fig. P3, determine the input distribution achieving the channel capacity. Evaluate channel
capacities for ε ¼ 0, 0.5, and 1. When this channel is cascaded n times, what would be the channel capacity of such
cascaded channel? Finally, in the limit n ! 1, determine the capacity of the cascaded channel.

1
x1 y1

1H

H
x2 y2
Fig. P3 The Z-channel transition probabilities

4. Let us consider the M-ary symmetric channel with M > 2, which represents the generalization of the binary symmetric
channel (BSC). The corresponding channel matrix is given by.
2 3
v p=ðM  1Þ p=ðM  1Þ ⋯ p=ðM  1Þ
6 p=ðM  1Þ v p=ðM  1Þ p=ðM  1Þ p=ðM  1Þ 7
6 7
PMSC ¼6 7, p þ v ¼ 1:
4 ⋮ ⋮ ⋮ ⋱ ⋮ 5
p=ðM  1Þ p=ðM  1Þ p=ðM  1Þ p=ðM  1Þ v

Determine the channel capacity as a function of M.

5. The binary symmetric channel (BSC) is described by the following channel matrix:
 
v p
PBSC ¼ , p þ v ¼ 1:
p v
Determine the channel matrices for two and three BSC channels serially connected (in a cascade). Determine the channel
capacity for the cascade of two BSCs and cascade of three BSCs.
6. Let us consider the time-invariant frequency-selective block-fading channel with three subchannels of bandwidth 2 MHz,
with frequency response amplitudes being 2, 4, and 6, respectively. For the transmit power 10 mW and noise power
spectral density of N0 ¼ 109 W/Hz, determine the corresponding Shannon capacity. What is the corresponding optimum
power allocation strategy to achieve this capacity?
7. Let us consider the communication over Nakagami-m channel of bandwidth 20 MHz. Determine the distribution of SNR
for this channel. Assuming that the channel state information is available at both transmitter and receiver sides (full CSI)
and the SNR mean is 12 dB, solve problems (a)–(d).
4.8 Problems 245

(a) Determine the optimum power adaptation policy achieving the Shannon capacity.
(b) Determine the Shannon capacities for the following values of parameter m ¼ 0.5, 1, and m ¼ (K + 1)2/(2 K + 1), with
K being 20 dB (convert to linear scale first). Discuss the results. How far we are from the AWGN channel capacity of
the same SNR for each of the cases?
(c) Repeat (b) but now assuming that only the receiver knows the CSI. Discuss the degradation compared to (b).
(d) Determine the zero-outage capacity and outage capacity, assuming that the outage probability is 0.02 for the same
cases as in (b). Discuss the degradation compared to (b).
8. In this problem, we study the frequency-selective fading channel with a total available bandwidth of 20 MHz and
corresponding coherence bandwidth being 5 MHz. Divide the total bandwidth into four subchannels with corresponding
bandwidth being equal to the coherence bandwidth and assume that each subchannel is a Nakagami-m flat-fading channel
with independent realizations for each subchannel. Assume the subchannels have average gains h|H1[i]|2i ¼ 1, h|
H2[i]|2i ¼ 0.75, h|H3[i]|2i ¼ 0.5, and h|H4[i]|2i ¼ 0.25. Assume a total transmit power of 50 mW and a receiver noise
spectral density of 0.002 μW/Hz. For m ¼ 0.5, 1, and 2, solve for (a) and (b).
(a) Determine the optimal two-dimensional water-filling power adaptation for this channel and the corresponding
Shannon capacity, assuming both the transmitter and the receiver know the instantaneous value of Hj[i], j ¼ 1, . . ., 4.
(b) Allocate now the equal average power to each subchannel and then perform water filling for each subchannel.
Compare the overall capacity against that in (a).
9. Let us now consider the time-invariant frequency-selective block-fading channel with 32 subchannels of bandwidth
2 MHz, with frequency response amplitudes being 1 for odd subchannels and √2 for even subchannels, respectively. For
the transmit power 10 mW and noise power spectral density of N0 ¼ 109 W/Hz, determine the OFDM channel capacity.
What is the corresponding optimum power allocation strategy based on OFDM to achieve this channel capacity?
10. In this problem, we are concerned with scintillation effects, and the i.i.d. FSO channel model, in which the samples of
irradiance are generated by gamma-gamma distribution.
(a) For (binary) intensity modulation with direct detection (IM/DD) system, determine the average bit error probability
when operating over turbulent FSO channel.
(b) For coherent detection, determine the ergodic capacity, assuming that channel state information is known at either
receiver (Rx) side or both (Tx and Rx) sides.
(c) Determine the water-filling, channel inversion, and truncated channel inversion capacities, assuming that coherent
detection is used.
Pn Pn
ðk Þ ðk Þ
11. The entropy of Markov source for moving k steps ahead from initial states is given as H ðkÞ ¼  pi pij log pij :
i¼1 j¼1
Prove the following property: H(k + 1) ¼ H(k) + H(1).
12. For the Markov source from Fig. P10, determine:
(a) Stationary transition and state probabilities.
(b) Entropy of this source.

0.7

S1

0.5 0.3

0.5
S3 S2
0.5

0.3 0.5

S4

0.7

Fig. P10 A particular Markov source

246 4 Capacities of Wireless and Optical Channels

13. Reproduce the results shown in Fig. 4.15 by using Monte Carlo integration.
14. Implement the algorithm described in Sect. 4.4. Reproduce the results shown in Fig. 4.15 and compare them against those
from Problem 4.11.
15. Plot the i.i.d. channel capacity against optical SNR for different memories in the forward step of the BCJR algorithm, in
the presence of both residual chromatic dispersion (11,200 ps/nm) and PMD with a DGD of 50 ps by observing the NRZ
transmission system operating at 10 Gb/s.
16. Plot the i.i.d. channel capacity in the presence of intrachannel nonlinearities against the number of spans, for dispersion
map shown in Fig. 4.17, for different memories in the forward step of the BCJR algorithm. The span length is set to
L ¼ 120 km, and each span consists of 2 L/3 km of D+ fiber followed by L/3 km of D fiber. Pre-compensation of
1600 ps/nm and corresponding post-compensation are also applied. The parameters of D+ and D fibers, used in
simulations, are given in Table 4.1. Assume that the RZ modulation format of a duty cycle of 33% is observed, the
extinction ratio is 14 dB, and the launched power is set to 0 dBm. EDFAs with a noise figure of 6 dB are deployed after
every fiber section, the bandwidth of optical filter (modeled as super-Gaussian filter of eight order) is set to 3Rl, and the
bandwidth of electrical filter (modeled as Gaussian filter) to 0.7Rl, with Rl being the line rate (defined as the bit rate
divided by a code rate). Assume that the bit rate is 40 Gb/s.
17. Reproduce the results shown in Figs. 4.19 and 4.20.
18. Let us now study by the information capacity by using the dispersion map shown in 4.18(a), but now replacing EDFAs
with a hybrid Raman/EDFA scheme described in Problem 16 of Chap. 2. Study the information capacities in a fashion
similar to Fig. 4.20. Discuss the improvements with respect to Fig. 4.20.
19. Repeat Problem 18, but now by taking quantization effects into account. Observe the information capacities for the
following number of bits: 8, 6, 4, and 2. Discuss the results.
20. Let us now study by the information capacity by using the dispersion map shown in Fig. 4.18(a), but now by replacing
EDFAs with distributed Raman amplifiers, with typical parameters as described in Chap. 3. Study the information
capacities in a fashion similar to Fig. 4.20. Discuss the improvements with respect to Fig. 4.20 and Problem 18.
21. Plot the ergodic channel capacity against SNR for 2  2 MIMO over MMF, assuming that MMF can support 500 modes.
Assume that channel coefficients follow zero-mean Gaussian distribution. Study different strategies: CSIT, CSIR, and full
CSI. Compare the results against the SISO system. Hint. You can use the MMF model described in Sect. 8.2.1 of Chap. 8.
22. Plot the ergodic channel capacity against SNR for different MIMO systems over FMF, assuming that FMF can support
2, 4, and 8 modes. Assume that channel coefficients follow zero-mean Gaussian distribution. Study different strategies
(CSIT, CSIR, and full CSI). Compare the results against the SISO system. Hint. You can use the FMF model described in
Sect. 8.2.1 of Chap. 8.

References
1. I.B. Djordjevic, Advanced coding for optical communication systems, in Optical Fiber Telecommunications VI, ed. by I. Kaminow, T. Li,
A. Willner, (Elsevier/Academic Press, Boston, May 2013), pp. 221–296
2. I.B. Djordjevic, W. Ryan, B. Vasic, Coding for Optical Channels (Springer, New York/Dordrecht/Heidelberg/London, 2010)
3. M. Cvijetic, I.B. Djordjevic, Advanced Optical Communication Systems and Networks (Artech House, Norwood, MA, 2013)
4. I.B. Djordjevic, On advanced FEC and coded modulation for ultra-high-speed optical transmission. IEEE Communications Surveys and
Tutorials 18(3), 1920–1951, third quarter 2016 (Aug 19, 2016). https://doi.org/10.1109/COMST.2016.2536726
5. T.M. Cover, J.A. Thomas, Elements of Information Theory (Wiley, New York, 1991)
6. F.M. Reza, An Introduction to Information Theory (McGraw-Hill, New York, 1961)
7. R.G. Gallager, Information Theory and Reliable Communication (Wiley, New York, 1968)
8. F.M. Ingels, Information and Coding Theory (Intext Educational Publishers, Scranton, 1971)
9. S. Haykin, Communication Systems (Wiley, New York, 2004)
10. A. Goldsmith, Wireless Communications (Cambridge University Press, Cambridge, 2005)
11. D. Tse, P. Viswanath, Fundamentals of Wireless Communication (Cambridge University Press, Cambridge, 2005)
12. S. Benedetto, E. Biglieri, Principles of Digital Transmission with Wireless Applications (Kluwer Academic/Plenum Publishers, New York,
1999)
13. E. Biglieri, Coding for Wireless Channels (Springer, New York, 2005)
14. J. Wolfowitz, Coding Theorems of Information Theory, 2nd edn. (Springer, Berlin/New York, 1964)
15. T.M. Duman, A. Ghrayeb, Coding for MIMO Communication Systems (Wiley, Chichester, 2007)
16. J.G. Proakis, Digital Communications (McGraw-Hill, Boston, MA, 2001)
17. C.E. Shannon, A mathematical theory of communication. Bell Syst. Tech. J. 27, 379–423. 623–656 (July, Oct 1948)
18. B. McMillan, The basic theorems of information theory. Ann. Math. Statistics 24, 196–219 (1952)
References 247

19. A.I. Khinchin, Mathematical Foundations of Information Theory (Dover Publications, New York, 1957)
20. D. Arnold, A. Kavcic, H.-A. Loeliger, P. O. Vontobel, W. Zeng, Simulation-based computation of information rates: upper and lower bounds, in
Proceedings of the IEEE International symposium information theory (ISIT 2003), 2003, p. 119
21. D. Arnold, H.-A. Loeliger, On the information rate of binary-input channels with memory, in Proceedings of the 2001 International conference
communications, Helsinki, Finland, 11–14 June 2001, pp. 2692–2695
22. H.D. Pfitser, J. B. Soriaga, P. H. Siegel, On the achievable information rates of finite state ISI channels, in Proceedings of the Globecom 2001,
San Antonio, TX, 25–29 Nov 2001, pp. 2992–2996
23. I.B. Djordjevic, B. Vasic, M. Ivkovic, I. Gabitov, Achievable information rates for high-speed long-haul optical transmission. IEEE/OSA
J. Lightw. Technol. 23, 3755–3763 (Nov 2005)
24. M. Ivkovic, I.B. Djordjevic, B. Vasic, Calculation of achievable information rates of long-haul optical transmission systems using instanton
approach. IEEE/OSA J. Lightw. Technol. 25, 1163–1168 (May 2007)
25. M. Ivkovic, I. Djordjevic, P. Rajkovic, B. Vasic, Pulse energy probability density functions for long-haul optical fiber transmission systems by
using instantons and Edgeworth expansion. IEEE Photon. Technol. Lett. 19(20), 1604, 15 Oct 2007–1606
26. I.B. Djordjevic, N. Alic, G. Papen, S. Radic, Determination of achievable information rates (AIRs) of IM/DD systems and AIR loss due to
chromatic dispersion and quantization. IEEE Photon. Technol. Lett. 19(1), 12–14 (1 Jan 2007)
27. I.B. Djordjevic, L.L. Minkov, H.G. Batshon, Mitigation of linear and nonlinear impairments in high-speed optical networks by using LDPC-
coded turbo equalization. IEEE J. Sel. Areas Comm. Optical Comm. Netw. 26(6), 73–83 (Aug 2008)
28. I.B. Djordjevic, L.L. Minkov, L. Xu, T. Wang, Suppression of fiber nonlinearities and PMD in coded-modulation schemes with coherent
detection by using turbo equalization. IEEE/OSA J. Opt. Comm. Netw. 1(6), 555–564 (Nov 2009)
29. R.-J. Essiambre, G.J. Foschini, G. Kramer, P.J. Winzer, Capacity limits of information transport in fiber-optic networks. Phys. Rev. Lett. 101,
163901-1–163901-4 (17 Oct 2008)
30. R.-J. Essiambre, G. Kramer, P.J. Winzer, G.J. Foschini, B. Goebel, Capacity limits of optical fiber networks. J. Lightw. Technol. 28(4), 662–701
(Feb 2010)
31. D.P. Bertsekas, Nonlinear Programming, 2nd edn. (Athena Scientific, Belmont, 1999)
32. E.K.P. Chong, S.H. Zak, An Introduction to Optimization, 3rd edn. (Wiley, New York, 2008)
33. I.B. Djordjevic, H. G. Batshon, L. Xu, T. Wang, Coded polarization-multiplexed iterative polar modulation (PM-IPM) for beyond 400 Gb/s
serial optical transmission, in Proceedings of the OFC/NFOEC 2010, Paper No. OMK2, San Diego, CA, 21–25 Mar 2010
34. H.G. Batshon, I.B. Djordjevic, L. Xu, T. Wang, Iterative polar quantization based modulation to achieve channel capacity in ultra-high-speed
optical communication systems. IEEE Photon. J. 2(4), 593–599 (Aug 2010)
35. C. Lin, I.B. Djordjevic, D. Zou, Achievable information rates calculation for optical OFDM transmission over few-mode fiber long-haul
transmission systems. Opt. Express 23(13), 16846–16856 (2015)
36. J.M. Kahn, K.-P. Ho, Spectral efficiency limits and modulation/detection techniques for DWDM systems. IEEE Selected Topics Quantum
Electron. 10, 259–272 (Mar/Apr 2004)
37. K.S. Turitsyn, S.A. Derevyanko, I.V. Yurkevich, S.K. Turitsyn, Information capacity of optical fiber channels with zero average dispersion.
Phys. Rev. Lett. 91(20), 203–901 (Nov 2003)
38. J. Tang, The channel capacity of a multispan DWDM system employing dispersive nonlinear optical fibers and an ideal coherent optical
receiver. J. Lightw. Technol. 20, 1095–1101 (July 2002)
39. E.E. Narimanov, P. Mitra, The channel capacity of a fiber optics communication system: perturbation theory. J. Lightw. Technol. 20, 530–537
(Mar 2002)
40. A. Mecozzi, M. Shtaif, On the capacity of intensity modulated systems using optical amplifiers. IEEE Photon. Technol. Lett. 13, 1029–1031
(Sept 2001)
41. P.P. Mitra, J.B. Stark, Nonlinear limits to the information capacity of optical fiber communications. Nature 411, 1027–1030 (June 2001)
42. J. Tang, The Shannon channel capacity of dispersion-free nonlinear optical fiber transmission. J. Lightw. Technol. 19, 1104–1109 (Aug 2001)
43. J. Tang, The multispan effects of Kerr nonlinearity and amplifier noises on Shannon channel capacity for a dispersion-free nonlinear optical
fiber. J. Lightw. Technol. 19, 1110–1115 (Aug 2001)
44. K.-P. Ho, J. M. Kahn, Channel capacity of WDM systems using constant-intensity modulation formats, in Optical Fiber Communications
Conferences (OFC ‘02), 2002, paper ThGG85
45. J.M. Kahn, K.-P. Ho, Ultimate spectral efficiency limits in DWDM systems, in Optoelectronics communication conference, Yokohama, Japan,
8–12 July 2002
46. I.B. Djordjevic, G.T. Djordjevic, On the communication over strong atmospheric turbulence channels by adaptive modulation and coding. Opt.
Express 17(20), 18250–18262 (25 Sept 2009)
47. I.B. Djordjevic, Adaptive modulation and coding for free-space optical channels. IEEE/OSA J. Opt. Commun. Netw. 2(5), 221–229 (May 2010)
48. A. Goldsmith, S.-G. Chua, Adaptive coded modulation for fading channels. IEEE Trans. Comm. 46, 595–601 (May 1998)
49. M.D. Yacoub, The α-μ distribution: a physical fading model for the Stacy distribution. IEEE Trans. Vehic. Techn. 56(1), 27–34 (Jan 2007)
50. M.A. Al-Habash, L.C. Andrews, R.L. Phillips, Mathematical model for the irradiance probability density function of a laser beam propagating
through turbulent media. Opt. Eng. 40, 1554–1562 (2001)
51. L.C. Andrews, R.L. Philips, Laser Beam Propagation through Random Media (SPIE Press, 2005)
52. I.B. Djordjevic, S. Denic, J. Anguita, B. Vasic, M.A. Neifeld, LDPC-coded MIMO optical communication over the atmospheric turbulence
channel. IEEE J. Lightw. Technol. 26(5), 478–487 (2008)
53. I.B. Djordjevic, Coding and modulation techniques for optical wireless channels, in Advanced Optical Wireless Communication Systems, ed. by
S. Arnon, J. Barry, G. Karagiannidis, R. Schober, M. Uysal, (Cambridge University Press, May 2012), pp. 11–53
54. A.A. Farid, S. Hranilovic, Upper and lower bounds on the capacity of wireless optical intensity channels, in Proceedings of the ISIT 2007,
pp. 2416–2420, Nice, France, 24 June–29 June 2007. A.R. Shah, R.C.J. Hsu, A.H. Sayed, B. Jalali, Coherent optical MIMO (COMIMO).
J. Lightw. Technol., 23(8), 2410–2419, Aug 2005
 Advanced Modulation and Multiplexing
Techniques 5

Abstract
This chapter is devoted to advanced modulation and multiplexing techniques suitable for both wireless and optical
communication systems. The chapter starts with signal space theory concepts applied to wireless communication systems.
After the geometric representations of signals, we describe various multidimensional modulators and demodulators
suitable for wireless communication applications, in particular the Euclidean distance, correlation, and matched filter-
based detectors, followed by the description of frequency-shift keying (FSK). Both continuous-time (CT) and discrete-time
(DT) implementations for pulse amplitude modulation schemes, suitable for wireless communications, are described as
well. After that, the focus is moved to multilevel schemes suitable for both wireless and optical communication
applications, including M-ary PSK, star-QAM, square-QAM, and cross-QAM. Regarding the optical communications,
the transmitters for M-ary PSK, star-QAM, and square
/cross-QAM are described in detail. The next topic in the chapter is
related to multicarrier modulation, including the description of multicarrier systems with both nonoverlapping and
overlapping subcarriers as well as the introduction of various approaches to deal with fading effects at the subcarrier
level. The concept of OFDM is also introduced; however, details are provided in Chap. 7. The MIMO fundamentals are
then provided, including the description of key differences with respect to diversity scheme, as well as the introduction of
array, diversity, and multiplexing gains. The parallel MIMO channel decomposition is briefly described. The details on
MIMO signal processing are postponed for Chap. 8. In the section on polarization-division multiplexing (PDM) and four-
dimensional (4-D) signaling, we describe key differences between PDM and 4-D signaling and describe how both types of
schemes can be implemented in both wireless and optical communications. The focus is moved to the spatial-division
multiplexing (SDM) and multidimensional signaling. We describe how SDM can be applied in wireless communications
first. Then we describe how various degrees of freedom including amplitude, phase, frequency, polarization states, and
spatial modes can be used to convey the information in the optical domain. In the same sections, the SDM concepts for
fiber-optic communications are described as well. The section concludes with SDM and multidimensional signaling
concepts applied to free-space optical (FSO) communications. The next topic is devoted to the signal constellation design,
including iterative polar modulation (IPM), signal constellation design for circular symmetric optical channels, energy-
efficient signal constellation design, and optimum signal constellation design (OSCD). The final section of the chapter is
devoted to the nonuniform signaling, in which different signal constellation points are transmitted with different
probabilities. The set of problems is provided after the concluding remarks.

5.1 Signal Space Theory in Wireless Communications

Before we start with a detailed description of advanced modulation schemes, we will present a generic view of optical digital
communication system, explain the geometric representation of a signal, and introduce signal constellation diagrams.
Additionally, we will provide generic configurations of modulators and demodulators.

# The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 249
I. B. Djordjevic, Advanced Optical and Wireless Communications Systems, https://doi.org/10.1007/978-3-030-98491-5_5
250 5 Advanced Modulation and Multiplexing Techniques

5.1.1 Geometric Representation of Signals

The key idea behind the geometric representation of the signals can be summarized as follows [1–7]: represent any set of
M energy (complex-valued) signals {si(t)} as a linear combination of N orthonormal basis functions {Φj}, where N  M, as
follows:

X
N 
0tT
si ðtÞ ¼ sij Φ j ðtÞ, , ð5:1Þ
j¼1 i ¼ 1, 2, . . . , M

where the coefficients of expansion represent projections along the basis functions:

ðT 
i ¼ 1, 2, . . . , M
sij ¼ hΦ j ðtÞjsi ðtÞi ¼ si ðtÞΦj ðtÞdt, : ð5:2Þ
j ¼ 1, 2, . . . , N
0

The basis functions satisfy the principle of orthogonality given as [1–9]

ðT 
1, i ¼ j
Φi ðtÞΦj ðtÞdt ¼ δij ¼ : ð5:3Þ
0, i 6¼ j
0

The signal vector can now be represented in a vector form as

2 3
si1
6s 7
6 i2 7
si ¼ jsi i ¼ 6 7; i ¼ 1, 2, . . . , M, ð5:4Þ
4 ...5
siN

where the jth component (coordinate) represents a projection along the jth basis function. The inner (dot) product of a signal
vector with itself is defined by

X
N
hsi jsi i ¼ ksi k2 ¼ s{i si ¼ jsij j2 ; i ¼ 1, 2, . . . , M: ð5:5Þ
j¼1

The signal energy of the ith symbol can be now expressed as

ðT ðT "X
N
#"
X
N
#
Ei ¼ jsi ðt Þj dt ¼ 2
sij Φ j ðt Þ sik Φk ðt Þ dt
j¼1 k¼1
0 0
ðT N   ð5:6Þ
P
N P
N P sij 2 ¼ ksi k2 :
¼ sij sik Φ j ðt ÞΦk ðt Þdt ¼
j¼1 k¼1 j¼1
0
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
1, k¼j

The inner product of signals si(t) and sk(t) is related to the correlation coefficient between them as

ðT
hsi jsk i ¼ si ðt Þsk ðt Þdt ¼ s{i sk , ð5:7Þ
0
5.1 Signal Space Theory in Wireless Communications 251

where we used { to denote the Hermitian conjugate. The Euclidean distance between signals si(t) and sj(t) can be
determined as

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi v ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uðT
u N
uX  2 u
ks i
s k k ¼ t sij
sik  ¼ u t jsi ðt Þ
sk ðt Þj2 dt : ð5:8Þ
j¼1
0

On the other hand, the angle between two signal vectors can be determined as
!

1 s{i sk
θik ¼ cos : ð5:9Þ
ksi kksk k

By using Eqs. (5.6) and (5.9), we can write that

ð
1

si ðt Þsk ðt Þdt

s{i sk
1
cos θi,k ¼ ¼2 31=2 2 31=2 : ð5:10Þ
ks i kks k k ð
1 ð
1
4 jsi ðt Þj2 dt5 4 jsk ðt Þj2 dt5

1
1

Since j cos θi, k j  1, the following inequality can be derived from (5.10):
1  21 31=2 2 1 31=2
ð  ð ð
 
 s ðt Þsk ðt Þdt  4 jsi ðt Þj2 dt5 4 jsk ðt Þj2 dt5 : ð5:11Þ
 i 
 

1
1
1

The arbitrary set of orthonormal basis functions {Φj} can be used in the geometric representation of signals. As an
illustration, the geometric representation of two sets of signals, also known as signal constellations, is shown in Fig. 5.1. The
one-dimensional (for N ¼ 1) ternary signal constellation (for M ¼ 3) is shown in Fig. 5.1(a), while the two-dimensional
(N ¼ 2) quaternary signal constellation, also known as quaternary quadrature amplitude modulation (4-QAM) or quadriphase-
shift keying (QPSK), is shown in Fig. 5.1(b).

Fig. 5.1 Illustration of geometric

representation of signals for (a) Decision
N ¼ 1, M ¼ 3 and (b) N ¼ 2, boundaries d
M¼4 s2
s1
s1 s2 s3 )1
D2 D1
-1 /2
-1/2 0 1/
/
1/2 1 )1
D1 D2 D3 d
D3 D4

Decision s3 s4
regions
(a) (b)
252 5 Advanced Modulation and Multiplexing Techniques

5.1.2 Modulators and Demodulators

To generate a given signal constellation point si, we can use Eq. (5.4), commonly referred to as the synthesis equation; we can
use the modulator (synthesizer) shown in Fig. 5.2, in the context of wireless communication systems. For binary sources, we
need to use parallel-to-serial (P/S) converter, whose output is used as an address for the N-dimensional mapper, implemented
as a look-up table (LUT). In this case, every M-ary symbol carries m ¼ log2M bits.
At the receiver side, decision boundaries should be established in the signal space to identify constellation points being
transmitted. If we assume the equal probability of signal transmission, the decision boundaries (see Fig. 5.1) between states
can be positioned as straight lines perpendicular to lines si
sk and passing at distance d/2 from si. Neighboring decision
boundaries determine a given decision region. In such a way, the signal space is split into M decision regions, denoted as Di.
When received vector r, obtained as a sum of transmitted vector sj and noise vector w, falls within the decision regionDi, we
decide in favor of signal point si. As an illustrative example, applicable to M-ary PAM suitable for optical communication
systems with direct detection, when the signals used in signal space do not have negative amplitude, we provide the signal
constellation and decision boundaries in Fig. 5.3.
Another decision strategy will be to use the Euclidean distance receiver [1], for which the Euclidian distance square
between the received signal r(t) and the mth transmitted signal sm(t) is defined as

ðT
d2E ðr, sm Þ ¼ jr ðt Þ
sm ðt Þj2 dt: ð5:12aÞ
0

Fig. 5.2 The configuration of the Sequence of source symbols

M-ary N-dimensional modulator
(synthesizer) for wireless
communications. DAC digital-to-
analog converter, PA power
The N-dimensional mapper (LUT)
amplifier
Coordinates: si,1 si,2 si,3 … si,N

)1 )2 )3 } )N
Transmit
antenna
N-dimensional
modulator
PA

N DAC
si  t  ¦ s ) t 
j 1
ij j

Fig. 5.3 Signal constellation, Decision boundaries

decision regions, and decision
boundaries for M-ary PAM for Decision regions
Decision region
optical communication systems
with direct detection
d

0 d d
2d …. M 2)d (M-
(M-
(M-2)d M 1)d
(M-1)d

Decision regions
5.1 Signal Space Theory in Wireless Communications 253

Fig. 5.4 Euclidean distance Receive

receiver for wireless antenna
communications. LNA low-noise
amplifier |˜|2 ³(˜)dt
 t=T
Selected
r(t) s1(t) Comparator: message

…
select the smallest mˆ mk
LNA
|˜|2 ³(˜)dt
 t=T
sM(t)

Fig. 5.5 The correlation receiver Receive

for wireless communications antenna
³(˜)dt
t=T
s1(t)  E1/2 Selected
r(t)
…

…
Comparator: message

LNA select the largest mˆ mk

³(˜)dt
t=T
sM(t)  EM/2

In the Euclidean distance receiver, we decide in favor of signal sk(t) closest to the received signal in Euclidean distance, based
on Eq. (5.12a), and the corresponding receiver is shown in Fig. 5.4. The decision strategy can also be described as follows:

Hm ðT Hm ð
T
< 2
<
d2E ðr, sm Þ d2E ðr, sn Þ , jr ðt Þ
sm ðt Þj dt jr ðt Þ
sm ðt Þj2 dt, ð5:12bÞ
> >
Hn 0 Hn 0

where we use Hm to denote the hypothesis that signal sm(t) was transmitted.
By squaring the bracket terms, assuming that signals are real valued, and after rearranging, we obtain that

ðT Tðs H m Tðs ðT
1 2 > 1
r ðt Þsm ðt Þdt
jsm ðt Þj dt r ðt Þsn ðt Þdt
jsn ðt Þj2 dt : ð5:13Þ
2 < 2
0 0 Hn 0 0
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} |fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}
Em En

The corresponding receiver, based on Eq. (5.13), is known as the correlation receiver, and it is shown in Fig. 5.5.
The received vector for this receiver, also known as the observation vector, can be obtained by projecting the received
signal along different basis functions as follows (assuming that signals are complex valued):
2 3
r1
6r 7 ðT
6 27
r ¼ 6 7, r n ¼ r ðt ÞΦn ðt Þdt: ð5:14Þ
4⋮5
0
rN

The N-dimensional demodulator (analyzer) is constructed based on Eq. (5.14) and provides the estimate of signal
constellation coordinates. Based on coordinates (projections of received signal), the Euclidean distance receiver assigns
the observation vector to the closest constellation point within Euclidean distance. The corresponding decision rule can be
defined as
254 5 Advanced Modulation and Multiplexing Techniques

Fig. 5.6 The reduced-complexity Observation: r1 r2 r3 … rN

correlation receiver: (a) detector/
demodulator and (b) signal
transmission decoder Samplers tk=kT
configurations
T T T T

³ ˜ dt
0
³ ˜ dt
0
³ ˜ dt
0
³ ˜ dt
0

) *1 )*2 )*3 } )*
Receive N

antenna
N-dim. demodulator:
correlation detector
Power splitter

LNA

(a)

Estimate m̂

Decision circuit (comparator):

Select the largest input

E1/2 E2/2 EM/2

Re{r †s1} Re{r † s2} Re{r † sM}

s1 s2 … sM

Dot-product
calculator
Observation
r vector
(b)

X
N  
bsm ¼ arg min kr
sk k2 ¼ arg min r j
skj 2 : ð5:15Þ
sk sk
j¼1

By rearranging Eq. (5.15), we obtain an alternative representation of the decision rule:

bsm ¼ arg max r{ sk
E k =2: ð5:16Þ

sk

The corresponding correlation receiver, based on Eq. (5.16), is shown in Fig. 5.6. This scheme has lower complexity since
there are N < M branches in detector/demodulator, as shown in Fig. 5.6(a).
An alternative receiver to the scheme discussed above is based on the so-called matched filter-based demodulator/detector
[1–7], as shown in Fig. 5.7. The output of the nth matched filter having impulse response hn ðt Þ ¼ Φn ðT
t Þ is given as

ð
1 ð
1

r n ðt Þ ¼ r ðτÞhn ðt
τÞdτ ¼ r ðτÞΦn ðT
t þ τÞdτ: ð5:17Þ

1
1
5.1 Signal Space Theory in Wireless Communications 255

Fig. 5.7 Matched filter-based Observation: r1 r2 r3 … rN

demodulator/detector
configuration
Samplers tk=kT

h1  t  h2  t  h3  t  hN  t 
)1* T  t  )*2 T  t  )*3 T  t  )*N T  t 

Receive N-dim. demodulator:

} matched filter-based
antenna
detector

Power splitter

By sampling at time instance t ¼ T, we obtain that

ð
1 ðT
r n ðT Þ ¼ r ðτÞhn ðT
τÞdτ ¼ r ðτÞΦn ðτÞdτ, ð5:18Þ

1 0

which is the same form as the nth output of detector in the correlation receiver. Therefore, the correlation and matched filter-
based receivers are equivalent to each other.
Even though in this section we are focused on wireless communications, by replacing the transmit antenna with
corresponding electro-optical modulator, and receive antenna with corresponding coherent optical detector, all modulators
and demodulators described here are also applicable in optical communications as well.

5.1.3 Frequency-Shift Keying (FSK)

The M-ary FSK signal set is defined as

rffiffiffiffiffiffi h i
2E π
si ðt Þ ¼ cos ðn þ iÞt , 0  t  T; i ¼ 1, 2, . . . , M, ð5:19Þ
T T

where n is an integer. Clearly, the individual signals are mutually orthogonal since each signal contains an integer number of
cycles within the symbol interval T and the number of cycles between adjacent FSK signals differs by one cycle, so that the
orthogonality principle among symbols is satisfied:

ðT
sm ðt Þsn ðt Þdt ¼ δmn : ð5:20Þ
0

Since the FSK signals are mutually orthogonal, they can be used as basis functions as follows:

1 0tT
Φm ðt Þ ¼ pffiffiffiffiffi sm ðt Þ, , ð5:21Þ
Es m ¼ 1, 2, . . . , M

where Es is the symbol energy. The minimum distance between adjacent FSK signals is given by dmin ¼ (2Es)1/2. The adjacent
frequency spacing is 1/(2 T ), so that the total bandwidth is given as B ¼ M/(2 T ). The corresponding bandwidth efficiency, for
data rate Rb, can be expressed as
256 5 Advanced Modulation and Multiplexing Techniques

Rb 2 log 2 M
ρ¼ ¼ : ð5:22Þ
B M

Clearly, the bandwidth efficiency decreases as the signal constellation size increases for all values M > 4.

5.1.4 M-ary Pulse Amplitude Modulation (PAM)

The M-ary pulse amplitude modulation in baseband signaling is one-dimensional signal set with the basis function being a
unit energy pulse p(t) of duration Ts. The resulting PAM signal can be represented as a train of rectangular pulses
X
sðt Þ ¼ n
sðnÞpðt
nT s Þ, ð5:23Þ

where s(n) 2 {Ai ¼ (2i
1
L )d; i ¼ 1, 2, . . ., L} for wireless communications and optical communications with coherent
optical detection or s(n) 2 {Ai ¼ (i
1)d, i ¼ 1, 2, . . ., L} for optical communications with direct detection. The pulse shape
p(t) is assumed to have a rectangular form with an amplitude A for
1/2 < t/T < 1/2 (and zero value otherwise), but other pulse
shapes are possible such as Gaussian pulse shape. When the number of amplitude levels is L ¼ 2 and direct detection is used,
the corresponding modulation scheme is known as on-off keying (OOK) or unipolar signaling. On the other hand, when the
number of amplitude levels is also two but coherent detection is used, the corresponding scheme is known as the antipodal
signaling scheme. Assuming an equally probable  transmission
 of symbols, the average symbol energy for coherent detection
P
PAM is given by Es ¼ ð1=M Þ Li¼1 A2i ¼ ð1=3Þ L2
1 d 2 : On the other hand, for the direct detection scheme with PAM, the
P
average symbol energy is given as E s ¼ ð1=M Þ Li¼1 A2i ¼ Es ¼ ðL
1Þð2L
1Þd2 =6: The corresponding bit energy Eb is
related to symbol energy as Eb ¼ Es/log2L. The number of bits per symbol equals log2L. Finally, the bit rate Rb can be
expressed by symbol rate Rs ¼ 1/T as Rb ¼ Rs log2L.
We will now turn our attention to the description of both continuous-time (CT) and discrete-time (DT) implementations of
the PAM scheme, described in the context of wireless communications. The CT PAM modulator and demodulator are shown
in Fig. 5.8. At the output of the serial-to-parallel (S/P) conversion block, the log2L bits are used to select the point from the
PAM mapper, implemented as the look-up table (LUT). The output of LUT can be represented as

Transmit
antenna

log2 L
PA
Sequence S/P bits Mapper Pulse shaper
of bits convertor (LUT) h(t)=p(t)

(a)
Receive
antenna

r(t)=sm(t)+z(t) Matched y(kT) Decision log2 L bits

Filter sˆ arg min y kT Am
h(t)=p(-t) corresponding to ŝ
t=kT Am

(b)
Fig. 5.8 The CT PAM configurations of (a) modulator and (b) demodulator
5.1 Signal Space Theory in Wireless Communications 257

X
sδ ðt Þ ¼ n
sðnÞδðt
nT Þ, ð5:24Þ

where δ(t) is the Dirac delta function. The output of the pulse shaper of the impulse response h(t) ¼ p(t) is obtained as a
convolution of the input a(t) and the pulse shaper impulse response:
P
sðt Þ ¼ P
sδ ðt Þ  pðt Þ ¼ n sðnÞδðt
nT Þ  pðt Þ
ð5:25Þ
¼ n sðnÞpðt
nT Þ,

which is the same form as the one given by Eq. (5.23). The continuous-time correlation PAM demodulator/detector is shown
in Fig. 5.8(b).
The output of the correlator, which presents the projection along the basis function p(t), is used in the Euclidean distance
receiver to determine the closest amplitude level Ai from the signal constellation set. For the additive channel model, where the
received signal r(t) ¼ s(t) + z(t), where z(t) is the additive white Gaussian noise, the output of the correlator can be written as

ð
1

yðt Þ ¼ r ðτÞpðτ
t Þdτ

1
ð
tþnT ð
tþnT
ð5:26Þ
P
¼ n s ð nÞ pðτ
nT Þpðτ
t Þdτ þ zðτÞpðτ
t Þdτ:
nT nT
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Rp ðt
nT Þ z0 ðt Þ

which after sampling becomes

X
yðkT Þ ¼ n
sðnÞRp ðkT
nT Þ þ z0 ðkT Þ ¼ sðkÞ þ z0 ðkT Þ: ð5:27Þ

We assume in the equation above that the pulse shape is properly chosen in accordance to the Nyquist criterion [1] and there is
no intersymbol interference at the output of the matched filter so that we can write

1, l ¼ 0
Rp ðlT s Þ ¼ :
0, l 6¼ 0

The DT implementation of the PAM modulator and demodulator is shown in Fig. 5.9. At the output of the serial-to-parallel
(S/P) conversion, the log2L bits are used to select the point from the PAM mapper, which is implemented as the look-up table.

log2 L
Sequence bits DT pulse
S/P Mapper
of bits convertor (LUT) U shaper DAC
h(n)=p(n T)
U=T / T

(a)

r(t)=sm(t)+z(t) r(n T) Matched y(n T) y(kT) Decision log2 L bits

ADC Filter sˆ arg min y kT Am
h(n)=p( n T) corresponding to ŝ
n=k U Am

(b)
Fig. 5.9 The DT PAM configurations of (a) modulator and (b) demodulator for wireless communications
258 5 Advanced Modulation and Multiplexing Techniques

The output of a mapper is upsampled for a factor U ¼ T/ΔT, where ΔT is the sampling period. The upsampling is performed
by inserting (U
1) zero samples between two neighboring samples in original sequence. The output of the pulse shaper can
be obtained by a DT convolution sum of the upsampler output ∑ks(k)δ(nΔT
kTs) and the pulse shaper impulse response
h(n) ¼ p(nΔT), which gives the following result:
X
sδ ðnΔT Þ ¼ k
sðkÞδðnΔT
kT Þ: ð5:28Þ

As a summary, after digital-to-analog conversion (DAC) at the transmitter side, we obtain the modulating signal given by
Eq. (5.25). On the receiver side, after performing ADC, matched filtering, and sampling, we obtain the input to the decision
circuit given by Eq. (5.27). The decision circuit operates in the same fashion as in the CT case described above.

5.1.5 Passband Digital Wireless/Optical Transmission

The passband modulated signal s(t) can be described in terms of its in-phase and in-quadrature (which is often referred to just
as quadrature) components, in standard (or canonical) form [1, 2], as follows:

sðt Þ ¼ sI ðt Þ cos ð2π f c t Þ
sQ ðt Þ sin ðωc t Þ, ð5:29Þ

where sI(t) and sQ(t) represent the in-phase and quadrature components, respectively, while ωc denotes the carrier frequency
[rad/s]. Because the sine and cosine signals are orthogonal to each other in the kth symbol interval (kT  t  (k + 1)T ), the
corresponding signal space is two-dimensional. To facilitate further discussion, we can use the complex envelope (low-pass
(LP)) representation of the signal s(t) given by

esðt Þ ¼ sI ðt Þ þ jsQ ðt Þ: ð5:30Þ

In the frequency domain, we have the following relation between the power spectral densities related to standard form
(denoted as PSDs) and to complex envelope representation (denoted as PSDLP):

1
PSDs ðωÞ ¼ ½PSDLP ðω
ωc Þ þ PSDLP ðω þ ωc Þ, ð5:31Þ
4

where the indices “S” and “LP” stand for “standard” and “low pass” representations, respectively. The standard form can also
be represented by polar coordinates in a form

h i1=2 
sQ ðt Þ
sðt Þ ¼ aðt Þ cos ½ωc t þ ϕðt Þ; aðt Þ ¼ s2I ðt Þ þ s2Q ðt Þ , ϕðtÞ ¼ tan
1 : ð5:32Þ
sI ðt Þ

If the information-bearing signal is imposed on amplitude, the corresponding modulated signal is called amplitude-shift
keying (ASK) one. On the other hand, if the information-bearing signal is imposed on phase, the corresponding modulating
format is called phase-shift keying (PSK). In addition, if the modulating signal is imposed on frequency, the modulation
format is called frequency-shift keying (FSK). It is also possible to use polarization orientation to modulate the carrier. The
corresponding modulation format is called polarization shift keying (PolSK). Finally, multiple degrees of freedom can be used
to impose the information-bearing signal, and the corresponding modulation scheme is known as hybrid one. For instance, if
amplitude and phase are used simultaneously, the corresponding modulation scheme is known as quadrature amplitude
modulation (QAM). The signal space of QAM is 2-D. By combining the in-phase and quadrature basis functions with two
orthogonal polarization states, the resulting signal space becomes four-dimensional (4-D).
The wireless/optical signal, which is used as a carrier, can be presented as a monochromatic electromagnetic wave whose
electric field is given as

Eðt Þ ¼ pA cos ½2πft þ ϕðt Þ, 0  t < T: ð5:33Þ

5.2 Multilevel (Two-Dimensional) Modulation Schemes 259

Each of the parameters in Eq. (5.33) (amplitude A, frequency f, phase ϕ, and polarization orientation p) can be utilized to carry
the information. The corresponding modulation formats implemented on RF/optical carrier are known as ASK, FSK, PSK,
and PolSK, respectively, as indicated already above.
As an illustration, in M-ary ASK, the signal waveforms can be represented as

sm ðt Þ ¼ Am pðt Þ cos ωc t; 0  t < T; m ¼ 1, 2, ⋯, M, ð5:34Þ

where p(t) is the pulse shape, while {Am| m ¼ 1, 2, . . ., M} represents a set of possible amplitudes. For M ¼ 2, we can set
A1 ¼ 0 and A2 ¼ A. For M > 2, we can use the PAM representation as follows: Am ¼ (2 m
1-M)d/2, indicating that the
distance between adjacent amplitudes is d.
Similar to signal, any linear time-invariant (LTI) passband system can be represented by its canonical impulse response:

hðt Þ ¼ hI ðt Þ cos ðωc t Þ
hQ ðt Þ sin ðωc t Þ, ð5:35Þ

where hI(t) and hQ(t) represent the in-phase and quadrature components, respectively. The complex envelope representation of
LTI band-pass systems is given by

e
hðt Þ ¼ hI ðt Þ þ jhQ ðt Þ: ð5:36Þ

The complex envelope of signal at the output of the band-pass system is obtained by CT convolution of the complex forms:

1
eyðt Þ ¼ ehðt Þ  exðt Þ: ð5:37Þ
2

Finally, the canonical form of the output of the LTI band-pass system can be obtained as

yðt Þ ¼ Re ½eyðt Þ exp ð jωc t Þ: ð5:38Þ

An important parameter of any modulation scheme is its bandwidth efficiency, defined as S ¼ Rb/B[bits/s/Hz], where B is
the bandwidth occupied by modulated signal, determined from PSDLP shape (often as the width of the main lobe in the PSD).
The spectrally efficient modulation schemes are aimed to maximize the bandwidth efficiency. In the next part, we will describe
the spectrally efficient modulation schemes that can be used in both wireless and optical communication systems.

5.2 Multilevel (Two-Dimensional) Modulation Schemes

In this section, we describe the multilevel and multidimensional modulation schemes suitable for and/or in use for both
wireless and optical communication systems.

5.2.1 Two-Dimensional Signal Constellations for Wireless Communications

M-ary QAM is a 2-D generalization of M-ary PAM, with two orthogonal passband basis functions given as
rffiffiffiffiffi
2
ΦI ðt Þ ¼ cos ðωc t Þ, 0  t  T
Ts
rffiffiffiffiffi ð5:39Þ
2
ΦQ ðt Þ ¼ sin ðωc t Þ, 0  t  T:
Ts

The mth signal constellation point in signal (ΦI, ΦQ) space can be represented by (Imdmin/2, Qmdmin/2), where dmin is the
minimum Euclidean distance among signal constellation points, which is related to the minimum energy signal constellation
260 5 Advanced Modulation and Multiplexing Techniques

1=2
point Emin by dmin =2 ¼ E min : With (Im,Qm), we denoted the coordinates of the mth signal constellation point. The transmitted
signal in the kth symbol interval can be represented as
rffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffi
2E min 2Emin 0tT
sk ðt Þ ¼ I cos ðωc t Þ
Q sin ðωc t Þ, : ð5:40Þ
Ts k Ts k k ¼ 0,  1,  2, . . .

Therefore, two phase-quadrature carriers are being employed, each modulated by the corresponding set of discrete amplitudes.
All QAM constellations can be categorized into three broad categories: (i) square-QAM in which the number of bits per
symbol is even (signal constellation size is a square number, M ¼ L2, where L is a positive integer); (ii) cross constellations in
which the number of bits per symbol is odd; and (iii) star-QAM constellations in which signal constellation points are placed
along concentric circles of different radiuses. The M-ary PSK can be considered as a special case of star-QAM with
constellation points placed uniformly on a single circle.
The square-QAMconstellations can be interpreted as being obtained as a two-dimensional Cartesian product of 1-D-PAM,
described in Subsect. 5.1.4, as follows:

X 2 ¼ X  X ¼ fðx1 , x2 Þjxn 2 X, 1  n  2g, ð5:41Þ

where X ¼ {2l
1
L| l ¼ 1, 2, . . ., L}, while L is the size of 1-D-PAM, which is related to the size of QAM square
constellation as L2 ¼ MQAM. In other words, the coordinates of the kth signal constellation point for square M-ary QAM signal
constellation are given by
2 3
ð
L þ 1, L
1Þ ð
L þ 3, L
1Þ ... ðL
1, L
1Þ
6 ð
L þ 1, L
3Þ ð
L þ 3, L
3Þ ... ðL
1, L
3Þ 7
6 7
fI k , Q k g ¼ 6 7: ð5:42Þ
4 ... ... ... ... 5
ð
L þ 1,
L þ 1Þ ð
L þ 3,
L þ 1Þ . . . ðL
1,
L þ 1Þ

As an illustration, for 16-ary square constellation, the signal constellation points are given as
2 3
ð
3, 3Þ ð
1, 3Þ ð1, 3Þ ð3, 3Þ
6 ð
3, 1Þ ð
1, 1Þ ð1, 1Þ ð3, 1Þ 7
6 7
fI k , Qk g ¼ 6 7:
4 ð
3,
1Þ ð
1,
1Þ ð1,
1Þ ð3,
1Þ 5
ð
3,
3Þ ð
1,
3Þ ð1,
3Þ ð3,
3Þ

The corresponding constellation is shown in Fig. 5.10, together with 64-QAM.

The cross constellation can be obtained by properly combining two square constellations; the inner square constellation
carries m
1 bits per symbol, while the outer constellation is split into four regions, with each region carrying m-3 bits/
symbol. The symbols are placed along each side of inner constellation as shown in Fig. 5.11(a). Another way to interpret the
cross signal constellation is to start with the Cartesian product of 1-D-PAM to obtain L2 square constellation and remove the
needed number of points in the corners to obtain a desired cross constellation. For example, 32-ary cross constellation can be
obtained by starting from 36-ary square constellation and by removing the four points in the corners, as shown in Fig. 5.11(b).
The star-QAM, also known as amplitude phase-shift keying (APSK), can be obtained by placing the constellation points
along the concentric rings. The M-ary phase-shift keying (PSK) is just a special case of star-QAM having only one ring. The
transmitted M-ary PSK signal, which can be considered as a special case of star-QAM, for the mth symbol can be expressed as
8 rffiffiffiffiffiffi

< 2E 2π
cos ωc t þ ðm
1Þ ,0  t  T
sm ðtÞ ¼ T M ; m ¼ 1, 2, . . . , M: ð5:43Þ
:
0, otherwise

As an illustration, the 8-PSK and 8-star-QAM constellation diagrams are shown in Fig. 5.12. Different mapping rules can
be used to map binary sequence to corresponding constellation points including natural mapping, Gray mapping, and anti-
Gray mapping. In Gray mapping rule, the neighboring signal constellation points differ in only one bit position. For additive
5.2 Multilevel (Two-Dimensional) Modulation Schemes 261

16-QAM 8 64-QAM
3
6
2 4

1 2

0 0

Q
Q

-2
-1
-4
-2
-6
-3 -8

-3 -2 -1 0 1 2 3 -8 -6 -4 -2 0 2 4 6 8

I I

(a) (b)
Fig. 5.10 Square-QAM constellations: (a) 16-QAM and (b) 64-QAM

6 32-QAM

Outer 4
constellation
(2m-3 points)
2
constellation
constellation

Inner square
(2m-3 points)
(2m-3 points)

constellation
Outer
Outer

0
(2m-1
Q

constellation
points) -2

Outer
constellation -4
(2m-3 points)
-6
-6 -4 -2 0 2 4

(a) (b)
Fig. 5.11 The cross signal constellation design: (a) derived from square constellations and (b) 32-QAM

white Gaussian noise (AWGN) and for moderate to high SNRs, the Gray mapping is the optimum one, as the single symbol
error results in a single bit error. The corresponding Gray mapping rule for 8-PSK is shown in Fig. 5.12(a). Strictly speaking,
the Gray mapping for 8-star-QAM cannot be defined. However, the mapping shown in Fig. 5.12(b) is a good approximation of
Gray mapping. In Table 5.1, we specify how the 3-bit sequence in 8-PSK and 8-star-QAM is mapped to the corresponding
constellation point.
262 5 Advanced Modulation and Multiplexing Techniques

Q
Q

011 011
010
001

010
001 r1

110 000 I 110 000 I

r0

100 c c c
111 2 1 0

100
111
101
c2 c1 c
0 101

(a) (b)
Fig. 5.12 The constellation diagrams for (a) 8-PSK and (b) 8-star-QAM

Table 5.1 Gray mapping rules for 8-PSK and 8-star-QAM

8-PSK 8-QAM
Input bits (c2c1c0) ϕk Ik Qk Ik Qk
000 0 1 0 1 + √3 0
001 π/4 √2/2 √2/2 1 1
011 π/2 0 1 0 1 + √3
010 3π/4 -√2/2 √2/2
1 1
110 π
1 0
(1 + √3) 0
111 5π/4
√2/2
√2/2
1
1
101 3π/2 0
1 0
(1 + √3)
100 7π/4 √2/2
√2/2 1
1

The CT implementation of the QAM modulator and demodulator, suitable for wireless communication applications, is
shown in Fig. 5.13. The m ¼ log2M bits from an S/P converter are used to select a point from the QAM mapper by using the
appropriate mapping rule. The in-phase I(t) and in-quadrature Q(t) outputs of the upper and lower branch pulse shaper can be,
respectively, written as
X X
I ðt Þ ¼ I l δð t
lT Þ  p ð t Þ ¼ I pðt
lT Þ,
l l
Xl X ð5:44Þ
Q ðt Þ ¼ Q δðt
lT Þ  pðt Þ ¼
l l
Q pðt
lT Þ,
l l

where the pulse shaper impulse response p(t) (commonly the rectangular pulse of unit energy) is appropriately chosen to
satisfy the Nyquist criterion for zero ISI, which is Rp(lTs) ¼ 1 for l ¼ 0 and Rp(lTs) ¼ 0 for other l-values.
The outputs of pulse shapers, I(t) and Q(t), are used to modulate the in-phase and quadrature carriers so that the modulated
signal at the output of I/Q modulator can be represented as
rffiffiffiffi rffiffiffiffi
X 2 2
sðt Þ ¼ I pðt
lT Þ cos ðωc t Þ
Q pðt
lT Þ sin ðωc t Þ, ð5:45Þ
l T l T l

which is consistent with Eq. (5.40). On the receiver side (see Fig. 5.13(b)), the outputs of matched filters represent projections
along basis functions, which after sampling can be written as
5.2 Multilevel (Two-Dimensional) Modulation Schemes 263

l
Il t lT I t
Pulse shaper
h(t)=p(t)
2 / T cos c t
+ s t
Sequence S/P log2 M Mapper Oscillator
of bits converter bits (LUT)
/2

Pulse shaper
Q t 2 / T sin c t
Ql t lT h(t)=p(t)
l
I/Q modulator

(a)
I t
r t s t
rI t CT matched I lT Iˆl
filter
z t 2 / T cos t h(t)=p( t) t=lT
c Euclidean
Local
distance- Demapper
oscillator (LO) log2M bits
based (LUT) corresponding to ŝ
/2
detector
2 / T sin c t
Q t Qˆ l
CT matched I lT
filter
rQ t h(t)=p( t) t=lT

(b)
Fig. 5.13 The CT implementations of (a) QAM modulator and (b) QAM demodulator suitable for wireless communication applications

X
bI ðlT s Þ ¼ I m Rp ðmT
lT Þ þ zI ðlT Þ,
Xm ð5:46Þ
b ðlT s Þ ¼
Q Qm Rp ðmT
lT Þ þ zQ ðlT Þ,
m

where zI and zQ represent the projection of the noise process along the in-phase and quadrature basis functions, respectively.
Since p(t) has the rectangular shape described above and in the absence of noise, we can perfectly recover the in-phase and
quadrature coordinates of the transmitted QAM constellation point, as Rp(mTs) ¼ 1 for m ¼ 0, and Rp(mTs) ¼ 0 for m 6¼ 0.
However, in the presence of noise and channel distortions,
we use the Euclidean
 distance receiver and decide in favor of a
b ðlT Þ .
candidate constellation point closest to the received point bI ðlT Þ, Q
The corresponding DT implementation of the QAM modulator and demodulator is shown in Fig. 5.14. The m ¼ log2M bits
from an S/P converter are used to select a point from the QAM mapper. The outputs of a mapper are upsampled for a factor
U ¼ T/ΔT, where ΔT is the sampling period.
The outputs of pulse shapers can be obtained as the DT convolution sums of inputs and impulse responses h(i) ¼ p(iΔT) of
corresponding pulse shapers, to obtain
X
I ðiΔT Þ ¼ I l pðiΔT
lT Þ,
Xl ð5:47Þ
QðiΔT Þ ¼ Q pðiΔT
kT Þ:
l l

These outputs are used to modulate in-phase and quadrature DT carriers, generated by direct digital synthesizer. The output of
DT I/Q modulator can be represented as
rffiffiffiffiffi rffiffiffiffi
X 2 2
sðiTÞ ¼ I pðiT
lTÞcos ðΩc iÞ
Q pðiT
lTÞsin ðΩc iÞ, ð5:48Þ
l Ts l T l
264 5 Advanced Modulation and Multiplexing Techniques

l
Il t lT DT pulse I i T
U shaper
h(i)=p(i T)
log2M 2 / T cos i
U=T/ T c
+ s i T
Sequence S/P bits Mapper Direct digital DAC
converter (LUT) synthesizer (DDS)
2 / T sin c i
DT pulse Q i T
U shaper
Ql t lT h(i)=p(i T) I/Q modulator
l
U=T/ T

(a)
yI i T DT I i T
matched I lT
filter Euclidean
h(i)=p(-i T) i =lU
2 / Ts cos c i distance-based
Estimated
Direct digital detector
ADC sequence
synthesizer
+
2 / Ts sin i
c
DT matched Q i T I lT Demapper
filter
yQ i T h(i)=p(-i T) i =lU

(b)
Fig. 5.14 The DT implementations of (a) QAM modulator and (b) QAM demodulator for wireless communication applications

which is just a sampled version of waveform given by Eq. (5.45); after DAC. The parameter Ωc is discrete frequency
(expressed in radians per sample) related to the carrier frequency ωc [rad/s] by Ωc ¼ ωcΔT, where ΔT is the sampling period.
The operation of the direct digital synthesizer (DDS) is described in Chap. 3. On the receiver side, after band-pass filtering
(BPF) and ADC, discrete-time projections along DT basis functions are obtained as follows:
X
bI ðiT Þ ¼ I l Rp ðlT
iT Þ þ zI ðiT Þ,
Xl ð5:49Þ
b ðiT Þ ¼
Q Q R ðlT
iT Þ þ zQ ðiT Þ,
l l p

where zI and zQ, similar to the CT case, represent the projections of the noise process along the in-phase and quadrature basis
functions, respectively. After resampling the matched filters’ outputs at i ¼ lU, we can obtain the estimates of the transmitted
QAM signal constellation point coordinates as follows:
X
bI ðlT Þ ¼ I m Rp ðmT
lT Þ þ zI ðlT Þ ¼ I l þ zI ðlT Þ,
Xm ð5:50Þ
b ðlT Þ ¼
Q Qm Rp ðmT
lT Þ þ zQ ðlT Þ ¼ Ql þ zQ ðlT Þ,
m

which are identical to the corresponding CT case equations given by (5.46).

Before concluding this section, let us discuss the bandwidth efficiency of PSK and QAM formats. The power spectral
density (PSD) of LP representation of M-PSK is given by

LP ðf Þ ¼ 2E s sinc ðTf Þ ¼ 2E b log2 Msinc ð f T b log2 MÞ,

PSDPSK ð5:51Þ
2 2

where we assume that the pulse shaper impulse response is a rectangular pulse of duration T. With Es we denoted symbol
energy, Eb is the bit energy, while Tb denotes the bit duration. If we use the null-to-null bandwidth definition, corresponding to
the main lobe in the PSD, we can establish relation B ¼ 2/T between the bandwidth occupied by the modulated signal B and
the symbol duration T. Since Rb ¼ 1/Tb, it follows that
5.2 Multilevel (Two-Dimensional) Modulation Schemes 265

B ¼ 2Rb = log 2 M: ð5:52Þ

The bandwidth efficiency is then obtained as

S ¼ Rb =B ¼ ð log 2 M Þ=2, ð5:53Þ

which means that it increases as the signal constellation size grows, but the dependence is logarithmic. As an illustration, we
can calculate that the bandwidth efficiency for 4-QAM (QPSK) is equal to 1 bit/s/Hz. If 3 dB bandwidth definition of the
spectrum is used instead, the bandwidth efficiency values will be higher.
The transmitter configurations for optical communications are more complicated than those for wireless communications
[6, 7, 9]. In principle, a single electro-optical (E-O) I/Q modulator, as described in Chap. 3 (see Fig. 3.10), can be used,
providing that sufficiently small amplitude variations of both in-phase and in-quadrature channels are used so that the
approximation given by Eq. (3.22) can be applied. A mapper, implemented as an LUT, is then sufficient to provide Cartesian
coordinates of a corresponding M-PSK/M-QAM constellation. However, large constellations are quite challenging to
implement with this approach. In the incoming sections, we describe various transmitters for optical communication
applications, based on [7, 9] (and references therein). The readers not familiar with external modulators for optical
communications are referred to Sect. 3.1.2.3 in Chap. 3.

5.2.2 M-ary PSK Transmitters for Optical Communications

There are several ways to generate M-PSK signals by using the following: (i) concatenation (serial connection) of M E-O
phase modulators (PMs), described in Fig. 3.9 of Chap. 3; (ii) combination of E-O I/Q and phase modulators; and (iii) single
E-O I/Q modulator. The serial connection of PMs is shown in Fig. 5.15. After serial-to-parallel conversion (or 1:m DEMUX),
m output bits are properly mapped into output signals as follows:

Vπ X
1
V i ðt Þ ¼ ci ðkÞpðt
kT s Þ; i ¼ 1, 2, ⋯, m: ð5:54Þ
2i
1 k¼
1

If the optical transmission system is sensitive to cycle slips, then differential encoding should be used. With p(t) we denoted
the pulse shape, which is obtained by the convolution of the impulse response of pulse shaper and impulse response of the
driver amplifier. The output amplitude of the ith pulse shaper (PS) equals Vπ/2i
1. The phase change after the first phase
modulator is equal to π, thus indicating that the first PM generates a BPSK format. The phase change introduced by the
second PM equals π/2, which makes the resulting signal constellation after the second PM to be in a QPSK format. Next, the

bm cm Pulse shaper (PS)

…

Sequence S/P b3 c3
convertor Mapper PS
of bits b2 c2
(1:m DEMUX) PS
b1 c1
PS
Pulse shaper V1(t), V2(t), V3(t), Vm(t),
V V /2 V /4 V /2m 1

CW
MZM PM PM PM PM
Laser

Sinusoidal
signal

Fig. 5.15 The serial configuration of multilevel PSK transmitter for optical communications. PM E-O phase modulator, MZM Mach-Zehnder
modulator, PS pulse shaper
266 5 Advanced Modulation and Multiplexing Techniques

third PM introduces the phase change π/4, so that the signal constellation after the third PM has an 8-PSK format. In a
similar fashion, it can be shown that the signal constellation after the mth E-O PM has an M ¼ 2m-PSK format.
The Mach-Zehnder modulator (MZM) is optional in an M-PSK transmitter setup; its purpose would be to perform the
NRZ-to-RZ conversion. The overall electric field at the output of the serial M-PSK transmitter can be represented as
!
π X
pffiffiffiffiffiffi m
ET ðt Þ ¼ PT e jðωT tþϕT,PN Þ exp j V ðt Þ , ð5:55Þ
V π i¼1 i

where PT denotes the transmitted power, ωT is the carrier emitting frequency, while ϕT,PN denotes the phase noise introduced
by the CW transmitter laser. Equation (5.55) also indicates that it is possible, at least in principle, to perform a summation of
various signals described by Eq. (5.54) in the electrical domain while employing only a single E-O PM.
The differential encoding is an efficient way to deal with the cycle slips at the expense of SNR performance degradation.
The differential PSK eliminates the need for frequency locking of the local laser oscillator at the receiving side with the
transmitting laser, since it combines two basic operations at transmitter: differential encoding and phase-shift keying. For
instance, in differential BPSK (DBPSK), to send bit 0 we advance in phase the current signal waveform by π radians, while to
send bit 1 we leave the phase unchanged. The receiver measures the relative phase difference received during two successive
bit intervals by using either the Mach-Zehnder interferometer, which is followed by either direct detection or conventional
coherent difference detection. The coherent optical detection scheme is also suitable for the compensation of linear channel
impairments while being insensitive to cycle slips. The transmitted signal in this case can be represented as
(
pffiffiffiffiffiffi pðtÞ, 0  t  T b
jðωT tþϕT,PN Þ
E 1 ðtÞ ¼ PT e
pðtÞ, T b  t  2T b
( ð5:56Þ
pffiffiffiffiffiffi pðtÞ, 0  t  T b
jðωT tþϕT,PN Þ
E 0 ðtÞ ¼ PT e ,

pðtÞ, T b  t  2T b

where the subscript “T” denotes the transmitted bit. If DBPSK transmission is interpreted as symbol transmission of duration
2Tb, it is evident from Eq. (5.56) that two symbols are orthogonal to each other. The following differential encoding rule is
applied: (i) if the incoming bit at the kth time instance, denoted as bk, is 1, leave the symbol ck unchanged with respect to the
previous bit; and (ii) if the incoming binary symbol bk is 0, we complement the symbol ck with respect to the previous bit. In
L
other words, we can write that ck ¼ bk ck
1 , where the overbar denotes the complement operation to the summation per
mod 2.
Another interpretation of M-PSK/DPSK, which requires the use of only one phase modulator but more complicated
electrical signal generation, can be described as follows. In a coherent detection scheme for M-ary PSK, the data phasor
ϕl 2 {0,2π/M, . . ., 2π(M
1)/M} is sent in each lth transmission interval. In both direct detection and coherent difference
detection for M-ary DPSK, the modulation is differential and the data phasor ϕl ¼ ϕl
1 + Δϕl is sent instead, where Δϕl 2
{0,2π/M, . . ., 2π(M
1)/M} is determined by the sequence of m ¼ log2M input bits using an appropriate mapping rule.

5.2.3 Star-QAM Transmitters for Optical Communications

The star-QAM transmitter can be interpreted as a generalization of the M-PSK one. In star-QAM, instead of having only one
circle in constellation diagram, we typically have several concentric circles with constellation points uniformly distributed
around. For instance, the signal constellation for 8-star-QAM is composed of two concentric circles of radiuses r0 and r1, with
four points per circle, as illustrated in Fig. 5.12(b). For 16-star-QAM, we have different options: two circles with 8 points
each, four circles with 4 points each, three circles with 6 points, 5 and 5 points, and so on. The ratio of radiuses does not need
to be fixed, while the number of points per circle does not need to be uniform. In Fig. 5.16, we provide a 16-star-QAM
transmitter configuration for the signal constellation containing two circles with 8 points each, which can be interpreted as a
generalization of the scheme presented in Fig. 5.15. The first three E-O PMs in Fig. 5.16 are used to generate 8-PSK. The
MZM is controlled with bit c4, and for c4 ¼ 0, it generates the signal of amplitude that corresponds to the inner circle of radius
r0, while for c4 ¼ 1, it generates the signal of amplitude that corresponds to the outer circle of radius r1. A similar strategy can
be applied for an arbitrary star-QAM signal constellation.
5.2 Multilevel (Two-Dimensional) Modulation Schemes 267

b4 c4
PS

Sequence S/P b3 c3
convertor Mapper PS
of bits b2 c2
(1:m DEMUX) PS
b1 c1
PS
Pulse shaper V1(t), V2(t), V3(t),
V /2 V /4 VMZM(t)
V
CW
MZM PM PM PM MZM
Laser

Sinusoidal
signal

Fig. 5.16 The operation principle of a concatenated (serial) configuration of a 16-star-QAM transmitter for optical communications

Let us say that signal constellation contains Mc concentric circles with Mp points per circle. The number of required E-O
PMs is mp ¼ log2Mp, while the number of required bits to select a given circle is mc ¼ log2Mc. Only one MZM will be needed,
with driving signal having Mc amplitude levels. The number of bits per constellation point is mc + mp, where the first mc bits
are used to select a circle, while the remaining mp bits are used to select a point within a given circle. The set of Mp phase
modulators, each driven with binary sequence, can be replaced with a single one dealing with the Mc-level signal. This
configuration corresponds to the polar modulator shown in Fig. 3.11 in Chap. 3. Alternatively, a single I/Q modulator can be
used, but requires the mapper implemented as an LUT followed by a DAC of sufficient resolution.
For differential encoding, the mc bits that are used to select a given circle must stay in natural mapping form, while the
remaining mp bits can be differentially encoded. The 16-star-QAM modulator can also be implemented by concatenating one
E-O I/Q modulator, one E-O PM, and one MZM. The E-O I/Q modulator is used to generate a QPSK constellation, while the
E-O PM is used to introduce an additional phase shift of π/4, so that 8-PSK is generated. Finally, the MZM is used to select
one out of two 8-PSK constellations. Notice that the use of more than two amplitude levels in MZM degrades the
overall OSNR.
The 16-ary star-QAM modulator shown in Fig. 5.16 can be generalized by using mp of E-O PMs and one MZM. This
modulator is able to generate a star-QAM constellation composed of two 2mp-ary PSK constellations. The output electric field
of this star-QAM modulator is given by

mp
! 
pffiffiffiffiffiffi π X πV MZM ðt Þ
ET ðt Þ ¼ PT e jðωT tþϕT,PN Þ exp j V ðt Þ cos , ð5:57Þ
V π i¼1 i 2Vπ

where Vi(t) (i ¼ 1,2,. . .,mp) is given by Eq. (5.54), while VMZM(t) is given by
" #
2V π X

1
V MZM ðt Þ ¼ cos ðr 0 =r 1 Þ cmp þ1 ½kpðt
kT s Þ
1 : ð5:58Þ
π k

5.2.4 Square- and Cross-QAM Transmitters for Optical Transmission

The square
/cross-QAM constellations can be generated in a number of different ways by using (i) single E-O I/Q modulator
(Fig. 3.10), (ii) single E-O polar modulator (Fig. 3.11), (iii) improved E-O I/Q modulator having one E-O PM and one MZM
per phase branch, (iv) parallel E-O I/Q modulators, (v) so-called tandem-QAM transmitter, and so on. The configuration
employing a single E-O I/Q modulator is shown in Fig. 5.17. The m ¼ log2M bits are used to select a QAM constellation point
stored in a QAM mapper, implemented as an LUT. The in-phase and quadrature driving signals are given as
268 5 Advanced Modulation and Multiplexing Techniques

Fig. 5.17 The operation bm

principle of an E-O I/Q Pulse shaper (PS)

…
modulator-based square-QAM Sequence S/P
transmitter suitable for optical QAM
convertor b2 Mapper
communications of bits (1:m DEMUX) PS
b1

VI(t) VQ(t)

CW
MZM I/Q Modulator
Laser

Sinusoidal
signal

Fig. 5.18 The operation bm Pulse shaper

principle of an E-O polar
(PS)

…
modulator-based square-QAM S/P
Sequence QAM
transmitter for optical convertor b2 Mapper
communications of bits (1:m DEMUX) PS
b1

VMZM(t) VPM(t)

CW
MZM MZM PM
Laser

Sinusoidal
signal

" #
V X
1
π
V I ðt Þ ¼ 2 π sin
1
ðb1 ðk ÞÞpðt
kT s Þ
,
π k¼
1
2
" # ð5:59Þ
V X
1
V Q ðt Þ ¼ 2 π sin
1 ðb2 ðkÞÞpðt
kT s Þ
1 :
π k¼
1

It is evident, based on Eqs. (3.23) and (3.24), that polar modulator can be used instead of an E-O I/Q modulator to generate
an arbitrary square/cross constellation. The polar modulator-based square-QAM transmitter is shown in Fig. 5.18. The
transmitted electric field can be written as
 
pffiffiffiffiffiffi jðωT tþϕ Þ πV MZM ðt Þ πV ðt Þ
E T ðt Þ ¼ PT e T,PN
cos exp j PM , ð5:60Þ
2Vπ Vπ

where with VMZM and VPM we denoted the MZM and PM driving signals, respectively. However, the polar modulator-based
square/cross transmitter requires the nonlinear transformation of the QAM constellation points’ coordinates.
The square
/cross-QAM transmitter employing an alternative E-O I/Q modulator is shown in Fig. 5.19. In the same figure,
the configuration of the alternative E-O I/Q modulator is shown. The key difference with respect to conventional E-O I/Q
modulator is the introduction of additional E-O PM in an in-phase branch. Notice that, from a communication theory point of
view, this additional PM is redundant. However, its introduction provides more flexibility in the sense that simpler driving
signals can be used to generate a desired QAM constellation. The MZM and E-O PM driving signals in an I-branch are
given by
5.2 Multilevel (Two-Dimensional) Modulation Schemes 269

bm Phase shaper (PS)

…
Sequence S/P QAM PS
convertor b2 Mapper V(MZM)I(t)
of bits (1:m DEMUX) PS
b1
PS V(PM)I(t)

MZM PM

CW
MZM 3 dB V(MZM)Q(t) V(PM)Q(t) 3 dB
Laser

Sinusoidal MZM PM - /2
signal

Fig. 5.19 The operation principle of an alternative I/Q modulator-based square-QAM transmitter. PS pulse shaper; I/Q modulator is shown in
Fig. 5.17; 3 dB, 3 dB coupler

" #
ðMZM Þ V X
1
π
VI ðt Þ ¼2 π sin
1
ðjb1 ðkÞjÞpðt
kT s Þ
,
π k¼
1
2
ð5:61Þ
ðPM Þ V X
1
V I ðt Þ ¼ π ð1
signðb1 ðkÞÞÞpðt
kT s Þ,
2 k¼
1

while the MZM and PM driving signals in a Q-branch are given by

" #
ðMZM Þ V X
1
π
VQ ðt Þ ¼2 π sin
1
ðjb2 ðkÞjÞpðt
kT s Þ
,
π k¼
1
2
ð5:62Þ
ðPM Þ V X
1
V Q ðt Þ ¼ π ð1
signðb2 ðkÞÞÞpðt
kT s Þ:
2 k¼
1

The output electric field of the alternative I/Q modulator-based square
/cross-QAM transmitter can be represented in polar
coordinates as

pffiffiffiffiffiffi jðωT tþϕ Þ

E T ðt Þ ¼ PT e T,PN
Rðt Þ exp ð jΨðt ÞÞ, ð5:63Þ

where the envelope R(t) and phase Ψ (t) are given, respectively, as

h i1=2
1 2
Rðt Þ ¼ RI ðt Þ þ R2Q ðt Þ þ 2RI ðt ÞRQ ðt Þ cos ðΨI ðt Þ
ΨQ ðt ÞÞ ,
2   ð5:64Þ
RI ðt Þ sin ΨI ðt Þ
RQ ðt Þ sin ΨQ ðt Þ
Ψðt Þ ¼ tan
1 ,
RI ðt Þ cos ΨI ðt Þ
RQ ðt Þ cos ΨQ ðt Þ

with RI (RQ) and ΨI (ΨQ) defined by

ðMZMÞ
! ðPMÞ
1 πV I,Q ðt Þ 1 πV I,Q ðt Þ
RI,Q ðt Þ ¼ cos , ΨI,Q ðt Þ ¼ :
2 Vπ 2 Vπ
270 5 Advanced Modulation and Multiplexing Techniques

bm Pulse shaper (PS)

…
…
Sequence S/P QAM PS
convertor b2 Mapper V(1)I(t)
of bits (1:m DEMUX) PS

…
b1
PS V(1)Q(t)
1

I/Q Modulator Attenuator

CW Power Power

…
MZM V(n)I(t) V(n)Q(t)
Laser splitter combiner

Sinusoidal I/Q Modulator Attenuator

signal
n

Fig. 5.20 The operation principle of a parallel I/Q modulator-based square-QAM transmitter. Attenuation coefficient of the ith branch is specified
by αi ¼ (i
1)  6 dB

bm Pulse shaper (PS)

Sequence PS
S/P bm-1
convertor
of bits bm-2 PS
(1:m DEMUX) QAM
…

b1 Mapper
PS
VI(t) VQ(t) V1(t), V2(t),
V V /2

CW
MZM I/Q Modulator PM PM
Laser

Sinusoidal
signal

Fig. 5.21 The operation principle of a tandem-QAM transmitter for optical communications

The parallel I/Q modulator-based square
/star-QAMtransmitter is shown in Fig. 5.20. The CW signal is split into
n branches, with each branch containing one I/Q modulator and attenuator. The attenuation coefficient of the ith branch is
given as αi ¼ (i
1)  6 dB. The key idea behind this approach is to simplify a generation of larger constellation sizes by
properly combing smaller constellation sizes. For instance, 16-QAM can be generated by summing up two QPSK signals,
while one of the QPSK signals is being attenuated by 6 dB.
The tandem-QAM transmitter is shown in Fig. 5.21, based on [9]. The I/Q modulator is used to generate the initial QAM
constellation residing in the first quadrant only. Two consecutive PM modulators are used to introduce the phase shifts of π/
2 and π, respectively. With this additional manipulation, we essentially rotate the constellation diagram residing in the first
quadrant for π/2 and  π, respectively, thus creating the resulting QAM constellation residing in all four quadrants. The first
m-2 bits are used to select a point from the initial signal constellation residing in the first quadrant. The bit bm
1 is used to
control the first PM so that, after the pulse shaper (PS), a phase shift of π/2 is introduced. The bit bm is used to control the
second PM so that after the PS a phase shift of π is introduced. The resulting signal constellation has 2m constellation points
located in all four quadrants.
5.3 Multicarrier Modulation 271

5.3 Multicarrier Modulation

The key idea behind the multicarrier modulation is to divide the transmitted bit stream into many lower-speed substreams,
which are typically orthogonal to each other, and send these over many different subchannels [1–3, 5–7]. For wireless
communication applications, the number of substreams is chosen to ensure that each subchannel has a bandwidth smaller than
the coherence bandwidth of the channel, so that each subchannel will experience the flat fading. This is an efficient way to deal
with frequency-selective fading effects. For fiber-optics communications, the guard interval is chosen in such a way that the
delay spread due to dispersion effects is smaller than the guard interval so that these dispersion effects can be successfully
compensated.
The discrete implementation of multicarrier modulation, well known as orthogonal frequency division multiplexing
(OFDM) [1–3, 5–7, 13–15], can be successfully used to deal with ISI introduced by various fading and dispersion effects
through the use of concept of cyclic prefix. As the name suggests, the OFDM can be considered as a modulation technique, a
multiplexing technique, or both. This gives the OFDM flexibility that does not exist in conventional single-carrier systems.
The full details of OFDM can be found in Chap. 7; here, we are concerned with some fundamental concepts of multicarrier
modulation. We describe two versions of multicarrier systems: (i) multicarrier systems with nonoverlapping subcarriers and
(ii) multicarrier systems with overlapping subcarriers.

5.3.1 Multicarrier Systems with Nonoverlapping Subcarriers

The multicarrier transmitter configuration is provided in Fig. 5.22, assuming that real-valued signals are transmitted, to
facilitate explanations. If we assume that Nsc subcarriers are used, each having an Rsc symbol rate, the total symbol rate Rs for
multicarrier systems can be calculated as

Rs ¼ N sc Rsc : ð5:65Þ

Let BW denote the signal bandwidth. The corresponding bandwidth per subcarrier will be BWsc ¼ BW/Nsc. When the
wireless channel is frequency selective with coherence bandwidth Bc > BW, for a sufficient number of subcarriers, we can
ensure that the Bc < < BWsc, so that each subcarrier will experience the flat fading. In other words, the symbol duration at the
subcarrier level Tsc ¼ 1/Rsc ffi 1/BWsc > > 1/Bc
1/Δτ, where Δτ is the multipath delay spread of the wireless channel. When
raised-cosine pulses, with roll-off factor β, are used at each subcarrier, the bandwidth occupied per subcarrier will be
BWsc ¼ (1 + β)Rsc. For the multicarrier system with nonoverlapping subcarriers, the guard band Δ/Rsc (Δ < <1) must be
introduced to limit co-channel crosstalk. By choosing the subcarrier frequencies as follows, fn ¼ f1 + (n
1)BWsc (n ¼ 1,
. . ., Nsc), the subcarrier channels will be orthogonal to each other, with each subcarrier channel occupying the bandwidth
BWsc. The total bandwidth of the multicarrier system with nonoverlapping subcarriers will be then

BW ¼ N sc ð1 þ β þ ΔÞRsc : ð5:66Þ

The modulated subcarriers are multiplexed together in an FDM fashion (see Fig. 5.22), and this multicarrier multiplexing can
be expressed as

X
N sc
sðt Þ ¼ sn pðt Þ cos ð2π f n t þ φn Þ, f n ¼ f 1 þ ðn
1ÞBW sc ðn ¼ 1, . . . , N sc Þ, ð5:67Þ
n¼1

where sn is the symbol transmitted by the nth subcarrier. After that, the multicarrier signal is amplified and transmitted with the
help of corresponding antenna.
On the receiver side, to reconstruct the nth subcarrier, the band-pass filter of bandwidth BWsc, centered at fn, is sufficient to
reject the neighboring multicarrier subchannels, as illustrated in Fig. 5.23. Following down-conversion to the baseband, the
rest of the receiver per subcarrier is the same as described earlier.
272 5 Advanced Modulation and Multiplexing Techniques

Subcarrier Symbol Pulse s1(t)

rate Rs/Nsc mapper 1 shaper p(t)
cos 1t

S/P converter
Data rate Rs Oscillator
s(t)

…
Subcarrier Symbol
Pulse sNsc(t)
mapper
rate Rs/Nsc shaper p(t)
Nsc cos Nsct

Oscillator

Fig. 5.22 Multicarrier transmitter configuration for wireless communications (for real-valued signals)

CT matched Symbol
Band-pass
filter
filter 1 t=kTsc demapper 1
h(t)=p( t)

P/S converter
Local Reconstructed
Rates Rs/Nsc=Rsc

…
…

oscillator (LO) 1
data stream
…

at rate Rs
CT matched Symbol
Band-pass
filter demapper Nsc
filter 1
h(t)=p( t)
t=kTsc
Local
oscillator Nsc

Fig. 5.23 The multicarrier receiver for nonoverlapping subcarriers (for real-valued signals)

5.3.2 Multicarrier Systems with Overlapping Subcarriers

Regarding the multicarrier receiver with overlapping subcarriers, the subcarrier frequencies are properly chosen such that the
following principle of orthogonality is satisfied:

Tðsc    
1 m n
cos 2π f1 þ t þ φi cos 2π f 1 þ t þ φn dt
T sc T sc T sc
0
ð5:68Þ
Tðsc 
1 m
n 1
¼ cos 2π t þ φm
φn dt þ . . . ffi δðm
nÞ,
T sc TN 2
0

even for random phase shifts per subcarrier, providing that f1Tsc > > 1. Thanks to this orthogonality of subcarriers, we can
allow for partial overlap of neighboring subcarriers, and we can choose the subcarrier frequencies as fn ¼ f1 + (n
1)
Rsc (n ¼ 1, . . ., Nsc). By assuming again that raised-cosine pulse shape is used on the subcarrier level, with roll-off factor β,
the total occupied bandwidth will be

BW ¼ N sc Rsc þ ðβ þ ΔÞRsc
N sc Rsc : ð5:69Þ

Clearly, for a sufficient number of subcarriers, the bandwidth efficiency of the multicarrier system with overlapping
subcarriers is (1 + β + Δ) times higher than that of the multicarrier system with nonoverlapping subcarriers.
5.3 Multicarrier Modulation 273

CT matched Symbol
filter
t=kTsc demapper 1
h(t)=p( t)

P/S converter
Local Reconstructed
Rates Rs/Nsc=Rsc

…
… oscillator (LO) 1
data stream

…
at rate Rs
CT matched Symbol
filter demapper Nsc
h(t)=p( t)
t=kTsc
Local
oscillator Nsc

Fig. 5.24 Configuration of the multicarrier receiver with overlapping subcarriers (for real-valued signals)

The transmitter configuration of the multicarrier system with overlapping subcarriers is the same that shown in Fig. 5.22,
providing that the principle of orthogonality, given by Eq. (5.68), is satisfied. On the other hand, the receiver configuration of
the multicarrier system with overlapping subcarriers is shown in Fig. 5.24. So the key difference of this receiver with respect
to the corresponding receiver shown in Fig. 5.23 is that the band-pass filters can be omitted, thanks to the orthogonality
principle. To illustrate that subcarrier channels can be properly demultiplexed, without co-channel crosstalk, the input to the
symbol mapper in the nth branch of the receiver is given by

Tðsc " #
X
N sc
bsn ¼ sm pðt Þ cos ð2π f m t þ φm Þ pðt Þ cos ð2π f n t þ φn Þdt ¼
m¼1
0
Tðsc    
P
N sc
m n ð5:70Þ
¼ sm p ðt Þ cos 2π f 0 þ
2
t þ ϕm cos 2π f 0 þ t þ φn dt
m¼1 T sc T sc
0
P
N sc
¼ sm δðm
nÞ ¼ sn ,
m¼1

and clearly, the nth subcarrier symbol is properly detected. Even in the presence of selective fading, the nth subcarrier will see
the flat fading with corresponding channel coefficient H( fn), and the corresponding channel model is given by

r n ¼ H ð f n Þsn þ zn ; n ¼ 1, 2, ⋯, N sc , ð5:71Þ

where zn is the AWGN sample with variance N0BWsc. Clearly, with multicarrier modulation, by employing the orthogonality
principle, we have successfully decomposed the wireless channel into Nsc parallel subchannels.
For complex-valued signals, the multicarrier signal can be represented as
8 (   )
>
< Re
n¼NP
s =2
1
n þ 0:5
snþN sc =2 exp j2π f c
t , 0  t  Ts
sðt Þ ¼ n¼
N s =2 Ts , ð5:72Þ
>
:
0, otherwise

where fc is the carrier frequency and Ts is the symbol duration of the multicarrier signal. The corresponding complex envelope
representation is given by
8 
< P
> n¼N s =2
1
n
s nþN sc =2 exp j2π t , 0  t  Ts
esðt Þ ¼ n¼
N s =2 Ts : ð5:73Þ
>
:
0, otherwise
274 5 Advanced Modulation and Multiplexing Techniques

To perform the transition into the discrete-time domain, we sample the multicarrier signal with a sampling interval ΔT ¼ Ts/N,
where N is the power of 2, and after sampling, we obtain
8 
< P
> n¼N s =2
1
n
s nþN sc =2 exp j2π kΔT , 0  t  Ts
esðkΔT Þ ¼ n¼
N s =2 Ts : ð5:74Þ
>
:
0, otherwise

Given that Ts/ΔT ¼ N, the previous equation can be rewritten as

8
< P
> n¼N s =2
1
s nþN sc =2 exp ½ j2πnk=N , 0  t  Ts
esðkΔT Þ ¼ n¼
N s =2 ð5:75Þ
>
:
0, otherwise

and represents the inverse discrete Fourier transform (IDFT) of 2-D (in-phase-quadrature) sequence {sn}. Therefore, to
efficiently implement multicarrier transmitter and receiver, the fast Fourier transform (FFT) can be used, and this version of
the multicarrier system is commonly referred to as OFDM [1–3, 5–7, 13–15]. For additional details on OFDM as well as
various applications, an interested reader is referred to Chap. 7, devoted to OFDM systems.

5.3.3 Dealing with Fading Effects on Subcarrier Level

As discussed above, in multicarrier modulation, each subcarrier channel is narrowband, so that each subcarrier experiences the
flat fading, which mitigates the multipath effects. Nevertheless, each subchannel experiences flat fading, which can introduce
the deep fade on certain subcarriers. The various methods to compensate for the flat fading effects on subcarrier channels can
be categorized as follows: (i) frequency equalization, which is described in Chap. 6; (ii) precoding; (iii) adaptive loading, also
described in Chap. 6; and (iv) coding with interleaving in time and frequency domains, which is described in Chap. 9. Here
briefly describe the basic multicarrier modulation concepts. In frequency equalization, the flat fading H( fn) on the nth
subcarrier is essentially inverted at the receiver side, the received signal is multiplied by 1/ H( fn), and the resulting resultant
signal power is independent on flat fading since |H( fn)|2Pn/|H( fn)|2 ¼ Pn. Even though the impact of flat fading effects on the
signal is removed, unfortunately, in equalization, in particular for zero-forcing equalizers, the noise is enhanced because the
noise power is proportional to N0BWsc/|H( fn)|2. Different equalization schemes are subject of study in Chap. 6.
Precoding employs the same idea as in frequency equalization, except that the flat fading is inverted at the transmitter side
instead of the receiver. The channel state information related to the nth subcarrier needs to be fed back to the transmitter side
so that transmitter transmits with power Pn/|H( fn)|, so that after the flat fading channel, the receiver will see constant power
since [Pn/|H( fn)|2]|H( fn)|2 ¼ Pn. Because the channel inversion takes place at the transmitter side instead of the receiver, the
noise power at the receiver side will be N0BWsc, and the noise enhancement problem is avoided.
Adaptive loading is based on the adaptive modulation techniques.
The key idea is to vary the data rate and power assigned to each subchannel relative to that subchannel gain.
In adaptive loading techniques, the power and rate on each subcarrier is adapted in such a way to maximize the total rate of
the system by using adaptive modulation techniques, described with more details in Chap. 6. The capacity of the multicarrier
system with Nsc independent subchannels of bandwidth BWsc and subcarrier channel coefficients {H( fn), n ¼ 1, . . ., Nsc},
based on Chap. 4, is given by

X
N sc 
jH ð f n Þj2 Pn
C¼ P
max BW sc log 2 1þ : ð5:76Þ
Pn : Pn ¼P n¼1 N 0 BW sc

The optimum power allocation policy is the water filling in the frequency domain:

Pn 1=ρtsh
1=ρn , ρn  ρtsh
¼ , ρn ¼ jH ð f n Þj2 P=N 0 BW sc , ð5:77Þ
P 0, otherwise

where ρtsh is the threshold SNR. After substituting Eq. (5.77) into (5.76), we obtain
5.4 MIMO Fundamentals 275

X 
ρn
C¼ BW sc log 2 : ð5:78Þ
n:ρn >ρtsh
ρtsh

The key idea of coding with interleaving is to encode information bits into code words, interleave the coded bits over both
time and frequency domains, and then transmit the coded bits over different subcarriers such that the coded bits within a code
word all experience independent flat fading. In such a way, the channel code will be correcting random errors only, instead of
both random and burst errors introduced by flat fading. For additional details on coding with interleaving, an interested reader
is referred to Chap. 9.

5.4 MIMO Fundamentals

Multiple transmitters can be used either to improve the performance through array gain, related to improvement in SNR, and
diversity gain, related to improving the BER slope, or to increase data rate through multiplexing of independent data streams.
In multiple-input multiple-output (MIMO) systems, the transmitters and receivers can exploit the knowledge about the
channel to determine the rank of so-called channel matrix H, denoted as RH, which is related to the number of independent
data streams that can be simultaneously transmitted. When identical data stream is transmitted over multiple transmit
antennas, the corresponding scheme is a spatial diversity scheme. In spatial diversity scheme, M either transmit or receive
branches are properly combined in a linear combiner to improve the SNR. This improvement in SNR is known as the array
gain and can be defined as the linear combiner average SNR ρL to single branch average SNR ρ ratio:

ρL
AG ¼  M, ð5:79Þ
ρ

and the maximum possible array gain is equal to M. The average symbol error probability can be determined from
SNR-dependent expression of symbol error probability Ps(ρ) by averaging out over all possible values of ρ:

ð
1

Ps ¼ Ps ðρÞ f ρL ðρÞdρ, ð5:80Þ

0

where f ρL ðρÞ is the PDF of the linear combiner SNR. The reduction in the symbol error rate due to diversity can be
described by more rapid slope in the corresponding symbol error probability curve. For very high SNRs, the symbol error
probability is reversely proportional to ρ, so that the diversity gain can be defined as

log Ps ðρÞ
DG ¼
lim : ð5:81Þ
ρ!1 log ρ

In MIMO systems, multiple independent data streams are transmitted simultaneously. To specify improvement due to
MIMO, we can define the multiplexing gain as the asymptotic slope in SNR of the outage capacity:

Coutage,p ðρÞ
MG ¼
lim , ð5:82Þ
ρ!1 log 2 ρ

where Coutage, p (ρ) is the p percentage outage capacity at signal-to-noise ratio ρ. The p percentage capacity is defined as the
transmission rate that can be supported in (100-p)% of the channel realizations.
As an illustration, in Fig. 5.25, we provide the basic concept of wireless MIMO communication, with the number of
transmit antennas denoted as MTx and the number of receive antennas denoted as MRx. The 2-D symbol transmitted on the mth
transmit antennas is denoted by xm (m ¼ 1, 2, . . ., MTx), while the received symbol on the nth received antenna is denoted as
yn. The channel coefficient relating the received symbol on the nth receive antenna and transmitted symbol on the mth transmit
antenna is denoted as hmn. By using this model, the output of the nth receive antenna can be represented as
276 5 Advanced Modulation and Multiplexing Techniques

h11
x1 h12 y1
h1M Tx
h21
h22
x2 y2
h2M Tx

…
…
hM Rx 2 hM Rx 1

…
xM Tx hM Rx M Tx yM Rx

Fig. 5.25 The wireless MIMO concept

X
M Tx
yn ¼ hnm xm þ zn , ð5:83Þ
m¼1

where zn is the zero-mean Gaussian noise sample with variance σ2 ¼ N0B, with N0/2 being the double-sided power spectral
density and B being the channel bandwidth.
A narrowband point-to-point wireless communication system composed of MTx transmit and MRx receive antennas can be
represented by the following discrete-time model, which is a matrix representation of Eq. (5.83):

y ¼ Hx þ z, ð5:84Þ

where x, y, and z represent the transmitted, received, and noise vectors defined as
2 3 2 3 2 3
x1 y1 z1
6x 7 6y 7 6z 7
6 2 7 6 2 7 6 2 7
x¼6 7, y¼6 7, z ¼ 6 7: ð5:85Þ
4... 5 4... 5 4... 5
xM Tx yM Rx zM Rx

The channel noise samples zn are complex Gaussian with zero mean and covariance matrix σ2IM Rx, where σ2 ¼ N0B. If the
noise variance is normalized to one then the transmitted power P can be normalized as follows: P/ σ2 ¼ ρ, where ρ denotes the
average SNR per receiver antenna, assuming unity channel gain. The total normalized transmitted power would be then
PM Tx   
m¼1 i xi ¼ ρ: We use H ¼ ½hmn M Rx M Tx in Eq. (5.84) to denote the so-called channel matrix:
x

2 3
h11 h12 ... h1M Tx
6 h h22 ... h2M Tx 7
6 21 7
H¼6 7: ð5:86Þ
4 ... ... ... ... 5
hM Rx 1 hM Rx 2 . . . hM Rx M Tx

The multiplexing gain of a MIMO system, introduced by Eq. (5.82), originates from the fact that a MIMO channel can be
decomposed into a number of parallel independent channels, so that the data rate can be increased RH times, where RH is the
rank of the channel matrix H.
Let us perform the singular value decomposition(SVD) of channel matrix H given by Eqs. (5.84), and (5.86). In SVD, we
have the following: (i) the ∑
matrix corresponds to scaling operation; (ii) the columns of matrix U, obtained as eigenvectors
5.5 Polarization-Division Multiplexing (PDM) and 4-D Signaling 277

…
x x y y

Fig. 5.26 The parallel decomposition of wireless MIMO channel

of the Wishart matrix HH{, correspond to the rotation operation; and (iii) the other rotation matrix V that has for columns the
eigenvectors of H{H. The rank of matrix H corresponds to the multiplexing gain. In few-mode fiber (FMF) and few-core fiber
(FCF)-based applications, we can ensure the full rank of channel matrix, which is not guaranteed when either regular
multimode fiber links or wireless communication links are employed.
The parallel decomposition of the MIMO wireless channel, illustrated in Fig. 5.26, is obtained by introducing the
following two transformations to the channel input vector x and channel output vector y: (i) transmit precoding, in which
the input vector of symbols x to transmitter is linearly transformed by pre-multiplying with rotation matrix V; (ii) receiver
shaping, in which the receiver vector y, upon optical coherent detection, is multiplied by rotation matrix U{. With these two
operations, we effectively performed the following manipulation:
0 1
~y ¼ U { @|{z}
H x þ zA ¼ U { UΣV { V~ U { z ¼ Σ~
x þ |{z} x þ ~z, ð5:87Þ
UΣV { ~z

and successfully decomposed the wireless MIMO channel into RH ¼ rank(H) parallel single-input single-output (SISO)
channels. The equivalent wireless/optical channel model is shown in Fig. 5.27. Therefore, with the help of precoding on the
transmitter side and receiver shaping on the receiver side, we were able to decompose the MIMO channel into RH independent
parallel channels and therefore increase the data rate of the SISO system RH times.
The readers interested in MIMO communications are referred to Chap. 8 for additional details.

5.5 Polarization-Division Multiplexing (PDM) and 4-D Signaling

The electric field of plane linearly polarized electromagnetic waves propagating in the z-direction can be presented as follows:

Eðz, t Þ ¼ Ex ðz, t Þ þ Ey ðz, t Þ ¼ ex E 0x cos ðωt
kzÞ þ ey E0y cos ðωt
kz þ δÞ, ð5:88Þ

where k ¼ 2π/λ is the wave propagation vector magnitude (with λ being the operating wavelength) and δ is the relative phase
difference between the two orthogonal polarizations. In Eq. (5.88), Ex and Ey denote x- (horizontal) and y- (vertical)
polarization states, respectively. By using the Jones vector representation of polarization wave [6, 7], we can represent the
electric field as follows:

 " pffiffiffiffiffiffiffiffiffiffiffiffiffi #
E x ðt Þ 1
pr jðωt
kzÞ
Eðt Þ ¼ ¼ E pffiffiffiffiffi jδ e , ð5:89Þ
E y ðt Þ pr e

where pr is the power splitting ratio between states of polarizations (SOPs), while the complex phasor term is omitted. Since x-
and y-polarization states are orthogonal, they can be used as the additional basis functions. Given the fact that in-phase and
quadrature channels are orthogonal as well, we can conclude that the corresponding space becomes four-dimensional (4-D).
This enables us to use 4-D constellations instead of 2-D ones (PSK and QAM). As compared to 2-D constellations with the
same symbol energy, 4-D constellations can increase the Euclidean distance among neighboring constellation points and thus
improve the SNR [10, 11]. In addition, as shown in [12], 4-D signal constellations, when applied to fiber-optic
communications, are more robust to fiber nonlinearities.
278 5 Advanced Modulation and Multiplexing Techniques

x1 y1

z1
1

x2 y2

z2
2
…

…
xN yRH

z RH
RH

Fig. 5.27 Equivalent parallel MIMO channel. The λi denotes the ith eigenvalue of the Wishart matrix HH {

The generic scheme suitable for wireless communication applications is provided in Fig. 5.25.
In this particular example, which employs dual-polarization transmit and receive antennas, two independent data streams
are combined properly together in a polarization beam combiner. The configurations of RF transmitters corresponding to x-
and y-polarization states are identical to that from Fig. 5.13(a), except from the antenna element. Due to multipath effect, the
initial orthogonality of polarization states is no longer preserved on a receiver side, and we can use the channel coefficients as
shown in Fig. 5.28 to describe this depolarization effect. The receiver now needs to compensate for cross polarization effects
and then perform demodulation in a similar fashion as shown in Fig. 5.13(b). The polarization-division combiner could be just
multiplexing two independent data streams or perform polarization-time coding, in a fashion similar to space-time coding,
which is described in Chap. 8.
The 4-D electro-optical modulator is shown in Fig. 5.29. The CW laser signal is separated into two orthogonal
polarizations by polarization beam splitter (PBS). Each polarization branch contains either single I/Q or polar modulator.
In polarization-division multiplexing (PDM) applications, the QAM constellation coordinates in x- and y-polarization
branches are used, after pulse shaping, as in-phase (I) and quadrature (Q) inputs of the corresponding I/Q modulator. The
independent QAM streams are multiplexed together by a polarization beam combiner (PBC). The I/Q modulator can be
replaced by polar modulators, as indicated in Fig. 5.29(b). Alternatively, various 2-D modulators described in Sect. 5.2 can be
used in polarization branches. In 4-D modulation schemes, a single 4-D mapper, implemented as an LUT, is used to provide
four coordinates for a 4-D modulator. In either case, the first two coordinates (after pulse shaping) are used as inputs to an I/Q
modulator in an x-polarization branch, while the other two coordinates (after pulse shaping) are used as inputs to an I/Q
modulator in a y-polarization branch. The corresponding signal constellation point of the ith symbol interval can be
represented in a vector form as follows:
5.5 Polarization-Division Multiplexing (PDM) and 4-D Signaling 279

y-polarization y-pol.
Tx antenna hyy Rx antenna

hxy
Transmitted hyx

sequence Polarization Estimated seq.

RF Tx Receiver
-division (x-pol)
(x-pol) (x-pol.) hxx
combiner
x-polarization x-pol.

Estimated seq.
Tx antenna Rx antenna

(y-pol)
RF Tx
(y-pol.)
Transmitted
seq. (y-pol)

Fig. 5.28 Generic polarization-division multiplexing scheme for wireless communication applications

Fig. 5.29 The configuration of a

4-D modulator suitable for Ix,k
polarization-division multiplexing
(PDM) and 4-D optical
transmission: (a) Cartesian 3 dB MZM
coordinate-based modulator and
(b) polar coordinate-based
MZM /2
modulator
I/Q modulator to SMF
DFB PBS Qx,k PBC

MZM /2
Qy,k
MZM

4-D modulator Iy,k

(a)

sx,k
v x ,k
j
sx,k e x ,k

AM PM
to SMF
DFB PBS PBC

AM PM j
v y ,k sy,k e y ,k

sy,k
(b)
280 5 Advanced Modulation and Multiplexing Techniques

Fig. 5.30 The PDM vs. 4-D Ix

scheme: (a) PDM transmitter and 2-D symbol
(b) 4-D transmitter data stream Mapper (x pol.)
(x-pol.) Qx
I/Q
modulator
Transmit (x-pol.)
to SMF
laser PBS PBC
diode I/Q
modulator
(y-pol.)
Iy
2-D symbol
data stream Mapper (y pol.)
(y-pol.) Qy

(a)
Ix 4-D modulator

Qx
I/Q
modulator
4-D Transmit (x-pol.) to SMF
4-D symbol
laser PBS PBC
Mapper diode
data stream I/Q
(LUT) modulator
(y-pol.)
Iy

Qy

(b)

0 1 0 1 0 1
Re ðE x,i Þ I x,i jsx,i j cos θx,i
B ImðE x,i Þ C B Q C B jsx,i j sin θx,i C
B  C B x,i C B   C
si ¼ B C¼B C¼B C, ð5:90Þ
@ Re E y,i A @ I y,i A @ sy,i  cos θy,i A
   
Im E y,i Qy,i sy,i  sin θy,i

where I and Q refer to the in-phase and quadrature components, respectively, while x and y refer to the x- and y-polarization
states. In Eq. (5.90), |s| represents the amplitude of the signal constellation point, while θ represents the corresponding phase in
a given polarization. The key differences between PDM and 4-D schemes are outlined in Fig. 5.30. In PDM (see Fig. 5.30(a)),
we employ two independent mappers that drive two independent 2-D data streams through I/Q modulators. The independent
2-D streams are multiplexed together by a polarization beam combiner and transmitted over an optical fiber line. The same
CW laser signal is used for both polarizations, which are separated by polarization beam splitter prior to the modulation. On
the other hand, in 4-D signaling, only one 4-D mapper is used, which provides four coordinates that are used as inputs to the
corresponding I/Q modulators, as shown in Fig. 5.30(b). It is evident from Fig. 5.30(b) that a 4-D modulator is composed of
two I/Q modulators (one per polarization), one polarization beam splitter (PBS), and one polarization beam combiner (PBC).
As an illustration, the 16-point 4-D constellation is described by the set {1,1,1,1} [10]. At the same time, the 32-4-D
constellation contains 16 points mapped to different combinations of vectors {0.5,0.5,0.5,0.5}, 4 points mapped to
combinations of {1,0,0,1}, 4 points mapped to combinations of {0,1,1,0}, and 8 points mapped to the different
combinations of {1,0,0,0}. Even though 4-D schemes are more popular in fiber-optic communications than wireless
communications, the scheme shown in Fig. 5.29 is also applicable in 4-D wireless communication-based signaling.
5.6 Spatial-Division Multiplexing and Multidimensional Signaling 281

It also possible to transmit the same data stream in both polarizations, and we refer to this scheme as the polarization
diversity scheme. The purpose of the polarization diversity scheme is to avoid the need for PMD compensation at the expense
of deteriorating the OSNR sensitivity by 3 dB. The various PMD compensation schemes will be described in Chap. 6. Another
interesting interpretation of a polarization diversity modulation is to consider it as a 2  2 multi-input multi-output (MIMO)
scheme and employ the MIMO signal processing to deal with PMD, chromatic dispersion, polarization-dependent loss (PDL),
and various filtering effects.
The spectral efficiency of 4-D modulation schemes is given by

S ¼ Rs log 2 M 4
D , ð5:91Þ

where Rs is the symbol rate and M4-D is the 4-D signal constellation size. On the other hand, the spectral efficiency of the
polarization PDM scheme is given as

S ¼ 2Rs log 2 M 2
D , ð5:92Þ

where factor 2 accounts for polarization states, while M2-D is the 2-D signal constellation size used in each polarization. Signal
constellation sizes for the same aggregate data rate are related by M 4
D ¼ M 22
D .
For instance, 16-ary 4-D constellation corresponds to PDM-QPSK. Interestingly enough, in the presence of nonlinearities,
4-D modulation schemes of larger constellation sizes show better robustness compared to a PMD-QAM scheme of the same
aggregate rate, as demonstrated in 0.
The 4-D signaling can also be combined with subcarrier multiplexing as described in 0. The characteristics of a dual-
polarization receiver were described in Chap. 3. Another alternative is to combine 4-D signaling with OFDM 0. Also, instead
of subcarrier multiplexing, we can use fiber mode multiplexing, and the corresponding scheme can be called mode-
multiplexed 4-D modulation. Finally, the polarization diversity multiplexing can be used in combination with the polariza-
tion-time (PT) coding, which is in particular suitable for use in combination with OFDM 0. The multidimensional coded
modulation (CM) studies [1, 14, 16, 17] indicate that the simultaneous employment of multiple photon degrees of freedom in
an optical communication channel can enable a substantial increase in bandwidth capacity. Given the high potential of
multidimensional signaling schemes, the next section will be devoted to spatial-division multiplexing and multidimensional
signaling.

5.6 Spatial-Division Multiplexing and Multidimensional Signaling

In spatial-division multiplexing (SDM), we transmit multiple independent data streams over either multipath channels in
wireless communications or spatial modes in SDM fiber-based optical communication systems.

5.6.1 SDM in Wireless Communications

The SDM techniques with MTx transmit antennas and MRx receiver antennas can achieve the spatial rates equal to min (MTx,
MRx). The generic scheme for wireless SDM scheme is shown in Fig. 5.31.
Incoming information data stream might be first encoded by conventional channel code (not shown in the figure), serving
as an outer code. The encoded data stream is forwarded to 1:L demultiplexer, where serial-to-parallel conversation takes place.
The key transmitter performs simultaneously channel encoding and modulation process thus performing L to MTx mapping.
The mth (m ¼ 1, 2, . . ., MTx) transmit antenna emits the symbol xm. These MTx symbols are transmitted over wireless MIMO
channel of interest such as rich-scattering channel, described by the channel matrix H. The maximum number of independent
streams is determined by the rank of channel matrix RH ¼ rank(H), as described in Sect. 5.4, which is upper bounded by
RH  min(MTx, MRx). In order to properly detect independent data streams, typically the number of receive antennas must be
larger than the number of transmitted antennas so that the maximum possible number of independent data streams is upper
bounded by MTx. Let RTx denote the code rate of both outer and inner channel codes (obtained as a product of individual code
rates); then the spectral efficiency (S) of the SDM scheme is upper bounded by
282 5 Advanced Modulation and Multiplexing Techniques

Rich-scattering
channel
1 h11 1

1 x1 x1 1
h1M Tx
(Serial-to-parallel converter)

(Parallel-to-serial converter)
2 2
Demultiplexer 1:L

Channel Channel

Multiplexer L:1
x2

SDM detector
2 x2 2
Incoming encoder decoder
hM Rx 1
data stream + +

…
…

…
…

…
Modulation MTx MRx Demodulation

L hM Rx M Tx L
xM Tx xM Tx

Fig. 5.31 Generic SDM scheme for wireless MIMO communications

SSDM  M Tx RTx log 2 ðM Þ ½bits=s=Hz, ð5:93Þ

where M is the signal constellation size (assuming that all independent data streams employ the same constellation size). At
the receiver side, each receive antenna detects the complex linear combination of signal from all transmit antennas. The
purpose of SDM detector is to compensate for spatial interference and provide accurate estimates of transmitted symbols.
After SDM detection, demodulation takes place, followed by the channel decoding. The L estimated data streams are
multiplexed together, and the corresponding output is passed to outer channel decoder, which delivers estimated information
sequence to the end users. The SDM concept is also applicable to both free-space (FSO) optical channels and fiber-optic
communications. Below, we describe the hybrid multidimensional coded modulation scheme, introduced in [16–21].

5.6.2 SDM and Multidimensional Signaling in Fiber-Optic Communications

The hybrid multidimensional modulation schemes can be effectively implemented with the help of SDM-based fibers; in
particular, the few-mode fibers (FMFs) and few-core fibers (FCFs) are excellent candidates for this purpose. It is well known
that photon can carry both spin angular momentum and orbital angular momentum (OAM) [22–29, 40, 46]. The spin angular
momentum is associated with polarization, while OAM is associated with the azimuthal phase of the complex electric field.
The angular momentum, L, of the classical electromagnetic field can be written as [22].
ð ð X
1 1
L¼ ðE  AÞdV þ E ðr
k¼x,y,z k
 —ÞAk dV, ð5:94Þ
4πc V 4πc V

where E is the electric field intensity, A is the vector potential, and c is the speed of light. A is related to the magnetic field
intensity H by H ¼ ∇  A and to the electric field intensity by E ¼
c
1∂A/∂t. The second term in Eq. (5.94) is identified as
the “OAM” due to the presence of the angular momentum operator r  —. In general, among various optical beams that can
carry OAM, the Laguerre-Gauss (LG) vortex beams/modes can easily be implemented. The field distribution of an LG beam
traveling along the z-axis can be expressed in cylindrical coordinates (r,ϕ,z) (r denotes the radial distance from propagation
axis, ϕ denotes the azimuthal angle, and z denotes the propagation distance) as follows [28, 29]:
5.6 Spatial-Division Multiplexing and Multidimensional Signaling 283

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  pffiffiffi jmj  2

2p! 1 r 2 2r
um,p ðr, ϕ, zÞ ¼ Lm
p 
π ðp þ jmjÞ! wðzÞ wðzÞ w 2 ðzÞ ð5:95Þ
jkr 2 z
2


2r
e ð Þe
2 z2 þz2 jð2pþjmjþ1Þ tan
1 zz
jmϕ
e w ðzÞ R eR ,
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where wðzÞ ¼ w0 1 þ ðz=zR Þ2 (w0 is the zero-order Gaussian radius at the waist), zR ¼ πw20 =λ is the Rayleigh range (with λ
being the wavelength), k ¼ 2π/λ is the wave number, and Lmp() is the associated Laguerre polynomial, with p and
m representing the radial and angular mode numbers, respectively. It can be seen from Eq. (5.95) that the mth mode of the
LG beam has the azimuthal angular dependence given by the term exp(
jmϕ), where m is the azimuthal mode index. In free
space, for m ¼ 0, u(r,ϕ,z) becomes a zero-order Gaussian beam known as TEM00 mode. For p ¼ 0, Lmp() ¼ 1 for all m’s, so
that the intensity of an LG mode is a ring of radius proportional to |m|1/2.
It can be shown that for fixed p, the following principle of orthogonality is satisfied [22–26, 29]:
ð
 
um,p jun,p ¼ um,p ðr, ϕ, zÞun,p ðr, ϕ, zÞrdrdϕ
8ð
< u 2 rdrdϕ, n ¼ m ð5:96Þ
m,p
¼ :
:
0, n 6¼ m

Evidently, different OAM “states” corresponding to a fixed index p are all mutually orthogonal, and they can be used as the
basis functions for an OAM modulation. Instead of OAM modes, in fiber optics, other spatial modes can be used such as LP
modes [30–39]. However, on a quantum level, only SAM and OAM of the photon can be defined. Moreover, other spatial
modes can be represented in terms of OAM modes.
Different signal constellations can be used in combination with OAM modulations, thus producing multidimensional
hybrid modulation schemes. These schemes are called hybrid in [16–21] since they employ all available degrees of freedom:
the amplitude, the phase, the polarization, and the OAM. As an illustration, for each polarization state, we can employ
L electrical basis functions, such as L orthogonal subcarriers [14], so we have that

Φl ðnT Þ ¼ exp ½ j2π ðl
1ÞΔT=T ; l ¼ 1, ⋯, L, ð5:97Þ

where Φl(nT) define the basis functions, T is the symbol duration, and ΔT denotes the sampling interval that is related to the
symbol duration by ΔT ¼ T/U, with U being the oversampling factor. The corresponding space described by Eq. (5.97) is 2 L-
dimensional. In addition, we can apply N orthogonal either OAM states or other spatial modes and two polarization states
for use in multidimensional modulation. Therefore, the corresponding signal space is D ¼ 4 LN-dimensional. By increasing
the number of dimensions, we can also increase the aggregate data rate of the system while ensuring reliable transmission at
these ultrahigh speeds using capacity-approaching LDPC codes, described in Chap. 9.
The D-dimensional space, if compared to the conventional 2-D space, can provide larger Euclidean distances between
signal constellation points, resulting in improved BER performance and better tolerance to fiber nonlinearities. We should
outline that modified orthogonal polynomials or any other set of orthogonal complex basis functions can be used as well, such
as Slepian sequences based on [18–21]. The overall multidimensional transmitter configuration is depicted in Fig. 5.32, while
the corresponding multidimensional receiver in Fig. 5.33. The D-dimensional modulator, whose configuration is shown in
Fig. 5.32(a), generates the signal constellation points by

X
D
si ¼ ϕi,d Φd , ð5:98Þ
d¼1

where ϕi,d denotes the dth coordinate (d ¼ 1, . . ., D) of the ith signal constellation point and the set {Φ1,. . .,ΦD} denotes the
basis functions introduced above. The transmitter architecture is provided in Fig. 5.32(a). A continuous wave laser diode, such
as DFB laser, signal is split into N branches by using a power splitter to feed 4 L-dimensional electro-optical modulators, each
corresponding to one out of N spatial modes, such as LP or OAM modes. The 4 L-dimensional electro-optical modulator is
implemented as shown in Fig. 5.32(b). The SDMA multiplexer combines N independent data streams for transmission over
284 5 Advanced Modulation and Multiplexing Techniques

Coordinates
from 1 to 4L
…
4LN-dim. mod.
4L-dimensional
modulator SDM fiber
DFB Power Spatial-division

…
laser splitter multiplexer
4L-dimensional
modulator
…
Coordinates
from 1 to 4L

(a)
Complex coordinates
I … L

Up-
U U=T/ T U sampl.

DT pulse- DT pulse-
Coordinates shaping filter shaping filter
from 1 to 2L hM(n) = L(n T)
h1(n) = 1(n T)
…
2L-dimensional
modulator
From laser To SDM
PBS 4L-dim. mod. PBC 2L-dim.
diode multiplexer Re{} Im{}
2L-dimensional
modulator modulator
… DAC DAC

Coordinates to PBC
From PBS
from 1 to 2L I/Q MOD

(b) (c)
Fig. 5.32 The overall transmitter architecture of a discrete-time energy-efficient hybrid D-dimensional modulation scheme: (a) transmitter
configuration, (b) 4 L-dimensional modulator, and (c) 2 L-dimensional modulator. PBS(C) polarization beam splitter (combiner)

SDM fibers, such as FMF, FCF, and few-mode and few-core fibers. As an illustration, when OAM modes are employed as
spatial modes, the corresponding OAM multiplexer is composed of N waveguides, taper-core fiber, and FMF, all properly
designed to excite orthogonal OAM modes in FMF [17–19]. The 4 L-dimensional modulator is composed of two 2 L-
dimensional modulators, one for each polarization, whose discrete-time (DT) implementation is shown in Fig. 5.32(c). The
signal constellation point coordinates after upsampling are passed through corresponding DT pulse shaping filters having
impulse responses hm(n) ¼ Φm(nΔT ), whose outputs are combined together into a single complex data stream. The real and
imaginary parts of this complex data stream, after digital-to-analog conversion (DAC), drive RF inputs of I/Q modulator. The
outputs from M-dimensional modulators are combined into a single SDM stream by the polarization beam combiner (PBC).
The obtained N SDM streams are combined by the spatial-division multiplexer.
5.6 Spatial-Division Multiplexing and Multidimensional Signaling 285

Complex coordinate estimates

Estimated coordinates
’1 (mT) … ’L(mT)
from 1 to 2L
…
Resampling:
2L-dimensional
m=n/U
4L-dimensional demodulator
demodulator Matched Matched
SDM fiber Balanced coherent filter filter
Spatial-division optical detector
From SDM h1(n)= * ( n T) hM(n) = * ( n T)

…
1 L
demultiplexer PBS 4L-dimensional demodulator
demultiplexer Balanced coherent Re{} +j Im{}
4L-dimensional optical detector
demodulator 2L-dim.
ADC ADC
demod.
(a) 2L-dimensional
demodulator From PBS Balanced coherent
… optical detector
Estimated coordinates
from 1 to 2L

(b) (c)

Fig. 5.33 Receiver architecture of a discrete-time energy-efficient hybrid D-dimensional modulation scheme: (a) receiver, (b) 4 L-dimensional
demodulator, and (c) 2 L-dimensional demodulator configurations

On the other hand, on the receiver side, the 4 LN-dimensional demodulator architecture is shown in Fig. 5.33(a).
We first perform SDM/OAM mode demultiplexing in the SDM/OAM-demultiplexing block, whose outputs are 4 L-
dimensional projections along N spatial modes/OAM states. Each spatial/OAM mode undergoes polarization diversity
coherent detection, as shown in Fig. 5.33(b), and the corresponding outputs are forwarded to 2 L-dimensional demodulators,
implemented as shown in Fig. 5.33(c). After polarization diversity detection, we recover Re and Im parts, which are after
ADCs combined into a single complex data stream. The same complex data stream is applied to the inputs of L complexed
matched filters with impulse responses hl(n) ¼ Φ*l(
nT). The corresponding outputs after resampling represent projections
along basis functions Φl.

5.6.3 SDM and Multidimensional Signaling in Free-Space Optical (FSO) Communications

Thanks to the mutual orthogonality among the electric fields of OAM modes with different integer OAM states, as given by
Eq. (5.96), one can in principle superimpose as many OAM modes as one wishes [24–26]. OAM state superposition using
diffractive elements enables implementation of OAM-based multiplexers as an integral part of an FSO communication link.
The simplest mathematical expression of an OAM mode and zero-order Gaussian beam with an incidence angle θ can be
written, respectively, as [25, 26]

uðr, ϕ, zÞ ¼ exp ð
jmϕÞ exp ð
jkzÞ and u0 ¼ exp ðjkx sin θÞ exp ð
jkzÞ: ð5:99Þ

A computer-generated hologram (CGH) is a recording of the interference pattern between two incident beams, in our case a
zero-order Gaussian beam and the beam we want to generate. The resulting interference pattern, assuming that z ¼ 0, can be
expressed as

I ¼ juðr, ϕ, z ¼ 0Þ þ u0 j2 ¼ 2 þ 2 cos ðkx sin θ
mϕÞ: ð5:100Þ

Equation (5.100) is a sinusoidal grating pattern, which is easy to generate but with low diffraction efficiency [35]. It is well
known that blazed grating can obtain 100% diffraction efficiency, whose interference pattern can be expressed as
I ¼ kx sin θ
mϕ mod 2π..
As an illustration, the SLM display programmed with blazed interference pattern is shown in Fig. 5.34(a)–(c). The
experimentally generated (Fig. 5.34(d)–(f)) and analytical (Fig. 5.34(g)–(i)) intensity distributions of LG beams can be
compared. In Fig. 5.35, the detection efficiency of individual OAM mode is illustrated in terms of intensity profile after the
detection. The imperfections in experimental generation of OAM modes, visible in Figs. 5.34 and 5.35, are contributed to the
finite resolution of SLMs. To our best knowledge, even though phase-only SLMs are used, it is impossible to realize in
286 5 Advanced Modulation and Multiplexing Techniques

Fig. 5.34 The phase pattern,

experimental, and analytical
intensity distributions: (a), (d), (g)
for OAM state 1; (b), (e), (h) for
OAM states 2 and
3; (c), (f), (i)
for OAM states 4 and
4

practice phase-only modulation without any residual amplitude modulation. This residual amplitude modulation may generate
the residual ring outside the desirable OAM mode in the far field.
The overall configurations of OAMN-dimensional modulator/multiplexer and demultiplexer, based on spatial light
modulators (SLMs), are shown in Figs. 5.36(a) and (b), respectively. A CW laser signal is split into N branches with the
help of 1:N power splitter, whose every output is used as an input to the E/O MOD. After E/O optical conversion, in each
branch, CGH (based on SLM) is used to impose the corresponding OAM mode. The OAM-based N-dimensional modulator is
composed of power splitter, N E/O MODs, N SLMs, power combiner, and corresponding waveguides.
On the receiver side, the OAM multiplexed signal is demultiplexed with the help of corresponding complex-conjugate
CGHs (see Fig. 5.36(b)), so that the ith output represents a projection along the ith OAM mode. The local laser is then used to
coherently detect the signals transmitted in all OAM modes. To improve OAM modes’ orthogonality, the MIMO signal
processing approaches, described in Chap. 8, can be used.

5.7 Optimum Signal Constellation Design

Different optimization criteria can be used in optimum signal constellation design including minimum MSE (MMSE)
[41, 42], minimum BER, and maximum information rate (channel capacity) [43].
5.7 Optimum Signal Constellation Design 287

Fig. 5.35 Experimental (top) and numerical (bottom) detection of OAM states from the superimposed OAM modes: (a), (e) OAM state 1 is
detected using OAM state
1; (b), (f) superposition of OAM states 4,
4 is detected using OAM state 4; (c), (g) superposition of OAM states 2,
2,
6,
6 is detected using OAM state 2; (d), (h) superposition of OAM states 2,
2, 6,
6 is detected using OAM state
6. Imperfections in
experimental generation of OAM modes are contributed to finite resolution of SLMs

Fig. 5.36 SLM-based Coordinate 1

configurations of (a) OAM N-
…
dimensional modulator and
multiplexer and (b) OAM
OAM N-dim. modulator OAM multiplexer
demultiplexer E/O MOD CGH 1
…

Laser Power Power

E/O MOD CGH 2
diode splier combiner
…

E/O MOD CGH N

Coordinate N
(a)
OAM demultiplexer
CGH* 1
…

Power CGH* 2
splier
…

CGH* N

(b)
288 5 Advanced Modulation and Multiplexing Techniques

5.7.1 Iterative Polar Modulation (IPM)

We first describe the iterative polar modulation (IPM) based on iterative polar quantization (IPQ) [42]. We can assume that,
for an ASE noise dominated channel model, the optimum information source is Gaussian with distribution:
 2
1 I þ Q2
pðI, QÞ ¼ exp
, ð5:101Þ
2πσ 2 2σ 2

where (I,Q) denote in-phase and quadrature coordinates. By expressing Cartesian coordinates through polar coordinates (r,θ)
as I ¼ rcosθ and Q ¼ rsinθ, it can be shown that radius r will follow a Rayleigh distribution [42], while the phase stays
uniform. This indicates that the optimum distribution, in an MMSE sense, contains constellation points placed on circles that
follow a Rayleigh distribution. The number of points per circle varies, while the distribution of points on a circle is uniform.
Let us also introduce the following denotations: Li for the number of constellation points per circle of radius mi; Lr for the
Pr
number of circles in the constellation; Q for the total number of signal constellation points (Q ¼ Li¼1 Li ); and p(r) for the
Rayleigh distribution function of radius r.
The IPMsignal constellationdesign algorithm can be formulated through the following steps [41, 42]:

Step 0: Choose an arbitrary signal constellation of size Q for initialization.

Step 1: Determine the number of constellation points per ith circle as
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi3
u rðiþ1 u rðiþ1
u u
6X 1 u
L r
u
3 3 7
Li ¼ tmi 2 pðr Þdr =4 tm2i pðr Þdr 5; i ¼ 1, 2, . . . , Lr : ð5:102Þ
i¼2
Q
ri ri

Step 2: Determine the radius of the ith circle as

2 rð
iþ1
3 2 rð
iþ1
3
mi ¼ 42 sin ðΔθi =2Þ rpðr Þdr 5=4Δθi pðr Þdr 5,
ð5:103Þ
ri ri
Δθi ¼ 2π=Li ; i ¼ 1, 2, . . . , Lr :

Iterate steps 1–2 until convergence. The limits of integration, which correspond to decision boundaries, in steps 1–2 are
determined by
   
r i ¼ π m2i
m2i
1 =2 =½mi Li sin ðΔθi =2Þ
mi
1 Li
1 sin ðΔθi
1 =2Þ;
ð5:104Þ
i ¼ 1, 2, . . . , Lr :

Therefore, in addition to IPM signal constellation in polar coordinates, this algorithm can also provide the decision
boundaries. As an illustration of an IPM algorithm, the 64-IPQ/IPM signal constellation is shown in Fig. 5.37. With r3 and r5
we denoted the decision boundaries for circle of radius m4. The details of this constellation are provided in Table 5.2. With
index i we denoted the index of a circle counted from the center. The 256-ary IPM has been experimentally evaluated in [43],
and it exhibits much better tolerance compared to 256-QAM.
We should point out that in the 64-IPQ example we did not use the signal constellation point placed in origin, and for large
signal constellations it does not really matter if the point placed in origin is used or not. However, for moderate and small
signal constellation sizes, it is an advantage, in terms of energy efficiency, to place the point in the origin. Such obtained signal
constellations can be called centered-IPQ/IPM (CIPQ/CIPM). As an illustration, 16-ary and 32-ary CIPQ constellations are
shown in Fig. 5.38.
5.7 Optimum Signal Constellation Design 289

Fig. 5.37 Details of a 64-IPQ/ 3

IPM signal constellation
IPQ-64 r5
2

Qudrature coordinate, Q
r4

1 m5
m3
0

-1

-2

-3
-3 -2 -1 0 1 2 3
In-phase coordinate, I

Table 5.2 Details of 64-IPQ/IPM constellation

i 1 2 3 4 5 6
ri 0 0.54 1.02 1.55 2.25 4.5
mi 0.33 0.78 1.26 1.84 2.61
Li 5 11 15 17 16

2.0 16-CIPQ
32-CIPQ
Quadrature coordinate, Qx (Qy)

2.0
1.5
Quadrature coordinate, Qx (Qy)

1.5
1.0
1.0
0.5
0.5
0.0
0.0
-0.5
-0.5
-1.0 -1.0
-1.5 -1.5
-2.0 -2.0
-2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
In-phase coordinate, Ix (Iy) In-phase coordinate, Ix (Iy)

(b)
Fig. 5.38 The CIPQ-based signal constellations for (a) 16 points and (b) 32 points
290 5 Advanced Modulation and Multiplexing Techniques

5.7.2 Signal Constellation Design for Rotationally Symmetric Optical Channels

Another algorithm for the signal constellation design algorithm for “rotationally symmetric optical channels” with coherent
detection is introduced in [43]. The rationale of this algorithm is dictated by the fact that the optical channel has memory
which requires the knowledge of conditional probability density functions (PDFs), which are typically estimated by the
evaluation of histograms. (As a consequence, gradient-based optimization methods cannot be directly applied in such a case.)
It was proposed in [43] to define the admissible set of constraints as follows:
( )
X
n X
n pffiffiffiffiffi
Ωðp; r Þ ¼ p: pi r 2i  Pa , pi ¼ 1, 0  r i  Pp ,
i¼1 i¼1
ð5:105Þ
i ¼ 1, 2, . . . , Lr

where with Pa we denoted the average power (AP) constraint and with Pp the peak power (PP) constraint. Further, ri denotes
the ith “mass point” (radius), while pi denotes the corresponding probability of occurrence. These constraints are introduced to
take the physical limitations of the system into account and minimize the effect of nonlinearities. The optimization problem is
solved by the following split-step optimization algorithm. After the initialization by conventional 2-D constellation, we
initiate the iterative steps below until either convergence is achieved or a predetermined number of iterations is reached [43]:

Step 1: piupdate rule:

pi ¼ arg max fI ðpÞ : p 2 Ωðp; r i
1 Þg: ð5:106Þ

p

Step 2: riupdate rule:

r i ¼ arg max fI ðpÞ : r 2 Ωðpi ; r Þg: ð5:107Þ

r

It can be shown that the sequence {I( pi)}, where I() denotes the mutual information, is a nondecreasing and converges to
the channel capacity.

5.7.3 Energy-Efficient Signal Constellation Design

Yet another signal constellation algorithm aimed for improved energy efficiency was introduced in [17] (see also [8]). The
basic energy-efficient optical communication problem can be formulated as follows. The set of symbols X ¼ {x1, x2, . . ., xQ}
that occur with a priori probabilities p1,. . .,pQ [pi ¼ Pr(xi)] and carrying energies E1,. . .,EQ are to be transmitted over the
optical channel of interest. The symbols from the set X satisfy the following two constraints: (i) ∑ipi ¼ 1 (probability
constraint) and (ii) ∑ipiEi  E (energy constraint). We can use the Lagrangian method [8, 47] in maximizing the mutual
information I(X,Y ), defined as I(X,Y) ¼ H(X)-H(X|Y ), so we have that [17].

H ðX Þ
0 H ðXjY Þ
1
zfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflffl{
X zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
X X
X 
X 
B C ð5:108Þ
L ¼
i pi log pi
@
i pi j
P ij log Q ji A þ η i
p i
1 þ μ i
p i E i
E ,

where Pij ¼ Pr(yj|xi) can be determined by the evaluation of histograms and Qji ¼ Pr(xi|yj) from Bayes’ rule [48]: Qji ¼ Pr(xi,
yj)/Pr(yj) ¼ Pijpi/∑kPkjpk. The energy-efficient signal constellation designalgorithm (EE-SCDA) can be now formulated as
follows [17].

Step 0 (Initialization step): Choose an arbitrary auxiliary input distribution and signal constellation, with a number of
constellation points Ma much larger than target signal constellation M.
5.8 Nonuniform Signaling 291

Step 1: Qjiupdate rule:

ðt Þ ðt Þ
X ðt Þ  
Qji ¼ Pij pi = P p , Pij
k kj k
¼ Pr y j jxi : ð5:109Þ

Step 2: piupdate rule:

 
ðtþ1Þ exp
μE i
H ðtÞ ðxi jY Þ
pi ¼P  ðt Þ
,
k exp
μE k
H ðxk jY Þ ð5:110Þ
X
H ðxi jY Þ ¼
k Pik log Qki :

Iterate steps 1–2 until convergence.

Step 3: Determine the constellation points of target constellation as a center of mass of closest Ma/M constellation points in the
auxiliary signal constellation.

Notice that the well-known Arimoto-Blahut algorithm [48, 49] does not impose an energy constraint and yields the
optimum source distribution only. Using EE-SCDA, we can obtain the optimized signal constellation while taking the energy
constraint into account. Both optimum source distribution and EE signal constellation are obtained from this algorithm.

5.7.4 Optimum Signal Constellation Design (OSCD)

Finally, we will describe the optimum signal constellation design (OSCD)algorithm, introduced in [50] (see also [51]). This
algorithm is convenient since it does not require the employment of constrained optimization software. Instead, a simple
Monte Carlo simulation can be used in signal constellation design. Given that the IPM algorithm is derived under assumptions
that are valid for large constellations, we can say that OSCD-generated constellations are the most favorable for small and
medium sizes of signal constellations. The OSCD can be formulated as follows:

Step 0 (Initialization): Choose an arbitrary signal constellation of size Q.

Step 1: Apply the Arimoto-Blahut algorithm to determine an optimum source distribution.
Step 2: Generate a long sequence of samples from the optimum source distribution. Group the samples from this sequence into
Q clusters. The membership of the cluster is determined based on a squared Euclidean distance between the sample point
and the signal constellation points from previous iteration. Each sample point is assigned to the cluster with the smallest
squared distance.
Step 3: Determine the signal constellation points as the center of mass for each cluster.

Repeat steps 2–3 until convergence.

As an illustration, we show 32-ary and 64-ary constellations obtained by employing this algorithm in Fig. 5.39.

5.8 Nonuniform Signaling

In the conventional data transmission schemes, each point in a given signal constellation is transmitted equally likely [1–7,
44, 45]. However, if we take the different energy costs of various constellation points into account and transmit each point in
the constellation diagram with different probabilities, the resulting modulation technique is called either nonuniform signaling
[5, 52, 53] or probability shaping-based signaling [54]. The basic idea of the nonuniform signaling is to transmit symbols of
larger energy with lower probabilities, so that the energy efficiency of conventional modulation schemes can be improved
[52]. This improvement in energy efficiency is commonly referred to as the shaping gain. In any data transmission scheme, the
goal is to transmit at a higher bit rate, with higher reliability and as low transmitter power as possible. When the distribution of
constellation points matches Maxwell-Boltzmann distribution, the ultimate shaping gain of 1.53 dB can be achieved
[53]. Moreover, the nonuniform signaling scheme may be more suitable for optical communication because the symbols
with small energies, transmitted with higher probabilities, suffer less fiber nonlinearity effects.
292 5 Advanced Modulation and Multiplexing Techniques

32-OSCD 3 64-OSCD
2
Quadrature coordinate, Qx (Qy)

Quadrature coordinate, Qx (Qy)

2
1
1

0 0

-1 -1

-2
-2
-3
-2 -1 0 1 2 -3 -2 -1 0 1 2
In-phase coordinate, Ix (Iy) In-phase coordinate, Ix (Iy)

(a) (b)
Fig. 5.39 Signal constellations obtained by an OSCD algorithm: (a) 32-ary signal constellation and (b) 64-ary signal constellation

In this section, based on our recent invited paper [52], we introduce the nonuniform signaling techniques suitable for use in
coherent optical communication systems. There are two basic ways to map the uniform binary source bits to nonuniform
distributed symbols. The first method is to use arithmetic code [54]. This method can be applied to existing signal
constellation sets like 16-QAM, 64-QAM, and so on, to map the uniform binary source to nonuniform symbols based on
arbitrary distributions, which are based on different criteria. Even though this scheme can easily fit to current optical systems,
its complexity is high. Moreover, it is only applicable to square-QAM constellations.
The second method is to use the Huffman code, which is based on a binary tree structure and has much lower complexity.
This method can be combined with either binary or nonbinary FEC code, but it cannot be applied to arbitrary constellation sets
[55]. The structure of the constellation is determined by the Huffman code so that the design of new signal constellation set is
needed. One of the corresponding constellation sets, 9-QAM, has the same spectral efficiency with traditional star-8-QAM.
On the other hand, the LDPC-coded 9-QAM scheme outperforms the LDPC-coded 8-QAM case by at least 0.8 dB [55].
The fundamental method of nonuniform signaling with arithmetic code is introduced in [54] by Böcherer, and then
Fehenberger and Böcherer proposed applying this approach to optical communication systems [56, 57]. In [57], the authors
extended the work on nonuniform signaling and investigated the effect of constellation shaping on nonlinearities in WDM
systems [57]. The block diagram of the corresponding nonuniform scheme employing arithmetic coding is shown in Fig. 5.40.
As an illustration, the receiver constellation diagrams of the nonuniform signaling 16-QAM and regular 16-QAM are also
shown in the same figure. The key part of the system is the distribution matcher, which is used for the nonuniform signaling
based on the arithmetic code. Note that the distribution matcher has been added outside the FEC encoder and decoder, which
is the reason why this scheme requires minor modifications regarding the current optical communication system. Because of
the flexibility of arithmetic code, the distribution matcher can be designed based on different criteria, such as Maxwell-
Boltzmann distribution and maximizing mutual information rate. According to [57], the proposed nonuniform signaling
scheme increases the transmission distance by 8% with the help of binary LDPC code and bit-interleaved coded modulation.
However, the use of arithmetic code increases the overall overhead of this nonuniform signaling scheme.
In the rest of this section, we investigate the method of achieving nonuniform signaling schemes for the transmission of
binary data by using the Huffman code. In this section, we mainly focus on the 2-D 9-point constellation sets as an illustrative
example. The conceptual diagram of the nonuniform signaling process is illustrated in Fig. 5.40. In this scheme, each symbol
can carry two, three, or four bits per symbol, and symbols are selected for transmission with different probabilities. It is
obvious that the center point carrying 00 bits has the largest probability of 0.25 and the symbols carrying 3 or 4 bits are
transmitted with probabilities 0.125 and 0.0625, respectively. The mapping rule is determined by employing the Huffman
procedure described as a tree diagram provided in Fig. 5.41. The receiving constellation diagram is shown in Fig. 5.42.
5.8 Nonuniform Signaling 293

Fig. 5.40 The nonuniform

signaling scheme based on an Laser
arithmetic code diode

Distribution FEC QAM I/Q Optical

matcher encoding mapping modulator channel

Inverse
FEC Coherent
distribution Demapper
decoder detection
matcher

Uniform Nonuniform

Fig. 5.41 The tree diagram of the Huffman code

It is obvious that the constellation structure of the nonuniform constellation sets is dictated by the structure of the binary
tree of the Huffman code. In order to determine the signal constellation that is suitable for nonuniform signaling scheme, an
overall search algorithm maximizing the constellation figure of merit (CFM) is used in signal constellation design. For
additional details, an interested reader is referred to [58, 59]. The key advantage of this scheme, compared to the
corresponding scheme with arithmetic coding, is the fact that overhead coming from the Huffman code is placed into larger
constellation so that overall code rate is identical to the traditional uniform signaling scheme.
The nonbinary LDPC-coded modulation scheme for nonuniform signaling based on the Huffman code is shown in
Fig. 5.43, based on ref. [59]. With the help of interleaver and 9-QAM [58], binary bits are mapped to the nonuniform
symbols, which are then passed to the nonbinary LDPC encoder. The power distributer block will avoid the presentence of
long zero energy sequences. After that, the I and Q RF signals will be used as inputs to the I/Q modulator to perform the
electrical-to-optical conversion, and such obtained nonuniform signal will be transmitted over fiber-optic communication
system of interests. On the receiver side, the homodyne coherent optical detection is used and followed with the compensation
of linear and nonlinear impairments and carrier phase estimator (CPE). The sliding-MAP equalizer provides soft symbol
LLRs, which are sent to the nonbinary LDPC decoder. Once decoding is completed, we can get information bits after
294 5 Advanced Modulation and Multiplexing Techniques

Fig. 5.42 The receiving constellation diagram of nonuniform 9-QAM

DFB Laser

SMF

Uniform distributed 9-QAM GF(9) Nonbinary-LDPC Power MZM

Interleaver ….
binary bits (Source Input) Mapper encoder Distributer Modulator
EDFA

Compensation Local
of linear Laser
GF(9)
Received Non binary 9-QAM Power and nonlinear Balanced
DeInterleaver Redistributer impairments Detector
bits LDPC demapper
decoder & From
CPE SMF

Fig. 5.43 The nonbinary LDPC-coded modulation scheme based on Huffman code-based nonuniform signaling

deinterleaving. The results of Monte Carlo simulation can be found in [58] for binary LDPC coding and [59] for nonbinary
LDPC coding. For the experimental validation results, an interested reader is referred to [60].
A reader interested to study how to combine nonuniform signaling and LDPC-coded modulation, nonuniform signaling-
based Shannon limits, hybrid probabilistic-geometric shaping, and combining probabilistic shaping and pulse shaping
schemes is referred to refs. [61–70].

5.9 Concluding Remarks

This chapter has been devoted to advanced modulation and multiplexing techniques suitable for both wireless and optical
communication systems. In Sect. 5.1, the signal space theory concepts have been applied to wireless communication systems.
After the geometric representations of signals (Sect. 5.1.1), various multidimensional modulators and demodulators suitable
for wireless communication applications have been described, in particular the Euclidean distance, correlation, and matched
filter-based detectors, followed by the description of FSK scheme in Sect. 5.1.3. Both continuous-time (CT) and discrete-time
(DT) implementations for pulse amplitude modulation schemes, suitable for wireless communications, have been described
as well.
After that, the focus has been moved in Sect. 5.2 to multilevel schemes suitable for both wireless and optical communica-
tion applications, including M-ary PSK, star-QAM, square-QAM, and cross-QAM, all described in Sect. 5.2.1. Regarding the
5.10 Problems 295

optical communications, the transmitters for M-ary PSK (Sect. 5.2.2), star-QAM (Sect. 5.2.3), and square
/cross-QAM (Sect.
5.2.4) have been described in detail.
The next topic in the chapter, in Sect. 5.3, has been related to the multicarrier modulation, including the description of
multicarrier systems with both nonoverlapping (Sect. 5.3.1) and overlapping (Sect. 5.3.2) subcarriers as well as the introduc-
tion of various approaches to deal with fading effects at the subcarrier level (Sect. 5.3.3). The concept of OFDM has been
introduced as well; however, details have been provided in Chap. 7 instead.
The MIMO fundamentals have been then provided in Sect. 5.4, including the description of key differences with respect to
diversity scheme, as well as the introduction of array, diversity, and multiplexing gains. The parallel MIMO channel
decomposition has been briefly described as well, while the details on MIMO signal processing have been postponed for
Chap. 8.
In the section on polarization-division multiplexing (PDM) and 4-D signaling, Sect. 5.5, we have described key differences
between PDM and 4-D signaling and have described as well how both types of schemes can be implemented in both wireless
and optical communications.
The focus has been moved in Sect. 5.6 to the spatial-division multiplexing (SDM) and multidimensional signaling. We
have described in Sect. 5.6.1 how SDM can be applied in wireless communications. Then we have described in Sect. 5.6.2
how various degrees of freedom including amplitude, phase, frequency, polarization states, and spatial modes can be used to
convey the information in the optical domain. In the same section, the SDM concepts for fiber-optic communications have
been described as well. The section concludes with SDM and multidimensional signaling concepts applied to free-space
optical (FSO) communications.
The next topic in Sect. 5.7 has been devoted to the signal constellation design, including iterative polar modulation (IPM)
in Sect. 5.7.1, signal constellation design for circular symmetric optical channels in Sect. 5.7.2, energy-efficient signal
constellation design (Sect. 5.7.3), and optimum signal constellation design (OSCD) in Sect. 5.7.4.
Section 5.8 has been devoted to the nonuniform signaling, in which different signal constellation points have been
transmitted with different probabilities.
The set of problems is provided in the incoming section.

5.10 Problems

5.1. Determine the Euclidean distance between the pair of signals provided below:
(a) Antipodal (bipolar) signals.
(b) Unipolar (on-off) signals.
(c) Orthogonal signals.

(a)
p1(t)
p2(t)
1T

0 T
0
T
1T

(b)
p1(t)
p2(t)
1T

0 T
0
T

(c)
p1(t) p2(t)
1T
1T
T
0 0
T
1T
296 5 Advanced Modulation and Multiplexing Techniques

5.2. Let us observe the 4-ary digital communication system, with symbols represented by the following waveforms:
p1(t) ¼ 0, p2(t) ¼ rect((t
T/2)/T ), p3(t) ¼ 2rect((t
T/2)/T ), p4(t) ¼ 3rect((t
T/2)/T). The sequence of pair of bits is
used to select the waveform to be sent, according to the following mapping rule:

00 ! p1 ðt Þ 01 ! p2 ðt Þ 10 ! p3 ðt Þ 11 ! p4 ðt Þ:

Determine the average energy per bit if the a priori probabilities are given by P(00) ¼ 1/16, P(01) ¼ P(10) ¼ 3/16,
P(11) ¼ 9/16.
P
5.3. One possible implementation of QPSK modulators is given in Fig. P3, where bi ðt Þ ¼ bik rectððt
kT Þ=t Þ, bik 2
k
f
1, 1g, i 2 fI, Qg. Show that the output of this modulator can be represented as
rffiffiffiffiffiffiffi h
i
2E s π m2π
sm ðt Þ ¼ cos ωc t
þ :
T0 4 4

bI(t)

cos( ct) bI(t) cos( ct)+bQ(t)sin( ct)

- /2

bQ(t)

Fig. P3 The QPSK modulator

5.4. The signal constellation of a ternary signal set is obtained by placing the signal constellation points in vertices of an
equilateral triangle with side length a. Sketch the signal constellation and determine the average signal energy as the
function of a by assuming that all symbols are equally likely. Determine the optimum decision regions. Select the
arbitrary basis functions ϕ1(t) and ϕ2(t) and express the signals from signal constellation diagram in terms of basis
functions. Determine the corresponding minimum energy signal constellation.
5.5. A ternary one-dimensional signal constellation is given by {
1,0,1}. Sketch the signal constellation and decision regions
and determine the average symbol energy assuming equal probable transmission. Further, determine the optimum
receiver. Finally, derive the exact symbol error probability.
5.6. A ternary set of signals is given by

d, 0  t  T
s0 ðt Þ ¼ 0; 0  t  T; s1 ðt Þ ¼
s2 ðt Þ ¼ :
0, otherwise

5.7. If P(s0) ¼ 0.4 and P(s1) ¼ P(s2) ¼ 0.3, determine the average symbol energy. Determine the basis functions and sketch
corresponding constellation diagram. Determine the optimum decision regions. Derive the expression for average
symbol error probability as the function of signal-to-noise ratio, defined as the average signal energy-to-power spectral
density of AWGN.
5.8. Let us consider a ternary transmission by using 9-ary two-dimensional (2-D) constellation over FSO channel containing
both multiplicative component represented by gamma-gamma distribution of irradiance and additive zero-mean Gauss-
ian noise of power spectral density N0/2, assuming that coherent optical detection is used. The 8 constellation points in
this 2-D constellation are placed on a circle of radius 1. The last (ninth) point is placed in the origin. Assume that the point
in the origin appears with probability 0.5, all others with probability 0.5/8.
(a) Determine the average symbol energy of this constellation.
(b) How many ternary digits can be transmitted per symbol?
(c) Determine the average energy per ternary digit.
(d) If the symbol rate is 25 giga symbols/s, determine the aggregate ternary data rate.
5.10 Problems 297

(e) Determine the Gray mapping rule to map ternary digits to different constellation points in 9-ary 2-D constellation.
(f) Determine the average symbol error probability for this modulation format in the presence of scintillation
and AWGN.
(g) Determine the optimum receiver for this 2-D constellation in the presence of scintillation and AWGN.
(h) In a fading wireless communication environment, the received signal power varies randomly over distance or time
due to multipath fading. Let us assume that the same signal constellation is used to transmit messages over wireless
multipath channel, when there is no line of sight. In this case, the distribution of envelope of received signal is
Rayleigh. Determine the average symbol error probability for this modulation format in the presence of Rayleigh
fading and AWGN. Compare the severity of fading and severity of turbulence in (g).
Note: Do not solve for integrals; just express them in terms of either Q-function or erfc-function.
5.9. Suppose we want to transmit the following data sequence: 1101000110 by binary DPSK. If s(t) ¼ Acos(ωct + θ)
represents the transmitted signal in any signaling interval of duration T, determine the phases of transmitted signal for
this data sequence. Begin with the phase θ ¼ 0 rad for the phase of the first bit to be transmitted.
5.10. Suppose we want to transmit the following data sequence: 0101110011010110 by differential 16-PSK. Determine the
phases of transmitted signal for this data sequence. Assume that the referent phase is 0 rad. Describe how the transmitted
sequence can be recovered on the receiver side. Provide the block diagrams of transmitter and receiver.
5.11. Let us consider the binary transmission with a bit rate of 10 Gb/s. The bit 0 is represented by the absence of signal, while
bit 1 is represented by a rectangular pulse of amplitude 1 V. The binary sequence, in which symbols 0 and 1 are equally
likely, is transmitted over a zero-mean additive white Gaussian noise (AWGN) channel of variance 0.2 W. We assume
that signal is attenuated during transmission in addition to being affected by AWGN. Let the received amplitude when
symbol 1 is transmitted be A ¼ 0.8 V. Determine the average time that elapses between two bits being in error. If the
low-noise amplifier (LNA) is used, the signal amplitude get increased to A ¼ 1.6 V. Determine the average time that
elapses between two bits being in error, when LNA is used (ignore the noise introduced by LNA).
5.12. Suppose we want to transmit a multilevel digital signal at rate 12 giga symbols/s through a channel described by transfer
function

H 0 , jf j  8 GHz
H c ðf Þ ¼ f ,
0, otherwise

where H0 2 (0,1] is a constant. Design the near optimum transmit and receive filters so that transmission is ISI free and
probability of error is minimized.
5.13. In this problem, we are concerned with noncoherent detection of binary signals in AWGN, when the phase is unknown.
In this communication system, the zero symbol is represented by the absence of signal, while the symbol one as a cosine
signal of amplitude A and frequency ω0. The transmission of both symbols is affected by the AWGN n(t) of PSD N0/2.
The received signal r(t) is represented by

nðt Þ, when 0 istransmitted
r ðt Þ ¼
A cos ðω0 t þ θÞ þ nðt Þ, when 1 istransmitted

where the distribution of the phase θ is uniform.

(a) Determine the optimum receiver minimizing bit error probability.
(b) Derive the expression for bit error probability in this case.
5.14. The cube signal constellation is obtained by placing the signal constellation points in vertices of the cube of side length
a. Sketch the signal constellation and determine the average signal energy as the function of a, assuming that all
symbols are equally likely. Select the arbitrary basis functions ϕi(t) (i ¼ 1,2,3) and express the signals from signal
constellation diagram in terms of basis functions. Determine the corresponding minimum energy signal constellation.
How many bits per symbol can be transmitted? Apply the Gray mapping and show how the bits should be mapped to
symbols.
5.15. In an M-ary pulse amplitude modulation (MPAM) system, the transmitted signal x takes values from the discrete set
X ¼ {0,d,. . .,(M-1)d} (M  2), where d is the Euclidean distance between two neighboring points in corresponding
signal constellation diagram. If all signal constellation points are equally probable, determine the average signal energy.
Derive the probability of symbol error as the function of signal-to-noise ratio per bit and signal constellation size. Sketch
298 5 Advanced Modulation and Multiplexing Techniques

the signal constellation diagram, and denote the decision regions and decision boundaries. Determine an equivalent
minimum energy MPAM system, and determine its average energy.
5.16. The 5-QAM signal constellation is shown in Fig. P16. We are interested in evaluating its performance against that of
5-PSK with constellation points located on a circle of radius a. Assuming that all constellation points are equally likely:
(a) Compare the energy efficiencies of these two signal constellations.
(b) Which signal constellation is more robust to phase errors? Justify your answer.
(c) Determine the decision boundaries and decision regions.

a/2

a/2

a/2

Fig. P16 The 5-QAM signal constellation diagram

5.17. Prove that the rectangular pulse shape p(t) ¼ rect(t/Ts) satisfies the Nyquist criterion for zero ISI at the output of the

1, l ¼ 0
matched filter: Rp ðlT s Þ ¼ , where Rp(τ) is the autocorrelation function of p(t). By using the result above and
0, l 6¼ 0
the fact that the output of I/Q modulator can be represented as
rffiffiffiffiffi rffiffiffiffiffi
X 2 2
sðt Þ ¼ I pðt
kT s Þ cos ðωc t Þ
Q pðt
kT s Þ sin ðωc t Þ,
k Ts k Ts k

prove that at the receiver side the outputs of matched filters after sampling can be written as
X X
I 0 ðkT s Þ ¼ I R ðlT s
kT s Þ þ vI ðkT s Þ, Q0 ðkT s Þ ¼
l l p l
Ql Rp ðlT s
kT s Þ þ vQ ðkT s Þ:

Describe the operation principle of the DT QAM modulator and demodulator and relate the output of the DT
demodulator with sampled version of CT demodulator output.
5.18. Based on concatenated transmitter configuration of 16-star-QAM, provide the corresponding transmitter configuration
for star-8-QAM signal constellation composed of two circles of radius 1 and r each carrying 4 points. Fully describe the
operation of this transmitter. Determine r so that the average signal energy is lowest possible. Provide the Gray
mapping rule.
5.19. The peak-to-average ratio (PAR) for a given waveform s(t), to be used in digital communications, is defined as
PAR ¼ max|s(t)|2/E{|s(t)|2}, where both max() and expectation E{} operators are defined with respect to time t.
Determine the PAR of a raised-cosine pulse when the roll-off parameter is either 0 or 1. Which pulse shape has the
lowest PAR? Is this pulse shape more or less sensitive to timing errors compared to others?
5.20. Explain how the multidimensional hybrid modulation scheme, described in Sect. 5.6.2, can be used as both 4 TbE and
10 TbE enabling technology by using the devices operating at a symbol rate of 40 GS/s. Discuss different alternatives
and their corresponding complexities.
5.21. Provide details of 128-CIPQ constellation in a fashion similar to Table 5.2 in Sect. 5.7.1. Use the algorithm described in
the same section.
5.22. Explain how the pi update rule given by Eq. (5.110) in an energy-efficient signal constellation design algorithm is
derived. Provide fully the steps in this derivation.
5.23. Provide the system parameters of the CO-OFDM system that is capable of compensating the accumulated chromatic
dispersion over 250 km of SMF and corresponding PMD. The aggregate data rate should be at least 800 Gb/s, and the
OFDM signal bandwidth should be either 25 GHz or 40 GHz. Use polarization-division multiplexing and square-QAM
constellations. Assume that the chromatic dispersion parameter is 16 ps/(nmkm) and that the PMD parameter is
References 299

Dp ¼ 0.05 pskm
1/2. Ignore the second-order GVD and fiber nonlinearities. Estimate the OSNR penalty due to guard
interval compared to back-to-back configuration. Determine the spectral efficiency of your design. Is it possible to find
the optimum number of subcarriers to maximize the spectral efficiency?
5.24. OFDM samples will follow Gaussian distribution for a sufficient number of subcarriers. The envelope will clearly
follow a Rayleigh distribution, while the power will have a chi-square distribution. Determine the cumulative
distribution function of envelope. Determine the probability that PAPR is below a certain threshold ztsh.
5.25. In order to study the efficiency of CO-OFDM in PMD compensation, the 2  2 Jones matrix for the first-order PMD is
typically used. Generalize this model for higher-order PMD studies.
5.26. Based on parallel decomposition of MIMO optical channel, similar to Fig. 5.27, explain how MIMO-OFDM can be
used in multi-Tb/s optical transport.

References
1. J.G. Proakis, Digital Communications, 5th edn. (McGraw Hill, New York, 2007)
2. S. Haykin, Communication Systems, 4th edn. (Wiley, New York, 2001)
3. S. Hayking, Digital Communications Systems (Wiley, Hoboken, 2014)
4. B.P. Lathi, Z. Ding, Modern Digital and Analog Communication Systems, 4th edn. (Oxford University Press, Oxford, 2009)
5. A. Goldsmith, Wireless Communications (Cambridge University Press, Cambridge, 2005)
6. I.B. Djordjevic, W. Ryan, B. Vasic, Coding for Optical Channels (Springer, New York/Dordrecht/Heidelberg/London, 2010)
7. M. Cvijetic, I.B. Djordjevic, Advanced Optical Communication Systems and Networks (Artech House, Norwood, 2013)
8. I.B. Djordjevic, L. Xu, T. Wang, Statistical physics inspired energy-efficient coded-modulation for optical communications. Opt. Lett. 37(8),
1340–1342 (2012)
9. M. Seimetz, High-Order Modulation for Optical Fiber Transmission (Springer, Berlin, 2009)
10. H.G. Batshon, I.B. Djordjevic, T. Schmidt, Ultra high speed optical transmission using subcarrier-multiplexed four-dimensional LDPC-coded
modulation. Opt. Express 18, 20546–20551 (2010)
11. I. Djordjevic, H.G. Batshon, L. Xu, T. Wang, Four-dimensional optical multiband-OFDM for beyond 1.4 Tb/s serial optical transmission. Opt.
Express 19(2), 876–882 (2011)
12. Y. Zhang, M. Arabaci, I.B. Djordjevic, Evaluation of four-dimensional nonbinary LDPC-coded modulation for next-generation long-haul
optical transport networks. Opt. Express 20(8), 9296–9301 (2012)
13. I.B. Djordjevic, L. Xu, T. Wang, PMD compensation in coded-modulation schemes with coherent detection using Alamouti-type polarization-
time coding. Opt. Express 16(18), 14163–14172 (2008)
14. I.B. Djordjevic, M. Arabaci, L. Xu, T. Wang, Generalized OFDM (GOFDM) for ultra-high-speed optical transmission. Opt. Express 19(7),
6969–6979 (2011)
15. W. Shieh, I. Djordjevic, OFDM for Optical Communications (Elsevier/Academic Press, Burlington, 2009)
16. I.B. Djordjevic, M. Arabaci, L. Xu, T. Wang, Spatial-domain-based multidimensional modulation for multi-Tb/s serial optical transmission. Opt.
Express 19(7), 6845–6857 (2011)
17. I.B. Djordjevic, Energy-efficient spatial-domain-based hybrid multidimensional coded-modulations enabling multi-Tb/s optical transport. Opt.
Express 19(17), 16708–16714 (2011)
18. I.B. Djordjevic, A. Jovanovic, Z.H. Peric, T. Wang, Multidimensional optical transport based on optimized vector-quantization-inspired signal
constellation design. IEEE Trans. Commun. 62(9), 3262–3273 (Sept 2014)
19. I.B. Djordjevic, On the irregular nonbinary QC-LDPC-Coded hybrid multidimensional OSCD-modulation enabling beyond 100 Tb/s optical
transport. IEEE/OSA J. Lightw. Technol. 31(16), 2969–2975 (15 Aug 2013)
20. I.B. Djordjevic, M. Cvijetic, C. Lin, Multidimensional signaling and coding enabling multi-Tb/s optical transport and networking. IEEE Sig.
Proc. Mag. 31(2), 104–117 (Mar 2014)
21. I.B. Djordjevic, On advanced FEC and coded modulation for ultra-high-speed optical transmission. IEEE Commun. Surv. Tutor. 18(3),
1920–1951, third quarter 2016 (19 Aug 2016). https://doi.org/10.1109/COMST.2016.2536726
22. I.B. Djordjevic, M. Arabaci, LDPC-coded orbital angular momentum (OAM) modulation for free-space optical communication. Opt. Express
18(24), 24722–24728 (22 Nov 2010)
23. I.B. Djordjevic, J. Anguita, B. Vasic, Error-correction coded orbital-angular-momentum modulation for FSO channels affected by turbulence.
IEEE/OSA J. Lightw. Technol. 30(17), 2846–2852 (1 Sept 2012)
24. Z. Qu, I.B. Djordjevic, 500 Gb/s free-space optical transmission over strong atmospheric turbulence channels. Opt. Lett. 41(14), 3285–3288
(15 July 2016)
25. I.B. Djordjevic, Z. Qu, Coded orbital angular momentum modulation and multiplexing enabling ultra-high-speed free-space optical transmis-
sion, in Optical Wireless Communications – An Emerging Technology, ed. by M. Uysal, C. Capsoni, Z. Ghassemlooy, A. Boucouvalas, E. G.
Udvary, (Springer, 2016), pp. 363–385
26. I.B. Djordjevic, Z. Qu, Coded orbital-angular-momentum-based free-space optical transmission, in Wiley Encyclopedia of Electrical and
Electronics Engineering, (Feb 2016). http://onlinelibrary.wiley.com/doi/10.1002/047134608X.W8291/abstract
27. J.D. Jackson, Classical Electrodynamics (Wiley, New York, 1975)
28. Z. Bouchal et al., Selective excitation of vortex fiber modes using a spatial light modulator. New J. Phys. 7, 125 (2005)
300 5 Advanced Modulation and Multiplexing Techniques

29. J.A. Anguita, M.A. Neifeld, B.V. Vasic, Turbulence-induced channel crosstalk in an orbital angular momentum-multiplexed free-space optical
link. Appl. Opt. 47(13), 2414–2429 (2008)
30. A. Li, A. Al Amin, X. Chen, W. Shieh, Reception of mode and polarization multiplexed 107-Gb/s CO-OFDM signal over a two-mode fiber, in
Proceedings of the OFC/NFOEC, Postdeadline Papers (OSA, 2011), Paper PDPB8
31. L. Gruner-Nielsen, Y. Sun, J.W. Nicholson, D. Jakobsen, R. Lingle, B. Palsdottir, Few mode transmission fiber with low DGD, low mode
coupling and low loss, in Proceedings of the OFC/NFOEC, Postdeadline Papers (OSA, 2012), Paper PDP5A.1
32. R. Ryf, et al., Low-loss mode coupler for mode-multiplexed transmission in few-mode fiber, in Proceedings of the OFC/NFOEC, Postdeadline
Papers (OSA, 2012), Paper PDP5B.5
33. N.K. Fontaine, et al., Space-division multiplexing and all-optical MIMO demultiplexing using a photonic integrated circuit, in Proceedings of
the OFC/NFOEC, Postdeadline Papers (OSA, 2012), Paper PDP5B.1
34. X. Chen, A. Li, J. Ye, A. Al Amin, and W. Shieh, Reception of dual-LP11-mode CO-OFDM signals through few-mode compatible optical
add/drop multiplexer, in Proceedings of the OFC/NFOEC, Postdeadline Papers (OSA, 2012), Paper PDP5B.4
35. S. Randel, et al., Mode-multiplexed 620-GBd QPSK transmission over 1200-km DGD-compensated few-mode fiber, in Proceedings of the
OFC/NFOEC, Postdeadline Papers (OSA, 2012), Paper PDP5C.5
36. R. Ryf, et al., Space-division multiplexed transmission over 4200 km 3-core microstructured fiber, in Proceedings of the OFC/NFOEC,
Postdeadline Papers (Optical Society of America, 2012), Paper PDP5C.5
37. B.J. Puttnam, W. Klaus, Y. Awaji, N. Wada, A. Kanno, T. Kawanishi, K. Imamura, H. Inaba, K. Mukasa, R. Sugizaki, T. Kobayashi, , and
M. Watanabe, 19-core fiber transmission of 19x100x172-Gb/s SDM-WDM-PDM-QPSK signals at 305Tb/s, in Proceedings of the
OFC/NFOEC, Postdeadline Papers (OSA, 2012), Paper PDP5C.1
38. P.M. Krummrich, Optical amplifiers for multi mode/multi core transmission, in Proceedings of the OFC/NFOEC (OSA, 2012), Paper OW1D.1
39. R. Essiambre, A. Mecozzi, Capacity limits in single mode fiber and scaling for spatial multiplexing, in Proceedings of the OFC/NFOEC (OSA,
2012), Paper OW3D
40. J.A. Anguita, J. Herreros, I.B. Djordjevic, Coherent multi-mode OAM superpositions for multi-dimensional modulation. IEEE Photonics
Journal 6(2), 900811 (Apr 2014)
41. Z.H. Peric, I.B. Djordjevic, S.M. Bogosavljevic, M.C. Stefanovic, Design of signal constellations for Gaussian channel by iterative polar
quantization, in Proceedings of the 9th Mediterranean electrotechnical conference, vol. 2, pp. 866–869, Tel-Aviv, Israel, 18–20 May 1998
42. I.B. Djordjevic, H.G. Batshon, L. Xu, T. Wang, Coded polarization-multiplexed iterative polar modulation (PM-IPM) for beyond 400 Gb/s
serial optical transmission, in Proceedings of the OFC/NFOEC 2010, Paper No. OMK2, San Diego, CA, 21–25 Mar 2010
43. X. Liu, S. Chandrasekhar, T. Lotz, P. Winzer, H. Haunstein, S. Randel, S. Corteselli, B. Zhu, Generation and FEC-decoding of a 231.5-Gb/
s PDM-OFDM signal with 256-Iterative-Polar-Modulation achieving 11.15-b/s/Hz intrachannel spectral efficiency and 800-km reach, in
Proceedings of the OFC/NFOEC, Postdeadline Papers (OSA, 2012), Paper PDP5B.3
44. Zhang, J., I.B. Djordjevic, Optimum signal constellation design for rotationally symmetric optical channel with coherent detection, in
Proceedings of the OFC/NFOEC 2011, Paper No. OThO3, LA, USA, 6–10 Mar 2011
45. R.-J. Essiambre, G. Kramer, P.J. Winzer, G.J. Foschini, B. Goebel, Capacity limits of optical fiber networks. J. Lightwave Technol. 28(4),
662–701 (2010)
46. N.J.A. Sloane, R.H. Hardin, T.S. Duff, J.H. Conway, Minimal-energy clusters of hard spheres. Discrete Comput. Geom. 14, 237–259 (1995)
47. G. Wannier, Statistical Physics (Dover Publications, New York, 1987)
48. T. Cover, J. Thomas, Elements of Information Theory (Wiley, 1991)
49. R.E. Blahut, Computation of channel capacity and rate distortion functions. IEEE Trans. Inform. Theory IT-18, 460–473 (1972)
50. I.B. Djordjevic, T. Liu, L. Xu, T. Wang, Optimum signal constellation design for high-speed optical transmission, in Proceedings of the
OFC/NFOEC 2012, Paper no. OW3H.2, Los Angeles, CA, USA, 6–8 Mar 2012
51. T. Liu, I.B. Djordjevic, L. Xu, T. Wang, Feedback channel capacity inspired optimum signal constellation design for high-speed optical
transmission, in Proceedings of the CLEO 2012, Paper no. CTh3C.2, San Jose, CA, 6–11 May 2012
52. T. Liu, Z. Qu, C. Lin, I.B. Djordjevic, Nonuniform Signaling based LDPC-coded modulation for high-speed optical transport networks, in OSA
Asia Communications and Photonics Conference (ACP) 2016, 2–5 Nov 2016, Wuhan, China. (Invited Paper)
53. F. Kschischang, S. Pasupathy, Optimal Nonuniform Signaling for Gaussian Channels. IEEE Trans. Inform. Theory 39, 913–929 (1993)
54. T. Fehenberger, D. Lavery, R. Maher, A. Alvarado, P. Bayvel, N. Hanik, Sensitivity gains by mismatched probabilistic shaping for optical
communication systems. IEEE Photo. Technol. Lett. 28(7), 786–789 (2016)
55. T. Liu, I.B. Djordjevic, LDPC-coded BICM-ID based nonuniform signaling for ultra-high-speed optical transport, in Optical Fiber Communi-
cation Conference, OSA Technical Digest (online) (Optical Society of America, 2016), paper M3A.3
56. G. Böcherer, Probabilistic signal shaping for bit-metric decoding, in Proceedings of the 2014 IEEE International Symposium on Information
Theory (ISIT 2014), (Institute of Electrical and Electronics Engineers, New York, 2014), pp. 431–435
57. T. Fehenberger, G. Böcherer, A. Alvarado, and N. Hanik, LDPC coded modulation with probabilistic shaping for optical fiber systems, in
Optical Fiber Communication Conference, OSA Technical Digest (online) (Optical Society of America, 2015), paper Th2A.23
58. T. Liu, C. Lin, I.B. Djordjevic, Advanced GF(32) nonbinary LDPC coded modulation with non-uniform 9-QAM outperforming star 8-QAM.
Opt. Express 24(23), 13866–13874 (2016)
59. C. Lin, D. Zou, T. Liu, I.B. Djordjevic, Capacity achieving nonbinary LDPC coded non-uniform shaping modulation for adaptive optical
communications. Opt. Express 24(16), 18095–18104 (2016)
60. Z. Qu, C. Lin, T. Liu, I. Djordjevic, Experimental study of nonlinearity tolerant modulation formats based on LDPC coded non-uniform
signaling, in OFC 2017, Paper W1G.7, 19–23 Mar 2017, Los Angeles, CA, USA
61. Z. Qu, I.B. Djordjevic, J. Anderson, Two-dimensional constellation shaping in fiber-optics communications. Appl. Sci. 9(9), 1889 (8 May 2019)
62. Z. Qu, I.B. Djordjevic, On the probabilistic shaping and geometric shaping in optical communication systems. IEEE Access 7, 21454–21464
(4 Feb 2019)
63. X. Han, M. Yang, I.B. Djordjevic, Y. Yue, Q. Wang, Z. Qu, J. Anderson, Joint probabilistic-nyquist pulse shaping for LDPC-coded 8-PAM
signal in DWDM data center communications. Appl. Sci. 9(23), 4996 (20 Nov 2019)
References 301

64. X. Han, Y. Yue, Z. Qu, R. Holmes, I. Djordjevic, Probabilistic shaped LDPC-coded 400G PM-64QAM DWDM transmission in 50-GHz Grid,
in 26th Optoelectronic and Communications Conference (OECC), 3–7 July 2021
65. X. Han, I.B. Djordjevic, A. Jovanovic, A hybrid PGS LDPC-coded scheme for PM-16QAM modulation in 140 km DWDM metro network
communication, in Proceedings of the CLEO 2021, paper SF2D.5, 9–14 May 2021
66. X. Han, Y. Yue, Z. Qu, I.B. Djordjevic, Probabilistic-shaping-enabled ADC/DAC resolution tolerance improvement in optical PAM/QAM
communication systems, in SPIE OPTO, Photonics West 2021, San Francisco, CA, 6–11 Mar 2021
67. X. Han, I.B. Djordjevic, Probabilistic-shaped, Nyquist-pulse-shaped four-dimensional LDPC-coded modulation scheme for 100 km DWDM
metro network transmission, in Proceedings of the 22nd international conference on transparent optical networks ICTON 2020, paper Mo.E1.7,
Bari, Italy, 19–23 July 2020
68. X. Han, M. Yang, I. Djordjevic, A. Li, Probabilistically-shaped four-dimensional LDPC-coded modulation in 100 km DWDM optical
transmission for metro network applications, in Asia Communications and Photonics Conference (ACP) 2019, 2–5 Nov 2019, Chengdu,
Sichuan, China
69. X. Han, M. Yang, I.B. Djordjevic, Y. Yue, Q. Wang, Z. Qu, J. Anderson, Combined probabilistic shaping and nyquist pulse shaping for PAM8
signal transmission in WDM systems, in Proceedings of the CLEO 2019, 5–10 May 2019, San Jose, California, USA
70. X. Han, I.B. Djordjevic, Probabilistically shaped 8-PAM suitable for data centers communication, in Proceedings of the ICTON 2018, Paper Th.
B3.4, Bucharest, Romania, 1–5 July 2018
 Advanced Detection Techniques
and Compensation of Channel Impairments 6

Abstract
This chapter is devoted to advanced detection and channel impairments compensation techniques applicable to both
wireless and optical communication systems. To better understand the principles behind advanced detection and channel
impairments compensation, we start the chapter with the fundamentals of detection and estimation theory including
optimum receiver design in symbol error probability sense and symbol error probability derivations. In the section on
wireless communication systems performance, we describe different scenarios including outage probability, average error
probability, and combined outage-average error probability scenarios. We also describe the moment-generating function-
based approach to average error probability calculation. We also describe how to evaluate performance in the presence of
Doppler spread and fading effects. In the section on channel equalization techniques, after short introduction, we first
describe how zero-forcing equalizers can be used to compensate for ISI, chromatic dispersion, polarization mode
dispersion (PMD) in fiber-optics communication; ISI and multipath fading effects in wireless communications; and ISI
and multipath effects in indoor optical wireless communications. After that, we describe the optimum linear equalizer
design in the minimum mean-square error sense as well as Wiener filtering concepts. The most relevant post-compensation
techniques, applicable to both wireless and optical communications, are described next including feedforward equalizer,
decision-feedback equalizer, adaptive equalizer, blind equalizers, and maximum likelihood sequence detector (MLSD),
also known as the Viterbi equalizer. The turbo equalization technique is postponed for Chap. 9. The next section is devoted
to the relevant synchronization techniques. In the section on adaptive modulation techniques, we describe how to adapt the
transmitter to time-varying wireless/optical channel conditions to enable reliable and high spectral-efficient transmission.
Various scenarios are described including data rate, power, code rate, error probability adaptation scenarios, as well as a
combination of various adaptation strategies. In particular, variable-power variable-rate modulation techniques and
adaptive coded modulation techniques are described in detail. Next, the Volterra series-based equalization to deal with
fiber nonlinearities and nonlinear effects in wireless communications is described. In the section on digital
backpropagation, we describe how this method can be applied to deal with fiber nonlinearities, chromatic dispersion,
and PMD in a simultaneous manner. In the section on coherent optical detection, we describe various balanced coherent
optical detection schemes for two-dimensional modulation schemes; polarization diversity and polarization demultiplexing
schemes; homodyne coherent optical detection based on phase-locked loops; phase diversity receivers; and the dominant
coherent optical detection sources including laser phase noise, polarization noise, transimpedance amplifier noise, and
amplifier spontaneous emission (ASE) noise. In the section on compensation of atmospheric turbulence effects, we
describe various techniques to deal with atmospheric turbulence including adaptive coding, adaptive coded modulation,
diversity approaches, MIMO signal processing, hybrid free-space optical (FSO)-RF communication approach, adaptive
optics, and spatial light modulator (SLM)-based backpropagation method. In particular, linear adaptive optics techniques
are described in detail. The set of problems is provided after the concluding remarks.

# The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 303
I. B. Djordjevic, Advanced Optical and Wireless Communications Systems, https://doi.org/10.1007/978-3-030-98491-5_6
304 6 Advanced Detection Techniques and Compensation of Channel Impairments

6.1 Detection and Estimation Theory Fundamentals

To better understand the principles of advanced detection and channel impairments compensation techniques, in this section,
we provide the fundamentals of detection and estimation theory.

6.1.1 Geometric Representation of Received Signals

Earlier in Sect. 5.1, we mentioned that in a digital wireless/optical communication system, there is a source of the message that
at any time instance generates a symbol mi (i ¼ 1, 2, . . ., M ) from the set of symbols {m1, m2, . . ., mM}. These symbols are
generated with a priori probabilities: p1, p2, . . ., pM, where pi ¼ P(mi). A transmitter then converts the output mi from the
message source into a distinct real signal si(t) of duration T, which is suitable for transmission over the wireless/optical
channel. We also outlined that this set of signals {si(t)} can be represented as properly weighted orthonormal basis functions
{Φn} (n ¼ 1, . . ., N ) (wherein N  M), so that [1–5]

X
N
si ðt Þ ¼ sin Φn ðt Þ, 0  t < T; i ¼ 1, 2, . . . , M, ð6:1Þ
n¼1

where sin is the projection of the ith signal along the nth basis function. The received signal r(t) can also be represented in
terms of projections along the same basis functions as follows:

X
N ðT
rðtÞ ¼ r n Φn ðtÞ, r n ¼ rðtÞΦn ðtÞdt ðn ¼ 1, 2, . . . , NÞ: ð6:2Þ
j¼1
0

Therefore, both received and transmitted signal can be represented as vectors r and si in the signal space, respectively, as

r ¼ ½r 1 r 2 ⋯ r N T , si ¼ jsi i ¼ ½ si1 si2 ⋯ siN T : ð6:3Þ

In Chap. 5, we used the Euclidean distance receiver to decide in favor of a signal constellation point closest to the received
signal vector. In the linear regime for both wireless and optical communications, we can use the additive (linear) channel
model and represent the received signal as addition of transmitted signal si(t) and accumulated additive noise z(t):

rðtÞ ¼ si ðtÞ þ zðtÞ, 0  t < T; i ¼ 1, 2, . . . , M: ð6:4Þ

We can represent the projection of received signal r(t) along the nth basis function as follows:

ðT ðT
r n ¼ rðtÞΦn ðtÞdt ¼ ½si ðtÞ þ zðtÞΦn ðtÞdt
0 0
ðT Tðs
ð6:5Þ
¼ si ðtÞΦn ðtÞdt þ zðtÞΦn ðtÞdt ¼
0 0
|fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl}
sin zn
¼ sin þ zn ; n ¼ 1, ⋯, N:

We can also rewrite Eq. (6.5) into a compact vector form as

6.1 Detection and Estimation Theory Fundamentals 305

Fig. 6.1 Two equivalent Observation: r1 r2 r3 … rN

implementations of detector
(or demodulator) for wireless
communications: (a) correlation Samplers tk=kT
detector and (b) matched filter-
T T T T
based detector
dt dt dt dt
0 0 0 0

Receive 1 2 3 N

antenna

N-dim. demodulator:
Power splitter correlation detector

LNA
(a)
Observation: r1 r2 r3 … rN

Samplers tk=kT

h1 t h2 t h3 t hN t
1 T t 2 T t 3 T t N T t

Receive
antenna
N-dim. demodulator:
matched filter-based
Power splitter detector

(b)

2 3 2 3 2 3
r1 si1 z1
6r 7 6 7 6 7
6 2 7 6 si2 7 6 z2 7
r ¼ 6 7 ¼ 4 5 þ 4 5 ¼ si þ z: ð6:6Þ
4⋮5 ⋮ ⋮
siN zN
rN |fflffl{zfflffl} |fflffl{zfflffl}
si z

Therefore, the received vector, also known as observation vector, can be represented as a sum of transmitted vector and
noise vector. The observation vector can be obtained by using one of two possible alternative detection methods: matched
filter-based detector and correlation detector, as summarized in Fig. 6.1.
The output of the nth matched filter having impulse response hn(t) ¼ Φn(T  t) is given as the convolution of the matched
filter input and impulse response:

ð
1 ð
1

r n ðt Þ ¼ r ðτÞhn ðt  τÞdτ ¼ r ðτÞΦn ðT  t þ τÞdτ: ð6:7Þ

1 1

By performing a sampling at the time instance t ¼ T, we obtain

306 6 Advanced Detection Techniques and Compensation of Channel Impairments

ð
1 ðT
r n ðT Þ ¼ r ðτÞhn ðT  τÞdτ ¼ r ðτÞΦn ðτÞdτ, ð6:8Þ
1 0

which has the same form as the nth output of the correlation receiver. Therefore, the correlation and matched filter-based
receivers are equivalent to each other.
Let us observe the random process obtained as a difference of received signal and signal obtained by projecting the
received signal along the basis functions. This difference is in fact the “remainder” of the received signal:

X
N X
N
r 0 ðt Þ ¼ r ðt Þ  r n Φ n ðt Þ ¼ si ðt Þ þ zðt Þ  ðsin þ zn ÞΦn ðt Þ
n¼1 n¼1
ð6:9Þ
X
N
0
¼ zðt Þ  zn Φn ðt Þ ¼ z ðt Þ:
n¼1

We can notice that the remainder of the received signal is only noise dependent. In all equations above, we have
represented the noise in terms of the same basis functions used to represent the transmitted signal. However, it is more
appropriate to represent the noise by using the Karhunen-Loève expansion [2], by taking the correlation properties of the noise
process into account. However, for additive white Gaussian noise (AWGN), the noise projections along the basis functions of
transmitted signals are uncorrelated and independent, as we will show shortly. Consequently, the basis used to represent noise
is irrelevant in the selection of the decision strategy in the detection process.
Let R(t) denote the random process established by sampling of the received signal r(t). The projection Rn of R(t) along the
nth basis function will be a random variable represented by the correlator output rn (n ¼ 1, 2, . . ., N ). The corresponding mean
value of Rn is given by

mRn ¼ hRn i ¼ hsin þ Z n i ¼ sin þ hZ n i ¼ sin , ð6:10Þ

and it is dependent only on the nth coordinate of transmitted signal sin. With Zn we denoted the random variable at the output
of the nth correlator (matched filter) originating from the noise component z(t), and with hi being the mathematical
expectation operator. In a similar fashion, the variance of Rn is also only noise dependent since it is
D E  
σ 2Rn ¼ Var½Rn  ¼ ðRn  sin Þ2 ¼ Z 2n : ð6:11Þ

The variance of Rn can be determined as follows:

*ðT ðT +
σ 2Rn ¼ hZ 2n i ¼ ZðtÞΦn ðtÞdt ZðuÞΦn ðuÞdu
0 0
*ðT ðT +
¼ Φn ðtÞΦn ðuÞZðtÞZðuÞdtdu
0 0
ð6:12Þ
ðT ðT
¼ Φn ðtÞΦn ðuÞhZðtÞZðuÞidtdu
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}
0 0 RZ ðt, uÞ

ðT ðT
¼ Φn ðtÞΦn ðuÞRZ ðt, uÞdtdu:
0 0

In Eq. (6.12), RZ() presents the autocorrelation function of the noise process Z(t), which is otherwise given as [1, 2]
6.1 Detection and Estimation Theory Fundamentals 307

N0
RZ ðt, uÞ ¼ δðt  uÞ: ð6:13Þ
2

Now, after substitution of autocorrelation function (6.13) into Eq. (6.12), we obtain that

ðT ðT ðT
N N0 N
σ 2Rn ¼ 0 Φn ðt ÞΦn ðuÞδðt  uÞdtdu ¼ Φ2n ðt Þdt ¼ 0 : ð6:14Þ
2 2 2
0 0 0
|fflfflfflfflffl{zfflfflfflfflffl}
1

Therefore, all matched filter (correlator) outputs have the variance equal to the power spectral density of the noise process
Z(t), which is equal to N0/2. The covariance between the nth and kth (n6¼k) outputs of the matched filters is given by

Cov½Rn Rk  ¼ hðRn  sin ÞðRk  sik Þi ¼ hRn Rk i

*ð0 ð0 +
¼ ZðtÞΦ j ðtÞdt ZðuÞΦk ðuÞdt
T T
*ð0 ð0 +
¼ Φn ðtÞΦk ðuÞZðtÞZðuÞdtdu
T T
ð0 ð0
¼ Φn ðtÞΦk ðuÞhZðtÞZðuÞidtdu
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} ð6:15Þ
T T RZ ðtuÞ

ð0 ð0 ð0 ð0
N
¼ Φn ðtÞΦk ðuÞRZ ðt  uÞdtdu ¼ 0 Φn ðtÞΦk ðuÞδðt  uÞdtdu
|fflfflfflfflffl{zfflfflfflfflffl} 2
T T N0 T T
2 δðtuÞ

ð0
N0
¼ Φn ðtÞΦk ðtÞdt ¼ 0:
2
T
|fflfflfflfflfflfflfflfflfflffl
ffl{zfflfflfflfflfflfflfflfflfflfflffl}
0, n6¼k

As a consequence, the outputs Rn of correlators (matched filters) are mutually uncorrelated. In conclusion, the outputs of
correlators, given by Eq. (6.6), are Gaussian random variables. These outputs are statistically independent so that the joint
conditional PDF of the observation vector R can be written as a product of conditional PDFs of its individual outputs from
correlators as follows:

N
f R ðrjmi Þ ¼ ∏ f Rn ðr n jmi Þ; i ¼ 1, ⋯, M: ð6:16Þ
n¼1

The components Rn’s are independent Gaussian random variables with mean values sin and variance equal to N0/2, so that
the corresponding conditional PDF of Rn is given as
 
1 1 n ¼ 1, 2, . . . , N
f Rn ðr n jmi Þ ¼ pffiffiffiffiffiffiffiffiffi exp  ðr n  sin Þ ;
2
: ð6:17Þ
πN 0 N 0 i ¼ 1, 2, . . . , M

By substituting the conditional PDFs into joint PDF, and after simple arrangements, we obtain
" #
1 X
N
N=2 2
f R ðrjmi Þ ¼ ðπN 0 Þ exp  ðr  sin Þ ; i ¼ 1, ⋯, M: ð6:18Þ
N 0 n¼1 n
308 6 Advanced Detection Techniques and Compensation of Channel Impairments
P
We can conclude that the elements of random vector given by Eq. (6.6) completely characterize the term Nn¼1 r n Φn ðt Þ:
What left to be characterized is the remainder z’(t) of the received signal given by Eq. (6.9), which is only noise dependent.
It can be easily shown that the sample of the noise process Z’(tm) is statistically independent on outputs Rj of the correlators; in
other words,

n ¼ 1, 2, . . . , N
hRn Z 0 ðt m Þi ¼ 0, : ð6:19Þ
0  tm  T s

Therefore, the outputs of correlators (matched filters) are only relevant in statistics of the decision-making process. In other
words, as long as the signal detection in AWGN is concerned, only the projections of the noise onto basis functions used to
represent the signal set {si(t)} (i ¼ 1, . . ., M ) affect the statistical process in the detection circuit; the remainder of the noise is
irrelevant. This claim is sometimes called the theorem of irrelevance [1–4]. According to this theorem, the channel model
given by Eq. (6.4) can be represented by an equivalent N-dimensional vector given by Eq. (6.6).

6.1.2 Correlation and Matched Filter-Based Receivers Maximizing the Signal-to-Noise Ratio

The schemes from Fig. 6.1 can serve as a base of the receiver decision circuit, as shown in Fig. 6.2. The observation vector
obtained by using either correlation of matched filter detector is used as the input vector. The observation vector is used as
input to M dot-product calculators. The ith product calculator evaluates the inner product between the observation vector r and
the ith signal vector si (corresponding to transmitted message mi) as follows:

ðT X
N
hrjsi i ¼ rðtÞsi ðtÞdt ¼ rsTi ¼ r n sin : ð6:20Þ
n¼1
0

If it happens that different symbols have different energies after dot-product calculations, we need to remove the half of the
energy of the corresponding symbol in each of the branches. Decision circuit decides in favor of the symbol with the largest
input and provides the estimate to the N-dimensional demapper. For nonbinary transmission, the demapper block can be
omitted. The configurations of both correlation detector and matched filter-based detector, both shown in Fig. 6.1, and the
configuration of signal transmission decoder/detector, shown in Fig. 6.2, have been derived based on the Euclidean distance
receiver in Chap. 5.

Estimated sequence of bits

The N-dimensional demapper

Estimate m̂

Decision circuit:
Select the largest input

-0.5E1 -0.5E2 -0.5EM

r Ts
1 r Ts2 r TsM
s1 s2 … sM Dot-product
calculator

Observation
r vector

Fig. 6.2 The configuration of the signal transmission detector/decoder for optimum receiver
6.1 Detection and Estimation Theory Fundamentals 309

In the rest of this section, we will show that the matched filter is optimum from the signal-to-noise ratio (SNR) point of
view. For this proof, we will use the Schwartz inequality given by Eq. (5.11). Let us observe the nth output from the matched
filter shown in Fig. 6.1b, which is the convolution of received signal and impulse response of the matched filter evaluated at
t ¼ T:

r n ðTÞ ¼ rðtÞ  hn ðtÞjt¼T ¼ ½sn ðtÞ  hn ðtÞjt¼T þ ½zðtÞ  hn ðtÞjt¼T

|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
an ðTÞ zn ðTÞ ð6:21Þ
¼ an ðTÞ þ zn ðTÞ,

where an(T ) represents the signal portion of the output from matched filter, while zn(T) denotes the noise portion of the
matched filter output. The SNR can be defined as
 2 
a ðT Þ
SNR ¼  2n  : ð6:22Þ
zn ðT Þ

The variance of the noise process can be expressed as

*1
ð ðT +
 
z2n ðT Þ ¼ hn ðT  τÞzðτÞdτ  hn ðT  uÞzðuÞdu
1 0
ð 1
1 ð ð6:23Þ
¼ hzðτÞzðuÞihn ðT  τÞhn ðT  uÞdτdu,
|fflfflfflfflfflffl{zfflfflfflfflfflffl}
1 1 Rz ðτuÞ

where Rz(τ-u) is the autocorrelation function of AWGN introduced by Eq. (6.13). Now we have that Rz(τ-u) ¼ (N0/2) δ (τ-u),
which upon substitution in Eq. (6.23) gives the following:

ð
1 ð
1
  N
z2n ðT Þ ¼ 0 δðτ  uÞhn ðT  τÞhn ðT  uÞdτdu
2
1 1
ð6:24Þ
ðT
N0 2
¼ hn ðT  uÞdu:
2
0

On the other hand, the variance of the signal at the output of matched filter is given as

 2  D
2 E
2
an ðT Þ ¼ sn ðt Þ  hn ðt Þjt¼T ¼ sn ðt Þ  hn ðt Þjt¼T ¼
21 32
ð ð6:25Þ
¼ 4 hn ðT  τÞsn ðτÞdτ5 :
1

After substitution of corresponding variances into Eq. (6.22), we obtain the following expression for SNR:
2 32
ð
1
4 hn ðT  τÞsn ðτÞdτ5
1
SNR ¼ : ð6:26Þ
ðT
N0
2 h2n ðT  uÞdu
0

By applying the Schwartz inequality to numerator in Eq. (6.26), becomes upper bounded by
310 6 Advanced Detection Techniques and Compensation of Channel Impairments

2T 32
ð ðT ðT
4 hn ðT  τÞsn ðτÞdτ5 h2n ðT  τÞdτ s2n ðτÞdτ
0 0 0
SNR ¼ 
ðT ðT
N0 2 N0 2
hn ðT  τÞdτ hn ðT  τÞdτ ð6:27Þ
2 2
0 0
Tðs
2 2E n
¼ s2n ðτÞdτ ¼ ¼ SNRn, max :
N0 N0
0

The equality sign in Eq. (6.27) is satisfied when hn(T  τ) ¼ Csn(τ), where C is an arbitrary constant (for convenience, we
can assume that C ¼ 1). Accordingly, the impulse response of the matched filter can be expressed as

hn ðτÞ ¼ sn ðT  τÞ: ð6:28Þ

In conclusion, we can say that the matched filter maximizes the SNR. It can be straightforwardly shown that the output
signal of a matched filter is proportional to a shifted version of the autocorrelation function of the input signal, namely,

r n ðt Þ ¼ Rsn ðt  T Þ: ð6:29Þ

It is also important to mention that the spectrum of the output signal of a matched filter is proportional to the spectral
density of the input signal:

FT fsn ðtÞg¼Sn ð f Þ
Hð f Þ ¼ FT fsn ðtÞ  sn ðT  τÞg ¼ Sn ð f ÞSn ð f Þexp ½j2πf T
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}
jSn ð f Þj2 ð6:30Þ
¼ jSn ð f Þj2 exp ½j2πf T:

Therefore, we have just shown that the Euclidean distance receiver, the correlation receiver, and matched filter-based
receivers are equivalent to each other for Gaussian-like channels.

6.1.3 Optimum and LLR Receivers

In this section, we are concerned with the receivers minimizing symbol error probability, commonly referred to as the
optimum receivers. The joint conditional PDF given by fR(r| mi) can be used in a decision-making process. It is often called
likelihood function and denoted as L(mi). For Gaussian channels, the likelihood function contains exponential terms that can
lead to numerical instabilities. It is more convenient to work with its logarithmic version, which is known as log-likelihood
function:

lðmi Þ ¼ log Lðmi Þ; i ¼ 1, 2, . . . , M: ð6:31Þ

For an AWGN channel, the log-likelihood function can be obtained by substituting Eq. (6.18) into Eq. (6.31) (by ignoring
independent terms in the channel statistics) as follows:

1 X
N
lðmi Þ ¼  ðr  sij Þ2 ; i ¼ 1, 2, . . . , M: ð6:32Þ
N 0 j¼1 j

One can notice that the log-likelihood function is similar to the Euclidean distance squared introduced earlier.
The log-likelihood function is quite useful in the decision-making procedure. Since we are particularly concerned with
optimum receiver design, we can use the log-likelihood function as a design tool. Namely, given the observation vector r, the
6.1 Detection and Estimation Theory Fundamentals 311

optimum receiver performs the mapping from the observation vector r to the estimate value m b of the transmitted symbol (say,
mi) so that the symbol error probability is minimized. The probability Pe of errorness decision, in other words the probability
of detecting mb ¼ mi when it was not transmitted, can be calculated by the following relation:

Pe ðmi jrÞ ¼ Pðmi isnot sentjrÞ ¼ 1  Pðmi issentjrÞ, ð6:33Þ

where P(.|.) denotes the conditional probability.

The optimum decision strategy, also known as the maximum a posteriori probability (MAP) rule, can be formulated as the
following decision rule:

b ¼ mi if
Set m
ð6:34Þ
Pðmi issentjrÞ  Pðmk issentjrÞ, 8k 6¼ i:

At this point, it is useful to provide the geometric interpretation of optimum detection problem. Let D denote the N-
dimensional signal space of all possible observation vectors (D is also known as the observation space). The total observation
space is partitioned into M nonoverlapping N-dimensional decision regions D1, D2, . . ., DM, which are defined as

Di ¼ fr : Pðmi sentjrÞ > Pðmk sentjrÞ8k 6¼ ig: ð6:35Þ

In other words, the ith decision region Di is defined as the set of all observation vectors for which the probability that mi is
sent is larger than any other probability. As an illustration, in Fig. 6.3, we show decision regions for QPSK (also known as
4-QAM), assuming an equal probability transmission. Whenever the observation vector falls within the decision region Di, we
can decide in favor of symbol mi.
The optimum decision strategy given by Eq. (6.35) is not quite suitable from a practical point of view. Let us consider an
alternative practical representation for M ¼ 2. The average error probability for M ¼ 2 is given by
ð ð
P e ¼ p1 f R ðrjm1 Þdr þ p2 f R ðrjm2 Þdr, ð6:36Þ
D1 D2

where pi ¼ P(mi) represents the a priori probability that symbol mi is transmitted, while Di ¼ D  Di denotes the complement
of Di (“not Di decision region”).

s2 d
s1

D2 D1

d
D3 D4

s3 s4

Fig. 6.3 Decision regions for QPSK

312 6 Advanced Detection Techniques and Compensation of Channel Impairments

Since D ¼ D1+D2, we can rewrite Eq. (6.36) as

ð ð
Pe ¼ p1 f R ðrjm1 Þdr þ p2 f R ðrjm2 Þdr ¼
DD1 D1
ð ð6:37Þ
¼ p1 þ ½p2 f R ðrjm2 Þ  p1 f R ðrjm1 Þdr:
D1

In order to minimize the average error probability given by Eq. (6.37), we can apply the following decision rule: if p1fR(r|
m1) is greater than p2fR(r|m2), then assign r to D1 and accordingly decide in favor of m1. Otherwise, we assign r to D2 and
decide in favor of m2. This decision strategy can be generalized for M > 2 as follows:

b ¼ mi if
Set m
ð6:38Þ
pk f R ðrjmk Þ is maximum for k ¼ i:

When all symbols occur with the same probability pi ¼ 1/M, then the corresponding decision rule is known as the
maximum likelihood (ML) rule and can be formulated as

b ¼ mi if
Set m
ð6:39Þ
lðmk Þ is maximum for k ¼ i:

In terms of the observation space, the ML rule can also be formulated as

Observationvector r liesindecisionregion Di when

ð6:40Þ
lðmk Þ is maximum for k ¼ i:

For an AWGN channel, by using the likelihood function given by Eq. (6.32) (which is related to the Euclidean distance
squared), we can formulate the ML rule as

Observation vector r lies in region Di if

ð6:41Þ
the Euclidean distance krsk k is minimum for k ¼ i:

From Euclidean distance in signal space (see Chap. 5), we know that

X
N X
N X
N X
N
kr  s k k2 ¼ jr n  skn j2 ¼ r 2n  2 r n skn þ s2kn , ð6:42Þ
n¼1 n¼1 n¼1 n¼1
|fflfflffl{zfflfflffl}
Ek

and by substituting this equation into Eq. (6.41), we obtain the final version of the ML rule as follows:

Observation vector r lies in decisionregion Di if

XN
1 ð6:43Þ
r n skn  E k is maximum for k ¼ i:
n¼1
2

The receiver configuration shown in Fig. 6.2 is an optimum one only for an AWGN channel (such as wireless channel and
optical channel dominated by ASE noise), while it is suboptimum for Gaussian-like channels. As for non-Gaussian channels,
we need to use the ML rule given by Eq. (6.39).
As an example, for modulation schemes with M ¼ 2, we can use the decision strategy based on the following rule:

  m1  
f R ðrjm1 Þ > p
LLRðrÞ ¼ log log 2 , ð6:44Þ
f R ðrjm2 Þ < p1
m2
6.1 Detection and Estimation Theory Fundamentals 313

Fig. 6.4 The log-likelihood ratio If the threshold is exceeded,

Observation Log-likelihood logL(r) Decision decide in favor of H1 (choose m1)
receiver circuit
vector r ratio computer (comparator) Otherwise, decide in favor of H2 (choose m2)

Threshold log(p2/p1)

where LLR() denotes the log-likelihood ratio. Equation (6.44) is also known as Bayes’ test [2], while the factor log( p2/p1)
is known as a threshold of the test. The scheme of the corresponding log-likelihood ratio receiver is shown in Fig. 6.4. When
p1 ¼ p2 ¼ 1/2, the threshold of the test is 0, while the corresponding receiver is the ML one.

6.1.4 Symbol Error Probability Calculation

In this section, we describe how to calculate symbol error probability and provide two illustrative examples: (i) calculation of
bit error probabilities for binary signal constellations and (ii) calculation of symbol error probability for M-ary QAM. For
large signal constellations, when it becomes quite difficult to calculate symbol error probabilities, we describe a union-bound
approximation.

6.1.4.1 Calculation of Symbol Error Probability

The average symbol error probability can be calculated as

X
M X
M ð
Pe ¼ pi Pðr does not lie in Di jmi sentÞ ¼ pi f R ðrjmi Þdr, ð6:45Þ
i¼1 i¼1 Di

where pi¼P(mi) represents the a priori probability that symbol mi is transmitted, while Di ¼ D  Di denotes the complement
of Di (not Di decision region). By using this definition of the complement of decision region, Eq. (6.45) can be rewritten as

X
M ð X
M ð
Pe ¼ pi f R ðrjmi Þdr ¼ pi f R ðrjmi Þdr
i¼1 Di i¼1 DDi
ð ð6:46Þ
X
M
¼1 pi f R ðrjmi Þdr:
i¼1 Di

For an equal probability transmission of symbols ( pi ¼ 1/M ), the average error probability can be calculated as

M ð
1 X
Pe ¼ 1  f ðrjmi Þdr: ð6:47Þ
M i¼1 Di R

In the ML detection, the probability Pe of the symbol error depends only on the relative Euclidean distances between the
message points in the constellation. Since AWGN is spherically symmetric in all directions in the signal space, the changes in
the orientation (with respect to both coordinate axes and the origin) do not affect the probability of the symbol error.
Therefore, the rotation and translation of signal constellation produce another signal constellation that has the same symbol
error probability as the one observed against the minimum Euclidean distance.
Let us first observe the rotation operation and denote the original complex-valued signal constellation by {si}. The rotation
matrix U is a Hermitian matrix (UU{ ¼ I, where I is the identity matrix). The corresponding constellation obtained after
application of rotation matrix is given by {Usi}. The Euclidean distance between the observation vector r and a constellation
point from rotated signal constellation, say, Usi, is given as

krrotate  si,rotate k ¼ kUsi þ z  Usi k ¼ kzk ¼ kr  si k: ð6:48Þ

314 6 Advanced Detection Techniques and Compensation of Channel Impairments

Therefore, the Euclidean distances between constellation points are preserved after rotation since it is only noise
dependent. Accordingly, the symbol error probability is invariant to the rotation process. We need also to verify if the
noise vector is sensitive to rotation. The variance of rotated noise vector can be expressed as
D E D E    
zrotate z{rotate ¼ UzðUzÞ{ ¼ Uzz{ U { ¼ U zz{ U {
|ffl{zffl}
ðN 0 =2ÞI ð6:49Þ
N N  
¼ 0 UU { ¼ 0 I = zz{ ,
2 2

which means that variance does not change during rotation. The mean value of the rotated noise vector is still equal to zero,
i.e.,

hzrotate i ¼ hUzi ¼ U hzi ¼ 0: ð6:50Þ

Therefore, the AWGN exhibits spherical symmetric property.

Regarding the translation operation, let all constellation points be translated for the same translation vector {si-a}. The
Euclidean distance between the observation vector and transmitted point si is given as

krtranslate  si,translate k ¼ ksi  a þ z  ðsi  aÞk ¼ kzk ¼ kr  si k, ð6:51Þ

and does not change with translation. Accordingly, the symbol error probability, as a function of the minimum Euclidean
distance, does not change either. However, the average symbol energy changes with translation, and upon translation it
becomes

X
M
Etranslate ¼ ks i  ak2 pi : ð6:52Þ
i¼1

Since ksi  ak2 ¼ ksik2  2a{si + kak2, we can rewrite Eq. (6.52) as follows:

X
M X
M X
M
E translate ¼ ksi k2 pi  2a{ s i pi þ kak2 pi ¼ E  2a{ hsi þ kak2 : ð6:53Þ
i¼1 i¼1 i¼1
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} |fflfflffl{zfflfflffl}
E hsi

The minimum energy constellation can be obtained by differentiating Etranslate with respect to a, by setting this result to be
zero, and by solving for the optimum translate value, which leads to amin ¼ hsi. The minimum energy signal constellation is
then obtained as

si,translate ¼ si  hsi: ð6:54Þ

The average symbol energy that corresponds to optimum constellation is derived after substitution of Eq. (6.54) into
Eq. (6.52) to obtain Etranslate, min ¼ E ‐ khsik2. Therefore, we can rotate and translate the signal constellation as needed to
simplify the symbol error probability calculations while not affecting the symbol error probability expressed as the function of
minimum distance in the signal constellation.
At this point, it is convenient to introduce the Q-function, defined as a probability that zero-mean Gaussian random variable
X of unit variance is larger than u:
ð1  
1 x2 =2 1 u
QðuÞ ¼ PðX > uÞ ¼ pffiffiffiffiffi e dx ¼ erfc pffiffiffi , ð6:55Þ
u 2π 2 2

wherein the complementary error function (erfc-function) is defined by

6.1 Detection and Estimation Theory Fundamentals 315

ð1
2
ex dx:
2
erfcðuÞ ¼ pffiffiffi
π u

6.1.4.2 Calculation of Bit Error Probability of Binary Signal Constellations

Let us now consider binary signal constellations. We know that the output of matched filter (correlator) r is a Gaussian
random variable since the input is a Gaussian random process. In any linear filter, the Gaussian input produces the output that
is a Gaussian random process as well, but with different mean value and variance [1–4]. The statistics of this Gaussian random
variable, r, changes depending on whether the transmitted signal was s1 (corresponding to message m1) or s2 (corresponding to
message m2). If s1 was sent, then r is a Gaussian random variable with the mean value μ1 and variance σ2 given by

N0
μ1 ¼ hrjm1 i ¼ E1  ρ12 , σ 2 ¼ ðE1 þ E2  2ρ12 Þ, ð6:56Þ
2
ÐT
where ρ12 is the correlation coefficient ρ12 ¼ 0 s1 ðt Þs2 ðt Þdt.
If s2 was sent, then r has mean value and variance given by

N0
μ2 ¼ hrjm2 i ¼ E 2 þ ρ12 , σ 2 ¼ ðE 1 þ E 2  2ρ12 Þ: ð6:57Þ
2

For an equal probability of symbol transmission, the optimum decision thresholds can be determined in intersection of
corresponding conditional probability density functions (PDFs). We can now use Eqs. (6.45), (6.56), and (6.57) to express the
error probability as
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
E1 þ E2  2ρ12
Pe fD1 jm2 g ¼ Pe fD2 jm1 g ¼ Q : ð6:58Þ
2N 0

The average bit error probability, for an equal probability transmission, is obtained as
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 E 1 þ E 2  2ρ12
Pb ¼ ðPe fD1 jm2 g þ Pe fD2 jm1 gÞ ¼ Q : ð6:58Þ
2 2N 0

We now pay attention to three representative examples of binary communication systems known as antipodal signaling,
P
M
orthogonal signaling, and unipolar signaling. The average symbol energy is given by Es ¼ Ek pk , which for equally likely
k¼1
transmission of binary symbols becomes Es ¼ 0.5E1 + 0.5E2. For binary transmission, the bit energy Eb is the same as average
symbol energy, i.e., Eb ¼ Es/log2M ¼ Es.
In antipodal signaling, such as BPSK, the transmitted symbols are s1(t) ¼
+ s(t) and s2 (t) ¼  s(t). In this case, E1 ¼ E2 ¼
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Eb, and ρ12 ¼  E, so that the average error probability is given by Pe ¼ Q 2E b =N 0 :
In orthogonal signaling, such as FSK, we have qthat qffiffiffifficoefficient ρ12 ¼ 0 and E1 ¼ E2 ¼ E. In this case, the
ffiffiffiffi correlation
E Eb
average probability of error probability is Pe ¼ Q N0 ¼ Q N 0 , which is about 3 dB worse than in antipodal signaling.
In unipolar signaling, such as on-off keying, transmitted symbols are represented as s1(t) ¼ s(t) and
qffiffiffiffiffiffi  s2(t)
¼q0.ffiffiffiffiIn this case,
E Eb
E1 ¼ E, E2 ¼ 0, Eb ¼ E/2, and ρ12 ¼ 0, so that the average error probability is given by Pe ¼ Q 2N 0 ¼Q N0 , which is
about 3 dB worse than in antipodal signaling.
Let us now establish the relationship between the average error probability and the Euclidean distance. From the definition
of Euclidean distance, we have that

ðT ðT
2
d2E ðs1 , s2 Þ ¼ ½s1 ðt Þ  s2 ðt Þ dt ¼ s21 ðt Þ  2s1 ðt Þs2 ðt Þ þ s22 ðt Þdt
ð6:59Þ
0 0
¼ E1 þ E2  2ρ12 ,

so that the average probability of error can be written as

316 6 Advanced Detection Techniques and Compensation of Channel Impairments

 
d E ðs1 , s2 Þ
Pb ¼ Q pffiffiffiffiffiffiffiffi : ð6:60Þ
2N 0

It can be shown that when E1 ¼ E2 ¼ E, dE(s1, s2) is maximized (while Pe is minimized) when s1(t) ¼  sp2(t), ffiffiffiffi which
corresponds to antipodal signaling. For antipodal signaling,
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi ρ 12 ¼  E, the Euclidean distance is d E ðs 1 , s 2 Þ ¼ 2 E and the
corresponding error probability is Pe ¼ Q 2E=N 0 , which is consistent with the expression obtained above.

6.1.4.3 Calculation of Symbol Error Probability of M-ary QAM

ffiffiffiffiffi 5, we know that M-ary square-QAM signal constellation can be obtained as two-dimensional Cartesian product of
FrompChap.
L ¼ M -ary pulse amplitude modulation (PAM) signals:

X  X ¼ fðx1 , x2 Þjxi 2 X g, X ¼ fð2i  1  LÞd; i ¼ 1, 2, ⋯, Lg: ð6:61Þ

The error probability for M-QAM can be expressed as

Pe ¼ 1  P2e,PAM , ð6:62Þ

wherein for in-phase and quadrature channels we have two independent PAM constellations. This assumption is correct for
medium and high SNRs. The neighboring constellation points in PAM are separated by 2d. In order to derive the expression of
average PAM symbol error probability Pe, PAM, we should first determine the probability of error for the following two
symbols: (i) the edge symbol having only one neighbor and (ii) the inner symbol which has two neighbors. The inner symbol
error probability can be expressed as
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pe ðmi Þ ¼ 2Pðkzk > dÞ ¼ 2Q 2d 2 =N 0 ; i ¼ 2, 3, ⋯, L  1: ð6:63Þ

On the other hand, the outer symbol error probability is just half of the inner symbol error probability. Accordingly, the
average symbol error probability of L-ary PAM can be expressed as
0sffiffiffiffiffiffiffi1 0sffiffiffiffiffiffiffi1 0sffiffiffiffiffiffiffi1
2 @ 2d2 A L  2 @ 2d2 A 2ðL  1Þ @ 2d2 A
Pe,PAM ¼ Q þ 2Q ¼ Q : ð6:64Þ
L N0 L N0 L N0

The average symbol energy of PAM constellations is given by

1X
L
1 
E s,PAM ¼ ð2i  LÞ2 d2 ¼ L2  1 d 2 : ð6:65Þ
L i¼1 3

By expressing d2 in terms of the average energy, we can write the average symbol error probability of PAM as
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2ð L  1Þ 6 E s,PAM
Pe,PAM ¼ Q
L L2  1 N 0
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð6:66Þ
L1 3 Es,PAM
¼ erfc :
L L  1 N0
2

By assuming that all L amplitude levels of in-phase or quadrature channels are equally likely, the average symbol energy of
M-ary QAM is simply

2 2  2
Es ¼ 2E s,PAM ¼ L  1 d 2 ¼ ðM  1Þd2 : ð6:67Þ
3 3

Similarly, by expressing d2 in terms of the average energy, we can write the average symbol error probability of QAM as
6.1 Detection and Estimation Theory Fundamentals 317

( pffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi)2 ( pffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi)2

M1 3 Es M1 3 Es
Pe ¼ 1  1  2 pffiffiffiffiffi Q ¼ 1  1  pffiffiffiffiffi erfc : ð6:68Þ
M M  1 N0 M 2ðM  1Þ N 0

For sufficiently high SNRs, we can neglect the erfc2{} term in the previous equation so that we can write that
pffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
M1 3 Es
Pe ffi 2 pffiffiffiffiffi erfc : ð6:69Þ
M 2ð M  1Þ N 0

6.1.4.4 Union-Bound Approximation

Although the equations derived above allow us to accurately calculate average symbol error probabilities, it is quite
challenging to find a closed form solution for large signal constellations. In these situations, we use the union-bound
approximation [1, 4], which gives us an expression that is only function of Euclidean distances among signal constellation
points. This approximation is accurate only for sufficiently high SNRs. Let Aik denote the event that the observation vector r is
closer to the signal vector sk than to si, when the symbol mi (vector si) is sent, which means that kr  skk < kr  sik. The
constellation point si is detected correctly if kr  sik < kr  skk 8 k 6¼ i, which means that the average error probability can be
upper bounded as

  X
M
M
Pe ðmi sentÞ ¼ P [ Aik  PðAik Þ: ð6:69Þ
k¼1, k6¼i
k¼1, k6¼i

The probability of event Aik can be evaluated as

PðAik Þ ¼ Pðkr  sk k < kr  si kjmi Þ ¼ Pðksi  sk þ zk < kzkÞ: ð6:70Þ

In other words, P(Aik) corresponds to the probability that the noise vector is closer to the vector si-sk than to the origin.
Given the fact that the ith noise component is a zero-mean Gaussian random variable of variance N0/2, what matters is just a
projection n of the noise vector on line containing vector si-sk. This projection has zero mean and variance equal to N0/2, so
that P(Aik) can be expressed as
0 1
ð
1
B C 1
pffiffiffiffiffiffiffiffiffi eu =2 du
2
PðAik Þ ¼ P@z > ksi  sk k=2A ¼
|fflfflfflfflffl{zfflfflfflfflffl} πN 0
dik d ik =2 ð6:71Þ
   
d ik 1 dik
¼ Q pffiffiffiffiffiffiffiffi ¼ erfc pffiffiffiffiffiffi :
2N 0 2 2 N0

The error probability related to symbol mi can be now upper bounded as

 
1 X
M
d
Pe ðmi Þ  erfc pikffiffiffiffiffiffi ; i ¼ 1, ⋯, M: ð6:72Þ
2
k¼1, k6¼i
2 N0

Finally, the average symbol error probability is upper bounded as

X  
1X X
M M M
dik
Pe ¼ pi Pe ðmi Þ  p erfc pffiffiffiffiffiffi , ð6:73Þ
i¼1
2 i¼1 i 2 N0
k¼1, k6¼i

and this inequality is known as union bound. For an equal probability transmission of symbols, pi ¼ 1/M, the union bound
simplifies to
318 6 Advanced Detection Techniques and Compensation of Channel Impairments

 
1 X X
M M
dik
Pe  erfc pffiffiffiffiffiffi : ð6:74Þ
2M i¼1
k¼1, k6¼i
2 N0

By defining the minimum distance of constellation as dmin ¼ mini, kdik, we can obtain a looser bound as
 
M1 dmin
Pe  erfc p ffiffiffiffiffi
ffi : ð6:75Þ
2 2 N0

For circularly symmetric signal constellation, we can use the following bound:

 
1 X
M
d
Pe  erfc pikffiffiffiffiffiffi : ð6:76Þ
2
k¼1, k6¼i
2 N0

Finally, the nearest neighbor approximation is given as

 
M dmin d
Pe
erfc pminffiffiffiffiffiffi , ð6:77Þ
2 2 N0

where M dmin is the number of neighbors at distance dmin from observed constellation point. Regarding the bit error probability
Pb, when Gray-like mapping rule [1] is used for sufficiently high SNRs, we can use the following approximation:

Pb
Pe = log 2 M: ð6:78Þ

If we would like to compare various signal constellation sizes, the bit energy-to-power spectral density ratio is typically
used, Eb/N0. In optical communications, very often we use the optical SNR (OSNR), introduced by

Ps Ps
OSNR ¼ ¼ , ð6:79Þ
PASE 2nsp hνðG  1ÞBop

instead of electrical SNR. In Eq. (6.79), Ps denotes the signal power, while PASE is the amplified spontaneous emission (ASE)
noise power determined by PASE ¼ 2nsphν(G  1)Bop, where nsp is the spontaneous emission factor of an erbium-doped fiber
amplifier (EDFA), G is the EDFA gain, Bop is the bandwidth of optical passband filter that follows EDFA, and hν is a single
photon energy. To compare different modulation schemes with different signal constellation sizes, it is convenient to
introduce OSNR per bit and per a single-polarization state and define it as [11]

OSNR R
OSNRb ¼ ¼ s SNRb , ð6:80Þ
log 2 M 2Bop

where M is the signal constellation size, Rs is the symbol rate, and Bop is the referent optical bandwidth (it is commonly
assumed that Bop ¼ 12.5 GHz, which corresponds to a 0.1 nm spectral width around a carrier wavelength of 1550 nm). The
parameter SNRb ¼ Eb/N0 denotes the signal-to-noise ratio per bit of information, where Eb is the bit energy and N0 is a power
spectral density (PSD) originating from ASE noise. For convenience, it is assumed that photodiode responsivity is 1 A/W so
that N0 ¼ NASE, where NASE is the PSD of ASE noise in a single-polarization state. The OSNR per single polarization is
commonly defined as OSNR ¼ E/(2NASEBop), where E is the symbol energy. Additionally, in optical communications, it is
very common to use the Q-factor as the figure of merit instead of SNR (that has nothing to do with resonators’/oscillators’
Q-factor), which is related to the bit-error rate (BER) on an AWGN as follows [11]:
 
1 Q
BER ¼ erfc pffiffiffi : ð6:81Þ
2 2

Finally, for a binary modulation scheme, the Q-factor in optical communications is defined as
6.1 Detection and Estimation Theory Fundamentals 319

jμ2  μ1 j
Q¼ , ð6:82Þ
σ2 þ σ1

where μi and σ i denote the mean value and standard deviation of symbol si (message mi), wherein i ¼ 1, 2.

6.1.5 Estimation Theory Fundamentals

So far, our focus was on detection of signal in the presence of additive noise, such as ASE or Gaussian noise. In this section,
we will pay attention to the estimation of a certain signal parameter, such as phase, frequency offset, and so on, important from
the detection point of view. Let the received signal be denoted by r(t) while the transmitted signal is s(t,p), where p is the
parameter to be estimated. The received signal in the presence of additive noise z(t) can be written as

r ðt Þ ¼ sðt, pÞ þ zðt Þ, 0  t  T, ð6:83Þ

where T is the observation interval (symbol duration for modulation schemes). The operation of assigning a value bp to an
unknown parameter p is known as parameter estimation; the value assigned is called an estimate, and the algorithm used to
perform this operation is called an estimator.
There are different criteria used to evaluate the estimator, including minimum mean-square (MMS) estimate, maximum a
posteriori probability (MAP) estimate, and maximum likelihood (ML) estimate. We can define the cost function in MMS
estimate (MMSE) as an integral of quadratic form [2]
ð
C MMSE ðεÞ ¼ ðp  b pÞ2 f ðpjrÞdp, ð6:84Þ
|fflfflfflffl{zfflfflfflffl} P
ε2

where ε ¼ b p  p, while fP( p|r) is the a posteriori probability density function (PDF) of random variable P. The observation
vector r in Eq. (6.84) is defined by its ri components, where the ith component is determined by

ðT
r i ¼ r ðt ÞΦi ðt Þdt; i ¼ 1, 2, ⋯, N ð6:85Þ
0

{Φi(t)} denotes the set of orthonormal basis functions. The estimate that minimizes the average cost function given by
Eq. (6.84) is known as Bayes’ MMS estimate [2, 3]. The estimate that maximizes the a posteriori probability density function
fP( p|r) is known as the MAP estimate. Finally, the estimate which maximizes the conditional PDF fr(r|p), which is also known
as the likelihood function, is called the ML estimate. The MAP estimate can be obtained as a special case of Bayes’ estimation
by using the uniform cost function C(ε) ¼ (1/δ)[1-rect(ε/δ)] as follows:
2 3
ð ð
^pδ=2 ð
1
16 7
hC ðεÞi ¼ C ðεÞ f P ðpjrÞdp ¼ 4 þ 5 f P ðpjrÞdp
δ
1 ^pþδ=2
2 3
ð
^pþδ=2
ð6:86Þ
16 7
¼ 41  f P ðpjrÞdp5 ffi ð1=δÞ½1  δ f P ðpjrÞ
δ
^pδ=2
1
¼  f P ðpjrÞ
δ

where the approximation above is valid for sufficiently small δ. The quality of an estimate is typically evaluated in terms of
expected mean value hb pi and variance of the estimation error. We can recognize several cases with respect to the expected
value. If the expected value of the estimate equals the true value of the parameter, hb pi ¼ p , we say that the estimate is
unbiased. If the expected value of estimate differs from true parameter by fixed value b, i.e., if hb pi ¼ p þ b, we say that the
estimate has a known bias. Finally, if the expected value of estimate is different from true parameter by variable amount b( p),
320 6 Advanced Detection Techniques and Compensation of Channel Impairments

i.e., if hb
pi ¼ p þ bðpÞ, we say that the estimate has a variable bias. The estimation error ε is defined as the difference between
true value p and estimate bp, namely, ε ¼ p  b p: A good estimate is one that simultaneously provides a small bias and a small
variance of estimation error. The lower bound on the variance of estimation error can be obtained from the Cramér-Rao
inequality [1, 2] which is
D E
1
pÞ 2  D
ðp  b 2
E, ð6:87Þ
∂ log f r ðrjpÞ
∂p2

where we assumed that the second partial derivative of conditional PDF exists and that it is absolutely integrable. Any
estimate that satisfies the Cramér-Rao bound defined by the right side of Eq. (6.87) is known as an efficient estimate.
In order to explain the ML estimation, we can assume that the transmitted signal s(t,p) can also be presented as a vector s in
signal space, with the nth component given as

ðT
sn ðpÞ ¼ sðt, pÞΦn ðt Þdt; n ¼ 1, 2, ⋯, N: ð6:88Þ
0

The noise vector z can be presented in a similar fashion

2 3
z1
6z 7 ðT
6 27
z ¼ 6 7, zn ¼ zðt ÞΦn ðt Þdt; n ¼ 1, 2, ⋯, N: ð6:89Þ
4⋮5
0
zN

Since the components zn of the noise are zero-mean Gaussian with PSD equal to N0/2, the components of observation
vector will also be Gaussian with mean values sn( p). The joint conditional PDF for a given parameter p will be

1 N 2
f R ðrjpÞ ¼ N=2
∏ e½rn sn ðpÞ =N 0 : ð6:90Þ
ðπN 0 Þ n¼1

The likelihood function can be defined as

ðT ðT
f r ðrjpÞ 2 1
L½r ðt Þ, p ¼ lim ¼ r ðt Þsðt, pÞdt  s2 ðt, pÞdt: ð6:91Þ
N!1 f r ðrÞ N0 N0
0 0

The ML estimate is derived by maximizing the likelihood function, which is done by differentiating the likelihood function
with respect to the estimate b
p and setting the result to zero, so we obtain [3]

ð0
∂sðt, ^
pÞ
½rðtÞ  sðt, ^
pÞ dt ¼ 0, ð6:92Þ
∂^p
T

and this condition is known as the likelihood equation.

Example. Let us consider the following sinusoidal signal whose amplitude a and frequency ω are known, while the phase is
unknown: a cos(ωt+θ). Based on the likelihood equation (6.92), we obtain the following ML estimate of the phase in the
presence of AWGN:
2ð T 3
6 rðtÞcos ðωtÞdt 7
^θ ¼ tan1 6
6 ð0T
7
7:
4 5
rðtÞsin ðωtÞdt
0
6.2 Wireless Communication Systems Performance 321

6.2 Wireless Communication Systems Performance

In a fading environment, as shown in Chap. 2, the envelope of the signal varies with the propagation distance as the
environment and consequently multipath and shadowing effects vary. Accordingly, the signal-to-noise ratio ρ also varies with
propagation distance, and these variations can be described with the probability density function f(ρ). Given that SNR
becomes random variable, the symbol error probability Ps(ρ) becomes random [5, 8, 14, 15] as well, and we can define
different scenarios to characterize these variations:

(i) The outage probability, denoted as Poutage, represents the probability that SNR falls below a given threshold value ρtsh,
corresponding to the maximum tolerable symbol error probability.
(ii) The average error probability corresponds to the symbol error probability averaged over SNR distribution.
(iii) Combined outage and average error probability scenario, when shadowing and multipath fading effects are simulta-
neously present.

6.2.1 Outage Probability Scenario

The outage probability scenario is applicable when the symbol duration Ts is comparable to the coherence time Tc so that the
signal fade level is roughly constant for the duration of the symbol. As indicated above, the outage probability Poutage
represents the probability that SNR falls below a given threshold value ρtsh:

ρðtsh

Poutage ¼ Pðρ  ρtsh Þ ¼ f ðρÞdρ, ð6:93Þ

0

where f(ρ) is the PDF of SNR ρ. For Rayleigh fading, the corresponding outage probability becomes

ρðtsh
1 ρ=ρ
Poutage ¼ Pðρ  ρtsh Þ ¼ e dρ ¼ 1  eρtsh =ρ : ð6:94Þ
ρ
0

The required SNR to achieve a target outage probability Poutage is given by

ρ
ρ¼  tsh : ð6:95Þ
 log 10 1  Poutage

From (6.95), we can determine the fade margin (FM) defined as SNR increase required to maintain acceptable performance
more than (1-Poutage) 100% of the time as follows:
 

FM ¼ 10 log 10  log 10 1  Poutage : ð6:96Þ

6.2.2 Average Error Probability Scenario

The average error probability scenario is applicable when the symbol duration T is comparable to wireless channel coherence
time Tc, and it can be computed by averaging the AWGN error probability for different fading realizations; in other words,

ð
1

Ps ¼ hPs ðρÞi ¼ Ps ðρÞf ðρÞdρ, ð6:97Þ

0
322 6 Advanced Detection Techniques and Compensation of Channel Impairments

where Ps(ρ) is the AWGN symbol error probability, with ρ being the symbol SNR. The generic expression applicable to many
2-D and 1-D signal constellations is given by
pffiffiffiffiffiffiffiffiffiffi
Ps ðρÞ ’ C M Q DM ρ , ð6:98Þ

where CM and DM are the modulation format-dependent parameters, with M being the signal constellation size. For instance,
for QPSK, C4 ¼ 2 and D4 ¼ 1. For M-PSK, the corresponding parameters are CM ¼ 2 and DM ¼ 2sin2(π/M ). On the other
hand, the corresponding parameters for M-QAM are CM ¼ 4(1-1/√M) and DM ¼ 3/(M-1). Finally, for M-PAM, the
corresponding parameters are CM ¼ 2(M-1)/M and DM ¼ 6/(M 2-1). For Gray mapping rule, to estimate the bit error
probability, we need to divide symbol error probability by log2M. Since symbol SNR ρ and bit SNR ρb are related by
ρ ¼ ρblog2M, the corresponding expression for bit error probability will be
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
CM
P b ð ρb Þ
Q DM ρb log 2 M : ð6:99Þ
log 2 M

For Rayleigh fading, the expression for average symbol error probability can be written as

ð
1 " sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi#
pffiffiffiffiffiffiffiffiffiffi DM ρ=2
1 ρ=ρ CM
Ps ¼ hPs ðρÞi ’ CM Q DM ρ e dρ ’ 1 : ð6:100Þ
ρ 2 1 þ DM ρ=2
0

For very high SNR, the following asymptotic expression can be used: Ps ρ!1
! C M =ð2DM ρÞ.
As an illustration, we show in Fig. 6.5 bit error probability vs. average SNR per bit for different constellation sizes for
AWGN and Rayleigh fading channels, by employing approximate formulas (6.98) for an AWGN channel and (6.100) for a
Rayleigh fading channel. Clearly, significant degradation in Pb can be found in the presence of Rayleigh fading.

Fig. 6.5 Average bit error

probability vs. average SNR per M-QAM in AWGN:
bit for AWGN and Rayleigh
fading channels
M= 4
100 M=16
M=64
10-1
M-QAM in Rayleigh fading:
Average bit error probability, Pb

10-2 M= 4
M=16
10-3 M=64

10-4

10-5

10-6

10-7

10-8

10-9
0 5 10 15 20 25 30 35 40 45
Average signal-to-noise ratio per bit, b
6.2 Wireless Communication Systems Performance 323

6.2.3 Combined Outage and Average Error Probability Scenario

The combined outage and average error probability scenario is applicable when the fading environment can be represented as a
superposition of both fast fading (shadowing effects) and slow fading (such as Rayleigh fading). In this scenario, we can define
the average error probability with some probability 1  Poutage. Therefore, when not in outage, in other words, the symbol SNR
is averaged over fast fading for fixed path loss and random shadowing ρ  ρtsh (where ρtsh is the average threshold SNR), the
average probability of error is obtained by averaging over the conditional PDF of the fast fading as follows:

ð
1

Ps ¼ Ps ðρÞf ðρjρÞdρ: ð6:101Þ

0

The average threshold SNR ρtsh is obtained from Eq. (6.101) by setting Ps ¼ Ps,tsh , where Ps,tsh is the maximum tolerable
average symbol error probability, and solving for ρ.

6.2.4 Moment-Generating Function (MGF)-Based Approach to Average Error Probability

Calculation

The moment-generating function (MGF) for a nonnegative random variable ρ with PDF fρ(ρ) is defined as

ð
1

MGF ρ ðsÞ ¼ esρ f ρ ðρÞdρ: ð6:102Þ

0

Clearly, the MGF is related to the Laplace transform (LT) of PDF by

 
LT f ρ ðρÞ ¼ MGF ρ ðsÞ:

As the name suggests, the MGF is related to the moments hρni by


∂
n 
h ρn i ¼ MGF ðs Þ  : ð6:103Þ
∂s n ρ 
s¼0

The MGF for Nakagami-m fading is given by [15]

1
MGF ρ ðsÞ ¼ : ð6:104Þ
ð1  sρ=mÞm

Rayleigh fading is a special case of Nakagami-m fading with m ¼ 1. On the other hand, the MGF for Rician fading with
Rician factor K is given by [15]
 
1þK Ksρ
MGF ρ ðsÞ ¼ exp : ð6:105Þ
1 þ K  sρ 1 þ K  sρ

As an illustration of this method, let us consider symbol error probability given by Eq. (6.98). For convenience, let us use an
alternative definition of Q-function [14, 15]:

ð
π=2  
1 z2
Q ðzÞ ¼ exp  dϕ: ð6:106Þ
π 2 sin 2 ϕ
0
324 6 Advanced Detection Techniques and Compensation of Channel Impairments

Equation (6.98) can be now rewritten as

pffiffiffiffiffiffiffiffiffiffi ð
π=2  
CM D ρ
Ps ðρÞ ’ CM Q DM ρ ¼ exp  M 2 dϕ: ð6:107Þ
π 2 sin ϕ
0

The average error probability in a fading environment will be then

ð
1
pffiffiffiffiffiffiffiffiffiffi
s ffi
P C M Qð DM ρÞf ðρÞdρ
0

ð 1
π=2 ð   
C D ρ
¼ M exp  M2 f ðρÞdρ dϕ )
π 2sin ϕ
0 0 ð6:108Þ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
h i
DM
MGF ρ 
2sin2 ϕ

ð
π=2  
C DM
¼ M MGF ρ  dϕ:
π 2sin2 ϕ
0

Equation (6.108) still requires numerical integration, but the corresponding integral is easier to evaluate.

6.2.5 Performance Evaluation in the Presence of Doppler Spread and Fading

When differential detection is used, the Doppler spread introduces an error floor phenomenon caused by the correlation of
symbols, which is a function of Doppler frequency νD and symbol duration T. The Doppler spread is accounted by the
correlation coefficient for different Doppler power spectrum models, defined as ρC ¼ R(T )/R(0), and for the Clark-Jakes
model, described in Chap. 2, we have that

ρC ¼ RðT Þ=Rð0Þ ¼ J 0 ð2πBD T Þ, ð6:109Þ

where BD is the Doppler spread.

The average probability of error for DPSK in the presence of Doppler spread and Rician fading is given by
 
ðDPSKÞ 1 1 þ K þ ρb ð 1  ρC Þ Kρb
Pb ¼ exp  : ð6:110Þ
2 1 þ K þ ρb 1 þ K þ ρb

For Rayleigh fading, the corresponding average probability of error is obtained from Eq. (6.110), by setting K ¼ 0 to obtain
b ð1ρC Þ
Pb ¼ 12 1þρ1þρ . The error floor, in Rician fading, can be estimated as the following limit:
b


ðDPSKÞ ðDPSKÞ  ð1  ρC ÞeK
Pb,floor ¼ Pb  ¼ : ð6:111Þ
ρ!1 2

The average probability of error for DQPSK in the presence of Doppler spread and in Rician fading is upper bounded by
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1
0
u pffiffiffi 2
u ðρC  ρ= 2Þ
 @1  t
ðDQPSKÞ 1
b
P
2 pffiffiffi 2 A:
ð1 þ  ρÞ2  ðρC  ρ= 2Þ
" # ð6:112Þ
pffiffiffi
ð2  2ÞK ρ=2
exp  pffiffiffi :
ð1 þ 
ρs Þ  ðρC  ρ= 2Þ
6.3 Channel Equalization Techniques 325

Fig. 6.6 Average bit error

probability vs. average bit SNR in T=0.2:
10-1 D
DPSK for different Rician K= 0
Rayleigh fading
parameters and different Doppler 10-2 K= 9

Average bit error probability, P b

frequencies under the Clark-Jakes
(uniform scattering) model 10-3
K=18
D
T=0.02:
10-4 Rician
K= 0
fading
10-5 K= 9
K=18
10-6

10-7

10-8

10-9

10-10

10-11
0 8 16 24 32 40 48
Average bit SNR, b
[dB]

The corresponding error floor is obtained as the following limit:

0 vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1 "  #
 u  pffiffiffi pffiffiffi
ðDQPSKÞ 
u ρC = 2 2 2  2 K=2
ðDQPSKÞ
Pb,floor ¼ Pb ¼
1@
1 t A pffiffiffi :
  pffiffiffi2 exp   ð6:113Þ
ρ!1 2 1 ρ = 2 C
1  ρC = 2

The DPSK error floor for the Clark-Jakes (uniform scattering) model in the presence of Doppler spread and Rayleigh fading
can be obtained upon substitution of Eq. (6.109) into Eq. (6.113) as follows:

ðDQPSKÞ 1  J 0 ð2πνD T Þ
Pb,floor ¼ , ð6:114Þ
2

where νD is the Doppler frequency.

As an illustration, in Fig. 6.6, we provide the average DPSK bit error probabilities for different Rician fading parameters
K 2 {0, 9, 18} and normalized Doppler frequencies νDT 2 {0.2, 0.02} under the Clark-Jakes (uniform scattering) model.
Clearly, in the presence of Rayleigh fading and Doppler spread, the DPSK system is facing significant performance
degradation and very strong error floor phenomenon.

6.3 Channel Equalization Techniques

In this section, we describe various equalization techniques suitable to deal with different channel impairments in both
wireless and optical communication systems.

6.3.1 ISI-Free Digital Transmission and Partial-Response Signaling

We will start with the equivalent wireless/optical channel model by observing only one polarization state (either x- or
y-polarization state), which is shown in Fig. 6.7. The function hTx(t) denotes the equivalent impulse response of transmitter,
which is in wireless communications the convolution of pulse shaper and power amplifier. In optical communications, hTx(t)
is obtained as the convolution of pulse shaper impulse response, driver amplifier, and Mach-Zehnder modulator (MZM)
impulse responses. In the same figure, hc(t) denotes the impulse response of wireless/optical channel. hRx(t) denotes the
326 6 Advanced Detection Techniques and Compensation of Channel Impairments

z(t)

a(t) b(t) c(t)

Transmitter filter Optical/wireless Receiver filter
hT(t) channel hc(t) hRx(t)
r(t)
a t bk t kT b t bk hTx t kT c t bk hTx hc hRx t kT
k
k k h

Fig. 6.7 Equivalent low-pass wireless/optical channel model (only single polarization is observed to facilitate explanations)

equivalent impulse response of the receiver, which is in wireless communications obtained as the convolution of receive
antenna impulse response and low-noise amplifier impulse response. In optical communications with coherent optical
detection, hRx(t) is obtained as the convolution of impulse responses of photodetector impulse response, front end with
transimpedance amplifier impulse response, and impulse responses of both optical and electrical filters that are employed. The
electrical filter can be replaced by matched filter. Finally, z(t) denotes the equivalent noise process dominated by Gaussian
noise in wireless communications and ASE noise in optical communications.
The equivalent impulse response of this system is given by convolution h(t) ¼ (hTx  hc  hRx)(t), so that the output of the
system can be written as
X
cð t Þ ¼ bk hðt  kT Þ, ð6:115Þ
k

where bk is the transmitted sequence of symbols, while T denotes the symbol duration. By sampling the system output, we
obtain
X X
cðmT Þ ¼ bk hðmT  kT Þ ¼ bm hð0Þ þ bk hðmT  kT Þ
k k6¼m
X
¼ bm hð 0Þ þ bk hðkT Þ, ð6:116Þ
k6¼0
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}
ISI term

where the term bmh(0) corresponds to the transmitted symbol, while the second term represents the intersymbol interference
(ISI) term. The second term can disappear if ISI-free optical transmission is achieved, which is possible when transmitter and
receiver are properly designed such that the equivalent impulse response satisfies the following condition:

h0 ¼ 1, k ¼ 0
hðkT Þ ¼ : ð6:117Þ
0, k 6¼ 0

If the frequency response of equivalent system function H( f ) ¼ FT[h(t)], where FT stands for the Fourier transform,
satisfies the following condition

X
1  
k
H fþ ¼ T, ð6:118Þ
k¼1
T

then there will not be ISI at sampling time instances. This claim is known as Nyquist’s criterion for zero-ISI and can be used in
the design of all channel equalizers.
The above claim is quite straightforward to prove. Let us start with inverse Fourier transform definition and let us split the
integration interval into frequency bands of bandwidth 1/T, so we have that

ð2kþ1Þ=2T
ð
1
X
1 ð
1 j2πft j2πft
hðt Þ ¼ FT fH ð f Þ g ¼ H ð f Þe df ¼ H ð f Þe df : ð6:119Þ
k¼1
1 ð2k1Þ=2T
6.3 Channel Equalization Techniques 327

By sampling the equivalent impulse response from Eq. (6.119), we obtain

ð2kþ1Þ=2T
X
1 ð
j2πfmT
hðmT Þ ¼ H ð f Þe df : ð6:120Þ
k¼1
ð2k1Þ=2T

Now, by substituting f’ ¼ f-k/T, Eq. (6.120) becomes

X
1 ð
1=2T
ð zfflfflfflfflfflfflfflfflfflfflfflfflfflffl
X1
1=2T ffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
0
j2π f 0 mT
hðmT Þ ¼ H ð f 0 þ k=T Þe j2π f mT d f 0 ¼ H ð f 0 þ k=T Þ e df0
k¼1 k¼1
1=2T 1=2T
ð6:121Þ
ð
1=2T (
j2π f 0 mT sin ðmπ Þ 1 m¼0
¼ Te df0 ¼ ¼ :
mπ 0 m 6¼ 0
1=2T

We can see from Eq. (6.121) that the ISI term disappeared, which proves that Nyquist’s criterion indeed yields to zero-ISI
among symbols. The channel with frequency response satisfying Eq. (6.118) with a single term is known as the ideal Nyquist
channel. The transfer function of the ideal Nyquist channel has a rectangular form given as H( f ) ¼ (1/Rs)rect( f/Rs) (Rs is the
symbol rate defined as Rs ¼ 1/T ), while the corresponding impulse response is given as

sin ðπt=T Þ
hð t Þ ¼ ¼ sincðt=T Þ, ð6:122Þ
πt=T

where the sinc-function is defined as sinc(u) ¼ sin (πu)/(πu).

The ideal rectangular frequency response has sharp transition edges, which is difficult to implement in practice. The
frequency response (spectrum) satisfying Eq. (6.118) with additional terms (k ¼ 1, 0, 1) is known as the raised-cosine
(RC) spectrum given by
(
T, 0  jf j  ð1  βÞ=2T
H RC ðf Þ ¼ T 1 ð6:123Þ
f1  sin ½πTðjf j  Þ=βg, ð1  βÞ=2T  jf j  ð1 þ βÞ=2T
2 2T

The impulse response of the raised-cosine spectrum is given as

sin ðπt=T Þ cos ðβπt=T Þ

hRC ðt Þ ¼ : ð6:124Þ
πt=T 1  4β2 t 2 =T 2

The parameter β from Eqs. (6.123) and (6.124) is known as the roll-off factor and determines the bandwidth required to
accommodate the raised-cosine spectrum, which is given by BW ¼ (1+β)/(2T ). The raised-cosine spectrum reduces to the
ideal Nyquist channel if β ¼ 0. It is evident from the raised-cosine filter impulse response that in addition to zero-crossings at
T, 2T, . . ., there are additional zero-crossings at points 3T/2, 5T/2, . . . . The RC impulse responses and spectra for
several roll-off factors are illustrated in Fig. 6.8.
In fiber-optics communications, when different wavelength responses are sinc-functions, the corresponding scheme is
commonly referred to as the Nyquist-WDM scheme. In this scheme, the pulse shapes in the time domain are rectangular.
When the Nyquist-WDM channel responses are of raised-cosine type in wavelength domain, the corresponding WDM
channel spacing will be (1+β)Rs.
In Sect. 6.1, we studied different approaches to minimize symbol error probability and learned that employment of the
matched filter is optimum for Gaussian noise-dominated channels. On the other hand, in this section, we learned that for
ISI-free transmission, the receive filter should be chosen in such a way that the overall system function has the raised-cosine
328 6 Advanced Detection Techniques and Compensation of Channel Impairments

0.8

0.6

0.4
h(t)
0.2

-0.2

-0.4 t / Ts
-4 -3 -2 -1 0 1 2 3 4

(a)
1

0.9

0.8

0.7

0.6

H ( f ) 0.5

0.4

0.3

0.2

0.1

0 f Ts
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

(b)
Fig. 6.8 Raised-cosine function: (a) impulse response and (b) frequency response

form. By combining these two approaches, we can come up with a near-optimum transmit and receive filters. Namely, from
the zero-ISI condition, we require that

hðt Þ ¼ ðhTx  hc  hRx Þðt Þ ¼ hRC ðt Þ, ð6:125Þ

or in the frequency domain

jH RC j ¼ jH Tx H c H Rx j: ð6:126Þ

The condition for matching filter can be written as

6.3 Channel Equalization Techniques 329

Table 6.1 Common partial-response signaling schemes

Type
of class N w0 w1 w2 w3 w4 Class is also known as:
I 2 1 1 Duobinary
II 3 1 2 1
III 3 2 1 1
IV 3 1 0 1 Modified duobinary
V 5 1 0 2 0 1

jH Rx ð f Þj ¼ jH Tx ð f ÞH c ð f Þj: ð6:127Þ

When these two criteria are combined, we obtain that the following is valid:
pffiffiffiffiffiffiffiffiffi
H RC pffiffiffiffiffiffiffiffiffi
jH Tx j ¼ , jH Rx j ¼ H RC : ð6:128Þ
jH c j

Notice that this still represents an approximate solution, as two problems get solved independently and then get combined.
In our discussion above, we considered the ISI as a detrimental effect. However, by inserting the ISI in a controllable
manner, it is possible to transmit signal at a Nyquist rate of 2BW, where BW is the bandwidth of the channel. The systems
using this strategy are known as partial-response (PR) signaling schemes, and they are widely used in magnetic recording and
various digital communications [1, 32, 33]. The impulse response of PR systems is given as

X
N 1
hð t Þ ¼ wn sincðt=T  nÞ: ð6:129Þ
n¼0

The common PR signaling schemes are listed in Table 6.1. For instance, the scheme with N ¼ 2 is known as a duobinary
scheme, while the scheme with N ¼ 3 and weight coefficients 1, 0, 1 is known as a modified duobinary scheme.

6.3.2 Zero-Forcing Equalizers

To compensate for linear distortion effects, such as chromatic dispersion in fiber-optics communications and multipath effects
in wireless communications, we may use the circuit known as equalizer, connected in cascade with the system in question, as
shown in Fig. 6.9a. The overall frequency response is equal to Hc(ω)Heq(ω), where Hc(ω) is the system transfer function,
while Heq(ω) is the equalizer transfer function. We assume that balanced coherent detection is used. For distortionless
transmission, we require that

H c ðωÞH eq ðωÞ ¼ ejωt0 , ð6:130Þ

so that the frequency response of the equalizer is inversely proportional to the system transfer function, as given below:

ejωt0
H eq ðωÞ ¼ : ð6:131Þ
H c ð ωÞ

A functional block that is well suited for equalization in general is tapped-delay-line (transversal) equalizer, or the finite
impulse response (FIR) filter, which is shown in Fig. 6.9b. The total number of filter taps is equal to N+1.
The FIR equalizer design can be done by using some of the methods described in [26], which are (i) the symmetry method,
(ii) the window method, (iii) the frequency sampling method, and (iv) the Chebyshev approximation method. The FIR filters
obtained by symmetry method have a linear phase so that their discrete impulse response h[n] and system function H(z) satisfy
the following symmetry property [26]:
330 6 Advanced Detection Techniques and Compensation of Channel Impairments

Fig. 6.9 (a) Equalization Hc( ) Heq( )

principle and (b) tapped-delay-
line equalizer (FIR filter)
Dispersive
channel Equalizer

Distortionless transmission
system
(a)
x[n]
z-1 z-1 z-1
h[1] h[N]
h[0] h[2]

y[n]
(b)

h½n ¼ h½N  1  n; n ¼ 0, 1, ⋯, N  1

  ð6:132Þ
H ðzÞ ¼ Z fh½ng ¼ zðN1Þ H z1 ,

where Z{} presents the z-transform of impulse response h[n], as defined in Chap. 3. From the symmetry property (6.132), we
conclude that the roots of H(z) must be the same as the roots of H(z-1), which indicates that the roots of H(z) must occur in
reciprocal pairs: zk, zk-1. Moreover, for filter coefficients to be real valued, the roots must appear in complex-conjugate pairs:
zk, zk-1, z*k, z*k-1.
The windowing method starts with a specified filter order N (N is typically an even integer), and for a given sampling
interval Ts, the following steps are taken: (i) set the constant time delay t0 ¼ (N/2)/Ts; (ii) take the inverse Fourier transform of
Heq(ω) to obtain a desired impulse response heq(t); and (iii) set h[n] ¼ w[n]heq[nTs], where w[n] is a window of length (N+1).
The simplest windowing function is a “rectangular window” given by

1, n ¼ 0, 1, . . . , N  1
wðnÞ ¼ : ð6:133Þ
0, otherwise

Therefore, the frequency response of smoothed equalizer can be obtained as the convolution H(Ω) ¼ Heq(Ω)  W(Ω),
where Ω ¼ ωTs is the discrete frequency (expressed in rad/sample). For large filter order having rectangular window, the
Gibbs phenomenon in H(Ω) can be noticed [26], which comes from abrupt truncation of heq[n]. For better results, the tapered
windows can be used. The commonly used tapered windowing functions include Blackman, Hamming, Hanning, Kaiser,
Lanczos, and Tukey windows [26]. For instance, the Hanning window is given by the function

1 2πn
w½n ¼ 1  cos : ð6:134Þ
2 N1

Tapered windows reduce the Gibbs ringing but increase the transition bandwidth (they provide better smoothing but less
sharp transition).
In fiber-optics communications, the first-order polarization mode dispersion (PMD) can be compensated by using four FIR
filters. The output symbols in x-polarization can be determined by [21]

X
N 1  
x½k  ¼ hTxx x0 þ hTxy y0 ¼ hxx ½nx0 ½k  n þ hxy ½ny0 ½k  n , ð6:135Þ
n¼0
6.3 Channel Equalization Techniques 331

x’ x
hxx

hxy

hyx
y
hyy
y’

Fig. 6.10 The compensation of polarization-dependent impairments in fiber-optics communications by FIR filters

where hij (i,j 2 {x,y}) are responses of FIR filters, each with N taps. The corresponding equation to determine the symbols in
y-polarization can be written in a similar fashion, and the PDM compensation procedure can be summarized as shown in
Fig. 6.10.
The design of an FIR equalizer to compensate for chromatic dispersion in fiber-optics communications is a straightforward
procedure when the windowing method is used. As shown in Chap. 2, the impact of chromatic dispersion can be described by
the following transfer function:
 

H c ðωÞ ¼ exp j β2 ω2 =2 þ β3 ω3 =6 L , ð6:136Þ

where β2 and β3 are the group-velocity dispersion (GVD) and second-order GVD parameters, respectively, while L is the fiber
length. In particular, when the second-order GVD parameter can be neglected, by taking inverse Fourier transform of
Eq. (6.136), and applying the windowing method, the FIR equalizer coefficients become
sffiffiffiffiffiffiffiffiffiffiffi    
j k j k jDjλ2 L
jcT 2 πcT 2 2 N N
h½n ¼ exp j 2 n ,  n ,N ¼ 2 þ 1, ð6:137Þ
Dλ2 L Dλ L 2 2 2cT 2

where D is the dispersion parameter, related to GVD parameter β2 by D ¼(2πc/λ2)β2.

The frequency sampling method is based on sampling of frequency response of the equalizer as follows:

H ðk Þ ¼ H ðΩÞjΩk ¼2πk=N : ð6:138Þ

The function H(Ω) can be expressed by discrete-time Fourier transform as

X
N 1
H ð ΩÞ ¼ h½nejΩn , ð6:139Þ
n¼0

and, since the frequency response is related to z-transform by setting z ¼ exp(jΩ), we can easily determine the system function
H(z). This is particularly simple to do when the system function is decomposable as follows:

1  zN X
N 1
H ðk Þ
H ðzÞ ¼ ¼ H comb ðzÞH bank ðzÞ, ð6:140Þ
N ffl} k¼0 1  e j2πk=N z1
|fflfflfflffl{zfflfflffl
H comb ðzÞ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
H bank ðzÞ

where Hcomb(z) represents the all-zero system or a comb filter, with equally spaced zeros located at the unit circle (zk ¼ ej2π(k
+α)/N
, with α ¼ 0 or 0.5), while Hbank(z) consists of a parallel bank of single-pole filters with resonant frequencies equal to pk ¼
ej2π(k+α)/N. The unit-sample impulse response h[n] needs now to be determined from H(z). The main drawback of this method
is nonexistence of control of H(Ω) at frequencies in between points Ωk.
332 6 Advanced Detection Techniques and Compensation of Channel Impairments

In practice, the optimum equiripple method is used to control H(Ω). Namely, the frequency response H(Ω) of actual
equalizer is different from desired function Hd(Ω). Consequently, the error function E(Ω) can be found as the difference
between the desired and actual frequency responses:

E ðΩÞ ¼ W ðΩÞ½H d ðΩÞ  H ðΩÞ, ð6:141Þ

where W(Ω) is known as a weighting function. The key idea behind this method is to determine the h[k] by minimizing the
maximum absolute error (the minimax criterion), which can be expressed as
 
b
h½n ¼ arg min max jEðΩÞj , ð6:142Þ
over hðnÞ Ω2S

where S denotes the disjoint union of frequency bands over which the optimization is performed. For linear phase FIR filter
design, it is possible to use the Remez algorithm [26]. On the other side, for problems involving both quadratic and cubic
phase terms, the Chebyshev approximation method can be used instead.
We can conclude from Eq. (6.137) that chromatic dispersion compensator is essentially an all-pass filter with constant
magnitude of the response, |H(Ω)|¼1, 0Ωπ. An important property of all-pass filters is that zero and poles are reciprocals
of one another; in other words,

Aðz1 Þ X N
H ðzÞ ¼ zN , AðzÞ ¼ 1 þ a½nzn ð6:143Þ
AðzÞ n¼1


with response magnitudes jH ðωÞj2 ¼ H ðzÞH ðz1 Þz¼e jΩ ¼ 1. Upon factorization, the system function of an all-pass filter is
given by [26]

Nr z1  αk N c ðz1  βk Þðz1  βk Þ

H ðzÞ ¼ ∏ ∏ ;
k¼1 1  αk z1 k¼1 ð1  βk z1 Þð1  βk z1 Þ ð6:144Þ
jαk j < 1, jβk j < 1,

where Nr is the number of sections with simple poles, while Nc is the number of sections with complex-conjugate poles. By
interpreting the chromatic dispersion compensator design as the all-pass filter design, it is expected that the number of taps will
be smaller than that needed in FIR filter design (for which the number of taps is a linear function of fiber length). It is evident
from Eq. (6.144) that the all-pass filter is essentially infinite impulse response (IIR) filter. Additional details on IIR all-pass
filter design can be found in [27] and [28].
The function of FIR/IIR equalizers described above is to ensure the zero-ISI at all instances t ¼ nT, except for n ¼
0. Because of that, they are also known as the zero-forcing equalizers. However, the problem in their employment is that the
Gaussian noise, chromatic dispersion, and PMD in fiber-optics communications all act together, similar to multipath fading
and Gaussian noise in wireless communication, affecting the behavior of a transmission system in a combined manner, while
transversal equalizers ignore the effect of channel noise. This leads to the noise enhancement phenomenon, which can be
explained as follows. Let us again consider the model shown in Fig. 6.9, but now with the impulse response of receive filter
being decomposed into matched filter impulse response and equalizer impulse response:

hRx ðt Þ ¼ hTx ðt Þ  heq ðt Þ: ð6:145Þ

The Fourier transform of system output, assuming Heq( f ) ¼ FT{heq(t)} ¼ 1/Hc( f ), is given by

Cð f Þ ¼ ½H T ð f ÞH c ð f Þ þ N ð f ÞH T ð f ÞH eq ð f Þ
N ð f ÞH T ð f Þ
¼ j H T ð f Þj 2 þ : ð6:146Þ
Hcð f Þ
|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
N0 ð f Þ
6.3 Channel Equalization Techniques 333

The corresponding power spectral density of colored Gaussian noise N’( f ) is given by

2
N 0 j H T ð f Þj
PSDN 0 ð f Þ ¼ 2
: ð6:147Þ
2 j H c ð f Þj

We can see that the ISI is compensated for, but the noise is enhanced unless we have that |HTx( f )|¼|Hc( f )|.

6.3.3 Optimum Linear Equalizer in MMSE Sense

The better approach for receiver design, which avoids the noise enhancement phenomenon, would be to use the minimum
mean-square error (MMSE) criterion to determine the equalizer coefficients, which provides a balanced solution to the
problem by taking the effects of both channel noise and ISI into account. Let hRx(t) be the receiver filter impulse response and
x(t) be the channel output determined by
X
xðt Þ ¼ sk qðt  kT Þ þ zðt Þ, qðt Þ ¼ hTx ðt Þ  hc ðt Þ, ð6:148Þ
k

where hTx(t) is the transmit filter impulse response, hc(t) is the channel impulse response, sk is the transmitted symbol at the kth
time instance, and z(t) is the additive channel noise. (As before, T denotes the symbol duration.) The receive filter output can
be determined by the convolution of receive filter impulse response and corresponding input as

ð
1

yð t Þ ¼ hRx ðτÞxðt  τÞdτ: ð6:149Þ

1

By sampling at t ¼ nT, we obtain the following:

X ð
1

yðnT Þ ¼ ζ n þ zn , ζ n ¼ sk hRx ðτÞqðnT  kT  τÞdτ,

k
1
ð6:150Þ
ð
1

zn ¼ hRx ðτÞzðnT  τÞdτ:

1

The signal error can be defined as a difference between receive sample and transmitted symbol as follows:

en ¼ yðnT Þ  sn ¼ ζ n þ zn  sn : ð6:151Þ

The corresponding mean-square error is

1  2 1  2 1  2 1  2
MSE ¼ e ¼ ζ n þ zn þ sn þ hζ n zn i  hζ n sn i  hzn sn i: ð6:152Þ
2 n 2 |{z} |fflffl{zfflffl}
ð 1 |fflffl{zfflffl} |fflffl{zfflffl}
2 2
¼1 ¼0 ¼0
hRx ðτÞqðτÞdτ
1

Assuming the stationary environment, the equation above can be rewritten as

334 6 Advanced Detection Techniques and Compensation of Channel Impairments

Transversal equalizer
x(t)
q(-t) z-1 z-1 z-1 z-1
c-N c-N+1 cn cN-1 cN
Matched
filter

y(t) y(nT)

Fig. 6.11 Optimum linear equalizer (in the MMSE sense)

ð
1 ð
1
h i
1 1 N
MSE ¼ þ Rq ðt  τÞ þ 0 δðt  τÞ hRx ðt ÞhRx ðτÞdtdτ
2 2 2
1 1
ð6:153Þ
ð
1

 hRx ðt Þqðt Þdt,

1

where Rq(τ) is the autocorrelation function of q(t) and N0 is the power spectral density of Gaussian noise. In order to determine
the optimum filter in the MMSE sense, we have to find the derivative of MSE with respect to hRz(t) and set it to zero, which
yields to

ð
1
h i
N
Rq ðt  τÞ þ 0 δðt  τÞ hRx ðτÞdτ ¼ qðt Þ: ð6:154Þ
2
1

By applying the Fourier transform (FT) of the equation above and solving it for receive filter transfer function HRx( f ) ¼ FT
[hRx(t)], we obtain the following:

Q ð f Þ 1
H Rx ð f Þ ¼ ¼ Q ð f Þ , ð6:155Þ
Sq ð f Þ þ N20 Sq ð f Þ þ N20
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}
H eq ð f Þ

where the impulse response of equalizer is obtained as heq(t) ¼ FT-1[Heq( f )], followed by the windowing method described
above. Therefore, the optimum linear receiver, in the MMSE sense, consists of the cascade connection of matched filter and
transversal equalizer, as shown in Fig. 6.11.
If delay Td is equal to the symbol duration T, the corresponding equalizer is known as symbol-spaced equalizer. If the
symbol rate (Rs ¼ 1/T ) is smaller than 2BW (BW is the channel bandwidth), the equalizer will need to compensate for both
channel distortions and aliasing effect. On the other hand, when Td is chosen to satisfy the condition 1/Td  2BW > Rs, aliasing
problem is avoided, and equalizer needs to compensate for channel distortions only. This type of equalizer is known as
fractionally spaced equalizer, and the most common case is the one for which Td ¼ T/2.

6.3.4 Wiener Filtering

We have already discussed in Sect. 6.1.5 how to estimate a certain signal parameter in the presence of additive noise. In this
section, we extend that study to the estimation of random vector s (of length N ) from the received vector r. We are specifically
interested in linear (unbiased) minimum error variance estimation [2]. Namely, we would like our estimate to be a linear
function of a received vector, which can be expressed as
6.3 Channel Equalization Techniques 335

bs ¼ Ar þ b, ð6:156Þ

where A and b are the matrix and vector, respectively, that need to be determined. The vector b can be determined from the
requirement for estimate to be unbiased as follows:

hbsi ¼ ms = A hri þ b = Amr þ b, ð6:157Þ

|{z} |{z}
hsi mr

and by solving for b, we obtain

b = ms  Amr : ð6:158Þ

The error made by unbiased estimator will be then defined as

ε =bs  s = ðbs‐ms Þ  ðs‐ms Þ ¼ Aðr  mr Þ  ðs‐ms Þ: ð6:159Þ

|fflffl{zfflffl}
Aðrmr Þ

The matrix A can be determined from the condition that trace of covariance matrix, defined as
 
CovðεÞ ¼ εεΤ , ð6:160Þ

is minimized. By setting the first derivative of trace of covariance matrix with respect to matrix A from Eq. (6.160) to be zero,
we derive the following orthogonality principle:
 
ðr  mr ÞεΤ ¼ 0: ð6:161Þ

The orthogonality principle indicates that error vector is independent of the data. By substituting Eq. (6.159) into
Eq. (6.161) and by solving for A, we obtain

A ¼ CTrs C1 T T
r ; Crs ¼ Efðr  mr Þðx  ms Þ g, Cr ¼ Efðr  mr Þðr  mr Þ g, ð6:162Þ

where Crs and Cr are the corresponding covariance matrices, and E{}¼<> denotes the expectation operator. Now, by
substituting Eqs. (6.162) and (6.158) into Eq. (6.156), we obtain the following form of the estimated transmitted vector:

bs ¼ CTrs C1
r ðr  mr Þ þ ms , ð6:163Þ

which represents the most general form of the Wiener filter. The minimum error variance of this estimator is given by
 
εεΤ ¼ Cs  Csr C1
r Csr :
T
ð6:164Þ

The Wiener filtering is also applicable in polarization-division multiplexed (PDM) systems. In PDM, the transmitted vector
s can be represented as s ¼ [sx sy]T, where the subscripts x and y are used to denote x- and y-polarization states. In a similar
fashion, the received vector r can be represented as r ¼ [rx ry]T. The only difference in Wiener filtering problem is just in
dimensionality of vectors s and r, which in PDM have twice more components (2N ). The dimensionality of matrix A is also
higher, namely, 2N2N.

6.3.5 Adaptive Equalization

So far, we have assumed that different channel impairments in either wireless or optical communication systems are time
invariant, which is not completely true, especially with respect to multipath fading in wireless communications as well as the
336 6 Advanced Detection Techniques and Compensation of Channel Impairments

x[n] x[n-1] x[n-N+1] x[n-N]

z-1 z-1 … z-1

… Variable
w0 w1 wN-1 wN weights

…
y[n]
e[n]
Error signal
d[n]
Desired response
Fig. 6.12 Adaptive equalizer

PMD and polarization-dependent losses (PDL) in fiber-optics communications. The adaptive filtering offers an attractive
solution to address the channel equalization [1, 3, 4, 26]. An adaptive filter has a set of adjustable filter coefficients, as shown
in Fig. 6.12, which are adjusted based on the algorithms described below. The widely used adaptive filter algorithms are the
steepest descent algorithm and the least-mean-square (LMS) algorithm [3, 4, 26]. These algorithms can be used to determine
the coefficients of transversal equalizer.
According to the steepest descent algorithm, we update the kth filter coefficient wk, shown in Fig. 6.12, by making
correction of the present value in the direction opposite to the gradient ∇k, in other words, in the direction of the steepest
descent on the error-performance surface [3], as follows:

1
wk ½n þ 1 ¼ wk ½n þ μf∇k ½ng; k ¼ 0, 1, . . . , N, ð6:165Þ
2

where the real-valued parameter μ determines the convergence speed.

When the error signal e[n] is defined as the difference of desired signal (commonly the received sequence of a training
sequence in a back-to-back configuration) and the output y[n] of corresponding FIR filter output, we can write

X
N
e ½ n ¼ d ½ n  wk ½nx½n  k: ð6:166Þ
k¼0

The gradient ∇k can be determined as

∂
∇k ½n ¼ he2 ½ni ¼ 2he½nx½n  ki ¼ 2REX ½k;
∂wk ½n ð6:167Þ
k ¼ 0, 1, . . . , N,

where REX[k] is the cross-correlation function between the error signal and the adaptive filter input. By substituting Eq. (6.167)
into Eq. (6.165), we obtain the following form of steepest descent algorithm:

wk ½n þ 1 ¼ wk ½n þ μREX ½k; k ¼ 0, 1, . . . , N: ð6:168Þ

It can be shown that for this algorithm to converge the following condition for the parameter μ should be satisfied:

0 < μ < μmax , ð6:169Þ

where μmax is the largest eigenvalue of the correlation matrix of the filter input RX:
6.3 Channel Equalization Techniques 337

Fig. 6.13 Decision-feedback Transmitted symbol

equalizer (DFE) estimate
Input sequence
+ r[n]
Feedforward Decision
x[n] section device
b̂ n

Feedback
section

2 3
R X ð 0Þ RX ð1Þ . . . R X ðN Þ
6 7
6 7
6 R X ð 1Þ RX ð0Þ . . . RX ðN  1Þ 7
6 7
RX ¼ 6
6
7,
7
6 ... ... ... ... 7 ð6:170Þ
6 7
4 5
R X ðN Þ RX ðN  1Þ . . . R X ð 0Þ

RX ðlÞ ¼ hx½nx½n  li:

The main drawback of the steepest descent algorithm is that it requires the knowledge of gradient ∇k in each iteration. On
the other side, the key idea in LMS algorithm is to approximate the operator of averaging h.i in Eq. (6.167) by its instantaneous
value e[n]x[n-k], so that the LMS weight update rule is simply

ðLMSÞ ðLMSÞ
wk ½n þ 1 ¼ wk ½n þ μe½nx½n  k; k ¼ 0, 1, . . . , N: ð6:171Þ

6.3.6 Decision-Feedback Equalization

Another relevant equalizer is the decision-feedback equalizer (DFE), which is shown in Fig. 6.13. The key idea of DFE is to
use the decision made on the basis of precursors of the channel impulse response in order to take care of post-cursors. The
assumption in DFE is that decisions in decision circuit are correct. Let the channel impulse response in discrete form be
denoted by hc[n]. The response of the channel, in the absence of noise, to an input sequence x[n] is given by
X
y½n ¼ hc ½k x½n  k
k
X X
¼ hc ½0x½n þ hc ½kx½n  k þ hc ½kx½n  k: ð6:172Þ
k<0 k>0
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
precursors postcursors

The first term on the right side h[0]x[n] represents the desired data symbol; the second term is a function of previous
samples only (where the channel coefficients are known as precursors), while the third term is a function of incoming samples,
with channel coefficients being known as post-cursors. The DFE is composed of a feedforward equalizer (FFE) section, a
feedback equalizer (FBE) section, and a decision device, as shown in Fig. 6.13.
The feedforward and feedback sections can be implemented as FIR filters (transversal equalizers) and can also be adaptive.
The input to the detector can be written as

X
N FFE X
N FBE

r ½m ¼ f ½nx½mT  nT   g½nb

b½m  n, ð6:173Þ
n¼1 n¼1
338 6 Advanced Detection Techniques and Compensation of Channel Impairments

x[n]
z 1 z 1 z 1

f [-3] f [-2] f [-1] f [0]

Decision
device
-
Error
signal +
accumulators
T T T

g[3] g[2] g[1]

Fig. 6.14 An adaptive decision-feedback equalizer example

z(t)
Combined channel h(t)=hTx(t) hc(t)
d(t) c(t) Estimated data
Transmitter filter Optical/wireless Matched filter MLSD
hTx(t) channel hc(t) h(-t) (MLSE) sequence sL
r(t)
d t sk t kTs
k

Fig. 6.15 Equivalent system model to study MLSD (to facilitate explanations only single-polarization case is studied)

where f[n] and g[n] are values of adjustable taps in feedforward and feedback sections, respectively. The NFFE (NFBE) denotes
the number of taps in feedforward (feedback) section.
We can see that the feedforward section is a fractionally spaced equalizer, which operates with rate equal to an integer
multiple of the symbol rate (1/Td ¼ m/T, m is an integer). The taps in feedforward and feedback sections are typically chosen
based on MSE criterion, by employing the LMS algorithm. An example of the adaptive decision-feedback equalizer is shown
in Fig. 6.14. Since the transmitted symbol estimate is dependent on previous decisions, this equalizer is a nonlinear functional
module. If previous decisions are in error, the error propagation effect will occur. However, the errors do not persist
indefinitely, but rather occur in bursts.

6.3.7 MLSD (MLSE) or Viterbi Equalization

Another very important equalizer is based on maximum likelihood sequence detection (estimation) (MLSD or MLSE) [34–
36]. Since this method estimates the sequence of transmitted symbols, it avoids the problems of noise enhancement and error
propagation. The MLSE chooses an input sequence of transmitted symbols {sk} in a way to maximize the likelihood function
of received signal r(t). The equivalent system model that illustrates MLSD is shown in Fig. 6.15.
The received signal r(t) can be expressed in terms of a complete set of orthonormal basis functions {Φn(t)} as follows:

X
N
r ðt Þ ¼ r n Φn ðt Þ, ð6:174Þ
n¼1

where
6.3 Channel Equalization Techniques 339

X
1 X
L
rn ¼ sk hnk þ zn ¼ sk hnk þ zn ;
k¼1 k¼0
ð
LT ð
LT ð6:175Þ
hnk ¼ hð t  kT ÞΦn ðt Þdt; zn ¼ zðt ÞΦn ðt Þdt; hðt Þ ¼ hTx ðt Þ  hc ðt Þ:
0 0

The impulse responses hTx(t) and hc(t) used in the equation above denote (as before) impulse responses of transmitter and
channel, respectively. The N denotes the cardinality of the basis set, L is the channel memory expressed in the number of
symbols, while h(t) is the combined channel impulse response (the convolution of transmit and channel impulse responses).
Since zn are Gaussian random variables, the distribution of r[1,N] ¼ [r1 r2 . . . rN]T is a multivariate Gaussian:
8  2 2 39
N <  XL  =
1 1  
f ðr½1, N js½1, L, hðt ÞÞ ¼ ∏ exp 4 r n  sk hnk  5 , ð6:176Þ
n¼1 :πN 0 N0  k¼0
 ;

where N0 is the power spectral density of the Gaussian noise. The MLSD decides in favor of the symbol sequence s[1,L] ¼ [s0
s1 . . . sL] that maximizes the likelihood function given by Eq. (6.176), so that we can write

 2 2 3
XN  XL 
 
bs½0, L ¼ arg max f ðy½1, N js½0, L, hðt ÞÞ ¼ arg max 4 r n  sk hnk  5
s½0, L s½0, L n¼1
 k¼0

( " # ) ð6:177Þ
X X N XX X
N
¼ arg max 2 Re sk r n hnk  sk sm hnk hnm :
s½0, L k n¼1 k m n¼1

By using substitution u(t) ¼ h(t)  h(t) and noticing that the following is valid:

X
N ð
1

r n hnk ¼ r ðτÞh ðτ  kT Þdτ ¼ c½k ,

n¼1
1
ð6:178Þ
X
N ð
1

hnk hnm ¼ hðτ  kT Þh ðτ  mT Þdτ ¼ uðmT  kT Þ ¼ u½m  k

n¼1
1

Eq. (6.177) can be simplified to

( # " )
X XX
 
bs½0, L ¼ arg max 2 Re sk c½k   sk sm u½m  k : ð6:179Þ
s½0, L k k m

Equation (6.179) can efficiently be calculated by the Viterbi algorithm [1] (see also Chap. 9). However, the Viterbi
algorithm provides the hard decisions and as such is not suitable for use with soft-decision decoding/detection schemes. To
fully exploit the advantages of soft decoding, the soft reliabilities are needed. These reliabilities can be obtained by soft-output
Viterbi algorithm (SOVA) [37], BCJR algorithm [38], or by Monte Carlo-based equalization [39]. To further improve the
BER performance, we can perform the iteration of extrinsic information between soft equalizer and soft decoder, the
procedure known as turbo equalization [40–42], which will be described in Chap. 9. The turbo equalization scheme can be
used in fiber-optics communications to simultaneously compensate for chromatic dispersion, PMD, PDL, and fiber
nonlinearities, as shown in [43].
If Gaussian-like approximation is not valid, which is the case with the presence of strong fiber nonlinearities in fiber-optics
communications or in satellite communications, the conditional PDF f(r[1, N]| s[0, L], h(t)) should be estimated from
histograms, by propagating the sufficiently long training sequence.
340 6 Advanced Detection Techniques and Compensation of Channel Impairments

6.3.8 Blind Equalization

In MMSE equalizer, the use of training sequences is needed for initial adjustment of equalizer coefficients. However, the
adjustment of equalizer coefficients can be done without using the training sequences, and this approach is known as blind
equalization (or self-recovery equalization) [44–50]. All blind equalizers can be classified into three broad categories [1]:
(i) steepest descent algorithm based, (ii) high-order statistics based, and (ii) maximum likelihood approach based.
To facilitate explanation of blind equalization, let us observe a signal polarization state. The output of optical fiber channel
upon balanced coherent detection, assuming only either x- or y-polarization, can be expressed as

X
K
r ½ n ¼ h½ka½n  k  þ z½n, ð6:180Þ
k¼0

where h[n] represents the optical channel impulse response (in either x- or y-polarization), a[n] is the transmitted sequence of
symbols, while z[n] are samples of additive Gaussian noise. The joint PDF of received sequence r[1,L] ¼ [r1 r2 . . . rL]T, under
Gaussian noise assumption and for given impulse response vector h[0,K] ¼ [h0 h1 . . . hK]T and transmitted sequence a[1,L] ¼
[a1 . . . aL], can be written as
2 3
a1 0 0 ⋯ 0
6 7
  6 a2 a1 0 ⋯ 0 7
1 6 7
f R ðrjh, aÞ ¼ ðπN 0 Þ exp  kr  Ahk , A ¼ 6
L 2
6 a3 a2 a1 0 77: ð6:181Þ
N0 6 7
4⋮ ⋮ ⋮ ⋱ ⋮ 5
aL aL1 aL2 ⋯ aLK

When training sequence is used and data vector a is known, the ML estimate of channel impulse response can be obtained
by maximizing the likelihood function (6.181) to obtain

 1
hML ðaÞ ¼ A{ A AT r: ð6:182Þ

On the other hand, when the channel impulse response is known, we can use the Viterbi equalizer, described in the section
above, to detect the most likely transmitted sequence a. However, when neither a nor h is known, we will need to determine
them in such a way that the likelihood function from Eq. (6.181) is maximized. As an alternative solution, we can estimate
h from fR(r|h), which is obtained by averaging over all possible data sequences (MK in total, where M is the signal
constellation size). In other words, the conditional PDF can be determined as

X
MK 
f R ðrjhÞ ¼ f R ðrjh, aÞP aðkÞ : ð6:183Þ
k¼1

The estimate of h that maximizes this new likelihood function from Eq. (6.183) can be obtained from the first derivative of
fR(r|h) with respect to h, which is set to zero, to obtain
" #
X  1 X 
ðk Þ{ ðk Þ krAðkÞ hðiÞ k =N 0
AðkÞ{ rekrA h k =N 0 P aðkÞ :
2 ðk Þ ði Þ 2
ðiþ1Þ ðk Þ
h ¼ A A e P a ð6:184Þ
k k

Given that the corresponding equation is transcendental, we can use an iterative procedure to determine h recursively. Once
h is determined, the transmitted sequence a can be estimated by the Viterbi algorithm that maximizes the likelihood function
from Eq. (6.181) or minimizes the Euclidean distance kr  AhMLk2:
6.3 Channel Equalization Techniques 341

b
a ¼ arg min kr  AhML k2 : ð6:185Þ
a

It is clear that this approach is computationally extensive. Moreover, since h is estimated from average conditional PDF, its
estimate is not going to be that accurate as compared to the case when training sequences are used. The better strategy in this
case will be to perform channel and data estimation jointly.
In joint channel and data estimation, the process is done in several stages. In the first stage, the corresponding ML estimate
of h is determined for each candidate data sequence a(k), which is
 1
hML aðkÞ ¼ AðkÞ{ AðkÞ AðkÞT r; k ¼ 1, ⋯, M K : ð6:186Þ

In the second stage, we select the data sequence that minimizes the Euclidean distance for all channel estimates:
  2
 
a ¼ arg min r  AðkÞ hML aðkÞ  :
b ð6:187Þ
a ðk Þ

For efficient implementation of joint channel and data estimation, a generalized Viterbi algorithm (GVA) in which the best
B (1) estimates of transmitted sequence are preserved is proposed in [47]. In this algorithm, the conventional Viterbi
algorithm up to the Kth stage is applied, which performs an exhaustive search. After that, only B surviving sequences are
retained. This algorithm performs well for B ¼ 4 and for medium SNR values, as shown in [47].
In stochastic-gradient blind equalization algorithms, a memoryless nonlinear assumption is typically used. The most
popular from this class of algorithms is Godard’s algorithm [48], also known as the constant modulus algorithm (CMA). In
fiber-optics communications, this algorithm has been used for PMD compensation [21, 29, 31], which is performed in
addition to compensation of I/Q imbalance and other channel impairments. In conventional adaptive equalization, we use a
training sequence as desired sequence, so it is d[n] ¼ a[n]. However, in blind equalization, we have to generate a desired
sequence d[n] from the observed equalizer output based on a certain nonlinear function:

gðb
a½nÞ, memorylesscase
d ½ n ¼ : ð6:188Þ
gðb
a½n, b
a½n  1, ⋯, b
a½n  mÞ, memoryoforder m

The commonly used nonlinear functions are as follows:

(1) Godard’s function [48] given as

0 D E 1
b
a½ n @ a½nj4
jb
gðb
a½nÞ ¼ a½nj þ D
jb E jb a½nj3 A,
a½nj  jb ð6:189Þ
jb
a½ n j jb
a½nj 2

(2) Sato’s function [49] given as

D E
½ Re ða½nÞ2
gðb
a½nÞ ¼ signðb
a½nÞ , ð6:190Þ
hj Re ða½nÞji

and
342 6 Advanced Detection Techniques and Compensation of Channel Impairments

Decision a n
r n circuit
Adaptive â n
equalizer
Nonlinear d [n]
function,
w n 1 w n g( )
e n r n - +
Error signal
e[n]

Fig. 6.16 The description of generic stochastic-gradient blind equalization algorithm

Fig. 6.17 Godard’s scheme for

simultaneous blind equalization
From either balanced rI n
and carrier phase tracking
coherent optical â n a n
detector or Adaptive Decision
wireless coherent rQ n equalizer circuit
detector
j ˆk
e
Carrier
tracking

(3) Benveniste-Goursat’s function [50] given as

2 D E 3
½ Re ða½nÞ2
gðb
a½ n Þ ¼ b
a½n þ c1 ðb
a½n  a½nÞ þ c2 jb a½nj4signðb
a½ n  e a½ n Þ a½n5,
e ð6:191Þ
hj Re ða½nÞji

where c1 and c2 are the properly chosen positive constants, while ea½n is the decision circuit output as shown in Fig. 6.16. In
the same figure, we also provide the update rule for adaptive equalizer tap coefficients, which is applicable to all three
nonlinear functions described above.
Godard’s algorithm belongs to the class of steepest descent algorithms, and it is widely used when the training sequence is
not available. Godard’s scheme is suitable for simultaneous blind adaptive equalization and carrier phase tracking, as shown in
Fig. 6.17. With rI[n] and rQ[n] we denoted the in-phase and quadrature components of the received signal, after either wireless
coherent detection or optical coherent detection. The equalizer output is represented by discrete-time convolution of the input
complex sequence r[n] ¼ (rI[n], rQ[n]) and the equalizer tap coefficients w[n] given by

X
K
b
a½k  ¼ w½nr ½k  n: ð6:192Þ
n¼K


b k , where ϕ
This output is multiplied with exp jϕ b k is the carrier phase estimate at the kth time instance (symbol
interval). The error signal can be defined as

bk b
e½k ¼ a½k  exp jϕ a½k , ð6:193Þ

which assumes that a[k] were known. The MSE can be minimized with respect to equalizer tap coefficients and carrier phase
estimate to become
6.4 Synchronization Techniques 343

   2 
b  b 
ϕk , w ¼ arg min E a½k   exp jϕk ba½k  : ð6:194Þ
bϕk , w

The LMS-like algorithm can be used to determine the carrier phase estimate and equalizer tap coefficients, respectively, as
n o
bk ¼ ϕ
ϕ b k þ μϕ Im a½kb a ½k  exp jϕ bk ,
 ð6:195Þ
b kþ1 ¼ w
w b k þ μ w a½ k   b a½k  exp jϕ bk :

However, since the desired sequence a[n] is not known, the algorithm above will not converge. To solve this problem, we
can use the cost function that is independent of the carrier phase, defined as
D E
a½kjp  ja½k jp  ,
2
CðpÞ ¼ ½jb ð6:196Þ

where p is a positive integer ( p ¼ 2 is the most used value). However, only signal amplitude will be equalized in this case.
Another more general cost function was introduced for Godard’s algorithm, defined as [48]
D
2 E
D ð pÞ ¼ a½kjp  Rp ,
jb ð6:197Þ

where Rp is a positive real number to be determined. The minimization of cost function from Eq. (6.197) with respect to
equalizer tap coefficients can be done by using the steepest descent algorithm, with the weight coefficient update rule as
follows:

dD ðpÞ
wiþ1 ¼ wi  μp : ð6:198Þ
dwi

After performing a derivation of variance and omitting the expectation operator hi, the following LMS-like adaptation
algorithm can be used:
D E
  a½k j2p
jb
wkþ1 ¼ wk þ μp rk b a½k jp2 Rp  jb
a½kjb a½kjp , Rp ¼ : ð6:199Þ
a½k jp i
hjb

As we can see, the knowledge of carrier phase is not needed to determine equalizer tap coefficients. The algorithm is
a½kjp2 ¼ 1: The algorithm from Eq. (6.199) will converge if all tap coefficients have
particularly simple for p ¼ 2, since jb
been initialized to zero, but with exception of the center tap that was chosen to satisfy the following inequality:
D E
j a½ k  j 4
jw0 j2 > nD Eo2 , ð6:200Þ
2jhmax j ja½kj2

where hmax is the channel impulse response sample with the largest amplitude. The inequality from Eq. (6.200) represents the
necessary but not the sufficient condition for the convergence.

6.4 Synchronization Techniques

There are two basic forms of synchronization employed in digital optical receivers: (i) carrier synchronization, in which the
frequency and phase of optical carrier are estimated; and (ii) symbol synchronization, in which instants of time at which
modulation changes its state are estimated. The synchronization problem can be interpreted as an estimation problem with
344 6 Advanced Detection Techniques and Compensation of Channel Impairments

Fig. 6.18 The Mth power loop The M-PSK signal from either
for phase recovery (in M-ary PSK
wireless coherent detector or M-th power- Low-pass
systems) law device filter
balanced coherent optical detector

Voltage-controlled
Oscillator (VCO) PLL

Frequency
divider (by M)

Phase-shift network

The M reference signals

respect to statistical parameter. For instance, maximum likelihood (ML) estimation can be straightforwardly applied here. We
can classify the various synchronization methods as data-aided and nondata-aided ones, depending on whether a preamble is
used for synchronization. In this section, we will describe nondata-aided synchronization methods. These methods can further
be classified as (i) conventional, phase-locked loop (PLL)-based approaches and (ii) algorithmic approaches.
As for M-ary PSK format, we can use a variant of the Costas loop of the Mth power loop, as illustrated in Fig. 6.18. This
method can easily be generalized and extended to star-QAM constellations. However, it suffers from phase ambiguities within
the [0,2π] interval, as well as the cyclic slips. Both of these problems can be solved by differential encoding, but at the expense
of SNR degradation.
The algorithmic approaches are based on specially developed DSP algorithms. The M-ary PSK signal after wireless
coherent detector or balanced coherent optical detection can be represented as
 
r ðt Þ ¼ A cos ð2π f IF t þ θ þ ϕk Þp t  t g þ zðt Þ; ϕk ¼ k2π=M ðk ¼ 0, ⋯, M  1Þ, ð6:201Þ

where A is an amplitude, θ is the random phase due to either laser phase noise in optical communications or multipath delay in
wireless communications, tg is the group delay, p(t) is the pulse shape used on the transmitter side, z(t) is the equivalent
Gaussian noise, while fIF denotes the intermediate frequency, which is the difference in carrier frequencies of received signal
and local oscillator. The basis functions to be used at the receiver side are given by
rffiffiffiffi
2
Φ1 ðt Þ ¼ cos ð2π f IF t Þ, t g  t  t g þ T
T
rffiffiffiffi ð6:202Þ
2
Φ2 ðt Þ ¼ sin ð2π f IF t Þ, t g  t  t g þ T:
T

The received signal can now be represented as a two-dimensional vector:

"  # ð tg þT
  r1 tg  
r tg ¼   , r 1,2 t g ¼ r ðt ÞΦ1,2 ðt Þdt: ð6:203Þ
r2 tg tg

In a similar fashion, the transmitted signal portion of r(tg) can be represented as

"  #
  s1 ak , θ, t g
s ak , θ, t g ¼   ,
s2 ak , θ, t g
ð tg þT ð6:204Þ
 
s1,2 ak , θ, t g ¼ A cos ð2π f IF t þ θ þ ϕk ÞΦ1,2 ðt Þdt
tg
6.4 Synchronization Techniques 345

where ak is a transmitted symbol, which is equal to exp( jϕk) for M-ary PSK. By choosing the frequencies of transmit laser and
local laser oscillator so that fIF is an integer multiple of symbol rate (1/T), Eq. (6.204) can be simplified to
" pffiffiffiffi #
  E cos ðθ þ ϕk Þ
s ak , θ, t g
pffiffiffiffi : ð6:205Þ
 E sin ðθ þ ϕk Þ

The noise signal z(t) can also be represented in a vector form as


 ð tg þT
z1
z¼ , z1,2 ¼ zðt ÞΦ1,2 ðt Þdt ð6:206Þ
z2 tg

so that the equivalent channel model in a vector form is given by

   
rk t g = s ak , θ, t g þ z: ð6:207Þ

In fiber-optics communications, when the chromatic dispersion, PMD, and PDL effects are properly compensated for, the
resulting noise process is Gaussian-like with the power spectral density of N0 so that the corresponding conditional probability
density function is Gaussian. Similarly, after compensation of multipath fading effects in wireless communications, the
resulting distribution is Gaussian as well:

  1 krk ðtg Þsðak ,θ,tg Þk2

f R rjak , θ, t g ¼ e : ð6:208Þ
πN 0

For non-Gaussian channels, we will need to use the method of histograms to estimate the conditional probability density
function fR(r| ak, θ, tg). For convenience, we can use the likelihood function defined as
 
  f R rjak , θ, t g
¼ eN 0 k ð g Þ ð k g Þ N 0 k ð k g Þ k :
2 T 2
r t s a ,θ,t  1 s a ,θ,t
L ak , θ, t g ¼ ð6:209Þ
f R ðrjak ¼ 0Þ

In M-ary PSK, the term ks(ak, θ, tg)k2 is constant and can be ignored, so that the previous equation becomes
 
  2 k T   
L ak , θ, t g ¼ exp r t g s ak , θ, t g : ð6:210Þ
N0

If the sequence of L statistically independent symbols, a¼[a0 . . . aL-1]T, is transmitted, the corresponding likelihood
function will be
 
  L1 2 T   
L a, θ, t g ¼ ∏ exp r t s al , θ, t g : ð6:211Þ
l¼0 N0 l g

To avoid numerical overflow problems, the log-likelihood function should be used instead, so that we can write

    XL1
2 T   
l a, θ, t g ¼ log L a, θ, t g ¼ rl t g s al , θ, t g : ð6:212Þ
l¼0
N0

The first derivative of log-likelihood function with respect to θ is given by

∂lða, θ, t g Þ 2E 1=2 X
L1
ak r k ejθ g; r k ¼ r 1,k þ jr 2,k ,
^ ^
ffi Im fb
θ N 0 k¼0 ð6:213Þ
b k þ θÞ, r 2,k ¼ E1=2 sin ðϕ
r 1,k ¼ E 1=2 cos ðϕ b k þ θÞ,
346 6 Advanced Detection Techniques and Compensation of Channel Impairments

rn
Detector

Error
exp( ) estimation
e[n]
ˆ n

Loop-filter
ˆ n 1
z-1

Fig. 6.19 The recursive implementation of the Costas loop

where b b k is the estimate of ϕk. Now we can formulate an LMS-like algorithm that can be used to
ak is the estimate of ak, while ϕ
estimate θ. Namely, the error signal in the current iteration can be estimated as
 
 ^ jb θ½n
e½n ¼ Im bak r k e , ð6:214Þ

where b
θ½n is the previous estimate of a random phase, while an updated estimate can be obtained as

b
θ ½ n þ 1 ¼ b
θ½n þ μe½n, ð6:215Þ

where μ is the step-size parameter. This algorithm essentially implements the recursive Costas loop as depicted in Fig. 6.19,
where the loop filter is in fact the first-order filter. As a consequence, this loop can have a static error, which can be avoided if
the second-order filter is used as a loop filter.
For timing synchronization, we need to know the intermediate frequency fIF. The starting point in developing an algorithm
for timing synchronization is again the likelihood function given by Eq. (6.211). However, the likelihood function is
dependent on random phase θ so that it needs to be averaged since we know that distribution of the frequency offset. In
optical communications, the Gaussian distribution can be used to describe the laser phase noise, so that the average likelihood
function can be defined as
ð
    1
pffiffiffiffiffi eθ =2σ θ dθ
2 2
L ak , t g ¼ L ak , θ, t g
σ θ 2π
ð   ð6:216Þ
2 k T    1
pffiffiffiffiffi eθ =2σθ dθ,
2 2
¼ exp r t g s ak , θ, t g
N0 σ θ 2π

where the variance of the phase noise is given by σ 2θ ¼ 2πΔfT, with Δf being the frequency offset. For laser phase noise, we
can set Δf ¼ Δν, where Δν is the sum of linewidths of transmit laser and local laser oscillator. The integral above can be solved
numerically or by Monte Carlo integration. The further procedure is similar as for carrier synchronization since we have to
first create the likelihood function for the sequence of L statistically independent symbols, a¼[a0 . . . aL-1]T, given by

  L1  
L a, t g ¼ ∏ L ak , t g : ð6:217Þ
l¼0

   
The next step is to create the log-likelihood function by l a, t g ¼ log L a, t g : Finally, we can properly choose the error
signal and create an LMS-like algorithm
  for timing synchronization from the derivative of log-likelihood function with respect
to group delay, namely, from ∂l a, t g =∂t g :
6.5 Adaptive Modulation Techniques 347

Transmit
antenna Receive
h[i] antenna
Power
amplifier LNA
Sequence y[i] Est.
Mapper + Power I/Q
Encoder Decoder
I/Q modulator control, P[i] x[i]
demodulator seq.
y[i]
Channel
ĝ i
2
ĝ i hˆ i estimator
Feedback
channel delay

Fig. 6.20 Adaptive wireless communication system model

6.5 Adaptive Modulation Techniques

Adaptive modulation techniques [5, 11, 14, 51–56] enable spectrally efficient and reliable transmission over time-varying
wireless and optical channels. As illustrated in Fig. 6.20, the basic premise is to estimate the channel state information (CSI) at
the receiver side and feedback this CSI estimate to the transmitter, which can be adapted relative to the time-varying channel
conditions. The nonadaptive modulation techniques are designed for the worst-case scenario and require a fixed system
margin to maintain acceptable performance when the channel quality is poor.
There are many parameters that can be varied at the transmitter side depending on the channel SNR. The most common
parameters to adapt include data rate, launch power, error correction strength (code rate), error probability, as well as
combinations of various adaptive techniques. In this section, we describe the adaptive modulation techniques in the context
of wireless communication systems. However, these concepts can easily be modified for optical communication applications
[53–56]. For this purpose, for discrete-time interval i, the SNR can be defined as ρ[i] ¼ Pg[i]/N0BW, where P is the average
power, g[i] ¼ |h[i]|2 is the channel gain, N0 is the additive noise power spectral density, and BW is the channel bandwidth. The
data rate and spectral efficiency of M-ary modulation scheme are defined, respectively, as R[i] ¼ log2 M[i]/T ¼ BW log2 M [i]
[bits/s] and R[i]/BW ¼ log2M[i] [bits/s/Hz], where we assumed that the channel bandwidth is equal to the symbol rate, namely,
BW ¼ 1/T.
In variable-rate modulation technique, the data rate R[ρ] is varied relative to the channel SNR ρ, which can be done by
(1) fixing the symbol rate to Rs ¼ 1/T and vary signal constellation sizes for the same modulation format or employing
multiple modulation formats and (2) fixing the modulation format while the symbol rate is varied. Symbol rate variation is
difficult to implement in wireless communications because of the bandwidth sharing problem. However, this is typically not a
problem in optical communication systems.
The variable-power modulation technique is used to compensate for SNR variation due to fading effects, to enable
constant SNR at the receiver side, which is equivalent to the constant receiver side bit error probability. Therefore, with
variable-power adaptation technique, we effectively invert the channel fading effects so that the channel appears as an AWGN
channel at the receiver side, as described in Chap. 4. The power adaptation policy for channel inversion is given by

PðρÞ ρ0
¼ , ð6:218Þ
P ρ

wherein the constant ρ0 is determined from the following condition:

ð ð
P ð ρÞ ρ
f ðρÞdρ ¼ 0 f ðρÞdρ ¼ 1, ð6:219Þ
P ρ

with f(ρ) being the PDF of SNR, as follows:

ρ0 ¼ 1=h1=ρi: ð6:220Þ
348 6 Advanced Detection Techniques and Compensation of Channel Impairments

For Rayleigh fading, when h1/ρi ! 1, the target bit error probability cannot be achieved by the channel inversion
technique. To solve for this problem, we can invert the fading effects only when SNR is larger than a given threshold (cutoff)
SNR value ρtsh, and corresponding variable-power adaptation techniques are known as the truncated channel inversion policy,
which can be formulated as

PðρÞ ρ0
 ¼ ρ , ρ > ρtsh
P
ð6:221Þ

wherein the parameter ρ0 is determined from

ð
1
1
ρ0 ¼ 1=h1=ρiρtsh , h1=ρiρtsh ¼ f ðρÞdρ: ð6:222Þ
ρ
ρtsh

In variable-error probability technique,

Ð we vary the signal constellation size and/or modulation format so that the average
bit error probability, defined as Pb ¼ Pb ðρÞf ðρÞdρ, is smaller than target Pb.
In adaptive coding, a stronger channel code is used when channel conditions (or SNR) are (is) bad, while a weaker channel
code is used when channel conditions are favorable (or SNR is high). Various adaptive coding approaches include (i) code
multiplexing, in which different channel codes are used for different SNR ranges; (ii) rate-compatible punctured convolutional
(RCPC) codes or rate-compatible punctured LDPC codes, in which error correction strength is reduced by puncturing a certain
number of parity bits; (iii) shortening approach, in which the smaller number of information bits is transmitted for the same
number of parity bits to improve error correction strength; (iv) re-encoding approach, in which a certain number of bits/
symbols that form the previous codeword are re-encoded with incoming information bits to enable almost a continuous tuning
of error correction strength (code rate) [11]; and (v) fountain codes [57], in particular Raptor codes [58], in which error
correction strength is improved by increasing the codeword length.
In hybrid adaptive modulation techniques, we adapt multiple parameters of the wireless/optical communication transmis-
sion system including code rate, power, data rate, and bit error probability. To satisfy a given performance metric, the joint
optimization of the different modulation techniques is employed. For instance, the data rate adaptation and power adaptation
can be simultaneously performed to maximize the spectral efficiency. In adaptive modulation and coding (AMC), the
modulation format/constellation size is combined with code rate adaptation.
In the incoming sections, we describe variable-power variable-rate modulation techniques and adaptive coded modulation
with more details.

6.5.1 Variable-Power Variable-Rate Modulation Techniques

In variable-power variable-rate modulation technique, we simultaneously adapt the transmitted power and data rate. For
illustrative purposes, we study a family of M-ary QAM signal constellations, but the described concepts are also applicable to
any other M-ary (M – the number of points in corresponding signal constellation) modulation scheme. We assume that the
symbol duration T is related to the channel bandwidth BW by T ¼ 1/BW based on the ideal Nyquist pulse shaping. Let P, N0,
and ρ ¼ Pg/N0BW denote the average power, the PSD of AWN, and the average SNR, respectively. The average symbol SNR
is given by the following equation:

Es PT
¼ ¼ ρ: ð6:223Þ
N0 N0

We use the following bit error probability bound, which is accurate within 1 dB penalty for M  4 [5]:

Pb  0:2 exp ½1:5ρ=ðM  1Þ, 0  ρ  30 dB, M  4: ð6:224Þ

6.5 Adaptive Modulation Techniques 349

Let us now adapt the transmit power P(ρ) depending on symbol SNR ρ, subject to the average power constraint P and an
instantaneous probability of error constraint Pb(ρ) ¼ Pb. Given that transmitted power is adapted, the received SNR will be ρ
P(ρ)/P, while the corresponding Pb bound can be written as
 
1:5ρ PðρÞ
Pb  0:2 exp , 0  ρ  30 dB, M  4: ð6:225Þ
M ð ρÞ  1 P

By solving Eq. (6.225) for M(ρ), we obtain

1:5 PðρÞ PðρÞ

MðρÞ ¼ 1 þ ρ  ¼ 1 þ Cfρ  ,
log ð5Pb Þ P P
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} ð6:226Þ
Cf

0  ρ  30 dB, M  4,

where Cf ¼ 1.5/(-log(5Pb)) is the correction factor, which is used to account for the fact that non-optimized signal constellation
is used. The spectral efficiency is maximized by optimizing the hlog2Mi [bits/symbol], subject to (s.t.) the corresponding
power constraint as follows:

ð
þ1  
PðρÞ
hlog2 Mi ¼ max log2 1 þ C f ρ  f ðρÞdρ,
P
0
ð6:227Þ
ð
þ1

s:t: 
PðρÞf ðρÞdρ ¼ P:
0

Based on Chap. 4, as expected, the optimum power adaptation policy is the water-filling method:
(
PðρÞ 1=ρtsh  1=ρC f , ρ  ρtsh =C f
¼ ,
P 0, ρ < ρtsh =C f
( ð6:228Þ
C f P ð ρÞ 1=ρC f  1=ρ, ρ  ρC f
¼ , ρC f ¼ ρtsh =C f :
P 0, ρ < ρC f

After substituting the optimum power allocation into power constraint, we obtain the following equation:

ð 
þ1 
1 1
 f ðρÞdρ ¼ C f , ð6:229Þ
ρC f ρ
ρC f

which can be used to determine ρCf. On the other hand, after substituting Eq. (6.228) into Eq. (6.227), we obtain the following
expression for instantaneous data rate [bits/symbol]:
 
ρ
log 2 M ðρÞ ¼ log 2 : ð6:230Þ
ρC f

The corresponding expression for spectral efficiency is obtained by averaging out the instantaneous data rate as follows:

   ð
1  
R ρ ρ
¼ h log 2 M ðρÞi ¼ log 2 ¼ log 2 f ðρÞdρ: ð6:231Þ
BW ρC f ρC f
ρC f
350 6 Advanced Detection Techniques and Compensation of Channel Impairments

Fig. 6.21 Average spectral

efficiencies for adaptive M-QAM
in Rayleigh fading and Nakagami-
2 fading

As an illustration, in Fig. 6.21, we provide the spectral efficiencies vs. average SNR for Rayleigh fading and Nakagami-
2 fading for adaptive M-ary QAM. Because the M-ary QAM is not capacity modulation format, there is the gap with respect to
Shannon capacity, related to the constant Cf.
The optimization given by Eq. (6.229) can be computationally extensive. To solve for this problem, we can apply channel
inversion-based power adaptation. With channel inversion, the receiver will experience the constant SNR, and the transmitter
can employ a single fixed data rate M-QAM modulation archiving the target Pb. The constellation size M that satisfies the
target Pb can be obtained by substituting the channel inversion power adaptation given P(ρ)/P ¼ ρ0 / ρ, ρ0 ¼ 1/h1/ ρi. The
corresponding spectral efficiency for the channel inversion-based power adaptation is given as follows:
" # " #
R PðρÞ 1
¼ log2 M ¼ log2 1 þ C f  ¼ log2 1 þ C f ,
BW P h1=ρi
8
> 1:5ρ ð6:232Þ
< log ð5Pb Þ , M  4
>
Cf ¼ :
>
> 1:5
: ,M < 4
log ð0:5Pb Þ

To avoid the problem associated with small SNRs tending to zero, the truncated channel inversion-based power
adaptation can be used in which we do not transmit when SNR falls below the threshold ρtsh, which is determined to
maximize the spectral efficiency below:
" #
R 1
¼ max log 2 1 þ C f Pðρ > ρtsh Þ: ð6:233Þ
BW ρtsh h1=ρiρtsh

As an illustration, in Fig. 6.22, we provided the average spectral efficiencies for various power adaptation strategies related
to M-QAM in Rayleigh fading. Clearly, the truncated channel inversion, although not optimal, performs comparable to water-
filling method in low- and medium-SNR regimes. On the other hand, the channel inversion strategy faces significant
performance degradation in the spectral efficiency sense. The corresponding spectral efficiency plots related to Nakagami-
2 fading are provided in Fig. 6.23.
6.5 Adaptive Modulation Techniques 351

Fig. 6.22 Average spectral

efficiencies for various power
adaptation strategies related to
M-QAM in Rayleigh fading

Fig. 6.23 Average spectral

efficiencies for various power
adaptation strategies related to
M-QAM in Nakagami-2 fading
352 6 Advanced Detection Techniques and Compensation of Channel Impairments

To further simplify the optimum and suboptimum adaptation policies, the discrete-rate adaptation should be used, which
restricts the adaptive modulations to a limited set of signal constellations. For instance, for square-QAM, we can restrict the
signal constellation sizes to M0 ¼ 0, M1 ¼ 2, and Mm ¼ 22(m1), m ¼ 2, . . ., N  1 (for some N > 2). To determine which signal
constellation to use to for a given SNR ρ, we have to discretize the range of SNRs due to fading effects into N SNR regions
Δρm ¼ [ρm-1, ρm), m ¼ 0, . . ., N  1, wherein we set ρ 1 ¼ 0 and ρN1 ! 1. Whenever the signal SNR falls into region Rm,
we transmit the signal with signal constellation size Mm, and the corresponding spectral efficiency in this case is log2Mm [bits/
s/Hz] (for m > 0). The key problem for discrete-rate adaptation is the SNR boundaries of the Δρm regions. Even with numerical
optimization, this process could be computationally extensive. To solve for this problem, we can use a suboptimum technique
in which we express the signal constellation sizes M(ρ) ¼ ρ/ρ*, where ρ* > 0 is a single SNR parameter to be determined by
optimizing the spectral efficiency. For an instantaneous ρ, we transmit the largest signal constellation from the set of possible
constellations {Mm | m ¼0, . . ., N} that is smaller than M(ρ), namely, Mm  M(ρ) < Mm+1. The corresponding SNR boundaries
are defined by {ρm¼ ρ* Mm+1 | m ¼ 0, . . ., N-2}. The optimum power adaptation policy is again water filling, but with slightly
modified expression (6.228):
8 
M ðρÞ 1 
C f PðρÞ <  , M m < ρ=ρ  M mþ1 ðM m  1Þð1=ρÞ, M m < ρ=ρ  M mþ1
¼ ρ ρ ¼ : ð6:234Þ
P : 0, otherwise
0, otherwise

After substituting Eq. (6.234) into Eq. (6.227), we obtain the following expression for the spectral efficiency for discrete-
rate adaptation:

X
N 1  
R ρ
¼ log 2 ðM m ÞP M m   < M mþ1 : ð6:235Þ
BW m¼1 ρ

The parameter ρ* is determined from the power constraint condition:

ρ M
ðmþ1
X
N1
P m ð ρÞ
f ðρÞdρ ¼ 1: ð6:236Þ
m¼1 P
ρ M m

The discrete-rate policy can be further simplified by employing the constant power for each SNR region, when signal
constellation size Mm is used; and the corresponding adaptation policy can be called discrete-power discrete-rate adaptive
modulation technique.
The optimization given by Eq. (6.236) can be still time-consuming. To reduce the complexity further, we can use the
channel inversion-based discrete-rate adaptation for a fixed set of signal constellation sizes {Mm | m ¼ 0, 1, . . ., N-1}.
However, the corresponding spectral efficiency, given below, is reduced:
 
R 1
¼ log 2 1 þ C f , ð6:237Þ
BW h1=ρi fM m jm¼0,1,⋯,N1g

where we used the notation bcMm to denote the largest signal constellation in the set {Mm} smaller than or equal to the
enclosed quantity. The corresponding truncated channel inversion-based discrete-rate adaptation policy will have the
following spectral efficiency:
$ %
R 1
¼ max log 2 1 þ C f Pðρ > ρtsh Þ: ð6:238Þ
BW ρtsh h1=ρiρtsh
fM m jm¼0,1,⋯,N1g

The choice of the number of SNR regions is highly dependent on hardware constraints and how fast the fading changes,
dictating how many signal constellations can be supported on the transmitter side and at what rate to adapt modulation format
and launch power. In particular, by specifying how long SNR is going to stay in a given SNR region gives us an information at
6.5 Adaptive Modulation Techniques 353

which rate to adapt and how many SNR regions to use. This information is provided by the average fade duration concept
introduced in Chap. 2. We use Tm to denote the average fade region duration that SNR ρ stays within the mth SNR region. The
finite-state Markov model, also described in Chap. 2, provides a reasonable complexity model as it assumes that the SNR ρ
remains within one SNR region for the duration of symbol period and that from the given SNR region the transition is possible
only to neighboring regions. Under these assumptions, the transition probabilities between the regions are given by [60]

Lmþ1 T L T
pm,mþ1 ¼ , pm,m1 ¼ m , pm,m ¼ 1  pm,mþ1  pm,m1 , ð6:239Þ
πm πm

where π m is the probability of being in the mth SNR region, namely, π m ¼ P(Δρm  ρ  Δρm + 1). With Lm we denoted the mth
SNR level (Δρm) crossing rate, which is for Rician fading given by [13]

 1=2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ρ
Lm ¼ 2πðK þ 1ÞνD m

ρ
   rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð6:240Þ
ρ ρ
exp K  ðK þ 1Þ m I 0 2 KðK þ 1Þ m ,

ρ 
ρ

where K is the Rician parameter and νD is the Doppler frequency. For Rayleigh fading, by setting K ¼ 2, the previous equation
simplifies to

 1=2  
pffiffiffiffiffi ρm ρm
LðmRayleighÞ ¼ 2π νD exp  : ð6:241Þ
ρ ρ

The average fade duration for the mth SNR region can be determined as

T πm
Tm ¼ ¼ : ð6:242Þ
pm,mþ1 þ pm,m1 Lmþ1 þ Lm

Clearly, Tm is reversely proportional to the Doppler frequency.

6.5.2 Adaptive Coded Modulation

When compared to adaptive modulation alone, the additional coding gain can be achieved by superimposing either trellis
codes or coset codes on top of the adaptive modulation scheme. The key idea behind this adaptive coded modulation is to
exploit the separability of coding stage and constellation design inherent to the coset codes [61, 62], which are described in
Chap. 9. The instantaneous SNR varies with time, which will change the distance in the received signal constellation d0(t),
and, therefore, the corresponding minimum distance between coset sequences ds and the minimum distance between coset
points dc will be time varying as well. By using adaptive modulation in tandem with the coset codes, we can ensure that
distances ds and dc do not vary as channel conditions change by adapting the signal constellation size M(ρ), transmit power
P(ρ), and/or symbol rate Rs(ρ) of the transmitted signal constellation relative to the SNR, subject to an average transmit power
P constraint on SNR-dependent transmit power P(ρ). As minimum distance does not vary as channel conditions change
dmin(t) ¼ min{ds(t), dc(t)}¼dmin, the adaptive coded modulation will exhibit the same coding gain as coded modulation
developed for a Gaussian-like channel with corresponding minimum codeword distance dmin. Therefore, this adaptive coded
modulation scheme requires small changes in the generic coded modulation scheme described above, as shown in Fig. 6.24.
Clearly, the coding stage does not change at all. The number of cosets does not change as well. On the other hand, the number
of points within the coset is channel SNR dependent. When channel condition get improved, we enlarge the coset size. In such
a way, we transmit more uncoded bits while not affecting the distance of the channel code. Additionally, this approach is
hardware friendly.
354 6 Advanced Detection Techniques and Compensation of Channel Impairments

Fig. 6.24 Generic adaptive One point

coded modulation scheme One point selected out of
Additional selected out of 2[n( )+r] / (N/2)
n( )-k info 2n( )+r possible possible Adaptive two- To
bits Signal point Constellation dimensional channel
selection mapper modulator
from coset M( ), P( )
One coset selected Modulation stage
out of 2k+r possible

k+r coded Binary

bits encoder k info bits
Coset
selector (block or
convolutional)
Coding stage

The following approximation for bit error probability for M-ary QAM systems can be used:
 
1:5ρ  Gc PðρÞ
Pb  0:2 exp , M  4, 0  ρ ½dB  30 dB, ð6:243Þ
M ð ρÞ  1 P

where Gc is the coding gain. We adjust the signal constellation size M(ρ) depending on channel SNR and target BER as
follows:

1:5 P ð ρÞ P ð ρÞ
M ðρÞ ¼ 1 þ ρ  Gc ¼ 1 þ C f ρ  Gc , ð6:244Þ
 log ð5Pb Þ P P
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}
Cf

where Cf is again the correction factor, defined as Cf ¼  1.5/ log (5Pb), which is used to account for the fact that
non-optimized signal constellation is used. The number of uncoded bits needed to select a point within the coset is determined
by

nðρÞ  2k=N ¼ log 2 M ðρÞ  2ðk þ r Þ=N: ð6:245Þ

The normalized transmission rate is given by log2M(ρ), while the normalized adaptive data rate is determined by log2M(ρ)-
2r/N. To determine the optimum launch power, we need to optimize hlog2M(ρ)i, subject to the average power constraint, and
not surprisingly the optimum power adaptation strategy is the water-filling law:
8
< 1=ρ0  1=ρc , ρ  |fflffl{zfflffl}
ρ0 =Cc
P ð ρÞ
¼ ρc , C c ¼ C f Gc , ð6:246Þ
P :
0, ρ < ρ0 =C c

whereρ0 is the threshold signal-to-noise ratio for an AWGN channel. After substituting Eq. (6.246) into Eq. (6.244), we obtain
the following equation for adaptive signal constellation size:

M ðρÞ ¼ ρ=ρc : ð6:247Þ

The corresponding spectral efficiency is given by

ð
1  
R ρ
¼ h log 2 M ðρÞi ¼ log 2 f ðρÞdρ: ð6:248Þ
BW ρc
ρc
6.6 Volterra Series-Based Equalization 355

6.6 Volterra Series-Based Equalization

One important application of the Volterra series is in compensation of nonlinear effects in fiber-optics communications [63–
70] or to deal with nonlinearities on satellite channels. The nonlinear compensation can be performed either in time domain
[63, 70] or frequency domain [67, 91]. The frequency domain is particularly attractive, as the theory behind it is well
developed [67, 68]. Expansion in the Volterra series is given as [64]

X
1
yð t Þ ¼ yn ðt Þ,
n¼0
ð
1 ð
1 ð6:249Þ
yn ð t Þ ¼ ⋯ hn ðτ1 , τ2 , ⋯τn Þxðt  τ1 Þxðt  τ2 Þ⋯xðt  τn Þdτ1 dτ2 ⋯dτn ,
1 1

where x(t) is the input to the nonlinear system, while the nth term yn(t) represents the n-fold convolution of the input and the n-
fold impulse response hn(τ1, ⋯τn). The set {h0, h1(τ1), h2(τ1, τ2), ⋯, hn(τ1, τ2, ⋯τn), ⋯} of impulse responses is known as
the Volterra kernel of the system. The 0th order corresponds to d.c. component, which is not represented here, while the first-
order term corresponds to impulse response of the linear system. The first-order impulse response can be expressed in
analytical form, where the second-order GVD can be neglected. The frequency domain representation of the Volterra series is
given by [67]

1 ð ð
1 1
X
Y ð ωÞ ¼ ⋯ H n ðω1 , ⋯ωn ÞX ðω1 Þ⋯X ðωn ÞX ðω  ω1  ⋯ωn Þdω1 ⋯dωn : ð6:250Þ
n¼1
1 1

As an example, if H3 denotes the three-dimensional Fourier transform of kernel h3, it can be estimated as [64]

Sxxxy ð f , g, hÞ
H 3 ð f , g, hÞ ¼ ,
6Sxx ð f ÞSxx ðgÞSxx ðhÞ ð6:251Þ
n o  
1
Sxx ð f Þ ¼ E jX i ð f Þj2 , Sxxxy ð f , g, hÞ ¼ E X i ð f ÞX i ðgÞX i ðhÞY i ð f þ g þ hÞ ,
T

where X( f ) and Y( f ) are the Fourier transforms of x(t) and y(t), T is the observation interval, while the subscript i denotes
the index of record of duration T.
The propagation over the SMF is governed by the nonlinear Schrödinger equation (NLSE), as described in Chap. 2. Since
we are concerned here with compensation of chromatic dispersion and nonlinearities, let us observe the following inverse
version of NLSE that can be obtained from original NLSE by changing the sign of the terms:

∂E ðz, t Þ  ^ ^ 
¼ D  N Eðz, tÞ,
∂z
2 3
ð6:252Þ
^ ¼  α  j β2 ∂ þ β3 ∂ , hatN ¼ jγjE2 j,
D
2 2 ∂t 2 6 ∂t 3

where E is the signal electric field, Db and Nb denote the linear and nonlinear operators, while α, β2, β3, and γ represent the
attenuation coefficient, GVD, second-order GVD, and nonlinear coefficient, respectively. By setting that X(ω) ¼ E(ω,z) ¼ FT
{E(z,t)}, the electric field from Eq. (2.252) can be presented as [69]
ðð
Eðω, 0Þ
H 1 ðω, zÞEðω, zÞ þ H 3 ðω1 , ω2 , ω  ω1 þ ω2 , zÞEðω1 , zÞE  ðω2 , zÞEðω  ω1 þ ω2 , zÞdω1 ω2 , ð6:253Þ

where the linear kernel in the frequency domain is given as

356 6 Advanced Detection Techniques and Compensation of Channel Impairments

H 1 ðω, zÞ ¼ exp αz=2  jβ2 ω2 z=2 , ð6:254Þ

while the second-order GVD is neglected. The third-order kernel in the frequency domain is given by [69]

1  exp ½αz  jβ2 ðω1  ωÞðω1  ω2 Þz

H 3 ðω1 , ω2 , ω  ω1 þ ω2 , zÞ ¼ jγH 1 ðω, zÞ : ð6:255Þ
α þ jβ2 ðω1  ωÞðω1  ω2 Þ

As we see in Eq. (6.255), fiber parameters of opposite sign are used in the design of the nonlinear equalizer based on the
Volterra series. For ultra-long-haul transmission systems’ studies, the second-order GVD must be included. Also, the fifth-
order kernel term should be calculated as well. Regarding the complexity of this method, the following rule of thumb can be
used. The number of operations related for the nth kernel is approximately equal to the nth power of the number of operations
needed for the first-order kernel.

6.7 Digital Backpropagation in Fiber-Optics Communications

Another method to simultaneously compensate for chromatic dispersion and fiber nonlinearities is known as the digital
backpropagation compensation [72, 73]. The key idea of this method is to assume that received signal in the digital domain
can now propagate backward through the same fiber. This virtual backpropagation is done by using the fiber with parameters
just opposite in sign to real fiber parameters (applied in forward propagation). By this approach, we will be able in principle to
compensate for fiber nonlinearities if there was not signal-to-noise nonlinear interaction. This virtual backpropagation
(BP) can be performed either on transmitter side or receiver side, as illustrated in Fig. 6.25. In the absence of noise, those
two approaches are equivalent to each other.
The backpropagation is governed by inverse NLSE, given by Eq. (6.252). The backpropagation method operates on the
signal electric field, and it is, therefore, universal and independent on modulation format. It uses a split-step Fourier method of
reasonable high complexity to solve it. Both asymmetric and symmetric split-step Fourier methods [74] can be used. As an
illustration, we briefly describe the iterative symmetric split-step Fourier method. Additional details can be found in Chap. 2.
The key idea of this method is to apply the linear and nonlinear operators in an iterative fashion.
The linear operator, which corresponds to multiplication in the frequency domain, is expressed as

Receiver side
back-propagation
SMF

Optical Optical
Tx Rx
, 2 , 3 , , 2, 3,

ˆ
D N ˆ

(a)
Transmitter side
back-propagation
SMF

Optical Optical
Tx Rx
, 2, 3, , 2 , 3 ,
ˆ
D N ˆ

(b)

Fig. 6.25 Digital backpropagation (BP) method illustration: (a) receiver side BP and (b) transmitter side BP
6.7 Digital Backpropagation in Fiber-Optics Communications 357

h i      
b 1 α β2 2 β3 3
exp ΔzD Eðz þ Δz, t Þ ¼ FT exp   þ j ω þ j ω Δz FT ½E ðz þ Δz, t Þ , ð6:256Þ
2 2 6

where Δz is the step size, while FT (FT-1) denotes the Fourier transform (inverse Fourier transform). On the other hand, the
nonlinear operator, which performs the nonlinear phase “rotation” in the time domain, is expressed as

^
exp ðΔzξNÞEðz þ Δz, tÞ ¼ exp jξγΔzjEðz þ Δz, tÞj2 Eðz þ Δz, tÞ, ð6:257Þ

where 0  ξ  1 is the correction factor, which is needed to account for ASE noise-signal nonlinear interaction during
propagation.
It is evident that the nonlinear operator depends on the magnitude of the electric field at location z, which should be
determined. The electric field can be found by using the trapezoidal rule, as proposed in [74], so that we can write
0 1
ðz
^
Eðz, tÞ ¼ exp ðDΔz=2Þexp @ξ Nðz0 Þdz0Aexp ðDΔz=2ÞEðz
^ þ Δz, tÞ
zþΔz ð6:258Þ
 
^ þ ΔzÞ þ NðzÞ
Nðz ^
^

exp ðDΔz=2Þexp ξ ^
Δz exp ðDΔz=2ÞEðz þ Δz, tÞ:
2

b ðzÞ ¼ jγ jE ðz, t Þj2 is dependent on the output, Eq. (6.258) can be solved in an iterative fashion, as
Since the operator N
illustrated by the flowchart in Fig. 6.26.

E(z+ z, t)

Linear half-step operator

exp(-D z / 2)
Estimate
N(z)
Nonlinear step operator
exp{ - [N(z)+N(z+ z) ] z / 2 }

Linear half-step operator

exp(-D z / 2)

Eprev(z, t)

No
|E-Eprev| / |E|<e

Yes
E(z, t)

Fig. 6.26 The iterative symmetric split-step Fourier method to solve the inverse nonlinear Schrödinger equation
358 6 Advanced Detection Techniques and Compensation of Channel Impairments

The operating frequency range of interest in BP method must be smaller than the sampling rate, and therefore, the
oversampling is needed. Regarding the complexity issues, FFT complexity scales with the factor NlogN, while the filtering
and nonlinear phase rotation operations have complexity proportional to N.
As for the multichannel BP, there are two options that can be applied. The first one is to combine different WDM channels
together so that the electric field can be written as
X
jn2πΔf
Eðz, t Þ ¼ E ðz, t Þe
n n
, ð6:259Þ

where Δf is the channel spacing. This approach can be called the “total-field BP” [73]. The second approach is to consider
individual WDM separately and then solve the set of coupled inverse NLSE [74]:

∂Ei ðz, t Þ b 
b i E i ðz, t Þ,
¼ Di  ξN
∂z
! ð6:260Þ
α ∂ β ∂
2
β ∂
3 X 2
b i ¼   β1,i  j
D 2
þ 3 b ¼ jγ jE i j þ
,N 2 E j  :
2 ∂t 2 ∂t 2 6 ∂t 3 i
j6¼i

Although this method appears to be more extensive, the synchronization of all WDM channels is not required. Both
methods are suitable for point-to-point application. A different situation is in mesh networks, since some wavelength channels
might be dropped and added several times along the route. As a consequence, if the exact situation along the route is not
known at the receiver side, the application of multiband BP might even lead to system performance degradation.
The BP method can also be used in polarization-division multiplexed (PDM) systems. Since the PMD rotation matrix
changes in a random fashion from fiber section to fiber section, it is quite challenging to predict the individual rotation
matrices. As a consequence, it is expected that the BP method will not be that efficient in simultaneous compensation of
nonlinear effects and PMD. In addition, the complexity of BP method to address PMD is higher as now the electric field
scalars used in single-channel single-polarization BP should be replaced by Jones matrices. The inverse NSE for PDM
systems will be now

∂Eðz, t Þ b 
b Eðz, t Þ,
¼  D  ξN
∂z
ð6:261Þ
2 3   
Db ¼  α  β1 ∂  j β2 ∂ þ β3 ∂ , N
b ¼ jγ jEj2 I‐ 1 E{ YE Y ,
2 ∂t 2 ∂t 2 6 ∂t 3 3

where E ¼ [Ex Ey]T denotes the Jones vector of the electric field, α denotes the attenuation coefficients in x- and
y-polarizations, while β1, β2, and β3 denote the group-velocity, GVD, and second-order GVD parameters in x- and
y-polarizations, respectively. Finally, with Y we denoted the Pauli-Y matrix in quantum mechanics notation [75]. Namely,
the Pauli matrices are different in quantum mechanics than in representation used in papers and books related to optics
[73, 74]. In our case, the Pauli matrices are defined as
     
0 1 0 j 1 0
X¼ ,Y ¼ , Z¼ : ð6:262Þ
1 0 j 0 0 1

If there is no possibility to track the Jones vector of each section, then β1 should be set to zero. Otherwise, β1 is related to
the polarization rotation matrix R as follows:

δ
β1 ðθk , εk Þ ¼ Rðθk , εk Þ k Z R{ ðθk , εk Þ,
" 2 #
cos θk cos εk  j sin θk sin εk sin θk cos εk þ j cos θk sin εk ð6:263Þ
R ð θ k , εk Þ ¼ ,
sin θk cos εk þ j cos θk sin εk cos θk cos εk  j sin θk sin εk

where 2θk and 2εk are the azimuth and ellipticity angles of the kth fiber section and δk corresponds to the differential group
delay (DGD) of the section. The solution of Eq. (6.236) should be obtained by the split-step Fourier method
6.8 Optical Communication Systems with Coherent Optical Detection 359

h i
Eðz, t Þ ¼ exp Δz Db þ ξN
b Eðz þ Δz, t Þ, ð6:264Þ

where the complexity of the iteration process is higher than in the single-polarization case.

6.8 Optical Communication Systems with Coherent Optical Detection

The principles of coherent optical detection have been already introduced in Chap. 3. In particular, in Sect. 3.1.8, we described
balanced coherent optical detectors for 1-D and 2-D modulation schemes. The balanced coherent detector suitable for
homodyne, intradyne, and heterodyne detections based on two Y-junctions and two 3 dB couplers was described. In this
section, we describe other alternatives for coherent balanced detection, in particular 2  4 optical hybrid-based balanced
coherent optical detector, polarization-division demultiplexing scheme, homodyne coherent optical detection based on
phased-locked loop (PLL), phase diversity concepts, as well as dominant coherent optical detection noise sources.

6.8.1 Balanced Coherent Optical Detection for 2-D Modulation Schemes

Various alternatives suitable for balanced coherent optical detector implementation include (i) 2  4 optical hybrid, (ii) 3 dB
couplers and phase shifter, (iii) 3 dB coupler and polarization beam splitter (PBS), (iv) 3  3 coupler, (v) 4  4 multimode
interference (MMI) coupler, and (vi) Y-junction, 3 dB coupler, and phase shifter-based solutions, to mention a few. The
option (v) has already been described in Sect. 3.1.9. In Fig. 6.27, we provide the description of basic 2  2 optical hybrid, with
two phase trimmers, suitable for implementation of 2  4 optical hybrid. Let the input to the coherent detector be given by Sn
¼ |Sn|exp[j(ωTt+ϕPN,T+θn)], where the subscript n is used to denote the nth symbol being transmitted, ωT denotes the transmit
laser frequency (in rad/s), ϕPN,T denotes the laser phase noise of transmitter laser, and |Sn| and θn denote the amplitude and the
phase of the kth symbol for M-ary QAM/PSK (or any other 2-D signal constellation). Let the local laser (oscillator) output be
denoted by L ¼ |L|exp[j(ωLOLt+ϕPN,LOL)], where ωLOL denotes the local oscillator laser frequency and ϕPN,LOL denotes the
laser phase noise of local laser. The output port 1 electric field, denoted as Eo,1, is clearly related to the input port fields Sn and
L by
 jϕ1
pffiffiffiffiffiffiffiffiffiffiffi
E o,1 ¼ Sn e þL 1  κ, ð6:265Þ

where κ is the power splitting ratio and ϕ1 is the phase shift introduced by the control voltage 1 (phase trimmer 1). On the other
hand, the output port 2 electric field, denoted as Eo,2, is related to the input port fields Sn and L by

Fig. 6.27 The basic 2  2 optical Sn

hybrid with two phase trimmers

Control
voltage ( 1)

1 Phase
trimmer Eo,1

Eo,2

Control
voltage ( 2)

L
360 6 Advanced Detection Techniques and Compensation of Channel Impairments

Eo,1 i1
Sn 2 2 optical i=i1–i2
hybrid ( , 1, 2)
L i2
E0,2
Balanced
detector (BD)

Fig. 6.28 The coherent balanced optical detector based on 2  2 optical hybrid

 jϕ2
pffiffiffi
E o,2 ¼ Sn þ e L κ, ð6:266Þ

where ϕ2 is the phase shift introduced by the control voltage 2 (phase trimmer 2). The previous two equations can be
represented in the matrix form by
" # " pffiffiffiffiffiffi pffiffiffiffiffiffi
#" # " #
E o,1 e jϕ1 1κ 1κ Sn Sn
¼ pffiffi pffiffi ¼H ,
E o,2 κ e jϕ1 κ L L
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
H ð6:267Þ
" pffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffi #
jϕ1
e 1κ 1κ
H¼ pffiffiffi pffiffiffi ,
κ e jϕ1 κ

where H is the scattering matrix. For power splitting ratio κ ¼ 1/2, the scattering matrix simplifies to
 
1 e jϕ1 1
Hjκ = 1=2 = pffiffiffi jϕ2
: ð6:268Þ
2 1 e

Let us now observe the balanced coherent optical detector shown in Fig. 6.28, which is based on 2  2 optical hybrid
shown in Fig. 6.27, characterized by power splitting ration κ and phase shifts ϕ1 and ϕ2. Assuming that both photodetectors
have the same responsivity R and the power splitting ratio is κ ¼ 1/2, the resulting photocurrent i ¼ i1-i2 can be written as

i ¼ i1  i2 ¼ RjEo,1 j2  RjE o,2 j2

R  ð6:269Þ
¼ S e jϕ1 L þ Sn ejϕ1 L  Sn L ejϕ2  Sn Le jϕ2
,
2 n

where we ignored the photodiode shot noise to facilitate the explanations. Now by setting ϕ1 ¼ 0 and ϕ2 ¼ π, we obtain

R    
iðϕ1 ¼ 0, ϕ2 ¼ π Þ ¼ iI ¼ Sn L þ Sn L þ Sn L þ Sn L ¼ R Sn L þ Sn L
2 2 3
ð6:270Þ
¼ 2R Re fSn L g ¼ 2RjSn jjLj cos 4ðωT  ωLO Þt þ ϕ T,PN  ϕLO,PN þ θ n 5,
|fflfflfflfflfflfflfflfflfflffl
ffl{zfflfflfflfflfflfflfflfflfflfflffl}
ϕPN

which is clearly the in-phase component iI of the 2-D modulated signal. On the other hand, by ϕ1 ¼ ϕ2 ¼ π/2, we clearly
detect the quadrature component iQ of 2-D modulated signal:
 
iðϕ1 ¼ π=2, ϕ2 ¼ π=2Þ ¼ iQ ¼ R jSn L  jSn L
ð6:271Þ
¼ 2RImfSn L g ¼ 2RjSn jjLj sin ½ðωT  ωLO Þt þ ϕPN þ θn :
6.8 Optical Communication Systems with Coherent Optical Detection 361

2 4 optical hybrid
Eo,1 i1
2 2 optical hybrid iI=i1–i2
Y-junction
Sn ( =1/2, 1=0, 2= )
E0,2 i2
L i3
Eo,3
2 2 optical hybrid iQ=i3–i4
( =1/2, 1= /2, 2= /2)
i4
E0,4

Fig. 6.29 The coherent balanced optical detector for demodulation of 2-D signals based on 2  4 optical hybrid

3 dB directional couplers
based 2 4 optical hybrid
i1
Eo,1
Sn 3 dB /2
3 dB iI=i1–i2
coupler coupler
i2
E0,2
Eo,3 i3
L 3 dB 3 dB iQ=i3–i4
coupler coupler
E0,4 i4

Fig. 6.30 The coherent balanced optical detector for demodulation of 2-D signals based on 24 optical hybrid implemented with the help of four
 
1 j
3 dB directional couplers. The scattering matrix of 3 dB directional coupler is given by H 3 dB = p1ffiffi2
j 1

By combining these two balanced coherent detectors, we obtain the balanced coherent detector suitable of demodulation of
2-D modulated signals as shown in Fig. 6.29. The configuration of corresponding 2  4 optical hybrid is provided as well. To
take the presence of Y-junctions into account, the corresponding expressions for detected in-phase iI and quadrature iQ
components become, respectively,

1  
iI ¼ R Sn L þ Sn L ¼ RjSn jjLj cos ½ðωT  ωLO Þt þ ϕPN þ θn ,
2 ð6:272Þ
1  
iQ ¼ R jSn L  jSn L ¼ RjSn jjLj sin ½ðωT  ωLO Þt þ ϕPN þ θn :
2

This scheme is applicable to (i) homodyne detection by setting ωT ¼ ωLO; (ii) heterodyne detection by setting ωT¼ωLO+ωIF
wherein ωIF is the intermediate frequency larger than the symbol rate Rs, that is, ωIF>2πRs; and (iii) intradyne detection by
properly choosing the frequency difference as follows: |ωT–ωLO|2πRs.
The 3 dB directional coupler-based 24 optical hybrid and corresponding balanced coherent optical detectors are
provided in Fig. 6.30. By using the basic matrix multiplication, the output electric fields Eo,i are related to the received
optical field Sn and local oscillator laser output field L as follows:

1 1 1
Eo,1 ¼ ð jSn þ jLÞ, E o,2 ¼ ðSn þ LÞ, Eo,3 ¼ ð jSn  LÞ,
2 2 2 ð6:273Þ
1
E o,4 ¼ ðSn þ jLÞ:
2

The corresponding in-phase iI and quadrature iQ photocurrents of balanced detector branches are equal to
362 6 Advanced Detection Techniques and Compensation of Channel Impairments

1
iI ¼ RðSn L þ Sn LÞ ¼ RjSn jjLjcos ½ðωT  ωLO Þt þ ϕPN þ θn ,
2 ð6:274Þ
1
iQ ¼  Rð jSn L  jSn LÞ ¼ RjSn jjLjsin ½ðωT  ωLO Þt þ ϕPN þ θn ,
2

and are identical to the corresponding equations given by Eq. (6.272), except for the sign of quadrature component.
The 33 coupler, implemented in either integrated optics or fiber optics, with the following scattering matrix [76]
2 3
1 1 1
1 6 2π 7
pffiffi
ffi 41 e j3 ej 3 5,
2π
H 33 ¼ ð6:275Þ
3
1 ej 3 e j3
2π 2π

can also be used in coherent optical detection by using Sn and L as the first two inputs. The corresponding output port electric
fields are
2 3 2 32 3 2 3
Eo,1 1 1 1 Sn Sn þ L
6 7 1 6 j2π 2π 76 7 1 6 7
4 Eo,2 5 ¼ pffiffiffi 4 1 ej 3 54 L 5 ¼ pffiffiffi 4 Sn þ e j 3 L 5:
2π
e 3 ð6:276Þ
3 j2π 3
e j3 Sn þ ej 3 L
2π 2π
Eo,3 1 e 3 0

The corresponding photocurrents are given by


R R
io,1 ¼ RjEo,1 j2 ¼ jSn þ Lj2 ¼ jSn j2 þ jLj2 þ Sn L þ Sn L ,
3 3
 2 
R j2π  R
jSn j2 þ jLj2 þ Sn ej 3 L þ Sn e j 3 L ,
2π 2π
io,2 2
¼ RjEo,2 j ¼ Sn þ e 3 L ¼ ð6:277Þ
3 3
 2 
R j2π  R
jSn j2 þ jLj2 þ Sn e j 3 L þ Sn ej 3 L :
2 2π 2π
io,3 ¼ RjEo,3 j ¼ Sn þ e 3 L ¼
3 3

Now by subtracting the first two photocurrents, we obtain


R
Sn L þ Sn L  Sn ej 3 L  Sn e j 3 L
2π 2π
io,1  io,2 ¼
3
2R
¼ pffiffiffi jSn jjLj cos ½ðωT  ωLO Þt þ ϕPN þ θn  π=6 ð6:278Þ
3
2R :
¼ pffiffiffi jSn jjLj sin ½ðωT  ωLO Þt þ ϕPN þ θn þ π=6¼iQ ,
3

and this expression represents the quadrature component. On the other hand, the in-phase component can be obtained as
follows:

2R :
io,1 þ io,2  2io,3 ¼ pffiffiffi jSn jjLj cos ½ðωT  ωLO Þt þ ϕPN þ θn þ π=6¼iI : ð6:279Þ
3

The 3 dB directional coupler together with two PBS can also be used to implement 24 optical hybrid required in 2-D
modulation schemes, as illustrated in Fig. 6.31. We use |%i and |Ri to denote the +π/4 and right-circular polarization states,
defined, respectively, as
   
1 1 1 1 1 1
j%i ¼ pffiffiffi ¼ pffiffiffi ðjxi þ jyiÞ, jRi ¼ pffiffiffi ¼ pffiffiffi ðjxi þ jjyiÞ, ð6:280Þ
2 1 2 2 j 2

where |xi and |yi are the x- and y-polarization states.

Therefore, the inputs to the 3 dB directional coupler are given by
6.8 Optical Communication Systems with Coherent Optical Detection 363

Fig. 6.31 The 24 optical hybrid implemented with the help of one 3 dB directional coupler and two polarization beam splitters

S L
Ei,1 ¼ Sn j%i ¼ pnffiffiffi ðjxi þ jyiÞ, Ei,2 ¼ LjRi ¼ pffiffiffi ðjxi þ jjyiÞ: ð6:281Þ
2 2

The corresponding output port electric fields are given, respectively, by

1 1
Eo,1 ¼ ðSn þ jLÞjxi, Eo,2 ¼ ðSn  LÞjyi,
2 2 ð6:282Þ
1 1
Eo,3 ¼ ðjSn þ LÞjxi, Eo,4 ¼ ðjSn þ jLÞjyi:
2 2

The in-phase component can be obtained by placing balanced detector in front of output ports 2 and 4 to obtain

R
io,2  io,4 ¼ ðjS  Lj2  jSn þ Lj2 Þ
4 n
R ð6:283Þ
¼  ðSn L þ Sn LÞ
2
:
¼ RjSn jjLjcos ½ðωT  ωLO Þt þ ϕPN þ θn ¼iI :

On the other hand, the quadrature component can be obtained by placing balanced detector in front of output ports 1 and
3 to obtain

R
io,1  io,3 ¼ jSn þ jLj2  jjSn þ Lj2
4 ð6:284Þ
R  :
¼ jSn L þ jSn L ¼ RjSn jjLj sin ½ðωT  ωLO Þt þ ϕPN þ θn ¼iQ :
2

Regarding the implementation of the 24 optical hybrid with the help of 44 MMI, to [77, 78] an interested reader is
referred.

6.8.2 Polarization Diversity and Polarization-Division Demultiplexing

The coherent optical receivers require matching the state of polarization (SOP) of the local laser with that of the received
optical signal. In practice, only the SOP of local laser can be controlled; one possible configuration of the polarization control
receiver is shown in Fig. 6.32. Polarization controller from Fig. 6.32 is commonly implemented by using four squeezers as
described in [20]. Therefore, by polarization transformer, the SOP of local laser signal is changed until it matches the SOP of
received optical signal.
Receiver resilience with respect to polarization fluctuations can be improved if the receiver derives two demodulated
signals from two orthogonal polarizations of the received signal, as illustrated in Fig. 6.33. This scheme is known as a
polarization diversity receiver, and it has been derived from Fig. 6.31. In polarization diversity schemes, however, the same
data sequence is transmitted over both polarization states, which means that spectral efficiency is the same as in receivers
364 6 Advanced Detection Techniques and Compensation of Channel Impairments

Fig. 6.32 The configuration of SMF link

the polarization control receiver Data detected
Coherent optical Polarization Coherent
transmitter transformer
optical
Local oscillator Polarization receiver Controller
laser transformer

Fig. 6.33 The configuration of Photodetector

polarization diversity receiver. Decision
PBS, polarization beam splitter Received PBS
circuit
optical signal Photodetector
3 dB
directional
coupler

Local Used in balanced detection

oscillator
laser

iI(x) x-polarization
Detected sequence
2 4 state
(x-pol. state)
optical demodulator
SMF iQ(x) Compensation of
x hybrid
chromatic
PBS
dispersion,
y iI(y) y-polarization
PMD, and Detected sequence
x 2 4 state
PDL effects (y-pol. state)
Local PBS optical demodulator
laser iQ(y)
y hybrid

Fig. 6.34 The polarization-division demultiplexing-based receiver configuration

without polarization diversity. To double the spectral efficiency of polarization diversity schemes, the polarization-division
multiplexing can be applied [22, 23].
In polarization-division multiplexing (PDM) [22, 23] both polarizations carry independent multilevel modulated streams,
as described in Chap. 5, Sect. 5.5 , which doubles the spectral efficiency. As indicated in Chap. 5, the M-ary PSK, M-ary
QAM, and M-ary 2-D modulation schemes, in general, achieve the transmission of log2M ¼ m bits per symbol. In M-ary PSK,
the data phasor θn 2 {0,2π/M, . . ., 2π(M-1)/M} is sent at each nth transmission interval. In M-ary QAM, the mapper is
typically used that takes m bits at time instance n to determine the corresponding M-ary (M ¼ 2m) constellation point sn ¼ (In,
Qn) ¼ |sn|exp(jθn), and corresponding coordinates after DAC are used in RF inputs of electro-optical I/Q modulator. In PDM,
two I/Q modulators (with configuration provided in Chap. 3) and two 2-D mappers are needed, one for each polarization. The
outputs of I/Q modulators are combined using the polarization beam combiner (PBC). The same transmit DFB laser is used as
CW source, with x- and y-polarizations being separated by a polarization beam splitter (PBS).
The corresponding coherent optical detector receiver architecture is shown in Fig. 6.34, while the balanced coherent optical
detector architecture is described already above. The balanced output photocurrents of I- and Q-channel branches for
x-polarization at the time instance n can be written as
6.8 Optical Communication Systems with Coherent Optical Detection 365

0 1
B C
iI,n ’ RjSðnxÞ jjLðxÞ jcos B
ðxÞ
ϕT,PN  ϕLO,PN C
ðxÞ ðxÞ ð xÞ
@ωIF t þ θn þ |fflfflfflfflfflfflfflfflfflffl
ffl{zfflfflfflfflfflfflfflfflfflfflffl}A
,
ϕPN
ðxÞ
ð6:285Þ
ðxÞ ð xÞ
iQ,n ’ RjSðnxÞ jjLðxÞ jsin ðωIF t þ θðnxÞ þ ϕPN Þ,
ðxÞ ðxÞ ðxÞ
ϕPN ¼ ϕT,PN  ϕLO,PN ,

where R is the photodiode responsivity, while ϕT,PN and ϕLO,PN represent the laser phase noise of transmitting and local laser
oscillator, respectively. ωIF ¼ |ωT  ωLO| denotes the intermediate frequency, while Sn(x) and L(x) represent the incoming
signal (at the nth symbol interval) in x-polarization and x-polarization output from local laser oscillator, respectively. Similar
expressions hold for y-polarization state:

   
iI,n ’ RSðnyÞ LðyÞ  cos ωIF t þ θðnyÞ þ ϕPN ,
ðyÞ ðyÞ

    ð6:286Þ
iQ,n ’ RSðnyÞ LðyÞ  sin ωIF t þ θðnyÞ þ ϕPN :
ðyÞ ðyÞ

The heterodyne receiver can also be implemented based only on single balanced detector as described in Chap. 3. For ASE
noise-dominated scenario, both heterodyne schemes perform comparably. However, in shot noise-dominated scenario, the
heterodyne design with single balanced detector performs better. In homodyne detection, it is set that ωIF ¼ 0. After the
in-phase and quadrature signals are detected in both polarization states, we perform ADC and compensate for various
dispersion effects including chromatic dispersion, polarization mode dispersion (PMD), and polarization-dependent losses
(PDL). After timing recovery and phase estimation steps in each polarization state, the demodulation procedure takes place.

6.8.3 Homodyne Coherent Optical Detection Based on PLLs

Homodyne coherent optical detection receivers based on PLL can be classified as either residual carrier receivers or
suppressed carrier receivers [14]. Corresponding PLLs used in optical communications are commonly referred to as optical
PLLs (OPLLs). In systems with residual carrier, shown in Fig. 6.35, the phase deviation between the mark- and space-state
bits in BPSK is less then π/2 rad, so that the part of transmitted signal power is used for the non-modulated carrier transmission
and as consequence some power penalty occurs. Local oscillator used in various OPLL schemes is tunable.
The Costas loop and decision-driven loop (DDL) for BPSK [18], shown in Fig. 6.36, are two alternatives to the receivers
with residual carrier. Both these alternatives employ a fully suppressed carrier transmission, in which the entire transmitted
power is used for data transmission only. However, at the receiver side, a part of the power is used for the carrier extraction, so
some power penalty is incurred with this approach too. The corresponding decision-directed OPLL for QPSK is shown in
Fig. 6.37, with details provided in [79].
The coherent optical detection receivers with OPLL are very sensitive to laser linewidth, as discussed in [18–20]. Another
big challenge is to implement the optical voltage-controlled oscillator (OVCO). To avoid this problem, typically DSP
approaches, such as those described in Sect. 6.4, are used. Alternatively, the subcarrier-based OPLL, proposed in [80, 81],
described in Fig. 6.38 can be used, which replaces the loop filter and local laser from Fig. 6.37. In OVCO, external modulator,

Fig. 6.35 Balanced loop-based Received

receiver for BPSK
optical signal

PD 1 + Data
Data
LPF
hybrid PD 2 - detector output

PD: photodetector
Local Loop
laser filter LPF: lowpass filter
366 6 Advanced Detection Techniques and Compensation of Channel Impairments

Fig. 6.36 BPSK coherent optical Received

detection receivers with (a)
decision-driven (decision- optical signal
Data output
directed) loop and (b) Costas Data
PD 1 LPF
loop. PD photodetector, LPF detector
/2
low-pass filter
hybrid
PD 2 Delay line
PD: photodetector
LPF: low-pass filter
Local Loop
laser filter
(a)
Received
optical signal

PD 1 + Data
Data
LPF
PD 2 - detector output
Optical
hybrid
PD 3 +
LPF
PD 4 -
PD: photodetector
LPF: low-pass filter

Local Loop
laser filter
(b)

Fig. 6.37 Decision-directed Received

OPLL for QPSK. The
optical signal
configuration of balanced detector Data
(BD) is shown in Fig. 6.28 Data
BD 1 LPF
detector I output I
2x4
hybrid Data
Data
BD 2 LPF
- detector Q output Q

Local Loop I(t)

laser filter
BD: balanced detector

Local laser
signal L

Optical VCO (OVCO)

Mach- I(t)
Local CW Loop
Zehnder
laser filter
modulator

Fig. 6.38 Configuration of the subcarrier-based OPLL

6.8 Optical Communication Systems with Coherent Optical Detection 367

Fig. 6.39 Phase diversity E1

receivers (a) and demodulator S . PD LPF Demodulator
.
configurations (b) Optical . Ek
. PD LPF Demodulator LPF
N-port hybrid .
L . EN
PD LPF Demodulator
(a)
ASK
Squarer

DPSK
Squarer
Demodulator
Delay, T
CPFSK
Squarer

Delay,
(b)

such as the Mach-Zehnder modulator, is used to change the OVCO output frequency by ωCW  ωVCO, where ωCW is the
angular frequency of continuous-wave (CW) laser and ωVCO is the angular frequency of the electrical VCO.

6.8.4 Phase Diversity Receivers

To avoid the problems associated with OPLL-based receivers, the phase diversity receivers can be used. The general
architecture of a multiport homodyne receiver is shown in Fig. 6.39(a) [9, 20].
The incoming optical signal and local laser output signal can be written as

j½ωc tþϕS ðt Þ j½ωc tþϕLO ðt Þ

Sðt Þ ¼ aE s e , Lðt Þ ¼ E LO e , ð6:287Þ

where the information is imposed either in amplitude a or phase ϕs. Both incoming optical signal S and local laser output
signal L are used as inputs to N output ports of an optical hybrid, which introduces fixed phase difference k(2π/N ) (k ¼ 0, 1,
. . ., N-1) between the ports, so that the output electric fields can be written as
h i
1
Ek ðt Þ ¼ pffiffiffiffi Sðt Þejk N þ Lðt Þ :
2π
ð6:288Þ
N

The corresponding photodetector outputs are as follows:

ik ðt Þ ¼ RjE k ðt Þj2 þ I nk
n h io ð6:289Þ
R 2π
¼ jE LO j2 þ ajE S j2 þ 2ajES jjE LO j cos ϕS ðt Þ  ϕLO ðt Þ þ k þ I sn,k ,
N N

where Isn,k is the kth photodetector shot noise sample. Different versions of demodulators for ASK, DPSK, and CPFSK are
shown in Fig. 6.39(b). For ASK we have simply to square photodetector outputs and add them together to get

X
N
y¼ I 2k : ð6:290Þ
k¼1
368 6 Advanced Detection Techniques and Compensation of Channel Impairments

PSDx( )
2Ps /

Ps /

Fig. 6.40 Lorentzian spectrum of the laser diode

6.8.5 Dominant Coherent Optical Detection Noise Sources

In this section, we will outline the noise processes due to phase, polarization, and intensity variations in coherent optical
detector; a more detailed description of other optical relevant channel impairments has been already provided in Chap. 2.

6.8.5.1 Laser Phase Noise

Noise due to spontaneous emission causes phase fluctuations in lasers, leading to a nonzero spectral linewidth Δν. In
semiconductor lasers, the linewidth Δν is dependent on the cavity length and the value of the linewidth enhancement factor
and can vary from 10 KHz up to the megahertz range. The output from the semiconductor laser can be expressed as
pffiffiffiffiffi j½ω0 tþϕn ðt Þ 
xð t Þ ¼ PS e , ð6:291Þ

where ϕn(t) is the laser phase noise process. We can also assume that the corresponding PSD of x(t) can be expressed as the
Lorentzian curve. For this purpose, we can express the PSD of Eq. (6.292) in a simplified form as
  1
2PS ν  ν0 2
PSDx ðνÞ ¼ 1þ 2 , ð6:292Þ
π Δν Δν

which is shown in Fig. 6.40. The laser phase noise, expressed through the spectral linewidth, will cause bit error probability
degradation.
As an illustration, the impact of the laser phase noise to BPSK performance is illustrated in Fig. 6.41. As we can see, the
effect of phase noise on BER curves is twofold: (i) the BER curves are shifted to the right, and (ii) BER floor appears. In the
design of advanced coherent detection schemes, the impact of the laser phase noise is mitigated by digital signal processing
that takes place after photodetection, and it does impose a serious limitation if spectral linewidth is measured by hundreds of
kilohertz. Moreover, for higher multilevel modulation schemes, such as M-ary QAM, the laser phase noise can cause larger
performance degradation.

6.8.5.2 Polarization Noise

Polarization noise, which comes from discrepancies between the SOP of incoming optical signal and local laser oscillator, is
another factor that causes performance degradation in coherent receiver schemes. In order to mitigate its impact, several
avoidance techniques have been considered. These techniques can be classified as follows: (i) polarization control,
(ii) employment of polarization-maintaining fibers, (iii) polarization scrambling, (iv) polarization diversity, and
(v) polarization-division multiplexing. The employment of polarization control, polarization diversity, and polarization-
division demultiplexing has already been discussed above.
The polarized electromagnetic field launched into the fiber can be represented as
6.8 Optical Communication Systems with Coherent Optical Detection 369

Fig. 6.41 Probability of

error vs. number of photons per bit
in the presence of the laser phase
noise for homodyne BPSK
signaling

 
ex ðt Þ jωc t
E ðt Þ ¼ e , ð6:293Þ
ey ðt Þ

where ex and ey represent two orthogonal SOP components and ωc is the optical carrier frequency. The received field can be
represented by
" #
0 ex ðtÞ jωc t
ES ðtÞ ¼ H e , ð6:294Þ
ex ðtÞ

where H’ is the Jones matrix of birefringence, introduced in Chap. 2. In order to match the SOPs of local laser with that of
incoming optical signal, an additional transformation is needed, and it can be represented with transformation matrix H”, so
we have that
" # # "
ex ðtÞ ex ðtÞ
E0s ðtÞ ¼ |ffl00
Hffl{zffl
Hffl} 0
e jωc t
¼H e jωc t
, H ¼ H 00 H 0 : ð6:295Þ
ey ðtÞ ey ðtÞ
H

The SOP of local laser oscillator can be represented by Stokes coordinates as SLO ¼ (S1,LO S2,LO S3,LO), as introduced in
Chap. 2. The full heterodyne mix is possible only if SLO ¼ SR, where SR is the SOP of the received optical signal. The action
of birefringence corresponds to rotating the vector, which represents the launched SOP, on the surface of the Poincaré sphere.
This rotation can also be represented in terms of rotation matrices Ri (i ¼ 1, 2, 3) for angles αi around axes si. These matrices
can be represented as [75]
" #
cos ðα1 =2Þ j sin ðα1 =2Þ
R1 ðα1 Þ ¼ ,
j sin ðα1 =2Þ cos ðα1 =2Þ
" #
cos ðα2 =2Þ  sin ðα2 =2Þ
R2 ðα2 Þ ¼ , ð6:296Þ
sin ðα2 =2Þ cos ðα2 =2Þ
" #
ejα3 =2 0
R3 ðα3 Þ ¼ :
jα3 =2
0 e
370 6 Advanced Detection Techniques and Compensation of Channel Impairments

If we assume that the SOP of local laser oscillator is aligned with s1-axis (whose Jones vector is given by ELO ¼ [ex(t) 0]T)
in the Stokes space, we can establish the relationship between received and local laser SOPs in terms of spherical coordinates
(r, 2ε, 2η) (see Fig. 2.2 in Chap. 2) as
" # " #" #" #
e0x ðtÞ cos ðεÞ sin ðεÞ ejη 0 ex ðtÞ
¼ ¼
sin ðεÞ cos ðεÞ jη
e0y ðtÞ 0 e 0
|fflfflffl{zfflfflffl} |fflfflffl{zfflfflffl}
ES ELO ð6:297Þ
" jη
#
e cos ðεÞ
¼ ex ðtÞ:
ejη sin ðεÞ

The ratio between the power of heterodyned component (aligned with local laser) and the total power is given by

jejη cos ðεÞj

2
Phet 1
phet ðη, εÞ ¼ ¼ ¼ j cos εj2 ¼ ½1 þ cos ð2εÞ: ð6:298Þ
Ptot jejη cos ðεÞj2 þ jejη sin ðεÞj2 2

The probability density function of η ¼ ∡ (SR, SLO) is given as [20]

 
sin ð2ηÞ  A22 ½1cos ð2ψÞ 4R2 PS PLO
f ðηÞ ¼ e 4σ 1þ ½1 þ cos ð2ηÞ ,
2 4qRPLO ð6:299Þ
η 2 ½0, π:

6.8.5.3 Transimpedance Amplifier Noise

The typical optical receiver is based on transimpedance FET stage, as illustrated in Fig. 6.42. The PSD of receiver noise can be
written as [19]
 2
4kB T a f
PSDrec ðf Þ ¼ þ 4k B T a gm Γ , f T,eff ¼ gm =2πC T , C T ¼ C i þ CPIN , ð6:300Þ
Rf f T,eff

where gm is the transconductance, Γ is the FET channel-noise factor, and CT is the total capacitance (FET input capacitance Ci
plus PIN photodetector capacitance CPIN). (RL is the load resistor, Rf is the feedback resistor, Ta is the absolute temperature,
and kB is the Boltzmann constant.) This amplifier stage is popular because it has large gain-bandwidth product (GBWi) defined
as [19]

+V

RL
Rf
uout

Fig. 6.42 Transimpedance FET receiver

6.8 Optical Communication Systems with Coherent Optical Detection 371

   
 gm RL  1 þ gm RL 

GBW i ¼ R f   ¼ f T,eff R f RL : ð6:301Þ
1 þ gm RL  2πC T R f þ Ri

The intensity noise comes from the variation of optical power of transmitting laser P ¼ hPi + ΔP ( hi denotes, as before,
the statistical averaging). The relative intensity noise is defined as RIN ¼ hΔP2i/hPi2. Because the power of transmitting laser
fluctuates, the photocurrent fluctuates as well, I ¼ hIi + ΔI. The corresponding shot noise PSD can be determined as
2qhIi ¼ 2qRhPi. The intensity noise PSD is simply hΔI2i ¼ R2RINhPi2. The SNR in the presence of shot, receiver, and RIN
can be determined as signal power over total noise power:

2R2 PS PLO
SNR ¼ , ð6:302Þ
2qRPLO BW þ RIN  R2 P2LO BW þ PSDrec BW

where BW is the receiver bandwidth and PSDrec is introduced by Eq. (6.300). The SNR for balanced coherent optical reception
can be written as

2R2 PS PLO
SNR ¼ RIN
, ð6:303Þ
2qRPLO B þ CMRR R2 P2LO B þ 2PSDrec B

where CMRR is the common mode rejection ratio of FET. Therefore, the RIN of balanced receiver is significantly reduced by
balanced detection.

6.8.5.4 Coherent Optical Detection in the Presence of ASE Noise

In the rest of this section, we describe the coherent optical detection in the presence of amplified spontaneous emission (ASE)
noise. To facilitate explanations, we return to the basic balanced coherent optical detector, shown in Fig. 6.28. The 22
optical hybrid output port electric field for κ ¼ 1/2, ϕ1 ¼ 0, ϕ2 ¼ π can be written as follows:

1 1
Eo,1 ¼ pffiffiffi ðSn þ Z þ LÞ, E o,2 ¼ pffiffiffi ðSn þ Z  LÞ, ð6:304Þ
2 2

where we use Z to denote the ASE noise vector, which can be represented as Z ¼ z exp(jωTt), where ωT is the angular
frequency of the transmit laser. Further, z can be decomposed in terms of x- and y-polarization states by z ¼ zx |xi+zy |yi. The
photocurrent i1 now becomes

R  
i1 ¼ RjE o,1 j2 ¼ ðS þ Z þ LÞ Sn þ Z  þ L þ I 1,sh ¼
2 n
R R 2
¼ j Sn j 2 þ j Lj
2
|fflffl{zfflffl} 2ffl{zfflffl}
|ffl
Signalpower LO laserpower
ð6:305Þ
R 2 R   R  R
þ jZ j þ Sn L þ Sn L þ Sn Z  þ Sn Z þ ðLZ  þ L Z Þ þ I 1,sh ,
2ffl{zfflffl}
|ffl 2
|fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl} 2
|fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl} 2
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
ASE‐ASE noise Signal‐LO laserbeating Signal‐ASE noisebeating LO laser‐ASE noisebeating

where we have identified various beating terms as ASE-ASE noise interaction, signal-LO laser beating, signal-ASE noise
beating, and LO laser-ASE noise beating components. With I1,sh we denoted the shot noise of photodiode 1. In a similar
fashion, the photocurrent i2 can be expressed as
372 6 Advanced Detection Techniques and Compensation of Channel Impairments

R  
i2 ¼ RjE o,2 j2 ¼ðSn þ Z  LÞ Sn þ Z   L þ I 2,sh ¼
2
R 2 R 2
¼ j Sn j þ j Lj
2
|fflffl{zfflffl} 2ffl{zfflffl}
|ffl
Signalpower LO laserpower
ð6:306Þ
R 2 R  R  R
þ jZ j þ Sn L  Sn L þ Sn Z  þ Sn Z þ ðLZ   L Z Þ þ I 2,sh :
2ffl{zfflffl}
|ffl 2
|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} 2
|fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl} 2
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
ASE‐ASE noise Signal‐LO laserbeating Signal‐ASE noisebeating LO laser‐ASE noisebeating

The balanced detector photocurrent will be now

 
i ¼ i1  i2 ¼ R Sn L þ Sn L þ RðLZ  þ L Z Þ þ I 1,sh  I 2,sh : ð6:307Þ

Based on Eq. (6.307), we conclude that in this balanced detector, ASE-ASE noise components and signal-ASE noise
component cancel each other, providing that photodetectors in balanced detector have identical parameters. Therefore, the LO
laser signal-ASE noise beating is the dominant noise source, which can be modeled as zero-mean Gaussian with PSD given by

PSDLO‐ASE ¼ 2R2 jLj2 N 0 , ð6:308Þ

where N0 is the PSD of ASE noise, as described in Chap. 2.

6.9 Compensation of Atmospheric Turbulence Effects

The relevant approaches to deal with atmospheric turbulence effects include adaptive coding and adaptive coded modulation
[5, 9, 44–53], hybrid RF-FSO communications [9, 53], diversity techniques, MIMO signal processing [5, 8, 9, 12, 16],
adaptive optics approaches [82–88], and spatial light modulator (SLM)-based backpropagation method [88], to mention a few.
In adaptive coding approaches, the error correction strength (code rate) is adapted based on time-varying channel
conditions. When channel conditions are bad, a stronger channel code is used; on the other hand, when channel conditions
are favorable, a weaker channel code is used. In adaptive modulation and coding, both modulation format and code rate are
simultaneously adapted. These schemes have been already described in Sect. 6.5 (see also Chaps. 9 and 10 in ref. [9]).
The key idea behind the diversity-combining techniques [5, 8, 9, 12, 16] is to use several independent realizations of that
same transmitted signal to improve tolerance to channel impairments. The independent signal copies transmitted over
different paths or degrees of freedom will have a low probability of experiencing the same level of channel distortion. In
multiple-input multiple-output (MIMO) systems, the transmitters and receivers can exploit the knowledge about the channel
to determine the rank of so-called channel matrix H, which is related to the number of independent data streams that can be
simultaneously transmitted. The diversity and MIMO techniques are subject of study in Chap. 8.
In hybrid FSO-RF communications, FSO and RF subsystems cooperate to overcome the shortcomings of each other
[9, 53]. The atmospheric turbulence, as described in Chap. 2, is one of the most important factors that degrade the performance
of an FSO link, which is present even in clear sky conditions. Another relevant factor that impacts FSO transmissions is the
Mie scattering, which occurs in the near-infrared wavelength region when particle sizes due to fog, haze, and pollution
(aerosols) become comparable to the operating wavelength. Fog itself is the most detrimental factor that impacts the reliability
of an FSO link. In contrast, RF signals are not impacted by these problems, but are affected by other issues, particularly rain
and snow. This suggests that the two transmission media (FSO and RF) can be operated in a complementary fashion,
depending on the predominant weather and atmospheric conditions. During certain channel conditions, FSO and RF link
performance becomes comparable, when it makes sense to simultaneously and cooperatively communicate over the two
channels so as to maximize the overall link performance. For additional details, an interested reader is referred to Chap. 4 (see
Sect. 4.6).
Atmospheric turbulence effects can also be handled through the use of the azimuthal phase correction method, similar to
the Gerchberg-Saxton (GS) phase retrieval algorithm [82, 83]. This method can be described as follows [89]. In the first stage,
we determine the azimuthal phase of a given spatial mode, φ(x,y), in the absence of turbulence, based on the magnitude
response of the Fourier transform of exp[jφ(x,y)]. In the second stage, we determine the azimuthal phase of spatial modes in
6.9 Compensation of Atmospheric Turbulence Effects 373

the presence of turbulence. The distortions introduced by atmospheric turbulence can be compensated for by using the GS
phase retrieval algorithm. However, there is no evidence that this method can be used to deal with strong turbulence effects.
In the remainder of this chapter, we describe the usage of adaptive optics and SLM-based backpropagation method to deal
with atmospheric turbulence effects.

6.9.1 Adaptive Optics Techniques

A linear adaptive optics system consists of three subsystems: (i) wavefront sensor to detect the atmospheric distortions,
(ii) wavefront corrector to compensate for turbulence effects, and (iii) the control processor to monitor the wavefront sensor
information and control the wavefront corrector.
Before we describe each individual subsystem, we provide the Zernike representation of the atmospheric turbulence.

6.9.1.1 Zernike Representation of Atmospheric Turbulence

The polynomial expansion of the arbitrary wavefront φ(r) over the circular aperture of radius R can be represented as [85, 86]:

X
1
φð r Þ ¼ al Z l ðrÞ, ð6:309Þ
l¼1

where al are the coefficients of expansion. Zernike polynomials Zl(r) form a set of orthogonal polynomials expressed as
product of radial Rm
n and azimuth Vml components as follows [85]:

pffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffi
Z l ðrÞ ¼ 2δm0 n þ 1 Rm ðr=RÞV ml ðθÞ ¼ cmn Rm
n ðr=RÞV ml ðθ Þ, ð6:310Þ
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} n
cmn

where radial components Rm

n are defined as

ðnm   !
XÞ=2 ð1Þs ðn  mÞ! XÞ=2
ðnm
s ns
n  2s
Rm
n ð ρÞ ¼   nm  ρ n2s
¼ ð1Þ nm ρn2s , ð6:311Þ
s! nþm  s !  s ! s s
s¼0 2 2 s¼0 2

while azimuth components Vml as


cos ðmθÞ, l ¼ 0, 2, 4, ⋯ðl isevenÞ
V ml ¼ : ð6:312Þ
sin ðmθÞ, l ¼ 1, 3, 5, ⋯ðl is oddÞ

Zernike polynomials are orthogonal within a circular aperture of radius R:

ZZ
Z l ðrÞZ k ðrÞd 2 r ¼ cl δlk : ð6:313Þ
rR

By employing the orthogonality principle, from Eq. (6.309), we can write

ZZ X
1 ZZ
φðrÞZ l ðrÞd r ¼
2
ak Z l ðrÞZ k ðrÞd2 r , ð6:314Þ
k¼1
rR rR
|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
cl δlk

and determine the coefficient al as follows:

374 6 Advanced Detection Techniques and Compensation of Channel Impairments

Table 6.2 Zernike polynomials sorted by Noll index l

Noll index l Radial degree n Azimuthal frequency m Zl Typical name
1 0 0 1 Piston
2 1 1 2ρ cos θ Tip, lateral position, tilt-x
3 1 1 2ρ sin θ Tilt, lateral position, tilt-y
pffiffiffi 2
4 2 0 3ð2ρ  1Þ Defocus, longitudinal position
pffiffiffi 2
5 2 2 6ρ sin 2θ Oblique astigmatism
pffiffiffi 2
6 2 2 6ρ cos 2θ Vertical astigmatism
pffiffiffi 2
7 3 1 8ð3ρ  2ρÞ sin θ Vertical coma
pffiffiffi 2
8 3 1 8ð3ρ  2ρÞ cos θ Horizontal coma
pffiffiffi 3
9 3 3 8ρ sin 3θ Vertical trefoil
pffiffiffi 3
10 3 3 8ρ cos 3θ Oblique trefoil
pffiffiffi 4
11 4 0 5ð6ρ  6ρ2 þ 1Þ Primary spherical
pffiffiffiffiffi 4
12 4 2 10ð4ρ  3ρ2 Þ cos 2θ Vertical secondary astigmatism
pffiffiffiffiffi 4
13 4 2 10ð4ρ  3ρ2 Þ sin 2θ Oblique secondary astigmatism
pffiffiffiffiffi 4
14 4 4 10ρ cos 4θ Oblique quadrafoil
pffiffiffiffiffi 4
15 4 4 10ρ sin 4θ Vertical quadrafoil

ZZ
1
al ¼ φðrÞZ l ðrÞd 2 r: ð6:315Þ
πR2
rR

During propagation over the atmospheric turbulence channels, the wavefront φ(r) get randomly distorted with samples
generated as described in Chap. 2, so that we can use the Gaussian approximation to describe the random phase screen
variations. To characterize this Gaussian process, we need to determine the correlation matrix with corresponding correlation
coefficients as follows:
ZZ ZZ
1
hal ak i ¼ 2
Z l ðrÞ hφðrÞφðr0 ÞiZ l ðr0 Þd 2 r0 d 2 r: ð6:316Þ
ðπR2 Þ
rR r0 R

The first several Zernike polynomials, ordered by Noll index l, are summarized in Table 6.2, based on [90]. The following
recurrence relation can be used to calculate all other radial terms:

Rm m m
n ¼ f2ðn  1Þ½2nðn  2Þρ  m  nðn  2ÞRn2 ðρÞ  nðn þ m  2Þðn  m  2ÞRn4 ðρÞg=½ðn þ mÞðn  mÞðn  2Þ:
2 2

ð6:317Þ

The convenience of the Zernike polynomials is that we can determine the power associated with each Zernike term. Once a
certain number of Zernike terms get compensated for, we can determine the residual aberration as follows:
ZZ
ΔL ¼ jφðrÞ  φL ðrÞj2 d2 r, ð6:318Þ
r<R

where φL(r) is the approximation with the first L terms:

X
L
φ L ðrÞ ¼ al Z l ðrÞ: ð6:319Þ
l¼1

When L > 10, the following approximation due to Noll can be used [90]:
6.9 Compensation of Atmospheric Turbulence Effects 375

CCD array Lenslet

(photodetectors) array
Unaberrated
location

Control processor

Distorted wavefront
Unaberrated
location
f

Fig. 6.43 Illustration of the Shack-Hartmann sensor

pffiffi
 5=3
D 2

ΔL
0:2944L 3=2
rad , ð6:320Þ
r0

where r0 is the Fried parameter introduced in Chap. 2.

6.9.1.2 Shack-Hartmann Wavefront Sensor

The most widely used wavefront sensor is the Shack-Hartmann sensor, whose basic layout is illustrated in Fig. 6.43.
This wavefront sensor consists of an array of collecting lenslets, in whose focal plan an array of photodetectors is placed.
The subapertures (lenslets) can be arranged in rows, the hexagon packing can be used, etc. Each subaperture focuses the
corresponding portion of the light beam to the photodetector plane. The signals from different photodetectors are used to
determine the displacement of every sub-beam from the optical axis of the corresponding subaperture, which is proportional to
the mean tilt of the wavefront within the subaperture. When sufficient number of photodetectors in the array is used, such as
CCD array, the displacement of the focal spot is determined as the displacement of its centroid. The focal spot centroid vector
within subaperture A can be determined as [90] (see also [85])
ZZ
kf
rc ¼ IðrÞ∇φd2 r, ð6:321Þ
P
A

where f is the focal length of the lenslet and P is the total radiation power incident to the subaperture A. The normalized phase
gradient can be expressed in terms of the gradients of the real and imaginary parts of the complex amplitude E as follows [85]:
" #
2 1 Im ðEÞ

IðrÞ∇φ ¼ jEj ∇tan ¼ Re ðEÞ∇Im ðEÞ  ∇ Re ðEÞ Im ðEÞ: ð6:322Þ

Re ðEÞ

After substituting Eq. (6.322) into Eq. (6.321), we obtain the following equation for the local tilt:
376 6 Advanced Detection Techniques and Compensation of Channel Impairments

ZZ
1       
 
s ¼ rc =kf ¼ Re E ∇Im E  ∇ Re E Im E d 2 r: ð6:323Þ
P
A

The wavefront modal reconstruction of the sensor aperture can be obtained from the estimates of local tilts, as described in
[92] (see also [85]). The starting point is the approximation of φ(r) in terms of Zernike polynomials Zl, given by Eq. (6.319).
The local tilts of the mth subaperture sm (m ¼ 1, 2, . . ., M) are fitted to the estimates obtained in the wavefront sensor as
follows:
" #2
X
M X
L
εL ¼ al Zlm  sm , ð6:324Þ
m¼1 l¼1

where Zlm is the local tilt of the lth basis function Zl on the mth subaperture. The variation method can be applied to determine
the expansion coefficients ak, by setting the partial derivatives of the error ε with respect to ak to zero [85]:
" #2
∂ X X
M L
∂εL
¼ al Zlm  sm ¼ 0: ð6:325Þ
∂ak ∂ak m¼1 l¼1

The corresponding set of linear equations can be written in a matrix form as follows:

X
M X
M
Aa ¼ b, Akl ¼ ZTlm Zkm , bk ¼ sm Zkm : ð6:326Þ
m¼1 m¼1

The solution of expansion coefficient can be obtained by the following matrix inversion and multiplication [85]:

a ¼ A1 b: ð6:327Þ

To determine the local tilts according to Eq. (6.323) would require quadrature sensors, which is not possible by the Shack-
Hartmann sensor, and this equation can only be used for modeling purposes. With the Shack-Hartmann sensor we measure the
displacements of the focal spots, denoted as Δrm, with respect to reference positions. The local tilts are directly proportional to
the displacement and reversely proportional to the focal length f, that is, sm ¼ Δrm /f. Local tilts are then substituted in
Eq. (6.236) to determine the coefficients of the expansion al. Assuming the intensity of aperture incident beam I(x,y), the
position of the image spot in each small region of the grid is determined by evaluating the centroid:

P Ny
Nx P P Ny
Nx P
xi I ij yi I ij
i¼1 j¼1 i¼1 j¼1
r ¼ ðx, yÞ; x ¼ Ny
,y ¼ Ny
, ð6:328Þ
P
Nx P P
Nx P
I ij I ij
i¼1 j¼1 i¼1 j¼1

where Nx (Ny) is the number of focal spots along the x-direction (y-direction).

6.9.1.3 Wavefront Correctors

Regarding the wavefront correctors, the most popular are adaptive mirrors, which can be classified as either deformable
mirrors or segmented mirrors. In modal corrector, the residual aberration error, based on Eq. (6.318), can be defined as

ZZ 2
X
L 

ΔL ¼ WðρÞjφðρÞ  al Z l ðρÞ d 2 ρ, ð6:329Þ
l¼1


where the aperture function W(r) corresponds to the circle of the radius R:
6.9 Compensation of Atmospheric Turbulence Effects 377

Reflecting surface
Glass
Front (inner)
electrode PZT (lead-zirconate-titanate)

Back
electrodes

Fig. 6.44 Illustrating the cross section of the biomorphic deformable mirror


1, ρ  R
W ðρÞ ¼ : ð6:330Þ
0, ρ > R

By employing the orthogonality principle, the expansion coefficients can be determined as

RR
φðρÞZ l ðρÞd2 ρ
ρR
al ¼ RR : ð6:331Þ
Z 2l ðρÞd2 ρ
ρR

Alternatively, the variational method, similar to that given by Eqs. (6.325, 6.326 and 6.327), can be used. The only
difference is that the elements of matrix A and vector b are determined, respectively, as
ð ð
Akl ¼ WðρÞZ k ðρÞZ l ðρÞd2 ρ, bk ¼ WðρÞZ k ðρÞφðρÞd 2 ρ: ð6:332Þ

The most commonly used deformable mirrors are those with discretely arranged points of force (force moment). The simplest
and most popular method is to control the mirror surface with the help of actuators. The actuators, performing the deformation
of the deformable mirror, could be the discrete ones, acting either perpendicular to the surface or affecting the bending
moments for edge actuators. The actuators can also be continuous such as in a membrane mirror or a bimorph mirror. The
bimorph deformable mirror, whose cross section is illustrated in Fig. 6.44, is composed of either glass with reflecting surface
or metal mirror faceplate bonded to a crystal of piezoelectric ceramics, such as lead-zirconate-titanate (PZT)-based ceramics,
which is polarized normal to its surface. The bonded (glued) interface between glass and PZT ceramics contains inner
electrode. The back electrodes are attached to the PZT-based ceramics. The glass is polished and coated to create the surface of
the deformable mirror. The wires are attached to the electrodes.
By applying the voltage V across front and back electrodes, the corresponding dimension of the PZT-based ceramics
changes. The curvature R of the deformable mirror is directly proportional to the voltage applied and reversely proportional to
the thickness t of the PZT-based ceramics, that is [94],

V
R ¼ d13 , ð6:333Þ
t2

where d13 is the piezoelectric tensor coefficient. The actuators are organized in either square or hexagonal array. Square arrays
come in the following sizes: 4, 9, 16, 21 (55-4 on corners), 37 (77-12 on corners), 69 (99-24), 97 (55-4), etc. On the
other hand, hexagonal arrays come in the following sizes: 3, 7, 19, 37, 61, 91, 127, etc., with generic formula being 1+6∑n n,
where n is a positive integer. For additional details, an interested reader is referred to [93, 94].
The mirror surface can be described as the weighted sum of the response functions fl [85]:

X
L
S ð ρÞ ¼ al f l ðρ  ρl Þ, ð6:334Þ
l¼1

wherein the response functions are determined experimentally, based on the system of mechanical equations. To simplify the
problem, the Gaussian response approximation is used in numerous papers, such as [85, 93]:
378 6 Advanced Detection Techniques and Compensation of Channel Impairments

 
f l ðρÞ ¼ exp ρ2 =w2l , ð6:335Þ

where the halfwidth wl is determined experimentally and typically ranges from 0.7d to 0.8d, with d being the spacing between
actuators. The approximation of the surface required by the deformable mirror, with a given set of response functions, can be
obtained as the minimization problem of the following variance, which is similar to Eq. (6.329):

ZZ  2
 X
L 

σ 2L ¼ WðρÞφðρÞ  al f l ðρ  ρl Þ d2 ρ: ð6:336Þ
l¼1


To reduce the computation time, the truncated Gaussian response function can be used:
(
exp ðρ2 =w2l Þ, ρ  wl
f l ðρÞ ¼ ð6:337Þ
0, ρ > wl

Now we can apply the variation method given by Eqs. (6.325, 6.326 and 6.327), wherein the elements of matrix A and
vector b are determined, respectively, as
X      
Akl ¼ W ρi,j f k ρi,j  ρk f l ρi,j  ρk ,
jρi,j ρk j<2w
X       ð6:338Þ
bk ¼ W ρi,j f k ρi,j  rk φ ρi,j ,
k

where we used the subscripts i, j to denote the grid (i ¼ 1, . . ., I; j ¼ 1, . . ., J). The actuator fastening points are typically placed
on an equidistant grid with node coordinates (xl, yl) being determined by [85]
h pffiffiffiipffiffiffi n pffiffiffiopffiffiffi
xl ¼ R þ l= L Ld, yl ¼ R þ l= L Ld, ð6:339Þ

where the brackets [] are used to denote the integer part of enclosed entity, while the brackets {} are used to denote the
fractional part.
Now we have all elements required for an adaptive optics system, which can be placed at either transmitter or receiver side.
The receiver side adaptive optics system is illustrated in Fig. 6.45. The wavefront errors are measured with the help of the
Shack-Hartmann wavefront sensor, and to correct for the aberration errors, the deformable mirror is used in a closed loop,
controlled by the control subsystem. The pilot beam is used to estimate the distortions introduced by atmospheric turbulence
channels and deformable mirror. Once the residual aberrations are minimized, the signal beam is reflected from deformable
mirrors and the atmospheric turbulence-induced wavefront errors are corrected for, providing that resolution is sufficiently
good, and time needed for deformable mirror adjustment is shorter than the channel coherence time.
The size of aperture in front of photodetector is properly chosen, depending on FSO link length. For signal spatial mode
transmission, a single photodetector might be sufficient. Alternatively, a photodetector array can be used to exploit spatial
diversity, by properly selecting the compression telescope, collimating lens, the aperture size in front of photodetector array,
and separation of photodetectors so that the different optical paths will exhibit different turbulence effects. In such a way, we
combine the adaptive optics with spatial diversity to deal with strong turbulence effects. Alternatively, the low-cost wavefront
sensorless adaptive optics can be used to reduce the effects due to turbulence, and to deal with remaining turbulence effects,
the adaptive LDPC coding can be used, as described in [88]. By employing the multiple spatial laser sources, the MIMO
signal processing can be combined with adaptive optics.

6.9.2 SLM-Based Backpropagation Method

The receiver side adaptive optics can be used to deal with atmospheric turbulence effects in both single-mode and multimode
free-space optical communication systems including classical, QKD, and quantum communication systems. However, it has
6.9 Compensation of Atmospheric Turbulence Effects 379

Fig. 6.45 Typical receiver side

adaptive optics system. BS, beam Photodetector
splitter array

Lenslet
array

BS CCD
Compressing
telescope

Collimating

subsystem
Signal beam lens

Control
Pilot beam

Deformable mirror

Receiver side adaptive optics

been shown in [87] that the residual wavefront error is limited by the finite resolution of the deformable mirror (DM). To deal
with strong turbulence effects, the resolution of existing DMs is not sufficient. To solve for this problem, a liquid crystal
spatial light modulator (LC-SLM) can be used as a promising alternative to the DM [87]. As an illustration, an LC-SLM with a
10001000 pixel array in a 1010 mm2 area has been reported in [94]. The SLMs can also be used to emulate atmospheric
turbulence effects [95]. However, because the LC-SLM is polarization sensitive, we need two SLMs, one for each orthogonal
polarization state. Alternatively, SLM can be segmented into two areas corresponding to each SOP. To deal with strong
turbulence effects, the SLM-based backpropagation mode (BPM) method has been proposed in [87]. In the SLM-based BPM
scheme, the pilot beam is propagated backward from the receiver to the transmitter to probe the atmospheric channel. The
distorted wavefront is now determined at the transmitter side with the help of wavefront sensor, as illustrated in Fig. 6.46.
For the signal beam, SLM (SLMs) is (are) used to transmit the data sequence into specific spatial mode, which is the
conjugation of the determined wavefront. Even though there exist amplitude-modulating liquid crystal on silicon (LCOS)
devices, it has been shown in [87] that the low-cost phase-only SLMs to modulate the signal beams in the SLM-based BPM
scheme are sufficient to deal with turbulence effects in a strong atmospheric turbulence regime. Compared to receiver side
adaptive optics scheme, in which the segmented DM applies modal correction with superposition Zernike modes, the SLM is
programed to generate the conjugated phase of the piecewise wavefront measured by the wavefront sensor directly
[87]. Essentially, SLM operates as a zonal corrector to take advantage of its high spatial resolution. Therefore, the wavefront
of the signal beam is pre-shaped so that after propagation the light will focus at a predetermined position, i.e., the target
detector. This scheme is applicable in both single transmitter-single receiver and MIMO systems. In multi-spatial mode
scenario, when we use one probing source for each parallel channel, the same number of wavefront sensors will be needed,
which result in high complexity of the transmitter optics. To reduce the complexity, fewer probing sources can be used
with linear phase interpolation, as proposed in [87]. When the number of probe sources is smaller than the number of
spatial mode-based transmitters, a linear phase term is needed to modify the conjugation wavefront to provide the necessary
signal displacement to the position of the target detector at the receiver side. Therefore, the pre-shaped wavefront Ushaped is
given by [87]
( )
xðxdet  xprob Þ þ yðydet  yprob Þ
U shaped ¼ U {backpropagation exp j2π , ð6:340Þ
λf

where (xdet, ydet) is the position of a target detector, while (xprob, yprob) is the position of the closest probing source to the target
detector, and f is the focal length of the receiver (compressing) telescope. Additional details of SLM-based BPM can be found
in [87].
380 6 Advanced Detection Techniques and Compensation of Channel Impairments

Fig. 6.46 Illustration of the Optical

SLM-based backpropagation transmitter
method Expanding
telescope

Collimating
lens

Signal beam
Pilot beam
SLM

Lenslet array
CCD
Control
subsystem

6.10 Concluding Remarks

This chapter has been devoted to the advanced detection and channel impairments compensation techniques applicable to both
wireless and optical communication systems. To better understand the principles behind advanced detection and channel
impairments compensation, the fundamentals of detection and estimation theory have been described in Sect. 6.1, including
geometric representation of received signals (Sect. 6.1.1), optimum receiver design (Sects. 6.1.2 and 6.1.3) in either maximum
SNR sense (Sect. 6.1.2) or symbol error probability sense (Sect. 6.1.3), and symbol error probability derivations (Sect. 6.1.4).
Fundaments of estimation theory have been provided in Sect. 6.1.5.
In the section on wireless communication systems performance, Sect. 6.2, different scenarios have been described
including outage probability scenario (Sect. 6.2.1), average error probability scenario (Sect. 6.2.2), and combined outage-
average error probability scenario (Sect. 6.2.3). The moment-generating function-based approach to average error probability
calculation has been described in Sect. 6.2.4. The performance calculation in the presence of Doppler spread and fading
effects has been described in Sect. 6.2.5.
In the section on channel equalization techniques, Sect. 6.3, after introductory remarks on ISI-free transmission and partial-
response signaling (Sect. 6.3.1), the zero-forcing equalizers suitable for compensation of ISI, chromatic dispersion, PMD in
fiber-optics communication; ISI and multipath fading effects in wireless communications; and ISI and multipath effects in
indoor optical wireless communications have been described. The optimum linear equalizer design in the minimum mean-
square error sense has been provided in Sect. 6.3.3, while the Wiener filtering concepts have been introduced in Sect. 6.3.4.
The most relevant post-compensation techniques, applicable to both wireless and optical communications, have been
described next including adaptive equalizer (Sect. 6.3.5), decision-feedback equalizer (Sect. 6.3.6), MLSD or Viterbi
equalizer (Sect. 6.3.7), and blind equalization techniques (6.3.8). Relevant synchronization techniques have been described
in Sect. 6.4.
In the section on adaptive modulation techniques, Sect. 6.5, various techniques to adapt the transmitter to time-varying
wireless/optical channel conditions, enabling reliable and high spectral-efficient transmission, have been described. Various
scenarios have been described including data rate, power, code rate, error probability adaptation scenarios, as well as a
combination of various adaptation strategies. Various variable-power variable-rate modulation techniques have been
described in Sect. 6.5.1 including water-filling method, channel inversion, and truncated channel inversion. The discrete-
rate and discrete-power discrete-rate approaches have been discussed as well. The adaptive coded modulation techniques have
been described in Sect. 6.5.2.
6.11 Problems 381

The Volterra series-based equalization to deal with fiber nonlinearities and nonlinear effects in wireless communications
has been described in Sect. 6.6. In Sect. 6.7, the digital backpropagation has been described as a method to deal with fiber
nonlinearities, chromatic dispersion, and PMD in a simultaneous manner.
Section 6.8 has been devoted to the coherent optical detection. The following topics have been covered: various balanced
coherent optical detection schemes for two-dimensional modulation schemes in Sect. 6.8.1; polarization diversity and
polarization demultiplexing schemes in Sect. 6.8.2; homodyne coherent optical detection based on phase-locked loops in
Sect. 6.8.3; phase diversity receivers in Sect. 6.8.4; and the dominant coherent optical detection sources, such as laser phase
noise, polarization noise, transimpedance amplifier noise, and ASE noise, in Sect. 6.8.5.
In the section on compensation of atmospheric turbulence effects, Sect. 6.9, various techniques to deal with atmospheric
turbulence have been described including adaptive coding, adaptive coded modulation, diversity approaches, MIMO signal
processing, and hybrid FSO-RF communication approach. In Sect. 6.9.1, the adaptive optics approaches to deal with
turbulence effects have been described in detail. The SLM-based backpropagation method has been described in Sect. 6.9.2.
The set of problems is provided in the incoming section to help the reader gain deeper understanding of material described
in this chapter.

6.11 Problems

6.1. A low-pass, zero-mean, Gaussian random process X(t) has a PSD of


A jfj < B
GX ð f Þ ¼ :
0 jfj  B

Ð T=2
Let Y ¼ ð1=T Þ T=2 X ðt Þdt. Assuming that T >> 1/B, calculate σ 2Y and compare it against σ 2X .
6.2. The AWGN signal is passed through the system whose transfer function can be modeled as in Fig. P6.1. What is the
output noise power? Determine the autocorrelation function of the random noise process at the output of the system.
Assume that R/L ¼ 8 and 1/(LC) ¼ 25.

R L

+
+
x(t) C y(t)

Fig. P6.1 The circuit under study

6.3. Determine the average energy per symbol for 4-ASK, assuming that each of the four symbols is equally likely. Derive the
expression for symbol error probability. Compare it against QPSK.
6.4. The signal constellation is obtained by placing the signal constellation points in vertices of a rhombus (rhomb) of side
length a. Sketch the signal constellation and determine the average signal energy as the function of a, assuming that all
symbols are equally likely. Determine the optimum decision regions. Select the arbitrary basis functions ϕ1(t) and ϕ2(t)
and express the four signals (from signal constellation diagram) in terms of basis functions. Determine the optimum
receiver.
6.5. In this problem, we study the performance of 5-QAM signal constellation shown in Fig. P6.2 against that of 5-PSK with
constellation points located on a circle of radius a. Assuming that all constellation points are equally likely:
(a) Determine the decision boundaries and decision regions for both 5-QAM and 5-PSK.
(b) Derive the expression for average error probability of 5-QAM.
(c) Derive the expression for average error probability of 5-PSK and compare it against that of 5-QAM.
382 6 Advanced Detection Techniques and Compensation of Channel Impairments

a/2

a/2

a/2

Fig. P6.2 The 5-QAM signal constellation diagram

6.6. The cube signal constellation is obtained by placing the signal constellation points in vertices of the cube of side length
a. Assuming that all symbols are equally likely, derive the expression for average symbol error probability. Compare it
against 8-star-QAM constellation shown in Fig. 5.12 (see Chap. 3).
6.7. Let r(Ts) (Ts is the symbol duration) be the output of the optimum binary receiver. Prove that the following is valid:

m1 ¼ E frjm1 g ¼ E 1  ρ12 ,

m2 ¼ EfrjS2 g ¼ E2 þ ρ12 ,

8T 9
<ð =
σ 2 ¼ Varfrjm1 g ¼ Varfrjm2 g ¼ Var nðt Þ½s1 ðt Þ  s2 ðt Þdt
: ;
0
N0
¼ ðE 1 þ E 2  2ρ12 Þ
2

where

Tðs Tðs Tðs

E1 ¼ s21 ðt Þdt, E 2 ¼ s22 ðt Þdt, andρ12 ¼ s1 ðt Þs2 ðt Þdt:

0 0 0

P
D
6.8. Let us consider the random process R(t) and its expanded form Rðt Þ ¼ r j Φ j ðt Þ þ Z ðt Þ, 0  t  T s , where Z(t) is
j¼1
the remainder of noise term. {Φj(t)} represents the set of orthonormal basis functions, and rj is the projection of R(t)
along the jth basis function. Let Z(tm) be the random variable obtained by observing Z(t) at t ¼ tm. Show that the sample
j ¼ 1, 2, . . . , D
of the noise process Z(tm) is statistically independent on correlators’ outputs {Rj}:E½R j Zðt m Þ ¼ 0, f :
0  tm  T s
6.9. An orthogonal set of signals is characterized by the property that the dot product of any pair of signals is zero. Provide a
pair of signals that satisfies this condition. Construct the corresponding signal constellations.
6.10. A set of 2M biorthogonal signals can be derived from the set of orthogonal signals by augmenting it with the negative
sign signals from orthogonal set. Does the biorthogonal signal construction increase the dimensionality of original
signal set? The construction of the biorthogonal signal constellation for orthogonal signal constellation is developed in
Problem 6.9.
6.11. Formulate the signal constellations for the following lines codes: unipolar NRZ, polar NRZ, unipolar RZ, and
Manchester line code. For Manchester line code, derive the expression for error probability when the ML rule is
applied in an AWGN channel.
6.12. The simplex signals are equally likely and highly correlated signals. When derived from M orthogonal signals, the
correlation coefficient between any two pairs of signals in the set is given by
6.11 Problems 383


1=ðM  1Þ, k 6¼ l
ρkl ¼ :
1, k ¼ l

The simplest way to construct the simplex signal set is to start with a set of orthogonal signal set, with each element
from the set having the same energy, and then create minimum energy signal set. Consider now the signal set with
constellation points placed on equilateral triangle vertices. Prove that these three signals represent a simplex signal set.
6.13. Let us observe the M-ary cross-constellation described in Chap. 5 (see Fig. 5.11). Prove that for sufficiently high SNRs,
 pffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffi
the average symbol error probability can be estimated as Pe ffi 2 1  1= 2M erfc E 0 =E 0 , where E0 is the
smallest symbol energy in constellation. Determine also the union bound of this constellation and discuss its accuracy.
Finally, determine the nearest neighbor approximation.
6.14. Let us observe M-ary PSK signal constellation. Derive the expression for the average symbol error probability.
Determine also the union-bound approximation. Finally, determine the nearest neighbor approximation. Discuss the
accuracy of these approximations against the exact expression for average symbol error probability.
6.15. Let us consider a signal constellation that is symmetric with respect to the origin. Assume that there are M signal
constellation points in the set, which are equally likely. Using the upper bound approach, determine the corresponding
average symbol error of this signal constellation.
6.16. Let us reconsider the example from Sect. 6.1.5. Show how the equation in the example of this section is derived. By
using it, deduce how the phase-locked loop can be implemented.
6.17. Derive the orthogonality principle given by Eq. (6.161). By using it, derive the most general form of the Wiener filter
(6.163). Finally, derive the variance error of the Wiener filter. Show the steps of your derivations.
6.18. This problem is useful in MIMO signal processing and filter design. Let x be transmitted, r be received, and n be noise
vectors, which are related by r ¼ Hx + n, where H is the channel matrix. Derive the Wiener estimate and determine
its minimum variance of error.
6.19. In this problem, we study the carrier phase recovery by data-aided method. In data-aided method, the receiver has the
knowledge of preamble of length L, namely, fal gL1 l¼0 : Determine the ML estimate of the carrier phase. Provide the block
diagram of this ML phase estimator.
6.20. Let us consider heterodyne coherent detection designs with two balanced detectors and single balanced detector, as
shown in Fig. P6.3. Derive vI and vQ outputs in both designs, assuming that the incoming optical signal is given by
pffiffiffiffiffi
S ¼ P ja je jðωs tþϕi þϕS,PN Þ , where ω is the carrier frequency of transmitting laser, (a ,ϕ ) are the polar coordinates of
i s i s i i
two-dimensional signal constellation being transmitted at the ith time instance (symbol interval), while ϕs,PN is the laser

From /2
vI
SMF

From local
vQ
oscillator laser

vI
From LPF
SMF
cos( IF t)

- /2
From local
vQ
oscillator laser LPF

Fig. P6.3 Two optical heterodyne detection schemes: (a) two balanced detector-based scheme and (bottom) one balanced detector-based
scheme
384 6 Advanced Detection Techniques and Compensation of Channel Impairments

phase noise of transmitting laser. Assume that the local oscillator (LO) laser signal is given by L ¼
pffiffiffiffiffiffiffiffi jðωLO tþϕLO,PN Þ
PLO e , where ωLO is the carrier frequency of local laser oscillator and ϕLO,PN is the corresponding
laser phase noise. Compare the receiver sensitivities of these two schemes in the following two regimes: (i) ASE noise-
dominated scenario and (ii) shot noise-dominated scenario.
6.21. The signal constellation of a ternary signal set is obtained by placing the signal constellation points in vertices of an
equilateral triangle of side length a.
(a) Sketch the signal constellation and determine the average signal energy as the function of a, assuming that all
symbols are equally likely. Determine the optimum decision regions.
(b) Select the arbitrary basis functions ϕ1(t) and ϕ2(t) and express the signals from signal constellation diagram in terms
of basis functions.
(c) Determine the corresponding minimum energy signal constellation.
(d) What would be the optimum receiver in an AWGN channel?
(e) Derive an expression for average symbol error probability against average symbol energy for an AWGN channel.
(f) In a fading environment, the received signal power varies randomly over distance or time due to shadowing and/or
multipath fading. Determine the average symbol error probability when in Nakagami m ¼ 2 fading.
(g) Let us now assume that we would like to transmit video signal, by employing the same signal constellation, with
symbol rate larger than coherence bandwidth over the Nakagami-m fading. What would be the outage probability?
(h) Determine the required signal-to-noise ratio corresponding to the outage probability of 1%, for the outage
probability expression derived in (g). Do not evaluate exact value; just write down the corresponding equation to
be solved numerically. Now, in special case for m ¼ 1, determine the fade margin.
6.22. A four-dimensional (4-D) signal constellation is described by the following set of points {a, a, a, a} (a > 0).
Different signal constellation points are used with the same probability. Determine the average symbol energy. Choose
the appropriate basis functions {Φ1(t), Φ2(t), Φ3(t), Φ4(t)} and express the signal constellation points in terms of basis
functions. For an AWGN channel, determine the average symbol error probability as a function of the average energy
(Eav). Evaluate now the average symbol error probability in the presence of Rayleigh fading.
6.23. This problem is related to the adaptive modulation applied to 9-ary 2-D constellation described below, transmitted over
wireless channel in the presence of Nakagami fading and AWGN. The eight constellation points in this 2-D constella-
tion are placed on a circle of radius 1. The last (ninth) point is placed in the origin. Assume that the point in the origin
appears with probability 0.5, all others with probability 0.5/8.
(a) Determine the average symbol error probability for this modulation format in the presence of fading and AWGN.
(b) Determine the optimum power adaptation strategy as well as the corresponding spectral efficiency of this scheme in
the presence of fading and AWGN.
(c) Describe the channel inversion technique for this signal constellation and determine the corresponding spectral
efficiency in the presence of fading and AWGN.
(d) Describe the truncated channel inversion technique for this signal constellation and determine the corresponding
spectral efficiency in the presence of fading and AWGN.
6.24. A five-signal configuration in a two-dimensional signal space has been provided already in Fig. P6.2. The probability of
occurrence of the constellation point placed in the origin is 2/5. Other constellation points occur with the same
probability. Determine the average symbol energy. Choose the appropriate basis functions Φ1(t) and Φ2(t) and express
the five signals in terms of basis functions. In the signal space, sketch the optimum decision regions, assuming an
AWGN channel model. Provide the optimum receiver configuration. Determine the symbol error probability as a
function of the average energy (Eav) of the optimum receiver for an AWGN channel. Evaluate the average symbol error
probability of this scheme in the presence of AWGN and Nakagami-m fading for m ¼ 0.5, 1, and 2, assuming that the
average SNR is 12 dB.
6.25. This problem is a continuation of Problem 1 in Chap. 2. We are again concerned with a study of the propagation of
Gaussian and LG beams over the atmospheric turbulent channel, but now taking the adaptive optics (AO) into account.
The simulation conditions are the same as in Problem 1 of Chap. 2. Figure P6.4 shows a typical wavefront sensorless
AO-based FSO link. When a collimated Gaussian beam with a beam diameter of 5 mm is transmitted over 1 km
turbulent channel, the captured on-axis intensity will be affected by turbulence accordingly. An AO system based on
stochastic parallel gradient descent (SPGD) algorithm can be used to adjust the phase profile of the deformable mirror
(square aperture with 88 mm size, 5050 pixels). If the algorithm converging rate is higher than the turbulence
changing rate, a stable and near-perfect Gaussian beam can be then recovered, in a weak turbulence regime.
6.11 Problems 385

Fig. P6.4 Wavefront sensorless AO-based turbulence mitigation

Simulate the whole AO process for the following refractive structure parameters: Cn2 ¼ 1015, 1014, and 1013
2/3
m .
(a) Plot the monitored intensity evolution over each SPGD algorithm step.
(b) When the monitored intensity converges to a certain level, plot the intensity profiles of the distorted Gaussian beam
affected by the turbulence and recovered Gaussian beam after the turbulence mitigation.
(c) Plot the intensity probability density function and compare it with the case when no AO is used.
6.26. In this problem, which is a continuation of the previous problem, you are expected to simulate the wavefront sensor
(WFS)-based AO system, shown in Fig. P6.5. Assume that WFS is based on 2-D Shack-Hartmann sensor. To represent
the atmospheric turbulence effects, you are suggested to observe at least 56 Zernike terms. To determine the local tilts,
you can use either of the equations below:
ZZ
1
s ¼ rc =kf ¼ IðrÞ∇φd 2 r
P
A
ZZ
1
¼ fRe ðEÞ∇Im ðEÞ  ∇½Re ðEÞIm ðEÞgd2 r,
P
A

where E is the reference beam complex amplitude of the electric field and I is the corresponding intensity. The
surface of deformable mirror S(ρ) can be represented as

X
N
S ð ρÞ ¼ An f n ðρ  ρn Þ,
n¼1

where An is the deflection

 of the mirror
 surface at the point of fastening of the nth actuator ρ n and fn() are the response
functions f n ðρÞ ¼ exp ρ2 =w2n , with wn being the halfwidth, ranging from 0.7d to 0.8d, where d denotes the spacing
between actuators.
By setting S(ρ) ¼ φ(ρ), repeat steps (a), (b), and (c) from Problem 6.25 and discuss the improvements with respect to
the wavefront sensorless AO. Also discuss the improvement for a different number of Zernike terms in WFS.
6.27. Repeat Problem 6.26 for LG mode with azimuthal mode index m ¼ 3 and radial index p ¼ 0.
386 6 Advanced Detection Techniques and Compensation of Channel Impairments

Photodetector
array

Lenslet
array

BS CCD
Compressing
telescope

Collimating

subsystem
Signal beam

lens

Control
Pilot beam

Deformable mirror

Receiver side adaptive optics

Fig. P6.5 Receiver side AO system

References
1. J.G. Proakis, Digital Communications (McGraw-Hill, Boston, 2001)
2. R.N. McDonough, A.D. Whalen, Detection of Signals in Noise, 2nd edn. (Academic Press, San Diego, 1995)
3. S. Haykin, Digital Communications (John Wiley & Sons, 1988)
4. S. Haykin, Communication Systems, 4th edn. (John Wiley & Sons, 2001)
5. A. Goldsmith, Wireless Communications (Cambridge University Press, 2005)
6. D. Tse, P. Viswanath, Fundamentals of Wireless Communication (Cambridge University Press, Cambridge, 2005)
7. S. Benedetto, E. Biglieri, Principles of Digital Transmission with Wireless Applications (Kluwer Academic/Plenum Publishers, New York,
1999)
8. E. Biglieri, Coding for Wireless Channels (Springer, New York, 2005)
9. I. Djordjevic, W. Ryan, B. Vasic, Coding for Optical Channels (Springer, 2010)
10. W. Shieh, W. Djordjevic, OFDM for Optical Communications (Elsevier, 2009)
11. I.B. Djordjevic, On Advanced FEC and Coded Modulation for Ultra-High-Speed Optical Transmission. IEEE Communications Surveys and
Tutorials 18(3), 1920–1951, third quarter 2016, August 19, 2016 (2016). https://doi.org/10.1109/COMST.2016.2536726
12. T.M. Duman, A. Ghrayeb, Coding for MIMO Communication Systems (John Wiley & Sons, Chichester, 2007)
13. G.L. Stüber, Principles of Mobile Communication, 3rd edn. (Springer, New York, 2012)
14. M.K. Simon, M.-S. Alouini, A unified approach to the performance analysis of digital communication over generalized fading channels.
Proceedings of the IEEE 86(9), 1860–1877 (1998)
15. M.K. Simon, M.-S. Alouini, Digital Communication over Fading Channels: A Unified Approach to Performance Analysis (John Wiley & Sons,
New York, 2000)
16. M. Cvijetic, I.B. Djordjevic, Advanced Optical Communication Systems and Networks (Boston-London, Artech House, 2013)
17. G. Jacobsen, Noise in Digital Optical Transmission Systems (Artech House, Boston, 1994)
18. Djordjevic, I. B., and Stefanovic, M. C., “Performance of optical heterodyne PSK systems with Costas loop in multichannel environment for
nonlinear second order PLL model,” IEEE J. Lightw. Technol., vol. 17, no. 12, pp. 2470-2479, Dec. 1999.
19. P.W. Hooijmans, Coherent Optical System Design. J. Wiley and Sons (1994)
20. L. Kazovsky, S. Benedetto, A. Willner, Optical Fiber Communication Systems (Artech House, Boston, 1996)
21. S.J. Savory, Digital filters for coherent optical receivers. Opt. Express 16, 804–817 (2008)
22. I.B. Djordjevic, L. Xu, T. Wang, PMD compensation in multilevel coded-modulation schemes with coherent detection using BLAST algorithm
and iterative polarization cancellation. Opt. Express 16(19), 14845–14852, Sept. 15 (2008)
23. I.B. Djordjevic, L. Xu, T. Wang, PMD compensation in coded-modulation schemes with coherent detection using Alamouti-type polarization-
time coding. Opt. Express 16(18), 14163–14172, 09/01/2008 (2008)
References 387

24. I.B. Djordjevic, M. Arabaci, L. Minkov, Next generation FEC for high-capacity communication in optical transport networks. IEEE/OSA
J. Lightw. Technol. accepted for publication. (Invited Paper.)
25. I.B. Djordjevic, L. Xu, T. Wang, Beyond 100 Gb/s optical transmission based on polarization multiplexed coded-OFDM with coherent
detection. IEEE/OSA J. Opt. Commun. Netw. 1(1), 50–56, June 2009 (2009)
26. J.G. Proakis, D.G. Manolakis, Digital Signal Processing: Principles, Algorithms, and Applications, 4th edn. (Prentice-Hall, 2007)
27. C.-C. Tseng, Design of IIR digital all-pass filters using least pth phase error criterion. IEEE Trans. Circuits Syst. II, Analog Digit. Signal Process
50(9), 653–656, Sep. 2003 (2003)
28. G. Goldfarb, G. Li, Chromatic dispersion compensation using digital IIR filtering with coherent detection. IEEE Photon. Technol. Lett 19(13),
969–971, July 1, 2007 (2007)
29. E. Ip, A. Pak, T. Lau, D.J.F. Barros, J.M. Kahn, Coherent detection in optical fiber systems. Opt. Express 16(2), 753–791, 21 January 2008
(2008)
30. E. Ip, J.M. Kahn, Digital equalization of chromatic dispersion and polarization mode dispersion. J. Lightw. Technol. 25, 2033–2043, Aug. 2007
(2007)
31. M. Kuschnerov, F.N. Hauske, K. Piyawanno, B. Spinnler, M.S. Alfiad, A. Napoli, B. Lankl, DSP for coherent single-carrier receivers. J. Lightw.
Technol 27(16), 3614–3622, Aug. 15, 2009 (2009)
32. J.G. Proakis, Partial response equalization with application to high density magnetic recording channels, in Coding and Signal Processing for
Magnetic Recording Systems, ed. by B. Vasic, E. M. Kurtas, (CRC Press, Boca Raton, 2005)
33. I. Lyubomirsky, Optical duobinary systems for high-speed transmission, in Advanced Technologies for High-Speed Optical Communications,
ed. by L. Xu, (Research Signpost, 2007)
34. N. Alic, G.C. Papen, R.E. Saperstein, L.B. Milstein, Y. Fainman, Signal statistics and maximum likelihood sequence estimation in intensity
modulated fiber optic links containing a single optical preamplifier. Opt. Express 13, 4568–4579, 13 June 2005 (2005)
35. G. Colavolpe, T. Foggi, E. Forestieri, G. Prati, Multilevel optical systems with MLSD receivers insensitive to GVD and PMD. IEEE/OSA J. of
Lightw. Technol. 26, 1263–1273 (2008)
36. M. Ivkovic, I.B. Djordjevic, B. Vasic, Hard decision error correcting scheme based on LDPC codes for long-haul optical transmission, in Proc.
Optical Transmission Systems and Equipment for Networking V-SPIE Optics East Conference, vol. 6388, (Boston, 2006),
pp. 63880F.1–63880F.7, 1–4 October 2006
37. Hagenauer, J., and Hoeher, P., “A Viterbi algorithm with soft-decision outputs and its applications,” in Proc. IEEE Globecom Conf., Dallas,
pp. 1680-1686, Nov. 1989.
38. L.R. Bahl, J. Cocke, F. Jelinek, J. Raviv, Optimal decoding of linear codes for minimizing symbol error rate. IEEE Trans. Inform. Theory IT-20,
284–287 (Mar. 1974)
39. H. Wymeersch, M.Z. Win, Soft electrical equalization for optical channels. Proc. ICC’08, 548–552, 19–23 May 2008 (2008)
40. C. Douillard, M. Jézéquel, C. Berrou, A. Picart, P. Didier, A. Glavieux, Iterative correction of intersymbol interference: turbo equalization’. Eur.
Trans. Telecommun. 6, 507–511 (1995)
41. M. Tüchler, R. Koetter, A.C. Singer, Turbo equalization: principles and new results. IEEE Trans. Commun. 50(5), 754–767 (May 2002)
42. M. Jäger, T. Rankl, J. Spiedel, H. Bulöw, F. Buchali, Performance of turbo equalizers for optical PMD channels. IEEE/OSA J. Lightw. Technol.
24(3), 1226–1236 (Mar 2006)
43. I.B. Djordjevic, L.L. Minkov, L. Xu, T. Wang, Suppression of fiber nonlinearities and PMD in coded-modulation schemes with coherent
detection by using turbo equalization. IEEE/OSA J. Opt. Commun. Netw. 1(6), 555–564 (Nov 2009)
44. M.H. Hayes, Statistical Digital Signal Processing and Modeling (John Wiley & Sons, Inc, 1996)
45. A. Haykin, Adaptive Filter Theory, 4th edn. (Pearson Education, 2003)
46. D.G. Manolakis, V.K. Ingle, S.M. Kogon, Statistical and Adaptive Signal Processing (McGraw-Hill, Boston, 2000)
47. N. Seshadri, Joint data and channel estimation using fast blind trellis search techniques. IEEE Trans. Comm., col. COMM-42, 1000–1011,
Feb/Mar/Apr (1994)
48. D.N. Godard, Self-recovering equalization and carrier tracking in two-dimensional data communication systems. IEEE Trans. Comm 28(11),
1867–1875 (Nov. 1980)
49. Y. Sato, A method of self-recovering equalization for multilevel amplitude-modulation systems. IEEE Tran. Comm. 23(6), 679–682 (Jun 1975)
50. A. Benveniste, M. Goursat, G. Ruget, Robust identification of a nonminimum phase system. IEEE Trans. Auto. Control AC-25, 385–399 (June
1980)
51. A. Goldsmith, S.-G. Chua, “Adaptive coded modulation for fading channels,” IEEE Trans. Comm., vol. 46, pp. 595-601, May 1998.
52. I.B. Djordjevic, Adaptive modulation and coding for free-space optical channels. IEEE/OSA J. Opt. Commun. Netw. 2(5), 221–229 (May 2010)
53. I.B. Djordjevic, G.T. Djordjevic, On the communication over strong atmospheric turbulence channels by adaptive modulation and coding.
Optics Express 17(20), 18250–18262, September 25, 2009 (2009)
54. I.B. Djordjevic, On the Irregular Nonbinary QC-LDPC-Coded Hybrid Multidimensional OSCD-Modulation Enabling Beyond 100 Tb/s Optical
Transport. IEEE/OSA J. Lightwave Technol 31(16), 2969–2975, Aug.15, 2013 (2013)
55. D. Zou, I.B. Djordjevic, FPGA-based rate-adaptive LDPC-coded modulation for the next generation of optical communication systems. Optics
Express 24(18), 21159–21166 (2016)
56. D. Zou, I.B. Djordjevic, FPGA-based rate-compatible LDPC codes for the next generation of optical transmission systems. IEEE Photonics
Journal 8(5), 7906008-1–7906008-8 (Oct. 2016). https://doi.org/10.1109/JPHOT.2016.2602063
57. D.J.C. MacKay, Fountain codes. IEE Proc.-Commun 152, 1062–1068 (Dec 2005)
58. A. Shokrollahi, Raptor codes. IEEE Inf. Theory 52, 2551–2567 (June 2006)
59. I.B. Djordjevic, Spatial-domain-based hybrid multidimensional coded-modulation schemes enabling multi-Tb/s optical transport. IEEE/OSA
J. Lightw. Technol 30(14), 2315–2328, July 15, 2012 (2012)
60. H.S. Wang, N. Moayeri, Finite-state Markov channel-a useful model for radio communication channels. IEEE Trans. Veh. Technol. 44(1),
163–171 (1995)
61. G.D. Forney, R.G. Gallager, G.R. Lang, F.M. Longstaff, S.U. Qureshi, Efficient modulation for band-limited channels. IEEE J. Sel. Areas
Comm SAC-2(5), 632–647 (1984)
388 6 Advanced Detection Techniques and Compensation of Channel Impairments

62. G.D. Forney, Coset codes-part I: introduction and geometrical classification. IEEE Trans. Infrom. Theory 34(5), 1123–1151 (Sept 1988)
63. R.D. Nowak, B.D. Van Veen, Volterra filter equalization: a fixed point approach. IEEE Tran. Sig. Proc. 45(2), 377–388 (1997)
64. M.C. Jeruchim, P. Balaban, K.S. Shanmugan, Simulation of Communication systems: Modeling, Methodology, and Techniques, 2nd edn.
(Kluwer Academic/Plenum Pub, N. York, 2000)
65. M. Nazarathy, R. Weidenfeld, Nonlinear impairments in coherent optical OFDM and their mitigation, in Impact of Nonlinearities on Fiber Optic
Communication, ed. by S. Kumar, (Springer, Mar. 2011), pp. 87–175
66. F.P. Guiomar, J.D. Reis, A.L. Teixeira, A.N. Pinto, Mitigation of intra-channel nonlinearities using a frequency-domain Volterra series
equalizer, in Proc. ECOC 2011, (Paper Tu.6.B.1, 18–22 September 2011, Geneva, 2011)
67. K.V. Peddanarappagari, M. Brandt-Pearce, Volterra series transfer function of single-mode fibers. J. Lightw. Technol. 15, 2232–2241 (Dec
1997)
68. B. Xu, M. Brandt-Pearce, Modified Volterra series transfer function method. IEEE Photon. Technol. Lett. 14(1), 47–49 (2002)
69. F.P. Guiomar, J.D. Reis, A.L. Teixeira, A.N. Pinto, Digital post-compensation using Volterra series transfer function. IEEE Photon. Technol.
Lett 23(19), 1412–1414, October 1, 2011 (2011)
70. Y. Gao, F. Zgang, L. Dou, Z. Chen, A. Xu, Intra-channel nonlinearities mitigation in pseudo-linear coherent QPSK transmission systems via
nonlinear electrical equalizer. Opt. Commun. 282, 2421–2425 (2009)
71. L. Liu, L. Li, Y. Huang, K. Cui, Q. Xiong, F.N. Hauske, C. Xie, Y. Cai, Intrachannel nonlinearity compensation by inverse Volterra series
transfer function. J. Lightw. Technol 30(3), 310–316, Feb. 1, 2012 (2012)
72. E. Ip, J.M. Kahn, Compensation of dispersion and nonlinear impairments using digital backpropagation. J. Lightw. Technol. 26(20), 3416–3425
(2008)
73. E. Ip, J.M. Kahn, Nonlinear Impairment Compensation using Backpropagation, in Optical Fibre, New Developments, ed. by C. Lethien,
(In-Tech, Vienna Austria, Dec. 2009)
74. G.P. Agrawal, Nonlinear Fiber Optics, 3rd edn. (Academic Press, San Diego)
75. I.B. Djordjevic, Quantum Information Processing and Quantum Error Correction: An Engineering Approach (Elsevier/Academic Press, Apr.
2012)
76. R.G. Priest, Analysis of fiber interferometer utilizing 3  3 fiber coupler. IEEE Journal of Quantum Electronics QE-18(10), 1601–1603 (1982)
77. M.R. Paiam, R.I. MacDonald, Design of phased-array wavelength division multiplexers using multimode interference couplers. Applied Optics
36(21), 5097–5108 (1997)
78. M. Seimetz, C.M. Weinert, Options, feasibility and availability of 24 90o hybrids for coherent optical systems. IEEE Journal of Lightwave
Technology 24(3), 1317–1322 (2006)
79. S. Norimatsu et al., An 8 Gb/s QPSK optical homodyne detection experiment using external cavity laser diodes. IEEE Photonics Technology
Letters 4(7), 765–767 (1992)
80. S. Camatel et al., Optical phase-locked loop for coherent detection optical receiver. Electronics Letters 40(6), 384–385 (2004)
81. S. Camatel et al., 2-PSK homodyne receiver based on a decision driven architecture and a sub-carrier optical PLL, in Proceedings of Optical
Fiber Communication Conference (OFC), (Paper OTuI3, 2006)
82. A. Jesacher, A. Schwaighofer, S. Frhapter, C. Maurer, S. Bernet, M. Ritsch-Marte, Wavefront correction of spatial light modulators using an
optical vortex image. Optics Express 15(5), 5801–5808 (Apr. 2007)
83. R.W. Gerchberg, W.O. Saxton, A practical algorithm for the determination of phase from image and diffraction plane pictures. Optik 35,
237–246 (1972)
84. N. Chandrasekaran and J. H. Shapiro, “Turbulence-induced crosstalk in multiple-spatial-mode optical communication,” in Proc. CLEO 2012,
San Jose, CA, May 6–11, 2012, Paper CF3I.6.
85. V.P. Lukin, B.V. Fortes, Adaptive Beaming and Imaging in the Turbulent Atmosphere (SPIE Press, Bellingham, Washington, 2002)
86. A. Glindemann, S. Hippler, T. Berkfeld, W. Hackenberg, Adaptive optics on large telescopes. Experimental Astronomy 10(1), 5–47 (2000)
87. X. Sun, I.B. Djordjevic, M.A. Neifeld, Multiple spatial modes based QKD over marine free-space optical channels in the presence of
atmospheric turbulence. Optics Express 24(24), 27663–27673, 28 Nov 2016 (2016)
88. Z. Qu, I.B. Djordjevic, 500 Gb/s Free-Space Optical Transmission over Strong Atmospheric Turbulence Channels. Optics Letters 41(14),
3285–3288, July 15 2016 (2016)
89. I.B. Djordjevic, J. Anguita, B. Vasic, Error-Correction Coded Orbital-Angular-Momentum Modulation for FSO Channels Affected by
Turbulence. IEEE/OSA J. Lightw. Technol 30(17), 2846–2852, September 1 2012 (2012)
90. R.J. Noll, Zernike polynomials and atmospheric turbulence. J. Opt. Soc. Am. 66(3), 207–211 (1976)
91. Tatarskii VI (1967), The Effects of the Turbulent Atmosphere on Wave Propagation (Translated by Israel Program for Scientific Translations in
1971), U.S. Dept. of Commerce, National Technical Information Service, Springfield.
92. R. Cubalchini, Modal wave-front estimation from phase derivative measurements. J. Opt. Soc. Am. 69, 972–977 (1979)
93. M.A. Ealey, J.A. Wellman, Deformable mirrors: design fundamentals, key performance specifications, and parametric trades (Proc. SPIE 1543,
Active and Adaptive Optical Components, 1992), pp. 36–51. https://doi.org/10.1117/12.51167. January 13, 1992
94. R.K. Tyson, Principles of Adaptive Optics (CRC Press, Boca Raton, 2015)
95. Z. Qu, I.B. Djordjevic, M.A. Neifeld, RF-subcarrier-assisted Four-state Continuous-variable QKD Based on Coherent Detection. Optics Letters
41(23), 5507–5510, Dec. 1, 2016 (2016)
 OFDM for Wireless and Optical Communications
7

Abstract
This chapter is devoted to OFDM fundamentals and applications to wireless and optical communications. The chapter
starts with the basic of OFDM including the generation of OFDM signal by inverse FFT. After that several basic
manipulations on OFDM symbols, required to deal with multipath effects in wireless communications and dispersion
effects in fiber-optics communications, including guard time and cyclic extension as well as the windowing operation, are
described. Next, the bandwidth efficiency of OFDM is discussed. The next topic in the chapter is devoted to the OFDM
system design, description of basic OFDM blocks, and OFDM-based parallel channel decomposition, including multipath
fading radio and dispersion-dominated channel decomposition. Following the description of coherent optical OFDM
(CO-OFDM) principles, the basic estimation and compensation steps, common to both wireless and optical communica-
tion systems, are described including DFT windowing, frequency synchronization, phase estimation, and channel estima-
tion. Regarding the channel estimation, both pilot-aided and data-aided channel estimation techniques are described.
Following the differential detection in OFDM description, the focus is on various applications of OFDM in wireless
communications including digital audio broadcasting (DAB), digital video broadcasting (DVB), Wi-Fi, LTE, WiMAX,
and ultra-wideband (UWB) communication. Next, optical OFDM applications are discussed starting with description of
basic OFDM system types such as CO-OFDM and direct detection optical OFDM (DDO-OFDM) systems. Regarding the
high-speed spectrally efficient CO-OFDM systems, the polarization-division multiplexed OFDM systems and OFDM-
based superchannel optical transmission systems are described. Before, the concluding remarks, the application of OFDM
in multimode fiber links is described. The set of problems is provided after the concluding remarks.

7.1 Introduction, OFDM Basics, and Generation of Subcarriers Using Inverse FFT

7.1.1 Introduction to OFDM

Orthogonal frequency division multiplexing (OFDM) belongs to the multicarrier transmission, where a single data stream is
transmitted over a number of lower-rate orthogonal subcarriers. As the name suggests, the OFDM can be considered as a
modulation technique, a multiplexing technique, or both. This gives the OFDM flexibility that does not exist in conventional
single-carrier systems. The OFDM has been introduced by Chang [1] in his seminal paper in 1966, followed by his patent in
1970 [2]. The OFDM is already a part of numerous standards for mobile wireless systems, wireless LANs, asymmetric digital
subscriber lines, and coax cables [3, 4]. In optical communications [5–7], the OFDM occurred around 1996 [8], and the paper
on optical OFDM was sporadic [9, 10], until recent resurgence [11–16]. The OFDM is in these days intensively studied as
enabler of ultrahigh speed optical communications [15–19]. Regarding the optical communications, there are two basic
classes of OFDM systems, namely, direct detection optical (DDO) OFDM [11, 13] and coherent optical (CO) OFDM [12],
each of them with a number of variants.
There are several fundamental advantages of OFDM as compared to single-carrier systems, such as robustness against
various multipath effects in wireless communication, infrared (IR) indoor and visible light communication (VLC) channels,

# The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 389
I. B. Djordjevic, Advanced Optical and Wireless Communications Systems, https://doi.org/10.1007/978-3-030-98491-5_7
390 7 OFDM for Wireless and Optical Communications

robustness to dispersion effects (chromatic dispersion, PMD, multimode dispersion) and polarization-dependent loss (PDL)
for fiber-optics communications, adaptivity to time-varying channel conditions, flexibility for software-defined radio (SDR)
and software-defined optical transport, straightforward channel estimation, and low complexity compared to conventional
equalizer schemes. However, it is sensitive to frequency offset and phase noise, and it exhibits large peak-to-average power
ratio. Since the OFDM experts have already figured out how to deal with these drawbacks, the OFDM remains to be a hot
research topic, in particular for fiber-optics communications. In addition, given the resurgence of optical fibers that support
spatial multiplexing (FMF and FCF), the OFDM can be seen as a key enabling technology for next-generation optical
transport [20–25].

7.1.2 OFDM Basics

An OFDM signal can be generated as a sum of orthogonal subcarriers that are modulated using conventional two-dimensional
(2D) signal constellations such as PSK and QAM. The word orthogonality indicates that there exists a precise mathematical
relationship among the frequencies of the carriers in the OFDM system. Namely, each subcarrier has exactly an integer
number of cycles within the symbol interval Ts, and the number of cycles between adjacent subcarriers differs by exactly one,
so that k-th subcarrier, denoted as fk, and the l-th one, denoted as fl, are related as follows:

f k
f l ¼ n=T s , ð7:1Þ

where n is an integer. The k-th subcarrier waveform can be represented by

j2π f k t 1,
1=2 < t  1=2
sk ðt Þ ¼ rectðt=T s Þe , rectðt Þ ¼ ð7:2Þ
0, otherwise

The following principle of orthogonality between k-th and l-th (k 6¼ l) subcarrier waveforms is valid only when the
subcarriers are related by Eq. (7.1):

ZT s =2
1 sin ðπ ð f k
f l ÞT s Þ
hsl , sk i ¼ sk ðt Þsl ðt Þdt ¼
Ts π ð f k
f l ÞT s ð7:3Þ

T s =2

¼ sincðð f k
f l ÞT Þj f k
f l ¼n=T s ¼ 0,

where sinc(x) ¼ sin(πx)/(πx). Therefore, in this case the orthogonal subcarriers present the basis functions. And, as common,
an arbitrary signal can be represented in terms of projections along basis functions. In other words, by using PSK or QAM
constellation points as projections along basis functions, we can generate an arbitrary OFDM waveform as
8  
< P
> k¼N sc =2
1
k
X kþN sc =2 exp j2π t ,
T s =2 < t  T s =2
sðt Þ ¼ k¼
N sc =2 Ts ð7:4Þ
>
:
0, otherwise

where with Xk we denoted the QAM/PSK symbol imposed on k-th subcarrier and Nsc denotes the number of subcarriers. Based
on this interpretation, we can represent the OFDM modulator as shown in Fig. 7.1a. The QAM serial data stream is
demultiplexed into Nsc parallel streams, which are multiplied with different subcarrier basis functions to generate the
OFDM waveform, shown in Fig. 7.1b. The real and imaginary part of this OFDM signal after digital-to-analog conversions
can be used as RF inputs of I/Q modulator. Therefore, this OFDM modulator is consistent with synthesizer, described in
Chap. 5. In this particular configuration, the OFDM principle is employed to perform modulation. On the other hand, if Nsc
independent QAM streams for different sources are multiplied with corresponding subcarrier basis functions before being
combined together, the OFDM synthesizer will serve as multiplexer.
The demodulation of l-th subcarrier QAM symbol can be performed by down-conversion by l/Ts, in other words by
correlation receiver from Chap. 5 as follows:
7.1 Introduction, OFDM Basics, and Generation of Subcarriers Using Inverse FFT 391

Fig. 7.1 The OFDM signal: (a) exp[j Nsc t /Ts ]

waveform generation and (b)
Nyquist overlapping subcarriers

…
QAM OFDM
S/P
data converter signal
exp[-j (Nsc -2)t/Ts]

(a) (b)

ZT s =2   N scX
=2
1 
1 l k
exp
j2π t X kþN s =2 exp j2π t dt
Ts T Ts
k¼
N sc =2

T s =2

X
N sc =2
1 ZT s =2   ð7:5Þ
1 k
l
¼ X kþN sc =2 exp j2π t dt ¼ X lþN s =2
Ts T
k¼
N sc =2

T s =2
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
T s , k¼l ð0, k6¼lÞ

Therefore, thanks to the orthogonality principle, the QAM symbol transmitted over l-th subcarrier has been demodulated.
However, in the presence frequency offset, the initial orthogonality of subcarriers will be violated, and there will be a crosstalk
among basis functions (subcarriers), commonly referred to as inter-carrier interference (ICI).

7.1.3 Generation of OFDM Signals

In order to perform transition in DT domain, let us sample the OFDM CT signal, given by Eq. (7.4), with a sampling interval
T ¼ Ts/N, where N is an integer of the form 2n (n > 1). The samples can be expressed as
8  
>
< P
k¼N sc =2
1
k
X kþN sc =2 exp j2π mT , T s =2 < t  T s =2
sðmT Þ ¼ k¼
N sc =2 Ts ð7:6Þ
>
:
0, otherwise

Since Ts/T ¼ N, we can rewrite the previous equation as

8
h i
< P
> k¼N sc =2
1
2π
X kþN sc =2 exp j km , T s =2 < t  T s =2
sðmT Þ ¼ k¼
N sc =2 N ð7:7Þ
>
:
0, otherwise

which represents the inverse discrete Fourier transform (IDFT) of input QAM sequence. The IDFT can be efficiently
implemented by inverse fast Fourier transform (IFFT). This discrete-time implementation is an important reason of high
appealing of OFDM scheme. The DT implementation is very flexible as several subcarriers can be used for channel
estimation. In addition, the radix-2 and radix-4 algorithms can be used to reduce the complexity of IDFT. Namely, in
radix-2 algorithm an N-point IFFT requires (N/2)log2N complex multiplications, while radix-4 algorithm requires only (3 N/8)
(log2N-2) complex multiplications or phase rotations. On the other hand, the regular IDFT requires N2 complex
multiplications. Unfortunately, the samples used to calculate IFFT are not sufficient for practical implementation since the
use of digital-to-analog conversion (DAC) is needed. To avoid aliasing effect, the oversampling must be used. This is
typically performed by zero padding. The zero samples should be added in the middle of QAM data sequence rather than
appending them at the edges. With this approach we will ensure that data QAM points are mapped around d.c. component,
while zero samples will be mapped close to plus/minus sampling rate. After the IFFT calculation and parallel-to-serial
conversion, DAC process is performed.
392 7 OFDM for Wireless and Optical Communications

The better spectral efficiency of OFDM, as compared to single-carrier systems, can be justified as follows. From Eq. (7.4)
we conclude that the spectrum of a single OFDM symbol, shown in Fig. 7.1b, can be obtained as a convolution of the
spectrum of a square pulse having duration Ts and given by sinc( fTs), with a sum of Dirac pulses located at the subcarrier
frequencies:

k¼NP
sc =2
1
 
k
Sðf Þ ¼ FTfsðtÞg ¼ FT rectðt=T s Þ X kþN sc =2 exp j2π t
k¼
N sc =2 T s

k¼NP sc =2
1
 
1
¼ sinc ð f T s Þ  X kþN sc =2 δ f
k ð7:8Þ
k¼
N sc =2 T s

k¼NPsc =2
1
  
1
¼ X kþN sc =2 sinc f
k T :
k¼
N sc =2 Ts s

where FT denotes Fourier transform operator. Therefore, the overlapping sinc spectra of individual subcarriers (OFDM
spectrum) satisfy the Nyquist criterion for an intersymbol interference (ISI)-free signaling [26, 27]. Even though the
neighboring subcarrier pulses partially overlap at the observed OFDM subcarrier location, the neighboring subcarrier pulse
shapes cross zero at that location so that there is no inter-carrier interference. The key difference as compared to zero-ISI
signaling is that the pulse shape of OFDM subcarrier is observed in the frequency-domain rather than in the time-domain.
Since the partial overlapping of neighboring subcarrier pulses is allowed, the bandwidth efficiency can be improved.

7.2 Guard Time, Cyclic Extension, and Windowing

In this section we describe several basic manipulations on OFDM symbol required to deal with dispersion effects in optical
communications and multipath effects in wireless communications (including indoor optical wireless and IR
communications), as well as out-of-band spectrum reduction.

7.2.1 Guard Time and Cyclic Extension

One of the key advantages of OFDM is that it can straightforwardly be used to deal with various dispersion and multipath
effects. To reduce the ISI among OFDM symbols, the guard time is introduced for every OFDM symbol. The guard time is
typically chosen to be longer than the expected dispersion/multipath widening so that dispersion/multipath components from
one symbol are not able to interfere with the neighboring symbols. The guard time could be just an empty time interval;
however, the problem of inter-carrier interference (ICI) will arise in that case as the subcarrier components cease to be
orthogonal. In order to eliminate the ICI introduced by dispersive optical and multipath wireless channels, the OFDM symbol
is commonly cyclically extended in the guard time to restore the orthogonality of the subcarrier components. The symbol
duration, after cyclic extension, is determined as Ts ¼ TG + TFFT, where TG is the guard interval duration, while TFFT is the
duration of effective portion of OFDM symbol (FFT part) related to subcarrier spacing Δfsc by TFFT ¼ 1/Δfsc. The cyclic
extension can be performed by repeating the last NG/2 samples (corresponding to TG/2) of the FFT frame (of duration TFFT
with NFFT samples) as a prefix and repeating the first NG/2 samples (out of NFFT) as a suffix, which is illustrated in Fig. 7.2.
Another alternative is to use only either prefix or suffix of duration TG.
In fiber-optics communications, for complete elimination of ISI, the total delay spread Δt due to chromatic dispersion and
differential group delay (DGD) should be shorter than the guard interval [5]:

c
Δt ¼ jβ2 jLtot Δω þ DGDmax ¼ jDt jN FFT Δ f sc þ DGDmax  T G , ð7:9Þ
f2

where Dt is the accumulated dispersion, c is the speed of the light, f is the central frequency (typically set to 193.1 THz), Ltot is
the total transmission distance, DGDmax is the maximum DGD, and β2 is the second-order group velocity dispersion (GVD)
7.2 Guard Time, Cyclic Extension, and Windowing 393

NG /2 samples NG /2 samples
Original NFFT samples

Preffix
Suffix

OFDM symbol after cyclic extension, NFFT + NG samples

Fig. 7.2 The illustration of cyclic extension procedure

parameter related to chromatic dispersion parameter D by β2 ¼
(2πc/λ2)D, with λ being the operating wavelength. The
received electrical field, in the presence of chromatic dispersion, can be written as


 NX
FFT
j2π f k t jϕCD ðk Þ
E s ¼ exp j 2π f LD t þ ϕPN,S Xk e e , ð7:10Þ
k¼1

where ϕCD(k) denotes the phase factor of k-th subcarrier

ω2k jβ2 jL 4π 2 f 2k λ2LD πc

ϕCD ðkÞ ¼ ¼ DL ¼ 2 Dt f 2k , ð7:11Þ
2 2 2πc f LD

with fk being the k-th subcarrier frequency (corresponding angular frequency is ωk ¼ 2πfk), fLD being the transmitting laser
frequency (the corresponding wavelength is λk), Dt being the total dispersion (Dt ¼ DL, D dispersion coefficient, L the fiber
length), and c being the speed of light. The output of the OFDM balanced coherent optical detector can be represented as

X
N FFT

vðt Þ ’ RPIN RL ef j½2π ð f LD
f LO ÞtþϕPN,S
ϕPN,LO g

Xk e j2π f k t
e jϕCD ðk Þ
þ N ðt Þ, ð7:12Þ
k¼1

where fLO denotes the carrier frequency of the local laser oscillator, RPIN is the photodetector responsivity, RL is feedback
resistance in trans-impedance amplifier configuration, while ϕS,PN and ϕL,PN represent the laser phase noise of the transmitting
laser and local laser oscillator, respectively. These two noise sources are commonly modeled as Wiener-Lévy process [7],
which is a zero-mean Gaussian process with variance 2π(ΔνS + ΔνL)|t|, where ΔνS and ΔνL are the laser linewidths of
transmitting laser and local oscillator, respectively, while N¼NI-jNQ represent the noise process, mostly dominated by ASE
noise. The received symbol Yk of k-th subcarrier can be represented as


Y k ¼ X k exp j ϕPN,S
ϕPN,LO exp ½ jϕCD ðk Þ þ N k , N k ¼ N k,I þ N k,Q ð7:13Þ

where Nk represents a circular Gaussian process due to ASE noise expressed in a complex form. The estimation of the
transmitted symbol on k-th subcarrier can be done by relation


e k ¼ Y k exp
j ϕPN,S
ϕPN,LO exp ½
jϕCD ðkÞ,
X ð7:14Þ

where the chromatic dispersion phase factor of k-th subcarrier ϕCD(k) is estimated by training-based channel estimation while
the phase factor ϕPN, S
ϕPN, LO is determined by pilot-aided channel estimation. The additional details of various OFDM
channel estimation techniques can be found in Sec. 7.5. Therefore, with proper guard interval selection, we can decompose the
dispersive channel into NFFT parallel channels, which will be described later with additional details. To include the
394 7 OFDM for Wireless and Optical Communications

polarization mode dispersion (PMD) in the model, we employ the Jones vector representation. Namely, the electric field of
plane linearly polarized waves propagating in the z-direction can be presented as

Eðz, tÞ ¼ EH ðz, tÞ þ EV ðz, tÞ

ð7:15Þ
¼ eH E0H cos ðωt
kzÞ þ eV E 0V cos ðωt
kz þ δÞ,

where k ¼ 2π/λ is the magnitude of the wave propagation vector (λ is the operating wavelength) and δ is the relative phase
difference between the two orthogonal polarizations. With EH and EV, we denoted H- and V-polarization states, respectively.
If using the Jones vector representation of polarization wave [28], we can write that

  " pffiffiffiffiffiffiffiffiffiffiffiffiffi #
E H ðt Þ 1
pr jðωt
kzÞ
E ðt Þ ¼ ¼ E pffiffiffiffiffi jδ e , ð7:16Þ
E V ðt Þ pr e

where pr is the power splitting ratio between states of polarizations (SOPs) while the complex phasor term is typically omitted.
From Eq. (7.16), the equivalent channel model can be expressed as

  " ðH Þ
# " ðH Þ #
ðH Þ
Yk Xk 
 Nk
ðV Þ ¼ Uð f k Þ ðV Þ
exp j ϕPN,S
ϕPN,LO exp ½ jϕCD ðk Þ þ ðV Þ
, ð7:17Þ
Yk
k Xk Nk

where U( fk) denotes the Jones matrix of fiber link, defined as

 
U HH ð f k Þ U HV ð f k Þ
Uð f k Þ ¼ : ð7:18Þ
U VH ð f k Þ U VV ð f k Þ

In wireless communications, to deal with multipath effects, the guard interval must be two to four times longer than the root
mean square (RMS) of the multipath delay spread.
We will now provide the interpretation of cyclic extension from digital signal processing (DSP) [7, 29, 30] point of view.
The N-point discrete Fourier transform (DFT) of discrete-time sequence x(n) (0  n  N-1) and corresponding inverse DFT
(IDFT) are defined as

X
N
1
DFTfxðnÞg ¼ X k ¼ x½ne
j2πnk=N , 0  k  N
1
n¼0
ð7:19Þ
1 X
N
1
j2πnk=N
IDFTfX k g ¼ xðnÞ ¼ X e , 0nN
1
N k¼0 k

The output y(n) of a linear time-invariant (LTI) system can be obtained by convolution of the input x(n) and LTI dispersive
channel impulse response h(n):
X
y ð nÞ ¼ x ð nÞ  hð nÞ ¼ hð nÞ  x ð nÞ ¼ hðkÞxðn
kÞ: ð7:20Þ
k

On the other hand, the circular convolution of x(n) and h(n) is defined as
X
yðnÞ ¼ xðnÞ  hðnÞ ¼ hðkÞxðn
kÞN ,
k ð7:21Þ
ðn
kÞN  ðn
kÞmod N,

where with (n-k)N we denoted the periodic extension of x(n-k), with period N. Therefore, the resulting sequence obtained by
circular convolution is periodic with a period of N. DFT of circular convolution of x(n) and h(n) is product of corresponding
DFTs of individual terms:
7.2 Guard Time, Cyclic Extension, and Windowing 395

Copy-paste the last L symbols as prefix

x(N-L) x(N-L+1) … x(N-1) x(0) x(1) x(2) … x(N-L-2) x(N-L-1) x(N-L) x(N-L+1) … x(N-1)
Cyclic prefix The original sequence of length N

Fig. 7.3 The illustration of cyclic prefix creation for FFT-based OFDM

DFTfyðnÞ ¼ xðnÞ  hðnÞg ¼ X k H k , 0  k  N
1: ð7:22Þ

In this case, the channel output is not a circular but rather a linear convolution of the channel input and its impulse response.
However, the linear convolution can be converted to circular by adding a cyclic extension. The cyclic extension can be
inserted as a prefix or suffix or can be split into prefix and suffix portions. The cyclic extension of {x(n)}, inserted as a prefix,
is defined by {x(N-L ),. . .,x(N-1)}, where L presents the last L samples of sequence, as illustrated in Fig. 7.3.
When the cyclic extension is appended as a prefix (to the beginning of the sequence x(n)), the convolution of the resulting
sequence and impulse response h(n) yields to the circular convolution:

X
L X
L
yðnÞ ¼ xcircular ðnÞ  hðnÞ ¼ hðkÞxcircular ðn
kÞ ¼ hðk Þxðn
k ÞN
k¼0 k¼0
ð7:23Þ
¼ xðnÞ  hðnÞ,

where with xcircular(n) we denoted the cyclic extended sequence. The DFT of circular convolution of x(n) and h(n) is given as

DFTfyðnÞ ¼ xðnÞ  hðnÞg ¼ X k H k , 0  k  N
1: ð7:24Þ

The estimation of the input sequence, denoted here as exðnÞ, can be done by



Yk DFTfyðnÞg
exðnÞ ¼ IDFT ¼ IDFT , ð7:25Þ
Hk DFTfhðnÞg

where Hk can be obtained by pilot-aided channel estimation [5]. The function y(n), used in Eq. (7.25), has the length N + L.
However, the first L samples in y(n), (where n ¼
L,. . .,-1) are not required to recover the original sequence x(n) (where
n ¼ 0,1,. . .,N-1). By adding the cyclic prefix, we have solved the problem of circular convolution, but we have introduced an
overhead of L/N, while the overall data rate has been reduced by N/(L + N ) times. Instead of using DFT/IDFT in OFDM, we
can apply FFT/IFFT. The length of sequence N in discussion above will correspond to NFFT. In OFDM, we add the cyclic
prefix to every OFDM symbol. As a summary, with the help of cyclic prefix, we can simultaneously address the problems of
both ISI and circular convolution.

7.2.2 Windowing

Another important procedure with respect to OFDM is the windowing, which is illustrated in Fig. 7.4. The purpose of the
windowing is to reduce the out-of band spectrum, which is important for multiband coherent optical (CO) OFDM
applications, often known as superchannel optical OFDM structure. The windowing is also important in wireless communi-
cation to reduce out-of-band spectrum so that the neighboring wireless channels can be placed closer. A commonly used
window type has the raised cosine shape:
8
> 1
>
> ½1
cos πðt þ T win þ T G =2Þ=T win ,
T win
T G =2  t <
T G =2
<2
wðtÞ ¼ 1:0,
T G =2  t < T FFT þ T G =2 ð7:26Þ
>
>
>1
: ½1
cos πðt
T FFT Þ=T win , T FFT þ T G =2  t < T FFT þ T G =2 þ T win
2
396 7 OFDM for Wireless and Optical Communications

TG/2 TFFT TG/2

Effective part

Twin

T=TFFT+TG+Twin

OFDM symbol duration

kT

Fig. 7.4 The illustration of the windowing operation

where TG is the guard interval duration and Twin is the duration of windowing interval. The window shape expressed above
can be represented in digital signal processing by various digital functions (also known just as windows), such as Hamming,
Hanning, Bartlett, etc. [28]. The windowing interval needs to be carefully chosen not to violate the orthogonality principle
since a partial overlap of neighboring OFDM is allowed in practice. Please notice that digital filtering techniques can be used
instead of windowing, to reduce the out-of-band spectrum. However, the digital filtering requires at least several
multiplications per sample, while windowing needs several multiplications per symbol, which means that the windowing is
simpler to implement. The OFDM signal after windowing can be represented as

X
1 X
N FFT =2
1
j2π T k
ðt
nT s Þ
sB ðt Þ ¼ wðt
nT s Þ X n,k  e FFT , ð7:27Þ
n¼
1 k¼
N FFT =2

where we used the index n to denote the n-th OFDM symbol while index k denotes the k-th OFDM subcarrier. For single-band
OFDM systems, the windowing operation can be omitted. (In Eq. (7.27), subscript “B” stands for the baseband signal.)
In summary, the OFDM signal can be generated as follows. The NQAM input QAM symbols are zero-padded to obtain NFFT
input samples used to calculate the IFFT. The last NG/2 samples of the IFFT output are inserted at the start of the OFDM
symbol, while the first NG/2 samples are appended at the end. The OFDM symbol is multiplied by corresponding windowing
function. After that, the real and imaginary parts are separated and digital-to-analog conversion performed on both the real and
imaginary parts. Finally, after amplification in the driver, the real and imaginary parts are used as inputs to I and Q ports of an
I/Q modulator.

7.3 Bandwidth Efficiency of OFDM

If we assume that Nsc subcarriers are used, each having Rsc symbol rate, the total symbol rate Rs for CO-OFDM and wireless
OFDM systems can be calculated as

Rs ¼ N sc Rsc : ð7:28Þ

The bandwidth occupied by OFDM symbol, without employing the windowing operation, can be expressed as

N sc
1
BW OFDM ¼ þ 2Rsc , ð7:29Þ
T FFT

where TFFT is the effective portion of OFDM symbol (carrying data), as shown in Fig. 5.39. Since the subcarrier spacing is
given by Δfsc ¼ 1/TFFT, the bandwidth occupied by subcarriers equals (Nsc-1)Δfsc. On the other hand, the bandwidth portion
7.4 OFDM Parameter Selection, OFDM Building Blocks, and Parallel Channel Decomposition 397

Leave the guard interval empty on transmitter side

Copy-paste the samples of received trail as prefix

on receiver side

Fig. 7.5 The illustration of the zero-prefix OFDM

due to finite rise and falling edges of the first and last subcarriers is given by 2Rsc. Therefore, the overall bandwidth efficiency
of CO-OFDM systems can be expressed as

bN sc Rsc bN sc Rsc
ρ¼2 ¼2 , ð7:30Þ
BW OFDM 2Rsc þ ðN sc
1Þ=T FFT

where b denotes the number of bits per QAM symbol while factor 2 originates from polarization-division multiplexing
(PDM). In wireless communications, typically, the PDM is not used so that the factor 2 in (7.30) should be omitted. The
subcarrier spacing can be determined as Δfsc ¼ 1/(Ts-TG), and it is system design dependent. The guard interval is strongly
dependent on the amount of dispersion/multipath fading the system needs to compensate for. Therefore, the true bandwidth
efficiency can be determined only if all details of OFDM system design are known. Nevertheless, the following rule of thumb
can be used for an initial study. Typically, the ratio Rs /BCO-OFDM is around 8/9 so that the bandwidth efficiency is about
(1.778b) bits/s per Hz [5]. As an illustration, when all subcarriers carry 16-QAM symbols, the bandwidth efficiency is ~7.112
bits/s/Hz. Notice that transmission of cyclic prefix, in addition of reducing the bandwidth efficiency, also reduces the power
efficiency.
The power efficiency of OFDM can be improved by using all-zero prefix, which is illustrated in Fig. 7.5. Namely, on the
transmitter side we leave the guard intervals empty. The optical channel will introduce the trails in guard intervals due to
chromatic dispersion and PMD. These trails need to be “copy-pasted” as prefix of OFDM symbols before OFDM demodula-
tion takes place. In principle, there is no difference in dispersion compensation capabilities of this scheme as compared to
cyclic prefix in CO-OFDM systems. However, the power efficiency can be improved (N + L )/N times. The drawback of this
scheme is that ASE noise gets approximately doubled, as the noise from trail is transferred to already noisy prefix portion of
OFDM symbol. On the other hand, the bandwidth efficiency can be improved through the so-called no-guard interval OFDM
[17, 18]. The no-guard interval OFDM, however, requires the use of conventional linear equalizer to compensate for
distortions introduced by dispersion effects.

7.4 OFDM Parameter Selection, OFDM Building Blocks, and Parallel Channel Decomposition

7.4.1 OFDM Parameter Selection

The selection of various OFDM parameters is a trade-off between various, very often conflicting requirements. In wireless
communications, the main requirements to start with include the bandwidth, bit rate, and multipath delay spread. The
multipath delay spread directly dictates the guard time: two to four times RMS of delay spread. The symbol duration is
typically chosen to be at least five times the guard time, to ensure 1 dB SNR loss because of the guard time [3, 4]. The number
of subcarriers can be determined as the ratio of the required – 3 dB bandwidth and the subcarrier spacing. The subcarrier
spacing is determined as the inverse of symbol duration minus the guard time, i.e., Δfsc ¼ 1/(Ts-TG). Alternatively, the number
of subcarriers may be determined as the ratio of required bit rate and the bit rate per subcarrier [3, 4].
Example. Let us design the OFDM system, for wireless communication applications, with required bit rate of 40 Mbps and
tolerable multipath delay spread of 100 ns. Let us assume that the available channel bandwidth is 25 MHz. The guard time is
chosen to be four times the multipath delay spread, which is 400 ns. The OFDM symbol duration can be chosen to be 6  the
guard time, which is 2.4 μs. The subcarrier spacing is determined as 1/(2.4–0.4) μs ¼ 500 kHz. The number of bits per OFDM
symbol is determined as the required bit rate divided by the OFDM symbol rate ¼ 40 Mb/s / (1/2.4 μs) ¼ 96 bits per OFDM
symbol. When 8-QAM is used with 2/3 rate coding, the required number of subcarriers will be 48, with 3  2/3 ¼ 2
398 7 OFDM for Wireless and Optical Communications

information bits per subcarrier symbol. The bandwidth occupied will be 48  500 kHz ¼ 24 MHz, which is smaller than the
available bandwidth (25 MHz). In this particular example, the 64-point radix-4 FFT/IFFT can be used, leaving 16 unused
subcarriers to provide the oversampling. Additional requirement is that an integer number of samples must be used within
FFT/IFFT interval and in the symbol interval. By choosing the number of samples per symbol to be 72, the sampling rate will
be 72/2.4 μs ¼ 30 MHz. Finally, the FFT interval will be 64/30 MHz ¼ 2.134 μs.

7.4.2 OFDM Building Blocks

The generic OFDM system suitable for wireless applications is shown in Fig. 7.6. The binary data sequence is first LDPC or
turbo encoded, depending on application of interest, and the encoder output is written into interleaver. For turbo coding
s-random interleaver can be used, while for LDPC coding the block-interleaver can be used. For block-interleaver, log2(M),
where M is signal constellation size, codewords are written in row-wise fashion interleaver. Notice that arbitrary
two-dimensional (2D) constellation can be used. Then log2(M ) bits are taken into column-wise fashion from the interleaver
and used to select a point from 2D constellation such as QAM, by applying appropriate mapping rule. The QAM symbol data
stream is converted from serial to parallel form, and with the help of subcarrier mapper, different subcarriers are loaded with
different QAM symbols. The inverse FFT is then applied, followed by the cyclic extension and windowing operations. After
digital-to-analog conversion (DAC), the corresponding RF signal is up-converted, amplified, and transmitted by using a
transmit antenna. On the receiver side, after low-noise amplification, down-conversion, and analog-to-digital conversion
(ADC), the timing and frequency-offset compensation are performed. On such obtained symbol sequence, with proper timing
and frequency-offset compensated, the cyclic removal step is applied, followed by serial-to-parallel (S/P) conversion. Further,
FFT demodulation is performed followed by symbol detection and QAM demapping. Further, deinterleaving is preformed,
followed by turbo or LDPC decoding. Deinterleaving and turbo/LDPC decoding are typically jointly performed.
On the other hand, the generic CO-OFDM system architecture (suitable for optical communications) is shown in Fig. 7.7.
The QAM data stream, obtained after encoding of binary data stream and performing the corresponding binary-to-QAM
mapping, arrives as serial symbol flow to OFDM RF transmitter, as shown in Fig. 7.7a. The symbols in QAM stream belong to
the set of constellation points {0,1,. . ., M-1}, where M is the QAM signal constellation size. When nonbinary coding is used,
the codeword symbols can be interpreted as QAM symbols, thus providing that encoder nonbinary field size is the same as the
size of the QAM constellation. The serial QAM stream is converted into parallel stream in P/S converter (or DEMUX), with
Nsc symbols provided to the subcarrier mapper. The subcarrier mappers serve as a look-up table (LUT), which assigns the
in-phase and quadrature coordinates to every input symbol, and create the corresponding complex numbers passed to IFFT
OFDM modulator in parallel fashion. After the FFT portion of OFDM symbol is created, the cyclic extension is performed.
The cyclically extended OFDM symbol undergoes P/S conversion and up-sampling operation, before being passed to the
DACs. The real and imaginary outputs of DACs are (after pulse shaping and amplification, denoted as low-pass filtering

QAM
symbol OFDM RF transmitter
data stream
RF transmitter
S/P converter
(or DEMUX)

Binary
Subcarrier

LDPC/turbo/
mapper

Cyclic extension
data stream QAM IFFT DAC
convolutional Interleaver … +
coding mapper modulator
windowing

(a)
Symbol
timing Estimated binary
OFDM RF receiver Estimated
data stream
QAM data stream
RF receiver

Timing and Cyclic LDPC/turbo/

S/P FFT Symbol P/S QAM
ADC extension
…

frequency convolutional
…

Deinterleaving
…

converter demodulator detection converter demapping

synchronization removal decoding

Frequency-offset
Corrected signal

(b)
Fig. 7.6 The block diagram of OFDM transceiver suitable for wireless applications: (a) OFDM transmitter configuration and (b) OFDM receiver
configuration
7.4 OFDM Parameter Selection, OFDM Building Blocks, and Parallel Channel Decomposition 399

OFDM RF transmitter

I/Q Modulator
I

S/P converter
DAC LPF to SMF

Subcarrier
mapper
QAM
symbol
IFFT Cyclic extension
data stream
… modulator insertion
DAC LPF
Q

(a)
From
OFDM RF receiver
SMF DFT/frequency
Balanced LPF ADC FFT demodulator, Estimated
synchronization, P/S QAM

…
…
channel estimation,

…
coherent converter
cyclic removal, data stream
detector LPF ADC symbol detection
From S/P conversion

local laser

(b)
Fig. 7.7 The CO-OFDM system architecture: (a) OFDM transmitter configuration, and (b) CO-OFDM receiver configuration

operation) used as inputs of an optical I/Q modulator. The optical OFDM data stream is transmitted over optical communica-
tion system of interest. On the receiver side [see Fig. 7.7b], the received optical signal is mixed with local laser signal in
balanced coherent detector. The balanced coherent detector provides the projections along in-phase and quadrature basis
functions, denoted as vI and vQ in Fig. 7.7b, which are after amplification and ADC interpreted as real and imaginary parts of
corresponding complex number. Upon the S/P conversion, the FFT demodulation is performed, followed by QAM symbol
detection. The QAM symbol stream is after P/S conversion delivered to the end user. In OFDM RF receiver from Fig. 7.7b,
the OFDM signal after DACs undergoes the following synchronization steps: (i) FFT window synchronization, in which
OFDM symbols are properly aligned to avoid any ISI, and (ii) frequency synchronization, in which the frequency offset is
estimated and compensated. Additionally, after FFT demodulation the subcarrier recovery is performed, where the channel
coefficients at subcarrier level are estimated and compensated for. Full details of these steps, together with various channel
estimation techniques, will be provided in incoming sections.
The OFDM signal at the output of transmitter shown in Fig. 7.7a can be expressed as

X
1 X
N FFT =2
1  
j½2π f LD tþϕPN,S  t
nT s j2π T k
ðt
nT s Þ
sðtÞ ¼ e  X n,k rect e FFT , ð7:31Þ
n¼
1 k¼
N FFT =2
Ts

where we omitted the windowing function. In Eq. (7.31), fLD denotes the optical carrier frequency of transmitter laser diode,
while ϕPN,S denotes the stochastic laser phase noise process. The other parameters in Eq. (7.31) are already described in
previous sections.

7.4.3 OFDM Parallel Channel Decomposition and Channel Modelling

The transmitted optical OFDM signal propagates over optical medium of impulse response h(t,τ), where the parameter τ is
used to denote that the impulse response is time-variant, so that the received signal can be represented as

r ðt Þ ¼ sðt Þ  hðt, τÞ þ wðt Þ, ð7:32Þ

where w(t) denotes the ASE noise process that originates from EDFAs periodically deployed to compensate for fiber loss.
Upon the balanced coherent detection, the corresponding signal can be written as

vðt Þ ’ RPIN RL e jf2πð f LD
f LO ÞtþϕPN,S
ϕPN,LO g

sB ðt Þ  hðt, τÞ þ wðt Þ, ð7:33Þ
400 7 OFDM for Wireless and Optical Communications

where RPIN denotes the photodiode responsivity and RL is load resistance. With sB(t) we denoted the baseband portion of
(7.33):

X
1 X
N FFT =2
1  
t
nT s j2π T k
ðt
nT s Þ
sB ðt Þ ¼ X n,k rect e FFT : ð7:34Þ
n¼
1 k¼
N FFT =2
Ts

It is evident from Eq. (7.33) that estimation and compensation of frequency offset Δfoff ¼ fLD-fLO and the phase error
Δϕ ¼ ϕPN,S-ϕPN,LO are of crucial importance. We can now assume the following: (i) The channel is quasi-static and does not
change for the duration of several OFDM symbols, which is always true for typical CO-OFDM systems, so we can write that
h(t,τ) ffi h(t); (ii) the guard interval is properly chosen so that there is no ISI between neighboring OFDM symbols. The OFDM
demodulating can be performed by a bank of matched filters, and the output of k-th branch can be obtained as

kTZ
s þT FFT
1
Y n,k ¼ vðtÞe
j2πiðt
kT s Þ=T FFT dt
T FFT
t¼kT s
2 3 ð7:35Þ
kTZ
s þT FFT ZΔt
1 4
j2πnt
kT s
¼ hk ðτÞsB ðt
τÞdτ þ wðtÞ5e T FFT
dt,
T FFT
t¼kT s 0

where for simplicity of presentation we omitted RPINRL term and assumed that frequency offset and phase error were
compensated for.
By substituting sB(t) with expression given by Eq. (7.34), we obtain
2 3
ZT FFT ZΔt
P
N=2
1
1
Y n,k ¼ X n0 ,k 4 hk ðτÞe
j2πn ðσ
τÞ=T FFT dτ5e
j2πnσ=T FFT dσ þ wn,k
0

n0 ¼
N=2 T FFT
σ¼0 0
ZT FFT ZΔt ð7:36Þ
P
N=2
1
1 0 0
¼ X n0 ,k hk ðτÞe
j2πn τ=T FFT dτe
j2πðn
n Þσ=T FFT dσ þ wn,k
n0 ¼
N=2 T FFT
σ¼0 0
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
H n0 ,k

where Hn,k denotes the transfer function term of k-th subcarrier in the n-th OFDM symbol. This term is not a function of outer
integral and can be moved out so that the equation above becomes

1, n¼n0
zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
X
N=2
1 ZT FFT
1 0
Y n,k ¼ X n0 ,k H n0 ,k e
j2πðn
n Þσ=T FFT dσ þwn,k , ð7:37Þ
T FFT
n0 ¼
N=2
σ¼0

By using the orthogonality principle, we obtain

Y n,k ¼ H n,k X n,k þ wn,k , ð7:38Þ

which represents the parallel decomposition of an optical dispersive channel in frequency-domain, as illustrated in Fig. 7.8.
If the phase error in coherent optical system is not compensated for, the corresponding parallel optical channel decompo-
sition is given by

Y n,k ¼ X n,k H n,k e jðϕPN,S
ϕPN,LO Þ þ wn,k : ð7:39Þ

The channel coefficient Hn,k accounts for chromatic dispersion, filtering effects, and other linear distortion effects. The
chromatic dispersion portion has the form exp[jϕCD(k)], where ϕCD(k) is defined earlier. We can now substitute Hn,k ¼ H’n,k
exp[jϕCD(k)] to obtain
7.4 OFDM Parameter Selection, OFDM Building Blocks, and Parallel Channel Decomposition 401

…
Hn,k wn,k

Xn,k Yn,k

Hn,k+1 wn,k+1

Xn,k+1 Yn,k+1
…

Fig. 7.8 The parallel dispersive optical channel decomposition in frequency-domain

Y n,k ¼ X n,k H 0 n,k e jðϕPN,S
ϕPN,LO Þ e jϕCD ðkÞ

þ wn,k : ð7:40Þ

The generalization of Equation (7.40) to PDM systems can be done by introducing the Jones matrix of fiber link U( fk), so
we obtain the following channel model:

Y n,k ¼ U ð f k ÞH 0 n,k e jðϕPN,S
ϕPN,LO Þ e jϕCD ðk Þ

Xn,k þ W n,k , ð7:41Þ

where the corresponding symbols get replaced with two-component column vectors, with the first component representing
x-polarization state and the second component representing the y-polarization state. This model can be used to study
CO-OFDM systems and various channel distortion compensation methods. For instance, to study the impact of first-order
PMD, we can use the following Jones matrix [5, 47]:
" #
U xx ðωÞ U xy ðωÞ
U¼ ¼ RPðωÞR
1 ,
U yx ðωÞ U yy ðωÞ
"
jωτ=2 # ð7:42Þ
e 0
PðωÞ ¼ ,
0 e jωτ=2

where τ is DGD, ω is the angular frequency, and R ¼ R(θ,ε) is the rotational matrix
2    3
θ jε=2 θ
jε=2
6 cos 2 e sin e 7
6 2 7
R¼6     7 ð7:43Þ
4 θ jε=2 θ
jε=2 5

sin e cos e
2 2

with θ being the polar angle, while ε is the azimuthal angle. The magnitude of Uxx and Uxy coefficients of Jones channel
matrix against normalized frequency fτ (the frequency is normalized with DGD τ so that the conclusions are independent on
the data rate) is shown in Fig. 7.9 for two different cases: (i) θ ¼ π/2 and ε ¼ 0 and (ii) θ ¼ π/3 and ε ¼ 0. In the first case,
402 7 OFDM for Wireless and Optical Communications

Fig. 7.9 Magnitude response of

Uxx and Uxy Jones matrix |Uxx| |Uxy| = /2, =0
coefficients against the
normalized frequency for: (a)
θ ¼ π/2 and ε ¼ 0 and (b) θ ¼ π/3 1.0
and ε ¼ 0

0.8

Magnitude 0.6

0.4

0.2

0.0
0 2 4 6 8 10 12
Normalized frequency, f
(a)
|Uxx| |Uxy| = /3, =0

1.0

0.8
Magnitude

0.6

0.4

0.2

0.0
0 2 4 6 8 10 12
Normalized frequency, f
(b)
7.4 OFDM Parameter Selection, OFDM Building Blocks, and Parallel Channel Decomposition 403

channel coefficient Uxx completely fades away for certain frequencies, while in the second case, it never completely fades
away, thus suggesting that the first case represents the worst-case scenario.
For multipath fading radio channel, we can denote the impulse response as h(t,τ), where the parameter τ is used to denote
that the impulse response is time-variant. Therefore, Eq. (7.32) is also applicable here. By assuming that the guard interval is
longer than the maximum delay spread, while that the wireless channel is quasi-static for at least the duration of one OFDM
symbol, we can perform the multipath fading channel decomposition in similar fashion as given by Eqs. (7.35)–(7.38). The
time-varying impulse response, as already shown in Chap. 2, can be represented as a sum of multipath components arriving at
the receiver side:
X

jθn ðt Þ
hðτ, t Þ ¼ h ðt Þe
n n
δ½τ
τn ðt Þ, ð7:44Þ

where hn, θn, and τn represent the n-th propagation path amplitude, phase, and delay, respectively. The Fourier transform
(FT) of the time-varying impulse response with respect to propagation path delay τ will give us time-varying transfer function:

Z1 X
H ð f , tÞ ¼ hðτ, t Þe
j2πf τ ¼ h ðt Þe
n n

j½2πf τn ðt Þþθn ðt Þ
, ð7:45Þ

1

which changes rapidly in both time and frequency, exhibiting therefore both time- and frequency-selective fading. Given that
the transfer function is time-varying, it makes sense to define the autocorrelation function of the transfer function:

RH ð f 1 , f 2 ; t 1 , t 2 Þ ¼ hH  ð f 1 , t 1 ÞH ð f 2 , t 2 Þi: ð7:46Þ

This autocorrelation function gets simplified under wide-sense stationary-uniform scattering (WSS-US) assumption (see
Sect. 2.3 of Chap. 2 for additional details):

RH ðΔf ; Δt Þ ¼ hH  ð f , t ÞH ð f þ Δf , t þ Δt Þi: ð7:47Þ

The ranges in time and frequency over which the transfer function exhibits significant correlation are known as coherence
time and coherence bandwidth. The coherence time is defined as the time range when the spaced-time correlation function
RH(Δt) ¼ RH(0;Δt) drops below 0.9. On the other hand, the coherence bandwidth is defined as the frequency range when the
spaced-frequency correlation function RH(Δf ) ¼ RH(Δf;0) drops below 0.9. The delay power spectrum (DPS), denoted as
Rh(τ), is the inverse Fourier transform of the space-frequency correlation function RH(Δf ) ¼ RH(Δf;0). The following simple
model can be used to study OFDM systems for wireless applications [4]:
8
< ρ δðτÞ, τ ¼ 0
> 2

Rh ðτÞ ¼ Π, 0 < τ  τc ð7:48Þ

>
:
γðτ
τc Þ
Πe , τ > τc

where ρ2 is the normalized power of the line-of-site (LOS) component, Π is the normalized power density of the constant-level
portion of DPS, τc is the duration of the DPS constant level, and γ is the decay exponent of the exponentially decreasing
portion of the DPS. Given that the magnitude R ¼ |H( f,t)| of time-varying transfer function H( f,t) for WSS-US model follows
the Rician distribution,
 2   
r r þ s2 rs
f R ðrÞ ¼ 2 exp
I0 2 ,
σ 2σ 2 σ
Z2π ð7:49Þ
1
I 0 ðxÞ ¼ exp ðxcos ψÞdψ,
2π
0

where I0(x) is the modified Bessel function of the first kind and zero order, s ¼ |hH( f,t)i|, and σ2 is the variance of the real and
imaginary components of the complex Gaussian process. As already   discussed in Chap. 2, the total power in multipath
P
components different from the LOS component is 2σ 2 ¼ n,n6¼0 h2n , while the total power is given by
404 7 OFDM for Wireless and Optical Communications

Z1
PRx ¼ r 2 f R ðr Þdr ¼ s2 þ 2σ 2 : ð7:50Þ
0

The Rician distribution can also be expressed in terms of Rician parameter K, defined by K ¼ s2/(2σ 2), as follows:

  sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
2r ðK þ 1Þ ðK þ 1Þr 2 K ð K þ 1Þ
f R ðr Þ ¼ exp
K
I 0 2r , r
0: ð7:51Þ
PRx PRx PRx

The Rician parameter can be used as a measure of the fading severity: the small K implies severe fading, a large K implies
mild fading, and when K tends to infinity, there is no fading at all. The parameters q
of ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
the model ffigiven by Eq. (7.48) can easily
be related to K, PRx, and root mean square (RMS) delay spread, defined as τRMS ¼ hτ2 i
hτi2, as described in [4]. Based on
this model, we can build the frequency-domain channel simulator, based on Gaussian random generator, with details found in
[4], which agrees very well with the measurement results.

7.5 CO-OFDM Principles, DFT Windowing, Frequency Synchronization, Phase Estimation,

and Channel Estimation

7.5.1 Principles of Coherent Optical OFDM (CO-OFDM)

The operation principles for wireless communications have already been provided in Sect. 7.4.2. The operation principle of
CO-OFDM system is illustrated in Fig. 7.10. Several basic building blocks can be identified including RF OFDM transmitter,
optical I/Q modulator (to perform electro-optical conversion), dispersion map of interest, balanced coherent detector, and RF
OFDM receiver. Upon the balanced coherent detection, the received signal can be expressed as
( " #)
rðtÞ ’ Aexp j ðωTx
ωLO Þt þ ϕPN,Tx
ϕPN,LO  sBB ðtÞ  hðtÞ þ wðtÞ, ð7:52Þ
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
Δω Δϕ

where A is the balanced coherent detection constant, h(t) is optical channel impulse response, w(t) is additive noise process
(dominated by ASE noise), while sBB(t) denotes the baseband (BB) portion of the signal:

X
1 X
N FFT =2
1  
t
nT s j2π T k
ðt
nT s Þ
sBB ðt Þ ¼ X n,k rect e FFT : ð7:53Þ
n¼
1 k¼
N FFT =2
Ts

The Xn,k denotes a two-dimensional signal constellation (such as QAM) point transmitted on the k-th subcarrier in the n-th
OFDM symbol, Ts is symbol duration, and TFFT is the effective FFT part of OFDM symbol. The ωTx and ωLO denote the
carrier frequencies of transmit laser and local laser oscillator, respectively. Finally, ϕPN,Tx and ϕPN,LO denote the transmit
laser and local oscillator laser phase noise, respectively. It is evident from Eqs. (7.52) and (7.53) that estimation and
compensation of frequency offset Δω ¼ ωTx
ωLO and the phase error Δϕ ¼ ϕPN, Tx
ϕPN, LO are of crucial importance.
After this review of the optical OFDM signal processing, at conceptual level, we describe the following RF OFDM signal
processing steps: DFT window synchronization, frequency synchronization, channel estimation, and phase noise estimation.
In DFT window synchronization, the OFDM symbols are properly aligned to avoid ISI. In frequency synchronization, we
estimate the frequency offset between transmit laser and local oscillator laser frequency in order to compensate for it. In
subcarrier recovery, we estimate the channel coefficients for each subcarrier and compensate for channel distortion. Finally, in
phase noise estimation, we are trying to estimate the phase noise originating from laser phase noise of transmit and local lasers
and trying to compensate for it. (The reader is advised to look at references [5, 30–37] if needed.)
7.5 CO-OFDM Principles, DFT Windowing, Frequency Synchronization, Phase Estimation. . . 405

Estimate of
transmitted data
Data input

Multiplexer RF OFDM
Demultiplexer … receiver
… Data symbol
decisions
RF OFDM transmitter

Constellation
mapper …
… FFT demodulator
IFFT modulator … •DFT window synchronization
S/P converter and •Frequency synchronization
…
CE remover •Subcarrier recovery
P/S converter and
cyclic extension (CE) Im{ } Re{ }

Re{ } Im{ } ADC

DAC DAC ADC

LPF LPF LPF LPF

Dispersion map
Local
Laser Balanced
I/Q modulator oscillator
diode Coherent detector
Nx laser

Fig. 7.10 The operation principle of CO-OFDM system

Segment 1 Segment 2 OFDM symbol 1 GI DFT window OFDM symbol N

Schmidl-Cox OFDM symbol n

preamble

Fig. 7.11 The organization of OFDM frame showing Schmidl-Cox preamble for DFT windowing

In the rest of this section, we describe the receiver operations common to both wireless OFDM communication and
coherent optical OFDM communication including DFT window synchronization, frequency synchronization, phase estima-
tion, and channel estimation.

7.5.2 DFT Window Synchronization

DFT windowing [32], in which a preamble of pilot symbols consisting of two identical segments is transmitted, has become
very attractive from the application perspective. This method is illustrated in Fig. 7.11. The autocorrelation method is used to
identify the start of DFT window. It is also possible to use cyclic extended guard interval and autocorrelation method to
determine the starting point of DFT window, as discussed in [11]. However, the guard interval is typically shorter than the
preamble used in DFT synchronization, which might result in inferior performance. The DFT synchronization preamble can
be described as [5, 32]

pm ¼ pm þ N sc =2; m ¼ 1, 2, ⋯, N sc =2 ð7:54Þ

where pm denotes the m-th pilot symbol and Nsc denotes the number of subcarriers. By sampling the received signal, based on
parallel channel decomposition model described earlier, we obtain
406 7 OFDM for Wireless and Optical Communications

jðΔωmT FFT =N sc þΔϕÞ

r m ¼ r ðmT FFT =N sc Þ ¼ pm e þ wm , ð7:55Þ

where TFFT is the effective portion of OFDM symbol (see Fig. 7.11). The following correlation function can be used in DFT
windowing:

NX
sc =2

Rd ¼ r mþd r mþdþN sc =2 , ð7:56Þ

m¼1

with the peak for d ¼ 0. The start of FFT effective part can easily be identified, providing that frequency offset Δω is
reasonably small. Another approach is to define normalized magnitude squared of correlation function as [5]
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uN =2
uX sc NX sc =2
 
M d ¼ jRd =Sd j , Sd ¼ t
2
jr mþd j2 r mþdþN =2 2 ,
sc
ð7:57Þ
m¼1 m¼1

and determine the optimum starting point by maximizing Md, which is

b
d ¼ arg max M d : ð7:58Þ
d

7.5.3 Frequency Synchronization in OFDM Systems

The frequency synchronization has two phases: (i) frequency acquisition phase and (ii) frequency tracking phase. Typically,
telecommunication lasers are locked to references in accordance with ITU-T-defined frequency grid. Regarding the optical
communications, a wavelength locker that is commonly used has accuracy of ~2.5 GHz. Any frequency offset between

5 GHz and 5GHz will cause serious problems for OFDM system operation, as discussed in [5]. The purpose of frequency
acquisition is to perform course estimation of frequency offset and to reduce it to fall into a range equal several times of
OFDM subcarrier spacing. After that, the frequency tracking phase can be initiated.
The frequency acquisition can be performed based on pilots [38], ML estimation [39], or cyclic prefix [40], which is
illustrated in Fig. 7.12. If the cyclic extension procedure is implemented as shown in Fig. 7.2, the prefix and the last NG/
2 samples of the effective OFDM symbol portion are identical. On the other hand, the first NG/2 samples of the effective
OFDM symbol portion and the suffix, which are separated by TFFT seconds, are also identical. We are able to estimate the
frequency offset and compensate for it by cross-correlating the corresponding segments. At the same time, correlation peaks
obtained after every OFDM symbol can be used for timing. However, this approach, given the duration of guard interval, is
inferior to Schmidl-Cox timing approach proposed in [32].
The frequency offset that remained upon frequency acquisition can be estimated by using Schmidl-Cox approach. By
substituting Eq. (7.55) into Eq. (7.56), we can rewrite Rd in terms of the frequency offset by

2π f offset !
z}|{
j Δω T FFT =2þΔϕ
NX
sc =2 |ffl{zffl}
2
Rd ¼ jr mþd j e 1=Δ f sc
þ nd
m¼1 ð7:59Þ
NX
sc =2
jðπ f offset =Δ f sc þΔϕÞ
¼ jr mþd j2 e þ nd ,
m¼1

Fig. 7.12 The frequency Frequency

· dt Maximum
acquisition by using the cyclic TG
correlation offset
prefix
Delay
Conjugation
TFFT Timing
7.5 CO-OFDM Principles, DFT Windowing, Frequency Synchronization, Phase Estimation. . . 407

where nd is the equivalent noise term. By ignoring the phase noise error Δϕ, we can estimate the frequency offset foffset as
follows [5, 32]:

bf offset ¼ Δ f sc ∡R , ð7:60Þ
π bd

where ∡Rb is the angle of correlation function (in radians) while b

d is obtained from Eq. (7.58). Once the frequency offset is
d
estimated, it can be compensated from received signal by

b
r comp ðt Þ ¼ e
j2πf offset r ðt Þ: ð7:61Þ

After the time synchronization and frequency synchronization were completed, the subcarrier recovery takes place. The
subcarrier recovery involves phase estimation and channel estimation.

7.5.4 Phase Estimation in OFDM Systems

In the equivalent OFDM channel model, described earlier, the received symbol vector of the k-th subcarrier in the n-th OFDM
symbol, denoted as rn, k, can be represented as

jϕi
r n,k ¼ H k sn,k e þ wn,k , ð7:62Þ

where sn, k denotes the transmitted symbol of the k-th subcarrier in the i-th OFDM symbol, Hk is the channel coefficient of k-th
P sc
subcarrier, and ϕn ¼ ð1=N sc Þ Nk¼1 ϕnk is the common phase error (CPE). Parameter wn,k denotes the additive noise term
(again dominated by AWGN). The likelihood function in this case is given as

N sc 
 !
jϕi 2
X r i,k
H k si,k e
f ðr i1 , r i2 , ⋯, r i,N sc jH 1 , H 2 , ⋯, H N sc ; ϕi Þ ¼ C exp
, ð7:63Þ
k¼1
2σ 2k

where C is the normalization constant while σ 2k is the variance of noise. The corresponding log-likelihood function is given by

N sc 

jϕi 2
X r i,k
H k si,k e
lðr i1 , ⋯, r iN jH 1 , ⋯, H N sc ; ϕi Þ ¼
: ð7:64Þ
k¼1
2σ 2k

By minimizing the log-likelihood function, we can obtain the following solution for the common phase error [5]:

X
N sc 
^ n ¼ arg
ϕ r n,k H k sn,k =σ 2k : ð7:65Þ
k¼1

7.5.5 Channel Estimation in OFDM Systems

There exist a number of algorithms for channel estimation proposed so far. They could be time-domain based, frequency-
domain based, or combination of the two. The channel estimation techniques can be classified as either data-aided or pilot-
aided. The data-aided channel estimation techniques are prone to the error propagation effect and as such will not be discussed
here with much details.
408 7 OFDM for Wireless and Optical Communications

7.5.5.1 Pilot-Aided Channel Estimation

In pilot-aided channel estimation, a portion of subcarriers are allocated for channel estimation. There are two types of pilot
arrangements in one-dimensional pilot-aided channel estimation [35–37], block type and comb type, as illustrated in Fig. 7.13.
The block-type arrangement is suitable when channel conditions change with slow pace, which is the case with chromatic
dispersion effect in fiber-optics communications. The comb-type arrangement is suitable when channel conditions change
fast, like in the case when PMD is considered, in fiber-optics communication. In this case, the pilots are inserted into a certain
number of subcarriers of each OFDM symbol. It is possible to use the two-dimensional (2D) channel estimation as well, which
is illustrated in Fig. 7.14. For instance, the preamble can be used for chromatic dispersion compensation, while the comb-pilot
arrangement is used for PMD compensation. In 2D channel estimation, the pilot symbol spacings in time-domain (TD) and
frequency-domain (FD) must satisfy the following constraints: (1) the TD-symbol spacing ΔsTD < 1/BD, where BD is the
Doppler spread, and (2) the FD-symbol spacing ΔsFD < 1/τmax, where τmax is the maximum delay spread.
Various channel estimators typically belong to the class of either ML estimators or MMSE estimators. Let us again
consider the model expressed by Eq. (7.62), but now we will set the common phase error to zero and omit the subscript
corresponding to the index of OFDM symbol (the assumption is that optical channel does not change too fast). The likelihood
function, assuming Gaussian noise-dominated scenario, is given by
!
X
N sc
j r k
H k sk j 2
f ðr 1 , r 2 , ⋯, r N sc jH 1 , H 2 , ⋯, H N sc Þ ¼ C exp
, ð7:66Þ
k¼1
2σ 2

where C is the normalization constant and σ 2 is the variance of noise. The corresponding log-likelihood function is given as

Fig. 7.13 Pilot arrangements for Data Pilot

OFDM channel estimation: (left)
symbols symbols
block-type arrangement and
(right) comb-type arrangement
Subcarrier, k
Subcarrier, k

Pilot
symbols

Data
symbols

Block Time, i Time, i

Preamble
Subcarrier, k

Time, i
Fig. 7.14 Pilot arrangements for two-dimensional OFDM channel estimation
7.5 CO-OFDM Principles, DFT Windowing, Frequency Synchronization, Phase Estimation. . . 409

X
N sc
jr k
H k sk j2
lðr 1 , r 2 , ⋯, r N sc jH 1 , H 2 , ⋯, H N sc Þ ¼

k¼1
2σ 2
ð7:67Þ
X
N sc
ðr k
H k sk Þðr k
H k sk Þ
¼

k¼1
2σ 2

To determine the channel coefficients, someone needs to perform maximization of log-likelihood function with respect to
all Hk’s, which needs an extensive computation time. Instead, by setting the first derivative of likelihood function with respect
to Hk to be zero, and by solving it for Hk, we will obtain the following solution:

rk w
H k,LS ¼ ¼ Hk þ k : ð7:68Þ
sk sk

The channel estimate obtained by Eq. (7.68) is commonly referred to as the least square (LS) solution. The estimated
information symbols by LS estimator are given as

rk H s þw w w0 s
sk ¼ ¼ k k w0 k k  sk þ k
k 0k , ð7:69Þ
H k,LS H k þ s0 k H k Hk s k

where we used s’k and w’k to denote the transmitted symbol and noise sample generated when the LS estimate of Hk is
obtained. Since there are two additive terms, it is evident that noise gets enhanced, indicating that this scheme represents an
instance of zero-forcing estimator.
In MMSE channel estimation, we intend to linearly transform the LS solution by pre-multiplying HLS with a properly
chosen matrix A as

H MMSE ¼ AH LS , ð7:70Þ

which will minimize the MSE. The estimator error can be defined as

ε ¼ AH LS
H: ð7:71Þ

By employing the orthogonality principle from Sect. 6.3.4 in Chap. 6, and by claiming that the error matrix is independent
on “data,” we obtain

hðAH LS
HÞH { i ¼ 0⟺AhH LS H{ i
hHH { i ¼ 0, ð7:72Þ

|fflfflfflfflffl{zfflfflfflfflffl} |fflfflffl{zfflfflffl}
RH,HLS RH

where RH denotes the correlation matrix of H while RH,HLS denotes the cross-correlation matrix between H and HLS. By
solving Eq. (7.72) for A, we obtain

A ¼ RH R
1
H,H LS , ð7:73Þ

and after substituting it in Eq. (7.70), we derive the MMSE channel estimation as

HMMSE ¼ RH R
1
H,H LS H LS , ð7:74Þ

which resembles a quite general Wiener filter solution. Based on Eq. (7.68), the cross-correlation matrix RH,HLS ffi
 
1
RH,HLS þ σ 2n SS{ , where σ 2n is the variance of the noise and S is the diagonal pilot matrix. Now, upon substitution into
Eq. (7.74), we obtain a simpler equation for MMSE estimation of H:
410 7 OFDM for Wireless and Optical Communications

n o

1
1
H MMSE ¼ RH RH,HLS þ σ 2n hSS{ i H LS : ð7:75Þ

The discussion above, related to MMSE estimation, is applicable to block-type pilot arrangements. To reduce the
complexity and lower the overhead, the combo-type and 2D pilot arrangements should be used instead, as already shown
in Fig. 7.14. The expressions for MMSE estimates are still applicable, but we need just to replace every H with Hp, with
subscript p referring to the pilots. For instance, Eq. (7.74) for combo-type/2D arrangements can be rewritten as

^ p,MMSE ¼ RHp R
1

H ^
Hp ,Hp,LS H p,LS , ð7:76Þ

where Hp, LS is given by [36, 37]



b p,LS ¼ H p,LS ð1Þ⋯H p,LS N p T , H p,LS ðiÞ ¼ Y p ðiÞ ; i ¼ 1, ⋯, N p
H ð7:77Þ
X p ði Þ

In Eq. (7.77), we use Np to denote the number of pilot subcarriers, while Xp(i) and Yp(i) denote the i-th transmitted and
received pilot symbols (i ¼ 1,. . .,Np), respectively. Once the pilots have been determined in combo-type and 2D
arrangements, we need to perform the interpolation to estimate the channel coefficients for data subcarriers. Various
interpolation methods, such as linear interpolation, second-order interpolation, low-pass interpolation, spline cubic interpola-
tion, and time-domain interpolation, have been proposed so far. For instance, in linear interpolation the channel estimation for
k-th subcarrier is given by [35]

  l
b ðk Þ ¼ H p ðm þ 1Þ
H p ðmÞ
H þ H p ðmÞ
N sc =N p
ð7:78Þ
N N N
m sc < k < ðm þ 1Þ sc ; 0 < l < sc
Np Np Np

where we assumed the uniform spacing of pilots, X(mNsc/Np + l ) ¼ Xp(m), with X(i), i ¼ 1,. . ., Nsc, being the symbol carried
by i-th subcarrier. The time-domain interpolation method [35, 48] represents a high-resolution method based on zero-padding
FFT/IFFT. The IFFT is applied to H b p,LS in order to perform conversion into time-domain [35]

NX
p
1
j2πkn=N p
h p ð nÞ ¼ H p,LS e ; n ¼ 0, 1, ⋯, N p
1 ð7:79Þ
k¼0

and interpolate to Nsc points as follows [35]:

8
> Np
>
> hp ðnÞ, 0  n < 2
1
>
>
<
b N N
hðnÞ ¼ 0, p  n < N sc
p þ 1 ð7:80Þ
>
> 2 2
>
>
: hp n
N sc þ N p þ 1
, N sc
N p þ 1  n  N sc
1
>
2

Then the FFT is then applied to convert Nsc-sequence to frequency-domain:

NX
sc
1
b ðk Þ ¼
H b
hðnÞe
j2πkn=N sc ; k ¼ 0, 1, ⋯, N sc
1 ð7:81Þ
n¼0

and such obtained channel coefficient is used to compensate for channel impairments.
The calculation of channel coefficients by using Eq. (7.74) is computationally extensive. The following approximation has
been shown to be effective for both multi-Tb/s superchannel CO-OFDM optical transport [36, 37] and multi-Tb/s CO-OFDM
transport over few-mode fibers [41]:
7.6 Differential Detection in OFDM Systems 411

 
1
^ αðMÞ ^ p,LS ,
H p,MMSE ¼ RHp RHp þ H ð7:82Þ
SNR
 2 
D 2 E 
1 
where parameter α(M ) is defined as αðM Þ ¼ X p ðmÞ X p ðmÞ , and it is dependent on both the signal constellation
D 2 E
type and the size. The signal-to-noise ratio is defined as SNR ¼ X p ðmÞ =σ 2n (σ 2n is the noise variance). This method is also
effective in dealing with multipath fading effects.

7.5.5.2 Data-Aided Channel Estimation

So far the pilot-aided channel estimation techniques suitable for both wireless coherent and optical coherent detections have
been studied. The use of pilots reduces the overall spectral efficiency. In other words, the power efficiency can be improved if
the pilot subcarriers are not transmitted. In data-aided channel estimation, the first OFDM symbol must be known at the
receiver side to start the detection process. The subcarrier symbols from previous OFDM symbol detection are used to predict
the channel for the current OFDM symbol. Namely, when the channel is slowly time-varying, there will be a large correlation
of current OFDM symbol with previous OFDM symbol, so that the channel estimation based on previous OFDM symbol will
be reliable. Otherwise, the error propagation effect will occur. For additional details on data-aided channel estimation, the
interested reader is refereed to [50, 51].

7.6 Differential Detection in OFDM Systems

The key idea behind various channel estimation techniques described so far, based on coherent detection, is to determine
accurate magnitude and phase estimates of the subcarriers’ channel coefficients. These estimates are then used to compensate
for channel distortions. On the other hand, the differential detection does not perform any channel estimation, since the
information is written in the phase difference between the current and previous sub-symbol. The overall complexity and
latency have been reduced at the cost of about 3 dB penalty in signal-to-noise ratio. The configuration of OFDM receiver with
differential detection, suitable for wireless applications, is shown in Fig. 7.15.
This scheme can easily be adopted in direct-detection OFDM, to be described later. The absolute phase reference is not
needed, as the differential detector compares the current sub-symbol with previous sub-symbol. Given that the OFDM is a
two-dimensional scheme, the differential encoding can be applied in either time-domain or frequency-domain.
In time-domain differential encoding, the information is imposed in phase difference between two sub-symbols of the
current and previous OFDM symbols within the same subcarrier. Therefore, the differential encoding is applied on subcarrier
level. For M-ary PSK (M-PSK), the differentially encoded phase related to the k-th subcarrier of the n-th OFDM symbol,
denoted as θn,k, can be written as

X
n
θn,k ¼ ϑm,k mod2π, ð7:83Þ
m¼0

where we use ϑn, k to denote the corresponding absolute phase. The corresponding transmitted PSK symbol is X n,k ¼ e jθn,k . At
the receiver side, based on parallel channel decomposition, described by Eq. (7.38), the received sub-symbol n in k-th
subcarrier can be represented as

Y n,k ¼ jH n,k je jϕn,k  |ffl{zffl}

e jθn,k þ wn,k , ð7:84Þ
|fflfflfflfflfflffl{zfflfflfflfflfflffl}
H n,k X n,k

Estimated binary
data stream
RF receiver

LDPC/turbo/
S/P FFT Differential P/S
ADC
…
…

Deinterleaving convolutional
…

converter demodulator detection converter

decoding

Fig. 7.15 The simplified configuration of OFDM receiver with differential detection, suitable for wireless applications
412 7 OFDM for Wireless and Optical Communications

with wn,k representing an AWGN noise sample, as before. The n-th phase differential detection can be performed by

jðθn,k
θ n
1,k Þ
   jðϕ
ϕ Þ |fflfflfflfflfflfflfflfflfflfflϑ{zfflfflfflfflfflfflfflfflfflffl}
Y n,k Y n
1,k ¼ jH n,k jH n
1,k e n,k n
1,k e n,k þ jH n,k je jϕn,k
wn
1,k ð7:85Þ
  
jϕ

þ H n
1,k e  n
1,k 
wn,k þ wn,k wn
1,k

Even though that the phase ϑn, k is properly detected, there is the phase disturbance term ϕn,k-ϕn-1,k, which is introduced by
the time-selective fading. When OFDM symbol duration is properly chosen to small relative to the phase coherence time, this
phase disturbance can be neglected. On the other hand, the signal level gets affected by fading, while the additive noise gets
doubled, resulting in at least 3 dB degradation in SNR compared to the coherent detection case. On the other hand, when pilot-
aided channel estimation is imperfect, the differential detection penalty is actually lower than 3 dB. Based on Clark-Jakes
model described in Chap. 2, the normalized correlation coefficient corresponding to the n-th OFMD symbol would be

RðnÞ ¼ RðnT s Þ=PRx ¼ J 0 ð2πνD, max nT s Þ, ð7:86Þ

where Ts if OFDM symbol duration while ν D,max is the maximum Doppler frequency. This correlation coefficient can be used
to estimate the maximum tolerable Doppler frequency. The difference between received sub-symbols can be related to the
correlation coefficient as follows [3]:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Y n,k
Y n
1,k ¼ ½1
Rð1ÞY n
1,k þ 1
R2 ð1Þzn,k , ð7:87Þ

where zn,k is the unit variance noise sample. When yn-1,k and zn,k are uncorrelated, the signal-to-distortion ratio (SDR) can be
estimated as [3]

1
SDR ¼ , ð7:88Þ
2ð1
Rð1ÞÞ

and must much larger than SNR not to affect the BER performance. For a SNR penalty of 1 dB, the SDR should be at least
10 dB, so that the corresponding correlation coefficient is R(1)  0.95, corresponding to Doppler frequencyOFDM symbol
duration product of νD,maxTs  0.07 or, equivalently, the maximum tolerable Doppler frequency of νD,max0.07/Ts. Since the
maximum Doppler frequency is related to the user speed u by νD,max ¼ fcu/c ( fc is the carrier frequency), the maximum
tolerable user velocity would be

νD, max 0:07

umax ¼ c c : ð7:89Þ
fc f cT s

As an illustration, for carrier frequency of 3 GHz and OFDM symbol duration 125 μs, the maximum tolerable car speed
would be 56 m/s.
In frequency-domain differential encoding, the information is imposed in phase difference between two neighboring
subcarrier symbols of the same OFDM symbol. The differentially encoded phase θn,k is related to the absolute phase ϑn, k by

X
k
θn,k ¼ ϑn,l mod2π: ð7:90Þ
l¼0

The k-th subcarrier differential detection can be performed by

jðθn,k
θ n,k
1 Þ
   j ϕ
ϕ |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
Y n,k Y n,k
1 ¼ jH n,k jH n,k
1 e ð n,k n,k
1 Þ e ϑn,k
þ jH n,k je jϕn,k
wn,k
1 ð7:91Þ
  
jϕ
þH n,k
1 e n,k
1 wn,k þ wn,k wn,k
1
7.7 OFDM Applications in Wireless Communications 413

Not surprisingly, this equation has similar form as Eq. (7.85). The phase distortion is now related to the phase difference
between the n-th and (n-1)-th subcarrier. In other words, this phase disturbance is now introduced by the frequency-selective
fading. For the exponentially decreasing multipath power delay profile, the frequency-domain (FD) normalized correlation
function is given by [3]

1
RFD ðk Þ ¼ Rðk=T Þ=PRx ¼ , ð7:92Þ
1 þ j2πτRMS k=T

τRMS is the RMS multipath delay spread, and the subcarrier channel spacing is given by 1/T. The corresponding frequency-
domain SDR for k ¼ 1 would be

1
SDRFD ¼ : ð7:93Þ
2ð1
RFD ð1ÞÞ

For the FD SDR of 10 dB, the corresponding correlation will be 0.95, which results in normalized subcarrier spacing (1/T)/
(1/τRMS) of 0.03. In other words, the maximum tolerable RMS delay spread is 3% of the FFT period T. Clearly, in frequency-
domain differential detection, the robustness to multipath fading is FFT period dependent and, therefore, significantly worse
than coherent detection where the tolerance to multipath fading is guard interval dependent.

7.7 OFDM Applications in Wireless Communications

The OFDM has found various applications in wireless communications including digital audio broadcasting (DAB) [51],
digital video broadcasting (DVB) [52, 53], wireless LAN (WLAN) [3], WiMAX, Wi-Fi, WiGig, and ultra-wideband (UWB)
communications [54], to mention few.

7.7.1 OFDM in Digital Audio Broadcasting (DAB)

The digital audio broadcasting has several advantages compared to FM including the improved sound quality (comparable to
that of a compact disk), higher spectral efficiency, it enables single-frequency network (SFN) operation, it offers new data
services, and it provides an unequal error protection. The DAB standard is described in ETSI standard ETS300401, where the
data structure, the FEC, and coded-OFDM modulation are described. There are four selectable transmission modes in DAB, as
summarized in Table 7.1 [3, 49]; the first three are related to the different frequency bands, while the fourth one is introduced
to provide a better coverage range for mode II. The SFN capability can significantly improve the spectrum efficiency. For a
receiver at a given location, it is possible to receive the signal from two different DAB transmitters located at distances d1 and
d2. When the difference in arrival times Δτ ¼ |d1-d2|/c is smaller than the guard interval duration TG, the signals coming from
different DAB transmitters can be interpreted as the transmit diversity scheme to improve SNR and reduce the BER. The
audio signals in DAB are encoded into either MPEG-1 (frame duration of 24 ms) or MPEG-2 (48 ms) frames. During the

Table 7.1 Different DAB modes

Mode Mode I Mode II Mode III Mode IV
Number of subcarriers 1536 384 192 768
Subcarrier spacing 1 kHz 4 kHz 8 kHz 2 kHz
Symbol duration (no guard interval) 1 ms 250 μs 125 μs 500 μs
Guard interval duration 246 μs 62 μs 31 μs 123 μs
Frame length 96 ms 24 ms 24 ms 48 ms
(76 symbols) (76 symbols) (153 symbols) (76 symbols)
Frequency range, carrier frequency Band III VHF, < 375 MHz L band, < 1.5 GHz L band, <3 GHz L band, <1.5 GHz
Application Single-frequency Multi-frequency Satellite SFN
networks (SFN) network (MFN)
Maximum propagation path difference 100 km 25 km 12.5 km 50 km
414 7 OFDM for Wireless and Optical Communications

MPEG-1,2 compatible portion of the frame

DAB extension

Bit SCF Scale SCF F-

STU
…

FF
Header CRC SI
Sub-band samples X-PAD
allocation factors CRC PAD

Variable size
Byte L-1 Byte L
4 bytes

F-
X-PAD SCF-CRC
PAD
Fixed size

Fig. 7.16 DAB audio frame including DAB extension

compression, from 1.5 Mb/s down to 64–384 kb/s, the perceptual coding is used in which the audio components that cannot be
detected by human earing system are omitted. MPEG-compatible DAB audio frame is illustrated in Fig. 7.16.
The header is 32 bits long, while the CRC checksum is 16 bits long. The bit allocation field, scale factors, and sub-band
samples follow the CRC field. The audio sampling rate in MPEG-1 is maximum 48 kHz, which is higher than 44.1 kHz used
in the standard CD. For mono sound the data rates range from 32 to 192 kb/s, while for stereo, dual sound, and joint stereo, the
data rates range between 64 kb/s and 384 kb/s. The MPEG-2 frame represents the MPEG-1 frame supplemented with MPEG-
2 extension. The sampling rate in MPEG-2 is 24 kHz; thus, the frame duration is twice longer (48 ms). The sub-band coding of
audio signals is used. After the staffing bytes field, the DAB extension is used in DAB standard. In DAB extension, the
program-associated data (PAD) are transmitted, with possible versions extended-PAD (X-PAD), which is variable in size with
minimum size of 4 bytes, and fixed-PAD (FPAD), which is of fixed size (2 bytes). The PAD contains the information related
to the music/voice identifier, program-related text, and additional scale factor-CRC bits (SCF-CRC). The DAB audio rates
used in the UK are 256 kb/s for classical music, 128 kb/s for popular music, and 64 kb/s for the voice signal.
The coded-OFDM frame, in DAB, consists of 77 coded-OFDM symbols, the null symbol is used to denote the start of
DAB frame, the first symbol is used as the phase reference for phase/frequency synchronization on the receiver side, the data
OFDM symbols start with the symbol number 2. The coded-OFDM symbol duration ranges from 125 μs, for mode III, to
1 ms, for mode I. To deal with multipath fading effects, the guard interval is appended to the OFDM symbol, with duration of
guard interval being approximately 1/4 of duration of OFDM symbol. The first several symbols of total rate 96 kb/s compose
the so-called fast information channel (FIC), with a number of OFDM symbols dictated by the DAB mode (sampling rate).
The remaining data OFDM symbols of rate 2.304 Mb/s compose the so-called main service channel (MSC). The number of
subcarrier per OFDM symbol is dictated by DAB mode and ranges from 192 for mode III to 1536 for mode I. Each subcarrier
carries π/4-shift-DQPSK symbols, with corresponding constellation diagram shown in Fig. 7.17a. The carrier phase gets
shifted for π/4 from symbol to symbol. On such a way, we move from QPSK to rotated QPKS for π/4 rad from symbol to
symbol. From Fig. 7.17a, it is evident that with π/4 phase shift, we move to 00 and 10 correction symbols (states), while with
3π/4 shifts, we move to 01 and 11 states. Clearly there is no transition of π rad. For comparison purposes, the conventional
DQPSK signal constellation is shown as well. The transition from 00 state to 11 state results in the phase shift of π rad, which
changes amplitude of the carrier causing additional symbol errors in detection stage, due to filtering effects. After the
computation of projections of received signal along the basis functions ϕ1(t) and ϕ2(t), to get the in-phase and quadrature
estimates, the corresponding detector is composed of the following blocks: (i) arctangent computation block to determine the
absolute phase, (ii) the phase-difference block, and (iii) modulo-2π correction block to correct the errors due to phase angles
wrapping around the ϕ1(t)-basis function.
The bandwidth of DAB signal is 1.536 MHz, and together with the guard band, it occupies in total 1.75 MHz, which is 1/4
of the 7 MHz channel available in band III (174–240 MHz). Therefore, four DAB channels get frequency multiplexed into one
VHF channel.
7.7 OFDM Applications in Wireless Communications 415

Fig. 7.17 The constellation

2
diagrams for: (a) π/4-shift- 2
DQPSK and (b) DQPSK
01 00 00
01

1
1

11 10 11 10
(a) (b)

Fig. 7.18 The configuration of

FEC 1
DAB modulator
Convolutional Time
Scrambler encoder interleaver
S/P converter

…
ETI OFDM
signal FEC n modulator
Convolutional Time
Scrambler encoder interleaver

Pre-compensation

Power amplifier
encoder and
Frequency
interleaver

Differential

and BPF
From
mapper

Guard I/Q mod.

FEC IFFT FIR
… … … Interval &
blocks modulator
insertion IF/RF up-conv.

Fig. 7.19 The configuration of OFDM modulator for DAB

The DAB signal is composed of up to 64 subchannels carrying audio, data, or a combination of the two signals,
multiplexed together into the common interleaved frame, with the help of the ensemble multiplexer. The data rate per
subchannel is a multiple of 8 KHz. Such obtained ensemble transfer interface (ETI) signal is used as the input to the DAB
modulator, shown in Fig. 7.18, composed of n FEC schemes, one per subchannel, providing unequal error protection (UEP),
and an OFDM modulator. Clearly, the DAB modulator is in fact the coded-OFDM modulator. The configuration of OFDM
modulator is shown in Fig. 7.19. The frequency interleaver is used to deal with frequency-selective fading effects. The
differential encoding and mapping is already described earlier; the pre-compensation block is used to partially pre-compensate
for the nonlinearities introduced by the amplifier. After I/Q modulation and intermediate frequency (IF)/RF up-conversion, the
OFDM signal is amplified, followed by the band-pass filtering (BPF), and broadcasted by a transmit antenna. The OFDM
demodulator is similar to that shown in Fig. 7.15. The organization of DAB packet is shown in Fig. 7.20, which is self-
explanatory.
To avoid synchronization problems on the receiver, originating from sequences of all zeroes/ones, the scrambler shown in
Fig. 7.21 is used. The scrambler has been reset every 24 ms by the all-ones pattern into the shift register.
416 7 OFDM for Wireless and Optical Communications

Main stream data (MSD)

Header FIC SBC1 SBC2 … SBCn … SBCN

FIC: Fast information channel

FIB: Fast Sub-channel (SBC) packet
FIB MCI, SI, FIDC FIB information 24/48/72/92 bytes Packet header & ID
block

FIB: Fast information block (256 bits) Packet Packet … Packet

payload payload payload
CRC
FIB data field (16 bits) Multiple of 8 kb/s

FIG i … FIG j … FIG k Padding

MCI: Multiplex configuration info.

SI: Service information
FIG FIG Data field of FIDC: Fast info. Data channel
type length FIG

Fig. 7.20 The organization of the DAB packet

Fig. 7.21 The description of the Scrambler Scrambler

scrambler
input output

z-1 z-1 z-1 z-1 z-1 z-1 z-1 z-1 z-1

The FEC is based on convolutional coding, as illustrated in Fig. 7.22. The convolutional codes have been described in
Chap. 9; here we briefly describe the puncturing mechanism. To achieve the unequal error protection, the code rate R is varied
by changing the puncturing index (PI) according to the following equation:

8
R¼ , PI 2 f1, 2, ⋯, 24g: ð7:94Þ
8 þ PI

Therefore, the code rate ranges from 8/9 to 8/32. Typically, the FIC is protected by code rate 1/3, while the code rates
related to the data payload are chosen from the set {2/8, 3/8, 4/8, 6/8}, depending on the priority level.
7.7 OFDM Applications in Wireless Communications 417

Puncturing

z-1 z-1 z-1 z-1 z-1 z-1 Code rate

adjustment

Fig. 7.22 The convolutional encoder configuration and puncturing mechanism. The code rate is determined by R ¼ 8/(8 + PI), where PI is the
puncturing index, which is dependent on the protection level. The set of code rates of this punctured convolutional code is {8/9, 8/10, . . ., 8/31, 8/32}

7.7.2 Coded-OFDM in Digital Video Broadcasting (DVB)

Today there exist numerous standards related to DVB [52, 53]. DVB-S, DVB-C, and DVB-T represent the DVB standards for
the first-generation satellite, cable, and terrestrial transmission, respectively. On the other hand, DVB-S2, DVB-C2, and
DVB-T2 represent the second-generation standards for satellite, cable, and terrestrial transmission, respectively. DVB-H
standard represents an extension of DVB-T for handheld mobiles. DVB-SH standard describes the hybrid method for
handheld mobiles via satellite and terrestrial transmission systems. DVB-IP is the standard for video stream transmission
over the IP networks. DVB-SNG represents the satellite transmission standard for professional applications. DVB-T,
DVB-C2, and DVB-T2 standards are coded-OFDM based, while DVB-S, DVB-C, DVB-SNG, and DVB-S2 are single-
carrier-based standards.
In the remainder of this section, we briefly describe the basics of DVB-T and DVB-T2. DVB-T channel has a bandwidth of
6, 7, or 8 MHz. There are two operating modes in DVB-T: 2 K mode employing 2046-point-based IFFT and 8 K mode
employing 8192-point-based IFFT. The basic system parameter in DVB-T is the IFFT sampling frequency for 8 MHz
bandwidth, denoted as fs IFFT, 8MHz and determined by

f s IFFT,8MHZ ¼ 64=7 MHz ¼ 9:142857143 MHz: ð7:95Þ

The corresponding sampling frequencies for 7 MHz and 6 MHz channels are given respectively by

64 7
f s IFFT,7 MHz ¼  MHz ¼ 8 MHz
7 8 ð7:96Þ
64 6
f s IFFT,6 MHz ¼  MHz ¼ 6:857142857 MHz
7 8

The subcarrier spacing can be obtained by dividing the sampling frequency with the number of subcarriers N:

Δ f sc ¼ f s IFFT,8=7=6 MHz =N,

ð7:97Þ
Δ f 2K ¼ f s IFFT,8=7=6 MHz =2048, Δ f 8K ¼ f s IFFT,8=7=6 MHz =8192:

The symbol duration (duration of IFFT portion) can be determined as the reciprocal of the subcarrier spacing:
418 7 OFDM for Wireless and Optical Communications

Table 7.2 DVB-T system parameters for 8 MHz channel bandwidth

Mode 2K 8K
Number of subcarriers Nsc actually used 1705 6817
Subcarrier spacing 4.464285714 kHz 1.116071429 kHz
Signal bandwidth 7.61161 MHz 7.608259 MHz
Symbol duration Δt (no guard interval) 224 μs 896 ms
Guard interval durations gΔt, g2(1/4,1/8,1/16,1/32) g ¼ 1/4: 56 μs g ¼ 1/4: 224 ms
g ¼ 1/8: 28 μs g ¼ 1/8: 112 ms
g ¼ 1/16: 14 μs g ¼ 1/16: 56 ms
g ¼ 1/32: 7 μs g ¼ 1/32: 28 ms
Total symbol durations, (1 + g) Δt g ¼ 1/4: 280 μs g ¼ 1/4: 1120 ms
g ¼ 1/8: 252 μs g ¼ 1/8: 1008 ms
g ¼ 1/16: 238 μs g ¼ 1/16: 952 ms
g ¼ 1/32: 231 μs g ¼ 1/32: 924 ms
Symbol rates (per single subcarrier) g ¼ 1/4: 3.571429 kS/s g ¼ 1/4: 0.8928571 kS/s
g ¼ 1/8: 3.968254 kS/s g ¼ 1/8: 0.9920635 kS/s
g ¼ 1/16: 4.201681 kS/s g ¼ 1/16: 1.05042 kS/s
g ¼ 1/32: 4.329004 kS/s g ¼ 1/32: 1.082251 kS/s

Δt ¼ 1=Δ f sc , Δt 2K ¼ 1=Δ f 2K , Δt 8K ¼ 1=Δ f 8K : ð7:98Þ

The guard interval TG is chosen to be 1/4, 1/8, 1/16, or 1/32 part of the symbol duration. The total symbol duration (OFDM
symbol duration) is then given by

1þg
T s ¼ T G þ Δt ¼ gΔt þ Δt ¼ , g 2 f1=4, 1=8, 1=16, 1=32g: ð7:99Þ
|{z} Δ f sc
gΔt

The subcarrier symbol rate would be then

Δ f sc
Rsubcarrier ¼ 1=T s ¼ , g 2 f1=4, 1=8, 1=16, 1=32g: ð7:100Þ
1þg

The number of subcarriers actually used in 2 K mode is Nsc ¼ 1705, while the number of subcarriers used in 8 K mode is
Nsc ¼ 6817. Therefore, the bandwidth occupied BWDVB-T would be then

BW DVB‐T ’ N sc Δ f sc : ð7:101Þ

Out of the 1705 subcarriers used in 2 K mode, there are 1512 payload subcarriers, carrying QPSK, 16-QAM, or 64-QAM
symbols, 142 scattered pilots (whose position changes), 45 continuous pilots, and 17 transmission parameter signaling (TPS)
subcarriers. On the other hand, out of the 6817 subcarriers used in 8 K mode, there are 6048 payload subcarriers, 568 scattered
pilots, 177 continuous pilots, and 68 TPC carriers.
The DVB-T system parameters are summarized in Table 7.2, for 8 MHz channel bandwidth. By using Eq. (7.96), it is easy
to calculate the corresponding parameters for 7 MHz and 6 MHz channel bandwidths.
The FEC used in DVB-T, DVB-C, and DVB-S is illustrated in Fig. 7.23. For completeness of the presentation, the source
coding and MPEG-2 multiplexing is described as well. The energy dissipation is used to decorrelate the byte TS sequence.
Reed Solomon RS (204,188) code, capable of correcting up to 8 bytes in error, is used as an outer code. The convolution
interleaver is used to deal with burst errors introduced by deep fade (please refer to Chap. 9 for additional details). The
convolutional encoder is used as an inner code. The puncturing block is used to puncture parity bits to deal with time-varying
channel conditions.
The code rates of punctured convolutional code are selected from the following set {1/2, 2/3, 3/4, 5/6, 7/8}. Two different
MPEG-TS streams can be transmitted simultaneously through the so-called hierarchical modulation. The high-priority
(HP) path is used to transmit a low-data rate stream, lower-quality signal with higher compression ratio, but with stronger
FEC code and more robust modulation format such as QPSK. On the other hand, the low-priority (LP) path is used to transmit
MPEG-TS with a higher data rate, employing a weaker FEC scheme, but higher modulation format order, such as 16-QAM
7.7 OFDM Applications in Wireless Communications 419

Program
Video streams
encoder

Multiplexer
Program
PS1

(MUX)
Audio
encoder

Multiplexer (MUX)
Transport
Data stream (TS) TS1

Transport
encoder PS2
Splitter
TS2

…
Source coding
+ MPEG-2 MUX PSn

(a)
MUX Coded
TSi adaptation, data
RS(204,188) Convolutional Convolutional Puncturing
Energy
encoder interleaver encoder block sequence
(i=1,2) dissipation
(scrambling)

The same as in DVB-C Code rate

adaptation

The same as in DVB-S

(b)
Fig. 7.23 The source coding and FEC in DVB-T: (a) source coding and MPEG-2 multiplexing and (b) FEC in DVB-T, DVB-C, and DVB-S

and 64-QAM. At the receiver side, the TV set can decide which mode to select (LP or HP), depending on the received signal
quality.
The aggregate data information rate (net data rate), denoted as RD, can be determined as

Δ f sc 188
RD ¼ N sc R log 2 M  cf BW , g 2 f1=4, 1=8, 1=16, 1=32g, ð7:102Þ
1 þ g 204 PCC

where RPCC is the punctured convolutional code rate, M is the signal constellation size, and cfBW is channel bandwidth-
dependent correction factor, which is based on Eq. (7.96), determined as
8
< 1, BW ¼ 8 MHz
>
cf BW ¼ 7=8, BW ¼ 7 MHz :
>
:
6=8, BW ¼ 6 MHz

The DVB-T FEC decoder configuration is shown in Fig. 7.24. After symbol/bit deinterleaving, the Viterbi decoder inserts
dummy bits on positions where the parity bits get punctured on the transmitter side. After the Viterbi decoding, the
convolutional deinterleaver breaks the burst errors, so that RS (204,188) decoder deals only with random errors and short
bursts. If the number of errors exceeds the error correction capability of the RS decoder, the transport error indicator gets
420 7 OFDM for Wireless and Optical Communications

Energy
From Symbol/bit Viterbi Convolutional RS(204,188) dissipation TS
demapper deinterleaver decoder deinterleaver decoder removal & output
adaptation

Code rate
adaptation

Fig. 7.24 The DVB-T FEC decoder

Fig. 7.25 The configuration of Pilots & TPS

the DVT-B transmitter insertion

Interval insertion
TS1 Low-priority

interleaver
interleaver

adaptation
Demux

Symbol

Mapper
FEC

Guard
Frame
IFFT

Bit

…
…

…
TS2 High-priority modulator
FEC

Pre-correction Transmit
antenna

conversion
IF/RF up-

amplifier
Band-

Power
block

DAC Pass
filter

activated so that the corresponding packet in error cannot be processed by MPEG-2 decoder. After energy dissipation removal
and re-synchronization, the MPEG-2 TS get recreated.
The overall configuration of DVB-T transmitter is shown in Fig. 7.25. Clearly, by comparing with Fig. 7.6, we conclude
that the core of DVB-T transmitter is coded-OFDM RF transmitter. The corresponding low-priority and high-priority data
streams after FEC encoding, as described in Fig. 7.23b, undergo the demultiplexing, followed by bit and symbol interleaving
steps. After subcarrier QAM mapping, the frame adaptation is performed by appropriately inserting the pilots used on the
receiver side for channel estimation and TPS subcarrier symbols.
The remaining steps are identical to those described in Fig. 7.6a and include the guard interval insertion and DAC, while
the pre-correction step to deal with amplifier nonlinearities is optional. After the IF/RF up-conversion, power amplification,
and band-pass filtering, the corresponding DVB-T signal is transmitted over a transmit antenna.
The TPS subcarriers contain information about the mode (2 K or 8 K), duration of the guard interval (1/4, 1/8, 1/16, or 1/32
of IFFT part), convolutional code rate (1/2, 2/3, 3/4, 5/6, 7/8), the type of modulation (QPSK, 16-QAM, 64-QAM), and the
employment of the hierarchical modulation. To transmit this information, the differential BPSK (DBPSK) is used. As
indicated above, 17 subcarriers in 2 K mode and 68 subcarriers in 8 K mode are used as TPC subcarriers. All TPC subcarriers
within the same OFDM symbol carry the same bit. On the receiver side, the decision rule is a majority voting rule. We can
interpret this approach as a repetition code. The TPS information sequence is transmitted over 68 OFDM symbols, with the
first one serving as a reference, meaning that the corresponding TPS sequence length is 67 bits. Positions 1–16 represent a
synchronization word, while bits 25 and 26 describe signal constellation size as follows: 00 denotes QPSK, 01 denotes
16-QAM, and 10 denotes 64-QAM. Further, bits 27–29 are related to the hierarchical modulation information, with
000 denoting non-hierarchical modulation (conventional QAM), while 001, 010, and 011 correspond to hierarchical
modulation with expansion factor ef ¼ 1, 2, and 4, respectively. Bits 36 and 27 represent the (normalized) guard interval
duration as follows: 00 represents 1/32, 01 represents 1/16, 10 represents 1/8, and 11 represents 1/4. The transmission mode is
transmitted by bits 38 and 39, with 00 representing 2 K mode and 01 representing 8 K mode. Bits 33–35 are used to represent
the convolutional code rate of the LP mode, with 000 denoting 1/2 rate, 001 representing 2/3 rate, 010 denoting 3/4 rate,
011 representing 5/6 rate, and 100 denoting the 7/8 code rate. In similar fashion, bits 3–32 are used to represent the code of the
HP stream.
7.7 OFDM Applications in Wireless Communications 421

1
1

Continuous/ Continuous/ TPS

TPS
scattered scattered subcarrier
subcarrier
subcarrier subcarrier

(a) (b)

2
2

1
1

(c) (d)
Fig. 7.26 DVT-B signal constellation diagrams for (a) QPSK and (b) 16-QAM (for ef ¼ 1). Hierarchical 16-QAM modulation for expansion
factors: (c) ef ¼ 2 and (d) ef ¼ 4

The continuous (continual) pilots are based on BPSK and are transmitted with 3 dB higher power compared to the average
transmit power, and they are used on the receiver side for phase reference in automatic frequency control (AFC) block.
Within the OFDM symbol, every 12th subcarrier is the scattered pilot used in channel estimation. In the next OFDM
symbol, the positions of the scattered pilots are increased for three positions. Therefore, it is possible that the position of the
scattered pilot coincides with the position of the continuous pilot.
The payload subcarriers carry QPSK, 16-QAM, or 64-QAM symbols. To illustrate the coordinates of the TPS and
continuous/scattered subcarriers relative to payload subcarriers signal constellation, in Fig. 7.26 QPSK and 16-QAM signal
constellations are provided.
To ensure that reliable transmission is still possible, when the received signal is weak (SNR is small), the hierarchical
modulation can be used as an option in DVB-T. In hierarchical modulation, there are two modulation paths in DVB-T
transmitter, as mentioned earlier, the LP and HP modulation paths. The HP path employs smaller constellation size (QPKS)
and lower code rate, such as 1/2, to guarantee the reliable transmission. In principle, these two modulation paths can carry the
same video signal with different compression ratios and consequently different data rates. On the other hand, HP and LP
modulation paths can carry different video signals. The HP path represents an embedded QPSK in either 16-QAM or
64-QAM. As an illustration, in Fig. 7.26a, b, the four constellation points in each quadrant can be interpreted as points in
QPSK for HP modulation path. The quadrant information carries 2 bits, so the remaining 2 bits in 16-QAM can be used to
identify the point within the quadrant. In 64-QAM, we can decompose the constellation into 4 quadrants each with 16 points.
The quadrant information can be used for HP stream, while with 4 bits we can select 1 of the 16 points in each quadrant, which
corresponds to the LP stream. The expansion factor increases the distance among center-of-mass points for each quadrant, as
422 7 OFDM for Wireless and Optical Communications

Fig. 7.27 The block diagram of Scattered TPS

Cont.
the DVB-T receiver. NCO: pilots subcarriers
pilots
numerically controlled oscillator.
The channel decoder is described Frequency Channel TPS
in Fig. 7.24 /2 NCO correction estimation decoder

compensation
Impairments
demodulator

Demapper
Channel
Delay

Decoder
Channel
Switch
Recovered

FFT
TS
FIR

Receive
antenna

RF/IF down-
conversion
Low- Band-
ADC Pass Pass
filter filter

Clock Local
oscillator
fs=4 32/7 MHz
fs/4=32/7 MHz

illustrated in Fig. 7.26c, d. By increasing the expansion factor, the interference between HP and LP streams is lower. On the
other hand, for LP stream the constellation points are getting closer to each other, and the LP stream is more sensitive to
crosstalk and low SNR values.
The configuration of DVB-T receiver is provided in Fig. 7.27. The first module represents the tuner, to convert the selected
RF channel down to the intermediate frequency (IF). At IF the signal is band-pass filtered, typically by a surface acoustic wave
(SAW) filter. Further, the second IF down-conversion stage takes place, with center IF frequency being at 32/7 MHz. The I/Q
demodulation in DVB-T is based on fs/4-method, where fs is the sampling frequency in the ADC, which is equal to fs ¼ 432/
7 MHz. The time synchronization is based on autocorrelation method described in Fig. 7.12. In parallel with the time
synchronization, the signal after the ADC is used as the input to the changeover switch, whose outputs get selected in an
alternative fashion. Therefore, the corresponding branch data rate is fs/2. There is the offset between two signal branches for
the half of the sampling cycle. To compensate for this, the FIR filter is used in the lower branch and delay line in the upper
branch (to make sure that delays in I and Q channels are properly aligned in time-domain). The output data streams of the
upper and lower branches are used as the input of the numerically controlled oscillator (NCO). The NCO and mixer are
essentially used as a digital PLL. However, since the oscillators are not of very high precision (to ensure that the TV set is
inexpensive), the AFC approach is used by employing the information conveyed by the continuous pilots after FFT
demodulation. When transmit and receive frequencies differ, there will be either clockwise or anti-clockwise rotation of the
constellation diagram. The phase difference between continuous pilots from symbol to symbol is used as a control variable of
the NCO, which changes the output in the direction of reducing the difference between incoming frequency and digital PLL
frequency. Once this difference becomes zero, the rotation of constellation diagram is compensated for, and we employ the
scattered pilots to compensate for the channel impairments. The detection of TPS subcarriers is differential and therefore
independent on the phase error. The repetition coding and additional BCH coding is used to ensure that the TPS sequence is
error-free. After that, the demapping takes place followed by channel decoding, with details of channel decoder provided
already in Fig. 7.24. The output of decoder represents the recovered TS MPEG-2 stream.
The fs/4 method for I/Q demodulation is illustrated in Fig. 7.28. The QAM signal with carrier frequency fIF is used as input
to ADC clocked at fs ¼ 4fIF. On the other hand, the 1-to-2 switch is clocked at twice lower frequency ( fs/2 ¼ 2fIF) so that two
samples per I and two samples per Q channel are obtained at the upper and lower output branches of the switch, respectively.
Since the second sample in either branch is of opposite sign that is needed, we flip the signs with rate fIF. The FIR is used in the
lower branch to ensure the proper synchronization of the upper and lower branches. The outputs of the upper and lower
branches represent the estimates of in-phase and quadrature samples, used at the input of the demodulator.
In order to increase the overall data rates of DVB-T, the DVB-T2 standard is introduced [52, 55]. In addition to channel
bandwidths already in use in DVB-T, the 10 MHz bandwidth is introduced. Stronger FEC is used in DVB-T2, employing
7.7 OFDM Applications in Wireless Communications 423

fs/2
1
I
Delay

Switch
QAM
ADC
signal Q
FIR

fs=4fIF
1
Fig. 7.28 The fs/4 method based I/Q demodulation

From BCH LDPC Bit

scrambler encoder encoder interleaver

Fig. 7.29 The FEC used in DVB-T2 is based on concatenated code in which LDPC code is used as the inner code and BCH code is used as the
outer code

Fig. 7.30 The FEC frame used in

DVB-T2 Header Data field
(80 bits) (MPEG-2 or other data)

Padding

kBCH

BCH
(parity bits)

kLDPC

LDPC
(parity bits)

LDPC codeword length (64800 or 16200 bits)

LDPC coded as the inner code and BCH code as the outer code, which is illustrated in Fig. 7.29. The following code rates are
supported: 1/2, 3/5, 2/3, 4/5, and 5/6. The corresponding FEC frame is shown in Fig. 7.30. Both BCH and LDPC codes belong
to the class of systematic codes, in which information portion remains unchanged, while generalized parity bits are added to
the information word to form a codeword. The number of parity bits in BCH code is 16 t, where t is the error-correcting
capability of BCH code (8, 10, or 12). The LDPC codeword length is either 64,800 or 16,200 bits. The information word
length for the LDPC code, denoted as kLPDC, is determined from the target code rate and FEC frame length. Moreover, the
LDPC code is adaptive; the FEC strength can be increased (code rate decreased) when channel conditions get worse. The
adaptive coding can be combined with adaptive modulation in the so-called adaptive modulation and coding (AMC) [6, 7, 30].
The following modulation formats are supported in DVB-T2: QPSK, 16-QAM, 64-QAM, and 256-QAM. In DVB-T2
Gray mapping and coherent detection are used. For 8 MHz channel bandwidth, the signal bandwidth is 7.61 MHz. Therefore,
the date rates for guard interval fraction g ¼ 1/128 range from 4 Mb/s for QPSK with code rate R ¼ 1/2 to 50.32 Mb/s for
256-QAM with R ¼ 5/6 [52]. The coordinates for these normal QAM constellations can be found in textbooks [26]. Another
type of signal constellations used in DVB-T2 belongs to the class of the so-called rotated Q-coordinate-delayed signal
constellations, whose operation principles are illustrated in Fig. 7.31. The m ¼ log2M bits from input sequence are used to
select the point from M-ary QAM, by employing the Gray mapping rule, with the corresponding coordinates stored in a look-
up table (LUT), with coordinates being I’k and Q’k. The rotator block rotates the signal constellation by ϕ [deg] as follows:
424 7 OFDM for Wireless and Optical Communications

Fig. 7.31 The illustration of log2M I’k Ik Ik

rotated Q-coordinate delayed
signal constellation principle Sequence bits Q-
S/P Mapper Rotator
converter (LUT) Q’k block Qk coordinate Qk 1
of bits delay

Qk 1

Fig. 7.32 The organization of

the DV-T2 frame P2
P1 PLP0 PLP1 … Payload symbols PLPn
symbols

DVB-T2 frame

8
>
>
29 deg, forQPSK
>
< 16:8 deg, for 16‐QAM
ϕ¼
>
> 8:6 deg, for 64‐QAM
>
:
tan
1 ð1=16Þ deg, for 256‐QAM

The Q-coordinate delay block, in the k-th transmission instance, transmits the following coordinates (Ik, Qk-1). Therefore,
the k-th signal constellation point of rotated QAM (Ik, Qk) is transmitted into two consecutive instances (Ik, Qk-1) in the k-th
time instance, also referred to as “cell” k, and (Ik + 1, Qk) in the (k + 1)-st time instance, also referred as cell k + 1. Since,
DVB-T2 is based on coded-OFDM, the Q-coordinate of a given signal constellation point is on the next subcarrier. On such a
way, better tolerance to time-varying channel conditions and fading effects is obtained.
In addition to 10 MHz channel bandwidth introduced in DVB-T2, for cellular (mobile) telephony applications, additional
two channel bandwidths are introduced as well: 1.7 MH and 5 MHz. The number of different FFT modes for coded-OFDM in
DVB-T2 is larger compared to DVB-T, and these are denoted as 1 K, 2 K, 4 K, 8 K, 16 K, and 32 K. The numbers of
subcarriers in the 8 MHz channel are 853 for mode 1 K, 1705 for mode 2 K, 3409 for 4 K, 6817 for 8 K, 13633 for 16 K, and
27,625 for mode 32 K. Because sinc(x) has tails, the out-of-band spectrum is typically filtered by corresponding windowing
functions. As the number of subcarriers increases, the transition in frequency-domain is sharper. Nevertheless, for modes
larger or equal to 8 K, the spectrum moves toward the spectrum of neighboring channel. To solve for this problem, the
extended carrier mode is introduced in which additional channel bandwidth is allocated. The numbers of subcarriers in
extended carrier mode are 6913 in 8 K mode, 13,921 in 16 K mode, and 27,861 in 32 K mode. As before, the script K is used
to denote the size of FFT block of 1024. For instance, the size of FFT block in 32 K is 32,768.
The organization of DVB-T2 frame is shown in Fig. 7.32. The symbol P1 is used for synchronization and frame detection
purposes. The symbols P2 are used for signaling purposes on the receiver side to determine the parameters of OFDM frame,
and the number of P2 symbol is a function of the FFT mode: 16 P2 symbols are used in 1 K mode, 8 in 2 K, 4 in 4 K, 2 in 8 K
mode, and 1 in 16 K and 32 K modes. The payload symbols that follow after P2 symbols are organized into the so-called
physical layer pipes (PLPs), and up to 255 input data streams are possible. The length of DVB-T2 frame in terms of the
number of OFDM symbols is dependent on FFT mode and guard interval fraction g and, for 8 MHz channel and g ¼ 1/16,
ranges from 64 in 32 K mode to 2098 in 1 K mode.
Before the IFFT modulation takes place, different types of interleavers are used: bit interleaver is used just after FEC
encoder, cell interleaver as described in Fig. 7.31, and the time interleaver is used after the cell interleaver. The symbol streams
from different PLPs are frequency interleaved before the IFFT modulation, as shown in Fig. 7.33.
The coded-OFDM portion of transmitter is very similar to generic coded-OFDM transmitter shown in Fig. 7.6a, except
from P1 symbol insertion. To deal with time-selective fading effects, the time interleaver is used. To deal with frequency-
selective fading effects, the frequency interleaver is used. To deal with multipath fading, the multiple-input single-output
(MISO) processing is used, based on Alamouti coding Scheme [56, 57]. Additional details about Alamouti coding scheme can
be found in Chap. 8.
7.7 OFDM Applications in Wireless Communications 425

PLP0

Mode/stream adaptation
Concatenated FEC Mapper cell time

Frequency interleaver
Frame builder block
stream + bit interleaver (LUT) interleaver interleaver
PLP1
Concatenated FEC Mapper cell time
+ bit interleaver (LUT) interleaver interleaver
…

…
…

…
PLPn Concatenated FEC Mapper cell time
+ bit interleaver (LUT) interleaver interleaver

Guard P1 To RF
Pilot IFFT PAPR
interval symbol DAC
MISO processing

insertion modulator reduction Tx1

insertion insertion

Guard P1 To RF
Pilot IFFT PAPR
interval symbol DAC
insertion modulator reduction Tx2
insertion insertion

Fig. 7.33 The configuration of DVB-T2 transmitter based on coded-OFDM and multiple input single output (MISO)

Since in OFDM different subcarriers are modulated independently, some of the subcarriers can add constructively leading
to high peak-to-average power ratio (PAPR), defined as
0 1
max jsOFDM ðt Þj2
PAPR ¼ 10 log 10 @ D E A ½dB,
0t<T OFDM
ð7:103Þ
jsOFDM ðt Þj2

where sOFDM(t) represents the OFDM frame of duration TOFDM. High PAPR can introduce nonlinearities in driver amplifier
and DAC and ADC blocks. To solve this problem, either tone reservation or active constellation extension (ACE) is used in
DVB-T2. In tone reservation, certain subcarriers are planed for payload symbols, which are not used as pilots, but reserved to
reduce the PAPR. These tones are not used unless the PAPR is too high, when they get activated to reduce the PAPR. In ACE
the outmost constellation points are displaced further away from the original constellation in an effort to reduce the PAPR.
The pilots used in DVB-Tw can be categorized as edge pilots (at the beginning and end of spectrum), continuous pilots,
scattered pilots, P2 pilots, and frame-closing pilots. The continuous and scattered pilots have the same role as in DVB-T, for
AFC and channel estimation, respectively. On the other hand, the P2 symbols have the same role as TPS subcarriers in
DVB-T. There exist eight pilot patterns (PPs) in DVB-T2 depending on the distance between neighboring subcarriers in time-
domain and frequency-domain. Let d1 and d2 denote the distance between scattered pilot subcarrier positions and distance
between scattered pilots within the same symbol, respectively. Let d3 denote the number of symbols involved in the pilot
sequence. Then the PPs can be specified by the following triple (d1, d2, d3). The PP1 is specified by the triple (3,12,4), PP2 by
(6,12,2), PP3 by (6,24,4), PP4 by (12,24,2), PP5 by (12,48,4), PP6 by (24, 48, 2), PP7 by (24, 96, 4), and PP8 by the triple
(6, 96, 16). The scattered pilot patterns PP1 and PP2 are transmitted with normalized amplitude of 4/3, PP3 and PP4 with 7/4,
while PP5-PP7 with normalized amplitude 7/3. As an illustration, the pilot patterns used in MISO mode for 32 K with g ¼ 1/
16 are PP2 and PP8. The corresponding PPs used in single-input single-output (SISO) mode are PP2, PP8, and PP4.
426 7 OFDM for Wireless and Optical Communications

7.8 OFDM for Wi-Fi, LTE, and WiMAX

The OFDM is widely used in Wi-Fi, Long-Term Evolution (LTE), and Worldwide Interoperability for Microwave Access
(WiMAX). The Wi-Fi Alliance was established in 1999 to serve as a non-profit organization to promote the Wi-Fi technology
and corresponding product certification. The Wi-Fi is based on IEEE 802.11 standard family, representing the set of MAC
and PHY standards implementing WLAN computer communications in 0.9, 2.4, 3.6, 5, and 60 GHz frequency bands
[58–60]. The first wireless networking standard from this family was 802.11–1997; however, the 802.11b was widely
accepted, which was followed by 802.11a, 802.11 g, 802.11n, and 802.11 ac. Other standards in this family (such as c–f,
h, j) represent the service amendments to the existing ones that are used to extend the scope, or eventually the amendments
represent the corrections to the previous specification. The basic properties of some relevant IEEE 802.11 PHY standards are
summarized in Table 7.3, based on references [58–63].
For additional details, the interested reader is referred to [58–63] and references therein. The OFDM system parameters in
IEEE 802.11a are summarized in Table 7.4. Eight OFDM channels with bandwidth of 20 MHz can be supported in UNIII
bands I and II of 5GHz band (5.15–5.35 GHz), and an additional four OFDM channels can be supported in the upper UNII
band III (5.725–5.825 GHz). The block diagram of OFDM transceiver is similar to that shown in Fig. 7.6. The preamble,

Table 7.3 Some relevant IEEE 802.11 PHY standards

Approximate
range Release date/
IEEE 802.11 [Frequency (GHz), Modulation [indoor (m), expected release
standard bandwidth (MHz)] technique Aggregate data rate (Mb/s) outdoor (m)] date
802.11–1997 [2.4, 22] DSSS, FHSS 1, 2 [20, 100] Jun. 1997
802.11a [5, 20] OFDM 6, 9, 12, 18, 24, 36, 48, 54 [35, 120] Sept. 1999
802.11b [2.4, 22] DSSS 1, 2, 5.5, 11 [35, 140] Sept. 1999
802.11 g [2.4, 20] OFDM 6, 9, 12, 18, 24, 36, 48, 54 [38, 140] Jun. 2003
802.11n [2.4/5, 20/40] MIMO- For 400 ns guard interval (GI) [70, 250] Oct. 2009
OFDM and bandwidth (BW) of 20 MHz:
7.2, 14.4, 21.7,28.9, 43.3, 57.8, 65, 72.2
GI ¼ 800 ns, BW ¼ 20 MHz: 6.5, 13, 19.5, 26, 39,
52, 58.5, 65
802.11 ac [5, 20/40/80/160] MIMO- GI ¼ 400 ns, BW ¼ 20 MHz: 7.2, 14.4, 21.7, 28.9, [35,
] Dec. 2013
OFDM 43.3, 57.8, 65, 72.2,
86.7, 96.3
GI ¼ 800 ns, BW ¼ 20 MHz: 6.5, 13, 19.5, 26, 39,
52, 58.5, 65, 78, 86.7
GI ¼ 400 ns, BW ¼ 160 MHz: 65, 130, 195, 260,
390, 520, 585, 650, 780, 866.7
GI ¼ 800 ns, BW ¼ 160 MHz: 58.5, 117, 175.5,
234, 351, 468, 702, 780
802.11ad [60, 2160] OFDM, 6912 [60, 100] Dec. 2012
single-carrier
802.11ay [60, 8000] OFDM, 100,000 [60, 1000] 2017
single-carrier

Table 7.4 OFDM system parameters in IEEE 802.11a PHY standard

OFDM symbol duration 4 μs
Number of subcarriers actually used/number of pilots 52/4
Subcarrier spacing 312.5 kHz
Signal bandwidth 16.66 MHz
Guard interval duration 800 ns
Channel spacing 20 MHz
Modulation formats BPSK, QPSK, 16-QAM, 64-QAM
Code rates ½, 9/16, 2/3, 3/4
Aggregate data rates 6, 9, 12, 18, 36, 48, 54 Mb/s
7.8 OFDM for Wi-Fi, LTE, and WiMAX 427

which precedes the OFDM packed, is composed of three sections of durations 8 μs, 8 μs, and 4 μs, respectively. The first
section is used in automatic gain control (AGC) and coarse frequency offset estimation, and it is composed of ten training
symbols with duration of 800 ns each. These training symbols are generated by employing only nonzero QAM symbols, with
located subcarrier indices being a multiple of 4, namely, {
24,
20,
16,
12,
8,
4, 4, 8, 12, 16, 20, 24}, wherein the
subcarrier indices range from
26 to 26. Given that for the digital signal with a symbol duration of T the shortest measurable
frequency offset is 1/(2 T ), we conclude that the offset up to 1/(2  800 ns) ¼ 625 kHz can be estimated. If the whole FFT
duration is used for estimation, namely, 4 μs–0.8 μs ¼ 3.2 μs, the frequency offset of 156 kHz can be estimated, so that the
corresponding relative frequency error of 26.89 ppm at carrier frequency 5.8 GHz can be measured. The sequence of short
training symbols is also used for frame detection by correlating to neighboring short symbols (pulses), and the frame will be
detected once correlator outputs exceed a desired threshold. The AGC circuit can adjust operation once the frame is detected.
The sequence of short training symbols is followed by a long training symbol (with duration of 8 μs), composed for
52 QPSK-modulated subcarriers, which is used for fine-frequency offset estimation, timing, and channel estimation, with
corresponding concepts explained already above. Given the duration of this long training symbol, it is possible to place 8 μs/
3.2 μs ¼ 2.5 FFT frames and therefore perform the cyclic extension. This first portion with duration of 1.6 μs is used as guard
interval and contains the copy of the last 1.6 μs of FFT part. By measuring the frequency drift of two samples separated by
3.2 μs, the fine-frequency offset estimation is possible. Moreover, the long training symbol is used to estimate the amplitude
for coherent (synchronous) detection. The training symbols are properly selected so that the PAPR of 3 dB is obtained,
significantly lower than in data carrying OFDM symbols, thus avoiding any nonlinear distortion due to power amplifier.
The last 4 μs of preamble contains the information about modulation type, code rate employed, and the duration of the
packet length.
Even after the fine-frequency offset compensation, there will be the uncompensated frequency drift causing the common
phase error (drift) for all subcarriers, and to compensate for the common phase error, 4 out of 52 subcarriers are used as pilots,
located at subcarrier indices {
21,
7, 7, 21}. The pilots are scrambled by a pseudorandom sequence of length 127 to avoid
the spectral lines in power spectral density of the OFDM spectrum.
To adjust the error correction strength to time-varying channel conditions, the punctured convolutional code is used with
possible code rates already specified in Table 7.4. Clearly, the lowest data rate would be [(52–4)/410
6]  1/2 b/s ¼ 6 Mb/s.
On the other hand, the highest possible data rate in this standard would be (48/410
6)  (log264)  3/4 b/s ¼ 54 Mb/s.
The IEEE 802.11 g PHY standard is very similar to 801.11a, except that it is related to the 2.4 GHz unlicensed ISM band
(2.4–2.497 GHz).
To increase the aggregated data rates in IEEE 802.11n and 802.11 ac PHY standards, the MIMO-OFDM concept,
explained in Chap. 8, is used. Namely, through MIMO concept, several independent data streams can be simultaneously
transmitted increasing the aggregate data rates. The IEEE 802.11n PHY standard is applicable to both 2.4 and 5 GHz bands
while 802.11 ac to 5 GHz band only. In IEEE 802.11n standard, the available channel bandwidths are {20 MHz, 40 MHz},
while in 802.11 ac the available channel bandwidths are {20, 40, 80, 160} MHz. The highest data rate in 802.11 ac standard,
for 160 MHz bandwidth and GI of 400 ns, is 780 Mb/s.
Recent 802.11ax (Wi-Fi 6E) standard, also known as a high-efficiency Wi-Fi, adopted in 2020, is designed to operate in
license-exempt bands (1–7.125 GHz), including commonly used 2.4 and 5 GHz as well as the 6 GHz band (5.925–7.125 GHz
in the USA) [64], and it is based on OFDMA and multi-user (MU)-MIMO. The purpose of the Wi-Fi 6E standard is to enhance
throughput/area for high-density scenarios including shopping malls, corporate offices, and dense residential areas. The
modulation formats range from BPSK to 1024-QAM, with code rates from 1/2 to 5/6. The data rates range from 135 Mb/s (for
20 MHz channels) to 1.201 Gb/s (for 120 MHz channels). The guard intervals are extended to 0.8 μs, 1.6 μs, or 3.2 μs so that
the corresponding systems are more tolerant to the delay spread. Compared to the 802.11 ac, the OFDM symbol duration is
increased four times (extend to 12.8 μs), while the subcarrier spacing is reduced four times.
The third Generation Partnership Project (3GPP) represents an international standardization body working on 3G Universal
Terrestrial Access Network (UTRAN) and GSM specifications. This standardization body developed the 3G radio access
widely known as Long-Term Evolution (LTE) or Evolved UTRAN as well as the evolved packet access core network in the
System Architecture Evolution (SAE) [60]. The 3GPP standards are reported as so-called Releases. For instance, Release
8 (2008 Q4) represents the first LTE release and all-IP SAE, based on OFDMA, frequency-domain equalization (FDE), and
MIMO-based interface, not compatible with previous CDMA-based interfaces. Release 10 (2011 Q1) describes the LTE
Advanced standard satisfying the International Mobile Telecommunications-Advanced (IMT-Advanced) 4G requirements
and being backward compatible with Release 8. The LTE Advanced can use up to 8  8 MIMO signal processing and
128-QAM in downlink direction, and with 100 MHz in aggregated bandwidth, it provides almost 3.3 Gb/s in peak download
rates per sector of the base station under perfect channel conditions. LTE Advanced Pro represents the 3GPP Releases 13 and
428 7 OFDM for Wireless and Optical Communications

Table 7.5 OFDM system parameters and other relevant parameters in WiMAX IEEE 802.16e standard
Channel bandwidth 1.25 MHz 5 MHz 7 MHz 8.75 MHz 10 MHz 20 MHz
Sampling frequency 1.4 5.6 MSa/s 8 MSa/s 10 MSs/s 11.2 MSa/s 22.4 MSa/s
The FFT size 128 512 1024 1024 1024 2048
Number of subchannels 2 8 16 16 16 32
Subcarrier spacing (Δf) 10.94 kHz 10.94 kHz 7.81 kHz 9.77 kHz 10.94 kHz 10.94 kHz
Symbol duration (T = 1/Δf) 91.4 μs 91.4 μs 128 μs 102.4 μs 91.4 μs 91.4 μs
Guard interval duration 11.425 μs 11.425 μs 16 μs 12.8 μs 11.425 μs 11.425 μs
(TG = T/8)
OFDM symbol duration 102.825 μs 102.825 μs 144 μs 115.2 μs 102.825 μs 102.825 μs
(Ts = T + TG)
Number of OFDMA symbols 48 48 34 43 48 48
per frame
Frame duration 4.94 ms ffi 5 ms 4.94 ms ffi 5 ms 4.89 ms ffi 5 ms 4.95 ms ffi 5 ms 4.94 ms ffi 5 ms 4.94 ms ffi 5 ms
Modulation QPSK, 16-QAM, 64-QAM
Channel Convolutional coding (CC): 1/2, 2/3,3/4
Coding Block turbo coding (BTC): 1/2, 3/4
Rates LDPC coding: 1/2, 2/3,3/4, 5/6
Convolutional turbo coding (CTC) in both downlink (DL) and uplink (UL): For QPSK and 16-QAM: 1/2, 3/4;
for 64-QAM: 1/2, 2/3,3/4, 5/6

14. Another relevant 3GPP standard-air interface for the 5G networks is the 5G NR (New Radio), also known as Release
15 [65]. The 5G NR employs two frequency bands: frequency range 1 (FR1) 410 MHz–7.125 GHz and frequency range
2 (FR2) 24.25–52.6 GHz. Early FR1 deployments employ 5G NR software on 4G hardware, which are slightly better than
newer 4G systems, and essentially represent 4.5G technology. The 5G-Advanced standard is currently under development by
3GPP, known as Release 18 [66].
WiMAX represents one of the most popular broadband wireless (BWA) technologies, aiming to provide high-speed
bandwidth access for wireless MANs [60, 69–71]. WiMAX is based on IEEE 802.16 set of standards, in particular IEEE
802.16-2004 (fixed-WiMAX) and IEEE 802.16e-2005 (mobile-WiMAX) standards, which provide multiple PHY and MAC
options. WiMAX represents the last mile wireless broadband access alternative to the cable and DSL services, and it is a direct
competitor of the LTE Advanced standard. The IEEE 802.16e-2005 provides several improvements compared to the IEEE
802.16-2004 including [69–71] (1) providing support for mobility by enabling the soft and hard handover between base
stations; (2) introducing the concept of scalable OFDM (SOFDM) by scaling the FFT to the channel bandwidth to ensure that
the carrier spacing constant across different channel bandwidths, ranging from 1.25 MHz through 5 MHz and 10 MHz all way
to 20 MHz, and thus improving the spectral efficiency; (3) introducing the advanced antenna diversity schemes and hybrid
automatic repeat-request (HARQ); (4) introducing the adaptive antenna systems (AAS) and MIMO signal processing;
(5) providing the denser sub-channelization and therefore improving the indoor penetration; (6) introducing downlink
subchannelization, in order to trade the coverage for capacity or vice versa; (7) introducing the LDPC coding to improve
the error correction strength; and (8) introducing an extra QoS class for VoIP applications. The basic OFDM parameters and
other relevant system parameters in IEEE 802.16e-2005 are summarized in Table 7.5.
For OFDM PHY, the following FEC schemes are used: (1) mandatory concatenated RS-convolutional code (RS), in which
RS code is used as the outer code and rate-compatible CC code as the inner code; (2) optional block turbo coding (BTC); and
(3) optional convolutional turbo coding (CTC). The RS-CC concatenated code has already been discussed above; for
additional detail the interested reader is referred to Chap. 9. The BTC and CTC schemes are fully described in Chap. 9. In
the burst mode, when we are concerned with correcting both burst and random errors, the following FEC schemes are used:
(1) mandatory tail-biting CC code as specified in IEEE 802.16 standard or zero-tailing CC in WiMAX profiles; (2) CTC,
which is optional in IEEE 802.16 standards, but mandatory in WiMAX profiles; (3) optional TC; and (4) optional LDPC
coding. In September 2018 the IEEE 802.16-2017 standard was released, relevant for combined fixed and mobile point-to-
multipoint BWA systems providing multiple services [67]. The IEEE 802.16 Working Group ceased their activities on
9 March 2018 [68].
7.9 OFDM in Ultra-Wideband Communication (UWC) 429

7.9 OFDM in Ultra-Wideband Communication (UWC)

Ultra-wideband (UWB) communication represents a fast-emerging RF technology enabling short-range high-bandwidth

communications with high energy efficiency while occupying a large RF bandwidth [72–75]. In the USA, UWB radio
communication links operate in the range from 3.1 GHz up to 10.6 GHz, satisfying FCC 15.517(b,c) requirements (for indoor
applications). The available bandwidth is 7.5 GHz, and the maximum allowable output power is
41.3 dBm/MHz. There are
two major UWB technologies: (1) impulse radio UWB (IR-UWB) and (2) multiband OFDM (MB-OFDM). The UWB
technologies are fully described in Chap. 10; here we briefly introduce the MB-OFDM concept [72, 73, 75].
The multiband modulation represents approach to modulate information in UWB systems, by splitting the 7.5 GHz band
into multiple smaller frequency bands. The key idea is to efficiently utilize UWB spectrum by simultaneously transmitting
multiple UWB signals. The MB-OFDM system can be interpreted as a combination of OFDM and frequency hopping. With
this approach, the information bits are “spread” across the whole UWB spectrum, exploiting, therefore, frequency diversity
and providing the robustness against multipath fading and narrowband interference, originating from Wi-Fi channels. The
7.5 GHz UWB spectrum can be divided into 14 bands, each with bandwidth of Δf ¼ 528 MHz, as illustrated in Fig. 7.34. The
center frequency of the nth band fc,n, the low boundary frequency fl,n, and the upper boundary frequency fh,n can be
determined, respectively, as

f c,n ¼ f 0 þ nΔf , f l,n ¼ f c,n
Δf =2, f h,n ¼ f c,n þ Δf =2,

ð7:104Þ
f 0 ¼ 2904 MHz:

The MB-OFDM transmitter configuration is provided in Fig. 7.35. Clearly, the transmitter configuration is very similar to
conventional OFDM transmitter, shown in Fig. 7.6a, except from the time-frequency coding block. The time-frequency code
(TFC) denotes which band within the band group is occupied at a given transmission interval. As an illustration, the TFCs of
length 6 for band group 1 are given as [72]: {[1 2 3 1 2 3], [1 3 2 1 3 2], [1 1 2 2 3 3], [1 1 3 3 2 2], [1 2 1 2 1 2], [1 1 1 2 2 2]}.
When TFC code 1 is used, the first OFDM symbol is transmitted on band 1, the second OFDM symbol on band 2, the third
OFDM symbol on band 3, the fourth OFDM symbol on band 1, the firth OFDM symbol on band 2, and the sixth OFDM
symbol on band 3; and this sequence of bands within the band group is repeated until the TFC is changed in time-frequency
coding block.
The binary data bits are scrambled and then encoded employing a convolutional code of rate 1/3, with constraint length
K ¼ 7, and generating polynomials g0 ¼ 1338, g1 ¼ 1658, and g2 ¼ 1718 (the subscript denotes octal number system).
Higher-rate convolutional codes (of rate 11/32, 1/2, 5/8, and 3/4) can be obtained by puncturing. The number of subcarriers
used in MB-OFDM is 128, 100 of which are used to transmit QPSK data, 12 are pilots, and 10 are guard tones (the remained
subcarriers are set to 0). The pilots are used for channel estimation and carrier-phase tracking on the receiver side. To improve
the performance of MB-OFDM scheme, LDPC-coded MB-OFDM can be used instead. It has been shown [76, 77] that the
transmission distance over wireless channel of LDPC-coded MB-OFDM can be increased by 29%–73% compared to that of
MB-OFDM with convolutional codes. After pilot insertion, serial-to-parallel conversion (S/P), inverse FFT (IFFT) operation,
and parallel-to-serial (P/S) conversion, cyclic extension and windowing are performed. After D/A conversion (DAC) and
time-frequency coding, the RF signal is transmitted over transmit (Tx) antenna. The transmitted MB-OFDM signal can be
written as [5, 73]

Band group # 1 Band group # 2 Band group # 3 Band group # 4 Band group # 5

Band Band Band Band Band Band Band Band Band Band Band Band Band Band
#1 #2 #3 #4 #5 #6 #7 #8 # 9 # 10 # 11 # 12 # 13 # 14

Mandatory Optional Optional Optional Optional

3432 3960 4488 5016 5544 6072 6600 7128 7656 8184 8712 9240 9768 10296 f [MHz]

Fig. 7.34 The frequency allocation in MB-OFDM for UWB communications

430 7 OFDM for Wireless and Optical Communications

Fig. 7.35 The MB-OFDM

transmitter configuration for Data modulation section
UWB communications
Binary Encoding & Bit QAM
Scrambler
input puncturing interleaver mapping
data

Cyclic extension Pilot

DAC P/S IFFT S/P
& windowing insertion

OFDM Tx

RF Tx
exp(j2 fc)
Time-frequency
coding

cos( ct)
AGC
VGA
Pre-select Timing
filter LNA LPF ADC
Timing and Remove
frequency cyclic
synchronization extension
LPF ADC
Frequency-
corrected
sin( ct) signal
OFDM Rx
Binary QAM Channel
Descrambler Decoding Deinterleaver P/S FFT S/P
output data de-mapping correction

Fig. 7.36 The MB-OFDM receiver configuration for UWB communications

( )
X
N
1 h i
sMB‐OFDM ðt Þ ¼ Re sOFDM,n ðt
nT OFDM Þ exp j2π f ðn mod 6Þ t , ð7:105Þ
n¼0

where N is the number of OFDM symbols transmitted while sOFDM, n(t) denotes the nth OFDM symbol of duration TOFDM
transmitted on carrier frequency f n mod 6. As discussed above, the carrier frequency changes over three frequencies assigned to
the band group, organized in sequences of length 6 (TFCs). The nth OFDM symbol is generated in similar fashion as we
described in Sect. 7.1.2.
The receiver configuration, shown in Fig. 7.36, is very similar to conventional OFDM receiver [Fig. 7.6b], except from
pre-select filter that is used to select OFDM signal of 528 MHz bandwidth at a given carrier frequency.

7.10 Optical OFDM Applications

In this section, we describe several selected applications of OFDM relevant in optical communications. For a more detailed
and comprehensive description, the interested reader is referred to [5] (and references therein).

7.10.1 Optical OFDM System Types

Various OFDM types can be placed into two broad categories [5]: (i) coherent optical OFDM (CO-OFDM) and (ii) direct
detection-based optical OFDM (DDO-OFDM).
7.10 Optical OFDM Applications 431

7.10.1.1 Coherent Optical OFDM (CO-OFDM) Systems

The basic principles of coherent optical OFDM (CO-OFDM) have already been introduced in Fig. 7.10. Here we provide
some additional details. As indicated in [5], the synergy between OFDM and coherent optical detection is twofold: (i) coherent
optical detection brings the required linearity of RF-to-optical up-conversion and optical-to-RF down-conversion and
(ii) brings the efficiency for channel and phase noise estimation. Given that basic operation principles of CO-OFDM have
already been described, here we provide more details of various modules within CO-OFDM system. The basic modules for
CO-OFDM systems are (1) RF OFDM transmitter, (2) RF OFDM receiver, (3) RF-to-optical (RF2O) up-converter based on
either MZM or optical I/Q modulator, and (4) optical-to-RF (O2RF) down-converter based on coherent optical detection. The
details of RF OFDM transmitter and receiver have been provided already in Figs. 7.7 and 7.10, together with the description of
operation principles. There are two options for RF-to-optical up-conversion, as indicated in [5], namely, direct RF2O
up-conversion and RF2O up-conversion through RF intermediate frequency, with both options shown in Figs. 7.37 and
7.38. The detailed description of operation principles of balanced coherent optical detector, Mach-Zehnder modulator (MZM),
and optical I/Q modulator has been provided already in Chap. 3. To perform RF2O up-conversion in direct up-conversion
configuration, the optical I/Q modulator is used [Fig. 7.37a]; and to perform O2RF down-conversion, the balanced coherent
optical detector is used [Fig. 7.37b]. On the other hand, to implement CO-OFDM transmitter in RF intermediate-frequency-
based configuration, RF up-conversion is combined with optical RF-to-optical converter based on single MZM followed by
an optical band-pass filter (OBPF), as shown in Fig. 7.38a. To implement CO-OFDM receiver in RF intermediate-frequency-
based configuration, the optical-to-RF down-converter, implemented with the help of an OBPF and an optical π-hybrid-based
balanced optical detector, is combined with an RF down-converter, as shown in Fig. 7.38b.
The advantages of direct up-configuration are twofold [5]: (1) the required electrical bandwidth in transmitter and receiver
is significantly lower, and (2) there is no need for the image rejection filters (BPF and OBPF) in transmitter and receiver. The
MZMs in either configuration are operated at null point (see Chap. 3 for additional details). It is also possible to use direct RF-
to-optical up-conversion on the transmitter side and RF intermediate-frequency-assisted configuration on the receiver side and
vice versa.
By omitting the distortions due to imperfect channel and phase estimations, the receiver sensitivity will be identical to that
of single-carrier in ASE noise-dominated scenario. For instance, the BER performance of CO-OFDM with QPSK can be
described as

Fig. 7.37 The block diagram of RF-to-optical up-convertor

CO-OFDM system in direct In-phase (I) RF component
optical-to-RF up-conversion
configuration. Architectures of (a) Optical link
transmitter and (b) receiver for MZM
CO-OFDM Data in RF OFDM Laser
Tx diode
MZM /2
Y-junction

Quadrature (Q) RF component I/Q optical modulator

(a)

Optical-to-RF down-convertor

From optical link /2 I’

Estimated
RF OFDM
Q’ receiver
Local data
Oscillator
Laser

Balanced coherent optical detector

(b)
432 7 OFDM for Wireless and Optical Communications

Fig. 7.38 The block diagram of RF up-converter

CO-OFDM system in optical-to- RF-to-optical up-converter
Data in RF OFDM
RF up-conversion through RF Optical link
BPF
intermediate frequency. Tx
Architectures of (a) transmitter
Laser
and (b) receiver for CO-OFDM RF MZM OBPF
diode
oscillator

(a)
Optical-to-RF down-convertor

RF local
From optical link oscillator
OBPF
Estimated
RF OFDM
-hybrid BPF
Local receiver data
oscillator RF down-converter
laser

Balanced optical detector

(b)

pffiffiffiffiffiffiffiffi σ2
1
BER ’ erfc ρ=2 , ρ ¼ 2s ð7:106Þ
2 σ wn

where ρ is signal-to-noise ratio, defined as the ratio of the signal variance σ 2s and the white noise variance σ 2wn . The symbol
SNR (for QPSK) is related to OSNR, number of subcarriers Nsc, and subcarrier spacing Δf by [5, 78]

Bref
ρ ffi 2OSNR , ð7:107Þ
N sc Δf

where Bref is the reference bandwidth (12.5 GHz in 0.1 nm at 1550 nm).

7.10.1.2Direct Detection Optical OFDM (DDO-OFDM)

There exist many variants of DDO systems [5, 8–11, 79–81], which can be classified into broad categories [5]: (i) linearly
mapped DDO-OFDM in which there exists a linear mapping between baseband OFDM and optical field and (ii) nonlinearly
mapped DDO-OFDM in which there exists a linear mapping between the baseband OFDM and optical intensity instead.
In linearly mapped DDO-OFDM, the optical carrier frequency should be at least the OFDM signal bandwidth from the
main optical carrier, so the transmitted signal can be represented in complex from as
 j2πΔ f G
 j2π f o t
sðt Þ ¼ 1 þ ae sBB ðt Þ e , ð7:108Þ

where ΔfG is the guard frequency band between the main optical carrier frequency fo and OFDM signal, a is the coefficient
related to the ratio of OFDM signal power and the main optical carrier, while the baseband OFDM signal is defined by

NX
sc =2
j2π f k t
sBB ðt Þ ¼ xk e , ð7:109Þ
k¼
N sc =2þ1

with xk being the symbol carried by the k-th subcarrier fk. The received signal, in the presence of chromatic dispersion, can be
represented as
" #
X
N sc =2
jΦCD ð
Δ f G Þ j2πΔ f G j2π f k tþjΦCD ð f k Þ j2π f o tþjϕðtÞ
rðtÞ ¼ e þ ae xk e e ð7:110Þ
k¼
N sc =2þ1
7.10 Optical OFDM Applications 433

where ΦCD( fk) is the phase distortion of the k-th subcarrier introduced by chromatic dispersion and ϕ(t) is the random phase
shift of the main optical carrier. When the second-order GVD can be neglected, the phase distortion is quadratic [5]:

ΦD ð f k Þ ¼ πcDtot ð f k = f o Þ2 , ð7:111Þ

where Dtot denotes the accumulated chromatic dispersion (expressed in ps/pm) and c is the speed of light. After the PIN
photodetector, the photo current I(t) will be proportional to the magnitudes squared of the received optical field:
 2
 NX
sc =2

 
I ðt Þ / jr ðt Þj2 ¼ e jΦCD ð
Δ f G Þ
þ ae j2πΔ f G t
xk e j2π f k tþjΦCD ð f k Þ

 k¼
N sc =2þ1 
8 9
< NX
sc =2 =
¼ 1 þ 2a Re e j2πΔ f G t
xk e j2π f k tþj½ΦCD ð f k Þ
ΦCD ð
Δ f G Þ ð7:112Þ
: k¼
N sc =2þ1
;
NX
sc =2
j2π ð f k1
f k2 Þþj½ΦCD ð f k1 Þ
ΦCD ð f k2 Þ
þjaj2 xk1 xk2 e
k 1 , k2 ¼
N sc =2þ1

The DC component can easily be removed by DC blocker, while the third-term represents the second-order nonlinearity
that must be properly canceled out. The second-term contains the desired OFDM signal. To improve the spectral efficiency,
the single-side band (SSB) transmission is advocated in [11, 13]. To remove one of the bands (either the upper or lower band),
the optical filtering approach is used in [11, 13]. Alternatively, the Hilbert transform approach for SSB generation can be used
at the expense of increased complexity in digital domain on the transmitter side [26, 27]. Namely, the quadrature signal will
represent the OFDM baseband signal passed through the Hilbert transform filter, and in passband representation, the
quadrature signal will represent an interference canceling the desired upper/lower band of the double-side band (DSB) signal.
The quadratic term is then filtered out in electrical domain (after the photodetection takes place). Alternatively, the parameter
a in Eq. (7.108) should be chosen sufficiently low so that the quadratic term becomes negligible. It is also possible to estimate
the quadratic term in an iterative fashion, by using the estimate linear term to cancel the quadratic term [82]. It is also possible
to transmit QAM/PSK symbols only on odd subcarrier; the second-order term will introduce the crosstalk only on even
subcarrier, not carrying any information [83]. Each of these approaches to deal with the second-order distortion has
advantages and disadvantages as discussed in [5].
As indicated earlier, in nonlinearly mapped DDO-OFDM, we establish the linear mapping between the baseband OFDM
signal, defined by (7.109), and optical intensity by employing the direct modulation of the laser diode as follows:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
j2π f o t
Eðt Þ ¼ jE ðt Þje , j E ðt Þj ¼ E 0 1 þ a Re fe j2π f IF t sBB ðt Þg, ð7:113Þ

where E(t) is the complex representation of laser diode output field and fIF is the RF intermediate frequency (IF). By ignoring
the dispersion effects, the PIN photodetector photocurrent would be proportional the OFDM baseband signal:
  
I ðt Þ / jE ðt Þj2 ¼ E 20 1 þ a Re e j2π f IF t sBB ðt Þ
8 8 99
< < NX
sc =2 == ð7:114Þ
¼ E20 1 þ a Re e j2π f IF t xk e j2π f k t :
: : k¼
N =2þ1
;;
sc

For additional details on schemes belonging to this type, the interested reader is referred to [5] (and references therein).

7.10.2 High-Speed Spectrally Efficient CO-OFDM Systems

Among various alternatives considered in the previous section, the CO-OFDM offers better spectral efficiency, receiver
sensitivity, and tolerance to chromatic dispersion. By employing polarization states, by either performing the polarization-
434 7 OFDM for Wireless and Optical Communications

division multiplexing (PDM) of two independent optical data streams or four-dimensional (4D) signaling, the spectral
efficiency can be doubled.

7.10.2.1 Polarization-Division Multiplexed OFDM

In conventional polarization-division multiplexed CO-OFDM scheme, two independent two-dimensional (2D) signal con-
stellation data streams are used as inputs of two RF OFDM transmitters, which is similar to the OFDM transmitter
configuration described earlier (see Figs. 7.7 and 7.10). One transmit laser diode is used for both polarizations, with output
being split by polarization beam splitter (PBS). Also, two I/Q modulators are used, one for each polarization, as shown in
Fig. 7.39(a). The 2D (QAM data streams) are processed in RF OFDM transmitter in the same fashion as described in Sects.
7.4.2, 7.4.3, and 7.5. The outputs of RF OFDM transmitter present the real and imaginary parts of OFDM signal, which are
used as I and Q inputs to corresponding I/Q modulator. Upon electro-optical conversion by I/Q modulator, two independent
optical signals corresponding to x- and y-polarizations are combined together by polarization beam combiner (PBC) and
transmitted over optical transmission system in question. At the receiver side, the incoming optical signal is split into two
orthogonal polarizations by PBS, which is also done with the signal coming from local laser oscillator. The x-polarization
outputs of PBSs are used as inputs to the upper branch of the balanced coherent detector, while the y-polarization outputs are
used as inputs to the down-branch of the balanced coherent detector, as shown in Fig. 7.39b. The configuration of coherent
balanced detector has been provided already in Sec. 7.10.1.1 (see also Chap. 3). The outputs of balanced coherent detectors
represent estimates of real and imaginary parts of corresponding RF OFDM receiver, whose configuration is shown in
Figs. 7.7 and 7.10. After FFT demodulations in x- and y-polarization branches, chromatic dispersion compensation,
PMD/PDL compensation, and symbol detection have been performed. In the presence of polarization-dependent loss
(PDL), we also need to perform the matrix inversion by using the parallel decomposition (see Fig. 7.8 and corresponding
text related to it). This operation is not computationally extensive given the fact that the channel matrix size is 2  2.
Another alternative is to use polarization-time (PT) coding. One such scheme is shown in Fig. 7.40. The scheme, which is
based on Alamouti code described later in Chap. 8 (see also [56]), was proposed in [57] (see also ref. [5]). The operations of all
blocks on the transmitter side, except that of the PT encoder, are similar to those shown in Fig. 7.39. The functional blocks at
the receiver side are identical to the blocks shown in Fig. 7.39, except from symbol detection that is based on the Alamouti-
type combiner. The operation principle of the encoder from Fig. 7.40 can be described as follows: (i) in the first half of i-th
time instance (“the first channel use”), symbol sx is sent by using x-polarization channel, while symbol sy is sent by using
y-polarization channel; (ii) in the second half of i-th time instance (“the second channel use”), symbol –s*y is sent by using
x-polarization channel, while symbol s*x is sent by using y-polarization.
The received symbol vectors for the first and second channel use can be written as

ðmÞ ðmÞ jðϕT
ϕLO Þ ðmÞ

ri,k ¼ U ðkÞsi,k e þ zi,k , m ¼ 1, 2 ð7:115Þ

where the Jones (channel) matrix U(k) has been already introduced by Eqs. (7.18) and (7.42) (we use again index k to denote
h iT
ðmÞ ðmÞ ðmÞ
the frequency ωk of k-th subcarrier), ri,k ¼ r x,i,k , r y,i,k denotes the received symbol vector in the m-th (m ¼ 1,2) channel use
ðmÞ ðmÞ ðmÞ T
of i-th OFDM symbol and k-th subcarrier, while zi,k ¼ ½zx,i,k , ny,i,k  denotes the corresponding noise vector. We also use
 T h iT
ð1Þ ð2Þ
si,k ¼ sx,i,k , sy,i,k to denote the symbol transmitted in the first channel use and si,k ¼
sy,i,k , sx,i,k to denote the symbol
transmitted in the second channel use (of the same symbol interval). Because the symbol vectors transmitted in the first and the
second channel use of i-th time instance are orthogonal, Eq. (7.115) can be rewritten by grouping separately x- and
y-polarizations, so we have that
" ð1Þ
# " #  " ð1Þ #
r x,i,k jϕPN jϕPN zx,i,k
U xx e U xy e sx,i,k
¼ þ ð2Þ , ð7:116Þ
ð2Þ
rx,i,k U xy e
jϕPN
U xx e
jϕPN sy,i,k zx,i,k
2 3 " # 2 3
ð1Þ  ð1Þ
r y,i,k U yx e jϕPN
U yy e jϕPN
sx,i,k zy,i,k
4 5¼ þ 4 ð2Þ 5: ð7:117Þ
ð2Þ
r  y,i,k U yy e
jϕPN
U yx e
jϕPN sy,i,k z
y,i,k
7.10 Optical OFDM Applications 435

QAM symbol
data stream RF OFDM
(x-pol.) transmitter (x pol.)
4-D E/O modulator

I/Q
modulator
Transmit (x-pol.)
to fiber
laser PBS PBC
diode I/Q
modulator
(y-pol.)
QAM symbol
data stream RF OFDM
(y-pol.) transmitter (y pol.)
(a)

From Balanced coherent

PBS detector (y pol.)
SMF

Local Balanced coherent

PBS detector (y pol.)
laser

Estimated QAM
ADC LPF Cyclic extension Chromatic dispersion data stream
FFT
…

removal compensation, (x-pol.)

ADC LPF
PMD/PDL compensation
& Estimated QAM
symbol detection by
ADC LPF Cyclic extension Alamouti-type PT data stream
FFT
…
…

ADC LPF removal decoder (y-pol.)

(b)
Fig. 7.39 Polarization-division multiplexed CO-OFDM scheme: (a) transmitter architecture and (b) receiver architecture. PBS(C) polarization
beam splitter (combiner)

where ϕPN ¼ ϕT – ϕLO. If only one polarization is used, we could solve either Eq. (7.116) or Eq. (7.117). However, the use of
only one polarization would result in 3 dB penalty with respect to the case when both polarizations are employed. Following
the derivation similar to that described in the previous section, it can be shown that the estimates of transmitted symbols at the
output of PT decoder (for ASE noise-dominated scenario, neglecting fiber nonlinearities) can be obtained as
436 7 OFDM for Wireless and Optical Communications

Ix
QAM [ s*y sx ] DAC LPF
S/P IFFT Cyclic extension

…
symbols conv. insertion Qx
block
DAC LPF
(x-pol.)
Alamouti-type
PT encoder Iy
QAM
DAC LPF
symbols S/P IFFT Cyclic extension

…
conv. block insertion Qy
(y-pol.) [s *
x sy ] DAC LPF

Fig. 7.40 The transmitter architecture of PT coding scheme combined with CO-OFDM

ð1Þ ð2Þ ð1Þ ð2Þ

esx,i,k ¼ U xx r x,i,k e
jϕPN þ U xy r x,i,k e jϕPN
þ U yx r y,i,k e
jϕPN þ U yy r y,i,k e jϕPN
, ð7:118Þ

ð1Þ ð2Þ ð1Þ ð2Þ

esy,i,k ¼ U xy r x,i,k e
jϕPN
U xx r x,i,k e jϕPN
þ U yy r y,i,k e
jϕPN
U yx r y,i,k e jϕPN
, ð7:119Þ

where esx,i and esy,i denote the PT decoder estimates of symbols sx,i and sy,i transmitted in ith time instance. If only one
polarization were used (i.e., x-polarization), the last two terms in Eqs. (7.118) and (7.119) must be omitted.

7.10.2.2 OFDM-Based Superchannel Transmission

The growth of the Internet traffic does not appear to be leveling off any time soon, and it is projected to continue to grow
exponentially in the years to come [84–90]. This exponential-like traffic growth places enormous pressure on the underlying
information infrastructure at every level, from the core to access networks. The growth is driven mainly by new services,
particularly high-quality video and cloud computing, which demand huge bandwidths. As for the market demand, the IP
traffic in nationwide networks has grown so rapidly that some large Internet service providers have already reported huge
demands in router-to-router trunk connectivity exceeding 100 Gb/s while expressing the desire to have Terabit Ethernet (TbE)
ports in the near future. As for the standardization activities, several industry forums (ITU-T, IEEE 802.3ba, and OIF) have
completed the work on 100 Gb/s Ethernet (100 GbE), while the activities for an early adoption of 400 Gb/s Ethernet and 1 Tb/
s Ethernet and beyond are underway, with the expectation to be completed in the next couple of years.
As the data rates exceed 100 Gb/s, the electrical bandwidth required for OFDM becomes a bottleneck [5, 85, 86]. To
overcome this problem, the orthogonal band multiplexed OFDM (OBM-OFDM) is proposed in [84]. This method might
come under different names such as multiband OFDM [88, 89] and superchannel OFDM [86, 90]. The key idea behind this
approach is to divide the entire OFODM spectrum into multiple orthogonal bands [5, 84]. To achieve so, electrical domain-
based approaches, optical domain-based approaches, or a combination of the two is possible [5, 84, 91–93]. Let Δf be
subcarrier spacing and ΔfG the frequency band guard between neighboring OFDM bands. The orthogonality condition among
different OFDM bands would be satisfied when the frequency band guard is a multiple of subcarrier spacing [5]:

Δ f G ¼ nΔf , n
1: ð7:120Þ

A generic superchannel transponder employing WDM is shown in Fig. 7.41. The term superchannel is used here to denote
a collection of optical carriers modulated and multiplexed together at the transmitter side [86, 90]. As we see from Fig. 7.41,
the superchannel content is generated by applying 4D signals (two phase channels and two polarizations) on a single optical
carrier (or better to say an optical subcarrier) by using 4D modulator, being composed of two I/Q modulators, one polarization
beam splitter and one polarization beam combiner, as shown in Fig. 7.39a. This 4D modulator can be used either for PDM of
independent OFDM data streams or fully 4 D signaling. The OFDM signals of all optical subcarriers within the superchannel
are spectrally multiplexed, in an all-optical OFDM fashion, to create the superchannel. The superchannels are then spectrally
multiplexed together to create even larger entities, called here superchannel band groups. Finally, the superchannel band
groups are further multiplexed together by employing a frequency-locked WDM multiplexer.
The process at the receiving side is just opposite, and it is composed of three-step spectral demultiplexing, which all results
in the selection of individual optical subcarriers loaded with 4D electrical signals. It is also possible to use the anti-aliasing
7.10 Optical OFDM Applications 437

Content of superchannel band group N3

Content of superchannel 1
OFDM Tx
(x pol.)

4-D E/O
LD 1
modulator
Transmitter side DSP

OFDM Tx
(y pol.)

…
Power
…

combiner
OFDM Tx
(x pol.)

Spectral
4-D E/O multiplexer
LD N1 SMF
modulator

OFDM Tx WDM
(y pol.) multiplexer
…

Content of superchannel N2
…

Content of superchannel band group N3

(a)
Rx side corresponding to

Rx side corresponding to superchannel band group 1

LOL 1
superchannel 1

PD 1 RF OFDM Rx 1
Polarization (x pol.)
diversity PD 2
optical PD 3
hybrid RF OFDM Rx 1
Receiver side DSP

PD 4 (y pol.)
Power
…

splitter LOL
…
…

N1

Spectral
PD 1 RF OFDM Rx 1
demultiplexer Polarization (x pol.)
diversity PD 2
optical PD 3
hybrid RF OFDM Rx 1
From SMF WDM (y pol.)
…

PD 4
demultiplexer
…

Rx side corresponding to superchannel N2

…

Rx side corresponding to superchannel band group N3

(b)
Fig. 7.41 The generic superchannel OBM-OFDM enabling ultrahigh-speed optical transport: (a) transmitter side and (b) receiver side
architectures. N1 denotes the number of frequency-locked laser signal within the superchannel, N2 denotes the number of superchannels within
the superchannel band group, and N3 denotes the number of superchannel band groups. LD laser diode, LOL local oscillator laser, PD photodetector
438 7 OFDM for Wireless and Optical Communications

Superchannel Superchannel
Superchannel
band group # 1 band group # 10
band group # 4

Superchannels Superchannels Superchannels

Spectral Multiplexing

#1 #2 # 10 # 31 # 32 # 40 # 91 # 92 # 100

… … … … … Bit rate

1 TbE 10 Tb/s 10 Tb/s

40 Tb/s

Spectral content of all-optical OFDM multiplexer output (100 Tb/s)

Fig. 7.42 The digital hierarchy of superchannel OBM-OFDM enabling up to 100 Tb/s serial optical transport

optical filters to select the desired superchannel as well as the desired optical supercarrier within the superchannel. Further
processing of the 4D signals is done in a conventional coherent detection scheme with balanced optical receivers, which is
followed by the ADC and DSP processing blocks. The RF OFDM transmitter operates as described in Sects. 7.4.2, 7.4.3, and
7.5. On the other hand, the configurations of OFDM receivers, corresponding to x- and y-polarizations, are provided in
Fig. 7.7. In this particular configuration, the spectral multiplexer is implemented with the help of an all-optical OFDM
multiplexer. Clearly, all-optical OFDM is used here not for modulation, but purely for multiplexing purposes, to reduce the
overall system cost.
Now we describe the digital hierarchy enabling ultrahigh-speed optical transport. In the illustrative example shown in
Fig. 7.42, the signal frame is organized into ten superchannel band groups with center frequencies being orthogonal to each
other. Each spectral component carries 1 Tb/s Ethernet (1 TbE) superchannel, while each superchannel band group carries
10 TbE traffic. With 10 superchannel band groups, 100 Tb/s serial optical transport is possible over SMF links. In this
example, we set N1 ¼ N2 ¼ N3 ¼ 10. For some initial experimental results related to multi-Tb/s superchannel optical transport,
the interested reader is referred to [90] (and references therein).
By using the concept of orthogonal division multiplexing (ODM), where orthogonal division multiplexer is implemented
with the help of FBG-based optical filters with orthogonal impulse response, it is possible to achieve beyond 1 Pb/s serial
optical transport over SMF links as proposed in [91–93]. Now by employing multiple spatial modes in few-mode fibers
(FMFs), it is possible to achieve even beyond 10 Pb/s serial optical transport over FMF links [92, 93].
Other relevant high-speed spectrally efficient CO-OFDM systems include spectrally multiplexed generalized OFDM
(GOFDM) [88], spectrally multiplexed 4-D OFDM [94], and no-guard interval CO-OFDM [95, 96] systems.
Other applications of optical OFDM include radio-over-fiber (RoF) systems [97, 98], passive optical networks (PONs)
[99, 100], indoor wireless optical communications [101, 102], both infrared and visible light communications (VLCs), and
free-space optical communications [103]. Regarding the multimode fiber (MMF) applications, the OFDM can be used for
short-reach applications [104], such as data centers and optical broadcasting in MIMO fashion [105, 106], as well as for
medium-haul and long-haul applications [107, 108]. Given that this book is not devoted to OFDM only, in the next section we
describe the use OFDM for communication over short-reach MMF links. The reader interested on other optical OFDM
applications is referred to reference [5].

7.10.3 OFDM in Multimode Fiber Links

OFDM as technique can utilize adaptive modulation and coding as a mechanism to deal with time-varying channel conditions,
such as ones in multimode fiber (MMF) links. This scheme can be based on direct detection to reduce the system cost.
However, it still requires the use of RF up- and down-converters. A particular version of OFDM, discrete multitone (DMT)
modulation, is considered as a cost-effective solution for transmission over multimode fiber links in LAN applications
[18]. Although the same sequence is transmitted twice if DMT modulation is applied, which results in reduced data rate, DMT
offers the following two advantages as compared to traditional single-carrier approaches: (i) it has the ability to maximize the
7.10 Optical OFDM Applications 439

Fig. 7.43 The DMT system for

transmission over MMF. P/S
Data input Estimate of input data
parallel-to-serial, S/P serial-to-
parallel, DAC digital-to-analog
converter, ADC analog-to-digital
converter
Demultiplexer Multiplexer

… …
Constellation Constellation
mapper demapper
… …
IFFT FFT
modulator demodulator
… …
P/S converter and S/P converter and
cyclic extension (CE) CE remover

DAC ADC
MMF link

Bias Laser diode Photodetector

information rate by tailoring the information-bearing distribution across the channel in accordance with the channel
conditions; and (ii) it can adapt to time-varying channel conditions, which is achieved by the virtue that original data stream
is divided among many subcarriers, while the subcarriers affected by the channel conditions can be avoided through adaptive
modulation. The block diagram of DMT system for transmission over MMF links is shown in Fig. 7.43. The demultiplexer
converts incoming information data stream into parallel form. Constellation mapper maps parallel data into N subcarriers
using QAM. The modulation is performed by applying the IFFT, which transforms frequency-domain parallel data into time-
domain parallel data. The complex-valued time-domain sequence sk after IFFT can be expressed as

 
1 X
N
1
n
sk ¼ pffiffiffiffi Cn exp j2πk ; k ¼ 0, 1, . . . , N
1 ð7:121Þ
N n¼0 N

where Cn (n ¼ 0,1,. . .,N-1) is frequency-domain input QAM sequence of symbols. In DMT, the time-domain sequence is real-
valued, which is achieved by employing 2 N-IFFT instead, with input values satisfying the Hermitian symmetry property [6]:

C2N
n ¼ C n ; n ¼ 1, 2, . . . , N
1; ImfC 0 g ¼ ImfC N g ¼ 0 ð7:122Þ

meaning that the second half of input sequence (going to the IFFT block) is complex conjugate of the first half. In addition,
0-th and N-th subcarriers must be real-valued as given in Eq. (7.122). Therefore, the 2 N-point IFFT in DMT is obtained as

 
1 X
2N
1
n
sk ¼ pffiffiffiffiffiffi C n exp j2πk ; k ¼ 0, 1, . . . , 2N
1 ð7:123Þ
2N n¼0 2N
440 7 OFDM for Wireless and Optical Communications

where sk is now real-valued sequence of length 2 N. The corresponding, discrete-time signal upon parallel-to-serial conversion
can be written as

   
1 X
2N
1
T kT
s k ¼ pffiffiffiffiffiffi Cn exp j2πn ; k ¼ 0, 1, . . . , 2N
1 ð7:124Þ
2N 2N n¼0 2N

where T is the time duration of DMT frame. In P/S block (see Fig. 7.43), the cyclic extension is performed in similar fashion as
one described in previous sections. The length of cyclic extension should be longer than the maximum pulse spread due to
multimode dispersion. The digital-to-analog converter, which typically contains a transmit filter, performs the conversion
from digital into analog domain. Direct modulation of the laser diode by DMT signal is used to reduce the system cost, as
shown in Fig. 7.43. Also, the bias voltage is used to convert the negative portion of DMT signal into a positive one since the
negative signals cannot be transmitted if an intensity modulation is used in combination with direct detection. The signal
clipping can be used to improve power efficiency. At the receiver side, upon optical-to-electrical conversion by photodetector,
DC bias is blocked by photodetection, while ADC, cyclic removal, and S/P conversion have been performed. The demodula-
tion is performed by 2 N-point FFT with the output

X
2N
1  
b n ¼ p1ffiffiffiffiffiffi n
C ck exp
j2πk ; n ¼ 0, 1, . . . , 2N
1 ð7:125Þ
2N k¼0 2N

where Cb n (n ¼ 0,1,. . .,N-1) presents the estimation of transmitted sequence. Although the DMT frames can have high PAPR
values, they occur with certain probability and can be characterized using the complementary cumulative distribution function
Pr(PAPR>PAPRref), where PAPRref is the referent PAPR (usually expressed in dBs). It has been shown in [18] that Pr
(PAPR>15 dB) (when 512 subcarriers are used) is 10
4, indicating that Pr(PAPR>27 dB) (when all subcarriers add
constructively) would be even smaller. Therefore, there is no need to accommodate the full dynamic range. By limiting the
dynamic range of DAC and ADC to an appropriately chosen range, the optimum performance can be obtained, as shown in
[42]. The use of different numbers of subcarriers leads to different probability density functions of PAPR. It has been shown in
[18] that the use of a larger number of subcarriers provides better robustness to poor channel response by allowing for longer
guard intervals, for the same overhead.
The adjustment of the dynamic range of DAC and ADC can simply be achieved by clipping, as given below


sDMT ðt Þ, jsDMT ðt Þj  A
sclipped ðt Þ ¼ ð7:126Þ
0, otherwise

where sclipped(t) denotes the clipped DMT signal while A denotes the maximum tolerable amplitude before clipping. We can
determine the optimum clipping level that minimizes the BER if we measure the BER against the clipping level CL, based on
Eq. (). As an example, it was shown in [18] that for 256 subcarriers and 64-QAM, the optimum clipping ratio for different
ADC/DAC resolutions is between 8.5 and 9.5 dB. On the other side, the clipping introduces distortion to DMT signals. It is
possible to reduce the PAPR by using distortionless PAPR reduction [43–45]. Another approach to reduce the PAPR would
be through coding [46]. Some of these distortionless methods have complexity significantly higher than ones allowing
clipping. This is the reason why medium-level complexity methods, such as selective mapping [43, 44], have been proposed
in [18]. The selective mapping is based on dual use of real and imaginary parts of a complex FFT in a procedure described
below. Let us first observe the complex sequence Dn (n ¼ 0,1,. . .,N-1) of length N. Next, assume the sequence of complex
numbers of length 2 N satisfies the symmetry property

D2N
n ¼
Dn ; n ¼ 1, 2, . . . , N
1; ImfD0 g ¼ ImfDN g ¼ 0 ð7:127Þ

The sequence given by Eq. (7.127) has purely imaginary values for the 2 N-point IFFT. The IFFT of the sequence
Xn ¼ Cn + jDn (Cn is the transmitted sequence as defined above) is now given as
7.10 Optical OFDM Applications 441

 
1 X
2N
1
n
sk ¼ pffiffiffiffiffiffi X n exp j2πk ; k ¼ 0, 1, . . . , 2N
1
2N n¼0 2N
" #
X
2N
1   X
2N
1   ð7:128Þ
1 n n
¼ pffiffiffiffiffiffi C n exp j2πk þj Dn exp j2πk
2N n¼0 2N n¼0
2N
¼ Re fsk g þ jImfsk g

We have just created a two-channel input-output modulator with a single 2 N-point IFFT. By using a gray mapping for Cn,
anti-gray mapping for Dn, and applying the same input sequence Dn ¼ Cn, we will get two DMT frames with different PAPR
values. We can then transmit the DMT frame with smaller PAPR value. Receiver has to know which sequence was used at the
transmitter side. It can be shown that the same probability Pr(PAPR>PAPRref) of 10
4 is now obtained for PAPRref ~ 13.2 dB
(for 512 subcarriers and 64-QAM) [18].
The capacity of the DMT system, having N independent subcarriers with bandwidth BN and subcarrier gain {gi, i ¼ 0, . . .,
N
1}, can be expressed as

X
N
1  
g ðP þ Pbias Þ
C¼ P
max BN log 2 1 þ i i , ð7:129Þ
Pi : Pi ¼P i¼0 N 0 BN

where Pi is the optical power allocated to i-th subcarrier, N0 is the power spectral density of transimpedance amplifier, while
Pbias represents the corresponding power used to transmit the bias. The signal bias is used to convert the negative values of
DMT signal into positive ones. The subcarrier gain gi can be calculated as gi ¼ R2|Hi|2, with R being the photodiode
responsivity and Hi being the MMF transfer function amplitude of the i-th subcarrier. Since the direct detection is used,
Eq. (7.129) represents essentially a lower bound on channel capacity. Nevertheless, it can be used as an initial figure of merit.
By using the Lagrangian method, it can be shown that [5]
(
Pi þ Pbias 1=ρc
1=ρi , ρi
ρc
¼ , ρi ¼ gi ðP þ Pbias Þ=N 0 BN ð7:130Þ
P 0, otherwise

where ρi is the signal-to-noise ratio of i-th subcarrier while γ c is the threshold SNR. By substituting Eq. (7.128) into
Eq. (7.127), the following expression for channel capacity can be obtained:

X  
ρi
C¼ BN log2 : ð7:131Þ
i:ρi >ρc
ρc

The i-th subcarrier is used when the corresponding SNR is above the threshold. The number of bits per i-th subcarrier is
determined by mi ¼ bBNlog2(ρi/ρc)c, where bc denotes the largest integer smaller than the enclosed number. Therefore, the
signal constellation size and power per subcarrier are determined based on MMF channel coefficients. When subcarrier’s SNR
is high, larger constellation sizes are used, and power per subcarrier is chosen in accordance with Eq. (7.130); smaller
constellation sizes are used when the subcarrier’s SNR is low; and nothing is transmitted when the subcarrier SNR falls below
a certain threshold value. An example of adaptive QAM is shown in Fig. 7.44.

Fig. 7.44 Illustration of adaptive

QAM constellation size

QAM-based subcarrier mapping

64

32

16
Subcarrier index, n

75 150 225 300 375 450 525

442 7 OFDM for Wireless and Optical Communications

7.11 Concluding Remarks

The main subject of this chapter has been on the OFDM fundamentals and OFDM application to wireless and optical
communications. The chapter starts with Sect. 7.1, where the basics of OFDM, including the generation of OFDM signal by
inverse FFT, have been described. After that, in Sect. 7.2, several basic manipulations on OFDM symbols, required to deal
with multipath effects in wireless communications and dispersion effects in fiber-optics communications, including guard
time and cyclic extension as well as the windowing operation, have been described. Next, the bandwidth efficiency of OFDM
has been discussed in Sect. 7.3. The next topic in the chapter, Sect. 7.4, has been devoted to the OFDM system design,
description of basic OFDM blocks, and OFDM-based parallel channel decomposition, including multipath fading radio and
dispersion-dominated channel decomposition. In Sect. 7.5, following the description of CO-OFDM principles (Sect. 7.5.1),
the basic estimation and compensation steps, common to both wireless and optical communication systems, have been
described including DFT windowing (Sect. 7.5.2), frequency synchronization (Sect. 7.5.3), phase estimation (Sect. 7.5.4), and
channel estimation (Sect. 7.5.5). Regarding the channel estimation, both pilot-aided and data-aided channel estimation
techniques have been described. Following the differential detection in OFDM description (Sect. 7.6), the focus has moved
in Sect. 7.7 on various applications of OFDM in wireless communications including DAB (Sect. 7.7.1) and DVB (Sect. 7.7.2).
Wi-Fi, LTE, and WiMAX have been described in Sect. 7.8, while the OFDM application in UWB has been described in Sect.
7.9. Next, the optical OFDM applications have been discussed in Sect. 7.10, starting with description of basic OFDM system
types, Sect. 7.10.1, such as CO-OFDM systems (Sect. 7.10.1.1) and DDO-OFDM systems (Sect. 7.10.1.2). Regarding the
high-speed spectrally efficient CO-OFDM systems, described in Sect. 7.10.2, the polarization-division multiplexed OFDM
systems (Sect. 7.10.2.1) and OFDM-based superchannel optical transmission systems (Sect. 7.10.2) have been described.
Section 7.10.3 has been devoted to the application of OFDM in multimode fiber links. In Sect. 7.11 concluding remarks have
been provided. The set of problems is provided in incoming section.

7.12 Problems

1. Design the OFDM system for wireless communications with the required bit rate of 20 Mbps, tolerable delay spread of
200 ns, and bandwidth occupied being less than 16 MHz. Discuss different options. Determine corresponding bandwidth
efficiencies.
2. OFDM is the basis for the 5 GHz standard selected by the IEEE 802.11 standardization group in July 1998 to support data
rates ranging from 6 to 54 Mbps. Describe the corresponding OFDM system design. Which modulation formats and coding
rates are used for 6 and 54 Mbps?
3. An alternative description of OFDM is based on matrix representation:
2 3
xN
1
2 3 2 36 ⋮ 7 2 3
yN
1 h0 h1 ⋯ hm ⋯ 6 7 zN
1
0
6 0
7
6 yN
2 7 6 0 h0 ⋯ hm
1 hm ⋯ 0 76 x0 7 6 zN
2 7
6 7¼6 76 7þ6 7
4 ⋮ 5 4 ⋮ ⋮ ⋱ ⋱ ⋱ ⋱ ⋮ 56 x
1 7 4 ⋮ 5,
6 7
y0 ⋯ h0 ⋯ hm
1 hm 6 7 z0
|fflfflffl{zfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}4 ⋮ 5 |fflfflffl{zfflfflffl}
0 0

y H z
x
m

where the last m samples from signal vector x ¼ [xN-1 . . . x0]T correspond to the cyclic prefix: x
1 ¼ xN-1, x
2 ¼ xN-2, . . ., x-
μ ¼ xN-m. In the equation above, H denotes the channel matrix, y is the received vector, and z is the noise vector. Often, the
received symbols y
1,. . .,y-μ get disregarded from consideration on the receiver side because they are not really needed to
recover the input signal and at the same time they are affected by the intersymbol interference (ISI). By re-arranging the
above matrix equation, we obtain the following equivalent matrix representation:
2 3
h0 h1 ⋯ hm 0 ⋯ 0
2 60 3 2
0 7 xN
1
3 2 3
yN
1 6 h0 ⋯ hm
1 hm ⋯
7 zN
1
6y 7 6 ⋮ ⋮ ⋱ ⋱ ⋱ ⋱ ⋮ 76
76 xN
1 7 6 7
6 N
2 7 6 6 7 6 zN
2 7
6 7 ¼ hμ 76 þ6
4 ⋮ 5 6
0 ⋯ 0 h0 ⋯ hm
1
74 ⋮ 7 5 4 ⋮ 5
7
6⋮ ⋮ ⋱ ⋱ ⋱ ⋱ ⋮7
6 7
y0 4 h2 h3 ⋯ hm ⋯ h0 h1 5 x0 z0
h1 h2 ⋯ hm
1 ⋯ 0 h0
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
e
H
7.12 Problems 443

Clearly, the inserted cyclic prefix allows us to model the channel as a circulant convolution matrix over the N samples
of interest. By using the corresponding eigenvalue decomposition of H, e describe how the inter-carrier interference (ICI)
between subcarriers can be compensated for.
4. By using the singular value decomposition (SVM) of the channel matrix H, describe how the concept from parallel
decomposition of wireless MIMO channel, described in Sect. 5.4 of Chap. 5, can be used to compensate for the ICI among
the subcarriers. This approach allows the guard intervals to be empty, thus improving the power efficiency. Describe
drawbacks of this approach compared to the conventional OFDM described in this chapter.
5. The frequency acquisition can be performed based on pilot tones, maximum likelihood estimation, or cyclic prefix.
Describe the details of these schemes. Discuss advantages and disadvantages of each scheme against each other.
6. Let us consider 16-QAM transmission for 64-subcarriers OFDM over two-ray channel model. Assume that 16 subcarriers
are used as pilots for LS channel estimation. In two-ray channel model, reflection coefficient is
1, and attenuation of
reflected ray is 6 dB. The OFDM system design is similar to Problem 1. Show how the reconstructed 16-QAM
constellation will look like for the following delays:
(a) Delay is one-half of guard interval.
(b) Delay exceeds the FFT interval by 2.5%.
(c) Delay exceeds the FFT interval by 10%.
7. Repeat Problem 6 assuming that there is the second reflected component of attenuation 4 dB with delay being 1/2 of that
of the first reflected component. The reflection coefficient for the second reflected ray is 0.5.
8. Let us study OFDM channel estimation for DAB employing mode III, with parameters provided in Table 7.1. Assuming a
two-ray model between the transmitter and receiver wherein the reflected ray is 6 dB lower than the LOS ray, by using
16-QAM compare the LS and MMSE channel estimation efficiencies for different SNRs. Assume that the reflection
coefficient is
1. Use the delay relative to the guard interval as a parameter.
9. The small-scale wireless channel models are valid within small local areas, ranging from 5λ to 40λ. To describe the
channel’s behavior in the local area, the following three average parameters can be used: Rician factor K, RMS deal
spread τRMS, and normalized received power Pn. The channel impulse response (IR) can be modeled as the sum of
P
discrete multipath components arriving at the receiver hðτ, t Þ ¼ i αi ðt Þe
jθi ðtÞ δðτ
τi ðt ÞÞ, where αi is the i-th propaga-
tion path amplitude and τi is the corresponding propagation path delay. θi denotes the i-th propagation path phase. The IR
is time-varying. Power delay profile (PDP) represents the distribution of power among the multipath components
P P
PDPðτÞ ¼ α2i δðτ
τi Þ. The normalized received power is simply Pn ¼ α2i . The Rician factor is the ratio of the
i i
α2max
dominant’s power to the power in the scattered multipath components K ¼ Pn
α2max , αmax ¼ max fαi g. Channel transfer
i
R1
j2πf τ
P
function (TF) is just the Fourier transform (FT) of the IR H ð f , t Þ ¼ hðτ, t Þe dτ ¼ αi ðt Þe
j½2πf τi ðtÞþθi ðtÞ. Under

1 i
the Gaussian wide-sense stationary uncorrelated scattering (WSSUS) model, assume that different multipath components
are uncorrelated, and for a sufficient number of rays with similar amplitudes (except for the line-of-sight ray), the TF will
have a complex Gaussian distribution. The corresponding magnitude R ¼ |H( f,t)| will have Rician PDF. Establish the
connection between small-scale model parameters and Rician distribution.
10. The channel correlation function for small-scale model described in Problem 9, under WSS assumption, is defined as
RH(Δf, Δt) ¼ hH( f, t)H( f + Δf, t + Δt)i. Delay power spectrum (DPS), denoted as Rh(τ), is defined as the inverse FT of
the spaced-frequency correlation function RH(Δf ) ¼ RH(Δf,0). For the small-scale model, we can use the approximation
of the DPS given by Eq. (7.48).
8
>
< ρ2 δðτÞ, τ ¼ 0
Rh ðτÞ ¼ Π, 0 < τ  τc
>
:
Πe
γðτ
τc Þ , τ > τc

where ρ2 is the normalized power of the line-of-site (LOS) component, Π is the normalized power density of the
constant-level portion of DPS, τc is the duration of the DPS constant level, and γ is the decay exponent of the exponentially
decreasing portion of the DPS. Establish the connection between this model and the actual channel parameters. By using
this model, develop the corresponding simulator, and evaluate it for indoor use.
444 7 OFDM for Wireless and Optical Communications

11. Apply the model developed in Problem 10 to the OFDM system for use in Wi-Fi according to the IEEE 802.11a PHY
standard, with OFDM system parameters provided in Table 7.4. Provide BER results for different modulation formats and
different aggregate data rates.
12. Let us now consider the time-invariant frequency-selective block-fading channel with 32 subchannels of bandwidth
2 MHz, with frequency response amplitudes being 1 for odd subchannels and √2 for even subchannels, respectively. For
the transmit power of 20 mW and noise power-spectral density of N0 ¼ 10
9 W/Hz, determine the OFDM channel
capacity. What is the corresponding optimum power allocation strategy based on OFDM to achieve this channel capacity?
Provide the corresponding OFDM transmitter and receiver configurations.
13. Provide the system parameters of CO-OFDM system that is capable of compensating the accumulated chromatic
dispersion over 1000 km of SMF and corresponding PMD. The aggregate data rate should be at least 400 Gb/s, and
the OFDM signal bandwidth should be either 25 GHz or 40 GHz. Use polarization-division multiplexing and square-
QAM constellations. Assume that chromatic dispersion parameter is 16 ps/(nmkm) and that PMD parameter is
Dp ¼ 0.05 pskm-1/2. Ignore the second-order GVD and fiber nonlinearities. Estimate the OSNR penalty due to guard
interval compared to back-to-back configuration. Determine the spectral efficiency of your design. Is it possible to find the
optimum number of subcarriers to maximize the spectral efficiency?
14. OFDM samples at the output of OFDM transmitter follow Gaussian distribution for a sufficient number of subcarriers.
The envelope will clearly follow Rayleigh distribution, while the power will have chi-square distribution. Determine the
cumulative distribution function of the envelope. Determine the probability that PAPR is below certain threshold ztsh. Plot
the logarithm of this probability against PAPR [dB] for the following number of subcarriers: 32, 64, 128, 256, and 512.
Finally, discuss possible approaches to reduce the PAPR.
15. This problem is related to direct detection OFDM (DD-OFDM). Let us generalize the OFDM equation by introducing the
RF frequency term exp.(jωRFt) as follows:
8
>
< P
k¼N sc =2
1
j2π Tks t jωRF t
X kþN sc =2 e e ,
T s =2 < t  T s =2
sðt Þ ¼ k¼
N sc =2
>
:
0, otherwise

16. The real part of this signal, after appropriate bias addition, is used as input of an MZM. On the receiver side, only one
photodetector is used to perform optical-to-electrical conversion. Provide the corresponding block diagram of this
DD-OFDM system, and describe the operation principle of each block. Explain why the addition of bias is needed.
This scheme is known as double-side band (DSB) DD-OFDM scheme. Describe how the DSB scheme can be converted
to single-side band (SSB) scheme by using either time-domain or frequency-domain approach. Provide the block
diagrams of corresponding SSB schemes. Discuss the implementation complexity of these two SSB schemes.
17. Eq. (7.33) describes the OFDM signal upon coherent detection in the presence of frequency offset and phase noise. In this
problem we study the influence of frequency offset on BER performance of QPSK-OFDM system. The frequency offset
introduces the inter-carrier interference (ICI). By using a sufficient number of subcarriers, the ICI can be approximated as
Gaussian process by invoking the central limit theorem. The BER of QPSK in the absence of ICI is given by BER ¼
pffiffiffiffiffiffiffiffi
0:5erfc ρ=2 , where ρ is signal-to-noise ratio (SNR) and erfc() is the complementary error function. In the presence
of ICI, we need to increase the SNR to achieve the same BER. Plot the BER against SNR [dB] for different values of
normalized frequency offset δ ¼ Δfoff/(1/Ts), where Δfoff is frequency offset and Ts is symbol duration.
18. Repeat the previous problem but now in terms of phase error. Plot the BER for QPSK-OFDM against SNR [dB] for
different values of normalized laser linewidth Δν/(1/Ts), where Ts is the symbol duration.
19. In this problem we study the efficiency of DFT window synchronization for CO-OFDM systems. Implement the
CO-OFDM system in Matlab or C/C++ to perform the following Monte Carlo simulations. The OFDM system
parameters should be set as follows: the symbol period of 25.6 ns, the guard time 3.2 ns, the number of subcarriers
256, the aggregate data rate of 10 Gb/s with BPSK format being used, and the laser linewidths of transmit and local
oscillator lasers equal to 100 kHz. Plot the timing metric against the timing offset (in samples) for accumulated chromatic
dispersion of 34,000 ps/nm and OSNR of 6 dB. Repeat the simulation when QPSK is used instead. Finally, repeat the
simulation when the first-order PMD is present for different DGD values: 100 ps, 500 ps, and 1000 ps.
20. Here we study the efficiency of Alamouti-type polarization-time (PT) coding in CO-OFDM systems. Alamouti code is
described in Sect. 8.2.4.1 of Chap. 8. Design a CO-OFDM system, based on Alamouti-type PT coding, capable of
References 445

100 Gb/s serial optical transmission over 6500 km of SMF. Perform Monte Carlo simulations in linear regime, by taking
the following effects into account: chromatic dispersion, the first-order PMD, and laser linewidth of 100 kHz (for both
transmitter and local lasers). Use the typical fiber parameters given in previous chapters. Compare this scheme with an
equivalent polarization-division multiplexed (PDM) scheme, based on LS channel estimation, of the same aggregate data
rate. Compare and discuss the results. What are the advantages and disadvantages of PT coding with respect to PDM?

References
1. R.W. Chang, Synthesis of band-limited orthogonal signals for multichannel data transmission. Bell Syst. Tech. J. 45, 1775–1796 (1966)
2. R. W. Chang, Orthogonal frequency division multiplexing, US Patent no. 3488445 (1970)
3. R. Van Nee, R. Prasad, OFDM wireless multimedia communications (Artech House, Boston-London, 2000)
4. R. Prasad, OFDM for wireless communications systems (Artech House, Boston-London, 2004)
5. W. Shieh, I. Djordjevic, OFDM for optical communications (Elsevier, Oct. 2009)
6. I. Djordjevic, W. Ryan, B. Vasic, Coding for optical channels (Springer, 2010)
7. M. Cvijetic, I.B. Djordjevic, Advanced optical communications and networks (Artech House, 2013)
8. Q. Pan, R.J. Green, Bit-error-rate performance of lightwave hybrid AM/OFDM systems with comparision with AM/QAM systems in the
presence of clipping noise. IEEE Photon. Technol. Lett. 8, 278–280 (1996)
9. B.J. Dixon, R.D. Pollard, S. Iezekiel, Orthogonal frequency-division multiplexing in wireless communication systems with multimode fiber
feeds. IEEE Tran. Microw. Theory Techn. 49(8), 1404–1409 (Aug. 2001)
10. A.J. Lowery, J. Armstrong, 10 Gbit/s multimode fiber link using power-efficient orthogonal-frequency-division multiplexing. Opt. Express
13(25), 10003–10009 (Dec. 2005)
11. I.B. Djordjevic, B. Vasic, Orthogonal frequency division multiplexing for high-speed optical transmission. Opt. Express 14, 3767–3775 (May
1, 2006)
12. W. Shieh, C. Athaudage, Coherent optical frequency division multiplexing. Electron. Lett. 42, 587–589 (2006)
13. A. J. Lowery, L. Du, J. Armstrong, Orthogonal frequency division multiplexing for adaptive dispersion compensation in long haul WDM
systems, in Proc. OFC Postdeadline Papers, Paper no. PDP39, (Mar. 2006)
14. I.B. Djordjevic, L. Xu, T. Wang, Beyond 100 Gb/s optical transmission based on polarization multiplexed coded-OFDM with coherent
detection. IEEE/OSA J. Opt. Commun. Netw. 1(1), 50–56 (June 2009)
15. Y. Ma, Q. Yang, Y. Tang, S. Chen, W. Shieh, 1-Tb/s single-channel coherent optical OFDM transmission over 600-km SSMF fiber with
subwavelength bandwidth access. Opt. Express 17, 9421–9427 (2009)
16. X. Liu, S. Chandrasekhar, Beyond 1-Tb/s superchannel transmission, in Proc. IEEE Photonics Conference (PHO) 2011, pp. 893–894, 9–13
Oct. 2011
17. S. Sano et al., No-guard-interval coherent optical OFDM for 100-Gb/s long-haul WDM transmission. J. Lightw. Technol. 27(16), 3705–3713
(2009)
18. B. Zhu, X. Liu, S. Chandrasekhar, D.W. Peckham, R. Lingle, Ultra-long-haul transmission of 1.2-Tb/s multicarrier no-guard-interval
CO-OFDM superchannel using ultra-large-area fiber. IEEE Photon. Technol. Lett. 22(11), 826–828 (2010)
19. S.C.J. Lee, F. Breyer, S. Randel, H.P.A. Van den Boom, A.M.J. Koonen, High-speed transmission over multimode fiber using discrete
multitone modulation [invited]. J. Opt. Netw. 7(2), 183–196 (Feb. 2008)
20. I.B. Djordjevic, M. Arabaci, L. Xu, T. Wang, Spatial-domain-based multidimensional modulation for multi-Tb/s serial optical transmission.
Opt. Express 19(7), 6845–6857 (2011)
21. A. Li, A. Al Amin, X. Chen, W. Shieh, Reception of mode and polarization multiplexed 107-Gb/s CO-OFDM signal over a two-mode fiber, in
Proc. OFC/NFOEC, Postdeadline Papers (OSA, 2011), Paper PDPB8
22. L. Gruner-Nielsen, Y. Sun, J. W. Nicholson, D. Jakobsen, R. Lingle, B. Palsdottir, Few mode transmission fiber with low DGD, low mode
coupling and low loss, in Proc. OFC/NFOEC, Postdeadline Papers (OSA, 2012), Paper PDP5A.1
23. Ryf, R., et al., “Low-loss mode coupler for mode-multiplexed transmission in few-mode fiber,” in Proc. OFC/NFOEC, Postdeadline Papers
(OSA, 2012), Paper PDP5B.5
24. Fontaine, N. K., et al., Space-division multiplexing and all-optical MIMO demultiplexing using a photonic integrated circuit, in Proc. OFC/
NFOEC, Postdeadline Papers (OSA, 2012), Paper PDP5B.1
25. X. Chen, A. Li, J. Ye, A. Al Amin, W. Shieh, Reception of dual-LP11-mode CO-OFDM signals through few-mode compatible optical
add/drop multiplexer, in Proc. OFC/NFOEC, Postdeadline Papers (OSA, 2012), Paper PDP5B.4
26. J.G. Proakis, Digital communications, 5th edn. (McGraw Hill, New York, 2007)
27. S. Haykin, Communication Systems, 4th edn. (John Wiley & Sons, 2001)
28. E. Collett, Polarized light in fiber optics (SPIE Press, 2003)
29. J.G. Proakis, D.G. Manolakis, Digital signal processing: principles, algorithms, and applications, 4th edn. (Prentice-Hall, 2007)
30. A. Goldsmith, Wireless communications (Cambridge University Press, Cambridge, 2005)
31. W. Shieh, X. Yi, Y. Ma, Q. Yang, Coherent optical OFDM: Has its time come? [invited]. J. Opt. Netw. 7, 234–255 (2008)
32. T.M. Schmidl, D.C. Cox, Robust frequency and time synchronization for OFDM. IEEE Trans. Commun. 45, 1613–1621 (1997 Dec)
33. Minn, H., Bhargava, V. K., and Letaief, K. B., “A robust timing and frequency synchronization for OFDM systems,” IEEE Trans Wireless
Comm., vol. 2, pp. 822–839, July 2003
34. T. Pollet, M. Van Bladel, M. Moeneclaey, BER sensitivity of OFDM systems to carrier frequency offset and Wiener phase noise, IEEE Trans
Comm., vol. 43, pp. 191–193, Feb./Mar./Apr. 1995
446 7 OFDM for Wireless and Optical Communications

35. S. Coleri, M. Ergen, A. Puri, A. Bahai, Channel estimation techniques based on pilot arrangement in OFDM systems. IEEE. Trans.
Broadcasting 48(3), 223–229 (2002 Sept)
36. D. Zou, I. B. Djordjevic, Multi-Tb/s optical transmission based on polarization-multiplexed LDPC-coded multi-band OFDM, in Proc. 13th
International Conference on Transparent Optical Networks (ICTON) 2011, Paper Th.B3.3, 26–30 June 2011, Stockholm
37. D. Zou, I. B. Djordjevic, Beyond 1Tb/s superchannel optical transmission based on polarization multiplexed coded-OFDM over 2300 km of
SSMF, in Proc. 2012 Signal Processing in Photonics Communications (SPPCom), Paper SpTu2A.6, 17–21 June, 2012, Colorado Springs
38. H. Sari, G. Karam, I. Jeanclaude, Transmission techniques for digital terrestrial TV broadcasting. IEEE Commun. Mag. 33(2), 100–109 (1995)
39. P. Moose, A technique for orthogonal frequency division multiplexing frequency offset correction. IEEE Trans. Commun. 42, 2908–2914
(1994)
40. M. Sandell, J. J. Van de Beek, P. O. Börjesson, Timing and frequency synchronization in OFDM systems using cyclic prefix, in Proc. Int. Symp.
Synchron., pp. 16–19, Saalbau, Essen, Germany, Dec. 1995
41. D. Zou, C. Lin, I. B. Djordjevic, LDPC-coded mode-multiplexed CO-OFDM over 1000 km of few-mode fiber, in Proc. CLEO 2012, Paper
no. CF3I.3, San Jose, CA, 6–11 May 2012
42. X.Q. Jin, J.M. Tang, P.S. Spencer, K.A. Shore, Optimization of adaptively modulated optical OFDM modems for multimode fiber-based local
area networks [invited]. J. Opt. Netw. 7(3), 198–214 (2008 Mar)
43. D.J.G. Mestdagh, P.M.P. Spruyt, A method to reduce the probability of clipping in DMT-based transceivers. IEEE Trans. Commun. 44,
1234–1238 (1996)
44. P.V. Eetvelt, G. Wade, M. Thompson, Peak to average power reduction for OFDM schemes by selected scrambling. IEE Electron. Lett. 32,
1963–1964 (1996)
45. M. Friese, Multicarrier modulation with low peak-to-average power ratio. IEE Electron. Lett. 32, 712–713 (1996)
46. J.A. Davis, J. Jedwab, Peak-to-mean power control in OFDM, Golay complementary sequences, and Reed-Muller codes. IEEE Trans. Inf.
Theory 45, 2397–2417 (1999)
47. I. B. Djordjevic, L. Xu, T. Wang, PMD Compensation in Multilevel Coded-Modulation Schemes with Coherent Detection using BLAST
Algorithm and Iterative Polarization Cancellation, Optics Express, vol. 16, no. 19, pp. 14845–14852, Sept. 15, 2008
48. Y. Zhao, A. Huang, A novel channel estimation method for OFDM mobile communication systems based on pilot signals and transform-
domain processing, in Proc. IEEE 47th Vehicular Technology Conference, 1997, vol. 3, pp. 2089–2093, Phoenix, AZ, 1997
49. P.K. Frenger, N. Arne, B. Svensson, Decision-directed coherent detection in multicarrier systems on Rayleigh fading channels. IEEE Trans.
Veh. Technol. 48(2), 490–498 (Mar 1999)
50. T. Kella, Decision-directed channel estimation for supporting higher terminal velocities in OFDM based WLANs, in Proc. IEEE Global
Telecommunications Conference 2003 (GLOBECOM '03), vol. 3, pp. 1306–1310, 1–5 Dec. 2003
51. W. Hoeg, T. Lauterbach, Digital audio broadcasting: principles and applications of digital radio (Wiley, Chichester, 2003)
52. W. Fischer, Digital video and audio broadcasting technology: a practical engineering guide, 3rd edn. (Springer, Heidelberg-
Dordrecht-London-New York, 2010)
53. ETSI Standard: EN 300 744 V1.5.1, Digital Video Broadcasting (DVB); Framing structure, channel coding and modulation for digital
terrestrial television
54. M. Ghavami, L.B. Michael, R. Kohno, Ultra wideband signal and systems in communication engineering, 2nd edn. (Wiley, Chichester, 2007)
55. ETSI Standard: EN 302755, digital video broadcasting (DVB), frame structure, channel coding and modulation for a second generation digital
terrestrial video broadcasting system, ETSI, 2009
56. S.M. Alamouti, A simple transmit diversity technique for wireless communications. IEEE J. Sel. Areas Commun. 16(8), 1451–1458 (1998)
57. I.B. Djordjevic, L. Xu, T. Wang, PMD compensation in coded-modulation schemes with coherent detection using Alamouti-type polarization-
time coding. Opt. Express 16(18), 14163–14172 (09/01/2008)
58. IEEE Standard for Information technology—Telecommunications and information exchange between systems Local and metropolitan area
networks—Specific requirements Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, in IEEE
Std 802.11–2012 (Revision of IEEE Std 802.11–2007), pp. 1–2793, March 29 2012, https://doi.org/10.1109/IEEESTD.2012.6178212
59. J. Heiskala, J. Terry, OFDM wireless LANs: a theoretical and practical guide (Sams Publishing, 2002)
60. L. Hanzo, Y. Akhtman, L. Wang, M. Jiang, MIMO-OFDM for LTE, WiFi, and WiMAX (IEEE/Wiley, 2011)
61. M.S. Gast, 802.11 wireless networks: the definitive guide, 2nd edn. (O’Reilly Media, Inc, Sebastopol, 2005)
62. M.S. Gast, 802.11ac: A Survival Guide (O’Reilly Media, Inc, Sebastopol, 2013)
63. M.S. Gast, 802.11an: A Survival Guide (O’Reilly Media, Inc, Sebastopol, 2012)
64. Unlicensed Use of the 6 GHz Band; Expanding Flexible Use in Mid-Band Spectrum Between 3.7 and 24 GHz, available at: https://docs.fcc.
gov/public/attachments/FCC-20-51A1.pdf
65. 5G NR (Rel-15), available at: https://www.3gpp.org/lte-2
66. 3GPP Release 18, available at: https://www.3gpp.org/release18
67. IEEE Standard for Air Interface for Broadband Wireless Access Systems. IEEE Std 802.16–2017 (Revision of IEEE Std 802.16–2012),
pp.1–2726, 2 March 2018
68. IEEE 802.16 Working Group on Broadband Wireless Access Standards, see https://grouper.ieee.org/groups/802/16/
69. L. Nuaymi, WiMAX: technology for broadband wireless access (Wiley, Chichester, 2007)
70. IEEE 802.16–2004, IEEE Standard for Local and Metropolitan Area Networks, Air Interface for Fixed Broadband Wireless Access Systems,
October 2004
71. IEEE 802.16e, IEEE Standard for Local and Metropolitan Area Networks, Air Interface for Fixed Broadband Wireless Access Systems,
Amendment 2: Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands and Corrigendum
1, February 2006
72. M. Ghavami, L.B. Michael, R. Kohno, Ultra wideband signals and systems in communication engineering (Wiley, 2007)
73. A. Batra, J. Balakrishnan, A. Dabak, R. Gharpurey, J. Lin, P. Fontaine, J.-M. Ho, S. Lee, M. Frechette, S. March, H. Yamaguchi, “Multi-band
OFDM physical layer proposal for IEEE 802.15 Task Group 3a,” Doc. IEEE P802.15-03/268r3, Mar. 2004
74. H. Nikookar, R. Prasad, Introduction to Ultra Wideband for Wireless Communications (Springer Science+Business Media B.V, 2009)
References 447

75. W.P. Siriwongpairat, K.J.R. Liu, Ultra-wideband communications systems: multiband OFDM approach (Wiley, Hoboken, 2007)
76. P. K. Boon, X. Peng, F. Chin, Performance studies of a multi-band OFDM system using a simplified LDPC code, in Proc. 2004 International
Workshop on Ultra Wideband Systems & 2004 Conference on Ultrawideband Systems and Technologies (UWBST & IWUWBS),
pp. 376–380, 18–21 May 2004
77. S.-M. Kim, J. Tang, K. K. Parhi, Quasi-cyclic low-density parity-check coded multiband-OFDM UWB systems, in Proc. 2005 IEEE
International Symposium on Circuits and Systems, pp. 65–68, 23–26 May 2005
78. W. Shieh, R.S. Tucker, W. Chen, X. Yi, G. Pendock, Optical performance monitoring in coherent optical OFDM systems. Opt. Express 15,
350–356 (2007)
79. J.M. Tang, K.A. Shore, Maximizing the transmission performance of adaptively modulated optical OFDM signals in multimode-fiber links by
optimizing analog-to-digital converters. J. Lightwave Technol. 25, 787–798 (2007)
80. B.J.C. Schmidt, A.J. Lowery, J. Armstrong, Experimental demonstrations of 20 Gbit/s direct-detection optical OFDM and 12 Gbit/s with a
colorless transmitter. In Opt. Fiber Commun. Conf. 2007, paper no. PDP18, San Diego (2007)
81. M. Schuster, S. Randel, C.A. Bunge, et al., Spectrally efficient compatible single-sideband modulation for OFDM transmission with direct
detection. IEEE Photon. Technol. Lett. 20, 670–6722 (2008)
82. W. Peng, X. Wu, V.R. Arbab, et al., Experimental demonstration of 340 km SSMF transmission using a virtual single sideband OFDM signal
that employs carrier suppressed and iterative detection techniques (Opt. Fiber Commun. Conf. 2008, paper no. OMU1, San Diego, 2008)
83. W.R. Peng, X. Wu, V.R. Arbab, et al., Experimental demonstration of a coherently modulated and directly detected optical OFDM system
using an RF-tone insertion (In: Opt. Fiber Commun. Conf. 2008, paper no. OMU2, San Diego, 2008)
84. W. Shieh, Q. Yang, Y. Ma, 107 Gb/s coherent optical OFDM transmission over 1000-km SSMF fiber using orthogonal band multiplexing. Opt.
Express 16, 6378–6386 (2008)
85. I. B. Djordjevic, W. Shieh, X. Liu, M. Nazarathy, Advanced Digital Signal Processing and Coding for Multi-Tb/s Optical Transport [From the
Guest Editors], IEEE Sig. Proc. Mag., vol. 31, no. 2, pages 15 and 142, Mar. 2014
86. I.B. Djordjevic, M. Cvijetic, C. Lin, Multidimensional signaling and coding enabling multi-Tb/s optical transport and networking. IEEE Sig.
Proc. Mag. 31(2), 104–117 (Mar. 2014)
87. I. B. Djordjevic, “On the irregular nonbinary QC-LDPC-coded hybrid multidimensional OSCD-modulation enabling beyond 100 Tb/s optical
transport,” IEEE/OSA J. Lightwave Technol., vol. 31, no. 16, pp. 2969–2975, Aug.15, 2013
88. I.B. Djordjevic, M. Arabaci, L. Xu, T. Wang, Generalized OFDM (GOFDM) for ultra-high-speed optical transmission. Opt. Express 19(7),
6969–6979 (2011)
89. I. B. Djordjevic, Spatial-domain-based hybrid multidimensional coded-modulation schemes enabling multi-Tb/s optical transport, J. Lightwave
Technol., vol. 30, no. 14, pp. 2315–2328, July 15, 2012
90. X. Liu, S. Chandrasekhar, P.J. Winzer, Digital signal processing techniques enabling multi-Tb/s superchannel transmission: An overview of
recent advances in DSP-enabled superchannels. IEEE Signal Process. Mag. 31(2), 16–24 (March 2014)
91. I. B. Djordjevic, A. H. Saleh, F. Küppers, Design of DPSS based fiber Bragg gratings and their application in all-optical encryption, OCDMA,
optical steganography, and orthogonal-division multiplexing, Opt. Express, vol. 22, no. 9, pp. 10882–10897, 5 May 2014
92. I. B. Djordjevic, A. Jovanovic, Z. H. Peric, T. Wang, “Multidimensional optical transport based on optimized vector-quantization-inspired
signal constellation design,” IEEE Trans. Commun., vol. 62, no. 9, pp. 3262–3273, Sept. 2014
93. I. B. Djordjevic, On advanced FEC and coded modulation for ultra-high-speed optical transmission, IEEE Communications Surveys and
Tutorials, vol. 18, no. 3, pp. 1920–1951, third quarter 2016, August 19, 2016, https://doi.org/10.1109/COMST.2016.2536726
94. I. Djordjevic, H.G. Batshon, L. Xu, T. Wang, Four-dimensional optical multiband-OFDM for beyond 1.4 Tb/s serial optical transmission. Opt.
Express 19(2), 876–882 (01/17/2011)
95. A. Sano, E. Yoshida, H. Masuda, et al., 30  100-Gb/s all-optical OFDM transmission over 1300 km SMF with 10 ROADM nodes (Eur. Conf.
Opt. Commun. 2007 (ECOC 2007), paper no. PD 1.7, Berlin, 2007)
96. E. Yamada, A. Sano, H. Masuda, et al., Novel no-guard interval PDM CO-OFDM transmission in 4.1 Tb/s (50 88.8-Gb/s) DWDM link over
800 km SMF including 50-GHz spaced ROADM nodes (Opt. Fiber Commun. Conf. 2008 (OFC 2008), paper no. PDP8, San Diego, 2008)
97. G Singh, A Alphones. OFDM modulation study for a radio-over-fiber system for wireless LAN (IEEE 802.11a). In: Proc. ICICS-PCM.
Singapore. p. 1460–4 (2003)
98. J.B. Song, A.H.M.R. Islam, Distortion of OFDM signals on radio-over-fiber links integrated with an RF amplifier and active/passive
electroabsorption modulators. J. Lightwave Technol. 26(5), 467–477 (2008)
99. L. Xu, D. Qian, J. Hu, W. Wei, T. Wang, OFDMA-based passive optical networks (PON) (2008 Digest IEEE LEOS Summer Topical Meetings,
2008), pp. 159–160
100. A. Emsia, M. Malekizandi, T. Q. Le, I. B. Djordjevic, F. Kueppers, 1 Tb/s WDM-OFDM-PON power budget extension techniques, in Proc.
IEEE Photonics Conference 2013 (IPC 2013), Paper WG3.3, 8–12 September 2013, Seattle, USA
101. J. Grubor, V. Jungnickel, K.-D. Langer, Capacity analysis in indoor wireless infrared communication using adaptive multiple subcarrier
transmission (Proc ICTON, 2005), pp. 171–174
102. J. Grubor, O.C. Gaete Jamett, J.W. Walewski, S. Randel, K.-D. Langer, High-speed wireless indoor communication via visible light, in ITG
Fachbericht 198, (VDE Verlag, Berlin, 2007), pp. 203–208
103. I.B. Djordjevic, B. Vasic, M.A. Neifeld, LDPC coded OFDM over the atmospheric turbulence channel. Opt. Express 15(10), 6332–6346 (May
2007)
104. I.B. Djordjevic, LDPC-coded OFDM transmission over graded-index plastic optical fiber links. IEEE Photon. Technol. Lett. 19(12), 871–873
(2007)
105. A.R. Shah, R.C.J. Hsu, A.H. Sayed, B. Jalali, Coherent optical MIMO (COMIMO). J. Lightwave Technol. 23(8), 2410–2419 (2005)
106. R.C.J. Hsu, A. Tarighat, A. Shah, A.H. Sayed, B. Jalali, Capacity enhancement in coherent optical MIMO (COMIMO) multimode fiber
links. J. Lightwave Technol. 23(8), 2410–2419 (2005)
448 7 OFDM for Wireless and Optical Communications

107. Z. Tong, Q. Yang, Y. Ma, W. Shieh. 21.4 Gb/s coherent optical OFDM transmission over multimode fiber, post-deadline papers technical
digest. 13th Optoelectronics and Communications Conference (OECC) and 33rd Australian Conference on Optical Fibre Technology
(ACOFT), Paper No. PDP-5. (2008)
108. F. Yaman, S. Zhang, Y.-K. Huang, E. Ip, J. D. Downie, W. A. Wood, A. Zakharian, S. Mishra, J. Hurley, Y. Zhang, I. B. Djordjevic, M.-F.
Huang, E. Mateo, K. Nakamura, T. Inoue, Y. Inada, T. Ogata, First Quasi-Single-Mode Transmission over Transoceanic Distance using
Few-mode Fibers, in Proc. OFC Postdeadline Papers, Paper Th5C.7, 22–26 March 2015, Los Angeles
 Diversity and MIMO Techniques
8

Abstract
The purpose of this chapter is to describe various diversity and MIMO techniques capable of improving single-input single-
output system performance. Through MIMO concept it is also possible to improve the aggregate data rate. The diversity
and MIMO techniques described are applicable to wireless MIMO communications as well as free-space optical and fiber-
optic communications with coherent optical detection. Various diversity schemes are described including polarization
diversity, spatial diversity, and frequency diversity schemes. The following receiver diversity schemes are described:
selection combining, threshold combining, maximum-ratio combining, and equal-gain combining schemes. Various
transmit diversity schemes are described, depending on the availability of channel state information on transmitter side.
Most of the chapter is devoted to wireless and optical MIMO techniques. After description of various wireless and optical
MIMO models, we describe the parallel decomposition of MIMO channels, followed by space-time coding (STC)
principles. The maximum likelihood (ML) decoding for STC is described together with various design criteria. The
following classes of STC are described: Alamouti code, orthogonal designs, linear space-time block codes, and space-time
trellis codes. The corresponding STC decoding algorithms are described as well. After that we move our attention to spatial
division multiplexing (SDM) principles and describe various BLAST encoding architectures as well as multi-group space-
time coded modulation. The following classes of linear and feedback MIMO receiver for uncoded signals are subsequently
described: zero-forcing, linear minimum MSE, and decision-feedback receivers. The next topic in the chapter is related to
suboptimum MIMO receivers for coded signals. Within this topic we first describe linear zero-forcing (ZF) and linear
MMSE receivers (interfaces) but in the context of STC. Within decision-feedback and BLAST receivers, we describe
various horizontal/vertical (H/V) BLAST architecture including ZF and MMSE ones. The topic on suboptimum receivers
is concluded with diagonal (D)-BLAST and iterative receiver interfaces. The section on iterative MIMO receivers starts
with brief introduction of concept of factor graphs, followed with the description of factor graphs for MIMO channel and
channels with memory. We then describe the sum-product algorithm (SPA) operating on factor graphs, followed by
description of SPA for channels with memory. The following iterative MIMO receivers for uncoded signals are described:
ZF, MMSE, ZF H/V-BLAST, and LMMSE H/V-BLAST receivers. After a brief description of factor graphs for linear
block and trellis codes, we describe several iterative MIMO receivers for space-time coded signals. The section on
broadband MIMO describes how frequency-selectivity can be used as an additional degree of freedom, followed by
description of MIMO-OFDM and space-frequency block-coding principles. The focus is then moved to MIMO channel
capacity calculations for various MIMO channel models including deterministic, ergodic, and non-ergodic random
channels, as well as correlated channel models. The concepts of ergodic and outage channel capacity are described. The
section on MIMO channel capacity ends with MIMO-OFDM channel capacity description. In MIMO channel estimation
section, the following MIMO channel estimation techniques are described: ML, least squares (LS), and linear minimum
MSE MIMO channel estimation techniques. After the conclusion section, the set of problems is provided for deeper
understanding of the chapter material.

# The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 449
I. B. Djordjevic, Advanced Optical and Wireless Communications Systems, https://doi.org/10.1007/978-3-030-98491-5_8
450 8 Diversity and MIMO Techniques

8.1 Diversity Techniques

The key idea behind the diversity-combining techniques [1–3] is to use several independent realizations of that same
transmitted signal to improve tolerance to channel impairments. The independent signal copies transmitted over different
paths or degrees of freedom will have a low probability of experiencing the same level of channel distortion. In wireless
communications, the probability that all independent fading paths are in deep fade will be low. The independent realizations of
the same signal will be properly combined to improve overall reliability of transmission.

8.1.1 Basic Diversity Schemes

There are many different ways of transmitting the independent signal paths. In polarization diversity, for optical
communications, the laser signal can be split into two polarizations by the polarization beam splitter (PBS), and the same
data signal can be imposed on two orthogonal polarization states. In wireless communications, two transmit or receive
antennas with horizontal and vertical polarizations can be used to implement the polarization diversity. The coherent optical
receivers require matching the state of polarization (SOP) of the local oscillator laser (LOL) with that of the received optical
signal. In practice, only the SOP of local laser can be controlled, and one possible polarization control receiver configuration is
shown in Fig. 8.1. Polarization controller is commonly implemented by using four squeezers [4, 5]. Insensitivity with respect
to polarization fluctuations is possible if the receiver derives two demodulated signals from two orthogonal polarizations of
the received signal, which is illustrated in Fig. 8.2. This scheme is known as polarization diversity receiver. There two main
issues with respect to this approach: (i) there only two polarization diversity branches and (ii) since the same signal (s) used in
both polarizations, the spectral efficiency of this scheme is twice lower than that of the polarization division multiplexing
(PDM) scheme.
In frequency diversity, the same narrowband signal is transmitted over different subcarriers, while the frequency channel
spacing is larger than the coherence bandwidth. In wavelength diversity, similarly, several copies of the same signal are
transmitted over different wavelength channels. The main drawback of this scheme is poor spectral efficiency.
In time diversity, the same signal is transmitted over different time -slot, while the slot separation is larger than the
coherence time or reciprocal of the channel Doppler spread. This scheme also suffers from the poor spectral efficiency. To
overcome this problem, the multidimensional signaling [6, 7] can be employed, namely, the multiple degrees of freedom, such
as time or frequency slots, can be used as dimensions for multidimensional signaling. Now instead of using this degrees of
freedom to multiplex independent data streams, we can define the signal constellations in multidimensional space, as
discussed in a chapter on advanced modulation schemes. If some of coordinates are affected by deep fade, for instance, the
other will not and the signal constellation point will not be completely destroyed, which will improve tolerance to channel
impairments. In another approach, it is possible to combine the coding with interleaving.
In directional/angle diversity, for wireless applications, the receive antenna beamwidth is restricted to a given angle
increment. If the angle increment is sufficiently small, it is possible to ensure that only one multipath array reaches the
antenna’s beamwidth. On such a way, we can efficiently deal with the multipath fading. On the other hand, the complexity of
such receiver diversity scheme will be high. Thanks to high directivity of laser beams, this method might be suitable for free-

Fig. 8.1 Polarization control Data

receiver configuration for fiber- Polarization
Coherent transmitter
optic communications transformer
Receiver
Local Polarization
laser transformer Controller

Fig. 8.2 Polarization diversity Photodetector

receiver configuration in fiber- Received signal Decision
PBS circuit
optic communications. PBS 3-dB Photodetector
polarization beam splitter splitter
LO Used for balanced detection
8.1 Diversity Techniques 451

space optical communications to deal with atmospheric turbulence effects. However, such system has a sensitive beam
wandering effect. The so-called smart antennas are built from an array of directional antenna elements, wherein the phase of
each element can be independently adjusted. By steering the antenna element to the direction of the strongest multipath
component, we can effectively deal with the multipath fading.
In space/spatial diversity, for wireless applications, multiple transmit or receive antennas are used, and the antennas are
sufficiently separated so that the corresponding fading paths are uncorrelated and independent of each other. In free-space
optical communications, the space diversity can be achieved through multiple laser diodes on the transmitter side and multiple
photodetectors on the receiver side. In fiber-optic communications, the space diversity can be achieved through multiple fiber;
however, the total system capacity will be reduced. In spatial division multiplexing (SDM) systems, in multicore fibers,
different cores can be used to transmit the same data to achieve the spatial diversity. In few-mode fibers (FMFs), the different
spatial modes can be used to enable the spatial diversity. Through space/spatial diversity, the overall signal-to-noise ratio
(SNR) can be improved, and simultaneously the symbol error probability (Ps) can be reduced. Therefore, we can define two
parameters specifying the improvement of various diversity techniques against the single-branch system, namely, array gain
and diversity gain. The array gain is defined as the improvement in the signal-to-noise ratio when diversity is used. On the
other hand, the diversity gain is defined as the change in the slope of symbol error probability curve.

8.1.2 Receiver Diversity

In receiver (Rx) diversity, the independent propagation paths after reception are properly combined, and such obtained signal
is demodulated. We are mostly concerned with the linear combiners, shown in Fig. 8.3, in which multipath propagation paths
are properly weighted according to certain criterion. In wireless applications, the i-th receiver Rxi is represented by the receive
antenna followed by a low-noise amplifier. In free-space optical communications with direct detection, the i-th receiver
corresponds to the photodetector, be it a PIN or APD. In SDM systems, the i-th receiver is essentially a coherent optical
detector. The i-th receiver output can be represented as r i exp ð jϕi Þs, where ϕi is the random phase shift introduced over the i-
th propagation path, ri is the corresponding attenuation coefficient, and s denotes the transmitted signal constellation point.
The linear combiner requires the co-phasing, after the coherent detection, which can be performed by multiplying by wiexp
(jϕi) for some real-valued weights (wi). The output of linear combiner is then r ¼ Σi wi ri s. Assuming that all receivers are
identical with equivalent power spectral density N0, the SNR after the linear combiner, denoted as ρL, can be estimated as:

M 2
P
wi r i
ρL ¼
i¼1
, ð8:1Þ
P
M
N0 w2i
i¼1

where the nominator represents the signal power, while the denominator represents the total noise power in all branches. To
determine the maximum possible SNR at the output of the linear combiner, let us ignore the distortion introduced during the

Fig. 8.3 The generic linear

combiner configuration in Rx1 Rx2 … RxM-1 RxM
receiver diversity

r1ej 1s r2ej 2s rM-1ej M-1s rMej Ms

w1e-j 1
w2e-j 2 … wM-1e-j M-1 wM e-j M

Combiner output
452 8 Diversity and MIMO Techniques

propagation, such as the fading in wireless channels, so that the output of the matched filter in the i-th branch is proportional to
pffiffiffiffiffi
E s , where Es is the average symbol energy and the SNR of linear combiner can be upper bounded by:

M 2
P pffiffiffiffiffi
wi E s
i¼1 Es
ρ  M ¼ Mρ, ð8:2Þ
P
M N0
N0 w2i |{z}
i¼1 ρ

where ρ is the average SNR of any branch. Therefore, the array gain can be now defined as the linear combiner SNR to single-
branch SNR:

ρ
AG ¼  M, ð8:3Þ
ρ

and the maximum possible array gain is M. The average symbol error probability can be determined from SNR-dependent
expression of symbol error probability Ps(ρ) by averaging out over all possible values of ρ:

Z1
Ps ¼ Ps ðρÞ f ρ ðρÞdρ, ð8:4Þ
0

where f ρ ðρÞ is the PDF of linear combiner SNR. The outage probability can be derived from the cumulative distribution
function (CDF):

Zρ0
 
Poutage ¼ Pr ρ  ρ0 ¼ f ρ ðρÞdρ, ð8:5Þ
0

where ρ0 is the target SNR. The linear diversity combining helps getting a more favorable distribution of combine SNR,
f ρ ðρÞ; the average symbol rate and outage probability get reduced. This reduction in the symbol error rate can be described
by more rapid slope in the corresponding symbol error probability curve. We know that for very high SNRs, the symbol error
probability is reversely proportional to ρ, so the diversity gain can be defined as:

log Ps ðρÞ
DG ¼  lim : ð8:6aÞ
ρ!1 log ρ

For some diversity schemes, symbol error probability is proportional to ρ-M, and based on Eq. (8.6a), the diversity gain for
such schemes is DG ¼ M. Clearly, the maximum diversity order is M, and when achieved it is commonly referred to as the full
diversity order.
The most popular receiver diversity schemes include selection combining, threshold combining, maximum-ratio combin-
ing (MRC), and equal gain combining (EGC). In selection combining, the linear combiner selects the branch with the largest
SNR. In threshold combining, the combiner scans the branches and select the first branch with SNR larger than threshold
SNR. In MRC, the weighting coefficients in the linear combiner are properly chosen such that the combiner SNR is
maximized. Finally, in EGC all branches have the same weight. Additional details of various receiver diversity schemes
are provided in subsequent sections.

8.1.2.1 Selection Combining

As indicated earlier, in selection combining (SC), the linear combiner selects the branch with the maximum SNR. If all
receivers are identical, the noise power would be the same, and equivalently the linear combiner in SC can instead select the
branch with the largest total power, which is easier to measure. Given that only one branch is used in this case, there is no need
to use the co-phasing so that the SC is applicable to both differential and coherent detection cases. As discussed in the previous
section, in order to determine either the average symbol error probability or the outage probability, the first step is to determine
8.1 Diversity Techniques 453

the distribution of SNR of the combiner, which can be obtained as the first derivative of CDF with respect to SNR. The CDF is
related to the probability that the combiner SNR ρ is smaller than a given SNR ρ, which is equivalent to the probability that
the maximum of SNRs from all branches is smaller than ρ. When the maximum SNR is smaller than a given ρ, then SNRs
from other branches will be smaller than ρ and we can write:

  M
CDF ρ ðρÞ ¼ Pr ρ < ρ ¼ Prð max ½ρ1 , ρ1 , . . . , ρM  < ρÞ ¼ ∏ Prðρi < ρÞ: ð8:6bÞ
i¼1

The PDF of combiner SNR is then the first derivative of CDF with respect to ρ:

dCDF ρ ðρÞ
f ρ  ð ρÞ ¼ : ð8:7Þ
dρ

Based on Eqs. (8.4) and (8.5), the average symbol error probability and the outage probability are expressed with respect to
CDF as follows:

Z1
dCDF ρ ðρÞ
Ps ¼ Ps ðρÞ dρ, Poutage ¼ CDF ρ ðρ0 Þ, ð8:8Þ
dρ
0

where ρ0 is the target SNR. The PDF of combiner SNR is particularly easy to determine for Rayleigh fading. It has been
shown earlier that the PDF of SNR in the i-th branch is given by:

1 ρi =ρi
f ð ρi Þ ¼ e , ð8:9Þ
ρi

while the corresponding CDF of the i-th branch is:

CDF ρi ðρÞ ¼ 1  eρ=ρi : ð8:10Þ

By using Eq. (8.6b), the CDF of combiner SNR is given by:

M M M  
CDF ρ ðρÞ ¼ ∏ Prðρi < ρÞ ¼ ∏ CDF ρi ðρÞ ¼ ∏ 1  eρ=ρi : ð8:11Þ
i¼1 i¼1 i¼1

Assuming that all branches are identical with the same SNR ρi ¼ ρ, the CDF of SC SNR becomes:

M    M
CDF ρ ðρÞ ¼ ∏ 1  eρ=ρ ¼ 1  eρ=ρ : ð8:12Þ
i¼1

The PDF of SC SNR is then:

dCDF ρ ðρÞ  M  M1

d M
f ρ  ð ρÞ ¼ ¼ 1  eρ=ρ ¼ 1  eρ=ρ eρ=ρ : ð8:13Þ
dρ dρ ρ

The average SC SNR is given by:

Z1 Z1  M1 XM
M 1
ρ ¼ ρ f ρ ðρÞdρ ¼ ρ 1  eρ=ρ eρ=ρ dγ ¼ ρ , ð8:14Þ
ρ i¼1
i
0 0
454 8 Diversity and MIMO Techniques
P 1
and clearly the array gain AG ¼ M i¼1 i increases with the number of branches, but not in a linear fashion. The largest
improvement is when we move from the case of single-branch (no diversity at all) to two-branch case diversity. The outage
M  
probability is obtained from Eq. (8.12) by Poutage ¼ CDF ρ ðρ0 Þ ¼ ∏ 1  eρ0 =ρi , while the average symbol error
i¼1
probability for different modulation formats can be obtained by substituting (8.12) into (8.8). For instance, for BPSK the
average bit error probability, in Rayleigh fading, for SC is given by:

Z1 pffiffiffiffiffi Z1 pffiffiffiffiffi  M1

M
Pb ðBPSKÞ ¼ Q 2ρ f ρ ðρÞdρ ¼ Q 2ρ 1  eρ=ρ eρ=ρ dρ: ð8:15Þ
ρ
0 0

For DPSK, in Rayleigh fading, the average bit error probability can be found in closed form:
 
M1
Z1  M1
M X
M 1
1 ρ M m
Pb ðDPSKÞ ¼ e 1  eρ=ρ eρ=ρ dρ ¼ ð1Þm : ð8:16aÞ
2 ρ
|ffl{zffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} 2 m¼0
1 þ m þρ
0
Pb ðρÞ f ρ ðρÞ

In Fig. 8.4 we show the outage probability vs. normalized SNR, defined as ρ=ρ0 , of the selection combining scheme for
different numbers of diversity branches M in Rayleigh fading. Clearly, the largest improvement is obtained by moving from
single-branch case (no diversity at all) to two-branch-based SC. On the other hand, in Fig. 8.5 we show the average bit error
probability against average SNR ρ for BPSK in Rayleigh fading when SC is used for different numbers of diversity branches.
The results are obtained by numerical evaluation of integral in Eq. (8.15). The largest improvements over single diversity
branch are obtained for M ¼ 2 and M ¼ 4 diversity branches. Going above M ¼ 8 diversity branches does not make sense, as
the further increasing number of branches provides diminishing improvements.

Fig. 8.4 The outage 10

0

probability vs. normalized SNR

(ρ=ρ0 ) of selection combining Selection combining M= 1
scheme for different number of M= 2
diversity branches in Rayleigh -1
fading
10 M= 4
M= 8
Outage probability, Pout

-2
M=16
10 M=20

-3
10

-4
10

-5
10

-6
10
-10 0 10 20 30 40 50
Normalized SNR, av
/ 0
[dB]
8.1 Diversity Techniques 455

Fig. 8.5 The average bit error

probability vs. average SNR (ρ) of SC:
selection combining scheme for M= 1
BPSK for different numbers of
diversity branches in Rayleigh 10
-1 M= 2
fading M= 4

Average bit-error probability, Pb

M= 8
10
-2 M=16
EGC:
M=2
-3
10

-4
10

-5
10

-6
10
0 4 8 12 16 20 24 28
Average SNR, av
[dB]

8.1.2.2 Threshold Combining

In threshold combining (TC), the combiner scans different branches and select the first branch with SNR larger than threshold
SNR, denoted as ρTsh. The TC outputs the signal from this branch as long as the SNR is larger than threshold SNR. Once the
SNR falls below ρTsh, the TC scans sequentially other branches until it finds another branch with SNR larger than ρTsh. The TC
with only two branches is commonly referred to as switch-and-stay combiner (SSC). The CDF for SSC can be determined as
[1, 9]:

CDF ρ1 ðρTsh ÞCDF ρ2 ðρÞ, ρ < ρTsh

CDF ρ ðρÞ ¼ ð8:16bÞ
PrðρTsh  ρ1  ρÞ þ CDF ρ1 ðρTsh ÞCDF ρ2 ðρÞ, ρ  ρTsh

The first case in (8.16b), ρ < ρTsh, corresponds to the case that when the SNR in branch 1, ρ1, dropped below the threshold, we
switched to branch 2; however, the SNR on branch 2, ρ2, is also below the threshold. The second case in (8.16b), ρ  ρTsh,
corresponds to the situation when either SNR on the first branch ρ1 is larger than threshold but smaller than ρ or ρ1 dropped
below the threshold and we moved to branch 2, while the SNR on branch two was above the threshold.
As an illustration, the CDF for Rayleigh fading, providing that ρi ¼ ρ, becomes:
(
1  eρTsh =ρ  eρ=ρ þ eðρTsh þρÞ=ρ , ρ < ρTsh
CDF ρ ðρÞ ¼ ð8:17Þ
1  2eρ=ρ þ eðρTsh þρÞ=ρ , ρ  ρTsh

The PDF of SSC in Rayleigh fading can be found as the first derivative with respect to ρ:
( 
dCDF ρ ðρÞ 1  eρTsh =ρ ð1=ρÞeρ=ρ , ρ < ρTsh
f ρ  ð ρÞ ¼ ¼   ð8:18Þ
dρ 2  eρTsh =ρ ð1=ρÞeρ=ρ , ρ  ρTsh

The outage probability can be obtained from Eq. (8.17), by setting ρ ¼ ρ0, namely, Poutage ¼ CDF ρ ðρ0 Þ. The average symbol
error probability can be expressed as follows:
456 8 Diversity and MIMO Techniques

Z1
Ps ¼ Ps ðρÞ f ρ ðρÞdρ, ð8:19Þ
0

and for Rayleigh fading, we can write:

ZρTsh Z1
Ps ¼ Ps ðρÞ f ρ ðρÞdρ þ Ps ðρÞ f ρ ðρÞdρ ¼
0 ρTsh
ð8:20Þ
ZρTsh   Z1  
ρTsh =ρ ρ=ρ
¼ Ps ðρÞ 1  e ð1=ρÞe dρ þ Ps ðρÞ 2  eρTsh =ρ ð1=ρÞeρ=ρ dρ:
0 ρTsh

For DPSK, the average bit error probability can be determined in closed form:

Z1  
1 ρ 1
Pb ðDPSKÞ ¼ e f ρ ðρÞdρ ¼ 1  eρTsh =ρ þ eρTsh eρTsh =ρ : ð8:21Þ
2 2ð 1 þ ρÞ
0

When SNR is much larger than threshold SNR, Pb ðDPSKÞ ! 2ð1þρ

1
Þ, representing the single-branch result. Therefore, SSC is
inferior compared to SC.

8.1.2.3 Maximum-Ratio Combining

In maximum-ratio combining (MRC), the outputs of different diversity branches, shown in Fig. 8.3, are properly weighted so
that the combiner SNR is maximized. Assuming that all branches are composed of identical devices so that the noise PSD N0
is the same, the combiner output is given by r ¼ Σi wi r i s, where s is the transmitted 2D symbol of energy normalized to one.
Further assuming that the channel bandwidth  symbol duration product is one, the total noise power on all branches will be
PM 2
i¼1 r i N 0 , and the combiner SNR will be upper bounded by:

M 2
P P
M P
M
wi r i w2i r 2i X
M
i¼1 i¼1 i¼1
ρ ¼  ¼ ρi : ð8:22Þ
P
M P
M
i¼1
N0 w2i N0 w2i
i¼1 i¼1

The equality sign is achieved when the squared weight of the i-th branch is proportional to the i-th branch SNR, namely,
w2i ’ r 2i =N 0 . Therefore, in MRC the combiner SNR is equal to the sum of SNRs of individual branches.
The PDF of ρ can be obtained as “inverse” Laplace transform of moment-generating functions (MGFs) of SNRs for
individual branches. The PDF in Rayleigh fading is χ 2 with 2 M degrees of freedom, expected value Mρ, and variance 2Mρ:

ρM1 eρ=ρ
f ρ ðρÞ ¼ , ρ0 ð8:23Þ
ρM ðM  1Þ!

The outage probability in this case can be determined in closed form:

Zρ0 XM
  ðρ0 =ρÞm1
Poutage ¼ Pr ρ  ρ0 ¼ f ρ ðρÞdρ ¼ 1  eρ0 =ρ : ð8:24Þ
m¼1
ðm  1Þ!
0

The average symbol error probability for Rayleigh fading can be determined as:
8.1 Diversity Techniques 457

Z1
ρM1 eρ=ρ
Ps ¼ P s ð ρÞ dρ, ð8:25Þ
ρM ðM  1Þ!
0 |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
f ρ ðρÞ

and for BPSK the average bit error probability can be found in closed form:

Zρ0 pffiffiffiffiffi M1 ρ=ρ

ρ e
Pb ðBPSKÞ ¼ Q 2ρ M dρ
ρ ðM  1Þ!
0 |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
f ρ ðρÞ ð8:26Þ
 M X ! m
1  ρn
M M1þm 1 þ ρn pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ , ρn ¼ ρ=ð1 þ ρÞ
2 k¼1 m 2

The upper bound on symbol error probability for M-ary QAM, obtained by employing the Chernoff bound [Q(x)  exp
(x2/2)], is given by:
pffiffiffiffiffiffiffiffiffiffiffi  
Ps ¼ αM Q βM ρ  αM exp βM ρ =2 ¼ αM exp ðβM ½ρ1 þ . . . þ ρM =2Þ ð8:27Þ

so that the average symbol error probability can be upper bounded by:

Z1 M 1
Ps ¼ Ps ðρÞ f ρ ðρÞdρ  αM ∏ : ð8:28Þ
i¼1 1 þ βM ρi =2
0

When all branches in MRC scheme are identical, for high SNR, the upper bound is simply:

Ps  αM ðβM ρ=2ÞM , ð8:29Þ

indicating that the diversity gain is given by:

log Ps
DG ¼  lim ¼ M, ð8:30Þ
ρ!1 log ρ

and the MRC schemes achieves the full diversity.

For other fading distributions, the MGF approach is more suitable. The MGF for a nonnegative random variable ρ with
PDF fρ(ρ) is defined as:

Z1
MGF ρ ðsÞ ¼ esρ f ρ ðρÞdρ: ð8:31Þ
0

Clearly, the MGF is related to the Laplace transform (LT) of fρ(ρ) by LT{fρ(ρ)} ¼ MGFρ(s). As the name suggests, the
moments of ρn, denoted as E[ρn], can be determined as:

∂
n 
E½ρ  ¼ n MGF ρ ðsÞ :
n
ð8:32Þ
∂s s¼0

The MGF for Nakagami-m fading is given by:

458 8 Diversity and MIMO Techniques

1
MGF ρ ðsÞ ¼ , ð8:33Þ
ð1  sρ=mÞm

while for the Rician fading is given as:


1þK Ksρ
MGF ρ ðsÞ ¼ exp , ð8:34Þ
1 þ K  sρ 1 þ K  sρ

with K being the Rician parameter K, defined by K ¼ s2/(2σ 2) (s2 is the power in line of sight and 2σ 2 is the power in multipath
components). Based on Eq. (8.22), we concluded that in MRC, the combiner SNR is equal to the sum of SNRs of individual
branches, ρ ¼ ∑iρi. Assuming that SNRs at different branches are independent of each other, the joint PDF can be
represented as the product of PDFs:

f ρ1 ,...,ρM ðρ1 , . . . , ρM Þ ¼ f ρ1 ðρ1 Þ f ρ2 ðρ2 Þ  . . .  f ρM ðρM Þ: ð8:35Þ

When the symbol error probability is of the form:

Ps ðρÞ ¼ α exp ðβρÞ, ð8:36Þ

where α and β are constants, such as for DPSK and M-ary QAM for high SNRs (see Eq. (8.27)); the average symbol error
probability can be related to MGF as follows:

Z1 Z1 Z1
Ps ¼ ⋯ α exp ½βðρ1 þ . . . þ ρM Þ f ρ1 ðρ1 Þ . . . f ρM ðρM Þdρ1 . . . dρM
0 0 0
Z1 Z1 Z1 M M
Z1
¼α ⋯ ∏ exp ½βρi  f ρi ðρi Þdρi ¼ α ∏ exp ½βρi  f ρi ðρi Þdρi ð8:37Þ
i¼1 i¼1
0 0 0 0
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
MGF ρi ðβÞ
M
¼ α ∏ MGF ρi ½β:
i¼1

As an illustration, the average bit error probability for DPSK will be obtained from (8.32) by setting α ¼ 1/2 and β ¼ 1 to
obtain Pb ¼ 12 ∏Mi¼1 MGF ρi ½1.
Another relevant case is when the symbol error probability is of the form [1]:

Zd
P s ð ρÞ ¼ α exp ðβðuÞρÞdu, ð8:38Þ
c

which is applicable to M-ary PSK and M-ary QAM, namely, an alternative expression for M-ary PSK is due to Craig [8]:

Z
ππ=M  
1 sin 2 ðπ=M Þ
P s ð ρÞ ¼ exp  ρ dϕ, ð8:39Þ
π sin 2 ϕ
0

and can be related to (8.38) by α ¼ 1/π, β(ϕ) ¼ sin2(π/M)/sin2ϕ, c ¼ 0, and d ¼ π  π/M. Assuming again that SNRs at
different branches are independent of each other, the average symbol error probability is as follows:
8.1 Diversity Techniques 459

Z0 Z0 Z0 Z c Z0 Z0 Z0 Z c M
s ¼
P ⋯ αexpðβðxÞρÞdxf ρ ðρÞdρ ¼ ⋯ ∏ αexpðβðuÞρi Þ f ρi ðρi Þdudρ
i¼1
1 1 1 d 1 1 1 d
ð8:40Þ
Zc M
Z1 Zd M
¼ ∏ αexpðβðuÞρi Þ f ρi ðρi Þdudρ ¼ α ∏ MGF ρi ðβðuÞÞdu
i¼1 i¼1
d 0 c

The average symbol error probability for M-ary PSK will be then:

Z
ππ=M  
1 M sin 2 ðπ=M Þ
Ps ðMPSKÞ ¼ ∏ MGF ρi  dϕ: ð8:41Þ
π i¼1 sin 2 ϕ
0

This MGF-based method is suitable only for MRC schemes, since the combiner SNR in this case is equal to the sum of
SNRs of individual branches, which is not the case for other diversity combining schemes.
In Fig. 8.6 we show the outage probability vs. normalized SNR, also defined as ρ=ρ0 , of the maximum-ratio combining
scheme for different numbers of diversity branches M in Rayleigh fading. By comparing Figs. 8.6 and 8.4, we conclude that
MRC significantly outperforms the SC scheme.
On the other hand, in Fig. 8.7 we show the average bit error probability against average SNR ρ for BPSK in Rayleigh
fading when MRC is used for different numbers of diversity branches. Similarly as in SC, the largest improvements over
single diversity branch are obtained for M ¼ 2 and M ¼ 4 diversity branches. Going above M ¼ 8 diversity branches does not
make sense, as the further increasing in number of branches provides diminishing improvements. By comparing Figs. 8.7 and
8.5, we conclude that MRC significantly outperforms SC. As an illustration, for M ¼ 4 and BPSK, the improvement of MRC
over SC is about 4.8 dB at average bit error probability of 106.

Fig. 8.6 The outage 10

0

probability vs. normalized SNR

(ρ=ρ0 ) of maximum-ratio MRC:
combining scheme for different M= 1
numbers of diversity branches in -1
10
Rayleigh fading as well as for M= 2
EGC with two diversity branches M= 4
Outage probability, Pout

-2
M= 8
10
M=16
EGC:
-3
10
M=2

-4
10

-5
10

-6
10
-10 0 10 20 30 40 50
Normalized SNR, av
/ 0
[dB]
460 8 Diversity and MIMO Techniques

Fig. 8.7 The average bit error

probability vs. averaged SNR (ρ) MRC:
of maximum-ratio combining M= 1
scheme for different number of
diversity branches in Rayleigh
-1
10 M= 2
fading as well as for EGC with M= 4

Average bit-error probability, Pb

two diversity branches
M= 8
-2
10 M=16
EGC:
M=2
-3
10

-4
10

-5
10

-6
10
-4 0 4 8 12 16 20 24 28
Average SNR, av
[dB]

8.1.2.4 Equal-Gain Combining

pffiffiffiffiffidiversity branches ri (i ¼ 1,2,. . .,
In equal-gain combining (EGC), once the co-phasing is performed, the signals from different
PM
M) are combined with equal weights, so the EGC output can be written as r ¼ i¼1 r i = M . Assuming that all branches are
composed of identical components, the combiner output SNR would be then:
!2
X
M
ρ ¼ ri =ðN 0 M Þ: ð8:42Þ
i¼1

Unfortunately, the PDF of combiner SNR cannot be found in closed form, except for Rayleigh fading channel with two
diversity branches [1, 9]. The CDF of combiner SNR for Rayleigh fading channel with two diversity branches can be
determined as [9]:
pffiffiffiffiffiffiffiffiffiffi ρ=ρ  pffiffiffiffiffiffiffiffiffiffi
CDF ρ ðρÞ ¼ 1  e2ρ=ρ  πρ=ρ e 1  2Q 2ρ=ρ : ð8:43Þ

The corresponding outage probability will be then Poutage ðρ0 Þ ¼ CDF ρ ðρ0 Þ. The PDF for this case can be found as the first
derivative of CDF with respect to combiner SNR:
 rffiffiffi rffiffiffiffiffiffi
dCDF ρ ðρÞ 1 2ρ=ρ pffiffiffi ρ=ρ 1 1 ρ ρ
f ρ ðρÞ ¼ ¼ e  πe pffiffiffiffiffiffiffiffi  1  2Q 2 : ð8:44Þ
dρ ρ 4ρρ ρ ρ ρ

The average symbol error probability can be now determined by:

Z1   rffiffiffi rffiffiffiffiffiffi
1 pffiffiffi 1 1 ρ ρ
Ps ¼ Ps ðρÞ e2ρ=ρ  π eρ=ρ pffiffiffiffiffiffiffiffi  1  2Q 2 dρ: ð8:45Þ
ρ 4ρρ ρ ρ ρ
0 |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
f ρ ðρÞ

For BPSK, the average error probability expression can be derived in closed form:
8.1 Diversity Techniques 461

2 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi3
Z1 pffiffiffiffiffi  2
1 1
Pb ðBPSKÞ ¼ Q 2ρ f ρ ðρÞdρ ¼ 41  1  5 ð8:46Þ
2 1þρ
0

Based on Fig. 8.5, we conclude that the power penalty of EGC with respect to MGC, for medium SNRs, when the outage
probability is concerned for M ¼ 2 case, is about 0.6 dB at Pout of 106. Interestingly enough, based on Fig. 8.6, the power
penalty of EGC for M ¼ 2 and BPSK, in Rayleigh fading, is also about 0.6 dB. On the other hand, as in Fig. 8.5, we conclude
that EGC scheme for M ¼ 2 (with BPSK) outperforms the corresponding SC scheme for almost 0.9 dB at average bit error
probability of 106.

8.1.3 Transmitter Diversity

In transmitter diversity multiple transmitters are used, wherein the transmit power is divided among these transmitters. In
optical communications, these transmitters are either laser diodes or LEDs, while in wireless communications, these are
transmit antennas. Two relevant cases of transmitter diversity include (i) the case when the channel state information (CSI) is
available on transmitter side and (ii) when the CSI is not available on transmitter side, the Alamouti-type Scheme [10] is used
in this case.

8.1.3.1 Transmitter Diversity when CSI Is Available on Transmitter Side

Let us consider a transmit diversity system with M transmitters and one receiver. For optical communications with coherent
optical detection and wireless communications, the CSI associated with the i-th transmitter (laser diode, antenna), given by
riexp( jθi), is known at the transmitter side. The transmitted signal constellation point s is pre-multiplied by a complex channel
coefficient wiexp(jθi) (0  wi  1), before being sent with the help of the i-th transmitter (laser diode, antenna). The
P
corresponding weight coefficients (in electrical domain) satisfy the condition i w2i ¼ 1. Clearly, the co-phasing is performed
on a transmitter side. The weights wi maximizing the SNR on a receiver side are given by [1, 9]:

!1=2
X
M
wi ¼ r i = r 2i : ð8:47Þ
i¼1

The receiver side SNR can be expressed now as:

Es X 2 X
M M
ρ ¼ ri ¼ ρi : ð8:48Þ
N 0 i¼1 i¼1

Clearly, the reviver side SNR is equal to the sum of SNRs originating from different transmit branches, the result identical to
that in MRC, indicating that the corresponding combiner PDF for Rayleigh fading is given by Eq. (8.23). When the samples
originating from different transmitters are identical, ri ¼ r, the overall SNR is ρ ¼ Mr2Es/N0, and the corresponding array
gain is AG ¼ M.
The upper bound on symbol error probability for M-ary QAM, obtained by employing the Chernoff bound, is given by:
!
  pffiffiffiffiffiffiffiffiffiffiffi   X
M
Ps ρ ¼ αM Q βM ρ  αM exp βM ρ =2 ¼ αM exp βM ρi =2 : ð8:49Þ
i¼1

The average error probability can be determined by averaging over different SNR-values:

Z1 M 1
Ps ¼ Ps ðρÞ f ρ ðρÞdρ  αM ∏ : ð8:50Þ
i¼1 1 þ βM ρi =2
0
462 8 Diversity and MIMO Techniques

When the average SNR values originating from different transmitter branches are identical, ρi ¼ ρ , the average error
probability is upper bounded by:

Ps  αM ðβM ρ=2ÞM , ð8:51Þ

and the corresponding diversity gain is given by DG ¼ M, indicating that the full diversity is possible, the same as for MRC
scheme.

8.1.3.2 Transmitter Diversity when CSI Is Not Available: Alamouti Scheme

When the CSI is not available on the transmitter side, we can employ the Alamouti scheme [10], shown in Fig. 8.7, illustrated
for wireless communication applications. The code matrix of Alamouti code is given by:

x1 x 2
X¼ ð8:52Þ
x2 x 1

where the columns correspond to 2D signal constellation points transmitted on antennas Tx1 and Tx2, while the rows
correspond to two channel uses. Therefore, the symbols transmitted on antennas Tx1 and Tx2 in the first channel use are x1
and x 2 , respectively. On the other hand, the symbols transmitted on antennas Tx1 and Tx2 in the second channel use are x2
and x 1 , respectively
The signals received in two consecutive time slots, denoted as y1 and y2 , are given by:

y1 ¼ h1 x1 þ h2 x2 þ z1 , y2 ¼ h1 x 2 þ h2 x 1 þ z2 , ð8:53Þ

where h1 is the channel coefficient between receiver antenna and transmitter antenna Tx1, while h2 is the channel coefficient
between receiver antenna and transmitter antenna Tx2. We use z1 and z2 to denote the zero-mean circularly symmetric
Gaussian noise samples at two consecutive channel uses.
The combiner, as shown in Fig. 8.8, operates as follows:

ex1 ¼ h 1 y1 þ h2 y 2 , ex2 ¼ h 2 y1  h1 y 2 , ð8:54Þ

where ex1 and ex2 are estimations of transmitted symbols x1 and x2. After the substitutions of Eq. (8.53) into (8.54), we obtain:

~x1 ¼ h 1 ðh1 x1 þ h2 x2 þ z1 Þ þ h2 ðh 1 x2 þ h 2 x1 þ z 2 Þ

¼ ðjh1 j2 þ jh2 j2 Þx1 þ ðh 1 z1 þ h2 z 2 Þ ð8:55Þ
2 2
~x2 ¼ ðjh1 j þ jh2 j Þx2 þ ðh 2 z1  h1 z 2 Þ:

Clearly, after the combiner, the spatial interference is compensated for, while the noise power is doubled. The receiver SNR is
given by:

Fig. 8.8 The Alamouti scheme Tx1

for two transmit and one receive
antennas h1
Rx x1
x1,x2
Transmitter h2 Combiner
x2

Tx2
8.1 Diversity Techniques 463

Source
channels sx , –s*y

1 LDPC encoder
r1=k1/n
. . m MZM to fiber
. … . Interleaver Mapper PT-encoder DFB PBS PBC
. . mxn
MZM
m LDPC encoder
rm=km/n

sy , s*x

(a)

APP Demapper
From Coherent 1
PBS LDPC Decoder 1

PT-decoder

Calculation
SMF detector .

Bit LLRs
4
.
.
From Coherent
local laser
PBS
detector LDPC Decoder m m

(b)
Fig. 8.9 The architecture of PT-coding scheme concatenated with LDPC coding for multilevel modulations: (a) transmitter architecture and (b)
receiver architecture

 
jh1 j2 þ jh2 j2 E
s
ρ¼ , ð8:56Þ
2 N0

and represents 1/2 of the sum of SNRs from both transmitted branches. Clearly, the array gain is 1. Since the spatial
interference is compensated for (see Eq. (8.55)), the Alamouti scheme achieves the full diversity (DG ¼ 2).
In the rest of this section, we describe how this scheme can be used to compensate for polarization mode dispersion (PMD)
and polarization-dependent loss (PDL) in fiber-optic communications [11] (see also [5]). The polarization time (PT) encoder
operates as follows [11]. In the first half of i-th time instance (“the first channel use”), it sends the symbol sx to be transmitted
using x-polarization channel and the symbol sy to be transmitted using y-polarization channel, as illustrated in Fig. 8.9a. In the
second half of i-th time instance (“the second channel use”), it sends the symbol s y to be transmitted using x-polarization
channel and the symbol s x to be transmitted using y-polarization. Therefore, the PT-coding procedure is similar to the
Alamouti scheme [10]. Notice that Alamouti-type PT-coding scheme has the spectral efficiency comparable to coherent
OFDM with polarization diversity scheme. When the channel is used twice during the same symbol period, the spectral
efficiency of this scheme is twice higher than that of polarization diversity OFDM. Notice that the hardware complexity of PT
encoder/decoder is trivial compared to that of OFDM. The transmitter complexity is slightly higher than that that of
polarization diversity receiver, as it requires additional PT encoder, a polarization beam splitter (PBS), and a polarization
beam combiner (PBC). On the receiver side, we have the option to use only one polarization or to use both polarizations. The
receiver architecture employing both polarizations is shown in Fig. 8.9b. The received symbol vectors in the i-th time instance
for the first (ri,1) and second channel use (ri,2) can be written, respectively, as follows:

jðϕT ϕLO ÞþjϕCD ðk Þ jðϕT ϕLO ÞþjϕCD ðkÞ

ri,1 ¼ Hsi,1 e þ ni,1 , ri,2 ¼ Hsi,2 e þ ni,2 , ð8:57Þ

where H is the Jones (channel) matrix, ri,1(2) ¼ [rx,i,1(2) ry,i,1(2)]T denotes the received symbol vector in the first (second)
channel use of i-th time instance, while ni,1(2) ¼ [nx,i,1(2) ny,i,1(2)]T is the amplified spontaneous emission (ASE) noise vector
corresponding the first (second) channel use in the i-th time instance. We use si,1 ¼ [sx,i sy,i]T to denote the symbol transmitted
h iT
in the first channel use of i-th time instance and si,2 ¼ s y,i s x,i to denote the symbol transmitted in the second channel use
464 8 Diversity and MIMO Techniques

(of the same time instance). Because the symbol vectors transmitted in the first and the second channel use of i-th time instance
are orthogonal, sþ
i,1 si,2 ¼ 0, Eq. (8.57) can be rewritten by grouping separately x- and y-polarizations as follows:

" # " #" # " #

r x,i,1 hxx hxy sx,i nx,i,1
jðϕT ϕLO ÞþjϕCD ðkÞ
¼ e þ ,
r x,i,2 h xy h xx sy,i n x,i,2
" # " #" # " # ð8:58Þ
r y,i,1 hyx hyy sx,i ny,i,1
jðϕT ϕLO ÞþjϕCD ðk Þ
¼ e þ :
r y,i,2 h yy h yx sy,i n y,i,2

If only one polarization is to be used, we can solve either of the equations in Eq. (8.58). However, the use of only one
polarization results in 3 dB penalty with respect to the case when both polarizations are used. Following the derivation similar
to that performed by Alamouti [10], it can be shown that the optimum estimates of transmitted symbols at the output of PT
decoder, for ASE noise-dominated scenario, can be obtained as follows:

esx,i ¼ h xx r x,1 þ hxy r x,2 þ h yx r y,1 þ hyy r y,2 ,

ð8:59Þ
esy,i ¼ h xy r x,1  hxx r x,2 þ h yy r y,1  hyx r y,2 ,

where esx,i and esy,i denote the PT-decoder estimates of symbols sx,i and sy,i transmitted in i-th time instance. In the case that only
one polarization is to be used, say x-polarization, the last two terms in Eq. (8.59) are to be omitted. For additional details, an
interested reader is referred to [11].

8.2 MIMO Optical and Wireless Techniques

Multiple transmitters can be used either to improve the performance through array and diversity gains or to increase data rate
through multiplexing of independent data streams. In multiple-input multiple-output (MIMO) systems, the transmitters and
receivers can exploit the knowledge about the channel to determine the rank of the so-called channel matrix H, denoted as RH,
which is related to the number of independent data streams that can be simultaneously transmitted. In MIMO systems, we can
define the multiplexing gain as the asymptotic slope in SNR of the outage capacity:

Coutage,p ðρÞ
MG ¼  lim , ð8:60Þ
ρ!1 log 2 ρ

where Coutage, p (ρ) is the p percentage outage capacity at signal-to-noise ratio ρ. The p percentage capacity is defined as the
transmission rate that can be supported in (100-p)% of the channel realizations.
As an illustration, in Fig. 8.10 we provide the basic concept of wireless MIMO communication, with the number of
transmit antennas denoted as MTx and the number of receive antennas denoted as MRx. The 2D symbol transmitted on the m-th
transmitter antennas is denoted by xm (m ¼ 1, 2,. . ., MTx), while the received symbol on the n-th receiver antenna is denoted as
yn. The channel coefficient relating the received symbol on n-th receiver antenna and transmitted symbol on the m-th
transmitter antenna is denoted as hmn. By using this model, the output of the n-th receiver antenna can be represented as:

X
M Tx
yn ¼ hnm xm þ zn , ð8:61Þ
m¼1

where zn is the zero-mean Gaussian noise sample with variance σ2 ¼ N0B, with N0/2 being the double-sided power-spectral
density and B being the channel bandwidth.
8.2 MIMO Optical and Wireless Techniques 465

h11
x1 h12 y1
h1M Tx
h21
h22
x2 y2
h2M Tx

…
…
hM Rx 2 hM Rx 1

…
xM Tx hM Rx M Tx yM Rx

Fig. 8.10 The wireless MIMO concept

8.2.1 MIMO Wireless and Optical Channel Models

A narrowband point-to-point wireless communication system composed of MTx transmitter and MRx receiver antennas can be
represented by the following discrete-time model, which is a matrix representation of Eq. (8.61):

y ¼ Hx þ z, ð8:62Þ

where x, y, and z represent the transmitted, received, and noise vectors defined as:
2 3 2 3 2 3
x1 y1 z1
6x 7 6y 7 6z 7
6 2 7 6 2 7 6 2 7
x¼6 7, y¼6 7, z ¼ 6 7, ð8:63Þ
4... 5 4... 5 4... 5
xM Tx yM Rx zM Rx

As indicated above channel noise samples (zn) are complex Gaussian with zero-mean and covariance matrix σ2IMr, where
σ2 ¼ N0B. If the noise variance is normalized to one that the transmitted power P can be normalized as follows P/σ2 ¼ ρ,
where ρ denotes the average SNR per receiver antenna, assuming unity channel gain, the total normalized transmitted power
P Tx  
would be then M m¼1 x i x i ¼ ρ: The correlation matrix of transmitted vector x is given by Rx ¼ hxx{i. The trace of correlation
matrix is clearly Tr(Rx) ¼ ρ. We use H ¼ ½hmn M Tx M Rx in Eq. (8.63) to denote the so-called channel matrix:

2 3
h11 h12 . . . h1M Tx
6h h22 . . . h2M Tx 7
6 21 7
H¼6 7: ð8:64Þ
4... ... ... ... 5
hM Rx 1 hM Rx 2 . . . hM Rx M Tx

The concept of MIMO optical communications has already been introduced in a section on PDM, namely, the polarization-
division multiplexing in which two orthogonal polarizations are used to transmit two independent QAM streams. Therefore,
the PDM represents an example of 2  2 MIMO communication system. On the other hand, in multimode fibers the light
propagates in spatial modes, each with different propagation constant and group velocity. The spatial mode excitement at the
fiber entrance depends on the launch conditions, while the conditions during propagation depend on the initial conditions and
the mode coupling, as analyzed in Sect. 3.6.1. Since a single pulse of light excites a number of spatial modes, it arrives at the
output of the multimode fiber as a sum of multiple pulses due to the modal dispersion, an effect that resembles the multipath
466 8 Diversity and MIMO Techniques

effect in wireless communications. Therefore, different methods already in use in wireless communications are also applicable
in this case including equalization (either in electrical or optical domain), multicarrier modulation and OFDM, spread
spectrum technique, diversity, and MIMO. Some of those methods have already been studied for use in both point-to-point
links and broadcasting/multicasting applications [14–21, 36]. The key disadvantage of MMF-based MIMO is the problem of
coupling between too many modes, which makes it very complex to perform all calculations needed when MIMO signal
processing is applied. This is the reason why the application of conventional MMFs is limited to short-reach applications.
However, by using the so-called few-mode fibers (FMFs) and few-core fibers (FCFs), the mode-coupling related calculations
became simpler, which makes them excellent candidates for high-capacity long-haul transmission. An example of OFDM-
MIMO signal transmission over FMF/FCF, by employing both polarizations in coherent detection scheme, is shown in
Fig. 8.11. Figure 8.11a shows an example of MIMO system with MTx transmitters and MRx receivers, while for simplicity
sake, we can assume that MTx ¼ MRx ¼ M. For FMF/FCF applications, we can interpret this MIMO scheme as one employing
M OAM modes. Since this system employs both polarizations, it is able to carry 2 M independent two-dimensional (QAM)
data streams. The laser diode CW output signal is split among MRx branches, each containing a transmitter, whose
configuration is given in Fig. 8.11b. Two independent data streams are used as inputs to each branch OFDM transmitter. In
fact, we can assume that there exist two independent transmitters in each branch (denoted as x and y), one aimed for
transmission over x-polarization and the other for transition over y-polarization. To convert OFDM signals to optical domain,
I/Q modulators are used. The optical OFDM signals from both polarizations are combined by polarization beam combiner.
The PDM data streams are further mode-multiplexed, as illustrated in Fig. 8.11a, and transmitted over FMF/FCF-based
system of interest. At the receiver side, M different outputs from FMF/FCF are used as inputs to M coherent receivers, after
mode demultiplexing is done. The basic configuration of a coherent detector scheme is shown in Fig. 8.11c, with additional

Data Data
1x 1y Output 1x
Coherent
Transmitter 1 receiver 1
Mode-demultiplexer
Mode-multiplexer

FMF/MMF Output 1y

Source Power … (OFDM)

… … … …
laser diode splitter MIMO
signal
processing
Transmitter MTx
Output MRx, x-pol.
Local Coherent
Data Data
laser diode receiver MRx
MTx, x-pol. MTx, y-pol.
Output MRx, y-pol.

(a)
Data ix RF OFDM transmitter x I
x-pol. state
Optical BD
I Q
From hybrid
I/Q Modulator ix PBS BD
FMF Q
From
transmit PBS PBC
laser From I
I/Q Modulator iy local PBS BD
Optical
laser
I Q hybrid
BD
Data iy RF OFDM transmitter y y-pol. state
Q
(c)
Fig. 8.11 Coherent optical OFDM-MIMO over FMF/FCF: (a) system block diagram, (b) i-th branch transmitter configuration, (c) coherent
receiver configuration. PBS, polarization beam splitter; PBC, polarization beam combiner; BD, balanced detector. RF OFDM transmitter and
receiver configurations on the diagram are described in the chapter on OFDM
8.2 MIMO Optical and Wireless Techniques 467

Delay 1 Delay 1 … Delay 1

Delay 2 Delay 2 … Delay 2

Delay 3 Delay 3 … Delay 3

Delay 4 Delay 4 … Delay 4

Section 1 Section 2 Section K

Fig. 8.12 Conceptual model of FMF in the strong coupling regime

details to be provided in OFDM chapter. One local laser oscillator is used for all coherent receivers, to reduce the system cost.
The in-phase (I) and quadrature (Q) channel outputs from both polarizations are used as inputs to OFDM-MIMO signal
processing block. The use of OFDM elevates simultaneously the impacts of chromatic dispersion, mode-coupling effects, and
PMD. The scheme shown in Fig. 8.11 can be used in both point-to-point and multipoint-to-multipoint applications. If optical
MIMO system over FMF/FCF has MTx transmitters and MRx receivers, the corresponding input-output relation can be
expressed as:

M Tx X
X P  
jωc ðtτp,k Þ
yn ð t Þ ¼ hmn ðk Þe xm t  τg,k þ zn ðt Þ, ð8:65Þ
m¼1 k¼1

where yn(t) is the signal received by n-th receiver (n ¼ 1,2,..,MRx), hmn(k) is the channel gain from m-th transmitter through the
k-th mode, while τp,k and τg,k denote the phase and group delay associated with k-th mode, respectively. Also, ωc is the optical
carrier frequency, P is the number of spatial modes, zn presents the noise process, while xm(t) denotes the transmitted signal
originating from m-th transmitter. We should notice that Eq. (8.65) deals with a single polarization state. When the group
delay spread Δτg ¼ τg,P  τg,1 is small as compared to symbol duration, we can assume that x(t  τg,k)
x(t  τg), meaning
that all paths arrive at approximately the same time, which is valid as distance does not exceed a certain critical value. The
sampled baseband equivalent of function given by Eq. (8.65), with sampling interval Ts, can be written in a matrix form as
follows:

yðlÞ ¼ HxðlÞ þ zðlÞ, ð8:66Þ

where:
2 3 2 3 2 3
x1 ðlT s Þ y1 ðlT s Þ z1 ðlT s Þ
6 ... 7 6 ... 7 6 ... 7 X P
xðlÞ ¼ 4 5, yðlÞ ¼ 4 5, zðlÞ ¼ 4 5, ½Hnm ¼ hnm ðkÞejωc τp,k : ð8:67Þ
k¼1
xM Tx ðlT s Þ yM Rx ðlT s Þ zM R ðlT s Þ

We can assume that, based on previous results presented in [15], the number of modes is large so that each transmitter/receiver
launch/sample from sufficiently different groups of modes and that the product of carrier frequency and phase-delay spread is
large enough to satisfy relationship ωcτp,k  2π. In such a case, each element in H structure can be considered as having a
uniform distribution of phase. Accordingly, the elements of H will have a complex Gaussian distribution, while their
amplitudes follow Rayleigh distribution [1]. Accordingly, it is feasible sending the estimated channel state information
(CSI) back to the transmitter since the multimode channel transfer function varies at relatively slower rate as compared to data
rate that is being transmitted.
The group delays (GDs) among the spatial modes in FMFs are proportional to the square root of link length if there is a
strong coupling among the spatial modes, which is similar to dependence of the first-order PMD on the length of fiber link
[22, 23]. Accordingly, the model with optical fiber sections, used in analysis of PMD coupling, can also be applied to FMF by
presenting the total link length by the number of independent fiber sections. By applying such a model, joint probability
density function (PDF) of GDs in FMF has been derived in [23]. The FMF is divided into K sections, as shown in Sect. 8.12,
468 8 Diversity and MIMO Techniques

and the coarse-step modeling method is adopted, which is a standard technique for modeling the polarization effects in single-
mode fibers. The length of each section should be larger than the fiber correlation length. Ideally, the section lengths should be
randomly distributed, to avoid periodic artifacts in the wavelength domain. Let us consider FMF operating in strong mode
coupling regime. The propagation characteristics of k-th section of FMF having D modes can be modeled by a D  D channel
matrix with terms Mk(ω) given as [23, 24] (see also [7]):

H k ðωÞ ¼ V k Λk ðωÞU {k , k ¼ 1, 2, ⋯, K ð8:68Þ

where Vk and Uk are frequency-independent random complex unitary matrices representing the strong random coupling of
modes in neighboring sections in the absence of mode-independent gain or loss; Λk ðωÞ ¼ diag½ejωτ1 , ejωτ2 , . . . , ejωτD  is
the diagonal matrix describing the uncoupled modal group delays; while τi is the group delay of the i-th mode. (K denotes the
number of fiber sections.) The number of sections (K ) depends on the length of the fiber and correlation length, so the total
frequency domain response can be represented by concatenating the individual sections, as follows:

H total ðωÞ ¼ H 1 ðωÞ⋯H k ðωÞ⋯H K ðωÞ: ð8:69Þ

The corresponding optical MIMO scheme for FSO channels is illustrated in Fig. 8.13. We assume that the beam spots on
the receiver side are sufficiently wide to illuminate a whole photodetector array, which simplifies the transmitter-receiver
pointing problem. The transmitter side is composed of MTx optical transmitters, with configuration of the m-th transmitter
shown in Fig. 8.13b. To facilitate explanations, only the single polarization is shown. The external modulator is used to
impose the data on CW laser beam signal. Alternatively, signal CW laser can be used for all optical transmitters with the help
of power splitter. For direct detection applications, the external modulator is typically Mach-Zehnder modulator, imposing
on-off keying (OOK) signal. For coherent detection, the external modulator is either PSK or I/Q modulator. After amplifying
the modulated optical signal, by an optical amplifier, such as erbium-doped fiber amplifier (EDFA), the signal is expanded by
an expanding telescope, properly designed so that the whole optical receiver array is illuminated. After transmission through
atmospheric turbulence channel, each optical receiver after the compressing telescope is converted into electrical domain by
using either direct or coherent optical detection. For direct detection, the n-th receiver is composed of a compressing telescope
and PIN photodetector, followed by a trans-impedance amplifier (TIA). For coherent optical detection, as shown in Fig. 8.13c,
after compressing telescope, the incoming optical signal is mixed with local oscillator (LO) laser signal in an optical hybrid,
followed by two balanced detectors (BDs) to determine I- and Q-channels’ estimates. Alternatively, a single LO laser can be
used for the whole receiver array, with the help of a power splitter.
The channel coefficients for direct detection are real numbers. In the case of the repetition MIMO, the n-th optical receiver
output, for OOK, can be represented by [5, 25]:

X
M Tx
yn ðlÞ ¼ xðlÞ I nm þ zn ðlÞ; n ¼ 1, . . . , M Rx ; xðlÞ 2 f0, Ag ð8:70Þ
m¼1

where A denotes the intensity of the pulse in the absence of scintillation and x(l ) denotes data symbol at the l-th time slot. Inm
represents the intensity channel coefficient between the n-th photodetector (n ¼ 1,2,. . .,N ) and the m-th (m ¼ 1,2,. . .,M)
optical source, which is described by the Gamma-Gamma probability density function. The optical transmitters and receivers
are positioned in such a way that different transmitted symbols experience different atmospheric turbulence conditions. In
Eq. (8.70), zn denotes the n-th receiver TIA thermal noise that is modeled as a zero-mean Gaussian process with double-sided
power spectral density N0/2.

8.2.2 Parallel Decomposition of Optical and Wireless MIMO Channels

The multiplexing gain of MIMO system, introduced by Eq. (8.60), originates from the fact that a MIMO channel can be
decomposed into a number of parallel independent channels, so that the data rate can be increased RH-times, where RH is the
rank of the channel matrix H. Let us perform the singular value decomposition of channel matrix H given by Eqs. (8.64),
(8.67), and (8.69). In singular value decomposition, we have the following: (i) the ∑  matrix corresponds to scaling
operation, (ii) the columns of matrix U, obtained as eigenvectors of the Wishart matrix HH{, correspond to the rotation
8.2 MIMO Optical and Wireless Techniques 469

Fig. 8.13 Optical MIMO scheme

for FSO channels: (a) generic Optical Optical
MIMO scheme, (b) m-th
transmitter 1 receiver 1

MIMO processing
transmitter configuration, and (c)
n-th receiver configuration for

…
…
coherent optical detection. EDFA,
erbium-doped fiber amplifier; CL, Optical Optical
compressing lenses; OF, optical receiver n
transmitter m
fiber; BD, balanced detector

…
…
Optical Optical
transmitter MTx receiver MRx

Atmospheric turbulence
channel
(a)
The m-th data stream
EDFA
CL
Laser External
diode m modulator m OF
Expanding
telescope Light beam through
The m-th optical transmitter
turbulent channel

(b)

LO laser
diode n

I-channel
Optical BD
hybrid Q-channel
BD
Compressing
telescope The n-th optical receiver

(c)

operation, and (iii) the other rotation matrix V that has for columns the eigenvectors of H{H. The rank of matrix H corresponds
to the multiplexing gain. In FMF/FCF we can ensure the full rank of channel matrix all the time by employing OAM
eigenstates, which is not guaranteed if regular multimode fiber links are employed.
The parallel decomposition of the optical channel, illustrated in Fig. 8.14, is obtained by introducing the following two
transformations to the channel input vector x and channel output vector y: (i) transmit precoding, in which the input vector of
symbol x to transmitter is linearly transformed by pre-multiplying with rotation matrix V, and (ii) receiver shaping, in which
the receiver vector y, upon optical coherent detection, is multiplied by rotation matrix U{. With these two operations, we
effectively performed the following manipulation:
0 1
~y ¼ U { @|{z}
H x þ zA ¼ U { UΣV { V~ U { z ¼ Σ~
x þ |{z} x þ ~z, ð8:71Þ
UΣV { ~z
470 8 Diversity and MIMO Techniques

I/Q
coherent

Mode-demultiplexer
modulators

Mode-multiplexer
FMF/MMF detectors

Complex … x = Vx y = U†y

…
symbols

…
x x y y

Fig. 8.14 The parallel decomposition of optical MIMO channel. The received vector y can be written as y ¼ Hx + z (where we omitted the
photodiode responsivity and load resistance terms for simplicity of explanations), with z denoting the equivalent noise process

x1 y1

z1
1

x2 y2

z2
2
…

xN yRH

z RH
RH

Fig. 8.15 Equivalent parallel optical MIMO channel. The λi denotes the i-th eigenvalue of the Wishart matrix HH {

Fig. 8.16 The beam-forming Transmitter Receiver

strategy in optical MIMO. Optical array Strong-coupling in FMFs/MMFs array
v1 u*1
transmitter and receiver
Tx 1 Rx 1
configurations are similar to those h11
shown in Fig. 8.13b, c
…

h1MTx
Mode-demultiplexer
Mode-multiplexer

v2 u*2
x y
Tx 2 Rx 2
vM Tx uM* Rx
hM Rx 1
…

hM Rx M Tx
…

Tx MTx Rx MRx

and successfully decomposed the optical MIMO optical channel into RH ¼ rank(H) parallel single-input single-output (SISO)
channels. The equivalent optical channel model is shown in Fig. 8.15.
If there is no concern about multiplexing gain, a simpler scheme known in wireless communication literature as beam-
forming [1–3] can be used instead. In this scheme, as illustrated in Fig. 8.16, the same symbol x, weighted by a complex scale
8.2 MIMO Optical and Wireless Techniques 471

Fig. 8.17 The parallel Transmit

decomposition of wireless MIMO antennas Receive
channel I/Q antennas
mod.

…
Complex x = Vx y = U†y

…
…

…
symbols

…
x x y y
y = Hx + z

Fig. 8.18 The beam-forming h11

strategy in wireless MIMO
v1 u*1
h1MTx

v2 u*2
x y

vM Tx hM Rx 1 uM* Rx
…

…
hM Rx M Tx

factor, is sent over all spatial modes. Therefore, by applying the beam-forming strategy, the precoding and receiver matrices
become just column vectors, so it is V ¼ v and U ¼ u. The resulting received signal is now given by:

y ¼ u{ Hvx þ u{ z, kuk ¼ kvk ¼ 1, ð8:72Þ

where with |||| we denoted the norm of a vector. When the optical channel matrix H is known at the receiver side, the received
SNR can be optimized by choosing u and v as the principal left and right singular vectors of the channel matrix H, while the
corresponding SNR is given by σ2maxρ, where σmax is the largest eigenvalue
 of the Wishart matrix HH{. The channel capacity
beam-forming scheme can be expressed as C ¼ B log 1 þ σ max SNR , which corresponds to the capacity of a single SISO
2 2

channel having the equal channel power gain σ 2max. On the other hand, when the channel matrix is not known, we will need to
do maximization only with respect to u. The array gain of beam-forming diversity is between max(MTx,MRx) and MTxMRx,
while the diversity gain equals MTxMRx:
The wireless MIMO channel can be parallelized in a fashion similar to that shown in Fig. 8.14, and this parallelization is
illustrated in Fig. 8.17. Clearly, the parallel wireless MIMO channel model is identical to that shown in Fig. 8.15. The
corresponding beam-forming strategy for wireless MIMO is shown in Fig. 8.18.

8.2.3 Space-Time Coding: ML Detection, Rank Determinant, and Euclidean Distance Criteria

The symbol transmitted over the coherent MIMO optical channel with spatial modes or over the wireless MIMO channel is
rather a vector than a scalar since channel has the input-output relationship given as y ¼ Hx + z. In the case when the signal
design extends over both space coordinate (by multiple spatial modes in FMF/MMF or multiple transmit antennas) and time
coordinate (by multiple symbol intervals), we can refer to this design as the space-time coding. Under the assumption that
space-mode channel is static for the duration of Ns symbols, its input and output become matrices, with dimensions
corresponding to space coordinate (spatial modes) and time coordinate (the symbol intervals). The output matrix is then
given as:
472 8 Diversity and MIMO Techniques

Y ¼ HX þ Z, ð8:73Þ

where:
     
X ¼ ½x1 , x2 , . . . , xN s  ¼ X ij M Tx N s , Y ¼ y1 , y2 , . . . , yN s ¼ Y ij M Rx N s ,
  ð8:74Þ
Z ¼ ½z1 , z2 , . . . , zN s  ¼ Z ij M Rx N s :

As before, we use MTx to denote the number of transmitters/transmitter antennas and MRx to denote the number of receivers/
receiver antennas. Therefore, the transmitted matrix X is obtained by concatenating transmitted column vectors in Ns channel
uses. On the other hand, the received matrix Y is obtained by concatenating received column vectors in Ns symbol intervals.
Finally, the Z matrix entries are circularly symmetric zero-mean
 complex Gaussian random variables with variance N0. The
{
corresponding correlation matrix is given by RZ ¼ ZZ ¼ N s N 0 IM Rx , where IM Rx is the identity matrix of size MRx,
indicating that the additive noise affecting the received signal is spatially and temporally independent.
Let us observe a case with space-time coding in which the receiver has the knowledge of the channel matrix H. We also
assume that the channel matrix H is fixed for the duration of the entire space-time codeword. To analyze such a channel, we
need to know the PDF of MTxMRx channel matrix entries. There exist several simplified models suitable for MIMO processing
studies:

(i) Rich scattering model, in which channel matrix H entries are independent circularly symmetric complex zero-mean
Gaussian random variables [2, 26, 28]. In other words, the H entries are zero-mean spatially white (ZMSW) [1].
(ii) Fully correlated model, in which the H entries are correlated circularly symmetric complex zero-mean Gaussian random
variables.
(iii) Separately correlated model, in which the H entries are correlated circularly symmetric complex zero-mean Gaussian
random variables, wherein the correlation between the entries are separable into two factors, corresponding to receive and
transmit correlations:
 
½H mn ½H  m0 n0 ¼ ½Rmm0 ½T nn0 , ð8:75Þ

where R (MRx  MRx) and T (MTx  MTx) are Hermitian nonnegative definite matrices. This model is applicable when the
surrounding objects on transmitter and receiver sides introduce the local antenna-element correlations while not having
influence on the other end of the link [2, 26, 30, 37]. The channel matrix for separately correlated model can be
decomposed as:

H ¼ R1=2 H ZMSW T 1=2 , ð8:76Þ

where HZMSW is the matrix of circularly symmetric complex zero-mean Gaussian random variables with unit variance.
Unless otherwise stated we will adopt the rich scattering model. As an illustration, the simplest space-time code, the delay
diversity Scheme [2, 38, 39], is essentially rate-1/MTx repetition code, whose transmitted code matrix for MTx ¼ 2 is given by:

x0 x1 x2 ⋯
X¼ : ð8:77Þ
0 x0 x1 ⋯

Given that each transmitted symbol transverses in total MTxMRx independent paths, this scheme achieves the full diversity
with the cost of very low data rate, only 1 symbol/channel use.
Under maximum likelihood (ML) detection, the optimum transmit matrix of the space-time code is obtained as the result of
the following minimization problem [1, 2, 7, 26]:

ð8:78Þ
8.2 MIMO Optical and Wireless Techniques 473

where the minimization is performed over all possible space-time input matrices X. In Eq. (8.78), we use the operation kAkF to
denote the Frobenius (Euclidean) norm of matrix A = [A]mn defined as:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X X
: pffiffiffiffiffiffiffiffiffiffiffiffiffi
kAkF ¼ ðA, AÞ ¼ TrðAA{ Þ ¼ m
ja j2 ,
n mn
ð8:79Þ

P P
where we use (A,B) to denote the scalar product of matrices A = [A]mn and B = [B]mn, defined as ðA, BÞ ¼ m n amn bmn .
Because it is extremely difficult to calculate the exact error probability, we employ the union bound [2]:

ð8:80Þ

where the piece-wise error probability (PEP) is given by:

   2 
b  b  Y¼HXþZ
P X ! X ¼ Pr Y  H X < kY  HX kF 2
¼
F
 2 
 b 2
¼ Pr HX þ Z  H X  < k HX þ Z  HX k F
F
0 2 1 ð8:81Þ
 
     
B  C
¼ PrB H X  X b þ Z < kZk2 C ¼ Pr kHD þ Zk2 < kZk2 ,
@  FA F F
 |fflfflfflfflffl{zfflfflfflfflffl} 
 D

F

b By expressing the Frobenius norm in terms of scalar product, we can rewrite the previous
where D is defined as D = X  X.
expression as follows:
 
b ¼ PrððHD þ Z, HD þ ZÞ < ðZ, ZÞÞ
P X!X
     
¼ Pr Tr ðHD þ ZÞ H { D{ þ Z{  ZZ{ < 0 ð8:82Þ
 
¼ Pr kHDk2F þ 2ðH, ZÞ < 0 :

The variance of Gaussian random variable (H, Z) is given by:

*2 0 132 +
D E D E
ðH, ZÞ2 ¼ 4Tr@ H Z{ A5 ¼ ½TrðH I ZI þ HQ ZQ Þ2
|{z} |{z}
HI þjH Q ZI jZQ
nX X o2 
¼ m
½H I mn ½ZI nm þ ½H Q mn ½ZQ nm
n
XX D E D E
¼ ½HI 2mn ½ZI 2nm þ ½H Q 2nm ½ZQ 2nm
m n |fflfflfflfflffl{zfflfflfflfflffl} |fflfflfflfflffl{zfflfflfflfflffl} ð8:83Þ
N 0 =2 N 0 =2
N 0 XX
¼ ½H I 2mn þ ½H Q 2nm
2 m n
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
kH k2E
N0
¼ kH k2E
2

The PEP can now be written as:

474 8 Diversity and MIMO Techniques

* 0sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1+
  kHDk2F A
b ¼
P X!X Q@ : ð8:84Þ
2N 0

   
By representing kHDk2F as kHDk2F ¼ Tr HDD{ H { ¼ Tr H { HDD{ , we can conclude that after the spatial interference
cancellation H { H = IM Tx , and the PEP is upper bounded by:
0sffiffiffiffiffiffiffiffiffiffiffi1
  2
b  Q@ kDkF A:
P X!X ð8:85Þ
2N 0

By using the following approximation for Q-function Q(x)  exp(x2/2), we can upper bound the Q-function in Eq. (8.84) by:
0sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1 " #
k HD k 2
k HD k 2
Q@ FA
 exp  F
: ð8:86Þ
2N 0 4N 0

Under Rayleigh fading assumption, in other words assuming that channel coefficients are i.i.d. complex Gaussian ,
the PEP will be upper bounded by:
!M Rx
  1
b 
P X!X h {
i : ð8:87Þ
det IM Tx þ DD
4N 0

The PDF of i.i.d. complex Gaussian entries of a random matrix A, of size M  N, is given by [2, 40]:
  
π MN exp Tr AA{ , ð8:88Þ

This result can be used to prove the following theorem [2, 41].

Theorem If A is a given M  M Hermitian matrix such that IM + A > 0, and B is an M  N matrix whose entries are i.i.d. with
, then:
    
exp Tr ABB{ ¼ detðIM þ AÞN : ð8:89Þ

To prove this theorem, let us represent matrix B as follows: B ¼ [b1, b2, . . ., bN], where bn denotes the n-th column of B.
The expected value of exp[Tr(ABB{)] can be determined with the help of Eq. (8.88) as follows:
Z
       
exp Tr ABB{ ¼ π MN exp Tr ABB{  BB{ dB

Z N ð8:90Þ
  
¼ π M exp Tr ðA þ IM Þbb{ db :

After the following transformation b ¼ (A + IM)1/2c, since the Jacobian of this transformation is det(IM + A)1, we arrive at
Eq. (8.89).
Eq. (8.84) can be now upper bounded by:
* "  #+
  Tr H { HDD{
b 
P X!X exp  , ð8:91Þ
4N 0

and by directly applying (8.89), we derive Eq. (8.87). Because the determinant of a matrix can be determined as the product of
its eigenvalues, we can rewrite Eq. (8.87) as:
8.2 MIMO Optical and Wireless Techniques 475

!M Rx
  MTx
b  ∏ 1
P X!X λk
, ð8:92Þ
k¼1 1 þ 4N 0

where λk (k ¼ 1,2,. . .,MTx) are eigenvalues for DD{. Let us use to denote the set of nonzero eigenvalues, with the size of set
corresponding to rank of DD{. For sufficiently high SNR, defined as γ ¼ 1/4N0, the inequality (8.92) simplifies to:

ð8:93Þ

The first term in PEP upper bound is independent on SNR, while dependent of product of eigenvalues of DD{ as well as power
MRx. Because this term is independent of SNR, it does not change the slope of PEP but rather shifts the PEP curve to the left; in
a similar fashion as a channel code, it can be related to the coding gain of the space-time code. To improve the coding gain, we
need to maximize the minimum of the determinant of DD{ over all possible pairs X, X b ; we refer to this criterion as the
determinant criterion. On the other hand, the second term is SNR dependent, and the cardinality of affects the slope of PEP
curve indicating that diversity gain is given by . Thus to maximize the diversity gain, we have to design space-time
code on such a way that the matrix DD{ is of full rank equal to MTx, and we refer to this design criterion as the rank criterion.
When DD{ is of the full rank, we can write:

M Tx  
∏ λk ¼ det DD{ , ð8:94Þ
k¼1

and the necessary condition for matrix DD{ to be of the full rank is that space-time code block length must be equal or higher
than the number of transmit antennas.
The determinant in Eq. (8.87) can be written as follows:
     
det IM Tx þ γDD{ ¼ 1 þ γTr DD{ þ ⋯ þ γ M Tx det DD{ : ð8:95Þ

Clearly, for small SNRs with γ  1 the determinant and consequently PEP depend predominantly on Tr(DD{), in other words
on Euclidean distance squared between X and X b . This is the reason why the space-time code design criterion based on
Eq. (8.95) is commonly referred to as the Euclidean distance criterion. When SNR is high with γ  1, the PEP is dominated
by det(DD{), in other words the product of eigenvalues of DD{, and the rank determinant criterion is relevant here. Therefore,
when SNR is small, the Euclidean distance among codewords is relevant, in a similar fashion as in a channel code, and the
PEP performance is dictated by the additive noise rather than fading. For high SNRs, the rank of DD{ dictates the PEP
performance, and fading effects are relevant.
To study the error probability in waterfall region, we can allow the number of received antennas to tend to infinity, for a
fixed number of transmitted antennas. In that case, the spatial interference get cancelled as H { H ! IM Tx , and the PEP, based
on inequality (8.85), tends to AWGN case:
0sffiffiffiffiffiffiffiffiffiffiffi1
  M !1
b Rx! Q@ kDkF A,
2
P X!X ð8:96Þ
M Tx <1 2N 0

indicating that union bound on error probability is only dependent on Euclidean distances among space-time codewords. In
this particular case, the PEP is not affected by the number of transmitted antennas, but only the multiplexing gain that grows
linearly with the number of transmitted antennas.
When the number of received antennas tends to infinity, we need to change the variance of entries in H from 1 to 1/MRx to
avoid the divergence problems. The PEP needs to be modified as follows:
0 1M Rx
 
b @ 1
P X!X h iA : ð8:97Þ
DD{
det IM Tx þ 4M Rx N 0

The corresponding determinant in (8.97) can be represented now as:

476 8 Diversity and MIMO Techniques

  M Tx
γ γ   γ  
det IM Tx þ DD{ ¼ 1 þ Tr DD{ þ ⋯ þ det DD{ : ð8:98Þ
M Rx M Rx M Rx

When the SNR increases at the same rate as the number of received antennas, the rank and determinant criteria are applicable.
On the other hand, when SNR is fixed and finite, the Euclidean distance criterion should be used instead.

8.2.4 Relevant Classes of Space-Time Codes (STCs)

In this section we describe several relevant classes of space-time codes including Alamouti code, orthogonal designs, linear
space-time block codes, and trellis space-time codes.

8.2.4.1 Alamouti Code (Revisited) and Orthogonal Designs

The Alamouti code is given by the following coding matrix [10]:

x1 x 2
X¼ , ð8:99Þ
x2 x 1

in which, during the first time interval, symbol x1 is transmitted using spatial mode (antenna 1) 1, while symbol x2 occupies the
spatial mode (antenna) 2. In the second time interval, the symbol x 2 is transmitted by spatial mode (antenna) 1, while the
symbol x 1 goes over spatial mode (antenna) 2. The corresponding spatial mode/antenna channel matrix for two transmitted
modes (antennas) and two received modes (antennas), as illustrated in Fig. 8.19, can be written as:

h11 h12
H¼ : ð8:100Þ
h21 h22

The received current signals after coherent optical detection as well as the receive antenna outputs can be organized in
matrix form as follows:
" # " #" # " #
y11 y12 h11 h12 x1 x 2 z11 z12
¼ þ
y21 y22 h21 h22 x2 x 1 z21 z22
" # ð8:101Þ
h11 x1 þ h12 x2 þ z11 h11 x 2 þ h12 x 1 þ z12
¼
h21 x1 þ h22 x2 þ z21 h21 x 2 þ h22 x 1 þ z22

The combiner outputs, described in [10], can be obtained as follows:

 
ex1 ¼ h 11 y11 þ h12 y 12 þ h 21 y21 þ h22 y 22 ¼ jh11 j2 þ jh12 j2 þ jh21 j2 þ jh22 j2 x1 þ noise
  ð8:102Þ
ex2 ¼ h 12 y11  h11 y 12 þ h 22 y21  h21 y 22 ¼ jh11 j2 þ jh12 j2 þ jh21 j2 þ jh22 j2 x2 þ noise

indicating that the spatial interference is cancelled.

h11
h21
x1
h12
x1,x2
Transmitter Combiner
h22 x2

Fig. 8.19 The Alamouti scheme for two transmitted and two received antennas
8.2 MIMO Optical and Wireless Techniques 477

The general Alamouti code with two transmit antennas and arbitrary number of receive antennas can be obtained by
generalization.
The received signal in the Alamouti code with two transmitters and one receiver can be rewritten in the following form:

y 1 ¼ h1 x 1 þ h2 x 2 þ z 1 ,
ð8:103Þ
y 2 ¼ h 1 x2 þ h 2 x1 þ z 2 :

The corresponding matrix representation is given by:

   
y1 h1 h2 x1 z1
¼ þ e þ ez:
¼ Hx ð8:104Þ
y 2 h 2 h 1 x2 z 2
|fflfflfflfflfflffl{zfflfflfflfflfflffl}|ffl{zffl} |ffl{zffl}
x
e
H
^
z

e { from the left, we obtain:

By multiplying both sides of MIMO equation y ¼ Hx + z by H
 
e {y ¼
H He H {
e { z = j h1 j 2 þ j h2 j 2 x þ H
xþH e { z, ð8:105Þ
|ffl{zffl}
ðjh1 j 2
þjh2 j ÞI22

and the spatial interference is successfully compensated for. By using the Alamouti design, we were able to determine the
modified channel matrix H e orthogonal to the channel matrix H up to the multiplication constant |h1|2 + |h2|2. This is the reason
why the Alamouti code belongs to the class of orthogonal designs.
Another interesting example of orthogonal design for MTx ¼ 3, MRx ¼ 1, Ns ¼ 4 is given by the code matrix:
2 3
x1 x 2 x 3 0
6 7
X ¼ 4 x2 x 1 0 x 3 5: ð8:106Þ
x3 0 x 1 x 2

The corresponding MIMO model of this system is given by:

2 3 2 3
y1 2 3 z1
6y 7 x1 x 2 x 3 0 6 7
6 27 6 7 6 z2 7
y ¼ 6 7 ¼ HX þ z ¼ ½h1 h2 h3 4 x2 x 1 0 x 3 5 þ 6 7: ð8:107Þ
4 y3 5 4 z3 5
x3 0 x 1 x 2
y4 z4

Let us rewrite the MIMO model as follows:

2 3 2 3 2 3
y1 h1 x 1 þ h2 x 2 þ h3 x 3 2 z1 3 2 3
h1 h2 h3 2 3 z1
6 7 6 7 6 7 x1
6 y2 7 6 h1 x2 þ h 2 x1 7 6 z2 7 6 h h 0 7 6 z 7
6 7 6 7 6 7 6 2 76 7 6 2 7
6 7=6 7þ6 7¼6 7 4 x2 5 þ 6 7
1

6 y 7 6 h x3 þ h x1 7 6 z 7 4 h3 0 h 1 5 4 z3 5
4 35 4 1 3 5 4 35 x3 ð8:108aÞ
h3
h2 |fflffl{zfflffl}

z 4

y4 h2 x3 þ h 3 x2

z 4 0
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl} x |fflffl{zffl ffl}
e
H
^
z

e þ ez:
¼ Hx

{
e from the left, we obtain:
By multiplying both sides of MIMO Eq. (8.107) by H
 
e {y ¼
H He {H e { z = jh1 j2 þ jh2 j2 þ jh3 j2 x þ H
xþH e { z, ð8:108bÞ
|ffl{zffl}
ðjh1 j
2
þjh2 j þjh3 j ÞI3
2 2
478 8 Diversity and MIMO Techniques

and the spatial interference is compensated for. Because three symbols are transmitted in four symbol intervals, the code rate
of this space-time code is 3/4 signal per channel use, which is smaller than in Alamouti scheme whose rate is 1. Moreover, it
has been shown in [42] that the code rate of orthogonal designs with the number of transmit antennas larger than two cannot
exceed 3/4. However, by concatenating a linear encoder with a layered space-time mapper, it is possible simultaneously to
achieve the full-diversity and full-rate for any number of transmitted and received antennas as shown in [43].

8.2.4.2 Linear Space-Time Block Codes

The Alamouti code belongs to the class of linear space-time block codes [12]. In linear space-time codes, the L symbols x1,
x2,..,xL are transmitted by using MTx transmitters in Ns time intervals, and their code matrix has the form [2]:

X
L
X¼ ðal Al þ jbl Bl Þ, al ¼ Refxl g, bl ¼ Imfxl g ð8:109Þ
l¼1

where Al and Bl are the complex matrices of dimension MTx  Ns. For instance, the Alamouti code can be expressed in the
following form:
" # " #
x1 x 2 a1 þ jb1 a2 þ jb2
X¼ ¼
x2 x 1 a2 þ jb2 a1  jb1
" # " # " # " # ð8:110Þ
1 0 1 0 0 1 0 1
¼ a1 þ jb1 þ a2 þ jb2
0 1 0 1 1 0 1 0

Let us define the transmitted vector as follows:

x ¼ ½ a1 b 1 . . . a L b L  T :
e ð8:111Þ

Let the noise vector ez be obtained by stacking the columns of the noise matrix Z on the top of each other, and we denote this
manipulation by ez = vectðZÞ. By using the same manipulation, the received matrix Y can be then converted to received vector
ey as follows:

X
L
ey ¼ vecðY Þ ¼ vecðHX þ ZÞ ¼ e x þ ez,
ðal vecðHAl Þ þ bl vecðjHBl Þ Þ ¼ He ð8:112Þ
l¼1

e is defined as:
wherein L  NsMRx and the modified channel matrix H

e ¼ ½vecðHA1 Þ vecðjHB1 Þ . . . vecðHAL Þ vecðjHBL Þ:

H ð8:113Þ

e
We now perform the QR factorization of H:

H e R,
e ¼Q e ð8:114Þ

e being the unitary matrix and R

with Q e being the upper triangular matrix. By multiplying both sides of MIMO Eq. (8.112) by
{
e from the left, we obtain:
Q

e {ey = Re
Q e {ez,
ex þ Q ð8:115Þ

e x is proportional to bL and can be detected. The next-to-last entry is a linear combination of aL and
wherein the last entry in Re
bL; since bL has already been detected, the aL can be detected as well. The third from the last entry is a linear combination of
bL  1, aL, and bL, so that bL  1 can be detected. This iterative procedure is conducted until all L symbols get detected. For the
best performance, it is a good idea to rearrange the received symbols according to SNRs, so that the received symbol with the
8.2 MIMO Optical and Wireless Techniques 479

highest SNR is placed at the bottom. This approach has certain similarities with the so-called Bell Labs Layered Space-Time
(BLAST) architecture [28].

8.2.4.3 Space-Time Trellis Codes (STTCs)

Another class of space-time codes that are applicable in wireless communication as well as SDM systems are trellis space-
time codes [1, 2, 12]. These codes are similar to trellis-coded modulation (TCM) Schemes [2], in which transition among
states, described by a trellis branch, is labeled by MTx signals, each associated with one transmitter. At each time instant t,
depending on the state of the encoder and the input bits, a different transition branch is chosen. If the label of this branch is
x1t x2t . . . xM
t
Tx
, then transmitter m is used to send signal constellation point xm
t , and all these transmission processes are
simultaneous. The spectral efficiency is determined by the modulation format size M and it is log2(M) bits/symbol. As an
illustration, a QPSK trellis space-time coding schemes with MTx ¼ 2 are shown in Fig. 8.20. Four-state and eight-state STTCs’
examples are provided. QPSK symbols are denoted as {0, 1, 2, 3}, each symbol carrying two bits from the following set {00,
01, 10, 11}. The input symbols select a branch from a given state. The branches are numerated by {0,1,. . .,M-1}. The states
are denoted by the circles. The left column of states denotes the current time instance t, while the right column of states

Fig. 8.20 QPSK trellis space- Time instances

time code with MTx ¼ 2 and Input symbols: 0 1 2 3 t t+1 States
diversity of 2MRx: (a) four-state
STTC and (b) eight-state STTC 00 01 02 03 1
output symbols

10 11 12 13 2
Encoded

20 21 22 23 3

30 31 32 33 4
(a)

Input symbols: 0 1 2 3 t t+1 States

00 01 02 03 1

10 11 12 13 2
Encoded output symbols

20 21 22 23 3

30 31 32 33 4

22 23 20 21 5

32 33 30 31 6

02 03 00 01 7

12 13 10 11 8
(b)
480 8 Diversity and MIMO Techniques

denotes the next time instance t + 1. The initial state and incoming symbol dictate the encoded symbols. The first encoded
symbol is transmitted on the first antenna and the second symbol on the second antenna. The trellis with larger number of
states for the same modulation format and the transmit antennas represents a stronger STTC in BER sense.
Assuming an ideal channel state information (CSI) and that the path gains hnm (m ¼ 1,. . .,MTx; n ¼ 1,. . .,MRx) are known to
the detector, the branch metric for a transition labeled x1t x2t . . . xM
t
Tx
is given as:
 2
M Rx 
X 
 n X
M Tx

 yt  hnm xm
t  : ð8:116Þ
n¼1
 m¼1


where ynt is the received signal at n-th receiver at the time instance t. The Viterbi algorithm [27] is further used to compute the
path with the lowest accumulated metric. Clearly, the trellis space-time codes can achieve higher code rates than orthogonal
designs and other linear space-time block codes; however, the receiver complexity grows exponentially as the number of
transmit antennas increases.

8.2.5 Spatial Division Multiplexing (SDM)

In SDM, we transmit multiple independent data streams over either multipath channels in wireless communications or spatial
modes in SDM fiber-based optical communication systems, as described already in Sects. 8.2.1 and 8.2.2. Compared to space-
time coding that is designed to achieve spatial diversity and transmit at most one symbol per modulation period, the SDM
techniques can achieve the spatial rates equal to min (MTx, MRx).
Generic scheme for wireless SDM scheme is shown in Fig. 8.21. Incoming information data stream might be first encoded
by conventional channel code (not shown in figure), serving as an outer code. The encoded data stream is forwarded to 1:
L demultiplexer, where serial-to-parallel conversation takes place. The key transmitter performs simultaneously STC
encoding, inner channel encoding, and modulation process, thus performing L to MTx mapping. The m-th (m ¼ 1,2,. . .,
MTx) transmitted antenna emits the symbol xm. These MTx symbols are transmitted over wireless MIMO channel of interest
such as rich-scattering channel, described by the channel matrix H (see Eq. (8.64)). The maximum number of independent
streams is determined by the rank of channel matrix RH ¼ rank(H), as described in Sect. 8.2.2, which is upper bounded by
RH  min(MTx, MRx). In order to properly detect independent data streams, the number of received antennas must be larger
than the number of transmitted antennas so that the maximum possible number of independent data streams is upper bounded

Rich-scattering
channel
1 h11 1

1 x1 x1 1
h1MTx
STC STC
(Serial-to-parallel converter)

(Parallel-to-serial converter)

2 2
encoder decoder
Demultiplexer 1:L

Multiplexer L:1

x2
SDM detector

2 + x2 + 2
Incoming
Channel hM Rx 1 Channel
data stream
…
…

encoder decoder
…

MTx MRx
+ +

L
Modulation hM Rx M Tx Demodulation
L
xM Tx xM Tx
Fig. 8.21 Generic SDM scheme for wireless MIMO communications
8.2 MIMO Optical and Wireless Techniques 481

4DTx 1
2m-ary source
… 4D Tx/Rx: Four-dimensional transmitter/receiver
4-D mapper
m bits
4-D modulator 4-D Rx 1
Laser Power Mode- Mode-

…
diode splitter multiplexer demultiplexer
…

…
FMF
4D Tx MTx 4-D Rx MRx
m bits

(a)
Qx,i Ix,i

3 dB MZM

MZM /2
From DFB/ to fiber/power
PBS PBC
power splitter combiner
MZM /2
MZM: Mach-Zehnder modulator
MZM PBS(C): polarization-beam splitter
4D modulator (combiner)
Qy,i Iy,i

(b)
Extrinsic LLRs
Balanced ADC
Mode-
PBS coherent 1
Calculation

demultiplexer LDPC Decoder 1

Bit LLRs

detector ADC
APP
Balanced ADC demapper
Local laser coherent LDPC Decoder m m
PBS
diode detector ADC

(c)
Fig. 8.22 Mode-multiplexed 4D coded modulation scheme for FMF-based optical communication systems: (a) overall transmitter configuration,
(b) 4D modulator configuration, and (c) receiver configuration

by MTx. Let RTx denote the code rate of both outer and inner channel codes (obtained as a product of individual code rates);
then the spectral efficiency (SE) of the SDM scheme is upper bounded by:

SESDM  M Tx RTx log 2 ðM Þ ½bits=s=Hz, ð8:117Þ

where M is the signal constellation size (assuming that all independent data streams employ the same constellation size). At
the receiver side, each receive antenna detects the complex linear combination of signal from all transmit antennas, as
illustrated by Eq. (8.61). The purpose of SDM detector is to compensate for spatial interference and provides accurate
estimates of transmitted symbols. The SDM detector configurations are subject of study in incoming sections, which are more
general and applicable not only SDM scheme but also STC and uncoded MIMO schemes.
482 8 Diversity and MIMO Techniques

The main purpose of this section is to introduce relevant SDM schemes and describe their transmitter configurations. After
SDM detection, demodulation takes place, followed by STC decoding and inner channel decoding. The L estimated data
streams are multiplexed together, and corresponding output is passed to outer channel decoder, which delivers estimated
information sequence to the end users.
The corresponding SDM scheme relevant in optical communications has already been described in Sects. 8.2.1 and 8.2.2
(see, e.g., Fig. 8.11, with details related to STC and channel coding omitted). Here we briefly describe one of SDM schemes
introduced in [28]. Mode-multiplexed four-dimensional (4D) coded modulation system architecture is shown in Fig. 8.22a.
Given that this scheme is based on FMF, the full rank of channel matrix can easily be achieved so that we can write
MTx ¼ MRx ¼ N. The corresponding 4D modulator and 4D receiver configurations are depicted in Fig. 8.22b, c. The
m ¼ log2(M ) independent streams of information are encoded using LDPC codes of code rates ri ¼ ki/n (i ¼ 1,. . .,m) where ki
and n represent the information block length and codeword length, respectively. The codewords are written row-wise into
block interleaver, while m bits at every symbol interval read column-wise from the block interleaver. These m-bits are used to
determine four coordinates of corresponding signal constellation point stored inside the LUT. The N 4D streams are combined
together using the mode multiplexer and transmitted over FMF transmission system of interest. On the receiver side, after
mode demultiplexing (Fig. 8.22a), the 4D detection is performed by employing polarization diversity receiver (Fig. 8.22c),
followed by ADCs. Then compensation of mode-coupling effects is performed. The 4D coordinate estimates are forwarded to
the APP demapper where the symbol LLRs are calculated, which are further used to calculate bit LLRs needed in LDPC
decoding.

8.2.5.1 BLAST Encoding Architectures

In the rest of this section, we will describe the family of Bell Laboratories Layered Space-Time (BLAST) [29–32] architecture,
followed by multi-group space-time coding techniques. These techniques are applicable to both wireless MIMO
communications and optical MIMO communications with coherent optical detection. The family of BLAST architectures
includes:

(i) Vertical BLAST (V-BLAST) architecture.

(ii) Horizontal BLAST (H-BLAST) architecture.
(iii) Diagonal BLAST (D-BLAST) architecture.

In V-BLAST encoding architecture, shown in Fig. 8.23, the incoming information data stream is used as input to time-
domain (TD)/temporal encoder. The corresponding codeword is demultiplexed into MTx parallel streams. Each parallel stream
is then modulated with the help of I/Q modulator, and after corresponding amplification, such modulated signal is transmitted
toward received antennas. The keyword parallel originates from serial-to-parallel conversion, after which the parallel data
streams are depicted vertically. Given that each layer has its own I/Q modulator, it is possible to implement this scheme such
that different layers can have different rates. Because at each symbolling interval MTx symbols can be transmitted in total, the
SE is given by Eq. (8.117), where RTx is now the code rate of temporal code. It has been shown in [33] that diversity ranges

Fig. 8.23 V-BLAST encoding Antennas

principles. TD: time-domain V-BLAST encoding architecture
(temporal) 1
I/Q
modulator 1
(Serial-to-parallel converter)

…
Demultiplexer 1:MTx

m
Incoming I/Q
TD encoder
data stream & symbol mapper modulator m
…
…

MTx
I/Q
modulator MTx
8.2 MIMO Optical and Wireless Techniques 483

Fig. 8.24 H-BLAST encoding Antennas

principles H-BLAST encoding architecture
1
TD encoder 1 I/Q
& symbol mapper modulator 1

…
…
m
Incoming
Demultiplexer 1:MTx TD encoder m I/Q
data stream (Serial-to-parallel conv.) & symbol mapper modulator m

…
…
MTx
TD encoder MTx I/Q
& symbol mapper modulator MTx

Fig. 8.25 The illustration of D-BLAST encoding architecture

D-BLAST architecture of
encoder. TD time-domain Antennas
TD encoder 1 I/Q
(temporal) & symbol mapper modulator 1 1

…
Stream
Incoming rotator
Demultiplexer 1:MTx TD encoder m I/Q

…
MTx
data stream (Serial-to-parallel conv.) & symbol mapper modulator m

TD encoder MTx I/Q

& symbol mapper Modulator MTx

from MRx  MTx + 1 to MRx, depending on the layer detected. Given that overall performance is dominated by the smallest
spatial diversity level, the maximum diversity gain is about MRx  MTx + 1.
The encoding process in H-BLAST, illustrated in Fig. 8.24, can be interpreted as follows. Each of the parallel sub-streams
undergoes TD/temporal block encoding, with codeword length Ns. Because every coded symbol is detected by all MRx
received antennas, H-BLAST can achieve at most a diversity order of MRx.
To restore the full diversity order MTxMRx, we need to modify the H-BLAST by inserting the stream rotator [1, 45]. In this
scheme, instead of transmitting the independent space-time codewords on separate antennas, by stream rotator the codeword
symbols from a given space-time codeword are rotated across transmit antennas, so that a codeword is effectively spread over
all MTx antennas, as illustrated in Fig. 8.25. Such obtained BLAST scheme is commonly referred to as diagonal-BLAST or
D-BLAST Scheme [1, 2, 26, 45]. Let the mth time-domain (TD)/temporal encoder codeword be denoted by xm ¼
½ xm1 xm2 ⋯ xmN s . The stream rotator transmits n-th codeword symbol on n-th antenna, where n ¼ 1,2,. . .,MTx. When
the TD codeword length exceeds MTx, then the rotation begins again from the first antenna. The corresponding code matrix
can be described as:
2 3
x11 x21 x31 ⋯ xN s 1 0 ⋯
60 x12 x22 x32 ⋯ xN s 2 ⋯7
6 7
X¼6 7: ð8:118Þ
40 0 ⋱ ⋱ ⋱ ⋱ ⋯5
0 0 0 x1M Tx x2M Tx x3M Tx ⋯

Even though that D-BLAST scheme archives full diversity, for sufficiently long codewords, there are many unused space-
time dimensions, denoted by zeros in Eq. (8.118). This problem can be solved by the so-called threaded space-time
architecture, in which the symbols are distributed in the codeword matrix on such a way to achieve full spatial span MTx
(to guarantee desired spatial diversity order) and full temporal span Ns (to guarantee desired temporal diversity order to deal
with fast fading effects). The key idea is to properly fill symbols from previous and incoming codewords so that there are no
zero symbols in X.
484 8 Diversity and MIMO Techniques

Fig. 8.26 The illustration of MG-STCM encoding architecture Antennas

MG-STCM architecture of
encoder 1
b1 m1

Space-time I/Q mod. 1

(Serial-to-parallel converter)
component m1

…
…

…
Demultiplexer 1:MTx
encoder 1 I/Q mod. m1
Incoming Outer
MTx (mg-1)

…
channel
data stream encoder bg mg
Space-time I/Q mod.
component MTx

…
…

…
encoder g I/Q mod. MTx

8.2.5.2 Multi-group Space-Time Coded Modulation (MG-STCM)

The BLAST architectures, mentioned in previous subsections, do not employ any STC but instead represent SDM schemes.
The schemes without STC do not provide a good diversity order. There were many attempts to develop a scheme that would
be simultaneously SDM scheme while providing excellent diversity orders. One such scheme get introduced by Tarokh,
Naguib, Seshadri, and Calderbank in 1999 [34], which can be called multi-group space-time coded modulation (MG-STCM).
The encoder architecture for MG-STCM is provided in Fig. 8.26. Input information sequence is optionally encoded by an
outer channel encoder of codeword length b. The channel encoder codeword sequences is split into g groups of bits each
having b1, b2, . . ., bg bits, respectively. The i-th (i ¼ 1,2,. . .,g) group of bi bits is fed to space-time component encoder i,
generating the codeword length mi.
The space-time encoders could be of either STBC or STTC or could be of different types. Clearly, the following two
conditions hold:

X
g X
g
bi ¼ b, mi ¼ M Tx : ð8:119Þ
i¼1 i¼1

Similarly to BLAST schemes suffering from interlayer spatial interference, this scheme suffers from inter-group spatial
interference. The interference cancellation must be performed on a group-by-group basis rather than on a layer-by-layer basis.
The outputs of space-time encoders are mapped to corresponding constellation points with the help of I/Q modulators whose
corresponding outputs drive transmit antennas. The codeword matrix of MG-STCM encoder can be represented as:
2 3
x11 ⋯ x1N s
6⋮ ⋮ 7
6 7 2 3
6 7 XC1
6 xm 1 1 ⋯ xm 1 N s 7
6 7 6 XC 7
6 7 6 27
X ¼ 6⋮ ⋮ 7¼6 7, ð8:120Þ
6 7 4⋮ 5
6 xM m þ1,1 ⋯ xM Tx mg þ1,N s 7
6 Tx g 7 XCg
6 7
4⋮ ⋮ 5
xM Tx 1 ⋯ xM Tx N s

where the i-th (i ¼ 1,2,. . .,g) code Ci codeword matrix is given by:
2 3
xm1 þ⋯mi1 þ1,1 ⋯ xm1 þ⋯mi1 þ1,N s
6 7
X Ci ¼ 4 ⋮ ⋮ 5: ð8:121Þ
xm1 þ⋯mi1 þmi ,1 ⋯ xm1 þ⋯mi1 þmi ,N s

The channel matrix for MG-STCM case be represented as:

8.2 MIMO Optical and Wireless Techniques 485

2 3
h11 ⋯ h1m1 ⋯ x1,M Tx mg þ1 ⋯ h1M Tx
6 7
H¼6
4⋮ ⋮ ⋮ 7
5 ð8:122Þ
hM Rx 1 ⋯ hM Rx m1 ⋯ xM Rx ,M Tx mg þ1 ⋯ hM Rx M Tx
 
¼ H C1 H C2 ⋯ H C g ,

where H Ci is the submatrix corresponding to the i-th space-time component code of dimensions MRx  mi. The corresponding
MIMO channel model is given by Y = HX + Z, as for STCs. The truncated channel matrix in which the first m1 + . . . + mi
columns are omitted can be defined as:
 
HCC1,2...,i ¼ H Ciþ1 H Ciþ2 ⋯ HCg : ð8:123Þ

Based on Fig. 8.26, we conclude that the rank of H CC1 is upper bounded by rankðH C1 Þ  M TX  m1 : The dimensionality of
the null space of H CC1 , NullðH CC1 Þ, and rank of H C1 is related by [34, 35]:

dim½NullðH CC1 Þ þ rankðH C Þ ¼ M Rx : ð8:124Þ

Therefore, the dimensionality of the null space of H CC1 is lower bounded by:

dim½NullðH CC1 Þ  M Rx  ðM TX  m1 Þ: ð8:125Þ

 
Let the set of orthonormal vectors spanning the null space of H CC1 be denoted as v1 , v2 , ⋯, vM Rx ðM TX m1 Þ : Let us arrange
the vector as the rows of the Hermitian matrix Θ, which reminds us to the parity-check matrix of an linear block code:
2 3
v1
6 v2 7
:6 7
ΘCC1 ¼6 7: ð8:126Þ
4⋮ 5
vM Rx ðM TX m1 Þ

Because vectors vi span the null space H CC1 , they are orthogonal to H CC1 and we can write:

ΘCC1 H CC1 ¼ 0: ð8:127Þ

Decoding at the receiver side is performed in an iterative fashion. The key idea of decoding process is to decode each code
separately while suppressing the spatial interference from other component codes. Because space-time code C1 employs m1
antennas, the signals from remaining MTx  m1 antennas represent the spatial interference in decoding process for C1. The
zero-forcing (ZF) spatial interference approaches described below can be employed. Let us multiply the received matrix from
the left by ΘCC1 to obtain:
0 1
B H
|{z} X þ Z C
B " |{z} # C
ΘCC1 Y ¼ ΘCC1 B H C1 H CC1 X C1 C ¼ ΘCC1 ðHX þ ZÞ
@ A
X CC1 ð8:128Þ
¼ ΘCC1 H C1 X C1 þ ΘCC1 H CC1 X CC1 þ ΘCC1 Z
|fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |fflfflfflffl{zfflfflfflffl}
¼0 eZ
e
¼ ΘCC1 H C1 X C1 þ Z,

and we have been able to successfully compensate for the spatial interference from other groups of STCs, in other words the
:
e C1 ¼
inter-group spatial interference. We can denoted this result by Y ΘCC1 Y. The transmitted block can now be estimated to
minimize the Euclidean distance squared derived from (8.128):
486 8 Diversity and MIMO Techniques

 2
 
 
 
b C1 ¼ arg min ΘCC1 Y  ΘCC1 HC1 XC1  :
X ð8:129Þ
|fflfflfflffl{zfflfflfflffl} 
fX C 1 g 
 e


Y C1
F

b C1 , it can be used to cancel the spatial interference from C1 as

Once the estimate of codeword matrix for C1 is obtained, X
follows:

b C1 :
Y C1 ¼ Y  H C1 X ð8:130Þ

In a similar fashion as for C1, we can now multiply Y C1 by another matrix whose rows span null space of HCC1,2 , denoted as
ΘCC1 , to obtain:
0 1
H X þ Z  H b
X
! B |{z} 2 |{z} 3
C1 C1 C
b C1 B C
: Y  H C1 X
|{z} B H C1 H C2 H CC1,2 X C1 C
e C2 ¼
Y ΘCC1,2 Y C2 ¼ ΘCC1,2 ¼ ΘCC1,2 B 6 7 C
HXþZ B 6 XC2 7 C
B 4 5 C
@ A
XCC1,2
b C1
¼ ΘCC1,2 H C1 XC1 þ ΘCC1,2 H C2 XC2 þ ΘCC1,2 H CC1,2 XCC1,2 þ ΘCC1,2 Z  ΘCC1,2 H C1 X
|fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl}
¼0
 
b C1 þ ΘCC1,2 H C2 XC2 þ ΘCC1,2 Z,
¼ ΘCC1,2 H C1 X C1  X
ð8:131Þ

b C1 . The ML estimate of X
indicating that the spatial interference from other groups is cancelled providing that XC1 ’ X b C2 can
be determined as follows:
 2
b C2 ¼ arg min 
X Ye C2  ΘCC1,2 H C2 X C2 
 : ð8:132Þ
F
fX C 2 g

The i-th codeword (i > 2) is estimated by first cancelling the spatial interference originating from the (i  1)-st STC:

b Ci1 :
Y Ci1 ¼ Y Ci2  H Ci1 X ð8:133Þ
 
Then we need to determine the orthonormal basis vectors that span Null H CC1,2,...,i , which are arranged as the rows of the
matrix denoted by ΘCC1,2,...,i . We now multiply Y Ci1 with ΘCC1,2,...,i from the left side to obtain:
 
:
e Ci ¼ b Ci1 ¼
Y ΘCC1,2,...,i Y Ci1 ¼ ΘCC1,2,...,i Y Ci2  H Ci1 X
X
i1   ð8:134Þ
¼ ΘCC1,2,...,i H Ci XCi þ b C j þ ΘCC1,2,...,i Z,
ΘCC1,2,...,i H C j XC j  X
j¼1

which indicates that the spatial interference from all other previous groups ( j ¼ 1,2,. . .,i  1) is cancelled out providing that
b C j . The ML estimate of X
the following is valid X C j  X b Ci can be determined as follows:

 2
b Ci ¼ arg min 
X Ye Ci  ΘCC1,2,...,i H Ci X Ci 
 ; i ¼ 2, 3, ⋯, g ð8:135Þ
F
fX C i g
8.2 MIMO Optical and Wireless Techniques 487

8.2.6 Linear and Decision-Feedback MIMO Receivers for Uncoded Signals

We will now discuss several suboptimum strategies [2, 7, 26], in order to reduce the ML detector complexity. We will start our
discussion with linear and decision-feedback receivers, applied to the schemes without space-time coding. The key idea
behind linear receivers is to perform certain linear processing operation on received vector so that the transformed channel
matrix does not have off-diagonal elements, which is:

ey = AðH Þy ¼ AðH ÞHx þ AðH Þz: ð8:136Þ

The union bound for the error probability in this case is given by:

1X X
Pe 
M x bx Pðx ! b
xÞ, ð8:137Þ

where the conditional PEP for Euclidean metric is expressed as:

 
Pðx ! b
xjH Þ ¼ P kAy  b
xk2 < kAy  xk2 jH
 
¼ P kAHx  bx þ Azk2 < kAHx  x þ Azk2 jH ð8:138Þ
 
¼ P kdk2 þ 2ððAH‐IÞx, dÞ þ 2ðAz, dÞ < 0jH ,

x, while (a,b) denotes the dot product of vectors a and b. Since Var{(Az, d)} ¼ N0kA{dk2, the PEP becomes:
where d = x  b

* 0vu
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1+
  ukdk2 þ 2ððAH  IÞx, dÞ
b ¼
P X!X Q @t  2 A : ð8:139Þ
2N 0 A{ d

In the zero-forcing receiver, the linear matrix A is chosen to have the form:
8 " þ#
> Σ
>
<V  
U { , M Rx  M Tx
A = Hþ ¼ 0 , Σþ ¼ diag σ 1 , σ 1
, . . . , σ 1
p , 0, . . . , 0
>
>
1 2
: þ { ð8:140Þ
V½ Σ 0 U , M Rx > M Tx
pffiffiffiffi
{
H ¼ U ½ Σ 0 V ; σ 1  σ 2  . . .  0, σ i ¼ λi , Σ ¼ diagðσ 1 , σ 2 , . . . , Þ
   
U ¼ eigenvectors HH { , V ¼ eigenvectors H { H

where we use the superscript + to denote the Moore-Penrose pseudoinverse of the matrix.
Since, for MRx  MTx, the matrix H+H is invertible:

{  1  1
H þ ðH þ Þ ¼ H { H ) Hþ ¼ H{ H H{ , ð8:141Þ

we can compensate for spatial interference in wireless links or mode coupling FMF/MMF links by pre-multiplying the MIMO
equation with H+ from the left:

H þ y ¼ x þ H þ z: ð8:142Þ

The variance for PEP expression is given by:

n o n o
N0 { N 21
VarfðΗ þ z, dÞg ¼ Tr d{ Η þ ðΗ þ Þ d ¼ 0 Tr d{ ðΗ þ Η Þ d , ð8:143Þ
2 2

indicating that the price to be paid during zero-forcing of spatial interference/mode coupling is the noise enhancement effect.
488 8 Diversity and MIMO Techniques

An alternative strategy would be to minimize jointly off-diagonal elements of transformed channel matrix AH and colored
noise Az. In this case, the receiver is known as the minimum mean square error (MMSE) one, with A-matrix chosen as:

 1
AMMSE = H { H þ I=SNR H { , ð8:144Þ

where SNR ¼ E/N0 is the signal to noise ratio for the average energy E of one component of vector x.
The last method to be described in this subsection is related to decision-feedback receivers, which is similar to that
described in the section on linear space-time block codes. The key idea is to perform preprocessing of H by QR factorization
H ¼ QR, Q{ Q ¼ IM Tx , so that we can transform the received vector as:

ey = Q{ y ¼ Rx þ Q{ z, ð8:145Þ

wherein the transformed noise vector retains the properties of z. Since R is an upper-triangular matrix, we can perform the
following optimization:

b
x ¼ arg min ky  RxkF : ð8:146Þ
x

 2
The optimization starts with detection of xM Tx by minimizing the eyM Tx  RM Tx ,M Tx xM Tx  ; next, decision bxM Tx is used to
 2  2
detect xM Tx 1 by minimizing eyM Tx 1  RM Tx 1,M Tx 1 xM Tx 1  RM Tx 1,M Tx bxM Tx  þ eyM Tx  RM Tx ,M Tx bxM Tx  and so on. The
problem with this approach is related to the error propagation.

8.2.7 Suboptimum MIMO Receivers for Coded Signals

In this section we study linear detector strategies, such as zero-forcing and linear MMSE (LMMSE) interfaces [2, 7, 26], but
described in the context of space-time coding.

8.2.7.1 Linear and Zero-Forcing Receivers (Interfaces)

Similarly as in previous section, the key idea behind linear receivers is to perform linear transformation A(H) on received
matrix so that the spatial interference is cancelled:

e = AðH ÞY ¼ AðH ÞHX þ AðH ÞZ:

Y ð8:147Þ

The conditional PEP is given by, obtained by generalization of (8.139):

* 0v
u
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1+
  ukΔk2 þ 2ððAH‐IÞX, DÞ
b ¼
P X!X Q@t  2 A , D = X  X:
b ð8:148Þ
2N 0 A{ D

In zero-forcing interface we chose the linear transformation matrix A to be pseudo-inverse of H, i.e., A ¼ H+, in a fashion
similar to one from Eq. (8.142), so that spatial interference in MIMO wireless communication and mode coupling effect in
FMF/MMF links can be compensated by:

Hþ Y ¼ X þ Hþ Z ð8:149Þ

The Euclidean metrics is given by kHþ Y  X kF, and based on Eq. (8.141), we conclude that the conditional PEP is given by:
2

sffiffiffiffiffiffiffiffiffiffiffi!
kDk2
PðX ! XjHÞ ¼ Q , ð8:150Þ
2σ 2

where the variance is given by, obtained by generalization of (8.143):

8.2 MIMO Optical and Wireless Techniques 489

n o
N0 21
VarfðΗ þ Z, DÞg ¼ Tr D{ ðΗ þ Η Þ D : ð8:151Þ
2

Clearly, the price to be paid for compensation of spatial interference by zero-forcing is the noise enhancement effect.

8.2.7.2 Linear MMSE Receiver (Interface)

In LMMSE interface (receiver), the linear transformation matrix A is chosen in such a way to minimize the MSE, defined by:

MSEðAÞ ¼ hkAY  Xk2F i

¼ hTr½ððAH  IM Tx ÞX þ AZÞððAH  IM Tx ÞX þ AZÞ{ i ð8:152Þ
{ {
¼ Tr½ρðAH  IM Tx ÞðAH  IM Tx Þ þ N 0 AA ,

where the third line in the above equation originates from the i.i.d. zero-mean assumption of X, whose second moment is
denoted by ρ. By setting the MSE variation with respect to A to be zero [2], i.e.:
n h i   
δðMSE Þ ¼ Tr δA ρH ðAH  IM Tx Þ{ þ N 0 A{ þ ρðAH  IM Tx ÞH{ þ N 0 A δA{ ¼ 0, ð8:153Þ

we obtain the following solution for A:

 1  { 1
AMMSE = H { HH { þ I M Rx =SNR ¼ H H þ I M Tx =SNR H { , ð8:154Þ

where the equivalent SNR is defend as SNR ¼ ρ/N0.

The corresponding piece-wise error probability, needed for estimation of upper bound of error probability, is now given as:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
* u !+
u IM Tx 1 {
ukDk þ 2ð½ðH { H þ SNR
2
Þ H H  IM Tx X, DÞ
PðX ! XÞ ¼ Q u  2 : ð8:155Þ
t  IM Tx 1 
2N 0 HðH { H þ SNR Þ D
F

When SNR tends to plus infinity, we obtain the same result as in zero-forcing receiver, as it was expected.
When the number of transmitters is fixed and finite, while the number of receiver tends to plus infinity, as discussed earlier,
the spatial interference/mode coupling is compensated for and we can write H { H ! IM Tx . Therefore, in this case there is no
performance loss with respect to ML detector. On the other hand, when the number of both transmitters and receivers tends to
plus infinity by their ratio as constant MTx/MRx!c > 0, the corresponding PEP tends to [2]:
0sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1
  M !1, M !1 M kDk2 A
b Tx ! Rx
P X!X Q@ j1  cj Rx , ð8:156Þ
M Tx =M Rx !c 2N 0

indicating that there is a performance loss with respect to ML detector. For the same PEP the SNR loss is 10log10|
1  c|1 [dB].

8.2.7.3 Decision-Feedback and BLAST Receivers (Interfaces)

In the linear interface with nonlinear decision-feedback method, illustrated in Fig. 8.21, we first process the received signal Y
linearly by matrix A and then subtract the estimate of the spatial interference/mode-coupling BX b where X b is a preliminary
 
e 
decision on transmitted codeword. The preliminary decision has been made by using the following metric Y  X , where
e ¼ AY  BX:
Y b It is important that the diagonal entries of matrix B are set to zero, so that spatial interference and coupling
effects are effectively removed. We should notice that this strategy has already been used to deal with PMD effect in
polarization division multiplexed coherent optical systems [44]. This scheme is essentially a generalization of decision-
feedback equalizers, used to compensate for ISI.
490 8 Diversity and MIMO Techniques

Y AY Y AY BXˆ
A
-
X̂ Decision
B
circuit

Fig. 8.27 The illustration of linear interface with nonlinear decision feedback

We will describe two attractive versions of the BLAST interface: (i) minimizing MSE in the absence of noise and
(ii) minimizing the MSE in the presence of noise. The first scheme is also known as the zero-forcing vertical/horizontal-
BLAST (ZF V/H-BLAST), while the second scheme is known as MMSE V-BLAST.
In ZF V/H-BLAST, we first perform QR factorization of channel matrix, H ¼ QR, and then determine A and B matrices
from Fig. 8.27 so that the MSE is set to zero. Because in zero-forcing receiver we ignore the presence of additive noise, the
MSE is defined as:
 2    2 
e   b 
MSE ðA, BÞ ¼ Y  X  ¼ AHX  BX  X  : ð8:157Þ
F

For sufficiently high SNRs and when the level of spatial interference/mode
D {E  coupling is not too high, the following is valid
 D {E  {
b b
X
X, so that we can use the following two approximations XX
XX , X { bXb
XX , and after employing them
in (8.157), we obtain:
D E
MSE ðA, BÞ ¼ kðAH  B  IM Tx ÞX k2F
D n oE
¼ Tr ðAH  B  IM Tx Þ{ ðAH  B  IM Tx ÞXX { ð8:158Þ
hXX{ i¼N s ρIMTx D E
¼ N s ρ kAH  B  IM Tx k2F :

By representing the channel matrix as H ¼ QR, the MSE becomes for the following values for matrices A and B:

AZF V‐BLAST ¼ diag1 ðRÞQ{ , BZF V‐BLAST ¼ diag1 ðRÞR  IM Tx , ð8:159Þ

where matrix B is strictly upper-triangular ([B]mn ¼ 0, m  n). The soft estimate of this receiver, based on Fig. 8.26, is
given as:
 
e ¼ diag1 fRgQ{ Y  diag1 fRgR{  IM Tx X
Y b
  ð8:160Þ
= X þ diag1 fRgR{  IM Tx D þ diag1 fRgQ{ Z,

where the first term is the useful term, the second term corresponds to remain mode-coupling fractions from other modes to the
observed mode, while the last term is the colored Gaussian noise.
In MMSE V/H-BLAST, we minimize the MSE by taking into account the presence of additive noise:
 2  D E
e 
MSE ðA, BÞ ¼ Y  X ¼ kAY  BX  Xk2F
F
n o
¼ E kðAH  B  IM Tx ÞX þ AZk2 ð8:161Þ
n o
¼ N s ρ kAH  BX  IM Tx k2 þ SNR1 kAk2 , SNR ¼ ρ=N 0

The minimization of the MSE is performed into two steps. We first minimize MSE with respect to A to obtain:
8.2 MIMO Optical and Wireless Techniques 491

 1
AMMSE ¼ ðB þ IM Tx Þ H { H þ SNR1 IM Tx H { : ð8:162Þ

The residual MSE becomes:

h  1 i
MSE ðBÞ ¼ N s N 0 Tr ðB þ IM Tx Þ H { H þ SNR1 IM Tx ðB þ IM Tx Þ{ : ð8:163Þ

In the second step, the remaining MSE is then minimized over MTx  MTx strictly upper-triangular matrices ([B]mn ¼ 0,
m  n) by Cholesky factorization [2, 26] of:

H { H þ SNR1 IM Tx ¼ S{ S, ð8:164Þ

where S is an upper-triangular matrix. The remaining MSE is now:

 2   2
MSE ðBÞ ¼ N s N 0 ðB þ IM Tx ÞS1   N s N 0 diag ðB þ IM Tx ÞS1 
  2 X
M Tx   ð8:165Þ
¼ N s N 0 diag S1  ¼ N s N 0 ½S 2
m,m
m¼1

 
The minimum is obtained when ðB þ IM Tx ÞS1 ¼ diag S1 : In conclusion, the matrices A and B minimizing the MSE in the
presence of noise are given as follows:

AMMSE V‐BLAST ¼ diag1 ðSÞS{ H { , BMMSE V‐BLAST ¼ diag1 ðSÞS  IM Tx : ð8:166Þ

The soft estimate of this receiver, based on Fig. 8.27, is given by:
 
e ¼ diag1 fSgS{ H { Y  diag1 fSgS  IM Tx X
Y b
    ð8:167Þ
= IM Tx  diag1 fSgS{ X þ diag1 fSgS  IM Tx D þ diag1 fSgS{ H{ Z

where the first term is a useful term, the second term corresponds to residual mode-coupling from other modes to the observed
mode, and the last term is the colored Gaussian noise.
Because in both cases the matrix B is strictly upper diagonal, the steps in V-BLAST algorithms can be summarized as
follows [2]. We use [A]m toh denote i the m-th row in A and [A]mn to denote the m-th row and n-th column. h i We first solve for the
MTx-th row in X b from Y e ¼ ½AY M Tx : And then we solve (MTx-1)-th row in X b from Y e ¼ ½AY M Tx 1 
h i M Tx M Tx 1
½BM Tx 1,M Tx Xb : Since the MTx-th and (MTx  1)-th rows are known, the (MTx  2)-th row in X b can now be determined
h i M Tx h i h i
from Y e ¼ ½AY M Tx 2  ½BM Tx 2,M Tx Xb b
 ½BM 2,M Tx 1 X : We continue this iterative procedure, until we
M Tx 2 M t h i Tx M Tx 1 h i  
reach the first row in X, b which is determined from Y e ¼ ½AY   ½B b b .
1 1,M Tx X  . . .  ½B1,2 X
1 M Tx 2
The receiver in D-BLAST architecture first detects x11 symbol that does not suffer from spatial interference. And then the
symbol x12 is detected, which suffers from spatial interference from the symbol x21, which can be compensated for by a zero-
forcing equalizer. This procedure is continued until all symbols for codeword x1 are detected. Once all symbols of codeword
x1 get detected, their contribution get subtracted from the next received word corresponding to codeword x2 and so forth.

8.2.7.4 Spatial Interference Cancellation by Iterative Receiver (Interface)

Yet another scheme of canceling out the spatial interference in wireless MIMO systems and mode coupling FMF/MMF links
in an iterative fashion is illustrated in Fig. 8.28. This scheme is suitable for compensation of spatial interference in fading
channels. A similar scheme has been proposed in [44] to deal with PMD effects in polarization division multiplexing coherent
optical systems. The key idea behind this scheme is to cancel the spatial interference/mode coupling in an iterative fashion.
The iterative mode-coupling cancellation algorithm can be formulated as follows:
492 8 Diversity and MIMO Techniques

Fig. 8.28 The operation

BXˆ
l+1 l
l + Y Y
principle of iterative spatial
Y
interference/mode-coupling
cancellation scheme - B AH diag AH
X̂ Decision 1
B A = diag R Q†
circuit

e ð0Þ ¼ 0, Y
Step 0 (initialization): Set A¼ diag1 ðRÞQ{ , X e ð0Þ ¼ Y:
 ðk Þ 2
Step 1 (horizontal update rule): Xb ðkÞ ¼ arg min 
Ye  X  :
X
ðkþ1Þ ðk Þ
e
Step 2 (vertical update rule): Y e  ðAH  diagðAH ÞÞX
¼Y b :

We have to iterate steps 1 and 2 until the convergence.

ðk Þ
b
It can be shown that in the absence of nonlinearities and for moderate SNRs, X converges to transmitted X.

8.3 Iterative MIMO Receivers

In this section we study iterative MIMO processing of received signals by employing the factor graph theory [2, 26, 46–49].

8.3.1 Factor Graph Fundamentals

Factor graphs are suitable to describe the factorization of a global function as the product of local function, with each local
function being a product of subset of the variables. As an illustration let us consider maximum a posteriori probability (MAP)
detection/decoding problem. Let x denote the transmitted sequence of length N and y the received sequence. The optimum
receiver (in symbol error probability sense) assigns bxn to the value x from the signal set X that maximizes the APP P(xn ¼ x|y)
given the received sequence y as follows:

bxn ¼ arg max f ðxn ¼ xjyÞ: ð8:168Þ

x2X

This maximization is easy to perform once the APP functions f(xn|y) are known. Assuming that conditional PDF f(x|y) is
known, the APP functions f(xn|y) can be determined as follows:
X X X X X
f ðxn jyÞ ¼ x1 x2
⋯ xn1 xnþ1
⋯ xN
f ðxjyÞ: ð8:169Þ

Clearly, the summation in Eq. (8.169) is performed over all variables except variable xn, and we can denote this summation as
~xn so that we can rewrite Eq. (8.169) as:
X
f ðxn jyÞ ¼ x~n
f ðxjyÞ: ð8:170Þ

This operation is known as marginalization, and once performed we can make a decision based on maximization given by
Eq. (8.168).
Evidently, the complexity of marginalization operation grows exponentially with sequence length N. The marginalization
operation can be greatly simplified if the function to be marginalized ( f ) can be factored as a product of local functions, each
function of the subset of variables. As an illustration let us consider the function f(x1,. . .,x4), which can be factorized as
follows:

f ðx1 , ⋯, x4 Þ ¼ g1 ðx1 , x2 Þg2 ðx1 , x3 Þg3 ðx1 , x4 Þ: ð8:171Þ

The marginal f1(x1) can be computed as:

8.3 Iterative MIMO Receivers 493

Fig. 8.29 The factor graph of x1

f(x1, ⋯x4) ¼ g1(x1, x2)g2(x1, x3)
x1 x1,
g3(x1, x4): (a) non-normal form x2 g1 g2 x3 x2 g1 = g2 x3
and (b) normal form
x1,,

g3 g3

x4 x4
(a) (b)

X
f 1 ð x1 Þ ¼ g1 ðx1 , x2 Þg2 ðx1 , x3 Þg3 ðx1 , x4 Þ
x~1
XX
¼ g1 ðx1 , x2 Þg2 ðx1 , x3 Þg3 ðx1 , x4 Þ ð8:172Þ
x2 x3
X X X
¼ g1 ð x 1 , x 2 Þ g2 ð x 1 , x 3 Þ g3 ðx1 , x4 Þ:
x2 x3 x4

The corresponding factor graph describing this marginalization operation is shown in Fig. 8.29a. Each node represents a
processor performing a given function with arguments being the labels of incoming edges, while each edge serves as a link
over which the nodes exchange the messages. The edge connected to only one node is called the half-edge, and it is terminated
P
by the filled circle. The node g1 performs the following calculation locally: x2 g1 ðx1 , x2 Þ. On the other hand, the nodes g2 and
P P
g3 perform the summations x3 g2 ðx1 , x3 Þ and x4 g3 ðx1 , x4 Þ, respectively.
The factor graph composed only of nodes, edges and half-edges is known as normal graph. The normal graph for graph
shown in Fig. 8.29a is shown in Fig. 8.29b. This normal graph contains the repetition node ¼ , which repeats the same
variable multiple times. This repetition node can be also represented in terms of the so-called Iverson function, denoted by [P]
with value 1 when proposition P is true, and zero otherwise:

: 1, P is true
½P¼ ð8:173Þ
0, P is false

The repetition function for three arguments, denoted as f¼, can be now defined as:
 : 
f ¼ x1 , x001 , x001 ¼ x1 ¼ x001 ¼ x001 : ð8:174Þ

If multiple propositions Pn (n ¼ 1,2,. . .,N ) need to be simultaneously satisfied, we can represent this as follows:

½P1 ^ P2 ^ ⋯ ^ PN  ¼ ½P1 ½P1 ⋯½PN : ð8:175Þ

8.3.2 Factor Graphs for MIMO Channels and Channels with Memory

Let us now observe the MIMO channel, given by Eq. (8.62), namely, y ¼ Hx + z. Providing that the number of transmit
antennas (optical transmitters) is MTx and the number of receive antennas (optical receivers) is MRx, the output-input
relationship can be described in terms of conditional PDF:

h i M Rx  
f ðyjxÞ ¼ C exp ky  Hxk2F =N 0 ¼ C ∏ exp jyn  hn xj2
n¼1 |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
f ðyn jxÞ ð8:176Þ
M Rx
¼ C ∏ f ðyn jxÞ,
n¼1
494 8 Diversity and MIMO Techniques

x1 x2 xM Tx

= = … =

f(y1|x) f(y2|x) … f yM R | x

Fig. 8.30 The factor graph of the optical/wireless MIMO channel

x1 x2 xN

= = … =

f(y1|x) f(y2|x) … f yN |x

Fig. 8.31 The factor graph of the linear ISI channel (linear channel with memory)

where C is the normalization constant and we use hn to denote the n-th row of H, namely, hn ¼ [H]n. The corresponding factor
graph of this MIMO channel is shown in Fig. 8.30.
In linear intersymbol interference (ISI) channel, the output (N+μ)-component vector y and the input N-component vector x
are related by the discrete-time convolution:

X
μ
yn ¼ hk xnk þ zn , ð8:177Þ
k¼0

where μ is the channel memory and h0, h1,..,hμ are channel coefficients; while zn is a zero-mean Gaussian noise sample. The
matrix representation of Eq. (8.177) is given by:

y ¼ Hx þ z, ð8:178Þ

where H is the corresponding channel matrix of format (N + μ)  N. The conditional PDF, describing output-input
relationship, is then:
0  2 1
h i Nþμ  XN 
 
f ðyjxÞ ¼ C exp ky  Hxk2F =N 0 ¼ C ∏ exp @yn  xk hnk  A
n¼1  k¼1

|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} ð8:179Þ
f ðyn jxÞ
Nþμ
¼ C ∏ f ðyn jxÞ,
n¼1

which is very similar to Eq. (8.176), and not surprisingly the corresponding factor graph shown in Fig. 8.31 is very similar to
the one shown in Fig. 8.30.
8.3 Iterative MIMO Receivers 495

x1 x2 xN

f(y1|x1) f(y2|x2) … f(yN|xN)

Fig. 8.32 The factor graph of a channel without memory (memoryless channel)

xn g xn xn g x1 , , xN g xn ' xn '
~ xn n' n

g(x1,…,xN)

…
x1 xN

x1 g x1 xN g xN

Fig. 8.33 The illustration of the sum-product rule

When the channel model is memoryless, μ ¼ 0, yn ¼ h0xn + zn, the conditional PDF is simply:

N
f ðyjxÞ ¼ C ∏ f ðyn jxn Þ: ð8:180Þ
n¼1

The factor graph of the memoryless channel is shown in Fig. 8.32, and it is composed of N disconnected subgraphs.

8.3.3 Sum-Product Algorithm

To compute marginal functions efficiently, the sum-product algorithm (SPA) [2, 5, 7, 26, 47, 50, 51] is used, which is suitable
when the factor graph is free of cycles. In SPA, with each edge we associate two types of messages being the functions of
variable xn, denoted as μ(xn), one for each direction. These messages are vectors, with components calculated for different
values of xn. For binary signaling, the message has the form μ(xn) ¼ (μ0, μ1) ¼ μ0(1, μ1/μ0). Alternatively the logarithm can be
used so that the message is then μ0 (xn) ¼ logμ(xn) ¼ (0,log(μ1/μ0)), and clearly the first component can be omitted. Let u
observe the node function g(x1,. . .,xN), as illustrated in Fig. 8.33. The message to be sent along the edge xn from node g,
denoted as μg!xn ðxn Þ, can be calculated as:
X
μg!xn ðxn Þ ¼ gðx1 , . . . , xN Þ ∏ μg!x0n ðxn0 Þ, ð8:181Þ
xn n0 6¼n

and this rule is commonly referred to as the sum-product rule. In other words, μg!xn ðxn Þ is the product of the local function
g and all messages associated with incoming edges to g, except xn, summed over all variables except xn. The half-edge
message is considered to be 1, while for only one variable the product in (8.181) does not exist and we can write
μg!xn ðxn Þ ¼ gðxn Þ. The message originating from the repetition node sent over edge xn is simply the product of incoming
messages, except xn:
496 8 Diversity and MIMO Techniques

μ f ¼!xn ðxn Þ ¼ ∏ μg!xn0 ðxn0 Þ: ð8:182Þ

n0 6¼n

The SPA can also be applied when factor graph has the cycles, by specifying initial values of local nodes and a stopping
criterion [2, 5, 50, 51]. However, in this case the SPA is a suboptimum algorithm, which performs well when the length of the
shortest cycle in the factor graph, the so-called girth, is sufficiently high.

8.3.4 Sum-Product Algorithm for Channels with Memory

We return our attention now to the MAP detection problem, given by Eq. (8.168). To calculate the f(x|y), we invoke Bayes’
rule:

f ðyjxÞf ðxÞ
f ðxjyÞ ¼ , ð8:183Þ
f ðyÞ

where f(y|x) is conditional PDF and f(x) is the a priori probability of input sequence x, which when the symbols are
independent factors as f ðxÞ ¼ ∏Nn¼1 f ðxn Þ , where N is the length of the sequence to be detected. Given that the PDF of
received sequence of symbols is independent of x, we consider 1/f(y) as constant so that we can rewrite Eq. (8.183) as:

f ðxjyÞ
f ðyjxÞf ðxÞ ¼ f ðyjxÞ∏Nn¼1 f ðxn Þ: ð8:184Þ

Let us define the prior information as the intrinsic message μi(xn) ¼ f(xn) and the extrinsic message (denoted by superscript
e), from symbol demapper point of view, related to edge symbol xn by:
X X
μ e ð xn Þ ¼ f ðyjxÞ∏n0 6¼n f ðxn0 Þ ¼ f ðyjxÞ∏n0 6¼n μi ðxn0 Þ, ð8:185Þ
x~
|fflffl{zfflffl} x~
n n
μi ðxn0 Þ

And then the corresponding sum-product for general channel with memory (including ISI channel) is illustrated in Fig. 8.34.

8.3.5 Iterative MIMO Receivers for Uncoded Signals

Based on similarities of MAP detection problem for channel with memory and MIMO channels, f(x|y) for MIMO channels can
be factorized as follows:

x
!
zfflfflfflfflfflfflffl}|fflfflfflfflfflfflffl{
f ðxjyÞ
f ðyjxÞf ðxÞ ¼ f yj x1 , ⋯, xM Tx ∏M m¼1 f ðxm Þ,
Tx

|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
M Rx
∏ f ðyn jxÞ
ð8:186Þ
n¼1

M Rx  
¼ ∏ f ðyn jxÞ∏M
m¼1 f ðxm Þ, f ðyn jxÞ ¼ exp jyn  hn xj :
Tx 2
n¼1

Fig. 8.34 The illustration of the i

xn f xn
sum-product algorithm for MAP
detection (“equalization”) in f(x1) … f(xn) … f(xN)
channels with memory
i
i
x1 e
x1
i
xn e
xn xN e
xN f y| x i
xn '
n' N
~ xN
x1 xn xN

Symbol demapper f(y|x)

8.3 Iterative MIMO Receivers 497

f(x1) f(x2) f ( xM Tx )

x1 x2 xMTx
= = … =

f(y1|x) f(y2|x) … f yM Rx | x

Fig. 8.35 The illustration of factor graph for MAP detection of uncoded signals for MIMO channels

Fig. 8.36 The illustration of

factor graphs for MIMO receivers f(x1) f(x2) f ( xM Tx )
with (a) ZF and (b) LMMSE xMTx
interfaces x1 x2
f ( xM Tx ) = = … =
f(x1) f(x2)

x1 x2 xMTx
…

f y1 | x1 f y2 | x2 … f yM Tx | xM Tx f y1 | x f y2 | x … f yM Tx | x
(a) (b)

The corresponding factor graph for MAP detection of uncoded signals at the output of MIMO channel is shown in Fig. 8.35.
Because of the existence of cycles, the corresponding SPA is suboptimum. By increasing the number of antennas sufficiently,
the factor graph girth can be increased, and performance of suboptimum iterative MIMO receiver based on SPA can be
improved. However, the complexity of computation would be high.
To reduce the complexity, we can perform pre-compensation of spatial interference by employing suboptimum linear
receivers (interfaces) described in Sects. 8.2.6 and 8.2.7. The key idea, given by Eq. (8.136), is to pre-multiply the received
vector y by a matrix being function of channel matrix H, denoted as A(H), to obtain ey ¼ AðH Þy, so that the spatial interference
is either compensated for or reduced.
The A(H) matrix can be chosen as follows, depending on the corresponding interface:

• In zero-forcing interface as the Moore-Penrose pseudoinverse of H: AZF(H) ¼ H+.

• In linear MMSE (LMMSE) interface, A(H) is chosen on such a way to minimize MSE:

 1
AMMSE = H { H þ I=SNR H {

In zero-forcing H/V-BLAST architecture, after QR factorization of H, the transform matrix is chosen as:

AZF V‐BLAST ¼ diag1 ðRÞQ{

In linear minimum MSE H/V-BLAST architecture, after Cholesky factorization of H { H þ SNR1 IM Tx ¼ S{ S, the transform
matrix is chosen to minimize MSE:

AMMSE V‐BLAST ¼ diag1 ðSÞS{ H { :

After the initial compensation of spatial interference, the corresponding factor graphs for ZF and LMMSE interfaces are
shown in Fig. 8.36.
498 8 Diversity and MIMO Techniques

f(x1) f ( xM Tx 2 ) f ( xM Tx 1 ) f ( xM Tx )

x1 xM Tx 1 xM Tx 1
xMTx
= … = = =

f1 … f M Tx 2 f M Tx 1
f M Tx

fn f yn | xn , xn 1 , , xM Tx

Fig. 8.37 The illustration of factor graph for MIMO receivers with H/V-BLAST architecture

In Fig. 8.36b we use thick edges to denote the diagonal terms after spatial interference cancellation and thin edges to denote
the remaining spatial interference terms. When the spatial interference is well compensated, the following approximation can
be used to further reduce the complexity of the factor graph:

f ðeyn jxÞ ¼ f ðeyn jx1 , ⋯, xM Tx Þ ’ f ðeyn jxn Þ: ð8:187Þ

This approximation is valid when MRx  MTx [2, 52–54].

The factor graph for H/V-BLAST architectures is shown in Fig. 8.37, which is obtained by simplifying the factor graph
shown in Fig. 8.35 by using approximation given by Eq. (8.187). The detection algorithm outputs the solution after the
following steps are executed:
 
bxM Tx ¼ arg max f eyM Tx jxM Tx ,
x2X
ð8:188Þ
bxn ¼ arg max f ðeyn jxn , exnþ1 , ⋯, exM Tx Þ; n ¼ M Tx  1, M Tx  2, ⋯1:
x2X

Before moving to the description of iterative MIMO receivers for space-time coded signals, we briefly describe the factor
graphs for linear block and trellis channel codes.

8.3.6 Factor Graphs for Linear Block and Trellis Channel Codes

The linear block codes (LBCs) can be described as a K-dimensional subspace C of the N-dimensional vector space of N-
tuples, F N2 , over the binary field F2 [55]. For this K-dimensional subspace, we can find the basis B ¼ {g0, g1, . . ., gK  1} that
spans C so that every codeword v can be written as a linear combination of basis vectors v ¼ m0g0 + m1g1 + . . . + mK  1gK  1
for message vector m ¼ (m0, m1, . . ., mK  1); or in compact form we can write v ¼ mG, where G is the so-called generator
matrix with the i-th row being gi. The (N-K)-dimensional null space C⊥ of G comprises all vectors x from F2N such that
xGT ¼ 0, and it is spanned by the basis B⊥ ¼ {h0, h1, . . ., hn  k  1}. Therefore, for each codeword v from C, vhTi ¼ 0 for
every i; or in compact form, we can write vHT ¼ 0, where H is the so-called parity-check matrix whose i-th row is hi.
Therefore, the parity-check matrix of LBC can be written as:
2 3
h0
6h 7
6 1 7
H¼6 7: ð8:189Þ
4⋮ 5
hNK1
8.3 Iterative MIMO Receivers 499

The codeword must satisfy all parity-check equations as discussed above, and this condition can be represented with the help
of Iverson function as follows:

  NK1  
½v 2 C ¼ vhT0 ¼ 0, ⋯, vhTNK1 ¼ 0 ¼ ∏ vhTn ¼ 0 : ð8:190Þ
n¼0

The n-th factor in right-hand side of Eq. (8.190) can be represented as mod-2 adder , connecting the edges involved in the n-
th parity-check equation. A bipartite (Tanner) graph is a graph whose nodes may be separated into two classes (variable and
check nodes) and where undirected edges may only connect two nodes not residing in the same class. The Tanner graph of a
code is drawn according to the following rule: check (function) node c is connected to variable (bit) node v whenever element
hcv in a parity-check matrix H is a 1. In an (N  K )N parity-check matrix, there are N-K check nodes and N variable nodes.
As an illustrative example, consider the H matrix of the Hamming (7,4) code:
2 3
1 0 0 1 0 1 1
6 7
H¼4 0 1 0 1 1 1 0 5
0 0 1 0 1 1 1

For any valid codeword v ¼ [v0 v1. . .vn  1], the checks used to decode the codeword are written as:

• Equation (c0): v0 + v3 + v5 + v6 ¼ 0 (mod 2)

• Equation (c1): v1 + v3 + v4 + v5 ¼ 0 (mod 2)
• Equation (c2): v2 + v4 + v5 + v6 ¼ 0 (mod 2)

The bipartite graph (Tanner graph) representation of this code is shown in Fig. 8.38. Both non-normal and normal versions of
Tanner graph have been shown.
The circles represent the bit (variable) nodes while -nodes represent the check (function) nodes. For example, the variable
nodes v0, v3, v5, and v6 are involved in (c0) and therefore connected to the check node c0. A closed path in a bipartite graph
comprising l edges that close back on itself is called a cycle of length l. The shortest cycle in the bipartite graph is called the
girth. The shortest cycle in Tanner graph in Hamming (7,4) code is 4; therefore, the girth is 4.
Codes described by trellis can also be described using the concept of normal factor graphs. A trellis can be interpreted as a
set of triples (sn  1, vn, sn), describing the transition from state sn  1 to state sn driven by the incoming channel symbol vn at

v0 v1 v2 v3 v4 v5 v6

v0 v1 v2 v3 v4 v5 v6
= = = = = = =
bit-nodes

Check-nodes
c0 c1 c2 c0 c1 c2
(a) (b)
Fig. 8.38 Tanner graph of (7,4) Hamming code: (a) non-normal and (b) normal versions
500 8 Diversity and MIMO Techniques

Fig. 8.39 A node in TWLK 0

graph representing a given trellis vn
section
1
0
sn-1 sn Tn={{0, ,0}, {0, ,3},
1 2 {1, ,1}, {1, ,2}}

3
sn-1 vn { , } sn

trellis
0 0 0

1 1 1
=
Tanner graph
=
TWLK graph

Fig. 8.40 Trellis, TWLK graph, and normal Tanner graph representations of repetition (3,1) code

time instance n  1 (n ¼ 1,2,. . .,N ). Let the set of branch labels connecting the states sn  1 and sn be denoted by Tn. And then
the local function corresponding to the n-th trellis section, shown in Fig. 8.39, is given by:

½ðsn1 , vi , sn Þ 2 T n , ð8:191Þ

and the whole trellis can be described as a product of corresponding Iverson functions of individual trellis sections:

N
½v 2 C  ¼ ∏ ½ðsn1 , vi , sn Þ 2 T n : ð8:192Þ
n¼1

Such factor graph description of trellis is commonly referred to as Tanner-Wilberg-Loeliger-Koetter (TWLK) graph.
As an illustration in Fig. 8.40, we provide trellis, TWLK, and normal Tanner graph representations of (3,1) repetition code,
which has only two codewords (0,0,0) and (1,1,1).
Additional details on decoding algorithms based on Tanner graph and trellis graph representations can be found in Chap. 9.
Here we briefly describe decoding process based on sum-product algorithm. Given that transmitted codewords v from the code
C are equally likely, we have that f(v) ¼ 1/|C| when v is a codeword and zero otherwise, and we can write f(v)
[v 2 C]. The
conditional PDF f(v|y) can be represented as:

f ðvjyÞ
f ðyjvÞ f ðvÞ ¼ f ðyjvÞ½v 2 C : ð8:193Þ
|ffl{zffl}
½v2C 

The factor graph describing the factorization of f(v| y) can be obtained by combining the code graph, describing function
[v2C] and the channel function, and describing the function f(y| v), as shown in Fig. 8.41. The intrinsic (denoted by
superscript e) and extrinsic (denoted by superscript i) messages are defined respectively as:
X
μ i ð vn Þ ¼ v~
f ðyjvÞ∏n0 6¼n μe ðvn0 Þ,
Xn ð8:194Þ
μ e ð vn Þ ¼ v~n
½v 2 C ∏n0 6¼n μi ðvn0 Þ:
8.3 Iterative MIMO Receivers 501

Fig. 8.41 The SPA illustrating

decoding over an arbitrary Channel decoder [v C]
channel
… …
e i e i e
v1 v1 vn vn vN i
vN
v1 vn vN

Symbol demapper f(y|v)

Fig. 8.42 The factor graph for

space-time coded MIMO (with Space-time decoder [X C]
interleaver)
i e i e
i
x1,1
e
x1,1
i
vn e
vn vN vN vN vN
… … …
x1,1 xM Tx ,1 x1, N s xM Tx , N s

Interleaver

… …
… xM Tx , N s
x1,1 xM Tx ,1 x1, N s

f y1 | x1 … f yNs | xNs Symbol demapper

8.3.7 Iterative MIMO Receivers for Space-Time Coded Signals

A codeword generated by a space-time encoder, as already described in Sect. 8.2.3, can be described as MTx  Ns matrix
X ¼ [x1, . . ., xNs], where xn (n ¼ 1,..,Ns) denotes the n-th column of X, and the length of vector xn is MTx. Clearly, rows
correspond to transmit antennas (space dimensions) while columns to time intervals. The received signal is represented as
MRx  Ns matrix Y ¼ [y1, . . ., yNs]. The received and transmitted signals are related by Y¼HX + Z, where H is the channel
matrix and Z is a noise matrix of independent zero-mean symmetric complex Gaussian random variables. (For additional
details on STC-based MIMO system, please refer to Sect. 8.2.3.) In analogy with Eq. (8.193), the conditional PDF f(X|Y) can
be factorized as follows:

Ns
f ðXjY Þ
f ðYjXÞ½X 2 C ¼ ½X 2 C ∏ f ðyn jxn Þ, ð8:195Þ
n¼1

and the corresponding factor graph describing this factorization is shown in Fig. 8.42. The purpose of the interleaver (which is
optional) is to increase the girth of the factor graph. (We use tilde below variable to denote that the corresponding variable is
obtained after interleaving.) To facilitate the explanations, below we omit the interleaver.
From space-time decoder point of view, the extrinsic massage is calculated by:
X
μe ðxm,n Þ ¼ ~
xm,n
½X 2 C ∏ðm0 ,n0 Þ6¼ðm,nÞ μi ðxm0 ,n0 Þ, ð8:196Þ

where μi(xm, n) are messages output by the functional nodes f(ym|xm) computed as:
X
μi ðxm,n Þ ¼ ~
xm,n
f ðyn jxn Þ∏m0 6¼m μe ðxm,n Þ: ð8:197Þ

Once the above messages converge to exact values, the APPs can be computed as follows:
502 8 Diversity and MIMO Techniques

f ðxm,n jY Þ
μi ðxm,n Þμe ðxm,n Þ: ð8:198Þ

This iterative SPA computes repeatedly the two-way messages associated with the edges, after proper interleaving/
deinterleaving, until a predetermined number of iterations is reached or valid space-time codeword is obtained, after which
APPs are computed and used in MAP decoding as described earlier.
To reduce complexity, similarly as in Sect. 8.3.5, it might be a good idea to perform preprocessing as follows:

e ¼ AðH ÞY ¼ AðH ÞHX þ Z,

Y ð8:199Þ

in a similar fashion as described in Sect. 8.2.7.1, to compensate for spatial interference. When the spatial interference is
properly compensated, the factor graph corresponding to each function f ðyn jx Þ get disconnected, which simplifies the SPA.
n
The space-time codes to be used in MIMO receiver processing should be properly matched to the corresponding symbol
demapper. The EXIT chart analysis [2, 5, 50, 51, 56] should be used, which is subject of study in the next chapter.

8.4 Broadband MIMO

The assumption in previous sections was that all frequency components will experience the same channel coefficients. This is
never the correct assumption for fiber-optic communications, because dispersion effects and mode-coupling effects are
frequency-dependent. On the other hand, most MIMO wireless communications operate at data rates when flat fading
assumption is incorrect. Based on Chap. 2 we know that when the symbol rate Rs is larger than (channel) coherence
bandwidth Bc, the channel is frequency-selective; in other words, different frequency components will have different channel
coefficients. The 50% coherence bandwidth Bc, 50% is related to the root-mean-squared (RMS) multipath delay spread σ τMDS by
[57] Bc,50% ’ ð5σ τMDS Þ1 .
We have to modify the MIMO channel model to take into account the number of symbols over which the channel
coefficients can be considered static. If we divide the RMS multipath delay spread σ τMDS with the symbol duration Ts, we can
determine the number of symbols over which the channel coefficients can be considered fixed, denoted as L, namely, L¼ σ τMDS
/Ts. The quasi-static MIMO channel model can be now described as [58]:

M Tx X
X L1
y n ðk Þ ¼ hðnm
lÞ
xm ðk  lÞ; n ¼ 1, ⋯, M Rx ; k ¼ 1, ⋯, N sf ð8:200Þ
m¼1 l¼0

where Nsf is number of symbols in frame (for which the channel can be considered fixed).
To deal with frequency-selective fading, different techniques can be classified as either single-carrier or multicarrier. The
single-carrier systems typically employ the equalization to deal with frequency selectivity, which was subject of investigation
in Chap. 6. On the other hand, the multicarrier techniques are typically based on OFDM, which is a subject of study in
Chap. 7. In this section, we mainly deal with relevant MIMO topics omitted in previous sections and Chaps. 6 and 7, which are
related on MIMO-OFDM and space-frequency block coding (SFBC) [58–60].
The frequency selectivity can be used as an additional degree of freedom, frequency diversity, as discussed at the beginning
of this chapter. The maximum number of possible diversity channels, denoted as MD,max, can be summarized as follows:

M Tx M Rx , for flat fading channels

M D, max ¼ ð8:201Þ
M Tx M Rx L, for frequency‐selective channels

This equation suggests that multiplexing gain of frequency-selective channels is higher than that of flat fading channels, which
was confirmed in [58]. This paper inspired the design of space-time codes to exploit the diversity related to frequency-
selective fading [59, 60].
8.4 Broadband MIMO 503

MIMO OFDM Tx architecture Antennas

1
RF x1t .d . n
x1 k
Optional OFDM I/Q mod. 1 Frequency

Serial-to-parallel converter
Tx 1
space-time -selective

…
Incoming Outer encoder
channel

…
…
channel
data stream encoder +
xtM.dTx. n MTx model
Symbol xM Tx k RF
OFDM I/Q mod. MTx
mapper Tx MTx

(a)
Receive
antennas
MIMO OFDM Rx architecture

y1t .d . n RF y1 k

Parallel-to-serial converter
OFDM STC
Rx 1 demodulator/
…

decoder Estimated
…

MRx + data stream

…

Channel
y tM.dRx. n RF yM Tx k
OFDM decoder
Rx MRx

(b)
Fig. 8.43 MIMO-OFDM system architecture for frequency-selective channels: (a) transmitter and (b) receiver architectures. t.d. time domain

8.4.1 MIMO-OFDM Scheme

The concept of MIMO-OFDM is shown in Fig. 8.43. After serial-to-parallel (S/P) conversion, we perform symbol mapping,
and the corresponding symbols are mapped to different subcarriers in conventional OFDM, as described in the previous
chapter. Optionally the space-time coding can be used as well as the channel coding. The OFDM concept is used to deal with
frequency-selective fading, so that each subcarrier will see flat fading. The multiple transmitter and multiple receiver antennas
are used to improve multiplexing and diversity gains as in traditional MIMO systems. The space-time coding can help exploit
the frequency diversity gain.
In a chapter on OFDM, we have shown that when the guard interval is larger than the RMS multipath delay spread
(together with perfect time and carrier synchronization), we can decompose the wireless fading channel into Nsc parallel
channels, where Nsc is the number of subcarriers used in OFDM. For SMF-based optical communications, as shown in [61],
the guard interval (cyclic prefix duration) must be longer than RMS due to chromatic dispersion, plus RMS due to differential
group delay of two polarization states. Finally, in FMF-based optical applications, we need to account for RMS of mode group
delay. Further assuming that the number of subcarriers is larger than or equal to the number of resolvable paths due to selective
fading (L ), the corresponding OFDM parallelization model we developed can be extended as follows:
504 8 Diversity and MIMO Techniques

yðk Þ ¼ H ðkÞxðkÞ þ zðk Þ; k ¼ 0, 1, ⋯, N sc  1 ð8:202Þ

where x(k), y(k), and z(k) represent the transmitted, received, and noise vectors for subcarrier index k:
2 3 2 3 2 3
x1 ð k Þ y1 ð k Þ z1 ðk Þ
6 x ðk Þ 7 6 y ðk Þ 7 6 z ðk Þ 7
6 2 7 6 2 7 6 2 7
xð k Þ ¼ 6 7 , yð k Þ ¼ 6 7 , zðk Þ ¼ 6 7, ð8:203Þ
4... 5 4... 5 4... 5
xM Tx ðkÞ yM Rx ðkÞ zM Rx ðk Þ

The channel matrix now is also a function of subcarrier index:

2 3
h11 ðkÞ h12 ðkÞ ... h1M Tx ðk Þ
6 h ðk Þ h22 ðkÞ ... h2M Tx ðk Þ 7
6 21 7
H ðk Þ ¼ 6 7; k ¼ 0, 1, ⋯, N sc  1: ð8:204Þ
4... ... ... ... 5
hM Rx 1 ðkÞ hM Rx 2 ðk Þ . . . hM Rx M Tx ðkÞ

Regarding the OFDM subsystem design, it is very similar to what we explained in the previous chapter, except that the total
data rate Rb needs to be split among MTx antennas (optical transmitters for optical communication applications) as follows:

Rb ¼ M Tx Rs log 2 M ½bits=symbol: ð8:205Þ

For FMF-based optical communications, we need to insert a factor 2 in Eq. (8.205), to account for two polarization states used,
while MTx will now account for a number of spatial modes.
In equivalent channel model, given by Eq. (8.202), we interpret the symbols as being transmitted in frequency domain; this
is why the corresponding spatial encoding scheme can be called the space-frequency coding scheme.

8.4.2 Space-Frequency Block Coding-Based MIMO

One possible version of space-frequency block coding (SFBC)-based MIMO is shown in Fig. 8.44.

Fig. 8.44 The SFBC architecture Tx antennas

for frequency-selective channels
1
IDFT Frequency
I/Q mod. 1
Space-frequency block 1
-selective
…

Incoming sequence block encoder

…

channel
of symbols + MTx
IDFT model
Symbol mapper I/Q mod. MTx
block MTx
1

DFT
block 1
…

SFBC Estimated
…

MRx
decoder sequence of symbols

DFT
block MRx
Rx antennas
8.5 MIMO Channel Capacity 505

A block of Ns information symbols is arriving at the input of SFBC encoder. In principle, any STC can be used. The STC
generates MTx encoded outputs. Each output is used as an input of inverse DFT (IDFT) block, composed of S/P converter,
IFFT, and P/S converter. In principle, RF OFDM transmitters can be used in a similar fashion as shown in Fig. 8.43. On the
receiver side, the n-th (n ¼ 1,2,. . .,MRx) receiver antenna signal after A/D conversion undergoes S/P conversion, FFT, and P/S
conversion. And then SFBC decoding takes place, and the estimated sequence of symbols is delivered to the end user. There
are different options; single SFBC decoder can be used or MRx SFBC decoders can be used operating in parallel. The SFBC
principles can be combined with various BLAST architectures.
For additional details on SFBC systems, an interested reader is referred to [59, 60].

8.5 MIMO Channel Capacity

The channel capacity of single-input single-output (SISO) channels has already been studied in Chap. 4. Here we are
concerned with MIMO channel capacity, where the channel model is defined by Eq. (8.62), namely, y ¼ Hx + z, where
x, y, and z denote the transmitted, received, and noise vectors, respectively.
MIMO channel capacity evaluation depends on what is known about the channel matrix or its distribution at the transmitter
and/or receiver sides. We assume that channel state information at the receiver (CSIR) side is always available. We consider
the following relevant cases [62]:

• The channel matrix H is deterministic (static).

• The channel matrix H is random, and at every channel use an independent realization of H is generated; in other words the
channel is ergodic.
• The channel matrix is random but fixed for the duration of transmission; in other words, the channel is non-ergodic.

8.5.1 Capacity of Deterministic (Static) MIMO Channels

The MIMO channel capacity for deterministic channel can be defined as the following maximization problem:

C¼ max ½H ðyÞ  H ðyjxÞ, ð8:206Þ

f X ðx1 , ⋯, xM Tx Þ

where the maximization is performed over the multivariate PDF fX(x). Given that the components of noise vector
are independent on x and zero-mean Gaussian, we can simplify the problem of estimation of entropy H(y| x) ¼
H(Hx + z| x) ¼ H(z), so that the previous equation can be rewritten as:

C¼ max ½H ðyÞ  H ðzÞ: ð8:207Þ

f X ðx1 , ⋯, xM Tx Þ

Given that components of noise vector are i.i.d. and complex circular symmetric Gaussian, the corresponding distribution
of z is a multivariate Gaussian [62]:

{ 1
eðzμz Þ Rz ðzμz Þ ,
1
f Z ðzÞ ¼ ð8:208Þ
detðπRz Þ
D { E  { 
: :  
where μz ¼ hzi and Rz is the covariance matrix of z, defined as Rz ¼ z  μz z  μz ¼ zz  μz μ{z : The corresponding
differential entropy can be determined as:
506 8 Diversity and MIMO Techniques

P P
* ðz μ zm Þ½R1 z mn ðzn μzn Þ
zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{ +
m n m

 {  
H ðzÞ ¼ h log 2 f Z ðzÞi ¼ log 2 detðπRz Þ þ z  μz R1 z z  μz log 2 e
X X D  E 1 

¼ log 2 detðπRz Þ þ log 2 e  z m  μ z z n  μ z Rz mn ð8:209Þ
m n m n
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
½Rz mn

¼ log 2 detðπRz Þ þ log 2 e  TrðIM Rx Þ

¼ log 2 detðeπRz Þ:

This result is applicable to any complex Gaussian multivariate. For i.i.d. zero-mean complex Gaussian multivariate, the
following are valid μz ¼ 0, Rz ¼ σ 2 IM Rx , so that the differential entropy becomes:
 
H ðzÞ ¼ log 2 det eπσ 2 IM Rx : ð8:210Þ

The correlation matrix of y can be determined as:

 
Hxþz
Ry ¼ z}|{ { ¼ HRx H { þ σ 2 IM Rx : ð8:211Þ
y y

The differential entropy of y can be obtained by substituting Ry into (8.209) to obtain:

    
H ðyÞ ¼ log 2 det eπRy ¼ log 2 det eπ HRx H { þ σ 2 IM Rx : ð8:212Þ

After substitution of Eqs. (8.210) and (8.212) into (8.207), we obtain:

     
C¼ max log 2 det eπ HRx H { þ σ 2 IM Rx  log 2 det eπσ 2 IM Rx
f X ðx1 , ⋯, xM Tx Þ
 
HRx H { þ σ 2 IM Rx
¼ max log 2 det ð8:213Þ
f X ðx1 , ⋯, xM Tx Þ σ 2 IM Rx
 
1
¼ max log 2 det IM Rx þ 2 HRx H { ½bits=channel use,
Rx :TrðRx Þ¼P σ
P Tx  
where the transmit power is constrained by M i¼1 xi xi  P, so that the maximization is performed over all input covariance
matrices satisfying this power constraint. If we want to convert units into bit(s), we need to multiply (8.213) by the channel
bandwidth B.
This optimization process is dependent on the availability of CSI on the transmitter side. As in Chap. 4, we know that when
CSI is available on transmitter (CSIT) side, the optimum adaptation strategy is the water-filling. As already discussed in Sect.
8.2.2, by performing the SVD of the channel matrix we obtain:

H ¼ UΣV { ; U { U ¼ IM Rx ; V { V ¼ IM Tx ; Σ ¼ diagðσ i Þ,
pffiffiffiffi ð8:214Þ
σ i ¼ λi ,

where λi are eigenvalues of the Wishart matrix HH{. As discussed in Sect. 8.2.2, we can now decompose the MIMO channel
into RH parallel channels, where RH is the rank of channel matrix. The MIMO channel capacity would be then:

X
RH   σ2 P
P
C¼ P
max log 2 1 þ i SNRi , SNRi ¼ i 2 : ð8:215Þ
Pi : P P i¼1
P σ
i i

The water-filling power allocation strategy achieves the channel capacity:

Pi 1=SNRc  1=SNRi , SNRi  SNRc

¼ ð8:216Þ
P 0, otherwise
8.5 MIMO Channel Capacity 507

where SNRc is the SNR cutoff value (we do not transmit when SNR is below the cutoff value). By substituting (8.216) into
(8.215), we obtain:

X  
SNRi
C¼ log 2 ð8:217Þ
i:SNRi >SNRc
SNRc

When the CSI is not known on the transmitter side, there is no reason to transmit on one antenna with higher launch power
than on the others, and there is no need to introduce the correlation among transmit antennas. Therefore, it makes sense to
apply the uniform power allocation strategy. Assuming the rich scattering scenario, the input covariance matrix is a scaled
version of the identity matrix Rx ¼ ðP=M T x ÞIM T x , and after substitution into (8.213) we obtain:
 
P {
I ðx, yÞ ¼ log 2 det IM Tx þ HH : ð8:218Þ
M Tx σ 2

In the limit, as the number of transmit antennas tends to plus infinity, we have that HH { =M Tx ! IM Tx , so that the mutual
information tends to C ¼ M Rx log 2 ð1 þ P=σ 2 Þ, and the capacity is a linear function in M ¼ min(MTx, MRx). By performing
the SVD of channel matrix, Eq. (8.218) can also be written in the following form:

X
RH  
σ 2 P=σ 2
I ðx, yÞ ¼ log 2 1 þ i , RH ¼ rankðH Þ  min ðM Tx , M Rx Þ ð8:219Þ
i¼1
M Tx

When the transmitter does not know the channel matrix, it can select a fixed data rate R to transmit, which might exceed the
mutual information and an outage will occur. The probability that the actual data rate R exceeds mutual information I(x,y) is
known as the outage probability, defined as:
    
P=σ 2
Poutage ðRÞ ¼ Pr log 2 det IM Rx þ HH { <R : ð8:220Þ
M Tx

The maximum data rate that can be supported at a given outage probability is commonly referred to as the outage capacity.

8.5.2 Capacity of Random MIMO Channels

So far we assumed that the channel matrix is static, which is rarely true in optical and wireless communication channels. The
channel matrix describing the chromatic dispersion is frequency selective, but it is closest to the deterministic assumption.
Other fiber-optic channel impairments such as PMD in SMFs and mode coupling in FMFs are time varying and random. The
FSO channel matrix is time varying with the correlation time ranging typically from 10 μs to 10 ms. The wireless channel
matrix is time varying due to the dynamic nature of the channel and movement through the spatially varying field. As
discussed earlier, we consider two scenarios: (i) the ergodic channel capacity, in which for each channel use the new
realization of channel matrix is drawn, and (ii) non-ergodic channel capacity, in which the channel matrix H is randomly
chosen and kept fixed for the duration of transmission.

8.5.2.1 Ergodic Capacity

When the receiver has the perfect knowledge of the CSI, the mutual information between channel input x and channel output
y, given the channel matrix H, can be determined as [2, 63]:

Iðx; y, HÞ ¼ Iðx; HÞ þ Iðx; yjHÞ ¼ Iðx; yjHÞ

|fflfflffl{zfflfflffl}
¼0 ð8:221Þ
0
¼ hIðx; yjH ¼ H ÞiH0 ,
508 8 Diversity and MIMO Techniques

where we use the subscript H0 in expectation operator to denote different realizations H0 of the random channel matrix H.
The term I(x; H) ¼ 0 as the input vector and the channel matrix are independent of each other.
Based on Eqs. (8.213) and (8.221), we can write the following expression for ergodic capacity:
  
1
C¼ max log 2 det IM Rx þ 2 HRx H { : ð8:222Þ
Rx :TrðRx Þ¼P σ H

When the CSI is known on the transmitter side, assuming that the short-term power constraint P is satisfied, we perform the
SVD of the channel matrix, so that based on (8.215) and (8.222) we obtain:
*  +
X
RH
Pi σ2P
C¼ P
max log 2 1 þ SNRi , SNRi ¼ i 2 : ð8:223Þ
Pi : P P i¼1 P σ
i i H

Clearly, the optimum strategy is water-filling in space domain.

When we are concerned with the less restrictive long-term power constraint [1].

PH
*zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{ +
XM Tx  

i¼1
xi ðH Þxi ðH Þ P ð8:224Þ
H

at each realization, we can use channel matrix-dependent power PH so that the ergodic channel capacity is given by [1]:
*  +
X
RH
Pi σ 2 PH
C¼ max P
max log 2 1 þ SNRi , SNRi ¼ i 2 : ð8:225Þ
PH :hPH iP Pi : P PH i¼1
PH σ
i i H

In this case, the optimum strategy is two-dimensional water-filling in space and time.
When transmitter does not know the CSI, it assumes that the elements of the channel matrix are i.i.d. complex symmetric
circular Gaussian; in other words the envelope of entries follow Rayleigh distribution while the phase is uniform, so that the
optimization problem is to maximize ergodic capacity subject to transmit power constraint [1]:
D h  iE
1
C¼ max log 2 det IM Tx þ 2 HRx H { : ð8:226Þ
Rx :TrðRx Þ¼P σ H

The optimum covariance function that maximizes ergodic capacity is a properly scaled identity matrix so that the ergodic
channel capacity expression simplifies to:
    
P=σ 2
C¼ log 2 det IM Rx þ HH { : ð8:227Þ
M Tx H

For a fixed number of receive antennas, and as the number of transmit antennas tends to infinity, we have that:

     
1 P P M Rx P
HH { ! IM Rx ) log 2 det IM Rx þ 2 IM Rx ¼ log 2 1 þ 2 ¼ M Rx log 2 1 þ 2 , ð8:228Þ
M Tx M Tx !1 σ σ σ

indicating that ergodic channel capacity grows linearly with MRx.

The upper bound can be determined by applying Jensen’s inequality [63] on Eq. (8.227) as follows:
    
P=σ 2
C¼ log 2 det IM Rx þ HH {
M Tx
  2
 H ð8:229Þ
P=σ   
 log 2 det IM Rx þ HH { H ¼ M Rx log 2 1 þ P=σ 2 :
M Tx

By recalling that HH{ and H{H have the same spectrum (the set of eigenvalues), then the following is valid:
8.5 MIMO Channel Capacity 509

   
P=σ 2 { P=σ 2 {
det IM Rx þ HH ¼ det IM Rx þ H H , ð8:230Þ
M Tx M Tx

and after substitution of (8.230) into (8.227), Jensen’s inequality yields to:
    
P=σ 2 {
C ¼ log 2 det IM Rx þ H H
M Tx
   H   ð8:231Þ
P=σ 2  {  M
 log 2 det IM Rx þ H H H ¼ M Tx log 2 1 þ Rx P=σ 2 :
M Tx M Tx

By combining these two inequalities into one, we obtained the following upper bound:

 
  M Rx
C  min M Rx log 2 1 þ P=σ , M Tx log 2 1 þ
2
P=σ 2
: ð8:232Þ
M Tx

For the exact computation of (8.227), we can use Monte Carlo integration approach. For semi-analytical computation, an
interested reader is referred to [2].

8.5.2.2 Non-ergodic Capacity

As discussed at the beginning of MIMO channel capacity section, in non-ergodic channel, the channel matrix is generated at
random but kept fixed for the duration of the transmission. Since the average channel capacity does not have any meaning
here, we can define the instantaneous channel capacity as a function of the channel matrix:
 
P=σ 2 {
CðH Þ ¼ log 2 det IM Rx þ HH , ð8:233Þ
M Tx

and determine the outage probability as the probability that the transmission rate exceeds instantaneous channel capacity
C(H) [2]:

Poutage ¼ PrfCðH Þ < Rg: ð8:234Þ

The maximum rate that can be supported by this channel, for the outage probability Poutage, is called the outage capacity. More
formally, the p-percent outage capacity is defined as the data capacity that is guaranteed in (1  p)  100% of channel
realizations [45]:
 
Pr C ðH Þ  Coutage,p ¼ p: ð8:235Þ

Given that Coutage,p occurs in (1  p)% of channel uses, the throughput of the system associated with, denoted as Rp, can be
determined as [45]:

Rp ¼ ð1  pÞC outage,p : ð8:236Þ

An interesting special case of non-ergodic channels is the so-called block-fading channel model. In this channel model, the
channel matrix is selected randomly and kept fixed of duration of Ns symbols. After Ns symbols a new channel matrix
realization is drawn. The B block-fading channel model can be described as:

yb ½n ¼ H b xb ½n þ zb ½n; b ¼ 1, ⋯, B; n ¼ 1, ⋯, N s ð8:237Þ

where b is the index of the block and n is the index of the symbol within that block. Other assumptions for input, output, and
noise vectors are the same as before. Again, we can perform the SVD of channel matrix Hb in a similar fashion to Eq. (8.214):
510 8 Diversity and MIMO Techniques

H b ¼ U b Σb V {b , ð8:238Þ

and by pre-multiplying (8.237) by U {b , the output-input relationship becomes:

U{b yb ½n ¼ U {b U b Σb V {b xb ½n þ U {b zb ½n

|fflfflfflffl{zfflfflfflffl}
Hb ð8:239Þ
¼ Σb V {b xb ½n þ U {b zb ½n:

when transmitted vector xb[n] get preprocessed on transmitter side, Vbxb[n], the output-input relationship (8.239) becomes
free of spatial interference. As the number of blocks tends to plus infinity, the block-fading model becomes ergodic, and in the
case of CSIR, the channel capacity can be evaluated as:
X  
P=σ 2 2
C¼ log 2 1þ σ , ð8:240Þ
i M Tx i

and this equation is similar to (8.223), so in the case of CSIT, the optimum power allocation strategy is water-filling.
When the number of blocks in block-fading model is finite, the channel capacity calculation is more involved; therefore, an
interested reader is referred to specialized books on MIMO communication principles, such as [2, 26].

8.5.2.3 Correlated Fading Channel Capacity

In a separately correlated MIMO model, introduced earlier, the H entries are correlated circularly symmetric complex zero-
mean Gaussian random variables, and the channel matrix can be decomposed as:

H ¼ R1=2 H ZMSW T 1=2 , ð8:241Þ

where HZMSW is the matrix of circularly symmetric complex zero-mean Gaussian random variables with unit variance. The
CSIR ergodic channel capacity, based on Eq. (8.227), is given by:
    
P=σ 2
C ¼ log 2 det IM Rx þ HH {
M Tx
     ð8:242Þ
P=σ 2
¼ log 2 det IM Rx þ H ZMSW TH {ZMSW R :
M Tx

For high SNRs, when MTx ¼ MRx ¼ M, the following asymptotic approximation is valid [64]:
 
P=σ 2
C
M log 2 þ log ðM!Þ  j log 2 detðTRÞj, ð8:243Þ
M

with the last term representing the loss due to local spatial correlation. Clearly, the spatial correlation degrades the system
performance, while the linear growth with respect to the minimum number of transmit/receive antennas is preserved.

8.5.2.4 MIMO-OFDM Channel Capacity

The channel matrix for MIMO-OFDM system can be generalized as a block diagonal matrix with i-th block diagonal element
corresponding to MTx  MRx MIMO channel [13]:
2 3
H ð 0Þ ... 0
6 7
H ¼ 4... ... ... 5, ð8:244Þ
0 . . . H ðN sc  1Þ
8.6 MIMO Channel Estimation 511

where Nsc is the number of subcarriers. The corresponding ergodic channel capacity expressions given above can be
generalized to OFDM MIMO systems with Nsc subcarriers as follows [13]:
 h  i
1 1
C¼ max log 2 det IM Rx N sc þ 2 HΣH { , ð8:245Þ
N sc TrðΣÞP σ H

where Σ is the covariance matrix of the Gaussian input vector, defined as:

P
Σ ¼ diagfΣk gNk¼0
c 1
, Σk ¼ I ; k ¼ 0, 1, . . . , N sc  1 ð8:246Þ
M Rx N xc M Rx

With uniform power allocation, the OFDM MIMO channel capacity is given by [13]:
*    +
1 X
N sc 1
P=σ 2 {
C¼ log 2 det IM Rx þ H ðkÞH ðk Þ : ð8:245Þ
N sc k¼0 M Rx N sc
H

Various MIMO channel capacity scenarios described above in addition to wireless MIMO communications are also
applicable to free-space optical communications with coherent detection. Regarding fiber-optic communications, these
approaches can be used only in linear regime. Once the fiber nonlinearities become relevant, the generalized nonlinear
Schrödinger equation should be used to study the propagation effects, nonlinear signal-ASE noise interaction, nonlinear
interaction between polarization states, as well as nonlinear interaction among spatial modes in FMF links. By modeling such
nonlinear fiber-optic channel as nonlinear channel with memory, the method proposed in [67] can be used to evaluate the
achievable information rates.

8.6 MIMO Channel Estimation

So far we assumed that CSI was perfect. In this section we discuss how to estimate the MIMO channel coefficients. We have
seen above that parallel MIMO channel decomposition has similarities with parallel decomposition of OFDM channels. So it
is not surprising that similar methods can be used for channel estimation. Here we describe three techniques to estimate
MIMO channel matrix:

(i) Maximum likelihood (ML) MIMO channel estimation.

(ii) Least squares (LS) MIMO channel estimation.
(iii) Linear minimum MSE (LMMSE) MIMO channel estimation.

The common denominator to all three channel estimation techniques is the assumption of using a sequence of Np training
symbols as a preamble. The transmitted pilot matrix Xp has the format MTx  Np, the received matrix has Yp, and the noise
matrix have MRx  Np. The channel matrix H dimensionality is the same as before (MTx  MRx). We also assume that channel
is quasi-static; in other words the H does not change during transmission of pilot symbols. The corresponding output-input
relationship is given by:

Y p ¼ HX p þ Z: ð8:246Þ

8.6.1 Maximum Likelihood (ML) MIMO Channel Estimation

Given that Yp is a random function of the channel matrix H, the conditional multivariate PDF f(Yp| H) can be used for ML
estimation of H as follows:
 
b ML ¼ arg max f Y p jH :
H ð8:247Þ
H
512 8 Diversity and MIMO Techniques

Because the entries of noise matrix Z are ZMSW, from detection theory [65], with basics provided in Chap. 6, we know that
maximizing ML is equivalent to minimizing the Euclidean (Frobenius) norm, and the optimization problem can be
simplified to:
 
b ML ¼ arg min Y p  HX p 2 :
H ð8:248Þ
F
H

Defining now the MSE by:

 2 h  { i
MSE ðH Þ ¼ Y p  HX p F ¼ Tr Y p  HX p Y p  HX p
h i ð8:249Þ
¼ Tr Y p Y {p  Y p X{p H {  HX p Y {p þ HX p X{p H { ,

b ML by solving the equation ∂MSE(H)/∂H ¼ 0. By using the following property of matrices ∂Tr(AB)/
we can determine H
T {
∂A ¼ B , ∂Tr(AB )/∂B ¼ 0, we obtain the following equation for ∂MSE(H)/∂H:

∂MSE ðH Þ  T  T
¼  X p Y {p þ X p X{p H{ ¼ 0: ð8:250Þ
∂H

From Eq. (8.250), we obtain the following solution for H{:

 1
H { ¼ Xp X{p Xp Y {p , ð8:251Þ

or equivalently, after applying the {-operation, we arrive at the following solution for H:
 1
b ML ¼ Y p X { X p X {
H : ð8:252Þ
p p

 1
From matrix theory we know that in order for Xp X{p to exist, the following condition most be satisfied: Np  MTx.

8.6.2 Least Squares (LS) MIMO Channel Estimation

In LS MIMO estimation, we determine the channel matrix H that minimizes the squared error between the actual received
b p ; in other words:
signal Yp and the estimated received signal HX
 2
b LS ¼ arg min 
H b p
Y p  HX  : ð8:253Þ
F
b
H

b is obtained after solving

Clearly, this equation is almost identical to Eq. (8.248), and not surprisingly, the same solution for H
 2
 b p b = 0, which is:
the equation ∂Y p  HX  =∂H
F

 1
b LS ¼ Y p X { X p X {
H : ð8:254Þ
p p

The LS estimation method can be justified as follows. For reasonably high SNRs, the MIMO channel model is given by
Yp ’ HXp. The Xp is not a square matrix and thus it is not invertible. However, if we pre-multiply this MIMO equation by X {p
from the right, we obtain the modified MIMO model Y p X {p ’ HX p X {p , and Xp X{p is an invertible square matrix. By multiplying
 1
now the modified MIMO model by Xp X{p from the right, we obtain the following solution for the channel matrix H ’
 1
Y p X{p X p X {p , which is the same as the LS solution given by Eq. (8.254).
8.6 MIMO Channel Estimation 513

8.6.3 Linear Minimum Mean Square Error (LMMSE) MIMO Channel Estimation

In LMMSE MIMO channel estimation, we try to determine the channel matrix minimizing the MSE between true channel
b
matrix H and channel matrix estimate H:
 2 
b  b 
H LMMSE ¼ arg min H  H  : ð8:255Þ
F
b
H

b is the linear transformation of received signal Yp; in other

To simplify the solution, we assume that channel matrix estimate H
words:

b = Y p A,
H ð8:256Þ

where A is the transformation matrix to be determined. After substitution of (8.256) into (8.255), the equivalent optimization
problem becomes:
D  E D h   iE
b MMSE ¼ arg min H  Y p A2 ¼ arg min Tr H  Y p A H  Y p A { :
H ð8:257Þ
F
A A

As Telatar has shown [62], the expectation and trace operations commute, so we can rewrite the previous equation as follows:
hD  { Ei
b MMSE ¼ arg min Tr
H H  YpA H  YpA
A
2 3
  D E   D E ð8:258Þ
6 7
¼ arg min Tr4 HH {  HA{ Y {p  Y p AH { þ Y p AA{ Y {p 5,
A |fflfflffl{zfflfflffl}
RH

where we used RH to denote the correlation matrix of H, namely, RH ¼ hHH{i. Let us define the MSE by:
D E   D E
MSE ðAÞ ¼ RH  HA{ Y {p  Y p AH { þ Y p AA{ Y {p : ð8:259Þ

By assuming that hHi ¼ 0, hZi ¼ 0 after substitution of Yp ¼ HXp + Z into the previous equation, we obtain:

MSE ðAÞ ¼ RH  A{ X {p RH  Xp ARH þ Xp AA{ X{p RH þ AA{ RZ , ð8:260Þ

 
where we used RZ to denote the correlation matrix of Z, namely, RZ ¼ ZZ{ ¼ N p σ 2 IN p : By taking the partial derivative
with respect to A, we obtain:

∂MSE ðAÞ
¼ Xp RH þ X p A{ X {p RH þ A{ RZ ¼ 0, ð8:261Þ
∂A

and after solving for A, we obtained the optimum solution for transformation matrix as:
 1
Aoptimum ¼ RH X {p X p þ RZ RH X {p : ð8:262Þ

Based on Eq. (8.256), LMMSE estimate of the channel matrix becomes:

 1
b LMMSE ¼ Y p RH X{ Xp þ RZ
H RH X{p : ð8:263Þ
p

Clearly, this result is applicable to any random channel matrix, not necessarily ZMSW model as long as the following is valid
hHi ¼ 0.
514 8 Diversity and MIMO Techniques

8.6.4 Selection of Pilot Signals

So far we have not put any restriction on pilot matrix, except for the corresponding invertible matrix which we required that
Np  MTx. If we choose X p X {p to be a scaled identity matrix:

Np
X p X {p ¼ I , ð8:264Þ
M Tx M Tx

the LS solution becomes:

 
b LS ¼ M Tx HX p þ Z X { ¼ H þ M Tx ZX { ,
H ð8:265Þ
N p |fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} p Np p
Yp

and we can draw the following conclusions:

• The estimate error increases as the number of transmit antennas increases.

• The estimate error decreases as the pilot sequence length increases.
• The estimate error decreases as SNR increases.
• The LS estimate is unbiased when the noise samples are zero-mean Gaussian.

Interestingly enough, the authors in [66] used information theoretic approach and concluded that rows in training matrix
should be orthogonal among each other, which represents just the generalization of Hadamard matrices, namely, Hadamard
matrix, which is defined as a square matrix with entries being either 1 or  1, whose rows are mutually orthogonal. Hadamard
matrices can be defined recursively as follows:

1 1
H 1 ¼ ½1, H 2 ¼ , H 2k ¼ H 2  H 2k1 , ð8:266Þ
1 1

where  denotes the Kronecker product. Therefore, the simplest way to design the pilot matrices satisfying Eq. (8.264), when
MRx > MTx, is to start with Hadamard matrix of size MTx  MTx and add additional MRx  MTx rows that are orthogonal to
already existing rows in Hadamard matrix.

8.7 Massive MIMO

The massive MIMO is a generalization of multiuser MIMO in which the base station (BS) antenna array has hundreds of
antenna elements, supporting tens of single-antenna users [68–75].

8.7.1 Massive MIMO Concepts

The massive MIMO concept is illustrated in Fig. 8.45, in which the number of antenna elements in BS M is much larger (could
be orders of magnitude) than the number of active users’ devices N. The large number of antenna elements allows highly
directive links between the BS and user equipment; thus more energy can be concentrated, and consequently the spectral
efficiency and throughput can be improved. Thanks to higher directivity, the required power per link can be reduced, and thus
a lower cost power amplifier per antenna element can be used. The use of a large number of antenna elements smoothens the
frequency response dependences so that the power control to improve the spectral efficiency can be simplified. To come close
to the channel capacity, thanks to a large number of antenna elements, low complexity precoder and decoder can be used.
For uplink transmission N single-antenna users send either data or pilots for channel estimation toward the base station.
After decoding the base station can detect data sequences belonging to different users, and this process is illustrated on
Fig. 8.46a. Let the transmitted vector originating from N users be denoted by x2C N , where C is the set of complex numbers.
8.7 Massive MIMO 515

Fig. 8.45 Illustrating massive

MIMO concept: (a) uplink and (b) Base station
downlink operations

Massive MIMO
uplink

(a)
Base station

Massive MIMO
downlink

(b)

The received vector at the base station is denoted by y2C M . Let the channel matrix of size M  N be denoted by H. A simple
channel mode for uplink communication will be then:

y ¼ Hx þ iuplink þ n, ð8:267Þ

where n is a noise vector and iuplink describes the interference from simultaneous transmissions. In the simplest approximation,
the elements of the channel matrix can be assumed to zero-mean spatially white.
The downlink channel is used to send the data to different users or to estimate the channel. After the training phase is
completed, the channel state information (CSI) is used in the precoder to transmit the data streams for different users as
illustrated in Fig. 8.46b. The simplified downlink channel model from the BS to the n-th user can be represented by:

yn ¼ hn xn þ idownlink þ n, ð8:268Þ

where xn2C M is the signal vector transmitted by the BS intended for the n-th user, hn is the row channel vector between the BS
and the n-th user, and yn is the n-th user received vector. As before, we use n to denote the noise vector and idownlink to describe
the interference coming from other users.
516 8 Diversity and MIMO Techniques

Fig. 8.46 Massive MIMO Channel

operations of (a) uplink and (b)
downlink model
1 h11
User 1
Data stream 1 1
h1M
2
Base station User 2
Data stream 2 2
with an
Decoder array of hN 1

…
…
antenna

…
…
M
elements User N
Data stream N M hNM
H
(a)
Channel
model
CSI 1 h11
User 1
Data stream 1 1 1
h1M
2
Base station User 2
Data stream 2 2 2
with an
User
Precoder array of hN 1
…

…
scheduler
antenna
…

M
elements User N
Data stream N M M
hNM
H

(b)

Even though that the massive MIMO has many promises, there are some challenges. As an illustration, given that the
wireless channel coherence time is finite, the number of possible orthogonal pilots is limited so that the pilot sequences must
be reused in the neighboring cells resulting in the pilot contamination problem, which is illustrated in Fig. 8.47. In this
example, two single-antenna users, belonging to neighboring cells, employ the same orthogonal pilots to estimate the channels
which results in interference. The pilot contamination affects the channel estimation, resulting in imperfect CSI. The BS needs
the accurate CSI to take the advantages of the massive MIMO concepts. To focus the spatial radiation patterns to respective
users, as illustrated in Fig. 8.45b, the beamforming is used enabled by precoding as described in Sect. 8.2.2.
The CSI needs to be accurate enough and not outdated for precoding to be effective in spatial interference cancelation and
be able to improve the spectral efficiency and thus the pilot contamination affects the precoding as well. Finally, various
detection techniques described in Sect. 8.2 require the CSI so that the pilot contamination affects the detection as well. Other
challenges for massive MIMO include hardware impairments and user scheduling. Massive MIMO employs a large number
of antenna elements at the BS, which increase the overall system complexity and cost. To reduce the cost and complexity,
lower-complexity algorithms and low-cost hardware should be used. Unfortunately, the low-cost hardware introduces various
impairments including the phase noise, I/Q imbalance, and amplifier distortions, to mention a few. When the number of users
increases, the multi-access interference grows, and we need to do proper users’ scheduling, which is illustrated in Fig. 8.48.
8.7 Massive MIMO 517

BS BS

Fig. 8.47 Illustrating the pilot contamination problem

Fig. 8.48 Illustrating massive Channel

MIMO user scheduling
model
CSI 1 h11
User 1
Data stream 1 1 1
h1M
2
Base station User 2
Data stream 2 2 2
with an
User
Precoder array of hN 1
…

…
scheduler
antenna
…

M
elements User N
Data stream N hNM
M M
H

To solve for the pilot contamination problem, different approaches have been proposed [74–78]. The channel estimation
can be based either on OFDM channel estimation described in Sect. 7.5.5 or MIMO channel estimation described in Sect. 8.6.
Alternatively, the combination of these two approaches is relevant too. In particular, the two-dimensional OFDM channel
estimation, described in Fig. 7.14, is applicable here. In this scheme, multiple subcarriers within the OFDM symbol and
multiple OFDM symbols are used for the channel estimation. Since the number of pilots is limited, they need to be reused in
neighboring cells. The simplest approach is to place them in different time and frequency slots so that there is no overlap. For
the time-division duplexing (TDD), the authors in [74] proposed to use time-shifted pilot scheme in which cells are divided
into several groups and for each group the pilots are placed in non-overlapping time slots. In this scenario, neighboring cells
should belong to different groups to avoid the pilot contamination, which is illustrated in Fig. 8.49.
In this example, we assumed that three time slots are used for uplink transmission, three time slots for pilots, one time slot
to process pilots and obtain CSI, and four time lots for downlink transmission. Various OFDM- and MIMO-based channel
estimation techniques have already been described in respective section; we move our attention to lower-complexity detection
schemes suitable for massive MIMO applications.
518 8 Diversity and MIMO Techniques

Fig. 8.49 Illustrating the time- Processing

Uplink Pilots Downlink
domain shifted pilot technique for
massive MIMO to avoid the pilot
contamination problem

8.7.2 Massive MIMO Detection Schemes

Let us consider the channel model given by Eq. (8.267) in which the noise and interference vectors get combined into a z
vector:

y ¼ Hx þ z, ð8:269Þ

which is similar to that given by Eq. (8.62). The zero-forcing linear detector performs the Moore-Penrose pseudo-inverse of
the channel matrix H+ and provides the estimate of transmitted vector x by Eq. (8.142), that is:

xZF = Hþ y ¼ x þ H þ z,
b ð8:270Þ

which exhibits noise enhancement problem and operates well in interference-limited regime. On the other hand, the MMSE
detector minimizes the MSE between H{y and transmitted vector x; it takes the noise into account but has high complexity for
massive MIMO applications since:

 1
N
xMMSE = A{MMSE y, AMMSE ¼ HH { þ
b I H{ , ð8:271Þ
SNR

where N is the number of active users. The main issue arises from the matrix inversion above, which is of huge complexity for
massive MIMO. To solve this problem, the approximate matrix inversion was proposed in [79].
When the number of users is small compared to the number of antenna elements in the BS (N  M ), the small-scale fading
effects dominate, and for sufficiently high SNR, our problem reduces to the matrix inversion of the Gram matrix G = H{H. In
the Neumann series (NS) approach, we decompose the Gram matrix in terms of diagonal matrix D and off-diagonal matrix O,
that is, G ¼ D + O. We then perform the following NS expansion [80]:

X
1  k
G1 ¼ D1 O D1 : ð8:272Þ
k¼0

The diagonal matrix inversion is trivial, and for sufficiently large k, we have that –D1O ! 0 as k ! 1.
In the Gauss-Seidel (GS) method, we further decompose the off-diagonal matrix into strictly upper U and strictly lower L
triangular matrices, that is, O¼U + L, and detect the transmitted vector in an iterative fashion [81]:
 
xðkÞ ¼ ðD þ LÞ1 H { y  Uxðk1Þ ; k ¼ 1, 2, ⋯ ð8:273Þ

In the successive over-relaxation (SOR) method, generalize the GS method as follows [82]:
8.7 Massive MIMO 519

 1 h {   i
xðkÞ ¼ r 1 D þ L H y þ r 1  1 D  Uxðk1Þ ; k ¼ 1, 2, ⋯ ð8:274Þ

where the relaxation parameter r, affecting the convergence rate, is selected as r2(0,2). Clearly, for r ¼ 1 the SOR method
reduces down to the GS method.
In the Jacobi method, we apply the following iterative formula to detect the transmitted signal [83]:
h i
xðkÞ ¼ D1 H { y þ ðD  AÞxðk1Þ ; k ¼ 0, 1, ⋯ ð8:275Þ

which is applicable when I-D1A ! 0 as k ! 1. We use matrix A to denote the matrix GΛ + N0I, where Λ is the signal power
xð0Þ ¼ D1 H { y.
matrix and N0 is the PSD of the noise. The starting point is b
In the Richardson method, we apply the following iterative relation [84]:
 
xðkÞ ¼ xðkÞ þ r y  HxðkÞ ; k ¼ 1, 2, ⋯ ð8:276Þ

where the relaxation parameter is chosen as r2(0,2/λmax) with λmax being the largest eigenvalue of the symmetric positive
definite matrix H. x(0) is all zeros vector.
In the Newton iteration method, the matrix inversion of the Gram matrix G is obtained by the following recursive relation
[85]:
 
X 1 1 1
k ¼ X k1 2I  GX k1 , ð8:277Þ
 
which converges quadratically providing that the following is valid I  GX 1 0
 < 1.
A large number of antenna elements in massive MIMO increase the number of degrees of freedom, so many elements in the
channel matrix are either zero or negligible. This gives an idea to transform the channel matrix H into a sparse matrix Hs with
the help of two unitary matrices U and V as follows [86, 87]:

H ¼ UH s V: ð8:278Þ

Now by exploiting the properties of this sparse matrix, we can estimate the channel matrix by employing the compressive
sensing (CS) concepts [87]. In CS we reconstruct the sparse signals s from the compressed measurements y ¼ Φs, where
y2C m , through a convex programming, where Φ is the sensing matrix. If the original signal x2C n , m < n, is not sparse, we
need to sparsify it by s ¼ Ψx, where Ψ is the sparsifying matrix. Therefore, y and x are connected by y ¼ ΦΨx = Ax with
A = ΦΨ being the measurement matrix. The solution for x will be x ¼ (ATA)1ATy. Unfortunately, the system of equations is
underdetermined, that is, the number of equations is smaller than the number of unknowns. To reconstruct x we can apply the
following combinatorial optimization problem:

b
x ¼ arg minm kxk0 subject to Ax ¼ y, ð8:279Þ
x2C

where ||||0 denotes the l0-norm (the number of nonzero components). Unfortunately, this problem in general is NP-hard. A
very popular method to solve this problem is the basis pursuit or l1-minimization:

b
x ¼ arg minm kxk1 subject to Ax ¼ y, ð8:280Þ
x2C

Because the l1-norm is convex, the efficient methods from convex optimization are applicable. For instance, the greedy
methods such as the orthogonal matching pursuit (OMP) can be employed [88]. This optimization is applicable even in the
presence of noise, that is, when y ¼ Ax + z, which is similar to the massive MIMO channel model. As an illustration, a
multipath matching pursuit (MMP) detector, employing the sphere decoding (SD) detector, is used in ref. [89]. The SD search
is performed by [89, 90]:
520 8 Diversity and MIMO Techniques

Fig. 8.50 Illustrating the post-

detection sparse error recovery
y Conventional x̂ x e x = xˆ eˆ
(PDSR) scheme detection
scheme

Estimated ê
H

Sparse error
y' = H x xˆ z recovery
e

n o
b
x ¼ arg minm ky  Rxk2 < d 2 , ð8:281Þ
x2C

where R is the upper triangular matrix in QR factorization of the channel matrix, that is, H ¼ QR. The distance d is properly
chosen to get a near-ML solution. In [91] the CS is used to improve the performance of the conventional receiver, with the
corresponding post-detection sparse error recovery (PDSR) scheme provided in Fig. 8.50. This scheme assumes that the CSI is
perfect. We first perform the conventional massive MIMO detection to get b x. Since the error vector e = x  b
x is sparse, we can
apply the CS techniques, such as MMP or OMP, to estimate it. The estimate error is added to the detected vector to get the
improved detected vector ex=b x þ e.
Various multiuser detection techniques described in Chap. 10 (see Sect. 10.3) are also applicable in massive MIMO
applications. The decision-feedback and BLAST receivers described in Sect. 8.2.7.3 are applicable as well. Spatial interfer-
ence cancelation scheme described in Sect. 8.2.7.4 can also be used in massive MIMO. The successive interference
cancelation (SIC) Schemes [71, 92] belong to this category.
The performance of linear detectors can be improved by employing the lattice reduction-based (LRB) algorithms [71, 93,
94]. The complex-valued lattice of rank m in n-dimensional complex space C n can be defined as follows:

L ¼ fyjy ¼ Hxg, x ¼ ½x1 ⋯xn T , ð8:282Þ

where H is the basis of the lattice (with columns representing the basis vectors). In other words, L is the set of received signal
points not perturbed by the noise. The column vectors in the channel matrix are non-orthogonal. The goal of LRB algorithms
is to find an improved basis and then perform detection in this improved basis. By multiplying the channel matrix with a
unimodular matrix T (a square integer matrix whose determinant is either +1 or 1), we obtain the improved basis, that is,
e = TH. The decision regions in modified basis are the parallelograms around the lattice points, while the Voronoi regions in
H
original space could be of highly irregular shape. We can modify the MIMO channel model given by Eq. (8.269) as follows:

y ¼ |{z}
HT |ffl{zffl} e x þ z,
T 1 x þ z ¼ He ð8:283Þ
e
H ex

and then perform the detection of e

x. As an illustration, the LRB zero-forcing detector will be:

b e þy ¼ e
xZF = H e þ z:
xþH ð8:284Þ

Various LRB algorithms are reviewed in [93].

Various machine learning algorithms (MLAs) can be also used in massive MIMO communications, in particular to
simplify the optimizations. As an illustration, MLAs have been already studied for use in channel estimation, detection,
MIMO beamforming, load balancing, and optimization of available spectrum [71, 72, 95–98].
8.9 Problems 521

8.8 Concluding Remarks

The purpose of this chapter has been to describe various diversity and MIMO techniques capable of improving SISO system
performance as well as the system throughput. The various diversity and MIMO techniques that have been described can be
used not only in wireless MIMO communications but also in free-space optical and fiber-optic communications with coherent
optical detection.
Various diversity schemes including polarization diversity, spatial diversity, and frequency diversity schemes are
summarized in Sect. 8.1.1. The following receiver diversity schemes have been described in Sect. 8.1.2: selection combining,
threshold combining, maximum-ratio combining, and equal-gain combining schemes. In Sect. 8.1.3, various transmit
diversity schemes have been described, depending on the availability of channel state information on the transmitter side.
Most of the chapter has been devoted to wireless and optical MIMO techniques. After description of various wireless and
optical MIMO models in Sect. 8.2.1, we describe the parallel decomposition of MIMO channels in Sect. 8.2.2, followed by
space-time coding (STC) principles in Sect. 8.2.3. The maximum likelihood (ML) decoding for STC is described together
with various design criteria. The following relevant classes of STC have been described in Sect. 8.2.4: Alamouti code,
orthogonal designs, linear space-time block codes, and space-time trellis codes. The corresponding STC decoding algorithms
have been described as well in the same section.
After that the attention has moved to SDM principles in Sect. 8.2.5, where various BLAST encoding architectures (Sect.
8.2.5.1) as well as multi-group space-time coded modulation (Sect. 8.2.5.2) have been introduced. The following classes of
linear and feedback MIMO receiver for uncoded signals have been subsequently described in Sect. 8.2.6: zero-forcing, linear
minimum MSE, and decision-feedback receivers. The next topic in the chapter has been related to the suboptimum MIMO
receivers for coded signals (see Sect. 8.2.7). Within this topic, the linear zero-forcing (ZF) and linear MMSE receivers
(interfaces) have been described in the context of STC. Within decision-feedback and BLAST receivers, various H/V-BLAST
architectures, including ZF and MMSE ones, have been introduced. The topic on suboptimum receivers has been concluded
with diagonal (D)-BLAST and iterative receiver interfaces.
The section on iterative MIMO receivers (Sect. 8.3) starts with a brief introduction of concept of factor graphs (Sect. 8.3.1),
followed by the description of factor graphs for MIMO channel and channels with memory (Sect. 8.3.2). And then the sum-
product algorithm (SPA) operating on factor graphs has been described in Sect. 8.3.3, followed by description of SPA for
channels with memory (Sect. 8.3.4). The following iterative MIMO receivers for uncoded signals have been described in Sect.
8.3.5: ZF, MMSE, ZF H/V-BLAST, and LMMSE H/V-BLAST receivers. After a brief description of factor graphs for linear
block and trellis codes in Sect. 8.3.6, several iterative MIMO receivers for space-time coded signals have been described in
Sect. 8.3.7.
The section on broadband MIMO (Sect. 8.4) has been devoted to the description on how frequency selectivity can be used
as an additional degree of freedom, followed by a description of MIMO-OFDM in Sect. 8.4.1 and space-frequency block-
coding principles in Sect. 8.4.2.
The focus has been then moved to MIMO channel capacity calculations in Sect. 8.5 for various MIMO channel models
including deterministic (Sect. 8.5.1), ergodic and non-ergodic random channels, as well as correlated channel models,
described in Sect. 8.5.2. The concepts of ergodic and outage channel capacity have been introduced. The section on
MIMO channel capacity ends with MIMO-OFDM channel capacity description (Sect. 8.5.2.4).
In MIMO channel estimation section (Sect. 8.6), the following MIMO channel estimation techniques have been described:
ML MIMO channel estimation (Sect. 8.6.1), least squares (LS) MIMO channel estimation (Sect. 8.6.2), and linear minimum
MSE MIMO channel estimation (Sect. 8.6.3). The massive MIMO concepts have been described in Sect. 8.7.
The set of problems is provided in incoming section to help reader gain deeper understanding of material described in this
chapter.

8.9 Problems

8.1. A four-dimensional (4D) signal constellation is described by the following set of points {a, a, a, a} (a > 0).
Different signal constellation points are used with the same probability. Determine the average symbol energy. Choose
the appropriate basis functions {Φ1(t), Φ2(t), Φ3(t), Φ4(t)} and express the signal constellation points in terms of basis
functions. For AWGN channel, determine the average symbol error probability as a function of the average energy (Eav).
Evaluate now the average symbol error probability in the presence of Rayleigh fading. Finally, evaluate the symbol error
probability in the presence of Rayleigh fading when the maximum-ratio combining diversity is used.
522 8 Diversity and MIMO Techniques

8.2. In this problem, we study the receiver diversity based on equal gain combining (EGC). Let us assume that either BPSK or
QPSK is used to transmit messages over a generalized fading wireless channel. According to the generalized fading
channel model, the signal envelope at the i-th receiver branch (i ¼ 1,2,. . .,L ) follows α-μ distribution with probability
density function (PDF) given by:
 
αi μi μi r i αi μi 1 r i αi
f Ri ðr i Þ ¼ exp μ i αi ,
br i αi μi Γðμi Þ br i

where αi > 0 is the parameter of nonlinearity, Γ(.) is the Gamma function, and μi > 0 is the inverse of the normalized
variance
p of r i αi , namely, μi ¼ E2 fr i αi g=Varfr i αi g, where E{.} is the expectation operator and br i is a αi-root mean value
ffiffiffiffiffiffiffiffiffiffiffiffiffiffi
br i ¼ αi Efr i αi g. If the phase estimation is done from unmodulated carrier by using phase-locked loop (PLL) and if only
the Gaussian noise is present in the phase-locked loop circuit, then the PDF of this phase error is:
exp ðςi cos ðφi ÞÞ
pφi ðφi Þ ¼ 2π
1
I0 ðςi Þ ,  π < φi  π, where I0(x) is the modified Bessel function of the first kind and zero order for
the argument x, ζ i is the SNR in the PLL circuit at the i-th receiver branch.
Write the expressions for the average bit error probability of uncoded BPSK and QPSK signal detections in the presence
of phase error and α-μ fading assuming that EGC receiver is used.
8.3. Based on parallel decomposition of MIMO optical channel, explain how MIMO-OFDM can be used in multi-Tb optical
transport.
8.4. In the section on space-time coding for MIMO communications, we described the Alamouti scheme employing two
transmitters and two receivers. Describe how this Alamouti scheme can be generalized to a number of receivers larger
than two. Provide the corresponding combiner rule. Prove it by mathematical induction.
8.5. Discuss the advantages and disadvantages of Alamouti-type polarization-time (PT) coding scheme against that of
conventional PDM-OFDM scheme.
8.6. Provide the star-8-QAM trellis space-time coding scheme with two transmitters and diversity of 2MRx, where MRx is the
number of transmitters.
8.7. The space-time code for a number of transmit antennas MTx ¼ 4, number of receive antennas MRx ¼ 1, and number of
channel’s uses N ¼ 4 is described by the following signaling matrix:
2 3
x1 x 2 x 3 0
6x x 1 x 3 7
6 2 0 7
X¼6 7
4 x3 0 x 1 x 2 5
0 x 3 x 2 x 1

(a) Can this space-time block code be used to design the receiver that will result in spatial interference cancellation? If so
describe the corresponding receiver configuration. What is the code rate of this space-time code?
(b) Represent the signaling matrix X in the same fashion as for linear space-time (LST) codes. If the channel matrix is
given by H ¼ [h1 h2 h3 h4] using detection approach for LST, determine the equations for estimated symbols bxi .
8.8. Figure 8.P8 depicts a 2  2 polarization-time coding MIMO scheme, which employs dual-polarization transmitter and
receiver antennas. Due to multipath effect, the initial orthogonality of polarization states is no longer preserved on a
receiver side, and we can use the channel coefficients as shown in Fig. 8.P8 to describe this depolarization effect. Show
that Alamouti 2  2 scheme can be used to deal with depolarization effect. Determine the array, diversity, and
multiplexing gains of this scheme. How would you determine the channel capacity of this scheme? Consider now a
MIMO scheme employing two dual-polarization Tx antennas and two dual-polarization Rx antennas. How would you
approach this system to deal simultaneously with depolarization and multipath effects? How would you determine the
channel capacity of this scheme?
8.9 Problems 523

Fig. 8.P8 A 2  2 polarization- y-pol.

y-polarizaon
time coding MIMO scheme hyy Rx antenna
Tx antenna

hxy hyx

Informaon Polarizaon-me Receiver

symbols Block code hxx
x-polarizaon x-pol.
Tx antenna Rx antenna

8.9. This problem is related to the derivation of AMMSE matrix given by Eq. (8.144). Let us define the MSE by:
D E D h iE
MSE ðAÞ ¼ kAy  xk2 ¼ Tr ððAH  I M Tx Þx þ AzÞððAH  I M Tx Þx þ AzÞ{ :

By assuming that components of x are i.i.d. zero-mean with the second moment corresponding to the average energy,
determine the optimum (in MMSE sense) linear transformation A. Derive the corresponding PEP. Discuss its
asymptotic behavior for (a) finite MT, MR- > 1 and (b) MT- > 1, MR- > 1, while MT/MR- > a.
8.10. In this problem we deal with the derivation of matrices A and B in ZF V-BLAST. The MSE in the absence of noise is
  2   2 
e   b  X
given by MSE ðA, BÞ ¼ Y  X ¼ AHX  BX  : By using the high SNR approximation and QR
factorization of channel matrix, determine the matrices A and B so that the MSE is minimized. Once you determine
these matrices, demonstrate that mode coupling is indeed canceled out. Fully describe the detection procedure.
8.11. Here we deal with the derivation of matrices A and B in MMSE V-BLAST. The MSE in the presence of noise is
 2  D E D E
e 
given by: MSE ðA, BÞ ¼ Y  X ¼ kAY  BX  Xk2 ¼ kðAH  L  IM Tx ÞX þ AZk2 : By using the

Cholesky factorization of H { H þ SNR1 IM Tx ¼ S{ S, determine the matrices A and B so that the MSE is minimized.
What is the remaining MSE error? Once you determine these matrices, demonstrate that mode coupling is indeed
canceled out.
8.12. Let us study a MIMO system with the following channel matrix:”
2 3
0:1 0:2 0:3
6 7
H ¼ 4 0:3 0:4 0:5 5:
0:7 0:6 0:5

(a) Write the channel matrix in terms of its singular value decomposition (SVD) H ¼ UΣV{.
(b) Based on SVD develop an equivalent MIMO systems composed of RH ¼ rank(H) independent channels. After
determining RH, devise pre-coding and receiver shaping matrices suitable to transform the original system into an
equivalent system.
(c) Find optimum power allocation Pi (i ¼ 1,2,3) across the RH channels found in (b), and find the corresponding total
capacity of the equivalent system assuming SNR of 22 dB and a system bandwidth of 1 MHz.
(d) When the channel is not known to the transmitter, the equal power is allocated to each antenna. Compare the
corresponding capacity part (c) with that for equal power allocation.
8.13. Consider the channel with the following channel matrix:
2 3
0:1 0:1 0:4
6 7
H ¼ 4 0:2 0:2 0:5 5:
0:6 0:3 0:7

Under assumption that ρ ¼ 12 dB, determine the output SNR when beamforming is used over the channel with equal
weights on each transmit antenna and optimal weights at the receiver side. Compare the SNR with the corresponding
SNR under beamforming when optimum weights are applied at both sides.
524 8 Diversity and MIMO Techniques

8.14. Based on the SVD properties show that, for a MIMO channel that is known to the transmitter and receiver side (full-
CSI), corresponding capacity expression:
  
C=BW ¼ max log 2 det IM Rx þ HRx H { ,
Rx :TrðRx Þ¼ρ

simplifies to:
X  
C=BW ¼ P
max log 2 1 þ σ 2i ρi ,
ρi : ρ ρ
i i i

for singular values {σi} and SNR values ρi.

8.15. Let us consider 16  16, 64  64, and 256  256 massive MIMO systems for which entries of the channel matrix are
generated using the simulator developed in Problem 13 of Chap. 2 for Doppler frequency of 20 Hz. Estimate the channel
by using both LS and LMMSE channel estimators. Evaluate BER performance of Neumann series, Gauss-Seidel,
successive over-relaxation, Jacobi, and Richardson detection methods for different SNRs. Which detection scheme
performs the best? Discuss the complexity of this scheme.
8.16. Repeat Problem 15 but now using the lattice reduction-based approach applied to ZF and MMSE detectors. Evaluate the
BER performance against different schemes discussed in Problem 16.
8.17. Here we study the efficiency of V-BLAST scheme in CO-OFDM systems. The V-BLAST scheme is described in Sect.
8.2.7.3. Design a CO-OFDM system, capable of 100 Gb(s) serial optical transmission over 10,000 km of SMF. Perform
Monte Carlo simulations in linear regime, by taking the following effects into account: chromatic dispersion, the first-
order PMD, and laser linewidth of 100 kHz (for both transmitter and local lasers). Use the typical fiber parameters given
in previous chapters. To compensate for PDM and chromatic dispersion, use the V-BLAST approach. Compare this
scheme with an equivalent PDM scheme, based on LS channel estimation, of the same aggregate data rate. Compare
and discuss the results. What are the advantages and disadvantages of V-BLAST detection scheme with respect
to PDM?
8.18. Repeat Problem 17, but now in the nonlinear regime, by solving the generalized nonlinear Schrödinger equation
(GNLSE) for both polarizations. Study the BER performance as the function of total transmission distance.
8.19. In this problem we are interested in improving the BER performance and nonlinearity tolerance of V-BLAST-based and
conventional PDM CO-OFDM systems by using MMSE channel estimation. Plot BER vs. total transmission distance
for several cases: (i) V-BLAST-LS estimation, (ii) conventional PDM-LS estimation, and (iii) V-BLAST-MMSE
estimation, conventional PDM-MMSE estimation. For MMSE estimation use the first- and second-order interpolation.
Discuss results.
8.20. In this problem we are concerned with 1 Tb optical transport over 1500 km of few-mode fiber (FMF) by using mode-
multiplexed PDM CO-OFDM. For modeling of FMF, use the model described in Sect. 8.2.1 of this chapter (see
Fig. 8.12). Design a CO-OFDM system with aggregate data rate exciding 1 Tb by employing two spatial and two
polarization modes. Use five orthogonal OFDM bands for simulations, with 200 Gb in aggregate data rate per band.
Perform Monte Carlo simulations for the following cases of interest: (i) V-BLAST-LS estimation, (ii) conventional
PDM-LS estimation, and (iii) V-BLAST-MMSE estimation, conventional PDM-MMSE estimation. For MMSE
estimation use the first- and second-order interpolation. Discuss results.
8.21. For FSO channels for IM/DD, the repetition MIMO outperforms any space-time coding scheme. Explain this
counterintuitive claim.
8.22. Let us consider 3  3 FSO MIMO system, in which three independent data streams are transmitted by sufficiently
separated FSO transmitters. Assume that each FSO transmitter illuminates both FSO receivers. Assume that coherent
detection is used and that the simulation parameters are the same as in Problem 1 of Chap. 2.
(a) For 3  3 FSO MIMO system with either zero-forcing (ZF) or minimum MSE (MMSE) receiver, for the refractive
structure parameter of Cn2 ¼ 1015, perform Monte Carlo simulations for M ¼ 2, 4, and 16. Plot BERs versus
optical SNR per information bit. Determine the coding gains at BER of 105. Assume that channel matrix H is
available on the receiver side.
(b) Repeat (a) but now for different values of the refractive structure parameter. At BER of 106, for M ¼ 2, 4, 16,
determine improvements when MMSE receiver is used compared to the ZF receiver case.
References 525

References
1. A. Goldsmith, Wireless Communications (Cambridge University Press, Cambridge, 2005)
2. E. Biglieri, Coding for Wireless Channels (Springer, New York, 2005)
3. D. Tse, P. Viswanath, Fundamentals of Wireless Communication (Cambridge University Press, Cambridge, 2005)
4. L. Kazovsky, S. Benedetto, A. Willner, Optical Fiber Communication Systems (Artech House, Boston, 1996)
5. I. Djordjevic, W. Ryan, B. Vasic, Coding for Optical Channels (Springer, New York, Mar. 2010)
6. I.B. Djordjevic, M. Cvijetic, C. Lin, Multidimensional signaling and coding enabling multi-Tb/s optical transport and networking. IEEE Sig.
Proc. Mag. 31(2), 104–117 (Mar. 2014)
7. M. Cvijetic, I.B. Djordjevic, Advanced Optical Communications and Networks (Artech House, Norwood, Jan. 2013)
8. J.W. Craig, A new, simple and exact result for calculating the probability of error for two-dimensional signal constellations, in Proceedings of
the IEEE Military Communications Conference 1991 (MILCOM’'91), McLean, VA, vol 2, pp. 571–575, 4–7 Nov. 1991
9. G.L. Stüber, Principles of Mobile Communication, 3rd edn. (Springer, New York, 2012)
10. S.M. Alamouti, A simple transmit diversity technique for wireless communications. IEEE J. Sel. Areas Commun. 16(8), 1451–1458 (1998)
11. I.B. Djordjevic, L. Xu, T. Wang, PMD compensation in coded-modulation schemes with coherent detection using Alamouti-type polarization-
time coding. Opt. Express 16(18), 14163–14172 (5 Sep. 2008)
12. V. Tarokh, N. Seshadri, A.R. Calderbank, Space–time codes for high data rate wireless communication: performance criterion and code
construction. IEEE Trans. Inf. Theory 44(2), 744–765 (Mar. 1998)
13. H. Bölcskei, D. Gesbert, A.J. Paulraj, On the capacity of wireless systems employing OFDM-based spatial multiplexing. IEEE Trans. Commun.
50, 225–234 (Feb. 2002)
14. A.R. Shah, R.C.J. Hsu, A.H. Sayed, B. Jalali, Coherent optical MIMO (COMIMO). J. Lightwave Technol. 23(8), 2410–2419 (Aug. 2005)
15. R.C.J. Hsu, A. Tarighat, A. Shah, A.H. Sayed, B. Jalali, Capacity enhancement in coherent optical MIMO (COMIMO) multimode fiber
links. J. Lightwave Technol. 23(8), 2410–2419 (Aug. 2005)
16. A. Tarighat, R.C.J. Hsu, A.H. Sayed, B. Jalali, Fundamentals and challenges of optical multiple-input multiple output multimode fiber links.
IEEE Commuun. Mag. 45, 57–63 (May 2007)
17. N.W. Bikhazi, M.A. Jensen, A.L. Anderson, MIMO signaling over the MMF optical broadcast channel with square-law detection. IEEE Trans.
Commun. 57(3), 614–617 (Mar. 2009)
18. A. Agmon, M. Nazarathy, Broadcast MIMO over multimode optical interconnects by modal beamforming. Opt. Express 15(20), 13123–13128
(26 Sept. 2007)
19. E. Alon, V. Stojanovic, J.M. Kahn, S.P. Boyd, M. Horowitz, Equalization of modal dispersion in multimode fibers using spatial light
modulators, in Proceedings of the IEEE Global Telecommunication Conference 2004, Dallas, TX, Nov. 29–Dec. 3, 2004
20. R.A. Panicker, J.M. Kahn, S.P. Boyd, Compensation of multimode fiber dispersion using adaptive optics via convex optimization. J. Lightwave
Technol. 26(10), 1295–1303 (15 May 2008)
21. S. Fan, J.M. Kahn, Principal modes in multi-mode waveguides. Opt. Lett. 30(2), 135–137 (15 Jan. 2005)
22. H. Kogelnik, R.M. Jopson, L.E. Nelson, Polarization-mode dispersion, in Optical Fiber Telecommunications IVB: Systems and Impairments,
ed. by I. Kaminow, T. Li, (Academic, San Diego, 2002)
23. K.-P. Ho, J.M. Kahn, Statistics of group delays in multimode fiber with strong mode coupling. J. Lightwave Technol. 29(21), 3119–3128 (2011)
24. Zou, D., Lin, C., Djordjevic, I.B., LDPC-coded mode-multiplexed CO-OFDM over 1000 km of few-mode fiber, in Proceedings of CLEO 2012,
Paper no. CF3I.3, San Jose, CA, 6–11 May 2012
25. I.B. Djordjevic, S. Denic, J. Anguita, B. Vasic, M.A. Neifeld, LDPC-coded MIMO optical communication over the atmospheric turbulence
channel. IEEE/OSA J. Lightwave Technol. 26(5), 478–487 (1 Mar. 2008)
26. E. Biglieri, R. Calderbank, A. Constantinides, A. Goldsmith, A. Paulraj, H.V. Poor, MIMO Wireless Communications (Cambridge University
Press, Cambridge, 2007)
27. J.G. Proakis, Digital Communications, 5th edn. (McGraw Hill, New York, 2007)
28. I.B. Djordjevic, Spatial-domain-based hybrid multidimensional coded-modulation schemes enabling multi-Tb/s optical transport. IEEE/OSA
J. Lightwave Technol. 30(14), 2315–2328 (15 July 2012)
29. G.J. Foschini, Layered space-time architecture for wireless communication in a fading environment when using multi-element antennas. Bell
Labs Tech. J. 1, 41–59 (1996)
30. P.W. Wolniansky, G.J. Foschini, G.D. Golden, R.A. Valenzuela, V-BLAST: an architecture for realizing very high data rates over the rich-
scattering wireless channel, in Proceedings of 1998 URSI International Symposium on Signals, Systems, and Electronics (ISSSE 98), Pisa,
pp. 295–300, 29 Sept.–2 Oct. 1998
31. X. Li, H. Huang, G.J. Foschini, R.A. Valenzuela, Effects of iterative detection and decoding on the performance of BLAST, in Proceedings of
the IEEE Global Telecommunications Conference 2000. (GLOBECOM '00), vol 2, San Francisco, CA, pp. 1061–1066, 2000
32. G.J. Foschini, D. Chizhik, M.J. Gans, C. Papadias, R.A. Valenzuela, Analysis and performance of some basic space-time architectures. IEEE
J. Sel. Areas Commun. 21(3), 303–320 (Apr. 2003)
33. S. Loyka, F. Gagnon, Performance analysis of the V-BLAST algorithm: An analytical approach. IEEE Trans. Wirel. Commun. 3(4), 1326–1337
(July 2004)
34. V. Tarokh, A. Naguib, N. Seshadri, A.R. Calderbank, Combined array processing and space-time coding. IEEE Trans. Inf. Theory 45(4),
1121–1128 (May 1999)
35. R.A. Horn, C.R. Johnson, Matrix Analysis (Cambridge University Press, New York, 1988)
36. A.L. Moustakas, H.U. Baranger, L. Balents, A.M. Sengupta, S.H. Simon, Communication through a diffusive medium: coherence and capacity.
Science 287, 287–290 (14 Jan. 2000)
37. S. Da-Shan, G.J. Foschini, M.J. Gans, J.M. Kahn, Fading correlation and its effect on the capacity of multielement antenna systems. IEEE Trans.
Commun. 48(3), 502–513 (Mar. 2000)
526 8 Diversity and MIMO Techniques

38. A. Wittneben, Basestation modulation diversity for digital simulcast, 41st IEEE Vehicular Technology Conference, 1991. Gateway to the Future
Technology in Motion, pp. 848–853, St. Louis, MO, 19–22 May 1991
39. N. Seshadri, J.H. Winters, Two signaling schemes for improving the error performance of frequency-division-duplex (FDD) transmission
systems using transmitter antenna diversity, 43rd IEEE Vehicular Technology Conference, 1993, pp. 508–511, Secaucus, NJ, 18–20 May 1993
40. A. Edelman, Eigen Values and Condition Numbers of Random Matrices. PhD Dissertation, MIT, Cambridge, May 1989
41. A.T. James, Distributions of matrix variates and latent roots derived from normal samples. Ann. Math. Stat. 35(2), 475–501 (1964)
42. H. Wang, X.-G. Xia, Upper bounds of rates of complex orthogonal space-time block codes. IEEE Trans. Inf. Theory 49(10), 2788–2796 (Oct.
2003)
43. X. Ma, G.B. Giannakis, Full-diversity full-rate complex-field space-time coding. IEEE Trans. Signal Process. 51(11), 2917–2930 (Nov. 2003)
44. I.B. Djordjevic, L. Xu, T. Wang, PMD compensation in multilevel coded-modulation schemes with coherent detection using BLAST algorithm
and iterative polarization cancellation. Opt. Express 16(19), 14845–14852 (15 Sept. 2008)
45. J.R. Hampton, Introduction to MIMO Communications (Cambridge University Press, Cambridge, 2014)
46. N. Wiberg, H.A. Loeliger, R. Kotter, Codes and iterative decoding on general graphs, in Proceedings of 1995 IEEE International Symposium on
Information Theory, p. 468, Whistler, BC, 1995
47. F.R. Kschischang, B.J. Frey, H.A. Loeliger, Factor graphs and the sum-product algorithm. IEEE Trans. Inf. Theory 47(2), 498–519 (Feb. 2001)
48. H.A. Loeliger, An introduction to factor graphs. IEEE Signal Process. Mag. 21(1), 28–41 (Jan. 2004)
49. B.J. Frey, F.R. Kschischang, H.-A. Loeliger, N. Wiberg, Factor graphs and algorithms, in Proceedings of 35th Allerton Conference on
Communications, Control, and Computing, Allerton House, Monticello, IL, pp. 666–680, 29 Sept.–1 Oct. 1997
50. S. Lin, D.J. Costello, Error Control Coding: Fundamentals and Applications, 2nd edn. (Pearson Prentice Hall, Upper Saddle River, 2004)
51. W.E. Ryan, S. Lin, Channel Codes: Classical and Modern (Cambridge University Press, Cambridge, 2009)
52. E. Biglieri, G. Taricco, A. Tulino, Performance of space-time codes for a large number of antennas. IEEE Trans. Inf. Theory 48(7), 1794–1803
(July 2002)
53. E. Biglieri, G. Taricco, A. Tulino, Decoding space-time codes with BLAST architectures. IEEE Trans. Signal Process. 50(10), 2547–2552 (Oct.
2002)
54. A. Pulraj, R. Nabar, D. Gore, Introduction to Space-Time Wireless Communications (Cambridge University Press, Cambridge, 2003)
55. I.B. Djordjevic, On advanced FEC and coded modulation for ultra-high-speed optical transmission, IEEE Commun. Surv. Tutorials 99, 1–31,
1 Mar. 2016. https://doi.org/10.1109/COMST.2016.2536726
56. S. Ten Brink, Convergence behavior of iteratively decoded parallel concatenated codes. IEEE Trans. Commun. 49(10), 1727–1737 (2001)
57. T.S. Rappaport, Wireless Communications: Principles and Practice, 2nd edn. (Prentice Hall, Hoboken, 2002)
58. Z. Zheng, T.M. Duman, E.M. Kurtas, Achievable information rates and coding for MIMO systems over ISI channels and frequency-selective
fading channels. IEEE Trans. Commun. 52(10), 1698–1710 (Oct. 2004)
59. R. Schober, W.H. Gerstacker, L.H.J. Lampe, Performance analysis and design of STBCs for frequency-selective fading channels. IEEE Trans.
Wirel. Commun. 3(3), 734–744 (May 2004)
60. W. Zhang, X.G. Xia, P.C. Ching, High-rate full-diversity space–time–frequency codes for broadband MIMO block-fading channels. IEEE
Trans. Commun. 55(1), 25–34 (Jan. 2007)
61. W. Siheh, I.B. Djordjevic, OFDM for Optical Communications (Elsevier/Academic Press, Amsterdam/Boston, 2009)
62. E. Telatar, Capacity of multi-antenna Gaussian channels. Eur. Trans. Telecommun. 10, 585–595 (1999)
63. T.M. Cover, J.A. Thomas, Elements of Information Theory (Wiley, New York, 1991)
64. H. Shin, J.H. Lee, Capacity of multiple-antenna fading channels: Spatial fading correlation, double scattering, and keyhole. IEEE Trans. Inf.
Theory 49(10), 2636–2647 (Oct. 2003)
65. R.N. McDonough, A.D. Whalen, Detection of Signals in Noise, 2nd edn. (Academic Press, San Diego, 1995)
66. B. Hassibi, B.M. Hochwald, How much training is needed in multiple-antenna wireless links? IEEE Trans. Inf. Theory 49(4), 951–963 (Apr.
2003)
67. C. Lin, I.B. Djordjevic, D. Zou, Achievable information rates calculation for optical OFDM transmission over few-mode fiber long-haul
transmission systems. Opt. Express 23(13), 16846–16856 (2015)
68. E.G. Larsson, O. Edfors, F. Tufvesson, T.L. Marzetta, Massive MIMO for next generation wireless systems. IEEE Commun. Mag. 52(2),
186–195 (2014)
69. T. Marzetta, E.G. Larsson, H. Yang, H.Q. Ngo, Fundamentals of Massive MIMO (Cambridge University Press, Cambridge, 2016)
70. H.H. Yang, T.Q.S. Quek, Massive MIMO Meets Small Cell: Backhaul and Cooperation (Springer Nature, Cham, 2017)
71. M.A. Albreem, M. Juntti, S. Shahabuddin, Massive MIMO detection techniques: A survey. IEEE Commun. Surv. Tutorials 21(4), 3109–3132
(2019)
72. R. Chataut, R. Akl, Massive MIMO systems for 5G and beyond networks-overview, recent trends, challenges, and future research direction.
Sensors (Basel) 20(10), 2753 (2020)
73. S.A. Khwandah, J.P. Cosmas, P.I. Lazaridis, et al., Massive MIMO systems for 5G communications. Wireless Pers. Commun. 120, 2101–2115
(2021)
74. F. Fernandes, A. Ashikhmin, T.L. Marzetta, Inter-cell interference in noncooperative TDD large scale antenna systems. IEEE J. Sel. Areas
Commun. 31(2), 192–201 (2013)
75. J.L.L. Morales, S. Roy, Channel estimation using time-shifted pilot sequences in non-cooperative cellular TDD networks with large antenna
arrays. In Proceedings of 2013 Asilomar Conference on Signals, Systems and Computers, pp. 1258–1262, 2013
76. V. Saxena, G. Fodor, E. Karipidis, Mitigating pilot contamination by pilot reuse and power control schemes for massive MIMO systems. In
Proceedings of the IEEE VTC Spring, Glasgow, UK, 11–14 May 2015, pp. 1–6, 2015
77. R. Muller, L. Cottatellucci, M. Vehkapera, Blind pilot decontamination. IEEE J. Sel. Top. Signal Proc. 8, 773–786 (2014)
78. O. Elijah, C.Y. Leow, T.A. Rahman, S. Nunoo, S.Z. Iliya, A comprehensive survey of pilot contamination in massive MIMO-5G system. IEEE
Commun. Surv. Tutorials 18, 905–923 (2016)
79. M. Wu, B. Yin, A. Vosoughi, C. Studer, J.R. Cavallaro, C. Dick, Approximate matrix inversion for high-throughput data detection in the large-
scale MIMO uplink, in Proceedings of 2013 IEEE International Symposium on Circuits and Systems (ISCAS), pp. 2155–2158, 2013
References 527

80. B. Kang, J.-H. Yoon, J. Park, Low complexity massive MIMO detection architecture based on Neumann method, in Proceedings of 2015
International SoC Design Conference (ISOCC), pp. 293–294, 2015
81. Z. Wu, Y. Xue, X. You, C. Zhang, Hardware efficient detection for massive MIMO uplink with parallel Gauss-Seidel method, in Proceedings of
2017 22nd International Conference on Digital Signal Processing (DSP), pp. 1–5, 2017
82. X. Gao, L. Dai, Y. Hu, Z. Wang, Z. Wang, Matrix inversion-less signal detection using SOR method for uplink large-scale MIMO systems, in
Proceedings of 2014 IEEE Global Communications Conference, pp. 3291–3295, 2014
83. B.Y. Kong, I.-C. Park, Low-complexity symbol detection for massive MIMO uplink based on Jacobi method, in Proceedings of 2016 IEEE 27th
Annual International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC), pp. 1–5, 2016
84. X. Gao, L. Dai, Y. Ma, Z. Wang, Low-complexity near-optimal signal detection for uplink large-scale MIMO systems. IEEE Electron. Lett.
50(18), 1326–1328 (2014)
85. M. Ylinen, A. Burian, J. Takala, Direct versus iterative methods for fixed-point implementation of matrix inversion, in Proceedings of 2004
IEEE International Symposium on Circuits and Systems (IEEE Cat. No.04CH37512), pp. III–225, 2004
86. G. Wunder, H. Boche, T. Strohmer, P. Jung, Sparse signal processing concepts for efficient 5G system design. IEEE Access 3, 195–208 (2015)
87. M. Masood, L.H. Afify, T.Y. Al-Naffouri, Efficient coordinated recovery of sparse channels in massive MIMO. IEEE Trans. Signal Process.
63(1), 104–118 (2015)
88. S. Foucart, H. Rauhut, A Mathematical Introduction to Compressive Sensing (Springer, New York, 2013)
89. A.K. Sah, A.K. Chaturvedi, An MMP-based approach for detection in large MIMO systems using sphere decoding. IEEE Wirel. Commun. Lett.
6(2), 158–161 (2017)
90. M.A.M. Albreem, M.F.M. Salleh, Regularized lattice sphere decoding for block data transmission systems. Wirel. Pers. Commun. 82(3),
1833–1850 (2015)
91. J.W. Choi, B. Shim, New approach for massive MIMO detection using sparse error recovery, in Proceedings of 2014 IEEE Global
Communications Conference, pp. 3754–3759, 2014
92. Y. Jiang, M.K. Varanasi, J. Li, Performance analysis of ZF and MMSE equalizers for MIMO systems: an in-depth study of the high SNR regime.
IEEE Trans. Inf. Theory 57(4), 2008–2026 (2011)
93. D. Wübben, D. Seethaler, J. Jaldén, G. Matz, Lattice reduction. IEEE Signal Process. Mag. 28(3), 70–91 (2011)
94. S. Shahabuddin, J. Janhunen, Z. Khan, M. Juntti, A. Ghazi, A customized lattice reduction multiprocessor for MIMO detection, in Proceedings
of 2015 IEEE International Symposium on Circuits and Systems (ISCAS), pp. 2976–2979, 2015
95. J. Liu, R. Deng, S. Zhou, Z. Niu, Seeing the unobservable: CHANNEL learning for wireless communication networks, in Proceedings of 2015
IEEE Global Communications Conference (GLOBECOM), San Diego, CA, USA, pp. 1–6, 2015
96. M. Koller, C. Hellings, M. Knödlseder, T. Wiese, D. Neumann, W. Utschick, Machine learning for channel estimation from compressed
measurements, in Proceedings of 2018 15th International Symposium on Wireless Communication Systems (ISWCS), Lisbon, Portugal,
pp. 1–5, 2018
97. D. Neumann, T. Wiese, W. Utschick, Learning the MMSE channel estimator. IEEE Trans. Signal Process. 66, 2905–2917 (2018)
98. A. Zappone, L. Sanguinetti, M. Debbah, User association and load balancing for massive MIMO through deep learning, in Proceedings of 2018
Asilomar Conference on Signals, Systems and Computers, Pacific Grove, CA, USA, pp. 1262–1266 , 2018
 Advanced Coding and Coded Modulation
Techniques 9

Abstract
This chapter is devoted to advanced channel coding and coded modulation techniques. After linear and BCH codes
fundamentals, we provide the trellis description of linear block codes and describe the corresponding Viterbi decoding
algorithm. After describing the fundamentals of convolutional, RS, concatenated, and product codes, we describe coding
with interleaving as an efficient way to deal with burst of errors and fading effects. Significant space in the chapter is
devoted to codes on graphs, in particular turbo, turbo product, and LDPC codes are described together with corresponding
decoding algorithms. Regarding LDPC codes, both binary and nonbinary (NB) LDPC codes are introduced, and
corresponding decoding algorithms as well as their FPGA implementation are described. Additionally, LDPC code design
procedures are described, followed by rate adaptation. Rate-adaptive FPGA implementations of LDPC and generalized
LDPC codes are described as well. The next portion of the chapter is devoted to coded modulation (CM) and unequal error
protection (UEP). After providing the coded modulation fundamentals, we describe trellis-coded modulation (TCM),
multilevel coding and UEP, bit-interleaved coded modulation (BICM), turbo TCM, and various hybrid multidimensional
coded modulation schemes suitable for ultrahigh-speed optical transmission including multilevel nonbinary LDPC-coded
modulation. After coded modulation sections, the focus of the chapter is on multidimensional turbo equalization. The
following topics are described: nonlinear channels with memory, nonbinary MAP detection, sliding-window multidimen-
sional turbo equalization, simulation study of multidimensional turbo equalization, and several experimental
demonstrations including time-domain 4D-NB-LDPC-CM and quasi-single-mode transmission over transoceanic
distances. In section on optimized signal constellation design and optimized bit-to-symbol mappings-based coded
modulation, we describe multidimensional optimized signal constellation design (OSCD), EXIT chart analysis of OSCD
mapping rules, nonlinear OSCD-based coded modulation, and transoceanic multi-Tb/s transmission experiments enabled
by OSCD. Finally, in adaptive coding and adaptive coded modulation section, we describe adaptive coded modulation,
adaptive nonbinary LDPC-coded multidimensional modulation suitable for high-speed optical communications, and
adaptive hybrid free-space optical (FSO)-RF-coded modulation. For better understanding, the set of problems is provided
after concluding remarks.

9.1 Linear Block Codes Fundamentals

The linear block code (n,k) satisfies a linearity property, which means that a linear combination of arbitrary two codewords
results in another codeword. If we use the terminology of vector spaces, it can be defined as a subspace of a vector space over
finite (Galois) field, denoted as GF(q), with q being the prime power. Every space is described by its basis – a set of linearly
independent vectors. The number of vectors in the basis determines the dimension of the space. Therefore, for an (n,k) linear
block code, the dimension of the space is n, and the dimension of the code subspace is k.
Let m ¼ (m0 m1. . .mk  1) denote the k-bit message vector. Any codeword x ¼ (x0 x1 . . . xn  1) from the (n,k) linear block
code can be represented as a linear combination of k basis vectors gi (i ¼ 0,1,..,k  1) as follows:

# The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 529
I. B. Djordjevic, Advanced Optical and Wireless Communications Systems, https://doi.org/10.1007/978-3-030-98491-5_9
530 9 Advanced Coding and Coded Modulation Techniques

m0 m1…mk-1 b0 b1…bn-k-1
n-k-
k1

Message bits Parity bits

Fig. 9.1 Systematic codeword structure

2 3 2 3
g0 g0
6g 7 6g 7
6 1 7 6 1 7
x ¼ m0 g0 þ m1 g1 þ ⋯ þ mk1 gk1 ¼ m6 7 ¼ mG; G ¼ 6 7, ð9:1Þ
4... 5 4 ... 5
gk1 gk1

where G is the generator matrix (of dimensions k  n), in which every row represents basis vector from the coding subspace.
Therefore, in order to be encoded, the message vector m has to be multiplied with a generator matrix G to get the codeword,
namely, x ¼ mG.
The code may be transformed into a systematic form by elementary operations on rows in the generator matrix, i.e.,

Gs ¼ ½Ik jP, ð9:2Þ

where Ik is the unity matrix of dimensions k  k and P is the matrix of dimensions k  (n  k) with columns denoting the
positions of parity checks:
2 3
p00 p01 ... p0,nk1
6 p p11 ... p1,nk1 7
6 10 7
P¼6 7:
4 ... ... ... 5
pk1,0 pk1,1 ... pk1,nk1

The codeword of a systematic code is obtained as

x ¼ ½mjb ¼ m½Ik jP ¼ mG, G = ½Ik jP, ð9:3Þ

and has the structure as shown in Fig. 9.1. The message vector stays unaffected during systematic encoding, while the vector
of parity checks b is appended having the bits that are algebraically related to the message bits as follows:

bi ¼ p0i m0 þ p1i m1 þ . . . þ pk1,i mk1 , ð9:4Þ

where

1, if bi depends on m j
pij ¼ :
0, otherwise

The optical/wireless channel introduces the errors during transmission, and the received vector r can be represented as
r ¼ x + e, where e is the error vector (error pattern) whose components are determined by


1, if an error occurred in the ith location
ei ¼
0, otherwise

In order to verify if the received vector r is a codeword one, in incoming subsection we will introduce the concept of a
parity-check matrix as another useful matrix associated with the linear block codes.
9.1 Linear Block Codes Fundamentals 531

Let us expand the matrix equation x ¼ mG in a scalar form as follows:

x 0 ¼ m0
x 1 ¼ m1
...
xk1 ¼ mk1
ð9:5Þ
xk ¼ m0 p00 þ m1 p10 þ . . . þ mk1 pk1,0
xkþ1 ¼ m0 p01 þ m1 p11 þ . . . þ mk1 pk1,1
...
xn1 ¼ m0 p0,nk1 þ m1 p1,nk1 þ . . . þ mk1 pk1,nk1

By using the first k equalities, the last n  k equations can be rewritten in terms of the codeword elements as follows:

x0 p00 þ x1 p10 þ . . . þ xk1 pk1,0 þ xk ¼ 0

x0 p01 þ x1 p11 þ . . . þ xk1 pk1,0 þ xkþ1 ¼ 0
ð9:6Þ
...
x0 p0,nkþ1 þ x1 p1,nk1 þ . . . þ xk1 pk1,nkþ1 þ xn1 ¼ 0

The equations presented above can be rewritten through matrix representation as

2 3T
p00 p10 ... pk1,0 1 0 ... 0
6p 7
6 01 p11 ... pk1,1 0 1 ... 0 7
½x0 x1 . . . xn1 6 7 ¼0
|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl}4 ... ... ... ... 5
x ð9:7Þ
p p ... p k1,nk1 0 0 ... 1
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
0,nk1 1,nk1

HT
T T
⟺xH ¼ 0, H ¼ ½P Ink ðnkÞxn

The H-matrix in Eq. (9.7) is known as the parity-check one. We can easily verify that G and H matrices satisfy equation
 
P
GH T ¼ ½ Ik P ¼ P þ P ¼ 0, ð9:8Þ
Ink

meaning that the parity-check matrix H of an (n,k) linear block code has rank n-k and dimensions (n  k)  n whose null space
is k-dimensional vector with basis forming the generator matrix G.
Every (n,k) linear block code with generator matrix G and parity-check matrix H has a dual code, this time having
generator matrix H and parity-check matrix G. As an example, (n,1) repetition code and (n,n  1) single-parity-check code are
dual ones.
In order to determine the error correction capability of the linear block code, we have to introduce the concepts of
Hamming distance and Hamming weight [1, 2]. The Hamming distance d(x1,x2) between two codewords x1 and x2 is defined
as the number of locations in which these two vectors differ. The Hamming weight wt(x) of a codeword vector x is defined as
the number of nonzero elements in the vector. The minimum distance dmin of a linear block code is defined as the smallest
Hamming distance between any pair of code vectors in the code space. Since the zero vector is also a codeword, the minimum
distance of a linear block code can be defined as the smallest Hamming weight of the nonzero code vectors in the code.
We can write the parity-check matrix in a form H ¼ [h1 h2 . . . hn], where hi presents the i-th column in the matrix structure.
Since every codeword x must satisfy the syndrome equation xHT ¼ 0, the minimum distance of a linear block code is
determined by the minimum number of columns in the H-matrix whose sum is equal to zero vector. As an example, (7,4)
Hamming code discussed above has a minimum distance dmin ¼ 3 since the sum of first, fifth, and sixth columns leads to zero
vector.
532 9 Advanced Coding and Coded Modulation Techniques

r r
xi xji xi xj
t t r t
t

(a) (b)
Fig. 9.2 The illustration of Hamming distance: (a) d(xi,xj)  2 t + 1 and (b) d(xi,xj) < 2 t + 1

The codewords can be represented as points in n-dimensional space, as shown in Fig. 9.2. Decoding process can be
visualized by creating the spheres of radius t around codeword points. The received word vector r in Fig. 9.2a will be decoded
as a codeword xi because its Hamming distance d(xi,r)  t is closest to the codeword xi. On the other hand, in example shown
in Fig. 9.2b, the Hamming distance satisfies relation d(xi,xj)  2 t, and the received vector r that falls in intersection area of
two spheres cannot be uniquely decoded. Therefore, (n,k) linear block code of minimum distance dmin can correct up to t errors
if, and only if, t  ⌊1/2(dmin  1)c or dmin  2 t + 1 (where ⌊c denotes the largest integer smaller or equal to the enclosed
quantity). If we are only interested in detecting ed errors, then the minimum distance should be dmin  ed + 1. However, if we
are interested in detecting ed errors and correcting ec errors, then the minimum distance should be dmin  ed + ec + 1.
Accordingly, the Hamming (7,4) code is a single error correcting and double error detecting code. More generally, Hamming
codes are (n,k) linear block codes with the following parameters:

• Block length: n ¼ 2m  1
• Number of message bits: k ¼ 2m  m  1
• Number of parity bits: n  k ¼ m
• dmin ¼ 3

where m  3. Hamming codes belong to the class of perfect codes, the codes that satisfy the Hamming inequality given as
[1–7]

Xt  
nk n
2  : ð9:9Þ
i¼0 i

This bound gives how many errors t can be corrected with a specific (n,k) linear block code.
The coding gain a linear (n,k) block is defined as the relative saving in the energy per information bit for a given bit error
probability when coding is applied. Since the total information word energy kEb must be the same as the total codeword
energy nEc (where Ec is the transmitted bit energy, while Eb is the information bit energy), we can establish the following
relationship:

E c ¼ ðk=nÞE b ¼ RE b ð9:10Þ

where R defines the coding rate. As an example, the probability of error for BPSK on an AWGN channel, when coherent hard-
decision (bit-by-bit) demodulator is used, can be obtained as
rffiffiffiffiffiffi rffiffiffiffiffiffiffiffi
1 Ec 1 RE b
p ¼ erfc ¼ erfc , ð9:11Þ
2 N0 2 N0

where erfc(x) function is defined in Chap. 2. By using the Chernoff bound [3], we obtain the following expression for the
coding gain when hard decision is applied in decoding process:

ðEb =N 0 Þuncoded
 Rðt þ 1Þ, ð9:12Þ
ðE b =N 0 Þcoded

where t is the error correction capability of the code. The corresponding soft-decision coding gain can be estimated by [2, 3,
6].
9.1 Linear Block Codes Fundamentals 533

ðEb =N 0 Þuncoded
 Rd min : ð9:13Þ
ðE b =N 0 Þcoded

By comparing Eqs. (9.12) and (9.13), we can see that the soft-decision coding gain is about 3 dB better than hard-decision
coding gain since the minimum distance dmin  2 t + 1.
In fiber-optics communications, it is common to use the expression for BER on an AWGN for on-off keying and express
both the coding gain (CG) and net coding gain (NCG) as follows [1, 2, 6–12]:

CG ¼ 20log10 erfc1 ð2BERout,t Þ  20log10 erfc1 ð2BERin Þ ½dB ð9:14Þ

NCG ¼ CG þ 10 log 10 R ½dB ð9:15Þ

where BERin denotes the BER at the input of FEC decoder, BERout denotes the BER at the output of FEC decoder, while BERt
denotes the target BER (which is typically 1015).
In the rest of this section, we describe several important coding bounds including Hamming, Plotkin, Gilbert-Varshamov,
and Singleton ones [2, 3]. The Hamming bound has already been introduced for binary linear block codes (LBC) and was
expressed by Eq. (9.9). The Hamming bound for q-ary (n,k) LBC is given as
        
n 2 n i n t n
1 þ ðq  1 Þ þ ðq  1Þ þ . . . þ ð q  1Þ þ . . . þ ðq  1Þ qk  qn , ð9:16Þ
1 2 i t
 
n
where t is the error correction capability and ðq  1Þi is the number of received words that differ from a given codeword
i
in i symbols. The codes satisfying the Hamming bound with equality sign are known as perfect codes. Hamming codes are
perfect codes since n ¼ 2n  k  1, which is equivalent to (1 + n)2k ¼ 2n, so that the relationship between the right and left side
of Eq. (9.16) is expressed by equality sign. The (n,1) repetition code is also an example of a perfect code.
The Plotkin bound is defined by the following relation for minimum distance:

n2k1
dmin  : ð9:17Þ
2k  1

Namely, if all codewords are written as the rows of a 2k  n matrix, each column will contain 2k  1 “zeros” and 2k  1 “ones,”
with the total weight of all codewords being equal to n2k  1.
Gilbert-Varshamov bound is based on the property that the minimum distance dmin of a linear (n,k) block code and can be
determined as the minimum number of columns in H-matrix that adds to zero, i.e.,
     
n1 n1 n1
þ þ ... þ < 2nk  1: ð9:18Þ
1 2 dmin  2

The Singleton bound is defined by inequality:

d min  n  k þ 1: ð9:19Þ

This bound is straightforward to prove. Let only one bit of value 1 be present in information vector. If it is involved in n-k
parity checks, the total number of ones in codeword cannot be larger than n  k + 1. The codes satisfying the Singleton bound
with equality sign are known as the maximum-distance separable (MDS) codes (e.g., RS codes are MDS codes).
534 9 Advanced Coding and Coded Modulation Techniques

9.2 BCH Codes Fundamentals

The BCH codes, the most famous cyclic codes, were discovered by Hocquenghem in 1959 and by Bose and Chaudhuri in
1960 [2–5]. Among many different decoding algorithms, the most important are Massey-Berlekamp algorithm and Chien’s
search algorithm. An important subclass of BCH is a class of Reed-Solomon codes proposed in 1960.
Equipped we the knowledge of Galois fields from Appendix, we can continue our description of structure of BCH codes.
Let the finite field GF(q) (symbol field) and extension field GF(qm) (locator field), m  1, be given. For every m0 (m0  1) and
Hamming distance d, there exists a BCH code with the generating polynomial g(x), if and only if it is of smallest degree with
coefficients from GF(q) and the roots from the extension field GF(qm) as follows [2, 5]:

αm0 , αm0 þ1 , ⋯, αm0 þd2 ð9:20Þ

where α is from GF(qm). The codeword length is determined as the least common multiple of orders of roots. (The order of an
element β from finite field is the smallest positive integer j such that β j ¼ 1.)
It can be shown that for any positive integer m (m  3) and t (t < 2m  1), there exists a binary BCH code having the
following properties:

• Codeword length: n ¼ 2m  1
• Number of parity bits: n  k  mt
• Minimum Hamming distance: d  2 t + 1

This code is able to correct up to t errors. The generator polynomial can be found as the LCM of the minimal polynomials
of αi:

gðxÞ ¼ LCM½Pα1 ðxÞ, Pα3 ðxÞ, . . . , Pα2t1 ðxÞ ð9:21Þ

where α is a primitive element in GF(2m) and Pαi ðxÞ is the minimal polynomial of αi.
Let c(x) ¼ c0 + c1x + c2x2 + . . . + cn  1xn  1 be the codeword polynomial, and let the roots of generator polynomial be α,
α ,. . ., α2t, where t is the error correction capability of BCH code. Because the generator polynomial g(x) is the factor of
2

codeword polynomial c(x), the roots of g(x) must also be the roots of c(x):
 
c αi ¼ c0 þ c1 αi þ . . . þ cn1 αðn1Þi ¼ 0; 1  i  2t ð9:22Þ

This equation can also be written as inner (scalar) product of codeword vector c ¼ [c0 c1 . . . cn  1] and the following vector
[1 αi α2i . . . α2(n  1)i]:
2 3
1
6 αi 7
6 7
½c0 c1 . . . cn1 6 7 ¼ 0; 1  i  2t ð9:23Þ
4... 5
αðn1Þi

Equation (9.23) can also be written in the following matrix form:

2 3T
αn1 αn2 ... α 1
6 ðα2 Þn1 ðα2 Þn2 ... α2 1 7
6 7
½c0 c1 . . . cn1 6 3 n1
1 7 ¼ cH ¼ 0,
T
n2
6 ðα Þ ðα 3 Þ ... α3
7 ð9:24Þ
4 ... ... ... ... ... 5
n1 n2
ðα2t Þ ðα2t Þ ... α2t 1
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
HT
9.2 BCH Codes Fundamentals 535

where H is the parity-check matrix of BCH code. Using the property 2 of GF(q) from Appendix, we conclude that αi and α2i
are the roots of the same minimum polynomial, so that the even rows in H can be omitted to get the final version of the parity-
check matrix of BCH codes:
2 3
αn1 αn2 ... α 1
6 ðα3 Þn1 ðα3 Þ
n2
. . . α3 1 7
6 7
6  n1  5 n2 7
H¼6
6 α
5
α . . . α5 1 77 ð9:25Þ
6 7
4... ... ... ... ...5
n1 n2
ðα2t1 Þ ðα2t1 Þ . . . α2t1 1

For example, (15,7) 2-error correcting BCH code has the generator polynomial [6]

gðxÞ ¼ LCM ½Pα ðxÞ, Pα3 ðxÞ

    

¼ LCM x4 þ x þ 1, x þ α3 x þ α6 x þ α9 x þ α12
¼ x8 þ x7 þ x6 þ x4 þ 1,

wherein the primitive polynomial used to design this code is p(x) ¼ x4 + x + 1. The corresponding parity-check matrix is given
by
 
α14 α13 α12 ⋯ α 1
H¼ :
α42 α39 α36 ⋯ α3 1

By using the GF(24) property α15 ¼ 1, we can rewrite the parity-check matrix as
 
α14 α13 α12 α11 α10 α9 α8 α7 α6 α5 α4 α3 α2 α 1
H¼ :
α12 α9 α6 α3 1 α12 α9 α6 α3 1 α12 α9 α6 α3 1

The BCH code is able to correct up to t errors. Let us assume that error polynomial e(x) does not have more than t errors,
which can then be written as

j1 j2
eðxÞ ¼ e j1 x þ e j2 x þ . . . þ e jl x jl þ . . . þ e jv x jv ; 0  v  t ð9:26Þ

ej1 is the error magnitude, α jl is the error-location number, while jl is the error location. Notice that the error magnitudes are
from symbol field, while the error-location numbers are from extension field. The corresponding syndrome components can
be obtained from (9.26) and (9.24) as follows:
 i j1  i j2  i  jl  j
Si ¼ e j1 α þe j2 α þ ... þ e jl α þ . . . þ e jv αi v ; 0  v  t ð9:27Þ

In order to avoid the double indexing, let us introduce the following notation X l ¼ α jl , Y l ¼ e jl. The pair (Xl,Yl) completely
identifies the errors (l2[1,v]). We have to solve the following set of equations:

S1 ¼ Y 1 X 1 þ Y 2 X 2 þ . . . þ Y v X v
S2 ¼ Y 1 X 21 þ Y 2 X 22 þ . . . þ Y v X 2v
ð9:28Þ
...
S2t ¼ Y 1 X 2t
1 þ Y 2X2 þ . . . þ Y vXv :
2t 2t

The procedure to solve this system of equations represents the corresponding decoding algorithm. Direct solution of
this system of equations is impractical. There exists many different algorithms to solve the system of Eq. (9.28), ranging
from iterative to Euclidean algorithms [3–7]. The very popular decoding algorithm of BCH codes is Massey-Berlekamp
536 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.3 A shift register that

generates the syndromes Sj

- 1 - 2 - v-1 - v

SV+j-1 SV+j-2 Sj

algorithm [2–7]. In this algorithm the BCH decoding is observed as shift register synthesis problem: given the syndromes Si,
we have to find the minimal length shift register that generates the syndromes. Once we determine the coefficients of this shift
register, we construct the error locator polynomial [2–7]:

v
σ ðxÞ ¼ ∏ ð1 þ X i xÞ ¼ σ v xv þ σ v1 xv1 þ . . . þ σ 1 x þ 1, ð9:29Þ
i¼1

where the σ i’s, also known as elementary symmetric functions, are given by Viète’s formulas:

σ1 ¼ X1 þ X2 þ . . . þ Xv
X
σ2 ¼ XiX j
i<j
X
σ3 ¼ XiX jXk ð9:30Þ
i<j<k
...
σv ¼ X1X2 . . . Xv

Because {Xl} are the inverses of the roots of σ(x), σ(1/Xl) ¼ 0 8 l, and we can write [2–7]
 
X vl σ X 1
l ¼ X vl þ σ 1 X v1
l þ . . . þ σv: ð9:31Þ

By multiplying the previous equation by X lj and performing summation over l for fixed j, we obtain

Svþj þ σ 1 Svþj1 þ . . . þ σ v S j ¼ 0; j ¼ 1, 2, . . . , v ð9:32Þ

This equation can be rewritten as follows:

X
v
Svþj ¼  σ i Svþji ; j ¼ 1, 2, . . . , v ð9:33Þ
i¼1

where Sv + j represents the output of shift register as shown in Fig. 9.3. The Massey-Berlekamp algorithm is summarized by
flowchart shown in Fig. 9.4, which is self-explanatory.
Once the locator polynomial is determined, we have to find the roots and invert them to obtain the error locators. To
determine the error magnitudes, we have to define another polynomial, known as the error-evaluator polynomial, defined as
follows [6]:

ςðxÞ ¼ 1 þ ðS1 þ σ 1 Þx þ ðS2 þ σ 1 S1 þ σ 2 Þx2 þ . . . þ ðSv þ σ 1 Sv1 þ . . . þ σ v Þxv : ð9:34Þ

The error magnitudes are then obtained from [5]

 
ς X 1
l
Yl ¼ v  : ð9:35Þ
∏ 1 þ X i X 1
l
i¼1, i6¼l

For binary BCH codessince Yl ¼ 1 we do not need to evaluate the error magnitudes.
9.3 Trellis Description of Linear Block Codes and Viterbi Decoding Algorithm 537

Fig. 9.4 Flowchart of Massey-

Initialization:
Berlekamp algorithm. Δ denotes
the error (discrepancy) between
=0, (x)= 0=1, L=0, a(x)=1
the syndrome and the shift register
output; a(x) stores the content of
shift register (normalized by Δ1) =1,2,…,2t
prior the lengthening

i S i
i 0

0 No

Yes

' x x xa x

No
2L

Yes
1
a x x
a x xa x a x xa x
L L

x ' x

Yes
deg x L More than t errors
occurred

No

(x) is the inverse

locator polynomial

9.3 Trellis Description of Linear Block Codes and Viterbi Decoding Algorithm

Since the bits in codeword are statistically dependent, the encoding process can be described by the finite state machine, with
precisely determined initial and terminal states. We can further describe the encoding by means of trellis [13] and perform
maximum likelihood decoding using the Viterbi algorithm [2–7].
Let linear block code (n,k) be defined over GF(2). The corresponding parity-check matrix of dimension (n  k)  n can be
written as
538 9 Advanced Coding and Coded Modulation Techniques

H ¼ ½h1 h2 . . . hn , ð9:36Þ

where hi denotes the i-th column of parity-check matrix. The codeword vector c ¼ (c1c2. . .cn) satisfies the following equation:

cH T ¼ 0: ð9:37Þ

Let the code be systematic with the first k bits representing the information bits and last n  k bits representing the parity bits.
The state at depth j is denoted by Sj. The number of states in trellis is determined by 2n  k, each represented in binary form as a
vector of length n  k. Let the information vector be denoted by m ¼ (m1m2. . .mk). The new state at depth j, Sj, is related to the
previous state Sj  1 by [13]

Sj ¼ S j1 þ c jh j, ð9:38Þ

where cj ¼ mj (for j  k). For j ¼ k + 1,. . .,n the corresponding parity bits are obtained by generalized parity checks according
to the generator matrix G. For terminal state Sn to be equal to initial state S0, Eq. (9.37) is to be satisfied. Different paths
throughout the trellis correspond to 2k different codewords.
For example, the parity-check and generator matrices of Hamming (7,4) code are given, respectively, as
2 3
2 3 1 0 0 0 1 1 1
1 1 0 1 1 0 0 60
6 7 6 1 0 0 1 0 17
7
H ¼ 41 0 1 1 0 1 05 G¼6 7
40 0 1 0 0 1 15
1 1 1 0 0 0 1
0 0 0 1 1 1 0

The corresponding trellis is shown in Fig. 9.5. The solid edges correspond to cj ¼ 1, while dashed edges to cj ¼ 0. There are
2n  k ¼ 8 states in trellis represented as 3-tuples. The total number of paths through trellis is 2k ¼ 16, and these paths
correspond to different codewords. This trellis is created based on Eq. (9.38). For example, for j ¼ 1 and c1 ¼ 0, the terminal
state is obtained by S1 ¼ S0 + 0h1 ¼ (000) + 0(111) ¼ (000), while for c1 ¼ 1 the terminal state is obtained by
S1 ¼ S0 + 1h1 ¼ (000) + 0(111) ¼ (111). For j ¼ 2, initial state S1 ¼ (111), and c2 ¼ 1, we arrive at terminal state
S2 ¼ S1 + 1h2 ¼ (111) + 1(101) ¼ (010) and so on. After j ¼ 3 we can see that trellis is fully developed; there exist 16 paths

j
Sj 0 1 2 3 4 5 6 7
(000)

(001)

(010)

(011)

(100)

(101)

(110)

(111)
Fig. 9.5 The trellis representation of Hamming (7,4) code
9.3 Trellis Description of Linear Block Codes and Viterbi Decoding Algorithm 539

leading to 8 nodes (with 2 edges reaching the terminal state). These 16 paths correspond to 16 possible information sequences
of length 4. The number of possible transitions after j ¼ k reduces dramatically because the remaining bits c5, c6, and c7 are
parity bits, which are algebraically related to the previous information bits. For q-ary linear block codes, once the trellis is fully
developed, every node will have q outgoing and q incoming edges. It is interesting to notice that the trellis for liner block
codes is time-variant.
The decoding rule for choosing an estimate of the code vector C, given the received vector r, is optimum when the
probability of decoding error is minimized. The maximum likelihood (ML) decoder decision rule can be formulated as
follows:
Choose the estimate of C if the log-likelihood function logpR j C(r| c) is maximum.
The log-likelihood function, given the transmitted codeword c ¼ (c1c2. . .cn), assuming that components of receiver word
r ¼ (r1r2. . .rn) are statistically independent and noise is additive can be written as follows:

X
n
X
n
log pRjC ðrjcÞ ¼ log pN i ðr i jci Þ ¼  M i ðr i , ci Þ≜  MðCÞ, ð9:39Þ
i¼1 i¼1

where pR|C(r|c) is the joint probability density function (PDF), pN i ðr i jci Þ is the noise PDF, and Mi is the branch metric. The ML
decoder provides the codeword C that minimizes the path metric M(C). Consider BPSK coherent optical communication
system with homodyne balanced detection and amplified spontaneous emission (ASE) noise dominated scenario. The
corresponding PDF function will be
pffiffiffiffiffiffiffiffi 2
1 ðr c 2R P P Þ
 i i PD 2 S LO
pN i ðr i jci Þ ¼ pffiffiffiffiffi e 2σ , ð9:40Þ
σ 2π

where RPD is the photodiode responsivity; Ps and PLO are the average powers of incoming and local laser, respectively; and
σ 2 is the variance of equivalent noise process dominated by ASE noise. The corresponding log-likelihood function can be
obtained by

X
n pffiffiffiffiffiffiffiffiffiffiffiffiffi 2
ðr i  ci 2RPD PS PLO Þ n  
log pRjC ðrjcÞ ¼   log πσ 2
i¼1
2σ 2 2
ð9:41Þ
Xn
¼ A r i ci þ B ¼ Aðr  cÞ þ B,
i¼1

where A and B are constants independent on transmitted codeword. The path metric is therefore the inner (scalar) product of
the received word and transmitted codeword.
The brute force decoding method would mean to try all possible codewords 2k and select one that minimizes the path metric
M(C) ¼ rc. The better option would be to use Viterbi algorithm [2–7, 14], which is a recursive algorithm, whereby many
codewords can be discarded from consideration in determination of c that minimizes the path metric. Alternatively, we can use
the square Euclidean metric, which is equivalent to the correlation metric as shown by Eq. (9.41). The square Euclidean metric
can also be used for ASK transmission, wherein ci2{0,1}. Below we provide an illustrative example that employs the squared
Euclidean metric and ASK. To keep the exposition simple, we perform the following normalization, 2σ 2 ¼ 1 and
2RPD√(PsPLO) ¼ 1, and assume that Hamming (7,4) code (with corresponding trellis shown in Fig. 9.5) is used. Let the
received word be as follows r ¼ (0.4, 0.2, 0.5, 0.1, 0.2, 0.1, 0.2), while the all-zero codeword was transmitted. The first four
samples correspond to information bits, and last three to the parity bits. The decoding starts when all branches merge together
for the first time, which occurs at trellis depth j ¼ 4. The cumulative metric for (000)- > (000) path at depth j ¼ 4,
corresponding to the information sequence 0000, is (0.4  0)2 + (0.2  0)2 + (0.5  0)2 + (0.1  0)2 ¼ 0.46; while the
cumulative metric for (000)- > (000) path at depth j ¼ 4, corresponding to the information sequence 0111, is
(0.4  0)2 + (0.2  1)2 + (0.5  1)2 + (0.1  1)2 ¼ 1.86. Out of two possible paths, the path with lower accumulated metric
(0.46) is preserved and is commonly referred to as the survivor path. The other path is discarded from further consideration.
The similar procedure is repeated for all state nodes at level j ¼ 4, and the survivor paths are denoted as bolded letters in
Table 9.1. In step 2 ( j ¼ 5), we repeat the similar procedure using the accumulated metric from survived paths as the starting
540 9 Advanced Coding and Coded Modulation Techniques

Table 9.1 Illustrating the soft Viterbi decoding of Hamming (7,4) code
Step 1 Step 2 Step 3 Step 4
Received samples ! 0.4, 0.2, 0.5, 0.1 0.2 0.1 0.2
Paths
#
(000) ! (000) 0000: 0.46 0: 0.46 + (0.2–0)2 ¼ 0.5 0: 0.5 + (0.1–0)2 ¼ 0.51 0: 0.51 + (0.2–0)2 ¼ 0.55
0111: 1.86 1: 0.66 + (0.2–1)2 ¼ 1.3 1: 1.3 + (0.1–1)2 ¼ 2.11 1: 1.31 + (0.2–1)2 ¼ 1.95
(000) ! (001) 1110: 1.26 0: 1.26 + (0.2–0)2 ¼ 1.3 0: 1.3 + (0.1–0)2 ¼ 1.31
1001: 1.46 1: 1.06 + (0.2–1)2 ¼ 1.7 1: 0.5 + (0.1–1)2 ¼ 1.31
(000) ! (010) 1100: 1.26 0: 1.26 + (0.2–0)2 ¼ 1.3
1011: 1.46 1: 1.06 + (0.2–1)2 ¼ 1.7
(000) ! (011) 0010: 0.46 0: 0.46 + (0.2–0)2 ¼ 0.5
0101: 1.86 1: 0.66 + (0.2–1)2 ¼ 1.3
(000) ! (100) 1101: 2.06
1010: 0.66
(000) ! (101) 0011: 1.26
0100: 1.06
(000) ! (110) 0001: 1.26
0110: 1.06
(000) ! (111) 1111: 2.06
1000: 0.66
Adopted from [5]

point. When two paths have the same accumulated metric, we flip the coin to determine which one to choose without affecting
the final result. We see that in step 5 the only survived path is all-zeros path, and the transmitted codeword is properly
decoded. We can also introduce a single error, for example, by setting 0.5 sample to 0.6, and repeat the decoding procedure.
The all-zero codeword will be detected again. However, when two errors are introduced, the Viterbi decoder will fail to
determine the correct codeword. On the other hand, by checking the syndrome equation, we can detect the double error.
If the probability of error is independent of the transmitted bit value, the corresponding channel is called the binary
symmetric channel (BSC). This channel is completely described by transition (crossover) probability: P(ri ¼ 1| ci ¼ 0) ¼ P(-
ri ¼ 0| ci ¼ 1) ¼ p. Because

p, r i 6¼ ci
pðr i jci Þ ¼ ,
1  p, r i ¼ ci

the corresponding log-likelihood function can be written as follows:

log pðrjcÞ ¼ d H log p þ ðn  dH Þ log ð1  pÞ

 
p ð9:42Þ
¼ dH log þ n log ð1  pÞ,
1p

where dH is the Hamming distance between the received word r and transmitted codeword c. Therefore, for p < 1/2 the ML
rule for BSC can be formulated as:
Choose the estimate C that minimizes the Hamming distance between the received vector r and codeword c.
For intensity modulation with direct detection (IM/DD), it was found in [15] that distribution of PDF tails is exponential
rather than Gaussian, so that square root metric is more suitable than Euclidean metric. Some other metrics suitable for IM/DD
systems are discussed in [16]. For high-speed long-haul optical communication system, such as those operating at 40 Gb/s and
above, the influence of self-phase modulation (intrachannel four-wave mixing in systems with direct detection and nonlinear
phase noise in systems with coherent detection) is important, and we have to use the metric given by Eq. (9.42), providing that
interleaving is used so that neighboring symbols are independent of each other. The PDF has to be estimated by histogram
methods. An alternative would be to use the edgeworth expansion method as we described in [17].
So far we were concerned with different classes of linear block codes. Another important class of codes is based on
convolutional codes. Convolutional codes of practical importance are commonly (n,1) codes, whose code rate is low 1/n, and
as such are not suitable for fiber-optics communication, because the penalty due to chromatic dispersion is highly severe at
9.4 Convolutional Codes 541

high-data rates. The code rate can be increased by puncturing, at the expense of performance degradation. On the other hand,
the recursive systematic convolutional (RSC) codes are usually employed as component codes for turbo codes that are used in
wireless and deep-space communications.

9.4 Convolutional Codes

In a systematic (n,k) linear block code, information bits get unchanged during the encoding process; only n  k parity-check
bits are added that are algebraically related to the information bits. If the information bits are statistically independent, the
parity (redundant) bits will only be dependent on the information block of current codeword, so that the codewords are
statistically independent. On the other hand, in an (n,k,M) convolutional code, the parity bits are not only the function of
current information k-tuple but also the function of previous m information k-tuples. The statistical dependence is therefore
introduced in a window of K ¼ n(M + 1) symbols, which is known as the constraint length. The constrained length represents
the number of coded symbols influenced by a single-message symbol. The general architecture of a convolutional code is
shown in Fig. 9.6. The k information bits are taken during encoding process from a serial-to-parallel converter (S/P) that
follows an information buffer as the input of the encoder memory of size k-by-M. The same bits are also used as the input to
the combinational logic, which determines the n-outputs based on current k-tuple and previous Mk bits. The logic outputs are
written in parallel to the output shift register, and the codeword is transmitted in a serial fashion over the channel.
An example of nonsystematic convolutional (2,1,2) code is shown in Fig. 9.7. This encoder has the code rate R ¼ 1/2 and
has M ¼ 2 flip-flops as memory elements. Let the input sequence to the encoder be denoted by m ¼ (m0 m1 . . . ml . . .).

1 2 3 … M-1 M
1
2
Information S/P
… … …
buffer converter
k-1
k
…

… LOGIC

…
n-bit shift register

Fig. 9.6 Illustrating the operational principle of convolutional codes

Fig. 9.7 Nonsystematic

convolutional (2,1,1) encoder

c(1)

c(2)
542 9 Advanced Coding and Coded Modulation Techniques
 
ð1Þ ð1Þ ð1Þ ð2Þ ð2Þ ð2Þ
The corresponding output sequences are denoted with cð1Þ ¼ c0 c1 . . . cl . . . and cð2Þ ¼ c0 c1 . . . cl . . . , respec-
tively. The encoder sequence is obtained by multiplexing two output sequences as follows: c ¼

ð1Þ ð2Þ ð1Þ ð2Þ ð1Þ ð2Þ
c0 c0 c1 c1 . . . cl cl . . . . For convolutional (2,1,2) code shown in Fig. 9.7, the output sequences are related to the
input sequence by

ð1Þ ð2Þ
cl ¼ ml þ ml2 , cl ¼ ml þ ml1 þ ml2 :

Another approach to describe the encoding operation is by convolution of the input sequence and corresponding generator
sequences g(1) ¼ (101) and g(2) ¼ (111) as follows:

cð1Þ ¼ m gð1Þ , cð2Þ ¼ m gð2Þ ,

where the generator sequences are obtained as the response to the unit step sequence (100. . .). The generator sequence,
representing the impulse response of corresponding input-output path can also be determined as the sequence of connections
from the shift register to the pertinent adder, with a 1 representing the existence of connection and a 0 representing the absence
of connection.
The encoding process in convolutional encoder can also be represented in similar fashion to that of the block codes, as
matrix product of the input message vector m and generator matrix G:
2 ð1Þ ð2Þ ð1Þ ð2Þ ð 1Þ ð2Þ
3
g0 g0 g1 g1 g2 g2 0 0 0 0 ...
6 7
60 g0
ð1Þ
g0
ð2Þ
g1
ð 1Þ ð2Þ
g1 g2
ð1Þ
g2
ð2Þ
... 7
6 0 0 0 7
6 7
c ¼ mG, G ¼ 6 0 0 0 0 g0
ð 1Þ ð2Þ
g0 g1
ð1Þ
g1
ð2Þ
g2
ð1Þ
g2
ð2Þ
... 7 :
6 7
6 7
4⋮ ⋮ ⋱ ⋱ ⋮ ⋮ 5
ð1Þ ð2Þ ð1Þ ð2Þ ð1Þ ð 2Þ
0 0 0 0 ⋯ g0 g0 g1 g1 g2 g2

The generator matrix of an (n,1,m) convolutional code, of rate R ¼ 1/n, can be represented by
2 3
G0 G1 ⋯ GM
6 G0 G1 ⋯ GM 7
6 7
G¼6 7, ð9:43Þ
4 G0 G1 ⋯ GM 5
⋱

ð1Þ ð2Þ ðnÞ
where Gl ¼ gl gl . . . gl ; l ¼ 0, 1, . . . , M . Since the number of memory elements in shift register is M, there exist
n generator sequences g(i) (i ¼ 1,2,. . .,n) of length M + 1, and n output sequences c(l ) (l ¼ 1,..,n) are multiplexed before being
transmitted over the channel. The j-th output symbol and l-th input symbol are related by the following convolution
relationship:

ð jÞ
X
M
ð jÞ
cl ¼ mli gi : ð9:44Þ
i¼0

(The summation in (9.44) is to be performed by mod 2.)

By employing the delay operator D (or z1 in digital signal processing books), we can simplify the encoding process and
perform the multiplication in transform domain instead of convolution in time domain. The transform-domain representations
of message sequence m and j-th output sequence can, respectively, be written as

mðDÞ ¼ m0 þ m1 D þ m2 D2 þ . . .
ð jÞ ð jÞ ð jÞ
ð9:45Þ
cð jÞ ðDÞ ¼ c0 þ c1 D þ c2 D2 þ . . .
9.4 Convolutional Codes 543

For (2,1,2) convolutional encoder shown in Fig. 9.7, the corresponding transform-domain representations of generator
polynomials can be written as

g1 ðDÞ ¼ 1 þ D2 g2 ð D Þ ¼ 1 þ D þ D 2 :

The corresponding transform-domain generator matrix G(D) will be

GðDÞ ¼ ½ g1 ðDÞ g2 ð D Þ  ¼ 1 þ D 2 1 þ D þ D2 :

The output sequences in transform domain can be written as

cð1Þ ðDÞ ¼ mðDÞg1 ðDÞ cð2Þ ðDÞ ¼ mðDÞg2 ðDÞ:

The multiplexed sequence in transform domain is as follows:

cðDÞ ¼ cð1Þ ðDÞ þ Dcð2Þ ðDÞ:

For an arbitrary (n,1,m) convolutional code, the corresponding generator matrix in transform domain can be represented as

GðDÞ ¼ ½ g1 ðDÞ g2 ðDÞ ⋯ gn ðDÞ , ð9:46Þ

while the transform domain of output sequence can be represented by

h i
cðDÞ ¼ mðDÞGðDÞ, cðDÞ ¼ cð1Þ ðDÞ cð2Þ ðDÞ ⋯ cðnÞ ðDÞ : ð9:47aÞ

For an arbitrary (n,k,M) convolutional code, we have to split the input sequence into k subsequences m ¼ [m1 m2 . . . mk]
and define the generator matrix as
2 3
G0 G1 ⋯ GM
6 G0 G1 ⋯ GM 7
6 7
G¼6 7, ð9:47bÞ
4 G0 G1 ⋯ GM 5
⋱

where
2 ð11Þ ð12Þ ð1nÞ
3
gl gl ⋯ gl
6 ð21Þ 7
6g ð22Þ
gl ⋯ gl 7
ð2nÞ
Gl ¼ 6
6
l 7,
7
4⋮ ⋮ ⋱ ⋮ 5
ðk1Þ ðk2Þ ðknÞ
gl gl ⋯ gl

and proceed in similar fashion as above.

The convolutional codes can also be put in systematic form, by transforming the generator matrix and representing it into
the following form:

GðDÞ ¼ ½ I PðDÞ , ð9:48Þ

where I is k  k identity matrix and P(D) is kx(n  k) matrix of polynomials. The nonsystematic generator matrix of linear
block codes can easily be transformed in systematic form by elementary operations per rows, while this operation is not
always possible to do for the convolutional codes. Namely, the two generator matrices are equivalent if they generate the same
544 9 Advanced Coding and Coded Modulation Techniques

convolutional code. Let us again observe the (2,1,2) convolutional code shown in Fig. 9.7, whose generator matrix is
G(D) ¼ [1 + D2 1 + D + D2]. We can perform the following transformation:
 

  1 þ D þ D2
cðDÞ ¼ mðDÞGðDÞ ¼ mðDÞ 1 þ D2 1 þ D þ D2 ¼ mðDÞ 1 þ D2 1
1 þ D2
¼ m0 ðDÞG0 ðDÞ,

where
 
0

0 1 þ D þ D2
m ðDÞ = mðDÞT ðDÞ; T ½D = 1 þ D , G ðDÞ ¼ 1
2
:
1 þ D2

Because both generator matrices, G(D) and G0 (D), will generate the same output sequence for the same input sequence, we can
say that these two matrices are equivalent, which represents a necessary condition. For two generator matrices to be
equivalent, the T(D) matrix should be invertible, which represents the sufficient condition. In previous example, m(D) can
be obtained from m0 (D) by
 
mðDÞ ¼ m ðDÞT 1 ðDÞ; T 1 ½D ¼
0 1
:
1 þ D2

Therefore, two generator matrices G(D) and G0 (D) ¼ T(D)G(D) are equivalent if the matrix T(D) is invertible. In the example
above, G0 (D) is in systematic form; unfortunately its elements are not polynomials but rather rational functions. The parity
sequence can be generated by first multiplying the input sequence by 1 + D + D2 and then dividing by 1 + D2. One possible
implementation is shown in Fig. 9.8.
Similar to the linear block codes, we can define the parity-check matrix in transform domain H(D) (of size (n-k)xk) that
satisfies the following equation:

GðDÞH T ðDÞ ¼ 0: ð9:49Þ

Every codeword c(D) satisfies the following equation:

cðDÞH T ðDÞ ¼ 0: ð9:50Þ

If G(D) is the generator matrix of a systematic convolutional code, given by Eq. (9.48), the corresponding parity-check matrix
can be written in the following format:

H ðDÞ ¼ PT ðDÞ I , ð9:51Þ

where I is (n  k)  (n  k) identity matrix. For example, shown in Fig. 9.8, the parity-check matrix in transform domain can
be written as
 
1 þ D þ D2
H ðD Þ ¼ 1 :
1 þ D2

For nonsystematic codes, the code sequence on receiver c(D) should be multiplied by G1(D) to determine the information
sequence:

cðDÞG1 ðDÞ ¼ mðDÞGðDÞG1 ðDÞ ¼ mðDÞDl , ð9:52Þ

which can be done if the following is valid:

9.4 Convolutional Codes 545

Fig. 9.8 The systematic encoder m

equivalent to the nonsystematic
encoder shown in Fig. 9.7

c(1)

c(2)

GðDÞG1 ðDÞ ¼ IDl , l  0 ð9:53Þ

where the multiplication with Dl corresponds to the delay. Therefore, the generator matrix must be invertible. If not, the
corresponding convolutional code is catastrophic, that is, the finite number of channel errors will introduce infinite number of
decoding errors. The sufficient condition for the convolutional code to be non-catastrophicis that
 the greatest common divisor
n
(GCD) of all submatrices of size kxk be equal to Dl. (The number of these submatrices is .) The systematic codes cannot
k
be catastrophic because their generator matrix is invertible. Any non-catastrophic convolutional code can be transformed into
systematic equivalent. For our example shown in Fig. 9.8, the GCD can be found as
 
GCD 1 þ D2 , 1 þ D þ D2 ¼ 1 ¼ D0 :

Therefore, for the (n,1,M) codes, it is sufficient not to have any common roots among generator polynomials except probable
one of the form Dl.
The convolutional codes can be considered as being a subclass of so-called trellis codes. The trellis codes can be described
as finite state machines, with the output being the function of Mk previous bits and k current bits, whose output can be
described in the form of trellis. If the trellis encoder is time-invariant and linear, the corresponding code is called the
convolutional code.
Similarly as in block codes where the error correction capability is related to the minimum Hamming distance and decoding
algorithm, the performance of convolutional codes is determined by decoding algorithm and so-called free distance of the
code, dfree. The free distance of convolutional codes is defined as the minimum Hamming distance between any two
codewords in the code. Because the convolutional codes contain all-zero codeword and belong to the class of linear codes,
the free distance can be determined as the minimum weight output codeword sequence caused by a certain nonzero input
sequence. The following three algorithms are commonly used in decoding of convolutional codes: (i) majority logic decoding,
(ii) sequential decoding, and (iii) decoding by Viterbi algorithm. The Viterbi algorithm is the most popular and it is already
explained in the previous section. The only difference is in trellis, which is in case of convolutional codes time-invariant, so
that Viterbi decoder of convolutional codes is simpler to implement than that of block codes. The description of other two
decoding algorithms can be found in any textbook on channel coding [3–7]. The calculation of free distance of convolutional
codes is described below.
The calculation of free distance of convolutional codes is closely related to the generating function of a convolutional code
[3–7], i.e., the transfer function of the encoder with respect to state transitions. We have shown in Fig. 9.6 that convolutional
codes can be represented using the finite state machine (FSM) approach, and its operation can be described using the state
diagram. The state of convolutional encoder of rate 1/n is defined by the sequence of M most recent bits (mj-M,mj  M + 1,. . .,
mj  1,mj) moved into encoder’s shift register, and it is denoted by (mj  1,. . .,mj  M), where mj denotes the current bit. For
example, for convolutional encoder shown in Fig. 9.9a with M ¼ 2 flip-flops, there are 2M ¼ 4 possible states denoted by
a (00), b (10), c (01), and d (11). The corresponding state diagram of this encoder is shown in Fig. 9.9b, c. The FSM states
correspond to the 2-bit states of shift register. The pair of labels above edges, in Fig. 9.9b, represents the outputs of encoder for
a given input. For example, if the content of shift register is 10 (we are in state 10) and incoming bit is 1, the generated output
sequence is 01, and in the next clock cycle, the content of shift register will be 11, meaning that the terminal state is 11.
546 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.9 Convolutional

(2,1,2) code: (a) encoder, (b) state
diagram, and (c) version 2 of state
diagram

m c

(a)
10/1 10

11 11
d
01/1 01/0 01 01

10/0 10
10 01 b 10 01 c
00/1 00
11/1 11/0 11 11

a
00 00

00/0 00
(b) (c)

Fig. 9.10 The modified state The exponent of I represents the Hamming weight
DLI
diagram of convolutional (2,1,2) Modified state diagram of corresponding input
code from Fig. 9.9a
d The exponent of D represents the Hamming weight
of the encoder output corresponding to that branch

The exponent of L represents the branch length

DLI DL
(always one)

D2LI DL D2L

a0 b c a1

Another equivalent representation is given in Fig. 9.9c, where dashed lines correspond to input bit 1 and the solid lines
correspond to input bit 0.
The state diagram can be used as an effective tool to determine the generating function of the code. First of all we have to
modify the state diagram as shown in Fig. 9.10. The all-zero state a is split into an initial state a0 and a final state a1. The labels
9.4 Convolutional Codes 547

All-zero path

Error events

Fig. 9.11 Introducing the concept of error event

of edges are changed as well. The label of each edge is composed of three letters with corresponding exponents. The exponent
of I represents the Hamming weight of input (for bit zero we obtain I0 ¼ 1), the exponent of L is the branch length, and the
exponent of D represents the Hamming weight of the encoder output corresponding to the input of observed edge. With
different paths from initial to final state in modified state diagram, we associate the corresponding label by concatenating the
edge labels. For example, the path a0bdca1 has the label D6L4I3. The path starting at a0 and ending at a1 is called the
fundamental path. If the number of paths starting at a0 and ending at a1 with label DdLlIi is denoted by Td,l,i, the following
generating function will represent the complete path enumerator [2–7, 18]:

X
1 X
1 X
1
T ðD, L, I Þ ¼ T d,l,i Dd Ll I i : ð9:54Þ
d¼1 l¼1 i¼1

For example, shown in Fig. 9.10, we can determine the T(D,L,I ) by using Mason’s gain formula [19] from signal-flaw graph
theory to obtain [18]

D5 L3 I
T ðD, L, I Þ ¼ ¼ D5 L3 I þ D6 L4 I 2 ð1 þ LÞ þ D7 L5 I 3 ð1 þ LÞ2 þ . . . ð9:55Þ
1  DLI ð1 þ LÞ

From Eq. (9.55) we conclude that there are no fundamental paths at distance 0, 1, 2, or 3 from the all-zero path; and there is a
single fundamental path at distance 5 from all-zero path meaning that the free distance of this convolutional code is dfree ¼ 5.
This implies that Viterbi decoder can correct up to two errors in decoded sequence.
The modified state diagram can also be used to estimate the bit error ratios (BERs). Without loss of generality, for an
AWGN channel model, we will assume that the all-zero codeword was transmitted (c ¼ 0). We will say that an error event has
occurred if a nonzero codeword survives at state 0, therefore eliminating the correct path (all-zero codeword), which is
illustrated in Fig. 9.11.
The probability that the first error event E of weight d occurs is given by [6]
8 !
>
> Pd d
>
> pi ð1  pÞdi , d is odd
>
>
>
> dþ1 i
>
< i¼
2
PðE, d Þ ¼ ! ! ð9:56Þ
>
> d Xd d i
>
>
1 d=2
p ð 1  pÞ þ d=2
p ð1  pÞdi , d is even
>
> 2 d=2
>
> d i
>
: i¼ þ 1
2

which is always smaller than P(E, d ) < 2dpd/2(1  p)d/2, where p is the crossover probability of BSC. The event error
probability upper bound can be determined by [6].

X
1
PE < T d PðE, d Þ, ð9:57Þ
d¼dfree

where Td denotes the number of codewords of weight d. Combining (9.56) and (9.57), we obtain
548 9 Advanced Coding and Coded Modulation Techniques

PE < T d ½4pð1  pÞd=2 ¼ T ðD, I ÞjI¼1,D¼pffiffiffiffiffiffiffiffiffiffiffiffi

4pð1pÞ
: ð9:58Þ

The BER of a convolutional code of rate r ¼ k/n is upper bounded by [6]

1 X
1
BER  βd PðE, dÞ ð9:59Þ
k d¼d
free

where βd denotes the number of nonzero data bits carried by all Td paths of weight d. By setting L ¼ 1 in (9.55), we obtain

X
1 X
1
T ðD, I Þ ¼ T d,i Dd I i : ð9:60Þ
d¼dfree i¼1

By differentiating T(D,I) with respect to I and by setting I ¼ 1 upon differentiation, we obtain


∂T ðD, I Þ X1 X1 X1 X
1

 ¼ iT d,i Dd ¼ βd Dd , βd ¼ iT d,i ð9:61Þ

∂I I¼1 d¼d i¼1 d¼d i¼1
free free

Finally, by comparing (9.59) and (9.61), we conclude that BER is upper bounded by

1 X
1
1 ∂T ðD, I Þ
BER   ¼ β d Dd : ð9:62Þ
k ∂I I¼1
k d¼d
free

For binary symmetric channel, the crossover probability is

rffiffiffiffiffiffi  
1 Eb 1 Eb
p ¼ erfc  pffiffiffi exp  , ð9:63Þ
2 N0 2 π N0

where Eb/N0 is electrical bit energy per power spectral density ratio, and the parameter D from (9.62) can be estimated as
 
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Eb
D ¼ 4pð1  pÞ  1:06 exp  : ð9:64aÞ
N0

By substituting (9.64a) into (9.62), we obtain

 
1 X 1 X
1 1
d d dRE b
BER  βd D  1:06 βd exp  , ð9:64bÞ
k d¼d k d¼d 2N 0
free free

where R is the code rate. For high SNRs only the first term in summation above dominates:
   
1 d free RE b E
dfree
BER  1:06 βd exp  for large b ð9:65Þ
k 2N 0 N0

The corresponding expression for uncoded BER at high SNR can be estimated by
 
E
BERunc  0:282 exp  b : ð9:66Þ
N0

By comparing the exponents in (9.65) and (9.66), we conclude that argument in coded case is dfreeR/2 times larger, so that
corresponding hard-decision asymptotic coding gain is
9.5 RS Codes, Concatenated Codes, and Product Codes 549

 
d free R
Gh ¼ 10 log 10 ½dB: ð9:67Þ
2

For binary input AWGN channel, we can repeat the similar procedure and obtain the following asymptotic coding gain for soft
decoding:

Gs ¼ 10 log 10 ðd free RÞ ½dB, ð9:68Þ

which is 3 dB better than that for hard-decision decoding.

9.5 RS Codes, Concatenated Codes, and Product Codes

The Reed-Solomon (RS) codes were discovered in 1960 and represent a special class of nonbinary BCH codes [2–7, 20]. RS
codes represent the most commonly used nonbinary codes. Both the code symbols and the roots of generating polynomial are
from the locator field. In other words, the symbol field and locator field are the same. The codeword length of RS codes is
determined by n ¼ qm  1 ¼ q  1, so that these codes are relatively short codes. The minimum polynomial for some element
β is Pβ (x) ¼ x  β. If α is the primitive element of GF(q) (q is a prime or prime power), the generator polynomial for t-error
correcting Reed-Solomon code is given by
   
gðxÞ ¼ ðx  αÞ x  α2 ⋯ x  α2t : ð9:69Þ

The generator polynomial degree is 2t and it is the same as the number of parity symbols n  k ¼ 2t, while the block length of
the code is n ¼ q  1. Since the minimum distance of BCH codes is 2 t + 1, the minimum distance of RS codes is
dmin ¼ n  k + 1, satisfying therefore the Singleton bound (dmin  n  k + 1) with equality, thus belonging to the class of
maximum-distance separable (MDS) codes. When q ¼ 2m, the RS code parameters are n ¼ m(2m  1), n  k ¼ 2 mt, and
dmin ¼ 2 mt + 1. Therefore, the minimum distance of RS codes, when observed as binary codes, is large. The RS codes may be
considered as burst error correcting codes and as such are suitable for bursty-error-prone channels. This binary code is able to
correct up to t bursts of length m. Equivalently, this binary code will be able to correct a single burst of length (t  1)m + 1.
The weight distribution of RS codes can be determined as [2, 5, 14]

  X
id  
n min
i1
Ai ¼ ðq  1Þ ð1Þ j qidmin j , ð9:70Þ
i j¼0 j

and by using this expression, we can evaluate the undetected error probability as [2, 5, 14]

X
n
P u ð eÞ ¼ Ai pi ð1  pÞni , ð9:71Þ
i¼1

where p is the crossover probability and Ai denotes the number of codewords of weight i. The codeword weights can be
determined by McWilliams identity [2, 5, 7, 14].

Example Let the GF(4) be generated by 1 + x + x2. The symbols of GF(4) are 0, 1, α, and α2. The generator polynomial for
RS(3,2) code is given by g(x) ¼ x  α. The corresponding codewords are 000, 101, α0α, α20α2, 011, 110, α1α2, α21α, 0αα,
1αα2, αα0, α2α1, 0α2α2, 1α2α, αα21, and α2α20. This code is essentially the even parity-check code (α2 + α + 1 ¼ 0). The
generator polynomial for RS(3,1) is g(x) ¼ (x  α)(x  α2) ¼ x2 + x + 1, while the corresponding codewords are 000, 111,
ααα, and α2α2α2. Therefore, this code is in fact the repetition code.

Since RS codes are special class of nonbinary BCH codes, they can be decoded by using the same decoding algorithm as
one already explained in the previous section.
550 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.12 The concatenated (Nn, 2k-ary super channel

Kk,Dd) code
Outer (N,K,D) code Inner (n,k,d) code Inner Outer
Channel
over GF(2k) over GF(2) decoder decoder

n1
k1

Checks on
k1k2 information

rows
k2
symbols
n2

Checks
Checks on columns on
checks
Fig. 9.13 The structure of a codeword of a product code

To further improve the burst error correction capability of RS codes, RS code can be combined with an inner binary block
code in a concatenation scheme as shown in Fig. 9.12. The key idea behind the concatenation scheme can be explained as
follows. Consider the codeword generated by inner (n,k,d) code (with d being the minimum distance of the code) and transmitted
over the bursty channel. The decoder processes the erroneously received codeword and decodes it correctly. However,
occasionally the received codeword is decoded incorrectly. Therefore, the inner encoder, the channel, and the inner decoder
may be considered as a superchannel whose input and output alphabets belong to GF(2k). The outer encoder (N,K,D) (D – the
minimum distance of outer code) encodes input K symbols and generates N output symbols transmitted over the superchannel.
The length of each symbol is k information digits. The resulting scheme is known as concatenated code [2, 5, 7, 14]. This scheme
is an (Nn,Kk,Dd) code with the minimum distance of at least Dd. For instance, RS(255,239,8) code can be combined with the
(12,8,3) single-parity-check code in the concatenation Scheme (12255,2398,24). The concatenated scheme from Fig. 9.12
can be generalized to q-ary channels, where the inner code operates over GF(q) and outer over GF(qk).
Another way to deal with burst errors is to arrange two RS codes in a product manner as shown in Fig. 9.13. A new product
code is an (n1n2,k1k2,d1d2) scheme in which codewords form an n1 x n2 array such that each row is a codeword from C1 (n1,k1,d1)
code, while each column is a codeword from another C2 (n2,k2,d2) code, where ni, ki, and di (i ¼ 1,2) are the codeword length,
dimension, and minimum distance of i-th component code, respectively. Turbo-product codes were originally proposed in
[29]. Both binary (such as binary BCH codes) and nonbinary codes (such as RS codes) may be arranged in a product code
manner. It is possible to show that the minimum distance of a product codes is the product of minimum distances of component
codes. Accordingly, it can be shown that the product code is able to correct the burst error of length b ¼ max(n1b2,n2b1), where bi
is the burst error capability of component code i ¼ 1,2.
The results of Monte Carlo simulations for several RS concatenation schemes on an optical on-off keying AWGN channel
are shown in Fig. 9.14. The net coding gain is measured by difference in Q-values for a specific BER. As we can see, the
concatenation Scheme RS(255,239) + RS(255,223) of code rate R ¼ 0.82 outperforms all other combinations.

9.6 Coding with Interleaving

An interleaved code is obtained by taking L codewords (of length n) of a given code xi ¼ (xi1,xi2,. . .,xin) (i ¼ 1,2,. . .,L ) and
forming the new codeword by interleaving the L codewords as yi ¼ (x11,x21,..,xL1, x12,x22,..,xL2,. . .,x1n,x2n,..,xLn). The process
of interleaving can be visualized as the process of forming an L  n matrix of L codewords written row by row and
transmitting the matrix column by column, as given below:
9.6 Coding with Interleaving 551

Fig. 9.14 BER performance of

concatenated RS codes on an
optical on-off keying channel

x11 x12 ... x1n

x21 x22 ... x2n
... ... ... ...
xL1 xL2 ... xLn

The parameter L is known as the interleaving degree. The transmission must be delayed until L codewords are collected. To be
able to transmit a column whenever a new codeword becomes available, the codewords should be arranged down diagonals as
shown below. This interleaving scheme is known as the delayed interleaving (1-frame delayed interleaving):

xiðn1Þ,1 ... xi2,1 xi1,1 xi,1

xiðn1Þ,2 ... xi2,2 xi1,2 xi,2
... ... ... ... ...
xiðn1Þ,n1 xiðn2Þ,n1
xiðn1Þ,n xiðn2Þ,n

Each new codeword completes one column of this array. In the example above, the codeword xi completes the column (frame)
xi,1,xi-1,2,. . .,xi-(N  1),n. A generalization of this scheme, in which the components of i-th codeword xi (say xi,j and xi,j + 1) are
spaced λ frames apart, is known as λ-frame delayed interleaving.
Interleaving is an efficient way to deal with error bursts introduced by fading effects in wireless communication channels.
In the presence of burst of errors, the code designed for random error needs to deal with both random errors and burst errors.
By using coding with interleaving, we spread error bursts due to deep fades over many codewords such that each received
codeword only exhibits at most a few simultaneous symbol errors, which can be corrected for. The interleaver depth must be
large enough so that fading is independent across a received codeword. Slowly fading channels require large interleavers,
which can introduce the large decoding delays. In the block interleaver discussed above, during transmission, two neighboring
codeword symbols, originating the same codeword, are separated by L  1 symbols from other codewords, so that the
symbols from the same codeword will experience approximately independent fading conditions providing that their separa-
tion in time is greater than the channel coherence time Tc:

LT s  T c  1=BD , ð9:72Þ

where Ts is the symbol duration and BD is the Doppler spread. The block coding with interleaving is illustrated in Fig. 9.15.
The codewords from block encoder are written into block interleaver in column-wise fashion. After L codewords are
written, the modulator reads bits from the interleaver in column-wise fashion, starting from the first column. On receiver side,
552 9 Advanced Coding and Coded Modulation Techniques

Modulator Channel

column-wise
Codewords

fashion
read in
x11 x12 … x1k … x1n

Block interleaver
Source Codewords written in
Block x21 x22 … x2k … x2n
bits encoder row-wise fashion

…
…

…
xL1 xL2 … xLk … xLn
k information bits n-k parity
bits

From
Demodulator
channel
deinterleaver

wise fashion
Written into

in column-

y11 y12 … y1k … y1n

Read from deinterleaver Block Estimation of
y21 y22 … y2k … y2n
in row-wise fashion decoder transmitted bits
…

yL1 yL2 … yLk … yLn

Fig. 9.15 Illustrating block coding with interleaving and deinterleaving principles

after demodulation, for hard decision the bits upon hard decision are written into deinterleaver in column-wise fashion. Once
the deinterleaver is filled, the block decoder reads the content of deinterleaver in row-wise fashion, and decoding process takes
place. On the other hand, for soft-decision decoding, corresponding samples are written into deinterleaver in column-wise
fashion. For two-dimensional signaling schemes, the two samples, corresponding to in-phase and quadrature channel, are
written in column-wise fashion. The i-th row and j-th element of deinterleaver are then represented as yij ¼ (yij,I, yij,Q), where
the first component corresponds to the sample in in-phase channel, while the second component corresponds to the sample in
quadrature channel. These channel samples are read in row-wise fashion by soft-decision block decoder.
When convolutional code is used instead of block code, the corresponding convolutional interleaver is provided in
Fig. 9.16. The convolutional encoder output is multiplexed into the stage of L buffers of increasing size, from zero buffering
size to the buffer of size (L – 1). We use z1 to denote one unit delay of Ts seconds. The buffer stage output is multiplexed into
the channel in a similar fashion. On receiver side, the reverse operation is performed. Clearly, any two code symbols are
separated by LTs, and providing that the buffering delay is larger than the coherence time Tc, the interleaver is efficient in
dealing with fading effects. The total number of memory elements per buffering stage is 0.5 L(L  1).
9.7 Codes on Graphs Basics 553

Buffering stage
Buffering stage
z (L 1)

z 1

…
Convolutional Convolutional
Channel
encoder decoder

…
z 1

z (L 1)

Fig. 9.16 Illustration of convolutional coding with interleaving and deinterleaving principles

Fig. 9.17 Bipartite graph Sub-code m (nm,km) described

examples: (a) bipartite graph for a Constraint nodes by H-matrix Hm
generalized LDPC (GLDPC) code
and (b) bipartite graph for parallel Degree n1 Degree n2 Degree nm
concatenation of codes. The
constraint node could be any Interleaver
linear code

Bit nodes

(a)
Constraint nodes

Interleaver

Bit nodes … … … … …
Parity bits Data bits Data bits Parity bits

(b)

9.7 Codes on Graphs Basics

A graph is typically defined as a set of nodes connected by edges. There are many ways to describe a code by using graphical
approach including the factor graph approach [21]; one such approach is described below. We associate two types of nodes to
a given code: bit (variable) nodes and constraint (function) nodes. The bit nodes can only be connected to the function nodes
and vice versa. Such a graph, which allows only the nodes of different type to be connected, is known as a bipartite graph
[1, 2, 4, 7, 22]. Two examples of bipartite graphs, corresponding to generalized LDPC (GLDPC) codes and parallel
concatenation of codes, are shown in Fig. 9.17. In GLDPC coding, shown in Fig. 9.17a, each constraint node, also known
as subcode, represents an (ns, ks) (s ¼ 1,. . .,m) linear block code such as single-parity-check (SPC) code, BCH code (including
Hamming code), or Reed-Müller (RM) code. There is one-to-one correspondence between the set of variable nodes and the set
of all bits in a codeword. However, we might allow certain number of variable nodes to be punctured (corresponding bits are
not transmitted), to adapt to time-varying optical channel conditions. In parallel concatenation of codes (Fig. 9.17b), constraint
nodes are typically convolutional codes. Very often only two constraint nodes (constituent codes) are used in conventional
turbo codes [20, 23–27].
The codes on graphs of interest in wireless and optical communications include turbo codes [20, 23–28], turbo-product
codes [2, 8, 9, 28, 30], and LDPC codes [1, 2, 4, 5, 7, 10–12, 31–51]. As fiber-optics communications are affected by fiber
nonlinearities, the code rate should be kept sufficiently high (0.75).
554 9 Advanced Coding and Coded Modulation Techniques

9.8 Turbo Codes

A turbo encoder comprises the concatenation of two (or more) convolution encoders, while corresponding decoders consist of
two (or more) convolutional soft decoders in which extrinsic probabilistic information is iterated back and forth among soft
decoders. Parallel turbo encoder, shown in Fig. 9.18a, consists of two rate one-half convolutional encoders arranged in parallel
concatenation scheme. In a serial turbo encoder, shown in Fig. 9.18b, the serially concatenated convolutional encoders are
separated by K/R0-interleaver, where R0 is the code rate of the outer encoder. The interleaver takes incoming block of bits and
arranges them in pseudorandom fashion prior encoding by the second encoder, so that the same information bits are not
encoded twice by the same recursive systematic convolutional (RSC) code (when identical RSC codes are used). The
generator matrix of an RSC code can be written as GRSC ðxÞ ¼ ½ 1 g2 ðxÞ=g1 ðxÞ  , where g2(x)/g1(x) denotes the transfer
function of the parity branch of encoder (see Fig. 9.18a). As an example, the RSC encoder described by GRSC ðxÞ ¼

1 ð1 þ x4 Þ=ð1 þ x þ x2 þ x3 þ x4 Þ is shown in Fig. 9.18c. The output of the RSC encoder in parallel turbo code, the
codeword, can be represented as
2 3
h i
x¼ 4x1 x2 ⋯ xk
|{z}
⋯xK 5 ðinfoÞ ðparityÞ ðinfoÞ ðparityÞ
¼ x1 , x1 , x2 , x2
ðinfoÞ ðparityÞ
, . . . , uK , x K , ð9:73Þ
½xðkinfoÞ , xðkparityÞ 

Fig. 9.18 Turbo codes encoder u

configurations for (a) parallel and
(b) serial turbo codes. (c) RSC p
G2(D)/G1(D)
encoder with GRSC ðxÞ ¼
RSC1
p, q
Puncturer
K-bit interleaver

q
G2(D)/G1(D)

RSC2
(a)
u

K/R0-bit
p interleaver
G2(D)/G1(D) Punct. G2(D)/G1(D) Punct.

RSC1 RSC2
(b)
uk uk

ck

mod 2
Flip-flop

pk

(c)

1 ð1 þ x4 Þ=ð1 þ x þ x2 þ x3 þ x4 Þ
9.8 Turbo Codes 555

Fig. 9.19 Parallel turbo decoder

configuration. MUX multiplexer, L1(uk) - Le12(u’k)
DEC decoder u u
Le
Interleaver u u
Lc(uk) 12(uk)
DEC1 - DEC2
M c
U c -
Lc(pk) Le21(u’k) L2(u’k)
X Le 21(uk)
Deinterleaver
-

Lc(uk) Interleaver M
U
Lc(qk) X

ðinfoÞ ðparityÞ
where xk is k-th information bit (k ¼ 1,2,. . .,K ) and xk is the k-th parity bit. For antipodal signaling, we can represent
ðinfoÞ ðparityÞ
xk xk
the modulation process as uk ¼ ð1Þ , pk ¼ ð1Þ so that the transmitted word c is represented as

c ¼ ½c1 , c2 , . . . , cK  ¼ ½u1 , p1 , u2 , p2 , . . . , uK , pK , c k ¼ ½ uk , pk  ð9:74Þ

The iterative (turbo) decoder, shown in Fig. 9.19, interleaves two soft-input/soft-output (SISO) decoders, exchanging the
extrinsic information iteratively and cooperatively. The role of iterative decoder is to iteratively estimate the a posteriori
probabilities (APPs) Pr(uk|y), where y is a received codeword plus noise n:

y ¼ c þ n ¼ ½y1 , y2 , . . . , yK  ¼ yu1 , yp1 , yu2 , yp2 , . . . , yuK , ypK , yk ¼ yuk , ypk ð9:75Þ

with yuk being the received sample corresponding to k-th data bit, while ypk being the sample corresponding to the k-th parity bit.
In iterative decoders, each component decoder receives extrinsic or soft information for each uk from its companion
decoder, which provides “the prior” information, as illustrated in Fig. 7.20. The key idea behind extrinsic information is that
decoder DEC2 provides the soft information to DEC1 for each uk by using only information that is not available to DEC1. The
knowledge of APPs allows the optimum decisions for information bits uk to be made by the maximum a posteriori probability
(MAP) rule:

þ1  
Pðuk ¼ þ1jyÞ > Pðuk ¼ þ1jyÞ
1 ⟺ uk ¼ sign ½Lðuk Þ, Lðuk Þ ¼ log , ð9:76Þ
Pðuk ¼ 1jyÞ < Pðuk ¼ 1jyÞ
1

where L(uk) is the log-APP also known as log-likelihood ratio (LLR). By applying Bayes’ rule, we can write that
   
Pðbk ¼ þ1jyÞ pðyÞ Pðyjbk ¼ þ1Þ Pðbk ¼ þ1Þ
Lðbk Þ ¼ log ¼ log
Pðbk ¼ 1jyÞ pðyÞ Pðyjbk ¼ 1Þ Pðbk ¼ 1Þ
    ð9:77Þ
Pðyjbk ¼ þ1Þ Pðbk ¼ þ1Þ
¼ log þ log ,
Pðyjbk ¼ 1Þ Pðbk ¼ 1Þ

where bk 2 {uk, pk} and the second term represents a priori information. The first term in Eq. (9.77) represents the extrinsic
information for iterative decoders obtained from corresponding companion decoder. Notice that for conventional decoders we
typically have P(bk ¼ +1) ¼ P(bk ¼ 1).
The key idea behind extrinsic information is to provide only soft information about bk (that is not already available) to the
companion decoder. Therefore, although initially the prior information is zero, it becomes nonzero after the first iteration. As
shown in Fig. 7.20, the extrinsic information to be sent from DEC1 to DEC2, which is denoted as Le12 ðuk Þ, can be calculated by
subtracting from DEC1 output LLR L1(uk) both the channel reliability Lc(uk) and extrinsic information Le21 ðuk Þ already
received from DEC2. On the other hand, the extrinsic information Le21 ðuk Þ, to be sent from DEC2 to  DEC1, is obtained by
subtracting both the interleaved channel reliability Lc(uk) and interleaved extrinsic information Le12 u0k (already received from
DEC1) from DEC2’s output LLR L2(uk). Since the DEC2 operates on interleaved sequence, this extrinsic information needs to
be deinterleaved before being sent to DEC1. This exchange of extrinsic information is performed until the successful decoding
is done or until the predetermined number of iterations is reached. Decoders DECi (i ¼ 1,2) operate on trellis description of
556 9 Advanced Coding and Coded Modulation Techniques

encoder, by employing the BCJR algorithm [13]. The BCJR algorithm can also be used for MAP detection, and as such it will
be described in section on turbo equalization. Since the resulting code rate of parallel turbo encoder is low (R ¼ 1/3), the
puncturer deletes the selected bits in order to reduce the coding overhead, as shown in Fig. 9.18a. The resulting code rate
becomes R ¼ K/(K + P), where P is the number of parity bits remained after puncturing. On the other hand, the puncturing
reduces the minimum distance of original code, which leads to performance degradation and an early error floor as shown in
[28]. Because of low-code rate and early error floor, the turbo codes are not widely used in fiber-optics communications.
However, they are widely used in wireless communication systems as well as deep-space optical communications.
The details of the BCJR decoder are provided below. We assume that RSC is represented using the trellis description, in
similar fashion as described in Sect. 9.3. The key difference with respect to the linear block codes is that trellis of RSC is time-
invariant. Let s0 denote one of the states in current stage of the trellis, while s denote the corresponding state in the next stage of
trellis. The transition from the state s0 to state s can be represented as the pair (s0 , s). Let U+ (U) denote the set of pairs (s0 ,s) for
the state transitions (sk  1 ¼ s0 ) ! (sk ¼ s) corresponding to uk ¼ +1 (uk ¼ 1). The LLR of the k-th data bit can be
represented as
P
pðsk1 ¼ s0 , sk ¼ s, yÞ
Uþ
Lðuk Þ ¼ log P : ð9:78Þ
pðsk1 ¼ s0 , sk ¼ s, yÞ
U

Therefore, we have to compute p(s’,s,y) for all state transitions and then sum up over appropriate transitions in numerator and
denominator. The BCJR algorithm is composed of two recursions, forward and backward recursion as illustrated in Fig. 9.21.
To efficiently compute LLRs, we factor p(s0 ,s,y) as follows:

pðs0 , s, yÞ ¼ αk1 ðs0 Þ  γ k ðs0 , sÞ  βk ðsÞ, ð9:79Þ

where the state probabilities in forward recursion αk(s), the transition probabilities, γ k(s0 ,s), and the state probabilities in
backward recursion βk(s) are defined, respectively, as
   
αk ðsÞ ¼ p sk ¼ s, yk1 , γ k ðs0 , sÞ ¼ pðsk ¼ s, yk jsk1 ¼ s0 Þ, βk ðsÞ ¼ p yKkþ1 jsk ¼ s ð9:80Þ

where we used yba to denote [ya, ya + 1, . . ., yb]. In the forward recursion, we calculate the state probabilities for the next trellis
as follows:
X
αk ðsÞ ¼ αk1 ðs0 Þγ k ðs0 , sÞ, ð9:81Þ
s0

where the summation is performed over all edges merging into state s, as illustrated in Fig. 9.20a. The transition probabilities
γ k(s0 ,s) are determined as

pðs, yk js0 Þpðs0 Þ pðs0 , s, yk Þ pðs0 , sÞ pðs0 , sÞ pðs0 , s, yk Þ

γ k ðs0 , sÞ ¼ pðs, yk js0 Þ ¼ ¼ ¼
pð s 0 Þ pðs0 Þ pðs0 , sÞ pð s 0 Þ pð s 0 , s Þ ð9:82Þ
0 0 0
¼ pðsjs Þpðyk js , sÞ ¼ pðuk Þ pðyk js , sÞ,

where p(s|s0 ) equals 1/2 for a valid (existing) transition, while p(yk|s0 ,s) is determined from channel conditional PDF. For an
AWGN channel and antipodal signaling, the transition probabilities are simply

ðyuk uk Þ þðypk pk Þ

2 2
kyk ck k2
1  1 
0
γ k ðs , sÞ ¼ pðuk Þ pffiffiffiffiffi e 2σ 2
k ¼ pðuk Þ pffiffiffiffiffi e 2σ 2
k : ð9:83Þ
σ 2π σ 2π

In the backward recursion, the state probabilities in the previous trellis stage are determined by, as illustrated in Fig. 9.20b,
9.8 Turbo Codes 557

k-1(s0) (s0,s’0) k(s0)

s’0 s0

s’1

(s0,s’m)
…

s’m k-1(s’m)

k(s0)= (s0,s’0) k-1(s0) + (s0,s’m) k-1(s’m) +…

(a)
(s0,s’0) k(s0)
k-1(s0) s’0 s0

s’1 s1

(sl,s’0) …
k-1(s’0)= (s0,s’0) k(s0) + (sl,s’0) k(sl) +… sl k(sl)

(b)

Fig. 9.20 Illustration of forward (a) and backward (b) recursions in BCJR algorithm

X
βk1 ðs0 Þ ¼ γ k ðs0 , sÞβk ðsÞ, ð9:84Þ
s

where the summation is performed over all incoming edges to state s’ in backward direction. Clearly, the BCJR algorithm is
composed of two Viterbi-like algorithms running in opposite directions. Now we have to calculate the LLRs, as follows:
P
αk1 ðs0 Þ  γ k ðs0 , sÞ  βk ðsÞ
Uþ
Lðuk Þ ¼ log P , ð9:85Þ
αk1 ðs0 Þ  γ k ðs0 , sÞ  βk ðsÞ

U

and this expression is obtained by substitution of Eq. (9.79) into (9.78).

The factorization claim (9.79) can be proved by applying conditional probability formula two times and removing
redundant variables.
   K 0  0 
pðs0 , s, yÞ ¼ p s0 , s, yk1 K k1 k1
1 , yk , ykþ1 ¼ p ykþ1 js , s, y1 , yk  p s , s, y1 , yk
     0 k1 
¼ p yKkþ1 js0 , s, yk1
1 , yk  p s, yk js , y1
0 k1
 p s , y1
 K 0
¼ p ykþ1 js  pðs, yk js0 Þ  pðs0 , yk1 ð9:86Þ
1 Þ
|fflfflfflfflfflffl{zfflfflfflfflfflffl} |fflfflfflfflffl{zfflfflfflfflffl} |fflfflfflfflfflffl{zfflfflfflfflffl
ffl}
βk ðsÞ γ k ðs0 , sÞ αk1 ðs0 Þ

¼ αk1 ðs0 Þ  γ k ðs0 , sÞ  βk ðsÞ:

On the other hand, the forward recursion step (9.81) has been derived as follows:
P P
αk ðsÞ ¼ pðsk ¼ s, yk1 Þ ¼ pðs, yk1 Þ ¼ pðs0 , s, yk1 Þ ¼ pðs, yk js0 , yk1 0 k1
1 Þpðs , y1 Þ
0 0 |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflffl{zfflfflfflfflfflffl}
s s
P
γ k ðs0 , sÞ αk1 ðs0 Þ ð9:89Þ
0 0
¼ αk1 ðs Þγ k ðs , sÞ:
s0
558 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.21 Log-domain BCJR algorithm steps: (a) forward recursion, (b) backward recursion, (c) LLR calculation step

In similar fashion, the backward recursion step (9.84) is derived as follows:

  X  K 
βk1 ðs0 Þ ¼ p yKk js0 ¼ p yk , sjs0
s
X   X  
¼ p yKkþ1 js0 , s, yk pðs, yk js0 Þ ¼ p yKkþ1 js pðs, yk js0 Þ
s s |fflfflfflfflffl{zfflfflfflfflffl}|fflfflfflfflffl{zfflfflfflfflffl} ð9:90Þ
βk ðsÞ γ k ðs0 , sÞ
X
¼ βk ðsÞγ k ðs0 , sÞ:
s

The initialization can be performed as

 
1, s¼0 1, s¼0
α0 ðsÞ ¼ β K ðsÞ ¼ ð9:91Þ
0, s 6¼ 0 0, s 6¼ 0
9.8 Turbo Codes 559

The calculation of state probabilities as well as LLRs involves a product of small probabilities, which can cause numerical
problems, and to solve for this problem, the log-domain BCJR algorithm can be formulated as follows. The forward metric in
log domain is introduced by taking the logarithm of probability-domain forward metric by
" # " #
X X
0 0 0 0
~k ðsÞ ¼ log ½αk ðsÞ ¼ log
α αk1 ðs Þγ k ðs , sÞ ¼ log exp ð~
αk1 ðs Þ þ ~γ k ðs , sÞÞ ð9:92Þ
s0 s0

On the other hand, the backward metric in log domain is introduced by taking the logarithm of probability-domain backward
metric by
" #
P
β~k1 ðs0 Þ ¼ log ½βk1 ðs0 Þ ¼ log γ k ðs0 , sÞβk ðsÞ
s0
" # ð9:93Þ
P
¼ log 0 ~
exp ð~γ k ðs , sÞ þ βk ðsÞÞ :
s0

The branch metric in log domain is obtained by taking the logarithm of corresponding transition probability as

γ k ðs0 , sÞ ¼ log ½γ k ðs0 , sÞ ¼ log ½Pðuk Þpðyk juk Þ

e
ð9:94Þ
¼ log ½Pðuk Þ þ log ½pðyk juk Þ:

For an AWGN channel and antipodal signaling, the branch metric gets simplified to

pffiffiffiffiffi ky  c k2
γ k ðs0 , sÞ ¼  log 2σ 2π  k 2 k :
e ð9:95Þ
2σ

The LLRs can be expressed in terms of forward, backward, and branch metrics as follows:
P
αk1 ðs0 Þγ k ðs0 , sÞβk ðsÞ
Uþ
Lðuk Þ ¼ log P
αk1 ðs0 Þγ k ðs0 , sÞβk ðsÞ
U
(P )
αk1 ðs0 Þ γ k ðs0 , sÞ βk ðsÞ ð9:96Þ
¼ log Uþ |fflfflfflffl{zfflfflfflffl} |fflfflffl{zfflfflffl} |ffl{zffl} 
αk1 ðs0 Þexp ½~γ k ðs0 , sÞexp ½β~k ðsÞ
exp ½~
8 9
<P =
αk1 ðs0 Þ γ k ðs0 , sÞ βk ðsÞ
log  |fflfflfflffl{zfflfflfflffl} |fflfflffl{zfflfflffl} |ffl{zffl} ,
:U αk1 ðs0 Þexp ½~γ k ðs0 , sÞexp ½β~k ðsÞ
exp ½~
;

which can be written in more compact form as

(X h i)
Lðuk Þ ¼ log exp α γ k ðs0 , sÞ þ e
ek1 ðs0 Þ þ e β k ðsÞ
Uþ
X h i ð9:97Þ
 log exp e γ k ðs0 , sÞ þ e
αk1 ðs0 Þ þ e βk ðsÞ :
U

Clearly, in (9.35) we recognize the following operation log(ex + e y), which we can denote as max*(x,y). The max*-operation
can be efficiently calculated as

log ðex þ ey Þ ¼ max ðx, yÞ ¼ max ðx, yÞ þ log 1 þ ejxyj , ð9:98Þ

560 9 Advanced Coding and Coded Modulation Techniques

where the second term is a correction term and can efficiently be implemented with the help of a look-up table (LUT) in
hardware. The max*-operator for more than two terms can be applied recursively, to terms at the time. For instance, for three
entries x, y, and z, the max*-operator is applied as

max ðx, y, zÞ ¼ max ð max ðx, yÞ, zÞ: ð9:99Þ

To prove the claim (9.98), we can express the max(x,y) as follows:



?ex þ ey
max ðx, yÞ ¼ log : ð9:100Þ
1 þ ejxyj

Let x > y, then the left side of equation is clearly x, while the right side of previous equation becomes

   !
ex þ ey ex 1 þ eðxyÞ
log ¼ log ¼ log ðex Þ ¼ x, ð9:101Þ
1 þ ejxyj 1 þ eðxyÞ

which proves the correctness of claim (9.98). With the introduction of max*-operation, the forward recursion (metric)
simplifies to
" #
X
0 0
e
αk ðsÞ ¼ log exp ðe
αk1 ðs Þ þ e
γ k ðs , sÞÞ ¼ max
0
αk1 ðs0 Þ þ e
½e γ k ðs0 , sÞ, ð9:102Þ
s
s0

and it is illustrated in Fig. 9.21a. On the other hand, the backward recursion (metric) becomes
" #
X  h i
e 0
βk1 ðs Þ ¼ log γ k ðs , sÞ þ e
exp e 0
βk ðsÞ ¼ max e
βk ðsÞ þ eγ k ðs0 , sÞ , ð9:103Þ
s
s0

and it is illustrated in Fig. 9.21b. Finally, the LLR computation step simplifies to
h i h i
Lðuk Þ ¼ 0
max eαk1 ðs0 Þ þ eγ k ðs0 , sÞ þ e
β k ðsÞ  0
max e γ k ðs0 , sÞ þ e
αk1 ðs0 Þ þ e β k ðsÞ ð9:104Þ
ðs , sÞ: uk ¼0 ðs , sÞ: uk ¼1

and it is illustrated in Fig. 9.21c. The initialization step is obtained by taking logarithm of (9.91)
 
0, s¼0 0, s¼0
e
α0 ðsÞ ¼ log ½α0 ðsÞ ¼ e
βK ðsÞ ¼ log ½βK ðsÞ ¼ ð9:105Þ
1, s 6¼ 0 1, s 6¼ 0

The log-domain BCJR algorithm can be summarized as follows:

0. Initialization step:
( (
0, s¼0 0, s¼0
e
α 0 ðsÞ ¼ e
βK ðsÞ ¼
1, s 6¼ 0 1, s 6¼ 0

1. For k ¼ 1,2,. . .,K, take the samples from the channel yk ¼ yuk , ypk , and compute the branch metric
γ k(s0, s) ¼ log [p(yk| uk)] + log [P(uk)].
9.9 Turbo-Product Codes 561

2. For k ¼ 1,2,. . .,K, apply the forward recursion:

e
αk ðsÞ ¼ max
0
αk1 ðs0 Þ þ e
½e γ k ðs0 , sÞ:
s

3. For k ¼ K, K  1,. . .,2, apply the backward recursion:

e
βk1 ðsÞ ¼ max ½e γ k ðs0 , sÞ:
βk ðsÞ þ e
s

4. For k ¼ 1,2,. . .,K, compute LLRs:

h i h i
Lðuk Þ ¼ max eαk1 ðs0 Þ þ eγ k ðs0 , sÞ þ e
β k ðsÞ  max γ k ðs0 , sÞ þ e
αk1 ðs0 Þ þ e
e β k ðsÞ
ðs0 , sÞ: uk ¼þ1 ðs0 , sÞ: uk ¼1

5. Make decisions for k ¼ 1,2,. . .,K:

b
uk ¼ sign½Lðuk Þ:

If a valid codeword is obtained or maximum number of iterations is reached, then stop; otherwise, go to step 1 to update the
branch metric with new extrinsic information log[P(uk)].
The BCJR algorithm is a basic building block for implementing DEC1 and DEC2 in Fig. 9.20.

9.9 Turbo-Product Codes

A turbo-product code (TPC), which is also known as block turbo code (BTC) [2, 8, 9, 29], is an (n1n2,k1k2,d1d2) code. The ni,
ki, and di (i ¼ 1,2) denote the codeword length, dimension, and minimum distance, respectively, of the i-th component code
(i ¼ 1,2). A TPC codeword, shown in Fig. 9.13, forms an n1  n2 array in which each row represents a codeword from an (n1,
k1,d1) code C1, while each column represents a codeword from an (n2,k2,d2) code C2. The code rate of TPC is given by
R ¼ R1R2 ¼ (k1k2)/(n1n2), where Ri is the code rate of the i-th (i ¼ 1,2) component code, namely, Ri ¼ ki/ni. The corresponding
overhead (OH) of each component code is defined as OHi ¼ (1/Ri  1)100%. The TPC can be considered as a particular
instance of serial concatenated block codes. In this interpretation, rows belong to the outer code and columns to the inner code,
and the interleaver is a deterministic column-row interleaver. The corresponding TPC decoder is shown in Fig. 9.22.
In soft-decision decoding, the extrinsic reliabilities are iterated between soft-input/soft-output (SISO) decoders for C1 and
C2. The extrinsic information in iterative decoders is obtained from corresponding companion decoder. The key idea behind
extrinsic information is to provide to the companion decoder only with soft information not already available to it. In fiber-
optics communications, TPCs based on extended BCH (including extended Hamming) component codes were intensively
studied, e.g., [8, 9, 30]. The product codes were proposed by Elias [52], but the term “turbo” is used when two SISO decoders
exchange the extrinsic information. As already mentioned above, the minimum distance of a product code is the product of
minimum distances of component codes, d ¼ d1d2. The constituent codes are typically extended BCH codes, because with
extended BCH codes, we can increase the minimum distance for d1 + d2 + 1 compared to the nominal BCH codes. The

Channel sample rk

Lrow Lrce
Interleaver
Row - Column
-
decoder Deinterleaver decoder
Lcre L column

Fig. 9.22 The decoder of a turbo-product code

562 9 Advanced Coding and Coded Modulation Techniques

expressions for bit error probability Pb and codeword error probability Pcw under maximum likelihood (ML) decoding, for
additive white Gaussian noise (AWGN) channel, can be estimated as follows [2]:
rffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffi
wmin RdE b A RdE b
Pb ’ erfc , Pcw ’ min erfc , ð9:106Þ
2k 1 k2 N0 2 N0

where wmin is the minimum weight of all Amin TPC codewords at the minimum distance d. Notice that wmin and Amin are quite
large, resulting in excellent bit error rate (BER) performance of BTCs. Unfortunately, due to high complexity, maximum
likelihood (ML) decoding is typically not used in BTC decoding, but simple Chase II-like decoding algorithms have been
used instead [2, 29]. One such algorithm, which is independent of the channel model [2], is described below. Let vi be i-th bit
in a codeword v ¼ [v1. . .vn] and yi be the corresponding received sample in a received vector of samples y ¼ [y1. . .yn]. The role
of decoder is to iteratively estimate the a posteriori probabilities (APPs) Pr(vi|yi). The knowledge of APPs allows for optimum
decisions on the bits vi by MAP rule

0
Pðvi ¼ 0jyÞ >
1, ð9:107Þ
Pðvi ¼ 1jyÞ <
1

or, more conveniently, we can write


1  sign½Lðvi Þ þ1, x  0
bvi ¼ , sign½x ¼ ð9:108Þ
2 1, x < 0

where L(vi) is the log a posteriori probability (log-APP), commonly referred to as the log-likelihood ratio (LLR), defined as
 
Pðvi ¼ 0jyÞ
Lðvi Þ ¼ log : ð9:109Þ
Pðvi ¼ 1jyÞ

The component SISO decoders for C1 and C2 calculate L(vi) and exchange extrinsic information as explained above. The
initial/channel bit LLRs, to be used in SISO decoding including Chase II algorithm, can be calculated by

Pðvi ¼ 0jyi Þ
Lch ðvi Þ ¼ log , ð9:110Þ
Pðvi ¼ 1jyi Þ

where the conditional probability P(vi|yi) is determined by

Pðyi jvi ÞPðvi Þ

Pðvi jyi Þ ¼ , ð9:111Þ
Pðyi jvi ¼ 0ÞPðvi ¼ 0Þ þ Pðyi jvi ¼ 1ÞPðvi ¼ 1Þ

In the presence of fiber nonlinearities, the conditional P(yi|) can be evaluated by estimation of histograms. On the other hand,
the channel LLRs for AWGN channel can be calculated as follows:

Lch ðvi Þ ¼ 2yi=σ2 , for binary input AWGN ð9:112Þ

where σ2 is the variance of the Gaussian distribution of the AWGN. It has been shown in experimental study for beyond
100 Gb/s transmission [53, 54] that in the links without in-line chromatic dispersion (CD) compensation, the distribution of
samples upon compensation of CD and nonlinear phase compensation is still Gaussian-like, which justifies the use of the
Gaussian assumption (9.112) in amplified spontaneous emission (ASE) noise-dominated scenario.
The constituent SISO decoding algorithms are based on modified Chase II decoding algorithm [2]:

1. Determine p least reliable positions starting from either (9.110) or (9.112). Generate 2p test patterns to be added to the hard-
decision word obtained after (3).
9.10 LDPC Codes 563

2. Determine the i-th (i ¼ 1,. . .,2p) perturbed sequence by adding (modulo-2) the test pattern to the hard-decision word
(on least reliable positions).
3. Perform the algebraic or hard decoding to create the list of candidate codewords. Simple syndrome decoding is suitable for
high-speed transmission.
4. Calculate the k-th candidate codeword ck LLRs by

X
n  
eð1vk ðiÞÞLðvk ðiÞÞ
L½vk = ðvk ð1Þ⋯vk ðnÞÞ ¼ log , ð9:113Þ
i¼1
1 þ eLðvk ðiÞÞ

where vk(i) denotes the i-th bit in the k-th candidate codeword and L() is a corresponding LLR.
5. Calculate the extrinsic bit reliabilities, denoted as Le(), for the next decoding stage using

Le ðvi Þ ¼ L0 ðvi Þ  Lðvi Þ, ð9:114Þ

where
"P #
0 vk ðiÞ¼0 Lðvk Þ
L ðvi Þ ¼ log P : ð9:115Þ
vk ðiÞ¼1 Lðvk Þ

In (9.115), summation in numerator (denominator) is performed over all candidate codewords having 0 (1) at position i.
6. Set L(vi) ¼ L0 (vi) and move to step 1.

The following “max-star” operator can be applied to (9.113) and (9.115) recursively as follows:

max ðx, yÞ≜ log ðex þ ey Þ ¼ max ðx, yÞ þ log 1 þ ejxyj : ð9:116Þ

Given this description of SISO constituent decoding algorithm, the TPC decoder, shown in Fig. 9.22, operates as follows. Let
Lerc,j denote the extrinsic information to be passed from row to column decoder, and let Lecr,j denote the extrinsic information to
be passed in opposite direction. Then, assuming that the column decoder operates first, the TPC decoder performs the
following steps:

0. Initialization: Lecr,i ¼ Lerc,i ¼ 0 for all i.

 
1. Column decoder: Run the SISO decoding algorithm described above with the following inputs L v j þ Lerc,j to obtain
 
{Lcolumn(vi)} and Lecr,i , as shown in Fig. 9.22. The extrinsic information is calculated by (9.114). Pass the extrinsic
n o
information Lecr,j to companion row decoder.
 
2. Row decoder: Run the SISO decoding algorithm with the following inputs Lðvi Þ þ Lecr,i to obtain {Lrow(vi)} and Lerc,i .
 
Pass the extrinsic information Lerc,i to companion column decoder.
3. Bit decisions: Repeat the steps 1 to 2 until a valid codeword is generated or a predetermined number of iterations have been
reached. Make the decisions on bits by sign[Lrow(uk)].

9.10 LDPC Codes

9.10.1 Binary LDPC Codes

Because LDPC codes belong to the class of linear block codes, they can be described as a k-dimensional subspace C of the n-
dimensional vector space of n-tuples, F n2 , over the binary field F2 ¼ GF(2). For this k-dimensional subspace, we can find the
basis B ¼ {g0, g1, . . ., gk  1} that spans C so that every codeword v can be written as a linear combination of basis vectors
v ¼ m0g0 + m1g1 + . . . + mk  1gk  1 for message vector m ¼ (m0, m1, . . ., mk  1); or in compact form, we can write v ¼ mG,
where G is so-called generator matrix with the i-th row being gi. The (n  k)-dimensional null space C⊥ of G comprises all
564 9 Advanced Coding and Coded Modulation Techniques

vectors x from F n2 such that xGT ¼ 0, and it is spanned by the basis B⊥ ¼ {h0, h1, . . ., hn  k  1}. Therefore, for each codeword
v from C, vhTi ¼ 0 for every i; or in compact form, we can write vHT ¼ 0, where H is so-called parity-check matrix whose i-th
row is hi.
An LDPC code can now be defined as an (n, k) linear block code whose parity-check matrix H has a low density of 1’s. A
regular LDPC code is a linear block code whose H-matrix contains exactly wc 1’s in each column and exactly wr ¼ wcn/
(n  k) 1’s in each row, where wc
n  k. The code rate of the regular LDPC code is determined by R ¼ k/n ¼ 1  wc/wr. The
graphical representation of LDPC codes, known as bipartite (Tanner) graph representation, is helpful in efficient description of
LDPC decoding algorithms. A bipartite (Tanner) graph is a graph whose nodes may be separated into two classes (variable
and check nodes) and where undirected edges may only connect two nodes not residing in the same class. The Tanner graph of
a code is drawn according to the following rule: check (function) node c is connected to variable (bit) node v whenever
element hcv in a parity-check matrix H is a 1. In an m  n parity-check matrix, there are m ¼ n  k check nodes and n variable
nodes. LDPC codes were invented by Robert Gallager (from MIT) in 1960, in his PhD dissertation [31], but received no
attention from the coding community until the 1990s [32]. As an illustrative example, consider the H-matrix of the Hamming
(7,4) code:
2 3
1 0 0 1 0 1 1
6 7
H¼4 0 1 0 1 1 1 0 5
0 0 1 0 1 1 1

For any valid codeword v ¼ [v0 v1. . .vn  1], the checks used to decode the codeword are written as

• Equation (c0): v0 + v3 + v5 + v6 ¼ 0 (mod 2)

• Equation (c1): v1 + v3 + v4 + v5 ¼ 0 (mod 2)
• Equation (c2): v2 + v4 + v5 + v6 ¼ 0 (mod 2)

The bipartite graph (Tanner graph) representation of this code is given in Fig. 9.23a.
The circles represent the bit (variable) nodes, while squares represent the check (function) nodes. For example, the variable
nodes v0, v3, v5, and v6 are involved in (c0) and therefore connected to the check node c0. A closed path in a bipartite graph
comprising l edges that closes back on itself is called a cycle of length l. The shortest cycle in the bipartite graph is called the
girth. The girth influences the minimum distance of LDPC codes, correlates the extrinsic LLRs, and therefore affects the
decoding performance. The use of large (high) girth LDPC codes is preferable because the large girth increases the minimum
distance and prevents early correlations in the extrinsic information during decoding process. To improve the iterative
decoding performance, we have to avoid cycles of lengths 4 and 6, which are illustrated in Fig. 9.23b, c, respectively.
The code description can also be done by the degree distribution polynomials μ(x) and ρ(x), for the variable node (v-node)
and the check node (c-node), respectively [2]:

X
dv X
dc
μ ð xÞ ¼ μd xd1 , ρð x Þ ¼ ρd xd1 , ð9:117Þ
d¼1 d¼1

where μd and ρd denote the fraction of the edges that are connected to degree-d v-nodes and c-nodes, respectively, and dv and
dc denote the maximum v-node and c-node degrees, respectively.
The most obvious way to design LDPC codes is to construct a low-density parity-check matrix with prescribed properties.
Some important LDPC designs, among others, include Gallager codes (semi-random construction) [31], MacKay codes
(semi-random construction) [32], combinatorial design-based LDPC codes [34], finite-geometry-based LDPC codes [68–70],
and array [also known as quasi-cyclic (QC)] LDPC codes [41], to mention a few. The generator matrix of a QC-LDPC code
can be represented as an array of circulant submatrices of the same size B indicating that QC-LDPC codes can be encoded in
linear time using simple shift-register-based architectures [35]. A QC-LDPC code can be defined as an LDPC code for which
every sectional cyclic shift to the right (or left) for l2[0,B  1] places of a codeword v ¼ [v0 v1 . . . vB  1] (each section vi
contains B elements) results in another codeword. Additional details on LDPC code designs are postponed for Sect. 9.10.5.
9.10 LDPC Codes 565

Fig. 9.23 (a) Bipartite graph of v0 v1 v2 v3 v4 v5 v6

Hamming (7,4) code described by
H-matrix above. Cycles in a
Tanner graph: (b) cycle of length Bit (variable) nodes
4 and (c) cycle of length 6

Check (function) nodes

c0 c1 c2
An example of irregular codes A cycle of length four

(a)
v0 v1 v0 v1
...
c0 1 ... 1
H ... ... ...
c1 1 ... 1
... c0 c1
(b)
v0 v1 v2 v0 v1 v2
...
c0 1 1
H c1 ... 1 1 ...
c2 1 1
... c0 c1 c2
(c)

9.10.2 Decoding of Binary LDPC Codes

The sum-product algorithm (SPA) is an iterative LDPC decoding algorithm in which extrinsic probabilities are iterated
forward and back between variable and check nodes of bipartite (Tanner) graph representation of a parity-check matrix [1, 2,
4, 5, 7, 36, 37, 65]. To facilitate the explanation of the various versions of the SPA, we use N(v) [N(c)] to denote the
neighborhood of v-node v (c-node c) and introduce the following notations:

• N(c) ¼ {v-nodes connected to c-node c}

• N(c)\{v} ¼ {v-nodes connected to c-node c except v-node v}
• N(v) ¼ {c-nodes connected to v-node v}
• N(v)\{c} ¼ {c-nodes connected to v-node v except c-node c}
• Pv ¼ Pr(v ¼ 1|y)
• E(v): the event that the check equations involving variable node v are satisfied
• M(c0 )\{c} ¼ {messages from all c’-nodes except node c}
• qvc(b) ¼ Pr(v ¼ b | E(v),y,M(c0 )\{c})
• M(v0 )\{v} ¼ {messages from all v0 -nodes except node v}
• rcv(b) ¼ Pr(check equation c is satisfied | v ¼ b, M(v0 )\{v})
• Lvc: the extrinsic likelihood to be sent from v-node v to c-node c
• Lcv: the extrinsic likelihood to be sent from c-node c to v-node v

We are interested in computing the APP that a given bit in a transmitted codeword v = [v0 v1 . . . vn  1] equals 1, given the
received word y ¼ [y0 y1 . . . yn  1]. Let us focus on decoding of the variable node v and we are concerned in computing LLR:
566 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.24 Illustration of half-

iterations of the SPA: (a)
c0 c0 c1 c2
subgraph of bipartite graph
corresponding to the H-matrix
with the 0-th row [1 1 0 0 1 1
0. . .0], (b) subgraph of bipartite
graph corresponding to the H-
matrix with the 0-th column [1 1 v0
1 0. . .0]

v0 v1 v4 v5 y0 (channel sample)

(a) (b)

c c’
c
qvc(b) rc’v(b)

qv’c(b) v
rcv(b)

v v’ y
(a) (b)

Fig. 9.25 Illustration of calculation of (a) extrinsic information to be passed from node c to variable node v, (b) extrinsic information (message) to
be passed from variable node v to check node c regarding the probability that v ¼ b, b2{0,1}

 
Prðv ¼ 0jyÞ
LðvÞ ¼ log , v 2 fv0 , ⋯, vn1 g: ð9:118Þ
Prðv ¼ 1jyÞ

The SPA, as indicated above, is an iterative decoding algorithm based on Tanner graph description of an LDPC code. We
interpret the v-nodes as one type of processors and c-nodes as another type of processors, while the edges as the message paths
for LLRs.
The subgraph illustrating the passing of messages (extrinsic information) from c-node to v-nodes in the check-node update
half-iteration is shown in Fig. 9.24a. The information passed is the probability that parity-check equation c0 is satisfied. The
information passed from node c0 to node v1 represents the extrinsic information it had received from nodes v0, v4, and v5 on the
previous half-iteration. We are concerned in calculating the APP that a given bit v equals 1, given the received word y and the
fact that all parity-check equations involvinghbit v are satisfied;
i with this APP is denoted as Pr(v ¼ 1| y, E(v)). Instead of APP
Prðv¼0jy, E ðvÞÞ
we can use log-APP or LLR defined as log Prðv¼1jy, EðvÞÞ :
In the v-node update half-iteration, the messages are passed in opposite direction – from v-nodes to c-nodes, as depicted in
subgraph of Fig. 9.24b. The information passed concerns log[Pr(v0 ¼ 0|y)/Pr(v0 ¼ 1|y)]. The information being passed from
node v0 to node c2 is information from the channel (via y0) and extrinsic information node v0 had received from nodes c0 and c1
on a previous half-iteration. This procedure is performed for all v-node/c-node pairs.
The calculation of the probability that c-th parity-check equation is satisfied given v ¼ b, b2{0,1}, denoted as rcv(b), is
illustrated in Fig. 9.25a. On the other hand, the calculation of the probability that v ¼ b given extrinsic information from all
check nodes, except node c, and given channel sample y, denoted as qvc(b), is illustrated in Fig. 9.25b.
In the first half-iteration, we calculate the extrinsic LLR to be passed from node c to variable node v, denoted as Lvc ¼ L(rcv),
as follows:

  " X #
Lcv ¼ ∏0 αv0 c ϕ ϕðβv0 c Þ ; α ¼ sign ½L , β ¼ jL j ð9:119Þ
vc vc vc vc
v v0 2NðcÞ∖v

where with ϕ(x) we denoted the following function:

9.10 LDPC Codes 567

Fig. 9.26 The plot of function ϕ(x)

  x 
x e þ1
ϕðxÞ ¼  log tanh ¼ log x , ð9:120Þ
2 e 1

which is plotted in Fig. 9.26. In the second half-iteration, we calculate the extrinsic LLR to be passed from variable node v to
function node c regarding the probability that v ¼ b, denoted as Lvc ¼ L(qvc), as
X
Lvc ¼ Lðqvc Þ ¼ Lch ðvÞ þ Lðr c0 v Þ: ð9:121Þ
c0 2N ðvÞ∖c

For the derivation of Eqs. (9.119) and (9.121), an interested reader is referred to [37].
Now we can summarize the log-domain SPA as follows:

0. Initialization: For v ¼ 0,1,. . .,n  1, initialize the messages Lvc to be sent from v-node v to c-node c to channel LLRs Lch(v),
namely, Lvc ¼ Lch(v).
1. c-node update rule: For c ¼ 0,1,. . .,n  k  1, compute Lcv ¼ þ Lvc : The box-plus operator is defined by
N ðcÞ∖fvg

!
2 X
2
L1 þ L2 ¼ ∏ signðLk Þ  ϕ ϕðjLk jÞ ,
k¼1 k¼1

where ϕ(x) ¼ log tanh (x/2). The box operator for |N(c)\{v}| components is obtained by recursively applying
two-component version defined above.
P
2. v-node update rule: For v ¼ 0,1,. . .,n  1, set Lvc ¼ Lch ðvÞ þ Lcv for all c-nodes for which hcv ¼ 1.
N ðvÞ∖fcg
P
3. Bit decisions: Update L(v) (v ¼ 0,. . .,n  1) by LðvÞ ¼ Lch ðvÞ þ Lcv and set bv ¼ 1 when L(v) < 0 (otherwise, bv ¼ 0).
N ðvÞ
T
If bvH ¼ 0 or predetermined number of iterations has been reached, then stop; otherwise go to step 1.

Because the c-node update rule involves log and tanh functions, it is computationally intensive, and there exist many
approximations. The very popular is the min-sum-plus-correction-term approximation [39]. Namely, it can be shown that
“box-plus” operator þ can also be calculated by

2
L1 þ L2 ¼ ∏ signðLk Þ  min ðjL1 j, jL2 jÞ þ cðx, yÞ, ð9:122Þ
k¼1

where c(x,y) denotes the correction factor defined by

568 9 Advanced Coding and Coded Modulation Techniques

   
cðx, yÞ ¼ log 1 þ ejxþyj  log 1 þ ejxyj , ð9:123Þ

commonly implemented as a look-up table (LUT). Given the fact that |c(x,y)| < 0.693, very often this term can be ignored.
Alternatively, the following approximation can be used:
8
< d, j x  y j< 2 \ j x þ y j> 2 j x  y j
>
cðx, yÞ ’ d, j x þ y j< 2 \ j x  y j> 2 j x þ y j ð9:124Þ
>
:
0, otherwise

with typical d being 0.5.

Another popular decoding algorithm is min-sum algorithm in which we simply ignore the correction term in (9.123).
Namely, the shape of ϕ(x), shown in Fig. 9.26, suggests that the smallest βvc in the summation (9.119) dominates and we can
write:
2 3
X   
ϕ4 ϕðβv0 c Þ5 ffi ϕ ϕ min
0
β 0
vc ¼ min
0
β v0 c : ð9:125Þ
v v
v0 2N ðcÞ∖v

Therefore, the min-sum algorithm is thus the log-domain algorithm with step 1 replaced by
!
Lcv ¼ ∏ α c0 v  min
0
β v0 c : ð9:126Þ
v0 2N ðcÞ∖v v

9.10.3 FPGA Implementation of Binary LDPC Decoders

In this section, we present an FPGA-based partial-parallel implementation of QC-LDPC decoder. The decoder of girth-10
QC-LDPC (34635, 27708) with column weight Wc ¼ 3, row weight Wr ¼ 15, and submatrix size B ¼ 2309 has been
implemented by using BEEcube system BEE4. This rapid prototyping FPGA system contains four fully integrated Xilinx
Virtex-6 FPGAs. Four LDPC decoders have been implemented per single FPGA, resulting in total 16 LDPC decoders per
BEE4 system.
As shown in Fig. 9.27a, the implementation consists of three types of memories: Mem, MemInit, and MemBit. The Wc  Wr
Mem units store the messages to be interchanged between variable nodes and check nodes, Wr MemInit units store the initial
LLRs from the channel, and Wr MemBit units store the decoded bits. There are Wr variable node processors (VNPs) and Wc
check node processors (CNPs) in our implementation. While the VNPs read the extrinsic information from Mem and prior
information from MemInit and write the updated information to the corresponding addresses of Mem and MemBit in the first
half-iteration, the CNPs read the information from Mem and MemBit and write the updated extrinsic information to the
corresponding addresses of Mem in the second half-iteration. Decoding stops when either the predefined maximum number of
iterations has been reached or decoder outputs a codeword.
With above described architecture, the lowest decoding throughput per decoder and FPGA can be expressed as follows:

N  f clock
Throughput ¼ ’ 143 Mb=s
B þ ðB þ B þ δÞ  Iter

where N ¼ 34635 is the codeword length, fclock ¼ 200 MHz is the clock frequency, B ¼ 2309 is the size of the submatrix,
Iter ¼ 20 is the maximum number of iterations, and δ ¼ 6 is the critical path delay of the processing unit. The first B clock
cycles used to initialize the MemInit, Wr VNPs need B clock cycles to process Wr  B variable nodes, while Wc CNPs need
B clock cycles to process Wc  B check nodes in each iteration. The aggregate minimum throughput of 16 LDPC decoders
9.10 LDPC Codes 569

Fig. 9.27 (a) Implementation

architecture of FPGA-based
partial-parallel LDPC decoder and CNP 3 MemInit1 MemInit2 … MemInit15 VNP 15
(b) BER performance comparison
of simulated and FPGA-based
LDPC decoder. CS computer
Mem1,1 Mem1,2 … Mem1,15
simulations, FPGA FPGA results,

…
MS min-sum algorithm, MSC CNP 2 Mem2,1 Mem2,2 … Mem2,15
min-sum with correction term
algorithm, MSS min-sum Mem3,1 Mem3,2 … Mem3,15
algorithm with scaling
VNP 2

CNP 1 MemBit1 MemBit2 … MemBit15 VNP 1

(a)
-1
10
CS-MS
10
-3 CS-MSS
CS-MSC
-5 FPGA-MS
10
FPGA-MSS
Bit-error rate, BER

FPGA-MSC
-7
10

-9
10

-11
10

-13
10

-15
10
1.6 2.0 2.4 2.8 3.2
Signal-to-noise ratio, SNR [dB] (per bit)
(b)

implemented on BEE4 system is ffi143 Mb/s  4  4 ¼ 2.288 Gb/s. Since the average number of iterations until a valid
codeword is reached is 8.8, the average total throughput is actually higher.
Figure 9.27b presents the BER performances of the FPGA-based min-sum algorithm (MS), min-sum with scaling factor
(MSS), and min-sum with correction term (MSC). The computer simulation (CS) results are provided as well for verification.
In practice, it is also possible to implement other decoding algorithms to reduce the complexity, described above. Here, the
input LLRs and extrinsic information are represented by 8 soft-decision bits (1 sign bit, 5 integer bits, and 2 fractional bits). It
can be observed that there is no error floor down to 1015, which is consistent with ref. [55], which claims that girth-10 LDPC
codes do not exhibit error floor phenomenon in the region of interest for optical communications (down to 1015). The net
coding gain (NCG) at BER of 1015 is 11.8 dB. The good BER performance of our QC-LDPC code is contributed to the large
minimum distance and large girth. For short LDPC codes, error floor can be suppressed by (i) BCH with corresponding
interleaver or RS codes [56, 57] as outer code in concatenation fashion or (ii) post-processing techniques [58].
570 9 Advanced Coding and Coded Modulation Techniques

9.10.4 Decoding of Nonbinary LDPC Codes

The q-ary sum-product algorithm (qSPA), where q is the cardinality of the finite field, and a low-complexity version via fast
Fourier transformation (FFT-qSPA) were proposed for decoding nonbinary LDPC codes [59, 60]. Log-domain qSPA
(Log-qSPA) was presented in [61] where multiplication is replaced by addition. Later, a mixed-domain version of the
FFT-qSPA (MD-FFT-qSPA) decoder was proposed [62], where large-size look-up tables are required for the purpose of
domain conversion. To avoid the complicated operations involved in the aforementioned decoding algorithms, extended min-
sum algorithm [63] and min-max algorithm [64] were widely adopted.
Following the same notation used in previous section, let Lch, v(a), Lvc(a), Lcv(a), Lk,l
cv and Lv(a) represent the prior
information of v-node v, the message from v-node v to c-node c, the message from c-node c to v-node v, and the APP value
concerning symbol a 2 GF(q), respectively. The nonbinary min-max algorithm can be summarized as follows:

0. Initialization: For v ¼ 0,1,. . .,n  1, initialize the messages Lvc(a) to be sent from v-node v to c-node c to channel LLRs Lch,
v(a), namely,Lvc(a) ¼ Lch, v(a).
 

1. c-node update rule: For c ¼ 0,1,. . .,n  k  1, compute Lcv ðaÞ ¼ min a0 0 2I ðcjav Þ max v0 2N ðcÞ∖v Lvc a0v0 , where I(c| av)
P v

denotes the set of codewords such that v0 2N ðcÞ∖v hcv0 a0v0 ¼ hcv av. Here hcv denotes the c-th row, v-th column element of the
parity-check matrix.
2. v-node update rule: For v ¼ 0,1,. . .,n  1, set Lvc(a) ¼ Lch, v(a) + ∑N(v)\cLcv(a) for all c-nodes for which hcv 6¼ 0.
3. Post-processing: Lvc ðaÞ ¼ Lvc ðaÞ  min a0 2GFðqÞ Lvc ða0 Þ, it is necessary for numerical reasons to ensure the non-divergence
of the algorithm.
4. Symbol decisions: Update Lv(a) (v ¼ 0,. . .,n  1) by Lv(a) ¼ Lch, v(a) + ∑N(v)Lcv(a), and set bv ¼ arg min a2GF ðqÞ ðLv ðaÞÞ: If
bvH T ¼ 0 or predetermined number of iterations has been reached, then stop; otherwise go to step 1.

The BER vs. SNR performances of the decoding algorithms are presented in Fig. 9.28.
All the decoders have been simulated with floating point precision for GF(4) (8430, 6744) nonbinary LDPC code over
binary input AWGN channel. The maximum number of iteration Imax is set to 20. The two decoding algorithms, Log-QSPA
and MD-FFT-qSPA, achieve the best performance as they are based on theoretical derivations, while the reduced-complexity

Fig. 9.28 BER performance of

NB-LDPC code over BI-AWGN
9.10 LDPC Codes 571

decoding algorithms, min-max, max-log, and min-sum algorithms, face the performance degradation at BER of 107 by
~0.13 dB, ~0.27 dB, and ~ 0.31 dB, respectively.
The details of corresponding FPGA implementation of nonbinary LDPC decoders can be found in [66, 67]. In the same
papers, the FPGA resource utilization has been discussed in detail. For instance, the reduced-complexity nonbinary decoding
algorithms and finite precision introduce the error floor phenomenon, which has been successfully eliminated by an outer
RS code.

9.10.5 Design of LDPC Codes

The most obvious way to design LDPC codes is to construct a low-density parity-check matrix with prescribed properties.
Some important designs among others include (i) Gallager codes (semi-random construction) [31], (ii) MacKay codes (semi-
random construction) [32], (iii) finite-geometry-based LDPC codes [68–70], (iv) combinatorial-design-based LDPC codes
[34], (v) quasi-cyclic (array, block-circulant) LDPC codes [2, 41], (vi) irregular LDPC codes [71], and (vii) Tanner codes
[22]. In this section we describe briefly several classes of LDPC codes including Gallager, Tanner, MacKay, and quasi-cyclic
LDPC codes.

9.10.5.1 Gallager Codes

The H-matrix for Gallager code has the following general form:

H ¼ ½ H1 H2 ⋯ H wc  T ð9:127aÞ

where H1 is pxp∙wr matrix of row weight wr, and Hi are column-permuted versions of H1-submatrix. The row weight of H is
wr, and column weight is wc. The permutations are carefully chosen to avoid the cycles of length 4. The H-matrix is obtained
by computer search.

9.10.5.2 Tanner Codes and Generalized LDPC (GLDPC) Codes

In Tanner codes [22], each bit node is associated with a code bit, and each check node is associated with a subcode whose
length is equal to the degree of node, which is illustrated in Fig. 9.29. Notice that so-called generalized LDPC codes [72–78]
were inspired by Tanner codes.
Let us consider the following example:

Fig. 9.29 An illustration of Sub-code m (nm,km) described

Tanner codes Check nodes
by H-matrix Hm
Degree n1 Degree n2 Degree nm

Interleaver

Bit nodes
572 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.30 Illustration of construction of LZ-GLDPC codes

The Tanner code design in this example is performed in two stages. We first design the global code by starting from an
identity matrix Im/2 and replace every nonzero element with n1 ones and every zero element by n1 zeros. The lower submatrix
is obtained by concatenating the identity matrices In1. In the second stage, we substitute all-one row vectors (of length n1) by
the parity-check matrix of local linear block code (n1,k1), such as Hamming, BCH, or RM code. The resulting parity-check
matrix is used as the parity-check matrix of an LDPC code.
The GLDPC codes can be constructed in similar fashion. To construct a GLDPC code, one can replace each single parity-
check equation of a global LDPC code by the parity-check matrix of a simple linear block code, known as the constituent
(local) code, and this construction is proposed by Lentmaier and Zigangirov [77], and we will refer to this construction as
LZ-GLDPC code construction. One illustrative example is shown in Fig. 9.30. In another construction proposed by Boutros
et al. in [76], referred here as B-GLDPC code construction, the parity-check matrix, H, is a sparse matrix partitioned into
W submatrices H1,. . .,HW. The H1-submatrix is a block-diagonal matrix generated from an identity matrix by replacing the
ones by a parity-check matrix H0 of a local code of codeword length n and dimension k:
2 3
H0 0
6 H0 7
T 6 7
H ¼ ½H T1 H T2 . . . H TW  , H 1 ¼ 6 7 ð9:127bÞ
4 ... 5
0 H0

Each submatrix Hj in (9.127) is derived from H1 by random column permutations. The code rate of a GLDPC code is lower
bounded by

R ¼ K=N  1  W ð1  k=nÞ, ð9:128Þ

where K and N denote the dimension and the codeword length of a GLDPC code, W is the column weight of a global LDPC
code, and k/n is the code rate of a local code (k and n denote the dimension and the codeword length of a local code). The
GLDPC codes can be classified as follows [72]: (i) GLDPC codes with algebraic local codes of short length, such as Hamming
codes, BCH codes, RS codes, or Reed-Muller codes, (ii) GLDPC codes for which the local codes are high-rate regular or
irregular LDPC codes with large minimum distance, and (iii) fractal GLDPC codes in which the local code is in fact another
GLDPC code. For more details on LZ-GLDPC and B-GLDPC like codes and their generalization-fractal GLDPC codes
(a local code is another GLDPC code), an interested reader is referred to [33, 37] (and references therein).

9.10.5.3 MacKay Codes

Following MacKay (1999) [32] below are listed several ways to generate the sparse matrices in order of increasing algorithm
complexity (not necessarily improved performance):

1. H-matrix is generated by starting from an all-zero matrix and randomly inverting wc (not necessarily distinct bits) in each
column. The resulting LDPC code is an irregular code.
2. H-matrix is generated by randomly creating weight-wc columns.
9.10 LDPC Codes 573

3. H-matrix is generated with weight-wc columns and uniform row weight (as near as possible).
4. H-matrix is generated with weight-wc columns, weight-wr rows, and no two columns having overlap larger than one.
5. H-matrix is generated as in (4), and short cycles are avoided.
6. H-matrix is generated as in (5) and can be represented H ¼ [H1|H2], where H2 is invertible or at least has a full rank.

The construction via (5) may lead to an H-matrix that is not of full rank. Nevertheless, it can be put in the following form by
column swapping and Gauss-Jordan elimination:
" #
PT I
H¼ , ð9:129Þ
0 0

and by eliminating the all-zero submatrix, we obtain the parity-check matrix of systematic LDPC code:

e ¼ PT
H I : ð9:130Þ

Among various construction algorithms listed above, the construction (5) will be described with more details.

Outline of Construction Algorithm (5)

1. Choose code parameters n, k, wc, wr, and g (minimum cycle length girth). The resulting H-matrix will be an m x n (m ¼ n  k)
matrix with wc ones per column and wr ones per row.

2. Set the column counter to ic ¼ 0.

3. Generate a weight-wc column vector and place it in ic-th column of H-matrix.
4. If the weight of each row  wr, the overlap between any two columns 1, and all cycle lengths are g, then increment the
counter ic ¼ ic + 1.
5. If ic ¼ n, stop; else go to step 3.

This algorithm could take hours to run with no guarantee of regularity of H-matrix. Moreover, it may not finish at all, and
we need to restart the search with another set of parameters. Richard and Urbanke (in 2001) proposed a linear complexity in
length technique based on H-matrix [33].
An alternative approach to simplify encoding is to design the codes via algebraic, geometric, or combinatoric methods. The
H-matrix of those designs can be put in cyclic or quasi-cyclic form, leading to implementations based on shift registers and
mod-2 adders.

9.10.5.4 Large-Girth Quasi-Cyclic (QC) Binary LDPC Codes

Based on Tanner’s bound for the minimum distance of an LDPC code [22]
8 
> wc
<1 þ ðwc  1Þ⌊ðg2Þ=4c  1 , g=2 ¼ 2m þ 1
wc  2
d  ð9:131Þ
: 1 þ wc
> ðwc  1Þ⌊ðg2Þ=4c  1 þ ðwc  1Þ⌊ðg2Þ=4c , g=2 ¼ 2m
wc  2

(where g and wc denote the girth of the code graph and the column weight, respectively, and where d stands for the minimum
distance of the code), it follows that large girth leads to an exponential increase in the minimum distance, provided that the
column weight is at least 3. (⌊c denotes the largest integer less than or equal to the enclosed quantity.) For example, the
minimum distance of girth-10 codes with column weight r ¼ 3 is at least 10. The parity-check matrix of regular QC LDPC
codes [2, 10, 12] can be represented by
574 9 Advanced Coding and Coded Modulation Techniques

2 3
I I I ... I
6 S½0 7
6P PS½1 PS½2 . . . PS½c1 7
6 2S½0 7
H¼6
6P P2S½1 P2S½2 . . . P2S½c1 7,
7 ð9:132Þ
6 7
4 ... ... ... ... ... 5
Pðr1ÞS½0 Pðr1ÞS½1 Pðr1ÞS½2 . . . Pðr1ÞS½c1

where I is B  B (B is a prime number) identity matrix, P is B  B permutation matrix given by P ¼ ( pij)B  B, pi,i + 1 ¼ pB,1 ¼ 1
(zero otherwise), and r and c represent the number of block rows and block columns in (9.132), respectively. The set of
integers S are to be carefully chosen from the set {0,1,. . .,B  1} so that the cycles of short length, in the corresponding Tanner
(bipartite) graph representation of (9.132), are avoided. We have to avoid the cycles of length 2 k (k ¼ 3 or 4) defined by the
following equation:

S½i1  j1 þ S½i2  j2 þ ⋯ þ S½ik  jk ¼ S½i1  j2 þ S½i2  j3 þ ⋯ þ S½ik  j1 mod B, ð9:133Þ

where the closed path is defined by (i1,j1), (i1,j2), (i2,j2), (i2,j3), . . ., (ik,jk), (ik,j1) with the pair of indices denoting row-column
indices of permutation blocks in (2) such that lm 6¼ lm + 1, lk 6¼ l1 (m ¼ 1,2,..,k; l2{i,j}). Therefore, we have to identify the
sequence of integers S[i]2{0,1,. . .,B-1} (i ¼ 0,1,. . .,r-1; r < B) not satisfying Eq. (9.133), which can be done either by
computer search or in a combinatorial fashion. We add an integer at a time from the set {0,1,. . .,B  1} (not used before) to the
initial set S and check if Eq. (9.133) is satisfied. If Eq. (9.133) is satisfied, we remove that integer from the set S and continue
our search with another integer from set {0,1,. . .,B  1} until we exploit all the elements from {0,1,. . .,B  1}. The code rate
of these QC codes, R, is lower bounded by

jSjB  rB
R ¼ 1  r=jSj, ð9:134Þ
jSjB

and the codeword length is |S|B, where |S| denotes the cardinality of set S. For a given code rate R0, the number of elements
from S to be used is br/(1  R0)c. With this algorithm, LDPC codes of arbitrary rate can be designed.
As an illustration, by setting B ¼ 2311, the set of integers to be used in (2) is obtained as S ¼ {1, 2, 7, 14, 30, 51, 78,
104, 129, 212, 223, 318, 427, 600, 808}. The corresponding LDPC code has rate R0 ¼ 1–3/15 ¼ 0.8, column weight 3, girth
10, and length |S|B ¼ 15∙2311 ¼ 34,665. In the example above, the initial set of integers was S ¼ {1,2,7}, and the set of row to
be used in (6.25) is {1,3,6}. The use of a different initial set will result in a different set from that obtained above.

9.10.6 Rate-Adaptive LDPC Coding

In this section, we first describe the nonbinary (NB) irregular QC-LDPC code design derived from PBDs, introduced in [48],
defined over GF(q). Let a set V of size v represent a set of elements (points), with any subset being called a block. Then, a
pairwise balanced design PBD(v,K,δ) is defined as a collection of blocks of different sizes taken from set K, such that every
pair of points is contained in δ of the blocks. For additional details on PBDs, an interested reader is referred to [79]. The parity-
check matrix of irregular NB-QC-LDPC codes based on PBDs is given by [48]
2 3
α0 I ðb00 ÞI α1 I ðb10 ÞI ⋯ αc1 I ðbc1,0 ÞI
6 c1 7
6 α I ðb01 ÞPS½0 α0 I ðb11 ÞPS½1 ⋯ αc2 I ðbc1,1 ÞPS½c1 7
6 7
6 7
H ¼ 6 αc2 I ðb02 ÞP2S½0 αc1 I ðb12 ÞP2S½1 ⋯ αc3 I ðbc1,2 ÞP2S½c1 7, ð9:135Þ
6 7
6 7
4⋮ ⋮ ⋱ ⋮ 5
αcðr1Þ I ðb0,r1 ÞPðr1ÞS½0 αcrþ2 I ðb1,r1 ÞPðr1ÞS½1 ⋯ αcr I ðbc1,r1 ÞPðr1ÞS½c1

where I is B  B (B is a prime number) identity matrix, P is B  B permutation matrix given by P ¼ ( pij)B  B, pi,i + 1 ¼ pB,1 ¼ 1
(zero otherwise), and r and c represent the number of block rows and block columns in (9.135), respectively. In (9.135), {bij}
are points of the i-th block in PBD(r,K,δ), with the largest size of block k in set of sizes K satisfying the inequality k  r. αi are
nonzero elements of Galois field of size q, denoted as GF(q). Finally, the I(bij) denotes the indicator function, which has the
value 1 for the existing point within the i-th block and 0 for the non-existing point. Therefore, only those submatrices for
9.10 LDPC Codes 575

which indicator function is 1 will be preserved from template, regular, QC-LDPC code design. Notice the repetition of PBD
blocks in (9.135) is allowed. Given the fact that PBDs have regular mathematical structure that can be algebraically described,
the irregular NB-QC-LDPC codes derived from PBDs have the complexity comparable or lower to that of regular NB-QC-
LDPC code design. As an illustration, the irregular NB-QC-LDPC code derived from PBD(5, {3,2,1}, 1) ¼ {{0,1,3}, {1,2,4},
{1,2}, {0,4}, {3,4}, {0}, {1}} has the following form:
2 3
α0 I 0 0 α3 I 0 α5 I 0
6 6 7
6α I α0 PS½1 α1 PS½2 0 0 0 α5 PS½5 7
6 7
H¼6
60 α6 P2S½1 α0 P2S½2 0 0 0 0 7:
7
6 4 7
4α I 0 0 0 α1 P3S½4 0 0 5
0 α4 P4S½1 0 α6 P4S½3 α0 P4S½4 0 0

Since both the identity matrix and the power of permutation matrix have a single 1 per row, the block size of the i-th block
from PBD determines i-th block-column weight. In the example above, the first two block columns have column weight 3; the
third, fourth, and fifth have the column weight 2; and the last two block columns have weight 1. Notice that for GF
(4) ¼ {0,1,α,α2}, we have that α3 ¼ 1, α4 ¼ α, α5 ¼ α2, and α6 ¼ α3.
We now describe how to derive different classes of codes proposed for next generation of optical networks from (9.135).
Let the template H-matrix of (4,8) irregular QC code be given as
2 3
α0 I ðb00 ÞI α1 I ðb10 ÞI ⋯ α7 I ðb70 ÞI
6 α7 I ðb ÞI α0 I ðb11 ÞPS½1 ⋯ α6 I ðb71 ÞPS½7 7
6 01 7
H¼6 6 7:
4 α I ðb02 ÞI α7 I ðb12 ÞP2S½1 ⋯ α I ðb72 ÞP
5 2S½7 5

α I ðb03 ÞI
5
α6 I ðb13 ÞP3S½1 ⋯ α4 I ðb73 ÞP3S½7

By employing the following PBD, {{0}, {0,1}, {0,1,2}, {0,1,2,3},{1,2,3},{2,3},{3}}, we obtain the NB staircase LDPC
code:
2 3
I α1 I α2 I α3 I α4 I
6 PS½1 α1 PS½2 α2 PS½3 α3 PS½4 α4 PS½5 7
6 7
H¼6 7,
4 P2S½2 α1 P2S½3 α2 P2S½4 α3 P2S½5 α4 P2S½6 5
P3S½3 α1 P3S½4 α2 P3S½5 α3 P3S½6 α4 P3S½7

which represents the generalization of binary staircase-like codes. The corresponding spatially coupled LDPC code, derived
from the following PBD {{0,1,2},{1,2,3}}, is given as

The convolutional code can be obtained from the template code as follows:
2 3
I
6 α7 I PS½1 7
6 7
6 6 7
6α I α7 P2S½1 P2S½2 7
6 7
6 7
H ¼ 6 α5 I α6 P3S½1 α7 P3S½2 P3S½3 7:
6 7
6 α5 P4S½1 α6 P4S½2 α7 P4S½3 P4S½4 7
6 7
6 5 5S½2 6 5S½3 7
4 α P α P α7 P5S½4 P5S½5 5
⋱ ⋱ ⋱
576 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.31 The principles of rate-

adaptive LDPC coding with ,QIRUPDWLRQGDWD .0 .0 .0 5HHQFRGHGGDWD
re-encoding

VW FRGHZRUG 0 1.

QG FRGHZRUG 0

UG FRGHZRUG 0

'DWDWUDQVPLWWHG 10 10 10

Therefore, the staircase-like, spatially coupled-like, and some convolutional LDPC codes can be interpreted as different
arrangements derived from irregular QC-LDPC codes.
In the rest of this section, we discuss different strategies for rate adaptation. The code rate adaptation is performed by
partial reconfiguration of decoder based on (9.135), for a fixed codeword length, by changing the size of the permutation
matrix P while keeping the number of block rows constant, and/or by varying the number of employed block rows while
keeping P fixed. Another alternative to change code rate is to use the same regular parity-check matrix as template but employ
the PBDs corresponding to different code rates. It is also possible to perform the puncturing of parity symbols in original code
(n,k,d ) to obtain a linear block code (n  p,k,dp), dp  d, where p is the number of removed parity symbols. Notice that this
approach when applied to LDPC codes can introduce an early error floor phenomenon. This is the reason why the re-encoding
approach has been introduced in [12, 80] instead of using the conventional shortening or puncturing approach [81]. As shown
in Fig. 9.31, to encode, the adaptive LDPC encoder encapsulates the last M symbols of the proceeding codeword and the
incoming K  M information symbols into a K-symbol vector. In other words, each codeword is generated continuously by
re-encoding the last M symbols of its preceding codeword. (The first M symbols in the first codeword are set as known since
no proceeding codeword exists and the first M re-encoded symbols of each codeword are not transmitted.) Therefore, the
actual code rate is R0 ¼ (K  M )/(N  M ), where 0 < M < K and can be tuned easily in the range (0, R] (R is the code rate of
template code) by adjusting the re-encoded data size M. Notice that the template code should be systematic to leverage the
advantage of re-encoding in decoding.
The decoder employs the aforementioned nonbinary LDPC decoding algorithm [82]. Correspondingly, decoding of each
codeword utilizes both the LLRs of the M re-encoded symbols produced by the decoding of its preceding codeword and the
initial LLRs of the received K-M symbols. As LLRs of the M re-encoded symbols are converged LLRs and thus of higher
confidence, it helps improve the error correction capability of the LDPC code. With higher number of symbols re-encoded, the
lower code rate R0 or equivalently the stronger error correction capability can be obtained.
As an illustration, we provide Monte Carlo simulation results for an adaptive LDPC code based on the quasi-cyclic
nonbinary LDPC (12128, 10991) code [82] of girth 10 over GF(8) and code rate R ¼ K/N ¼ 0.906, for polarization-division
multiplexed (PDM) 8-QAM of aggregate date information rate over 100 Gb/s. The FFT-based 8-ary MD-FFT-SPA [82] is
employed in simulations as the decoding algorithm with 50 iterations. By choosing the proper parameter M, we have
constructed the LDPC codes of rate 0.875 and 0.833, the BER vs. OSNR performances of which along with that of the
mother code are shown in Fig. 9.32. By increasing the LDPC overhead from 10.3% to 14.3% and 20%, LDPC coding gain can
be raised by 0.45 and 0.85 dB, respectively, at BER¼108. The rate-adaptive coding has been used in Section IX.C to deal
with wavelength-dependent Q-factor degradation.

9.10.7 Rate-Adaptive Coding Implementations in FPGA

9.10.7.1 Rate-Adaptive LDPC Coding in FPGA

The basic rate adaptation via either shortening or puncturing is widely used everywhere in communication systems and can be
introduced in both block and convolutional codes. In this subsection, we use shortening to achieve rate-adaptive LDPC coding
since it can allow a wide range of rate adjustment with unified decoder architecture through a set of reconfigurable registers in
9.10 LDPC Codes 577

Fig. 9.32 BER performance of -1

10
polarization-division multiplexed NB-LDPC (12128,K) coded
rate-adaptive NB-LDPC codes -2
10 8-QAM:
R=0.906
-3
10 R'=0.875
R'=0.833

Bit-error rate, BER

-4
10
-5
10
-6
10
-7
10
-8
10
-9
10

4.5 5.0 5.5 6.0

Optical signal-to-noise ratio, OSNR [dB] (per bit)

Fig. 9.33 Architecture of rate-

adaptive binary LDPC decoder:
(a) overall schematic diagram, (b)
schematic diagram of CNU

FPGA [67]. Because of the quasi-cyclic structure of our nonbinary LDPC codes, we shorten the entire sub-block by adding the
least number of logics’ blocks. For example, we start from a (3,15)-regular nonbinary LDPC codes with rate of 0.8, and we
can obtain a class of shortened regular nonbinary LDPC codes with column weight and row weight of {(3,14), (3,13), (3,12),
(3,11), (3,10)}, which correspond to code rates of {0.786, 0.77, 0.75, 0.727, 0.7}.
As shown in Fig. 9.33a, the binary LDPC decoder consists of two major memory blocks (one stores channel LLR and
another stores a posteriori probability (APP) messages), two processing blocks (variable node unit (VNU) and check node unit
(CNU)), an early termination unit (ETU), and a number of mux blocks, wherein its selection of output signal can be software
reconfigurable to adjust the shortening length. The memory consumption is dominated by LLR message with size of n  WL
578 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.34 Architecture of rate-

adaptive nonbinary LDPC
decoder: (a) overall schematic
diagram, (b) schematic diagram
of CNU

Table 9.2 Logic resource utilization

Resources Binary LDPC decoder Nonbinary LDPC decoder
Occupied slices 2969 out of 74,400 (3%) 16,842 out of 74,400 (22%)
RAMB36E1 113 out of 1064 (10%) 338 out of 1064 (31%)
RAMB18E1 14 out of 2128 (1%) 221 out of 2128 (9%)

APP messages of size c  n  WR, where WL and WR denote the precisions used to represent LLR and APP messages. The
logic consumption is dominated by CNU, as shown in Fig. 9.33b. The ABS block first takes the absolute value of the inputs,
and the sign XOR array produces the output sign. In the two least minimums’ finder block, we find the first minimum value via
binary tree and trace back the survivors to find the second minimum value as well as the position of the first minimum value.
This implies that we can write three values and r sign bits back to the APP memories instead of r values. However, we will not
take advantage of memory reduction techniques for comparison in the following sections.
Similar to the rate-adaptive binary LDPC decoder architecture discussed above, the architecture of the layered min-max
algorithm (LMMA)-based nonbinary LDPC decoder is presented in Fig. 9.34a. There are four types of memories used in
implementation: memory Rc,v with size of c  n  q  WR stores the information from check nodes to variable nodes, memory
for Lv with size of n  q  WL stores the initial log-likelihood ratios, memory for bc with size n  log2q stores the decoded bits,
and memories inside each CNU store the intermediate values. The same notations are borrowed from previous subsection
except that q denotes the size of Galois field. As shown in Fig. 9.34b, it is obvious that CNU is the most complex part of the
decoding algorithm, which consists of r inverse permutators, r BCJR-based min-max processors and r permutators, and two
types of the first-in-first-out (FIFO) registers. The inverse permutator block shifts the incoming message vector cyclically. The
first FIFO is used to perform the parallel-to-serial conversion. After the min-max processor, which is implemented by
low-latency bidirectional BCJR algorithm, the processed data is fed into another FIFO block performing serial-to-parallel
conversion, followed by the permutator block. Because of high complexity of CNU design and high memory requirements of
nonbinary decoder than that of binary decoder, reduced-complexity architectures and selective version of MMA can be further
exploited.
We compare three rate-adaptive schemes based on binary LDPC codes and nonbinary LDPC codes. These three
architectures can be software-defined by initializing configurable registers in FPGA. The resource utilization is summarized
in Table 9.2. One can clearly notice that the LMMA-based nonbinary LDPC codes consume 3.6 times larger memory size than
the binary one because of large field size and high quantization precision, while the occupied number of slices is five times
larger than that in binary case because of higher complexity of CNU.
9.10 LDPC Codes 579

Fig. 9.35 BER performance of

rate-adaptive binary LDPC
decoder implemented in FPGA

Fig. 9.36 BER performance of

rate-adaptive nonbinary LDPC
decoder implemented in FPGA

The BER vs. Q-factor performances of the rate-adaptive binary and nonbinary LDPC code are presented in Figs. 9.35 and
9.36. The FPGA-based emulation was conducted over binary (BI)-AWGN channel, and 6 and 8 bits precision are used in
binary and nonbinary LDPC decoder, respectively. A set of column weight and row weight configurations of {(3, 15), (3, 14),
(3, 13), (3, 12), (3, 11), (3, 10)}, which corresponds to the code rates of {0.8, 0.786, 0.77, 0.75, 0.727, 0.7}, can be achieved by
580 9 Advanced Coding and Coded Modulation Techniques

software-based reconfiguration of specific register in FPGA. The girth-10 regular (34635, 27710, 0.8) binary and nonbinary
mother code can achieve a Q-limit of 5.2 and 5.14 dB at BER of 1015, which corresponds to NCG of 11.83 and 11.89 dB.
The rate-adaptive nonbinary LDPC codes outperform the binary LDPC codes by approximated 0.06 dB in all range of rate
from 0.7 ~ 0.8. In addition, we believe this gap will be larger when combined with higher modulation schemes enabling
100Gbits/s (with QPSK) and 400Gbits/s (with 16-QAM) optical communication systems.

9.10.7.2Rate-Adaptive GLDPC Coding in FPGA

The key idea behind the GLDPC codes [72–78], as discussed earlier, is to replace the parity-check equations in a parity-check
matrix of a global LDPC code by a linear block code. The decoding is based on several low-complexity soft-input/soft-output
(SISO) linear block decoders operating in parallel. The SISO decoders are commonly implemented based on maximum a
posteriori probability (MAP) decoders, and as such they provide accurate estimates of bit reliabilities for a global LDPC
decoder after small number of iterations. Due to high complexity of the BCJR decoder, the GLDPC coding is limited to simple
linear block component codes such as Hamming, binary BCH, and Reed-Muller (RM) codes. To keep the code rate
reasonably high, the global code of GLDPC code must be of column weight (variable node degree) 2. Unfortunately, the
global codes with parity-check matrix of column weight 2 are known as weak LDPC codes as they exhibit an early error floor
phenomenon. In principle, multiple local codes can be used instead of parity checks of a global code. The proper selection of
these local codes can lead to further improvement of BER performance. However, the high-code rate requirement problem
important for certain applications, such as optical communications and magnetic recording, will not be solved. As solution to
high-code rate requirement, while employing global codes of column weight 3, instead of 2 used in [72, 73], here we describe
the possibility not to replace all parity checks by local codes, but only certain, selected, number of parity checks. Such
obtained GLDPC codes will contain two types of so-called super-nodes, simple parity-check equations for which MAP
decoders are trivial and selected number of reasonable-complexity linear block codes (not necessary identical ones) [74]. This
type of GLDPC codes shows remarkable flexibility for code rate adaptation as it will be shown later in the subsection.
For instance, if only single-parity-check codes and (n,k) local code are used in regular GLDPC code design, the
corresponding code rate R can be bounded as follows [74]:
 
W W kþ1
1  1  R  RG , ð9:136Þ
n d n

where W is the column weight of the global regular code and parameter d denotes that every d-th row in a global code is
replaced by (n,k) local code, while the remaining rows from the global code are interpreted as single-parity-check codes in
GLDPC code. In (9.136), RG denotes the code rate of the global regular code, which is RG ¼ 1 2 W/n. In this subsection, for
global code we chose large-girth quasi-cyclic LDPC code, thanks to its simplicity for FPGA implementation.
For the sake of completeness of the presentation, we provide the layered decoding algorithm of the GLDPC codes [75],
which is suitable for hardware implementations. This algorithm is applicable to any linear block code to be used as the local
code. Let Rðc,v
k,lÞ ðk,lÞ
, Lv,c , Lv represent the check c to variable v extrinsic information, the variable v to check c at k-th iteration and
l-th layer extrinsic information, and the log-likelihood ratio (LLR) from the channel of variable (bit) node v, respectively,
where k ¼ 1,2,. . .,Imax and l ¼ 1,2,. . .,c. The layered min-max algorithm is adopted for FPGA implementation, whose data
flow can be summarized as follows:

0. Initialization step:

Rðcvk,lÞ ¼ 0, k ¼ 1; l ¼ 1, 2, . . . , c ð9:137Þ

1. Bit decision step:

(
X 0 0, Lðvk,lÞ  0
Lðvk,lÞ ¼ Lv þ Rðk,l Þ ,
l0 cv
bcv ¼ ð9:138Þ
1, Lðvk,lÞ < 0
9.10 LDPC Codes 581

2. Variable node processing rule:

X 
0 0 k  1, l < l0
Lðvck,lÞ ¼ Lv þ Rðk ,l Þ , k0 ¼
l0 6¼l cv
ð9:139Þ
k, l  l0

3. Check node processing rule:

• If it is a local code: BCJR-based APP updating rule is applied:

Rðcvk,lÞ ¼ BCJR Lðvck,lÞ : ð9:140Þ

• Else: scaled min-sum updating rule is applied:

  
ðk,lÞ  ðk,lÞ 
Rðc,v
k,lÞ
¼ a∏v0 6¼v sign Lv0 ,c min
0
 Lv0 ,c , ð9:141Þ
v 6¼v

where a is the attenuation factor.

The criterion of early termination is achieved when decoded bits converge to transmitted codeword, since only 1/c portion
of check nodes involved in each layer.
The implementation of BCJR-based MAP decoder from Eq. (9.140) can be divided into three parts. As shown in Fig. 9.37,
the first part calculates forward and backward recursion likelihoods, the second part corresponds to the memories storing the
intermediate data α and β, and the third part is a combiner calculating the output. Since the trellis derived from a block code is
time-variant, it implies that a selection signals should be pre-stored in ROM so that we can select an appropriate feedback
output to the input of forward and backward recursion blocks. Aiming to keep reasonable complexity and latency, we replace
the max-star operation by max operation and adopt a bidirectional recursion scheme, which minimizes the memory sizes for α
and β as well as the latency. To be more specific, for illustrative purposes, Fig. 9.38 shows the timing diagram of decoding of

Fig. 9.37 BCJR-based APP

decoder architecture Sum-Max
Forward Forward
Selector Recursion

Mem Combiner
Sum-Max
Forward Forward
Selector Recursion

I0 I1 I2 I3 I4 I5 I6 I7 I8 I9 I10 I11 I12 I13 I14

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14

I14 I13 I12 I11 I10 I9 I8 I7 I6 I5 I4 I3 I2 I1 I0

15 14 13 12 11 10 9 8 7 6 5 4 3 2 1

O7 O6 O5 O4 O3 O2 O1 O0

O8 O9 O10 O11 O12 O13 O14

Fig. 9.38 Timing diagram of the BCJR processor for Hamming local code
582 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.39 BER

performance vs. Q-factor with
maximum number of iterations set
to 10

the (15, 11) local Hamming code. The first half of the forward and backward recursion is updated and stored in memories, and
current α7 and current β8 are combined to the output O7. After that, each cycle will generate two outputs, one is obtained by
combining current α with β from memory blocks, while the other one is obtained by combining the current β with α from the
memory. This technique reduces the total latency to the length of local codes plus the latency due to parallel-to-serial
conversion of the input and serial-to-parallel conversion of the output. In summary, the complexity of MAP decoder is
reasonably low and thus makes the proposed GLDPC code very promising.
To prove the advantages of the proposed rate-adaptive GLDPC codes, we start from a well-designed LDPC code, which is
a quasi-cyclic (3, 15)-regular, girth-10 (34635, 27710, 0.8) binary LDPC code and choose a simple (15, 11) Hamming code as
component code. For the purpose of ease of implementation, we sweep different parameters d in the range of {1, 127, 63, 31,
15}, which corresponds to the code rates of {0.8, 0.7953, 0.7906, 0.7807, 0.7601}. The precision of LLRs, variable-to-check
message, and check-to-variable message are set to 5-bit, 5-bit, and 6-bit, respectively; and the maximum number of iterations
is set to either 10 or 15. The bit error rate (BER) performances vs. Q-factor are present in Fig. 9.39 with 10 iterations and
Fig. 9.40 with 15 iterations. The net coding gain of the designed mother is 11.61 dB and 11.71 dB for 10 and 15 iterations,
which demonstrates its fast convergence. One can clearly observe from the figure that the BER performance is enhanced as
d decreases; thus fine-tuning of code rate can be achieved.

9.11 Coded Modulation and Unequal Error Protection

Combining channel coding and modulation into a single entity is commonly referred to as the coded modulation [1, 11, 12, 48,
49, 51, 82, 83]. The coded modulation concept was introduced by Massey in 1974 [84]. When channel coding in coded
modulation is based on block codes, we refer to such scheme as block-coded modulation (BCM) [85–92]. On the other hand,
when coded modulation is based on convolutional codes, we refer to such a scheme as trellis-coded modulation (TCM)
[93]. The key idea behind the coded modulation is to expand the signal constellation to obtain the redundancy for channel
coding and then use FEC to increase the minimum Euclidean distance between sequences of modulated signals. One very
popular version of BCM was introduced by Imai and Hirakawa [85], and this BCM scheme is well known as multilevel coding
(MLC) scheme. In MLC we map codewords to bit positions, with the help of binary partition. For 2m-ary signal constellation,
9.11 Coded Modulation and Unequal Error Protection 583

Fig. 9.40 BER

performance vs. Q-factor with
maximum number of iterations set
to 15

we use m different error correcting codes, one per each label. On receiver side, we perform so-called multistage decoding, in
which the LLRs from higher decoding stage, corresponding to stronger code, are forwarded to lower decoding stage,
corresponding to a weaker code. When the same code is used on all levels, we refer to such coded modulation scheme as
bit-interleaved coded modulation (BICM) scheme, which was introduced by Caire et al. [89]. In TCM, introduced by
Ungerboeck in [93], we apply a natural mapping of bits to symbols through so-called set partitioning. Given a trellis
description of corresponding code, we assign symbol sequences to the trellis paths. On receiver side, we perform the Viterbi
decoding of trellis describing the trellis code. It is also possible to trade the constellation expansion with bandwidth expansion
in so-called hybrid coded modulation Scheme [12, 50].

9.11.1 Coded Modulation Fundamentals

Let dunc denote the minimum Euclidean distance in original, uncoded signal constellation diagram, and let dmin denote the
minimum distance in sequence of symbols after coding. When the uncoded normalized average energy, denoted as Eunc, is
different from the normalized average energy after coding, denoted as Ecoded, the asymptotic coding gain can be defined as
follows:

Eunc =d 2unc E unc d2min

Ga ¼ ¼  2
Ecoded =d2min E |fflfflcoded d unc
{zfflffl} |{z} ð9:142Þ
GC GD
¼ GC  GD ,

where GC is the constellation expansion factor and GD is the increased distance gain. The generic coded modulation scheme,
applicable to both BCM and TCM, is illustrated in Fig. 9.41, which is based on Forney’s interpretation [94, 95] (see also
[96, 98]). Two-dimensional (2-D) modulator such as M-ary QAM (I/Q) modulator or M-ray PSK modulator is used to impose
the sequence of symbols to be transmitted. The N-dimensional lattice is used for code design. A lattice represents a discrete set
of vectors in real Euclidean N-dimensional space, which forms a group under ordinary vector addition, so the sum or
584 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.41 Generic coded One point

modulation scheme selected
One point out of
Additional n-k selected out of 2(n+r) / (N/2) Two-
info bits Signal point 2 possible
n+r
To channel
Constellation possible dimensional
selection mapper modulator
from coset (MQAM,
One coset selected Modulation stage MPSK, …)
out of 2k+r possible

k+r coded Binary

bits encoder k info bits
Coset
selector (block or
convolutional)
Coding stage

difference of any two vectors in the lattice is also in the lattice. A sub-lattice is a subset of a lattice that is itself a lattice. The
sequence space of uncoded signal is a sequence of points from N-cube, obtained as a Cartesian product of 2-D rectangular
lattices with points located at odd integers. Once the densest lattice is determined, we create its decomposition into partition
subsets known as cosets. The k information bits are used as input to binary encoder, be block or convolutional. The binary
encoder generates the codeword of length k + r, where r is the number of redundant bits. With k + r bits, we can select one out
of 2k + r possible cosets. Each coset contains 2n  k points, and therefore, n  k additional information bits are needed to select
a point within the coset. Given that the lattice is defined in N-dimensional space, while 2-D modulator imposes the data in 2-D
space, N/2 consecutive transmissions are needed to transmit a point from the coset. Therefore, information bits are conveyed
in two ways: (i) through the sequence of cosets from which constellation points are selected and (ii) through the points selected
within each coset. Let dpoints denote the minimum distance of points within the coset and dseq denote the minimum distance
among coset sequences; then the minimum distance of the code can be determined as dmin ¼ min(dpoints,dseq). Since the
fundamental volume per Nffi dimensions is 2r, where r is the number of redundant bits, the normalized volume per two
pffiffiffiffi
dimensions would be N=2
2r ¼ 22r=N ; the coding gain of this coded modulation scheme can be estimated as

d 2min
G¼ ¼ 22r=N d 2min : ð9:143Þ
22r=N

Clearly, the normalized redundancy per two dimensions is equal to r/(N/2) ¼ 2r/N. When the constellation is chosen to be N-
sphere-like to reduce the average energy, we need to account for an additional gain, often called the shaping gain, denoted
with Gs, which measures the reduction in energy. Calderbank and Sloane have shown that the shaping gain of an N-sphere
over N-cube, when N is even, is given by [97]

π ðN=2Þ þ 1
Gs ¼ : ð9:144Þ
6 ½ðN=2Þ!1=ðN=2Þ

For instance, for N ¼ 2 and 4, the shaping gains are π/3 (0.2 dB) and π/23/2 (0.456 dB), respectively. The shaping gain in
limiting case, as N tends to infinity, is πe/6 (1.533 dB).
Based on Forney’s guidelines [94], it is possible to generate maximum-density N-dimensional lattices by using a simple
partition of a 2-D lattice with corresponding block/convolutional code, when N ¼ 4, 8, 16, and 24, by applying the subset
partition rules as shown in Fig. 9.42. The 4-D lattice is obtained by taking all sequences of pair of points from the same subset,
such as (A,A) or (B,B). The 8-D lattice is obtained by taking all sequences of four points of either type A or type B. Each point
in subset has a single subscript. Within each of four-point subset, the point subscripts satisfy the parity-check equation
i1 + i2 + i3 + i4 ¼ 0, so that the sequence subscripts must be codewords in the (4,3) parity-check code, which has a minimum
Hamming distance of two. Therefore, three information bits and one parity bit are used to determine the lattice subset. The
16-D lattice is obtained by taking all sequences of eight points of either type A or type B. Each point has two subscripts. The
subscripts in the subset belong to a codeword from extended Hamming code (16,11) of min Hamming distance 4. Finally, the
24-D lattice is obtained by taking all sequences of 12 points of either type A or type B. Each point has three subscripts (i,j,k).
The subscripts (i,j) in 12-point subset form a codeword from Golay (24,12) code of minimum Hamming distance 8. The third
9.11 Coded Modulation and Unequal Error Protection 585

Fig. 9.42 Subset partition for up . . . . . . . . . . . . . . . .

to 24-dimensional lattice
B000 A000 B110 A110 B000 A000 B110 A110 B000 A000
A101 B010 A010 B101 A101 B010 A010 B101 A101 B010
B111 A111 B001 A001 B111 A111 B001 A001 B111 A111
A011 B100 A100 B011 A011 B100 A100 B011 A011 B100
B000 A000 B110 A110 B000 A000 B110 A110 B000 A000
A101 B010 A010 B101 A101 B010 A010 B101 A101 B010
B111 A111 B001 A001 B111 A111 B001 A001 B111 A111
A011 B100 A100 B011 A011 B100 A100 B011 A011 B100
B000 A000 B110 A110 B000 A000 B110 A110 B000 A000
A101 B010 A010 B101 A101 B010 A010 B101 A101 B010
. . . . . . . . . . . . . . . .

Fig. 9.43 Illustration of the Q

two-state TCM trellis section for
8-ary signal constellation S ¼ (s0, s0 s3
s2
Seven s1
s1,..,s7) (a) and corresponding
8-PSK signal constellation (b) s4 Seven dmin

s6 s5 s4 s0 I
s1 s2
s3
Sodd Sodd s5 s7
s7
s6
(a) (b)

subscript k represents overall parity check; for B-points even parity check is used and for A-points odd-parity check is used
(or vice versa). The corresponding coding gains are 1.5 dB, 3 dB, 4.5 dB, and 6 dB for lattice codes of dimensionalities 3, 8,
16, and 24, respectively [94].

9.11.2 Trellis-Coded Modulation

In TCM, introduced by Ungerboeck [93], the redundancy is introduced by expanding the signal constellation into a larger
signal constellation. Typically, if the size of original constellation is M, the size of extended constellation would be 2 M. The
TCM can be described as follows. We define the trellis and then assign signal constellation points to different branches in
trellis, as illustrated in Fig. 9.43a. Unlike convolutional codes where only one edge is allowed between two states, in TCM
parallel transitions are allowed. This particular example is applicable to both 8-QAM and 8-PSK (shown in Fig. 9.43b). This
trellis has two states, denoted as Seven and Sodd. When the encoder is in state Seven, only a signal constellation point s with even
subscript can be generated, in other words s2{s0,s2,s4,s6}. On the other hand, when the encoder is in state Sodd, only a signal
constellation point s with odd subscript can be generated, in other words s2{s1,s3,s5,s7}. The original constellation is QPSK,
defined as SQPSK ¼ {s0,s2,s4,s6}. The Euclidean distance square for QPSK is d2unc ¼ 2Eunc , Eunc ¼ 1: On the other hand, the
Euclidean distance in 8-PSK is d 28‐PSK ¼ 4 sin 2 ðπ=8ÞEcoded , E coded ¼ 1: As the generalization of the free distance of
convolutional codes, the free distance of TCM can be defined as the minimum Euclidean distance between any two
codewords in the code. The path corresponding to the minimum distance from (s0,s0)-path is (s2,s1) path. Therefore, the
ffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
free distance of this TCM scheme is d free ¼ d 2E ðs2 , s0 Þ þ d2E ðs1 , s0 Þ ¼ 2 þ 4 sin 2 ðπ=8Þ, where dE(si, sj) is the Euclidean
586 9 Advanced Coding and Coded Modulation Techniques

s0
S0
s4
s2 s2
s6 s6

S1 s1
s0
s7 s4

s5
S2
s7

s3 s3
s1
S3
Fig. 9.44 The four-state TCM trellis section for 8-ary signal constellation

distance between signal constellation points si and sj. By using Eq. (9.142), we can calculate the asymptotic coding gain as
follows:

E unc d2min d2free 2 þ 4 sin 2 ðπ=8Þ

Ga ¼ ¼ ¼ ¼ 1 þ 2 sin 2 ðπ=8Þ
Ecoded d2unc d2unc 2
|fflffl{zfflffl}
¼1
¼ 1:29289 ð1:1156 dBÞ:

To increase the coding gain, we can use trellis with larger number of states. For instance, the four-state TCM trellis is
shown in Fig. 9.44. Two shortest paths require three steps to return back to the state s0. The shortest path with respect to the
all-zero path is (s4,s0,s0)-path and the corresponding Euclidean distance square is 4. The asymptotic coding gain for four-state
TCM scheme is determined as follows: Ga ¼ d2free =d 2unc ¼ 4=2 ¼ 2 ð3 dBÞ:
The mapping of constellation points should be achieved on such a way to maximize the asymptotic coding gain, in other
words to maximize the free Euclidean distance. One popular way to achieve so is through so-called set partitioning. A 2m-ary
signal constellation set is partitioned into m-levels. At the i-th partition level (0  i  m  1), the signal Si is split into two
subsets Si(0) and Si(1) so that the intra-set squared Euclidean distance d2i is maximized. With each subset choice Si(ci), we
associate a bit label ci. The end result of this process is unique m-label c0c1. . .cm  1 associated with each constellation point
s(c0c1. . .cm  1). By employing this Ungerboeck (standard) set partitioning of 2m-ary signal constellation, the intra-set
distances per levels are arranged in nondecreasing order as level index increases; in other words:

d 20  d 21  ⋯d2m1 :
9.11 Coded Modulation and Unequal Error Protection 587

Fig. 9.45 The set partitioning of 010

8-PSK with natural mapping rule 011 001=c2c1c0
d0=
2sin( /8)
100 000

c0=0 c0=1
101 111
110

d1= 2

c1=0 c1=1 c1=0 c1=1

c2=0
c2=0 c2=0
c2=0
c2=1
d2=2
c2=1 c2=1
c2=1

As an illustration, the set partitioning of 8-PSK with natural mapping rule is shown in Fig. 9.45. Clearly, in level 3 there are
four subsets (cosets) with each subset containing two points. To select a given coset, two bits are needed. Rate ½
convolutional code can be used for this purpose as illustrated in Fig. 9.46a. To select a point within a given subset (coset),
one information bit is needed. The overall code rate of this trellis-coded 8-PSK modulation scheme is ¾. The corresponding
trellis diagram is shown in Fig. 9.46b. The current state is defined by the content of shift register (D0D1), while the next state
by (u0D1). The first two bits (c0c1) are used to select the coset, while the information bit u1 (code bit c2) is used to select a point
within that coset.
The general encoder for (m  1)/m TCM scheme is shown in Fig. 9.47. The k information bits are used as input to rate k/
(k + 1) convolutional code, whose codeword (c0c1. . .ck) is used to select a coset (signal constellation subset). Additional m-
(k + 1) information bits are used to select a point within the coset.
To decode the most likely TCM sequence, the Viterbi algorithm can be used, in similar fashion as described in Sect. 9.3.
For the branch metric, either correlation metric or Euclidean distance can be used. The survivor path memory must include the
m  (k + 1) information (uncoded) bits. Because of the parallel edges, the computation complexity of the corresponding
Viterbi algorithm is higher than that for either convolutional or linear block codes.
The symbol error rates of TCM schemes can be evaluated in similar fashion as for convolutional codes, with the help of
weight-enumerated sequence obtained from the state diagram of trellis code. The key difference is the replacement of
Hamming distance with Euclidean distance. Another relevant difference is that in TCM we have parallel branches in trellis
diagram.

9.11.3 Multilevel Coded Modulation and Unequal Error Protection

Multilevel coding (MLC) is initially proposed by Imai and Hirakawa in 1977 [85]. The key idea behind the MLC is to protect
individual bits using different binary codes and use M-ary signal constellations, as illustrated in Fig. 9.48. The i-th component
code, denoted as Ci, is (n,ki,di(H)) (i ¼ 1,2,. . .,m) linear block code of Hamming distance di(H). The length of information word
ui for the i-th component code is ki, while the corresponding codeword length n is the same for all codes. The bits-to-symbol
mapper can be implemented with the help of block interleaver, in which the codewords from different component codes are
written in row-wise fashion into interleaver and read out m bits at the time in column-wise fashion. With this interpretation, the
symbol at j-th time instance ( j ¼ 1,2,. . .,n) is obtained after the mapping sj ¼ s(cj), where cj ¼ (ci,j ci,2 . . . ci,m) and s() is the
588 9 Advanced Coding and Coded Modulation Techniques

8-PSK signal
mapping
c0
00111100

u0
D0 D1

Partition label
c1
00110011

u1 c2
01010101

04371526
Signal index

(a)
000
00 00
011 100 011
111
111

01 111 01

001
101
10 100 10

010 010
110 001 110
11 11
101
(b)

Fig. 9.46 Encoder for four-state rate-2/3 trellis-coded 8-PSK modulation (a) and corresponding trellis diagram (b)

k info bits Rate k/(k+1) (k+1) coded bits Select the coset
convolutional (signal constellation
code subset)

m (k+1) info bits Select the Signal index

point within
the coset

Fig. 9.47 General encoder for a rate (m  1)/m TCM scheme

corresponding mapping rule. The 2m-ary 2-D modulator, such as I/Q and PSK modulators, is used to impose the selected
signal constellation point si ¼ ðsi,I , si,Q Þ, where si, I and si, Q are corresponding in-phase and quadrature components, on
corresponding carrier signal. For instance, the signal index can be used as the address for look-up table (LUT) where the
coordinates are stored. Once the coordinates are selected, the in-phase and quadrature components are used as inputs of
corresponding pulse shapers, followed by driver amplifiers. For optical communications, the optical I/Q modulator is further
used to perform electrical-to-optical conversion. For wireless applications, after the I/Q modulator, the corresponding signal is
after power amplifier used as input to the transmit antenna.
9.11 Coded Modulation and Unequal Error Protection 589

Information c1
Encoder 1
sequence 1, u1
c2 Bits-to- s(c) M-ary To channel
u2 Encoder 2 symbol 2-D
mapping modulator

…
um cm
Encoder m

Fig. 9.48 Multilevel encoder for MLC/MCM scheme

û 1
Decoder 1 Delay 1

û 2
Received Decoder 2 Delay 2
Delay TD1
samples, r

…
…
From Decoder m-1
uˆ m
Delay TDm Decoder m

Fig. 9.49 Multistage decoder for MLC/MCM scheme

The overall code rate (spectral efficiency) of this MCM scheme, expressed in bits/symbol, is given by

Xm Xm Pm
i¼1 k i
R¼ R
i¼1 i
¼ k =n
i¼1 i
¼ : ð9:145Þ
n

The minimum squared Euclidean distance (MSED) of this scheme is lower bounded by [85].
n o
ðH Þ
MSED  min di di , ð9:146Þ
1im

where di is the Euclidean distance at the i-th level of set partitioning. As an illustration, the Euclidean distances at i-th level of
set partitioning (i ¼ 1,2,3) have been already provided in Fig. 9.45. When the component n codes are selected as follows,
o
pffiffiffi2
(8,1,8), (8,7,2), (8,8,1), the MSED for coded 8-PSK is lower bounded by MSED  min 0:76542  8, 2  2, 22  1 ¼ 4,
indicating that the asymptotic coding gain with respect to uncoded QPSK is at least 3 dB. When properly designed LDPC
codes are used as component codes, the capacity-achieving performance can be obtained [99, 100].
Each component code can be represented as trellis diagram as described in Sect. 9.3. By taking the Cartesian product of
individual trellises and then performing mapping based on set partitioning, we obtain a trellis description of this MCM
scheme. Now we can apply either Viterbi algorithm or BCJR algorithm to perform decoding on such trellis. Unfortunately, the
complexity of such decoding is too high to be of practical importance. Instead, the decoding is typically based on so-called
multistage decoding (MSD) algorithm [7] in which the decisions from prior (lower) decoding stage are passed to next (higher)
stage, which is illustrated in Fig. 9.49.
The i-th decoder operation (i ¼ 2,. . .,m) is delayed until the (i  1)-th decoder completes the decoding algorithm. Further,
after the decoding, the i-th decoder (i ¼ 1,. . .,m  1) output is delayed until the m-th decoder decoding algorithm is completed.
Clearly, even though the complexity of MSD algorithm is lower compared to Viterbi/BCJR algorithm operating on MLC
trellis diagram, since the i-th component decoder operates on trellis of Ci code, the overall decoding latency is still high.
Moreover, when the component codes are weak codes, the error not corrected in prior decoding stage will affect the next stage
and result in error multiplicity. On the other hand, when component codes are properly designed LDPC codes, this problem
can be avoided [99, 100]. Moreover, it has been shown in [100], when the code rates and degree distribution are optimized for
a given mapping rule, BER performance degradation when component LDPC decoders operate independently can be made
arbitrary small compared to MSD algorithm. This decoding algorithm in which all component decoders operate independently
of each other is commonly referred to as parallel independent decoding (PID) [99, 100]. The PID decoding architecture is
illustrated in Fig. 9.50, assuming that properly designed LDPC codes are used as component codes.
590 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.50 Parallel independent Extrinsic LLRs

decoding (PID) for MLC/MCM
scheme
û 1
LDPC decoder 1
APP
Received Symbol LLRs- û 2
demapper LDPC decoder 2
samples
to-bit LLRs
rl=(rl,I, rl,Q) (Symbol LLRs

…
calculation
calculation)
LDPC decoder m uˆ m

The received signal from receive antenna, after down conversion and digital-to-analog conversion in the observed
symbolling interval, can be represented as r ¼ ðr I , r Q Þ, where rI is the in-phase coordinate and rQ is the quadrature coordinate
of received symbol. In optical communication, after the optical coherent balanced detection, as shown earlier, the received
symbol can be represented in the same form. In the a posteriori probability (APP) demapper, the symbols’ si (i ¼ 1,2,..,2m)
LLRs are calculated as follows:

Pðsi jr Þ
Lðsi Þ ¼ log , ð9:147Þ
Pðs0 jr Þ

where P(s|r) is determined by using Bayes’ rule:

Pðrjsi ÞPðsi Þ
Pðsi jr Þ ¼ P : ð9:148Þ
s0 Pðrjs ÞPðs Þ
0 0

In the presence of fiber nonlinearities, P(r|si) in (9.148) needs to be estimated evaluating the histograms, by propagating a
sufficiently long training sequence. With P(si) we denote the a priori probability of symbol si, while s0 is a reference symbol.
By substituting (9.148) into (9.147), we obtain
     
Pðrjsi ÞPðsi Þ Pðrjsi Þ Pðsi Þ
Lðsi Þ ¼ log ¼ log þ log
Pðrjs0 ÞPðs0 Þ Pðrjs0 Þ Pðs Þ
    0 ð9:149Þ
Pðrjsi Þ P ðsi Þ
¼ log þ La ðsi Þ, La ðsi Þ ¼ log ,
Pðrjs0 Þ Pðs0 Þ

where with La(si) we denote the prior reliability of symbol si. Let us denote by sij the j-th level of the transmitted symbol si
represented as si ¼ (si1,. . .,sim), where sij2{0,1}. The prior symbol LLRs for the next iteration are determined by

m
∏ Pðsij Þ
Pðsi Þ j¼1
La ðsi Þ ¼ log ¼ log m ¼
Pðs0 Þ
∏ Pðsij ¼ 0Þ ð9:150Þ
j¼1

P
m Pðsij Þ Xm
Pðsij Þ
¼ log ¼ La ðsij Þ, La ðsij Þ ¼ log
j¼1 Pðsij ¼ 0Þ j¼1
Pðs ij ¼ 0Þ

where we assumed that reference symbol is s0 ¼ (0. . .0).

The prior symbol estimate for the next outer iteration can be obtained from

X
m 
La ðbsi Þ ¼ LD,e bsij , ð9:151Þ
j¼1

where
9.11 Coded Modulation and Unequal Error Protection 591

  
ðt Þ ðt1Þ
LD,e bsij ¼ L sij  L sij : ð9:152Þ


ðt Þ
In (9.152), we use L sij to denote the LDPC decoder output in current iteration (iteration t). The symbol LLRs
corresponding to the j-th decoder L(sij) are determined from symbol LLRs (9.149) by

P P 
 si :sij ¼1 exp ½Lðsi Þ exp si :sik ¼0,k6¼j La ðsik Þ
L bsij ¼ log P P : ð9:153Þ
s :sij ¼0 exp ½Lðsi Þ exp
i s :sik ¼0,k6¼j La ðsik Þ
i

Therefore, the j-th position reliability in (9.153) is calculated as the logarithm of the ratio of a probability that sij ¼ 1 and
probability that sij ¼ 0. In the nominator, the summation is done over all symbols si having 1 at the position j, while in the
denominator over all symbols si having 0 at the position j. With La(sik) we denoted the prior (extrinsic) information determined
from the APP demapper. The inner summation in (9.153) is performed over all positions of symbol si, selected in the outer
summation, for which sik ¼ 0, k 6¼ j. The j-th position LLRs are forwarded to corresponding binary LDPC decoders. The
iteration between the APP demapper and LDPC decoders is performed until the maximum number of iterations is reached or
the valid codewords are obtained.
Clearly, in MLC different bits transmitted over the channel are differently protected, and such the MLC is suitable for
unequal error protection (UEP). The MLC channel encoder consists of m different binary error-correcting codes C1, . . ., Cm
with decreasing codeword distances. The i-th priority bit stream is used as the input to the i-th encoder, which generates the
coded bits ci. The rest is the same as described above. As an illustration, in image compression, one type of channel code can
be used for bits corresponding to the low-resolution reproduction of the image, whereas high-resolution bits that simply refine
the image can be protected by a different channel code. This is in particular important in JPEG 2000 image compression
standard, in which the image is decomposed into a multiple-resolution representation. The JPEG 2000 provides efficient code-
stream organizations that are progressive by pixel accuracy and by image resolution. Thanks to such organization, after a
smaller portion of the whole file has been received, the viewer can see a lower-quality version of the picture. The quality of the
picture then gets improved progressively through downloading additional data bits from the source.
For UEP, the nested coding can also be used. Nested coding can be interpreted as follows. Let us consider m different
information vectors ui (i ¼ 1, . . ., m) of length km. We would like to jointly encode these information vectors on such a way
that each information vector is associated with a codeword from a different subcode. The i-th subcode Ci is represented by the
generator matrix Gi of rate Ri ¼ ki/n. The overall generator matrix is given by
2 3
G1
6G 7
6 2 7
G¼6 7: ð9:154Þ
4⋮ 5
Gm

The overall codeword c can be obtained as follows:

2 3
G1
6 7
6 G2 7
6 7
cT ¼ ½uT1 uT2 . . . uTm G ¼ ½ uT1 uT2 . . . uTm 6 7
6⋮ 7 ð9:155Þ
4 5
Gm
¼ uT1 G1  uT2 G2 ⋯ uTm Gm ,

where we use  to denote the bitwise XOR operation. If we are interested in unequal error correction, by setting ui ¼ u, by
varying the number of generator matrices Gi, we can achieve different levels of protection. The lowest level of protection
would be to use only one generator matrix. The highest level of protection, corresponding to high-priority bits, will be
achieved by encoding the same information vector u m-times.
592 9 Advanced Coding and Coded Modulation Techniques

To improve the spectral efficiency of MLC scheme, the time-multiplexed coded modulation can be used. The key idea is to
use different coded modulation schemes for different priority classes of information bits. Let Ti be the fraction of time in which
the i-th priority code Ci is employed, and let Rs,i denote the corresponding symbol rate. The overall symbol rate Rs of this
scheme would be
P
i Rs,i T i
Rs ¼ P : ð9:156Þ
iT i

When all component codes in MCM scheme have the same rate, the corresponding scheme in [12], and related papers, is
called as block-interleaved coded modulation, as it contains the block interleaver. This particular version, when combined
with optimum signal constellation and optimum mapping rule designs [12, 123–125], performs comparable to MLC, but has
even lower complexity. At the same time, it is suitable for implementation in hardware at high speed, as m LDPC decoders
operate in parallel at bit rate equal to symbol rate Rs. In conventional bit-interleaved coded modulation (BICM) Scheme [89],
described in the next section, a single LDPC decoder is required operating at rate m Rs, which can exceed the speed of existing
electronics for fiber-optics communications, where the information symbol rate is typically 25 GS/s.
For convergence behavior analysis of the block-interleaved coded modulation for optical communication applications, the
EXIT chart analysis should be performed. In order to determine the mutual information (MI) transfer characteristics of the
demapper, we model the a priori input LLR, LM,a, as a conditional Gaussian random variable [101]. The MI between c and LM,
a is determined numerically as explained in [101] (see also [90, 91, 99, 100, 102, 103]). Similarly, the MI IL M,e between c and
LM,e is calculated numerically, but with the p.d.f. of c and LM,e determined from histogram obtained by Monte Carlo
simulation, as explained in [101]. By observing the IL M,e as a function of the MI of IL M,a and optical signal-to-noise ratio,
OSNR, in dB, (SNR in wireless applications) the demapper EXIT characteristic (denoted as TM) is given by

IL M,e ¼ T M ðI L M,a , OSNRÞ: ð9:157Þ

The EXIT characteristic of LDPC decoder (denoted as TD) is defined in a similar fashion as

IL D,e ¼ T D ðI L D,a Þ: ð9:158Þ

The “turbo” demapping-based receiver operates by passing extrinsic LLRs between demapper and LDPC decoder. The
iterative process starts with an initial demapping in which LM,a is set to zero, and as a consequence, IL M,a becomes zero as
well. The demapper output LLRs, described by

IL M,e ¼ IL D,a : ð9:159Þ

are fed to LDPC decoder. The LDPC decoder output LLRs, described by

IL D,e ¼ IL M,a ð9:160Þ

are fed to the APP demapper. The iterative procedure is repeated until the convergence or the maximum number of iterations
has been reached. This procedure is illustrated in Fig. 9.51, where the APP demapper and LDPC decoder EXIT charts are
shown together on the same graph. Three modulation formats (8-PSK, 16-PSK, 16-QAM) are observed, as well as the
following mappings: natural, Gray, and anti-Gray. The EXIT curves have different slopes for different mappings. The
existence of “tunnel” between corresponding demapping and decoder curves indicates that iteration between demapper and
decoder will be successful. The smallest OSNR, at which iterative scheme starts to converge, is known as threshold (pinch-
off) limit [101]. The threshold limit in the case of 16-PSK (Fig. 9.51b) is about 3 dB worse as compared to 8-PSK (Fig. 9.51a).
The 16-QAM mapping curve is well above the 16-PSK curve (see Fig. 9.1b), indicating that 16-QAM scheme is going to
significantly outperform the 16-PSK one.
9.11 Coded Modulation and Unequal Error Protection 593

Fig. 9.51 EXIT charts for 1.0

different mappings and
modulation formats

0.8

IM,e->ID,a
0.6
Decoder:
LDPC(4320,3242,0.75)
8-PSK, OSNR=8 dB / 0.1nm:
0.4 Demapper:
Gray
Natural
Anti-Gray
0.2
0.0 0.2 0.4 0.6 0.8 1.0

ID,e->IM,a
(a)
1.0

0.8
IM,e->ID,a

Decoder:
0.6 LDPC(4320,3242,0.75)
16-PSK, OSNR=11 dB / 0.1nm:
Demapper:
Gray
0.4 Natural
16-QAM, OSNR=11 dB / 0.1nm:
Demapper:
Gray
0.2
0.0 0.2 0.4 0.6 0.8 1.0

ID,e->IM,a
(b)
594 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.52 Bit-interleaved coded

modulation (BICM) scheme: (a) m
BICM encoder and (b) BICM
Pseudorandom bits Bits-to- s(c) 2m-ary To
decoder configurations Information Binary c
bit interleaver symbol 2-D
sequence u encoder channel
mapping modulator

(a)

Demodulator Symbol LLRs- û

From Deinterleaver Binary
and to-bit LLRs
channel
1 decoder
APP demapper
calculation

(b)

Fig. 9.53 Pragmatic coded m k bits

modulation scheme
Information Store and
Turbo Interleaver
sequence u distribute
encoder Puncturer
m k bits
m k k
bits bits

Bits-to- s(c) 2m-ary To

symbol 2-D
channel
mapping modulator

9.11.4 Bit-Interleaved Coded Modulation (BICM)

In bit-interleaved coded modulation (BICM) [89], a single binary encoder is used followed by a pseudorandom bit interleaver,
as shown in Fig. 9.52. The output of bit interleaver is group in block of m bits, and during signaling interval these m bits are
used as input of corresponding mapper, such as Gray, anti-Gray, and natural mapping, and after the mapper, m bits from
mapper are used to select a point from a 2m-ary signal constellation. The rest of transmitter is similar to that for MLC scheme.
On receiver side, the configurations of demodulator, APP demapper, and symbol LLRs-to-bit LLRs calculation block are
similar to those for MLC scheme. After the deinterleaving of bit LLRs, corresponding bit reliabilities are forwarded to a single
binary decoder. For this scheme, the choice of mapping rule is of high importance. For iterative decoding and demapping, the
Gray mapping rule is not necessarily the best mapping rule. The EXIT chart analysis should be used to match the APP
demapper and LDPC decoder choice.

9.11.5 Turbo Trellis-Coded Modulation

There exist different approaches to coded modulation based on turbo codes such as pragmatic coded modulation [104], turbo
TCM with symbol interleaving [105, 106], and turbo TCM with beat interleaving [88].
In so-called pragmatic coded modulation [104], encoders and decoders are binary, as shown in Fig. 9.53. Therefore, this
scheme has certain similarities with BICM scheme. Turbo encoder sequence is split into blocks of m  k bits each, which are
then used as input to storage and distribution block.
The m  k bits at the output of this block are forwarded directly to interleaver, while at the same time m  k bits are
punctured down to k bits and then forwarded to the interleaver Π. Therefore, the interleaver collects m bits at the time. After
interleaving, m bits are taken from interleaver to select a point from corresponding 2m-ary signal constellation. The rest of
transmitter is similar to that of BICM scheme.
9.12 Hybrid Multidimensional Coded Modulation Scheme for High-Speed Optical Transmission 595

Fig. 9.54 Symbol-interleaved m (k+1) bits info bits

turbo TCM: (a) encoder and (b) Alternate
decoder architectures Mapping
selector
RSC 1
k bits k+1
Symbol
bits
deinterleaver
-1
Symbol Even-to-even positions s

interleaver Odd-to-odd positions

s

Mapping
RSC 1

(a)
Symbol
Decoder 1 interleaver
s

Decision
step block
Symbol
interleaver Decoder 2
s

(b)

The symbol interleaving TCM scheme is shown in Fig. 9.54. The encoder architecture is shown in Fig. 9.54a. Clearly, the
interleaver operates on symbol level, each symbol contains m bits, even positions in interleaver get mapped to even positions,
and odd positions get mapped to odd positions. Two TCM encoders are coupled by the symbol interleaver. The RSC codes
have the rate k/(k + 1). The alternate selector selects the upper and lower branches in an alternative fashion. The m-(k + 1)
uncoded bits and k + 1 coded bits from RSC are used to select a point in a signal constellation diagram. Given the existence of
two alternative modulation branches, the careful puncturing of the redundant symbols is needed.
The symbol-interleaved TCM decoder architecture is shown in Fig. 9.54b. Clearly, the symbol-interleaved TCM decoding
principle is very similar to that of turbo decoding principle, and the key difference is that interleavers operate on a symbol
level. Additional details on this scheme can be found in [105, 106].
The bit-interleaved turbo TCM, shown in Fig. 9.55, was introduced by Benedetto et al. in 1996 [88]. In this scheme, instead
of symbol interleaving and puncturing of redundant symbols, multiple bit interleavers are employed. After RSC1 we select m/
2 systematic bits and puncture the remaining ones. In RSC2 we select the bit positions punctured in RSC1 as the output and
puncture the remaining ones. The permutation block Π1 permutes the bits selected by RSC1. On the other hand, the
permutation block Π2 permutes the bits punctured by RSC1. The remaining m/2 bits from RSC1 and m/2 bits from RSC2
are used to select a point from 2m-ary signal constellation.

9.12 Hybrid Multidimensional Coded Modulation Scheme for High-Speed Optical Transmission

It has been shown that multilevel coding with parallel independent decoding (MLC/PID) and iterative detection and decoding
(IDD) perform very well under diverse optical fiber transmission scenarios [2, 10–12]. Recently, nonbinary LDPC-coded
modulation (NB-LDPC-CM) has been shown to outperform its corresponding MLC/PID+IDD scheme with performance gaps
increasing as the underlying signal constellation size increases [82]. NB-LDPC-CM employs a single component code defined
596 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.55 Turbo TCM encoder

(m k 1)/2 bits
with bit interleaving m/2 bits
RSC 1 Bits-to-
(m k 1)/2 bits
symbol
To
mapping
channel
+
Interleaver Interleaver
2-D
1 2
modulator
m/2 bits
RSC 2

over an alphabet that matches in size to the underlying constellation size, which in return eliminates the need for iterative
demapping-decoding but increases the decoding complexity. In this section, we described alternative, hybrid multidimen-
sional LDPC-coded modulation scheme, which represents the generalization of hybrid coded modulation scheme introduced
in [50] that lies essentially in between the two extremes described in [10], employing only binary LDPC codes, and [82] –
employing a single nonbinary LDPC code over a large finite field – in terms of decoding complexity and error correction
performance. Having such alternatives at hand can provide the much needed flexibility during link budget analysis of optical
communication systems rather than forcing the system designer to opt into one of the two extremes. Additionally, various
coded modulation schemes including bit-interleaved coded modulation, MLC, nonbinary LDPC CM, and multilevel
nonbinary LDPC-coded modulation (NB-LDPC-CM) described in Subsection C, to mention few, are just particular instances
of hybrid coded modulation scheme.

9.12.1 Hybrid Coded Modulation

A generic adaptive hybrid D-dimensional NB-LDPC-coded modulation scheme to be used in combination with irregular
NB-LDPC coding, discussed in Sec. 9.10.4, is shown in Fig. 9.56. To facilitate explanations, the details related to
compensation of dispersion and nonlinear effects, carrier frequency, and carrier phase estimations are omitted. It is applicable
to various spatial division multiplexing (SDM) fibers: few-mode fibers (FMFs), few-core fibers (FCFs), and few-mode-few-
core fibers (FMFCFs). It is also applicable to single-mode fibers (SMFs), which can be considered as SDM fibers with only
one fundamental mode and two polarization states. Both electrical and optical basis functions are applicable. Electrical basis
functions include modified orthogonal polynomials and orthogonal prolate spheroidal wave functions (also known as Slepian
sequences in discrete-time domain), while optical basis functions include spatial modes and polarization states. This scheme is
very flexible and can be used in various configurations ranging from spatial multiplexing of various multidimensional signals
to fully D-dimensional signaling.
Generic Tx configuration for the hybrid multidimensional LDPC-coded modulation scheme is presented in Fig. 9.56a. The
MLC scheme, shown in Fig. 9.57a, is a degenerate case of the hybrid CM scheme and occurs when source channels are binary,
while LDPC codes at different levels have different code rates. When all LDPC codes shown in Fig. 9.57a have the same code
rate, we refer to this scheme as the block-interleaved coded modulation (BICM). The block interleaver can be replaced by
more advanced interleavers [20, 24]. The conventional bit-interleaved coded modulation is a serial concatenation of channel
encoder, interleaver, and multilevel modulator [89]. This scheme is based on a single LDPC code, which operates at rate mRs,
where m is the number of bits per symbol and Rs is the symbol rate. For currently available fiber-optics communication
systems operating at 25 Gs/s, this scheme will require the implementation of LDPC encoder and decoder operating at rates
m  25 Gb/s, which would be quite challenging to implement with currently available commercial electronics. To avoid this
problem, m parallel LDPC encoders and decoders should be used instead in fiber-optics communications, with one possible
implementation shown in Fig. 9.57a (see also Fig. 9.56a), as indicated above. Alternatively, the source channels shown in
Fig. 9.57a may originate from 1:m demultiplexer. Moreover, the block interleaver can be replaced with m interleavers, one per
LDPC encoder, decomposing, therefore, the BICM scheme from Fig. 9.57a into m parallel conventional bit-interleaved coded
modulation schemes due to Caire et al. [89]. The corresponding schemes perform comparable under AWGN channel
assumption.
9.12 Hybrid Multidimensional Coded Modulation Scheme for High-Speed Optical Transmission 597

Fig. 9.56 Hybrid D-dim.

multidimensional coded constellation
b1 b1
modulation: (a) transmitter NB LDPC encoder

D-dimensional

D-dimensional
configuration, (b) receiver GF(2b1) Level 1

modulator
mapper
configuration, (c) configuration of To SDM
Source

…
D-dimensional modulator, and (d)

…
channels fiber
D-dimensional demodulator bm bm
configuration. SDM spatial NB LDPC encoder
division multiplexing. SDM fibers GF(2b1) Level m
could be FMF, FCF, FMFCF, or
SMF. The corresponding signal (a)
space is D ¼ 4N1N2-dimensional,
where N1 is the number of Extrinsic LLRs

…
sub-bands in a superchannel with calculation
orthogonal center frequencies and D-dim. const.
N2 is the number of spatial modes. samples NB LDPC

pol. diversity Rx
The factor 4 originates from two decoder

D-dimensional
Mode demux +

APP demapper

LLRs calculation
D-dimensional
demodulator

Extrinsic level
polarization states and in-phase GF(2b1) Level 1
From SDM

…
and quadrature channels

…
…
fiber
NB LDPC
decoder
GF(2bm) Level m

(b)
Content of spatial mode 1

Transmitter side DSP

Content of spatial mode N2

DAC

DAC

DAC

DAC

DAC

DAC

DAC

DAC
LDN1

LD1
…
QN1x

IN1x

Q1x
Q1y

I1x
I1y
IN1y
QN1y

modulator

modulator
4-D E/O

4-D E/O
Power-
combiner

Spatial division
multiplexer
D-dimensional modulator
SDM fiber

(c)
SDM fiber

D-dimensional demodulator
Spatial division
demultiplexer

Power-
splitter 4N1-dimensional
… demodulator 1 …

Local 4-D polarization 4-D polarization Local 4N1-dimensional

LD1 diversity Rx 1
… diversity Rx N1 LDN1 demodulator N2

I’1x Q’1x I’1y Q’1y

…I’N 1x Q’N1x I’N1y Q’N1y …
…
D-dimensional coordinate samples

(d)
598 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.57 MLC-based D-dim.

multidimensional coded constellation
modulation: (a) transmitter BI LDPC encoder 1,

D-dimensional
configuration of MLC and block-

D-dimensional
R1=k1/n

Interleaver or
n m block-
interleaver

modulator
interleaved coded modulation Source m

mapper
To SMF /

…
schemes and (b) the multistage channels

…
decoding principle SDM fiber
BI LDPC encoder m,
Rm=km/n

(a)
D-dim. const.
samples
D
APP
LDPC decoder 1
demapper 1

D-dimensional
pol. diversity Rx

demodulator
Mode demux +
D
APP LDPC
From SDM
demapper 2 decoder 2

…
fiber

…
D
APP
LDPC decoder m
demapper m

(b)

On the other hand, the NB-LDPC-CM of [82], whose generalization to D-dimensional space is shown later in Fig. 9.87, is a
particular instance of hybrid CM scheme employing a single level with an M-ary LDPC code and performing just a single pass
through the detector and the decoder without any feedback. In hybrid multidimensional CM scheme, the number of levels is
between 1 and log2(M).  
The hybrid CM approach combines component codes defined over GF 2bi , 1  i  m in a way that the condition
b1 + . . . + bm ¼ log2(M) is satisfied. Thus, the resulting NB-LDPC-coded modulation schemes are not homogenous in the field
orders of their component codes, but are rather heterogeneous. They enable the interaction of component codes over various
fields and hence the name hybrid LDPC-coded modulation scheme.
The configuration of D-dimensional modulator is shown in Fig. 9.56c. Clearly, conventional PDM scheme is just a special
case corresponding to fundamental mode only by setting N1 ¼ N2 ¼ 1. Transmitter side DSP provides coordinates for D-
dimensional signaling. The 4-D modulator is composed of polarization beam splitter, two I/Q modulators corresponding to x-
and y-polarizations, and polarization beam combiners.
One spatial mode contains the data streams of N1 power combined 4-D data streams. The N2 data streams, corresponding to N2
spatial modes, are combined into a signal data stream by spatial division multiplexer and transmitted over SDM fiber. On the other
hand, the D-dimensional demodulator, shown in Fig. 9.56d, provides the samples corresponding to signal constellation estimates.
The LLRs of symbols si (i ¼ 1,. . .,M ) are calculated in an APP demapper block as follows:

Pðsi jrÞ
Lðsi Þ ¼ log , ð9:161Þ
Pðs0 jrÞ

where P(s|r) is determined by using Bayes’ rule:

Pðrjsi ÞPðsi Þ
Pðsi jrÞ ¼ P : ð9:162Þ
Pðrjs0 ÞPðs0 Þ
s0

Notice that si is the transmitted signal constellation point at time instance i, and rij ( j ¼ 1,2,. . .,D) are corresponding samples
obtained at the output of D-dimensional demodulator, shown in Fig. 9.56d. In the presence of fiber nonlinearities, P(r|si) in
(36) needs to be estimated evaluating the histograms, by propagating a sufficiently long training sequence. With P(si) we
denote the a priori probability of symbol si, while s0 is a reference symbol. By substituting (9.162) into (9.161), we obtain
     
Pðri jsi ÞPðsi Þ Pðri jsi Þ Pðsi Þ
Lðsi Þ ¼ log ¼ log þ log
Pðri js0 ÞPðs0 Þ Pðri js0 Þ Pðs0 Þ
    ð9:163Þ
Pðri jsi Þ P ðsi Þ
¼ log þ La ðsi Þ, La ðsi Þ ¼ log ,
Pðri js0 Þ Pðs0 Þ
9.12 Hybrid Multidimensional Coded Modulation Scheme for High-Speed Optical Transmission 599

where with La(si) we denote the prior reliability of symbol si. Let us denote by sij the j-th component of the transmitted symbol
si represented as si ¼ (si1,. . .,sim), where sij2GF(2mi). The prior symbol LLRs for the next iteration are determined by

m  
∏ P sij
Pðsi Þ j¼1
La ðsi Þ ¼ log ¼ log m  ¼
Pðs0 Þ
∏ P sij ¼ 0 ð9:164Þ
j¼1
   
Xm
P sij Xm     P sij
¼ log  ¼ La sij , La sij ¼ log  
j¼1
P sij ¼ 0 j¼1
P sij ¼ 0

where we assumed that reference symbol is s0 ¼ (0. . .0).

Finally, the prior symbol estimate can be obtained from

X
m  
La ðbsi Þ ¼ LD,e bsij , ð9:165Þ
j¼1

where

   
ðt Þ ðt1Þ
LD,e bsij ¼ L sij  L sij ð9:166Þ


ðt Þ
In (9.166), we use L sij to denote the LDPC decoder output in current iteration (iteration t).
The symbol LLRs corresponding to the j-th decoder L(sij) are determined from symbol LLRs (9.163) by

P P 
  si :sij ¼s exp ½Lðsi Þ exp si :sik ¼0,k6¼j La ðsik Þ
L bsij ¼ log P P : ð9:167Þ
si :sij ¼0 exp ½Lðsi Þ exp si :sik ¼0,k6¼j La ðsik Þ

Therefore, the j-th position reliability in (9.167) is calculated as the logarithm of the ratio of a probability that sij ¼ s, s2GF
(2mi)\{0} and probability that sij ¼ 0. In the numerator, the summation is done over all symbols si having s2GF(2mi)\{0} at the
position j, while in the denominator over all symbols si having 0 at the position j.
With La(sik) we denoted the prior (extrinsic) information determined from the APP demapper. The inner summation in
(9.167) is performed over all positions of symbol si, selected in the outer summation, for which sik ¼ 0, k 6¼ j. The j-th position
LLRs are forwarded to corresponding LDPC decoders operating over GF(2mj). The iteration between the D-dimensional APP
demapper and NB-LDPC decoders is performed until the maximum number of iterations is reached or the valid codewords are
obtained.
Instead of parallel independent decoding (PID) fashion, described above, iterative demapping and decoding can be
performed in a multistage decoding fashion, in which the reliabilities from the higher i-th (i ¼ 1,2,. . .,m  1) decoding
stage are forwarded to the lower (i + 1)-th decoding stage, as shown in Fig. 9.57b. In this scheme the code rates of LDPC codes
are arranged in increasing order: R1 < R2 < . . . < Rm. However, this decoding scheme exhibits higher decoding complexity and
higher latency compared to PID.

9.12.2 Multilevel Nonbinary LDPC-Coded Modulation (ML-NB-LDPC-CM) for High-Speed Optical

Transmissions

Still we restrict our analysis of coded modulation schemes over a predetermined modulation format of constellation size
M ¼ 2m. We compare multilevel NB-LDPC-CM with block-interleaved LDPC-coded modulation (BI-LDPC-CM) and
NB-LDPC-CM as presented in Fig. 9.58. The transmitter/receiver architectures of BI-LDPC-CM are depicted in
Fig. 9.58a. The source information entering the transmitter is composed of m parallel binary information streams of length
K bits each. The m parallel bit streams are separately encoded into m codewords through m binary LDPC encoders of identical
600 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.58 Transmitter and

Block-interleaver

2m-ary LDPC
LDPC Encoder

Optical I/Q
Modulator
receiver structures of (a)

Encoder
m I

Optical I/Q

Mapper
Modulator
m I m

Mapper
mxN
BI-LDPC-CM, (b) NB-LDPC- LDPC Encoder
CM, and (c) multilevel … Q
NB-LDPC-CM. These are Q
particular instances of hybrid LDPC Encoder
multidimensional coded
modulation scheme shown in
LDPC Decoder

2m-ary LDPC
Fig. 9.56 applied to 2-D signaling

Demapper
Demapper
I’

Coherent

Coherent
Calculate
Detector

Detector

Decoder
I’

Bit LLR
schemes
LDPC Decoder
… Q’
Q’
LDPC Decoder

(a) (b)
q q

Block-interleaver
2q-ary LDPC Encoder

Optical I/Q
Modulator
q q I

m/q x N

Mapper
2q-ary LDPC Encoder m
…
Q
q q
2q-ary LDPC Encoder

q
2q-ary LDPC Encoder

Symbol LLR
Demapper

Converter
Coherent

I’
Detector

q
2q-ary LDPC Encoder
Q’ …
q
2q-ary LDPC Encoder
(c)

code rate R ¼ K/N. The codewords are then placed in an m  N block interleaver row-wise. The mapper reads the data out of
the interleaver column-wise and maps the m-bit tuple to a 2m-ary constellation symbol (Ii, Qi), i ¼ 1,2,...,2m, corresponding to
the in-phase and quadrature components which are further modulated onto the optical carrier through the optical I/Q
modulator. For NB-LDPC-CM in Fig. 9.58b, the m parallel binary information streams can be regarded as a single nonbinary
stream over an alphabet of Q ¼ 2m symbols. This Q-ary information stream of length K is encapsulated into a codeword of
length N symbols by a single Q-ary LDPC encoder of code rate R ¼ K/N. The mapper then maps each Q-ary codeword symbol
to a Q-ary (or 2m-ary) constellation point s ¼ (Ix, Qx, Iy, Qy). The following optical modulator operates in the same way as
described before. As shown in Fig. 9.58c, different from NB-LDPC-CM, the multilevel NB-LDPC-CM employs nonbinary
LDPC code with smaller Q, i.e., Q ¼ 2q, while 1 < q < m. The m parallel information bit streams are first grouped into L ¼ m/q
sets of bit streams, while L represents the number of decoding levels.
Each set of bit streams is considered as a stream of 2q-ary symbols. The L 2q-ary information streams of length K are then
encoded into codewords of length N separately by L 2q-ary LDPC encoders and later are block-interleaved through an L  N
block interleaver, similar to that for BI-LDPC-CM. Then each tuple of L 2q-ary symbols is mapped to a 2m-ary constellation
point followed with the optical I/Q modulation. We have noticed that BI-LDPC-CM and NB-LDPC-CM are in fact two
special cases of the proposed multilevel NB-LDPC-CM for q ¼ 1 and q ¼ m, respectively. Obviously, the selection of
q becomes a tradeoff between performance and complexity.
At the receiver, the demapper first calculates the symbol LLRs L(s) for each received symbol with s2{0,1,...,M-1}
indicating the log-likelihood of sending a symbol on the constellation map. Since L-level Q-ary symbols have been used, it
is natural to represent a symbol s by a tuple of L Q-ary symbols as S ¼ {s0,s1,...,sL  1} with si2{0,1,...,Q  1} for any
i 2{0,1,...,L  1}. Corresponding to the configuration of the transmitter, L(s) with s2{0,1,...,M  1} are then converted into
L parallel Q-ary symbol LLRs L(si) with si2{0,1,...,Q  1} and i 2{0,1,...,L  1} according to the following equation:
!
X
M 1
LðsÞ
Lðsi Þ ¼ log e ð9:168Þ
s¼0, si 2S
9.12 Hybrid Multidimensional Coded Modulation Scheme for High-Speed Optical Transmission 601

Each GF(Q) decoder then performs decoding separately after which L Q-ary symbol estimates are converted into the one
M-ary symbol and later m bits to count bit errors. As noticed in our previous analysis, the BI-LDPC-CM and NB-LDPC-CM
schemes serve as two special cases of multilevel NB-LDPC-CM; the above receiver side processing can also be applied to BI-
LDPC-CM and NB-LDPC-CM by setting q ¼ 1 and q ¼ m, respectively. Notice that for BI-LDPC-CM, iterative decoding
between demapper and LDPC decoder can be applied.
Coded modulation with the adoption of nonbinary LDPC codes achieves promising net coding gain (NCG) performance at
the expense of very high complexity in both hardware and computation. As hardware complexity analysis is affected by many
practical tradeoff factors like throughput and parallelization and hence not easy to get a clear expression, analysis of decoding
computation is mostly executed instead. Particularly, an expression of the total number of additions involved in decoding a
quasi-cyclic LDPC codeword based on MD-FFT-SPA (which is already known as hardware-friendly) is given as [82]
 
1
2ρQðN  K Þ q þ 1  , ð9:169aÞ
2ρ

where ρ is the number of nonzero elements in the row of the parity-check matrix. For detailed derivation of the decoding
complexity computation, we refer readers to [82]. We adopt the expression to analyze the complexity reduction of the
proposed scheme over the conventional NB-LDPC-CM. Hence, the complexity ratio (CR) of the conventional NB-LDPC-CM
over that of the proposed scheme is given as

2ρ2m ðN  K Þ m þ 1  2ρ
1
CR ¼ : ð9:169bÞ
m q
q 2ρ2 ðN  K Þ q þ 1  2ρ
1

Figure 9.59 plots the complexity ratio with ρ ¼ 24, q ¼ 2, m ¼ 2,3,...,7 and shows the potential of the ML-NB-LDPC-CM
scheme in reducing computation complexity. Notice that we use q ¼ 2 for the ML-NB-LDPC-CM scheme for the purpose of
revealing the potential of complexity reduction, it may not be the optimal choice in terms of performance, and the selection of
q should also take the performance into account for practical applications.
Figure 9.60 presents the simulated 25-Gbaud 16-QAM optical transmission system in the ASE noise-dominated scenario,
based on which we evaluate the aforementioned three coded modulation schemes. We have applied identical coded
modulation scheme to both polarizations. Parameters for the adopted quasi-cyclic LDPC codes are N ¼ 13,680,
K ¼ 11,970, R ¼ 0.875, and girth 8. The simulated optical transmission data rate is 4  2  25  0.875 ¼ 175 Gb/s. For
BI-LDPC-CM, we use 25 iterations in binary LDPC decoders based on the sum-product algorithm and perform 3 outer
iterations between the demapper and the decoders as iterative decoding is adopted. For both NB-LDPC-CM and multilevel
NB-LDPC-CM, we use 50 iterations for the nonbinary LDPC decoders based on MD-FFT-SPA [82]. For NB-LDPC-CM, we
use 16-ary LDPC code to match the constellation size. For multilevel NB-LDPC-CM, we set q ¼ 2 and employ a couple of

Fig. 9.59 Computation 25

complexity ratio of the
conventional NB-LDPC-CM over
multilevel NB-LDPC-CM with
20
q¼2

15
CR

10

0
2 3 4 5 6 7

m
602 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.60 BER vs. OSNR -1

10
performance of LDPC-coded
PDM-16-QAM optical
transmission -2
10

-3
10

-4
10

BER
-5
10

-6
10 BI-LDPC-CM
NB-LDPC-CM
-7 Multi-level NB-LDPC-CM
10

-8
10
6.4 6.6 6.8 7.0 7.2 7.4 7.6 7.8 8.0 8.2
OSNRb [dB/0.1nm] (per information bit)

identical 4-ary LDPC codes. Figure 9.60 shows the performances of LDPC-coded PDM 16-QAM optical transmission system
in terms of BER vs. OSNRb. As shown in Fig. 9.60, the ML-NB-LDPC-CM outperforms BICM-LDPC-CM by ~0.65 dB,
while achieving nearly identical performance as the conventional NB-LDPC-CM. The results have indicated that the ML-NB-
LDPC-CM scheme has reduced the complexity by adopting low-complexity nonbinary LDPC codes over smaller finite field
while keeping the performance unaffected.

9.13 Multidimensional Turbo Equalization

In this section we describe turbo equalization [107–112] with the employment of LDPC codes named as LDPC-coded turbo
equalization (TE) [2, 10, 12, 107, 108], as a universal scheme that can be used to deal with the imperfectly compensated linear
effects and nonlinear effects. This scheme was initially introduced to deal with ISI in wireless channels [112]. However, given
its exponential complexity, this scheme is more suitable to deal with nonlinear effects rather than linear ones. In fiber-optics
communications, it can be used to deal with imperfectly compensated: (i) fiber nonlinearities, (ii) residual polarization-mode
dispersion (PMD), (iii) residual polarization-dependent loss (PDL), (iii) residual CD, (iv) residual I/Q-imbalance effects in
multilevel coded modulation schemes, and (v) residual mode coupling in FMFs.

9.13.1 Nonlinear Channels with Memory

The multidimensional Bahl-Cocke-Jelinek-Raviv (BCJR) algorithm-based equalizer [12] to be described below operates on a
discrete dynamical trellis description of the optical channel. Notice that this equalizer is universal and applicable to any
multidimensional signal constellation. The BCJR algorithm [13] has been used earlier in this chapter as a MAP decoding
algorithm; however, its application is more general. This dynamical trellis is uniquely defined by the following triplet: the
previous state, the next state, and the channel output. The state in the trellis is defined as sj ¼ (xj  m,xj  m + 1,..,xj,xj + 1,. . .,
xj + m) ¼ x[j  m,j + m], where xk denotes the index of the symbol from the following set of possible indices X ¼ {0,1,. . .,
M  1}, with M being the number of points in corresponding M-ary signal constellation. Every symbol carries b ¼ log2M bits,
using the appropriate mapping rule. The memory of the state is equal to m1 + m2 + 1, with m1 (m2) being the number of
symbols that influence the observed symbol from the left (the right) side. An example of such dynamic trellis of memory
m1 + m2 + 1 for arbitrary multidimensional constellation of size M is shown in Fig. 9.61. The trellis has M m1 þm2 þ1 states, each
of which corresponds to the different (m1 + m2 + 1)-symbol patterns (symbol configurations). The state index is determined by
9.13 Multidimensional Turbo Equalization 603

Fig. 9.61 A portion of trellis to m1 previous m2 previous

be used in multidimensional symbols symbols
BCJR detector with memory
m1 + m2 + 1. The m1 previous 0…0 s’0…0 0/0 s0…0 0…000…0
symbols and m2 next symbols
0…01 s’0…01 2/2

…
relative to the observed symbol
Symbol position to be detected
define the memory of the detector

0…02 s’0…02 s0…020…2 0…020…2

0…03 s’0…03

…
.
. s0…010…0 0…010…0
.

…
0…0100 s’0…010 s0…0120…0 0…0120…0

0…0101 s’0…0101

…
0…0102 s’0…0102 0/2 s0…020…1020 0…020…1020

…
.
. 3/M-1 s0…0 M-1 0 2 0…03 0…0 M-1 20…03

…
.

considering (m1 + m2 + 1) symbols as digits in numerical system with the base M. The left column in dynamic trellis represents
the current states, and the right column denotes the terminal states. The branches are labeled by two symbols, the input symbol
is the last symbol in initial state (the blue symbol), and the output symbol is the central symbol of terminal state (the red
symbol). Therefore, the current symbol is affected by both previous and incoming symbols.
The dynamic trellis shown in Fig. 9.61 is generated by assuming that the symbols are coming from the right and that the
size of window of symbols is m1 + m2 + 1.
For instance, let us observe the symbol 10. . .0 M-3 0. . .023, with the center of the window denoted in bold. For the same
window size, for incoming symbol 5, we will move to the state 0. . .0 M-3 0. . .0235. By using this strategy, the trellis portion
shown in Fig. 9.61 is generated. During detection process instead of detecting whole sequence of length n, we instead detect
the subsequence (the window) of length w > m1 + m2 + 1, with each window containing m2 samples from previous window
and m1 samples from the next window. Since the detection process is performed by splitting the sequence to be detected into n/
w subsequences containing w symbols each, the detection algorithm is called the sliding window. This approach is suitable for
parallelization as n/w BCJR processors can perform detection in parallel.
For the complete description of the dynamical trellis, the transition PDFs p(yj|xj) ¼ p(yj|s), s2S are needed, where S is the
set of states in the trellis and yj is the vector of samples (corresponding to the transmitted symbol index xj). The conditional
PDFs can be determined from collected histograms or by using instanton-Edgeworth expansion method [17]. The number of
edges originating in any of the left-column states is M, and the number of merging edges in arbitrary terminal state is also M.

9.13.2 Nonbinary MAP Detection

Before we describe the LDPC-coded turbo equalization scheme, we provide the basic concepts of optimum detection of
nonbinary signaling in minimum probability of error sense [2, 12, 109]. Let x denote the transmitted sequence and y the
received sequence. The optimum receiver (in symbol error probability sense) assigns bxk to the value x2GF(2m) that maximizes
the APP P(xk ¼ x|y) given the received sequence y as follows:
604 9 Advanced Coding and Coded Modulation Techniques

bxk ¼ arg max Pðxk ¼ xjyÞ: ð9:170Þ

x2GF ð2m Þ

The corresponding detection algorithm is commonly referred to as the MAP detection algorithm. In practice, it is common to
use the logarithmic version of Eq. (9.170a) as follows:
 
Pðxk ¼ xjyÞ
bxk ¼ arg max log ¼ arg max Lðxk ¼ xjyÞ, ð9:171Þ
x2GF ð2m Þ Pðxk ¼ 0jyÞ m
|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl} x2GF ð2 Þ
Lðxk ¼xjyÞ

where L(xk ¼ x|y) is the conditional LLR that the symbol at position k is x, defined as
 
Pðxk ¼ xjyÞ
Lðxk ¼ xjyÞ ¼ log , ð9:172Þ
Pðxk ¼ 0jyÞ

where x is a nonzero element from GF(2m). Clearly, for each position we need to calculate 2m  1 conditional LLRs. (Notice
that in this formulation L(0|y) is always zero.) To calculate the P(xk ¼ x|y) needed in either (9.171) or (9.172), we invoke
Bayes’ rule:

X X PðyjxÞPðxÞ
Pðxk jyÞ ¼ PðxjyÞ ¼ , ð9:173Þ
8x:xk ¼x 8x:x ¼x
PðyÞ
k

where P(y|x) is the conditional probability density function (PDF) and P(x) is the a priori probability of input sequence x, in
which the symbols are independent factors as PðxÞ ¼ ∏ni¼1 Pðxi Þ, where n is the length of the sequence to be detected. By
substituting (9.173) into (9.172), the conditional LLR can be written as
2 P n 3
pðyjxÞ ∏ Pðxi Þ
68x:xk ¼x i¼1 7
Lðxk ¼ xjyÞ ¼ log 6
4 P n
7
5
pðyjxÞ ∏ Pðxi Þ ð9:174Þ
8x:xk ¼0 i¼1

¼ Lext ðxk ¼ xjyÞ þ La ðxk ¼ xÞ,

where the extrinsic information about xk contained in y Lext(xk ¼ x|y) and the a priori LLR La(xk ¼ x) are defined, respectively,
as
2 P n 3
pðyjxÞ ∏ Pðxi Þ
68x:xk ¼x i¼1, i6¼k 7
Lext ðxk ¼ xjyÞ ¼ log 6
4 P n
7,
5
pðyjxÞ ∏ Pðxi Þ ð9:175Þ
8x:xk ¼0 i¼1, i6¼k
 
Pðxk ¼ xÞ
La ðxk ¼ xÞ ¼ log :
Pðxk ¼ 0Þ

From (9.175), it is clear that computation of conditional LLRs can be computationally intensive. One possible computation is
based on BCJR algorithm, with log-domain version when applied to the multilevel/multidimensional modulation schemes
being described in the following section.

9.13.3 Sliding-Window Multidimensional Turbo Equalization

The multidimensional LDPC-coded turbo equalizer is composed of two ingredients: (i) the MAP detector based on multilevel/
multidimensional BCJR detection algorithm and (ii) the LDPC decoder. The multidimensional transmitter configuration is
9.13 Multidimensional Turbo Equalization 605

Fig. 9.62 Multidimensional

LDPC-coded turbo equalization Extrinsic LLRs

…
scheme architecture calculation
D-dim. const.
samples
NB LDPC decoder

Multidimensional

LLRs calculation
pol. diversity Rx
Mode demux +

D-dimensional

BCJR detector

Extrinsic level
GF(2b1) Level 1

demodulator
From SDM

…
…
fiber
NB LDPC decoder
GF(2bm) Level m

identical to one shown in Fig. 9.56a. The receiver configuration of multidimensional LDPC-coded turbo equalizer is shown in
Fig. 9.62.
Clearly, the D-dimensional APP demapper from Fig. 9.56b is replaced by multidimensional BCJR detector, whose
operation principle is described next, and represents the generalization of [2, 107, 108]. The forward metric is defined as
αj(s) ¼ log{p(sj ¼ s,y[1,j])} ( j ¼ 1,2,. . .,n); the backward metric is defined as βj(s) ¼ log{p(y[j + 1,n]|sj ¼ s)}; and the branch
metric is defined as γ j(s0 ,s) ¼ log[p(sj ¼ s,yj,sj  1 ¼ s0 )]. The corresponding metrics can be calculated iteratively as follows
[2, 107, 108]:
0

α j ðsÞ ¼ max
0
α j1 ðs Þ þ γ j ðs0 , sÞ , ð9:176Þ
s

β j1 ðs Þ ¼ max β j ðsÞ þ γ j ðs0 , sÞ , ð9:177Þ

s

γ j ðs0 , sÞ ¼ log p y j jx½ j  m, j þ m P x j : ð9:178Þ

The max*-operator used in (9.176) and (9.177) can be calculated efficiently according to (9.98), while the correction factor log
[1 + exp(|x  y|)] is commonly approximated or implemented using a look-up table. p(yj|x[j  m1,j + m2]) is obtained, by
either collecting the histograms or by instanton-Edgeworth expansion method, and P(xj) represents a priori probability of
transmitted symbol xj.
In the first outer iteration, P(xj) is set to either 1/M (because equally probable transmission is observed) for an existing
transition from trellis given in Fig. 9.61 or to zero for a non-existing transition. The outer iteration is defined as the calculation
of symbol LLRs in multidimensional BCJR detector block, the calculation of corresponding LLRs needed for LDPC
decoding, the LDPC decoding, and the calculation of extrinsic symbol LLRs needed for the next iteration. The iterations
within LDPC decoder are called here inner iterations.
The initial forward and backward metrics values are set to [2, 107, 108]
 
0, s ¼ s0 0, s ¼ s0
α0 ðsÞ ¼ , β n ðsÞ ¼ , ð9:179Þ
1, s 6¼ s0 1, s 6¼ s0

where s0 is an initial state.

Let s0 ¼ x[j  m1  1,j + m2  1] represent the previous state, s ¼ x[j  m,j + m] the present state, x ¼ (x1,x2,. . ., xn) the
transmitted word of symbols, and y ¼ (y1,y2,. . .,yn) the received sequence of samples. The LLR, denoting the reliability, of
symbol xj ¼ δ ( j ¼ 1,2,. . .,n) can be calculated by [2, 107, 108].
  0

L x j ¼ δ ¼ max
0
α j1 ðs Þ þ γ j ðs0 , sÞ þ β j ðsÞ
ðs , sÞ:x j ¼δ
0

ð9:180Þ
 0max α j1 ðs Þ þ γ j ðs0 , sÞ þ β j ðsÞ ,
ðs , sÞ:x j ¼δ0

where δ represents the observed symbol (δ2{0,1,. . .,M  1}\{δ0}) and δ0 is the reference symbol. The forward and backward
metrics are calculated using the (9.176) and (9.177). The forward and backward recursion steps of multidimensional BCJR
606 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.63 Forward/backward ,

j 1( s 0 ) ,
recursion steps for s ,
0 j ( s 0 , s)
multidimensional BCJR detector
of arbitrary signal constellation
size M: (a) the forward recursion , ,
step and (b) the backward j 1( s1 ) j ( s1 , s )
s,
Forward recursion
recursion step 1 s
, , , , ,
j ( s M -2 , s )

…
j s max* j 1( s 0 ) j ( s 0 , s ), j 1( s1 ) j ( s1 , s),...,
,
j 1( s M -2 ) , , , ,
,
sM -2 j 1( s M -1 ) j ( s M -2 , s ), j 1( s M -1 ) j ( s M -1 , s)
,
j ( s 4 , s)

,
j 1( s M -1 )
,
sM -1
(a)
j (s0 )
j ( s', s 0 ) s
0

j ( s ', s1 ) j ( s1 )
s
j 1( s') 1
Backward recursion s'

…
j ( s ', s M -2 )
, ,
j 1( s') max* j (s0 ) j ( s , s 0 ), j ( s1 ) j ( s , s1 ),..., j ( s M -2 )
s
M -2
, ,
j ( s M -2 ) j ( s , s M -2 ), j ( s M -1 ) j ( s , s M -1 )
j ( s ', s M -1 )

j ( s M -1 )
s
M -1
(b)

MAP detector for signal constellation size M are illustrated in Fig. 9.63a, b, respectively. In Fig. 9.63a, s denotes an arbitrary
terminal state, which has M edges originating from corresponding initial states, denoted as s00 , s01 , s02 , . . . , s0M1. Notice that the
first term in branch metric is calculated only once, before the detection/decoding takes place, and stored. The second term, log
(P(xj)), is recalculated in every outer iteration. The forward
  of state s in the j-th step ( j ¼ 1,2,. . .,n) is updated by
metric
preserving the maximum term (in max*-sense) α j1 s0k þ γ j s, s0k (k ¼ 0,1,. . .,M  1). The procedure is repeated for every
state in column of terminal states of j-th step. The similar procedure is used to calculate the backward metric of state s’,
βj  1(s0 ) (in ( j  1)-th step), as shown in Fig. 9.63b, but now proceeding in backward direction ( j ¼ n,n  1,. . .,1).
To improve the overall performance of LDPC-coded turbo equalizer, we perform the iteration of extrinsic LLRs between
LDPC decoder and multidimensional BCJR equalizer.
To reduce the complexity of sliding-window BCJR detector for large constellation sizes and/or larger memories, the
reduced-complexity BCJR algorithm should be used instead. For instance, we do not need to memorize all branch metrics but
several largest ones. In forward/backward metrics’ update, we need to update only the metrics of those states connected to the
edges with dominant branch metrics. Moreover, when max*(x,y) ¼ max(x,y) + log[1 + exp(|x  y|)] operation, required in
forward and backward recursion steps, is approximated by max(x,y) operation, the forward and backward BCJR steps become
the forward and backward Viterbi algorithms, respectively. In systems with digital back-propagation [113, 114], since the
channel memory is reduced before BCJR detection takes place, we can set m2 to zero as shown in [115]. Moreover, after
digital back-propagation, the statistics of samples is Gaussian-like as discussed earlier in this chapter, and instead of histogram
approach to estimate the conditional PDFs, the multivariate Gaussian assumption can be used instead without too much
degradation in performance as shown in [115]. Instead of a single BCJR detector operating on whole sequence to be detected,
several sliding-window BCJR detectors can be operated in parallel.
9.13 Multidimensional Turbo Equalization 607

9.13.4 Nonlinear Propagation Simulation Study of Turbo Equalization

In this section we provide the propagation simulation study results for nonbinary LDPC-coded turbo equalization applied to
rate-adaptive four-dimensional signaling [116]. The transmitter/receiver structures of the 4D nonbinary LDPC-coded modu-
lation (4D-NB-LDPC-CM) are depicted in Fig. 9.64. As introduced before, the source signal entering the transmitter is indeed
composed of m parallel binary information streams of length K bits each. These m parallel bit streams are considered as a
single nonbinary stream over an alphabet of Q ¼ 2m symbols. This Q-ary information stream of length K is encapsulated into a
codeword of length N symbols by a Q-ary LDPC encoder of code rate R ¼ K/N. The mapper maps each Q-ary codeword
symbol to a Q-ary 4D constellation point s ¼ (Ix, Qx, Iy, Qy), where Ix and Qx correspond to the in-phase and quadrature
components in x-polarization, while Iy and Qy correspond to those in y-polarization. Rate adaptation is achieved by using
software to configure the mapper which then selects the proper 4D modulation format (tuning the number of bits per symbol).
The output of the mapper is used to drive the 4D modulator, composed of a distributed feedback (DFB) laser, a polarization
beam splitter (PBS), two I/Q modulators, and a polarization beam combiner (PBC). At the Rx side, the received optical signal
is first split into two polarizations using the PBS, and the resulting signals are fed to two balanced coherent detectors. Outputs
of the coherent detectors are sampled at the symbol rate to obtain the estimates on the coordinates corresponding to the
transmitted symbols and then passed to the equalizer employing BCJR algorithm. Compared to the conventional 2D schemes
where each polarization branch is equipped with a separate BCJR equalizer block, the corresponding 4D scheme uses a single
BCJR equalizer which handles both polarizations simultaneously, and hence, it can compensate for the nonlinear crosstalk
between polarizations, whereas the conventional 2D schemes lack such capability. The symbol LLRs at the output of the
BCJR equalizer are finally forwarded to a Q-ary LDPC decoder matched to the encoder at the transmitter side. The estimates
of the m information sequences sent by the transmitter are then passed to the sink.
The 4D-NB-LDPC-CM scheme is evaluated in the optical transmission system depicted in Fig. 9.64 by using VPI
Transmission Maker v8.5, for dispersion map shown in Fig. 9.65 [116], based on the nonlinear propagation model described
in Section III. Our fiber model takes into account dispersion effects, ASE noise, and impairments due to nonlinear phase noise,
Kerr nonlinearities, and stimulated Raman scattering. To explore the fundamental limits on the transmission distances that
could be achieved by the 4D-LDPC-coded modulation schemes (both binary and nonbinary) and compare them against those
achievable by its counterpart polarization-division multiplexing 2D-LDPC-coded modulation schemes, perfect CD and PMD
compensation are assumed. In order to achieve this in our simulations, we perform inline CD compensation via dispersion-
compensating fibers (DCFs) and adopt polarization tracking. As shown in Fig. 9.65, each span in the dispersion map consists

Fig. 9.64 (a) Transmitter and (b)

Ix Qx
LDPC Encoder

receiver configurations of 4D-NB- Source

To Fiber
4D Mapper
Nonbinary

LDPC-CM
m m DFB MZM
PBS PBC
Laser MZM
Iy Qy
(a)
Ix'
LDPC Decoder

From Coherent
PBS Qx' Output
Nonbinary
Equalizer

fiber Detector
BCJR

m
Iy'
Local Coherent
PBS Qy'
laser Detector
(b)

Fig. 9.65 Dispersion map of the DMF EDFA SMF EDFA

SMF EDFA DCF EDFA
4D-LDPC-CM transmission
system under test
Transmitter Receiver

N spans
608 9 Advanced Coding and Coded Modulation Techniques

of a 40-km-long SMF (dispersion 16 ps/nmkm, dispersion slope 0.08 ps/nm2km, effective cross-sectional area 80 μm2,
attenuation 0.22 dB/km) and a 7.11-km-long DCF (dispersion 90 ps/nmkm, dispersion slope  0.45 ps/nm2km, effective
cross-sectional area 30 μm2, attenuation 0.5 dB/km) each followed by an erbium-doped fiber amplifier (EDFA). For both SMF
and DCF, we set nonlinear refractive index as 2.6  1020 m2/W and PMD coefficient 0.1  1012 s/km1/2. We also applied
pre-compensation of 320 ps/nm and the corresponding post-compensation in our transmission link. The optimal launch
power of the system is found to be near 5 dBm. The operating wavelength and the laser linewidth are set to 1552 nm and
100 KHz, respectively. Bandwidths of the optical and electrical filters are set to 3Rs and 0.75Rs, respectively, where Rs is the
symbol rate which is defined as the information bit rate Ri ¼ 40 Gb/s divided by the FEC code rate R, which is set to R ¼ 0.8,
corresponding to 25% overhead. For efficient encoding and hardware-friendly decoding, we employ regular and QC-LDPC
codes. The software-configured 4D signal constellations are of sizes 16, 32, and 64, denoted as 16-, 32-, and 64-4D,
respectively. The 16-4D constellation is described by the set { 1, 1, 1, 1}, which is indeed the set of vertices of a
tesseract (i.e., a regular octachoron). Also, each EDFA in Fig. 9.65 has a noise figure of 5 dB.
The coherent receiver is composed of an optical filter, two π-hybrids, eight photodiodes, four electrical filters, and four
samplers followed by a BCJR equalizer and LDPC decoder(s). In Fig. 9.66a, we present first the BER against transmission
distance performances of 4D block-interleaved LDPC-coded modulation (4D-BI-LDPC-CM) and the corresponding PDM 2D
block-interleaved LDPC-coded modulation (PDM-2D-BI-LDPC-CM) schemes under the dispersion map given in Fig. 24.
Our comparisons are performed against conventional modulation formats, i.e., 16-4D vs. PDM-QPSK and
64-4D vs. PDM-8-QAM. As shown in Fig. 9.66a, we achieve more than 8000 km with 16-4D at 160 Gb/s aggregate bit
rate and 2800 km with 64-4D at 240 Gb/s aggregate bit rate. The corresponding PDM-QPSK and 8-QAM distances are limited
to around 6000 km and 2000 km, respectively. Similar observation can be obtained from the comparison between 4D-NB-
LDPC-CM and PDM 2D nonbinary LDPC-coded modulation (PDM-2D-NB-LDPC-CM) in Fig. 9.66b. As revealed by both
figures, 4D scheme exhibits solid advantage over the corresponding PDM-2D scheme. Specifically, such advantage is even
more significant when nonbinary LDPC-coded modulation is adopted. To put it numerically, 4D-NB-LDPC-CM maintains
more than 2000 km additional transmission reach over PDM-2D-NB-LDPC-CM during transmission at aggregate bit rates of
both 160 Gb/s and 240 Gb/s. Such an improvement in transmission reach is just a well-grounded justification why 4D schemes
should be adopted rather than PDM-2D schemes for long-haul transmission.
In order to further manifest the superiority of 4D-NB-LDPC-CM over 4D-BI-LDPC-CM, we performed comparisons
between the two for three different 4D modulation formats, i.e., 16-, 32-, and 64-4D. Our results are presented in Fig. 9.67.
The 4D-NB-LDPC-CM significantly outperforms 4D-BI-LDPC-CM especially as the underlying constellation size, and
hence the spectral efficiency, and hence the aggregate bit rates increase. For example, for transmission at 240 Gb/s, 4D-NB-
LDPC-CM can provide more than 1500 km additional reach compared to 4D-BI-LDPC-CM. In addition to improving
transmission reach compared to 4D-BI-LDPC-CM, 4D-NB-LDPC-CM can also improve the latency at the receiving ends
since it avoids costly turbo equalization steps which pass through the inherently complex BCJR equalization block about 3 to
5 times for acceptable performance in BI-LDPC-CM schemes.

9.13.5 Experimental Study of Time-Domain 4D-NB-LDPC-CM

The transmitter/receiver architectures of time-domain 4D-NB-LDPC-CM, introduced in [117], are depicted in Fig. 9.68. The
source signal entering the transmitter is composed of m parallel binary information streams of length K bits each. Again, we
consider these m parallel bit streams as a single nonbinary stream over an alphabet of Q ¼ 2m symbols. This Q-ary information
stream of length K is encapsulated into a codeword of length N symbols by a Q-ary LDPC encoder of code rate R ¼ K/N. The
mapper maps each Q-ary codeword symbol to a Q-ary 4D constellation point (Ix,t, Qx,t, Ix,t + 1, Qx,t + 1). We then serialize the
four coordinates into two pairs, (Ix,t, Qx,t) and (Ix,t + 1, Qx,t + 1), each of which will occupy half symbol duration. The
coordinates Ix (Iy) and Qx (Qy) correspond to the in-phase and quadrature components in x-polarization (y-polarization). The
output of P/S is then used to generate RF I and Q signals in both time instances t and t + 1 to drive the I/Q modulator.
A typical coherent receiver and standard offline DSP algorithm were used for recovering the original transmitted signals, I 0x
, Q0x, I 0y, and Q0y. Corresponding to the transmitter, each continuous two pairs of (I 0x, Q0x) is grouped as (I 0x,t , Q0x,t , I 0x,tþ1 , Q0x,tþ1),
which represents the estimate of a received 4D symbol and is then passed to the 4D equalizer employing BCJR algorithm, as
described above. The symbol LLRs at the output of the BCJR equalizer are finally forwarded to a Q-ary LDPC decoder
matched to the encoder at the transmitter side. The estimates of the m information sequences sent by the transmitter are then
obtained. The information sequences transmitted on the y polarization are recovered in a similar fashion.
9.13 Multidimensional Turbo Equalization 609

Fig. 9.66 BER vs. transmission

distance comparisons between (a)
-1
4D-BI-LDPC-CM and PDM-2D- 10
BI-LDPC-CM and (b) 4D-NB-
LDPC-CM and PDM-2D-NB-
LDPC-CM

Bit-error rate, BER

-2
10

-3
10

16-4D
-4
64-4D
10 PDM-QPSK
PDM-8QAM
600 km 2300 km
-5
10
2000 4000 6000 8000 10000
Total transmission distance, L [km]
(a)

-1
10
Bit-error rate, BER

-2
10

-3
10
16-4D
64-4D
-4
PDM-QPSK
10
PDM-8QAM
2100 km 2700 km
-5
10
2000 4000 6000 8000 10000

Total transmission distance, L [km]

(b)

The counterpart of the time-domain 4D-LDPC-CM scheme is PDM nonbinary LDPC-coded modulation, which employs
the conventional 2D QAM. The PDM-QPSK system operates using two 2D equalizers one for each time interval under
consideration. On the other hand, the 16-4D system operates with one 4D equalizer. Both 2D and 4D equalizers operate on
trellis description of the optical channel with memory. Therefore, the 4D equalization is able to compensate for the
intrachannel nonlinear effects, which expand over the time domain.
610 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.67 BER against

transmission distance comparison
-1
between 4D-NB-LDPC-CM and 10
4D-BI-LDPC-CM

Bit-error rate, BER

-2
10

-3
10 NB-16-4D
BI-16-4D
NB-32-4D
10
-4 BI-32-4D
NB-64-4D
BI-64-4D
-5
10
4000 6000 8000 10000

Total transmission distance, L [km]

Fig. 9.68 Time-domain 4D-NB-

LDPC-CM: (a) transmitter and (b)
receiver architecture. P/S parallel-
to-serial, S/P serial-to-parallel

Figure 9.69 shows the experiment setup of the eight-channel DWDM optical transmission system at 50 GHz channel
spacing as well as the optical spectrum at the transmitter and receiver sides. The eight 100-kHz line-width external cavity
lasers with λ1 ¼1550.12 nm and λ8 ¼1550.40 nm are divided into odd and even groups, each of which is modulated with
16 Gbaud 4D signals. The 4D signals are generated by digital-to-analog converters (DACs) using coordinates that are
produced according to the time-domain 4D-LDPC-CM described before.
9.13 Multidimensional Turbo Equalization 611

Fig. 9.69 Experimental setup (top figure) and optical spectra (at 0.1 nm RBW): (a) Tx output WDM signal, (b) Tx output signal in C-band, (c)
WDM signal after 7680 km. ECL external cavity laser, DAC digital-to-analog converter, OTF optical tunable filter, LO local oscillator laser, BPD
balanced photodetector

A de-correlated copy of signal is combined with its original signal for emulating PDM signals. The odd and even channels
are then combined and launched into the fiber recirculating loop that consists of 8  60-km spans of DMF with an inline
EDFA after each span. Signal loading and circulation are controlled by a mechanical transmitter switch and loop switch. At
the receiver side, the desired channel is extracted by an optical tunable filter and down-converted to electrical baseband
domain by combining it with a local laser centered at the proper frequency using a polarization-diversity 90 optical hybrid
followed by balanced photodetectors. The equalized symbols are forwarded to the 4D equalizer to compensate the
intrachannel nonlinearities and later decoded by the nonbinary LDPC decoder. In this experiment, the 4D modulation format
of constellation size 16 (16-4D) and 16-ary (16935, 13550) LDPC code achieve 16  109  4  2  0.8 ¼ 102.4 Gb/s per
channel and 2.05 b/s/Hz. The adopted decoding algorithm is MD-FFT-SPA [82]. For comparison, we also transmit the
counterpart 32-Gbaud PDM-QPSK signals with NB-LDPC-CM, which adopts conventional 2D equalizer and 4-ary (16935,
13550) LDPC code at the same data rate and spectral efficiency.
Figure 9.70 shows the back-to-back BER performance of time-domain 4D-LDPC-CM of 16-4D and NB-LDPC-CM of
PDM QPSK, in which the former is ~0.4 dB better, at BER of 104, than the latter in terms of OSNR due to the adoption of
16-ary LDPC code instead of 4-ary. The result confirms that nonbinary LDPC codes in a larger Galois Field exhibit stronger
error correction capability.
To further illustrate the advantage of time-domain 4D-LDPC-CM in compensating intrachannel nonlinear effects, we
present the transmission performance in terms of Q-factor (derived from BER) vs. transmission distance in Fig. 9.71. We
show Q-factor values of the received 16-4D and PDM-QPSK signals before and after the corresponding equalization. For
PDM-QPSK with NB-LDPC-CM, the 2D equalizer does not help improve Q-factor, while for 16-4D with 4D-LDPC-CM,
Q-factor is improved by ~0.8 dB after 4D equalization as the intrachannel nonlinear effects are compensated by the 4D
equalizer. Namely, because 4D equalizer operates in 4D space, with basis functions being two neighboring symbols and
in-phase and quadrature channels, it can compensate for the nonlinear interaction between neighboring symbols. In this
work, we collect the samples of 4  106 symbols after every six spans. No countable error is found after transmitting
7680 km at the output of LDPC decoder for the 4D scheme. In contrast, the NB-LDPC-CM with PDM QPSK can only reach
6240 km. Therefore, the optical transmission reach has been extended by 1440 km by employing the time-domain
4D-LDPC-CM.
612 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.70 BER vs. OSNR

performance of time-domain -1 Uncoded
4D-LDPC-CM of 16-4D and NB- 10
LDPC-CM of PDM-QPSK NB-LDPC-CM PDM-QPSK
4D-LDPC-CM 16-4D
-2
10

-3

BER
10

-4
10

-5
10

9 10 11 12 13 14 15 16 17 18 19
OSNR [dB/0.1nm]

Fig. 9.71 Q-factor vs. distance

performance of time-domain 10 PDM-QPSK:
16-4D and PDM-QPSK before Before Equalizer
and after BCJR equalization
9 After Equalizer
16-4D:
8 Before Equalizer
After Equalizer
7 LDPC threshold:
Q [dB]

4-ary code
6 16-ary code

2880 3840 4800 5760 6720 7680

Distance [km]

9.13.6 Quasi-Single-Mode Transmission Over Transoceanic Distances Using Few-Mode Fibers

High-speed optical systems rely solely on single-mode fibers for transmission beyond a few kilometers reach, and state-of-the
art single-mode fibers are required to achieve spectral efficiencies beyond 6 b/s/Hz over transoceanic distances. Recently,
there has been a renewed interest in transmission over FMFs to increase the capacity over a single fiber strand by the multitude
of the spatial modes. However, such an approach requires a radical transformation of optical communication infrastructure
which has been built on the premise of transmission over a single spatial mode. It was proposed that FMFs can be integrated
9.13 Multidimensional Turbo Equalization 613

into the otherwise single-mode infrastructure by transmitting only in the fundamental mode, while taking advantage of their
significantly larger effective area. The feasibility of using FMFs was demonstrated in a recent transmission experiment over
2600 km. In our article [118], 101.6-km-long hybrid spans consisting of low-loss silica-core FMFs and single-mode fibers are
used to achieve a new record spectral efficiency of 6.5 b/s/Hz in unidirectional transmission configuration over the
transoceanic distance of 6600 km using EDFAs only. To the best of our knowledge, this is the first time FMFs are used to
outperform state-of-the-art single-mode fibers in a transoceanic transmission experiment. To make the FMFs compatible with
single-mode transmission, single-mode fiber jumpers are spliced to each end. This excites only the fundamental mode of the
FMF at launch and also strips any high-order modes excited during transmission at the exit splice. Such systems may suffer
from multipath interference (MPI) even with inherently low-mode-coupling FMFs. In effect, by using FMFs the nonlinear
fiber impairments are traded for linear impairments from MPI which are easier to manage. While the improvements possible
from mitigating nonlinear impairments by DSP are fundamentally limited in scope and computationally costly, it was
demonstrated recently that most of the penalty from the MPI can be mitigated with much less computational complexity by
including an additional decision-directed least mean-square (DD-LMS) equalizer to the standard digital coherent DSP. The
differential modal group delay (DMGD) of a FMF can be of the order of 1 ns/km, and the required number of taps in the
DD-LMS equalizer may be several hundreds to thousands, which not only makes them less stable and computationally
demanding but also degrades its performance. In this work, instead of using standard single-carrier modulation (SCM), a
digital multi-subcarrier modulation (MSCM) signal with 32 subcarriers at 1 Gbaud is used to reduce the number of equalizer
taps per subcarrier by more than an order of magnitude while increasing the temporal extent of MPI that can be mitigated. To
further reduce MPI, a hybrid FMF-single-mode fiber span design is used for the first time with the FMF at the beginning of the
span to reduce nonlinearity and the single-mode fiber at the end to avoid excessive MPI. Hybrid spans also reduce the DSP
complexity as they have smaller modal group delay per span compared to all FMF spans.
The experimental setup is shown in Fig. 9.72. At the transmitter side, digital-to-analog converters (DACs) with a sampling
rate of 64 GHz are used to generate WDM channels tuned to a 33 GHz grid. WDM channels are prepared in two groups. The
first group consists of six neighboring channels generated with tunable ECLs. These six channels are tuned together across the
C-band, and only the center channel which uses a narrow linewidth (<1 kHz) laser is measured. Each of the even and odd
subgroups of the six tunable channels is modulated separately by four independent streams of data for all in-phase and
quadrature (I/Q) rails in both polarizations. The 111 loading channels are modulated with only independent I/Q rails followed
by polarization multiplexing emulators. The tunable channels and the dummy channels are combined using a wavelength
selective switch (WSS) with a 1-GHz grid resolution and a coupler. Using the DACs, either a 32 Gbaud, Nyquist-shaped SCM
16-QAM, or a MSCM signal of 32  1Gbaud 16-QAM subcarriers with 10 MHz guard-band is generated per wavelength.
The 218–1 PRBS binary data are encoded by 16-ary irregular quasi-cyclic nonbinary LDPC encoder of girth 10 to generate
codewords at three overheads 14.3%, 20%, and 25%, to maximize the capacity across C-band. The codeword length is 13,680
for 14.3% and 20% OHs and 74,985 for 25% OH. Encoder output is uniformly distributed into 32 subcarriers of MSCM
signal.
At the receiver side, the WDM channel under test is filtered and captured with a standard offline coherent receiver, using a
narrow-linewidth (<1 kHz) laser as the local oscillator and a real-time 80 GSa/s sampling scope. In the case of a SCM signal,
after resampling and chromatic dispersion compensation (CDC), the signal is fed into a multi-modulus algorithm for initial
convergence of polarization demultiplexing followed by carrier phase recovery. Subsequently a second stage of T-spaced
equalizer using DD-LMS is applied to the 32Gbaud SCM signal to mitigate MPI. In the case of MSCM, after frequency offset
estimation and CDC, each subcarrier is digitally filtered and processed individually in the same way as the single-carrier
signal. The data of all subcarriers is combined and is further improved by the BCJR equalization and then sent to nonbinary
LDPC decoder for detecting post-FEC bit errors.
The circulating loop test bed is composed of five hybrid spans obtained by splicing 51.3-km-long FMFs with 50.3-km-long
Vascade® EX3000 fiber with average attenuations of 0.157 dB/km and 0.153 dB/km, respectively. The FMFs support the
LP11 mode albeit with a slightly higher attenuation, and the differential modal delay is estimated to be 0.95 ns/km with the
LP01 lagging. The average effective areas of the single-mode fiber and the FMF portions are 151 μm2 and an estimated
200 μm2, respectively. A short piece of Vascade® EX3000 is used as a bridge fiber between the input end of the FMF and the
standard-single-mode fiber pigtail of the EDFA output. The average span loss is 16.48 dB corresponding to an additional
0.7 dB incurred by the four splices and the two connectors. The mid-band chromatic dispersion is approximately 21 ps/nm/
km. Span loss is compensated by C-band EDFAs, and the accumulated amplifier gain tilt is compensated by a WSS each loop.
The Q-factor is measured across the C-band at 6600 km for a total of 111 WDM channels, from 1534.5 nm to 1563.6 nm.
Q-factors averaged over 32 carriers, both polarizations, and over 10 separate measurements are plotted in Fig. 9.73 along with
the received spectrum. The Q-factor varies from 4.5 dB to 5.7 dB from the shorter to the longer wavelengths. To maximize the
614 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.72 System setup: (a) DAC11 DAC12

transmitter with 6 tunable
66 GHz
channels and 111 loading
channels, on a 33 GHz grid; (b) 55 DFB I/Q Pol.
recirculating loop with 5  101.6- Lasers Mod MUX
Coupler
km-long hybrid spans of FMF and EDFA DAC8 DAC9
Vascade EX3000 WSS
66 GHz
56 DFB I/Q Pol.
Lasers Mod MUX

DAC1 DAC2 DAC3 DAC4

Odd
3 Dual-polarization
ECLs I/Q mod
Coupler
DAC5 DAC6 DAC7 DAC8
Even
3 Dual-polarization
ECLs I/Q mod

(a)
Switch (SW)
Coherent
OTF
Rx

SW

Pol.
GEQ
scrambler Vascade FMF
EX3000 5

(b)

capacity with such a variation of Q-factor, three different FEC OHs at 14.3%, 20%, and 25% are used for 28, 64, and 19 WDM
channels across the C-band, as shown in Fig. 9.73, resulting in an average spectral efficiency of 6.5 b/s/Hz. For each channel,
more than 25 million bits were processed for LDPC decoding, and all the channels were decoded error-free in the LDPC
decoder assisted with BCJR equalizer. The coding performance is further evaluated in the BTB configuration as a function of
pre-FEC Q-factor derived from bit error counting as shown in Fig. 9.74. To achieve sufficient confidence level, over
100 million bits are considered for each BER point, and exponential fitting is used for extrapolating the measured BER
data and thus estimating the FEC limit as 5.5 dB, 4.9 dB, and 4.35 dB Q-factor for 14.3%, 20%, and 25% OH, respectively.
To make direct performance comparisons, we next constructed three comparable loops from either all single-mode fiber, all
FMF, or hybrid spans. Four of the hybrid spans are broken up and re-spliced to obtain either two spans of all-single-mode fiber
spans or all-FMF spans with almost identical span lengths and losses. For the hybrid case, three spans are removed from the
loop. Figure 9.75 shows the comparison after 10 and 20 circulations using the same MSCM signals and the same receiver DSP
with 101-tap-long CMA-based filer. Because the FMFs have larger effective area, the optimum OSNR is larger by about 1 dB
compared to the all-SMF case. However, some of this improvement is lost because the impact of MPI cannot be removed
altogether which is more evident in portions of the curves dominated by the linear noise, especially when all-FMF spans are
used. This is most clear at 4060 km in the low OSNR regime. However, for the hybrid span case, the nonlinear improvement
from the large effective area more than compensates for the penalty due to residual MPI affording 0.4 dB improvement over
the all-single-mode fiber configuration. For additional details on this experiment, an interested reader is referred to [118].
9.13 Multidimensional Turbo Equalization 615

Fig. 9.73 (Top)

Q-factor vs. wavelengths with 5.8
14.3% (green), 20% (blue), and
25% (red) FEC OH after 6600 km. 5.5

Q (dB)
(Bottom) Received spectrum.
(After ref. [118]; # IEEE 2015; 5.2
reprinted with permission)
4.9
4.6
4.3
Power (dB) -30
-40
-50 Received Spectrum

-60
1535 1540 1545 1550 1555 1560 1565
Wavelength (nm)

Fig. 9.74 Back-to-back FEC

coding performance. (After ref.
[118]; # IEEE 2015; reprinted
with permission) -3
10
Post-FEC BER

-6
10

-9
10

-12
10 14% OH
20% OH
-15
25% OH
10
3.5 3.7 3.9 4.1 4.3 4.5 4.7 4.9 5.1 5.3 5.5
Pre-FEC Q (dB)
616 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.75 Q-factor versus OSNR

at 2030 km and 4060 km, all- 9
single-mode fiber (blue), hybrid
(red), and all FMF (black). (After
ref. [118]; # IEEE 2015;
reprinted with permission)
8

7 2030 km

Q (dB)
6

4 4060 km

All-single-mode-fiber
3 Hybrid
All-few-mode fiber
2
16 17 18 19 20 21 22 23 24 25 26 27
Received OSNR @ 0.1 nm (dB)

9.14 Optimized Signal Constellation Design and Optimized Bit-to-Symbol Mapping-Based

Coded Modulation

Different optimization criteria can be used in optimum signal constellation design including minimum mean-square error
(MMSE) [119, 120], minimum symbol error rate [121], maximum information rate [122], and maximum LLR [123]. Recently,
in [120], we used the iterative polar quantization (IPQ) procedure that minimizes the mean-squared quantization error of
optimum source distribution. For non-Gaussian channels, Arimoto-Blahut algorithm (ABA) [124] should be used to
determine optimum source distribution. For an ASE noise-dominated channel, the optimum information source, maximizing
the mutual information, is Gaussian:
 2 
1 I þ Q2
pðI, QÞ ¼ exp  , ð9:181Þ
2πσ 2 2σ 2s

where (I,Q) denote in-phase and quadrature coordinates and σ 2s is the variance of the source. By expressing Cartesian
coordinates by using polar coordinates (r,θ) as follows I ¼ rcosθ and Q ¼ rsinθ, it is straightforward to show that radius will
follow Rayleigh distribution, while phase distribution is uniform. This indicates that the optimum distribution, in mutual
information (MI) sense, contains constellation points placed on circles of radii that follow Rayleigh distribution. The number
of points per circle is nonuniform, wherein the distribution of point on a circle is uniform. For corresponding algorithm an
interested reader is referred to [119, 120], which is accurate for medium and large signal constellation sizes. Namely, the
approximation used in algorithm described in [119, 120] is not very accurate for small constellation sizes. This signal
constellation has been extended in [122] to arbitrary signal constellation sizes and to multidimensional signal space
[125]. Since it closely approaches the channel capacity in ASE noise-dominated scenario, it has been called the optimum
signal constellation design (OSCD), which is optimum in MI sense. Namely, the OSCD algorithm is formulated on such a way
to maximize the MI for ASE noise-dominated scenario, which justifies the name introduced. The OSCD has been extended in
[126, 127] to the phase noise-dominated channels. The optimum mapping rule design for OSCDs is the subject of ref.
[125]. Finally, in [128] the vector quantization concepts have been used to large signal constellation design for multidimen-
sional signaling.
9.14 Optimized Signal Constellation Design and Optimized Bit-to-Symbol Mapping-Based. . . 617

9.14.1 Multidimensional Optimized Signal Constellation Design

Given the space limitations, here we briefly describe the OSCD for multidimensional signaling, the optimum mapping rule
determination, as well as the EXIT chart analysis [125] of OSCDs. In Sect. 9.14.3, we describe the nonlinear optimized signal
constellation design (NL-OSCD), suitable for SPM-dominated channels. After the initialization stage by signal constellation
obtained by sphere-packing method, we generate the multidimensional training sequences from the optimum source
distribution and split them into the clusters of points according to the Euclidean distance squared from constellation obtained
in previous iteration. New constellation points are obtained as the center of mass of such obtained clusters. This procedure is
repeated until convergence or until a predetermined number of iterations have been reached.
As an illustration in Fig. 9.76, we provide the 8-ary and 16-ary 2D signal constellations obtained by using the OSCD
algorithm for ASE noise-dominated scenario as well as the optimized mapping rules. Corresponding 3D constellations and
mapping rules are provided in Fig. 9.78. The 3D constellations are relevant in SDM-based applications.

9.14.2 EXIT Chart Analysis of OSCD Mapping Rules

For the convergence behavior analysis, the EXIT chart analysis should be performed [101]. In order to determine the MI
transfer characteristics of the demapper, we model the a priori input LLR, LM,a, as a conditional Gaussian random variable.
The MI between c (c denotes the initial bits) and LM,a is determined numerically as explained in [101]. Similarly, the MI
between c and demapper extrinsic LLR, denoted as LM,e, is also calculated numerically, but with the PDF of c and LM,
e determined from histogram obtained by Monte Carlo simulation. By observing the MI of demapper extrinsic likelihoods ILM,
e as a function of the MI of the prior LLRs LM,a, denoted as ILM,a, and OSNR, in dB, the demapper EXIT characteristic
(denoted as TM) is given by
 
I LM,e ¼ T M I LM,a , OSNR : ð9:182Þ

The EXIT characteristic of LDPC decoder (denoted as TD) is defined in a similar fashion as
 
I LD,e ¼ T D I LD,a , ð9:183Þ

where we use ILD,e to denote MI of decoder extrinsic likelihoods and ILD,a to denote the MI of decoder the prior LLRs.
The EXIT chart analysis of optimized mapping, optimized as described in [125] (and references therein), for 2D 8-ary
OSCD is shown in Fig. 9.77a. Namely, the optimization criterion for these mapping rules can be summarized as follows [125]:
find the mapping rule in such a way that every two symbols that differ in only one bit position has larger distance than the
minimum Euclidean distance of the constellation. The channel OSNR (for symbol rate of 25 GS/s) equals 5.0 dB, and we use a
QC-LDPC (16935, 13550) decoder. It is obvious to see that in this case only 8-ary OSCD constellation can operate with the
turbo decoding process. Other mapping rules, for this OSNR, will result in turbo decoding errors. In Fig. 9.78a, we can see the
tradeoff performance for MI in optimized mapping. The MI at the output of the demapper increases over iterations when using
the optimized mapping. The corresponding EXIT chart for 8-ary 3D OSCD is shown in Fig. 9.78b.
The optimized mappings for OSCDs are studied for use in adaptive polarization-division multiplexed/mode-
division multiplexed LDPC-coded modulation schemes to enable both 400GbE and 1 TbE. The channel symbol rate was
set to 31.25 GS/s, and QC-LDPC (16935, 13550) code of girth 8 and column weight 3 was used in simulations. All results are
obtained for 20 LDPC decoder (inner) iterations and 3 APP demapper-LDPC decoder (outer) iterations and summarized in
Fig. 9.79. Clearly, the OSCD constellations outperform QAM constellations in both 16-ary and 8-ary cases. Also, the
optimized mapping rules outperform randomly selected mapping for OSCDs, and the gap, at BER of 108, is almost
0.5 dB in the 8-ary case and 0.6 dB in the 16-ary case. Note that we used natural mapping for 8-QAM and 16-QAM
constellations.
618 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.76 The optimum signal

constellations and mapping rules
for (a) 8-ary and (b) 16-ary 2D
signal space

(a)

(b)
9.14 Optimized Signal Constellation Design and Optimized Bit-to-Symbol Mapping-Based. . . 619

Fig. 9.77 3D (a) 8-ary and (b)

16-ary optimum signal
constellations and mapping rules.
X, Y, Z are used to denote three 011
either electrical (Slepian 1.5
sequences) or optical (spatial 111
modes) basis functions to be used
in SDM applications 1 000

0.5
010
Z
0 100

-0.5
110
-1 101
001 1
-1.5
1.5 0
1 0.5 0 -0.5 -1
-1 -1.5
X
Y
(a)

(b)
620 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.78 EXIT chart analysis 1.0

for (a) 8-ary 2D OSCD and (b)

channel and extrinsic ourput ID,a of demapper

8-ary 3D OSCD. Random
mapping: randomly selected
mapping rule
0.8

0.6

Channel OSNR=5.0 dB
0.4 8OSCD Optimized mapping
8OSCD Random mapping
8PSK Gray mapping
8PSK Natural mapping
0.2
8QAM Natural mapping
The Decoder characteristic
16935 LDPC Decoder
0.0
0.0 0.2 0.4 0.6 0.8 1.0
a priori input ID,e to demapper

(a)
1.0
channel and extrinsic ourput ID,a of demapper

0.8

0.6
Channel OSNR=3.0 dB
(16935,13550) LDPC decoder
3D-8-Sphere packing
3D-8-OSCD with Random mapping
3D-8-OSCD with Optimal mapping
0.4
0.0 0.2 0.4 0.6 0.8 1.0
a priori input ID,e to demapper
(b)
9.14 Optimized Signal Constellation Design and Optimized Bit-to-Symbol Mapping-Based. . . 621

Fig. 9.79 BER performance for 10

-1

optimal mapping rules of LDPC-

coded PDM 2D-OSCDs. PDM,
polarization-division multiplexing 10
-2

-3
10
P D M L D P C -c ode d Q A M:

B it-to-erro r ra te , B E R
-4
10 1 6 -Q A M
8 -Q A M
-5
10 P D M L D P C -c ode d OSCD:
1 6 -Random mapping
10
-6
1 6 -O ptim iz e d m a pping
( by C os t F unc tion)
10
-7 8 -Random mapping
8-O ptim iz e d m a pping
-8
10

-9
10
4 5 6 7 8 9 10

O ptic a l s ig na l to nois e ra tio, O S N R [dB ] (per bit)

9.14.3 Nonlinear Optimized Signal Constellation Design-Based Coded Modulation

In this subsection, we describe NL-OSCD algorithm suitable for SPM dominated channel. The NL-OSCD algorithm can be
summarized as follows [127]:

0. Initialization: Choose the signal constellation of size M that will be used for initialization.
1. Generate the training sequence from the optimum source distribution, denoted as {xj ¼ (xj,I, xj,Q)| j ¼ 0,. . .,n-1}.
2. Group the samples from this sequence into M clusters. The membership to the cluster is determined based on LLR of
sample point and candidate signal constellation points from previous iteration. Each sample point is assigned to the
cluster with the largest log-likelihood (LL) function, defined below. Given the m-th cluster (subset) A bm ¼
   
y j ¼ y j,I , y j,Q j j ¼ 0, 1, . . . , find the LL function of the m-th cluster as follows:

X
n1
LLm ¼ n1 max Lðxk , yÞ, ð9:184Þ
k¼0 y2b
Am

where L(xj,y) is the nonlinear phase noise-averaged LLR defined as

(   jΦ 
2
1 XN s 1 xk,I  Re yI þ yQ j e NL,i
Lðxk , yÞ ¼ 
Ns i¼0 2δ2
  
2 ) ð9:185Þ
xk,Q  Im yI þ yQ j ejΦNL,i
þ
2δ2

where Ns denotes the number of nonlinear phase noise samples ΦNL, i.

3. Determine the new signal constellation points as center of mass for each cluster.
622 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.80 The optimized 2D 0.02

16-ary signal constellations for the
SPM-dominated scenario (after
2000 km of SMF and for launch
power of – 6 dBm)

Quadrature coodinate Qx (Qy)

0.01

0.00

-0.01

-0.02
-0.02 -0.01 0.00 0.01 0.02
In-phase coordinates Ix (Iy)

Repeat the steps 2–3 until convergence is achieved. Equation (9.185) is written by having in mind a 2D signal constellation
design, but it can straightforwardly be generalized for arbitrary dimensionality.
As an illustration, the 16-ary NL-OSCD constellation, obtained for launch power of 6 dB and transmission distance of
2000 km, is shown in Fig. 9.80. The optimal launch power (in uncoded BER sense) for 16-ary NL-OSCD is 4 dBm, as shown
in Fig. 9.81a, and the total transmission distance for LDPC-coded 16-NL-OSCD exceeds 4400 km, as shown in Fig. 9.81b.
Figure 9.81 indicates that 16-ary NL-OSCD has better nonlinear tolerance than 16-QAM. On the other hand, Fig. 9.81b
indicates that LDPC-coded 16-ary NL-OSCD outperforms 16-QAM by almost 1000 km. The signal constellations obtained
by using the NL-OSCD algorithm clearly have better nonlinear tolerance and can significantly extend the total transmission
distance of QAM. The main reason for this improvement is due to the fact that during the signal constellation design, the
nonlinear nature of the channel is taken into account.

9.14.4 40 Tb/s Transoceanic Transmission Enabled by 8-ary OSCD

By using 8-ary OSCD, we have demonstrated in [129] a record to date 714 Gb/s capacity-per-amplifier (40 Tb/s in aggregated
data rate) transoceanic transmission over 6787 km is demonstrated. The system is based on dual-carrier 400G transmission
with C + L-band EDFA amplification only (at 121 km span length). In this subsection, we have briefly described this
experimental verification.
Figure 9.82 outlines the experimental setup. At the transmitter side, a total of 109 C-band DFB lasers spanning from
1529.88 to 1564.55 nm and 91 L-band lasers spanning from 1572.40 to 1602.81 nm at 40.2GHz spacing are used and
independently modulated by the data bar outputs of 4 64Gs/s DACs. For the measurement, instead of DFB lasers, eight
external cavity lasers (ECLs) are separated into odd and even channels and are modulated separately by four 64Gs/s digital-to-
analog converters (DACs) driving two I/Q modulators to produce 40Gbaud Nyquist-shaped NB-LDPC-coded 8-ary OSCD
signals. First, a PRBS of order 218–1 was encoded by the 8-ary irregular quasi-cyclic LDPC (81126, 67605) encoder of girth
10 to generate codewords at 20% overhead and then mapped into 8-ary OSCD symbols via optimal mapping described above.
For additional details on experimental setup including recirculating loop, an interested reader is referred to [129].
The offline DSP algorithm digitally compensates for the coarse frequency offset between the transmitter and LO lasers and
then filters out the testing carrier and resamples the signals back to two samples per symbol. After resampling, either CD
compensation (CDC) is performed in single stage in frequency domain or nonlinear compensation (NLC) is applied using
digital back-propagation with one step per span [130]. Received signal is then fed into multi-modulus algorithm for initial
convergence of polarization demultiplexing. Due to the non-regular constellation layout for the 8-ary OSCD, the blind phase
searching method [131, 132] is applied for carrier phase recovery in the stage of DD-LMS algorithm. The recovered symbols
are then sent to nonbinary LDPC decoder for detecting post-FEC bit errors. MD-FFT-SPA is employed with 50 iterations
or less.
9.14 Optimized Signal Constellation Design and Optimized Bit-to-Symbol Mapping-Based. . . 623

Fig. 9.81 (a) Uncoded

BER vs. launch power for 16-ary
NL-OSCD and 16-QAM, (b) -1
BER vs. transmission length plot 10
for LDPC-coded 16-ary
NL-OSCD and 16-QAM.
LLR-OSCD: constellation

Bit Error Rate

obtained for optimization in the
presence of linear phase noise. -2
The parameters (44000) mean that 10
launch power was 4 dBm and
target distance for optimization
was 4000 km

-3
10

Length = 2000 km
QAM
-4
NL-OSCD(4,4000)
10
-8 -6 -4 -2 0 2 4 6
Launch Power (dBm)
(a)
-1
10

-2
10

-3
10
Bit Error Rate

-4
10

-5
10

-6 Launch Power = 4dBm

10
Span length=80km
-7
QAM
10 LLR-OSCD
NL-OSCD(4,4200)
-8
10
3000 3500 4000 4500 5000 5500
Total Length
(b)

The coding performance of using regular 8-QAM and 8-ary OSCD modulation formats is compared in the back-to-back
(BTB) scenario as a function of pre-FEC Q-factor derived from BER counting. The regular 8-QAM constellations can be
found in [133]. To achieve sufficient confidence level, over 100 million bits are considered for each BER point, and
exponential fitting is used for extrapolating the measured BER data and thus estimating the FEC limit. As observed in
Fig. 9.83, with the same LDPC codewords transmitted, the FEC limit achieved by using optimal 8-ary OSCD is 4.05 dB at
20% overhead, whereas 4.6 dB for regular 8-QAM. Conforming to the findings in [48], there is ~0.55 dB improvement when
utilizing optimal nonbinary LDPC-coded 8-QAM due to the joint coded modulation involving bit mapping, coding, and
modulation format.
624 9 Advanced Coding and Coded Modulation Techniques

DAC3 DAC4
C-band
110 I/Q
Lasers Mod 3 C+L SW
coupler WSS-C
Coherent
DAC1 DAC2 Pol.
PM-EDFA OTF
L-band MUX Rx
WSS-L
92 I/Q
Lasers Mod 4

DAC1 DAC2
SW L C

LSPC
Odd
4 I/Q Pol.
ECLs Mod 1 MUX C or L 4
DAC3 DAC4 coupler
Even 121.2 km
4 I/Q Pol.
ECLs Mod 2 MUX
GEQ-C C
GEQ-L
L

P o w e r (1 0 d B /d i v ) (a)

1530 1545 1560 1575 1590 1605

Wavelength (nm)

(b)
Fig. 9.82 (a) Experimental system setup, (b) 8-ary OSCD constellation in back-to-back (BTB) configuration and received spectrum after 6787 km
at 0.1 nm resolution

In the bidirectional transmission, C- and L-band signals are transmitted in opposite directions over the same fiber over
6780 km. Figure 9.84 shows the measured Q-value for the 200 wavelengths. The Q-values are shown for both cases where
either CDC is used or NLC is performed. Even though most channels fall below the FEC limit when only CDC is performed,
after an average of 0.4 dB improvement by NLC, for the C-band, and 0.8 dB improvement for the L-band, all channels are
recovered error-free after FEC decoding assisted with BCJR detector. The NLC improvement is different for the two bands
since C-band is operated slightly below the optimum channel power and L-band is operated at a slightly above the optimum
power to avoid excessive Raman interaction between bands and also to minimize gain tilt at the L-band.
Figure 9.85 shows the Q-value vs. received OSNR at 1551.2 nm as an example, where CDC only and NLC are compared
as the channel power is swept freely. [The Q-factor is obtained from measured BERs]. The lower two figures in Fig. 9.85
depict the constellation density after CDC and NLC, respectively. These results show that NLC not only increases system
margin but provides additional flexibility for system design and power budget allocation. In addition, the coded 8-OSCD
could benefit more from the NLC since the coded modulation has lower FEC limit and the Q-factor improvement from NLC is
sufficient to let the Q-factor move away (to the right) from the LDPC waterfall region, thus avoiding bit errors. Further
experiments employing OSCD-based constellations can be found in [134, 135].
9.15 Adaptive Coding and Adaptive Coded Modulation 625

Fig. 9.83 Experimental

comparison of the coding
performance between regular -2
8-QAM and 8-OSCD
10
FEC
FEC
FEClimit
limit limit
Optimal
8-OSCD Regular

Post-FEC BER
8QAM 8QAM
-4
10

-6
10
Exp Fit for 8-OSCD
Exp Fit for regular 8QAM
-8
10
3.6 3.8 4 4.2 4.4 4.6
pre-FEC Q [dB]

Fig. 9.84 Measured transmission

performance for all channels with
5
CDC and NLC after 6780 km. We CDC
use the term “FEC limit” to denote
that not any countable error is NLC
found in experiment after LDPC
decoder for this particular Q-value
4.5
Q-factor [dB]

3.5
“FEC limit” of optimal coded-8QAM

3
1540 1560 1580 1600
Wavelength [nm]

9.15 Adaptive Coding and Adaptive Coded Modulation

As discussed in Chap. 5, adaptive modulation and adaptive coding enable both robust and spectrally efficient transmission
over time-varying wireless and optical channels [67, 75, 96, 136–140]. The key idea is to estimate the channel state at the
receiver and feed this estimate back to the transmitter side, so that the transmitter can adapt its parameters relative to the
channel characteristics. In case when modulation format and channel code are fixed, the transmission system margin must be
taken into consideration, to enable the proper system operation when the channel conditions are poor. Additionally, the
component aging effects should be taken into account. Clearly, the system with fixed modulation format and fixed channel
626 9 Advanced Coding and Coded Modulation Techniques

Fig. 9.85 Measured transmission

performance for the channel at 5.5
1551.2 nm with CDC and NLC as
a function of received OSNR
NLC
5 CDC

4.5

Q-factor [dB]
4

3.5

2.5
16 16.5 17 17.5 18 18.5
OSNR/0.1nm [dB]

code is designed for the worst-case scenario, which results underutilization of the system most of the time. Adaptive
modulation techniques have already been discussed in Chap. 5. In adaptive coding, different FEC codes are used to provide
different levels of protection to the transmitted bits. For instance, a stronger FEC code with smaller coder rate (larger
overhead) can be used when conditions are poor, while a weaker FEC code can be used when the channel conditions are good.
Adaptive coding can be implemented by combining together FEC codes with different error correction capabilities, using
UEP and multilevel coding approach. Additionally, the shortening, puncturing, and re-encoding approaches can be used as
well. Rate-adaptive LDPC coding has already been discussed in Sect. 9.10.6, while corresponding FPGA implementations are
described in Sect. 9.10.7. The unequal error protection and multilevel coding have already been described in Sect. 9.11. In the
remainder of this chapter, we describe the adaptive coded modulation scheme in which trellis or coset codes are superimposed
on the top of adaptive modulation scheme (Sect. 9.15.1), the adaptive modulation and coding scheme in which both the signal
constellation size and error correction strength (code rate) are simultaneously adapted to come as close as possible to the
channel capacity (Sect. 9.15.2), and hybrid adaptive modulation and coding scheme in which FSO and RF channels are
operated in a cooperative fashion compensating for shortcomings of each other (Sect. 9.15.2).

9.15.1 Adaptive Coded Modulation

Compared to adaptive modulation alone, the additional coding gain can be achieved by superimposing either trellis codes or
coset codes on top of the adaptive modulation scheme. The key idea behind this adaptive coded modulation is to exploit the
9.15 Adaptive Coding and Adaptive Coded Modulation 627

Fig. 9.86 Generic adaptive One point

coded modulation scheme One point selected out of
Additional selected out of 2[n(SNR)+r] / (N/2)
n(SNR)-k info 2n(SNR)+r possible possible Adaptive two- To
bits Signal point Constellation dimensional channel
selection mapper modulator
from coset M(SNR), P(SNR)
One coset selected Modulation stage
out of 2k+r possible

k+r coded Binary

bits encoder k info bits
Coset
selector (block or
convolutional)
Coding stage

separability of coding stage and constellation design inherent to the coset codes. The instantaneous SNR or OSNR varies with
time, which will change the distance in the received signal constellation d0(t), and, therefore, the corresponding minimum
distance between coset sequences ds and the minimum distance between coset points dc will be time-varying as well. By using
adaptive modulation in a tandem with the coset codes, we can ensure that distances ds and dc do not vary as channel
conditions change by adapting the signal constellation size M(SNR), transmit power P(SNR), and/or symbol rate Rs(SNR) of
the transmitted signal constellation relative to the SNR, subject to an average transmit power P constraint on SNR-dependent
transmit power P(SNR). Minimum distance does not vary as channel conditions change dmin(t) ¼ min{ds(t), dc(t)} ¼ dmin; the
adaptive coded modulation will exhibit the same coding gain as coded modulation developed for a Gaussian-like channel with
corresponding minimum codeword distance dmin. Therefore, this adaptive coded modulation scheme requires small changes
in generic coded modulation scheme described in Sect. 9.11.1, as shown in Fig. 9.86. Clearly, the coding stage does not
change at all. The number of cosets does not change as well. On the other hand, the number of points within the coset is
channel SNR dependent. When channel condition gets improved, we enlarge the coset size. On such a way, we transmit more
uncoded bits, while not affecting the distance of the channel code. Additionally, this approach is hardware-friendly.
The following approximation for bit error probability for M-ary QAM systems can be used:
 
1:5SNR  Gc PðSNRÞ
Pb  0:2 exp , M  4, 0  SNR ½dB  30 dB, ð9:186Þ
M ðSNRÞ  1 P

where Gc is the coding gain. We adjust the signal constellation size M(SNR) depending on channel SNR and target BER as
follows:

1:5 PðSNRÞ Pðγ Þ

M ðSNRÞ ¼ 1 þ SNR  Gc ¼ 1 þ C f SNR  Gc , ð9:187Þ
 log ð5Pb Þ P P
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}
Cf

where Cf is the correction factor, defined as Cf ¼  1.5SNR/ log (5Pb), which is used to account for the fact that non-optimized
signal constellation is used. The number of uncoded bits needed to select a point within the coset is determined by

nðγ Þ  2k=N ¼ log 2 M ðγ Þ  2ðk þ r Þ=N: ð9:188Þ

The normalized transmission rate is given by log2M(SNR), while the normalized adaptive data rate is determined by log2M
(SNR)-2r/N. To determine the optimum launch power, we need to optimize hlog2M(SNR)i, subject to the average power
constraint, and not surprisingly the optimum power adaptation strategy is the water-filling law:
8
< 1=SNR0  1=SNRc , SNR  SNR 0 =C c
|fflfflfflfflffl{zfflfflfflfflffl}
PðSNRÞ
¼ SNRc , C c ¼ C f Gc ð9:189Þ
P :
0, SNR < SNR0 =C c
628 9 Advanced Coding and Coded Modulation Techniques

where SNR0 is the signal-to-noise ratio for AWGN channel. After substituting (9.189) into (9.187), we obtain the following
equation for adaptive signal constellation size:

M ðSNRÞ ¼ SNR=SNRc : ð9:190Þ

The corresponding spectral efficiency is given by

Z1  
R SNR
¼ log 2 f ðSNRÞdSNR, ð9:191Þ
B SNRc
SNRc

for which the distribution of SNR, denoted as f(SNR), is needed.

9.15.2 Adaptive Nonbinary LDPC-Coded Multidimensional Modulation for High-Speed Optical

Communications

In this scheme, which employs a single NB-LDPC encoder on Tx side and a single NB-LDPC decoder operating over GF of
order matching the signal constellation size on Rx side, as depicted in Fig. 9.87, the signal constellation size M ¼ 2m and code
rate R adaptation have been performed such that mR  C, where C is the multidimensional optical channel capacity.
The simulation results for SDM NB irregular QC-LDPC-coded 2D constellations are summarized in Fig. 9.88, for
information symbol rate Rs, info ¼ 25 GS/s and aggregate data rate of 2mNRRs, where m is the number of bits/symbol, N is
the number of spatial modes, R is the code rate, and Rs is the channel symbol rate (Rs,info/R). (To compensate for mode-
coupling effects, the MIMO signal processing is used as described in [1].) The corresponding signal constellation size and
code rate have been chosen based on the estimate of channel OSNR per bit. Both conventional QAM and constellations from
optimized signal constellation design (OSCD) have been observed (see chapter on advanced modulation formats for
additional details on OSCD constellations). In Fig. 9.88 fixed irregular (34635, 27708,0.8) code is observed. The results
for adaptive NB-QC-LDPC coded modulation, for code rates ranging from 0.667 to 0.8 are summarized in Fig. 9.89.

Fig. 9.87 A generic adaptive

hybrid D-dimensional irregular Adaptive hybrid D-dim. modulator
NB-QC-LDPC-coded modulation
scheme employing single LDPC m -ary
encoder/decoder operating over Source
GF of order matching the signal SDM Adaptive Adaptive symbols Adaptive NB
D-dim. LDPC symbol
constellation size
D-dim. …
fiber mapper encoder sequence
modulator
R=K / N
D-dim.
symbols

(a)
Adaptive hybrid D-dim. Dem.

D-
dimen. Adaptive SDM
Adaptive NB
APP D-dim.
LDPC dec. demod. fiber
demap.
Extrinsic
LLRs

(b)
9.15 Adaptive Coding and Adaptive Coded Modulation 629

Fig. 9.88 BER performance of

irregular NB-QC-LDPC (34635, NB-LDPC (34635,27708,0.8):
10
-2 Regular QC-LDPC coded:
27708,0.8)-coded 2D
16-OSCD
constellations
-3 16-QAM
10
8-QAM
Irregular QC-LDPC coded:

Bit-error rate, BER

-4
10 8-OSCD
16-OSCD
-5
10 BI reg. LDPC(34635,27710,0.8):
LDPC-coded 8-QAM
10
-6 LDPC-coded 16-QAM

-7
10

-8
10

-9
10
4.0 4.5 5.0 5.5 6.0 6.5
Optical signal-to-noise ratio, OSNR [dB] (per bit)

Fig. 9.89 BER performance of -1

10 Rate-0.8 NB-LDPC code:
irregular NB-QC-LDPC-coded
8-OSCD
2D constellations for code rates -2
10 16-OSCD
ranging from 0.667 to 0.8
32-OSCD
10
-3 16-QAM
Bit-error rate, BER

-4 Rate-0.667 NB-LDPC code:

10
16-OSCD
-5 16-QAM
10 32-OSCD
32-QAM
-6
10
-7
10
-8
10
-9
10
3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5
Optical signal-to-noise ratio, OSNR [dB] (per bit)

9.15.3 Adaptive Hybrid FSO-RF-Coded Modulation

In a hybrid FSO/RF system, FSO and RF channels exhibit drastically different dynamics and transmission speeds, so they
naturally require different coding and modulation techniques. Determining the optimum split of information bits between the
two is quite challenging. Rate-adaptive LDPC-coded modulation can be used for this purpose by adjusting the error correction
strength according to the time-varying conditions of the two channels [137]. The LDPC code design has been performed
based on the channel capacity of the hybrid link, with some of the symbols transmitted over FSO and the rest over the RF
channel. In one possible solution, one LDPC decoder has been used for both subsystems.
To illustrate high potential of hybrid techniques, combined together with adaptive hybrid LDPC-coded modulation, in
Fig. 9.90 we provide spectral efficiency results per single hybrid Tx-Rx pair. The PAM is used in FSO subsystem with direct
detection, while QAM is used in RF subsystem. The hybrid coded modulation scenario is compared against corresponding
FSO system with RF feedback. The coding gain over adaptive modulation at bit error probability Pb ¼ 106 for spectral
efficiency R/B ¼ 4 bits/s/Hz is 7.2 dB in both (weak and strong) turbulence regimes. Further improvements can be obtained by
increasing the girth of LDPC codes and employing better modulation formats. The increase in codeword length does not
improve spectral efficiency performance that much. It is interesting to notice that by employing adaptive hybrid coded
modulation, the communication under saturation regime is possible. Moreover, for variable-rate variable-power scheme, there
is no degradation in saturation regime compared to strong turbulence regime. Overall improvement from adaptive modulation
630 9 Advanced Coding and Coded Modulation Techniques

Hybrid FSO-RF system

FSO with RF feedback: Overall improvement: -6
LDPC(16935,13550)-coded, Pb=10
2
R
=0.04: 10.5 dB-38.9 dB Variable-rate variable-power:
Variable-rate variable-power (at 4 bits/s/Hz & 10-6) =2, =1 (Rayleigh):
2 2 4 2
Truncated channel inversion R
=0.04 R
=10 R
=4.0
2
=4.0: =2, =2 (Nakagami m=2):
R 2 2 4
=0.04 =10
Variable-rate variable-power R R

Channel inversion =3, =2:

2 4
Truncated channel inversion R
=0.04 R
=10
2 4 Truncated channel inversion:
R
=10 :
=2, =1 (Rayleigh):
Variable-rate variable-power 2
=0.04
2
=10
4
5
Spectral efficiency, R/B [bits/s/Hz]

R R
LDPC(16935,13550)
=3, =2:
LDPC(100515,80412) 2 4
=10
Channel inversion

Spectral efficiency, R/B [bits/s/Hz]

R
4
5 5.25 dB (Rayleigh fading)
better then FSO
3 Upper FSO bound
4
(without scintillation) 7.2 dB (Rayleigh fading)
better then uncoded
2 3

2
1
-6
Pb=10 1
0
0 5 10 15 20 25 30 5 10 15 20 25 30
Signal-to-noise ratio, Es/N0 [dB] Signal-to-noise ratio, Es/N0 [dB]

Fig. 9.90 Spectral efficiencies against symbol SNR for hybrid LDPC-coded modulation

and coding for spectral efficiency of 4 bits/s/Hz at Pb ¼ 106 over non-adaptive uncoded modulation ranges from 10.5 dB
(3.3 dB from adaptive modulation and 7.2 dB from coding) in the weak turbulence regime to 38.9 dB in the strong turbulence
regime (31.7 dB from adaptive modulation and 7.2 dB from coding). For additional details on this approach, an interested
reader is referred to [2, 137].

9.16 Concluding Remarks

This chapter has been devoted to advanced channel coding and coded modulation techniques. After linear block codes
fundamentals described in Sect. 9.1, BCH codes fundamentals have been described in Sect. 9.2, including Massey-Berlekamp
algorithm. The trellis description of linear block codes and the corresponding Viterbi decoding algorithm have been described
in Sect. 9.3. The fundamentals of convolutional codes have been provided in Sect. 9.4. The RS, concatenated, and product
codes have been described in Sect. 9.5. Section 9.6 has been devoted to the description of coding with interleaving as an
efficient way to deal with burst of errors and fading effects.
Significant space in the chapter has devoted to codes on graphs, starting with codes on graph fundamentals in Sect. 9.7. The
turbo codes and BCJR algorithm-based decoding procedure have been described in Sect. 9.8, while turbo-product codes have
been described in Sect. 9.9. LDPC codes and corresponding decoding algorithms have been described in Sect. 9.10.
Regarding LDPC codes, both binary and nonbinary (NB) LDPC codes have been introduced, and corresponding decoding
algorithms as well as their FPGA implementation have been described. Section on LDPC codes has been organized as
follows: binary LDPC codes have been described in Sect. 9.10.1, decoding of binary LDPC codes in Sect. 9.10.2, FPA
implementation of binary LDPC decoders in Sect. 9.10.3, decoding of NB-LDPC codes in Sect. 9.10.4, design of LDPC codes
in Sect. 9.10.5, rate-adaptive LDPC coding in Sect. 9.10.6, and rate-adaptive coding implementations in FPGA in Sect. 9.10.7.
The next section in the chapter, Sect. 9.11, has been devoted to coded modulation and unequal error protection. After
9.17 Problems 631

providing the coded modulation fundamentals (Sect. 9.11.1), TCM has been described in Sect. 9.11.2, multilevel coding and
UEP in Sect. 9.11.3, BICM in Sect. 9.11.4, and turbo TCM in Sect. 9.11.5. Various hybrid multidimensional coded
modulation schemes suitable for ultrahigh-speed optical transmission have been described in Sect. 9.12. Hybrid coded
modulation has been described in Sect. 9.12.1, while multilevel NB-LDPC-coded modulation in Sect. 9.12.2.
After coded modulation sections, the focus of the chapter is on multidimensional turbo equalization, Sect. 9.13. The
following topics have been described: nonlinear channels with memory in Sect. 9.13.1, nonbinary MAP detection in Sect.
9.13.2, sliding-window multidimensional turbo equalization in Sect. 9.13.3, simulation study of multidimensional turbo
equalization in Sect. 9.13.4, and several experimental demonstrations including time-domain 4D-NB-LDPC-CM and quasi-
single-mode transmission over transoceanic distances in Sects. 9.13.5 and 9.13.6, respectively. In section on optimized signal
constellation design and optimized bit-to-symbol mapping-based coded modulation, Sect. 9.14, multidimensional OSCD
(Sect. 9.14.1), EXIT chart analysis of OSCD mapping rules (Sec. 9.14.2), nonlinear OSCD-based coded modulation (Sect.
9.14.3), and transoceanic multi-Tb/s transmission experiments enabled by OSCD (Sect. 9.14.4) have been described.
Finally, in adaptive coding and adaptive coded modulation section (Sect. 9.15) the following coded modulation schemes
are described: adaptive coded modulation (Sect. 9.15.1), adaptive nonbinary LDPC-coded multidimensional modulation
suitable for high-speed optical communications (Sect. 9.15.2), and adaptive hybrid FSO-RF coded modulation (Sect. 9.15.3).
The set of problems is provided in incoming section to help reader gain deeper understanding of material described in this
chapter.

9.17 Problems

9.1. The binary symmetric channel (BSC) is described by the following channel matrix:
 
q p
PBSC ¼ , p þ q ¼ 1:
p q

(a) Determine the channel capacity of this channel, and plot its dependence against the crossover probability p.
Observe now the channel that is obtained by cascading several BSC channels.
(b) Determine the channel matrix after cascading two and three BSC channels.
(c) Plot the channel capacity of cascaded channels against p. Discuss the results.
9.2. For the channel shown in Fig. 9.P2, with the prior probabilities P(x1) ¼ p, P(x2) ¼ q, and the transition probabilities
being different p1 6¼ p2, determine the channel capacity. This channel is known as the binary asymmetric channel
(BAC), and the optical channel for IM/DD belongs to this category.

1-p1
x1 y1

p1

p2
x2 y2
1-p2

Fig. 9.P2 The binary asymmetric channel

9.3. An (n,k) linear block code is described by the following parity-check matrix:
2 3
0 0 0 0 0 0 0 1 1 1 1 1 1 1 1
60 17
6 0 0 1 1 1 1 0 0 0 0 1 1 1 7
H¼6 7:
40 1 1 0 0 1 1 0 0 1 1 0 0 1 15
1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
632 9 Advanced Coding and Coded Modulation Techniques

(a) Determine the code parameters: codeword length, number of information bits, code rate, overhead, minimum
distance, and error correction capability.
(b) Represent the H-matrix in systematic form, and determine the generator matrix G of the corresponding systematic
code.
9.4. Consider a (7,4) code with generator matrix:
2 3
0 1 0 1 1 0 0
61 0 07
6 1 0 1 0 7
G¼6 7
40 1 1 0 0 1 05
1 1 0 0 0 0 1

(a)
Find all the codewords of the code.
(b)
What is the minimum distance of the code?
(c)
Determine the parity-check matrix of the code.
(d)
Determine the syndrome for the received vector [1101011].
Assuming that an information bit sequence of all 0 s is transmitted, find all minimum weight error patterns e that
(e)
result in a valid codeword that is not the all-zero codeword.
(f) Use row and column operations to transform G to systematic form, and find its corresponding parity-check matrix.
Sketch a shift register implementation of this systematic code.
(g) All Hamming codes have a minimum distance of 3. What is the error correction and error detection capability of a
Hamming code?
9.5. Consider the following set:

C ¼ fð000000Þ, ð001011Þ, ð010101Þ, ð011110Þ, ð100111Þ, ð101100Þ, ð110010Þ, ð111001Þg

(a) Show that C is vector space over GF(2).

(b) Determine the dimension of this set.
(c) Determine the set of basis vectors.
If the set C is considered as the code book of an LBC:
(d) Determine the parameters of this code.
(e) Determine the generator and parity-check matrices in systematic form.
(f) Determine the minimum distance and error correction capability of this code.
9.6. Let C be an (n,k) LBC of minimum Hamming distance d. Define a new code Ce by adding an additional overall parity-
check equation. Such obtained code is known as extended code:
(a) If the minimum distance of original code is d, determine the minimum distance of extended code.
(b) Show that extended code is an (n + 1,k) code.
9.7. Determine an (8,4) code obtained by extending the (7,4) Hamming code. Determine its minimum distance and error
correction capability. Show that (8,4) code and its dual are identical. Such codes are called self-dual and are of high
importance in quantum error correction.
m m
9.8. The maximum length Cm dual is the dual of (2  1,2  1  m,3) of corresponding Hamming code. Let m ¼ 3:
(a) List all codewords of this code.
(b) Show that C3dual is an (7,3,4) LBC.
9.9. Let G ¼ [Ik | P] be the generator matrix of a systematic (n,k) LBC. Show that this code is self-dual if and only if P is a
square and PPT ¼ I. Show that for dual codes, n ¼ 2 k. Finally, design self-dual codes for n ¼ 4, 6, and 8.
9.10. Prove the following properties of standard array:
(a) All n-tuples of row are distinct.
(b) Each n-tuple appears exactly once in the standard array.
(c) There are exactly 2n  k rows in the standard array.
(d) For perfect codes (satisfying the Hamming bound with equality sign), all n-tuples of weight t ¼ int[(dmin  1)/2] or
less appear as coset leaders (int[x] is the integer part of x).
(e) For quasi-perfect codes, in addition to all n-tuples of weight t or less, some but not all n-tuples of weight t + 1 appear
as coset leaders.
9.17 Problems 633

(f) All elements in the same row (coset) have the same syndrome.
(g) Elements in different rows have different syndromes.
(h) There are 2n  k different syndromes corresponding to 2n  k rows.
9.11. For extended code from Problem 9.7, determine the standard array and syndrome decoding table. If the received word
was (10101010), what would be the result of standard array decoding? Determine the weight distribution of coset
leaders and word error probability for crossover probability of BSC being 104. Finally, determine the coding gain at
target BER of 106.
9.12. The (n,k) LBC can be shortened by deleting the information symbols, which is equivalent to the deleting of parity
columns in the parity-check matrix H ¼ [PT|In  k]. Let us remove all columns of even weight in PT of the (2m  1,
2m  1  m) Hamming code. Derive the new code parameters in terms of the original code, and determine the minimum
distance of the shortened code. What is the error correction capability of this shortened code?
9.13. Prove that an LBC can correct all error patterns having ec or fewer errors, and simultaneously detect all error patterns
containing ed (ed  ec) or fewer errors if the minimum distance dmin satisfies the following inequality: dmin  ed + ec + 1.
9.14. Let the polynomial x7 + 1 over GF(2) be given. Determine the generator and parity-check polynomials of cyclic (7,4)
code. By using these polynomials, create the corresponding generator and parity-check matrices. Encode the informa-
tion sequence (1101) using this cyclic code. Provide the encoder and syndrome circuits.
9.15. Consider RS (31,15) code:
(a) Determine the number of bits per symbol in the code and determine the block length in bits.
(b) Determine the minimum distance of the code and the error correction capability.
9.16. Let us observe RS (n,k) code:
(a) Prove that minimum distance is given by dmin ¼ n  k + 1.
(b) Prove that the weight distribution of RS codes can be determined by

  X
id  
n min
i  1 idmin j
Ai ¼ ð q  1Þ ð1Þ j q :
i j¼0 j

9.17. Design a concatenated code using (15,9) 3-symbol-error-correcting RS code as the outer code and (7,4) binary
Hamming code as the inner code. Determine the overall codeword length and the number of binary information bits
contained in the codeword. Determine the error correction capability of this code.
9.18. This problem is related to coded modulation:
(a) Describe how to determine the symbol log-likelihood ratios (LLRs): (i) assuming Gaussian approximation and
(ii) assuming that conditional PDF is obtained by collecting the histograms.
(b) Describe the transmitter and receiver configuration for polarization-multiplexed bit-interleaved LDPC-coded
modulation.
(c) Describe the transmitter and receiver configuration for multilevel coding (MLC) scheme with LDPC codes as
channel codes.
(d) Providing that symbol LLRs are determined as in (a), describe how to determine the bit LLRs required for LDPC
decoding. Describe how to determine the extrinsic information at the output of LDPC decoder for APP demapper.
Describe how to determine the prior symbol LLRs for next APP-LDPC decoder iteration step.
9.19. This problem is related to the sum-product algorithm. Let qij(b) denote the extrinsic information (message) to be passed
from variable node vi to function node fj regarding the probability that ci ¼ b, b2{0,1}, as illustrated in Fig. 9.P19 on the
left. This message is concerned about probability that ci ¼ b, given extrinsic information from all check nodes, except

fj
fj
qij(b) rji(b)

qij(b)
vi rji(b)

vi
yi

Fig. 9.P19 Illustration of the sum-product algorithm

634 9 Advanced Coding and Coded Modulation Techniques

node fj, and given the channel sample yj. Let rji(b) denote extrinsic information to be passed from node fj to node vi, as
illustrated in figure below on the right. This message is concerned about probability that j-th parity-check equation is
satisfied given that ci ¼ b, while other bits have separable distribution given by {qij’}j’ 6¼ j:
– Show that rji(0) and rji(1) can be calculated from qi’j(0) and qi’j(1) as follows:

1 1   1 1  
r ji ð0Þ ¼ þ ∏ 1  2qi0 j ð1Þ , r ji ð1Þ ¼  ∏ 1  2qi0 j ð1Þ
2 2 i0 2V j ∖i 2 2 i0 2V j ∖i
– Show that qji(0) and qji(1) can be calculated as follows:

qij ð0Þ ¼ ð1  Pi Þ ∏ r j0 i ð0Þ, qij ð1Þ ¼ Pi ∏ r j0 i ð1Þ

j0 2C i ∖j j0 2C i ∖j

9.20. The probability domain sum-product algorithm has the following disadvantages: (i) multiplications are involved
(additions are less costly to implement) and (ii) the calculation of many product probabilities could become numerically
unstable (especially for long codes with more than 50 iterations). The log-domain version of sum-product algorithm is a
preferable option. Instead of beliefs in log domain, we use the log-likelihood ratios (LLRs), which for codeword bits ci
(i ¼ 2,. . .,n) are defined by
 
Prðci ¼ 0jyi Þ
Lðci Þ ¼ log ,
Prðci ¼ 1jyi Þ

where yi is a noisy sample corresponding to ci. The LLRs corresponding to rji and qij are defined as follows:
   
  r ji ð0Þ   qij ð0Þ
L r ji ¼ log , L qij ¼ log :
r ji ð1Þ qij ð1Þ

By using tanh[log( p0/p1)/2] ¼ p0  p1 ¼ 1  2p1, show that

1 X
Lðr ji Þ ¼ 2tanh1 f ∏ tanh ½ Lðqi0 j Þg and Lðqij Þ ¼ Lðci Þ þ Lðr j0 i Þ:
i0 2V j ∖i 2 0
j 2Ci ∖j

9.21. This problem is related to the LDPC decoder implementation using the min-sum-with-correction term algorithm:
(a) Implement the LDPC decoder using the min-sum-with-correction-term algorithm. By assuming that zero
codeword was transmitted over an additive white Gaussian noise (AWGN) channel for the LDPC code described
in example of Sect. 9.10.5.4, show BER dependence against Q-factor for on-off keying. Determine the coding
gains at BER of 106 and 109. Discuss the results.
(b) Implement now corresponding encoder, use sufficiently long pseudorandom binary sequence, encode it, transmit
over the AWGN channel, and repeat the simulation from (a). Discuss the differences.
(c) You are asked now to transmit an LDPC encoded sequence over the realistic optical channel. Generate the PRBS
sequence of sufficient length to accurately estimate BER down to 106. Encode this sequence using the LDPC
encoder you developed in problem (b), and transmit over the optical fiber transmission systems as described
below. Represent the symbol 1 pulse by a Gaussian pulse with FWHM of 12.5 ps. Assume the launch power of
0 dBm, the extinction ratio of 20 dB, and data rate of 40 Gb/s. Transmission system is 6000 km in length, with
dispersion map composed of N sections of SMF of length 80 km (with dispersion and dispersion slope parameters
being 16 ps/(nmkm) and 0.06 ps/(nm2km), respectively) and 40 km of DCF. The DCF is chosen to exactly
compensate for residual chromatic dispersion, and dispersion map slope is matched. The nonlinear coefficient of
both fibers is γ ¼ 2.5 (Wkm)1. The EDFAs with NF of 5 dB and BW ¼ 3Rb (Rb is the bit rate) are periodically
deployed every fiber section to exactly compensate for fiber loss. The SMF loss is 0.2 dB/km and DCF loss is
0.5 dB/km. The operating wavelength is 1550 nm and photodiode responsivity is 1 A/W. The receiver electronics
can be model as Gaussian filter of 3 dB bandwidth equal BW ¼ 0.75 Rb. Vary the launch power from 6 dBm to
6 dBm in steps of 2 dBm, and plot how the BER changes as a function of the launch power before and after LDPC
decoding. Determine how much transmission distance can be extended by using the LDPC code. Provide
BER vs. total transmission plots for both uncoded and LDPC-coded cases, for different launch powers.
9.17 Problems 635

9.22. This problem is related to the QAM with coherent detection in a back-to-back configuration. The transmitter and
receiver are shown in Fig. 9.P22a, b, respectively. Determine by simulation both bit error rate Pb and symbol error rate
Ps against optical SNR per bit for the following signal constellation sizes: M ¼ 4, 16, and 64.
9.23. This problem is related again to the QAM with coherent detection, but now in a long-haul optical transmission system
with dispersion map as shown in Fig. 9.P22c. The transmitter and receiver configurations are identical to those shown in
Fig. 9.P22a, b.
Assume the average symbol launch power of 0 dBm, the symbol rate is 25 Giga symbol/s (25 GS/s), and that either
QPSK or 16-QAM is used. Transmission system is composed of N sections of SMF of length 80 km (with dispersion
and dispersion slope parameters being 16 ps/(nmkm) and 0.06 ps/(nm2km), respectively) and 40 km of DCF. The
DCF is chosen to exactly compensate for residual chromatic dispersion, and dispersion map is slope matched. The
nonlinear coefficient of both fibers is γ ¼ 2.5 (Wkm)1. The EDFAs of NF of 5 dB and BW ¼ 3Rb (Rb is the bit rate)
are periodically deployed every fiber section to exactly compensate for fiber loss. The SMF loss is 0.2 dB/km and
DCF loss is 0.5 dB/km. The operating wavelength is 1550 nm and photodiode responsivity is 1 A/W. The receiver
electronics can be model as Gaussian filter of 3 dB bandwidth equal BW ¼ 0.75 Rb. Vary the launch power from
6 dBm to 6 dBm in steps of 2 dBm, and plot how the BER changes as a function of total transmission length by
using the launch power as a parameter. Show both bit error and symbol error rates with both natural and Gray
mappings.

Ii
1 3 dB To
Source . m MZM SMF
Interleaver QAM DFB
Channels .
mxN mapper laser
.
m MZM /2
Qi
(a)
Coherent balanced detector

from /2 vI
SMF
Symbol
vQ detector
from
Local laser

(b)
N spans

SMF DCF

Transmitter Receiver

EDFA

(c)
Fig. 9.P22 QAM with coherent detection: (a) transmitter configuration, (b) receiver configuration, and (c) dispersion map. 3dB 3dB
coupler, MZM Mach-Zehnder modulator

9.24. This problem is related to the coded modulation with coherent detection in a back-to-back configuration. Transmitter
configuration is shown in Fig. 9.P22a, and receiver configuration is shown in Fig. 9.P24. Let us observe again QAM,
but this time in combination with LDPC codes. The outputs of LDPC encoders are written in corresponding block
interleaver by block row-wise order, as shown in Fig. 9.P22a. The m ¼ log2M bits are taken from block interleaver
636 9 Advanced Coding and Coded Modulation Techniques

column-wise to select the corresponding constellation point using the Gray mapping rule. The upper branch at the input
of modulator represents the in-phase channel (I) and lower the quadrature channel (Q). The a posteriori probability
(APP) block determines the symbol log-likelihood ratios (LLRs). Bit LLRs block calculates the bit LLRs needed for
LDPC decoding:
(a) Derive the symbol LLRs. How to calculate bit LLRs based on symbol LLRs?
(b) Perform Monte Carlo simulations for M ¼ 2, 4, and 16 for both uncoded case and LDPC-coded case. Plot BERs
versus optical SNR per information bit.
(c) Determine the coding gains at BER of 106.
(d) Iterate extrinsic information between LDPC decoders and APP demapper 3 times, and provide BER plots vs. optical
SNR per bit, down to 106. Determine the OSNR improvement with respect to (c).

Coherent balanced detector Extrinsic LLRs

from /2 vI
SMF LDPC Decoder 1 1

Calculation
Bit LLRs
APP …
demapper
vQ LDPC Decoder m m
from
Local laser

Fig. 9.P24 Receiver configuration for bit-interleaved LDPC-coded QAM

9.25. This problem is related to the LDPC-coded QAM with coherent detection, but now in a long-haul optical transmission
system with dispersion map as shown in Fig. 9.P22c. The transmitter and receiver configurations are identical to those
shown in Fig. 9.P24. The fiber parameters are identical to those given in Problem 9.23. Vary the launch power from
6 dBm to 6 dBm in steps of 2 dBm, and plot how the BER changes as a function of total transmission length by using
the launch power as a parameter. Show both bit error and symbol error rates with both natural and Gray mappings.
Identify the extension in transmission distance from LDPC coding. Provide results for one and three outer (APP
demapper-LDPC decoder) iterations. Use 25 inner (LDPC decoder) iterations. Discuss the improvement when three
outer iterations are used with respect to the only one outer iteration.
9.26. This problem is related to the chromatic dispersion compensation for QAM with coherent detection. The transmitter and
receiver are shown in Fig. 9.P26a, b, respectively. The transmission system is composed of SMF of length L followed
by corresponding EDFA to compensate for fiber losses. Determine by simulation bit error rate BER against optical SNR
per bit for the following signal constellation sizes M ¼ 4 and 16 and for the following distances L ¼ 40, 60, 80, and
100 km. Calculate BER for both uncoded and LDPC-coded cases. Use the same LDPC code, which was provided to you
earlier. Assume that symbol rate is 40 GS/s. Discuss the channel memory assumption on BER results.
9.27. This problem is related to PMD compensation in QAM systems with coherent detection. The transmitter and receiver
are shown in Fig. 9.P27a, b, respectively. For the first-order PMD study, the Jones matrix, neglecting the polarization-
dependent loss and depolarization effects, can be represented by

  " #
hxx ðωÞ hxy ðωÞ ejωτ=2 0
H¼ ¼ RPðωÞR1 , PðωÞ ¼ ,
hyx ðωÞ hyy ðωÞ 0 ejωτ=2

where τ denotes DGD, ω is the angular frequency, and R ¼ R(θ,ε) is the rotational matrix:
2     3
θ jε=2 θ jε=2
6 cos e sin e 7
6 2 2 7
R¼6     7,
4 θ jε=2 θ jε=2 5
 sin e cos e
2 2
9.17 Problems 637

Ii
1 LDPC encoder To
. R=K / N m
3 dB MZM SMF
Source QAM DFB
Channels . … Interleaver
. mxN mapper laser
m LDPC encoder MZM /2
R=K / N
Qi

(a)
Coherent balanced detector Extrinsic LLRs

/2 vI
from
SMF LDPC Decoder 1 1

Calculation
Bit LLRs
BCJR …
equalizer
vQ LDPC Decoder m m
from
Local laser

(b)
Fig. 9.P26 Bit-interleaved LDPC-coded QAM with turbo equalization: (a) Tx and (b) Rx configurations

LDPC encoder
1x
. Rx=Kx / N mx
Source Interleaver QAM
Channels . … I/Q MODx
mxxN mapper x
(x-pol.) .
mx LDPC encoder
Rx=Kx / N
DFB laser PBS PBC
LDPC encoder
1y
. Ry =Ky / N . my
Source QAM
. … . Interleaver I/Q MODy
Channels my xN mapper y
. .
(y-pol.) my LDPC encoder
Ry=Ky / N

(a)

Ix
1x
calculaon

Coherent LDPC decoder 1x

(x-pol.)

Bit LLRs

detector (x-pol.) .. .
Qx
LDPC decoder mx mx
From SMF Local MAP equalizaon
PBS PBS
DFB laser (BCJR equalizer) 1y
calculaon

LDPC decoder 1y
(y-pol.)

Bit LLRs

.. .
Coherent
detector (y-pol.) LDPC decoder my my

Extrinsic LLRs

(b)
Fig. 9.P27 Polarization-multiplexed bit-interleaved LDPC-coded QAM with turbo equalization: (a) Tx and (b) Rx configurations
638 9 Advanced Coding and Coded Modulation Techniques

with θ being the polar angle and ε being the azimuth angle. For the worst-case scenario, determine by simulation bit
error rate BER against optical SNR per bit for the signal constellation sizes of 4 and 16 and for DGD values of 25 ps,
50 ps, and 75 ps. Calculate BER for both uncoded and LDPC-coded cases. Use the same LDPC code as in previous
examples. Assume that symbol rate is 40 GS/s. Discuss the channel memory assumption on BER.
9.28. This problem is related to the intrachannel nonlinearities compensation by turbo equalization. The corresponding long-
haul optical transmission system under study has dispersion map as shown in Fig. 9.P28. The transmitter and receiver
configurations are identical to those shown in Fig. 9.P22a, b. Assume the average symbol launch power of 0 dBm, the
symbol rate is 40 GS/s, and that either QPSK or 16-QAM is used. Transmission system is composed of N sections of
SMF of length 80 km (with dispersion and dispersion slope parameters being 16 ps/(nmkm) and 0.06 ps/(nm2 km),
respectively) and 40 km of DCF. The DCF is chosen to exactly compensate for residual chromatic dispersion, and
dispersion map is slope matched. The nonlinear coefficient of both fibers is γ ¼ 2.5 (Wkm)1. The EDFAs with NF of
5 dB and BW ¼ 3Rs (Rs is the symbol rate) are periodically deployed every fiber section to exactly compensate for the
fiber loss. The SMF loss is 0.2 dB/km and DCF loss is 0.5 dB/km. The operating wavelength is 1550 nm and
photodiode responsivity is 1 A/W. The receiver electronics can be model as Gaussian filter of 3 dB BW ¼ 0.75Rs.
Vary the launch power from 6 dBm to 6 dBm in steps of 2 dBm, and plot how the BER changes as a function of total
transmission length by using the launch power as a parameter. Show bit error rates with both natural and Gray
mappings. Provide results for both uncoded and LDPC-coded cases. Discuss the channel memory assumption on
BER results. Compare the obtained results with those from previous problems where APP demapper was used instead
of BCJR equalizer. Discuss results.

N spans

SMF DCF

Transmitter Receiver

EDFA

Fig. 9.P28 Dispersion map under study

9.29. In this problem you are asked to transmit an LDPC encoded sequence over the realistic free-space optical (FSO)
channel, by using the model developed in Problem 2.7 of Chap. 2:
(a) Generate the PRBS sequence of sufficient length to accurately estimate BER down to 106. Encode this sequence
using the LDPC encoder you developed in 9.21, and transmit over the FSO channel as described below. Represent
the symbol 1 pulse by a Gaussian pulse of FWHM of 20 ps. Assume the launch power of 0 dBm, the extinction ratio
of 25 dB, and data rate of 25 Gb/s. FSO transmission link is 1 km in length. The EDFA of NF of 5 dB and bandwidth
BW ¼ 3Rb (Rb is the bit rate) is used to exactly compensate for FSO channel loss. The receiver electronics can be
model as Gaussian filter of 3 dB BW ¼ 0.75 Rb. Vary the launch power from 6 dBm to 6 dBm in steps of 1 dBm,
and plot how the BER changes as a function of launch power before and after LDPC decoding, assuming that the
refractive structure parameter C 2n ¼ 1015 . Study the tolerance to atmospheric turbulence by using the refractive
structure parameter as a variable. To summarize, provide BER vs. the launch power for both uncoded and LDPC-
coded cases, for different refractive structure parameters. The Gaussian beam parameters should be the same in
Problem 1 of Chap. 2.
(b) To improve the tolerance to atmospheric turbulence effects, repeat the (a) but now performing the shortening
approach to reduce the code rate of provided LDPC code.
(c) Now assuming that the 2  2 repetition FSO MIMO is used, repeat (a), and discuss the improvements in tolerance to
atmospheric turbulence effects compared to adaptive coding.
9.30. This problem is related to the LDPC-coded modulation with coherent detection for communication over FSO channel.
Transmitter (Tx) and receiver (Rx) configurations are shown in Fig. 9.P26. Let us observe the QAM, combined with
LDPC coding. The outputs of LDPC codewords are written in corresponding block-interleaver block in row-wise
fashion, as shown in Fig. 9.P26. The m ¼ log2M bits are taken from block interleaver in column-wise fashion to select
the corresponding constellation point using the Gray mapping rule. The upper branch at the input of modulator
represents the in-phase channel (I) and lower the quadrature (Q) channel. The a posteriori probability (APP) block
9.17 Problems 639

determines the symbol log-likelihood ratios (LLRs). Bit LLRs block calculates the bit LLRs needed for LDPC
decoding:
(a) Assuming that the refractive structure parameter is C 2n ¼ 1015 , perform the Monte Carlo simulations for M ¼ 2,
4, and 16 for both uncoded case and LDPC-coded case. Plot BERs versus optical SNR per information bit.
Determine the coding gains at BER of 106. The simulation conditions are identical to those described in Problem
9.29(a).
(b) Iterate extrinsic information between LDPC decoders and APP demapper 3 times, and provide BER plots vs. optical
SNR per bit, down to 106. Determine the OSNR improvement with respect to (a).
(c) Perform adaptive coding with shortening approach, and show improvements by adaptive coding for M ¼ 2, 4, and
16 for different turbulence strengths.
9.31. This problem is continuation of Problem 9.30, but now in MIMO environment. Let us consider 2  2 FSO MIMO
system, in which two independent data streams are transmitted by two sufficiently separated FSO transmitters. Assume
that each FSO transmitter illuminates both FSO receivers:
(a) For 2  2 FSO MIMO system with either zero-forcing (ZF) or minimum MSE (MMSE) receiver, for the refractive
structure parameter of C2n ¼ 1015 , perform Monte Carlo simulations for M ¼ 2, 4, and 16 for both uncoded case
and LDPC-coded case. Plot BERs versus optical SNR per information bit. Determine the coding gains at BER of
106. Assume that channel matrix H is available on receiver side. The simulation conditions are identical to those
described in Problem 9.29(a).
(b) Repeat (a) but now for different values of the refractive structure parameter. At BER of 106, for LDPC-coded
QAM, determine improvements when MMSE receiver is used compared to the ZF receiver case.
(c) Discuss the improvements of LDPC-coded QAM when MMSE receiver is used with respect to the adaptive coding
case discussed in Problem 9.31(c) for different turbulence strengths.
9.32. This problem is related to coded modulation applied to 9-ary 2-D constellation described in Problem 5.8 of Chap. 5, in
the presence of scintillation and AWGN:
(a) Describe the transmitter and receiver configuration for polarization-multiplexed 9-ary LDPC-coded modulation.
(b) Determine the symbol LLRs required for 3-ary LDPC decoding. Describe how to determine the extrinsic informa-
tion at the output of 3-ary LDPC decoder for APP demapper. Describe how to determine the prior symbol LLRs for
next APP-LDPC decoder iteration step.
9.33. This problem is related to the adaptive modulation and coding applied to 9-ary 2-D constellation described in Problem
5.8 of Chap. 5, in the presence of scintillation and AWGN. First of all derive an error probability bound for 9-ary 2-D
constellation in the presence AWGN to be used throughout the problem:
(a) Determine the optimum power adaptation strategy as well as the corresponding spectral efficiency of this scheme, in
the presence of scintillation and AWGN.
(b) Describe the channel inversion technique for this signal constellation, and determine the corresponding spectral
efficiency, in the presence of scintillation and AWGN.
9.34. This problem is related to the adaptive modulation and coding applied to 9-ary 2-D constellation described below,
transmitted over wireless channel in the presence of Nakagami fading and AWGN. The eight constellation points in this
2-D constellation are placed on a circle of radius 1. The last (ninth) point is placed in the origin. Assume that the point in
the origin appears with probability 0.25, all others with probability 0.75/8. For channel coding use the LDPC code from
Problem 9.21:
(a) Determine the average symbol error probability for this modulation format in the presence of fading and AWGN.
(b) Estimate the coding gain for this LDPC code and use it in incoming (c)–(e) problems.
(c) Determine the optimum power adaptation strategy as well as the corresponding spectral efficiency of this scheme, in
the presence of fading and AWGN.
(d) Describe the channel inversion technique for this signal constellation, and determine the corresponding spectral
efficiency, in the presence of fading and AWGN.
(e) Describe the truncated channel inversion technique for this signal constellation, and determine the corresponding
spectral efficiency, in the presence of fading and AWGN.
9.35. Implement the parallel (1152, 1024) turbo decoder. The turbo code should be of rate R ¼ 8/9. Use two identical RSC
codes for encoding, whose encoder in octal form is given by (g1,g2) ¼ (7,5). To achieve the desired code rate, perform
the puncturing (e.g., you can save only 1 parity bit from block of 16 parity bits). Perform Monte Carlo simulations on an
additive white Gaussian channel (AWGN) model by employing BPSK. Plot BER against signal-to-noise ratio (SNR)
(in dB scale) for both uncoded case and turbo-coded case. Determine the coding gain at BER of 105.
640 9 Advanced Coding and Coded Modulation Techniques

9.36. This step is related to the coded modulation. The transmitter and receiver are shown in Fig. 9.P36a, b, respectively. The
noise from different amplifiers at receiver side can be modeled as AWGN.
Let us assume that M-ary QAM is used. The output of parallel turbo encoder is written into a buffer. The m ¼ log2M bits
are taken from the buffer and used to select the corresponding constellation point using the Gray mapping rule. After
D/A conversion (DAC), M-QAM modulator is used. The upper branch at the input of modulator represents the
in-phase channel (I) and lower the quadrature channel (Q). After up-conversion and amplification signal is emitted
by transmit antenna. Assume that there exists a line of sight (LOS) between transmitter and receiver, and ignore the
multipath fading and shadowing effect. On the receiver side, the signal after antenna is amplified, the I- and
Q-channel signals obtained upon demodulation are converted into digital form (assume double precision). The a
posteriori probability (APP) block determines the symbol log-likelihood ratios (LLRs). Bit LLRs block calculates
the bit LLRs needed for turbo decoding:
(a) Derive the symbol LLRs. How to calculate bit LLRs based on symbol LLRs?
(b) Perform Monte Carlo simulations for M ¼ 2, 4, and 16 for both uncoded case and turbo-coded case. Plot BERs
versus SNR per information bit.
(c) Determine the coding gains at BER of 105.

Source
m DAC
channel Gray I/Q Transmit
Turbo encoder Buffer Mapper modulator
DAC filter

Local
oscillator

(a)
cos( c t)

Pre-select
APP Demapper

filter LNA LPF ADC

Calculation
Bit LLRs

End
Turbo Decoder
user
LPF ADC

sin( c t)

(b)
Fig. 9.P36 Bit-interleaved LDPC-coded modulation-based wireless system: (a) transmitter configuration and (b) receiver configuration. DAC
digital-to-analog converter, LNA low-noise amplifier, LPF low-pass filter, ADC analog-to-digital converter

9.37. Repeat the Problem 9.36 by modeling the wireless channel using the Rayleigh fading simulator you have developed in
Problem 2.13 of Chap. 2. In addition to questions (a)–(c), determine the penalty coming from wireless channel at BER
of 105.
9.38. This problem is related to the adaptive modulation. We are concerned with adaptive MQAM transmission over
Rayleigh fading channel, as shown in Fig. 9.P38. The channel estimate from receiver side is transmitted back to
transmitter that adapts its power and signal constellation size according to the channel conditions. When channel
conditions are favorable, more data is transmitted. When channel conditions are poor, less data is transmitted; and when
SNR falls below threshold, data are not transmitted at all. Derive the optimum power adaptation policy. Determine the
spectral efficiency for optimum power adaptation policy. Write a MATLAB program that determines the spectral
efficiency vs. SNR for target bit error probability of 105. For spectral efficiency of 2 bits/s/Hz, determine how far away
this adaptation policy is from the channel capacity. When the truncated channel inversion is used, determine SNR
degradation for spectral efficiency of 2 bits/s/Hz with respect to the optimum adaptation strategy. The noise from
different amplifiers at receiver side can be modeled as AWGN.
9.39. Repeat the Problem 9.38 but now observing the Nakagami m ¼ 2 fading instead. Compare the conclusions between
Problems 9.38 and 9.39.
9.17 Problems 641

Local
oscillator

Source m DAC
Buffer Gray I/Q Transmit
data Mapper modulator filter
DAC

Constellation
Power
size control control

Feedback
channel

Channel
estimation cos( ct)

Pre-select
filter LNA LPF ADC

Demapper
Output
data
LPF ADC

sin( ct)

Fig. 9.P38 The variable-power variable-rate adaptation system. DAC digital-to-analog converter, LNA low-noise amplifier, LPF low-pass filter,
ADC analog-to-digital converter

9.40. We now observe the adaptive turbo-coded modulation. Corresponding system configuration is shown in Fig. 9.P40. For
turbo code you developed in Problem 9.35 and for spectral efficiency of 2 bits/s/Hz, determine how far away this
adaptive coding scheme is from the channel capacity. When the truncated channel inversion is used, determine SNR
degradation for spectral efficiency of 2 bits/s/Hz with respect to the optimum adaptation strategy. Write a MATLAB or
C/C++ program that plots the spectral efficiency against SNR when turbo coding is used. Determine the improvement
due to adaptive coding.

Local
oscillator

Source
Turbo m DAC
data Buffer Gray I/Q Transmit
encoder Mapper modulator filter
DAC

Constellation
Power
size control
control

Feedback
channel

Channel
estimation cos( c t)

Pre-select
Bit LLR calculation

APP Demapper

filter LNA LPF ADC

Turbo Output
+

decoder data
LPF ADC

sin( c t)

Fig. 9.P40 The turbo-coded variable-power variable-rate adaptation system. DAC digital-to-analog converter, LNA low-noise amplifier, LPF
low-pass filter, ADC analog-to-digital converter, APP, a posteriori probability
642 9 Advanced Coding and Coded Modulation Techniques

References
1. M. Cvijetic, I.B. Djordjevic, Advanced Optical Communication Systems and Networks (Artech House, Norwood, Jan. 2013)
2. I. Djordjevic, W. Ryan, B. Vasic, Coding for Optical Channels (Springer, New York, 2010)
3. S.B. Wicker, Error Control Systems for Digital Communication and Storage (Prentice Hall, Englewood Cliffs, 1995)
4. R.H. Morelos-Zaragoza, The Art of Error Correcting Coding, 2nd edn. (Wiley, Chichester, 2006)
5. D. Drajic, P. Ivanis, Introduction to Information Theory and Coding, 3rd edn. (Akademska Misao, Belgrade, 2009)
6. J.B. Anderson, S. Mohan, Source and Channel Coding: An Algorithmic Approach (Kluwer Academic Publishers, Boston/Dordrecht/London,
1991)
7. S. Lin, D.J. Costello, Error Control Coding: Fundamentals and Applications, 2nd edn. (Pearson Prentice Hall, Upper Saddle River, 2004)
8. T. Mizuochi et al., Forward error correction based on block turbo code with 3-bit soft decision for 10 Gb/s optical communication systems.
IEEE J. Sel. Top. Quantum Electron. 10(2), 376–386 (Mar./Apr. 2004)
9. T. Mizuochi, et al., Next generation FEC for optical transmission systems, in Proceedings of OFC 2003, Paper ThN1
10. I.B. Djordjevic, M. Arabaci, L. Minkov, Next generation FEC for high-capacity communication in optical transport networks. J. Lightwave
Technol. 27, 3518–3530 (2009)
11. I.B. Djordjevic, Advanced coding for optical communication systems, in Optical Fiber Telecommunications VI, ed. by I. Kaminow, T. Li,
A. Willner, (Elsevier/Academic Press, Boston, May 2013), pp. 221–296
12. I.B. Djordjevic, On advanced FEC and coded modulation for ultra-high-speed optical transmission. IEEE Commun. Surv. Tutorials 18(99),
1–31 (1 Mar. 2016). https://doi.org/10.1109/COMST.2016.2536726
13. L.R. Bahl, J. Cocke, F. Jelinek, J. Raviv, Optimal decoding of linear codes for minimizing symbol error rate. IEEE Trans. Inf. Theory IT-20(2),
284–287 (Mar. 1974)
14. F.J. MacWilliams, N.J.A. Sloane, The Theory of Error-Correcting Codes (North Holland, Amsterdam, 1977)
15. M. Ivkovic, I.B. Djordjevic, B. Vasic, Calculation of achievable information rates of long-haul optical transmission systems using instanton
approach. IEEE/OSA J. Lightwave Technol. 25, 1163–1168 (May 2007)
16. G. Bosco, P. Poggiolini, Long-distance effectiveness of MLSE IMDD receivers. IEEE Photon. Tehcnol. Lett. 18(9), 1037–1039 (1 May 2006)
17. M. Ivkovic, I. Djordjevic, P. Rajkovic, B. Vasic, Pulse energy probability density functions for long-haul optical fiber transmission systems by
using instantons and edgeworth expansion. IEEE Photon. Technol. Lett. 19, 1604–1606 (15 Oct. 2007)
18. S. Haykin, Communication Systems (Wiley, Hoboken, 2004)
19. M.E. van Valkenburg, Network Analysis, 3rd edn. (Prentice-Hall, Englewood Cliffs, 1974)
20. B. Vucetic, J. Yuan, Turbo Codes-Principles and Applications (Kluwer Academic Publishers, Boston, 2000)
21. B.P. Smith, F.R. Kschischang, Future prospects for FEC in fiber-optic communications. IEEE J. Sel. Top. Quantum Electron. 16(5),
1245–1257 (Sept.–Oct. 2010)
22. R.M. Tanner, A recursive approach to low complexity codes. IEEE Tans. Inf. Theory IT-27, 533–547 (1981)
23. C. Berrou, A. Glavieux, P. Thitimajshima, Near Shannon limit error-correcting coding and decoding: turbo-codes, in Proceedings of the IEEE
International Conference on Communications 1993 (ICC ‘93), vol 2, pp. 1064–1070, 23–26 May 1993
24. C. Heegard, S. Wicker, Turbo Codes (Kluwer Academic Press, Boston, 1999)
25. B. Moision, J. Hamkins, Deep space optical communications downlink budget: Modulation and coding, Jet Propulsion Laboratory, Pasadena,
CA, Interplanetary Netw. Progr. Rep., pp. 42–154, Apr.–Jun. 2003
26. Available at: http://ipnpr.jpl.nasa.gov/progress_report/42-154/154K.pdf
27. B. Moision, J. Hamkins, Coded modulation for the deep space optical channel: serially concatenated pulse-position modulation, Jet Propulsion
Laboratory, Pasadena, CA, Interplanetary Netw. Progr. Rep. 42–161, pp. 1–25, 15 May 2005. Available at: http://ipnpr.jpl.nasa.gov/progress_
report/42-161/161T.pdf
28. W.E. Ryan, Concatenated convolutional codes and iterative decoding, in Wiley Encyclopedia in Telecommunications, ed. by J. G. Proakis,
(Wiley, Hoboken, 2003)
29. R.M. Pyndiah, Near optimum decoding of product codes. IEEE Trans. Commun. 46, 1003–1010 (1998)
30. O.A. Sab, V. Lemarie, Block turbo code performances for long-haul DWDM optical transmission systems, in Proceedings of the OFC 2001,
Paper ThS5
31. R.G. Gallager, Low Density Parity Check Codes (MIT Press, Cambridge, MA, 1963)
32. D.J.C. MacKay, Good error correcting codes based on very sparse matrices. IEEE Trans. Inf. Theory 45, 399–431 (1999)
33. T. Richardson, R. Urbanke, Modern Coding Theory (Cambridge University Press, Cambridge, 2008)
34. B. Vasic, I.B. Djordjevic, R. Kostuk, Low-density parity check codes and iterative decoding for long haul optical communication
systems. J. Lightwave Technol. 21(2), 438–446 (Feb. 2003)
35. Z.-W. Li, L. Chen, L.-Q. Zeng, S. Lin, W. Fong, Efficient encoding of quasi-cyclic low-density parity-check codes. IEEE Trans. Commun.
54(1), 71–81 (Jan. 2006)
36. W.E. Ryan, An introduction to LDPC codes, in CRC Handbook for Coding and Signal Processing for Recording Systems, ed. by B. Vasic,
(CRC Press, London, 2004)
37. W.E. Ryan, S. Lin, Channel Codes: Classical and Modern (Cambridge University Press, Cambridge, 2009)
38. F.R. Kschischang, B.J. Frey, H.-A. Loeliger, Factor graphs and the sum-product algorithm. IEEE Trans. Inf. Theory 47, 498–519 (Feb. 2001)
39. J. Chen, A. Dholakia, E. Eleftheriou, M. Fossorier, X.-Y. Hu, Reduced-complexity decoding of LDPC codes. IEEE Trans. Commun. 53,
1288–1299 (2005)
40. L. Lan, L. Zeng, Y.Y. Tai, L. Chen, S. Lin, K. Abdel-Ghaffar, Construction of quasi-cyclic LDPC codes for AWGN and binary erasure
channels: a finite field approach. IEEE Trans. Inf. Theory 53, 2429–2458 (2007)
41. J.L. Fan, Array-codes as low-density parity-check codes, Second International Symposium on Turbo Codes, Brest, France, pp. 543–546,
Sept. 2000
References 643

42. S. Kudekar, T.J. Richardson, R.L. Urbanke, Threshold saturation via spatial coupling: why convolutional LDPC ensembles perform so well
over the BEC. IEEE Trans. Inf. Theory 57(2), 803–834 (Feb. 2011)
43. M. Lentmaier, G. Fettweis, K.S. Zigangirov, D. Costello, Approaching capacity with asymptotically regular LDPC codes, in Proceedings of the
Information Theory and Applications 2009, San Diego, CA, USA, pp. 173–177, Feb. 2009
44. K. Sugihara, et al., A spatially-coupled type LDPC code with an NCG of 12 dB for optical transmission beyond 100 Gb/s, in Proceedings of the
OFC/NFOEC 2013, Paper OM2B.4
45. D. Chang, et al., LDPC convolutional codes using layered decoding algorithm for high speed coherent optical transmission, in Proceedings of
OFC/NFOEC 2012, Paper OW1H.4
46. Y. Zhang, I.B. Djordjevic, Staircase rate-adaptive LDPC-coded modulation for high-speed intelligent optical transmission, in Proceedings of
the OFC/NFOEC 2014, Paper M3A.6
47. A.J. Felström, K.S. Zigangirov, Time-varying periodic convolutional codes with low-density parity-check matrix. IEEE Trans. Inf. Theory
45(6), 2181–2191 (1999)
48. I.B. Djordjevic, On the irregular nonbinary QC-LDPC-coded hybrid multidimensional OSCD-modulation enabling beyond 100 Tb/s optical
transport. J. Lightwave Technol. 31(16), 2969–2975 (15 Aug. 2013)
49. M. Arabaci, I.B. Djordjevic, L. Xu, T. Wang, Nonbinary LDPC-coded modulation for rate-adaptive optical fiber communication without
bandwidth expansion. IEEE Photon. Technol. Lett. 23(18), 1280–1282 (15 Sept. 2012)
50. M. Arabaci, I. B. Djordjevic, L. Xu, T. Wang, Hybrid LDPC-coded modulation schemes for optical communication systems, in Proceedings of
the CLEO 2012, Paper CTh3C.3
51. M. Arabaci, I.B. Djordjevic, R. Saunders, R.M. Marcoccia, High-rate non-binary regular quasi-cyclic LDPC codes for optical
communications. J. Lightwave Technol. 27(23), 5261–5267 (1 Dec. 2009)
52. P. Elias, Error-free coding. IRE Trans. Inf. Theory IT-4, 29–37 (Sept. 1954)
53. Y. Zhao, J. Qi, F.N. Hauske, C. Xie, D. Pflueger, G. Bauch, Beyond 100G optical channel noise modeling for optimized soft-decision FEC
performance, in Proceedings of the OFC/NFOEC 2012, Paper OW1H.3
54. F. Vacondio, et al., Experimental characterization of Gaussian-distributed nonlinear distortions, in Proceedings of the ECOC 2011, Paper
We.7.B.1
55. F. Chang, K. Onohara, T. Mizuochi, Forward error correction for 100 G transport networks. IEEE Commun. Mag. 48(3), S48–S55 (Mar. 2010)
56. Y. Miyata, K. Sugihara, W. Matsumoto, K. Onohara, T. Sugihara, K. Kubo, H. Yoshida, T. Mizuochi, A triple-concatenated FEC using soft-
decision decoding for 100 Gb/s optical transmission, in Proceedings of the OFC/NFOEC 2010, Paper OThL3
57. T. Mizuochi et al., Experimental demonstration of concatenated LDPC and RS code by FPGAs emulation. IEEE Photon. Technol. Lett. 21(18),
1302–1304 (Sept. 2009)
58. Z. Zhang, et al., Lowering LDPC error floors by postprocessing, in Proceedings of the IEEE GLOBECOM, 2008
59. G.J. Byers, F. Takawira, Fourier transform decoding of non-binary LDPC codes, in Proceedings Southern African Telecommunication
Networks and Applications Conference (SATNAC), Sept. 2004
60. M. Davey, D.J.C. MacKay, Low-density parity check codes over GF(q). IEEE Commun. Lett. 2(6), 165–167 (1998)
61. L. Barnault, D. Declercq, Fast decoding algorithm for LDPC over GF(2q), in Proceedings of the ITW2003, pp. 70–73
62. C. Spagnol, W. Marnane, E. Popovici, FPGA implementations of LDPC over GF(2m) decoders. IEEE Workshop on Signal Processing
Systems, Shanghai, China, pp. 273–278, 2007
63. D. Declercq, M. Fossorier, Decoding algorithms for nonbinary LDPC codes over GF(q). IEEE Trans. Commun. 55(4), 633–643 (2007)
64. V. Savin, Min-max decoding for non-binary LDPC codes, in Proceedings of the ISIT 2008, Toronto, ON, Canada, pp. 960–964,
6–11 July 2008
65. A. Voicila, F. Verdier, D. Declercq, M. Fossorier, P. Urard, Architecture of a low-complexity non-binary LDPC decoder for high order fields,
in Proceedings of the ISIT 2007, pp. 1201–1206
66. D. Zou, I.B. Djordjevic, FPGA implementation of concatenated non-binary QC-LDPC codes for high-speed optical transport. Opt. Express
23(10), 14501–14509 (18 May 2015)
67. D. Zou, I. B. Djordjevic, FPGA implementation of advanced FEC schemes for intelligent aggregation networks, in Proceedings of the
PhotonicsWest 2016, OPTO, Optical Metro Networks and Short-Haul Systems VIII, SPIE, San Francisco, CA, USA, vol. 9773, pp. 977309-
1–977309-7, 13–18 Feb. 2016. (Invited Paper)
68. I.B. Djordjevic, S. Sankaranarayanan, B. Vasic, Projective plane iteratively decodable block codes for WDM high-speed long-haul transmis-
sion systems. IEEE/OSA J. Lightwave Technol. 22, 695–702 (Mar. 2004)
69. I.B. Djordjevic, B. Vasic, Performance of affine geometry low-density parity-check codes in long-haul optical communications. Eur. Trans.
Telecommun. 15, 477–483 (2004)
70. I.B. Djordjevic, B. Vasic, Projective geometry low-density parity-check codes for ultra-long haul WDM high-speed transmission. IEEE
Photon. Technol. Lett. 15, 784–786 (May 2003)
71. I.B. Djordjevic, S. Sankaranarayanan, B. Vasic, Irregular low-density parity-check codes for long haul optical communications. IEEE Photon.
Technol. Lett. 16, 338–340 (Jan. 2004)
72. I.B. Djordjevic, O. Milenkovic, B. Vasic, Generalized low-density parity-check codes for optical communication systems. J. Lightwave
Technol. 23(5), 1939–1946 (2005)
73. I.B. Djordjevic, L. Xu, T. Wang, M. Cvijetic, GLDPC codes with Reed-Muller component codes suitable for optical communications. IEEE
Commun. Lett. 12(9), 684–686 (2008)
74. I.B. Djordjevic, T. Wang, Multiple component codes based generalized LDPC codes for high-speed optical transport. Opt. Express 22(14),
16694–16705 (2014)
75. D. Zou, I. B. Djordjevic, An FPGA design of generalized low-density parity-check codes for rate-adaptive optical transport networks. In
Proceedings of the SPIE PhotonicsWest 2016, optical metro networks and short-haul systems VIII, SPIE, San Francisco, CA, USA, vol. 9773,
pp. 97730M-1–97730M-6, 13–18 Feb. 2016
76. J. Boutros, O. Pothier, G. Zemor, Generalized low density (Tanner) codes, in Proceedings of the IEEE Intternational Conference on
Communication (ICC’99), pp. 441–445, 1999
644 9 Advanced Coding and Coded Modulation Techniques

77. M. Lentmaier, K.S. Zigangirov, On generalized low-density parity-check codes based on Hamming component codes. IEEE Commun. Lett.
3(8), 248–250 (1999)
78. V.A. Zyablov, R. Johannesson, M. Loncar, Low-complexity error correction of Hamming-code-based LDPC codes. Probl. Inf. Transm. 45(2),
95–109 (2009)
79. I. Andersen, Combinatorial designs: construction methods, in Mathematics and Its Applications, (Ellis Horwood, Chichester, 1990)
80. M.-F. Huang, A. Tanaka, E. Ip, Y.-K. Huang, D. Qian, Y. Zhang, S. Zhang, P.N. Ji, I.B. Djordjevic, T. Wang, Y. Aono, S. Murakami,
T. Tajima, T.J. Xia, G.A. Wellbrock, Terabit/s Nyquist superchannels in high capacity fiber field trials using DP-16QAM and DP-8QAM
modulation formats. J. Lightwave Technol. 32(4), 776–782 (15 Feb. 2014)
81. G.-H. Gho, J.M. Kahn, Rate-adaptive modulation and low-density parity-check coding for optical fiber transmission systems. J. Opt. Commun.
Netw. 4(10), 760–768 (Oct. 2012)
82. M. Arabaci, I.B. Djordjevic, R. Saunders, R.M. Marcoccia, Polarization-multiplexed rate-adaptive non-binary-LDPC-coded multilevel
modulation with coherent detection for optical transport networks. Opt. Express 18(3), 1820–1832 (2010)
83. L. Beygi, E. Agrell, P. Johannisson, M. Karlsson, A novel multilevel coded modulation scheme for fiber optical channel with nonlinear phase
noise, in Proceedings of the IEEE GLOBECOM, 2010
84. J.L. Massey, Coding and modulation in digital communications, in Proceedings of the International Zurich Seminar on Digital
Communications, Zurich, Switzerland, pp. E2(1)–E2(4), 1974
85. H. Imai, S. Hirakawa, A new multilevel coding method using error correcting codes. IEEE Trans. Inf. Theory IT-23(3), 371–377 (May 1977)
86. U. Wachsmann, R.F.H. Fischer, J.B. Huber, Multilevel codes: theoretical concepts and practical design rules. IEEE Trans. Inf. Theory 45(5),
1361–1391 (July 1999)
87. K.R. Narayanan, J. Li, Bandwidth efficient low density parity check codes using multilevel coding and iterative multistage decoding, in
Proceedings of the 2nd Symposium on Turbo Codes and Related Topics, Brest, France, pp. 165–168, 2000
88. S. Benedetto, D. Divsalar, G. Montorsi, H. Pollara, Parallel concatenated trellis coded modulation, in Proceedings of the IEEE International
Conference on Communications, Dallas, TX, USA, pp. 974–978, 1996
89. G. Caire, G. Taricco, E. Biglieri, Bit-interleaved coded modulation. IEEE Trans. Inf. Theory 44(3), 927–946 (May 1998)
90. J. Tan, G.L. Stüber, Analysis and design of symbol mappers for iteratively decoded BICM. IEEE Trans. Wirel. Comm. 4(2), 662–672 (Mar.
2005)
91. J. Hou, P.H. Siegel, L.B. Milstein, H.D. Pfitser, Capacity-approaching bandwidth-efficient coded modulation schemes based on low-density
parity-check codes. IEEE Trans. Inf. Theory 49(9), 2141–2155 (2003)
92. L. Beygi et al., Coded modulation for fiber-optic networks. IEEE Sig. Proc. Mag. 31(2), 93–103 (Mar. 2014)
93. G. Ungerboeck, Channel coding with multilevel/phase signals. IEEE Trans. Inf. Theory IT-28, 55–67 (1982)
94. G.D. Forney, R.G. Gallager, G.R. Lang, F.M. Longstaff, S.U. Qureshi, Efficient modulation for band-limited channels. IEEE J. Sel. Areas
Comm. SAC-2(5), 632–647 (1984)
95. G.D. Forney, Coset codes-part I: introduction and geometrical classification. IEEE Trans. Inf.. Theory 34(5), 1123–1151 (Sept. 1988)
96. A. Goldsmith, Wireless Communications (Cambridge University Press, Cambridge, 2005)
97. A.R. Calderbank, N.J.A. Sloane, New trellis codes based on lattices and cosets. IEEE Trans. Inf. Theory IT-33, 177–195 (1987)
98. G.D. Forney, Trellis shaping. IEEE Trans. Inf. Theory 38(2), 281–300 (Mar. 1992)
99. I.B. Djordjevic, B. Vasic, Multilevel coding in M-ary DPSK/differential QAM high-speed optical transmission with direct detection. IEEE/
OSA J. Lightwave Technol. 24(1), 420–428 (Jan. 2006)
100. J. Hou, P.H. Siegel, L.B. Milstein, H.D. Pfitser, Capacity approaching bandwidth-efficient coded modulation schemes based on low-density
parity-check codes. IEEE Trans. Inf. Theory 49(9), 2141–2155 (Sept. 2003)
101. S. ten Brink, Designing iterative decoding schemes with the extrinsic information transfer chart. Intern. J. Electron. Commun. 54, 389–398
(Dec. 2000)
102. J. Hou, P.H. Siegel, L.B. Milstein, Design of multi-input multi-output systems based on low-density parity-check codes. IEEE Trans. Commun.
53, 601–611 (Apr. 2005)
103. I.B. Djordjevic, M. Cvijetic, L. Xu, T. Wang, Using LDPC-coded modulation and coherent detection for ultra high-speed optical transmission.
IEEE/OSA J. Lightwave Technol. 25, 3619–3625 (Nov. 2007)
104. S. Le Goff, A. Glavieux, C. Berrou, Turbo-codes and high-spectral efficiency modulation, Proceedings of the 1994 IEEE International
Conference on Communications (ICC’94), New Orleans, LA, USA, pp. 645–649, 1994
105. P. Robertson, T. Wörz, Coded modulation scheme employing turbo codes. Electron. Lett. 31, 1546–1547 (Aug. 1995)
106. P. Robertson, T. Wörz, Bandwidth-efficient turbo trellis-coded modulation using punctured component codes. IEEE J. Sel. Areas Commun.
16(2), 206–218 (Feb. 1998)
107. I.B. Djordjevic, B. Vasic, Nonlinear BCJR equalizer for suppression of intrachannel nonlinearities in 40 Gb/S optical communications systems.
Opt. Express 14, 4625–4635 (29 May 2006)
108. I.B. Djordjevic, L.L. Minkov, L. Xu, T. Wang, Suppression of fiber nonlinearities and PMD in coded-modulation schemes with coherent
detection by using turbo equalization. IEEE/OSA J. Opt. Commun. Netw. 1(6), 555–564 (Nov. 2009)
109. R. Koetter, A.C. Singer, M. Tüchler, Turbo equalization. IEEE Sig. Proc. Mag. 21(1), 67–80 (Jan. 2004)
110. C. Duan, et al., A low-complexity sliding-window turbo equalizer for nonlinearity compensation, in Proceedings of the OFC 2012, Paper
JW2A.59
111. T. Fujimori, et al., A study on the effectiveness of turbo equalization with FEC for nonlinearity compensation in coherent WDM transmissions,
in Proceedings OECC/PS 2013, Paper OM2B
112. C. Douillard, M. Jézéquel, C. Berrou, A. Picart, P. Didier, A. Glavieux, Iterative correction of intersymbol interference: turbo equalization. Eur.
Trans. Telecommun. 6, 507–511 (1995)
113. E. Ip, J.M. Kahn, Compensation of dispersion and nonlinear impairments using digital backpropagation. J. Lightwave Technol. 26(20),
3416–3425 (2008)
114. E. Ip, J.M. Kahn, Nonlinear impairment compensation using backpropagation, in Optical Fibre, New Developments, ed. by C. Lethien,
(In-Tech, Vienna, Dec. 2009) (Invited Chapter)
References 645

115. Y. Zhang, S. Zhang, I.B. Djordjevic, On the pragmatic turbo equalizer for optical communications, in Proceedings of the CLEO-PR&OECC/
PS 2013, Paper ThR2-4
116. Y. Zhang, M. Arabaci, I.B. Djordjevic, Evaluation of four-dimensional nonbinary LDPC-coded modulation for next-generation long-haul
optical transport networks. Opt. Express 20(8), 9296–9301 (9 Apr. 2012)
117. Y. Zhang, S. Zhang, I.B. Djordjevic, F. Yaman, T. Wang, Extending 100G transatlantic optical transmission over legacy DMF fibers using
time-domain four-dimensional nonbinary LDPC-coded modulation, In Proceedings of ECOC 2013, Paper P.4.18
118. F. Yaman, S. Zhang, Y.-K. Huang, E. Ip, J.D. Downie, W.A. Wood, A. Zakharian, S. Mishra, J. Hurley, Y. Zhang, I.B. Djordjevic, M.-F.
Huang, E. Mateo, K. Nakamura, T. Inoue, Y. Inada, T. Ogata, First quasi-single-mode transmission over transoceanic distance using few-mode
fibers, in Proceedings of the OFC Postdeadline Papers, Paper Th5C.7, Los Angeles, CA, USA, pp. 22–26, Mar. 2015
119. Z.H. Peric, I.B. Djordjevic, S.M. Bogosavljevic, M.C. Stefanovic, Design of signal constellations for Gaussian channel by iterative polar
quantization, in Proceedings of the 9th Mediterranean Electrotechnical Conference, Tel-Aviv, Israel, vol. 2, pp. 866–869, 18–20 May, 1998
120. I.B. Djordjevic, H.G. Batshon, L. Xu, T. Wang, Coded polarization-multiplexed iterative polar modulation (PM-IPM) for beyond 400 Gb/s
serial optical transmission, in Proceedings of the OFC/NFOEC 2010, Paper OMK2
121. A.P.T. Lau, J.M. Kahn, Signal design and detection in presence of nonlinear phase noise. J. Lightwave Technol. 25(10), 3008–3016 (Oct.
2007)
122. T. Liu, I.B. Djordjevic, On the optimum signal constellation design for high-speed optical transport networks. Opt. Express 20(18),
20396–20406 (27 Aug. 2012)
123. T. Liu, I.B. Djordjevic, Optimal signal constellation design for ultra-high-speed optical transport in the presence of nonlinear phase noise. Opt.
Express 22(26), 32188–32198 (29 Dec. 2014)
124. R.E. Blahut, Computation of channel capacity and rate distortion functions. IEEE Trans. Inf. Theory IT-18, 460–473 (1972)
125. T. Liu, I.B. Djordjevic, Multi-dimensional optimal signal constellation sets and symbol mappings for block-interleaved coded-modulation
enabling ultra-high-speed optical transport. IEEE Photonics J. 6(4), Paper 5500714 (Aug. 2014)
126. T. Liu, I.B. Djordjevic, T. Wang, Optimal signal constellation design for ultra-high-speed optical transport in the presence of phase noise, In
Proceedings of the CLEO 2014, Paper STu3J.7
127. T. Liu, I.B. Djordjevic, M. Li, Multidimensional signal constellation design for channel dominated with nonlinear phase noise, in Proceedings
of the 12th International Conference on Telecommunications in Modern Satellite, Cable and Broadcasting Services – TELSIKS 2015, Nis,
Serbia, pp. 133–136, 14–17 Oct. , 2015
128. I.B. Djordjevic, A. Jovanovic, M. Cvijetic, Z.H. Peric, Multidimensional vector quantization-based signal constellation design enabling beyond
1 Pb/s serial optical transport networks. IEEE Photonics J. 5(4), Paper 7901312 (Aug. 2013)
129. S. Zhang, F. Yaman, M.-F. Huang, Y.-K. Huang, Y. Zhang, I. B. Djordjevic, E. Mateo, 40Tb/s transoceanic transmission with span length of
121km and EDFA-only amplification, in Proceedings of the OECC/ACOFT 2014, Paper THPDP1-3
130. S. Chandrasekhar, X. Liu. B. Zhu, D.W. Peckham, Transmission of a 1.2-Tb/s 24-carrier no-guard-interval coherent OFDM superchannel over
7200-km of ultra-large-area fiber, in Proceedings of the ECOC 2009, Paper PD2.6
131. T. Pfau, S. Hoffmann, R. Noé, Hardware-efficient coherent digital receiver concept with feedforward carrier recovery for M-QAM
constellations. J. Lightwave Technol. 27, 989–999 (2009)
132. X. Liu, S. Chandrasekhar, et al., Digital coherent superposition for performance improvement of spatially multiplexed coherent optical OFDM
superchannels. Opt. Express 20, B595–B600 (2012)
133. J.G. Proakis, Digital Communications, 4th edn. (McGraw-Hill, New York, 2001)
134. X. Liu, S. Chandrasekhar, P.J. Winzer, Digital signal processing techniques enabling multi-Tb/s superchannel transmission. IEEE. Sig. Proc.
Mag. 31(2), 16–24 (Mar. 2014)
135. T.H. Lotz et al., Coded PDM-OFDM transmission with shaped 256-iterative-polar modulation achieving 11.15-b/s/Hz intrachannel spectral
efficiency and 800-km reach. J. Lightwave Technol. 31, 538–545 (2013)
136. A. Goldsmith, S.-G. Chua, Adaptive coded modulation for fading channels. IEEE Trans. Commun. 46, 595–601 (May 1998)
137. I.B. Djordjevic, G.T. Djordjevic, On the communication over strong atmospheric turbulence channels by adaptive modulation and coding. Opt.
Express 17(20), 18250–18262 (25 Sept. 2009)
138. I.B. Djordjevic, Adaptive modulation and coding for free-space optical channels. IEEE/OSA J. Opt. Commun. Netw. 2(5), 221–229 (May
2010)
139. I.B. Djordjevic, Adaptive modulation and coding for communication over the atmospheric turbulence channels, in Proceedings of the IEEE
Photonics Society Summer Topicals 2009, Paper TuD3. 3, Newport Beach, CA, USA, 20–22 July 2009
140. I.B. Djordjevic, G.T. Djordjevic, Adaptive modulation and coding for generalized fading channels, in Proceedings of the IEEE TELSIKS 2009,
Nis, Serbia, pp. 418–422, 7–9 Oct. 2009
 Spread Spectrum, CDMA, and Ultra-Wideband
Communications 10

Abstract
The subject of this chapter is devoted to spread spectrum (SS), CDMA, and ultra-wideband (UWB) communication
systems. After the introduction of SS systems, both direct sequence-spread spectrum (DS-SS) and frequency-hopping-
spread spectrum (FH-SS) systems are described in detail. Regarding the DS-SS systems, after typical transmitter and
receiver configurations are described, we describe in detail the synchronization process including acquisition and tracking
stages. To deal with the multipath fading effects, the use of RAKE receiver is described. The section on DS-SS systems
concludes with the spreading sequences description, in particular pseudo-noise (PN) sequences. In this section, after
introduction of the basic randomness criterions, we introduce maximal-length sequences (m-sequences) and describe their
basic properties. In the same section, the concept of nonlinear PN sequences is introduced as well. Regarding the FH-SS
systems, both slow-frequency hopping (SFH) and fast-frequency hopping (FFH) systems are introduced. We provide the
detailed description of FH-SS systems, including both transmitter and demodulator as well as various time acquisition
schemes. In the same section we describe how to generate the FH sequences. After the introduction of CDMA systems, we
describe various signature waveforms suitable for use in DS-CDMA systems including Gold, Kasami, and Walsh
sequences. We then describe relevant synchronous and asynchronous CMDA models, with special attention devoted to
discrete-time CDMA models. In the section on multiuser detection (MUD), we describe how to deal with interference
introduced by various CDMA users. The conventional single-user detector is used as the reference case. The various MUD
schemes are described then, with special attention being paid to the jointly optimum MUD. The complexity of jointly
optimum MUD might be prohibitively high for certain applications, and for such applications we provide the description of
the decorrelating receiver, which can compensate for multiuser interference but enhances the noise effect. To solve for this
problem, the linear minimum mean-square error (MMSE) receiver is introduced. The section on MUD concludes with
description of the nonlinear MUD schemes employing decisions on bits from other users (be preliminary or final), with
description of successive cancelation and multistage detection schemes. In the section on optical CDMA (OCDMA), we
describe both incoherent and coherent optical detection-based OCDMA schemes. Regarding the incoherent optical
detection-based schemes, we first describe how to design various unipolar signature sequences, known as optical
orthogonal codes (OOC), followed by the description of basic incoherent OCDMA schemes including time spreading,
spectral amplitude coding (SAC), and wavelength hopping (WH) schemes. Related to coherent optical detection-based
OCDMA systems, we describe both pulse laser-based and CW laser-based OCDMA systems. Further, the hybrid OFDM-
CDMA systems are described, taking the advantages of both OFDM and CDMA concepts into account, particularly in
dealing with frequency selective fading and narrowband interference, while at the same time supporting multiple users. The
following broad classes of hybrid OFDM-CDMA systems are described: multicarrier CDMA (MC-CDMA), OFDM-
CDMA, and OFDM-CDMA-FH systems. In the section on UWB communications, we describe the concepts and
requirements as well as transmitter and receiver configurations. Further, various modulation formats, suitable for UWB
communications, are described and categorized into time-domain category (pulse-position nodulation and pulse-duration
modulation) and shape-based category [on-off keying, pulse-amplitude modulation, bi-phase modulation (BPM) or BPSK,
and orthogonal pulse modulation]. Regarding the pulse shapes suitable for orthogonal pulse modulation (OPM), the
modified Hermite polynomials, Legendre polynomials, and orthogonal prolate spheroidal wave functions (OPSWF) are

# The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 647
I. B. Djordjevic, Advanced Optical and Wireless Communications Systems, https://doi.org/10.1007/978-3-030-98491-5_10
648 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

described. Regarding the UWB channel models suitable for indoor applications, the model due to Saleh and Valenzuela is
described. Both types of UWB systems are described: the impulse radio UWB (IR-UWB), typically DS-CDMA-based, and
multiband OFDM (MB-OFDM). After the conclusion section, the set of problems is provided for deeper understanding of
the chapter material.

10.1 Spread Spectrum Systems

The main issues in digital communications are related to the efficient use of the bandwidth and power. However, in certain
situations, in a very hostile environment, in the presence of eavesdroppers and jammers, we may want to scarify these
efficiencies in order to enable the secure and reliable transmission. One way to achieve this goal is through the spread
spectrum communications [1–7]. Signals belonging to the spread spectrum class must simultaneously satisfy the following
spreading and despreading criterions. The generic spread spectrum digital communication system is illustrated in Fig. 10.1.
The spreading criterion indicates that information-bearing signal is spread in frequency domain, on a transmitter side, as
shown in Fig. 10.1, with the help of the spreading signal so that the resulted signal spectrum is much larger than the bandwidth
of the original information-bearing signal. On the other hand, the despreading criterion indicates that the recovery of
information-bearing signal from noisy spread spectrum received signal is achieved by correlating the received signal with
the synchronized replica of the spreading signal, as illustrated in Fig. 10.1. The channel encoder and decoder, the basic
ingredients of a digital communication system, are shown as well. If there is no any interference/jammer present, an exact
copy (replica) of the original signal can be recovered on the receiver side, and the spread spectrum system performance is
transparent to the spreading-despreading process.
In the presence of the narrowband interference signal, introduced during transmission, during the despreading process, the
bandwidth of the interference signal will be increased by the receiver side spreading signal. Since the filter that follows
despreading matches the information-bearing signal, the interfering spread signal will be significantly filtered, and the output
signal-to-interference ratio (SIR), often called signal-to-noise ratio (SNR), will be improved. This improvement in SIR due to
spread spectrum processing is commonly referred to as the processing gain (PG) and will be formally introduced below.

10.1.1 Spread Spectrum Systems Fundamentals

In a typical digital communication problem, we require the dimensionality of the system D to be smaller than or equal to the
number of signaling waveforms M. In spread spectrum systems, on the other hand, we require the dimensionality N to be
much larger than M. Let us observe M linearly independent signaling real-valued waveforms {sm(t)} fsm ðt Þ; 0  t < T gMm¼1 :
At every signaling interval T, one signaling waveform is selected at random, indicating that the data rate is log2M/T. The
dimensionality D  M is related to the signal set bandwidth BD by D ffi 2BDT. We are concerned with embedding this signal
set into much larger signal space of dimensionality N> > D. The required bandwidth of the spread spectrum (SS) signal,
denoted as BSS, can be determined from the following relationship N ffi 2BSST. We can now represent the m-th transmitted
signaling waveform in terms of orthonormal basis functions fϕn ðt Þ; 0  t < T gNn¼1 as follows:

X
N
sm ðt Þ ¼ smn ϕn ; m ¼ 1, 2, ⋯, M; 0  t < T, ð10:1Þ
n¼1

where

Fig. 10.1 The generic spread Information-bearing Estimated

Channel Channel
spectrum digital communication sequence encoder
Modulator Channel Demodulator
decoder sequence
system. The spreading sequences
on transmitter and receiver sides
are synchronized
Spreading
Spreading
sequence
sequence
10.1 Spread Spectrum Systems 649

ZT
smn ¼ sm ðt Þϕn ðt Þdt: ð10:2Þ
0

The basis functions satisfy the orthogonality principle:

ZT 
1, n ¼ k
ϕn ðt Þϕk ðt Þdt ¼ δnk ¼ : ð10:3Þ
0, n 6¼ k
0

We also assume that all signaling waveforms have the same energy Es, in other words:

ZT X
N 
  
s2m ðt Þ dt ¼ s2mn ≜Es , ð10:4Þ
n¼1
0

where we use the notation hi to denote the ensemble averaging. To better hide the signaling waveform into N-dimensional
space, we chose to select the coefficients of expansion smn independently so that they have zero mean and the correlation given
by [6]:

Es
hsmn smk i ¼ δ : ð10:5Þ
N nk

Now we can represent narrowband jamming (interference) signal in terms of basis functions as:

X
N
J ðt Þ ¼ J n ϕn , 0  t < T, ð10:6Þ
n¼1

with corresponding energy, denoted as EJ, being:

ZT X
N
J 2 ðt Þdt ¼ J 2n ≜EJ : ð10:7Þ
n¼1
0

When the jamming signal is additive to transmitted signal, and by ignoring the noise, the received signal is given by:

X
N
r ðt Þ ¼ sm ðt Þ þ J ðt Þ ¼ ðsmn þ J n Þϕn ; m ¼ 1, 2, ⋯, M; 0  t < T: ð10:8Þ
n¼1

The output of the m-th branch of correlator receiver can be determined as:

ZT ZT X
N  
rm ¼ r ðt Þsm ðt Þdt ¼ ½sm ðt Þ þ J ðt Þsm ðt Þdt ¼ s2mn þ smn J n : ð10:9Þ
n¼1
0 0

The mean value of rm given sm will be then:

0 1
XN 
B 2  C X 2 
N
hr m jsm i ¼ @ smn þ hsmn i hJ n iA ¼ smn ¼ E s , ð10:10Þ
n¼1
|ffl{zffl} n¼1
¼0

while the mean value of rm, assuming equal probable transmission:

650 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

hr m i ¼ hr m jsm iPðsm Þ ¼ E s =M: ð10:11aÞ

|fflfflffl{zfflfflffl}|fflffl{zfflffl}
Es 1=M

The variance of rm given sm can be determined now by:

D E D E XN D 2 E
ðr m  hr m jsm iÞ2 jsm ¼ ðr m  Es Þ2 jsm ¼ s2mn þ smn J n  s2mn
n¼1
X
N    ð10:11bÞ
¼ s2mn J 2n ¼ E s E J =N:
n¼1 |ffl{zffl}
E J =N

The variance of rm is then:

D E
EE EE
ðr m  E s Þ2 ¼ s J Pðsm Þ ¼ s J : ð10:12Þ
N MN

The SIR can be then determined as:

E s 2
hr m i2 Es N E
SIR ¼ D E¼ M
¼ ¼ s  PG, ð10:13Þ
ðr m  E s Þ2
Es EJ M
EJ |{z} EJ
MN
PG

where PG is the processing gain, also known as the spreading factor. By assuming that M ¼ D, we can express the PG as
follows:

N 2BSS T BSS
PG ¼ ffi ¼ , ð10:14Þ
D 2BD T BD

indicating that PG is in fact the ratio of spread spectrum bandwidth to information signal bandwidth. In practice, the PG ranges
from 10 to 1000.
Depending on how the spread spectrum signals have been generated, the SS systems can be classified into two broad
categories: (i) direct sequence-spread spectrum (DS-SS) and (ii) frequency hopping-spread spectrum (FH-SS).
In DS-SS modulation process, the modulated signal s(t) is multiplied by a wideband spreading signal, also known as
spreading code/sequence, denoted by sc(t). Typically, the spreading code has a constant value + or  1 over a time duration Tc,
which is much shorter than the symbol duration Ts. The bits in spreading code are commonly referred to as chips, while the
duration of the bit in spreading code Tc is known as the chip time. Finally, the rate 1/Tc is called the chip rate. The bandwidth
of spreading signal, denoted as Bc, is approximately 1/Tc, so that the processing gain can be estimated as PGffiBSS/
BD ffi (1/Tc)/(1/Ts) ¼ Ts/Tc. The multiplication of s(t) and sc(t) in time domain, namely, s(t)sc(t), can be represented as the
convolution in frequency domain S( f )*Sc( f ), where S( f ) is the Fourier transform (FT) of s(t), while Sc( f ) is the FT of sc(t).
The corresponding bandwidth of the resulting signal is approximately equal to Bc + BD. In the presence of additive white
Gaussian noise (AWGN) process z(t), the received SS signal is given by r(t) ¼ s(t)sc(t) + z(t). After despreading, the resulting
signal becomes:

sc ðt Þr ðt Þ ¼ sðt Þ s2c ðt Þ þ sc ðt Þzðt Þ ¼ sðt Þ þ z0 ðt Þ: ð10:15Þ

|ffl{zffl} |fflfflfflffl{zfflfflfflffl}
¼1 z0 ðt Þ

Since z'(t) still has the white Gaussian statistics, the despreading does not have effect on signal transmitted over AWGN.
When the jamming (interference) signal is wideband, it can be approximated by an equivalent AWGN process, and the end
effect will be reduction of overall SNR. On the other hand, when the jamming signal is narrowband, in frequency domain after
despreading it can be represented as Sc( f )*J( f ), where J( f) ¼ FT{J(t)}. Therefore, the power of the jamming signal
gets distributed over SS bandwidth, and the corresponding jamming signal power after the matched filter is approximately
10.1 Spread Spectrum Systems 651

BSS/BD ffi PG times lower. This indicates that SS approach is effective in dealing with narrowband interference. We will now
show that SS approach is also effective in dealing with multipath fading-induced intersymbol interference (ISI).
For simplicity, let us observe the two-ray multipath model, introduced in Chap. 2, whose impulse response is given by
h(t) ¼ aδ(t) + bδ(t  τ), where the first component corresponds to the line of sight (LOS) component, while the second
component is the reflected component, which arrives at receiver side with delay τ. The corresponding frequency response is
given by H( f) ¼ a + b exp.(j2πfτ). The received signal, in the absence of noise, after two-ray multipath channel, can be
represented in frequency domain as [Sc( f )*S( f )]H( f ). The corresponding time domain representation is given by asc(t)
s(t) + bsc(t-τ)s(t-τ). After despreading we can write as(t) + b sc(t)sc(t-τ)s(t-τ). Since the reflected component is asynchronous
with respect to the spreading signal, it will not be despread, and the power of reflected component after the matched filter will
be reduced approximately PG times. This indicates that DS-SS systems are effective in dealing with multipath fading-
induced ISI.
In FH-SS, we occupy a large bandwidth by randomly hopping input data-modulated carrier form one frequency to another.
Therefore, the spectrum is spread sequentially rather than instantaneously. The term sequentially is used to denote that
frequency hops are arranged in pseudo-randomly ordered code sequence sc(t). The chip time Tc dictates the time elapsed
between frequency hops. The FH-SS is suitable to combine with M-ary frequency-shift keying (MFSK). When N different
carrier frequencies are used in FH-SS systems, the bandwidth occupied is given by NBD, where BD is the bandwidth of original
information-bearing signal, as before. The FH-SS systems are attractive for defense/military applications. Depending on the
rate at which frequency hops occur, we can classify the FH-SS systems into two broad categories:

• The slow-frequency hopping (SFH), in which the symbol rate of MFSK Rs ¼ 1/Ts is an integer multiple of the hop rate
Rc ¼ 1/Tc. In other words, several symbols are transmitted with the same carrier frequency (for each frequency hop).
• The fast-frequency hopping (FFH), in which the hop rate Rc is an integer multiple of the MFSK symbol rate Rs. In other
words, the carrier frequency gets changed several times during the transmission of the same symbol. Clearly, in multipath
fading environment, FFH scheme provides inherent frequency diversity.

In the presence of AWGN, the FH does not have influence on AWGN performance. Let us consider the narrowband
interference at carrier frequency fn of bandwidth B¼BD. When the FH frequency coincides with the interference carrier, the
particular symbol transmitted will be affected by the interference. However, this occurs 1/N fraction of time. The power of
interference is effectively reduced approximately 1/N times. However, compared to DS-SS systems, where the power of
narrowband interference is reduced all the time, the FH-SS systems are affected by the full power of the jammer a fraction of
the time. In FFH systems, the interference affects the portion of symbol only, and this problem can be solved by coding, to
correct for occasional errors. On the other hand, in SFH systems many symbols are affected by the interference, so that the
coding with interleaving is required to solve for this problem.
Regarding the influence of multipath fading-induced ISI, for two-ray channel, on FH-SS systems, whenever τ > Tc, the
reflected component will exhibit a different carrier frequency than LOS component, and will not affect the detection of LOS
component. However, when τ < Tc, the FFH system will exhibit the flat fading, while the SFH system will exhibit either slow-
varying flat fading for interference bandwidth B < 1/τ or slow-varying frequency-selective fading for B > 1/τ.

10.1.2 Direct Sequence-Spread Spectrum (DS-SS) Systems

As we discussed in the previous section, in DS-SS systems, the modulated signal s(t) is multiplied by a wideband spreading
signal sc(t), very often derived from pseudo-noise (PN) sequences [1–6], which possess noise-like properties which are as
follows: (i) the mean value is approximately zero, and (ii) the autocorrelation function is approximately δ-function. Additional
details on PN sequences will be provided later in this section. More generally, DS refers to the specific way to generate SS
signals satisfying the following three conditions [5]:

(i) The chip waveform of duration Tc, denoted as pTc, is chosen such that:

Z1
pT c ðt ÞpT c ðt  nT c Þdt ¼ 0, 8n ¼ 1, 2, ⋯ ð10:16Þ
1

(ii) The number of chips is N.

(iii) The binary spreading sequence is of length N, namely, b ¼ (b1,b2,. . .,bN), and does not carry any information.
652 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

The DS-SS waveform (signal) can be then generated as:

X
N
sc ðt Þ ¼ A c ð2bi  1ÞpT c ½t  ðn  1ÞT c , ð10:17Þ
n¼1

where Ac is the normalization constant. Very often the rectangular pulses of duration Tc are used as the chip waveform:

1, 0  t < 1
pT c ðt Þ ¼ rectðt=T c Þ, rectðt Þ ¼ : ð10:18Þ
0, otherwise

However, there exist significant spectral components outside the spectral null 1/Tc. When someone is concerned with
spectral efficiency, in particular for CDMA systems, the following waveform can be used for the chip waveform [5]:

pT c ðt Þ ¼ sincð2t=T c  1Þ, sincðt Þ ¼ sin ðπt Þ=ðπt Þ: ð10:19Þ

10.1.2.1 DS-SS Transmitters and Receivers

To implement DS-SS system, we can start from either spreading sequence generator or spreading signal generator. The DS-SS
modulator for QPSK, based on PN sequence generator, is shown in Fig. 10.2a.
The spreading sequence generators used for in-phase and quadrature channels are typically identical and generate sequence
b ¼ (b1b2. . .bN). The block channel encoder (n,k) generates the codeword c ¼ (c1c2. . .cn). The sequence after mod-2 adder can
L
be obtained by bi ci. The in-phase component before product modulator can be represented as:

Fig. 10.2 DS-SS QPSK signal Mod-2

generation based on (a) binary adder
Binary Polar
spreading sequence generator and spreading NRZ
(b) spreading signal generator sequence encoder
cos ct

Data Adder DS-SS QPSK

Channel Wireless
Oscillator channel
sequence encoder signal
/2

Binary Polar sin ct

spreading NRZ
sequence encoder

(a)

Spreading
signal
generator
Polar
NRZ cos ct

Data encoder Adder

Channel Wireless
Oscillator channel
sequence encoder
Polar
NRZ /2
encoder
Spreading sin ct
signal
generator

(b)
10.1 Spread Spectrum Systems 653

Fig. 10.3 Various in-phase

r(t) Matched
Channel Estimated
signal demodulator architectures
filter ADC
for DS-SS signals: (a) matched- decoder sequence
p*Tc(Tc-t)
filter and spreading sequence- 2bi 1
based demodulator, (b) correlator
Spreading
and spreading sequence-based Chip-rate
demodulator, and (c) spread sequence
clock
signal-based demodulator with generator
correlator
(a)
r(t) Tc
Channel Estimated
· dt ADC
decoder
0 sequence
2bi 1
p*Tc(t) Chip-rate
Spreading
sequence
clock
generator

(b)

r(t) Tc
Estimated
Channel
· dt ADC
decoder
0 sequence
(2bi 1)pTc(t iTc)
p*Tc(t)
Spreading
Chip-rate
signal
clock
generator

(c)

I i ðt Þ ¼ ð2bi  1Þð2ci  1ÞpT c ðt  iT c Þ: ð10:20Þ

Similar equation holds for the quadrature channel. After product modulation, the in-phase and quadrature SS signals are
added together and transmitted toward remote destination by a transmit antenna. The spreading signal generator-based
configuration of DD-SS QPSK modulator is shown in Fig. 10.2b. The spreading signal generator generates the signal
according to Eq. (10.17). The mod-2 adder version is easier to implement instead of waveform multiplication.
On the receiver side, we can also employ either spreading sequence generator or spreading signal generator to perform
despreading. Additionally, either matched-filter receiver or correlation receiver configurations can be used. To facilitate
explanations, only in-phase demodulator structures are summarized in Fig. 10.3. The quadrature demodulator structure is very
similar to the one shown in Fig. 10.3.
The received in-phase equivalent low-pass signal can be represented by:

r ðt Þ ¼ ð2bi  1Þð2ci  1ÞpT c ðt  iT c Þ þ zðt Þ, ð10:21Þ

where z(t) originates from the jamming signal and/or Gaussian noise. When z(t) is Gaussian noise or when interference can be
modeled as equivalent zero-mean Gaussian process with power spectral density (PSD) N0/2 ¼ PJTc/2, where PJ is the average
power of the jamming (interference) signal, we can use either matched-filter or correlator-based receiver, with corresponding
spreading sequence-based configurations provided in Fig. 10.3a, b. The corresponding spreading signal generator-based
demodulator with correlator is shown in Fig. 10.3c. Given that (2bi  1)2 ¼ 1 in either case, we have successfully performed
the despreading operation. Based on Chap. 6, we conclude that bit-error probability Pb for uncoded binary PSK (BPSK) can
be estimated as:
654 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

rffiffiffiffiffiffi
rffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
1 Eb 1 Eb 1 E b =T b T b
Pb ¼ erfc ¼ erfc ¼ erfc
2 N0 2 PJ T c 2 PJ T c
rffiffiffiffiffiffiffiffiffiffiffi
ð10:22Þ
1 Pb T b
¼ erfc , Pb ¼ Eb =T b ,
2 PJ T c

where Pb is the average signal power and Tb is the bit duration.

Given that the ratio Tb/Tc is the processing gain, we can define the jamming margin [dBs], denoted as MJ, as follows:

Eb
M J ½dB ¼ PG½dB  10 log 10  L½dB, ð10:23Þ
N0 min

where (Eb/N0)min is the minimum required bit energy-to-jamming “noise” PSD ratio to support a prescribed bit error
probability and L [dB] represents implementation and other losses. Therefore, the jamming margin describes the residual
advantage that SS system exhibits against the jammer once we subtract (Eb/N0)min [dB] and L [dB] from PG [dB].
For M-ary PSK, the codeword error probability can be upper-bounded by the union bound as follows [3]:

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

1X
M
Eb
PM  erfc Rw , ð10:24Þ
2 m¼2 N0 c m

where Rc is the coder rate of block code, defined by Rc ¼ k/n, and wm is the weight of the codeword Cm.
We now explain how the DS-SS systems can be efficient in dealing with multipath fading-induced ISI. Let us assume that
wireless channel introduces several multipath components so that the corresponding impulse response can be represented by
h(t) ¼ aLOSδ(t-τLOS) + a1δ(t-τ1) + a2δ(t-τ2) + . . ., wherein aLOS > a1 > a2 > . . . . Let us further denote the spreading sequence
(code) by sc(t), while the transmitted sequence by {sk}. The baseband modulated signal can be represented by esðt Þ ¼
P
k sk gðffi t  kT s Þ, where g(t) is the baseband modulator impulse response, which is assumed to be rectangular gðt Þ ¼
pffiffiffiffiffiffiffiffiffi
2=T s , 0  t < T s : We also assume that the chip waveform is rectangular given by Eq. (10.18). For receiver shown in
Fig. 10.3(c), the input to the correlator can be written as:

b
esðt Þ ¼ f½esðt Þsc ðt Þ cos ðωc t Þ  hðt Þgsc ðt  τÞ cos ðωc t þ ϕÞ
ð10:25Þ
þzðt Þsc ðt  τÞ cos ðωc t þ ϕÞ:

Now by further assuming that multipath components am, m ¼ 2,3,. . . can be neglected, while the receiver locks to the LOS
component (τLOS ¼ 0, ϕ ¼ 0), we can rewrite Eq. (10.25), after substitution for impulse response h(t), as follows:

^~sðtÞ ¼ aLOS~sðtÞ s2 ðtÞ cos2 ðωc tÞ þ a1~sðt  τ1 Þsc ðt  τ1 Þcos½ωc ðt  τ1 Þsc ðtÞcosðωc tÞ
c
|ffl{zffl}
¼1 ð10:26Þ
þzðtÞsc ðtÞcosðωc tÞ:

We are concerned in demodulation of the m-th symbol, and assuming that multipath fading-induced ISI originates from the
transmission of the (m-k)-th symbol, we can set τ1 ¼ kTs to obtain:
10.1 Spread Spectrum Systems 655

ZT s ZT s
b 2
bsm ¼ esðt Þg ðT s  t Þdt ¼ aLOS sm cos 2 ðωc t Þ dt
Ts |fflfflfflfflfflffl{zfflfflfflfflfflffl}
2½1þ cos ð2ωc t Þ
0 0 1

ZT s
2
þa1 smk sc ðt Þsc ðt  τ1 Þ cos ðωc t Þ cos ½ωc ðt  τ1 Þdt
Ts |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} ð10:27Þ
2½ cos ðωc τ 1 Þþ cos ð2ωc tωc τ1 Þ
0 1

rffiffiffiffiffi ZT s
2
þ zðt Þsc ðt Þ cos ðωc t Þdt :
Ts
0
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
zm

Given that the double-frequency terms in (10.27) will not survive the low-pass filtering (integration), the previous equation
simplifies to:

ZT s
1
bsm ffi aLOS sm þ a1 smk cos ðωc τ1 Þ sc ðt Þsc ðt  τ1 Þdt þ zm , ð10:28Þ
Ts
0
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
ρc ðt, tτ1 Þ

where ρc(t, t  τ1) is the autocorrelation function of the spreading sequence (code), defined as:

ZT s
1 X
N
1
ρc ðt, t  τÞ≜ sc ðtÞsc ðt  τÞdt ¼ s ðnT c Þsc ðnT c  τÞ, N ¼ T=T c : ð10:29Þ
Ts N n¼1 c
0

When the autocorrelation of the spreading sequence is periodic in T, then autocorrelation becomes only function in τ:

ZT s
1
ρc ðt, t  τÞ ¼ sc ðt Þsc ðt  τÞdt ¼ ρc ðτÞ, ð10:30Þ
Ts
0

and we can rewrite (10.28) as:

bsm ffi aLOS sm þ a1 smk ρc ðτ1 Þ cos ðωc τ1 Þ þ zm : ð10:31Þ

Clearly, when T ¼ Ts and ρc(τ1) ¼ δ(τ1), the multipath fading-induced ISI gets completely eliminated by the despreading
procedure. So, one of the key topics in DS-SS systems is to properly design the spreading sequence (code) of finite length,
whose autocorrelation over Ts is approximately δ-function.
In the discussion above, we assumed that the receiver is perfectly locked to the LOS, which is not always the case in
practice, and we need to study the synchronization process of synchronizing the locally generated spreading sequence with the
received SS signal.

10.1.2.2 Acquisition and Tracking

The synchronization process is generally accomplished in two steps [2, 8]: (i) acquisition or coarse synchronization step and
(ii) tracking or fine synchronization step. In the acquisition step, we are coarsely aligning the timing of local spreading
sequence with incoming SS signal. Once the acquisition step is completed, we perform the fine timing alignment with a help
of phase-locked loop (PLL)-based approaches.
The DS acquisition can be performed in either serial or parallel search fashion. The serial search acquisition, illustrated in
Fig. 10.4, represents a low-cost solution, in which we use a single correlator or matched filter to serially search for the correct
phase of DS spread sequence. The locally generated spreading sequence is correlated with incoming spreading signal. The
656 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

Threshold

kTc
Received
· dt Comparator
DS-SS signal 0

Spreading
Feedback
sequence
search control
generator

Fig. 10.4 The DS-SS serial search acquisition

Spreading
sequence
generator
sc(t)
kTc

· dt
0

sc(t Tc/2)
kTc Comparator
Received
· dt (select the Output
DS-SS signal 0 largest signal
input)
sc[t (2Nc 1)Tc/2]
…

kTc

· dt
0

Fig. 10.5 The DS-SS parallel search acquisition

incoming signal is compared with the preset threshold at search dwell time intervals kTc, where k> > 1. If the comparator
output is below the threshold value, the delay of locally generated spreading sequence is incremented by 0.5Tc and correlation
is reexamined.
Once the threshold is exceeded the acquisition step will be completed, while the tracking step will be initiated. Assuming
that the uncertainty between the local spreading sequence and received one is Nc chips, the maximum acquisition time to
perform a fully serial DS search is:
 
T acquisition max
¼ 2N c kT c : ð10:32Þ

The mean acquisition time for a serial DS search system has been found in [9] to be:

  ð2  PD Þð1 þ KPFA Þ
T acquisition ¼ N c kT c , ð10:33Þ
PD

where PD and PFA are probabilities of the correct detection and false alarm, respectively. We assumed that the time required to
verify detection is NckTc.
To reduce the acquisition time of DS serial search, the parallel search acquisition, illustrated in Fig. 10.5, can be used. In
this acquisition scheme, the locally generated spreading sequence sc(t) is available together with 2Nc delayed versions, spaced
Tc/2 apart. Clearly, 2Nc correlators (matched filters are needed), and the comparator after kTc seconds select the largest input.
As the number of chips k increases, the probability of synchronization error decreases. When for each k the correlator outputs
are examined to determine the largest one, the maximum time interval required for a fully parallel search is:
 
T acquisition max
¼ kT c : ð10:34Þ

If an incorrect output is chosen, additional k chips will be required to determine the correct output, so that the average
acquisition time can be estimated as [10, 11]:
10.1 Spread Spectrum Systems 657

 
T acquisition ¼ kT c PD þ 2kT c PD ð1  PD Þ þ 3kT c PD ð1  PD Þ2 þ ⋯
kT ð10:35Þ
¼ c,
PD

where PD is the probability of detection. The complexity of fully parallel acquisition is high; however, the average acquisition
time is much shorter compared to the serial search acquisition.
Another acquisition technique, which has rapid acquisition time, but employs sequential estimation, is shown in Fig. 10.6.
It is commonly referred to as rapid acquisition by sequential estimation (RASE) search technique [12]. With switch in
position 1, the first n chips are detected and stored inside n-stage PN sequence generator. The PN generator has a unique
property that the next combination of the register states depends only on the present states, so, when the first n chips are
correctly estimated, all next chips from a local PN generator will be correctly generated, and the switch can be moved to
position 2. Now we correlate the output from n-stage PN sequence generator with that of incoming SS signal. Once the
threshold is exceeded after kTc seconds, we assume that correct coarse synchronization is achieved. On the other hand, when
the output of correlator is below the threshold, we move the switch in position 1, and the register is reloaded with the estimates
of next n received chips, and the procedure is repeated. Clearly, when the noise level is sufficiently low, the minimum
acquisition time will be nTc. The RASE provides rapid acquisition time; however, it is sensitive to noise and jamming signals.
Other approaches to acquisition can be found in [10, 11].
Once the acquisition step is completed, we move to the tracking step. There are two types of tracking spreading sequence
(code) loops, noncoherent loops and coherent ones. In coherent loops, the carrier frequency and phase are perfectly known so
that the loop can operate on baseband level. Given that the carrier frequency and phase are not known due to Doppler and
various multipath effects, the noncoherent loops to track PN sequences are more common. The popular noncoherent tracking
loops include (i) the delay-locking loop (DLL), also known as full-time early-late tracking loop, and (ii) the tau-dither loop
(TDL), also known as time-shared early-late tracking loop. The DLL for BPSK signals is provided in Fig. 10.7. Given that the

Fig. 10.6 The rapid acquisition Threshold

by sequential estimation (RASE)
search technique
kTc

· dt Comparator
0

2
Received n-stage PN
PN sequence
sequence
DS-SS signal chip detector Pos. 1 Switch generator

Feedback
search control

Fig. 10.7 The delay-locked loop Despreading

(DLL) for DS-SS signal tracking
correlator
Estimated
Demodulator
data
sc t
Early correlator
Received Square-law
BPF
DS-SS signal detector

PN sequence
VCO Loop-filter
T generator
sc t c +
2 Tc
sc t
2
Band-pass Square-law
filter (BPF) detector
Late correlator
658 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

Fig. 10.8 The tau-dither loop Estimated

(TDL) for DS-SS signal tracking Demodulator
data

Received Square-law
BPF
DS-SS signal detector

Tau-dither
generator
Early path
PN sequence
VCO Loop-filter
generator
Late path

tracking step follows the acquisition step, we assume that local PN sequence generator is offset from the incoming spreading
sequence by τ < Tc/2. The local PN generator provides two PN sequences: sc(t + Tc/2 + τ) for early correlator and sc(t-Tc/2 + τ)
for late correlator. The square-law detector is applied to erase the modulation data, since |  1|2 ¼ 1. The feedback signal at the
output of comparator indicates whether the frequency of voltage control oscillator (VCO) should be increased, when the late
correlator output is larger in magnitude, or decreased. The increase in VCO frequency is proportional to 1/τ, namely, when
VCO frequency is increased, τ is decreased and vice versa. On such a way, the τ is getting smaller and smaller, and in limit we
have that sc ðt Þsc ðt þ τÞ ! 1. For sufficiently small τ, the dispreading correlator output signal represents the dispread signal
τ!0
that is further demodulated in BPSK demodulator. The main problem DLL is facing is related to imperfect gain balancing of
early and late correlator branches, which can cause the offset in feedback signal, when magnitudes are expected to be the
same, resulting in nonzero error signal.
To avoid this problem, the tau-dither loop, shown in Fig. 10.8, can be used, in which only one correlator is used for both
code tracking and dispreading functions. The incoming sequence is correlated with either early or late version of locally
generated PN sequence, which is dictated by tau-dither generator. The tau-dither signal is a square wave signal that alternates
between +1 and  1 and controls the switch that alternatively passes early path (advanced) and late path (delayed) versions of
PN sequence. Since the same BPF and square-law detector are used with both paths, the problem of identical transfer
functions of DLL is solved.

10.1.2.3 RAKE Receiver

The DS-SS receiver, described in Fig. 10.3, will synchronize to one of the multipath components, not necessarily the strongest
one. To solve this problem, the RAKE receiver [13], shown in Fig. 10.9, should be used. The assumption is here that there are
M-1 multipath components, in addition to LOS, with delay being a multiple of chip interval Tc. Therefore, the impulse
P
response of such multipart system, based on Chap. 2, can be represented as hðt Þ ¼ M m¼1 am δðt  mT c Þ, where am is the gain
related to the m-th multipath component. The component with m ¼ 0 corresponds to the LOS component. The spreading
sequence and different delayed versions serve as referent signals. It is also possible to perform the down-conversion first prior
to the multiplication with spreading signal. We also assume that co-phasing has been performed after dispreading, to ensure
that different multipath components add constructively, and the weights wm are chosen according to the desired diversity
scheme. The coherent detection is then performed. For selection combining, we select the branch with strongest gain. With
equal gain combining, the demodulators’ outputs are summed up with equal weights. In maximal ratio combining (MRC), the
weights are proportional to corresponding pertinent signal strengths. To be more precise, in MRC the weight in each branch is
proportional to the signal-to-interference-and-noise power ratio. Each branch in RAKE receiver gets synchronized to a
corresponding multipath component. The m-th branch demodulator output can be written as:

ðmÞ
X
M
ðmÞ
bsk ffi am s k þ am0 sk,m0 ρc ½ðm  m0 ÞT c  þ zk , ð10:36Þ
m0 ¼1, m0 6¼m

which represents the generalization of Eq. (10.31). In Eq. (10.31), symbol sk gets transmitted over the k-th symbol interval,
while sk,m’ symbol gets transmitted over [(k-m’)Ts, (k + 1-m’)Ts] symbol interval. When ρc(τ) ffi 0, for |τ| > Tc, and the ISI in
each branch is negligible, the performance of RAKE receiver will be identical to the performance of various diversity schemes
described in Chap. 8 (see also ref. [14]).
In the rest of this section, we describe spreading sequences suitable for DS-SS systems [15–17] and beyond.
10.1 Spread Spectrum Systems 659

Fig. 10.9 The generic Phase and gain

configuration of RAKE receiver
adjustments
Coherent
w0, 0
demodulator

sc t
Coherent
w1,

Diversity combiner
1
demodulator

sc t Tc Estimated

…
…
Received

…
DS-SS signal data
Coherent
wm, m
demodulator

…
…

…
sc t mTc
Coherent
wM, M
demodulator

sc t MTc

Clock

s0 s1 s2 sm
Output
1 2 m
sequence
Flip-flop

The feedback logic

Fig. 10.10 The generic block scheme of the feedback shift register

10.1.2.4 Spreading Sequences

A pseudo-noise (PN) sequence is a binary sequence, typically generated by means of a feedback shift register, having a noise-
like waveform. The most popular PN sequences are maximal-length or m-sequences, Gold sequences, Walsh sequences, and
Kasami sequences. The pseudo-random sequences of 1 and  1 must satisfy the following randomness criterions [16, 17]:

(i) Balance property indicates that the number of binary symbols of one type must be approximately equal to the number of
symbols of other type (more precisely, the difference should not be larger than one).
(ii) Run property indicates that shorter successive runs of symbols of the same type are more probable. More precisely,
one-half of runs has length 1, one-fourth of runs has length 2, one-eighth of runs has length 3, and so on.
(iii) Correlation property indicates that autocorrelation function must have a large peak in the origin and to decay fast as
separation index from origin increases. The most relevant is the pseudo-random sequence whose autocorrelation function
has two values:

1, k ¼ 0
ρ c ðk Þ ¼ , ð10:37Þ
const < 1, 0 < jk j < L

where L is the sequence length. The pseudo-random sequences satisfying Eq. (10.37) are commonly referred to as the
perfect sequences.
The generic block scheme of the feedback shift register is shown in Fig. 10.10. Let si(k) denote the state of the i-th flip-flop
after the k-th clock pulse. Then the state of the shift register can be described as s ¼ s1(k)s2(k). . .sm(k). Given that the number
660 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

flip-flops is m, the number of possible states of the shift register is 2m. Clearly, the sequence generated by shift register must be
periodic, with a period at most 2m.
The logic can be nonlinear, and corresponding sequence will be binary nonlinear PN sequence. The possible number of
m
Boolean functions of m variables will be 22 . The feedback branch symbol is obtained as the arbitrary Boolean function of the
shift register content (state). De Bruijn (or Good) graph representation [18, 19] can be used to describe this calculation, which
represents the state diagram. State corresponds to the content of the shift register, and edges represent possible transitions.
Each state has two incoming and two outgoing edges. As an illustration, we provide de Bruijn graphs for m ¼ 1 and m ¼ 2 in
Fig. 10.11, where we assumed that the next bit has been appended from the left. The maximal length (2m) sequence is obtained
by finding the path starting from one state and which visits every state only once and reading out the output symbols. De
m1
Bruijn has shown that the number of such sequences, known as de Bruijn sequences, is N m ¼ 22 m .
The de Bruijn sequences satisfy the following correlation property:

Rð0Þ ¼ 2n =L, RðiÞ ¼ 0 ði ¼ 1, 2, ⋯, m  1Þ, RðmÞ 6¼ 0: ð10:38Þ

The de Bruijn sequences are suitable for encryption. There exist many algorithms to generate the Bruijn sequences, such
as [20].
When the feedback logic in Fig. 10.10 is composed of modulo-2 adders only, the corresponding register is commonly
referred to as the linear feedback-shift register (LFSR). Because the all-zero state will always generate all-zeros sequence,
such a state is not allowed, and the period of a PN sequence cannot exceed 2m-1. When the period is exactly L ¼ 2m-1, the
corresponding PN sequence is known as a maximal-length sequence or m-sequence. The generic LFSR register is provided in
Fig. 10.12. The coefficients fi (i ¼ 1,2,. . .,m-1) have the value 1 or 0, depending whether the corresponding switch is closed or

0/00
0/0

00
0
0/01

0/10
1/0 0/1 10 01
1/01

1 1/10 0/11

11
1/1

1/11
m=1
m=2

Fig. 10.11 The de Bruijn graphs for m ¼ 1 and m ¼ 2

Fig. 10.12 The generic block Clock

scheme of the linear-feedback
shift register (LFSR)

Output
sm sm-1 s2 s1
sequence

f0=1 f1 f2 fm 2 fm 1 fm=1

mod-2
adder
10.1 Spread Spectrum Systems 661

open. The switches f0 and fm are always closed ( f0 ¼ fm ¼ 1). Assuming that initial content of the shift register smsm-1. . .s2s1 is
different from all-zero word, the i-th feedback output symbol can be determined by:

X
m
si ¼ f 1 si1 f 2 si2 ⋯ f m sim ¼ f j sij : ð10:39Þ
j¼1

After certain number of clocks, the initial content of register will repeat. The switches should be properly chosen such that
period is maximum possible L ¼ 2m-1.
By introducing the delay operator x ¼ z1, we can define the generating function as follows:

X
1
GðxÞ ¼ s i xi , ð10:40Þ
i¼0

where s0 is the very first bit being generated. Based on recurrent relation (10.39), we can represent the generating function as:

P
m  
f i xi si xi þ si1 xði1Þ þ ⋯ þ s1 x1
GðxÞ ¼ i¼1 P
m : ð10:41Þ
1 f i xi
i¼1

Since the initial content of the shift register can be any sequence of length m different from all-zero word, we can set
s1 ¼ s2 ¼ . . . ¼ s-(m-1) ¼ 0, s-m ¼ 1 (similar to unit response) to obtain:

1 1
G ð xÞ ¼ X
m ¼ , ð10:42Þ
f ð xÞ
1 f i xi
i¼1
|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
f ð xÞ

where f(x) is the characteristic polynomial. It can be shown that when f(x) is irreducible (not divisible with any polynomial of
lower degree), the period L will be independent of initial conditions, and L can be found as the smallest positive integer for
which 1-xL is divisible with the characteristic polynomial f(x) [16]. In other words, the characteristic polynomial is one of
factors of 1-xL. Let us assume that the characteristic polynomial is the product of two irreducible polynomials, that is, f(x) ¼ f
(x1)f(x2). Then we can perform the partial fraction expansion of 1/f(x) as follows:

1 a ð x Þ a2 ð x Þ
¼ 1 þ ; deg½ai ðxÞ < deg½ f i ðxÞ; i ¼ 1, 2 ð10:43Þ
f ð xÞ f 1 ð xÞ f 2 ð xÞ

The maximum possible period will be the least common multiplier of possible periods 2m1 þm2 ¼ 2m , 2m1 , 2m2 . Since

ð2m1  1Þð2m2  1Þ ¼ 2m1 þm2  2m1  2m2 þ 1 < 2m  2  2 þ 1 < 2m  1, ð10:44Þ

we conclude that the necessary condition to obtain the m-sequence is to make sure that the characteristic polynomial is
irreducible. However, this is not a sufficient condition. The sufficient condition will be satisfied when the characteristic
polynomial is primitive. The primitive polynomial is an irreducible polynomial with corresponding roots being the primitive
elements of finite field GF(2m). Finally, a primitive element is a field element that generates all nonzero elements of the field as
its successive powers.
As an illustration, to get m-sequence of length L ¼ 23–1 ¼ 7, we need to factorize x7  1 as follows:
  
x 7  1 ¼ ð x  1Þ x 3 þ x þ 1 x 3 þ x 2 þ 1 :
662 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

g1(x) = 1 + x + x3

s3 s2 s1

(a)
g2(x) = 1 + x2 + x3

s3 s2 s1

(b)

Fig. 10.13 Two different ways to generate m-sequences of length 7

Both polynomials g1(x) ¼ x3 + x + 1 and g2(x) ¼ x3 + x2 + 1 are irreducible and primitive and as such can be used to
generate m-sequences of length 7. The corresponding LSFRs are provided in Fig. 10.13. Assuming that initial content of shift
register is 001, the corresponding m-sequences are 0111001 and 1101001, respectively.
The generation of m-sequences can also be described using a matrix approach as follows [16]. The transformation matrix
corresponding to g1(x) ¼ x3 + x + 1 is given by:
2 3
1 1 0
6 7
M ¼ 40 0 1 5:
1 0 0

Let si be the content of register in the i-th step. The content of shift register in the next step will be si + 1 ¼ siM. In general,
the transformation matrix has the form:
2 3
f1 1 0 ⋯ 0
6 7
6 f2 0 1 ⋯ 0 7
6 7
M¼6
6⋮ ⋮ ⋮ ⋱ ⋮77, ð10:45Þ
6 7
4 f m1 0 0 ⋯ 1 5
fm 0 0 ⋯ 0

where the first column are coefficients of the characteristic polynomial, corresponding to the feedback. The rest of the matrix
is an (m-1)
(m-1) identity submatrix, while the elements in the last row, except the first element, are zeros. It can be shown
that the characteristic equation for this matrix is:
2 3
6  7
pðλÞ ¼ detðM  λIÞ ¼ ð1Þmþ1 41  f 1 x þ f 2 x2 þ ⋯ þ f m xm 5=xm , ð10:46Þ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
f ð xÞ

where x ¼ 1/λ. (Notice that, in GF(2),  sign can be replaced with + sign.)
10.1 Spread Spectrum Systems 663

R( )
LTc LTc

… 1/L …

2Tc

Fig. 10.14 Illustration of the autocorrelation function of the m-sequence

The period of m-sequence can be determined as the smallest power of M yielding to the identity matrix:

M L ¼ I: ð10:47Þ

Since every matrix satisfies its characteristic equation p(M) ¼ 0, if p(x) is the factor of 1-xL, then M will be the root of
eq. I-M L, given that M is the root of p(M). Therefore, when p(x) divides 1-xL, then ML ¼ I.
The autocorrelation function of the m-sequence is dependent on the chip waveform. When the chip waveform is a
rectangular pulse of duration Tc, given by Eq. (10.18), the continuous autocorrelation function will have a triangle shape
around the origin of duration 2Tc, in other words:
8
> L þ 1 jτ j
<1  , jτ j < T c
Rmsequence ðτÞ ¼ L Tc : ð10:48Þ
>
:  1 , jτj > T
c
L

On the other hand, the autocorrelation function of random binary sequence of symbols of duration Tc and amplitude either
+1 or  1 is given by:
8
< jτ j
1 , jτ j < T c
Rrandom ðτÞ ¼ Tc : ð10:49Þ
:
0, jτj > T c

Clearly, Eq. (10.49) can be derived from (10.48) by letting L to tend to plus infinity. Given that m-sequence is periodic with
period LTc, the corresponding autocorrelation function will be periodic as well and can be sketched as shown in Fig. 10.14.
The corresponding power spectral density (PSD) can be obtained by calculating the Fourier transform (FT) of the autocorre-
lation function. From FT theory, we know that the FT of the triangle function is the sinc2 function of frequency, so we expect
the continuous PSD to be sinc2 function, while the discrete portion of PSD will contain the series of δ-functions spaced
1/(NTc) apart, with magnitude dictated by the sinc2( fT) function. The magnitude of the D.C. component of PSD can be
determined by calculating the area below the autocorrelation curve, shown in Fig. 10.14, which is 1/L2. Therefore, the PSD of
m-sequence will be:

X
1

1 Lþ1 l l
PSDmsequence ðτÞ ¼ δð f Þ þ 2 sinc2 δ f : ð10:50Þ
L2 L L LT c
l¼1, l6¼0

10.1.3 Frequency Hopping-Spread Spectrum (FH-SS) Systems

As discussed in Sect. 10.1.1, in FH-SS systems, the large bandwidth occupation is achieved by randomly hopping input data-
modulated carrier from one frequency to another. Clearly, the spectrum is spread sequentially; in other words, the frequency
664 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

hops are arranged in pseudo-randomly ordered code sequence sc(t), and the time elapsed between frequency hops is dictated
by chip time Tc. Depending on the hope rate relative to symbol rate, the FH-SS systems can be categorized into two broad
categories: (i) the slow-frequency hopping (SFH), in which the symbol rate Rs is an integer multiple of the hop rate Rc ¼ 1/Tc,
and (ii) the fast-frequency hopping (FFH), in which the hop rate Rc is an integer multiple of the symbol rate Rs. The generic
FH-SS scheme is provided in Fig. 10.15. The FH spreading sequence (code), to be described below, is used as an input to the
frequency synthesizer, which generates the carrier hopping signal A cos [ωkt + ϕk(t)]. The carrier hoping signal is then used as
input to the modulator, which can be coherent, noncoherent, or differential, to perform the up-conversion to the carrier signal.
Such modulated signal is transmitted toward remote destination by transmit antenna, over wireless channel exhibiting
multipath fading, described by impulse response h(t), noise z(t), and jamming (interference) signal J(t). On the receiver
side, following the antenna reception, a synchronizer is used to synchronize the locally generated spreading sequence (code)
to that of the incoming SS signal. The synchronization procedure also has two steps, acquisition and tracking steps. Once
synchronization is achieved, the demodulation process takes place. When either differential or noncoherent modulation is
employed, it is not required to synchronize phases of receive and transmit signals.
For FFH-MFSK, we provide additional details of noncoherent demodulator in Fig. 10.16. For completeness of presenta-
tion, the configuration of FH-MSFK system is shown as well. After frequency dehopping, we perform filtering to select the

Wireless channel
FH-SS transmier (Tx) z(t) FH-SS receiver (Rx)
t=iTs
Sequence of Tx signal sˆi
Decision
Modulator h(t) Demodulator
symbols {si} s(t) circuit

A cos t t A cos t ˆ t
k k k k
J(t)
Frequency Local
synthesizer frequency
synthesizer
sc(t) Synchronizer
sc(t )
Spreading
sequence
generator Local spreading
sequence generator

Fig. 10.15 Generic FH-SS system architecture

Fig. 10.16 Configurations of (a) Data MFSK FH Wireless FH MFSK Estimated

FH-MSFK system and (b) modulator modulator channel demodulator modulator data
FFH-MFSK demodulator
PN PN
generator generator

(a)
Matched filters (MFs) &
envelope detectors (EDs)
Received
MF &
Clipper Accumulator
FH-SS signal ED 1
Decision
stage sˆi
MF &
Filter Clipper Accumulator (select the
ED 2
largest
input)
…

…
…

Frequency
hopper
MF &
Clipper Accumulator
ED M

(b)
10.1 Spread Spectrum Systems 665

Fig. 10.17 Parallel FH Received

Square-law Delay of
acquisition scheme BPF
FH-SS signal detector k1 hops

f1

Square-law Delay of
BPF
detector k2 hops

…
f2

…
Square-law Delay of kN
BPF
detector hops

fN

Received
FH-SS signal

Square-law
BPF Integrator
detector

Frequency
synthesizer/
hopper Threshold Comparator

Search
PN code
feedback
generator
control

Fig. 10.18 FH serial search acquisition scheme

desired bandwidth of data signal only. After that, signal is split into M branches, one for each frequency fk (k ¼ 1,. . .,M). The
k-th branch is composed of matched filter for frequency fk, envelope detector, clipper, and accumulator. The clipper is needed
in the presence of jamming signal. Demodulator does not make a decision on a chip level but rather on a symbol level. The
accumulator sums up the energies after all chips. In the decision stage, the accumulator output of highest cumulative energy is
selected.
The parallel FH acquisition scheme is shown in Fig. 10.17. The received FH-SS signal is split into N branches, each
corresponding to one of possible hopping frequencies. Correlators can be replaced by matched filters. After match filtering,
the square-low detection takes place, and in each branch the corresponding delays are introduced, depending on the spreading
sequence. Once the delays are properly chosen, the strong peak gets generated at the output, indicating that the local spreading
sequence matches that of the received signal.
To reduce the complexity of this acquisition scheme, single correlator/matched filter can be used, combined with frequency
synthesizer/hopper, as shown in Fig. 10.18. The PN generators control the frequency synthesizer/hopper. Once the correct
frequency hopping sequence is found, the comparator exceeds the threshold, and the acquisition indication is generated. The
tracking step is very similar to that of DS-SS systems.
We have discussed the impact of multipath fading on FH-SS systems already in Sect. 10.1.1. When multipath effects cause
either flat or frequency-selective fading, in the absence of interference signal, the performances are similar to those of not-FH
systems. In the presence of interference, let us assume that SFH is used and that interference (jamming) signal occupies K out
of N frequency bands. The symbol error probability Ps can be estimated as follows:

Ps ¼ Prðsymbol errorjno jammer ÞPrðno jammerÞ

|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
NK
N
ð10:51Þ
þPrðsymbol errorjjammer ÞPrðjammerÞ:
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}
K
N
666 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

Let us assume that during the dispreading, in the absence of jammer, the influence of multipath components is mitigated,
and the symbol SNR ρs is affected by the channel gain, so that we can use the expression of symbol error probability for
AWGN channel. As an illustration, for M-ary
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi QAM/PSK with coherent detection, we can use the following approximate
expression PðsAWGNÞ ’ aðM ÞQ bðM Þρs , with parameters a(M ) and b(M) being modulation and signal constellation size
dependent as shown in Chap. 6. What remains is to determine the symbol error probability in the presence of jammer, denoted
as PðsJammerÞ. When narrowband jammer inserts the Gaussian noise within the bandwidth of modulated signal B, such a jammer,
known as partial band noise jammer, it reduces the SNR, and corresponding SNR can be determined as:

N 0B
ρðsJammerÞ ¼ ρ, ð10:52Þ
N 0 B þ N Jammer B s

where NJammer is the PSD of the jammer noise source. The corresponding symbol error probability will be PðsJammerÞ ¼
 
PðsAWGNÞ ρðsJammerÞ :The overall symbol error probability, based on Eq. (10.51), can be determined as:

N K N0B K
Ps ðρs Þ ¼ PðsAWGNÞ ðρs Þ ðJammerÞ
þ Ps ρ : ð10:53Þ
N N 0 B þ N Jammer B s N

Let now the jamming signal be the tone at the hopped carrier frequency with some phase offset δϕi. The demodulator
output at the i-th symbol interval can be represented as:
pffiffiffiffiffi jδϕi
bsi ¼ ai si þ zi þ PJ e , ð10:54Þ

where ai is the channel gain coefficient after dispreading and PJ is the average power of the jamming tone. For M-ary PSK,
assuming that Arg{si} ¼ 0, the symbol error probability can be estimated by:
  pffiffiffiffiffi  
Ps ¼ 1  Pr Arg ai si þ zi þ PJ e jδϕi  π=M , ð10:55Þ

with additional details to be found in [10, 11].

In the rest of this section, we describe how to generate sequences for FH-SS systems. When we are concerned with
synthesis of sequences for multiuser FS-SS systems, the following parameters are relevant [17, 21]: number of available
frequency slots Q, sequence period L, cardinality of sequence set M, and maximum Hamming cross-correlation Hmax. The
Hamming cross-correlation for two sequences {x} and {y} of length L each is defined as:

X
L1  
H xy ðiÞ ¼ h xl ylþi , ð10:56Þ
l¼0


1, a ¼ b
where h(a,b) is the Hamming metrics defined as hða, bÞ ¼ . The hopping pattern can be defined as L-tuple:
0, a 6¼ b
 
fL ¼ f j1 , f j2 , ⋯, f jL , ð10:57Þ

where f jl is the symbol selected from a finite set of Q elements. The maximum number of patterns, for synchronous multiuser
case, with no more than k hits between any two patterns, is Qk + 1 [21]. For asynchronous transmission, the maximum number
of patterns reduces down to Qk + 1/L. The upper limit for maximum sequence length is given by [21]:

Qkþ1  1
L , ð10:58Þ
Qk  1

which indicates that for k ¼ 1 we have that Lmax ¼ Q + 1, while for k > 1, the Lmax ¼ kQ. Compared to the Singleton bound for
maximum distance separable (MDS) codes, this bound is not tight bound. The Singleton bound, as shown in Chap. 9, for
codeword length n and number of information bits kinfo, the Singleton bound is given by:
10.1 Spread Spectrum Systems 667

H max  n  k info þ 1: ð10:59Þ

The set of patterns with k hits is equivalent to the MDS code with kinfo ¼ k + 1. So only for k ¼ 1, these two bounds are in
agreement. The lower bound for Hmax is given by [22]:

ðL  bÞðL þ b þ QÞ
H max  , b ¼ L mod Q: ð10:60Þ
Q ð L  1Þ

It has been shown in Chap. 9 that Reed-Solomon (RS) codes satisfy the Singleton bound with equality sign and belong to
the class of MDS codes. If we are concerned with asynchronous multiuser transmission, and sequences of length L ¼ Q-1, we
can generate Qk sequences as follows [17, 23–25]. Let Q be a prime power for the finite field GF(Q) defined, and let α be a
primitive element in GF(Q). We need to determine the polynomial:

X
k
PðxÞ ¼ pi xi , pi 6¼ 0, ð10:61Þ
i¼0

such that the set of sequences:

 
fai g ¼ P αi ; i ¼ 0, 1, 2, ⋯, Q  2 ð10:62Þ

have autocorrelation with maximum of k-1. In particular, for one-coincidence sequences (k ¼ 1), the polynomial P(x) is
simple:

PðxÞ ¼ p0 þ p1 x, p0 2 GF ðQÞ, p1 ¼ const 2 GF ðQÞ: ð10:63Þ

The LSFR in the so-called parallel configuration can be used to determine different powers of primitive element as the
content of shift register. Now by adding p0 to the shift register and reading out the content as integer mod Q, we obtain the
desired frequency hopping sequence.
As an illustration, let us observe the GF(2) and extended field GF(Q ¼ 23), which can be obtained from primitive
polynomial g(x) ¼ 1 + x + x3, whose parallel implementation is shown in Fig. 10.19. The representation of GF(8) elements in
GF(2) is given by:

αi GF(2) representation Decimal representation

0 000 0
1 001 1
α 010 2
α2 100 4
α3 ¼ α + 1 011 3
α4 ¼ α2 + α 110 6
α5 ¼ α2 + α + 1 111 7
α6 ¼ α2 + 1 101 5

g(x) = 1 + x + x3

Fig. 10.19 Illustration of generation successive primitive element power from parallel configuration of LSFR
668 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

Now by calculating the polynomial P(x) ¼ x + p0, p02GF(8), we obtained the hopping sequences as follows [17, 25]:

Polynomial Hopping sequence

x+0 1,243,675
x+1 0352764
x+α 3,061,457
x + α2 5,607,231
x + α3 2,170,546
x + α4 7,425,013
x + α5 6,534,102
x + α6 4,716,320

10.2 Code-Division Multiple Access (CDMA) Systems

The spread spectrum systems can be used in multiaccess communications, in which multiple users share the spectrum.
Multiaccess communications suffer from multiaccess interference, which can be both intentional and unintentional, such as
crosstalk or multipath fading. Moreover, the system design, such as in code-division multiple access (CDMA), can introduce
the crosstalk among the users. When spread spectrum principles are used in multiaccess communication, sometimes we refer
to such scheme as spread spectrum multiple access (SSMA), which is one particular instance of CDMA. To deal with
multiuser interference, originating from multiple interfering signals, we use the multiuser detection (demodulation) [5], which
is also known as interference cancelation or co-channel interference suppression, and it is subject of investigation in Sect.
10.3. The performances of multiaccess communication are also dependent on the fact if the multiaccess communication link is
a downlink channel (one transmitter to many receivers scenario) or an uplink channel (many transmitters to one receiver
scenario). The downlink channel is also known as the forward or broadcast channel as well as point-to-multipoint channel, and
all signals are typically synchronous as they originate from the same transmitter. In contrast, in the uplink channel, also known
as multiaccess channel or reverse link as well as multipoint-to-point channel, the signals are typically asynchronous as they
originate from different users at different locations. In frequency-division multiple access (FDMA), different non-overlapped
frequency bands are allocated to different users. On the other hand, in time-division multiple access (TDMA), the time scale is
partitioned into time slots, which are shared among different users in a round-robin fashion. The TDMA requires precise time
synchronism. Therefore, in both FDMA and TDMA, users are required not to overlap each other in either time domain or
frequency domain. However, it is possible for users’ signals to be orthogonal while simultaneously overlapping in both time
and frequency domains, such as the one illustrated in Fig. 10.20a, and we refer to such multiaccess scheme as CDMA. Clearly,
the signature waveforms p1(t) and p2(t), assigned to two different users, are orthogonal to each other since:
Z T
hp2 jp1 i ¼ p1 ðt Þp2 ðt Þdt ¼ 0: ð10:64Þ
0

Now let both users apply antipodal signaling with aik2{1,+1}, i ¼ 1,2. Then modulated signals can be represented as:
X X
s1 ðt Þ ¼ a p ðt  iT Þ, s2 ðt Þ ¼
i 1i 1
a p ðt
i 2i 2
 iT Þ, hs2 js1 i ¼ 0: ð10:65Þ

The corresponding CDMA system model is provided in Fig. 10.20(b). Because the signature waveforms (“codes”) p1(t)
and p2(t) are orthogonal to each other, and perfect synchronism is assumed, the optimum receiver will be matched filter or
correlation receiver.
The FDMA can be considered as the spatial case of CDMA in which signature waveforms are defined as:

pk ðt Þ ¼ gðt Þ cos ð2π f k t Þ, j f k  f m j ¼ l=T, ð10:66Þ

where g(t) is the pulse of unit energy and duration T while l is an integer. In similar fashion, TDMA can be considered as a
special case of CDMA with signature waveforms being defined as:

1 t  ðk  1ÞT=K
pk ðt Þ ¼ pffiffiffiffiffiffiffiffiffi rect ; k ¼ 1, 2, ⋯, K ð10:67Þ
T=K T=K

In the next subsection, we discuss different signature waveforms suitable for use in CDMA systems.
10.2 Code-Division Multiple Access (CDMA) Systems 669

Fig. 10.20 (a) Two orthogonal p1(t)

signature waveforms assigned to p2(t)
two different users. (b) Two users
1 1
CDMA system model

t
T/2 T T

(a)

iT a1i+zi aˆ1i
· dt Decision
circuit
i 1T
s1(t) z(t)
r(t) p1(t iT)
Channel

s2(t) iT a2i+zi aˆ2i

· dt Decision
circuit
i 1T

p2(t iT)

(b)

10.2.1 Signature Waveforms

The spreading sequences suitable for DS-SS systems have already been discussed in Sect. 10.1.2, in particular in Sect.
10.1.2.4. On the other hand, the spreading sequences suitable for both FH-SS- and FH-based CDMA systems have been
discussed in Sect. 10.1.3. The DS-SS waveform given by Eq. (10.17) can be generalized as follows [5]:

X
N
sc ðt Þ ¼ Ac αn ϕn ðt Þ, ð10:68Þ
n¼1

where [α1,. . ., αn,. . ., αN] is the spreading code with components not necessarily binary and the basis functions {ϕi(t)} satisfy
the orthogonality conditions:

hϕm jϕn i ¼ δnm :

For DS-SS systems, the basis functions are chip waveforms given by ϕn ðt Þ ¼ pT c ½t  ðn  1ÞT c  and spreading code
components αn2{1,+1}. In FH-SS, the chips are modulated in frequency domain:

X
N 
cos ð2π f n t Þ, ðn  1ÞT c  t < nT c
sðcFHÞ ðt Þ ¼ Ac ϕn ðt, f n Þ, ϕn ðt, f n Þ ¼ ð10:69Þ
n¼1 0, otherwise

The dual for DS chip waveforms ϕn ðt Þ ¼ pT c ½t  ðn  1ÞT c  can be defined in frequency domain as:
670 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

 
n1
Φn ð f Þ ¼ p1=T f , ð10:70Þ
T

which can be written in time domain as:

n o
j2π ðn1Þt=T
ϕn ð f Þ ¼ e P1=T ðt Þ, P1=T ðt Þ ¼ FT 1 p1=T ð f Þ : ð10:71Þ

The corresponding scheme is known as multicarrier CDMA.

K
In CDMA systems, we assign different signature sequences (spreading codes) to different users fsck g . The cross-
k¼1
correlation of different spreading codes defines the amount of interference among users. In asynchronous CDMA, the signals
for different users arrive at different times at the receiver side, and we can define the cross-correlation of the spreading codes
assigned to user k and k0 as follows:

Z
1 X
Ts N
1
ρkk ðt, t  τÞ ¼
0 sck ðtÞsck0 ðt  τÞdt ¼ s ðnT c Þsck0 ðnT c  τÞ, ð10:72Þ
Ts 0 N n¼1 ck

where τ denotes the difference in time arrivals. Given that spreading codes fsck g are periodic with the period Ts, then cross-
correlation is dependent only on difference in time arrivals τ, and we can use ρkk'(τ) to denote ρkk'(t, t  τ). In synchronous
CDMA, the users’ signals arrive at the same time at the receiver side, and the corresponding cross-correlation becomes:

Z
1 X
Ts N
1
ρkk0 ð0Þ ¼ sck ðtÞsck0 ðtÞdt ¼ sck ðnT c Þsck0 ðnT c Þ
Ts 0 N n¼1 |fflfflfflffl{zfflfflfflffl}|fflfflfflfflffl{zfflfflfflfflffl}
ð1Þckn ð1Þck0 n ð10:73Þ
1 X
N
¼ ð1Þckn ð1Þck0 n ,
N n¼1

where we used ð1Þckn to denote sck ðnT c Þ. Clearly, the negative logic is used here: 1 to denote binary 1 and + 1 to denote
binary 0. Ideally, we would like in synchronous (asynchronous) CDMA to have ρkk’(0) ¼ 0 [ρkk’(τ) ¼ 0] whenever k 6¼ k’. The
codes set satisfying this cross-correlation property can be called as orthogonal spreading code set, otherwise non-orthogonal
spreading code set. The literature on PN sequences is quite rich [10, 11, 16–26]. Some examples of pseudo-random sequences
include Gold codes, Kasami sequences [10, 17], Barker sequences, etc. The cross-correlation between two synchronous
signature sequences can be determined as:

1 X
N
1
ρkk0 ð0Þ ¼ ð1Þckn ð1Þck0 n ¼ ð#agreements  #disagreementsÞ
N n¼1 N
1
¼ ½#agreements  ðN  #agreementsÞ
N ð10:74Þ
2
¼ 1 þ ð#agreementsÞ
N
2 X
n
¼ 1 þ ½c ¼ ck0 n ,
N n¼1 kn

where we used [P] to denote the Inversion function, which has value 1 when proposition P is true and zero otherwise:

: 1, P is true
½P¼ ð10:75Þ
0, P is false

For an m-sequence, defined as (c1,c2,. . .,cN), there are 2m-1 zeros and 2m-1 ones in the sequence. The sequence length is
N ¼ 2m-1, and when we let the m-sequence to repeat, the corresponding correlation function is given by:
10.2 Code-Division Multiple Access (CDMA) Systems 671

(
1 X
N 1, l ¼ 0, N, 2N, ⋯
RðlÞ ¼ ð1Þcn ð1Þcnþl ¼ 1 ð10:76Þ
N n¼1  , otherwise
N

The spreading codes for asynchronous CDMA must satisfy the following two constraints: (i) the out-of-phase autocorre-
lation must be low and (ii) the cross-correlation must be low as well. The m-sequences do satisfy the constraint (i); however,
their cross-correlation constraint is not that good. The Welch has shown that the maximum cross-correlation for any two
spreading (periodic) binary sequences of length N, taken from the set of size M, are lower bounded as follows [27]:

 1=2 pffiffiffiffi
M1
Rðk Þ  N  N: ð10:77Þ
MN  1

The Gold sequences (codes) [28, 29] have better cross-correlation properties than m-sequence but worse autocorrelation
properties. At this point it is convenient to introduce the concept of preferred pairs of m-sequences, which represent a pair of
m-sequences {a} and {b} of length N with a periodic cross-correlation function taking three possible values:
8
(
< 1=N
>
1 þ 2ðnþ1Þ=2 , n  odd
ρa,b ¼ t ðnÞ=N , t ð nÞ ¼ ð10:78Þ
>
: 1 þ 2ðnþ2Þ=2 , n  even
½t ðnÞ  2=N

Let us observe two sequences of length N ¼ 2m-1; the second is obtained from the first {an} by decimation (taking every q-
th element): {bn} ¼ {aqn}. When N and q are co-prime, with either q ¼ 2k + 1 or q ¼ 22k–2k + 1, andwhen m ¼ 0 mod 4 with
1, m  odd
the greatest common divider (GCD) for m and k satisfying the following constraint GCDðm, kÞ ¼ , then it
2, m  even
can be shown that cross-correlation between these two sequences will have three possible values, as specified by Eq. (10.78).
Therefore, {an} and {bn} ¼ {aqn} are preferred pair of sequences. They can be used to create the Gold sequences:
   
fag, fbg, fa þ bg, fa þ Dbg, a þ D2 b , ⋯, a þ DN1 b , ð10:79Þ

where D ¼ z1 is a delay operator, having trivalent cross-correlation (10.78). The first two sequences are m-sequences, while
the remaining N are not. For sufficiently large N, the maximum cross-correlation is either √2 (m-odd) or two times (m-even)
larger than that predicted by Welch bound. As an illustration, in Fig. 10.21, we describe how to generate Gold sequences of
length N ¼ 31 starting from polynomials g1(x) ¼ 1 + x2 + x5 and g2(x) ¼ 1 + x + x2 + x4 + x5.
Kasami sequences are also derived from m-sequences but have better cross-correlation properties than Gold codes [2, 10,
17]. There two different sets of Kasami sequences, used to generate Kasami codes, namely, the large set and the small set. The
small set is obtained with the help of the following generating polynomial: f(x)f 1(x), where f(x) corresponds to m-sequence {a}
of length L ¼ 2m-1 (m-even) and f 1(x) corresponds to an {a1} sequence obtained from {a} by decimation with q ¼ 2m/2 + 1.
The new sequence is shorter L ¼ 2m/2–1, and there are 2m/2 different sequences. The cross-correlation values are 3-valued as
well {1,-(2m/2 + 1), (2m/2–1)}/N, and the cross-correlation values satisfy the Welch bound with equality sign. The large set
is obtained with the help of generating polynomial: f(x)f 1(x) f 2(x), where f 2(x) corresponds to an {a2} sequence obtained from
{a} by decimation with q ¼ 2(m + 2)/2 + 1. The cross-correlation has five different values:
8
< 1=N
>  
ρa,b ¼ 1  2m=2 =N , ð10:80Þ
>
:  m=2 
1  2 þ 1 =N

and the number of sequences is large 23m/2 for m ¼ 0 mod 4 and 23m/2 + 2m/2 for m ¼ 2 mod 4.
For synchronous CDMA applications, the orthogonal sequences can be used. The orthogonality of two sequences,
represented as vectors |ai ¼ [a1,. . .,aN]T and |bi ¼ [b1 . . . bN]T, can be defined in similar fashion to orthogonality principle
for signals [see Eq. (10.64)] as follows:
672 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

X
N
hbjai ¼ an bn ¼ 0:
n¼1

The very popular orthogonal sequences are Walsh functions (sequences) [30–32], the set of bipolar sequences of size
N ¼ 2m, defined recursively as follows:

x1 ¼ fðþ1Þg, x2 ¼ fðþ1, þ1Þ, ðþ1, 1Þg,

ð10:81Þ
x3 ¼ fðþ1, þ1, þ1, þ1Þ, ðþ1, þ1, 1, 1Þ, ðþ1, 1, þ1, 1Þ, ðþ1, 1, 1, þ1Þg, ⋯

An alternative way to obtain the Walsh sequences is to interpret them as the rows (columns) of the Hadamard matrices,
defined recursively as:
2 3
þ1 þ1 þ1 þ1
" # 6 7
þ1 þ1 6 þ1 þ1 1 1 7
6 7
H 1 ¼ ½þ1, H 2 ¼ , H 4 ¼ H 2 ¼6 7, ⋯,
þ1 1
2
6 þ1 1 þ1 1 7
4 5 ð10:82Þ
þ1 1 1 þ1
" #
Hn Hn
H 2n ¼ H 2
n ¼ ,
Hn H n

where we use () to denote the Kronecker product of matrices. Clearly, the Hadamard matrix is symmetric and has equal
number of +1 and  1 in every row (column), except the first one. Additionally, any two rows (columns) are orthogonal to
each other. In conclusion, the rows (columns) of Hadamard matrix can be used as Walsh sequences.

10.2.2 Synchronous and Asynchronous CDMA Models

We first describe the basic models, followed by discrete-time models.

10.2.2.1 Basic CDMA Models

In synchronous CDMA, because the signals originating from different users are perfectly time-aligned, it is sufficient to
consider a single symbol interval [0, T]. To facilitate explanations, let us denote the k-th (k ¼ 1,2,. . .,K ) user antipodally
modulated signature waveform of unit energy with sk(t) instead of sck ðt Þused before. As the generalization of Fig. 10.20 to
K users, the received signal can be represented as:

X
K
r ðt Þ ¼ Ak bk sk ðt Þ þ σzðt Þ, ð10:83Þ
k¼1

where Ak is the amplitude of the k-th user at receiver side, bk2{+1, 1} is the k-th user information bit, and z(t) is a zero-mean
Gaussian noise of unit variance. When the real-valued signature waveforms are orthogonal to each other, the performance of

Fig. 10.21 Generation of one set

of Gold sequences form
polynomials g1(x) ¼ 1 + x2 + x5
and g2(x) ¼ 1 + x + x2 + x4 + x5
1 + x2 + x5

1 + x + x2 + x4 + x5
10.2 Code-Division Multiple Access (CDMA) Systems 673

the k-th user is dependent on signal-to-noise ratio Ak/σ, where σ signal is the standard deviation of Gaussian noise process.
When the signature waveforms are quasi-orthogonal (or even non-orthogonal), we can use the cross-correlation to quantify the
orthogonality:

ZT
ρkk0 ¼ hsk0 jsk i ¼ sk ðtÞsk0 ðtÞdt, ð10:84Þ
0

and since the waveform signatures are real-valued, the complex-conjugate operation can be omitted. By applying the
Bunjakowski-Cauchy-Schwartz inequality, the following upper bound on magnitude of cross-correlation is obtained:

Z 0 ZT


jρkk0 j ¼ sk ðtÞsk0 ðtÞdt  ksk kksk0 k ¼ 1, ksk k ¼ jsk ðtÞj2 dt ¼ 1:
2
ð10:85Þ

T 0

The cross-correlation matrix R can be defined as:

2 3
ρ11 ρ12 ⋯ ρ1K
6ρ ρ22 ⋯ ρ2K 7
6 21 7
R¼6 7; ρkk ¼ 1, k ¼ 1, ⋯, K ð10:86Þ
4⋮ ⋮ ⋱ ⋮ 5
ρK1 ρ2K ⋯ ρKK

It is straightforward to show that correlation matrix is non-negative definite, since for arbitrary vector |ai ¼ [a1 . . . aN]T we
have that:
2 3
ρ11 ρ12 ⋯ ρ1K 2 3
a
6 7 1
6 ρ21 ρ22 ⋯ ρ2K 7 6 a2 7 P
K P K
6 76 7
hajRjai ¼ ½a1 a2 ⋯aK 6 76 7 ¼ ak ak0 ρkk0
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}6 ⋮ ⋮ ⋱ ⋮ 7 4 ⋮ 5 k¼1 k0 ¼1
haj
4 5
aK
ρK1 ρ2K ⋯ ρKK |fflffl{zfflffl}
jai
ð10:87Þ
ZT ZT ! !
P
K P
K X
K X
K
¼ ak ak0 sk ðtÞsk0 ðtÞdt ¼ ak sk ðtÞ ak0 sk0 ðtÞ dt
k¼1 k 0 ¼1 k¼1 k 0 ¼1
0 0
 2
P 
K 
¼  ak sk ðtÞ  0:
k¼1 

In asynchronous CDMA, we need to specify delays in arrival for each user τk2[0,T]; k ¼ 1,2,. . .,K. Because different users’
signals arrive at different times at the receiver side, it is not sufficient to observe one symbol but block of symbols. Let us
denote the block of symbols for the k-th user containing 2 M + 1 symbols as follows [5]: bk ¼ [bk [M], . . ., bk [0], . . ., bk [M]].
The basic asynchronous CDMA model can be now represented as:

X
K X
M
r ðt Þ ¼ Ak bk ½isk ðt  iT  τk Þ þ σzðt Þ: ð10:88Þ
k¼1 i¼M

We further assume that information block length bk is sufficiently large so that edge effects are not relevant, so that we can
observe delays modulo-T. Clearly, when signals from all users arrive at the same time (τk ¼ τ), the asynchronous case reduces
down to the synchronous chase. Another interesting special case is when all amplitudes are the same (Ak ¼ A) and all users use
the same signature waveform, namely, sk(t) ¼ s(t). If we further assume that delays can be represented as τk ¼ (k-1)T/K, like in
TDMA, clearly, Eq. (10.88) reduces down to ISI case:
674 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

bk[i]
sk (current symbol)
k-th user
bm[i 1] bm[i]
sm (prev. sym.) sm (current sym.)
m-th user

mk km t

k m

Fig. 10.22 Illustration of asynchronous cross-correlations’ definitions, assuming that τk < τm

X
K X
M

k1
r ðt Þ ¼ Abk ½is t  iT  T þ σzðt Þ
k¼1 i¼M
K
X X
K M
ð10:89Þ
k1
¼ Ab½iK þ k  1s t  iT  T þ σzðt Þ:
k¼1 i¼M
|fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} K
bk ½i

Because the signals from users k and m have different delays, τk and τm, we have to specify two cross-correlations,
assuming that τ ¼ τm-τk > 0 (the signal from the k-th user arrives earlier than the signal from the m-th user for τ seconds), as
illustrated in Fig. 10.22. The cross-correlation ρkm is related to cross-correlation between the current symbol of k-th user and
current symbol of the m-th user, and it is defined as:

ZT
ρkm ¼ sk ðt Þsm ðt  τÞdt: ð10:90Þ
τ

On the other hand, the cross-correlation ρmk is related to cross-correlation between the current symbol of k-th user and
previous symbol of the m-th user, and it is defined as:

Zτ
ρmk ¼ sk ðt Þsm ðt þ T  τÞdt: ð10:91Þ
0

10.2.2.2 Discrete-Time CDMA Models

Given that most of the signal processing in CDMA systems is performed in the discrete-time (DT) domain, it makes sense to
develop the DT CDMA models.
Similar to the single-user case, the samples from correlators (matched filters) multiuser synchronous CDMA case, as shown
in Fig. 10.23, represent a sufficient statistics. The k-th correlator output can be represented as:

ZT ZT "X
K
#
rk ¼ r ðt Þsk ðt Þdt ¼ Ak bk sk ðt Þ þ σzðt Þ sk ðt Þdt
k¼1
0 0

X
K ZT ZT
¼ Ak bk þ Am bm sm ðt Þsk ðt Þdt þ σzðt Þsk ðt Þdt ð10:92Þ
m¼1, m6¼k
0 0
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
ρmk zk

X
K
¼ Ak bk þ Am bm ρmk þ zk ; k ¼ 1, 2, ⋯, K:
m¼1, m6¼k
10.2 Code-Division Multiple Access (CDMA) Systems 675

Fig. 10.23 Illustration of DT iT

r1[i]
synchronous CDMA model · dt
i 1T t=iT

…
…
b1s1(t) z(t) s1(t iT)

…
…
r(t) iT
rk[i]=Akbk +
bksk(t) Channel · dt
i 1T t=iT m k Ambm mk+zk

…
bKsK(t)

…
…
sk(t iT)

…
iT rK[i]
· dt
i 1T t=iT

sK(t iT)

This set of K-equations can also be written in more compact matrix form:
2 3
r1 2 3 2 3
b1 z1
6 7
6 r2 7 6 b2 7 6 z2 7
6 7 6 7 6 7
jr i ¼ 6 7 ¼ R diagðA1 , ⋯, AK Þ 6 7 þ 6 7
6⋮7 |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl} 4 ⋮ 5 4 ⋮ 5
4 5 A
bK zK
rK |fflffl{zfflffl} |fflffl{zfflffl} ð10:93Þ
jbi jzi

¼ R A jbi þ jzi
|{z} |{z}
b z

¼ R A b þ z:

It can easily be shown that the following is valid:

 
hzi ¼ 0, CovðzÞ ¼ z zT ¼ σ 2 R, ð10:94Þ

indicating that components in |zi are correlated, and in analogy with single-user ISI channel, we can perform whitening by
Cholesky factorization of the cross-correlation matrix [3, 33, 34], assuming that signals are real-valued:

R ¼ FT F, ð10:95Þ

where F is the lower triangular matrix with positive diagonal elements. Now, by multiplying the Eq. (10.93) by (FT)1 from
the left side, we obtain:

0 1 1 1
jr i ¼ ðFT Þ jri ¼ ðFT Þ FT F A b þ ðFT Þ z
|fflfflfflfflfflffl{zfflfflfflfflfflffl} |fflfflfflffl{zfflfflfflffl}
I z0 ð10:96Þ
0
¼F Abþz:

Given that:

1 1
hz0 i ¼ hðFT Þ zi ¼ ðFT Þ hzi ¼ 0,
1 1
Covðz0 Þ ¼ hz0 z0 T i ¼ hðFT Þ zzT F1 i ¼ ðFT Þ hzzT i F1
|ffl{zffl}
σ2 R ð10:97Þ
T 1 2 T 1
R F1 ¼ σ 2 ðF Þ FT FF
¼ ðF Þ σ |{z} 1
¼ σ 2 I,
|fflfflfflfflfflffl{zfflfflfflfflfflffl}|fflffl{zfflffl}
FT F I I

we conclude that components of |z’i are uncorrelated, and the whitening of the colored Gaussian noise has been successfully
conducted.
676 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

This synchronous model is useful when K signature waveforms are quasi-orthogonal and known on the receiver side.
However, in certain situations, such as antenna arrays, a different set of orthogonal functions is used, say fϕd ðt ÞgD
d¼1 . To
solve for this problem, we can project the signature waveforms along this new basis functions as follows:

ZT
skd ¼ sk ðt Þϕd ðt Þdt; d ¼ 1, ⋯, D: ð10:98Þ
0

In similar fashion, the received signal can be represented in terms of projection as:

ZT
rd ¼ r ðt Þϕd ðt Þdt; d ¼ 1, ⋯, D: ð10:99Þ
0

Now, after substituting the Eq. (10.83) into (10.99), we obtain:

ZT ZT "X
K
#
rd ¼ r ðt Þ ϕd ðt Þdt ¼ Ak bk sk ðt Þ þ σzðt Þ ϕd ðt Þdt
|{z} k¼1
0 P
K 0
Ak bk sk ðt Þþσzðt Þ
k¼1
ð10:100Þ
X
K ZT ZT X
K
¼ Ak bk sk ðt Þϕd ðt Þdt þ σzðt Þϕd ðt Þdt ¼ Ak bk skd þ zd ; d ¼ 1, ⋯, D:
k¼1 k¼1
0 0
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl}
skd zd

The corresponding matrix representation will be:

2
3 2 3 2 3
r1 sk1 z1
6r 7 X K 6 sk2 7 6 z2 7 X
K
! 6 2 7 ! !
r ¼6 7¼ A k bk 6
4⋮
7þ6 7¼
5 4⋮5 A k bk s k þ z : ð10:101Þ
4 ⋮ 5 k¼1 k¼1
skD zD
rD |fflffl{zfflffl} |fflffl{zfflffl}
! !
sk z

Let us now group the signature vectors into a signature matrix:

h i
! ! !
S ¼ s1 s2 ⋯ sK , ð10:102Þ

which is clearly of dimensionality D

K. By using this signature matrix, with the k-th column corresponding to the k-th user
!
signature s k , we can rewrite Eq. (10.101) as follows:

! !
r ¼ SAb þ z , ð10:103Þ
!
where matrix A and vector b are the same as introduced before. Let us now determine the covariance matrix of r as follows:
* +
  D T E D E D TE
! !! ! !T T !!
Cov r ¼ r r ¼ SAb þ z SAb þ z ¼ SAbðSAbÞ þ zz
|fflfflffl{zfflfflffl} |fflfflfflffl{zfflfflfflffl}
bT AT ST σ2 I
  ð10:104Þ
¼ SA bbT AT ST þ σ 2 I,
|fflffl{zfflffl}
I
T T
= SAA S þ σ 2 I,
10.2 Code-Division Multiple Access (CDMA) Systems 677

where we used in the second line the fact that users’ bits are independent. Clearly, the noise samples are independent and
Eq. (10.103) is valid regardless of whether the orthogonal basis fϕd ðt ÞgD K
d¼1 span the signature waveforms fsk ðt Þgk¼1 or not.
Regarding the DT representation of asynchronous CDMA system, without loss of generality, we assume that delays can be
arranged as τ1  τ2  ⋯  τK. The correlators’ outputs can be written as:

τk þZðiþ1ÞT

r k ½i ¼ r ðt Þsk ðt Þdt

τk þiT
τk þZðiþ1ÞT τk þZðiþ1ÞT
X
K X
M
¼ Am bm ½i sm ðt  iT  τk Þsk ðt Þdt þ σzðt Þdt
m¼1 i¼M ð10:105Þ
τk þiT τk þiT
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
zk ½i
τk þZðiþ1ÞT
X
K X
M
¼ A k bk ½ i  þ Am bm ½i sm ðt  iT  τk Þsk ðt Þdt þ zk ½i:
m¼1, m6¼k i¼M
τk þiT

Now we have to group the crosstalk terms, relative to the k-th user signal, into two categories, as illustrated in Fig. 10.24:
(i) the crosstalk terms for which τm > τk and (ii) crosstalk terms for which τl < τk. By taking the asynchronous cross-
correlations’ definitions, given by (10.90 and 10.91), into account, we can rewrite the equation above as follows:

X
K X
K
r k ½i ¼ Ak bk ½i þ Am bm ½iρkm þ Am bm ½i  1ρmk
m>k m>k
ð10:106Þ
X
K X
K
þ zk ½i þ Al bl ½i þ 1ρkl þ Al bl ½iρlk
l<k l<k

Fig. 10.24 Illustration of

classification of crosstalk terms
T
relative to channel k
bk[i]
sk (current symbol)
k-th user
bm[i 1] bm [ i ]
sm (prev. sym.) sm (current sym.)
m-th user

mk km

bl[i] bl[i+1]
sl (current symbol) sl (next symbol)
l-th user

lk
kl

l k m
678 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

Let the elements of correlation matrices R[0] and R[1] be defined as [5]:
8
< 1, m ¼ k
> 
0, m  k
Rmk ½0 ¼ ρmk , m < k , Rmk ½1 ¼ ð10:107Þ
>
: ρkm , m < k
ρkm , m > k

We can now represent the K-equations from (10.106) in matrix form by:

r½i ¼ RT ½1Ab½i þ 1 þ RT ½0Ab½i þ R½1Ab½i  1 þ z½i: ð10:108Þ

The noise vector z[i] is a zero-mean Gaussian with covariance matrix:

8 2 T
>
> σ R ½1, j ¼ i þ 1
>
< σ 2 RT ½0, j ¼ i
hz½iz½ ji ¼ ð10:109Þ
> σ 2 RT ½1, j ¼ i  1
>
>
:
0, otherwise

Let us now apply the z-transform on Eq. (10.108) to obtain:

 
rðzÞ ¼ RT ½1z þ RT ½0 þ R½1z1 AbðzÞ þ zðzÞ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
SðzÞ ð10:110Þ
¼ SðzÞAbðzÞ þ zðzÞ:

Similar to synchronous case, let us factorize S(z) as follows [5, 33]:

 
SðzÞ ¼ FT z1 FðzÞ, FðzÞ ¼ F½0 þ F½1z1 , ð10:111Þ

where F[0] is lower triangular and F [1] is upper triangular with zero diagonal, such that [5, 33]:
 Z 1 
1   j2πf

detF½0 ¼ exp log detS e df : ð10:112Þ
2 0

Moreover, when detS(e jω) > 0, 8 ω 2 [π, π], then the filter with transfer function F1[z] is causal and stable [5, 33]. The
corresponding MIMO asynchronous CDMA model in the z-domain is shown in Fig. 10.25.
Let us now multiply the Eq. (10.110) by FT(z1), from the left side, to obtain:

r0 ðzÞ ¼ FT ðz1 ÞrðzÞ ¼ FT ðz1 ÞFT ðz1 ÞFðzÞAbðzÞ þ FT ðz1 ÞzðzÞ
|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl}
I wðzÞ ð10:113Þ
¼ FðzÞAbðzÞ þ wðzÞ:

w(z)

F T (z 1)

z(z)
b(z) A S(z) r(z)

Fig. 10.25 The asynchronous CDMA model in the z-domain

10.3 Multiuser Detection 679

z(z)

F T (z 1)

w(z)
b(z) A F(z) r’(z)

Fig. 10.26 Equivalent asynchronous CDMA model in the z-domain with MIMO whitening filter

It can be shown that the samples of noise process w(z) are uncorrelated, indicating that the filter with transfer function
FT(z1)is in fact MIMO whitening filter. The corresponding equivalent asynchronous CDMA model in z-domain with
MIMO whitening filter, based on Eq. (10.113), is shown in Fig. 10.26.
The time-domain representation of asynchronous CDMA system can be obtained by inverse z-transform of Eq. (10.113),
which yields to:

r0 ½i ¼ F½0Ab½i þ F½1Ab½i  1 þ w½i: ð10:114Þ

Finally, from Eqs. (10.110) and (10.111), we can establish the following relationships between R[0], R[1] and F[0], F[1]:

R½0 ¼ FT ½0F½0 þ FT ½1F½1, R½1 ¼ FT ½0F½1: ð10:115Þ

10.3 Multiuser Detection

The interference originating from multiple users should not be treated as a noise process. Instead, the knowledge of signature
waveforms should be exploited to mitigate multiaccess interference. In particular, when all users are detected simultaneously,
the interference from other users can be estimated and cancelled out. The CDMA receiver can also exploit the structure of
multiuser interference in detection of desired signal, which is commonly referred to as the multiuser detection. The multiuser
detection was pioneered by Sergio Verdú, in a series of publications [5, 35–42].
We start our discussion with conventional single-user detection, when other users are considered as interference channels,
which is illustrated in Fig. 10.23. This case is suboptimum, because the output of the k-th correlator (matched filter) does not
represent the sufficient statistics for making a decision on bit transmitted bk.

10.3.1 Conventional Single-User Correlator Receiver

Suppose that signature sequences in synchronous CDMA system are orthogonal to each other. In this case, the k-th user can be
operated independently on other users, and the optimum receiver ffiffiffi be conventional matched filter receiver, and the
pwill
probability of error for the k-th user will be P1,k ¼ 0:5erfc Ak =σ 2 , where erfc function is introduced in Chap. 4 as
pffiffiffi R 1
erfcðzÞ ¼ ð2= π Þ z exp ðu2 Þdu: When signature sequences are quasi-orthogonal or non-orthogonal, we need to determine
cross-correlations among different users and apply the CDMA model provided in Fig. 10.23. Even though this receiver is
suboptimum, it can be used for reference. Based on Eq. (10.92), we know that the output of the k-th matched filter can be
P
represented as r k ¼ Ak bk þ Km¼1,m6¼k Am bm ρmk þ zk , k ¼ 1, ⋯, K , and corresponding CDMA model for two-user case
(K ¼ 2) is simply:

r 1 ¼ A1 b1 þ A2 b2 ρ þ z1
ð10:116Þ
r 2 ¼ A2 b2 þ A1 b1 ρ þ z2 ,
680 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

f(r1|b1= 1, b2=1) f(r1|b1=1, b2= 1)

f(r1|b1= 1, b2= 1) f(r1|b1=1, b2=1)

r1

A1 A2 A1 A1+ A2 0 A1 A2 A1 A1+ A2

b1= 1 b1=1

Fig. 10.27 Probability density functions for user 1 bit, b1, by taking into account the bit originating from user 2, b2

where for simplicity we have written ρ12 ¼ ρ21 ¼ ρ. Given that noise samples are zero-mean Gaussian with the same variance
σ 2, we sketched in Fig. 10.27 the probability density functions (PDFs) for user 1 when transmitted bit b1 was either +1 or  1,
which takes into account the bit b2 transmitted by user 2. The probability of error for user 1 can be determined as:
n o
Pe,1 ¼ Pr bb1 6¼ b1
ð10:117Þ
¼ Prfr 1 < 0jb1 ¼ þ1gPrfb1 ¼ þ1g þ Prfr 1 > 0jb1 ¼ 1gPrfb1 ¼ 1g:

However, r1 conditioned only on b1 is not Gaussian, since it is also dependent on b2. So we need to condition on b2 as well,
so that the last term can be written as:
8 9
< =
Prfr 1 > 0jb1 ¼ 1g ¼ Pr r > 0 jb ¼ 1, b2 ¼ 1 Prfb2 ¼ 1g
: |fflffl1ffl{zfflfflffl} 1 ;
z1 >A1 þA2 ρ
8 9 ð10:118Þ
< =
þPr r 1 > 0 jb1 ¼ 1, b2 ¼ þ1 Prfb2 ¼ þ1g
: |fflfflffl{zfflfflffl} ;
z1 >A1 A2 ρ

Based on Fig. 10.27, we conclude that:

1 A1 þ A2 ρ
Prfz1 > A1 þ A2 ρjb1 ¼ 1, b2 ¼ 1g ¼ erfc pffiffiffi
2 σ 2

ð10:119Þ
1 A1  A2 ρ
Prfz1 > A1  A2 ρjb1 ¼ 1, b2 ¼ þ1g ¼ erfc pffiffiffi ,
2 σ 2

and after substitution in (10.118), assuming Pr{b2 ¼ 1} ¼ Pr{b2 ¼ +1} ¼ 1/2, we obtain:

1 A1 þ A2 ρ 1 A1  A2 ρ
Prfr 1 > 0jb1 ¼ 1g ¼ erfc pffiffiffi þ erfc pffiffiffi : ð10:120Þ
4 σ 2 4 σ 2

Because of symmetry as shown in Fig. 10.27, we can write Pr{r1 < 0| b1 ¼ + 1} ¼ Pr {r1 > 0| b1 ¼  1}, so that the
expression for probability of error for user 1 becomes:

1 A1 þ A2 ρ 1 A1  A2 ρ
Pe,1 ¼ erfc pffiffiffi þ erfc pffiffiffi , ð10:121Þ
4 σ 2 4 σ 2
10.3 Multiuser Detection 681

while based on Eq. (10.116), we conclude that probability of error for user 2 will be the same as probability of error for user
1, namely, Pe,2 ¼ Pe,1. Given that the first term in (10.121) cannot be smaller than the second term, the probability of error can
be upper-bounded by:

1 A  A jρj
Pe,1 ¼ Pe,2  erfc 1 pffiffi2ffi , ð10:122Þ
2 σ 2

indicating that probability of error will be smaller than ½ (Pe,1 < 1/2) when A1 > A2|ρ|, and this condition is known as the open
eye condition (when observed on the scope, the eye diagram will be open). In this case, the useful term is dominant over
crosstalk term, and error probability can be approximated as:

1 A  A jρj
Pe,1 ¼ Pe,2 ’ erfc 1 pffiffi2ffi ,
2 σ 2

and this expression is similar to single-user link, except that SNR is lower due to the presence of crosstalk. In opposite
situation, when A1 < A2|ρ|, the eye diagram will be completely closed, and the crosstalk will dominate over useful signal,
which reminds to the near-far crosstalk problem. Since the reception will be dominated by interference, the receiver 1 will
detect the bit transmitted by user 2, and probability of error will be close to ½. When A1 ¼ A2ρ, from Eq. (10.121), we obtain
the following expression for error probability:

1 2A1 1
Pe,1 ¼ erfc pffiffiffi þ , ð10:123Þ
4 σ 2 4

indicating that with probability ½ the interference dominates so that probability of error is ½ in that case, while with
probability of ½ the b2 doubles the amplitude for channel 1 resulting in improvement in performance. As an illustration of
error probability performance of conventional detector, in Fig. 10.28, we provide Pe versus signal-to-noise ratio (SNR) in
channel 1, wherein the ratio A2/A1 is used as a parameter for cross-correlation of ρ ¼ 0.25. As expected, when the crosstalk

Fig. 10.28 Probability of error in

channel 1, for conventional
single-user correlator -1
10
receiver, vs. SNR for different A2/
A1 ratios, assuming the cross-
correlation of 0.25
-2
10
Probability of error, Pe

-3
10

-4 =0.25:
10
A2/A1=4
A2/A1=2
-5
10 A2/A1=1
A2/A1=1/2
-6
10 A2/A1=0

-7
10
0 4 8 12 16 20
Signal-to-noise ratio, A1/ [dB]
682 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

dominates, it is not possible to detect the transmitted bits correctly. On the other hand, when useful signal dominates over
crosstalk, it is possible to detect the transmitted bits correctly, without any error floor phenomenon.
By using analogy, we can generalize two-user error probability calculation to K-user case, with details provided in [5], as
follows:

Pe,k ¼ Prfr k < 0jbk ¼ þ1gPrfbk ¼ þ1g þ Prfr k > 0jbk ¼ 1gPrfbk ¼ 1g
n X o n X o
1 1
¼ Pr zk < Ak  m6¼k
Am b m ρ mk þ Pr z k > A k  m6¼k
A m b m ρ mk
2n o 2
X
¼ Pr zk > Ak  A b ρ ð10:124Þ
m6¼k m m mk
P

1 X X X 1 Ak  m6¼k Am αm ρmk
¼ K1 ⋯ ⋯ erfc pffiffiffi ,
2 α1 2f1,þ1g αm 2f1,þ1g αK 2f1,þ1g 2 σ 2

where we performed conditioning over all possible interfering bits. Again, for the open eye condition, the following upper
bound can be used:
P

1 Ak  m6¼k Am jρmk j X
Pe,k  erfc pffiffiffi , Ak > A ρ : ð10:125Þ
2 σ 2 m6¼k m mk

For sufficient number of crosstalk terms, we can use the Gaussian approximation, according to central limit theorem, for
∑m 6¼ kAmρmk, and corresponding expression for error probability will be:
0 1

ee,k ’ 1 erfcB Ak C
P @qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P ffipffiffiffiA, ð10:126Þ
2 σ 2 þ m6¼k A2m ρ2mk 2

which gives more pessimistic results, in high SNR regime, as shown in [5].
The probability of error analysis for asynchronous CDMA is similar, except that we need to account for: (i) left and right
crosscorrelations and (ii) the fact that each bit is affected by 2 K-2 interfering bits. The resulting probability of error expression
will have increased number of terms twice [5]:
P

1 X X X 1 Ak  m6¼k Am ðαm ρmk þ βm ρkm Þ

Pe,k ¼ ⋯ ⋯ erfc p ffiffi
ffi : ð10:127Þ
4K1 2 σ 2
α1 , β1 2f1, þ1g2 αm , βm 2f1, þ1g2 2
αK , βK 2f1, þ1g

When the user k signal is dominant, the following open eye condition will be valid:
X
Ak > A ðjρmk j
m6¼k m
þ jρkm jÞ: ð10:128Þ

However, since the arrival times τk are random variables, corresponding cross-correlations ρkm and ρmk will be random
variables as well so that the open eye condition cannot be guaranteed even in the absence of noise. This represents a key
drawback of conventional single-user correlator (matched filter) receiver approach, and we move the focus to the optimum
multiuser detection.

10.3.2 Optimum Multiuser Detection

Prior to the mid-1980s, it was commonly believed that conventional single-user matched filter receiver was optimal or near-
optimal. The main argument was that the interference term in DT synchronous CDMA model is binomial for a finite number
of users K or Gaussian for infinitely a large number of users, according to the central limit theorem. However, as we know
from the detection theory [33], the sufficient statistic in multiuser detection for making decisions on (b1,. . .,bK) is (r1,..,rK), not
just rk, and the use of non-optimum receiver might result in significant performance degradation, in particular for asynchro-
nous CDMA in the presence of multipath fading.
10.3 Multiuser Detection 683

In our study on optimum multiuser detection for synchronous CDMA, we assume that the receiver has full knowledge
of signature waveforms of all users and received amplitudes Ak for all users, bit-timing information is known, and the variance
of the noise σ 2 is known as well. We apply DT synchronous CDMA model, given by Eq. (10.83), namely,
r(t) ¼ ∑kAkbksk(t) + σz(t).We start our study with a two-user case, in similar fashion as we have done before:

r ðt Þ ¼ A1 b1 s1 ðt Þ þ A2 b2 s2 ðt Þ þ σzðt Þ, t 2 ½0, T Þ: ð10:129Þ

From detection theory [33] (see also Chap. 6), we know that the maximum a posteriori probability (MAP) decision rule
minimizes the decision error probability, and the individually optimum decision rules can be formulated as:

b
b1 ¼ arg max Prfr 1 jr ðt Þg,
b1 2f1, þ1g
ð10:130Þ
b
b2 ¼ arg max Prfr 2 jr ðt Þg:
b2 2f1, þ1g

We are more interested in the jointly optimum decision rule, formulated by:
 
b1 , b
b b2 ¼ arg max Prfðb1 , b2 Þjr ðt Þg: ð10:131Þ
ðb1 , b2 Þ2f1, þ1g2

For sufficiently high SNR, we would expect these two decision rules to agree. When each of the possibilities for (b1,b2) is
equally likely, the MAP decision rule becomes the maximum likelihood (ML) decision rule:
 
bb1 , b
b2 ¼ arg max f ðr ðt ÞÞ j ðb1 , b2 Þ
ðb1 , b2 Þ

ZT
 1
2σ 2
½r ðt Þ  A1 b1 s1 ðt Þ  A2 b2 s2 ðt Þ2 dt
¼ arg max e 0
ðb1 , b2 Þ

ZT ð10:132Þ
1 2
¼ arg min 2 ½r ðt Þ  A1 b1 s1 ðt Þ  A2 b2 s2 ðt Þ dt
ðb1 , b2 Þ 2σ
0
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
d2E ½rðt Þ, A1 b1 s1 ðt ÞþA2 b2 s2 ðt Þ
1 2
¼ arg min d ½r ðt Þ, A1 b1 s1 ðt Þ þ A2 b2 s2 ðt Þ:
ðb1 , b2 Þ 2σ 2 E

Therefore, when the variance is not known, the Euclidean distance receiver is equivalent to ML decision rule for AWGN,
which is consistent with Chap. 6. Instead of using waveforms, it will be much more convenient to work with matched filter
(MF) outputs r1 and r2, representing projections of received signal r(t) onto signatures s1(t) and s2(t), as illustrated in
Fig. 10.23. The corresponding ML decision rule will be now:
 
b
b1 , b
b2 ¼ arg max Prfr 1 , r 2 jðb1 , b2 Þg
ðb1 , b2 Þ
T
C1 ðrmÞ
¼ arg max e2σ2 ðrmÞ
1
ð10:133Þ
ðb1 , b2 Þ

¼ arg min ðr  mÞT C1 ðr  mÞ,

ðb1 , b2 Þ

where r ¼ [r1 r2]T, m ¼ [A1b1 + A2b2ρ A2b2 + A1b1ρ]T and C is the covariance matrix:
 
1 ρ
C ¼ Covðr 1 , r 2 Þ ¼ σ 2
:
ρ 1
684 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

The performance of such ML decision rule can be improved if we project to orthogonal basis functions {ϕ1(t), ϕ2(t)},
instead of non-orthogonal signal set {s1(t), s2(t)}, obtained by Gram-Schmidt procedure:

ϕ1 ðt Þ ¼ s2 ðt Þ,
s ðt Þ  ρs1 ðt Þ ð10:134Þ
ϕ2 ðt Þ ¼ 2 pffiffiffiffiffiffiffiffiffiffiffiffiffi :
1  ρ2

These new projections, denoted as ðer 1 ,er 2 Þ, will have the mean and covariance as follows:
   
A 1 b1 þ A 1 b1 ρ 1 0
e=
m e
pffiffiffiffiffiffiffiffiffiffiffiffiffi , C ¼ Covðer 1 ,er 2 Þ ¼ σ 2
: ð10:135Þ
A 2 b2 1  ρ2 0 1

Corresponding ML decision rule becomes:

 
b1 , b
b e 1 ðer  m
e ÞT C
b2 ¼ arg min ðer  m e Þ: ð10:136Þ
ðb1 , b2 Þ

Since the covariance matrix is diagonal, we can simplify the decision rule to:
 
bb1 , b
b2 ¼ arg min ker  m
e k2 , ð10:137Þ
ðb1 , b2 Þ

and with this decision rule, we can split the signal space into decision regions, similarly in K-hypothesis decision theory.
The following bounds for error probability for ML decision for two-user channel have been derived in [5]:
8 0qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi19
<
A21 þ A22  2A1 A2 jρj =
1 A1 1 @ A  Pe
max erfc pffiffiffi , erfc pffiffiffi
:2 σ 2 4 σ 2 ;
0qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1 ð10:138Þ

A21 þ A22  2A1 A2 jρj
1 A 1
Pe  erfc p1ffiffiffi þ erfc@ pffiffiffi A:
2 σ 2 4 σ 2

In Fig. 10.29, we plotted these probabilities of error ML bounds and compared against conventional receiver, assuming A2/
A1 ¼ 1.5 and ρ ¼ 0.45. First of all we notice that these ML bounds agree very well. Secondly, the ML receiver significantly
outperforms the conventional receiver.
The ML decision rule (10.137) can further be written as:

    pffiffiffiffiffiffiffiffiffiffiffiffiffi2 
b b 2
b1 , b2 ¼ arg min ðer 1  A1 b1  A2 b2 ρÞ þ er 2  A2 b2 1  ρ2
ðb1 , b2 Þ
 pffiffiffiffiffiffiffiffiffiffiffiffiffi  ð10:139Þ
¼ arg max er 1 A1 b1 þ er 1 A2 b2 ρ þ er 2 A2 b2 1  ρ2  A1 b1 A2 b2 ρ :
ðb1 , b2 Þ

From (10.134), we have the following relationship between new and old coordinates:
pffiffiffiffiffiffiffiffiffiffiffiffiffi
er 1 ¼ r 1 ,er 2 ¼ ðr 2  ρr 1 Þ= 1  ρ2 , ð10:140Þ

and after substitution we obtained the following ML decision rule:

 
bb1 , b
b2 ¼ arg max ðr 1 A1 b1 þ r 2 A2 b2  A1 b1 A2 b2 ρÞ: ð10:141Þ
ðb1 , b2 Þ

It has been shown in [5] that this expression can be expressed equivalently as:
10.3 Multiuser Detection 685

Fig. 10.29 Probability of error in

a two-user channel for ML against
conventional receiver for A2/ 10
-1
A1 ¼ 1.5 and ρ ¼ 0.25

-2
10

Probability of error, Pe
-3
10

-4
10

-5
10
=0.45, A2/A1=1.5:
-6 Conventional
10
ML lower-bound
ML upper-bound
-7
10
0 2 4 6 8 10 12 14 16
Signal-to-noise ratio, A1/ [dB]

T r1 r1A1 b̂1
· dt
0 t=T
r(t) Decision
A1 circuit
s1(t) ||
T r2
r2A2
· dt A1A2
0 t=T

s2(t) A2 1/2
||

Fig. 10.30 Implementation of jointly optimum multiuser detector for user 1 in a two-user system

 
b 1 1
b1 ¼ sgn r 1 A1 þ jr 2 A2  A1 A2 ρj  jr 2 A2 þ A1 A2 ρj
2 2 ð10:142Þ
 
b 1 1
b2 ¼ sgn r 2 A2 þ jr 1 A1  A1 A2 ρj  jr 1 A1 þ A1 A2 ρj ,
2 2

and corresponding implementation for decision on b1 is shown in Fig. 10.30.

Clearly, the jointly optimum multiuser detector requires sufficient statistics (r1, r2), amplitudes A1 and A2, as well as cross-
correlation ρ.
In this rest of this section, we describe the jointly optimum detector for K-user synchronous CDMA case, described by
Eq. (10.83), r(t) ¼ ∑kAkbksk(t) + σz(t). The aim is to choose the information vector b ¼ [b1,. . .,bK] maximizing the likelihood
function f(r(t)|b), namely:

b
b ¼ arg max f ðrðt ÞjbÞ: ð10:143Þ
686 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

Assuming that all 2K possibilities are equally likely, the MAP detector becomes ML one. Similar to Eq. (10.132), the ML
decision rule for K-user case can be written as:

ZT
 12
2σ
½r ðt Þ  A1 b1 s1 ðt Þ  A2 b2 s2 ðt Þ  ⋯AK bK sK ðt Þ2 dt
b
b ¼ arg max e 0
b
ZT " X
K
#2
¼ arg min r ðt Þ  Ak bk sk ðt Þ dt ð10:144Þ
b k¼1
0
X
K K X
X K
¼ arg max 2 A k bk  Ak bk Am bm ρkm :
b k¼1 k¼1 m¼1
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
LK ðbÞ

The optimization in (10.144) is a combinatorial optimization problem, with details provided in [5]. By introducing the
matrix notation, already used in Sect. 10.2.2.2, namely:
2 3 2 3 2 3
r1 b1 1 ρ12 ⋯ ρ1K
6r 7 6b 7 6ρ ⋯ ρ2K 7
6 2 7 6 27 6 21 1 7
r ¼ 6 7 ¼ RA b þ z, b ¼ 6 7, A ¼ diagðA1 , ⋯, AK Þ, R ¼ 6 7
4⋮5 4⋮ 5 4⋮ ⋮ ⋱ ⋮ 5
rK bK ρK1 ρ2K ⋯ 1

we can rewrite the last line in Eq. (10.144) in matrix form as follows:

b
b ¼ arg max 2bT Ar  bT ðARAÞb, ð10:145Þ
b

where ARA is the un-normalized correlation matrix.

Additional details on optimum multiuser detection of asynchronous CDMA as well as noncoherent CDMA can be found
in [5].

10.3.3 Decorrelating and Linear MMSE Multiuser Detectors

By going back to the matrix representation of DT synchronous model, given by Eq. (10.93), namely:
 
r ¼ R A b þ z, hzi ¼ 0, CovðzÞ ¼ z zT ¼ σ 2 R, ð10:146Þ

Assuming that covariance matrix R is invertible, we can multiply (10.146) by R1 from the left side to obtain:

R1 r ¼ |ffl
Rffl1
{zfflR
1 1
ffl} A b þ R z = A b þ R z, ð10:147Þ
I

we conclude that multiuser interference is compensated for, and the decision can be made by simply looking for the signs of
R1r as follows:
 
b
bk = sign R1 r k ; k ¼ 1, ⋯, K; ð10:148Þ

where we used the notation ()k to denote the k-th element of the vector. Clearly, in the absence of noise, b
bk = signðAbÞk ¼
bk , Ak > 0, providing that amplitudes are non-negative. In the presence of noise, from equalization (Chap. 6) and spatial
10.3 Multiuser Detection 687

Fig. 10.31 Decorrelating

detector for synchronous CDMA
systems
T r1 b̂1
· dt
0 t=T

…
…

…
s1(t)

…
r(t) T
rk bˆk
· dt R 1
0 t=T

…
…

…
sk(t)

…
T rK bˆK
· dt
0 t=T

Bank of correlators
sK(t) (matched fitters) Bank of decision
(threshold) circuits

interference cancelation in MIMO systems (Chap. 8), we know that the noise process described by R1z gets enhanced.
The configuration of decorrelating detector for synchronous CDMA system is provided in Fig. 10.31. The covariance matrix
inversion can be computationally intensive for a large number of users. Two interesting properties of decorrelating receiver
are as follows: (i) the knowledge of amplitudes Ak is not needed, as evident from (10.148), and (ii) the demodulation process
can be parallelized by operating with transformed set of basis functions in bank of decelerators (matched filters). Namely, the
k-th element of R1r can be rewritten as:
* +
 1  X
K   X
K  
R r k¼ R1 km r m ¼ R 1
s jr
km m
|{z}
m¼1 m¼1
hsm jri |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
esk ð10:149Þ
X
K  
¼ hesk jr i,esk ¼ R1 s :
km m
m¼1

Clearly by using the set fesk g instead of {sk} to implement the bank of correlators, we can greatly simplify the receiver
architecture, as shown in Fig. 10.32. As long as the set of signals {sk} is linearly independent, the decorrelating set is a unique
one. The new signal set has a unit dot product with the original signal set since:

ZT X
K    
hesk jsk i ¼ R1 s ðt Þsk ðt Þdt
km m
¼ R1 R kk ¼ 1: ð10:150Þ
m¼1
0

However, when the cross-correlations are not known a priori, due to channel distortion, we would need to generate them
locally on receiver side, as well as to calculate elements of R1.
The decorrelating linear transformation can be related to the whitening matched filter approach. Namely, we can perform
the Cholesky factorization R ¼ FTF, and apply approach similar to Eq. (10.96), with additional details provided in [5].
We mentioned earlier that the decorrelation receiver does not require the knowledge of amplitudes. Assuming equal
probable transmission, the ML decision will be:
688 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

Fig. 10.32 Modified correlator

bank-based decorrelating detector
for synchronous CDMA systems T b̂1
· dt
0 t=T

…
…

…
s1 t

…
r(t) T bˆk
· dt
0 t=T

…
…

…
sk t
…
T bˆK
· dt
0 t=T

Modified bank of
sK t correlators Bank of decision
(threshold) circuits
 
b b
b, A ¼ arg max f ðrjb, AÞ
ML ðb, AÞ

¼ arg min kr  RAbk2 ¼ arg min ðr  RAbÞT ðr  RAbÞ

ðb, AÞ ðb, AÞ
ð10:151Þ
¼ arg max ðRAbÞ r þ r ðRAbÞ  ðRAbÞT RAb
T T
ðb, AÞ

= arg max ðRAbÞT ðr  RAbÞ þ rT ðRAbÞ,

ðb, AÞ

and the maximum is obtained by setting r  RAb = 0 or equivalently (Ab)optimum = R1r. Therefore, the decision rule given
by Eq. (10.148) is optimum when no information about amplitudes {Ak} is available.
As indicated above, one of the main drawbacks of the decorrelating detector is the noise enhancement problem due to
transformation R1z. To avoid this problem, we will determine the transformation (“weight”) matrix W applied on the
received vector r on such a way to minimize the mean squared error (MSE). We are in particular interested in linear (or affine)
transformations. The corresponding multiuser detection is commonly referred to as the linear minimum MSE (MMSE)
multiuser detection. Instead of minimizing the MSE related to the k-th user, namely,
D E  
min bk  e
bk ¼ min bk  wTk r ,
wk wk

we will minimize the sum of these MSE expectations simultaneously:

10.3 Multiuser Detection 689

* +
X
K   X
K  
min bk  wTk r ¼ min bk  wTk r
f wk g f wk g
k¼1 k¼1
2 3
wT1
6 T7 ð10:152Þ
D E 6w 7
6 27
2
¼ min kb  Wrk , W ¼ 6 7:
W 6⋮ 7
4 5
wTK

We again use the matrix representation of DT synchronous model, given by (10.146), and invoke the orthogonality
principle from detection theory [33], claiming that error vector is independent on the data vector; see also Sect. 6.3.4 in
Chap. 6, so that we can write:
     
ðb  WrÞrT ¼ 0 , brT ¼ W rrT : ð10:153Þ

Form one side, the expectation of brT can be represented as:

 T          
br ¼ bðRAb þ zÞT ¼ bðRAbÞT þ bzT ¼ bbT AR ¼ bbT AR = AR: ð10:154Þ
|fflffl{zfflffl} |fflffl{zfflffl}
0 I

From the other side, we can write:

   
W rrT ¼ W ðRAb þ zÞðRAb þ zÞT
       
¼ W RAbbT AR þ W RAbzT þ W zðRAbÞT þ W zzT
|fflfflfflfflffl{zfflfflfflfflffl} |fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} |ffl{zffl}
0 0 σ2 R
  ð10:155Þ
¼ WRA bbT AR þ Wσ 2 R
|fflffl{zfflffl}
I
 
= W RA2 R þ σ 2 R :

Since (10.153) and (10.155) represent the left and right sides of Eq. (10.153), by equating them, we obtain:
 
AR = W RA2 R þ σ 2 R , ð10:156Þ

and by multiplying (10.156) by R1 from the right, we arrive at:

 
A = W RA2 þ σ 2 I : ð10:157Þ

By solving for W, we obtain the optimum solution as:

 1
W optimum ¼ A RA2 þ σ 2 I , ð10:158Þ

and the optimal joint linear MMSE multiuser detector performs the following operations:

 1
e
b ¼ W optimum r ¼ A RA2 þ σ 2 I r: ð10:159Þ

The k-th user decision is given by:

nh  1 i o
b
bk ¼ sign A RA2 þ σ 2 I r ; k ¼ 1, ⋯, K: ð10:160Þ
k
690 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

Fig. 10.33 The linear MMSE

multiuser detector for
synchronous CDMA systems
T r1 b̂1
· dt
0 t=T

…
…

…
s1(t)

…
r(t) T
rk bˆk
· dt [R+ A 2] 1
0 t=T

…
…

…
sk(t)

…
T rK bˆK
· dt
0 t=T

Bank of correlators
sK(t) (matched fitters) Bank of decision
(threshold) circuits

The sign of b bk will not be affected if we factor out matrix A ¼ diag(A1,. . .,AK), since attenuation coefficients are
nonnegative, so that the final decision for the k-th user is made according to:
nh 1 i o
b
bk ¼ sign R þ σ 2 A2 r ; k ¼ 1, ⋯, K: ð10:161Þ
k

The corresponding linear MMSE multiuser detector is provided in Fig. 10.33. Compared to decorrelating receiver from
Fig. 10.31, we replaced linear transformation R1 of decorrelating detector by (R + σ 2A2)1, wherein
 
σ2 σ2 σ2
σ 2 A2 ¼ diag , ⋯, , ⋯ ,
A21 A2k A2K

and we do not require anymore the covariance matrix to be nonsingular. When the variance tends to zero, the decision
reduces to:
n o
b
bk ! sign R1 r k ,
σ 2 !0

and we can say that MMSE detector becomes decorrelating detector, when noise variance tends to be zero (or equivalently
SNR tends to be plus infinity).
The linear MMSE multiuser detector has certain similarities with linear MMSE interface described in Sect. 8.2.6
(in Chap. 8). By denoting RA as H, our DT synchronous CDMA model becomes MIMO model r ¼ Hb + z, wherein the
channel matrix is quadratic and of full rank. Therefore, different MIMO techniques, but of lower complexity, are applicable
here as well. Moreover, the iterative interface described in Sect. 8.2.7.4, in which spatial interference is cancelled in an
iterative fashion is also applicable here.

10.3.4 Decision-Driven Multiuser Detectors

Nonlinear multiuser detection scheme can employ decisions on bits from other users, to estimate the interference and remove
it from the received signal [5, 43–53]. Both tentatives of final decision on other users’ bits may be employed. These decision-
driven multiuser detectors can be classified into several categories [5, 43–53]: (i) successive cancellation Schemes [45–47],
(ii) multistage detection [48, 51], (iii) continuous-time tentative decoding [5], and (iv) decision-feedback multiuser detection
Schemes [52, 53]. The first two schemes will be briefly described here.
10.3 Multiuser Detection 691

Fig. 10.34 Two-user b̂1

r̂ t T
r1
synchronous successive · dt
cancellation scheme, employing 0 t=T
conventional correlators s1(t)
A2

r(t) T
r2
· dt
0 t=T b̂2

s2(t)

T
r1
b̂1
· dt
0 t=T
s1(t)
r(t) T

· dt A2
0 t=T
s2(t)
T
r2
· dt
0 t=T b̂2

Fig. 10.35 An alternative implementation of two-user successive cancellation scheme

In successive cancellation scheme, the symbol of a given user is detected by observing the signals for other users as noise.
Such obtained decision is then used to re-modulate the signal from that user and subtract it from the received signal. This
procedure repeated in an iterative fashion until the transmitted bits from all users are demodulated. Before the successive
cancelation takes place, the users
hare arranged iwith  the decreased average energies at the outputs of matched filters, for
RT 2 P
synchronous CDMA defined as 0 r ðt Þsk ðt Þdt ¼ σ 2 þ A2k þ m6¼k A2m ρ2mk : However, this is not the best strategy since
the cross-correlation is not taken into account. As an illustration, let us observe two-user synchronous successive cancellation
  conventional correlators (matched filters), with corresponding scheme being provided in Fig. 10.34.
scheme, employing
Assuming that r 22 is higher, we first make decision on b2 by:

b
b2 ¼ signfr 2 g: ð10:162Þ

We used this decision to estimate the crosstalk originating from user 2 by:

r ðt Þ  A 2 b
b2 s2 ðt Þ ¼ br ðt Þ, ð10:163Þ

and after subtracting the crosstalk from user 2, we perform match filtering for user 1 and perform decision by:
8 T 9
<Z =
^b1 ¼ sign ½rðtÞ  A2 ^b2 s2 ðtÞs1 ðtÞdt
: ;
0
0 1 ð10:164Þ
B C
¼ sign@ r1 b2 ρA ¼ signðA1 b1 þ ρA2 ðb2  ^
 A2 ^ b2 Þ þ σhs1 jziÞ,
|{z} |{z}
A1 b1 þρA2 b2 þσhs1 jzi signðr 2 Þ

with corresponding implementation being shown in Fig. 10.35.

692 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

Fig. 10.36 A two-user

synchronous two-stage detection
T
r1
b̂1
scheme, with the first stage being · dt
conventional receiver
0 t=T
s1(t)
r(t) T
A1
· dt
A2
0 t=T
s2(t)
T b̂2
· dt
r2
0 t=T
Successive cancelation
Bank of correlators stage
(matched fitters)

Fig. 10.37 A two-user T b̂1

r1
synchronous two-stage detection · dt
scheme, with the first stage being 0 t=T
d̂1
the decorrelating detector s1(t)
r(t) T
A1
· dt
0 t=T A2

s2(t)
d̂ 2

T b̂2
· dt r2
0 t=T
Successive cancelation
Decorellating detector stage

In the K-user successive cancelation scheme, assuming that decisions on bits related to the users k + 1,. . .,K are correct, the
decision on the bit related to the k-th user, while ignoring the presence of other users, will be:
( )
X
K
b
bk ¼ sign r k  Am ρmk b
bm : ð10:165Þ
m¼kþ1

In the K-user successive cancelation scheme, the decorrelating detectors can be used as well.
In multistage detection scheme, the various decisions get generated in different stages. In one illustrative two-user example,
shown in Fig. 10.36, the first stage is the conventional bank of correlators (matched filters), followed by the successive
cancelation stage for both users.
The successive cancelation stage, based on Eq. (10.165), can be repeated as many times as needed. In another interesting
example, shown in Fig. 10.37, the first stage is the decorrelating detector, while the second stage is the successive cancelation
stage for both users. For the K-user case, the outputs of decorrelation stage are given by:
n o
b
dk ¼ sign R1 r : ð10:166Þ
k

These decisions are then used as tentative decisions used in the successive cancelation stage, operating as described by
Eq. (10.165), so that the corresponding decisions in successive cancelation will be made now as:
( )
X
K
b
bk ¼ sign r k  b
Am ρmk d m : ð10:167Þ
m¼kþ1

For two-user case, the corresponding decisions are made based on:

b
b1 ¼ signfr 1  A2 ρ signðr 2  ρr 1 Þg,
ð10:168Þ
b
b2 ¼ signfr 2  A1 ρ signðr 1  ρr 2 Þg:
10.4 Optical CDMA Systems 693

10.4 Optical CDMA Systems

In this section, we briefly describe various optical CDMA Schemes [54–75]. Similar to wireless communications, optical
CDMA allows users to share the same bandwidth at the same time using the unique codes or signature sequences. Some
unique advantages of optical CDMA include the following [66–69]: (i) privacy and inherent security in transmission,
(ii) simplified network control (no centralized network control is required), (iii) scalability of network, (iv) multimedia
applications support, and (v) providing provision for quality of service (QoS). Optical CDMA can be either direct detection-
based [56–64] or coherent optical detection-based [65]. The basic intensity modulation/direct detection (IM/DD) optical
CDMA scheme, suitable for LAN applications, is provided in Fig. 10.38.
The optical source could be either laser diode or LED. The light beam gets modulated by electro-optical modulator, which
can be either electro-absorption (EA) or Mach-Zehnder modulator (MZM). Different users get assigned with unique signature
codes, typically referred to optical orthogonal codes (OOCs), which are strictly speaking quasi-orthogonal. The OOCs must
satisfy the following two constraints [56–58, 66, 68, 69]: (i) each sequence (code) must be easily distinguishable from a
shifted version of itself and (ii) each code must be easily be distinguished from any other sequence in the OOC family (and
possibly corresponding shifted version as well). The OOC can be defined in either time-domain, typically called time-
spreading encoding, or in spectral (wavelength)-domain, commonly referred to as spectral amplitude coding (SAC).
Moreover, the frequency hopping (coding) can also be employed, in similar fashion as already employed in wireless
FH-CDMA systems. The scheme shown in Fig. 10.38 is suitable for both time-spreading encoding, typically based on
pulse laser, and SAC, wherein the optical source is an LED. The corresponding OOCs are imposed optically in either time-
domain or spectral-domain. Independent signals are after time-domain/spectral-domain encoding combined together with the
help of N
N optical coupler, such as optical star coupler. On the receiver side, the signal for the intended user is decoded and
converted into electrical domain by photodetector.
The optical orthogonal codes, denoted as OOC(v,k,λa,λc), are unipolar sequences of length v, weight k, autocorrelation
constraint λa, and crosscorrelation constraint λc [56, 58, 63, 66–69]. For unipolar sequences fxn gv1 v1
n¼0 and fyn gn¼0 , the
autocorrelation and cross-correlation constraints are defined as [56, 66]:

• Autocorrelation constraint:

X
v1
xn xnþm  λa ; m ¼ 0, 1, ⋯, v  1 ð10:169Þ
n¼0

• Cross-correlation constraint:

X
v1
xn ynþm  λc ; m ¼ 0, 1, ⋯, v  1: ð10:170Þ
n¼0

Fig. 10.38 Optical CDMA- Optical Optical

based LAN Encoder 1 Decoder 1 Photodetector
source Modulator

Data N N optical
…
…

user 1 coupler

Optical Optical Encoder N Decoder N Photodetector

source modulator

Data
user N
694 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

The strict orthogonality would mean that λa ¼ λc ¼ 0; however, when sequences are unipolar, the cardinality of set of
unipolar sequences will be too small. In practice, we typically require λa ¼ λc ¼ 1. There are different ways how to design
OOCs including algebraic constructions [58, 61], combinatorial constructions [66, 69], and finite geometries [68], to mention
a few. The weight of codewords in OOC do not need to be of fixed weight; the variable weight can be used as a way to provide
different QoS values for different user types [71, 72]. To improve the cardinality of OOCs, the two-dimensional (2-D)
wavelength-time OOCs have been advocated in [70, 73, 74]. The 2-D wavelength-time OOCs (m
n, k, λa,λc) are 2-D
unipolar sequences, described as m
n arrays, with weight k, and autocorrelation and cross-correlation constraints being λa
and λc. For 2-D time-wavelength OOCs {xi, j} and {yij}, the autocorrelation and cross-correlation constraints are defined,
respectively, as [70, 73, 74]:

• Autocorrelation constraint:

X
m1 X
n1 
k, l¼0
xi,j xi,ð jþlÞmod n ¼ ð10:171Þ
i¼0 j¼0  λa , 1  l  n  1

• Cross-correlation constraint:

X
m1 X
n1
xi,j yi,ð jþlÞmod n  λc , 0  l  n  1 ð10:172Þ
i¼0 j¼0

The incoherent optical CDMA systems can be categorized into three categories: time-spreading encoding, spectral-
amplitude-coding, and wavelength-hopping optical CDMA systems. The generic time-spreading encoding scheme, together
with the configuration of encoded/decoder, is provided in Fig. 10.39. Clearly, the pulse laser is needed with pulse duration
corresponding to the chip duration. The pulse laser is directly modulated by data sequence to be transmitted. The optical
encoder is based on 1-D OOC sequence and composed of 1:n power splitter (star coupler), n optical delay lines (ODLs), and
n:1 power combiner (gain optical star). The ODLs are properly chosen so that the pulse appears on chip positions according to
the OOC sequence, and after power combining, a desired unipolar sequence gets generated. Decoder operates with
complementary ODLs so that pulses get aligned when encoder and decoders are serially connected.
The spectral-amplitude coding employs a similar encoding principle, but in spectral (wavelength) domain, as illustrated in
Fig. 10.40. The SACs are typically synchronous schemes, with OOCs characterized by in-phase cross-correlation. For two
unipolar sequences {a} and {b}, the in-phase cross-correlation is defined by Ra,b ¼ Σn anbn. The balanced optical detector
shown in Fig. 10.40 has been used to deal with intensity fluctuations introduced by broadband source, and for this scheme,
typically the OOCs with fixed in-phase cross-correlations are used. The splitting ratio 1:a is properly chosen to maximize the
eye opening upon balanced optical detector, described in Chap. 3. The broadband source spectrum is sliced into n frequency
bands, and nonzero positions in OOC determine which frequency bands are used. We denoted this encoding process by A( f ).
The spectrum complementary to A( f ) is denoted by Â( f ). The fiber Bragg grating (FBG) represents an ideal device to perform
the spectral amplitude encoding.

Fig. 10.39 Generic time- Pre-amp

spreading scheme together with a
configuration of encoder/decoder Data Pulsed Optical Optical Optical
Photodiode
source laser encoder network decoder

...
Encoder/decoder:
...

...
...
10.4 Optical CDMA Systems 695

SAC
Spectral decoder
amplitude 1
encoder A(f)
Intensity Star 1:a Balanced Threshold
LED A(f) detector detector
modulator coupler coupler
Â(f)
a
Data
source

Fig. 10.40 Generic spectral amplitude coding (SAC) scheme with balanced optical detector

address

ODLn
...

...
ODL1

SOA1
...

...

AWG
SOAn
Broadband
source

Fig. 10.41 Reconfigurable SAC encoder/decoder based on AWG and semiconductor optical amplifiers (SOAs)

In particular, for SAC decoding process, the reflected component from FBG will represent the decoded signal, while the
transparent will perform the complementary function Â( f ). It is also possible to implement SAC schemes with non-fixed
in-phase cross-correlations [71]; however, SAC decoder complexity is higher. Moreover, the SAC schemes for multimedia
applications require the use of constant-length variable-weight OOCs. Instead of FBGs, arrayed waveguide gratings (AWGs)
can be used in SAC Schemes [72], as illustrated in Fig. 10.41. The signal from broadband source is demultiplexed into
different wavelength channels λ1,. . .,λn. The active wavelengths are determined by the OOC, and only those semiconductor
optical amplifiers (SOAs) are turned on. Different ODLs are used to ensure that all wavelength channels experience the same
delay. Clearly, this SAC encoder/decoder is reconfigurable. Moreover, ODLs and SOAs can be used to implement 2-D
wavelength-time encoders/decoders.
The wavelength hopping (WH) CDMA schemes, such as the one shown in Fig. 10.42, are conceptually very similar to
frequency hopping (FH)-CDMA schemes used in wireless communications and are described above. Therefore, the same
spreading sequences are applicable. The pulse from broadband source is narrow and will generate a narrow impulse response
from Bragg grating. Identical Bragg gratings are written to the same wavelength and are tuned to the desired wavelength [see
Fig. 10.42a]. The target wavelength can be changed in either slow or fast fashion so that we can also define slow and fast
WH. Either one or multiple wavelengths can be activated during the symbol duration. At the receiver side, we need to tune the
wavelengths in opposite order to the WH encoder, to ensure that wavelengths are properly aligned at the output of WH
decoder.
Regarding the coherent optical CDMA systems, they can be classified into two broad categories [65]: (i) pulse laser-based
and (ii) CW laser-based. The pulse laser-based coherent optical CDMA systems, shown in Fig. 10.43a, are conceptually
similar to the time-spreading incoherent schemes.
Namely, the incoming pulse from the laser is split into n branches by an 1:n power splitter. Each branch contains an ODL
and a phase shifter (PS). The ODLs are used to choose the proper locations of chips, while phase shifters are used to introduce
696 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

Impulse response of a grating Chip duration

Broadband Intensity
source modulator

Identical Bragg gratings written to the same

Data wavelength and tuned to the different wavelengths
source
WH encoder
(a)
WH decoder
Received
signal

Gratings are tuned to inverse order of encoder

wavelengths
Desired
signal
MAI

Threshold
detector
(b)
Fig. 10.42 Wavelength hopping (WH) CDMA schemes based on FBGs

the appropriate phase shifts. Various carrier synchronization algorithms described in Chap. 6 can be implemented in DSP.
Another alternative scheme based on optical PLL is described in [65]. On the other hand, the CW laser-based coherent optical
scheme, shown in Fig. 10.43b, is similar to DS-CDMA schemes from wireless communications. After the spread spectrum
signal is generated by DS approach, such signal is converted into optical domain by electro-optical (E/O) modulator. For
BPSK, either MZM or optical phase modulator can be used, while for QPSK and higher-order modulation schemes, the E/O
I/Q modulator should be used. On the receiver side, we perform coherent optical balanced detection as described in Chap. 3,
and the remaining steps are the same as in DS wireless CDMA systems, including the dispreading and demodulation.
The readers interested in optical CDMA are referred to reach literature, including [54–75], in particular to books
specialized in optical CDMA, such as [55].

10.5 Hybrid OFDM-CDMA Systems

The hybrid OFDM-CDMA systems [76–80] are trying to take advantages of both OFDM and CDMA concepts, particularly in
dealing with frequency selective fading and narrowband interference, while at the same time supporting multiple users. They
can be categorized into three broad classes: (i) multicarrier CDMA (MC-CDMA), (ii) OFDM-CDMA, and (iii) OFDM-
CDMA-FH systems.
In multicarrier CDMA [79, 81–83], on the transmitter side, shown in Fig. 10.44a, the data sequence is coming in serial
fashion, in blocks of length P each, and after serial-to-parallel (S/P) conversion, each bit (symbol) of the k-th user, say bkp, is
multiplied by a spreading sequence of the k-th user [sk1 ... skN] to obtain bkp[sk1 ... skN]. Such frequency spread [bkp sk1 ... bkp
skN] data is then used as the input of inverse FFT (IFFT) block of OFDM subsystem. Clearly, the required number of
subcarriers is NP. The remaining steps on the transmitter side are the same as described in Chap. 8 (this is why full details of
OFDM transmitter are not provided here). Clearly, the spreading is applied here in frequency-domain rather than time-domain
in DS-CDMA systems. Because of that, each user benefits from frequency diversity since the same user bit (symbol) gets
transmitted over multiple subcarriers. The transmitted signal of the k-th user (k ¼ 1,. . .,K ) can be represented as:
10.5 Hybrid OFDM-CDMA Systems 697

Data Pulse
source local
laser
Pre-amp

Pulse E/O Optical Optical Optical Coherent optical

laser modulator encoder network decoder balanced detector

Encoder/decoder: ODLn PSn 0 0

...
ODLi PSl

...
ODL1 PS1

(a)
CW
Data Spreading Despreading
source
local
sequence sequence
laser
Pre-amp

CW E/O Optical Coherent optical

laser modulator network Demodulator
balanced detector

(b)
Fig. 10.43 Coherent optical detection CDMA systems: (a) pulse laser-based and (b) CW laser-based schemes

ðMCCDMAÞ
1 X
X P1 X
N 1
j2πðnPþpÞΔf ðtiT s Þ
sk ðtÞ ¼ bkp skn pðt  iT s Þe ,
i¼1 p¼0 n¼0
( ð10:173Þ
1,  T G  t < T s  T G
pðtÞ ¼
0, otherwise

where Δf is the subcarrier spacing, determined by Δf ¼ 1/(Ts  TG), with Ts being OFDM symbol duration related to original
symbol period by Ts ¼ PT, while TG is the guard interval duration. Clearly, the bandwidth occupied by K-user MC-CDMA
signal is:
BMCCDMA ¼ ðNP  1Þ=ðT s  T G Þ þ 2=T s : ð10:174Þ

The received signal can be represented as:

1 X
X P1 X
N 1 X
K
r ðMC‐CDMAÞ ðt Þ ¼ H ðnpkÞ bkp skn pðt  iT s Þe j2π ðnPþpÞΔf ðtiT s Þ
þ zðt Þ, ð10:175Þ
i¼1 p¼0 n¼0 k¼1

where H ðnpkÞ is the channel coefficient corresponding to the (nP + p)-th subcarrier of the k-th user and z(t) is the noise process.
The configuration of the k-th user receiver is shown in Fig. 10.44b. After the down-conversion and analog-to-digital
698 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

Data stream,
User 1 MC-CDMA transmitter
user 1

User k MC-CDMA transmitter

...

...
...
sk1

bk1
sk2

...
...
Serial-to- Guard
...

Data stream, parallel bkp IFFT Parallel-to-

interval
serial DAC
user k conversion block insertion
...

(1:P) bkP ... skN

conversion
(TG)

...
cos( ct)
...

...
...

Data stream,
User K MC-CDMA transmitter
user K

(a)

User k MC-CDMA receiver

Despreading
...

w1sk1

bˆk 1
w2sk2
...
...

bˆkp
...

Estimated data, Guard

P/S FFT S/P
interval ADC
user k conversion block conversion
...

removal
w2skN
...

...

bˆkP cos( ct)

...

(b)
Fig. 10.44 Generic K-user MC-CDMA system: (a) transmitter configuration and (b) k-th user receiver configuration

conversion (DAC), we perform timing recovery and frequency synchronization (not shown in the figure), followed by cyclic
extension removal. We further perform S/P conversion, followed by FFT. The next step is to perform dispreading together
with a proper weighting, according to some of diversity schemes, such as maximum ratio combining or MMMSE combining.
The generic hybrid OFDM-CDMA-FH Scheme [76–78, 80] is provided in Fig. 10.45. The transmitter, shown in
Fig. 10.45a, is composed of four stages: (i) data modulation stage, (ii) DS-CDMA stage, (iii) OFDM stage, and (iv) FH
stage. When FH stage gets replaced with up-convertor, the corresponding scheme can be called hybrid OFDM-CDMA
scheme. On the other hand, when DS-CDMA stage is omitted, the corresponding system will be hybrid OFDM-FH scheme.
10.5 Hybrid OFDM-CDMA Systems 699

Fig. 10.45 Generic hybrid Spreading

DS-CDMA
OFDM-CDMA-FH scheme Data modulation stage sequence user k
stage
related to the k-th user: (a)
transmitter and (b) receiver Data stream, Channel QAM Data
configurations Interleaver
user k coding mapper scrambling

Guard
P/S S/P Pilot
DAC interval IFFT
conv. conv. insertion
insertion
OFDM Tx stage

Tx antenna

RF Tx
FH
stage
Frequency
synthesizer

PN code
generator

(a)

DS-CDMA Despreading
Data demodulation stage stage sequence user k

Estimated data, Channel QAM Data

Deinterleaver
user k decoder demapper descrambling

Guard
S/P P/S Channel
ADC Synch. interval FFT
conv. conv. correction
removal

OFDM Rx stage
Symbol timing

Rx antenna

RF Rx
FH
stage
Frequency
synthesizer

PN code
generator

(b)

The binary input sequence gets first encoded using one of the following alternatives: (1) convolutional code, (2) concatenated
convolutional-RS code, (3) turbo code, or (4) LDPC code. The interleaver is employed to randomize the burst errors due to the
deep fades on some subcarriers. After that, the I/Q modulator is used followed by QAM mapper. After that, the scrambling is
used, typically based on PN sequences. After that, the spreading is achieved by multiplying the scrambled signal with a
spreading sequence.
The PN sequence is not suitable for spreading since the scheme needs to be combined with OFDM and the period is not
sufficient to ensure good autocorrelation and cross-correlation properties. Instead, the Walsh sequences should be used. After
that, the spread signal is used as the input to the OFDM transmitter, which starts with pilot insertion, which is used on the
receiver side for channel estimation to compensate for channel impairments as well as for synchronization purposes. Further,
after pilot insertion, we perform S/P conversion, IFFT, P/S conversion, and guard interval insertion, followed by DAC. In FH
700 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

stage, the frequency synthesizer, which is phase-locked to the carrier frequency oscillator in RF Tx, changes the frequency
according to the FH code. The FH codes, described in Sect. 10.1.3, are applicable here as well.
The steps in the receiver are in reverse order compared to the transmitter. The receiver also has four stages: FH stage,
OFDM Rx, DS-CDMA Rx, and the data demodulation stage. Additional details about this scheme as well as its performance
comparison against MC-CDMA can be found in [80].

10.6 Ultra-Wideband (UWB) Communications

The UWB communication represents a fast emerging RF technology operating at very low energy levels, suitable for short-
range high-bandwidth applications, by occupying a large RF spectrum [84–93]. UWB technology can be considered as a
complement to other longer range wireless communication/networking technologies such as WiFi, WiMAX, and cellular
networks. It is also suitable for localization, radar, imaging, and medical applications. In the USA, UWB radio transmissions,
satisfying the transmitted power constraints, can legally operate in the range from 3.1 GHz up to 10.6 GHz [84–89]. Therefore,
UWB communication can support high data rates at short range (up to 10 m or so) with limited interference to other wire
devices. The UWB technologies can be classified into two broad categories: (i) impulse radio UWB (IR-UWB), typically
DS-CDMA-based, and (ii) multiband OFDM (MB-OFDM). The spectral mask for UWB communications, as regulated by
FCC 15.517(b,c), is provided in Fig. 10.46.
Two popular pulse shapes used in IR-UWB communications are Gaussian monocycle and doublet, which are derived as the
first and second derivative of Gaussian pulse g1(t) ¼ C1 exp[(t/ τ)2], as follows:

h i
h i
2t 2 2 2t 2 2
g2 ð t Þ ¼ C 2 exp  ð t=τ Þ , g3 ð t Þ ¼ C 3 1  exp  ð t=τ Þ : ð10:176Þ
τ2 τ2 τ

In order to occupy the whole UWB spectrum, ranging from 3.1 to 10.6 GHz, the pulse duration should be in the order of
hundreds of ps. Different modulation schemes suitable for use in IR-UWB communications can further be classified in two
broad categories, as illustrated in Fig. 10.47: (i) time-domain-based schemes such as pulse position modulation (PPM) and
pulse duration modulation (PDM) and (ii) shape-based schemes such as bi-phase modulation (BPM) or BPSK, pulse-
amplitude modulation (PAM), on-off keying (OOK), and orthogonal pulse modulation (OPM).
The signal constellation for M-ary PPM, where M is the signal constellation size, can be represented by the signal set
{si ¼ p(t-τi)} (i ¼ 1,. . .,M), where τi is delay parameter related to the i-th signal constellation point and p(t) is the pulse shape.

Fig. 10.46 The spectral mask for

UWB EIRP level, S [dBm/MHz]

UWB communications as 40
mandated by FCC

1.99 GHz
50

For indoor
UWB
60
communications

70

3.1 GHz 10.6 GHz

80
2 4 6 8 10
0.99 GHz
Frequency, f [GHz]
1.61 GHz
10.6 Ultra-Wideband (UWB) Communications 701

Fig. 10.47 Classification of

modulation formats for UWB UWB modulation formats
communications

Time-based techniques Shape-based techniques

Pulse position Pulse duration Bi-phase Pulse amplitude On-off keying Orthogonal pulse
modulation (PPM) modulation (PDM) modulation (BPM) modulation (PAM) (OOK) modulation (OPM)

The signal constellation for BPM (or BPSK) is represented by the following binary signal set {si ¼ bip(t)} (i ¼ 1,2;
bi2{1,1}). The signal constellation for OOK is represented by the signal set {si ¼ bip(t)} (i ¼ 1,2; bi2{0,1}). Finally, the
signal constellation for M-ary OPM is represented by the signal set {pi(t)} (i ¼ 1,. . .,M ), wherein the waveforms pi(t) are
mutually orthogonal. As an illustration, the Modified Hermite polynomials are mutually orthogonal and can be used as
symbols for M-OPM as follows [84, 85]:

  ½n=2 
X  n2i
dn t2 =2τ2 1 i ðt=τÞ
hn ðt Þ ¼ kn et =4τ2 2
¼ k n et =4τ τ2
2 2 2
τ n e  , ð10:177Þ
dt i¼0
2 ðn  2iÞ!i!

where the coefficient kn is related to the pulse energy En of hn by:

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
En
kn ¼ pffiffiffiffiffi:
τn! 2π

Typically, a UWB signal is in fact a sequence of sub-nanosecond pulses of low duty cycle (about 1%), with the bandwidth
being in the order of several GHz. Because the spectrum of UWB signal is spread over a wide range of frequencies, its power
spectral density (PSD) is low, so that the interference it causes to other existing wireless systems is extremely low.
Another class of orthogonal polynomials suitable for UWB applications are Legendre polynomials [95], which can be
obtained by applying the Rodrigues’ formula:

1 dn  2 n 
Pn ðxÞ ¼ n n x  1 , x ¼ t=τ, ð10:178Þ
2 n! dx

where τ is the scaling factor. It is trivial to show that the 0-th and the first terms are P0(x) ¼ 1 and P1(x) ¼ x. The higher-order
terms can be obtained by applying the following recurrent relation:

ðn þ 1ÞPnþ1 ðxÞ ¼ ð2n þ 1ÞxPn ðxÞ  nPn1 ðxÞ: ð10:179Þ

The Legendre polynomials satisfy the following orthogonality principle:

Z1
1
Pn ðxÞPm ðxÞdx ¼ δ : ð10:180Þ
2n þ 1 nm
1

The final class of functions to be considered here, to be used in OPM, are orthogonal prolate spheroidal wave functions
(OPSWF) [96]. The OPSW functions are simultaneously time-limited to symbol duration Ts and bandwidth-limited to band Ω
and can be obtained as solutions of the following integral equation [96, 97]:

ZT s =2
sin Ωðt  uÞ
Ψ n ð uÞ du ¼ ηn Ψn ðt Þ, ηn 2 ð0, 1 ð10:181Þ
π ð t  uÞ
T s =2

where the coefficient ηn is related to the energy concentration in the interval [Ts/2,Ts/2], and it is defined by:
702 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

ZT s =2
Ψn ðt Þdt
T s =2
ηn ¼ Z1 : ð10:182Þ
Ψn ðt Þdt
1

The OSPWs satisfy double-orthogonality principle:

ZT s =2 Z1
Ψn ðuÞΨm ðuÞdu ¼ ηn δnm , Ψn ðuÞΨm ðuÞdu ¼ δnm , ð10:183Þ
T s =2 1

and as such these functions are ideal for both UWB [85] and optical telecommunication [97] applications. The second
orthogonality condition indicates that these systems are more tolerant to intersymbol interference (ISI), since they stay
orthogonal outside of symbol duration. The following unique properties make OPSWFs ideal for UWB and optical
communications [85, 97]: the pulse duration is independent on index n, the pulse bandwidth does not change much with
index n, they satisfy orthogonality and double-orthogonality principles, for sufficiently large index n the D.C. component
tends to zero, and the pulse duration and bandwidth occupied can be tuned simultaneously.
Similarly to CDMA systems, to each user in UWB communications, we allocate a unique PN sequence of length N that is
used to encode the pulses in either position (PPM) or polarity (BPM). Let the PN sequence be denoted by {s} and binary
sequence by {b}. Then the PPM-based UWB signal can be represented as:

1 X
X N 1
sPPM‐UWB ðt Þ ¼ A pðt  iNT s  nT s  cn T c  τbi Þ, ð10:184Þ
i¼1 n¼0

where Ts is the symbol duration, while Tc is the chip duration. With p(t) we denoted the arbitrary pulse shape, with some of the
pulse shapes being discussed already. Clearly, for PPM-based UWB communication, the PN sequence modulates the
positions in increments/decrements of multiple of Tc, while the information bit bm introduces additional time shift bmτ. In
BPM (BPSK)-based UWB communications, the corresponding UWB signal can be represented as:

1 X
X N 1
sBPM‐UWB ðt Þ ¼ A bi cn pðt  iNT s  nT s Þ, bi 2 f1, 1g, ð10:185Þ
i¼1 n¼0

which is the same as in DS-SS systems. In other words, in BPM-based UWB communication, the PN sequence modulates the
polarity of the pulse within each symbol of duration Ts, while the information bit bn changes the polarity of the whole block of
N pulses simultaneously.
The block diagram of UWB receiver and demodulator of UWB signals is provided in Fig. 10.48, with operation principle
very similar to DS-SS systems, while the operation of automatic gain control (AGC) is described in Chap. 3. For PPM-based
UWB system, the signal corresponding to symbols 0 and 1, to be used in correlator shown in Fig. 10.48, is given by:

X
N 1 X
N 1
s0 ðt Þ ¼ pðt  nT s  cn T c þ τÞ, s1 ðt Þ ¼ pðt  nT s  cn T c  τÞ: ð10:186Þ
n¼0 n¼0

On the other hand, for BPM-based UWB system, the signal corresponding to symbols 0 and 1 is given by:

X
N 1
s1 ðt Þ ¼ cn pðt  nT s Þ, s0 ðt Þ ¼ s1 ðt Þ: ð10:187Þ
n¼0
10.6 Ultra-Wideband (UWB) Communications 703

Fig. 10.48 A typical UWB Rx

receiver configuration. LNA
low-noise amplifier, AGC
automatic gain control, ADC AGC
analog-to-digital conversion,
DAC digital-to-analog converter 1 1 Tb bˆi
ADC · dt
iTb t=iT
LNA s1(t) s0(t)
Synchronization,
control
DAC

Regarding the UWB channel models suitable for indoor applications, the model due to Saleh and Valenzuela [94] has been
adopted in IEEE 802.15.3a. According to Saleh-Valenzuela model, the multipath components arrive in clusters, while each
cluster is composed of several multipath rays. Clearly, the model has similarities with wideband multipath channel model
described in Chap. 2. The impulse response can be represented as:
XX
hð t Þ ¼ c
a δ ðt
r c,r
 τc  τc,r Þ, ð10:188Þ

where ac,r is normalized multipath gain related to the ray r within the cluster c. In (10.188), τc denotes the arrival time of the
first ray within the cluster c, while τc,r denotes the arrival time of the ray r within the cluster c. The magnitude distribution in
IEEE 802.15.3a has been assumed to be log-normal. The arrival time of the c-th cluster τc, given τc-1, follows the exponential
distribution:

Prfτc jτc1 g ¼ ΛeΛðτc τc1 Þ , ð10:189Þ

where Λ is the cluster arrival rate. On the other hand, the arrival time of the r-th ray within the c-th cluster follows as well the
exponential distribution:

Prfτc,r jτc,r1 g ¼ λeλðτc,r τc,r1 Þ , ð10:190Þ

where now λ is the arrival rate of the rays within the cluster. The power delay profile (PDP) is defined by double negative
exponential decaying functions:

τc τc,r
Pc,r ¼ P0,0 e Γ e γ , ð10:191Þ

where Γ and γ denote the cluster and ray time decay constants, respectively. For additional details of this model, an interested
reader is referred to [85, 94].
The multiband modulation represents approach to modulate information in UWB systems, by splitting the 7.5 GHz band
into multiple smaller frequency bands. The key idea is to efficiently utilize UWB spectrum by simultaneously transmitting
multiple UWB signals. This is applicable to both pulsed multiband and OFDM. To deal with multipath effects, mentioned
above, in IR-UWB, we can use the RAKE receiver, described in Sect. 10.1.2.3. On the other hand, in MB-OFDM, to deal with
multipath effects, we can use approaches described in Chap. 7.
The MB-OFDM system can be interpreted as a form of hybrid OFDM-FH system, similar to the one described in previous
section, when the DS-CDMA section is omitted. The key idea of MB-OFDM is to “spread” the information bits across the
whole UWB spectrum, exploiting, therefore, frequency diversity and providing the robustness against multipath fading and
narrowband interference, originating from WiFi channels. The 7.5 GHz UWB spectrum can be divided into 14 bands, each of
bandwidth of Δf ¼ 528 MHz, as illustrated in Fig. 10.49. The center frequency of the nth band fc,n, the low boundary
frequency fl,n, and upper boundary frequency fh,n can be determined, respectively, as:

f c,n ¼ f 0 þ nΔf , f l,n ¼ f c,n  Δf =2, f h,n ¼ f c,n þ Δf =2, f 0 ¼ 2904 MHz:
704 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

Band group # 1 Band group # 2 Band group # 3 Band group # 4 Band group # 5

Band Band Band Band Band Band Band Band Band Band Band Band Band Band
#1 #2 #3 #4 #5 #6 #7 #8 # 9 # 10 # 11 # 12 # 13 # 14

Mandatory Optional Optional Optional Optional

3432 3960 4488 5016 5544 6072 6600 7128 7656 8184 8712 9240 9768 10296 f [MHz]

Fig. 10.49 The frequency allocation in MB-OFDM for UWB communications

Fig. 10.50 The MB-OFDM

transmitter configuration for Data modulation section
UWB communications Binary
Encoding & Bit QAM
Scrambler
input data puncturing interleaver mapping

Cyclic extension Pilot

DAC P/S IFFT S/P
& windowing insertion

OFDM Tx
Tx antenna
RF Tx
exp(j2 fc)
Time-frequency
coding

The MB-OFDM transmitter configuration is provided in Fig. 10.50, and as expected it is very similar to conventional
OFDM transmitter, described in Chap. 7 [see Fig. 7.6a], except from the time-frequency coding block. The time-frequency
code (TFC) denotes which band within the band group is occupied at given transmission interval. As an illustration, the TFCs
of length 6 for band group 1 are given as [85] {[1 2 3 1 2 3], [1 3 2 1 3 2], [1 1 2 2 3 3], [1 1 3 3 2 2], [1 2 1 2 1 2], [1 1 1 2 2 2]}.
When TFC code 1 is used, the first OFDM symbol is transmitted on band 1, the second OFDM symbol on band 2, the third
OFDM symbol on band 3, the fourth OFDM symbol on band 1, the firth OFDM symbol on band 2, and the sixth OFDM
symbol on band 3; and this sequence of bands within the band group is repeated until the TFC is changed in time-frequency
coding block. The FH codes described in Sect. 10.1.3 can be also used as TFC codes in MB-OFDM-based UWB
communications. The binary data bits are scrambled, and then encoded employing a convolutional code of rate 1/3, with
constraint length K ¼ 7, and generating polynomials g0 ¼ 1338, g1 ¼ 1658, and g2 ¼ 1718 (the subscript denotes octal
number system) [93]. Higher-rate convolutional codes (of rate 11/32, 1/2, 5/8, and 3/4) can be obtained by puncturing. The
number of subcarriers used in MB-OFDM is 128, 100 of which are used to transmit QPSK data, 12 are pilots, and 10 are guard
tones (the remained subcarriers are set to zero). The pilots are used for channel estimation and carrier-phase tracking on
receiver side. To improve performance of MB-OFDM scheme, LDPC-coded MB-OFDM can be used instead. It has been
shown in [92, 93] that the transmission distance over wireless channel of LDPC-coded MB-OFDM can be increased by
29–73% compared to that of MB-OFDM with convolutional codes. After pilot insertion, and serial-to-parallel conversion
(S/P), inverse FFT (IFFT) operation, and parallel-to-serial (P/S) conversion, cyclic extension and windowing are performed.
After DAC and time-frequency coding, the RF signal is transmitted over transmit (Tx) antenna. The transmitted MB-OFDM-
based UWB signal can be written as:
( )
X
N 1 h i
sMB‐OFDM‐UWB ðt Þ ¼ Re sOFDM,n ðt  nT OFDM Þ exp j2π f ðn mod 6Þ t , ð10:192Þ
n¼0
10.6 Ultra-Wideband (UWB) Communications 705

cos( ct)
AGC
VGA Timing
Pre-select
filter LNA LPF ADC
Timing and Remove
frequency cyclic
synchronization extension
LPF ADC

Frequency-
sin( ct) corrected signal
OFDM Rx
Output
QAM Channel
Descrambler Decoding Deinterleaver P/S FFT S/P
data de-mapping correction

Fig. 10.51 The MB-OFDM receiver configuration for UWB communications

where N is the number of OFDM symbols transmitted, while sOFDM, n(t) denotes the nth OFDM symbol of duration TOFDM
transmitted on carrier frequency f n mod 6. As discussed above, the carrier frequency changes over three frequencies assigned to
the band group, organized in sequences of length 6 (TFCs). The n-th OFDM symbol is generated in similar fashion as we
described in Chap. 7.
The receiver configuration, shown in Fig. 10.51, is very similar to conventional OFDM receiver [Fig. 7.6b], except from
preselected filter that is used to select OFDM signal of 528 MHz bandwidth at given carrier frequency.
One of the important issues related to the UWB communications is the coexistence with other wireless communication
systems and the narrowband interference (NBI) influence on UWB system performance. Even though that NBI bandwidth is
low compared to UWB signal bandwidth, the power of NBI is significantly higher in that frequency band. To deal with
multipath effects, we can use the RAKE receiver, shown in Fig. 10.9, where different outputs of correlators are properly
weighted with weight vector w, so that the decision statistics will be based on the following:

w{ r, r ¼ s þ i þ z ð10:193Þ

where s is the desired signal vector, i is the interference vector, and z is the noise vector. The receiver output signal-to-
interference-plus-noise ratio, denoted by SINR, when NBI is ignored, by using maximum ratio combining (MRC), is given by
[85]:

N s E s hh{    
SINRMRC ¼ {
; Ri ¼ ii{ , Rz ¼ zz{ , ð10:194Þ
h ðRi þ Rz Þh

where Es is the symbol energy, h is channel fading coefficient, and Ns is the number of bits per symbol. (Ri and Rz denote the
covariance matrices of interference and noise, respectively). In MRC, however, the influence of interference is ignored,
representing sub-optimum solution. The optimum weighting must be determined by minimizing the MSE, to obtain the
following optimum weight vector for the minimum MSE combiner (MMSEC) [85]:

N s E s ðRi þ Rz Þ1 h
w= , ð10:195Þ
1 þ N s E s h{ ðRi þ Rz Þ1 h

and the corresponding SINR is given by:

SINRMMSEC ¼ N s Es h{ ðRi þ Rz Þ1 h: ð10:196Þ

It has been shown in [85] that MRC results in significant performance degradation compared to MMSEC, because it
ignores the presence of NBI (due to WLAN device).
706 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

Another effective but low-complexity approach to deal with NBI is to use the notch filter of bandwidth B with transfer
function [85]:

H ð f Þ ¼ 1  ½uð f  f i þ B=2Þ  uð f  f i  B=2Þ, ð10:197Þ

where fi is the center frequency of the NBI and u(x) is the unit-step function. The corresponding impulse response is obtained
as inverse Fourier transform (FT) of H( f ):

hðt Þ ¼ FT 1 fH ð f Þg ¼ δðt Þ  Be j2π f i t

sincðBt Þ: ð10:198Þ

Another relevant issue for UWB communications is their short reach. To extend the transmission distance, the radio-over-
fiber (RoF) technologies can be used, as described in [98].

10.7 Concluding Remarks

The subject of this chapter has been devoted to spread spectrum (SS) communications systems, described in Sect. 10.1;
CDMA systems, described in Sect. 10.2; multiuser detection schemes, described in Sect. 10.3; optical CDMA systems,
described in Sect. 10.4; hybrid OFDM-CDMA systems, introduced in Sect. 10.5; and UWB communication systems,
described in Sect. 10.6.
In the section on SS systems (Sect. 10.1), after the introduction of SS systems (Sect. 10.1.1), both DS-SS and FH-SS
systems have been described in detail. Regarding the DS-SS systems (Sect. 10.1.2), after typical transmitter and receiver
configurations have been described in Sect. 10.12.1, the synchronization process, including acquisition and tracking stages,
has been provided in Sect. 10.1.2.2. To deal with the multipath fading effects, the use of RAKE receiver has been described in
Sect. 10.1.2.3. In the subsection on spreading sequences (Sect. 10.1.2.4), after the introduction of PN sequences as well as the
basic randomness criterions, we have introduced maximal-length sequences (m-sequences), including their basic properties.
In the same subsection, the concept of nonlinear PN sequences has been introduced as well. In Sect. 10.1.3, devoted to the
FH-SS systems, we have provided the detailed description of FH-SS systems, including both transmitter and demodulator as
well as various time acquisition schemes. In the same section, we have described how to generate the FH sequences.
In Sect. 10.2, after the basic introduction of CDMA systems, we have introduced the various signature waveforms (Sect.
10.2.1.) suitable for use in DS-CDMA systems including Gold, Kasami, and Walsh sequences. Further, the relevant
synchronous and asynchronous CDMA models have been described in Sect. 10.2.2, with special attention devoted to
discrete-time CDMA models.
In the section on multiuser detection (MUD), Sect. 10.3, we have described how to deal with interference introduced by
various CDMA users. The conventional single-user detector, introduced in Sect. 10.3.1, has been used as the reference case.
The various MUD schemes have been described then on Sect. 10.3.2, with special attention being paid to the jointly optimum
MUD. The complexity of jointly optimum MUD might be prohibitively high for certain application, and for such application,
we have introduced decorrelating receiver in Sect. 10.3.3, which can compensate for multiuser interference but enhances the
noise effect. To solve for this problem, the linear MMSE receiver has been introduced as well in the same section capable of
dealing simultaneously with both multiuser interference and noise effects. In Sect. 10.3.4, we have described the successive
cancelation and multistage detection schemes, taking either preliminary or final decisions from other users in detection
process.
In the section on optical CDMA (OCDMA), Sect. 10.4, we have introduced both incoherent and coherent optical detection-
based OCDMA schemes. Regarding the incoherent optical detection-based schemes, we first have described how to design
various unipolar signature sequences, optical orthogonal codes (OOC), followed by the description of basic incoherent
OCDMA schemes including time-spreading, spectral amplitude coding (SAC), and wavelength hopping (WH) schemes.
Related to coherent optical detection-based OCDMA systems, we have described both pulse laser-based and CW laser-based
OCDMA systems.
In Sect. 10.5, the hybrid OFDM-CDMA systems have been described, taking the advantages of both OFDM and CDMA
concepts into account, particularly in dealing with frequency selective fading and narrowband interference, while at the same
time supporting multiple users. The following three classes of hybrid OFDM-CDMA systems have been introduced:
multicarrier CDMA (MC-CDMA), OFDM-CDMA, and OFDM-CDMA-FH systems.
10.8 Problems 707

In the section on UWB communications, Sect. 10.6, we have described the concepts and requirements for UWB
communication as well as transmitter and receiver configurations. In the same section, various modulation formats suitable
for UWB communications have been described. These have been categorized into time-domain category (pulse-position
nodulation and pulse-duration modulation) and shape-based category (on-off keying, pulse-amplitude modulation, bi-phase
modulation (BPM) or BPSK, and orthogonal pulse modulation). Regarding the pulse shapes suitable for orthogonal pulse
modulation (OPM), the modified Hermite polynomials, Legendre polynomials, and orthogonal prolate spheroidal wave
functions (OPSWF) have been introduced. Further, the Saleh-Valenzuela UWB channel model has been described, suitable
for indoor applications. Both types of UWB systems have been described in detail: the impulse radio UWB (IR-UWB),
typically DS-CDMA-based, and multiband OFDM (MB-OFDM).
The set of problems is provided in the following section to help the reader gain deeper understanding of the material
described in this chapter.

10.8 Problems

1. The jammer signal is composed of three tones:

J(t) ¼ J cos (2πfct + θ) + J cos (2π( fc  Δf )t + θ) + J+ cos (2π( fc + Δf )t + θ),and applied to the input of DS-SS with
QPSK. Determine SIR and PG as a function of Δf.
2. Assuming that in M-ary PSK DS-SS the block code of rate R ¼ k/n and codeword weight distribution {wm} is used, derive
the upper bound for the codeword error probability.
3. Let us assume that the DS-SS system is based on m-sequence with m being 5 with the chip rate of 106 chips/s.
(a) Determine PN sequence length, period, and chip duration.
(b) Sketch the autocorrelation function and PSD of this PN sequence.
(c) By using (a), determine the average acquisition time for a serial DS search under assumptions that the detection
probability is 0.95 and maximum tolerable false alarm probability is 0.1.
4. Let us consider the following two m-sequences’ generators for shift-registers of length 8: (i) the feedback taps are taken
after the flip-flops {8,6,5,1} and (ii) the feedback taps are taken after flip-flops {8,6,4,3,2,1}.
(a) Calculate and plot autocorrelation and cross-correlation functions.
(b) Calculate and plot the corresponding PSDs.
5. Let us consider the example to generate the Gold sequences from polynomials g1(x) ¼ 1 + x2 + x5 and
g2(x) ¼ 1 + x + x2 + x4 + x5, provided in Fig. 10.21. Generate the Gold sequences using Eq. (10.79).
(a) Calculate and plot autocorrelation and cross-correlation functions.
(b) Calculate and plot the corresponding PSDs.
6. Let us consider a QPSK DS-SS in which m-sequence has parameter m ¼ 23. Under the requirement that the BER caused
by the interfering jammer cannot be larger than 106, determine the processing gain and antijam margin.
7. The multipath channel is composed of the LOS component of amplitude a0 and two reflected components with amplitudes
a1 and a2 with corresponding delays with respect to the LOS being τ1 and τ2, respectively. The amplitudes follow
Nakagami-m fading with m ¼ 0.5 and 1 each occurring with probability 0.5. The corresponding average powers of the
LOS related to m ¼ 0.5 and m ¼ 1 are 10 and 20. The average powers of the first reflected components related to m ¼ 0.5
and m ¼ 1 are 1 and 5. Finally, the average powers of the second reflected components related to m ¼ 0.5 and m ¼ 1 are
5 and 10. Let us assume that the SS receiver get locked to the i-th multipath component in the absence of other
components with corresponding SNR being ai2 (thus normalized N0 ¼ 1). The bit duration of the BPSK scheme is Tb.
The m-sequence is used to implement the DS-SS system.
(a) When the receiver gets locked to the LOS component, determine τi (i ¼ 1, 2) so that the corresponding multipath
components are attenuated by 1/N (N-the number of chips).
(b) By observing a single-branch SS receiver locked to the LOS component, determine the outage probability
corresponding to the instantaneous probability of error Pb of 104.
(c) For the same instantaneous probability and three-branch RAKE receiver, each locked to one multipath component,
determine the outage probability when selection combining is used. Compare the result against (a).
(d) Repeat (c) when the maximum ratio combining is used instead.
8. In this problem, we are interested in FH based on M-ary FSK. The bandwidth allocated to our system is BW ¼ 100 MHz.
Let us assume that the multipath channel is a two-ray model channel, wherein the reflected ray is 6 dB lower than the LOS
ray. Assume that the reflection coefficient is 0.2. Let us use the delay relative to the 1/BW as a parameter.
708 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

(a) Evaluate the BER performance for the following values of delay of reflected component relative to the LSO τ2{0.25,
0.5, 1, 2, 4}/BW for the following values of M2{2, 4, 8, 16}. Select the chip interval Tc so that the reflected component
will not affect the detection of the LOS component that much.
(b) Repeat (a) but selecting Tc such the system will exhibit slow-varying flat fading.
(c) Repeat (a) but selecting Tc such the system will exhibit slow-varying frequency-selective fading.
9. In this problem we are interested in evaluation performance of FH-SS system with M-ary PSK in the presence jammer.
The bandwidth of modulated signal is B ¼ 20 MHz. The jammer can vary the power PJ, center frequency, and the
bandwidth of his jamming signal. Evaluate the performance of this system in the presence of AWGN and jammer activity
as describe above vs. SNR. Use the PJ and various jammer strategies as parameters.
PN
10. Let us consider the DS-SS waveform given by sc ðt Þ ¼ Ac αn pT c ½t  ðn  1ÞT c , where [α1,. . ., αn,. . ., αN] is the
n¼1
nonbinary spreading code (real-valued components) with A being the normalization amplitude such that ||sc|| ¼ 1. If the
maximum tolerable cross-correlation between any two signature sequences is ρmax, prove that the maximum number of
users that can be supported with N chips per symbol is upper-bounded by [5, 99]:
 
1  ρ2max N ðN þ 2Þ⋯ðN þ 2kÞ
K max  min   ,
k ð2k þ 1Þ  ðN þ 2kÞρ2max ½1  3  . . .  ð2k  1Þ

with minimization being performed over all ks for which the denominator is positive. Specify the particular case for
which ρ2max ¼ 1=N.
11. The OFDM can be considered as a special case of the synchronous CDMA model in which the signature waveforms are
given by si(t) ¼ T1/2 exp [j2π(i  1  0.5(K  1))t/T], where T is the symbol duration and K is number of users.
(a) Show that any two signature waveforms are orthogonal.
(b) In the absence of noise, the received signal is given by r(t) ¼ A∑ibisi(t). Show that the K-point DFT given by [r(0) r(T/
K ) . . . r((K-1)T/K )] reconstructs the transmitted sequence.
12. This problem is related to diversity in which we receive M copies of transmitted signal s(t) affected by noise σz(t) and
Rayleigh fading coefficients Ri as follows: ri(t) ¼ ARis(t) + σz(t), where i ¼ 1,. . .,M.
(a) In selection combining, we select the branch with maximum SNR (equivalently maximum Ri2). Determine the
corresponding cumulative distribution function of the resulting SNR.
(b) In equal-gain combining, all diversity branches will have the same weight in resulting signal. Determine the
corresponding moment-generating function of the SNR.
(c) In maximum-ratio combining, we weigh the output of the i-th branch by Ri before summing them up. Determine the
corresponding moment-generating function of the SNR.
13. In this problem we are concerned with the decomposition of cross-correlation matrix by R ¼ FTF. Show that the i-th
column in the lower triangular matrix F depends only on si,. . .,sK.
14. Solve the equations in (10.115) for F[0] and F [1] in a two-user asynchronous channel model.
15. Let us consider K-user synchronous channel under assumption that any two users’ cross-correlation is the same and equal
to ρ and the amplitudes for all users are the same (Ak ¼ A). Determine the range of cross-correlation for which the
correlation matrix is positive-definite.
(a) Determine the bit error probability when single-user matched filter is used.
(b) Apply the Gaussian approximation given by Eq. (10.126) and discuss the accuracy of this approximation for K ¼ 2.
16. The optimum two-user detection scheme is one that performs the following maximization:
 
bb1 , b
b2 ¼ arg max ðr 1 A1 b1 þ r 2 A2 b2  A1 b1 A2 b2 ρÞ:
ðb1 , b2 Þ

Show that this expression can be expressed equivalently as:

 
b 1 1
b1 ¼ sgn r 1 A1 þ jr 2 A2  A1 A2 ρj  jr 2 A2 þ A1 A2 ρj
2 2
 
b 1 1
b2 ¼ sgn r 2 A2 þ jr 1 A1  A1 A2 ρj  jr 1 A1 þ A1 A2 ρj ,
2 2
References 709

17. Let us consider a three-user synchronous case in which the signature waveforms s1 and s2 are orthogonal, while s3 is the
superposition s3 ¼ s1 + s2/√2. Sketch the decision regions when the jointly optimum detection is used.
18. The ML decision rule for K-users’ synchronous case is given by.
b
b ¼ arg max 2bT Ar  bT ðARAÞb where ARA is the un-normalized correlation matrix. Show that equivalently the
b
optimization can be represented by:

x ¼ arg min xT Bx þ W T x,
b
x2f0, 1gK

where B ¼ ARA-V and W¼V-ARA[1. . .1]T-Ar, with V being arbitrary diagonal matrix V ¼ diag (v1, . . ., vK). See
[100] for additional details of this representation.
19. Let us go back to the K-user synchronous channel model with M diversity branches, described in Problem 12, namely:
X
r m ðt Þ ¼ A b s ðt Þ
k mk k k
þ σzm ðt Þ; m ¼ 1, . . . , M
RT
Show that the following set of variables {rmk} with r m ðt Þsk ðt Þdt represents the sufficient statistics.
0
P
20. Let us study the K-user asynchronous channel. Prove that, when the open eye condition jr k j > Ki¼1,i6¼k Ai ðjρik j þ jρki jÞ is
satisfied, the jointly optimum detector is simply b bk ¼ sgn ðr k Þ.
21. Let us observe a two-user synchronous channel model. Determine ρ, A1/σ, and A2/σ in such a way that the decorrelation
detector outperforms single-user matched filter (in BER sense), while the signal-to-interference ratio (SIR) in decorrelator
is lower than that in the single-used matched detector.
! ! !
22. Let us consider the discrete time synchronous model introduced by Eq. (10.103), namely, r ¼ SAb þ z . Show that Sþ r ,
where S is the pseudo-inverse operation, is the solution of the following minimization problem
+
   
þ! ! T 1 þ! ! !
min
!
SS r  r C SS r  r , where C is the covariance matrix of z .
Sþ r 2RK
1 P
23. Determine the coefficients ak in the following equation: ½R þ σ 2 I ¼ Kk¼1 ak Rk .
24. Derive the implementations of the multistage detectors with conventional and decorrelating two stages in terms of the
model discussed in Problem 22.
25. Describe key advantages and disadvantages of incoherent optical CDMA systems with respect to corresponding coherent
detection schemes. Discuss possible applications of incoherent CDMA systems.
26. Discuss the advantages of the hybrid OFDM-CDMA systems compared to CDMA-only and OFDM-only systems.
27. Let us consider the system model given by Eq. (10.193) in which multipath fading is compensated by RAKE receiver, but
when in addition to AWGN noise the narrowband interference is present. Derive optimum weights minimizing the MSE
in maximum-ratio combining receiver in this scenario.

References
1. S. Glisic, B. Vucetic, Spread Spectrum CDMA Systems for Wireless Communications (Artech House, Inc, Boston/London, 1997)
2. A. Goldsmith, Wireless Communications (Cambridge University Press, Cambridge/New York, 2005)
3. J.G. Proakis, Digital Communications, 4th edn. (McGraw-Hill, Boston, 2001)
4. S. Haykin, Digital Communication Systems (Wiley, Hoboken, 2014)
5. S. Verdú, Multiuser Detection (Cambridge University Press, Cambridge/New York, 1998)
6. R. Pickholtz, D. Schilling, L. Milstein, Theory of spread-Spectrum communications – a tutorial. IEEE Trans. Commun. 30(5), 855–884 (1982)
7. D. Tse, P. Viswanath, Fundamentals of Wireless Communication (Cambridge University Press, Cambridge, 2005)
8. B. Sklar, Digital Communications: Fundamentals and Applications, 2nd edn. (Prentice-Hall, Inc, Upper Saddle River, 2001)
9. J.K. Holmes, C.C. Chen, Acquisition time performance of PN spread-spectrum systems. IEEE Trans. Commun. 25(8), 778–784 (1977)
10. M.K. Simon, J.K. Omura, R.A. Scholtz, B.K. Levit, Spread Spectrum Communications (Computer Science Press, Inc., Rockville, 1985)
11. D. Torrieri, Principles of Spread-Spectrum Communication Systems, 3rd edn. (Springer International Publishing Switzerland, 2015)
710 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

12. R.B. Ward, Acquisition of pseudonoise signals by sequential estimation. IEEE Trans. Commun. Technol. 13(4), 475–483 (1965)
13. R. Price, P.E. Green, A communication technique for multipath channels. Proc. IRE 46(3), 555–570 (1958)
14. G.L. Turin, Introduction to spread spectrum antimultipath techniques and their application in urban digital radio. Proc. IEEE 68(3), 328–353
(1980)
15. D.V. Sarwate, M.B. Pursley, Crosscorrelation properties of pseudorandom and related sequences. Proc. IEEE 68(5), 593–619 (1980)
16. S.W. Golomb, Shift Register Sequences: Secure and Limited-Access Code Generators, Efficiency Code Generators, Prescribed Property
Generators, Mathematical Models, 3rd Revised edn. (World Scientific Publishing Co., Singapore, 2017)
17. D.B. Drajic, Introduction to Statistical Theory of Telecommunications (Akademska Misao, Belgrade, 2003). (In Serbian)
18. N.G. De Bruijn, A combinatorial problem. Proc. Sect. Sci. Koninklijke Nederlands Akademie wan Wetenschappen 49(2), 758–764 (1946)
19. I.J. Good, Normal recurring decimals. J. London Math. Soc. 21(3), 169–172 (1946)
20. T. Etzion, A. Lempel, Algorithms for the generation of full-length shift-register sequences. IEEE Trans. Inform. Theory 30(3), 480–484 (1984)
21. T.S. Seay, Hopping patterns for bounded mutual interference in frequency hopping multiple access, in Proceedings of the IEEE Military
Communications Conference – Progress in Spread Spectrum Communications (MILCOM ’82), (Boston, 1982), pp. 22.3-1–22.3-6
22. A. Lempel, H. Greenberger, Families of sequences with optimal Hamming correlation properties. IEEE Trans. Inform. Theory 20(1), 90–94
(1974)
23. I.S. Reed, K-th order near-orthogonal codes. IEEE Trans. Inform. Theory 17(1), 116–117 (1971)
24. G. Solomon, Optimal frequency hopping sequences for multiple access. Proc. Syp. Spread Spectrum Commun. 1, 33–35 (1973)
25. B.M. Popovic, Comparative Analysis of Sequences for Asynchronous Code-division Multiplex with Frequency Hopping, Master Thesis,
University of Belgrade, Serbia. (in Serbian) (1989)
26. A.A. Shaar, P.A. Dvaiees, A survey of one-coincidence sequences for frequency-hopped spread-spectrum systems. IEEE Proc. Commun.
Radar Signal Proc. Part F 131(7), 719–724 (1984)
27. L.R. Welch, Lower bounds on the maximum cross correlation of signals (Corresp.). IEEE Trans. Inform. Theory 20(3), 397–399 (1974)
28. R. Gold, Optimal binary sequences for spread spectrum multiplexing (Corresp.). IEEE Trans. Inform. Theory 13(4), 619–621 (1967)
29. R. Gold, Maximal recursive sequences with 3-valued recursive cross-correlation functions (Corresp.). IEEE Trans. Inform. Theory 14(1),
154–156 (1968)
30. J.L. Walsh, A closed set of normal orthogonal functions. Am. J. Math. 45(1), 5–24 (1923)
31. H.F. Harmuth, Applications of Walsh functions in communications. IEEE Spectr. 6(11), 82–91 (1969)
32. K.G. Beauchamp, Walsh Functions and Their Applications (Academic, London, 1975)
33. R.N. McDonough, A.D. Whalen, Detection of Signals in Noise, 2nd edn. (Academic, San Diego, 1995)
34. Vasic B, Digital Communications II, Lecture Notes, 2005
35. S. Verdú, Optimum Multiuser Signal Detection, PhD dissertation, University of Illinois, Urbana-Champaign, 1984)
36. S. Verdú, Minimum probability of error for asynchronous Gaussian multiple-access channels. IEEE Trans. Inform. Theory 32(1), 85–96 (1986)
37. R. Lupas, S. Verdú, Linear multiuser detectors for synchronous code-division multiple-access channels. IEEE Trans. Inform. Theory 35(1),
123–136 (1989)
38. S. Verdú, Multiuser detection, in Advances in Statistical Signal Processing: Signal Detection, ed. by H. V. Poor, J. B. Thomas, (JAI Press,
1993), pp. 369–410
39. S. Verdú, Computational complexity of optimum multiuser detection. AIgorithmica 4, 303–312 (1989)
40. S. Verdú, Optimum multiuser asymptotic efficiency. IEEE Trans. Commun. 34, 890–897 (1986)
41. S. Verdú, Multiple-access channels with point-process observations: Optimum demodulation. IEEE Trans. Inform. Theory 32, 642–651 (1986)
42. S. Verdú, Adaptive multiuser detection, in Code Division Multiple Access Communications, ed. by S. G. Glisic, P. A. Leppanen, (Kluwer
Academic, 1995)
43. A. Duel-Hallen, J. Holtzman, Z. Zvonar, Multiuser detection for CDMA systems. IEEE Pers. Commun. 2(2), 46–58 (1995)
44. J.G. Andrews, Interference cancellation for cellular systems: A contemporary overview. IEEE Wirel. Commun. 12(2), 19–29 (2005)
45. Y.C. Yoon, R. Kohno, H. Imai, A spread-spectrum multiaccess system with cochannel interference cancellation for multipath fading channels.
IEEE J. Sel. Areas Commun. 11(7), 1067–1075 (1993)
46. P. Patel, J. Holtzman, Analysis of a simple successive interference cancellation scheme in a DS/CDMA system. IEEE J. Sel. Areas Commun.
12(5), 796–807 (1994)
47. D. Divsalar, M.K. Simon, D. Raphaeli, Improved parallel interference cancellation for CDMA. IEEE Trans. Commun. 46(2), 258–268 (1998)
48. M.K. Varanasi, B. Aazhang, Near-optimum detection in synchronous code-division multiple access systems. IEEE Trans. Commun. 39,
725–736 (1991)
49. M.K. Varanasi, B. Aazhang, Multistage detection in asynchronous code-division multiple-access communications. IEEE Trans. Commun.
38(4), 509–519 (1990)
50. J. Luo, K.R. Pattipati, P.K. Willett, F. Hasegawa, Near-optimal multiuser detection in synchronous CDMA using probabilistic data association.
IEEE Commun. Lett. 5(9), 361–363 (2001)
51. X. Wang, H.V. Poor, Wireless Communication Systems: Advanced Techniques for Signal Reception (Prentice-Hall, Emglewood Cliffs, 2004)
52. A. Duel-Hallen, Decorrelating decision-feedback multiuser detector for synchronous code-division multiple-access channel. IEEE Trans.
Commun. 41(2), 285–290 (1993)
53. A. Duel-Hallen, A family of multiuser decision-feedback detectors for asynchronous code-division multiple-access channels. IEEE Trans.
Commun. 43(2/3/4), 421–434 (1995)
54. P. Prucnal, M. Santoro, T. Fan, Spread spectrum fiber-optic local area network using optical processing. J. Lightw. Technol. 4(5), 547–554
(1986)
55. P.R. Prucnal, Optical Code Division Multiple Access: Fundamentals and Applications (CRC Press, Boca Raton, 2006)
56. J.A. Salehi, Code division multiple-access techniques in optical fiber networks-part I: Fundamental principles. IEEE Trans. Commun. 37(8),
824–833 (1989)
57. Fathallah H, Rusch LA, and S. LaRochelle S (1999), Passive optical fast frequency-hop CDMA communication system. J. Lightwave Technol.
17: 397–405
References 711

58. J.A. Salehi, F.R.K. Chung, V.K. Wei, Optical orthogonal codes: Design, analysis, and applications. IEEE Trans. Inform. Theory 35(3),
595–605 (1989)
59. R. Petrovic, Orthogonal codes for CDMA optical fiber LAN’s with variable bit interval. IEE Electron. Lett. 26(10), 662–664 (1990)
60. S.V. Maric, A new family of optical code sequences for use in spread-spectrum fiber-optic local area networks. IEEE Trans. Commun. 41(8),
1217–1221 (1993)
61. S.V. Maric, New family of algebraically designed optical orthogonal codes for use in CDMA fiber-optic networks. IEE Electron. Lett. 29(6),
538–539 (1993)
62. G.-C. Yang, W.C. Kwong, Performance analysis of optical CDMA with prime codes. IEE Electron. Lett. 31(7), 569–570 (1995)
63. Z. Wei, H. Ghafouri-Shiraz, Proposal of a novel code for spectral amplitude-coding optical CDMA systems. IEEE Photon Technol. Lett. 14,
414–416 (2002)
64. S. Yengnarayanan, A.S. Bhushan, B. Jalali, Fast wavelength-hopping time-spreading encoding/decoding for optical CDMA. IEEE Photon.
Technol. Lett. 12(5), 573–575 (2000)
65. W. Huang, M.H.M. Nizam, I. Andonovic, M. Tur, Coherent optical CDMA (OCDMA) systems used for high-capacity optical fiber networks-
system description, OTDMA comparison, and OCDMA/WDMA networking. J. Lightw. Technol. 18(6), 765–778 (2000)
66. I.B. Djordjevic, B. Vasic, Novel combinatorial constructions of optical orthogonal codes for incoherent optical CDMA systems. J. Lightw.
Technol. 21(9), 1869–1875 (2003)
67. I.B. Djordjevic, D.F. Geraghty, A. Patil, C.H. Chen, R. Kostuk, B. Vasic, Demonstration of spectral-amplitude encoding/decoding for
multimedia optical CDMA applications. J. Opt. Commun. 27(2), 121–124 (2006)
68. I.B. Djordjevic, B. Vasic, Unipolar codes for spectral-amplitude-coding optical CDMA systems based on projective geometries. IEEE Photon.
Technol. Lett. 15(9), 1318–1320 (2003)
69. I.B. Djordjevic, B. Vasic, Combinatorial constructions of optical orthogonal codes for OCDMA systems. IEEE Commun. Lett. 8(6), 391–393
(2004)
70. A. Patil, I.B. Djordjevic, D.F. Geraghty, Performance Assessment of Optical CDMA Systems based on Wavelength-Time Codes from
Balanced Incomplete Block Designs, in Proceedings of the 7th International Conference on Telecommunications in Modern Satellite, Cable
and Broadcasting Services-TELSIKS 2005, (Nis, Serbia and Montenegro, 2005), pp. 295–298, September 28–30
71. I.B. Djordjevic, B. Vasic, J. Rorison, Multi-weight unipolar codes for spectral-amplitude-coding optical CDMA systems. IEEE Commun. Lett.
8(4), 259–261 (2004)
72. I.B. Djordjevic, B. Vasic, J. Rorison, Design of multi-weight unipolar codes for multimedia optical CDMA applications based on pairwise
balanced designs. IEEE/OSA J. Lightw. Technol. 21(9), 1850–1856 (2003)
73. E.S. Shivaleela, K.N. Sivarajan, A. Selvarajan, Design of a new family of two-dimensional codes for fiber-optic CDMA networks. J. Lightw.
Technol. 16(4), 501–508 (1998)
74. E.S. Shivaleela, A. Selvarajan, T. Srinivas, Two-dimensional optical orthogonal codes for fiber-optic CDMA networks. J. Lightw. Technol.
23(2), 647–654 (2005)
75. L. Tancevski, I. Andonovic, J. Budin, Secure optical network architectures utilizing wavelength hopping/time spreading codes. IEEE Photon.
Technol. Lett. 7(5), 573–575 (1995)
76. M. Jankiraman, R. Prasad, Hybrid CDMA/OFDM/SFH: a novel solution for wideband multimedia communications, in Proceedings of the 4th
ACTS Mobile Communication Summit ’99, (Sorento, Italy, 1999), pp. 1–7
77. M. Jankiraman, R. Prasad, A novel solution to wireless multimedia application: the hybrid OFDM/CDMA/SFH approach, in Proceedings of
the 11th IEEE International Symposium on Personal Indoor and Mobile Radio Communications (PIMRC 2000), 2, (London, UK, 2000),
pp. 1368–1374
78. M. Jankiraman, R. Prasad, Wireless multimedia application: the hybrid OFDM/CDMA/SFH approach, in Proceedings of the 2000 IEEE Sixth
International Symposium on Spread Spectrum Techniques and Applications (ISSTA 2000), 2, (Parsippany, NJ, 2000), pp. 387–393
79. R. Van Nee, R. Prasad, OFDM Wireless Multimedia Communications (Artech House, Boston/London, 2000)
80. R. Prasad, OFDM for Wireless Communications Systems (Artech House, Boston/London, 2004)
81. N. Yee, J.-P. Linnartz, G. Fettweis, Multi-carrier CDMA in indoor wireless networks, in Proceedings of the Fourth International Symposium
on Personal, Indoor and Mobile Radio Communications (PIMRC’93), (Yokohama, Japan, 1993), pp. 109–113
82. A. Chouly, A. Brajal, S. Jourdan, Orthogonal multicarrier techniques applied to direct sequence spread spectrum CDMA systems, in
Proceedings of the IEEE Global Telecommunications Conference 1993 (GLOBECOM ’93), including a Communications Theory Mini-
Conference, 3, (Houston, TX, 1993), pp. 1723–1728
83. K. Fazel, Performance of CDMA/OFDM for mobile communication system, in Proceedings of the 2nd IEEE International Conference on
Universal Personal Communications, 2, (Ottawa, Canada, 1993), pp. 975–979
84. H. Nikookar, R. Prasad, Introduction to Ultra Wideband for Wireless Communications (Springer ScienceþBusiness Media B.V, 2009)
85. M. Ghavami, L.B. Michael, R. Kohno, Ultra Wideband Signals and Systems in Communication Engineering (Wiley, Chichester, 2007)
86. K. Siwiak, D. McKeown, Ultra-Wideband Radio Technology (Wiley, Chichester, 2004)
87. F. Nekoogar, Ultra-Wideband Communications: Fundamentals and Applications (Prentice Hall Press, Upper Saddle River, 2005)
88. A. Batra, J. Balakrishnan, A. Dabak, R. Gharpurey, J. Lin, P. Fontaine, J.-M. Ho, S. Lee, M. Frechette, Y.H. March, Multi-Band OFDM
Physical Layer Proposal for IEEE 802.15 Task Group 3a, Doc (IEEE P802.15-03/268r3, 2004)
89. A. Batra, J. Balakrishnan, G.R. Aiello, J.R. Foerster, A. Dabak, Design of a multiband OFDM system for realistic UWB channel environments.
IEEE Trans. Microwave Theory Tech. 52(9), 2123–2138 (2004)
90. O.-O. Shin, S.S. Ghazzemzadeh, L.J. Greenstein, V. Tarokh, Performance evaluation of MB-OFDM and DS-UWB systems for wireless
personal area networks, in Proceedings of the 2005 IEEE International Conference on Ultra-Wideband, 1, (Zurich, Switzerland, 2005),
pp. 214–219
91. M.Z. Win, R.A. Scholtz, Ultra-wide bandwidth time-hopping spread-spectrum impulse radio for wireless multiple–access communications.
IEEE Trans. Commun. 48(4), 679–691 (2000)
712 10 Spread Spectrum, CDMA, and Ultra-Wideband Communications

92. K.-B. Png, X. Peng, F. Chin, Performance studies of a multi-band OFDM system using a simplified LDPC code, in Proceedings of the 2004
International Workshop on Ultra Wideband Systems & 2004 Conference on Ultrawideband Systems and Technologies (UWBST & IWUWBS),
1, (Kyoto, Japan, 2004), pp. 376–380
93. S.-M. Kim, J. Tang, K.K. Parhi, Quasi-cyclic low-density parity-check coded multiband-OFDM UWB systems, in Proceedings of the 2005
IEEE International Symposium on Circuits and Systems, 1, (Kobe, Japan, 2005), pp. 65–68
94. A.A.M. Saleh, R. Valenzuela, A statistical model for indoor multipath propagation. IEEE J. Sel. Areas Commun. 5(2), 128–137 (1987)
95. G.B. Arfken, H.J. Weber, Mathematical Methods for Physicists, 6th edn. (Elsevier/Academic, Amsterdam/Boston, 2005)
96. D. Slepian, Prolate spheroidal wave functions, Fourier analysis and uncertainty V: The discrete case. Bell Syst. Tech. J. 57, 1371–1430 (1978)
97. I.B. Djordjevic, On the irregular nonbinary QC-LDPC-coded hybrid multidimensional OSCD-modulation enabling beyond 100 Tb/s optical
transport. IEEE/OSA J. Lightw. Technol. 31(16), 2969–2975 (2013)
98. W. Shieh, I. Djordjevic, OFDM for Optical Communications (Elsevier/Academic, Amsterdam/Boston, 2010)
99. J.E. Mazo, Some theoretical observations on spread-spectrum communications. Bell Syst. Tech. J. 58(9), 2013–2023 (1979)
100. Z.-L. Shi, W. Du, P.F. Driessen, A new multistage detector for synchronous CDMA communications. IEEE Trans. Commun. 44(5), 538–541
(1996)
 Physical-Layer Security for Wireless and Optical
Channels 11

Abstract
This chapter is devoted to the physical-layer security. The chapter starts with discussion on security issues, followed by the
introduction of information-theoretic security and comparison against the computational security. In the same section,
various information-theoretic security measures are introduced, including strong secrecy and weak secrecy conditions.
After that, Wyner’s wiretap channel model, also known as the degraded wiretap channel model, is introduced. In the same
section, the concept of secrecy capacity is introduced as well as the nested wiretap coding. Further, the broadcast channel
with confidential messages is introduced, and the secrecy capacity definition is generalized. The focus is then moved to the
secret-key generation (agreement), the source- and channel-type models are introduced, and corresponding secret-key
generation protocols are described. The next section is devoted to the coding for the physical-layer security systems,
including both coding for weak and strong secrecy systems. Regarding the coding for weak secrecy systems, the special
attention is devoted to two-edge-type LDPC coding, punctured LDPC coding, and polar codes. Regarding the coding for
strong secrecy systems, the focus is on coset coding with dual of LDPC codes and hash functions/extractor-based coding.
The attention is then moved to information reconciliation and privacy amplification. In wireless channels physical-layer
security (PLS) section, the following topics are covered: wireless MIMO PLS and secret-key generation in wireless
networks. In section on optical channels PLS, both PLS for spatial division multiplexing (SDM)-fibers-based systems and
free-space optical (FSO) systems are discussed. For better understanding, the set of problems is provided.

11.1 Security Issues

Public key cryptography has several serious drawbacks such as the following: it is difficult to implement it in devices with low
memory and low process constraints, Internet is becoming more and more mobile, security schemes are based on unproven
assumptions of intractability of certain functions, and the assumption of limiting computing resources of the Eve is not always
applicable, to mention few. The open system interconnection (OSI) reference model defines seven layers. However, only five
layers relevant to security issues are provided in Fig. 11.1. The original OSI model does not specify the security issues at all.
The security issues are addressed in X.800 standard (security architecture for OSI) [1]. Even though the physical-layer
security (PLS) is not discussed in this standard, the services specified in these five layers can be enhanced by employing the
PLS as discussed in [2–5]. The PLS scheme can also operate independently. The distortions and noise effects introduced by
channel can be exploited to reduce the number of bits extracted by Eve. Compared to conventional cryptographic approaches
where strong error control coding (ECC) schemes are used to provide reliable communication, the transmission in PLS
scenario needs to be simultaneously reliable and secure. This indicates that different classes of ECC must be developed, and
these codes will be discussed later in this chapter. Alternatively, similarly to QKD, the randomness of the channel can be
exploited to generate the key, and this approach is commonly referred to as the secret-key agreement [3–6], and this concept is
described in Fig. 11.2, inspired by [3, 4, 7].
Alice and Bob monitor the Alice-Bob channel capacity (also known as the capacity of the main channel) CM and the
secrecy capacity CS, defined as a difference between main channel capacity and eavesdropping channel capacity CE. When the

# The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 713
I. B. Djordjevic, Advanced Optical and Wireless Communications Systems, https://doi.org/10.1007/978-3-030-98491-5_11
714 11 Physical-Layer Security for Wireless and Optical Channels

Fig. 11.1 Security mechanisms

at different layers in OSI model. Application End-to-end cryptography
(Only security-relevant layers are
shown)
Transport Secure socket layer (SSL) & Transport layer security (TLS)

Network Virtual private networks, Internet protocol security (IPSec)

Data Link
End-to-end cryptography
Control

Physical Physical-layer security (PLS)

Fig. 11.2 Secret-key generation

(agreement) protocol suitable for Alice Bob
wireless communications Transmission phase
Transmitter Receiver
CS CS,tsh, CM CM,tsh
X Y
Direct information
Systematic Parity reconciliation
LDPC decoder
LDPC encoder bits CS<CS,tsh, CM<CM,tsh
X
X

Privacy Privacy amplification Privacy

amplification stage amplification stage

K K
Key K based
secure communication

secrecy capacity is sufficiently above the threshold value CS,tsh and the main channel capacity is well above threshold value
CM,tsh, Alice transmits Gaussian-shaped symbols X to Bob. When the secrecy capacity and main channel capacity are both
below corresponding thresholds due to deep fading, Alice and Bob perform information reconciliation of previously
transmitted symbols, which is based on strong ECC scheme to ensure that errors introduced by either channel or Eve can
be corrected for. Similar to QKD Schemes [8–15], a systematic low-density parity-check (LDPC) code can be used (that does
not affect information bits but generates the parity-check bit algebraically related to the information bits) to generate the parity
bits and transmit them over an authenticated public channel. There exist direct and reverse information reconciliation
schemes. In direct reconciliation, shown in Fig. 11.2, Alice performs LDPC encoding and sends the parity bits to Bob.
The Bob performs the LDPC decoding to get the correct key X. In reverse reconciliation, Bob performs LDPC encoding
instead. Privacy amplification is then performed between Alice and Bob to distil from X a smaller set of bits K (secure key),
whose correlation with Eve’s string is below a desired threshold. One way to accomplish privacy amplification is through the
use of universal hash functions G [10, 13, 15, 16], which map the set of n-bit strings X to the set of m-bit strings K such that for
any distinct X1 and X2 from the set of corrected keys, when the mapping g is chosen uniformly at random from G, the
probability of having g(X1) ¼ g(X2) is very low.
11.2 Information-Theoretic vs. Computational Security 715

Fig. 11.3 Shannon’s model of Secret key K

secrecy system
ALICE BOB
k-bit message M k-bit decrypted
Encryption Decryption message M
C C
C
Cryptogram
C=f(M,K) Eve

K K
Key generator
…1010… …1010…
Alice Bob
M C Mb
…1110… …0100… …1110…
XOR C XOR
Eve

Fig. 11.4 One-time pad scheme

11.2 Information-Theoretic vs. Computational Security

Shannon’s model of secrecy system [17] is depicted in Fig. 11.3. The purpose of this system is to reliably transmit the message
M, which is at the same time secret for Eve’s point of view. To do so, Alice and Bob have access to the random key K, which is
not known to Eve, and it is used by Alice to encrypt the message into the cryptogram C. On receiver side Bob decrypts the
cryptogram C with the help of key K and determines the transmitted message M.

11.2.1 Information-Theoretic (Perfect) Security

We say that the message is transmitted with perfect security, if given Eve’s knowledge about the cryptogram it does not help
her deciphering the message. In other words, Pr(M | Eve’s knowledge) ¼ Pr(M). Clearly, in perfect security sense, the
codeword C is statistically independent of the message M, and we can write H(M|C) ¼ H(M). Equivalently, the mutual
information between the codeword C and the message M is zero, that is, I(M;C) ¼ H(M|C)-H(M) ¼ 0. In this scenario, the best
strategy for the Eve is to guess the message, and the probability of success is 2-k and decreases exponentially as the key length
k increases. By employing the chain rule from information theory [2, 3], we can write:

H ðM Þ ¼ H ðMjCÞ
 H ðM, KjCÞ ¼
¼ H ðKjCÞ þ H ðMjC, K Þ ð11:1Þ
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}
¼0
¼ H ðKjCÞ ¼ H ðK Þ,

where we used the fact that message is completely specified when both the cryptogram and the key are known, that is,
H(M|K,C) ¼ 0. Therefore, the perfect secrecy condition is given by:

H ðM Þ  H ðK Þ: ð11:2Þ

In conclusion, the entropy (uncertainty) of the key cannot be lower than the entropy of the message, for an encryption
scheme to be perfectly secure.
One-time pad encryption scheme (Vernam cipher) [18, 19], shown in Fig. 11.4, operates by adding mod 2 message M bits
and uniform random key bits K and satisfies the perfect security condition [20]. Namely, the key bits randomize the encoding
716 11 Physical-Layer Security for Wireless and Optical Channels

process and ensure that the statistical distribution of cryptogram is independent of the message. On the other hand, the key
length is the same as the message length, and given complete randomness and statistical independence of the cryptogram and
the message, we have that H(C) ¼ log2k ¼ k ¼ H(M), satisfying the inequality (11.2) with equality sign. Unfortunately,
one-time pad scheme has several drawbacks [10, 14]: (i) it requires the secure distribution of the key, (ii) the length of the key
must be at least as long as the message, (iii) the key bits cannot be reused, and (iv) the keys must be delivered in advance,
securely stored until used, and destroyed after the use.
Given that condition (11.2) is difficult to satisfy, in conventional cryptography, instead of perfect security the computa-
tional security is used [15, 21]-[25].

11.2.2 Computational Security

The computational security introduces two relaxations with respect to information-theoretic security [22--24]:

• Security is guaranteed against an efficient eavesdropper running the cryptanalytic attacks for certain limited amount of
time. Of course, when eavesdropper has enough computational resources and/or enough time, he/she will be able to break
the security of the encryption scheme.
• Eavesdroppers can be successful in breaking the security protocols, but with small success probability.

There are two general approaches to precisely define the computational relaxations [22]:

• The concrete approach, in which an eavesdropper running the eavesdropping algorithm for time interval no more than
T seconds can be successful with probability at most ε. We could call this approach the concept of (T,ε)-security.
• The asymptotic approach, in which a probabilistic polynomial-time eavesdropper caring out an eavesdropping strategy
(cryptanalytic attack), for every polynomial p(n) there exists an integer N such that when n > N, will have the success attack
probability that is smaller than 1/p(n).

A reader interested to learn more about computational security is referred to an excellent book due to Katz and Lindell [22].

11.2.3 Information-Theoretic Secrecy Metrics

Before concluding this section, we describe several relevant information-theoretic secrecy metrics [4, 5, 26]. The generic
metric D(,), measuring the distance between the joint distribution of M and C = [C1,C2,. . .,Ck], denoted as fM,C, and the
product of independent distributions for M and C, denoted as fM and fC, can be used to define the secrecy requirement as
follows:
 
lim D f M,C , f M f C ¼ 0: ð11:3Þ
k!1

As an illustration, the Kullback-Leibler (KL) distance [27], representing the measure of inefficiency of assuming that
distribution is fMfC when true distribution is fM,C, leads to the well-known mutual information:

   
 f M,C
D f M,C  f M f C ¼ E f M,C log ¼ I ðM, CÞ, ð11:4Þ
fM fC

where E denotes an expectation operator. The L1-distance between the joint distribution fM,C and the product of independent
distribution fMfC, known as the variational distance [5], can also be used as the secrecy metric:
 

V f M,C , f M f C ¼ E f M,C

f M,C  f M f C
: ð11:5Þ
11.3 Wyner’s Wiretap Channel 717

Mk Binary data (k bits)

Kk One-time pad (k l bits)

Ck Protected data (k l bits) Unprotected data (l bits)

Fig. 11.5 Illustration of the weak secrecy concept

Given that mutual information I(M,C) measures the average amount of information about message M leaked in C, the
following requirement:

lim I ðM, CÞ ¼ 0 ð11:6Þ

k!1

is commonly referred to as the strong secrecy condition. From practical point of view, given that the strong secrecy condition
is difficult to satisfy, instead of requesting the mutual information to vanish, we can soften the requirement and request that the
rate of information leaked to Eve tends to zero:

1
lim I ðM, CÞ ¼ 0 ð11:7Þ
k!1 k

This average information rate about the message M leaked to C is well known as the weak secrecy condition. To illustrate
the difference, let us consider the following illustrative example, provided in Fig. 11.5. The message M of length k, denoted as
Mk, is to be encrypted. To encrypt the message, we apply the one-time pad on first k-l bits, while the remaining l bits are
unprotected, where l is fixed number. The strong secrecy metric can be expressed as:
   
I M, C k ¼ H ðM Þ  H MjCk ¼ k  ðk  lÞ ¼ l,

and since the mutual information does not tend to zero as k ! 1, the strong secrecy condition is not satisfied. On the other
hand, the weak secrecy metric will be:

1   1   k  ðk  lÞ l
I M, Ck ¼ H ðM Þ  H MjCk ¼ ¼ ,
k k k k

and clearly tends to zero as k ! 1, indicating that weak secrecy condition is satisfied.
From this example we conclude that some secrecy metrics are mathematically stronger than others. For instance, the mutual
information metric is stronger than variational distance secrecy metric, while variational metric is stronger than weak secrecy
metric as shown in [28].

11.3 Wyner’s Wiretap Channel

Shannon’s model is too pessimistic as it assumes that no noise has been introduced during transmission. Wyner introduced the
so-called wiretap channel [29], now also known as a degraded wiretap channel model, in which Eve’s channel is the degraded
version of the Alice-to-Bob channel (main channel), as indicated in Fig. 11.6. Alice encodes the message M into a codeword
Xn of length n and sends it over the noisy channel, represented by conditional probability density function (PDF) f(y|x) toward
the Bob. On the other hand, Eve observes the noisy version of the signal available to Bob. Therefore, the wiretap channel is a
degraded channel represented by the conditional PDF f(z|y). More formally, the discrete memoryless degraded wiretap
channel is specified by input alphabet X ; two output alphabets Y , Z ; and transition probabilities f(y|x) and f(z|y) such that
joint distributions for main and wiretap channels are independent, that is, we can write:

n n
f YZ ðyn zn jxn Þ ¼ ∏ f ðym jxm Þ ∏ f ðzm jym Þ: ð11:8Þ
m¼1 m¼1
718 11 Physical-Layer Security for Wireless and Optical Channels

Fig. 11.6 Wyner’s wiretap

Main
channel model. DMS, discrete DMS U Codeword
memoryless source ALICE channel BOB
Xn Yn decoded
Message M
Encoder f(y|x) Decoder
message MB

f(z|y) Wiretap
channel
Zn

Eve

Wyner suggested to use the equivocation rate, defined as (1/n)H(M|Zn), instead of the entropy of the message H(M). So the
secrecy condition in Wyner’s sense will be:

1 1 1
H ðM Þ  H ðMjZn Þ ¼ I ðM, Zn Þ ! 0, ð11:9Þ
n n n n!1

which is clearly the weak secrecy condition. In addition to secrecy condition, the reliability condition must be satisfied as well:

PrðM B 6¼ MjY n Þ ! 0: ð11:10Þ

n!1

In other words, the probability that Bob’s message is different from the message sent by Alice tends to zero as the
codeword length n ! 1. The channel codes to be used in this scenario must satisfy both reliability and secrecy conditions and
are sometimes referred to as the wiretap codes [30]. For instance, LDPC, polar, and lattice codes can be used to design the
wiretap codes. The (n,k) wiretap code C n of rate R ¼ k/n is specified by [4, 30]: (i) the set of messages M of size 2nR, (ii) the
local random source U with distribution f U , (iii) the encoder performing the mapping M  U ! X n (mapping the message
and a random realization of the local source into a codeword), and (iv) the decoder performing the following mapping
Y n ! M [{Er} (mapping the received word into a message or an error Er).
The transmission rate R-equivocation rate Req pair (R,Req) is achievable in Wyner’s sense for the sequence of codes {C n} of
code rate R if both reliability and secrecy conditions are satisfied, defined, respectively, as:

PrðM B 6¼ MjC n Þ ¼ Pe ðC n Þ ! 0,
n!1
1 ð11:11Þ
lim H ðMjZn C n Þ ¼ lim ERðC n Þ  Req ,
n!1n
|fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} n!1
ERðC n Þ

where the first line represents the probability of error when wiretap code C n is used, while the second term represents the
equivocation rate (ER). The largest transmission rate at which both reliability and secrecy conditions are simultaneously
satisfied is commonly referred to as the secrecy capacity. For any distribution fx of X from set of distributions P (R  0) for
which I(X,Y)  R, Wyner has defined the merit function, which can be called a secrecy rate, as follows:

SRðRÞ ¼ sup I ðX, YjZ Þ: ð11:12Þ

f x 2P ðRÞ

Since random variables X, Y, Z form the Markov chain, we can apply the chain rule to obtain:

I ðX, YjZ Þ ¼ H ðXjZ Þ  H ðXjY, Z Þ ¼

|fflfflfflfflfflffl{zfflfflfflfflfflffl}
H ðXjY Þ ð11:13Þ
¼ I ðX, Y Þ  I ðX, Z Þ,

so that the alternative expression for C(R) is obtained by:

11.3 Wyner’s Wiretap Channel 719

SRðRÞ ¼ sup ½I ðX, Y Þ  I ðX, Z Þ: ð11:14Þ

f x 2P ðRÞ

Wyner further proved the following lemma [29]: The secrecy rate SR(R), 0  R  Cm (Cm is the capacity of the main
channel) satisfies the following properties:

• The supremum in definitions above is in fact maximum, i.e., for each R there exists distribution from P ðRÞ such that I(X,Y|
Z ) ¼ SR(R).
• SR(R) is concave and continuous function of R.
• SR(R) is nonincreasing in R.
• SR(R) is upper bounded by Cm and lower bounded by Cm-Ce, where Ce is the capacity of the main-wiretap channel cascade;
that is, Cm-Ce  SR(R)  Cm.

Wyner also proved the following theorem [29]: The achievable region R is determined by:
  
R ¼ R, Req : 0  R  C m , 0  ER  H ðM Þ, RReq  SRðRÞH ðM Þ : ð11:15Þ

The typical achievable region is provided in Fig. 11.7. As expected from the lemma, the achievable region is not convex,
and there are two critical corner points: (Cm,Hm ¼ SR(R)H(M)/Cm) and (CM, H(M)). The secrecy capacity can now be
defined as:

CS ¼ max R: ð11:16Þ
ðR, H ðM ÞÞ2R

Leung-Yan-Cheong has shown in [31] that the secrecy capacity is equal

 to the lower  bound in lemma, that is, CS ¼ Cm-Ce,
when both main and wiretap channels are weakly symmetric. The channel X , Y , f Y jX is weakly symmetric when the rows of
the channel transition matrix are permutations of each other, while the summations per column are independent of y. The
binary symmetric channel (BSC) is a weakly symmetric channel. As an illustration, let us assume that both main and wiretap
channels can be both modeled as binary symmetric channels, as shown in Fig. 11.8, but with two different error probabilities
( p and q). From Chap. 4 we know that the capacity of BSC is given by Cm ¼ 1-h( p), where h( p) is the binary entropy function
h( p) ¼ p log p-(1-p)log p. Given that cascade of two BSCs is also a BSC with equivalent error probability of p + q-2pq, the
secrecy capacity for this case will be:

C S ¼ Cm  Ce ¼ 1  hðpÞ  ½1  hðp þ q  2pqÞ

¼ hðp þ q  2pqÞ  hðpÞ:

If we instead use the strong secrecy condition, as it was done in [4], defined as follows:

Req

H(M) Achievable
region
Hm

R
CM Cm
Fig. 11.7 Typical achievable region
720 11 Physical-Layer Security for Wireless and Optical Channels

BSC(p)

x0 1 p
y0
p p

x1 y1
1 p
Main channel Y
X
Alice BSC(p) Bob

BSC(q) Wiretap
channel
Z

Eve

Fig. 11.8 The wiretap channel model when both main and wiretap channels are BSCs

Req

I(X,Y|Z)

Hm Achievable region for

strong secrecy
R
I(X,Y|Z) I(X,Y)

Fig. 11.9 The typical rate-equivocation achievable region in strong secrecy scenario

 
lim H ðMjZn C n Þ  nReq  0,
n!1
ð11:17Þ

together with the same reliability condition, the corresponding achievable rate-equivocation region R 0 becomes convex and
can be defined as a sequence of (n,nR) codes {C n } satisfying both strong secrecy constraint and reliability constraint and be
determined by [4, 32]:
  
R0 ¼ R, Req : 0  Req  R  I ðX, Y Þ, 0  Req  I ðX, YjZ Þ : ð11:18Þ

The typical shape of achievable region in strong secrecy scenario is provided in Fig. 11.9. As long as we are below the I(X,
Y|Z ), it is possible to find codes satisfying both full transmission and equivocation rates. However, above I(X,Y|Z ) the
equivocation rate saturates, while the transmission rate can be increased up to I(X,Y ). The secrecy capacity in strong secrecy
scenario can be determined by [4]:

CS ¼ max ½I ðX, Y Þ  I ðX, Z Þ: ð11:19Þ

fx

For corresponding proofs of (11.18) and (11.19), an interested reader is referred to [4].
Before completing this section, we discuss the wiretap codes with more details. Let us observe an example of the wiretap
channel model, mentioned in Wyner’s paper [29], in which the main channel is noiseless, while the wiretap channel is the
binary erasure channel (BEC) with erasure probability α as shown in Fig. 11.10. The BEC has two inputs x0 and x1 and three
outputs y0, y1, and Er. The output symbol Er is called the erasure. A fraction α of the incoming bits get erased by the channel.
The channel capacity of such channel is 1-α. The secrecy capacity for this wiretap channel model will be then CS ¼ 1-
(1-α) ¼ α. Therefore, out of n transmitted bits, nα bit will be erased. Clearly since the secrecy capacity is lower than 1, Eve is
capable to learn the content of relevant number of bits. To solve for this problem, Alice can apply the stochastic encoding as
follows. Let us assume that Alice wants to transmit a binary message M2{0,1}, by encoding the message zero as codeword
xn0 ¼ ðx00 , x01 , ⋯, x0,n1 Þ, while encoding the message one to codeword xn1 ¼ ðx10 , x11 , ⋯, x1,n1 Þ. Since these codewords get
transmitted over the noiseless channel, Bob will receive the codeword correctly and will be able to conclude which message
11.4 Broadcast Channel with Confidential Messages and Wireless Channel Secrecy Capacity 721

X BEC( )
Alice Bob 1
x0 y0
BEC( ) Wiretap
channel Er
Z
x1 y1
Eve 1

Fig. 11.10 An example of wiretap channel model in which the main channel is noiseless, while the wiretap channel is a BEC

xn(1,1) xn(1,2) … xn(1,2nRU)

indexed by M
2nR subcodes
xn(2,1) xn(2,2) … xn(2,2nRU)

…
…

…
…
xn(2nR,1) xn(2nR,2) xn(2nR, 2nRU)

2nRU codewords within the subcode indexed by U

Fig. 11.11 Illustrating the nested structure of a wiretap code

was transmitted by simply taking the parity check. However, since Eve’s channel is BEC and in her received word there will
be nα erasures, so she will not be able to determine what was transmitted by taking the parity check. So the conditional entropy
will be lower bounded by:

H ðMjZ n Þ  1  ð1  αÞn ,

and tends to the entropy of the message 1 as n ! 1 (for sufficiently small erasure probability). On the other hand, the code
rate of such stochastic encoding scheme is only 1/n. However, this concept can be generalized as shown in Fig. 11.11. For
each of 2nR possible messages indexed by M, there exist 2nRU codewords indexed by a local random generator U. The set of
codewords corresponding to a message forms a subcode of the wiretap code. Given that the main channel is noiseless, we need
to ensure that the total code rate R + RU  1.

11.4 Broadcast Channel with Confidential Messages and Wireless Channel Secrecy Capacity

We first define the broadcast channel with confidential messages.

11.4.1 Broadcast Channel with Confidential Messages

Wyner’s wiretap channel gets generalized and refined by Csiszár and Körner [32], and the corresponding model, now known
as the broadcast channel with confidential messages (BCC), is provided in Fig. 11.12. The broadcast channel is assumed to be
discrete and memoryless and characterized by input alphabet X , output alphabets Y and Z (corresponding to Bob and Eve,
respectively), and transition PDF f(yz|x). So, the channel itself is modeled by a joint PDF for Bob’s and Eve’s observations,
f(yz|x), given the channel input. Since the BCC is a memoryless channel, the PDF can be decomposed as:

n
f ðyn zn jxn Þ ¼ ∏ f YZjX ðym zm jxm Þ: ð11:20Þ
m¼1

In this scenario, Alice wishes to broadcast a common message Mc to both Bob and Eve and a confidential message M to
Bob. The corresponding stochastic code C n of codeword length n is composed of:
722 11 Physical-Layer Security for Wireless and Optical Channels

Fig. 11.12 The broadcast

ALICE
channel model with confidential Mc
messages (BCC) Codeword BOB
M Yn
Xn Decoded
Encoder f(yz|x) Decoder
message MB
Zn
• Confidential message M for Bob Eve
• Common message Mc for both
Bob and Eve

• Two message sets: the common message set M c of size 2nRc and the confidential message set M of size 2nR.
• The encoding (stochastic) function en: M c  M ! X n , which maps the message pair (mc,m)2 M c  M into a codeword
xn2X n .
• The decoding functions dn: Y n ! M c  M , dE,n: Z n ! M c , with the first one mapping the observation vector yn to a
message pair ðm b Þ, while the second one the observation zn to common message estimate m
b c, m b c.

Similarly to Wyner’s approach, the confidential message M secrecy with respect to Eve is measured in terms of the
equivocation rate as follows:

1
ERðC n Þ ¼ H ðMjZn C n Þ: ð11:21Þ
n

The probability of error is defined in similar fashion as for Wyner’s approach:

   
Pe ðC n Þ ¼ Pr d n ðY n Þ ¼ Mb c, M
b 6¼ ðM c , M Þor dE,n ðY n Þ ¼ M
b c 6¼ M c jC n ð11:22Þ

We further request that the (stochastic) code C n must satisfy both reliability and secrecy constraints, specified,
respectively, as:

lim P ðC n Þ ¼ 0, lim ERðC n Þ  Req : ð11:23Þ

n!1 e n!1

The rate tuple (Rc,R,Req) for the BCC is achievable if and only if there exists a (stochastic) wiretap code Cn of length
n satisfying simultaneously both reliability and security constraints, defined above.
Csiszár and Körner proved the following theorem [32] (see also [4]): For the joint distribution fUVX on U  V  X that
factorizes as fUfU|VfX|V, the region of achievable rate tuples (Rc,R,Req), denoted as R , is determined by:
8
> 0  Req  R
>
>
<   Req  I ðV, YjU Þ  I ðV, ZjU Þ
R ¼ [ Rc , R, Req : ð11:24Þ
f U f V jU f XjV >
> R þ Rc  I ðV, YjU Þ þ min ½I ðU, Y Þ, I ðU, Z Þ
>
:
0  Rc  min ½I ðU, Y Þ, I ðU, Z Þ

U and V are auxiliary random variables so that random variables U, V, X, and YZ form the Markov chain U ! V ! X ! YZ.
Finally, the cardinalities of U and V maybe limited to |U|  |X | + 3 and |V |  |X |2 + 4|X | + 3.
Csiszár and Körner provided the following corollary [32] (see also [4]): Secrecy capacity is determined as the difference of
mutual information for Alice-Bob and Alice-Eve links, when the rate of the common message is set to zero, that is:

Cs ¼ max ½I ðV, Y Þ  I ðV, Z Þ, ð11:25Þ

f VX
V!X!YZ

where the maximization is performed over all possible joint distributions fVX(v,x) and V, X, and YZ form a Markov chain
V ! X ! YZ.
11.4 Broadcast Channel with Confidential Messages and Wireless Channel Secrecy Capacity 723

Clearly, the secrecy capacity is strictly positive when Bob’s channel is less noisy than Eve’s channel, i.e., I(X;Y ) > I(X;Z ).
Namely, by setting V ¼ X, the secrecy capacity expression becomes C s ¼ max ½I ðX, Y Þ  I ðX, Z Þ, which is clearly strictly
fX
positive when I(X;Y ) > I(X;Z).
Similarly as in Wyner’s wiretap channel case, the secrecy metric can be replaced with strong secrecy metric.

11.4.2 Wireless Channel Secrecy Capacity

Wireless communication scenario is illustrated in Fig. 11.13. Due to fading effects [33–35], the signal-to-noise ratios (SNRs)
in both main and Eve’s channels are time-varying. For certain time periods, the SNR in main channel is higher than that in
Eve’s channel and vice versa. Moreover, if Alice and Bob have access to Eve’s channel state information (CSI), they can
exploit this knowledge to get an advantage over Eve.
As indicated in [36, 37], the general goal is to maximize the transmission rate between Alice and Bob R ¼ H(Mk)/n, while
minimizing the Eve’s equivocation rate ER ¼ H(Mk| Zn)/H(Mk), wherein the error probability tends to zero. Secrecy capacity
is defined as the maximum transmission rate R for which Eve’s equivocation rate is equal to 1. The generic fading wireless
BCC model is provided in Fig. 11.14. Clearly, both main and Eve’s channels are simultaneously additive and multiplicative.
The additive noise is a zero-mean Gaussian, with variance in main (Alice-Bob) channel being σ 2m and variance in Eve’s
channel being σ 2e .
The multiplicative components are described by (i) main channel coefficients hmn ¼ [hm [1], . . ., hm[n]] and (ii) Eve’s
channel coefficients hen ¼ [he [1], . . ., he[n]]. For ergodic-fading channel model, channel coefficients change every channel
use, that is, hm[i] and he[i] (i ¼ 1,. . .,n) are mutually independent. For quasi-static channel model, employed in the rest of this
section, the channel coefficients do not change for duration of whole codeword, that is, hm ¼ hm[i], he ¼ he[i]. The block-
fading channel model describes the scenario between the two cases just described, in which the channel coefficients do not
change for certain number of symbols, but still change many times during the codeword transmission. For additional details
on wireless channel models, an interested reader is referred to [33–35]. The SNRs for main channel, denoted as ρm, and Eve’s
channel, denoted as ρe, assuming quasi-static fading channel are simply:

ρm ¼ jhm j2 P=σ 2m ¼ gm P=σ 2m , ρe ¼ jhe j2 P=σ 2e ¼ ge P=σ 2e , ð11:26Þ

|ffl{zffl} |ffl{zffl}
gm ge

SNR

Alice-Eve link
Alice-Bob link

Time

Fig. 11.13 Illustration of the wireless communication scenario

Fig. 11.14 The generic wireless hmn

BCC wmn
ALICE BOB
Mk Xn Yn ˆk
M
Encoder Decoder

Zn
Decoder
Eve
hen wen
724 11 Physical-Layer Security for Wireless and Optical Channels

wherein P is the transmit power, gm is the gain coefficient in main channel (gm ¼ |hm|2), and ge is the gain coefficient in Eve’s
channel (ge ¼ |he|2). Based on discussion related to (11.25), the instantaneous secrecy capacity for quasi-static fading channel
is determined by:

log ð1 þ ρm Þ  log ð1 þ ρe Þ, ρm > ρe
CS ¼ ð11:27Þ
0, otherwise

The average secrecy capacity for quasi-static channel can be obtained by averaging out the gain coefficients as follows:

 S ¼ E g ,g fC s ðgm , ge Þg,
C m e

h  gm P
 
ge P
 þ ð11:28Þ
C s ðgm , ge Þ ¼ log 1 þ 2  log 1 þ 2 ,
σm σe

where [x]+ ¼ max(0,x). For formal proof, an interested reader is referred to [36].
Another relevant figure of merit related to secrecy, important in wireless communications, is the probability of positive
secrecy capacity defined as:

Pþ
S ¼ Pr ðC S > 0Þ: ð11:29Þ

11.5 Secret-Key Generation (Agreement) Protocols

Alice and Bob can employ the noisy channel to generate correlated random sequences and subsequently use a public
authenticated error-free feedback channel to agree on a secret key [38–41], and this approach is commonly referred to as
the secret-key agreement [3–6, 42]. Two types of models are typically considered for secret-key agreement [39]:

• Source-type model, in which terminals observe the correlated output of the source of randomness without having control
of it.
• Channel-type model, in which one terminal transmits random symbols to other terminals using a broadcast channel. This
scenario is similar to the wiretap channel model with feedback channel, which is an authenticated public noiseless channel.

Both of these models are very similar to QKD [8–15], except that raw key is transmitted over classical channel, while in
QKD over the quantum channel. The source-type scenario is illustrated in Fig. 11.15. Even if the feedback message F is
available to Eve, it is still possible to generate secure key such that H(K|ZnF) is arbitrarily close to H(K). In this specific
example, Alice, Bob, and Eve get correlated noisy observations, Xn, Yn, and Zn, respectively. Alice and Bob are able to
generate the common key K based on their observations and a set of feedback messages F. Even if Eve’s channel is better than
Bob’s channel, it is possible to achieve positive secrecy capacity, as explained by Maurer [38]. The corresponding procedure
that Alice and Bob need to follow to determine the secret key, can be called secret-key agreement (generation) protocol. The
rate at which secret key is generated can be called the same way as in QKD, the secret-key rate (SKR). If the protocols exploit
the public messages sent in one direction only (from either Alice to Bob or Bob to Alice), the corresponding secret-key rate
(SKR) is said to be achievable with one-way communication; otherwise, the SKR is said to be achievable with two-way
communication.

Fig. 11.15 The generic source

model secret-key generation
K Key Xn Yn Key K̂
(agreement) scenario
Correlation
generation generation
Alice Zn Bob

Eve
F
Feedback message F
Public authenticated noiseless channel
11.5 Secret-Key Generation (Agreement) Protocols 725

Fig. 11.16 The generic channel

model secret-key generation ALICE
BOB
(agreement) scenario
K Yn K̂
Xn
Encoder f(yz|x) Decoder
Zn

Eve
Feedback message F
Public authenticated noiseless channel

The generic channel-type scenario is described in Fig. 11.16, which is clearly the generalization (extension) of the Csiszár-
Körner wiretap model. The channel is characterized by conditional PDF f(yx|x), and its input is controlled by Alice by
generating n symbols Xn. Bob and Eve observe the channel outputs and get Yn and Zn realizations. Additionally, Alice and
Bob communicate over two-way authenticated noiseless channel of unlimited capacity. By exchanging the random symbols
over the channel and feedback symbols F over the public channel, Alice and Bob are able to generate the secret key not known
to Eve. To get the secret key, Alice and Bob apply a set of steps that can also be called the secret-key generation protocol.
This protocol should be properly designed such that in addition to reliability condition, defined as:
 
b ! 0,
Pe ¼ Pr K 6¼ K ð11:30Þ
n!1

and secrecy condition, defined as:

L ¼ I ðK, Zn FÞ ! 0, ð11:31Þ
n!1

the uniformity condition is also satisfied, which is defined as:

 

U ¼
H ðK Þ  log 2nR
! 0, ð11:32Þ
n!1

where R is the SKR. As before, the reliability condition tells us that the Alice and Bob must agree on secret key with the high
probability. The security condition is important to ensure the key Alice and Bob finally agreed is indeed secret with respect to
the Eve. Finally, the uniformity condition ensures that the secret key is uniformly distributed within the corresponding
keyspace. The corresponding parameters defined above are known as probability of error (Pe), the leakage of information (L ),
and uniformity of the keys (U ), respectively. We say that the secret-key rate R is achievable if there exist a sequence of secret-
key generation protocols satisfying all three constraints as n ! 1. The supremum of achievable SKRs is commonly referred
to as the secret-key capacity, denoted here as CSK. Given two-way communication over the authenticated public channel, it is
difficult to derive an exact expression for CSK; however based on [38, 39], it can be bounded from both sides as follows:

max max ½I ðX, Y Þ  I ðX, Z Þ, I ðY, X Þ  I ðY, Z Þ  C SK  max ½I ðX, YjZ Þ ð11:33Þ
fX fX

The upper-bound term indicates the secret-key capacity when Bob has access to Eve’s observations. The lower-bound term
max[I(X,Y )-I(X,Z )] indicates that direct reconciliation is employed, while the lower-bound term max[I(Y,X)-I(Y,Z)] indicates
that reverse reconciliation is employed instead.
Let us now provide the following example due to Maurer [38] to illustrate how the noiseless feedback channel can help
getting positive secrecy capacity even when Eve’s channel is less noisy. We assume that both main and Eve’s channels are
BSCs with bit error probabilities pm and pe, respectively. Alice transmits bit X over BSC( pm) to Bob, and this transmission
was received by Eve over BSC( pe). The errors introducedL by channels can be represented
L by corresponding Bernoulli random
variables Em and Ee, added mod 2 to X to get Y ¼ X Em at Bob’s side and Z ¼ X Ee on Eve’s side. Let us assume Bob’s
channel is noisier than Eve’s channel, that is, Pr(Em ¼ 1) > Pr(Ee ¼ 1). Bob employs the feedback channel as follows. To send
726 11 Physical-Layer Security for Wireless and Optical Channels

Fig. 11.17 The illustration of the X Y V Em V

channel-type secret-key Alice BSC(pm) Bob Alice BSC(pm) Bob
generation scheme (on the left)
and an equivalent wiretap channel Z V Em Ee
model (on the right) F BSC(pe) Eve
BSC(pe) Eve

a bit V to Alice, Bob first adds the noisy observation Y, to obtain F¼V  Y ¼ V  X  Em, and sends it over the authenticated
public channel as a feedback bit F. Since Alice already knows X, she adds it to the received F to obtain
F  X ¼ V  Y  X ¼ V  Em. On the other hand, Eve’s estimation will be F  Z ¼ V  Y  Z ¼ V  X  Em  Z ¼ V  Em  Ee,
Y  Z ¼ V  X  Em  Z ¼ V  Em  Ee, which is more noisy than Bob’s channel given that equivalently V was sent over
cascade of main BSC and Eve’s BSC, as illustrated in Fig. 11.17. In the absence of feedback, the schemes become the wiretap
channel model, and we have already shown that secrecy capacity of such channel will be:

C S ¼ ½C m  C e þ ¼ ½1  hðpm Þ  ð1  hðpe ÞÞþ ¼ ½hðpe Þ  hðpm Þþ : ð11:34Þ

So if pm > pe, clearly, the secrecy capacity will be zero. On the other hand, when feedback is used, based on Fig. 11.17
(right), we conclude that corresponding secrecy capacity will be:
h iþ
ðF Þ
C S ¼ CðmFÞ  C ðeF Þ ¼ ½1  hðpm Þ  ð1  hðpm þ pe  2pm pe ÞÞþ
ð11:35Þ
¼ ½hðpm þ pe  2pm pe Þ  hðpm Þþ ,

which is strictly positive unless pm ¼ 1/2, pe ¼ 0.

To distill for a shorter shared secret key, Bennet [16] (see also [6, 10–15]) proposed the following three-phase approach:

• Advantage distillation phase, in which Alice and Bob out of a set of correlated observations select the positions over which
they have an advantage over the Eve. In QKD this step is also known as a sifting procedure.
• Information reconciliation phase, in which these “advantage” observations are further processed to reconcile the
discrepancies by employing the error correction.
• Privacy amplification phase, in which Alice and Bob perform further processing of corrected key to remove redundancy
and get a uniformly distributed key sequence with no leakage to Eve, and this is done with the help of hash functions, as
discussed in introductory part of this chapter.

We can extend these three phases with the 0th phase representing the raw key transmission, which in the source-type model
can be called the randomness sharing [4]. Now we provide more details on both source-type and channel-type secret-key
agreement scenarios.

11.5.1 Source-Type Secret-Key Generation

The source model for secret-key agreement scenario is provided in Fig. 11.18. Here we use the interpretation due to Bloch and
Barros [4], which is slightly different than originally proposed in [39], but it has some similarities with [16]. This protocol is
very similar to the entanglement-assisted QKD [13–15]. The main point of this scenario is discrete memoryless source (DMS),
not controlled by involved parties, described by joint PDF fXYZ, with three output ports X , Y , Z corresponding to Alice, Bob,
and Eve, respectively. In other words, Alice, Bob, and Eve have access to correlated n-component realizations Xn, Yn, and Zn,
governed by the joint PDF. Additionally, Alice and Bob communicate over two-way public authenticated noiseless channel of
unlimited capacity. This side channel does not contribute to the secret key, but helps Alice and Bob to distill for the shorter
secure key. As mentioned above, this feedback channel is authenticated to prevent Eve tampering with the key. Finally, Alice
and Bob employ the local sources of randomness, denoted as R x and R y , respectively, to randomize the message they
transmit. These sources of randomness are described by corresponding PDFs fRx and fRy. As already indicated above, the set of
steps Alice and Bob need to follow constitute the secret-key generation protocol, which can also be called the key-distillation
protocol (strategy) [4].
11.5 Secret-Key Generation (Agreement) Protocols 727

Fig. 11.18 The source model for

secret-key generation (agreement) RX RY
DMS
scenario K Encoder/ Xn Yn Encoder/ K̂
fXYZ
decoder decoder
Alice Zn Bob
Am Eve
Bm

The generic key-generation (key-distillation) protocol Pn for DMS, denoted as (XYZ , fXYZ), of secret-key rate R and
codeword length n, with m rounds over the authenticated public channel, employs:
The following alphabets:

• A key alphabet K of cardinality 2nR.

• Alice alphabet A used to communicate over the authenticated public channel.
• Bob alphabet B used to communicate over the authenticated public channel.

The following sources of the local randomness:

• A source of local randomness for Alice (R X , fRY).

• A source of local randomness for Bob (R Y , fRY).

The following encoding functions:

ðAÞ
• The m Alice encoding functions ei :

ðAÞ
ei : X n  Bi1  RX ! A, i 2 f1, 2, ⋯, mg ð11:36Þ

ðBÞ
• The m Bob encoding functions ei :

ðBÞ
ei : Y n  Ai1  RY ! B, i 2 f1, 2, ⋯, mg ð11:37Þ

The following key-distillation functions:

• Alice key-distillation function kA, defined as the following mapping:

k A : X n  Bm  RX ! K ð11:38Þ

• Bob key-distillation function kB, defined as the following mapping:

k B : Y n  Am  RY ! K ð11:39Þ

The key-distillation protocol Pn is composed of the following steps:

• Alice observes n realizations of the source xn, while Bob observes yn.
• Alice generates local random realization rx, while Bob generates ry.
• In round i2[1,m], Alice transmits ai and Bob bi by:

ðAÞ  n  ðBÞ  
ai ¼ e i x , bi1 , r x , bi ¼ ei yn , ai1 , r y ; ð11:40Þ
728 11 Physical-Layer Security for Wireless and Optical Channels

• After round m Alice computes key kA and Bob key kB by:

k ¼ k A ðxn , bm , r x Þ, ^
k ¼ kB ðyn , am , r y Þ: ð11:41Þ

• Alice and Bob perform information reconciliation, by employing the error correction, to correct for the discrepancies in
symbols at different positions on the raw keys.
• Finally, Alice and Bob perform privacy amplification to remove any correlation with the Eve.

At the end of protocol Pn, Alice and Bob end with the common key kc ¼ b
kc , which is secret with respect to the Eve.
This source-type key-generation protocol Pn is evaluated in terms of:

• The average probability of error:

 
b n ,
Pe ðPn Þ ¼ Pr K 6¼ KjP ð11:42Þ

• The information leakage to the Eve:

LðPn Þ≜ IðK; Zn Ar Br jPn Þ, ð11:43Þ

• The uniformity of the key:

 
U ðPn Þ≜ log 2nR  HðKjPn Þ: ð11:44Þ

Similarly as before, we can define both strong and weak SKR. A strong secret-key rate R is achievable for a source/channel
model if there exist a sequence of key-distillation protocols {Pn} of length n ¼ 1,2,. . . such that the reliability, secrecy, and
uniformity conditions for {Pn} are simultaneously satisfied; in other words the following constraints are valid:

Pe ðPn Þn!1
! 0, LðPn Þn!1
! 0, U ðPn Þn!1
! 0: ð11:45Þ

On the other hand, a weak secret-key rate R is achievable for a source/channel model if there exist a sequence of
key-distillation protocols {Pn} of length n ¼ 1,2,.. such that the reliability constraint, weak secrecy constraint, and weak
uniformity constraint are simultaneously satisfied. The reliability constraint is the same as in strong SKR case, while weak
secrecy and uniformity constraints are defined, respectively, as follows:

LðPn Þ U ðP n Þ
! 0, ! 0: ð11:46Þ
n n!1 n n!1

The corresponding keys are weak secret keys. Given that we introduced both weak and strong secrecy constraints, we can
also define the weak and strong secret-key capacity. The weak secret-key capacity of a source-type model with joint PDF fXYZ
is defined as:

ðweakÞ
CSK ≜ supfR : R is an achievable weak SKRg ð11:47Þ

On the other hand, the strong secret-key capacity of a source-type model with joint PDF fXYZ is defined as:

ðstrongÞ
C SK ≜ supfR : R is an achievable strong SKRg ð11:48Þ
11.5 Secret-Key Generation (Agreement) Protocols 729

The following theorem has been proved by Maurer, Ahlswede, and Csiszár [38, 39] (see also [4]): The weak secret-key
capacity of a source-type model (XYZ, fXYZ) satisfies the following lower and upper bounds:

ðweakÞ
max ðI ðX; Y Þ  I ðX; Z Þ, I ðY; X Þ  I ðY; Z ÞÞ  C SK  min ðI ðX; Y Þ, I ðX, YjZ ÞÞ ð11:49Þ

The lower bound can also be rewritten as I(X; Y )  min (I(X; Z ), I(Y; Z )). Moreover, for the one-way communication, the
following SKR is achievable [38, 39]:

ðone‐way communicationÞ
C SK ¼ max ðI ðX; Y Þ  I ðX; Z Þ, I ðY; X Þ  I ðY; Z ÞÞ ð11:50Þ

Proof. As discussed above, the lower bound corresponds to the Csiszár-Körner wiretap channel model with a feedback. We
have also shown an example related to Fig. 11.17 that this channel model can be equivalently represented as Wyner’s channel
model, by following the similar procedure explained in text related to Fig. 11.17, which starts with Alice observing the
channel realization x and adding uniformly generated integer u mod-|X |. We know that the secrecy capacity of Wyner’s
channel model is higher than I(X,Y )-I(X,Z ), that is, we can write:

ðweakÞ
CSK  I ðX; Y Þ  I ðX; Z Þ: ð11:51Þ

Since the communication over authenticated public channel is two-way, the roles of Alice and Bob can be reversed by
applying the reverse reconciliation so that the secret-key capacity is now lower limited by:

ðweakÞ
CSK  I ðY; X Þ  I ðY; Z Þ: ð11:52Þ

By combining these two inequalities, we obtain the lower bound of secret-key capacity as follows:

ðweakÞ
C SK  max ðI ðX; Y Þ  I ðX; Z Þ, I ðY; X Þ  I ðY; Z ÞÞ: ð11:53Þ

To determine the upper bound, we observe the scenario in which Bob has access to Eve’s observations and can create new
random variably Y0 ¼ YZ. Since X, Y0 , and Y form the Markov chain, we can apply the chain rule to obtain:

I ðX, YjZ Þ ¼ I ðX, YZ Þ  I ðX, Z Þ ¼ I ðX, Y Þ þ I ðX, ZjY Þ  I ðX, Z Þ

|fflfflfflffl{zfflfflfflffl} |fflfflfflfflffl{zfflfflfflfflffl}
I ðX, Y ÞþI ðX, ZjY Þ 0 ð11:54Þ
¼ I ðX, Y Þ  I ðX, Z Þ,

which is clearly the upper bound of secrecy capacity. Moreover, from this equation we conclude that in this scenario
I(X, Y| Z)  I(X, Y). However, since the participants do not have control of DMS, there are realizations for which I(X,
Y ) > I(X,Y|Z ). Let us take the following example due to Maurer [38]. When X, Y are binary independent with Pr(X ¼ 0) ¼ Pr
(Y ¼ 0) ¼ 1/2 and Z is formed by Z ¼ X  Y, we have that I(X,Y ) ¼ 0 and I(X,Y|Z ) ¼ H(Z ) ¼ 1. So to account for both
scenarios, our upper bound becomes:

ðweakÞ
C SK  min ðI ðX, Y Þ, I ðX, YjZ ÞÞ: ð11:55Þ

11.5.2 Channel-Type Secret-Key Generation

The channel model for secret-key agreement scenario is provided in Fig. 11.19. Here we again use the interpretation due to
Bloch and Barros [4] for consistency. In this scenario, Alice provides the input X to the discrete memoryless channel (DMC)
described by the conditional PDF fYZ|X. Bob and Eve do not have any control of the channel outputs; they can just observe
random realizations at corresponding output ports. We again assume that Alice and Bob can communicate over authenticated
730 11 Physical-Layer Security for Wireless and Optical Channels

Fig. 11.19 The channel model

for secret-key generation RX RY
DMC
(agreement) scenario K Encoder/ Xn Yn Encoder/ K̂
fYZ|X
decoder decoder
Alice Zn Bob
Am Eve
Bm

public noiseless channel to which Eve can have access. As before, we assume that Alice and Bob employ the local sources of
randomness, denoted as R x and R y , respectively, to randomize their communication.
In analogy with the key generation protocols for DMS, we can define the corresponding generic protocol for DMC, which
is briefly described in text related to Fig. 11.1. This scenario is very similar to weak coherent states-based QKD [13–15]. The
generic key-generation (key-distillation) protocol Pn for DMC, described by the conditional PDF fYZ|X, of secret-key rate R and
codeword length n, with m rounds over the authenticated public noiseless channel and n transmissions over DMC, employs:
The following alphabets:

• A key alphabet K of cardinality 2nR.

• Alice alphabet A used to communicate over the authenticated public channel.
• Bob alphabet B used to communicate over the authenticated public channel.

The following sources of the local randomness:

• A source of local randomness for Alice (R X , fRY).

• A source of local randomness for Bob (w, fRY).

The following encoding functions to be used over the public channel:

ðAÞ
• The m Alice encoding functions ei :

ðAÞ
ei : Bi1  RX ! A, i 2 f1, 2, ⋯, mg ð11:56Þ

ðBÞ
• The m Bob encoding functions ei :

ðBÞ
ei : Y j  Ai1  RY ! B, i 2 f1, 2, ⋯, mg ð11:57Þ

The index j is used to denote the number of available DMC channel outputs up to now.
The following encoding functions to be used over the DMC:

ðDMC Þ
el : Bk  RX ! X, l 2 f1, 2, ⋯, ng ð11:58Þ

The index k is used to denote the number of bits received from Bob over the public channel up to now.
The following key-distillation functions:

• Alice key-distillation function kA, defined as the following mapping:

k A : X n  Bm  RX ! K ð11:59Þ
11.5 Secret-Key Generation (Agreement) Protocols 731

• Bob key-distillation function kB, defined as the following mapping:

k B : Y n  Am  RY ! K ð11:60Þ

The key-distillation protocol Pn is composed of the following steps:

• Alice generates local random realization rx, while Bob generates ry.
• In round i2[1,m], Alice transmits ai and Bob bi by:

ðAÞ  i1  ðBÞ  

ai ¼ e i b , r x , bi ¼ ei ai1 , r y ; ð11:61Þ

• In the l-th DMC transmission interval, Alice transmits symbol xl over DMC:

ðDMC Þ  k 
xl ¼ el b , rx , ð11:62Þ

Bob receives yl and Eve receives zl. (Once more, k denotes the number of bits received from Bob over the public channel up
to now.) The transmission over DMC takes place when channel conditions are favorable.

• After all DMC and public channel transmissions are completed, Alice computes key kA and Bob key kB by:

 
k ¼ k B ðxn , bm , r x Þ, b
k ¼ k B y n , am , r y : ð11:63Þ

• Alice and Bob perform information reconciliation, by employing the error correction to correct for the errors introduced by
the DMC and Eve.
• Finally, Alice and Bob perform information reconciliation to remove any correlation with the Eve.

At the end of protocol Pn, Alice and Bob end up with the common key kc ¼ b kc , which is not known to the Eve. This
channel-type key-generation protocol Pn is evaluated in terms of probability of error, information leakage to the Eve, and
uniformity of the key.
The weak secret-key capacity for DMC-type scenario can be defined as follows:

ðweakÞ
CSK ≜ supfR : R isanachievableweak SKRg ð11:64Þ

On the other hand, the strong secret-key capacity of a channel-type model is defined as:

ðstrongÞ
CSK ≜ supfR : R isanachievablestrong SKRg ð11:65Þ

The following theorem has been proved by Ahlswede and Csiszár in [39] (see also [4]): The weak secret-key capacity of a
channel-type model satisfies the following lower and upper bounds:
 
max max ½I ðX; Y Þ  I ðX; Z Þ, max ½I ðY; X Þ  I ðY; Z Þ 
fX fX
ð11:66Þ
ðweakÞ
 CSK  max min ðI ðX; Y Þ, I ðX, YjZ ÞÞ
fX

By employing the information reconciliation and privacy amplification steps, the strong secrecy condition can be obtained
from the weak secrecy condition.
732 11 Physical-Layer Security for Wireless and Optical Channels

11.6 Coding for Physical-Layer Security Systems

We have already discussed the concept of nested code in Sect. 11.3. In this coding scenario, the message M is used to select
the subcode (n, nR). After that the source of local randomness U is used to select the codeword within the subcode. For each
message there are 2nRU possible codewords selected at random. In this section we discuss (i) coding for weak secrecy,
(ii) coding for strong secrecy, and (iii) information reconciliation.

11.6.1 Coding for Weak Secrecy Systems

Ozarow and Wyner considered in [43] the wiretap channel model in which the main channel is noiseless, while the wiretap
(Eve’s) channel is noisy. Clearly, assuming the binary transmission, the secrecy of this wiretap channel model is 1-Cw, where
Cw is the capacity of the wiretap channel. Let the binary linear block code of code rate R0 be denoted by C 0 . For the parity-
check matrix H, each codeword xn satisfies the parity-check equations: Hxn ¼ 0. For each message m, we define the coset Cm
as follows:

Cm ¼ fxn : Hxn ¼ mg: ð11:67Þ

To encode Alice randomly chooses the n-tuple xn from the coset Cm. Since there are 2n n-tuples and the cardinality of the
coset Cm is 2nR0, the number of possible cosets is 2n/ 2nR0t> ¼ 2n(1-R0), indicating that the code rate of this scheme is 1-R0.
Clearly, when C 0 is the capacity-achieving code, the equivocation and wiretap rate can be achieved arbitrarily close to the
secrecy capacity. The nested codes get extended to multiterminal scenarios in [44].
The nested wiretap code idea gets extended to noisy main channel in [45] as follows. Let H be an (1-R12)n  n parity-check
matrix of full rank representing the code C , composed of two submatrices H1 and H2:

H1
H¼ , ð11:68Þ
H2

where H1 is (1-R1)n  n submatrix representing the code C 1 , with R1 > R12, and H2 is Rnn submatrix. Clearly, (1-R1)
n + Rn ¼ (1-R12)n so that we can write R ¼ R1-R12. The linear block code (LBC) C is a subcode of the LBC C 1 , while the
distinct cosets of C in C 1 represent the partition of C. Alice wants to send the nR-bit message M to Bob. To do so she randomly
selects n-tuple Xn from the following coset:
   
n H1 n 0
CM ¼ X : X ¼ : ð11:69Þ
H2 M

To decode transmitted message, Bob performs the syndrome decoding as follows. By using the observation vector Yn, he
obtains the estimate of transmitted n-tuple, denoted as X b n , by running the corresponding decoding algorithm to satisfy the
parity-check equations H 1 Xb n ¼ 0: Finally, Bob computes the message estimate by H 2 X b n ¼ M:
b The wiretap code just
described can be denoted by C n .
Clearly, we can partition C 1 , using the cosets of C into 2nR disjoint subsets, and thus represent the generalization of
Wyner’s construction, since C 1 is equivalent to C 0 . Assuming that C 1 comes from the capacity-achieving channel code over
the main channel and C forms the channel capacity-achieving code over the wiretap channel, authors in [45] (see also [46])
were able to show that this coset encoding scheme satisfies both the reliability condition (Pe ! 0 as n ! 1) and the secrecy
capacity condition (1/n)I(M,Zn) ! 0 as n ! 1. Because both codes C 1 and C are capacity achieving, the probability of error
condition is satisfied. To prove the weak secrecy requirement, let us apply the chain rule for I(M,Xn,Zn) in two different ways
[27]:

I ðM, Zn Þ þ I ðX n , Zn jM Þ ¼ I ðXn , Zn Þ þ I ðM, Zn jXn Þ ¼ I ðX n , Zn Þ, ð11:70Þ

|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
¼0

wherein I(M,Zn|Xn) ¼ 0, since M ! Xn ! Zn is a Markov chain. We can rewrite the previous equation as:
11.6 Coding for Physical-Layer Security Systems 733

I ðM, Zn Þ ¼ I ðX n , Zn Þ  I ðX n , Zn jM Þ ¼
|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
H ðX n jM ÞH ðX n jM, Zn Þ ð11:71Þ
n n n n n
¼ I ðX , Z Þ  H ðX jM Þ þ H ðX jM, Z Þ:

Because H(Xn|M) ¼ nR12, I(Xn,Zn)  nCw, and from Fano’s inequality [27, 46] H(Xn| Zn, M)  h(Pe) + PenR12, the
previous equation becomes the following inequality:

I ðM, Zn Þ  nC w  nR12 þ hðPe Þ þ Pe nR12 , ð11:72Þ

and from the weak secrecy requirement definition, we obtain:

h i
1 1
lim I ðM, Zn Þ ¼ lim C w  R12 þ hðPe Þ þ Pe R12 ¼ 0, ð11:73Þ
n!1 n n!1 n

where we employed the reliability condition (Pe ! 0 as n ! 1).

In incoming subsections, we describe several codes’ designs capable of achieving the weak secrecy capacity.

11.6.1.1 Two-Edge-Type LDPC Coding

The two-edge-type LDPC coding [47] is the most natural way to implement the nested coding. It can be considered as an
instance of generalized LDPC coding [48–54]. As discussed in Chap. 9, we can describe an LDPC code in terms of the degree
distribution polynomials μ(x) and ρ(x), for the variable node (v-node) and the check node (c-node), respectively, as follows:

X
dv X
dc
μðxÞ ¼ μd xd1 , ρð x Þ ¼ ρd xd1 , ð11:74Þ
d¼1 d¼1

where μd and ρd denote the fraction of the edges that are connected to degree-d v-nodes and c-nodes, respectively, and dv and
dc denote the maximum v-node and c-node degrees, respectively. For sufficiently large codeword length n, the most of parity-
check matrices in a {μ,ρ} ensemble of LDPC codes of the same girth exhibit similar performance under the log-domain sum-
product algorithm (SPA) described in Chap. 9. Now we can define the two-edge LDPC code ensemble as a {μ1, μ2,ρ1, ρ2}
two-edge-type LDPC coding ensemble of codeword length n, wherein the variable nodes are connected to μ1 (μ2) check nodes
of type 1 (2) and all type 1 (2) check nodes have
 degree
ρ1 (ρ2) [47]. The parity-check matrix of two-edge-type LDPC codes
H1
has the form introduced above, namely, H ¼ , where H1 (H2) define connections between the variable nodes and the
H2
check nodes of type 1 (2). Since the two-edge LPDC codes represent the instances of LBC codes described above, they
achieve the weak secrecy capacity. Let G be the generator matrix that corresponds to the parity-check matrix H. As discussed
above, each coset Cm of C in C 1 contains solutions of xHT ¼ [xH1T xH2T] ¼ [0 m] for message m. Let G* ¼ [g1T g2T . . .
gnRT]T be the submatrix with rows being linearly independent and when augmented with G to form the basis for code with
parity-check matrix H1. To encode nR-bit message m, Alice randomly chooses nR12-bit auxiliary message m’. By calculating:

0 G
x ¼ ½m m  , ð11:75Þ
G

Alice randomly chooses x among all possible solutions of xHT ¼ [0 m]. Bob applies the decoding procedure described in
the paragraph below Eq. (11.69) to estimate the transmitted message.

11.6.1.2 Punctured LDPC Coding

The puncturing in communication systems is typically used to change the code rate of original code. The punctured bits are not
transmitted. So the key idea of using puncturing in secrecy coding is to assign the punctured bits to message bits, denoted by m
[55, 56], which is illustrated in Fig. 11.20.
Further auxiliary dummy bits get generated, denoted as m’, and then parity bits get generated by LDPC encoding to get p.
Alice then sends the codeword x’ ¼ [m’ p] to Bob over the noisy channel, and Bob performs log-domain SPA trying to
734 11 Physical-Layer Security for Wireless and Optical Channels

Auxiliary dummy
random bits m’

Message LDPC
Puncturer
bits m encoder

Alice source encoder

Fig. 11.20 The punctured LDPC encoding for wiretap channels

Fig. 11.21 Defining the

secrecy gap Pe,Eve,min
Bit-error ratio, BER

Reliable
Secure region
region

Pe,Bob,max

Security gap

SNREve,min SNRBob,min

Signal-to-noise ratio, SNR [dB]

reconstruct the punctured (message) bits. Clearly, the number of message bits km is smaller than the number of information
bits k used in LDPC encoder of rate R ¼ k/n. The number of dummy bits is k-km. Therefore, the secrecy rate Rs ¼ km/n is lower
than the LDPC code rate R. The key problem here is to determine the optimized puncturing distribution, which is subject of
investigation in [56]. For Gaussian wiretap channel, the reliability region is defined by the smallest signal-to-noise ratio
(SNRBob,min) for Bob for which the probability of error (Pe,Bob,max) is maximally tolerable, as illustrated in Fig. 11.21. The
secure region is defined as the highest SNR of Eve (SNREve,min) for which the probability of error (Pe,Eve,min) is minimum that
can be tolerated (it is close to 0.5). The security gap is defined as a difference (in dB scale), SNRBob,min-SNREve,min. The goal is
to optimize the puncturing function to minimize the security gap. For additional details an interested reader is referred to [56].
Another approach advocated in [55] was to choose which bits to puncture in an LDPC codeword such that the codewords
with the same punctured bits form the subcodebook, forming thus the nested structure. Let us consider (n’,k’) LPDC code C 0
whose parity-check matrix can be represented in the form H ¼ [H1 H2], where H2 is a lower-triangular matrix of size (n’-
k’)  (n’-k’). The codewords for C 0 are represented in the form x ¼ [m m’ p], where m is the message vector of length k < k’,
m’ is the axillary dummy message vector of length k’-k, and p is the vector of parity checks of length n’-k’. The (n,k) punctured
code C is composed of punctured codewords x’ ¼ [m’ p], which get transmitted toward Bob. The partition of code C can be
done based on punctured bits’ values in the message m. Bob based on the received word performs LDPC decoding to
determine the punctured bits.
11.6 Coding for Physical-Layer Security Systems 735

11.6.1.3 Polar Codes

The polar codes, introduced by Arikan in [57], are low-complexity linear block codes being capacity achieving over binary
input symmetric channels (rows in transition channel matrix are permutations of each other). Let W: {0,1} ! Y , (x ! y), be a
generic binary input discrete memoryless channel (BI-DMC) characterized with transition probabilities W(y|x). For the block
length n ¼ 2m (m is a positive integer), we define the n  n matrix Gn as follows:

: 1 0
Gn ¼ G
m
2 , G2 ¼ , ð11:76Þ
1 1

where
denotes the Kronecker product. Let us consider the sequence u of n-tuples from the binary field F 2, that is, u 2 F n2.
Each bit xi in a codeword x ¼ uGn is transmitted over BI-DMC with transition probability W(y|x), and the resulting output
vector is denoted by y. For each bit ui in u, we define the bit channel W(i): U ! Y n  U i1 , indicating that the bit ui can be
decoded by using the received sequence y and previously decoded bits u1,u2,. . .,ui-1. The i-th bit channel function is
defined by:
  X
W ðiÞ y, ui1 jui ¼ W n ðyjuGn Þ, ð11:77Þ
uNiþ1 2f0, 1gni

whereas before we use the notation ui-1 to denote the sequence [u1 u2 . . . ui-1]. Similarly we use the notation uniþ1 to denote the
sequence [ui + 1 ui + 2 . . . un]. Since the channel is memoryless, the Wn(y|x) term above can be written as product of individual
transition probabilities, that is, W n ðyjxÞ ¼ ∏ni¼1 W ðyi jxi Þ: Arikan has shown that these bit channels “polarize” in the sense that
they become either almost noiseless or almost completely noisy. This “polarization” effect can be described in terms of the
Bhattacharya parameter Z(W ) of the binary input channel W, which is defined as:
X pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ZðWÞ ¼ y
Wðyj0ÞWðyj1Þ 2 ½0, 1, ð11:78Þ

with Z(W ) ! 0 meaning that the channel is noiseless, while for Z(W) ! 1 meaning that the channel is totally noisy.
Let C(W ) denote the channel capacity of channel W. Arikan has shown in [57] that for any fixed δ2(0,1) as n ! 1, the
fraction of indices i2{1,2,. . .,n} for which Z(W(i))2(1-δ,1] tends to C(W ), while the fraction of indices for which Z(W(i))2[0,δ)
tends to 1-C(W ). The bit channels of the first type (noiseless) can be called good bit channels and denoted by G, while the bit
channels of the second type can be called bad bit channels denoted by B ¼ {1,2,. . .,n}\G. This polarization phenomenon can
be employed to design the polar codes. Given that bad bit channels are too noisy, they should not be used to transmit any
reliable information; instead they should be frozen to a fixed value. On the other hand, the good bit channels are almost noise-
free and reliable for transmission of information. So, the main task in the polar code design is to identify then indices of good
bit channels G. By using the Bhattacharya parameter, we can select the indices of the bit channels with the k least Bhattacharya
parameter to design an (n,k) polar code.
Regarding the decoding procedures of polar codes, we can use either sum-product algorithm or sequential decoding. The
sequential decoding operates as follows. For indices belonging to the good bit channels, namely, i2G, we apply the following
decoding rule:
8  
> ðiÞ n bi1
> W y , u j1
< 1,  >1
b
ui ¼ W ðiÞ
y n, b
u i1
j0 ð11:79Þ
>
>
:
0, otherwise

A typical polar code n-combiner architecture for n  4, as proposed by Arikan [57] (see also [30]),
 is shown inFig. 11.22.
Clearly, the combining process proceeds in two steps. In step 1, we pair un into sequence of pairs ( u21 , u43 , ⋯, unn1 ) and then
apply G2 to each pair. This step can be described by the following transformation [30]:
736 11 Physical-Layer Security for Wireless and Optical Channels

Fig. 11.22 Description of

Arikan’s combiner in the Gn
polar code u1 W y1
Permutations
u2 W y2

u3 W y3
Gn/2
u4 W y4

…
…

…
un/2 W yn/2 1
1

un/2 yn/2
W
un/2+1 W yn/2+1

un/2+2 W yn/2+2
…

…
…
un 3 Gn/2 W yn 3

un 2 W yn 2

un 1 W yn 1

un yn
W

2 3
G2 0 0
6 7
In=2
G2 ¼ 4 0 ⋱ 0 5: ð11:80Þ
0 0 G2

In step 2, we feed even bits (u2,u4,. . .,un) to the odd positions of the bottom n/2-stage of the combiner. At the same time, we
feed n/2 sum bits (u1  u2, u2  u3,. . ., un-1  un) to the even positions of the top n/2-stage of the combiner. This step can be
N
described as Rn(I2 Gn/2), where Rn is the permutation matrix separating the even bits and the sum bits.
If Eve’s channel is degraded, it can be shown that Eve’s set of good bits, denoted as G E, is the subset of Bob’s set of good
bits G B [58, 59]. Clearly, in this scenario, Eve will be able to decode properly only a subset of bits that Bob is able to decode.
So we can group the bit channels into three groups: (i) bit channels decodable by Bob only, denoted as G B ∖G E ; (ii) bit
channels decodable by both Bob and Eve, denoted as GBE; and (iii) Bob’s set of bad bits denoted as B B. We can now reorder
information word as follows uG B ∖G E uG BE uBB , wherein Bob’s bad bits are frozen (fixed). The codewords
 
uG B ∖G E uG BE uBB Gn can now be partitioned into different cosets characterized by different values of the uG B ∖G E , thus
representing the nested structure. To encode Alice will place the message bits m into uG B ∖G E information bits block and
auxiliary dummy bits m’ into uG BE block and transmit the codeword ½ m m0 uBB Gn. On receiver side, Bob will perform the
sequential decoding as described above.
Some other classes of codes suitable for use in physical-layer security are provided in [60–62].

11.6.2 Coding for Strong Secrecy Systems

This subsection is devoted to the coding for strong secrecy [63–66], and we describe two classes of codes capable of achieving
the strong secrecy requirement.
11.6 Coding for Physical-Layer Security Systems 737

11.6.2.1 Coset Coding with Dual of LDPC Codes

Let us consider an (n,k) LDPC code C , with the parity-check matrix H belonging to the ensemble of codes with distribution
(μ,λ) and with the BEC threshold α*(μ,λ) [63, 64], suitable for the binary erasure wiretap channel model shown in Fig. 11.10.
(As a reminder, for BEC all errors with error probability α below the threshold α* are correctable with the sum-product (belief-
propagation) algorithm.) Let C ⊥ be the dual of LDPC code C, in other words the (n,n-k) low-density generator matrix
(LDGM) code with generator matrix being H. The 2k cosets of C ⊥ in n-tuple space form a partition, thus representing the
nested structure. To encode a k-bit message m, the encoder chooses the coset with syndrome m ¼ xGT and then randomly
selects a codeword from that coset. To simplify the codeword selection process, we can proceed as follows. The linearly
independent rows of H (representing the basis for C ⊥ code) can be augmented with k linearly independent vectors {g1,g2,. . .,
 T
gk} to form the basis for the n-tuple space (in {0,1}n), representing the matrix G ¼ gT1 ⋯gTk , and the encoding process can
be represented as follows:

G 0
x ¼ ½m m  , ð11:81Þ
H

where m’ is the dummy auxiliary vector. The decoding process is of a reasonable complexity as discussed in [63].
Let PðeC ,BECð1αÞÞ denote the block error probability of the code C over the BEC with erasure probability 1-α. It can be
shown that the following inequality holds [65]:

1 k
I ðM, Z n Þ  PðeC ,BECð1αÞÞ ð11:82Þ
n n

indicating that for 1-α < α*(μ,λ) this construction guarantees the weak secrecy. However, it has been shown in [64], when an
ensemble of large-girth LDPC codes is used, whose girth grows exponentially with the codeword length n, the probability of
error has the desired decay for 1-α < α*(μ,λ), and strong secrecy can be achieved.
When the main channel is noiseless, while the wiretap channel is binary additive, it has been shown in [67] that coset
coding with parity-check matrix generated uniformly at random can also achieve the strong secrecy.

11.6.2.2 Hash Functions and Extractor-Based Coding

The code constructions based on hash functions and extractors can achieve the strong secrecy as discussed in [68–70]. The key
idea behind these code constructions is to employ the universal2 families of hash functions [16, 68–71]. We say that the family
of functions F : (0,1)r ! (0,1)k is a universal2 family of hash functions if for any two distinct x1 and x2 from (0,1)r, the
probability that f(x1) ¼ f(x2) when f is chosen uniformly at random from F is smaller than 2-k. Let us now consider (n,r)
reliability code C and choose f uniformly at random from F . To encode a secret message m from (0,1)k, Alice selects the
sequence u uniformly at random from (0,1)r in the set f 1(m) and encodes it using the code C. It has been shown in [68] that
any choice of f guarantees the strong secrecy when the code rate k/n is lower than secrecy capacity. The problem with this
approach is that function f(x) is not invertible or it is very difficult to invert. To solve for this problem, the use of invertible
extractors is advocated in [69, 70].

11.6.3 Information Reconciliation

We turn our attention now to the source-type secret-key generation (agreement) protocol, depicted in Fig. 11.18 and described
in Sect. 11.5.1. In this model, Alice, Bob, and Eve have access to the respective output ports X , Y , Z of the source
characterized by the joint PDF fXYZ. Each of them sees the i.i.d. realizations of the source denoted, respectively, as Xn, Yn,
and Zn. Given that Alice’s and Bob’s realizations are not perfectly correlated, they need to correct the discrepancies’ errors.
This step in secret-key generation as well as in QKD is commonly referred to as the information reconciliation (or just
reconciliation). The reconciliation process can be considered as a special case of the source coding with the side information
[27, 72]. To encode the source X, we know from Shannon’s source coding theorem that the rate Rx must be as small as possible
but not lower than the entropy of the source, in other words Rx > H(X). To jointly encode the source (XY , fXY), we need the rate
R > H(X,Y ). If we separately encode the sources, the required rate would be R > H(X) + H(Y ). However, Slepian and Wolf
have shown in [72] that the rate R ¼ H(X,Y ) is still sufficient even for separate encoding of correlated sources. This claim is
738 11 Physical-Layer Security for Wireless and Optical Channels

Ry

H(X,Y)
Slepian-Wolf region
H(Y)

H(Y|X)
Rx

H(X|Y) H(X) H(X,Y)

Fig. 11.23 The typical Slepian-Wolf achievable rate region for a correlated source (XY , fXY)

Fig. 11.24 The reconciliation

problem represented as a source Alice DMS Bob
coding with syndrome-based side yn x̂
Syndrome xn
information fXYZ Decoder
s=Hx
zn
s Eve
Authenticated
public channel

commonly referred to as the Slepian-Wolf theorem and can be formulated as follows [27, 72]: The achievable rate region R
(for which the error probability tends to zero) for the separate encoding of the correlated source (XY , fXY) is given by:
8 9
>
< Rx  H ðXjY Þ >
=
: 
R ¼ Rx , Ry : Ry  H ðYjX Þ ð11:83Þ
>
: >
;
Rx þ Ry  H ðX, Y Þ

The typical shape of the Slepian-Wolf region is provided in Fig. 11.23. Now if we assume that (X,fX) should be compressed
when (Y,fY) is available as the side information, then H(X|Y ) is sufficient to describe X.
Going back to our reconciliation problem, Alice compresses her observations Xn, and Bob decodes them with the help of
correlated side information Yn, which is illustrated in Fig. 11.24. The source encoder can be derived from capacity-achieving
channel code. For an (n,k) LDPC code, the parity-check matrix H of size (n-k)  n can be used to generate the syndrome by
s ¼ Hx, where x is Alice’s DMS observation vector of length n. Since x is not a codeword, the syndrome vector is different
from all-zeros vector. The syndrome vector is of length n-k and can be represented as s ¼ [s1 s2 . . . sn-k]T. Clearly, the
syndrome vector is transmitted toward Bob over the (error-free) authenticated public channel.
Once Bob receives the syndrome vector s, his decoder tries to determine the vector x based on s and Bob’s observation
vector y, which maximizes the posterior probability:

b
x ¼ max Pðyjx, sÞ, ð11:84Þ

and clearly, the decoding problem is similar to the maximum a posteriori probability (MAP) decoding. The compression rate
is (n-k)/n ¼ 1-k/n, while the linear code rate is k/n. The number of syndrome bits required is:
11.7 Privacy Amplification 739

n  k  H ðX n jY n Þ ¼ nH ðXjY Þ: ð11:85Þ

In practice, the practical reconciliation algorithms introduce the overhead OH > 0, so that the number of transmitted bits
over the public channel is nH(X|Y )(1 + OH). Given that Alice and Bob at the end of the reconciliation step share with high
probability the common sequence Xn with entropy nH(X), the reconciliation efficiency β will be then:

nH ðX Þ  nH ðXjY Þð1 þ OH Þ H ðXjY Þ

β¼ ¼ 1  OH  1: ð11:86Þ
nI ðX, Y Þ I ðX, Y Þ

The MAP decoding complexity is prohibitively high when LDPC coding is used; instead the low-complexity sum-product
algorithm (SPA) (also known as belief-propagation algorithm) should be used as described in [73]. Given that Alice
observation x is not a codeword, the syndrome bits are nonzero. To account for this problem, we need to change the sign
of log-likelihood to be sent from the c-th check node to the v-th variable node when the corresponding syndrome bit sc is
1. Based on Chap. 2 and ref. [74], we can summarize the log-domain SPA below. To improve the clarity of presentation, we
use the index v ¼ 1,. . .,n to denote variable nodes and index c ¼ 1,. . .,n-k to denote the check nodes. Additionally, we use the
following notation (introduced in Chap. 9): (i) N(c) to denote {v-nodes connected to c-node c}, (ii) N(c)\{v} denoting {v-
nodes connected to c-node c except v-node v}, (iii) N(v) to denote {c-nodes connected to v-node v}, and (iv) N(v)\{c}
representing {c-nodes connected to v-node v except c-node c}.
The formulation of the log-domain SPA for source coding with the side information is provided now:

1. Initialization: For the variable nodes v ¼ 1,2,. . .,n, initialize the extrinsic messages Lvc to be sent from variable node
(v-node) v to the check node (c-node) c to the channel log-likelihood ratios (LLRs), denoted as Lch(v), namely, Lvc ¼ Lch(v).
2. The check-node (c-node) update rule: For the check nodes c ¼ 1,2,. . .,n-k, compute Lcv ¼ ð1  2sc Þ þ Lvc : The
N ðcÞ∖fvg
box-plus operator is defined by:
!
2 X
2
L1 þ L2 ¼ ∏ signðLk Þ  ϕ ϕðjLk jÞ ,
k¼1 k¼1

where ϕ(x) ¼  log tanh (x/2). The box operator for |N(c)\{v}| components is obtained by recursively applying
two-component version defined above.
P
3. The variable-node (v-node) update rule: For v ¼ 1,2,. . .,n, set Lvc ¼ Lch ðvÞ þ Lcv for all c-nodes for which the c-th
N ðvÞ∖fcg
row/v-th column element of the parity-check matrix H is one, that, is hcv ¼ 1.
P
4. Bit decisions: Update L(v) (v ¼ 1,. . .,n) by LðvÞ ¼ Lch ðvÞ þ Lcv and make decisions as follows bxv ¼ 1 when L(v) < 0
N ðvÞ
(otherwise set bxv ¼ 0).

So, the only difference with respect to conventional log-domain SPA is the introduction of the multiplication factor (1-2sc)
in the c-node update rule. Moreover, the channel likelihoods Lch(v) should be calculated from the joint distribution fXY by
Lch(v) ¼ log [fXY(xv ¼ 0| yv)/fXY(xv ¼ 1| yv)]. Because the c-node update rule involves log and tanh functions, it can be
computationally intensive, and to lower the complexity, the reduced complexity approximations should be used as described
in Chap. 9.

11.7 Privacy Amplification

The purpose of the privacy amplification is to extract the secret key from the reconciled (corrected) sequence. The key idea is
to apply a well-chosen compression function, Alice and Bob have agreed on, to the reconciled binary sequence of length nr as
follows g: (0,1)n ! (0,1)k, where k < n, so that Eve gets negligible information about the secret key [3, 4, 15, 16,
75]. Typically, g is selected at random from the set of compression functions G so that Eve does not know which g has
been used. In practice, the set of functions G is based on the family of universal hash functions, introduced by Carter and
Wegman [76, 77].
740 11 Physical-Layer Security for Wireless and Optical Channels

We have already introduced the concept of hash functions but repeat the definition here for completeness of the section. We
say that the family of functions G: (0,1)n ! (0,1)k is a universal2 family of hash functions if for any two distinct x1 and x2 from
(0,1)n, the probability that g(x1) ¼ g(x2) when g is chosen uniformly at random from G is smaller than 2-k.
As an illustrative example, let us consider the multiplication in GF(2n) [16]. Let a and x be two elements from GF(2n). Let
the g function be defined as the mapping (0,1)n ! (0,1)k that assigns to argument x the first k bits of the product ax2GF(2n). It
can be shown that the set of such functions is a universal class of functions for k2[1,n].
The analysis of privacy amplification does not rely on Shannon entropy, but Rényi entropy [78–80]. For discrete random
variable X, taking the values from the sample space {x1,. . .,xn} and distributed according to PX ¼ {p1,. . .,pn}, the Rényi
entropy of X of order α is defined by [79]:
nX o
1 α
R α ðX Þ ¼ log 2 ½P X ð x Þ  : ð11:87Þ
1α x

The Rényi entropy of order two is also known as the collision entropy. From Jensen’s inequality [27], we conclude that the
collision entropy is upper limited by the Shannon entropy, that is, R2(X)  H(X). In limit when α ! 1, the Rényi entropy
becomes the min-entropy:

lim R ðX Þ ¼  log 2 max PX ðxÞ ¼ H 1 ðX Þ: ð11:88Þ

α!1 α x2X

Finally, by applying l’Hôpital’s rule as α ! 1, the Rényi entropy becomes the Shannon entropy:
X
lim Rα ðX Þ ¼  x PX ðxÞ log 2 PX ðxÞ ¼ H ðX Þ: ð11:89Þ
α!1

The properties of Rényi entropy are listed in [79]. Regarding the conditional Rényi entropy, there exist different
definitions; three of them have been discussed in detail in [80]. We adopt the first one. Let the joint distribution for (X, Y )
be denoted as PX,Y, and the distribution for X denoted as PX. Further, let the Rényi entropy of the conditional random variables
X |Y ¼ y, distributed according to PX|Y ¼ y, be denoted as Rα(X |Y ¼ y). Then the conditional Rényi entropy can be defined by:
X
Rα ðXjYÞ ¼ P ðyÞRα ðXjY
y Y
¼ yÞ
ð11:90Þ
1 X X
¼ PY ðyÞlog P ðxjyÞα :
x XjY
1α y 2

The mutual Rényi information can be defined then as:

I α ðX, Y Þ ¼ Rα ðX Þ  Rα ðXjY Þ, ð11:91Þ

and it is not symmetric to Rα(Y )-Rα(Y|X). Moreover, it can even be negative as discussed in [79, 80].
Bennett et al. have proved the following privacy amplification theorem in [16] (see theorem 3): Let X be a random variable
with distribution PX and Rényi entropy of order two R2(X). Further, let G be the random variable representing uniform
selection at random of a member of universal hash functions (0,1)n ! (0,1)k. Then the following inequalities are satisfied [16]:
h i
H ðGðX ÞjGÞ  R2 ðGðX ÞjGÞ  k  log 2 1 þ 2kR2 ðX Þ
ð11:92Þ
2kR2 ðX Þ
k :
ln 2

Bennett et al. have also proved the following corollary of the privacy amplification theorem [16] (see also [4]): Let S2
{0,1}n be a random variable representing the sequence shared between Alice and Bob, and let E be a random variable
representing the knowledge Eve was able to get about the S, with a particular realization of E denoted by e. If the conditional
Rényi entropy of order two R2(S|E ¼ e) is known to be at least r2 and Alice and Bob choose the secret key by K ¼ G(S), where
G is a hash function chosen uniformly at random from the universl2 family of hash functions (0,1)n ! (0,1)k, then the
following is valid:
11.8 Wireless Channels Physical-Layer Security 741

2kr2
H ðKjG, E ¼ eÞ  k  : ð11:93Þ
ln 2

Cachin has proved the following Rényi entropy lemma in [81] (see also [82]): Let X and Q be two random variables and
s > 0. Then with probability of at least 1–2-s, the following is valid [81, 82]:

R2 ðX Þ  R2 ðXjQ ¼ qÞ  log jQj þ 2s þ 2 ð11:94Þ

The total information available to Eve is composed of her observations Zn from the source of common randomness and
additional bits exchanged in information reconciliation phase over the authenticated public channel, represented by random
variable Q. Based on Cachin’s lemma, we can write:

R2 ðSjZn ¼ zn Þ  R2 ðSjZn ¼ zn , Q ¼ qÞ  log jQj þ 2s þ 2 with probability 1  2s ð11:95Þ

We know that R2(X)  H(X), so we conclude that R2(S|Zn ¼ zn)  H(S|Zn ¼ zn) and therefore we can upper bound R2(S|
Z ¼ zn) as follows:
n

R2 ðSjZn ¼ zn Þ  nH ðXjZ Þ: ð11:96Þ

Based on (11.95), we can lower bound R2(S|Zn ¼ zn,Q ¼ q) by:

R2 ðSjZn ¼ zn , Q ¼ qÞ  nH ðXjZ Þ  log jQj  2s  2 with probability 1  2s ð11:97Þ

Since the number of bits exchanged over the public channel log|Q| is approximately nH(X|Y)(1 + OH), for sufficiently large
n, the previous inequality becomes:

R2 ðSjZn ¼ zn , Q ¼ qÞ  nH ðXjZ Þ  nH ðXjY Þð1 þ OH Þ  2s  2

|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
r2 ð11:98Þ
s
with probability 1  2

From Eq. (11.93) we conclude that k-r2 ¼ k2 < 0, which guarantees that for k2 > 0 Eve’s uncertainty of the key is lower
bounded by:

H ðKjE Þ  k  2k2 = ln 2 with probability 1  2s ð11:99Þ

11.8 Wireless Channels Physical-Layer Security

The wireless channel secrecy capacity has been already discussed in Sect. 11.4.2, see also [85, 86], while the corresponding
coding schemes have been described in Sect. 11.6 as well as in [87]. Here we concentrate on (i) PLS for MIMO channels and
(ii) key generation techniques for wireless channels. We first introduce the wireless MIMO systems fundamentals.

11.8.1 PLS for Wireless MIMO Channels

The MIMO wiretap channel consists of Alice’s MTx antennas, Bob’s MRx antennas, and Eve’s ME receive antennas. The
matrix representation of the main (Alice-to-Bob) channel is given by:

y B ¼ H B xA þ z B , ð11:100Þ

while the corresponding matrix representation of the Alice-to-Eve’s channel is:

742 11 Physical-Layer Security for Wireless and Optical Channels

y E ¼ H E xA þ z E , ð11:101Þ
 
where xA is the Alice transmitted vector, with covariance Rx ¼ E xA x{A , yB is Bob’s received vector, HB is Alice-to-Bob
channel matrix of size MRx  MTx, and HE is Alice-to-Eve channel matrix of size ME  MTx. When the additive noise is
Gaussian, the optimum source to achieve secrecy capacity is Gaussian, and the corresponding secrecy capacity expression
under the average power constraint Tr(Rx)  P will be [88]:

Cs ¼ max ½I ðX A , Y B Þ  I ðX A , Y E Þ: ð11:102Þ

Rx , TrðRx ÞP

For Gaussian source, if the noise is spatially white with unit variance, we can write:
h    i
Cs ¼ max log 2 det I þ H B RX H {B  log 2 det I þ H E RX H {E : ð11:103Þ
Rx , TrðRx ÞP

From equation above we conclude that we still need to determine the transmit covariance matrix Rx maximizing the secrecy
rate. Now we consider two scenarios, when the CSI is perfectly and partially known.
Assuming that transmitter has the perfect CSI for both Bob and Eve, for the multiple-input single-output multiple-Eve
(MISOME) case, the optimal solution is transmit beamforming [88]:

Rx ¼ Pjλm ihλm j, hλm j ¼ ðjλm iÞ{ , ð11:104Þ

where the unit-norm column-eigenvector |λmi corresponding to the eigenvalue λm is obtained as the solution to eigenvalue
equation [88]:
   
I þ Ph{B hB jλm i ¼ λm I þ PH {E H E jλm i, ð11:105Þ

where hB is Bob’s channel vector. The solution for the secrecy rate maximization problem of Eq. (11.103) is given by [89]:
h  i
1
CS ¼ log 2 λmax I þ Ph{B hB , I þ PH {E H E , ð11:106Þ
2

where λmax is the largest generalized eigenvalue of the following two matrices I þ Ph{B hB and I þ PH {E HE . Under the matrix
power covariance constraint Rx  S, where “” denotes less or equal to in the positive semidefinite partial ordering sense of
the symmetric matrices, the MIMO secrecy capacity is given by [90]:
X  
1=2 { 1=2 {
C S ðSÞ ¼ i
log 2 λ i I þ S H B H B S1=2
, I þ S H E H E S1=2
, ð11:107Þ

where λi are greater than one eigenvalue of matrices I þ S1=2 H {B H B S1=2 and I þ S1=2 H {E HE S1=2. The average power constraint
is much less restrictive than the matrix power constraint, and the secrecy capacity of the MIMO wiretap channel under the
average power constraint can be found by the exhaustive search over the set {S|S  0), Tr(S)  P}; in other words, we can
write [88, 91]:

CS ðPt Þ ¼ max CS ðSÞ: ð11:108Þ

S0, TrðSÞPt

In the practical schemes, due to limited feedback capabilities, we must deal with the imperfect CSI or partial CSI
[92, 93]. A realistic model will be to assume that actual channel realizations are not known to Alice and Bob, but are
known to lie in an uncertainty set of possible channels [92, 93], and these channels are known as compound channels [92–
95]. Now we have to perform the reliability and security studies over all possible realizations in the uncertainty sets. The
uncertainty sets for Bob and Eve channel matrices’ estimates can be defined, respectively, as [88, 93]:
11.8 Wireless Channels Physical-Layer Security 743


HB ¼ H B j H B ¼ H þ ΔH, kΔH kF  ε , ð11:109Þ

HE ¼ H E j kΔH E kF  ε , ð11:110Þ

where ||||F is the Frobenius norm, H is the estimate of Bob’s channel matrix, and ΔH is the estimation error not larger than ε.
Clearly, we assume that the estimate of Eve’s channel matrix is not known, only the maximum estimation error. This scenario
is applicable when Eve cannot approach the transmitter beyond certain protection distance. In other words, this compound
model is applicable when Eve is outside of the exclusion zone. Interestingly enough, the secrecy capacity for such compound
model can be determined for the nondegraded case for Gaussian MIMO channels [93]. Based on (11.103), (11.109), and
(11.110), we conclude that the compound MIMO wiretap secrecy capacity can be determined by:
    
Cs ¼ max min log 2 det I þ H B RX H {B  max log 2 det I þ H E RX H {E : ð11:111Þ
TrðRx ÞP H B 2HB H E 2HE

Clearly, the maximum secrecy rate is limited by the worst channel to Bob and the best channel to Eve.
The concept of broadcast channel with confidential messages has been already introduced in Sect. 11.4.1. In broadcast
scenario one user sends the message to multiple receivers, and this scenario is very common in downlink phase of cellular
communication, in which the base station sends information to multiple mobile users. In a broadcast channel with confidential
messages, Alice sends confidential message to Bob, which should be kept as secret as possible from other users/
eavesdroppers. The broadcast channel with parallel independent subchannels is considered in [96], and the corresponding
optimal source allocation, achieving the boundary of secrecy capacity region, is determined. The transmission of two
confidential messages over discrete memoryless broadcast channel is studied in [97], wherein each receiver serves as an
eavesdropper of the other. The general MIMO Gaussian case over the matrix power constraint is studied in [98]. For two-user
broadcast channel, with each user receiving the corresponding confidential message, it has been shown that the secrecy
capacity region (R1,R2) is rectangular for the matrix power constraint Cx  S. Assuming so-called secret dirty-paper coding
[98], the corner point rate (R*1,R*2) for secrecy capacity is given by [88, 98]:
X  
1=2 {
R 1 ¼ i
log 2 λ i , λ i ¼ eigenvalues I þ S H B H B S 1=2

X   ð11:112Þ
0 0 1=2 {
R2 ¼  j
log 2 λ j , λ j ¼ eigenvalues I þ S H E H E S1=2

wherein λi are larger than one eigenvalue of I þ S1=2 H {B H B S1=2 , while λ’j are smaller than one eigenvalue of I þ
S1=2 H {E H E S1=2 . Some more recent results on MIMO broadcast channels with confidential messages can be found in [99].
In a multiaccess channel, multiple users send information toward the same receiver, and this situation occurs in cellular
communications when multiple mobile users transmit data toward the base station. In multiaccess channel with confidential
messages, multiple senders Alice 1, Alice 2, etc. transmit confidential messages M1, M2, . . . to a single receiver Bob. Each
confidential message should be decodable by Bob, but without any leakage to other transmitters. This problem is considered in
[100], wherein the inner and outer bounds for the region of secret rates have been determined, while the secrecy capacity
region has been left unsolved.
The interference channel describes a scenario in which multiple transmitter-receiver communication links are simulta-
neously active introducing the crosstalk to each other. The interference channel with confidential messages represents the
situation in which multiple transmitters want to transmit confidential message to the respective receivers in such a way to keep
those messages secure from the counterpart receivers [93, 97, 101, 102]. As an illustration, in Fig. 11.29 the system model for
two-user MIMO interference channel with confidential messages is provided. The Hi is used to denote the channel matrix of
the i-th (i ¼ 1,2) direct channel, while Hji to denote the channel matrix of the crosstalk channel j. The figure of merit suitable to
study the scaling behavior of the aggregate secrecy rate R in multiuser networks with signal-to-noise ratio at transmitter side,
denoted as ρ, is the secret multiplexing gain, denoted as SMG, and can be defined as [88, 101]:

R
SMG ¼ ρ!1
lim : ð11:113Þ
log ρ
744 11 Physical-Layer Security for Wireless and Optical Channels

Fig. 11.25 The model for

two-user MIMO interference Alice 1 Bob 1
channel with confidential M1 X1 Y1 M̂1
messages Encoder 1 H1 Decoder 1
H21

Alice 2 Bob 2
H12
M2 X2 Y1 M̂ 2
Encoder 2 H2 Decoder 1

The SMG, also known as the secure degree of freedom (DoF), has been studied in [101, 102] for the K-users Gaussian
interference channel, where K  3, assuming that multiuser network is composed of K transmitter-receiver pairs, employing
F frequency bands, and it has been found that SMG ¼ (K-2)/(2 K-2) secure DoFs per frequency-time slot per user are almost
surely achievable [101]. On the other hand, when an external eavesdropper is present, each user can achieve SMG ¼ (K-1)/
(2 K ) secure DoFs per frequency-time slot [101].
The two-user MIMO interference channel, with each node having arbitrary number of antennas, has been studied in
[103, 104], and the corresponding interference channel model illustrated in Fig. 11.25 can be represented as follows:

y1 ¼ H 1 x1 þ H 21 x2 þ z1 ,
ð11:114Þ
y2 ¼ H 2 x2 þ H 12 x1 þ z2 :

The achievable secrecy regions have been derived in [103, 104] under different CSI assumptions and various noncoopera-
tive and cooperative scenarios. In noncooperative scenarios, it is assumed that both transmitters have perfect CSI of both the
direct and crosstalk channels. The generalized singular value decomposition (GSVD) can be employed, as explained in
[103, 104]. Given the direct and crosstalk channel matrices H1 and H12, the GSVD procedure gives unitary matrices URx1 and
UE1, positive semidefinite diagonal matrices Λ1 and D1, and A1 matrix of size MRx1  d, where d ¼ min(MRx1,MTx1 + MTx2),
such that the following relationships are valid [103, 104]:

H 1 A1 ¼ U Rx1 Λ1 ,
ð11:115Þ
H 12 A1 ¼ UE1 D1 :

The nonzero elements in Λ1 are ordered in ascending order, while in decreasing order in D1; and these two diagonal
matrices are related by ΛT1 Λ1 þ DT1 D1 ¼ I: The transmitter Tx1 then precodes x1 as follows:

x1 ¼ A1 s1 , s1  CN ð0, Σ1 Þ, ð11:116Þ

where CN is the multivariate complex circular symmetric zero-mean Gaussian (normal) distribution with covariance matrix
Σ1 being positive semidefinite diagonal matrix. The diagonal elements in Σ1 represent the powers allocated to different
antennas. The transmitter Tx2 applies an equivalent procedure. The achievable secrecy rate for Tx1 is given by [103, 104]:

0  R1  log
I þ URx1 Λ1 Σ1 Λ{1 U{Rx1 þ U E2 D2 Σ2 D{2 U {E2

ð11:117Þ

 log
I þ U E2 D2 Σ2 D{2 U{E2
 log
I þ U E1 D1 Σ1 D{1 U {E1
:

The similar achievable secrecy rate inequality holds for Tx2. A reader interested in cooperative scenario is referred to refs
[103, 104].
The relay is used to support communication between transmitter and receiver, by extending the range/coverage or enabling
higher transmission rate. The relay channel with confidential messages [88, 93, 105–107] represents scenario in which Alice
wants to deliver the confidential message to Bob with the help of the relay. Given that relay can be trusted or untrusted, it is
possible to classify the various MIMO wiretap networks into two broad categories: trusted and untrusted relay wiretap
networks. The cooperative jamming strategies enabling secure communication without employing the external helpers have
11.8 Wireless Channels Physical-Layer Security 745

Fig. 11.26 The two-hop MIMO

relay system architecture with an
external eavesdropper

…
…
Transmission Relay Transmission
phase 1 antenna phase 2
array

MR

…
Bob’s

…
Alice antenna
antenna array
array MB

MA
ME
…
Eve’s antenna array

been discussed in [107], with corresponding two-hop MIMO relay network architecture provided in Fig. 11.26. The number
of antennas available to Alice, Bob, relay, and Eve is denoted, respectively, as MA, MB, MR, and ME. In cooperative jamming
strategies, Alice and Bob transmit jamming signals in stages when they do not typically transmit any data. The relay
retransmits the signal received from Alice. Eve has access to both Alice and Bob’s transmissions over wireless channel.
The signals received by relay and Eve in transmission phase 1 can be represented as [107]:

yR ¼ HAR ðT A xA þ T 0 A zA Þ þ H BR T 0 B zB þ zR ,
ð11:118Þ
yE1 ¼ H AE ðT A xA þ T 0 A zA Þ þ H BE T 0 B zB þ zE1 ,

where xA is the data signal transmitted by Alice toward the relay, while zA and zB are the jamming signals transmitted by Alice
and Bob, respectively. (The relay and Eve’s antennas noise sources are denoted by zR and zE1, respectively.) We use TA and
T’B to denote the beamformers corresponding to Alice and Bob. On the other hand, the signals received by Bob and Eve in
transmission phase 2 can be represented as [107]:

yR ¼ H RB ðT R xR þ T 0 R zR Þ þ HAB T 0 A2 zA2 þ zB ,
ð11:119Þ
yE1 ¼ H AE ðT R xR þ T 0 R zR Þ þ H AE T 0 A2 zA2 þ zE1 ,

where xR is the data signal transmitted by relay and TR is the corresponding beamformer. We use the zR and zA2 to denote the
jamming signals transmitted by relay and Alice, respectively. (T’R and T’A2 are corresponding jammers’ beamformers.) When
both Alice and Bob’s jammer signals are different from zero, that is, zA 6¼ 0 and zB 6¼ 0 in transmit phase 1, the corresponding
scheme is called the fully cooperating jamming (FCJ) scheme. Otherwise, if one of them is zero, the corresponding scheme is
called the partially cooperating jamming (PCJ) scheme. As expected, in terms of secrecy rate, the FCJ scheme significantly
outperforms both PCJ scheme and the scheme without the jamming. Moreover, the FCJ does not exhibit the secret rate
saturation effect for very high SNRs. For full details on cooperative jamming schemes for secure communication in relay
MIMO networks, an interested reader is referred to [107].

11.8.2 Secret-Key Generation in Wireless Networks

The secret-key generation (SKG) protocols have already been described in Sect. 11.5, where source- and channel-type SKG
protocols are introduced and analyzed. Here we concentrate on SKG protocols in context of wireless networks. The wireless
channels themselves can serve as a source of common randomness as discussed in [93, 108–111]. The SKG protocols for
wireless channels are similar to those considered in Sect. 11.5; namely, Alice and Bob want to generate the secret with as little
746 11 Physical-Layer Security for Wireless and Optical Channels

correlation with Eve as possible. To do so they perform transmission over wireless fading channel and employ authenticated
public channel for information reconciliation and privacy amplification steps. The key idea is to exploit the fading channel
coefficients reciprocity between Alice and Bob, that is, hAB ¼ hBA ¼ hm, for the channel coherence time Tc. Even if the
reciprocity is not perfect, the occasional errors can be corrected for by Slepian-Wolf-like coding. In training phase, Alice sends
symbol sequence xA of duration TA, followed by Bob’s transmission of symbol sequence xB of duration Tc-TA. During Bob’s
transmission Bob’s and Eve’s received vectors, denoted, respectively, as yB and yE, can be represented as:

y B ¼ hm xA þ z B ,
ð11:120Þ
yE ¼ hAE xA þ zE ,

with zB and zE being the multivariate complex circular symmetric zero-mean Gaussian (normal) distributions with covariance
matrix Σ being σ2I. On the other hand, when Bob transmits the symbol sequence xB, Alice and Eve’s received vectors can be
written as:

y A ¼ h m xB þ z A ,
ð11:121Þ
yE ¼ hBE xB þ zE :

Let us assume that the main (Alice-to-Bob) channel coefficient is generated from a zero-mean Gaussian random variable of
variance σm2. The Alice and Bob’s channel coefficient estimates will be:

x{B
hðmAÞ ¼ hm þ zA ,
kxB k2
ð11:122Þ
x{A
hðmBÞ ¼ hm þ zB :
kxA k2
 2
Clearly, the Alice (Bob) channel coefficient is a zero-mean Gaussian random variable with variance σ ðmAÞ ¼
   2  
σ 2m þ σ 2 =x2B  [ σ ðmBÞ ¼ σ 2m þ σ 2 =x2A  ]. Assuming that both Alice and Bob’s average transmit powers are the same and
 2  
equal to P, we have that x  ¼ T A P, x2  ¼ ðT c  T A ÞP: The secret-key rate (SKR) can be estimated by [110]:
A B

   2  
1 ðAÞ ðBÞ 1 σ þ σ 2m T A P σ 2 þ σ 2m ðT c  T A ÞP
SKR ¼ I hm , hm ¼ log 2 : ð11:123Þ
Tc Tc σ 4 þ σ 2 σ 2m T c P

Obviously, the optimum frame duration for Alice is TA ¼ Tc/2, so that the previous expression simplifies to:
" #
1 σ 2m ðT c PÞ2
SKR ¼ log 2 1 þ  4  : ð11:124Þ
Tc 4 σ þ σ 2 σ 2m T c P

From Eq. (11.124) we conclude that SKR increases as the average power P increases and decreases as the coherence time
increases. Therefore, the fast fading is beneficial, while slow varying fading results in low SKR.
To improve further the SKRs, the joint source-channel secret-key generation was proposed in [111]. In this scheme, the
key generation protocol, illustrated in Fig. 11.27, is composed of two phases: (i) the source-type training phase of duration Ts,
which is used to generate the source-type key and estimate the main (Alice-Bob) channel coefficient at the same time and

Tc

Training source-type Transmission channel-

phase type phase
fTc (1 f)Ts
Tc Ts
Ts

Fig. 11.27 The illustration of joint source-channel key generation scheme

11.9 Optical Channels Physical-Layer Security 747

(ii) the transmission of duration Tc-Ts that is used to generate the channel-type secret key. The total SKR will be then a
summation of source- and channel-type SKRs. The fraction 0 < f < 1 of training interval Ts is used by Alice to transmit the
training sequence xA, while the portion of training interval (1-f )Ts is used by Bob to transmit the training symbol sequence xB.
Therefore, this training source-type phase is very similar to the fading channel reciprocity-based protocol discussed above.
The source-type SKR can be then calculated using an expression similar to Eq. (11.123) as follows:

   2  
1 ðAÞ ðBÞ 1 σ þ σ 2m fT s Ps σ 2 þ σ 2m ð1  f ÞT s Ps
SKRs ¼ I hm , hm ¼ log 2 , ð11:125Þ
Tc Tc σ 4 þ σ 2 σ 2m T c P

where Ps is the average power used during the training source-type phase. Given that the training interval Ts is shorter than
coherence time, it is a good idea to use the minimum mean square error (MMSE) channel estimation approach [33, 112]
instead of the least square method discussed above. Bob’s MMSE estimate of the main channel coefficient will be [111]:

eðBÞ σ 2m
hm ¼ x{ y : ð11:126Þ
σ 2 þ fPs T s σ 2m A B

The true channel coefficient value can be represented as:

ðBÞ
hm ¼ e
hm þ Δhm , ð11:127Þ
 
where the Δhm is the estimation error, which is a zero-mean Gaussian random variable of variance σ 2m = σ 2 þ fPs T s σ 2m :.
In second phase authors in [111] consider scheme in which Alice does not perform both power and rate adaptation to
decouple the first and the second phases, since Eve can learn the Alice channel coefficient estimate for the phase 1. So, Alice
transmits the symbol sequence to Bob with constant power, and the SKR for channel-type phase can be determined by [112]:
h  iþ
Tc  Ts ðBÞ
SKRch ¼ I X A , Y B je
hm  I ðX A , Y E jhAE Þ
Tc
8 2 0  ðAÞ 2 13 9
>
> e   >
>
< 6 B hm Pch C7 h 2
P = ð11:128Þ
Tc  Ts
E6 B C 7 AE ch
¼ 4 log 2@ 1 þ A 5  log 1 þ ,
Tc > > σ m Pch
2 2
σ 2 >
>
: σ þ 2
2 ;
σ þ fPs T s σ 2m

where the second term is an upper bound for Eve’s mutual information, while for the first term, we need to perform the
averaging for different channel coefficient estimates. The following overall SKR for this two-phase-based secret-key
generation protocol is achievable [110]:

SKR ¼ max ðSKRs þ SKRch Þ

f , Ps , T s
ð11:129Þ
s:t:Ps T s þ Pch ðT c  T s Þ  T c P,

where the maximization is performed for all possible parameters f, Ps, and Ts.
For additional details on joint source-channel key generation protocols and relay-assisted key generation, an interested
reader is referred to [110] and references therein.

11.9 Optical Channels Physical-Layer Security

Thanks to its flexibility, security, immunity to interference, high-beam directivity, and energy efficiency, the free-space
optical (FSO) technology represents an excellent candidate for high-performance secure communications. Despite these
advantages, large-scale deployment of FSO systems has so far been hampered by reliability and availability issues due to
atmospheric turbulence in clear weather (scintillation), low visibility in foggy conditions, and high sensitivity to misalignment
[33, 113]. Because of high directivity of optical beams, the FSO systems are much more challenging to intercept compared to
748 11 Physical-Layer Security for Wireless and Optical Channels

RF systems. Nevertheless, the eavesdropper can still apply the beam splitter on transmitter side, the blocking attack, or exploit
beam divergence at the receiver side. The research on FSO physical-layer security is getting momentum, which can be judged
based on increased number of recent papers related to this topic [83, 84, 114–117]. Most of papers on the physical-layer
security for FSO communications are based on direct detection and employ wiretap channel approach introduced by
Wyner [29].
In our recent papers [83, 84, 116, 117], we introduced a different strategy. It is well known that we can associate with a
photon both spin angular momentum (SAM), related to polarization, and OAM, related to azimuthal dependence of the
complex electric field [118–120]. Given that OAM eigenstates are orthogonal, these additional degrees of freedom can be
utilized to improve both spectral efficiency and the physical-layer security in optical networks [12, 33, 74, 83, 84, 118–
120]. Given that the spatial modes in spatial division multiplexing (SDM) fibers, such as few-mode fibers (FMFs), few-core
fibers (FCFs), and few-mode-few-core fibers (FMFCFs), can be decomposed in terms of OAM eigenkets, the OAM can be
used to enable the physical-layer security in both FSO and fiber-optics-based optical networks. Because OAM states provide
an infinite basis state, while SAM states are two-dimensional only, the OAM can also be used to increase the security of QKD
as we described in [124]. Therefore, the OAM eigenkets can be employed to provide physical-layer security on classical,
semiclassical, and quantum levels.

11.9.1 SDM-Fibers-Based Physical-Layer Security

Three types of physical-layer security schemes based on spatial modes are possible, classical, semiclassical, and QKD
schemes, depending on the desired level of security. It is well known that classical protocols rely on the computational
difficulty of reversing the one-way functions and in principle cannot provide any indication of Eve’s presence at any point in
the communication process. However, the optical communication links can be operated at a desired margin from the receiver
sensitivity, and for known channel conditions, Eve’s beam-splitting attack can be detected as it will cause sudden decrease in
secrecy capacity CS, defined as:

C S ¼ C AB  CAE , ð11:130Þ

where CAB is the instantaneous capacity of Alice-to-Bob channel and CAE is the instantaneous capacity of Alice-to-Eve
channel. Another relevant probabilistic measure that will be used in this paper to characterize the security of optical
communication link is the probability of strictly positive secrecy capacity, defined as:

Pþ
S ¼ Pr ðC S > 0Þ: ð11:131Þ

Our assumption for DWDM applications is that we are concerned with the physical-layer security of a particular WDM
channel and that we cannot manipulate other WDM channels. From recent studies of spatial division multiplexing (SDM)
systems, such as [121–123], we have learned that channel capacity can be increased linearly with number of spatial modes N,
rather than logarithmically with signal-to-noise ratio for conventional 2-D schemes. These observations motivate us to employ
the spatial modes to dramatically improve secrecy capacity when compared to conventional 2-D schemes. The use of SDM
schemes to increase the secret-key rates is always sensitive to the crosstalk among spatial modes, and potential eavesdropper
can compromise the security by relying on spatial coupling, without being detected by Alice and Bob. To solve for this
problem, in addition to compensating for spatial modes coupling effects, it is possible to employ the multidimensional
signaling. In multidimensional signaling, the spatial modes are used as bases functions, and by detecting the signal in any
particular spatial mode, Eve will not be able to compromise security as only a single coordinate will be detected. Since the
multidimensional signaling based on spatial modes has been already described in [33, 74, 121–123], here we just briefly
describe the corresponding multidimensional scheme to be used for raw key transmission, which is shown in Fig. 11.28.

Esmate of
Muldim. muldim.
symbols FMF symbols
Binary/nonbinary Mul- Mul-
Muldimensional Mode- Coherent Detected
sequence dimensional dimensional
…

modulator demulplexer detecon sequence

mapper demapper

Fig. 11.28 The physical-layer security scheme employing mode-multiplexing-based multidimensional signaling
11.9 Optical Channels Physical-Layer Security 749

Fig. 11.29 SPML-semiclassical Random binary sequence

physical-layer security scheme:
(a) optical encryption stage and FMF
(b) optical decryption stage. APD, Opcal Spaal domain mux
avalanche photodiode switch
Laser

… …
…
MZM (randomly
diode Taper-core
select

…
the output)
Opcal encrypon stage

(a)
Opcal decrypon stage

FMF Spaal domain demux

APD 1
Decision circuit Raw key

…
…

…
Taper-core APD n (select the

…
largest input)

…
APD N

(b)

The configurations of spatial-modes-based multidimensional modulator and mode-demultiplexer are already provided in
[122, 123]. Alice generates the binary sequence randomly. The multidimensional mapper can be implemented as a look-up
table (LUT). For signal constellation size M, the log2M bits are used to find the coordinates of multidimensional signal
constellation, obtained as described in [123]. The multidimensional coordinates are used as the inputs to corresponding Mach-
Zehnder modulators (MZMs) of multidimensional modulator. After the mode-multiplexing, the signal is transmitted over
SDM system of interest. On receiver side, after mode-demultiplexing and coherent detection, the estimated multidimensional
coordinates are used as inputs of multidimensional a posteriori probability (APP) demapper, which provides the most
probable symbol being transmitted, and the detected sequence is delivered to Bob. After that information reconciliation,
based on systematic LDPC coding, is performed in similar fashion as already proposed for QKD applications [124]. To distill
from the generated key a smaller set of bits whose correlation with Eve’s string falls below the desired threshold, the privacy
amplification is performed with the help of the universal hash functions.
In the rest of this subsection, we describe the corresponding semiclassical physical-layer security scheme. The simplest
version, based on spatial position modulation-like (SPML) approach, is illustrated in Fig. 11.29.
Alice generates a random binary sequence, which is with the help of MZM converted into optical domain. The MZM’s
output drives the optical switch (OS) with ns switching speed, such as one introduced in [125]. The OS output is randomly
selected. The OS outputs are used as inputs of spatial-domain multiplexer, whose configuration is provided in
[122, 123]. Such encrypted signal is sent over FMF system to Bob. On receiver side, in the simplest (direct detection)
version, Bob applies spatial-domain demultiplexer, whose each output branch drives an avalanche photodiode (APD). In
decision, circuit, the largest output is selected as transmitted symbol. For Ms spatial modes, log2 Ms bits of raw key are
transmitted. In this simple scheme, MZM is used for framing purpose only. If coherent detection is used instead and MZM is
replaced by I/Q modulator (with two RF inputs), we can transmit log2 M + log2Ms bits of raw key per channel use, where M is
the size of I/Q constellation. The polarization state (not shown in figure) has not been used for raw key transmission, but to
detect the presence of Eve, since the semiclassical encryption scheme is operated close to the quantum limit. After raw key
transmission is completed, Alice encodes the raw key by employing multilevel nonbinary LDPC-coded modulation (ML-NB-
LDPC-CM) [75]-based information reconciliation. The privacy amplification is then performed to distill for the shorter key
with negligible correlation with Eve. This key is then used for secure communication, based on one-time pad or any
symmetric cipher.

11.9.2 FSO Physical-Layer Security

For classical OAM-based physical-layer security scheme, the N raw key-carrying TEM00 modes are illuminated on a series of
computer-generated holograms (CGHs), implemented, for instance, with the help of the spatial light modulators (SLMs), each
750 11 Physical-Layer Security for Wireless and Optical Channels

Computer generated holograms (CGHs)

Hs)
N-dimensional OAM
Adaptive receiver
optics
Expanding Compressing
CGH 1 telescope CGH* 1
… CGH N telescope
… CGH* N
N-dimensional
FSO
OAM
channel
transmitter

… Mach-Zehnder
modulators Coherent
… detection
receivers
Alice multidimensional
signal generator Bob’s multidimensional
demodulator

Fig. 11.30 The OAM-based FSO physical-layer security scheme

programmed to one out of N OAM modes in use, as illustrated in Fig. 11.30. The corresponding diffraction angles are properly
adjusted so that the coaxial propagation of outgoing OAM beams is obtained, and the resulting superposition beam is
expanded by an expanding telescope. To impose N coordinates for multidimensional signaling, a series of MZMs is used on a
transmitter side. The number of bits required to select a point from multidimensional signal constellation of size M is given by
log2M. Therefore, log2M bits are used to select a point from multidimensional signal constellation, whose coordinates are
stored in N-dimensional mapper, which can be implemented with a help of an arbitrary waveform generator (AWG). On
receiver side, after compressing telescope, we pass the signal to adaptive optics subsystem to compensate for atmospheric
turbulence effects. After that a series of conjugate volume holograms recorded on SLMs are used to determine the projections
along corresponding OAM modes. These OAM projections (in optical domain) are used as inputs of corresponding coherent
detectors to estimate the coordinates of transmitted multidimensional signal. For efficient implementation, one local laser is
used for all coherent detectors. After coherent detection, the corresponding analog-to-digital converters (ADCs) outputs are
passed to an N-dimensional APP demapper, in which symbol LLRs are calculated. After raw key transmission is completed,
Alice encodes the raw key by employing ML-NB-LDPC-CM-based information reconciliation. The privacy amplification is
further performed, to distill for the shorter key with negligible correlation with Eve. This key is then used for secure FSO
communication.
To demonstrate high potential of the OAM-based physical-layer security scheme, the propagation of OAM modes at
1550 nm with the azimuthal index l from 20 to 20 and radial index p set to 0 is simulated in [83] by performing split-step
propagation method [113]. The FSO link of length 1 km is observed. To model the strong turbulence effects, the use of
11 random phase screens is sufficient [83]. The phase power spectral density (PSD) used to generate random phase screens is a
modified version of the Kolmogorov spectrum which includes both inner and outer scales and is given as [113]:

h i
eκ =κl
2 2
7=6
Φn ðκ Þ ¼ 0:033C2n 1 þ 1:802ðκ=κl Þ  0:254ðκ=κ l Þ  11=6 , ð11:132Þ
κ2 þ κ20

where κ is the spatial frequency, κl ¼ 3.3/l0, κ 0 ¼ 1/L0, and C 2n is the refractive index structure parameter that indicates the
turbulence strength; l0 and L0 are the inner and outer scale of the turbulence, respectively. The propagation of Laguerre-
Gaussian (LG) beams with n 2 S ¼ {20, . . ., 1, 0, 1, . . ., 20} of N ¼ 41 OAM modes in total is simulated [83] for three
OAM subsets: S' ¼ {18, . . ., 3, 0, 3, . . ., 18}, S'' ¼ {20, . . ., 5, 0, 5, . . ., 20}, and S' ' ' ¼ {14, 7, 0, 7, 14} with
spacing 2, 4, and 6 between selected modes, respectively, following by the calculation of the aggregate secrecy rate. The
results of calculation are summarized in Fig. 11.31 [83]. The worst-case scenario, when Eve is located on a transmitter side
and taps the portion re of the transmitted optical power, is considered. As we can see, in the weak turbulence regime, the total
secrecy capacity can be improved to close to two orders of magnitude by using OAM multiplexed beam. However, as the
refractive index structure parameter C 2n increases, the secrecy capacity decreases due to orthogonality loss caused by
turbulence effects. Comparing with the secrecy capacity of using single channel (denoted as LG00 mode), the use of OAM
multiplexing is not beneficial in very strong turbulence regime without adaptive optics.
11.9 Optical Channels Physical-Layer Security 751

Fig. 11.31 The aggregate

secrecy capacity vs. refractive
index structure parameter. Solid
lines: equal power per channel.
Dashed lines: fixed system power
(fixed transmitted power equally
divided among OAM channels).
The portion of transmitted power
taped by Eve is set to 0.01. Signal-
to-noise ratio is set to 20 dB.
(After ref. [83]; # IEEE 2016;
reprinted with permission)

Fig. 11.32 The probability of

positive secrecy
capacity vs. Eve’s interception
fraction. Solid lines, equal power
per channel; dashed lines, fixed
system power. Rytov variance is
set to 4 (corresponding to strong
turbulence). (After ref. [83];
# IEEE 2016; reprinted with
permission)

We can also use the Rytov variance σ 2R ¼ 1:23C2n k7=6 L11=6 to represent the turbulence strength, which takes the propagation
distance L, the operating wavelength, and the refractive structure parameter into account. For our FSO system under study
(a path of 1 km and wavelength at 1550 nm), the range of σ 2R spans from 0.002 to 10, while for C 2n from 1016 to 5  1013. As
an illustration, the probability of strictly positive secrecy capacity Pþ S by Eq. (11.131) is calculated, which represents a
probabilistic metric for characterization the secrecy for the communication. The results for Pþ S corresponding to strong
turbulence regime are summarized in Fig.11.32 [83].
In an ideal homogeneous medium, the vorticity of OAM modes is preserved as they propagate and their wavefronts remain
orthogonal even after undergoing diffraction. However, in an atmospheric turbulence channel, their orthogonality is no longer
752 11 Physical-Layer Security for Wireless and Optical Channels

Fig. 11.33 The simulated OAM modes from order 0 to order +5 after 1 km of propagation distance through Cn2 ¼ 1014 turbulence. The top row
contains sample beam profiles for BS modes, and the bottom row contains sample beam profiles for LG modes. (After ref. [84]; # IEEE 2016;
reprinted with permission)

preserved due to fading and time-varying refractive index fluctuations that cause intensity scintillations in the beam profiles.
This results in crosstalk as the power gets transferred from one particular mode to its neighbors. An intriguing possibility for
FSO links that’s been considered for quite some time has been to communicate using Bessel beams. In addition to forming
orthogonal solutions to the free-space Helmholtz equation, these beams also have the property of being non-diffracting for
extremely long distances and self-regenerating if partially blocked by obstructions along their propagation axis [126]. How-
ever theoretical Bessel beams contain an infinite number of nearly equal energy rings, and by definition they have an infinite
energy. Thus the physically realizable analog to the Bessel beam is the Bessel-Gaussian (BS) beam that can also
non-diffractively propagate but only over a finite depth of focus distance. An experimental demonstration of this promising
property being harnessed for use in transmitting data over an FSO channel was reported in [127]. However, the robustness of
BS beams remains tenuous and very much depends on the nature of the obstructions that the beams encounter. While it’s
known that these beams can still regenerate when clipped by a discrete point object blocking a small section of their
wavefront, simulation studies have indicated that they cannot overcome strong turbulence-induced phase changes that distort
their entire profile [128]. Nonetheless under certain scenarios such as a weak to medium turbulence channel of distance on the
order of several kilometers, it’s possible to create BS beams that can propagate intact, as demonstrated in [84].
The BS and LG beams of orders from 15 to +15 have been individually propagated in consecutive manner over a 1 km
turbulence channel of specified strength, with details provided in [84]. Monte Carlo trials are done for each Cn2 value with the
particular values being intentionally chosen to facilitate the plotting of the Cn2 points on a semi-logarithmic horizontal scale.
Figure 11.33 shows the simulated intensity profiles for several of the OAM modes after traveling through a strong turbulence
channel. The complex electric field values at the final step are stored, and from these data the channel crosstalk matrices are
generated by pair-wise computation of overlap integrals between the modal electric fields. An identity matrix would
correspond to the ideal case where no channel crosstalk occurs. On the other hand, the turbulence would cause power leakage
from each of the diagonal cells in the matrix to neighboring cells in the same row. Figure 11.34 contains crosstalk matrix
results arranged in increasing turbulence strength for several Cn2 values. A visual inspection shows slightly more power
leakage among the LG modes than among the BS modes when the Cn2 value is below 1015 m-2/3. To quantify these effects,
the aggregate secrecy capacity is calculated over the entire set of modes and corresponding crosstalk matrices using the
information-theoretic security formulas, with the results being summarized in Fig. 11.35 [84]. Two curves, one for the case of
Eve near the transmitter and one for Eve near the receiver, are shown. As expected, the aggregate CS values decrease with
increasing turbulence strength. For reference, the maximum achievable aggregate secrecy capacity is 175 bits/sec/Hz when
using consecutive orders from 15 to +15 in a perfect channel having no crosstalk regardless of whether Eve is on the transmit
or receive sides.
11.9 Optical Channels Physical-Layer Security 753

Fig. 11.34 The simulation crosstalk matrices for BS modes (top) and LG modes (bottom) for Cn2 ¼ 1017, 1016, 1.75x1016, 1.75x1015 (going
left to right). Transmit orders are labeled along the vertical axis and the orders used to receive them are labeled along the horizontal axis. (After ref.
[84]; # IEEE 2016; reprinted with permission)

Fig. 11.35 The plots of aggregate secrecy capacity calculated from crosstalk matrices over orders 15 to +15 as a function of turbulence strength.
The shaded bands surrounding the data points represent error bars over the Monte Carlo runs. (After ref. [84]; # IEEE 2016; reprinted with
permission)

Comparing the data points in the BS and LG curves, we find that the CS values are typically higher by 10 to 30 bits/sec/Hz
when using BS beams, which indicates better PLS performance than when using LG beams. However, this improvement is
mostly negated in strong turbulence when the Cn2 value is above 1015 m-2/3. Under these channel conditions, we observe the
capacity approaches 0 bit/sec/Hz if the Eve is located near the receiver and becomes negative if Eve is located near transmitter.
For corresponding experimental results, an interested reader is referred to [117].
754 11 Physical-Layer Security for Wireless and Optical Channels

11.10 Concluding Remarks

This chapter has been devoted to the physical-layer security fundamentals. The chapter starts with discussion on security
issues (Sect. 11.1), followed by Sect. 11.2 where the information-theoretic security is introduced and comparison against the
computational security has been performed. In the same section, various information-theoretic security measures have been
introduced, including strong secrecy and weak secrecy conditions. After that, in Sect. 11.3 Wyner’s wiretap channel model,
also known as the degraded wiretap channel model, has been introduced. In the same section, the concept of secrecy capacity
has been introduced as well as the nested wiretap coding. Further, in Sect. 11.4, the broadcast channel with confidential
messages has been introduced, and the secrecy capacity definition has been generalized. The focus has been then moved to the
secret-key generation (agreement) protocols in Sect. 11.5, where the source- and channel-type models have been introduced,
and corresponding secret-key generation protocols have been described. Section 11.6 has been devoted to the coding for the
physical-layer security systems, including both coding for weak and strong secrecy systems. Regarding the coding for weak
secrecy systems, described in Subsection 11.6.1, the special attention has been paid to two-edge-type LDPC coding,
punctured LDPC coding, and polar coding. Regarding the coding for strong secrecy systems, described in Subsection
11.6.2, the focus has been on coset coding with dual of LDPC codes and hash functions/extractor-based coding. In Subsection
11.6.3, the information reconciliation has been introduced and described. In Sect. 11.7, the concept of privacy amplification
has been introduced and described. In wireless channels physical-layer security section, Sect. 11.8, the following topics have
been covered: wireless MIMO PLS (Subsection 11.8.1) and secret-key generation in wireless networks (Subsection 11.8.2). In
Sect. 11.9, related to the optical channels PLS, both PLS for spatial division multiplexing (SDM)-fibers-based systems
(subsection 11.9.1) and FSO systems (Subsection 11.9.2) have been discussed. The set of problems is provided in incoming
section to help reader gain deeper understanding of material described in this chapter.

11.11 Problems

1. The wiretap channel model in which the main (Alice-to-Bob) channel is BSC( p) and wiretap (Bob-to-Eve) channel is
BSC(q) is provided in Fig. 11.8. Prove that equivalent Alice-to-Eve channel is a BSC( p + q-2pq), and calculate the
corresponding secrecy capacity.
2. Let us study the wiretap channel model provided in Fig. 11.P2, in which Alice-to-Bob channel is BSC( p), while the
wiretap channel is Z(q) channel (shown in Fig. 11.P2). Determine secrecy capacity for this model.
3. Let us study the wiretap channel model provided in Fig. 11.P3, in which Alice-to-Bob channel is BSC( p), while the
wiretap channel is the binary erasure channel BEC(q) (shown in Fig. 11.P2). Determine secrecy capacity for this model.

BSC(p)

x0 1 p
y0
p p

x1 y1
1 p
X Y
Alice BSC(p) Bob
Main channel
x0 1
y0
Wiretap
Z(q)
channel q
x1 y1
1 q
Z Z(q)
Eve

Fig. 11.P2 The wiretap channel model under study

11.11 Problems 755

4. Describe how the wiretap code, described in Sect. 11.3, should be modified for the wiretap channel model provided in
Fig. 11.P3.

BSC(p)

x0 1 p
y0
p p

x1 y1
1 p
X Y
Alice BSC(p) Bob
Main channel
x0 1 q y0
Wiretap q
BEC(q)
channel e
q
x1 y1
Z 1 q

Eve BEC(q)

Fig. 11.P3 The wiretap channel model in which main channel is BSC and wiretap channel is BEC

5. Prove the Csiszár-Körner corollary for a broadcast channel with confidential messages, given by Eq. (11.25).
6. Calculate the FSO secrecy capacity for IM/DD in the weak turbulence regime where the distribution of irradiance is
governed by the log-normal distribution, and thermal noise of transimpedance front-end is the dominant noise source.
7. Let us consider indoor infrared communication with IM/DD in which LED is used a Lambertian source and it is placed on
a ceiling. Bob’s receiver is placed directly below the LED. The impulse response of the main channel is composed of the
LOS and four dominant reflections from the walls. What would be the best strategy for Eve to locate her receiver?
Describe how to calculate the secrecy capacity for various Eve’s scenarios.
8. This problem is related to physical-layer security (PLS) applied to 9-ary 2-D constellation described in Problem 8 of
Chap. 5, in the presence of scintillation and AWGN. By employing Wyner’s wiretap PLS model, determine the secrecy
capacity assuming:
(a) Eve is located on transmitter side and employs a beam-splitting attack.
(b) Eve is located on receive side and employs the beam divergence attack.
(c) Eve is located in the middle of the FSO link and exploits Mie scattering effects.
9. In this problem we explore the simulator developed in Problem 1 of Chap. 2. Assume that Eve employs any of three
strategies from Problem 8 and that QPSK is used in the main channel, which is imposed on Gaussian beam by the I/Q
modulator:
(a) Determine the secrecy capacity for different Eve’s scenario assuming that operating wavelength in the main channel is
1550 nm for different turbulence strengths (as specified in Problem 1 of Chap. 2).
(b) Let us assume that the symbol rate per wavelength is 20 GS/s. Assume further dense WDM is used with channel
spacing of 25 GHz. Calculate the crosstalk matrix for different turbulence strengths (as specified in Problem 1 of
Chap. 2), wherein the number of wavelength channels is used as a parameter.
(c) By using the crosstalk matrix developed in (b), calculate overall secrecy capacity for different number of channels and
different Eve’s strategies.
10. This problem is a continuation of the previous problem. Assume now that adaptive optics (AO) is used in main channel.
Under assumption that Eve does not use the AO in her channel, repeat the steps (a)–(c) from the previous problem, and
discuss the improvements. The AO system should be based on your solution from Problem 26 from Chap. 6.
11. In ref. [129], the atmospheric turbulence is used, among others, as the source of randomness for secret-key generation
(SKG). Here we consider one possible application of this scenario. In Fig. 11.P11 a simplified version of the source-type
SKG scheme is provided. Alice and Bob simultaneously send the CW laser beams with the help of compressing
telescopes over the time-varying atmospheric turbulence channel. On receiver side, Alice and Bob after the compressing
telescope and beam splitter detect the received optical signals by using either direct detection or coherent detection
receiver. With direct detection, the fluctuations in intensity (scintillation) can be used as the common source of
756 11 Physical-Layer Security for Wireless and Optical Channels

Atmospheric
Compressing Expanding
turbulence
telescope telescope

Laser Beam External Beam Laser

diode splitter modulator splitter diode
Expanding
Compressing
telescope
telescope
Optical Optical
receiver Alice receiver
Bob

x y

Fig. 11.P11 Atmospheric turbulence-based secret-key generation

randomness. They can quantize the intensity samples, because the reciprocity of the channel are the same. Eve will
experience different turbulence conditions, and her intensity samples are uncorrelated with Alice and Bob’s ones. Assume
that turbulence conditions are weak, and the log-normal distribution of irradiance can be used. Assuming that both Alice
and Bob employ identical transimpedance front-ends, the thermal noise from electrical amplifiers will be Gaussian.
Determine the weak secret-key capacity for this scheme.
12. In this problem we are concerned with a particular coding scheme for the weak secrecy systems, the punctured LDPC
coding, which is described in Sect. 11.6.1.2. In this scheme, the message bits m and the dummy random bits m’ are
combined before the LDCP coding takes place. In punctured FEC schemes, we typically puncture the parity bits, and
these punctured bits are not transmitted. In this scheme, however, we puncture the message bits m instead. Please refer to
the section just cited for additional details. Let us observe an LDPC code example provided in Sect. 9.10.5.4, which is of
rate 0.8 and codeword length 34,665. Assuming that the target BER for Bob is 106 for the AWGN channel, determine
the security gap if the minim BER of Eve that can be tolerated is 0.3.
13. This problem is a continuation of the previous problem in the presence of atmospheric turbulence, modeled with the
gamma-gamma distribution of irradiance for different turbulence strength: weak turbulence with Rytov variance 0.01,
medium turbulence with Rytov variance 1, and strong turbulence with Rytov variance of 20. Determine the security gap
for different turbulence strengths with other parameters being specified in the previous problem.
14. Design a polar code with similar code rate and codeword length as the LDPC code used in Problem 12. Under the same
assumptions as in Problem 12, determine the security gap, and compare it against that obtained in Problem 12.
15. Repeat Problem 13 but now employing the polar code designed in Problem 14. Discuss the security gaps for polar and
LDPC codes.
16. Let the LDPC and polar codes used in Problems 12 and 14, respectively, be used in information reconciliation as
described in Sect. 11.6.3. Determine reconciliation efficiencies of these two codes and discuss which one is better.
17. Let us consider 2  2 MIMO wireless system with elements of channel matrix being spatially white. Assuming the perfect
CSI for Bob and Eve for multiple-input single-output multiple-Eve (MISOME) case, determine the MIMO secrecy
capacity.
18. Let us now consider the FSO channel with coherent detection with distribution of irradiance following the gamma-gamma
distribution and distribution of variance being Gaussian. For different turbulence strengths and MISOME case, determine
the corresponding secrecy capacities.
19. Let us consider the K-user wireless communication channel. For the same conditions as in Problem 17, determine the
secret multiplexing gain.
20. Let us consider the K-user FSO communication channel. For the same conditions as in Problem 18, determine the secret
multiplexing gain.

References
1. X.800: Security architecture for Open Systems Interconnection for CCITT applications, Recommendation X.800 (03/91), available at: https://
www.itu.int/rec/T-REC-X.800-199103-I
2. I.B. Djordjevic, Physical-Layer Security and Quantum Key Distribution (Springer, Cham, 2019)
3. M. Bloch, Physical-Layer Security. PhD dissertation, School of Electrical and Computer Engineering, Georgia Institute of Technology (2008)
References 757

4. M. Bloch, J. Barros, Physical-Layer Security: From Information Theory to Security Engineering (Cambridge University Press, Cambridge,
2011)
5. M. Bloch, Fundamentals of physical layer security, in Physical Layer Security in Wireless Communications, ed. by X. Zhou, L. Song,
Y. Zhang, (CRC Press, Boca Raton/London/New York, 2014), pp. 1–16
6. A. Chorti et al., Physical layer security: A paradigm shift in data confidentiality, in Physical and Data-Link Security Techniques for Future
Communications Systems, Lecture notes in electrical engineering 358, (Springer, 2016), pp. 1–15
7. M. Bloch, J. Barros, M.R.D. Rodrigues, S.W. McLaughlin, Wireless information-theoretic security. IEEE Trans. Inform. Theory 54(6),
2515–2534 (2008)
8. C.H. Bennet, G. Brassard, Quantum cryptography: Public key distribution and coin tossing, in Proceedings of IEEE International Conference
on Computers, Systems, and Signal Processing, (Bangalore, 1984), pp. 175–179
9. C.H. Bennett, Quantum cryptography: Uncertainty in the service of privacy. Science 257, 752–753 (1992)
10. I.B. Djordjevic, Quantum Information Processing and Quantum Error Correction: An Engineering Approach (Elsevier/Academic Press,
Amsterdam-Boston, 2012)
11. X. Sun, I.B. Djordjevic, M.A. Neifeld, Multiple spatial modes based QKD over marine free-space optical channels in the presence of
atmospheric turbulence. Opt. Express 24(24), 27663–27673 (2016)
12. Z. Qu, I.B. Djordjevic, Four-dimensionally multiplexed eight-state continuous-variable quantum key distribution over turbulent channels.
IEEE Photonics Journal 9(6), 7600408-1-7600408-8 (2017)
13. I.B. Djordjevic, FBG-based weak coherent state and entanglement assisted multidimensional QKD. IEEE Photonics Journal 10(4), 7600512-1-
7600512-12 (2018)
14. M.A. Neilsen, I.L. Chuang, Quantum Computation and Quantum Information (Cambridge University Press, Cambridge, 2000)
15. G. Van Assche, Quantum Cryptography and Secrete-Key Distillation (Cambridge University Press, Cambridge/New York, 2006)
16. C.H. Bennett, G. Brassard, C. Crepeau, U. Maurer, Generalized privacy amplification. IEEE Inform. Theory 41(6), 1915–1923 (1995)
17. C.E. Shannon, Communication theory of secrecy systems. Bell Sys. Tech. J. 28(4), 656–715 (1949)
18. G.S. Vernam, Cipher printing telegraph systems for secret wire and radio telegraphic communications. Trans. Am. Inst. Electr. Eng. 1, 295–301
(1926)
19. D. Kahn, The Codebreakers: The Story of Secret Writing (Macmillan Publishing Co., Ney York, 1967)
20. G.D. Forney, On the role of MMSE estimation in approaching the information theoretic limits of linear Gaussian channels: Shannon meets
Wiener, in Proceedings of 41st Annual Allerton Conference on Communication, Control, and Computing, (Monticello, 2003), pp. 430–439
21. B. Schneier, Applied Cryptography, Second Edition: Protocols, Algorithms, and Source Code in C (Wiley, Indianapolis, 2015)
22. J. Katz, Y. Lindell, Introduction to Modern Cryptography, 2nd edn. (CRC Press, Boca Raton, 2015)
23. J.-P. Aumasson, Serious Cryptography: A Practical Introduction to Modern Encryption (No Starch Press, San Francisco, 2018)
24. J. Sebbery, J. Pieprzyk, Cryptography: An Introduction to Computer Security (Prentice Hall, New York, 1989)
25. H. Delfs, H. Knebl, Introduction to Cryptography: Principles and Applications (Information Security and Cryptography), 3rd edn. (Springer,
Heidelberg/New York, 2015)
26. M. Bloch, M. Hayashi, A. Thangaraj, Error-control coding for physical-layer secrecy. Proc. IEEE 103(10), 1725–1746 (2015)
27. T.M. Cover, J.A. Thomas, Elements of Information Theory (Wiley, New York, 1991)
28. M. Bloch, J.N. Laneman, On the secrecy capacity of arbitrary wiretap channels, in Proceedings of 46th Annual Allerton Conference on
Communication, Control, and Computing, (Monticello, 2008), pp. 818–825
29. A.D. Wyner, The wire-tap channel. Bell Syst. Tech. J. 54(8), 1355–1387 (1975)
30. F. Lin, F. Oggier, Coding for wiretap channels, in Physical Layer Security in Wireless Communications, ed. by X. Zhou, L. Song, Y. Zhang,
(CRC Press, Boca Raton/London/New York, 2014), pp. 17–32
31. S.K. Leung-Yan-Cheong, Multi-User and Wiretap Channels Including Feedback. PhD dissertation, Stanford University (1976)
32. I. Csiszár, J. Körner, Broadcast channels with confidential messages. IEEE Trans. Inf. Theory 24(3), 339–348 (1978)
33. I.B. Djordjevic, Advanced Optical and Wireless Communications Systems (Springer, Cham, 2017)
34. A. Goldsmith, Wireless Communications (Cambridge University Press, Cambridge, 2005)
35. D. Tse, P. Viswanath, Fundamentals of Wireless Communication (Cambridge University Press, Cambridge, 2005)
36. J. Barros, M.R.D. Rodrigues, Secrecy capacity of wireless channels, in Proceedings of IEEE International Symposium on Information Theory
2006 (ISIT 2006), (Seattle, 2006), pp. 356–360
37. J. Barros, M. Bloch, Strong secrecy for wireless channels, in Information-Theoretic Security, Lecture Notes in Computer Science, vol. 5155,
(Springer, Berlin, 2008), pp. 40–53
38. U.M. Maurer, Secret key agreement by public discussion from common information. IEEE Trans. Inf. Theory 39(3), 733–742 (1993)
39. R. Ahlswede, I. Csiszár, Common randomness in information theory and cryptography-part I: Secret sharing. IEEE Trans. Inf. Theory 39(4),
1121–1132 (1993)
40. U.M. Maurer, S. Wolf, Unconditionally secure key agreement and intrinsic conditional information. IEEE Trans. Inf. Theory 45(2), 499–514
(1999)
41. I. Csiszár, P. Narayan, Secrecy capacities for multiple terminals. IEEE Trans. Inf. Theory 50(12), 3047–3061 (2004)
42. X. Zhou, L. Song, Y. Zhang (eds.), Physical layer security in wireless communications (CRC Press, Boca Raton/London/New York, 2014)
43. L.H. Ozarow, A.D. Wyner, Wire-tap channel II. AT&T Bell Lab. Tech. J. 63(10), 2135–2157 (1984)
44. R. Zamir, S. Shamai, U. Erez, Nested linear/lattice codes for structured multiterminal binning. IEEE Trans. Inf. Theory 48(6), 1250–1276
(2002)
45. A. Thangaraj, S. Dihidar, A.R. Calderbank, S.W. McLaughlin, J. Merolla, Applications of LDPC codes to the Wiretap Channel. IEEE Trans.
Inform. Theory 53(8), 2933–2945 (2007)
46. M. Andersson, Coding for Wiretap Channels. Thesis, School of Electrical Engineering, Royal Institute of Technology (KTH), Sweden (2011)
47. V. Rathi, R. Urbanke, M. Andersson, M. Skoglund, Rate-equivocation optimal spatially coupled LDPC codes for the BEC wiretap channel, in
Proceedings of 2011 IEEE International Symposium on Information Theory Proceedings, (St. Petersburg, 2011), pp. 2393–2397
48. R.M. Tanner, A recursive approach to low complexity codes. IEEE Tans. Inf. Theory 27, 533–547 (1981)
758 11 Physical-Layer Security for Wireless and Optical Channels

49. J. Boutros, O. Pothier, G. Zemor, Generalized low density (Tanner) codes. In Proc. IEEE Int. Conf. Comm. 99, 441–445 (1999)
50. M. Lentmaier, K. Zigangirov, On generalized low-density parity-check codes based on hamming component codes. IEEE Comm. Lett. 3(8),
248–250 (1999)
51. I.B. Djordjevic, O. Milenkovic, B. Vasic, Generalized low-density parity-check codes for optical communication systems. J. Lightw. Technol.
23(5), 1939–1946 (2005)
52. I.B. Djordjevic, L. Xu, T. Wang, M. Cvijetic, GLDPC codes with reed-Muller component codes suitable for optical communications. IEEE
Comm. Lett. 12(9), 684–686 (2008)
53. I.B. Djordjevic, T. Wang, Multiple component codes based generalized LDPC codes for high-speed optical transport. Opt. Express 22(14),
16694–16705 (2014)
54. D. Zou, I.B. Djordjevic, An FPGA design of generalized low-density parity-check codes for rate-adaptive optical transport networks, in
Proceedings of SPIE PhotonicsWest 2016, Optical Metro Networks and Short-Haul Systems VIII, SPIE, vol. 9773, (San Francisco, 2016),
p. 97730M-1-97730M-6
55. C.W. Wong, T. Wong, J. Shea, Secret-sharing LDPC codes for the BPSK constrained Gaussian wiretap channel. IEEE Trans. Inform. Forensics
Sec. 6(3), 551–564 (2011)
56. D. Klinc, J. Ha, S.W. McLaughlin, J. Barros, B. Kwak, LDPC codes for the Gaussian Wiretap Channel. IEEE Trans. Inform. Forensics and
Security 6(3), 532–540 (2011)
57. E. Arikan, Channel polarization: A method for constructing capacity-achieving codes for symmetric binary-input memoryless channels. IEEE
Trans. Inform. Theory 55(7), 3051–3073 (2009)
58. H. Mahdavifar, A. Vardy, Achieving the secrecy capacity of wiretap channels using polar codes. IEEE Trans. Inform. Theory 57(10),
6428–6443 (2011)
59. O.O. Koyluoglu, H. El Gamal, Polar coding for secure transmission and key agreement. IEEE Trans. Inform. Forensics and Security 7(5),
1472–1483 (2012)
60. R. Liu, Y. Liang, H.V. Poor, P. Spasojevic, Secure nested codes for type II wiretap channels, in Proceedings of 2007 IEEE Information Theory
Workshop, (Tahoe City, 2007), pp. 337–342
61. J. Belfiore, F. Oggier, Secrecy gain: A wiretap lattice code design, in Proceedings of 2010 International Symposium On Information Theory &
Its Applications, (Taichung, 2010), pp. 174–178
62. F. Oggier, P. Solé, J. Belfiore, Lattice codes for the wiretap Gaussian Channel: Construction and analysis. IEEE Trans. Inform. Theory 62(10),
5690–5708 (2016)
63. A. Thangaraj, S. Dihidar, A.R. Calderbank, S.W. McLaughlin, J.-M. Merolla, Applications of LDPC codes to the wiretap channel. IEEE Trans.
Inform. Theory 53(8), 2933–2945 (2007)
64. A. Subramanian, A. Thangaraj, M. Bloch, S.W. McLaughlin, Strong secrecy on the binary erasure wiretap channel using large-girth LDPC
codes. IEEE Trans. Inform. Forensics Sec. 6(3), 585–594 (2011)
65. A.T. Suresh, A. Subramanian, A. Thangaraj, M. Bloch, S.W. McLaughlin, Strong secrecy for erasure wiretap channels, in Proceedings of IEEE
Information Theory Workshop 2010 (ITW2010), (Dublin, 2010)
66. A. Thangaraj, Coding for wiretap channels: Channel resolvability and semantic security, in Proceedings of 2014 IEEE Information Theory
Workshop (ITW 2014), (Hobart, 2014), pp. 232–236
67. G. Cohen, G. Zémor, Syndrome-coding for the wiretap channel revisited, in Proceedings of IEEE Information Theory Workshop 2006,
(Chengdu, 2006), pp. 33–36
68. M. Hayashi, R. Matsumoto, Construction of wiretap codes from ordinary channel codes, in Proceedings of IEEE International Symposium
Information Theory (ISIT) 2010, (Austin, 2010), pp. 2538–2542
69. M. Bellare, S. Tessaro, A. Vardy, Semantic security for the wiretap channel, in Advances in Cryptology—CRYPTO 2012, Lecture Notes in
Computer Science, vol. 7417, (Springer, Berlin/Heidelberg), pp. 294–311
70. M. Cheraghchi, F. Didier, A. Shokrollahi, Invertible extractors and wiretap protocols. IEEE Trans. Inform. Theory 58(2), 1254–1274 (2012)
71. W.K. Harrison, J. Almeida, M.R. Bloch, S.W. McLaughlin, J. Barros, Coding for secrecy: An overview of error-control coding techniques for
physical-layer security. IEEE Signal Proc. Mag. 30(5), 41–50 (2013)
72. D. Slepian, J.K. Wolf, Noiseless coding of correlated information sources. IEEE Trans. Inf. Theory 19(4), 471–480 (1973)
73. A.D. Liveris, Z. Xiong, C.N. Georghiades, Compression of binary sources with side information using low-density parity-check codes, in
Proceedings of IEEE Global Telecommunications Conference 2002 (GLOBECOM '02), vol. 2, (Taipei, 2002), pp. 1300–1304
74. I.B. Djordjevic, On advanced FEC and coded modulation for ultra-high-speed optical transmission. IEEE Commun. Surveys and Tutorials
18(3), 1920–1951 (2016). https://doi.org/10.1109/COMST.2016.2536726
75. C. Cachin, U.M. Maurer, Linking information reconciliation and privacy amplification. J. Cryptol. 10, 97–110 (1997)
76. J.L. Carter, M.N. Wegman, Universal classes of hash functions. J. Comput. Syst. Sci. 18(2), 143–154 (1979)
77. M.N. Wegman, J. Carter, New hash functions and their use in authentication and set equality. J. Comp. Sci. Syst. 22, 265–279 (1981)
78. A. Rényi, On measures of entropy and information, in Proceedings of 4th Berkeley Symposium on Mathematical Statistics and Probability, vol.
1, (University of California Press, 1961), pp. 547–561
79. I. Ilic, I.B. Djordjevic, M. Stankovic, On a general definition of conditional Rényi entropies, in Proceedings of 4th International Electronic
Conference on Entropy and Its Applications, vol. 4, (Sciforum Electronic Conference Series, 2017). https://doi.org/10.3390/ecea-4-05030
80. A. Teixeira, A. Matos, L. Antunes, Conditional Rényi entropies. IEEE Trans. Information Theory 58(7), 4273–4277 (2012)
81. C. Cachin, Entropy Measures and Unconditional Security in Cryptography. PhD dissertation, ETH Zurich, Hartung-Gorre Verlag, Konstanz
(1997)
82. U. Maurer, S. Wolf, Information-Theoretic Key Agreement: From Weak to Strong Secrecy for Free, in Advances in Cryptology —
EUROCRYPT 2000. EUROCRYPT 2000, Lecture Notes in Computer Science, ed. by B. Preneel, vol. 1807, (Springer, Berlin/Heidelberg,
2000)
83. X. Sun, I.B. Djordjevic, Physical-layer security in orbital angular momentum multiplexing free-space optical communications. IEEE
Photon. J. 8(1), paper 7901110 (2016)
References 759

84. T.-L. Wang, J. Gariano, I.B. Djordjevic, Employing Bessel-Gaussian beams to improve physical-layer security in free-space optical
communications. IEEE Photon. J. 10(5), paper 7907113 (2018). https://doi.org/10.1109/JPHOT.2018.2867173
85. E. Telatar, Capacity of multi-antenna Gaussian channels. Eur. Trans. Telecomm. 10, 585–595 (1999)
86. E. Biglieri, Coding for Wireless Channels (Springer, New York, 2005)
87. J.R. Hampton, Introduction to MIMO Communications (Cambridge University Press, Cambridge, 2014)
88. A. Mukherjee, S.A.A. Fakoorian, J. Huang, L. Swindlehurst, MIMO signal processing algorithms for enhanced physical layer security, in
Physical Layer Security in Wireless Communications, ed. by X. Zhou, L. Song, Y. Zhang, (CRC Press, Boca Raton/London/New York, 2014),
pp. 93–114
89. A. Khisti, G.W. Wornell, Secure transmission with multiple antennas I: The MISOME wiretap channel/part II: The MIMOME wiretap channel.
IEEE Trans. Inf. Theory 56(7), 3088-3104/5515–5532 (2010)
90. R. Bustin, R. Liu, H.V. Poor, S. Shamai(Shitz), An MMSE approach to the secrecy capacity of the MIMO Gaussian wiretap channel. EURASIP
J. Wirel. Commun. Netw. 2009, 370970 (2009). https://doi.org/10.1155/2009/370970
91. P. Narayan, H. Tyagi, Multiterminal secrecy by public discussion. Found. Trends Commun. Inf. Theory 13, 129–275 (2016)
92. R.F. Schaefer, H. Boche, H.V. Poor, Secure communication under channel uncertainty and adversarial attacks. Proc. IEEE 103(10), 1796–1813
(2015)
93. H.V. Poor, R.F. Schaefer, Wireless physical layer security. PNAS 114(1), 19–26 (2017)
94. I. Bjelakovic, H. Boche, J. Sommerfeld, Secrecy results for compound wiretap channels. Probl. Inf. Transm. 49(1), 73–98 (2013)
95. R.F. Schaefer, S. Loyka, The secrecy capacity of compound MIMO Gaussian channels. IEEE Trans. Inf. Theory 61(10), 5535–5552 (2015)
96. H.D. Ly, T. Liu, Y. Liang, Multiple-input multiple-output Gaussian broadcast channels with common and confidential messages. IEEE Trans.
Inform. Theory 56(11), 5477–5487 (2010)
97. R. Liu, I. Maric, P. Spasojevic, R.D. Yates, Discrete memoryless interference and broadcast channels with confidential messages: Secrecy rate
regions. IEEE Trans. Inf. Theory 54(6), 2493–2507 (2008)
98. R. Liu, T. Liu, H.V. Poor, S. Shamai (Shitz), Multiple-input multiple-output Gaussian broadcast channels with confidential messages. IEEE
Trans. Inform. Theory 56(9), 4215–4227 (2010)
99. R. Liu, T. Liu, H.V. Poor, S. Shamai (Shitz), New results on multiple-input multiple-output broadcast channels with confidential messages.
IEEE Trans. Inf. Theory 59(3), 1346–1359 (2013)
100. Y. Liang, H.V. Poor, Multiple-access channels with confidential messages. IEEE Trans. Inf. Theory 54(3), 972–1002 (2008)
101. O.O. Koyluoglu, H. El Gamal, L. Lai, H.V. Poor, Interference alignment for secrecy. IEEE Trans. Inf. Theory 57(6), 3323–3332 (2011)
102. X. He, A. Aylin Yener, K-user interference channels: Achievable secrecy rate and degrees of freedom, in Proceedings of 2009 IEEE
Information Theory Workshop on Networking and Information Theory, (Volos, 2009), pp. 336–340
103. S.A.A. Fakoorian, A.L. Swindlehurst, MIMO interference channel with confidential messages: game theoretic beamforming designs, in 2010
Conference Record of the Forty Fourth Asilomar Conference on Signals, Systems and Computers, (Pacific Grove, 2010)
104. S.A.A. Fakoorian, A.L. Swindlehurst, MIMO Interference Channel with confidential messages: Achievable secrecy rates and Precoder design.
IEEE Trans. Inform. Forensics and Security 6(3), 640–649 (2011)
105. Y. Oohama, Capacity Theorems for Relay Channels with Confidential Messages, in Proceedings of 2007 IEEE International Symposium on
Information Theory, (Nice, 2007), pp. 926–930
106. X. He, A. Yener, Cooperation with an untrusted relay: A secrecy perspective. IEEE Trans. Inf. Theory 56(8), 3801–3827 (2010)
107. J. Huang, A.L. Swindlehurst, Cooperative jamming for secure communications in MIMO relay networks. IEEE Trans. Signal Process. 59(10),
4871–4884 (2011)
108. R. Wilson, D. Tse, R.A. Scholtz, Channel identification: Secret sharing using reciprocity in ultrawideband channels. IEEE Trans. Inf. Forensics
and Security 2(3), 364–375 (2007)
109. C. Ye, S. Mathur, A. Reznik, W. Trappe, N. Mandayam, Information theoretic key generation from wireless channels. IEEE Trans Inf.
Forensics and Security 5(2), 240–254 (2010)
110. L. Lai, Y. Liang, H.V. Poor, W. Du, Key generation from wireless channels, in Physical Layer Security in Wireless Communications, ed. by
X. Zhou, L. Song, Y. Zhang, (CRC Press, Boca Raton/London/New York, 2014), pp. 47–68
111. L. Lai, Y. Liang, H.V. Poor, A unified framework for key agreement over wireless fading channels. IEEE Trans. Inf. Forensics and Security
7(2), 480–490 (2012)
112. P.K. Gopala, L. Lai, H. El Gamal, On the secrecy capacity of fading channels. IEEE Trans. Inf. Theory 54(10), 4687–4698 (2008)
113. L.C. Andrews, R.L. Philips, Laser Beam Propagation through Random Media (SPIE Press, Bellingham, 2005)
114. V.G. Sidorovich, Optical countermeasures and security of free-space optical communication links. In Proceedings of European Symposium on
Optics Photonics Defence Security 2004, pp. 97–108, Int. Soc. opt. Photon (2004)
115. F.J. Lopez-Martinez, G. Gomez, J.M. Garrido-Balsells, Physical-layer security in free-space optical communications. IEEE Photonics J. 7(2),
paper 7901014 (2015)
116. IB Djordjevic, X. Sun, Spatial modes-based physical-layer security. In Proceedings of IEEE ICTON 2016, Paper Mo.C1.3, Trento, Italy,
(Invited Paper) (2016)
117. T.-L. Wang, I.B. Djordjevic, Physical-layer security of a binary data sequence transmitted with Bessel-Gaussian beams over an optical Wiretap
Channel. IEEE Photon. J. 10(6), 7908611 (2018)
118. I.B. Djordjevic, M. Arabaci, LDPC-coded orbital angular momentum (OAM) modulation for free-space optical communication. Opt. Express
18(24), 24722–24728 (2010)
119. I.B. Djordjevic, A. Anguita, B. Vasic, Error-correction coded orbital-angular-momentum modulation for FSO channels affected by turbulence.
IEEE/OSA J. Lightw. Technol. 30(17), 2846–2852 (2012)
120. I.B. Djordjevic, Z. Qu, Coded orbital-angular-momentum-based free-space optical transmission. In Wiley encyclopedia of electrical and
electronics engineering (J. G. Webster, Ed., (2016)), pp. 1–12, https://doi.org/10.1002/047134608X.W8291
121. I.B. Djordjevic, Spatial-domain-based hybrid multidimensional coded-modulation schemes enabling multi-Tb/s optical transport. IEEE/OSA
J. Lightw. Technol. 30 (14): 2315–2328 (2012). (Invited Paper)
760 11 Physical-Layer Security for Wireless and Optical Channels

122. I.B. Djordjevic, M. Cvijetic, C. Lin, Multidimensional signaling and coding enabling multi-Tb/s optical transport and networking. IEEE Sig.
Proc. Mag. 31(2), 104–117 (2014)
123. I.B. Djordjevic, A. Jovanovic, Z.H. Peric, T. Wang, Multidimensional optical transport based on optimized vector-quantization-inspired signal
constellation design. IEEE Trans. Comm. 62(9), 3262–3273 (2014)
124. I.B. Djordjevic, Integrated optics modules based proposal for quantum information processing, teleportation, QKD, and quantum error
correction employing photon angular momentum. IEEE Photonics Journal 8(1), paper 6600212 (2016) https://ieeexplore.ieee.org/document/
7393447
125. I.B. Djordjevic, R. Varrazza, M. Hill, S. Yu, Packet switching performance at 10Gb/s across a 44 optical Crosspoint switch matrix. IEEE
Photon. Technol. Lett. 16(1), 102–104 (2004)
126. D. McGloin, K. Dholakia, Bessel beams: Diffraction in a new light. Contemp. Phys. 46(1), 15–28 (2005)
127. S. Chen, S. Li, Y. Zhao, J. Liu, L. Zhu, A. Wang, J. Du, L. Shen, J. Wang, Demonstration of 20-Gbit/s high-speed Bessel beam encoding/
decoding link with adaptive turbulence compensation. Opt. Lett. 41(20), 4680–4683 (2016)
128. W. Nelson, J.P. Palastro, C.C. Davis, P. Sprangle, Propagation of Bessel and airy beams through atmospheric turbulence. J. Opt. Soc. Am. A
31(3), 603–609 (2014)
129. I.B. Djordjevic, Atmospheric turbulence-controlled cryptosystems. IEEE Photon. J. 13(1), 7900909 (2021)
 Appendix

This chapter is mostly devoted to the abstract algebra fundamentals [1–15], needed in Chap. 9. Only the most important topics
needed for better understanding of the error correction are covered. The chapter is organized as follows. In Sect. A.1, we
introduce the concept of groups and provide their basic properties. In Sect. A.2, we introduce the concept of fields, while in
Sect. A.3, the concept of vector spaces. Further, Sect. A.4 is devoted to algebra of finite fields, which is of high importance in
forward error correction. Section A.5 is related to the pulse-position modulation (PPM), relevant in UWB communications
and deep-space optical communications. The fundamentals of the z-transform, needed in Chaps. 3 and 6, are provided in
Sect. A.6.

A.1 Groups

Definition D1 A group is the set G that together with an operation, denoted by “+,” satisfies the following axioms:

1. Closure: 8 a, b 2G ) a + b 2G.
2. Associative law: 8 a, b, c 2G ) a + (b + c) ¼ (a + b) + c.
3. Identity element: ∃ e 2G such that a + e ¼ e + a ¼ a 8 a 2G.
4. Inverse element: ∃ a1 2G 8 a 2G, such that a + a1 ¼ a1 + a ¼ e.

We call a group Abelian or a commutative group, if the operation “+” is also commutative: 8 a, b 2G ) a + b ¼ b + a.

Theorem T1 The identity element in a group is the unique one, and each element in the group has a unique inverse element.

Proof If the identity element e is not a unique one, then there exists another one e’: e’ ¼ e’ + e ¼ e, implying that e’ and e are
identical. In order to show that a group element a has a unique inverse a1, let us assume that it has another inverse element
a11 : a11 ¼ a11 + e ¼ a11 + (a + a1) ¼ (a11 + a) + a1 ¼ e + a1 ¼ a1, thus implying that the inverse element in the
group is unique.

Examples
• F2 ¼ {0,1} and “+” operation is in fact the modulo-2 addition, defined by 0 + 0 ¼ 0, 0 + 1 ¼ 1, 1 + 0 ¼ 1, 1 + 1 ¼ 0. The
closure property is satisfied, 0 is identity element, 0 and 1 are their own inverses, and operation “+” is associative. The set
F2 with modulo-2 addition forms, therefore, a group.
• The set of integers forms a group under the usual addition operation on which 0 is identity element, and for any integer n,
n is its inverse.
• Consider the set of code words of binary linear (N,K) block code. Any two code words added per modulo-2 form another
code word (according to the definition of the linear block code). All-zeros code word is identity element, and each code
word is its own inverse. The associative property also holds. Therefore, all group properties are satisfied, and the set of code
words form a group, and this is the reason why this is called the group code.

# The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 761
I. B. Djordjevic, Advanced Optical and Wireless Communications Systems, https://doi.org/10.1007/978-3-030-98491-5
762 Appendix

The number of elements in the group is typically called the order of the group. If the group has a finite number of elements,
it is called the finite group.

Definition D2 Let H be a subset of elements of group G. We call H subgroup of G if the elements of H themselves form the
group under the same operation as that defined on elements from G.
In order to verify that the subset S is the subgroup, it is sufficient to check the closure property and that an inverse element
exists for every element of subgroup.

Theorem T2 (Lagrange Theorem) The order of a finite group is an integer multiple of any of its subgroup order.

Example Consider the set Y of N-tuples (received words on a receiver side of a communication system), each element of the
N-tuple being 0 or 1. It can easily be verified that elements of Y form a group under modulo-2 addition. Consider a subset C of
Y with elements being the code words of a binary (N,K ) block code. Then C forms a subgroup of Y, and the order of group Y is
divisible by the order of subgroup C.
There exist groups, whose all elements can be obtained as power of some element, say a. Such a group G is called the cyclic
group, and the corresponding element a is called the generator of the group. The cyclic group can be denoted by G ¼ <a > .

Example Consider an element α of a finite group G. Let S be the set of elements S ¼ {α, α2, α3,. . ., αi,. . .}. Because G is
finite, S must be finite as well, and therefore, not all powers of α are distinct. There must be some l,m (m > l) such that αm ¼ αl
so that αm α-l ¼ αl α-l ¼ 1. Let k be the smallest such power of α for which αk ¼ 1, meaning that α, α2, . . ., αk are all distinct.
We can verify now that set S ¼ {α, α2, . . ., αk ¼ 1} is a subgroup of G. S contains the identity element, and for any element αi,
the element αk-i is its inverse. Given that any two elements αi, αj2S the corresponding element obtained as their product,
αi + j2S if i + j  k. If i + j > k, then αi + j1 ¼ αi + j α-k, and because i + j-k  k, the closure property is clearly satisfied. S is,
therefore, the subgroup of group G. Given the definition of cyclic group, the subgroup S is also cyclic. The set of code words
of a cyclic (N,K ) block code can be obtained as a cyclic subgroup of the group of all N-tuples.

Theorem T3 Let G be a group and {Hi|i2I} be a nonempty collection of subgroups with index set I. The intersection \i2IHi
is a subgroup.

Definition D3 Let G be a group and X be a subset of G. Let {Hi|i2I} be the collection of subgroups of G that contain X. Then
the intersection \i2IHi is called the subgroup of G generated by X and is denoted by <X > .

Theorem T4 Let G be a group and X nonempty subset of G with elements {xi | i2I} (I is the index set). The subgroup of
G generated by X, <X>, consists of all finite product of the xi. The xi elements are known as generators.

Definition D4 Let G be a group and H be a subgroup of G. For any element a2G, the set aH ¼ {ah|h2H} is called the left
coset of H in G. Similarly, the set Ha ¼ {ha|h2H} is called the right coset of H in G.

Theorem T5 Let G be a group and H be a subgroup of G. The collection of right cosets of H, Ha ¼ {ha|h2H} forms a
partition of G.

Instead of formal proof, we provide the following justification. Let us create the following table. In the first row, we list all
elements of subgroup H, beginning with the identity element e. The second column is obtained by selecting an arbitrary
element from G, not used in the first row, as the leading element of the second row. We then complete the second row by
“multiplying” this element with all elements of the first row from the right. Out of not previously used elements, we arbitrarily
select an element as the leading element of the third row. We then complete the third row by multiplying this element with all
elements of the first row from the right. We continue this procedure until we exploit all elements from G. The resulting table is
as follows:

h1 ¼ e h2 ... hm-1 hm
g2 h2g2 ... hm-1 g2 hmg2
g3 h2g3 ... hm-1 g3 hmg3
... ... ... ... ...
gn h2gn ... hm-1gn hmgn
Appendix 763

Each row in this table represents a coset, and the first element in each row is a coset leader. The number of cosets of H in
G is in fact the number of rows, and it is called the index of H in G, typically denoted by [G:H]. It follows from table above |
H| ¼ m, [G:H] ¼ n, and |G| ¼ nm ¼ [G:H]|H|, which can be used as the proof of Lagrange’s theorem. In other words, [G:H] ¼ |
G|/|H|.

Definition D5 Let G be a group and H be a subgroup of G. H is normal subgroup of G if it is invariant under conjugation, that
is, 8 h2H and g2G, the element ghg12H.
In other words, H is fixed under conjugation by the elements from G, namely, gHg1 ¼ H for any g2G. Therefore, the left
and the right cosets of H in G coincide: 8 g2G, gH¼Hg.

Theorem T6 Let G be a group and H a normal subgroup of G. If G|H denotes the set of cosets of H in G, then the set G|H
with coset multiplication forms a group, known as the quotient group of G by H.
The coset multiplication of aH and bH is defined as aH*bH ¼ abH. It follows from Lagrange’s theorem that |G/H| ¼ [G:
H]. This theorem can straightforwardly be proved by using the table above.

A.2 Fields

Definition D6 A field is a set of elements F with two operations, addition “+” and multiplication “∙”, such that:

1. F is an Abelian group under addition operation, with 0 being the identity element.
2. The nonzero elements of F form an Abelian group under the multiplication operation, with 1 being the identity element.
3. The multiplication operation is distributive over the addition operation:

8a, b, c 2 F ) a  ðb þ cÞ ¼ a  b þ a  c

Examples
• The set of real numbers, with operation + as ordinary addition, and operation ∙ as ordinary multiplication, satisfy above
three properties, and it is therefore a field.
• The set consisting of two elements {0,1}, with modulo-2 multiplication and addition, given in the table below, constitutes a
field known as Galois field and is denoted by GF(2).

+ 0 1 ∙ 0 1
0 0 1 0 0 0
1 1 0 1 0 1

• The set of integers modulo-p, with modulo-p addition and multiplication, forms a field with p elements, denoted by GF( p),
providing that p is a prime.
• For any q that is an integer power of prime number p (q ¼ pm, m - an integer), there exists a field with q elements, denoted
as GF(q). (The arithmetic is not modulo-q arithmetic, except when m ¼ 1.) GF( pm) contains GF( p) as a subfield.

Addition and multiplication in GF(3) are defined as follows:

+ 0 1 2 ∙ 0 1 2
0 0 1 2 0 0 0 0
1 1 2 0 1 0 1 2
2 2 0 1 2 0 2 1
764 Appendix

Addition and multiplication in GF(22):

+ 0 1 2 3 ∙ 0 1 2 3
0 0 1 2 3 0 0 0 0 0
1 1 0 3 2 1 0 1 2 3
2 2 3 0 1 2 0 2 3 1
3 3 2 1 0 3 0 3 1 2

Theorem T7 Let Zp denote the set of integers {0,1,. . .,p-1}, with addition and multiplication defined as ordinary addition and
multiplication modulo-p. Then Zp is a field if and only if p is a prime.

A.3 Vector Spaces

Definition D7
Let V be a set of elements with a binary operation “+” and let F be a field. Further, let an operation “∙” be defined between the
elements of V and the elements of F. Then V is said to be a vector space over F if the following conditions are satisfied 8 a, b 2
F and 8 x, y 2 V:

1. V is an Abelian group under the addition operation.

2. 8 a 2 F and 8 x 2 V, then a ∙ x 2 V.
3. Distributive law:

a  ðx þ yÞ ¼ a  x þ a  y

ða þ bÞ  x ¼ a  x þ b  x

4. Associative law:

ða  bÞ  x ¼ a  ðb  xÞ

If 1 denotes the identity element of F, then 1 ∙x ¼ x. Let 0 denote the zero element (identity element under +) in F, and 0 the
additive element in V, 1 the additive inverse of 1, the multiplicative identity in F. It can easily be shown that the following
two properties hold:

0x¼0
x þ ð1Þ  x ¼ 0

Examples
• Consider the set V, whose elements are n-tuples of the form v ¼ (v0,v1,. . .,vn-1), vi2F. Let us define the addition of any two
n-tuples as another n-tuple obtained by component-wise addition and multiplication of n-tuple by an element from F. Then
V forms vector space over F, and it is commonly denoted by Fn. If F ¼ R (R – the field of real number), then Rn is called the
Euclidean n-dimensional space.
• The set of n-tuples whose elements are from GF(2), again with component-wise addition and multiplication by an element
from GF(2), forms a vector space over GF(2).
• Consider the set V of polynomials whose coefficients are from GF(q). Addition of two polynomials is the usual polynomial
addition, addition being performed in GF(q). Let the field F be GF(q). Scalar multiplication of a polynomial by a field
element from GF(q) corresponds to the multiplication of each polynomial coefficient by the field element, carried out in GF
(q). V is then a vector space over GF(q).
Appendix 765

Consider a set V that forms a vector space over a field F. Let v1,v2,. . .,vk be vectors from V, and a1,a2,. . .,ak be field
elements from F. The linear combination of the vectors v1,v2,. . .,vk is defined by:

a1 v 1 þ a2 v 2 þ . . . þ ak v k

The set of vectors {v1,v2,. . .,vk} is said to be linearly independent if there does not exist a set of field elements a1,a2,. . .,ak,
not all ai ¼ 0, such that:

a1 v 1 þ a2 v 2 þ . . . þ ak v k ¼ 0

Example The vectors (0 0 1), (0 1 0), and (1 0 0) (from F3) are linearly independent. However, the vectors (0 0 2), (1 1 0), and
(2 2 1) are linearly dependent over GF(3) because they sum to zero vector.
Let V be a vector space and S be subset of the vectors in V. If S is itself a vector space over F under the same vector addition
and scalar multiplication operations applicable to V and F, then S is said to be a subspace of V.

Theorem T7 Let {v1,v2,. . .,vk} be a set of vectors from a vector space V over a field F. Then the set consisting of all linear
combinations of {v1,v2,. . .,vk} forms a vector space over F and is, therefore, a subspace of V.

Example Consider the vector space V over GF(2) given by the set {(0 0 0), (0 0 1), (0 1 0), (0 1 1), (1 0 0), (1 0 1), (1 1 0), (1 1
1)}. The subset S ¼ {(0 0 0), (1 0 0), (0 1 0), (1 1 0)} is a subspace of V over GF(2).

Example Consider the vector space V over GF(2) given by the set {(0 0 0), (0 0 1), (0 1 0), (0 1 1), (1 0 0), (1 0 1), (1 1 0), (1 1
1)}. For the subset B ¼ {(0 1 0), (1 0 0)}, the set of all linear combinations is given by S ¼ {(0 0 0), (1 0 0), (0 1 0), (1 1 0)} and
forms a subspace of V over F. The set of vectors B is said to span S.

Definition D8 A basis of a vector space V is set of linearly independent vectors that spans the space. The number of vectors in
a basis is called the dimension of the vector space.
In example above, the set {(0 0 1), (0 1 0), (1 0 0)} is the basis of vector space V and has dimension 3.

A.4 Algebra of Finite Fields

A ring is a set of elements R with two operations, addition “+” and multiplication “∙”, such that:

(i) R is an Abelian group under addition operation, (ii) multiplication operation is associative, and (iii) multiplication is
associative over addition.

The quantity ais said to be congruent to b to modulus n, denoted as a  b (mod n), if a-b is divisible by n. If x  a (mod n),
then a is called a residue to x to modulus n. A class of residues to modulus n is the class of all integers congruent to a given
residue (mod n), and every member of the class is called a representative of the class. There are n classes, represented by (0),
(1), (2), . . ., (n-1), and the representative of these classes is called a complete system of incongruent residues to modulus n. If
i and j are two members of a complete system of incongruent residues to modulus n, then addition and multiplication between
i and j are defined by:

i þ j ¼ ði þ jÞðmod nÞ and i  j ¼ ði  jÞðmod nÞ

A complete system of residues (mod n) forms a commutative ring with unity element. Let s be a nonzero element of these
residues. Then s posses inverse if and only if n is prime, p. Therefore, if p is a prime, a complete system of residues (mod p)
forms a Galois (or finite) field and is denoted by GF( p).
Let P(x) be any given polynomial in x of degree m with coefficients belonging to GF( p), and let F(x) be any polynomial in
x with integral coefficients. Then F(x) may be expressed as:
766 Appendix

FðxÞ ¼ f ðxÞ þ p  qðxÞ þ PðxÞ  QðxÞ, where f ðxÞ ¼ a0 þ a1 x þ a2 x2 þ ⋯ þ am1 xm1 , ai 2 GFðpÞ:

This relationship may be written as:

F(x) f(x) mod {p,P(x)}, and we say that f(x) is the residue of F(x) modulus p and P(x). If p and P(x) are kept fixed but f(x)
varied, pm classes may be formed, because each coefficient of f(x) may take p values of GF( p). The classes defined by f(x)
form a commutative ring, which will be a field if and only if P(x) is irreducible over GF( p) (not divisible with any other
polynomial of degree m1 or less).
The finite field formed by pm classes of residues is called a Galois field of order pm and is denoted by GF( pm). The function
P(x) is said to be minimum polynomial for generating the elements of GF( pm) (the smallest-degree polynomial over GF( p)
having a field element β2GF( pm) as a root). The nonzero elements of GF( pm) can be represented as polynomials of degree at
most m-1 or as powers of a primitive root α such that:
m
αp 1
¼ 1, αd 6¼ 1 ðfor d dividing pm  1Þ

A primitive element is a field element that generates all nonzero field elements as its successive powers. A primitive
polynomial is irreducible polynomial that has a primitive element as its root.

Theorem T8 Two important properties of GF(q), q ¼ pm are:

1. The roots of polynomial xq-1-1 are all nonzero elements of GF(q).

2. Let P(x) be an irreducible polynomial of degree m with coefficients from GF( p) and β be a root from the extended field GF
(q ¼ pm). Then all the m roots of P(x) are:

m1
β, βp , βp , . . . , βp
2

To obtain a minimum polynomial, we divide xq-1 (q ¼ pm) by the least common multiple of all factors like xd-1, where d is a
divisor of pm-1. Then we get the cyclotomic equation – the equation having for its roots all primitive roots of equation xq-1-
1 ¼ 0. The order of this equation is ϕ ( pm-1), where ϕ (k) is the number of positive integers less than k and relatively prime to
it. By replacing each coefficient in this equation by least nonzero residue to modulus p, we get the cyclotomic polynomial of
order ϕ ( pm-1). Let P(x) be an irreducible factor of this polynomial; then P(x) is a minimum polynomial, which is in general
not the unique one.

Example Let us determine the minimum polynomial for generating the elements of GF(23). The cyclotomic polynomial is:




x 7  1 =ð x  1Þ ¼ x 6 þ x 5 þ x 4 þ x 3 þ x 2 þ x þ 1 ¼ x 3 þ x 2 þ 1 x 3 þ x þ 1

Hence, P(x) can be either x3 + x2 + 1 or x3 + x + 1. Let us choose P(x) ¼ x3 + x2 + 1.

ϕ ð7Þ ¼ 6, deg½PðxÞ ¼ 3

Three different representations of GF(23) are given in the table below.

Power of α Polynomial 3-tuple

0 0 000
α0 1 001
α1 α 010
α2 α2 100
α3 α2 + 1 101
α4 α2+ α + 1 111
α5 α+1 011
α6 α2+ α 110
α7 1 001
Appendix 767

A.5 Pulse-Position Modulation (PPM)

Pulse-position modulation (PPM) is a standard modulation technique that has been considered for different purposes [16–21].
In PPM, we employ M pulse-position basis functions defined as follows:
 
1 t  ð j  1ÞT s =M
Φ j ðt Þ ¼ pffiffiffiffiffiffiffiffiffiffiffiffi rect ; j ¼ 1, ⋯, M;
T s =M T s =M

where Ts is a symbol duration and rect(t) is defined as:


1, 0  t < 1
rectðt Þ ¼ :
0, otherwise

The signal space of M-ary PPM is therefore M-dimensional, and constellation points are located on axes. Only two amplitude
levels per basis function are used. The time-domain representation of M-ary PPM signal intensity is given by [17]:
rffiffiffiffiffi
X
1
Ts
sðt Þ ¼ MP Φ ðt  kT s Þ,
k¼1
M ak

where {ak} represents a sequence of symbols being transmitted, while P denotes the average power. The symbol error
probability for equal probability of transmitted symbols is given as [17]:
 rffiffiffiffiffiffiffiffiffiffiffiffi
P M
Ps ’ ðM  1Þerfc ,
2 2Rs σ 2

where erfc() is the complementary error function introduced earlier, Rs is the symbol rate related to bit rate by Rb by Rs ¼ Rb/
log2M, and σ 2 is the noise variance. Let us assume that all symbol errors are equally likely, and therefore, they occur with
probability Ps/(M-1). There are 2 log 2 M1 cases in which i-th bit in a given symbol is in error. Therefore, the bit-error
probability can be estimated as:

2 log 2 M1 M=2

BER ’ Ps ¼ P:
M1 M1 s

Bandwidth occupied by PPM signal is B ¼ 1/(Ts/M) since there are M possible time slots (also known as chips), each with
duration Ts/M. The bandwidth efficiency of M-ary PPM is given by:

Rb Rb R log 2 M
ρ¼ ¼ ¼ b ¼ ,
B 1=ðT s =M Þ MRs M

which is lower than the efficiency of M-ary QAM. The low-bandwidth efficiency of M-ary PPM can be improved by multi-
pulse PPM [19], where w (w > 1) out of M possible time slots are used. The symbol interval Ts is now related to bit interval
Tb by:

M
T s ¼ T b log 2 :
w

The bandwidth efficiency of multi-pulse PPM is given by:


M
log 2
Rb Rb Rb w
ρ¼ ¼ ¼   ¼ :
B 1=ðT s =M Þ M M
1= T b log 2 =M
w
768 Appendix

Another alternative approach to improve the bandwidth efficiency of PPM is based on differential PPM (DPPM) [20], also
known as truncated PPM (TPPM), in which the new PPM symbol begins as soon as the slot containing the pulse is over. This
scheme, however, suffers from variable-rate code design problem and catastrophic error propagation.
Finally, to improve bandwidth efficiency of PPM, so-called multidimensional PPM [21] can be used. In this scheme, the
pulse positions are used as basis functions, as given above, and then L amplitude levels are imposed on each basis function.
The constellation points, in this scheme, occupy all axes not just one as in ordinary PPM. This scheme can carry log2(ML) bits
per symbol, thus improving bandwidth efficiency by L times.

A.6 The z-Transform

The z-transform plays the same role in discrete time (DT) systems as the Fourier and Laplace transforms play in continuous
time (CT) systems. It converts the difference equations into algebraic ones in z-plane. By using the z-transform, the
convolution sum in time domain can be converted to product of z-transforms of input and impulse response. This section
is based on references [22–24].

A.6.1 Bilateral z-Transform

The bilateral z-transform of DT signal x(n) is defined by:

X
1
Z fx ð nÞ g ¼ xðnÞzn ¼ X ðzÞ:
n¼1

The set of values for which the summation above converges is called the region of convergence (ROC). Generally speaking,
the ROC is an annular region of the entire complex plane defined by r1 < |z| < r2. Two important types of sequences are left-
handed and right-handed ones. A right-handed sequence is one for which x(n) ¼ 0 for all n < n0, where n0 is a positive or
negative finite integer. If n0  0, the resulting sequence is causal or a positive-time sequence. A left-handed sequence is one
for which x(n) ¼ 0 for all n  n0, where n0 is positive or negative finite integer. If n0  0, the resulting sequence is anticausal
or a negative-time sequence.
Suppose that a causal sequence x(n) can be written as sum of complex exponentials:

X
N
x ð nÞ ¼ ðai Þn uðnÞ:
i¼1

By applying the definition equation and geometric progression formulas, the corresponding z-transform can be written as:

X
N
z
X ðzÞ ¼ ,
i¼1
z  ai

where the ROC of i-th term is given as the exterior of a circle with radius jai j : Ri ¼ {z:| z| > | ai| }. The overall ROC is
obtained as intersection of ROCs Ri as:

ROC : R ¼ \Ni¼1 Ri ¼ fz : jzj > max ðjai jÞg:

Suppose now that an anticausal sequence x(n) can be written as a sum of time-shifted complex exponentials:

X
N
x ð nÞ ¼ bni uðn  1Þ:
i¼1
Appendix 769

Im{z}
z-plane

min | bi |
max | ai | x
x x Re{z}
x x x
x Right-hand x
poles
x
Lef t-hand Lef t-hand
poles poles

ROC

Fig. A1 Determination of the ROC

By applying again the definition equation and geometric progression formulas, the corresponding z-transform becomes:

X
N
z
X ðzÞ ¼ ,
i¼1
z  bi

where the ROC of i-th term is given as the interior of a circle of radius jbi j : Li ¼ {z:| z| < | bi| }. The overall ROC is obtained
as intersection of ROCs Ri by:

ROC : L ¼ \Ni¼1 Li ¼ fz : jzj < min ðjbi jÞg:

Finally, suppose now that a sequence x(n) contains both right- and left-handed sub-sequences. Its ROC is determined by
intersection of right- and left-handed subsequences’ ROCs as follows:

ROCtotal ¼ R \ L ¼ fz : max ðjai jÞ < jzj < min ðjbi jÞg,

and corresponding illustration is shown in Fig. A1.

A.6.2 Properties of z-Transform and Common z-Transform Pairs

The basic operations on sequences include the addition, the time-shifting (or translation), multiplication, and convolution. The
corresponding properties are summarized in Table A1. Some common z-transform pairs are summarized in Table A2.

A.6.3 The Inversion of the z-Transform

The inverse z-transform is defined as:

1
x ð nÞ ¼ ∳X ðzÞzn1 dz,
2πj C

where C is the closed contour (observed counterclockwise, encircling the origin) within the ROC. The methods commonly
used to evaluate the inverse z-transform are (i) contour integration, (ii) partial-fraction expansion, and (iii) division. The
contour integration method is based on the Cauchy’s formula:
770 Appendix

Table A1 The z-transform properties

Property Time Domain Z-Domain ROC
Notation x(n) X(z) ROC:r2 < |z| < r1
x1(n), x2(n) X1(z), X2(z) ROC1, ROC2
Linearity a1x1(n) + a2x2(n) a1X1(z) + a2X2(z) ROC ¼ ROC1\ROC2
Time shifting x(n-k) z-kX(z) ROC: the same as X(z) except z ¼ 0 if k > 0, or z- > 1 if
k<0
Scaling in the z-domain anx(n) X(a1z) |a|r2 < |z| < |a|r1
Time reversal x(n) X(z1) 1/r1 < |z| < 1/r2
Conjugation x*(n) X*(z*) ROC
Real and imaginary parts Re{x(n)}, Im{x(n)} [X(z) + X*(z*)]/2, j[X(z)- Includes ROC
X*(z*)]/2
Differentiation in the nx(n) -zdX(z)/dz ROC
z-domain
Convolution x1(n)*x2(n) X1(z)X2(z) At least ROC1\ROC2
Correlation rx1, Rx1,x2(z) ¼ X1(z)X2(z1) At least intersection of ROCs for X1(z) and X2(z1)
x2(l) ¼ x1(l ) x2(l )
*

Initial value theorem If x(n) is causal xð0Þ ¼ lim X ðzÞ

z!1


Multiplication x1(n)x2(n) 1
2πj ∳X 1 ð vÞX 2 vz v1 dv At least: r1lr2l < |z| < r1ur2u
C
Parseval’s relation P
1
1
x1 ðnÞx2 ðnÞ ¼ 2πj
1
∳X 1 ðvÞX 2 v v1 dv
n¼1 C

Table A2 The common z-transform pairs

Signal, x(n) Z-transform, X(z) ROC
(n) 1 All z
u(n) 1/(1-z1) |z| > 1
anu(n) 1/(1-az1) |z| > |a|
nanu(n) az-1/(1-az1)2 |z| > |a|
anu(n-1) 1/(1-az1) |z| < |a|
nanu(n-1) az-1/(1-az1)2 |z| < |a|
cos(ωn)u(n) (1-z1cosω)/(1-2z1cosω + z2) |z| > 1
sin(ωn)u(n) z1sinω/(1-2z1cosω + z2) |z| > 1
ancos(ωn)u(n) (1-az1cosω)/(1-2az1cosω + a2z2) |z| > |a|
ansin(ωn)u(n) az1sinω/(1-2az1cosω + a2z2) |z| > |a|


1 f ðzÞ f ðz0 Þ, z0 inside of C
∳ dz ¼ ,
2πj C z  z0 0, z0 outside of C

where C is a closed path, while f’(z) exists on and inside of C. If there are (k + 1)-th order derivatives of f(z), for poles of
multiplicity k that are enclosed by the contour C, while f(z) does not have the poles outside C, then the following is valid:
8
< 1 dk1
1 f ðzÞ ½ f ðzÞjz¼z0 , z0 inside of C
∳ dz ¼ ðk  1Þ! dzk1
2πj C ðz  z0 Þk :
0, z0 outside of C

The partial-fraction method can be used to determine the inverse z-transform when X(z) is expressed as a rational function
of z. The key idea behind this method is to express X(z) as a sum:

X ðzÞ ¼ α1 X 1 ðzÞ þ α2 X 2 ðzÞ þ . . . þ αK X K ðzÞ,

where the inverse z-transforms of i-th terms (i ¼ 1,. . .,K) are available in the table of z-transform pairs (see Table A2).
References 771

References

1. I.B. Djordjevic, Quantum Information Processing and Quantum Error Correction: An Engineering Approach (Elsevier/Academic, Amsterdam/
Boston, 2012)
2. I. Djordjevic, W. Ryan, B. Vasic, Coding for Optical Channels (Springer Science+Business Media, New York, 2010)
3. W.E. Ryan, S. Lin, Channel Codes: Classical and Modern (Cambridge University Press, Cambridge/New York, 2009)
4. S. Lin, D.J. Costello, Error Control Coding: Fundamentals and Applications, 2nd edn. (Pearson Prentice Hall, Upper Saddle River, 2004)
5. C.C. Pinter, A Book of Abstract Algebra (Dover Publications, New York, 2010). (reprint)
6. J.B. Anderson, S. Mohan, Source and Channel Coding: An Algorithmic Approach (Kluwer Academic Publishers, 1991)
7. P.A. Grillet, Abstract Algebra (Springer, 2007)
8. A. Chambert-Loir, A Field Guide to Algebra (Springer, 2005)
9. S. Lang, Algebra (Addison-Wesley Publishing Company, Reading, 1993)
10. F. Gaitan, Quantum Error Correction and Fault Tolerant Quantum Computing (CRC Press, Boca Raton, 2008)
11. D. Raghavarao, Constructions and Combinatorial Problems in Design of Experiments (Dover Publications, Inc., New York, 1988). (Reprint)
12. D.B. Drajic, P.N. Ivanis, Introduction to Information Theory and Coding (Akademska Misao, Belgrade, 2009). (in Serbian)
13. M. Cvijetic, I.B. Djordjevic, Advanced Optical Communications and Networks (Artech House, Boston/London, 2013)
14. S.B. Wicker, Error Control Systems for Digital Communication and Storage (Prentice Hall, Englewood Cliffs, 1995)
15. B. Vucetic, J. Yuan, Turbo Codes-Principles and Applications (Kluwer Academic Publishers, Boston, 2000)
16. Infrared Data Association. Infrared Data Association serial infrared physical layer specification. Version 1.4, available at: www.irda.org, 2001
17. S. Hranilovic, Wireless Optical Communication Systems (Springer, Boston, 2005)
18. H. Hemmati, A. Biswas, I.B. Djordjevic, Deep-space optical communications: Future perspectives and applications. Proc. IEEE 99(11),
2020–2039 (2011)
19. S.G. Wilson, M. Brandt-Pearce, Q. Cao, M. Baedke, Optical repetition MIMO transmission with multipulse PPM. IEEE Sel. Areas Commun.
23(9), 1901–1910 (2005)
20. S.J. Dolinar, J. Hamkins, B.E. Moision, V.A. Vilnrotter, in Optical Modulation and Coding, in Deep Space Optical Communications, ed. by
H. Hemmati, (Wiley, 2006)
21. I.B. Djordjevic, Multidimensional pulse-position coded-modulation for deep-space optical communication. IEEE Photon. Technol. Lett. 23(18),
1355–1357 (2011)
22. S. Haykin, B. Van Veen, Signals and Systems, 2nd edn. (Wiley, 2003)
23. J.G. Proakis, D.G. Manolakis, Digital Signal Processing: Principles, Algorithms, and Applications, 4th edn. (Prentice-Hall, 2007)
24. L.L. Ludeman, Fundamentals of Digital Signal Processing (Harper & Row, Publishers, Inc., New York, 1986)
 Index

A Average fade duration (AFD), 110–112, 110–112

Access network, 6, 160 Average power constraint, 742
A configuration of OFDM modulator for DAB, 415 Average probability of error, 728
Acousto-optical filters, 176 Average symbol error probability, 215, 452–460, 521, 215,
Acquisition, 655–658, 664, 665, 706, 707 452–460, 521
Adaptive coded modulation, 353–354, 372, 380, 381, 625–631
Adaptive coding, 348, 372, 381, 625–631, 638, 639, 641
Adaptive equalization, 335–337, 341, 342 B
Adaptive hybrid FSO-RF coded modulation, 629–631, 629–631 Backward metric, 559, 605, 606
Adaptive loading, 274 Backward recursion, 556, 558, 560, 561, 581, 582, 605, 606
Adaptive modulation Balanced coherent optical detection for 2-D modulation schemes,
power adaptation by channel inversion, 347, 347, 350 358–371, 358–362, 359–363
truncated channel inversion policy, 348 Bandwidth efficiency, 255, 256, 259, 264, 265, 272
variable-power modulation technique, 347 Basic 2  2 optical hybrid with two phase trimmers, 359
variable-rate modulation technique, 347 Bayes’ rule, 496
Adaptive modulation and coding (AMC), 235, 237, 241, 242, 235, BCH codes, 534–536, 549, 550, 553, 561, 572, 630
237, 241, 242 BCJR-algorithm-based equalizer, 602, 606–608, 614, 638, 602, 606–
Adaptive nonbinary LDPC-coded multidimensional 608, 614, 638
748, 754, 755, 10, 12, 595, 631, 479–486, 521, 522, 524, 748, BCJR decoder, 556, 580
749, 501–502, 521, 364, 365, 488, 550, 556, 559, 562, 570, 596, Bell Laboratories Layered Space-Time (BLAST) architectures
628, 634, 639, 640modulation, 628, 631, 628, 631 diagonal BLAST (D-BLAST), 482, 483, 491
Adaptive QAM-based subcarrier mapping, 441 horizontal BLAST (H-BLAST), 482, 483, 482, 483
Additive white Gaussian noise (AWGN), 532, 533, 547, 549, vertical BLAST (V-BLAST), 482, 491
532, 533, 547, 549, 550, 556, 559, 562, 570, 596, 628, 634, Binary LDPC codes, 563–564, 573–574, 578, 580, 582, 596, 630
639, 640 Binary phase shift-keying (BPSK), 653, 657, 658, 696, 700–702, 707,
Advantage distillation, 726 653, 657, 658, 696, 700–702, 707
Alamouti code, 476–478, 521 Binary symmetric channel (BSC), 540, 547, 548, 631, 633, 540, 547,
Alamouti code (revisited), 462, 476–478, 521 548, 631, 633
Alamouti scheme, 462–464, 476, 478, 522 Bipartite (Tanner) graph, 499, 564, 565
All-zero prefix, 397 Bi-phase modulation (BPM), 700–702, 707
Amplification factor, 172, 173 Bit error probability, 313, 315–316, 318, 322, 325, 347, 348, 354, 368
Amplified spontaneous emission (ASE) noise, 158 Bit-error rate (BER), 533, 548, 550, 551, 562, 569–571, 576, 577, 579,
Andrews spectral density, 74 580, 582, 583, 589, 602, 608–612, 614, 617, 621–623, 627, 629,
Antenna arrays, 196–199, 202, 676 633–636, 638–640
Antipodal signaling, 315, 316 Bit-interleaved coded modulation (BICM), 583, 592, 594, 596, 631,
Antireflection (AR) coatings, 173 583, 592, 594, 596, 631
Arithmetic coding-based nonuniform signaling, 291–294 BLAST receivers (interfaces), 488–491, 497, 520, 521, 488–491, 497,
Array factor (AF), 198, 199 520, 521
Array gain (AG), 275, 451, 452, 454, 461, 463, 471 Blind equalization (BE)
Array waveguide grating (AWG), 177, 178, 695, 177, 178, 695 Benveniste-Goursat’s function, 342
Asymmetric split-step method (ASSM), 64–66, 69, 64–66, 69 Godard’s algorithm or constant-modulus algorithm (CMA), 341,
Asymptotic coding gain, 548, 549, 583, 586, 589 341, 342
Asynchronous CMDA, 670–679, 682, 686, 706 high-order statistics based BE, 340
Asynchronous coherent optical detection, 179 maximum-likelihood approach based BE, 340
Atmospheric turbulence, 63–85, 133, 149, 150 Sato’s function, 341
Autocorrelation constraint, 693, 694 steepest descent algorithms based blind equalization, 342
Autocorrelation function of the impulse response, 114 Block coded modulation (BCM), 582, 583, 582, 583
Autocorrelation function of the spreading sequence (code), 655 Block coding with interleaving, 551, 551, 552
Automatic gain control (AGC), 157–183, 200–202, 157–183, 200–202 Block-interleaved coded modulation, 592, 596, 598, 592, 596, 598
Avalanche photodiode (APD), 13, 168, 169, 202, 204 Block-interleaver, 398
Average error probability, 311–313, 315, 317, 321–324, 380, 381 Block turbo code (BTC), 561, 562

# The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 773
I. B. Djordjevic, Advanced Optical and Wireless Communications Systems, https://doi.org/10.1007/978-3-030-98491-5
774 Index

Booster amplifier, 173 punctured LDPC coding, 733, 734

BPM (BPSK)-based UWB communication, 702 reliability region, 734
Branch metric, 480, 539, 559–561, 587, 605, 606 secure region, 734
Broadband MIMO, 502–505, 521 sequential decoding, 735
Broadcast channel with confidential messages (BCC), 721–724, 743, two-edge type LDPC coding, 733
754, 755, 721–724, 743, 754, 755 Coding gains, 532, 533, 576, 584–586, 626, 627, 629, 633, 634, 636,
Burst error codes, 549–551 639, 640
Coding with interleaving in time- and frequency-domains, 274
Coefficients of reflections, 36
C Coefficients of transmission, 36
Calculation of information capacity by the forward recursion Coherence, 51–63, 77–79, 115–118, 149
of the BCJR algorithm, 230–231 Coherent balanced optical detector based on 2  2 optical hybrid, 360
Capacity of channels with memory, 224–230 Coherent balanced optical detector for demodulation of 2-D signals
Capacity of continuous channels, 211–214, 242, 211–214, 242 based on 2  4 optical hybrid, 360, 361
Capacity of deterministic (static) MIMO channels, 505–507, 505–507 Coherent optical (CO) OFDM (CO-OFDM), 389, 393, 395, 400,
Capacity with outage, 215, 216, 242 404–411, 430–432, 389, 393, 395, 400, 404–411, 430–432
Cascade-integrator-comb (CIC) filter, 192, 192, 193 Coherent optical balanced detectors, 180–183, 202, 180–183, 202
Cascade realization, 186 Coherent optical detection, 160, 178–179, 181, 202
Catastrophic CC, 545 Coherent optical detection based OCDMA, 706
Cell coverage area, 100–101, 151 Coherent optical OFDM-MIMO, 466
Cellular concept, 2, 12, 13, 17 Collision entropy, 740
Channel capacity, 209–224, 228, 230, 232, 234, 235, 237, 240–242, Combined outage and average error probability scenario, 321, 323,
244–246 321, 323
Channel capacity with outage, 216 Communication hardware revolutions, 2
Channel coding theorem, 210 Compensation of atmospheric turbulence effects
Channel distribution information, 215 adaptive coded modulation, 371, 372, 380
Channel equalization, 325–343, 380 adaptive coding, 372, 381
Channel estimation in OFDM, 407–411 adaptive optics approaches, 371, 372, 381
Channel inversion, 219, 220, 237–240, 242, 245 control processor, 373
Channel inversion-based discrete rate adaptation, 352 diversity techniques, 372
Channel matrix, 464, 465, 468, 469, 471, 472, 476–478, 480, 482, 484, hybrid RF-FSO communication, 372
487, 488, 490, 494, 497, 501, 504–513, 515, 518–520, 522–524 MIMO signal pro-cessing, 372, 378, 381
Channel modelling for OFDM, 399–401, 403, 404 Shack-Hartmann sensor, 375–379
Channel side information at receiver (CSIR), 215, 216, 246, 215, spatial light modulator (SLM)-based back-propagation method, 372
216, 246 wavefront corrector, 373, 376–378
Channel side information or channel state information (CSI), 214–221, wavefront sensor, 375, 376, 378, 379
224, 242, 244–246, 214–221, 224, 242, 244–246 Zernike polynomials, 373, 374, 376
Characteristic polynomial, 661, 662 Zernike representation of atmospheric turbulence, 372–377,
Chase II decoding algorithm, 562 373–379
Check nodes, 499, 564–566, 568, 571, 577, 578, 581, 633 Complete path enumerator, 547
Chip rate, 650, 707 Compound channels, 742
Chip time, 650, 651, 664 Computational security, 715–717, 754
Cholesky factorization, 675, 687 Concatenated codes, 549–550, 633
Chromatic dispersion, 160, 179, 204, 205 Concatenated (serial) configuration of 16-star-QAM transmitter, 267
material dispersion, 119 Connector return loss, 119
waveguide dispersion, 119 Constellation expansion factor, 583
Chromatic dispersion compensator, 332, 332 Constraint length, 541
CIC filter-based upsampling and downsampling, 192, 192, 193 Continuous-time (CT) PAM transmitter and receiver, 256
Circular polarization, 29 Conventional single-user correlator receiver, 679–682, 679–682
Clark-Jakes model, 106–107, 151 Convolutional codes (CC), 540–549, 553, 576, 582, 585, 587, 630
Code-division multiple access (CDMA), 2, 10, 12, 16, 17, 648–707, Convolutional encoder configuration and puncturing mechanism, 417
709, 2, 10, 12, 16, 17, 648–707, 709 Convolutional interleaver, 552
Coded modulation, 529–642 Convolutional LDPC code, 576
Coded-OFDM in digital video broadcasting (DVB), 413, 417, 442, CO-OFDM operation principles, 404–411
417–425, 417–425 Cooperating jamming, 745
Code rate, 540, 541, 548, 550, 553, 554, 556, 561, 564, 572, 574, 576, Core network, 6, 160
577, 579, 580, 582, 587, 589, 596, 599, 600, 607, 608, 626, 628, Correlation detector, 305, 308
629, 632, 638, 639 Correlation receiver, 253–255
Codes on graphs, 553, 630 Correlator bank-based decorrelating detector, 688
Coding for physical-layer security systems Cosets, 584, 587, 626, 627, 632, 633
coding for strong secrecy systems Cramér-Rao bound, 320
coset coding with dual of LDPC codes, 737, 737 Crosscorrelation constraint, 693
hash functions and extractors based coding, 737 Cross-correlation function of the in-phase and quadrature
coding for weak secrecy systems components, 106
degree distribution polynomials, 733 Cryptogram, 715, 716
polar codes, 735, 736 Cryptography, 713, 716
Index 775

CW laser based OCDMA, 695–697, 706 Discrete-time (DT) CDMA models, 674–679
Cyclic extension, 389–396, 398, 406, 427, 429, 440, 442 Discrete-time (DT) PAM transmitter and receiver, 258
Cyclic prefix, 395, 397, 406, 442, 443 Discrete-time system realizations, 185–187, 185–187
Cylindrical polar coordinate system, 25 Dispersion compensation
NZDSF with large effective area
DCF, 120
D guided modes, 123
Data-aided channel estimation, 407, 411, 442, 407, 411, 442 leaky modes, 123
3 db-coupler, 176, 177, 180 optical mode, 123
De Bruijn sequences, 660 radiation modes, 123
Decision boundaries, 252, 288, 298 weak guidance condition, 123
Decision-feedback equalizer (DFE), 337, 338, 379, 337, 338, 380 NZDSF with reduced dispersion slope, 120
Decision-feedback MIMO receivers for uncoded signals, 487, 488, Distributed Bragg reflector (DBR), 8, 8, 13
496–498, 521, 487–488, 487–488 Distributed feedback (DFB) lasers, 8, 8, 13
Decision-feedback receivers (interfaces) for coded signals, 487, Diversity gain, 275, 451, 452, 457, 462, 464, 471, 475, 483, 503
487, 488, 521 Diversity techniques
Decision region, 252, 296, 298 directional/angle diversity, 450
Decoder of a turbo product code, 561 frequency diversity, 450, 502, 503, 521
Decoding of binary LDPC codes, 565–568, 630, 565–568, 630 polarization diversity, 450, 463, 482, 521
Decoding of nonbinary LDPC codes, 570–571, 570–571 space/spatial diversity, 451
Decorrelating detector for synchronous CDMA, 687, 688, 687, 688 time diversity, 450
Decorrelating receiver, 687, 690, 706 Dominant coherent optical detection noise sources, 358, 367–371, 359,
Decryption, 749 368–372
Degree distribution polynomials, 564 Doppler shift, 86–87, 102, 116
Degree of polarization, 31 Doppler spread, 324–325, 380
Delay-locking loop (DLL), 657, 658 Downlink channel, 10, 668
Delay spread, 11 Downsampling, 189–194, 202
Demodulator, 249, 252–257, 262–264, 285, 294, 298 DS-SS parallel search acquisition, 656
Design of IIR filters (subsystems), 157–207, 157–183, 185, 187, 195, DS-SS serial search acquisition, 656
202, 206 Duobinary scheme, 329
Design of LDPC codes
Gallager codes, 571
generalized LDPC (GLDPC) codes, 571, 572, 571, 572 E
large-girth quasi-cyclic (QC) binary LDPC codes, 573, 573, 574 Edge network, 6, 160
MacKay codes, 572, 573 Electrical scalar potential, 40, 42, 40, 42, 43
Tanner codes, 571, 572 Electro-absorption modulator (EAM), 8, 13, 160, 164, 165, 202, 8, 13,
Detection theory, 304–320 160, 164, 165, 202
DFT windowing, 404–411, 442 Electromagnetic potentials, 40–51, 149
Differential detection in OFDM Electromagnetic waves, 26–85, 124, 149
frequency-domain differential encoding, 412 Elliptical polarization, 28, 37
time-domain differential encoding, 411 Encoder for a rate (m 1)/m TCM scheme, 588
Differential entropy, 211–213 Encryption, 715, 716, 749
Diffraction-based optical filter, 177 Energy-efficient signal constellation design, 617, 290–291, 295, 299,
Diffraction-based WDM demultiplexers, 177 290–291, 295, 298
Diffraction in optics, 51–63 Entropy, 209, 210, 212, 214, 224, 227–229, 231, 242, 245
Digital audio broadcasting (DAB), 414, 413–416, 442, 443, 413–416, Equal gain combining (EGC), 452, 459–461, 522
442, 443 Erbium-doped fiber amplifiers (EDFAs), 158, 160, 174, 175, 205, 158,
Digital backpropagation in fiber-optics communications, 355–358, 160, 174, 175, 205
356–359 Ergodic capacity, 215, 245
Digital signal processing (DSP), 183–191, 193, 202, 183–191, Ergodic Markov chain, 226
193, 202 Error correcting capability of LBC, 533
Digital video broadcasting (DVB), 413, 417, 442 Error-evaluator polynomial, 536
Direct canonic forms, 185, 186 Error event, 547
Direct detection receiver, 159, 203 Error locator polynomial, 536
Direct digital synthesizer (DDS), 187–189, 202, 187–189, 202 Error polynomial, 535
Directional coupler-based coherent optical detection schemes, 181 Estimation theory, 304, 319–320, 380
Directivity gain, 48, 49 Euclidean distance criterion, 471–476, 471–476
Direct modulation, 158, 163 Euclidean distance receiver for wireless communications, 253
Direct optical modulators, 8 EXIT chart analysis of OSCD mapping rules, 617, 631, 617, 631
Direct sequence-spread spectrum (DS-SS) systems, 651–663, EXIT charts, 592–594, 617, 620
651–663 Exponential distribution of arrival time, 703
Discrete memoryless channel (DMC), 209, 210, 209, 210 External modulation, 159, 163
Discrete-power discrete-rate adaptive modulation technique, 352 External modulators, 8, 13, 159, 160, 164–165, 167, 175, 202–204
Discrete-rate adaptation, 352 Extrinsic absorption, 118
Discrete-time (DT) and continuous-time (CT) Fourier Extrinsic information, 555, 561–564, 566, 568, 569, 580, 604, 633, 634,
transforms, 183 636, 639
776 Index

F Frequency-hopping–spread spectrum (FH-SS) systems, 650, 651, 650,

Fabry-Perot (FP) filters, 176 651, 663–669, 706
Fabry-Pérot (FP) lasers, 8, 8, 13 Frequency-selective, 103
Factor graph for MAP detection of uncoded signals for MIMO Frequency-selective channels, 502–504, 502–504
channels, 497 Frequency-selective fading, 116
Factor graph for MIMO receiver with LMMSE interface, 497 Frequency-selective fading channel capacity, 221–224, 221–224, 242
Factor graph for MIMO receivers with H/V-BLAST architecture, 498 Frequency-shift keying (FSK), 255–256, 258, 259, 295, 255–256, 258,
Factor graph for MIMO receiver with zero-forcing interface, 487–490, 259, 294
497, 487–490, 497 Frequency synchronization, 399, 404–411, 414, 442
Factor graph for space-time coded MIMO (with interleaver), 501 Fresnel diffraction, 58, 65, 66
Factor graph of a channel without memory (memoryless channel) Fresnel-Kirchhoff integral formula, 57–59, 57–59
495, 495 Fresnel knife-edge diffraction model, 91
Factor graph of the linear ISI channel (linear channel with Fresnel’s equation, 36
memory), 494 Fried parameter, 75
Factor graph of the optical/wireless MIMO channel, 494 Frobenius (Euclidean) norm, 473
Factor graphs’ theory, 492–495, 497–500, 521, 492–495, 497–500, 521 Fully correlated model, 472
Fade margin, 321, 384 Fundamental mode, 171
Fading channel reciprocity, 747
Far-field region/Fraunhofer region, 43
Fast-frequency hopping (FFH) systems, 651, 664, 651, 664 G
FEC frame used in DVB-T2, 423 Gain saturation, 172
Feedback shift register, 659 Galois field (GF), 529, 534, 535, 537, 549, 550, 563, 570, 574–576,
FFT window synchronization, 399 578, 599, 601, 603, 604, 611, 628, 632, 633
FH serial search acquisition, 665 Gamma-gamma distribution, 79–85, 150
Fiber attenuation, 118–119, 132, 149, 152 Generalized fading model, 110
Fiber Bragg grating (FBG), 694, 695 General ray tracing (GRT) model, 91–92, 91–92
Fiber insertion losses, 118, 119 Generating function, 661
Fiber nonlinearity, 5, 9, 13, 15 Generating function of CC, 545–547
anomalous dispersion region, 121, 121, 129 Generator matrix, 530, 531, 538, 542–545, 554, 563, 564, 591, 632
cross-phase modulation (XPM), 127, 129, 130, 127, 129, Generator polynomial, 534, 535, 543, 545, 549
130, 149 Generator sequences, 542
effective cross-sectional area, 96, 127, 128, 96, 127, 128 Geometric representation of received signals, 304–308, 378,
four-wave mixing (FWM), 127, 130, 131 304–308, 380
generalized nonlinear Schrödinger equation (GNLSE), 129, Geometric representation of signals, 250–251, 250–251
133–137, 129, 133, 149 Gilbert-Varshamov bound, 533
Kerr effect, 127–129 Girth, 496, 497, 499, 501
Lorentzian spectral profile, 132 Gold sequences, 659, 671, 672, 707
nonlinear optical scattering, 127 Guard time, 389–397, 442, 444
Raman threshold, 132
self-phase modulation (SPM), 127–129, 149, 127–129, 149
wavelength conversion through the process of phase , H
conjugation,131 Hadamard matrices, 672
Fiber splices and fiber connectors, 118 Half-wave plate (HWP), 38
Finite impulse response (FIR) systems (filters), 185 Hamming bound, 533, 632
Finite-state Markov channel (FSMC) model, 113, 113 Hamming code, 531–533, 553, 572, 582, 584, 632, 633
Flat-fading channel capacity, 214–221 Hamming cross-correlation, 666
Forward metric, 559, 605, 606 Hamming distance, 531, 532, 534, 540, 545, 584, 587, 632
Forward recursion, 556–558, 560, 561, 606 Helmholtz equations, 41, 47, 81
Fourier split-step beam propagation method, 64 Hertzian dipole, 45–46
Four-level pumping schemes, 161 Heterodyne detection, 179
FPGA-based rate-adaptive coding High-impedance receiver front end, 202, 167–169, 202, 167–169, 202
rate-adaptive GLDPC coding in FPGA, 580, 582, 580, 582 Homodyne coherent optical detection based on PLLs, 358,
rate-adaptive LDPC coding in FPGA, 576–580, 576–580 359, 365–367
FPGA implementation of binary LDPC decoders, 568–569, 568–569 Homodyne detection, 179, 180, 183
4-D signaling, 277–281, 295 Huffman code, 292, 293
40Tb/s transoceanic transmission enabled by 8-ary OSCD, 622–624, Huygens’ principle, 57
622–624 Hybrid adaptive modulation techniques, 348
Fractionally spaced equalizer, 334, 334, 338 Hybrid free-space optical (FSO)-RF channel capacity, 235–241
Fraunhofer diffraction, 62, 65–67, 149 Hybrid multidimensional coded modulation, 595–602, 631,
Free distance of CC, 545, 547, 585 595–602, 631
Free distance of TCM, 585 Hybrid OFDM-CDMA, 696–700, 706, 709
Free-space optical (FSO) PLS, 748, 748, 750
Free-spectral range, 162, 173, 177, 178
Frequency acquisition by using the cyclic prefix, 406 I
Frequency-division multiple access (FDMA), 668, 668 IEEE 802.11 PHY standards, 426
Frequency-domain equalization, 274 Impulse radio UWB (IR-UWB), 429, 700, 703, 707
Index 777

Incoherent OCDMA, 706 L

Individually optimum decision rules, 683 Laser beam propagation, 63–85, 149, 150
Indoor optical wireless communication Laser operation principle, 161
diffuse link, 145, 146 Laser phase noise, 137–140, 149, 344, 346, 359, 365, 368, 369, 381,
point-to-point link, 145, 146 383, 384
quasi-diffuse link, 145–147 Lasers
Infinite impulse response (IIR) systems (filters), 185 distributed Bragg reflector (DBR) lasers, 160, 162, 163, 160, 162,
Information capacity, 212–214, 229–234, 242, 246 163, 202
Information capacity of systems with coherent optical detection, distributed feedback (DFB), 160, 162, 163, 202, 160, 162, 163, 202
232–234 Fabry-Perot laser, 160–162, 173, 176, 202, 203
Information capacity theorem, 209–214, 209–214, 242 Lattice, 583–585
Information leakage to the Eve, 728, 728, 731 LDPC codes, 553, 563–582, 589, 596, 597, 599–602, 611, 629, 630,
Information reconciliation, 714, 726, 728, 731, 732, 737–739, 741, 746, 633–636, 638, 639
749, 750, 754, 756 Legendre polynomials, 701, 707
Information-theoretic secrecy metrics Level crossing rate, 111
mutual information, 716 Light-emitting diodes (LEDs), 8
variational distance, 716 Lightwave path, 7
Information-theoretic security, 716, 752, 754, 716, 752, 754 Likelihood equation, 320
In-line amplifier, 172, 173 Likelihood function, 310, 312, 319, 320, 338–340, 345, 346
Interference, 51–63, 101, 104, 105, 113, 143, 144, 149 Linear block code (LBC), 485, 498, 529, 531–533, 537, 541, 553, 564,
Interference-based optical filter, 177 572, 576, 580, 587, 631–633
Interference-based WDM demultiplexers, 177 Linear combiners, 451, 452
Interleaved code, 550 Linear feedback-shift register (LFSR), 660, 660
Interleaving degree, 551 Linear interface with nonlinear decision-feedback, 489, 489, 490
Intermodal dispersion, 119, 151 Linear MIMO receivers for uncoded signals, 487, 488
Internal reflection, 28–40, 122, 149 Linear minimum mean-square error (MMSE) receiver, 706
Intersymbol interference (ISI) channel, 494 Linear minimum MSE (MMSE) multiuser detection, 688
Intradyne detection, 179 Linear MMSE receiver (interface), 489, 489
Intrinsic absorption Linear polarizer, 37, 39
infrared absorption, 118 Linear receivers (interfaces), 487, 488, 487, 488, 497
ultraviolet absorption, 118 Linear space-time block codes, 476, 478–480, 488, 521, 476, 478–480,
Inverse fast Fourier transform (IFFT), 391, 395, 396, 398, 410, 417, 488, 521
420, 424, 429, 439–441, 391, 395, 396, 398, 410, 417, 420, 424, LLR receivers, 310–313
429, 439–441 LMS algorithm, 337, 338
Inversion function, 670 Locator field, 534, 549
I/Q modulator, 8, 13, 165–167 Log a posteriori probability (log-APP), 555, 562, 555, 562, 566
Irregular LDPC codes, 571, 572 Log-domain BCJR algorithm, 558–560
ISI-free digital transmission, 325–329, 325–329 Log-domain SPA for source coding with the side information, 739
Isoplanatic angle, 77 Log-domain sum-product algorithm (SPA), 565–567, 565–567
Iterative MIMO receivers, 492–502, 521 Log-likelihood ratio (LLR), 555, 556, 558, 560, 562, 563, 565–567,
Iterative MIMO receivers for channel coded signals, 498 577, 578, 580, 592, 604, 605, 616, 617, 621
Iterative MIMO receivers for space-time coded signals, 498, Log-normal distribution, 84, 151
498, 501–502, 521 Long-term evolution (LTE), 426–428, 442
Iterative MIMO receivers for uncoded signals, 496–498, 521
Iterative polar modulation (IPM), 288–289, 291, 295, 288,
289, 291, 295 M
Iterative receiver (interface), 488–492, 497, 521, 488–492, 497, 521 Mach-Zehnder (MZ) interferometer filters, 176
Iverson function, 493, 499, 500 Mach-Zehnder modulator (MZM), 8, 13, 160, 164–166, 202, 204, 205,
8, 13, 160, 164–166, 202, 204, 205
Magnetic vector potential, 40–45, 47, 49–51
J Malus law, 33
Jamming margin, 654 MAP detection problem, 496
Joint source-channel key generation, 746, 746, 747 Marginalization operation, 492, 493
Jointly optimum detector for K-user synchronous CDMA, 685 Markoffian approximation, 139
Jointly optimum MUD, 706 Markov sources, 225, 227, 242, 245
Jones vector representation, 29 M-ary PSK, 458, 459
M-ary PSK transmitters for optical communications,
concatenation (serial connection) of M electro-optical (E-O)
K phase modulators, 265
Kasami sequences, 659, 670, 671 combination of E-O I/Q and phase modulators, 265
Key-generation (key-distillation) protocol, 727, 727, 730 single E-O I/Q modulator, 265
Kirchhoff integral theorem, 56 M-ary QAM, 457, 458, 461
Kolmogorov refractive-index spectral density, 74 Mason’s gain formula, 547
Kolmogorov theory of turbulence, 69–77, 69–77 Massey-Berlekamp algorithm, 534–537, 630
K-user successive cancelation, 692, 692 Matched filter based detector, 308
Matched filters-based demodulator/detector, 254, 254, 255
778 Index

Maximal-length sequence/m-sequence, 660 Multichannel propagation, 135

Maximum a posteriori probability (MAP) decision rule, 683 Multidimensional optimized signal constellation design, 617
Maximum a posteriori probability (MAP) detection/decoding, 492 Multidimensional signaling in fiber-optic communications, 281–285
Maximum a posteriori probability (MAP) estimate, 319 Multidimensional signaling in FSO communications, 282, 286–288,
Maximum a posteriori probability (MAP) rule, 311, 311, 555 282, 285–287
Maximum-distance separable (MDS) codes, 533, 533, 549 Multidimensional turbo equalization, 604–606, 631, 602–614, 631,
Maximum likelihood (ML) decision rule, 683 602–614, 631
Maximum likelihood (ML) decoder decision rule, 539 Multi-group space-time coded modulation (MG-STCM), 484–486, 521,
Maximum likelihood (ML) detection, 472 484–486, 521
Maximum likelihood (ML) estimate, 319 Multi input multi output (MIMO), 3–5, 11, 13–17, 3–5, 11, 13–17
Maximum likelihood sequence detection (estimation) (MLSD or Multilevel coded modulation (MCM), 587–592, 602, 587–592, 602
MLSE), 338 Multilevel coding (MLC), 582, 587, 589–592, 594–596, 598, 626,
Maximum-ratio combining (MRC), 452, 456–462, 521, 452, 631, 633
456–462, 521 Multilevel encoder for MLC/MCM scheme, 589
Maxwell’s equations, 21–24, 27, 40, 122, 123, 148, 149 Multilevel nonbinary LDPC-coded modulation (ML-NB-LDPC-CM),
McMillan-Khinchin model, 228, 230, 242 596, 599–602, 596, 599–602
McMillan sources, 228, 242 Multimode dispersion (or intermodal dispersion), 160, 171, 160, 171
Metal-semiconductor-metal (MSM) photodetector, 8, 13, 8, 13, 168 Multipath fading, 10, 11, 13–17
MGF-based average error probability calculation, 323–324 Multipath signal components, 89
Michelson filters, 176 Multiple input multiple output (MIMO), 275–278, 281, 282, 287, 295,
Mie scattering, 118 299, 449–525, 275–278, 281, 282, 286, 295, 299, 449–525
MIMO channel capacity, 505–511, 521 Multiplexed coding and decoding, 219, 242, 219, 242
MIMO channel estimation Multiplexing gain, 275–277, 295, 464, 468–470, 475, 502, 522
least squares (LS) MIMO channel estimation, 512, 521, 512, 521 Multiplexing techniques, 14
linear minimum MSE (LMMSE) MIMO channel estimation, 513, Multiplicative noise components, 137
513 Multirate DSP, 189–195, 202
maximum likelihood (ML) MIMO channel estimation, 511, 512, Multistage decoder, 589
511, 512 Multistage detection, 690, 692, 706
MIMO concept, 465, 514–517, 521 Multiuser detection (demodulation), 10, 10, 16
MIMO OFDM, 502–504, 510, 521, 522 Multiuser detection (MUD), 668, 679–692, 706
MIMO OFDM channel capacity, 510–511, 521 Mutual coherence function (MCF) of irradiance, 52, 52, 54, 77
MIMO optical channels, 470, 471, 522 Mutual information, 209–214, 224, 229, 230, 242, 243
MIMO wireless channels, 452, 507, 516, 522
MIMO wiretap channel, 741, 742
Min-entropy, 740 N
Minimal polynomial, 534 Nakagami distribution, 110
Minimum distance, 531–533, 549, 550, 556, 561, 562, 564, 569, 572, Narrowband channel model, 104, 105, 107
573, 583–585, 627, 632, 633 Nested code, 732
Minimum energy signal constellation, 314, 314, 384 Nested coding, 591
Minimum mean-square (MMS) estimate, 319 Net coding gain, 533, 550, 569, 582, 601
Minimum MSE combiner (MMSEC), 705, 705 Noble Identities, 192–194
Minimum squared Euclidian distance (MSED), 589, 589 Noise sources, 7, 9, 13, 17
Min-sum algorithm, 568–570 Nonbinary MAP detection, 603–604, 631
Min-sum-plus-correction-term approximation, 567 Nonbinary min-max algorithm, 570
Mobile telephone switching office (MTSO), 12, 12 Nonlinear channels with memory, 602–603, 631, 602–603, 631
Modal noise, 137, 140, 149 Nonlinear OSCD-based coded modulation, 631
Mode-multiplexed four-dimensional (4-D) coded-modulation, 481, Nonlinear propagation simulation of turbo equalization, 607, 608,
481, 482 607, 608
Mode-partition Noise, 137, 138, 149 Non-normal form of factor graph, 493
Modified duobinary scheme, 329 Non-overlapping subcarriers, 271, 272, 271, 272
Modified Hermite polynomials, 701, 701, 707 Nonuniform signaling, 291–295
Modified state diagram of CC, 546, 546, 547 Normal graph, 493
Modified von Kármán spectral density, 74, 79, 74, 79 Normalized frequency (V parameter or V number), 171, 171
Modulator, 249, 252–257, 262–270, 278–280, 283, 284, 286, 287, 293, Normalized propagation constant, 171, 171
294, 296, 298 Notch filter, 706
Moment generating function (MGF), 323–324, 457, 458, 323–324, Numerical aperture (NA), 37, 169, 170
457, 458 Nyquist’s criterion for zero-ISI, 326
Mueller matrix representation, 39, 39, 40
Multiaccess channel, 743
Multiaccess communication, 10, 13, 17, 668 O
Multiaccess interference (MAI), 10, 668, 10, 668, 679 OAM demultiplexer, 287
Multiband modulation, 703 OAM modulation, 283
Multiband OFDM (MB-OFDM), 429, 430, 700, 703–705, 707 OAM multidimensional modulator and multiplexer, 287
Multicarrier CDMA (MC-CDMA), 670, 696–698, 700, 706, 670, 696– OAM multiplexing, 282–287
698, 700, 706 Observation vector, 305, 307, 308, 310, 311, 313, 314, 317, 319, 320
Multicarrier modulation, 271–275, 295 OFDM bandwidth efficiency, 392, 396, 397
Index 779

OFDM-based super channel transmission, 436, 438, 436, 438 Parity-check matrix, 530, 531, 535, 537, 538, 544, 564, 565, 570–574,
OFDM building blocks, 397–404 576, 580, 601, 631–633
OFDM-CDMA, 696, 706 Partial-response signaling, 325–329
OFDM-CDMA FH systems, 696, 706 Passband digital wireless/optical transmission, 258–259
OFDM demodulation, 397 Path loss, 10, 13, 17
OFDM parallel channel decomposition, 399–404, 399–404 Pattern multiplication theorem, 198
OFDM parameters selection, 397–404 Peak-to-average power ratio (PAPR), 390, 425, 427, 440, 441, 444,
OFDM signal generation, 391, 392 390, 425, 427, 440, 441, 444
OFDM system types, 430–433 Percentage outage capacity, 464
OFDM waveform generation, 391 Perfect codes, 532, 533, 632
One-time pad encryption, 715 Perfect secrecy condition, 715
One-way communication, 724, 729 Phase diversity receivers, 367, 381
On-off keying (OOK), 700, 701, 707 Phase estimation in OFDM systems, 407, 407
Open eye condition, 681, 682, 709 Phase matching condition, 161, 162, 177
Operating points of Mach-Zehnder modulator, 166 Phase modulator, 13
Optical add-drop multiplexer (OADM), 158 Physical-layer security (PLS), 713–756, 713–756
Optical CDMA (OCDMA), 693–696, 706, 709 Piece-wise error probability (PEP), 473–475, 487–489, 523, 473–475,
Optical channels PLS, 754 487–489, 523
Optical communication systems with coherent optical detection, π hybrid, 180
359–372 π/2 hybrid, 180
Optical couplers, 172, 175–177, 202 π/4-shift-DQPSK OFDM in digital audio broadcasting (DAB), 414, 415
Optical cross-connect (OXS), 158 Pilot-aided channel estimation
Optical fiber core and cladding, 159, 159, 170 block-type channel estimation, 408, 410, 408, 410
Optical fibers, 36–37, 118–138, 140, 147, 149 comb-type channel estimation, 408, 408
MMF, 159, 170 least square (LS) channel estimation, 409, 409
SMF, 159, 170 MMSE channel estimation, 408–410
Optical filters, 158, 160, 175–177, 202–204 time-domain interpolation method, 410, 410
Optical hybrids, 160, 179–180, 202, 205 P-i-n photodiode, 8, 168, 169, 202
Optical orthogonal codes (OOC), 693–695, 706, 693–695, 706 Plane linearly polarized waves, 28, 28, 52
Optical receivers, 158, 160, 167–170, 179, 180, 202 Plane waves, 26–28, 45, 58, 59, 61, 62, 75, 78, 79, 82, 83, 131, 149
Optical SNR (OSNR), 318 Plotkin bound, 533
Optical transmitters, 157, 160–167, 172, 202 PMD compensator, 330, 341
Optimized bit-to-symbol mappings-based coded modulation, 616, 616 Poincaré sphere, 31, 32
Optimized signal constellation design (OSCD), 287–292, 295, Polar modulators, 160, 165–167, 202
616–624, 628, 631, 616–624, 628, 631 Polarization diversity and polarization-division demultiplexing
Optimum equiripple method, 332 363–364, 363–365
Optimum linear equalizer in MMSE sense, 333–334, 333–334 Polarization-division multiplexed OFDM, 434–436, 442, 434–436, 442
Optimum multiuser detection, 682–686, 682–686 Polarization-division multiplexing (PDM), 277–281, 295, 299,
Optimum receivers, 308, 310, 311, 380, 381, 384, 492, 603 277–281, 295, 298
Optimum signal constellation design (OSCD), 286–292, 295 Polarization ellipse, 32
Orbital angular momentum (OAM), 748, 750–752, 748, 750–752 Polarization-mode dispersion (PMD)
Orbital annular momentum (OAM), 282–288, 282–287 anisotropic stress, 125
Orthogonal designs, 476–478, 480, 521 differential group delay (DGD), 125, 125, 136
Orthogonal frequency division multiplexing (OFDM), 389–445, geometric birefringence, 125
389–445 Maxwellian distribution, 126
Orthogonality principle, 649, 671, 689, 701 polarization-maintaining fiber (PMF), 125
Orthogonal prolate spheroidal wave functions (OPSWF), 701, 701, 707 second-order PMD, 126
Orthogonal pulse modulation (OPM), 700, 701, 707, 700, 701, 707 Polarization noise, 368–370, 381
Orthogonal signaling, 315 Polarization-time (PT) coding, 434, 444, 522, 434, 444, 522, 523
Orthonormal basis functions, 250, 250, 251 Polyphase decomposition, 193, 202, 207
Outage probability, 100–101, 151, 215, 220, 245, 321, 380, 384, Polyphase filters-based resampling, 193, 195, 193, 195
452–456, 459–461, 507, 509 Population inversion, 160, 161, 172, 174
Overlapping subcarriers, 271–274 Power adaptation, 216, 217, 219, 220, 237, 238, 242, 245
Power attenuation coefficient, 118
Power delay profile (PDP), 115–116
P Poynting vector, 27, 28, 44, 48, 49, 95
Parallel decomposition of optical and wireless MIMO channels PPM-based UWB communication, 702
468–471, 468–471 Pre-amplifier, 167, 173, 204
Parallel decomposition of the MIMO wireless channel, 277 Precoding, 274, 277
Parallel independent decoding (PID), 589, 590, 595, 599, 589, 590, Preferred pairs of m-sequences, 671
595, 599 Primitive polynomial, 661, 667
Parallel I/Q modulators’ based square/star QAM transmitter, 270, 270 Principle of orthogonality, 250, 272, 273, 283, 390
Parallel realization, 187, 206 Privacy amplification, 714, 726, 728, 731, 739–741, 746, 749, 750, 754
Parallel turbo decoder, 555 Privacy amplification theorem, 740, 740
Parametric amplifiers, 158, 172 Probability of strictly positive secrecy capacity, 748, 748, 751
Processing gain, 648, 650, 654, 707
780 Index

Product codes, 549–550, 561, 630 Roll-off factor, 327

Propagation constant, 162, 171, 176, 177 Rotated Q-coordinate delayed signal constellation principle, 424
Propagation effects in wireless communication link, 144 Rotationally symmetric optical channels, 289–290, 290
Pseudo-noise (PN) sequences, 651 Rotation operation, 313, 358
Pulse-amplitude modulation (PAM), 252, 256–259, 295, 298, 700, 707, Rotator, 37, 38, 40, 149
252, 256–259, 294, 297, 700, 707 Rytov method of small perturbations, 77–79, 77–79
Pulse-duration modulation (PDM), 700, 700, 707 Rytov variance, 79, 84, 85, 150
Pulse laser based OCDMA, 695, 697, 706
Pulse-position nodulation (PPM), 700, 702, 707, 700, 702, 707
Punctured codes, 553, 594, 595 S
Saleh-Valenzuela indoor UWB channel model, 707
Scattering of electromagnetic waves, 93
Q Schmidl-Cox preamble for DFT windowing, 405
Q-factor, 318 Scintillation index, 79, 80, 84, 85, 150
QR-factorization, 478, 488, 490, 497, 520, 523 SDM-fibers-based PLS, 748–749
Quadrature amplitude modulation (QAM), 251, 258–260, 262–264, SDM in free-space optical (FSO) communications, 286–288,
267, 268, 270, 277, 278, 298, 251, 258–260, 262–264, 267, 268, 285–287, 295
270, 277, 278, 298 SDM in optical communications, 281
Quantum key distribution (QKD), 713, 714, 724, 726, 730, 737, SDM spectral efficiency, 481
713, 714, 724, 726, 730, 737, 748, 749 Secrecy capacity, 713, 714, 718–726, 729, 732, 733, 737, 741–743,
Quarter-wave plate (QWP), 38 748, 750–756
Quasi-single-mode transmission, 612–614, 631, 612–614, 631 Secret-key capacity, 725, 728, 729, 731, 756
Secret-key generation (agreement), 713, 714, 724–731, 737, 745–747,
754–756, 713, 714, 724–731, 737, 745–747, 754–756
R channel-type model (scenario), 729–731, 729–731
Radiation patterns, 46, 48, 87, 196–198 source-type model (scenario), 726–729, 726–729
Radix-2 and radix-4 algorithms, 391 Secret-key rate (SKR), 724, 725, 727–730, 746, 747
Raised cosine (RC) shape, 395 Secret multiplexing gain, 743, 756
Raised cosine (RC) spectrum, 327 Selection combining (SC), 452–455, 521
RAKE receiver, 658, 659, 703, 705–707, 709 Selection of pilot signals, 514
Raman amplifier gain, 174 Sellmeier equation, 119, 121
Raman amplifiers, 158, 160, 173–174, 202, 203 Semiconductor lasers, 6, 8
Random MIMO channels capacity Semiconductor optical amplifiers (SOAs), 160, 172–174, 160, 172–174
correlated fading channel capacity, 510, 510 Separately correlated model, 472, 472
ergodic capacity, 507–509 Set partitioning, 583, 586, 587, 589
non-ergodic capacity, 509, 510 Shadowing, 10, 13, 17
Randomness criterions, 659, 706 Shannon capacity, 215, 216, 220, 223, 224, 242, 244, 245
Rank determinant criterion, 475 See also Ergodic capacity
Rapid acquisition by sequential estimation (RASE), 657, 657 Shape-based UWB communications, 707
Rate-adaptive LDPC coding, 574–580, 626, 630 Shaping gain, 584
Rate-adaptive LDPC coding with re-encoding, 576 Signal constellations, 249, 251–253, 256, 257, 259–261, 264–267, 270,
Rayleigh, 51, 84, 108–112, 150–152 277, 278, 280–283, 288–293, 295–298
Rayleigh fading, 453–456, 459–461, 474, 521 Signal hierarchy of super channel OBM-OFDM enabling up to 100 Tb/s
Rayleigh scattering, 118 serial optical transport, 438
Ray tracing approximation, 89–91 Signal space theory, 249–259, 294
Receiver bandwidth, 204 Signature waveforms, 668–673, 676, 677, 679, 683, 706, 708, 709
Receiver diversity, 450–461, 521, 522 Singleton bound, 533, 549
Receiver shaping, 277 Singular value decomposition (SVD), 276, 276
Receivers maximizing signal-to-noise ratio, 308 SISO decoding, 562, 563
Reconciliation efficiency, 739, 756 Slepian-Wolf theorem, 738
Recursive systematic convolutional (RSC) code, 554 Sliding-window BCJR equalizer, 606, 606, 631
Reed-Solomon (RS) codes, 549 Sliding-window multidimensional turbo equalization, 604–606, 631
Reflection-induced noise, 137, 138 Slow-frequency hopping (SFH) systems, 651, 664, 665, 651, 664, 665
Regular LDPC codes, 564 Snell’s law of refraction, 28–40, 149, 28–40, 149
Regular Markov chain, 225, 226 Source coding with the side information, 737
Relative intensity noise (RIN), 137–140, 149, 180, 137–140, 149, 180 Space-frequency block coding (SFBC)-based MIMO, 504
Relay channel with confidential messages, 744 Space-time coding, 471–476, 479, 480, 482, 487, 488, 503, 521,
Reliability condition, 718, 720, 725, 732, 733 522, 524
Rényi entropy, 740, 741 Space-time decoder extrinsic massage, 501
Repetition MIMO, 468, 524 Space-time trellis codes (STTCs), 479–480, 521, 479–480, 521
Repetition node, 493, 495 SPA decoding over an arbitrary channel, 501
Resampling, 189–195, 202 Spatial coherence, 54, 78, 80
Retarded potentials, 42 Spatial diversity scheme, 275
Rich scattering model, 472
Index 781

Spatial division multiplexing (SDM), 5, 7, 9, 13–15, 17, 451, Tanner-Wilberg-Loeliger-Koetter (TWLK) graph, 500
5, 7, 9, 13–15, 17, 451, 479–486, 521 Tatarskii’s approach, 72
Spatial-division multiplexing (SDM) in wireless communications, Tau-dither loop (TDL), 657, 658
281–282, 281–282, 295 The fs/4 method based on I/Q demodulation, 422
Spatial interference cancellation, 474, 491–492, 498, 522, 474, Theorem of irrelevance, 308
491–492, 498, 522 Thermal noise, 137, 138, 140, 145, 149
Spatial light modulator (SLM), 286, 285, 286 Threaded space-time architecture, 483
Spatially-coupled LDPC codes, 575, 575, 576 Threshold combining (TC), 452, 455–456, 521
Spectral amplitude coding (SAC), 693, 695, 706, 693, 695, 706 Time-division multiple access (TDMA), 668, 673, 668, 673
Spectral efficiency of MCM, 592, 611, 614, 629 Time-domain 4D-NB-LDPC-CM, 608–611, 631
Spherical coordinates, 25–26, 43, 56, 72, 73, 148 Time-domain UWB modulations, 700–706
Spontaneous emission noise and noise beating components Time-frequency code (TFC), 704
interchannel crosstalk, 143 Time-invariant channel capacity, 221
intrachannel, 143, 144 Time-multiplexed coded modulation, 592
optical amplifier noise figure, 141 Time-selective channel, 103
optical demultiplexer, 143 Time-spreading encoding, 693, 694
optical filter, 141, 143 Time-varying impulse response, 11
optical pre-amplifiers, 141 Total reflection, 36
quantum shot noise, 142 Tracking, 655–658, 664, 665, 704, 706
Spread spectrum (SS), 648–707 Transform-domain representation, 542, 542, 543
Spread spectrum multiple access (SSMA), 668 Trans-impedance amplifier noise, 369–370, 379, 370–371, 381
Square- and cross-QAM transmitters for optical communications, 268 Trans-impedance receiver front end, 202, 204
Square-QAM, 260, 261, 268–270, 292, 294, 298 Transition probabilities, 556, 559, 631
Star-QAM, 260, 266, 267, 270, 294, 295 Translation operation, 314
Star-QAM transmitters for optical communications, 266–267 Transmit antenna, 41, 43–46, 49, 87–89
Stationary property, 228 Transmit diversity, 461, 521
Steepest descent algorithm, 336, 337 Transmit precoding, 277
Stimulated emission, 160, 161, 163, 171, 173 Transmitter diversity when CSI is available on transmitter side,
Stimulated Raman scattering (SRS), 173, 173 461–462, 461–462
Stokes vector representation, 30 Transverse electric (TE) polarization, 34, 35, 39, 90, 151, 34, 35,
Stream rotator, 483 39, 90, 151
Strong secrecy condition, 717, 719, 731 Transverse magnetic (TM) polarization, 34–36, 39, 90, 34–36, 39, 90
Subcarrier-multiplexing (SCM), 7, 9, 13, 7, 9, 13, 17 Trellis coded modulation (TCM), 582, 583, 585–588, 594,
Subharmonics’ method, 76, 76 582, 583, 585–588, 594, 595, 631
Suboptimum MIMO receivers for coded signals, 488–492, 521, Trellis description of linear block codes, 537–541, 630, 537–541, 630
488–492, 521 Trellis space-time codes, 476, 479, 480
Successive cancelation, 691, 692, 706 Truncated channel inversion policy, 220
Sum-product algorithm (SPA), 495–496, 500, 521, 733, 735, 739, Truncated channel matrix, 485
495–496, 500, 521, 733, 735, 739 Tunable lasers, 8
Sum-product algorithm for channels with memory, 496, 496 Turbo codes, 541, 553–561, 594, 630, 639, 641
Sum-product rule, 495 Turbo encoding, 554, 556, 594, 640, 641
Superchannel OBM-OFDM enabling ultra-high-speed Turbo equalization, 556, 602, 603, 605, 607–608, 637, 638
optical transport, 437 Turbo TCM encoder with bit interleaving, 596
Survivor path, 539, 587 Turbo trellis coded modulation, 594–595, 594–595
Symbol error probability, 310, 311, 313–319, 321–323, 327, 380–384 Turbo-product code (TPC), 561–563
Symbol field, 534, 535, 549 Two-dimensional signal constellations, 259–265
Symbol interleaved turbo TCM, 595 Two-sate/four-state trellis section, 585
Symmetric split-step method (SSSM), 64, 65, 68, 69, 76, 64, 65, 68, Two-user synchronous successive cancellation, 691, 691
69, 76
Synchronization, 655–657, 664, 696, 698, 699, 706
carrier synchronization, 343, 346 U
Costas loop, 344, 346, 365, 366 Ultra-wideband (UWB), 648–707
symbol synchronization, 343 Ultra-wideband communication (UWC), 413, 429–430, 442, 413,
Synchronizer, 664 429–430, 442
Synchronous CDMA, 670–679, 682, 683, 690, 691, 708 Undetected error probability, 549
Synchronous coherent optical detection, 179 Unequal error protection, 582–595, 626, 630
Syndrome, 531, 535–537, 540, 563, 632, 633 Uniformity condition, 725, 728
System (transfer) function, 183–184, 202 Uniformity of the key, 725, 728, 731
Systematic code, 530, 545, 632 Uniform power allocation strategy, 507, 507, 511
Systematic form of convolutional code, 543 Union bound approximation, 317, 383
Unipolar signaling, 315
Universal hash functions, 714, 739, 740, 749
T Uplink channel, 10, 668
Tandem-QAM transmitter, 267, 270 Upsampling, 181, 189–194, 202
Tanner graph cycle, 499, 500 UWB receiver configuration, 703
782 Index

V time division multiple access (TDMA)-based standards, 2,

Variable-error probability technique, 348 2, 10, 12
Variable nodes, 499, 553, 564–568, 577, 578, 580, 581, 633 3G (third generation)
Variable-power variable-rate Modulation technique, 348 CDMA2000, 2
Vector derivatives dispersion compensating fibers (DCFs), 5
curls, 25 dispersion shifted fiber (DSF), 5, 5
del/nabla operator, 23 Erbium doped fiber amplifiers (EDFAs), 5
divergence, 23 high speed packet access (HSPA), 2
gradient, 23 IMT-2000, 2
Laplacian, 25 1.55 μm telecommunication window, 4
Vernam cipher, 715 optical add-drop multiplexers (OADMs), 5
Vertical cavity surface emitting laser (VCSEL), 8, 8, 163 optical solitons, 5
Visible light communications (VLCs), 144, 147–149, 144, 147–149 3GPP long term evolution (LTE), 2
Viterbi decoding algorithm, 537–541, 630 UMTS, 2
Voltage control oscillator (VCO), 187, 187 wavelength division multiplexing (WDM), 5
Volterra series based equalization, 354, 355, 379, 355, 356, 381 4G (fourth generation)
dense wavelength division multiplexing (DWDM), 5
DWDM multiplexers/demultiplexers, 5
W IMT-advanced, 3
Walsh sequences, 659, 672, 699, 706 layered control plane, 5
Water-filling method, 217, 218, 242, 506, 508, 510 LTE advanced, 3
Wave equation, 21–26, 41, 42, 45, 49, 51, 55, 56, 63, 68, 123, 148 optical cross-connects (OXCs), 5
Wavelength bands, 7, 8 Raman amplifiers, 5
Wavelength division multiplexing (WDM), 5, 8, 158, 172, 174, reconfigurable OADMs (ROADMs), 5, 5, 6
177–178, 204, 205, 5, 8, 158, 172, 174, 177–178, 204, 206 10 Gb/s and 40 Gb/s data rates, 5
Wavelength hopping (WH) schemes, 706 WirelessMAN-advanced (IEEE 802.16m), 3
Wave structure function, 79 5G (fifth generation)
WDM multiplexers/demultiplexers, 160, 160, 202 advanced FEC, 7
WDM optical network, 6 coherent optical detection, 5
Weak secrecy condition, 717, 718, 731, 754 digital signal processing (DSP), 5
Weight distribution of RS codes, 549, 549, 633 400 Gb/s Ethernet, 5
Wideband channel, 104, 113–114 100 GbE, 5
Wiener filtering, 334–335, 380 100 Gb/s to 1 Tb/s data rates, 5
Wi-Fi, 426–428 massive MIMO, 3, 16
WiGig, 413 multilevel modulation, 5
WiMAX, 413, 426–428, 442 Next Generation Mobile Networks Alliance, 3
Wimax (IEEE 802.16), 2, 4, 12, 15 1 Tb/s Ethernet, 5
Windowing, 392–396, 398, 399, 424, 429, 442 optical transport network (OTN), 5
Wireless channels PLS, 713–756 orthogonal frequency division multiplexing (OFDM), 5
Wireless communication systems generations polarization-multiplexing, 5
1st generation (1G) wireless software-defined network (WSDN), 3
advanced mobile phone service (AMPS), 2, 2 Wireless LAN (WLAN), 413, 426
0.8 μm–0.9 μm telecommunication window, 4 WiFi (IEEE 802.11), 3
multimode fiber (MMF), 4, 13 Wireless system performance, 358
multimode lasers, 4 Wiretap channel, 717, 719–721, 724, 726, 729, 732, 734, 737, 748,
total access communication system (TACS), 2 754, 755
2G (second generation) Wiretap codes, 718, 720–722, 732, 755
code division multiple access (CDMA)-based standards, 2 Wyner’s wiretap channel, 717–721, 723, 754, 717–721, 723, 754
EDGE, 2
GPRS, 2
GSM, 2 Z
IS-95, 2 Zero-forcing equalizers, 329–333, 380
IS-136, 2 Zero-forcing receivers (interfaces), 488–489, 488–489
1.3 μm telecommunication window, 4 Zero-outage capacity, 220, 242, 245
single-mode fiber (SMF), 4
single-mode lasers, 5