Doppler Compensation

Algorithms for DSP-based

Implementation of OFDM
Underwater Acoustic
Communication Systems

Ammar Abdelkareem
Newcastle University
School of Electrical, Electronic, and Computer Engineering,
Newcastle upon Tyne, UK.

A thesis submitted for the degree of

Doctor of Philosophy
2012

To my loving parents and my wife

Science is organised knowledge. Wisdom is organised life.

-Immanuel Kant

Acknowledgements

Throughout my PhD research in this University, I was fortunate to have

the opportunity to work with such a high quality supervision team. This
team included Prof. Bayan Sharif who introduced me to working on an
embedded system implementation for Doppler compensation algorithms.
I would like to express my sincere gratitude and appreciation for his
valuable comments, guidance and suggestions.
I wish to direct my special gratitude to Dr. Charalampos Tsimenidis for
suggestions that contributed to the development of a research project of
high quality and for his help during the PhD research journey. Thanks
are also due to my third supervisor, Mr. Jerey Neasham.
I am grateful to my friends for their support and companionship. Additionally, I wish to express my thanks to Iterative Solutions Inc. for their
CML toolbox.
I greatly appreciate the assistance provided in the form of a scholarship
from the Iraqi Governments Ministry of Higher Education and Scientic Research. Without their funding and support, it would have been
impossible to commence and successfully complete this research work.
Special thanks to my wife and my children, Yasir, Ali and Rahaf, for
their support and patience during my study.
Finally, I would like to express my deep gratitude to my parents who
have supported me throughout my education. Praise God, for all things
that have happened to me in my life.

Declaration

I declare that this thesis is my own work and it has not been previously
submitted, either by me or by anyone else, for a degree or diploma at any
educational institute, school or university. To the best of my knowledge,
this thesis does not contain any previously published work, except where
another persons work used has been cited and included in the list of
references.

Abstract

In recent years, orthogonal frequency division multiplexing (OFDM) has

gained considerable attention in the development of underwater communication (UWC) systems for civilian and military applications. However,
the wideband nature of the communication links necessitate robust algorithms to combat the consequences of severe channel conditions such
as frequency selectivity, ambient noise, severe multipath and Doppler
Eect due to velocity change between the transmitter and receiver. This
velocity perturbation comprises two scenarios; the rst induces constant
time scale expansion/compression or zero acceleration during the transmitted packet time, and the second is time varying Doppler-shift. The
latter is an increasingly important area in autonomous underwater vehicle (AUV) applications. The aim of this thesis is to design a low complexity OFDM-based receiver structure for underwater communication
that tackles the inherent Doppler eect and is applicable for developing real-time systems on a digital signal processor (DSP). The proposed
structure presents a paradigm in modem design from previous generations of single carrier receivers employing computationally expensive
equalizers. The thesis demonstrates the issues related to designing a
practical OFDM system, such as channel coding and peak-to-average
power ratio (PAPR). In channel coding, the proposed algorithms employ
convolutional bit-interleaved coded modulation with iterative decoding
(BICM-ID) to obtain a higher degree of protection against power fading caused by the channel. A novel receiver structure that combines
an adaptive Doppler-shift correction and BICM-ID for multi-carrier systems is presented. In addition, the selective mapping (SLM) technique
has been utilized for PAPR. Due to their time varying and frequency
selective channel nature, the proposed systems are investigated via both
laboratory simulations and experiments conducted in the North Sea o

the UKs North East coast. The results of the study show that the proposed systems outperform block-based Doppler-shift compensation and
are capable of tracking the Doppler-shift at acceleration up to 1m /s2 .

Contents
Nomenclature

xv

List of Symbols

xix

1 Introduction

1.1

Underwater Channel Characteristics . . . . . . . . . . . . . . . . . . .

1.2

Advances in Doppler shift compensation for UWC systems . . . . . .

1.3

Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.4

Publications Arising From This Research . . . . . . . . . . . . . . . .

1.5

Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2 Background

11

2.1

Introduction to Digital Modulation . . . . . . . . . . . . . . . . . . . 11

2.2

Literature survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.3

2.2.1

Advances in Underwater Acoustic communications

2.2.2

Tools and Algorithms for real-time implementation . . . . . . 13

2.2.3

Channel coding . . . . . . . . . . . . . . . . . . . . . . . . . . 15

OFDM system transmitter . . . . . . . . . . . . . . . . . . . . . . . . 15

2.3.1

Peak-to-average power ratio (PAPR) . . . . . . . . . . . . . . 15

2.3.2

Pulse shaping . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.3.3

Guard interval in the OFDM systems . . . . . . . . . . . . . . 17

2.3.3.1

2.4

. . . . . . 13

Cyclic prex OFDM . . . . . . . . . . . . . . . . . . 17

The Underwater Acoustic Channel . . . . . . . . . . . . . . . . . . . 20

2.4.1

Attenuation and ambient noise . . . . . . . . . . . . . . . . . 20

2.4.2

Doppler eect . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.4.3

Time varying multipath channel . . . . . . . . . . . . . . . . . 22

2.4.3.1

Doppler spread and coherence time . . . . . . . . . . 23

vii

CONTENTS
2.4.3.2

Delay spread and coherence bandwidth . . . . . . . . 24

2.5

BICM-ID OFDM system . . . . . . . . . . . . . . . . . . . . . . . . . 26

2.6

The Eects of the Interleaver . . . . . . . . . . . . . . . . . . . . . . 31

2.7

DSP platform selection issues . . . . . . . . . . . . . . . . . . . . . . 32

2.8

Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3 Low-complexity symbol-by-symbol Doppler shift compensation

3.1

3.2

36

OFDM system description . . . . . . . . . . . . . . . . . . . . . . . . 37

3.1.1

System and channel models . . . . . . . . . . . . . . . . . . . 37

3.1.2

Doppler shift in wideband communication . . . . . . . . . . . 39

Doppler compensation techniques . . . . . . . . . . . . . . . . . . . . 41

3.2.1

Block length-based Doppler compensation . . . . . . . . . . . 41

3.2.2

One-shot Doppler shift compensation . . . . . . . . . . . . . . 43

3.2.3

3.2.4

3.2.2.1

Coarse Doppler estimation . . . . . . . . . . . . . . . 44

3.2.2.2

Peak localization . . . . . . . . . . . . . . . . . . . . 45

Adaptive Doppler compensation . . . . . . . . . . . . . . . . . 47

3.2.3.1

Weighting coecients . . . . . . . . . . . . . . . . . 48

3.2.3.2

Fractional CFO Estimation . . . . . . . . . . . . . . 49

Proposed Doppler shift compensation . . . . . . . . . . . . . . 50

3.2.4.1

Doppler shift variation adjustment . . . . . . . . . . 51

3.2.4.2

Fine timing estimation . . . . . . . . . . . . . . . . . 52

3.2.4.3

Channel estimation and decoding . . . . . . . . . . . 52

3.2.4.4

Complexity analysis . . . . . . . . . . . . . . . . . . 53

3.3

Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.4

Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.4.1

Experiment setup . . . . . . . . . . . . . . . . . . . . . . . . . 56

3.4.2

Performance evaluation . . . . . . . . . . . . . . . . . . . . . . 57

3.5

Real-time implementation of BICM-ID . . . . . . . . . . . . . . . . . 60

3.6

Hardware platform description . . . . . . . . . . . . . . . . . . . . . . 63

3.7

3.6.1

Processing time optimization . . . . . . . . . . . . . . . . . . 64

3.6.2

Memory allocation . . . . . . . . . . . . . . . . . . . . . . . . 67

3.6.3

Real-time experimental results . . . . . . . . . . . . . . . . . . 67

Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

viii

CONTENTS
4 Time varying Doppler shift compensation
4.1

4.2

4.3

69

Time varying Doppler shift model . . . . . . . . . . . . . . . . . . . . 70

4.1.1

Sampling frequency errors . . . . . . . . . . . . . . . . . . . . 71

4.1.2

Carrier frequency oset errors . . . . . . . . . . . . . . . . . . 72

Signal processing in the proposed receiver

. . . . . . . . . . . . . . . 72

4.2.1

Coarse timing metric estimation . . . . . . . . . . . . . . . . . 73

4.2.2

Time varying Doppler shift estimation . . . . . . . . . . . . . 73

Doppler extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.3.1

Linear prediction of the symbol timing oset . . . . . . . . . . 75

4.3.2

Fine symbol timing oset and synchronization . . . . . . . . . 76

4.3.3

Tracking the Doppler shift . . . . . . . . . . . . . . . . . . . . 77

4.3.4

Residual Doppler shift estimation . . . . . . . . . . . . . . . . 78

4.4

Pilot-based channel estimation . . . . . . . . . . . . . . . . . . . . . . 78

4.5

Experimental results . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.5.1

Proposed receiver performance

. . . . . . . . . . . . . . . . . 81

4.5.2

Eect of weighting coecients

. . . . . . . . . . . . . . . . . 84

4.5.3

Performance evaluation with improved coarse timing estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

4.5.4

Performance evaluation based on two point correlation . . . . 88

4.6

Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

4.7

Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

5 Adaptive time varying Doppler shift compensation

96

5.1

Acceleration eects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

5.2

Adaptive OFDM receiver structure . . . . . . . . . . . . . . . . . . . 98

5.2.1

Estimation of Symbol timing expansion/compression . . . . . 99

5.2.1.1

5.2.2

5.3

Control range and PM algorithms . . . . . . . . . . . 100

Early termination search algorithm . . . . . . . . . . . . . . . 101

5.2.2.1

Selection of step-size () and correction factor (Ki ) . 102

5.2.2.2

Time-varying Doppler shift estimation and tracking . 104

5.2.2.3

Residual Doppler shift estimation

. . . . . . . . . . 105

System design, simulation, and experimental results . . . . . . . . . . 107

5.3.1

System design parameters . . . . . . . . . . . . . . . . . . . . 110

5.3.1.1

Transmitted packet structure . . . . . . . . . . . . . 110

ix

CONTENTS
5.3.1.2

5.4

Parameters of cyclic prex . . . . . . . . . . . . . . . 110

5.3.2

Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . 112

5.3.3

Experimental results . . . . . . . . . . . . . . . . . . . . . . . 114

5.3.3.1

Experimental setup and channel characteristics . . . 114

5.3.3.2

Performance evaluation of the proposed receiver . . . 115

5.3.3.3

Search points, exponents and CFO range selection

. 119

Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

6 Conclusion and Future Work

124

References

127

List of Figures
2.1

Block diagram of SLM method for PAPR reduction. . . . . . . . . . . 16

2.2

Time domain characteristics of raised-cosine function for dierent values of roll-o factor . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.3

Normalized CIR of a 500 m range channel. . . . . . . . . . . . . . . . 25

2.4

Normalized CIR of a 1000 m range channel. . . . . . . . . . . . . . . 25

2.5

System diagram of BICM-ID. . . . . . . . . . . . . . . . . . . . . . . 28

2.6

Convolutional encoder of rate Rc =1/2. . . . . . . . . . . . . . . . . . 29

2.7

Performance comparison of BICM-ID OFDM and uncoded system.

2.8

SHARC ADSP-21364. . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.1

Proposed transmitter structure, where the operator represents the

. 30

real part of the signal . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.2

Open loop Doppler correction. . . . . . . . . . . . . . . . . . . . . . . 42

3.3

Packet length measurement using chirp correlation. . . . . . . . . . . 42

3.4

Receiver structure of the proposed system.

3.5

Correlation operation in (3.21), and i = ( 2 ) represent the leading

. . . . . . . . . . . . . . 43

and trailing edge of the OFDM frame, respectively. . . . . . . . . . . 44

3.6

A set of OFDM symbols showing the closed and far neighbour to

symbol i. At time n, symbol i estimates i (n) . . . . . . . . . . . . . 47

3.7

Packet structure for Nc = 1024. . . . . . . . . . . . . . . . . . . . . . 54

3.8

Performance comparison between one-shot algorithm and the algorithm proposed by Kim in [1]. . . . . . . . . . . . . . . . . . . . . . . 55

3.9

Anticipated correlation window before and after smoothing for packet

1, symbol 3 at speed -0.25 m/s from the experiment. . . . . . . . . . 55

3.10 Performance comparison between the adaptive scheme and block Doppler
compensation for xed and variable speeds. . . . . . . . . . . . . . . . 56

xi

LIST OF FIGURES
3.11 Conguration of the experiment in the North Sea. . . . . . . . . . . . 57
3.12 Bit error rate over each packet of 8920 bits. . . . . . . . . . . . . . . 57
3.13 Estimation of the Doppler scaling factor over each block for packets
( 2, 12) of the adaptive algorithm. . . . . . . . . . . . . . . . . . . . . 59
3.14 Estimation of the Doppler scaling factor over each block for packets
( 5, 6, 10, 16) of one-shot algorithm.

. . . . . . . . . . . . . . . . . . 59

3.15 Changing of speed through the packets time. . . . . . . . . . . . . . 60

3.16 Performance of the proposed system from the experiment. . . . . . . 62
3.17 General System Specication. . . . . . . . . . . . . . . . . . . . . . . 63
3.18 SHARC ADSP-21364 system architecture block diagram. . . . . . . . 63
3.19 Receiver tasks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.20 Output of real-time BICM-ID 3 Iterations. . . . . . . . . . . . . . . . 68
4.1

Proposed receiver structure . . . . . . . . . . . . . . . . . . . . . . . 72

4.2

Acceleration eect over Doppler frequency change during each symbol

time at fc =12 kHz. (a) a=0.5 m/s2 , (b) a=1 m/s2 . . . . . . . . . . 74

4.3

Estimation of timing oset during the packet time. . . . . . . . . . . 75

4.4

Tracking the Doppler within the OFDM symbol. . . . . . . . . . . . . 77

4.5

Received signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

4.6

Sample of normalized channel impulse response for 1000 m channel

range. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

4.7

Performance of the proposed system at 1024 sub-carriers. . . . . . . . 83

4.8

Estimated speed variation during OFDM symbol. . . . . . . . . . . . 84

4.9

Constellation output from equalizer and iterative receiver. . . . . . . 85

4.10 Eect of weighting coecients on estimation. . . . . . . . . . . . . . . 86

4.11 Performance of the proposed system with improved coarse timing
estimation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.12 Improved time-varying speed estimation during OFDM symbol 7 of
packet 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.13 Performance of the improved proposed system. . . . . . . . . . . . . . 93
4.14 Performance comparison of block based and proposed techniques. . . 94
5.1

OFDM symbol structure due to Doppler eects. . . . . . . . . . . . . 97

5.2

Receiver structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

5.3

Eect of exponents on step-size and correction factor convergence. . . 104

xii

LIST OF FIGURES
5.4

Structure of the transmitter used in simulation. . . . . . . . . . . . . 107

5.5

Simulation model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

5.6

Scenarios of time-varying Doppler shift. . . . . . . . . . . . . . . . . . 110

5.7

Correlation lag variation due to time scale expansion/compression at

an acceleration of 1 m/s2 . . . . . . . . . . . . . . . . . . . . . . . . . 111

5.8

BERs performance for dierent acceleration and cyclic prex lengths

at SNR=15 dB, max = 10 ms, and Nc = 1024. . . . . . . . . . . . . . 113

5.9

Eect of the maximum delay spread max on the BER performance

at a = 0.5 m/s2 and Tg = 16 ms. . . . . . . . . . . . . . . . . . . . . 113

5.10 BER performance with the system parameters Tg = 16 ms, Nc = 1024

at max =10 ms and a = 1 m/s2 , where Da denotes deceleration case. 114
5.11 Channel measurements for 1000 m range. . . . . . . . . . . . . . . . . 116
5.12 Channel measurements for 500 m range. . . . . . . . . . . . . . . . . 117
5.13 BER performance of adaptive time varying Doppler shift compensation receiver over 1000 m channel range for dierent sub-carriers
length. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.14 BER performance of adaptive time varying Doppler shift compensation receiver over 500 m channel range for dierent sub-carriers length.120
5.15 Eect of exponents on the estimation of i and Ki , search points and
smoothing the tracking step at Nc = 1024 over 1000 m channel range. 121

xiii

List of Tables
2.1

DMA operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.2

Receiver operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3.1

Correlation complexity estimates. . . . . . . . . . . . . . . . . . . . . 53

3.2

Performance results of the experiment . . . . . . . . . . . . . . . . . . 61

3.3

Average BER and error statistics comparison of the experimental

results for dierent Doppler shift compensation techniques . . . . . . 61

3.4

Total Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

4.1

Average BER comparison of the experimental results at dierent settings of weighting coecients between the proposed and block-based
Doppler shift techniques for Nc =1024

4.2

. . . . . . . . . . . . . . . . . 82

Performance of the experimental results between the improved and

block-based Doppler shift techniques for Nc =1024 . . . . . . . . . . . 92

4.3

Performance of the experimental results between the improved and

block-based Doppler shift techniques for Nc =512 . . . . . . . . . . . . 94

4.4

OFDM symbol structure and the corresponding data rates . . . . . . 94

5.1

Main system specications . . . . . . . . . . . . . . . . . . . . . . . . 114

xiv

Nomenclature
Acronyms
ACS Add Compare Select
ADC Analogue to Digital Converter
AET Automatic Early Termination
AP P A Posteriori Probabilities
AU V Autonomous Underwater Vehicle
AW GN Additive White Gaussian Noise
BBER Block Bit Error Rate
BCJR Bahl Cocke Jelinek Raviv
BER Bit Error Rate
BICM Bit Interleaved Coded Modulation
BICM ID BICM with Iterative Decoding
CC

Convolutional Code

CDM A Code Division Multiple Access

CE

Channel Estimation

CF O Carrier Frequency Oset

CIR Channel Impulse Response
COF DM Coded Orthogonal Frequency Division Multiplexing
CP

Cyclic Prex
xv

Nomenclature
CRC Cyclic Redundancy Check
CSI Channel State Information
DAC Digital to Analogue Converter
DDW S Direct Digital Wave Synthesis
DEU Doppler Extraction Unit
DF E Decision Feedback Equalizer
DM

Data Memory

DP SK Dierential Phase Shift Keying

DSP Digital Signal Processing
ECC Error Correction Coding
ER

Expectation Range

F EC Forward Error Correction

F F T Fast Fourier Transform
F IR Finite Impulse Response
F P GA Field Programmable Gate Array
F SK Frequency Shift Keying
HT LS Hankel Total Least Square
ICI

Intercarrier Interference

ISI

Intersymbol Interference

LF M Linear Frequency Modulated

LLR Log Likelihood Ratio
LM

Learning Mode

LM S Least Mean Square

LOS Line of Sight
xvi

Nomenclature
LP

Learning and Punishment

M AP Maximum a Posteriori
M IM O Multiple Input Multiple Output
M LE Maximum Likelihood Estimator
M LSE Maximum Likelihood Sequence Estimation
M M LE Marginal Maximum Likelihood Estimation
M M SE Minimum Mean Square Error
M SE Mean Square Error
N SC Non Systematic Code
OF DM Orthogonal Frequency Division Multiplexing
P AP R Peak to Average Power Ratio
P DP Power Delay Prole
P LL Phase Lock Loop
PM

Punishment Mode

P SF Pulse Shaping Filter

QAM Quadrature Amplitude Modulation
QP SK Quadrature Phase Shift Keying
RF

Radio Frequency

RLS Recursive Least Square

SICM Symbol Interleaved Coded Modulation
SISO Soft-in Soft-out
SN R Signal to Noise Ratio
SOV A Soft Output Viterbi Algorithm
SP I

Serial Port Interface

xvii

Nomenclature
SRAM Static Random Access Memory
T CB Transfer Control Block
T CM Trellis Coded Modulation
TI

Texas Instruments

TL

Transmission Loss

U AC Underwater Acoustic Channel

U W C Underwater Communication
V HDL Very high speed Hardware Description Language
W SS Wide Sense Stationary
ZF E Zero Forcing Equalizer
ZP

Zero Padding

xviii

List of Symbols
a(f )

attenuation loss

Acceleration in m/s2

a(t)

Time-varying acceleration in m/s2

The forward-state metric

Signal bandwidth in kHz

Bcoh

Coherence bandwidth in Hz

bi

Information bits

Roll-o factor

The reverse-state metric

Propagation speed of acoustic signals

ck

group of interleaved bits

Set of complex numbers

Doppler shift

Doppler shift estimation error

(t)

Time varying Doppler shift

Doppler shift estimate

The anticipated observation window

Residual Doppler shift estimate

Sub-carrier bandwidth

fc

Carrier frequency

fn

Sub-carrier frequency

fs

Sampling frequency

Fd

Doppler frequency shift

Shift in OFDM samples

Estimate shift in OFDM samples

Previous estimation of ne symbol timing oset

xix

Nomenclature
c

Current estimation of ne symbol timing oset

Sampling frequency oset estimate at the leading edge

Sampling frequency oset estimate at the trailing edge

step

The tracking step

The centroid-based 1st order moment estimate

Correlation-based 1st order moment estimate over

yy

Correlation-based 1st order moment estimate over full CP samples

Average shift in samples between c and p

CP

Expected control range (in samples) for positive a

N P

Expected control range (in samples) for negative a

Fine tuned 1st order moment of the correlation lag

th

Threshold for the correlation peaks

The branch metric

hl

Path amplitude

p
H

Estimated channel at pilot symbols

LS [n] Least square channel estimate at each sub-carrier

H
K

Constraint length

Kd

Length of information bits

Kc

Length of coded bits

Sampling frequency oset of one sample drift

Total number of paths

Lint

Interleaver length

La

A priori LLRs

Le

Extrinsic LLRs

Lf

The transmitted passband samples length

Lf

The Doppler shifted received passband samples length

(m, i) Timing function at window range m and i

M

Constellation size

Slop of a linear equation

Vector of maximum peak positions

Nc

Number of sub-carriers

Nd

Data bearing sub-carriers

Np

Pilot sub-carriers

xx

Nomenclature
Ng

Guard samples

NS

Number of samples per symbol

Group of OFDM symbols

prc

Pulse shaping lter

Position estimate of the maximum Doppler shift

Interleaver

De-Interleave

Relative sampling frequency oset estimate

(I)

Integer relative sampling frequency oset estimate

(F)

Fractional relative sampling frequency oset estimate

Rc

Code rate

r(t)

Received signal in passband

Correlation vector

w2

Noise variance

Ts

Sampling time

Tcoh

Coherence Time

Tg

Guard time

l (t)

Time varying path delay

max

Maximum delay spread

Sequence of phases for PAPR

Uopt

Optimal sequence of PAPR phase

Weighting coecients vector

X(n)

Transmitted frequency domain OFDM

Yi [n]

Received frequency domain OFDM at index i

Yp

The received pilot symbols after the FFT

Frame synchronization point

Fine tuned synchronization point

xxi

Chapter 1
Introduction
Since 1919, when the rst scientic paper on underwater sound was published by
German scientists who presented a theoretical description of the bending of sound
rays and their importance in determining sound ranges [2], remarkable interest has
been shown in the exploitation of acoustic waves for both military and commercial purposes. In military applications for instance, they are employed extensively
by submarines and for navigation; while as far as commercial applications are concerned, acoustic waves are utilised in oshore oil exploration, monitoring underwater
pipelines, sh nding and as an aid to divers. The emergence of digital signal processors has made possible the development of complex methods of signal processing
for underwater communication (UWC) systems applications. In recent years, researchers and engineers have devoted great eorts towards achieving high data rates
and reliable wireless communication systems for data, video and voice without incurring any loss of information. However, wireless UWC systems have to overcome
many physical obstacles such as temperature and pressure variations, salinity, ambient noise, multipath and the Doppler eect. The Doppler eect is caused by
current and wave motions and is due to the relative motion between the transmitter/receiver. In order to attain reliable communication in such severe circumstances,
as far as the underwater acoustic channel is concerned, it is necessary to take into
account these obstacles. The characteristics of the underwater acoustic channel and
the most notable accomplishments in underwater communication in recent years are
presented in the following sections. In this thesis, practical Doppler shift compensation techniques for OFDM-based receivers are adopted. In addition, BICM-ID is
employed to protect the data against the channel eects and it is considered as an

1.1 Underwater Channel Characteristics

application for ADSP-21364 SHARC DSP.

1.1

Underwater Channel Characteristics

Although the interference and crosstalk properties from other users are stationary
in a wired communications system, it fails to achieve mobility and exibility in
terms of maintenance, particularly in deep water; therefore, wireless communication
is a more practical alternative. Underwater acoustic communications are preferred
on radio frequency (RF) communications because electromagnetic waves do not
propagate over long distances underwater, except at high power. This is a direct
constraint which makes acoustic waveforms the best solution for transmitting data
undersea. However, these waveforms present challenges as far as achieving reliable
communication is concerned due to their special properties such as low propagation
speed.
Dealing with underwater acoustic channels is a daunting challenge, but one
which should be considered in order to achieve more reliable wireless communications. These channels are time-varying in nature and delay dispersive to the order
of 100 symbol time. Furthermore, due to the low propagation speed of acoustic
signals (c) of approximately 1500 m/s, the transmitted signal is more vulnerable
to the Doppler eect when compared to other communication systems. Therefore,
even a slow movement between the transmitter/receiver and/or the inherent current waves motion can stretch or press the transmitted signal, depending on the
direction of motion, and consequently destroy the synchronization. Many receiver
structures have been proposed to deal with the time varying multipath that causes
inter-symbol interference (ISI). All of them adopt channel estimation and equalization schemes, coding algorithms and spread spectrum systems. These receivers are
based on a time-domain view of the channel and they require a highly complex structure. Recently, an alternative multi-carrier communication system in the form of
OFDM has gained considerable interest for communication over frequency-selective
channels, where the symbol duration is made much larger than the delay spread.
Furthermore, the OFDM is also attractive due to its simplicity in terms of modulation/demodulation by means of fast Fourier transform (FFT). However, the relative
Doppler shift in the channel, due to the transmitter/receiver motion (v) with the
acoustic signal propagation and the sensitivity of the OFDM to the Doppler eect,
2

1.1 Underwater Channel Characteristics

means that delicate synchronization algorithms are required.
The available bandwidth in UAC is restricted by the transmission loss and signalto-noise ratio (SNR). The transmission loss is proportional to the range and SNR
is limited by the noise level caused by the ambient such as the sound of ships engines, bubbles and acceleration thrust. Thus, for the desired transmission range,
this frequency susceptibility to transmission loss ultimately adds a constraint on
the choice of the carrier frequency [3]. Consequently, underwater acoustic communication links can be divided into three types according to their range. Firstly, a
long-range system, operating over several tens of kilometres, is limited to a few kHz
of bandwidth. Secondly, a medium-range system, operating over several kilometres,
has a bandwidth of the order of ten kHz. Thirdly, a short- range system, operating
over several tens of metres, may have available bandwidths of more than a hundred
kHz [3]. Additionally, multipath propagation is prevalent in UAC channels. These
channels characteristics vary with time [4] and hence they are also called doubly
selective channels, as the transmitted signal undergoes dispersion in both the time
and frequency domains. Furthermore, the multipath arrivals are highly dependent
on the location of the transmitter and receiver in the given geometry and density
of the medium. While the vertical channels exhibit little multipath, the horizontal
channels may have extremely long multipath spreads. Establishing reliable communication and combating the underwater multipath is considered, without exception,
to be a daunting task for any underwater acoustic communication system.
Two main problems are encountered when considering the underwater channel.
The rst one is multipath propagation due to reection from the surface and the
seabed; the second one is the Doppler eect [5], arising due to relative motion
between the transmitter and receiver. In multipath propagation, each transmitted
signal is received via multiple paths and delays and is accompanied by random
uctuations in phase and amplitudes. The multipath structure depends on the
channel geometry. For instance, the multipath spread in a vertical channel is very
small compared with its counterpart in a horizontal channel which attains a hundred
times the multipath spreads experienced in radio channels. For example, a multipath
spread of 10 ms in a shallow water channel causes the ISI to extend over 100 symbols,
if the system is operating at a rate of 10 kilo-symbols per second (ksps) [6]. In single
carrier transmission, the multipath eects are dealt with by adopting equalizers
at the receiver, which has the eect of adding a burden to the complexity. An
3

1.1 Underwater Channel Characteristics

alternative, with a lower degree of complexity, is the multi-carrier transmission in
the form of OFDM. However, it is very sensitive to the Doppler shift [7] caused
either by the sampling rate mismatch between the transmitter/receiver pair or their
motion.
The time-varying multipath exhibits random uctuations resulting in a spectral
broadening of the received signal, known as Doppler spreading. Due to the motion
of the transmitter-receiver pair relative to each other, in addition to the motion of
waves, the Doppler spread is increased by an additional frequency shift in the signal.
The Doppler eect due to motion on any given frequency can be modelled as [8]
fc = fc (1 ) ,

(1.1)

where fc is the transmitted carrier frequency, is the Doppler shift dened as a

ratio of

v
c

and fc is the received frequency. The (+) sign indicates an expansion sit-

uation where the distance between the transmitter-receiver pair increases, resulting
in decreasing frequency and vice versa. For narrow-band signals, i.e. (f c >> B) [8],
Doppler shift translates all OFDM sub-carriers by the same amount of carrier frequency and (1.1) is often used as an approximation; whereas in the case of wide-band
signals ( fBc 1), each sub-carrier is shifted non-uniformly [7]. In such a case, the
Doppler eect is modelled as a complete time scaling (expansion or compression) of
the signal waveform [8]
r (t) = x((1 )t).

(1.2)

where x(t) and r (t) are the source and Doppler shifted received signals, respectively.
Here, (-) sign indicates decreasing distance between the transmitter and receiver,
resulting in compression of the signal and vice versa. In this case, symbol synchronization is of equal importance to carrier synchronization. Furthermore, due
to severe multipath distortion and fading, conventional synchronization techniques
in single carrier transmission, such as phase-locked loop (PLL) [9], when coupled
with equalization, are very unreliable in underwater acoustic communications [10].
In addition, receivers adopting PLL and equalizer are considered to be highly complex. The alternative OFDM-based receivers are less complex. If the normalized
Doppler spread, dened as the ratio of the Doppler shift to sub-carrier spacing f ,
is large, which is a typical scenario for most of the underwater channels [6], special care should be taken as far as the residual frequency oset is concerned, which
4

1.2 Advances in Doppler shift compensation for UWC systems

should be << f [7] in order to mitigate inter-carrier interference (ICI), reserve the
orthogonality and achieve reliable communication systems. To demonstrate how
severe the Doppler eect can be in an underwater channel, an example comparing
both a highly mobile radio system and an underwater communication system may
be considered. Suppose a vehicle in the RF communication moving at a speed of 250
km/h, where c is 3 108 m/s, resulting in = 2.3 107 ; as opposed to this situation, consider a stationary acoustic system which may experience an unintentional
motion of 0.5 m/s which is equivalent to 1 knot, resulting in = 3.0104 . Rapidly
moving platforms, such as AUV, present a more serious problem where the factor
will be in the order of 103 [8]. Moreover, an acceleration in the order of (0.25 m
/s2 ) during the symbol time has a signicant inuence on how rapidly the channel
changes with time. Therefore, it is evident that the eect of time expansion-dilation
cannot be ignored in underwater communication. As the time varying multipath,
combined with non-negligible Doppler eect, severely distorts the signal transmission, sophisticated signal processing algorithms are required to establish error free
communication.

1.2

Advances in Doppler shift compensation for

UWC systems

Several time-domain receivers which adopt coherent modulation with an emphasis on channel equalization to increase communication reliability have been suggested. However, the time-varying doubly-spread characteristic of the UAC requires
a highly complex equalizer. An alternative low-complexity, high-speed communication scheme is the multi-carrier system in the form of OFDM in which fast Fourier
transform (FFT) is used for modulation. This system resists the frequency selectivity of the channel by dividing the broadband data into parallel narrowband channels.
Also, in a delay-dispersive environment, adopting CP of a length greater than the
maximum delay spread provides an excellent way to assure the orthogonality of the
carriers. However, the wireless propagation is considered time-varying, and thus
time-selective due to the Doppler shift in which one sub-carrier may pose ICI with
the adjacent sub-carriers. The Doppler shift sensitivity is inversely proportional to
the OFDM symbol duration; therefore, slightly moving platforms can cause serious

1.2 Advances in Doppler shift compensation for UWC systems

synchronization impairments in long sub-carriers.
Previous studies on UWC have addressed several approaches for synchronization
in the presence of Doppler distortion. For single carrier transmission, a block-based
approach [8] has been used to estimate and compensate the Doppler shift. In this approach, two LFMs are used for coarse estimation of the time scaling factor and then
an equalizer is used for residual Doppler shift compensation. Such an approach is
well suited for constant speed. An alternative adaptive Doppler compensation technique has been suggested by [11] to accommodate AUV. This closed-loop Doppler
correction necessitates high complexity when it is applied to OFDM systems because
there is a demodulation requirement. For multi-carrier transmission, the authors in
[12] utilized the principle in [8] and null sub-carriers for re-sampling factor estimation
and residual Doppler compensation, respectively. Although these algorithms attain
precise estimation by adopting preamble and post-amble, the bandwidth utilization
factor is compromised.
A point estimate of the Doppler shift is adopted in [13], therefore it is suitable for
situations where the Doppler shift stays constant or varies slowly during the packet
time. The concept in [14] has been extended to work in UAC by [1], with an iterative
cyclic prex correlation. The author employs the symmetry of the guard interval
with its replica in order to estimate the Doppler shift. This parameter is estimated
iteratively depending on the peak location and its phase with respect to the new
sampling interval and, for this reason, it is a computationally expensive search.
Authors in [15] have examined iterative Doppler estimation, channel estimation and
decoding.
The carrier frequency oset (CFO) is estimated using null sub-carriers as in [12].
Although re-sampling the signal removes the Doppler shift, a major problem with
its residual is destruction of the orthogonality of the sub-carriers due to the resulting ICI. A considerable amount of literature has been published on the subject of
combating ICI. These studies [16], [17] have presented the conclusion that successful
communication will result in mitigating ICI.
A number of previous studies have based their criteria for Doppler shift and CFO
estimation on utilizing signal space and statistics. For instance, the authors in [18]
have used maximal likelihood estimation (MLE) and estimation of signal parameter
via rotational invariance technique (ESPRIT) to estimate both CFO and Doppler
shift in wideband OFDM, while in [19], HTLS (Hankel) Total Least Square has been
6

1.3 Contributions
used for joint channel and Doppler estimation. The system requires no estimation of
the CFO and there is no need to re-sample the signal. An extension to [13] has been
suggested by [20] for symbol by symbol Doppler estimation. This method adopts
marginal maximum likelihood estimation (MMLE) to track the Doppler variation
between symbols. Despite its precise estimation, MLE has a number of associated
problems in terms of hardware implementation, where it necessitates a high degree
of complexity.
All the aforementioned researches assume the Doppler shift is constant during
the symbol period and all paths have equal Doppler shift; hence re-sampling the
signal with a unique time scaling factor is valid and therefore a symbol by symbol
approach works satisfactorily. A recent study by [21] has highlighted the need to
estimate the optimal time scaling factor in a multipath channel of dierent Doppler
shift in each path. However, in our proposed method, it is assumed that the channel
variation is mainly caused by the motion of both transmitter and receiver, leading to
a signicant time varying Doppler shift. This will consequently create acceleration
that may exceed 1 m/s2 due to speed alterations, and therefore ignoring this eect
yields a signicant ICI.

1.3

Contributions

This dissertation presents the algorithms that design low-complexity, high data rate
OFDM-based receiver structures for underwater acoustic communication that compensate the inherent Doppler shift and are applicable in the development of real-time
systems. Previous single-carrier receivers [22],[6], and [10] adopted beamforming to
attain an acceptable performance; however, these receivers are considered costly.
Furthermore, this thesis provides remedies for a signicant problem in the underwater acoustic channel (UAC) which is called time-varying Doppler shift. This
problem is caused due to the acceleration that accompanies the applications of the
autonomous vehicles. It should be stressed that this problem has not been discussed
yet by other researchers, where most of the studies assume the speed is constant i.e.,
the acceleration is zero. However, we assume the acceleration is time-varying up to
1 m/s2 during 5.5 seconds.
This thesis intends to determine the extent to which the multi-carriers modulation in the form of an OFDM and whether it can combat the channel impairments
7

1.3 Contributions
with less complexity than single carrier receivers that employ equalizers. The application of an adaptive Doppler shift compensation with DFE and beamformer was
presented in [11]. This approach accommodates AUV, moving at up to 2.6 m/s.
Instead of beamformers, the authors in [23] suggested BICM-ID and adaptive DFE
in order to reduce the receiver complexity. For Doppler compensation, they combined an adaptive technique proposed by [11]. However, the complexity still exists
due to the employment of RLS algorithm which demands extensive execution time
and memory requirements. Multi-carrier communications, [12] and [13] adopt the
block length-based technique in [8] to estimate the Doppler shift and, subsequently,
ne tune the CFO.
The proposed OFDM-based techniques are of less complexity than single carrier receivers, and consider the acceleration within a packet duration, something
which was not considered by all the aforementioned receivers [12], [13], and [20].
The proposed techniques were designed using the ZFE and BICM-ID for the channel equalization and decoding, respectively. Furthermore, this research presents a
detailed comparison between the proposed techniques and the block length-based
method at dierent channel conditions. Additionally, an adaptive iterative time
varying Doppler shift compensation receiver is suggested and investigated under
dierent scenarios and at various channel ranges. The results of the real-time implementation of the BICM-ID on the SHARC DSP are also presented. In summary,
the following points indicate the contribution of the work:
1. Evaluate coded OFDM (COFDM) with velocity variation in AUV systems.
2. Design and development of a low-complexity compensation technique for inherent Doppler shift.
3. The performance of the receiver with time-varying Doppler shift is assessed.
4. The performance of an adaptive iterative receiver approach under the inuence
of time-varying acceleration is analysed.
5. The SHARC DSP in implementing a real-time system for the BICM-ID algorithm, which is the most challenging part in the system is applied.

1.4 Publications Arising From This Research

1.4

Publications Arising From This Research

1. A. E. Abdelkareem, B. S. Sharif, C. C. Tsimenidis, and J. A. Neasham,

Compensation of Linear Multi-scale Doppler for OFDM-based Underwater
Acoustic Communication Systems, Journal of Electrical and Computer Engineering, submitted (under reviews).
2. A. E. Abdelkareem, B. S. Sharif, C. C. Tsimenidis, and J. A. Neasham,
Adaptive Time-varying Doppler shift compensation for OFDM-Based Underwater Acoustic Communication Systems, (under preparation to submit to
IEEE Journal of Ocean Engineering.).
3. A. E. Abdelkareem, B. S. Sharif, C. C. Tsimenidis, and J. A. Neasham,
Low-complexity Doppler compensation for OFDM-based Underwater Acoustic Communication Systems, in Proc. IEEE Oceans, Santander, Spain, 6-9
June 2011.
4. A. E. Abdelkareem, B. S. Sharif, C. C. Tsimenidis, and J. A. Neasham,
Adaptive Doppler-shift compensation for OFDM Underwater Acoustic Communications system, in Proc. UAM, Kos, Greece, 20-24 June 2011.
5. A. E. Abdelkareem, B. S. Sharif, C. C. Tsimenidis, and J. A. Neasham,
Time varying Doppler-shift compensation for OFDM-based shallow Underwater Acoustic Communication systems, in Proc. IEEE MASS, Valencia,
Spain, 17-22 October 2011.

1.5

Thesis Outline

The thesis is organized as follows: Chapter 2 describes the background of the practical OFDM system that will be used throughout this thesis. It also surveys the
research literature for the state of the art in UWC systems. Chapter 3 presents
dierent Doppler compensation techniques that will serve as a base in developing and improving other techniques. In addition, it presents mainly a symbol by
symbol approach to compensate the Doppler shift. Furthermore, this chapter suggests two approaches to dealing with the residual Doppler shift. Additionally, this
chapter provides an application of the SHARC Digital signal processor for the realtime implementation of BICM-ID, which is the bottleneck of the proposed receiver.
9

1.5 Thesis Outline

Chapter 4 introduces a new approach that tackles a time varying Doppler shift due
to the acceleration. This chapter addresses the main problems of such type of the
Doppler shift. Chapter 5 focuses on the solutions to the problems outlined in chapter 4 due to the acceleration eects of broadband Doppler shift on the performance
of the receiver. This chapter produces a model for the time varying Doppler shift
under dierent scenarios in order to investigate the proposed system by extensive
simulation to analyse the performance, in addition to the trial over dierent channel ranges. Finally, conclusions are drawn in Chapter 6 and the thesis ends with
suggestions for future works.

10

Chapter 2
Background
The focus of this chapter is the provision of a background to the related fundamental
materials that are used throughout this thesis such as CP-OFDM, PAPR, and pulseshaping in the transmitter and BICM-ID on the receiver side. The chapter begins
with an introduction to digital modulation schemes, which is followed by a literature
survey to complement the survey discussed in chapter 1. In this chapter, the survey
covers the advances in underwater communication, real-time implementation tools
and channel coding. Details regarding some mathematical background of BICM-ID
are also provided. In terms of hardware implementation, the aspects of the ADSP21364 SHARC processor selection to implement the decoder are discussed.

2.1

Introduction to Digital Modulation

In a digital transmission, the information is either available in binary form or it

is obtained by sampling an analogue signal. Particularly, a speech signal is rst
sampled and then quantized to appropriate signal levels to obtain binary information. In either case, it is not possible to transmit the information directly. Digital
modulation is the process by which digital symbols are transformed into waveforms
that are compatible with the characteristics of the channel. In the case of baseband
modulation, these waveforms usually take the form of shaped pulses. However, in
the case of bandpass modulation, these shaped pulses modulate a sinusoid known as
a carrier wave. For radio transmission, the carrier is converted to an EM wave for
propagation to the desired destination. The transmission of the EM wave through
the space is accomplished by the use of antennas. The wavelength and the application govern the antenna size. For mobile communication, antennas are typically
11

2.1 Introduction to Digital Modulation

/4 in size, where wavelength is equal to c/f , and c, the speed of light is 3108 m/s.
Let us consider a baseband signal where f = 3 kHz is coupled directly to an antenna
without carrier modulation: in this case, the antenna size will be 2.5 104 m, which
is not practical. However, if the baseband signal is rst modulated onto a higher
frequency carrier, e.g. a 900 MHz carrier wave, the antenna size will be 8 cm. For
this reason, carrier wave or bandpass modulation is an essential step for all systems
involving radio transmission. In the case of UWC, acoustic waves are transmitted
as the EM waves cannot propagate through sea where the carrier frequency is in the
range of 12 to 20 kHz and the speed of sound is 1500 m/s.
The choice of a modulation system is dependent upon many factors. This is
because the signal is transmitted on an imperfect channel, aggravated by the addition
of noise, and is subject to variations in the amplitude of the received signal, which is
called fading due to the rapid change in the propagation conditions. The size of the
constellation M is the main parameter that is associated with the modulation. When
M = 2, which is binary phase shift keying (BPSK) in the M-ary phase shift keying
(M-PSK) technique, each binary information bit is mapped to a polar format of 1.
Another scheme, where M = 4, is known as quadrature phase shift keying (QPSK)
which maps two consecutive bits to a symbol taken from the QPSK alphabet. The
PSK modulation scheme oers only one degree of freedom as the amplitude of all
the symbols is the same, whereas the quadrature amplitude modulation (QAM)
oers two degrees of freedom as the symbols exhibit dierent amplitude and phase
distribution. In the case of UWC, where the bandwidth is limited, it is desirable
to use higher order modulation schemes such as 8-PSK or 16-QAM where more
information bits can be packed into a symbol to increase the bit rate of the system.
However, due to the limitations of the channel which exhibits severe multipath
and low SNR, it is prohibitive to utilize such schemes as a higher data rate will
lead to more ISI. The choice of the modulation system can therefore only result
from a compromise adapted to a particular application. In UWC, QPSK or 4QAM schemes are considered. The performance of the QPSK scheme under various
channel conditions will be discussed later in the thesis.

12

2.2 Literature survey

2.2
2.2.1

Literature survey
Advances in Underwater Acoustic communications

It has been proven that phase coherent modulation techniques [24], such as PSK
and QAM is the best technique to achieve high-speed data transmission over underwater acoustic channels (UAC) compared with non-coherent frequency shift keying
(FSK) or dierentially phase shift keying (DPSK) in term of performance; however it requires dicult carrier recovery. Phase-coherent communication systems
are presented by the Woods Hole Oceanographic Institution (WHOI) in U.S.A. [25].
The modulation format is QPSK, and the signals are transmitted at 5kbps, using
carrier frequency of 15 kHz. Its real-time operation contains 4 Texas instruments
TMS320C40 DSPs. The eect of multipath is processed by a decision-feedback
equalizer (DFE) operating under RLS algorithm.
An approach for multipath rejection at the receiver end was investigated at
the University of Newcastle [22]. The researchers used adaptive beamforming with
least mean square (LMS) type to steer the reected wave. It was found that the
beamformer encounter diculties as the range increases relative to the depth because
they used 64 point correlation sequence. The real time system was implemented
using multiple DSPs. The system was tested in shallow water at 9.975kbps, and
resulted bit error rate (BER) of 2.2 102 to less than 103 .
Spread spectrum is also involved as an attractive approach to mitigate the multipath eect in UWC [26], [27], and [28]. In [28], the authors adopt spread spectrum
signals to suppress the multiple access interference, which utilize spreading codes
to recognize between users and improve the performance against multipath eects.
These techniques perform very well; however they come with the cost of reducing
the throughput in a band-limited channel.

2.2.2

Tools and Algorithms for real-time implementation

Signal processing functions, such as Viterbi decoding, can be implemented using

DSPs. Particularly, Analogue Devices [28], TigerSHARC ADSP-101S and SHARC
ADSP-21065L can be used in the baseband modem implementation. The rst of
these manipulates the Viterbi Decoder in 0.86 MIPS and 1024-point complex FFT
in 32.75 s, and has been used as a multiprocessor structure by [29] with eld
13

2.2 Literature survey

programmable gate array (FPGA) to implement an OFDM underwater acoustic
communication system. The second manipulates 1024-point FFT in 0.274 ms. In
addition, TMS320C6416 is designed for 3rd Generation Partnership Project (3GPP)
turbo code and is capable of decoding up to 12 Mbps (6 Iterations) [30]. Furthermore, in [31] the transmitter has been implemented with multiple DSPs of type
ADSP-TS101s and FPGA as the logical control. It has been proven from the experiment in the pool and lake that the signal transmitter satises requirements of signal
transmission for OFDM in real-time in underwater multi-carrier acoustic communication, and the signal synthesis method, DDWS (Direct Digital Wave Synthesis)
is adopted to realize producing OFDM. Furthermore, the authors in [32] have used
optimized built-in code for turbo decoder implementation.
Other algorithms have been implemented using Texas Instruments (TI) platforms. The authors in [33] and [34] have implemented an OFDM acoustc modem with a 225 MHz TI TMS320C6713 DSP board and it has been successfully
tested with in-air communication. In [35], TMS320C6713 and TMS320C6416 have
been utilized to implement single-input single-output (SISO) and multi-input multioutput (MIMO) OFDM acoustic receivers. These receivers are coded and investigated with low-density parity-check (LDPC) and convolutional code (CC).
FPGAs can be used to implement COFDM in addition to DSPs in acoustic
communications. For instance, they have been used in underwater communications to achieve reliable high rate data communication (1-10Kbps) in very shallow
waters [36]. The researcher has implemented the transmitter and receiver of the suggested modem using VHDL on FPGA, and has employed dierential PSK (DPSK)
modulation to avoid channel estimation at the receiver, which results in reducing
implementation complexity. DSP chips can be distinguished by their xed-point
or oating-point architectures. Currently, xed-point processors are either 16-bit
or 24-bit devices, while oating-point processors are 32-bit or 40-bit devices. The
oating-point devices are dominant in the communications systems that have large
dynamic range. The faster development cycle for a oating-point device may easily
outweigh the extra cost of the DSP device itself [37].

14

2.3 OFDM system transmitter

2.2.3

Channel coding

To achieve reliable communication over acoustic channel, channel coding in the

form of block or convolutional coding of the source bit stream should be involved
to achieve reliable communication system [38]. In 1982, Ungerboeck introduced a
trellis-coded modulation (TCM) system as a bandwidth-ecient signalling over an
additive white Gaussian noise (AWGN) channel [39]. This study set out with the
aim of assessing the importance of mapping. The most interesting nding was that
coding reduces noise about 3-4 dB compared with uncoded with the same transmitted information. However these results were not very encouraging in undersea
channel, thus for fading channels, the diversity order of the coded modulation system
should be high; therefore the performance of TCM is degraded in such channels [40],
but it can be improved by adding symbol interleaver. However, the limitation of the
diversity order in symbol interleaved coded modulation and the cost of increasing
the complexity of the code results in nding dierent approach called BICM. It was
suggested by Zahavi [41] to improve the performance of coded modulation over fading channels. It was shown that the diversity order can be increased to a minimum
number of distinct bits rather than symbols by using bitwise interleaving. It was
shown in [42], [43], [44], [45], and [46] that with iterative decoding (ID), BICM can
be used to provide excellent performance over any channel provided well designed
signal mapping. In [47], the author has exploited the diversity that has been oered
by the channel coding with convolutional BICM-ID to improve the reliability of the
UWA channel.

2.3
2.3.1

OFDM system transmitter

Peak-to-average power ratio (PAPR)

One of the major obstacles in multicarrier transmission is the high PAPR of the
transmit signal. There are many techniques [48], [49]have been used to avoid high
PAPR signals based on computation complexity, BER, bandwidth expansion. The
block diagram of SLM technique for PAPR reduction is predicted in Fig. (2.1).
The input data X = [X[1], X[2], X[Nc ]] is multiplied with U phase sequences
[ejn , ....ejn ] where n [0, 2] for n = 1, 2, Nc and u = 1, 2, , U . An OFDM
1

symbol data then contains a modied phase X u = [X u [1], X u [2], X u [Nc ]] is ob15

2.3 OFDM system transmitter

U (1)

Data
Source

Partition
into
blocks
and
serial
to
parallel

X(1)
IFFT
X

one

U (2)
X(2)
U

(n)

..
.
X(n)

conversion

Select

IFFT

with
minimum

PAPR
IFFT

Figure 2.1: Block diagram of SLM method for PAPR reduction.

tained. Converting to the time domain at each sequence U by taking the IFFT
results a corresponding sequences xu = [x[1], x[2], , x[Nc ]]. In order to select the
minimum PAPR for the transmission, each time domain OFDM symbol is produced
according to its corresponding sequence, hence the discrete-time PAPR is calculated
as
P AP R ,

max|x[n]|2
,
En [|x[n]|2 ]

(2.1)

and the minimum phase is selected as shown [50]

u = arg min (P AP R).
u=1,2, ,U

(2.2)

On the receiver, side information is transmitted to recover the optimal sequence.

Considering the hardware implementation issue, our project buers the optimal
PAPR phase sequence as a vector and then provide this information to the receiver.
It is worth to mention that this approach is useful when the transmitted message is
knew, otherwise, side information is an alternative in the case of random message
transmission. In this case, U IFFT operations are needed, whereas log2 U bits are
required as a side information.

2.3.2

Pulse shaping

The UAC is band limited, thus it constraints the transmitted signal and consequently
an increase in the decoding error due to the ISI is most likely to occur at the receiver
side. Therefore, these types of channels necessitate employing the pulse shaping in
order to preserve the bandwidth and minimize the decoding errors.
The roll-o factor govern the performance of the pulse shaping lter (PSF)
[51]. In Fig. (2.2), when = 0, it oers the most ecient use of bandwidth, but

16

2.3 OFDM system transmitter

1
0.8

Amplitude

0.6
0.4
0.2
0

=0.1
=0.5
=0.98

0.2
0.4

10

20
30
(k) Samples

40

50

Figure 2.2: Time domain characteristics of raised-cosine function for dierent values
of roll-o factor .
comes with an increasing ripples relative to > 0 in which there is an increase in
the transmitted bandwidth on the cost of reducing the ripple magnitude. The pulse
shaping prc (t) which is realized as an up-sampled raised cosine FIR lter is dened
in [52] as

prc (t) =

2.3.3

[
1
2Ts

0,

(
1 + cos

Ts

0 |t|
|t|

Ts (1)
2

Ts (1)
2

])]

Ts (1)
2

|t|

Ts (1+)
2

(2.3)

otherwise.

Guard interval in the OFDM systems

There are two main approaches that can be used to insert the guard interval in
OFDM systems. First is the zero padding (ZP) that pads zeros in the guard interval.
The second approach is to preface the OFDM symbol by the last samples, this
approach is called cyclic prex (CP). Referring to [53] and calling that we aim to
design a paradigm of a low complexity receiver that is applicable in the hardware
implementation, selecting a CP approach is now feasible.
2.3.3.1

Cyclic prex OFDM

Let X(n) be the frequency domain OFDM symbol modulates the sub-carrier with a
frequency of fn = n/T , for n = 0, 1, 2, , Nc . The duration of this symbol X(n)
17

2.3 OFDM system transmitter

was originally Ts and its length has been extended to T = Nc Ts by transmitting
parallel Nc sub-carriers. This OFDM symbol comprises a signal corresponds to Nc
sub-carriers that has a duration of T . Consider the ith OFDM signal
xi (t) =

Nc 1
1
Xi (n)ej2fn (tiT ) ,
Nc n=0

(2.4)

for iT t < iT + kTs .

For the channel with an impulse response of hi (t), the received signal is given as

yi (t) = xi (t)hi (t)+wi (t) =

hi ( )xi (t )dt+wi (t), iT t < iT +kTs , (2.5)

where wi (n) is the additive white Gaussian noise. Sampling at kTs = kTd /Nc , (2.5)
can be represented in a discrete time as follows
yi (k) = xi (k) hi (k) + wi (k) =

hi (m)xi (k m) + wi (k).

(2.6)

m=0

Let Td be the duration of the OFDM symbol without a guard interval. Then,
Td = Nc Ts = 1/f , where f = 1/(Nc Ts ) = B/Nc and B = 1/Ts . Due to the
extension of the symbol duration by Nc Ts , the eect of the multipath is reduced
on the OFDM symbol. However, there is a trade o between the symbol duration
and the CFO. The longer duration means tight sub-carrier spacing and more eect
of the Doppler shift. It is worth to mention that the impairments are still eective
among the sub-carriers due to the ISI and thus inserting a CP as a guard between
two consecutive OFDM is essential. By copying the last samples of the OFDM
symbol into its front and thus extending the OFDM symbol is called CP. Hence,
the OFDM symbol duration is changed to be T = Td + Tg , where Tg represents the
guard time. Once the length of the guard interval is selected longer than or equal
max , the ISI eect of the ith OFDM is limited within the guard interval, hence,
the i + 1th OFDM symbol is protected against the interference from the previous
symbol. However, if the length of the guard interval is selected shorter than max ,
the tail of the ith symbol aects the leading edge of the i + 1th OFDM symbol and
yields to an ISI. Now, assuming that the FFT window is completely synchronized,
i.e., within the CP interval, then the received samples yi (k) of the ith symbol after
FFT can be written as
18

2.3 OFDM system transmitter

Yi [n] =
=

N
c 1

k=0
{
N
c 1

m=0
k=0
{
N
1
c

i=0

yi (k)ej2kn/Nc

1
Nc

}
hi [m]xi [k m] + wi [k] ej2kn/Nc
hi [m]

m=0
N
c 1

l=0

{{

{N 1
c

}}
Xi [l]ej2l(km)/Nc

l=0

hi [m]ej2lm/Nc

}
Xi [l]

m=0

ej2kn/Nc + Wi [n]
}
ej2(nl)k/Nc

ej2kn/Nc + Wi [n]

k=0

= Hi [n]Xi [n] + Wi [n],

(2.7)
where the transmitted symbol Xi [n], the received symbol Yi [n] , the channel response
Hi [n], and noise Wi [n] are in frequency domain at the nth sub-carrier. Thus, it can be
inferred that once the cyclic prex is appended to the channel input, the convolution
in time domain will be converted to a multiplication in frequency domain and a
circular convolution is obtained. Therefore, taking the DFT of the channel output
yields
Y [n] = DFT {y[k] = x[k] ~ h[k]} = Hi [n]Xi [n], for 0 6 n 6 Nc 1

(2.8)

as desired in the receiver to enable a single-tap equalizer by dividing the received

symbol by the channel.
It should be stressed that increasing the CP length does not mean completely
eliminate the ISI and/or ICI. Additionally, adding a CP comes with a cost. To
be more specic, once the Doppler shift exist, the symbol time is expanded or
compressed depending on the direction of the movement, hence, the FFT window
start point is misaligned and therefore a synchronization impairments are occurred.
That is, if the FFT window comes earlier than the lagged end of the previous symbol,
ISI occurs; if the FFT window position is later than the beginning of a symbol, an
ISI and ICI are the consequences [54]. Adding a CP comes with a cost of an overhead
of Ng /Nc , where Ng is the CP length. Furthermore, the redundant data caused a
loss in the transmitted power.

19

2.4 The Underwater Acoustic Channel

2.4

The Underwater Acoustic Channel

Unlike the existing communication channels, the UACs have not characterized yet as
a standard channels. However, there are a main characteristics that should be considered in the case of establishing a reliable communication system for underwater
acoustic. These characteristics are [2]:
1. Attenuation and ambient noise
2. Doppler eect
3. Time varying multipath channels

2.4.1

Attenuation and ambient noise

The acoustic signal experiences a power attenuation while travelling through its link.
This attenuation is due to:
Energy spreading, and
Sound absorption loss.
Let r be the range from the source in km, then the attenuation of the signal is
proportional to

1
r2

when the propagation is line-of-sight. This type of spreading is

called spherical or inverse square law. When the acoustic wave is propagated via a
reection within a boundaries of the sea surface and the bottom, then the attenuation is proportional to 1r . This type of attenuation is called cylindrical spreading.
For short-range < 1 km, spherical spreading is dominant while cylindrical spreading
refers to the case of medium (1-10 km) and long-range (10-100 km) transmission.
Practically, it is interesting to formulate an equation as a function of the signal frequency that includes not only cylindrical but also spherical spreading and
absorption loss, i.e., the total transmission loss (TL) is given by [55]:
T L(f ) = k log10 (r) + a(f )r 103 ,

(2.9)

where k is 20 for spherical and 10 for cylindrical, a(f ) is the attenuation coecient
in dB/km. Several formulas for the absorption coecients have been derived in [2]

20

2.4 The Underwater Acoustic Channel

out of which, one can be given as
a(f ) = Af 2 +

Bf0
Cf1
( )2 +
( )2
1 + ff0
1 + ff1

(2.10)

where the rst term on the right hand side is fresh water attenuation, second term
is magnesium sulphate relaxation and the third term is boric acid relaxation. Additionally,
A = 2.1 1010 (T 38)2 + 1.38 107 ,

(2.11)

B = 2S 105 ,

(2.12)

C = 1.2 104 ,

(2.13)

f0 = 50(T + 1),

(2.14)

T 4

f1 = 10 100 ,

(2.15)

where S is the salinity in parts per thousand, T is the temperature in Celsius and f
is the operating frequency in kHz. In practice, a useful design rule for determining
at which point absorption losses become substantial is [56]
a(f )r < 10 dB.

(2.16)

In the ocean, another factor that aects the received SNR is the noise. It can
be classied into man-made noise and ambient noise. The latter comes from seismic
events, marine life, ships engine, rainfall, breaking waves [57] and so on. The
majority of these types are approximated as Gaussian statistics. It should be stressed
that the ambient noise is time-varying particularly in shallow water. Furthermore,
the noise level is inversely proportional to the depth. Interested reader can refer to
[57], (ch. 6) for more information.

2.4.2

Doppler eect

In general, the Doppler eect is inherent in the UAC due to the currents and wave
motions. The movement of the receiver (Rx) and/or the transmitter (Tx) also yields
to a shift of the received frequency, called Doppler frequency shift. This Doppler

21

2.4 The Underwater Acoustic Channel

shift can be represented as
v
Fd = fc .
c

(2.17)

The Doppler shift frequency is positive when the Tx and Rx move towards each
other, where the transmitted signal will be compressed resulting in an escalated
frequency. That is, the received frequency can be formulated as
v
fc = fc [1 ] = fc || .
c

(2.18)

It is obvious that the frequency shift depends on the direction of the wave, and must
lie in the range fc max fc + min , where max = fc v/c. Furthermore, due
to this movement, the path lengths are also aected, therefore, we assume here all
paths have the same Doppler shift.
Since the propagation speed c is relatively small, as mentioned earlier, it seems
the Doppler shift has a signicant inuences on the underwater link. This eect is
arisen with an OFDM systems due to its sensitivity to Doppler shift. The sensitivity
is proportional to the sub-carriers spacing which in turn depends on the symbol
length. We assume the channel is quasi-static in this thesis. However, the Doppler
shift is varying linearly within the OFDM symbol time as a result of the acceleration
eect. This acceleration adds a burden on the receiver and needs a special signal
processing such as an adaptive algorithm or iterative receiver to deal with this
obstacle.

2.4.3

Time varying multipath channel

The time varying channel can be characterized by an impulse response h(t, ) and in
a frequency domain is characterized as H(t, f ). Furthermore, the time varying multipath channel is also characterized in terms of delay-Doppler spread C(Fd , ). All
aforementioned characteristics are related via a two-dimensional Fourier transforms.
In particular, H(t, f ) and C(Fd , ) are related to h(t, ) as [58]

H(t, f ) =

h(t, )e

j2f t

dt

C(Fd , ) =

22

h(t, )ej2Fd t dt

(2.19)

2.4 The Underwater Acoustic Channel

Let us assume that the channel is wide sense stationary (WSS) process in t and f ,
then the time-frequency correlation function
RH (t, f ) = E {H(t + t, f + f )H (t, f )} .

(2.20)

In terms of signal distortion, H(t, f ) represents the eect of the channel in time
and frequency. Considering this, there are a set of channel spread parameters Tcoh
and Bcoh that can be used to identify the channel variations in time and frequency,
respectively.
2.4.3.1

Doppler spread and coherence time

Although the maximum delay spread max and coherence bandwidth Bcoh do not
provide information about the time-varying nature of the channel due to the relative
motion between the transmitter and receiver, it can be considered as an indicator
of the time dispersive nature of the channel. Practically, the channel is changing
with time, therefore; it is very crucial to understand these time variations and how a
digital communications system responds to it. In practice, the receiver is equipped
with a mechanism to estimate the time variations due to the Doppler shift discussed
in Chapter 4 in more detail.
The Doppler spread (Fd ) of the channel is the range over which the Doppler spectrum is non-zero. The Doppler spread and coherence time are inversely proportional
to each other. That is [9]
Tcoh ,

1
.
Fd

(2.21)

The amount of the spectral broadening depends on Fd which is relative to the velocity
of the mobile.
The value of Tcoh is actually a statistical measure of the time duration over
which the channel impulse response can be considered invariant. In other words,
coherence time means that the duration over which the channel is highly correlated.
The channel variation in time is captured by the Doppler spread, therefore, this
spreading factor is used to indicate the rapidity of the channel variation with time.
Accordingly, a channel can be classied into a fast time-varying and a slow fading
channel.
If Tcoh is less than the OFDM symbol time Td , the channel will change during

23

2.4 The Underwater Acoustic Channel

the transmission of a symbol and corrupt the source signal severely. This is called
fast or time selective fading. However, if Tcoh is longer than the symbol duration,
the CIR will change very slowly compared to the symbol rate. This type of fading
is called slow fading. If this is the case, the channel can be assumed to be constant
over the signalling interval. Another way to characterise the channel is in terms of
the product Fdmax max which is known as the spread factor [59] of the channel. The
channel requires to full that the spread factor is 1 to be considered as underspread
in which case the channel can be measured without any ambiguity. However, if this
condition is not fullled, i.e., the value of the spread factor is greater than 1, then,
the channel is called overspread which makes it dicult to measure the CIR.
2.4.3.2

Delay spread and coherence bandwidth

While max of the channel can be computed as the time dierence between the latest
to earliest arrival, the channel coherence bandwidth, Bcoh , is derived from the delay
spread. That is, max is inversely related to the channel coherence bandwidth as [9]
Bcoh ,

1
max

(2.22)

The coherence bandwidth is a measure of the frequency selectivity in a multipath

propagation. The frequency non selective fading is the range over which the channel remains almost constant in frequency. If the channel has a constant gain and
phase over the transmitted signal bandwidth (B) which is less than the coherence
bandwidth (Bcoh ), then the received signal will undergo at fading. However, the
received signal undergoes frequency selective fading when B is much greater than
the Bcoh of the channel, which means that all the transmitted frequency components
will be similarly attenuated. This implies that the symbol duration is smaller than
the maximum delay spread of the channel resulting in ISI in the received signal
and the OFDM modulation technique is suitable for such case to mitigate these
eects. In reality, the assumption of WSSUS does not hold for all the values t and
. Therefore, the concept of a quasi-WSSUS channel is used which implies that the
WSSUS assumptions hold for some values of t and observed by a communication
system. In a simpler context, the channel remains constant for the duration of the
transmitted signal but it can change for the next packet.
Fig. 2.3 and 2.4 illustrate typical normalized channel impulse responses for 500
24

2.4 The Underwater Acoustic Channel

Normalized magnitude

1
0.8
0.6
0.4
0.2
0

0.002 0.004 0.006 0.008 0.01 0.012

Delay (s)

Figure 2.3: Normalized CIR of a 500 m range channel.

Normalized magnitude

1
0.8
0.6
0.4
0.2
0

0.005
0.01
Delay (s)

0.015

Figure 2.4: Normalized CIR of a 1000 m range channel.

25

2.5 BICM-ID OFDM system

and 1000 m range link set-ups, respectively. These CIRs were measured by means
of the transmission of a linear frequency modulated (LFM) chirp signal and then
correlation of the incoming chirp signal at the receiver end. The 500 m range channel
exhibited a delay spread in the order of 10 ms which translates into 40 symbols of ISI
at the data rate of 4 ksps. It was conrmed that the 1000 m range channel exhibited
a delay spread of 5 ms, which is 2 times less than that of the 500 m range channel
and translates into 20 symbols of ISI at the data rate of 4 ksps. There are many
ways in which mitigation of channel impairments can be achieved; for example, by
employing trellis based equalizers, a lter based equalizer, OFDM or beamforming.
The problem is exacerbated when the channels vary rapidly within time. In such
cases, it is necessary to employ a mechanism that is capable of estimating time
variations at the receiver end. In this thesis, an OFDM technique is employed to
address these time variations due to its low complexity as far as implementing the
receiver is concerned.

2.5

BICM-ID OFDM system

In a multipath fading environment, the forward error correction (FEC) is used to

reduce the error probability. However, this error performance comes with the cost of
transmission bandwidth reduction and receiver complexity. An alternative spectralecient TCM is introduced by [39] which improves the performance of the bandlimited communication systems by jointly optimizing the coding and modulation
scheme. A reduction in the bandwidth expansion and signicant coding gain are
obtained from this approach.
Practically, a robust communication system should perform well not only in an
AWGN such as TCM, but also in a fading channel. This is due to the burst of errors
that accompanies the fading channel which is out of the capability of many FEC
types in the case of long burst of errors. Hence, the emergence of interleaving has
devised to reduce the eect of such type of error with a coded system in a multipath
fading channels. The main eect of the interleaving with a coded modulation is
to randomise the positions of the errors within a burst and ultimately convert this
burst of errors into a random error that can be sorted by the FEC. This is the main
task of the interleaver at the transmitter. The length of the interleaver Lint plays an
important rule in making the fading independent and the longer is the best, however
26

2.5 BICM-ID OFDM system

the cost is the implementation complexity.
In order to achieve robust performance over a dierent channel conditions, interleaved coded modulation should exhibit both large Euclidean and Hamming distances [47]. Therefore, in a fading channels, maximizing the diversity order is the
main target of designing an interleaved coded modulation scheme. Unlike block
and convolutional codes, there are two options for interleaving in coded modulation. First is to interleave the bits and then map them to modulated symbols. This
option is called BICM and it is now a dominant technique for coded modulation
in fading channels. Alternatively, the modulation and coding can be done jointly
as in coded modulation for AWGN channels and the resulting symbols interleaved
prior to transmission. This technique is called symbol-interleaved coded modulation
(SICM).
A major breakthrough in the design of coded modulation for fading channels
was the discovery of BICM [41], [60]. In BICM the code diversity equals to the
smallest number of distinct bits rather than channel symbols along any error event.
This is achieved by bit-wise interleaving at the encoder output prior to symbol
mapping, with an appropriate soft-decision bit metric as an input to the Viterbi
decoder. While this breaks the coded modulation paradigm of joint modulation and
coding, it provides much better performance than SICM. Moreover, analytical tools
for evaluating the performance of BICM as well as design guidelines for good performance are provided in [60]. BICM is now the dominant technique for improving
the performance of coded modulation in fading channels [61].
As mentioned earlier, the conventional BICM achieves good performance providing high diversity order. However, due to the randomization in the modulation
caused by the bitwise interleaver, the Euclidean distance is reduced. Therefore, there
is a degradation in the performance of the BICM in Gaussian channels compared
to TCM [41]. To counter this problem, an iteratively decoded BICM with carefully
selecting signal mapping, referred to as BICM-ID has been proposed in [62], [63],
[64], and [65]. The idea behind BICM-ID is to increase the Euclidean distance of
the BICM code and to exploit the full advantage of bit interleaving by performing
soft iterative decoding technique [46]. BICM-ID was shown to be better than TCM
and BICM in both AWGN and uncorrelated Rayleigh fading channels [66].
The system diagram of BICM-ID is illustrated in Fig. 2.5. The transmitter
contains a serial concatenation of FEC, an interleaver and and signal mapping.
27

2.5 BICM-ID OFDM system

bi

FEC

ck

Mapping

sn

wn
Le (sk )

bi

Max
Log
MAP

La (k)

La (k)

Le (sk )
1

rk
Soft
Demapper

Figure 2.5: System diagram of BICM-ID.

The block diagram of a FEC that contains non-systematic convolutional (NSC) code
is illustrated in Fig. 2.6. Unlike systematic code, NSC has a better error performance
at large SNR. A sequence of Kd information bits, b = (b0 , , bKd 1 ) are sent by
the source. In order to protect this message, a convolutional code of rate Rc (0, 1]
is used to produce a sequence c = (c0 , , cKc 1 ) of Kc = (Kd + K0 )/Rc coded
bits, where K0 0 is the overhead introduced by the encoder, i.e. a termination
sequence to set the nal state of the encoder to zero.
An important characteristic of convolutional code is the constraint length K,
which is dened as the number of stages that shift the message bits. In an encoder
of M -stage shift register, then M = K 1 ip-ops are required. This type of
convolutional encoder thus contains 2M states that denoted in binary or decimal
form. Thus the state of an encoder shown in Fig. 2.6 contains two ip-ops can take
22 values. The interleaver is used to randomize the codeword and ultimately reduces
the burst errors introduced in the transmission. The de-interleaver is an opposite
procedure providing that the positions of the scrambling operations are known by
the receiver.
The encoded bits are permuted by a random interleaver of length Lint and the
output bit sequence, ck , is grouped to form the sub-sequences Cn , [cn,1 , , cn,m ],
where m = log2 M represents the number of bits per symbol and M is the constellation size of the utilized modulation scheme. Subsequently, each Cn is mapped to M ary symbols, sn , taking values from the M -ary symbol alphabet = {1 , , M },
where i C and C denotes the set of complex number.
This chapter considers signal transmission on AWGN channel to simplify the
receiver derivation. The application of BICM-ID in the case of frequency selective
and Doppler shift channels is considered in next chapter. Therefore, the received
28

2.5 BICM-ID OFDM system

(1)

ci

bi

bi1

ci

bi2

(2)

ci

Figure 2.6: Convolutional encoder of rate Rc =1/2.

signal can be written as
rn = sn + wn ,

(2.23)

where wn is the complex zero mean Gaussian noise with variance w2 in each real
dimension.
At the receiver, the demapper processes the received complex symbols rn , the
P [Cn,i = 0]
of the coded bits and outputs
corresponding a priori LLRs La [Cn,i ] = log
P [Cn,i = 1]
extrinsic LLRs
Le [Cn,i ] = log

P [cn,i = 0|rn , La (Cn )]

La (Cn,i ),
P [cn,i = 1|rn , La (Cn )]

(2.24)

where Cn,i denotes the binary random variable with realizations cn,i {0, 1}.
Let ib denote the subset of symbols sn , whose bit labels have the value
b {0, 1} in position i {1, 2, ..m}. Using Bayes rule and taking the expectation
of p(
rn |sn ) over P [sn |Cn,i = b], sn ib yields

xn i0

p(
rn |sn )P [sn |Cn,i = 0]

sn i1

p(
rn |sn )P [sn |Cn,i = 1]

Le [Cn,i ] = log

(2.25)

The rst term p(

rn |sn ) is computed according to the channel model assuming a
Gaussian distribution [67]
|
rn sn |2
1 2 2
w
.
p(
rn |sn ) =
e
2w2

(2.26)

The second term P [sn |Cn,i = b] is computed from the a priori information of the

29

2.5 BICM-ID OFDM system

0

10

1st Iteration
3rd Iteration
UnCoded

Bit error rate (BER)

10

10

10

10

3
4
EbNo (dB)

Figure 2.7: Performance comparison of BICM-ID OFDM and uncoded system.

individual bits [45]
P [sn |Cn,i = b] =

1+
j=1,j=i

eLa (Cn,j )

eLa (Cn,j )cn,j .

(2.27)

The extrinsic estimates Le [Cn,i ] are deinterleaved and applied to the Log MAP
channel decoder. Performing iterative decoding, extrinsic information about the
coded bits from the decoder is fed back and regarded as a priori information La [Cn,i ]
at the demapper. During the initial demapping step, the a priori LLRs are set to
zero.
The MAP or BCJR [68] decoder is preferred in implementing such soft-in-soft-out
(SISO) algorithm because it has better performance. In this algorithm, we need to
compute the forward-state metric , the reverse-state metric , the branch metric
. The forward state probabilities k (s) can be calculated as [69]
k (s) =

k1
(
s)k (
s, s).

(2.28)

Thus, once the k (

s, s) values are known, the k (s) values can be calculated recursively. Assuming that the trellis has the initial state S0 = 0, the initial conditions
for this recursion are
0 (S0 = 0) = 1,
0 (S0 = s) = 0, s = 0.
30

(2.29)

2.6 The Eects of the Interleaver

The backward state probabilities k (s) can similarly be calculated recursively as

k1
(
s) =

s, s).
k (s)k (

(2.30)

The initial conditions for this recursion are

N
(SN = 0) = 1,

N
(SN = s) = 0, s = 0.

(2.31)

The transition probabilities k (

s, s) can be calculated using the received sequence
and available a priori information, which can be written as
k (
s, s) = P (yk |xk )P (uk ),

(2.32)

where uk is an input bit necessary to cause the transition for state Sk1 = s to state
Sk = s; P (uk ) is a priori probability of this bit, xk is the transmitted codeword
associated with this transition, yk is the received codeword associated with this
transition and k is the time index.
However, it is complex in a real time environment. A sub-optimal version of MAP
is max-log-MAP or linear approximation to log-MAP is adopted, which performs
the max operations (Jacobian logarithm) dened as [70]
max (x, y) = max(x, y) + log(1 + e|xy| ),

(2.33)

where x and y represent the modulated encoded bits and received intrinsic information with noise, respectively. Sometimes, the expression log(1 + e|xy| ) is
approximated using a constant or ignored.

2.6

The Eects of the Interleaver

Due to the impairments of the signal transmission that are caused by the multipath
fading channel, the received signal arrives at a dierent phase and with distortion.
In addition, UWA channels suer from ambient noise and other burst noise (e.g.
ship engine noise, noise of sh and humans). All of these impairments result in a
dependency among successive symbol transmission. That is, the disturbances cause
31

2.7 DSP platform selection issues

errors that occur in burst rather than individual events. This case is an example of
the channel with memory which cannot be considered as a single random bit error
and it causes degradation in error performance. Therefore, such burst errors of the
channel are dealt with by the use of interleaving. The idea behind the employment
of interleaving the coded bits before transmission and a corresponding deinterleaving
after reception is to spread in time the burst errors caused by the deep fade of the
channel, hence translating them into random errors, and thus enabling the FEC to
work eectively in correcting burst errors [71].
In this thesis, the aim is to use a BICM-ID as a tool to mitigate bursts of channel
error. The iterative decoding algorithms are derived assuming that all input LLRs
to the SISO modules are reliable. However, the system performance degrades when
the Doppler eect exists. When trying to detect a particular code bit and calculate
its output LLR, a SISO module uses the input LLRs for the nearby code bits in
the computations. This means that the correlation between nearby input LLRs will
cause performance degradation and hence this scrambling operation is necessary
between the SISO modules.
There has been much focus on interleaver design regarding BICM-ID and processing of the signals [72] and [73]. Considering BICM-ID for a time-varying channel
is dicult because the inner code, i.e. the channel, cannot be considered as known
when designing the interleaver. The interleaver should therefore be as random as
possible. Since the focus of this work is to deal with the Doppler eect and tackling
the channel by designing BICM-ID system as a tool, therefore it is considered that
the bit redistribution pattern is known by the receiver for the purpose of deinterleaving before decoding. Furthermore, the proposed system uses a block interleaver,
implying that the interleaver operates on a block of coded bits at a time. Unless
otherwise stated, the length of the interleaver is assumed to be equal to the length
of the codeword.

2.7

DSP platform selection issues

Selecting the most appropriate DSP processor and tackling a real-time signal is an
important issue. Programmable DSP is more exible, of a lower cost and a higher
speed than other processors, so it has become the best solution for many communication, medical, and industrial products because traditional microprocessors are
32

2.7 DSP platform selection issues

Figure 2.8: SHARC ADSP-21364.

inappropriate for such applications. The main aspects of selecting a DSP processor
are as follows: data format, memory bandwidth, CPU architecture, and million integer operation per second (MIPS) or million oating point operations (MFLOPS)
[74].
In terms of data format, xed point DSPs are generally cheaper, but produce
higher quantization noise. This will be added to the signal and lower the signal to
noise ratio of the system. In addition, extra code has to be written to overcome the
overow or underow and the programmer should be aware of what scaling needs
to take place. In comparison, oating point devices have better precision, a higher
dynamic range, and a shorter development cycle [74].
As there is iterative decoding in the suggested receiver and the algorithm spends
most of the execution time, especially the SISO algorithm because of the add compare select (ACS), it is important to take advantage of some of the available architecture, such as super Harvard architecture (SHARC), as shown in Fig. (2.8),
because it includes an instruction cache in the central processing unit (CPU) and
has split instruction and data buses. This feature is important with regard to avoiding any conict between data and instruction transfer during the fetch cycle, and
to ensure the program memory does not have to be accessed for the instructions to
be restored. Consequently, all of the memory for CPU information transfers can be
accomplished in a single cycle, which results in a high memory access bandwidth.
Additionally, on-chip memory is a key factor to be considered when deciding which
DSP device to use, because the memory should be sucient enough to hold the
digitized samples.
The third aspect of selecting a DSP is the CPU architecture. For instance, traditional architecture uses single memory for both data and instruction, whereas some
DSPs have very long instruction word (VLIW) core architectures; thus they execute

33

2.7 DSP platform selection issues

Table 2.1: DMA operation
Receive Process Transmit
Block A

Block B

Block A

Block C

Block B

Block A

Block A

Block C

Block B

Block B

Block A

Block C

Block C

Block B

Block A

multiple instructions in parallel, resulting in fast operations. However, these types

of architectures [75] dissipate more power than conventional DSP architectures. In
contrast, SHARC has been improved by using separate memories for data and instruction. In addition, it includes a high speed I/O controller to support direct
memory access (DMA). Furthermore, SHARC utilises shadow registers for all the
CPUs registers. They are used to accomplish the interrupt quickly by moving the
entire register contents to these registers in a single clock cycle.
SHARC ADSP-21364 has been selected to use the direct memory access (DMA)
chaining facility, which allows the DMA controller to auto-initialize itself between
multiple DMA transfers. A section of internal memory, called the transfer control
block (TCB), is where the DMA attributes are stored for each DMA operation.
A chain pointer is also associated with each DMA operation. Basically, the chain
pointer (an address to a TCB) links one DMA operation to the next. To properly set
up and initiate a chained DMA, the TCBs should rst be set up with the appropriate
attribute information. To enable the chained DMA, the DMA enable and chain
enable bits in the corresponding DMA control register should be set simultaneously.
The DMA controller will auto-initialize itself with the rst TCB, then start the rst
transfer. When this transfer is over, if the current chain pointer register is non-zero,
it will be used as a pointer to a new TCB and the process will begin again as shown
in Table 2.1.
The challenge of any oating point architecture for the purpose of real-time
application is the number of operations that can be carried out simultaneously. A
benchmark has been used to express the speed of a microprocessor as a number.
For example, [74] has pointed out that oating point devices can be specied by
MFLOPS and MIPS to specify xed point devices. This gauge is useful only in terms

34

2.8 Chapter Summary

Table 2.2: Receiver operation
Receiver stages
Additions Multiplications
BPF

49

50

Symbol likelihood

102

31

Decoding

4698

627

of a single, known, processor architecture; so MIPS and MFLOPS is misleading [76]

because the amount of processing required by an instruction can vary depending on
the instruction format of that processor. Also, it ignores subscripting, memory trac
and the countless other overheads associated with program execution. However,
it is useful to determine the minimum specications of the platform. Therefore,
in the current application, eorts have been focused on how many operations are
performed in the receiver, where it contains the most complex parts such as iterative
ACS in the SISO decoder. Table 2.2 demonstrates the number of operations (add
and multiply) required for each stage in the receiver. It is noticeable from this
table that the minimum number of operations are in the band pass lter (BPF) and
maximum number of operations in the decoding stage. Therefore, using available
DSPs could help in calibrating the system and make a rst estimation of the required
specication of the proposed system.

2.8

Chapter Summary

This chapter provides a background of a coded CP-OFDM system with an overview

of the channel characteristics for wireless communication systems. A feature of the
OFDM BICM-ID system is provided with a comparison in terms of performance
against an uncoded OFDM system. The main characteristics of a multipath channel are explained. The key points in selecting the DSP platform are discussed in
detail. Based on these aspects, the selection of an ADSP-21364 SHARC processor
is justied.

35

Chapter 3
Low-complexity symbol-by-symbol
Doppler shift compensation
In this chapter, low-complexity Doppler shift compensation techniques for OFDMbased UWAC receivers are proposed. Three techniques are demonstrated in this
chapter in order to establish a base for developing further Doppler shift compensation
algorithms. The rst method is based on a one-shot estimation that independently
manipulates the Doppler shift for each OFDM symbol within the packet. The second
approach presents an algorithm to cope with a time variation of the Doppler shift
between each OFDM symbol, depending on its preceding neighbours values. This
algorithm relies on the concept of the nearest neighbour rule to facilitate smoothing
between symbols, and a dynamic symbol synchronization point to update the time
scaling factor. To accomplish this, an adaptation step is derived, involving the
weight of the nearest neighbours time scaling in estimating the integer part of the
re-sampling factor. The fractional part of the Doppler frequency shift is considered
as a CFO. Based on this, a proposed approach that accommodates a broadband
Doppler shift is devised. This algorithm exploits the integer and the fractional part
of the time expansion/compression measured within a fraction of a sample period
in each OFDM to jointly estimate the Doppler shift and its residual.
All aforementioned methods, instead of utilizing the whole guard interval, have
exploited a nite length window of the cyclic prex for correlation in each OFDM
symbol in order to estimate the Doppler shift frequently. No iterative computation
is required for the interpolation factor estimation. Furthermore, the proposed algorithms need to only buer one OFDM frame before data demodulation, instead

36

3.1 OFDM system description

QPSK

x(t)

FEC
M
U
X

Mod
Pilots

Chirp

<{}

PSF

SLM
MUX

bi

ui

IFFT
CP

Silent

ej2fc nTs
Figure 3.1: Proposed transmitter structure, where the operator represents the
real part of the signal .
of buering the whole data packet. Thus, it demands a lesser degree of complexity and memory requirements. Moreover, all proposed techniques rely on a single
preamble of a packet consisting of multiple OFDM symbols to detect the start of
the packet; hence, the throughput is increased. An experiment was conducted in the
North Sea during 2009 and the algorithms were compared with the block Doppler
compensation technique. Results revealed that there is variation in speed during the
packet time; therefore, the proposed system surpasses the block technique. It was
conrmed that the time scaling factor of the adaptive system was estimated for each
OFDM symbol, whereas the block approach failed in estimating these variations.

3.1
3.1.1

OFDM system description

System and channel models

The proposed system to be investigated contains the transmitter depicted in Fig.3.1.

At each instant i, the encoder receives a vector of information bits bi of length Kd
at its input to produce a binary code of length Kc = Kd /Rc encoded bits, where
Rc (0, 1] is the coding rate of NSC code. The coded bits are permuted by a random
interleaver, then converted in groups of m successive bits into alphabet symbols of
constellation size M = 2m . This mapping operation generates a sequence of Nd =
Kc /m : s = {s0 ....sNd 1 }, where si C and C denotes the set of complex symbols.
Subsequently, in the OFDM symbol to be constructed, pilot symbols of phase shift
keying (PSK) with unit amplitude are embedded with the data symbols in a comb
method. These pilot symbols are used for the purpose of channel response estimation
at the receiver. A PAPR reduction is introduced using the SLM technique [50]. To
37

3.1 OFDM system description

implement this technique, a sequence of phases U are added in the transmitted signal
to be multiplied by the input data sequences and the symbol sequence of minimum
PAPR is selected for transmission. The resulting OFDM symbol, containing Np
pilots and Nd data-bearing sub-carriers, where Nd + Np = Nc , is then modulated by
an IFFT of size Nc and the last samples are copied and prefaced to the symbol to
form the CP-OFDM frame. The guard interval of length Ng is chosen to be longer
than the channel dispersion time in order to minimize the inter-symbol interference
(ISI). The resulting frame is pulse shaped, using a pulse shape lter (PSF), and then
up-converted using carrier modulation. Let Td denote the OFDM symbol duration
and Tg the guard interval. The total OFDM frame duration is T = Td + Tg . Let
fn = fc +nf , being the carrier frequency corresponding to each of the sub-carriers of
the OFDM spectrum, where f = 1/Td is the frequency separation between alternate
sub-carriers and fc is the carrier frequency, so the bandwidth is B = Nc f . The
time-domain representation of the ith OFDM symbol is given by

xi (t) =

1
j2 Tn (tTg iT )
d
di (n)uopt
prc (t iT ),
i (n)e
Nc nI

(3.1)

for iT t < (i + 1)T,

where di (n) is the symbol transmitted over the nth sub-carrier, Uopt is the optimum
phase set [ui (1), ui (2), ....ui (n)] for lower PAPR with ui (n) = ejn , n [0, 2], I
denotes the set of modulated sub-carriers and prc (t iT ) is the pulse shaping lter,
which is realized as an up-sampled raised cosine FIR lter. An equivalent passband
model of (3.1) is
}

n
1
(tT
iT
)
j2
g
Td

prc (t iT ) ,
x(t) = ej2fc t
di (n)uopt
i (n)e
N
c nI
i=0
{
}
1
j2fn (tTg iT )

=
di (n)uopt
prc (t iT ) ,
i (n)e
N
c
i=0
nI
{

(3.2)

It is assumed that the signal is transmitted over a multipath fading channel characterized by
h(, t) =

L1

hl (t)[ l (t)],

l=0

38

(3.3)

3.1 OFDM system description

where hl (t) are the path amplitudes, l (t) are the time-varying path delays and L
is the total number of paths. As in [12], we assume the path delay l and the gains
hl are constant over the frame duration T . For perfect OFDM synchronization, and
providing that the maximum delay spread is within the guard interval, the received
passband signal can be written as
{

1
j2fn t
di (n)uopt
i (n)e
Nc nI
}
L1

hl prc (t l )ej2fn l + wi (t),

r(t) =

(3.4)

l=0

where wi (t) is a white Gaussian noise with variance 2 ; hence down-conversion and
removing the CP yields the received baseband signals which are thus expressed by

r(n) =

Hi (n)xi (n) + wi (n),

(3.5)

i=0 nI

3.1.2

Doppler shift in wideband communication

When the Doppler is present, a transmitted signal is received as:

v
r(t) = x[(1 )t l ],
c

(3.6)

where the (+) sign indicates an expansion of the signal since the distance is increased
and vice versa. The magnitude of the spectrum of r(t) can be written as
|R(f )| = |X[

f
]|.
(1 v/c)

(3.7)

Sampling at an integer multiple of the carrier frequency assuming zero Doppler and
replacing the time index with k gives
r(k) = x(

k
l ).
fs

(3.8)

With Doppler, the received signal is then given by

r(k) = x[

k(1 vc )
l ],
fs

39

(3.9)

3.1 OFDM system description

Adjusting the sampling frequency by the same Doppler shift gives a new sampling
frequency of
v
fs = fs (1 ),
c

(3.10)

k(1 vc )
r (k) = x[
l ],
fs

(3.11)

which forms a new received signal

Substituting (3.10) in (3.11) yields,

r (k) = x(

k
l ) = r(k).
fs

(3.12)

At this point the results are identical and processing of the data can proceed as in
the zero Doppler case.
For narrow-band signals, (i.e, f c >> B), Doppler shift translates all OFDM
sub-carriers by the same amount of carrier frequency, whereas in the case of wideband signals, (f c = 1.5B), Doppler shift translates each sub-carrier by a dierent
amount. Let Ts be the sampling period: in such a case, the Doppler eect is modelled
in discrete time as a complete sampling period scaling (interpolation or decimation)
of the signal waveform [8]
r[kTs ] = x[k(1 )Ts ],

(3.13)

where, k is an integer, and x(kTs ), r(kTs ), are the sampled signals transmitted and
Doppler shifted received sampled signals, respectively. This wide-band model results
in an inevitable symbol timing error and CFO. Equivalent to (3.13), the Doppler
shifted received frame is modelled by
Lf = (Lf ),
where Lf =

Nc
BTs

(3.14)

represents the transmitted passband samples length and Lf is

the Doppler shifted received passband samples length. To remove both CFO and
symbol shift, an inverse time scaling of the received (compressed/expanded) signal
should be achieved, providing that the amount of Doppler shift is known. This is
equivalent to changing the sampling rate of the passband signal by 1 + in discrete40

3.2 Doppler compensation techniques

time processing. From (3.14), it can be inferred that increasing or decreasing the
length of samples is equivalent to re-sampling the sampling frequency fs by 1 + ;
thus an equivalent to (3.13) is rewritten as
fs = fs /(1 ).

(3.15)

We assume that all paths have a similar , therefore the received signal in (3.4)
can be rewritten as
{

1
j2fn (1+)t
di (n)uopt
i (n)e
Nc nI
}
L1

hl prc ((1 + )t l )ej2fn l + wi (t),

r(t) =

(3.16)

l=0

The passband signal model in (3.16) is modulated at fc , thus the corresponding

{
}
baseband model r(t) such that r(t) = r(t) ej2fc t can be written as

r(t) =

j2nf t j2fn t
di (n)uopt
e
i (n)e

nI

L1

hl prc [(1 + )t l ] ej2fn l + wi (t)

l=0

(3.17)

j2nf t j2fn t
Hi (n)di (n)uopt
e
+ wi (t),
i (n)e

i=0 nI

where Hi (n) is the channel transfer function of the ith symbol at nth sub-carrier
and can be written as

Hi (n) =

L1

hl ej2fn l prc [(1 + )t l )] .

(3.18)

l=0

3.2
3.2.1

Doppler compensation techniques

Block length-based Doppler compensation

In order to compensate the Doppler shift on a received signal, it is necessary to adopt

a method that is capable of estimating the interpolation factor and then apply its

41

3.2 Doppler compensation techniques

Doppler
estimator

Input
signal

Interpolator
1+

Receiver

Figure 3.2: Open loop Doppler correction.

inverse on the received signal. This system is shown in Fig. (3.2) [8]. This approach
provides a generic preprocessor that can be used with wide-band receiver structures.
The interpolator structure can be used either on bandpass or baseband signals.
For a complex baseband interpolator structure, the carrier frequency oset must
be removed prior to demodulation. Baseband interpolation oers a considerable
computational saving for relatively narrow-band signals; however, for an underwater
communication system, which is inherently broadband, this saving is not signicant.
LFM
CHIRP

DATA PACKET

LFM
CHIRP

Ttp Trp

Figure 3.3: Packet length measurement using chirp correlation.

Therefore, the target is to estimate the interpolation factor precisely. In
[8], a novel block length-based approach was presented in order to estimate the
Doppler shift for single carrier transmission by comparing a prior knowledge of the
transmitted data packet duration (Ttp ) with the received Doppler shifted packet
can be written as
(Trp ), as shown in Fig. (3.3). The Doppler shift estimate
= Trp 1.

Ttp

(3.19)

This equation can be considered as a coarse estimation of the Doppler shift for both
single and multi-carriers transmission.
Block length-based algorithm can be summarised for an OFDM system as follows:
1. Design a chirp signal of duration 50 ms with a bandwidth in the range
[fc B/2, fc + B/2].
2. Formulate a packet that contains 10 OFDM symbols with a chirp at the pre
and post-amble. A silent period of the same LFM duration is set after and
42

3.2 Doppler compensation techniques

BPF

ADC

PreAmp

Transducer

Packet
Synchronization

r[k(1 + )]

Doppler
Extraction

Resample

Smoothing

Int{.}
e j2(fc +)kTs

r(k)

Frac{.} 
new

Fd (coa)

(a)
ui
r(k)

-CP

FFT

ZFE

Soft
Demaper

BICM-ID

bi

(b)

Figure 3.4: Receiver structure of the proposed system.

before the pre-amble and post-amble, respectively.
3. For simulation purposes, set an array of dierent speeds in accordance with
each OFDM symbol in the transmitted packet.
4. FIR correlate the received signal with the LFM signal to detect the maximum
peaks that associate the pre-amble and post-amble chirps.
5. Apply (3.19) to estimate the interpolation factor.
In the block length based approach, the resolution of the Doppler shift estimation
is proportionate to the packet duration. It is worth mentioning that this approach
is very accurate for a xed speed situation, in which the estimated speed represents
the average or the mid-point of the packet. However, this case is not pragmatic, particularly in a medium range, where the Doppler spreads are already found undersea,
regardless of the systems mobility [4]. Additionally, in the OFDM systems, as the
sub-carriers bandwidths are mostly tight, such a method is not considered due to
the residual Doppler shift or CFO impairments. Therefore, all suggested techniques
in this thesis consider the CFO to achieve reliable communication.

3.2.2

One-shot Doppler shift compensation

The receiver structure of the suggested technique is depicted in Fig. (3.4). In the
preprocessing stage, a bandpass lter 8-16 kHz is designed to remove unwanted
43

3.2 Doppler compensation techniques

i = /2

Tg
CP

Td

111
000
000
111
000
111

000
OFDM Symbol 111
min

max

(i)

Figure 3.5: Correlation operation in (3.21), and i = ( 2 ) represent the leading and
trailing edge of the OFDM frame, respectively.
sidelobes. After bandpass ltering, the received samples are passed through a FIRcorrelator to detect the start of the packet. The resulting Doppler shifted samples
r[k(1 + )] are then given as input to the Doppler extraction unit (DEU).
In one-shot algorithm, the Doppler shift is estimated on symbol-by-symbol basis
and independently. This method estimates the Doppler shift and its CFO based
of the current OFDM symbol only, regardless of
on the estimated Doppler shift
a change in the speed from symbol to symbol during the packet time. In order to
the DEU mentioned earlier is designed to
estimate the re-sampling parameter ,
comprise two main stages that are employed as a preprocessor for all techniques in
this chapter. These stages will be explained in detail in the following sections.
3.2.2.1

Coarse Doppler estimation

Due to the Doppler eect, errors in the symbol timing will be increased or decreased
proportionally to . To align the symbol within its period, samples should be
removed, if ( > 0), or added, if ( < 0), at regular intervals [77]. For an OFDM
symbol with Nc =1024 sub-carriers and =0.0013, the OFDM symbol drift will be
15.97 samples per OFDM symbol, which is equivalent to a Doppler shift of 15.6
Hz. Therefore, with these samples drift, there is no need to consider the whole CP
window; hence a massive reduction in complexity is obtained.
Accordingly, the redundancy introduced by the guard interval is exploited and
the drift in the received passband samples is measured by correlating the rst
Ng samples with the anticipated observation window denoted by . Let rg =
[r[0]...r[Ng ]] be a vector of Ng received samples, known as guard vector, r be a
vector of the received samples within the observation window and denote the
44

3.2 Doppler compensation techniques

frame synchronization point, as shown in Fig. (3.5), therefore

rD = [ + Nc ( ) + i, + Nc + Ng ( ) + i],
2
2

(3.20)

is the search range of the useful block, where is an even integer and i . In
(3.20), it is apparent that when i = ( 2 ), the OFDM symbol is received within its
period; otherwise the frame length drifts by samples. In the proposed estimator,
the covariance between rg and r is exploited through the observation window to
detect the peaks as
Ng 1

(i) = |

rg (n)r ( + n + Nc + i)|,

(3.21)

n=0

where both rg and r are real samples. It should be noted that the envelope of the
correlation must be smoothed in order to improve detection.
3.2.2.2

Peak localization

Undesired correlation sidelobes are produced due to the inhomogeneities of the signal
fragment window rD with the guard interval. These inhomogeneities occur due to
the correlation of dierent data symbols which are aected by the existence of ISI.
Consequently, the correlation produces uncertainty in the peak location, depending
on the channel conditions. This time position uncertainty in the maximum peak will
pose signicant uctuation in estimating the symbol timing oset. To tackle this
random process, the proposed algorithm adopts a threshold and utilizes a weighted
centroid algorithm [78]. It is assumed that th represents this threshold. Due to the
diculty of determining th analytically, it has been chosen empirically,
th =

max {}
.
2

(3.22)

Thus, all peaks that exceed this threshold are accumulated in a temporary buer wi .
This will enable the positional approximation of the weights within the split-buer
to be determined. Let i be a vector of these time positions; hence the coordinates
of all peaks that attain th can be formulated as
i = arg {(i) > th } , i = 1 . . . n.

45

(3.23)

3.2 Doppler compensation techniques

After the peaks locations have been gathered, along with its corresponding
weight, the goal is to estimate the unknown position P (i , yi ) of the maximum
Doppler shift which is equivalent to the time expansion/compression in the OFDM
frame. Let yi denote the weights wi (amplitude) reserved in a split-buer which corresponds to each location; it follows that the time position of the maximum Doppler
shift can be approximated as
n
wi i
P (, w) = i=1
,
n
i=1 wi

(3.24)

and this correlational behaviour is called localization [79]. It should be stressed that
the estimation error of P (, y) results in timing misalignment and, consequently, it
degrades the FFT demodulation.
In order to estimate the time scaling factor, it is necessary to estimate the timing
which can be derived based on (3.24) as
oset of the OFDM block ,

= ( ) P (, w).
2

(3.25)

This timing metric is estimated independently on a symbol-by-symbol basis. In the

case of no gradient, a unique Doppler shift estimation for each symbol, based on
the estimated timing oset, is a feasible solution. However, in a worse case such as
velocity acceleration, this approach does not hold because the timing oset changes
over time. The next section will present this situation.
The Doppler shift manifests itself as a complete time expansion/compression,
therefore it can be estimated using (3.25)
=

Lf
Lf

(3.26)

where the transmitted frame length Lf is known. For the sake of simplicity, only
the sign (+) will be considered in this chapter. The parameter in (3.26) represents
the Doppler shift based on a one shot estimation that can be divided into an integer
part in order to re-sample the received signal and a fractional part for CFO.

46

3.2 Doppler compensation techniques

i2

i1

i (m)
i

OFDM Symbols

Figure 3.6: A set of OFDM symbols showing the closed and far neighbour to symbol
i. At time n, symbol i estimates i (n)

3.2.3

Adaptive Doppler compensation

So far, the timing metric and its associated Doppler shift, have been estimated on a
symbol-by-symbol basis and assumed to be independent; thus it was called one-shot
algorithm. An alternative adaptive system is suggested in order to estimate these
parameters in accordance with their preceding neighbours values. This approach
aims to design a basis that can accommodate a more realistic situation than oneshot algorithm by considering the speed change between OFDM symbols. As in the
preceding scheme, the coarse Doppler shift is estimated using (3.26). On the other
hand, by utilizing the concept of the nearest neighbour rule [80], this Doppler shift is
further performed to deal with a slow Doppler variation during the OFDM symbol.
To formulate the adaptation step, the following assumption is adopted:
Assumption 1 : Due to the Doppler shift, the OFDM symbol could be expanded
towards the far edge or compressed in the opposite direction. That means estimating
the average Doppler shift (i.e., at the middle) as in the block length-based approach
is not hold.
In both cases, the edge between the current symbol and its nearest neighbour
should be smoothed to mitigate both the channel and the noise eects. After taking
into consideration a set of N OFDM symbols, the adaptation equation for the symbol
timing is

(n)
=

p1

p),
W (n p)(n

(3.27)

p=0

where W (n p) is a weighting coecients vector of p N symbols. The rst

symbol, which is estimated in (3.25), excludes this equation as it contributed to the
1)=0, (n
2)=0, and for each
initialization process. Thus, start initializing (n
OFDM symbol repeat
2) = (n
1); (n
1) = (n).

(n
47

(3.28)

3.2 Doppler compensation techniques

3.2.3.1

Weighting coecients

Although there are many possible choices of weights W in the literature [81], the
weighting coecients in this approach have been chosen in accordance with the
concept of nearest neighbour rule [80] and the premised acceleration. To be more
specic, the following assumption should be taken into consideration:
Assumption 2 : If the OFDM symbol time Td is 256 ms, then it needs approximately 4Td to accelerate the speed to (1 m/s), providing the initial speed is zero
and the acceleration is (1 m /s2 ).
From this assumption, we can infer that the maximum speed in each OFDM
symbol is approximately 0.25 m/s, i.e., within the symbol time. This leads to the
use of the concept of nearest neighbour rule [80], where the cardinality [81] is proportional to how close the neighbour symbol is to the current one. This is practically
true and therefore assigning the nearest symbol (n 1) higher weight, means that
its reliability is high on the assumption that the Doppler shift is variable from symbol to symbol, while the weight decreases with time. All of these assumptions are
made because the change of the Doppler shift, or the symbol timing oset between
previous OFDM symbols, will contribute to predicting the subsequent values; hence,
resorting to involve preceding symbols in order to reinforce the estimation accuracy
of the current parameters. Consequently, each OFDM symbol can be assigned a
weight; however, this is inexpedient due to the convergence speed. Alternatively,
a dedicated group of symbols from the transmitted packet is considered as shown
in Fig. (3.6). This group consists of only the information related to the two previous symbols timing metric with their weights to be involved in estimating the new
timing metric. This weighting vector of tripartite coecients should satisfy

0 < Wi < 1,

Wi = 1 i n, W (n p) = 0 if p
/ N,

(3.29)

i=1

where, p represents the number of symbols in the group. Therefore, smoothing in

(3.27) by using the coecients in (3.29) contributes to improvements in the Doppler
shift estimation. It is worth mentioning that these weighting coecients can also
be utilized to smooth the Doppler shift and the following sections will discuss this
case.

48

3.2 Doppler compensation techniques

3.2.3.2

Fractional CFO Estimation

Thus far, only the time scaling factor has been estimated. Based on this factor,
the coarse estimation of the Doppler frequency shift, as shown in Fig. (3.4), is then
approximated as
Fd

(coa)

c.
= (1 )f

(3.30)

This parameter represents the integer part of the Doppler shift obtained by the
adaptive algorithm as shown in Fig. (3.4)(a), where it is assigned a dashed square.
In order to only re-sample the integer part, this coarse estimate is quantized. Let
. denote rounding toward the lower integer, then Fd

(quant)

new = fc Fd (quant) .

fc

= Fd

(coa)

+ 0.5, and

(3.31)

An ecient sample-by-sample linear interpolation method is used in the receiver to

new . r can be expressed
re-sample the OFDM block with a re-sampling factor
k
mathematically as
r (k) = (n 1) rm +1 + n rm ,

(3.32)

where m {1, 3, 5...}, k {1, 2, 3...} and n = 1 for n = 1.

The subsequent stage is to compute the residual Doppler shift based on the
coarse frequency estimation.
= Fd

(coa)

Fd

(quant) .

(3.33)

After re-sampling and CFO compensation, the channel estimation was implemented
using the least square (LS) method.
It is obvious from (3.30) and (3.33) that the carrier frequency contributes in the
estimation of the Doppler shift and then in the estimation of its residual. However, involving the carrier frequency in estimating such parameters for a channel
of a broadband nature results in an inaccurate approximation of these parameters;
therefore, an ICI is produced and, consequently, it is necessary to resort to exploiting the sample time expansion/compression in order to increase the accuracy and
ultimately improve performance.

49

3.2 Doppler compensation techniques

3.2.4

Proposed Doppler shift compensation

Unlike preceding schemes, the proposed technique derives the Doppler shift based
on a sampling frequency estimate.
As the Doppler shift is evidenced by a frame time expansion/compression [8], it
can be inferred that the rate of sampling frequency will be changed. Accordingly,
joining the samples drift, in estimating the Doppler shift and CFO, is now feasible.
To accomplish this, let be the sampling frequency oset of one sample drift caused
by an expansion, which can be formulated as

= fs (

Lf + 1
) fs ,
Lf

(3.34)

is given by
therefore, the relative sampling frequency oset

= (1 )fs ,

(3.35)

where the coarse Doppler shift is approximated as in (3.26). Clearly, a = 1 causes

a sampling frequency error; hence a drift in the symbol timing. Therefore, the
relative oset in (3.35) represents the samples drift which causes the timing error.
To perform Doppler shift compensation, the samples drift is exploited and divided
(I) is used to estimate
into an integer part and a fractional part. The integer part,
a new interpolation factor and is given by

= Lf (I) ,

Lf

(3.36)

(I) =
is rounded toward the lower integer.
where
(F) is exploited to estimate the CFO,
At the same time, the fractional part
where the fractional deviation of the samples drift is approximated as
(F) = (

(I) ),

(3.37)

It can be noticed that the main factor which destroys the orthogonality is the
fractional drift of the sub-carrier spacing f . This is based on the misalignment of
the symbol which degrades the FFT demodulation and consequently an inter-carrier
interference (ICI) will result.
Hence, estimating the fractional drift in (3.37) is crucial to the approximation of
50

3.2 Doppler compensation techniques

the CFO, where
(F) ( fc ),
=
fs

(3.38)

is the residual Doppler shift. Subsequently, compensating for in (3.16) after resampling, we obtain
r(t) = r(t)ej2 ,

(3.39)

where in this case, the orthogonality is preserved. It should be stressed that this is
an approximation of the ICI free received signal.
3.2.4.1

Doppler shift variation adjustment

Due to the wideband nature of the UA channels, each sub-carrier will be shifted
non-uniformly [12]. Furthermore, if the relative velocity between the innermost and
the outermost edge of the symbol were not constant (i.e., with acceleration) over
the symbol duration, then an error in Doppler estimation will result and will need
to be considered. Hence, adjusting this velocity perturbation necessitates frequent
estimations of the re-sampling factor or reduction of the symbol length. However,
in OFDM signal design there is a trade-o between the number of sub-carriers,
carrier frequency, scaling factor resolution and complexity. In such cases, reducing
the symbol length does not only cause a reduction in the bandwidth eciency, but
also mitigates the immunity against the ISI. In this method, there is a compromise
between these system specications. These circumstances of speed variations are
dealt with by employing weighting coecients to smooth the edges between symbols.
These coecients are chosen based on the principle of the nearest neighbour rule,
discussed earlier. Consider a set of N OFDM symbols; the adaptation equation for
the Doppler shift is

(n)
=

p1

p),
W (n p)(n

(3.40)

p=0

where W (n p) is a weighting coecients vector of p N symbols. The Doppler

shift of the rst symbol, which is estimated in (3.36), excludes this equation as it
1)=0, (n

contributes to the initialization process. Thus, start initializing (n
2)=0, and for each OFDM symbol time n repeat Algorithm 1.

The receiver then

re-samples the OFDM symbol with a re-sampling factor obtained after smoothing.
51

3.2 Doppler compensation techniques

input : Set weighting coecients W1 , W2 , W3 such that

Wi = 1
n
output: A smoothed Doppler shift
// <Temp is a temporary buffer>
T emp
n
if F lag > 1 then // <Flag represents the symbol index>
W1 +

n
n
n 1 W2 + n 2 W3
else
if F lag = 1 then

W1 +
W2

n
n
n1
end
end

n 2 n 1 ; n 1 T emp
Algorithm 1: Smothing algorithm
It is worth pointing out
Algorithm 1 is also applied to smooth the timing oset .
that the weighting coecients W1 , W2 and W3 are empirically obtained.
3.2.4.2

Fine timing estimation

There is a noticeable degradation in the FFT demodulation due to the fractional

part which accompanies in (3.9). Accordingly, this part is considered for the
[0]
purpose of updating the synchronization point . Therefore, starting with m =0,

and then for each OFDM symbol m in the packet, repeat

m = m + m ,

(3.41a)

m+1 = m + m + Lf ,

(3.41b)

m = m m ,

(3.41c)

where the operator . denotes truncation to the nearest integer.

3.2.4.3

Channel estimation and decoding

After re-sampling and CFO compensation, the channel estimation is implemented

using the least square (LS) method.
p (n) = D[Xp (n)]1 Yp (n), n = 0 . . . Np 1,
H

(3.42)

p (n) are the estimated pilot channel values, D[Xp (n)] is a diagonal matrix
where H
constructed using the known transmitted pilot symbols, and Yp (n) are the received

52

3.3 Simulation Results

Table 3.1: Correlation complexity estimates.
Operation Proposed
Add
8
Multiply
Ng

pilot symbols after the FFT operation. After removing the channel eect, the signal
is then passed through the soft de-mapper to produce the extrinsic estimates to be
deinterleaved and then applied to the BCJR algorithm in order to decode convolutional codes. The output of the BCJR in the rst iteration is fed to the cyclicredundancy-check (CRC). Accordingly, the symbols with errors can be corrected
by re-encoding the detected information from the rst iteration. This procedure is
called BICM-ID.
3.2.4.4

Complexity analysis

It is now appropriate to consider the computational complexity of (3.20) and (3.21)

for the proposed method in estimating the Doppler shift. In this analysis, a conventional approach is followed where the number of operations, such as addition
and multiplication, are counted as a benchmark for this purpose. It is worth pointing out that the proposed algorithm is implemented in passband, and therefore the
benchmark for real operations only. Furthermore, the proposed method requires no
iteration to estimate the Doppler shift. Table 3.1 shows the complexity estimation
of the proposed technique.

3.3

Simulation Results

The performance of the proposed system was tested over a multipath channel impulse response, h(n) = 0.6708(n) + 0.5(n 1) + 0.3873(n 2) + 0.3162(n 3) +
0.2236(n 4), and the corresponding delays at time n to n 4 were 0, 2.5, 5, 7.5, 10
ms. In these simulations, transmission was organized in packets of equal duration,
each containing single 50 ms LFM followed by a 12.5 ms silent period, and then
10 CP-OFDM frames as shown in Fig (3.7). A total of 8920 information bits were
transmitted in each setting. The carrier frequency was set to 12 kHz, whereas the
sampling frequency was fs = 4fc . Nc =1024 sub-carriers were used along with

53

3.3 Simulation Results

2.795 s
10 OFDM
Chirp Silent 1 2 3

10 Silent

50 ms 12.5 ms

12.5 ms
Copy

CP

OFDM

16 ms

256 ms

(16 + 1024)

1
B

= 272 ms

Figure 3.7: Packet structure for Nc = 1024.

bandwidth B =4 kHz, which led to a sub-carrier spacing of 3.90625 Hz. The guard
interval was set as Tg =16 ms. A rate 1/2 NSC code and interleaver was adopted
in this simulation to map 892 data bits to 1792 interleaved bits. The achieved data
rate [82] was 3.2794 kb/s
R=

Rc Nd log2 M
,
Tg + Td

(3.43)

and the bandwidth utilization factor was 0.8198 bits/sec/Hz for the QPSK modulation scheme.
=

R
B

bits/sec/Hz.

(3.44)

Fig. (3.8) shows the performance comparison of the CP-based Doppler shift
compensation between one-shot and the algorithm in [1]. The channel frequency
and phase responses are depicted in Fig. (3.8)(a) and used for both algorithms
to unify the comparison. It can be shown that the centroid-based normalization of
the CP-based correlation in estimating the Doppler shift outperforms the estimation
algorithm in [1]. In addition, due to the computational unlimited search on the angle
of the correlation, it can be inferred that the proposed scheme reduces complexity
and is more pragmatic than [1]. Fig. (3.9) shows the CP-correlation output in
the proposed scheme and its smoothing to improve the detection of the maximum
peaks. However, these correlation peaks are aected by the ISI and the Doppler
shift variation between symbols.
A comparison between the proposed system and the block Doppler technique

54

10

10

0
10
20

0.2
0.4
0.6
0.8
1
Normalized Frequency ( rad/sample)

BER

Phase (degrees)

Magnitude (dB)

3.3 Simulation Results

10

50
0
50

100

[Kim], 4 Hz at f =10 kHz

c

Proposed, 4 Hz at f =10 kHz

0.2
0.4
0.6
0.8
1
Normalized Frequency ( rad/sample)

10

10

15

SNR dB

(a) CIR.

(b) BER

Figure 3.8: Performance comparison between one-shot algorithm and the algorithm
proposed by Kim in [1].

0.03

Magnitude

0.03
0.02
0.02

0.01

0.01

50
Correlation lag

100

(a) Before smoothing.

50

100

(b) After smoothing

Figure 3.9: Anticipated correlation window before and after smoothing for packet
1, symbol 3 at speed -0.25 m/s from the experiment.
was made using variable and xed speeds during the packet time. To investigate
each OFDM symbol, an array of speeds was set to equal [1 1 1 2 2 2 0.5 0 1 1] and
[1 1 1 1 1 1 1 1 1 1] m/s for both variable and xed speeds, respectively.
The BERs of the simulation are plotted in Fig.3.10, which indicates that the block
Doppler compensation technique outperforms adaptive approach by about 2 dB in
a xed speed and moderate SNR. This is because the length of time left between
two LFMs increases the resolution of the average scaling factor estimate; hence,

55

3.4 Experimental Results

0

10

Bit Error Rate (BER)

10

10

10

AdaptiveFixed
AdaptiveVariable
BlockFixed
BlockVariable
No Doppler

10

8
10
EbNo dB

12

14

16

Figure 3.10: Performance comparison between the adaptive scheme and block
Doppler compensation for xed and variable speeds.
the Doppler shift estimate at the mid-point is approximately equal to the actual
speed. Furthermore, since the bandwidth is comparable to the carrier frequency in
a broadband Doppler shift, it follows that employing this frequency in estimating the
CFO cannot be accounted by the receiver. On the other hand, the block technique
fails to track the Doppler variation from symbol to symbol because the average
estimate is no longer capable of tracking the variation between each OFDM symbol.

3.4
3.4.1

Experimental Results
Experiment setup

During the summer of 2009, an experiment was conducted in the North Sea to
evaluate the system performance. The trial setup is illustrated in Fig. (3.11). The
transmitter and receiver were set at 10 and 5 m from the sea surface, respectively.
The transmitter power was set to 180 dB re 1Pa. There was a rapid time varying
multipath channel in that area due to the hard surface of the seabed. In the trial,
transmission was organized in packets of equal duration, each containing one 50 ms
LFM followed by a 12.5 ms silent period, and then 10 CP-OFDM frames. A total
of 8920 information bits were transmitted in each setting. A total of 20 packets of
2.795 s were sent. The carrier frequency was set to 12 kHz, whereas the sampling
frequency was 4fc . 1024 sub-carriers were employed and the system bandwidth was
56

3.4 Experimental Results

Figure 3.11: Conguration of the experiment in the North Sea.

1
0.1

BER

0.01
0.001
0.0001
0
BLK
1

AD
9
13
Packet index

Oneshot
17

20

Figure 3.12: Bit error rate over each packet of 8920 bits.
4 kHz, which led to a sub-carrier spacing of 3.90625 Hz. The guard interval was set
as Tg =16 ms.

3.4.2

Performance evaluation

Fig. (3.12) shows the BERs performance comparisons of the block, one-shot and
adaptive techniques. It is obvious from this gure and claried in Table 3.3 that
the adaptive algorithm which employs the weighting coecients outperforms block
length-based and one-shot methods by 83.6383, 63.9932 %, respectively. At the
57

3.4 Experimental Results

same time, Table 3.3 shows that one-shot algorithm surpasses block length-based
approach by 54.5594 %. It can be seen from Fig. (3.12) that only the packets (12,
18) have high decoding errors and no signicant reduction in the bit error rate was
found with the adaptive technique compared with the block method. This is due to
an error in estimating the time scaling factor due to the noise and the channel which
caused an ambiguity in estimating the Doppler shift at a xed speed. Looking at
Fig. 3.13, it is apparent that the speed of packet (P12) was xed; thus, estimation
of the average Doppler shift during a long packet time is approximately equal to
the actual value, and consequently the block technique outperforms our adaptive
algorithm. Fig. 3.15 presents, however, evidence that there was a variation in the
speed during the packet time; therefore, the adaptive technique surpasses the block
method. This is evidenced by Fig. 3.13, where the time scaling factor of the adaptive
technique in packet 2 (P2-AD) has been estimated for each OFDM symbol; whereas
in the same packet, the block approach (P2-BLK) fails in estimating this variation.
Furthermore, compared with a one-shot approach which processes symbol-by-symbol
independently, it is shown in Fig. 3.13 that there is an improvement in the performance due to the adoption of the weighting coecients that smooth the estimated
parameters. Locking at Fig. 3.14, in the uppermost graph, it is apparent that the
scaling factors for packets (P6, P12) were changing in one direction at what can
be considered semi-xed speeds, thus, estimation of the average speed during long
packet time indicates that the block technique is outperforming our proposed system. Furthermore, it is obvious from this gure, in the lower graph, that the scaling
factors of (P16, P10) have been changed from compression to expansion or vice versa
during the packet time. In packet 5, it is clear that the Doppler shift is very small;
therefore the algorithms have similar performances.
Turning now to the experimental evidence on the performance of the proposed
technique, the technique combines the measuring of the time expansion/compression
of the sample period and utilization of the weighting coecients in estimating and
smoothing the Doppler shift, respectively. To evaluate the performance, a comparison of the proposed algorithm and the block method are depicted in Fig. (3.16)(a).
It is apparent that the suggested method surpasses the block technique in 18 out of
20 packets. In the proposed scheme, it can be seen that only in packets (6, 12) are
the BERs high compared with the block technique. This is a synchronization issue,
where the proposed algorithm is based on the assumption that the Doppler shift
58

3.4 Experimental Results

1.0002

P12AD

P12BLK

1.0001
1
0.9999

10

Scaling factor

1.0001

1
P2AD
0.9999

P2BLK

5
6
7
Block index

10

Figure 3.13: Estimation of the Doppler scaling factor over each block for packets (
2, 12) of the adaptive algorithm.
1.0005
1
P12

Scaling factor

0.9995

P6
4

10

10

4
6
8
Transmitted blocks per packet

10

1.0001
1
P5
0.9999

1.0005
P16

P10

1
0.9995

Figure 3.14: Estimation of the Doppler scaling factor over each block for packets (
5, 6, 10, 16) of one-shot algorithm.
varies during the symbol time and consequently during the packet time; therefore,
this variation degrades the receiver performance if it is not taken into consideration.
Hence, the last symbol in the packet should also be involved in the smoothing algorithm. This case is evidenced in Fig. (3.16)(b), where it is apparent that the error
in the OFDM symbol of index 1 comes from the Doppler variation of the last symbol
in the previous packet. In symbols of indices (7, 8), the case is dierent, where there
is an error in estimating the Doppler shift during the OFDM symbol. As shown in

59

3.5 Real-time implementation of BICM-ID

0.5

Estimated speed (m/s)

0.3
0.1
0.1
0.3
0.5

40

80
120
160
OFDM symbol index

200

Figure 3.15: Changing of speed through the packets time.

Fig. 3.16(c), the speed at the end of OFDM symbol index 7 starts changing its direction and this necessitates considering the slope variation. In addition, it is shown in
Fig. 3.16(c) that the average speed of the packet is approximately constant; hence,
estimating the average Doppler (i.e., at the mid-point) outperforms the proposed
method. The BER results of the proposed method in Fig. (3.16)(a) were obtained
with two iterations, whereas in the block method the iterations were 10. This gure,
in conjunction with the summarized results in Tables 3.2 and Tables 3.3, conrms
the improvements of the proposed technique over counterpart techniques. The improvement ratio was 93 % between the proposed and the block technique, whereas
this ratio was 57.2363 % between the proposed and the adaptive algorithm. The underlying reason for this improvement is due to the broadband nature of the channel
in conjunction with a tight sub-carrier spacing; the receiver becomes very sensitive
to the Doppler shift and, consequently, it is not capable of accounting for a shift in a
carrier frequency. Therefore, adopting the complete sample time of the transmitted
frame to compensate for the integer and the fractional part of the Doppler shift was
the main contribution in this improvement.

3.5

Real-time implementation of BICM-ID

As mentioned earlier, the proposed system adopts iterative decoding at the receiver.
This iterative decoding is computationally expensive and requires long execution
time therefore, it is interested to benchmark the BICM-ID implementation. The
60

3.5 Real-time implementation of BICM-ID

Table 3.2: Performance results of the experiment

Packet
index
Block
one shot
Adaptive
Proposed

10

0
0
0
0

119
0
0
0

423
132
84
3

347
170
91
2

3
9
9
0

44
344
14
32

11
0
0
0

505
6
9
4

39
31
21
22

2443
90
245
210

Packet
index
Block
one shot
Adaptive
proposed

11

12

13

14

15

16

17

18

19

20

24
18
6
3

0
802
384
141

33
169
10
9

178
77
49
3

21
53
45
0

1702
383
6
11

21
267
36
3

0
176
20
0

471
123
30
7

119
105
5
5

Table 3.3: Average BER and error statistics comparison of the experimental results
for dierent Doppler shift compensation techniques
Method
Block
One-shot
Adaptive
Proposed

Error statistics
Errors/178400
Average BER
6503
0.0365
2955
0.0165
1064
0.0059
455
0.0025

61

Improvement ratio

54.5594 %
63.9932, 83.6383 %
57.2368, 93 %

3.5 Real-time implementation of BICM-ID

100

0.1

80
Number of errors

BER

0.01
0.001
0.0001

60
40
20

0
BLK

Proposed

9
13
Packet index

17

20

(a) BER performance.

4
6
8
OFDM symbol index

10

(b) Error statistics of packet 12.

0.3
0.2

Speed (m/s)

0.1
0
0.1
0.2
0.3
0.4

4
6
8
OFDM symbol index

10

(c) Estimated speed of packet 12.

Figure 3.16: Performance of the proposed system from the experiment.

general system compromises a single DSP in transmitter (Tx) and receiver (Rx) as
shown in Fig. 3.17. Data streams of length bi = 2047 bits are generated to form
the input of the encoder. It is composed of FEC with convolution NSC, which has
a code rate of 1/2 and K = 5, then passes to an interleaver. In this interleaver,
Lint = 4120 bits is used to permute the encoder output and consequently randomize
error. The digital modulation technique is QPSK mapped to Gray mapping. The
encoded data are transmitted with the carrier frequency of f c = 10 kHz and the
symbol rate of 4 ksps. On the receiver side, ADC data, in 24-bit unsigned integer format, must rst be converted to oating point representation. Additionally,
signal amplitude should be scaled from the ADC values to a normalized +/- 1.0
range for the subsequent signal processing stages. After bandpass FIR lter and
62

3.6 Hardware platform description

Rx
SHARC Data
DSP

Tx
Data SHARC
DSP

Figure 3.17: General System Specication.

'W>

:d',

<

D

^ZD

&




^h


W
'
d:



WW

^W

d

^W
K

ZW


^W/




s

&>'

/

^W/&/

ss



^W/&K

K

WZ

^W/

W

^/

^K

,

&>^,

Zy

Zy

:

Figure 3.18: SHARC ADSP-21364 system architecture block diagram.

frame synchronization, soft demodulation is required. These soft information are
de-interleaved and then decoded with a BICM-ID.

3.6

Hardware platform description

Fig. 3.18 shows the platform ADSP-21364-EZLITE Kit SHARC family from Analog
Devices. In this section, we will describe some features that have been actually
used in the implementation, the readers are referred to [83] for more details information. ADSP-21364 SHARC is a 32/40-bit oating point processor optimized for
high performance automotive audio applications with large on chip SRAM (3M bit)
and ROM (4M bit), multiple internal buses to eliminate input-output (I/O) bottlenecks, and an innovative digital audio interface (DAI). This interface is crucial for
the processor to communicate with the DAC/ADC or sometimes called CODEC.
One of the key components of the acoustic modem is the audio signal input/output
module. The ADSP-21364 development board used has a built-in module for sampling audio signal. The task is handled by the integrated Analog Device AD183x
CODEC family [83]. Data transfer word lengths of 16, 20, 24, and 32 bits, with sampling rates from 8 kHz to 96 kHz, are supported. The operation mode of AD183x
63

3.6 Hardware platform description

can be programmed with a set of control registers. For sampling rate of 48 kHz, if
the processors buer holds 1024 samples, then it has a frame acquisition interval
of 21.33 ms (i.e., 1024 20.833s). Here the DSP has 21.33 ms to complete all the
required processing tasks for that frame of data. Three buers of size 1024 have
been used to exchange these samples between CODEC, DMA and serial ports as
shown in Table 2.1, so data sampling and processing can be done simultaneously
and no incoming signals are missed even if the DSP is processing previously received
data.
Serial peripheral interface (SPI) is an industry standard synchronous serial data
link named by Motorola. It provides full-duplex synchronous serial interface to communicate with DAI and the core processor. This processor achieves an instruction
cycle time of 3.0 ns at 333 MHz.

3.6.1

Processing time optimization

In a simulation environment, we can process a whole communication packet in one

simulation time instance regardless of its size. In addition, the time synchronization
between transmitter and receiver can be assumed to be perfect. However, in a realtime system, we have to process the communication signals in frames rather than
packets due to the limitation of the internal memory of DSP systems and stringent
real-time constraints. Frames of length Nf = 2060 of coded and modulated symbols
are utilized. The transmitter sends a block of 1024 samples at each interrupt time to
a DAC of sampling rate 48 kHz. As the processor is running at 333 MHz, then the
number of clock cycles the processor can perform before an interrupt are 7104000
cycles for a block of 1024 samples. On the receiver side, soft de-mapping, log-MAP
decoder, and de-interleaver are complex algorithms, particularly with iterations,
therefore, they require more processing time to be run by the DSP than the frame
duration allows, which is 515 ms. These algorithms should convene the DSP realtime requirements. This constraint can be relaxed in twofold. Firstly, by setting the
silent period time between two consecutive frames equal to the decoding time and
greater than frame duration; however this reduces bandwidth. Alternatively, design
a silent period of minimum length according to the required time to empty the
bandpass lter taps and introducing one frame duration delay (rst frame) during
reception. In this mode, the DSP collects the received samples in data memory

64

3.6 Hardware platform description

TK1 TK2 TK3 TK4 TK5 TK6

Init. LLR 1 0

Figure 3.19: Receiver tasks.

instead of processing them immediately. Each symbol time in the frame duration
was exploited to manipulate a specic task of previous frame duration as shown in
Fig. (3.19).
That is, optimization of the processing time is dealt with based on the disposition
of the frame time according to the processing time of each stage in the outer receiver.
Moreover, this structure adopts pipelining the analysis and processing phases of
incoming frames and relies on the internal memory only. In this gure, the time
of task1 (TK1) is set in accordance with the measured decoding cycles in the DSP
board. Particularly, to infer how many symbol times need to be set for TK1, we need
to know the maximum cycles of each task. Thus, it is apparent that the maximum
cycles spent in TK5 are 28250768, in order to process Nf symbols from start to end
through all 16 states of BCJR. The details of this decoder, which is denoted relative
to its authors, are mentioned in [68].
Based on the time of each task in the decoder, we can nd a relation for each
repetition as shown below. This relationship includes all stages of the data symbols
during the SISO decoding. We can also note that, in (3.45), the number of iterations
I, has an inuential role in determining the number of necessary symbols that can
be exploited at every stage. Except the rst phase, I has no eect because this
phase is used for the purpose of array initialization,
Step[I] = (686, 20I, 158I, (8I) 1, 2I, 158I),

(3.45)

where I=1,. . . , 3. In the second phase (soft de-mapper), a period of twenty-symbols

has been exploited in the iteration to be done. We can infer that, a period of forty
and sixty symbols has been exploited in the second and third iterations, respectively,
to accomplish it. In our system, the time required for 20 symbols is 5 ms for the
case of I = 1, therefore in each symbol time of the current incoming symbol, we

65

3.6 Hardware platform description

process 103 symbols of the previous frame in the second stage.
Np (i) =

Nf
,
Step(i)

(3.46)

where i=2,. . . ,6, and Np is the number of symbols per stage.

By applying (3.46) to all stages of the decoder, we can nd the maximum symbols, in the iteration from the previous frame as below:

NI = Ni + I

Np (i),

(3.47)

i=2

where NI , is the maximum symbol length in the iteration after processing time
optimization. For Ni =4, the maximum symbol length in iterations 1, 2 and 3 are
1423, 1424 and 1429 symbols, respectively.
From (3.46), we can nd also that, in TK5 or Step(5), the maximum symbol
length is 1030 symbols. This result leads to determining the maximum cycles which
can be stolen from each symbol in one iteration, as below:

M AXc
.
N s
28250768
=
= 27428 cycles,
1030

M AXS =

(3.48)
(3.49)

where M AXS , M AXC , N s is the maximum stealing for each symbol, the maximum cycles and the maximum symbol length in stage TK5, respectively.
For the purpose of determining the stealing ratio, the total cycles were measured.
By applying the relation below, we can calculate how many cycles have been stolen
by each iteration:

Ctot
NT
38679420
=
= 27181 stolen cycles,
1423

CStol =

(3.50)
(3.51)

where CStol , Ctot and NT are the cycle stolen, the total cycles and the total symbols
in iteration one, respectively. Therefore, the stealing ratio of the iteration is:

66

3.6 Hardware platform description

CStol
,
SC
27181
=
= 32.65%,
83250

SR =

(3.52)
(3.53)

where SC is the symbol cycles.

Table 3.4: Total Cycles
Iteration 1 Iteration 2 Iteration 3
3,8679,420

3.6.2

7,7238,370

11,5857,252

Memory allocation

The 3Mbit onchip static random access memory (SRAM) of the DSP is split into 4
blocks, of dierent sizes. For 32 bit words, these blocks are allocated as the following:
32K (0x8000) of data memory (DM) space in memory block1, 16K (0x4000) of
program memory (PM) space in memory block2, 8K (0x2000) of heap space in
memory block3, 8K (0x2000) of stack space in memory block4. In this system, the
challenge is to process Nf symbols in the stage. All symbols, from start to end in
this stage should be processed through all 16 states. Therefore, a size of more than
32k of DM should be available. To tackle insucient memory space, stage memory
has been buered into both DM and heap. In addition, overwriting technique has
been used for the sake of memory optimization, especially to buer the whole frame
in the case of block interleaver and deinterleaver.

3.6.3

Real-time experimental results

The performance of the proposed BICM-ID receiver is investigated real-time in tank

with 3-iterations only as shown in Fig.3.20. This scatter diagram is used to present
the experimental output of the iterative decoding. A evidence can be inferred from
this gure that the system met the real time requirements and the iterative receiver
suited the underwater channel. What is interesting in these requirements is that,
the RAM has been determined (including interleaver), which is 21Nf . Comparisons
between these memory requirements and [84] results our system outperform their
system in one block memory requirement, where they used 22Nf .
67

3.7 Chapter summary

Scatter plot
2
1.5

Quadrature

1
0.5
0
0.5
1
1.5
2
2

0
InPhase

Figure 3.20: Output of real-time BICM-ID 3 Iterations.

3.7

Chapter summary

The following key points have been discussed in this chapter:

Three low-complexity Doppler compensation techniques are presented.
Comparison of the CP-correlation is demonstrated by simulation.
Re-sample with unique Doppler shift in each OFDM, but variable between
symbols within the packet is adopted in all suggested algorithms.
Real-time implementation requirements of the most computionally extensive
part of the proposed BCJR decoder has been presented with SHARC ADSP21364 processor.
In one-shot technique, it has been shown it cannot account for the Doppler
shift and CFO based on the carrier frequency due to the wideband nature of
the channel.
It has been proven through the experimental results that the weighting coefcients improve the Doppler shift estimation and accommodate the change in
speed between the OFDM symbols.
A proposed system which combines the using of the weighting coecients and
adopting the sample time expansion/compression is devised among all previous
schemes to establish a basis for other techniques.
The conclusion is the Doppler shift cannot be considered constant and it needs
an algorithm that is capable of tracking the speed variation within the OFDM
symbol time. The next chapter addresses the frequency estimation.
68

Chapter 4
Time varying Doppler shift
compensation
Traditional techniques employed in order to compensate for the Doppler shift in
conventional receivers are based on the assumption that there is a common Doppler
shift during the OFDM symbol time. In particular cases, such as acceleration, it is
dicult to ignore the time-varying Doppler scale during the packet time, a factor
which necessitates the use of a tracking algorithm to enable frequent estimation
of this multi-scale parameter. Therefore, this chapter aims to design a receiver
structure that is capable of accomplishing such time-varying Doppler compensation.
In this chapter, two approaches are taken into consideration in order to estimate
the symbol timing oset parameter. The rst method employed to achieve an estimate of this particular parameter is based upon centroid localization mentioned
in chapter 3 and this prediction is reinforced by a second technique which utilises
linear prediction, based on the assumption that the speed changes linearly during
the OFDM symbol time. Subsequently, the two estimations of the symbol timing
oset parameter are smoothed in order to obtain a ne tuned approximation of the
Doppler scale. Additionally, the eects of weighting coecients discussed in chapter
3 on smoothing the Doppler scale and on the performance of the receiver are also
investigated. The proposed receiver is investigated, incorporating an improvement
that includes ne tuning of the coarse timing synchronization in order to accommodate the time-varying Doppler. Based on this ne-tuned timing synchronization,
an extension to the improved receiver is presented to assess the performance of two
point correlations. The proposed algorithms performances were investigated using

69

4.1 Time varying Doppler shift model

real data obtained from an experiment that took place in the North Sea in 2009.

4.1

Time varying Doppler shift model

Based on the assumption that the speed of the motion changes linearly during the
ith OFDM symbol interval t [iT, T (i + 1)), the Doppler shift is varied with time,
therefore the constant does not hold to accommodate this variation and it should
be replaced by (t). Thus, the time varying Doppler shift can be modelled as

(t) =

v(t)
,
c

(4.1)

where v(t) represents the speed variation during the symbol time. Therefore, the
received passband signal in (3.16) can be rewritten as
{

1
j2fn (1+(t))t
di (n)uopt
i (n)e
Nc nI
}
L1

hl prc [(1 + (t))t l ]ej2fn l + wi (t),

r(t) =

(4.2)

l=0

and its corresponding complex baseband signal model in (3.17) can be written as

r(t) =

j2nf t j2(t)fn t
Hi (n)di (n)uopt
e
+ wi (t),
i (n)e

(4.3)

i=0 nI

where Hi (n) is the channel transfer function of the ith symbol at nth sub-carrier
with a time varying Doppler shift that can be written as

Hi (n) =

L1

hl ej2fn l prc [1 + (t)t l ] .

(4.4)

l=0

As referred to in [85], it is obvious in (4.3) that the eect of the Doppler shift
on the received signal is twofold. First, it scales the received OFDM frame duration
T by a factor of 1 + (t), yielding sampling frequency errors that result in symbol
timing error [86]. Second, there is a time varying CFO.

70

4.1 Time varying Doppler shift model

4.1.1

Sampling frequency errors

In discrete time, the sampled transmitted signal x[kTs ] in (3.13) is equivalent to a

scaling of the sampling period (interpolation or decimation)
r[kTs ] = x[k(1 (t))Ts ],

(4.5)

where k is an integer, and Ts and r(kTs ) are the sampling period and Doppler-shifted
received sampled signals respectively. The bidirectional eect of the Doppler shift
causes symbol timing errors, which will be increased or decreased proportionally to
(t). To align the symbol within its period, samples should be removed if ( > 0)
or added if ( < 0) at regular intervals [77]. Let be the deviation of samples
of the received sequence for each OFDM symbol due to the speed change. The
sampling period results in expansion or compression of the samples length, hence
the Doppler-shifted received frames length is modelled by
Lf = (Lf ),
where Lf =

Nc
BTs

(4.6)

represents the transmitted passband samples length. It is apparent

that Lf is only aected by Ts and any expansion/compression in the timescale will

result in . Therefore, (4.6) is implicitly equivalent to (4.5). To remove both CFO
and symbol shift, an inverse time scaling of the received (compressed/expanded)
signal should be achieved providing that the amount of Doppler shift (t) is known.
This is equivalent to changing the sampling rate of the passband signal by 1 + (t)
in discrete-time processing. From (4.6), we can infer that increasing or decreasing
the length of samples is equivalent to adjusting the sampling frequency fs by the
same Doppler shift 1 + (t); thus (4.5) is rewritten as
r[k] = x[

k(1 (t))
],
fs

(4.7)

where fs = fs (1 (t)). By substituting fs in (4.7), r[k] = x[k], i.e. the signal

received is then in conformity with the transmitted signal.

71

4.2 Signal processing in the proposed receiver

ui
r(k)

Resample

e j2(fc +)kTs
bi
(k)

(k)
r[k(1 + (k))]

r(k)

CP
Correlation

Doppler
Extraction

-CP

BICM-ID

(a)

FFT

ZFE
Soft
Demaper

(b)

Figure 4.1: Proposed receiver structure

4.1.2

Carrier frequency oset errors

The factor ej2(t)fn t in the received signal in (4.3) represents a time varying CFO,
where (t)fn = (t)fc +(t)nf . The CFO () is due to the residual Doppler shift.
It is destructive because it deviates the sub-carrier spacing f and introduces ICI,
which must be removed prior to the FFT to design an optimum receiver [86]. The
re-sampling process removes the Doppler shift and converts the wideband system
into narrowband. However, the residual Doppler shift produced by the fractional
part of the time expansion/compression degrades the receiver.

4.2

Signal processing in the proposed receiver

To utilize the available bandwidth eciently, the algorithm employs a low-complexity

blind technique to estimate the Doppler shift based on estimating the coarse timing
metric for each OFDM symbol by exploiting the inherent periodicity of the CP.
Centroid-based localization has been used to rene the maximum amplitude of the
timing metric; i.e. the timing oset, as explained in chapter (3). Using this coarse
timing metric, the Doppler shift and its residual are frequently estimated by deriving
a tracking step in the Doppler extraction unit (DEU). This unit comprises linear
expectation of the timing oset, ne tuning of the estimated parameters, tracking
the Doppler shift, and CFO estimation. In this technique, the fractional deviation
of the sub-carrier spacing, which is the source of ICI, is estimated by exploiting the
fractional part of the normalized sampling frequency oset; whereas the integer part
of this oset is used to estimate the integer Doppler shift.

72

4.2 Signal processing in the proposed receiver

4.2.1

Coarse timing metric estimation

The receiver structure of the proposed system is depicted in Fig. 4.1. The received
signal r(t) in (4.2) is fed through the transducer, pre-amplier and analogue-todigital converter, and then ltered in the frequency band [fc B/2, fc + B/2].
The resultant Doppler shifted passband signal r[k(1 + (k))] is correlated with the
Doppler tolerant-training (chirp) to detect the start of the packet that contains
several OFDM symbols. Based on the existing guard interval, the drift in the received Doppler-shifted signal r[k(1 + (k))] is measured by correlating the guard
samples (Ng Ns ) with an anticipated observation window in order to estimate the
coarse timing metric for each OFDM symbol within the packet, as in chapter 3. In
the case of time varying Doppler shift, i.e. multi time scaling factor, the resulting timing metric is aected by the velocity perturbation. Consequently, there is
a demand on estimating this timing metric of the same OFDM symbol, but using
an alternative approach to increase the accuracy of the Doppler shift estimation.
Therefore, in Fig. 4.1(a), linear prediction is adopted to extract the Doppler shift
for the purpose of reinforcing the symbol timing oset parameter that was estimated
using CP correlation.

4.2.2

Time varying Doppler shift estimation

Thus far, the timing metric has only been considered for the case of a common
Doppler shift during the OFDM symbol time. A worst case scenario may occur
when there is a velocity that accelerates or de-accelerates within the symbol period.
This situation can be explained in Fig. (4.2). This gure shows that the start of
the OFDM symbol undergoes a dierent speed relative to the speed at the end of
the symbol due to the acceleration, in which the speed is changing linearly with
time. As a result, a linear multi Doppler shift during the OFDM symbol period is
produced. In addition, the acceleration is a useful indication of how fast the change
is, where in Fig. (4.2)(a) the Doppler frequency shift is 1.12 Hz at OFDM symbol
1 and it increases to 11.2 at OFDM symbol 10. The same case is demonstrated in
Fig. (4.2)(b), where the acceleration is 1 m/s2 and the Doppler frequency at OFDM
symbol 10 is 22.4 Hz, in terms of time-selectivity measurement which is given as:
Td Fd > 1,
73

(4.8)

4.2 Signal processing in the proposed receiver

V =0.14 m/s

OF DM #1

Fd =11.2 Hz

OF DM #2

V =0.28 m/s

OF DM #10
(a)

OF DM #1 OF DM #2

Fd =22.4 Hz

OF DM #10
(b)

Figure 4.2: Acceleration eect over Doppler frequency change during each symbol
time at fc =12 kHz. (a) a=0.5 m/s2 , (b) a=1 m/s2
This rapid change within the symbol duration gives an indicator of the amount of
distortion caused by the channel on the signal.
Alternatively, frequent estimation of the Doppler shift within the OFDM symbol
or reducing the frame length are viable solutions. However, in OFDM signal design,
there is a trade-o between the number of sub-carriers, Doppler estimation resolution
and sensitivity to the CFO. Hence, frequent estimation of the interpolation factor
is more feasible than shortening the OFDM symbol length.
When the channel has a velocity that accelerates or de-accelerates in both directions (up or down) within the symbol period, the following assumption is considered:
Assumption : If Td is 256 ms and the maximum acceleration 1 m/s2 starting from
initial speed v0 , then the symbol needs approximately 4Td to attain the maximum
speed v0 + 1 m/s. From this assumption, it can be inferred that the maximum speed
change in each OFDM symbol is approximately 0.25 m/s.
For a system of 12 kHz carrier frequency, 48 kHz sampling frequency and a
symbol time of 0.256 seconds, such speed variation causes a Doppler frequency shift
Fd to increase by 2 Hz within each symbol up to 20 Hz by symbol number 10. In
such circumstances, estimating a common timing metric may not hold to attain
acceptable performance. Alternatively, a better solution and more accurate Doppler
compensation can be realized by adopting a frequent estimation of the Doppler shift
within the OFDM symbol.

74

4.3 Doppler extraction

E [i ]

i1
i2

i Symbol index

i1

Figure 4.3: Estimation of timing oset during the packet time.

4.3

Doppler extraction

The Doppler extraction unit in Fig. (4.1)(a) comprises linear prediction of the symbol
timing oset, ne symbol timing oset, tracking the Doppler shift and CFO or
residual Doppler shift estimation.

4.3.1

Linear prediction of the symbol timing oset

As the transmission structure contains multiple OFDM frames within a packet, the
synchronization between these frames is paramount to reduce both the ISI and ICI
on the receiver side. In the proposed technique, an improvement is obtained by
involving the estimated timing oset at time i 1 in predicting the timing oset at
time i. To accomplish this, it is assumed that due to the rst order Doppler shift, the
OFDM frame could be expanded towards the leading edge or compressed towards
the trailing edge in the range [T (1 + (t)) + max , T (1 (t)) + max ], respectively.
Therefore, the linear part of the speed variation can be formulated by the rst order
equation
y = m i + b,
where m =

i i1
xi xi1

(4.9)

denotes the slope and xi is the OFDM symbol at index i, as

shown in Fig. 4.3. Accordingly, the gradient will vary gradually in accordance with
the speed change and, subsequently, the output value yi is obtained. The slope
here is determined based on the previous two OFDM symbols estimated in (4.11)
and subsequently used to predict the timing oset for the next OFDM symbol.
Therefore the rst order predicted timing oset of the current OFDM symbol can
be formulated as:
E [i ] = 2i1 i2 .

75

(4.10)

4.3 Doppler extraction

4.3.2

Fine symbol timing oset and synchronization

Thus far, two estimations of the same parameter have been obtained. It should
be stressed that attaining accurate timing oset estimation may be dicult in the
presence of noise and/or ISI, especially with a short observation window. Therefore,
for the purpose of increasing the reliability of estimation, smoothing the timing
oset is adopted. This yields the following ne tuned estimated timing oset
i = i W1 + E [i ] W2 ,

(4.11)

where the coecients W1 and W2 are empirically obtained and satisfy the condition
of 0 < W1 + W2 1. These coecients are designed to attain a trade-o between
estimation accuracy and tracking capabilities. It is crucial to mention that these
coecients have an eect on adapting the slope variation, where W1 = 1, W2 = 0
indicates fast slop variation and the linear expectation does not hold. At the same
time, W1 = 0, W2 = 1 accommodates a constant gradient between symbols. The
estimated ne timing oset in (4.11) still represents the average. Assuming the
change in the time scale is linear within the OFDM symbol, the change in the speed
is considered unidirectional. This will enable tracking of the Doppler shift caused by
speed variation within the OFDM symbol time. Performing such tracking demands
knowledge of the timing oset at both edges of the symbol in order to determine the
tracking step. By involving previous estimation of ne symbol timing oset p and
current ne symbol timing oset c , the oset at the leading edge can be formulated
as
p + c
.
s =
2

(4.12)

At the same time, the sampling frequency oset at the trailing edge e is determined
as
e = 2c s ,

(4.13)

where p and c represent the average ne timing oset estimate from (4.11). It
should be stressed that the estimation accuracy of these two parameters plays an
important role in increasing the ability to compensate for the Doppler shift and its
residual eects in the subsequent stages.

76

4.3 Doppler extraction

T ime

Figure 4.4: Tracking the Doppler within the OFDM symbol.

4.3.3

Tracking the Doppler shift

If the relative velocity between the transmitter and receiver during the packet time
is constant, i.e. for zero acceleration, then the Doppler shift estimate computed can
be used to compensate for the entire OFDM symbol. In time varying Doppler shift,
however, a unique interpolation factor for the whole symbol does not hold due to the
resulting non-negligible sampling frequency errors which must be tracked. Therefore,
the sampling frequency oset aects channel estimation, which is computed over
pilot sub-carriers, due to the dierent delays of the positions of these pilots. By
searching for the delay in the 1st signicant arrival of the estimated CIR [87], an
approach to tracking the fractional sampling clock frequency oset due to a symbol
timing error is possible. However, in the case of time varying Doppler shift, it is
necessary to estimate the sampling frequency oset frequently.
An alternative realistic Doppler shift estimator, which can be realized by adopting frequent estimation of this parameter during the symbol time in the timedomain, is proposed here. In order to track the Doppler shift, it is necessary to
derive a tracking step that corresponds to the sampling frequency oset change over
s < s + Ts < e . In such a case, the tracking step is given as
step =

e s
,
Lf

(4.14)

where Lf represents the up-sampled sub-carriers. As shown in Fig. (4.4), each

OFDM symbol is identied by the two parameters of sampling frequency oset s
and e , based on the assumption that the speed changes linearly. Accordingly, the
estimated timing oset at the leading edge is updated at each sample time k, based
on the step in (4.14). At the same time, the integer Doppler shift can be computed
as

(I)
Lf (k)

(k)
=
,
Lf
77

(4.15)

4.4 Pilot-based channel estimation

where is the sampling frequency oset initialized with s , and then updated at
each sample as:
(k) = (k 1) + step ,

(4.16)

(I) =
is rounded towards the nearest integer, respectively. This integer
and
re-sampling factor is delivered to the sample-by-sample Lagrange Quadratic interpolation unit, as shown in Fig. 4.1(b), and the fractional part is dealt with as a carrier
frequency oset. It should be stressed that the resolution of the interpolation factor
in (4.15) is entirely dependent on the transmitted frame length.

4.3.4

Residual Doppler shift estimation

Ecient Doppler shift compensation relies on how accurately the re-sampling factor
estimation reduces the residual Doppler. This residual Doppler has a direct impact
on the performance of the receiver. Taking this eect into account involves nding
the amount of the fractional part of the estimated samples that shifts the sub-carrier

(I) ), and therefore

spacing fractionally. This deviation can be modelled as ((k)

(I) ]f fc ,
(k) = [(k)

fs

(4.17)

is the residual frequency estimate. The residual Doppler shift is not constant at
each sample within the OFDM symbol and thus it is dealt with by determining the
standard deviation across the fractional part of the estimated Doppler shift. Once
the Doppler shift and its residual have been estimated and compensated, the output
signal r(k) is delivered to the outer receiver in Fig. 4.1(b). This signal is rstly down
sampled and then its cyclic prex is discarded. The PAPR phases ui are removed
prior to FFT demodulation. The zero forcing equalizer (ZFE) and least square (LS)
method for channel estimation purposes are adopted by utilizing pilots which are
embedded in a comb method. After removing the channel eect, the subsequent
stage is BICM-ID.

4.4

Pilot-based channel estimation

In the channel estimation of the OFDM symbol, a comb type arrangement of the
training sequence (pilot) is adopted. In this scheme, specic tone indices are allo-

78

4.4 Pilot-based channel estimation

cated on all transmitted OFDM symbols and the rest for data transmission. Unlike
a block-based training sequence, the comb type is quite convenient for fast fading
channels. Additionally, with the comb type, all pilots and data are transmitted
simultaneously on all symbols. It is worth pointing out that in order to increase the
accuracy of the channel estimation, the residual Doppler shift should be eliminated
[88]. This is due to an induced ICI which destroys the orthogonality among subcarrier frequency components and ultimately the diagonal of the channel matrix. In
OFDM systems, the advantage of increasing the symbol duration in reducing the ISI
eect can conict with increasing the ICI impact, as a consequence of sub-carrier
spacing reduction. Therefore, after re-sampling and CFO compensation, all subcarriers are orthogonal (i.e. ICI free). Then the training symbols for Nc sub-carriers
can be represented by the following diagonal matrix:

Xp [0]
0

Xp [1]
,
X=
..

..
.
.

0 Xp [Np 1]

(4.18)

where Xp (n) represents pilot tones at the nth sub-carrier. This diagonal representation of X is based on the assumption that the sub-carriers are orthogonal. Let
Yp (n) be the received pilot symbols after the FFT operation, then

Y [0]
X [0]
0

0
Hp [0]
W [0]

..

Y [1] 0
Hp [1] W [1]
Xp [1]
.
=

Y =

..

..
..
..
..
.

.
0
.
.

Y [Np 1]
0

0 Xp [Np 1] Hp [Np 1]
W [Np 1]
= XH + W ,
(4.19)
where Hp = [Hp [0], Hp [1], , H[Np 1]]T is a channel vector and W denotes the
p (n) are the
noise vector which is given as W = [W [0], W [1], , W [Np 1]]T . H
estimated pilot channel values, D[Xp (n)] is a diagonal matrix constructed using
the known transmitted pilot symbols. This zero forcing estimator [89] is simple;
however, it has a high mean square error. The channel estimation was implemented

79

4.4 Pilot-based channel estimation

using the least square (LS) method. [90]
X(n) =X(mL + l)

X (m),
p
=
X (m),

l=0
l = 1 , L 1,

where L =

Nc
Np

(4.20)

and Xp (n) is the nth pilot sub-carrier value. Let Hp (n) be the

frequency response of the channel for n = 0 Np 1 at pilot sub-carriers. The

p (n) is given as
estimate of the channel at pilot sub-carriers H
p (n) = D[Xp (n)]1 Yp (n), n = 0 . . . Np 1.
H

(4.21)

is estimated by minimizing the folIn the least square estimation, the channel H
lowing cost function

2


J = Y X H

H (Y X H)

= (Y X H)

(4.22)

XHY H
H + XHH
H X H,

= Y HY Y HXH
where H denotes conjugate transpose. For minimization of J in (4.22), let

J
H
H

= 0,

then
J
J
H ) + J (X H H
H X H)

=
(X H Y H
H
H

H
H
H

= X H Y + X H X H

(4.23)

= 0,
= X H Y , therefore the LS estimation is written as:
we have X H X H
LS = (X H X)1 X H Y = X 1 Y .
H

(4.24)

LS is written as:
For sub-carriers n = 0, 1, 2, , Nc 1, LS channel estimation H
LS [n] = Y [n] .
H
X[n]

(4.25)

The mean square error of the LS channel estimation is considered high when

80

4.5 Experimental results

0.4
Normalized amplitude

0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
1.5

2.5
3
3.5
Received samples

4
4

x 10

Figure 4.5: Received signal

compared with the minimum mean-square error (MMSE) estimate [91]. However,
LS is attractive in implementing real-time systems due to its simplicity. In order
to increase the reliability of the channel estimation, an interpolation in frequency
domain between each pilot and data sub-carriers is adopted. It is well known that
the LS is the rst step of the channel frequency response estimation for the known
pilots and should be followed by interpolation to obtain a non-pilot sub-carriers
frequency response.

4.5

Experimental results

The setting of this experiment was mentioned in chapter 3. Fig. (4.5) and Fig. (4.6)
show the channel measurements over a range of 1000 m. These gures show a
received frame structure and the normalized CIR of a packet that exhibits maximum
delay spread of the order of 6 ms, respectively. This multipath delay is equivalent
to an ISI of 24 symbols for a system bandwidth of 4 kHz and this delay spread
is inversely proportional to the range. In addition to the silent period shown in
Fig. (4.5), the CP guard time also contributes towards reducing the ISI eect.

4.5.1

Proposed receiver performance

To evaluate the performance of the proposed system, the experimental results for
both block-based and proposed techniques are depicted in Fig. 4.7(a). The performances of both receivers are presented in terms of bit error rate (BER). It can
81

4.5 Experimental results

Normalized magnitude

1
0.8
0.6
0.4
0.2
0

0.005
0.01
Delay (s)

0.015

Figure 4.6: Sample of normalized channel impulse response for 1000 m channel
range.
Table 4.1: Average BER comparison of the experimental results at dierent settings of weighting coecients between the proposed and block-based Doppler shift
techniques for Nc =1024
Method

Error statistics
Errors

BER

Block

6503

0.0365

Proposed-set1

772

0.004

Proposed-set2

105

0.0006

be seen that for all packets the proposed technique outperforms the block based
method. Error statistics for both schemes are presented in Table 4.1. It can be
seen that compensating the time-varying Doppler scale and its residual leads to a
reduction in the BER from 0.0365 to 0.0006, which is equivalent to 98.4%. This
is further claried in Table 4.2 which shows that the proposed technique achieves
acceptable performance in reducing errors in all packets compared with the block
technique. However, Fig. 4.7 shows high decoding error in packet 6.
In Fig. 4.7, the bit errors are high only in two blocks within packet 6, as shown in
Fig. 4.7(b). This is due to the noise eect which aects the Doppler scale estimation
when estimating the timing oset. Evidence for this is shown in Figs. 4.8(b) and (d)
where in packet 6, there is a mismatch in estimating the speed at the end of symbol

82

4.5 Experimental results

25

0.1

20
Number of errors

BER

0.01
0.001
0.0001

15
10
5

0
Proposed

9
13
Packet index

Block

17

20

(a) BER for each packer over 1000 m range.

3 4 5 6 7 8
OFDM symbol index

9 10

(b) Error statistics for packet 6; 2 out of 10

have decoding errors

Figure 4.7: Performance of the proposed system at 1024 sub-carriers.

3 and at the start of symbol 4. Therefore, a decoding error results in symbol 4.
Furthermore, it can be seen from Fig. 4.8(a) that there is a relatively high deceleration of 0.9 m/s2 during the symbol time, which adds an error in approximating
the correlation-based Doppler scale estimation. This result shows that there is a
limitation on the acceleration that can be adopted in this algorithm.
Fig. 4.8 demonstrates that the adopted system is capable of precisely tracking
the speed variation in each symbol. Particularly, in Fig. 4.8(a), the speed in symbol
3 of packet 6 has been changed three times during 0.256 s, whereas in (c) the speed
is constant. However, changing the direction of velocity within the packet period,
along with higher acceleration, can produce higher noise levels in the system. The
source of this noise is the mismatch introduced by the transition from acceleration to
deceleration, or vice versa. The proposed system detects this critical point through
the CP correlation-based Doppler scale estimation and the linear expectation has
no eect on this scenario. However, linear expectation reduces the channel and/or
noise eect on the CP correlation. Consequently, accurate Doppler scale estimation
is obtained.
Fig. (4.9) shows the performance of BICM-ID and ZFE in the experiment. In
terms of (b), the gure shows that the ZFE delivers reliable information to the
decoder. The reliability depends on how accurate the Doppler shift compensation
is. It was mentioned earlier that the channel estimation is aected by the presence
of residual Doppler shift which can cause ICI and, as a result, the orthogonality is
83

4.5 Experimental results

0.35
0.3

Estimated speed (m/s)

Estimated speed (m/s)

0.3
0.25
0.2
0.15
0.1

0.2

0.1

0.05
0

0.1

0.2

0.3

0.1

OFDM frame time (s)

(a) Packet 6; OFDM symbol 3.

0.3

(b) Packet 6; OFDM symbol 4.

0.35

0.3

0.3

0.25
Estimated speed (m/s)

Estimated speed (m/s)

0.2

OFDM frame time (s)

0.2

0.1

0.2
0.15
0.1
0.05

0.1

0.2

0.3

OFDM frame time (s)

(c) Packet 6; OFDM symbol 6.

0.1
0.2
OFDM frame time (s)

0.3

(d) Packet 6; OFDM symbol 7.

Figure 4.8: Estimated speed variation during OFDM symbol.

destroyed. Consequently, the iterative decoding stage can generate unreliable LLRs
[92]. Thus, it can be seen that there is an improvement in the second iteration (d)
compared with the rst iteration in (c). At this stage, further iterations are pointless
and no more gain is expected.

4.5.2

Eect of weighting coecients

As mentioned in chapter 3, the weighting coecients play an important role in

the accuracy of the Doppler scale estimation. For this reason, special settings of
these parameters are required in order to achieve acceptable performance. It can
be shown that there is a trade-o between the value of the weighting coecients

84

4.5 Experimental results

(b) Output IQ Constellation EQout
2

1.5

1.5

0.5

0.5

Q Channel

Q Channel

(a) Output IQ Constellation EQIn

2

0
0.5

0
0.5

1.5

1.5

2
2

0
I Channel

2
2

1.5

1.5

0.5

0.5

0
0.5

1.5

1.5
0
I Channel

0.5
1

2
2

0
I Channel

(d) Output IQ Constellation Iter2

Q Channel

Q Channel

(c) Output IQ Constellation Iter1

2
2

0
I Channel

Figure 4.9: Constellation output from equalizer and iterative receiver.

and the receiver performance. To be more specic, by appointing the symbol timing
oset, estimated by linear expectation, a lower weighting coecient than correlationbased symbol timing oset estimation means there is a constant acceleration or
deceleration between symbols, and vice versa. As shown in (4.11) and (4.14), the
Doppler scale is approximated based on estimating the ne symbol timing oset and
its tracking step is derived based on the sampling frequency oset at the start and
end of the OFDM symbol. This means that the weighting coecients have a direct
eect on the estimation of the time varying Doppler scale (t).
Fig. (4.10) shows two settings of these parameters and their eect on the performance of the receiver. In set 1, where W1 = 0.5 and W2 = 0.5, it can be seen that
the receiver performance is poor. In Fig. (4.10) (a), it is obvious that packets 5 and
6 in set 1 exhibit a high BER of 271/8920 and 71/8920, receptively. The reason for
this degradation is that increasing the weight of the linear expectation in a channel
leads to signicant acceleration that can cause maladjustment of the interpolation
85

4.5 Experimental results

250

P5Set1

200

P6Set1
P6Set2

150
100
50
0

4
5
6
7
8
9
10
OFDM symbol index
(b) Symbol index 7, P6
(c) Symbol index 1, P5
0.8

Speed (m/s)

0.25

Set2
Set1

Speed (m/s)

Number of bit errors

(a) Error statistics of packets 5 and 6 (P5, P6)

0.2

0.15

0.1

0.1
0.2
Symbol time (s)

0.6
0.4
0.2
0

0.3

Set2
Set1

0.1
0.2
Symbol time (s)

0.3

Figure 4.10: Eect of weighting coecients on estimation.

factor and make the tracking of the Doppler scale change coarsely. This is shown
in Fig. (4.10) (c). Although both sets have the same slope, there is a mismatch
between them at the start and end of speed estimation. In set 2, on the other hand,
W1 = 0.85 and W2 = 0.15, there is a great improvement in the performance as
shown in Fig. (4.10) (a), with 0 errors in packet 5 and 33 bits in packet 6. Table
4.1 shows the performance of the receiver for the sub-carriers 1024 over a range of
1000 m using two dierent settings of the weighting coecients. In set 2, it can be
observed that the error decreases by about 86.4 % compared with set 1.

4.5.3

Performance evaluation with improved coarse timing

estimation

As mentioned earlier, the impairments in the channel estimation due to synchronization failure will result in unreliable LLRs as a consequence of the Doppler eect.
In contrast, estimating and compensating the Doppler scale precisely causes the received OFDM symbol to coincide with its transmitted period; thus improving the
channel estimation and delivering reliable symbols to the decoder. Therefore, the

86

4.5 Experimental results

target is to improve the Doppler scale estimation and ultimately reduce the burden
on the channel estimation. In order to extract the Doppler scale successfully, it is
important to increase the reliability of estimating the symbol timing.
Considering the eect of acceleration on the chirp correlation is small, in the case
of multiple OFDM symbols within a packet, the symbol timing error in each OFDM
block is accumulated with acceleration during the packet time. Hence, adopting
a single estimation of for the whole packet is no longer accurate. Therefore, in
order to mitigate the acceleration eect on the symbol timing error, needs to be
ne tuned. Performing the ne tuning necessitates updating the position of after
each symbol time. Let m, i denote the range of the timing oset around the leading
and the trailing edge during the OFDM symbol, respectively. It follows that a two
dimensional timing function is written as
Ng 1





(m, i) , 
r( + m + n) r( + n + N + i)


n=0

(4.26)

m {W/2 W/2} ; i { /2 /2} ,

then, m,i can be estimated from obtaining the maximum peak of the multiplication and it can be written as

m,i = arg max (max (m, i)T )

m,i

(4.27)

m {W/2 W/2} ; i { /2 /2} ,

and the ne tuned is obtained. The implementation of this ne tuning algorithm
of the coarse packet synchronization can be summarized as follows:
1. compute the coarse packet synchronization point which represents the time
position of the maximum peak of the chirp correlation,
2. compute the timing function (m, i) for m [W/2, W/2] , i [ /2, /2],
3. choose the maximum of (m, i) as the estimated packet timing oset,
4. update to be ne tuned which is given as

= + .
87

(4.28)

4.5 Experimental results

It should be noted that a two dimensional search (i.e. m and i) is included in
the proposed timing function (m, i). This is the main dierence from the single
synchronization point estimation in [93], where only coarse estimation of the packet
synchronization point is adopted. The rst search parameter is m, corresponding
to the rst search region in the range around the coarse synchronization point .
Meanwhile, the second search parameter is i, corresponding to the range in the region
around the tail of the OFDM symbol which yields the expected Doppler shift. Once
the ne tuned is obtained, the subsequent stage is the estimation of the rst
In existing techniques, [14] and [94], due to the acceleration and
order moment .
the inherent ISI, there is a uctuation in the maximum of the timing function and the
channel conditions have a direct eect on this maximum. Therefore, centroid-based
because it reduces the position uncertainty
localization is adopted to estimate ,
caused by the fading channel, and the search range is built on the ne tuned ,
which can be written as

rD [ + Ng + N ( ) + i, + N ( ) + i],
2
2

(4.29)

and the centroid-based rst order moment l is given as in chapter 3. Fig. (4.11)
shows that ne tuning this parameter results in reducing the BER. It can be inferred from this gure that adjusting the misalignments of the symbol timing due
to the time varying Doppler scale results in an improvement in the reliability of
the re-sampling factor estimation, which in turn reduces the noise that accompanies accumulated errors from symbol to symbol within each packet and ultimately
a reduction in BER is obtained.

4.5.4

Performance evaluation based on two point correlation

Fine tuning of the coarse symbol timing facilitates an alternative approach to estimating the rst order moment of the correlation lag. The suggested approach here
aims to increase the condence of estimation by considering the rst order moment
that results from two correlation lags. The rst correlation lag is estimated by means
of centroid-based localization, in accordance with the anticipated window mentioned
earlier. This type of correlation gives an accurate indication of the fractional part

88

4.5 Experimental results

1
0.1

BER

0.01
0.001
0.0001
0
Improved2

Block

9
13
Packet index

17

20

Figure 4.11: Performance of the proposed system with improved coarse timing estimation.
of the time-scale expansion/compression. However, the centroid-based localization
is severely aected due to the velocity perturbation. This perturbation degrades
the estimation performance of the timing function and ultimately l . Therefore,
an alternative approach has been adopted by involving another estimation point
based upon full cross correlation of the CP with its replica. The addition of this
correlation is based on the idea of increasing the certainty of the rst order moment
estimation. This correlation is based on the assumption that the OFDM timing is
approximately aligned due to the ne tuning of the packet synchronization . By
denition, the cross correlation between a pair of energy signals, x [n] and y [n], is
given by [95]

rxy =

x [n] y [n ] ,

= 0, 1, 2, ,

(4.30)

n=

where the parameter is called lag and it indicates the time-shift between the pair.
Based on this theory, the time-shift in samples for either expansion or compression
can be measured with respect to a reference sequence length of the guard interval
Ng . In the case of the existence of Doppler shift, the received samples are shifted to
the right in expansion or left for compression with respect to the reference. To be
more specic, once the start of the packet is identied, it can be deemed that the
symbol timing identication is reliable and the correlation between the received CP

89

4.5 Experimental results

and its replica is computed to measure the time-shift in the samples
Ng 1






c , 
r( + n) r( + n + N ) ,


n=0

(4.31)

= 0, 1, 2, .
Considering that the reference sequence of the transmitted CP is Ncp Ns , the rst
order moment of the Doppler shift x can be approximated as
x = arg max c Ncp Ns ,

(4.32)

= 0, 1, 2, .
Adopting such a scenario requires extraction of a ne tuned correlation lag. This
necessitates involvement of two parameters of weighting coecients to perform such
a smoothing approach, as mentioned earlier. The coecients W1 and W2 are empirically obtained from the experiment to accommodate the measured channel condition.
Therefore, , which represents the ne tuned rst order moment of the correlation
lag, is given as
= x W1 + l W2 .

(4.33)

This ne tuned parameter is then delivered to the Doppler extraction in Fig. 4.1(b)
in order to estimate the Doppler shift. Accordingly, the estimated Doppler shift,
which comprises both an integer and fractional part, is considered and utilized for
compensation. Therefore, the estimated re-sampling factor requires no extraction
of the fractional part to estimate the residual Doppler shift, as shown in Fig. 4.1(b);
hence the CFO is approximated as
0.5fc f /fs

f
,
8

(4.34)

where fs = 4fc . For sub-carrier spacing of 3.90625, as in the case of 1024 sub-carriers,
is 0.4883 Hz. These two-point estimations of l and x , in conjunction with ,
contribute towards improving the Doppler shift estimation and thus eliminate the
need to determine the CFO.
Fig. (4.12) demonstrates the implications of improving the Doppler shift estimation. It is obvious in this gure that there are two estimations that show the
deceleration in velocities over the symbol time. With respect to the improved sys90

4.5 Experimental results

0.24
Proposed
Improved2

0.22

Speed (m/s)

0.2
0.18
0.16
0.14
0.12
0.1

0.1

0.2
Symbol time (s)

0.3

Figure 4.12: Improved time-varying speed estimation during OFDM symbol 7 of

packet 6.
tem, the gradient is estimated smoothly. This conrms that an accurate estimation
of the drift in samples results in an accurate estimation and tracking of the time
varying Doppler shift. On the other hand, this gure illustrates that perturbations
in estimating the variation of speed within the OFDM symbol can lead to inaccurate
re-sampling factor estimation. In particular, it can be inferred from this gure that
there is a time varying Doppler shift during the symbol time which decelerates in
the order of 0.25 m/s2 . This deceleration is estimated by smoothing estimation.
However, in the proposed system, the deceleration is approximated to 0.48 m/s2 over
the same symbol. For the sake of clarity, the proposed system refers to the system
before the improvements and the improved system refers to the proposed system
after improving estimation. Table 4.2 illustrates the performance comparison between block based Doppler compensation, time varying Doppler shift compensation
and its improvements. The achieved BER decreases signicantly in the improved
system compared with the block based approach. Likewise, there are additional
improvements in the BERs of 83.8 % compared with the proposed technique. This
is shown in Fig. (4.13)(a), where the BER of packet 6 is reduced compared with
Fig. (4.7)(a). Additionally, the error statistics of packet 6, shown in Fig. (4.7)(b)
and Fig. (4.13)(b), conrm that estimating multi-lags contributes to an increase in
the accuracy of the speed estimation.
As demonstrated in Fig. (4.13)(c), the experimental results show that the investigation was also successful with 512 sub-carriers, as it was able to improve perfor91

4.5 Experimental results

Table 4.2: Performance of the experimental results between the improved and blockbased Doppler shift techniques for Nc =1024
Packet
index
Block
Proposed
improved

10

0
0
0

119
0
0

423
3
0

347
6
0

3
0
0

44
33
3

11
0
0

505
6
0

39
9
0

2443
0
5

Packet
index
Block
Proposed
improved

11

12

13

14

15

16

17

18

19

20

24
0
0

0
0
0

33
6
0

178
14
0

21
5
9

1702 21
0
0
0
0

0
0
0

471
23
0

119
0
0

mance by about 86%. This was an expected result, because reducing the symbol
length entails increasing the sub-carrier spacing and reducing the sensitivity to the
Doppler shift. Additionally, reducing the symbol length enables more frequent tracking of the Doppler shift. However, severe consequences accompany this reduction in
the symbol time, since it mitigates immunity against ISI, in addition to reducing the
available bandwidth. This performance reveals that improving the synchronization
and adopting smoothing produces low BER. Furthermore, compensating residual
Doppler shift or CFO preserves the orthogonality of the sub-carriers and ultimately
contributes towards mitigating decoding errors.
However, it is worthwhile mentioning that this approximation of the CFO cannot
be extrapolated to all cases, as in the case of higher acceleration where a special
signal processing method, such as an adaptive weighting coecients selection and/or
iterative-based estimation of the Doppler shift, should be adopted due to the eect
of the time varying Doppler shift and the inherent ISI on the correlation peak.
Another problem with this approach is that it fails to compensate for an abrupt
change in the direction of velocity, as it needs at least two symbols to self-adapt to
this sudden variation which causes a decoding error. In terms of the achieved data
rate, Table 4.4 presents two types of OFDM sub-carrier allocation that account for
the transmission overhead due to pilots, channel coding, and guard period.

92

4.6 Simulation results

Number of errors

Bit Error Rate

0.1
0.01
0.001
0.0001

1
0
Improved

Block

9
13
Packet index

17

20

(a) BER for each packer over 1000 m range

1 2 3 4 5 6 7 8 9 10
OFDM block index

(b) Packet 6; symbol 7 error statistics

0.01

Bit Error Rate

0.001

0.0001

0
Improved

9
13
Packet index

Block

17 19

(c) BER for each packer over 1000 m range

at 512 sub-carriers.

Figure 4.13: Performance of the improved proposed system.

4.6

Simulation results

Fig. (4.14) shows, in terms of BERs, the performance comparison between the blockbased approach and the proposed technique obtained by simulations. For the blockbased approach, two scenarios of the transmitted packet structure are investigated.
The rst structure includes 20 ms chirp, followed by a silent period then 10 CPOFDM symbols. The second structure comprises only a single CP-OFDM frame.
The former structure is investigated in the experiment; therefore the second structure is considered here for the purpose of the simulation. It can be seen that the
performance of the block approach is poor in the case of multi-scale Doppler within
the OFDM symbol. When the speed is low, as shown in the OFDM symbols indices
93

4.6 Simulation results

Table 4.3: Performance of the experimental results between the improved and blockbased Doppler shift techniques for Nc =512
Packet 1
index
Block
0
Improved 0

10 11 12 13 14 15 16 17 18 19

0
0

0
0

0
0

23 31 0
0 4 9

6
0

0
0

0
0

0
0

0
0

13 14 0
0 0 0

0
0

3
0

3
0

0
0

Table 4.4: OFDM symbol structure and the corresponding data rates
Nc

Nd

Np

Nb

data rates (kb/s)

512

448

64

20

3.0833

1024

896

128

10

3.2794

10

10

10
BER

10

10

10

10

Proposed
Block
1

9
13
OFDM symbol index

17

20

Figure 4.14: Performance comparison of block based and proposed techniques.

1 and 20, the block algorithm performance is approximately identical to that of
the proposed scheme. However, as the speed increases, the BER also increases in
the block-based approach, whereas the proposed algorithm demonstrates less performance error despite an escalation in speed. The degradation in the BERs in the
proposed algorithm is due to the eect of the acceleration on the CP correlation.

94

4.7 Chapter summary

4.7

Chapter summary

The performance of time-varying Doppler shift compensation for an OFDM-based

UWA communication system has been investigated. The algorithm accommodates
for channels with linear acceleration during a packet of multiple OFDM frames.
Unlike existing Doppler compensation methods, the proposed scheme is more pragmatic, as it considers the notion that the speed is changing linearly during the
OFDM symbol time. Additionally, under the assumption of linear speed during the
packet time, it has been shown that using the linear equation approach to predict
the rst order Doppler shift as a reinforcement parameter leads to acceptable performance over other techniques. Furthermore, it has been shown that employing
weighted coecients improves the performance as it ne tunes the estimated parameters. However, an approach to ne tuning these parameters adaptively and
in accordance with the acceleration is required and will be discussed in the next
chapter.

95

Chapter 5
Adaptive time varying Doppler
shift compensation
This chapter presents an adaptive approach to address the two main problems associated with the time varying Doppler shift, the rst being the acceleration eects
on the CP correlation and the second, the eect of a sudden change in the velocity direction between packets on the entire OFDM symbols. In addition, this
chapter considers the residual Doppler shift or CFO that was estimated iteratively
within a range according to a design based on the sub-carrier spacing using pilots,
which are basically utilized for the purpose of channel estimation. Furthermore, the
proposed receiver adopts three estimations of the symbol timing oset. These estimations are centroid-based localization, rst order expectation and autocorrelation
of the received cyclic-prex with its replica. Subsequently, a penalization algorithm
is applied in order to drop the anomalous parameter among them. Therefore, the
consequences of the inection point that accompanies the abrupt change in the velocity are mitigated and a reliable time varying Doppler shift is obtained. This
Doppler shift is ne tuned in an iterative manner. The proposed receiver was evaluated through simulations and sea trials conducted over 500 m and 1000 m channel
ranges. In simulations, a model was designed to imitate the time varying Doppler
shift with two scenarios (expansion/compression) in combination with a multipath.

96

5.1 Acceleration eects

=0

Tg

(i 1)th

Symbol

11
00
0
1
00
11
0CP
1
00
11
0
001
11
0
1
min

Td

111
000
000
111
CP
0
1
000
111
0
0001
111
0
1

ith

Symbol

(i + 1)th

Symbol

max

Compression case (min )

Expansion case

(max )

Figure 5.1: OFDM symbol structure due to Doppler eects.

5.1

Acceleration eects

A pragmatic underwater communication system which adopts Doppler shift estimation and compensation should consider the change of speed with time, which is
called acceleration (a). The eect of acceleration as a result of the mobility of the
transmitter and/or receiver and causes time-varying Doppler shift. Therefore, this
type of Doppler shift can be modelled [8] as

(t) = 0 +

a(t)t
,
c

(5.1)

where 0 is the initial Doppler shift which accompanies the platform velocity and
a(t) is time varying acceleration. The acceleration eects can be twofold. First, an
eect on the chirp signal detection, particularly when the change in velocity during
the chirp period is greater in magnitude than the platform velocity. Consequently,
this mismatch aects the correlation peak of the chirp when detecting the start of
the packet. A more signicant eect of acceleration to be considered is its eect over
the whole symbol or packet. In this case, the cyclic prex and its replica undergo
dierent Doppler-shifts. This results in uncertainty of the correlational behaviour
and consequently adds an error to the rst order moment estimate l .
There are three dierent cases of an OFDM symbol subject to Doppler shift as
shown in Fig. (5.1). The rst case is when there is no Doppler shift. In this case, the
OFDM symbol coincides with the exact timing, preserving the orthogonality among
sub-carrier frequency components. In the compression case, the symbol time is reduced and the sampling frequency must be increased to compensate for the Doppler
shift whereas in the expansion case, the symbol time is increased and the sampling

97

5.2 Adaptive OFDM receiver structure

frequency is reduced. In addition, the received signal within the FFT window contains a part of the current OFDM symbol and part of the next one. This causes an
ISI and an ICI, which implies that the orthogonality has been compromised.
Let us assume without loss of generality that the initial velocity in (5.1) is zero,
therefore, the Doppler shift at the end of the packet is:

aTpac
.
c

(5.2)

In addition, it is mentioned in [8] that the Doppler shift estimation error is related
to the acceleration and the chirp duration. In the proposed CP-based Doppler shift
compensation, this error is modelled as
=

aTg
.
c

(5.3)

Therefore, for practical acceleration levels (1 m/s2 ), reducing the length of the cyclic
prex is more useful. However, this reduces the sensitivity to low acceleration. The
other crucial implication of acceleration over the symbol length to be considered is
the residual Doppler shift. In the case of constant acceleration, this eect is dealt
with by adaptive equalization in a single carrier transmission, which can not be used
with OFDM. In addition, for constant acceleration, the estimated Doppler represents
the average velocity, i.e, the maximum residual Doppler shift at the symbol ends is
given as [8]
max(residual ) =

aTu
.
2c

(5.4)

It can be inferred that, to mitigate the residual Doppler shift, the symbol length
should be reduced, and hence, the sub-carrier spacing is increased. However, in
OFDM system design, reducing the symbol length entails reducing the immunity
against the ISI.

5.2

Adaptive OFDM receiver structure

The receiver structure is comprised of an acquisition stage, an estimation of the cyclic

prex position (symbol timing), an adaptive Doppler shift estimation and compensation, and channel decoding. The receiver block diagram is presented in Fig. (5.2).
In open-loop receivers, the Doppler shift is approximated based on one-shot esti-

98

5.2 Adaptive OFDM receiver structure

r [t(1 + (t))]

r0 (k)

(k)

r(k)

-CP
FFT
000

ej2(fc +

ZFE

)kTs

er

00
Update


MUX
0

Update

Re-encode

min(
er )

CRC
BICM-ID
bi

Pilots

Phase
Correction

(a)
ReMapping

La (C)
BCJR

bi

Le (C)

Soft
Demapper

(b)

Figure 5.2: Receiver structure

mation. The iterative receiver, instead of depending on a single estimation of the
centroid-based localization and linear prediction to estimate the Doppler shift, combines conventional autocorrelation and then averaging based on three estimations.
Furthermore, the pilot has been utilized for phase error detection and correction in
addition to channel estimation. The proposed system adopts an iterative estimation
of the rst order moment that results in minimum phase and decoding errors. In a
practical communication system, there is a CRC to detect bit errors after decoding
and an action such as retransmission or repeat decoding is taken. Additionally, the
iterations rely on the criteria of minimum phase error estimation to compensate the
residual Doppler shift. In the proposed technique, the estimation errors are subject
to penalization by enabling a learning and punishment (LP) action to ne tune
iteratively. Only the minimum phase error which accompanies the ith iteration is
chosen with its associated , therefore, an accurate Doppler shift is obtained. In
learning mode (LM), the acceleration of the previous packet is observed to designate
an adaptive expectation range (ER), whereas the punishment mode (PM) drops an
out of range estimation.

5.2.1

Estimation of Symbol timing expansion/compression

In this chapter, instead of a single estimation of the rst order moment, it has been
estimated by the collaboration of centroid-based localization l , auto correlation of
99

5.2 Adaptive OFDM receiver structure

the cyclic prex with its Doppler-shifted replica yy and rst order expectation E .
In such a case, the conditional expectation of is given
[
]
= E /(0),

= E(/)
/(1),
/(2)
,

(5.5)

[
]
where the symbol E denotes the expectation operator and = l , yy , E is a row
vector of scalar real values noisy measurements. For the rst OFDM symbol j, the
estimation of the rst order moment j is based on averaging l and yy . However
an additional parameter is added which is based on the linear expectation E as
mentioned in chapter 4, therefore
l + yy + E
j =
,
3
5.2.1.1

for j > 2

(5.6)

Control range and PM algorithms

The parameter E can only be considered reliable with increasing or decreasing gradient, i.e., when the speed change is unidirectional during a packet time. However,
this is an unrealistic condition, where the speed could be steepening and levelling
o during the packet time. Therefore, it is crucial to govern the estimation within
a specic range to detect anomalous situations. This range is the rst part of the
PM and it is built on the assumption that the speed is increasing with the packet
time at constant acceleration.
Based on that, the system is capable of predicting the drift in samples in the next
symbol. Let us dene a new variable a to buer the absolute dierence between c
and p
a = |c p |,

(5.7)

where c and p represent the current and previous estimation at time j and j
1, respectively; determining the mean value a of (5.7) over the OFDM blocks.
Accordingly, we formulate a general expected range in samples C and it can be
written as
C j1 2|a |,

(5.8)

where the (+) sign indicates an acceleration in the expansion of the signal since the
distance is increased and vice versa. Algorithm 2 is developed to deal with these
scenarios.
100

5.2 Adaptive OFDM receiver structure

input : Parameters a , j1
output: Range for CP , CN
1
2
3

if F lag > 10 then // <Flag represents the symbol index>

CP j1 + 2|a |
CN j1 2|a |
else
CP j1 + 4
CN j1 4

4
5
6
7

end
Algorithm 2: Range algorithm

It can be noticed from algorithm 2 that CP and N P ranges are assigned for the
positive and negative acceleration, respectively. Particularly, if j1 = 5 samples
and the average drift in samples of the the previous 10 OFDM blocks were a = 2
samples, therefore, it is expected to be j1 2. Accordingly, in algorithm (2),
lines 2 and 3, we expand the range to a square half of this coecient. In this case the
range is expressed as [j1 2|a |, j1 +2|a |] instead of [j1 |a |, j1 +|a |]. In
lines 5 and 6 on the other hand, j1 4 is based on the assumption that a = 1m/s2 .
In this case the speed will change 0.25m/s in each OFDM symbol and this can be
interpreted in terms of samples to 2 samples. As in lines 2 and 3, the tolerance is
also increased by 2.
The second step in the PM is to set the conditions that are needed to make an
action to correct the estimation. There are three cases adopted here to perform the
PM. In each case, two out of three parameters are considered and the third one is
dropped. This procedure is resorted to in order to accommodate the abrupt change
in the direction of the velocity, hence, the range control detects this perturbation
in the speed while the PM applies the appropriate action by ignoring the nuisance
parameter. Consequently, the average of the reliable parameters are considered and
utilized in the search. This procedure of PM is shown in algorithm (3).

5.2.2

Early termination search algorithm

In this algorithm, we are trying to estimate and compensate the time varying
Doppler shift recursively. An adaptive step-size is formulated in accordance with a
number of iterations to obtain an optimal search that results in minimum errors.
The criteria of optimality is adopted here in the sense of performance investigation,

101

5.2 Adaptive OFDM receiver structure

input : Parameters (i),

CN , CP
output: Parameters within the expected range
1
2
3
4
5
6
7
8
9

for i = 1 : 3 do
< CN (i)
> CP ) then
if ((i)
switch (i) do
case (1)
0.5((2)
+ (3))

(1)
case (2)
0.5((1)
+ (3))

(2)
case (3)
0.5((1)
+ (2))

(3)

10

endsw

11
12
13
14
15

else

(i)
end
end
Algorithm 3: PM Algorithm

therefore, the CRC is employed to terminate the search swiftly once there is zero
decoding errors. On the other hand, this search algorithm reveals the minimum
phase error and their accompanied parameters that give the lowest BER to be utilized later in the outer iteration. This outer iteration is enabled when the search
algorithm fails to produce zero decoding errors.
5.2.2.1

Selection of step-size () and correction factor (Ki )

For a closed-form system that contains several instantaneous variables, the estimation of the required parameter is generally not possible [96]. An alternative solution
to approximate the parameter is adopting an iterative approach. The estimation of
the parameter at iteration i represents the initial expectation and then this estimation is resumed recursively to improve it. Based on this approach, the parameter
i1 which is ne tuned earlier to produce minimum error among three estimation
agents, is utilized. The adaptation factor is shown as
= 0.33(

sgn(i/2) |i/2| n
) ,
0.5

(5.9)

where n is a positive integer exponent and represents the search points. It should be
stressed that equation (5.9) is empirically obtained. The search is chosen to converge

102

5.2 Adaptive OFDM receiver structure

the correction term given in (5.10) towards minimizing the phase error and reliable
This correction term is initialized to 1 and then approximated
estimation of .
iteratively. The idea behind selecting a cubic exponent step size is to search in a
convergent manner as the cardinality of estimation is high at the beginning and once
it diverges the estimation error is expected to be increased. In terms of complexity,
this search algorithm is better than a linear approach, where it requires higher
execution time. In addition, this type of search has an automatic early termination
(AET) condition. This termination depends on:
1. The CRC results 0 errors,
2. Reaching the maximum search points.
Resorting to the iteration is to practice another step-size and correction term
that should be selected closer to those at previous iteration. In this manner, it is
devised that the search algorithm diverges one step per iteration around the range
[/2] towards left and [/2] towards right. An action is taken in case of reaching the
full range by considering the estimated at iteration i.

Ki =

+ 1

mod(i, 2) = 0

otherwise

(5.10)

As shown in Fig. (5.15), the exponent n of the step-size in (5.9) plays an important role on reducing the errors. Although a higher degree of exponent indicates that
the estimation is reliable, there is a level at which no improvement gain is obtained.
It is shown in Fig. (5.3)(b) that at n = 3, the horizontal asymptote start smoothly
during the rst 10 candidate points of the search range, whereas, the smoothness
period is smaller when n = 2. On the other hand, in a linear case n = 1, the step-size
is constant.
The implications of the step-size in (5.9) are shown in Fig. (5.3)(a). In this gure,
the correction factor Ki is changing in accordance with the step size to ultimately
enforce c to converge. However, failing to attain an improvement and ultimately
converging to AET condition 1 results in an increased estimation error, hence, the
correction term in (5.10) diverges and then the search algorithm starts to choose a
larger step-size.

103

0.35

0.95

0.3

0.9

0.25
Amplitude

Amplitude

5.2 Adaptive OFDM receiver structure

0.85
0.8
0.75

0.2
0.15
0.1

n=3
n=2
n=1

0.7
0.65

n=3
n=2
n=1

10

0.05
20
30
Search points

40

50

(a) Change of correction factor as a function

of the exponents.

10

20
30
Search points

40

50

(b) Change of step-size as a function of the

exponents.

Figure 5.3: Eect of exponents on step-size and correction factor convergence.

5.2.2.2

Time-varying Doppler shift estimation and tracking

In terms of performance, when n = 3, it can be inferred that the reliability of

estimating c in (5.15) is high and is only needed to ne tune the approximation,
thus it necessitates adjusting the step size closer towards the left or right around
the middle of the search range. In this case ne tuned c is obtained. It represents
the timing oset at the start of the OFDM symbol, which is approximated as
i.
c = K

(5.11)

The parameter c contributes to the improvement of other dependent parameters,

particularly, the tracking step. Therefore, the iterative approach represented by
the search is important to approximate the Doppler shift estimation. To be more
specic, let us assume that the speed between the transmitter and receiver is 1 m/s,
which is equivalent to 8 Hz for a carrier frequency of 12 kHz and sampling frequency
=1.0006 and the estimated should be 2.048 for 12288 FFT up4fc , therefore,
sampling. Actually, these calculations yield that there is a demand on estimating
and compensating such Doppler shift that has a fraction of a variable time expansion
and/or compression. Therefore, dealing with such time-varying Doppler shift needs
to track this variation within the symbol time. In (4.14), this Doppler is dealt with
by deriving a tracking step to estimate this variation based on dividing the timevarying Doppler shift into an integer part for re-sampling and its residual or CFO
104

5.2 Adaptive OFDM receiver structure

is represented by the fractional part of the Doppler shift and smoother Doppler
shift estimation is obtained. However, in the proposed adaptive system, recursive
iteration to ne tune c is adopted and the time-varying Doppler shift contains
the re-sampling factor with its residual. Furthermore, it has been dealt with CFO
estimation separately hence, the Doppler shift is given as

Lf (k)

(k)
=
,
Lf

(5.12)

where (k) is the sampling frequency oset initialized with s at k=1 and its update
is approximated as
(k) = (k 1) + step ,

(5.13)

where step is given in (4.14).

Utilizing 4th order Lagrange interpolation polynomial [97] for re-sampling based

upon the parameter (k)

to produce r (k). This re-sampled signal can be mathematically written as

r (k) =

rm [k(1 + (k))]Vi (m ),

(5.14)

i=0

where
Vi (m ) =

m mn
,

m
n
i
i=0,n=i

(5.15)

where, Vi (m ) represents the polynomial of degree N associated with each node i.

where m = m + (k)
initiated with 3, n {1, 2, ...N }, R = m and mi = Ri2 .
Therefore, for ve points N = 4, the current point, the two previous points and next
two points are considered to t the interpolation curve.
5.2.2.3

Residual Doppler shift estimation

Post-FFT CFO estimation is adopted. When all angles of the received nth pilots
Yp (n) are shifted by the same angle, the ZFE is capable of correcting the rotation.
However, this is not the case where each sub-carrier is rotated depending on the
residual Doppler shift. In order to estimate this residual, a range of these parameters are assumed. Start with and for each candidate i, the phase error vector is

105

5.2 Adaptive OFDM receiver structure

determined as
Np 1

ei (n) =

(Yp (n)) (Xp (n)),

(5.16)

n=0

considering the mean phase error between the transmitted and received pilots that
is given as

Np 1
n=0

ei =

ei (n)

Np

(5.17)

therefore, the estimated residual phase error at the pilots sub-carriers indices can
be formulated as
Np 1

er =

|
ei (n) ei |,

(5.18)

n=0

and the CFO can be approximated as a function of this pilot-based residual phase
error estimation
= ( er ).

(5.19)

min

This criteria denotes the CFO candidate that accompanies the lowest phase error.
Hence, after re-sampling, the resulting received signal r (t) is then down-converted
to baseband with the chosen and can be written as

r(t) = r (t)ej2(fc + )kTs .

(5.20)

After compensating the CFO in (5.20), the resulting signal r(t) is converted to
the frequency domain and delivered to the ZFE. To improve the receiver performance, post-FFT tracking is useful to mitigate the remaining CFO [86]. Although
this residual is small, it degrades the receiver performance due to the accumulated
phase rotation consequences along each OFDM symbol in the packet [98]. On the
assumption that the CFO is compensated earlier, it is worth eliminating its phase
rotation eect. Here in this proposed technique, we utilize the pilots to estimate the
residual phase error of the current OFDM symbol which can be written as
i = (Y (n)) ei ,

106

(5.21)

5.3 System design, simulation, and experimental results

+CP
bi

s0n

PSF

Pilots

IFFT

Sk
.
S/P ..
MAPPING

MUX

FEC

sn

Carrier

Re{}

Modulation

Figure 5.4: Structure of the transmitter used in simulation.

and the residual phase error correction is written as
r = Y (n) = eji ,

(5.22)

which is the OFDM signal after residual phase correction. This yields to deliver
reliable information to the BICM-ID decoder.
Based on the CRC, a decision is made to terminate the search or resume the
iteration. In Fig. (5.2), it can be seen that once the search points are completed
and there is an error after performing the CRC, which is assigned a dashed line, the
outer iteration is enabled as a nal trial. To deal with this case, the phase error
given in (5.16) is utilized to buer the associated parameters c and that result in
minimum phase error during previous iterations. Exploiting c updates the Doppler
shift and produces a new interpolation factor whereas is utilized in compensating
the CFO. In order to distinguish the CFO at each stage, we use the variable ,
which denotes the output of the multiplexer among three estimations.

5.3

System design, simulation, and experimental

results

Fig. (5.4) depicts the structure of the transmitter used in simulation. For the sake
of simplicity, it is assumed that the system under consideration does not account for
the PAPR, as in the experiment. The binary information bits bi of length Kd are
applied to the FEC of type NSC to produce a codeword, sn , of length Kc = Kd /R
encoded bits, where R (0, 1] is the coding rate. The coded bits are permuted
by a random interleaver of length LI to generate bit sequence cn , then converted
in groups of m successive bits into alphabet symbols of constellation size M = 2m .
This mapping operation induces a sequence of Nu = Kc /m : s = {s0 ....sNu 1 },
where si C and C denotes the set of complex symbols. Subsequently, in the

107

5.3 System design, simulation, and experimental results

OFDM symbol to be constructed, pilot symbols of phase shift keying (PSK) with
unit amplitude are embedded with the data symbols in a comb method. These
pilot symbols are multi-purpose. Firstly, they are used for the purpose of channel
response estimation at the receiver and secondly, as a reference for phase correction.
The resulting OFDM symbol containing pilots and data-bearing sub-carriers is then
modulated by an IFFT of size Nc and the last samples are copied and preface the
symbol to form the CP-OFDM frame. The resulting frame is pulse shaped using a
PSF and then up-converted using carrier modulation. The passband model of the
transmitted OFDM ithsymbol x(t) is mathematically represented as
}

n
1
j2 (tTg iT )

xi (t) = ej2fc t
prc (t iT )
di (n)e Td
Nc nI
i=0
{
}
1
j2fn (tTg iT )

=
di (n)e
prc (t iT ) ,
Nc nI
i=0
{

(5.23)

where di (n) is the symbol transmitted over the nth sub-carrier, prc (t iT ) is the
pulse shaping lter.
The transmitted signal x(t) in (5.23) is passed through the channel model shown
in Fig. (5.5). This model is adopted to imitate the case of the time varying Doppler
shift with constant acceleration. Performing this type of simulation necessitates designing a packet structure which contains multiple OFDM symbols to accommodate
the required acceleration. As mentioned earlier, the LFM signal is utilized for packet
synchronization, however, the eect of acceleration on the chirp is not negligible with
such a type of packet transmission. Therefore, the chirp also undergoes this eect in
the simulation, hence it is involved in the acceleration and deceleration of the rst
and second packet, respectively. It should be stressed that there is an acceleration
in the expansion case or in the compression, similarly for deceleration. This is illustrated in Fig. (5.6) where the uphill and downhill of the solid line mean there is a
change in the direction of the velocity with time i.e., inection point from acceleration to deceleration whereas the at line means that the relative velocity between
the transmitter and receiver is constant or zero acceleration over the duration of
the packet. Likewise for the dotted line, in this case, the velocity increases towards
the negative in the rst packet then starts decreasing towards the positive in the
second packet. Accordingly, the simulation uses two consecutive packets to imitate
108

5.3 System design, simulation, and experimental results

hl

AWGN

(t)
xi (t)

r[(1 + (t))t]

Acceleration/
Deacceleration
Channel

Figure 5.5: Simulation model.

the proposed system, namely packet 1 and packet 2. The rst one was accelerated
and the second one decelerated.
In simulating the time varying Doppler shift, the speed was assumed initially
equal to zero and then the terminal speed of the packet is given as

Vmax (t) =

a(t) Lpac
fs

(5.24)

= a(t) Tpac .
Therefore, the associated Doppler shift at the end of the packet relative to the
propagation speed can be written as

max (t) =

c Vmax (t)
,
c

(5.25)

where max (t) represents the Doppler shift at t = Tpac . Based on the assumption
that the speed is changing linearly during the packet duration, then at each sample
time within the OFDM symbol, the rst order Doppler shift is formulated as

step =

max (t) 0
,
Lpac

(5.26)

where t = Tpac . Based on this step, the Doppler shift is speeding up until arriving
at the last symbol in the rst packet and then starts slowing down. It is well known
that there are lots of UAC channels that have been characterized yet, however
there are no standard as in the case of RF channels [99], therefore, channel A is
adopted, hence the subsequent stage is convolving this time dispersion channel with
the Doppler-shifted incoming signal and then adding the AWGN to investigate a

109

5.3 System design, simulation, and experimental results

Speed (m/s)

Zero
2
0.25 m/s
2
0.5 m/s
2
0.75 m/s
2
1 m/s

2
3
4
Packet Time (seconds)

Figure 5.6: Scenarios of time-varying Doppler shift.

more realistic case.

5.3.1

System design parameters

5.3.1.1

Transmitted packet structure

The system bandwidth of 8 kHz (8 kHz- 16 kHz) is swept by a 50 ms chirp which

prexes a packet. The signal packet comprises 10 CP-OFDM frames of QPSK.
The length of each QPSK OFDM frame plays an important role in controlling the
performance of the Doppler shift estimation and compensation. In addition, required
Doppler resolution and acceleration contribute in determining the OFDM frame and
packet length, respectively.
5.3.1.2

Parameters of cyclic prex

Due to the symmetry of the CP with its replica, there is a good correlation property
of this guard interval denoted as cyclostationary because there is a cyclic convolution with the channel in the time domain. However, depending on the transmitted
data, resulting envelops of the correlation peaks and their sidelobes are varied. Particularly, if the transmitted data are random, the peaks and side-lobes are variable
whereas with symmetrical data (the start and end of the frame contain the same
data) the peak-to-average power ratio is symmetrical. Since the Doppler is changing

110

5.3 System design, simulation, and experimental results

60
Zero
Comp.
Exp.

50

Magnitude

40
30
20
10
0

50
Correlation lag

100

Figure 5.7: Correlation lag variation due to time scale expansion/compression at an

acceleration of 1 m/s2 .
with time, there is a mismatch in the CP correlation, where the Doppler aects the
rst part of the CP by dierent amount of the second CP part. This mismatch
appears in the position of the CP and this case is dealt with by searching within
a specic window around the leading and trailing edge of the OFDM symbol as
discussed in chapter 4. Furthermore, there is a mismatch in the length of the CP
windows. To be more specic, let us assume the acceleration is 1 m/s2 and the
Doppler frequency shift at CP1 is Fd , then we expect Fd 2Hz at CP2 when the
symbol time is 0.25 s. This frequency shift is ignored relative to fs and Ng Ns .
Therfore, reducing the CP length could be useful in terms of its sensitivity to the
Doppler shift and bandwidth.
The bandwidth of the cyclic prex is chosen to accommodate the channel impairments and to minimize the loss of data rate. In this case, the coherence bandwidth
lower bound is given as

Bc (lower) =

Nc
f.
Ng

(5.27)

Furthermore, the BT product is subject to the required amount of gain to achieve

reliable detection. A gain of (18 dB) (BT=64) was determined in accordance with
the OFDM signal design to be sucient, and therefore, the cyclic prex is a 16 ms
period.

111

5.3 System design, simulation, and experimental results

5.3.2

Simulation results

In this section, we present the simulation results of the proposed system based on the
simulation model described in Fig. (5.5). The CIR was h(n) = 0.6708(n)+0.5(n
1)+0.3873(n2)+0.3162(n3)+0.2236(n4). In order to investigate the system
performance, two scenarios are considered, acceleration (expansion) and deceleration
(compression) up to 1.1 m/s2 . Fig. (5.7) shows the output of the centroid-based
correlation. It can be seen in this gure that the length of the correlation window
in the x-axis is 100 samples and the centre is 50 as in the case of zero Doppler
shift, therefore, any drift relative to this centre due to an expansion or compression
is exploited to estimate the timing oset. Fig. (5.8) shows a plot of the BERs at
SNR=15 dB and a maximum delay spread of 10 ms for Nc = 1024. In order to assess
the proposed system with the two scenarios, dierent accelerations and various CP
lengths were used. For CP=32 or 8 ms, the system fails in all scenarios and at
dierent accelerations. This is due to the severe ISI that introduces a delay spread
greater than the CP length. However, for CP=64 and 128, the receiver achieves a
satisfactory performance through all anticipated accelerations and scenarios. A clear
benet of increasing the CP length over shorter CP is in the low acceleration case.
This is palpable in Fig. (5.8)(a) at a = 0.3 m/s2 . That means, at low accelerations
we need to increase the resolution of the estimation by extending the CP length.
In contrast, at higher accelerations, the impact of increasing the CP length on the
performance is marginal as shown in Fig. (5.8)(b).
Fig. (5.9) shows the performance of the time varying Doppler shift compensation
versus the delay spread of the channel. There was a signicant reduction in the BER
over a short delay spread. A possible explanation for this might be that shorter delay
spread increases the certainty of the CP correlation peaks due to the increased area of
the ISI free region; hence an accurate Doppler shift is obtained. Accordingly, these
simulation results conrm a trade-o in the design of an OFDM frame structure
between the spectral eciency, the desired acceleration and ICI reduction. That
means the guard time Tg should not only be chosen to achieve the condition max ,
rather it should also consider what is the required maximum velocity that attain
the optimal performance. Therefore, with the design parameters of Tg = 16 ms and
Nc = 1024, it is veried from these simulation results that the system can attain a
BER of 105 at a = 0.5 m/s2 and max =5 ms.

112

5.3 System design, simulation, and experimental results

10

10

10

10

10

10

10

0.3

CP=128
CP=64
CP=32

BER

BER

CP=128
CP=64
CP=32

0.5
0.7
0.9
Acceleration (m/s2)

10

1.1

0.3

(a) Acceleration.

0.5
0.7
0.9
Acceleration (m/s2)

1.1

(b) Deacceleration.

Figure 5.8: BERs performance for dierent acceleration and cyclic prex lengths at
SNR=15 dB, max = 10 ms, and Nc = 1024.
0

10

10

BER

10

10

=5ms
=10ms
=15 ms

10

10

10
12
EbNo (dB)

14

16

Figure 5.9: Eect of the maximum delay spread max on the BER performance at
a = 0.5 m/s2 and Tg = 16 ms.
Fig. (5.10) depicts alternative scenarios when a higher acceleration is simulated.
In order to investigate the system performance at higher acceleration, the simulation results are rstly obtained with AWGN at a = 1 m/s2 and then under the
inuence of the multipath channel. It can be shown that the performance of the
proposed system in a combination of broadband time varying Doppler shift and multipath channels can achieve an acceptable performance. These results suggest that
the proposed adaptive receiver can accommodate a multipath channel of max =10

113

5.3 System design, simulation, and experimental results

0

10

BER

10

10

AWGN
a= +1
Da= +1
Da= 1
a= 1
a=0

10

10

10
EbNo (dB)

15

Figure 5.10: BER performance with the system parameters Tg = 16 ms, Nc = 1024
at max =10 ms and a = 1 m/s2 , where Da denotes deceleration case.
ms and time varying Doppler shift of a = 1 m/s2 in the case of expansion and
compression with an acceptable BER. The maximum speed that associates this acceleration at the terminal (OFDM symbol number 10) of 272 ms packet duration is
2.72 m/s.
Table 5.1: Main system specications
Parameter
Value
System bandwidth

4 kHz

Carrier frequency

12 kHz

Sampling frequency

48 kHz

Cosine roll-o factor

0.98

Code rate

1/2

Convolutional code polynomial

[23, 35]8

5.3.3

Experimental results

5.3.3.1

Experimental setup and channel characteristics

This experiment was conducted by Newcastle University at the UK coast. The

system parameters for each OFDM frame are summarized in Table 5.1. The transmission power was set at 108 dB 1Pa. It is known that in UAC, the multipath
delay spread is inversly proportional to the distance between the transmitter and
114

5.3 System design, simulation, and experimental results

receiver and the delay spread of the short range channel is usually long. Furthermore, the presence of the Doppler shift due to the transmitter/receiver pair motion
increases the burden on the receiver due to the synchronization impairments consequences. Fig. (5.11) and (5.12) show the detected channel proles for the 100
and 500 m ranges, respectively. These gures depict the time varying CIR which
is normalized to unity, the CIR of a single packet which is selected randomly and
the time varying spectral before and after the BPF. The impulse response of the
channels is determined by the FIR-correlator of the LFM chirp. It can be seen that
the maximum delay spread is up to 11 and 6 ms, for 500 and 1000 m channel ranges
respectively. However, upon a closer look to the CIR of 500 m range, it can be
inferred that it is more severe as it exhibits longer delay spread and some of their
paths have a comparable amplitude relative to the direct path. Consequently, synchronization impairments challenge the receiver performance, particularly, for the
ZF estimator. Furthermore, in the case of 500 m range, it can be observed that the
time dierence of the arrival paths is quite signicant; therefore, the CIR becomes
larger [100]. Moreover, Fig. 5.11(c) and 5.12(c) show a visual representation of the
received acoustic signal. It is noticeable throughout the spectrogram that the noise
levels of the 500 m channel range are higher than the 1000 m channel range. The
BPF is capable to mitigate these frequencies as shown in Fig. 5.11(d) and 5.12(d).
5.3.3.2

Performance evaluation of the proposed receiver

In addition to the simulation mentioned in the preceding section, this experiment

is carried out to asses the system performance. The performance of the adaptive
receiver is evaluated using the packet structure shown in Fig. (3.7) transmitted
through the multipath channels described in Section 5.3.3.1. The chirp was used
for the purpose of packet synchronization. This synchronization is achieved by
correlating the LFM signal with its replica after the BPF. The highest correlation
peak has been chosen to indicate the start of the packet. The performance of the
proposed receiver is evaluated based on the criteria of the decoding BERs over 500
and 1000 m channel ranges and dierent OFDM data structures. CP-OFDM was
used with a guard interval of Tg = 16 ms for each OFDM symbol. The number of
sub-carriers used in the experiment were Nc =512, 1024 and 2048 with a sub-carrier
spacing f = 7.81, 3.906 and 1.9531 Hz, and OFDM symbol duration Tu = 1/f =

115

5.3 System design, simulation, and experimental results

Normalized amplitude

0.005

0.01
Delay (s)

0.015 0

0.8
0.6
0.4
0.2

Tim

0
0

20

e(
pa
ck

40

ets

0.5

Normalized magnitude

0.005
0.01
Delay (s)

(a) Time varying CIR.

(b) Sample of CIR.

0.6
50
0.4

0.8

0
Frequency

Frequency

0.8

0.015

50

0.6
100
0.4
150

100

0.2

0.2
200

0.5

1.5
Time

2.5
6

x 10

(c) Before BPF.

0.5

1.5
Time

2.5
6

x 10

(d) After BPF.

Figure 5.11: Channel measurements for 1000 m range.

0.128 s, 0.256 s, and 0.512 s respectively. The code rate was 1/2 NSC and a QPSK
modulation scheme was utilized. Each packet comprises respectively, 20, 10 and 5
OFDM frames Nf for Nc =512, 1024, and 2048 where the frame includes CP and
OFDM symbol. The total number of information bits per packet are 8880, 8920 and
8940 for Nc = 512, 1024, and 2048 respectively. For Nc = 512, 19 packets were sent
whereas 20 packets were sent for both cases of 1024 and 2048 sub-carriers. Each
group of them was sent separately and at dierent time intervals over both channel
ranges.
Fig. (5.13) presents the average BERs over each transmitted packet for 1000 m
channel range and dierent sub-carriers spacing. It can be shown from this gure
that the proposed scheme surpasses the block-based Doppler shift compensation
116

5.3 System design, simulation, and experimental results

Amplitude

0.8
0.6

20
0.005

0.01
Delay (s)

0.015

Tim
e

0
0

0.8
0.6
0.4
0.2

(pa

40

0.2

cke
t

0.4

Normalized magnitude

0.002 0.004 0.006 0.008 0.01 0.012

Delay (s)

(a) Time varying CIR.

(b) Sample of CIR.

1
0

0.8

0.8

50

0.4

100

0.2
0

Frequency

Frequency

50
0.6

1.5
Time

100
0.4
150

0.2

150

0.5

0.6

2.5
6

x 10

(c) Before BPF.

0.5

1.5
Time

2.5
6

x 10

(d) After BPF.

Figure 5.12: Channel measurements for 500 m range.

in approximately all received packets. In total, 2 out of 20 packets have decoding errors in the case of 1000 m channel range and 1024 sub-carriers as shown in
Fig. (5.13)(a). The number of errors in these packets were marginal and contributed
by OFDM symbol number 2 and 6 of packets 8 and 19, respectively. This result
is very encouraging, especially with 1000 m range and the consequences of the associated transmission loss. On the other hand, for 512 sub-carriers, 3 out of 19
packets have decoding errors as shown in Fig. (5.13)(b). This is an expected result,
where the immunity of the OFDM sub-carriers against the ISI are proportional to
the OFDM symbol length while the sensitivity to the Doppler shift is increased with
symbol time.
Fig. (5.13)(c) presents the BER performance comparison of the proposed system
117

5.3 System design, simulation, and experimental results

technique with a block-based approach for long OFDM symbol times. It can be
observed that 13 out of 20 packets are error-free. It is evident that the adaptive
system is capable of achieving satisfactory performance despite its tight sub-carrier
bandwidth. Additionally, this result reveals that without a CFO compensation, the
receiver deteriorates as in the case of a block-based method, therefore, estimating
the coarse re-sampling factor is not enough to achieve reliable communication.
It can be seen in Fig. (5.13)(d), there is a recurrent high BERs in OFDM symbol index 1 for the whole transmitted packets in the case of 1000 m range and
Nc =2048. The underlying reason for this case is due to the false alarm in detecting
the maximum peak of the chirp signal which results in synchronization impairments.
Furthermore, block error rates (BLERs) in this gure showed that the system capable of recovering the rest of the symbols within the packet and tracking the Doppler
shift variations, where the errors are reduced dramatically with the packet time as
shown in OFDM symbol 2 and 3. The reduction of the errors admit that the PM is
working perfectly in the process of dropping the extraordinary parameters.
Turning now to the experimental evidence over 500 m channel range. Comparing
the two results in the case of Nc = 1024 and 2048, it can be seen in Fig. (5.14) (a)
and (b) that despite a signicant reduction in the BERs with respect to the blockbased technique, the performance of 1000 m range surpasses 500 m. These results
are not surprising because it conrms that the associated ISI in a long distance is
short due to the reduction in the delay spread length and vice versa. Through the
eective contribution of the CP in estimating the Doppler shift, this delay spread of
the channel will aect the certainty of the correlation peaks, therefore, room for CP
correlation should be given. The suggestion here is to choose the CP longer than the
channel dispersion time by at least 5 ms to achieve reliable communication, hence
this revels a trade-o between the bandwidth eciency and CP length. Furthermore,
longer OFDM symbol time increases the immunity against the ISI, therefore, in the
case of 500 m channel range and Nc = 512, the system fails. The aforementioned
results conrm that the proposed scheme achieves near error-free transmission over
1000 m range with a sub-carriers length of Nc = 1024 or 512. In addition, although
there is a narrow sub-carrier bandwidth, a satisfactory performance is achieved with
Nc = 2048.

118

0.1

0.1

0.01

0.01
BER

BER

5.3 System design, simulation, and experimental results

0.001
0.0001

0.001
0.0001

0
Block

9
13
Packet index

Adaptive

17

Adaptive

20

9
13
Packet index

17 19

(b) Nc = 512.

0.1

0.1

0.01

0.01

Average BLER

BER

(a) Nc = 1024.

Block

0.001
0.0001
0

0.001
0.0001
0

Block

9
13
Packet index

Adaptive

17

20

(c) Nc = 2048.

3
Block index

(d) Average BER per symbol (Nc = 2048).

Figure 5.13: BER performance of adaptive time varying Doppler shift compensation
receiver over 1000 m channel range for dierent sub-carriers length.
5.3.3.3

Search points, exponents and CFO range selection

The search points, exponents selection and CFO have a direct impact on the success of the receiver operation. In the search points, reducing the complexity of the
receiver is crucial, and in such a case, a limited number of inner iterations is preferred. Consequently, we resort to early termination strategy in order to reduce
the computation complexity. That is, the iterations are terminated once the BER
of the current OFDM symbol attains zero error on the CRC output, otherwise,
the receiver resumes to reach its predetermined iterations. At each iteration, two
parameters are updated adaptively in accordance with the search points and the
exponents. These parameters, i and Ki ne tune c and an accurate Doppler shift
119

0.1

0.1

0.01

0.01
BER

BER

5.3 System design, simulation, and experimental results

0.001
0.0001

0.001
0.0001

0
Block

9
13
Packet index

Adaptive

17

Block

20

(a) Nc = 1024.

9
13
Packet index

Adaptive

17

20

(b) Nc = 2048.

Figure 5.14: BER performance of adaptive time varying Doppler shift compensation
receiver over 500 m channel range for dierent sub-carriers length.
is obtained. Fig. (5.15)(a) demonstrates the adaptive change of the tracking parameters c , s , and e of symbol index 8 of packet 6 with time. It is palpable
from this gure that at the rst iteration, the dierence between them is big which
results in a large tracking step. Accordingly, the CRC test does not indicate zero
error, therefore the iterations are continuing to investigate the system with further
ne tuned parameters. Obviously, the step size should be controlled to correct the
values around the estimated parameters. However, as the iterations increase, the
step size is diverged. This is shown in Fig. (5.15)(a), where e starts to change its
step size automatically after iteration 5.
This step size, through the use of exponents order n given in (5.9) will contribute
to how large the correction term Ki at the next iteration is and will contribute ultimately to update the interpolation factor adaptively. In particular, Fig. (5.15)(b)
shows two values of the exponents that have been chosen to investigate the performance of packet 6 over 1000 m channel range. In this gure, it can be seen that
at n = 2, the system exhibits an acceleration of 1 m/s2 due to the change of
estimated speed, which is given as
vr (t) = [(t) 1] 1500,

(5.28)

during the symbol time that results in decoding errors in OFDM symbol 8 of
packet 6 even with 10 inner iterations and 2 outer iterations. On the other hand,

120

5.3 System design, simulation, and experimental results

1.8

0.35
c

n=3
n=2

Estimated speed (m/s)

Estimated

2.2
2.4
2.6
2.8

0.3

0.25

0.2

3
3.2

4
6
Search points

(a) Estimation of the parameter at n = 3.

1
2
Symbol time (s)

(b) Exponents eect on speed estimation

of packet 6, symbol 8.

1
Pac6

Pac8

CFO (Hz)

0.5

0.5

4
6
8
OFDM symbol index

10

(c) CFO estimation for packets 6 and 8.

Figure 5.15: Eect of exponents on the estimation of i and Ki , search points and
smoothing the tracking step at Nc = 1024 over 1000 m channel range.
set n = 3 demonstrates the speed is changing very smoothly during the symbol
time and the acceleration is almost zero, therefore, an evidence of the smoothness
caused by the exponents order is interpreted in the resulting of zero error in packet 6
with 8 iterations as shown in Fig. (5.13)(a). It is worth mentioning that no specic
rule has been adopted to choose the exponents order n in the proposed receiver.
However, it has been noticed the order increases proportionally with the sub-carriers,
therefore, n was set equal to 2, 3 and 4 for Nc = 512, 1024, and 2048, respectively.
This is basically true on the assumption that it is likely the velocity changes with
longer symbol time, especially with high acceleration, hence an adaptive step size is
required.
121

5.4 Chapter Summary

The residual Doppler shift or CFO [12] destroys the orthogonality among subcarrier frequency components. Consequently, in order to maintain the receiver performance, this CFO is estimated. As explained in Section 5.2, a range of CFO
candidates are chosen based on the sub-carrier spacing for each OFDM symbol
length and a CFO that results in minimum phase error is selected among the range.
Starting with = f fc Ts = 0.25f , where fs = 1/Ts = 4fc , and for each candidate
i repeat
= i ,

(5.29)

where {i R : 1 6 i 6 1} for Nc = 1024 and 2048 and {i R : 2 6 i 6 2}

for Nc = 512. To reduce the CFO search points, a step based on f was employed
and can be formulated as f /8 in the case of Nc = 1024 and 2048, whereas the step
was f /16 for Nc = 512. Fig. (5.15)(c) illustrates how the CFO changes within the
duration of packets 6 and 8. It can be inferred that there is no relation that governs
the change in the residual Doppler shift between packet 6 and 8; the CFO changes
randomly but sometimes constant. The residual Doppler shift is depending on the
accuracy of estimating the re-sampling factor.

5.4

Chapter Summary

The focus of the chapter was on the Doppler eect caused by the acceleration due to
the relative motion between the transmitter and receiver. This acceleration aects
the correlation behaviour of the cyclic prex and destroys the orthogonality of the
sub-carriers due to the synchronization impairments. The proposed system is assessed through simulations at dierent scenarios and at dierent channel conditions
to imitate the realistic case. Additionally, the suggested method is investigated with
an o-line data that was recorded and processed from an experiment at the North
Sea. This chapter presented a technique to tackle the eect of a broadband time
varying Doppler shift in UWC. This technique adopted a learning and punishment
approach to iteratively estimate the Doppler shift parameters. These parameters
were estimated by the cooperation of a two point estimation of the normalized correlation of the rst order moment in addition to the linear prediction of the speed
change. This method is very robust when the relative velocity is changing linearly
and capable of dealing with the velocity inection. The suggested method is capable

122

5.4 Chapter Summary

of tackling an acceleration up to 1 m/s2 during the packet time and correcting a
speed up to 3 m/s.

123

Chapter 6
Conclusion and Future Work
Single carrier receivers for UWA that adopt high complexity algorithms such as
beamformer achieve reliable communication over severe channel conditions. However, these types of receivers are highly complex. An alternative multicarrier receiver
in the form of OFDM requires less complexity as far as real-time implementation is
concerned. Furthermore, such types are attractive to combat channel impairments.
This project was undertaken to design Doppler shift compensation algorithms for
a single element multi-carrier receiver that would be applicable in real-time implementation and to evaluate the performance of utilizing the COFDM in the presence
of this Doppler shift.
The present study, however, proposed several noteworthy contributions in order
to track and compensate the time-varying Doppler shift. In order to accommodate
this change, it was assumed that the speed is changing linearly within the OFDM
symbol; therefore, a linear equation model was proposed to govern the rst order
Doppler shift. This linear prediction technique reinforces the cyclic prex based
Doppler shift estimation and an improvement in the BER was obtained. However,
it was shown that as the acceleration increases, the accuracy of the Doppler shift
estimation decreases, thus necessitating an advanced signal processing technique in
order to tackle this challenge.
Therefore, an adaptive iterative receiver is suggested in this thesis to deal with
time-varying acceleration during a period of 5.4 s. The suggested algorithm is capable of tackling the induced linear Doppler shift variation due to the acceleration
and deceleration. In addition, the inection point between them was dealt with by
adopting a control range, which is derived based on the measured acceleration of

124

the previous two OFDM symbols and on the linearity of the speed change assumption. It has been shown through an extensive simulation and an experimental trial
that the suggested algorithm is robust in the presence of a time-varying multipath
channel and it can achieves an acceptable error performance.
The performance of the proposed Doppler compensation scheme was compared
with a block-based technique, in both cases using a COFDM-based receiver. It was
shown that the performance of the proposed receiver depends to a large extent on
the delay spread of the channel and on the acceleration. The Doppler estimator
accuracy aects the compensation of the residual Doppler shift which causes ICI
and, in turn, aects the channel estimation, resulting in a reduction in the eciency
of decoding due to the resultant unreliable LLRs values. It was shown that weighting
coecients improve the Doppler shift estimation and accommodate the change in the
speed between the OFDM symbols. The performance of the cyclic prex correlation
in estimating the Doppler shift depends on the range. Reducing the range entails
increasing in the ISI and ultimately aects the correlation accuracy. Hence, adopting
the weighting coecients reduces the noise and channel eect on the Doppler shift
estimation and a ne tuned re-sampling factor is obtained. The reason for the
improvements is the weighting coecients act as a smoothing lter which improves
the estimation. The promising feature of the proposed receiver is its capability of
delivering acceptable performance and a high data rate is achieved even with a tight
sub-carrier spacing.
Although the research emphasis was to design a paradigm modem that is applicable for DSP-based real-time implementation, the eect of Doppler shift on the
performance of the iterative receiver was also investigated. It is more pragmatic to
consider the acceleration eects on the estimation of the Doppler shift; therefore, an
adaptive compensation technique for time-varying was utilized to achieve reliable
communication. It has been shown that with a linear velocity change, adopting
learning and punishment approach provides a robust error performance. Results
of simulations and experiments at various channel ranges reveal that the proposed
adaptive time-varying technique is capable of tracking the Doppler shift with an
acceleration range of 1 m/s2 during the packet time and correcting a speed up to
3 m/s.

125

Future Work
It is recommended that further research be undertaken in the following areas:
1. Robust signal processing techniques are required to combat the Doppler eect
at each path to deliver acceptable performance.
2. It is suggested that the association of the channel parameters is investigated in
future studies, particularly the delay spread to improve predicting the Doppler
shift.
3. Developing a channel model that fulls the state of the art research in this
eld would be of great help in designing better algorithms.
4. Generalized deriving weighting coecients adaptively.
5. Further experimental investigations are needed in order to consider more realistic model of the velocity with higher acceleration.
6. In the DSP-based real-time implementation, this research concentrated primarily on the decoding stage, which is represented by BCJR with iterative
decoding, because it was the most restrictive in achieving on-line processing.
Therefore, involving the rest of the system for the hardware implementation
would be useful for future work in order to compare the multi-carrier system
with a single carrier one in terms of performance and hardware implementation, where the code is written in c language.

126

References
[1] B. Kim and I. Lu., Parameter study of ofdm underwater communications
system, in Proc. IEEE Oceans, Providence, RI, Sep. 2000, pp. 12511255.
[2] R. J. Urick, Principles of Underwater Sound. McGraw-Hill, Inc., 3rd edition,
1983.
[3] L. Brekhovskikh and Y. Lysanov, Fundamentals of Ocean Acoustics. Springer,
1982.
[4] R. S. H. Istepanian and M. Stojanovic, Underwater Acoustic Digital Signal
Processing and Communication Systems. Kluwer Academic Publisher, 2002.
[5] A. C. Singer, J. K. Nelson, and S. S. Kozat, Signal processing for underwater
acoustic communications, IEEE Commun. Mag., vol. 47, pp. 9096, Jan.
2009.
[6] M. Stojanovic, Recent advances in high-speed underwater acoustic communications, IEEE J. Ocean. Eng., vol. 21, pp. 125136, Apr. 1996.
[7] , Low-complexity ofdm detector for underwater channels, in Proc. IEEE
Oceans, Sep. 2006, pp. 1821.
[8] B. S. Sharif, J. Neasham, O. Hinton, and A. E. Adams, A computationally
ecient doppler compensation system for underwater acoustic communications, IEEE J. Ocean. Eng., vol. 25, No. 1, pp. 5261, Jan. 2000.
[9] J. G. Proakis, Digital Communications. McGraw Hill, Fifth edition, 2008.
[10] J. A. Neasham et al., Combined equalization and beamforming to achieve
20 kbits/s acoustic telemetry for rovs, in Proc. IEEE Oceans, Oct 1993, pp.
988993.

127

REFERENCES
[11] B. S. Sharif, J. Neasham, O. Hinton, A. E. Adams, and J. Davies, Adaptive
doppler compensation for coherent acoustic communication, IEE Proceedings
of Radar, Sonar and Navig., vol. 147, No. 1, pp. 239246, Oct. 2000.
[12] B. Li, S. Zhou, M. Stojanovic, L. Freitag, and P.Willett, Multicarrier communication over underwater acoustic channels with nonuniform doppler shifts,
IEEE J. Ocean. Eng., vol. 32, No.(2), pp. 198209, 2008.
[13] S.Mason, C.Berger, S. Zhou, and P. Willett, Detection, synchronization, and
doppler scale estimation with multicarrier waveforms in underwater acoustic
communication, IEEE Journal on Selected Area on Communications, vol. 26,
No.(9), pp. 16381649, Dec. 2008.
[14] J. van de Beek, M. Sandell, and P. Borjesson, Ml estimation of time and
frequency oset in ofdm systems, IEEE Transactions on Signal Processing,
vol. 45, No. 7, pp. 18001805, Jul. 1997.
[15] T. Kang and R. Iltis, Fast-varying doppler compensation for underwater
acoustic ofdm systems, in Proc. IEEE Asilomar Conf. on Signals, Systems
and Computers, 2008, pp. 933 937.
[16] J. Armstrong, Analysis of new and existing methods of reducing intercarrier
interference due to carrier frequency oset in ofdm, IEEE Trans. Comm.,
vol. 47, pp. 365 369, mar. 1999.
[17] S. Ahmed and H. Arslan, Estimation and compensation of doppler eect
using variable sub-carrier spacing in multiband uac ofdm systems, in Proc.
IEEE Oceans, Sep. 2010, pp. 14.
[18] A. B. Salberg and A. Swami, Doppler and frequency-oset synchronization
in wideband ofdm, vol. 4(6), pp. 28702881, Nov. 2005.
[19] A. Y. Kibangou, C. Siclet, and L. Ros, Joint channel and doppler estimation for multicarrier underwater communications, in Proc. IEEE ICASSP,
MARCH 2010, pp. 56305633.
[20] P. et. al, Symbol by symbol doppler rate estimation for highly mobile underwater ofdm, in Proc. WUWNET, Nov. 2009.

128

REFERENCES
[21] S. Yerramalli and U. Mitra, Optimal resampling of ofdm signals for multiscale
multilag underwater acoustic channels, IEEE J. Ocean. Eng., vol. 36, No.(1),
pp. 126138, Jan. 2011.
[22] G. S. Howe, P. D. Tarbit, O. Hinton, B. S. Sharif, and A. E. Adams, Sub-sea
acoustic remote communications utilising an adaptive receiving beamformer
for multipath suppression, in Proc. IEEE Oceans, 1994, pp. 313316.
[23] C. P. Shah, C. C. Tsimenidis, B. S. Sharif, and J. A. Neasham, Low complexity iterative receiver design for shallow water acoustic channels, EURASIP
Journal on Advances in Signal Processing, vol. 2010, Article ID 590458,13
pages, 2010.
[24] M. Stojanovic, J. Catipovic, and J. Proakis, Phase-coherent digital communication for underwater acoustic channels, IEEE J. Ocean. Eng., vol. 19, pp.
100111, Jan. 1994.
[25] M. Johnson, D. Herold, and J. Catipovic, The design and performance of a
compact underwater acoustic network node, in Proc. IEEE Oceans, 1994, pp.
467471.
[26] M. Stojanovic and L. Freitag, Multichannel detection for wideband underwater acoustic cdma communications, IEEE J. Ocean. Eng., vol. 31, No.(3),
pp. 685 695, 2006.
[27] M. Stojanovic, Design and capacity analysis of cellular-type underwater
acoustic networks, IEEE J. Ocean. Eng., vol. 33, No.(2), pp. 171 181, 2008.
[28] L. Linton, P. Conder, and M. Faulkner, Multiuser communication for underwater acoustic networks using mimo-ofdm-idma, in Proc. International
Conference on signal Processing and Communication systems ICSPCS, Dec.
2008, pp. 18.
[29] J. H. Z. Yan and C. He, Implementation of an ofdm underwater acoustic
communication system on an underwater vehicle with multiprocessor structure, Frontiers of Electrical and Electronic Engineering, China, vol. 2, No. 2,
pp. 151 155, Jan. 2007.

129

REFERENCES
[30] M.R.Solaimani, Y. Gao, and U. Vilaipornsawai, Turbo Coding for satellite and
wireless Communications. Springer, 2002.
[31] L. Kaizhuo, Y. Zhenhua, H. Jing, and H. Jianguo, Design and implementation
of underwater ofdm acoustic communication transmitter, in ICALIP, July
2008, pp. 609 613.
[32] R.Kothandaraman and M. Lopez, An ecient implementation of turbo decoding on adi tigersharc ts201 dsp, in IEEE SPCOM International Conference, 2004, pp. 344 348.
[33] H. Yan, S. Zhou, Z. Shi, and B. Li, A dsp implementation of ofdm acoustic
modem, in Proc. WUWNet, September 14 2007.
[34] H. Yan, L. Wan, S. Zhou, Z. Shi, J. Cui, J. Huang, and H. Zho, Dsp based receiver implementation for ofdm acoustic modems, Elsevier Journal on Physical Communication, vol. 5, No. 1, pp. 22 32, Sep. 2011.
[35] H. Yan, S. Zhou, Z. Shi, J. Cui, L. Wan, J. Huang, and H. Zho, Dsp implementation of siso and mimo ofdm acoustic modem, in Proc. IEEEOcean,
May 2010.
[36] K. Zhong, S. Q. Swee, A. K. Tiong, and A. T. Bien, A real-time coded ofdm
acoustic modem in very shallow underwater communications, in Proc. IEEE
Oceans, Asia-Pacic, May 2007, pp. 1 5.
[37] S. M. Kuo and B. H. Lee, Real-time digital signal processing implementations,
applications and experiments with the TMS320C55X.

John wiley and sons,

LTD, 2001.
[38] J. G. Proakis, Coded modulation for digital communications over rayleigh
fading channels, IEEE J. Ocean. Eng., vol. 16, pp. 6673, Jan. 1991.
[39] G. Ungerboeck, Channel coding with multi-level/phase signals, IEEE Trans.
Inf. Theory, vol. 28, pp. 5567, Jan. 1982.
[40] D. Divsalar and M. K. Simon, The design of trellis coded mpsk for fading
channels: performance criteria, IEEE Trans. Commun., vol. 36, pp. 1004
1012, Sep. 1988.
130

REFERENCES
[41] E. Zehavi, 8-psk trellis codes for a rayleigh fading channel, IEEE Trans.
Commun., vol. 40, pp. 873883, May 1992.
[42] E. M. Sozer, J. G. Proakis, and F. Blackmon, Iterative equalization and decoding techniques for shallow water acoustic channels, in Proc. IEEE Oceans,
Nov. 2004, pp. 22012208.
[43] S. L. Go, A. Glavieux, and C. Berrou, Turbo codes and high spectral efciency modulation, in Proc. IEEE ICC, New Orleans, LA, May 1994, pp.
645649.
[44] S. L. Go, K. B. Kien, C. C. Tsimenidis, and B. S. Sharif, Constellation shaping for bandwidth-ecient turbo-coded modulation with iterative receiver,
IEEE Wireless Communications, vol. 6, pp. 22232233, June 2007.
[45] F. Schreckenbach, N. Gortz, J. Hagenauer, and G. Bauch, Optimized symbol
mappings for bit-interleaved coded modulation with iterative decoding, in
Proc. IEEE Globecom Conference, Dec. 2003, pp. 33163320.
[46] X. Li, A. Chindapol, and J. A. Ritcey, Bit-interleaved coded modulation with
iterative decoding and 8 psk signalling, IEEE Transactions on Communication, vol. 50, No.(5), pp. 12501257, Aug. 2002.
[47] K. Pelekanakis, Exploiting space-time-frequency diversity with mimo-ofdm
for underwater acoustic communications, IEEE J. Ocean. Eng., vol. 36,
No.(4), pp. 502513, Oct. 2011.
[48] T. Jiang and Y. Wu, An overview: peak-to-average power ratio techniques
for ofdm signals, IEEE Transactions on Broadcasting, vol. 54, No. 2, pp.
257268, June 2008.
[49] S. Han and J. Lee, An overview of peak-to-average power ratio reduction
techniques for multicarrier tranmission, IEEE Wireless Communications, vol.
12, No. 2, pp. 5665, Feb. 2005.
[50] R. W. Bauml, R. F. H. Fisher, and J. B. Huber, Reducing the peak-to-average
power ratio of multicarrier modulation by selected mapping, Electronic Letter, vol. 32, No. 22, pp. 20562057, 1996.

131

REFERENCES
[51] A.

Devices,

2009.

Digital

[Online].

pulse-shaping

Available:

lter

basics

AN-922,

Jun.

http://www.analog.com/en/processors-

dsp/sharc/processors/application-notes/resources/index.html
[52] P. Tan and B. N.C, A novel pulse-shaping for reduced ici in ofdm systems,
in Proc. IEEE VTC, Sep. 2004, pp. 456 459.
[53] B. Muquet, Z. Wang, G. Giannakis, and M. Courville, Cyclic prexing or
zero padding for wireless multicarrier transmission? IEEE Transactions on
Communications, vol. 50, pp. 21362148, Dec. 2002.
[54] A. Peled and A. Ruiz, Frequency domain data transmission using reduced
computational complexity algorithms, in Proc. IEEE ICASSP, 1980, pp. 964
967.
[55] W. S. Burdic, Underwater Acoustic System Analysis. Prentice Hall, 1991.
[56] J. A. Catipovic, Performance limitations in underwater acoustic telemetry,
IEEE J. Ocean. Eng., vol. 15, pp. 205216, 1990.
[57] R. F. W. Coates, Underwater acoustic systems. Macmillan, 1990.
[58] F. Hlawatsch and G. Matz, Wireless Communications Over Rapidly TimeVarying Channels. Elsevier Ltd., 2011.
[59] A. F. Molisch, Wireless Communications. John Wiley, 2005.
[60] G. Caire, G. Taricco, and E. Biglieri, Bit-interleaved coded modulation,
IEEE Trans. Inf. Theory, vol. 44, pp. 927946, May 1998.
[61] A. Goldsmith, Wireless Communications. Cambridge University Press, 2005.
[62] X. Li and J. A. Ritcey, Bit-interleaved coded modulation with iterative decoding, IEEE Commun. Lett., vol. 1, pp. 169171, Nov. 1997.
[63] A. Chindapol and J. A. Ritcey, Design, analysis, and performance evaluation
for bicm-id with square qam constellations in rayleigh fading channels, IEEE
J. Sel. Areas Commun., vol. 19, pp. 944957, May 2001.
[64] X. Li and J. A. Ritcey, Trellis-coded modulation with bit interleaving and iterative decoding, IEEE Journal on Selected Area on Communications, vol. 17,
pp. 715724, Apr. 1999.
132

REFERENCES
[65] , A soft-input soft-output app module for iterative decoding of concatenated code, Electronic Letter, vol. 34, pp. 942943, May 1998.
[66] L. Hanzo, C. H. Wong, and M. S. Yee, Adaptive Wireless Transceivers: TurboCoded, Turbo-Equalized and Space-Time Coded TDMA, CDMA, and OFDM
Systems. Wiley-IEEE Press, 2002.
[67] S. ten Brink, J. Speidel, and R. Yan, Iterative demapping and decoding for
multilevel modulation, in Proc. IEEE Globecom Conference, Nov. 1998, pp.
579584.
[68] L. Bahl, J. Cocke, F. Jelinek, and J. Raviv, Optimal decoding of linear codes
for minimizing symbol error rate, IEEE Trans. Inf. Theory, vol. 20, pp. 284
287, Mar. 1974.
[69] J. P. Woodard and L. Hanzo, Comparative study of turbo decoding techniques: An overview, IEEE Journal on Vehicular Technology, vol. 49, pp.
22082233, Nov. 2000.
[70] H. Malepati, Digital media processing DSP algorithms using c. Elsevier Ltd.,
2010.
[71] B. Sklar, Digital communications Fundamentals and Applications.

Prentice

Hall P T R, Second edition, 2001.

[72] A. S. Alvarado, L. Agrell, and E. Svensson, On bicm-id with multiple interleavers, in 14th European Wireless Conf., Sep. 2010.
[73] X. Zou, M. Wang, and G. Feng, A new interleaver design for iteratively
decoded bit-interleaved coded modulation, IJSC, vol. 3, pp. 338343, 2008.
[74] S. Smith, The Scientest and Engineers Guide to Digital Signal Processing.
Analog Device, 1998.
[75] W. H. Park, M. H. Sunwoo, and S. K. Oh, Ecient dsp architecture for
viterbi decoding with small trace back latency. IEICE Transactions.
[76] P. Lapsley and G. Blalock, How to estimate dsp processor performance,
IEEE Spectrum Magazine, vol. 33, No. 7, pp. 7478, Jul. 19962.

133

REFERENCES
[77] T. Pollet, P. Spruyt, and M. Moeneclaey, The ber performance of ofdm systems using non-synchronized sampling, in Proc. IEEE Globecom Conference,
1994, pp. 253257.
[78] J. B. et al., Weighted centroid localization in zigbee-based sensor networks,
in IEEE WISP, 2007, pp. 16.
[79] F. Reichenbach and D. Timmermann, Indoor localization with low complexity in wireless sensor networks, in Proc. IEEE International Conf.Industrial
Informatics, Aug. 2006, pp. 10181023.
[80] A. Jadbabaie, J. Lin, and A. Morse, Coordination of groups of mobile autonomous agents using nearest neighbor rules, IEEE Transactions on Automatic Control, vol. 48, No. 6, pp. 9881001, June 2003.
[81] F. S. Cattivelli, C. Lopes, and A. H. Sayed, Diusion recursive least-squares
for distributed estimation over adaptive networks, IEEE Transactions on
Signal Processing, vol. 56, No. 5, pp. 1865 1877, May 2008.
[82] B. Li, S. Zhou, J. Huang, and P. Willett, Scalable ofdm design for underwater
acoustic communications, in Proc. IEEE ICASSP, Las Vegas, 2008, pp. 5304
5307.
[83] A. Devices, Adsp-21364 ez-kit lite evaluation system manual, Jul. 2007.
[Online]. Available: http://www.analog.com
[84] Y. Or, G. Kutz, A. Chass, A. Gubeskys, and E. Pollak, Iterative decoding
algorithms for real time software implementation in wireless communication
systems, in IEEE Conf.,VTS, Jun. 2001, pp. 18841888.
[85] B. Li, S. Zhou, M. Stojanovic, L. Freitag, and P. Willett, Non-uniform doppler
compensation for zero-padded ofdm over fast-varying underwater acoustic
channels, in Proc. IEEE Ocean, Aberdeen, 2007, pp. 16.
[86] M. Speth, S. Fechtel, G. Fock, and H. Meyr, Optimum receiver design for
wireless broad-band systems using ofdm-part ii, IEEE Trans. Comm., vol.
49(4), pp. 571578, 2001.
[87] B. Yang, K. B. Letaief, R. Cheng, and Z. Cao, Timing recovery for ofdm
transmission, IEEE SA. Comm., vol. 49, pp. 467479, Nov. 2000.
134

REFERENCES
[88] M. K. Ozdemir, L. B. wireless solutions INC., and H. Arslan, Channel estimation for wireless ofdm systems, IEEE Comm. surveys, vol. 9, No. 2, 2nd
quarter 2007.
[89] V. de Beek, J. J., O. Edfords, and M. Sandell, On channel estimation in ofdm
system, in Proc. IEEE VTC, 1995, pp. 815819.
[90] S. Coleri, M. Ergen, A. Puri, and A. Bahai, Channel estimation techniques
based on pilot arrangement in ofdm systems, IEEE Transactions on Broadcasting, vol. 48, No. 3, pp. 223229, Sep. 2002.
[91] M. Hsieh and C. Wei, Channel estimation for ofdm systems based on combtype pilot arrangement in frequency selective fading channels, IEEE Transactions on Consumer Electronic, vol. 44, pp. 217225, Feb. 1998.
[92] C. P. Shah, C. C. Tsimenidis, B. S. Sharif, and J. A. Neasham, Lowcomplexity iterative receiver structure for time-varying frequency-selective
shallow underwater acoustic channels using bicm- id:design and experimental results, IEEE J. Ocean. Eng., vol. 36, pp. 406421, 2011.
[93] A. E. Abdelkareem, B. S. Sharif, C. C. Tsimenidis, J. A. Neasham, and
O. Hinton, Low-complexity doppler compensation for ofdm-based underwater acoustic communication systems, in Proc. IEEE Oceans, Jun. 2011, pp.
16.
[94] A. E. Abdelkareem, B. S. Sharif, C. C. Tsimenidis, and J. A. Neasham, Time
varying doppler-shift compensation for ofdm-based shallow underwater acoustic communication systems, in Proc. IEEE MASS, Oct. 2011, pp. 885891.
[95] S. K. Mitra, Digital signal processing A computer based approach. McGrawHill, 3rd edition,2006.
[96] A. H. Sayed, Adaptive Filters. Wiley-IEEE Press, 2008.
[97] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions with
Formulas, Graphs, and Mathematical Tables. New York: Dover,, 1972.
[98] W. Kim and D. C. Cox, Residual frequency oset and phase compensation
for ofdm systems, in Proc. IEEE VTC, 2007, pp. 22092213.
135

REFERENCES
[99] R. Otnes, T. Jenserud, J. E. Voldhaug, and C. Solberg, A roadmap to ubiquitous underwater acoustic communications and networking, in 3rd UAM:
Technologies and Results, Nafplion, Greece, Jun. 2009.
[100] A. Zielinski, Y. H. Yoon, and L. Wu, Performance analysis of digital acoustic
communication in a shallow water channel, IEEE J. Ocean. Eng., vol. 20, pp.
293299, Oct. 1995.

136