Digital Signal Processing

Digital Signal
Processing
Concepts and Applications

Bernard Mulgrew, Peter Grant and

John Thompson
Department of Electronics and Electrical Engineering
The University of Edinburgh
© Bernard Mulgrew, Peter M. Grant and John S. Thompson 1999

All rights reserved. No reproduction, copy or transmission of this

publication may be made without written permission.
No paragraph of this publication may be reproduced, copied or
transmitted save with written permission or in accordance with the
provisions of the Copyright, Designs and Patents Act 1988, or under
the terms of any licence permitting limited copying issued by the
Copyright Licensing Agency, 90 Tottenham Court Road,
London WIP 9HE.
Any person who does any unauthorised act in relation to this
publication may be liable to criminal prosecution and civil claims
for damages.

The authors have asserted their right to be identified as the authors

of this work in accordance with the Copyright, Designs and Patents
Act 1988.

First published 1999 by

MACMILLAN PRESS LTD
Houndmills, Basingstoke, Hampshire RG21 6XS
and London
Companies and representatives throughout the world

ISBN 978-0-333-74531-1 ISBN 978-1-349-14944-5 (eBook)

DOI 10.1007/978-1-349-14944-5
A catalogue record for this book is available from the British Library.

This book is printed on paper suitable for recycling and made from
fully managed and sustained forest sources.

10 9 8 7 6 5 4 3 2 1
08 07 06 05 04 03 02 01 00 99
Contents

Preface, x
Acknowledgements, xii
Abbreviations, xiii
Principal symbols, xvi
Special functions, xx
Introduction, xxi

1 Signal representation and system response, 1

1.1 Introduction, 1
1.2 Signal classification, 3
1.2.1 Energy signals, 3
1.2.2 Power signals, 4
1.3 Fourier series, 5
1.4 The Fourier transform, 13
1.5 Laplace transform, 18
1.6 Transform analysis oflinear systems, 21
1.6.1 Superposition, 21
1.6.2 Linear ordinary differential equations, 22
1.6.3 Response of linear system to a periodic input, 23
1.6.4 General approach, 26
1.7 Transfer function, 27
1.8 Summary, 30
1.9 Problems, 30

2 Time-domain description and convolution, 33

2.1 Introduction, 33
2.2 The impulse response, 33
2.2.1 The impulse, 34
2.2.2 Signal representation, 35
2.2.3 System response to an impulse, 36
2.3 Convolution, 39
2.4 Properties of convolution, 45
2.4.1 Time delay, 47
2.5 Summary, 53
2.6 Problems, 53
vi Contents

3 Transfer function and system characterisation, 56

3.1 Introduction, 56
3.2 Transfer function, poles and zeros, 56
3.3 Transfer function and frequency response, 58
3.3.1 Frequency response from pole/zero diagram, 59
3.3.2 Fourier transform of periodic signals, 61
3.3.3 Measurement offrequency response, 64
3.3.4 Bode plots, 65
3.3.5 Fourier and Laplace, 69
3.4 Transfer function and impulse response, 71
3.5 Time-domain response of first and second order systems, 73
3.5.1 First order systems, 74
3.5.2 Second order systems, 75
3.6 Rise time and bandwidth, 79
3.7 Summary, 81
3.8 Problems, 82

4 Sampled data systems and the z-transform, 85

4.1 Introduction, 85
4.2 Sampled data systems and aliasing, 89
4.2.1 Sampling theorem - the Nyquist criterion, 98
4.2.2 Practical sampled data systems, 98
4.3 The z-transform, 99
4.3.1 The inverse z-transform, 103
4.3.2 Delay theorem, 105
4.4 Digital filters and discrete convolution, 106
4.4.1 Discrete convolution, 109
4.5 Poles and stability, 112
4.6 Frequency response of a digital filter, 113
4.7 Example of a complete system, 117
4.8 Summary, 122
4.9 Problems, 123

5 Infinite impulse response digital filters, 126

5.1 Introduction, 126
5.2 Analogue prototype filters, 127
5.2.1 Introduction, 127
5.2.2 Butterworth prototype polynomials, 128
5.2.3 Chebyshev prototype polynomials, 130
5.3 Digital filter structures, 132
5.3.1 Introduction, 132
5.3.2 The canonicalform, 133
5.3.3 Parallel and cascade realisations, 136
5.4 Filter design methods, 137
5.4.1 Introduction, 137
5.4.2 The bilinear z-transform, 138
 Contents Vll

5.4.3 Filter transformation, 143

5.5 Finite precision effects, 145
5.5.1 Filter coefficient quantisation errors, 145
5.5.2 Limit cycles, 147
5.5.3 IIRfilter hardware, 147
5.6 Summary, 148
5.7 Problems, 148

6 Finite impulse response digital filters, 150

6.1 Introduction, 150
6.2 Finite theory and frequency response, 150
6.2.1 Transfer junction, 150
6.2.2 Frequency response, 151
6.2.3 Phase response, 152
6.3 Linear phase filters, 152
6.3.1 Principles, 152
6.3.2 Linear and nonlinear phase filters, 155
6.4 Linear phase filter design, 156
6.4.1 Fourier series method, 156
6.4.2 Window effects, 159
6.4.3 Design summary, 163
6.4.4 Design optimisation techniques, 165
6.5 Finite precision effects, 168
6.5.1 Noise reduction through the filter, 168
6.5.2 Filter coefficient quantisation errors, 169
6.5.3 FIRfilter hardware, 172
6.6 FIR filter applications, 172
6.6.1 Matched filter detector, 172
6.6.2 Matched filter applications, 173
6.6.3 Other receiver designs, 174
6.7 Summary, 174
6.8 Problems, 175

7 Random signal analysis, 176

7.1 Introduction, 176
7.2 Random processes, 176
7.3 Averages and spectral representations, 181
7.3.1 Autocorrelation and autocovariance, 183
7.3.2 Correlation and dependence, 185
7.3.3 Power spectral density, 185
7.3.4 Alternative representations of a random process, 187
7.4 Random signal and discrete linear systems, 189
7.4.1 Cross-correlation between the input and output of a filter, 191
7.5 Spectral factorisation, inverse and whitening filters, 192
7.5.1 Inverse filters, 194
7.5.2 Noise whitening, 195
Vlll Contents

7.5.3 Cross-correlation between two filter outputs, 196

7.6 Filter noise calculations, 197
7.6.1 Quantisation noise, 198
7.6.2 Dynamic range, 200
7.7 Summary, 203
7.8 Problems, 203

8 Adaptive filters, 206

8.1 Introduction, 206
8.2 Wiener filters, 207
8.2.1 Wiener FIR filter, 208
8.2.2 Application to channel equalisation, 211
8.3 Algorithms for adaptive filtering, 215
8.3.1 Recursive least squares, 217
8.3.2 Stochastic gradient methods, 219
8.3.3 A comparison of algorithms, 227
8.4 Applications, 230
8.4.1 Adaptive line enhancement, 231
8.4.2 Adaptive tone suppression, 233
8.4.3 Noise whitening, 233
8.4.4 Echo cancellation, 234
8.4.5 Channel equalisation, 235
8.5 Summary, 238
8.6 Problems, 238

9 The Fourier transform and spectral analysis, 240

9.1 Development of the discrete Fourier transform, 240
9.1.1 The continuous Fourier transform, 240
9.1.2 Fourier transform of a finite length data record, 241
9.1.3 Definition of the DFT, 243
9.1.4 Properties of the DFT, 245
9.2 Computation of the discrete Fourier transform, 246
9.2.1 DFT matrix coefficient values, 246
9.2.2 Matrixformulation of the DFT, 246
9.2.3 Analogiesfor the DFT, 251
9.3 Resolution and window responses, 253
9.3.1 Resolution, 253
9.3.2 Leakage effects, 253
9.3.3 The rectangular window, 254
9.3.4 Hanning window, 257
9.3.5 Hamming window, 259
9.3.6 The Dolph-Chebyshev window, 259
9.3.7 Window comparisons, 261
9.4 Fundamentals of spectral analysis, 262
9.5 Classical spectral analysis, 264
9.6 Modem spectral analysis, 267
 Contents ix

9.6.1 Introduction to parametric techniques, 267

9.6.2 Autoregressive spectrum analysis, 268
9.7 Comparison of spectral analysis techniques, 272
9.8 Application of AR techniques in speech coders, 273
9.9 Summary, 276
9.10 Problems, 276

10 The fast Fourier transform, 278

10.1 Introduction, 278
10.2 Partitioning of the DFT into two half-size matrices, 278
10.3 Radix-2 FFT, 284
10.3.1 Decimation-in-time algorithm, 284
10.3.2 Decimation-in-frequency algorithm, 288
lOA Implementation considerations, 288
1004.1 Complex multiplier hardware, 289
1004.2 Alternative radix arithmetic approaches, 289
1004.3 Real valued data, 290
100404 Inverse transforms, 293
10.5 Applications, 294
10.6 Summary, 297
10.7 Problems, 297

11 Multirate signal processing, 299

11.1 Introduction, 299
11.2 Decimation, interpolation, imaging and aliasing, 300
11.2.1 Decimation, 303
11.2.2 Bandpass sampling, 306
11.2.3 Interpolation, 310
11.3 Applications of multirate systems, 311
11.3.1 Transmultiplexers, 311
11.3.2 Analysis and synthesis filterbanks, 311
11.3.3 Filterbank design approaches, 314
1104 Audio and speech processing, 319
1104.1 Speech and image signal coding, 321
11.5 Summary, 325
11.6 Problems, 325

Appendix A - Matrix theory revision, 327

Appendix B - Signal transforms, 330
Solutions to self assessment questions, 332
Bibliography, 348
Index, 354
Preface

Digital signal processing (DSP) provides a rapidly advancing portfolio of filtering and
estimating techniques or algorithms which are used in signal analysis and processing.
Significant current applications are in the development of mobile communications
equipment, particularly for personal use and design of sophisticated radar systems. The
aim of this book is to provide an introduction to the fundamental DSP operations of fil-
tering, estimation and signal analysis as used in signal processing.
Most of the chapters include substantive numerical examples to illustrate the
material developed, as well as self assessment questions which have been designed to
help readers aid their comprehension of this material. All the chapters conclude with
further problem questions for the student.
Chapters 1 to 3 cover basic analogue signal theory as a prerequisite to this DSP
text. Chapter 4 extends these concepts to sampled-data systems and here discrete con-
volution is introduced. Chapters 5 and 6 explore digital filters, both infinite and finite
impulse response, which implement the convolution operation and include both analyti-
cal design and software optimisation techniques. The two chapters conclude with a
brief discussion on the problems of finite precision arithmetic. Chapters 7 and 8 intro-
duce, at a more mathematical level, the concept of random signals, correlation and
spectral density. Chapter 8 covers adaptive or self learning filters which alter their char-
acteristics dependent on the signal scenario which is present. They find widespread
application in communications systems as equalisers and echo cancellers. The final
three chapters deal with spectral analysis techniques. Chapter 9 covers the discrete
Fourier transform (DFT), its derivation and the design of DFT processors. This chapter
then investigates the application of DFT processors in classical spectrum analysis
equipment before introducing the modern analysers which are based on the adaptive fil-
ter technique of Chapter 8. Chapter 10 deals with the fast Fourier transform which is
the most widely applied implementation of the DFT processing function. Chapter 11
introduces multirate techniques to extend the capabilities of these analysers to speech
and image processing applications.
With this balance between signal theory, processor design and systems applica-
tions we hope that this text will be useful both in academia and in the rapidly growing
commercial signal processing community. Advanced DSP is of fundamental impor-
tance in implementing many high performance systems such as personal communica-
tions and radar.
To aid the class instructor, the authors can provide a printed set of outline solu-
tions to the end-of-chapter problems and these are available via the WWW on password
access. MATLABTM source code is also provided on an open access basis to assist the
 Preface Xl

instructor with presentation of the material and the student in understanding of the
material. In general the source code provides a computer animation of some of the fig-
ures in the book. For example the m-file Ilfig1_4.m" contains MATLAB code which
produces an animation of the complex phasor of Figure 1.4. These are identified within
the text by the g symbol. These software and solutions to problems are available via
the menu at the Edinburgh WWW server address:
http://www.ee.ed.ac.ukrpmg/SIGPRO/index.htmi. Subsequent corrections to this text
will also be available at the same WWW address.

Edinburgh Bernard Mulgrew, Peter Grant and John Thompson

May 1998
Acknowledgements

Parts of this book have been developed from BEng, MEng, MSc and industrial training
courses provided by the Department of Electronics and Electrical Engineering at the
University of Edinburgh. These courses were also taught by Professor Colin Cowan
and Dr James Dripps, and we acknowledge their contribution to the initial shaping of
these courses which is reflected in the book's content and structure. We are grateful to
Professor Cowan for having provided a draft version of Chapter 5 and Dr Dripps for
having provided a draft version of parts of Chapter 9 and for assistance with many of
the problem solutions. We are also grateful to Dr Ian Glover at the University of Brad-
ford and Prentice-Hall for permission to include the material on bandpass sampling
within Chapter 11, and to the lEE for permission to reproduce Figure 9.22.
We would like to thank all those other colleagues at the University of Edinburgh
who have provided detailed comments on sections of this text. Thanks must go to the
many students who have read and commented on earlier versions of this material and
helped to refine the end-of-chapter problems, particularly to Miss Oh who generated the
initial version of many of the diagrams. We also gratefully acknowledge the generous
assistance of Dr Jonathon Chambers of Imperial College in carefully reviewing and
editing this text. In addition we acknowledge the assistance of Philip Yorke at
Chartwell Bratt publishing and training in encouraging us to develop, in the 1980s, the
preliminary version of this material.
Special thanks are due to Joan Burton, Liz Paterson and Diane Armstrong for
their perseverance over several years in typing the many versions of the individual
chapters, as they have evolved into their current form. We also acknowledge Bruce
Hassall's generous assistance with the preparation of the final version of the text in the
appropriate typefont and text format.
Finally we must thank our respective families: Fiona and Maeve; Marjory, Lind-
say and Jenny; and Nadine - for the considerable time that we required to prepare this
book and the associated WWW supporting material.

Bernard Mulgrew, Peter Grant and John Thompson

Abbreviations

AC Alternating current (implying sinusoidal signal)

ACF Autocorrelation function
AID Analogue to digital (converter)
ADPCM Adaptive DPCM
AGC Automatic gain control
AM Amplitude modulation
AR Autoregressive

BLMS Block least mean squares

BP Bandpass
BPF Bandpass filter
BS Bandstop

CD Compact disc
CDMA Code division mUltiple access
CELP Codebook of excited linear prediction
CMOS Complementary metal oxide silicon (transistor)
COFDM Coded orthogonal frequency-division multiplex

D/A Digital to analogue (converter)

DC Direct current (implying a 0 Hz component)
DCT Discrete cosine transform
DFS Discrete Fourier series
DFT Discrete Fourier transform
DIP Decimation in frequency
DIT Decimation in time
DM Delta modulation
DPCM Differential pulse code modulation
DPSK Differential phase shift keying
DSB Double sideband
DSP Digital signal processing
DTFT Discrete-time Fourier transform

ESD Energy spectral density

XIV Abbreviations

EVR Eigenvalue ratio

FFT Fast Fourier transfonn

FIR Finite impulse response
FM Frequency modulation
FS Fourier series; Federal Standard
FSK Frequency shift keying
FT Fourier transfonn

HP High pass

I Imaginary (quadrature signal) component

IDFT Inverse DFT
IF Intennediate frequency
IIR Infinite impulse response
lSI Inter-symbol interference

KCL Kirchhoff's current laws

LMS Least mean squares

LO Local oscillator
LOS Line of sight
LP Low pass
LPC Linear predictive coding
LS Least squares
LTI Linear time invariant

MA Moving average
MAC Multiply and accumulate
MATLAB MATrix LABoratory commercial DSP software product
MFSK Multiple frequency shift keying
MMSE Minimum mean square error
MODEM Modulator/demodulator
MOS Mean opinion score (for speech quality assessment)
Metal oxide silicon (transistor)
MPE Multipulse excitation
MSE Mean square error

NATO North Atlantic Treaty Organisation

NPSD Noise power spectral density

PAM Pulse amplitude modulation

PCM Pulse code modulation
pdf Probability density function
PFE Partial fraction expansion
 Abbreviations xv

PLL Phase locked loop

PM Phase modulation
PN Pseudo-noise
PO Percentage overshoot
PPM Pulse position modulation
PR Perfect reconstruction
PSD Power spectral density
PSK Phase shift keying
PWM Pulse width modulation

Q Quantiser

R Real (in-phase signal) component

RAM Random access memory
ROM Read only memory
RLS Recursive least squares
RMS Root mean square

SAQ Self assessment question

SBC Sub-band coder
sa Stochastic gradient
SIR Sample and hold
SNR Signal-to-noise ratio

TDM Time division multiplex

THD Total harmonic distortion

VLSI Very large scale integrated (circuit)

WWW World Wide Web

ZOH Zero order hold

Principal symbols

a tap weight vector

Qi digital filter weight coefficient value for tap i
A A-law PCM compander constant
An nth trigonometric Fourier component
A(k) kth real Fourier coefficient

bi digital filter recursive weight coefficient value for tap i

B signal bandwidth in Hz
Bn nth trigonometric Fourier component
B(k) kth imaginary (real valued) Fourier coefficient

FIR filter coefficient value

windowed coefficient value
constant, capacitance (Farads)
Chebyshev polynomial of order n

d lag

e error vector
e(n) scalar error signal at data sample n
E energy of signal x(t)

10 centre frequency
11 passband cut-off frequency
i2 stopband edge frequency
hdB half power bandwidth
Ib bit rate
Ie centre frequency
IH highest frequency component
fL lowest frequency component
lLO local oscillator frequency
Is sample frequency (in Hz)
Fm(z) z-plane reconstruction filter m response
 Principal symbols xvii

speech formant N frequency component

frequency response

g(t) baseband signal

g(kTs) estimate of g(t)
g(kTs) prediction of g(t)
G amplifier gain
Gp processing gain

h(n) filter impulse response vector

h(n) filter impulse response for sample value n
h(t) filter impulse response
HnIH(s) system transfer function
HA(m) real frequency response
HD(m) required or desired frequency response
H(s) Laplace transfer function
H(z) z-transfer function
H(m) angular frequency response

lo[ ] modified Bessel function of first kind and order zero

k Boltzmann's constant
K constant

L upsampling ratio
L(m) weighting function

M number of feedback taps in an IIR filter; downsampling ratio

n type of semiconductor material

N orderofDFf
N number of feedforward taps in an IIRlFIR filter
No noise power spectral density

P type of semiconductor material

PI filter pole number 1
p(v) probability density function of variable v
P power

q quantisation step size

Q quality factor; fHIB

ryx cross-correlation vector

Ryy correlation matrix
R resistance in ohms
XVlll Principal symbols

RD dynamic range
RC resistor-capacitor time constant 'Cc

Laplace variable
power spectral density

t time
/).t sample period

eigenvector of autocorrelation matrix

information signal
voltage waveform of symbol i

FIR filterlDFf window or weight coefficient n

kth value of the N roots of unity
DFf of window function

x vector x
x(n) sampled input signal
fen) estimate of input signal
Xn complex Fourier coefficient
X(f/OJ) Fourier (voltage) spectrum of x(n)
IX(f/OJ)1 Fourier amplitude spectrum
X(k) DFf output value for bin or sample number k
X(s) Laplace transform of x(t)

yen) processed output signal

YD(n) decimated signal
y/(n) interpolated signal

filter zero number n

a attenuation; forgetting factor or window taper

p constant

Dirac delta
impulse function
impulse train
voltage difference
gradient vector

error voltage; Chebyshev design parameter

 Principal symbols XIX

quantisation error
mean-square quantisation error
filter damping factor
MSE cost function

TJ efficiency
TJ(n) additive noise signal

eigenvalue i

step-size scaling constant

p normalised correlation coefficient; sum of squared error cost function

p(n) vector norm or deviation

real part of Laplacian; standard deviation

variance of signal x

time constant

autocorrelation sequence x(n)

autocorrelation matrix for x(n)
cross-correlation vector

filter centre frequency

analogue filter cut-off frequency
lower bandpasslbandstop cut-off frequency
upper bandpasslbandstop cut-off frequency
digital filter cut-off frequency
(angular) sample frequency (in rad/s)
filter transition bandwidth, DFf bin spacing
Special functions

E[ ] statistical expectation operator

[ ,] scalar product
() time average
* convolution operation
a• complex conjugate of a
F[,] Fourier transform operator
g operator corresponding to imaginary part of ...
f(.) integer part of ...
L[,] Laplace transform operator
9t operator corresponding to real part of ...
sa (x) sampling function
sgn (x) signum (sign) function
sinc (x) sinc function
u(t) unit step function
Z-I inverse z transform
Z z transform
Introduction

Real life signals are generally continuous in time and analogue, i.e. they exist at all time
instances and can assume any value, within a predefined range, at these time instances.
There are many kinds of analogue signals appearing in nature:

Electrical signals: voltages, currents, electric and magnetic fields.

• Acoustic signals: mechanical vibrations, sound waves.
• Mechanical signals: displacements, angular motion, velocities, forces,
moments, pressures.

Acoustic signals such as sound waves are converted to electrical voltages or cur-
rents by sensors or transducers, (i.e. a microphone) in order for them to be processed in
an electronic system. Analogue processing involves linear operations such as amplifi-
cation, filtering, integration and differentiation, as well as various forms of nonlinear
processing such as squaring or rectification. This text does not cover the field of non-
linear signal processing, but, in practice, saturation in amplifiers or mixing of signals
often introduces nonlinearities. Limitations of practical analogue processing operations
are:

Restricted accuracy.
Sensitivity to noise.
• Restricted dynamic range.
Poor repeatability due to component variations with time, temperature, etc.
• Inflexibility to alter or adjust the processing functions.
• Problem in implementing accurate nonlinear and time-varying operations.
• Limited speed of operation.
• High cost of storage for analogue waveforms.

However, in the 1970s, analogue signal processing in the form of surface acoustic
wave and charge-coupled sampled-data devices delivered very sophisticated matched
filter and correlator parts which were widely used in military equipment. At this time it
was not possible to match the analogue speed/performance capability with digital
devices so the accuracy of analogue processors was severely limited. The initial imple-
mentations of Dolby noise reduction systems also employed analogue filter techniques
and the early Y.21-V.29 data modems, for transmission of digital data over telephone
lines, used exclusively analogue filters and modulators.
XXll Introduction

Digital signal processing (DSP) is achieved by sampling the analogue signal at

regular intervals and representing each of these sample values with a binary number.
These are subsequently passed to a specialised digital processor to perform numerical
or computational operations on these signals. Operations in digital signal processing
systems encompass additions, multiplications, data transfers, logical operations and can
be extended to implementation of complex matrix manipulations. The essential opera-
tions of many DSP systems include:

Converting analogue signals into a sequence of digital binary numbers,

which requires both sampling and analogue-to-digital (NO) conversion.
• Performing numerical manipulations, predominantly multiplications, on the
digital information data stream.
Converting the digital information back to analogue signal, by digital-to-
analogue (D/A) conversion and filtering.

The basic components of the digital processor, which is equivalent to the ana-
logue processing function, is considered later in Chapter 4. The DSP function is gener-
ally described as an algorithm or program which defines the in-built arithmetic opera-
tions. DSP is in fact an extension of the conventional microprocessor function except
that a fast multiplier is added as a hardware accelerator element.
The main attraction of digital processors is that their accuracy, which is con-
trolled by the quantisation step size or word length employed in the NO converter, is
extendable only at the cost of greater complexity for the ensuing processing operations,
i.e. additions and multiplications. However, floating-point number representations per-
mit the accuracy to be maintained over a wide dynamic range. Further, digital proces-
sors are generally repeatable and much less sensitive to noise than analogue processors
as they are dealing with binary symbols and the processed outputs must always possess
binary values. The microelectronics industry has been reducing continuously silicon
VLSI circuit geometrics and hence improving the speed, complexity and storage capac-
ity. The feature size in the 1970s was 5,um while in the year 2000 we will be designing
VLSI circuits with 0.18,um feature sizes and single chip complexities of 40 M individ-
ual transistors, and by the year 2010 0.7,um feature size circuits are predicted to have 3
GHz clock rates.
VLSI permits the design of application specific integrated circuits with excep-
tional cost/performance capabilities to execute sophisticated DSP processing functions.
Current developments in microelectronics are delivering an order of magnitude increase
in processor operating speed, coupled with a 30-fold reduction in power consumption,
every eight years.
Some of the key benefits which derive from digital signal processing are:

Efficient implementation of linear phase filters for communication and

radar receiver design.
• Easy realisation of Fourier transform processors for signal analysis.
Possibility of implementing complicated processing functions involving
matrix manipulations.
 Introduction xxiii

These advances in DSP, complexity and functionality over analogue signal pro-
cessing are not obtained without some penalties. High accuracy, high speed AID and
D/A conversion hardware is expensive and it can lead to noise and distortion problems.
We always require bandlimiting filters before the sampling function and this introduces
some loss of information. For some applications, e.g. RF signals, digital processors
cannot achieve the necessary high speed sampling requirements. Even with these draw-
backs the use of DSP is becoming ubiquitous, partly because of the 40% per year cumu-
lative growth of the DSP since 1988, with predicted market sales of 6,000,000,000 dol-
lars in 2000. The generic application areas are:

Speech and audio: noise reduction (Dolby), coding, compression (MPEG), recogni-
tion, speech synthesis.

Music: recording, playback and mixing, synthesis of digital music, CD players.

Telephony: speech, data and video transmission by wire, radio or optical fibre.

Radio: digital modulators and modems for cellular telephony.

Signal analysis: spectrum estimation, parameter estimation, signal modelling and clas-
sification.

Instrumentation: signal generation, filtering, signal parameter measurement.

Image processing: 2-D filtering, enhancement, coding, compression, pattern recogni-

tion.

Multimedia: generation, storage and transmission of sound, motion pictures, digital

TV, HDTV, DVD, MPEG, video conferencing, satellite TV.

Radar: filtering, target detection, position and velocity estimation, tracking, imaging,
direction finding, identification.

Sonar: as for radar but also for use in acoustic media such as the sea.

Control: servomechanisms, automatic pilots, chemical plant control.

Biomedical: analysis, diagnosis, patient monitoring, preventive health care,

telemedicine.

Transport: vehicle control (braking, engine management) and vehicle speed measure-
ment.

Navigation: Accurate position determination, global positioning, map display.

XXIV Introduction

The key attraction of DSP devices is that their in-built programmability allows
one standard part to be applied to most of these processing functions by changing the
stored program or instruction set.
In terms of specific uses, the modem car has many tens of processors attached to
sensors for fuel injection, engine management, passenger compartment climate control,
braking system protection, detection of component failure, etc. In fact a major part of
the modem automobile comprises electronics rather than mechanical engineering.
Modem mobile and cellular telephone systems rely on advanced signal processing to
detect and decode the received signals, to minimise received errors, to control the
mobile transmissions, to enhance battery life, etc. The TV set and new set-top boxes
contain a large number of DSP chips to detect the low power signals received and to
decompress the video traffic efficiently.
In the late 1990s there are a tremendous number of applications for DSP and, in
the future, we confidently expect these to increase as the continuous growth in proces-
sor power and novel algorithm development unlocks even more sophisticated process-
ing algorithms which cannot be contemplated today. DSP is thus now firmly
entrenched in the consumer marketplace.