77un/ Edition

DIGITAL SIGNAL PROCESSING

Principles, Algorithms, m l Applications

J o h n G. Proakis Dimitris G. M anolakis

Digital Signal Processing

Principles, Algorithms, and Applications
Third E dition

John G. Proakis
Northeastern U niversity

Dimitris G. Manolakis
Boston C ollege

PRENTICE-HALL INTERNATIONAL, INC.

This edition may be sold only in those countries to which it is consigned by Prentice-Hall International. It is not to be reexported and it is not for sale in the U .S.A ., Mexico, or Canada. 1996 by Prentice-Hall, Inc. Simon & Schuster/A Viacom Company U pper Saddle River, New Jersey 07458

All rights reserved. No part of this book may be reproduced, in any form or by any means, without permission in writing from the publisher.

The author and publisher of this book have used their best efforts in preparing this book. These efforts include the development, research, and testing of the theories and programs to determine their effectiveness. The author and publisher make no warranty of any kind, expressed or implied, with regard to these programs or the documentation contained in this book. The author and publisher shall not be liable in any event for incidental or consequential damages in connection with, or arising out of. the furnishing, performance, or use of these programs.

Printed in the United States of America 10 9 8 7 6 5

ISBN 0-13-3TM33fl-c l

Prentice-Hall International (U K ) Limited. L ondon Prentice-Hall of Australia Pty. Limited, Sydney Prentice-Hall Canada, Inc., Toronto Prentice-Hall Hispanoamericana. S.A., M exico Prentice-Hall of India Private Limited, N ew D elhi Prentice-Hall of Japan, Inc., T okyo Simon & Schuster Asia Pie, Ltd., Singapore Editora Prentice-Hall do Brasil, Ltda., R io de Janeiro Prentice-Hall, Inc, Upper Saddle River, N ew Jersey

Contents

PREFACE 1 INTRODUCTION
1.1
S ignals, S ystem s, and S ignal P ro cessin g 2 1.1.1 Basic Elements of a Digital Signal Processing System. 4 1.1.2 A dvantages of Digital over Analog Signal Processing, 5 C lassificatio n o f Signals 6 1.2.1 Multichannel and Multidimensional Signals. 7 1.2.2 Continuous-Time Versus Discrete-Time Signals. 8 1.2.3 Continuous-Valued Versus Discrete-Valued Signals. 10 1.2.4 Determ inistic Versus Random Signals, 11 T h e C o n c e p t o f F re q u e n c y in C o n tin u o u s -T im e an d D isc re te -T im e S ignals 14 1.3.1 Continuous-Time Sinusoidal Signals, 14 1.3.2 Discrete-Time Sinusoidal Signals. 16 1.3.3 Harmonically Related Complex Exponentials, 19 A n a lo g -to -D ig ita l an d D ig ita l-to -A n a lo g C o n v e rs io n 21 1.4.1 Sampling of Analog Signals, 23 1.4.2 The Sampling Theorem , 29 1.4.3 Q uantization of Continuous-Am plitude Signals, 33 1.4.4 Quantization of Sinusoidal Signals, 36 1.4.5 Coding of Quantized Samples, 38 1.4.6 Digital-to-Analog Conversion, 38 1.4.7 Analysis of Digital Signals and Systems Versus Discrete-Time Signals and Systems, 39 S u m m a ry a n d R e fe re n c e s 39

xiii
1

1.2

1.3

1.4

Problems

40
iii

iv

Contents

DISCRETE-TIME SIGNALS AND SYSTEMS

2.1 D isc rete-T im e S ignals 43 2.1.1 Some Elem entary Discrete-Time Signals, 45 2.1.2 Classification of Discrete-Time Signals, 47 2.1.3 Simple Manipulations of Discrete-Time Signals, 52 D isc re te -T im e S ystem s 56 2.2.1 Input-O utput Description of Systems, 56 2.2.2 Block Diagram Representation of Discrete-Time Systems, 59 2.2.3 Classification of Discrete-Time Systems, 62 2.2.4 Interconnection of Discrete-Tim e Systems, 70 A n alysis o f D isc re te -T im e L in e a r T im e -In v a ria n t S ystem s 72 2.3.1 Techniques for the Analysis of Linear Systems, 72 2.3.2 Resolution of a Discrete-Time Signal into Impulses, 74 2.3.3 Response of LTI Systems to A rbitrary Inputs: The Convolution Sum, 75 2.3.4 Properties of Convolution and the Interconnection of LTI Systems, 82 2.3.5 Causal Linear Tim e-Invariant Systems. 86 2.3.6 Stability of Linear Tim e-Invariant Systems, 87 2.3.7 Systems with Fim te-D uration and Infinite-Duration Impulse Response. 90 D isc rete-T im e System s D e s c rib e d by D iffe re n c e E q u a tio n s 91 2.4.1 Recursive and Nonrecursive Discrete-Tim e Systems, 92 2.4.2 Linear Time-Invariant Systems Characterized by Constant-Coefficient Difference Equations, 95 2.4.3 Solution of Linear Constant-Coefficient Difference Equations. 100 2.4.4 The Impulse Response of a Linear Tim e-Invariant Recursive System, 108 Im p le m e n ta tio n o f D isc re te -T im e S ystem s 111 2.5.1 Structures for the Realization of Linear Tim e-Invariant Systems, 111 2.5.2 Recursive and Nonrecursive Realizations of FIR Systems, 116 C o rre la tio n of D isc re te -T im e S ignals 118 2.6.1 Crosscorrelation and A utocorrelation Sequences, 120 2.6.2 Properties of the A utocorrelation and Crosscorrelation Sequences, 122 2.6.3 Correlation of Periodic Sequences, 124 2.6.4 Com putation of Correlation Sequences, 130 2.6.5 Input-O utput Correlation Sequences, 131 S u m m ary a n d R e fe re n c e s 134

43

2.2

2.3

2.4

2.5

2.6

2.7

Problems

135

Contents

THE Z-TRANSFORM AND ITS APPLICATION TO THE ANALYSIS OF LTI SYSTEMS

3.1 T h e r-T ra n sfo rm 151 3.1.1 The Direct ^-Transform. 152 3.1.2 The inverse : -Transform, 160 P ro p e rtie s o f th e ; -T ra n sfo rm 161

151

3.2 3.3

R a tio n a l c-T ran sfo rm s 172 3.3.1 Poles and Zeros, 172 3.3.2 Pole Location and Time-Domain Behavior for Causal Signals. 178 3.3.3 The System Function of a Linear Tim e-Invariant System. 181 In v e rs io n o f th e ^ -T ra n sfo rm 184 3.4.1 The Inverse ; -Transform by Contour Integration. 184 3.4.2 The Inverse ;-Transform by Power Series Expansion. 186 3.4.3 The Inverse c-Transform by Partial-Fraction Expansion. 188 3.4.4 Decomposition of Rational c-Transforms. 195 T h e O n e -sid e d ^ -T ra n sfo rm 197 3.5.1 Definition and Properties, 197 3.5.2 Solution of Difference Equations. 201 A n aly sis o f L in e a r T im e -In v a ria n t S ystem s in th e --D o m a in 3.6.1 Response of Systems with Rational System Functions. 203 3.6.2 Response of P ole-Z ero Systems with Nonzero Initial Conditions. 204 3.6.3 Transient and Steady-State Responses, 206 3.6.4 Causality and Stability. 208 3.6.5 P ole-Z ero Cancellations. 210 3.6.6 M ultiple-Order Poles and Stability. 211 3.6.7 The Schur-C ohn Stability Test, 213 3.6.8 Stability of Second-Order Systems. 215 S u m m ary an d R e fe re n c e s P ro b le m s 220 219 203

3.4

3.5

3.6

3.7

FREQUENCY ANALYSIS OF SIGNALS AND SYSTEMS

4.1 F re q u e n c y A n aly sis o f C o n tin u o u s-T im e Signals 230 4.1.1 The Fourier Series for Continuous-Time Periodic Signals. 232 4.1.2 Power Density Spectrum of Periodic Signals. 235 4.1.3 The Fourier Transform for Continuous-Time Aperiodic Signals, 240 4.1.4 Energy Density Spectrum of Aperiodic Signals. 243 F re q u e n c y A n aly sis o f D isc re te -T im e Signals 247 4.2.1 The Fourier Series for Discrete-Time Periodic Signals, 247

230

4.2

V)

Contents

Power Density Spectrum of Periodic Signals. 250 The Fourier Transform of Discrete-Time Aperiodic Signals. 253 Convergence of the Fourier Transform. 256 Energy Density Spectrum of Aperiodic Signals, 260 Relationship of the Fourier Transform to the i-Transform , 264 The Cepstrum, 265 The Fourier Transform of Signals with Poles on the Unit Circle, 267 4.2.9 The Sampling Theorem Revisited, 269 4.2.10 Frequency-Domain Classification of Signals: The Concept of Bandwidth, 279 4.2.11 The Frequency Ranges of Some N atural Signals. 282 4.2.12 Physical and M athematical Dualities. 282 4.3 P ro p e rtie s of th e F o u rie r T ra n s fo rm fo r D isc re te -T im e S ignals 286 4.3.1 Symmetry Properties of the Fourier Transform, 287 4.3.2 Fourier Transform Theorems and Properties, 294 F re q u e n c y -D o m a in C h a ra c te ristic s of L in e a r T im e -In v a ria n t S ystem s 305 4.4.1 Response to Complex Exponential and Sinusoidal Signals: The Frequency Response Function. 306 44.2 Steady-State and Transient Response to Sinusoidal Input Signals. 314 4.4.3 Steady-State Response to Periodic Input Signals, 315 4.4.4 Response to Aperiodic Input Signals. 316 4.4.5 Relationships Between the System Function and the Frequency Response Function. 319 4.4.6 Com putation of the Frequency Response Function. 321 4.4.7 Input-O utput Correlation Functions and Spectra, 325 4.4.8 Correlation Functions and Power Spectra for Random Input Signals. 327 L in e a r T im e -In v a ria n t S ystem s as F re q u e n c y -S e le c tiv e F ilters 330 Ideal Filter Characteristics, 331 4.5.1 4.5.2 Lowpass, Highpass, and Bandpass Filters, 333 4,5.3 Digital Resonators, 340 4.5.4 Notch Filters, 343 4.5.5 Comb Filters. 345 4.5.6 All-Pass Filters. 350 4.5.7 Digital Sinusoidal Oscillators, 352 In v e rse S y stem s an d D e c o n v o lu tio n 355 4.6.1 Invertibility of Linear Tim e-Invariant Systems, 356 4.6.2 Minimum-Phase. Maximum-Phase, and Mixed-Phase Systems. 359 4.6.3 System Identification and Deconvolution, 363 4.6.4 Hom om orphic Deconvolution. 365

4.2.2 4.2.3 4.2.4 4.2.5 4.2.6 4.2.7 4.2.8

4.4

4.5

4.6

Contents

vii S u m m ary a n d R e fe re n c e s P ro b le m s 368 367

4.7

THE DISCRETE FOURIER TRANSFORM: ITS PROPERTIES AND APPLICATIONS

5.1 F re q u e n c y D o m a in Sam pling: T h e D isc re te F o u rie r T ra n s fo rm 394 5.1.1 Frequency-Dom ain Sampling and Reconstruction of Discrete-Time Signals. 394 5.1.2 The Discrete Fourier Transform (DFT). 399 5.1.3 The D FT as a Linear Transform ation. 403 5.1.4 Relationship of the DFT to O ther Transforms, 407 P ro p e rtie s o f th e D F T 409 5.2.1 Periodicity. Linearity, and Symmetry Properties, 410 5.2.2 Multiplication of Two DFTs and Circular Convolution. 415 5.2.3 Additional DFT Properties, 421 L in e a r F ilte rin g M e th o d s B ased on th e D F T 5.3.1 Use of the DFT in Linear Filtering. 426 5.3.2 Filtering of Long Data Sequences. 430 F re q u e n c y A n aly sis o f S ignals U sing th e D F T S u m m ary an d R e fe re n c e s P ro b le m s 440 440 425

394

5.2

5.3

5.4 5.5

433

EFFICIENT COMPUTATION OF THE DFT: FAST FOURIER TRANSFORM ALGORITHMS

6.1 E fficien t C o m p u ta tio n of th e D F T : F F T A lg o rith m s 448 6.1.1 Direct Com putation of the DFT, 449 6.1.2 D ivide-and-Conquer Approach to Com putation of the DFT. 450 6.1.3 Radix-2 FFT Algorithms. 456 6.1.4 Radix-4 FFT Algorithms. 465 6.1.5 Split-Radix FFT Algorithms, 470 6.1.6 Im plem entation of FFT Algorithms. 473 A p p lic a tio n s o f F F T A lg o rith m s 475 6.2.1 Efficient Com putation of the D FT of Two Real Sequences. 475 6.2.2 Efficient Com putation of the D FT of a Z N -Point Real Sequence, 476 6.2.3 Use of the FFT Algorithm in Linear Filtering and Correlation, 477 A L in e a r F ilte rin g A p p ro a c h to C o m p u ta tio n o f th e D F T 6.3.1 The Goertzel Algorithm, 480 6.3.2 The Chirp-z Transform Algorithm, 482 479

448

6.2

6.3

viii 6.4

Contents

Q u a n tiz a tio n E ffects in the C o m p u ta tio n o f th e D F T 486 6.4.1 Quantization Errors in the Direct Com putation of the DFT. 487 6.4.2 Quantization Errors in FFT Algorithms. 489 S u m m ary an d R e fe re n c e s P ro b le m s 494 493

6.5

7 IMPLEMENTATION OF DISCRETE-TIME SYSTEMS

7.1 7.2 S tru c tu res fo r th e R e a liz a tio n o f D isc re te -T im e S ystem s S tru c tu res fo r F IR System s 502 7.2.1 Direcl-Form Structure, 503 7.2.2 Cascade-Form Structures. 504 7.2.3 Frequency-Sampling S tructures1. 506 7.2.4 Lattice Structure. 511 S tru c tu re s for IIR S ystem s 519 7.3.1 Direct-Form Structures. 519 7.3.2 Signal Flow Graphs and Transposed Structures. 521 7.3.3 Cascade-Form Structures, 526 7.3.4 Parallel-Form Structures. 529 7.3.5 Lattice and Lattice-Ladder Structures for IIR Systems, 531 S tate-S p a ce System A n aly sis a n d S tru c tu re s 539 7.4.1 State-Space Descriptions of Systems Characterized by Difference Equations. 540 7.4.2 Solution of the State-Space Equations. 543 7.4.3 Relationships Between Input-O utput and State-Space Descriptions, 545 7.4.4 State-Space Analysis in the z-Domain, 550 7.4.5 Additional State-Space Structures. 554 R e p re s e n ta tio n of N u m b e rs 556 7.5.1 Fixed-Point Representation of Numbers. 557 7.5.2 Binary Floating-Point R epresentation of Numbers. 561 7.5.3 E rrors Resulting from R ounding and Truncation. 564 Q u a n tiz a tio n of F ilte r C o e fficien ts 569 7.6.1 Analysis of Sensitivity to Quantization of Filter Coefficients. 569 7.6.2 Q uantization of Coefficients in FIR Filters. 578 7.7 R o u n d -O ff E ffects in D igital F ilte rs 582 7.7.1 Limit-Cycle Oscillations in Recursive Systems. 583 7.7.2 Scaling to Prevent Overflow, 588 7.7.3 Statistical Characterization of Q uantization Effects in Fixed-Point Realizations of Digital Filters. 590 S u m m ary a n d R e fe re n c e s P ro b le m s 600 598 500

500

7.8

Contents

ix

DESIGN OF DIGITAL FILTERS

8.1 G e n e ra l C o n s id e ra tio n s 614 8.1.1 Causality and Its Implications. 615 8.1.2 Characteristics of Practical Frequency-Selective Filters. 619 D e sig n o f F IR F ilters 620 8.2.1 Symmetric and Antisym m eiric FIR Filters, 620 8.2.2 Design of Linear-Phase FIR Filters Using Windows, 623 8.2.3 Design of Linear-Phase FIR Filters by the Frequency-Sampling M ethod, 630 8.2.4 Design of Optimum Equiripple Linear-Phase FIR Filters, 637 8.2.5 Design of FIR Differentiators, 652 8.2.6 Design of Hilbert Transformers, 657 8.2.7 Comparison of Design M ethods for Linear-Phase FIR Filters, 662 D esig n o f I I R F ilters F ro m A n a lo g F iiters 666 8.3.1 IIR Filter Design by Approxim ation of Derivatives. 667 8.3.2 IIR Filter Design by Impulse Invariance. 671 8.3.3 IIR Filter Design by the Bilinear Transform ation, 676 8.3.4 The M atched-; Transform ation, 681 8.3.5 Characteristics of Commonly Used Analog Filters. 681 8.3.6 Some Examples of Digital Filter Designs Based on the Bilinear Transform ation. 692 F re q u e n c y T ra n s fo rm a tio n s 692 8.4.1 Frequency Transform ations in the Analog Dom ain, 693 8.4.2 Frequency Transform ations in the Digital Dom ain. 698 D esig n o f D ig ital F ilters B a sed on L e a st-S q u a re s M e th o d 8.5.1 Pade Approxim ation Method, 701 8.5.2 Least-Squares Design Methods, 706 8.5.3 FIR Least-Squares Inverse (W iener) Filters, 711 8.5.4 Design of IIR Filters in the Frequency Dom ain, 719 S u m m ary an d R e fe re n c e s P ro b le m s 726 724 701

614

8.2

8.3

8.4

8.5

8.6

SAMPLING AND RECONSTRUCTION OF SIGNALS

9.1 S am p lin g o f B a n d p a ss S ignals 738 9.1.1 R epresentation of Bandpass Signals, 738 9.1.2 Sampling of Bandpass Signals, 742 9.1.3 Discrete-Time Processing of Continuous-Time Signals, 746 A n a lo g -to -D ig ita l C o n v e rsio n 748 9.2.1 Sample-and-Hold. 748 9.2.2 Quantization and Coding, 750 9.2.3 Analysis of Q uantization Errors, 753 9.2.4 Oversampling A /D Converters, 756

738

9.2

Contents

9.3

D ig ita l-to -A n a lo g C o n v e rsio n 763 9.3.1 Sample and Hold, 765 9.3.2 First-Order Hold. 768 9.3.3 Linear Interpolation with Delay, 771 9.3.4 Oversampling D/A Converters, 774 S u m m ary an d R e fe re n c e s P ro b le m s 775 774

9.4

10 MULTIRATE DIGITAL SIGNAL PROCESSING

10.1 10.2 10.3 10.4 10.5 In tro d u c tio n 783 784 787 790

782

D e c im a tio n by a F a c to r D In te rp o la tio n by a F a c to r /

S am p lin g R a te C o n v e rsio n by a R a tio n a l F a c to r I ID F iite r D esig n an d Im p le m e n ta tio n for S a m p lin g -R ate C o n v e rsio n 792 10.5.1 Direct-Form FIR Filter Structures, 793 10.5.2 Polyphase Filter Structures, 794 10.5.3 Time-Variant Filter Structures. 800

10.6 10.7

M u ltistag e Im p le m e n ta tio n o f S a m p lin g -R a te C o n v e rs io n

806

S a m p lin g -R a te C o n v e rsio n o f B a n d p a ss S ignals 810 10.7.1 Decim ation and Interpolation by Frequency Conversion, 812 10.7.2 M odulation-Free Method for Decimation and Interpolation. 814 S a m p lin g -R a te C o n v e rsio n by an A rb itra ry F a c to r 815 10.8.1 First-O rder Approxim ation, 816 10.8.2 Second-Order Approximation (Linear Interpolation). 819 A p p lic a tio n s o f M u ltira te Signal P ro c essin g 821 10.9.1 Design of Phase Shifters. 821 10.9.2 Interfacing of Digital Systems with Different Sampling Rates, 823 10.9.3 Im plem entation of Narrowband Lowpass Filters, 824 10.9.4 Im plem entation of Digital Filter Banks. 825 10.9.5 Subband Coding of Speech Signals, 831 10.9.6 Q uadrature M irror Filters. 833 10.9.7 Transmultiplexers. 841 10.9.8 Oversampling A/D and D /A Conversion, 843 S u m m ary an d R e fe re n c e s
P ro b le m s 846

10.8

10.9

10.10

844

Contents

xi

11 LINEAR PREDICTION AND OPTIMUM LINEAR FILTERS

11.1 In n o v a tio n s R e p re s e n ta tio n o f a S ta tio n a ry R a n d o m P ro c e ss 852 11.1.1 Rational Power Spectra. 854 11.1.2 Relationships Between the Filter Param eters and the Autocorrelation Sequence, 855 F o rw a rd an d B a ck w ard L in e a r P re d ictio n 857 11.2.1 Forw ard Linear Prediction, 857 11.2.2 Backward Linear Prediction, 860 11.2.3 The Optimum Reflection Coefficients for the Lattice Forward and Backward Predictors, 863 11.2.4 Relationship of an A R Process to Linear Prediction. 864 S o lu tio n o f th e N o rm al E q u a tio n s 864 11.3.1 The Levinson-Durbin Algorithm. 865 11.3.2 The Schiir Algorithm. 868 P ro p e rtie s o f th e L in e a r P re d ic tio n -E rro r F ilte rs 873

852

11.2

11.3

11.4 11.5

A R L attice an d A R M A L a ttic e -L a d d e r F ilters 876 11.5.1 AR LaLtice Structure. 877 11.5.2 A RM A Processes and Lattice-Ladder Filters. 878 W ie n e r F ilters fo r F ilterin g a n d P re d ictio n 880 11.6.1 FIR W iener Filter, 881 11.6.2 Orthogonality Principle in Linear M ean-Square Estimation, 884 11.6.3 IIR W iener Filter. 885 11.6.4 Noncausal Wiener Filter. 889 S u m m ary an d R e fe re n c e s P ro b le m s 892 890

11.6

11.7

12 POWER SPECTRUM ESTIMATION

12.1 E stim a tio n o f S p e c tra from F in ite -D u ra tio n O b s e rv a tio n s o f Signals 896 12.1.1 Com putation of the Energy Density Spectrum. 897 12.1.2 Estim ation of the Autocorrelation and Power Spectrum of Random Signals: The Periodogram. 902 12.1.3 The Use of the DFT in Power Spectrum Estim ation, 906 N o n p a ra m e tric M e th o d s fo r P o w er S p ectru m E s tim a tio n 908 12.2.1 The B artlett Method: Averaging Periodograms, 910 12.2.2 The Welch Method: Averaging Modified Periodogram s, 911 12.2.3 The Blackman and Tukey Method: Smoothing the Periodogram, 913 12.2.4 Perform ance Characteristics of N onparam etric Power Spectrum Estim ators, 916

896

12.2

xii

Contents

12.2.5 Com putational Requirem ents of Nonparam etric Power Spectrum Estimates, 919 12.3 P a ra m e tric M e th o d s fo r P o w er S p e c tru m E stim a tio n 920 12.3.1 Relationships Between the A utocorrelation and the Model Param eters, 923 12.3.2 The Y ule-W alker M ethod for the A R Model Param eters, 925 12.3.3 The Burg M ethod for the A R Model Param eters, 925 12.3.4 Unconstrained Least-Squares M ethod for the A R Model Param eters, 929 12.3.5 Sequential Estim ation M ethods for the A R Model Param eters, 930 12.3.6 Selection of A R Model O rder, 931 12.3.7 MA Model for Power Spectrum Estim ation, 933 12.3.8 A R M A Model for Power Spectrum Estim ation, 934 12.3.9 Some Experim ental Results, 936 M in im u m V a rian ce S p ectral E stim a tio n 942

12.4 12.5

E ig e n an aly sis A lg o rith m s fo r S p e c tru m E stim a tio n 946 12.5.1 Pisarenko Harm onic Decom position M ethod, 948 12.5.2 Eigen-decomposition of the A utocorrelation Matrix for Sinusoids in White Noise, 950 12.5.3 MUSIC Algorithm. 952 12.5.4 ESPR IT Algorithm, 953 12.5.5 O rder Selection Criteria. 955 12.5.6 Experim ental Results, 956 S u m m ary an d R e fe re n c e s P ro b le m s 960 959

12.6

A RANDOM SIGNALS CORRELATION FUNCTIONS, AND POWER SPECTRA B RANDOM NUMBER GENERATORS C TABLES OF TRANSITION COEFFICIENTS FOR THE DESIGN OF LINEAR-PHASE FIR FILTERS D LIST OF MATLAB FUNCTIONS REFERENCES AND BIBLIOGRAPHY INDEX

A1 B1

C1 D1 R1
11

L j_ Preface

T h is b o o k w as d e v e lo p e d b ased on o u r te ach in g o f u n d e rg ra d u a te and g ra d u a te level co u rse s in d ig ital signal p ro cessin g o v er th e p a s t several y ears. In this b o o k w e p re se n t th e fu n d a m e n ta ls o f d isc re te -tim e signals, system s, and m o d e rn d ig ital p ro cessin g a lg o rith m s an d a p p lic a tio n s fo r stu d e n ts in electrical e n g in e e r ing. c o m p u te r en g in eerin g , a n d c o m p u te r science. T h e b o o k is su itab le fo r e ith e r a o n e -se m e s te r o r a tw o -se m e ste r u n d e rg ra d u a te level c o u rse in d isc re te system s a n d dig ital signal p ro cessin g . It is also in te n d e d fo r use in a o n e -se m e s te r first-year g ra d u a te -le v e l co u rse in digital signal processing. It is a ssu m ed th a t th e s tu d e n t in electrical and c o m p u te r e n g in e e rin g has h ad u n d e rg ra d u a te c o u rses in a d v an ce d calculus (in clu d in g o rd in a ry d iffe re n tia l e q u a tio n s). an d lin ear sy stem s fo r c o n tin u o u s-tim e signals, including an in tro d u c tio n to th e L ap lace tran sfo rm . A lth o u g h the F o u rie r se ries a n d F o u rie r tra n sfo rm s of p e rio d ic an d a p e rio d ic signals a re d escrib ed in C h a p te r 4, we ex p ect th a t m any s tu d e n ts m ay have h ad th is m a te ria l in a p rio r course. A b ala n c e d co v erag e is p ro v id e d of b o th th e o ry an d p ra c tic a l ap p licatio n s. A larg e n u m b e r o f w ell d esigned p ro b le m s a re p ro v id e d to h e lp th e s tu d e n t in m a ste rin g th e su b ject m a tte r. A so lu tio n s m a n u a l is av ailab le fo r th e b en efit o f th e in stru c to r an d can be o b ta in e d fro m th e p u b lish er. T h e th ird e d itio n o f th e b o o k covers basically th e sa m e m a te ria l as th e se c o n d e d itio n , b u t is o rg an ized d ifferen tly . T h e m a jo r d ifferen ce is in th e o rd e r in w hich th e D F T a n d F F T alg o rith m s are co v ered . B a sed o n su g g estio n s m a d e by se v era l rev iew ers, w e n o w in tro d u c e th e D F T a n d d esc rib e its efficient c o m p u ta tio n im m e d ia te ly fo llo w ing o u r tr e a tm e n t of F o u rie r analysis. T his re o rg a n iz a tio n h as also allo w ed us to elim in a te re p e titio n o f so m e to p ics c o n cern in g th e D F T and its ap p licatio n s. In C h a p te r 1 w e d escrib e th e o p e ra tio n s in v o lv ed in th e an alo g -to -d ig ital c o n v ersio n o f an alo g signals. T h e p ro cess o f sa m p lin g a sin u so id is d escrib ed in so m e d e ta il an d th e p ro b le m o f aliasing is ex p lain ed . Signal q u a n tiz a tio n an d d ig ita l-to -a n a lo g co n v ersio n a re also d escrib ed in g e n e ra l term s, b u t th e analysis is p re s e n te d in su b s e q u e n t c h a p te rs. C h a p te r 2 is d e v o te d e n tire ly to th e c h a ra c te riz a tio n a n d analysis o f lin e a r tim e -in v a ria n t (sh ift-in v arian t) d isc re te -tim e system s a n d d isc re te -tim e signals in th e tim e d o m a in . T h e co n v o lu tio n sum is d e riv e d a n d system s a re categ o rized a c co rd in g to th e d u ra tio n of th e ir im p u lse re sp o n s e as a fin ite -d u ra tio n im p u lse xiii

xiv

Preface

re sp o n se (F IR ) an d as an in fin ite -d u ra tio n im pulse re sp o n se ( II R ) . L in e a r tim ein v a ria n t sy stem s c h a ra c te riz e d by d ifferen ce e q u a tio n s are p r e s e n te d an d th e so lu tio n o f d ifferen ce e q u a tio n s w ith initial c o n d itio n s is o b ta in e d . T h e c h a p te r co n clu d es w ith a tre a tm e n t o f d isc re te -tim e c o rre la tio n . T h e z -tra n sfo rm is in tro d u c e d in C h a p te r 3. B o th th e b ila te ra l an d th e u n ila te ra l z -tra n sfo rm s are p re se n te d , a n d m e th o d s fo r d e te rm in in g th e in v erse z -tra n sfo rm are d esc rib e d . U se o f the z -tra n s fo rm in the analysis o f lin ear tim ein v a ria n t sy stem s is illu stra te d , an d im p o rta n t p ro p e rtie s o f system s, su c h as c a u s a l ity a n d stab ility , a re re la te d to z-d o m ain ch aracteristics. C h a p te r 4 tr e a ts th e analysis o f signals and sy stem s in th e fre q u e n c y d o m ain . F o u rie r se ries an d th e F o u rie r tra n sfo rm a re p re s e n te d fo r b o th co n tin u o u s-tim e an d d isc rete-tim e signals. L in e a r tim e -in v a ria n t (L T I) d isc rete sy stem s are c h a r a c terized in th e fre q u e n c y d o m a in by th e ir freq u e n c y resp o n se fu n c tio n an d th e ir re sp o n se to p e rio d ic an d a p e rio d ic signals is d e te rm in e d . A n u m b e r of im p o rta n t ty p es o f d isc re te -tim e system s are d esc rib e d , in clu d in g re s o n a to rs , n o tc h filters, co m b filters, all-p ass filters, a n d o scillato rs. T h e desig n of a n u m b e r of sim ple F IR a n d IIR filters is also co n sid ered . In a d d itio n , th e stu d e n t is in tro d u c e d to th e co n c e p ts o f m in im u m -p h a se , m ix ed -p h ase, an d m a x im u m -p h a se system s an d to th e p ro b le m o f d e c o n v o lu tio n . T h e D F T . its p ro p e rtie s an d its a p p licatio n s, a re th e topics c o v e re d in C h a p te r 5. T w o m e th o d s a re d e sc rib e d fo r using th e D F T to p e rfo rm lin e a r filtering. T h e use o f th e D F T to p e rfo rm fre q u e n c y analysis o f signals is also d escrib ed . C h a p te r 6 co v ers th e efficient c o m p u ta tio n o f th e D F T . In c lu d e d in this c h a p te r are d e sc rip tio n s o f radix-2, ra d ix -4, a n d sp lit-ra d ix fast F o u rie r tra n sfo rm (F F T ) alg o rith m s, a n d a p p lic a tio n s o f th e F F T a lg o rith m s to th e c o m p u ta tio n o f c o n v o lu tio n a n d c o rre la tio n . T h e G o e rtz e l alg o rith m a n d the ch irp -z tra n sfo rm are in tro d u c e d as tw o m e th o d s fo r c o m p u tin g th e D F T using lin e a r filtering. C h a p te r 7 tre a ts th e re a liz a tio n o f I I R an d F IR system s. T h is tre a tm e n t in clu d es d irect-fo rm , cascad e, p a ra lle l, lattice, a n d la ttic e -la d d e r re a liz a tio n s. T h e c h a p te r in clu d es a tr e a tm e n t o f sta te -sp a c e analysis an d s tru c tu re s fo r d isc rete-tim e system s, an d ex am in es q u a n tiz a tio n effects in a d igital im p le m e n ta tio n o f F IR and I IR system s. T e c h n iq u e s fo r d esign o f digital F IR a n d IIR filters are p r e s e n te d in C h a p te r 8. T h e d esign te c h n iq u e s in clu d e b o th d irect design m e th o d s in d isc re te tim e an d m e th o d s involv in g th e co n v ersio n o f an a lo g filters in to digital filters by v ario u s tra n sfo rm a tio n s. A lso tre a te d in this c h a p te r is th e d esig n o f F I R a n d IIR filters by le a st-sq u a re s m e th o d s. C h a p te r 9 fo cu ses o n th e sam pling o f c o n tin u o u s-tim e sig n a ls a n d th e r e c o n s tru c tio n o f such signals fro m th e ir sam ples. In th is c h a p te r, w e d eriv e th e sam p lin g th e o re m fo r b a n d p a ss co n tin u o u s-tim e -sig n a ls an d th e n co v e r th e A /D an d D /A co n v ersio n te c h n iq u e s, including o v e rsam p lin g A /D a n d D /A co n v erters. C h a p te r 10 p ro v id e s an in d e p th tre a tm e n t o f sa m p lin g -ra te c o n v ersio n and its a p p lic a tio n s to m u ltira le d ig ital signal p ro cessin g . In a d d itio n to d escrib in g d e c im atio n a n d in te rp o la tio n by in te g e r facto rs, we p re s e n t a m e th o d o f sa m p lin g -rate

Preface

xv

co n v e rsio n by an a rb itra ry facto r. S ev eral a p p licatio n s to m u ltira te signal p ro c e ss ing a re p re s e n te d , in clu d in g th e im p le m e n ta tio n o f d igital filters, su b b a n d cod in g o f sp e ech sig n als, tra n sm u ltip le x in g , an d o v ersam p lin g A /D a n d D /A c o n v e rte rs. L in e a r p re d ic tio n an d o p tim u m lin e a r (W ien er) filters a re tr e a te d in C h a p te r 11. A lso in clu d ed in this c h a p te r are d escrip tio n s o f th e L e v in s o n -D u rb in alg o rith m a n d Schiir a lg o rith m fo r solving th e n o rm a l e q u a tio n s , as w ell as th e A R la ttic e a n d A R M A la ttic e -la d d e r filters. P o w e r sp e c tru m e stim a tio n is th e m ain to p ic of C h a p te r 12. O u r co v erag e in clu d es a d e s c rip tio n o f n o n p a ra m e tric an d m o d el-b ased (p a ra m e tric ) m e th o d s. A lso d e s c rib e d a re e ig e n -d e c o m p o sitio n -b a se d m e th o d s, in clu d in g M U S IC an d E S P R IT . A t N o r th e a s te r n U n iv ersity , w e h av e u se d th e first six c h a p te rs o f this b o o k fo r a o n e -se m e s te r (ju n io r level) c o u rse in d isc rete sy stem s a n d d ig ital signal p r o cessing. A o n e -s e m e s te r se n io r level c o u rse fo r stu d e n ts w h o h av e h a d p rio r e x p o su re to d isc rete sy stem s can u se th e m a te ria l in C h a p te rs 1 th ro u g h 4 for a q u ick rev iew a n d th e n p ro c e e d to co v er C h a p te r 5 th ro u g h 8. In a first-v ear g ra d u a te level c o u rse in digital signal p ro cessin g , th e first five c h a p te rs p ro v id e th e s tu d e n t w ith a goo d rev iew of d isc re te -tim e system s. T h e in stru c to r can m o v e q u ick ly th ro u g h m o st o f th is m aterial a n d th e n co v e r C h a p te rs 6 th ro u g h 9, fo llo w ed by e ith e r C h a p te rs 10 and 11 o r by C h a p te rs 11 an d 12. W e h a v e in c lu d e d m an y ex am p les th ro u g h o u t th e b o o k an d a p p ro x im a te ly 500 h o m e w o rk p ro b le m s. M an y o f th e h o m ew o rk p ro b le m s can b e so lv ed n u m e r ically on a c o m p u te r, using a so ftw are p ack ag e such as M A T L A B . T h e se p r o b lem s a re id e n tifie d by an asterisk . A p p e n d ix D co n tain s a list o f M A T L A B fu n c tio n s th a t th e s tu d e n t can use in solving th e se p ro b lem s. T h e in s tru c to r m ay also w ish to c o n s id e r th e u se o f a s u p p le m e n ta ry b o o k th a t c o n ta in s c o m p u te r b ased exercises, su c h as th e b o o k s Digilal Signal Processing Us ing M A T L A B (P.W .S. K e n t, 1996) by V. K. In g le a n d J. G . P ro a k is a n d C o m p u te r- B a s e d Exercises f o r S ignal P ro cessing Using M A T L A B (P re n tic e H all, 1994) by C. S. B u rru s e t al. T h e a u th o rs a re in d e b te d to th e ir m an y facu lty c o lleag u es w ho h av e p ro v id e d v alu ab le su g g e stio n s th ro u g h review s o f the first an d se co n d ed itio n s o f this b o o k . T h e se in clu d e D rs. W . E . A le x a n d e r, Y. B re sle r, J. D e lle r, V. Ingle, C. K eller, H . L e v -A ri, L. M e ra k o s , W. M ik h a e l, P. M o n ticcio lo , C. N ikias, M . S ch etzen , H . T ru ssell, S. W ilso n , a n d M. Z o lto w sk i. W e a re also in d e b te d to D r. R , P ric e fo r re c o m m e n d in g th e in clu sion o f sp lit-ra d ix F F T alg o rith m s a n d re la te d su g g estio n s. F in ally , w e w ish to ac k n o w le d g e th e su g g e stio n s an d c o m m e n ts o f m an y fo rm e r g ra d u a te s tu d e n ts , a n d especially th o se by A . L. K ok, J. L in an d S. S rin id h i w ho assisted in th e p r e p a r a tio n o f several illu stra tio n s an d th e so lu tio n s m an u al. J o h n G . P ro a k is D im itris G , M a n o lak is

Introduction

D ig ital signal p ro cessin g is an are a o f science a n d e n g in e e rin g th a t h a s d ev e lo p e d rap id ly o v e r th e p ast 30 y ears. T his rap id d e v e lo p m e n t is a resu lt o f th e signif ican t ad v an ce s in digital c o m p u te r tech n o lo g y an d in te g ra te d -c irc u it fab rica tio n . T h e digital c o m p u te rs an d asso ciated digital h ard w are of th re e d e c a d e s ago w ere relativ ely larg e an d ex p en siv e and, as a co n seq u en ce, th e ir use w as lim ited to g e n e ra l-p u rp o s e n o n -re a l-tim e (o ff-line) scientific c o m p u ta tio n s an d business a p p licatio n s. T h e ra p id d ev e lo p m e n ts in in te g ra te d -c irc u it te c h n o lo g y , sta rtin g with m ed iu m -scale in te g ra tio n (M S I) an d p ro g ressin g to large-scale in te g ra tio n (L S I), a n d now , v ery -larg e-scale in te g ra tio n (V L S I) of e le c tro n ic circuits has sp u rre d th e d e v e lo p m e n t o f p o w erfu l, sm a ller, faster, an d c h e a p e r digital c o m p u te rs an d sp e cial-p u rp o se d igital h a rd w a re . T h e se in ex p en siv e an d re lativ ely fast digital c ir cuits h av e m a d e it p o ssib le to co n stru c t highly so p h istic a te d digital system s cap ab le o f p e rfo rm in g co m p lex digital signal p ro cessin g fu n ctio n s a n d tasks, w hich are u su ally to o difficult a n d /o r to o expensive to be p e rfo rm e d by an a lo g circuitry or a n alo g signal p ro cessin g system s. H e n c e m an y of th e signal p ro cessin g task s th a t w ere c o n v en tio n ally p e rfo rm e d by an alo g m e a n s a re realized to d a y by less ex p en siv e an d o fte n m o re re lia b le digital h a rd w a re . W e do n o t w ish to im ply th a t digital signal p ro cessin g is th e p ro p e r so lu tio n fo r all signal p ro cessin g p ro b lem s. In d e e d , fo r m a n y signals w ith e x tre m e ly w ide b a n d w id th s, real-tim e p ro cessin g is a re q u ire m e n t. F o r such signals, a n a log o r, p e rh a p s, o p tical signal p ro cessin g is th e only p o ssib le so lu tio n . H o w ev er, w h ere dig ital circuits are av ailab le an d h av e sufficient sp e e d to p e rfo rm th e signal p ro cessin g , th ey a re usually p re fe ra b le . N o t only d o d igital circuits yield c h e a p e r an d m o re re lia b le system s fo r signal p ro cessin g , th e y h av e o th e r a d v an tag es as w ell. In p a rtic u la r, digital pro cessin g h a rd w a re allow s p ro g ra m m a b le o p e ra tio n s. T h ro u g h so ftw are, on e can m o re easily m o d ify th e sig n al p ro cessin g fu n ctio n s to b e p e rfo rm e d by th e h a rd w a re . T h u s dig ital h a rd w a re a n d a s so ciated so ftw are p ro v id e a g re a te r d eg re e o f flexibility in sy stem d esign. A lso , th e re is o ften a h ig h e r o rd e r of p re c isio n ach iev ab le w ith d ig ital h a rd w a re an d so ftw are c o m p a re d w ith an alo g circu its a n d an alo g signal p ro cessin g system s. F o r all th e se re a so n s, th e re h as b e e n an explosive grow th in d ig ital signal p ro cessin g th e o ry a n d a p p licatio n s o v e r th e p ast th re e decades.

Introduction

Chap. 1

In this b o o k o u r o b jectiv e is to p re se n t an in tro d u c tio n o f th e basic analysis to ols an d te c h n iq u e s fo r d igital p ro cessin g o f signals. W e b eg in by in tro d u c in g so m e o f th e n ecessa ry term in o lo g y an d by d escrib in g th e im p o rta n t o p e ra tio n s asso ciated w ith th e p ro cess of c o n v ertin g an an alo g signal to d ig ital fo rm su itab le fo r d igital p ro cessin g . A s we shall se e, digital p ro cessin g o f a n a lo g signals has som e d raw b ack s. F irst, an d fo re m o st, c o n v ersio n o f an a n a lo g signal to digital fo rm , acco m p lish ed by sa m p lin g th e signal an d q u a n tiz in g th e sa m p le s, resu lts in a d isto rtio n th a t p re v e n ts us fro m re c o n stru c tin g th e o rig in a l a n a lo g signal fro m the q u a n tiz e d sam p les. C o n tro l o f th e a m o u n t o f th is d isto rtio n is ach ie v e d by p ro p e r choice o f th e sam p lin g ra te a n d th e p recisio n in th e q u a n tiz a tio n p ro cess. S eco n d , th e re a re finite p re c isio n effects th a t m u st be c o n s id e re d in th e d igital pro cessin g o f th e q u a n tiz e d sam p les. W hile th e se im p o rta n t issues are c o n s id e re d in som e d etail in this b o o k , th e em p h asis is on th e analysis a n d d esig n o f digital signal p ro cessin g sy stem s a n d c o m p u ta tio n a l te ch n iq u es.

1.1 SIGNALS, SYSTEMS, AND SIGNAL PROCESSING

A signal is d efin ed as any physical q u a n tity th a t varies w ith tim e, sp ace, o r any o th e r in d e p e n d e n t v ariab le o r variables. M a th em atic ally , we d e sc rib e a signal as a fu n ctio n o f o n e o r m o re in d e p e n d e n t variab les. F o r e x am p le, th e fu n ctio n s
* i( r ) = 5/

(1.1.1) S2(t) = 20 r d escrib e tw o signals, o n e th a t varies lin early w ith the in d e p e n d e n t v ariab le t (tim e) an d a seco n d th a t v aries q u a d ra tic a lly w ith t. A s a n o th e r ex a m p le , co n sid e r the fu n ctio n v) = 3x + 2 x y + 1 0 y 2 (1.1.2) T his fu n ctio n d escrib es a signal o f tw o in d e p e n d e n t v a riab les x a n d y th a t could r e p re s e n t th e tw o sp a tia l c o o rd in a te s in a p lan e. T h e signals d e sc rib e d by (1.1.1) an d (1.1.2) b e lo n g to a class o f signals th a t are p recisely d efin ed by specifying th e fu n c tio n a l d e p e n d e n c e on th e in d e p e n d e n t v ariab le. H o w ev er, th e re are cases w h ere such a fu n c tio n a l re la tio n sh ip is u n k n o w n o r to o highly c o m p licated to be o f any p ractical use. F o r ex am p le, a sp e ech signal (see Fig. 1.1) c a n n o t be d e s c rib e d fu n ctio n ally by ex p ressio n s such as (1.1.1). In g e n eral, a se g m e n t o f sp e ech m ay be re p re se n te d to a high d eg re e o f accu racy as a sum of se v era l sin u so id s o f d iffe re n t am p litu d e s a n d freq u e n cies, th a t is, as N
A j ( t ) s i n [ 2 ; r f } ( r ) f + #,(/)]
(1.1.3)

i=i w h ere {/!,(/)}, {F ,(r)j, a n d {t9,(r)} a re th e se ts of (p o ssib ly tim e -v a ry in g ) a m p litu d es, freq u e n cies, an d p h a se s, resp ectiv ely , o f th e sinusoids. In fact, o n e w ay to in te rp re t th e in fo rm a tio n c o n te n t o r m essag e co n v ey ed by an y sh o rt tim e se g m e n t o f th e

Sec. 1.1

Signals, Systems, and Signal Processing

^ #

#
S

I
A

Th

... ^
ft
N D

i j ^

---------- ,|* y y y > v y y m w

'

'r r m

w m

1
Figure 1.1 Example of a speech signal.

' W W W .......... ............

sp e e c h signal is to m e a s u re the a m p litu d es, freq u e n cies, a n d p h a se s c o n ta in e d in th e sh o rt tim e se g m e n t o f the signal. A n o th e r ex am p le o f a n a tu ra l signal is an e le c tro c a rd io g ra m (E C G ). Such a signal p ro v id e s a d o c to r w ith in fo rm a tio n a b o u t th e co n d itio n o f the p a tie n t's h e a rt. S im ilarly, an e le c tro e n c e p h a lo g ra m (E E G ) signal p ro v id es in fo rm a tio n a b o u t th e activ ity o f th e b rain . S p eech , e le c tro c a rd io g ra m , a n d e le c tro e n c e p h a lo g ra m signals a re ex am p les o f in fo rm a tio n -b e a rin g signals th a t evolve as fu n ctio n s o f a single in d e p e n d e n t v ariab le, n am elv , tim e. A n ex am p le o f a signal th at is a fu n ctio n o f tw o in d e p e n d e n t v ariab les is an im age signal. T h e in d e p e n d e n t v ariab les in th is case are th e sp atial c o o rd in a te s. T h e se a re b u t a few ex am p les o f th e co u n tless n u m b e r of n a tu ra l signals e n c o u n te re d in practice. A s so c ia te d w ith n a tu ra l signals are the m ean s by w hich such signals are g e n e ra te d . F o r ex am p le, sp e ech signals are g e n e ra te d by fo rcin g air th ro u g h th e vocal co rd s. Im ag es a re o b ta in e d by ex p o sin g a p h o to g ra p h ic film to a scene o r an o b ject. T h u s signal g e n e ra tio n is usually asso ciated w ith a sy stem th a t re sp o n d s to a stim u lu s o r fo rce. In a sp e ech signal, th e system consists o f th e vocal cords a n d th e vocal tra c t, also called th e vocal cavity. T h e stim ulus in c o m b in a tio n w ith th e sy stem is called a signal source. T h u s w e have sp eech so u rces, im ag es so u rces, an d v ario u s o th e r ty p es o f signal sources. A sy stem m ay also be defin ed as a physical device th a t p e rfo rm s an o p e r a tio n on a signal. F o r e x am p le, a filter u sed to red u c e th e n o ise an d in te rfe re n c e co rru p tin g a d e s ire d in fo rm a tio n -b e a rin g signal is called a system . In this case th e filter p e rfo rm s so m e o p e ra tio n (s ) on th e signal, w hich h as th e effect o f red u cin g (filterin g ) th e n o ise a n d in te rfe re n c e from th e d e sire d in fo rm a tio n -b e a rin g signal. W h en w e pass a signal th ro u g h a system , as in filterin g , w e say th a t we h av e p ro c e sse d th e signal. In this case th e p ro cessin g of th e signal involves filtering th e n o ise an d in te rfe re n c e fro m th e d e s ire d signal. In g e n e ra l, th e system is c h a ra c te riz e d by th e ty p e o f o p e ra tio n th a t it p e rfo rm s on th e signal. F o r ex am p le, if th e o p e ra tio n is lin ear, th e system is called linear. If th e o p e ra tio n o n th e signal is n o n lin e a r, th e system is said to be n o n lin e a r, a n d so fo rth . S uch o p e ra tio n s a re u su a lly re fe rre d to as signal processing.

Introduction

Chap. 1

F o r o u r p u rp o se s, it is c o n v en ien t to b r o a d e n th e d efin itio n o f a system to include n o t o n ly physical devices, b u t also so ftw are re a liz a tio n s o f o p e ra tio n s on a signal. In d igital p ro cessin g o f signals on a digital c o m p u te r, th e o p e ra tio n s p e r fo rm e d on a signal co n sist of a n u m b e r of m a th e m a tic a l o p e ra tio n s as specified by a so ftw are p ro g ram . In this case, th e p ro g ra m r e p re s e n ts an im p le m e n ta tio n o f the system in software. T h u s we h ave a system th a t is re a liz e d on a d igital c o m p u te r by m ean s o f a se q u en ce o f m a th e m a tic a l o p e ra tio n s; th a t is, w e h av e a digital signal p ro cessin g system realized in so ftw are. F o r e x am p le, a d ig ital c o m p u te r can be p ro g ra m m e d to p e rfo rm digital filtering. A lte rn a tiv e ly , th e d igital processing o n th e signal m ay be p e rfo rm e d by digital h ard w a re (logic circu its) co nfigured to p e rfo rm th e d e sire d specified o p e ra tio n s. In such a re a liz a tio n , w e h av e a physical d ev ice th a t p e rfo rm s th e specified o p e ra tio n s. In a b r o a d e r se n se, a digital system can be im p le m e n te d as a c o m b in a tio n o f digital h a rd w a re an d so ftw are, each of w hich p e rfo rm s its ow n set of specified o p e ra tio n s. T h is b o o k d eals w ith th e p ro cessin g o f signals by digital m e a n s, e ith e r in so ft w are o r in h a rd w a re . Since m an y of the signals e n c o u n te re d in p ra c tic e are analog, w e will also co n sid er th e p ro b lem of c o n v ertin g an a n a lo g signal in to a digital sig n al fo r pro cessin g . T h u s we will be d ealin g p rim a rily w ith d ig ital system s. T he o p e ra tio n s p e rfo rm e d by such a system can u su ally be specified m ath em atically . T h e m e th o d o r set o f ru les for im p le m e n tin g th e sy stem by a p ro g ra m th a t p e r fo rm s th e c o rre sp o n d in g m a th e m a tic a l o p e ra tio n s is called an algorithm. U sually, th e re are m an y w ays o r alg o rith m s by w hich a system can be im p le m e n te d , e ith e r in so ftw are o r in h a rd w a re , to p e rfo rm th e d e sire d o p e ra tio n s a n d c o m p u tatio n s. In p ra c tic e , we h av e an in te re st in devising a lg o rith m s th a t are c o m p u ta tio n a lly efficient, fast, an d easily im p lem en ted . T h u s a m a jo r to p ic in o u r stu d y o f digi tal signal p ro cessin g is th e discussion o f efficient a lg o rith m s fo r p e rfo rm in g such o p e ra tio n s as filterin g , c o rre la tio n , an d sp e c tra l analysis.

1.1.1 Basic Elements of a Digital Signal Processing System

M o st o f th e signals e n c o u n te re d in science an d e n g in e e rin g a re a n a lo g in n a tu re. T h a t is. th e signals a re fu n ctio n s of a c o n tin u o u s v a ria b le , such as tim e o r space, an d u su ally ta k e o n v alues in a co n tin u o u s ran g e. S uch signals m ay be p ro cessed directly by a p p ro p ria te an alo g system s (such as filters o r fre q u e n c y an aly zers) or fre q u e n c y m u ltip lie rs for th e p u rp o se of ch an g in g th e ir c h a ra c te ristic s o r ex tractin g so m e d esired in fo rm a tio n . In such a case w e say th a t th e signal h as b e e n p ro cessed d irectly in its an alo g fo rm , as illu strated in Fig. 1.2. B o th th e in p u t signal a n d the o u tp u t signal a re in an a lo g form .

Analog input signal

Analog signal processor

Analog output signal Figure 1.2 A nalog signal processing.

Sec. 1.1

Signals, Systems, and Signal Processing

5
Analog output signal

Analog input

signal
Digital input signal Figure 1.3 Digital output signal

Block diagram of a digital signal processing system.

D ig ital signal p ro cessin g p ro v id e s an a lte rn a tiv e m e th o d fo r p ro cessin g th e a n a lo g signal, as illu stra te d in Fig. 1.3. T o p e rfo rm th e p ro cessin g digitally, th e re is a n e e d fo r an in te rfa c e b e tw e e n th e an a lo g signal a n d th e digital p ro cesso r. T h is in te rfa c e is called an analog-to-digital ( A / D ) converter. T h e o u tp u t of th e A /D c o n v e rte r is a d ig ital signal th a t is a p p ro p ria te as an in p u t to th e d igital p ro cesso r. T h e dig ital signal p ro c e ss o r m ay be a larg e p ro g ra m m a b le digital c o m p u te r o r a sm all m ic ro p ro c e s so r p ro g ra m m e d to p e rfo rm th e d e s ire d o p e ra tio n s on th e in p u t signal. It m ay also be a h a rd w ire d digital p ro c e ss o r co n fig u red to p e rfo rm a specified se t o f o p e ra tio n s on th e in p u t signal. P ro g ra m m a b le m ach in es p r o v id e th e flexibility to ch an g e th e signal p ro cessin g o p e ra tio n s th ro u g h a ch an g e in th e so ftw are, w h e re a s h a rd w ire d m ach in es a re difficult to reco n fig u re. C o n s e q u e n tly , p ro g ra m m a b le signal p ro c e ss o rs a re in very c o m m o n use. O n th e o th e r h an d , w h en signal p ro cessin g o p e ra tio n s are w ell d efin ed , a h a rd w ire d im p le m e n ta tio n o f th e o p e ra tio n s can be o p tim ized , re su ltin g in a c h e a p e r signal p ro c e sso r a n d , u su ally , o n e th a t ru n s fa ste r th a n its p ro g ra m m a b le c o u n te rp a rt. In a p p li c atio n s w h e re th e d ig ital o u tp u t fro m th e d igital signal p ro c e sso r is to be given to th e u se r in an alo g form , such as in sp e ech co m m u n icatio n s, w e m ust p r o vid e a n o th e r in te rfa c e fro m th e digital d o m a in to th e a n a lo g d o m ain . S uch an in te rfa c e is called a digital-to-analog ( D / A ) converter. T h u s th e signal is p r o v id ed to th e u se r in an a lo g form , as illu stra te d in th e b lo ck d iag ram o f Fig. 1.3. H o w e v e r, th e re a re o th e r p ractical a p p lic a tio n s involving signal analysis, w h ere th e d e s ire d in fo rm a tio n is co n v ey ed in digital form a n d n o D /A c o n v e rte r is re q u ire d . F o r ex am p le, in th e d igital p ro cessin g o f r a d a r signals, th e in fo rm a tio n e x tra c te d fro m th e ra d a r signal, such as th e p o sitio n o f th e aircra ft a n d its sp e ed , m ay sim ply b e p rin te d on p a p e r. T h e re is n o n e e d fo r a D /A c o n v e rte r in th is case.

1.1.2 Advantages of Digital over Analog Signal Processing

T h e re a re m an y re a so n s w hy d ig ital signal p ro cessin g o f an an alo g signal m ay be p re fe ra b le to p ro cessin g th e signal directly in th e an a lo g d o m ain , as m e n tio n e d briefly e a rlie r. F irst, a digital p ro g ra m m a b le sy stem allow s flexibility in r e c o n figuring th e digital signal p ro cessin g o p e ra tio n s sim ply by ch anging th e p ro g ra m .

Introduction

Chap. 1

R e c o n fig u ra tio n o f an an a lo g system usually im plies a re d e sig n o f th e h a rd w a re follow ed by te stin g a n d v erification to see th a t it o p e ra te s p ro p e rly . A ccu racy c o n s id e ra tio n s also p lay an im p o rta n t role in d e te rm in in g th e fo rm o f th e signal p ro cesso r. T o le ra n c e s in an alo g c ircu it c o m p o n e n ts m a k e it e x tre m e ly difficult fo r th e system d esig n er to co n tro l th e accu racy o f an an a lo g signal p r o cessing system . O n th e o th e r h an d , a digital system p ro v id e s m uch b e tte r c o n tro l o f accu racy re q u ire m e n ts . Such re q u ire m e n ts , in tu rn , re s u lt in specifying th e a c cu racy r e q u ire m e n ts in th e A /D c o n v e rte r a n d th e d igital sig n a l p ro c e sso r, in te rm s o f w ord le n g th , flo atin g -p o in t v ersu s fix ed -p o in t arith m e tic , a n d sim ilar facto rs. D ig ita l signals are easily sto re d o n m a g n e tic m ed ia (ta p e o r disk) w ith o u t d e te rio ra tio n o r loss o f signal fidelity b e y o n d th a t in tro d u c e d in th e A /D co n v ersio n . A s a c o n se q u e n c e , th e signals b e c o m e tra n s p o rta b le an d can b e p ro cessed off-line in a re m o te la b o ra to ry . T h e digital signal p ro cessin g m e th o d also allow s for th e im p le m e n ta tio n o f m o re so p h istic a te d signal p ro cessin g alg o rith m s. It is usually very difficult to p e rfo rm p recise m a th e m a tic a l o p e ra tio n s on signals in a n a lo g fo rm b u t th ese sam e o p e ra tio n s can b e ro u tin e ly im p le m e n te d on a d ig ital c o m p u te r using so ftw are. In so m e cases a d igital im p le m e n ta tio n of th e signal p ro cessin g system is c h e a p e r th a n its an a lo g c o u n te rp a rt. T h e lo w er cost m ay be d u e to th e fact th a t th e dig ital h a rd w a re is c h e a p e r, o r p e rh a p s it is a re su lt o f th e flexibility fo r m o d ifications p ro v id e d by th e digital im p le m e n ta tio n . A s a c o n se q u e n c e o f th ese ad v a n ta g e s, d igital signal p ro c e ssin g has b e e n a p p lied in p ractical sy stem s co v erin g a b ro a d ra n g e of d iscip lin es. W e cite, fo r ex am p le, th e a p p licatio n o f d igital signal p ro cessin g te c h n iq u e s in sp e ech p ro cessin g an d signal tran sm issio n o n te le p h o n e ch an n els, in im age p ro c e ssin g an d tra n sm is sio n , in seism o lo g y an d geophysics, in oil e x p lo ra tio n , in th e d e te c tio n of n u c le a r ex p lo sio n s, in th e p ro cessin g of signals receiv ed fro m o u te r sp a ce, an d in a vast v ariety o f o th e r a p p licatio n s. S om e o f th e se a p p lic a tio n s a re cited in su b s e q u e n t ch ap ters. A s a lre a d y in d icated , h o w ev er, digital im p le m e n ta tio n has its lim itatio n s. O n e p ractical lim ita tio n is th e sp e ed o f o p e ra tio n o f A /D c o n v e rte rs a n d digital signal p ro cesso rs. W e shall see th a t signals hav in g e x tre m e ly w id e b a n d w id th s re q u ire fa st-sam p lin g -rate A /D c o n v e rte rs an d fast d igital signal p ro cesso rs. H e n c e th e re a re an alo g signals w ith larg e b a n d w id th s fo r w hich a digital p ro cessin g a p p ro a c h is b ey o n d th e s ta te of th e a rt o f digital h a rd w a re .

1.2 CLASSIFICATION OF SIGNALS

T h e m e th o d s we use in p ro cessin g a signal o r in an aly zin g th e re s p o n s e o f a system to a sig n al d e p e n d h eavily on th e ch a ra c te ristic a ttr ib u te s o f th e specific signal. T h e re a re te c h n iq u e s th a t ap p ly only to specific fam ilies o f signals. C o n seq u en tly , an y in v estig atio n in signal p ro cessin g sh o u ld sta rt w ith a classification o f th e signals in v o lv ed in th e specific ap p licatio n .

Sec. 1.2

Classification of Signals

1.2.1 Multichannel and Multidimensional Signals

A s e x p lain ed in S ectio n 1.1, a signal is d escrib ed by a fu n c tio n o f o n e o r m o re in d e p e n d e n t v ariab les. T h e v alue of th e fu n ctio n (i.e., th e d e p e n d e n t v ariab le) can be a re a l-v a lu e d sc alar q u a n tity , a co m p lex -v alu ed q u a n tity , o r p e rh a p s a v ecto r. F o r e x am p le, th e signal si( r ) = A sin37rr is a re a l-v a lu e d signal. H o w e v e r, th e signal s2(f) = A e ji7Tt = A cos 37t t j'A sin 3:r r is co m p lex v alu ed . In so m e a p p lic a tio n s, signals a re g e n e ra te d by m u ltip le so u rces or m u ltip le sen so rs. Such signals, in tu rn , can be re p re s e n te d in v e c to r fo rm . F ig u re 1.4 show s th e th re e c o m p o n e n ts of a v e c to r signal th a t re p re se n ts th e g ro u n d a c c e le ra tio n d u e to an e a r th q u a k e . T h is a c c e le ra tio n is the re su lt of th re e basic ty p es of elastic w aves. T h e p rim a ry (P ) w aves an d th e se co n d a ry (S) w aves p ro p a g a te w'ithin th e b o d y o f rock a n d a re lo n g itu d in al a n d tra n sv e rsa l, resp ec tiv ely . T h e th ird ty p e o f elastic w ave is called th e su rface w ave, b e c a u se it p ro p a g a te s n e a r th e g ro u n d su rface. If $*(/). k = 1. 2. 3. d e n o te s th e electrical signal from th e th se n so r as a fu n ctio n o f tim e, th e se t of p = 3 signals can be re p re se n te d by a v e c to r S?(f )< w h ere r si (O ' S;,(r) =
Si(t)

-Sl(t) J W e re fe r to such a v e c to r o f signals as a m u ltich a n n el signal. In e le c tro c a rd io g ra p hy. for ex am p le, 3 -lead an d 12-lead e le c tro c a rd io g ra m s (E C G ) are o ften used in p ractice, w hich resu lt in 3 -ch an n el a n d 12-channel signals. L e t us n o w tu rn o u r a tte n tio n to th e in d e p e n d e n t v a ria b le (s). If the signal is a fu n ctio n o f a single in d e p e n d e n t v ariab le, th e signal is called a o ne-d im en sio n a l signal. O n th e o th e r h a n d , a signal is called M -d i m e n s i o n a l if its v alu e is a fu n ctio n of M in d e p e n d e n t v ariab les. T h e p ic tu re sh o w n in Fig. 1.5 is an ex am p le of a tw o -d im e n sio n al signal, since th e in ten sity o r b rig h tn e ss I ( x . y) a t each p o in t is a fu n ctio n of tw o in d e p e n d e n t v ariab les. O n th e o th e r h a n d , a b la c k -a n d -w h ite telev isio n p ic tu re m ay be r e p r e se n te d as I ( x . y . t ) since th e b rig h tn e ss is a fu n ctio n of tim e. H e n c e th e T V p ic tu re m ay b e tr e a te d as a th re e -d im e n s io n a l signal. In c o n tra st, a co lo r T V p ic tu re m ay b e d e sc rib e d by th re e in te n sity fu n ctio n s of th e fo rm Ir (x, y. ?), Is (x. y. t ), a n d I i , ( x . y , t ) , c o rre sp o n d in g to th e b rig h tn e ss of the th re e p rin cip al colors (red . g re e n , b lu e) as fu n ctio n s o f tim e. H e n c e th e co lo r T V p ic tu re is a th re e -c h a n n e l, th re e -d im e n s io n a l signal, w hich can b e re p re s e n te d by th e v e c to r -/,(* ,> . O ' I U , y. t) . l b(x, v ,r ) _ In this b o o k we d e a l m ainly w ith sin g le-ch an n el, o n e -d im e n sio n a l real- or co m p lex -v alu ed signals a n d w e re fe r to th e m sim ply as signals. In m a th e m a tic a l

Introduction
Up

Chap. 1

/ % East

7 jJL______
South

bouth

I ____ i

] ____ I.

1 ____ 1 ____ I

f S waves

t Surface waves

_4 P waves
I I

i_______ I ,

I____ _ J _______I _______ I----------- 1 -----------1 ----------- 1 -----------1 ----------- 1 -----------1

r1 -2

10

12

14

16

18

20

22

24

26

28

30

Time (seconds) (b) Figure 1.4 Three components of ground acceleration measured a few kilometers from the epicenter of an earthquake. (From Earthquakes, by B. A . Bold. 1988 by W. H. Freeman and Company. Reprinted with permission of the publisher.)

te rm s th ese signals are d escrib ed by a fu n ctio n o f a single in d e p e n d e n t v ariable. A lth o u g h th e in d e p e n d e n t variab le n e e d n o t be tim e, it is c o m m o n p ractice to use t as th e in d e p e n d e n t v ariab le. In m an y cases th e signal p ro c e ssin g o p e ra tio n s and a lg o rith m s d e v e lo p e d in this tex t for o n e -d im e n sio n a l, sin g le-ch an n el signals can b e e x te n d e d to m u ltic h a n n e l an d m u ltid im e n sio n a l signals.

1.2.2 Continuous-Time Versus Discrete-Time Signals

Signals can b e fu rth e r classified in to fo u r d iffe re n t c a te g o rie s d e p e n d in g on the ch a ra c te ristic s o f th e tim e (in d e p e n d e n t) v a ria b le an d th e v alu es th ey tak e. Con tin u o u s -tim e signals o r a nalog signals a re d e fin ed for ev e ry value o f tim e an d

Sec. 1.2

Classification of Signals

Figure 1.5

Example of a two-dimensional signal.

th ey ta k e on v alu es in the co n tin u o u s in terv al ( a . b ). w h e re a can be oc a n d b can be oc. M a th em atic ally , th ese signals can be d e sc rib e d by fu n ctio n s o f a c o n tin u o u s v ariab le. T h e sp eech w av efo rm in Fig. 1.1 an d th e signals x i(r) = c o s 7 it, xj{t) = e ^ 1 '1 , oc < t < oq are ex am p les o f an alo g signals. Discrete-time signals a re d efin ed o n ly at c e rta in specific v alu es o f tim e. T h e se tim e in sta n ts n eed n o t be e q u id ista n t, b u t in p ractice th ey are usually ta k e n a t e q u a lly sp a ced in terv als fo r c o m p u ta tio n a l c o n v en ien ce an d m a th e m a tic a l tra c ta b ility . T h e signal x(t) = n = 0, 1 , 2 , . . . p ro v id es an ex am p le o f a d isc re te -tim e signal. If we use th e in d ex n o f th e d isc rete-tim e in sta n ts as th e in d e p e n d e n t v ariab le, th e signal v alu e b eco m es a fu n ctio n o f an in te g e r v ariab le (i.e., a se q u e n c e of n u m b e rs). T h u s a d isc re te -tim e signal can be re p re s e n te d m a th e m a tic a lly by a se q u e n c e of real o r c o m p lex n u m b ers. T o em p h asize the d isc rete-tim e n a tu r e o f a signal, w e sh all d e n o te such a signal as x{n) in ste a d o f x ( t ) . If th e tim e in stan ts t are e q u ally sp a ced (i.e., t = n T ), th e n o ta tio n x ( n T ) is also used. F o r ex am p le, th e se q u e n c e x(n) if n > 0 o th erw ise
( 1 .2 . 1 )

is a d isc re te -tim e signal, w hich is r e p re s e n te d g rap h ically as in Fig. 1.6. In ap p licatio n s, d isc rete-tim e signals m ay arise in tw o ways: 1. B y se lectin g v alu es o f an an alo g signal a t d isc re te -tim e in stan ts. T his p ro c e ss is called s am plin g an d is discussed in m o re d etail in S ectio n 1.4. A ll m e a s u r ing in stru m e n ts th a t ta k e m e a s u re m e n ts at a re g u la r in te rv a l o f tim e p ro v id e d isc rete-tim e signals. F o r ex am p le, th e signal x ( n ) in Fig. 1.6 can be o b ta in e d

10
x{n)

Introduction

Chap. 1

I I T
Figure 1.6 Graphical representation of the discrete time signal x[n) = 0.8" for n > 0 and x(n) = 0 for n < 0.

bv sa m p lin g th e an a lo g signal x ( t ) 0 .8 ', t > 0 an d x ( t ) = 0. t < 0 once ev ery seco n d .

2. By accu m u latin g a v a ria b le o v e r a p e rio d o f tim e. F o r e x a m p le , c o u n tin g th e

n u m b e r o f cars using a given s tre e t every h o u r, o r re c o rd in g th e v alu e of gold ev ery day, resu lts in d isc re te -tim e signals. F ig u re 1.7 show s a g rap h o f the W o lfer su n sp o t n u m b e rs. E a c h sam ple o f this d isc re te -tim e signal p ro v id es th e n u m b e r o f su n s p o ts o b se rv e d d u rin g an in te rv a l o f 1 y e a r.

1.2.3 Continuous-Valued Versus Discrete-Valued Signals

T h e v alu es o f a c o n tin u o u s-tim e or d isc re te -tim e signal can be c o n tin u o u s or d is crete. If a signal ta k e s on all po ssib le values on a finite or an infinite ran g e, it

Year Figure 1.7 W olfer annual sunspot num bers (1770-1869).

Sec. 1.2

Classification of Signals

11

is said to b e c o n tin u o u s-v a lu e d signal. A lte rn a tiv e ly , if th e signal ta k e s on v alu es fro m a finite se t o f p o ssib le values, it is said to be a d isc re te -v a lu e d signal. U su ally , th e s e v alu es a re e q u id ista n t a n d h en ce can be e x p ressed as an in te g e r m u ltip le of th e d istan c e b e tw e e n tw o successive values. A d isc re te -tim e signal hav in g a set of d isc rete v alu es is called a digital signal. F ig u re 1,8 show s a d igital signal th a t ta k e s o n o n e o f fo u r p o ssib le values. In o r d e r fo r a signal to be p ro c e sse d digitally, it m u st be d isc rete in tim e a n d its v alu es m u st b e d isc re te (i.e., it m ust b e a digital sig n al). If th e signal to b e p ro c e sse d is in an a lo g fo rm , it is c o n v e rte d to a digital signal by sam pling th e an alo g signal at d isc rete in stan ts in tim e, o b ta in in g a d isc re te -tim e signal, and th e n by q ua n tizin g its v alu es to a set o f d isc re te v alu es, as d e sc rib e d la te r in th e c h a p te r. T h e p ro cess o f co n v e rtin g a c o n tin u o u s-v a lu e d signal in to a d isc re te -v a lu e d signal, called quan tizatio n, is basically an a p p ro x im a tio n p ro cess. It m ay be acco m p lish ed sim ply bv ro u n d in g o r tru n c a tio n . F o r ex am p le, if th e allo w ab le signal v alu es in th e d ig ital signal are in teg ers, say 0 th ro u g h 15, the c o n tin u o u s-v a lu e signal is q u a n tiz e d in to th ese in te g e r values. T h u s th e signal v alue 8.58 will be a p p ro x im a te d by th e v alu e 8 if th e q u a n tiz a tio n p ro cess is p e rfo rm e d by tru n c a tio n o r by 9 if th e q u a n tiz a tio n p ro c e ss is p e rfo rm e d by ro u n d in g to th e n e a re s t in teg er. A n ex p la n a tio n o f th e a n a lo g -to -d ig ita l co n v ersio n p ro cess is given la te r in th e c h a p te r.

Figure 1.8

Digital signal with four different amplitude values.

1.2.4 Deterministic Versus Random Signals

T h e m a th e m a tic a l an aly sis an d p ro cessin g o f signals re q u ire s th e availability o f a m a th e m a tic a l d e sc rip tio n fo r th e signal itself. T h is m a th e m a tic a l d e scrip tio n , o fte n re fe rre d to as th e signal m o d e l, lead s to a n o th e r im p o rta n t classification of signals. A n y signal th a t can b e u n iq u ely d esc rib e d by an explicit m a th e m a tic a l ex p ressio n , a tab le o f d a ta , o r a w ell-defined ru le is called deterministic. T his te rm is used to em p h asize th e fact th a t all p ast, p re se n t, an d fu tu re values o f th e signal are kno w n precisely , w ith o u t an y u n c e rta in ty . In m a n y p ractical ap p lic a tio n s, h o w ev er, th e re are sig n a ls th a t e ith e r c a n n o t b e d esc rib e d to an y re a so n a b le d e g re e o f accu racy by explicit m a th e m a tic a l fo r m u las, o r such a d e sc rip tio n is to o c o m p licated to b e of any p ractical use. T h e lack

12

Introduction

Chap. 1

o f such a re la tio n sh ip im plies th a t such signals ev o lv e in tim e in an u n p re d ic ta b le m a n n e r. W e re fe r to th ese signals as ra n d o m . T h e o u tp u t o f a n oise g e n e ra to r, th e seism ic signal o f Fig. 1.4, an d th e sp e ech signal in Fig. 1.1 are ex am p les of ra n d o m signals. F ig u re 1.9 show s tw o signals o b ta in e d fro m th e sa m e n o ise g e n e ra to r an d th e ir asso ciated h isto g ram s. A lth o u g h th e tw o signals do n o t re s e m b le each o th e r visually, th e ir h isto g ra m s re v e a l som e sim ilarities. T his p ro v id e s m o tiv a tio n fo r

3 > 4 1 ---------------------------------------------------------- 1 ------- * ------------------- ----- ------ ---------------------------------------

200

400

600

800

1000

1200

1400

1600

(a)

400 f ----------------- ------------------- - -------------------------- ------------------- ------------------- 1

350 r
300 250 < 200 -

-j

150 100 -

(b)

Figure 1.9 tograms.

Two random signals from the same signal generator and their his

Sec. 1.2

Classification of Signals
4

13

----- ' 200

------------ '------------------------------------------- ------------------ --------1 400 600 800 1000 1200 1400 1600

< c)

Figure 1.9

C ontinued

th e an aly sis a n d d e sc rip tio n of ra n d o m signals using statistical te c h n iq u e s in stea d o f ex p licit fo rm u las. T h e m ath e m a tic a l fra m e w o rk fo r th e th e o re tic a l analysis of ra n d o m sig n als is p ro v id e d by th e th e o ry of p ro b a b ility a n d sto c h astic processes. S o m e b asic e le m e n ts o f this a p p ro a c h , a d a p te d to th e n e e d s o f this bo o k , are p re s e n te d in A p p e n d ix A . It sh o u ld b e em p h a siz e d a t th is p o in t th a t th e classification o f a real-world sig n al as d e te rm in is tic o r ra n d o m is n o t alw ays clear. S o m etim es, b o th a p p ro a c h e s le a d to m e a n in g fu l resu lts th a t p ro v id e m o re insight in to signal b eh av io r. A t o th e r

14

Introduction

Chap. 1

tim es, th e w ro n g classification m ay le a d to e rro n e o u s resu lts, since som e m a th e m atical to o ls m ay ap p ly o n ly to d e te rm in istic signals w hile o th e rs m ay apply only to ra n d o m signals. T h is will b e c o m e c le a re r as w e ex am in e specific m a th e m a tic a l tools.

1.3 THE CONCEPT OF FREQUENCY IN CONTINUOUS-TIME AND DISCRETE-TIME SIGNALS

T h e co n cep t o f fre q u e n c y is fam iliar to stu d e n ts in e n g in e e rin g a n d th e sciences. T h is co n cep t is basic in. fo r ex am p le, th e design of a rad io re c e iv e r, a high-fidelity sy stem , o r a sp e c tra l filter fo r co lo r p h o to g ra p h y . F ro m physics w e k n o w th a t freq u e n cy is closely re la te d to a specific ty p e o f p e rio d ic m o tio n called h arm o n ic o scillatio n , w hich is d e s c rib e d by sin u so id al fu n ctio n s. T h e c o n c e p t o f freq u e n cy is d irectly re la te d to th e c o n c e p t o f tim e. A ctu ally , it has th e d im en sio n of inverse tim e. T h u s w e sh o u ld e x p ect th a t th e n a tu re of tim e (c o n tin u o u s o r d isc rete) w ould affect th e n a tu re o f th e fre q u e n c y accordingly.

1.3.1 Continuous-Time Sinusoidal Signals

A sim ple h a rm o n ic o sc illatio n is m a th e m a tic a lly d escrib ed c o n tin u o u s-tim e sin u so id al signal: x a(t) = A cos(Q t + 0). oc < t < oc by th e follow ing (1.3.1)

show n in Fig. 1.10. T h e su b sc rip t a u se d w ith x { t ) d e n o te s an an a lo g signal. T his signal is co m p letely c h a ra c te riz e d by th re e p a ra m e te rs: A is th e a m p litu d e of the sin u so id . ft is th e fre q u e n c y in ra d ia n s p e r se co n d (rad /s), a n d 6 is th e p h a se in rad ian s. In ste a d o f ft, we o fte n use th e fre q u e n c y F in cycles p e r seco n d o r h e rtz (H z), w h ere Q In F In term s o f F. (1.3.1) can be w ritte n as x a(t) = A cos(2 n F t + 6 ), oo < t < oc (1.3.3) (1.3.2)

W e will use b o th fo rm s, (1.3.1) an d (1.3.3), in re p re s e n tin g sin u so id al signals.

x j t ) = A cos(2 nFt + 8)

Figure 1.10 Example of an analog sinusoidal signal.

Sec. 1.3

Frequency Concepts in Continuous-Discrete-Tim e Signals

15

T h e an a lo g sin u so id al signal in (1.3.3) is c h a ra c te riz e d by th e follow ing p r o p erties: A L F o r ev ery fixed v alu e o f th e fre q u e n c y F, x a(r) is p erio d ic. In d e e d , it can easily b e sh o w n , using e le m e n ta ry trig o n o m e try , th a t x a(.t + Tp ) = A(r) w h e re Tp = 1 / F is th e fu n d a m e n ta l p e rio d o f the sin u so id al signal. A 2 . C o n tin u o u s -tim e sin u so id al signals w ith d istin ct (d iffe re n t) fre q u e n c ie s a re th e m se lv e s d istin ct. A 3 . In c re a sin g th e fre q u e n c y F resu lts in an in crease in th e ra te o f o sc illatio n o f th e signal, in th e sense th a t m o re p e rio d s are included in a given tim e in terv al. W e o b se rv e th a t fo r F = 0. th e value Tp oc is co n sisten t w ith the fu n d a m e n ta l re la tio n F = 1 / T r . D u e to co n tin u ity o f th e tim e v a ria b le r, w e can in crease th e fre q u e n c y F, w ith o u t lim it, w ith a c o rre sp o n d in g increase in th e ra te o f oscillatio n . T h e re la tio n sh ip s we h ave d e sc rib e d fo r sin u so id al signals carry o v er to th e class o f co m p lex e x p o n e n tia l signals xa {t) - A e JlSi,+{" (1.3.4)

T h is can easily b e seen by e x p ressin g th e se signals in te rm s o f sinusoids using th e E u le r id en tity ej4: = cos < p i j sin < p (1.3.5)

B y d efin itio n , freq u e n c y is an in h e re n tly p o sitiv e physical q u an tity . T his is o b v io u s if we in te r p re t fre q u e n c y as the n u m b e r o f cycles p e r u n it tim e in a p e rio d ic signal. H o w e v e r, in m any cases, only fo r m a th e m a tic a l co n v en ien ce , w e n e e d to in tro d u c e n e g a tiv e freq u e n cies. T o see this w e recall th a t th e sin u so id al signal (1.3.1) m ay be e x p ressed as xa (t) = A c o s (^ r + 6 ) = j eJ(Q,+f>) + ~ e - J(a+9) (1.3.6)

w hich follow s fro m (1.3.5). N o te th a t a sin u so id al signal can be o b ta in e d by ad d in g tw o e q u a l-a m p litu d e c o m p lex -co n ju g ate e x p o n e n tia l signals, so m e tim e s called p h aso rs, illu stra te d in Fig. 1.1 1. A s t i m e p ro g re sse s th e p h a s o rs ro ta te in o p p o site d ire c tio n s w ith a n g u la r fre q u e n c ie s 2 ra d ia n s p e r se co n d . Since a positive f r e q u e n c y c o rre sp o n d s to co u n te rc lo c k w ise u n ifo rm a n g u la r m o tio n , a negative f r e q u e n c y sim ply c o rre sp o n d s to clockw ise a n g u la r m o tio n . F o r m a th e m a tic a l c o n v en ien ce , w e use b o th n e g a tiv e a n d po sitiv e freq u e n cies th ro u g h o u t th is b o o k . H e n c e th e fre q u e n c y ra n g e fo r a n a lo g sinusoids is oo < F < oo.

16

Introduction

Chap. 1

Re

Figure 1.11 Representation of a cosine function by a pair of complex-conjugate exponentials (phasors).

1.3.2 Discrete-Time Sinusoidal Signals

A d isc rete-tim e sin u so id al signal m ay be e x p ressed as x ( n ) A cos (ton + 8), oo < n < oc (1.3.7)

w h ere n is an in te g e r v ariab le, called th e sam p le n u m b e r. A is th e am plitu de o f the sin u so id , co is th e fr e q u e n c y in rad ian s p e r sa m p le , and 8 is th e p h a s e in radians. If in ste a d o f a> we use the freq u e n cy v ariab le / defin ed by a; = 2 n f th e re la tio n (1.3.7) b eco m es x( n ) A cos(2 n f n + 8). oc < n < oc (1.3.9) (1.3.8)

T h e freq u e n cy / h as d im en sio n s o f cycles p e r sa m p le . In S ectio n 1.4. w here we co n sid e r th e sam p lin g o f an alo g sinusoids, w e re la te th e fre q u e n c y v ariab le / o f a d isc re te -tim e sin u so id to th e fre q u e n c y F in cycles p e r se co n d fo r the an a lo g sin u so id . F o r th e m o m e n t we co n sid er th e d isc re te -tim e sin u so id in (1.3.7) in d e p e n d e n tly of th e co n tin u o u s-tim e sinusoid given in (1.3.1). F ig u re 1.12 show s a sin u so id w ith fre q u e n c y co n /6 ra d ia n s p e r sa m p le ( f ~ ~ cycles p e r sam ple) a n d p h ase 8 n / 3 .
x(n) - A cos (urn + 8)

Figure 1.12 Example of a discrete-time sinusoidal signal (w 7t / 6 and b 7r/3).

Sec. 1.3

Frequency Concepts in Continuous-Discrete-Tim e Signals

17

In c o n tra s t to co n tin u o u s-tim e sinusoids, th e d isc re te -tim e sin u so id s are c h a r a c te riz e d by th e fo llow ing p ro p e rtie s: B l . A discrete-time sinu so id is p er iod ic o n ly i f its fr e q u e n c y f is a rational n u m b e r . B y d efin itio n , a d isc rete-tim e signal x( n ) is p erio d ic w ith p e rio d N ( N > 0) if a n d only if x ( n + N ) = x(n ) fo r all n (1.3.10)

T h e sm a llest v alu e o f N fo r w hich (1.3.10) is tru e is called th e f u n d a m e n ta l p eriod . T h e p r o o f o f th e p erio d icity p ro p e rty is sim ple. F o r a sin u so id w ith fre q u e n c y /o to b e p e rio d ic , we sh o u ld have cos[27t /o( A7 + n) + 8} co s(2 ,t/o + 6) T h is re la tio n is tru e if an d only if th e re exists an in te g e r k such th a t 2 n f ) N = 2kn o r, eq u iv alen tly . /o = 4 N (1.3.11)

A cc o rd in g to (1.3.11). a d isc rete-tim e sin u so id al signal is p e rio d ic only if its f re q u e n c y /o can be ex p re sse d as th e ra tio o f tw o in teg ers (i.e.. / () is ra tio n a l). T o d e te rm in e th e fu n d a m e n ta l p e rio d N o f a p e rio d ic sin u so id , w e e x p ress its fre q u e n c y /o as in (1.3.11) an d cancel com m on facto rs so th a t k a n d N are relativ ely p rim e . T h e n th e fu n d a m e n ta l p e rio d o f the sin u so id is e q u a l to N. O b serv e th a t a sm all ch an g e in fre q u e n c y can resu lt in a large change in th e p erio d . F o r ex am p le, n o te th a t f \ = 3 1 /6 0 im plies th at N\ = 60, w h ereas f i = 3 0 /6 0 resu lts in Nz = 2. B 2. Disc rete-tim e sinu so ids whose fre q u en cies are separa ted by an integer m ultiple o f 2 n are identical. T o p ro v e this assertio n , let us c o n sid er th e sin u so id cos(oo + 0). It easily fo llow s th a t cos[(wo + 2 n )n + B\ = cos(wo/i + 2 n n + 9) co%{a)Qn + 9) A s a re su lt, all sin u so id al se q u en ces x k(n) A cos(a>tn + 8). w h ere Wk = cl> c + 2k n , 7r < o>o < n k = 0 ,1 ,2 ,... (1.3.13) (1.3.12)

are in distinguishable (i.e., identical). O n th e o th e r h a n d , th e se q u e n c e s o f any tw o sin u so id s w ith fre q u e n c ie s in th e ra n g e - n < a> < n or ^ < f < ~ a re distinct. C o n s e q u e n tly , d isc rete-tim e sinusoidal signals w ith fre q u e n c ie s M < n o r | / | < \

18

Introduction

Chap. 1

are u n iq u e. A n y se q u e n c e re su ltin g fro m a sin u so id w ith a fre q u e n c y M > n , or | / | > j , is id en tical to a se q u e n c e o b ta in e d fro m a sin u so id al signal w ith fre q u e n c y \co\ < n . B e c a u se o f th is sim ilarity, we call th e sin u so id h av in g th e freq u e n cy M > tt an alias o f a c o rre sp o n d in g sinusoid w ith fre q u e n c y jwj < n . T h u s w e re g a rd freq u e n cies in th e ra n g e tt < a> < tt, o r 1 < / < 1 as u n iq u e a n d all freq u e n cies |o>[ > t t , o r | / | > ~, as aliases. T h e r e a d e r sh o u ld n o tice th e d iffe re n c e b e tw e e n d isc rete-tim e sin u so id s an d co n tin u o u s-tim e sinusoids, w h e re th e la tte r re su lt in d istin ct signals fo r 2 o r F in th e e n tir e ran g e oc < 2 < oc o r oc < F < oc. B 3. The highest rate o f oscillation in a discrete-time sin u s o i d is attained whe n to 7i (or cu = tt) or, eq uivalently , f \ (or f = \ ) . T o illu stra te th is p ro p e rty , let u s in v estig ate th e c h a ra c te ristic s of th e sin u so id a l signal se q u e n c e x ( n ) = cos c^on w h en th e fre q u e n c y v aries fro m 0 to tt. T o sim plify th e a rg u m e n t, we ta k e values o f (l> o = 0, 7 t / 8, tt/4 , jt/2 , n c o rre sp o n d in g to / = 0, 5, w hich resu lt in p erio d ic se q u e n c e s h av in g p e rio d s N = oc, 16, 8, 4, 2. as d e p ic te d in Fig. 1.13. W e n o te th a t th e p e rio d o f th e sinusoid d e c re a se s as th e fre q u e n c y in creases. In fact, we can see th a t th e ra te o f o scillatio n in crease s as th e fre q u e n c y increases.
,, xin)

Figure 1.13

Signal x ( n ) = c o s cu^n for various values of the frequency

cdq.

Sec. 1.3

Frequency Concepts in Continuous-Discrete-Time Signals

19

T o see w h at h a p p e n s for tt < ioq < 2tt . we co n sid e r th e sinusoids w ith fre q u e n c ie s a> \ = a > (> a n d 0 J2 2 n ojq. N o te th at as co\ varies from tt to 2n . a> z v aries fro m i r to 0. it can be easily se en th at = A cos co} n A cos won X2 (n) = A cos uhn A cos(27r coo)n = A c o s (coqii) x \ (h) H e n ce ur is an alias o f w\. If we h ad u sed a sine fu n ctio n in stea d o f a cosine fu n c tio n , th e resu lt w ou ld basically be th e sam e, ex cep t fo r a 180' p h a se d ifferen ce b etw een th e sin u so id s A]() and xi ( n ) . In any case, as we increase th e re lativ e freq u e n cy coo o f a d isc re te -tim e sin u so id from tt to 27r. its ra te of o sc illatio n d e creases. F o r coo = 2 tt th e re su lt is a c o n sta n t signal, as in th e case fo r oju = 0. O b v io u sly , fo r co{) = tt (o r f = k) w e h ave th e hig h est ra te o f oscillation. A s fo r th e case o f co n tin u o u s-tim e signals, n eg ativ e fre q u e n c ie s can be in tro d u c e d as w ell for d isc rete-tim e signals. F o r this p u rp o se w e use th e id en tity
x (n ) = Acos(con + 0) = (>Jiwn+0) +

(1.3.14)

(1.3.15)

Since d isc re te -tim e sinusoidal signals w ith fre q u e n c ie s th a t are se p a ra te d by an in teg er m u ltip le o f 27r are iden tical, it follow s th a t th e fre q u e n c ie s in any in terv al co] < a> < co\ + 2 tt c o n stitu te all the existing d isc rete-tim e sinusoids o r com plex e x p o n en tials. H en ce th e fre q u e n c y ran g e fo r d isc rete-tim e sinusoids is finite with d u ra tio n 2 n . U sually, we choose th e ran g e 0 < co < 2n o r tt < co < tt ({) < f < 1. 1 < / < | ) , w hich we call th e f u n d a m e n t a l range.

1.3.3 Harmonically Related Complex Exponentials

S in u so id al signals a n d com plex e x p o n e n tia ls play a m a jo r role in th e analysis o f signals an d system s. In so m e cases we deal w ith sets of h arm on ic a lly related c o m plex e x p o n e n tia ls (o r sin u so id s). T h e se are sets of p e rio d ic co m p lex e x p o n e n tia ls w ith fu n d a m e n ta l fre q u e n c ie s th a t are m u ltip le s o f a single positive freq u e n cy . A lth o u g h we confine o u r discussion to com plex e x p o n e n tia ls, th e sam e p r o p e r ties clearly hold fo r sin u so id al signals. W e co n sid e r h a rm o n ically re la te d co m p lex e x p o n e n tia ls in b o th co n tin u o u s tim e and d isc rete tim e.

Continuous-time exponentials.
h arm o n ically re la te d e x p o n e n tia ls are sk(t) = ejkno' = e ll7TkFn'

T h e basic signals for co n tin u o u s-tim e ,

jt = 0 . l . 2 . . . .

(1.3.16)

W e n o te th a t for ea c h value o f k, s^U) is p erio d ic w ith fu n d a m e n ta l p e rio d 1 /(kFo) = Tp/ k o r fu n d a m e n ta l freq u e n cy kFo. Since a signal th a t is p e rio d ic w ith p e rio d Tp / k is also p e rio d ic w ith p erio d k ( T p/ k ) = Tp fo r any positive in te g e r k, w e see th a t all o f th e s*(r) h av e a co m m o n p e rio d of Tp, F u rth e rm o re , acco rd in g

20

Introduction

Chap. 1

to S ectio n 1.3.1, Fo is allo w ed to ta k e any v alu e an d all m e m b e rs of th e set are d istin ct, in th e se n se th a t if k\ ^ k2, th en 5*1 (7) ^ F ro m th e basic signals in (1.3.16) we can c o n s tru c t a lin e a r c o m b in a tio n of h arm o n ically re la te d co m p lex ex p o n e n tia ls o f th e form
cc x ( ' ) = k oc = k ~ oc SC Ckelkiltit ( 1 -3 -1 7 )

w h ere ck, k = 0, 1 , 2 . . . . are a rb itra ry co m p lex co n sta n ts. T h e signal x a(t) is p erio d ic w ith fu n d a m e n ta l p erio d Tp = l / f o , a n d its r e p re s e n ta tio n in term s o f (1.3.17) is called th e F ourier series ex p an sio n fo r x a (t). T h e co m p lex -v alu ed co n sta n ts are th e F o u rie r se rie s coefficients a n d th e signal sk (r) is called th e fcth h a rm o n ic o f x (l(t).

Discrete-time exponentials. Since a d isc re te -tim e co m p le x e x p o n e n tia l is p erio d ic if its relativ e fre q u e n c y is a ra tio n a l n u m b e r, w e ch o o se f Q 1/A' an d we d efine th e sets o f h arm o n ically re la te d co m p lex e x p o n e n tia ls by
sk (n) = ej2* kf" \ k = 0. 1 . 2 , . . . (1.3.18)

In c o n tra st to th e c o n tin u o u s-tim e case, we n o te th at sk+Nln) = eJ7* n'k+N,/N = e ^ s k (n) = sk (n) T his m ean s th a t, c o n siste n t w ith (1.3.10), th e re are only N d istin ct p e rio d ic com plex e x p o n en tials in th e se t d e sc rib e d by (1.3.18). F u rth e rm o re , all m e m b e rs of the set h av e a co m m o n p e rio d o f N sam ples. C learly, w e can ch o o se a n y co n secu tiv e A' co m p lex e x p o n e n tia ls, say from k = no to k no 4- N 1 to fo rm a h arm o n ically re la te d set w ith fu n d a m e n ta l freq u e n cy /(, = 1 / N . M o st o fte n , fo r co n v en ien ce , we ch o o se th e set th a t c o rre sp o n d s to no = 0 , th a t is, th e set sk(n) = ejl n k n / s . * = 0 . 1 . 2 .........N - 1 (1.3.19)

A s in th e case o f c o n tin u o u s-tim e signals, it is o b v io u s th a t th e lin e a r co m b in atio n \-l

k= 0

N- 1
Jt= 0

x ( n ) = c * s * ( n ) = Y L ckei2nkn' N resu lts in a p e rio d ic signal w ith fu n d a m e n ta l p e rio d N . A s w e shall see later, this is th e F o u rie r series re p re s e n ta tio n for a p e rio d ic d isc re te -tim e se q u e n c e w ith F o u rie r co efficien ts {q}. T h e se q u e n c e sk (n) is called th e /tth h a rm o n ic o f x(n ). Example 1.3.1 Stored in the memory of a digital signal processor is one cycle of the sinusoidal signal
x ( n ) = sin I

. ( 2nn

+ 6

where 6 2 n q / N , where q and N are integers.

Sec. 1.4

Analog-to-Digital and Digital-to-Analog Conversion

21

(a) Determ ine how this table of values can be used to obtain values of harmonically related sinusoids having the same phase.
(b) D eterm ine how this table can be used to obtain sinusoids of the same frequency

but different phase.

Solution

(a) Let

denote the sinusoidal signal sequence ( 2yrnk xk(n ) = sin I --------- ! V N

This is a sinusoid with frequency f k = k / N. which is harmonically related to x{n). But xk{n) may be expressed as xk{n) = sin
2 tt ( k n )

= x(kn) Thus we observe that ,vt (0) = .v(0). **(1) = x ( k ) . x k (2) = x ( 2 k ) . and so on. Hence the sinusoidal sequence can be obtained from the table of values of x ( n ) by taking every k th value of * ( ) . beginning with .v(0). In this m anner we can generate the values of all harmonically related sinusoids with frequencies fk = k / N for k = 0. 1 ....... N - 1 .
(b) We can control the phase 8 of the sinusoid with frequency j\ k / N by taking

the first value of the sequence from memory location q 9 N/ 2 tt. where q is an integer. Thus the initial phase 6 controls the starling location in the table and we wrap around the table each time the index (kn) exceeds N.

1.4 ANALOG-TO-DIGITAL AND DIGITAL-TO-ANALOG CONVERSION

M o st sig n als o f p ractical in terest, such as sp e ech , bio lo g ical signals, seism ic signals, ra d a r signals, so n a r signals, and v ario u s co m m u n ic a tio n s signals such as au d io a n d v id eo signals, are an alo g . T o p ro cess an a lo g signals by digital m e a n s , it is first n ecessa ry to c o n v e rt th e m into digital form , th a t is, to c o n v e rt th e m to a se q u e n c e o f n u m b e rs h av in g finite precision. T his p ro c e d u re is called analog-to-digital ( A / D ) co n v e r sio n , an d th e c o rre sp o n d in g devices a re called A / D converters ( A D C s ) . C o n c e p tu a lly , w e view A /D co n v ersio n as a th re e -s te p p ro cess. T his p ro cess is illu stra te d in Fig. 1.14. 1. S am p lin g . T his is th e c o n v ersio n o f a c o n tin u o u s-tim e signal in to a d is c re te tim e signal o b ta in e d by tak in g sa m p le s" o f th e c o n tin u o u s-tim e signal at d isc re te -tim e in stan ts. T h u s, if x a(t) is th e in p u t to th e sa m p le r, th e o u tp u t is x a ( n T ) = x ( n ), w h ere T is called th e s a m p lin g interval. 2. Q ua ntiza tio n . T h is is th e co n v ersio n o f a d isc re te -tim e c o n tin u o u s-v a lu e d signal in to a d isc rete-tim e, d isc rete-v alu ed (d ig ital) signal. T h e v alu e o f ea c h

22 A/D converter

Introduction

Chap. 1

01011...
"7

Analog signal

Discrete-time signal

Quantized signal

Digital signal

Figure 1.14 Basic parts of an analog-to-digital (A /D ) converter. signal sam p le is re p re s e n te d by a v alu e se lected from a finite set o f p o ssi b le values. T h e d ifferen ce b e tw e e n th e u n q u a n tiz e d sa m p le x ( n ) an d the q u a n tiz e d o u tp u t x q(n) is called th e q u a n tiz a tio n erro r. 3. Coding. In th e co d in g p ro cess, each d isc rete v alu e x q{n) is re p re s e n te d by a 6 -b it b in ary se q u en ce. A lth o u g h w e m o d e l th e A /D c o n v e rte r as a sa m p le r follow ed by a q u a n tiz e r an d co d er, in p ractice th e A /D co n v ersio n is p e rfo rm e d by a single device th at ta k e s x a(t) an d p ro d u c e s a b in a ry -c o d e d n u m b e r. T h e o p e ra tio n s of sa m p lin g an d q u a n tiz a tio n can be p e rfo rm e d in e ith e r o rd e r b u t. in p ractice, sa m p lin g is alw ays p e rfo rm e d b e fo re q u a n tiz a tio n . In m an y cases o f p ractical in te re st (e.g., sp e ech p ro cessin g ) it is d esirab le to co n v ert th e p ro cessed digital signals in to an a lo g form . (O b v io u sly , we can n o t listen to th e se q u e n c e of sa m p le s re p re se n tin g a sp e ech signal o r see th e n u m b ers co rre sp o n d in g to a T V signal.) T h e p ro cess o f co n v e rtin g a digital signal in to an an alo g signal is kno w n as digital-to-analog ( D / A ) co nversion. A ll D /A c o n v e rte rs c o n n ect th e d o ts " in a digital signal by p e rfo rm in g so m e kind of in te r p o la tio n , w hose accu racy d e p e n d s on th e q u ality of th e D /A c o n v ersio n process. F ig u re 1.15 illu strates a sim ple fo rm o f D /A c o n v ersio n , called a z e ro -o rd e r hold o r a sta ircase a p p ro x im a tio n . O th e r a p p ro x im a tio n s a re p o ssib le, such as lin early co n n ectin g a p a ir o f successive sa m p le s (lin e a r in te rp o la tio n ), fittin g a q u a d ra tic th ro u g h th re e successive sa m p le s (q u a d ra tic in te rp o la tio n ), an d so on. Is th e re an o p tim u m (ideal) in te rp o la to r? F o r signals hav in g a limited f r e q u e n c y co ntent (finite b an d w id th ), th e sa m p lin g th e o re m in tro d u c e d in th e follow ing se c tio n specifies the o p tim u m form o f in te rp o la tio n . S am p lin g a n d q u a n tiz a tio n are tr e a te d in th is sectio n . In p a rtic u la r, we d e m o n s tra te th a t sa m p lin g d o e s n o t re su lt in a loss of in fo rm a tio n , n o r d o es it in tro d u c e d isto rtio n in th e signal if th e signal b a n d w id th is finite. In p rin cip le , th e an alo g signal can b e re c o n stru c te d from th e sam ples, p ro v id e d th a t th e sam p lin g ra te is sufficiently high to avoid th e p ro b le m co m m o n ly called aliasing. O n th e o th e r h an d , q u a n tiz a tio n is a n o n in v e rtib le o r irre v e rsib le p ro c e ss th a t resu lts in signal d isto rtio n . W e shall sh o w th a t th e a m o u n t o f d isto rtio n is d e p e n d e n t on

Sec. 1.4

Analog-to-Digital and Digital-to-Analog Conversion

23

Original Signal

Staircase Approximation

/ / /

V *o /

/ -/ I I o
Time Figure 1.15 Zero-ordcr hold digital-to-analog (D /A ) conversion.

67-

th e accu racy , as m e a s u re d by th e n u m b e r of bits, in th e A /D c o n v ersio n process. T h e facto rs affe ctin g the choice of th e d esired accu racy of th e A /D c o n v e rte r are cost an d sam p lin g ra te . In g en eral, th e cost in crease s w ith an in crease in accuracy a n d /o r sa m p lin g rate.

1.4.1 Sampling of Analog Signals

T h e re are m an y ways to sa m p le an an a lo g signal. W e lim it o u r discussion to p er iod ic o r u n i f o r m s a m p li n g , w hich is th e ty p e of sa m p lin g used m o st o ften in p ractice. T h is is d escrib ed by the re la tio n
x(n) = xa (nT). o c < n < c c

(1.4.1)

w h ere x ( n ) is th e d isc re te -tim e signal o b ta in e d by ta k in g sa m p le s o f th e an alo g signal x aU) ev ery T se co n d s. T his p ro c e d u re is illu stra te d in Fig. 1.16. T h e tim e in te rv a l T b etw e e n successive sa m p les is called th e sa m p li n g p er io d o r sa m p le interval an d its recip ro c al 1 / 7 Fs is called the s a m p lin g rate (sam p les p e r second) o r th e sa m p lin g f r e q u e n c y (h ertz). P e rio d ic sa m p lin g estab lish es a re la tio n sh ip b e tw e e n th e tim e v ariab les t an d n o f c o n tin u o u s-tim e an d d isc re te -tim e signals, resp ec tiv ely . In d e e d , th e s e v a ri ab les a re lin early re la te d th ro u g h th e sa m p lin g p e rio d T or, e q u iv alen tly , th ro u g h th e sa m p lin g ra te Fs l / 7 \ as (1.4.2) A s a c o n s e q u e n c e o f (1.4.2), th e re exists a re la tio n sh ip b e tw e e n th e freq u e n cy v a ria b le F (o r Q) fo r an a lo g signals a n d the freq u e n c y v a ria b le / (o r co) for d isc re te -tim e signals. T o estab lish th is re la tio n sh ip , co n sid e r an an alo g sinusoidal signal o f th e fo rm
x B{t) = A c o s ( 2 t t F t + 8 )

(1.4.3)

24

Introduction

Chap. 1

Analog signal

*(n) = xa(nT) Fs = 1IT Sampler

Discrete-time signal

Figure 1.16

Periodic sampling of an analog signal.

w hich, w h en sa m p le d p erio d ically at a ra te Fs 1 / 7 sa m p le s p e r se co n d , yields x a( n T ) == x{ n) = A cos(27 t F n T -f 9) /2 n n F \ = A cos ( + 9 \ (1-4.4)

If we c o m p are (1.4.4) w ith (1.3.9). w e n o te th a t th e fre q u e n c y v ariab les F an d / a re lin early re la te d as (1.4.5) Fs=sampling frequency F=frequency of analoa f=frequency of digital signal (1.4.6) =relative or normalized frequency

Is or, eq u iv alen tly , as co = Q T

T h e re la tio n in (1.4.5) ju stifies th e n am e relative o r n o r m a li z e d f r e q u e n c y , w hich is so m e tim e s u sed to d esc rib e th e fre q u e n c y v a ria b le / . A s (1.4.5) im p lies, w e can use / to d e te rm in e th e fre q u e n c y F in h e rtz only if th e sa m p lin g fre q u e n c y Fs is k now n. W e recall fro m S ectio n 1.3.1 th a t th e ran g e o f th e fre q u e n c y v a ria b le F o r 2 fo r co n tin u o u s-tim e sin u so id s are oc < F < oo oc < 2 < 00 ( I .4 .7 )

H o w e v e r, th e situ a tio n is d iffe re n t fo r d isc re te -tim e sinusoids. F ro m S ectio n 1.3.2 w e recall th a t 2 c / < \ 2 (1.4.8)

n < co < n

B y su b stitu tin g fro m (1.4.5) a n d (1.4.6) in to (1.4.8), w e find th a t th e freq u e n cy o f th e co n tin u o u s-tim e sin u so id w h e n sa m p le d a t a ra te Fs = 1 / 7 m u st fall in

Sec. 1.4 th e ra n a e

Analog-to-Digitai and Digital-to-Analog Conversion

25

(1.4.9) or, e q u iv alen tly .

-----------=

n F, <

. -Q < _Ti F,

(1.4.10)

T h e se re la tio n s a re su m m a riz e d in T ab le 1.1.

TA B LE 1.1 RELATIO NS AM ONG FR E Q U E N C Y VARIABLES Discrete-time sienals

Continuous-time sienals

Q = radians sec

F
u .. \

c o = 2,t f radians cycles sample sample

1A 5 IA 1

u > = nT ,f=F /F,

/ ' - /

/
\ - o c < fi < oc oc < f' < oc

= to/T.F~f- Fs

.................. - ..... ..

t t /T < Q < ,T /r -f-2,/2 5 F < /-;/-

F ro m th ese re la tio n s w e o b se rv e th at the fu n d a m e n ta l d ifferen ce b etw een c o n tin u o u s-tim e an d d isc rete-tim e signals is in th e ir ran g e of v alu es o f th e fre q u en cy v a riab les F an d / , o r Q and w. P erio d ic sa m p lin g o f a c o n tin u o u s-tim e signal im p lies a m a p p in g of th e infinite freq u e n cy ran g e fo r th e v ariab le F (o r 2) in to a finite fre q u e n c y ran g e for the v ariab le / (o r a>). Since th e highest freq u e n cy in a d isc re te -tim e signal is co tt o r / = *, it follow s th a t, w ith a sa m p lin g rate Fs, th e c o rre sp o n d in g h ig h est values o f F and 2 are

~ 2T

_J _

(1.4.11)

^max ft Fs T h e re fo re , sa m p lin g in tro d u c e s an am b ig u ity , since th e h ig h est fre q u e n c y in a co n tin u o u s-tim e signal th a t can be u n iq u ely d istin g u ish ed w h en such a signal is sa m p le d at a ra te Fs = l / T is Fm & li F J 2 , o r Qmax = n F s. T o see w h at h a p p e n s to fre q u e n c ie s a b o v e F J 2, let us co n sid er the follow ing ex am p le. Example 1.4.1 The implications of these frequency relations can be fully appreciated by considering the two analog sinusoidal signals xi (!) cos 2ji(l0)t xi(t) cos2;r(50)f (1.4.12)

26

Introduction

Chap. 1

which are sampled at a rate Fs = 40 Hz. The corresponding discrete-time signals or sequences are

(1.4.13)

However. cos5,t/2 = cos(2^n + 7rn/2) co s 7 rn / 2 . Hence = *i(n)- Thus the sinusoidal signals are identical and, consequently, indistinguishable. If we are given the sampled values generated by cos( 7r/'2 )n, there is some ambiguity as to whether these sampled values correspond to x\{i) or xz(D- Since x2(r) yields exactly the same values as when the two are sampled at F, 40 samples per second, we say that the frequency F2 50 Hz is an alias of the frequency F\ = 10 Hz at the sampling rate of 40 samples per second. It is im portant to note that F2 is not the only alias of F]. In fact at the sampling rate of 40 samples per second, the frequency F3 = 90 Hz is also an alias of F], as is the frequency F4 = 130 Hz, and so on. All of the sinusoids cos2tt(Fj -f 40k)i. k = 1. 2. 3. 4 . . . . sampled at 40 samples per second, yield identical values. Consequently, they are all aliases of F\ = 10 Hz. In g en eral, th e sam pling o f a c o n tin u o u s-tim e sin u so id al signal x a(t) = A cos(27rF{)/ + 8) w ith a sam pling rate F v = 1 / T resu lts in a d isc re te -tim e signal x ( n ) = A cos(27r/ 0 -f 6)
(1.4.15) (1 .4 .1 4 )

w h ere /o = F()/F , is th e re la tiv e fre q u e n c y o f th e sin u so id . If w e assum e th a t - F s /2 < Fo < F J 2 . th e fre q u e n c y f 0 o f x ( n ) is in th e ran g e ^ < /o < L w hich is th e freq u e n cy ran g e fo r d isc re te -tim e signals. In this case, the re la tio n sh ip b etw een Fo an d f {) is o n e -to -o n e , a n d h en ce it is p o ssib le to identify (o r re c o n stru c t) th e a n alo g signal xa (t) from th e sa m p les x ( n) . O n th e o th e r h a n d , if th e sinusoids x a {t) = A w h ere Fk = F0 + k F s . k = l. 2 . (1.4.17)
c o s (27 z F kt

+ 6)

(1.4.16)

are sam p led a t a ra te F,, it is c le a r th a t th e fre q u e n c y F* is o u tsid e th e fu n d a m e n ta l fre q u e n c y ran g e Fs / 2 < F < F J 2 . C o n se q u e n tly , th e sa m p le d signal is

= A

c o s (27zn F o / F s

+ 6 + 2nkn)

A c o s ( 2 n f o n + 6)

Sec. 1.4

Analog-to-Digital and Digital-to-Analog Conversion

27

w hich is id en tical to th e d isc re te -tim e signal in (1.4.15) o b ta in e d by sam p lin g . (1.4.14). T h u s an infinite n u m b e r of c o n tin u o u s-tim e sin u so id s is re p re se n te d by sa m p lin g th e sa m e d isc re te -tim e signal (i.e.. by th e sam e se t o f sam p le s). C o n se q u e n tly , if w e a re given th e se q u en ce an am b ig u ity exists as to w hich c o n tin u o u s-tim e signal x a(t) th e se v alu es re p re se n t. E q u iv a le n tly , we can say th a t th e fre q u e n c ie s Fk F o + k F s, oo < k < oo (k in te g e r) a re in d istin g u ish a b le fro m th e fre q u e n c y Fo a fte r sa m p lin g an d h e n c e th ey are aliases o f Fo. T he re la tio n sh ip b e tw e e n th e fre q u e n c y v a riab les o f th e c o n tin u o u s-tim e a n d d isc rete-tim e signals is illu stra te d in Fig. 1.17. A n ex a m p le o f aliasing is illu stra te d in Fig. 1.18. w h e re tw o sinusoids w ith fre q u e n c ie s F 0 = | H z an d Fj = | H z yield id en tical sa m p le s w hen a sam p lin g ra te o f Fs 1 H z is used. F ro m (1.4.17) it easily follow s th a t fo r k = - 1 , Fo = F, + Fs = ( | + 1) H z = i H z.

Figure 1.17 Relationship between the continuous-time and discrete-time fre quency variables in the case of periodic sampling.

Figure 1.18

Illustration of aliasing.

28

Introduction

Chap. 1

Since Fsf2. w hich c o rre sp o n d s to w = tt, is th e hig h est fre q u e n c y th a t can be r e p re s e n te d u n iq u ely w ith a sam p lin g ra te Fs , it is a sim ple m a tte r to d e te rm in e th e m ap p in g o f any (alias) fre q u e n c y above Fs/2 (co tt) in to th e e q u iv a le n t freq u e n cy b elo w Fs /2. W e can use F J 2 or a) t t as the p iv o ta l p o in t a n d reflect or fo ld th e alias freq u e n c y to the ran g e 0 < w < tt. Since th e p o in t o f reflectio n is Fsj 2 (co = t t ) , th e fre q u e n c y F J 2 (cu = re) is called th e f o l d i n g frequency. Example 1.4.2 Consider the analog signal Xait) = 3 cos IOOtt/ (a) Determ ine the minimum sampling rate required to avoid aliasing.

(b) Suppose that the signal is sampled at the rate Fs = 200 Hz.

What is the discrete-time signal obtained after sampling? (c) Suppose that the signal is sampled at the rate Fs ~ 75 Hz. W hat is the discretetime signal obtained after sampling?

(d) What is the frequency 0 < F < FJ2 of a sinusoid that yields samples identical to those obtained in part (c)? Solution (a) The frequency of the analog signal is F 50 Hz. Hence the minimum sampling rate required to avoid aliasing is Ff = 100 Hz. (b) If the signal is sampled at Fs = 200 Hz. the discrete-time signal is lOOtf TT x{n) = j cos n j cos n 200 2 (c) If the signal is sampled at F, = 75 Hz. the discrete-time signal is 100?r A tt x(n) 3 cos j ^ n 3 cos n

3 cos ~n 3 (d) For the sampling rate of Fs ~ 75 Hz. we have F = f F, = 7 5 / The frequency of the sinusoid in part (c) is / |. Hence F = 25 Hz Clearly, the sinusoidal signal ya(i) 3 cos I n Ft 3 cos 50 7 ti sampled at Fs 75 samples/s yields identical samples. H ence F 50 Hz is an alias of F = 25 Hz for the sampling rate Fs = 75 Hz.

Sec. 1.4

Analog-to-Digital and Digital-to-Analog Conversion

29

1.4.2 The Sampling Theorem

G iv en an y an a lo g signal, how sh o u ld w e select th e sa m p lin g p e rio d T or, eq u iv alen tly , th e sa m p lin g ra te FJ. T o an sw er this q u e stio n , w e m u st h ave som e in fo rm a tio n a b o u t th e c h aracteristics of th e signal to be sa m p le d . In p a rtic u la r, we m u st h av e so m e g e n e ra l in fo rm a tio n c o n cern in g th e fr e q u e n c y con tent o f th e sig nal. S uch in fo rm a tio n is g en erally av ailab le to us. F o r e x am p le, w e k n o w g en erally th a t th e m a jo r fre q u e n c y co m p o n e n ts o f a sp eech signal fall b elo w 3000 H z. O n th e o th e r h a n d , telev isio n signals, in g e n eral, c o n ta in im p o rta n t fre q u e n c y co m p o n e n ts u p to 5 M H z. T h e in fo rm a tio n c o n te n t of such signals is c o n ta in e d in th e a m p litu d e s, fre q u e n c ie s, an d p h ases o f the v ario u s fre q u e n c y co m p o n e n ts, b ut d e ta ile d k n o w le d g e o f th e c h aracteristics of such signals is n o t a v a ilab le to us p rio r to o b ta in in g th e signals. In fact, the p u rp o se of p ro cessin g the signals is usually to e x tra c t th is d e ta ile d in fo rm a tio n . H o w ev er, if w e k n o w th e m ax im u m freq u e n cy c o n te n t o f th e g e n e ra l class o f signals (e.g.. th e class of sp e ech signals, the class o f v id eo signals, etc.). w e can specify th e sam pling ra te n ecessa ry to co n v ert the a n alo g signals to dig ital signals. L et us su p p o se th a t any an alo g signal can be r e p re s e n te d as a sum o f sin u so id s o f d iffe re n t a m p litu d e s, freq u e n cies, a n d p h ases, th a t is.
N

(1.4.18) w h ere N d e n o te s th e n u m b e r o f freq u e n cy c o m p o n e n ts. A ll signals, such as speech an d v id eo , len d th em se lv e s to such a re p re s e n ta tio n o v er an y sh o rt tim e segm ent. T h e a m p litu d e s, freq u e n cies, a n d p h ases usually ch an g e slow ly w ith tim e from one tim e se g m en t to a n o th e r. H o w e v e r, su p p o se th a t th e fre q u e n c ie s do n o t exceed som e k n o w n fre q u e n c y , say Fmax. F o r ex am p le, F max = 3000 H z fo r th e class o f sp e ech signals a n d Fmax = 5 M H z fo r telev isio n signals. Since th e m ax im u m freq u e n cy m ay v ary slightly fro m d iffe re n t re a liz a tio n s am o n g signals of any given class (e.g., it m ay vary slightly from s p e a k e r to sp e a k e r), w e m ay wish to e n su re th a t Fmax d o e s n o t ex ceed som e p re d e te rm in e d v alue by passin g th e an a lo g signal th ro u g h a filter th a t se v ere ly a tte n u a te s freq u e n cy c o m p o n e n ts ab o v e Fmax. T hus we a re c e rta in th a t no signal in the class co n tain s fre q u e n c y c o m p o n e n ts (having significant a m p litu d e o r p o w e r) above Fmax. In p ra c tic e , such filtering is com m only u sed p rio r to sam p lin g . F ro m o u r k n o w led g e o f Fmax, w e can se lect th e a p p ro p ria te sam pling rate. W e k n o w th a t th e h ig h est freq u e n cy in an an alo g signal th a t can be u n a m b ig u ously re c o n s tru c te d w h en th e signal is sa m p le d a t a ra te F, = 1 / T is F J 7. A ny fre q u e n c y a b o v e Fsf 2 o r b elo w - F J 2 resu lts in sa m p le s th a t a re id en tical w ith a c o rre sp o n d in g fre q u e n c y in th e ra n g e F J 2 < F < Fs/2. T o avoid th e am b ig u ities re su ltin g fro m aliasin g , we m u st se lect th e sa m p lin g ra te to be sufficiently high. T h a t is, w e m u st select F J 2 to be g re a te r th an Fmax. T h u s to avoid th e p ro b le m o f aliasin g , Fs is se le c te d so th a t Fs > 2 Fmax (1.4.19)

30

Introduction

Chap. 1

w h ere Fmax is th e larg est fre q u e n c y c o m p o n e n t in the a n a lo g signal. W ith the sa m p lin g ra te se le c te d in this m a n n e r, any fre q u e n c y c o m p o n e n t, say |F ;| < Fmax, in th e an alo g signal is m a p p e d in to a d isc re te -tim e sinusoid w ith a freq u e n cy

1
2 o r, eq u iv alen tly ,

Fs ~ 2

(1.4.20)

tt < a)j = 2n f < 7 r

(1.4.21)

S ince, | / | = \ o r \co\ = n is th e h ig h est (u n iq u e ) freq u e n c y in a d isc re te -tim e signal, th e choice o f sam p lin g ra te acco rd in g to (1.4.19) avoids th e p ro b le m of aliasing. In o th e r w o rd s, th e co n d itio n Fs > 2 Fmax e n s u re s th a t all th e sin u so id al c o m p o n e n ts in th e a n a lo g signal are m a p p e d in to c o rre sp o n d in g d isc re te -tim e freq u e n cy c o m p o n e n ts w ith fre q u e n c ie s in th e fu n d a m e n ta l in terv al. T h u s all the freq u e n cy c o m p o n e n ts o f th e a n alo g signal a re re p re s e n te d in sa m p le d fo rm w ith o u t am b i guity, a n d h en ce th e an alo g signal can be re c o n stru c te d w ith o u t d isto rtio n from th e sa m p le v alu es u sing an a p p r o p ria te in te rp o la tio n (d ig ita l-to -a n a lo g c o n v e r sio n ) m e th o d . T h e ap p ro p riate o r ideal in te rp o la tio n fo rm u la is specified by the s a m p ling theorem. Sam pling T h eorem . If the h ig h est fre q u e n c y c o n ta in e d in an an alo g signal x a (t) is Fmax = B a n d th e signal is sa m p le d at a ra te F, > 2 F max = 2 B. th e n A .u(r) can b e exactly re c o v e re d fro m its sa m p le values using th e in te rp o la tio n fu n ctio n s i n 2 jr B / 8(t) = T h u s jcfl(f) m ay be e x p ressed as * . ( ) * ( ' - ) w h ere x a(n / F s ) = x a( n T ) = Jt(rc) a re th e sa m p les o f x a(t). W h en th e sa m p lin g of x a(t) is p e rfo rm e d at the m in im u m sam p lin g ra te Fs = 2 B , th e re c o n stru c tio n fo rm u la in (1.4.23) b eco m es ^ / n \ sin2nB (t n flB )
(

_ , ( }

2n B t

(1-4.23)

, ,

zb)

(1'4 2 4 )

T h e sa m p lin g r a te F ^ = 2 B = 2 Fmax is called th e N y q u is t rate. F ig u re 1.19 illus tra te s th e ideal D /A c o n v ersio n p ro cess using th e in te rp o la tio n fu n c tio n in (1.4.22). A s can b e o b se rv e d from e ith e r (1.4.23) o r (1.4.24), th e re c o n stru c tio n o f x a(t) fro m th e se q u e n c e x ( n ) is a c o m p lic a te d p ro cess, involving a w e ig h te d sum o f the in te rp o la tio n fu n ctio n g (t) an d its tim e -sh ifte d v ersio n s g ( t n T ) fo r oo < n < oo, w h ere th e w eig h tin g facto rs a re th e sa m p le s x ( n ) . B e c a u se o f th e co m p lex ity an d th e infinite n u m b e r o f sa m p les re q u ire d in (1.4.23) o r (1.4.24), th e s e re c o n stru c tio n

Sec. 1.4

Analog-to-Digital and Digital-to-Analog Conversion

sample of ,v(n

31

(/[ ^ /

{ri l )l

fll

\t -r l 1 1

Figure 1.19 Ideal D /A conversion (interpolation).

fo rm u ias are p rim a rily o f th e o re tic a l in te re st. P ra ctical in te rp o la tio n m e th o d s are given in C h a p te r 9. Example 1.4.3 Consider the analog signal xu(r) = 3cos50;rz 10sin300;n cos 100tt?

What is the Nvquist rate for this signal? Solution The frequencies present in the signal above are F = 25 Hz. F: = 150 Hz. F, = 50 Hz

Thus Fnm = 150 Hz and according to (1.4.19), F > 2 Fm ax = 300 Hz The Nvquist rale is FA = 2 Fm ;. Hence Fs = 300 Hz Discussion It should be observed that the signal component 10sin300;r/. sampled at the Nvquist raie FA - = 300, results in the samples 10 sin 7r/j. which are identically zero. In other words, we are sampling the analog sinusoid at its zero-crossing points, and hence we miss this signal component completely. This situation would not occur if the sinusoid is offset in phase by some am ount 8. In such a case we have lOsinGOOin -ffl) sampled at the Nvquist rate FA - = 300 samples per second, which yields the samples
+ -c o 10 sin(7rn+ ^) = 10(sin n n cos & stth

sin f>)

= lOsin 6 cos nn

Thus if 6 ^ 0 or tt, the samples of the sinusoid taken at the zero. However, we still cannot obtain the correct amplitude the phase 9 is unknown. A simple rem edy that avoids this situation is to sample the analog signal at a rate higher than Example 1.4.4 Consider the analog signal

Nvquist rate are not all from the samples when potentially troublesome the Nvquist rate.

*a(t) = 3cos2000irf + 5 sin6000;rr + lOcos 12.000;?;

Introduction

Chap. 1

(a) What is the Nvquist rate for this signal?

(b) Assume now that we sample this signal using a sampling rate Fs = 5000
samples/s. What is the discrete-time signal obtained after sampling? (c) What is the analog signal y(r) we can reconstruct from the samples if we use ideal interpolation?

= 2.5 kHz
2

and this is the maximum frequency that can be represented uniquely by the sampled signal. By making use of (1.4.2) we obtain

= 3 cos 2jt( j )h + 5 sin 2n- (^ ) + 10 cos 2n(^)n = 3 c o s2 ;r(|)n + 5 sin 2 7 r(l ) 4- 10cos2,t(1 4- ^)u = 3cos2jr({)n 4- 5sin27r(1) 4- 1 0 c o s 2 ^ (|) Finally, we obtain x(n) = 13 cos2^({)/i - 5sin27r( = )fl The same result can be obtained using Fig. 1.17. Indeed, since F, = 5 kHz. the folding frequency is FJ2 = 2.5 kHz. This is the maximum frequency that can be represented uniquely by the sampled signal. From (1.4.17) we have Fti = Fk kFs. Thus Fo can be obtained by subtracting from Fk an integer m ultiple of Fs such that F t/2 < F0 < F J 2. The frequency F: is less than Fsf2 and thus it is not affected by aliasing. However, the other two frequencies are above the folding frequency and they will be changed by the aliasing effect. Indeed. F '2 = Fj ~ Fs = - 2 kHz
f;

= Fj Fs = 1 kHz h and h = ^ which are in agreem ent

From (1.4.5) it follows that /] = with the result above.

Sec. 1.4

A nalog-toD igital and Digital-to-Analog Conversion

33

(c) Since only the frequency components at 1 kHz and 2 kHz are present in the sampled signal, the analog signal we can recover is
x(t)

= 13 cos 2000^r - 5sin400()-Tf

which is obviously different from the original signal xU). This distortion of the original analog signal was caused by the aliasing effect, due to the low sampling rate used. A lth o u g h aliasin g is a pitfall to be av o id ed , th e re are tw o useful p ractical a p p licatio n s b ased on th e ex p lo itatio n of the aliasing effect. T h e se a p p licatio n s are th e stro b o sc o p e an d the sam pling oscilloscope. B o th in stru m e n ts are d esigned to o p e ra te as aliasin g devices in o rd e r to re p re se n t high fre q u e n c ie s as low f re q u en cies. T o e la b o ra te , co n sid er a signal w ith h ig h -freq u en cy c o m p o n e n ts confined to a given fre q u e n c y b an d B\ < F < B2. w h ere Bz B\ = B is d efined as the b a n d w id th o f th e signal. W e assum e th a t B < < B\ < B 2. T h is co n d itio n m ean s th a t th e fre q u e n c y c o m p o n e n ts in the signal are m uch larg er th an th e b an d w id th B of th e signal. Such signals are usually called p a ssb a n d or n a rro w b a n d signals. N ow . if this signal is sa m p le d at a rate Fs > 2B. b u t F^ << B\. th e n all th e f re q u en cy c o m p o n e n ts c o n ta in e d in the signal will be aliases of fre q u e n c ie s in the ran g e 0 < F < F J 2 . C o n seq u en tly , if we o b se rv e the freq u e n cy c o n te n t of the signal in th e fu n d a m e n ta l range 0 < F < F J 2 . we k now precisely the freq u e n cy co n te n t o f th e an a lo g signal since we k now the fre q u e n c y b an d B\ < F < B2 u n d e r c o n sid e ra tio n . C o n se q u e n tly , if the signal is a n a rro w b a n d (p a ss b a n d ) signal, we can re c o n stru c t th e o riginal signal from the sam p le s, p ro v id e d th a t the signal is sa m p le d at a ra te Fs > 2 B. w h ere B is th e b an d w id th . T h is s ta te m e n t c o n stitu te s a n o th e r fo rm o f th e sam pling th e o re m , w hich we call the p a s s b a n d f o r m in o rd e r to d istin g u ish it fro m th e p rev io u s form o f the sa m p lin g th e o re m , w hich ap p lies in g en eral to all ty p es of signals. T he la tte r is so m e tim es called th e bas e ban d f or m. T h e p a s s b a n d f o r m o f th e sam pling th e o re m is d escrib ed in d e ta il in S ectio n 9.1.2.

1.4.3 Quantization of Continuous-Amplitude Signals

A s w e h av e se en , a dig ital signal is a se q u en ce of n u m b e rs (sa m p le s) in w hich each n u m b e r is re p re s e n te d by a finite n u m b e r of digits (finite p recisio n ). T h e p ro c e ss o f co n v ertin g a d isc rete-tim e c o n tin u o u s-a m p litu d e signal in to a dig ital signal by ex p ressin g each sa m p le value as a finite (in ste a d of an infinite) n u m b e r o f d igits, is called quant izati on. T he e rro r in tro d u c e d in re p re se n tin g th e c o n tin u o u s-v a lu e d signal by a finite set o f d isc rete v alu e levels is called quant i zat i on error o r quant i zat i on noise. W e d e n o te th e q u a n tiz e r o p e ra tio n o n th e sa m p le s x ( n ) as Q[x{n)] an d let x q{n) d e n o te th e se q u en ce o f q u a n tiz e d sam ples a t th e o u tp u t o f th e q u a n tiz e r. H e n ce
Xq(n)

= Q[x(n)]

34

Introduction

Chap. 1

T h e n th e q u a n tiz a tio n e r ro r is a se q u e n c e eq (n) d efin ed as th e d iffe re n c e b etw e e n th e q u a n tiz e d v alu e a n d th e actu al sa m p le value. T h u s eq (n) = x q {n) - x ( n ) W e illu strate th e q u a n tiz a tio n p ro cess w ith a n ex am p le. d isc rete-tim e signal (1.4.25) L e t us co n sid e r the

o b ta in e d by sa m p lin g th e an a lo g e x p o n e n tia l signal x a( t ) = 0 .9 ', t > 0 w ith a sam p lin g freq u e n cy f , = 1 H z (see Fig. 1.20(a)). O b se rv a tio n o f T a b le 1.2, w hich show s th e v alu es o f th e first 10 sa m p les o f x ( n ) , rev eals th a t th e d e sc rip tio n o f the sam p le v alu e x{n) re q u ire s n significant digits. It is o b v io u s th a t th is signal can n o t

(a)

Figure 1.20

Illustration of quantization.

Sec. 1.4

Analog-to-Digital and Digital-to-Analog Conversion

NU M ER IC A L ILLU S TR A TIO N O F Q U A N TIZA TIO N W ITH ONE

35

TA B LE 1.2

S IG N IF IC A N T D IG IT USING T R U N C A T IO N O R R O UN DING
x ( n) n

D iscrete-tim e signal 1 0.9 0.81 0.729 0.6561 0.59049 0.531441 0.4782969 0.43046721 0.387420489

x,(n ) (Truncation)
1.0 0.9 0.8 0.7 0.6 0.5 0.5 0.4 0.4 0.3

.voi) (Rounding)
1.0 0.9 0.8 0.7 0.7 0.6 0.5 0.5 0.4 0.4
-

(Rounding)
0.0 0.0 0.01 0.029 0.0439 0.00951 0.031441 0.0217031 0.03046721 0.012579511

0 1 2 3 4 5 6 7 8 9

be p ro cessed by u sing a c a lcu lato r o r a digital c o m p u te r since only the first few sam p les can be sto re d an d m a n ip u la te d . F o r ex am p le, m ost c alcu lato rs process n u m b e rs w ith only eig h t significant digits. H o w e v e r, let us assum e th a t w e w an t to use only o n e significant digit. To elim in ate th e excess digits, w e can e ith e r sim ply d iscard th em ( tru n ca tion ) o r dis card th e m bv ro u n d in g th e resu ltin g n u m b e r (ro un din g). T h e resu ltin g q u an tized signals x q (n) a re show n in T ab le 1.2. W e discuss only q u a n tiz a tio n by ro u n d in g , alth o u g h it is ju st as easy to tr e a t tru n c a tio n . T h e ro u n d in g p ro c e ss is g raphically illu stra te d in Fig. 1.20b. T h e values allo w ed in th e digital signal are called the quan tizatio n levels, w h ereas the d istan c e A b etw een tw o successive q u a n tiz a tio n levels is called th e q uantization step siz e o r resolution. T h e ro u n d in g q u a n tiz e r assigns each sa m p le of x ( n ) to th e n e a re s t q u a n tiz a tio n level. In c o n tra st, a q u a n tizer th a t p e rfo rm s tru n c a tio n w ould have assigned each sa m p le of jc(/z) to the q u a n tiz a tio n level b elo w it. T h e q u a n tiz a tio n e r ro r eq (n) in ro u n d in g is lim ited to th e ra n g e of A /2 to A /2 , th a t is,

- y <<?,()<f

(1A26)

In o th e r w o rd s, th e in sta n ta n e o u s q u a n tiz a tio n e r ro r c a n n o t exceed half of the q u a n tiz a tio n ste p (see T a b le 1.2). If jcmjn an d j:max r e p re s e n t th e m in im u m an d m ax im u m v alu e of x (n ) a n d L is th e n u m b e r o f q u a n tiz a tio n levels, th en A = Xmax ~ ^ L - 1 (1.4.27)

W e d efin e th e d y n a m i c range of th e signal as jrmax -*min- 1 o u r ex am p le we h av e Jtmax = 1, Jtmjn = 0, a n d L 11, w hich leads to A = 0.1. N o te th a t if the d y n am ic ran g e is fixed, in creasin g th e n u m b e r o f q u a n tiz a tio n levels, L resu lts in a d e c re a se o f th e q u a n tiz a tio n ste p size. T h u s th e q u a n tiz a tio n e r ro r d e c re a s e s and th e accu racy o f th e q u a n tiz e r in crease s. In p ra c tic e w e can re d u c e th e q u a n tiz a tio n

36

Introduction

Chap. 1

error to an insignificant am ount by choosing a sufficient num ber o f quantization levels. T h eoretically, quantization o f analog signals always resu lts in a loss o f in form ation. T his is a result o f the am biguity introduced by quantization. Indeed, quantization is an irreversible or noninvertible process (i.e., a m any-to-on e m ap ping) since all sam ples in a distance A /2 about a certain quantization level are assigned the sam e value. This am biguity m akes the exact quantitative analysis of quantization extrem ely difficult. This subject is discussed further in C hapter 9, where w e use statistical analysis.

1.4.4 Quantization of Sinusoidal Signals

Figure 1.21 illustrates the sam pling and quantization o f an an alog sinusoidal signal x a (/) = A cos using a rectangular grid. H orizontal lines w ithin the range of the quantizer indicate the allow ed levels o f quantization. V ertical lines indicate the sam pling tim es. Thus, from the original analog signal x a{t ) w e obtain a discrete tim e signal x ( n ) = x a {nT) by sam pling and a discrete-tim e, discrete-am plitude signal x q ( nT) after quantization. In practice, the staircase sign al xv (r) can be obtained by using a zero-order hold. This analysis is useful b ecau se sinusoids are used as test signals in A /D converters. If the sam pling rate Fs satisfies the sam pling theorem , quantization is the only error in the A /D conversion process. Thus w e can evalu ate the quantization error

Time Figure L21 Sampling and quantization of a sinusoidal signal.

Sec. 1.4

Analog-to-Digital and Digital-to-Analog Conversion

37

bv q u a n tiz in g th e an alo g signal x(t) in stead of th e d isc re te -tim e signal a ( ) = x a( nT) . In sp e c tio n o f Fig. 1.21 in d icates th at the signal x u {t) is alm o st lin ear b e tw e e n q u a n tiz a tio n levels (see Fig. 1.22). T he c o rre sp o n d in g q u a n tiz a tio n e rro r eq (t) x u(t) x q(t) is show n in Fig. 1.22. In Fig. 1.22. r d e n o te s th e tim e th at x a {t) stay s w ithin th e q u a n tiz a tio n levels. T he m e a n -sq u a re e r ro r p o w e r Pq is Pq = 2 t / Since eq (i) = ( A / 2 r ) t . t

e{!)cJl = 7 j we have

(1.4.28)

< t <

t.

If th e q u a n tiz e r has b bits of accuracy an d the q u a n tiz e r co v ers the e n tire range 2A . th e q u a n tiz a tio n step is A = 2 A / 2 h. H en ce
A 2/ 3

P., = T h e a v erag e p o w e r o f the signal xu(D is 1 f 1' , A2 P, = / (A cos Qi,i r d i = ~ TP Jo 2

(1.4.30)

(1.4.31)

T h e q u ality o f th e o u tp u t o f the A /D c o n v e rte r is usually m e a s u re d by th e signalto- quanti zati on noise ratio ( S Q N R ) . w hich pro v id es th e ratio o f th e signal p o w er to th e no ise po w er: S Q N R = = - 22b P " > ' i/ E x p re s se d in d ecib els (d B ), th e S Q N R is S Q N R (d B ) = 101og1(l S Q N R = 1.76 + 6.026 (1.4.32)

T his im p lies th a t th e S Q N R in crease s a p p ro x im ately 6 dB fo r ev ery bit a d d e d to th e w o rd le n g th , th a t is. fo r each d o u b lin g of th e q u a n tiz a tio n levels. A lth o u g h fo rm u la (1.4.32) was d eriv ed fo r sin u so id al signals, w e shall see in C h a p te r 9 th a t a sim ilar resu lt holds fo r every signal w hose d y n am ic ran g e sp a n s the ran g e o f th e q u a n tiz e r. T h is re la tio n sh ip is ex trem ely im p o rta n t b e c a u se it d ictates

Figure 1.22

T he quantization error eq (t) x a (t) - x q (t).

38

Introduction

Chap. 1

th e n u m b e r of bits re q u ire d by a specific a p p lic a tio n to assu re a given signal-ton o ise ratio . F o r ex am p le, m o st co m p a c t disc p lay ers use a sa m p lin g freq u e n cy o f 44.1 k H z an d 16-bit sa m p le re so lu tio n , w hich im plies a S Q N R of m o re th an 96 dB .

1.4.5 Coding of Quantized Samples

T h e co d in g p ro cess in an A /D c o n v e rte r assigns a u n iq u e b in a ry n u m b e r to each q u a n tiz a tio n level. If we h av e L levels w e n e e d at least L d iffe re n t binary' n u m b ers. W ith a w ord len g th o f b bits w e can c re a te 2b d iffe re n t b in ary n u m b e rs. H e n c e we h av e 2h > L. o r e q u iv alen tly , b > log 2 L. T h u s th e n u m b e r of bits re q u ire d in the co d e r is th e sm allest in te g e r g re a te r th a n o r eq u al to log 2 L. In o u r ex am p le it can easily be seen th a t we n eed a c o d e r w ith b = 4 bits. C o m m ercially av ailab le A /D c o n v e rte rs m ay be o b ta in e d w ith finite p recisio n of b 16 o r less. G e n e ra lly , the h ig h er th e sam p lin g sp e ed a n d th e fin er th e q u a n tiz a tio n , th e m o re ex p en siv e the d evice becom es.

1.4.6 Digital-to-Analog Conversion

T o co n v ert a d igital signal in to an an a lo g signal we can use a d ig ital-to -an alo g (D /A ) co n v erter. A s sta te d p rev io u sly , th e task o f a D /A c o n v e rte r is to in te rp o la te b e tw e e n sam ples. T h e sam p lin g th e o re m specifies th e o p tim u m in te rp o la tio n fo r a bandlim ited signal. H o w ev er, this ty p e o f in te rp o la tio n is to o c o m p lic a te d an d . h en ce im p ractical, as in d icated p rev io u sly . F ro m a p ractical v iew p o in t, the sim p lest D /A co n v e rte r is th e z e ro -o rd e r h o ld show n in Fig. 1.15. w hich sim p ly holds c o n sta n t th e v alu e o f o n e sam p le u n til th e n ex t o n e is receiv ed . A d d itio n a l im p ro v e m e n t can b e o b ta in e d by u sing lin e a r in te rp o la tio n as show n in Fig. 1.23 to c o n n e c t successive sa m p les w ith stra ig h t-lin e se g m en ts. T h e z e ro -o rd e r hold an d lin ear in te rp o la to r are an aly zed in S ectio n 9.3. B e tte r in te rp o la tio n can be ach iev ed by u sing m o re so p h isticated h ig h e r-o rd e r in te rp o la tio n tech n iq u es. In g en eral, su b o p tim u m in te rp o la tio n te c h n iq u e s re su lt in p assin g fre q u e n c ie s ab o v e th e fo ld in g freq u e n cy . S uch fre q u e n c y c o m p o n e n ts are u n d e s ira b le a n d are u su ally rem o v ed by p assin g th e o u tp u t o f th e in te rp o la to r th ro u g h a p ro p e r an alo g

Am plifude

Sec. 1.5

Summ ary and References

39

filter, w hich is called a postfilter or sm o o th in g filter. T h u s D /A c o n v ersio n usually involves a su b o p tim u m in te rp o la to r follow ed by a p ostfilter. D /A c o n v e rte rs are tre a te d in m o re d e ta il in S ectio n 9.3.

1.4.7 Analysis of Digital Signals and Systems Versus Discrete-Time Signals and Systems
W e h av e seen th a t a d igital signal is defin ed as a fu n ctio n o f an in te g e r in d e p e n d e n t v ariab le an d its v alu es are ta k e n from a finite set of po ssib le values. T he usefulness of such signals is a c o n s eq u en ce of th e possibilities o ffe re d by digital co m p u ters. C o m p u te rs o p e ra te on n u m b ers, w hich are re p re s e n te d by a strin g of 0 's an d l's . T h e len g th of th is strin g (w o r d length) is fixed an d finite an d usually is 8. 12. 16. or 32 bits. T h e effe cts o f finite w ord len g th in c o m p u ta tio n s cause co m p licatio n s in th e an aly sis of d ig ital signal p ro cessin g system s. T o avoid th ese co m p licatio n s, we neg lect th e q u a n tiz e d n a tu re of digital signals an d system s in m uch o f o u r analysis an d c o n sid e r th em as d isc rete-tim e signals an d system s. In C h a p te rs 6 . 7. and 9 we in v estig ate th e c o n se q u e n c e s o f using a finite w ord len g th . T h is is an im p o rta n t topic, since m any digital signal p ro cessin g p ro b lem s are solved w ith sm all c o m p u te rs o r m icro p ro cesso rs th at em p lo y fix ed -p o in t arith m etic. C o n se q u e n tly , o n e m ust look carefully at the p ro b le m of fin ite-p recisio n arith m e tic an d a c c o u n t for it in th e design of so ftw are an d h a rd w a re th at p e rfo rm s the d esired signal p ro cessin g tasks.

1.5 SUMMARY AND REFERENCES

In th is in tro d u c to ry c h a p te r w e have a tte m p te d to p ro v id e the m o tiv a tio n for digital signal p ro cessin g as an a lte rn a tiv e to a n a lo g signal pro cessin g . W e p re se n te d the basic e le m e n ts o f a digital signal p ro cessin g system an d d efin ed th e o p e ra tio n s n e e d e d to c o n v e rt an an alo g signal in to a digital signal re a d y fo r processing. O f p a rtic u la r im p o rta n c e is the sam pling th e o re m , w hich w as in tro d u c e d by N vquist (1928) an d la te r p o p u la riz e d in the classic p a p e r by S h a n n o n (1949). T h e sam pling th e o re m as d e sc rib e d in S ection 1.4.2 is d eriv ed in C h a p te r 4. S in u so id al signals w ere in tro d u c e d p rim a rily fo r the p u rp o se of illu stra tin g th e aliasin g p h e n o m e n o n an d fo r th e s u b s e q u e n t d e v e lo p m e n t o f th e sa m p lin g th e o re m . Q u a n tiz a tio n effects th a t are in h e re n t in the A /D co n v e rsio n of a signal w ere also in tro d u c e d in th is c h a p te r. Signal q u a n tiz a tio n is b est tre a te d in statistical term s, as d esc rib e d in C h a p te rs 6 , 7. an d 9. F in ally , th e to p ic o f signal re c o n stru c tio n , o r D /A co n v e rsio n , w as d escrib ed briefly. Signal re c o n stru c tio n b ased on sta ircase o r lin e a r in te rp o la tio n m eth o d s is tre a te d in S ectio n 9.3. T h e re a re n u m e ro u s p ractical a p p lic a tio n s of d igital signal processing. T he b o o k e d ite d by O p p e n h e im (1978) tre a ts a p p lic a tio n s to sp e ech p rocessing, im age p ro cessin g , ra d a r signal pro cessin g , so n a r signal p ro cessin g , a n d g eophysical signal p ro cessin g .

40

Introduction

Chap. 1

PROBLEMS

LI Classify the following signals according to whether they are (1) one- or multi dimensional; (2) single or multichannel, (3) continuous time or discrete time, and (4) analog or digital (in amplitude). Give a brief explanation. (a) Closing prices of utility stocks on the New York Stock Exchange. (b) A color movie. (c) Position of the steering wheel of a car in motion relative to cars reference frame. (d) Position of the steering wheel of a car in motion relative to ground reference frame. (e) Weight and height measurem ents of a child taken every month. 1.2 D eterm ine which of the following sinusoids are periodic and com pute their funda mental period. 30n \ ( 62m (a) cosO.OIjt/i (b) cs n -I (c) cos 3 (d) sin3 (e) sin 105 / ' ' V 10 1 3 Determ ine whether or not each of the following signals is periodic. In case a signal is periodic, specify its fundamental period. (a) xu(r) 3cos(5r + 7r/6) (b) = 3 cos(5n + ;r/6) (c) j:(h) = 2 e x p [j(n /6 - 7i)] (d) x(n) = cos(/8) cos(?rn/8) (e) x(n) = cos(7rn/2) sin(7rn/8) + 3cos(jrn/4 + 7t / 3) 1.4 (a) Show that the fundamental period Nr of the signals ,s>() = ei2nkr,IN. * = 0 .1 .2 ,...

is given by Np = N / C C D ( k . N ), where GCD is the greatest common divisor of k and N. (b) What is the fundam ental period of this set for N =71 (c) What is it for N = 16? 1.5 Consider the following analog sinusoidal signal: xa(t) 3 sin(1007rr) (a) Sketch the signal xa{t) for 0 < t < 30 ms. (b) The signal xa(t) is sampled with a sampling rate Fs = 300 samples/s. Determ ine the frequency of the discrete-time signal x{n) = xa( nT ), T = 1 /F. and show that it is periodic. (c) Compute the sample values in one period of .x(n). Sketch _*<n) on the same diagram with x(t). W hat is the period of the discrete-time signal in milliseconds? (d) Can you find a sampling rate Fs such that the signal x(n) reaches its peak value of 3? W hat is the minimum Fs suitable for this task? L6 A continuous-time sinusoid xa(t) with fundamental period Tp = 1/F0 is sampled at a rate F, = 1/ T to produce a discrete-time sinusoid x(n) = x(,nT). () Show that x(n) is periodic if T / T p = k / N (i.e., T/ T p is a rational num ber). (b) If x(n) is periodic, what is its fundam ental period Tp in seconds?

Chap. 1

Problem s

41

(c) Explain the statement: ,r(n) is periodic if its fundamental period Tr . in seconds,

is equal to an integer number of periods of .vu). 1.7 An analog signal contains frequencies up to 10 kHz. (a) W hat range of sampling frequencies allows exact reconstruction of this signal from its samples? (b ) Suppose that we sample this signal with a sampling frequency F, = 8 kHz. Ex amine what happens to the frequency F| = 5 kHz. (c) Repeat part (b) for a frequency F> = 9 kHz. 1.8 An analog electrocardiogram (ECG) signal contains useful frequencies up to 100 Hz. (a) W hat is the Nvquist rate for this signal? (b ) Suppose that we sample this signal at a rate of 250 samples/s. What is the highest frequency that can be represented uniquely at this sampling rate? 1.9 An analog signal a u ) = sin(480;rr) + 3sin(720:rr) is sampled 600 times per second. (a) D eterm ine the Nvquist sampling rate for x a{t). (b ) D eterm ine the folding frequency. (c) What are the frequencies, in radians, in the resulting discrete time signal t (/j )? (d) If is passed through an ideal D/A converter, what is the reconstructed signal v(n? 1.10 A digital communication link carries binary-coded words representing samples of an input signal (t) 3 cos 600.7 r 2 cos 1800jt /

The link is operated at 10.000 bits/s and each input sample is quantized into 1024 different voltage levels. (a) W hat is the sampling frequency and the folding frequency? ( b ) W hat is the Nvquist rate for the signal .*(;)? (c) What are the frequencies in the resulting discrete-time signal x(n)7 (d) W hat is the resolution A? 1.11 Consider the simple signal processing system shown in Fig. P I.11. The sampling periods of the A/D and D /A converters are T = 5 ms and T' = 1 ms. respectively. Determ ine the output vU) of the system, if the input is a:(n = 3 cos 100;rf -+ - 2 sin 250;rt (t in seconds)

The postfilter removes any frequency component above F J 2.

Figure P l . l l

1.12 (a) Derive the expression for the discrete-time signal .r(n) in Example 1.4.2 using the periodicity properties of sinusoidal functions. (b) W hat is the analog signal we can obtain from x(n) if in the reconstruction process we assume that Fs = 10 kHz?

42

Introduction

Chap. 1

1.13 The discrete-time signal x(n) = 6.35cos(jr/10)n is quantized with a resolution (a) A = 0.1 or (b) A = 0.02. How many bits are required in the A/D converter in each case? 1.14 D eterm ine the bit rate and the resolution in the sampling of a seismic signal with dynamic range of 1 volt if the sampling rate is Fs = 20 samples/s and we use an S-bit A/D converter? W hat is the maximum frequency that can be present in the resulting digital seismic signal? 1.15* Sampling o f sinusoidal signals: aliasing Consider the following continuous-time si nusoidal signal
XoO) = sin 2jr/r0f, oc < t < oo

Since xa(t) is described m athematically, its sampled version can be described by values every T seconds. The sampled signal is described by the formula

where Fs = l / T is the sampling frequency. (a) Plot the signal j:( n), 0 < n < 99 for F, = 5 kHz and Fu = 0.5, 2, 3, and 4.5 kHz. Explain the similarities and differences am ong the various plots. (b) Suppose that F0 = 2 kHz and Fs = 50 kHz. (1) Plot the signal x(n). W hat is the frequency / (l of the signal je(n)? (2) Plot the signal v(n) created by taking the even-numbered samples of x(n). Is this a sinusoidal signal? Why? If so, what is its frequency? 1.16* Quantization error in A /D conversion o f a sinuoidal signal Let xq(n) be the signal obtained by quantizing the signal x(n) = sin27r/on. The quantization error power PQ is defined by

The quality of the quantized signal can be measured by the signal-to-quantization noise ratio (SQ NR) defined by SQ N R = 10 log,0 a where Px is the power of the unquantized signal x (n). () For /o = 1/50 and N = 200, write a program to quantize the signal Jt(n), using truncation, to 64, 128, and 256 quantization levels. In each case plot the signals x (n), Xq(n), and e(n) and com pute the corresponding SQNR. (b) Repeat part (a) by using rounding instead of truncation. (c) Comment on the results obtained in parts (a) and (b). (d) Compare the experimentally m easured SQNR with the theoretical SQNR pre dicted by formula (1.4.32) and com m ent on the differences and similarities.

Discrete-Time Signals and Systems

In C h a p te r 1 w e in tro d u c e d th e re a d e r to a n u m b e r o f im p o rta n t ty p es o f signals an d d escrib ed th e sa m p lin g p ro cess bv w hich an a n a lo g signal is c o n v e rte d to a d isc rete-tim e signal. In a d d itio n , we p re s e n te d in so m e d etail th e c h a racteristics o f d isc re te -tim e sin u so id al signals. T h e sin u so id is an im p o rta n t e le m e n ta ry signal th a t se rv es as a b asic b u ild in g block in m o re co m p lex signals. H o w e v e r, th e re are o th e r e le m e n ta ry signals th a t are im p o rta n t in o u r tr e a tm e n t o f signal processing. T h ese d isc re te -tim e signals a re in tro d u c e d in this c h a p te r a n d are used as basis fu n ctio n s o r b u ild in g b locks 1o d escrib e m o re com plex signals. T h e m a jo r e m p h asis in this c h a p te r is th e c h a ra c te riz a tio n o f d isc rete-tim e sy stem s in g en era! a n d the class o f lin ear tim e -in v a ria n t (L T I) system s in p articu lar. A n u m b e r o f im p o rta n t tim e-d o m ain p ro p e rtie s o f L T I sy stem s a re d efined and d e v e lo p e d , an d an im p o rta n t fo rm u la, called th e c o n v o lu tio n fo rm u la, is d eriv ed w hich allow s us to d e te rm in e th e o u tp u t of an L T I system to any given a rb itra ry in p u t signal. In a d d itio n to th e c o n v o lu tio n fo rm u la , d iffe re n c e e q u a tio n s are in tro d u c e d as an a lte rn a tiv e m e th o d fo r describ in g th e in p u t- o u tp u t re la tio n sh ip of an L T I sy stem , an d in a d d itio n , recursive an d n o n re c u rsiv e re a liz a tio n s of LTI sy stem s are tre a te d . O u r m o tiv a tio n fo r th e em p h asis on th e stu d y o f L T I sy stem s is tw ofold. F irst, th e re is a larg e co llectio n o f m a th e m a tic a l te c h n iq u e s th a t can be ap p lied to the an aly sis o f L T I system s. S eco n d , m an y p ractical system s a re e ith e r L T I system s o r can b e a p p ro x im a te d by L T I system s. B ecau se of its im p o rta n c e in digital signal p ro cessin g a p p licatio n s an d its close re se m b la n c e to th e co n v o lu tio n form ula, we also in tro d u c e th e c o rre la tio n b e tw e e n tw o signals. T h e a u to c o rre la tio n and c ro ssc o rre la tio n o f signals a re defined an d th e ir p ro p e rtie s a re p re se n te d .

2.1 DISCRETE-TIME SIGNALS

A s w e d iscu ssed in C h a p te r 1, a d isc re te -tim e signal x{n) is a fu n ctio n o f an in d e p e n d e n t v a ria b le th a t is an in teg er. It is g rap h ically re p re s e n te d as in Fig. 2.1. It is im p o rta n t to n o te th a t a d isc re te -tim e signal is n o t defined at in sta n ts b etw een 43

44

Discrete-Time Signals and Systems

Chap. 2

Figure 2-1

Graphical representation of a discrete-time signal.

tw o successive sam p les. A lso , it is in c o rre c t to th in k th a t .v(n) is e q u a l to z e ro if n is n o t an in teg er. S im ply, th e signal x ( n ) is n o t defin ed fo r n o n in te g e r v alu es o f n. In th e se q u el w e will assu m e th a t a d isc re te -tim e signal is d efin ed fo r every in te g e r value n fo r oo < n < oc. By tra d itio n , w e re fe r to x( n) as th e n th sa m p le o f th e signal ev en if th e signal x( n) is in h e re n tly d isc re te tim e (i.e., n ot o b ta in e d by sam p lin g an a n a lo g signal). If, in d e e d , x ( n) w as o b ta in e d fro m sa m p lin g an a n alo g signal x a( t ), th e n .i(n ) = x a( nT) , w h ere T is th e sa m p lin g p e rio d (i.e., th e tim e b etw e e n successive sam p les). B e sid es th e g rap h ical re p re s e n ta tio n of a d isc re te -tim e signal o r se q u e n c e as illu strated in Fig. 2.1. th e re a re som e a lte rn a tiv e re p re s e n ta tio n s th at are o ften m o re co n v e n ie n t to use. T h e se are: 1. F u n ctio n al re p re s e n ta tio n , such as x(n) 2. T a b u la r r e p re s e n ta tio n , such as n x ( n) -2 0 -1 0 0 0 1 1 2 4 3 1 4 0 5 0

I4,

f 1, [ 0,

fo r n = 1, 3 fo r n = 2 else w h e re

(2 . 1.1)

3. S eq u en ce re p re s e n ta tio n A n in fin ite -d u ra tio n signal o r se q u e n c e w ith th e tim e o rig in (n = 0) in d ic a te d by th e sy m b o l | is re p re s e n te d as *<n) = { . . . 0 . 0 . 1 . 4 , 1 . 0 , 0 , . . . } T A se q u e n c e j:(n ), w hich is z e ro fo r n < 0, can be re p re s e n te d as
jc()

(2.1.2)

= { 0 ,1 .4 .1 .0 .0 ....} T

(2.1.3)

T h e tim e o rig in fo r a se q u e n c e x ( n ) , w hich is z e ro fo r n < 0, is u n d e rs to o d to be th e first (le ftm o st) p o in t in th e seq u en ce.

Sec. 2.1

Discrete-Time Signals

45

A fin ite -d u ra tio n se q u e n c e can be re p re se n te d as x i n) = {3. - 1 . - 2 . 5 .0 .4 . -1 }

(2.1.4)

w h ereas a fin ite -d u ra tio n se q u e n c e th a t satisfies the c o n d itio n x(/i) ~ 0 for n < 0 can be r e p re s e n te d as jc(n) = { 0 .1 .4 . 1)

(2.1.5)

T h e signal in (2.1.4) co n sists of seven sa m p le s or p o in ts (in tim e), so it is called or id en tified as a se v e n -p o in t se q u e n c e . S im ilarly, the se q u e n c e given by (2.1.5) is a f o u r-p o in t se q u e n c e .

2.1.1 Some Elementary Discrete-Time Signals

In o u r stu d y o f d isc re te -tim e signals an d system s th e re a re a n u m b e r o f b asic signals th at a p p e a r o ften a n d play an im p o rta n t role. T h ese signals a re d efin ed below . 1. T h e unit s a m p l e se quence is d e n o te d as < 5 (n) an d is defin ed as
< 5( / i )
=

0.

for n - 0 for n ^ 0

( 2 . 1. 6 )

In w o rd s, th e u n it sa m p le se q u e n c e is a signal th at is zero e v erv w h ere, ex cep t a t n 0 w h e re its value is unity. T his signal is so m e tim e s re fe rre d to as a unit impul se. In c o n tra st to the an alo g signal 8(t). w hich is also called a u n it im p u lse an d is d efin ed to be ze ro ev ery w h ere ex cep t / = 0. an d has unit a re a , th e u n it sa m p le se q u e n c e is m uch less m ath e m a tic a lly c o m p licated . T he g rap h ical re p re s e n ta tio n o f < 5(n ) is show n in Fig. 2.2. 2. T h e unil step signal is d e n o te d a s w (n ) an d is defin ed as u(n) = 1. 0. fo r n > 0 fo r n < 0 (2.1.7)

F ig u re 2.3 illu stra te s th e u nit ste p signal. 3. T h e uni t r a m p signal is d e n o te d as u r (n) an d is d efin ed as u r (n) T h is signal is illu stra te d in Fig. 2.4. fi(n) fo r n > 0 fo r n < 0
( 2 . 1.

Figure 2.2 G raphical rep resen tatio n of the unit sample signal.

46
u(n)

Discrete-Tim e Signals and Systems

Chap. 2

12

3 4 5 6 7

Figure 2 3 G raphical rep resen tatio n of the unit step signal.

ur(n)

T
Figure 2,4 G raphical rep resen tatio n of the unit ram p signal.

4. T h e exponent i al signal is a se q u e n c e o f th e fo rm
jr(n) = a fo r all n (2.1.9)

If th e p a r a m e te r a is real, th e n jr(n) is a real signal. F ig u re 2.5 illu stra te s x ( n) fo r v ario u s v alu es o f th e p a r a m e te r a. W h en th e p a ra m e te r a is co m p lex v a lu e d , it can b e e x p re ss e d as a s rejf1 w h e re r an d 6 a re n o w th e p a ra m e te rs. H e n c e w e c a n e x p ress x ( n ) as x ( n ) = r nej0n = r n (cos On -j- y s in # n )
(2 . 1. 10 )

T iin ilU

Figure 2.5

G raphical representation of exponential signals.

Sec. 2.1

Discrete-Time Signals

47

Since x (/j ) is now co m plex valu ed , it can be re p re s e n te d g rap h ically by p lo ttin g th e real p a rt x K(n) = r" cos6#fi as a fu n ctio n of n. an d se p a ra te ly p lo ttin g th e im ag in ary p a rt xi (n) = r 1sin 6n (2.1.12) (2.1.11)

as a fu n ctio n o f n. F ig u re 2.6 illu strates th e g ra p h s o f x R(n) a n d x / ( n ) for r 0.9 an d 6 = tt/1 0 . W e o b se rv e th a t th e signals x R(?i) a n d x / ( n ) a re a d a m p e d (decaying e x p o n e n tia l) co sin e fu n ctio n an d a d a m p e d sine fu n c tio n . T h e angle v ariab le 6 is sim ply th e freq u e n c y of th e sinusoid, p rev io u sly d e n o te d by th e (n o rm alized ) fre q u e n c y v ariab le w. C learly, if r 1. th e d a m p in g d is a p p e a rs an d x K(n). x/ ( n). an d A'(n) h av e a fixed am p litu d e, w hich is unity. A lte rn a tiv e ly , th e signal .*(/?) given by (2.1.10) can be re p re s e n te d g raphically by th e a m p litu d e fu n ctio n |.i(h )| = A{n) = r an d th e p h ase fu n ctio n _ v (/;) = < f>{n) = Bn (2.1.14) (2.1.13)

F ig u re 2.7 illu stra te s -4(/?) an d 0 (/i) fo r r 0.9 a n d B = ,t/1 0 . W e o b se rv e th at th e p h ase fu n ctio n is lin ear w ith n. H o w ev er, th e p h ase is d efin e d only o v e r the in terv al n < B < t t o r. eq u iv alen tly , o v er th e in terv al 0 < 6 < 2 t t . C o n seq u en tly , by co n v e n tio n 4>(n) is p lo tte d o v er th e finite in terv al n < B < t t o t Q < $ < 2tt. In o th e r w o rd s, w e su b tra c t m u ltip lies o f I n fro m <p(n) b e fo re p lo ttin g . In one case. <p(n) is c o n s tra in e d to th e range n < 0 < n a n d in th e o th e r case <p(n) is c o n stra in e d to th e ran g e 0 < fi < 2n. T h e su b tra c tio n o f m u ltip le s o f 2n from <p(n) is e q u iv a le n t to in te rp re tin g th e fu n ctio n 4>(n) as 4>{n), m o d u lo 2n. T he g rap h for <p{n). m o d u lo 2 n . is show n in Fig. 2.7b.

2.1.2 Classification of Discrete-Time Signals

T h e m a th e m a tic a l m e th o d s e m p lo y ed in th e analysis of d isc re te -tim e signals and system s d e p e n d on th e c h aracteristics o f th e signals. In this se ctio n we classify d isc rete-tim e signals acco rd in g to a n u m b e r of d iffe re n t c h aracteristics.

Energy signals and power signals.

d efin ed as
OC

T h e en e rg y of a signal x( n ) is

E=
n=-oc

|jc(n)|2

(2.1.15)

W e h av e u se d th e m a g n itu d e -sq u a re d v alu es o f jr(n), so th a t o u r d efin itio n applies to c o m p le x -v a lu e d signals as w ell as re a l-v a lu e d signals. T h e e n e rg y of a signal can be fin ite o r in fin ite. If E is finite (i.e., 0 < E < oo), th e n x ( n ) is called an energy

Figure 2.6 signal.

G raph of the real and im aginary com ponents of a com plex-valued exponential

Sec. 2.1

Discrete-Time Signals

49

-1 0

1 2

~ S

11
9

(a* Gra ph of A {n I = r" . r = 0.9

( b t G r a p h ol

n . m o d u l o 2ir p l o n e d in t h e r a n g e I t t , it I

Figure 2.7 G r a p h o f a m p l i tu d e and p h a s e function of a co m p l e x -v a l u e d e x p o n e n tial sicnal: ( a ) gr ap h of A i m = r " . 4 = 0 . 9 : ( b ) g rap h of c/>in) = I . t / I O v j . m o d u lo 2ji p lo tt e d in the ran ge i - j i . t t J .

signal. S o m etim es w e add a su b scrip t .v to an d w rite , to em p h asize th a t , is the en e rg y o f th e signal x( n). M anv signals th a t possess infinite en erg y , h av e a finite av e ra g e p o w er. T he av erag e p o w e r o f a d isc re te -tim e signal x( n) is defin ed as P = 1 lim * * 2 N -j- 1 n - -,\

(2 .1.1 6)

If we define th e signal en e rg y o f x(?i) o v er th e finite in terv al N < n < N as (2.1.17)

th e n w e can e x p ress th e signal energy as = lim /v' ,\ oc (2.1.18)

an d th e a v erag e p o w e r o f th e signal x ( n ) as 1 En P = lim n -+x 2 N + 1

(2.1.19)

50

Discrete-Tim e Signals and Systems

Chap. 2

C learly , if E is finite. P = 0. O n th e o th e r h a n d , if E is infinite, th e av erag e p o w e r P m ay be e ith e r finite o r infinite. If P is finite (an d n o n z e ro ), th e signal is called a p o w e r signal. T h e follow ing e x am p le illu stra te s such a signal. Example 2.1.1 Determ ine the power and energy of the unit step sequence. The average power of the unit step signal is

A 1 -** 2 N

N + 1 lim --------------

a 9c 2

1 + l/N 1 lim ------------- --- 4- l / N 2

Consequently, the unit step sequence is a power signal. Its energy is infinite. Sim ilarly, it can b e show n th a t th e co m p lex e x p o n e n tia l se q u e n c e x(n') = A e JU Jan h as av erag e p o w e r A 2, so it is a p o w e r signal. O n th e o th e r h an d , th e unit ra m p se q u en ce is n e ith e r a p o w e r signal n o r an en e rg y signal.

Periodic signals and aperiodic signals. A s d efin e d o n S ection 1.3, a signal x( n ) is p erio d ic w ith p e rio d N ( N > 0) if a n d only if
x( n + N ) = x( n ) fo r all n (2 . 1.20 )

T h e sm allest v alu e o f N fo r w hich (2.1.20) h o ld s is called th e (fu n d a m e n ta l) p erio d . If th e re is no v alu e o f N th a t satisfies (2.1.20), th e signal is called nonper i odi c or aperiodic. W e h av e a lread y o b se rv e d th a t th e sin u so id al signal o f th e form x ( n) = A sin 2n f o n is p e rio d ic w h en /J, is a ra tio n a l n u m b e r, th a t is, if /o can be e x p re sse d as
(2 .1. 22 )
( 2 . 1 .21 )

w h ere k an d N a re integers. T h e e n erg y o f a p erio d ic signal x ( n ) o v e r a single p e rio d , say. o v er th e in terv al 0 5 n < N - 1, is fin ite if x () ta k e s on finite v alu es o v e r th e p e rio d . H o w ev er, the e n e rg y o f th e p e rio d ic signal fo r oc < n < oo is infinite. O n th e o th e r h an d , the a v e ra g e p o w e r o f th e p erio d ic signal is finite an d it is e q u a l to th e av erag e p o w er o v e r a single p e rio d . T h u s if x () is a p e rio d ic signal w ith fu n d a m e n ta l p e rio d N an d ta k e s o n fin ite v alues, its p o w e r is g iven by

(2.1.23)
n=0 C o n s e q u e n tly , p erio d ic signals are p o w e r signals.

Sec. 2.1

Discrete-Time Signals

51

Symmetric (even) and antisymmetric (odd) signals.

n al x ( n ) is called sy m m etric (ev en ) if jr(-n ) =
j

A real v a lu ed sig

()

(2.1.24)

O n th e o th e r h a n d , a signal x ( n ) is called a n tisy m m etric (o d d ) if .v( - h ) = - x ( n ) (2.1.25)

W e n o te th a t if .v(/7) is odd, th e n x(0) = 0. E x am p le s o f signals w ith ev en an d odd sy m m etry are illu stra te d in Fig. 2.8. W e w ish to illu strate th a t any a rb itra ry signal can be e x p re sse d as th e sum of tw o signal c o m p o n e n ts , o n e o f w hich is even an d th e o th e r o d d . T h e ev en signal co m p o n e n t is fo rm e d by ad d in g x(/i) to x ( n) and div id in g by 2. th a t is. = j[.v{7!) + x ( - ) ] .v(n] (2.1.26)

< ' 4 *

ll
-4-3-2-I 0
(a)

!I j T ] ! l M ! ....................
3 4 i,

12

.r(n)

- 5 - 4 - 3 - 2 -1

.ill'
12 3 4 5 rt

< >

(b) Figure 2.8 Exam ple of even (a) and odd (b) signals.

52

Discrete-Time Signals and Systems

Chap. 2

Clearly, x e(n ) satisfies the sym m etry con d ition (2.1.24). Sim ilarly, w e form an odd signal com p onent x(n) according to the relation x(n) = j[.x(n) - * ( - / i ) ] (2.1.27)

A gain , it is clear that x 0(n ) satisfies (2.1.25); hence it is in d eed odd. N ow , if we add the tw o signal com p onents, defined by (2.1.26) and (2.1.27), w e ob tain ;c(n), that is, *(n ) = x e(n) + x n{n) Thus any arbitrary signal can be exp ressed as in (2.1.28). (2.1.28)

2.1.3 Simple Manipulations of Discrete-Time Signals

In this section w e consider som e sim ple m odifications or m anipulations involving the in d ep en dent variable and the signal am plitude (depend en t variable).

Transformation of the independent variable (time). A signal x ( n ) may be shifted in tim e by replacing the in d ep en dent variable n by n k, w here k is an integer. If A : is a positive integer, the tim e shift results in a delay of the signal by k units o f tim e. If k is a negative integer, the tim e shift results in an advance of the signal by \k\ units in time.

Example 2.L2
A signal x ( n ) is graphically illustrated in Fig. 2.9a. Show a graphical representation of the signals x ( n 3) and x ( n -I- 2).

Solution The signal x (/i 3) is obtained by delaying ;t(n) by three units in time. The
result is illustrated in Fig. 2.9b. On the other hand, the signal x(n + 2 ) is obtained by advancing x ( n ) by two units in time. The result is illustrated in Fig. 2.9c. Note that delay corresponds to shifting a signal to the right, whereas advance implies shifting the signal to the left on the time axis. If the signal x ( n ) is stored on m agnetic tape or on a disk or, perhaps, in the m em ory o f a com puter, it is a relatively sim ple operation to m odify the base by introducing a delay or an advance. O n the other hand, if the signal is not stored but is b ein g generated by som e physical p h en om en on in real tim e, it is not p ossible to advance the signal in tim e, since such an op eration in volves signal sam ples that have not yet b een generated. W hereas it is alw ays possib le to insert a delay into signal sam ples that have already b een generated, it is physically im possible to view the future signal sam ples. C on sequ en tly, in real-tim e signal processing applications, the operation o f advancing the tim e base o f the signal is physically unrealizable. A n o th er useful m odification o f the tim e base is to replace th e in d ep en dent variable n by n. T h e result o f this op eration is a f o l d i n g or a reflection o f the signal about the tim e origin n = 0.

Sec. 2.1

Discrete-Time Signals
xin)
4 i

53

j 1

I 4 3 2 1 0

1 2

x i n - 3)

il
T -l 0 1 2 3 4 S 6

(b)

xin +2)

- 6 - 5 - 4 -3 - 2 - 1

Figure 2.9 Graphical representation of a signal, and its delayed and advanced versions.

Example 2.1.3
Show the graphical representation of the signal x { - n ) and x i - n + 2). where x ( n ) is the signal illustrated in Fig. 2.10a.

Solution

The new signal yin) = x ( n) is shown in Fig. 2.10b. Note that y(0) = *(0). v ( l ) = x( 1). y(2) = _t( 2). and so on. Also, y (1) = jc(1 ) , v(2) = x(2), and so on.

Therefore, yin) is simply xin) reflected or folded about the lime origin n = 0. The signal yin) x ( n + 2) is simply x ( n) delayed by two units in time. The resulting signal is illustrated in Fig. 2.10c. A simple way to verify that the result in Fig. 2.10c is correct is to compute samples, such as y(0) = *(2), y (l) = jc(1 >, v(2) = ;t(0), v(1) = jr(3). and so on.

It is im p o rta n t to n o te th a t th e o p e ra tio n s o f folding an d tim e d elay in g (o r ad v an cin g ) a signal a re n o t co m m u ta tiv e . If we d e n o te th e tim e-d e la y o p e ra tio n by T D a n d th e fo ld in g o p e ra tio n by F D . we can w rite T D i[jc(n )l = jc(n - k) FD [jc(/j)] = x ( n) N ow k > 0 (2.1.29)

T D A(FD[.r(n)]l = T D *[.t(-n)] = x ( - n + k)

(2.1.30)

54

Discrete-Time Signals and Systems

Chap. 2

y(n) = x(-n + 2 )

-2 -1

0 1 2 3 4 5 Figure 2.10 Graphical illustration of the folding and shifting operations.

in

(c)

w hereas F D {T D i[j:(n )]} = FD[j:(/i &)] = x ( n k ) (2.1.31)

N o te that becau se the signs o f n and k in x { n k) and Jt(-n + ik ) are different, the re sult is a shift o f the signals x ( n ) and x ( n) to the right by k sam p les, corresponding to a tim e delay. A third m odification o f the in d ep en dent variable in volves replacing n by fin, w here /x is an integer. W e refer to this tim e-b ase m odification as time scaling or dow nsam pling.

Example 2.L4
Show the graphical representation of the signal y(n) = x(2n), where x(n) is the signal illustrated in Fig. 2.11a.

Solution We note that the signal y(n) is obtained from x(n) by taking every other sample from jc(), starting with x(0). Thus y(0) = x(0), y{l) = x(2), y(2) = jc(4), and y ( - l ) = x(~ 2 ), y ( -2 ) = jc( 4), and so on. In other words, we have skipped

Sec. 2.1

Discrete-Time Signals

55

v(n) = .v(2n)

-4

! -2

0 I 2 3

(b)
Figure 2.11 Graphical illustration ot down-samplinc operation.

the odd-num bered samples in *() and retained the even-num bered samples. The resulting signal is illustrated in Fig. 2.11b. If th e signal ,\ () w as o riginally o b ta in e d by sa m p lin g an a n a lo g signal x a(t), th e n jc() = Xa(nT), w h ere T is the sa m p lin g in terv al. Nowr. v(n) = x ( 2n) = x a(2Tn). H e n c e th e tim e-scalin g o p e ra tio n d escrib ed in E x a m p le 2.1.4 is e q u iv a le n t to ch an g in g th e sam p lin g ra te from 1 /T to 1/27". th a t is, to d e c re a s in g the ra te by a facto r o f 2. T h is is a d o w n s a mp l i n g o p e ra tio n .

Addition, multiplication, and scaling of sequences. A m p litu d e m o d ifi catio n s in clu d e addition, multiplication, an d scaling o f d isc re te -tim e signals. A m p l i t u d e scaling o f a signal by a c o n sta n t A is acco m p lish ed by m u ltiplying th e v alu e o f ev ery signal sa m p le by A. C o n se q u e n tly , w e o b ta in
v(n) = Ax ( n ) oc < fi < oc

T h e s u m o f tw o signals xj ( n) a n d xz i n) is a signal _v(n), w h o se v alu e at any in stan t is equal to th e sum o f th e values o f th e se tw o signals a t th a t in sta n t, th a t is. y(/t) = jq (n) + X2 <n) oc < n < oc

T h e p r o d u c t o f tw o signals is sim ilarly defin ed o n a sa m p le -to -sa m p le basis as v(n) = X](n)X 2 (n) oo < n < oo

56

Discrete-Time Signals and Systems

Chap. 2

2.2 DISCRETE-TIME SYSTEMS

In m a n y ap p lic a tio n s o f d ig ital signal p ro cessin g w e w ish to desig n a d ev ice or an alg o rith m th a t p e rfo rm s so m e p re sc rib e d o p e ra tio n o n a d isc re te -tim e signal. S uch a d evice o r a lg o rith m is called a d isc re te -tim e system . M o re specifically, a discrete-time s y s t em is a d ev ice o r a lg o rith m th a t o p e ra te s on a d isc re te -tim e signal, called th e i nput o r excitation, acco rd in g to so m e w ell-defined ru le , to p ro d u c e a n o th e r d isc rete-tim e signal called th e out p u t o r response of th e system . In g en eral, we view a system as an o p e r a tio n o r a se t o f o p e ra tio n s p e rfo rm e d on th e in p u t sig n al x( n) to p ro d u c e th e o u tp u t signal _v(n). W e say th a t th e in p u t signal x( n) is t r ans f ormed by th e sy stem in to a signal >(), an d ex p ress th e g e n e ra l re la tio n sh ip b e tw e e n jc () a n d y ( n ) as y( n) = T[x{n) } (2 . 2. 1) w h ere th e sym bol T d e n o te s th e tra n s fo rm a tio n (also called an o p e r a to r) , o r p r o cessing p e rfo rm e d by th e sy stem on ;c(n) to p ro d u c e y(n). T h e m a th e m a tic a l re la tio n sh ip in (2.2.1) is d e p ic te d g rap h ically in Fig. 2.12. T h e re are v ario u s w ays to d escrib e th e c h a ra c te ristic s of th e system an d th e o p e ra tio n it p e rfo rm s on x( n ) to p ro d u c e y(n ). In this c h a p te r w e shall b e c o n c e rn e d w ith th e tim e -d o m a in c h a ra c te riz a tio n o f system s. W e shall begin w ith an in p u t-o u tp u t d e sc rip tio n o f th e system . T h e in p u t-o u tp u t d e sc rip tio n focuses on th e b e h a v io r at th e te rm in a ls of th e system a n d ignores th e d e ta ile d in te rn a l co n stru ctio n o r re a liz a tio n o f th e system . L a te r, in S ection 7.5. w e in tro d u c e th e sta te -sp a c e d e sc rip tio n o f a system . In this d e sc rip tio n w e d e v e lo p m a th e m a ti cal e q u a tio n s th a t n o t o nly d esc rib e th e in p u t- o u tp u t b e h a v io r o f th e sy stem b u t specify its in te rn a l b e h a v io r a n d stru c tu re .

2.2.1 Input-Output Description of Systems

T h e in p u t-o u tp u t d e sc rip tio n o f a d isc re te -tim e sy stem consists o f a m a th e m a tic a l ex p ressio n o r a ru le, w hich explicitly d efin es th e re la tio n b e tw e e n th e in p u t an d o u tp u t signals ( i n p u t - o u t p u t relationship). T h e ex act in te rn a l s tru c tu re o f th e sys tem is e ith e r u n k n o w n o r ig n o red . T h u s th e o n ly w ay to in te ra c t w ith th e sy stem is by u sin g its in p u t an d o u tp u t te rm in a ls (i.e., th e system is a ssu m ed to be a black b o x to th e u se r). T o reflec t th is p h ilo so p h y , w e u se th e g ra p h ic a l re p re se n ta -

x (n J

Input signal or excitation Figure 2.12

Discrete-time System Output signal or response

Block diagram representation of a discrete-tim e system.

Sec. 2.2

Discrete-Time Systems

57

tio n d e p ic te d in Fig. 2.12, an d th e g e n e ra l in p u t-o u tp u t re la tio n sh ip in (2.2.1) or, a lte rn a tiv e ly , th e n o ta tio n Jt(n) y(n) (2.2.2)

w hich sim ply m e a n s th a t v(n) is th e re sp o n se of the system T to th e e x c ita tio n x{n). T h e fo llo w in g e x am p les illu stra te se v era l d iffe re n t system s. Example 2.2.1 D eterm ine the response of the following sytems to the input signal x(n) = , A u, (a) (b) (c) (d) -3 < n < 3 otherwise

y(n) = x{n) v() = x in i) y(n) = x i n 4- i) y i n ) = j[A-(n + 1) + x ( n ) + x i n - D] (e) y ( n) = m a x { x( n + 1), x ( n ) . x ( n 1)1 (0 y ( n ) = Z L . x x ( k ) = x ( n ) + x ( n 1) + x{n 2) -t First, we determ ine explicitly the sample values of the input signal xin) = ( ....0 .3 ,2 .1 .0 .1 .2 , 3 ,0 ,...)
T

(2.2.3)

Solution

Next, we determ ine the output of each system using its input-output relationship. (a) In this case the output is exactly the same as the input signal. Such a system is known as the identity system. (b) This system simply delays the input by one sample. Thus its output is given by x{n) = { ...,0 ,3 .2 .1 ,0 ,1 .2 .3 ,0 ,..,) t (c) In this case the system advances the input one sample into the future. For example, the value of the output at time n = 0 is y(0) = *(1). The response of this system to the given input is x(n) = { ...,0 ,3 . 2 .1 .0 ,1 ,2 . 3 ,0 ....} t (d) The output of this system at any time is the mean value of the present, the im m ediate past, and the immediate future samples. For example, the output at time n = 0 is y(0) = + x(0) + jr(l)] = |[1 + 0 + 1] = |

R epeating this com putation for every value of n, we obtain the output signal >() = { ...0 ,1 , f , 2 , l j . l . 2 , , 1 .0 ,...) t

58

Discrete-Time Signals and Systems

Chap. 2

(e> This system selects as its output at time n the maximum value of the three input

samples x(n - l.l, .v(n). and ,r(n + 1). Thus the response of this system to the input signal .\{n) is
v(n) = {0.3. 3. 3. 2 .1 .2 . 3, 3, 3 . 0 . . . . ) t

(f) This system is basically an accumulator that computes the running sum of all the past input values up to present time. The response of this system to the given input is v(n) = {.. ..0 .3 . 5. 6. 6, 7, 9. 1 2 .0 ....} T W e o b se rv e th a t fo r several of th e sy stem s c o n sid e re d in E x a m p le 2.2.1 the o u tp u t at tim e n no d e p e n d s n ot only on th e v alu e of the in p u t at n = n (, [i.e., jc(o)]- b u t also on th e values o f the in p u t a p p lie d to th e system b e fo re an d after n = n (). C o n sid er, fo r in stan ce, th e a c c u m u la to r in th e ex a m p le . W e see th at the o u tp u t at tim e n = ?i() d e p e n d s n ot only on th e in p u t a t tim e n = no. b u t also on x ( n ) a t tim es n = no 1. no - 2, and so on. By a sim ple a lg e b ra ic m a n ip u latio n th e in p u t-o u tp u t re la tio n o f th e a c c u m u la to r can b e w ritte n as

= y(/i - 1) + x(ti) w hich justifies th e te rm accumul at or. In d e e d , th e system c o m p u te s th e c u rre n t v alu e o f th e o u tp u t by a d d in g (a ccu m u latin g ) th e c u rre n t v alu e o f th e in p u t to th e p rev io u s o u tp u t value. T h e re are so m e in te re stin g co n clu sio n s th a t can be d raw n by tak in g a close lo o k in to this a p p a re n tly sim ple system . S u p p o se th a t we are given th e in p u t signal x(rt ) fo r n > no. a n d we wish to d e te rm in e th e o u tp u t v(/i) o f th is system fo r n > no. F o r n = no. no + 1........ (2.2.4) gives v (n n) = v(o - 1) -f x (/i0)
_v(no + l l = v ( o ) + x ( n o + 1)

an d so on. N o te th a t we h ave a p ro b le m in c o m p u tin g y ( n a), since it d e p e n d s on y(o - 1). H o w ev er, y(t io - 1) = ^

k X.

x(k)

th a t is. y(no - 1) sum m arizes* th e effect on th e system from all the in p u ts w hich h ad b e e n ap p lied to th e system b efo re tim e no- T h u s th e re sp o n s e of th e system fo r n > no to th e in p u t x(/7j th a t is a p p lie d a t tim e no is th e c o m b in e d resu lt of this in p u t an d all in p u ts th a t h ad b e e n a p p lie d p re v io u sly to th e sy stem . C o n seq u en tly . y(/i), n > no is n o t u n iq u ely d e te rm in e d by th e in p u t x ( n ) fo r n > no.

Sec. 2.2

Discrete-Time Systems

59

T he additional inform ation required to d eterm ine y ( n) for n > no is the initial condi t i on y(no - 1). T his value sum m arizes the effect o f all p reviou s inputs to the. system . Thus the initial con d ition y(o - 1) togeth er with the input seq u en ce x ( n ) for n > no uniquely d eterm ine the output sequ en ce y(n ) for n > n0. If the accum ulator had no excitation prior to n 0, the initial con d ition is y(no 1) = 0. In such a case w e say that the system is initially relaxed. S ince y(no 1) = 0, the output seq u en ce y(n) d ep en d s only on the input seq u en ce x ( n ) for n > n0. It is custom ary to assum e that every system is relaxed at n = oo. In this case, if an input x ( n ) is applied at n = co, the corresponding output y( n) is solely and uni qu e l y determ ined by the given input.

Example 2.2.2
T he accum ulator described by (2.2.3) is excited by the sequence x(n) = nu(n). D e term ine its output under the condition that:

(a ) It is initially relaxed [i.e., v ( - l ) = 0]. <b) Initially, y ( 1 )= 1.

Solution The output of the system is defined as tl -] y(n) = ^ x(k) = x(k) +
*=-oc *=-oc

r
i= < l

= y (-l) + k=o But

x(k)

n(n -f 1)

(a) If the system is initially relaxed, v(1) = 0 and hence

n(n + l)
v (n) = -------- 2 -------" 0

(b) On the other hand, if the initial condition is y ( - l ) = 1, then n(n -I-1) n2 + n + 2 v(n) = 1 + ---- - - = ------------n > 0

2.2.2 Block Diagram Representation of Discrete-Time Systems

It is useful at this point to introduce a block diagram representation o f d iscrete tim e system s. For this purpose w e n eed to define som e basic building blocks that can b e intercon n ected to form com p lex system s.

An adder. F igure 2.13 illustrates a system (adder) that perform s the addi tion o f tw o signal seq u en ces to form another (th e sum ) seq u en ce, w hich w e d en ote

60
x|(n )

Discrete-Time Signals and Systems

Chap. 2

y(n) = i,( n ) + x2(n)

Figure 2.13 Graphical representation of an adder.

as y (n ). N o te th a t it is n o t n ecessa ry to sto re e ith e r o n e o f th e se q u e n c e s in o rd e r to p e rfo rm th e a d d itio n . In o th e r w ords, th e a d d itio n o p e ra tio n is memor yl ess .

A constant multiplier. T his o p e ra tio n is d e p ic te d by Fig. 2.14, an d sim ply re p re s e n ts ap p ly in g a scale fa c to r on th e in p u t x ( n) . N o te th a t th is o p e ra tio n is also m em oryless.
a ------------------------ v(n) = o i( n ) Figure 2.14 Graphical representation of a constant multiplier.

A signal multiplier. F ig u re 2.15 illu stra te s th e m u ltip lic a tio n o f tw o sig nal se q u en ces to fo rm a n o th e r (th e p ro d u c t) se q u e n c e , d e n o te d in th e figure as y (n ). A s in th e p re c e d in g tw o cases, w e can view th e m u ltip lic a tio n o p e ra tio n as m em o ry less.

---- -0 ^
x2( n )

A|(n)

v(n) = jT|(n)Ai(n)

Figure 2.1S Graphical representation of a signal multiplier.

A unit delay element. T h e u n it d elay is a sp e cial sy stem th a t sim ply d elay s th e signal passing th ro u g h it by one sam p le. F ig u re 2.16 illu stra te s such a system . If th e in p u t signal is x ( n) , th e o u tp u t is x( n 1). In fact, the sa m p le x{n 1) is sto re d in m em o ry at tim e n 1 a n d it is recalled fro m m e m o ry a t tim e n to form
v ( n ) = x( n - 1) T h u s th is basic b u ild in g b lock re q u ire s m em o ry . T h e use o f th e sym bol ; _1 to d e n o te th e u n it o f d eiay will b eco m e a p p a r e n t w h e n w e discuss th e z -tra n sfo rm in C h a p te r 3.
x(n) ------------------ _____
y (n ) = jr ( n 1)

Figure 2,16 Graphical representation of the unit delay element.

A unit advance element. In c o n tra st to th e u n it d e la y , a u nit ad v an ce m o v es th e in p u t x ( n ) a h e a d by o n e sa m p le in tim e to yield x ( n + 1). F ig u re 2.17 illu stra te s th is o p e ra tio n , w ith th e o p e r a to r ; b ein g used to d e n o te th e u n it advance.

Sec. 2.2

Discrete-Time Systems
x( n) y( n ) = x( n + I )

61

Figure 2.17 Graphical representation of the unit advance element.

W e o b se rv e th a t an y such ad v an ce is physically im possible in real tim e, since, in fact, it in v o lv es lo o k in g in to th e fu tu re o f th e signal. O n th e o th e r h an d , if we store th e signal in th e m em o ry o f th e c o m p u te r, w e can recall any sa m p le a t any tim e. In such a n o n re a l-tim e ap p lic a tio n , it is p o ssible to advance th e signal jr(?r) in tim e. Example 2.2.3 Using basic building blocks introduced above, sketch the block diagram representa tion of the discrete-time system described by the input-output relation.
v(n) = 3.v( - 1) + | x(n) + \x(n - 1)

where x(n) is the input and y(n) is the output of the system. Solution According to (2.2.5), the output v(n) is obtained by multiplying the input x(n) by 0.5, multiplying the previous input jr ( n - l) by 0.5. adding the two products, and then adding the previous output v(n 1) multiplied by j. Figure 2.18a illustrates this block diagram realization of the system. A simple rearrangem ent of (2.2.5). namely.
v ( ) =
5

.v(n -

)+

[jc(k) + x ( n -

)|

( 2 . 2 .6 )

leads to the block diagram realization shown in Fig. 2.18b. Note that if we treat "the system from the viewpoint of an input-output or an external description, we are not concerned about how the system is realized. On the other hand, if we adopt an
Black box

-i
x( n )

0.5

(a)
Black box

-i
x( n)

Figure 2.18 Block diagram realizations of the system y(n) = 0.25y(n 1) + 0.5 x{n) + 0.5j(n 1).

62

Discrete-Time Signals and Systems

Chap. 2

internal description of the system, we know exactly how the system building blocks are configured. In terms of such a realization, we can see that a system is relaxed at time n = n o if the outputs of all the delays existing in the system are zero at n = n{ ) (i.e., all memory is filled with zeros).

2.2.3 Classification of Discrete-Time Systems

In th e analysis as w ell as in th e design o f system s, it is d e s ira b le to classify the sy stem s acco rd in g to th e g e n e ra l p ro p e rtie s th a t th e y satisfy. In fact, th e m a th e m atical te c h n iq u e s th a t w e d e v e lo p in th is an d in s u b s e q u e n t c h a p te rs fo r analyzing an d d esig n in g d isc rete-tim e system s d e p e n d h eav ily on th e g e n e ra l ch aracteristics of th e system s th a t are b ein g c o n sid e re d . F o r th is re a so n it is n ecessa ry for us to d ev elo p a n u m b e r o f p ro p e rtie s o r c a te g o rie s th a t can be u se d to d escrib e th e g e n e ra l ch aracteristics o f system s. W e stress th e p o in t th a t fo r a sy stem to po ssess a given p ro p e rty , th e p ro p e rty m u st h o ld fo r ev ery p o ssible in p u t signal to th e system . If a p ro p e rty holds for so m e in p u t signals b u t n ot fo r o th e rs, th e system d o e s n ot p o ssess th a t p ro p erty . T h u s a c o u n te re x a m p le is sufficient to p ro v e th a t a system d o e s n ot possess a p ro p e rty . H o w ev er, to p ro v e th a t th e system has so m e p ro p e rty , we m ust prove th a t th is p ro p e rty h o ld s fo r every po ssib le in p u t signal.

Static versus dynamic systems. A d isc re te -tim e sy stem is called static o r m em o ry le ss if its o u tp u t at any in sta n t n d e p e n d s at m o st o n the in p u t sam ple at th e sam e tim e, b u t n o t o n p a st o r fu tu re sa m p le s o f th e in p u t. In any o th e r case, th e system is said to b e d y n a mi c o r to h av e m em o ry . If th e o u tp u t o f a system at tim e n is co m p letely d e te rm in e d by th e in p u t sa m p le s in th e in te rv a l fro m n - N to n ( N > 0), th e system is said to h av e m e m o r y o f d u ra tio n N . U N 0. th e sy stem is static. H 0 < N < oo, th e sy stem is said to have f ini te m e m o r y , w h ereas if N = oo, th e system is said to have infinite m e m o r y . T h e system s d escrib ed by th e follow ing in p u t- o u tp u t e q u a tio n s
y(n) = a x {n) y ( n ) = nx ( n ) + b x 3(n) (2.2.7) (2.2.8)

a re b o th static o r m em o ry less. N o te th a t th e re is n o n e e d to s to re any o f th e past in p u ts o r o u tp u ts in o rd e r to c o m p u te th e p re se n t o u tp u t. O n th e o th e r h an d , th e sy stem s d escrib ed by th e follow ing in p u t-o u tp u t re la tio n s y( n) = x ( n ) + 3 x ( n 1) y(n) = J ^ x ( n - k ) k=0
X

(2.2.9) (2.2.10)

y( n) = J 2 x ( n - k ) Jt=0

(2.2.11)

are dynamic systems or systems with memory. The systems described by (2.2.9)

Sec. 2.2

Discrete-Time Systems

63

an d (2.2.10) h av e fin ite m em o ry , w h ereas the sy stem d e sc rib e d by (2.2.11) has infinite m em o ry . W e o b se rv e th a t sta tic o r m em oryless system s are d e sc rib e d in g e n e ra l by in p u t-o u tp u t e q u a tio n s o f th e form y(n) = T [ x ( n ) , n] a n d th e y do n o t in clu d e d elay e le m e n ts (m em o ry ). (2.2.12)

Time-invariant versus time-variant systems. W e can su b d iv id e th e g en eral class o f sy stem s in to th e tw o b ro a d c a teg o ries, tim e -in v a ria n t system s and tim e -v a ria n t sy stem s. A system is called tim e -in v a ria n t if its in p u t- o u tp u t c h a ra c teristics d o n o t c h a n g e w ith tim e. T o e la b o ra te , su p p o se th a t w e h av e a system T in a re la x e d s ta te w hich, w h en ex cited by an in p u t signal x ( n) , p ro d u c e s an o u tp u t signal y(n). T h u s w e w rite
y( n) = T [ x { n) ) (2.2.13)

N o w su p p o se th a t th e sam e in p u t signal is d e la y e d by k u n its o f tim e to yield x (n - &), an d ag ain a p p lied to th e sam e system . If th e c h a ra c te ristic s of th e system d o n o t ch an g e w ith tim e, th e o u tp u t o f th e relax ed system will b e y( k). T h at is, th e o u tp u t will b e th e sa m e as th e resp o n se to x ( n) . ex cep t th a t it will be d elay ed by th e sam e k u n its in tim e th a t the in p u t w as d elay ed . T his lead s us to define a tim e -in v a ria n t o r sh ift-in v a ria n t system as follow s.

Definition.
o n ly if

A relax ed system T is time i nvariant o r shift i nvariant if and

x( n) im p lies th a t x {n k)

y(n)

y( n k)

(2.2.14)

fo r ev ery in p u t signal x (n ) a n d every tim e shift k.

T o d e te rm in e if an y given system is tim e in v a ria n t, w e n e e d to p e rfo rm the te st specified b y th e p re c e d in g definition. B asically, we ex cite th e system w ith an a rb itra ry in p u t se q u e n c e x ( n) , w hich p ro d u c e s an o u tp u t d e n o te d as y ( n) . N ext w e d elay th e in p u t se q u e n c e by sam e a m o u n t k an d re c o m p u te th e o u tp u t. In g e n eral, w e can w rite th e o u tp u t as y(, k) = T [ x ( n <:)] N o w if th is o u tp u t y{n, k) = y{n k), for all p o ssib le v alu es o f k, th e system is tim e in v a ria n t. O n th e o th e r h a n d , if th e o u tp u t y( n, k ) ^ y ( n k), ev en fo r o n e v alu e o f k, th e sy stem is tim e v arian t.

64 xin )

Discrete-Time Signals and Systems

Chap. 2

V I7 1) = X( 7 7I- V <7 1 - ]I

D ifferentiator"

- B

x(/t) Time" multiplier

v( n ) = xl - n )
Folder"

v(n J = .u n ) cos oiii

Figure 2.19 Examples of a lime-invariant (a) and some time-variant systems (h)-(d).

Example 2.2.4 Determ ine if the systems shown in Fig. 2.19 are time invariant or time variant. Solution (a) This system is described by the input^output equations y(7i) = T\ xin)} = x(n\ x(n - 1) (2.2.15)

Nov, if the input is delayed by k units in time and applied to the system, it is clear from the block diagram that the output will be
y i n . k) = x i n - k) x i n k 1)

(2.2.16)

On the other hand, from (2.2.14) we note that if we delay y (n) by k units in time, we obtain yin k) = x(n k) xin k 1) (2.2.17)

Since the right-hand sides of (2.2.16) and (2.2.17) are identical, it follows that v(n. k) = yin - k). Therefore, the system is time invariant.

Sec. 2.2

Discrete-Time Systems

65

(b) The input-output equation for this system is

y(n) = T[x(n)] = nx(n) The response of this system to x(n - Jt) is y(n, k) = nx(n - k) Now if we delay ;y(n) in (2.2.18) by k units in time, we obtain y(n - k) = (n k)x(n k)
(2 .2.20)
= nx(n k) - kx(n - k)

(2.2.18)

(2.2.19)

This system is time variant, since y(n, k) ^ y(n - k). (c) This system is described by the input-output relation >() = T[x(n)\ = x ( - n ) The response of this system to jr(n - k) is ;y(n, A ;) = T[x(n - *)] = x ( - n - k) (2.2.22) (2.2.21)

Now, if we delay the output ;y(n), as given by (2.2.21), by k units in time, the result will be y(n - k) = x ( - n + k) (2.2.23) Since y(n, k) ^ y{n - it), the system is time variant.

(d) The input-output equation for this system is

y(n) = j:(n) costDon The response of this system to x(n - k) is y(n, k) = x(n - k) cos o\)n (2.2.25) (2.2.24)

If the expression in (2.2.24) is delayed by k units and the result is compared to (2.2.25), it is evident that the system is time variant.

Linear versus nonlinear systems. The general class o f system s can also be subdivided into linear system s and nonlinear system s. A linear system is one that satisfies the su pe rp ositio n princ iple. Sim ply stated, the principle o f su p erp osi tion requires that the resp onse o f the system to a w eighted sum o f signals b e equal to the corresponding w eighted sum of the responses (outp u ts) of the system to each of the individual input signals. H en ce w e have the follow ing definition o f linearity.
Definition.
A r e la x e d T s y s te m is lin e a r if a n d only if T [ a i x i ( n ) + azx2{n)] = a\ T [ x \ ( n ) ] + a i T [ x 2{n)] (2.2.26)

for any arbitrary input seq u en ces x\ ( n) and x 2(n), and any arbitrary constants aj and 0 2 Figure 2.20 gives a pictorial illustration of the superposition principle.

66

Discrete-Time Signals and Systems

Chap. 2

if and only if v(n) = v'(n).

T h e su p e rp o sitio n p rin cip le e m b o d ie d in th e re la tio n (2.2.26) can be s e p a r a te d in to tw o p arts. F irst, su p p o se th a t a2 = 0. T h e n (2.2.26) re d u c e s to T{a\ X\ ( n) ] = a\ T [ x \ { n ) } = a\ vi(n) w h ere
vi (fl) = T [ x x(n)}

(2.2.27)

T h e re la tio n (2.2.27) d e m o n s tra te s the mul ti pli cat i ve o r scaling p r o p e r t y of a lin ear system . T h a t is, if th e re sp o n se o f th e system to th e in p u t x i ( n ) is vi(n ), the re sp o n se to a\X](n) is sim ply a i j i(n ). T h u s any scaling of th e in p u t resu lts in an id en tical scaling o f th e c o rre sp o n d in g o u tp u t. S eco n d , su p p o se th a t ai = a2 = 1 in (2.2.26). T h e n
T [ x \ ( n ) + x 2 (n)] = T [ x \ { n ) ] + T [ x \ ( n ) }

(2.2.28) = yi ( n) + yz(n) T his re la tio n d e m o n s tra te s th e additivity pr ope r t y o f a lin e a r sy stem . T h e ad ditivity an d m u ltip lic ativ e p ro p e rtie s c o n stitu te th e su p e rp o s itio n p rin c ip le as it ap p lies to lin ear system s. T h e lin earity co n d itio n em b o d ie d in (2.2.26) can b e e x te n d e d a rb itra rily to any w eig h ted lin e a r c o m b in a tio n o f signals by in d u ctio n . In g e n e ra l, we h av e M- 1 x( n) = ^ 2 GkXk(n)
k=l

M- 1 y ( n ) = ^ akyk (n)

(2.2.29)
i= l

w h ere ^ ( n ) = T [ x k(n)} k = 1, 2, , . . , M 1 (2.2.30)

Sec. 2.2

Discrete-Time Systems

67

W e o b se rv e fro m (2.2.27) th a t if a i 0, th e n y (n ) = 0. In o th e r w ords, a r e lax ed , lin e a r sy stem w ith z e ro in p u t p ro d u c e s a z e ro o u tp u t. If a system p ro d u c e s a n o n z e ro o u tp u t w ith a zero in p u t, th e system m ay be e ith e r n o n re la x e d o r n o n lin ear. If a re la x e d sy stem d o e s n o t satisfy th e su p e rp o s itio n p rin cip le as given by th e d efin itio n ab o v e, it is called nonlinear. Exam ple 2.2^ D eterm ine if the systems described by the following input-output equations are linear or nonlinear. (a) y(n) = ttx(n) (b) y(n) = *(n2) (e) y(n) = ex[n] (c) v(n) = v2(n)

<d) y( n ) = Ax{n) + B Solution

(a) For two input sequences jti(n) and

the corresponding outputs are (2.2.31)

Vi(n) = n.ifi(n) y2{n) = nx2(n) A iinear combination of the two input sequences results in the output Vj() = T[a\Xy (n) + oijMh)] = = ainxi (n) + a2nx2(n) () + (/i)]

(2.2.32) On the other hand, a linear combination of the two outputs in (2.2.31) results in the output a\ V ] (n) + a2y 2(n) = ainx\(n) + a2n,x2(n) (2.2.33)

Since the right-hand sides of (2.2.32) and (2.2.33) are identical, the system is iinear. (b) As in part (a), we find the response of the system to two separate input signals *i(n) and x 2(n). The result is v,(n) = X\(n2) (2.2.341 y2(rt) = X2(n2) The output of the system to a linear combination of Xi(n) and *;(?) is y3(n) = T\a\X\ () + a 2x 2(n)] = a hx,(n2) + a2x2(n2) Finally, a linear combination of the two outputs in (2.2.36) yields O ] \>i(n) + c 2V2(n) = a i JC i (n2) + < i2X2(n2) (2.2.36) (2.2.35)

By comparing (2.2.35) with (2.2.36). we conclude that the system is linear. (c) The output of the system is the square of the input. (Electronic devices that have such an input-output characteristic and are called square-law devices.) From our previous discussion it is clear that such a system is memoryless. We now illustrate that this system is nonlinear.

68

Discrete-Time Signals and Systems

Chap. 2

The responses of the system to two separate input signals are v,() = .vf(n) ' y2(n) = x;(n) (2.2.37)

The response of the system to a linear combination of these two input signals is >'3<n ) = T[a 1*1 (n) + a2x2(n)} = [fli*i(n) + a2x2(n)]2
= cfA 'fln) -I- 2a-la 2x i ( n ) x 2(n) + a 2x 2 { n )

(2.2.38)

On the other hand, if the system is linear, it would produce a linear combination of the two outputs in (2.2.37). namely, ai_Vi(n) + fl2.V 2(n) = (rt) + a2x2(n) (2.2.39)

Since the actual output of the system, as given by (2.2.38). is not equal to (2.2.39), the system is nonlinear.

(d) Assuming that the system is excited by x\(n) and x2in) separately, we obtain
the corresponding outputs
V](n) = AX](rt) + B

(2.2.40) y2(n) = A x 2(n) + B A linear combination of X\(n) and x 2{n) produces the output
V i( ) =
T[u^X ]

(/i) + a 2x 2(n>]

= A[a,x,(/i) + a 2x 2(n)} + B
= A a \ X \ ( n ) -I- a2A x 2(n) + B

(2.2.41)

On the other hand, if the system were linear, its output to the linear com bina tion of Ji(n) and x 2 (n) would be a linear combination of vj i n) and y2(n). that is.
ai yi (n ) + a 2y 2(n) = a ] A x ] { n ) + a \ B a 2A x 2{ n ) + a 2 B

(2.2.42)

Clearly. (2.2.41) and (2.2.42) are different and hence the system fails to satisfy the linearity test. The reason that this system fails to satisfy the linearity test is not that the system is nonlinear (in fact, the system is described by a linear equation) but the presence of the constant B. Consequently, the output depends on both the input excitation and on the param eter B ^ 0. Hence, for B ^ 0. the system is not relaxed. If we set B = 0, the system is now relaxed and the linearity test is satisfied. (e) Note that the system described by the input-output equation y(n) = e,1',) (2.2.43) is relaxed. If x(n) = 0, we find that y(n) = 1. This is an indication that the system is nonlinear. This, in fact, is the conclusion reached when the linearity test, is applied.

Causal versus noncausal systems.

d isc re te -tim e system s.

W e b eg in w ith th e d efin itio n o f causal

Sec. 2.2

Discrete-Time Systems

69

D e fin itio n , a system is said to b e causal if th e o u tp u t o f th e system at any tim e n [i.e., v(n)] d e p e n d s oniy on p re s e n t an d p ast in p u ts [i.e., x { n ), x(tt - 1), x(rt 2 ) , . . . ] , b u t d o e s n o t d e p e n d on fu tu re in p u ts [i.e., x( n + 1), x ( n + 2 ) , . . . ] . In m a th e m a tic a l te rm s, th e o u tp u t o f a cau sal sy stem satisfies an e q u a tio n o f th e form v(n) = F[x{n), x ( n - 1), x ( n - 2 ) , . . . ] w h ere /'[] is so m e a rb itra ry function. If a system d o e s n o t satisfy this d efin itio n , it is called noncausal . S uch a sy stem has an o u tp u t th a t d e p e n d s n o t oniy on p re s e n t a n d p ast in p u ts b u t also o n fu tu re in p u ts. It is a p p a re n t th a t in real-tim e signal p ro cessin g ap p licatio n s w e c a n n o t o b serv e fu tu re v alu es o f th e signal, and h e n c e a n o n cau sal system is physically u n re a l izab le (i.e., it c a n n o t b e im p le m e n te d ). O n th e o th e r h an d , if th e signal is re c o rd e d so th a t th e p ro cessin g is d o n e off-line (n o n re a l tim e ), it is p o ssible to im p lem en t a n o n cau sal sy stem , since all v alu es o f th e signal are av ailab le a t th e tim e o f p r o cessing. T h is is o fte n th e case in th e p ro cessin g o f g eophysical signals an d im ages. Example 2.2.6 D eterm ine if the systems described by the following input-output equations are causal or noncausal. (a) y(n) = x(n) - x(n - 1) (d) y(n') = x(n) + 3jr(n + 4) (g) }'(n) = x ( -n ) Solution The systems described in parts (a), (b), and (c) are clearly causal, since the output depends only on the present and past inputs. On the other hand, the systems in parts (d). (e), and (f) are clearly noncausal, since the output depends on future values of the input. The system in (g) is also noncausal, as we note by selecting, for example, n = - 1 , which yields v(1) = * 0 ) Thus the output at n = - 1 depends on the input at n = 1, which is two units of time into the future. (b) y(n) = (e) y(n) = x( n2) x(k) (c) y(n) = ax(n) (2.2.44)

(t) y(n) = x(2n)

Stable versus unstable systems. S tab ility is an im p o rta n t p ro p e rty th a t m u st b e c o n s id e re d in an y p ractical ap p lic a tio n o f a system . U n s ta b le system s u su ally ex h ib it e rra tic an d ex tre m e b e h a v io r an d cause overflow in an y p ractical im p le m e n ta tio n . H e re , w e define m a th e m a tic a lly w h at w e m e a n by a sta b le system , a n d la te r, in S ectio n 2.3.6, w e ex p lo re th e im plicatio n s o f this definition fo r lin ear, tim e -in v a ria n t system s.

Definition. A n a rb itra ry re la x e d system is said to be b o u n d e d in p u t-b o u n d e d o u tp u t (B IB O ) sta b le if a n d only if ev ery b o u n d e d in p u t p ro d u ces a b o u n d e d o u tp u t.

T h e c o n d itio n s th a t th e in p u t se q u e n c e x{n) a n d th e o u tp u t se q u e n c e y ( n) are b o u n d e d is tra n s la te d m a th e m a tic a lly to m e a n th a t th e re exist som e finite n u m b ers,

70

Discrete-Time Signals and Systems

Chap. 2

say M x an d M v. such th at j.v(ri)! < M K < oc < M x < dc (2.2.45)

fo r all n. If. fo r so m e b o u n d e d in p u t se q u e n c e ,v(), the o u tp u t is u n b o u n d e d (in fin ite), th e sy stem is classified as u n sta b le . Exam ple 2.2.7 Consider the nonlinear system described by the input-output equation
V(/I ) = 1) ,V(/i )

As an input sequence we select the hounded signal xin ) = C&(n ) where C is a constant. We also assume that y(1) = 0. Then the output sequence is y(0) = C, y (l) = C \ y(2) = Cd............ y(n) = C : "

Clearly, the output is unbounded when ] < ICl < oc. Therefore, the system is BIBO unstable, since a bounded input sequence has resulted in an unbounded output.

2.2.4 Interconnection ot Discrete-Time Systems

D isc rete-tim e sy stem s can be in te rc o n n e c te d to form larg er sy stem s. T h e re are tw o b asic ways in w hich system s can be in te rc o n n e c te d : in c a scad e (series) o r in p arallel. T h ese in te rc o n n e c tio n s are illu strated in Fig. 2.21. N o te th at th e tw o in te rc o n n e c te d sy stem s are d ifferen t. In th e cascad e in te rc o n n e c tio n the o u tp u t of th e first system is yiO?) = 7j[;r(/j)] (2.2.46)

xin)

: T\
r

y(n) T,

7,
(a)

v | (n )

(b)

Figure 2.21 Cascade (a) and parallel (b) interconnections of systems.

Sec. 2.2

Discrete-Time Systems

71

an d th e o u tp u t o f th e second system is v(n) = T2[\\(n)] (2.2.47)

= r 2{7 i[* (n )]}

W e o b se rv e th a t sy stem s 7"j a n d T2 can be co m b in ed o r c o n s o lid a te d in to a single o v e ra ll sy stem % = T271 C o n s e q u e n tly , w e can ex p ress th e o u tp u t o f th e co m b in ed sy stem as y( n) = Tc[x(n)] In g e n e ra l, th e o rd e r in w hich th e o p e ra tio n s T\ a n d T2 a re p e rfo rm e d is im p o rta n t. T h a t is, T27I # T,T2 fo r a rb itra ry system s. H o w ev er, if th e system s 7j a n d T2 a re iin e a r a n d tim e in v a ria n t, th e n (a) % is tim e in v arian t an d (b ) T2T\ = T \ T2, th a t is, th e o r d e r in w hich th e sy stem s p ro cess th e signal is n o t im p o rta n t. 7^71 a n d T \ T 2 yield id en tical o u tp u t se q u e n c e s. T h e p ro o f o f (a) follow s. T h e p ro o f o f (b ) is given in S e c tio n 2.3.4. T o p ro v e tim e in v a ria n c e , su p p o se th a t Tj and T2 a re tim e in v arian t; th e n x( n k ) an d Vi(n - k) ' + y( n - k) Thus x{n k) Tf y( n k ) vi(/i - k) (2.2.48)

an d th e re fo re , Tc is tim e in v arian t. In th e p a ra lle l in te rc o n n e c tio n , th e o u tp u t of th e sy stem T\ is ^ ( n ) an d the o u tp u t o f th e sy stem T2 is y2(n). H e n c e th e o u tp u t o f th e p a ra lle l in te rc o n n e c tio n is v3(n) = .V ] (n) + > > 2(n) = Ti[x{n)\ + T2[x(n)\ = (T\ + T2)[x(n)} = Tp[x(n)\ w h e re Tp = T\ + T2. In g e n e ra l, w e can u se p a rallel an d cascade in te rc o n n e c tio n o f sy stem s to c o n s tru c t la rg e r, m o re com plex system s. C o n v e rsely , w e can ta k e a la rg e r system a n d b re a k it d o w n in to sm a ller su b sy stem s fo r p u rp o se s o f an aly sis a n d im p le m e n ta tio n . W e sh all u se th e s e n o tio n s la te r, in th e design a n d im p le m e n ta tio n of d ig ital filters.

72

Discrete-Time Signals and Systems

Chap. 2

2.3 ANALYSIS OF DISCRETE-TIME LINEAR TIME-INVARIANT SYSTEMS

In S ectio n 2.2 we classified system s in a c c o rd a n c e w ith a n u m b e r of c h aracteristic p ro p e rtie s o r categ o ries, nam ely: lin e a rity , cau sality , stab ility , a n d tim e in variance. H av in g d o n e so. we now tu rn o u r a tte n tio n to th e analysis o f th e im p o rta n t class o f lin ear, tim e -in v a ria n t (L T I) system s. In p a rtic u la r, we shall d e m o n s tra te th a t such system s are c h a ra c te riz e d in th e tim e d o m ain sim ply bv th e ir resp o n se to a u n it sam p le se q u en ce. W e shall also d e m o n s tra te th a t any a rb itra ry in p u t signal can b e d eco m p o sed an d r e p re s e n te d as a w eig h ted sum of u n it sa m p le seq u en ces. A s a c o n s eq u en ce o f th e lin e a rity a n d tim e-in v arian ce p ro p e rtie s of the system , th e resp o n se o f th e system to any a rb itra ry in p u t signal can be e x p ressed in term s of th e u n it sam p le re sp o n se of th e system . T h e g en eral form o f the ex p ressio n th a t re la te s th e u n it sam ple re sp o n se o f the system an d th e a rb itra ry in p u t signal to th e o u tp u t signal, called th e c o n v o lu tio n sum o r th e c o n v o lu tio n fo rm u la, is also d eriv ed . T h u s we are ab le to d e te rm in e the o u tp u t o f any lin e a r, tim e-in v arian t system to any a rb itra ry in p u t signal.

2.3.1 Techniques for the Analysis of Linear Systems

T h e re are tw o basic m e th o d s for an aly zin g th e b e h a v io r o r re sp o n s e of a lin ear system to a given in p u t signal. O n e m e th o d is b a se d on th e d ire c t so lu tio n o f the in p u t-o u tp u t e q u a tio n for th e system , w hich, in g e n e ra l, has th e form v(ji) - F [v ( n - 1), v (?7 2 ) ........ y(n - N) , x ( n ) . x ( n 1).......... x ( n - M)] w h ere F[-] d e n o te s so m e fu n ctio n o f th e q u a n titie s in b ra c k e ts. Specifically, fo r an L T I system , w e shall see la te r th a t th e g e n e ra l form o f th e in p u t-o u tp u t re la tio n sh ip is N M (2.3.1)

w h ere an d {b * ,.} are c o n sta n t p a ra m e te rs th a t specify th e sy stem an d a re in d e p e n d e n t o f x( n) a n d y( n) . T h e in p u t-o u tp u t re la tio n sh ip in (2.3.1) is called a d ifferen ce e q u a tio n a n d re p re se n ts o n e w ay to c h a ra c te riz e th e b e h a v io r of a d isc re te -tim e L T I system . T h e so lu tio n o f (2.3.1) is th e su b je ct o f S ection 2.4. T h e seco n d m e th o d fo r analyzing th e b e h a v io r o f a lin e a r sy stem to a given in p u t signal is first to d e c o m p o se o r reso lv e th e in p u t signal in to a sum o f e le m e n ta ry signals. T h e e le m e n ta ry signals a re se le c te d so th a t th e re sp o n se o f the system to each signal c o m p o n e n t is easily d e te rm in e d . T h e n , u sin g th e lin earity p ro p e rty o f th e sy stem , th e re sp o n se s o f th e sy stem to th e e le m e n ta ry signals are ad d e d to o b ta in th e to ta l re s p o n s e of th e sy stem to th e given in p u t signal. T h is seco n d m e th o d is th e o n e d e sc rib e d in th is se ctio n .

Sec. 2.3

Analysis o f Discrete-Time Linear Tim e-Invariant Systems

73

T o e la b o ra te , su p p o se th a t th e in p u t signal x (n ) is reso lv ed into a w eig h ted sum o f elem en tary ' signal c o m p o n e n ts {*/. ()} so th a t

w h ere th e {ct} a re th e set of a m p litu d e s (w eighting coefficients) in the d e c o m p o sitio n o f th e signal x( n) . N ow su p p o se th a t th e re sp o n se o f th e system to the e le m e n ta ry signal c o m p o n e n t x k{n) is y k(n). T hus. y*(n) = T [j.t (n)] (2.3.3)

assu m in g th a t th e sy stem is re la x e d an d th a t th e re sp o n se to ckx k(n) is cky k(n). as a c o n s e q u e n c e o f th e scaling p ro p e rty o f th e lin e a r system . F in ally , th e to ta l re sp o n se to th e in p u t x( n ) is

(2.3.4)

In (2.3.4) we u se d th e ad d itiv ity p ro p e rty o f th e lin e a r system . A lth o u g h to a larg e e x te n t, th e ch o ice o f th e e le m e n ta ry signals a p p e a rs to b e a rb itra ry , o u r se lectio n is heavily d e p e n d e n t on th e class of input signals that we w ish to co n sid er. If w e place n o re stric tio n on th e c h aracteristics of th e input signals, its re so lu tio n in to a w eig h ted sum of u n it sam ple (im p u lse) se q u en ces p ro v e s to b e m a th e m a tic a lly c o n v e n ie n t an d c o m p letely g en eral. O n th e o th e r h a n d , if w e re stric t o u r a tte n tio n to a subclass o f in p u t signals, th e re m ay be a n o th e r set o f e le m e n ta ry signals th a t is m o re c o n v e n ie n t m a th e m a tic a lly in the d e te rm in a tio n o f th e o u tp u t. F o r ex am p le, if th e in p u t signal x( n) is p erio d ic w ith p e rio d N , w e have a lre a d y o b se rv e d in S ectio n 1.3.5 th a t a m a th em atically c o n v e n ie n t set o f elem en tary 7 signals is th e set o f e x p o n en tials x k (n) = eJluin k = 0. l , . . . , i V - l (2.3.5)

w h e re th e fre q u e n c ie s {cok } a re h arm o n ic a lly re la te d , th a t is. k = 0. 1.........N - 1 (2.3.6)

T h e fre q u e n c y 2 n / N is called th e fu n d a m e n ta l fre q u e n c y , an d all h ig h er-freq u en cy c o m p o n e n ts a re m u ltip le s o f th e fu n d a m e n ta l fre q u e n c y c o m p o n e n t. T h is subclass o f in p u t sig n als is c o n s id e re d in m o re d e ta il later. F o r th e re so lu tio n o f th e in p u t signal in to a w eig h ted su m o f u n it sam ple se q u en ces, w e m u st first d e te rm in e th e re sp o n se o f th e system to a u n it sa m p le se q u e n c e a n d th e n use th e scaling a n d m u ltip lic ativ e p ro p e rtie s of th e lin ear

74

Discrete-Time Signals and Systems

Chap. 2

sy stem to d e te rm in e th e fo rm u la fo r th e o u tp u t given any a rb itra ry input, T his d e v e lo p m e n t is d escrib ed in d etail as follow s.

2.3.2 Resolution of a Discrete-Time Signal into Impulses

S u p p o se w e h av e an a rb itra ry signal x ( n ) th a t we w ish to reso lv e in to a sum of unit sa m p le seq u en ces. T o utilize th e n o ta tio n e sta b lish e d in th e p re c e d in g se ctio n , we select th e e le m e n ta ry signals x k (n) to be x k(n) = 8{n - k) (2.3.7)

w h ere k re p re se n ts th e d elay o f th e u n it sam p le se q u e n c e . T o h a n d le an a rb itra ry signal x ( n ) th a t m ay h ave n o n z e ro v alu es o v er an infinite d u ra tio n , th e set of unit im p u lses m u st also b e infinite, to en co m p ass th e infinite n u m b e r of delays. N o w su p p o se th a t we m u ltip ly th e tw o se q u e n c e s x( n) a n d <5(n - k). Since 8{n k) is z e ro e v ery w h ere ex cep t a t n = k , w h e re its v alu e is u nity, the result o f th is m u ltip lic atio n is a n o th e r se q u e n c e th a t is z e ro e v ery w h ere e x c e p t at n k , w h ere its v alu e is x ( k) , as illu stra te d in Fig. 2,22. T h u s x( n) 8( n k) = x ( k) 8( n k) (2.3.8)

J t(n ) T T T 111 i l l i [ -2-10113 (a ) 6 (/i-J t) i , 1 I , Tt I i i 1 J C ( J c )

(b)
*(*) 6(n k )

k 0

Figure 2.22

M ultiplication of a signal x i n ) with a shifted unit sam ple sequence.

Sec. 2.3

Analysis of Discrete-Time Linear Time-Invariant Systems

75

is a se q u e n c e th a t is z e ro e v e ry w h e re ex cep t at n = k , w h e re its v a lu e is x( k ) . If we w ere to re p e a t th e m u ltip lic a tio n of x ( n ) w ith < 5(a? m ), w h ere m is a n o th e r d elay (im =6 k), th e re su lt will b e a se q u en ce th a t is z e ro e v e ry w h e re e x cep t at n = m, w h ere its v alu e is x ( m ) . H e n c e x( n ) 5 ( n m) = x ( m) 8( n m) (2.3.9)

In o th e r w o rd s, each m u ltip lic a tio n o f th e signal x( n ) by a u n it im p u lse at som e d elay k, [i.e., <5(n it)], in essen ce picks o u t th e single v alu e x ( k ) o f th e signal jc(n) at th e d e la y w h e re th e u n it im pulse is n o n z e ro . C o n s e q u e n tly , if w e re p e a t this m u ltip lic a tio n o v e r all p o ssib le delays, - o o < k < oo, a n d su m all th e p ro d u c t se q u e n c e s, th e re su lt will be a se q u e n c e e q u a l to th e se q u e n c e x ( n ) , th a t is, P C x( n) = ^ x ( k) 8( n k) k= oc W e e m p h a s iz e th a t th e rig h t-h an d side of (2.3.10) is th e infinite n u m b e r o f u n it sa m p le se q u en ces w h ere th e u n it sa m p le h as an a m p litu d e value o f x( k ) . T h u s th e rig h t-h a n d sid e o f re so lu tio n o f o r d e c o m p o s itio n o f any a rb itra ry signal jc(n) in to a sum o f sh ifted u n it sam p le seq u en ces. Exam ple 2.3.1 Consider the special case of a finite-duration sequence given as
jc()

(2.3.10) su m m a tio n of an se q u e n c e 6(n - k) (2.3.10) gives th e w e ig h te d (scaled)

= (2, 4, 0,3)
T

Resolve the sequence x(n) into a sum of weighted impulse sequences. Solution Since the sequence x(n) is nonzero for the time instants n = 1, 0. 2, we need three impulses at delays k = 1. 0, 2, Following (2.3.10) we find that x<n) = 2(5(n + 1) + 4<5(n) + 3<5(n 2)

2.3.3 Response of LTI Systems to Arbitrary Inputs: The Convolution Sum

H av in g re so lv e d an a rb itra ry in p u t signal x ( n ) in to a w eig h te d su m o f im pulses, w e a re no w re a d y to d e te rm in e th e re sp o n se of an y re la x e d lin e a r sy stem to any in p u t signal. F irs t, w e d e n o te th e re sp o n se v(n, k) o f th e sy stem to th e in p u t unit sa m p le se q u e n c e a t n = it by th e special sym bol h(n, k), oo < k < oo. T h a t is, y( n, k) = h(n. k ) = T[ S( n )] (2.3.11)

In (2.3.11) w e n o te th a t n is th e tim e index a n d k is a p a r a m e te r sh ow ing th e lo c a tio n o f th e in p u t im p u lse. If th e im pulse at th e in p u t is sc aled by an a m o u n t ct = jc(it), th e re sp o n s e of th e system is th e c o rre sp o n d in g ly sc aled o u tp u t, th a t is,

Ckh(n, k) = x(k)h(n, k)

(2.3.12)

76

Discrete-Time Signals and Systems

Chap. 2

F in ally , if th e in p u t is th e a rb itra ry signal x(/t) th a t is e x p re ss e d as a sum of w eig h ted im p u lses, th a t is. (2.3.13) th e n th e resp o n se o f the system to x(/i) is the c o rre sp o n d in g sum o f w eig h ted o u tp u ts, th a t is, y(rt) = 7"[.r(/;)] = T ^ x( k ) S( n k)

x ( k ) T [ 5 (m - k)]

(2.3.14)

ii= oc C learly , (2.3.14) follow s from th e su p e rp o sitio n p ro p e rty of lin e a r system s, and is k n o w n as th e superposit i on s u mma t i o n . W e n o te th a t (2.3.14) is an e x p ressio n for th e resp o n se o f a lin e a r system to any a rb itra ry in p u t se q u e n c e x( n) . T his ex p ressio n is a fu n ctio n of b o th .v() and th e resp o n ses h(n. k) of the system to th e unit im pulses Sin k) fo r oc < k < oc. In d eriv in g (2.3.14) w e used th e lin earity p ro p e rty o f th e system but not its tim ein v arian ce p ro p e rty . T h u s th e ex p ressio n in (2.3.14) ap p lies to any relax ed lin ear (tim e -v a ria n t) system . If. in a d d itio n , th e system is tim e in v a ria n t, th e fo rm u la in (2.3.14) sim plifies c o n sid erab ly . In fact, if the resp o n se o f th e L T I system to th e u n it sa m p le seq u en ce <5(rc) is d e n o te d as h(n). th a t is. h(n) = T [ b( n ) \ (2.3.15)

th en by th e tim e-in v arian ce p ro p e rty , th e resp o n se o f the system to the delay ed u n it sa m p le se q u e n c e <5(n - k) is h(n k) = T [ S( n A')] C o n seq u en tly , th e fo rm u la in (2.3.14) re d u c e s to (2.3.17) k=-oc N ow we o b serv e th a t th e relax ed L T I system is co m p letely c h a ra c te riz e d by a single fu n ctio n h(n), n am ely , its resp o n se to th e u n it sam p le se q u e n c e In c o n tra st, th e g en eral c h a ra c te riz a tio n of th e o u tp u t o f a tim e -v a ria n t, lin e a r sys tem re q u ire s an in fin ite n u m b e r o f u n it sa m p le re sp o n s e fu n ctio n s, h{n, k), o n e for ea c h p o ssib le d elay . T h e fo rm u la in (2.3.17) th a t gives th e re sp o n se y( n) of th e L T I system as a fu n c tio n o f th e in p u t signal x ( n ) a n d th e u n it sa m p le (im p u lse) re sp o n se h(n) is called a convol ut i on s um. W e say th a t th e in p u t jt(n ) is c o n v o lv ed w ith th e im p u lse (2,3.16)

Sec. 2.3

Analysis of Discrete-Time Linear Tim e-Invariant Systems

77

response h(n) to yield the output y in ). W e shall now explain the procedure for com puting the resp onse y (n ). both m athem atically and graphically, given the input x ( n ) and the im pulse response h(n) o f the system . Suppose that we wish to com pute the output of the system at som e time instant, say n = n 0. A ccordin g to (2.3.17), the resp onse at n = no is given as
OC

y (n 0) =

^ 2 x ( k ) h ( n 0 - k)
Jc=-oc

(2.3.18)

O ur first observation is that the index in the sum m ation is k , and h en ce both the input signal x ( k ) and the im pulse resp onse h(no - k) are fun ction s o f k. Second, w e ob serve that the sequ en ces x ( k ) and h(nQ k) are m ultiplied togeth er to form a product seq u en ce. T h e output >(o) is sim ply the sum over all valu es o f the product sequ en ce. T he seq u en ce h ( n 0 k) is ob tain ed from h ( k ) by, first, folding h (k) about k = 0 (the tim e origin), w hich results in the seq u en ce h ( k). The folded seq u en ce is then shifted by no to yield h(no k). T o sum m arize, the process o f com p utin g the con volu tion b etw een x ( k ) and h (k) in volves the follow in g four steps. 1. Folding. F old h(k) about k = 0 to obtain h ( ~ k ) . 2. Shifting, Shift h ( k) by n 0 to the right (left) if n o is p ositive (n egative), to obtain h(no ). 3. Multiplication. M ultiply v Jk) = x ( k ) h ( n 0 - k). by h (no k) to obtain the product sequ en ce

4. S u m m a t i o n . Sum all the values o f the product seq u en ce vnt)( k ) to obtain the value o f the output at tim e n = n 0. W e note that this p rocedure results in the resp onse o f the system at a sin gle tim e instant, say n = n 0. In gen eral, we are interested in evaluating the response o f the system over all tim e instants - o o < n < oo. C onsequently, steps 2 through 4 in the sum m ary m ust be rep eated , for all p ossible tim e shifts oo < n < oo. In order to gain a b etter understanding o f the procedure for evaluating the convolution sum , w e shall dem onstrate the p rocess graphically. T he graphs will aid us in explaining the four steps in volved in the com p utation o f the convolution sum.

Example 2.3.2
The impulse response of a linear time-invariant system is /!() = [1 .2 ,1 ,-1 }
T

(2.3.19)

Determine the response of the system to the input signal x(n) = {1,2.3,1} t
(2.3.20)

Discrete-Time Signais and Systems

Chap. 2

Solution We shall com pute the convolution according to the formula (2.3.17). but we shall use graphs of the sequences to aid us in the com putation. In Fig. 2.23a we illustrate the input signal sequence x(k) and the impulse response h{k) of the system, using k as the time index in order to be consistent with (2.3.17), The first step in the com putation of the convolution sum is to fold h(k). The folded sequence h(~k) is illustrated in Fig. 2.23b. Now we can compute the output at n = 0. according to (2.3.17), which is

v(0) = *=-cx

(2.3.21)

Since the shift n = 0, we use h( k) directly without shifting it. The product sequence = x(k)h(-k) (2.3.22)

h(k) 3 * -i

x(k 4i

T 1
10

-1 0 ! j

1 2 3

h(-k)
2 .
. -2

L 'n(k 1

T t . . . -1 0 1 2

.
(b)

Shift h(\-k)

vAk)

Product

To

111
(c)

,,. L |{t) 1 k (d) 0 12

Product sequence

T .

2 *

~3 T ! j-2 -1 0 1

T 7

Br

Figure 2.23

G raphical com putation of convolution.

Sec. 2.3

Analysis of Discrete-Time Linear Time-Invariant Systems

79

is also shown in Fig. 2.23b. Finally, the sum of all the terms in the product sequence yields () = vott) = 4

We continue the com putation by evaluating the response of the system at n = 1. According to (2.3.17), (2.3.23) The sequence h(\ k) is simply the folded sequence h( k) shifted to the right by one unit in time. This sequence is illustrated in Fig. 2.23c. The product sequence V ] (k} = x(k)h{l k) (2.3.24)

is also illustrated in Fig. 2.23c. Finally, the sum of all the values in the product sequence yields y(l) = ui(*) = 8

In a similar m anner, we obtain y(2) by shifting h ( - k ) two units to the right, forming the product sequence ih(A) = x(k)h(2 k) and then summing all the terms in the product sequence obtaining y(2) = 8. By shifting h(k) farther to the right, multiplying the corresponding sequence, and summing over all the values of the re sulting product sequences, we obtain v(3) = 3. v(4) = - 2 , y(5) = - 1 . For u > 5, we find that v(n) = 0 because the product sequences contain all zeros. Thus we have obtained the response y(n) for n > 0. Next we wish to evaluate v(n) for n < 0. We begin with n = Then (2.3.25) Now the sequence h ( 1 k) is simply the folded sequence h ( k ) shifted one time unit to the left. The resulting sequence is illustrated in Fig. 2.23d. The corresponding product sequence is also shown in Fig. 2.23d. Finally, summing over the values of the product sequence, we obtain V (-1) = 1 From observation of the graphs of Fig. 2.23, it is clear that any further shifts of h (1 - k) to the left always results in an all-zero product sequence, and hence y(n) = 0 for n 5 2

Now we have the entire response of the system for oc < n < oc. which we summarize below as y(n) = 0 ,0,1, 4. 8, 8. 3, - 2 , - 1 , 0 . 0 . . . .) t (2.3.26)

80

Discrete-Time Signals and Systems

Chap. 2

In E x am p le 2.3.2 w e illu stra te d th e c o m p u ta tio n o f th e co n v o lu tio n sum . using g rap h s o f th e se q u en ces to aid us in visualizing th e ste p s in volved in th e c o m p u ta tio n p ro c e d u re . B e fo re w o rk in g o u t a n o th e r ex a m p le , w e w ish to show th a t th e co n v o lu tio n o p e ra tio n is c o m m u ta tiv e in th e se n se th a t it is irre le v a n t w hich of th e tw o se q u e n c e s is fo ld ed a n d shifted. In d e e d , if w e b eg in w ith (2.3.17) a n d m ak e a ch an g e in th e v ariab le o f th e su m m a tio n , fro m k to m , by defin in g a new index m n k, th e n k = n m an d (2.3.17) b e co m es
CC

y(n) =

^2

x( n m ) h { m )

(2.3.27)

Since m is a d u m m y in dex, w e m ay sim ply re p la c e m by k so th a t y{n) = x(n-k)h(k) (2.3.28)

T h e ex p ressio n in (2.3.28) involves leav in g th e im p u lse re sp o n s e h( k) u n a lte re d , w hile th e in p u t se q u en ce is fo ld ed a n d sh ifted . A lth o u g h th e o u tp u t v(n) in (2.3.28) is id en tical to (2.3.17), th e p ro d u c t se q u e n c e s in th e tw o fo rm s o f th e co n v o lu tio n fo rm u la are not id en tical. In fact, if w e define th e tw o p ro d u c t se q u e n c e s as v(k) = x ( k ) h ( n k) w n(k) = x( n k) h(k) it can b e easily show n th at un() = w n (n k) an d th e re fo re ,
CC CC

v(n) =

^
k oc

^
oc

~ k)

since b o th se q u en ces co n tain th e sam e sa m p le v alu es in a d iffe re n t a rra n g e m e n t. Example 2.3.3 Determ ine the output y(n) of a relaxed linear tim e-invariant system with impulse response h(ri) = a"u(n), \a\ < 1 when the input is a unit step sequence, that is, x (n) = u(n ) Solution In this case both /j(n) and jc(n) are infinite-duration sequences. We use the form of the convolution formula given by (2.3.28) in which x (k) is folded. The

Sec. 2.3

Analysts of Discrete-Time Linear Tim e-Invariant Systems

81

h{k\

x(k)

TT i f
1 1 2 (b) 3 4 *

l * 1

x(- k)

I'oC A ')

-3

-2 -1

0
A(1

-1 -k) (c) >

i'i(i-) a

1
- 1 0 1 1':(!>

v( 2 k)

II

- 2 - 1 0 1 2 3 4 5
Figure 2.24

Graphical computation of convolution in Example 2.3.3.

sequences h(k), x(k). and x{k) are shown in Fig. 2.24. The product sequences vo(k). v\(k), and v2(k) corresponding to x ( k)h(k), x (l k)h(k), and x(2 - k)h(k) are illus trated in Fig. 2.24c, d. and e. respectively. Thus we obtain the outputs v(0) = 1 y(l) = 1 + a
y( 2) = 1 + a + a 1

82

Discrete-Time Signals and Systems

Chap. 2

Clearly, for n > 0, the output is y(n) = 1 + a 4- a2 + + a 1 _ an+1 (2.3,29) = 1-a On the other hand, for n < 0, the product sequences consist of all zeros. Hence v(n) = 0 n < 0 A graph of the output y(n) is illustrated in Fig. 2.24f for the case 0 < a < 1. Note the exponential rise in the output as a function of n. Since |a| < 1, the final value of the output as n approaches infinity is v(oo) = lim v(n) = ------n-*oc ' 1 a (2.3.30)

T o su m m arize, th e co n v o lu tio n fo rm u la p ro v id e s us w ith a m ean s fo r co m p u tin g th e re sp o n s e o f a re lax ed , lin ear tim e -in v a ria n t system to an y a rb itra ry in p u t signal x( n). It ta k e s o n e o f tw o e q u iv a le n t form s, e ith e r (2.3.17) o r (2,3.28), w h ere jt(n ) is th e in p u t sig n al to th e system , h (n ) is th e im p u lse re s p o n s e of th e system , an d y (n ) is th e out p u t o f th e system in re sp o n s e to th e in p u t signal x (n ). T he e v a lu a tio n o f th e co n v o lu tio n fo rm u la involves fo u r o p e ra tio n s , n am ely: f ol di ng e ith e r th e im p u lse re sp o n s e as specified by (2.3.17) o r th e in p u t se q u e n c e as sp ec ified by (2.3.28) to yield e ith e r h ( k) o r x { k). resp ec tiv ely , shifting th e folded se q u e n c e by n u n its in tim e to yield e ith e r h{n k ) o r x { n k ). mul t i pl yi ng the tw o se q u e n c e s to yield th e p ro d u c t se q u en ce, e ith e r x { k) h{ n k) o r x ( n - k ) h ( k ) , a n d finally s u m m i n g all th e v alu es in th e p ro d u c t se q u e n c e to y ield th e o u tp u t v (n ) o f th e sy stem a t tim e n. T h e folding o p e ra tio n is d o n e only o n ce. H o w ev er, th e o th e r th re e o p e ra tio n s a re re p e a te d fo r all p o ssib le shifts oc < n < oo in o rd e r to o b ta in y (n ) fo r oo < n < oc.

2.3.4 Properties of Convolution and the Interconnection of LTI Systems

In th is se ctio n w e in v estig ate so m e im p o rta n t p ro p e rtie s o f co n v o lu tio n an d in te r p re t th e se p r o p e rtie s in te rm s o f in te rc o n n e c tin g lin ear tim e -in v a ria n t system s. W e sh o u ld stress th a t th ese p ro p e rtie s h o ld fo r e v e ry in p u t signal. It is c o n v e n ie n t to sim plify th e n o ta tio n by using an a s te ris k to d e n o te the c o n v o lu tio n o p e ra tio n . T h u s
OC

y( n) = x{n) * h(n) = ^
Jt = - O C

x ( k ) h ( n k)

(2.3.31)

In th is n o ta tio n th e se q u e n c e follow ing th e aste risk [i.e., th e im p u lse re sp o n se /i()] is fo ld e d an d sh ifted . T h e in p u t to th e sy stem is ;c(n). O n th e o th e r h a n d , we also sh o w ed th a t
OC

>>(n) = h{n) * x( n) = ^

h ( k ) x ( n - k)

(2.3.32)

k=-oc

Sec. 2.3

Analysis of Discrete-Time Linear Time-Invariant Systems

hin)
h(n) Figure 2.25
v(n )

83

<

>

xin)

Interpretation of the commutative property of convolution.

In th is fo rm o f th e co n v o lu tio n fo rm u la, it is the in p u t signal th a t is fo ld ed . A lte r n ativ ely . we m ay in te r p re t this fo rm o f the c o n v o lu tio n fo rm u la as re su ltin g from an in te rc h a n g e o f th e ro les o f j:(n) an d h(n). In o th e r w ords, w e m ay re g a rd x( n) as th e im p u lse re sp o n se o f th e system an d h( n) as the e x c ita tio n o r in p u t signal. F ig u re 2.25 illu stra te s th is in te rp re ta tio n . W e can view co n v o lu tio n m o re ab stractly as a m a th e m a tic a l o p e ra tio n b e tw een tw o signal se q u e n c e s, say x( n ) a n d h(n), th a t satisfies a n u m b e r o f p ro p e rtie s . T h e p ro p e rty e m b o d ie d in (2.3.31) an d (2.3.32) is called th e c o m m u ta tiv e law'.

Commutative law
jf(n) * h{n) = h(n) * x( n) (2.3.33)

V iew ed m a th e m a tic a lly , the c o n v o lu tio n o p e ra tio n also satisfies th e asso cia tive law , w hich can be sta te d as follow s.

Associative law
[-v(/i) * /?[()] * h 2(/i) x( n) * [/*!() * (2.3.34)

F ro m a physical p o in t o f view, we can in te rp re t x (n) as the in p u t signal to a lin e a r tim e -in v a ria n t system w ith im pulse re sp o n se /j|(/i). T h e o u tp u t o f this sy stem , d e n o te d as v i(n ), b eco m es the in p u t to a seco n d lin e a r tim e -in v a ria n t sy stem w ith im p u lse re sp o n se hzin). T h en th e o u tp u t is y( n) V ] (n) * h 2(n) = [j:(fi) * h\(n)] * h2(n) w hich is p recisely th e left-h an d side of (2.3.34). T h u s th e le ft-h a n d side o f (2.3.34) c o rre sp o n d s to h av in g tw o lin ear tim e -in v a ria n t system s in cascad e. N ow th e righth an d side o f (2.3.34) in d icates th a t th e in p u t x ( n ) is a p p lied to an e q u iv a le n t system h av in g an im p u lse re sp o n se , say h(n), w hich is e q u a l to the co n v o lu tio n o f th e tw o im p u lse resp o n se s. T h a t is, h(n) = h](n) * h?(n) an d y(n) = x( n ) * h(n) F u rth e rm o re , sin ce th e co n v o lu tio n o p e ra tio n satisfies th e c o m m u ta tiv e p ro p e rty , o n e can in te rc h a n g e th e o r d e r o f th e tw o system s w ith re sp o n s e s h \ ( n ) a n d hzi n) w ith o u t a lte rin g th e o v erall in p u t-o u tp u t re la tio n sh ip . F ig u re 2.26 g rap h ically il lu stra te s th e asso ciativ e p ro p e rty .

84
xin) y()

Discrete-Time Signals and Systems

jr(n) h(n) = h\(n) * h2(n) y{/i)

Chap. 2

(a)

x(n)

h,(n)

y</i)

v(n)

h2(n)

h t(n)

(b)

Figure Z 2 6

Implications of the associative (a) and the associative and commuta tive (b) properties of convolution.

Example 2.3.4
D eterm ine the impulse response for the cascade of two iinear time-invariant systems having impulse responses h \ () = ( j W " ) and

Solution To determ ine the overall impulse response of the two systems in cascade,
we simply convolve h}{n) with h2(n). Hence

where h2(n) is folded and shifted. We define the product sequence v(k) = h\(k)h2(n - k) = ()*(*)"-* which is nonzero for k > 0 and n - k > 0 or n > k > 0. On the o th er hand, for n < 0, we have v(k) = 0 for all k, and hence h{n) = 0, n < 0 For n > k > 0. the sum of the values of the product sequence v(k) over all k yields hin) =
tscO

= Hr = (i)"(2"+l - 1 )

Sec. 2.3

Analysis of Discrete-Time Linear Tim e-Invariant Systems

85

T h e g e n e ra liz a tio n o f the associative law to m o re th a n tw o system s in cascade follow s easily fro m th e d iscu ssio n given ab o v e. T h u s if w e h av e L lin e a r tim ein v arian t sy stem s in cascade w ith im p u lse re sp o n se s h\ ( u) . h ^ i n ) ........ h L (n). th ere is an e q u iv a le n t lin e a r tim e -in v a ria n t sy stem hav in g an im p u lse re sp o n se th a t is e q u a l to th e (L l)-fo id co n v o lu tio n of th e im p u lse resp o n se s. T h a t is. h( n) = h \ { n ) * hzi n) * * h L (n) (2.3.35)

T h e c o m m u ta tiv e law im p lies th a t th e o rd e r in w hich the c o n v o lu tio n s a re p e r fo rm e d is im m a te ria l. C o n v e rsely , an y lin e a r tim e -in v a ria n t system can be d e c o m p o se d in to a cascad e in te rc o n n e c tio n o f subsystem s. A m e th o d for accom plishing th e d e c o m p o s itio n will be d escrib ed later. A th ird p ro p e rty th a t is satisfied by th e c o n v o lu tio n o p e ra tio n is the d istrib u tive law, w hich m ay be sta te d as follow s.

Distributive law
xin) * 4= .xin) * h\ {n) 4- x i n ) * hzi n) (2.3.36)

I n te rp r e te d physically, this law im plies th a t if we h ave tw o lin ear tim ein v a ria n t sy stem s w ith im p u lse re sp o n se s h \ i n ) a n d /?;() ex cited by the sam e in p u t signal .r(/;), th e sum of the tw o resp o n ses is identical to the resp o n se of an o v erall system w ith im pulse resp o n se h i n ) = )i\ in) 4- //:(/;) T h u s th e o v erall system is view ed as a p arallel co m b in a tio n of the tw o linear tim e -in v a ria n t sy stem s as illu stra te d in Fig. 2.27, T h e g e n e ra liz a tio n o f (2.3.36) to m o re th an tw o lin e a r tim e-in v arian t sys tem s in p a rallel follow's easily by m a th e m a tic a l in d u ctio n . T h u s th e in te rc o n n e c tio n of L lin ear tim e -in v a ria n t system s in p a ra lle l w ith im p u lse resp o n ses h\ i n) . h z i n ) .........h L{n) a n d ex cited by the sa m e in p u t x i n ) is e q u iv a le n t to o n e overall system w ith im p u lse re sp o n se
L

h(n) ^ hj i n) (=i

(2.3.37)

C o n v e rsely , an y lin ear tim e -in v a ria n t system can be d e c o m p o se d into a p arallel in te rc o n n e c tio n o f su b sy stem s.

Figure 2.27 Interpretation of the distributive property of convolution: two LTI systems connected in parallel can be replaced by a single system with h(n) =

(n) + k2{n).

86

Discrete-Time Signals and Systems

Chap. 2

2.3.5 Causal Linear Time-Invariant Systems

In S ectio n 2.2.3 w e d efin ed a causal system as o n e w hose o u tp u t at tim e n d e p en d s o n ly on p re s e n t an d p ast in p u ts b u t d o es n o t d e p e n d on f u tu re in p u ts. In o th e r w o rd s, th e o u tp u t o f the. sy stem at so m e tim e in s ta n t n, say n = no, d e p e n d s only on v alu es o f jc() fo r n < n 0In th e case o f a lin e a r tim e -in v a ria n t system , cau sality can b e tra n sla te d to a c o n d itio n o n th e im p u lse resp o n se. T o d e te rm in e this re la tio n sh ip , le t us co n sid e r a lin ear tim e -in v a ria n t system having an o u tp u t a t tim e n = no given by th e c o n v o lu tio n fo rm u la O C
v ( o ) =

^2
k - o c

h ( k ) x (n o ~ k )

S u p p o se th a t w e su b d iv id e th e sum in to tw o sets o f term s, o n e se t involving p re se n t a n d p a st v alu es o f th e in p u t [i.e.. x{n) for n < n 0] a n d o n e se t involving fu tu re valu es o f th e in p u t [i.e., n > no]. T h u s we o b ta in

O C
y (n u) ^ h ( k ) x ( n o - k) +
i=0

-I

^
k oc

h( k) x( nu - k)
1) + h ( 2 ) x ( n 0 - 2 ) + ]

= [/ ?(0)x (n ()) + h( \ ) x( r tu -

+ [/?( 1 )x((] + 1) + h ( - 2 ) x ( n o + 2) + ]

W e o b se rv e th a t th e term s in th e first sum involve jr(no), x( no 1).........w hich are th e p re s e n t a n d p ast values of th e in p u t signal. O n th e o th e r h a n d , th e te rm s in th e se co n d sum in volve th e in p u t signal c o m p o n e n ts ;c(no + l) , x { n o -f 2 ) .........N ow , if th e o u tp u t a t tim e n = n 0 is to d e p e n d only on th e p re s e n t a n d p a st in p u ts, th en , clearly , th e im p u lse re sp o n se o f th e system m u st satisfy th e c o n d itio n ft(n) = 0 n < 0 (2.3.38)

Since h{n) is th e re sp o n se o f th e re lax ed lin e a r tim e -in v a ria n t sy stem to a u nit im p u lse a p p lied a t n = 0, it follow s th a t h(n) = 0 fo r n < 0 is b o th a necessary an d a sufficient c o n d itio n fo r causality. H e n c e an L T I syst em is causal i f a n d onl y i f its i mpul se respons e is z e r o f o r negative values o f n. Since fo r a cau sal sy stem , h(n) = 0 fo r n < 0, th e lim its on th e su m m a tio n of th e co n v o lu tio n fo rm u la m ay be m odified to reflect th is re stric tio n . T h u s w e h ave th e tw o e q u iv a le n t form s
OC

y ( n ) = ^ h ( k ) x ( n - k) Jt=0 n = ^ x{k) h{n k) k=~oc

(2.3.39)

(2.3.40)

A s in d icated p rev io u sly , cau sality is re q u ire d in an y re a l-tim e signal p ro c e ss ing a p p licatio n , since at a n y given tim e n w e have n o access to f u tu re v alu es o f th e

Sec. 2.3

Analysis of Discrete-Time Linear Time-Invariant Systems

87

in p u t signal. O n ly th e p re s e n t a n d p ast v alu es o f the in p u t signal are av ailab le in c o m p u tin g th e p re s e n t o u tp u t. It is so m e tim e s c o n v e n ie n t to call a se q u en ce th a t is z e ro fo r n < 0, a causa! s e q u e n c e , an d o n e th a t is n o n z e ro fo r n < 0 a n d n > 0. a n on c aus al sequence. T his te rm in o lo g y m e a n s th a t su c h a se q u e n c e could be th e u n it sa m p le re sp o n se of a causal o r a n o n c a u sa l system , resp ectiv ely . If th e in p u t to a causal lin e a r tim e -in v a ria n t system is a causal se q u e n c e [i.e., if jr(n) = 0 fo r n < 0]. th e lim its on th e co n v o lu tio n fo rm u la a re fu rth e r restricted . In th is case th e tw o e q u iv a le n t form s o f th e c o n v o lu tio n fo rm u la b eco m e n y ( n ) = ^ h( k ) x ( n k) (2.3.41) *=o n = -k) (2.3.42) * = < ) W e o b se rv e th a t in th is case, th e lim its on the su m m a tio n s fo r the tw o a lte rn a tiv e fo rm s are id en tical, a n d th e u p p e r lim it is grow ing w ith tim e. C learly , th e resp o n se o f a cau sal sy stem to a causal in p u t se q u e n c e is causal, since y( n) 0 fo r n < 0. Example 2.3.5 Determ ine the unit step response of the linear time-invariant system with impulse response
h{n) = a " u( n ) \a\ < 1

Solution Since the input signal is a unit step, which is a causal signal, and the system is also causal, we can use one of the special forms of the convolution formula, either (2.3.41) or (2.3.42). Since x(n) = 1 for n > 0. (2.3.41) is simpler to use Because of the simplicity of this problem, one can skip the steps involved with sketching the folded and shifted sequences. Instead, we use direct substitution of the signals sequences in (2.3.41) and obtain y(n) = y ~ v *=(I 1 - a"*1 1 -ci and y(n) = 0 for n < 0. We note that this result is identical to that obtained in Ex ample 2.3.3. In this simple case, however, we computed the convolution algebraically without resorting to the detailed procedure outlined previously.

2.3.6 Stability of Linear Time-Invariant Systems

A s in d ic a te d p rev io u sly , sta b ility is an im p o rta n t p ro p e rty th a t m u st be co n sid e re d in an y p ra c tic a l im p le m e n ta tio n o f a system . W e d efin ed an a rb itra ry relax ed system as B IB O sta b le if a n d only if its o u tp u t se q u e n c e y () is b o u n d e d fo r every b o u n d e d in p u t x ( n) .

88

Discrete-Time Signals and Systems

Chap. 2

If x ( n ) is b o u n d e d , th e re exists a c o n s ta n t M x such th a t
l* (n )l < M x

<00

S im ilarly, if th e o u tp u t is b o u n d e d , th e re exists a c o n s ta n t M y such th a t

| ^ ( n ) | < My < OO

fo r all n. N ow , given such a b o u n d e d in p u t se q u e n c e x ( n ) to a lin e a r tim e -in v a ria n t sy stem , le t us in v e stig a te th e im p licatio n s of th e d efin itio n o f sta b ility o n th e c h a r acteristics o f th e system . T o w a rd th is en d , w e w o rk again w ith th e c o n v o lu tio n fo rm u la O O y( n) = h{k) x{n - k ) k= oc If w e ta k e th e a b so lu te value of b o th sides o f th is e q u a tio n , w e o b tain Lv()| =
Y
* = -O C

h( k ) x ( n k)

N ow , th e ab so lu te v alu e of th e sum o f term s is alw ays less th a n o r e q u a l to the sum o f th e a b so lu te v alu es o f th e term s. H e n c e
OC

\y(n)\ < Y
i= oc

IM*)II*(/1 - fc)l

If th e in p u t is b o u n d e d , th e re exists a finite n u m b e r M x such th a t |x(n)[ < M t , By su b s titu tin g th is u p p e r b o u n d fo r x( n ) in th e e q u a tio n ab o v e, w e o b ta in

OC'

|y (n )| < M X Y
k = -o c

W * )l

F ro m th is ex p ressio n w e o b se rv e th a t th e o u tp u t is b o u n d e d if th e im p u lse resp o n se o f th e sy stem satisfies th e co n d itio n

OO

Sh =

Y
k = -o c

IAWI < 00

< 2-3-43)

T h a t is, a linear time-invari ant s y s t em is stable i f its i mpu l s e respons e is absolutely s u m m a b l e . T h is c o n d itio n is n o t only sufficient b u t it is also n e c e ssa ry to e n su re th e sta b ility o f th e system . In d e e d , w e shall show th a t if 5* = 00, th e r e is a b o u n d e d in p u t fo r w hich th e o u tp u t is n o t b o u n d e d . W e c h o o se th e b o u n d e d in p u t \h*(-n)\ 0, h{n) ? 0 h(n) = 0

w h e re h*(n) is th e co m p lex c o n ju g a te o f h(n). I t is sufficient to show th a t th e re is o n e v alu e o f n fo r w h ich y( n) is u n b o u n d e d . F o r n = 0 w e h av e ,< 0 ,=

*= 00^

;
k = (x. 1 v /l

Thus, if Si, = 00, a bounded input produces an unbounded output since y(0) = 00.

Sec. 2.3

Analysis of Discrete-Time Linear Tim e-Invariant Systems

89 zero zero this, + N,

T h e c o n d itio n in (2.3.43) im plies th a t th e im p u lse re s p o n s e h(n) goes to as n a p p ro a c h e s infinity. A s a co n se q u e n c e , th e o u tp u t o f th e system g o es to as n a p p ro a c h e s infinity if th e in p u t is set to z e ro b ey o n d n > n 0. T o p ro v e su p p o se th a t |j ( h ) | < M x fo r n < no an d x ( n ) = 0 fo r n > no- T h e n , at n = no th e system o u tp u t is (no + AO = ^ h ( k ) x ( n 0 + N - k) + ^ *-=-oc h { k ) x ( n 0 + N - k)

B u t th e first su m is z e ro since * () = 0 fo r n > no- F o r th e rem ain in g p a rt, we ta k e th e a b so lu te v alu e o f th e o u tp u t, w hich is

j oc I sc

y (0 + AO| =

h( k)x(no + N )j <

|/i(/:)||jf(no + N )|

< Mx k=N

W *)!

N ow , as N a p p ro a c h e s infinity.

an d hen ce

lim ]\ {/i() + AO! = 0

T h is re su lt im p lies th a t any ex citatio n at th e in p u t to th e system , w hich is of a finite d u ra tio n , p ro d u c e s an o u tp u t th a t is tra n s ie n t in n a tu re ; th a t is, its am p litu d e d ecay s w ith tim e an d dies o u t e v en tu ally , w h en th e system is stab le. Example 2.3.6 Determ ine the range of values of the param eter a for which the linear time-invariant system with impulse response
hin) = a"u(n)

is stable. Solution First, we note that the system is causal. Consequently, the lower index on the summation in (2.3.43) begins with k = 0. Hence

Clearly, this geom etric series converges to

provided that \a\ < 1 . Otherwise, it diverges. Therefore, the system is stable if |a| < 1. Otherwise, it is unstable. In effect, h(.n) must decay exponentially toward zero as n approaches infinity for the system to be stable.

90

Discrete-Time Signals and Systems

Chap. 2

Example 23.7
D eterm ine the range of values of a and b for which the linear time-invariant system with impulse response ... ( a", n > 0 h(-n) = [ iw. f, n < n 0 is stable.

Solution This system is noncasual. The condition on stability given by (2.3.43) yields
OC OC

1
ia i" +
y i
n= -o c

t :
n= -o c

i*()f=
fi=s(J

|fe,n

From Example 2.3.6 we have already determ ined that the first sum converges for |a| < 1. The second sum can be manipulated as follows:

= p( l + p + p 2 + )=

I - p

where p = \f\b\ must be less than unity for the geometric series to converge. Conse quently, the system is stable if both \a\ < 1 and |fc| > 1 are satisfied,

2.3.7 Systems with Finite-Duration and Infinite-Duration Impulse Response

U p to this p oin t w e have characterized a linear tim e-invariant system in term s of its im pulse resp onse h(n). It is also con ven ien t, how ever, to subdivide the class o f linear tim e-invariant system s in to tw o types, those that h ave a finite-duration im pulse resp onse (F IR ) and those that have an infinite-duration im pulse response (IIR ). Thus an F IR system has an im pulse resp onse that is zero ou tsid e o f som e finite tim e interval. W ithout loss o f generality, w e focus our atten tion on causal FIR system s, so that h (n) = 0 n < 0 and n > M

The con volu tion form ula for such a system reduces to u-1 ? ( n ) = H h ^ x(^n ~ k ) t=0 A useful interpretation o f this exp ression is ob tain ed by ob serving that th e output at any tim e n is sim ply a w eigh ted iinear com b ination o f the in p ut signal sam ples x (n ), x { n - 1 ) , , x ( n - M + 1). In other words, the system sim ply w eigh ts, by the values o f the im pulse resp onse h(k), k = 0, 1, 1, the m ost recent M signal sam ples and sum s the resulting M products. In e ffe ct, th e system acts as a w in d o w that view s on ly the m ost recent M input signal sam p les in form ing the output. It n eg lects or sim ply forgets all prior input sam p les [i.e., x ( n M ),

Sec. 2.4

Discrete-Time Systems Described by Difference Equations

91

x ( n M 1). . . .] . T h u s we say th a t an F IR svstem has a finite m e m o ry o f len g th -M sam p les. In c o n tra st, an I I R lin e a r tim e-in v arian t system has an in fin ite -d u ra tio n im pu lse resp o n se. Its o u tp u t, b ased on th e c o n v o lu tio n fo rm u la, is v(n) = ^ >2h(k) x( n - k ) w h ere cau sality h as b e e n assu m ed , a lth o u g h this a ssu m p tio n is n o t n ecessary. N ow . th e system o u tp u t is a w eig h te d [by the im p u lse re sp o n se *(/:}] lin e a r co m b in a tio n o f th e in p u t sig n al sa m p le s .r(n), x{n - 1), x( n 2 ) ........ S ince this w eig h te d sum in v olves th e p re s e n t an d all th e p ast in p u t sam ples, w e say th a t th e sy stem has an infinite m em o ry . W e in v e stig a te th e c h aracteristics of F IR an d IIR sy stem s in m o re d etail in su b s e q u e n t c h a p te rs.

2.4 DISCRETE-TIME SYSTEMS DESCRIBED BY DIFFERENCE EQUATIONS

U p to this p o in t w e h ave tre a te d linear and tim e-in v arian t sy stem s th a t are c h a r a c terized by th e ir u n it sa m p le resp o n se h(n). In tu rn . h(n) allow s us to d e te rm in e th e o u tp u t v{n) o f th e system for any given in p u t se q u e n c e jc(/i) by m e a n s o f the c o n v o lu tio n su m m a tio n . (2.4.1) In g e n e ra l, th e n , we h av e show n th a t any linear tim e -in v a ria n t system is c h a r a c terized by th e in p u t- o u tp u t rela tio n sh ip in (2.4.1). M o re o v e r, th e co n v o lu tio n su m m a tio n fo rm u la in (2.4.1) suggests a m ean s fo r th e re a liz a tio n o f th e system . In th e case o f F IR sy stem s, such a realizatio n in v o lv es a d d itio n s, m u ltip lic atio n s, a n d a finite n u m b e r o f m e m o ry locations. C o n se q u e n tly , an F IR system is read ily im p le m e n te d d irectly , as im p lied by the c o n v o lu tio n su m m atio n . If th e sy stem is IIR , h o w ev er, its practical im p le m e n ta tio n as im p lied by c o n v o lu tio n is clearly im p o ssib le, since it re q u ire s an infinite n u m b e r o f m e m o ry lo catio n s, m u ltip lic a tio n s, an d ad d itio n s. A q u e stio n th a t n a tu ra lly arises, th e n , is w h e th e r o r n o t it is possible to realize IIR sy stem s o th e r th a n in the form su g g e sted by th e c o n v o lu tio n su m m atio n . F o rtu n a te ly , th e an sw er is yes. th e re is a p ra c tic a l an d c o m p u ta tio n a lly efficient m e a n s fo r im p le m e n tin g a fam ily o f IIR sy stem s, as will b e d e m o n s tra te d in th is se ctio n . W ith in th e g en eral class o f IIR sy stem s, this fam ily of d isc re te -tim e sy stem s is m o re c o n v en ien tly d e s c rib e d by d ifferen ce eq u atio n s. T h is fam ily o r subclass of IIR sy stem s is v ery u sefu l in a v ariety o f p ractical ap p licatio n s, in clu d in g th e im p le m e n ta tio n o f d ig ita l filters, a n d the m o d elin g o f physical p h e n o m e n a a n d physical system s.

92

Discrete-Time Signals and Systems

Chap. 2

2.4.1 Recursive and Nonrecursive Discrete-Time Systems

A s in d icated ab o v e, th e co n v o lu tio n su m m a tio n fo rm u la e x p re sse s th e o u tp u t of th e lin e a r tim e -in v a ria n t sy stem explicitly a n d only in te rm s o f th e in p u t signal. H o w e v e r, th is n e e d n o t b e th e case, as is sh o w n h e re . T h e re a re m an y system s w h ere it is e ith e r n ecessa ry o r d e sirab le to ex p re ss th e o u tp u t o f th e system n o t on ly in term s o f th e p re s e n t a n d p a st v alu es o f th e in p u t, b u t also in te rm s o f the a lre a d y av ailab le p a st o u tp u t values. T h e follow ing p ro b le m illu stra te s th is point. S u p p o se th a t w e w ish to c o m p u te th e cumul at i ve average o f a signal x ( n ) in th e in te rv a l 0 < k < n, d efin ed as 1 " v(n) = ------- = 0 . 1, . . . n + 1 ~f. (2.4.2)

A s im p lied b y (2.4.2), th e c o m p u ta tio n o f _v(n) re q u ire s th e sto ra g e o f all th e in p u t sa m p le s x (k ) fo r 0 < k < n. Since n is in creasin g , o u r m em o ry r e q u ire m e n ts grow lin early w ith tim e. O u r in tu itio n suggests, h o w ev er, th a t y( n) can be c o m p u te d m o re efficiently by utilizin g th e p re v io u s o u tp u t v alue y( n 1). In d e e d , by a sim p le alg eb raic re a rra n g e m e n t o f (2.4.2), we o b ta in it l ( + l)y (n ) = x(k) + x(n)

= ny( n - 1) + x {n) a n d h en ce y( n) = 1 -x(n) y(n - 1) + n + 1' n + 1' (2.4.3)

N ow , th e cu m u lativ e a v erag e v(n) can b e c o m p u te d recu rsiv ely b y m u ltip ly in g th e p re v io u s o u tp u t v a lu e y( n - 1) by n/ ( n 4-1 ), m u ltip ly in g th e p r e s e n t in p u t x ( n ) by 1 / (n + 1), an d a d d in g th e tw o p ro d u cts. T h u s th e c o m p u ta tio n o f y (n ) by m ean s o f (2.4.3) re q u ire s tw o m u ltip lic atio n s, o n e a d d itio n , a n d o n e m e m o ry lo catio n , as illu stra te d in Fig. 2.28. T his is an ex a m p le of a recursive s yst em. In g e n e ra l, a sy stem w hose o u tp u t v(n) a t tim e n d e p e n d s o n a n y n u m b e r o f p a s t o u tp u t values y( n - 1), y ( n - 2 ) , . . . is called a recu rsiv e system .
tin)

Figure 2.28

Realization of a recursive cum ulative averaging system.

Sec. 2.4

Discrete-Time Systems Described by Difference Equations

93

T o d e te rm in e th e c o m p u ta tio n of the recu rsiv e sy stem in (2.4.3) in m ore d etail, su p p o se th a t we begin th e p ro cess w ith n = 0 a n d p ro c e e d fo rw a rd in tim e. T h u s, acco rd in g to (2.4.3). we o b ta in y(0) jc(0)

v (l) = 5.v(0) + U ( l ) y(2) = y ( l ) + i* ( 2 ) an d so on. If o n e grow s fatig u ed w ith this c o m p u ta tio n an d w ishes to pass the p ro b le m to so m e o n e else at som e tim e, say n = no. th e only in fo rm a tio n th a t one n e e d s to p ro v id e his o r h e r su ccesso r is the p a st value y(o - 1) a n d the new' input sa m p le s jr(n), j (/j + 1 )........ T h u s the successor b eg in s w ith

an d p ro c e e d s fo rw ard in tim e until som e tim e, say n = n j. w h en he or she b e co m es fatig u ed an d passes the c o m p u ta tio n a l b u rd e n to so m e o n e else w ith the in fo rm a tio n o n th e value _v(/?i 1). an d so on. T h e p o in t wc wish to m ak e in th is discussion is th a t if o n e w ishes to co m p u te th e re sp o n se (in this case, the cu m u lativ e av e ra g e ) of the sy stem (2,4.3) to an input signal x( n ) a p p lied at n tiu. we n eed th e value y (n (, - 1) an d th e input sam ples x ( n ) for /? > /in. T h e term y(>i() 1) is called th e initial condi t i on for the system in (2.4.3) an d c o n ta in s all the in fo rm a tio n n e e d e d to d e te rm in e th e re sp o n se o f the sy stem for n > () to th e in p u t signal x (n ). in d e p e n d e n t o f w h at has o ccu rred in th e p ast. T h e fo llow ing ex am p le illu strates th e use o f a (n o n lin e a r) recu rsiv e system to co m p u te th e sq u a re ro o t of a n u m b e r. Exam ple 2.4.1 Square-Root Algorithm

Many computers and calculators compute the square root of a positive number A. using the iterative algorithm

sn ~ s_ j. T hen it easily follows that

where j_i is an initial guess (estimate) of \/~A. As the iteration converges we have = J~A. Consider now the recursive system (2.4.4)

which is realized as in Fig, 2.29. If we excite this system with a step of amplitude A [i.e.. x(n) = A u ( n )] and use as an initial condition y(1) an estimate of the response v() of the system will tend toward as n increases. Note that in contrast to the system (2.4.3), we do not need to specify exactly the initial condition. A rough estim ate is sufficient for the proper perform ance of the system. For example, if we

94

Discrete-Time Signals and Systems

Chap. 2

4n)

-0---- -0
v(n - I) Figure Z 2 9 Realization of the square-root system.

let A = 2 and y ( - l ) = 1, we obtain j(0 ) = >-(1) = 1,4166667, v(2) = 1,4142157. Similarly, for y ( 1) = 1.5, we have v(0) = 1,416667, y (l) = 1.4142157. Compare these values with the %/2, which is approximately 1.4142136. W e h av e no w in tro d u c e d tw o sim ple recu rsiv e system s, w h ere th e o u tp u t vf/i) d e p e n d s o n th e p re v io u s o u tp u t v alu e y( n 1) a n d th e c u rre n t in p u t ;r(n). B o th system s a re causal. In g e n e ra l, w e can fo rm u la te m o re co m p lex cau sal recu rsiv e system s, in w hich th e o u tp u t y ( n ) is a fu n ctio n o f se v era l p ast o u tp u t v alu es an d p re se n t a n d p ast in p u ts. T h e system sh o u ld h av e a finite n u m b e r o f d elay s o r, eq u iv alen tly , sh o u ld re q u ire a finite n u m b e r o f sto ra g e lo catio n s to be p ractically im p le m e n te d . T h u s th e o u tp u t o f a cau sal a n d p ractically re a liz a b le recursive system can be e x p ressed in g e n e ra l as y( n) = F[y( n 1), y( n 2 ) , . . . , y( n N ) , x ( n) , x ( n 1 ) , . . . , x ( n M ) \ (2.4.5)

w h ere F [ ] d e n o te s so m e fu n ctio n of its a rg u m e n ts. T h is is a re c u rsiv e e q u a tio n specifying a p ro c e d u re fo r co m p u tin g th e system o u tp u t in term s o f p rev io u s v alu es o f th e o u tp u t a n d p re s e n t an d p ast inputs. In c o n tra st, if y{n) d e p e n d s only o n th e p re s e n t an d p a st in p u ts, th e n y (n ) = F[ x ( n ) , x ( n 1)........ x(rt M) ] (2.4.6)

S uch a sy stem is called nonrecursi ve. W e h a s te n to ad d th a t th e c a u s a l F IR system s d escrib ed in S ectio n 2.3.7 in te rm s o f th e co n v o lu tio n sum fo rm u la h av e th e fo rm o f (2.4.6). In d e e d , th e c o n v o lu tio n su m m a tio n fo r a cau sal F IR sy stem is
M

y ( n) = i=0

- *)

= h( 0 ) x ( n ) -)- h ( \ ) x ( n 1) + - + h ( M ) x ( n M ) F [j:(n ), x( n 1), . . . , x ( n M )] w h ere th e fu n ctio n F[-] is sim ply a lin e a r w eig h ted su m o f p re s e n t a n d p ast in p u ts a n d th e im p u lse re sp o n s e values h( n), 0 < n < M , c o n s titu te th e w eig h tin g c o e f ficients. C o n se q u e n tly , th e cau sal lin e a r tim e -in v a ria n t F IR sy stem s d e s c rib e d by th e c o n v o lu tio n fo rm u la in S ectio n 2.3.7, a re n o n re c u rsiv e . T h e b a sic d ifferen ces b e tw e e n n o n re c u rsiv e a n d re c u rsiv e system s a re illu stra te d in Fig. 2.30. A sim ple in sp e ctio n o f th is figure re v e a ls th a t th e fu n d a m e n ta l d iffe re n c e b e tw e e n th e s e tw o

Sec. 2.4

Discrete-Time Systems Described by Difference Equations

95

x(n)

F[Mn). xin - 1),

........ ti n - .Wl]

V(H )

K n ) ........... v(/i -

M )\
]

1 --------------------------------- 1 (b)

1 ^

x{n)

| |

y{n)

Figure 2.30 Basic form for a causal and realizable (a) nonrecursive and (b) recursive system.

system s is the fe e d b ack loop in th e recu rsiv e system , w hich feed s b ack th e o u tp u t o f the system in to th e in p u t. T h is feed b ack loop co n tain s a d e la y ele m e n t. T he p resen c e o f this d elay is crucial for the realizab ility of the system , since the ab sen ce of this d e la y w ould force the system to c o m p u te yi n) in term s o f v(n). w hich is n o t po ssib le fo r d isc re te -tim e system s. T h e p re se n c e o f the fe ed b ack loop o r, eq u iv alen tly , the recu rsiv e n a tu re of (2.4.5) c re a te s a n o th e r im p o rta n t d ifferen ce b etw een recursive an d n o n re c u rsiv e system s. F o r e x am p le, su p p o se th a t we wish to c o m p u te th e o u tp u t y(o) of a system w h en it is ex cited by an in p u t ap p lied at tim e n = 0. If th e system is recu rsiv e, to co m p u te y ( 0). we first n e e d to c o m p u te all the p re v io u s v alu es y(0). y ( l ) ........ y(o - 1)- In c o n tra st, if the system is n o n recu rsiv e. we can c o m p u te the o u tp u t y (n 0) im m e d ia te ly w ith o u t having y(no - 1), y(o 2 )........ In co n clu sio n , th e o u tp u t o f a recu rsiv e system sh o u ld be c o m p u te d in o rd e r [i.e., v(0), y ( l) , y ( 2 ) . . . w h e re a s for a n o n re c u rsiv e system , th e o u tp u t can b e c o m p u te d in any o rd e r [i.e., y(200). y (15). y{3). y(300). etc.]. T his fe a tu re is d e sira b le in so m e p ractical a p p licatio n s.

2.4.2 Linear Time-Invariant Systems Characterized by Constant-Coefficient Difference Equations

In S ectio n 2.3 w e tre a te d lin e a r tim e -in v a ria n t system s a n d c h a ra c te riz e d th em in te rm s o f th e ir im p u lse resp o n ses. In this su b sectio n w e focus o u r a tte n tio n o n a fam ily o f iin e a r tim e -in v a ria n t system s d escrib ed by an in p u t- o u tp u t r e la tio n called a d iffe re n c e e q u a tio n w ith c o n s ta n t c o e ffic ie n ts . S ystem s d e sc rib e d by c o n s tan t-co efficien t lin e a r d ifferen ce e q u a tio n s are a subclass o f the recu rsiv e a n d n o n re c u rsiv e sy stem s in tro d u c e d in th e p re ced in g su b sectio n . T o b rin g o u t th e im p o rta n t id eas, w e b eg in by tre a tin g a sim ple re c u rsiv e sy stem d e sc rib e d by a first-o rd e r d iffe re n c e e q u a tio n .

96

Discrete-Time Signals and Systems

Chap. 2

<*>
a
Figure 131 Block diagram realization of a simple recursive system.

S u p p o se th a t w e h a v e a recu rsiv e system w ith an in p u t- o u tp u t e q u a tio n y( n) = ay( n - 1) + x ( n ) (2.4.7)

w h ere a is a c o n sta n t. F ig u re 2.31 show s a block d ia g ra m re a liz a tio n o f th e system . In co m p a rin g th is sy stem w ith th e cu m u la tiv e a v erag in g sy stem d e sc rib e d by the in p u t-o u tp u t e q u a tio n (2.4.3), w e o b se rv e th a t th e system in (2.4.7) h as a c o n sta n t co effic ie n t (in d e p e n d e n t o f tim e ), w h e re a s th e sy stem d e sc rib e d in (2.4.3) h as tim ev a ria n t coefficients. A s w e will show , (2.4.7) is an in p u t- o u tp u t e q u a tio n fo r a lin e a r tim e -in v a ria n t sy stem , w h e re a s (2.4.3) d escrib es a lin e a r tim e -v a ria n t system . N ow , su p p o se th a t w e ap p ly an in p u t signal x ( n ) to th e sy stem fo r n > 0. W e m a k e n o assu m p tio n s a b o u t th e in p u t sig n a l fo r n < 0, b u t w e d o assum e th e ex iste n ce o f th e in itial c o n d itio n v ( 1). S ince (2.4.7) d e s c rib e s th e system o u tp u t im plicitly, w e m u st solve this e q u a tio n to o b ta in an ex p licit ex p ressio n for th e sy stem o u tp u t. S u p p o se th a t w e co m p u te successive valu es o f y(n) fo r n > 0, b eg in n in g w ith y(0), T h u s y (0) = o j ( - l ) + x(0) v (l) = ay (0 ) + jc(3 ) = a 2y ( - 1) + o j:( 0 ) + *(1) y(2) = o j ( l ) + x ( 2 ) = o 3y (1) + a 2jr(0) + a ^ ( l ) + x ( 2 )

y( n) = a y( n - 1) + x( n) = e " +1y ( - l ) + a" x( 0) + fl" 1Jt(l) + + a x ( n - 1) + x( n) o r, m o re co m p actly , n > 0 (2.4.8)

T h e re sp o n se y ( n ) o f th e system as given by th e rig h t-h a n d sid e o f (2.4.8) co n sists o f tw o p arts. T h e first p a rt, w hich c o n ta in s th e te rm y ( 1), is a re su lt of th e in itial c o n d itio n y ( 1) o f th e system . T h e se co n d p a r t is th e re sp o n se o f the sy stem to th e in p u t sig n al x( n) . I f th e sy stem is initially re lax ed a t tim e n = 0, th e n its m e m o ry (i.e., the o u tp u t o f th e d e la y ) sh o u ld b e zero . H e n c e y ( 1) = 0. T h u s a rec u rsiv e system is re la x e d if it sta rts w ith z e ro in itial c o n d itio n s. B e c a u se th e m e m o ry o f th e system d escrib es, in so m e sen se, its sta te , w e say th a t th e system is a t z e ro s ta te an d its c o rre sp o n d in g o u tp u t is called th e zero-state respons e o r f o r c e d respons e , an d

Sec. 2.4

Discrete-Time Systems Described by Difference Equations

97

is d e n o te d by yzs(n). O b v iously, the z e ro -sta te re sp o n se o r fo rc e d resp o n se o f the system (2.4.7) is given by V/jittf) Y ^ a kx( n k) *=() n > 0 (2.4.9)

It is in te re stin g to n o te th at (2.4.9) is a co n v o lu tio n su m m a tio n involving the in p u t signal co n v o lv ed w ith the im p u lse re sp o n se h(n) = a':u ( n ) (2.4.10)

W e also o b se rv e th a t the sy stem d e sc rib e d by th e first-o rd e r d ifferen ce e q u atio n in (2.4.7) is cau sal. A s a resu lt, th e low er lim it on th e c o n v o lu tio n su m m atio n in (2.4.9) is k = 0. F u rth e rm o re , th e co n d itio n v (1) = 0 im plies th a t the in p u t signal can be a ssu m ed cau sal an d h en ce th e u p p e r lim it on th e co n v o lu tio n su m m atio n in (2.4.9) is n. since x( n - k) = 0 for k > n. In effect, we have o b ta in e d th e resull th a t th e re lax ed recu rsiv e system d esc rib e d by th e firs t-o rd e r d ifferen ce e q u atio n in (2.4.7), is a lin e a r tim e-in v arian t IIR svstem w ith im p u lse resp o n se given bv (2.4.10). N ow . su p p o se th at th e system d e sc rib e d by (2.4.7) is in itially n o n re la x e d [i.e.. y ( 1) ^ 0] an d th e in p u t x(/i) = 0 for all //. T h e n the o u tp u t of th e system with zero in p u t is called the zero- i nput response or nat ural r espons e an d is d e n o te d by yZj(/j). F ro m (2.4.7). w ith ,v() = 0 for oc < n < oc . we o b ta in
\Y,OM = ',+ l y ( 1)
n >

(2.4.11)

W e o b se rv e th a t a recu rsiv e system w ith n o n z e ro initial c o n d itio n is n o n relax ed in th e se n se th a t it can p ro d u c e an o u tp u t w ith o u t b ein g ex cited . N o te th at the z e ro -in p u t re sp o n se is d u e to th e m e m o ry of th e system . T o su m m a riz e , th e z e ro -in p u t re sp o n se is o b ta in e d by se ttin g the in p u t signal to z e ro , m ak in g it in d e p e n d e n t of the in p u t. It d e p e n d s only on th e n a tu re of the system a n d th e in itial co n d itio n . T h u s th e z e ro -in p u t re sp o n s e is a ch a ra c te ristic of th e sy stem itself, a n d it is also know n as th e nat ural o r f ree r esponse of th e svstem . O n th e o th e r h a n d , th e z e ro -sta te resp o n se d e p e n d s on th e n a tu re of th e system an d th e in p u t signal. Since th is o u tp u t is a re sp o n se fo rce d u p o n it by th e input signal, it is u su ally called th e f o r c e d response of th e system . In g en eral, th e total re sp o n se o f th e system can b e ex p ressed as v(n) = y Zi ( n ) + .Vzs( n). T h e sy stem d escrib ed by th e first-o rd e r d iffe re n c e e q u a tio n in (2.4.7) is the sim p lest p o ssib le recu rsiv e system in th e g en eral class o f re c u rsiv e system s d e sc rib ed by lin e a r co n stan t-co efficien t d iffe re n c e e q u a tio n s . T h e g e n eral form for such an e q u a tio n is
S M

v(/i ) = ^ ai,y{n - k) -j- ^ bkx ( n - k) i-=l k=0 or, e q u iv alen tly ,

Ar
M

(2.4,12)

' Y a ky{n - k) = ' Y ^ b kx( n - k) Jc=0 *=0

a0 m 1

(2.4.13)

98

Discrete-Time Signals and Systems

Chap. 2

T h e in te g e r N is called th e order o f the d ifferen ce e q u a tio n o r th e o r d e r of the system . T h e n eg ativ e sign on th e rig h t-h a n d side o f (2.4.12) is in tro d u c e d as a m a tte r o f co n v en ien ce to allow us to ex p ress th e d ifferen ce e q u a tio n in (2.4.13) w ith o u t any n eg ativ e signs. E q u a tio n (2.4.12) ex p re sse s th e o u tp u t of the system at tim e rt d irectly as a w eig h ted sum o f p ast o u tp u ts \ {n - 1), y( n - 2 ).........y(/i - N ) as well as past an d p re se n t in p u t signals sam p les. W e o b se rv e th a t in o rd e r to d e te rm in e y( n) fo r n > 0, we n e e d th e in p u t x ( n ) for all n > 0, an d th e initial c o n d itio n s y ( 1), y ( 2), y ( N ) . In o th e r w o rd s, th e initial c o n d itio n s su m m a riz e all th a t we n eed to k n o w a b o u t th e p a s t h isto ry o f th e re sp o n se o f th e sy stem to c o m p u te th e p re se n t an d fu tu re o u tp u ts. T h e g e n e ra l so lu tio n o f th e /V -order c o n stan tco efficien t d ifferen ce e q u a tio n is c o n sid e re d in th e follow ing su b se c tio n . A t this p o in t we r e s ta te th e p ro p e rtie s of lin earity , tim e in v a ria n c e , an d stab ility in th e c o n te x t o f recu rsiv e system s d e sc rib e d by lin ear c o n s ta n t-c o e ffic ie n t d ifferen ce e q u a tio n s. A s we h av e o b se rv ed , a recu rsiv e system m a y b e relax ed o r n o n re la x e d , d e p e n d in g o n th e initial co n d itio n s. H e n c e th e d e fin itio n s o f th ese p ro p e rtie s m u st ta k e in to a c c o u n t th e p re se n c e o f the initial co n d itio n s. W e begin w ith th e d e fin itio n o f lin earity . A system is lin e a r if it satisfies th e follow ing th re e re q u ire m e n ts: 1. T h e to tal re sp o n se is e q u a l to th e sum o f th e z e ro -in p u t a n d z e ro -sta te r e sp o n ses [i.e.. y ( n) = y 7 .\(n) + yzs(n)]. 2. T h e p rin cip le o f su p e rp o s itio n ap p lies to th e z e ro -sta te re s p o n s e ( zero-state linear). 3. T h e p rin cip le o f su p e rp o s itio n ap p lies to the z e ro -in p u t re s p o n s e ( zero- i nput linear). A system th a t d o es n o t satisfy all three se p a ra te r e q u ire m e n ts is by d efin itio n n o n lin ear. O b v io u sly , fo r a re lax ed system , yZj(n) = 0, an d th u s re q u ire m e n t 2, w hich is th e d efin itio n o f lin e a rity given in S ection 2.2.4, is sufficient. W e illu strate th e a p p lic a tio n o f th ese re q u ire m e n ts by a sim p le ex am p le. Example 2.4.2 D eterm ine if the recursive system defined by the difference equation v(n) = av(n 1)4- x(n) is linear. Solution as By combining (2.4.9) and (2.4.11), we obtain (2.4.8), which can be expressed

y(n) = yzi(n) + y*(n) Thus the first requirem ent for linearity is satisfied.

Sec. 2.4

Discrete-Time Systems Described by Difference Equations

99 +

To check for the second requirem ent, let us assume that ,t(n) = Then (2.4.9) gives

*= {}

= ('iy ^ (n) + C2V^'(n) Hence y(/i> satisfies the principle of superposition, and thus the system is zero-state linear. Now let us assume that y ( 1) = q vj(1) 4- f 2y;( 1). From (2.4.11) we obtain 1) 4- C;V;(1)] = vi (1)4v; ( 1)

= C i v'i '(n) + r ;y ^(/i) Hence the system is zero-input linear. Since the system satisfies all three conditions for linearity, it is linear. A lth o u g h it is so m e w h at ted io u s, the p ro c e d u re used in E x am p le 2.4,2 to d e m o n s tra te lin e a rity for th e system d escrib ed by th e first-o rd e r d ifferen ce e q u a tio n , ca rrie s o v er d irectly to th e g en eral recu rsiv e system s d e sc rib e d by th e c o n stan tcoefficient d ifferen ce e q u a tio n given in (2.4.13). H en ce , a recu rsiv e system d escrib ed by th e lin ear d iffe re n c e e q u a tio n in (2.4.13) also satisfies all th re e c o n d itio n s in th e d efin itio n o f lin earity , a n d th e re fo re it is linear. T h e n ex t q u e s tio n th a t arises is w h e th e r o r n o t the cau sal lin ear system d e sc rib e d by th e lin e a r c o n stan t-co efficien t differen ce e q u a tio n in (2.4.13) is tim e in v arian t. T h is is fairly easy, w h en d e alin g w ith system s d e sc rib e d by explicit i n p u t-o u tp u t m a th e m a tic a l re la tio n sh ip s. C learly, th e system d e sc rib e d by (2.4.13) is tim e in v a ria n t b ecau se th e coefficients ak an d bk are c o n stan ts. O n th e o th e r h a n d , if o n e o r m o re o f th e s e coefficients d e p e n d s on tim e, th e system is tim e v a ria n t, since its p ro p e rtie s ch an g e as a fu n ctio n of tim e. T h u s w e co n clu d e th at the recursive sy s t em descri bed b y a linear constant-coefficient difference equat i on is linear a n d t ime invariant. T h e final issu e is th e sta b ility of th e recursive system d e sc rib e d by th e lin ear, c o n stan t-co efficien t d iffe re n c e e q u a tio n in (2.4.13). In S ection 2.3.6 w e in tro d u c e d th e c o n c e p t o f b o u n d e d in p u t-b o u n d e d o u tp u t (B IB O ) sta b ility fo r re la x e d sys tem s. F o r n o n re la x e d system s th a t m ay b e n o n lin e a r, B IB O sta b ility sh o u ld be view ed w ith so m e care. H o w e v e r, in th e case o f a lin e a r tim e -in v a ria n t recursive system d e sc rib e d by th e lin e a r co n stan t-co efficien t d iffe re n c e e q u a tio n in (2.4.13), it suffices to s ta te th a t such a system is B IB O sta b le if a n d only if fo r every b o u n d e d in p u t a n d ev ery b o u n d e d initial c o n d itio n , th e to ta l system re sp o n se is bounded.

100 Example 2.43

Discrete-Time Signals and Systems

Chap. 2

Determ ine if the linear tim e-invariant recursive system described by the difference equation given in (2.4.7) is stable. Solution i*(n)l < Let us assume that the input signal x(n) is bounded in amplitude, that is, < oc for all n > 0. From (2.4.8) we have j v{n J | < |a',+l_y(1)| + ^ akx(n - k) , n>0

If n is finite, the bound M v is finite and the output is bounded independently of the value of a. However, as n -* oo, the bound My remains finite only if |aj < 1 because |a |B - 0 as n -* oc. Then M y = Ms j(\ - |o|). Thus the system is stable only if \a\ < 1. For the sim ple first-order system in E xam ple 2.4,3, w e w ere able to express the con d ition for B IB O stability in term s o f the system param eter a. nam ely \a\ < 1. W e should stress, how ever, that this task b ecom es m ore difficult for higher-order system s. F ortunately, as we shall see in su bsequent chapters, other sim ple and m ore efficient tech n iqu es exist for investigating the stability o f recursive system s.

2.4.3 Solution of Linear Constant-Coefficient Difference Equations

G iven a linear con stan t-coefficien t d ifferen ce eq u ation as the in p u t-o u tp u t rela tionship describing a linear tim e-invariant system , our objective in this subsection is to determ ine an explicit exp ression for the output y{n). T h e m eth od that is d ev elo p ed is term ed the direct m e t h o d . A n alternative m eth od based on the ztransform is described in Chapter 3. For reasons that will b eco m e apparent later, the z-transform approach is called the indirect m e th o d . Basically, the goal is to determ ine the output y(n ), n > 0, o f the system given a specific input x ( n ), n > 0, and a set o f initial con d ition s. T h e direct solution m ethod assum es that the total solu tion is the sum o f tw o parts: y ( n ) = y h(n ) + }'r (n) T he part y/,(n) is know n as the h o m o g e n e o u s or c o m p l e m e n ta r y solu tion , w hereas yp(n) is called the particular solution.

The homogeneous solution of a difference equation. W e begin the problem o f solving the linear con stan t-coefficien t d ifferen ce eq u ation given by

Sec. 2.4

Discrete-Time Systems Described by Difference Equations

101

(2.4.13) bv assu m in g th a t th e in p u t x (n ) = 0. T h u s w e will first o b ta in th e so lu tio n to th e h o m o g e n e o u s di ff erence equati on \ J 2 a ky ( n - k ) = 0 *=(> (2.4.14)

T h e p ro c e d u re fo r solving a lin e a r c o n s tan t-co efficien t d ifferen ce e q u a tio n d irectly is very sim ilar to th e p ro c e d u re fo r solving a iin e a r c o n stan t-co efficien t d iffe re n tia l e q u a tio n . B asically, we assum e th a t th e so lu tio n is in th e fo rm o f an e x p o n e n tia l, th a t is. yi,{n) = a" (2.4.15)

w h ere th e su b scrip t h o n \ {n) is used to d e n o te th e so lu tio n to th e h o m o g e n e o u s d ifferen ce e q u a tio n . If w e su b stitu te this assu m ed so lu tio n in (2.4.14), w e o b tain th e p o ly n o m ial e q u a tio n =0 fc=U or k" A (k* + ci)k^ ! + a2k^ * + - +
q n \ k

+ aw) = 0

(2.4.16)

T h e p o ly n o m ial in p a re n th e se s is called th e characteristic p o l y n o m i a l o f the system . In g e n e ra l, it h as N roots, w hich we d e n o te as Xi. k 2 ......... k ^ . T h e ro o ts can b e real o r co m p lex v alued. In p ra c tic e the co efficien ts a \ , a 2 ........ are usually re a l. C o m p le x -v a lu e d ro o ts o ccu r as c o m p le x -c o n ju g a te p airs. S om e o f th e N ro o ts m ay be id en tical, in w hich case w e have m u ltip Je -o rd e r ro o ts. F o r th e m o m e n t, let us assum e th a t th e ro o ts a re d istin ct, th a t is, th e re are n o m u ltip le -o rd e r ro o ts. T h e n th e m o st g e n eral so lu tio n to th e h o m o g e n e o u s d iffe re n c e e q u a tio n in (2.4.14) is yh(n) = C \ k \ + C2k 2 + + C^ k ^ , (2.4.17)

w h ere C i. C 2.........C s are w eig h tin g coefficients. T h e se co efficien ts are d e te rm in e d fro m th e in itial c o n d itio n s specified fo r th e sy stem . Since th e in p u t jr(n) = 0. (2.4.17) can b e u sed to o b ta in th e z ero- i nput response o f th e system . T h e follow ing ex am p les illu stra te th e p ro c e d u re . Example 2.4.4 D eterm ine the hom ogeneous solution of the system described by the first-order dif ference equation _v(n) + a\y(rt 1) = x(n) Solution The assumed solution obtained by setting x(n) = 0 is y*(n) = V (2.4.18)

102

D iscrete-Time Signals and System s

Chap. 2

When we substitute this solution in (2.4.18), we obtain [with x(n) = 0] X -f-1 2 ]An ^ = 0 A" *(A + ai) = 0 A= O) Therefore, the solution to the hom ogeneous difference equation is = CA" = C ( - a ,)" (2.4.19)

The zero-input response of the system can be determ ined from (2.4,18) and (2.4.19). With x(n) = 0, (2.4.18) yields v(0) =

-a, v ( - l )

On the other hand, from (2.4,19) we have v*(0) = C and hence the zero-input response of the system is Vii() = ( i)n+1 v{1) n >0 (2.4.20)

With a = ai, this result is consistent with (2.4,11) for the first-order system, which was obtained earlier by iteration of the difference equation.

Example 2AS
Determ ine the zero-input response of the system described by the homogeneous second-order difference equation y() - 3 y(n - 1) - 4y(n - 2) = 0 the solution to be the exponential yh(n ) = X " U pon substitution of this solution into (2.4.21), we obtain the characteristic equation
a"

(2.4.21)

Solution First we determ ine the solution to the homogeneous equation. We assume

- 3xn- ' - 4 r ~ 2 = o

A"-2(X2 3A 4) = 0 Therefore, the roots are X = - 1 , 4, and the general form of the solution to the homogeneous equation is }7i(n) = CjX" + C 2X2 = C1( - i r + C2(4)" The zero-input response of the system can be obtained from the homogenous solution by evaluating the constants in (2.4.22), given the initial conditions y (1) and y ( ~ 2). From the difference equation in (2.4.21) we have y(0) = 3y(1) + 4y( 2) y(l) = 3v(0) + 4_y(1) = 3 [3 y (-l)+ 4y(-2)] + 4y(-l) = 13y(1) + l 2y( 2) (2.4.22)

Sec. 2.4

Discrete-Time Systems Described by Difference Equations

103

On the other hand, from (2.4.22) we obtain v(0> = C) + C;

V ( 1 I = -C l +4C;
By equating these two sets of relations, we have C i + C ; = 3_v( 1) + 4y( 2) - C , + 4 C ; = 13 v{ 1) + 12 y( 2) The solution of these two equations is Ci = 7 v( 1 ) + ? v ( - 2 ) C; = X .V (-l) + T.v( -2 ) Therefore, the zero-input response of the system is

v,i(n) = [ 5 v( 1 >- i v(-2)](-l)''

(2.4.23) + [ x . v ( - l + t.v(-2)](4>" n >0 For example, if v{ 2) = 0 and y ( - l ) = 5. then C| = 1, C: = 16. and hence y-,i(H) = ( -1 ) " * 1 + (4)"+: n >0

T h e se e x am p les illu strate the m e th o d fo r o b ta in in g the h o m o g e n e o u s so lu tio n and th e z e ro -in p u t re sp o n se o f th e system w hen the ch a ra c te ristic e q u a tio n co n tain s d istin ct ro o ts. O n th e o th e r h an d , if th e c h a ra c te ristic e q u a tio n co n ta in s m u ltip le ro o ts, th e form o f th e so lu tio n given in (2.4.17) m ust be m odified. F o r ex am p le, if ai is a ro o t o f m u ltip licity m, th en (2.4.17) b eco m es

\ h(ii) = Ci A .'.' + C2/iX! + C v;:a',' + + Cm> 7 n'~lX'!

(2.4.24) + Cm+\)"m+\ + + C^Kl

The particular solution of the difference equation. T h e p a rtic u la r so lu tio n \ p (n) is re q u ire d to satisfy th e d ifferen ce e q u a tio n (2.4.13) fo r th e specific in p u t signal x (n ). n > 0. In o th e r w ords, y p(n) is any so lu tio n satisfying
N M

' Y ^ a ky p (n - k) ' Y ^ b kx ( n - k) A-0 *=0

an 1

(2.4.25)

T o solve (2.4.25). w e assu m e fo r yp (n), a fo rm th at d e p e n d s on th e fo rm o f th e in p u t j:(h ). T h e fo llow ing ex am p le illu strates the p ro c e d u re . Example 2.4.6 D eterm ine the particular solution of the first-order difference equation y(n) + fliy(n - 1) = x(n). | f l i | <l (2,4.26)

when the input x(n) is a unit step sequence, that is.

x(n) = u (n )

104

Discrete-Time Signals and System s

Chap. 2

Solution Since the input sequence x(n) is a constant for n > 0. the form of the solu tion that we assume is also a constant. Hence the assumed solution of the difference equation to the forcing function x(n), called the particular solution of the difference equation, is vr (n) = Ku(n) where A - is a scale factor determined so that (2.4.26) is satisfied. U pon substitution of this assumed solution into (2.4.26). we obtain
Ku i n ) + d]Ku{,n 1) = u ( n )

To determine K, we must evaluate this equation for any n > 1. where none of the terms vanish. Thus
K -t- O ]K 1

1 K = -------l+^i Therefore, the particular solution to the difference equation is v (n) = -------- u(n) ' 1 +fii (2.4.27)

In th is ex am p le, th e in p u t * ( ). n > 0. is a c o n s ta n t an d th e fo rm assu m ed for th e p a rtic u la r so lu tio n is also a c o n sta n t. If x{n) is an e x p o n e n tia l, w e w ould assu m e th a t th e p a rtic u ia r so lu tio n is also an e x p o n e n tia l. If x {n) w e re a sin u so id , th e n >/,() w o uld also be a sinusoid. T h u s o u r assu m ed fo rm fo r th e p a rtic u la r so lu tio n tak es th e basic fo rm of th e signal x{n). T a b le 2.1 p ro v id e s th e g en eral fo rm o f th e p a rtic u la r so lu tio n for sev eral ty p es of excitatio n . Exam ple 2.4.7 Determ ine the particuiar solution of the difference equation v(n) = | y(n - 1) - v(w - 2) + x(n) when the forcing function x(n) = 2". n > 0 and zero elsewhere.
TABLE 2.1 GENERAL FORM OF THE PARTICULAR SOLUTION FOR SEVERAL TYPES OF INPUT SIGNALS

Input Signal, x(n) A (constant) AM" AnM

Particular Solution, vr (n) K KM" Kan + K ^ " - ' 1 + . . . + KM

AnnM

An(K i)nM+ K\ttM 1

A cos W f> n A sin a>on

Kj cos toon -I- K2 sin won

Sec. 2.4

Discrete-Time Systems Described by Difference Equations

105

Solution

The form of the particular solution is yp (n ) = K2n n >0

Upon substitution of yP(n) into the difference equation, we obtain

K2"u( n) =
-

1)

- I K 2 n- 2u(n - 2 ) + 2 " ( )

To determ ine the value of K, we can evaluate this equation for any n > 2, where none of the terms vanish. Thus we obtain 4A " = \ ( 2K) - i f f + 4 and hence K = Therefore, the particular solution is yn(n) = ^2" n > 0

W e h av e n o w d e m o n s tra te d how to d e te rm in e th e tw o c o m p o n e n ts o f the so lu tio n to a d ifferen ce e q u a tio n w ith c o n s ta n t coefficients. T h e se tw o c o m p o n e n ts are th e h o m o g e n e o u s so lu tio n and th e p a rtic u la r so lu tio n . F ro m th e s e tw o co m p o n e n ts, we c o n s tru c t the to tal so lu tio n fro m w hich w e can o b ta in th e ze ro -sta te re sp o n se .

The total solution of the difference equation. T h e lin e a rity p ro p e rty o f th e lin ear c o n stan t-co efficien t d ifferen ce e q u a tio n allow s u s to a d d th e h o m o g e n e o u s so lu tio n an d the p a rtic u la r so lu tio n in o rd e r to o b ta in th e total solution. T h u s

v() =

y h(n)

+ \ p( n)

T h e r e s u lta n t sum v(n) co n tain s th e c o n s ta n t p a r a m e te r s {C/} e m b o d ie d in th e h o m o g e n e o u s so lu tio n c o m p o n e n t >v,(n). T h e se c o n s ta n ts can be d e te rm in e d to satisfy th e in itial co n d itio n s. T h e follow ing ex a m p le illu stra te s th e p ro c e d u re , Exam ple 2.4.8 D eterm ine the total solution y(/i), n > 0, to the difference equation y(n) +a i y ( n - 1) = x(n) (2.4.28)

when x( n) is a unit step sequence [i.e., x(n) = ()] and y (1) is the initial condition. Solution From (2.4,19) of Example 2.4.4, the hom ogeneous solution is

y*(n) = C(-fli)"
and from (2.4.26) of Example 2.4.6, the particular solution is

Consequently, the total solution is y(n) = C ( - a i)" + - 1 + fli n >0 (2.4.29)

where the constant C is determ ined to satisfy the initial condition y ( - l ) .

106

Discrete-Time Signals and Systems

Chap. 2

In particular, suppose that we wish to obtain the zero-siate response of the system described by the first-order difference equation in (2.4.28). T hen we set y( 1) = 0. To evaluate C. we evaluate (2.4.28) at n = 0 obtaining
y ( 0 ) + Oi y ( 1 ) = 1

y(0) = 1 On the other hand, (2.4.29) evaluated at n = 0 yields v(0) = C 4- i 1 + ax Consequently. 1 C + -------- = 1 1 + i

ai 1 + ai

Substitution for C into (2.4.29) yields the zero-state response of the system l-f-fli)" * ' V i*(/i) = ---- ----- 1 + i n >0

If we evaluate the param eter C in (2,4.29) under the condition that y( 1) ^ 0. the total solution will include the zero-input response as well as the zero-state response of the system. In this case (2.4.28) yields y(0) + a]V( 1) = 1
y (0) = fliy( 1) 4- 1

On the other hand. (2.4.29) yields ] 1+ U \ By equating these two relations, we obtain C + ^ = ai v(1) 4- 1
1 4 - a,

C = -g ] v( 1) +

-
1 4- a ]

Finally, if we substitute this value of C into (2.4.29). we obtain y(n) = ( a , r +1y(1) + -

= + Vzs( n)

^ 1 + a)

n >0

(2.4.30)

W e o b se rv e th a t th e system re sp o n se as given by (2.4.30) is c o n s iste n t w ith th e re sp o n se y ( n) given in (2.4.8) for th e first-o rd e r system (w ith a = - o i ) . w hich was o b ta in e d by solv in g th e d ifferen ce e q u a tio n iterativ e ly . F u rth e rm o r e , we n o te th a t th e v alu e o f th e c o n s ta n t C d e p e n d s b o th o n th e initial co n d itio n y (1) an d o n th e ex citatio n fu n ctio n . C o n se q u e n tly , th e v alue o f C in flu en ces b o th th e zeroin p u t re sp o n se a n d th e z e ro -sta te re sp o n se . O n th e o th e r h a n d , if w e w ish to

Sec. 2.4

Discrete-Time Systems Described by Difference Equations

107

obtain the zero-state resp onse only, w e sim ply solve for C under th e con d ition that y ( - l ) = 0, as d em onstrated in E xam ple 2.4.8. W e further ob serve that the particular solution to the d ifferen ce equation can b e o b tain ed from the zero-state resp onse o f the system . In d eed , if |o]| < 1, which is the con d ition for stability o f the system , as w ill be show n in S ection 2.4.4, the lim iting valu e o f >zs() as n approaches infinity, is the particular solu tion , that is, 1 ( n ) = lim >zs() = --------n-*Oo 1+ai Since this com p onent o f the system response d o es not go to zero as n approaches infinity, it is usually called the steady-state re spons e o f the system . T his response persists as lon g as the input persists. T he com p onent that d ies out as n approaches infinity is called the transient re sponse o f the system .

Example 2.4.9
D eterm ine the response y(n), n > 0, of the system described by the second-order difference equation v(n) - 3 y(n - 1) - 4v(n - 2) = x( n ) + 2x(n - 1) when the input sequence is x(n) =4"u(n) (2.4.31)

Solution We have already determ ined the solution to the homogeneous difference
equation for this system in Example 2.4.5. From (2.4.22) we have y*(n) = C , ( - l ) n + C 2(4)n (2.4.32)

The particular solution to (2.4.31) is assumed to be an exponential sequence of the same form as x(n). Normally, we could assume a solution of the form yp(n) = K(4,)au(n) However, we observe that ( n ) is already contained in the hom ogeneous solution, so that this particular solution is redundant. Instead, we select the particular solution to be linearly independent of the terms contained in the homogeneous solution. In fact, we treat this situation in the same manner as we have already treated multiple roots in the characteristic equation. Thus we assume that yp(n) = Kn{4)"u(n) Upon substitution of (2.4.33) into (2.4.31), we obtain Kn(4)"u(n) - 3 K ( n - l)(4 ),' - 1u(n - 1) - 4 K{n - 2)(4)n- 2u(n - 2) = (4)"u(n) + 2(4)"-1 m(/i 1) To determ ine K , we evaluate this equation for any n > 2, where none of the unit step term s vanish. To simplify the arithmetic, we select n = 2, from which we obtain K = | . Therefore, (2.4.33)

yP{n) |rt(4)n(/l)

(2.4.34)

108

Discrete-Time Signals and System s

Chap. 2

The total solution to the difference equation is obtained by adding (2.4.32) to (2.4.34). Thus

y() C ](1)" + C;(4)n + |w(4)"

n > 0

(2.4.35)

where the constants C \ and C2 are determined such that the initial conditions are satisfied. To accomplish this, we return to (2.4.31), from which we obtain v(0) = 3 y (- l) + 4y (2) + 1 y (1) = 3v(0) + 4_v(-l) + 6 = 1 3 y (- l) + 12y(-2) + 9 O n the other hand, (2.4.35) evaluated at n = 0 and n = 1 yields y(0) = Ci + C2 y (l) = -C] + 4C2 + f W e can now equate these two sets of relations to obtain C \ and C 2. In so doing, we have the response due to initial conditions y (1) and y (2) (the zero-input response), and the zero-state or forced response. Since we have already solved for the zero-input response in Exam ple 2.4.5. we can simplify the computations above by setting v (1) = y (2) = 0. Then we have C, + C; = 1 - C l + 4 C2 + f

=9

Hence C : = ^ and C 2 = Finally, we have the zero-statc response to the forcing function = (4)"(n) in the form

v B (n )=

+5< 4 )" +H 4)"

nZ0

(2.4.36)

The total response of the system, which includes the response to arbitrary initial conditions, is the sum of (2.4.23) and (2.4.36).

2.4.4 The Impulse Response of a Linear Time-Invariant Recursive System

T h e im p u lse re sp o n se o f a lin e a r tim e -in v a ria n t sy stem w as p re v io u sly defin ed as th e re sp o n se o f th e sy stem to a u n it sa m p le e x citatio n [i.e., x (n ) = <5(n)]. In th e case o f a recu rsiv e sy stem , h ( n ) is sim ply e q u a l to th e z e ro -s ta te re sp o n s e o f the system w h en th e in p u t j:(n ) = <5(n) an d th e system is initially re la x e d . F o r ex am p le, in th e sim ple first-o rd e r re c u rsiv e system g iv en in (2.4.7), th e z e ro -sta te re sp o n se g iv en in (2.4.8), is n

(2.4.37)
*=o W ith jr(n) = i ( ) is substituted into (2.4.37), we obtain n yzs(n) = Y 2 a k8(n - k)
= an n > 0

Sec. 2.4

Discrete-Time Systems Described by Difference Equations

109

H e n c e th e im p u lse re sp o n se o f th e first-o rd e r rec u rsiv e system d e sc rib e d by (2.4.7) is h(n) = a nu(n) (2.4.38)

as in d ic a te d in S ectio n 2.4.2. In th e g e n e ra l case o f an a rb itra ry , lin e a r tim e -in v a ria n t rec u rsiv e system , the z e ro -s ta te re sp o n s e ex p re sse d in te rm s o f th e co n v o lu tio n su m m a tio n is > zs(n) ^ 2 h ( k ) x ( n k)
i=0

n > 0

(2.4.39)

W h e n th e in p u t is an im p u lse [i.e., * ( ) = <5()], (2.4.39) re d u c e s to Vzs(n) = h(n) (2.4.40)

N o w , let us c o n s id e r th e p ro b le m o f d e te rm in in g th e im p u lse re sp o n se h i n ) given a lin e a r c o n s tan t-co efficien t d ifferen ce e q u a tio n d e sc rip tio n o f th e system . In term s o f o u r d isc u ssio n in th e p re c e d in g su b sectio n , we h av e e s ta b lis h e d th e fact th a t the to ta l re sp o n s e o f th e sy stem to any e x citatio n fu n ctio n co n sists o f th e sum o f tw o so lu tio n s o f th e d ifferen ce e q u atio n : th e so lu tio n to the h o m o g e n e o u s e q u atio n p lu s th e p a rtic u la r so lu tio n to th e e x c itatio n fu n ctio n . In th e case w h ere the exci ta tio n is an im p u lse, th e p a rtic u la r so lu tio n is z e ro , since x ( n ) = 0 for n > 0. th at is. yp (n) = 0 C o n s e q u e n tly , th e re sp o n se o f th e system to an im p u lse co n sists only o f th e so lu tio n to th e h o m o g e n e o u s e q u a tio n , w ith the ( Q ) p a r a m e te r s e v a lu a te d to satisfy th e in itial c o n d itio n s d ic ta te d by th e im pulse. T h e follow ing ex am p le illu strates th e p ro c e d u re fo r o b ta in in g h(n) given th e d ifferen ce e q u a tio n fo r the system . Example 2.4.10 D eterm ine the impulse response h(n) for the system described by the second-order difference equation
y(n ) 3v(n 1) 4 y ( 2) = x ( n ) + 2 x i n 1) (2.4.41)

Solution W e have already determ ined in Example 2.4.5 that the solution to the hom ogeneous difference equation for this system is ^ ( n ) = C, (-1 )" + C2(4)"
n > 0

(2.4.42)

Since the particular solution is zero when x(n) = 6(n), the impulse response of the sys tem is simply given by (2.4.42), where C] and C2 must be evaluated to satisfy (2.4.41). For n = 0 and n = 1, (2.4.41) yields v(0) = 1 y (1) = 3 y (0 )+ 2 = 5 where we have imposed the conditions y ( 1) = y ( 2) = 0. since the system must be relaxed. On the other hand, (2.4.42) evaluated at n = 0 and n = 1 yields y (0) = C, + C2

y (1) = Ci + 4C2

110

Discrete-Time Signals and Systems

Chap. 2

By solving these two sets of equations for C] and C2, we obtain = C: = 5

Therefore, the impulse response of the system is h{n) = [i ( 1)" + f (4)-}u(n) W e m ak e th e o b se rv a tio n th a t b o th th e sim ple firs t-o rd e r recu rsiv e system an d th e se c o n d -o rd e r recu rsiv e system h av e im pulse re sp o n se s th a t a re infinite in d u ra tio n . In o th e r w o rds, b o th o f th e s e recu rsiv e system s a re IIR system s. In fact, d u e to th e recu rsiv e n a tu re of th e system , an y recu rsiv e sy stem d e sc rib e d by a lin e a r co n stan t-co efficien t d iffe re n c e e q u a tio n is an I IR system . T h e co n v erse is n o t tru e, h o w ev er. T h a t is, n o t ev ery lin e a r tim e -in v a ria n t I I R sy stem can be d e sc rib e d by a lin e a r c o n s tan t-co efficien t d ifferen ce e q u a tio n . In o th e r w ords, recu rsiv e sy stem s d e sc rib e d by lin ear co n sta n t-c o e ffic ie n t d iffe re n c e e q u a tio n s are a su bclass o f lin e a r tim e -in v a ria n t I I R system s. T h e ex ten sio n o f th e a p p ro a c h th a t w e h av e d e m o n s tra te d fo r d e te rm in ing th e im p u lse re sp o n se o f th e first- a n d s e c o n d -o rd e r system s, g en e ra liz e s in a stra ig h tfo rw a rd m a n n e r. W h e n th e system is d e sc rib e d by an A 'th -o rd e r lin e a r d ifferen ce e q u a tio n o f th e ty p e given in (2.4.13), th e so lu tio n o f th e h o m o g e n e o u s e q u a tio n is

*=i w h en th e ro o ts {a*} o f th e c h a ra c te ristic p o ly n o m ial are d istin ct. H e n c e th e im pulse re sp o n se o f th e sy stem is id en tical in fo rm , th a t is. (2.4.43) w h ere th e p a ra m e te rs {Ctl are d e te rm in e d by se ttin g th e initial c o n d itio n s v ( 1) = . . . = y ( - N ) = 0. T h is fo rm o f h{n) allow s us to easily re la te th e stab ility of a system , d escrib ed by an N th -o rd e r d iffe re n c e e q u a tio n , to th e values o f th e ro o ts o f th e ch a ra c te ristic p o ly n o m ial. In d e e d , since B IB O sta b ility re q u ire s th a t th e im p u lse re sp o n s e be ab so lu te ly su m m ab le, th e n , fo r a causal system , w e h av e
x oc N N oo oo > (* )! = Y ^ C kXn k < |C * | |A * |" =0 n =0 n=0 Ik=-l k=\ n=0

N ow if |

j < 1 fo r all k, th e n

an d h en ce
OC

IM*)[ < oo

Sec. 2.5

im plementation of Discrete-Time Systems

111

O n th e o th e r h an d , if o n e o r m o re o f th e > 1, h{n) is n o lo n g e r ab so lu te ly su m m ab le, an d c o n se q u e n tly , th e sy stem is u n sta b le . T h e re fo re , a n ecessa ry an d sufficient c o n d itio n fo r th e sta b ility o f a causal IIR system d e sc rib e d by a lin ear c o n s tan t-co efficien t d iffe re n c e e q u a tio n , is th a t al! ro o ts o f th e c h a ra c te ristic p o ly n o m ial b e less th a n u n ity in m a g n itu d e. T h e re a d e r m ay verify th a t th is c o n d itio n ca rrie s o v er to th e case w h ere th e system h as ro o ts o f m u ltip lic ity m.

2.5 IMPLEMENTATION OF DISCRETE-TIME SYSTEMS

O u r tre a tm e n t o f d isc re te -tim e sy stem s h as b e e n fo cu sed on th e tim e -d o m a in c h a r ac te riz a tio n an d analysis o f lin ear tim e -in v a ria n t system s d esc rib e d by c o n stan tco efficien t lin e a r d ifferen ce e q u a tio n s. A d d itio n a l an aly tical m e th o d s a re d e v e l o p e d in th e n ex t tw o c h a p te rs, w h ere we c h a ra c te riz e an d an aly ze L TI sy stem s in th e fre q u e n c y d o m ain . T w o o th e r im p o rta n t to p ics th a t will b e tr e a te d la te r are th e d esig n an d im p le m e n ta tio n o f th e se system s. In p ractice, system design a n d im p le m e n ta tio n a re usually tr e a te d jo in tly r a th e r th a n se p a ra te ly . O fte n , th e system design is d riv en by th e m e th o d o f im p le m e n ta tio n an d by im p le m e n ta tio n c o n stra in ts, such as cost, h a rd w a re lim itatio n s, size lim ita tio n s, a n d p o w e r re q u ire m e n ts . A t th is p o in t, we h av e n o t as y e t d e v e lo p e d th e n ecessa ry analysis an d design tools to tr e a t such com plex issues. H o w ev er, we h ave d e v e lo p e d sufficient b a c k g ro u n d to c o n sid e r so m e b a sic im p le m e n ta tio n m e th o d s for re a liz a tio n s o f L TI sy stem s d escrib ed by lin ear c o n s tan t-co efficien t d iffe re n c e e q u atio n s.

2.5.1 Structures for the Realization of Linear Time-Invariant Systems

In th is su b sectio n we d esc rib e s tru c tu re s fo r th e re a liz a tio n o f system s d escrib ed by lin e a r c o n s tan t-co efficien t d ifferen ce e q u a tio n s. A d d itio n a l stru c tu re s fo r th ese sy stem s a re in tro d u c e d in C h a p te r 7. A s a b eg in n in g , let us c o n sid e r th e first-o rd e r system y(n) = a iy (n 1) + box ( n ) + b \ x ( n - 1) (2.5.1)

w hich is re a liz e d as in Fig. 2.32a. T h is re a liz a tio n uses s e p a ra te delays (m em o ry ) fo r b o th th e in p u t a n d o u tp u t signal sa m p le s a n d it is called a direct f o r m I structure. N o te th a t th is sy stem can b e view ed as tw o lin e a r tim e -in v a ria n t system s in cascad e. T h e first is a n o n re c u rsiv e , sy stem d e sc rib e d by th e e q u a tio n
u(n) = Zoj;(n) + 6 j j ( n 1) (2.5.2)

w h e re a s th e se c o n d is a rec u rsiv e system d e sc rib e d by th e e q u a tio n y ( n ) = - a i y ( n - 1) + t>{) (2.5.3)

H o w e v e r, as we h a v e se e n in S ectio n 2.3.4, if w e in te rc h a n g e th e o r d e r o f the c a s cad ed lin e a r tim e -in v a ria n t system s, th e o v erall system re sp o n s e re m a in s th e

112

Discrete-Time Signals and Systems

Chap. 2

sam e. T h u s if w e in te rc h a n g e the o rd e r o f the recu rsiv e an d n o n re c u rsiv e system s, we o b ta in an a lte rn a tiv e s tru c tu re fo r th e re a liz a tio n o f th e sv stem d e sc rib e d by (2.5.1). T h e re su ltin g system is show n in Fig. 2.32b. F ro m th is figure we o b tain th e tw o d ifferen ce e q u a tio n s w( n) fl] w( n 1) + jt() v(n) = t>nw(n) + b\ w{ n - 1) (2-5.4) (2.5.5)

w hich p ro v id e an a lte rn a tiv e a lg o rith m for co m p u tin g th e o u tp u t o f the system d e sc rib e d by th e single d ifferen ce e q u a tio n given in (2.5.1). In o th e r w o rd s, the tw o d ifferen ce e q u a tio n s (2.5.4) an d (2.5.5) are e q u iv a le n t to th e single differen ce e q u a tio n (2.5.1). A close o b se rv a tio n o f Fig. 2.32 re v e a ls th at th e tw o d elay e le m e n ts co n tain th e sam e in p u t w( n) an d h en c e the sam e o u tp u t w( n 1). C o n s e q u e n tly , these tw o e le m e n ts can be m e rg ed in to o n e delay, as show n in Fig. 2.32c. In c o n tra st
x(n)
-----------------~

b{ ,

r(n)

vim

(a)

(b.)

(c)

Figure 132 Steps in converting from the direct form I realization in (a) to the direct form II realization in (c).

Sec. 2.5

Implementation of Discrete-Time Systems

113

to th e d ire c t fo rm I s tru c tu re , th is new re a liz a tio n re q u ire s o n ly o n e d elay fo r th e au x iliary q u a n tity ui(n), an d h en c e it is m o re efficient in te rm s o f m em o ry re q u ire m e n ts . It is called th e direct f o r m I I structure a n d it is u sed ex ten siv ely in p ra c tic a l a p p licatio n s. T h e se stru c tu re s can read ily be g e n e ra liz e d fo r th e g e n e ra l lin e a r tim ein v a ria n t recu rsiv e sy stem d escrib ed by th e d iffe re n c e e q u a tio n
N M

y (n ) = - y ^ a ^ .y ( - k) +
*=l k=0

- k)

(2.5.6)

F ig u re 2.33 illu stra te s th e d ire c t fo rm I s tru c tu re fo r th is sy stem . T h is stru c tu re re q u ire s M + N d elay s a n d N + M + 1 m u ltip lic atio n s. I t c a n be v iew ed as the c ascad e o f a n o n re c u rsiv e system M i;() = Y bkx{n - k)
i=U

(2.5.7)

a n d a recu rsiv e system

s

y(n) = -

v(n ~ k) + v(n)

(2.5.8)

By rev ersin g th e o r d e r o f th ese tw o system s as was p rev io u sly d o n e fo r th e first-o rd e r sy stem , w e o b ta in th e d ire c t form II s tru c tu re sh o w n in Fig. 2.34 fo r

Figure 1 3 3

Direct form I structure of the system described by (2.5.6).

114

Discrete-Time Signals and Systems

w(n) b0 N

Chap. 2

y(n )

-1 U J{ /1 - 1) J -1
w(n - 2) b2

w(n - 3)

by

Figure 2 3 4

Direct form II structure for the system described by (2.5.6).

N > M. T h is stru c tu re is the cascad e o f a recu rsiv e system

N

w( n) = ^

ai W(n - k ) -f x( n)

(2.5.9)

*=1
fo llo w ed by a n o n re c u rsiv e sy stem M v(n) = vu(n - k ) (2.5.10)

W e o b serv e th a t if N > M . this s tru c tu re re q u ire s a n u m b e r o f d elay s e q u a l to th e o rd e r N o f th e sy stem . H o w e v e r, if M > N , th e re q u ire d m e m o ry is specified by M . F ig u re 2.34 can easily by m o d ified to h a n d le th is case. T h u s th e d ire c t form II stru c tu re re q u ire s M + N + 1 m u ltip lic a tio n s a n d m ax{M , jV} d elay s. B e cau se it re q u ire s th e m in im u m n u m b e r o f d elay s fo r th e re a liz a tio n o f th e system d e sc rib e d by (2.5.6), it is so m e tim es ca lle d a canoni c f o r m .

Sec. 2.5

Implementation of Discrete-Time Systems

115

A special case o f (2.5.6) occurs if w e set th e system p a ra m e te rs ak 0. it = 1.........N. T h e n th e in p u t-o u tp u t re la tio n sh ip fo r th e sy stem re d u c e s to M v(n) = k=0 w hich is a n o n re c u rsiv e lin e a r tim e -in v a ria n t system . T his sy stem view s only the m o st re c e n t M + 1 in p u t signal sa m p les a n d , p rio r to a d d itio n , w eights each sam ple by th e a p p ro p ria te co efficien t bk from th e set {i^}. In o th e r w ords, th e system o u tp u t is b asically a weight ed m o v i n g average of th e in p u t signal. F o r this reaso n it is so m e tim e s called a m o v i n g average ( M A ) syst em. S uch a system is an F IR sy stem w ith an im p u lse re sp o n s e h (k ) e q u a l to th e coefficients bk, th at is. - k) (2.5.11)

* '* ) = { o ! '

o to rw l"

'5 1 2

If w e r e tu rn to (2.5.6) a n d se t M = 0, th e g e n eral lin e a r tim e -in v a ria n t system red u ces to a p u re ly re c u rsiv e system d escrib ed by th e d ifferen ce e q u a tio n N y ( n ) = ~ Y a O (n ~ k) + bux(n) i =1 (2.5.13)

In th is case th e sy stem o u tp u t is a w eig h ted lin ear c o m b in a tio n o f N past o u tp u ts a n d th e p re s e n t in p u t. L in e a r tim e -in v a ria n t system s d esc rib e d by a se c o n d -o rd e r d ifferen ce e q u a tio n a re an im p o rta n t subclass o f th e m o re g e n eral system s d e sc rib e d by (2.5.6) o r (2.5.10) o r (2.5.13). T h e re a so n fo r th e ir im p o rta n c e will be ex p lain ed later w h en w e discuss q u a n tiz a tio n effects. Suffice to say at this p o in t th a t se c o n d -o rd e r sy stem s a re u su ally u sed as b a sic b u ild in g b lo ck s fo r realizin g h ig h e r-o rd e r system s. T h e m o st g e n e ra l se c o n d -o rd e r sy stem is d escrib ed by th e d ifferen ce e q u a tio n y (n ) = - a[v( n - 1) - a2v(n - 2) 4- b(,x(n) (2.5.14) + b\ x ( n 1) +- b 2x ( n 2) w hich is o b ta in e d fro m (2.5.6) b y se ttin g N = 2 an d M = 2. T h e d irect form II s tru c tu re fo r realizin g th is sy stem is sh o w n in Fig. 2.35a. If we se t a\ = = 0. th e n (2.5.14) re d u c e s to y( n) = box(n) + b\x(rt 1) + b 2x ( n - 2) (2.5.15)

w hich is a sp ecial case o f th e F IR system d esc rib e d by (2.5.11). T h e stru c tu re fo r realizin g th is sy stem is sh o w n in Fig. 2.35b. F inally, if w e set b\ = bz = 0 in (2.5.14), w e o b ta in th e p u re ly recu rsiv e se c o n d -o rd e r sy stem d esc rib e d by th e d iffe re n c e e q u a tio n y( n) = a \ y ( n 1) a 2y(n - 2) 4- box(rt) (2.5.16)

w hich is a sp e cial case o f (2.5.13). T h e stru c tu re fo r realizin g th is system is show n in Fig. 2.35c.

116

Discrete-Time Signals and Systems

Chap. 2

(a)

----- -0
(b)
x i n)

Y[n)

-^-1 H

I _a' -1

02 -1 <c)

Figure 2.35 Structures for the realization of second-order systems: (a) general second-order system; (b) FIR system; (c) purely recursive system

2.5.2 Recursive and Nonrecursive Realizations of FIR Systems

W e h a v e a lread y m a d e th e d istin ctio n b etw e e n F IR an d IIR system s, b ased on w h e th e r th e im p u lse re sp o n se h( n) o f th e system h a s a finite d u ra tio n , o r an infi n ite d u ra tio n . W e h av e also m a d e th e d istin c tio n b e tw e e n rec u rsiv e a n d n o n re c u r sive system s. B asically, a causal recu rsiv e system is d esc rib e d by an in p u t-o u tp u t e q u a tio n o f th e fo rm y(n) = F[v(^i 1).........y{n N ) , x ( n ) , ------ x ( n - M)] (2.5.17)

an d fo r a lin ear tim e -in v a ria n t system specifically, by th e d iffe re n c e e q u a tio n

N U

y { n ) = - Y ak>(n ~ k ) + Y , bkX(n ~ k ) *=i *=o

(2.5.18)

Sec. 2.5

Implementation of Discrete-Time Systems

117

O n th e o th e r h a n d , cau sal n o n recu rsiv e system s d o n o t d e p e n d o n p a st values of th e o u tp u t a n d h e n c e a re d escrib ed by an in p u t-o u tp u t e q u a tio n o f th e form y (n) = F [x (n ), jc(n - 1).........x ( n Mj ] (2.5.19)

a n d fo r iin e a r tim e -in v a ria n t system s specifically, by th e d iffe re n c e e q u a tio n in (2.5.18) w ith a k : = 0 fo r k = 1, 2 , . . . , N. In th e case o f F IR system s, we h av e a lre a d y o b se rv e d th a t it is aiw ays possible to re a liz e such sy stem s n o n recu rsiv ely . In fact, w ith at . = 0, k = 1, 2 , . . . , N , in (2.5.18), w e h av e a sy stem w ith an in p u t-o u tp u t e q u a tio n v() = ] p 6 * ;t( n ~ (2.5.20)

T h is is a n o n re c u rsiv e a n d F IR system . A s in d ic a te d in (2.5.12), th e im pulse re sp o n s e o f th e sy stem is sim ply e q u a l to th e coefficients {&*). H e n c e every F IR sy stem can b e re a liz e d n o n recu rsiv ely . O n th e o th e r h a n d , an y F IR system can also b e realized recu rsiv ely. A lth o u g h the g e n eral p ro o f o f th is s ta te m e n t is given la te r, w e shall give a sim ple ex am p le to illu stra te th e p o in t. S u p p o se th a t w e h av e an F IR system o f th e form 1 m y( n) = M + If-' - k) (2.5.21)

fo r co m p u tin g th e mo v i n g average of a signal jc(h). C learly , th is sy stem is F IR w ith im p u lse re sp o n se h( n) ~ 1 M + 1 0 < n < M

F ig u re 2.36 illu stra te s th e stru c tu re of th e n o n re c u rsiv e re a liz a tio n o f th e system . N ow , su p p o se th a t w e ex p ress (2.5.21) as 1 M y (n ) = --------- x ( n 1 k) M m + ^ 1^ Jt=0 + . [x(n) ~ x ( n - 1 - M )] M + 1 1 M + V [x(n) x (n 1 Af)] (2.5.22)

y(n - 1) +

Figure 2 3 6

N onrecursive realization of an FIR moving average system.

118

Discrete-Time Signals and Systems

Chap. 2

N ow , (2.5.22) re p re se n ts a recu rsiv e re a liz a tio n of th e F IR system . T h e stru c tu re o f th is recu rsiv e re a liz a tio n of th e m oving av erag e system is illu stra te d in Fig. 2.37. In su m m ary , w e can th in k of th e te rm s F IR a n d IIR as g e n e ra l ch aracteristics th a t distin g u ish a ty p e of lin e a r tim e -in v a ria n t system , and of th e term s recursive an d nonrecursi ve as d e sc rip tio n s of th e stru c tu re s fo r realizin g o r im p lem en tin g th e system .

vt/> - 1)
Figure 2 3 7 Recursive realization of an FIR moving averapc svstem.

2.6 CORRELATION OF DISCRETE-TIME SIGNALS

A m a th em atical o p e ra tio n th a t closely rese m b le s co n v o lu tio n is c o rre la tio n . Just as in th e case o f c o n v o lu tio n , tw o signal se q u e n c e s are in v o lv ed in c o rre la tio n . In c o n tra st to c o n v o lu tio n , h o w ev er, o u r o bjective in co m p u tin g th e c o rre la tio n b etw een th e tw o signals is to m e a su re th e d e g re e to w hich th e tw o signals are sim ilar an d th u s to e x tra c t so m e in fo rm a tio n th a t d e p e n d s to a large e x te n t on th e ap p licatio n . C o rre la tio n o f signals is o ften e n c o u n te re d in ra d a r, so n a r, digital co m m u n icatio n s, geolo g y, an d o th e r a re a s in science an d en g in eerin g . T o b e specific, le t us su p p o se th a t we have tw o signal se q u e n c e s x ( n ) an d y( n) th a t we w ish to co m p a re . In r a d a r an d active so n a r a p p lic a tio n s. x( n ) can re p re s e n t th e sa m p le d v ersio n o f th e tra n s m itte d signal and y{n) can re p re s e n t th e sam p led v ersio n o f th e receiv ed signal at th e o u tp u t o f the a n a lo g -to -d ig ita l (A /D ) c o n v erter. If a ta rg e t is p re se n t in th e space b e in g se a rc h e d by th e ra d a r o r so n a r, th e receiv ed signal y( n) consists of a d e la y e d v ersio n o f th e tra n sm itte d signal, reflec te d from th e ta rg e t, an d c o rru p te d b y a d d itiv e noise. F ig u re 2.38 d e p ic ts the ra d a r signal re c e p tio n p ro b le m . W e can re p re s e n t th e re c e iv e d signal se q u e n c e as y ( n ) = a x ( n D) + w( n ) (2.6.1)

w h ere a is som e a tte n u a tio n fa c to r re p re s e n tin g th e signal loss involved in th e ro u n d -trip tran sm issio n o f th e signal x (n ), D is th e ro u n d -trip d elay , w hich is

Sec. 2.6

Correlation of Discrete-Time Signals

119

assum ed to be an integer m ultiple o f the sam pling interval, and w (n) represents the additive n oise that is picked up by the antenna and any n oise generated by the electron ic com p onents and amplifiers contained in the front end o f the receiver. O n the other hand, if there is no target in the space searched by the radar and sonar, the received signal y(n ) consists o f noise alone. H avin g the tw o signal sequ en ces, x ( n ) , which is called the reference signal or transm itted signal, and y ( n ) , the received signal, the problem in radar and sonar d etection is to com p are y(n) and x ( n ) to determ ine if a target is present and, if so, to determ in e the tim e d elay D and com p ute the distance to the target. In practice, the signal x ( n D) is h eavily corrupted by the additive n oise to the point w here a visual inspection o f y ( n ) d oes n ot reveal the p resen ce or absence o f the desired signal reflected from the target. Correlation provides us with a m eans for extracting this im portant inform ation from y( n). D igital com m unications is an oth er area w here correlation is o ften used. In digital com m unications the inform ation to be transm itted from on e p oin t to an other is usually con verted to binary from , that is, a seq u en ce o f zeros and ones, which are then transm itted to the in tend ed receiver. T o transm it a 0 w e can trans m it the signal seq u en ce xo(n) for 0 < n < L 1, and to transm it a 1 w e can transmit the signal seq u en ce jti(n) for 0 < n < L 1, w here L is som e integer that d en otes the num ber o f sam p les in each o f the tw o sequ en ces. V ery often , x\ (n) is selected to be th e negative o f xo(n). T h e signal received by the in tend ed receiver m ay be represented as y (n ) = x,-(n) + w (n ) * = 0 ,1 0 < n < L 1 (2.6.2) w here now the uncertainty is w hether x 0(n) or *](n ) is the signal com p on en t in >(n), and w (n ) rep resents the additive n oise and other interferen ce inherent in

120

Discrete-Time Signals and Systems

Chap. 2

any co m m u n icatio n system . A g ain , such noise has its o rig in in th e e le ctro n ic c o m p o n e n ts c o n ta in e d in th e fro n t e n d o f th e receiv er. In an y case, th e receiv er k n o w s th e p o ssib le tra n sm itte d se q u e n c e s xo(n) a n d (n) a n d is fa c e d w ith the task o f co m p a rin g th e receiv ed signal y( n) w ith b o th xo(n) a n d Jti(n) to d e te rm in e w hich o f th e tw o signals b e tte r m a tc h e s y(n). T h is c o m p a riso n p ro c e ss is p e rfo rm e d by m ean s o f th e c o rre la tio n o p e ra tio n d e sc rib e d in th e follow ing su b se c tio n .

2.6.1 Crosscorrelation and Autocorrelation Sequences

S u p p o se th a t we h av e tw o real signal se q u e n c e s x ( n ) an d y ( n ) ea c h o f w hich has finite en erg y . T h e crosscorrelation o f x ( n ) and v(n) is a se q u e n c e rxy(l), w hich is d efin ed as r.tv(l)
n ~ x

~ 0

l = 0. 1 , 2 , . . .

(2.6.3)

or, eq u iv alen tly , as

OC

riy(t) Y , X ^n + 0 y ( )
n= -oc

1 = 0, 1 , 2 , . . .

(2.6.4)

T h e in d ex I is th e (tim e ) sh ift (o r lag) p a r a m e te r an d th e su b scrip ts x y on th e c ro ss c o rre la tio n se q u e n c e rxy(l) in d icate the se q u e n c e s b ein g c o rre la te d . T h e o rd e r of th e su b scrip ts, w ith .x p re c e d in g y, in d ic a te s th e d ire c tio n in w h ich o n e se q u en ce is sh ifted , re lativ e to th e o th e r. T o e la b o ra te , in (2.6.3), th e se q u e n c e x ( n ) is left u n sh ifte d an d y (n ) is sh ifted by / u n its in tim e, to th e rig h t fo r / p o sitiv e a n d to th e left for I n eg ativ e. E q u iv a le n tly , in (2.6.4), th e se q u en ce y ( ) is left u n sh ifted a n d x{n) is sh ifted by I u n its in tim e, to th e left fo r / p o sitiv e a n d to th e rig h t fo r / n eg ativ e. B u t sh iftin g x (n ) to th e left by / u n its relativ e to y (n ) is e q u iv a le n t to sh iftin g y() to th e rig h t by / u n its re lativ e to x ( n) . H e n c e th e c o m p u ta tio n s (2.6.3) an d (2.6.4) yield id en tical c ro ssc o rre la tio n se q u en ces. If w e rev erse th e ro les o f jr() an d y (n) in (2.6.3) an d (2.6.4) a n d th e re fo re re v e rse th e o rd e r o f th e indices xy. we o b ta in th e c ro ss c o rre la tio n se q u e n c e
OC

ryx(I) = n= 0 C o r, e q u iv alen tly ,

OC

y ( n ) x ( n I)

(2.6.5)

ryx(l) = Y
n d c

y ( n + l ) x ( n)

(2.6.6)

By c o m p a rin g (2.6.3) w ith (2.6.6) o r (2.6.4) w ith (2.6.5), we c o n c lu d e th a t rxy(l) = ryx( ~l ) (2.6.7)

T h e re fo re , r y i (l) is sim ply th e fo ld ed v ersio n o f rxy(l), w h e re th e fo ld in g is d o n e w ith re sp e c t to / = 0. H e n c e , ryx(l ) p ro v id e s exactly th e sam e in fo rm a tio n as rxv(l), w ith re sp e c t to th e sim ilarity o f x ( n ) to y(n ).

Sec. 2.6

Correlation of Discrete-Time Signals

121

Exam ple 2.6.1 D eterm ine the crosscorrelation sequence rxv(l) of the sequences x(n) = { ....0 .0 .2 . 1 .3 .7 .1 .2 . 3. 0. 0 ....}
t

v(n) = { . . . . 0 . 0 . 1 . - 1 . 2 . - 2 . 4 . 1 . - 2 . 5 .0 .0 ,...]
t

Solution

Let us use the definition in (2.6.3) to compute r rv(0) = Y x(ri)v(n)

.(/). For I = 0 w e have

The product sequence u()(n) = x ( n ) y ( r ) is u(n) = { ..., 0. 0. 2. 1. 6. -1 4 . 4. 2, 6, 0, 0. . . .)

t

and hence the sum over all values of n is r,v(0) = 7 For I > 0, we simpiy shift v() to the right relaLive to a ' ( h ) hy / units, compute the product sequence v/(n) = jr(n)_v(n I), and finally, sum o v er all va lu es o f the product sequence. Thus we obtain r t ,( l) = 13, rM5) = 5. rJV(2) = -1 8 . r ,v(6) = - 3 , r vv(3) = 16. rxy (/) = 0. r,,(4 ) = - 7 I >1

For / < 0, we shift y(n) to the left relative to jr(n) by / units, compute the product sequence v/(n) = j(n ) v(n I), and sum over all values of the product sequence. Thus we obtain the values of the crosscorrelation sequence rIV(1) = 0, rJV( -5 ) = 19, riy( - 2 ) = 33. fxv(-6) = - 9 , rly{ - 3) = -1 4 . rIV( - 7) = 10, rTV ( - 4 ) = 36 rxv(l) = 0, I < - 8

Therefore, the crosscorrelation sequence of x{n) and y(n) is r,AD = (10, - 9 ,1 9 , 36, -1 4 , 33,0, 7,13, -1 8 ,1 6 . - 7 , 5, - 3 )
t

T h e sim ilarities b e tw e e n th e c o m p u ta tio n o f th e c ro ss c o rre la tio n o f tw o se q u e n c e s a n d th e c o n v o lu tio n o f tw o se q u e n c e s is a p p a re n t. In th e c o m p u ta tio n of c o n v o lu tio n , o n e o f th e se q u e n c e s is fo ld ed , th e n sh ifted , th e n m u ltip lie d by th e o th e r se q u e n c e to fo rm th e p ro d u c t se q u e n c e fo r th a t shift, a n d finally, th e values o f th e p r o d u c t se q u e n c e a re su m m ed . E x c e p t fo r th e fo ld in g o p e ra tio n , th e co m p u ta tio n o f th e c ro ss c o rre la tio n se q u e n c e involves th e sa m e o p e ra tio n s: shifting o n e o f th e se q u e n c e s , m u ltip lic atio n o f th e tw o se q u e n c e s, an d su m m in g o v er all v alu es o f th e p r o d u c t se q u e n c e . C o n se q u e n tly , if w e h av e a c o m p u te r p ro g ra m th a t p e rfo rm s c o n v o lu tio n , w e can use it to p e rfo rm c ro ss c o rre la tio n by p ro v id in g

122

Discrete-Time Signals and Systems

Chap. 2

as in p u ts to th e p ro g ra m , th e se q u e n c e jc() an d th e fo ld ed se q u e n c e y ( n). T h e n th e c o n v o lu tio n o f x ( n ) w ith y ( n) yields th e c ro ss c o rre la tio n r rv(/). th a t is, rxv(D = x ( l ) * y ( - l ) (2.6.8)

In th e special case w h ere y( n) = x( n ) , we h av e th e aut ocorrel ati on o f *( ), w hich is d efined as th e se q u e n c e

rX x(l)=
o r, eq u iv alen tly , as

O C
x(n)j:(n - 0
ft ~ O C

(2.6.9)

O O
rxx{l) = ^ 2 , x (n +
n= oc

(2.6.10)

In d ealin g w ith fin ite -d u ra tio n se q u e n c e s, it is cu sto m a ry to ex p ress th e a u to c o rre la tio n an d c ro ss c o rre la tio n in te rm s of th e finite lim its on th e su m m a tio n . In p a rtic u la r, if x ( ) a n d v(n) a re causal se q u e n c e s o f le n g th N [i.e., .v(n) = y(n) = 0 for n < 0 an d n > N], th e c ro ssc o rre la tio n an d a u to c o rre la tio n se q u e n c e s m ay be ex p ressed as rxy(l) = an d A'-1*1-1 rxx( l ) = ^ x(n)y(n-l) (2.6.11)

Y
n=f

(2.6.12)

w h ere i = I, k = 0 fo r / > 0, a n d / = 0, k = I fo r / < 0.

2.6.2 Properties of the Autocorrelation and Crosscorrelation Sequences

T h e a u to c o rre la tio n an d c ro ssc o rre la tio n se q u e n c e s h av e a n u m b e r of im p o rta n t p ro p e rtie s th a t w e n o w p re se n t. T o d e v e lo p th e s e p ro p e rtie s , le t us assu m e th a t we h av e tw o se q u e n c e s x ( n ) a n d y (n) w ith finite en erg y from w hich w e fo rm the lin e a r c o m b in a tio n , ax(n) bv( n I)

w h ere a an d b a re a rb itra ry c o n s ta n ts a n d I is so m e tim e shift. T h e en erg y in this signal is

O C

O C
[ax(n) + by( n - I)]2 = a 2 ^
n = oc

O C
x 2( n ) + b 2 ^
/i oc

Y
fj -oc

y 2{n - I) (2 '6 -13)

. 2ab -j l ^ V""' x (in ) \ y (tn I) n +

n = oc

= a 2rxx(0) + b2r yy( 0) + l a b r xy{l)

Sec. 2.6

Correlation of Discrete-Time Signals

123

F irst, we n o te th a t rx x (0) = E x a n d /-vy(0) = v, w hich a re th e en e rg ie s o f x{n) an d y (n ), resp ectiv ely . It is o b v io u s th a t a 2rxx(0) -I- b 2r y> .(0) + 2abrxy(l) > 0 N ow , assu m in g th a t b ^ 0, w e can divide (2.6.14) by b 2 to o b ta in r (0 ) (2.6.14)

(^ )2

+ 2rxy(l) Q

+ r v_ v(0) > 0

W e view th is e q u a tio n as a q u a d ra tic w ith coefficients r XJ(0), 2 rxv(l), a n d ^ ,( 0 ) . Since th e q u a d ra tic is n o n n e g a tiv e , it follow s th a t th e d isc rim in a n t of this q u a d ra tic m u st b e n o n p o sitiv e , th a t is, 4 [ r ; v(/) - r , , ( 0 K v(0)] < 0 T h e re fo re , th e c ro ss c o rre la tio n se q u e n c e satisfies th e c o n d itio n th a t |r,v(/>| < y r ( 0 ) r , v(0) = In th e sp ecial case w h ere v(n) = x ( n ), (2.6.15) re d u c e s to \rxx( D \ < r xx(0) = E x (2.6.16) (2.6.15)

T his m ean s th a t th e a u to c o rre la tio n se q u e n c e o f a signal a tta in s its m ax im u m value at z e ro lag. T h is re su lt is c o n siste n t w ith th e n o tio n th a t a sig n a l m a tc h e s p erfe ctly w ith itself at z e ro shift. In th e case o f th e c ro ssc o rre la tio n se q u en ce, th e u p p e r b o u n d o n its v alu es is given in (2.6.15). N o te th a t if an y o n e o r b o th o f th e signals involved in th e c ro ssc o rre la tio n a re scaled , th e sh a p e o f th e c ro ss c o rre la tio n se q u e n c e d o e s n o t ch an g e, only the a m p litu d e s o f th e c ro ss c o rre la tio n se q u e n c e a re sc aled accordingly. S ince scaling is u n im p o rta n t, it is o ften d e s ira b le , in p ra c tic e , to n o rm a liz e th e a u to c o rre la tio n a n d c ro ss c o rre la tio n se q u e n c e s to th e ran g e fro m - 1 to 1. In th e case o f the a u to c o rre la tio n se q u e n c e , w e can sim ply d iv id e by ^ ( 0 ) . T h u s th e n o rm aliz ed a u to c o rre la tio n se q u e n c e is d efin ed as P. A D = (2-6 -17)

rixiO)

Sim ilarly, we d efin e th e n o rm a liz e d c ro ssc o rre la tio n se q u e n c e pXY(l) = r ' v(l) : v/ r xx(0 )rvv(0) (2.6.18)

N ow \pXI{l)\ < 1 a n d |/oX v(0! < 1, a n d h e n c e th ese se q u e n c e s a re in d e p e n d e n t of signal scaling. F in ally , as we h av e a lre a d y d e m o n s tra te d , th e c ro ss c o rre la tio n se q u e n c e sa t isfies th e p ro p e rty
r Xy ( l ) = f y x ( - 0

124

Discrete-Time Signals and Systems

Chap. 2

W ith y(n) = x ( n) , th is re la tio n resu lts in th e follow ing im p o rta n t p ro p e rty fo r the a u to c o rre la tio n se q u en ce
r { l ) = rx s ( . - l )

(2.6.19)

H e n c e th e a u to c o rre la tio n fu n ctio n is an ev en fu n ctio n . C o n s e q u e n tly , it suffices to co m p u te rx x (l) fo r / > 0. Example 2.6.2 Compute the autocorrelation of the signal x(n) = a"u(n), 0 < a < 1 Solution Since x (n) is an infinite-duration signal, its autocorrelation also has infinite duration. We distinguish two cases. If I > 0. from Fig. 2.39 we observe that
r t J (/) =
n= l

x ( n ) x ( n - /) =
n =i

' = a

1
n~l

Since a < 1, the infinite series cot i crges and we obtain r tl (/) ~ For / < 0 we have =
n=(l

^ -V 1 a-

I> 0

x(n)x(/i /) = a -1'S~'(rr)" = ------- -ti~! I - an=(l a ' 1'.

/ < 0
r i t (!)

But when / is negative, c r 1 into the following expression: rx,(!) =

Thus the two relations for

can be combined

,,a ,t: oc < / < oc 1 a~ The sequence rxx(l) is shown in Fig. 2.42(d). We observe that r ( ~ / | = rxAD and

(2.6.20)

rlt(0) =

1 a2

Therefore, the normalized autocorrelation sequence is r (/) p s,(!) - cr|,: ~ oc < I < oc rxxW)

(2.6.21)

2.6.3 Correlation of Periodic Sequences

In S ectio n 2.6.1 w e d efin ed th e c ro ss c o rre la tio n an d a u to c o r re la tio n se q u en ces of en erg y signals. In this se ctio n we co n sid e r th e c o rre la tio n se q u e n c e s o f p o w er signals an d , in p a rtic u la r, p e rio d ic signals. L et x( n ) a n d y(rc) b e tw o p o w e r signals. T h e ir c ro ss c o rre la tio n se q u e n c e is d efin ed as 1 M
rx \ 0 )

M -oc 2 M + 1

lim

----- - Y ]

x(n)y(n~l)

( 2 .6 .2 2 )

Sec. 2.6

Correlation of Discrete-Time Signals

xi n )

125

) i'

-2-10123
(a)

in .
/ >o

xin-I)

o
(b)

x(n - I )

l< 0

(c)

r(!) =

' , a1 '1 1 - a2

- 2 - 1 0
Figure 139

1 2 (d)

/
=

Compulation of the autocorrelation o f the signal xin)

a",

0 < a < 1. If x ( n ) = y(tt), we have the definition o f the autocorrelation seq u en ce of a p ow er signal as 1 M rxx(I) = iim . . . , 1 Y ] x ( n ) x ( n - l ) (2.6.23) Af-oo 2M + 1 In particular, if x ( n ) and y ( n ) are two periodic seq u en ces, each with period the averages indicated in (2.6.22) and (2.6.23) o ver the infinite interval, are identical

126

D iscrete-Time Signals and Systems

Chap. 2

to th e av erag es o v er a single p e rio d , so th a t (2.6.22) an d (2.6.23) re d u c e to (2.6.24) an d (2.6.25) It is c lear th a t r ry(l) an d rxx(l) are p e rio d ic c o rre la tio n T h e fa c to r 1 / N can b e v iew ed as a n o rm a liz a tio n scale In som e p ractical ap p licatio n s, c o rre la tio n is u se d an o b se rv e d physical signal w hich m ay be c o rru p te d by ex am p le, c o n sid er a signal se q u e n c e y( n) o f th e form y{n) = * (n ) + w(n) se q u e n c e s w ith p e rio d N . facto r. to id en tify p erio d icitie s in ra n d o m in te rfe re n c e . F o r (2.6.26)

w h ere jc(/i) is a p erio d ic se q u e n c e o f so m e u n k n o w n p e rio d N a n d w( n) re p re se n ts an ad d itiv e ran d o m in te rfe re n c e . S u p p o se th a t w e o b se rv e M sa m p le s o f y(n ), say 0 < n < M 1, w h ere M > > N. F o r all p ractical p u rp o se s, w e can assum e th a t y( n) 0 fo r n < 0 an d n > M. N ow th e a u to c o rre la tio n se q u e n c e of y(n), using th e n o rm aliz atio n facto r o f \ / M . is (2.6.27) If we su b stitu te for y(n) fro m (2.6.26) in to (2.6.27) w e o b ta in _ _1 ry.T(/) = j M- 1 Y ] x ( n ) x i n - /)

M
j M J

H - y ^ [ .r ( n ) itj (/2 /) + w{ n) x{n /)] M

(2.6.28)

T h e first fa c to r on th e rig h t-h a n d sid e of (2.6.28) is th e a u to c o rre la tio n se q u e n c e o f x i n) . S ince x( n) is p e rio d ic , its a u to c o rre la tio n s e q u e n c e ex h ib its th e sam e p erio d icity , th u s co n ta in in g relativ ely larg e p e a k s at / = 0, N , 2 N , a n d so on. H o w ev er, as th e shift I a p p ro a c h e s M , th e p e a k s a re re d u c e d in a m p litu d e d u e to th e fact th a t w e h av e a finite d a ta re c o rd o f M sa m p le s so th a t m any o f th e p ro d u c ts ;t(n)j:(rt /) a re zero . C o n s e q u e n tly , w e sh o u ld av o id c o m p u tin g r V T(/) fo r larg e lags, say, I > M f l .

Sec. 2.6

Correlation of Discrete-Time Signals

127

T h e c ro ss c o rre la tio n s rxu,(/) an d rwx(l ) b etw e e n th e signal a n d th e a d ditiv e ra n d o m in te rfe re n c e a re e x p ected to be relativ ely sm all as a resu lt of the e x p e c ta tio n th a t ;r(/i) a n d w ( n ) will be to ta lly u n re la te d . F in ally , the last term on th e rig h t-h a n d sid e of (2.6.28) is th e a u to c o rre la tio n se q u e n c e o f th e ra n d o m se q u e n c e w( n) . T h is c o rre la tio n se q u en ce will c e rta in ly c o n tain a p e a k a t / = 0, but b ecau se o f its r a n d o m ch aracteristics, r ww(l) is ex p e c te d to d ecay rap id ly to w ard zero . C o n s e q u e n tly , o nly rxx{l) is e x p e c te d to h av e large p e a k s fo r / > 0. T his b e h a v io r allow s us to d e te c t th e p re se n c e of th e p e rio d ic signal a (/ j ) b u rie d in the in te rfe re n c e u>(n) a n d to id e n tify its p e rio d . A n e x am p le th a t illu stra te s th e use of a u to c o rre la tio n to identify a h id d e n p e rio d ic ity in an o b se rv e d physical signal is show n in Fig. 2.40. T h is figure illus tra te s th e a u to c o rre la tio n (n o rm a liz e d ) se q u e n c e fo r th e W o lfe r su n sp o t n u m b e rs fo r 0 < / < 20, w h ere an y v alu e o f / c o rre sp o n d s to o n e y ear. T h e se n u m b e rs are given in T a b le 2.2 fo r th e 100-year p e rio d 1770-1869. T h e re is c lear ev id en ce in th is fig u re th a t a p e rio d ic tre n d exists, w ith a p e rio d o f 10 to 11 years. Example 2.6.3 Suppose that a signal sequence jr(n) = sin(7r/5)/i, for 0 < n < 99 is corrupted by an additive noise sequence Kn), where the values of the additive noise are selected independently from sample to sample, from a uniform distribution over the range
TABLE 2.2 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794

YEARLY WOLFER SUNSPOT NUMBERS

101 82 66 35 31 7 20 92 154 125 85 68 38 23 10 24 83 132 131 118 90 67 60 47 41 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 21 16 6 4 7 14 34 45 43 48 42 28 10 8 2 0 1 5 12 14 35 46 41 30 24 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 16 7 4 2 8 17 36 50 62 67 71 48 28 8 13 57 122 138 103 86 63 37 24 11 15 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 40 62 9X 124 96 66 64 54 39 21
7

4 23 55 94 96 77 59 44 47 30 16 7 37 74

128

Discrete-Time Signals and Systems

Chap. 2

Year

(a)

Figure 2.40 Identification of periodicity in the Wolfer sunspot numbers: (a) an nual Wolfer sunspot numbers; (b) autocorrelation sequence.

(A /2, A /2), where A is a param eter of the distribution. The observed sequence is y(n) = x(n) + w(n). D eterm ine the autocorrelation sequence rvt(n) and thus determine the period of the signal x(rt). Solution The assumption is that the signal sequence x(n) has some unknown period that we are attem pting to determ ine from the noise-corrupted observations {y(n)). Although x(n) is periodic with period 10, we have only a ftmte-duration sequence of

Sec. 2.6

Correlation of Discrete-Time Signals

129

length M = 100 [i.e.. 10 periods of jcm )]. The noise power level P in the sequence u'(n) is determ ined by the param eter A. We simply state that Pu = A: /12. The signal power level is P, = I. Therefore, the signal-to-noise ratio (SNR) is defined as P,

~ P ~ U ~ A : /1 2 A :

Usually, the SNR is expressed on a logarithmic scale in decibels (dB) as 101ogui ( PJPv) . Figure 2.41 illustrates a sample of a noise sequence w(n), and the observed sequence y(n) x(n) + it(n) when the SNR = 1 dB. The autocorrelation sequence

J ' t I l l Lt ttIiIt . Tllrii! ill ..tI 1. I i i h i 1 [Ijj-liii p p T T J ljw ^ IT 1 i( i| 4 * 4 i

j * 4 1 1

'

la]

(c)
Figure 2.41

Use of autocorrelation to detect the presence of a periodic signal corrupted by

noise.

130

Discrete-Tim e Signals and Systems

Chap. 2

w(n)

L TT T T T . T tT J IT .T T U T tT T ._L T .T. I .T .

pp V |V v li

'

t t T .1 .

(a)

(b)

rM .

1 ?

r![It x ij|il

i o1

rl It

r!lfit

F
(c)

1 t T T * ? 1 [1

Figure 2.42 Use of autocorrelation to detect the presence of a periodic signal corrupted by noise.

r vv(/) is illustrated in Fig. 2.41c. W e observe that the periodic signal *(/?), embedded in y(n), results in a periodic autocorrelation function rZI(l) with period N 10. The effect of the additive noise is to add to the peak value at / = 0. but for I ^ 0, the correlation sequence rwu,(l) = ^ 0 a s a result of the fact that values of w(ri) were gen erated independently. Such noise is usually called white noise. The presence of this noise explains the reason for the large peak at I = 0. The smaller, nearly equal peaks at I = 10, 2 0 ,... are due the periodic characteristics of x(n). Figure 2.42 illustrates the noise sequence w(n), the noise-corrupted signal y(n), and the autocorrelation sequence r vv(/) for the same signal, within which is embedded a signal at a smaller noise level. In this case, the SNR = 5 dB. E ven with this relatively small noise level, the periodicity of the signal is not easily determ ined from observa tion of y( n ). However, it is clearly evident from observation of the autocorrelation sequence ryy(n).

2.6.4 Computation of Correlation Sequences

A s indicated on Section 2.6.1, the procedure for com puting the crosscorrelation seq u en ce b etw een x ( n ) and y ( n ) in volves shifting on e o f the seq u en ces, say x(n),

Sec. 2.6

Correlation of Discrete-Time Signals

131

to o b ta in x( n - /), m u ltip ly in g th e sh ifted se q u e n c e by v(n) to o b ta in th e p r o d u ct se q u e n c e y ( n ) x ( n - /), an d th en su m m in g all th e v alu es o f th e p ro d u c t se q u en ce to o b ta in r yx(l). T his p ro c e d u re is r e p e a te d fo r d iffe re n t values o f th e lag /. E x c e p t fo r th e fo ld in g o p e ra tio n th a t is involved in c o n v o lu tio n , th ese b a sic o p e ra tio n s fo r co m p u tin g th e c o rre la tio n se q u e n c e a re id e n tic a l to th o se in co n v o lu tio n . T h e p ro c e d u re fo r co m p u tin g th e co n v o lu tio n is directly a p p licab le to c o m p u tin g th e c o rre la tio n o f tw o se q u en ces. S pecifically, if w e fo ld y (n) to o b ta in y ( n), th e n th e c o n v o lu tio n o f x ( n ) w ith v ( n) is id en tical to th e cro ssc o rre la tio n o f x ( n ) w ith y(n). T h a t is. rxy(l) = x(rt) * y ( n ) |n=/ (2.6.29)

A s a c o n se q u e n c e , th e c o m p u ta tio n a l p ro c e d u re d escrib ed for c o n v o lu tio n can be a p p lied d irectly to th e c o m p u ta tio n o f th e c o rre la tio n se q u en ce. W e no w d e sc rib e an a lg o rith m th a t can b e easily p ro g ra m m e d to c o m p u te th e c ro ss c o rre la tio n se q u e n c e of tw o fin ite -d u ra tio n signals * ( ), 0 < n < N ~ I, an d y ( n) , 0 < n < M 1. T h e alg o rith m c o m p u te s rJV(/) fo r p o sitiv e lags. A cc o rd in g to th e re la tio n rxy(l) = r Y X(I), th e v alu es o f rXY(l) fo r n eg ativ e lags can be o b ta in e d by using th e sam e a lg o rith m fo r p o sitiv e lags, an d in te rc h a n g in g th e roles o f jt( ) an d v(n). W e o b se rv e th a t if M < N, rXY{i) can be c o m p u te d by th e re la tio n s
M - \ + l

Y ' , x ( n ) y ( n /),

0 < 1 < N M (2.6.30)

O n th e o th e r h a n d , if M > N , th e fo rm u la fo r th e c ro ss c o rre la tio n becom es rxy(l) = Y x ( n ) v ( n - 1) 0 < 1< N -1 (2.6.31)

T h e fo rm u la s in (2.6.30) a n d (2.6.31) c a n b e c o m b in ed a n d c o m p u te d by m ean s o f th e fo llo w in g sim p le alg o rith m illu stra te d in th e flow chart in Fig. 2.43. By in terch an g in g th e ro les o f j ( ) a n d y(n) an d re c o m p u tin g th e cro ssc o rre la tio n se q u e n c e , w e o b ta in th e v alu es o f rXY(l) c o rre sp o n d in g to n eg ativ e shifts I. If w e w ish to c o m p u te th e a u to c o rre la tio n se q u e n c e rxx(l), w e se t y (n ) = jr(n) an d M = N in (2.6.31). T h e co m p u ta tio n o f rxx{l) can be d o n e by m e a n s of the sa m e a lg o rith m fo r p o sitiv e shifts only.

2.6.5 Input-Output Correlation Sequences

In th is se ctio n w e d e riv e tw o i n p u t- o u tp u t re la tio n sh ip s fo r L T I system s in th e c o rre la tio n d o m a in . L e t us assu m e th a t a signal x ( n ) w ith k n o w n a u to c o r re la tio n rxx(I) is a p p lied to an L T I system w ith im p u lse re sp o n s e h( n), p ro d u c in g th e

132

Discrete-Tim e Signals and Systems

Chap. 2

C j lD

Figure 2A3 Flowchart for software implementation of crosscorrelation.

Sec. 2.6

Correlation of Discrete-Time Signals

133

o u tp u t signal
v(n) = h( n) * .x (/?) = ^ h(k)x(>i k)

k= oc T h e c ro ss c o rre la tio n b e tw e e n th e o u tp u t a n d th e in p u t signal is rV J-(7) = y d ) * x ( - l ) h ( l ) * [*(/) * * ( - / ) ] or ryx(l) = /;(/) * rxx(l) (2.6.32)

w h ere w e h av e u se d (2.6.8) a n d th e p ro p e rtie s o f co n v o lu tio n . H e n ce th e c ro ssco r re la tio n b e tw e e n th e in p u t an d the o u tp u t of th e sy stem is th e co n v o lu tio n of the im p u lse re sp o n s e w ith the a u to c o rre la tio n of th e in p u t se q u e n c e . A lte rn a tiv e ly . ryx(l) m av b e view ed as th e o u tp u t of th e L T I sy stem w hen th e in p u t se q u en ce is rxx (/). T h is is illu stra te d in Fig. 2.44. If we rep lace / by - / in (2.6.32), w e o b tain r.tyU) = h ( - D * rxx(l) T h e a u to c o rre la tio n of th e o u tp u t signal can be o b ta in e d by using (2.6.8) w ith x(/i) = y(/i) an d th e p ro p e rtie s o f co n v o lu tio n . T h u s we have r lv(/) = v(/) * v (/) = [/;(/) * * (/)] * [/j ( / J * * ( /)] (2.6.33) = [/?(/) * h ( /)] * [*(/) * * ( /)] = rhh(l) * rxx(l) T h e a u to c o rre la tio n rhh(l) of th e im pulse resp o n se h(n) exists if th e system is stable. F u rth e rm o re , th e stab ility in su res th a t th e system d o e s n ot ch an g e th e ty p e (en erg y o r p o w e r) o f th e in p u t signal. By ev alu atin g (2.6.33) fo r / = 0 we o b ta in
CC

(2.6.34) w hich p ro v id e s th e en e rg y (o r p o w er) of th e o u tp u t signal in te rm s o f a u to c o r re latio n s. T h e se re la tio n sh ip s h o ld fo r b o th en erg y a n d p o w e r signals. T h e direct d e riv a tio n o f th e s e re la tio n sh ip s for en erg y an d p o w e r signals, a n d th e ir e x ten sio n s to co m p lex sig n als, are left as exercises fo r the stu d e n t.

Input
rxx<n )

LTI SYSTEM

Output rvr(n) Figure 2.44 Input-output relation for crosscorrelation ryx(n).

ft(n)

134

Discrete-Time Signals and Systems

Chap. 2

2.7 SUMMARY AND REFERENCES

T h e m a jo r th e m e o f th is c h a p te r is th e c h a ra c te riz a tio n o f d isc re te -tim e signals an d sy stem s in th e tim e d o m ain . O f p a rtic u la r im p o rta n c e is th e class o f lin e a r tim ein v a ria n t (L T I) sy stem s w hich a re w idely u se d in th e design a n d im p le m e n ta tio n o f d ig ital signal p ro cessin g system s. W e c h a ra c te riz e d L T I sy ste m s by th e ir u n it sa m p le re sp o n se h{n) an d d e riv e d th e co n v o lu tio n su m m a tio n , w hich is a fo rm u la fo r d e te rm in in g th e re sp o n se y( n) o f th e system c h a ra c te riz e d by h( n) to an y given in p u t se q u e n c e x(n). T h e class o f L T I system s c h a ra c te riz e d by lin e a r d ifferen ce e q u a tio n s w ith c o n s ta n t coefficients is by fa r th e m o st im p o rta n t of th e L T I sy stem s in th e th e o ry an d a p p licatio n o f d ig ital sig n a l p ro cessin g . T h e g e n e ra l s o lu tio n of a lin e a r dif fe re n c e e q u a tio n w ith c o n s ta n t coefficients w as d eriv e d in th is c h a p te r an d show n to co n sist of tw o c o m p o n en ts: th e so lu tio n o f th e h o m o g e n e o u s e q u a tio n w hich r e p re s e n ts th e n a tu ra l re sp o n se o f th e sy stem w h en th e in p u t is z e ro , an d th e p a r tic u la r so lu tio n , w hich re p re s e n ts th e re sp o n s e o f th e system to th e in p u t signal. F ro m th e d ifferen ce e q u a tio n , w e also d e m o n s tra te d h o w to d e riv e th e u n it sam ple re sp o n s e of th e L T I system . L in e a r tim e -in v a ria n t sy stem s w ere g en erally su b d iv id ed in to F IR (finited u ra tio n im p u lse re sp o n se ) a n d I I R (in fin ite -d u ra tio n im pulse re sp o n s e ) d e p e n d ing o n w h e th e r h(n) h as finite d u ra tio n o r infinite d u ra tio n , resp ec tiv ely . T he re a liz a tio n s o f such system s w ere briefly d escrib ed . F u rth e rm o re , in the re a liz a tio n o f F IR system s, we m a d e th e d istin ctio n b e tw e e n recu rsiv e a n d n o n recu rsiv e re a lizatio n s. O n th e o th e r h a n d , w e o b se rv e d th a t I I R system s can be im p le m e n te d recu rsiv ely , only. T h e re are a n u m b e r o f te x ts o n d isc re te -tim e signals a n d system s. W e m e n tio n as ex am p les th e b o o k s by M c G illem a n d C o o p e r (1984), O p p e n h e im a n d W illsky (1983), an d S ie b e rt (1986). L in e a r c o n sta n t-c o e ffic ie n t d iffe re n c e e q u a tio n s are tr e a te d in d e p th in th e b o o k s by H ild e b ra n d (1952) a n d L evy a n d L essm an (1961). T h e last to p ic in this c h a p te r, o n c o rre la tio n o f d isc re te -tim e signals, plays an im p o rta n t ro le in d ig ital signal p ro cessin g , esp ecially in a p p lic a tio n s d ealin g w ith digital c o m m u n ic a tio n s, ra d a r d e te c tio n a n d e stim a tio n , so n a r, a n d geophysics. In o u r tr e a tm e n t o f c o rre la tio n se q u e n c e s, w e a v o id e d th e use o f sta tis tic a l concepts. C o rre la tio n is sim ply d efin ed as a m a th e m a tic a l o p e r a tio n b e tw e e n tw o se q u en ces, w hich p ro d u c e s a n o th e r se q u e n c e , called e ith e r th e crosscorrelat ion s e quence w hen th e tw o se q u e n c e s a re d iffe re n t, o r th e aut ocorrel ati on sequence w h e n th e tw o se q u e n c e s are id en tical. In p ractical a p p lic a tio n s in w hich c o rre la tio n is u sed , o n e ( o r b o th ) o f th e se q u e n c e s is (a re ) c o n ta m in a te d by n o ise a n d , p e rh a p s , by o th e r fo rm s o f in te rfe r en ce. In such a case, th e noisy se q u e n c e is c alled a r a n d o m se q u e n c e a n d is c h a r a c te riz e d in sta tistical term s. T h e c o rre sp o n d in g c o rre la tio n se q u e n c e b e c o m e s a fu n ctio n o f th e sta tistical c h a ra c te ristic s of th e n o ise an d an y o th e r in te rfe re n c e . T h e statistical c h a ra c te riz a tio n o f se q u e n c e s a n d th e ir c o rre la tio n is tr e a te d in A p p e n d ix A . S u p p le m e n ta ry re a d in g o n p ro b a b ilis tic an d sta tistical c o n c e p ts deal-

Chap. 2

Problems

135

ing w ith c o rre la tio n can be fo u n d in th e b o o k s by D a v e n p o rt (1970). H e lstro m (1990). P ap o u lis (1984). an d P eeb les (1987).

PROBLEMS
2.1 A discrete-time signal x(n) is defined as

l + j, 1. 0,

3< n < 1 0 < n < 3 elsewhere

(a) Determ ine its values and sketch the signal .v(n). (b) Sketch the signals that result if we:

(1) First fold x{n) and then delay the resulting signal by four samples. (2) First delay xin) by four samples and then fold the resulting signal (c) Sketch the signal x ( n + 4 ). (d) Com pare the results in parts (b) and (c) and derive a rule for obtaining the signal ,v( n -t- k) from (e) Can you express the signal ,r(n) in term s of signals S(n) and u{n)l 2.2 A discrete-time signal ,v(n) is shown in Fig. P2.2. Sketch and label carefully each of the following signals.
.v(n)

J_

J_

L L _ ________ _
- 2 - 1 0 1 2 3 4
(a ) x(n - 2)

FigUre P2.2

(b) x(4 n) ( c)x(n + 2) (d) x(n)u(2 n) (e) x(n - 1 )8{n - 3) (F) x( n2) (g) even part of x{n) (h) odd part of x ( n ) 2 3 Show that (a ) &(n) = u(n) u(n 1) (b) u(n) =

8(k) =

2.4 Show that any signal can be decomposed into an even and an odd component. Is the decomposition unique? Illustrate your arguments using the stgnal x(n) = {2. 3, 4. 5. 6)
t

2.5 Show that the energy (power) of a real-valued energy (power) signal is equal to the sum of the energies (powers) of its even and odd components. 2.6 Consider the system
vf) = T[x( n) ] = x ( n 2)

(a) Determ ine if the system is time invariant.

136

Discrete-Time Signals and Systems

Chap. 2

(b) To clarify the result in part (a) assume that the signal _fl, \ 0, O 5 /1 < 3 elsewhere

is applied into the system. (1) Sketch the signal *(). (2) D eterm ine and sketch the signal y ( n ) = T[x(n)}, (3) Sketch the signal y'2(n) = y(n 2). (4) D eterm ine and sketch the signal x2(n) = x(n - 2). (5) Determ ine and sketch the signal = T \ x 2(n)]. (6) Com pare the signals ^ ( n ) and y(n - 2). W hat is your conclusion? (c) R epeat part (b) for the system
y(rt) = x(n) - x(n 1)

Can you use this result to make any statem ent about the time invariance of this system? Why? (d) Repeat parts (b) and (c) for the system y(rt) = T [ x ( n )] = nx(n) 2.7 A discrete-time system can be (1) Static or dynamic (2) Linear or nonlinear (3) Time invariant or time varying (4) Causal or noncausal (5) Stable o r unstable Examine the following systems with respect to the properties above. (a) v() = cos[jc(n)] (b) (c) (d) (e) v(n) = x(k> v(n) = x(n)cos(^n) y(n) ~ x ( n + 2) y(n) = Trun[;c(n)], where Trun[jc(n)] denotes the integer part of x(n), obtained by truncation (f) y(n) = Round[jc(n)], where Round[;c(n)] denotes the integer part of Jt(n) obtained by rounding Remark: The systems in parts (e) and (f) are quantizers that perform truncation and rounding, respectively. (g) y(B) = |*(n)| (h) v(rt) = x(n)u(n) (I) y(n) = x(n) + nx{n + 1) (j) y ( n ) = x ( 2 n )

(I) y( n ) = x ( - n ) (m) y(n) = sign[j:(n)] (n) The ideal sampling system with input xaU) and output x(n) = x a(nT), oc < n < oo 2J8 Two discrete-time systems 7] and T2 are connected in cascade to form a new system T as shown in Fig. P2.8. Prove or disprove the following statements.

Chap. 2

Problems

137

xin )

yin)

T i
T - T-. T 2

T; Figure P2.8

(a) If T\ and % are linear, then T is linear (i.e.. the cascade connection of two linear systems is linear). (b) If T\ and are time invariant, then T is time invariant. (c) If T[ and 7? are causal, then T is causal. <d) If T] and T2 are linear and time invariant, the same holds for T . (e) If 7] and T2 are linear and time invariant, then interchanging their order does not change the system T. (0 As in part (e) except that 7J, T2 are now time varying. (Hint: Use an example.) (g) If 7] and T2 are nonlinear, then T is nonlinear. (h) If T< and T2 are stable, then T is stable. (i) Show by an example that the inverse of parts (c) and (h) do not hold in general. 2.9 Let T be an LTI, relaxed, and BIBO stable system with input x{n) and output y(n). Show that: (a) If x(n) is periodic with period N [i.e., jr(n) = x{n + N) for all n > 0], the output y(n) tends to a periodic signal with the same period. (b) If x(n) is bounded and tends 10 a constant, the output will also tend to a constant. (c) If x{n) is an energy signal, the output y(n) will also be an energy signal. 2.10 The following in p u t-output pairs have been observed during the operation of a umeinvariam system: x,(n) = {1.0,2} ^
t

y,(fi) = (0, 1.2}

t

x,(n) = {0.0,3} ^
t

v; (n) = (0, 1.0,2}

t

x-\(n) = {0. 0, 0. 1}
t

vj(n) = (1,2, 1}
t

Can you draw any conclusions regarding the linearity of the system. W hat is the impulse response of the system? 2.11 The following input-output pairs have been observed during the operation of a linear system: Xi(n) = {-1. 2. 1}
t

y,(/i) = (1, 2. - 1 , 0. 1}
t

x2{n) = {1, - 1 , -1}

t x 3(n) = {0, 1, 1) t

\'2in) = {-1. 1, 0, 2}
t y i ( n) = {1, 2. 1} r

Can you draw any conclusions about the time invariance of this system? 2.12 The only available information about a system consists of N input-output pairs, of signals y,(rc) = T[xj(n )], / = 1, 2........N.

138

Discrete-Time Signals and Systems

Chap. 2

(a) What is the class of input signals for which we can determ ine the output, using
the information above, if the system is known to be linear? (b) The sam e as above, if the system is known to be tim e invariant. 2.13 Show that the necessary and sufficient condition for a relaxed LTI system to be BIBO stable is y . for some constant Mn. 2.14 Show that; (a) A relaxed linear system is causal if and only if for any input x(n) such that jc(n) = 0 for n < no => y(n) = 0 (b) A relaxed LTI system is causal if and only if h(n) = 0 for n < 0 for n < no i/i()i < Mh < oo

2.15 (a) Show that for any real or complex constant a, and any finite integer num bers M and N, we have n a M a h . if * !
1 < 3

N - M + 1, (b) Show that if |o| < 1 , then

if a = 1

y v

= - i -

l - a

2.16 (a) If y(n) = x (n) * h(n). show that v =

/ .- where ^ _w (b) Compute the convolution y(n) = x(n) * h(n) of the following signals and check the correctness of the results by using the test in (a). (1) jr(n) (1,2, 4), /)(n) = (1 ,1 ,1 ,1 ,1 } (2) x(n) = {1, 2, - 1 ) , h(n) = x (n) (3) x(n) = (0,1, - 2 , 3. - 4 ) . h(n) = {, i , 1, 1} (4) jc(n )= :{1 .2 .3.4.5J.A (n) = {l) (5) x(n) = (1, -2 ,3 } , h(n ) = (0, 0 .1 .1 ,1 ,1 ) t t (6) x(n) = { 0 ,0 ,1 ,1 ,1 ,1 ), h(n) = { 1 ,-2 . 3} t t (7) jr(u) = {0,1, 4, -31. h(n) = [1,0, - 1 , -1} t t (8) = [1,1,2], h(n) = u(n) t

(9) jt(n) = [1,1. 0,1,11, h(n) = {1, - 2 , - 3 , 4} t t (10) jc(n) = (1,2, 0,2 , l}/i(n) = Jt(n) t (11) *{n) = (i)"u(n), h(rt) = ( j ) nM(n)

2.17 Compute and plot the convolutions x(n) * h(n) and h(n) *x(n) for the pairs of signals
shown in Fig. P2.17.

Chap. 2

Problems

139

b TI f
0 12 3 j x(n) n

trln)
-

0 l 2 3 4 5 b

h\n)

Iiit.
1 2 3 x(n) i

-3-2-10 I 2 3 hin)

..III!..
3 4 5 6 J (n )

II
htnl

111

!
<di

2 3 4 5

2-1 Figure P2.17

II

2 . 1 8

Determ ine and sketch the convolution y(n) of the signals

-V(/l) =
h(n) -

0. 1, 0.

0 < Ji < 6 elsewhere 2 < n < 2 elsewhere

(a) Graphicallv
(b) Analytically

2.19 Compute the convolution y(n) of the signals

x (n) =

o'".

3 < n < 5

0. 1, 0,

elsewhere 0< n< 4 elsewhere

h(n) =

2.20 Consider the following three operations. (a) Multiply the integer numbers: 131 and 122. (b) Compute the convolution of signals: {1. 3.1) * (1,2. 2}. (c) Multiply the polynomials: 1 4- 3; + z2 and 1 4- 2z 4- 2z2. (d> Repeat part (a) for the numbers 1.31 and 12.2. (e) Comment on your results. 2.21 Com pute the convolution y(n) = x ( n ) * h(n) of the following pairs of signals. (a) x(n) = a"u(n ), h(n) = b"u{n) when a ^ b and when a ~ b
(b) x ( n) = 1. 2,
n = 2, 0, 1 n = 1

. 0,

elsewhere

h ( n ) = S(n) S (n 1) + S(n 4) + S(n 5)

140

Discrete-Time Signals and System s

Chap. 2

(c) x{n ) = u(n + 1) u(n 4) <5(n 5) h(n) = [u(n +2) u(n - 3)] (3 - |n|) (d ) x ( n ) = u ( n ) - u( n - 5) h(n) = u(n 2) u(n 8) 4- u(n 11) u(n 17) 2.22 Let x(n) be the input signal to a discrete-time filter with impulse response ht(n) and let y,(n) be the corresponding output. (a) Compute and sketch x(n) and \ j ( n ) in the following cases, using the same scale in all figures. x(n) = {1,4, 2. 3, 5, 3, 3. 4. 5. 7. 6. 9} h](n) = (1,1) h2(n) = {1,2.1} ]ij(n) = {i, j) *4() = {? { j)

(b) (c) (d)

(e)

Sketch x(n), yi(n), y 2(n) on one graph and *(). y3(n), y,j(n), y.s(n) on another graph W hat is the difference between yi() and \ 2(n). and between y^(n) and y^(n)? Comment on the smoothness of v2(/?) and v4(n). Which factors affect the sm ooth ness? Compare y4(n) with ysfn). What is the difference? Can you explain it? Let h(,(n) = {^, - j } . Com pute y6<n). Sketch v(n), y 2(n), and yft(n) on the same figure and comment on the results. v (n) = ny(n 1) + jr(n) n > 0

2.23 The discrete-time system

is at rest [i.e., v( 1) = 0]. Check if the system is linear time invariant and BIBO stable. 2.24 Consider the signal y(n) = a"u(n), 0 < a < 1. (a) Show that any sequence x{n) can be decomposed as

and express ck in terms of x(n). (b) Use the properties of linearity and time invariance to express the output y(n) = T[x(n)] in term s of the input x (n) and the signal g(n) = T[y(n)], where T [ ] is an LTI system. (c) Express the impulse response h(n) = T[B{n)} in terms of g(rt). 2.25 D eterm ine the zero-input response of the system described by the second-order dif ference equation x(n) - 3y(n - 1) - 4y(n - 2) = 0 2.26 Determ ine the particular solution of the difference equation y(n) = jv(ii - 1) - y(n - 2) +x ( n) when the forcing function is x(n) = 2"u(n).

Chap. 2

Problems

141

2 2 1 D eterm ine the response yin). n > 0. of the system described by the second-order difference equation yin) - 3v(n 1) 4y(/i 2) = xin) + 2x(n 1) to the input xin) = 4"w(n). 2.28 Determ ine the impulse response of the following causal system: y(n) 3y(n 1) 4v(n 2) = jr(n) + 2x(n 1) 2.29 Let xin). A'i < n < N2 and h(n), < n < M2 be two finite-duration signals. (a) Determ ine the range L\ < n < L 2 of their convolution, in term s of N\, N2, M\ and M2. (b) Determ ine the limits of the cases of partial overlap from the left, full overlap, and partial overlap from the right. For convenience, assume that h(n) has shorter duration than jc(). (c) Illustrate the validity of your results by computing the convolution of the signals -2 < n < 4 elsewhere -1 < n < 2 elsewhere 2.30 Determ ine the impulse response and the unit step response of the systems described by the difference equation (a) yin) = ().6y(;i - 1) - ().08v(n - 2) + xin) (b) _v( ) = 0.7y(;; - 1) - 0.1 yin - 2) -f 2xin) - xin - 2) 231 Consider a svstem with impulse response A,) = H r ' ( 0. - n - 4 elsewhere

Determ ine the input xin) for 0 < n < S that will generate the output sequence v(n) = 1 1 .2 .2 .5 .3 .3 .3 .2 .1 .0 ....}

t
232 Consider the interconnection of LTI systems as shown in Fig. P2.32. (a) Express the overall impulse response in terms of h \ (n), h2(n), h^in). and h^in). (b) D eterm ine h{n) when M ) = {j. 3 . 7 } h2{n) hy(n) = (n + 1 )u(n) fi4(n) = S(n 2)

Figure P 2J2

142

Discrete-Time Signals and Systems (c) Determ ine the response of the system in part (b) if
x(n) = &(n + 2) + 3S(n - 1) - 4 S(n - 3)

Chap. 2

2 3 3 Consider the system in Fig. P2.33 with h(n) = a"u(n ), 1 < a < 1. Determ ine the response y(n) of the system to the excitation *(n) = (n + 5) u(n 10)

x(n)

Figure P233
2 3 4 Com pute and sketch the step response of the system

U_ 1

2 3 5 Determ ine the range of values of the param eter a for which the linear time-invariant

system with impulse response

(Hint: The solution can be obtained easily and quickly by applying the linearity and tim e-invariance properties to the result in Exam ple 2.3.5.)
2 3 7 D eterm ine the response of the (relaxed) system characterized by the impulse response

h(n) = ( l ) u(n) to the input signal | 1, x(n) = { ^ 10, 0 < n < 10 ~ otherwise

2 3 8 D eterm ine the response of the (relaxed) system characterized by the impulse response

h(n) = ( j ) H u{n) to the input signals (a) x(n) = 2nu(n) (b) x(n) = u ( - n )

Chap. 2

Problems

143

239 T hree systems with impulse responses h\(n) 5(n) &(n 1). h2(rt) = h( n}. and = u(n), are connected in cascade. (a) W hat is the impulse response. of the overall system? (b ) Does the order of the interconnection affect the overall system? 2.40 (a) Prove and explain graphically the difference between the relations x(n )5(n no) = o) and x(n) *&(n - n0) = x(n - nu)

(b ) Show that a discrete-time system, which is described by a convolution summation,

is LTI and relaxed, (c) W hat is the impulse response of the system described by y(n) = x(n n())? 2.41 Two signals 5(n) and u(n) are related through the following difference equations j(n) + a\ j(n 1) + N) = bi)v(n)

Design the block diagram realization of: (a) The system that generates x(n) when excited by v(n). (b ) The system that generates u(n) when excited by s(n). (c) What is the impulse response of the cascade interconnection of systems in parts (a) and (b)? 2.42 Com pute the zero-state response of the system described by the difference equation y(n ) + ^ v(n 1) = x(n ) + 2x{n 2) to the input xin) = (1.2. 3. 4, 2, 1)

T
by solving the difference equation recursively.
2 .43 Determ ine the direct form II realization for each of the following LTI systems. (a) 2v(n) + y(n 1 ) - 4 v(n 3) = x(n) + 3x(n 5) (b) y(n) = Jr(n) x(n 1) + 2x(n 2) 3x(n - 4) 2 .4 4 Consider the discrete-time system shown in Fig. P2.44.

Figure P2.44

(a) Com pute the 10 first samples of its impulse response.

(b ) Find the input-output relation.

(c) Apply the input x(n) = { 1 .1 .1 ....} and com pute the first 10 samples of the output, t

144

Discrete-Time Signals and Systems

Chap. 2

(d) Com pute the first 10 samples of the output for the input given in part (c) by using convolution. (e ) Is the system causal? Is it stable? 2 ^ 5 Consider the system described by the difference equation y(n) = ay(n - 1) + bx(n)

(a ) Determ ine b in terms of a so that

(b ) Com pute the zero-state step response s(n) of the system and choose b so that

j(oo) = 1. (c) Com pare the values of b obtained in parts (a) and (b). W hat did you notice? 2*46 A discrete-time system is realized by the structure shown in Fig. P2.46. ( a) D eterm ine the impulse response. (b) Determ ine a realization for its inverse system, that is, the system which produces x( n) as an output when y(n) is used as an input.
x in )

-o
0.8

v(n)

Figure P2.46

2 .4 7 Consider the discrete-time system shown in Fig. P2.47.

v(n)

Figure P2^f7

(a) Com pute the first six values of the impulse response of the system.
(b ) Com pute the first six values of the zero-state step response of the system.

( c ) Determ ine an analytical expression for the impulse response of the system.
1 4 8 D eterm ine and sketch the impulse response of the following systems for n 0, 1........9. (a ) Fig. P2.48(a). (b ) Fig. P 2 .4 8 (b ). ( c ) Fig. P2.48(c).

Chap. 2

Problems

145

Cc)

Figure P2.48

(d) Classify the systems above as FIR or IIR.

(e) Find an explicit expression for the impulse response of the system in part (c).

2.49 Consider the systems shown in Fig. P2.49. (a) Determ ine and sketch their impulse responses /i|(n), h2(n ), and h3(n). (b) Is it possible to choose the coefficients of these systems in such a way that h\(n) = h2(n) = h3(n) 2.50 Consider the system shown in Fig. P2.50. (a) Determ ine its impulse response h(n). (b) Show that h(n) is equal to the convolution of the following signals. h] (n) = 6(n) + 6(n - 1) M " ) = (^)"u(n)

146

Discrete-Time Signals and Systems

Chap. 2

v (n )

yin)

2.51 Com pute the sketch the convolution y,(n) and correlation r,(n) sequences for the following pair of signals and comment on the results obtained. (a) *,{) = (1.2.4) A ,(n )= (1 ,1 .1 .1 . If
t t

(b) x2(n) = (0,1. - 2 . 3, - 4 ]

t

h2(n) = U. 1. 2 . 1 , i}
t

(c)

jt j( b )

= (1.2. 3, 4}
t

A3 (u )

= (4. 3. 2, 1)
t

(d) x4(n) = {1. 2, 3,4)

t

hA(n) = (1.2.3. 4)
t

2.52 The zero-state response of a causal LTI system to the input x ( n ) = {1,3, 3,1) is y(n) = (1,4, 6 ,4 ,1 ). Determ ine its impulse response. t

Chap. 2

Problems

147

2.53 Prove by direct substitution the equivalence of equations (2.5.9) and (2.5.10), which describe the direct form II structure, to the relation (2.5.6), which describes the direct form I structure. 2.54 D eterm ine the response y(n). n > 0 of the system described by the second-order difference equation
y ( n) 4 y(n - 1) + 4v(/i 2) = x( n) x( n 1)

when the input is x(n) = (-l)" u (n ) and the initial conditions are v(1) = y ( -2 ) = 0. 2.55 D eterm ine the impulse response h(n) for the system described by the second-order difference equation
v(ni 4v(;i 1 > + 4y(n 2) = x(r?) x(n 1)

2.56 Show that any discrete-tim e signal x(n) can be expressed as c(n) = [.v(A') x(k 1)]u(n - k)

where (/i - k ) is a unit step delayed by k units in time, that is,

u(/i k) = .

1. I 0,

n >k otherwise

2.57 Show that the output of an LTI system can be expressed in term s of its unit step response v(n) as follows. i'(n) = y " ' [,v(A :) x(k 1)]jr(fl k)

[x(AQ - x ( k - l) ] s ( /i - k)

c 2.58 Com pute the correlation sequences rIX(l) and rtv(l) for the following signal sequences.
j _ P nu - N < n < n {, + N

I 0,
f 1.
v(n) = 1

otherwise
-N <n < N

1 0,

otherwise

2.59 D eterm ine the autocorrelation sequences of the following signals. (a) x(n) = {1. 2.1.1)
t

(b) v(n) = i l . 1.2.1}

t

W hat is your conclusion? 2.60 W hat is the normalized autocorrelation sequence of the signal x(n) given by 1, x(n) = , 0, -jV < n < N otherwise

148

Discrete-Time Signals and Systems

Chap. 2

2.61 An audio signal j(r) generated by a loudspeaker is reflected at two different walls with reflection coefficients r ] and r2. The signal *(/) recorded by a microphone close to the loudspeaker, after sampling, is x (n) = s(n) + rxs(n k\) + r2s(n k2) where kt and k2 are the delays of the two echoes. (a) Determ ine the autocorrelation rzx(I) of the signal x(n). Can we obtain ri, r2, k\, and k2 by observing r,s (1)1 (c) W hat happens if r2 = 0?

(b )

2.62* Time-delay estimation in radar Let xa(t) be the transm itted signal and yfl(r) be the received signal in a radar system, where y(r) = axa(t - id) + vu(f) and va(t ) is additive random noise. The signals xa(t) and >(/) are sampled in the receiver, according to the sampling theorem , and are processed digitally to deter mine the time delay and hence the distance of the object. The resulting discrete-time signals are jr(n) = xa(nT) y(n) = y(nT) = axu(nT - DT) + vu(nT) = ax(n D) + u(n) (a) Explain how we can measure the delay D by computing the crosscorrelation r*,.(/). (b) Let x(n) be the 13-point Barker sequence
X

(n) = (+ 1,+1,+1,+1,+1, -1, -l.+l.+l, -1,+1. -1,+1)

and u(n) be a Gaussian random sequence with zero mean and variance a 2 = 0.01. Write a program that generates the sequence v(n), 0 < n < 199 for a = 0.9 and D = 20. Plot the signals jt(), y(n), 0 < n < 199. (c) Compute and plot the crosscorrelation rT V (/), 0 < / < 59. Use the plot to estimate the value of the delay D. (d) Repeat parts (b) and (c) for a 2 = 0.1 and a 2 = 1. <e) Repeat parts (b) and (c) for the signal sequence jt(n) = j _ l , _ l , - l , + i , + i , + i . + i , - i , + 1 . - l . + l . + l , - 1 ,- 1 ,+ 1 } which is obtained from the four-stage feedback shift register shown in Fig. P2.62,

Figure P2.61 register.

Linear feedback shift

Chap. 2

Problems

149

Note that x(n) is just one period of the periodic sequence obtained from the feedback shift register. (f) Repeat parts (b) and (c) for a sequence of period N 27 1, which is obtained from a seven-stage feedback shift register. Table 2.3 gives the stages connected to the modulo-2 adder for (maximal-length) shift-register sequences of length
N =2" TABLE 2.3 SHIFT-REGISTER CONNECTIONS FOR GENERATING MAXIMAL-LENGTH SEQUENCES m Stages Connected to Modu)o-2 Adder

1
2 3 4 5 6 7 H y id li 12 13 14 15 16 17

1
1. 2 1. 3 1. 4 S. 4 L6 1. 7 1, 5, 6. 7 1.6 1. K 1. 10 i. 7. y, 12 1. H), 11. 13 1. 5. 9. 14 1. 15 1. 5. 14, 16 1. 15

2.63* Implementation o f L T I systems by the difference equation

Consider the recursive discrete-time system described

_v(n) = U\ v( n 1) a;v (n 2) 4- b^x(rt)

where a\ - 0.8, u? = 0.64. and b() = 0.866. (a) Write a program to compute and plot the impulse response h{n ) of the system for 0 < n < 49. (b) Write a program to com pute and plot the zero-state step response s(n) of the system for 0 < n < 100. (c) Define an F IR system with impulse response ^ fir (h ) given by , , .
1 h(n),

10.

0 < n < 19 elsewhere

where h(n) is the impulse response computed in part (a). W rite a program to compute and plot its step response. (d) Com pare the results obtained in parts (b) and (c) and explain their similarities and differences.

150

Discrete-Time Signals and Systems

Chap. 2

2jS4* W rite a com puter program that computes the overall impulse response h(n) of the sys
tem shown in Fig. P2.64 for 0 < n < 99. The systems TU T2, T 3, and % are specified by Ti : hi(n) = {1 . 5 . t g.

55)

T2 : h 2(n) = {1,1,1,1,11 t Ts : + 5*(" - 1) + - 2)

T4 : y(n) = 0 .9y(n 1) 0.81y{n 2) + v(n) + u(n 1) Plot h(n) for 0 < n < 99.

Figure P2.64

The Z -Transform and Its Application to the Analysis of LTI Systems

T ra n s fo rm te c h n iq u e s a re an im p o rta n t tool in the analysis o f signals a n d lin e a r tim e -in v a ria n t (L T I) system s. In this c h a p te r w e in tro d u c e th e ^ -tran sfo rm , d ev elo p its p ro p e rtie s , and d e m o n s tra te its im p o rta n c e in th e analysis an d c h a ra c te riz a tio n o f lin ear tim e -in v a ria n t system s. T h e : -tra n sfo rm plays th e sam e role in th e analysis o f d isc re te -tim e signals an d L T I sy stem s as th e L ap lace tran sfo rm d o e s in th e analysis o f c o n tin u o u s-tim e signals a n d L T I svstem s. F o r ex am p le, w e sh a d see th a t in th e ^ -d o m ain (com plex z -p lan e) th e c o n v o lu tio n of tw o tim e-d o m ain signals is e q u iv a le n t to m u ltip lic atio n o f th e ir c o rre sp o n d in g ^ -tran sfo rm s. T h is p ro p e rty g reatly sim plifies the analysis o f th e re sp o n s e o f an L TI system to v ario u s signals. In ad d itio n , th e c-tran sfo rm p ro v id es us w ith a m ean s of ch a ra c te riz in g an L T I system , a n d its resp o n se to v ario u s signals, by its p o le - z e r o locations. W e b eg in th is c h a p te r by defining th e c-tran sfo rm . Its im p o rta n t p ro p e rtie s a re p re s e n te d in S ectio n 3.2. In S ection 3.3 the tra n sfo rm is u se d to ch aracterize signals in te rm s o f th e ir p o le - z e r o p a tte rn s. S ection 3.4 d escrib es m e th o d s fo r in v ertin g th e z-tra n sfo rm o f a signal so as to o b ta in th e tim e -d o m a in re p re s e n ta tio n o f th e signal. T h e o n e-sid ed :-tra n s fo rm is tr e a te d in S ectio n 3.5 a n d used to solve lin e a r d ifferen ce e q u a tio n s w ith n o n z e ro in itial co n d itio n s. T h e c h a p te r c o n clu d es w ith a d iscussion o n th e use of th e z -tra n sfo rm in th e analysis o f L TI system s.

3.1 THE Z-TRANSFORM

In th is sectio n w e in tro d u c e th e z -tra n sfo rm of a d isc re te -tim e signal, in v estig ate its c o n v e rg e n c e p ro p e rtie s , an d briefly discuss th e in v erse z-tran sfo rm .
151

152

The z-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

3.1.1 T h e D ire c t z - T r a n s f o r m T h e ^ -tran sfo rm o f a d isc re te -tim e signal jt(fi) is d efin ed as th e p o w e r series
OC

X(z) =

n= oo

x { n ) z ~'

(3-u )

w h ere z is a co m p lex v ariab le. T h e re la tio n (3.1.1) is so m e tim e s called th e direct z - t ransform b ecau se it tra n sfo rm s th e tim e -d o m a in signal x ( n ) in to its co m p lex p la n e re p re s e n ta tio n X (z). T h e in v erse p ro c e d u re [i.e., o b ta in in g x ( n ) fro m X (z)] is called th e i nverse z- t ransf orm a n d is e x a m in e d briefly in S e ctio n 3.1.2 a n d in m o re d etail in S ectio n 3.4. F o r c o n v en ien ce , th e z -tra n sfo rm o f a signal x ( n ) is d e n o te d by X (z) - Z U ( )} w h ereas th e re la tio n sh ip b e tw e e n x ( n ) a n d X ( z ) is in d ic a te d by
jc()

(3.1.2)

< -^ * (;)

(3.1.3)

S ince th e z-tra n sfo rm is an infinite p o w e r se ries, it exists only fo r th o se v alu es of z fo r w hich th is se ries co n v erg es. T h e region o f conver gence ( R O C ) o f X (z) is th e set o f all v alu es o f z fo r w hich X ( z ) a tta in s a finite value. T h u s an y tim e w e cite a z-tra n sfo rm w e sh o u ld also in d icate its R O C . W e illu strate th e se co n c e p ts by so m e sim ple ex am p les. Example 3.1.1 D eterm ine the ^-transforms of the following finite-duration signals. (a) jf,(n) = (1 ,2 .5 .7 ,0 ,1 }
(b) jf,() = ( I . 2 . 5 . 7 . 0 . 1)
t

(c) jc3(n) = (0 ,0 ,1 ,2 , 5, 7,0.1} (d) *4(h) = (2 .4 ,5 .7 .0 ,1 )

t

(e) x j ( n ) = S( n)

(f) x$(n) = <S(n k), k > 0 (g) x j ( n ) = &(n + k) , k > 0 Solution From definition (3.1.1), we have

(a) X](z) = 1 + 2z~' + 5z~2 + 7 z '3 + z~5, ROC: entire z-plane except z = 0 (b) X 2(z ) = z2 + 2z + 5 + 7c-1 + z-3, ROC: entire z-plane except z = 0 and z = oo (c) Xj(z) = z~2 + 2z-3 + 5z-4 + 7z-5 -I- z-7, ROC: entire z-plane except z = 0 (d) X4(z) = 2z2 -I- 4z -I- 5 4- 7z_1 -I- z-3, ROC: entire z-plane except z = 0 and z = oo (e) X;(z) = l[i.e S(n) * 1], ROC: entire z-plane (f) Xb(z) = z_ t[i.e &(n k) z_t], k > 0, ROC: entire z-plane except z = 0 (g) Xy(z) = zk[i.e &(n + k) z*], k > 0, ROC: entire z-plane except z = oo

Sec. 3.1

The z-Transform

153

F ro m th is ex a m p le it is easily se en th a t th e R O C o f a f i ni t e-durat i on signal is th e e n tire ;- p la n e , e x cep t p ossibly th e p o in ts z = 0 a n d /o r z oo. T h e se p o in ts a re e x c lu d ed , b e c a u s e z t (k > 0} b eco m es u n b o u n d e d fo r z = oc an d z ~ k01 > 0) b ec o m e s u n b o u n d e d fo r z = 0. F ro m a m a th e m a tic a l p o in t of view th e z-tra n sfo rm is sim p ly an a lte rn a tiv e r e p re s e n ta tio n o f a signal. T h is is nicely illu stra te d in E x a m p le 3.1.1, w h e re w e see th a t th e co effic ie n t o f z~", in a given tra n sfo rm , is th e v a lu e o f th e signal at tim e n. In o th e r w o rd s, th e e x p o n e n t o f z co n tain s th e tim e in fo rm a tio n w e n eed to id en tify th e sa m p le s o f th e signal. In m an y cases w e can ex p ress th e sum of th e finite o r infinite se rie s fo r th e z-tra n s fo rm in a c lo sed -fo rm e x p ressio n . In such cases th e z -tra n s fo rm o ffers a co m p act a lte rn a tiv e r e p re s e n ta tio n of th e signal. Example 3.1.2 D eterm ine the z-transform of the signal Jf(n) = (5 Solution The signal jc(n) consists of an infinite num ber of nonzero values

x(n)= ( l . a u ^ . U )'1 ....

The z-transform of x(n) is the infinite power series
X( z ) = 1 + U ' + ( ^) 2z - 2 + (| ) "z"" + ---

f t = < )

nail

This is an infinite geometric series. We recall that

1 + A + ,42 + A3 - t - - - - = if | A 1 < 1

1 A

Consequently, for | 1

< 1, or equivalently* for \z\ > X(z) =

J jZ

X(z) converges to

ROC: (zl > |

We see that in this case, the z-transform provides a compact alternative representation of the signal x(n). L e t us e x p ress th e co m p lex v a ria b le z in p o la r fo rm as z = r e }6 w h ere r = |z| a n d 6 = i^z- T h e n X ( z ) can be e x p ressed as
OC

(3.1.4)

X ( z ) \ t- r ' = Y x ( n ) r ~ ne - j en n*-00

154

The z-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

In th e R O C o f X U ). |X ( ;) | < oc. B ut

(3.1.5) 5 \ M n ) r - ne - ' * n \ = \ x( n) r ~n\

H e n c e |X (z)| is finite if th e se q u en ce x ( n ) r ~ ' is a b s o lu te ly su m m a b le . T h e p ro b le m o f finding th e R O C fo r X ( z ) is e q u iv a le n t to d e te rm in in g the ra n g e o f v alu es o f r fo r w hich th e se q u e n c e x ( n ) r ~ " is a b so lu te ly su m m ab le. T o e la b o ra te , let us e x p ress (3.1.5) as

(3.1.6)

If X ( z ) co n v erg es in so m e region of the co m p lex p la n e , b o th su m m a tio n s in (3.1.6) m u st be finite in th a t region. If the first sum in (3.1.6) co n v erg es, th e re m ust exist v alu es o f r sm all e n o u g h such th at the p ro d u c t se q u e n c e x ( n) r" . 1 < /; < oc, is a b s o lu te ly su m m ab le. T h e re fo re , the R O C for th e first sum co n sists o f all p o in ts in a circle of so m e rad iu s r^, w here / ] < oc, as illu stra te d in Fig. 3.1a. O n the o th e r h a n d , if th e seco nd sum in (3.1.6) co n v erg es, th e re m u st exist v alu es o f r larg e en o u g h such th a t th e p ro d u c t se q u e n c e x ( n ) / r " . 0 < n < oc, is a b so lu te ly su m m ab le. H en ce th e R O C fo r the se co n d sum in (3.1.6) co n sists o f all p o in ts o u tsid e a circle o f rad iu s r > r2. as illu stra te d in Fig. 3.1b. Since th e co n v erg en ce of X (c) re q u ire s th a t b o th sum s in (3.1.6) b e finite, it follow s th a t th e R O C o f X ( z ) is g en erally specified as th e a n n u la r region in th e ;- p la n e , r: < r < r\. w hich is the co m m o n reg io n w h e re b o th su m s are finite. T his reg io n is illu stra te d in Fig. 3.1c. O n th e o th e r h a n d , if > r\, th e r e is no co m m o n reg io n o f co n v erg en ce fo r th e tw o sum s an d h en ce X ( ;) d o es n o t exist. T h e follow ing ex am p les illu strate th e se im p o rta n t co n cep ts. Example 3.1.3 Determ ine the e-transform of the signal

Solution

From the definition (3.1.1) we have

If \az 11 < 1 or equivalently, |z| > |a |, this power series converges to 1/(1 - a ; -1).

Sec. 3.1

The z-Transform

155

Im(z)

Im(;)

Im(z)

Figure 3.1 R egion of convergence for X (z) and its corresponding causal and anticausa! components.

156

The z-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

lm(;)

Figure 3.2 The exponential signal xi n) = txnu{n) (a), and the ROC of its transform (b).

Thus we have the z-transform pair x(n) =a" u( n ) X (;) = 1

1 - a ;-1

ROC: |z| > |cr|

(3.1.7)

The R O C is the exterior of a circle having radius |a |. Figure 3.2 shows a graph of the signal .x(n) and its corresponding ROC. Note that, in general, or need not be real. If we set or = 1 in (3.1.7), we obtain the z-transform of the unit step signal x(n) - u(n ) Example 3.1.4 Determ ine the z-transform of the signal x(n) = a"u( n 1) = Solution 0, n >0 n < -1
oc

X(z) =

1 1

ROC: |zl > 1

(3.1.6

From the definition (3.1.1) we have

-I

no c

/= i

where / = n. Using the formula A + A: + A3 + = A(1 + A + A2 + )= when | > 1f < 1 gives * ( ;) = - 1 1 a ~ lz l - a ; '1 A
1 - A

provided that |cr_1z| < 1 or, equivalently, |z| < jar j. Thus 1 (3.1.9) ROC: [z| < |a | 1-azThe R O C is now the interior of a circle having radius |a|. This is shown in Fig. 3.3. x(n) = a"u( n 1) X(Z) = -

Sec. 3.1

The z-Transform
Im(c)

157

Figure 3.3 Anticausal signal jt(h) = -crnu( n - 1) (a), and the ROC of its transform (b).

E x a m p le s 3.1.3 an d 3.1.4 illu stra te tw o very im p o rta n t issues. T h e first c o n cern s th e u n iq u e n e s s o f th e "-tran sfo rm . F ro m (3.1.7) an d (3.1.9) we see that th e cau sal signal a nu ( n ) a n d th e an ticau sal signal a " u ( n 1) have id en tical clo sed -fo rm e x p re ssio n s fo r th e ^ -tran sfo rm , th a t is,
Z ( o " w ( ) ) = Z { a nu ( n 1) ) ------ --------

1 a c ' 1

T h is im p lies th a t a clo sed -fo rm e x p ressio n fo r th e z-tra n sfo rm d o e s n o t u n iq u ely specify th e signal in th e tim e do m ain . T h e am b ig u ity can b e reso lv ed only if in a d d itio n to th e c lo sed -fo rm ex p ressio n , th e R O C is specified. In su m m ary , a discrete-time signal x ( n ) is uni quel y d et er mi ned b y its z- t rans f orm A' (;) a n d the region o f conver gence o f X( z ) . In this te x t th e te rm z -tra n s fo rm " is u se d to re fe r to b o th th e clo sed -fo rm e x p ressio n an d th e c o rre sp o n d in g R O C . E x a m p le 3.1.3 also illu stra te s th e p o in t th a t the R O C o f a causal signal is the exterior o f a circle o f s o m e radius r 2 whi l e the R O C o f an ant icausal signal is the interior o f a circle o f s o m e radi us rj. T h e fo llow ing e x am p le co n sid ers a se q u en ce th a t is n o n z e ro for 00 < n < 00. Example 3.1.5 D eterm ine the z-transform of the signal x (n) = a"u(n) + bnu( n 1) Solution From definition (3.1.1) we have X(z) =
n=0 n= oc

b"z " =
n=0

+ Y ib -'z)1
i= l

The first pow er series converges if locz-11 < 1 or |z| > |a |. The second power series converges if \b~xz\ < 1 or |z| < |6j. In determ ining the convergence of XCz), we consider two different cases.

158

The z-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

Case 1 |b| < |a |: In this case the two R O C above do not overlap, as shown in Fig. 3.4(a). Consequently, we cannot find values of z for which both power series converge simultaneously. Clearly, in this case, X(c) does not exist. Case 2 |f>| > |a |: In this case there is a ring in the z-plane where both power series converge simultaneously, as shown in Fig. 3.4(b). Then we obtain X(z) = 1 ccz 1 1 bz 1 b a

(3.1.10)

a + b z abz~l The R O C of X(z) ts |cr| < \z\ < \b\. T h is e x am p le show s th a t i f there is a R O C f o r an infinite durat i on two-si ded signal, it is a ring ( annul ar region) in the z-plane. F ro m E x a m p le s 3.1.1, 3.1.3, 3.1.4, an d 3.1.5. w e see th a t th e R O C of a signal d e p e n d s o n b o th its d u ra tio n (finite or in fin ite) an d o n w h e th e r it is cau sal, a n tic a u sa l, o r tw o -sid ed . T h e se facts are su m m a riz e d in T a b le 3.1. O n e special case of a tw o -sid ed signal is a signal th a t h as infinite d u ra tio n on th e rig h t sid e b u t n o t on th e left [i.e., x( n ) = 0 fo r n < (l < 0], A sec on d case is a signal th a t has infinite d u ra tio n o n th e left side b u t n o t on the

plane

Ifcl < tot

X(z) does not exisl

Im(;)
krl < IA I

ReU)

ROC for X(z) Figure 3.4 R O C fo r z-transform in E xam ple 3.1.5.

Sec. 3.1

The ^-Transform
TABLE 3.1
CH A R A C TE R IS TIC FAM ILIES O F SIGN ALS W IT H T H E IR

159

C O R R E S P O N D IN G ROC

Signal

ROC Finite-Duration Signals

Two-sided

. . TTT l i t ,
0 Causal

Jnfinite-Duration

l l T t*

Anltcausal

T ]T 1
Two-sided

I . I t.....
rig h t [i.e., x{ n ) = 0 fo r n > n\ > 0]. A th ird sp ecial case is a signal th a t has finite d u ra tio n o n b o th th e left a n d rig h t sides [i.e., x ( n ) = 0 fo r n < no < 0 a n d n > n\ > 0]. T h e se ty p e s o f signals a re so m e tim e s c alled right-sided, left sided, a n d finite-d uration two-sided, signals, resp ec tiv ely . T h e d e te rm in a tio n o f th e R O C fo r th e s e th r e e ty p es o f signals is left as an e x ercise fo r th e r e a d e r (P r o b lem 3.5). F in ally , w e n o te th a t th e z -tra n sfo rm d efin ed b y (3.1.1) is so m e tim e s re fe rre d to as th e tw o-sided o r bilateral z-transform , to d istin g u ish it fro m th e o ne-sided o r

160

The z-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

unilateral z- t ransf orm given by X~(z) = ^ * ( ) z ~ " (3.1.11)

T h e o n e-sid ed z -tra n sfo rm is ex am in ed in S ection 3.5. In this tex t w e use the ex p ressio n z-tra n sfo rm exclusively to m ean th e tw o -sid ed z -tra n sfo rm d efined by (3.1.1). T h e te rm 'tw o -sid e d ' will b e u se d oniy in cases w h ere w e w an t to resolve any am b ig u ities. C learly , if x ( n ) is causal [i.e., * ( 72) = 0 for n < 0], th e o n e-sid ed and tw o -sid ed z-tran sfo rm s a re eq u iv a le n t. In any o th e r case, th e y a re d ifferen t.

3.1.2 The Inverse z-Transform

O ften , we h av e th e z-tra n sfo rm X ( z ) of a signal a n d w e m ust d e te rm in e th e signal se q u en ce. T h e p ro c e d u re fo r tra n sfo rm in g from th e z-d o m ain to th e tim e dom ain is called th e inverse z- t ransform. A n in v ersio n fo rm u la for o b ta in in g x( n) from X (z) can b e d eriv ed by using th e Cau c hy integral t h e o r e m , w hich is an im p o rta n t th e o re m in th e th e o ry o f co m p lex variables. T o b egin, we h av e th e z-tra n sfo rm defin ed by (3.1.1) as

S u p p o se th a t we m u ltip ly b o th sides o f (3.1.12) by z"~' an d in te g ra te both sides o v er a closed c o n to u r w ithin the R O C o f X ( z ) w hich en clo ses th e origin. Such a c o n to u r is illu stra te d in Fig. 3.5. T h u s we have (3.1.13) w h e re C d e n o te s th e clo sed c o n to u r in th e R O C o f A'(z). ta k e n in a c o u n te rc lo c k w ise d irectio n . Since th e se rie s c o n v erg es on th is c o n to u r, w e can in te rc h a n g e th e o rd e r o f in te g ra tio n an d su m m atio n o n th e rig h t-h a n d side o f (3,1.13). T h u s
im (o

Figure 3.5 (3.1.13).

Contour C for integral in

Sec. 3.2

Properties of the z-Transform

161

(3.1.13) b eco m es x ( z ) z n- ' d z = t= 00 x ( k ) ( h z n- l~kd z * (3.1.14)

N o w w e can in v o k e th e C a u ch y in te g ra l th e o re m , w hich s ta te s th a t

ftM5)
w h ere C is an y c o n to u r th a t en clo ses th e origin. B y ap p ly in g (3.1.15), th e righth a n d side o f (3.1.14) re d u c e s to 2 n j x ( n ) a n d h en c e th e d e sire d in v e rsio n fo rm u la x ( n ) = ^ - ^ X ( z ) z n_I d z (3.1.16)

A lth o u g h th e c o n to u r in te g ra l in (3.1.16) p ro v id es th e d e s ire d in v ersio n fo r m u la fo r d e te rm in in g th e se q u e n c e * ( ) fro m th e z-tra n sfo rm , w e sh all n o t use (3.1.16) d irectly in o u r e v a lu a tio n o f in v erse z -tran sfo rm s. In o u r tre a tm e n t w e d eal w ith signals a n d sy stem s in th e z-d o m ain w hich h av e ra tio n a l z -tra n s fo rm s (i.e., ztra n sfo rm s th a t a re a ra tio o f tw o p o ly n o m ials). F o r such z -tra n sfo rm s w e d e v e lo p a s im p ler m e th o d fo r in v ersio n th a t ste m s from (3.1.16) and e m p lo y s a ta b le lo o k u p .

3.2 PROPERTIES OF THE Z-TRANSFORM

T h e z -tra n sfo rm is a v ery p o w erfu l to o l fo r th e stu d y o f d isc re te -tim e signals an d system s. T h e p o w e r o f th is tra n sfo rm is a c o n s eq u en ce o f so m e v ery im p o rta n t p ro p e rtie s th a t th e tra n sfo rm possesses. In th is sectio n w e e x am in e som e o f th e se p ro p e rtie s . In th e tre a tm e n t th a t follow s, it sh o u ld b e re m e m b e re d th a t w h e n w e c o m b in e sev eral z -tra n sfo rm s, th e R O C o f th e o v erall tra n sfo rm is, at least, th e in te rse c tio n o f th e R O C o f th e in d iv id u al tra n sfo rm s. T h is will b eco m e m o re a p p a re n t la te r, w h en w e discu ss specific ex am p les.

Linearity.
an d

If Jt,(n) Xi ( z )

x 2(n) < -U X 2(z) th e n x( n) = a \ x \ ( n ) + a 2x 2(n) X( z ) = a j Xj ( z ) -h o2X 2(z) (3.2.1)

fo r an y c o n s ta n ts a i a n d a 2. T h e p ro o f o f th is p ro p e rty follow s im m e d ia te ly from th e d efin itio n o f lin e a rity a n d is left as an ex ercise fo r th e re a d e r. T h e lin e a rity p ro p e rty can easily be g e n e ra liz e d fo r an a r b itra r y n u m b e r o f signals. B asically , it im p lies th a t th e z -tra n sfo rm o f a lin e a r c o m b in a tio n o f signals is th e sa m e lin e a r c o m b in a tio n o f th e ir z-tran sfo rm s. T h u s th e lin e a rity p ro p e rty h e lp s u s to find th e z -tra n s fo rm o f a signal by ex p ressin g th e signal as a su m o f e le m e n ta ry signals, fo r ea c h o f w hich, th e z -tra n sfo rm is a lre a d y k n o w n .

162

The z-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

Example 3.2.1 Determ ine the z-transform and the R O C of the signal *() = [3(2") -4 (3 ")] ( ) Solution If we define the signals jti() = 2"u(n) and x2(n) = 3"u(n) then
jc()

can be written as x(n) = 3xi(n) 4 x2(n)

According to (3.2.1), its z-transform is X(;) = 3 X ,( c ) - 4 X 2(z) From (3.1.7) we recall that a'u(n ) 1 - az ROC: |z| > |a| (3.2.2)

By setting a = 2 and or = 3 in (3.2.2). we obtain Ai(n) = 2u(n) x2(n) = 3nu(ri) X,(z) = ^ X 2(z) ~ ^ __ , ROC: |zj > 2 ROC: |zl > 3

The intersection of the ROC of X,(z) and X2(:) is |z| > 3. Thus the overall transform X (z) is ROC: |;| > 3

Example 3.2.2 Determ ine the z-transform of the signals (a) x(n) = (COSoon)u(n) (b) x(n) = (sin w^n)u(n) Solution (a) By using E uler's identity, the signal .t(n) can be expressed as x (n) = (cosa>on)u(n) = \ e Jaln"u(.n) + ^e~JU Jn u(n) Thus (3.2.1) implies that
X( z ) = \ Z { e ^ u ( n ) ) + ^ { g - ^ u i n ) }

Sec. 3.2

Properties of the z-Transform

163

If we set a = e^^O ori = \ei< u\ = 1) in (3.2.2), we obtain eJU V "u(fi) and e - ,wnnu(n) ^ Thus X(z) = -------- ----------h I ------- - ------r 2 1 ^ f z -1 2 1 e~J^lz~I RO C : |;1 > 1
-------- T T T

---------------

R O C : |z[ > 1

1-

R O C : l; l >

A fter some simple algebraic manipulations we obtain the desired result, namely, (coscO(}f?)u(n) < - - ;---------------- 1 - 2z ~1cos ton + (b) From E ulers identity,
x(n) = (sino<o)(n) = ^-\eiun" u(n) e~,w""u()]
2j : 1 C_ ! COSW()

ROC: |-1 > 1

L r2.j)

,, ,, ,

Thus X(-) = ^ - f - ------1 ------- - ------- ------- - | 2j \ 1 r _1 1 and finally. z sm O H ) (sin coi,n)u(n) x - ^ ----- - -----------------1 - 2z_1 cos wo + z~ROC: |;| > 1 (3.2.4) ROC: |:| > 1

Time shifting.

If x( n) X (c)

th e n x{n - k ) z ~ kX{ z ) (3.2.5)

T h e R O C o f z ~kX ( z ) is th e sam e as th a t o f X ( z ) e x c e p t fo r j = 0 if k > 0 and z = oo if k < 0. T h e p r o o f o f th is p ro p e rty follow s im m e d ia te ly fro m the definition o f th e ^ -tra n sfo rm given in (3.1.1) T h e p r o p e rtie s o f lin e a rity an d tim e shifting a re th e k ey fe a tu re s th a t m ake th e z -tra n s fo rm e x tre m e ly u se fu l fo r th e analysis o f d isc re te -tim e L T I system s. Exam ple 3*23 By applying the time-shifting property, determ ine the z-transform of the signals .T 2(n,i and xi(n) in Exam ple 3.1.1 from the Z'transform of Jti(n). Solution

It can easily be seen that

x 2( n ) = * ]( + 2)

164
and

The ^-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

*i(n) = j:i(n - 2) Thus from (3.2.5) we obtain X2(z) = z2X, (z) = r + 2; + 5 + 7z ' 1 + z3 and X 3(z) = r 2X,(z) = z"2 + 2z"3 + 5z-4 + 7z"s + z-7 Note that because of the multiplication by z2, the R O C of X2(z) does not include the point z = oc, even if it is contained in the R O C of A^i(z). E x am p le 3.2.3 p ro v id es a d d itio n a l insight in u n d e rs ta n d in g th e m e an in g of th e shiftin g p ro p e rty . In d e e d , if we recall th a t th e coefficient o f z ~ n is th e sam ple v alu e at tim e n, it is im m e d ia te ly se en th a t d elay in g a signal by k( k > 0) sam ples [i.e., x ( n ) x ( n A')] c o rre sp o n d s to m u ltip ly in g all te rm s o f th e z -tra n sfo rm by z~ k. T h e co efficien t o f z~" b ec o m e s th e coefficient o f ~~tn+k). Example 3.2.4 Determ ine the transform of the signal
f 1, ' " H o . 0 < n < jV - 1 e lse w h e r e , <316>

Solution We can determ ine the z-transform of this signal by using the definition (3.1.1). Indeed, * -i f N, if z = l X(z) = ^ l z ^ = l + ; - ' + - - - + z - (A f- l l = l - r * -f . , , (3.2.7) * = < ' I 1 - z-1' 1 Since .v(n) has finite duration, its R O C is the entire z-plane, except z 0. Let us also derive this transform by using the linearity and time shifting prop erties. Note that x in) can be expressed in terms of two unit step signals x(n) = u(n) u(n N) By using (3.2.1) and (3.2.5) we have X(z) = Z{u(n )} - Z{u(n - N)) = (1 - z"*')Z{u ()} However, from (3.1.8) we have Z[u(n)) = ^ ROC: jzl > 1 (3.2.8)

which, when combined with (3.2.8), leads to (3.2.7). E x a m p le 3.2,4 h e lp s to clarify a v ery im p o rta n t issue re g a rd in g th e R O C o f th e c o m b in a tio n o f se v era l z-tran sfo rm s. If th e lin ear c o m b in a tio n of several sig n als h as finite d u ra tio n , th e R O C o f its z -tra n sfo rm is exclusively d ic ta te d by the fin ite -d u ra tio n n a tu re o f th is signal, n o t by th e R O C o f th e in d iv id u al tran sfo rm s.

Scaling in the z-domain.

x{n)

If
X{z )

ROC: ri < [z| < r2

Sec. 3.2

Properties of the z-Transform

165

th en a nx ( n ) X (c ~ z) R O C : \a\r\ < |z| < \a\r2 (3.2.9) for a n y c o n s ta n t a, re a l o r com plex. Proof. F ro m th e d efin itio n (3.1.1)
OC OC

Z{a" x( n) } =

a nx ( n ) z " =

x( n) ( a z) n

\o \r i < I;I < \a\r2 T o b e tte r u n d e rsta n d th e m ean in g a n d im plicatio n s o f th e scaling p ro p e rty , w e e x p re ss a a n d z in p o la r form as a = rae->a", z = r e i< 0, a n d w e in tro d u c e a new co m p lex v a ria b le w = a~^z- T h u s Z { x ( n ) ) = X ( z ) an d Z{a" x( n) } = X (if). It can easily b e seen th a t

T h is ch a n g e o f v a ria b le s re su lts in e ith e r sh rin k in g (if r 0 > 1) o r e x p a n d in g (if r 0 < 1) th e z -p lan e in co m b in a tio n w ith a ro ta tio n (if too # 2i-jr ) o f th e z-p lan e (see Fig. 3.6). T h is e x p lain s w hy w e h av e a ch an g e in th e R O C o f th e n ew tra n sfo rm w h e re |a | < 1. T h e case \a\ = 1, th a t is, a = e^w" is o f special in te re st b e c a u se it c o rre sp o n d s o n ly to ro ta tio n o f th e z-p lan e. Exam ple 3.2.5 Determ ine the z-transforms of the signals (a) x(n) = a"(cosu\in)u(n) (b) x(n) = tf"(sin ion)u(n) r-planc
Im(-) w-plane

(W -C^o 0 Re(z) 0
Re(w)

Figure 3.6
o _ 1 Z,

a roe^.

Mapping of the r-plane to the u -plane via the transformation ui =

166

The z-Transform and Its Application to the Analysis of LTI S ystem s Solution (a) From (3.2.3) and (3.2.9) we easily obtain 1 - a ; -1 cos wn
a (C O S a i |, ) u ( r t ) *-------* --------- ---------- ;--------------------------

Chap. 3

1 l a z r 1 co s ton -f a

(3.2.10)

(b) Similarly, (3.2.4) and (3.2.9) yield

a (sin wii/i)u(n) * --- ---- ; ---------

a r -1 sin to < i 1 2az~ cos to,, + a

|z! > la I

(3.2.11)

Time reversal.

If a (//) *-> X ( z ) R O C : r\ < < r;

th en j r ( - n ) <-i-> X { z ~ ]) R O C : < |z < r2

r\

(3.2.12)

P r o o f F ro m th e d efin itio n (3.1.1), w e have Z{a (/;)) = Y

h= - x.

-x { - !i ) z~" Y ' ( h ( z ~ t )~l Ar (z~ ') /= ->;

w h ere th e ch an g e o f v ariab le / = n is m ad e. T h e R O C o f A '( ;_1) is < |c 1f < r-i o r e q u iv alen tly < ];| < r2 n

N o te th a t th e R O C fo r x( n) is the inverse o f th a t fo r x ( n). T h is m ean s th a t if co b elo n g s to th e R O C o f x( n) , th en l/" o is in th e R O C fo r x ( n). A n in tu itiv e p ro o f o f (3.2.12) is th e follow ing. W h en w e fold a signal, the co efficien t o f z ~n b e co m es th e coefficient o f z n. T h u s, fo ld in g a signal is eq u iv alen t to rep lacin g ; by in th e z-tra n sfo rm fo rm u la. In o th e r w o rd s, reflectio n in the tim e d o m ain c o rre sp o n d s to inversion in th e z-d o m ain . Exam ple 3.2.6 Determ ine the z-transform of the signal
x i n ) = u( n)

Solution

It is known from (3.1.8) that u(n) <-U i ROC:

|zl

> 1

By using (3.2.12), we easily obtain u(n) * -----ROC: |z < 1 (3.2.13)

Differentiation in the z-domain.

If

Sec. 3.2

Properties of the z-Transform

167

then nx{n) < > - z ^ ^ y dz Proof . B y differentiating both sid es o f (3.1.1), w e have dX{z) n~ oc n= oc (3.2.14)

- z ~ l Z{nx(n)}
N o te that b oth transform s have the sam e RO C .

Example 3.2.7
D eterm ine the z-transform of the signal jr(n) = na"u(n )

Solution The signal j:(n) can be expressed as njcitn), where Xj(n) = a"u(n ). From
(3.2.2) we have that jri(rt) = aKu(n) < - X,(z) = - ----- 1 az~' Thus, by using (3.2.14), we obtain n<j"ii(n) dXji z) a : '1 X(z) = - z ^ - = ----------- r-r
az
(1 - az~' V

ROC: |z| > \a\

ROC: |z| > |a|

(3.2.15)

If we set a = 1 in (3.2.15), we find the z-transform of the unit ram p signal ntt(n)

ROC: jz| > 1

(3.2.16)

Example 3.2.8
D eterm ine the signal x(n) whose z-transform is given by X(z) = log(l -t- o z '1) |z! > |o(

Solution By taking the first derivative of X(z), we obtain

dX{z) - a z ~2 dz 1 + az~l Thus dX(z) dz = az 1 - ( - a ) z ->

>

\a]

The inverse z-transform of the term in brackets is (-a) ". The multiplication by z _1 implies a time delay by one sample (time shifting property), which results in ( - a ) " _ 1u(n 1). Finally, from the differentiation property we have nx(n ) = a ( a)n~lu(n 1)

168

The z -Transfomn and Its Application to the Analysis of LTI Systems

Chap. 3

x ( n) = ( - 1 } " +1 u(n - 1)

Convolution of two sequences.

Xi (rt) x 2(n)
th e n

If

X^z) X 2 {z)

x( n) = jfi(n) * x 2 (n)

X ( z ) = X \ ( z ) X 2 (z)

(3.2.17)

T h e R O C o f X (;) is, at least, th e in te rse c tio n of th a t fo r ATi(c) an d AS(z). Proof. T h e co n v o lu tio n o f x i(n ) a n d x 2 (n) is d efin e d as
OC

x(n) = T h e z-tra n sfo rm o f x( n ) is

oc

Y
i' = oc

x i ( k ) x 2(n - k)

oc

ac

AT{;) =

Y 2 x( n) z ~ " =
n -x

Y ,
n= - o c |_Jt = -cc

x \ ( k ) x 2(n - k)

U p o n in te rch an g in g th e o rd e r o f the su m m a tio n s a n d ap p ly in g th e tim e-sh iftin g p ro p e rty in (3.2.5). we o b tain X{z) = Y x ' {k) x 2(n k) z "

= X 2 (z) Y

^ ( k ) z ~ k = X 2 ( z ) X ](z)

Example 3.2.9
Compute the convolution x(rt) of the signals jt,(n) = { 1 .-2 ,1 ) _ f 1. 0 < n < 5

10 ,

elsewhere

Solution From (3.1.1), we have

* ,( ; ) = 1 - 2 ; ' + z ~ 2 X 2(z) = I + z ~ ' + z~2 + j ' 3 + z"4 + According to (3.2.17), we carry out the multiplication of X\(z) and X 2(z). Thus X(z) = X 1 (z)X 1 (z) = 1 - z "1 - z ~6 + z "7 Hence
x ( n) = { 1 . - 1 . 0 , 0, 0, 0 , - 1 , 1 }
t

Sec. 3.2

Properties of the z-Transform

169

The same result can also be obtained by noting that X,U) = (1 - ; ~ 'r x 2w Then X(z) = (1 - ; - )(l - z~*) = 1 - z "6 + :" 7

1 - ; -6 = 3 p

The reader is encouraged to obtain the same result explicitly by using the convolution summation formula (tim e-domain approach). T h e co n v o lu tio n p ro p e rty is o n e o f th e m o st p o w erfu l p ro p e rtie s o f th e ztra n sfo rm b ec a u se it c o n v e rts th e co n v o lu tio n o f tw o signals (tim e d o m a in ) to m u ltip lic a tio n o f th e ir tra n sfo rm s. C o m p u ta tio n of th e c o n v o lu tio n o f tw o signals, using th e z -tra n sfo rm , re q u ire s th e follow ing steps: 1. C o m p u te th e z -tra n s fo rm s o f th e signals to be co n v o lv ed . X i(z) = Z{ x \ ( n ) \ (tim e d o m ain * -.-dom ain) X 2 (z) = Z { x 2 (n) ]

2 . M u ltip ly th e tw o z -tran sfo rm s.

X (z) = X ,(z )X 2(;) 3. F in d th e in v e rse z -tra n s fo rm o f X (z). x ( n ) = Z _ , {X(z)) (z-d o m ain tim e d o m a in ) (z-d o m ain )

T h is p ro c e d u re is, in m a n y cases, c o m p u ta tio n a lly e a s ie r th a n th e d irect e v a l u a tio n o f th e co n v o lu tio n su m m atio n .

Correlation of two sequences.

xj(n) x 2 (n) th e n
OC

If X i(z) X 2 (z)

Rx,J2 (z) = X 1 (Z)X 2(Z"1)

OC

(3.2.18)

Proof . We recall that rXix2U) = x\ ( l ) * x 2( - l )

170

The z-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

U sing th e co n v o lu tio n an d tim e-rev ersa l p ro p e rtie s , we easily o b ta in = Z{x] ( / ) }Z{xz ( /)} = X l ( z ) X 2 ( z ~1) T h e R O C o f R X}X2(z) is at least the in te rse c tio n o f th a t fo r X i(c ) an d X 2("- 1 )A s in th e case o f co n v o lu tio n , th e c ro ssc o rre la tio n o f tw o signals is m o re easily d o n e via p o ly n o m ial m u ltip lic atio n acco rd in g to (3.2.18) a n d th en inverse tra n sfo rm in g th e resu lt. Exam ple 3.2.10 Determ ine the autocorrelation sequence of the signal
x (n ) =
a " u {n ).

1 <

< 1

Since the autocorrelation sequence of a signal is its correlation with itself, (3.2.18) gives
S o lu tio n

Rxs(z) = Z{rxJ I ) ) = X (z)X (;~l ) From (3.2.2) we have Ale) = ---------- J - az~' and by using (3.2.15). we obtain | = -.... 1 - az Thus = - = ------------ ------------ 1 az 1 1 - az 1 a (; + ; ) + fl ROC: \a\ < |;| < |c| ROC: |z| > \a\ (causal signal)

ROC: |;| < (a|

(anticausal signal)

Since the R O C of Rxx(z) is a ring, rIX{i) is a two-sided signal, even if x(n) is causal. To obtain rsJ l ) , we observe that the ^-transform of the sequence in Exam ple 3.1.5 with b = 1 /a is simply (1 a1)Rxx(z). Hence it follows that
rXI(l) = -- a ^
\ a-

oc < I < oo

The reader is encouraged to compare this approach with the time-dom ain solution of the same problem given in Section 2.6.

Multiplication of two sequences.

x \ ( n) x 2 (n) th e n r(n) = x i ( n ) x 2 (n) X(z) =

If ATi(z) X 2 {z)

( ) v~ ^ v

(3.2.19)

w h ere C is a clo sed c o n to u r th a t e n clo ses th e o rig in a n d lies w ith in th e reg io n of co n v erg en ce c o m m o n to b o th X](i>) a n d ^ ( l / u ) .

Sec. 3.2

Properties of the z-Transform

171

Proof . T h e z -tra n sfo rm o f J 3 (n) is

O C
X(z) = Y
ft

OC

x( n)z ~n = Y
n = oc

x ] ( n) x 2 ( n) z ~n

L e t us s u b s titu te th e in v erse tra n sfo rm * i(n ) = fo r in th e z -tra n sfo rm X (z) a n d in te rc h a n g e th e o r d e r o f su m m a tio n an d in te g ra tio n . T h u s w e o b ta in 1 X (z)= 2 / " oc v~ ^ v

T h e su m in th e b ra c k e ts is sim ply th e tra n sfo rm X 2 (:) e v a lu a te d a t z / v . T h e re fo re , X(z) = ^ - 6 x ^ x 2 2 n j Jc w hich is th e d e s ire d result. T o o b ta in th e R O C o f X (z) w e n o te th a t if X i(u ) co n v e rg e s fo r ru < |u| < r\ an d X 2 iz) c o n v erg es fo r r2! < |z| < r2ll, th e n th e R O C o f X 2 {z/v) is

H e n c e th e R O C fo r A"(z) is at least r\ir2i < |z[ < r u r2u (3.2.20)

A lth o u g h this p ro p e rty will n o t be used im m e d ia te ly , it will p ro v e u se fu l later, esp ecially in o u r tr e a tm e n t o f filter design b ased o n th e w in d o w te c h n iq u e , w h ere w e m u ltip ly th e im p u lse re sp o n s e o f an IIR system by a fin ite -d u ra tio n w in d o w w h ich se rv es to tru n c a te th e im p u lse re sp o n se o f th e I I R sy stem . F o r c o m p le x -v a lu e d se q u e n c e s .ti(rt) an d x 2 (n) w e c a n define th e p ro d u c t s e q u e n c e as x( n ) = Jti (nj x^i n) . T h e n th e c o rre sp o n d in g c o m p lex co n v o lu tio n in te g ra l b eco m es x ( n ) = x i ( n ) x 2 {n) X(z) = v~ldv (3.2.21)

T h e p r o o f o f (3.2.21) is left as an ex ercise fo r th e re a d e r.

Parsevals relation.
y

If jti(n ) an d x 2 (n) a re c o m p le x -v a lu e d se q u en ces, th e n v ~ 'd v (3.2.22)

* i(n )x 2 (n) = l j ( j ^ X i ( v ) X 2

p ro v id e d th a t r ^ r y < 1 < n ur2u, w h e re ry < |z| < r \ u a n d r ^ < |z| < r2u a re th e R O C o f X ](z) a n d X 2(z). T h e p ro o f o f (3.2.22) follow s im m e d ia te ly by ev alu atin g X ( z ) in (3.2.21) at z = 1.

172

The z-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

The Initial Value Theorem.

If j ( ;i ) is causal [i.e.. x ( n ) = 0 fo r n < 0], th e n (3.2.23)

.r(0) = lim X ( z ) : * sc Proof. Since x{n) is causal. (3.1.1) gives X (z) y x {n )z x (0) + x (1 )z 1 + x (2 )z =o +

O b v iou sly , as z oc. z ~" 0 since n > 0 an d (3.2.23) follow s. A ll th e p ro p e rtie s o f th e z -tra n sfo rm p r e s e n te d in th is s e c tio n are su m m arized in T a b le 3.2 fo r easy re fe re n c e . T h ey are listed in th e sam e o rd e r as th ey have b een in tro d u c e d in th e tex t. T h e c o n ju g atio n p ro p e rtie s a n d P a rse v a l's relatio n are left as ex ercise s fo r th e re a d e r. W e have now' d e riv e d m o st o f th e z -tra n sfo rm s th a t are e n c o u n te re d in m any p ractical ap p licatio n s. T h e se z -tra n sfo rm pairs a re su m m a riz e d in T ab le 3.3 for easy re fe re n c e . A sim ple in sp e ctio n o f th is ta b le show s th a t th e s e z-tran sfo rm s are all rational f u n c t i o n s (i.e., ratio s o f p o ly n o m ials in z _1). A s will soon b ecom e a p p a re n t, ratio n al z -tra n sfo rm s are e n c o u n te re d n o t only as th e z-tran sfo rm s of v ario u s im p o rta n t signals b u t also in th e c h a ra c te riz a tio n of d isc re te -tim e lin ear tim e -in v a ria n t sy stem s d esc rib e d by c o n s tan t-co efficien t d iffe re n c e e q u a tio n s.

3.3 RATIONAL Z-TRANSFORMS

A s in d ic a te d in S ectio n 3.2, an im p o rta n t fam ily o f z-tra n sfo rm s a re th o se fo r w hich X ( z ) is a ra tio n a l fu n ctio n , th a t is. a ra tio of tw o p o ly n o m ials in z _l (o r z). In this sectio n w e discuss som e very im p o rta n t issues re g a rd in g th e class o f ra tio n a l z-tran sfo rm s.

3.3.1 Poles and Zeros

T h e zeros of a z -tra n sfo rm X (z) a re th e v alu es of z fo r w hich X (z) = 0. T h e pol es o f a z-tra n sfo rm are th e v alu es o f z fo r w hich X (z) = oc. If X ( z ) is a ratio n al fu n ctio n , th e n

J K ,) . D( z) ao + a i z 1 + ----- \ - aNz ~ h

= --------* k=o

(3.3.1)

If ao / 0 an d bo ^ 0, w e can avoid th e n eg ativ e p o w e rs o f z by fa cto rin g o u t the te rm s boz~M a n d a$z~N as follow s:

X(z)

N( z)
=

b 0z ~ M z M + {bx/ bo) z M~ x + + b M/ b 0 a0z N Z N + ( a \ / a a) z fJ~ l H -------- \ - aN/ a Q

D( z)

TABLE 3.2

PROPERTIES OF THE Z-TRANSFORM

Property Notation

Time Domain *(n) xt (n) ^(n) aixy(n) + a2x 2(n) x{n - k) a"x(n) x(~rj) x'(rt) Relx(n)} lm{x(n)) nx(n) x i ( n ) * x 2(n) rx,x2(l) = * i(0 * x 2 (~l) If x(n) causal Xi(n)x 2(n) = 2? i i

z-Domain X(z) Xi(z) X 2(z) <t\* l(z ) + 2X 2(z) z~kX(z) X(a~' z) X(z~') X' ( z ' ) i[X (z) + X*U*)] i[X (z )-* (; )] dX( z ) Z dz Jf|(z)X 2(z) / W z ) = Xi<z)* 2<z_l) .r(O) = lim X(z) 2~J^X,(v)X2^ J v - ' d v

R OC ROC: r2 < |z) < r\ ROC, ROC2 At least the intersection of R O Q and R O C 2 T hat of X (z), except z = 0 if k > 0 and z = oo if k < 0 \a\r2 < |z| < |a|r[ -

Linearity Time shifting Scaling in the z-domain Time reversal Conjugation Real part Imaginary part Differentiation in the z-domain Convolution Correlation Initial value theorem Multiplication Parsevals relation

ROC Includes ROC Includes ROC r2 < |z| < r. A t least, the intersection of ROCi and R O C 2 A t least, the intersection of ROC of X,(z) and -K^z 1)

r1

< |z| <

r2

A t least rj/ry < |z| < r]r2 a

5X ,( t)) ^ (l/tJ ) v - ' dv

174

The z-Transform and Its Application to the Analysis of LTI Systems

TABLE 3.3 SOME COMMON Z-TRANSFORM PAIRS : -Transform, X (c) 1 1 1 - r-1 1 1 az~l 1:1 > 1 ROC A ll ;

Chap. 3

Signal. x(n) 1 2 3 4 5 6 7 8

5{n) u(n ) a"u(n) na"u(n ) au(n 1) na'u(n 1)

(cos aion)u(n) (sin a>nn)u(n)
( a" cosiH ) n ) u ( n )

\z\ > \a\ Izt > !a|

k l < la|

a:" 1 ( l - a z ^ )2
1 1 a : -1

a z~l

(1 - c : - 1)2
1 - Z '1 COSf^o

|;| < N
kl > 1

1 - 2; _1 costal + z ~2
sin
1-

2 ;" 1 cos tun + z ~2 1 az~] cos a*)

\z\ > 1

10

(a'1sini*in)u{n)

1 2az~ cos wo + a 2z ~2 a ; -1 sinaiii 1-

2a ; " 1 cos a*,

a 2z ~2

Since N ( z ) and D( z ) are polynom ials in z, they can be expressed in factored form as
X (-)
b() - - M + N

D( z ) u

~ ~ Z2) (Z ~ (z - p i ) ( z - P i ) (z -

Zm )
p n

n " z*} X ( z ) = G z N~ M^A -----------*=i

(3.3.2)

n<* ~ pjt)

w here G = 6o/ao- Thus X (r) has M finite zeros at z = z\, zi , . , z m (th e roots o f the num erator p olyn om ial), N finite p oles at z = p \ , p i .........P n (th e roots o f the d enom inator p olyn om ial), and \N M\ zeros (if N > M ) or p o le s (if N < M ) at the origin z = 0. P o les or zeros m ay also occur at z = 0 0 . A zero exists at z = oc if X ( 0 0 ) = 0 and a p o le exists at z = oc if X ( 0 0 ) = oc. If w e count the p oles and zeros at zero and infinity, w e find that X (z) has exactly the sam e num ber o f p o les as zeros. W e can represent X ( z ) graphically by a p o l e - z e r o p l o t (or pat t ern) in the com plex plane, which show s the location o f p oles by crosses ( x ) and the location o f zeros by circles (o). T he m ultiplicity of m ultipie*order p o les or zeros is indicated by a num ber clo se to the corresponding cross or circle. O b viou sly, by definition, the R O C o f a z-transform should not contain any poles.

Sec. 3.3

Rational z-Transform s

175

Exam ple 33.1 D eterm ine the pole-zero plot for the signal x(n) = a"u(n) Solution From Table 3.3 we find that X(z) = a >0

1 _____
1 a ; -1 - a

ROC; \z\ > a

Thus has one zero at n = 0 and one pole at pi = a. The pole-zero plot is shown in Fig. 3.7. Note that the pole p\ = a is not included in the R O C since the z-transform does not converge at a pole.

Re(;)

Figure 3.7

Pole-zero plot for the causal exponential signal ,v() = a"imu

Exam ple 3 3 .2 D eterm ine the pole-zero plot for the signal

1 0,
where a > 0. Solution

Of n < W - 1 elsewhere

From the definition (3.1.1) we obtain ^ X(z) = X W " ln

1}" = - r r ^ ------ r

l - t a ; ' 1) " z - a u T T t --------T 1 - az 1 = S z (z - a)

Since a > 0, the equation z M = a M has M roots at

Z t = a e ' 1* * ' " it = 0 , 1 . . . . . . M 1

The zero zo = a cancels the pole at z = a. Thus X(z) = ---------------- ^ - ---------------w ,

(Z - Z\ ) ( Z - Zl ) (Z - Z j t f - i )

which has M 1 zeros and M - 1 poles, located as shown in Fig. 3.8 for M = 8. Note th at the R O C is the entire z-plane except z = 0 because of the M - 1 poles located at the origin.

176

The z-Transfonm and Its Application to the Analysis of LTI Systems

Chap. 3

Im(c)

Red)

Figure 3.8 Pole-zero patlem for the finite-duration signal x(n ) = a", 0 < n < M l(a > 0). for M = 8. C learly , if we are given a p o le - z e r o p lo t, w e can d e te rm in e X ( ), by using (3.3.2), to w ithin a scaling fa c to r G. T his is illu stra te d in th e follow ing ex am p le. Example 3.3.3 Determine the c-transform and the signal that corresponds to the pole-zero plot of Fiji. 3.9. Solution There are two zeros (M = 2) at Z] = 0, Z2 = r cosojo and two poles (N = 2) al p] reJ , Pz = re~iw". By substitution of these relations into (3.3.2), we obtain X(z) = ( (; - Pi)(c - pi) = G(c - re}w u)(z - re~>w ")
Z(Z r C O S ttH i)

ROC:

A fter some simple algebraic manipulations, we obtain Xiz) = G-. 1 - 2 r - ~ l costou + r1 From Table 3.3 we find that x(n) = G(r" cosa>on)u(n) F ro m E x a m p le 3.3.3, w e see th a t th e p r o d u c t (; p \ ) ( z P 2 ) resu lts in a p o ly n o m ial w ith re a l coefficients, w h en p\ a n d p 2 a re c o m p lex co n ju g ates. In Im(2)

C O S a><)

ROC: 1-1 > r

Figure 3.9 P ole-zero p attern for Exam ple 3.3.3.

Sec. 3.3

Rational z - T ransforms

177

g e n e ra l, if a p o ly n o m ial h a s re a l coefficients, its ro o ts a re e ith e r real o r o ccu r in c o m p lex -co n ju g ate pairs. A s w e h av e seen , th e ; -tran sfo rm X ( z ) is a com plex fu n c tio n of th e com plex v ariab le z = R e ( z ) + j Im (r). O b v io u sly , |X (c )|, th e m a g n itu d e o f X ( z ), is a real and p o sitiv e fu n ctio n o f c. S ince : re p re se n ts a p o in t in th e co m p lex p la n e , |X ( )| is a tw o -d im e n sio n a l fu n c tio n an d d esc rib e s a su rfa c e . T h is is illu stra te d in Fig. 3.10(a) fo r th e z -tra n sfo rm
7 - 1 _ - - 2

(a)

Figure 3.10 Graph of |X (;)| for the ;;-Transform in (3.3.3). [Reproduced with permission fr om Introduction to Systems Analysis, by T. H. Glisson, 1985 by McGraw-Hill B ook Company.]

178

The z-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

w hich h as o n e z e ro a t z\ = 1 a n d tw o p o le s a t p \ , p 2 = 0. 9e J" /4. N o te the high p e a k s n e a r th e sin g u larities (p o les) a n d th e d e e p v alley close to th e zero. F ig u re 3.10(b) illu stra te s th e g rap h o f lX (z)j fo r z = eJO > .

3.3.2 Pole Location and Time-Domain Behavior for Causal Signals

In th is su b sectio n w e c o n sid e r th e re la tio n b e tw e e n th e z -p la n e lo catio n o f a pole p a ir a n d th e fo rm (sh a p e ) o f th e c o rre sp o n d in g signal in th e tim e d o m ain . T h e dis cu ssio n is b a s e d g en erally o n th e collectio n o f z -tra n s fo rm p a irs given in T a b le 3.3 an d th e resu lts in th e p re c e d in g su b sectio n . W e d e a l exclusively w ith real, causal signals. In p a rtic u la r, w e se e th a t th e c h a ra c te ristic b e h a v io r o f cau sal signals d e p e n d s o n w h e th e r th e p o les o f th e tra n sfo rm a re c o n ta in e d in th e reg io n |z| < 1 , o r in th e reg io n |z| > 1, o r on th e circle |z| = 1. S ince th e circle jz| = 1 h as a ra d iu s o f 1 , it is called th e unit circle. If a real signal h as a z-tran sfo rm w ith o n e p o le , th is p o le h a s to b e real. T he o n ly su ch signal is th e re a l e x p o n e n tia l x ( n ) = a nu{n) ^ X ( z ) = -- ---- r 1 az R O C : |z] > |o|

h av in g o n e z e ro a t zi = 0 an d o n e p o le a t pi a on th e re a l axis. F ig u re 3.11
x (n) x(n )

1 1

1 1 ! TTt

x(n)

x (n 0 i l l I I I r t

x(n)

'1

n i

1 -oi '

Figure 3.11 Time-domain behavior of a single-real pole causal signal as a function of the location of the pole with respect to the unit circle.

Sec. 3.3

Rational z-Transform s

179

illu stra te s th e b e h a v io r o f th e signal w ith re sp e c t to th e lo c a tio n o f th e p o le r e l ative to th e u n it circle. T h e signal is d ecay in g if th e p o le is inside the unit circle, fixed if th e p o le is o n th e u n it circle, an d gro w in g if th e p o le is o u t side th e u n it circle. In a d d itio n , a n eg ativ e p o le resu lts in a signal th a t a lte r n a te s in sign. O b v io u sly , causal signals w ith p o les o u tsid e th e u n it circle b e co m e u n b o u n d e d , cau se overflow in d igital system s, a n d in g e n e ra l, sh o u ld be av o id ed . A cau sal re a l signal w ith a d o u b le re a l p o le h as th e fo rm
x ( n ) = n a nu ( n )

(see T a b le 3.3) an d its b e h a v io r is illu stra te d in Fig. 3.12. N o te th a t in c o n tra s t to th e sin g le-p o le signal, a d o u b le real p o le o n th e u n it circle re su lts in an u n b o u n d e d signal. F ig u re 3.13 illu stra te s th e case of a p a ir o f c o m p le x -c o n ju g a te poles. A c c o rd ing to T a b le 3.3, th is co n fig u ratio n of p o le s resu lts in an e x p o n e n tia lly w eig h ted sin u so id al signal. T h e d istan c e r of th e p o les fro m th e o rigin d e te rm in e s th e e n v e lope o f th e sin u so id al signal an d th e ir angle w ith th e real p o sitiv e axis, its relative freq u e n cy . N o te th a t th e a m p litu d e o f th e signal is gro w in g if r > 1, c o n stan t if r 1 (sin u so id a l sig n als), a n d d ecaying if r < 1 .
x(n)

<

T T ....

xin)

. 0

T 1

Figure 3.12 Time-domain behavior of causal signals corresponding to a double (m = 2) real pole, as a function of the pole location.

180

The z-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

Figw e 3.13 A pair of complex-conjugate poles corresponds to causal signals with oscillatory behavior.

Finally, Fig. 3.14 show s the b eh avior o f a causal signal w ith a d ou b le pair of p o les on the unit circle. T his reinforces the corresponding results in Fig. 3.12 and illustrates that m ultiple p o les on the unit circle should b e treated with great care. T o sum m arize, causal real signals w ith sim ple real p oles or sim ple com plexconjugate pairs o f p oles, w hich are in sid e or on the unit circle are always bounded in am plitude. Furtherm ore, a signal with a p o le (or a com p lex-con ju gate pair o f p o les) near the origin decays m ore rapidly than on e a ssociated with a pole n ear (but inside) the unit circle. Thus the tim e behavior o f a signal depends strongly on the location o f its p o les relative to th e unit circle. Z eros also af fect the behavior o f a signal but n ot as strongly as p oles. F or exam ple, in the

Sec. 3.3

Rational z-Transform s

181

Figure 3.14 Causal signal corresponding to a double pair of complex-conjugate poles on the unit circle. case o f sin u so id al signals, th e p resen c e an d lo catio n o f ze ro s affects only th e ir p h ase. A t this p o in t, it sh o u ld be stressed th a t e v e ry th in g w e h av e said a b o u t causal signals ap p lies as w ell to causal L TI system s, since th e ir im p u lse re sp o n se is a causal signal. H en ce if a p o le of a system is o u tsid e the unit circle, th e im pulse resp o n se of th e system b eco m es u n b o u n d e d and. co n se q u e n tly , th e sy stem is u n sta b le .

3.3.3 The System Function of a Linear Time-Invariant System

In C h a p te r 2 w e d e m o n s tra te d th a t th e o u tp u t o f a (re la x e d ) lin e a r tim e -in v a ria n t sy stem to an in p u t se q u e n c e x( n) can be o b ta in e d by co m p u tin g th e co n v o lu tio n of jc(h) w ith th e u n it sa m p le resp o n se o f th e system . T h e co n v o lu tio n p ro p e rty , d eriv e d in S ectio n 3.2, allow s us to ex p ress th is re la tio n sh ip in th e z-d o m ain as Y( z ) = m z ) X ( z ) (3.3.4)

w h ere Y( z ) is th e z -tra n sfo rm o f th e o u tp u t se q u en ce v(n), X (z) is th e z-tra n sfo rm o f th e in p u t se q u e n c e x ( n ) an d H( z ) is th e z -tra n sfo rm o f th e u n it sa m p le resp o n se h{n). If w e k n o w h( n) an d x( n ) , w e can d e te rm in e th e ir c o rre sp o n d in g z-tra n sfo rm s H ( z ) a nd X ( z ) , m u ltip ly th e m to o b ta in Y(z), a n d th e r e fo r e d e te rm in e y( n) by e v a lu a tin g th e in v erse z -tra n sfo rm of K(z). A lte rn a tiv e ly , if w e k now x ( n ) an d we o b se rv e th e o u tp u t y( n) of th e system , w e can d e te rm in e th e u n it sa m p le resp o n se by first solv in g fo r H( z ) fro m th e re la tio n

a n d th e n e v a lu a tin g th e in v erse z -tra n sfo rm o f H( z ) . Since

OC

(3.3.6)

182

The z-Transform and Its Application to the Analysis of LTI System s

Chap. 3

it is c le a r th a t H ( z ) re p re s e n ts th e z -d o m ain c h a ra c te riz a tio n o f a system , w h ereas h{n) is th e c o rre sp o n d in g tim e -d o m a in c h a ra c te riz a tio n of th e system . In o th e r w o rd s, H{z ) an d h( n) are e q u iv a le n t d e s c rip tio n s o f a system in th e tw o d o m ain s. T h e tra n sfo rm H{ z ) is called th e s yst em f unct i on. T h e re la tio n in (3.3.5) is p a rtic u la rly u sefu l in o b ta in in g H{ z ) w h en th e system is d e s c rib e d b y a lin e a r c o n s tan t-co efficien t d iffe re n c e e q u a tio n o f th e fo rm M N (3.3.7) In th is case th e sy stem fu n ctio n can be d e te rm in e d d irectly fro m (3.3.7) by com p u tin g th e z -tra n s fo rm o f b o th sid es o f (3.3.7). T h u s, by a p p ly in g th e tim e-sh iftin g p ro p e rty , w e o b ta in M
y u

) = - J 2 a*Y ^ z ~k + E

b iX (z )r

M Y(z) X(z)

o r, eq u iv alen tly , > z-* (3.3.8)

T h e re fo re , a lin e a r tim e -in v a ria n t system d e s c rib e d by a c o n s ta n t-c o e ffic ie n t dif fere n ce e q u a tio n h a s a ra tio n a l system fu n ctio n . T h is is th e g e n e ra l fo rm fo r th e system fu n c tio n o f a sy ste m d e sc rib e d by a lin e a r c o n s tan t-co efficien t d ifferen ce e q u a tio n . F ro m th is g e n e ra l fo rm w e o b tain tw o im p o rta n t sp e cial fo rm s. F irst, if ajt = 0 fo r 1 < k < N , (3.3.8) re d u c e s to M H ( z ) = J 2 bkz ~k
k=0

(3.3.9)

In th is case, H ( z ) c o n ta in s M z ero s, w h o se valu es a re d e te r m in e d by the system p a ra m e te rs {*}, a n d an M th -o rd e r p o le a t th e o rigin z = 0. S ince the sy stem c o n ta in s o n ly trivial p o le s (a t z = 0) an d M n o n triv ia l z e ro s, it is called

Sec. 3.3

Rational z-Transform s

183

an all -zero syst em. C learly, such a system h as a fin ite -d u ra tio n im p u lse resp o n se (F IR ), a n d it is called an F IR system o r a m oving av erag e (M A ) system . O n th e o th e r h an d , if bk 0 fo r 1 < k < M, th e system fu n c tio n re d u c e s to

A'

*= 1 boz" at*.

k
(3.3.10) <io = l

In this case H ( z ) co n sists of N poles, w hose v alu es are d e te r m in e d by th e system p a ra m e te rs {a*} a n d an /V th-order zero at th e o rig in z = 0- W e usually do not m a k e re fe re n c e to th e se trivial zeros. C o n se q u e n tly , th e sy stem fu n c tio n in (3.3.10) c o n ta in s o n ly n o n triv ia l p o les an d th e c o rre sp o n d in g sy stem is called an all-pole syst em. D u e to th e p re se n c e o f poles, th e im p u lse re sp o n s e o f such a system is infinite in d u ra tio n , a n d h e n c e it is an I IR system . T h e g e n e ra l fo rm o f th e sy stem fun ctio n given by (3.3.8) c o n ta in s b o th poles a n d zero s, and h e n c e th e c o rre sp o n d in g sy stem is called a p o l e - z e r o s y s t e m , with N p o le s a n d M z ero s. P o les a n d /o r zero s at c = 0 an d z = oc a re im p lied b u t are n o t c o u n te d ex p licitly. D u e to th e p re se n c e o f p o les, a p o le - z e r o system is an IIR system . T h e fo llo w in g ex am p le illu strates th e p ro c e d u re fo r d e te rm in in g th e system fu n ctio n a n d th e u n it sa m p le resp o n se from the d iffe re n c e e q u a tio n . E xam ple 3.3.4 D eterm ine the system function and the unit sample response of the system described by the difference equation
v(n) =

1y(n -

1) + 2 x ( n )

Solution

By computing the -transform of the difference equation, we obtain Y(z) = z - > Y(z) + 2X(z)

H ence the system function is

xt:l

I - Jz -1

This system has a pole at z = \ and a zero at the origin. Using Table 3.3 we obtain the inverse transform h(n) = 2(i)"u()

This is the unit sample response of the system.

184

The z-T ransform and Its Application to the Analysis of LTI Systems

Chap. 3

W e h av e n o w d e m o n s tra te d th a t ra tio n a l z -tra n s fo rm s a re e n c o u n te re d in c o m m o n ly u se d sy stem s a n d in th e c h a ra c te riz a tio n o f lin e a r tim e -in v a ria n t sys tem s. In S ectio n 3.4 w e d esc rib e se v e ra l m e th o d s fo r d e te rm in in g th e inverse z -tra n sfo rm o f ra tio n a l fu n ctio n s.

3.4 INVERSION OF THE Z-TRANSFORM

A s w e saw in S ectio n 3.1.2, th e in v erse z -tra n s fo rm is fo rm ally given by x ( n ) = - < x ( z ) z n~ 1d z 2 njjt (3.4.1)

w h ere th e in te g ra l is a c o n to u r in te g ra l o v e r a clo sed p a th C th a t en clo ses the o rig in an d lies w ith in th e reg io n o f c o n v e rg e n c e o f ^ ( z ) . F o r sim plicity, C can be ta k e n as a circle in th e R O C o f X (z) in th e z-p lan e. T h e re a re th re e m e th o d s th a t a re o fte n u se d fo r th e e v a lu a tio n o f th e inverse z-tran sfo rm in practice: 1. D ire c t e v a lu a tio n o f (3.4.1), by c o n to u r in te g ra tio n . 2. E x p a n sio n in to a se rie s o f te rm s, in th e v a ria b le s z, an d z _1. 3. P a rtia l-fra c tio n ex p an sio n a n d ta b le lo o k u p .

3.4.1 The Inverse z-Transform by Contour Integration

In th is se ctio n w e d e m o n s tra te th e use o f th e C au ch y re sid u e th e o r e m to d e te rm in e th e in v erse z -tra n sfo rm d irectly fro m th e c o n to u r in teg ral.

Cauchy residue theorem. L e t / ( z ) b e a fu n c tio n o f th e c o m p lex v ariab le z an d C b e a clo sed p a th in th e z -p lan e. If th e d e riv a tiv e d f ( z ) / d z exists o n and inside th e c o n to u r C a n d if / ( z ) has no p o le s a t z = zo, th e n
2njjcz-zo

- L (f) J ! d z

= |^ < Z l,)' 10,

if zo is ou tsid e C

(3.4.2)

M o re g en erally , if th e (k + l ) - o r d e r d e riv a tiv e o f / ( z) exists a n d / ( z ) h a s n o p o les a t z = zo, th e n 1 d k- ' f ( z )

-!>'
0,

C
if zo is o u ts id e C

04.3)

T h e v alu es o n th e rig h t-h a n d sid e o f (3.4.2) a n d (3.4.3) a re c a lle d th e re sid u e s of th e p o le a t z = zo- T h e re su lts in (3.4.2) a n d (3.4.3) a re tw o fo rm s o f th e Cauc hy residue t heorem. W e can ap p ly (3.4.2) a n d (3.4.3) to o b ta in th e v a lu e s o f m o re g en eral c o n to u r in teg rals. T o b e specific, su p p o s e th a t th e in te g ra n d o f th e c o n to u r in te g ra l is

Sec. 3.4

Inversion of th e

2 - T ra n s fo rm

185

P(z) = f ( z ) / g ( z ) ~ w h e re f ( z ) h a s no p o les inside th e c o n to u r C an d g (z) is a p o ly n o m ial w ith d istin ct (sim p le) ro o ts c i, ^ 2 . ___-n inside C. T h e n

(3.4.4)
n

1=1
w h ere f(z) (3.4.5) A l (z) = ( z - z i ) P{ z ) = ( z - z l) J - 1x g( z) T h e v alu es (A, (-;,)} a re re sid u e s o f th e c o rre sp o n d in g p o le s at z = / = 1, 2 , . . . . n. H e n c e th e v alu e o f th e c o n to u r in te g ra l is e q u a l to th e sum o f th e resid u es o f all th e p o le s in sid e th e c o n to u r C. W e o b se rv e th a t (3.4.4) w as o b ta in e d by p e rfo rm in g a p a rtia l-fra c tio n e x p a n sion o f th e in te g ra n d an d ap p ly in g (3.4.2). W h en g(z) has m u ltip le -o rd e r ro o ts as w ell as sim p le ro o ts inside th e c o n to u r, th e p a rtia l-fra c tio n e x p a n sio n , w ith a p p ro p ria te m o d ifica tio n s, an d (3.4.3) can b e used to e v a lu a te th e resid u es at th e c o rre sp o n d in g p o les. In th e case o f th e in v erse z -tra n sfo rm , w e h ave

[resid u e of X (z )z n 1 a t z
a ll p o l e s U t) in s id e C

(3.4.6)

p ro v id e d th a t th e p o les {z,} a re sim ple. If X (z )r "_1 has no p o le s inside th e c o n to u r C fo r o n e o r m o re v alu es o f n, th e n x (n) = 0 fo r th e s e values. T h e fo llo w in g ex am p le illu stra te s th e e v a lu a tio n o f th e in v erse z-tra n sfo rm by u se o f th e C a u ch y re sid u e th e o re m . Exam ple 3.4.1 Evaluate the inverse z-transform of X(z) = ---------- r 1 az~ using the complex inversion integral. Solution We have kl > kil

where C is a circle at radius greater than |a|. We shall evaluate this integral using (3.4.2) with f ( z ) = z". We distinguish two cases.

186

The ^-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

L If n > 0, f ( z ) has only zeros and hence no poles inside C. T he only pole inside C is z = a. Hence *(*) = /(zo) = a n
n > 0

2. If n < 0, / ( z ) = z" has an nth-order pole at z = 0, which is also inside C. Thus there are contributions from both poles. For n = - 1 we have x (-l) = If n = 2, we have

1 S) 1 dz = 1 2 nj j c z ( z a) z a

+ -

= 0

2)

2 n j z H z - a ) dZ

dz ( z - f l )

= 0

By continuing in the same way we can show that *(n) = 0 for n < 0. Thus x (n) = a"u(n)

3.4.2 The Inverse z-Transform by Power Series Expansion

The basic idea in this m eth od is the follow ing: G iven a z-transform X ( z ) with its corresponding R O C , w e can expand X (z) into a p ow er series o f the form
OO

X(z) =

co

cz~n

(3.4.7)

w hich con verges in the given R O C . T h en , by the u n iqu en ess o f the z-transform, x ( n ) = c for all n. W hen X ( z ) is rational, the exp an sion can b e perform ed by long division. T o illustrate this tech n iqu e, w e w ill invert som e z-transform s involving the sam e expression for X ( z ) , but different R O C . T his w ill also serve to em phasize again the im portance o f the R O C in d ealing with z-transform s. Exam ple 3A 2 D eterm ine the inverse z-transform of 1 1.5z_1 + 0.5z 2 when (a) ROC: |z| > 1 (b) ROC: |z| < 0.5 Solution (a) Since the R O C is the exterior of a circle, we expect x(n) to be a causal signal. Thus we seek a power series expansion in negative powers of z. By dividing

Sec. 3.4

Inversion of the z-Transform

187

the num erator of X{z) by its denom inator, we obtain the power series A UI = ! _ 3 _ -; +
= 1+

+ TE; "4 + " '

By com paring this relation with (3.1.1), we conclude that

Note that in each step of the long-division process, we eliminate the lowestpower term of c~*. (b) In this case the ROC is the interior of a circle. Consequently, the signal x(n) is anticausal. To obtain a power series expansion in positive powers of c. we perform the long division in the following way: 2: 2 + 6c3 + 14c4 + 30cs + 62c* + + ill 1 - 3: + 2c: 3c - 2z 2 3c - 9c: + 6c3 l z 2 - 6c3 7 r - 21 c3 + 14c4 15c3 - 14c4 15c3 - 45c4 + 30cs 31c4 - 30c5 Thus
X (c) =

1_

------:------= 2c: + 6c3 + 14c4 + 30c5 + 62cfi + 1

In this case x(n) = 0 for n > 0. By comparing this result to (3.1.1), we conclude that Jt(n) = { 62. 30. 14.6,2, 0. 0}
t

We observe that in each step of the long-division process, the lowest-power term of c is eliminated. We emphasize that in the case of anticausal sig nals we simply carry out the long division by writing down the two poly nomials in reverse order (i.e., starting with the most negative term on the left).

F ro m th is e x a m p le w e n o te th a t, in g e n eral, th e m e th o d o f long d ivision will n o t p ro v id e a n sw ers fo r x( n) w h en n is larg e b e c a u se th e lo n g division b eco m es ted io u s. A lth o u g h , th e m e th o d p ro v id es a d irect e v a lu a tio n o f x( n ) , a clo sed -fo rm so lu tio n is n o t p o ssib le , ex cep t if th e resu ltin g p a tte r n is sim p le e n o u g h to infer th e g e n e ra l te rm x ( n ) . H e n c e th is m e th o d is used only if o n e w ish e d to d e te rm in e th e v a lu e s o f th e first few sa m p le s o f th e signal.

188

The z-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

Example 3.4.3 Determ ine the inverse z-transform of X(z) = log(l + az_1) Solution |zj > |a|
+ x ),

Using the power series expansion for log(l ^ ( - r v r "

with

|jc| < 1,

we have

Thus

0,

n < 0

Expansion of irrational functions into power series can be obtained from tables.

3.4.3 The Inverse z-Transform by Partial-Fraction Expansion

In th e tab le lo o k u p m e th o d , w e a tte m p t to e x p ress th e fu n c tio n X (z) as a linear c o m b in a tio n X ( z ) = a j X] (;) + 02 X 2 ( 1 ) + - - + c/Kx fc(z) (3.4.8)

w h ere X] ( ; ) , . . . . X k (z ) a re ex p ressio n s w ith in v erse tra n sfo rm s x \ ( n ) , . . . , x k W av ailab le in a ta b le o f z-tra n sfo rm p airs. If such a d e c o m p o s itio n is possible, th e n x( n) , th e in v erse z -tra n sfo rm o f X ( z ) , can easily b e fo u n d using th e lin earity p ro p e rty as x ( n) = at]Xi(n) + a 2x 2 (n) H -------- \ - a Kx K (n) (3.4.9)

T h is a p p ro a c h is p a rtic u la rly useful if X ( z ) is a ra tio n a l fu n ctio n , a s in (3.3.1). W ith o u t loss o f g e n erality , w e assu m e th a t ao = 1, so th a t (3.3.1) can b e e x p ressed as _ t o D(z) _ b +b,r' + + t r

l+ fliZ - 1 H ----- + a u z ~ N

N o te th a t if a 0 ^ 1. w e can o b ta in (3.4.10) fro m (3.3.1) by div id in g b o th n u m e ra to r an d d e n o m in a to r by aoA ra tio n a l fu n ctio n o f th e form (3.4.10) is called p r o p e r if a N ^ 0 an d M < N. F ro m (3.3.2) it fo llo w s th a t th is is e q u iv a le n t to saying th a t th e n u m b e r o f finite zero s is less th a n th e n u m b e r o f fin ite p o les. A n im p ro p e r ra tio n a l fu n ctio n ( M > N ) can alw ays b e w ritte n as th e sum of a p o ly n o m ial an d a p r o p e r ra tio n a l fu n ctio n . T h is p ro c e d u re is illu stra te d by the fo llow ing ex am p le. Example 3.4.4 Express the im proper rational transform l + 3 z - ' + n z - 2 + l 2-3 + in terms of a polynomial and a proper function.
1
5

1 + 6Z

Sec. 3.4

Inversion of the z-Transform

189

Solution First, we note that we should reduce the num erator so that the term s ; -2 and c- * ' are eliminated. Thus we should carry out the long division with these two polynomials written in reverse order. We stop the division when the order of the rem ainder becomes Then we obtain ^ = 1 + 2: - i +

In g e n e ra l, an y im p ro p e r ra tio n a l fu n ctio n (M > N ) can b e e x p ressed as X ( ;) = W i = Co + C i:" 1 + ' ' ' + Cu~n Z ~im~N) + (3 A ll)

T h e in v erse z -tra n s fo rm o f the p o ly n o m ial can easily b e fo u n d by in sp ectio n . W e fo cu s o u r a tte n tio n on th e inversion o f p r o p e r ra tio n a l tra n sfo rm s, since any im p ro p e r fu n c tio n can b e tra n sfo rm e d in to a p ro p e r fu n c tio n by using (3.4.11). W e carry o u t th e d e v e lo p m e n t in tw o steps. F irst, we p e rfo rm a p a rtia l fra c tio n e x p a n sio n o f th e p r o p e r ra tio n a l fu n ctio n a n d th e n w e in v ert each o f th e term s. L e t A"(c) b e a p ro p e r ra tio n a l fu n ctio n , th a t is, * (.-) = = - " - - ' " T ,1 + +buZ^ D( z) 1 + ^ iC + * -t~ w h ere aN ^ 0 an d M < N (3.4.12)

T o sim p lify o u r discu ssion w e e lim in ate n eg ativ e p o w ers of c by m ultip ly in g b o th th e n u m e r a to r a n d d e n o m in a to r of (3,4,12) by z N. T h is resu lts in
L N I L _ y v -l I I L N -M

X{z) =
CA 4- a \ z N + ------(- <a /v

Since N > M , th e fu n ctio n ,N -2 _ i_____ l

(3.4.13,

w hich c o n ta in s only p o sitiv e p o w e rs o f

(3,4.14) ; : w + tiic w- 1 + - - - + flW is also alw ays p ro p e r. O u r ta sk in p e rfo rm in g a p a rtia l-fra c tio n ex p an sio n is to ex p ress (3.4.14) o r, e q u iv a le n tly , (3.4.12) as a sum o f sim ple fractio n s. F o r th is p u rp o se w e first fa c to r th e d e n o m in a to r p o ly n o m ial in (3.4.14) in to facto rs th a t c o n tain th e poles P i, p 2, . . . , p n o f X (z). W e d istinguish tw o cases.

Distinct poles. S u p p o se th a t th e p o le s p \ , p 2 ........ p/v a re all d iffe re n t (dis tin ct). T h e n w e se e k an ex p an sio n of th e fo rm
z z p\ z P2 + -. - + ( 3 . z Pn 4 . 1 5 )

T h e p ro b le m is to d e te rm in e th e coefficients A i , A 2 , . - . , A s - T h e re a re tw o w ays to so lv e th is p ro b le m , as illu stra te d in th e follow ing exam p le.

190

The z-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

Example 3.4 .5 D eterm ine the partial-fraction expansion of the proper function (3'U 6 , Solution First we elim inate the negative powers, by multiplying both num erator and denom inator by z2. Thus
.2

Xiz)= z V < + i0 n.5 2- 1 1.5z The poles of X(z) are p\ = 1 and P2 = 0.5. Consequently, the expansion of the form (3.4,15) is X(z) z C z l)(z 0.5) A\ z 1 A2 z 0.5 (3.4.17)

A very simple method to determ ine A[ and A 2 is to multiply the equation by the denom inator term (z - l)(z - 0.5). Thus we obtain z = ( z - 0 .5 M i + ( z - 1 ) A 2 (3.4.18)

Now if we set z = p\ = 1 in (3.4.18), we eliminate the term involving A2. Hence 1 = (1 -0 .5 )A , Thus we obtain the result A i = 2. Next we return to (3.4.18) and set z = p 2 = 0.5, thus eliminating the term involving Ai, so we have 0.5 = ( 0 .5 - 1)A2 and hence Ai = 1. Therefore, the result of the partial-fraction expansion is X(z) 2 z- 1 1 z - 0.5 (3.4.19)

T h e exam ple given ab ove suggests that w e can determ ine the coefficients A \, A i , . . . , Afj , by m ultiplying b oth sides o f (3.4.15) by each o f the term s (z - Pk ). k = 1 , 2 , , . . , N , and evaluating the resulting exp ression s at the corresp on d ing pole p osition s, p \ , p i .........P n T h u s w e have, in general,

(Z- Pl)X(;) = (z-wM.i+ ... + /lt+,..+ fa-P)^

z C onsequently, w ith z = Pk, Ak = Exam ple 3.4.6 Determ ine the partial-fraction expansion of 1 17- ^ 0 .5 ,z - PI z - Pn Jtth coefficient as k =

(3420)
(3.4.21)

(3.4.20) yield s the

z \z-pl

( z ~ ^ )X (; )i

1, 2. .

,3A22)

Sec. 3.4

inversion of the z-Transform

191

Solution To eliminate negative powers o f ; in (3.4.22), we multiply both num erator and denom inator by Thus X(z) z+ 1 The poles of X(z) are complex conjugates
Px = \ + J i

z2 - ; + 0.5

and
Pi
=

\ ~ j \

Since p\ ^ p 2- we seek an expansion of the form (3.4.15). Thus

* (:) ; + l

A]

Ai

(z-p i)(:-p 2)

z-P\

z-pi

To obtain A , and A2, we use the formula (3,4.21), Thus we obtain (z-pi)X(z) At = ----------( z - p 2)X(z) A- = -----------; + 1 I
- P

?+

2 \ ^ P1

3+ J5-5+./5
1 1

; + l
-

Pi

:= P 2

T h e ex p a n sio n (3,4.15) a n d the fo rm u la (3.4.21) h o ld fo r b o th real a n d c o m p lex p o les. T h e o n ly c o n s tra in t is th a t all p o les be d istin ct. W e also n o te th at A; = A*. It can b e easily seen th a t th is is a c o n s e q u e n c e o f th e fact th a t p 2 = p ' . In o th e r w o rd s, comp l ex- conj ugat e pol es result in c o mpl ex- con j ugat e coefficients in the part ial-fraction expansi on. T h is sim ple resu lt will p ro v e v ery u se fu l la te r in o u r d iscu ssio n . M u ltip le - o rd e r p o l e s . If X U) h a s a p o le o f m u ltip lic ity /, th a t is, it co n ta in s in its d e n o m in a to r th e fa c to r (z - pk )1, th e n th e e x p a n s io n (3.4.15) is n o lo n g er tru e. In th is case a d iffe re n t ex p an sio n is n e e d e d . F irst, w e in v e stig a te th e case of a d o u b le p o le (i.e., 1 = 2). Exam ple 3.4.7 D eterm ine the partial-fraction expansion of

Solution

First, we express (3.4,23) in terms of positive powers of

in the form

* ( ;) = z2 : (z + 1)(; - l ) 2 X (z) has a simple pole at p\ = - 1 and a double pole pi = p$ = 1. In such a case the appropriate partial-fraction expansion is
= ______ t ______ = + (3424)

z (z + l)(z - 1)2 z+ 1 z - l (z -1 )2 The problem is to determ ine the coefficients A 1, A2, and A3.

192

The z-Transform and Its Application to the Analysis of LTI Systems

Chap. 3

We proceed as in the case of distinct poles. T o determ ine Aj, we multiply both sides of (3.4.24) by (z + 1) and evaluate the result at z = - 1 . Thus (3.4.24) becomes
------------------ = Ai + ------ - A 2 + ------- r - z A i

(z + l)X (z) z

z+ 1

z+ 1

z-1

( z - 1)2 1 4

which, when evaluated at z = - 1 , yields

A i = ------------------

(z + 1)X(z) z

Next, if we multiply both sides of (3.4.24) by (z - l ) 2, we obtain =

Z z+ 1

<3.4.25,

Now, if we evaluate (3.4.25) at z = 1, we obtain A 3. Thus (z l) 2X(z) I = 1

* Li 2 The remaining coefficient Az can be obtained by differentiating both sides of (3.4.25) with respect to z and evaluating the result at z = 1. Note that it is not necessary formally to carry out the differentiation of the right-hand side of (3.4.25