wei SIGNAL att tt TE ee Signals and Systems (ee GR Or elit : S Salivahanan i) Te C GnanapriyaInformation contained in this work has been obtained by Tata McGraw-Hill, from sources believed to be reliable. However, neither Tata McGraw-Hill nor its authors guarantee the accuracy or completeness of any information published herein, and neither Tata McGraw-Hill nor its authors shall be responsible for any errors, omissions, or damages arising out of use of this information. This work is published with the understanding that Tata McGraw-Hill and its authors are supplying information but are not attempting to render engineering or other professional services. If such services are required, the assistance of an appropriate professional should be sought. Tata McGraw-Hill © 2000, Tata McGraw-Hill Publishing Company Limited 21* reprint 2007 DZLCRRXYRCYZR No part of this publication may be reproduced in any form or by any means without the prior written permission of the publishers This edition can be exported from India by the publishers, Tata McGraw-Hill Publishing Company Limited ISBN 0-07-463996-X Published by Tata McGraw-Hill Publishing Company Limited, 7 West Patel Nagar, New Delhi 110 008, typeset at Anvi Composers, A1/33 Pashchim Vihar, New Delhi 110 063 and printed at A P Offset Pvt. Ltd., Naveen Shahdara, Delhi 110 032 Rael ee es ad aContents Foreword D Preface vit 1._ Classification of Signals and Systems i: 1.1___Introduction 1 1.2 Classification of Signals 3 ‘Singularity Functions 9 1.4 Amplitude and Phase Spectra 15 1.5 Classification of Systems 17 1.6 Simple Manipulations of Discrete-time Signals 21 1.7___ Representations of Systems 23 1.8 Analog-to-Digital Conversion of Signals 28 Review Questions 37 2. Fourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 40 2% Tatrodueti 0 2.2 Trigonometric Fourier Series 41 2.3 Complex or Exponential form of Fourier Series 52 2.4 Parseval’s Identity for Fourier Series 58 2.5 Power Spectrum of a Periodic Function 59 2.6 Fourier Transform 62 2.7 Properties of Fourier Transform 64 2.8 Fourier Transform of Some Important Signals 75 2.9 Fourier Transform of Power and Energy Signals 103 Review Questions 119 3._ Applications of Laplace Transform to System Analysis 127 3.1__Introduction 127 3.2 Definition _128 3.3 Region of Convergence (ROC) 128 3.4 Laplace Transforms of Some Important Functions 129 3.5 Initial and Final Value TE 3 3.6 Convolution Integral 138 3.7__ Table of Laplace Transforms 142 3.8 Partial Fraction Expansions 144 Copyrighted materialx Contents 3.10 _s-plane Poles and Zeros _147 3.11 Laplace Transform of Periodic Functions 154 3.12 Application of Laplace Transformation in Analysing Networks 157 Review Questions 183 4, z-Transforms 193 4.1 Introduction _193 2 Definition of the 24 fe 196 4.3 Properties of z-transform 203 Review Questions 228 5._Linear Time Invariant Systems 236 5.1 Introduction 236 5.2 Properties of a DSP System _238 5.3 Difference Equation and its Relationship with System Function, Impulse Response and Frequency Response 256 5.4 Frequency Response 260 Review Questions 272 6.3 Discrete-Time Fourier Transform (DTFT) 305 6.4 ‘ast Fourier Transform (FFT) 319 6.5___ Computing an Inverse DFT by Doing a Direct DFT 344 6.6 Composite-radix FFT 352 Review Questions 376 7. Finite Impulse Response (FIR) Filters 380 2.1 __Introduction 380 7.2 Magnitude Response and Phase Response of Digital Filters 381 7.3 Frequency Response of Linear Phase FIR Filters 384 7.4 Design Techniques for FIR Filters 385 7.5 Design of Optimal Linear Phase FIR Filters 409 Review Questions 414 8. Infinite Impulse Response (IIR) Filters 417 81 Introduction 417 8.2 __ IIR Filter Design by Approximation of Derivatives 418 Copyrighted materialContents xi 8.3 IIR Filter Design by Impulse Invariant Method 423 84 IIR Filter Design by the Bilinear Transformation 427 8.5 Butterworth Filters 432 8.6 Chebyshev Filters 439 8.7 Inverse Chebyshev Filters 444 8.8 Elliptic Filters 445 8.9 Frequency Transformation 446 Review Questions 450 9. Realisation of Digital Linear Systems 453 10. il. 9.1 Introduction 453 9.2 Basic Realisation Block Diagram and the Signal-flow Graph 453 9.3 Basic Structures for IIR Systems 455 9.4 Basic Structures for FIR Systems 482 Review Questions 489 Effects of Finite Word Length in Digital Filters 496 10.1 Introduction 496 10.2_Rounding and Truncation Errors 496 10.3 Quantisation Effects in Analog-to-Digital Conversion of Signals 499 10.4 Output Noise Power from a Digital System 502 10.5 Coefficient Quantisation Effects in Direct Form Realisation of IR & 505 10.6 Coefficient Quantisation in Direct Form Realisation of FIR Fi 508 10.7__Limit Cycle Oscillations 510 10.8 Product Quantisation 513 10.9 Scaling 518 10.10 Quantisation Errors in the Computation of DFT 519 Review Questions 621 Multirate Digital Signal Processing 523 11.1 Introduction 523 11.2 Sampling 524 11.3 Sampling Rate Conversion 525 11.4 Signal Flow Graphs 535 11.5 Filter Structures 539 11.6 Polyphase Decomposition 541 11.7 Digital Filter Design 551 11.8 Multistage Decimators and Interpolators 555 11.9 Digital Filter Banks 565 11.10 Two-channel Quadrature Mirror Filter Bank 572 11.11 Multilevel Filter Banks 578xii_ Contents 12. 13. 14, 15. Spectral Estimation 584 12.1 Introduction 584 12.2 Energy Density Spectrum 584 | 12.3 Estimation of the Autocorrelation and Power Spectrum | of Random Signals 586 12.4 DFT in Spectral Estimation 591 12.5 Power Spectrum Estimation: Non-Parametric Methods 593 12.6 Power Spectrum Estimation: Parametric methods 606 Review Questions 628 Adaptive Filters 631 13.1 Introduction 631 13.2 _ Examples of Adaptive filtering 637 13.3 The Minimum Mean Square Error Criterion 643 13.4 The Widrow LMS Algorithm 645 13.5 Recursive Least Square Algorithm 647 13.6 The Forward-Backward Lattice Method 650 183.7__Gradient Adaptive Lattice Method 654 Review Questions 655 Applications of Digital Signal Processing 658 14.1 Introduction 658 14.2 Voice Processing 658 14.3 Applicationsto Radar 671 14.4 Applications to Image Processing 673 14.5 Introduction to Wavelets 675 Review Questions 686 MATLAB Programs 688 15.1 Introduction 688 15.2 Representation of Basic Signals 688 15.3 Discrete Convolution 691 15.4 Discrete Correlation 693 15.5 StabilityTest 695 15.6 Sampling Theorem 696 15.7 Fast Fourier Transform 699 15.8 Butterworth Analog Filters 700 15.9 Chebyshev Type-1 Analog Filters 706 15.10 Chebyshev Type-2 Analog Filters 712 15.11 Butterworth Digital IIR Filters 718 15,12 Chebyshev Type-1 Digital Filters 724 15.13 Chebyshev Type-2 Digital Filters 729 15.14 FIR Filter Design Using Window Techniques 735 15.15 Upsampling a Sinusoidal Signal 750 15.16 Down Sampling a Sinusoidal Sequence 750Contents xiii 15.17 Decimator 751 15.18 Estimation of Power Spectral Density (PSD) 751 15.19 PSD Estimator 752 15.20 Periodogram Estimation 753 15.21 State-space Representation 753 15.22 Partial Fraction Decomposition 753 15.23 Inverse z-transform 754 15.24 Group Delay 754 15.25 Overlap-add Method 755 15.26 IIR Filter Design-impulse Invariant Method 756 15.27 IIR Filter Design-bilinear Transformation 756 15.28 Direct Realisation of IIR Digital Filters 756 15.29 Parallel Realisation of IIR Digital Filters 757 15.30 Cascade Realisation of Digital IIR Filters 757 15.31 Decimation by Polyphase Decomposition 758 15.32 Multiband FIR Filter Design 758 15.33 Analysis Filter Bank 759 15.34 Synthesis Filter Bank 759 15.35 Levinson-Durbin Algorithm 759 15.36 Wiener Equation’s Solution 760 15.37 Short-time Spectral Analysis 760 15.38 Cancellation of Echo produced on the Telephone—Base Band Channel 761 15.39 Cancellation of Echo Produced on the Telephone—Pass Band Channel 763 Review Questions 765 Appendix A 773 Appendix B 774 Appendix C 782 Index. 802Chapter 1 Classification of Signals and Systems 1.1 INTRODUCTION Signals play a major role in our life. In general, a signal can be a function of time, distance, position, temperature, pressure, etc., and it represents some variable of interest associated with a system. For example, in an electrical system the associated signals are electric current and voltage. In a mechanical system, the associated signals may be force, speed, torque, etc. In addition to these, some examples of signals that we encounter in our daily life are speech, music, picture and video signals. A signal can be represented in a number of ways. Most of the signals that we come across are generated naturally. However, there are some signals that are generated synthetically. In general, a signal carries information, and the objective of signal processing is to extract this information. Signal processing is a method of extracting information from the signal which in turn depends on the type of signal and the nature of information it carries. Thus signal processing is concerned with representing signals in mathematical terms and extracting the information by carrying out algorithmic operations on the signal. Mathematically, a signal can be represented in terms of basic functions in the domain of the original independent variable or it can be represented in terms of basic functions in a transformed domain. Similarly, the information contained in the signal can also be extracted either in the original domain or in the transformed domain. A system may be defined as an integrated unit composed of diverse, interacting structures to perform a desired task. The task may vary such as filtering of noise in a communication receiver, detection of range of a target in a radar system, or monitoring steam pressure in a boiler. The function of a system is to process a giv2n input sequence to generate an output sequence.2. Digital Signal Processing It is said that digital signal processing techniques origin in the seventeenth century when finite difference methods, numerical integration methods, and numerical interpolation methods were developed to solve physical problems involving continuous variables and functions. There has been a tremendous growth since then and today digital signal processing techniques are applied in almost every field. The main reasons for such wide applications are due to the numerous advantages of digital signal processing techniques. Some of these advantages are discussed subsequently. Digital circuits do not depend on precise values of digital signals for their operation. Digital circuits are less sensitive to changes in component values. They are also less sensitive to variations in temperature, ageing and other external parameters. In a digital processor, the signals and system coefficients are represented as binary words. This enables one to choose any accuracy by increasing or decreasing the number of bits in the binary word. Digital processing of a signal facilitates the sharing of a single processor among a number of signals by time-sharing. This reduces the processing cost per signal. Digital implementation of a system allows easy adjustment of the processor characteristics during processing. Adjustments in the processor characteristics can be easily done by periodically changing the coefficients of the algorithm representing the processor characteristics. Such adjustments are often needed in adaptive filters. Digital processing of signals also has a major advantage which is not possible with the analog techniques. With digital filters, linear phase characteristics can be achieved. Also multirate processing is possible only in the digital domain. Digital circuits can be connected in cascade without any loading problems, whereas this cannot be easily done with analog circuits. Storage of digital data is very easy. Signals can be stored on various storage media such as magnetic tapes, disks and optical disks without any loss. On the other hand, stored analog signals deteriorate rapidly as time progresses and cannot be recovered in their original form. For processing very low frequency signals like seismic signals, analog circuits require inductors and capacitors of a very large size whereas, digital processing is more suited for such applications. Though the advantages are many, there are some drawbacks associated with processing a signal in the digital domain. Digital processing needs ‘pre’ and ‘post’ processing devices like analog-to-digital and digital-to-analog converters and associated reconstruction filters. This increases the complexity of the digital system. Also, digital techniques suffer from frequency limitations. For reconstructing a signal from its sample, the sampling frequency must be atleast twice the highest frequency component present in that signal. The available frequency range of operation of a digital signal processor is primarilyClassification of Signals and Systems _ 3 determined by the sample-and-hold circuit and the analog-to-digital converter, and as a result is limited by the technology available at that time. The highest sampling frequency is presently around 1GHz reported by K.Poulton, etal., in 1987. However, such high sampling frequencies are. not used since the resolution of the A/D converter decreases with an increase in the speed of the converter. But the advantages of digital processing techniques outweigh the disadvantages in many applications. Also, the cost of DSP hardware is decreasing continuously. Consequently, the applications of digital signal processing are increasing rapidly. 1.2 CLASSIFICATION OF SIGNALS Signals can be classified based on their nature and characteristics in the time domain. They are broadly classified as (i) continuous-time signals and (ii) discrete-time signals. A continuous-time signal is a mathemati- cally continuous function and the function is defined continuously in the time domain. On the other hand, a discrete-time signal is specified only at certain time instants. The amplitude of the discrete-time signal between two time instants is just not defined. Figure 1.1 shows typical continuous-time and discrete-time signals. x(t) o (a) Continuous-time signal =r (n) aT (b) Discrete-time signal Fig. 1.1 Continuous-Time and Discrete-Time Signals4 Digital Signal Processing Both continuous-time and discrete-time signals are further classified as (i) Deterministic and non-deterministic signals (ii) Periodic and aperiodic signals (iii) Even and odd signals, and (iv) Energy and power signals. 1.2.1 Deterministic and Non-deterministic Signals Deterministic signals are functions that are completely specified in time. The nature and amplitude of such a signal at any time can be predicted. The pattern of the signal is regular and can be characterised mathematically. Examples of deterministic signals are (i) x(¢)=at This is a ramp whose amplitude increases linearly with time and slope is a. (ii) x(t) = A sin wt. The amplitude of this signal varies sinusoidally with time and its maximum amplitude is A. 1 nz0 0 otherwise amplitude is 1 for the sampling instants n 2 0 and for all other samples, the amplitude is zero. For all the signals given above, the amplitude at any time instant can be predicted in advance. Contrary to this, a non-deterministic signal is one whose occurrence is random in nature and its pattern is quite irregular. A typical example of a non-deterministic signal is thermal noise in an electrical circuit. The behaviour of such a signal is probabilistic in nature and can be analysed only stochastically. Another example which can be easily understood is the number of accidents in an year. One cannot exactly predict what would be the figure in a particular year and this varies randomly. Non-deterministic signals are also called random signals. 1.2.2 Periodic and Aperiodic Signals A continuous-time signal is said to be periodic if it exhibits periodicity, ie. (iti) x(n) = { This is a discrete-time signal whose x(t+T)=x(t), -e0 t) dt = 2 Je { <0 (1.20) (1.19) Since, the area of the impulse function is all concentrated at ¢ = 0, for any value of ¢ < 0 the integral becomes zero and for ¢ > 0, from Eq.1.18, the value of the integral is unity. The integral of the impulse function is also a singularity function and called the unit-step function and is represented as 0, ¢<0 t)= 1.21 u(t) a t>0 (1.21) The value at t = 0 is taken to be finite and in most cases it is unspecified. The discrete-time unit-step signal is defined as 0, n + — 1 2 4 A a (a) Unit-imputse function a(t) u(r) A A 3 a] | | / . -LTILT -_— OH or _ 0 t -3 2 aT o4 nr (b) Unit-step function ag) n(n) ; . 34 3 2{ 2 | Slope = 1 +4 ‘| t > | _ o 123 t -3 -2 -10 1 2 3 4a (c) Unit-ramp function A mo | | | -05 0 06 1 () Unit-pulse function Fig. 1.4 Singularity Functions (a) Unit-Impulse Function (b) Unit-Step Function (c) Unit-Ramp Function (d) Unit-Pulse FunctionClassification of Signals and Systems 13 Proof fixe) B(t — t.)] = x(t) 5(t — tp) + E(t) Blt — ty) = x(t) S(t — ty) + X (ty) B(t ty), ty < ty < ty Integrating, we get ty d ty . ty J Gest -t)1 dt = f (x(t) 5(¢~ to de + f Lilt) 8(t - to) dt i i 4 a (x(t) 8(¢ - ty)]? = { x(t) 8(t to) dt + (ty) LHS =0. o . Therefore, { x(t) 5(t - ty) dt + £(fq) = 0 4 t ie. j x(t) 5(t = ty) dt = ~ x (ty) 4 Similarly, eo J x0 5G -t) dt = 8ty) t Hence, j x(t) 8" (¢ = ty) dé = (-1)" x(t) 4 1.3.6 Representation of Signals In the signal given by x(at + b), i.e., x(a(t + b/a)), a is a scaling factor and b/a is a pure shift version in the time domain. If b/a is positive, then the signal x(t) is shifted to left. If b/a is negative, then the signal x(t) is shifted to right. If a is positive, then the signal x(t) will have positive slope. If a is negative, then the signal x(t) will have negative slope. Ifa is less than 0, then the signal x(t) is reflected or reversed through the origin. If |a| < 1, x(t) is expanded, and if |a| > 1, x(t) is compressed. Sketch the following signals (a) x(t) = M1(2t + 3) (b) x(t) = 21 (¢ - 1/4) (c) x(t) = cos(20 xt-5x) and = (d) x(t) = r (— 0.5t + 2) Solution (a) M1(2t + 3) = T1(2¢ + 3/2)) Here the signal shown in Fig. E1.2(a) is shifted to left, with centre at ~ 3 /2. Since a = 2, i.e. {a| > 1, the signal is compressed. The signal width becomes 1/2 with unity amplitude.14 Digital Signal Processing x() A -14 04 G4 Fig. E1.2 (a) Fig. E1.2 (6) (b) x(t) = 2M(¢ - 1/4) Here the signal shown in Fig. E1.2(b) is shifted to the right, with centre at 1 /4. Since a = 1, the signal width is 1 and amplitude is 2. (c) x(t) = cos(20 xt - 5x) = o«(20 (1-84) -(2r(-2) Here the signal x(t) shown in Fig. E1.2(c) is shifted by quarter cycle to the right. Fig. E1.2(c) (d) x(t)=r(—0.5t + 2) x0 -r(-05(+-2)} 0.5 =r(-0.5 (t- 4)) 2 The given ramp signal is reflected through the origin and 0 4 f shifted to right at t = 4. Fig. E1.2 (d) The signal is expanded by x = 2. When ¢ = 0, the magnitude of the signal x(t) = 2, shown in Fig, E1.2(d).Classification of Signals and Systems _\5 Ez Write down the corr nding equation for the given signal. x(t) Fig. E1.3 Solution Representation through addition of two unit step functions The signal x (t) can be obtained by adding both the pulses, i.e. x(t) = 2[u(t) — u(t — 2)]+{u(t — 3) — u(t - 5)) Representation through multiplication of two unit step functions x(t) = 2(u(t) u(—t + 2)] + [u(t - 3) u(-t + 5)] = 2(u(t) u(2 - t) + u(t — 3) u(5 - 2) 1.4 AMPLITUDE AND PHASE SPECTRA Let us consider a cosine signal of peak amplitude A, frequency f and phase shift ©, in order to introduce the concept of amplitude and phase spectra, i.e., x(t) = A cos (2nft + >) (1.27) The amplitude and phase of this signal can be plotted as a function of frequency. The amplitude of the signal as a function of frequency is referred to as amplitude spectrum and the phase of the signal as a function of frequency is referred to as phase spectrum of the signal. The amplitude and phase spectra together is called the frequency spectrum of the signal. The units of the amplitude spectrum depends on the signal. For example, the unit of the amplitude spectrum of a voltage signal is measured in volts, and the unit of the amplitude spectrum of a current signal is measured in amperes. The unit of the phase spectrum is usually radians. The frequency spectrum drawn only for positive values of frequencies alone is called a single-sided spectrum. The cosine signal can also be expressed in phasor form as the sum of the two counter rotating phasors with complex-conjugate magnitudes, 1.e.16 Digital Signal Processing j2Aft+o) , o- jax ft+o) x()= ae 2 From this the amplitude spectrum for the signal x(t) consists of two components of amplitude, viz. A/2 at frequency ‘f and A/2 at frequency “f’. Similarly, the phase spectrum also consists of two phase components one at ‘f’ and the other at ‘“-/’. The frequency spectrum of the signal, in this case, is called a double-sided spectrum. The following example illustrates the single-sided and double-sided frequency spectra of a signal. Sketch the single-sided and double- and phase spectra of the signal led amplitude xit)=8sin (20m -2),-w 0, The unit-impulse function is nothing but the derivative of the unit-step signal. Therefore, the impulse response of the system can also be obtained by computing the derivative of the step response of the system. 1.7.3 State-Variable Technique The state-variable technique provides a convenient formulation procedure for modelling a multi-input, multi-output system. This technique also facilitates the determination of the internal behaviour of the system very easily. The state of a system at time fy is the minimum information necessary to completely specify the condition of the system at time fy and it allows determination of the system outputs at any time t > to, when inputs upto time ¢ are specified. The state of a system at time fy is a set of values, at time fo, of a set of variables. These variables are called the state variables. The number of state variables is equal to the order of the system. The state variables are chosen such that they correspond to physically measurable quantities. It is also convenient to consider an n-dimensional space in which each coordinate is defined by one of the state variables x,, x», ..., x,, where n is the order of the system. This n-dimensional space is called the state space. The state vector is an n-vector x whose elements are the state variables. The state vector defines a point in the state space at any time t. As the time changes, the system state changes and a set of points, which is nothing but the locus of the tip of the state vector as time progresses, is called a trajectory of the system. A linear system of order n with m inputs and & outputs can be represented by n first-order differential equations and & output equations as shown below. a $y, Xt Oyg XQ to. + yy Xy t Oy, Uy t Ogu gt... + Dim Um oe Sg Xy + Ogg Xyt -1. + Ayn X qt Oy, Uy + Ogg U gt 0. + bam Um : (1.84) as, dt & Gy Xy + Aggy Lot... + Pay Xp t Ony Uy + Ong gt «.. + Dam UmClassification of Signals and Systems 27 and Wy =O yy Ny + C yy XQ F oe + Cy Xp_t dy Uy + Cyglgt ... + dy_, Up, Yo = C21 Xy + Cog Xyt ... + Con Xp + Ay Uy + Tog lg t ... + dom Um . (1.35) Ye = Ce Xp + Cyg Xo t .-. + Cyn Xpq + Ap Uy + Ago lg + ... + Tg Um where uj, i = 1, 2, ..., m are the system inputs, x,, i = 1, 2,3, ...,n are called the state variables and y;, i = 1, 2, 3, ...,# are the system outputs. Equations 1.34 are called the state equations, and Eqs 1.35 are the output equations. Equations 1.34 and 1.35 together constitute the state- equation model of the system. Generally, the a’s, b’s, c’s and d’s may be functions of time. The solution of such a set of time-varying state equations is very difficult. If the system is assumed to be time-invariant, then the solution of the state equations can be obtained without much difficulty. The state variable representation of a system offers a number of advantages. The most obvious advantage of this representation is that multiple-input, multiple-output systems can be easily represented and analysed. The model is in the time-domain, and one can obtain the simulation diagram for the equations directly. This is of much use when computer simulation methods are used to analyse the system. Also, a compact matrix notation can be used for the state model and using the laws of linear algebra the state equations can be very easily manipulated. For example, Eqs 1.34 and 1.35 expressed in a compact matrix form is shown below. Let us define vectors % uy vn x=|"|, u=|"? |, y=]? (1.36) Xn Um 'h and matrices My, Hg + + Oy by bp + + + bum Gy qn. + + Aan Bay bag. bam A=|. .... .) Bel. . ... . | aan Gq, Ome - + + Onn bay baa + + + Onm Cy Ces es Cy dy dig ss» dim Co Cg = + + Con dy dy... dom C= | D= a eo ny App - » » dem28 Digital Signal Processing Now, Eqs 1.34 and 1.35 can be compactly written as a =Ax+Bu (1.38a) y=Cx+Du (1.38b) where # = dx/dt. Equations 1.38 may be illustrated schematically as shown in Fig.1.10. The double lines indicate a multiple-variable signal flow path. The blocks represent matrix multiplication of the vectors and matrices. The integrator block consists of n integrators with appropriate connections specified by the A and B matrices. z is Fig. 1.10 Block Diagram of the State-Variable Model of Eq. 1.38 State Equations for Discrete-time Systems For a discrete-time system, the state equations form a set of first-order difference equations constituting a recursion relation. This recursion relation allows determination of the state of a system at the sampling time kT from the state of the system and the input at the sampling time (k — DT, where k is an integer. The state-equations for a discrete-time system can be modelled as shown below. X41 = Fx + Gu, (1.39a) 9, = Hx, + Ju, (1.39b) The dependence of these parameters on Tis suppressed for simplicity. For a single input, single output system u, and y, are scalars and G and H become vectors g and h, and d is a null in most cases. The state- variable modelling of a discrete-time system finds application in the digital simulation of a continuous time systems. 1.8 ANALOG-TO-DIGITAL CONVERSION OF SIGNALS A discrete-time signal is defined by specifying its value only at discrete times, called sampling instants. When the sampled values are quantised and encoded, a digital signal is obtained. A digital signal can be obtained from the analog signal by using an analog-to-digital converter. In the following sections the process of analog-to-digital conversion isClassification of Signals and Systems _29 discussed in some detail and this enables one to understand the relationship between the digital signals and discrete-time signals. Figure 1.11 shows the block diagram of an analog-to-digital converter. The sampler extracts the sample values of the input signal at the sampling instants. The output of the sampler is the discrete-time signal with continuous amplitude. This signal is applied to a quantiser which converts this continuous amplitude into a finite number of sample values. Each sample value can be represented by a digital word of finite word length. The final stage of analog-to-digital conversion is encoding. The encoder assigns a digital word to each quantised sample. Sampling, quantizing and encoding are discussed in the following sections. TC ——) f Tf | | >| Sampler | > Quantiser> >| Encoder {> er el | om | i Continuous-time Discrete-time Discrete-time Digital output continuous-amplitude Continuous-amplitude discrete-amplitude signal input signal signal signal Fig. 1.11 Analog-to-Digital Converter 1.8.1 Sampling of Continuous-time Signals Sampling is a process by which a continuous-time signal is converted into a discrete-time signal. This can be accomplished by representing the continuous-time signal x(t), at a discrete number of points. These discrete number of points are determined by the sampling period, 7, i.e. the samples of x(t) can be obtained at discrete points ¢ = nT, where n is an integer. The process of sampling is illustrated in Fig.1.12, The sampling unit can be thought of as a switch, where, to one of its inputs the continuous-time signal is applied. The signal is available at the output only during the instants the switch is closed. Thus, the signal at the output end is not a continuous function of time but only discrete samples. In order to extract samples of x(t), the switch closes briefly every T seconds. Thus, the output signal has the same amplitude as x(t) when the switch is closed and a value of zero when the switch is open. The switch can be any high speed switching device. The continuous-time signal x(t) must be sampled in such a way that the original signal can be reconstructed from these samples. Otherwise, the sampling process is useless. Let us obtain the condition necessary to faithfully reconstruct the original signal from the samples of that signal. The condition can be easily obtained if the signals are analysed in the frequency domain. Let the sampled signal be represented by x,(¢). Then, x, ()= x(t) g(t) (1.40) where g(t) is the sampling function. The sampling function is a continuous train of pulses with a period of T seconds between the pulses, and it models the action of the sampling switch. The sampling function is shown in Fig. 1.12(c) and (d). The frequency spectrum of the sampled30 Digital Signal Processing | i i 0 T aT 3T 4T 5T ert (a) Sampies of x (1) | sven «0 x 29 x() x,() es ott) (b) Modelling a sampler as a switch (c) Mode! of a sampler t 9 eee 1 i f 1 | | | | | ft ft ff Lil aL in i uy» ° T ar 3T aT ST 6T (4) Sampling function Fig. 1.12 The Sampling Process signal x,(t) helps in determining the appropriate values of TJ for reconstructing the original signal. The sampling function g(t) is periodic and can be represented by a Fourier series (Fourier Series and transforms are discussed in Chapter six), i.e. at)= SC, ert (1.41) sie where c,=2 faweve at (1.42) TE is the nth Fourier coefficient of g(t), and f, = 2 is the fundamental frequency of g(t). The fundamental frequency, /f, is also called the sampling frequency. From Eqs.1.40, and 1.41, we have xf =x(t) LC ei?! = FC, xlseirrh! (1.43) nace nase The spectrum of x,(t), denoted by X, (f ), can be determined by taking the Fourier transform of Eq. 1.43, i.e.Classification of Signals and Systems 34 XQ) = fx, (tel! dt (1.44) Using Eq.1.43 in the above equation, XP) = J LCyxle eM eae (1.45) Interchanging the order of integration and summation, Xf) = zG [atye tet -Me ae (1.46) nan iw But from the definition of the Fourier transform fae F086 a = X(F= nf) Thus, x= > C,X¢-nf) (1.47) From Eq. 1.47, it is understood that the spectrum of the sampled continuous-time signal is composed of the spectrum of x(t) plus the spectrum of x(¢) translated to each harmonic of the sampling frequency. The spectrum of the sampled signal is shown in Fig. 1.13. Each frequency translated spectrum is multiplied by a constant. To reconstruct the original signal, it is enough to just pass the spectrum of x(t) and suppress the spectra of other translated frequencies. The amplitude response of such a filter is also shown in Fig. 1.13. As this filter is used to reconstruct the original signal, it is often referred to as a reconstruction filter. The output of the reconstruction filter will be CX(f) in the frequency domain and x(t) in the time-domain. xh =F Gy —2fy -h -hO hf fs Fig. 1.13 Spectrum of Sampled Signal32_Digital Signal Processing The signal x(t), in this case, is assumed to have no frequency components above f,, i.e. in the frequency domain, X(f) is zero for If | 2f,- Such a signal is said to be bandlimited. From Fig.1.13, it is clear that in order to recover X(f) from X,(/), we must have f-h2th or equivalently, f, 2 2f,, hertz (1.48) That is, in order to recover the original signal from the samples, the sampling frequency must be greater than or equal to twice the maximum frequency in x(t). The sampling theorem is thus derived, which states that a bandlimited signal x(t) having no frequency components above f, hertz, is completely specified by samples that are taken at a uniform rate greater than 2f, hertz. The frequency equal to twice the highest frequency in x(t) , i.e. 2f,, is called the Nyquist rate. Sampling by Impulse Function The sampling function g(t), discussed above, was periodic, The pulse width of the sampling function must be very small compared to the period, 7. The samples in digital systems are in the form of a number, and the magnitude of these numbers represent the value of the signal x(t) at the sampling instants. In this case, the pulse width of the sampling function is infinitely small and an infinite train of impulse functions of period T can be considered for the sampling function. That is, g(t)= Y&(e-nT) (1.49) neo» The sampling function as given in Eq.1.49 is shown in Fig,1.14. When this sampling function is used, the weight of the impulse carries the sample value. The sampling function g(t) is periodic and can be represented by a Fourier series as in Eq.1.41, which is repeated here. git)= > 0, where os C, 3 BO) e Mae (1.50) nhs’ Since &(t) has its maximum energy concentrated at ¢ = 0, a more formal mathematical definition of the unit-impulse function may be defined as a functionalClassification of Signals and Systems 33 6 etEt thd het te -t -67 -5T-4T-3T -2T-7 0 T 2T 3T 4T ST 6T (a) Aen Fig. 1.14 (a) impulse Sampling Function (b) Spectrum of the Signal x(t) (c) Spectrum of Impulse Sampled Signal J +0 8) de =2(0) (1.51) where x(t) is continuous at ¢ = 0. Using Eq. 1.51 in Eq. 1.50, we have 1 1 Cie ne aaah (1.52) Thus C, is same as the sampling frequency f,, for all n. The spectrum of the impulse sampled signal, x,(¢) is given by XN=f, DXF - nf) (1.53) ne The spectra of the signal x(¢) and the impulse sampled signal X, (¢) are shown in Figs 1.14 (b) and (c). The effect of impulse sampling is same as sampling with a train of pulses. However, all the frequency translated spectra have the same amplitude. The original signal X(f) can be reconstructed from X,(f) using a low-pass filter. Figure 1.15 shows the effect of sampling at a rate lower than the Nyquist rate. Consider a bandlimited signal x(t), with f, as its highest frequency content, being sampled at a rate lower than the Nyquist rate, i.e., sampling frequency f, < 2f,. This results in overlapping of adjacent34 _Digital Signal Processing Ax /\ . “ho ty (a) Spectrum of the input signal 4 xan AKAALA., “ho ty thy ooh (b) Spectrum of the sampled signal's for f, > 2f, A xs (c) Sampled signal's spectrum for f, < 2f Fig. 1.15 Illustration of Aliasing spectra i.e., higher frequency components of X,(/) get superimposed on lower frequency components as shown in Fig.1.15. Here, faithful reconstruction or recovery of the original continuous time signal from its sampled discrete-time equivalent by filtering is very difficult because portions of X(f - f,) and X(f + f,) overlap X(f), and thus add to X() in producing X,(f). The original shape of the signal is lost due to undersampling, i.e. down-sampling. This overlap is known as aliasing or overlapping or fold over. Aliasing, as the name implies, means that a signal can be impersonated by another signal. In practice, no signal is strictly bandlimited but there will be some frequency beyond which the energy is very small and negligible. This frequency is generally taken as the highest frequency content of the signal. To prevent aliasing, the sampling frequency f, should be greater than two times the frequency /;, of the sinusoidal signal being sampled. The condition to be satisfied by the sampling frequency to prevent aliasing is called the sampling theorem. In some applications, an analog anti- aliasing filter is placed before sample/hold circuit in order to prevent the aliasing effect.Classification of Signals and Systems 35 A useful application of aliasing due to undersampling arises in the sampling oscilloscope, which is meant for observing very high frequency waveforms. 1.8.2 Signal Reconstruction Any signal x(t) can be faithfully reconstructed from its samples if these samples are taken at a rate greater than or equal to the Nyquist rate. It can be seen from the spectrum of the sampled signal, X,(¢) that it consists of the spectra of the signal and its frequency translated harmonics. Thus, if the spectrum of the signal alone can be separated from that of the harmonics then the original signal can be obtained. This can be achieved by filtering the sampled signal using a low-pass filter with a bandwidth greater than f;, and less than f, — f,. hertz. If the sampling function is an impulse sequence, we note from Eq.1.53 that the spectrum of the sampled signal has an amplitude equal to f,= VT. Therefore, in order to remove this scaling constant, the low-pass filter must have an amplitude response of I/f, = T. Assuming that sampling has been done at the Nyquist rate, i.e. 2f,, the bandwidth . of the low-pass filter will bef, = & . Therefore, the unit impulse response of an ideal filter for this bandwidth is fa A(t)=T fel? af (1.54) ~fyl2 That is A(t) = —D_ (eit ft _ ¢-J8ht) jane The above expression can be alternatively written as sin & f,t aft The ideal reconstruction filter is shown in Fig.1.16a. The input to this filter is the sampled signal x(nT) and the output of the filter is the reconstructed signal x(t). The output signal x(t) is given by h(t) = Tf, =sinc f,t (1.55) x(t)= Sx(nT)h(t - nT) n= Using Eq.1.55, we get x(t)= ¥ x(nT)sine f,(t - nT) (1.56) naae The above expression is a convolution expression and the signal x(t) is reconstructed by convoluting its samples with the unit-impulse response of the filter. Eq. 1.56 can also be interpreted as follows. The36 Digital Signal Processing original signal can be reconstructed by weighting each sample by a sinc function and adding them all. This process is shown in Fig. 1.16b. > x(nT)&(t= AT) | | new > | Ideal reconstruction ier x(0) pone ‘Sinc functions > (n-3)T (n=2)7 (a-1)T nT (ns )P (n42)7 (ns 3)T (b) Fig. 1.16 Signal Reconstruction (a) Reconstruction Filter (b) Time Domain Representation 1.8.3 Signal Quantisation and Encoding A discrete-time signal with continuous-valued amplitudes is called a sampled data signal, whereas a continuous-time signal with discrete- valued amplitudes is referred to as a quantised boxcar signal. Quantisation is a process by which the amplitude of each sample of a signal is rounded off to the nearest permissible level. That is, quantisation is conversion of a discrete-time continuous-amplitude signal into a discrete-time, discrete-valued signal. Then encoding is done by representing each of these permissible levels by a digital word of fixed wordlength. The process of quantisation introduces an error called quantisation error and it is simply the difference between the value of the analog input and the analog equivalent of the digital representation. This error will be small if there are more permissible levels and the width of these quantization levels is very small. In the analog-to-digital conversion process, the only source of error is the quantiser. Even if there are more quantisation levels, error can occur if the signal is at its maximum or minimum value for significant time intervals. Figure 1.17 shows how a continuous-time signal is quantised in a quantiser that has 16 quantising levels.Classification of Signals and Systems 37 Quantisationtevet | Encoded | 15 Wit 14 1110 13 1101 12 1100 " 1011 10 1010 9 1001 8 1000 7 0111 6 o110 5 0101 4 |_ e100 3 0011 2 0010 1 0001 | a A 0 F aT 3T 4T Fig. 1.17 Quantizing and Encoding rola REVIEW QUESTIONS 1.1 What are the major classifications of signals? 1.2 With suitable examples distinguish a deterministic signal from | a random signal. 1.3 What are periodic signals? Give examples. 1.4 Describe the procedure used to determine whether the sum of two periodic signals is periodic or not. 1.5 Determine which of the following signals are periodic and determine the fundamental period. (a) x,(t) = 10 sin 25 nt (b) xo(t) = 10 sin VB nt38 Digital Signal Processing (c) x3(t) = cos10nt (d) x(t) = x,(t) + x,ft) (e) x5(t) =x,(t) + x3 (t) (D x4(t) = xo(t) + x(t) 1.6 What are even signals? Give examples. 1.7 What are odd signals? Give examples. 1.8 What is energy signal? 1.9 What is power signal? 1.10 What are singularity functions? 1.11 Define unit-impulse function? 1.12 What is unit-step function? How it can be obtained from an unit-impulse function? 1.13 What is unit-ramp function? How it can be obtained from an unit-impulse function? 1.14 What is pulse function? 1.15 Evaluate (a) fe“? st-10)at ) fe Se+arae @ i 4oe** 5(t—-10)dt and (a) { e*” 6(t—10)a¢ Ans (a)e""* (8)0— (c) 40.077 (de 2 1.16 Explain the terms single-sided spectrum and double-sided spectrum with respect to a signal. 1.17 Sketch the single-sided and double-sided frequency spectra of the signals (a) x(t) = 10 sin (tome -25), —e > oO Tr \ ¥ t 2T t (b) “A an (a) A | | te > 0 Ti2 T t {o) Fig. 2.1, Waveforms Representing Periodic FunctionsFourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 41 Examples of periodic processes are the vibration of a tuning fork, oscillations of a pendulum, conduction of heat, alternating current passing through a circuit, propagation of sound in a medium, etc. Fourier series may be used to represent either functions of time or functions of space co-ordinates. In a similar manner, functions of two and three variables may be represented as double and triple Fourier series respectively. Periodic waveforms may be expressed in the form of Fourier series. Non-periodic waveforms may be expressed by Fourier transforms. 2.2 TRIGONOMETRIC FOURIER SERIES A periodic function f(t) can be expressed in the form of trigonometric series as F(8) = Fag + ay €05 Wnt + ay 608 2p t+ ay c05 Bay f+ see + 6, sin @o t+ by sin 2m) t + by sin Bayt + ... (2) where @) = 2nf= 3, f is the frequency and a’s and 6’s are the coefficients. The Fourier series exists only when the function f(t) satisfies the following three conditions called Dirichlet’s conditions. (i) f(t) is well defined and single-valued, except possibly at a finite number of points, i.e. f @ has a finite average value over the period T. (ii) f(¢) must posses only a finite number of discontinuities in the period T. (iii) f(t) must have a finite number of positive and negative maxima in the period T. Equation 2.1 may be expressed by the Fourier series fit)= 4 + Ya, cosnayt+ Sb, sin nat (2.2) n=l n=l where a, and b,, are the coefficients to be evaluated. Integrating Eq. 2.2 for a full period, we get riz 1 TI2 TI2 [fat = Say fat+ [ Ya, cos nage +b, sin n apt) dt -T/2 2 -TI2 -Ti2n=1 Integration of cosine or sine function for a complete period is zero. Tr. Therefore, fre dt = i aT’ -72 8 9 Ti Hence, ay== ffiede (2.3) “Tia42__Digitat Signal Processing T or, equivalently ag = a f(t)dt 0 Multiplying both sides of Eq. 2.2 by cos mm@pt and integrating, we have T/2 1 TZ J f()cos Mot dt = 3 Jao cos M@ot dt + -TI2 -TI2 TI2 @ TI2 J Sa, cos nag tcos maot d+ f >.b, sin naot cos mut dt _Tign=1 -Tign=1 1 TI Here, 2 fay cos m Wot dt = 0 -T2 TI2 a, Ti2 fon COS N Wot COS Mm@g dt = —" Jleos (m +N) t + cos (m — n) Wot] dt -TI2 2 -TI2 0, forme#n = Fa ny form=n TI2 bn TI2 Jen Sin N@ot Cos Mwytdt = — Jisin (m + n)@ot - sin (m - n)wot] dt -T/2 2 -T/2 =0 TI2 Ta Therefore, J £2) cos neoot dt = “,form=n -Te 2 TI2 Hence, a,= = J f(t) cos nagt dt (2.4) T ing in or, equivalently a, = 2 | fa)008 NW t dt 0 Similarly, multiplying both sides of Eq. 2.2 by sin m @)f and integrating, we get TI2 1 Ti? Jf) sin moo tat == [agsin magt dt -T/2 2 ig fea Th. + J Xa, cos nwot sin may t de + J X4, sin nwot sin mag t dt -Tign=1 -TIg n=.