Published by : LAXMI PUBLICATIONS (P) LTD 22, Golden House, Daryagani, ' New Dethi-110002, 011.23 26 23 68 Phones: | 011-23 26 23 70 _ for1.2325 2572 Faxes: {omza2sas 79 Branches : © 129/1, Hrd Main Road, IX Cross Chamrajpet, Bangalore (Phone : 080-26 61 1 61) * 26, Damodaran Street, 'T. Nagar, Chennai (Phone : 044-24 34 47 26) * St. Bonedict’s Road, Cochin (Phone : 0484-239 70 04) + Pan Bazar, Rani Bari, Guwahati (Phones ; 0361-254 36 69, 251 38 81) * 4-2-453, Ist Floor, Ramkote, Hyderabad (Phone : 040-2475 02 47) * Adda Tanda Chowk, N.D. 365, Jalandhar City (Phone : 0181-222 12 72) + 106/A, Ist Floor, S.N, Banerjee Road, Kolkata (Phones ; 093-22 27 37 79, 22 2452 47) + 18, Madan Mohan Malviya Marg, Lucknow (Phone : 0522-290 95 78) © 128A, Block 3, First Floor, Noorani Building, L.J. Road, Mumbai (Phone : 022}24 46 39 98) + Radha Govind Street, Tharpagna, Ranchi (Phone : 0651-230 77 64) EMAIL ; colaxmi@hotmail.com WEBSITE : www. laxmipublications.com ©All! Rights Reserved with the Author and the Publishers, EDS.0652-125-DIGITAL SIGNAL PROCESSING First Edition Price : Rs, 125.00 Oniy. C—9743/08/12 Laser Typesetted at : Goswami Printers, Delhi-110053 Printed at : Ajit Printers, Delhi-110053Contents Chapters Pages De Tn ction essen ennenen tensa snuennsnsunssnsnnesnnsntannansnne 128 1.1 Classification of Signals 1.2_Multi Channel s 8 & 1.4 Continuous-time Versus Discrete-time Signals .. 8 1.5 Frequency Concept is Continuous Time and Discrete Time Signals 1.5.1 Continuons-time Sinusoidal signals 1.5.2 Diserete-time Sinusoidal Signals. 1.5.3 Harmonically Related Complex Exponential: 1.6 Energy and Power Signals (Continuous time-instants) 1.7 Singularity Functions... 1.7.1 Unit-Impulse Function 1.7.2 Unit-Step Function 16 1.7.3 Unit-Ramp Funetion 7 (17.4 Unit-Pulge Pumetion oneness LP 1.8 Energy Signals and Power Signals (Discrete-time instants) . 19 1.9 Signal Processing 24 1,10 Analog Versus Digital Signal Processing Review Questions versus seisiacsivse RII acess snsnssssssnsssnsnnnnannsnsnsnssnsnssaroannssnnsnarssssnansssnasnsssssassssssssnunsnnsnsnesssasassssae 2D 2. Applications of Digital Signal Processing... 2.1 Introduction .. 2.2. Application to Speech Pro 10 10 12 wu 15 16 16 2.2.2 Speech Technology. 2.23 Parameters of Speech .. 2.2.4 Speech Analysis 2.2.5 Speech Coding 2.3 Application to Image Processing 2.3.1 Image Formation and Recording .. 2.3.2 Image Sampling and QuantizatioChapters Pages 2.3.3 Image Compression .. 2.3.4 Image Restoration Review Questions 3. Discrete Time Systems.. 3.1_Discrete-time Signals and Systems . a ition 3.1.2 Representations 3.1.3 Some Elemental 3.2 Classification of Discrete-time Signal 3.3 Sampling 3.4 Real and Complex Sequence... 8.5 Finite and Infinite Sequence . 3.6 Types of Infinite-length Sequence 8.7 Operations on Sequences 3.8 Sampling Rate Alteration 3.9 Classification Based on Symmetry Problem 3.9.1 Periodic Conjugate-symmetric Part and Periodic Conjugate Anti-symmetric Part 4 3.10 Sampling Process 3.11 Classification of Diserete-time Systems 3.12.1 Representation of a Discrete-time Sij 3.12.2 Discrete-time Unit Impulse Response and Convolution Sum Representation of LTI System. 2.13 The Convolution Process iui sasssssestssssssssastsasssns 3.14 Properties of Linear Time-invariant System 3.15 Causality and Stability Condition for LTI Discrete-time Syster 3.16 Classification of LT! System ... I BEB BEBRE Lobehebkbeeeee 3.18 Recursive and Non-recursive Discrete-time System .......... 3.19 Linear Constant Co-efficient Difference Equation 3.20.1 The Homogeneous Solution of a Difference Equation 3.20.2 The Particular Solution of the Difference Equation 3.20.3 The Total Solution the Difference i RBERBBBBBRERRSChapters Pages 3.22 Impulse Response... Review Questions Exercises. 4._Frequency Donain Characterization or Discrete-Time System. 4.1 Fourier Transform of discrete-time Signals 4.1.1 Fouricr Sorics for Discrete-time Periodic Signal 4.1.2 Condition for convergence of Fourier Transform 4.2. Frequency response of Discrete-time Systems 4.3. Properties of Frequency Response 4.4 Polar form of Frequency Response .. 4.5 Frequency Response of First order System. 4.6 Properties of Frequency Response 4.7 2-Transform 4.7.1 Definition of Z-transform 4.7.2 Region of Convergence . 4.73 Properties .. 4.7.4 Some Common One Sided Z-transform Pairs .. 4.8 Inverse Z-transform 48.1 The Inverse Z-transform Using Contour Integration 48,2 The Inverse 2-transform by Power Series Expan: or Via Long Division .. 4.8.3 The Inverse Z-transform by Partial Fraction Expansion 4.9 Solution of Difference Equation Using Z-Transform ... Review Questions Exercises 5. Frequency Analysis of Signals... 5.1 Frequeney Analysis of Continuous-time (Analog) Signals 5.2 Evaluation of Fourier Co-efficients 5.3 Symmetry Conditions for Periodic Signals 5.4 Exponential Fourier Series 5.4.1 Existence of Fourier Series 5.5 Fourier Spectrum .. 5.6 Properties of Continuous-time Fourier Series 5.7 Continuotis-Time Fourier Transform 5.8 Fourier Transform of a Periodic Signal 5.9 Properties of Continuous Time Fourier Transform 5.10 Frequency Domain Representation of Discrete Time Signal and System . 5.10.1 Frequency Analysis of Discrete Time Signals... 92-130 136Chapters Pages 5.10.2 Fourier Series for Diserete Time Periodic Signals 5,10,3 Expression for the Values of the Co-efficient a... 5.11 Discrete Time Fourier Transform .. 5.11.1 Inverse Discrete Time Fourier Transform .. 5.11.2 Condition for Convergence of Fourier Transform. 5.11.3 Energy Density Spectrum ....... 5.114 Properties of Discrete-Time Fourier Transform Review Questions , Exercises... 6.1 Introduction 6.5.1 Methods of Performing Circular Convolution ae a eee ge 1 Overlap Add Method 6.6.2 Overlap Save Method. 6.7 Computation of the DFT of Real Sequences . 6.7.1 N-point DFTs of Two Real Sequences using a Single N-point DFT. 6.7.2 2N-point DFT of a Real Sequence using a Single N-point DFT .. Fast Fourier Transforms Algorithms ..... GBD Dont ai eaten 6.8.2 Radix of FFT Algorithms 6.8.3 Radix-2 Algorithm 6.9 Decimation-in-time FFT Algorithms The 8-point DFT using Radix-2 DIT FFT G.11 Decimation in Frequency (DIF) Radix-2 FFT ... 6.12 Comparison of DIT and DIF .. Review Z Digital Processing of Continuous Signals. 11 Introduction 1.2 Sampling Process... 7.2.4 Analysis of Sampling Process in Frequency Domain ......Chapters 7.3 Sampling Theorem 7.4 Anti Aliasing Filter. 7.8 Signal Reconstruction 7.6 Zero-order Hold ... 7.6.1 Transfer Function of Zero Order Hold 7.7 Sampling of Band Pass Signals .. 7.8 Frequency Selective Filters and Filter Specifications. 7.8.1 Filter Specifications. 7.9 Analog Lowpass Filter Design 7.40 Analog Lowpass Butterworth Filter 7.11 Analog Lowpass Chebyshev Filters 7.11.1 Type-I Chebyshev Approximation .. 7.112 Pole Locations for Chebyshev Filter 7.11.3 Chebyshev Type-HI Filter ..... 7,12 Analog Frequency Transformation 7.13 Design Procedure for Analog Butterworth Lowpass Filter. 7.14 Design Procedure for Analog Chebyshev Lowpass Filter 7.15 Sample and Hold Cireuit.. 1.16 Analog-to-Digital Convertor 1.16.1 Flash A/D Converters . 7.16.2 Serial-Parallel A/D Converter 7.16.3 Successive-approximation A/D Converter 7.16.4 Counting A/D Converter. 7.16.5 Oversampling Sigma-Delta A/D Converter... Digital-to-Analog Converter .... 7.17.1 Weighted-Resistor D/A Converter 2.17.2 Resistor Ladder D/A Converter wuseiuseeueiieaanisaiisssse 2 7.17.8 Oversampling Signal-delta D/A Converter ... Review Questions... Exercises... 8. Digital Filter Structures 8.1 Introduction 295 8.2 System Describing Equations .. 3 Recurai Nom wed 8.4 Block Diagram Representations .... 8.4.1 First Order System Block Diagram Representation .. 8.5 Structure For IIR System 8.5.1 Direct Form Structures .. 8.5.2 Cascade Form Structure TY aChapters 8.6 Structures For FIR Systems... Examination Question Papers Index.Chapters : 1. Introduction 2, Applications of Digital Signal Processing DIGITAL SIGNAL PROCESSINGIntroduction Characterization and Classification of Signals Signal A ‘signal’ is defined as any physical quantity that varies with time, space and any other independent variable or variables. More precisely a signal is a function of a set of independent variables. The signal itself carries some kind of information available for observation. Processing By ‘processing’ we mean operating in some fashion on signal to extract some useful information. Digital ‘The word ‘digital’ shall mean that the processing is done with a digital computer or special purpose digital hardware. Digital Signal Processing Digital signal processing is concerned with the representation of signals by sequence of numbers or symbols and the processing of these sequence. ‘The purpose of such processing may be to estimate characteristic parameters or trans- form a signal into form which is in some sense more desirable. Application Bio-medical engineering, acousties, radar, speech communication, data communication, image processing, nuclear science and many others. 11 CLASSIFICATION OF SIGNALS ‘There are five methods of classifying signals based on different features : (a) Based on independent variable. (b) Depending upon the number of independent variable. (c) Depending upon the certainity by which the signal can be uniquely described. (d) Based on repetition nature. {e) Based on reflection.Digital _ Signal, Processing (a) Based on independent variable. Independent variables can be dpntinuous or dis- crete. | 1. Continuous Time Signal. It is also referred as analog signal i.e., thp signal is repre- sented continuously in time. In simple words, a signal x(t) is said th be a continuous time signal if it is defined for all time. 2. Diserete Time Signal. Signals are represented as sequence at discrete tie intervals. Thus, the independent variable has discrete values only. x0 xin) (a) Continuous time signal (b) Discrete time signal Fig. 1.1 eg. Speech signal is an example of analog signal. A discrete time signal which discrete-valued represented by a finite number of digits is referred to as a “digital signal”. eg. Digitized music signal stored in CD-ROM disk. (6) Depending upon the number of independent variable. (@) L-D Signals. It is a funetion of a single independent variable. e.g. (a) speech signal-independent variable is time, () music signal. (ii) 2-D Signal. It is a function of two independent variables. e.g. Photographie image signal~two independent variables a1 vafiables, Each frame of a black and white video signal is a 2D-image si tion of two discrete spatial variable, with each frame occu discrete instants of time. (iii) M-D Signal. It is a function of ‘M’ independent variable in timp. e.g. Video signal. ‘The black and white video signal ean be considered an example of a 3D three independent variables are two spatial variables and time. A colour video signal is a three-channel signal composed of three 3-Dignals represent- ing the three primary colours : red, green and blue (RGB). For transmission purpose, the RGB television signal is transformed isto another type of 3-channel signal is composed of luminance component and two chrominande components. (c) Depending upon the certainity by which the signal can be uniquely described the two spatial pal that is a fune- ing squentially at signal where the (i) Deterministic Signal. A signal that can be uniquely determinog by a well-defined process such as a mathematical expression or rule, or table Jook-up is called a deterministic signal.Introduction e.g. (a) A sinusoidal signal can be represented as, xt vlt)=¥,, sin of fort2 0. (6) A square signal can be defined as xt)=A for O0 Odd fx(n)) =-12 forn <0 =0 forn =O =U2 forn >0. Thus, Even fx(n)] = ght) + at-nd] (1.3) Odd [x(n = J ten) —at-nd) wf LdDigital |Signal Processing Properties of even and odd signal : 1. The sum of two even signals are even signal. 2. The sum of two odd signals are odd. | 3. The sum of an even signal and an odd signal is neither even norfodd signal. 4, The product of two even signal is even, | 5, The product of two odd signal is even, | 6, The product of even signal and an odd signal is add. | 1.2 MULTI CHANNEL Asignal can be generated by a single source or by multiple sourcesjor multiple sensors. In the former case, it is (single) scalar signal and in the later case it is 4 vector signal, often called a multichannel signal. ‘These type of signals can be represented in vector form as, fx x(t) = [=i] CL.) x;(t) Equation represents a 3-channel signal. eg. In clectrocardiography [ECG] for example 3-lead and 12-leadJelectrocardiographs 1.3 MULTI DIMENSIONAL SIGNALS Ifa signal is a function of a single independent variable, then it is falled as one-dimen- sional signal. Similarly, if signal is a function of N-independent variablks, it is called as N- dimensional signal. eg. * Picture signal is a two dimensional signal, since the intensity Ix, y) is a function of -wo independent variables x and y. e Black and white television picture is an example of 3-dimensjonal signal because brightness I¢x, y, ¢) is a function of three independent variables k, y and ¢ (time). Itis also possible to have multichannel and multidimensional signals simultaneously. For example, a colour TV picture is described by three intengity functions of form 1,(z,y, 0 Iredl, I, Gy, © [green], and I, (x, y, ¢) blue). Hence colour TV picture is a three dimensional and three channel Bignal, which can be represented by the vector. Ite, ys) T(x, y0) 41.6) Itz, 8) Ian yt 1.4 CONTINUOUS-TIME VERSUS DISCRETE-TIME SIGN. (4) Signals can be further classified into different categories depehding on the charac- teristies of the time (independent) variables and the values they take. Continuous * Continuous-time signals or analog signals are defined for evegy value of time and they take on values in the continuous interval (a, b).Introduction 9 where acan be == bean be +e, ‘Mathematically, these signals can be deseribed by functions of a continuous variable. www wy —— eg. Speech signals x(t) = cos at a{t)eeltl, mactcn. Diserete « Discrete time signals are defined only at certain specific values of time. These time instant need not be equidistant, but generally they are taken at equally spaced inter- vals for convenience. eg. xt,= ell n=0,41,42... index ‘n’ of the discrete-time instants as the independent variables In applications, discrete-time signal may arise in two ways « In practical setting, such sequence (n) can often arise from periodic sampling of an analog signal. In this case, the numeric value of the n“* number in the sequence is equal to the value of analog signal x,(t) at time nT ie,, a(n) =anT). ‘The quantity’T is called sampling period and its reciprocal is the sampling frequency. « By accumulating a variable aver a period of time. For example, counting the number of cars ina given street every hour, or recording the value of gold every day, results in discrete-time signal (2)Continuous-valued and Diserete-valued Signals. A signal is said to be continu- gus valued signal if it takes on all possible values on a finite or infinite range. On the other hand, if the signal allowed to take on values from the given set, it is said to discrete-valued signal. Normally, those values are equidistance and hence can be expressed as an integer multiple of the distance between two successive values. If the signal to be processed is in analog form, it is converted to a digital signal by sampling the analog signal at diserete instants in time, obtaining a discrete-time signal, and then by quantizing its values to a set of diserete values Quantization. The process of converting a continuous-valued signal into a discrete- valued signal, called quantization. (8) Deterministic Versus Random Signals Depending upon the certainity by which the signal can be uniquely described as (@) Deterministic Signal, A signal that can be uniquely determined by a well-defined process such as a mathematical expression or rule, or table look-up is called a deter- ministie signal. (0) eg. (a) A sinusoidal signal can be represented as, v(t) = V,, sin at for t 20.(6) A square signal can be defined as at)=A for O 0) if and only if ain +N) =2(n) (1.9) The smallest value of N is called the fundamental period. Proof : For a sinusoid with frequency f, to be periodic, we should have, xin + N) = cos[2nf,{N +n) + 6) = 00s [2nfyn + 6) ‘The above relation is true if and only if, there exists an integer k such that, 2n f,N = 2nk. bey oo fy so 1.10) Therefore, the discrete-time sinusoids are periodic only if its pressed as rational number (ratio of two integers), To determine the fundamental period N of a periodic sinusoid, should be relatively prime. Then the fundamental period of sinusoid is ple, jueney fy can be ex- and N is eqn. 1.10 ial to N. For exam-introduction Ds] . 21 be, fh? 90 then fundamental period N, is 40 and if 20 521 f= 40°20 then fundamental period N, is 2. We observe that a small change in frequency may result in a large change in the period. (2) Discrete-time sinusoids whose frequencies are separated by an integer multiple of 2x are identical. Proof : Consider a sinusoid cos (@,n + 0). If the frequencies are separated by 2n, then, cos {lo + 2x)n +8] = cos [oy n+ 2nn +6] = cos [a n+ 0. Therefore, all the sinusoid signals, x,(n)= A. cos (@,n+0); k=0,1,2 whore, w, = w, + 2rk, are identical (distinguishable). Conclusion. The discrete-time sinusoids with frequencies | |< or |f|x or 4 /| >1/2,are identical to the sequence obtained from the sinusoid with frequency |@| 0,f ade = 1. ‘The integral of the impulse function is also a sin- gularity function and called the unit-step function and is represented as, 0, <0 wef #30 Fig. 1.10 (a). Continuous time The value att = 0 is taken ta be finite and in most. unit step signal. cases it is unspecified. The discrete-time unit-step sig- ule nal is defined as ‘ fO, a RSS. 1.7.3 Unit-Ramp Function Fig. 1.10 (2). Diserete time unit The unit-ramp function, rit) can be obtained by in- step signal. tograting the unit-impulse function twice or integrating 0 the unit-step function once, ie. n= ff smarde. =f worda. rit) = f. da : jo, t<0 Fig, 1.11 (a), Continuous time ‘Thatia, no={e £0" ee amp signal. A ramp signal starts at t = 0 and increases linearly with time ‘t. In discrete-time domain, the unit-ramp sig- nal is defined as, (0) ray= (0 n<0 2n>0" Fig. 1.11 (6). Discrete time 1.7.4 Unit-Pulse Function ramp signal. An unit-pulse function, n(¢), is obtained from unit-step signal as shawn below. w(t) = ult + 2) =u(t= 1/2)Ant) | =1 05 os 1 Fig. 1.12, Unit pulse signal. The signal w(¢ + 1/2) and u(t - 1/2) are the unit-step signals shifted by {1/2 units in the time axis towards the left and right respectively. Advantage. The advantage of the singularity function is that any arbitfary'signal that Properties of &(t) ay [6d = (2) [anode =x10) Proof for (2): f. x(t) lim 8, (8) dt + B= jim 3,0) = lim, 3 [sorrow = lim, + Prit) ie Him, Pro = im 3f. a(t)dt = x(0). | According to pulse funtion property, P,it) = 1 (3) J xD 8e-4) dt =x) & [2008 - a — a0) (5) Bat) = a 5) (6) x(t) Bt - t) = aty) (7) x(ty) &(t = t4) = ) () fi x(C)8"(t- to) dt = (Ya (ty). Proof for (8): & [x(t) Sle - tg)) = at) 5 (e-eq) + € (4) Blt — tq) = xtt)B(E ~ fg) + (tg) Blt ~ fy), ty < fy < ty |Introduction ce Integrating, we get sd ff on i 47 CO Ble to de = J" tate) 8c f9)] de + J Lita 5(¢ - to) de te [x0 ae 40] = i x(t) Bit ty) dt + ity) ity * LHS.=0. Therefore, ——_f* #954) HH) = 0 ie, i x(0) it = ty) dt =~ (ty). Similarly, f x(t) Bt - ty) dé = Ht) Hence, i x(t) 8°Ct —ty) dt = (— I" x"Uty) 1.8 ENERGY SIGNALS AND POWER SIGNALS (DISCRETE-TIME INSTANTS) ‘The energy of a signal x(n) is defined as, B= Di lxmp. ane ‘The energy ofa signal can be finite or infinite. If E is finite, thenx(n) is an energy signal. Many signal that posses infinite energy, have a finite power. The average power of a discrete- time signal x(n) is defined as, P= — P. jim yaa dy If we define the signal energy of x(n) over the finite interval-N sn: x Ey= lame. a=oN then we can express the signal energy Eas, E= lim Ey and the average power of the signal x(71) * Ps tm Nel Ne 7s Clearly, if Ey is finite, P= 0 on the other hand, if Ey is infinite, the pawer P may be either finite or infinite. If Pis finite (and non-zero), the signal is called a “power signal”. Problem 1, Determine the power and energy of the unit-step sequence.Sol. The average power of the unit-step signal is, N " 1 Pe fn aid, un) P= lim 1+N . YN+1 Pe New QN¢1 N= 24+0iN) Problem 2. Determine which of the following signals are periodic. (a) x(t) = sin 15 nt (b) a(t) = sin 20 at (c) x(t) = sin J2 nt (d) xf) = sin 5xt (ex, (= 240 +200 Pa) =x,(0 42/0. Sol. (a) x,(0= sin 15nt is periodic, ‘The fundamental period is, 2n_ an 1, = 28 = 22 0.1533 sec. ong IK see (b) xy (t) = sin 20 xt is periodic, ‘The fundamental period is, T, = 2% - 2% ~ 0.1 see Pee SoG 20n (©) x4 (é) = sin J2nt is period, an _ 2m The fundamental period is, Ty = = i (d)x,4@) = sin Snt is periodic The fundamental period is, T, 2 (e) x,(t) = x, (t) #x,{t). The fundamental period of x,(t) = Ty, = 0.133 sec, and the fundamental period of x,(£) = Ty, = 0.1 see. The ratio of fundamental frequencies, ‘Toy _ 01333 Toa Hence, x, (#) is not periodic. (Px, (=x, b+ x,t). ‘The fundamental period of x,(t) = Tyg = 0.1 sec and the fundamental =0.4 sec. ‘The ratio of fundamental frequencies, Te HT” 71 OP be empressod as ratio of integers. Heneo, x(t) is Problem 3. Sketch of the following signal : (a) x(t) = n(2t + 3) (b) x) = 2xlt= 144) (e) x(t) = cos (20 nt - 5x) and (a) x) = r(-0.5t + 2). , cannot be expressed as a ratio of intbgers. iod of, (1)=Tpy iodic.Sol. (a) x) = (2t + 3) = n(2 (t + 3/2)] Here, the signal is shifted to left, with centre at ~3/2. Sincea = 2i.e., |a| >1, the signal is compressed. The signal width becomes U2 with unity amplitude. gat (b) a(t) = Q(t - 1/4). Here the signal is shifted to the right, with centre at 1/4. Sincea = 1, the signal width is Land amplitude is 2. 4x0 \ * TE) t (© #lt) = c08 (20 nt — 5x) = cos [2 (t— V/4)} Here the signal x(¢) is shifted by quarter cycle to the right. (d) xt) = rl—0.5¢ + 2) =f os(+- ll =r[-050-4)] The given ramp signal is reflected through the origin and shifted to the right at t = 4,Sol. Representation through addition of two unit step functions, thd signal x(¢) can be obtained by adding both the pulses, i.e., x(é) = Qfu(t) — u(¢ — 2)] + (ult - 3) - u(t- 5)) Representation through multiplication of two unit step functions, x(t) = ful) ul—t + 2)) + [ult -3) ul # + 5)) = 2[ult) u(2—£)) + [ult ~ 3) u5 = 8) Problem 5. Plot the following signals for the given x(n) = (6 —n)} (ul4 (y(n) = 4-1) (i) y,fn) = (2n ~ 3). Sol. The given x(n) = (5 —n) (u(n) — u(n —5)] is plotted as shown belgw, = ufn - 5] wi} u(n-5) ore? 3 7 % ot 2s is Fig. (b)Introduction 23 u(a) -uin=5) x(n) 12 3 «4 8 6 9 wn * 2401 2 3 4 8 t Fig. (c) Fig. (d) Now Fig. (c) is multiplied by seale factor (5 —n) thus we get wave Fig. (d). (iy y(n) = x(4-n) where, x(n) = (5 ~ n) [u(n) - ulin -5)] 5— 44 n) [u(4 —n)—ul4 —n —5)] =(1+n) u(4—n)—u-1-n)) u (4 —n) means the sequence will exist between -« 0 Evaluate the real and imaginary components of x(t). xfn) = 20 co:Digitg!_ Signal Processing 7, Determine the power and rms value for each of the following signals : (1) 20 eas fr00e + | (2) 20 sin 5¢ cos 10¢ (8) 10 c08 5¢ cos 10¢ (4) ei cos my, 8. Figure below shows a signal x(t). For this signal sketch. () tt ~ 4), (2) x(e/10) (3) 3t -2) (A) x8 -D.Applications of Digital Signal Processing 2.1 INTRODUCTION Because of the availability of high resolution spectral analysis, DSP has various appli- cation areas, which requires high speed processors to implement the FFT algorithm. Itis also popular due to availability of custom made DSP chip which is highly reliable. Speech process- ing, Audio processing, Radar signal processing and Image processing would be discussed in this chapter. 2.2 APPLICATION TO SPEECH PROCESSING ‘The signals of speech are one dimensional. DSP is applied to a wide range of problem in speech such as channel vocoders, spectrum analysis etc, Problems in speech processing can generally be divided into three classes, first is the speech analysis. The speech analysis is performed to extract some desirable information of speceh. This system starts with analysis of speech waveform and the desired result is used for speech recognization and speaker indentification. Second type of problem is speech synthesis. Init, inputis in written text form and the output is a speech signal. For example, an automatic reading machine for which the input is written text and the output is speech, Finally the third type is speech compression which involves speech analysis followed by speech synthesis. If the speech is transmitted by simply sampling and digitizing, the data rate required is in the order of 90,000 bits per second of speech. Through the use of appropriate coding this can be reduced by factor of 50, depending on the type of system used. 2.2.1 Vocal Mechanism Production of speech. The two important part responsible for human speech are (a) vocal cord and (6) voeal tract. (a) Vocal cord. It has two bands of tough, elastic tissue, which is located at the opening of the larynx. It vibrates when the air from the lungs passes between them producing sound waves which are emitted from the lips and to some extent from the nose ; these are sound waves heard as speech. (b) Vocal tract. It includes larynx, the pharnx and the nasal cavity. 29Kinds of Sounds @ Voiced sound Gi) Unvoiced (fricative) sound. 1 Voiced sounds are produced by quasi-periodic pulses of air exciting the vocal tract. Unvoiced sounds are produced at some point along the vocal tract, usually towards the mouth. 2.2.2 Speech Technology (a) Speech coding. “Speech Coding” is the process of capturing the spepeh of a person and processing it to transmit over a communication channel. Tho application of “speech coding” is in the area of telephony, narrow-band cellular radio, military communication ete. (6) Speech enhancement. This is the proccss of minimizing the derogatory effects of noise on the performance of speech communication, source coding etc. ‘The application of ‘speech enhancement’ isin the areas where the perforsnance of equip- ment is improved in noisy atmosphere like factories ete. Synthesizing speech lies in creating speech like waveforms from textual] words or sym- bols, using a model for speech production and time-varying parameters, The application of this are in voice alarms, reading machines for the flumb or blind, data-base enquiry services etc. (d) Speech recognition. The process of deriving the meaning from f speech input whereby a request can be made for information or service from a machinery by conversing with it, Application of “speech recognition” could be Banking from distant locatipn, information retrieval systems ete. (e) Speaker recognition. It means to recognize a particular person's identity with the sample specch dipping. 2.2.3 Parameters of Speech (i) Pitch ; Corresponds to frequency of sound (in Hz). (ié) Loudness : This relates to intensity of sound (in dB). (éii) Quality : This relates te harmonic constant of sound (in timbre). ‘Phonemes’ are the smallest unit of sound that are recognized by contrast| with their environment, these are forming the basic units of speech. ‘Dipones’ are sountls that stretch from the middle of one phoneme ta the centre of the next, there by spanning the transition region. 2.2.4 Speech Analysis The most common methods of speech analysis are as follows : (a) Short-time fourier analysis (6) Linear prediction. (c) Homonorphic filtering. Let us discuss about these three methods of speech analysiApplications of Digital Signal Processing Br] (a) Short-time fourier analysis. The short-time Fourier transform of a sampled speech signal represented by the sequence x(n) is given by a(n) =D xh) h(n = Wet (21) Kora Fig. 2.1 shows the short-time fourier analysis. in). Fig. 2.1. Short-time fourier analysis. There are two methods of obtaining the short-time fourier analysis. @ Through analog implementation of a filter bank. Gi) By digitally computing short-time fourier transform either by using a filter bank or by using the FFT algorithm. (6) Linear prediction. This method is based on “Auto regressive moving average” or pole-zoro model. If H@) is an all-pole transfer function given by Hw = — 4 — ABQ) 1-¥ ae mH We have the time-series discrete time signal as P adn) = ¥, a4 20-4) + ABR) 2.3) a where A is the gain factor, P for n>0, xin) = DY ay xin - 2) wf) mnthis approximate value, P F(x) = Fay aln—b) 2 > | v2.5) im \ ‘The corresponding error is given by | e(n) = x(n) = X(n) | : » | =atn)- Yay x(n-h)n > 0 2.6) a Mean squared error is given by SS Na Pp 2 Eys =, C= SY fxn) — PF. ay xn -2) (2.7) ” = mt The parameters {a,] that minimize E,,, are determined by partial dérivating E,,, with respect to each co-efficient a,, k Pp and equating to zero, i 9Eus that gives -(2.8) with m= hn-Daln-h), 12.9) oO (c) Homonorphic filtering (Cepstral Analysis). Since the excitation function and vocal tract impulse response are convolved to produce speech, this problem js thought of as a separation or deconvolution of speech into these two components. ‘The deconvolution speech is carried out by the non-linear filtering que, deseribed here as “Homonorphic filtering”. The convolution operation is converted intb addition which gives the output called the “complex cepstrum”. Please note that the “Cepstrum” of a signal represents the fourier transform of its power spectrum. Input { { , Output st Lp=s > xin) im ee ¥iny System Unear Inverse ‘System ‘System Sa ie 82) } Transtarmation * Output signal Fig. 2.4. Transform coding. Sub-band coding. The specch signal is appliod to an analysis filter bandpass filters. This di, i Fig. 2.5. Sub-band coding.Applications of Digital Signal Processing [es] additive recombination of the set of sub-band signals, the original speech signal ean be generated. Each band is separately quantized and coded using pulse code modulation and transmitted. ‘The schematic is shown in Fig. 2.5. 2.3 APPLICATION TO IMAGE PROCESSING Any function which bears two-dimensional information is called an image. Image can be represented by an array of real or complex (real and imaginary) numbers with finite number of bits with respect to speech signal (which are one-dimensional signals), image signals are two dimensional. Image can be divided into picture elements or pixels (smallest element of image). Manipulation af two-dimensional signal with the help of digital computer is called “{mage Processing”. Its purpose is to imprave the visual appearance of Image. A Digital Image is digitalization of picture. Normally two-dimensional image nas reso- lution 128 x 128, 256 x 256,512 x 512. Saimage can be processed using two-dimensional signal processing. The image processing including the following steps : (a) Image Formation and Recording. (6) Image Sampling and Quantization. (c) Image Compression, (d) Image Restoration. (e) Image Enhancement. All the operations are possible on advanced software artificial intelligence and high- tech digital computers. Let us discuss all the operations one by one. 2.3.1 Image Formation and Recording ‘The two-dimensional signal of image can be expressed by image function as atay= [0 [0 Ae - ay - wien de dy A245) Eqn. (2.15) governs a 2D linear time invariant system. Here system impulse response function h(x — x, 9) is commonly referred as point-spread function which is usually associ- ated with optical image. The function fix, y) is the accumulation of energy from the objects radiant energy distribution. ‘Two major technologies are used for image sensing and recording, which are photo- chemical recording and photo-electronic recording. Both of the technologies are exemplified by readily available products which are photo-graphic films and television respectively. (Here “television” is used in generic sense not commercial broad casting television). 2.3.2 Image Sampling and Quantization After formation and recording of an image, it is sampled and quantized for the suitabil- ity of digital processing. Ina system project, a spot of light with intensity I, incident on a film and intensity 1, reflected from the film and collected by photo-multiplier. The transmittance is defined by Tn Ter = (2.16)weve [ff Ale. y-y del aydr, dy, 217) Here h, is the intensity profile of the spot of light projected an film. § is the image on film and finally g, is actual sampled image. The sample matrix g,(k Ax, 1 Ay)}is the sampled or digital image. 2.3.3 Image Compression In a digital image 10° to 108 data are there. The processing of these higher value of image data is a very stupendous task. But a digital image has large numbfr of redundancy which can be reduced by image compression, So we can say that image comprdssion is a science of efficiently coding a digital image to reduce the number of bits, which required to represent it. Uncompressed image consumes memory space in a large amount so ft increases com- plexity in computational and need a very large transmission bandwidth. A cpmpressed image reduces the redundancy in image. There are mainly three types of redunflaney which are discussed one by one. The first type redundancy is Spatial Redundancy which arises due tp correlation be- tween neighbouring pixels. Second type is Spectral Redundancy which is correlation between various colour plans. And finally is Temporal Redundancy is the correlation Hetween different frames in an image sequence, In an Image Compression System, the original continuous time image signal is fed to /D (Analog to Digital) converter, which converts it into digital signal. Now a serial to parallel (S/P) converter decomposed signal into parallel channels which fed to a quantizpr. The S/P con- vertor is linear transformer or filter banks are used. This quantized output isteaded by the use of lossless coding device, whose output is compressed digital image signal] A simple block diagram of image compression system is shown in Fig. 2.6. Digital Decomposed ‘Quantized signal signal into output parallel channels Input (original continuous time, image signal) Tassie AD sip Quantzer ‘coc Converter Converter techniq iT Fig. 2.6. Image compression system. Applications of digital signal processing image compression system ‘There are mainly three types of compression technique based on the thethod of redun- dancy detection :Applications of Digital Signal Processing [7] (a) Direct data compression method (6) Transformation method (c) Parametric extraction method. 2.3.4 Image Restoration The process of image restoration is used for correcting imaging effect to recover an original signal. This type of effect (imaging effect) is due to variety of intermixing factors, which are defocusing imaging camera, relative motion between object and camera, noise in sensors cte., All types of imaging effects deteriorate image quality. ‘The process of image restoration is to attempt a image which should be sharp, clean and free from the degradation. The restoration process is also called Image Deblurring. The proc- ess of image formation and recording can be modelled as atx, y) =n{f[lae—any-vo rend 4s] +n, 9) (2.18) Hereg(a, y) is the actual image, R is the response characteristic of the recording process and n(x, y) is additive noise source. In the restoration of digital image following equation can be expressed in diecrete form : Reinet ap.d= > > fi Dh pig i (2.19) fm A largo set of simultaneous linear equations can be solved by DSP techniques such as linear filters and FFT algorithms which are computationally efficient tools for solving these. 2.3.5 Image Enhancement ‘This technique improves the appearance of image for human perception by choosing some image features like edges or contrast etc. Its main application is in biomedical engineer- ing field for computer aided mammographics studies, In image enhancement spatial filtering is mainly used whose operation is done on im- age to reduce noise contamination of the image signal. Image enhancement is composed of a variety of methods whose suitability depends upon the goals at hand when enhancement is originally applied. REVIEW QUESTIONS 1. Give the areas in which signal processing find 2, Explain the various stages in voice processing. 3. How is a speech signal generated ? 4. Give the model of speech production system ? 5. What is the need for short time spectral analysis ? 6. What is a vocoder ? Explain with a block diagram ? 7. Describe how targets can be detected using radar. 8, Give an expression for the following parameters related to radar (a) beam width, and (6) maximum unambiguous range. ‘application.Digital 9. Explain with the block diagram the modern radar system. 10, Give the various image processing applications. 11. Give the various coding techniques for images. 12. What is the need for image compression ? 13. Give the block diagram of basic restoration process. 14, What is sub-band coding ? 15. Explain the process of digital FM stereo signal generation. 16. Explain how privacy can be acheieved in telephone communications.Chapters : 3. Discrete Time Systems 4. Frequency Domain Characterization of Discrete-time Systems DIGITAL SIGNAL PROCESSINGDiscrete Time Systems 3.1 DISCRETE-TIME SIGNALS AND SYSTEMS: 3.1.1 Definition 1. A discrete-time signal is a sequence, that is a funetion defined on the positive and negative integers. 2. A discrete-time system is a mapping from the sct of acceptable discrete-time signals called the input set, to a set of discrete-time signals called output set. 3. A discrete-time signal whose values are from a finite set is called a digital signal. 4. A digital system is a mapping which assigns a digital output signal to every accept- able digital input signal. 3.1.2 Representations 1, Graphical. In digital signal processing, signals are represented as sequence of num- bers called samples. A sampled value of typical discrete-time signal or sequence is denoted by x(n) which is a function of independent variable that is an integer. It is graphically repre- sented in Fig. 3.1. x(6) Fig. 3.1, Graphical representation. atJignal Processing It is important to note that x(n) is defined only for integer values of fand undefined for non-integer values of n. for every integer ly denoted byx(k), In the signal we have assumed that a discrete-time sequence is definal value ofn for ~~ x(t) ‘This system is basically an accumulator that computes the running sym of all the past input values upto present time. The response of the system to the given inp¥t is, yl) = fn O, 8, 5, 6, 6, 79, 12, 12, ond 3.3 SAMPLING In some applications a discrete-time sequence x(t) is generated by periodically sam- pling a continuous time signal x,(¢) at uniform time intervals. x(n) 22,0) |, 47 =, (2T) jn = 2,-1,0,1,2 ‘The spacing T between two consecutive samples in eqn. (3.10) interval or sampling period. ‘The reciprocal of the sampling interval T, denoted as Fis called the sanspling frequency. wf3.20) call¢d the sampling 1 Fy= ay (cyclelsec or Hertz), 3.4 REAL AND COMPLEX SEQUENCE Real sequence. If X(n) is real for all values of n, then (x(n)) is a real sbquence, Complex sequence. If the n'® sample value is complex for one or mpre values of 7», then it is complex sequence. It can be defined as,Discrete Time Systems 7] (xin)) = (x, (79) + fle, ()- 3.1) where, s,,(n) and x,,(n) are the real part and the imaginary part of x(n). Complex conjugate sequence (eM(n)} = bx, (a) = flr, (n)). (8.12) 3.5 FINITE AND INFINITE LENGTH SEQUENCE ‘The discrete-time signal may be a finite-length or an infinite-length sequence. A finite length sequence is defined only for a finite time interval. NysnsN, (3.13) where, -« Ny bey Fig. 3.4, Left-sided sequence.afn) = 0 forn > N, where, Ny is a finite integer which can be positive or negative. In general ‘Two sided-sequence is defined for all values of n in the range =~ operation. If N < 0, it is an advancing operation. it is a delayingDiscrete Time Systems [ss] The device implementing the delay operation gn) ; wale) by one sample is called a “Unit delay” and its | #° |} schematic representation is shown in Fig. 3.8 i ‘The schematic representation of the unit advance *{0) win) operation is shown in Fig. 3.9 | — Fig. 3.8. Unit delay. w,{n) = x[n wgin] = xin + 1) Fig. 8.9. Advance operation. (0) Time-reversed or Folding. The time reversal operation, also called the folding operation, is another useful scheme to develop a new sequence. w,(n) = x(=n) (3.20) which is the time-reversed version of the sequence x(n). (vi) Pick-off node. It is used to provide multiple copies of a xn) x10) sequence. Problem 3. Consider the following two sequence of length 5 defined for O 1, the process is called “interpolation” and results in a sequence with a higher sampling rate. © On the other hand, if R < 1, the sampling rate is decreased “decimation”. ‘The basic operations employed in the sampling rate alteration profess are called up- sampling and down-sampling. These operations play important roles in fnultirate discrete time systems. Up-sampling. In up-sampling by an integer factor L> 1, L—1 equidistant zero-valued samples are inserted by the up-sampler between each two consequtive samples of the input sequence x(n] to develop an output sequence y(n) according to the relation, z,Inl= { a process called xin], w=0,tL,+2L...... oO otherwise samples, generating an output sequence y(n) according to the relation, y(n) = 2laM} The result in a sequence 4) whose sampling rate is (I/m)" that of x(h). 3.9 CLASSIFICATION BASED ON SYMMETRY PROBLEM Problem 6. Consider the finite length sequence of length 7 defined fo de(r)) = 10, 1 + j4,-2 + j3, 4-j2, - 5-56, -j2, 3) T 3ens3: To determine its g,, and £,_- Sol. (1) To determine conjugate symmetric part g,, (1) [g*(n)] = (0, 1-j4, - 2-3, 4 +72, - 5 +76 ,j2, 3) tDiscrete Time Systems ea ‘Whose time-reversed version is given by, (* (—n)) = (3,32, -5 +j6, 4 +72, -2-J3, 1-J4, 0) T 1 Bey (Rn) = 5 bein + 2—n)). (o,, (nD) = (1.5, 0.5 43, - 3.5 + j4.5, 4,-3.5 ~74.5, 0.5-J3, 1.5] T 2. To determine conjugate anti-symmetrie part g,, (n) qn) = fem -x*(—n)]. Wg (m)] = {- 1.5, 0.5 +j1, 1.5 1.5, ~j2,~ 1.5 -J1.5, tT 5 — jl, 1.5) It can be easily verified that, Bain) =B.,8(-n) and Beg (R) =~," (- 0). 3.9.1 Periodic Conjugate-Symmetric Part and Periodic Conjugate Anti-symmetric Part Periodic conjugate-symmetric part defined by, Spe) 3 [ein) +x*(—n)yJ]. ,OSnZN-1 (9.24) Periodic conjugate anti-symmetric part is defined by, L Xpeelt) = y Ieln) 24 [(-n)gl], OS SN-1 43.25) So that xin}=x,,(n) +x, (0), OSn)). We observe that, u"((-0),) =u* (0) = 1 = ja. ut D,) =u" (3) =~5 +6. uf ((~2),] =u*(2)= 44 j2 u*((—3),) =u) =- 2-3.{u*((—n),)) = {1 —j4, - 6 + j6, 4 + j2, -2-j3} 1 Myo) = slut) + wm) u,,, (nd = (1,-3.5 +) 4.54 ,- 3.54.5) tt, ., (1) = 4, LB =j1.5, =j2, = 1.5 ~j1.5) It can be easily verified that, Mpc 3.10 SAMPLING PROCESS ‘The discrete-time sequence is developed by uniformly sampling a continudus ti &,(t) as illustrated in Fig. (3.11). Fig. 3.11, Continuous tim: ‘The relation between the two signals is given by eqn. xin =2(0|, (nT) n= 0,~2,-1,0,1, -(3.27) discrete time signal only at diserete-time instants ¢, given by, t,=nT +328) with, ye 3 denoting the sampling frequency and 0, = 2#F, denoting the sampling angular frequency For example, if continuous ~ time signal is, x ,At) = A cos (2n fot + ¢) ,(t) = A.cos (Se + 9) the corresponding discrete-time signal is gives. by, afn] =x, [nT] = Acos [0, nT +91 = Asos[2afot non sg (3.29) x(n) =A cos [a,, +61Discrete Time Systems 33 ] Bry where, =O (3.30) It is the normalised angular frequency of the discrete-time signal x(n). Units ‘The unit of the normalised digital angular frequency a, is radians per sample. While, the unit of the normalised analog angular frequency @, is radians per sample and the unit analog frequency f, is hertz is the unit of the sampling period T is in seconds. Problem 8. Consider the three sequence generated by uniformly sampling the three co- sine functions of frequencies 3Hz, 7Hz and 13Hz respectively : g (t) = cost 6rt), g,(t) = cos (1431), and g,(t) = cos (26nt) with sampling rate of 10 Hz. ie,, with T = 0.1 sec, Find the derived sequence or discrete sequence, Sol. 08 (pt); 0 (5.27) (1) = cos (6 xn xT) 8(n)= c08 (0.6 21) Similarly, Bz (n) = €08 (1.480) and Bs(n) = c08 (2.6 nn). Problem 9. Determine the diserete time signal v(n) obtained by uniformly sampling at sampling rate of 200 Hz., a continuous time signal v,(t) composed of a weighted sum of five sinusoidal of frequencies 30 Hz, 150 Hz, 170 Hz, 250 Hz and 330 Hz as given below : v4 (t) = 6 cas (60nt) + 3 sinfZ00nt) + 2 cos (340nt) + 4 cos (500xt) + 10 sin (660nt). Sol. To find the sampling period (T): 1 1 T=—=—— =0.005 sec. F200 ace ‘The generated discrete-time signal o(n) is given by, vin) =6 cos (0,3nr) +3 sin (1.5nn) + 2 cos (1.7xn) + 400s (2.5) + 10 sin (3.352), = 6 cos (0.3nn) +3 sin (2x —0.5n)n] + 2 cos [(2x - 0.3) n] +4 cos [(2n + 0.5x)n] + 10 sin [(4n-~ 0.7n)n] = 6 cos [0.3nn} ~ 3 sin [0.5 an} + 2 cos [0.3An] + 4 cos {0.5xn} — 10 sin[0.72n} b(n) = [8 cos (0.3nn) + 5 eos (0.520 + 0.6435) - 10 sin (0.7A0)] The discrete-time signal v(m) is composed of a weighted sum of three-diserete-time sinusoidal signals of normalised angular frequencies : 0.3m, 0.5x and 0.71]. 3.11 CLASSIFICATION OF DISCRETE-TIME SYSTEMS Diserete-time aystems are classified according to their general properties and charac teristics. They are (1) Static and Dynamic systems. (2) Time-variant and time-invariant systems,(3) Causal and non-causal systems. (4) Stable and unstable systems. (5) Linear and non-linear systems. (6) FIR and IIR systems. 1. Static and Dynamic Systems A discrete-time system is called static or “memory les: In any other case, the system is said to be dynamic or to have memo ‘The systems described by the following equations, vn) = ax(n) yin) = ax%(n). are both static as they won’t require memory. On the other hand, the syster following equations pln) = x(n — 1) + x(n — 2) yn) = a(n) + x(n - 1) are dynamic systems as they require finite memory. 2. Time-variant and Time-invariant Systems A system is called time-invariant if its input-output characteristics flo not change with time, A “linear time-invariant” (LTT) discrete-time system satisfies both thie linearity and the time-invariance properties. To test if any given system is time-invariant, first apply an arbitra find yin). sequence x(n) and yn) = Then). Now delay the input sequence by k samples and find output sequenge, denote it as, y(n, k)=T be (n- kd) 8.8) Delay the output sequence by & samples, denote it as y(n ~ &). If yin, k) = y(n-= k) 3.32) For all possible values of &, the system is time-invariant on the othr hand the output, Hn, kya y(n —) (8.83) Even for one value of &, the system is time-variant xin) x(n =k) L$» yin. &) »{ System yea [Delay OES inky Fig. 3.12. Time invariant and time variant system. Problem 10. Determine if the following systems are time-invariantlor time-variant. @) y(n} = xfn) + x(n - I) (ii) y(n) = xf— n) (ii) y(n) = xf2n) (iv) yin) = xfn) sin wn.Discrete Time Systems [ss (n) + x(n - 1). We know that, Ixtn)]. = x(n) +x(n- 1). If the input is delayed by & units in time, we have, yn, k) = Then - 8) y(n, k) = x(n —h) + x(n —R- 1) a) If we delay the output by A units in time then, yln—k) = x(n —k) +x(n -k-1) 2) ()=@) Here, y(n, k) = y(n -k) So, the system is time-invariant. Gi) ym) = x(n) If the input is delayed by & units in time and applied to the system, we have, y(n, k) = Thea —&)] =21- a —4] 8) If the output is delayed by # samples, yn-k)=xl-(n-k)) =2[-n +k) ond) (3)# (4) Here, vn, k)# yn) So, the system is time-variant. Gid) y(n) = x(n) The system is described by input output equation, yn) = Then) yn) = x(2n) If the input is delayed by unit in time and applied to the system, y(n, k) = Then — b)] y(n, k) =x(2n - ke) ) Now, if we delay the output y(n) by # unit in time, the result will be ¥n—k) = x[2(n —k)) = x[2n - 2k] 2) Since y(n, )# y(n —2), the system is time variant. = fa) = x(2n) x(0) yo, ) = x(2n—h) — System) ‘ylo.= K) « x(@n = 2k) (iv) y(n) = x(n) sin wn If the dp is delayed by & unit in time and applied to the system, Wn, &) =2(n =k) sin @, 2. If we delay the output by & unit in time, then y(n — ) = x(n —&) sin w, (n — kd Since y(n — 2) #3(n, b). So the system is time variant.8, Causal and Non-causal System xin] for n Sng and does not depend on input samples for n > ny. This means thht, a system is said to be causal if the output of the system at any time n depends only on présent and past inputs, but does not depend on future inputs. This can be expressed mathemat{cally as, y(n) = Fia(n), x(n - Dy x r= 2) 0). Ifa system depends not only on present and past inputs but also on futufe inputs, then it is said to be a non-causal system. 4, Stable System ‘There are various definitions of stability we define a discrete-time system tobe stable if and only if, for every bounded input, the output is also bounded. This impljes that, if the response to.x(rt) is the sequence y(n) and if, len) xu 8a - —_ > eW Tw - wn we DY He) bln, b) woe B.41) ro in) ‘Thus according to eqn. (3.41) if we know the response of a linear system to the set of shifted unit impulse, we can construct the response to an arbitrary input. If fhe response of the ‘LTI system to the unit impulse &n) is denoted by h(n), that is, An) = T[5(n)] ‘Then by the time invariance property of the response of the system fo the delayed unit impulse 5(n - k) ; is, h (nh) = TIB(n —#)] Consequently eqn. (3.41) reduces to, yn)= zie ay (8.42) The eqn, (3.42) referred as the convolution sum or superposition sunt and the operation on R.HLS. is known as the convolution of the sequence x(n) and A(n), We Wwill represent the operation of convolution symbolically as,Discrete Time Systems [er] yin) = x(n) * h(n) --AB.43) ‘The operation of discrete-time convolution takes two sequences x(n) and h(n) and pro- duces a third sequence yin). 3.13 THE CONVOLUTION PROCESS CAN BE SUMMARISED INTO THE FOLLOWING STEPS Step 1. Choose an initial value of ‘n’, the starting time for evaluating the output se- quence y(n), Ifx{n) starts atn =n, and A(n) starts atn =n,, then n =n, +N, is a good choice. ‘Then express both sequence in terms of the index k. Step 2. Folding. Fold the A(e) about the origin and obtain h(- k). Step 3. Time shifting. Shift the h(— k) by n unit to right if n is positive and left if n is negative to obtain A[-(k - n)] = h(n - k). Step 4. Multiplication. Multiply x() by h(n — k) to obtain w,(k) = x(k) h(n - h). Step 5. Summation, Sum all the values of the product (2) to obtain the value of output y(n). Step 6. Increment the indexn, shift the sequence h(n — k) to right by one sample and do step 4. Step 7. Repeat step 6 until the sum of product is zero for all remaining values of n. Problem 15. Determine the output y(n) of a linear time invariant system with impulse response n(n) = 16, 8, 4, 3, 2, 1) t when the input is x(n) = (1, 1,1, 1 1 Sol. x(n) starts at n, = 0 and A(n) starts at n, =0. ‘Therefore, the starting value of n =n +"%y Step 1. Folding. The folding sequence hh) is illustrated in Fig. 3.14, XO) 2 ) e nib) n(-k) awusae o1234 5 k ke BAA B2-10 Fig. 2.14. Folding sequence.(0) = amie ) = 410) hl0) = (1x6) =6 Bao tesas x Forn=1, {b= 5, (HAG-&) toe ht —) 4-3-2101 i (1) = (0) ACD) + 21) ACO) = (1x 6)+(1x5)= 11. Forn=2, y= ¥ 0) aa-2). ‘ na- Ky yt2) s s 5 3 q 4 3 2 tl * sa O12 x Bore 8 ‘ (2) = x0) fe (2) + x(1) ACL) + x(2) ACO). 1 4) + (1 x 5) + (1 x 6) =4454+6= 15.Discrete Time Systems [ss] Forn=3, (3) z x(k) (3 — k) eee hiS—k) 6 y(3) 6 5 5 4 ‘ Fy a 2 11 — . “120123 k 20123 * 3) = xf0) A(3) + x(1) A(2) + x(2) ACL) + x(3) ACO) = (1x 3) + (1x 4) + (1% 5) + (1x 6) =344+54+6=18 Forn=4, wae Yh -h) (4 —k) nm uy 6 4) 5 5 4 4 3 3 2 2] i ores 4 t Sore 3 x yd) = 20) ACA) + x(1) ALB) + x(2) (2) + 23) ACL) + (4) AO) (1% 2) + (1x 8) + (1 x 4) 4 (1x 5) + (1x6) m2434445= 24 Forn=5, y= YD x(k] ~b) (5) shusue »s}-——* w}L—_— ~-—__ a}, wt zroanoa }—sr. ot y965) = <(0) ACS) + x1) ACA) + x(2} A(B) + 2(3) h(2) + xf p AD + x66) n(O) = (1x 14 (1x 2)4 (1x3) 4 (Lx 4) 4040 21424344210 Forn =6, wO= >) x AG-% ie n(6 - k) 8 5 4 3 2 ' o1es 456 orests (6) = x(0) (6) + (1) (5) + x(2) ACA) + x(3) ACB) + x(4) A(2) + (5) A(1) + x(62A(0) = O+(L= 1)+ (1x 2)+(1 «32404040 =14+2+3=6, Forn=7, w= Y, xk) ht7-2) ote nik) 6 5 4 a 2 1 1 o,23465 67 k ot2aa § WT) = (0) A(T) + (1) (6) + x(2) A(5) + 2(3) h(4) =0404+(1 «14 (1x 2)4+0404040 al+2=3. For n=8, y8)= DS) x(k) 8-4) y(B) = x(0) (8) + x(1) A(T) + x€2) h(6) + x(3) A(S) + x(4) A(4) + x(5) h(8) + 216) (2) + x(7)A(1) + x{8) h(0), 23465678 °~-& or?2a456 =0+04+04(1x 1)4+040+0+040 a1 Forn=9, wo= SH AO-%) yok) yea) Je Lk. 2s 4 788 o1v23s46 epeeue ¥(9) = x(0) (9) + x(1) A(8) + x(2) ACT) + x(3) ALE) + x(4) A(B) + x(5) (4) + x(6) (3) + 2x(7) h(2) + x(8) h(2) + x(9) h(0). =0+04040+04040+040. Similarly, (1) = 0. Now we summarize the entire response for — « i1 -4éns-8 no) = {5 elsewhere Ans. y(n) 4 (1,2, 2, 2, 1} 2. Find the convolution of the signals, i x(n)= 1 2,0,1 | =2 n=-1 | =0 elsewhere. | nin) = &(n) — Sin — 1) + Bn - 2)- hn—-3) ie, Alm) = {1-1 1, Ans. y(n) = (1, 1, 0,1, -2, 0,- 1). t 3.14 PROPERTIES OF LINEAR TIME-INVARIANT SYSTEM Since all linear-invariant systems are described by the convolution sum, of this class of system are defined by the properties of discrete-time eonvolutia (1) Commutative. The convolution operation is commutative an) = x(n) * fn) = Aln) + tn) he properties where, x(n) + hin) = SY) x(k) h(n) (8.44) foam and Hin) + x(n) =" hem) x(n —m) (8.45) ate In eqn. (3.44), the impulse response is folded on the other hand the input hignal is folded in egn. (3.46) Commutative operation can be shown by applying a substitution of Variables, with m= n-k. ya)= Fi hin- b= >} Aim) xn -m) 18.46) soe ieee (2) Distributive (Parallel connection) The convolution operation is also distributive over addition ie., x(n) * Uiy(n) + Ag(n)] = x(n) * h(n) + x(n) * hind (3.47) Distributive law implies that ifwe have two linear time invariant systems with impulse response A,(n) and h,(n) excited by the same input x(n) then the sum of thp two response identical to the response of an overall system with impulse response Aln) = hn) + h,(n) 3.48) Thus the overall system is viewed as parallel combination of two linearjtime-invariant system. The parallel operation is shown in Fig. 3.15Discrete Time Systems (o) (ke 2 a Tana xin) hm Fig. 3.15. Parallel operation. (3) Associative [Cascade connection] The associative property states that, when we are convolving three signals, we can convolve two of them and then convolve that result with the third signal. Let us assume that, x(n), h,(n) and h(n) are three arbitrary signals to be convolved, i.e., Lal) + hy(n)] + hglre) = ala) # tryin) + han (3.49) An interpretation of the associative property is illustrated in Fig. 3.16 yin) xtn) n) oho ESR} Ss <> Somme Fig. 3.16. Associative property. yin) = win) * hyn) y(n) = x(n) [h(n) #4,{0)] = le(n) * Ay(n)] * Afr). Problem 16. Prove that x(n) + {hy(a) + hyfn)} = (x(n) +h, (n)] + hegfn). Sol. x(n) * (y(n) * h(n) =n) * y hain} ate = yo] > bahte--t] vote [ante Let, k-l=m, - $20] x [ ¥ so him-12] itn Y faim - 1) dain -m] DY, [xen « Gm] hy (nm) = ela) * h(n) © yin). Problem 17. Find the convolution of two finite duration sequences, hn) = al u(nd for all x(n) =F u(r) for alt n (when a #6 (ii) when a =b. Sol. The impulse response h(n) = 0 for n <0, So the system is causal and x(n) = Oforn <0, hence the sequence is a causal sequence.About the Book This book deals with the analysis of Digital Signal Processing in a lucid and precise style. There are about 200 solved problems apart from exercises. This book covers the latest syllabus prescribed by the fe University for Electrical and Electronics engineering students. Exercise problems and review questions are included at the end of each chapter. All the above aspects should make this book extremely valuable for engineering students preparing for Anna University examinations as well as for practicing engineers. About the Author C, Ramesh Babu Durai graduated from Arulmigu Kalasalingam College of Engineering, Srivilliputhur and did his post graduate studies at Hindustan College of Engineering, Chennai. He is a faculty member of the department of Electrical and Electronics Engineering, Hindustan College of Engineering, Chennai. He has more than six years of teaching experience and the college has honoured him by conferring on him “The Best Teacher Award”. His field of interest includes Control Systems, Advanced Digital Signal Processing and Electromagnetic theory. 18-736-8 81-7003 LAXMI PUBLICATIONS (P) LTD |: || | reo!