EE 422G Notes: Chapter 8 Instructor: Zhang Chapter 8 Discrete-Time Signals and Systems 8-1 Introduction Most “real” signals and natural (physical) processes: continuous – time A : System Design Problem How the computer sees “ the rest”? an equivalent (Physical Process + Sensor +A/D + D/A )=> discrete-time system The Equivalent discrete-time system Modeled by a discrete-time model System Design (Design of the computer control Algorithm): Based on discrete-time model description. Needs for discrete-time system analysis and design tool: Z-Transform (Similar position as Laplace Transform for continuoustime system.) Page 8-1 EE 422G Notes: Chapter 8 Instructor: Zhang B. How does the computer understand the progress and behaviors of the process being monitored and controlled? By sampling the output of the continuous-time system! => How can we ensure that the sampled signal is a sufficient representation of its continuous-time origin. i.e., how fast we have to sample? A question we must answer before z-transform based analysis! C. Two basic parts of the chapter Part one : Theoretical frame work for determining how fast we have to sample. Part two : z-transform Part one: How fast 8.2A Analog-to-Digital Conversion 1. Sample Operation Needs to Know: (1) Sampling period: T (2) x(t) is sampled at t=nT (3) What do we mean by x(n) (4) Sampling function: p(t) (5) Sampled signal xs = x(t)p(t) Page 8-2 EE 422G Notes: Chapter 8 Instructor: Zhang 2. Mathematical Description of Sampling Process Sampled signal : xs(t) = x(t)p(t) Objective: Derivation of xs(t)’s Fourier Series Expression (Time Domain) Derivation : p (t ) C e n General Equation for any periodical signal Fourier Series Description of Sampling Function jn 2f st n Sampling function: A Periodical function, (thus can be expressed using Fourier series), with period T on fundamental frequency f s 1 T With Fourier series coefficients: 1 T /2 Cn p (t )e jn 2f s t dt T T / 2 general equation for Fourier coefficient of any periodical signal xs (t ) p(t ) x(t ) ( Cn e jn 2f st ) x(t ) n C n n general equation: always true for any r (width of the sampling pulse). x(t )e jn 2f st 3. Spectrum of sampled signal Objective: Find the spectrum of the sampled signal xs(t). Derivation : Take Fourier Transform for xs (t ) C x(t )e n jn 2f st n Page 8-3 EE 422G Notes: Chapter 8 Instructor: Zhang j 2ft x ( t ) e dt s xs ( f ) C x(t )e n C x(t )e n For n n e j 2ft dt j 2 ( f nf s ) t n dt C jn 2f s t n n X ( f nf s ) X ( f nf s ) 4. Spectral Characteristic of ‘Real Signal’ Most ‘real’ signals: continuous with time Highest frequency fh can be found X(f) = 0 if | f | f h 5. How ‘sampling process’ modifies the spectrum Xs( f ) Consider a f 0 f h Cn X ( f n X s ( f0 ) C n n nf s ) X ( f 0 nf s ) C 0 X ( f 0 ) [C1 X ( f 0 f s ) C 1 X ( f 0 f s )] If f0 fs fh X ( f0 f s ) 0 or or f0 fs fh X ( f0 f s ) 0 X s ( f 0 ) C0 X ( f 0 ) Modifier Page 8-4 EE 422G Notes: Chapter 8 If Instructor: Zhang f0 fs fh f0 fs fh or X ( f0 f s ) 0 X ( f0 f s ) 0 or X s ( f 0 ) C0 X ( f 0 ) No spectrum modification 6. How fast we have to sample in order to keep the spectrum: Xs( f ) X( f ) (| f | f h ) | f f s | f h | f f s | f h Condition Should we consider | f | f s ? Of course not! ( | f | f s implies that the real process changes faster than the sampling rate.) Consider | f | f s only => fs f fh fs 2 fh fs f fh => f f f f 2 f fs f fh h s h s Answer : f s 2 f h Sampling rate: at least twice as the highest frequency of the “original process” Sampling Theorem: …………. Constant with n 7. What about if r 0 ? (show Figure 8-4) 1 T /2 1 jn 2f s t p (t ) (t nT ) Cn ( t ) e dt fs T T/ 2 T n X s ( f ) fs X ( f nf s ) n 8. Practical sampling rate: f s 10 f h 1 X ( f nf s ) T n Similar as r 0 Page 8-5 EE 422G Notes: Chapter 8 Instructor: Zhang 8-2B Data Reconstruction 1. What’s Data Reconstruction? Original x(t) t 0 (anytime) Its samples xs(t) t = 0, T, 2T, … (Discrete time) Can we tell x(t) between sampled points ( nT < t < (n+1)T ) based on xs(t)? Data Reconstruction problem! 2. Data Reconstruction Method What’s a filter? A system which processes the input to generate an output. It could be an algorithm (mathematical equation/operation set) or circuit/analog computer, depending on the form of xs(t) (digital number or analogy signal .) Let’s see how a filter works! Output y (t ) xs (kT )h(t kT ) k x(kT )h(t kT ) k Discrete-time algorithm y (kT ) x(kT ) h(0) [ x(kT T )h(T ) x(kT T )h(T ) [ x(kT 2T )h(2T ) x(kT 2T )h(2T )] Weighted sum of the ‘should be point x(kT) and its surrounding points |h(0)| should > h() 0 and |h()| decreases as What is y (t ) x(kT )h(t kT ) ? (In addition to being an algorithm) k Let’s see: xs (t ) x(t ) (t kT ) k or xs (t ) x(kT ) (t kT ) k Page 8-6 EE 422G Notes: Chapter 8 Instructor: Zhang Consider xs(t)*h(t) : a system with impulse response (system parameter) h() and input xs(t)! xs (t ) * h(t ) x ( )h(t )d s x(kT ) ( kT )h(t )d k k x(kT ) ( kT )h(t )d x(kT )h(t kT ) k The reconstruction algorithm y (t ) x(kT )h(t kT ) k Reconstruction Filter: with h(t) as impulse response! Output of the reconstruction Filter (y(t)): Convolution of xs(t) and h() ! 3. Design of Reconstruction Filter: Ideal case Assumption : fs>2fh 1/2 fs > fh (xs(t) was generated at a frequency higher than the Nyquist rate). fh: highest frequency of the original signal Ideal Filter T H( f ) 0 | f | 0.5 f s Otherwise (| f | 0.5 f s ) Question : why do we need this low-pass filter to reconstruct x(t) from xs(t)? answer : xs(t) contains frequencies higher than f h 0.5 f s , but x(t)does not! Question : Will any spectrum (other than x(t)’s introduced by sampling operation remains after the filter? Answer: No. f s 2 f h , has ensured that no overlapping between x(t)’s frequencies and the undesired frequencies in xs(t) introduced by sampling! Page 8-7 EE 422G Notes: Chapter 8 Instructor: Zhang Implementation of Ideal Reconstruction Filter (Given the Impulse response of the filter) Inverse Fourier transform => h(t ) H ( f )e j 2ft df fs / 2 Te j 2ft fs / 2 df T fs / 2 e j 2ft df fs / 2 f fs / 2 T e j 2ft j 2t f f s / 2 T sin( f s t ) T 1 / f s sin( f s t ) t (f s t ) Characteristic of the Ideal Reconstruction Filter: Non causal! Output at t ( y(t) ) must be generated using xs() > t => Not good for real-time application! How to reconstruction x(t) from nT < t < nT + T ? Answer : t sin f s ( k ) T x(t ) y (t ) x(kT ) t k n l 1 f s ( k ) T nl for example t = nT + 0.5T x(nT 0.5T ) y (nT 0.5T ) n l k n l 1 x(kT ) sin f s (nT 0.5 k ) f s (nT 0.5 k ) l points before t = nT + 0.5T l points after t = nT + 0.5T (k=n-l+1,…,n) (k=n+1,…,n+l) Page 8-8 EE 422G Notes: Chapter 8 Instructor: Zhang Part Two 8-3A The z-Transform 1. Definition For Laplace transform, we are given a function x(t), For z-Transform, we are given a sampling sequence: x(0) , x(T), x(2T), … Definition: z-transform of a given sequence x(0) , x(T), x(2T), … is Z ( x(nT )) X ( z ) x(nT ) z n x(0) x(T ) z 1 x(2T ) z 2 n 0 Why do we define such a transform? L[ x(t )] x(t )e st dt x(t) 0 If we want to compute this Laplace transform by computer L[ x(t )] x(nT )e snT T n 0 T x(nT )(e sT ) n n 0 On the other hand Z [ x(nT )] x(nT ) z n z ~ e sT n 0 L[ x(t )] TZ ( xs (t )) z esT x(nT): samples of x(t) Relationship between z- and s-plane Basic Relationship : z e sT z e( j )T eT e jT (1) 0 (eT 1) | z | 1 (note | e jT | 1) l.h.p. (s- plane) inside the unit circle (z- plane) Page 8-9 EE 422G Notes: Chapter 8 (2) 0 s: r.h.p. (3) 0 Instructor: Zhang s: j axis. (4) s = 0 ( 0, 0 ) | z | 1 z: outside the unit circle | z | 1 z: unit circle eT 1 jT e cos T j sin T 1 z=1 Page 8-10 EE 422G Notes: Chapter 8 Instructor: Zhang Note the difference between (t ) and (n ) 2. Basic z-Transform pairs Example 8-4: z-transform of unit pulse (n) : n 0 ( n) n 0 1 x(nT ) 0 unbounded bounded Solution : Similar as Laplace transform 1 X ( z ) x(0) x(T ) z 1 Example 8-5 z-Transform of unit step sequence u(n): n0 n0 1 x(nT ) 0 Solution : X ( z ) x(0) x(T ) z 1 1 z 1 z 2 1 1 z 1 (| z 1 | 1 or | z | 1) How to understand? L[u (t )] Step function u(t) : 1 1 s j Does TX ( z ) z e sT give the same spectrum if T 0 ? TX ( z ) z e sT T 1 z 1 T 0 : z e sT T 1 e sT T 1 e sT s j T 1 cos T j sin T T 1 1 lim T 0 sin T j cos T 1 cos T j sin T j Page 8-11 EE 422G Notes: Chapter 8 Instructor: Zhang z-Transform gives the same spectrum as Laplace transform if the sampling rate Example 8-6 : z-transform of unit exponential sequence x(nT ) et |t nT enT (eT ) n ( 0) (k eT ) kn Solution: X ( z ) x(0) x(T ) z 1 x(2T ) z 2 1 kz 1 k 2 z 2 1 (k 1 z ) 1 (k 1 z ) 2 k 1 z 1 1 2 k 1 z 1 1 T z k e 1 1 1 1 1 kz 1 1 X ( z) T 1 1 e z Is this result reasonable? L[e t 1 s j 1 ] s j s j TX ( z ) | z e sT T 1 e T 0: T jT e T 1 e ( j )T T 1 e ( j )T 1 1 T 0 ( j )e ( j )T j lim Why? Because lim e ( j )T lim e T e jT T 0 T 0 lim e T (cos T j sin T ) 1 T 0 1 1 0 Page 8-12 EE 422G Notes: Chapter 8 Instructor: Zhang Example 8-6 B x(nT ) : k 0 k 1 k 2 1 => X ( z ) 1 kz 1 Summary: Basic z-transform pairs continuous Function sequence (t ) ( n) u (t ) u (n) : 1, 1, e t (e T ) n : 1, e t , e 2t , 1, k , k 2 , (| z | k ) z - transform 1 1 1 z 1 1 1 e t z 1 1 1 kz 1 Would these be sufficient? No! 3. Extended z – transform pairs Page 8-13 EE 422G Notes: Chapter 8 Instructor: Zhang 4. Find z-transform using symbolic tool box Example 8-7 n x(nT ) a n cos 2 Solution: X ( z) n 2 a n cos n 0 Analysis: (1) n : odd => cos(k n z 2 )0 (2) n : even => n 2k cos( 1 n ) cos(k ) 2 1 k : even k : odd n n X ( z ) a n cos z 2 n 0 a cosk z 2k 1 0 2 k 2 4 3 k 0 a0 a 4 z 4 a 8 z 8 5 a 2 z 2 a 6 z 6 a10 z 10 1 a 4 z 4 1 1 2 a 2 z 2 (1 a 4 z 4 a 8 z 8 ) 1 a 2 z 2 1 1 1 1 1 a 2 z 2 1 a 2 z 2 1 1 4 4 1 1 a z 1 a 2 z 2 Very complex! Using Symbolic ToolBox syms a n z xn = a^n*cos(n*pi/2); xz = ztrans (xn, n, z); xz (enter) % Declare symbolic % Define x(n) % Determine X(z) Page 8-14 EE 422G Notes: Chapter 8 Instructor: Zhang xz = z2 1 2 2 a z 1 a 2 z 2 z^2/(a^2+z^2) MatLab: always in terms of z instead of z-1. 8-3B Properties of z - transform 1. Linearity Z ( x1 (nT ) x2 (nT )) Z ( x1 (nT )) Z ( x2 (nT )) X ( z) 2. Initial Value x(0) lim z why? X ( z) x(0) x(1) z 1 (1 z 1 ) X ( z ) 3. Final value x() lim z 1 Why? x() lim sX ( s) s 0 But, s0 1 s s z 1 1 1 z 1 1 z 1 sX ( s ) (1 z 1 ) X ( z ) lim sX ( s ) lim(1 z 1 ) X ( z ) s 0 z 1 Page 8-15 EE 422G Notes: Chapter 8 Instructor: Zhang 8-3C Inverse z-Transform Two Basic Methods: (1) Express X(z) into “Definition Form” X ( z) x(0) x(1) z 1 (very simple, use long division or MatLab: n=8 X = dimpulse(num,den, n) (enter) gives the first n terms) (2) Express X(z) into partial-fraction from X (z ) partial-fraction expansion each term has an inverse transform what Terms? 1 1 z 1 What about if you have ? 1 z 1 1 z 1 1 1 T 1 1 e z 1 kz 1 1 Tz 1 What about if you have ? 1 2 1 2 (1 z ) (1 z ) T 1 1 Te z What about if you have ? T 1 2 T 1 2 (1 e z ) (1 e z ) Az 1 B T 1 2 T 1 2 T 1 (1 e z ) (1 e z ) 1 e z 1 Let’s see: Az 1 B Be T z 1 (1 e T z 1 ) 2 Can we now find A and B? What is the inverse z-transform of 1 (1 e T 1 2 z ) Page 8-16 EE 422G Notes: Chapter 8 What to do if you have Instructor: Zhang 1 ? 1 z 2 1 1 1 1 1 1 z 2 (1 jz 1 )(1 jz 1 ) 2 1 jz 1 1 jz 1 1 1 1 2 1 ( jz 1 ) 1 ( jz 1 ) (1) n / 2 1 1 1 n n n n x(nT ) Z ( ) [( j ) ( j ) ] ( j ) [( 1) 1] 2 2 2 1 z 0 1 n even n odd Important: before doing partial-fraction expansion, make sure the ztransform is in proper rational function of z 1 ! Example 8.9 z2 X ( z) ( z 1)( z 0.2) 1 A B Solution : X ( z ) (1 z 1 )(1 0.2 z 1 ) 1 z 1 1 0.2 z 1 Heaviside’s Expansion Method: X ( z) A B 1 z 1 1 0.2 z 1 1 B(1 z 1 ) z 1 A (1) (1 z ) X ( z ) 1 0.2 z 1 1 0.2 z 1 1 B0 1 A A 1.25 1 0.2 1 0.2 0.8 1 A(1 0.2 z 1 ) B (1 0.2 z 1 ) 1 A(1 0.2 z 1 ) ( 1 0 . 2 z ) X ( z ) B (2) 1 z 1 1 0.2 z 1 1 z 1 1 z 1 1 10.2 z 1 0 ( z 0.2 ) B 1 /(1 5) 1 / 4 0.25 Page 8-17 EE 422G Notes: Chapter 8 X ( z) Instructor: Zhang 1.25 0.25 x(nT ) 1.25 0.25(0.2) n 1 1 1 z 1 0.2 z Example 8-9B MatLab Method (1) Find partial-fraction expansion z2 1 X ( z) 2 1 z 1.2 z 0.2 1 1.2 z 0.2 z 2 b = 1; a = [1 –1.2 0.2]; [r, p, k] = residuez(b,a); r 1.25 r 0.25 p k 1.0 p pole 0.2 1.25 0.25 1 1 1.0 z 1 0.2 z 1 1.25 0.25 1 1 z 1 0.2 z 1 k (2) Directly Find Inverse Transform syms n, z; % Declare symbolic xz = 1/(1-1.2*z^(-1)+0.2*z^(-2)); % define X(z) xn = iztrans(xz,z,n); % compute x(n) xn xn = 5/4-(1/4)*(1/5)^n x(nT) = 1.25-0.25(0.2)n Example 8-10 Y ( z) 1 z 2 1.2 z 0.2 Solution : 2 z2 Question: Define X ( z ) 2 z 2Y ( z ) (or Y ( z ) z X ( z ) ) z 1.2 z 0.2 any relationship between Page 8-18 EE 422G Notes: Chapter 8 Instructor: Zhang x(nT ) z 1 ( X ( z )) and y(nT ) z 1 (Y ( z )) ? 1 z 2 z 2 Y ( z) 2 z 1.2 z 0.2 z 2 ( z 2 1.2 z 0.2) 1 1.2 z 1 0.2 z 2 5(1 1.2 z 1 0.2 z 2 ) 5(1 1.2 z 1 ) 5 6 z 1 5 1 1.2 z 1 0.2 z 2 1 1.2 z 1 0.2 z 2 5 6 z 1 5 6 z 1 V ( z) 1 1.2 z 1 0.2 z 2 (1 z 1 )(1 0.2 z 1 ) A B 1 1 z 1 0.2 z 1 5 6 z 1 A V ( z )(1 z ) | z 1 1 1 0.2 z 1 1 z 1 1 5 6 z 1 B V ( z )(1 0.2 z ) | z 1 5 1 z 1 1 1.25 0.8 1 z 1 5 5 30 6.25 4 1 1 6 . 25 1 z 1 1 0.2 z 1 y (nT ) 5 (n) 1.25 6.25(0.2) n Y ( z ) 5 V ( z ) 5 1.25 y(nT ) 5 (n) 1.25 6.25(0.2) n x(nT ) 1.25 0.25(0.2) n n=0 n=1 n=2 n=3 5 + 1.25 - 6.25 = 0 0 + 1.25 - 6.25*0.2 = 0 0 + 1.25 - 6.25*0.22 = 1 0 + 1.25 - 6.25*0.23 =1.2 1.25 - 0.25 = 1 1.25 - 0.25*0.2 = 1.2 1.25 - 0.25*0.22 = 1.24 1.25 - 0.25*0.23 = 1.248 Why? 6.25*0.2*0.2=0.25 =>1.25 6.25(0.2) n2 1.25 0.25(0.2) n y(n+2) = x(n)! Does Y ( z ) z 2 X ( z ) always imply Z 1 (Y ( z )) has two-step-delay than Z 1 ( X ( z )) ? Yes! Page 8-19 EE 422G Notes: Chapter 8 Instructor: Zhang z-1 : Delay operator! (Must Assume X(nT)(the sequence to be z^(-1) processed)=0 for n<0) 8-3D Delay operator : z-k ( k steps ) ( k > 0 ) Z ( x(nT )) X ( z ) x(nT ) z n n 0 Z ( x(nT kT )) x(nT kT ) z n n 0 z k x(nT kT ) z nk n 0 z k x(nT kT ) z ( n k ) n 0 We want to establish the relationship between Z(x(nT-kT)) and Z(x(nT)) ! Let’s see what’s x(nT kT ) z ( nk ) : n 0 (1) x(nT kT ) z ( nk ) Z ( x(nT )) ? Yes! n k 0 (2) x(nT kT ) z ( n k ) n 0 n k 1 ( n k ) kT x(nT kT ) z ( nk ) x(nT ) z n 0 0 nk n k nT kT 0 x(nT kT ) z ( n k ) n k 0 Z ( x(nT )) Z ( x(nT kT )) z k x(nT kT )) z ( nk ) z k Z ( x(nT )) n 0 x(nT kT ) Z ( x(nT kT )) k steps behind z k x(nT ) Z ( x(nT )) z k Z ( x(nT )) Page 8-20 EE 422G Notes: Chapter 8 Instructor: Zhang 1 Question : If Z ( x(nT )) , what’s 1 0.5 z 1 z 2 Z ( x(nT 2T )) ? Answer: 1 0.5 z 1 8-4 Difference Equation and Discrete-Time Systems Continuous-Time System: Differential Equation, Laplace Transform Discrete-Time System: Difference Equation, z-Transform Properties of Continuous-Time Systems Properties of Discrete-Time Systems 8-4A Properties of Discrete-Time Systems System : Processes input to generate output How to process : system-dependent General symbolic notation for Discrete-Time System: y( nT ) = H [ x(nT) ] what does this notation tell us? operator or Processor 1. Shift-Invariant System (Time-Invariant Systems for continuous-time or general) An example of time-varying system The “processing algorithm” which maps input to output changes! What do we mean by a time-invariant system? Shift-invariant systems: Physical: Mathematic: Assume x(nT): x(0), x(T), … has generated y(nT): y(0), y(T), … For example: Page 8-21 EE 422G Notes: Chapter 8 Instructor: Zhang 0 1 generated 1 1 2 1 3 1 4 x: 1 has y : 0.2 0.4 0.6 0.8 1 If we apply x(nT n0T ) as input look at if n0 2 x(nT 2T ) 0 1 x: 0 0 2 1 3 1 4 1 y (nT 2T ) y : 0 0 0.2 0.4 0.6 generated Question: Is this system shift-invariant? Yes! Question: Is this example telling us H [ x(nT 2T )] y (nT 2T ) ? Yes! Question: Is H [ x(nT 2T )] y (nT 2T ) or H [ x(nT n0T )] y(nT n0T ) always true for different systems? No! only for time-invariant systems! Shift-invariant system: if H [ x(nT n0T )] y(nT n0T ) true for any n0 . 2. Causal and noncausal systems Physical Description: A system is causal or nonanticipatory if the system’s response to an input does not depend on future values of the input. Mathematical Description: Causal system: x1 (nT ) x2 (nT ) What about for n > n0 ? for n n0 Page 8-22 EE 422G Notes: Chapter 8 Instructor: Zhang H [ x1 (nT )] H [ x2 (nT )] for n n0 Why? Although x1(nT) may not be the same as x2(nT) for n > n0 , such difference does not affect the output determined by input up to n = n0 . 3. Linear System Linear System H [1x1 (nT ) 2 x2 (nT )] 1H [ x1 (nT )] 2 H [ x2 (nT )] Linear Systems: can be modeled as y ( nT ) x(kT )h(nT kT ) or k y ( nT ) h(kT ) x(nT kT ) k Convolution h(kT ) : response of the shift-invariant linear system at t=kT to an impulse input applied at t=0. (Or the response at t kT k 0T to an impulse input applied at t k 0T ) Causal systems: h(kT ) 0 k 0 Linear+causal+ x(kT ) 0 y (nT ) (k 0) x(kT )h(nT kT ) k x ( kT ) 0 ( k 0 ) x(kT )h(nT kT ) k 0 causal n (nT kT 0) nT kT n x(kT )h(nT kT ) h(kT ) x(nT kT ) k 0 k 0 Example: Given x(0) = 1, x(T) = 2, x(2T) = 2, x(3T) = 1, … h(0) = 3, h(T) = 2, h(2T) = 1, h(3T) = 0, … MatLab: x = [1 2 2 1 1]; h = [3 2 1]; y = conv(x,h); y 3 8 11 9 7 3 1 Example 8-13: n 1 x ( nT ) u( n ) n n 2 1 1 y ( nT ) x ( nT ) * h( nT ) 3 2 n 2 3 1 h( nT ) u( n ) 3 Page 8-23 EE 422G Notes: Chapter 8 Instructor: Zhang Can you write a program (algorithm) to calculate y(nT) = x(nT)*h(nT) ? Example 8-13: Symbolic Tool Box syms n z xn =(1/2)^n hn = (1/3)^n xz = ztrans(xn, n, z) hz = ztrans(hn, n, z) yz = xz*hz yn = iztrans (yz, z, n); yn (enter) yn = 3*(1/2)^n-2*(1/3)^n % Declare Symbolic % x(n) % h(n) % z-transform of x(n) % z-transform of h(n) % multiply, not convolution % Do you know why? % y(nT)=3(1/2)n – 2(1/3)n * Analytic solution of convolution n0 x(nT ) x(nT ) n0 0 n0 h(nT ) h(nT ) n0 0 Z ( x(nT ) * h(nT )) X ( z ) Z ( x(nT )) H ( z ) Z (h(nT )) Z { x(kT )h(nT kT )} k n 0 k { x(kT )h(nT kT )} z n k n 0 x(kT ) h(nT kT ) z n<kn-k<0 n k 1 n x ( kT ) h ( nT kT ) z h(nT kT ) z n k 0 nk n 0 x ( kT ) 0 k 0 k 0 nk x(kT ) h(nT kT ) z n x(kT )z k k 0 h(nT kT ) z z ( nk ) n k n k 0 X ( z)H ( z) Page 8-24 EE 422G Notes: Chapter 8 Instructor: Zhang 1 i.e. x(nT ) * h(nT ) Z [ X ( z ) H ( z )] n 3 1 1 1 Example: x ( nT ) u( n 3) 4 4 4 n 5 1 1 1 h(nT ) u(n 5) 3 3 3 x(nT ) 0 h(nT ) 0 if n 0 n 3 u(n 3) n 5 u(n 5) Find x(nT)*h(nT) Solution: x(nT ) 0 h(nT ) 0 n 0 n 0 u(n 3) due to u(n 5) 3 n 1 3 3 1 1 1 X ( z ) z Z u (n) z 1 4 4 4 1 z 1 4 5 n 5 1 1 H ( z ) Z u (n 5) 3 3 3 5 n 1 5 1 z Z u (n) 3 3 5 1 1 z 5 1 3 1 z 1 3 3 5 1 1 1 X ( z ) H ( z ) z 8 1 1 4 3 (1 z 1 )(1 z 1 ) 4 3 1 3 4 V ( z) 1 1 1 1 (1 z 1 )(1 z 1 ) 1 z 1 1 z 1 4 3 4 3 Page 8-25 EE 422G Notes: Chapter 8 Instructor: Zhang 1 1 Z 1 (V ( z )) (4( ) n 3( ) n )u (n) 3 4 1 1 Z 1 ( z 8V ( z )) (4( ) n 8 3( ) n 8 )u (n 8) 3 4 1 1 1 1 x(nT ) * h(nT ) ( ) 3 ( ) 5 [4( ) n 8 3( ) n 8 )]u (n 8) 4 3 3 4 4. Stable system Consider linear shift-invariant systems only. Definition of BIBO stable: | y(nT ) | n for all bounded x(nT). Derivation of Criterion y ( nT ) x(kT )h(nT kT ) k x(kT) bounded => | x ( kT ) | M y (nT ) | x(kT ) || h(nT kT ) | k M | h(nT kT ) | k M | h(nT kT ) | k M | h(kT ) | k Criterion: | h(kT ) | k How to use this criterion: A h: h(0), h(1), … h(N), 0, 0, … h(k ) 0 k 0 Limited number of terms (causal) Page 8-26 EE 422G Notes: Chapter 8 Instructor: Zhang causal + Limited N => stable Why! y ( nT ) x(kT )h(nT kT ) k y (nT ) , N n k 0 n k n N for any fixed n in for example, n 100 100 k 90 N 10 | y (100T ) || 100 x(kT )h(100T kT ) | k 90 100 M | h(100T kT ) | k 90 N M | h(kT ) | stable k 0 In general y (nT ) x(kT )h(nT kT ) k n x(kT )h(nT kT ) k n N | y (nT ) | M n | h(nT kT ) | k n N N M | h(nT ) | k 0 Limited Terms Conclusion: limited terms of h => stable! Example : y ( nT ) n x(kT )h(nT kT ) stable? k 0 What about y ( nT ) n x(kT )h(nT kT ) ? k n 100 How to use this criterion: B If we know Z(h(nT)) = H(z) Page 8-27 EE 422G Notes: Chapter 8 H ( z) Instructor: Zhang 1 1 0.2 z 1 0.22 z 2 1 1 0.2 z h(0) h(1) h(2) | h(nT ) | n 0 Why? |0.2| < 1 ! What about H ( z ) 1 1 z 1 z 2 1 1 z Not BIBO stable! In general, deg(num) < deg(den) H ( z) (1 p1 z ) (1 p n z n ) 1 | p j | 1 stable (poles inside the unit circle!) Example 8-14: h(nT ) [4(1/ 3) n 3(1/ 4) n ]u(n) h(nT ) 0 n 0 Due to u (n) Solution: n n 0 | h(nT ) | | h(nT ) | n 0 n 0 | 4(1 / 3)n 3(1 / 4)n | 4(1 / 3)n 4 1 6 1 1/ 3 Stable 8-4B Difference Equations 1. Difference Equations Problem: determine the output of the system at the present time : t = nT y(nT) What information to use: Page 8-28 EE 422G Notes: Chapter 8 Instructor: Zhang (1) input: current input u(nT) previous input u(kT) (k < n) future input u(kT) (k > n) causal system : no future input! Previous input u (n 1), u (n 2), ..., u (n ) We do not need all of them use u(n-1), … , u(n-m) (2) output: previous output (its history): y(kT) (k < n) ? Yes. future output y(kT) (k >n) ? No, no future output previous outputs y (n 1), y (n 2), ..., y (n ) We do not need all of them! y(n-1), …. , y(n-r) would be sufficient! Mathematical Equation y(nT) : depends on u(n 1),, u(n m) y (n 1),..., y (n r ) Difference Equation linear system y (nT ) L0 x(nT ) L1 x(nT T ) ... Lm x(nT mT ) k1 y (nT T ) ... k r y (nT rT ) weights: Li , K i Larger weight: more important in determining y(nT) Would the weights be the same? No! (r, m): system’s order different systems: different order and weights (parameters) 2. z-transfer function Different Equation y (nT ) k1 y (nT T ) ... k r y (nT rT ) L0 x(nT ) L1 x(nT T ) ... Lm x(nT mT ) z-transform => Y ( z ) k1 z 1Y ( z ) ... kr z rY ( z ) L0 X ( z ) L1 z 1 X ( z ) ... Lm z m X ( z ) Page 8-29 EE 422G Notes: Chapter 8 Instructor: Zhang Y ( z ) L0 L1 z 1 ... Lm z m H ( z) X ( z) 1 k1 z 1 ... k r z r z-transfer function Y(z) = H(z)X(z) Why H(z) is the z-transform of impulse response h(nT) ? 8-4C Steady-State Frequency Response of a Linear Discrete-Time System ( X ( s) L[ x(t )]) x(t)’s spectrum X ( s) s j X ( ) Y ( ) x(nT)’s spectrum TX ( z ) z e sT TX ( z ) z e jT s j (Y ( s) L[ y(t )]) y(t)’s spectrum Y ( s) s j y(nT)’s spectrum TY ( z ) z e sT TY ( z ) z e jT s j System’s frequency response Y ( ) TY ( z ) |z e jT Y ( z ) X ( ) TX ( z ) |z e jT X ( z ) z e jT What is Y(z)/X(z) ? H(z) = Y(z)/X(z) System frequency response Y ( ) / X ( ) H ( z ) |z e jT H (e jT ) Page 8-30 EE 422G Notes: Chapter 8 Instructor: Zhang Property of frequency response H (e jT ) T: sampling period s 2f s 2 1 : sampling frequency T e j ( k s )T e jT e jk sT s T 2 e jT e jk 2 e jT (cos k 2 j sin k 2 ) e jT H (e j ( k s )T ) H (e jT ) Frequency Response H: periodic function with period s when the frequency increase by k s , the system’s frequency response does not change. Example: Input 1: 10 sin( 5t ) T = 1 second Input 2 : 10 sin( 7t ) Generate the same output amplitude? Normalized Frequency r / s s : frequency period s 2 e jT e j 2 / s r / s 1 T 2 / s T e j 2r Frequency Response in terms of r (argument) H (e j 2r ) Amplitude Response | H (e j 2r ) | or Phase Response H (e j 2r ) or | H (e jT ) | H (e jT ) Question: what are their physical meaning? Example 8-15: y(nT) = x(nT) + x(nT-2T) Solution : Y ( z ) X ( z ) z 2 X ( z ) Page 8-31 EE 422G Notes: Chapter 8 H ( z) Instructor: Zhang Y ( z) 1 z 2 X ( z) H (e jT ) 1 z 2 z e jT 1 e j 2T (e jT e jT )e jT 2 cos T e jT H (e j 2r ) 2 cos 2r e j 2r | H ( e j 2r ) | 2 | cos 2r | cos 2r 0 2r H ( e j 2r ) cos 2r 0 2r Comment: z-transform: good for analysis difference equation: computer program Page 8-32