1 Lecture 2: System’s description (LTI, causal, finite dimensional) Instructor: Dr. Gleb V. Tcheslavski Contact: gleb@ee.lamar.edu Office Hours: Room 2030 Class web site: http://ee.lamar.edu/gleb/dsp/ind ex.htm ELEN 5346/4304 DSP and Filter Design Fall 2008 2 The Signal Flow Graph Let a yn k bm xn-m for a BIBO stable LTI finite-dimensional system m Assumea0 1 yn a yn k bm xn m (2.2.1) m * Δ yn is a function of previous (and current) inputs and outputs, therefore the system is causal. If ak and bm are constants → Linear Constant coefficient Difference Equation yn + Signal Flow Graph (SFG): * + a1 a2 z-1 z-1 yn-1 xn-1 memory xn-2 b0 z-1 z-1 z-1 b1 b2 … … yn-2 z-1 xn Remark: because we discretize the signal by amplitude, the system becomes non-linear! ELEN 5346/4304 DSP and Filter Design Fall 2008 + + Δ 3 Difference equation yn a ynk bm xnm (2.3.1) m yn is a function of ak, bm, xn, memory = states. or yn hk xnk h0 xn hk xnk (2.3.2) If we know the initial conditions, we can solve the difference equation: x1,n y1,n predictionsof system ' sbehavior x2,n y2, n If (2.3.1) is a linear constant coefficient difference equation, the output consists of a homogeneous solution and a particular solution: yn = yh,n + yp,n (2.3.3) ELEN 5346/4304 DSP and Filter Design Fall 2008 4 Homogeneous (complementary) solution xn 0,n 0theoutputisduetotheinitialconditions N Homogeneous equation: a y k 0 k n k 0 (2.4.1) N Let ' sassume yn ak nk 0 n (2.4.2) k 0 Then nN (a0 N a1 N 1 an1 an ) 0is a characteristicequation (2.4.3) The characteristic equation has N characteristic roots: 1, … N • If the system is real, xn, yn, and all the coefficients (a, b) are real • For LTI systems hn are real → roots are either real or complex conjugate pairs N n • yh ,n ci i is a form of solution, where ci depend on initial conditions. (2.4.4) • if 1 = 2, solution in form of (2.4.5) i 1 ELEN 5346/4304 DSP and Filter Design yh,n c11n c2n1n Fall 2008 5 Particular solution “We apply an input and see what happens to the output for large n” Assumption: yp,n has the same form as xn: xn yp,n c k k0nM k1nM 1 An M cos 0norsin 0n k1 cos 0n k2 sin 0n cn knc n An (k0nM k1nM 1 An nM 5 1 1 yn 1 yn 2 xn xn 1 y1 6; y2 6;x1 1;xn 2n ,n 0 6 6 2 Characteristic equation: Ex.: kM yn n n kM ) (2.5.1) 5 1 1 1 1 1 1 1 0 01 ;2 yh,n c1 c2 ;n 0 formof solution 6 6 2 3 2 3 2 3 (2.5.2) 2 ELEN 5346/4304 DSP and Filter Design Fall 2008 6 Alternative form xn Memory is empty (relaxed system) xn-1n xn-1n z-1 + z-1 + … We can find a solution in form: yn = yzi,n + yzs,n (zero input + zero state) (2.6.1) The output will be linear iff both: zero input and zero state are linear. For a relaxed system: xn = n → yn = hn ELEN 5346/4304 DSP and Filter Design - zero state response! Fall 2008 (2.6.2) 7 Example… Let’s find yzi,n for the example in (2.5.1). Assuming zero input, yp,n = k (a constant) and must satisfy the LCCDE 5 1 k k k 0 6 6 Difference Equation: (2.7.1) (2.7.2) “large enough n” implies that all terms are “active”: in our case, n ≥ 2 n k 0 yzi ,n n 1 1 c1 c2 ;n 0 2 3 In form of homogeneous solution 0 0 5 1 1 5 1 1 1 1 n 0 y0 y1 y2 x0 x1 6 6 0 0 4 c1 c2 c1 c2 6 6 2 6 6 2 2 3 5 1 1 5 1 1 1 1 1 n 1 y1 y0 y1 x1 x0 4 6 0 0 2 c1 c2 6 6 2 6 6 2 3 2 3 4 1 7.5 c1 1 c 2 1 3 c 1/ 2 1/ 3 2 3 ELEN 5346/4304 DSP and Filter Design n (2.7.6) yzi ,n Fall 2008 (2.7.3) (2.7.4) (2.7.5) n 1 1 7.5 3 ;n 0 2 3 (2.7.7) 8 Example (cont) yn 5 1 1 yn 1 yn 2 xn xn 1 y1 6; y2 6;x1 1;xn 2n ,n 0 6 6 2 n Got so far: yh,n is LTI n n (2.8.1) n 1 1 1 1 c1 c2 ;n 0 (2.8.2) yzi ,n 7.5 3 ;n 0 2 3 2 3 Sincexn cn 2n y p,n kcn k 2n (2.8.3) (2.8.4) By substituting (2.8.4) into (2.8.1) for large n, we get: 5 1 1 10 1 k 2n k 2n 1 k 2n 2 2n 2n 12n 2 4k k 2n 2 (4 1) 6 6 2 6 6 1 n 4 1 k 5k 2 y p ,n 2 2 2 n n 1 1 Totalsolution : yn yh,n y p,n c1 c2 2 2n 2 3 5 1 y2 y1 y2 c1 6 6 DE y1 c1 ELEN 5346/4304 DSP and Filter Design (2.8.5) (2.8.6) n n 1 1 n c2 5.5 (2.8.7) y 5.5 2 2 2 n 0 c tot , n 2 3 (2.8.8) 2 c2 Fall 2008 9 Example (cont 2) What’s about the zero state solution yzs,n? xn 2n ,same y p ,n 2 2n ;n 0 (2.9.1) n n 1 1 initialconditions0 yzs ,n c1 c2 2 2n ;n 0 2 3 1 1 y 0 0 1 { (2.9.2 ) } c c 2 1 1 2 0 2 2 (2.8.1) y1 y2 0 y 5 y 0 2 1 {(2.9.2)} c1 c2 4 3 3 1 6 0 2 2 3 4 (2.9.2) (2.9.3) c1 2;c2 1 n n 1 1 yzs ,n 2 2 2n ;n 0 2 3 n n n n 1 1 1 1 ytot ,n yzs ,n yzi ,n 2 2n 5.5 2 2 2n n 0 2 3 2 3 (2.9.5)(2.8.8) ELEN 5346/4304 DSP and Filter Design Fall 2008 (2.9.4) (2.9.5) 10 Example (cont 3) n ytot ,n transient response n For large n ELEN 5346/4304 n steady state response n 0 ytr ,n 0; DSP and Filter Design (2.10.1) ytot ,n ytr ,n yss,n Another representation: 1 1 ytr ,n 5.5 2 2 3 yss ,n 2 2n n 1 1 5.5 2 2 2n n 0 2 3 (2.10.2) yss,n dominates Fall 2008 11 Example (cont 4) To evaluate a system’s unit pulse response hn 1 when xn n for the relaxed system : i.c. = 0 y p ,n k n ; yh ,n c11 c22 ;1 ;2 2 (2.11.1) 5 1 1 Particular: k n k n 1 k n 2 n n 1largenk 0 0cannot solve for k! 6 6 2 (2.11.2) n n 1 1 (2.11.3) ytot ,n c1 c2 k n ;n 0 2 3 5 1 1 h0 h1 h2 x0 x1 hn 0n 0(causalsystem) c1 c2 k (2.11.4) DE: 6 6 2 5 1 1 1 1 1 (2.11.5) h1 h0 h1 x1 x0 c1 c2 0 6 6 2 2 2 3 5 1 1 8 1 1 1 h2 h1 h0 x2 x1 c1 c2 0 6 6 2 6 6 2 3 2 ELEN 5346/4304 DSP and Filter Design Fall 2008 2 (2.11.6) 12 Example (cont 5) c1 6 c2 5 k 0 (2.12.1) n n n n 1 1 1 1 forn 0hn 6 5 6 5 un 2 3 2 3 hn hnun forcausalsystems n 1 Note :6 lasts forever !thisisanInfiniteImpulseResponse(IIR)system 2 h n thesystemisBIBOstable n ELEN 5346/4304 DSP and Filter Design Fall 2008 (2.12.2)