1 Lecture 7: Z-transform Instructor: Dr. Gleb V. Tcheslavski Contact: gleb@ee.lamar.edu Office Hours: Room 2030 Class web site: http://ee.lamar.edu/gleb/ds p/index.htm ELEN 5346/4304 DSP and Filter Design Fall 2008 2 Definitions Z-transform converts a discrete-time signal into a complex frequency-domain representation. It is similar to the Laplace transform for continuous signals. X ( z ) xn z n n is an integer time index; When the magnitude r =1, n (7.2.1) If (where) it exists! z re j is a complex number; - angular freq. z e j X ( z) X e j xne jn n If it exists! ELEN 5346/4304 DSP and Filter Design Fall 2008 (7.2.2) 3 Region of Convergence (ROC) The Region of convergence (ROC) is the set of points z in the complex plane, for which the summation is bounded (converges): n x z n (7.3.1) n Since z is complex: z re j (7.3.2) X ( z) xn r n e jn xn r n e jn n In general, z-transform exists for Im rELEN 5346/4304 DSP and Filter Design r+ (7.3.3) n r r r (7.3.4) r z r (7.3.5) Re Fall 2008 4 Region of Convergence (ROC) Examples of ROCs from Mitra ELEN 5346/4304 DSP and Filter Design Fall 2008 5 Region of Convergence (ROC) X ( z ) xn z n n xn ...,a 2 ,a 1 ,,a,a 2 ,... Example 7.1: Let xn = an There are no values of z satisfying: n xn z n Example 7.2: Let xn = an un – a causal sequence X ( z ) a un z n n n az 1 n n 0 1 1 az 1 (7.5.1) Im for az 1 1 z a ROC a We can modify (7.5.1) as X ( z) N ( z )0z 0zero( s) D( z )0z a pole( s) ELEN 5346/4304 DSP and Filter Design z N ( z) z a D( z ) Re (7.5.2) roots of numerator: X(z) = 0 roots of denominator: X(z) Fall 2008 x 6 Region of Convergence (ROC) X ( z ) xn z n n Example 7.3: Let xn = -an u-n-1 – an anticausal sequence X ( z ) a u n 1 z n n n az m 1 1 m 1 n 1 0n 1 a n z 1 n n a z 1 m 1 m a 1 z z 1 a 1 z z a (7.6.1) Im for a 1 z 1 z a x a Conclusion 1: z-transform exists only within the ROC! Conclusion 2: z-transform and ROC uniquely specify the signal. Conclusion 3: poles cannot exist in the ROC; only on its boundary. Note: if the ROC contains the unit circle (|z| = 1), the system is stable. ELEN 5346/4304 DSP and Filter Design Fall 2008 Re 7 The transfer function X ( z ) xn z n n Time shift: xnm xnm z ( nm) z m {l n m} z m xl z l z m X ( z ) z n l i LTI: a y Consider an LCCDE: and take z-transform utilizing time shift i n i b j xn- j (7.7.2) j i j a z Y ( z ) b z i j X ( z) i n (7.7.3) j yn hn xn Y ( z ) hm xn m z n hm xn m z ( nm ) z m H ( z ) X ( z ) z (7.7.1) m m (7.7.4) n X (z) M The system transfer function H ( z) Y ( z) X ( z) b z j 0 N i a z i i 0 ELEN 5346/4304 DSP and Filter Design Fall 2008 j j hn z n n (7.7.5) 8 Rational z-transform X ( z ) xn z n n Frequently, a z-transform can be described as a rational function, i.e. a ratio of two polynomials in z-1: P( z ) n0 n1 z 1 ... nM 1 z ( M 1) nM z M (7.8.1) H ( z) 1 ( N 1) N D( z ) d0 d1 z ... d N 1 z dN z Here M and N are the degrees of the numerator’s and denominator’s polynomials. An alternative representation is a ratio of two polynomials in z: M ( M 1) ... nM 1 z nM P( z ) N M n0 z n1 z H ( z) z D( z ) d0 z N d1 z N 1 ... d N 1 z d N (7.8.2) Finally, a rational z-transform can be written in a factorized form: M H ( z) n0 (1 z j z 1 ) j 1 N d0 (1 pi z 1 ) M z ( N M ) i 1 zeros: numerator = 0 n0 ( z z j ) j 1 N d 0 ( z pi ) (7.8.3) i 1 Poles: denominator = 0 ELEN 5346/4304 DSP and Filter Design Fall 2008 9 Notes on poles of a system function Positions of poles of a transfer function are used to evaluate system stability. Let assume a single real pole at z = . Therefore: Y ( z) 1 H ( z) zY z Y z X z X ( z) z (7.9.1) The difference equation is: yn1 yn xn yn yn1 xn1 (7.9.2) Therefore, the impulse response is: for hn hn1 n1 (7.9.3) n 0,1, 2,3,...hn 0,1, , 2 ,... (7.9.4) Iff || < 1, hn decays as n and the system is BIBO stable; otherwise, hn grows without limits. Therefore, poles of a stable system (and signals in fact) must be inside the unit circle. Zeros may be placed anywhere. Zeros at the origin produce a time delay. ELEN 5346/4304 DSP and Filter Design Fall 2008 The transfer function and the Frequency response BIBO: j h z 1 RO C H e H ( z) ze j n 10 X ( z ) xn z n n (7.10.1) n M H ( z) b z j 0 N b0 z j a z i 0 j i i M M (z z ) j 1 N j a0 z N ( z pi ) (7.10.2) i 1 where zj are zeros and pi are poles of the transfer function. M e j z j M H e j b0 1 jN1 b0 e j ( M N ) (e j z j ) a0 j 1 e j pi (7.10.3) H e j N i 1 a0 (e j pi ) M N i 1 j j j H e b0 a0 ( M N ) (e z j ) (e pi ) j 1 i 1 BIBO: ELEN 5346/4304 DSP and Filter Design Fall 2008 The transfer function and the Frequency response 11 X ( z ) xn z n n A good way to evaluate the system’s frequency response: H e j k 2 k p M P DFT bm 0 N P DFT ak 0 (7.11.1) Zero-padded When the frequency approaches a pole, the frequency response has a local maximum, a zero forces the response to a local minimum. For real systems, poles and zeros are symmetrical with respect to the real axis. Magnitude of H() 1.5 1 0.5 0 0 0.1 pole ELEN 5346/4304 DSP and Filter Design 0.2 0.3 Fractional frequency, 0.4 0.5 zero Fall 2008 12 The transfer function and the SFG zS ( z ) AS ( z ) bX ( z ) Sn 1 ASn bxn T T z y c S dx Y ( z ) c S ( z ) dX ( z ) n n n (7.12.1) zI A S ( z ) bX ( z )S ( z ) zI A bX ( z ) 1 Y ( z ) c T zI A bX ( z ) dX ( z ) (7.12.2) H ( z ) c T zI A b d (7.12.3) 1 1 Poles of H(z) correspond to the eigenvalues of the system matrix. ELEN 5346/4304 DSP and Filter Design Fall 2008 13 More on Transfer function M H ( z) P( z ) D( z ) b m0 N m z m k a z k k 0 b0 z M M (z z ) j 1 N j a0 z N ( z pi ) zeros M b0 ( M N ) z a0 i 1 (z z m 1 N m ) (7.13.1) (z p ) k 1 k poles 1) N > M: zeros at z = 0 of multiplicity N-M 2) M > N: poles at z = 0 of multiplicity M-N j ( M N ) e 1constmagnitude j ( M N ) e distortionless j ( M N ) (M N )lin. phase e H e j BIBO z | 1 ROCH ( z ) j H e j j z e H e H ( zi ) 0Y ( zi ) 0;H ( pi ) Y ( pi ) ELEN 5346/4304 DSP and Filter Design Fall 2008 (7.13.2) 14 Types of digital filters 1. FIR (“all-zero”) filter: bn n M H ( z ) bm z hn m 0 0otherwise M m All poles are at z = 0: a “nest of poles” ROC: the entire z-plane except of the origin (z = 0). FIR filters are stable. ELEN 5346/4304 DSP and Filter Design Fall 2008 (7.14.1) 15 Types of digital filters 2. IIR (“all-pole”) filter: H ( z) b0 N a z k 0 k All zeros are at z = 0: a “nest of zeros” ELEN 5346/4304 DSP and Filter Design Fall 2008 k (7.15.1) 16 Types of digital filters 3. General IIR (“zero-pole”) filter: M H ( z) b m0 N k a z k k 0 ELEN 5346/4304 DSP and Filter Design m z m Fall 2008 (7.16.1) 17 On test signals… Y ( z ) H ( z ) X ( z ) yn hn xn (7.17.1) z n hn (7.17.2) xn xk nk yn (7.17.3) k Calculate yn xk hnk and compare to yn k We don’t need any other that a delta function test signals since a unit-pulse response is a complete system’s description. ELEN 5346/4304 DSP and Filter Design Fall 2008 18 Types of sequences and convergence 1. Two-sided: G( z ) N n M gn z n Converges everywhere except of z = 0 and z = ELEN 5346/4304 DSP and Filter Design Fall 2008 (7.18.1) 19 Types of sequences and convergence 2. Right-sided: G( z ) Blows up at z = n M gn z n 1 n M gn z n gn z n (7.19.1) n 0 Assume: if converges at z = z0, converges for |z| > | z0| ROC: r - < |z| < - exterior ROC For a causal sequence: |z| > r - = max|pk| - a max pole of G(z) To be causal, a sequence must be right-sided (necessary but not sufficient) ELEN 5346/4304 DSP and Filter Design Fall 2008 20 Types of sequences and convergence 3. Left-sided: G( z ) N n gn z n 1 n N gn z n gn z n Converges at z0 n 0 Blows up at z = 0 ROC: 0 < |z| < r + - interior ROC When encountering an interior ROC, we need to check convergence at z = 0. If the sequence “blows up” at zero – it’s an anti-causal sequence ELEN 5346/4304 DSP and Filter Design Fall 2008 (7.20.1) 21 Properties from Mitra ELEN 5346/4304 DSP and Filter Design Fall 2008 22 Common pairs ELEN 5346/4304 DSP and Filter Design Fall 2008 23 Inverse z-transform xn 1 2 j X ( z ) z n1dz (7.23.1) c Where C is a counterclockwise closed path encircling the origin and is entirely in the ROC. Contour C must encircle all the poles of X(z). In general, there is no simple way to compute (7.23.1) A special case: C is the unit circle (can be used when the ROC includes the unit circle). The inverse z-transform reduces to the IDTFT. 1 xn 2 ELEN 5346/4304 DSP and Filter Design X e j e jn d Fall 2008 (7.23.2) 24 Inverse z-transform X ( z ) xn z n n A. Via Cauchy residue theorem xn i (7.24.1) i For all poles of X(z)zn-1 inside C (contour of integration) Where i are the residues of X(z)zn-1 for a pole of multiplicity k: Residue function: ELEN 5346/4304 DSP and Filter Design 1 d k 1i ( z ) i (k 1)! dz k 1 i ( z ) X ( z ) z Fall 2008 n 1 (7.24.2) z pi z pi k (7.24.3) 25 Inverse z-transform: Example X ( z ) xn z n n X ( z) Example: xn 1 2 j C z ; a z za z n1 1 z dz za 2 j C Im a n 0 zn dz z a a 0 n 0 0 is a residue of X(z)z-n-1 at z=0 – involves pole of x a C Re multiplicity –n wnen n < 0. 1 d k 1 zn k n a ( z a ) z (k 1)! dz k 1 z a k 1 z a an multiplicity 1 d k 1 k ( z a)1 1 k k n Let n k(n 0)0 z ( 1)( 2)...( ( k 1))( z a ) a a z 0 (k 1)! dz k 1 z k (k 1)! n a n 0 n xn n x a un n n a a 0n 0 ELEN 5346/4304 DSP and Filter Design Fall 2008 26 Inverse z-transform X ( z ) xn z n n B. Via recognition (table look-up) Sometimes, the z-transform can be modified such way that it can be found in a table… Example: X ( z) e n an n a a 1 X ( z) e z ; 1 z a z n 0 z n ! n 0 n ! a z Therefore: ELEN 5346/4304 DSP and Filter Design a z an xn un n! Fall 2008 27 Inverse z-transform X ( z ) xn z n n C. Via long division 1. Right-sided z-transform sequences can be expanded into a power series in z-1. The coefficient multiplying z-n is the nth sample of the inverse z-transform. Example: z 2 2 z 1 2 X ( z) ; z 1 z 1 1 Lower powers first: and long division: 2 2 z 1 z 2 X ( z) ; z 1 1 z 1 2 z2 z3 z4 ... 1 z 1| 2 2 z 1 z 2 2 2 z 1 ELEN 5346/4304 DSP and Filter Design z 2 z 2 z 3 Fall 2008 xn {2,0,1, 1,1,...} x0 x1 x2 x3 x4 28 Inverse z-transform X ( z ) xn z n n 2. Left-sided z-transform sequences – into a power series in z1… Example: z 2 2 z 1 2 X ( z) ; z 1 z 1 1 Multiply both numerator and denominator by z2 … z 2 z 2 2 z 1 2 1 2 z 2 z 2 X ( z) 2 z z 1 1 z z2 Long division… X ( z)z 1 1 z z 2 z3 z 4 ... x1 ELEN 5346/4304 DSP and Filter Design x0 x-1 x-2 Fall 2008 x-3 x-4 Non-causal 29 Inverse z-transform X ( z ) xn z n n Example: Example: z 2 2 z 1 2 1 X ( z ) 1 ; z 1 1 2 z 1 z 2 z X ( z) z 1 ; z 1 z 1 z 1 1 xn n 1 n (1)n u n 1 ELEN 5346/4304 DSP and Filter Design Fall 2008 not suitable for long division! 30 Inverse z-transform X ( z ) xn z n n D. Via partial fraction expansion (PFE) LetG ( z ) N ( z) ;r z r D( z ) (7.30.1) If the degree of the numerator is equal or greater than the degree of the denominator: M N, G(z) is an improper polynomial. Then: N ( z ) M N l N1 ( z ) G( z ) cl z D( z ) l 0 D( z ) (7.30.2) A proper fraction: M1 < N Then: G( z ) M N N cl z l 0 l l 1 l 1 pl z 1 Simple poles: multiplicity of 1. ELEN 5346/4304 DSP and Filter Design Fall 2008 (7.30.3) 31 Inverse z-transform X ( z ) xn z n n Here l is a residue l ( z pl )G ( z ) z n l z pl (7.31.1) poles Therefore: gn M N c l 0 l n p u if external ROC : p r l l n l n p u if internal ROC : p r l 1 l l n l N n l This method is suitable for complex poles. Problem: large polynomials are hard to manipulate… ??QUESTIONS?? ELEN 5346/4304 DSP and Filter Design Fall 2008 (7.31.2)