z-transforms Definition of the z-transform Chapter

Transcription

1 z-transforms Chapter 7 In the study of discrete-time signal and systems, we have thus far considered the time-domain and the frequency domain. The z- domain gives us a third representation. All three domains are related to each other. A special feature of the z-transform is that for the signals and system of interest to us, all of the analysis will be in terms of ratios of polynomials. Working with these polynomials is relatively straight forward. Definition of the z-transform Given a finite length signal as xn [ ], the z-transform is defined N Xz ( ) xk [ ]z k k 0 k 0 xk [ ]( z ) k (7.) where the sequence support interval is [0, N], and z is any complex number This transformation produces a new representation of denoted Xz ( ) Returning to the original sequence (inverse z-transform) xn [ ] requires finding the coefficient associated with the nth power of z N xn [ ] 7

2 Definition of the z-transform Formally transforming from the time/sequence/n-domain to the z-domain is represented as A sequence and its z-transform are said to form a z-transform pair and are denoted (7.2) In the sequence or n-domain the independent variable is n In the z-domain the independent variable is z Example: Using the definition Thus, N n-domain z-domain xn [ ] xk [ ]δ[ n k] Xz ( ) k 0 xn [ ] δ[ n n 0 ] N z z xn [ ] Xz ( ) z N k 0 xk [ ]z k Xz ( ) xk [ ]z k δ [ k n ]z k 0 z n 0 k 0 N k 0 δ[ n n 0 ] z n 0 z 7 2

3 The z-transform and Linear Systems Example: xn [ ] 2δ[ n] + 3δ[ n ] + 5δ[ n 2] + 2δ[ n 3] By inspection we find that Example: Xz ( ) 4 5z 2 + Xz ( ) 2 + 3z + 5z 2 + By inspection we find that z 3 2z 4 2z 3 xn [ ] 4δ[ n] 5δ[ n 2] + δ[ n 3] 2δ[ n 4] What can we do with the z-transform that is useful? The z-transform and Linear Systems The z-transform is particularly useful in the analysis and design of LTI systems The z-transform of an FIR Filter We know that for any LTI system with input impulse response hn [ ], the output is yn [ ] xn [ ]*h[ n] xn [ ] and (7.3) We are interested in the z-transform of FIR filter hn [ ] b k δ[ n k] k 0 hn [ ], where for an (7.4) 7 3

4 To motivate this, consider the input The output is The z-transform and Linear Systems The term in parenthesis is the z-transform of known as the system function of the FIR filter Like ( ) function as yn [ ] xn [ ] z n, < n < yn [ ] b k xn [ k ] b k z n k He jω k 0 k 0 b k z n z k k 0 k 0 b k z k n z (7.5) (7.6) hn [ ], also was defined in Lecture 5, we define the system Hz ( ) b k z k k 0 k 0 hk [ ]z k The z-transform pair we have just established is z hn [ ] Hz ( ) (7.7) k 0 b k δ[ n k] z k 0 b k z k Another result, similar to the frequency response result, is yn [ ] hn [ ]*z n Hz ( )z n (7.8) 7 4

5 The z-transform and Linear Systems Note if z, we in fact have the frequency response result of Chapter 6 e jωˆ The system function is an th degree polynomial in complex variable z As with any polynomial, it will have roots or zeros, that is there are values z 0 such that Hz ( 0 ) 0 These zeros completely define the polynomial to within a gain constant (scale factor), i.e., Hz ( ) b 0 + b z + + b z where ( z z )( z 2 z ) ( z z ) ( z z )( z z 2 ) ( z z ) z z k, k,, denote the zeros Example: Find the Zeros of hn [ ] δ[ n] + The z-transform is --δ [ n ] 6 --δ [ n 2] 6 Hz ( ) + --z z z --z 2 3 z z -- 2 z 3 7 5

6 The zeros of Hz ( ) are -/2 and +/3 The difference equation yn [ ] 6x[ n] + xn [ ] xn [ 2] has the same zeros, but a different scale factor; proof: Properties of the z-transform Properties of the z-transform The z-transform has a few very useful properties, and its definition extends to infinite signals/impulse responses The Superposition (Linearity) Property ax [ n] + bx 2 [ n] ax ( z) + bx 2 ( z) z (7.9) proof Xz ( ) ( ax [ n] + bx 2 [ n] )z n 0 N a x [ n ]z + b x 2 n n 0 ax ( z) + bx 2 ( z) N n 0 [ ]z 7 6

7 Properties of the z-transform The Time-Delay Property and xn [ ] z z Xz ( ) z xn [ n 0 ] z n 0 Xz ( ) (7.0) (7.) proof: Consider then Let so Similarly Xz ( ) α 0 + α z + + α z N N N xn [ ] α k δ[ n k] k 0 α 0 δ[ n] + α δ[ n ] + + α N δ[ n N] Yz ( ) z Xz ( ) α 0 z + α z α N z N yn [ ] α 0 δ[ n ] + α δ[ n 2] + + α N δ[ n N ] xn [ ] Yz ( ) z n 0 Xz ( ) yn [ ] xn [ n 0 ] 7 7

8 A General z-transform Formula having support inter- We have seen that for a sequence val 0 n N the z-transform is The z-transform as an Operator (7.2) This definition extends for doubly infinite sequences having support interval n to (7.3) There will be discussion of this case in next Lecture when we deal with infinite impulse response (IIR) filters The z-transform as an Operator The z-transform can be considered as an operator. Unit-Delay Operator N Xz ( ) xn [ ]z n n 0 xn [ ] Xz ( ) xn [ ]z n n xn [ ] xn [ ] Unit Delay z yn [ ] xn [ ] 7 8

9 The z-transform as an Operator In the case of the unit delay, we observe that yn [ ] z { xn [ ]} xn [ ] unit delay operator which is motivated by the fact that Yz ( ) z Xz ( ) Similarly, the filter yn [ ] xn [ ] xn [ ] can be viewed as the operator yn [ ] ( z ){ xn [ ]} xn [ ] xn [ ] since Yz ( ) Xz ( ) z Xz ( ) ( z )Xz ( ) (7.4) Example: Two-Tap FIR Using the operator convention, we can write by inspection that Yz ( ) b 0 Xz ( ) + b z Xz ( ) yn [ ] b 0 xn [ ] + b xn [ ] 7 9

10 Convolution and the z-transform Convolution and the z-transform The impulse response of the unity delay system is hn [ ] δ[ n ] and the system output written in terms of a convolution is yn [ ] xn [ ]*δ[ n ] xn [ ] The system function (z-transform of hn [ ]) is Hz ( ) z and by the previous unit delay analysis, Yz ( ) z Xz ( ) We observe that Yz ( ) Hz ( )Xz ( ) (7.5) proof: yn [ ] xn [ ]*h[ n] hk [ ]xn [ k] k 0 (7.6) We now take the z-transform of both sides of (7.6) using superposition and the general delay property Yz ( ) hk [ ]( z k Xz ( )) k 0 k 0 hk [ ]z k Xz ( ) Hz ( )Xz ( ) (7.7) 7 0

11 Convolution and the z-transform Note: For the case of xn [ ] a finite duration sequence, Xz ( ) is a polynomial, and Hz ( )Xz ( ) is a product of polynomials in z Example: Convolving Finite Duration Sequences Suppose that We wish to find yn [ ] by first finding Yz ( ) We begin by z-transforming each of the sequences We find We find Yz ( ) xn [ ] 2δ[ n] 3δ[ n 2] + 4δ[ n 3] hn [ ] δ[ n] + 2δ[ n ] + δ[ n 2] Yz ( ) Convolve directly? Xz ( ) 2 3z 2 + 4z 3 z 2 Hz ( ) + 2z + by direct multiplication Yz ( ) ( 2 3z 2 + 4z 3 )( + 2z + z 2 ) yn [ ] 2 + 4z z 2 2z 3 + 5z 4 + 4z 5 using the delay property on each of the terms of yn [ ] 2δ[ n] + 4δ[ n ] δ[ n 2] 2δ[ n 3] + 5δ[ n 4] + 4δ[ n 5] 7

12 Convolution and the z-transform This section has established the very important result that polynomial multiplication can be used to replace sequence convolution, when we work in the z-domain, i.e., z-transform Convolution Theorem yn [ ] hn [ ]*x[ n] Hz ( )Xz ( ) z Yz ( ) Cascading Systems We have seen cascading of systems in the time-domain and the frequency domain, we now consider the z-domain We know from the convolution theorem that Wz ( ) H ( z)xz ( ) It also follows that Yz ( ) H 2 ( z)wz ( ) so by substitution Yz ( ) [ H 2 ( z)h ( z) ]Xz ( ) [ H ( z)h 2 ( z) ]Xz ( ) (7.8) 7 2

13 Convolution and the z-transform In summary, when we cascade two LTI systems, we arrive at the cascade impulse response as a cascade of impulse responses in the time-domain and a product of the z-transforms in the z-domain hn [ ] h [ n]*h 2 [ n] H ( z)h 2 ( z) z Hz ( ) Factoring z-polynomials ultiplying z-transforms creates a cascade system, so factoring must create subsystems Example: Hz ( ) + 3z 2z 2 + z 3 Since Hz ( ) is a third-order polynomial, we should be able to factor it into a first degree and second degree polynomial We can use the ATLAB function roots() to assist us >> p roots([ 3-2 ]) p i i >> conv([ -p(2)],[ -p(3)]) ans i With one real root, the logical factoring is to create two polynomials as follows 7 3

14 Convolution and the z-transform H ( z) z H 2 ( z) ( ( j0.42)z ) The cascade system is thus: ( ( j0.42)z ) z z 2 xn [ ] z wn [ ] z z 2 yn [ ] Xz ( ) Wz ( ) Yz ( ) H ( z) H 2 ( z) As a check we can multiply the polynomials >> conv([ -p()],conv([ -p(2)],[ -p(3)])) ans.0000, , i, i The difference equations for each subsystem are wn [ ] xn [ ] x[ n ] yn [ ] wn [ ] w[ n ] w[ n 2] Deconvolution/Inverse Filtering In a two subsystems cascade can the second system undo the action of the first subsystem? For the output to equal the input we need Hz ( ) We thus desire H ( z)h 2 ( z) or H 2 ( z) H ( z) 7 4

15 Convolution and the z-transform Example: H ( z) az, a < The inverse filter is H 2 ( z) This is no longer an FIR filter, it is an infinite impulse response (IIR) filter, which is the topic of next Chapter We can approximate az H ( z) H 2 ( z) az as an FIR filter via long division + az + a 2 z 2 + az az az a 2 z 2 a 2 z 2 a 2 z 2 a 3 z 3 a 3 z 3 An + term approximation is H 2 ( z) a k z k k 0 7 5

16 Relationship Between the z-domain and the Frequency Domain Relationship Between the z-domain and the Frequency Domain He jωˆ ωˆ - Domain ( ) b k e jωˆ k k 0 versus z - Domain Hz ( ) b k z k k 0 Comparing the above we see that the connection is setting z e jωˆ in Hz ( ), i.e., He jωˆ ( ) Hz ( ) z e jωˆ (7.9) The z-plane and the Unit Circle If we consider the z-plane, we see that evaluating Hz ( ) on the unit circle z-plane Im z ωˆ j π -- 2 z He jωˆ ( ) e jωˆ corresponds to ωˆ ± π z unit circle ωˆ Re ωˆ 0 z z ωˆ j π -- 2 ECE 260 Signals and Systems 7 6

17 Relationship Between the z-domain and the Frequency Domain From this interpretation we also can see why He ( jωˆ ) is periodic with period 2π As ωˆ increases it continues to sweep around the unit circle over and over again The Zeros and Poles of H(z) Consider Hz ( ) + b z + b 2 z 2 + b 3 z 3 (7.20) where we have assumed that Factoring Hz ( ) results in b 0 Hz ( ) ( z z )( z 2 z )( z 3 z ) (7.2) ultiplying by z 3 z 3 allows to write Hz ( ) in terms of positive powers of z Hz ( ) z 3 + b z 2 + b 2 z + b 3 z (7.22) The zeros are the locations where Hz ( ) 0, i.e., z, z 2, z 3 The poles are where Hz ( ), i.e., z 0 Note that the poles and zeros only determine constant; recall the example on page 7-5 z 3 ( z z )( z z 2 )( z z 3 ) z 3 Hz ( ) to within a 7 7

18 Relationship Between the z-domain and the Frequency Domain A pole-zero plot displays the pole and zero locations in the z- plane z-plane Im z 2 Three poles at z 0 z 3 Re z 3 Example: Hz ( ) + 2z + 2z 2 + ATLAB has a function that supports the creation of a polezero plot given the system function coefficients >> zplane([ 2 2 ],) z 3 Imaginary Part Real Part 3 7 8

19 Relationship Between the z-domain and the Frequency Domain The Significance of the Zeros of H(z) The difference equation is the actual time domain means for calculating the filter output for a given filter input The difference equation coefficients are the polynomial coefficients in Hz ( ) n For xn [ ] z 0 we know that n yn [ ] Hz ( 0 )z 0, (7.23) so in particular if z 0 is one of the zeros of Hz ( ), Hz ( 0 ) 0 and the output yn [ ] 0 If a zero lies on the unit circle then the output will be zero for a sinusoidal input of the form n xn [ ] z 0 ( e jωˆ 0 n ) e jωˆ 0n (7.24) where ωˆ 0 is the angle of the zero relative to the real axis, which is also the frequency of the corresponding complex sinusoid; why? yn [ ] Hz ( ) jωˆ e 0n z e jωˆ 0 0 (7.25) Nulling Filters The special case of zeros on the unit circle allows a filter to null/block/annihilate complex sinusoids that enter the filter at frequencies corresponding to the angles the zeros make with respect to the real axis in the z-plane 7 9

20 Relationship Between the z-domain and the Frequency Domain The nulling property extends to real sinusoids since they are composed of two complex sinusoids at ±ωˆ 0, and zeros not on the real axis will always occur in conjugate pairs if the filter coefficients are real This nulling/annihilating property is useful in rejecting unwanted jamming and interference signals in communications and radar applications Example: Hz ( ) 2cos( ωˆ 0 )z + z 2, xn [ ] cos( ωˆ 0n) Factoring Hz ( ) we find that Expanding Hz ( ) e jωˆ 0 z e jωˆ 0 xn [ ] we see that xn [ ] The nulling action of Hz ( ) at ± ωˆ 0 will remove the signal from the filter output We can set up a simple simulation in ATLAB to verify this >> n 0:00; >> w0 pi/4; >> x cos(w0*n); >> y filter([ -2*cos(w0) ],,x); >> stem(n,x,'filled') >> hold Current plot held >> stem(n,y,'filled','r') >> axis([ ]); grid z z 2 --e jωˆ 0n e jωˆ 0n 2 z 7 20

21 Relationship Between the z-domain and the Frequency Domain 0.8 Input (blue) ωˆ 0 π Amplitude of x[n] and y[n] Output (red) Time Index n Since the input is applied at n 0, we see a small transient while the filter settles to the final output, which in this case is zero >> zplane([ -2*cos(w0) ],)% check the pole-zero plot 0.8 Imaginary Part ωˆ 0 π Real Part 7 2

22 Relationship Between the z-domain and the Frequency Domain Graphical Relation Between z and When we make the substitution z in Hz ( ) we know that we are evaluating the z-transform on the unit circle and thus obtain the frequency response If we plot say Hz ( ) over the entire z-plane we can visualize how cutting out the response on just the unit circle, gives us the frequency response magnitude ωˆ e jωˆ Example: L 9 Here we have oving Average Filter (9 taps/8th-order) Hz ( ) 9 -- z k 9 k ( e j2πk 9 z ) 9 k Im 8 Poles at z 0 create the tree trunk H e jωˆ z-plane agnitude Surface Re 7 22

23 Relationship Between the z-domain and the Frequency Domain >> zplane([ones(,9)]/9,) 9-Tap oving Average FIlter Imaginary Part Real Part >> w -pi:(pi/500):pi; >> H freqz([ones(,9)/9],,w); H(e jω ) H(e jω ) hat(ω) 7 23

24 Useful Filters Useful Filters The L-Point oving Average Filter The L-point moving average (running sum) filter has yn [ ] and system function (z-transform of the impulse response) Hz ( ) (7.26) (7.27) The sum in (7.27) can be simplified using the geometric series sum formula Hz ( ) Notice that the zeros of the equation L -- xn [ k] L k 0 L -- z k L k 0 L -- z k -- z L L L z k 0 Hz ( ) z L 0 z L (7.28) are determined by the roots of (7.29) The roots of this equation can be found by noting that e j2πk for k any integer, thus the roots of (7.29) (zeros of (7.28)) are z k e j2πk L These roots are referred to as the L roots of unity -- L z L ( z ) z L, k 02,,,, L (7.30) ECE 260 Signals and Systems 7 24

25 Useful Filters One of the zeros sits at z, but there is also a pole at z, so there is a pole-zero cancellation, meaning that the pole-zero plot of Hz ( ) corresponds to the L-roots of unity, less the root at z 0 z-plane 2π L L- Pole-zero cancellation occurs here L 8 shown We have seen the frequency response of this filter before The first null occurs at frequency H e jωˆ ωˆ 0 2π L π L 4π L π ωˆ ECE 260 Signals and Systems 7 25

26 Practical Filter Design A Complex Bandpass Filter see text A Bandpass Filter with Real Coefficients see text Practical Filter Design Here we will use fdatool from the ATLAB signal processing toolbox to design an FIR filter Properties of Linear-Phase Filters A class of FIR filters having symmetrical coefficients, i.e., b k b k for k 0,,, has the property of linear phase The Linear Phase Condition For a filter with symmetrical coefficients we can show that ( ) is of the form He jωˆ He ( jωˆ ) Re ( jωˆ )e jω 2 (7.3) where Re ( jωˆ ) is a real function The fact that Re ( jωˆ ) is real means that the phase of He ( jωˆ ) is a linear function of frequency plus the possibility of ±π phase jumps whenever Re ( jωˆ ) passes through zero 7 26

27 Properties of Linear-Phase Filters Example: Hz ( ) b 0 + b z + b 2 z 2 + b z 3 + b 0 z 4 By factoring out z 2 we can write Hz ( ) [ b 0 ( z 2 + z 2 ) + b ( z + z ) + b 2 ]z 2 We now move to the frequency response by letting z ( ) [ 2b 0 cos( 2ωˆ ) + 2b cos( ωˆ ) + b 2 ]e jωˆ 4 2 He jωˆ e jωˆ Note that here we have 4, so we see that the linear phase term is indeed of the form e jωˆ 2 and the real function Re ( jωˆ ) is of the form Re jωˆ ( ) b 2 + 2[ b 0 cos( 2ωˆ ) + b cos( ωˆ )] Locations of the Zeros of FIR Linear-Phase Systems Further study of Hz ( ) for the case of symmetric coefficients reveals that H( z) z Hz ( ) (7.32) A consequence of this condition is that for Hz ( ) having a zero at z 0 it will also have a zero at z 0 Assuming the filter has real coefficients, complex zeros occur in conjugate pairs, so the even symmetry condition further implies that the zeros occur as quadruplets z 0 z*, 0, ----, z * z

28 Properties of Linear-Phase Filters z-plane z ---- * 0 z 0 z 0 * Quadruplet Zeros for Linear Phase FIlters ---- z 0 Example: Hz ( ) 2z + 4z 2 2z 3 + z 4 >> zplane([ ],).5 Imaginary Part Real Part 7 28