ELEG–305: Digital Signal Processing Lecture 9: Inverse Systems and Deconvolution; Ideal Sampling and Reconstruction Kenneth E. Barner Department of Electrical and Computer Engineering University of Delaware
Fall 2008
K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Fall 2008
1 / 22
Outline 1
Review of Previous Lecture
2
Lecture Objectives
3
Inverse Systems and Deconvolution Inverse Systems Recursive Solution Min/Max/Mixed-Phase Systems Decomposition of Nonminimum–Phase Pole–Zero Systems
4
Sampling and Reconstruction of Signals Discrete–Time Processing of Continuous–Time Signals
K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Fall 2008
2 / 22
Review of Previous Lecture
Review of Previous Lecture Filter Design Through Pole–Zero Placement – Poles increase response and zeroes decrease response (magnitude) Lowpass, Highpass, and Bandpass Filters – Pass signal content at the appropriate frequencies; Lowpass-to-Highpass filter transformation Digital Resonators – Two–pole bandpass filters with a pair of complex–conjugate poles near the unit circle Notch filters – Systems with one or more nulls in their frequency response Comb Filters – Notch filter with nulls spaced periodically across the spectrum All–Pass Filter – Constant magnitude response systems
K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Fall 2008
3 / 22
Lecture Objectives
Lecture Objectives
Objective Determine the inverse of LTI Systems with closed form expressions; Derive recursive inverse solutions for systems without closed forms; Define Min/Max/Mixed–Phase systems; Decompose nonminimum phase systems into minimum phase and all–pass components; Review ideal D/A and A/D systems Reading Chapters 5 and 6 (5.6-6.2); Next lecture, Sampling, Quantization, and Reconstruction (Chapter 6)
K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Fall 2008
4 / 22
Inverse Systems and Deconvolution
Inverse Systems
Objective I: Develop equalizers, or inverse systems, that invert communication channel distortions. Similarly, deconvolution inverts the convolution affects of a systems. Objective II: Determine the characteristics of an unknown system through system identification Identity system
x(n)
y(n)
T
T-1
Direct system
w(n) = x(n)
Inverse system
The cascade of a system and its inverse is the identity system w(n) = T −1 [y(n)] = T −1 [T [x(n)]] = x(n) Note: Not all systems are invertible, e.g., y(n) = ax 2 (n) K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Inverse Systems and Deconvolution
Fall 2008
5 / 22
Inverse Systems
Denote the impulse response of the inverse system as h I (n) w(n) = hI (n) ∗ h(n) ∗ x(n) = x(n) =⇒ h(n) ∗ hI (n) = δ(n) =⇒ H(z)HI (z) = 1 =⇒ HI (z) =
1 H(z)
If H(z) has a rational system function H(z) =
B(z) A(z)
=⇒
HI (z) =
A(z) B(z)
Result: The zeroes of H(z) became the poles of the inverse system, and vice versa Question: Does this raise any stability constraints? K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Fall 2008
6 / 22
Inverse Systems and Deconvolution
Inverse Systems
Example Determine the inverse of the system with impulse response h(n) = δ(n) − 0.5δ(n − 1) H(z) = 1 − 0.5z −1 , ROC: |z| > 0 1 z ⇒ HI (z) = = z − 0.5 1 − 0.5z −1 Two possible inverse systems: ROC: |z| > 0.5, causal hI (n) = (0.5)n u(n) ROC: |z| < 0.5, anticausal hI (n) = −(0.5)n u(−n − 1) K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Inverse Systems and Deconvolution
Fall 2008
7 / 22
Recursive Solution
Problem: Suppose no closed form of H(z) exists Solution: Use recursion and assume h I (n) is causal n
h(k)hI (n − k) = δ(n)
k =0
1 h(0) n h(k)hI (n − k) = 0 [for n ≥ 1] =⇒ h(0)hI (n) + [let n = 0] =⇒ hI (0) =
k =1
n =⇒ hI (n) = −
k =1 h(k)hI (n
− k)
h(0)
Note: Recursion uses known samples, i.e., h(n) and h I (m) for m < n Method Problems: h(0) = 0 causes problems – solved by delaying to first nonzero sample; Bigger problem – roundoff errors accumulate K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Fall 2008
8 / 22
Inverse Systems and Deconvolution
Min/Max/Mixed-Phase Systems
Consider two systems 1 1 H1 (z) = 1 + z −1 and H2 (z) = + z −1 2 2 Since H2 (z) = z −1 H1 (z −1 ), the systems have reciprocal zeroes and 5 + cos ω |H1 (ω)| = |H2 (ω)| = 4 The phase responses, however, are unique Θ1 (ω) = −ω + tan−1
1 2
sin ω + cos ω
K. E. Barner (Univ. of Delaware)
and Θ2 (ω) = −ω + tan−1
ELEG–305: Digital Signal Processing
Inverse Systems and Deconvolution
sin ω 2 + cos ω
Fall 2008
9 / 22
Min/Max/Mixed-Phase Systems
Observations on H1 (z): H1 (z) = 1 + 12 z −1 has a pole inside the unit circle (z = − 12 ) H1 (z) has minimum phase change, i.e., Θ 1 (π) − Θ1 (0) = 0 Generalization: when all system zeroes are inside the unit circle, the net zero contribution to phase is 0 radians – minimum phase systems Observations on H2 (z): H2 (z) =
1 2
+ z −1 has a pole outside the unit circle (z = −2)
H2 (z) has maximum phase change, i.e., Θ 2 (π) − Θ2 (0) = π Generalization: when all system zeroes are outside the unit circle, the net zero contribution to phase is Mπ radians (M = # of zeroes) – maximum phase systems
K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Fall 2008
10 / 22
Inverse Systems and Deconvolution
Min/Max/Mixed-Phase Systems
Min/Max/Mixed-Phase Systems Definition (Minimum–Phase System) H(z) is minimum–phase if all its poles and zeros of are inside the unit circle. Definition (Maximum–Phase System) H(z) is maximum–phase if all its poles are inside the unit circle and all its zeros are outside the unit circle. Definition (Mixed–Phase System) H(z) is mixed–phase if all poles of H(z) are inside the unit circle and some, but not all, of its zeros are outside the unit circle. Note: Pole restriction inside the unit circle is required for stability of causal systems K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Inverse Systems and Deconvolution
Fall 2008
11 / 22
Decomposition of Nonminimum–Phase Pole–Zero Systems
Recall H(z) =
B(z) A(z)
=⇒
H −1 (z) =
A(z) B(z)
Observations: H(z) minimum–phase =⇒ H −1 (z) stable and minimum–phase H(z) maximum or mixed phase =⇒ H −1 (z) unstable Theorem (Decomposition of Nonminimum–Phase Pole–Zero Systems) Any nonminimum–phase pole-zero system can be expressed as H(z) = Hmin (z)Hap (z) where Hmin (z) is a minimum-phase system and Hap (z) is an all-pass system. Note: Magnitude response of H(z) completely captured by H min (z) K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Fall 2008
12 / 22
Inverse Systems and Deconvolution
Decomposition of Nonminimum–Phase Pole–Zero Systems
To prove, consider a causal and stable system H(z) =
B(z) A(z)
Factor B(z) into minimum & maximum phase components B(z) = B1 (z)
B2 (z)
min. phase max. phase
⇒ B2 (z −1 ) is min.–phase (zeroes reflected inside unit circle). Define Hmin (z) =
B1 (z)B2 (z −1 ) A(z)
⇒ H(z) =
B1 (z)B2 (z −1 ) A(z)
and Hap (z) =
B2 (z) B2 (z −1 )
B2 (z) B2 (z −1 )
= Hmin (z)Hap (z)
Note: Hap (z) is a stable, maximum–phase, all–pass system K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Inverse Systems and Deconvolution
Fall 2008
13 / 22
Decomposition of Nonminimum–Phase Pole–Zero Systems
Example Let H(z) =
1 − 3.5z −1 + 6z −2 + 4z −3 1 − √1 z −1 + 14 z −2 2
Plot the P/Z diagram, determine phase characteristics, plot magnitude and phase, and express as H(z) = H min (z)Hap (z) Use MATLAB. Thus » B=[1,-3.5,6,4] » A=[1,-1/sqrt(2),0.25] » abs(roots(A)) ans = 0.5000 0.5000
K. E. Barner (Univ. of Delaware)
» angle(roots(A))/pi ans = 0.2500 -0.2500 Thus p1,2 = 12 e±jπ/4
ELEG–305: Digital Signal Processing
Fall 2008
14 / 22
Inverse Systems and Deconvolution
Decomposition of Nonminimum–Phase Pole–Zero Systems
» abs(roots(B)) » angle(roots(B))/pi ans = ans = 2.8284 0.2500 2.8284 -0.2500 0.5000 1.0000 √ ±jπ/4 and z3 = − 12 Thus z1,2 = 2 2e » zplane(B,A) 2 1.5
Imaginary Part
1 0.5 0 −0.5 −1 −1.5 −2 −2
−1
0
1
2
3
Real Part
K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Inverse Systems and Deconvolution
H(z) =
Fall 2008
15 / 22
Decomposition of Nonminimum–Phase Pole–Zero Systems
1 − 3.5z −1 + 6z −2 + 4z −3 1 − √1 z −1 + 14 z −2 2
=
(1 − z1 z −1 )(1 − z2 z −1 )(1 − z3 z −1 ) B(z) = A(z) (1 − p1 z −1 )(1 − p2 z −1 )
√ where z1,2 = 2 2e ±jπ/4 and z3 = − 12 and p1,2 = 12 e±jπ/4 . Let B(z) = B1 (z)B2 (z), where B1 (z) = (1 − z3 z −1 )
and B2 (z) = (1 − z1 z −1 )(1 − z2 z −1 )
min. phase
max. phase
Then H(z) = Hmin (z)Hap (z), where Hmin (z) = = min = =⇒ z1,2
1 √ e±jπ/4 2 2
K. E. Barner (Univ. of Delaware)
B1 (z)B2 (z −1 ) A(z) (1 − z3 z −1 )(1 − z1 z)(1 − z2 z) (1 − p1 z −1 )(1 − p2 z −1 )
min = 1 e ±jπ/4 and z3min = − 12 and p1,2 2 ELEG–305: Digital Signal Processing
Fall 2008
16 / 22
Inverse Systems and Deconvolution
Decomposition of Nonminimum–Phase Pole–Zero Systems
Hmin (z) P/Z Diagram 1
» Zmin=[sqrt(2)/4*exp(j*pi/4), sqrt(2)/4*exp(-j*pi/4), -0.5] » Bmin=poly(Zmin); » zplane(Bmin,A) Converts roots to polynomial coefficients and plots P/Z diagram. Next Hap (z) =
0.8 0.6
Imaginary Part
0.4 0.2 0 −0.2 −0.4 −0.6 −0.8 −1 −1
−0.5
0 Real Part
0.5
1
Hap (z) P/Z Diagram
B2 (z) B2 (z −1 )
2 1.5 1 Imaginary Part
» temp=roots(B); » ZB2=temp(1:2); »zplane(poly(ZB2),poly(1./ZB2))
0.5 0 −0.5 −1 −1.5 −2 −2
−1
0
1
2
3
Real Part
K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Inverse Systems and Deconvolution
Plot magnitude & phase » W=-pi:pi/100:pi; » H=freqz(B,A,W); » plot(W/pi,abs(H)) » plot(W/pi,angle(H)) » H=freqz(Bmin,A,W); » plot(W/pi,abs(H)) » plot(W/pi,angle(H)) » H=freqz(poly(ZB2), poly(1./ZB2),W); » plot(W/pi,abs(H)) » plot(W/pi,angle(H)) Top to bottom, H, H min , Hap
Fall 2008
17 / 22
Decomposition of Nonminimum–Phase Pole–Zero Systems
|H(ω)|
Θ(ω)
16
4
14
3 2
12
1 10 0 8 −1 6
−2
4
2 −1
−3
−0.5
0
0.5
1
−4 −1
2
0.8
1.8
0.6
1.6
0
0.5
1
−0.5
0
0.5
1
−0.5
0
0.5
1
0.4
1.4
0.2
1.2
0
1
−0.2
0.8
−0.4
0.6
−0.6
0.4 −1
−0.5
−0.5
0
0.5
1
8
−0.8 −1
4 3
8
2 1
8
0 −1
8
−2 −3
8 −1
K. E. Barner (Univ. of Delaware)
−0.5
0
ELEG–305: Digital Signal Processing
0.5
1
−4 −1
Fall 2008
18 / 22
Sampling and Reconstruction of Signals
Discrete–Time Processing of Continuous–Time Signals
Discrete-Time Processing of Continuous-Time Signals
Questions: How do we design the prefilter, A/D converter, D/A converter? Approach: Ideal cases are reviewed first, followed by the introduction & analysis of practical approaches
K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Sampling and Reconstruction of Signals
Fall 2008
19 / 22
Discrete–Time Processing of Continuous–Time Signals
Ideal A/D Converter Case
x(n) = xa (t)|t=nT = xa (nT ) ∞ 1 Xa (F − kFs ) X (F ) = T
[Time domain] [Frequency domain]
k =−∞
where Fs = 1/T , the value of which should be chosen to satisfy the Sampling Theorem K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Fall 2008
20 / 22
Sampling and Reconstruction of Signals
Discrete–Time Processing of Continuous–Time Signals
Ideal D/A Converter Case
ya (t) =
∞
y(n)ga (t − nT )
[Time domain]
n=−∞
Ya (F ) = Ga (F )Y (F )
[Frequency domain]
For perfect reconstruction the Ideal Interpolator is used sin(πt/T ) F T , |F | ≤ Fs /2 ←→ Ga (F ) = ga (t) = 0, otherwise πt/T K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Fall 2008
21 / 22
Lecture Summary
Lecture Summary LTI System Inverse – closed form case H(z) =
B(z) A(z)
=⇒
HI (z) =
A(z) B(z)
Note P/Z location restrictions for stability Recursive inverse solution for systems without closed forms 1 h(0) n h(k)hI (n − k) [for n ≥ 1] =⇒ hI (n) = − k =1 h(0) [let n = 0] =⇒ hI (0) =
Min/Max/Mixed–Phase systems; H(z) = Hmin (z)Hap (z) decomposition; Ideal D/A and A/D systems Next Lecture – Sampling, Quantization, and Reconstruction (Chapter 6) K. E. Barner (Univ. of Delaware)
ELEG–305: Digital Signal Processing
Fall 2008
22 / 22