Fall 2006 Digital Signal Processing S.F.

Hsieh

V: Transform Analysis of LTI Systems

[1] Frequency response of LTI systems

1. Definition:

H(e ) ≡
jω
h[n]e−jωn
n

is the complex gain(eigenvalue) corresponding to a cisoid eigenfunction ejωn input.

2. Since any input x[n] can be written as a linear combination of cisoids with X(ejω )
being its projection onto the orthonormal eigen-basis, the input and output pro-
jections(transforms) are related by
Y (ejω ) = H(ejω ) · X(ejω )
or
|Y (ejω )| = |H(ejω )| · |X(ejω )|, gain
∠Y (ejω ) = ∠H(ejω ) + ∠X(ejω ), phase shift
|H(ejω )| is the amplitude response and ∠H(ejω ) the phase response.
V-1
3. Ideal frequency-selective filters:
 −jωn
e d , |ω| ∈ [ω , ω ]
jω l h
H(e ) =
0, otherwise
4. Group delay:
 
d   
τ (ω) = grd H(e ) = −
jω jω
arg H(e )
dω
[Pf]
(a) Consider a narrowband bandpass signal x[n] = s[n] cos(ω0n).
1
X(e ) = S(e ) ∗ [δ(ω − ω0) + δ(ω + ω0)]
jω jω
2
1
= [S(ej(ω−ω0)) + S(ej(ω+ω0))]
2
(b) The frequency response can be approximated by a piecewise linear function
around the center frequency ω0:

|H(ejω0 )| · e−j(φ0+ωnd), ω ∈ [ω0 − , ω0 + ]
H(e ) ≈
jω
|H(ejω0 )| · e−j(−φ0+ωnd), ω ∈ [−ω0 − , −ω0 + ]

V-2
(c) The output narrowband signal becomes
Y (ejω ) = H(ejω ) · X(ejω ), if |H(ejω0 )| = 1
1 1
= S(ej(ω−ω0)) · e−jωnd · e−jφ0 + S(ej(ω+ω0)) · e−jωnd · e+jφ0
2  2 
1 1
y[n] = δ[n − nd] ∗ DT F T −1 S(ej(ω−ω0)) · e−jφ0 + S(ej(ω+ω0)) · e+jφ0
 2 2 
1 1
= δ[n − nd] ∗ s[n] · ejω0n · e−jφ0 + s[n] · e−jω0n · e+jφ0
2 2
= δ[n − nd] ∗ {s[n] · cos(ω0n − φ0)}
= s[n − nd] · cos[ω0(n − nd) − φ0]

V-3
5. (Ex 5.1) when “Mi Do Re” becomes “(silence) Re Do”

V-3
6. early(≤ 0) and late(≥ 0) group delays
4

i/p:cos(0.945*pi*n)+cos(0.955*pi*n)
2

−2 early group delay

y=conv(x,[1 0.95]);

−4

late group delay

−6 y2=conv(x,[1 1/0.9]);

−8

−10
0 50 100 150 200 250 300

V-4
7. Images are phase-sensitive!

V-5
V-6
[2] Practical System functions and linear difference equations

[2.1] System function

The system (transfer) function of an LTI system is the Z-transform of its impulse
response H(z) = Z{h[n]}. For an arbitrary input sequence x[n], its output is
y[n] = h[n] ∗ x[n], Y (z) = H(z)X(z).
1. ROC: must be a ring bounded by, but not including, poles.
2. Causality: ROC must be outside the outermost pole
3. Stability: ROC must include the unit circle, since |h[n]z −n| < ∞ at |z| = 1
implies h[n] is absolutely summable.
4. If an LTI system is stable and causal, then all poles of H(z) must lie inside unit
circle in the Z-plane.

[2.2] Linear difference equations

For an LTI system which can be described by a linear constant coefficient difference
equations:
N 
M
ak y[n − k] = br x[n − r],
k=0 r=0
V-7
Take Z-transform,
(a0 + a1z −1 + · · · + aN z −N )Y (z) = (b0 + b1z −1 + · · · + bM z −M )X(z),
then the system function is
M −r M −1
Y (z) r=0 br z b0 i=1 (1 − ciz )
H(z) = = N
=( ) N
,
−k
j=1 (1 − dj z )
X(z) a0 −1
k=0 ak z
where ci’s and dj ’s are zeros and poles, respectively, of H(z).
[Ex 5.3]
1. difference eqns:
(a) causal: y[n] :=(assign) 52 y[n − 1] − y[n − 2] + x[n]
(b) stable: y[n − 1] := 25 y[n] + 25 y[n − 2] − 25 x[n]
(c) neither: y[n − 2] := −y[n] + 52 y[n − 1] + x[n]
2. system function: two repeated zeros at z = 0 and two poles at 1/2 and 2,
1 1 −3 4/3
H(z) = = = +
1 − 52 z −1 + z −2 (1 − 12 z −1)(1 − 2z −1) 1 − 12 z −1 1 − 2z −1
3. ROC’s:
(a) causal(outside the outermost pole): |z| > 2
V-8
 (b) stable(including the unit circle): 1 < |z| < 2
(c) neither: |z| < 1
4. impulse responses:
(a) causal: h[n] = −3( 12 )nu[n] + 43 (2)nu[n], but unstable
(b) stable: h[n] = −3( 12 )nu[n] − 43 (2)nu[−n − 1], but noncausal
(c) neither: h[n] = +3( 12 )nu[−n − 1] − 43 (2)nu[−n − 1], anticausal, unstable

[2.3] Inverse systems

Let Hi(z) be the inverse of H(z), i.e., H(z)Hi(z) = 1.

1
1. system functions: Hi(z) = H(z)

2. impulse responses: h[n] ∗ hi[n] = δ[n]

1
3. frequency responses: Hi(ejω ) = H(ejω )
Accordingly,
1
|Hi(ejω )| =
|H(ejω )|
∠Hi(ejω ) = −∠H(ejω )
   
grd Hi(e ) = −grd H(e )
jω jω

V-9
4. poles/zeros become zeros/poles: If H(z) is rational with poles dk and zeros ck ,
−1
k=1 (1 − ck z )
b0 ΠM
H(z) =
k=1 (1 − dk z )
a0 ΠN −1

then Hi(z) is also rational with zeros dk and poles ck ,

−1
a0 ΠN
k=1 (1 − d k z )
Hi(z) =
k=1(1 − ck z )
b0 ΠM −1

5. ROC’s of H(z) and Hi(z) must overlap.

1−0.5z −1 1−0.9z −1
(a) (Ex 5.4) H(z) = 1−0.9z −1
and Hi(z) = .
Two possible ROCs for H(z):
1−0.5z −1
• R1 : |z| > 0.9, causal & stable
• R2 : |z| < 0.9, anticausal & unstable
Two possible ROCs for Hi(z):
• Ri1 : |z| > 0.5, causal & stable
• Ri2 : |z| < 0.5, anticausal & unstable
Only three pairs of ROCs are possible, except for (ROC1, ROCi2) because
there is no intersection.
If R1 is chosen for H(z),(causal and stable), then its inverse Hi(z) is unique
with ROCi1, which is also causal and stable.
V - 10
 z −0.5 −1 1−0.9z −1
(b) (Ex 5.5) H(z) = 1−0.9z −1 and Hi (z) = z −1 −0.5 . Two possible ROCs for H(z):

• R1 : |z| > 0.9, causal & stable

• R2 : |z| < 0.9, anticausal & unstable
Two possible ROCs for Hi(z):
• Ri1 : |z| > 2, causal but unstable
• Ri2 : |z| < 2, anticausal but stable
Only three pairs of ROCs are possible, except for (ROC2, ROCi1) because
there is no intersection.
If R1 is chosen for H(z),(causal and stable), then its inverse Hi(z) is not
unique, with either causal but unstable ROCi1 or anticausal but stable ROCi2.
6. Conclude that
An LTI system is causal and stable and has a causal and stable inverse
if and only if both the poles and zeros of H(z) are inside the unit circle.

[2.4] Impulse response for rational system functions


M−N 
N
Ak
−r
H(z) = Br z + , partial fraction expansion
r=0
1 − dk z −1
k=1
V - 11
 
M−N 
N
h[n] = Br δ[n − r] + Ak dnku[n]
r=0 k=1
• Finite impulse response (FIR) system: if h[n] is finite in length.
• Infinite impulse response (IIR) system: if h[n] is of infinite length.

[2.5] Pole/Zero plot of H(z)

1. For a rational transfer function

Q(z) (z − q1)(z − q2) · · · (z − qm)
H(z) = =
P (z) (z − p1)(z − p2) · · · (z − pn)
• Poles: p1, p2, . . . , pn, because H(z)|z=pi = H(pi) = ∞.
• Zeros: q1, q2, . . . , qm, because H(z)|z=qj = H(qj ) = 0.
2. The no. of poles equals the no. of zeros, counting all poles/zeros at infinity
z = ∞ and the origin z = 0.
2 2z
[Ex] H(z) = 1−0.5z −1 = z−0.5 has one pole at 0.5 and one zero at 0.
3
[Ex] H(z) = z−0.5 has one pole at 0.5 and one zero at ∞.
3. Poles and Time responses: According to the location of poles of H(z), we can
have a rough picture about the behavior of its impulse response h[n]:
V - 12
(a) Single-order pole: p = α is real
A
⇔ Aα n
u[n]
1 − αz −1
(b) Single-order Complex-conjugate poles: p = αejω0 and p∗ = αe−jω0
E + F z −1 E + F z −1
−1 −2
=
1 + c1 z + c2 z (1 − αejω0 z −1)(1 − αe−jω0 z −1)
⇔ [E · (α)n cos ω0n + D(α)n sin ω0n] · u[n]
V - 13
 (c) The poles magnitude α determines the signal’s exponentially growing/decaying
rate in time. Poles close to unit circle |z| = 1, corresponds to a slow
growing(|α| > 1, outside unit circle) or decaying(|α| < 1, inside unit cir-
cle) time sequence.
(d) The angles of the complex poles, ω0, determines the oscillation frequency of
the time sequence.
(e) For a causal LTI system, all poles must be within unit circle,
|poles| < 1, to assure BIBO stability.

[2.6] Transient and steady-state response

b0
Suppose H(z) = 1−a1 z −1
.
1. Complex sinusoidal input(eigenfunction):
If x[n] = ejω0n, then y[n] = H(ejω0 )ejω0n
2. Causal(truncated) complex sinusoidal input:
If x[n] = ejω0n · u[n],
1
X(z) =
1 − ejω0 z −1

V - 14
 b0 a1 b0
a1 −ejω0 1−a1e−jω0
Y (z) = H(z)X(z) = partial fraction expansion = +
1 − a1z −1 1 − ejω0 z −1
b0 a1 b0
y[n] = · a n
1 u[n] + ejω0 n
u[n]
a1 − e jω 0 1 − a1 e−jω 0

= transient response + H(ejω0 ) · x[n]

as long as the pole falls into the unit circle |a| < 1.
3. For any N −th order IIR system with a causal sinusoidal input x[n] = ejω0nu[n],
its output can be written as


N |pk | < 1
y[n] = Ak · pnk u[n] + H(ejω0 ) · ejω0n u[n] =⇒ H(ejω0 ) · ejω0n
k=1 n→∞

V - 15
[3] Frequency response of LTI systems with rational system functions

M −1
r=1 (1 − zr z )
H(z) = K N
k=1 (1 − pk z )
−1

The frequency response:

M −jω M −jω j∠ )(1−zr e−jω )
(1 − zr e ) )K |1 − z r e |e )H(ejω )
jω
H(e ) = K r=1
= |K|e j∠ r=1
= |H(e )|e
jω j∠
,
N N )(1−pk e−jω )
k=1(1 − pk e
−jω ) |1 − p e −jω |ej∠
k=1 k
|1−z1 e−jω | |1−z2 e−jω | ··· |1−zM e−jω |
• Magnitude response: |H(e )| = jω
|K| |1−p e−jω | |1−p e−jω | ··· |1−p e−jω | , or
1 2 N


M 
N
20 log10 |H(ejω )| = 20 log10 |K| + 20 log10 |1 − zr e−jω | − 20 log10 |1 − pk e−jω |(dB)
r=1 k=1

• Phase response: ∠)H(ejω ) = ∠)K + M

r=1 ∠
)(1 − zr e−jω ) − N
k=1 ∠
)(1 − pk e−jω ),
• Group delay:
  d
grd H(e ) ≡ − ∠)H(ejω )
jω
dω

N
d M
d
−jω
= ∠)(1 − pk e ) − ∠)(1 − zr e−jω )
dω r=1
dω
k=1
V - 16
 
N
|pk |2 − {pk e−jω }  |zr |2 − {zr e−jω }
M
= −
1 + |pk | − 2 {pk e } r=1 1 + |zr |2 − 2 {zr e−jω }
2 −jω
k=1

In general, zeros close to unit circle cause dips in magnitude response, and poles near
the unit circle cause peaks in magnitude response.

[3.1] Frequency response with only one pole or zero

jθ −jω ejω − rejθ

H(e ) = 1 − re e
jω
= jω .
e −0
• Geometrical interpretation using Z-plane vectors(Text Fig. 5.9, p.260),
z − zero zero vector
H(ejω ) = = .
z − pole z=ejω pole vector
6 (z)
z = ejω
zero-vector
'$ >




 pole-vector

i
×
-
(z)
&%
unit circle

V - 17
 Ᏽm z-plane
␾3

v3

v2 v1
␪
␻
᏾e

Figure 5.9 z -plane vectors for a

first-order system function evaluated on
the unit circle, with r < 1.

From Discrete-Time Signal Processing, 2e

by Oppenheim, Schafer, and Buck
©1999-2000 Prentice Hall, Inc.
 Ᏽm z-plane

v3 v1

v1 ␾3 ␻
v3 ␾3
v2 ᏾e

Figure 5.10 z -plane vectors for a

first-order system function evaluated on
the unit circle, with θ = π , r < 1. The
pole vector v1 and the zero vector v3 are
shown for two different values of ω.

From Discrete-Time Signal Processing, 2e

by Oppenheim, Schafer, and Buck
©1999-2000 Prentice Hall, Inc.
 Ᏽm z-plane

Unit circle

␻=␲+⑀
␾3 ␻=␲–⑀
v1
v3 ␾3 ⑀
v2 ᏾e
v3 ⑀
v1

Figure 5.12 z -plane vectors for a zero

at z = −1 for two different frequencies
close to π (ω = π − ε and π + ε).

From Discrete-Time Signal Processing, 2e

by Oppenheim, Schafer, and Buck
©1999-2000 Prentice Hall, Inc.
 10

dB
–10

–20

–30
0 ␲ ␲ 3␲ 2␲
2 2
Radian frequency (␻)
(a)
2

From Discrete-Time Signal Processing, 2e by Oppenheim, Schafer, and Buck ©1999-2000 Prentice Hall, Inc.
1
Radians

–1

–2
0 ␲ ␲ 3␲ 2␲
2 2
Radian frequency (␻)
(b)
2

–2
Samples

–4

r = 0.5 –6
r = 0.7
–8
r = 0.9
r=1 –10
0 ␲ ␲ 3␲ 2␲
2 2
Radian frequency (␻)
(c)

Figure 5.11 Frequency response for a single zero, with θ = π , r = 1, 0.9, 0.7,
and 0.5. (a) Log magnitude. (b) Phase. (c) Group delay for r = 0.9, 0.7, and 0.5.
 Ᏽm z-plane
Unit circle

␻=␲+⑀

␾3 v3 v1 ␻=␲–⑀
␾3
v2 ⑀
⑀ ᏾e

v3 v1

Figure 5.14 z -plane vectors for a

single zero evaluated on the unit circle,
with θ = π , r > 1.

From Discrete-Time Signal Processing, 2e

by Oppenheim, Schafer, and Buck
©1999-2000 Prentice Hall, Inc.
 10

dB
–5

–10

–15

–20
0 ␲ ␲ 3␲ 2␲
2 2
Radian frequency (␻)
(a)

From Discrete-Time Signal Processing, 2e by Oppenheim, Schafer, and Buck ©1999-2000 Prentice Hall, Inc.
4

2
Radians

–2

–4
0 ␲ ␲ 3␲ 2␲
2 2
Radian frequency (␻)
(b)

12

10

8
Samples

r = 1/0.9
2
r = 1.25
r = 2.0 0
0 ␲ ␲ 3␲ 2␲
2 2
Radian frequency (␻)
(c)
Figure 5.13 Frequency response for a single real zero outside the unit circle,
with θ = π, r = 1/0.9, 1.25, 2. (a) Log magnitude. (b) Phase (principal value).
(c) Group delay.
 • fixed r = 0.9, θ = 0, π/2, π(Text Fig 5.8, pp 259), mag. response dips and sharp
phase change at ω = θ.
• fixed θ = π, r = 1, 0.9, 0.7, 0.5(Text Fig 5.11, pp 262), more attenuation and
sharper phase change as zero approaches unit circle.
• fixed θ = π, r = 1/0.9, 1.25, 2 outside the unit circle (Text Fig 5.13, pp 264),
more attenuation as zero approaches unit circle; negative phase slope and positive
group delay.
As to the contribution of a single pole, reverse the signs both in the log-dB magnitude
response and the phase response.
z −1 1+0.8z −1 1
[Ex] H1(z) = 1+0.9z −1 is a HPF, H 2 (z) = −1
(1−0.9z ) 2 a LPF, H 3 (z) = 1+0.81z −2
a BPF.

V - 18
[3.2] Multiple poles and zeros

Text Ex. 5.8, Fig. 5.16(a)

1. Magnitude response is raised(suppressed) by poles(zeros):
2. Higher-order IIR filters can realize frequency responses with flatter passband
and stopbands(less ripples) and sharper transition bands.

V - 19
[4] Relationship between magnitude and phase

Define C(z) = H(z)H ∗(1/z ∗ ), then the magnitude-squared response

|H(ejω )|2 = H(ejω )H ∗(ejω ) = H(z)H ∗(1/z ∗ )|z=ejω = C(z)|z=ejω .
If
M −1
r=1 (1 − zr z )
H(z) = K N
(with poles/zeros being pk , zr ),
k=1 (1 − pk z )
−1

then
∗ ∗ ∗
M
r=1(1 − zr∗z)
H (1/z ) = K N
k=1 (1 − p∗k z)
produces conjugate reciprocal poles and zeros(1/p∗k , 1/zr∗ ).
M −1 ∗
2 r=1 (1−zr z )(1−zr z)
• C(z) = |K| N −1 ∗ has conjugate reciprocal pairs of poles and zeros,
k=1 (1−pk z )(1−pk z)
respectively.
• z = rejθ −→ reflection about the unit circle −→ conjugate reciprocal 1/z ∗ =
(1/r)ejθ .
• Different H(z)’s can have the same magnitude response (within a scale factor),
so long as their conjugate reciprocal pairs of poles and zeros in C(z) are identical.
V - 20
1. (Ex 5.11, Fig 5.19)

2(1 − z −1)(1 + 0.5z −1

H1(z) =
(1 − 0.8ejπ/4z −1)(1 − 0.8e−jπ/4z −1)
(1 − z −1)(1 + 2z −1
H2(z) =
(1 − 0.8ejπ/4z −1)(1 − 0.8e−jπ/4z −1)
Both have same poles at 0.8e±π/4 but different zeros: (1, −0.5) and (1, −2).
Because their (common) system functions C1(z) = C2(z) have the same conju-
gate reciprocal pairs of poles and zeros, their magnitude responses |H1(ejω )| =
|H2(ejω )| are the same. However, their phase responses are different.

V - 21
2. (Ex 5.12, Fig 5.20) There are 3 pairs of poles and zeros, respectively, for C(z).

(a) Knowing H(z) is causal and stable, we must restrict the poles p1, p2, p3
within the unit circle.
(b) Knowing h[n] is real-valued, there are 4 possible choices of zeros for H(z),
which are (z1, z2, z3), (z1, z2, z5), (z4, z5, z3) and (z4, z5, z6).
(c) If the no. of poles/zeros are not restricted, choices for H(z) would be
unlimited, if we cascading H(z) with any no. of all-pass systems of the form:
z −1 −a∗
1−az −1
, whose pole/zero are conjugate-reciprocal and magnitude response
is unity.

V - 22
[5] Allpass filters: compensate for phase distortion

A stable system function with one conjugate reciprocal pair of (pole, zero)=(a, 1/a∗),
z −1 − a∗
Hap(z) = ,
1 − az −1

has a flat unity magnitude response:

∗ jω
jω −jω 1 − a e
Hap(e ) = e .
1 − ae−jω

A general formula for an allpass system with real-valued impulse response is

Mr Mc
z −1 − dk (z −1 − e∗k )(z −1 − ek )
Hap(z) = K ∗ −1 ,
1 − dk z −1 (1 − ek z )(1 − ek z )
−1
k=1 k=1
where dk ’s are the real poles and ek ’s are the complex poles.
1. Pole/zero appear in conjugate reciprocal pair: pole: a ⇐⇒ zero: 1/a∗.
2. Pole/zero plot (Text Fig. 5.21, p 275.)

'$o
x
o xx o
xo
&%

V - 23
3. For causal and stable allpass systems, both real and complex conjugate poles
must be within the unit circle: |dk | < 1 and |ek | < 1.
4. In all, there are M = N = 2Mc + Mr poles and zeros, respectively.
5. (Continuous)Phase: always nonpositive for 0 < ω < π.
Denote the pole as a = rjθ , then the phase response is
−jω −jθ
 
e − re −1 r sin(ω − θ)
∠)H(ejω ) = ∠) = · · · = −ω − 2 tan
1 − rejθ e−jω 1 − r cos(ω − θ)
We can show that the phase response of a causal allpass system is always nonpositive
for 0 < ω < π.
6. Group delay: always positive for stable causal allpass systems(Eq 5.95).
 
d 1 − r2
grd H(e ) ≡ − ∠)H(e ) = · · · =
jω jω
dω |1 − rejθ e−jω |2
Because r < 1(causal stable pole), the group delay due to a single pair of
pole/zero allpass system is always positive. With the property of positive group
delay, we can also prove that ∠)H(ejω ) < 0 for ω ∈ (0, π).
7. Allpass systems are useful in phase compensators, min-phase system theory, LPF
→ other filters design.
V - 24
Figure 5.18 Typical pole–zero plot for an all-pass system.

Discrete-Time Signal Processing, Third Edition Copyright ©2010, ©1999, ©1989 by Pearson Education, Inc.
Alan V. Oppenheim • Ronald W. Schafer All rights reserved.
Figure 5.19 Frequency response for all-pass filters with real poles at z = 0.9
(solid line) and z = −0.9 (dashed line).
(a) Log magnitude. (b) Phase (principal value). (c) Group delay.

Discrete-Time Signal Processing, Third Edition Copyright ©2010, ©1999, ©1989 by Pearson Education, Inc.
Alan V. Oppenheim • Ronald W. Schafer All rights reserved.
Figure 5.20 Frequency response of 2nd-order all-pass system with
poles at z = 0.9e±jπ/4. (a) Log magnitude. (b) Phase (principal value).
(c) Group delay.

Discrete-Time Signal Processing, Third Edition Copyright ©2010, ©1999, ©1989 by Pearson Education, Inc.
Alan V. Oppenheim • Ronald W. Schafer All rights reserved.
Figure 5.21 Frequency response for an all-pass system with the pole–
zero plot in Figure 5.18. (a) Log magnitude. (b) Phase (principal value).
(c) Group delay.

Discrete-Time Signal Processing, Third Edition Copyright ©2010, ©1999, ©1989 by Pearson Education, Inc.
Alan V. Oppenheim • Ronald W. Schafer All rights reserved.
[6] Minimum-phase systems

For a minimum-phase system with known magnitude response |H(e )| = C(z)|z=ejω ,
jω

then its system function H(z) is uniquely determined from the poles/zeros of C(z)
inside the unit circle.
[fact]
Any rational system function can be written as
H(z) = Hmin(z)Hap(z),
where
• Hmin(z) is the minimum-phase system containing
1. poles: all poles of H(z) inside the unit circle: p̂,
2. zeros: all zeros of H(z) inside the unit circle: ẑ, and all the conjugate-
reciprocal of zeros of H(z) outside the unit circle: z̃ = 1/z̆ ∗,
• Hap is an allpass system containing
1. zeros: all zeros of H(z) outside the unit circle: z̆,
2. poles: the reflected conjugate-reciprocal poles: p̃ = 1/z̆ ∗.

V - 25
 [Ex]
H(z) Hmin(z) Hap(z)
◦ ◦
'$ ◦
'$ ◦ '$

×(4) ◦ ×(4) ×
◦&% ×&%
◦ ◦
&%
◦ ◦

[Ex 5.14]
1 + 3z −1
H(z) =
1 + 12 z −1
   
?2 1 + 3z −1
= 1 −1 ·
1 + 2z ?1 ap
 
min  
1 −1 −1 1
3(1 + 3 z ) z +3
= ·
1 + 12 z −1 1 + 13 z −1
min ap

V - 26
 [Ex 5.15] Magnitude-response compensation of a nonminimum phase system. Con-
sider a nonminimum-phase FIR channel with zeros at 0.9e±j0.6π and 1.25e±j0.8π . We
1 ±j0.8π
can build an equalizer with poles at 0.9e±j0.6π (from the inside zeros) and 1.25 e
(conjugate-reciprocal of the outside zeros) to equalize its magnitude response. The
overall response(channel + equalizer) is simply the all-pass part of the channel, which
1 ±j0.8π
has poles at 1.25 e and zeros at 1.25e±j0.8π .

Properties of minimum phase systems

• Hmin(z) is causal and stable, its inverse 1/Hmin(z) is causal and stable, too.
• Hmin(z) has the least phase lag among all causal filters with the same magnitude
response: ∠)H(ejω ) = ∠)Hmin (ejω ) + ∠)Hap(ejω ), where ∠)Hap (ejω ) < 0 increase
the phase-lag (−∠)H(ejω )).
• Minimum group delay: − dω d
∠)H(ejω ) = − dωd
∠)Hmin (ejω ) − dω
d
∠)Hap(ejω ), where
the group for allpass system − dω
d
∠)Hap(ejω ) is always positive.
n 2 n 2
• Minimum energy delay: m=0 |h[m]| ≤ m=0 |hmin [m]| , for all n.

V - 27
1. Pole-zero plots and impulse responses

V - 28
2. Energy delay:

V - 29
[7] FIR linear phase systems

For a causal FIR system of impulse response: h[n], 0 ≤ n ≤ M , there are 4 types
of such filter to achieve linear phase, i.e., H(ejω ) = A(ejω )e−jωτ +jβ , where the
amplitude response A(ejω ) is nonnegative real, τ is the constant time delay.
1. 4 types of linear phase FIR filters:
(a) Type I: symmetric, M : even, delay = M/2
h[n] = h[M − n], 0 ≤ n ≤ M.
−jω −j2ω −j( M
2 −1)ω M −j M
jω
H(e ) = h[0] + h[1]e + h[2]e + ··· + − 1]e
h[ M2 + h[ 2 ]e 2 ω
+h[M ]e−jωM + h[M − 1]e−jω(M−1) + h[M − 2]e −jω(M−2)
+ ···
M
+h[ M2 + 1]e−j( 2 +1)ω
M
  M M
  M M

= e−jω 2 h[ M2 ] + h[0] ejω 2 + e−jω 2 + h[1] ejω( 2 −1) + e−jω( 2 −1) + · · ·
 jω −jω

+h[ 2 − 1] e + e
M
⎧ ⎫
M
⎨ 
M/2 ⎬
= e−jω 2 h[ M2 ] + 2 h[ M2 − k] cos ωk .
⎩ ⎭
k=1

V - 30
(b) Type II: symmetric, M : odd, delay= M/2
h[n] = h[M − n], 0 ≤ n ≤ M.
⎧ ⎫
⎪
⎨
M+1
2
⎪
⎬
M
H(ejω ) = e−jω 2 2 h[ M+1
2 − k] cos ω(k − 1
2 )⎪ .
⎪
⎩ k=1 ⎭

(c) Type III: antisymmetric, M : even,

h[n] = h[M − n], 0 ≤ n ≤ M, ( Note: h[ M2 ] = 0).
⎧ ⎫
⎪
⎨
M

2
⎪
⎬
M
H(ejω ) = je−jω 2 2 h[ M2 − k] sin ωk .
⎪
⎩ k=1 ⎪
⎭

(d) Type IV: antisymmetric, M : odd,

h[n] = −h[M − n], 0 ≤ n ≤ M.
⎧ ⎫
⎪
⎨
M+1
2
⎪
⎬
M
H(ejω ) = je−jω 2 2 h[ M+1
2 − k] sin ω(k − 1
2 ) .
⎪
⎩ k=1 ⎪
⎭

V - 31
2. Impulse responses of FIR linear phase systems:

V - 32
3. Zeros plot of FIR linear phase systems

V - 33
4. Properties of FIR linear phase systems:
(a) Symmetric: Type I & II: h[n] = h[M − n].

M 
M 0

H(z) = h[n]z −n = h[M − n]z −n = h[k]z k z −M = z −M H(z −1)
n=0 n=0 k=M
implies zeros of symmetric FIR linear phase systems H(z) occur in reciprocal
pairs, H(z) = H(1/z) = 0.
[Pf] If z0 is a zero, H(z0) = 0 ⇒ z0−M H(z0−1) = 0 ⇒ H(z0−1) = 0, thus z0−1
is a zero, too.
If, in addition, h[n] is real-valued, then zeros must occur in conjugate pairs,
too.
(b) antisymmetric: Type III & IV: h[n] = −h[M − n].
H(z) = −z −M H(z −1)
implies zeros of antisymmetric FIR linear phase systems H(z) occur in reciprocal
pairs, too.
(c) Zeros of a real-valued linear-phase FIR, symmetric or antisymmetric, must
occur in a quadruple of conjugate/reciprocal pairs. If z0 is a zero of H(z), so
are its conjugate, its inverse, and its conjugate inverse, i.e., {z0, z0∗, 1/z0 , 1/z0∗}
are all zeros of H(z). [Pf] H(1/z) = z M H(z).
V - 34
(d) Notes:
• For Type II, with M being odd, −1 is always a zero.
note that H(−1) = (−1)−M H(−1/1), if M is odd, H(−1) = −H(−1);
thus H(−1) = 0.
• For Type III, ±1 are always zeros.
note that H(±1) = −(±1)−M H(±1) = −H(±1), if M is even; thus
H(±1) = 0.
• For Type IV, 1 are always a zero.
note that H(1) = −(1)−M H(1) = −H(1) implies H(1) = 0.
(e) To implement different kinds of filters,
i. LPF(no zero at z = 1): Use type I or II.
ii. HPF(no zero at z = −1): Use type I or IV.
iii. BPF: Use any type.
iv. BSF(bandstop, no zero at z = ±1): use type I.

V - 35