You are on page 1of 581

# SOLUTIONS MANUAL

to accompany

Digital Signal Processing:
A Computer-Based Approach
Third Edition

Sanjit K. Mitra

Prepared by
Chowdary Adsumilli, John Berger, Marco Carli,
Hsin-Han Ho, Rajeev Gandhi, Chin Kaye Koh,
Luca Lucchese, and Mylene Queiroz de Farias

Not for sale.

1

Chapter 2
2.1

(a) x1 1 = 22.85, x1 2 = 9.1396, x1 ∞ = 4.81,
(b) x 2 1 = 18.68, x 2 2 = 7.1944, x 2 ∞ = 3.48.

2.2

1, n ≥ 0,
1, n < 0,
µ[ n] = ⎧⎨
Hence, µ[ −n − 1] = ⎧⎨
Thus, x[ n] = µ[ n] + µ[ − n − 1].
⎩0, n < 0.
⎩0, n ≥ 0.

2.3

(a) Consider the sequence defined by x[ n] = ∑ δ[ k ]. If n < 0, then k = 0 is not included

n

k = −∞

in the sum and hence, x[n] = 0 for n < 0. On the other hand, for n ≥ 0, k = 0 is included
in the sum, and as a result, x[n] =1 for n ≥ 0. Therefore,
n
1, n ≥ 0,
x[ n] = ∑ δ[ k ] = ⎧⎨
= µ[ n].
0
⎩ , n < 0,
k = −∞

2.4

2.5

1, n ≥ 0,
1, n ≥ 1,
it follows that µ[ n − 1] = ⎧⎨
Hence,
(b) Since µ[ n] = ⎧⎨
0
,
n
0
,
<

⎩0, n < 1.
1, n = 0,
µ[ n] − µ[ n − 1] = ⎧⎨
= δ[ n].
⎩0, n ≠ 0,
Recall µ[ n] − µ[ n − 1] = δ[ n]. Hence,
x[ n] = δ[ n] + 3δ[ n − 1] − 2δ[ n − 2] + 4δ[ n − 3]
= (µ[ n] − µ[ n − 1]) + 3(µ[ n − 1] − µ[ n − 2]) − 2(µ[ n − 2] − µ[ n − 3]) + 4(µ[ n − 3] − µ[ n − 4])
= µ[ n] + 2µ[ n − 1] − 5µ[ n − 2] + 6µ[ n − 3] − 4µ[ n − 4].
(a) c[ n] = x[ − n + 2] = {2 0 − 3 − 2 1 5 − 4},

(b) d[ n] = y[ − n − 3] = {− 2 7 8 0 − 1 − 3 6 0 0},

(c) e[ n] = w[ − n] = {5 − 2 0 − 1 2 2 3 0 0},

(d) u[ n] = x[ n] + y[ n − 2] = {− 4 5 1 − 2 3 − 3 1 0 8 7 − 2},

(e) v[ n] = x[ n] ⋅ w[ n + 4] = {0 15 2 − 4 3 0 − 4 0},

(f) s[ n] = y[ n] − w[ n + 4] = {− 3 4 − 5 0 0 10 2 − 2},

(g) r[ n] = 3.5 y[ n] = {21 − 10.5 − 3.5 0 2.8 24.5 − 7}.

2.6

(a) x[ n] = −4δ[ n + 3] + 5δ[ n + 2] + δ[ n + 1] − 2δ[ n] − 3δ[ n − 1] + 2δ[ n − 3],
y[ n] = 6δ[ n + 1] − 3δ[ n] − δ[ n − 1] + 8δ[ n − 3] + 7δ[ n − 4] − 2δ[ n − 5],
w[ n] = 3δ[ n − 2] + 2δ[ n − 3] + 2δ[ n − 4] − δ[ n − 5] − 2δ[ n − 7] + 5δ[ n − 8],
(b) Recall δ[ n] = µ[ n] − µ[ n − 1]. Hence,
x[ n] = −4(µ[ n + 3] − µ[ n + 2]) + 5(µ[ n + 2] − µ[ n + 1]) + (µ[ n + 1] − µ[ n])
− 2(µ[ n] − µ[ n − 1]) − 3(µ[ n − 1] − µ[ n − 2]) + 2(µ[ n − 3] − µ[ n − 4])

Not for sale.

2

= −4µ[ n + 3] + 9µ[ n + 2] − 4µ[ n + 1] − 3µ[ n] − µ[ n − 1] + 3µ[ n − 2] + 2µ[ n − 3] − 2µ[ n − 4],

2.7

(a)
x[n-1]

_
z 1

x[n]

x[n-2]

_
z 1

h[2]

h[1]

h[0]

y[n]

+

+

From the above figure it follows that y[ n] = h[ 0] x[ n] + h[1] x[ n − 1] + h[2] x[ n − 2].
(b)
x[n]

h[0]

w[n]

+
_1

+

y[n]

_1

z

z

β11

x[n _ 1]

+

_
z 1

+

_
z 1

β 21

x[n _ 2]

β12

w[n _ 1]

w[n _ 2]

β 22

From the above figure we get w[ n] = h[0]( x[ n] + β11 x[ n − 1] + β 21 x[ n − 2]) and
y[ n] = w[ n] + β12 w[ n − 1] + β 22 w[ n − 2]. Making use of the first equation in the second
we arrive at
y[ n] = h[0]( x[ n] + β11 x[ n − 1] + β 21 x[ n − 2])
+ β12 h[0]( x[ n − 1] + β11 x[ n − 2] + β 21 x[ n − 3])
+ β 22 h[0]( x[ n − 2] + β11 x[ n − 3] + β 21 x[ n − 4])
= h[0]( x[ n] + (β11 + β12 ) x[ n − 1] + (β 21 + β12 β11 + β 22 ) x[ n − 2]
+ (β12 β 21 + β 22 β11 ) x[ n − 3] + β 22 β 21 x[ n − 4]).
(c) Figure P2.1(c) is a cascade of a first-order section and a second-order section. The
input-output relation remains unchanged if the ordering of the two sections is
interchanged as shown below.
x[n]

0.6

w[n]

+
_

+
_
z 1

0.8

u[n]

+

y[n+1]
_
z 1

0.3

+

+

0.4

y[n]

w[n _ 1]
_
z 1
_ 0.5

0.2

w[n _ 2]

Not for sale.

3

The second-order section can be redrawn as shown below without changing its inputoutput relation.
x[n]

0.6

w[n]

+
_

+

+

+

y[n+1]
_
z 1

_
_
z 1
z 1
0.8
w[n _ 1]

0.3

0.4

y[n]

+

_
z 1

_
z 1
_ 0.5

u[n]

0.2

w[n _2]

The second-order section can be seen to be cascade of two sections. Interchanging their
ordering we finally arrive at the structure shown below:
0.6

+

x[n]

s[n]

_
z 1

_

0.3
x[n _ 1]

u[n]

+
0.8

_
z 1

x[n _ 2]

y[n+1]
_
z 1

_
z 1

+

+

+

0.4

u[n _ 1]

y[n]

_
z 1
_ 0.5

0.2

u[n _ 2]

Analyzing the above structure we arrive at
s[ n] = 0.6 x[ n] + 0.3 x[ n − 1] + 0.2 x[ n − 2],
u[ n] = s[ n] − 0.8u[ n − 1] − 0.5u[ n − 2],
y[ n + 1] = u[ n] + 0.4 y[ n].
From u[ n] = y[ n + 1] − 0.4 y[ n]. Substituting this in the second equation we get after some
algebra y[ n + 1] = s[ n] − 0.4 y[ n] − 0.18 y[ n − 1] + 0.8 y[ n − 2]. Making use of the first
equation in this equation we finally arrive at the desired input-output relation
y[ n] + 0.4 y[ n − 1] + 0.18 y[ n − 2] − 0.2 y[ n − 3] = 0.6 x[ n − 1] + 0.3 x[ n − 2] + 0.2 x[ n − 3].
(d) Figure P2.19(d) is a parallel connection of a first-order section and a second-order
section. The second-order section can be redrawn as a cascade of two sections as
indicated below:
x[n]

w[n]

+

+

_
_
z 1
z 1
_
0.8
w[n _ 1]

_
z 1
_ 0.5

+

y 2[n]

_
z 1

w[n _2]

Not for sale.

0.3

0.2

4

Interchanging the order of the two sections we arrive at an equivalent structure shown
below:
0.3

x[n]

_1

+

z

q[n]

y 2[n]

+

_
z 1

_

0.2

_
z 1

0.8

+

y 2 [n _ 1]
_
z 1

_ 0.5

y 2[n _ 2]

Analyzing the above structure we get
q[ n] = 0.3 x[ n − 1] + 0.2 x[ n − 2],
y2 [ n] = q[ n] − 0.8 y2 [ n − 1] − 0.5 y2 [ n − 2].
Substituting the first equation in the second we have
y2 [ n] + 0.8 y2 [ n − 1] + 0.5 y2 [ n − 2] = 0.3 x[ n − 1] + 0.2 x[ n − 2].

(2-1)

Analyzing the first-order section of Figure P2.1(d) given below
u[n] _ 1 u[n _ 1] 0.6
y 1 [n]
z
x[n]
+
0.4

we get

u[ n] = x[ n] + 0.4u[ n − 1],
y1[ n] = 0.6u[ n − 1].
Solving the above two equations we have
y1[ n] − 0.4 y1[ n − 1] = 0.6 x[ n − 1].
(2-2)
The output y[n] of the structure of Figure P2.19(d) is given by
y[ n] = y1[ n] + y2 [ n].
(2-3)
From Eq. (2-2) we get 0.8 y1[ n − 1] − 0.32 y1[ n − 2] = 0.48 x[ n − 2] and
0.5 y1[ n − 2] − 0.2 y1[ n − 3] = 0.3 x[ n − 3]. Adding the last two equations to Eq. (2-2) we
arrive at y1[ n] + 0.4 y1[ n − 1] + 0.18 y1[ n − 2] − 0.2 y1[ n − 3]
(2-4)
= 0.6 x[ n − 1] + 0.48 x[ n − 2] + 0.3 x[ n − 3].
Similarly, from Eq. (2-1) we get
− 0.4 y2 [ n − 1] − 0.32 y2 [ n − 2] − 0.2 y2 [ n − 3] = −0.12 x[ n − 2] − 0.08 x[ n − 3]. Adding this
equation to Eq. (2-1) we arrive at
y2 [ n] + 0.4 y2 [ n − 1] + 0.18 y2 [ n − 2] − 0.2 y2 [ n − 3]
(2-5)
= 0.3 x[ n − 1] + 0.08 x[ n − 2] − 0.08 x[ n − 3].
Adding Eqs. (2-4) and (2-5), and making use of Eq. (2-3) we finally arrive at the inputoutput relation of Figure P2.1(d) as:
y[ n] + 0.4 y[ n − 1] + 0.18 y[ n − 2] − 0.2 y[ n − 3] = 0.9 x[ n − 1] + 0.56 x[ n − 2] + 0.22 x[ n − 3].
Not for sale.

5

2.8

(a) x1* [ n] = {1 − j 4 − 2 − j 5 3 + j 2 − 7 − j 3 − 1 − j},

x1* [ −n] = {− 1 − j − 7 − j 3 3 + j 2 − 2 − j 5 1 − j 4}. Therefore

x1, cs [n] =

1
2

x1, ca [n] =

1
2

(x [n] + x [−n]) = { j1.5

− 4.5 + j 3 − 4.5 − j − j1.5},

*
1

1

(x1[n] − x1*[−n]) = {1 + j2.5

(b) x 2 [ n] = e

jπn / 3

. Hence, x 2* [ n] = e

(

2.5 + j 4 − j 2 − 2.5 + j 4 − 1 + j 2.5}.

− jπn / 3

)

and thus, x 2* [ −n] = e jπn / 3 = x 2 [ n].

Therefore, x2, cs [n] = x2 [n] + x2* [−n] = e j 2 πn / 3 = x2 [n], and
2
1

x2, ca [n] =

1
2

(x2 [n] − x2* [−n]) = 0.

(c) x 3 [ n] = j e − jπn / 5 . Hence, x 3* [ n] = − j e jπn / 5 and thus,

(

)

x 3* [ − n] = − j e − jπn / 5 = − x 3 [ n]. Therefore, x3, cs [n] = x3 [n] + x3*[− n] = 0, and
2
1

x3, ca [n] =
2.9

1
2

(x3[n] − x3*[−n]) = x3[n] = j e − jπn / 5 .

(a) x[ n] = {− 4 5 1 − 2 − 3 0 2}. Hence, x[ − n] = {2 0 − 3 − 2 1 5 − 4}.
Therefore, x ev [ n] =

1
( x[ n] +
2

x[ − n]) =

1
{− 2
2

5 − 2 − 4 − 2 5 − 2}

= {− 1 2.5 − 1 − 2 − 1 2.5 − 1}

and x od [ n] =

1
1
( x[ n] − x[ −n]) = {− 6
2
2

5 4 0 − 4 − 5 6}

= {− 3 2.5 2 0 − 2 − 2.5 3}.

(b) y[ n] = {0 0 0 0 6 − 3 − 1 0 8 7 − 2}. Hence,

y[ − n] = {− 2 7 8 0 − 1 − 3 6 0 0 0 0}.

Therefore, yev [ n] =
and y od [ n] =

1
( y[ n] +
2

y[ − n]) = {− 1 3.5 4 0 2.5 − 3 2.5 0 4 3.5 − 1}

1
( y[ n] − y[ −n]) = {1
2

− 3.5 − 4 0 3.5 0 − 3.5 0 4 3.5 − 1}.

(c) w[ n] = {0 0 0 0 0 0 0 0 0 0 3 2 2 − 1 0 − 2 5}. Hence,

w[ − n] = {5 − 2 0 − 1 2 2 3 0 0 0 0 0 0 0 0 0 0}. Therefore

wev [ n] =

1
(w[ n] + w[ − n])
2

Not for sale.

6

= {2.5 − 1 0 − 0.5 1 1 1.5 0 0 0 1.5 1 1 − 0.5 0 − 1 2.5} and

wod [ n] =

1
(w[ n] − w[ − n])
2

= {− 2.5 1 0 0.5 − 1 − 1 − 1.5 0 0 0 1.5 1 1 − 0.5 0 − 1 2.5}.

2.10

(a) x1[ n] = µ[ n + 2]. Hence, x1[ −n] = µ[ −n + 2]. Therefore,
n ≥ 3,
⎧⎪1 / 2,
1
x1, ev [ n] = (µ[ n + 2] + µ[ − n + 2]) = ⎨ 1, − 2 ≤ n ≤ 2, and
2
⎪⎩1 / 2,
− 3 ≤ n,
n ≥ 3,
⎧⎪ 1 / 2,
1
− 2 ≤ n ≤ 2,
x1, od [ n] = (µ[ n + 2] − µ[ − n + 2]) = ⎨ 0,
2
⎪⎩− 1 / 2,
− 3 ≤ n.
(b) x 2 [ n] = α n µ[ n − 3]. Hence, x 2 [ −n] = α − n µ[ −n − 3]. Therefore,

⎧ 1 αn,
n ≥ 3,
⎪⎪ 2
n
−n
α µ[ n − 3] + α µ[ −n − 3] = ⎨ 0,
− 2 ≤ n ≤ 2, and
1

n
⎪ α ,
− 3 ≤ n,
⎪⎩ 2
⎧ 1 αn,
n ≥ 3,
⎪⎪ 2
n
−n
α µ[ n − 3] − α µ[ −n − 3] = ⎨ 0,
− 2 ≤ n ≤ 2,
1
⎪− α − n ,
− 3 ≤ n.
⎪⎩ 2

x 2, ev [ n] =

1
2

(

)

x 2, od [ n] =

1
2

(

)

(c) x 3 [ n] = n α n µ[ n]. Hence, x 3 [ − n] = − n α − n µ[ − n]. Therefore,

(
(

)
)

1
n
1
n α n µ[ n] + ( − n) α − n µ[ − n] = n α
2
2
1
n
1
x 3, od [ n] = n α n µ[ n] − ( − n) α − n µ[ − n] = n α .
2
2

x 3, ev [ n] =

n

(d) x 4 [ n] = α . Hence, x 4 [ −n] = α
1

1

1

1

−n

n

and

= x 4 [ n]. Therefore,
n

x 4, ev [ n] = ( x 4 [ n] + x 4 [ − n]) = ( x 4 [ n] + x 4 [ n]) = x 4 [ n] = α
2
2

and

x 4, od [ n] = ( x 4 [ n] − x 4 [ −n]) = ( x 4 [ n] − x 4 [ n]) = 0.
2
2
2.11 x ev [ n] = (x[ n] + x[ − n]). Thus, x ev [ −n] = (x[ − n] + x[ n]) = x ev [ n]. Hence, x ev [n] is
2
2
1

1

an even sequence. Likewise, x od [ n] = (x[ n] − x[ − n]). Thus,
2
1

x od [ − n] =

1
2

(x[ −n] − x[ n]) = − xod [ n].

Hence, x od [n] is an odd sequence.

Not for sale.

7

2.12 (a) g[ n] = x ev [ n]x ev [ n]. Thus, g[ −n] = x ev [ − n]x ev [ − n] = x ev [ n]x ev [ n] = g[ n]. Hence,
g[n] is an even sequence.
(b) u[ n] = x ev [ n]x od [ n]. Thus, u[ − n] = x ev [ − n]x od [ − n] = x ev [ n](− x od [ n]) = −u[ n].
Hence, u[n] is an odd sequence.
(c) v[ n] = x od [ n]x od [ n]. Thus, v[ −n] = x od [ − n]x od [ − n] = (− x od [ n])(− x od [ n])
= x od [ n]x od [ n] = v[ n]. Hence, v[n] is an even sequence.
2.13 (a) Since x[n] is causal, x[ n] = 0, n < 0. Also, x[ − n] = 0, n > 0. Now,
Hence, x ev [0] = (x[0] + x[0]) = x[0] and
2
⎧2 x ev [ n], n > 0,

1
x ev [ n] = x[ n], n > 0. Combining the two equations we get x[ n] = ⎨ x ev [ n], n = 0,
2
⎪⎩ 0,
n < 0.

x ev [ n] =

1
2

(x[ n] + x[ −n]).

Likewise, x od [ n] =

x od [ n] =

1
2

1
2

1

(x[ n] − x[−n]).

Hence, x od [0] =

1
2

(x[0] − x[0]) = 0

and

2 x [ n], n > 0,
x[ n], n > 0. Combining the two equations we get x[ n] = ⎧⎨ ev
n ≤ 0.
⎩ 0,

(b) Since y[n] is causal, y[ n] = 0, n < 0. Also, y[ − n] = 0, n > 0. Let
y[ n] = yre [ n] + jyim [ n], where yre [n] and yim [n] are real causal sequences.
1

(

)

1

(

)

Now, yca [ n] = y[ n] − y ∗ [ − n] . Hence, yca [0] = y[0] − y ∗ [0] = jyim [0] and
2
2

yca [ n] =

1
y[ n], n
2

> 0. Since yre [0] is not known, y[n] cannot be fully recovered from

yca [n] .
1

(

)

1

(

)

Likewise, ycs [ n] = y[ n] + y ∗ [ − n] . Hence, ycs [0] = y[0] + y ∗ [0] = yre [0] and
2
2

ycs [ n] =

1
y[ n], n
2

> 0. Since yim [ 0] is not known, y[n] cannot be fully recovered from

ycs [n] .
2.14 Since x[n] is causal, x[ n] = 0, n < 0. From the solution of Problem 2.13 we have
⎧2 x ev [ n], n > 0, ⎧2 cos(ωo n), n > 0,

1
n = 0, = 2 cos(ωo n)µ[ n] − δ[ n].
x[ n] = ⎨ x ev [ n], n = 0, = ⎨
0
,
n < 0,
⎪⎩ 0,

n < 0, ⎩
2.15 (a) {x[ n]} = {Aα n } where A and α are complex numbers with α < 1. Since for

n < 0, α

n

can become arbitrarily large, {x[ n]} is not a bounded sequence.

Not for sale.

8

⎧ n
(b) y[ n] = Aα n µ[ n] = ⎨ Aα , n ≥ 0, where A and α are complex numbers with
n < 0,
⎩ 0,
n

α < 1. Here, α ≤ 1, n ≥ 0. Hence y[ n] ≤ A for all values of n. Hence, {y[ n]} is a
bounded sequence.
(c) {h[ n]} = Cβ n µ[ n] where C and β are complex numbers with β > 1. Since for

n > 0, β

n

can become arbitrarily large, {h[ n]} is not a bounded sequence.

(d) {g[ n]} = 4 cos(ωo n). Since g[ n] ≤ 4 for all values of n,{g[ n]} is a bounded
sequence.
1
⎧⎛
1
1
⎪⎜1 − 2 ⎞⎟, n ≥ 1,
(e) v[ n] = ⎨⎝ n ⎠
Since 2 < 1 for n > 1 and 2 = 1 for n = 1, v[ n] < 1 for
n
n
⎪⎩ 0,
n ≤ 0.
all values of n. Thus {v[ n]} is a bounded sequence.

∞ ( −1) n +1
∞ 1
(−1) n +1
µ[ n − 1]. Now ∑ x[ n] = ∑
= ∑
= ∞. Hence {x[ n]} is
2.16 x[ n] =
n
n
n = −∞
n =1
n =1 n
not absolutely summable.

2.17 (a) x1[ n] = α n µ[ n − 1]. Now

n = −∞

n =1

n =1

x 2 [ n] = ∑ α n = ∑ α

n

=

α
1− α

< ∞ , since

α < 1. Hence, {x1[ n]} is absolutely summable.

(b) x 2 [ n] = α n µ[ n − 1]. Now

α

2

n = −∞

n =1

n =1

x 2 [ n] = ∑ nα n = ∑ n α

n

α

=

(1 − α ) 2

< ∞, since

< 1. Hence, {x 2 [ n]} is absolutely summable.

(c) x 3 [ n] = n 2 α n µ[ n − 1]. Now
2

3

n = −∞

n =1

n =1

x 3 [ n] = ∑ n 2 α n = ∑ n 2 α

n

4

= α + 2 2 α + 32 α + 4 2 α + K
2

3

4

2

3

4

3

4

5

= ( α + α + α + α + K) + 3( α + α + α + K) + 5( α + α + α + K)

Not for sale.

9

4

5

6

+ 7( α + α + α + K) =

=
=

α
1− α

⎛ ∞
1
n⎞
⎜⎜ ∑ (2n − 1) α ⎟⎟ =
⎠ 1− α
⎝ n =1

1
1− α

α (1 + α )
(1 − α ) 3

+

2

1− α

+

3

+

1− α

4

1− α

+K

⎛ ∞
1
n
n⎞
⎜⎜ 2 ∑ n α − ∑ α ⎟⎟ =
n =1
⎠ 1− α
⎝ n =1

⎛ 2α
α

⎜ (1 − α ) 2 1 − α

< ∞. Hence, {x 3 [ n]} is absolutely summable.

∞ 1
∞ 1
1
1
2.18 (a) x a [ n] =
= ∑
=
= 2 < ∞. Hence,
µ[ n]. Now ∑ x a [ n] = ∑
n
n
1
n = −∞
n=0 2
n=0 2
2n
1−
2

{x a [ n]} is absolutely summable.

1
1
µ[ n]. Now ∑ x b [ n] = ∑
(n + 1)(n + 2)
n = −∞
n = 0 (n + 1)(n + 2)
∞ ⎛ 1
1 ⎞ ⎛ 1⎞ ⎛1 1⎞ ⎛1 1⎞ ⎛1 1⎞
= ∑ ⎜

⎟ = ⎜1 − ⎟ + ⎜ − ⎟ + ⎜ − ⎟ + ⎜ − ⎟ + K = 1 < ∞. Hence,
2⎠ ⎝2 3⎠ ⎝3 4⎠ ⎝ 4 5⎠
n=0 ⎝ n + 1 n + 2 ⎠ ⎝
{x b [ n]} is absolutely summable.

(b) x b [ n] =

2.19 (a) A sequence x[n] is absolutely summable if

n = −∞

x[ n] < ∞. By Schwartz inequality

⎞⎛ ∞

2 ⎛ ∞
x[ n] ≤ ⎜⎜ ∑ x[ n] ⎟⎟⎜⎜ ∑ x[ n] ⎟⎟ < ∞. Hence, an absolutely summable
n = −∞
⎝ n = −∞
⎠⎝ n = −∞

sequence is square summable and has thus finite energy.

we have

1
n

Now consider the sequence x[ n] = µ[ n − 1]. The convergence of an infinite series can
be shown via the integral test. Let a n = f (x ), where a continuous, positive and

decreasing function is for all x ≥ 1. Then the series ∑ a n and the integral ∫ f ( x )dx
n =1

1

1

1

1

both converge or both diverge. For a n = , f ( x ) = . But ∫ dx = (ln x ) 1 = ∞ − 0 = ∞ .
n
x
x
1

Hence,

n = −∞

∞ 1
n =1 n

x[ n] = ∑

1
n

does not converge. As a result, x[ n] = µ[ n − 1] is not

absolutely summable.

Not for sale.

10

(b) To show that {x[ n]} is square-summable, we observe that here a n =

1
n2

, and thus,

∞ 1
1
⎛ 1⎞
f ( x ) = 2 . Now, ∫ 2 dx = ⎜ − ⎟ = − + 1 = 1. Hence, ∑ 2 converges, or in other

x
⎝ x ⎠1
n =1 n
1x

1

1

1
n

words, x[ n] = µ[ n − 1] is square-summable.
2.20 See Problem 2.19, Part (a) solution.
2

∞ ⎛ cos ω n ⎞

cos ωc n
1
2
c
µ[ n − 1]. Now, ∑ x 2 [ n] = ∑ ⎜
2.21 x 2 [ n] =
. Since,
⎟ ≤ ∑
2
2
πn
n = −∞
n =1 ⎝ πn ⎠
n =1 π n
2
π 2 ∞ ⎛ cos ωc n ⎞
1
, ∑ ⎜
⎟ ≤ . Therefore x2 [ n] is square-summable.
∑ 2 =
6 n =1 ⎝ πn ⎠
6
n =1 n

1

Using the integral test (See Problem 2.19, Part (a) solution) we now show that x2 [ n] is

cos ωc x
∞ cos ω x
πx
1
c
⋅ cos int(ωc x )
dx = ⋅
not absolutely summable. Now, ∫
πx
π cos ωc x
1

where
1

∞ cos ω x
c

cos int is the cosine integral function. Since ∫

πx
diverges. Hence, x2 [ n] is not absolutely summable.

cos ωc n
also
πn
n =1

dx diverges, ∑

1

K
1
1
2
2
∑ x [ n] = lim
∑ x ev [ n] + x od [ n]
K → ∞ 2 K +1 n = − K
K → ∞ 2 K +1 K = −∞

(

2.22 Px = lim

(

)

)

K
1
2
2
∑ x ev [ n] + x od [ n] + 2 x ev [ n]x od [ n]
K → ∞ 2 K +1 n = − K

= lim

= Px ev + Px od +

K
1
1
lim
∑ x[ n] +
2 K → ∞ 2 K +1 n = − K

(

1

1⎛ K

x[ − n])(x[ n] − x[ −n])

⋅ ⎜⎜ ∑ x 2 [ n ] − ∑ x 2 [ − n ] ⎟⎟ = Px + Px
= Px ev + Px od + lim
ev
od
+
K
2
1
2 ⎝ n=− K
n = −∞

K →∞

as

K

K

n = −K

n = −K

2
2
∑ x [ n] = ∑ x [ − n]. Now for the given sequence,
K
K 1 6
1
1
2
∑ x od [ n] = lim
∑ ⎛⎜ ⎞⎟
K → ∞ 2 K +1 n = − K
K → ∞ 2 K +1 n = 0 ⎝ 3 ⎠

Px od = lim

6

K
1
1
= ⎛⎜ ⎞⎟ lim
∑1
⎝ 3 ⎠ K → ∞ 2 K +1 n =0

6
K +1
1 1 6
1 ⎛ 1 ⎞6
1
= ⎛⎜ ⎞⎟ . Hence, Px = Px − Px
= ⎛⎜ ⎞⎟ lim
=

10
⎜ ⎟ .
ev
od
2 ⎝3⎠
⎝ 3 ⎠ K → ∞ 2 K +1 2 ⎝ 3 ⎠

Not for sale.

11

N −1

2

N −1

2.23 x[ n] = sin( 2 πkn / N ), 0 ≤ n ≤ N − 1. Now E x = ∑ x[ n] = ∑ sin 2 (2πkn / N )
n=0

=

1 N −1
∑ 1 − cos(4πkn / N )
2 n=0

(

) = N2 − 12

n=0
N −1

N −1

∑ cos(4πkn / N ). Let C = ∑ cos(4 πkn / N ) and

n=0

N −1

N −1

n=0

n=0

S = ∑ sin( 4 πkn / N ). Then C + jS = ∑ e − j 4 πkn / N =

α

2.24 (a) x[ n] = A µ[ n]. Then E x a = ∑

n = −∞

2.25

1
n2

1− e

1 − e − j 4 πk / N

= 0. This implies

N
.
2

C = 0. Hence E x =

(b) x b [ n] =

n=0
− j 4 πkn

2
x a [ n] = A 2 ∑ α 2 n =

µ[ n − 1]. Then E x b ∑

n = −∞

2

A2

n=0

1− α

x b [ n] = ∑

1

2
n =1 n

= ∑

2

1

4
n =1 n

.
=

π4
.
90

(a) x1[ n] = (−1) n . Then average power
K
1

K → ∞ 2 K +1 n = − K

Px1 = lim

E x1 = ∑

1
(2 K + 1) = 1, and energy
K →∞ 2K + 1

2

n = −∞

2

x1[ n] = lim

x1[ n] = ∑ 1 = ∞.
n = −∞

(b) x 2 [ n] = µ[ n]. Then average power
K
K
K +1 1
1
1
2
Px 2 = lim
= , and energy
∑ x 2 [ n] = lim
∑ 1 = lim
K → ∞ 2K + 1 n = −K
K →∞ 2K + 1 n = 0 K → ∞ 2K + 1 2

E x2 = ∑

n = −∞

2

x 2 [ n] = ∑ 1 = ∞.
n=0

(c) x 3 [ n] = nµ[ n]. Then average power
K
K
K ( K + 1)(2 K + 1)
1
1
2
2
Px 3 = lim
= ∞,
∑ x3 [ n] = lim
∑ n = lim
6
K → ∞ 2K + 1 n = −K
K → ∞ 2 K + 1 n =1
K →∞

and energy E x 3 = ∑

n = −∞

2

x3 [ n] = ∑ n 2 = ∞.
n=0

K
1
2
∑ x 4 [ n]
K → ∞ 2K + 1 n = −K

(d) x 4 [ n] = A0 e jω 0 n . Then average power Px 4 = lim

Not for sale.

12

2
K
K
1
1
1
jω n
2
= lim
⋅ A02 (2 K + 1) = A02 ,
∑ A0 e 0
∑ A0 = lim
K → ∞ 2K + 1 n = −K
K →∞ 2K + 1 n = −K
K →∞ 2K + 1

= lim

and energy E x 3 = ∑

n = −∞

2

x 3 [ n] = ∑

n = −∞

A0 e jω 0 n

2

= ∑ A02 = ∞.
n = −∞

2 πn
(e) x 5 [ n] = A cos⎛⎜
+ φ ⎞⎟. Note x 5 [ n] is a periodic sequence. Then average power
M

Px 5 =

2
1 M −1
1 M −1
1 A 2 M −1⎛ ⎛ 4 πn
2 πn
2

+ 2 φ ⎞⎟ + 1⎟.

+ φ ⎞⎟ =
∑ ⎜ cos⎜
∑ A cos⎛⎜
∑ x 5 [ n] =
M
M

M n=0
M n=0
M 2 n=0 ⎝

M −1
M −1
4πn
4πn
Let C = ∑ cos⎛⎜
+ 2φ ⎞⎟ and S = ∑ cos⎛⎜
+ 2φ ⎞⎟. Then
M
M

n=0
n=0

j ⎛⎜
C + jS = ∑ e ⎝
n=0
M −1

4 πn
+ 2 φ ⎞⎟
M

=e

j 2φ

M −1

∑ e

j 4 πn / M

n =0

=e

j 2φ

1 − e j 4π
1 − e j 4π / M

= 0.

1 A 2 M −1 A 2

.
∑1 =
M 2 n=0
2
Since x 5 [ n] is a periodic sequence, it has infinite energy.

Hence C = 0. Therefore Px 5 =

2.26 In each of the following parts, N denotes the fundamental period and r is a positive
integer.

x [ n] = 4 cos(2πn / 5). Here N and r must satisfy the relation
⋅ N = 2 πr.
(a) ~
1

5

Among all positive solutions for N and r , the smallest values are N = 5 and r = 1.
Hence the average power is given by
Px1 =

2
1 4
1 N −1 ~
2
2 πn ⎞
⎟ = 8.
∑ x1[ n] = ∑ 4 cos⎛⎜
⎝ 5 ⎠
N n=0
5 n=0

x 2 [ n] = 3 cos(3πn / 5). Here N and r must satisfy the relation
(b) ~
⋅ N = 2πr.
5

Among all positive solutions for N and r , the smallest values are N = 10 and r = 3.
Hence the average power is given by
2
1 9
1 N −1 ~
2
3πn ⎞

Px 2 =
⎟ = 4 .5 .
∑ x 2 [ n] =
∑ 3 cos⎜
⎝ 5 ⎠
N n=0
10 n = 0

(c) ~
x3 [ n] = 2 cos(3πn / 7). Here N and r must satisfy the relation
⋅ N = 2πr.

7

Among all positive solutions for N and r , the smallest values are N = 14 and r = 3.
Hence the average power is given by
2
1 13
1 N −1 ~
2
3πn ⎞

Px 3 =
⎟ = 2.
∑ x 3 [ n] =
∑ 2 cos⎜
⎝ 7 ⎠
N n=0
14 n = 0

Not for sale.

13

x 4 [ n] = 4 cos(5πn / 3). Here N and r must satisfy the relation
(d) ~
⋅ N = 2 πr.
3

Among all positive solutions for N and r , the smallest values are N = 6 and r = 5.
Hence the average power is given by
2

1 5
1 N −1 ~
2
5πn ⎞
⎟ = 8.
∑ x 4 [ n] = ∑ 4 cos⎛⎜
⎝ 3 ⎠
N n=0
6 n=0
(e) ~
x 5 [ n] = 4 cos(2πn / 5) + 3 cos(3πn / 5). We first determine the fundamental period

Px 4 =

N1 of cos( 2 πn / 5). Here N1 and r must satisfy the relation
⋅ N1 = 2 πr. Among all
5
positive solutions for N1 and r , the smallest values are N1 = 5 and r = 1. We next
determine the fundamental period N 2 of cos(3πn / 5). Here N 2 and r must satisfy the

⋅ N2
5

= 2πr. Among all positive solutions for N 2 and r , the smallest values
are N 2 = 10 and r = 3. The fundamental period of ~
x 5 [ n] is then given by
relation

LCM ( N1 , N 2 ) = LCM (5,10) = 10.
Hence the average power is given by
2
1 4
1 N −1 ~
2
2 πn ⎞
3πn ⎞

Px 5 =
⎟ + 3 cos⎜

∑ x 5 [ n] =
∑ 4 cos⎜
⎝ 5 ⎠
⎝ 5 ⎠
N n=0
10 n = 0
11
11
1 ⎛ 11
3πn ⎞
2 πn ⎞
3πn ⎞ ⎞
2 πn ⎞
= ⎜⎜ ∑ 16 cos 2 ⎛⎜
⎟ ⎟ ≅ 8 + 4.5 + 0 = 12.5.
⎟ cos⎛⎜
⎟ + ∑ 24 cos⎛⎜
⎟ + ∑ 9 cos 2 ⎛⎜
⎝ 5 ⎠ ⎝ 5 ⎠ ⎟⎠
⎝ 5 ⎠ n=0
⎝ 5 ⎠ n=0
10 ⎝ n = 0

x6 [ n] = 4 cos(5πn / 3) + 3 cos(3πn / 5). We first determine the fundamental period
(f) ~

N1 of cos( 5πn / 3). Here N1 and r must satisfy the relation
⋅ N1 = 2πr. Among all
3
positive solutions for N1 and r , the smallest values are N1 = 6 and r = 5. We next
determine the fundamental period N 2 of cos(3πn / 5). Here N 2 and r must satisfy the

⋅ N2
5

= 2πr. Among all positive solutions for N 2 and r , the smallest values
are N 2 = 10 and r = 3. The fundamental period of ~
x6 [ n] is then given by
relation

LCM ( N1 , N 2 ) = LCM (6,10) = 30.
Hence the average power is given by
2
1 N −1 ~
1 29
2
5πn ⎞
3πn ⎞

Px 6 =
⎟ + 3 cos⎜

∑ x 6 [ n] =
∑ 4 cos⎜
⎝ 3 ⎠
⎝ 5 ⎠
N n=0
30 n = 0
29
30

1 ⎛ 29
⎜⎜ ∑ 16 cos 2 ⎛⎜ 5πn ⎞⎟ + ∑ 9 cos 2 ⎛⎜ 3πn ⎞⎟ + ∑ 24 cos⎛⎜ 5πn ⎞⎟ cos⎛⎜ 3πn ⎞⎟ ⎟⎟ ≅ 8 + 4.5 + 0 = 12.5.
=
⎝ 3 ⎠ ⎝ 5 ⎠⎠
⎝ 5 ⎠ n =0
⎝ 3 ⎠ n=0
30 ⎝ n = 0

2.27

Now , from Eq. (2.38) we have ~
y[ n] = ∑ x[ n + kN ]. Therefore
k = −∞

Not for sale.

14

~
y[ n + N ] = ∑ x[ n + kN + N ]. Substituting r = k + 1 we get
k = −∞

~
y[ n + N ] = ∑ x[ n + rN ] = ~
y[ n]. Hence ~
y[ n] is a periodic sequence with a period N .
r = −∞

2.28

x p [ n] in the range 0 ≤ n ≤ 4 is
(a) N = 5. Now ~
x p [ n] = ∑ x[ n + k 5]. The portion of ~
n = −∞

given by x[ n − 5] + x[ n] + x[ n + 5] = {0 0 − 4 5 1}
+ {− 2 − 3 0 2 0} + {0 0 0 0 0} = {− 2 − 3 − 4 7 1}, 0 ≤ n ≤ 4.
x p [ n] is given by {− 2 − 3 − 4 7 1}, 0 ≤ n ≤ 4.
Hence, one period of ~

y p [ n] in the range 0 ≤ n ≤ 4 is given by
Now ~
y p [ n] = ∑ y[ n + k 5]. The portion of ~
n = −∞

y[ n − 5] + y[ n] + y[ n + 5] = {0 0 0 0 6}
+ {− 3 − 1 0 8 7} + {− 2 0 0 0 0} = {− 5 − 1 0 8 13}, 0 ≤ n ≤ 4.
y p [ n] is given by {− 5 − 1 0 8 13}, 0 ≤ n ≤ 4.
Hence, one period of ~

~ [ n] in the range 0 ≤ n ≤ 4 is given by
~ [ n] = ∑
Now w
w[ n + k 5]. The portion of w
p
p
n = −∞

w[ n − 5] + w[ n] + w[ n + 5] = {0 0 0 0 0}
+ {0 0 3 2 2} + {− 1 0 − 2 5 0} = {− 1 0 1 7 2}, 0 ≤ n ≤ 4.
~ [ n] is given by {− 1 0 1 7 2}, 0 ≤ n ≤ 4.
Hence, one period of w
p

x p [ n] in the range 0 ≤ n ≤ 6 is
(b) N = 7. Now ~
x p [ n] = ∑ x[ n + k 7]. The portion of ~
n = −∞

given by x[ n − 7] + x[ n] + x[ n + 7] = {0 0 0 0 − 4 5 1}
+ {− 2 − 3 0 2 0 0 0} + {0 0 0 0 0 0 0}
= {− 2 − 3 0 2 − 4 5 1}, 0 ≤ n ≤ 6. Hence, one period of ~
x p [ n] is given by

{− 2 − 3 0 2 − 4 5 1}, 0 ≤ n ≤ 6.

y p [ n] in the range 0 ≤ n ≤ 6 is given by
Now ~
y p [ n] = ∑ y[ n + k 7]. The portion of ~
n = −∞

x[ n − 7] + x[ n] + x[ n + 7] = {0 0 0 0 0 0 6}
+ {− 3 − 1 0 8 7 − 2 0} + {0 0 0 0 0 0 0}
= {− 3 − 1 0 8 7 − 2 6}, 0 ≤ n ≤ 6. Hence, one period of ~
y p [ n] is given by
{− 3 − 1 0 8 7 − 2 6}, 0 ≤ n ≤ 6.

~ [ n] in the range 0 ≤ n ≤ 6 is given by
~ [ n] = ∑
Now w
w[ n + k 7]. The portion of w
p
p
n = −∞

w[ n − 7] + w[ n] + w[ n + 7] = {0 0 0 0 0 0 0}
+ {0 0 3 2 2 − 1 0} + {− 2 5 0 0 0 0 0}

Not for sale.

15

2.29

~ [ n] is given by
= {− 2 5 3 2 2 − 1 0}, 0 ≤ n ≤ 6. Hence, one period of w
p
{− 2 5 3 2 2 − 1 0}, 0 ≤ n ≤ 6.
~
x[ n] = A cos(ωo n + φ).
x[ n] = {1 − 1 − 1 1 1 − 1 − 1 1}. Hence A = 2 , ω = π / 2, φ = π / 4.
(a) ~
o

(b) ~
x[ n] = {0 − 3 0
3 0 − 3 0
3}. Hence A = 3 , ωo = π / 2,
φ = π / 2.
x[ n] = {1 − 0.366 − 1.366 − 1 0.366 1.366}. Hence A = 2 , ωo = π / 3,
(c) ~
φ = π / 4.
x[ n] = {2 0 − 2 0 2 0 − 2 0}. Hence A = 2, ωo = π / 2, φ = 0.
(d) ~
2.30

The fundamental period N of a periodic sequence with an angular frequency ω o
satisfies Eq. (2.47a) with the smallest value of N and r .
(a) ωo = 0.5π. Here Eq. (2.47a) reduces to 0.5πN = 2πr which is satisfied with
N = 4, r = 1.
(b) ωo = 0.8π. Here Eq. (2.47a) reduces to 0.8πN = 2πr which is satisfied with
N = 5, r = 2.
(c) We first determine the fundamental period N1 of Re{e jπn / 5} = cos(0.2πn). In
this case, Eq. (2.47a) reduces to 0.2πN 1= 2πr1 which is satisfied with N1 = 10, r1 = 1.
We next determine the fundamental period N 2 of Im{e jπn / 10 = j sin(0.1πn). In this
case, Eq. (2.47a) reduces to 0.1πN 2 = 2πr2 which is satisfied with N 2 = 20, r2 = 1.
Hence the fundamental period N of ~
x c [ n] is given by

LCM ( N1 , N 2 ) = LCM (10,20) = 20.
(d) We first determine the fundamental period N1 of 3 cos(1.3πn). In this case, Eq.
(2.47a) reduces to 1.3πN 1= 2πr1 which is satisfied with N1 = 20, r1 = 13. We next
determine the fundamental period N 2 of 4 sin( 0.5πn + 0.5π). In this case, Eq. (2.47a)
reduces to 0.5πN 2 = 2πr2 which is satisfied with N 2 = 4, r2 = 1. Hence the
x 4 [ n] is given by LCM ( N1 , N 2 ) = LCM (20,4) = 20.
fundamental period N of ~
(e) We first determine the fundamental period N1 of 5 cos(1.5πn + 0.75π). In this
case, Eq. (2.47a) reduces to 1.5πN 1= 2πr1 which is satisfied with N1 = 4, r1 = 3. We
next determine the fundamental period N 2 of 4 cos( 0.6 πn). In this case, Eq. (2.47a)
reduces to 0.6πN 2 = 2πr2 which is satisfied with N 2 = 10, r2 = 3. We finally
determine the fundamental period N 3 of sin( 0.5πn). In this case, Eq. (2.47a) reduces
to 0.5πN 3= 2πr3 which is satisfied with N 3 = 4, r3 = 1. Hence the fundamental period
x 5 [ n] is given by LCM ( N1 , N 2 , N 3 ) = LCM (4,10,4) = 20.
N of ~
2.31

The fundamental period N of a periodic sequence with an angular frequency ω o
satisfies Eq. (2.47a) with the smallest value of N and r .
Not for sale.

16

(a) ωo = 0.6π. Here Eq. (2.47a) reduces to 0.6πN = 2πr which is satisfied with
N = 10, r = 3.
(b) ωo = 0.28π. Here Eq. (2.47a) reduces to 0.28πN = 2πr which is satisfied with
N = 50, r = 7.
(c) ωo = 0.45π. Here Eq. (2.47a) reduces to 0.45πN = 2πr which is satisfied with
N = 40, r = 9.
(d) ωo = 0.55π. Here Eq. (2.47a) reduces to 0.55πN = 2πr which is satisfied with
N = 40, r = 11.
(e) ωo = 0.65π. Here Eq. (2.47a) reduces to 0.65πN = 2πr which is satisfied with
N = 40, r = 13.
2.32

ωo = 0.08π. Here Eq. (2.47a) reduces to 0.08πN = 2πr which is satisfied with
x 2 [ n] = sin(ω2 n) with a fundamental period of N = 25 ,
N = 25, r = 1. For a sequence ~
Eq. (2.47a) reduces to 25ω2 = 2πr. For example, for r = 2 we have
ω2 = 4π / 25 = 0.16π. Another sequence with the same fundamental period is obtained
by setting r = 3 which leads to ω3 = 6π / 25 = 0.24 π. The corresponding periodic
x [ n] = sin(0.16πn) and ~
x [ n] = sin(0.24 πn).
sequences are therefore ~
2

2.33

3

The three parameters A, Ω o , and φ of the continuous-time signal x a (t ) can be
determined from x[ n] = x a (nT ) = A cos(Ω o nT + φ) by setting 3 distinct values of n.
For example
x[ 0] = A cos φ = α,
x[ −1] = A cos(−Ω o T + φ) = A cos(Ω o T ) cos φ + A sin(Ω o T ) sin φ = β, ,

x[1] = A cos(Ω o T + φ) = A cos(Ω o T ) cos φ − A sin(Ω o T ) sin φ = γ.
Substituting the first equation into the last two equations and then adding them we get
β+γ
which can be solved to determine Ω o . Next, from the second
cos(Ω o T ) =

equation we have A sin φ = β − A cos(Ω o T ) cos φ = β − α cos(Ω o T ). Dividing this
β − α cos(Ω o T )
equation by the last equation on the previous page we arrive at tan φ =
α sin(Ω o T
which can be solved to determine φ. Finally, the parameter is determined from the first
equation of the last page.

Not for sale.

17

= 2Ω o . In this case x[ n] = A cos( nπ + φ) = β and
T
x[ n + 1] = A cos((n + 1)π + φ) = A cos(nπ + φ) = β . Since all sample values are equal, the
three parameters cannot be determined uniquely.

Now consider the case Ω T =

< 2Ω o . In this case x[ n] = A cos(Ω o nT + φ)
T
= A cos(ωo n + φ) implying ωo = Ω o T > π. As explained in Section 2.2.1, a digital

Finally consider the case Ω T =

sinusoidal sequence with an angular frequency ω o greater than π assumes the identity
of a sinusoidal sequence with an angular frequency in the range 0 ≤ ω < π. . Hence,
Ω o cannot be uniquely determined from x[ n] = A cos(Ω o nT + φ) .
2.34

x[ n] = cos(Ω o nT ). If x[n] is periodic with a period N , then
x[ n + N ] = cos(Ω 0 nT + Ω 0 NT ) = x[ n] = cos(Ω 0 nT ). This implies Ω o NT = 2πr with
r any nonzero positive integer. Hence the sampling rate must satisfy the relation
π
T = 2πr / Ω o N . If Ω o = 20, i.e., T = π / 8, then we must have 20 N ⋅ = 2 πr . The
8
smallest value of N and r satisfying this relation are N = 4 and r = 5. The
fundamental period is thus N = 4 .

2.35

(a) For an input xi [ n], i = 1,2, the output is
yi [ n] = b0 xi [ n] + b1 xi [ n − 1] + b2 xi [ n − 2] + a1 yi [ n − 1] + a 2 yi [ n − 2], i = 1,2. Then, for
an input x[ n] = Ax1[ n] + Bx2 [ n], the output is y[ n] = b0 ( Ax1[ n] + Bx 2 [ n])

+ b1 ( Ax1[ n − 1] + Bx2 [ n − 1]) + b2 ( Ax1[ n − 2] + Bx2 [ n − 2]) + a1 ( Ay1[ n − 1] + By2 [ n − 1])
+ a1 ( Ay1[ n − 2] + By2 [ n − 2]) = A(b0 x1[ n] + b1 x1[ n − 1] + b2 x1[ n − 2] + a1 y1[ n − 1]
+ a 2 y2 [ n − 1]) + B(b1 x 2 [ n] + b2 x 2 [ n − 1] + b3 x 2 [ n − 2] + a1 x 2 [ n − 1] + a 2 x 2 [ n − 2])
= Ay1[ n] + By2 [ n]. Hence, the system of Eq. (2.18) is linear.
⎧ x [ n / L ], n = 0, ± L, ± 2 L,L
(b) For an input xi [ n], i = 1,2, the output is yi [ n] = ⎨ i
otherwise.
⎩ 0,
For an input x[ n] = Ax1[ n] + Bx2 [ n], the output for n = 0, ± L, ± 2 L,K is
y[ n] = x[ n / L ] = Ax1[ n / L ] + Bx2 [ n / L ] = Ay1[ n] + By2 [ n] . For all other values of
n, y[ n] = A ⋅ 0 + B ⋅ 0 = 0. Hence the system of Eq. (2.20) is linear.

(c) For an input xi [ n], i = 1,2, the output is yi [ n] = x i [ n / M ], i = 1,2. Then, for an input

x[ n] = Ax1[ n] + Bx2 [ n], the output is y[ n] = Ax1[ n / M ] + Bx2 [ n / M ] = Ay1[ n] + By 2 [ n].
Hence the system of Eq. (2.21) is linear.

Not for sale.

18

(d) For an input xi [ n], i = 1,2, the output is yi [ n] =

1 M −1
∑ xi [ n − k ], i = 1,2. Then, for
M k =0

an input x[ n] = Ax1[ n] + Bx2 [ n], the output is y[ n] =

1 M −1
∑ ( Ax1[ n − k ] + Bx 2 [ n − k ])
M k =0

⎛ 1 M −1

⎛ 1 M −1
= A⎜⎜
∑ x 2 [ n − k ] ⎟⎟ = Ay1[ n] + By2 [ n]. Hence the system of
∑ x1[ n − k ] ⎟⎟ + B⎜⎜

⎝ M k =0

⎝ M k =0
Eq. (2.61) is linear.

(e) The first term on the RHS of Eq. (2.65) is the output of a factor-of-2 up-sampler.
The second term on the RHS of Eq. (2.65) is simply the output of an unit delay
followed by a factor-of-2 up-sampler, whereas, the third term is the output of an unit
advance operator followed by a factor-of-2 up-sampler. We have shown in Part (b) that
the up-sampler is a linear system. Moreover, the unit delay and the unit advance
operator are linear systems. A cascade of two linear systems is linear and the linear
combination of linear systems is also linear. Hence, the factor-of-2 interpolator of Eq.
(2.65) is a linear system.
(f) Following the arguments given in Part (e), we can similarly show that the factor-of3 interpolator of Eq. (2.66) is a linear system.
2.36

(a) y[ n] = n 3 x[ n]. For an input xi [ n], i = 1,2, the output is yi [ n] = n 3 x i [ n], i = 1,2.
Then, for an input x[ n] = Ax1[ n] + Bx2 [ n], the output is y[ n] = n 3 ( Ax1[ n] + Bx 2 [ n])
= Ay1[ n] + By2 [ n]. Hence the system is linear.
For an input x[ n] = δ[ n], the output is the impulse response h[ n] = n 3 δ[ n]. As
h[ n] = 0 for n < 0, and the system is causal.
Let x[ n] = 1 for all values of n. Then y[ n] = n 3 and y[n] → ∞ as n → ∞. Since a
bounded input results in an unbounded output, the system is not BIBO stable.
Finally, let y[n] and y1[ n] be the outputs for inputs x[n] and x1[ n], respectively. If

x1[ n] = x[ n − no ] then y1[ n] = n 3 x1[ n] = n 3 x[ n − no ]. However, y[ n − no ] =
( n − n o ) 3 x[ n − n o ]. Since y1[ n] ≠ y[ n − no ], the system is not time-invariant.

(b) y[ n] = ( x[ n]) 5 . For an input xi [ n], i = 1,2, the output is yi [ n] = ( x i [ n]) 5 , i = 1,2.
Then, for an input x[ n] = Ax1[ n] + Bx2 [ n], the output is y[ n] = ( Ax1[ n] + Bx 2 [ n])5

≠ A( x1 [ n]) 5 + B( x 2 [ n]) 5 . Hence the system is nonlinear.
For an input x[ n] = δ[ n], the output is the impulse response h[ n] = (δ[ n]) 5 . As
h[ n] = 0 for n < 0, and the system is causal.

Not for sale.

19

For a bounded input x[ n] ≤ B < ∞, the magnitude of the output samples are
5

y[ n] = ( x[ n]) 5 = x[ n] ≤ B 5 < ∞. As the output is also a bounded sequence, the

system is BIBO stable.
Finally, let y[n] and y1[ n] be the outputs for inputs x[n] and x1[ n], respectively. If
x1[ n] = x[ n − no ] then y1[ n] = ( x 1 [ n]) 5 = (x[ n − no ])5 = y[ n − no ]. Hence, the system
is time-invariant.

3

(c) y[ n] = β + ∑ x[ n − l] with β a nonzero constant. For an input xi [ n], i = 1,2, the
l=0

3

output is yi [ n] = β + ∑ x i [ n − l], i = 1,2. Then, for an input x[ n] = Ax1[ n] + Bx2 [ n],
l =0

3

3

3

l =0

l =0

l=0

the output is y[ n] = β + ∑ ( Ax1[ n − l] + Bx 2 [ n − l]) = β + ∑ Ax1[ n − l] + ∑ Bx 2 [ n − l]

≠ Ay1[ n] + By2 [ n]. Hence the system is nonlinear.

For an input x[ n] = δ[ n], the output is the impulse response h[ n] = β + ∑ δ[ n − l]. As
l=0

h[ n] ≠ 0 for n < 0, the system is noncausal.
For a bounded input x[ n] ≤ B < ∞, the magnitude of the output samples are

y[ n] ≤ β + 4 B < ∞. As the output is also a bounded sequence, the system is BIBO
stable.
Finally, let y[n] and y1[ n] be the outputs for inputs x[n] and x1[ n], respectively. If
3

x1[ n] = x[ n − no ] then y1[ n] = β + ∑ x[ n −n o −l] = y[ n − no ]. Hence, the system is
l=0

time-invariant.
(d) y[ n] = ln(2 + x[ n] ). For an input xi [ n], i = 1,2, the output is yi [ n] = ln(2 + xi [ n] ),
i = 1,2. Then, for an input x[ n] = Ax1[ n] + Bx2 [ n], the output is

y[ n] = ln (2 + Ax1[ n] + Bx2 [ n] ) ≠ Ay1[ n] + By2 [ n]. Hence the system is nonlinear.

For an input x[ n] = δ[ n], the output is the impulse response h[ n] + ln(2 + δ[ n] ) .
For n < 0, h[ n] = ln(2) ≠ 0. Hence, the system is noncausal.
For a bounded input x[ n] ≤ B < ∞, the magnitude of the output samples are

y[ n] ≤ ln(2 + B ) < ∞. As the output is also a bounded sequence, the system is BIBO
stable.
Finally, let y[n] and y1[ n] be the outputs for inputs x[n] and x1[ n], respectively. If

Not for sale.

20

x1[ n] = x[ n − no ] then y1[ n] = ln(2 + x[ n − no ] ) = y[ n −n o ]. Hence, the system is
time-invariant.
(e) y[ n] = αx[ − n + 2], with a nonzero constant. For an input xi [ n], i = 1,2, the output
is yi [ n] = αxi [ −n + 2], i = 1,2. Then, for an input x[ n] = Ax1[ n] + Bx2 [ n], the output is

y[ n] = Aαx1[ −n + 2] + Bαx 2 [ −n + 2] = Ay1[ n] + By2 [ n]. Hence the system is linear.
For an input x[ n] = δ[ n], the output is the impulse response h[ n] = αδ[ −n + 2]. For
n < 0, h[ n] = 0. Hence, the system is causal.
For a bounded input x[ n] ≤ B < ∞, the magnitude of the output samples are
y[ n] = αB < ∞. As the output is also a bounded sequence, the system is BIBO stable.
Finally, let y[n] and y1[ n] be the outputs for inputs x[n] and x1[ n], respectively. If
x1[ n] = x[ n − no ] then y1[ n] = αx1[ − n + 2] = αx[ −(n − no ) + 2] = y[ n −n o ]. Hence, the
system is time-invariant.
(f) y[ n] = x[ n − 4]. For an input xi [ n], i = 1,2, the output is yi [ n] = xi [ n − 4], i = 1,2.
Then, for an input x[ n] = Ax1[ n] + Bx2 [ n], the output is y[ n] = Ax1[ n − 4] + Bx2 [ n − 4]
= Ay1 [ n] + By2 [ n]. Hence the system is linear.
For an input x[ n] = δ[ n], the output is the impulse response h[ n] = δ[ n − 4]. For n < 0,
h[ n] = 0. Hence, the system is causal.
For a bounded input x[ n] ≤ B < ∞, the magnitude of the output samples are

y[ n] = B < ∞. As the output is also a bounded sequence, the system is BIBO stable.
Finally, let y[n] and y1[ n] be the outputs for inputs x[n] and x1[ n], respectively. If
x1[ n] = x[ n − no ] then y1[ n] = x[ n −n o −4] = y[ n − no ]. Hence, the system is timeinvariant.
2.37

Let y[n] and y1[ n] be the outputs of a median filter of length 2 K + 1 for inputs x[n]
and x1[ n] , respectively. If x1 [ n] = x[ n − no ] , then
y1[ n] = med{x1[ n − K ], K, x1[ n − 1], x1[ n], x1[ n + 1], K, x1[ n + K ]}
= med{x[ n − no − K ], K, x[ n − no − 1], x[ n − no ], x[ n − no + 1], K, x[ n − no + K ]}
= y[ n − no ]. Hence, the system is time-invariant.

2.38

y[ n] = x[ n + 1] − 2 x[ n] + x[ n − 1]. For an input x i [ n], i = 1,2, the output is
yi [ n] = x i [ n + 1] − 2 xi [ n] + xi [ n − 1], i = 1,2. Then, for an input x[ n] = Ax1[ n] + Bx2 [ n],
the output is
y[ n] = Ax1[ n + 1] + Bx 2 [ n + 1] − 2 Ax1[ n] − 2 Bx 2 [ n] + Ax1[ n − 1] + Bx 2 [ n − 1]
= Ay1[ n] + By2 [ n]. Hence the system is linear.
If x1[ n] = x[ n − no ], then y1[ n] = x[ n − no + 1] − 2 x[ n − no ] + x[ −no n − 1] = y[ n − no ].
Hence, the system is time-invariant.

Not for sale.

21

The impulse response of the system is h[ n] = δ[ n + 1] − 2δ[ n] + δ[ n − 1] . Now
h[ −1] = δ[0] = 1. Since h[ n] ≠ 0 for all values of n < 0, the system is noncausal.
2.39

y[ n] = x 2 [ n] − x[ n − 1]x[ n + 1]. For an input xi [ n], i = 1,2, the output is
yi [ n] = x i2 [ n] − x i [ n − 1]x i [ n + 1], i = 1,2. Then, for an input x[ n] = Ax1[ n] + Bx2 [ n], the

output is y[ n] = ( Ax1[ n] + Bx 2 [ n])2 − ( Ax1[ n − 1] + Bx 2 [ n − 1])( Ax1[ n + 1] + Bx 2 [ n + 1])
≠ Ay1[ n] + By2 [ n]. Hence the system is nonlinear.
If x1[ n] = x[ n − no ], then y1[ n] = x12 [ n] − x1[ n − 1]x1[ n + 1]
= x 2 [ n − no ] − x[ n − no − 1] x[ n − no + 1] = y[ n − no ]. Hence, the system is timeinvariant.

The impulse response of the system is h[ n] = δ 2 [ n] − δ[ n − 1]δ[ n + 1] = δ[ n]. Since
h[ n] = 0 for all values of n < 0, the system is causal.
2.40

1⎛
x[ n] ⎞
⎜⎜ y[ n − 1] +
⎟. Now for an input x[ n] = αµ[ n], the output y[n]
2⎝
y[ n − 1] ⎟⎠
converges to some constant K as n → ∞. The input-output relation of the system as
α⎞
1⎛
n → ∞ reduces to K = ⎜ K + ⎟ from which we get K 2 = α or in other words
2⎝
K⎠
K = α.
x [ n] ⎞
1⎛
⎟, i = 1,2. Then,
For an input xi [ n], i = 1,2, the output is y i [ n] = ⎜⎜ yi [ n − 1] + i
2⎝
yi [ n − 1] ⎟⎠
y[ n] =

for an input x[ n] = Ax1[ n] + Bx2 [ n], the output is y[ n] =

Ax [ n] + Bx 2 [ n] ⎞
1⎛
⎜⎜ y[ n − 1] + 1
⎟⎟.
y[ n − 1]
2⎝

On the other hand,

Bx 2 [ n] ⎞
Ax1 [ n] ⎞ 1 ⎛
1⎛
⎟ ≠ y[ n].
⎟⎟ + ⎜⎜ By 2 [ n − 1] +
⎜⎜ Ay1 [ n − 1] +
y 2 [ n − 1] ⎟⎠
y1 [ n − 1] ⎠ 2 ⎝
2⎝
Hence the system is nonlinear.
x[ n − no ] ⎞
1⎛
⎟ = y[ n − no ]. Hence, the
If x1[ n] = x[ n − no ], then y1[ n] = ⎜⎜ y1[ n − 1] +
y1[ n − 1] ⎟⎠
2⎝
system is time-invariant.
Ay1[ n] + By2 [ n] =

2.41

y[ n] = x[ n] − y 2 [ n − 1] + y[ n − 1].
For an input xi [ n], i = 1,2, the output is yi [ n] = x i [ n] − yi2 [ n − 1] + yi [ n − 1], i = 1,2.
Then, for an input x[ n] = Ax1[ n] + Bx2 [ n], the output is

y[ n] = Ax1[ n] + Bx 2 [ n] − y 2 [ n − 1] + y[ n − 1]. On the other hand, Ay1[ n] + By2 [ n]

Not for sale.

22

= Ax1[ n] − Ay12 [ n − 1] + Ay1[ n − 1] + Bx 2 [ n] − By22 [ n − 1] + By2 [ n − 1] ≠ y[ n]. Hence the
system is nonlinear.
Let y[n] be the output for an input x[n] , i.e., y[n] = x[n] − y 2 [n] + y[n − 1]. Then, for an
input x[n − no ] the output is given by y[ n − n o ] = x[ n − n o ] − y 2 [ n − n o −1] + [ n − n o −1] , or
in other words, the system is time-invariant.
Now, for an input x[ n] = αµ[ n] , the output y[n] converges to some constant K as n → ∞ .
The difference equation describing the system as n → ∞ reduces to K = α − K 2 + K , or
K 2 = α, i.e., K = α .
2.42

The impulse response of the factor-of-3 interpolator of Eq. (2.66) is the output for an
input x u [ n] = δ[ n] and is given by
2
3
2
, 1,
3

1
3

h[ n] = δ[ n] + (δ[ n − 1] + δ[ n + 1]) + (δ[ n − 2] + δ[ n + 2]) or equivalently by
{h[ n]} =

2.43

{

1
,
3

2
,
3

}

1
, − 2 ≤ n ≤ 2.
3

The input-output relation of a factor-of- L interpolator is given by
L −1 L − k
(xu [ n − k ] + xu [ n + k ]). Its impulse response is the output for
y[ n] = xu [ n] + ∑
k =1 L
L −1 L − k

an input xu [ n] = δ[ n] and is thus given by h[ n] = δ[ n] + ∑

{,

k =1

L

K,

2
,
L

(δ[ n − k ] + δ[ n + k ])

or equivalently by

{h[ n]} =
2.44

1
L

2
,
L

K,

L −2
,
L

L −1
,
L

1,

L −1
,
L

L −2
,
L

1
L

}, − L + 1 ≤ n ≤ L − 1.

The impulse response h[n] of a causal discrete-time system satisfies the difference
equation h[ n] − a h[ n − 1] = δ[ n]. Since the system is causal, we have h[ n] = 0 for
n < 0. Evaluating the above difference equation for n = 0, we arrive at
h[0] − ah[ −1] = 1 and thus h[0] = 1. Next, for n = 1 , we have h[1] − a h[ 0] = 0 and thus

h[1] = a. Continuing we get for n = 2, h[ 2] − ah[1] = 0 , i.e., h[2] = ah[1] = a 2 . Assume

h[ n − 1] = a n −1 with n > 0. From the difference equation we then have
h[ n] − a h[ n − 1] = 0 , i.e., h[ n] = ah[ n − 1] = a n . Since the last equation holds for
n = 0,1, 2, by induction, it holds for n ≥ 3.

2.45

As x[n] and h[n] are right-sided sequences, assume x[ n] = 0 for all n < N1 and
h[ n] = 0 and n < N 2 . Hence, y[ n] = h[ n] O
* x[ n] = 0 for all n < N1 + N 2 and thus

Not for sale.

23

y[n] is also a right-sided sequence. Therefore,

=

n = N1 + N 2 k = N 2

n = N1 + N 2

* x[ n]
∑ h[ n] O

n = N1 + N 2

∑ h[ k ]x[ n − k ] = ∑ h[ k ]

k = N 2 n = N1 + N 2

m = N1 + N 2 − k

k = N2

m = N1

k =N2

∑ x[ n − k ]

n = N1 + N 2

∑ x[ m] = ∑ h[ k ] ∑ x[ m] as x[ m] = 0 for all m < N1 . Hence,

= ∑ h[ k ]
k = N2

∑ h[ k ]x[ n − k ] = ∑

∑ y[ n] =

⎞⎛

∑ y[ n] =⎜⎜ ∑ h[ n] ⎟⎟⎜⎜ ∑ x[ n] ⎟⎟.
n

⎠⎝ n
⎝n
2.46

2.47

⎧∑ n α k , n ≥ 0,
k
k
(a) α n µ[n] O
* µ[n] = ∑ α µ[k ]µ[n − k ] = ∑ α µ[n − k ] = ⎨ k = 0
0,
n < 0,
k = −∞
k =0

⎛ 1 − α n +1 ⎞
⎟µ[n].
=⎜
⎜ 1− α ⎟

⎧∑ n kα k , n > 0,
k
k
(b) nα n µ[ n] O
* µ[ n] = ∑ kα µ[ k ]µ[ n − k ] = ∑ kα µ[ n − k ] = ⎨ k = 0
0,
n ≤ 0.
k = −∞
k =0

Now from Eq. (2.72) an arbitrary input x[n] can be expressed as

x[ n] = ∑ x[ k ]δ[ n − k ] which can be rewritten using Eq. (2.41b) as
k = −∞

k = −∞

k = −∞

k = −∞

x[ n] = ∑ x[ k ](µ[ n − k ] − µ[ n − k − 1]) = ∑ x[ k ]µ[ n − k ] − ∑ x[ k ]µ[ n − k − 1].
Since s[n] is the response of an LTI system for an input µ[n], s[ n − k ] is the response
for an input µ[ n − k ] and s[ n − k − 1] is the response for an input µ[ n − k − 1]. Hence,
the output for an input

k = −∞

k = −∞

∑ x[ k ]µ[ n − k ] − ∑ x[ k ]µ[ n − k − 1] is given by

* s[ n] − x[ n − 1] O
* s[ n − 1].
y[ n] = ∑ x[ k ]s[ n − k ] − ∑ x[ k ]s[ n − k − 1] = x[ n] O
k = −∞

2.48

y[ n] =

k = −∞

x[ n − m]. Hence,
∑ h[ m]~

m = −∞

y[ n + kN ] =

x[ n + kN − m] =
∑ h[ m ]~

m = −∞

∑ h[ m]x[ n − m] = y[ n]. Thus, y[n] is also a

m = −∞

periodic sequence with a period N .
2.49

In this problem we make use of the identity δ[ n − m ] O
* δ[ n − r ] = δ[ n − m − r ].

Not for sale.

24

(a) y1[ n] = x1[ n] O
* h1[ n] = (3δ[ n − 2] − 2δ[ n + 1]) O
* (− δ[ n + 2] + 4δ[ n] + 2δ[ n − 1])
= −3δ[ n − 2] O
* δ[ n + 2]
* δ[ n + 2] + 12δ[ n − 2] O
* δ[ n] − 6δ[ n − 2] O
* δ[ n − 1] + 2δ[ n + 1] O
− 8δ[ n + 1] O
* δ[ n] + 4δ[ n + 1] O
* δ[ n − 1] . Hence

y1[ n] = −3δ[ n] + 12δ[ n − 2] − 6δ[ n − 3] + 2δ[ n + 3] − 8δ[ n + 1] + 4δ[ n]
= 2δ[ n + 3] − 8δ[ n + 1] + δ[ n] + 12δ[ n − 2] − 6δ[ n − 3].
(b) y 2 [ n] = x 2 [ n] O
* h2 [ n] = (5δ[ n − 3] + 2δ[ n + 1]) O
* (3δ[ n − 4] + 1.5δ[ n − 2] − δ[ n + 1])
= 15δ[ n − 3] O
* δ[ n − 4] + 7.5δ[ n − 3] O
* δ[ n − 2] − 5δ[ n − 3] O
* δ[ n + 1] + 6δ[ n + 1] O
* δ[ n − 4]
+ 3δ[ n + 1] O
* δ[ n − 2] − 2δ[ n + 1] O
* δ[ n + 1] = 15δ[ n − 7] + 7.5δ[ n − 5] − 5δ[ n − 2]

+ 6δ[ n − 3] + 3δ[ n − 1] − 2δ[ n + 2].
(c) y3 [ n] = x1[ n] O
* h2 [ n] = (− 3δ[ n − 2] − 2δ[ n + 1]) O
* (3δ[ n − 4] + 1.5δ[ n − 2] − δ[ n + 1])
= 9δ[ n − 2] O
* δ[ n − 4] + 4.5δ[ n − 2] O
* δ[ n − 2] − 3δ[ n − 2] O
* δ[ n + 1] − 6δ[ n + 1] O
* δ[ n − 4]
− 3δ[ n + 1] O
* δ[ n − 2] + 2δ[ n + 1] O
* δ[ n + 1] = 9δ[ n − 6] + 4.5δ[ n − 4] − 3δ[ n − 1]

− 6δ[ n − 3] − 3δ[ n − 1] − 3δ[ n − 1] + 2δ[ n + 2] = 2δ[ n + 2] − 6δ[ n − 1] − 6δ[ n − 3]
+ 4.5δ[ n − 4] + 9δ[ n − 6].
(d) y 4 [ n] = x 2 [ n] O
* h1[ n] = (5δ[ n − 3] + 2δ[ n + 1]) O
* (− δ[ n + 2] + 4δ[ n] − 2δ[ n − 1])
= −5δ[ n − 3] O
* δ[ n + 2] + 20δ[ n − 3] O
* δ[ n] − 10δ[ n − 3] O
* δ[ n − 1] − 2δ[ n + 1] O
* δ[ n + 2]
+ 8δ[ n + 1] O
* δ[ n] − 4δ[ n + 1] O
* δ[ n − 1] = −5δ[ n − 1] + 20δ[ n − 3] − 10δ[ n − 4] − 2δ[ n + 3]

+ 8δ[ n + 1] − 4δ[ n] = −2δ[ n + 3] + 8δ[ n + 1] − 4δ[ n] − 5δ[ n − 1] + 20δ[ n − 3] − 10δ[ n − 4].
2.50

(a) u[ n] = x[ n] O
* y[ n]

= {− 24, 42, − 5, − 20, − 45, 23, 66, − 25, − 42, − 17, 22, 14, − 4}, − 4 ≤ n ≤ 8.
(b) v[ n] = x[ n] O
* w[ n]

= {− 12, 7, 5, 10, − 16, − 3, − 28, 30, 13, − 6, − 15, − 4, 10}, − 1 ≤ n ≤ 11.
(c) g[ n] = w[ n] O
* y[ n]

= {18, 3, 3, − 14, 25, 26, 60, − 11, − 16, − 14, 26, 39, − 10}, 1 ≤ n ≤ 13.

Not for sale.

25

2.51

N2
y[ n] = ∑ m
g[ m]h[ n − m]. Now, h[ n − m] is defined for M1 ≤ n − m ≤ M 2 . Thus,
=N
1

for m = N1 , h[ n − m] is defined for M1 ≤ n − N1 ≤ M 2 , or equivalently, for
M1 + N1 ≤ n ≤ M 2 + N1 . Likewise, for m = N 2 , h[ n − m] is defined for
M1 ≤ n − N 2 ≤ M 2 , or equivalently, for M1 + N 2 ≤ n ≤ M 2 + N 2 . For the specified
sequences N1 = −3, N 2 = 4, M1 = 2, M 2 = 6. (a) The length of y[n] is
M 2 + N 2 − M1 − N1 + 1 = 6 + 4 − 2 − (−3) + 1 = 12 . (b) The range of n for y[ n] ≠ 0 is
min(M1 + N1 , M 2 + N 2 ) ≤ n ≤ max(M1 + N1 , M 2 + N 2 ) , i.e.,
M1 + N1 ≤ n ≤ M 2 + N 2 . For the specified sequences the range of n is − 1 ≤ n ≤ 10.

2.52

* x 2 [ n] = ∑ x 1 [ n − k ]x 2 [ k ]. Now,
y[ n] = x1 [ n] O
k = −∞

* x 2 [ n − N 2 ] = ∑ x1 [ n − N1 − k ] x 2 [ k − N 2 ]. Let k − N 2 = m.
v[ n] = x1 [ n − N1 ] O
k = −∞

Then v[ n] =
2.53

∑ x 1 [ n − N1
m = −∞

− N 2 − m ]x 2 [ m] = y[ n − N1 − N 2 ].

g[ n] = x1[ n] O
* x 2 [ n] O
* x 3 [ n] = y[ n] O
* x 3 [ n] where y[ n] = x1[ n] O
* x 2 [ n]. Now
v[ n] = x1[ n − N1 ] O
* x 2 [ n − N 2 ]. Define h[ n] = v[ n] O
* x 3 [ n − N 3 ]. Then from the

results of Problem 2.52, v[ n] = y[ n − N1 − N 2 ]. Hence,
h[ n] = y[ n − N1 − N 2 ] O
* x 3 [ n − N 3 ]. Therefore, making use of the results of Problem
2.52 again we get h[ n] = y[ n − N1 − N 2 − N 3 ].

2.54

* h[ n] = ∑ x[ n − k ]h[ k ]. Substituting k by n − m in this expression, we
y[ n] = x[ n] O
k = −∞

get y[ n] =

* x[ n]. Hence the convolution operation is
∑ x[ m]h[ n − m] = h[ n] O

m = −∞

commutative.

* (h1[ n] + h2 [ n]) = ∑ x[ n − k ](h1[ k ] + h2 [ k ])
Let y[ n] = x[ n] O
k = −∞

k = −∞

k = −∞

* h1[ n] + x[ n] O
* h2 [ n]. Hence the
= ∑ x[ n − k ]h1[ k ] + ∑ x[ n − k ]h2 [ k ] = x[ n] O

convolution operation is also distributive.
2.55

x 3 [ n] O
* x 2 [ n] O
* x1[ n] = x 3 [ n] O
* ( x 2 [ n] O
* x1[ n]). As x2 [ n] O
* x1[ n] is an unbounded

sequence, the result of this convolution cannot be determined. But

Not for sale.

26

x 2 [ n] O
* x 3 [ n] O
* x1[ n] = x 2 [ n] O
* ( x 3 [ n] O
* x1[ n]). Now x 3 [ n] O
* x1[ n] = 0 for all values

of n , and hence the overall result is zero. As a result, for the given sequences
x 3 [ n] O
* x 2 [ n] O
* x1[ n] ≠ x 2 [ n] O
* x 3 [ n] O
* x1[ n]
2.56

w[ n] = x[ n] O
* h[ n] = ∑ x[ k ]h[ n − k ] and
* h[ n] O
* g[ n]. Define y[ n] = x[ n] O
k

f [ n] = h[ n] O
* g[ n] = ∑ g[ k ]h[ n − k ]. Consider w1 [ n] = ( x[ n] O
* h[ n]) O
* g[ n]
k

= y[ n] O
* g[ n] = ∑ g[ m ]∑ x[ k ]h[ n − m − k ]. Now consider w 2 [ n] = x[ n] O
* (h[ n] O
* g[ n])
m

k

= x[ n] O
* f [ n] = ∑ x[ k ]∑ g[ m ]h[ n − k − m ]. The difference between the
k

m

expressions for w1[ n] and w2 [ n] is that the order of the summations is changed.
A) Assumptions: h[n] and g[n] are causal sequences, and x[ n] = 0 for n < 0. This
n
0,
for m < 0,

implies y[ m] = ⎨ m
Thus, w[ n] = ∑ g[ m]y[ n − m]
⎩∑ k = 0 x[k]h[m - k], for m ≥ 0.
m=0
n

n−m

m=0

k =0

= ∑ g[ m] ∑ x[ k ]h[ n − m − k ]. All sums have only a finite number of terms. Hence,
the interchange of the order of the summations is justified and will give correct results.
B) Assumptions: h[n] and g[n] are stable sequences, and x[n] is a bounded sequence

⎛ k2

with x[ n] ≤ B < ∞ . Here, y[ m] = ∑ ∞
k = −∞ h[ k ] x[ m − k ] = ⎜⎝ ∑ k = k1 h[ k ] x[ m − k ] ⎟⎠

+ε k1 , k 2 [ m] with ε k1 ,k 2 [ m ] ≤ ε n B. In this case, all sums have effectively only a finite
number of terms and the error ε k1 , k 2 [ m] can be reduced by choosing k1 and

k2 sufficiently large. As a result, in this case the problem is again effectively reduced
to that of the one-sided sequences. Thus, the interchange of the order of the
summations is again justified and will give correct results.
Hence, for the convolution to be associative, it is sufficient that the sequences be stable
and single-sided.
2.57

y[ n] = ∑ ∞
k = −∞ x[ n − k ]h[ k ]. Since h[k ] is of length M and defined for 0 ≤ k ≤ M − 1,
−1)
the convolution sum reduces to y[ n] = ∑ (kM
= 0 x[ n − k ]h[ k ]. y[n] will be nonzero for

all those values of n and for k which n − k satisfies 0 ≤ n − k ≤ N − 1. Minimum
value of n − k = 0 and occurs for lowest n at n = 0 and k = 0. Maximum value of
n − k = N − 1 and occurs for maximum value of k at M − 1. Thus n − k = M − 1
⇒ n = N + M − 2. Hence the total number of nonzero samples = N + M − 1.

Not for sale.

27

2.58

y[ n] = ∑ kN=−01x[ n − k ] x[ k ]. The maximum value of y[n] occurs at n = N − 1 when all

product terms are present. The maximum value is given by
y[ N − 1] = ∑ kN=−01a N −1− k a k .
2.59

y[ n] = ∑ kN=−01x[ n − k ]h[ k ]. The maximum value of y[n] occurs at n = N − 1 when all

product terms are present. The maximum value is given by
y[ N − 1] = ∑ kN=−01a N −1− k bk .
2.60

* hev [ n] = ∑ ∞
(a) y[ n] = g ev [ n] O
k = −∞ hev [ n − k ]g ev [ k ]. Now,

y[ −n] = ∑ ∞
k = −∞ hev [ − n − k ]g ev [ k ]. Replace k by − k. Then the summation on the left

becomes y[ − n] = ∑ ∞
k = −∞ hev [ − n + k ]g ev [ − k ] = ∑ k = −∞ hev [ −( n − k )]g ev [ − k ]
* hev [ n] is an even sequence.
= y[n]. Hence g ev [ n] O
* hod [ n] = ∑ ∞
(b) y[ n] = g ev [ n] O
k = −∞ hod [ n − k ]g ev [ k ]. Now,

y[ −n] = ∑ ∞
k = −∞ hod [ − n − k ]g ev [ k ] = ∑ k = −∞ hod [ − n + k ]g ev [ − k ]

= ∑∞
k = −∞ hod [ −( n − k )]g ev [ − k ] = − ∑ k = −∞ hod [ n − k ]g ev [ k ] = − y[ n].
* hod [ n] is an odd sequence.
Hence g ev [ n] O
* hod [ n] = ∑ ∞
(c) y[ n] = g od [ n] O
k = −∞ hod [ n − k ]g od [ k ]. Now,

y[ −n] = ∑ ∞
k = −∞ hod [ − n − k ]g od [ k ] = ∑ k = −∞ hod [ − n + k ]g od [ − k ]

= ∑∞
k = −∞ hod [ −( n − k )]g od [ − k ] = ∑ k = −∞ hod [ n − k ]g od [ k ] = y[ n].
* hod [ n] is an even sequence.
Hence g od [ n] O

2.61

The impulse response of the cascade is given by h[ n] = h1[ n] O
* h2 [ n] where

(

)

h1[ n] = α n µ[ n] and h2 [ n] = β n µ[ n]. Hence, h[ n] = ∑ nk = 0 α k β n − k µ[ n].
2.62

k

Now h[ n] = α n µ[ n]. Therefore y[ n] = ∑ ∞
k = −∞ h[ k ] x[ n − k ] = ∑ k = 0 α x[ n − k ]

k
k

= x[ n] + ∑ ∞
k =1 α x[ n − k ] = x[ n ] + α ∑ k = 0 α x[ n − 1 − k ] = x[ n] + α y[ n − 1].
Hence, x[ n] = y[ n] − αy[ n − 1]. Thus the inverse system is given by
y[ n] = x[ n] − αx[ n − 1]. The impulse response of the inverse system is given by
h[ n] = {1, α}, 0 ≤ n ≤ 1.

2.63

(

)

From the results of Problem 2.62 we have h[ n] = ∑ nk = 0 α k β n − k µ[ n]. Now,

Not for sale.

28

∞ ⎛ k
∞ ⎛ k

y[ n] = ∑ x[ n − k ]h[ k ] = ∑ ⎜⎜ ∑ α m β k − m ⎟⎟µ[ m]x[ n − k ] = ∑ ⎜⎜ ∑ α m β k − m ⎟⎟x[ n − k ]
k = −∞
k = −∞ ⎝ m = 0
k =0 ⎝ m =0

∞ ⎛ k

= x[ n] + ∑ ⎜⎜ ∑ α m β k − m ⎟⎟x[ n − k ]. Substituting r = k − 1 in the last expression we get
k =1 ⎝ m = 0

∞ ⎛ r
∞ ⎛ r +1

y[ n] = x[ n] + ∑ ⎜⎜ ∑ α m β r +1− m ⎟⎟x[ n − r − 1] = x[ n] + ∑ ⎜⎜ ∑ α m β r +1− m + α r +1 ⎟⎟x[ n − r − 1]
r =0 ⎝ m =0
r =0 ⎝ m =0

∞ ⎛ r

= x[ n] + β ∑ ⎜⎜ ∑ α m β r − m ⎟⎟x[ n − r − 1] + ∑ α r +1 x[ n − r − 1]
r =0 ⎝ m =0
r =0

= x[ n] + βy[ n − 1] + αx[ n − 1] + α 2 x[ n − 2] + α 3 x[ n − 3] + K. The inverse system is
therefore given by x[ n] = y[ n] − (α + β) y[ n − 1] + αβ y[ n − 2].
2.64

(a) h[ n] = h1[ n] O
* h2 [ n] O
* h3 [ n] O
* h3 [ n] + h1[ n] O
* h2 [ n] + h3 [ n] O
* h4 [ n].
(b) h[ n] = h4 [ n] +

2.65

h1[ n] O
* h2 [ n] O
* h3 [ n]
1 − h1[ n] O
* h2 [ n] O
* h5 [ n]

.

h[ n] = h1[ n] O
* h2 [ n] + h3 [ n]. Now
h1[ n] O
* h2 [ n] = (2δ[ n − 2] − 3δ[ n + 1]) O
* (δ[ n − 1] + 2δ[ n + 2])

= 2δ[ n − 2] O
* δ[ n − 1] − 3δ[ n + 1] O
* δ[ n − 1] + 2δ[ n − 2] O
* 2δ[ n + 2] − 3δ[ n + 1] O
* 2δ[ n + 2]

= 2δ[ n − 3] − 3δ[ n] + 4δ[ n] − 6δ[ n + 3] and
h3 [n] = 5δ[n − 5] + 7δ[n − 3] + 2δ[n − 1] − δ[n] + 3δ[n + 1]. Therefore,
h[ n] = 2δ[ n − 3] − 3δ[n] + 4δ[n] − 6δ[ n + 3] + 5δ[ n − 5] + 7δ[n − 3] + 2δ[ n − 1] − δ[ n] + 3δ[ n + 1]
= 5δ[n − 5] + 9δ[n − 3] + 2δ[n − 1] + 3δ[n + 1] − 6δ[n + 3].

2.66

3

1 ⎧
⎨ y[ n] − ∑ h[ k ]x[ n − k ]⎬ we
h[0] ⎩
k =0

(a) The length of x[n] is 8 − 4 + 1 = 5. Using x[ n] =
arrive at {x[ n]} = {3, − 2, 0, 1, 2}, 0 ≤ n ≤ 4.
(b) The length of x[n] is 7 − 4 + 1 = 4. Using x[ n] =

3

1 ⎧
⎨ y[ n] − ∑ h[ k ]x[ n − k ]⎬ we
h[0] ⎩
k =0

arrive at {x[ n]} = {1, 2, 3, 4}, 0 ≤ n ≤ 3.
(c) The length of x[n] is 8 − 5 + 1 = 4. Using x[ n] =

4

1 ⎧
⎨ y[ n] − ∑ h[ k ]x[ n − k ]⎬ we
h[0] ⎩
k =0

arrive at {x[ n]} = {1, − 2, 3, − 1}, 0 ≤ n ≤ 3.

Not for sale.

29

2.67

y[ n] = ay[ n − 1] + bx[ n]. Hence, y[0] = ay[ −1] + bx[ 0]. Next, y[1] = ay[0] + bx[1]

= a(ay[ −1] + bx[0]) + bx[1] = a 2 y[ −1] + abx[0] + bx[1]. Continuing further in a similar
way we obtained y[ n] = a n +1 y[ −1] + ∑ nk = 0 a n − k bx[ k ].
(a) Let y1[ n] be the output due to an input x1[ n]. Then

y1[ n] = a n +1 y[ −1] + ∑ nk = 0 a n − k bx1[ k ]. If x1[ n] = x[ n − no ], then
y1[ n] = a n +1 y[ −1] + ∑ nk = 0 a n − k bx[ k − no ] = a n +1 y[ −1] + ∑ rn=−0n o a n − n o − r bx[r ].
However,
y[ n − no ] = a n +1 y[ −1] + ∑ nk = 0 a n − k bx[ k − no ] = a n − n o +1 y[ −1] + ∑ rn=−0n o a n − n o − r bx[r ].
Hence y1[ n] ≠ y[ n − no ] if y[ −1] ≠ 0, i.e., the system is time-variant. The system is
time-invariant if and only if y[ −1] = 0, as then y1[ n] = y[ n − no ] .
(b) Let y1[ n] and y1[ n] be the outputs due to inputs x1[ n] and x1[ n] , respectively.
Let y[n] be the output due to an input αx1[ n] + βx 2 [ n]. However, αy1[ n] + βy2 [ n] =

α a n +1 y[ −1] + β a n +1 y[ −1] + α ∑ nk = 0 a n − k bx1[ k ] + β∑ nk = 0 a n − k bx 2 [ k ], whereas,
y[ n] = a n +1 y[ −1] + α ∑ nk = 0 a n − k bx1[ k ] + β∑ nk = 0 a n − k bx 2 [ k ]. Hence, the system is

nonlinear if y[ −1] ≠ 0 and is linear if and only if y[ −1] = 0.

(c) Generalizing the above result it can be shown that an N –th order causal discretetime system is linear and time-invariant if and only if y[ −r ] = 0,1 ≤ r ≤ N .
2.68

y[ n] = p0 x[ n] + p1 x[ n − 1] − d1 y[ n − 1] leads to x[ n] =

1
p0

y[ n] +

d1
p0

y[ n − 1] −

p1
p0

x[ n − 1],

which is the difference equation characterizing the inverse system. In other words, simply
solve the equation for x[n] in terms of present and past values of y[n] and x[n].
2.69

s[ n] = ∑ nk = 0 h[ k ]µ[ n − k ] = ∑ nk = 0 h[ k ], n ≥ 0, and s[ n] = 0, n < 0. Since h[k ] is

nonnegative, s[n] is a monotonically increasing function of n for n ≥ 0 , and is not
oscillatory. Hence, there is no overshoot.
2.70

(a) f [ n] = f [ n − 1] + f [ n − 2]. Let f [ n] = αr n , then the difference equation reduces to
αr n − αr n −1 − αr n − 2 = 0 which reduces further to r 2 − r − 1 = 0 resulting in

r=

n
n
1± 5
⎛ 1+ 5 ⎞
⎛ 1− 5 ⎞
+ α1 ⎜
.
. Thus, f [ n] = α1 ⎜

2
⎝ 2 ⎠
⎝ 2 ⎠

As f [ 0] = 0, hence α1 + α 2 = 0. Also f [1] = 1, , and hence

⎛⎜ α 1 + α 2 ⎞⎟ + 5 ⎛⎜ α 1 − α 2 ⎞⎟ = 1. Solving for α and α , we get α = −α = 1 . Hence,
1
2
1
2
5
⎝ 2 ⎠
⎝ 2 ⎠

Not for sale.

30

n

n

1 ⎛ 1− 5 ⎞
⎛ 1+ 5 ⎞
f [ n] =

⎟ − ⎜
⎟ .
5⎝ 2 ⎠
5⎝ 2 ⎠
1

(b) y[ n] = y[ n − 1] + y[ n − 2] + x[ n − 1]. As the system is LTI, the initial conditions are
equal to zero. Let x[ n] = δ[ n]. Then y[ n] = y[ n − 1] + y[ n − 2] + δ[ n − 1]. Hence,
y[0] = y[ −1] + y[ −2] = 0 and y[1] = y[ 0] + y[ −2] + δ[ 0] = 1. For n > 1, the
corresponding difference equation is y[ n] = y[ n − 1] + y[ n − 2] with initial conditions
y[ 0] = 0 and y[1] = 1, which are the same as those for the solution of the Fibonacci’s
n

n

1 ⎛ 1− 5 ⎞
⎛ 1+ 5 ⎞
sequence. Hence y[ n] =

⎟ − ⎜
⎟ . Thus denotes the impulse
5⎝ 2 ⎠
5⎝ 2 ⎠
response of a causal LTI system described by the difference equation
y[ n] = y[ n − 1] + y[ n − 2] + x[ n − 1].
1

2.71 y[ n] = αy[ n − 1] + x[ n]. Denoting y[ n] = yre [ n] + j yim [ n], and α = a + j b, we get
yre [ n] + j yim [ n] = (a + j b)( yre [ n − 1] + j yim [ n − 1]) + x[ n]. Equating the real and the
imaginary parts, and noting that x[n] is real, we get
yre [ n] = ayre [ n − 1] − byim [ n − 1] + x[ n], yim [ n] = byre [ n − 1] + ayim [ n − 1]. From the
second equation we have yim [ n − 1] =

1
b
yim [ n] − yre [ n − 1].
a
a

Substituting this

equation in the top left equation we arrive at
b
a

yre [ n] = ayre [ n − 1] − yim [ n] +

b2
a

yre [ n − 1] + x[ n], from which we get

byim [ n − 1] = − ayre [ n − 1] + ( a 2 + b 2 ) yre [ n − 2] + ax[ n − 1]. Substituting this equation

in the equation yre [ n] = ayre [ n − 1] − byim [ n − 1] + x[ n] we arrive at
yre [ n] = 2 ayre [ n − 1] − ( a 2 + b 2 ) yre [ n − 2] + x[ n] − ax[ n − 1] which is a second-order

difference equation representing yre [n] in terms of x[n].
2.72

The first-order causal LTI system is characterized by the difference equation
y[ n] = p0 x[ n] + p1 x[ n − 1] − d1 y[ n − 1]. Letting x[ n] = δ[ n] we obtain the difference
equation representation of its impulse response h[ n] = p0 δ[ n] + p1δ[ n − 1] − d1h[ n − 1].
Solving it for n = 0,1, 2, we get h[0] = p0 , h[1] = p1 − d1h[0] = p1 − d1 p0 , and
h[2] = −d1h[1] = − d1 p0 ( p1 − d1 p0 ). Solving these equations we get p0 = h[0],
d1 = −

2.73

h[ 2 ]
,
h[1]

and p1 = h[1] −

h[ 2 ]h[ 0 ]
.
h[1]

M

N

M

N

k =0

k =0
N d h[r
∑k =0 k

k =0

k =0

∑ pk x[ n − k ] = ∑ d k y[ n − k ]. Let x[ n] = δ[ n]. Then ∑ p k δ[ n − k ] = ∑ d k h[ n − k ].

Thus, pr =

− k ]. Since the system is assumed to be causal, h[r − k ] = 0

Not for sale.

31

for all k > r. Hence, pr = ∑ kN= 0 d k h[r − k ] = ∑ kN= 0 h[ k ]d r − k .
2.74

For a filter with a complex-valued impulse response, the first part of the proof is the
same as that for a filter with a real-valued impulse response. From

y[ n] = ∑ ∞
k = −∞ h[ k ]x[ n − k ] we get y[ n] = ∑ k = −∞ h[ k ]x[ n − k ] ≤ ∑ k = −∞ h[ k ] x[ n − k ] .

Since the input is bounded x[ n] ≤ B x . Therefore y[ n] ≤ B x ∑ ∞
k = −∞ h[ k ] . So if
∑∞
k = −∞ h[ k ] = S < ∞, then y[ n] ≤ B x S indicating that y[n] is also bounded.

To prove the converse we need to show that if a bounded input is produced by a
bounded input then S < ∞. Consider the following bounded input defined by
∞ h * [ − k ]h[ k ]

h * [ − n]
. Then y[ n] = ∑
x[ n] =
= ∑ h[ k ] = S. Now since the output
h[ −n]
h[ k ]
k = −∞
k = −∞
is bounded, S < ∞. Thus for a filter with a complex impulse response is BIBO stable if
and only if ∑ ∞
k = −∞ h[ k ] = S < ∞.
2.75

The impulse response of the cascade is g[ k ] = ∑ ∞
k = −∞ h1[ k − r ]h 2 [r ]. Thus

⎞⎛ ∞
⎛ ∞
∑ h1[ k − r ] h2 [r ] ≤ ⎜⎜ ∑ h1[ k ] ⎟⎟ ⎜⎜ ∑ h2 [r ] ⎟⎟. Since
k = −∞
k = −∞ r = −∞

⎠ ⎝ r = −∞
⎝ k = −∞
h1[ n] and h2 [ n] are stable, ∑ h1[ k ] < ∞ and ∑ h2 [ k ] < ∞. Hence ∑ g[ k ] < ∞ and as a

∑ g[ k ] = ∑

k

k

k

result, a cascade of two stable LTI systems is also stable.
2.76

The impulse response of the parallel structure is g[ n] = h1[ n] + h2 [ n]. Now,

k = −∞

k = −∞

k = −∞

k = −∞

∑ g[ k ] = ∑ h1[ k ] + h2 [ k ] ≤ ∑ h1[ k ] + ∑ h2 [ k ] . Since h1[ n] and

h2 [ n] are stable, ∑ h1[ k ] < ∞ and ∑ h2 [ k ] < ∞. Hence ∑ g[ k ] < ∞ and as a result, a
k

k

k

parallel connection of two stable LTI systems is also stable.
2.77

Consider a cascade connection of two passive LTI systems with an input x[n] and an
output y[n]. Let y1[ n] and y2 [ n] be the outputs of the two systems for the input

x[n]. Now ∑ ∞
n = −∞ y1[ n] ≤ ∑ n = −∞ x[ n] and ∑ n = −∞ y2 [ n] ≤ ∑ n = −∞ x[ n] . Let
y1[ n] = y2 [ n] = x[ n] satisfying the above inequalities. Then y[ n] = y1[ n] + y2 [ n]
2

2

2

2

2

2

2

= 2 x[ n] and as a result, ∑ ∞
n = −∞ y[ n] = 4∑ n = −∞ x[ n] > ∑ n = −∞ x[ n] . Hence, the
parallel connection of two passive LTI systems may not be passive.

2.78

Consider a parallel connection of two passive LTI systems with an input x[n] and an
output y[n]. Let y1[ n] and y2 [ n] be the outputs of the two systems for the input
Not for sale.

32

x[n]. Now ∑ ∞
n = −∞ y1[ n] ≤ ∑ n = −∞ x[ n] and ∑ n = −∞ y2 [ n] ≤ ∑ n = −∞ x[ n] . Let
y1[ n] = y2 [ n] = x[ n] satisfying the above inequalities. Then y[ n] = y1[ n] + y2 [ n]
2

2

2

2

2

2

2

= 2 x[ n] and as a result, ∑ ∞
n = −∞ y[ n] = 4∑ n = −∞ x[ n] > ∑ n = −∞ x[ n] . Hence, the
parallel connection of two passive LTI systems may not be passive.

2.79

Let the difference equation ∑ kM= 0 p k x[ n − k ] = y[ n] + ∑ kN=1 d k y[ n − k ] represents the
causal IIR digital filter. For an input x[ n] = δ[ n], the corresponding output is then
y[ n] = h[ n], the impulse response of the filter. As the number of coefficients { p k } is
M + 1 and the number of coefficients {d k } is N , there are a total of N + M + 1
unknowns. To determine these coefficients from the impulse response samples, we
compute only the first N + M + 1 impulse response samples. To illustrate the method,
without any loss of generality, we assume N = M = 3. Then, from the difference
equation we arrive at the following N + M + 1 = 7 equations:
h[0] = p0 ,

h[1] + h[0]d1 = p1 ,
h[2] + h[1]d1 + h[0]d 2 = p2 ,
h[3] + h[2]d1 + h[1]d 2 + h[0]d 2 = p2 ,
h[ 4] + h[3]d1 + h[2]d 2 + h[1]d 2 = 0,
h[5] + h[ 4]d1 + h[3]d 2 + h[2]d 2 = 0,
h[6] + h[5]d1 + h[ 4]d 2 + h[3]d 2 = 0.
Writing the last three equations in matrix form we arrive at
−1
⎡ d1 ⎤
⎡ h[3] h[2] h[1] ⎤ ⎡h[ 4]⎤
⎡h[ 4]⎤ ⎡ h[3] h[2] h[1] ⎤ ⎡ d1 ⎤ ⎡0⎤
⎢ h[5]⎥ = ⎢h[ 4] h[3] h[2]⎥ ⎢d 2 ⎥ = ⎢0⎥, and hence, ⎢d 2 ⎥ = − ⎢h[ 4] h[3] h[2]⎥ ⎢ h[5]⎥.
⎢d ⎥
⎢⎣ h[5] h[ 4] h[3]⎥⎦ ⎢⎣h[6]⎥⎦
⎢⎣h[6]⎥⎦ ⎢⎣ h[5] h[ 4] h[3]⎥⎦ ⎢ d ⎥ ⎢⎣0⎥⎦
⎣ 3⎦
⎣ 3⎦
Substituting these values in the first four equations written in matrix form we get
⎡ p0 ⎤ ⎡h[0] 0
0
0 ⎤⎡ 1 ⎤
⎢ p1 ⎥ ⎢ h[1] h[0] 0
0 ⎥ ⎢ d1 ⎥
⎢ p ⎥ = ⎢h[2] h[1] h[0] 0 ⎥ ⎢d ⎥.
2
⎢ 2⎥ ⎢
⎢d ⎥

[
3
]
[
2
]
[
1
]
[
0
]
h
h
h
h
⎦⎣ 3 ⎦
⎣ p3 ⎦ ⎣

2.80

y[ n] = y[ −1] + ∑ ln= 0 x[l] =y[ −1] + ∑ ln= 0 lµ[l] =y[ −1] + ∑ ln= 0 l =y[ −1] +
(a) For y[ −1] = 0, y[ n] =

n( n +1)
.
2

n( n +1)
.
2

(b) For y[ −1] = −2, y[ n] = −2 +

n( n +1)
2

=

n2 +n−4
.
2

Not for sale.

33

2.81

y(nT ) = y((n − 1)T ) + ∫(nT
n −1)T x ( τ)dτ =y(( n − 1)T ) + T ⋅ x (( n − 1)T ). Therefore, the

difference equation representation is given by y[ n] = y[ n − 1] + T ⋅ x[ n − 1] where
y[ n] = y( nT ) and x[ n] = x ( nT ).
2.82

y[ n] = ∑ ln=1 x[l] = ∑ ln=−11 x[l] + x[ n], n ≥ 1. Now y[ n − 1] =
1
n

1
n

1
n

1
∑ n −1 x[l], n ≥ 1,
n −1 l =1

i.e,

∑ ln=−11 x[l] = (n − 1) y[ n − 1]. Thus, the difference equation representation is given by
1
n −1 ⎞
y[ n] = ⎛⎜
⎟ y[ n − 1] + x[ n].
n
⎝ n ⎠

2.83

y[ n] − 0.35 y[ n − 1] = 2.4µ[ n] with y[ −1] = 3. The total solution is given by
y[ n] = yc [ n] + y p [ n], where yc [n] is the complementary solution and y p [n] is the

particular solution.

yc [n] is obtained by solving yc [ n] − 0.35 yc [ n − 1] = 0. To this end we set y c [ n] = λn ,
which yields λn − 0.35λn −1 = 0 resulting in the solution λ = 0.35. Hence
y c [ n] = α(0.35) n .

For the particular solution we choose y p [ n] = β. Substituting this solution in the
difference equation representing the system we get β − 0.35β = 2.4µ[ n]. For n = 0 we
get β − 0.35β = 2.4 , i.e., (1 − 0.35)β = 2.4 and hence β = 2.4 / 0.65 = 48 / 13.
Therefore y[ n] = yc [ n] + y p [ n] = α(0.35) n +

y[ −1] = 3 = α(0.35) −1 +
y[ n] =
2.84

48
implying
13
48
−0.2423(0.35) n + , n ≥ 0.
13

48
,
13

n ≥ 0. For n = −1, we thus have

α = −0.2423. The total solution is thus given by

y[ n] − 0.3 y[ n − 1] − 0.04 y[ n − 2] = 3 n µ[ n] with y[ −1] = 2 and y[ −2] = 1. The total
solution is given by y[ n] = yc [ n] + y p [ n], where yc [n] is the complementary solution
and y p [n] is the particular solution.

yc [n] is obtained by solving yc [ n] − 0.3 yc [ n − 1] − 0.04 yc [ n − 2] = 0. To this end we
set y c [ n] = λn , which yields λn − 0.3λn −1 − 0.04 λn − 2 = 0 resulting in the solutions

λ = 0.4 or λ = −0.1. Hence y c [ n] = α1 (0.4) n + α 2 ( −0.1) n .
For the particular solution we choose y p [ n] = β(3) n . Substituting this solution in the
difference equation representing the system we get

β(3) n − 0.3β(3) n −1 − 0.04β(3) n − 2 = 3 n µ[ n]. For n = 0 we have
β − 0.3β(3) −1 − 0.04β(3) −2 = 1 which yields β = 1.1166 . Therefore

Not for sale.

34

y[ n] = yc [ n] + y p [ n] = α1 (0.4) n + α 2 (−0.1) n + 1.1166(3) n , n ≥ 0. For n = −1 and

n = −2 we thus have y[ −1] = α1 (0.4) −1 + α 2 (−0.1) −1 + 1.1166(3) −1 = 2 and
y[ −2] = α1 (0.4) −1 2 + α 2 (−0.1) −2 + 1.1166(3) −2 = 1. Solving these two equations we
get α1 = 0.5489 and α 2 = −0.0255. Hence,
y[ n] = 0.5489(0.4) n − 0.0255(−0.1) n + 1.1166(3) n , n ≥ 0.
2.85

y[ n] − 0.3 y[ n − 1] − 0.04 y[ n − 2] = x[ n] + 2 x[ n − 1] with x[ n] = 3 n µ[ n], y[ −1] = 2 and
y[ −2] = 1. The total solution is given by y[ n] = yc [ n] + y p [ n], where yc [n] is the

complementary solution and y p [n] is the particular solution. From the solution of
Problem 2.84, the complementary solution is of the form y c [ n] = α1 (0.4) n + α 2 ( −0.1) n .
To determine y p [n] we observe that the it is given by the sum of the particular solution

y p1[ n] of the difference equation y1[ n] − 0.3 y1[ n − 1] − 0.04 y1[ n − 2] = x[ n] = 3 n µ[ n]
and the particular solution y p2 [ n] of the difference equation

y2 [ n] − 0.3 y2 [ n − 1] − 0.04 y2 [ n − 2] = 2 x[ n − 1] = 2 ⋅ 3 n −1 µ[ n − 1]. From the solution of
Problem 2.84, we have y p1[ n] = β(3) n . Hence, y p2 [ n] = 2 y p1[ n − 1] = 2β(3) n −1 .
Therefore, y p [ n] = y p1[ n] + 2 y p2 [ n] = β(3) n + 2β(3) n −1 = 3 n µ[ n] + 2 ⋅ 3 n −1 µ[ n − 1].
For n = 1 the above equation reduces to 3β + 2β = 3 + 2. Thus, β = 1. Therefore, the
total solution is given by

y[ n] = yc [ n] + y p [ n] = α1 (0.4) n + α 2 (−0.1) n + (3) n + 2(3) n −1 , n ≥ 0. For n = −1 and

n = −2 we thus have y[ −1] = α1 (0.4) −1 + α 2 (−0.1) −1 + (3) −1 + 2(3) −2 = 2 and
y[ −2] = α1 (0.4) −2 + α 2 (−0.1) −2 + (3) −2 + 2(3) −3 = 1. Solving these two equations we
get α1 = 0.4883 and α 2 = −0.0224. Hence,
y[ n] = 0.4883(0.4) n − 0.0224(−0.1) n + (3) n + 2(3) n −1 , n ≥ 0.
2.86

h[ n] − 0.35h[ n − 1] = δ[ n]. The solution is given by h[ n] = hc [ n] + h p [ n], where hc [n]
is the complementary solution and h p [n] is the particular solution. If h[n] is the
impulse response, then h p [ n] = 0. From Problem 2.83 we note that hc [ n] = α(0.35) n .
Thus, h[0] − 0.35h[ −1] = h[0] = 1. This implies α = 1. Hence, h[ n] = (0.35) n , n ≥ 0.

2.87

The overall system can be regarded as the cascade of two causal LTI systems:
S1: y[n] − 0.3 y[n − 1] − 0.04 y[n − 2] = x1[n] and S2: x1[n] = x[n] + 2 x[n − 1] .

Not for sale.

35

The impulse response h1[ n] of the system S1 can be found by solving the
complementary solution of h1[ n] − 0.3h1[ n − 1] − 0.04h1[ n − 2] = δ[ n] . Let the
complementary solution be h1c [ n] = λn , we have λn − 0.3λn −1 − 0.04λn − 2 = 0
hence λ = {0.4, − 0.1} . Therefore, the impulse response h1[ n] is given by
h1[ n] = h1c [ n] = A(0.4) n + B( −0.1) n , n ≥ 0 . Solving constants A, B, we get A = 0.8

and B = 0.2. Hence h1[ n] = 0.8(0.4) n + 0.2(−0.1) n , n ≥ 0. .
The impulse response h2 [ n] of the system S2 is given by h2 [ n] = δ[ n] + 2δ[ n − 1]. .
The impulse response h[n] of the overall system is

(

)

(

)

h[ n] = h1[ n] * h2 [ n] = 0.8(0.4) n + 0.2(−0.1) n µ[ n] + 2 0.8(0.4) n −1 + 0.2(−0.1) n −1 µ[ n − 1]
= δ[ n] + 1.92(0.4)
2.88

n −1

µ[ n − 1] + 0.38(−0.1)

n −1

µ[ n − 1].

h[ n] = (−α) n µ[ n], 0 < α < 1. Step response is then given by s[ n] = h[ n] O
* µ[ n]

⎧∑ n (−α) k , n ≥ 0,
k
= (−α) n µ[ n] O
* µ[ n] = ∑ (−α) µ[ k ]µ[ n − k ] = ⎨ k = 0
0,
n<0
k = −∞

⎧⎪1− ( − α ) n +1
= ⎨ 1+ α , n ≥ 0,
⎪⎩
0,
n < 0.

2.89

K

K

n +1
= 1. Since there
λ i . Now lim
n→∞ n
1 + λi
A
< 1. Hence
exists a positive integer N o such that for all n > N o , 0 < n +1 <
2
An
A
n +1
Let An = n (λ i ) . Then n +1 =
An
n
K

n

∑∞
n =0 An converges.
2.90

{x[ n]} = {− 4, 5, 1, − 2, − 3, 0, 2}, − 3 ≤ n ≤ 3,

{y[ n]} = {6, − 3, − 1, 0, 8, 7, − 2}, − 1 ≤ n ≤ 5,
{w[ n]} = {3, 2, 2, − 1, 0, − 2, 5}, 2 ≤ n ≤ 8.
(a) rxx [l] = ∑ 3n = −3 x[ n]x[ n − l], − 6 ≤ l ≤ 6.
{rxx [l]} = {− 8, 10, 14, − 11, − 23, − 11, 59, − 11, − 23, − 11, 14, 10, − 8}, − 6 ≤ l ≤ 6,

ryy [l] = ∑ 5n = −5 y[ n]y[ n − l], − 6 ≤ l ≤ 6.
{ryy [l]} = {− 12, 48, 29, − 31, − 30, 27, 163, 27, − 30, − 31, 29, 48, − 12}, − 6 ≤ l ≤ 6,
rww [ l] = ∑ 6n = −6 w[ n]w[ n − l], − 6 ≤ l ≤ 6.

{rww [l]} = {15, 4, 6, − 12, 6, − 2, 47, − 2, 6, − 12, 6, 4, 15}, − 6 ≤ l ≤ 6,
Not for sale.

36

(b) rxy [l] = ∑ 3n = −5 x[ n]y[ n − l], − 8 ≤ l ≤ 4.

{rxy [l]} = {8, − 38, 1, 51, 4, − 30, − 68, 43, 31, − 3, − 20, − 6, 12}, − 8 ≤ l ≤ 4,
rxw [l] = ∑ 3n = −8 x[ n]y[ n − l], − 11 ≤ l ≤ 1.
{rxw [l]} = {− 20, 33, − 5, − 8, − 24, 7, 12, 12, − 7, − 14, − 5, 4, 6}, − 11 ≤ l ≤ 1,
2.91

n
n−l

(a) x1[ n] = α n µ[ n]. rx 1 x1 [l] = ∑ ∞
µ[ n − l]
n = −∞ x1[ n]x1[ n − l] = ∑ n = −∞ α µ[ n] α

⎧ α −l
, l < 0,
2n − l

, l < 0, ⎪⎪1− α 2
⎪∑ n = 0 α
2n − l

= ∑n=0 α
µ[ n − l] = ⎨
=⎨ l
2n − l
⎪⎩∑ ∞
, l ≥ 0, ⎪ α , l ≥ 0.
n=l α
⎪⎩1− α 2

Note for l ≥ 0, rx1 x1 [l] =

αl
1− α 2

, and for l < 0, rx1 x1 [l] =

in the second expression we get rx1 x1 [ −l] =

α −( −l )
1− α 2

=

α −l
1− α 2

αl
1− α 2

. Replacing

l with − l

= rx x [l]. Hence, rx x [l]
1 1
1 1

is an even function of l. Maximum value of rx1 x1 [l] occurs at l = 0 since α l is a
decaying function for increasing when α < 1.

1, 0 ≤ n ≤ N − 1,
Now rx 2 x 2 [l] = ∑ nN=−01x 2 [ n − l], where
(b) x 2 [ n] = ⎧⎨
0
,
otherwise.

for l < −( N − 1),
⎧ 0,
⎪ N + l, for − ( N − 1) ≤ l ≤ 0,

1, l ≤ n ≤ N − 1 + l,
x 2 [ n − l] = ⎧⎨
Therefore, rx 2 x 2 [l] = ⎨ N ,
for l = 0,
0
,
otherwise.

⎪ N − l, for 0 < N − l ≤ N − 1,
⎪⎩ 0,
for l > N − 1.
It follows from the above that rx 2 x2 [l] is a triangular function of l , and hence is an
even function with a maximum value of N at l = 0.
2.92

πn
(a) x1[ n] = cos⎛⎜ ⎞⎟ where M is a positive integer. Period of x1[ n] is 2 M , and
⎝M⎠
1 2 M −1
1 2 M −1 ⎛ πn ⎞ ⎛ π(n + l) ⎞
hence rx1 x1 [l] =

∑ x 2 [ n] x 2 [ n + l ] =
∑ cos⎜ ⎟ cos⎜
2M n =0
2M n =0
⎝M⎠ ⎝ M ⎠

=

1 2 M −1 ⎛ πn ⎞⎧ ⎛ πn ⎞ ⎛ πl ⎞
⎛ πn ⎞ ⎛ πl ⎞⎫
∑ cos⎜ ⎟⎨cos⎜ ⎟ cos⎜ ⎟ − sin⎜ ⎟ sin⎜ ⎟⎬
2M n =0
⎝ M ⎠ ⎝ M ⎠⎭
⎝ M ⎠⎩ ⎝ M ⎠ ⎝ M ⎠

=

1
⎛ πn ⎞
⎛ πl ⎞2 M −1
cos⎜ ⎟ ∑ cos 2 ⎜ ⎟.
2M
⎝M⎠
⎝ M ⎠ n=0

Not for sale.

37

Now
2 M −1

N −1
N −1 ⎧
⎛ 4 πn ⎞⎫ N 1 N −1 ⎛ 4 πn ⎞
2 ⎛ πn ⎞
2 ⎛ 4 πn ⎞ 1
∑ cos ⎜ ⎟ = ∑ cos ⎜
⎟ = ∑ ⎨1 + cos⎜
⎟⎬ = + ∑ cos⎜
⎟.
⎝ M ⎠ n=0
⎝ N ⎠ 2 n=0 ⎩
⎝ N ⎠⎭ 2 2 n = 0 ⎝ N ⎠
n=0
N −1
N −1 ⎛ 4 πn ⎞
N −1
⎛ 4πn ⎞
j ( 4 πn / N )
Let C = ∑ cos⎜
⎟ and S = ∑ sin⎜
⎟. Then C + jS = ∑ e
⎝ N ⎠
⎝ N ⎠
n=0
n=0
n=0

e j 4π − 1

=
e

j 4π / N

−1

= 0. This implies C = 0. Thus

Hence, rx1 x1 [l] =

2 M −1

2 ⎛ πn ⎞ N
∑ cos ⎜ ⎟ = = M .
⎝M⎠ 2
n=0

M
⎛ πl ⎞
⎛ πl ⎞ 1
cos⎜ ⎟ = cos⎜ ⎟.
2M
⎝M⎠
⎝M⎠ 2

(b) {x 2 [ n]} = 〈 n〉 6 = {0, 1, 2, 3, 4, 5}, 0 ≤ n ≤ 5. It is a periodic sequence with a
1

period 6. Thus, rx 2 x 2 [l] = ∑ 5n = 0 x 2 [ n]〈 x 2 [ n + l]〉 6 , 0 ≤ l ≤ 5. rx 2 x2 [l] is also a
6
periodic sequence with a period 6.

(

)

1
55
x [0]x 2 [0] + x 2 [1]x 2 [1] + x 2 [2]x 2 [2] + x 2 [3]x 2 [3] + x 2 [ 4]x 2 [ 4] + x 2 [5]x 2 [5] = ,
6 2
6
40
1
rx2 x2 [1] = x 2 [0]x 2 [1] + x 2 [1]x 2 [2] + x 2 [2]x 2 [3] + x 2 [3]x 2 [ 4] + x 2 [ 4]x 2 [5] + x 2 [5]x 2 [0] = ,
6
6
1
32
rx2 x2 [2] = x 2 [0]x 2 [2] + x 2 [1]x 2 [3] + x 2 [2]x 2 [ 4] + x 2 [3]x 2 [5] + x 2 [ 4]x 2 [0] + x 2 [5]x 2 [1] = ,
6
6
28
1
rx2 x2 [3] = x 2 [0]x 2 [3] + x 2 [1]x 2 [ 4] + x 2 [2]x 2 [5] + x 2 [3]x 2 [0] + x 2 [ 4]x 2 [1] + x 2 [5]x 2 [2] = ,
6
6
1
31
rx2 x2 [ 4] = x 2 [0]x 2 [ 4] + x 2 [1]x 2 [5] + x 2 [2]x 2 [0] + x 2 [3]x 2 [1] + x 2 [ 4]x 2 [2] + x 2 [3]x 2 [ 4] = ,
6
6
40
1
rx2 x2 [5] = x 2 [0]x 2 [5] + x 2 [1]x 2 [0] + x 2 [2]x 2 [1] + x 2 [3]x 2 [2] + x 2 [ 4]x 2 [3] + x 2 [5]x 2 [ 4] = .
6
6

rx 2 x 2 [0] =

(

)

(

)

(

)

(

)

(

)

(c) x 3 [ n] = ( −1) n is a periodic sequence with a period 2. Thus,
1
2
1
2

rx 3 x 3 [l] = ∑1n = 0 x3 [ n]x3 [ n + l], 0 ≤ l ≤ 1. Hence,
rx 3 x 3 [0] =

(x3 [0]x3 [0] + x3 [1]x3 [1]) = 1,

rx 3 x 3 [1] =

1
2

(x3 [0]x3 [1] + x3 [1]x3 [0]) = −1.

rx3 x3 [l] is also a periodic sequence with a period 2.
M2.1 (a) The input data entered during the execution of Program 2_2.m are:
Type in real exponent = -1/12
Type in imaginary exponent = pi/6
Type in gain constant = 1
Type in length of sequence = 41
(b) The input data entered during the execution of Program 2_2.m are:
Type in real exponent = -1/12
Type in imaginary exponent = pi/6
Type in gain constant = 1
Type in length of sequence = 41
Not for sale.

38

Real part

Imaginary part

0.5

0
Amplitude

Amplitude

0
-0.5
-1

-0.5
-1
-1.5

-1.5
0

10

20
Time index n

30

40

0

5

10
15
20
Time index n

25

30

M2.2 (a) ~
x a [ n] = e − j 0.4 πn . The plots generated using Program 2_2.m are shown below:
Imaginary part

Real part

1

1

0.5
Amplitude

Amplitude

0.5
0

0
-0.5

-0.5

-1

-1
0

10

20
Time index n

30

0

40

10

20
Time index n

30

40

(b) The code fragment used to generate ~
x b [ n] = sin(0.8πn + 0.8π) is as follows:
x = sin(0.8*pi*n + 0.8*pi);
The plot of the periodic sequence is given below:
1

Amplitude

0.5
0
-0.5
-1

0

10

20
n

30

(

40

) (

)

(c) The code fragment used to generate ~
x c [ n] = Re e jπn / 5 + Im e jπn / 10 is as
follows:
x = real(exp(i*pi*n/5)+ imag(exp(i*pi*n/10);
The plot of the periodic sequence is given below:

Not for sale.

39

2

Amplitude

1
0
-1
-2

0

10

20
n

30

40

(d) The code fragment used to generate ~
x d [ n] = 3 cos(1.3πn) − 4 sin(0.5πn + 0.5π) is as
follows:
x = 3*cos(1.3*pi*n)-4*sin(0.5*pi*n+0.5*pi);
The plot of the periodic sequence is given below:
10

Amplitude

5
0
-5
-10

0

10

20
n

30

40

(e) The code fragment used to generate
~
x e [ n] = 5 cos(1.5πn + 0.75π) + 4 cos(0.6πn) − sin(0.5πn) is as follows:
x = 5*cos(1.5*pi*n+0.75*pi)+4*cos(0.6*pi*n)-sin(0.5*pi*n);
The plot of the periodic sequence is given below:
10

Amplitude

5
0
-5
-10

0

10

20
n

30

40

M2.3 (a) L = input('Desired length = ');
A = input('Amplitude = ');
omega = input('Angular frequency = ');
phi = input('Phase = ');
n = 0:L-1;

Not for sale.

40

x = A*cos(omega*n + phi);
stem(n,x);
xlabel('Time Index'); ylabel('Amplitude');
title(['\omega_{o} = ',num2str(omega/pi),'\pi']);
(b)

ωo = 0.28π
1.5

1

1

0.5

0.5

Amplitude

Amplitude

ωo = 0.6π
1.5

0
-0.5

0
-0.5

-1
0

10

20
Time Index
ωo = 0.45π

30

-1.5

40

1.5

1

1

0.5

0.5

0
-0.5
-1
-1.5

0

10

20
Time Index

30

40

ωo = 0.55π

1.5

Amplitude

Amplitude

-1.5

-1

0
-0.5
-1

0

10

20
Time Index

30

-1.5

40

0

10

20
Time Index

30

40

ωo = 0.65π
1.5

Amplitude

1
0.5
0
-0.5
-1
-1.5

0

10

20
Time Index

30

40

M2.4 t = 0:0.001:1;
fo = input('Frequency of sinusoid in Hz = ');
FT = input('Sampling frequency in Hz = ');
g1 = cos(2*pi*fo*t);
plot(t,g1,'-');
xlabel('time'); ylabel('Amplitude'); hold
n = 0:1:FT;
gs = cos(2*pi*fo*n/FT);
plot(n/FT,gs,'o'); hold off
M2.5 t = 0:0.001:0.85;
Not for sale.

41

g1 = cos(6*pi*t); g2 = cos(14*pi*t); g3 =
cos(26*pi*t);
plot(t/0.85,g1,'-', t/0.85, g2, '--', t/0.85, g3,':');
xlabel('time'); ylabel('Amplitude'); hold
n = 0:1:8; gs = cos(0.6*pi*n); plot(n/8.5,gs,'o');
hold off
M2.6 As the length of the moving average filter is increased, the output of the filter gets more
smoother. However, the delay between the input and the output sequences also
increases (This can be seen from the plots generated by Program 2_4.m for various
values of the filter length.)
M2.7 alpha = input('Alpha = ');
y0 = 1; y1 = 0.5*(y0 + (alpha/y0));
while abs(y1-y0)>0.00001
y2 = 0.5*(y1+(alpha/y1));
y0 = y1; y1 = y2;
end
disp('Squre root of alpha is'); disp(y1);
M2.8 format long
alpha = input('Alpha = ');
y0 = 0.3; y = zeros(1,61);
L = length(y) - 1;
y(1) = alpha - y0*y0 + y0; n = 2;
while abs(y(n-1) - y0) > 0.00001
y2 = alpha - y(n-1)*y(n-1) + y(n-1);
y0 = y(n-1); y(n) = y2;
n = n+1;
end
disp('Square root of alpha is');disp(y(n-1));
m = 0:n-2;
err = y(1:n-1) - sqrt(alpha);
stem(m,err);
axis([0 n-2 min(err) max(err)]);
xlabel('Time index n'); ylabel('Error');
title(['\alpha = ',num2str(alpha)]);
The displayed out is
Square root of alpha is
0.84178104293115

Not for sale.

42

α = 0.7086
0.06

Error

0.04
0.02
0
-0.02
-0.04
0

5

10
15
Time index n

20

25

M2.9 {x[ n]} = {− 4, 5, 1, − 2, − 3, 0, 2}, − 3 ≤ n ≤ 3,
{y[ n]} = {6, − 3, − 1, 0, 8, 7, − 2}, − 1 ≤ n ≤ 5,
{w[ n]} = {3, 2, 2, − 1, 0, − 2, 5}, 2 ≤ n ≤ 8.
{rxx [ n]} = {− 8, 10, 14, −11, − 23, −11, 59, −11, − 23, −11, 14, 10, − 8}, − 6 ≤ n ≤ 6.
{ryy [ n ]} = {−12, 48, 29, − 31, − 30, − 27, 163, − 27, − 30, − 31, 29, 48, −12}, − 6 ≤ n ≤ 6.

{rww [ n]} = {15, 4, 6, −12, 6, − 2, 47, − 2, 6, −12, 6, 10, 15}, − 6 ≤ n ≤ 6.
ryy [n]

rxx [n]

200

60

150
Amplitude

Amplitude

40
20
0

50
0

-20
-40
-6

100

-4

-2

0
Lag index

2

4

-50
-6

6

-4

rww[n]

-2

0
Lag index

2

4

6

rxy [n]

60

60
40
20
Amplitude

Amplitude

40
20

0
-20
-40

0

-60
-20
-6

-4

-2

0
Lag index

2

4

6

Not for sale.

-80

-5

0
Lag index

5

43

rxw[n]

ryw[n]

40

60
40
Amplitude

Amplitude

20
0
-20
-40
-6

20
0
-20

-4

-2

0
Lag index

2

4

-40
-6

6

-4

-2

0
Lag index

2

4

6

M2.10 N = input('Length of sequence = ');
n = 0:N-1;
x = exp(-0.8*n);
y = rand(1,N)-0.5+x;
n1 = length(x)-1;
r = conv(y,fliplr(y));
k = (-n1):n1;
stem(k,r);
xlabel('Lag_index'); ylabel('Amplitude');
3

Amplitude

2
1
0
-1
-30

-20

-10

0
10
Lag index

Not for sale.

20

30

44

Chapter 3
3.1

−∞

−∞

X a ( jΩ) = ∫ x a (t ) e − jΩt dt. Now, ∫ x a (t ) dt < ∞. Hence,

X a ( jΩ) = ∫ x a (t ) e − jΩt dt ≤ ∫ x a (t ) e − jΩt dt
−∞

−∞

3.2

≤ ∫ x a (t ) dt < ∞.

1 −∞

−∞

(

)

(a) Ya ( jΩ) = ∫ cos(Ω o t ) e − jΩt dt = ∫ e jΩot + e − jΩot e − jΩt dt
2 −∞
−∞
=

1
2

(δ(Ω − Ωo ) + δ(Ω + Ωo )) .
∞ −α t
− jΩt

(b) U a ( jΩ) = ∫ e

=

1
α − jΩ

−∞

[

e

0

dt = ∫ e αt e − jΩt dt + ∫ e − αt e − jΩt dt
−∞

]

0
⋅ e ( α − jΩ ) t − ∞ −

1
α + jΩ

0
0
⋅ e − ( α + jΩ ) t − ∞ =

[

−∞

−∞

]

1
1

+
=
.
2
α − j Ω α + jΩ α + Ω 2

(c) Va ( jΩ) = ∫ e jΩ o t e − jΩt dt = ∫ e − j (Ω − Ω o )t dt = δ(Ω − Ω o ).
∞ ⎛ ∞
∞ ⎛ ∞

(d) Pa ( jΩ) = ∫ ⎜⎜ ∑ δ(t − lT ) ⎟⎟ e − jΩt dt = ∑ ⎜⎜ ∫ δ(t − lT )e − jΩt dt ⎟⎟ using the
l = −∞ ⎝ − ∞

− ∞ ⎝ l = −∞

linearity property of the CTFT. Next, using the shifting property of the CTFT we get

Pa ( jΩ) = ∑ e − jΩlT which can be alternately expressed in the form
l = −∞

Pa ( jΩ) =

2π ∞
2 πl
) making use of the results of Problem 3.2(c).
∑ δ(Ω −
T l = −∞
T

3.3

(a) Va ( jΩ) = ∫ e − jΩt dt = δ(Ω).
(b)

µ ( jΩ ) =

−∞

∞ ⎛ t
⎞ − jΩt
1
− j Ωt
µ
=
(
)
+ πδ(Ω) .
t
e
dt
dt =

∫ ⎜⎜ ∫ δ(τ) dτ ⎟⎟ e

j
−∞
−∞ ⎝ −∞

(c) The function x a (t ) is also denoted by rect (t ). Thus, X a ( jΩ) = ∫ x a (t ) e − jΩt dt

=

1/ 2

− jΩt
dt = −
∫ e

−1 / 2

(

)

−∞

1 − jΩ / 2
2 sin(Ω / 2) sin(Ω / 2)
− e jΩ / 2 =
=
e
= sinc(Ω / 2).

jΩ
Ω/2

Not for sale

45

⎧1 − 2 t , t < 0.5,
(d) y a (t ) = ⎨
A more convenient way to determine the CTFT of y a (t )
⎩ 0, t ≥ 0.5.
is to differentiate it twice with respect to t , determine the CTFT of

d 2 y a (t )
dt 2

and then

make use of the time-differentiation property given in Problem 3.6(e) and time-shifting
dy (t ) ⎧ 2, t < 0.5,
dy ( t )
property given in Problen 3.6(a). Now, a = ⎨
As a
has jump
dt
dt
t

2
,

0
.
5
.

discontinuities with a positive jump of value 2 at t = ±0.5, a negative jump of value − 4
at t = 0, and zero everywhere else,

d 2 y a (t )
dt 2

and an impulse of strength − 4 at t = 0, i.e.,

has only impulses of strength 2 at t = ±0.5,
d 2 y a (t )
dt 2

= 2δ(t + 0.5) − 4δ(t ) + 2δ(t − 0.5). If

Ya ( jΩ) denotes the CTFT of y a (t ) , then using of the time-differentiation property we
d 2 y a ( t ) CTFT

have

dt

↔ ( jΩ) 2 Ya ( jΩ). Using the time-shifting property, we arrive at the

2

d 2 y a (t )

CTFT of

dt

2

given by 2 e jΩ / 2 − 4 + 2 e − jΩ / 2 . Therefore ( jΩ) 2 Ya ( jΩ) =

− Ω 2Ya ( jΩ) = 2 e jΩ / 2 − 4 + 2 e − jΩ / 2 = 4(cos(Ω / 2) − 1) , i.e,

Ya ( jΩ) = −

3.4

h(t ) =

4

1
σ 2π

2

(cos(Ω / 2) − 1) =

1

sin 2 ⎛⎜ ⎞⎟ = sinc 2 ⎛⎜ ⎞⎟.
⎝4⎠ 2
⎝4⎠

8

2

2
2
⋅ e − (t − µ) / 2σ . Thus, H ( jΩ) = ∫

1

− ∞ σ 2π

Making a change of variable t − µ = τ we get H ( jΩ) =
=

1
σ 2π

=e
e

3.5

−(

1

−∞

σ 2π

2
2
e − jΩµ ∫ e − τ / 2 σ e − jΩτ dτ =

2

2
2
⋅ e − (t − µ ) / 2σ e − jΩt dt.

1

∫e

− τ 2 / 2 σ 2 − jΩ ( τ + µ )

σ 2π − ∞

e

2 2
⋅ e − j Ωµ ⋅ σ 2 π ⋅ e − σ Ω / 2

2

σ Ω
+ jΩµ )
2
.

For a zero mean impulse response, we then have the CTFT pair

− t / 2 σ CTFT
2

x a (t ) =

2

−σ2Ω2 / 2
.
↔ σ 2 πe

(

)

1 1 jΩt
1
sin( t )
1 ∞
jΩt
.
e jΩ − e − jΩ =
∫ X ( jΩ ) e dΩ =
∫ e dΩ =
2 π −1
πt
2π − ∞
j 2 πt

Not for sale

46

3.6

(a)

∫ x a (t − t o ) e

− j Ωt

−∞

dt = ∫ x a ( τ) e − jΩ( τ + t o ) dτ obtained using a change of variable
−∞

−∞

−∞

t − t o = τ. Therefore ∫ x a (t − t o ) e − jΩt dt = e − jΩt o ∫ x a (τ) e − jΩτ dτ = e − jΩt o X a ( jΩ).

(b)

∫ x a (t ) e

j Ω o t − j Ωt

e

−∞

(c) x a (t ) =

dt = ∫ x a (t ) e − j (Ω − Ω o )t dt = X ( j (Ω − Ω o ) ).
−∞

1
j Ωt
− j Ωt
dΩ.
∫ X a ( jΩ) e dΩ. Therefore 2 π x a ( −t ) = ∫ X a ( jΩ) e
2π − ∞
−∞

Interchanging t and Ω we get 2 π x a (−Ω) = ∫ X a ( jt ) e − jΩt dt.
−∞

(d) For a positive real constant a the CTFT of x a (at ) is given by

− jΩ t
dt =
∫ x a ( at ) e

−∞

1 ∞
− j (Ω / a )τ

∫ x a ( τ) e
a
−∞

=

1
a

Xa ( j

).
a

In a similar manner we

1

can show that for a negative constant a the CTFT of x a (at ) is given by − X a ( j ).
a
a
CTFT 1

X a ⎛⎜ j ⎞⎟.
Therefore x a (at ) ↔
a
⎝ a⎠

(e) Differentiating both sides of x a (t ) =

1 ∞
j Ωt
∫ X a ( jΩ) e dΩ get
2 π −∞

dx a (t ) CTFT
dx a (t )
1 ∞
j Ωt
=
↔ jΩ X a ( jΩ).
∫ jΩ X a ( jΩ) e dΩ. Therefore
dt
dt
2π − ∞

3.7

X a ( jΩ) = ∫ x a (t ) e − jΩt dt = X a ( jΩ) e jθ a (Ω) , where θ a (Ω) = arg{X a ( jΩ)}. Thus,
−∞

X a ( − jΩ) = ∫ x a (t ) e jΩt dt. If x a (t ) is a real function of then it follows from the
−∞

definition of X a ( jΩ) and the expression for X a (− jΩ) that X a ( jΩ) and X a (− jΩ) are
complex conjugates. Therefore X a (− jΩ) = X a ( jΩ) and θ a (−Ω) = −θ a (Ω). Or in
other words, for a real , the magnitude spectrum X a ( jΩ) is an even function of Ω and
the phase spectrum θa (Ω) is an odd function of Ω.
3.8

Not for sale

47

3.9

xˆ (t ) = ∫ hHT (t − τ)x ( τ)dτ. where hHT (t ) is the impulse response of the Hilbert
−∞

transformer. Taking the CTFT of both sides we get Xˆ ( jΩ) = H HT ( jΩ) X ( jΩ) where
Xˆ ( jΩ ) and H ( jΩ) denote the CTFTs of xˆ (t ) and h (t ), respectively. Rewriting
HT

(

HT

)

Xˆ ( jΩ) = H HT ( jΩ) X p ( jΩ) + X n ( jΩ) = − j X p ( jΩ) + j X n ( jΩ). As the magnitude and

phase of Xˆ ( jΩ ) are an even and odd function, xˆ (t ) is seen to be real signal. Consider
the complex signal y(t ) = x (t ) + j xˆ (t ) . Its CTFT is then given by
Y ( jΩ) = X ( jΩ) + j Xˆ ( jΩ) = 2 X ( jΩ).
p

3.10

[

]

2

1
1
The total energy ε x = ∫ e − αt dt = ∫ e − 2αt dt =
e − 2 αt 0 =
= 1.
− 2α
2 α α =1 / 2
−∞
−∞
The total energy can also be computed using using the Parsevals’ theorem

εx=

1 ∞
1
dΩ.

2
2

−∞ α +Ω

Ωc

1

1

dΩ
Therefore, the 80% bandwidth Ω c can be found by evaluating

2
2

α
+

−Ωc

1 ⎡1
−1 ⎛ Ω ⎞ ⎤ c
=
⋅ tan ⎜ ⎟
2 π ⎢⎣ α
⎝ α ⎠⎥⎦ − Ω

=

2
tan −1 (2Ω c ) = 0.8 .
π

=
c

1 ⎡
−1 ⎛ Ω c ⎞⎟ − tan −1 ⎛⎜ − Ω c
⎢ tan ⎜⎜

⎜ α
2 πα ⎣

⎝ α ⎠

⎞⎤
1
−1 ⎛ Ω c
⎟⎟ ⎥ = πα ⋅ tan ⎜⎜ α

⎠⎦

⎟⎟
⎠ α =1 / 2

1
0.8 π ⎞
Therefore, Ω c = ⋅ tan⎛⎜
⎟ = 1.5388.
2
⎝ 2 ⎠
1

1

1

3.11 y[ n] = µ[ n] = y ev [ n] + y od [ n], where yev [ n] = ( y[ n] + y[ − n]) = + δ[ n] and
2
2 2
1
2

1
2

1
2

1
2

yod [ n] = ( y[ n] − y[ −n]) = (µ[ n] − µ[ − n]) = µ[ n] − − δ[ n]. Now,

Yev (e jω ) =

⎤ 1
1⎡
1
2
(
2
k
)
+
=
π
δ(ω + 2πk ) + . Since
π
δ
ω
+
π

2 ⎣ k = −∞
2
k = −∞
⎦ 2
1

1

1

1

y od [ n] = µ[ n] − − δ[ n], yod [ n − 1] = µ[ n − 1] − − δ[ n − 1]. As a result,
2 2
2 2
1
2

1
2

yod [ n] − y od [ n − 1] = µ[ n] − µ[ n − 1] − δ[ n − 1] + δ[ n] =

1
2

(δ[ n] + δ[ n − 1]).

Taking the DTFT of both sides of the above equation, we get
1 ⎛ 1 + e − jω ⎞
1
1
1
Yod (e jω ) − e − jωYod (e jω ) = 1 + e − jω or Yod (e jω ) = ⎜
− .
⎟=

j
ω

j
ω
2 ⎝ 1− e
2
2
⎠ 1− e

(

)

Hence, Y (e jω ) = Yev (e jω ) + Yod (e jω ) =

1
1 − e − jω

+ π ∑ δ(ω + 2 πk ).
k = −∞

3.12 The inverse DTFT of X (e jω ) = ∑ 2 πδ(ω + 2 πk ) is given by
k = −∞

Not for sale

48

x[ n] =

1 π

jωn
= 1.
∫ 2πδ(ω) e dω =
2π − π

3.13 Y (e jω ) = ∑ α e − jωn with α < 1. Rewriting we get
n

n = −∞
−1

n = −∞

n=0

n =1

n=0

Y (e jω ) = ∑ α − n e − jωn + ∑ α n e − jωn = ∑ (α e jω ) n + ∑ (α e − jω ) n
=

αe

1 − α e jω

+

1
1 − α e − jω

=

1− α

2

1 − 2α cos ω + α 2

.

∞ ⎛
sin(ωc n) ⎞ − jωn
3.14 G(e jω ) = ∑ ⎜ δ[ n] −
= 1 − H LP (e jω ).
⎟e
n
π

n = −∞ ⎝

G(e jω )
1

3.15 x[ n] =

c

0

ωc

π

ω

1 π
1 π
j ω j ωn
jω − jωn
dω.
∫ X (e ) e dω. Hence, x * [ n] =
∫ X * (e ) e
2π − π
2π − π

(a) Since x[n] is real and even, we have X (e jω ) = X * (e jω ). Thus

x[ −n] =
x[ n] =

1 π
j ω − j ωn
dω. Therefore,
∫ X (e ) e
2π − π

1
1 π

− jω
( x[ n] + x[ −n]) =
).
∫ X (e ) cos(ωn)dω. As x[n] is even, X (e ) = X (e
2
2π − π

As a result, the term X (e jω ) cos n(ωn) inside the above integral is even, and hence

x[ n] =

∫ X (e ) cos(ωn)dω.
π0

(b) Since x[n] is real and odd, we have x[ n] = − x[ −n] and X (e jω ) = − X (e − jω ). Thus,

x[ n] =

j π
1

( x[ n] − x[ −n]) =
∫ X (e ) sin(ωn)dω. As a result, the term X (e ) sin(ωn)
2
2π − π

inside the above integral is even, and hence x[ n] =

Not for sale

j π

∫ X (e ) sin(ωn)dω.
2π 0

49

⎛ jω 0 n e jφ − e − jω 0 n e − jφ ⎞
⎟µ[ n]
3.16 x[ n] = Aα sin(ω0 n + φ)µ[ n] = Aα

2
j

n⎜ e

n

)

(

)

(

n
n
A
A jφ
e α e jω 0 µ[ n] − e − jφ α e − jω 0 µ[ n]. Therefore, the DTFT of x[n] is given
2j
2j
1
1
A
A jφ
by X (e jω ) =
− e − jφ
.
e
j
ω
j
j
ω
j2
j2
1− αe e 0
1 − α e ω e − jω 0

=

3.17 Let x[ n] = α n µ[ n1] with α < 1. Its DTFT was computed in Example 3.6 and is given by
X ( e jω ) =

1
1 − α e − jω

.

(a) x1[ n] = α n µ[ n − 1] with α < 1. Its DTFT is given by X1 (e jω ) = ∑ α n e − jωn
n =1

= ∑ ( α e − jω ) n = ∑ ( α e − jω ) n − 1 =
n =1

n=0

1
1− αe

− jω

−1 =

αe

− jω

1 − α e − jω

.

(b) x 2 [ n] = nα n µ[ n] with α < 1. Note x 2 [ n] = n x[ n] . Therefore, using the
differentiation-in-frequency property in Table 3.4 we get

dX (e jω )
d ⎛⎜
α e − jω
1
⎟=
X 2 ( e jω ) = j
= j
.

dω ⎜⎝ 1 − α e − jω ⎟⎠ (1 − α e − jω ) 2
(c) x 3 [ n] = α n µ[ n + 1] with α < 1. Its DTFT is given by
X 3 (e

n − jωn

) = ∑α e
n = −1

−1 jω

=α e

n − jωn

+ ∑α e
n=0

−1 jω

=α e

1 ⎛⎜ e jω − α
+
=
α ⎜⎝ 1 − α e − jω
1 − α e − jω
1

(d) x 4 [ n] = nα n µ[ n + 2] with α < 1. Its DTFT is given by

X 4 (e jω ) = ∑ nα ne − jωn = ∑ nα ne − jωn − 2α − 2e j 2ω − α −1e jω. From the results of Part
n = −2

n =0

(b) we observe that
X 4 ( e jω ) =

α e − jω

n=0

(1 − α e − jω ) 2

n − jω n
=
∑ nα e

α e − jω
(1 − α e

− jω 2

. Hence,

− 2 α − 2 e j 2 ω − α − 1e jω .

)

(e) x 5 [ n] = nα n µ[ − n − 1] with α > 1. Its DTFT is given by

Not for sale

50

⎟.

−1

1

n = −∞

m =1

m =0

1 − α − 1e j ω

X 5 (e jω ) = ∑ α n e − jωn = ∑ α − m e jωm = ∑ α − m e jωm − 1 =

−1 =

e jω
α − e jω

.

⎧⎪ n
(f) x 6 [ n] = ⎨α , n ≤ M , Its DTFT is given by
⎪⎩ 0, otherwise.
−1

M

X 6 ( e j ω ) = ∑ α n e − j ωn +
n=0

∑α

− n − jω n

e

n =−M

=

1 − α M +1e − jω( M +1)
1 − α e − jω

+ α M e j ωM ⋅

1 − α M +1e − jω( M +1)
1 − α e − jω

3.18 (a) x a [ n] = µ[ n] − µ[ n − 5]. Let denote µ (e jω ) the DTFT of µ[n]. Using the timeshifting property of the DTFT given in Table 3.4, the DTFT of x a [n] is thus given by
X a (e jω ) = (1 − e − j 5ω ) µ (e jω ) . From Table 3.3, we have

µ ( e jω ) =

1 − e − j 5ω
.
+ ∑ πδ(ω + 2πk ). Therefore, X a (e jω ) =
1 − e jω k = −∞
1 − e jω

1

(b) x b [ n] = α n (µ[ n] − µ[ n − 8]). Let x[ n] = α n µ[ n1] with α < 1. Its DTFT was
1
computed in Example 3.6 and is given by X (e jω ) =
. Now
1 − α e − jω
x b [ n] = x[ n] − x[ n − 8]. Using the time-shifting property of the DTFT given in Table
3.4, the DTFT of x a [n] is thus given by X b (e jω ) = (1 − e − j 8ω ) X (e jω ) =

1 − e − j 8ω
1 − α e − jω

.

(c) x c [ n] = ( n + 1)α n µ[ n] = n α n µ[ n] + α n µ[ n] with α > 1. We can rewrite it as
x c [ n] = x 2 [ n] + x[ n]. The DTFT of x2 [ n] was computed in Problem 3.17(b) and is
given by X 2 (e jω ) =

α e − jω
(1 − α e

and is given by X (e jω ) =
=

3.19

αe

− jω

(1 − α e

− jω 2

+

)

− jω 2

1− αe

)

1
1− αe

1
− jω

and the DTFT of x[n] was computed in Example 3.6

=

− jω

. Therefore X c (e jω ) = X 2 (e jω ) + X (e jω )
1

(1 − α e − jω ) 2

.

⎛ 1 − e − jω(2 N +1) ⎞
N
1, − N ≤ n ≤ N ,

Then Y1 (e jω ) = ∑ e − jωn = e − jωN ⎜
(a) y1[ n] = ⎧⎨
j

ω
0
,
otherwise.

n = −N
⎝ 1− e

1
sin⎛⎜ ω[ N + ] ⎞⎟
2 ⎠

=
.
sin(ω / 2)

Not for sale

51

.

− jω( N +1)

N
1− e
1, 0 ≤ n ≤ N ,
(b) y2 [ n] = ⎧⎨
Then Y2 (e jω ) = ∑ e − jωn =
⎩0, otherwise.
n =0
1 − e − jω

⎛ sin(ω[ N + 1] / 2) ⎞
= e − jωN / 2 ⎜⎜
⎟⎟ .
⎝ sin(ω / 2) ⎠
n
⎧⎪
(c) y3 [ n] = ⎨1 − N , − N ≤ n ≤ N , Assume N
⎪⎩ 0,
otherwise.
⎧⎪1, −
1
y 3 [ n] = y 0 [ n] O
* y 0 [ n] where y 0 [ n] = ⎨
N
⎪⎩0,

Y3 (e jω ) =

1
1
Y0 (e jω )Y0 (e jω ) = Y02 (e jω ).
N
N

to be odd. Then we can express
N −1
2

≤n≤

N −1
,
2

Therefore,

otherwise.

Now, from the results of Part (a), we have

sin(ωN / 2 )
1 sin 2 (ωN / 2 )
. Hence, Y3 (e jω ) = ⋅
.
N sin 2 (ω / 2)
sin(ω / 2)
Note: The above result also holds for N even.
Y0 (e jω ) =

⎧N + 1 − n ,
(d) y 4 [ n] = ⎨
0,

considered in Part (a) and

− N ≤ n ≤ N,
= y1[ n] + Ny3 [ n], where y1[ n] is the sequence
otherwise,
y3 [ n] is the sequence considered in Part (c). Hence,

1
sin⎛⎜ ω[ N + ]⎞⎟
2
2 ⎠ sin (ωN / 2 )

+
.
Y4 (e jω ) = Y1 (e jω ) + N ⋅ Y3 (e jω ) =
sin(ω / 2)
sin 2 (ω / 2)

cos(πn / 2 N ), − N ≤ n ≤ N ,
Then
(e) y5 [ n] = ⎧⎨
0,
otherwise.

1 N − j ( πn / 2 N ) − jωn 1 N j ( πn / 2 N ) − jωn
+
Y5 (e jω ) =
e
e
∑e
∑e
2 n=−N
2 n = −N

π

π

(

π

1

)

(

π

1

)

1 N − j ⎜⎝ ω − 2 N ⎟⎠ n 1 N − j ⎜⎝ ω + 2 N ⎟⎠ n 1 sin (ω − 2 N )( N + 2 ) 1 sin (ω + 2 N )( N + 2 )
=
+
= ⋅
+ ⋅
.
∑e
∑e
2 n = −N
2 n = −N
2 sin (ω − π ) / 2
2 sin (ω + π ) / 2
2N
2N
3.20

(

)

(

)

(n + m − 1)! n
α µ[ n] with α < 1. We shall prove by induction that the
n! (m − 1)!
1
DTFT of x m [n] is given by X m (e jω ) =
. From Table 3.3, it follows that
(1 − α e − jω ) m
it holds for m = 1. Let m = 2. Then
(n + 1)! n
x 2 [ n] =
α µ[ n] = (n + 1) x1[ n] = nx1[ n] + x1[ n]. Therefore,
n!
Denote x m [ n] =

Not for sale

52

X 2 ( e jω ) =

α e − jω
(1 − α e − jω ) 2

+ 1 − α e − jω =

1
(1 − α e − jω ) 2

, and it also holds for m = 2.

Now, assume that it holds for m. Consider next x m +1[ n] =

(n + m)! n
α µ[ n]
n! (m)!

1
⎛n+m⎞
⎛ n + m ⎞ (n + m − 1)! n
α µ[ n] = ⎜
=⎜
⎟ x m [ n] = ⋅ n ⋅ x m [ n] + x m [ n]. Hence,

m
⎝ m ⎠
⎝ m ⎠ n! (m − 1)!

α e − jω
1 d ⎛⎜
1
1
1
⎟+
=
+
X m + 1 ( e jω ) = j
m dω ⎜⎝ (1 − α e − jω ) m ⎟⎠ (1 − α e − jω ) m
(1 − α e − jω ) m +1 (1 − α e − jω ) m
1
.
=
(1 − α e − jω ) m +1

3.21 (a) X a (e jω ) = ∑ δ(ω + 2 πk ). Hence, x a [ n] =
k = −∞

(b) X b (e jω ) =

(

e jω 1 − e jωN
1 − e jω

) = e jω N∑−1e jωn .
n =0

1 π
jωn
∫ δ(ω) e dω = 1.
2π − π
− N +1

Let m = −n. X b (e jω ) = e jω ∑ e − jωm .
m =0

− N +1

Consider the DTFT X (e jω ) = ∑ e − jωm . Its inverse is given by
m =0

1, − ( N − 1) ≤ n ≤ 0,
x[ n] = ⎧⎨
Therefore, by the time-shifting property of the DTFT, the
otherwise.
⎩0,
1, − N ≤ n ≤ −1,
inverse DTFT of X b (e jω ) = e jω X (e jω ) is given by x b [ n] = x[ n + 1] = ⎧⎨
⎩0, otherwise.
⎧3, n = 0,
N
N

(c) X c (e jω ) = 1 + 2 ∑ cos(ωl) = 2 + ∑ e − jωl . Hence, x c [ n] = ⎨1, 0 < n ≤ N ,
l=0
l = −N
⎪⎩0, otherwise.
(d) X d (e jω ) =

X d ( e jω ) =

− α e − jω
(1 − α e − j

jω 2

)

with α < 1. We can rewrite X d (e jω ) as

dX o (e jω )
1
where X o (e jω ) =
. From Table 3.3, the inverse

1 − α e − jω

DTFT of X d (e jω ) is given by x o [ n] = α n µ[ n]. From Table 3.4, using the
differentiation-in-frequency property the inverse DTFT of X d (e jω ) is thus given by
x d [ n] = n x o [ n] = n α n µ[ n].

Not for sale

53

e j 4ω − e − j 4ω
2j

3.22 (a) H a (e jω ) = sin(4ω) =

1 j 4ω
e
2j

=

ha [ n] = {− j 0.5, 0, 0, 0, 0, 0, 0, 0,
(b) H b (e jω ) = cos(4ω) =

e j 4ω + e − j 4ω
2

e j 5ω − e − j 5 ω
2j

1 − j 4ω
e
. Therefore,
2j

j 0.5}, − 4 ≤ n ≤ 4.

= 0.5e j 4ω + 0.5e − j 4ω . Therefore,

hb [ n] = { 0.5, 0, 0, 0, 0, 0, 0, 0,
(c) H c (e jω ) = sin(5ω) =

=

0.5}, − 4 ≤ n ≤ 4.

1 j 5ω
e
2j

1 − j 5ω
e
. Therefore,
2j

hc [ n] = {− j 0.5, 0, 0, 0, 0, 0, 0, 0, 0, 0,
(d) H d (e jω ) = cos(5ω) =

e j 5 ω + e − j 5ω
2

j 0.5}, − 5 ≤ n ≤ 5.

= 0.5e j 5ω + 0.5e − j 5ω . Therefore,

hd [ n] = { 0.5, 0, 0, 0, 0, 0, 0, 0, 0, 0,

0.5}, − 5 ≤ n ≤ 5.

⎛ e jω + e − jω ⎞ ⎛ e j 2 ω + e − j 2 ω ⎞
3.23 (a) H1 (e jω ) = 1 + 2 cos(ω) + 3 cos(2ω) = 1 + 2⎜
⎟ + 3⎜

2
2

⎠ ⎝

= 1 + e jω + e − jω + 1.5e j 2ω + 1.5e − j 2 ω . Therefore,
{h1[ n]} = {1.5, 1, 1, 1, 1.5}, − 2 ≤ n ≤ 2.
ω
(b) H 2 (e jω ) = (3 + 2 cos(ω) + 4 cos(2ω)) cos⎛⎜ ⎞⎟ e − jω / 2
⎝2⎠

⎛ e jω + e − jω ⎞ ⎛ e j 2 ω + e − j 2 ω ⎞⎤ ⎛ e jω / 2 + e − jω / 2 ⎞ − jω / 2
= ⎢3 + 2⎜
⎟e
⎟ + 4⎜
⎟⎥ ⎜
2
2
2

⎠ ⎝
⎠⎦ ⎝

1
= 3 + e jω + e − jω + 2 e j 2 ω + 2 e − j 2 ω 1 + e − jω
2

(

)(

)

= 2 + 1.5 e jω + 2 e − jω + e j 2ω + 1.5 e − j 2ω + e − j 3ω . Hence,
{h2 [ n]} = {1, 1.5, 2, 2, 1.5, 1}, − 2 ≤ n ≤ 3.
(c) H 3 (e jω ) = j [3 + 4 cos( ω) + 2 cos( 2ω)]sin( ω)

⎛ e jω + e − jω ⎞ ⎛ e j 2 ω + e − j 2 ω ⎞⎤ ⎛ e jω − e − jω ⎞
= j ⎢3 + 4⎜

⎟ + 2⎜
⎟⎥ ⎜
2
2
2j

⎠ ⎝
⎠⎦ ⎝

=

1
2

(3 + 2 e jω + 2 e − jω + e j2ω + e − j2ω )(e jω − e − jω )

= 3 + 2 e jω + 2 e − jω + 0.5 e j 2ω + 0.5 e − j 2ω . Hence,
{hc [ n]} = {0.5, 2, 3, 2, 0.5}, − 2 ≤ n ≤ 2.

Not for sale

54

(d) H 4 (e jω ) = j[4 + 2 cos(ω) + 3 cos(2ω)]sin(ω / 2) e jω / 2

⎛ e jω + e − jω ⎞ ⎛ e j 2 ω + e − j 2 ω ⎞⎤ ⎛ e jω / 2 − e − jω / 2 ⎞ − jω / 2
= j ⎢4 + 2⎜
⎟e
⎟ + 3⎜
⎟⎥ ⎜
2
2
2j

⎠ ⎝
⎠⎦ ⎝

=

1
2

(4 + e jω + e − jω + 1.5 e j2ω + 1.5 e − j2ω )(1 − e − jω )

= 1.5 − 0.25 e jω − 1.5 e − jω + 1.5 e j 2ω − 0.5 e − j 2ω − 0.75 e − j 3ω . Hence,
{h4 [ n]} = {− 1.5, − 0.5, − 3, 3, − 0.5, − 3}, − 3 ≤ n ≤ 2.
3.24 Let H (e jω ) and G(e jω ) denote the DTFTs of the sequences h[n] and g[n] , respectively.
(a) Linearity Theorem: F {α h[ n ] + β g [ n ]} =

n = −∞

n = −∞

∑ ( α h[ n ] + β g [ n ]) e

− jω n

n = −∞

= α ∑ h[ n] e − jωn + β ∑ g[ n] e − jωn = α H (e jω ) + β G(e jω ).

(b) Time-reversal Theorem:

∑ h[ −n] e

n = −∞

(c) Time-shifting Theorem:

= e − jωn o

∑ h[ m] e

m = −∞

− jωm

∑ h[ n − no ] e

∑ h[ m] e

jωm

m = −∞

=

∑ h[ m] e

= H (e − jω ).

− jω( m + n o )

m = −∞

= e − jωn o H (e jω ).

= H (e

=

− jωn

n = −∞

(d) Frequency-shifting Theorem:
j ( ω − ωo )

− jωn

(

)

jω n
− jωn
= ∑ h[ n] e − j (ω − ω o ) n
∑ e o h[ n] e

n = −∞

n = −∞

).

3.25 Let H 1(e jω ) = F {h1[ n]}and H 2 (e jω ) = F {h2 [ n]}. From Example 3.8 we have
∞ ⎛ sin(ω n) ⎞
2
H 2 ( e jω ) = ∑ ⎜

πn ⎠
n = −∞ ⎝

⎧1, 0 ≤ ω ≤ ω2 ,
=⎨
From the result of Problem 3.14 we get
⎩0, ω2 < ω ≤ π.
∞ ⎛
sin(ω1n) ⎞ − jωn ⎧0, 0 ≤ ω ≤ ω1 ,
H1 (e jω ) = ∑ ⎜ δ[ n] −
=⎨
⎟e
πn ⎠
n = −∞ ⎝
⎩1, ω1 < ω ≤ π.
As the impulse response of the cascade is given by h[ n] = h1[ n] O
* h2 [ n] , using the
convolution theorem we obtain the DTFT of the cascade: H (e jω ) = H 1(e jω )H 2 (e jω )
Not for sale

55

⎧0, 0 ≤ ω < ω1 ,

= ⎨1, ω1 ≤ ω ≤ ω2 ,
⎪0, ω < ω ≤ π.
2

H(e jω )
1
_ π _ ω _ ω1
2

0

ω1

ω2 π ω

(

)
n = −∞

Y (e jω ) = ∑ y[ n] e − jωn = X ((e jω ) 4 ) = ∑ x[ n] (e − jωn ) 4 = ∑ x[ m / 4] e − jωm .
n = −∞
n = −∞
m = −∞

3.26 Y (e jω ) = X (e j 4ω ) = X (e jω )4 . Now, X (e jω ) = ∑ x[ n] e − jωn . Hence,

x[ n], n = 0, ± 4, ± 8, ± 16,K
Therefore y[ n] = ⎧⎨
otherwise.
⎩ 0,

3.27 X (e jω ) = ∑ x[ n] e − jωn . Therefore, X (e jω / 2 ) = ∑ x[ n] e − j (ω / 2)n and
n = −∞

n = −∞

X (−e jω / 2 ) = ∑ x[ n] (−1) n e − j (ω / 2)n . Thus, we can write
n = −∞

Y (e jω ) =

∑ y[ n] e − jωn =

n = −∞

1
2

{X (e jω / 2 ) + X (−e jω / 2 )}= 12 ∑∞ (x[n] + x[n](−1)n )e − j(ω / 2)n .
n = −∞

x[ n] e − j (ω / 2)n , for n even,
x[2n], for n even,
Thus, y[ n] = ⎨ ∑
Hence, y[ n] = ⎧⎨
n = −∞
for n odd.
⎩ 0,

0,
for n odd.

3.28 F {x * [ −n]} =
3.29 X ca (e jω ) =

1
2

∑ x * [ − n] e

n = −∞

[X (e

− jωn

n = −∞

n = −∞

= ∑ x * [ n] e jωn = ( ∑ x[ n] e − jωn )* = X *(e jω ) .

]

) − X * ( e − jω ) =

1⎛ ∞
= ⎜⎜ ∑ x re [ n] e − jωn + j ∑ x im [ n] e − jωn − ∑ x re [ n] e − jωn − j ∑ x im [ n] e − jωn ⎟⎟
2 n = −∞
n = −∞
n = −∞
n = −∞

ω, = j ∑ x im [ n] e − jωn = F {j xim [n]}.
n = −∞

3.30

Not for sale

56

|X(e jω )|

0

0.6π

1.4π

3.31 From Table 3.2 we observe that an even real-valued sequence has a real-valued DTFT
and an odd real-valued sequence has an imaginary-valued DTFT.
(a) Since x1[ n] is an odd sequence, it has an imaginary-valued DTFT.
(b) Since x 2 [ n] is an even sequence, it has a real-valued DTFT.

sin(−ωc n) − sin(ωc n) sin(ωc n)
=
=
= x3 [ n]. Since, x3 [ n] is an even
− πn
− πn
πn
sequence, it has a real-valued DTFT.

(c) x3 [ −n] =

(d) Since x 4 [ n] is an odd sequence, it has an imaginary-valued DTFT.
(e) Since x 5 [ n] is an odd sequence, it has an imaginary-valued DTFT.
3.32 From Table 3.2 we observe that an even real-valued sequence has a real-valued DTFT
and an odd real-valued sequence has an imaginary-valued DTFT.
(a) Since Y1 (e jω ) is a real-valued function of ω, its inverse is an even sequence.
(b) Since Y2 (e jω ) is an imaginary-valued function of ω, its inverse is an odd sequence.
(c) Since Y3 (e jω ) is an imaginary-valued function of ω, its inverse is an odd sequence.
3.33 (a) Since H1 (e jω ) is a real-valued function of ω, its inverse is an even sequence.
(b) Since H 2 (e jω ) is a real-valued function of ω, its inverse is an even sequence.
3.34 Let u[ n] = x * [ −n], and let X (e jω ) and U (e jω ) denote the DTFTs of x[n] and u[n],
respectively. From the convolution property of the DTFT given in Table 3.4, the DTFT

of y[ n] = x[ n] O
* u[ n] is given by Y (e ) = X (e )U (e ). From Table 3.1, U (e )

= X * (e jω ). Therefore, Y (e jω ) = X (e jω ) X * (e jω ) = X (e jω )

2

which is a real-valued

function of ω.
3.35 From the frequency-shifting property of the DTFT given in Table 3.4,

F {x[ n] e − jπn / 3} = X (e j (ω + π / 3) ). A sketch of this DTFT is shown below
Not for sale

57

X(e j(ω+π/3) )

n

3.36 {−α µ[ − n − 1]} = X (e

_ 2π _ π
3
3
−1

=

2

Now, 4 ∫ X (e jω ) dω = 2 ∫ X (e jω ) dω = 4π ∑
n

(b) X (e jω )

2

n = −∞

−π
n

0

⎛ e jω ⎞
⎟ .
∑ ⎜
n=0 ⎝ α ⎠

2

1
1− α e

= −α e

n

.

⇒ X (e jω ) =

1
2

1+ α − 2α cos(ω)

.

Hence α = −2. Therefore, x[ n] = (−2) n µ[ −n − 1].

π

2

/ α)

=

− 1 jω

2

1 π
2

∫ X (e ) dω = ∑ x[ n] .
2π − π
n = −∞

1
.
5 + 4 cos(ω)

π

1− (e

e

n =1

1

ω

− n jω n

= − ∑α

n = −∞

From Parseval’s relation,
2

n − j ωn

) = ∑−α e

For α > 1, X (e jω ) = −α −1e jω

(a) X (e jω )

π

0

2

−1

x[ n] = 4π ∑ (−2) n

2

n = −∞

1
1
= 4 π ∑ ⎛⎜ ⎞⎟ = π ∑ ⎛⎜ ⎞⎟ = .
3
n =1 ⎝ 4 ⎠
n=0 ⎝ 4 ⎠

π

=

1
.
3.25 − 3 cos(ω)

2

Now, ∫ X (e jω ) dω =
0

2
2

−1
1 π
2

n
∫ X (e ) dω = π ∑ x[ n] = π ∑ (1.5)
2 −π
n = −∞
n = −∞

n

Hence α = 1.5. Therefore, x[ n] = −(1.5) n µ[ −n − 1].

n

4π 9

4
4
= π ∑ ⎛⎜ ⎞⎟ =
⋅ = .
∑ ⎛⎜ ⎞⎟ =
9
5
5
9
9
9
n =1 ⎝ ⎠
n=0 ⎝ ⎠

(c) Using the differentiation-in-frequency property of the DTFT as given in Table 3.4,
− jω

d ⎛
1
⎟ = αe

the inverse DTFT of X (e jω ) = j
is x[ n] = −nα n µ[ −n − 1].

j
ω
2

j
ω
dω ⎝ 1 − α e
)
⎠ (1− α e
Hence, the inverse DTFT of
2

Y ( e jω ) =
π

1
( 5 − 4 e − jω ) 2

2

.

1
(1− α e − jω ) 2

is − (n + 1)α n µ[ −n − 1].

Hence α = 2. Therefore, y[ n] = −(n + 1)2 n µ[ −n − 1] .
π

2

Now, 4 ∫ X (e jω ) dω = 2 ∫ X (e jω ) dω = 4π ∑
0

n = −∞

−π

Not for sale

2

−1

x[ n] = 4π ∑ (n + 1) 2 ⋅ 2 2 n
n = −∞

58

n

9/4
1
= π ∑ ⎛⎜ ⎞⎟ ⋅ n 2 = π
= 4 π.
9 / 16
n=0 ⎝ 4 ⎠
2

6
2
2
dX (e jω )
dω = 2π ∑ n x[ n] = 2π ∑ n x[ n] = 152π. (Using Parseval’s relation with

−∞
−3
−π

π

3.37

differentiation-in-frequency property)

6

n = −∞

n = −2

3.38 (a) X (e j 0 ) = ∑ x[ n] = ∑ x[ n] = 10.
6

(b) X (e jπ ) = ∑ x[ n] e jπn = ∑ x[ n] e jπn = −6.
n = −∞

π

∫ X (e

(c)

n = −2

)dω = 2πx[0] = −2π.

−π
π

2

∫ X (e ) dω = 2 π ∑

(d)

n = −∞

−π

6

2

2

x[ n] = 2π ∑ x[ n] = 120π.
n = −2

π dX (e jω ) 2
6

2
2
dω = 2 π ∑ n x[n] = 2π ∑ n x[n] = 1724π.
(e) ∫

−∞
−2
−π

2

n = −∞

n = −6

3.39 (a) X (e j 0 ) = ∑ x[ n] = ∑ x[ n] = 12.
2

(b) X (e jπ ) = ∑ x[ n] e jπn = ∑ x[ n] e jπn = −12.
n = −∞

π

∫ X (e

(c)

n = −6

)dω = 2πx[0] = −4π.

−π
π

2

∫ X (e ) dω = 2 π ∑

(d)

n = −∞

−π
π

(e)

−π

dX (e jω )

2

2

2

2

x[ n] = 2π ∑ x[ n] = 160π.
n = −6

2

2

2

dω = 2π ∑ n x[ n] = 2π ∑ n x[ n] = 136π.
−∞

−6

3.40 From the differentiation-in-frequency property of the DTFT given in Table 3.4 we

have

∑ n x[ n] e

n = −∞

∑ n x[ n] = j

n = −∞

− jωn

dX (e jω )

=j

dX (e jω )
where X (e jω ) = F {x[ n]}. Therefore,

. From the definition of the DTFT X (e jω ) = ∑ x[ n] e − jωn
n = −∞

ω= 0

Not for sale

59

j

it follows that

j0
∑ x[ n] = X (e ). Therefore, C g =

n = −∞

From Table 3.3, X (e jω ) = F {α n µ[ n]} =
j

dX (e jω )

=
ω= 0

α e − jω
(1− α e − jω ) 2

=
ω= 0

1
1 − α e − jω

α
(1 − α ) 2

dX (e jω )

X (e

ω= 0

j0

.

)

, α < 1. As a result,

and X (e j 0 ) =

1
.
1− α

α

Hence, C g =
.
1− α

3.41 Let G1 (e jω ) = F {g1[ n]}.
(b) Note g 2 [ n] = g1 [ n] + g1 [ n − 4]. Hence, F {g 2 [ n]} = G2 (e jω )
= G1 (e jω ) + e − j 4ωG1 (e jω ) = 1 + e − j 4ω G1 (e jω ).

(

)

(c) Note g 3 [ n] = g1 [ −(n − 3)] + g1 [ n − 4]. Now, F {g1[ −n]} = G1 (e − jω ). Hence,

F {g3 [ n]} = G3 (e jω ) = e − j 3ωG1 (e − jω ) + e − j 4ωG1 (e jω ).
(d) Note g 4 [ n] = g1 [ n] + g1 [ −(n − 7)]. Hence, F {g 4 [ n]} = G 4 (e jω )

= G1 (e jω ) + e − j 7ωG1 (e − jω ).

⎞⎛ ∞
⎛ ∞
3.42 Y (e jω ) = X1 (e jω ) X 2 (e jω ), i.e., ∑ y[ n] e − jωn = ⎜⎜ ∑ x1[ n] e − jωn ⎟⎟⎜⎜ ∑ x 2 [ n] e − jωn ⎟⎟.
n = −∞

⎠⎝ n = −∞
⎝ n = −∞

⎞⎛

(a) Setting ω = 0 in the above we get ∑ y[ n] = ⎜⎜ ∑ x1[ n] ⎟⎟⎜⎜ ∑ x 2 [ n] ⎟⎟.
n = −∞

⎠⎝ n = −∞
⎝ n = −∞

⎛ ∞
⎞⎛ ∞

(b) Setting ω = π we get ∑ (−1) n y[ n] = ⎜⎜ ∑ (−1) n x1[ n] ⎟⎟⎜⎜ ∑ (−1) n x2 [ n] ⎟⎟.
n = −∞
⎝ n = −∞
⎠⎝ n = −∞

3.43 x[ n] = α n µ[ n], α < 1. From Table 3.3, F {x[ n]} = X (e jω ) =
of x[n] is E x =

1 π
1

− jω

− π 1− α e

80% bandwidth of the signal, we

2

dω = ∑ ( α 2 ) n =
n=0

1
set E x,80 =

ωc

1
1 − α e jω

. The total energy

4
. To determine the
1 − α 2 α =1 / 2 3
1

=

2
1

− jω
− ω c 1− α e

dω = 0.8 E x = 0.8 ⋅

4
3

⎡ 1 ωc

3.2
1
dω⎥
=
and solve for ω c , i.e., set E x,80 = ⎢
.

⎢⎣ 2 π − ω c (1− α cos ω) 2 + α 2 sin 2 ω ⎥⎦ α =1 / 2 3
A numerical solution of the above equation yields ωc = 0.5081π.

Not for sale

60

1
2

3.44 Recall x[ n] = x ev [ n] + x od [ n], where x ev [ n] = ( x[ n] + x[ −n]) and As x[n] is causal,

x[ n] = 0 for n < 0 and x[ −n] = 0 for n > 0. Hence, there is no overlap between the
nonzero portions of x[n] and x[ −n] except at n = 0, and we have
x[ n] = 2 x ev [ n]µ[ n] − x ev [0]δ[ n] and x[ n] = 2 x od [ n]µ[ n] + x od [0]δ[ n]. Moreover, since

x[n] is real, it follows from Table 3.2 that F {x ev [ n]} = X re (e jω ) and

F {x ev [ n]} = j X im (e jω ). Taking the DTFT of x[ n] = 2 x ev [ n]µ[ n] − x ev [0]δ[ n] we
arrive at

X ( e jω ) =

1 π

j (ω − ν )
)dν − x[0].
∫ X re (e ) µ (e
π
−π

From Table 3.3 we have

ω ⎞
1⎛
(
2
k
)
1
j
cot(
)
+
πδ(ω + 2πk ).
+
πδ
ω
+
π
=

2⎝
2 ⎠ k = −∞
1− e − jω k = −∞
Substituting the above in the equation preceding it we get

µ (e jω ) = F {µ[ n]} =
X (e jω ) =

1

1 π
⎡1
X re (e jν )⎢ ⎛⎜1 −

π
⎣2 ⎝
−π

ω − ν ⎞ ⎞⎤
j cot⎛⎜
⎟ ⎟ dν
⎝ 2 ⎠ ⎠⎥⎦
π

+ ∑

∫ X re (e

k = −∞ − π

=

) δ((ω − ν) + 2 πk ) dν − x[0]

j π
1 π
ω− ν ⎞

⎟dν + X re (e jω ) − x[0].
∫ X re (e ) dν −
∫ X re (e ) cot⎛⎜
⎝ 2 ⎠
2π − π
2π − π

Comparing the last equation with X (e jω ) = X re (e jω ) + j X im (e jω ), we arrive at
1 π
ω− ν ⎞

X im (e jω ) = −
⎟dν.
∫ X re (e ) cot⎛⎜
⎝ 2 ⎠
2π − π
Likewise, taking the DTFT of x[ n] = 2 x od [ n]µ[ n] + x od [0]δ[ n] we get

X ( e jω ) =

j π

j (ω − ν )
)dν + x[0].
∫ X im (e ) µ (e
π
−π

Substituting the expression for µ (e

X ( e jω ) =

j π
⎡1
X im (e jν )⎢ ⎛⎜1 −

π
⎣2 ⎝
−π

+j ∑

) given earlier in the above equation we get

ω − ν ⎞ ⎞⎤
j cot⎛⎜
⎟ ⎟ dν
⎝ 2 ⎠ ⎠⎥⎦
π

∫ X im (e

k = −∞ − π
1 π

) δ((ω − ν) + 2πk ) dν + x[0]

j π
ω− ν ⎞

⎟dν + jX im (e jω ) + x[0].
∫ X im (e ) dν + 2 π ∫ X im (e ) cot⎛⎜

2 ⎠

−π
−π
Comparing the last equation with X (e jω ) = X re (e jω ) + j X im (e jω ), we arrive at

=

Not for sale

61

X re (e jω ) =

1 π
ω− ν ⎞

⎟dν + x[0].
∫ X im (e ) cot⎛⎜
⎝ 2 ⎠
2π − π

3.45 If u[ n] = z n is the input to the LTI discrete-time system, then its output is given by

k = −∞

k = −∞

k = −∞

y[ n] = ∑ h[ k ] u[ n − k ] = ∑ h[ k ] z n − k = z n ∑ h[ k ] z − k = z n H ( z).
where H ( z) = ∑ h[ k ] z − k . Hence is an eigenfunction of the system.
k = −∞

n

If v[ n] = z µ[ n] is the input to the LTI discrete-time system, then its output is given by

n

k = −∞

k = −∞

k = −∞

y[ n] = ∑ h[ k ] v[ n − k ] = z n ∑ h[ k ] µ[ n − k ] z n − k = z n ∑ h[ k ] z − k .
Since in this case the summation depends upon is not an eigenfunction of the system.
3.46 F {h1[ n]} = H1 (e jω ) = F {δ[ n] + δ[ n − 1]} = 1 + 0.5e − jω ,
2
1

F {h2 [ n]} = H 2 (e jω ) = F {1 δ[ n] − 1 δ[ n − 1]} = 0.5 − 0.25e − jω ,
2

4

F {h3 [ n]} = H 3 (e ) = F {2δ[ n]} = 2,
n

F {h4 [ n]} = H 4 (e jω ) = F {−2⎛⎜ 1 ⎞⎟ µ[ n]} =

−2

.
⎝2⎠
1 − 0.5e − jω
The overall frequency response of the structure of Figure 2.35 is given by
H ( e jω ) = H 1 ( e jω ) + H 2 ( e jω ) H 3 ( e jω ) + H 2 ( e jω ) H 4 ( e jω )

= 1 + 0.5e − jω + 2(0.5 − 0.25e − jω ) −

2(0.5 − 0.25e − jω )
1 − 0.5e − jω

= 1.

3.46 Denote H i (e jω ) = F {h1[ n]}, 1 ≤ i ≤ 5.
(a) The overall frequency of Figure P2.2(a) is then given by
H i (e jω ) = H1 (e jω ) H 2 (e jω ) + H 3 (e jω ) H 4 (e jω ) + H1 (e jω )H 2 (e jω ) H 3 (e jω ) H 5 (e jω ).
(b) The structure of Figure P2.2(b) can be redrawn as shown below

h 4[n]
ho[n]

h3[n]

+

where the block with an impulse response represents the part of Figure P2.2(b) with a
feedback loop as shown below

Not for sale

62

w[n]

u[n]

+

h1[n]

h2[n]

v[n]

h5[n]
Let U (e jω ) = F {u[ n]}, V (e jω ) = F {v[ n]}, and W (e jω ) = F {w[ n]}. Then we have
W (e jω ) = U (e jω ) + H 3 (e jω )V (e jω ) and V (e jω ) = H 1(e jω )H 2 (e jω )W (e jω ).

Eliminating W (e jω ) from these two equations we get

(1 −H 1(e jω )H 2(e jω )H 5(e jω ))V (e jω ) =H 1(e jω )H 2(e jω )U (e jω )

which leads to the frequency response of the feedback structure given by

H o ( e jω ) =

V ( e jω )

=

H 1(e jω )H 2 (e jω )

U (e jω ) 1 − H 1(e jω )H 2 (e jω )H 5(e jω )
The overall frequency of Figure P2.2(a) is thus given by
H ( e jω ) = H 4 ( e jω ) + H o ( e jω ) H 3 ( e jω ) = H 4 ( e jω ) +

H 1(e jω )H 2 (e jω )

1 − H 1(e jω )H 2 (e jω )H 5(e jω )

3.48 F {h1[ n]} = H1 (e jω ) = F {2δ[ n − 2] − 3δ[ n + 1]} = 2e − j 2ω − 3e jω ,
F {h2 [ n]} = H 2 (e jω ) = F {δ[ n − 1] + 2δ[ n + 2]} = e − jω + 2e j 2ω ,

F {h3 [ n]} = H 3 (e jω ) = F {5δ[ n − 5] + 7δ[ n − 3] + 2δ[ n − 1] − δ[ n] + 3δ[ n + 1]}
= 5e − j 5ω + 7e − j 3ω + 2e − jω − 1 + 3e jω .
The overall frequency of Figure P2.3 is given by
H (e jω ) = H 3(e jω ) + H 1(e jω ) H 2 (e jω ) = 5e − j 5ω + 7e − j 3ω + 2e − jω − 1 + 3e jω

(

)(

)

+ 2e − j 2ω − 3e jω e − jω + 2e j 2ω = 5e − j 5ω + 9e − j 3ω + 2e − jω + 3e jω − 6e j 3ω .
3.49 Now, hev [n] is the inverse DTFT of H re (e jω ). Rewriting we get
⎛ e jω + e − jω ⎞ ⎛ e j 2ω + e − j 2ω ⎞ ⎛ e j 3ω + e − j 3ω ⎞
⎟ + 3⎜
⎟ + 4⎜

H re (e jω ) = 1 + 2⎜⎜
⎟ ⎜
⎟ ⎜

2
2
2

⎠ ⎝
⎠ ⎝

= 1 + e jω + e − jω + 1.5e j 2 ω + 1.5e − j 2ω + 2e j 3ω + 2e − j 3ω . Its inverse DTFT is
hev [ n] = δ[ n] + δ[ n + 1] + δ[ n − 1] + 1.5δ[ n + 2] + 1.5δ[ n − 2] + 2δ[ n + 3] + 2δ[ n − 3].

Since h[n] is real and causal, and its DTFT H (e jω ) exists, it is also absolutely
summable. Hence, we can reconstruct h[n] from hev [n] as

h[ n] = 2hev [ n]µ[ n] − hev [0]δ[ n] = 2(δ[ n] + δ[ n − 1] + 1.5δ[ n − 2] + 2δ[ n − 3]) − δ[ n]
= δ[ n] + 2δ[ n − 1] + 3δ[ n − 2] + 4δ[ n − 3].

Not for sale

63

.

π
π
1
1
πn
3.50 (a) y a [ n] = sin⎛⎜ ⎞⎟ cos(ωo n) = sin⎛⎜ (ωo + )n ⎞⎟ − sin⎛⎜ (ωo − )n ⎞⎟. Hence, the angular
2
3 ⎠ 2
3 ⎠
⎝ 3⎠

π

frequencies present in the output are (ωo ± )n.
3
1 1
(b) yb [ n] = cos3 (ωo n) = cos 2 (ωo n) cos(ωo n) = ⎡ + cos(2ωo n)⎤ cos(ωo n)
⎢⎣ 2 2
⎥⎦

= cos(ωo n) + cos(2ωo n) cos(ωo n) = cos(ωo n) + [cos(3ωo n) + cos(ωo n)]
2
2
2
4
1

1

3

1

1

1

= cos(ωo n) + cos(3ωo n). Hence, the angular frequencies present in the output are
4
4
3ωo and ωo .
(c) yc [ n] = cos(3ωo n). Hence, the angular frequency present in the output is 3ωo .
3.51 F {δ[ n] − αδ[ n − R]} = H (e jω ) = 1 − α e − jωR . Let α = α e jφ . Then the maximum value
of H (e jω ) is 1 + α and the minimum value is 1 − α . There are R peaks of H (e jω )
located at ω = 2 πk / R, 0 ≤ k ≤ R − 1, and R dips located at
ω = (2 k + 1)π / R, 0 ≤ k ≤ R − 1 in the frequency range 0 ≤ ω < 2π.
2

2

Magnitude

1.5
1
0.5
0

0

0.5

M −1

1
/

1.5

3.52 G(e jω ) = ∑ α n e − jωn =
n=0
j0

to have G(e

K=

1− α
1− α M

0
-1
-2

2

1 − α M e − jωM
1 − αe

1

− jω

0

0.5

1
/

1.5

2

. Note G(e jω ) = H (e jω ) for α = 1. In order

) = 1, the impulse response should be multiplied by a factor K , where

.

3.53 H (e jω ) = [a1 cos(2ω) + ( a 2 + a3 ) cos(ω) + a 3 ] + j [− a1 sin(2ω) + ( a 2 − a 3 ) sin(ω)].
The frequency response will have zero-phase for a 2 = a3 and a1 = 0 .
3.54 H (e jω ) = a1 + a 2 e − jω + a 3 e − j 2ω + a 4 e − j 3ω + a 5 e − j 4 ω

(

)

(

)

= a1e j 2 ω + a 5 e − j 2ω e − j 2 ω + a 2 e jω + a 4 e − jω e − j 2ω + a 3 e − j 2ω .

Not for sale

64

If a1 = a 5 and a 2 = a 4 , then we can rewrite the above equation as

H (e jω ) = [2 a1 cos(2ω) + 2 a 2 cos(ω) + a3 ]e − j 2 ω which is seen to have a linear phase.

3.55 H (e jω ) = a1e jkω + a 2 e j ( k −1)ω + a 2 e − j ( k − 2)ω + a1e − j ( k − 3)ω

= e jkω (a1 + a 2 e − jω + a 2 e − j 2ω + a1e − j 3ω )
3
2

j ( k − )ω

[

]

=e
a1 (e j 3ω / 2 + e − j 3ω / 2 ) + a 2 (e jω / 2 + e − jω / 2 ) . Hence, H (e jω ) will be a
real function of ω if k = 3 / 2, in which case we have

H (e jω ) = a1 (e j 3ω / 2 + e − j 3ω / 2 ) + a 2 (e jω / 2 + e − jω / 2 ).
3.56 F {aδ[ n] + b δ[ n − 1] + δ[ n − 2]} = H1 (e jω ) = a + b e − jω + e − j 2ω ,
1
1
F {c n µ[ n]} = H 2 (e jω ) =
, and F {d n µ[ n]} = H 3 (e jω ) =
.
− jω
1− ce
1 − d e − jω
The overall frequency response is then H (e jω ) = H1 (e jω ) H 2 (e jω ) H 3 (e jω )
=

a + b e − jω + e − j 2 ω
(1 − c e − jω )(1 − d e − jω )

H ( e jω )
=
=

2

=

. Therefore,

a + b e − jω + e − j 2 ω

a + b e jω + e j 2 ω

(1 − c e − jω )(1 − d e − jω ) (1 − c e jω )(1 − d e jω )

(a 2 + b 2 + 1) + 2b(a + 1) cos(ω) + 2 a cos(2ω)

(1 − 2c cos(ω) + c 2 )(1 − 2d cos(ω) + d 2 )

(a 2 + b 2 + 1) + 2b(a + 1) cos(ω) + 2 a cos(2ω)
(1 + c 2 + d 2 + c 2 d 2 + 2cd ) − 2(c + d )(1 + cd ) cos(ω) + 2cd cos(2ω)

. Hence,

2
H (e jω ) = 1, if a 2 + b 2 + 1 = 1 + c 2 + d 2 + c 2 d 2 + 2cd,

b( a + 1) cos( ω) = −(c + d )(1 + cd ), and a = cd. Substituting a = cd in the equation on the
left we get b = −(c + d ).
α

3.57 Y (e jω ) = X (e jω ) e j arg X (e

jω )

. Therefore, H (e jω ) =

Y ( e jω )
X ( e jω )

= X ( e jω )

α −1

.

Since H (e jω ) is real function of ω, it has zero-phase.
3.58 y[ n] = x[ n] − α y[ n − R ]. Taking the DTFT of both sides we get

Y (e jω ) = X (e jω ) − α e − jωR Y (e jω ). Hence, H (e jω ) =

Not for sale

Y ( e jω )
X ( e jω )

=

1
1+ α e − jωR

.

65

The maximum value of H (e jω ) is

1
1− α

and the minimum value is

1
.
1+ α

There are

peaks and dips in the range 0 ≤ ω < 2π. The locations of the peaks and the dips are given
α
by 1 − α e − jωR = 1 ± α or e − jωR = ± . The locations of the peaks are given by
α

ω = ωk =

2 πk
R

and the locations of the dips are given by ω = ω k =

(2 π +1)k
, 0 ≤ k ≤ R − 1.
R

Plots of the magnitude and the phase responses of H (e jω ) for α = 0.8 and R = 5 are
shown below:
5

1

Magnitude

4
3
2

0.5
0
-0.5

1
0

0.5

1
/

1.5

In this case the maximum value is

3.59 A(e

=

)=

b0 + b1e − jω
1 + a1e − jω

-1

2

1
1 − 0 .8

0

0.5

1
/

= 5 and the minimum value is

1.5

2

1
1+ 0.8

= 0.5556.

2

. Thus, we set A(e jω ) = A(e jω ) A * (e jω ) = A(e jω ) A(e − jω )

(b0 + b1e − jω )(b0 + b1e jω ) b02 + b12 + 2b0 b1 cos(ω)
=
= 1.
(1 + a1e − jω )(1 + a1e jω )
1 + a12 + 2a1 cos(ω)

Solution #1: b0 = ±1 and b1 = sgn(b0 )a1 . In which case A(e

)=±

1 + a1e − jω

= 1, a

1 + a1e − jω

trivial solution.
Solution #2: b1 = ±1 and b0 = sgn(b1 )a1 . In which case A(e

3.60 A(e

)=

b0 + b1e − jω
1 + a1e − jω

=

B0 e jφ 0 + B1e jφ1 e − jω
1 + A1e jθ e − jω

)=±

a1 + e − jω

1 + a1e − jω

.

. Thus, we set

2
B e jφ 0 + B1e jφ1 e − jω B0 e − jφ 0 + B1e − jφ1 e jω
A(e jω ) = A(e jω ) A * (e jω ) = 0

1 + A1e jθ e − jω
1 + A1e − jθ e jω

=

B02 + B12 + 2 B0 B1 cos(ω − φ1 + φ 0 )
1 + A12 + 2 A1 cos(ω − θ)

= 1.

Not for sale

66

Solution #1: B0 = ±1, B1 = sgn( B0 ) A1 , φ1 − φ 0 = θ. In which case

A(e jω ) =

⎛ 1 + A e jθ e − jω ⎞
± e jφ 0 ± A1 e j (φ 0 + θ) e − jω
1
⎟ = ±e jφ 0 implying
= ±e jφ 0 ⎜
jθ − jω
jθ − jω ⎟

1 + A1 e e
⎝ 1 + A1 e e

φ 0 = ± π. A trivial solution.
Solution #2: B1 = ±1, B0 = sgn( B1 ) A1 , φ1 − φ 0 = θ. In which case

± A1e j (φ1 − θ) ± e − jω

⎛ A e − jθ + e − jω ⎞
⎟ = ±e jφ1 implying φ = ± π.
= ± e j φ1 ⎜ 1
1
jθ − jω
jθ − jω ⎟

1 + A1 e e
⎝ 1 + A1 e e

⎛ a * + e − jω ⎞
⎟.
Hence, a non-trivia solution is A(e jω ) = e jπ ⎜ 1
⎜ 1 + a e − jω ⎟
1

A(e jω ) =

cos(ω / 2) cos(ω)
ω
3.61 (a) H a (e jω ) = cosec(ω) = cot⎛⎜ ⎞⎟ − cot(ω) =

⎝2⎠
sin(ω / 2) sin(ω)
=

j ( e jω / 2 + e − jω / 2 )
( e jω / 2 − e − jω / 2 )

=j

j (e jω + e − jω )
( e jω − e − jω )

(e jω / 2 + e − jω / 2 )(e jω − e − jω ) − (e jω / 2 − e − jω / 2 )(e jω + e − jω )

(e jω / 2 − e − jω / 2 )(e jω − e − jω )

e − jω − e − j 2 ω
⎟. Therefore, the input-output relation is given by
= j 2⎜
⎜ 1 − e − jω − e − j 4 ω + e − j 3ω ⎟

y[ n] − y[ n − 1] − y[ n − 2] + y[ n − 3] = j 2 x[ n − 1] − j 2 x[ n − 2].

1
2
2 e − jω
. Therefore, the input-output
=
=
cos(ω) e jω + e − jω 1 + e − j 2ω
relation is given by y[ n] + y[ n − 2] = 2 x[ n − 1].
(b) H b (e jω ) = sec(ω) =

(c) H c (e jω ) = cot(ω) =
=

j (1+ e − j 2 ω )
1− e − j 2 ω

.

⎛ e jω + e − jω ⎞ j ( e jω + e − jω )
cos(ω)
j e − j ω ( e j ω + e − jω )
⎟⎟ =
=
= j ⎜⎜
sin(ω)
e jω − e − jω
1− e − j 2 ω
⎝ e jω − e − jω ⎠

Therefore, the input-output relation is given by

y[ n] − y[ n − 2] = j x[ n] + j x[ n − 2].

sin(ω / 2)
− j + j e − jω
1 − e − jω
ω
e j ω / 2 − e − jω / 2
(d) H d (e jω ) = tan⎛⎜ ⎞⎟ =
=
=
.
==
⎝ 2 ⎠ cos(ω / 2) j (e jω / 2 + e − jω / 2 ) j (1+ e − jω )
1 + e − jω
Therefore, the input-output relation is given by y[ n] + y[ n − 1] = − j x[ n] + j x[ n − 1].

3.62 From Eq. (2.20), the input-output relation of a factor-of- L up-sampler is given by
x[ n / L ], n = 0, ± L, ± 2 L, ± 3L,K
y[ n] = ⎧⎨
The DTFT of y[n] is thus given by
otherwise.
⎩ 0,

Not for sale

67

n = −∞

n = −∞
n = mL

Y (e jω ) = ∑ y[ n] e − jωn = ∑ x[ n / L ] e − jωn =

∑ x[ m ] e

− jωmL

m = −∞

= X (e jLω ), where

X (e jω ) = F {x[ n]}.
3.63 G(e jω ) =

X ( e jω ) =

1
1− αe
1

− jLω
− jω

, α < 1. Thus, we can write G(e jω ) = X (e jLω ), where

. From Table 3.3, the inverse DTFT of X (e jω ) is x[ n] = α n µ[ n].

1− αe
Hence, from the results of Problem 3.62, it follows that
x[ n / L ], n = 0, ± L, ± 2 L, ± 3L,K
g[ n] = ⎧⎨
otherwise.
⎩ 0,
3.64 From Table 3.3, H (e jω ) =

1

. Thus, H (e jω ) =

1

and
1.25 − cos(ω)
1 − 0.5 e − jω
⎛ − 0.5 sin(ω) ⎞
⎟⎟. H (e ± jπ / 5 ) = 1.3504 m j 0.6664. Therefore
arg{H (e jω )} = θ(ω) = tan −1 ⎜⎜
1
0
.
5
cos(
)

ω

H (e ± jπ / 5 ) = 1.5059 and θ( ± π / 5) = m 0.4585 radians.

Now, for an input x[ n] = sin(ωo n)µ[ n], the steady-state output is given by
y[ n] = H (e jω o ) sin (ωo n + θ(ωo ) ). For ωo = π / 5, the steady-state output is therefore
π
π
π
given by y[ n] = H (e jπ / 5 ) sin⎛⎜ n + θ( ) ⎞⎟ = 1.5059 sin sin⎛⎜ n − 0.4585 ⎞⎟.
5
5

⎝5

3.65 H (e jω ) = h[0] + h[1] e − jω + h[0] e − j 2ω = h[0](1 + e − j 2ω ) + h[1] e − jω

= e − jω (2h[0] cos(ω) + h[1]). We require H (e j 0.3 ) = 2h[ 0] cos(0.3) + h[1] = 0
and H (e j 0.6 ) = 2h[ 0] cos(0.6) + h[1] = 1. Solving these two equations we get

h[0] = 3.8461 and h[1] = −6.3487.
3.66 H (e jω ) = h[0](1 + e − j 2ω ) + h[1] e − jω = e − jω (2h[0] cos(ω) + h[1]). We require
H (e j 0.3 ) = 2h[ 0] cos(0.3) + h[1] = 1 and H (e j 0.6 ) = 2h[ 0] cos(0.6) + h[1] = 0. Solving

these two equations we get h[0] = −3.8461 and h[1] = 7.3487.
3.67
3.68 H (e jω ) = h[0](1 + e − j 4ω ) + h[1] (e − jω + e − j 3ω ) + h[2] e − j 2ω

Not for sale

68

= e − j 2ω (2h[0] cos(2ω) + 2h[1] cos(ω) + h[2]). We require
H (e j 0.2 ) = 2 cos(0.4)h[ 0] + 2 cos(0.2)h[1] + h[ 2] = 0,
H (e j 0.5 ) = 2 cos(1.0)h[ 0] + 2 cos(0.5)h[1] + h[ 2] = 1,
H (e j 0.8 ) = 2 cos(1.6)h[ 0] + 2 cos(0.8)h[1] + h[ 2] = 0. Solving these three equations we

get h[ 0] = −13.4866, h[1] = 45.228, h[ 2] = −63.8089, i.e.,
{h[ n]} = {− 13.4866, 45.228, − 63.8089}, 0 ≤ n ≤ 2.
3.69 H (e jω ) = h[0] + h[1] e − jω + h[2] e − j 2ω + h[3] e − j 3ω . Therefore,

H (e j 0 ) = h[0] + h[1] + h[2] + h[3] = 2,
H (e jπ / 2 ) = h[0] + h[1] e − jπ / 2 + h[2] e − jπ + h[3] e − j 3π / 2
= h[ 0] − j h[1] − h[ 2] + j h[3] = 7 − j 3,
H (e jπ ) = h[0] − h[1] + h[2] − h[3] = 0. Since the impulse response is real, the value of
H (e jω ) at ω = 3π / 2 is the conjugate of its value at ω = π / 2, i.e.,
H (e j 3π / 2 ) = H * (e jπ / 2 ) = h[ 0] + j h[1] − h[ 2] − j h[3] = 7 + j 3. Writing the four
equations in matrix form we get
−1
⎡h[0]⎤ ⎡1 1
1
1 ⎤⎡ 2 ⎤ ⎡ 4 ⎤
1
1⎤ ⎡ 2 ⎤
⎡1 1
⎢ h[1] ⎥ ⎢1 − j − 1 j ⎥ ⎢7 − j 3⎥ 1 ⎢1 j − 1 − j ⎥ ⎢7 − j 3⎥ ⎢ 2 ⎥
⎢h[2]⎥ = ⎢1 − 1 1 − 1⎥ ⎢ 0 ⎥ = 4 ⎢1 − 1 1 − 1⎥ ⎢ 0 ⎥ = ⎢− 3⎥, and
⎢ h[3]⎥ ⎢⎣1 j − 1 − j ⎥⎦ ⎢⎣7 + j 3⎥⎦
⎢⎣1 − j − 1 j ⎥⎦ ⎢⎣7 + j 3⎥⎦ ⎢⎣ − 1⎥⎦

hence {h[ n] = {4, 2, − 3, − 1}, 0 ≤ n ≤ 3.

3.70 (a) H (e jω ) = h[0] + h[1] e − jω + h[2] e − j 2ω + h[3] e − j 3ω

= h[0] + h[1] e − jω − h[1] e − j 2ω − h[0] e − j 3ω . Therefore,
H (e jπ / 2 ) = h[0] + h[1] e − jπ / 2 − h[1] e − jπ − h[0] e − j 3π / 2
= h[ 0] − j h[1] + h[1] − j h[ 0] = −2 + j 2,
H (e jπ ) = h[0] − h[1] − h[1] + h[0] = 8. Solving these two equations we get
h[0] = 1 and h[1] = −3. Hence, {h[ n]} = {1, − 3, 3, − 1}, 0 ≤ n p 3.
3.71 (a) H (e jω ) = h[0] + h[1] e − jω + h[2] e − j 2ω = h[0] + h[1] e − jω + h[0] e − j 2ω . The two
conditions to be satisfied by the filter are:

H (e j 0.4 π ) = h[0] + h[1] e − j 0.4 π + h[0] e − j 0.8π = 0 and
H (e j 0 ) = h[0] + h[1] + h[0] = 1. Solving these two equations we get h[0] = 0.7236 and
h[1] = −0.4472.
(b) H (e jω ) = 0.7236 − 0.4472 e − jω + 0.7236 e − j 2ω .

Not for sale

69

2

2

Magnitude

1.5
1
0.5
0

0

0.2

0.4

0.6

0.8

1
0
-1
-2

1

0

0.2

0.4

ω/pi

0.6

0.8

1

ω/pi

3.72 (a) H (e jω ) = h[0] + h[1] e − jω + h[2] e − j 2ω + h[3] e − j 3ω

= h[0] + h[1] e − jω + h[1] e − j 2ω + h[0] e − j 3ω = e − j 3ω / 2 (2h[0] cos(3ω / 2) + 2h[1] cos(ω / 2)).
The two conditions to be satisfied by the filter are:
H (e j 0.2 π ) = 2h[ 0] cos(0.3π) + 2h[1] cos(0.1π) = 0.8,
H (e j 0.5π ) = 2h[ 0] cos(0.75π) + 2h[1] cos(0.25π) = 0.5. Solving these two equations we

get h[0] = 0.0414 and h[1] = 0.395.
(b) H (e jω ) = 0.0414 + 0.395 e − jω + 0.395 e − j 2ω + 0.0414 e − j 3ω .
1

4

Magnitude

0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

2
0
-2
-4

0

0.2

ω/pi

0.4

0.6

0.8

1

ω/pi

3.73 (a) H (e jω ) = h[0] + h[1] e − jω + h[2] e − j 2ω + h[3] e − j 3ω

= h[0] + h[1] e − jω − h[1] e − j 2ω − h[0] e − j 3ω = j e − j 3ω / 2 (2h[0] sin s(3ω / 2) + 2h[1] sin(ω / 2)).
The two conditions to be satisfied by the filter are:
H (e j 0.5π ) = 2h[ 0] sin( 0.75π) + 2h[1] sin s(0.25π) = 0.2,
H (e j 0.8π ) = 2h[ 0] sin(1.2 π) + 2h[1] cos(0.4 π) = 0.7. Solving these two equations we get

h[0] = −0.14 and h[1] = 0.2815.
(b) H (e jω ) = −0.14 + 0.2815 e − jω − 0.2815 e − j 2ω + 0.14 e − j 3ω .

Not for sale

70

1

0.8

0

Magnitude

1

0.6
0.4
0.2
0

-1
-2
-3

0

0.2

0.4

0.6

0.8

-4

1

0

0.2

0.4

ω/π

0.6

0.8

1

ω/π

3.74 (a) H (e jω ) = h[0] + h[1] e − jω + h[2] e − j 2ω + h[3] e − j 3ω + h[ 4] e − j 4ω
= h[0] + h[1] e − jω − h[1] e − j 3ω − h[0] e − j 4 ω = j e − j 2ω (2h[0] sin s(2ω) + 2h[1] sin( ω)).
The two conditions to be satisfied by the filter are:

H (e j 0.5π ) = 2h[ 0] sin( 0.8π) + 2h[1] sin s(0.4 π) = 0.8,
H (e j 0.8 π ) = 2h[ 0] sin(1.6 π) + 2h[1] cos(0.8π) = 0.2. Solving these two equations we get

h[0] = 0.112 and h[1] = 0.3514.

1

4

0.8

2

Magnitude

(b) H (e jω ) = 0.112 + 0.3514e − jω − 0.3514 e − j 3ω − 0.112 e − j 4ω .

0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

0
-2
-4

0

0.2

0.4

0.6

0.8

1

ω/ π

ω/π

3.75 (a) H A (e jω ) = 0.3 − e − jω + 0.3 e − j 2ω , H B (e jω ) = 0.3 + e − jω + 0.3 e − j 2ω .

Not for sale

71

|HA(e ω )|

|HB(e ω )|
j

2

2

1.5

1.5
Magnitude

Magnitude

j

1

1

0.5
0

0.5

0

0.2

0.4

0.6

0.8

0

1

0

0.2

0.4

ω/π

0.6

0.8

1

ω/π

It can be seen from the above plots that H A (e jω ) is a highpass filter, whereas H B (e jω )
is a lowpass filter.
(b) H C (e jω ) = H B (e jω ) = H A (e j (ω + π) ).
3.76 y[ n] = y[ n − 1] + 0.5(x[ n] + x[ n − 1]). Taking the DTFT of both sides we get

(

)

Y (e jω ) = e − jω Y (e jω ) + 0.5 X (e jω ) + e − jω X (e jω ) . Hence, the frequency response is
given by H trap (e jω ) =

3.77 y[ n] = y[ n − 2] +

1
3

Y ( e jω )
X (e jω )

=

1 ⎛⎜ 1 + e − jω ⎞⎟
.
2 ⎜⎝ 1 − e − jω ⎟⎠

(x[n] + 4 x[ n − 1] + x[ n − 2]).

− jω
+ e − j 2ω
1 ⎛1+ 4e
H simpson (e jω ) = ⎜
3⎜
1 − e − j 2ω

Hence,

⎟.

2
Trapezoidal
Simpson

Magnitude

1.5
1
0.5
0

0

0.2

0.4

0.6

0.8

1

ω/π

Note: To compare the performances of the Trapezoidal numerical integration formula
with that of the Simpson’s formula, we first observe that if the input is x a (t ) = e jωt , then
the result of integration is y a (t ) = e jωt . Thus, the desired ideal frequency response is

1

Not for sale

72

H (e jω ) =

1
. Hence, we take the ratio of the frequency responses of the approximation

to the ideal, and plot the two curves as indicated on the previous page. From this plot, it
is evident that the Simpson’s formula amplifies high frequencies, whereas, the
trapezoidal formula attenuates them. In the very low frequency range, both formulae
yield results close to the ideal. However, Simpson’s formula is reasonably accurate for
frequencies close to the midband range.
3.78 G (e jω ) = g 0 + g1e − jω + g 2 e − j 2 ω + g 3 e − j 3ω

(

= e − j 3ω / 2 g 0 e j 3ω / 2 + g1e jω / 2 + g 2 e − jω / 2 + g 0 e − j 3ω / 2

)

= e − j 3ω / 2 [ g 0 cos(3ω / 2) + jg 0 sin(3ω / 2) + g 3 cos( −3ω / 2) + jg 3 sin( −3ω / 2)
+ g1 cos(ω / 2) + jg1 sin(ω / 2) + g2 cos(−ω / 2) + jg2 sin(−ω / 2)] ]
= e − j 3ω / 2 [( g 0 + g3 ) cos(3ω / 2) + ( g1 + g 2 ) cos(ω / 2)
+ j ( g 0 − g3 ) sin(3ω / 2) + j ( g1 − g 2 ) cos(ω / 2)].

Thus, if g 0 = g3 and g1 = g 2 , then
G (e jω ) = e − j 3ω / 2 [( g 0 + g3 ) cos(3ω / 2) + ( g1 + g 2 ) cos(ω / 2)] which has a linear phase

θ(ω) = −

, and
2

hence, a constant group delay.

Alternately, if g 0 = − g3 and g1 = − g 2 , then
G (e jω ) = j e − j 3ω / 2 [( g 0 + g 3 ) sin(3ω / 2) + ( g1 + g 2 ) sin(ω / 2)] which has a linear phase

θ(ω) = −

3ω π
+ , and
2
2

hence, a constant group delay.

⎛ − b sin ω ⎞
⎟⎟ .
3.79 (a) H a (e jω ) = a + b e − jω = a + b cos ω − j b sin ω. Thus, θ H a (ω) = tan −1 ⎜⎜
⎝ a + b cos ω ⎠
d θ H a (ω)
d ⎛ − b sin ω ⎞
1
Hence, τ H a (ω) = −
=−

2 dω ⎝ a + b cos ω ⎠

b
sin

ω

1+ ⎜

⎝ a + b cos ω ⎠
=−
=−

(a + b cos ω) 2
2

(a + b cos ω) +(b sin ω)

2

− (a + b cos ω)b cos ω − (−b sin ω)(−b sin ω)

− ab cos ω − b 2 cos 2 ω − b 2 sin 2 ω
a 2 + b 2 cos 2 ω + 2 ab cos ω + b 2 sin 2 ω

Not for sale

(a + b cos ω) 2
=

b 2 + ab cos ω
a 2 + b 2 + 2 ab cos ω

.

73

(b) Let G b (e jω ) = 1 + c e − jω = 1 + c cos ω − j c sin ω. From the results of Part (a) we
have τ G b (ω) =

c 2 + c cos ω
2

1 + c + 2c cos ω

. Since H b (e jω ) =

1
G b (e

θ H b (ω) = −θGb (ω), we have τ H b (ω) = − τ G b (ω) = −
a + b e − jω

(c) H c (e jω ) =

1+ ce

− jω

=
)

1
1 + c e − jω

c 2 + c cos ω
1 + c 2 + 2c cos ω

,

.

= H a (e jω )H b (e jω ), where H a (e jω ) is the frequency

response of Part (a) and H b (e jω ) is the frequency response of Part (b). Thus,
θ H c (ω) = θ H a (ω) + θ H b (ω), and therefore, τ H c (ω) = τ H a (ω) + τ H b (ω)
=

b 2 + ab cos ω

c 2 + c cos ω

.
a 2 + b 2 + 2 ab cos ω 1 + c 2 + 2c cos ω
1
1
(d) H d (e jω ) =

= H b (e jω )H e (e jω ), where H b (e jω ) is the
− jω
− jω
1+ ce
1+ d e

frequency response of Part (b) and H e (e jω ) is similar in form to H b (e jω ) . Thus,
θ H d (ω) = θ H d (ω) + θ H e (ω), and therefore, τ H d (ω) = τ H b (ω) + τ H e (ω)
=

c 2 + c cos ω
1 + c 2 + 2c cos ω

+

d 2 + d cos ω
1 + d 2 + 2 d cos ω

.

3.80

The group delay of a causal LTI discrete-time system with a frequency response
dθ(ω)
H (e jω ) = H (e jω ) e j θ(ω) is given by τ g (ω) = −
. Now,

d H ( e jω )
dH (e jω )
dθ(ω)
. Hence,
= e jθ(ω)
+ j H (e jω ) e jθ(ω)

− j H (e

)e

jθ(ω)

d H (e jω ) dH (e jω )
dθ(ω)
jθ(ω)
, or, equivalently,
=e

d H ( e jω )
dθ(ω)
e jθ(ω)
dH (e jω )
1

=

j H (e jω ) e jθ(ω)
j H (e jω ) e jθ(ω)
=

1
j H ( e jω )

d H ( e jω )

+j

dH (e jω )
. The first term on the rightjω
ω
d
H (e )
1

hand side of the above equation is purely imaginary. Hence,

Not for sale

74

⎧ dH (e jω ) ⎫
⎪ j dω ⎪
dθ(ω)
.
τ g (ω) = −
= Re ⎨
jω ⎬

⎪ H (e ) ⎪

dH (e jω )
. Rewriting Eq.

*

⎛ dH (e jω ) ⎞ ⎫

*⎫
j
ω
j
ω
j
ω

⎪ dH (e ) ⎜ j
⎟ ⎪
⎪ dH (e ) ⎜ j dH (e ) ⎟ ⎪

dω ⎟ ⎪
1 ⎪ j dω
1 ⎪ j dω

d
ω
⎟ ⎬= ⎨
(3.127) we get τ g (ω) = ⎨
+⎜
+

j
ω
j
ω
j
ω
*
j
ω
2 ⎪ H (e ) ⎜ H ( e ) ⎟ ⎪ 2 ⎪ H ( e )
H
(
e
)

⎠ ⎪⎭

⎪⎩

*

jω ⎞

⎪ j dH (e ) H * (e jω ) + ⎜ j dH (e ) ⎟ H (e jω ) ⎪

dω ⎟

1⎪

= ⎨

2
2⎪

H (e )

*
1
=
Gre (e jω ) + j Gim (e jω ) H re (e jω ) + j H im (e jω )
2
2 H ( e jω )
Since G(e jω ) is the Fourier transform of ng[n], G(e jω ) = j

3.81

[(

1

=

2 H (e
=

)

2

)(

[2G

)

(

)(

+ Gre (e jω ) + j Gim (e jω ) H re (e jω ) + j H im (e jω )
re (e

) H re (e jω ) + 2Gim (e jω ) H im (e jω )

Gre (e jω ) H re (e jω ) + Gim (e jω ) H im (e jω )
H (e

)

2

]

)* ]

.

3.82 (a) H a (e jω ) = 1 + 0.4 e − jω = 1 + 0.4 cos ω − j 0.4 sin ω and thus,

{

}

{

}

Re H a (e jω ) = 1 + 0.4 cos ω and Im H a (e jω ) = −0.4 sin ω. .

G a ( e jω ) = j

{

dH a (e jω )
= j (− j 0.4 e − jω ) = 0.4 cos ω − j 0.4 sin ω. Thus,

}

{

}

Re G a (e jω ) = 0.4 cos ω and Im G a (e jω ) = −0.4 sin ω. . Therefore, using Eq. (3.128) we

get τ H a (ω) =
=

(1 + 0.4 cos ω)(0.4 cos ω) + (−0.4 sin ω)2
(1 + 0.4 cos ω)2 + (−0.4 sin ω)2

0.16 cos 2 ω + 0.16 sin 2 ω + 0.4 cos ω
1 + 0.16 cos 2 ω + 0.16 sin 2 ω + 0.8 cos ω

=

Not for sale

0.16 + 0.4 cos ω
.
1.16 + 0.8 cos ω

75

(b) Let Gb (e jω ) =

1
H b (e

)

= 1 + 0.6 e − jω . Then τ G b (ω) = −τ H b (ω). Then using the

same procedure as in Part (a) we get τ G b (ω) =
τ H b (ω) = −

0.36 + 0.6 cos ω
.
1.36 + 1.2 cos ω

0.36 + 0.6 cos ω
. Therefore,
1.36 + 1.2 cos ω

1
⎟ = G a (e jω )Gb (e jω ), where
(c) Let H c (e jω ) = (1 − 0.5 e − jω )⎜
⎜ 1 + 0.3 e − jω ⎟

1
. Therefore,
G a (e jω ) = 1 − 0.5 e − jω and Gb (e jω ) =
1 + 0.3 e − jω
τ H c (ω) = τ G a (ω) + τ G b (ω). Then using the same procedure as in Part (a) we get

0.25 − 0.5 cos ω
and using the same procedure as in Part (b) we get
1.25 − cos ω
0.09 + 0.3 cos ω
0.25 − 0.5 cos ω 0.09 + 0.3 cos ω
τ G b (ω) = −
. Hence, τ H c (ω) =
.

1.09 + 0.6 cos ω
1.25 − cos ω
1.09 + 0.6 cos ω
τ G a (ω) =

⎞⎛

1
1
⎟⎜
⎟ = G a (e jω )Gb (e jω ), where
(d) Let H d (e jω ) = ⎜
⎜ 1 − 0.3 e − jω ⎟⎜ 1 + 0.5 e − jω ⎟

⎠⎝

1
1
G a ( e jω ) =
and Gb (e jω ) =
. Therefore,
− jω
1 − 0.3 e
1 + 0.5 e − jω
τ H d (ω) = τ G a (ω) + τ G b (ω). Then using the same procedure as in Part (b) we get
τ G a (ω) =

− 0.09 + 0.3 cos ω
− 0.25 − 0.5 cos ω
and τ G a (ω) =
. Hence,
1.09 − 0.6 cos ω
1.25 + cos ω

⎛ 0.09 − 0.3 cos ω 0.25 + 0.5 cos ω ⎞
⎟.
+
τ G a (ω) = −⎜⎜
1.25 + cos ω ⎟⎠
⎝ 1.09 − 0.6 cos ω
1

3.83 From Table 3.3, H (e jω ) =
H ( e jω ) =

1 + 0.5 e − jω
1
2

=

(1 + 0.5 cos ω) + (0.5 sin ω)

2

1
. Thus,
1 + 0.5 cos ω − j 0.5 sin ω
=

1
1.25 + cos ω

and

⎛ 0.5 sin ω ⎞
⎛ − 0.5 sin ω ⎞
⎟⎟. Now
⎟⎟ = tan −1 ⎜⎜
arg{H (e jω )} = θ(ω) = − tan −1 ⎜⎜
⎝ 1 + 0.5 cos ω ⎠
⎝ 1 + 0.5 cos ω ⎠
1
H ( e jπ / 5 ) =
= 0.6821 + j 0.1427. Therefore,
1 + 0.5 cos(π / 5) − j 0.5 sin(π / 5)

Not for sale

76

⎛ 0.1427 ⎞
H (e jπ / 5 ) = 0.6969 and arg{H (e jπ / 5 )} = θ(π / 5) = tan −1 ⎜
⎝ 0.6821 ⎠

Since for a frequency response with real coefficient impulse response, H (e jω ) is an
even function of ω and θ(ω) is an odd function of ω, we have H (e − jπ / 5 ) = 0.6969 and
θ( − π / 5) = −0.2063.
Now, for an input x[ n] = sin(ωo n) µ[ n], the steady-state output is given by

y[ n] = H (e jω o ) sin (ωo n + θ(ωo ) ). Thus, for ωo = π / 5, the steady-state output is given
π
π
by y[ n] = H (e jπ / 5 ) sin⎛⎜ n + θ(π / 5) ⎞⎟ = 0.6969 sin⎛⎜ n + 0.2063 ⎞⎟.
⎝5

⎝5

(

)

3.84 H (e jω ) = h[0] 1 + e − j 2ω + h[1] e − jω = e − jω (2h[0] cos ω + h[1]). We require
2h[ 0] cos( 0.3) + h[1] = 1 and 2h[ 0] cos( 0.7) + h[1] = 0. Solving these two equations we
get h[0] = 2.6248 and h[1] = − 4.015.
2.5

Magnitude

2
1.5
1
0.5
0

0

0.2

0.4

0.6

0.8

1

1.2

ω

Not for sale

77

M3.1 r = 0.9, θ = 0.75
Imaginary part
2

6

0
Amplitude

Amplitude

Real part
8

4
2
0
-2

0

0.5
ω/π
Magnitude Spectrum

-8

1

0

0.5

1

ω/π
Phase Spectrum

0.5

6
Magnitude

-4
-6

8

4
2
0

-2

0

0.5
ω/π

0
-0.5
-1
-1.5
-2

1

0

0.5

1

ω/π

r = 0.7, θ = 0.5
Real part

Imaginary part
0

3

Amplitude

Amplitude

4

2
1
0

0

0.5
ω/π
Magnitude Spectrum

0

0.5

1

ω/ π
Phase Spectrum
0

4
Magnitude

-2

-3

1

5

3
2
1
0

-1

0

0.5
ω/π

1

-0.5

-1

-1.5

0

0.5

1

ω/ π

M3.2 It should be noted that Program 3_1.m uses the function freqz to determine the
samples of a DTFT that is rational function in e − jω , i.e., a ratio of polynomials in
e − jω . Their inverse DTFTs are two-sided sequences. However, all sequences of

Not for sale

78

Problem 3.19 except that in Part (b) are two-sided finite-length sequences of length

2 N + 1 , and their DTFTs have both positive and negative powers of e jω . As a result,
the frequency sample computed using freqz should be multiplied by the vector
e jω n N evaluated at the frequency points ω n used in freqz. In Parts (a), (c) and (d),
the phase spectra are the plots of the unwrapped phase obtained using the function
unwrap.

Moreover, the DTFTs of the sequences in Parts (a), (c) and (d) are real functions of ω
and thus have zero phase. More accurate plots of the DTFTs are obtained using the
function zerophase.

sin(21ω / 2 )
1, − 10 ≤ n ≤ 10,
Y1 (e jω ) =
. The plots obtained using
(a) y1[ n] = ⎧⎨
0
,
otherwise,
sin(ω / 2)

Program 3_1.m are shown below:
-14

20

1

10
0
-10

Magnitude

Amplitude

2

0

0.5
ω/π
Magnitude Spectrum

0.5

1

Phase Spectrum

20

0

15
10
5
0.5
ω/π

0

ω/π
2

0

Imaginary part

-1

25

0

x 10

0

-2

1

Amplitude

Real part
30

-2
-4
-6
-8

1

0

0.5

1

ω/π

The plot obtained using the function zerophase is shown below:
25

Amplitude

20
15
10
5
0
-5

0

0.2

0.4

0.6

0.8

1

ω/π

Not for sale

79

⎛ sin(ω[ N + 1] / 2) ⎞
1, 0 ≤ n ≤ N ,
(b) y2 [ n] = ⎧⎨
Then Y2 (e jω ) = e − jωN / 2 ⎜⎜
⎟⎟ . The plots
0
,
otherwise.

⎝ sin(ω / 2) ⎠
obtained using Program 3_1.m are shown below:
Real part

Imaginary part

15

8
6
Amplitude

Amplitude

10
5
0
-5

0

0.5
ω/π
Magnitude Spectrum

-2

1

0

0.5

1

ω/π
Phase Spectrum
3

Magnitude

2
0

15

10

5

0

4

0

0.5
ω/π

2
1
0
-1
-2

1

0

0.5

1

ω/π

n
⎧⎪
1 sin 2 (ωN / 2 )
. The plots obtained
(c) y3 [ n] = ⎨1 − N , − N ≤ n ≤ N , Y3 (e jω ) = ⋅
2
N
ω
sin
(
/
2
)
⎪⎩ 0,
otherwise.
using Program 3_1.m are shown below:
Imaginary part
1.5

5

1

Amplitude

Amplitude

Real part
10

0

-5

0

0.5
ω/π
Magnitude Spectrum

0.5

0

1

1

Phase Spectrum
4

8
Magnitude

0.5

ω/π

10

6
4
2
0

0

0

0.5
ω/π

1

3
2
1
0

0

0.5

1

ω/π

Not for sale

80

The plot obtained using the function zerophase is shown below:
25

Amplitude

20
15
10
5
0
-5

0

0.2

0.4

0.6

0.8

1

ω/π

1
sin⎛⎜ ω[ N + ] ⎞⎟
n
2
⎧⎪
2 ⎠ 1 sin (ωN / 2 )
+ ⋅
.
(d) y 4 [ n] = ⎨ N + 1 − N , − N ≤ n ≤ N , Y4 (e jω ) = N ⋅ ⎝
sin(ω / 2)
N sin 2 (ω / 2)
⎪⎩
0,
otherwise.
-14

Real part
40

2

20
10
0
-10

0

0.5
ω/ π
Magnitude Spectrum

-1

0

0.5

1

ω/ π
Phase Spectrum

5

30
Magnitude

0

-2

1

40

20
10
0

Imaginary part

1
Amplitude

Amplitude

30

x 10

0

0.5
ω/ π

0

-5

-10

1

0

0.5

1

ω/ π

The plot obtained using the function zerophase is shown below:
25

Amplitude

20
15
10
5
0
-5

0

0.2

0.4

0.6

0.8

1

ω/π

Not for sale

81

cos(πn / 2 N ), − N ≤ n ≤ N ,
(e) y5 [ n] = ⎧⎨
0,
otherwise.

Y5 (e

(

π

1

)

(

π

1

)

1 sin (ω − 2 N )( N + 2 ) 1 sin (ω + 2 N )( N + 2 )
+ ⋅
. The plots obtained using
)= ⋅
2 sin (ω + π ) / 2
2 sin (ω − π ) / 2

(

)

2N

(

)

2N

Program 3_1.m are shown below:
-14

Real part

x 10

15
0.5
Amplitude

Amplitude

10
5
0
-5

Imaginary part

0

0.5

0
-0.5

1

0

0.5

ω/π
Magnitude Spectrum

Phase Spectrum
15

Magnitude

15

10

5

0

1

ω/ π

0

0.5
ω/π

10
5
0
-5

1

0

0.5

1

ω/ π

The plot obtained using the function zerophase is shown below:
25

Amplitude

20
15
10
5
0
-5

0

0.2

0.4

0.6

0.8

1

ω/π

M3.3 (a) X (e jω ) =

0.2418(1 + 0.139 e − jω − 0.3519 e − j 2 ω + 0.139 e − j 3ω + e − j 4ω )
1 + 0.2386 e − jω + 0.8258 e − j 2ω + 0.1393 e − j 3ω + 0.4153 e − j 4 ω

The plots obtained using Program 3_1.m are shown below:

Not for sale

82

Real part

Imaginary part

1

1
0.5
Amplitude

Amplitude

0.5

0

-0.5

0

0.5
ω/ π
Magnitude Spectrum

0
-0.5
-1

1

1

Phase Spectrum
4

0.8
Magnitude

0.5

ω/π

1

0.6
0.4
0.2
0

0

0

(b) X (e jω ) =

0.5
ω/ π

2
0
-2
-4

1

0

0.5

1

ω/π

0.1397(1 − 0.0911 e − jω + 0.0911 e − j 2ω − e − j 3ω )
1 + 1.1454 e − jω + 0.7275 e − j 2 ω + 0.1205 e − j 3ω

.

The plots obtained using Program 3_1.m are shown below:
Real part

Imaginary part

0.8

0.6
0.4
Amplitude

Amplitude

0.6
0.4
0.2

0.2
0

0
-0.2

0

0.5
ω/ π
Magnitude Spectrum

-0.2

1

1

Phase Spectrum
3

Magnitude

0.5

ω/ π

0.8
0.6
0.4
0.2
0

0

0

0.5
ω/ π

1

2
1
0
-1

0

0.5

1

ω/ π

M3.4 % Property 1

Not for sale

83

N = 8; % Number of samples in sequence
gamma = 0.5; k = 0:N-1;
x = exp(-j*gamma*k); y = exp(-j*gamma*fliplr(k));
% r = x[-n] then y = r[n-(N-1)]
% so if X1(exp(jw)) is DTFT of x[-n], then
% X1(exp(jw)) = R(exp(jw)) = exp(jw(N-1))Y(exp(jw))
[Y,w] = freqz(y,1,512);
X1 = exp(j*w*(N-1)).*Y;
m = 0:511; w = -pi*m/512;
X = freqz(x,1,w);
% Verify X = X1
% Property 2
k = 0:N-1; y = exp(j*gamma*fliplr(k));
[Y,w] = freqz(y,1,512);
X1 = exp(j*w*(N-1)).*Y;
[X,w] = freqz(x,1,512);
% Verify X1 = conj(X)
% Property 3
y = real(x);
[Y3,w] = freqz(y,1,512);
m = 0:511; w0 = -pi*m/512;
X1 = freqz(x,1,w0);
[X,w] = freqz(x,1,512);
% Verify Y3 = 0.5*(X+conj(X1))
% Property 4
y = j*imag(x); [Y4,w] = freqz(y,1,512);
% Verify Y4 = 0.5*(X-conj(X1))
% Property 5
k = 0:N-1; y = exp(-j*gamma*fliplr(k));
xcs = 0.5*[zeros(1,N-1) x] + 0.5*[conj(y) zeros(1,N-1)];
xacs = 0.5*[zeros(1,N-1) x] - 0.5*[conj(y) zeros(1,N-1)];
[Y5,w] = freqz(xcs,1,512);
[Y6,w] = freqz(xacs,1,512);
Y5 = Y5.*exp(j*w*(N-1));
Y6 = Y6.*exp(j*w*(N-1));
% Verify Y5 = real(X) and Y6 = j*imag(X)
M3.5 N = 8; % Number of samples in sequence
gamma = 0.5; k = 0:N-1;
x = exp(gamma*k); y = exp(gamma*fliplr(k));
xev = 0.5*([zeros(1,N-1) x] + [y zeros(1,N-1)]);
xod = 0.5*([zeros(1,N-1) x] - [y zeros(1,N-1)]);
[X,w] = freqz(x,1,512);
[Xev,w] = freqz(xev,1,512);
[Xod,w] = freqz(xod,1,512);
Xev = exp(j*w*(N-1)).*Xev;

Not for sale

84

Xod = exp(j*w*(N-1)).*Xod;
% Verify real(X) = Xev, and j*imag(X) = Xod
M3.5 N = input('The length of the seqeunce = ');
k = 0:N-1; gamma = -0.5;
g = exp(gamma*k);
% g is an exponential seqeunce
h = sin(2*pi*k/(N/2));
% h is a sinusoidal sequence with period = N/2
[G,w] = freqz(g,1,512); [H,w] = freqz(h,1,512);
% Property 1
alpha = 0.5; beta = 0.25;
y = alpha*g+beta*h; [Y,w] =
freqz(y,1,512);
% Plot Y and alpha*G+beta*H to verify that they are equal
% Property 2
n0 = 12; % Sequence shifted by 12 samples
y2 = [zeros(1,n0) g];
[Y2,w] = freqz(y2,1,512);
G0 = exp(-j*w*n0).*G;
% Plot G0 and Y2 to verify they are equal
% Property 3
w0 = pi/2; % the value of omega0 = pi/2
r = 256; % the value of omega0 in terms of number of
samples
k = 0:N-1; y3 = g.*exp(j*w0*k);
[Y3,w] = freqz(y3,1,512);
k = 0:511;
w = -w0+pi*k/512; % creating G(exp(w-w0))
G1 = freqz(g,1,w);
% Compare G1 and Y3
% Property 4
k = 0:N-1; y4 = k.*g;
[Y4,w] = freqz(y4,1,512);
% To compute derivative we need sample at pi
y0 = ((-1).^k).*g;
G2 = [G(2:512)' sum(y0)]';
delG = (G2-G)*512/pi;
% Compare Y4, delG
% Property 5
y5 = conv(g,h);
[Y5,w] = freqz(y5,1,512);
% Compare Y5 and G.*H
% Property 6

Not for sale

85

y6 = g.*h;
[Y6,w] = freqz(y6,1,512,'whole');
[G0,w] = freqz(g,1,512,'whole');
[H0,w] = freqz(h,1,512,'whole');
% Evaluate the sample value at w = pi/2
% and verify with Y6 at pi/2
H1 = [fliplr(H0(1:129)') fliplr(H0(130:512)')]';
val = 1/(512)*sum(G0.*H1);
% Compare val with Y6(129) i.e., sample at pi/2
% Can extend this to other points similarly
% Parsevals theorem
val1 m = sum(g.*conj(h)); val2 = sum(G0.*conj(H0))/512;
% Compare val1 with val2
M3.7 The DTFT of nh[n] is j

dH ( e jω )
.

Hence, the group delay τ g (ω) can be computed at a

set of N discrete frequency points ω k = 2πk / N , 0 ≤ k ≤ N − 1, as follows:
DFT {n h[ n]} ⎞
τ g (ωk ) = Re⎛⎜
⎟,
⎝ DFT {h[ n]} ⎠
where all DFTs are N –points in length with N greater than or equal to the length of
{h[ n]}.
M3.8 h = [3.8461 -6.3487 3.8461];
[H,w] = freqz(h,1,512);
plot(w/pi,abs(H)); grid
xlabel('\omega/\pi'); ylabel('Magnitude');

Magnitude

15

10

5

0

0

0.2

0.4

0.6

0.8

1

ω/π

M3.9 h = [-13.4866 45.228 -63.8089 45.228 -13.4866];
[H,w] = freqz(h,1,512);
plot(w/pi,abs(H)); grid
xlabel('\omega/\pi'); ylabel(Magnitude);

Not for sale

86

200

Magnitude

150
100
50
0

0

0.2

0.4

0.6

0.8

1

ω/π

Not for sale

87

Chapter 4
4.1

Let φ(t ) be an arbitrary continuous-time function with a CTFT Φ ( jΩ ), where

~
Φ( jΩ) = ∫ φ(t ) e − jΩt dt. Let φT (t ) = ∑ φ(t + nT ) denote the periodic continuous-

n = −∞

−∞

time function with a period T obtained by a periodic extension of φ(t ). Note that
~
φT (t ) is also given by the convolution of φ(t ) with the periodic impulse train

~
p(t ) = ∑ δ(t + nT ), i.e., φT (t ) = ∫ φ(τ)p(t − τ)dτ.

n = −∞

−∞

~
~
Tthe CTFT of φT (t ) is then given by F{ φT (t )} = Φ( jΩ) ⋅ F { p(t ) }

= Φ ( jΩ ) ⋅

2π ∞
∑δ
T n = −∞

where Ω T =

( j(Ω − nΩT )) =

Φ( jnΩ T )δ
n = −∞ T

( j(Ω − nΩT )),

(4-1)

.
T

~
Now, the Fourier series expansion of φT (t ) = ∑ φ(t + nT ) is given by
n = −∞

~
φT (t ) = ∑ a n e jnΩ T t . A CTFT of both sides of this equation is then
n = −∞

~
F {φT (t )} = ∑ a n ⋅ 2πδ( j (Ω − nΩ T )).

(4-2)

n = −∞

1

Comparing Eqs. (4-1) and (4-2) we arrive at a n = Φ( jnΩ T ). Substituting this
T
~
expression in the Fourier expansion of φT (t ) we therefore arrive at the Poisson’s sum

~
1 ∞
jnΩ T t
formula φT (t ) = ∑ φ(t + nT ) =
.
∑ Φ( jnΩ T ) e
T
n = −∞

n = −∞

4.2

Consider the continuous-time signal g a (t ) = sin(Ω m t ) which is bandlimited to
Ω m . If we sample ga (t ) at a rate Ω T = 2Ω m starting at t = 0 , then all its samples
are zero. Hence, ga (t ) cannot be recovered from its samples obtained by sampling
it at the Nyquist rate Ω T = 2Ω m . As a result, g a (t ) = sin(Ω m t ) must be sampled
at a rate Ω T > 2Ω m to recover it fully from its samples.

4.3

(a) Now, the CTFT of y1 (t ) is given by Y1 ( jΩ) =

1
G ( jΩ ) O
* G a ( jΩ )
2π a

where

Ga ( jΩ) denotes the CTFT of g a (t ) and O
* denotes the frequency-domain
convolution. The highest frequency present in y1 (t ) is therefore twice that of
g a (t ) and hence, the Nyquist frequency of y1 (t ) is 2Ω m .

Not for sale

86

(b) The CTFT of y2 (t ) is given by Y2 ( jΩ) = ∫ g a ( ) e − jΩt dt
3
−∞
t

= 3 ∫ g a (τ) e − j 3Ωτ dτ = 3 G a ( j 3Ω). The highest frequency present in y2 (t ) is
−∞

therefore one-third of that of g a (t ) and hence, the Nyquist frequency of y2 (t ) is
Ω m / 3.

(c) The CTFT of y3 (t ) is given by Y3 ( jΩ) = ∫ g a (3t ) e − jΩt dt
−∞

=

1 ∞
− jΩτ / 3

∫ g a ( τ) e
3
−∞

1
3

3

= G a ( j ). The highest frequency present in y3 (t ) is

therefore three times of that of g a (t ) and hence, the Nyquist frequency of y3 (t ) is
3Ω m .
(d) The CTFT of y 4 (t ) is given by
∞ ⎡∞

⎡∞

Y4 ( jΩ) = ∫ ⎢ ∫ g a (t − τ)g a (τ)dτ⎥ e − jΩt dt = ∫ g a (τ) ⎢ ∫ g a (t − τ) e − jΩt dt ⎥ dτ
− ∞ ⎣− ∞
−∞

⎣− ∞

−∞

−∞

= ∫ g a (τ) e − jΩτ G a ( jΩ) dτ = G a ( jΩ) ∫ g a (τ) e − jΩτ dτ = G a ( jΩ)G a ( jΩ). The

highest frequency present in y 4 (t ) is therefore the same as that of g a (t ) and hence,
the Nyquist frequency of y 4 (t ) is Ω m .
1 ∞

(e) Now g a (t ) =
G a ( jΩ) e jΩt dΩ. Differentiating both sides of this equation

−∞
we get
y 5 (t ) =

dg a (t )
dt
dg a ( t )
dt

=

1 ∞
jΩt
∫ jΩ G a ( jΩ) e dΩ.

−∞

Hence, it follows that the CTFT of

is simply jΩ G a ( jΩ). The highest frequency present in y5 (t ) is

therefore the same as that of g a (t ) and hence, the Nyquist frequency of y5 (t ) is
Ωm .
4.4

By Parseval’s relation, the total energy of g a (t ) is given by

2

1 ∞

1

2

Ωm

2

E g a (t ) = ∫ g a (t ) dt =
∫ G a ( jΩ) dΩ = 2 π ∫ Ga ( jΩ) dΩ. Likewise, the

−Ωm
−∞
−∞

1 π

2

total energy of g[n] is given by E g[ n] = ∑ g[ n] =
∫ G ( e ) dω
2
π
−∞
−π

Not for sale

2

87

4.5

=

Ωm
2
1 π/T
1
1 π/T 1
2
2
G
(
j

)
d
(

T
)
=
G
(
j

)
d

=
∫ T a

∫ G a ( jΩ) dΩ
a
2 πT
2 πT

−π / T
−π / T
−Ωm

=

1
T

E g a (t ) .

Sampling period T =
FT =

1
T

2 .5
= sec. Hence, the sampling frequency is
5000

= 2000 Hz. Therefore, the highest frequency component that could be

present in the continuous-time signal has a frequency

4.6

20000
= 1000 Hz.
2

Since the continuous-time signal x a (t ) is being sampled at 2 kHz rate, the sampled
version of its i -th sinusoidal component with a frequency Fi will generate discretetime sinusoidal signals with frequencies F i ±2000n, − ∞ < n < ∞. Hence, the
frequencies Fim generated in the sampled version associated with the sinusoidal
components present in are as follows:

F1 = 300 Hz ⇒ F1m = 300, 1700, 2300,K Hz
F2 = 500 Hz ⇒ F2 m = 500, 1500, 2500,K Hz
F3 = 1200 Hz ⇒ F3m = 1200, 800, 3200,K Hz
F4 = 2150 Hz ⇒ F4 m = 2150, 150, 4150,K Hz
F5 = 3500 Hz ⇒ F5m = 3500, 1500, 5500, 500, 7500,K Hz
After filtering by a lowpass filter with a cutoff at 900 Hz, the frequencies of the
sinusoidal components in y a (t ) are 150, 300, 500, 800 Hz.
4.7

One possible set of values for the frequencies present in y a (t ) are: F1 = 350 Hz,

F2 = 575 Hz, F3 = 815 Hz, and F4 = 9650 Hz. Another possible set of values for
the frequencies present in y a (t ) are: F1 = 350 Hz, F2 = 575 Hz, F3 = 815 Hz,
and F4 = 10575 Hz. Hence, the solution is not unique.
4.8

t = nT =

n
.
50

Therefore,

20 πn ⎞
24 πn ⎞
120 πn ⎞
176 πn ⎞
x[ n] = 4 sin⎛⎜
⎟ − 5 cos⎛⎜
⎟ + 3 sin⎛⎜
⎟ + 2 cos⎛⎜

⎝ 50 ⎠
⎝ 50 ⎠
⎝ 50 ⎠
⎝ 50 ⎠
2 πn ⎞
12 πn ⎞
(10 + 2)πn ⎞
(100 −12)πn ⎞
= 4 sin⎛⎜
⎟ − 5 cos⎛⎜
⎟ + 3 sin⎛⎜
⎟ + 2 cos⎛⎜

5
25
⎝ 5 ⎠
⎝ 25 ⎠

2 πn ⎞
12 πn ⎞
2 πn ⎞
(12 πn ⎞
= 4 sin⎛⎜
⎟ − 5 cos⎛⎜
⎟ + 3 sin⎛⎜
⎟ + 2 cos⎛⎜
⎟.
⎝ 5 ⎠
⎝ 25 ⎠
⎝ 5 ⎠
⎝ 25 ⎠

Not for sale

88

4.9

Both channels are being sampled at 45 kHz. Therefore, there are a total of
2 × 45000 = 90000 samples/sec. Each sample is quantized using 12 bits. Hence,
the total bit rate of the two channels after sampling and digitization is 108 kpbs.

4.10 hr (t ) =

sin(Ω c t )
sin(Ω c nT )
. Therefore, hr (nT ) =
. Since T = 2π / ΩT , we have
ΩT t / 2
Ω T nT / 2

⎛ 2 πΩ c n ⎞
sin⎜

ΩT ⎠
sin( πn)

. For Ω c = Ω T / 2, we thus have hr (nT ) =
hr (nT ) =
= δ[ n].
πn
πn
4.11 The spectrum of the sampled signal is as shown below:
Xp(jΩ)
1/T
1/2T
_2Ω

Now, T =

2Ω m

=

m

π
Ωm

_

m
__
3

0

Ωm

. As a result, ωc =

Ωmπ
3Ω m

Ωm

2 Ωm

π
3

= . Hence after lowpass filtering

the spectrum of the output continuous-time signal y a (t ) will be as shown below:
Ya (jΩ)

m
_Ω
__

0

3

m
__
3

4.12 (a) Ω1 = 100π, Ω1 = 150π. Thus, ∆Ω = Ω 2 − Ω1 = 50π. Note ∆Ω is an integer
multiple of Ω 2 . Hence, we choose the sampling angular frequency as
Ω T = 2∆Ω = 100π =

2 ×150 π
,
M

which is satisfied for M = 3. The sampling

frequency is therefore 50 Hz. The CTFT X p ( jΩ) of the sampled sequence and the
frequency response H r ( jΩ) of the desired reconstruction filter are shown below.
X p(jΩ)
M=3
1/T

_ 200π _ 150π _ 100π

_ 50π

0

50π

Not for sale

100π

150π

200π

89

H r (jΩ)
T

_

150π

_ 100π

0

100π

150π

(b) Ω1 = 160π, Ω1 = 250π. Thus, ∆Ω = Ω 2 − Ω1 = 90π. Note ∆Ω is not an
integer multiple of Ω 2 . Hence, we extend the bandwidth to the left by assuming
the lowest frequency to be Ω 0 and choose the sampling angular frequency as
Ω T = 2∆Ω = 2(Ω 2 −Ω 0 ) =

2 × 250 π
,
M

which is satisfied for Ω 0 = 125π and M = 2.

The sampling frequency is therefore 125 Hz. The CTFT X p ( jΩ) of the sampled
sequence and the frequency response H r ( jΩ) of the desired reconstruction filter
are shown below.
X p (jΩ)

1/T

_

340π

_

250π _ 160π

_

90π

0

90π

160π

250π

340π

H r (jΩ)
T

_

250π

_ 160π

0

160π

250π

(c) Ω1 = 110π, Ω1 = 180π. Thus, ∆Ω = Ω 2 − Ω1 = 70π. Note ∆Ω is not an
integer multiple of Ω 2 . Hence, we extend the bandwidth to the left by assuming
the lowest frequency to be Ω 0 and choose the sampling angular frequency as
Ω T = 2∆Ω = 2(Ω 2 −Ω 0 ) =

2 ×180 π
,
M

which is satisfied for Ω 0 = 90π and M = 2.

The sampling frequency is therefore 90 Hz. The CTFT X p ( jΩ) of the sampled
sequence and the frequency response H r ( jΩ) of the desired reconstruction filter
are shown below.
X p (jΩ)

1/T

_

250π

_

180π

_ 110π

_

70π

0

Not for sale

70π

110π

180π

250π

90

H r (jΩ)
T

_

4.13

180π

_ 110π

0

110π

180π

α p = −20 log10 (1 − δ p ) dB and α s = −20 log10 δ s dB. Therefore,
−α p / 20

and δ s = 10 −α s / 20 .
(a) α p = 0.21 dB and α s = 52 dB. Hence, δ p = 0.0239 and δ s = 0.025.
(b) α p = 0.03 dB and α s = 69 dB. Hence, δ p = 0.0034 and δ s = 0.00355.

δ p = 1 − 10

(c) α p = 0.33 dB and α s = 57 dB. Hence, δ p = 0.0373 and δ s = 0.0014.
4.14

H a (s) =

a
a
, and hence,
. Thus, H a ( jΩ) =
jΩ + a
s+a

H a ( jΩ )

2

= H a ( jΩ ) H a ( − jΩ ) =

a
a
a2

= 2 2.
j Ω + a − jΩ + a
Ω +a

As Ω increases from
2

0 to ∞ , it can be seen that the square-magnitude function H a ( jΩ) and hence,
the magnitude function H a ( jΩ) =

a
Ω2 + a2

decreases monotonically from

H a ( j 0) = 1 to H a ( j∞) = 0. Let Ω c denote the 3 –dB cutoff frequency. Then
2

H a ( jΩ c ) =
4.15

G a (s ) =
G a ( jΩ )

a2
Ω 2c + a 2

1

= , which implies Ω c = a.
2

jΩ
s
, and hence,
. Thus, G a ( jΩ) =
jΩ + a
s+a
2

= G a ( jΩ)G a (− jΩ) =

− jΩ
jΩ
Ω2

= 2 2.
jΩ + a − jΩ + a
Ω +a

As Ω increases from

0 to ∞ , it can be seen that the square-magnitude function G a ( jΩ)
the magnitude function Ga ( jΩ) =

Ω2 + a2

2

and hence,

increases monotonically from

Ga ( j 0) = 0 to G a ( j∞) = 1. Let Ω c denote the 3 –dB cutoff frequency. Then
2

G a ( jΩ c ) =
4.16

Ω c2
Ω c2 + a 2

1

= , which implies Ω c = a.
2

s
a
1⎛
s−a⎞ 1
= ⎜1 −
⎟ = ( A1 ( s ) − A2 ( s )) and G a (s ) =
s+a 2⎝ s+a⎠ 2
s+a
1⎛ s−a⎞ 1
s−a
= ⎜1 +
. Now,
⎟ = ( A1 (s ) + A2 (s )), where A1 (s) = 1 and A2 (s ) =
2⎝ s+a⎠ 2
s+a
H a (s) =

Not for sale

91

j Ω − a − jΩ − a jΩ − a jΩ + a

=

= 1 for all
j Ω + a − jΩ + a jΩ + a jΩ − a
values of Ω, A1 (s) and A2 (s) are allpass functions.
2

A1 ( jΩ) = 1 and A 2 ( jΩ) =

4.17

H a (s) =

bs

jbΩ

. Thus, H a ( jΩ) =
2

and hence,
jbΩ + Ω o2 − Ω 2
− jbΩ
jbΩ
2

H a ( j Ω ) = H a ( jΩ ) H a ( − j Ω ) =
2
2
jbΩ + Ω o − Ω − jbΩ + Ω o2 − Ω 2

=

s 2 + bs + Ω o

b2Ω2
(Ω 2o − Ω 2 ) 2 + b 2 Ω 2

. At Ω = 0, H a ( j 0) = 0, at Ω = ∞, H a ( j∞) = 0, and at

Ω = Ω o , H a ( jΩ) has the maximum value of 1. Now,
d H a ( jΩ )

2

dΩ

=

2b 2 Ω(Ω o2 − Ω 2 )(Ω o2 + Ω 2 )

(

(Ω o2

2 2

−Ω ) +b Ω

frequency range 0 ≤ Ω < Ω o ,
Ω o < Ω < ∞,

2

d H a ( jΩ)

)

2 2

d H a ( jΩ)

. It therefore follows that in the

2

dΩ

> 0, and in the frequency range

2

< 0. Hence, in the frequency range 0 ≤ Ω < Ω o ,

dΩ

2

H a ( jΩ) is a monotonically increasing function of Ω and in the frequency range
2

Ω o < Ω < ∞, H a ( jΩ) is a monotonically decreasing function of Ω. Or in other
words, H a (s) has a bandpass magnitude response. The 3 –dB cutoff frequencies
are given by the solution of
2

2 2

2

2

2

2

(Ω o − Ω c ) + b Ω c = 2 b Ω c ,

b 2 Ω 2c
(Ω 2o − Ω 2c ) 2 + b 2 Ω 2c

=

1
,
2

or,

i.e., Ω c4 − (b 2 + 2Ω o2 )Ω c2 + Ω o4 = 0. Substituting

x = Ω 2c in the last equation we get x 2 − (b 2 + 2Ω 2o ) x + Ω o4 = 0. Let x1 = Ω12 and

x 2 = Ω 22 be the two roots of this quadratic equation. Then, x1 x 2 = Ω12 Ω 22 = Ω o4
and x1 + x 2 = Ω12 + Ω 22 = b 2 + 2Ω 2o . Therefore, Ω1Ω 2 = Ω 2o . From the last two
equations we get Ω12 + Ω 22 − 2Ω1Ω 2 = (Ω 2 − Ω1 ) 2 = b 2 + 2Ω 2o − 2Ω 2o = b 2 .
Hence, Ω 2 − Ω1 = b.

Not for sale

92

4.18

G a (s) =

s 2 + Ω 2o
2

Ω o2 − Ω

2

(Ω o2 − Ω 2 ) 2
(Ω o2

Now,

2 2

2

−Ω ) +b Ω
d G a ( jΩ )
dΩ

2

2

=

(

Ω o < Ω < ∞,
2

and hence,

Ω o2 − Ω 2

jbΩ + Ω o2 − Ω 2 − jbΩ + Ω o2 − Ω 2

2b 2 Ω 3 (Ω 2 − Ω o2 )

(Ω o2

d G a ( jΩ)
dΩ

−Ω

2

. Note G a ( j 0) = G a ( j∞) = 1 and G a ( jΩ o ) = 0.

2 2

2

−Ω ) +b Ω

frequency range 0 ≤ Ω < Ω o ,

G a ( jΩ )

jbΩ + Ω o2
2

s + bs + Ω o

G a ( jΩ) = G a ( jΩ)G a (− jΩ) =
=

Ω o2 − Ω 2

. Thus, G a ( jΩ) =
2

d G a ( jΩ)

)

2 2

. It therefore follows that in the

2

dΩ

< 0, and in the frequency range

2

> 0. Hence, in the frequency range 0 ≤ Ω < Ω o ,

is a monotonically decreasing function of Ω and in the frequency
2

range Ω o < Ω < ∞, G a ( jΩ) is a monotonically increasing function of Ω. Or in
other words, G a (s ) has a bandstop magnitude response.
The 3 –dB cutoff frequencies are given by the solution of

(Ω o2 − Ω c2 ) 2

=

1

2
(Ω o2 − Ω c2 ) 2 + b 2 Ω 2c
4
2
2
2
4
Ω c − (b + 2 Ω o )Ω c + Ω o = 0. This

,

or, 2(Ω o2 − Ω c2 ) 2 = (Ω o2 − Ω 2c ) 2 + b 2 Ω 2c , i.e.,
last equation is exactly the same as in solution of Problem 4.18 from which we get

4.19

Ω1Ω 2 = Ω 2o and Ω 2 − Ω1 = b.
bs
1 ⎛ s 2 − bs + Ω 2o ⎞⎟ 1
= ( A1 (s ) − A2 (s)) and
H a (s ) =
= ⎜1 −
s 2 + bs + Ω 2o 2 ⎜⎝ s 2 + bs + Ω 2o ⎟⎠ 2
s 2 + Ω o2
1 ⎛ s 2 − bs + Ω o2 ⎞⎟ 1
= ( A1 (s) + A2 (s)), where
G a (s) =
= ⎜1 +
s 2 + bs + Ω o2 2 ⎜⎝ s 2 + bs + Ω o2 ⎟⎠ 2

A1 (s) = 1 and A2 (s ) =

s 2 − bs + Ω o2

. Now, A1 ( jΩ) = 1 and A 2 ( jΩ)
2

2

s 2 + bs + Ω o

− Ω 2 − jbΩ + Ω o2 − Ω 2 + jbΩ + Ω o2
= A2 ( jΩ) A2 (− jΩ) =

= 1, for all values of
− Ω 2 + jbΩ + Ω o2 − Ω 2 − jbΩ + Ω 2o
Ω, A1 (s) and A2 (s) are allpass functions.

Not for sale

93

4.20

(a) Let Ai (s ) =

s + λ*i
. Since the pole of Ai (s ) is strictly in the left-half s –plane
s − λi

and hence, Ai (s ) is causal and stable. Now Ai ( jΩ)
=

2

= Ai ( jΩ) Ai* ( jΩ)

jΩ + λ*i − jΩ + λ i
jΩ + λ*i jΩ − λ i

=

= 1. Hence, Ai (s ) is an allpass
jΩ − λ i − jΩ − λ*i
jΩ − λ i jΩ + λ*i
N

function. Since, A(s ) = ∏ A i (s ), it is a product of causal, stable allpass functions,
i =1

and as a result, is also a causal, stable allpass function.

(b) Ai (s )

2

2

=

Ai (s ) Ai* (s ) =

2

s + λ i + 2 Re{sλ i }
s + λ*i s * +λ i

. Let
=
2
2
s − λ i s * −λ*i
s + λ i − 2 Re{s * λ i }

s = σ + jΩ and λ i = ai + jbi . Then Ai (s )

2

=

2

2

− 2Ωbi ) + 2σai

2

2

− 2Ωbi ) − 2σai

( s + λi
( s + λi
2

Since ai < 0, it follows from the above that Ai (s ) < 1 for σ > 0, Ai (s )
for σ = 0, and Ai (s )
4.21

H a ( jΩ )

2

=

follows then

1 + (Ω / Ω c )

d k D(Ω )
dΩ k

2N

. Define D(Ω) =

1
H a ( jΩ )

= 2 N (2 N − 1)L(2 N − k + 1)

Ω=0

= 1 + (Ω / Ω c ) 2 N . It

2

Ω2N − k
Ω 2c N

= 0 for k = 1, 2,K , N − 1, or, equivalently,

for k = 1, 2,K , N − 1.
4.22

=1

> 1 for σ < 0.

1

d k D(Ω )
dΩ k

2

2

.

. Therefore,

d k H a ( jΩ )
dΩ k

=0
Ω=0

⎛ 1 ⎞
⎟ = −0.25. Therefore, ε 2 = 10 0.025 − 1 = 0.0593. Next, from
10 log10 ⎜⎜
2⎟
⎝1+ ε ⎠
⎛ 1 ⎞
1 Ω
6
⎟ = −25, we get A 2 = 10 2.5 = 316.2278. Now, = s =
10 log10 ⎜⎜
=4
2⎟
k Ω p 1 .5
⎝A ⎠
log10 (1 / k1 )
A2 − 1
1
315.2278
= 3.0943.
=
= 72.9381. Hence, N =
=
log10 (1 / k )
k1
0.0593
ε2
We choose N = 4 as the filter order.
To verify using MATLAB, we use the code fragment
[N,Wn]=buttord(2*pi*1500,2*pi*6000,0.25,25,'s');

and

Not for sale

94

which yields N = 4 and Wn = 18365.51286.
4.23 The poles are given by p l = e j π ( 5 + 2 l ) / , 1 ≤ l ≤ 6 . Hence,

p1 = e j (7π / 12) = −0.2588 + j 0.9659, p2 = e j (9π / 12) = −0.7071 + j 0.7071,
p3 = e j (9π / 12 ) = −0.9659 + j 0.2588, p 4 = e j (11π / 12 ) = p3* = −0.9659 − j 0.2588,

p5 = e j (13π / 12 ) = p2* = −0.7071 − j 0.7071, p6 = e j (15π / 12 ) = p1* = −0.2588 − j 0.9659.
The poles can also be determined in MATLAB using the statement
[z,p,k]=buttap(6) which yields

p =
-0.2588
-0.2588
-0.7071
-0.7071
-0.9659
-0.9659

+
+
+
-

0.9659i
0.9659i
0.7071i
0.7071i
0.2588i
0.2588i

4.24 From Eq. (4.41) of text, TN (Ω) = 2ΩTN −1 (Ω) − TN − 2 (Ω), where TN (Ω) is defined in Eq.
(4.40).
Case 1: Ω ≤ 1. Making use of Eq. (4.40) in Eq. (4.41) we get

(

)

(
)
= 2Ω cos(N cos −1 Ω − cos −1 Ω ) − cos(N cos −1 Ω − 2 cos −1 Ω )
= 2Ω [cos( N cos −1 Ω) cos(cos −1 Ω) + sin( N cos −1 Ω) sin(cos −1 Ω)]
− [cos( N cos −1 Ω) cos(2 cos −1 Ω) + sin( N cos −1 Ω) sin(2 cos −1 Ω)]

T N (Ω) = 2Ω cos ( N − 1) ⋅ cos −1 Ω − cos ( N − 2) ⋅ cos −1 Ω

= 2Ω cos( N cos −1 Ω) cos(cos −1 Ω) − cos( N cos −1 Ω) cos(2 cos −1 Ω)

[

]

= 2Ω 2 cos( N cos −1 Ω) − cos( N cos −1 Ω) 2 cos 2 (cos −1 Ω) − 1

[

]

= cos( N cos −1 Ω) 2Ω 2 − 2Ω 2 + 1 = cos( N cos −1 Ω).
Case 2: Ω > 1. Making use of Eq. (4.40) in Eq. (4.41) we get

(

)

(

)

T N (Ω) = 2Ω cosh ( N − 1) ⋅ cosh −1 Ω − cosh ( N − 2) ⋅ cosh −1 Ω . Using the trigonometric
identities
cosh( A − B) = cosh( A) cosh( B) − sinh( A) sinh( B), sinh( 2 A) = 2 sinh( A) cosh( A), and

cosh(2 A) = 2 cosh 2 ( A) − 1, and following a similar algebra as in Case 1, we can show
T N (Ω) = cosh( N cosh −1 Ω).

4.25

From the solution of Problem 4.22, we have

Not for sale

1
1
= 4 and
= 72.9381. Hence,
k
k1

95

cosh −1 (1 / k1 )

= 2.4151. We choose the filter order as N = 3.
cosh −1 (1 / k )
The filter order obtained using the MATLAB statement
[N,Wn]=cheb1ord(2*pi*1500,2*pi*6000,0.25, 25, 's') results in N=3.
N=

⎛ 1
4.26 10 log10 ⎜⎜
⎝ 1 + ε2

⎛ 1 ⎞
⎟⎟ = −0.25, which yields ε = 0.2434. 10 log10 ⎜⎜
⎟⎟ = −25, which
⎝ A2 ⎠

Ω p 1500
0.2434
ε
yields A 2 = 316.2278. Now, k =
=
= 0.25 and k1 =
=
=
2
Ω s 6000
315
.
2278
A −1

= 0.0137. Substituting the value of k in Eq. (4..55a) we get k ' = 0.9682. Then from Eq.
(4.55b) we get ρ 0 = 0.004. Substituting the value ρ0 in Eq. (4.55c) we get ρ = 0.004.
Finally, from Eq. (4.54) we arrive at N = 2.0591. We choose the next higher integer as the
filter order N = 3.
The filter order obtained using the MATLAB statement
[N,Wn]=ellipord(2*pi*1500,2*pi*6000,0.25, 25, 's') results in N=3.
4.27 B N ( s ) = (2 N − 1) B N −1 ( s ) + s 2 B N − 2 (s ), where B1 (s) = s + 1 and B2 (s ) = s 2 + 3s + 3.
(a) Thus, B3 ( s ) = 5 B2 ( s ) + s 2 B1 ( s ) = 5( s 2 + 3s + 3) + s 2 ( s + 1) = s 3 + 6 s 2 + 15s + 15,
B4 ( s ) = 7 B3 ( s ) + s 2 B2 ( s ) = 7( s 3 + 6 s 2 + 15s + 15) + s 2 ( s 2 + 3s + 3)
= s 4 + 10 s 3 + 45s 2 + 105s + 105.

(b) B5 ( s ) = 9 B4 ( s ) + s 2 B3 ( s ) = 9( s 4 + 10 s 3 + 45s 2 + 105s + 105) + s 2 ( s 3 + 6 s 2 + 15s + 15)
= s 5 + 15s 4 + 105s 3 + 420 s 2 + 945s + 945.
ˆ = 2 π × 3. The mapping is thus s =
4.28 Ω p = 2π × 0.24 and Ω
p

ˆ
Ω pΩ
p

=

4 π 2 × 0.72
.

Denote K = 4 π 2 × 0.72 = 28.4245. Hence, the desired highpass transfer function is given
10
by H HP (sˆ ) = H LP (s ) s = K / sˆ =
3
2
⎛⎜ K ⎞⎟ + 4.309⎛⎜ K ⎞⎟ + 9.2835⎛⎜ K ⎞⎟ + 10
⎝ sˆ ⎠
⎝ sˆ ⎠
⎝ sˆ ⎠
10 sˆ 3
10 sˆ 3
=
=
K 3 + 4.309 K 2 sˆ + 9.2835Ksˆ 2 + 10 sˆ 3 10 sˆ 3 + 263.8785 sˆ 2 + 3481.5sˆ + 22966
=

sˆ 3
sˆ 3 + 26.38785 sˆ 2 + 348.15sˆ + 2296.6

.

ˆ = 2 π × 3. The mapping is thus s =
4.29 Ω p = 2π × 0.9 and Ω
p

ˆ
Ω pΩ
p

=

4π 2 × 2.7
. Denote

K = 4 π 2 × 2.7 = 106.5917. Hence, the desired lowpass transfer function is given by

Not for sale

96

H LP (s ) = H LP (s) s = K / sˆ =

=

=

⎛⎜ K ⎞⎟
⎝ sˆ ⎠
3

3

2

⎛⎜ K ⎞⎟ + 9.238⎛⎜ K ⎞⎟ + 40.087⎛⎜ K ⎞⎟ + 100
⎝ sˆ ⎠
⎝ sˆ ⎠
⎝ sˆ ⎠

K3
K 3 + 9.238 K 2 sˆ + 40.087 Ksˆ 2 + 100 sˆ 3
12110.735

.
sˆ 3 + 42.729 sˆ 2 + 1049.602 sˆ + 12110.735
ˆ = 2 π × 3 = 6 π, Ω
ˆ
ˆ
4.30 Ω p = 2 π × 0.25 = 0.5π, Ω
o
p 2 − Ω p1 = 2 π(0.5) = π. The mapping is thus
ˆ2
⎛ sˆ 2 + 36π 2 ⎞ sˆ 2 + 36 π 2
sˆ 2 + Ω
o
⎟=
= 0.5π⎜
s = Ωp
.
ˆ
ˆ )

ˆ
ˆ
π
s
s
2

sˆ(Ω
p2
p1

2
⎡⎛ ˆ 2

s + 36 π 2 ⎞
⎟⎟ + 367.93⎥
0.01⎢⎜⎜
⎢⎝ 2 sˆ

H BP (sˆ ) = H LP (s ) s = (sˆ 2 + 36π 2 ) / 2 sˆ =
2
⎛ sˆ 2 + 36 π 2 ⎞
⎛ sˆ 2 + 36 π 2 ⎞
⎟⎟ + 3.895
⎟⎟ + 2.269⎜⎜
⎜⎜

⎝ 2 sˆ

⎝ 2 sˆ
4
2
0.01(sˆ + 2182.33 sˆ + 126242.18)
=
.
sˆ 4 + 4.538 sˆ 3 + 726.19 sˆ 2 + 1612.38 sˆ + 126242.18
ˆ = 2π × 6.5 × 10 3 and Ω
ˆ = 2 π × 1.5 × 10 3.
4.31 Ω
p

s

⎛ 1 ⎞
10 log10 ⎜
⎟ = −0.5, and hence, ε 2 = 10 0.05 − 1 = 0.122.
2
⎝ 1+ ε ⎠
1
A2 − 1
⎛ 1 ⎞
=
= 286.2632.
10 log10 ⎜ ⎟ = −40, and hence, A 2 = 10 4. Therefore,
k1
⎝ A2 ⎠
ε2
ˆ

6.5 13
1 Ω
13
p
Set Ω p = 1. Then Ω s =
=
= . Thus, = s = . The order of the
ˆ
k Ωp
1 .5 3
3

s
prototype lowpass filter is thus given by N =

log10 (1 / k1 )
= 3.8579. As a result, we
log10 (1 / k )

choose the filter order as N = 4.
The order of the prototype lowpass filter obtained using the MATLAB statement
[N,Wn]=buttord(1,13/3,0.5, 40, 's') results in N=4.
The order of the desired highpass filter is also 4.
4.32

Fˆ p1 = 20 × 10 3 , Fˆ p2 = 45 × 10 3 , Fˆs1 = 10 × 10 3 , and Fˆs 2 = 50 × 10 3. Thus,

Not for sale

97

Fˆ p1 Fˆ p2 = 9 × 10 8 and Fˆs1 Fˆs 2 = 7.5 × 10 8. Since Fˆ p1 Fˆ p 2 > Fˆs1 Fˆs 2 , we can either
increase left stopband edge Fˆs1 or decrease the left passband edge Fˆ p1 to make
Fˆ p1 Fˆ p 2 = Fˆs1 Fˆs 2 . We choose to increase Fˆs1 to a new value given by Fˆs1 = 18 × 10 3 , in
which case Fˆ p1 Fˆ p2 = Fˆs1 Fˆs 2 = Fˆo2 = 9 × 10 8. The center angular frequency of the
ˆ = 2 π × 30 × 10 3. The passband width is
bandpass filter is therefore Ω
o

3
ˆ −Ω
ˆ
Bw = Ω
p1
p 2 = 2 π × 25 × 10 .

To determine the bandedges of the prototype lowpass filter we set Ω p = 1 and thus,
ˆ 2 −Ω
ˆ2

30 2 − 18 2
o
s1
Ωs = Ω p
=
= 1.28.
ˆ B
18 × 25

s1 w
Ωp
1
Now, k =
=
= 0.78125. Hence, k ' = 1 − k 2 = 0.62421826 .
Ω s 1.28

⎛ 1
Next, 10 log10 ⎜⎜
⎝ 1 + ε2

⎟⎟ = −0.25 or equivalently, log10 (1 + ε 2 ) = 0.025 which yields

⎛ 1 ⎞
⎟⎟ = −50
ε 2 = 10 0.025 − 1 = 0.05925372 5 or ε = 0.243421. Likewise, 10 log10 ⎜⎜
⎝ A2 ⎠

or, equivalently, log10 ( A 2 ) = 5 which yields A 2 = 10 5 = 100000. Therefore,
ε
1 − k'
k1 =
= 7.69768 × 10 − 4 , ρ 0 =
= 0.058635856. As a result,
2(1 + k ' )
A2 − 1
ρ = ρ 0 + 2(ρ 0 ) 5 + 15(ρ 0 ) 9 + 150(ρ 0 )13 = 0.058637246 . Hence,

2 log10 (4 / k1 )
= 6.0328. We choose N = 7 as the order of the prototype lowpass
log10 (1 / ρ)
filter.
Note that the order can also estimated using the specifications of the bandpass filter. To
this end, the statement to use is
[N,Wn]=ellipord([20 45],[15 50],0.25,50,'s') which also yields N=7 as
the order of the prototype lowpass filter. The order of the desired bandpass filter is
therefore 7 × 2 = 14.
N=

4.33 Fˆ p1 = 10 × 10 6 , Fˆ p2 = 70 × 10 6 ,
Fˆ Fˆ = 70 × 1013 and Fˆ Fˆ

Fˆs1 = 20 × 10 6 , and Fˆs 2 = 45 × 10 6. Thus,
13
ˆ ˆ
ˆ ˆ
p1 p2
s1 s 2 = 90 × 10 . Since F p1 F p 2 < Fs1 Fs 2 , we can either
increase left passband edge Fˆ or decrease the left stopband edge Fˆ to make
p1

s1

Fˆ p1 Fˆ p 2 = Fˆs1 Fˆs 2 . We choose to increase Fˆ p1 to a new value given by

Not for sale

98

Fˆ Fˆ
Fˆ p1 Fˆ p2 = s1 s 2 = 12.8571 × 10 6 , in which case Fˆ p1 Fˆ p2 = Fˆs1 Fˆs 2 = Fo2 = 700 × 1012.
Fˆ p2
ˆ −Ω
ˆ = 2 π × 25 × 10 6 and the center angular
The width of the stopband is Bw = Ω
s2
s1

frequency of the stopband is Ω o2 = 4 π 2 × 700 × 1012.
To determine the bandedges of the prototype lowpass filter we set Ω s = 1 resulting in its
ˆ B

p1 w
= 0.4375.
passband edge Ω p = Ω p
ˆ 2 −Ω
ˆ2

o
p1

⎛ 1
Now, 10 log10 ⎜⎜
⎝ 1 + ε2

⎟⎟ = −0.5 or equivalently, log10 (1 + ε 2 ) = 0.05 which yields

⎛ 1 ⎞
⎟⎟ = −30
ε 2 = 10 0.05 − 1 = 0.1220184543 or ε = 0.349114. Likewise, 10 log10 ⎜⎜
⎝ A2 ⎠

or, equivalently, log10 ( A 2 ) = 3 which yields A 2 = 10 3 = 1000. Therefore,

1
A2 − 1
999
1 Ωs
1
=
=
= 90.4836236.
=
=
= 2.2857 and
k1
ε
0.349114
k Ω p 0.4375
1
1
Substituting the values of and
in Eq. (4.43) we get
k
k1
cosh −1 (90.4836236 )

= 3.5408. We therefore choose N = 4 as the order of the
cosh −1 (2.2857)
prototype lowpass filter. The order of the desired bandstop filter is thus 8.
Using the statement [N,Wn]=cheb1ord(0.4375,1,0.5,30,'s') we get N=4.
Note that the order can also estimated using the specifications of the bandstop filter. To
This end, the statement to use is
[N,Wn]=cheb1ord([10 70],[20 45],0.5,30,'s') which also yields N=4 as
the order of the prototype lowpass filter.
N=

4.34 From Eq. (4.71), the difference in dB in the attenuation levels at Ω p and Ω s is given by
20 N log10 (Ω p / Ω s ). Hence, for Ω o = 2Ω p , the attenuation difference in dB is equal to

20 N log10 2 = 6.0206 N . Likewise, for Ω o = 3Ω p , the attenuation difference in dB is
equal to 20 N log10 3 = 9.5424 N . Finally, for Ω o = 4Ω p , the attenuation difference in
dB is equal to 20 N log10 4 = 12.0412 N .
4.35 The equivalent representation of the D/A converter of Figure 4.48 reduces to the circuit
shown below if j -th bit is ON and the remaining bits are OFF, i.e., a j = 1 and

a k = 0, k ≠ j.

Not for sale

99

Vo,j
_
2N 1

_
2j 1

2j

a j VR

_
2j 2

20

GL

+
Yin

In the above circuit, Yin is the total conductance seen by the load conductance GL which is
N −1

given by Yin = ∑ 2 i = 2 N − 1. The above circuit can be redrawn as indicated below:
i =0

Vo,j
_
Yin _ 2 j 1

_
2j 1

a j VR

GL

+

2 j −1
⋅ a j VR . Using the
Using the voltage-divider relation we then get Vo, j =
Yin + G L
superposition theorem, the general expression for the output voltage Vo is thus given by

N
RL
2 j −1
⋅ a j V R = ∑ 2 j −1 a j ⎜
⎜ 1 + (2 N − 1) R
j =1
j =1 Yin + G L
L

N

Vo = ∑

⎟V .
⎟ R

4.36 The equivalent representation of the D/A converter of Figure 4.49 reduces to the circuit
shown below on the left if N -th bit is ON and the remaining bits are OFF, i.e., a N = 1 and

a k = 0, k ≠ N ,
Vo,N
G
__
2

G
__
2

aNVR

Vo,N
G
__
2

GL

+

aNVR

G
GL + __
2

+

which simplifies to the circuit shown above on the right.
Using the voltage-divider relation we then get
Vo, N =

G
2
G
G
+GL +
2
2

⋅ a N VR =

RL
⋅ a N VR .
2( R + R L )

The equivalent representation of the D/A converter of Figure 4.49 reduces to the circuit
shown below on the left if ( N − 1) -th bit is ON and the remaining bits are OFF, i.e.,
a N −1 = 1 and a k = 0, k ≠ N − 1,
Not for sale

100

G

G

Vo,N _1

Vo,N _1

G
__
2

G
__
2

G
__
2
aN _ 1VR

G
__
2

G
__
2

GL

+

aN _ 1VR

__
GL + G
2

+

which simplifies to the circuit shown above on the right.
Its Thevenin equivalent circuit is indicated below:
Vo,N _1
G
__
2

G
GL + __
2

+
a___
N _1
VR
2
G
2

a
a
RL
⋅ N −1 V R =
⋅ N −1 V R .
G + GL
2
2( RL + R)
2
Following the same procedure we can show that if the l –th bit is ON and the remaining
bits are OFF, i.e., a l = 1, and a k = 0, k ≠ l, then
a
RL
Vo, l =
⋅ l VR .
2( RL + R) 2 N − l
Hence, in general we have
N
a
RL
⋅ l VR .
Vo = ∑
N −l
l =1 2( R L + R ) 2
Vo, N −1 =

4.37 From the input-output relation of the first-order hold, we get the expression for the impulse
δ(nT ) − δ(nT − T )
response as h f (t ) = δ(nT ) +
(t − nT ), nT ≤ t < (n + 1)T . In the range
T
t
δ(0) − δ(−T )
0 ≤ t < T , the impulse response is given by h f (t ) = δ(0) +
t =1+ .
T
T
Likewise, in the range T ≤ t < 2T , the impulse response is given by
δ(T ) − δ(0)
t
h f (t ) = δ(T ) +
(t − T ) = 1 − . Outside these two ranges, h f (t ) = 0. Hence we
T
T
have

Not for sale

101

h f (t)
2

t
, 0 < t < T,
T
t
h f (t ) = 1 _ , T < t < 2T ,
T
0,
otherwise.

{

1+

1

0

T

2T

t

_1

Using the step function we can write
t⎞
t⎞

h f (t ) = ⎜1 + ⎟ [µ(t ) − µ(t − T )] + ⎜1 − ⎟ [µ(t − T ) − µ(t − 2T )]
⎝ T⎠
⎝ T⎠
t
2( t − T )
(t − 2T )
= µ ( t ) + µ( t ) −
µ(t − T ) − 2µ(t − T ) − µ(t − 2T ) +
µ(t − 2T ) + 2µ(t − 2T ).
T
T
T
Taking the Laplace transform of the above equation we arrive at the transfer function
1
1
2 e − sT
e − sT e − 2 sT 1 e − 2 sT
e − 2 sT ⎛ 1 + sT
H f (s) = +
− ⋅
−2

+ ⋅
+2
=⎜
s Ts 2 T s 2
s
s
T
s
⎝ T
s2
Hence, the frequency response is given by
− jΩT
⎛ 1 + jΩT ⎞⎛⎜ 1 − e
H f ( jΩ ) = ⎜

jΩ
⎝ T
⎠⎜⎝

− sT

⎞ ⎛⎜ 1 − e

s
⎠ ⎜⎝

2

2

−1
⎟ = T 1 + Ω 2 T 2 ⎛⎜ 2 sin(ΩT / 2) ⎞⎟ e − jΩT e j tan ΩT . A

⎝ ΩT / 2 ⎠

plot of the magnitude responses of the zero-order hold and the first-order hold is shown
below:

Amplitude

1.5

← first-order hold

1

← zero-order hold

0.5

0

0

0.5

1

1.5

2

2.5

3

4.37

From the input-output relation of thelinear interpolator, we get the expression for the
δ(nT ) − δ(nT − T )
impulse response as h f (t ) = δ(nT − T ) +
(t − nT ), nT ≤ t < (n + 1)T . In the
T
δ(0) − δ(−T )
range 0 ≤ t < T , the impulse response is given by h f (t ) = δ(−T ) +
t. Likewise,
T

Not for sale

102

2

⎟ .

in the range T ≤ t < 2T , the impulse response is given by h f (t ) = δ(0) +

δ(T ) − δ(0)
(t − T ).
T

⎧ t
0 ≤ t < T,
⎪ T,
⎪⎪
t
Outside these two ranges, h f (t ) = 0. Hence we have h f (t ) = ⎨2 − , T ≤ t < 2T ,
⎪ T
otherwise.
⎪ 0,
⎪⎩
h f (t)
1

0

t

2T

T

Using the step function we can write
t⎞
t

h f (t ) = [µ(t ) − µ(t − T )] + ⎜ 2 − ⎟ [µ(t − T ) − µ(t − 2T )]
T⎠
T

t
2( t − T )
(t − 2T )
= µ(t ) −
µ(t − T ) +
µ(t − 2T ).
T
T
T
Taking the Laplace transform of the above equation we arrive at the transfer function
⎛ 1 − e − sT

+
= T⎜
H f (s) =
⎜ sT
s 2T
s 2T
s 2T

1

2 e − sT

e − 2 sT

2

⎟ . Hence, the frequency response is given

2

2
⎛ 1 − e − jΩT ⎞
⎛ sin(ΩT / 2) ⎞ − jΩT

by H f ( jΩ) = T
. A plot of the magnitude
= T⎜
⎟ e
⎜ j ΩT ⎟
⎝ ΩT / 2 ⎠

responses of the ideal filter, zero-order hold and the first-order hold is shown below:

Amplitude

1
0.8

← zero-order hold

0.6

← linear interpolator

0.4
0.2
0

0

0.5

1

1.5

2

2.5

3

M4.1 We use N = 4 and Wn = 18365.512865 computed in Problem 4.22 and use omega
= 0:2*pi:2*pi*10000; to evaluate the frequency points. The gain plot obtained using
Program 4_2 is shown below.

Not for sale

103

Analog Lowpass Filter

Passband Details
0.5

-10

Gain, dB

Gain, dB

0

-20

0

-0.5

-30
-40

0

2000

4000
6000
Frequency, Hz

8000

-1

10000

0

500

1000
1500
Frequency, Hz

2000

2500

M4.2 We use N = 3 computed in Problem 4.23 and Fp = 2*pi*1500 and Rp = 0.25
and use omega = 0:2*pi:2*pi*10000; to evaluate the frequency points. The gain plot
obtained using Program 4_3 is shown below.
Analog Lowpass Filter

Passband Details
0.5

0

Gain, dB

Gain, dB

-10
-20
-30

0

-0.5

-40
-50

0

2000

4000
6000
Frequency, Hz

8000

-1

0

500

1000
1500
Frequency, Hz

2000

M4.3 We replace the statement
Fp = input('Passband edge frequency in Hz = '); with
Fs = input('Stopband edge frequency in Hz = '); replace
Rp = input('Passband ripple in dB = ');
with Rs = input('Minimum stopband attenuation in dB = '); and
replace [num,den] = cheby1(N,Rp,Fp,'s'); with [num,den] =
cheby2(N,Rs,Fs,'s'); to modify Program 4_3.
Next, we run the modified program using N = 3 and Rs = 25, and Fs = 2*pi*6000. The
gain response plot generated by the modified program is shown below.

Not for sale

104

Passband Details

Analog Lowpass Filter

0.5

0

Gain, dB

Gain, dB

-10
-20
-30

0

-0.5

-40
-50

0

2000

4000
6000
Frequency, Hz

-1

8000

0

500

1000
1500
Frequency, Hz

2000

2500

The numerator and the denominator coefficients of the 3rd order Type 2 Chebyshev lowpass
filter can be obtained by typing num and den in the command window:
H LP (s ) =

10138.1864 s 2 + 4.8663294 × 1010
s 3 + 7030.25552 5 s 2 + 2.41983322 54 × 10 7 s + 4.8663294 × 1010

.

M4.4 We use N = 3 and Wn = 9424.777960769379 computed in Problem 4.26 in
Program 4_4 and use omega = [0: 200: 12000*pi]; to evaluate the frequency points. The
gain plot generated by running this program is shown below:
Analog Lowpass Filter

Passband Details
0.5

0

-20

Gain, dB

Gain, dB

-10

-30
-40

0

-0.5

-50
-60

0

1000

2000 3000 4000
Frequency, Hz

5000

6000

-1

0

500

1000
1500
Frequency, Hz

2000

M4.5 The MATLAB program used is as given below:
[N,Wn]=buttord(1,13/3,0.5, 40, 's');
[B,A] = butter(N,Wn,'s');
[num,den]=lp2hp(B,A,2*pi*6500);
figure(1)
[h,w]=freqs(B,A);gain = 20*log10(abs(h));
plot(w,gain);grid
xlabel('\Omega');ylabel('Gain, dB');
title('Analog Lowpass Filter');
figure(2)
[h,w]=freqs(num,den);gain = 20*log10(abs(h));
plot(w/(2*pi),gain);grid
xlabel('Frequency, Hz');ylabel('Gain, dB');
title('Analog Highpass Filter');
Not for sale

105

H LP (s ) =
H HP (s ) =

3.5262257
4

3

s + 3.58086432 s + 6.41129464 s 2 + 6.72423556 s + 3.5262257

,

s4
s 4 + 7.7880 s 3 + 3.0326485 × 10 9 s 2 + 6.91763168 × 1013 s + 7.8897418 × 1017

Prototype Lowpass Filter

Passband Details
0.5

0

Gian, dB

Gian, dB

-10
-20
-30

0

-0.5

-40
-50

0

1

2

3

4

-1

5

0

0.5

1

1.5

Analog Highpass Filter

Passband Details
0.5

0

Gain, dB

Gain, dB

-10
-20
-30

0

-0.5

-40
-50

0

0.5

1
Frequency, Hz

1.5

2
4

x 10

-1
0.5

1
1.5
Frequency, Hz

2
4

x 10

M4.6 The MATLAB program used is given below:
[N,Wn] = ellipord(1,1.28,0.25,50,'s');
[B,A] = ellip(N,0.25,50,Wn,'s');
[num,den] = lp2bp(B,A,2*pi*30e3,2*pi*25e3);
figure(1)
omega = [0:0.01:10];
h = freqs(B,A,omega);
gain = 20*log10(abs(h));
plot(omega,gain); grid; axis([0 5 -80 5]);
xlabel('\Omega'); ylabel('Gain, dB');
title('Analog Lowpass Filter');
figure(2)
omega = [0:200:100e3*2*pi];
h = freqs(num,den,omega);
gain = 20*log10(abs(h));
Not for sale

106

.

plot(omega/(2*pi),gain); grid; axis([0 60e3 -80 5]);
xlabel('Frequency in Hz'); ylabel('Gain, dB');
title('Analog Bandpass Filter');
Analog Bandpass Filter

0

0

-20

-20
Gain, dB

Gain, dB

Analog Lowpass Filter

-40

-60

-60
-80

-40

-80

0

1

2

3

4

5

0

1

2
3
4
Frequency in Hz

H LP (s ) =

6

4

5

6
4

x 10

2

0.0185 s + 0.1364 s + 0.2887 s + 0.186835
7

6

s + 1.36426 s + 2.9795 s 5 + 2.7545 s 4 + 2.70025 s 3 + 1.56215 s 2 + 0.7275 s + 0.1868

The numerator and denominator coefficients of can be obtained by typing num and den in
the Command Window.
M4.7 The MATLAB program used is given below:
[N,Wn] = cheb1ord(0.3157894, 1, 0.5, 30,'s');
[B,A] = cheby1(N,0.5, Wn,'s');
[num,den] = lp2bs(B,A,2*pi*sqrt(700)*10^6, 2*pi*15e6);
figure(1)
omega = [0:0.01:10];
h = freqs(B,A,omega);
gain = 20*log10(abs(h));
plot(omega, gain); grid; axis([0 4 -70 5]);
xlabel('\Omega'); ylabel('Gain, dB');
title('Analog Lowpass Filter');
figure(2)
omega = [0:10000:160e6*pi];
h = freqs(num,den,omega);
gain = 20*log10(abs(h));
plot(omega/(2*pi), gain); grid; axis([0 80e6 -70 5]);
xlabel('Frequency in Hz'); ylabel('Gain, dB');
title('Analog Bandstop Filter');

H LP (s ) =

0.02253823
s 3 + 0.3956566 s 2 + 0.1530643 s + 0.02253823

Not for sale

.

107

.

Analog Lowpass Filter

Passband Details
0.5

-20

Gain, dB

Gain, dB

0

-40

0

-0.5

-60
0

1

2

3

-1

4

0

0.1

0.2

0.3

0.4

0.5

Stopband Details

Analog Bandstop Filter

0.5

-20

Gain, dB

Gain, dB

0

-40

0

-0.5

-60

-1
0

2

4
Frequency in Hz

6

8
7

x 10

0

2

4
Frequency in Hz

6

8
7

x 10

M4.8 The MATLAB program to generate the plots of Figure 4.56 is given below:
% Droop Compensation
w = 0:pi/100:pi;
h1 = freqz([-1/16 9/8 -1/16],1,w);
h2 = freqz(9, [8 1], w);
w1 = 0;
for n = 1:101;
h3(n) = sin(w1/2)/(w1/2);
w1 = w1 + pi/100;
end
m1 = 20*log10(abs(h1));
m2 = 20*log10(abs(h2));
m3 = 20*log10(abs(h3));
plot(w/pi,m3,’-’,w/pi,m1+m2,’--’,w/pi,m2+m3,’-.’);grid
xlabel(‘Normalized frequency’);ylabel(Gain, dB’);

Not for sale

108

Chapter 5
5.1

N −1
N −1
~
~
~
Let ~
y[ n] = ∑ ~
x[r ]h [ n − r ]. Then ~
y[ n + kN ] = ∑ ~
x[r ]h [ n + kN − r ]. Since h [ n] is
r =0

r =0

~
~
periodic in n with a period N , h [ n + kN − r ] = h [ n − r ]. Therefore ~
y[ n + kN ]
N −1
~
= ∑ ~
x[r ]h [ n − r ] = ~
y[ n], hence ~
y[ n] is also periodic in n with a period N .
r =0

5.2

4
~
~
~
~
~
~
(a) ~
y[ 0] = ∑ ~
x[r ]h [ −r ] = ~
x[0]h [0] + ~
x[1]h [ 4] + ~
x[2]h [3] + ~
x[3]h [2] + ~
x[ 4]h [1] = −13,
r =0

4

~
~
~
~
~
~
~
y[1] = ∑ ~
x[r ]h [1 − r ] = ~
x[0]h [1] + ~
x[1]h [0] + ~
x[2]h [ 4] + ~
x[3]h [3] + ~
x[ 4]h [2] = −13,
r =0
4

~
~
~
~
~
~
~
y[ 2] = ∑ ~
x[r ]h [2 − r ] = ~
x[0]h [2] + ~
x[1]h [1] + ~
x[2]h [0] + ~
x[3]h [ 4] + ~
x[ 4]h [3] = −13,
r =0
4

~
~
~
~
~
~
~
y[3] = ∑ ~
x[r ]h [3 − r ] = ~
x[0]h [3] + ~
x[1]h [2] + ~
x[2]h [1] + ~
x[3]h [0] + ~
x[ 4]h [ 4] = −13,
r =0
4

~
~
~
~
~
~
~
y[ 4] = ∑ ~
x[r ]h [ 4 − r ] = ~
x[0]h [ 4] + ~
x[1]h [3] + ~
x[2]h [2] + ~
x[3]h [1] + ~
x[ 4]h [0] = −13.
r =0

y[ n] = {− 13, − 13, − 13, − 13, − 13,}, 0 ≤ n ≤ 4.
Therefore, ~
y[ n] = {1, 1, 1, 1, 1,}, 0 ≤ n ≤ 4.
(b) ~
5.3

~ [ n + rN ] = ψ
~ [ n], hence all the terms which are not in the range 0,1,K, N − 1,
Since ψ
k
k
~ [ n], where 0 ≤ k ≤ N − 1. Hence, in this case the Fourier
can be accumulated to ψ
k
series representation involves only N complex exponential sequences. Let
1 N −1 ~
j 2 πkn / N
~
x [ n] =
, then
∑ X[ k ] e
N k =0
N −1

1 N −1 ~ ⎛ N −1 j 2 π( k − r )n / N ⎞
1 N − 1 N −1 ~
j 2 π( k − r )n / N
⎟⎟.
=
x[ n] e − j 2 πrn / N =
∑ ~
∑ ∑ X[ k ] e
∑ X[ k ] ⎜⎜ ∑ e
N n=0 k =0
N k =0
n =0
⎝ n=0

Now, from Eq. (5.11), the inner summation is equal to N if k = r , otherwise it is equal
N −1
~
~
x[ n] e − j 2 πrn / N = X [r ]. Next, we observe X[ k + lN ]
to 0. Thus, ∑ ~
n=0

N −1

N −1
N −1
~
= ∑ ~
x[ n] e − j 2 π( k + lN )n / N = ∑ ~
x[ n] e − j 2 πkn / N e − j 2 πln = ∑ ~
x[ n] e − j 2 πkn / N = X[ k ].
n=0

n=0

Not for sale.

n=0

109

5.4

{

}

1
πn
~
x1[n] is N = 8.
x1[ n] = cos⎛⎜ ⎞⎟ = e jπn / 4 + e − jπn / 4 . The period of ~
⎝ 4⎠ 2
7

1⎧ 7
~
X1[ k ] = ⎨ ∑ e j 2 πn / 8 e − j 2 πkn / 8 + ∑ e − j 2 πn / 8 e − j 2 πkn / 8 ⎬
2 ⎩n = 0
n =0

(a)

=

7

1 ⎧ 7 − j 2 πn( k −1) / 8
+
e
e − j 2 πn( k +1) / 8 ⎬. Now, from Eqn. (5.11) we observe

2 ⎩n = 0
n =0

7

∑e

− j 2 πn( k −1) / 8

n=0

7
8, for k = 1,
8, for k = 7,
and ∑ e − j 2 πn( k +1) / 8 = ⎧⎨
= ⎧⎨
0
,
otherwise
,
⎩0, otherwise.

n=0

~
4, for k = 1,7,
Hence, X1[ k ] = ⎧⎨
0
⎩ , otherwise.
1
3
πn
πn
(b) ~
x 2 [ n] = sin⎛⎜ ⎞⎟ + 3 cos⎛⎜ ⎞⎟ = {e jπn / 3 − e − jπn / 3} + {e jπn / 4 + e − jπn / 4 }. The
2
2
j
⎝ 3⎠
⎝ 4⎠
πn
πn
x [ n] is the
is 6 and the period of cos
is 8. Hence, the period of ~
period of sin

( )

( )

3

4

2

GCM of (6,8) and is 24.

1 ⎧ 23 j8πn / 24 − j 2 πkn / 24 23 − j8πn / 24 − j 2 πkn / 24 ⎫
~
− ∑e
X 2 [k ] =
e
e
⎨ ∑e

2 j ⎩n = 0
n =0

23

3 ⎧ 23
+ ⎨ ∑ e j 6 πn / 24 e − j 2 πkn / 24 − ∑ e − j 6 πn / 24 e − j 2 πkn / 24 ⎬
2 ⎩n = 0
n=0

=

1 ⎧ 23 − j 2 πn( k − 3) / 24 23 − j 2 πn( k + 3) / 24 ⎫
− ∑e
⎨ ∑e

2 j ⎩n = 0
n=0

k = 3,
⎧− j12,
⎪ j12,
3 ⎧ 23 − j 2 πn( k − 4) / 24 23 − j 2 πn( k + 4) / 24 ⎫
~
k = 21,
+ ⎨ ∑e
− ∑e
⎬. Hence X 2 [ k ] = ⎨ 36,
= 4, 20,
k
2 ⎩n = 0
n =0

0
,
otherwise.

5.5

~
Let P[ k ] denote the coefficients of the Fourier series representation of ~
p[ n]. Since
~
p[ n] is periodic with a period N , then from Eq. (5.185b), we have
N −1
~
P[ k ] = ∑ ~
p[ n] e − j 2 πkn / N = 1. Hence, from Eq. (5.185a) we get

~
p[ n] =

5.6

n=0
1 N −1 ~

j 2 πln / N
=
∑ P[l] e

N l =0

~
X [ k ] = X ( e jω )

1 N −1 j 2 πln / N
.
∑ e
N l=0

ω = 2 πk / N

= X (e j 2 πk / N ) = ∑ x[ n] e − j 2 πk / N , − ∞ < k < ∞.
n = −∞

~
~
Now, X[ k + lN ] = X (e j 2 π( k + lN ) / N ) = X (e j 2 πk / N e jπl ) = X (e j 2 πk / N ) = X[ k ].

Not for sale.

110

1 N −1 ~
1 N −1⎛ ∞
j 2 πkn / N
− j 2 πln / N ⎞⎟ j 2 πkn / N
x [ n] =
=
Likewise, ~
∑ ⎜⎜ ∑ x[l] e
∑ X[ k ] e
⎟e
N k =0
N k = 0 ⎝ l = −∞

1 N −1 ∞
j 2 π( k − l )n / N
=
. Let l = n + rN. Then
∑ ∑ x[l] e
N k =0 l = −∞
N −1
⎛ N −1 − j 2 πkr ⎞
1 ∞
~
⎟⎟. But ∑ e − j 2 πkr = N . Hence,
x [ n] =
∑ x[ n + rN ]⎜⎜ ∑ e
N r = −∞
k =0

⎝ k =0

~
x[ n] = ∑ x[ n + rN ].
r = −∞

5.7

N −1
N −1
~
(a) G[ k ] = ∑ g~[ n] e − j 2 πkn / N = ∑ ~
x [ n ]~
y[ n] e − j 2 πkn / N . Now,
n=0
N −1 ~

n=0

1
− j 2 πrn / N
~
x [ n] =
. Therefore,
∑ X [r ] e
N r =0
1 N −1 N −1 ~ t
1 N −1 ~ N −1t
~
− j 2 π( k − r )n / N
− j 2 π( k − r ) n / N
=
G[ k ] =
∑ ∑ X[r ] y[ n] e
∑ X[r ] ∑ y[ n] e
N n=0 r =0
N r =0
n=0
1 N −1 ~ ~
=
∑ X[r ]Y [ k − r ].
N r =0
~
1 N −1 ~ ~
1 N −1 N −1 ~ ~
j 2 πkn / N
j 2 πk ( n − r ) / N
=
(b) h [ n] =
∑ X[ k ]Y [ k ] e
∑ ∑ x[r ]Y [ k ] e
N k =0
N k =0 r =0
N −1
N −1
⎛ 1 N −1 ~
j 2 πk ( n − r ) / N ⎞⎟
x[r ] ⎜⎜
x [r ] ~
y[ n − r ] .
=
= ∑ ~
∑~
∑ Y [k] e

N
r =0
k
r
=
0
=
0

5.8

(a) x a [ n] = sin(2πn / N ) = (e j 2 πn / N − e − j 2 πn / N ). Therefore,
2j
1

X a [k ] =

1 N −1 j 2 πn / N − j 2 πkn / N 1 N −1 − j 2 πn / N − j 2 πkn / N

e
e
∑ e
∑ e
2 j n=0
2 j n=0

1 N −1 − j 2 π( k −1)n / N 1 N −1 − j 2 π( k +1)n / N

. From Eq. (5.11), the first sum is
∑ e
∑ e
2 j n=0
2 j n=0
equal to N when k = 1 and 0 othereise. Likewise, from Eq. (5.11), the second sum is
equal to N when k = N − 1 and 0 otherwise. Therefore,
k = 1,
⎧ N / 2 j,

X a [ k ] = ⎨− N / 2 j, k = N − 1,
⎪⎩ 0,
otherwise.
=

N
2 πn ⎞ 1 1
4 πn ⎞
1
for
(b) x b [ n] = cos 2 ⎛⎜
⎟ = + cos⎛⎜
⎟. Now the N –point DFT of is
2
2
⎝ N ⎠ 2 2
⎝ N ⎠

( ) is

k = 0 and 0 otherwise. From Example 5.2, the N –point DFT of cos
k = 2 and k = N − 2 and 0 otherwise. Therefore,

Not for sale.

4 πn
N

N
2

for

111

k = 0,
⎧⎪ N / 2,
X b [ k ] = ⎨ N / 4, k = 2, N − 2,
⎪⎩ 0,
otherwise.
2 πn ⎞ 1
6 πn ⎞
(c) x c [ n] = cos 3 ⎛⎜
⎟ = cos⎛⎜
⎟+
⎝ N ⎠ 4
⎝ N ⎠
N
6 πn ⎞
DFT of cos⎛⎜
for k = 3 and
⎟ is
2
⎝ N ⎠

3
2 πn ⎞
cos⎛⎜
⎟.
4
⎝ N ⎠

From Example 5.2, the N –point

k = N − 3 and 0 otherwise. Likewise, from

N
2 πn ⎞
Example 5.2, the N –point DFT of cos⎛⎜
for k = 1 and k = N − 1 and 0
⎟ is
2
⎝ N ⎠
⎧ N / 8, k = 3, N − 3,

otherwise. Therefore, X c [ k ] = ⎨3N / 8, k = 1, N − 1,
⎪⎩ 0,
otherwise.

5.9

N −1

N −1

1 − αWNkN

n=0

n=0

1 − αWNk

(a) Ya [ k ] = ∑ α nWNkn = ∑ (αWNk ) n =

=

1− α
1 − αWNk

.

(b) Yb [ k ] = 2 ∑ WNkn − 3 ∑ WNkn . Assume first N is even, i.e., N = 2L. Then
k even

k odd

(

)

⎛ 1 − W kL
L −1
L −1
L −1
L −1
L
Yb [ k ] = 2 ∑ W2kL2r − 3 ∑ W2kL(2r +1) = 2 ∑ WLkr − 3W2kL ∑ WLkr = 2 − 3WNk ⎜
k

r =0
r =0
r =0
r =0
⎝ 1 − WL
L

L −1

r =0

r =0

⎟ = 0.

Next, assume N is odd, i.e., N = 2 L + 1. Then Yb [ k ] = 2 ∑ W2kL2r − 3 ∑ W2kL(2r +1)
=

L

2 ∑ WLkr
r =0

− 3W2kL

L −1

r =0

kL
⎛ 1 − W k ( L +1) ⎞

k ⎜ 1 − WL
L

=2
− 3W2 L
⎜ 1−W k ⎟
⎜ 1−W k
L
L

WLkr

k

⎟ = 2⎜ 1 − W L
⎜1− W k

L

⎟ = 2.

5.10 x[ n] = cos(ωo n) = (e jω o n + e − jω o n ), 0 ≤ n ≤ N − 1. Therefore,
1
2

X[ k ] =

1 N −1 jω o n − j 2 πkn / N 1 N −1
+ ∑ e − j jω o n e − j 2 πkn / N
e
∑ e
2 n=0
2 n=0

1 N −1 − j (
= ∑ e
2 n=0
=

2 πk
− ω o )n
N

2 πk
− j ⎛⎜
−ωo ⎞⎟ N
N

1 1− e

2

2 πk
− j ⎛⎜
−ωo ⎞⎟
N

1− e

1 N −1 − j (
+ ∑ e
2 n=0
+

2 πk
+ ω o )n
N

2 πk
− j ⎛⎜
+ωo ⎞⎟ N
N

1 1− e

2

2 πk
− j ⎛⎜
+ωo ⎞⎟
N

1− e

Not for sale.

112

⎛ 2 πk −ω ⎞⎛ N −1 ⎞

o ⎟⎜
N
⎠⎝ 2 ⎠

1 − j⎜
= e ⎝
2

(

sin (

N −1

5.11 X [ k ] = ∑ x[ n]WNnk =
n=0
( N / 2 ) −1

)+ 1 e
) 2

ωo N
2
πk
ωo

2
N

sin πk −

( N / 2 ) −1

2 πk
N −1 ⎞
+ ωo ⎞⎟⎛⎜
− j ⎛⎜

⎝ N
⎠⎝ 2 ⎠

sin

x[2r ]WN2rk +

( N / 2 ) −1

r =0
r =0
( N / 2 ) −1
= ∑ x[2r ]WNrk/ 2 + WNk ∑ x[2r + 1]WNrk/ 2
r =0
r =0
( N / 2 ) −1
( N / 2) −1
= ∑ x0 [r ]WNrk/ 2 + WNk ∑ x1[r ]WNrk/ 2
r =0
r =0
k
= X 0 [〈 k 〉 N / 2 ] + WN X1[ 〈 k 〉 N / 2 ], 0 ≤ k ≤ N − 1.
N −1

5.12 X [ k ] = ∑ x[ n]WNnk =
=

n=0
( N / 2 ) −1

( N / 2 ) −1

n=0

x[ n]WNnk + WN( N / 2 )k

x[ n]WNnk +

( N / 2 ) −1

n=0
( N / 2 ) −1 ⎛

n=0

x[

N
2

(

ωo N
2
πk ωo
+
N
2

sin πk +

(

)

).

x[2r + 1]WN(2r +1)k

N −1

nk
∑ x[ n]WN

n= N / 2

+ n]WNnk

⎞ nk
k N
⎜ x[ n] + (−1) x[ + n] ⎟WN . For k = 2l, we get
2

n=0 ⎝
( N / 2 ) −1 ⎛
( N / 2 ) −1 ⎛
N
N

X [2l] = ∑ ⎜ x[ n] + x[ + n] ⎟WN2 nl = ∑ ⎜ x[ n] + x[ + n] ⎟WNnl/ 2 = X 0 [l]
2
2

n=0 ⎝
n=0 ⎝
( N / 2 ) −1 ⎛
N

and for k = 2l + 1 we get X [2l + 1] = ∑ ⎜ x[ n] − x[ + n] ⎟WN(2 l +1)n
2

n=0 ⎝
=

=

( N / 2 ) −1 ⎛

N
N
⎞ n
nl
⎜ x[ n] + x[ + n] ⎟WN ⋅ WN / 2 = X1[l] where 0 ≤ l ≤ − 1.
2
2

n=0 ⎝

1
2

1
2

5.13 g[ n] = ( x[2n] + x[2 n + 1]), h[ n] = ( x[2n] − x[2n + 1]), 0 ≤ n ≤

N
− 1.
2

Solving for

x[2n] and x[2n + 1], we get x[2 n] = g[ n] + h[ n] and x[2 n + 1] = g[ n] − h[ n]. Therefore,
N −1

X [ k ] = ∑ x[ n]WNnk =
n=0
( N / 2 ) −1

( N / 2 ) −1

x[2r ]WN2rk +

( N / 2 ) −1

x[2r + 1]WN(2r +1)k

r =0
r =0
( N / 2 ) −1
= ∑ x[2r ]WNrk/ 2 + WNk ∑ x[2r + 1]WNrk/ 2
r =0
r =0
( N / 2 ) −1
( N / 2 ) −1
= ∑ ( g[ n] + h[ n]) WNrk/ 2 + WNk ∑ ( g[ n] − h[ n]) WNrk/ 2
r =0
r =0
( N / 2 ) −1
( N / 2 ) −1
= (1 + WNk ) ∑ g[ n]WNrk/ 2 + (1 − WNk ) ∑ h[ n]WNrk/ 2
r =0
r =0

Not for sale.

113

= (1 + WNk )G[ 〈 k 〉 N / 2 ] + (1 − WNk )H [ 〈 k 〉 N / 2 ], 0 ≤ k ≤ N − 1.

5.14 g[ n] = a1 x[2n] + a 2 x[2n + 1], h[ n] = a3 x[2 n] − a 4 x[2 n + 1], 0 ≤ n ≤

N
− 1,
2

with

a g[ n] − a 2 h[ n]
a1a 4 ≠ a 2 a3 . Solving for x[2n] and x[2n + 1], we get x[2n] = 4
and
a1a 4 − a 2 a3
− a3 g[ n] + a1h[ n]
x[2n + 1] =
. Therefore,
a1a 4 − a 2 a3
N −1

X [ k ] = ∑ x[ n]WNnk =
=

n=0
( N / 2 ) −1

( N / 2 ) −1

x[2r ]WN2rk +

r =0
( N / 2 ) −1
x[2r ]WNrk/ 2 + WNk ∑ x[2r
r =0

r =0
( N / 2 ) −1 ⎛ a

( N / 2 ) −1

r =0

x[2r + 1]WN(2r +1)k

+ 1]WNrk/ 2

( N / 2 ) −1 ⎛ − a g[ n] + a h[ n] ⎞
g[r ] − a 2 h[r ] ⎞ rk
1
⎜⎜ 4
⎟⎟ WN / 2 + WNk ∑ ⎜⎜ 3
⎟⎟ WNrk/ 2
r = 0 ⎝ a1a 4 − a 2 a3 ⎠
r = 0 ⎝ a1a 4 − a 2 a3 ⎠
1
=
(a 4 − a3WNk )G[〈 k 〉 N / 2 ] + (− a 2 + a1WNk )H[〈 k 〉 N / 2 ] , 0 ≤ n ≤ N − 1.
a1a 4 − a 2 a3

=

(

)

N −1

N −1

n=0

n=0

2 ln
5.15 (a) G[ k ] = ∑ x[ n]W2nk
N . For k even, i.e., k = 2l, G[ 2l] = ∑ x[ n]W2 N
N −1

= ∑ x[ n]2lnN = X[l], 0 ≤ l ≤ N − 1.
n=0

2 N −1

(b) H[ k ] = ∑ x[ n − N ]W2nk
N . Let m = n − N or n = m + N . Then
n=N

N −1

N −1

m + N )k
= (−1) k ∑ x[ m]W2mk
H[ k ] = ∑ x[ m]W2(N
N . For k even, i.e., k = 2l,
m=0
N −1

N −1

m=0

n=0

n=0

2 N −1

2 N −1

n=0

n=0

H[2l] = ∑ x[ n]W22Nln = ∑ x[ n]WNln = X[l], 0 ≤ l ≤ N − 1.

nk
5.16 Y [ k ] = ∑ y[ n]W2nk
N = ∑ ( g[ n] + h[ n]) W2 N = G[ k ] + H[ k ]. For k even, i.e.,

k = 2l, Y [ 2l] = G[ 2l] + H [ 2l] = 2 X [ l], 0 ≤ l ≤ N − 1.
N −1

N −1
n=0

N −1

n=0
N −1

n=0

n=0

2l +1)n
= ∑ x[ n]W2nN WNln and
For k odd, i.e., k = 2l + 1, G[2l + 1] = ∑ x[ n]W2(N
2l +1)n
H[2l + 1] = − ∑ x[ n]W2(N
= − ∑ x[ n]W2nN WNln = −G[2l + 1], 0 ≤ l ≤ N − 1.

Hence, for k = 2l + 1, Y [ 2l + 1] = G[ 2l + 1] − G[ 2l + 1] = 0, 0 ≤ l ≤ N − 1.

Not for sale.

114

5.17 Y [ k ] =

MN −1

N −1

nk
nk
∑ y[ n]WMN = ∑ x[ n]WMN . Thus,

n=0
N −1

n=0
N −1

n=0

n=0

nkM
Y [ kM ] = ∑ x[ n]WMN
= ∑ x[ n]WNnk = X[ k ].

5.18
N −1

N −1

5.19 (a) Now, X[ N / 2] = ∑ x[ n]WNnN / 2 = ∑ (−1) n x[ n]. Hence if x[ n] = x[ N − 1 − n]
n=0

n=0

N −1

and N is even, then X[ N / 2] = ∑ (−1) n x[ n] = 0.
n=0

N −1

(b) X[0] = ∑ x[ n]. Hence if x[ n] = − x[ N − 1 − n], then X [0] = 0.
n=0

N −1

( N / 2 ) −1

n =0

n=0

(c) X[2l] = ∑ x[ n]WN2 nl =
=

( N / 2 ) −1

n=0

x[ n]WN2 nl +

( N / 2 ) −1

n=0

x[ n +

x[ n]WN2 nl +

N −1

2 nl
∑ x[ n]WN

n = ( N / 2 ) −1

M −1

nl
= ∑ (x[ n] + x[ n + M ])W22M
. Hence if

N
]WN2 nl
2

n=0

x[ n] = − x[ n + M ], then X [2l] = 0 for 0 ≤ l ≤ M − 1.
N −1

5.20 X [2m] = ∑ x[n]W N2mn =
=

n=0
( N / 2 ) −1

n=0

0≤m≤

( N / 2 ) −1

n=0
N / 2 −1

x[n]W N2mn +

N −1

2 mn
∑ x[n]W N

n= N / 2
( N / 2 ) −1
N
x[n]W N2 mn + ∑ x[n + ]W N2mnW NmN = ∑ ⎛⎜ x[n] +
2
n=0 ⎝
n=0
N
N
− 1. This implies x[n] + x[ n + ] = 0.
2
2

x[n +

N
2

2 mn
⎟⎞ W N = 0,

5.21 (a) Using the circular convolution property of the DFT given in Table 5.3, we get
km1

DFT{x[〈 n − m1 〉 N } = W N

km2

X [k ] and DFT{x[〈 n − m2 〉 N } = W N
km1

Hence, W [k ] = DFT {w[n]} = W N
(b) g[n] =

1
2

km2

X [k ] + W N

km1

X [k ] = (W N

(x[n] + (−1) n x[n]) = 12 (x[n] + WN− ( N / 2)n x[n]).

X [k ].
km1

+ WN

) X [k ].

Using the circular

convolution property of the DFT given in Table 5.3, we get
G[k ] = DFT{g[n]} =

1
2

{X [k ] + X [〈k −

N

2 N

}.

(c) Using the circular convolution property of the DFT given in Table 5.3, we get
Y [k ] = DFT { y[n]} = X [k ] ⋅ X [k ] = X 2 [k ].

Not for sale.

115

{

}= W X [k ] = (−1) X [k ]. Hence,
⎧2 X [k ],
U [k ] = DFT{u[n]} = DFT{x[n] − x[〈 n − 〉 }= X [k ] − (−1) X [k ] = ⎨
⎩ 0,

5.22 (a) DFT x[〈 n −

N

2 N

k ( N / 2)
N

k

N
2 N

k

for k odd,
for k even.

N ⎫

(b) V [k ] = DFT{v[n]} = DFT ⎨ x[n] − x[n − ]⎬ = X [k ] + X [k ] = 2 X [k ].
2 ⎭

( N / 2) n
x[n]. Hence, using the circular frequency-shifting
(c) y[n] = (−1) n x[n] = W N

property of the DFT given in Table 5.3, we get

{

}

Y [k ] = DFT{ y[n]} = DFT W N( N / 2) n x[n] = X [〈 k −

N
〉 ].
2 N

5.23 (a) From the circular frequency-shifting property of the DFT given in Table 5.3, we get
− m1 n

IDFT{ X [〈 k − m1 〉 N } = W N

−m 2 n

x[n] and IDFT{ X [〈 k − m2 〉 N } = W N

x[n]. Hence,

w[n] = IDFT{W [k ]} = IDFT{αX [〈 k − m1 〉 N } + βX [〈 k − m2 〉 N }
− m1n

= αW N

− m2 n

x[n] + βW N

− m1n

x[n] = (αW N

(b) G[k ] = ( X [k ] + (−1) k X [k ]) =
1
2

1
2

− m2 n

+ βW N

(X [k ] + W

) x[n].

)

− ( N / 2) k
X [k ] .
N

Using the circular

time-shifting property of the DFT given in Table 5.3, we get
N
1
g[n] = IDFT{G[k ]} = ⎛⎜ x[n] + x[〈 n − 〉 N ⎞⎟.
2⎝
2

(c) Using the modulation property of the DFT given in Table 5.3, we get
y[n] = IDFT{Y [k ]} = N ⋅ x[n] ⋅ x[n] = N ⋅ x 2 [n].
N −1

( N / 2 ) −1

n=0

n=0

5.24 (a) X [2m] = ∑ x[n]W N2mn =

=
=
=

( N / 2 ) −1

x[n]W N2mn

x[n]W N2 mn +

n=0
( N / 2 ) −1
n=0
( N / 2 ) −1

+

x[n]W N2mn +

x[n +

x[n +

N
]W N2mnW Nmn
2

n=0
N / 2 −1
n=0

2 mn
∑ x[n]W N

n= N / 2

N

2m( n + )
N
2
]W N
2

N / 2 −1

N −1

( N / 2 ) −1
N
2 mn
⎛⎜ x[n] + x[n + N ]⎟⎞ W 2mn =
∑ (x[n] − x[n])W N = 0, 0 ≤ m ≤ − 1.
N
2 ⎠
2
n=0
n=0 ⎝

N −1

(b) X [4l] = ∑ x[n]W N4ln
n=0

=

( N / 4 ) −1

n=0

x[n]W N4ln +

( N / 2 ) −1

n= N / 4

x[n]W N4ln +

( 3 N / 4 ) −1

n= N / 2

Not for sale.

x[n]W N4ln +

N −1

4 ln
∑ x[n]W N

n = 3N / 4

116

N
−1
4 ⎛

4l ( n + )
4l ( n + )
4l ( n + ) ⎞⎟
N
N
3N
4 ln
4
2
4
= ∑ ⎜ x[n]W N + x[n + ]W N
+ x[n + ]W N
+ x[n + ]W N
⎟⎟
4
2
4

n=0
N

N

3N

N
−1
4

3N
N
N
= ∑ ⎛⎜ x[n] + x[n + ]W Nln + x[n + ]W N2ln + x[n + ]W N3ln ⎞⎟ W N4ln
4
2
4

n=0⎝
N
−1
4

= ∑ ( x[n] − X [n] + x[n] − x[n]) W N4ln = 0 as W NlN = W N2lN = W N3lN = 1.
n=0

N −1

N −1

n=0
N −1

n=0

5.25 (a) X [ N − k ] = ∑ x[n]W N( N − k ) n = ∑ x[n]W N− kn + X * [k ].
N −1

(b) X [0] = ∑ x[n]W N0 ∑ x[n] which is real.
n=0
N −1

N
2

n=0

N −1

(c) X [ ] = ∑ x[n]W N( N / 2) n = ∑ (−1) n x[n] which is real.
n=0

n=0

N −1

5.26 X [k ] = ∑ x[n]W Nnk .
n=0

N −1

(a) X * [k ] = ∑ x * [n]W N− nk . Replacing n by N − n in the summation we obtain
n=0

N −1

N −1

n=0

n=0

X * [k ] = ∑ x * [ N − n]W N− ( N − n) k = ∑ x * [ N − n]W Nnk . Thus,
DFT{x * [ N − n]} = DFT{x * [〈− n〉 N } = X * [k ].
1
2

(b) Re{x[n]} = {x[n] + x * [n]}. Taking the DFT of both sides and using the results

of Part (a) we get DFT{Re{x[n]}} =

1
{X [k ] +
2

X * [〈− k 〉 N }.

(c) j Im{x[n]} = {x[n] − x * [n]}. Thus, DFT{ j Im{x[n]}} =
1
2

1
{X [k ] −
2

X * [〈− k 〉 N }.

1

(d) x cs [n] = {x[n] + x * [〈− n〉 N ]}. Using the linearity property and results of Part (b)
2
1

we get DFT{xcs [n]} = { X [k ] + X * [k ]} = Re{ X [k ]}.
2
1

(e) x ca [n] = {x[n] − x * [〈− n〉 N ]}. Using the linearity property and results of Part (b)
2
1
2

we get DFT{xca [n]} = { X [k ] − X * [k ]} = j Im{X [k ]}.

Not for sale.

117

5.27 Since for a real sequence, x[ n] = x * [ n], taking the DFT of both sides we get
X [k ] = X * [〈− k 〉 N ]. This implies
Re{ X [k ]} + j Im{X [k ]} = Re{ X [〈− k 〉 N ]} − j Im{X [〈− k 〉 N ]}.
Comparing real and imaginary parts we get Re{ X [k ]} = Re{ X [〈− k 〉 N ]} and
Im{X [k ]} = − Im{X [〈− k 〉 N ]}.

Also, X [k ] = (Re{ X [k ]}) 2 + (Im{X [k ]}) 2
= (Re{ X [〈− k 〉 N ]}) 2 + (Im{X [〈− k 〉 N ]}) 2 = X [〈− k 〉 N ] and

⎛ − Im{X [〈− k 〉 N ]} ⎞
⎛ Im{X [k ]} ⎞
⎟⎟ = − arg{X [〈− k 〉 N ]}.
⎟⎟ = tan −1 ⎜⎜
arg{X [k ]} = tan −1 ⎜⎜
〈−

Re{
X
[
k
]}
Re{
X
[
k
]}

N

5.28 (a) x1[〈− n〉 9 ] = {4 3 − 5 1 − 2 − 2 1 − 5 3} = x1[n]. Thus, x1[n] is a
circular even sequence and hence, it has a real-valued 9 -point DFT.
(b) x2 [〈− n〉 9 ] = {0 − 5 − 1 − 4 3 − 3 4 1 5} = − x2 [n]. Thus, x2 [n] is a
circular odd sequence and hence, it has an imaginaryl-valued 9 -point DFT.
(c) x3 [〈−n〉 9 ] = {0 − 5 − 1 − 4 3 − 3 4 2 − 5} which is neither equal to

x 3 [n] nor equal to − x 3 [n]. Thus, x3 [ n] has a complex-valued 9 -point DFT.
(d) x4 [〈− n〉 9 ] = {− 5 5 − 2 2 4 4 2 − 2 5} = x 4 [n]. Thus, x4 [n] is a
circular even sequence and hence, it has a real-valued 9 -point DFT.
5.29 (a) h[n] = g[〈 n − 5〉 8 ]. Hence, H [k ] = W8−5k G[k ] = e j10πk / 8G[k ] = e j 5πk / 4 G[k ]

= {2.6 + j 4.1, e j 5π / 4 (3 − j 2.7), e j 5π / 2 (−4.2 + j1.4), e j15π / 4 (3.5 − j 2.6),
e j 5π (0.5), e j 25π / 4 (1.3 + j 4.4), e j15π / 2 (2.4 − j1.6), e j 35π / 4 (−3 + j1.6)}.
(b) H [k ] = G[〈 k + 3〉 8 ]. Hence, h[n] = W83 g[n] = e − j 6πn / 8 g[n] = e − j 3πn / 4 g[n]

= {−0.1 − j 0.7, e − j 3π / 4 (1.3 + j ), e − j 3π / 2 (2 + j 0.7), e − j 9π / 4 (1.1 + j 2.2),
e − j 3π (−0.8 + j 0.2), e − j15π / 4 (3.4 − j 0.1), e − j 9π / 2 (−1.2 + j 3.1), e − j 21π / 4 ( j1.5)}.
5.30 (a) y[ n] = αg[ n] + βh[ n]. Therefore,
N −1

N −1

N −1

n=0

n=0

n=0

Y [k ] = ∑ y[n]W Nnk = α ∑ g[n]W Nnk + β ∑ h[n]W Nnk = αG[k ] + βH [k ].
N −1

(b) x[n] = g[〈 n − no 〉 N ]. Therefore, X [k ] = ∑ g[〈 n − no 〉 N ]W Nnk
n=0

Not for sale.

118

n o −1

N −1

n=0

n = no

= ∑ g[ N + n − no ]W Nnk + ∑ g[n − no ]W Nnk

=

N −1

(n + no − N ) k

∑ g[n]W N

n = N − no

−ko n

(c) u[n] = W N

+

N − n −1

(n + no )k

∑ g[no ]W N

n=0

n k N −1

= WN o

n k

nk
∑ g[n]W N = W N o G[k ].

n=0

N −1

N −1

n=0

n=0

(k − k o )n

g[n]. Therefore, U [k ] = ∑ u[n]W Nkn = ∑ g[n]W N

⎧ N −1
(k − k o )n
,
if k ≥ k o ,
⎪⎪ ∑ g[n]W N
=
0
n
= ⎨ N −1
⎪ ∑ g[n]W ( N + k − k o ) n , if k < k o .
N
⎪⎩ n = 0
if k ≥ k o ,
⎧ G[k − k o ],
= G[〈 k − k o 〉 N ].
Thus, U [k ] = ⎨
⎩G[ N + k − k o ], if k < k o ,
N −1

N −1

(d) h[ n] = G[ n]. Therefore, H [k ] = ∑ h[n]W Nnk = ∑ G[n]W Nnk
N −1 N −1

N −1

n=0
N −1 ( k + r ) n

n=0 r =0

r =0

n=0

= ∑ ∑ g[r ]W Nnr W Nkr = ∑ g[r ] ∑ W N

n=0

. The second sum is nonzero only if

k = r = 0 or else if r = N − k and k ≠ 0. Hence,
if k = 0,
⎧ Ng[0],
H [k ] = ⎨
= Ng[〈− k 〉 N ].
⎩ Ng[ N − k ], if k > 0,
N −1

N −1 N − 1
n=0 m=0

N −1

m=0
N −1

N −1

m =0

n=0

m =0

(e) u[ n] = ∑ g[ m] h[〈− n − m〉 N ]. Therefore, U[ k ] = ∑ ∑ g[ m] h[〈− n − m〉 N ]WNnk

= ∑ g[ m] ∑ h[〈− n − m〉 N ]WNnk = ∑ g[ m] H[ k ]WNmk = H[ k ] G[ k ].
N −1

5.31

∑ g[ n] h * [ n] =

n=0

=

N −1
1 N −1 N − 1
1 N −1
− nk
− nk
∑ ∑ G[ k ]WN h * [ n] =
∑ G[ k ] ∑ h * [ n]WN
N n=0 k =0
N k =0
n=0

1 N −1
∑ G[ k ] H * [ k ].
N k =0

N −1⎛ N −1
N −1

5.32 DFT{ r xy [l]} = R xy [l] = ∑ r xy [l]WNlk = ∑ ⎜⎜ ∑ x[ n] y[〈l + n〉 N ⎟⎟ WNlk
l=0 ⎝ n=0
l=0

N −1
⎛ N −1
⎞ N −1
⎛ N −1

= ∑ x[ n]⎜⎜ ∑ y[〈l + n〉 N ]WNlk ⎟⎟ = ∑ x[ n]⎜⎜ ∑ y[ m]WNl( m − n) ⎟⎟
n =0
⎝ l=0
⎠ n=0
⎝ l=0

Not for sale.

119

N −1
⎛ N −1

= ∑ x[ n]WN− nk ⎜⎜ ∑ y[ m]WNlm ⎟⎟ = X * [ k ]Y [ k ].
n=0
⎝ l=0

5.33 Note X [k ] is the MN –point DFT of the sequence x e [n] obtained from x[n] by
appending it with M ( N − 1) zeros. Thus, the length- MN sequence y[n] is given by
M −1

y[ n] = ∑ x e [〈 n − Nl〉 MN ], 0 ≤ n ≤ MN − 1. Taking the MN –DFT of both sides we
l =0

⎛ M −1 Nkl ⎞
⎛ M −1

⎟⎟ X[ k ] = ⎜⎜ ∑ WNkl ⎟⎟ X [ k ].
get Y [ k ] = ⎜⎜ ∑ WMN
⎝ l=0

⎝ l=0

9

5.34 (a) X[0] = ∑ x[ n] = 30.
n=0
9

(b) X[5] = ∑ (−1) n x[ n] = 0.
n=0

9

(c)

∑ X[ k ] = 10 ⋅ x[0] = −30.

k =0

9
(d) The inverse DFT of e − j 2 πk / 5 X[ k ] is x[〈 n − 2〉 10 ]. Thus, ∑ e − j 2 πk / 5 X[ k ]
k =0

= 10 ⋅ x[〈0 − 2〉 10 ] = 10 ⋅ x[8] = −100.
9

9

2

2

(e) From Parseval’s relation, ∑ X[ k ] = 10 ⋅ ∑ x[ n] = 38600.
k =0

n=0

5.35 X[7] = X * [〈−7〉12 ] = X * [5] = 2 − j, X[8] = X * [〈−8〉12 ] = X * [ 4] = −3 − j 2,
X[9] = X * [〈−9〉12 ] = X * [3] = 6 − j3, X[10] = X * [〈−10〉12 ] = X * [2] = 1 + j12,
X[11] = X * [〈−11〉12 ] = X * [1] = 8 + j 2.
(a) x[0] =

1 11
∑ X[ k ] = 4.5,
12 k = 0

(b) x[6] =

1 11
k
∑ (−1) X[ k ] = −0.8333,
12 k = 0

11

(c)

∑ x[ n] = X[0] = 11,

n=0

−4 n
−4 n
x[ n]. Then, DFT{W12
x[ n]} = X [〈 k − 4〉 12 ].
(d) Let g[ n] = e j 2 πn / 3 x[ n] = W12

11

11

n=0

n=0

Thus, ∑ g[ n] = ∑ e j 2 πn / 3 x[ n] = X[〈0 − 4〉 12 ] = X[8] = −3 − j 2,
11

2

(e) From Parseval’s relation, ∑ x[ n] =
n =0

1 11
2
∑ X[ k ] = 74.8333.
12 k =0

Not for sale.

120

6

5.36 Now, yC [ n] = ∑ g[l] h[〈 n − l〉 7 ]. Hence,
l=0

yC [0] = g[0] h[0] + g[1] h[6] + g[2] h[5] + g[3] h[ 4] + g[ 4] h[3] + g[5] h[2] + g[6] h[1],
yC [1] = g[0] h[1] + g[1] h[0] + g[2] h[6] + g[3] h[5] + g[ 4] h[ 4] + g[5] h[3] + g[6] h[2],
yC [2] = g[0] h[2] + g[1] h[1] + g[2] h[0] + g[3] h[6] + g[ 4] h[5] + g[5] h[ 4] + g[6] h[3],
yC [3] = g[0] h[3] + g[1] h[2] + g[2] h[1] + g[3] h[0] + g[ 4] h[6] + g[5] h[5] + g[6] h[ 4],
yC [ 4] = g[0] h[ 4] + g[1] h[3] + g[2] h[2] + g[3] h[1] + g[ 4] h[0] + g[5] h[6] + g[6] h[5],
yC [5] = g[0] h[5] + g[1] h[ 4] + g[2] h[3] + g[3] h[2] + g[ 4] h[1] + g[5] h[0] + g[6] h[6],
yC [6] = g[0] h[6] + g[1] h[5] + g[2] h[ 4] + g[3] h[3] + g[ 4] h[2] + g[5] h[1] + g[6] h[0].
6

Likewise, y L [ n] = ∑ g[l] h[ n − l]. Hence,
l=0

y L [0] = g[0]h[0],
y L [1] = g[0]h[1] + g[1]h[0],
y L [2] = g[0]h[2] + g[1]h[1] + g[2]h[0],
y L [3] = g[0]h[3] + g[1]h[2] + g[2]h[1] + g[3]h[0],
y L [ 4] = g[0]h[ 4] + g[1]h[3] + g[2]h[2] + g[3]h[1] + g[ 4]h[0],
y L [5] = g[0]h[5] + g[1]h[ 4] + g[2]h[3] + g[3]h[2] + g[ 4]h[1] + g[5]h[0],
y L [6] = g[0]h[6] + g[1]h[5] + g[2]h[ 4] + g[3]h[3] + g[ 4]h[2] + g[5]h[1] + g[6]h[0],
y L [7] = g[1]h[6] + g[2]h[5] + g[3]h[ 4] + g[ 4]h[3] + g[5]h[2] + g[6]h[1],
y L [8] = g[2]h[6] + g[3]h[5] + g[ 4]h[ 4] + g[5]h[3] + g[6]h[2],
y L [9] = g[3]h[6] + g[ 4]h[5] + g[5]h[ 4] + g[6]h[3],
y L [10] = g[ 4]h[6] + g[5]h[5] + g[6]h[ 4],
y L [11] = g[5]h[6] + g[6]h[5],
y L [12] = g[6]h[6].
Comparing yC [n] with y L [n] we observe that
yC [0] = y L [0] + y L [7],
yC [1] = y L [1] + y L [8],
yC [2] = y L [2] + y L [9],
yC [3] = y L [3] + y L [10],
yC [ 4] = y L [ 4] + y L [11],
yC [5] = y L [5] + y L [12],
yC [6] = y L [6].
5.37 Since x[n] is a length-9 real sequence, its DFT satisfies X[ k ] = X * [〈− k 〉 9 ]. Therefore,
X[1] = X * [〈−1〉 9 ] = X * [8] = −7.7 + j 3.2,

Not for sale.

121

X[3] = X * [〈−3〉 9 ] = X * [6] = 8.6 + j 9.6,
X[5] = X * [〈−5〉 9 ] = X * [ 4] = −3.5 − j 5.3,
X[7] = X * [〈−7〉 9 ] = X * [2] = 1.2 + j 4.1.
5.38 X[1] = X * [〈−1〉 9 ] = X * [8] = 4.5 − j1.6,
X[ 4] = X * [〈−4〉 9 ] = X * [5] = −3.1 − j8.2,
X[6] = X * [〈−6〉 9 ] = X * [3] = −7.2 + 4.1,
X[7] = X * [〈−7〉 9 ] = X * [2] = 1.2 + j 2.3.
5.39 Since the DFT X [k ] is real-valued, x[n] is a circularly even sequence, i.e.,
x[ n] = x[〈− n〉12 ]. Therefore,
x[1] = x[〈−1〉12 ] = x[11] = −2,
x[ 4] = x[〈−4〉12 ] = x[8] = 9.3,
x[7] = x[〈−7〉 12 ] = x[5] = 4.1,
x[9] = x[〈−9〉 12 ] = x[3] = −3.25,
x[10] = x[〈−10〉12 ] = x[2] = 0.7.
5.40 Since the DFT X [k ] is imaginary-valued, x[n] is a circularly odd sequence, i.e.,
x[ n] = − x[〈− n〉12 ]. Therefore,
x[7] = − x[〈−7〉 12 ] = − x[5] = 9.3,
x[8] = − x[〈−8〉12 ] = − x[ 4] = −2.87,
x[9] = − x[〈−9〉12 ] = − x[3] = −4.1,
x[10] = − x[〈−10〉12 ] = − x[2] = 3.25,
x[11] = − x[〈−11〉12 ] = − x[1] = −0.7.
5.41 X[ k ] = X * [〈− k 〉174 ] = X * [174 − k ].
X [9] = X * [174 − 9] = X * [165] = −3.4 + j 5.9 ⇒ X [165] = −3.4 − j 5.9.
X [51] = X * [174 − 51] = X * [123] = 5 − j1.6 ⇒ X [123] = 5 + j1.6.
X [113] = X * [174 − 113] = X * [61] = 8.7 − j 4.9 ⇒ X [61] = 8.7 + j 4.9.
X [162 ] = X * [174 − 162 ] = X * [12 ] = 7.1 − j 2.4 ⇒ X [12 ] = 7.1 + j 2.4.
X[ k1 ] = 7.1 + j 2.4, X[ k 2 ] = 8.7 + j 4.9, X[ k3 ] = 5 + j1.6, X[ k 4 ] = −3.4 − j 5.9.
(a) Comparing these 4 DFT samples with the DFT samples given above we conclude
k1 = 12, k2 = 61, k3 = 123, k 4 = 165.
(b) dc value of {x[ n]} = X [0] = 11.
(c) x[ n] =

1 173
1
− kn
− 9n
− 51n
( X[0] + 2 Re{X[9]W174
} + 2 Re{X[51]W174
}
∑ X[ k ]W174 =
174 k = 0
174
−87 n
−113n
−162 n
+ X [87] W174
+ 2 Re{X [113] W174
} + 2 Re{X [162 ] W174
})

Not for sale.

122

173

(d)

2

∑ x[ n] =

n=0

1 173
2
∑ X[ k ] = 86.0279.
174 k = 0

5.42 X[ k ] = X * [〈− k 〉 126 ] = X * [126 − k ].
X [ 0] = 12.8 + jα.
X [13] = X * [126 − 13] = X * [113] = −3.7 + j 2.2 ⇒ X [113] = −3.7 − j 2.2.
X [ 51] = X * [126 − 51] = X * [75] = − j1.7 ⇒ X [ 75] = j1.7.
X [13] = X * [126 − 13] = X * [113] = −3.7 + j 2.2 ⇒ X [113] = −3.7 − j 2.2.
X[ k1 ] = X * [126 − k1 ] = 9.1 − j 5.4 ⇒ X[126 − k1 ] = 9.1 + j 5.4.
X[ k2 ] = X * [126 − k2 ] = 6.3 + j 2.3 ⇒ X[126 − k2 ] = 6.3 − j 2.3.
X[ k3 ] = X * [126 − k3 ] = γ + j1.7 ⇒ X[126 − k3 ] = γ − j1.7.

X[ k 4 ] = X * [126 − k 4 ] = −3.7 − j 2.2 ⇒ X[126 − k 4 ] = −3.7 + j 2.2.
(a) (b) Since x[n] is a real-valued sequence of length 126, X [0] and X[63] must be
real. Thus, α = 0 and β = 0. As X[126 − k1 ] and X[108] have the same imaginary
part, ε = 9.1 and k1 = 126 − 108 = 18. Likewise, as X[ k2 ] and X [79] have the same
real part, δ = −2.3 and k2 = 126 − 79 = 47. Similarly, as X[126 − k3 ] and X[51] have

the same imaginary part, γ = 0 and k3 = 126 − 51 = 75. Finally, as

X[ k 4 ] = X[113], k 4 = 113.
(c) dc value of x[n] is X[0] = 12.8.
(d) x[ n] =

1 125
1
− kn
− 26 πn / 126
X[0] + 2 Re{X[13]W126
}
∑ X[ k ]W126 =
126 k = 0
126

(

− 36 πn / 126
−94 πn / 126
− 63πn
+ 2 Re{X[18]W126
} + 2 Re{X[ 47]W126
} + X[63]W126
−150πn / 126
− 226πn / 126
+ 2 Re{X[75]W126
} + 2 Re{X[113]W126
} .

)

125

(e)

2

∑ x[ n] =

n=0

1 125
2
∑ X[ k ] = 5.767.
126 k = 0

5.43 Y [ k ] = W3−2 k X [ k ] = W9−6 k X [ k ]. Therefore, y[ n] = x[〈 n − 6〉 9 ]. Thus,
y[ 0] = x[3] = 4, y[1] = x[ 4] = −3, y[ 2] = x[ 5] = 5, y[3] = x[6] = −2, y[ 4] = x[ 7] = −2,
y[ 5] = x[8] = 4, y[6] = x[ 0] = 3, y[ 7] = x[1] = 5, y[8] = x[ 2] = 1.
5.44 H[ k ] = H * [〈− k 〉 9 ] = X * [9 − k ]. Hence, H [ 5] = H * [ 4] = −6.876 − j11.4883,
H [6] = H * [3] = − j8.6603, H [ 7] = H * [ 2] = 6.0346 + j1.957,
H [8] = H * [1] = 6.8414 + j 6.0572 .

Not for sale.

123

Now g[ n] = e j 2 πn / 3 h[ n] = e j 6 πn / 9 h[ n] = W9−6 n h[ n]. Therefore,
G[ k ] = H[〈 k − 6〉 9 ], 0 ≤ k ≤ 8. Thus, G[0] = H[〈−6〉 9 ] = H[3] = j8.6603,
G[1] = H[〈1 − 6〉 9 ] = H[ 4] = −6.876 − j11.4883,
G[2] = H[〈2 − 6〉 9 ] = H[5] = H * [ 4] = −6.876 + j11.4883,
G[3] = H[〈3 − 6〉 9 ] = H[6] = H * [3] = − j8.6603,
G[ 4] = H[〈 4 − 6〉 9 ] = H[7] = H * [2] = 6.0346 + j1.957,
G[5] = H[〈5 − 6〉 9 ] = H[8] = H * [1] = 6.8414 + j.0572,
G[6] = H[〈6 − 6〉 9 ] = H[0] = 15,
G[7] = H[〈7 − 6〉 9 ] = H[1] = 6.8414 − j 6.0572,
G[8] = H[〈8 − 6〉 9 ] = H[2] = 6.0346 − j1.957.
5.45 (a) y L [0] = g[0] h[0] = −4,
y L [1] = g[0] h[1] + g[1] h[0] = 10,
y L [2] = g[0] h[2] + g[1] h[1] + g[2] h[0] = −6,
y L [3] = g[0] h[3] + g[1] h[2] + g[2] h[1] = 8,
y L [ 4] = g[1] h[3] + g[2] h[2] = 7,
y L [5] = g[2] h[3] = −3.
(b) yC [0] = g e [0] h[0] + g e [1] h[3] + g e [2] h[2] + g e [3] h[1]
= g[ 0]h[ 0] + g[1]h[3] + g[ 2]h[ 2] = 3,
yC [1] = g e [0] h[1] + g e [1] h[0] + g e [2] h[3] + g e [3] h[2]
= g[0]h[1] + g[1]h[ 0] + g[ 2]h[3] = 7,
yC [2] = g e [0] h[2] + g e [1] h[1] + g e [2] h[0] + g e [3] h[3]
= g[ 0]h[ 2] + g[1]h[1] + g[2]h[ 0] = −6,
yC [3] = g e [0] h[3] + g e [1] h[2] + g e [2] h[1] + g e [3] h[0]
= g[0]h[3] + g[1]h[ 2] + g[ 2]h[1] = 8.
⎡Ge [0]⎤ ⎡1 1
1
1 ⎤⎡ 2 ⎤ ⎡ 4 ⎤
⎢ Ge [1] ⎥ ⎢1 − j − 1 j ⎥ ⎢− 1⎥ ⎢− 1 + j ⎥
=
=
,
(c) ⎢
G [2]⎥ ⎢1 − 1 1 − 1⎥ ⎢ 3 ⎥ ⎢ 6 ⎥
⎢ e ⎥ ⎢1 j − 1 j ⎥ ⎢ 0 ⎥ ⎢ − 1 − j ⎥
⎦⎣ ⎦ ⎣

⎣Ge [3]⎦ ⎣
⎡ H[0]⎤ ⎡1 1
1
1 ⎤ ⎡− 2⎤ ⎡ 3 ⎤
⎢ H [1] ⎥ ⎢1 − j − 1 j ⎥ ⎢ 4 ⎥ ⎢ − 4 − j 5⎥
⎢ H[2]⎥ = ⎢1 − 1 1 − 1⎥ ⎢ 2 ⎥ = ⎢ − 3 ⎥.
⎢ H [3]⎥ ⎢⎣1 j − 1 j ⎥⎦ ⎣⎢ − 1 ⎦⎥ ⎢⎣− 4 + j 5⎥⎦

⎡YC [0]⎤ ⎡Ge [0]H [0]⎤ ⎡ 12 ⎤
⎢ YC [1] ⎥ ⎢ Ge [1]H [1] ⎥ ⎢9 + j ⎥
⎢Y [2]⎥ = ⎢G [2]H[2]⎥ = ⎢ − 18 ⎥. Therefore
⎥ ⎢9 − j ⎥
⎢ C ⎥ ⎢ e

⎣YC [3]⎦ ⎣ Ge [3]H[3]⎦ ⎣

Not for sale.

124

⎡ yC [0]⎤
1
1 ⎤ ⎡ 12 ⎤ ⎡ 3 ⎤
⎡1 1
⎢ yC [1] ⎥ 1 ⎢1 j − 1 − j ⎥ ⎢9 + j ⎥ ⎢ 7 ⎥
⎢ y [2]⎥ = ⎢1 − 1 1 − 1⎥ ⎢ − 18 ⎥ = ⎢− 6 ⎥.
⎢ C ⎥ 4 ⎢1 − j − 1 j ⎥ ⎢9 − j ⎥ ⎢ 8 ⎥

⎦⎣
⎦ ⎣ ⎦
⎣ yC [3]⎦
(d) g e [ n] = [2, − 1, 3, 0, 0, 0], he [ n] = [− 2, 4, 2, − 1, 0, 0].
yC [0] = g e [0] he [0] + g e [1] he [5] + g e [2] h e [ 4] + g e [3] he [3] + g e [ 4] he [2] + g e [5] he [1]

= g[0]h[0] = −4 = y L [0],
yC [1] = g e [0] he [1] + g e [1] he [0] + g e [2] h e [5] + g e [3] he [6] + g e [ 4] he [3] + g e [5] he [2]
= g[0]h[1] + g[1]h[0] = 10 = y L [1],
yC [2] = g e [0] he [2] + g e [1] he [1] + g e [2] h e [0] + g e [3] he [5] + g e [ 4] he [ 4] + g e [5] he [3]
= g[0]h[2] + g[1]h[1] + g[2]h[0] = −6 = y L [2],
yC [3] = g e [0] he [3] + g e [1] he [2] + g e [2] h e [1] + g e [3] he [0] + g e [ 4] he [5] + g e [5] he [ 4]
= g[0]h[3] + g[1]h[2] + g[2]h[1] = 8 = y L [3],
yC [ 4] = g e [0] he [ 4] + g e [1] he [3] + g e [2] h e [2] + g e [3] he [1] + g e [ 4] he [0] + g e [5] he [5]
= g[1]h[3] + g[2]h[2] = 7 = y L [ 4],
yC [5] = g e [0] he [5] + g e [1] he [ 4] + g e [2] h e [3] + g e [3] he [2] + g e [ 4] he [1] + g e [5] he [0]
= g[2]h[3] = −3 = y L [5].
5.46 We need to show g[ n] O
N h[ n] = h[ n] O
N g[ n]. Let
N −1

x[ n] = g[ n] O
N h[ n] = ∑ g[ m ]h[ 〈 n − m〉 N ] and
m=0
N −1

n

N −1

m=0
N −1

m = n +1

y[ n] = h[ n] O
N g[ n] = ∑ h[ m ]g[ 〈 n − m〉 N ] = ∑ h[ m ]g[ n − m ] +
n

= ∑ h[ n − m]g[ m] +
m =0

m=0
N −1

∑ h[ m]g[ N + n − m]

∑ h[ N + n − m]g[ m] = ∑ h[〈 n − m〉 N ]g[ m] = x[ n].

m = n +1

m=0

N −1

5.47 (a) y[ n] = x1[ n] O
N x 2 [ n] = ∑ x1[ m ] x 2 [ 〈 n − m〉 N ]. Thus,
m=0

N −1

N −1

N −1

⎛ N −1
⎞⎛ N −1

∑ y[ n] = ∑ x1[ m] ∑ x[〈 n − m〉 N ] = ⎜⎜ ∑ x1[ n] ⎟⎟⎜⎜ ∑ x 2 [ n] ⎟⎟.
n=0
m =0
n =0
⎝ n=0
⎠⎝ n = 0

(b)
N −1

N −1

N −1

n
n
∑ (−1) y[ n] = ∑ x1[ m] ∑ x[〈 n − m〉 N ](−1)

n=0
⎛ N −1

m=0
n=0
⎞⎛ m −1
= ⎜⎜ ∑ x1[ m] ⎟⎟⎜⎜ ∑ x 2 [ N + m − n](−1) n
⎝ m =0
⎠⎝ n = 0

N −1

+ ∑ x 2 [ m − n](−1) n ⎟⎟. Replacing n by
n=m

Not for sale.

125

N + n − m in the first sum on the right-hand side and by n − m in the second sum on
the right-hand side we obtain
N −1
N −1
⎛ N −1
⎞⎛ m −1

n
n−N +m
+ ∑ x 2 [ n](−1) n + m ⎟⎟
∑ (−1) y[ n] = ⎜⎜ ∑ x1[ m] ⎟⎟⎜⎜ ∑ x 2 [ n](−1)
n =0
n=m
⎝ m =0
⎠⎝ n = 0

⎛ N −1
⎞⎛ N −1

= ⎜⎜ ∑ (−1) n x1[ n] ⎟⎟⎜⎜ ∑ (−1) n x 2 [ n] ⎟⎟.
⎝ n=0
⎠⎝ n = 0

5.48 y[ n] = x[3n], 0 ≤ n ≤

Now, x[ n] =

Y [k ] =

N
3

N
−1
3

N
−1
3

n=

n=

− 1. Therefore, Y [ k ] = ∑ y[ n]WNnk/ 3 = ∑ x[ 4 n]WNnk/ 3 .
N −1

N −1

1
1
− mn
− 3mn
=
∑ X [ m ] WN
∑ X [ m]WN / 3 . Hence,
N m =0
N m=0

N
−1
N −1
3

1 N −1

N
−1
3

1
( k − m )n
− mn nk
∑ ∑ X [ m ] WN / 3 WN / 3 =
∑ X[ m] ∑ WN / 3 . Since
N n =0 m =0
N m=0
n=0

N
−1
3

N
2N
⎧N
, k + N,
( k − m )n ⎪ , m = k , k + , k +
Thus,
=⎨3
∑ WN / 3
3
3
⎪⎩ 0,
n=0
elsewhere.
1
N
2N
Y [ k ] = ⎛⎜ X [ k ] + X [ k + ] + X [ k +
] + X [ k + N ] ⎞⎟.
3
3

3⎝
1
2

5.49 v[ n] = x[ n] + jy[ n]. Hence, X[ k ] = {V [ k ] + V * [〈− k 〉 8 ]} is the 8 –point DFT of x[n]

and Y [ k ] =

1
{V [ k ] − V * [〈− k 〉 8 ]} is the
2j

8 –point DFT of y[n].

V [ k ] = [3 + j 7, − 2 + j 6, 1 − j 5, 4 − j 9, 5 + j 2, 3 − j 2, j 4, − 3 − j8].
V * [〈− k 〉 8 ] = [3 + j 7, − 3 + j8, − j 4, 3 + j 2, 5 − j 2, 4 + j 9, 1 + j 5, − 2 − j 6].
Therefore,
5
1
9
7
7
7
7
5
1
9
X[ k ] = ⎡3 + j 7, − + j 7,
+ j ,
− j , 5,
+ j ,
− j , − − j 7 ⎤,
⎥⎦
⎢⎣
2
2
2
2
2
2
2
2
2
2
1
1
1
11
1
11
1
1
1
1
Y [ k ] = ⎡0, − 1 − j , − − j , − − j , 2, − + j , − + j , − 1 + j ⎤.
⎢⎣
2 ⎥⎦
2
2
2
2
2
2
2
2
2
The IDFT of V [k ] obtained using MATLAB is given by
Columns 1 through 4
1.3750 - 0.6250i
-0.9331 - 0.1187i

-0.8044 + 1.7223i

-1.7500 + 1.2500i

Columns 5 through 8
0.8750 + 2.6250i
-1.8169 + 1.1187i

2.5544 - 0.2223i

Not for sale.

3.5000 + 1.2500i

126

The same result is obtained by computing the IDFT x[n] of X [k ] and the IDFT y[n]
of Y [k ] using MATLAB and then forming x[ n] + jy[ n].
5.50 v[ n] = g[ n] + j h[ n] = [2 − j 2, − 1 + j 4, 3 + j 2, − j ]. Therefore,
⎡V [0]⎤ ⎡1 1
1
1 ⎤ ⎡ 2 − j 2 ⎤ ⎡ 4 + j3 ⎤
⎢ V [1] ⎥ ⎢1 − j − 1 j ⎥ ⎢− 1 + j 4⎥ ⎢ 4 − j 3 ⎥
⎢V [2]⎥ = ⎢1 − 1 1 − 1⎥ ⎢ 3 + j 2 ⎥ = ⎢ 6 − j 3 ⎥, i.e.,
⎢V [3]⎥ ⎢⎣1 j − 1 − j ⎥⎦ ⎢ − j ⎥ ⎢− 6 − j 5⎥

⎦ ⎣

V [ k ] = [4 + j3, 4 − j3, 6 − j3, − 6 − j 5].
Thus, V * [〈− k 〉 4 ] = [4 + j3, − 6 + j 5, 6 + j 3, 4 + j3]. Therefore,

) [

(

1
V [ k ] + V * [〈− k 〉 4 ] = 4,
2
1
H[ k ] =
V [ k ] − V * [〈− k 〉 4 ] = 3,
2j

G[ k ] =

) [

(

− 1 + j, 6, − 1 − j ] and

− 4 − j 5, − 3, − 4 + j 5].

5.51 Let p[ n] = IDFT{P[ k ]}. Thus,
⎡ p[0]⎤
1
1 ⎤ ⎡ − 5 ⎤ ⎡ − 1.25 ⎤
⎡1 1
⎢ p[1] ⎥ 1 ⎢1 j − 1 − j ⎥ ⎢− 2 + j 5⎥ ⎢− 4.75⎥
⎢ p[2]⎥ = ⎢1 − 1 1 − 1⎥ ⎢ 4 ⎥ = ⎢ 0.75 ⎥
⎥ 4 ⎢1 − j − 1 j ⎥ ⎢− 2 − j 5⎥ ⎣⎢ 0.25 ⎦⎥

⎦⎣

⎣ p[3]⎦
⎡d[0]⎤
1
1 ⎤⎡ 3 ⎤ ⎡ 1 ⎤
⎡1 1
⎢ d[1] ⎥ 1 ⎢1 j − 1 − j ⎥ ⎢4 + j ⎥ ⎢ 2 ⎥
⎢d[2]⎥ = 4 ⎢1 − 1 1 − 1⎥ ⎢ − 7 ⎥ = ⎢− 3⎥. Therefore,
⎢ d[3]⎥
⎢⎣1 − j − 1 j ⎥⎦ ⎢⎣4 − j ⎥⎦ ⎢⎣ 3 ⎥⎦

X ( e jω ) =

− 1.25 − 4.75 e − jω + 0.75 e − j 2 ω + 0.25 e − j 3ω

.
1 + 2 e − j ω − 3 e − j 2 ω + 3 e − j 3ω
5.52 Let p[ n] = IDFT{P[ k ]}. Thus,
⎡ p[0]⎤
1
1 ⎤ ⎡ 8 ⎤ ⎡ − 1.25 ⎤
⎡1 1
⎢ p[1] ⎥ 1 ⎢1 j − 1 − j ⎥ ⎢− 5 − j 6 ⎥ ⎢ 5.75 ⎥
⎢ p[2]⎥ = ⎢1 − 1 1 − 1⎥ ⎢ − 3 ⎥ = ⎢ 3.75 ⎥
⎥ 4 ⎢1 − j − 1 j ⎥ ⎢− 5 + j 6 ⎥ ⎢⎣− 0.25⎥⎦

[
3
]
p
⎦⎣

⎡d[0]⎤
1
1 ⎤ ⎡ 3 ⎤ ⎡ 2.75 ⎤
⎡1 1
⎢ d[1] ⎥ 1 ⎢1 j − 1 − j ⎥ ⎢4 + j ⎥ ⎢− 3.75⎥
⎢d[2]⎥ = 4 ⎢1 − 1 1 − 1⎥ ⎢ − 7 ⎥ = ⎢− 3.25⎥. Therefore,
⎢ d[3]⎥
⎢⎣1 − j − 1 j ⎥⎦ ⎢⎣4 − j ⎥⎦ ⎣⎢ − 1.75 ⎦⎥

X (e

)=

− 1.25 + 54.75 e − jω + 3.75 e − j 2 ω − 0.25 e − j 3ω
2.75 − 3.75 e − jω − 3.25 e − j 2ω − 1.75 e − j 3ω
N −1

N −1

n=0

n=0

.

5.53 X (e jω ) = ∑ x[ n] e − jωn and Xˆ [ k ] = ∑ x[ n] e − j 2 πkn / M . Now,

Not for sale.

127

xˆ[ n] =

1 M −1 ˆ
1 M −1 N −1
− j 2 πkm / M − nk
− nk
WM
∑ X [ k ] WM =
∑ ∑ x[ m] e
M k =0
M k =0 m =0

N −1

1 N −1
− j 2 πk ( m − n ) / M
= ∑ x[ n + rM ].
∑ x[ m] ∑ e
M m =0
k =0
r = −∞
Thus, xˆ[ n] is obtained by shifting x[n] in multiples of M and adding the shifted
copies. Since the new sequence is obtained by shifting in multiples of hence, to recover
the original sequence take any consecutive N samples in the range 0 ≤ n ≤ N − 1 for
any value of r. This would be true only if the shifted copies of x[n] did not overlap
with each other, that is, if and only if M ≥ N .

=

8

8

n=0

n=0

5.54 (a) X (e jω ) = ∑ x[ n] e − jωn . Therefore, X1[ k ] = ∑ x[ n] e − j 2 πkn / 12 . Hence,

x1[ n] =

1 11
1 11 ⎛ 8
j 2 πkn / 12
j 2 πkm / 12 ⎞⎟ j 2 πkn / 12
=
∑ X[ k ] e
∑ ⎜⎜ ∑ x[ m] e
⎟e
12 k = 0
12 k = 0 ⎝ m = 0

11

1 8
j 2 πk ( n − m ) / 12
= ∑ x[ n + 12r ] using the results of Problem 5.53.
∑ x[ m] ∑ e
12 m = 0
k =0
r = −∞
Since M = 12 and N = 9, M > N , and hence, x[n] is recoverable from x1[ n]. In fact
x1[ n] = {1, − 2, 3, − 4, 5, − 4, 3, − 2, 1, 0, 0, 0}, 0 ≤ n ≤ 11, and x[n] is
given by the first 9 samples of x1[ n].

=

8

(b) Here, X 2 [ k ] = ∑ x[ n] e − j 2 πkn / 8 . Hence,
n=0

x1[ n] =

7 ⎛ 8

1 7
j 2 πkn / 8 1
= ∑ ⎜⎜ ∑ x[ m] e j 2 πkm / 8 ⎟⎟ e j 2 πkn / 8
∑ X[ k ] e
8 k =0
8 k =0 ⎝ m =0

7

1 8
j 2 πk ( n − m ) / 8
= ∑ x[ n + 8r ] using the results of Problem 5.53.
∑ x[ m] ∑ e
8 m =0
k =0
r = −∞
Since M = 8 and N = 9, M < N , and hence, x[n] is not recoverable from x1[ n]. In
fact x 2 [ n] = [2, − 2, 3, − 4, 5, − 4, 3, − 2, 2, − 1, 1], 0 ≤ n ≤ 10.

=

1 N −1
− kn
∑ X[ k ]WN . Let n = N − m. Then,
N k =0
1 N −1
1 N −1
1
− k( N − m)
km
x[ m] =
=
∑ X [ k ] WN
∑ X[ k ]WN = F {X [ k ]}. Therefore,
N k =0
N k =0
N
1
1
F { F {x[ n]}}, or, F { F {x[ n]}} = N ⋅ x[ N − n]. Hence,
x[ N − n] = F {X [ k ]} =
N
N

5.55 x[ n] =

F { F { F { F {x[ n]}}}} = N 2 ⋅ x[ N − n].

Not for sale.

128

100

100

k =0
50

k =0

36

5.56 y[ n] = x[ n] O
* h[ n] = ∑ x[ k ] h[ n − k ] = ∑ h[ k ] x[ n − k ] = ∑ h[ k ] x[ n − k ].
k =17

36

u[ n] = x[ n] O
N h[ n] = ∑ h[ k ] x[ 〈 n − k 〉 51 ] = ∑ h[ k ] x[ 〈 n − k 〉 51 ].
k =0

k =17

Now, for n ≥ 36, x[〈 n − k 〉 51 ] = x[ n − k ]. Thus, y[ n] = u[ n] for 36 ≤ n ≤ 50.
5.57 (a) Overlap-add method: Since the impulse response is of length and the DFT size to
be used is 128, hence, the number of data samples required for each convolution will be
128 − 109 = 19. Thus the total number of DFTs required for the length- 1300 data
1300 ⎤
= 69. Also, the DFT of the impulse response needs to be computed
sequence is ⎡
⎢ 19 ⎥
once. Hence, the total number of DFTs used are = 69 + 1 = 70. The total number of
IDFTs used are = 69.
(b) Overlap-save method: In this case, since the first 110 − 1 = 109 points are lost, we
need to pad the data sequence with 109 zeros for a total length of 1409. Again, each
convolution will result in 128 − 109 = 19 correct values. Thus the total number of
1409 ⎤
= 75. Again, 1 DFT is required for the impulse
DFTs required for the data are ⎡
⎢ 19 ⎥
response. The total number of DFTs used are 75 + 1 = 76. The total number of IDFTs
used are = 75.

⎧ x[ n / L ], n = 0, L, 2 L,K, ( N − 1)L,
5.58 (a) y[ n] = ⎨
elsewhere.
⎩ 0,
NL −1

N −1

N −1

n=0

n=0

n=0

nk
nLk
= ∑ x[ n]WNL
= ∑ x[ n]WNnk . For k ≥ N , let k = k o + rN
Y [ k ] = ∑ y[ n]WNL
N −1

n( k o + rN )

where ko = 〈 k 〉 N . Then, Y [ k ] = Y [ k o + rN ] = ∑ x[ n]WN
n=0

N −1

nk o

= ∑ x[ n]WN
n=0

= X[ ko ] = X[〈 k 〉 N ].
(b) Since Y [ k ] = X[〈 k 〉 5 ] for 0 ≤ k ≤ 20, a sketch of Y [k ] is thus as shown below.
5.59 x 0 [ n] = x[2 n + 1] + x[2n], x1[ n] = x[2 n + 1] − x[2n], y0 [ n] = y[2n + 1] + y[2n], and

y1[ n] = y[2n + 1] − y[2n], 0 ≤ n ≤

N
2

− 1. Since x[n] and y[n] are real, symmetric

sequences, it follows that x 0 [ n] and y0 [ n] are real, symmetric sequences, and x1[ n]
and y1[ n] are real, antisymmetric sequences. Now, consider the

N
2

–length sequence

u[ n] = x 0 [ n] + y1[ n] + j (x1[ n] + y0 [ n]). Its conjugate sequence is given by
u * [ n] = x 0 [ n] + y1[ n] − j (x1[ n] + y0 [ n]). Next, we observe that
u[〈− n〉 N / 2 ] = x 0 [〈− n〉 N / 2 ] + y1[〈− n〉 N / 2 ] + j (x1[〈− n〉 N / 2 ] + y0 [〈− n〉 N / 2 ])
Not for sale.

129

= x 0 [ n] − y1[ n] + j (− x1[ n] + y0 [ n]). Its conjugate sequence is given by
u * [〈− n〉 N / 2 ] = x 0 [ n] − y1[ n] − j (− x1[ n] + y0 [ n]).
By adding the last 4 sequences we get
4 x 0 [ n] = u[ n] + u * [ n] + u[〈− n〉 N / 2 ] + u * [〈− n〉 N / 2 ].
From Table 5.3, if U [ k ] = DFT{u[ n]}, then U * [〈− k 〉 N / 2 ] = DFT{u * [ n]},
U * [ k ] = DFT{u * [〈− n〉 N / 2 ]}, and U[〈− k 〉 N / 2 ] = DFT{u[〈− n〉 N / 2 ]}. Thus,
1
4

X 0 [ k ] = DFT{x 0 [ n]} =

(U[k ] + U * [〈− k〉 N / 2 ] + U[〈− k〉 N / 2 ] + U * [k ]).

Similarly,

j 4 x1[ n] = u[ n] − u * [ n] − u[〈− n〉 N / 2 ] + u * [〈− n〉 N / 2 ]. Hence,

X1[ k ] = DFT{x1[ n]} =

(

)

1
U[ k ] − U * [〈− k 〉 N / 2 ] − U[〈− k 〉 N / 2 ] + U * [ k ] . Likewise,
j4

4 y1[ n] = u[ n] − u[〈− n〉 N / 2 ] + u * [ n] − u * [〈− n〉 N / 2 ]. Thus,
Y1[ k ] = DFT{y1[ n]} =

1
4

(U[k ] − U[〈− k〉 N / 2 ] + U * [〈− k〉 N / 2 ] − U * [k ]). Finally,

j 4 y0 [ n] = u[ n] + u[〈− n〉 N / 2 ] − u * [ n] − u * [〈− n〉 N / 2 ]. Hence,

Y0 [ k ] = DFT{y0 [ n]} =
N −1

(

5.60 X GDFT [ k, a, b] = ∑ x[ n] e
n=0

j
1 N −1
x[ n] =
∑ X[ k ] e
N k =0
j
1 N −1 N − 1
=
∑ ∑ x[r ] e
N k =0 r =0
N −1 j
1 N −1
=
∑ x[r ] ∑ e
N r =0
k =0
N −1

∑ e

j

2 π( n − r )( k + b )
N

k =0

)

1
U[ k ] + U[〈− k 〉 N / 2 ] − U * [〈− k 〉 N / 2 ] − U * [ k ] .
j4
−j

2 π( n + a )( k + b )
N
.

2 π( n + a )( k + b )
N

−j
1 N −1 N − 1
=
∑ ∑ x[r ] e
N k =0 r =0

2 π( n + a − r − a )( k + b )
N
2 π( n − r )( k + b )
N

=

2 π( r + a ) k + b )
N

j
1 N −1 N − 1
=
∑ ∑ x[r ] e
N k =0 r =0

j

e

2 π( n + a ) k + b )
N

2 π( n − r )( k + b )
N

1
⋅ x[ n] ⋅ N = x[ n], as from Eq. (5.23)
N

N , if n = r ,
= ⎧⎨
⎩ 0, otherwise.

5.61 (a) x[ n] = α g[ n] + β h[ n]. Thus,
N −1

N −1

πk (2 n +1) ⎞
πk (2 n + 1) ⎞
X DCT [ k ] = ∑ x[ n] cos⎛⎜
⎟ = ∑ (α g[ n] + β h[ n]) cos⎛⎜

⎝ 2N ⎠ n=0
⎝ 2N ⎠
n=0

N −1

N −1

πk (2 n +1) ⎞
πk (2 n + 1) ⎞
= ∑ α g[ n] cos⎛⎜
⎟ + ∑ β h[ n] cos⎛⎜
⎟ = α G DCT [ k ] + β H DCT [ k ].
⎝ 2N ⎠ n =0
⎝ 2N ⎠
n=0

Not for sale.

130

N −1

N −1

πk (2 n + 1) ⎞
πk (2 n + 1) ⎞
(b) G DCT [ k ] = ∑ g[ n] cos⎛⎜
⎟ ⇒ G *DCT [ k ] = ∑ g * [ n] cos⎛⎜
⎟.
⎝ 2N ⎠
⎝ 2N ⎠
n=0
n=0

DCT

Therefore, g * [ n] ↔ G *DCT [ k ].
N , if k = m = 0,

N −1
πk (2 n +1) ⎞
πm(2 n +1) ⎞ ⎪
(c) Note that ∑ cos⎛⎜
⎟ cos⎛⎜
⎟ = ⎨ N/2, if k = m and k ≠ 0, Now,
⎝ 2N ⎠ ⎝ 2N ⎠ ⎪
n=0
0, otherwise.

g[ n] g * [ n] =

1 N −1 N −1
N2

π(2 n +1)k ⎞
π(2 n +1)k ⎞
*
⎟ cos⎛⎜
⎟.
∑ α[ k ] α[ m] G DCT [ k ]G DCT [ m] cos⎛⎜
⎝ 2N ⎠ ⎝ 2N ⎠
k =0 m =0

Thus,
N −1

2

∑ g[ n] =

n=0

1 N −1 N − 1 N − 1
N2

π(2 n +1)k ⎞
π(2 n +1)k ⎞
*
⎟ cos⎛⎜
⎟.
∑ α[ k ] α[ m] G DCT [ k ]G DCT [ m] cos⎛⎜
2
N
2N

n=0 k =0 m =0

∑ ∑

Using the orthogonality property mentioned earlier we get
N −1

2

∑ g[ n] =

n=0

5.62 (a) HN

where I
H N H TN

2
1 N −1
∑ α[ k ] G DCT [ k ] .
2N k =0

⎡13
⎢17
=⎢
⎢13

⎣7
is the
⎡13
= ⎢17
⎢13
⎣⎢ 7

13
13
13 ⎤
− 7 − 17⎥⎥
7
. The matrix H N is orthogonal if H N H TN = cI
− 13 − 13 13 ⎥

− 17 17
−7⎦
4 × 4 identity matrix and c is a constant. Now,
13
13
13 ⎤ ⎡13 17
13
7 ⎤ ⎡676 0
0
0 ⎤
7
0 ⎥.
− 13 − 17⎥ = ⎢ 0 676 0
− 7 − 17⎥ ⎢13 7
0 676 0 ⎥
− 13 − 13 13 ⎥ ⎢13 − 7 − 13 17 ⎥ ⎢ 0
0
0 676⎦⎥
− 17 17
− 7 ⎦⎥ ⎣⎢13 − 17 13 − 7 ⎦⎥ ⎣⎢ 0

Hence, the matrix is orthogonal and all its rows have the same L 2 -norm.
1
1 ⎤
⎡1 1
⎢2 1 − 1 − 2 ⎥
⎥. Next, we observe
(b) G N = ⎢
⎢1 − 1 − 1 1 ⎥

⎣1 − 2 2 − 1 ⎦
1
1 ⎤ ⎡4 0 0 0 ⎤
1
1 ⎤ ⎡1 2
⎡1 1
1
1
1
2
1
1
2

⎥ ⎢0 10 0 0 ⎥
= 1 −1 −1 1 1 −1 −1 − 2
2⎥ = ⎢0 0 4 0 ⎥ which shows that

⎥⎢
⎢⎣1 − 2 2 − 1⎥⎦ ⎢⎣1 − 2 1 − 1⎥⎦ ⎢⎣0 0 0 10⎥⎦
the rows of G N are orthogonal but do not have the same L 2 -norms.
G N G TN

Not for sale.

131

1
1
1 ⎤ ⎡1 1 ⎤ = ⎡1
5.63 (a) H 2 = ⎡1 1 ⎤. Now, H t2 H 2 = ⎡1
2
2 ⎢⎣1 − 1⎦⎥ ⎢⎣1 − 1⎦⎥ ⎢⎣0
⎣⎢1 − 1⎥⎦
H2 ⎤
⎡H
. Thus,
H4 = ⎢ 2
⎣H 2 − H 2 ⎥⎦
⎡1
t
H t2 ⎤ ⎡H 2 H 2 ⎤ 1 ⎡4I 2 4I 2 ⎤ ⎢0
1 t
1 ⎡H 2
H H = ⎢ t
=
=
t ⎥ ⎢H
− H 2 ⎥⎦ 4 ⎢⎣4I 2 4I 2 ⎥⎦ ⎢0
4 4 4
4 ⎢H
⎣ 2 − H 2 ⎦⎥ ⎣ 2
⎢⎣0
1
N

H tN H N =

1
N

⎡H t
⎢ tN / 2
⎢⎣H N / 2

H tN / 2 ⎤ ⎡H N / 2

− H tN / 2 ⎥⎦ ⎢⎣H N / 2

0⎤ = I .
2
1 ⎥⎦

0
1
0
0

0
0
1
0

H N / 2 ⎤ 1 ⎡ NI N / 2
=
− H N / 2 ⎥⎦ N ⎢⎣ NI N / 2

0⎤
0⎥ = I .
4
0⎥
1⎥⎦
NI N / 2 ⎤
= IN .
NI N / 2 ⎥⎦

N −1
1
1
2
(b) From Eq. (5.172), ∑ x[ n] = x t x = ⎛⎜ H tN X Haar ⎞⎟ ⎛⎜ H tN X Haar ⎞⎟

⎠ ⎝N
⎝N
n=0
t

=

1
N2

X tHaar H N H tN X Haar =

1
N

X tHaar X Haar =

1 N −1
2
∑ X Haar [ k ]
N k =0

as from Eq.

(5.171) H N H tN + N .
N −1
2 πnk ⎞
2 πnk ⎞ ⎞
5.64 X DHT [ k ] = ∑ x[ n] ⎛⎜ cos⎛⎜
⎟ + sin⎛⎜
⎟ ⎟. Now,
⎝ N ⎠⎠
⎝ ⎝ N ⎠
n=0

2 πmk ⎞
2 πmk ⎞ ⎞
X DHT [ k ]⎛⎜ cos⎛⎜
⎟ + sin⎛⎜
⎟⎟
N
N ⎠⎠

N −1

2 πnk ⎞
2 πnk ⎞ ⎞⎛
2 πmk ⎞
2 πmk ⎞ ⎞
= ∑ x[ n] ⎛⎜ cos⎛⎜
⎟ + sin⎛⎜
⎟ ⎟⎜ cos⎛⎜
⎟ + sin⎛⎜
⎟ ⎟. Therefore,
⎝ N ⎠ ⎠⎝ ⎝ N ⎠
⎝ N ⎠⎠
⎝ ⎝ N ⎠
n=0
N −1

2 πmk ⎞
2 πmk ⎞ ⎞
⎟ + sin⎛⎜
⎟⎟
∑ X DHT [ k ]⎛⎜ cos⎛⎜
⎝ N ⎠⎠
⎝ ⎝ N ⎠
k =0
N −1
N −1
2 πnk ⎞
2 πnk ⎞ ⎞⎛
2 πmk ⎞
2 πmk ⎞ ⎞
= ∑ x[ n] ∑ ⎛⎜ cos⎛⎜
⎟ + sin⎛⎜
⎟ ⎟⎜ cos⎛⎜
⎟ + sin⎛⎜
⎟ ⎟.
⎝ N ⎠
⎝ N ⎠ ⎠⎝ ⎝ N ⎠
⎝ N ⎠⎠
n=0
k =0 ⎝

⎧ N , if m = n = 0,
⎪ N / 2, if m = n ≠ 0,
It can be shown that
=⎨
N / 2, if m = N − n,

⎩ 0, otherwise,
⎧ N , if m = n = 0,
N −1
π
π
2
nk
2
mk
⎞⎟ ⋅ sin⎛⎜
⎞⎟ = ⎪ N / 2, if m = n ≠ 0,
∑ sin⎛⎜

⎝ N ⎠
⎝ N ⎠ ⎪− N / 2, if m = N − n,
k =0
0, otherwise,

N −1

2 πnk ⎞
2 πmk ⎞
⎟ ⋅ cos⎛⎜

∑ cos⎛⎜
⎝ N ⎠
⎝ N ⎠
k =0

N −1

N −1

2 πnk ⎞
2 πmk ⎞
2 πnk ⎞
2 πmk ⎞
⎟ ⋅ cos⎛⎜
⎟ = ∑ cos⎛⎜
⎟ ⋅ sin⎛⎜
⎟ = 0.
∑ sin⎛⎜
⎝ N ⎠
⎝ N ⎠ k =0 ⎝ N ⎠
⎝ N ⎠
k =0

Hence, x[ m] =

1 N −1
2 πmk ⎞
2 πmk ⎞ ⎞
⎟ + sin⎛⎜
⎟ ⎟.
∑ X DHT [ n] ⎛⎜ cos⎛⎜
⎝ N ⎠⎠
⎝ ⎝ N ⎠
N n=0

Not for sale.

132

⎧ x[ n − no + N ], 0 ≤ n ≤ no − 1,
5.65 (a) y[ n] = x[〈 n − no 〉 N ] = ⎨
n o ≤ n ≤ N − 1.
⎩ x[ n − no ],
N −1
2 πnk ⎞
2 πnk ⎞ ⎞
YDHT [ k ] = ∑ y[ n] ⎛⎜ cos⎛⎜
⎟ + sin⎛⎜
⎟⎟
⎝ N ⎠⎠
⎝ ⎝ N ⎠
n=0
n −1

o
2 πnk ⎞
2 πnk ⎞ ⎞ N −1
2 πnk ⎞
2 πnk ⎞ ⎞
= ∑ x[ n − no + N ] ⎛⎜ cos⎛⎜
⎟ + sin⎛⎜
⎟ ⎟ + ∑ x[ n − no ] ⎛⎜ cos⎛⎜
⎟ + sin⎛⎜
⎟⎟ .
⎝ N ⎠⎠ n=n
⎝ N ⎠⎠
⎝ ⎝ N ⎠
⎝ ⎝ N ⎠
n=0
o

Replacing n − no + N by n in the first sum and n − no by n in the second sum we get

YDHT [ k ] =

N −1

2 π( n + n o )k ⎞
2 π( n + n o )k ⎞ ⎞
x[ n] ⎛⎜ cos⎛⎜
⎟ + sin⎛⎜
⎟⎟
N
N

⎠⎠
⎝ ⎝
n = N − no

n −1

o
2 π( n + n o ) k ⎞
2 π( n + n o ) k ⎞ ⎞
+ ∑ x[ n] ⎛⎜ cos⎛⎜
⎟ + sin⎛⎜
⎟⎟
N
N

⎠⎠
⎝ ⎝
n=0

N −1
2 π( n + n o ) k ⎞
2 π( n + n o ) k ⎞ ⎞
= ∑ x[ n] ⎛⎜ cos⎛⎜
⎟ + sin⎛⎜
⎟⎟
N
N

⎠⎠
⎝ ⎝
n=0
2 πn o k ⎞ N −1
2 πnk ⎞
2 πnk ⎞ ⎞

= cos⎛⎜
⎟ + sin⎛⎜
⎟⎟
⎟ ∑ x[ n] ⎜ cos⎛⎜
⎝ N ⎠⎠
⎝ N ⎠n=0
⎝ ⎝ N ⎠
2 πn o k ⎞ N −1
2 πnk ⎞
2 πnk ⎞ ⎞

+ sin⎛⎜
⎟ − sin⎛⎜
⎟⎟
⎟ ∑ x[ n] ⎜ cos⎛⎜
⎝ N ⎠⎠
⎝ N ⎠n=0
⎝ ⎝ N ⎠
2 πn o k ⎞
2 πn o k ⎞
= cos⎛⎜
[ −k ].
⎟ X DHT [ k ] + sin⎛⎜
⎟X
⎝ N ⎠
⎝ N ⎠ DHT

(b) The N –point DHT of x[〈− n〉 N ] is X DHT[ −k ] .
N −1

1 N-1 N −1
∑ ∑ X DHT [ k ]X DHT [l] ×
n=0
N 2 k =0 l=0
⎛ N −1⎛ ⎛ 2 πnk ⎞
2 πnk ⎞ ⎞⎛
2 πnl ⎞
2 πnl ⎞ ⎞ ⎞
⎜⎜ ∑ ⎜ cos⎜
⎟ + sin⎛⎜
⎟ ⎟⎜ cos⎛⎜
⎟ + sin⎛⎜
⎟ ⎟ ⎟.
⎝ N ⎠ ⎠⎝ ⎝ N ⎠
⎝ N ⎠ ⎠ ⎟⎠
⎝ n=0 ⎝ ⎝ N ⎠
Using the orthogonality property, we observe that the above product is equal to N if
N −1
1 N −1
2
k = l and is equal to zero if k ≠ l. Hence, ∑ x 2 [ n] =
∑ (X DHT [ k ]) .
N k =0
n=0
(c)

2
∑ x [ n] =

(

)

(

)

1
2 πnk ⎞ 1
2 πnk ⎞
5.66 cos⎛⎜
W − nk − WNnk . Therefore,
⎟ = W − nk + WNnk and sin⎛⎜
⎟=
⎝ N ⎠ 2 N
⎝ N ⎠ 2j N

X DHT [ k ] =

(

)

1 N −1
nk
nk
− nk
− nk
∑ x[ n] WN + WN − j WN + j WN
2 n=0
1
= X[ N − k ] + X[ k ] − j X [ N − k ] + j X[ k ] .
2

(

)

Not for sale.

133

N −1
⎛ ⎛ 2 πnk ⎞⎟ + sin⎛⎜ 2 πnk ⎞⎟ ⎞
5.67 y[ n] = x[ n] O
N g[ n]. Thus, YDHT [ k ] = ∑ y[ n] ⎜ cos⎜

⎝ N ⎠⎠
⎝ ⎝ N ⎠
n=0
N −1
N −1
2 πnk ⎞
2 πnk ⎞ ⎞
= ∑ x[r ] ∑ g[〈 n − r 〉 N ] ⎛⎜ cos⎛⎜
⎟ ⎟.
⎟ + sin⎛⎜
⎝ N ⎠⎠
⎝ ⎝ N ⎠
r =0
n=0

From the results of Problem 5.65 we have
N −1
2 πlk ⎞
2 πlk ⎞ ⎞
YDHT [ k ] = ∑ x[l]⎛⎜ G DHT [ k ] cos⎛⎜
⎟ + G DHT [〈− k 〉 N ] sin⎛⎜
⎟⎟
⎝ N ⎠
⎝ N ⎠⎠

l=0
N −1

N −1

2 πlk ⎞
2 πlk ⎞
= G DHT [ k ] ∑ x[l] cos⎛⎜
⎟ + G DHT [〈− k 〉 N ] ∑ x[l] sin⎛⎜

⎝ N ⎠
⎝ N ⎠
l=0
l =0

= G DHT [ k ] (X DHT [ k ] + X DHT [〈− k 〉 N ]) + G DHT [〈− k 〉 N ] (X DHT [ k ] − X DHT [〈− k 〉 N ])
=

1
2
1
[k]
X
2 DHT

(GDHT [k ] + GDHT [〈− k〉 N ]) +

1
2
1
[〈− k 〉 N ]
X
2 DHT

(GDHT [k ] − GDHT [〈− k〉 N ]) .

N −1

5.68 (a) y[ n] = ∑ (β1WN− nk + β 2WNnk ) X DCFT [ k ]
N −1

k =0
N −1

= ∑ x[ m] ∑ (α1β1WN( m − n)k + α 2 β 2WN− ( m + n)k + α1β 2WN( m + n)k + α1β 2WN− ( m − n)k )
m =0
N −1

k =0

= ∑ x[ m]N ((α1β1 + α 2 β 2 )δ[ m − n] + (α 2 β1 + α1β 2 )(δ[ m + n] + δ[ m + n − N ]))
m =0

= N [(α1β1 + α 2 β 2 ) x[ n] + (α 2 β1 + α1β 2 )(x[ n]δ[ n] + x[ N − n](1 − δ[ n])].
If we require
y[ n] = x[ n], 0 ≤ n ≤ N − 1,
then the following conditions must be satisfied:
α 2 β1 + α1β 2 = 0,
and
N (α1β1 + α 2 β 2 ) = 1.

(5-1)
(5-2)

(b) Let α12 − α 22 ≠ 0. Solving for β1 and β2 in Eqs. (5-1) and (5-2) we arrive at
α1
− α2
, β2 =
.
β1 =
N (α12 − α 22 )
N (α12 − α 22 )
Then, the inverse DCFT is given by
N −1
1
− nk
nk
x[ n] =
∑ (α1WN − α 2WN ) X DCFT [ k ], 0 ≤ n ≤ N − 1.
2
2
N (α 1 − α 2 ) k = 0
(c) If α1 = α *2 = α re + jα im with α re ≠ 0 and α im ≠ 0, the expression for
X DCFT [ k ] reduces to

Not for sale.

134

N −1
2 πnk ⎞
2 πnk ⎞ ⎞
X DCFT [ k ] = ∑ ⎛⎜ 2α re cos⎛⎜
⎟ + 2α im sin⎛⎜
⎟ ⎟x[ n]
⎝ N ⎠
⎝ N ⎠⎠
n=0 ⎝
which is real. The inverse DCFT is then given by
N −1
1
2 πnk ⎞
2 πnk ⎞ ⎞
x[ n] =
[ k ].
⎟ + 2α re sin⎛⎜
⎟ ⎟X
∑ ⎛⎜ 2α im cos⎛⎜
N
N ⎠ ⎠ DCFT

4 Nα1α 2 k = 0 ⎝

(d) It can be easily shown that the discrete Hartley transform (DHT) of Eq. (5.192) is a
1

.
special case of the real DCFT with α re = α im =
2N
1
1⎤
⎡1 1
1
1
1
1
1

5.69 (a) H 2 =
, H 4 = 1 1 − 1 − 11⎥, and
⎢⎣1 − 1⎥⎦

⎢⎣1 − 1 − 1 1 ⎥⎦
1
1
1 1
1
1 ⎤
⎡1 1
⎢1 − 1 1 − 1 1 − 1 1 − 1 ⎥
⎢1 1 − 1 − 1 1 1 − 1 − 1 ⎥
1 −1 −1 1
1 −1 −1 1 ⎥
H8 = ⎢
.
⎢1 1
1
1 − 1 − 1 − 1 − 1⎥
⎢1 − 1 1 − 1 − 1 1 − 1 1 ⎥
⎢1 1 − 1 − 1 − 1 − 1 1
1⎥
⎢⎣1 − 1 − 1 1 − 1 1
1 − 1⎥⎦
⎡H
(b) From the structure of H 2 , H 4 , and H 8 , it can be seen that H 4 = ⎢ 2
⎣H 2
H4 ⎤
⎡H
and H 8 = ⎢ 4
.
H

H 4 ⎥⎦
⎣ 4

H2 ⎤
,
− H 2 ⎥⎦

(c) X HT = H N x. Therefore, x = H −N1X HT = N ⋅ H TN X HT = N ⋅ H *N X HT . Hence,
N −1

x[ n] = ∑ X HT
k =0

l −1

∑ bi ( n ) bi ( k )

[ k ](−1) i =0

, where bi (r ) is the i –th bit in the binary

representation of r.
M5.1 (a) N = input('The value of N = ');
k = -N:N;
y = ones(1,2*N+1);
w = 0:2*pi/255:2*pi;
Y = freqz(y, 1, w);
Ydft = fft(y);
n = 0:1:2*N;
plot(w/pi,abs(Y),n*2/(2*N+1),abs(Ydft),'o');
xlabel('\omega/\pi'),ylabel('Amplitude');

Not for sale.

135

(b) Replace the statement k = -N:N; with k = 0:N;, y = ones(1,2*N+1);
with y = ones(1,N+1); the statement n = 0:1:2*N; with n = 0:N;, and
the statement plot(w/pi,abs(Y),n*2/(2*N+1),abs(Ydft),'o');
with plot(w/pi,abs(Y),n*2/(N+1),abs(Ydft),'o');,in the program of
Part (a).
(c) Add the statement y = y – abs(k)/N; below the statement y =
ones([1,2*N+1]); in the program of Part (a).
(d) Replace the statement y = ones(1,2*N+1); with y = N +
ones(1,2*N+1) – abs(k); in the program of Part (a).
(e) Replace the statement y = ones(1,2*N+1); with y = cos(pi*k/(2*N));
in the program of Part (a).

The plots generated for N = 4 are shown below where the circles denote the DFT
samples.
(a)
(b)
5

10

4
Amplitude

Amplitude

8
6
4

2
1

2
0

3

0
0

0.5

1
ω/ π

1.5

2

0

0.5

1

1.5

2

ω/ π

(d)
4

25

3

20
Amplitude

Amplitude

(c)

2
1
0

15
10
5

0

0.5

1
ω/π

1.5

2

0

0

0.5

1

1.5

2

ω/π

(e)

Not for sale.

136

6

Amplitude

5
4
3
2
1
0

0

0.5

1

1.5

2

ω/ π

M5.2 The code fragments to be used are as follows:

Y = fft(g).*fft(h);
y = iift(Y);
(a) y[ n] = g[ n] O
6 h[ n]. The output generated using the above code fragments is
y =
-6
9
-16
20
-4
45
5 v[ n]. The output generated using the above code fragments is
(b) w[ n] = x[ n] O
w =
Columns 1 through 3
11.0000 +25.0000i -9.0000 +48.0000i

3.0000 +17.0000i

Columns 4 through 5
29.0000 + 0.0000i -10.0000 +12.0000i
5 y[ n]. The output generated using the above code fragments is
(c) u[ n] = x[ n] O

u =
-23.0000

-69.0000

35.0000

105.0000

73.0000

M5.3 N = 8; % sequence length
gamma = 0.5;
k = 0:N-1;
x = exp(-gamma*k);
X = fft(x);
% Property 1
X1 = fft(conj(x));
G1 = conj([X(1) X(N:-1:2)]);
% Verify X1 = G1
% Property 2
x2 = conj([x(1) x(N:-1:2)]);
X2 = fft(x2);

Not for sale.

137

% Verify X2 = conj(X)
% Property 3
x3 = real(x);
X3 = fft(x3);
G3 = 0.5*(X+conj([X(1) X(N:-1:2)]));
% Verify X3 = G3
% Property 4
x4 = j*imag(x);
X4 = fft(x4);
G4 = 0.5*(X-conj([X(1) X(N:-1:2)]));
% Verify X4 = G4
% Property 5
x5 = 0.5*(x+conj([x(1) x(N:-1:2)]));
X5 = fft(x5);
% Verify X5 = real(X)
% Property 6
x6 = 0.5*(x-conj([x(1) x(N:-1:2)]));
X6 = fft(x6);
% Verify X6 = j*imag(X)
M5.4 N = 8;
k = 0:N-1;
gamma = 0.5;
x = exp(-gamma*k);
X = fft(x);
% Property 1
xpe = 0.5*(x+[x(1) x(N:-1:2)]);
xpo = 0.5*(x-[x(1) x(N:-1:2)]);
Xpe = fft(xpe);
Xpo = fft(xpo);
% Verify Xpe = real(X) and Xpo = j*imag(X)
% Property 2
X2 = [X(1) X(N:-1:2)];
% Verify X = conj(X2);
% real(X) = real(X2) and imag(X) = -imag(X2)
% abs(X)= abs(X2) and angle(X) = -angle(X2)
M5.5 N = 8; % N is length of the sequence(s)
gamma = 0.5;
k = 0:N-1;
g = exp(-gamma*k); h = cos(pi*k/N);
G = fft(g); H=fft(h);
% Property 1

Not for sale.

138

alpha=0.5; beta=0.25;
x1 = alpha*g+beta*h;
X1 = fft(x1);
% Verify X1=alpha*G+beta*H
% Property 2
n0 = N/2; % n0 is the amount of shift
x2 = [g(n0+1:N) g(1:n0)];
X2 = fft(x2);
% Verify X2(k)= exp(-j*k*n0)G(k)
% Property 3
k0 = N/2;
x3 = exp(-j*2*pi*k0*k/N).*g;
X3 = fft(x3);
G3 = [G(k0+1:N) G(1:k0)];
% Verify X3=G3
% Property 4
x4 = G;
X4 = fft(G);
G4 = N*[g(1) g(8:-1:2)]; % This forms N*(g mod(-k))
% Verify X4 = G4;
% Property 5
% To calculate circular convolution between
% g and h use eqn (3.67)
h1 = [h(1) h(N:-1:2)];
T = toeplitz(h',h1);
x5 = T*g';
X5 = fft(x5');
% Verify X5 = G.*H
% Property 6
x6 = g.*h;
X6 = fft(x6);
H1 = [H(1) H(N:-1:2)];
T = toeplitz(H.', H1); % .' is the nonconjugate transpose
G6 = (1/N)*T*G.';
% Verify G6 = X6.'
M5.6 g = input('Type in first sequence = ');
h = input('Type in second sequence = ');
x = g +i*h;
XF = fft(x);
XFconj = conj(XF);
N = length(g);
YF = zeros(N-1,1)';
for k = 1:N-1;
YF(k) = XFconj(mod(-k,N)+1);

Not for sale.

139

end
YF1 = [XFconj(1) YF];
GF = (XF + YF1)/2;
HF = (XF - YF1)/2;
M5.7 x = [-3 5 45 -15 -9 -19 -8 21 -10 23];
XF = fft(x);
k =0:9; YF = exp(-i*2*pi*k/5).*XF;
output = [XF(1) XF(6) sum(XF) sum(YF)];;
disp(output)
disp(sum(abs(XF).*abs(XF))

The output data generated by this program is
30
0
-30-0.0000i -100-0.0000i
38600

M5.8 X = [11 8-i*2 1-i*12 6+i*3 -3+i*2 2+i
k = 8:12; XF(k)=conj(X(mod(-k+2,12)));
XF = [X XF(8:12)];
x = ifft(XF);
n = 0:11; y = exp(i*2*pi*n/3).*x;
output = [x(1) x(7) sum(x) sum(y)];
disp(output)
disp(x.*x)

The output data generated by this program is
4.5000
-0.8333
11.0000
74.8333

15];

-3.0000-2.0000i

M5.9 n=0:255;
x = 0.1*n.*exp(-0.03*n);
plot(n,x);axis([0 255 0 1.3]);
xlabel('n');ylabel('Amplitude');
title('Original signal');
pause
z = [zeros(1,50) ones(1,156) zeros(1,50)];
y = 4*rand(1,256)-1;
YF = z.*fft(y);
yinv = ifft(YF);
s = x + yinv;
plot(n,s);axis([0 255 -2 4]);
xlabel('n');ylabel('Amplitude');
title('Noise corrupted signal');
pause
zc = [ones(1,50) zeros(1,156) ones(1,50)];
SF = zc.*fft(s);
xr = ifft(SF);
plot(n,xr);axis([0 255 0 1.3]);
xlabel('n');ylabel('Amplitude');

Not for sale.

140

title('Signal after Fourier-domain filtering');
Original signal

Noise corrupted signal
4

1.2
3
Amplitude

Amplitude

1
0.8
0.6
0.4

1
0
-1

0.2
0

2

0

50

100

150

200

-2

250

0

50

100

150

200

250

Signal after Fourier-domain filtering
1.2

Amplitude

1
0.8
0.6
0.4
0.2
0

M5.10

0

50

100

150

200

250

function y = overlapsave(x,h)
X = length(x);
%Length of longer sequence
M = length(h);
%length of shorter sequence
flops(0);
if (M > X)
%Error condition
disp('error');
end
%clear all
temp = ceil(log2(M)); %Find length of circular
convolution
N = 2^temp;
if(N > M)
for i = M+1:N
h(i) = 0;
end
end
m = ceil((-N/(N-M+1)));
while (m*(N-M+1) <= X)
if(((N+m*(N-M+1)) <= X)&((m*(N-M+1)) > 0))
for n = 1:N
x1(n) = x(n+m*(N-M+1));
end
end
if(((m*(N-M+1))<=0)&((N+m*(N-M+1))>=0))

Not for sale.

141

for n = 1:N
x1(n) = 0;
end
for n = m*(N-M+1):N+m*(N-M+1)
if(n > 0)
x1(n-m*(N-M+1)) = x(n);
end
end
end
for n = 1:N
x1(n) = 0;
end
for n = 1:(X-m*(N-M+1))
x1(n) =x (m*(N-M+1)+n);
end
end
w1 = circonv1(h,x1); %circular convolution using DFT
for i = 1:M-1
y1(i) = 0;
end
for i = M:N
y1(i) = w1(i);
end
for j = M:N
if((j+m*(N-M+1)) < (X+M))
if((j+m*(N-M+1)) > 0)
yO(j+m*(N-M+1)) = y1(j);
end
end
end
m = m+1;
end
%disp('Convolution using Overlap Save:');
y = real(yO);
function y = circonv1(x1,x2)
L1 = length(x1); L2 = length(x2);
if L1 ~= L2,
error('Sequences of unequal lengths'),
end
X1 = fft(x1);
X2 = fft(x2);
X_RES = X1.*X2;
y = ifft(X_RES);

The MATLAB program for performing convolution using the overlap-save method is
h = [1 1
R = 50;

1]/3;

Not for sale.

142

d = rand(1,R) - 0.5;
m = 0:1:R-1;
s = 2*m.*(0.9.^m);
x = s + d;
%x = [x x x x x x x];
y = overlapsave(x,h);
k = 0:R-1;
plot(k,x,'r-',k,y(1:R),'b--');
xlabel('Time index n');ylabel('Amplitude');
legend('r-', 's[n]','b--','y[n]');

The output plot generated by the above program is shown below:
8
s[n]
y[n]

Amplitude

6
4
2
0

0

10

20
30
Time index n

Not for sale.

40

50

143

Chapter 6

6.1 X ( z ) = ∑ x[ n] z − n = ∑ x[ n] z − n . Therefore, lim X ( z ) = lim ∑ x[ n] z − n
n = −∞

z→∞

n=0

z →∞ n=0

= lim x[0] + lim ∑ x[ n] z − n = x[0].
z→∞

z → ∞ n =1

6.2

(a) Z δ{[ n]} = ∑ δ[ n] z − n = δ[0] = 1, which converges everywhere in the z –plane.
n = −∞

(b) x[ n] = α n µ[ n]. From Table 6.1,

1

Z {x[ n]} = X ( z ) = ∑ x[ n]z − n =

, z > α . Let g[ n] = nx[ n]. Then,
1 − αz − 1

dX ( z )
= − ∑ ng[ n]z − n −1 . Hence,
Z {g[ n]} = G( z ) = ∑ nx[ n]z − n . Now,
dz
n = −∞
n = −∞
n = −∞

z

dX ( z )
dX ( z )
αz − 1
= − ∑ nx[ n]z − n = −G( z ), or, G ( z ) = − z
=
, z > α.
dz
dz
n = −∞
(1 − αz −1 ) 2

r n jω o n
(e
− e − jωo n ) µ[ n]. Using the results of Example
2j
6.1 and the linearity property of the z –transform we get

⎞ 1 ⎛
1 ⎛
1
1

⎟− ⎜
Z {r n sin(ωo n) µ[ n]} = ⎜
2 j ⎜ 1− r e jωo z −1 ⎟ 2 j ⎜ 1− r e − jωo z −1 ⎟

(c) x[ n] = r n sin(ωo n) µ[ n] =

=

6.3

r
( e jω o − e − jω o ) z − 1
2j
jω o
− jω o − 1
2 −2

1 − r (e

+e

)z

+r z

=

r sin(ωo )z −1
1 − 2r cos(ωo )z

−1

2 −2

+r z

,

∀ z > r.

(a) x1[ n] = α n µ[ n − 2]. Note, x1[ n] is a right-sided sequence. Hence, the ROC of its

z –transform is exterior to a circle. Therefore, X1 ( z ) = ∑ α n µ[ n − 2]z − n = ∑ α n z − n
Simplifying we get X1 ( z) =

1
1 − αz

−1

− 1 − αz − 1 =

n = −∞
2 −2

α z

1 − αz − 1

n=2

whose ROC is given by

z > α.
(b) x 2 [ n] = −α n µ[ − n − 3]. Note, x2 [ n] is a left-sided sequence. Hence, the ROC of
its z –transform is interior to a circle. Therefore,

−3

n = −∞

n = −∞

m =3

m=3

X 2 ( z) = − ∑ α n µ[ − n − 3]z − n = − ∑ α n z − n = − ∑ α − m z m = − ∑ ( z / α ) m

Not for sale.

147

Simplifying we get X 2 ( z ) =

(z / α )3
whose ROC is given by z < α .
1 − (z / α )

(c) x 3 [ n] = α n µ[ n + 4]. Note, x3[ n] is a right-sided sequence. Hence, the ROC of its

n = −∞

n = −4

z –transform is exterior to a circle. Therefore, X 3 ( z) = ∑ α n µ[ n + 4]z − n = ∑ α n z − n
Simplifying we get X 3 ( z ) =

(α / z )
whose ROC is given by z > α .
1 − (α / z )
−4

(d) x 4 [ n] = α n µ[ − n]. Note, x4 [ n] is a left-sided sequence. Hence, the ROC of its

0

n = −∞

n = −∞

z –transform is interior to a circle. Therefore, X 4 ( z) = ∑ α n µ[ − n]z − n = ∑ α n z − n

1
, z / α < 1. Therefore the ROC of X 4 ( z) is given by z < α .
1 − (z / α)

= ∑ α−mzm =
m=0

6.4

Z {(0.4) n µ[ n]} =

1
1 − 0 .4 z − 1

, z > 0.4; Z {( −0.6) n µ[ n]} =

1
1 + 0 .6 z − 1

, z > 0.6;

1

Z {(0.4) n µ[ − n − 1]} = −

, z < 0.4;
1 − 0. 4 z − 1
1
Z {( −0.6) n µ[ − n − 1]} = −
, z < 0.6;
1 + 0 .6 z − 1

(a) Z {x1[ n]} =

1
1 − 0 .4 z − 1

(b) Z {x 2 [ n]} =
(c) Z {x 3 [ n]} =
(d) Z {x 4 [ n]} =

+

1
1 − 0 .4 z − 1
1
1 − 0 .4 z − 1
1
1 − 0 .4 z − 1

1
1 + 0 .6 z − 1

+
+
+

=

1
1 + 0 .6 z − 1
1
1 + 0 .6 z − 1
1
1 + 0 .6 z − 1

1 + 0 .2 z − 1
(1 − 0.4 z −1 )(1 + 0.6 z −1 )

=
=
=

, z > 0 .6 .

1 + 0 .2 z − 1
(1 − 0.4 z −1 )(1 + 0.6 z −1 )
1 + 0 .2 z − 1
(1 − 0.4 z −1 )(1 + 0.6 z −1 )

, 0 .4 < z < 0 .6 .

, z < 0.4.

1 + 0 .2 z − 1
(1 − 0.4 z −1 )(1 + 0.6 z −1 )

. Since the ROC of

the first term is z < 0.4 and that of the second term is z > 0.6 . Hence, the z –
transform of x 4 [ n] does not converge.
6.5

(a) The ROC of Z {x1[ n]} is z > 0.3, the ROC of Z {x 2 [ n]} is z > 0.7, the ROC of
Z {x 3 [ n]} is z > 0.4, and the ROC of Z {x 4 [ n]} is z < 0.4.
(b) (i) The ROC of Z {y1[ n]} is z > 0.7,
(ii) The ROC of Z {y2 [ n]} is z > 0.4,

Not for sale.

148

(iii) The ROC of Z {y3 [ n]} is 0.3 < z < 0.4,
(iv) The ROC of Z {y 4 [ n]} is z > 0.7,
(v) The ROC of Z {x 2 [ n]} is z > 0.7, whereas, the ROC of Z {x 4 [ n]} is z < 0.4.
Hence, the z -transform of y5 [ n] does not converge.
(vi) The ROC of Z {x 3 [ n]} is z > 0.4, whereas, the ROC of Z {x 4 [ n]} is z < 0.4.
Hence, the z -transform of y6 [ n] does not converge.
6.6

v[ n] = α

n

= α n µ[ n] + α − n µ[ − n − 1]. Now, Z {α n µ[ n]} =

1
1 − αz − 1

−1

n = −∞

m =1

m=0

, z > α . (See Table

6.1) and Z {α − n µ[ − n − 1]} = ∑ α − n z − n = ∑ α m z m = ∑ α m z m − 1 =

1
−1
1 − αz

αz
1
α
(1 − α 2 )z −1
, αz < 1. Therefore, {v[ n]} = V ( z ) =
+
=
1 − αz
1 − αz −1 z −1 − α (1 − αz −1 )( z −1 − α )
with its ROC given by α < z < 1 / α .
=

6.7

(a) x1[ n] = α n µ[ n + 2] + β n µ[ n + 2] with β > α . Note that x1[ n] is a right-sided
sequence. Hence, the ROC of its z –transform is exterior to a circle. Now,

n = −2

n=0

Z {α n µ[ n + 2]} = ∑ α n z − n = ∑ α n z − n + (α / z) −1 + (α / z) − 2
=

1
(α / z ) −2
with its ROC given by z > α . Likewise,
+ ( α / z ) − 1 + (α / z ) − 2 =
1 − (α / z )
1 − (α / z )

Z {β n µ[ n + 2]} =

( β / z ) −2
with its ROC given by z > β . Hence,
1 − ( β / z)

Z {x1[ n]} = X1 ( z ) =
=

(α / z ) − 2 ( β / z ) − 2
+
1 − (α / z ) 1 − ( β / z )

(α −2 + β −2 ) − (αβ −2 + α −2 β )z −1
z − 2 (1 − αz −1 )(1 − βz −1 )

with its ROC given by z > β .

(b) x 2 [ n] = α n µ[ − n − 2] + β n µ[ n − 1] with β > α . Note that x2 [ n] is a two-sided
sequence. Now,
−2

n = −∞

m=2

m=0

Z {α n µ[ − n − 2]} = ∑ α n z − n = ∑ α − m z m = ∑ ( z / α ) m − 1 − ( z / α ) − ( z / α ) 2
=

(z / α)3
with its ROC given by z < α . Likewise,
1 − (z / α)

Not for sale.

149

1

n =1

n=0

1 − βz − 1

Z {β n µ[ n − 1]} = ∑ β n z − n = ∑ β n z − n − 1 =

−1 =

βz − 1
1 − βz − 1

with its ROC

given by z > β . Since the two ROCs do not intersect, Z {x 2 [ n]} does not converge.
(c) x 3 [ n] = α n µ[ n + 1] + β n µ[ − n − 2] with β > α . Note that x3[ n] is a two-sided
sequence. Now,
Z {β n µ[− n − 2]} =

=

−2

n = −∞

m=2

m=0

∑ β n z − n = ∑ β − m z m = ∑ ( z / β )m − 1 − ( z / β ) − ( z / β )2

(z / β )
with its ROC given by z < β . Likewise,
1− (z / β )
3

n =1

n =0

Z {α n µ[n − 1]} = ∑ α n z − n = ∑ α n z − n − 1 =

1
1 − αz −1

−1 =

αz −1
with its ROC given by
1 − αz −1

z > α.
6.8

The denominator factor ( z 2 + 0.3z − 0.18) = ( z + 0.6)( z − 0.3) has poles at z = −0.6 and
at z = 0.3, and the factor ( z 2 − 2 z + 4) has poles with a magnitude 2. Hence, the four
ROCs are defined by the regions: R 1 : 0 < z < 0.3, R 2 : 0.3 < z < 0.6,
R 3 = 0.6 < z < 2, and R 4 : z > 2. The inverse z –transform associated with the ROC
R 1 is a left-sided sequence, the inverse z –transforms associated with the ROCs
R 2 and R 3 are two-sided sequences, and the inverse z –transform associated with the
ROC R 4 is a right-sided sequence.

6.9

X (z ) = Z {x[ n]} with an ROC given by R x . Using the conjugation property of the –
transform given in Table 6.2, we observe that Z {x * [ n]} = X * ( z*) whose ROC is
1
2

given by R x . Now, Re{x[ n]} = ( x[ n] + x * [ n]). Hence, Z {Re x[ n]}

=

1
2

(X ( z) + X * ( z)) whose ROC is also R x .

Thus, Z {Im x[ n]} =
6.10

1
2j

Likewise, Im{x[ n]} =

1
( x[ n] −
2j

x * [ n]).

(X (z) − X * (z)) whose ROC is again R x .

{x[ n]} = {2, 3, − 1, 0, − 4, 3, 1, 2, 4}, − 2 ≤ n ≤ 6. Then,
~
~
X[ k ] = X ( z) z = e jπk / 3 = X ( z) z = e j 2πk / 6 = X (e jω )
. Note that X [ k ] is a
ω = 2πk / 6
periodic sequence with a period 6. Hence, from Eq. (5.49), the inverse of the discrete

t
~
Fourier series X [ k ] is given by x[ n] = ∑ x[ n + 6r ] = x[ n − 6] + x[ n] + x[ n + 6] for
r = −∞

0 ≤ n ≤ 5. Let y[ n] = x[ n − 6] + x[ n] + x[ n + 6], − 2 ≤ n ≤ 6. Now,

Not for sale.

150

{x[ n − 6]} = {0, 0, 0, 0, 0, 0, 2, 3, − 1} and
{x[ n + 6] = {1, 2, 4, 0, 0, 0, 0, 0, 0}. Therefore,
{y[ n]} = {3, 5, 3, 0, − 4, 3, 3, 5, 3}, − 2 ≤ n ≤ 6. Hence,
{~
x[ n]} = {3, 0, − 4, 3, 3, 5}, 0 ≤ n ≤ 5.
6.11

{x[ n]} = {4, 2, − 1, 5, − 3, 1, − 2, 4, 2}, − 6 ≤ n ≤ 2. Then,
~
~
X[ k ] = X ( z) z = e jπk / 3 = X ( z) z = e j 2πk / 6 = X (e jω )
. Note that X [ k ] is a
ω = 2πk / 6

periodic sequence with a period 6. Hence, from Eq. (5.49), the inverse of the discrete

t
~
Fourier series X [ k ] is given by x[ n] = ∑ x[ n + 6r ] = x[ n − 6] + x[ n] + x[ n + 6] for
r = −∞

0 ≤ n ≤ 5. Let y[ n] = x[ n − 6] + x[ n] + x[ n + 6], − 6 ≤ n ≤ 5. Now,
{x[ n − 6]} = {0, 0, 0, 0, 0, 0, 4, 2, − 1, 5, − 3, 1}, − 6 ≤ n ≤ 5,
{x[ n + 6] = {− 2, 4, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0}, − 6 ≤ n ≤ 5.
Therefore, {y[ n]} = {2, 6, 1, 5, − 3, 1, 2, 6, 5, − 3, 1}, − 6 ≤ n ≤ 5.
x[ n]} = {2, 6, 1, 5, − 3, 1}, 0 ≤ n ≤ 5.
Hence, {~
11

6.12

11

X ( z ) = ∑ x[ n] z − n . X 0 [ k ] = X ( z) z = e (2πk / 9 = ∑ x[ n] e − j 2πkn / 9 , 0 ≤ k ≤ 8.
n=0

Therefore, x 0 [ n] =

=

1 8
j 2πkn / 9
∑ X [k ] e
9 k =0 0

1 8 11
j 2 πk ( n − r ) / 9
∑ ∑ x[r ] e
9 k =0 r =0

From Eq. (5.11),

=

=

n=0
1 8 ⎛ 11
− j 2 πkr / 9 ⎞⎟ j 2 πkn / 9
∑ ⎜⎜ ∑ x[r ] e
⎟e
9 k =0 r =0

9
1 11
j 2 πk ( n − r ) / 9
∑ x[r ] ∑ e
9 r =0
k =0

1 8
− (r − n ) k
∑ W9
9 k −0

=

9
1 11
− (r − n ) k
.
∑ x[r ] ∑ W9
9 r =0
k =0

⎧1, for r − n = 9i,
Hence,
=⎨
otherwise.
⎩0,

⎧ x[0] + x[9], for n = 0
i.e.,
x 0 [ n] = ⎨
otherwise,
⎩ x[ n],
{x 0 [ n]} = {20 5 45 − 15 − 9 − 19 − 8 21 − 10}, 0 ≤ n ≤ 8.
6.13

n = −∞

n = −∞

r = −∞

(a) X ( z) = ∑ x[ n] z − n . Hence, X ( z 3 ) = ∑ x[ n] z − 3n = ∑ x[ m / 3] z − m . Define
m =3r

⎧ x[ m / 3], m = 0, ± 3, ± 6,K,
a new sequence g[ m] = ⎨
We can then express
otherwise.
⎩ 0,

Not for sale.

151

X ( z 3 ) = ∑ g[ n] z − n . Thus, the inverse z –transform of X ( z 3 ) is given by g[n].
n = −∞

⎧⎪(−0.5) n / 3 , n = 0, 3, 6,K,
For x[ n] = (−0.5) n µ[ n], g[ n] = ⎨
⎪⎩
0,
otherwise.

(b) Y ( z) = (1 + z −1 ) X ( z 3 ) = X ( z 3 ) + z −1 X ( z 3 ). Therefore,

y[n] = Z −1{Y ( z)} = Z −1 X ( z 3 ) + Z −1 z −1 X ( z 3 ) = g[ n] + g[ n − 1], where g[n] =
⎧⎪(−0.5) n / 3 , n = 0, 3, 6,K,
Hence,
Z −1 X ( z 3 ) . From Part (a), g[ n] = ⎨
⎪⎩
0,
otherwise.
⎪⎧(−0.5) ( n −1) / 3 , n = 1, 4, 7,K,
Therefore,
g[ n − 1] = ⎨
⎪⎩
0,
otherwise.
⎧ (−0.5) n / 3 ,
n = 0, 3, 6,K,
⎪⎪
y[ n] = ⎨(−0.5) ( n −1) / 3 , n = 1, 4, 7,K,

0,
otherwise.
⎪⎩
6.14

(a) X a (z) = Z {µ[ n] − µ[ n − 5]} =

1
1 − z −1

z −5
1 − z −1

=

1 − z −5
1 − z −1

= 1 + z −1 + z −2 + z −3 + z −4 . Since has all poles at the origin, the ROC is the entire
z –plane except the point z = 0 , and hence includes the unit circle. On the unit circle,

X a ( z ) z = e j ω = X a ( e j ω ) = 1 + e − j ω + e − j 2 ω + e − j 3ω + e − j 4 ω =

1 − e − j 5ω
1 − e − jω

.

(b) x b [ n] = α n µ[ n] − α n µ[ n − 8], α < 1. From Table 6.1,

z −8

1

X b (z) =

1 − z −8

. The ROC is exterior to the circle at

=
1 − αz − 1 1 − α z − 1 1 − αz − 1
z = α < 1. Hence, the ROC includes the unit circle. On the unit circle,

X b ( z) z = e jω = X b (e jω ) =

1 − e j 8ω
1 − α e jω

.

(c) x c [ n] = (n + 1)α n µ[ n] = nα n µ[ n] + α n µ[ n], α < 1. From Table 6.1,

X c ( z) =

αz − 1

1

1 + αz − 1

. The ROC is exterior to the circle at
+
=
1 − αz − 1 1 − αz − 1 1 − α z − 1
z = α < 1. Hence, the ROC includes the unit circle. On the unit circle,

X c (e jω ) =

1 + αe jω
1− e

.

Not for sale.

152

N
⎞ 1 − z −(2 N +1)
⎛ 2N
(a) Y 1( z ) = ∑ z − n = z N ⎜⎜ ∑ z − n ⎟⎟ =
. Y1 ( z) has N poles at
n=−N
⎠ z − N (1 − z −1 )
⎝ n=0
z = 0 and N poles at z = ∞. Hence, the ROC is the entire z –plane excluding the
points z = 0 and z = ∞, and includes the unit circle. On the unit circle,

6.15

Y1 ( z) z = e jω

1
sin⎛⎜ ω( N + ) ⎞⎟
2 ⎠

.
= Y1 (e jω ) =
=
sin(ω / 2)
e − jNω (1 − e − jω )

1 − e j (2 N + 1)ω

N

(b) Y 2( z) = ∑ z − n =

1 − z −( N + 1)

. Y 2( z) has N poles at z = 0. Hence, the ROC
1 − z −1
is the entire z –plane excluding the point z = 0. On the unit circle,
N +1 ⎞
ω⎟
sin⎛⎜
j ( N + 1)ω
1− e
2

⎠.

− Nω / 2
Y2 ( z ) z = e jω = Y2 (e ) =
=e
− jω
sin(ω / 2)
1− e
n
⎧⎪
(c) y3 [ n] = ⎨1 − N , − N ≤ n ≤ N , Now, y3 [ n] = y0 [ n] O
* y 0 [ n] where
⎪⎩ 0,
otherwise.
⎧⎪1, − N ≤ n ≤ N ,
(1 − z −( N + 1) ) 2
2
y 0 [ n] = ⎨
.
2
2 Therefore, Y3 ( z ) = Y0 ( z ) =
⎪⎩0,
z − N (1 − z −1 ) 2
otherwise.
n=0

Y3 ( z ) has

N
N
poles at z = 0 and
poles at z = ∞. Hence, the ROC is the entire z –
2
2

plane excluding the points z = 0 and z = ∞, and includes the unit circle. On the unit
N +1 ⎞ ⎞
sin 2 ⎛⎜ ω⎛⎜
⎟⎟
⎝ ⎝ 2 ⎠⎠ .
circle, Y3 (e jω ) = Y02 (e jω ) =
sin 2 (ω / 2)

⎧N + 1 − n , − N ≤ n ≤ N ,
= y1 [ n] + N ⋅ y 3 [ n], where y1[ n] is the
(d) y 4 [ n] = ⎨
0,
otherwise,

sequence of Part (a) and y 3 [ n] is the sequence of Part (c). Therefore,
Y4 ( z ) = Y1 ( z ) + N ⋅ Y3 ( z ) =

1 − z −(2 N + 1)

+N

(1 − z −( N +1) )2

. Since the ROCs of
z − N (1 − z −1 )2
z − N (1 − z −1 )
both Y1( z) and Y3 ( z ) include the unit circle, the ROC of Y4 ( z) also includes the unit
N +1 ⎞ ⎞
1
⎟⎟
sin⎛⎜ ω( N + ) ⎞⎟ sin 2 ⎛⎜ ω⎛⎜
2 ⎠
⎝ 2 ⎠⎠

+
.
circle. On the unit circle, Y4 (e ) =
sin(ω / 2)
sin 2 (ω / 2)

⎧cos(πn / 2 N ), − N ≤ n ≤ N ,
Therefore,
(e) y f [ n] = ⎨
0,
otherwise.

Not for sale.

153

1 N − j (πn / 2 N ) − n 1 N j (πn / 2 N ) − n
z +
z
∑e
∑e
2 n=−N
2 n=−N
e j (π / 2) z N ⎛⎜ 1 − e − j (2 N + 1)(π / 2 N ) z − (2 N + 1) ⎞⎟
=

2
1 − e − j (π / 2 N ) z −1

e − j (π / 2) z N ⎛⎜ 1 − e j (2 N +1)(π / 2 N ) z − (2 N +1) ⎞⎟
+
.
j (π / 2 N ) −1

2
e
z

1

Y f (z) has N poles at z = 0 and N poles at z = ∞. Hence, the ROC is the entire z –
Y f ( z) =

plane excluding the points z = 0 and z = ∞, and includes the unit circle. On the unit
π ⎞⎛
π ⎞⎛
1 ⎞
1 ⎞
sin⎛⎜ ⎛⎜ ω −
sin⎛⎜ ⎛⎜ ω +
⎟⎜ N + ⎞⎟ ⎟
⎟⎜ N + ⎞⎟ ⎟
2
N
2
2
N
2 ⎠⎠
1
1

⎠+

circle, Y f (e jω ) =
.
π ⎞ ⎞
π ⎞ ⎞
2
2

sin⎜ ⎜ ω −
sin⎜ ⎜ ω +
⎟ / 2⎟
⎟ / 2⎟
2N ⎠ ⎠
2N ⎠ ⎠
⎝⎝
⎝⎝
6.16

X ( z ) = −4 z 3 + 5 z 2 + z − 2 − 3 z − 1 + 2 z −3 , Y ( z ) = 6 z − 3 − z − 1 + 8 z − 3 + 7 z −4 − 2 z − 5 ,
W ( z ) = 3 z − 2 + 2 z − 3 + 2 z − 4 − z −5 − 2 z − 7 + 5 z −8 .
(a) U ( z ) = X ( z )Y ( z )
= (−4 z 3 + 5z 2 + z − 2 − 3z −1 + 2 z −3 )(6 z − 3 − z −1 + 8 z −3 + 7z −4 − 2 z −5 )
= −24 z 4 + 42 z 3 − 5z 2 − 20 z − 45 + 23z −1 + 66 z −2 − 25z −3 − 42 z −4 − 17 z −5
+ 22 z −6 + 14 z −7 − 4 z −8 . Hence,
{u[ n]} = {− 24, 42, − 5, − 20, − 45, 23, 66, − 25, − 42, − 17, 22, 14, − 4}, − 4 ≤ n ≤ 8.

(b) V ( z ) = X ( z )W ( z )

= (−4 z 3 + 5z 2 + z − 2 − 3z −1 + 2 z −3 )(3z −2 + 2 z −3 + 2 z −4 − z −5 − 2 z −7 + 5z −8 )
= −12 z + 7 + 5z −1 + 10 z −2 − 16 z −3 − 3z −4 − 28 z −5 + 30 z −6 + 13z −7 − 6 z −8
− 15 z −4 − 4 z −10 + 10 z −11 . Hence,
{v[ n]} = {− 12, 7, 5, 10, − 16, − 3, − 28, 30, 13, − 6, − 15, − 4, 10}, − 1 ≤ n ≤ 11.

(c) G ( z ) = W ( z )Y ( z )

= (3z −2 + 2 z −3 + 2 z −4 − z −5 − 2 z −7 + 5z −8 )(6 z − 3 − z −1 + 8 z −3 + 7z −4 − 2 z −5 )
= 18 z −1 + 3z −2 + 3z −3 − 14 z −4 + 25z −5 + 26 z −6 + 60 z −7 − 11z −8 − 16 z −9 − 14 z −10
+ 26 z −11 + 39 z −12 − 10 z −13 . Hence,
{g[ n]} = {18, 3, 3, − 14, 25, 26, 60, − 11, − 16, − 14, 26, 39, − 10}, 1 ≤ n ≤ 13.

6.17

2 N − 1 ⎛ N −1
N −1⎛ N −1

YL ( z) = ∑ ⎜⎜ ∑ x[ m]h[ n − m] ⎟⎟ z − n and YC ( z) = ∑ ⎜⎜ ∑ x[ m]h[〈 n − m〉 N ]⎟⎟ z − n .
n=0 ⎝ m =0
n=0 ⎝ m=0

Not for sale.

154

Now, YL (z) can be rewritten as
N −1⎛ N −1
2 N −1⎛ N −1

YL ( z ) = ∑ ⎜⎜ ∑ x[ m]h[ n − m] ⎟⎟z − n + ∑ ⎜⎜ ∑ x[ m]h[ n − m] ⎟⎟z − n
n=0 ⎝ m =0
n= N ⎝ m =0

N −1⎛ N −1
N −1⎛ N −1

= ∑ ⎜⎜ ∑ x[ m]h[ n − m] ⎟⎟z − n + ∑ ⎜⎜ ∑ x[ m]h[ k − m − N ] ⎟⎟z − ( k − N ) . Therefore,
n=0 ⎝ m =0
k =0 ⎝ m=0

N −1⎛ N −1
N −1⎛ N −1

〈YL ( z)〉 − N
= ∑ ⎜⎜ ∑ x[ m]h[ n − m] ⎟⎟z − n + ∑ ⎜⎜ ∑ x[ m]h[〈 k − m〉 N ] ⎟⎟z − k
(z
−1)
n=0 ⎝ m =0
k =0 ⎝ m=0

N −1⎛ N −1

= ∑ ⎜⎜ ∑ x[ m]h[〈 n − m〉 N ] ⎟⎟z − n = YC ( z ).
n=0 ⎝ m=0

6.18

G( z) = 2 − z −1 + 3z −2 , H ( z) = −2 + 4 z −1 + 2 z −2 − z −3 . Now,
Y L ( z) = G( z)H ( z) = (2 − z −1 + 3z −2 )(−2 + 4 z −1 + 2 z −2 − z −3 )
= −4 + 10 z −1 − 6 z −2 + 8 z −3 + 7 z −4 − 3z −5 . Therefore,
y L [ n] = {− 4 6 2 12 5 − 3}, 0 ≤ n ≤ 5.
Using the MATLAB statement y = conv([2 -1 3], [-2 4 2 -1]); we
obtain
y =
-4
10
-6
8
7
-3
which is seen to be the same result as given above.

YC ( z) = 〈YC ( z)〉

= 〈−4 + 10 z −1 − 6 z −2 + 8z −3 + 7z −4 − 3z −5 〉

( z − 4 −1)
−2
−3

( z − 4 −1)

= −4 + 10 z −1 − 6 z + 8 z + 7 − 3z −1 = 3 + 7 z −1 − 6 z −2 + 8 z −3 . Therefore,
yC [ n] = {1 3 2 12}, 0 ≤ n ≤ 3.
Using the MATLAB statement y = circonv([2 -1 3 0],[-2 4 2 -1]);
we obtain
y =
3
7
-6
8
which is seen to be the same result as given above.

6.19

G( z) =

ρl =

P( z)
P( z)
=
. The residue ρl of G ( z ) at the pole is given by
D( z ) (1 − λl z −1 ) R( z )

P( z)
. Now,
R( z ) z = λ
l

D' ( z ) =

dD( z)
dz

−1

=

d[(1 − λ l z −1 ) R( z)]
dz

−1

= − λl R( z ) + (1 − λl z −1 )

D' ( z) z = λl = − λl R( z) z = λ . Therefore, ρl = − λl
l

Not for sale.

dR( z )
dz −1

. Hence,

P( z)
.
D' ( z ) z = λ
l

155

ρ1
ρ2
3z
3z −1
=
=
+
,
( z + 0.6)( z − 0.3) (1 + 0.6 z −1 )(1 − 0.3z −1 ) 1 + 0.6 z −1 1 − 0.3z −1
3
3
10
3
3
10
where ρ1 =
=
= − , ρ2 =
=
= .
z − 0 . 3 z = − 0. 6 − 0 . 9
z + 0.6 z = 0.3 0.9 3
3

6.20 (a) X a ( z ) =

10 / 3

Therefore, X a ( z ) = −

10 / 3

+

.
1 + 0 .6 z − 1 1 − 0 .3 z − 1
There are three ROCs - R 1 : z < 0.3, R 2 : 0.3 < z < 0.6, R 3 : z > 0.6.
The inverse z –transform associated with the ROC R1 is a left-sided sequence:

Z −1{X a ( z)} = x a [ n] =

10 ⎛
n
⎜ ( −0.6)
3 ⎝

− (0.3) n ⎞⎟µ[ −n − 1].

The inverse z –transform associated with the ROC R 2 is a two-sided sequence:
Z −1{X a ( z)} = x a [ n] = − (−0.6) n µ[ −n − 1] + (0.3) n µ[ n].
3
3
The inverse z –transform associated with the ROC R 3 is a right-sided sequence:
10

10

(

)

Z −1{X a ( z)} = x a [ n] =
− (−0.6) n + (0.3) n µ[ n].
3
10

(b) X b ( z ) =

3z −1 + 0.1z −2 + 0.87 z −3
(1 + 0.6 z

−1

K = X b (0) = 0, ρ1 =

γ2 =
γ1 =

)(1 − 0.3z

−1 2

=K+

)

(1 − 0.3z −1 ) 2

1 + 0.6 z −1

1 + 0.6 z −1

+

γ1
1 − 0 .3 z

−1

+

γ2

(1 − 0.3z −1 ) 2

= 2.7279,
z = −0.6

= 0.6190,
z = 0.3
−2

1
d ⎛⎜ 3z −1 + 0.1z + 0.87 z − 3 ⎞⎟

− 0.3 dz −1 ⎜⎝
1 + 0 .6 z − 1

X b ( z) =

1 + 0 .6 z

−1

3z −1 + 0.1z − 2 + 0.87z − 3

3z −1 + 0.1z − 2 + 0.87z − 3

2.7279

ρ1

0.3469
1 − 0.3z −1

+

= −0.3469. Hence,
z = 0. 3

0.6190
(1 − 0.3z −1 ) 2

.

There are three ROCs - R 1 : z < 0.3, R 2 : 0.3 < z < 0.6, R 3 : z > 0.6.
The inverse z –transform associated with the ROC R1 is a left-sided sequence:
Z −1{ X b ( z )} = xb [n] = 2.7279(−0.6) n µ[−n − 1]

+ (− 0.3469 + 0.6190(n + 1))(0.3)n µ[ −n − 1].
The inverse z –transform associated with the ROC R 2 is a two-sided sequence:
Z −1{ X b ( z )} = xb [ n] = 2.7279( −0.6) n µ[ − n − 1]

+ (− 0.3469 + 0.6190(n + 1))(0.3) n µ[ n].
The inverse z –transform associated with the ROC R 3 is a right-sided sequence:
Not for sale.

156

.

Z −1{X b ( z )} = x b [ n] = 2.7279(−0.6) n µ[ n] + (− 0.3469 + 0.6190( n + 1))(0.3) n µ[ n].

G( z ) =

6.21

P (∞ )
P( z ) p0 + p1 z −1 + L + p M z − M
. Thus, G(∞ ) =
. Now, a partial=

1

N
D(∞ )
D( z )
1 + d1 z + L + d N z
ρl

N
fraction expansion of G ( z ) in z −1 is given by G( z) = ∑

l =1 1 − λl z

−1

, from which we obtain

N
p
G (∞ ) = ∑ ρ l = 0 .
d0
l =1

1

6.22 H ( z ) =

, z > r > 0. By using partial-fraction expansion we write
1 − 2r cos θz −1 + r 2 z − 2

e jθ
e − jθ
e jθ
e − jθ
1
1 ⎛⎜
⎟.

=

H (z) =

jθ ⎜
jθ −1
− jθ −1 ⎟ 2 sin θ ⎜
jθ − 1
− jθ − 1 ⎟
1 − re
z ⎠
z ⎠
(e − e − j ) ⎝ 1 − re z
1 − re
⎝ 1 − re z
1
r n ⎛⎜ e jθ ( n + 1) − e − jθ ( n + 1) ⎞⎟
Thus, h[ n] =
r n e jθ e jnθ µ[ n] − r n e − jθ e − jnθ µ[ n] =
µ[ n]

j 2 sin θ
sin θ ⎜⎝
2j

(

=

)

r n sin ((n + 1)θ )
µ[ n].
sin θ

−1

n = −∞

n = −∞

m =1

m =1

m =0

6.23 (a) X ( z ) = ∑ α n µ[ −n − 1]z − n = ∑ α n z − n = ∑ α − m z m = ∑ ( z / α) m = ∑ ( z / α) m − 1
=

1
z/α
=
, z < α.
1 − ( z / α ) 1 − αz −1

(b) Using the differentiation property, we obtain from Part (a),
Z{nx[ n]} = − z
=

α z −1
(1 − αz

−1 2

dX ( z )
α z −1
=
, z < α . Therefore, Z {y[ n]} = Z{nx[ n] + x[ n]}
dz
(1 − αz −1 ) 2
+

)

1
1 − αz

−1

=

1
(1 − αz −1 ) 2

, z < α.

6.24 (a) Expanding X1 ( z) in a power series we get X1 ( z) = ∑ z − 3n , z > 1. Thus,
n=0

⎧1, if n = 3k and n ≥ 0,
x1[ n] = ⎨
Alternately, using partial-fraction expansion we get
otherwise.
⎩0,

X1 ( z ) =

1
1 − z −3

=

1
3

1 − z −1

+

1
3
1
1+ (
2

+j

3 −1
)z
2

Not for sale.

+

1
3
1
1+ ( −
2

j

3 −1
)z
2

. Therefore,

157

1
1
1
x1[ n] = µ[ n] + ⎛⎜ − − j
3
3⎝ 2

=

1
3

3⎞
1⎛ 1
⎟ µ[ n] + 3 ⎜ − 2 +
2 ⎠

µ[ n] + e − j 2πn / 3 µ[ n] + e j 2 πn / 3 µ[ n] =
1
3

1
3

3⎞
⎟ µ[ n]
2 ⎠

j

1
3

2
3

µ[ n] + cos(2πn / 3) µ[ n]. Thus,

⎧1, if n = 3k and n ≥ 0,
x1[ n] = ⎨
otherwise.
⎩0,

(b) Expanding X 2 ( z) in a power series we get X 2 ( z ) = ∑ z − 4 n , z > 1. Thus,
n=0

⎧1, if n = 4k and n ≥ 0,
Alternately, using partial-fraction expansion we get
x 2 [ n] = ⎨
otherwise.
⎩0,
X 2 (z) =

x 2 [ n] =

1
4

1 − z −1
1
4

+

1
4

1 + z −1

+

1
4
1
2

1+ ( + j

3 −1
)z
2

1
1
1
µ[ n] + (−1) n µ[ n] + ⎛⎜ − − j
4
4⎝ 2

+

1
4

1+ (

1

2

j

3 −1
)z
2

3⎞
1
1
µ[ n] + ⎛⎜ − +

2 ⎠
4⎝ 2

j

. Thus,

3⎞
⎟ µ[ n]
2 ⎠

=

1
4

µ[ n] + ( −1) n µ[ n] + e − j 2 πn / 3 µ[ n] + e j 2 πn / 3 µ[ n]

=

1
4

⎧1, if n = 4k and n ≥ 0,
1
1
µ[ n] + ( −1) n µ[ n] + cos(2πn / 3) µ[ n]. Thus, x 2 [ n] = ⎨
4
2
otherwise.
⎩0,

1
4

1
4

1
4

6.25 (a) X1 ( z) = log(1 − αz −1 ), z > α . Expanding X1( z) in a power series we get

X 1 ( z ) = − αz

−1

∞ αn
α 2 z − 2 α 3 z −3

−K = − ∑
z − n . Therefore,
2
3
n =1 n

αn
µ[ n − 1].
n
⎛ α − z −1 ⎞
⎟ = log 1 − (αz ) −1 , z < α . Expanding X 2 ( z) in a power series
(b) X 2 ( z) = log⎜
⎜ α ⎟

x1[ n] = −

(

we get X 2 ( z ) = −(α z ) −1 −

)

∞ (α z ) − n
( αz ) − 2 ( αz ) − 3

−K = − ∑
. Therefore,
2
3
n
n =1

α −n
µ[ n − 1].
n

1
⎟, z > α . Expanding X 3 ( z ) in a power series we get
(c) X 3 ( z ) = log⎜⎜
−1 ⎟
⎝ 1 − αz ⎠

x 2 [ n] = −

X 3 ( z ) = αz − 1 +

∞ αn
α 2 z −2 α 3 z −3
αn
+
K= ∑
z − n . Therefore, x3 [ n] =
µ[ n − 1].
2
3
n
n =1 n

Not for sale.

158

(

)

⎛ α ⎞
⎟ = − log 1 − (αz) −1 , z < α . Expanding X 4 ( z) in a power
(d) X 4 ( z) = log⎜⎜
−1 ⎟
⎝α−z ⎠
series we get X 4 ( z) = (α z ) −1 +

x 4 [ n] =
6.26 H ( z ) =

α −n
µ[ n − 1].
n
z − 1 + 1 .7 z − 2
(1 − 0.3z −1 )(1 + 0.5z −1 )

∞ (α z ) − n
( α z ) − 2 ( αz ) − 3
. Therefore,
+
−K = ∑
2
3
n
n =1

=k+

ρ1

1 − 0 .3 z − 1

+

ρ2

1 + 0 .5 z − 1

, where k = H (0) = −

34
,
3

z −1 + 1.7z − 2
25
, ρ2 =
= 3.
1 − 0.3z −1 z = −0.5
1 + 0.5z −1 z = 0.3 3
The statement [r,p,k]=residuez([0 1 1.7],conv([1 -0.3],[1 0.6]);
yields
r =
3.0000
8.3333
ρ1 =

z −1 + 1.7z − 2

=

p =
-0.5000
0.3000
k =
-11.3333
34
25 / 3
3
. Hence, its inverse z –transform is given by
+
+
3 1 − 0 .3 z − 1 1 + 0 .5 z − 1
34
25
h[ n] = − δ[ n] + (0.3) n µ[ n] + 3(−0.5) n µ[ n].
3
3

Thus, H ( z ) = −

n = −∞

n = −∞

6.27 G( z ) = Z{g[ n] = ∑ g[ n]z − n with a ROC given by R g and H ( z ) = Z{h[ n] = ∑ h[ n]z − n
with a ROC given by R h .

n = −∞

n = −∞

(a) G * ( z ) = ∑ g * [ n]( z*) − n and G * ( z*) = ∑ g * [ n]z − n . Therefore,
Z{g * [ n]} = G * ( z*) with a ROC given by R g .

(b) Replace n by − m in the sum defining G ( z ). Then

∑ g[ − m]z

m = −∞

m

=

∑ g[ − m](1 / z)

m = −∞

−m

= G(1 / z). Thus, Z{g[ − n]} = G (1 / z ). Since z has

been replaced by 1 / z, the ROC of is given by 1 / R g .

Not for sale.

159

(c) Let y[ n] = αg[ n] + βh[ n]. Then
Y ( z ) = Z{αg[ n] + βh[ n] = αZ{g[ n] + βZ{h[ n] = αG ( z ) + βH ( z ). In this case Y ( z ) will
converge wherever both G ( z ) and H ( z ) converge. Hence the ROC of Y ( z ) is
R g ∩ Rh .
(d) y[ n] = g[ n − no ]. Hence,

Y ( z ) = ∑ y[ n]z − n = ∑ g[ n − n 0 ]z − n =
=z

n = −∞
− no ∞

∑ g[ n]z

n = −∞

−m

m = −∞

∑ y[ m]z

− (m + n o )

m = −∞

= z − n o G( z ). In this case, the ROC of Y ( z ) is same as that of G ( z )

except for the possible addition or elimination of the point z = 0 or z = ∞ (due to the
factor of z − n o ).

n = −∞

n = −∞

n = −∞

(e) y[ n] = α n g[ n]. Hence, Y ( z) = ∑ y[ n]z − n = ∑ αg[ n]z − n = ∑ g[ n]( z / α ) − n
= G ( z / α ). The ROC of Y ( z ) is α R g .

(f) y[ n] = n g[ n]. Hence, Y ( z) = ∑ y[ n]z − n = ∑ ng[ n]z − n . Now,
n = −∞

n = −∞

dG( z)
= − ∑ ng[ n]z − n −1 . Hence,
dz
n = −∞
n = −∞

dG( z)
dG( z )
. In this case, the ROC of
−z
= ∑ ng[ n]z − n −1 . Thus, Y(z) = Z{ng[ n]} = −
dz
dz
n = −∞
Y ( z ) is same as that of G ( z ) except possibly the point z = 0 or z = ∞ .
G( z ) = ∑ g[ n]z − n . Thus,

6.28 From Eq. (6.111), for N = 3, we get
⎡1 z −1 z − 2 ⎤
0
0

−1
D 3 = ⎢1 z1
z1− 2 ⎥. The determinant of D 3 is given by
⎢1 z −1 z − 2 ⎥
2
2 ⎦⎥
⎣⎢
1 z 0−1
det(D 3 ) = 1 z1−1
1 z 2−1

z 0− 2

1

z 2− 2

0 z 2−1 − z 0−1

z1− 2 = 0 z1−1 − z 0−1

= ( z1−1 − z 0−1 )( z 2−1 − z 0−1 )
=

z 0−1

1 z1−1 + z 0−1
1

z 2−1

+ z 0−1

z 0− 2

z −1 − z −1
z1− 2 − z 0− 2 = 1−1 0−1
z2 − z0
z 2− 2 − z 0− 2

z1− 2 − z 0− 2
z 2− 2 − z 0− 2

= ( z1−1 − z 0−1 )( z 2−1 − z 0−1 )( z 2−1 − z1−1 )

−1
−1
∏ ( z k − z l ).

2≥k ≥l≥0

From Eq. (6.111), for N = 4, we get

Not for sale.

160

1 z 0−1
D4 =

1 z1−1
1 z 2−1
1 z3−1

z 0− 2
z1− 2
z 2− 2
z3− 2

1 z 0−1
1 z1−1
det(D 4 ) =
1 z 2−1
1 z3−1
z1−1 − z 0−1

= z 2−1 − z 0−1
z3−1 − z 0−1

z 0− 3

z1− 3
. The determinant of D 4 is given by
z 2− 3
z3− 3
z 0− 2
z1− 2
z 2− 2
z3− 2

z1− 2 − z 0− 2

z 2− 2 − z 0− 2
z3− 2 − z 0− 2

z 0− 3

z 0−1

1

z 0− 2

z1− 3 0 z1−1 − z 0−1
=
z 2− 3 0 z 2−1 − z 0−1
z 3− 3 0 z3−1 − z 0−1

z1− 2 − z 0− 2
z 2− 2 − z 0− 2
z3− 2 − z 0− 2

z 0− 3

z1− 3 − z 0− 3
z 2− 3 − z 0− 3
z3− 3 − z 0− 3

z1− 3 − z 0− 2

z 2− 3 − z 0− 2
z3− 3 − z 0− 2
1 z1−1 + z 0−1

z1− 2 + z1−1 z 0−1 + z 0− 2

= ( z1−1 − z 0−1 )( z 2−1 − z 0−1 )( z3−1 − z 0−1 ) 1 z 2−1 + z 0−1
1 z 3−1 + z 0−1

z 2− 2 + z 2−1 z 0−1 + z 0− 2
z 3− 2 + z3−1 z 0−1 + z 0− 2

1 z1−1 + z 0−1

z1− 2 + z1−1 z 0−1 + z 0− 2

= ( z1−1 − z 0−1 )( z 2−1 − z 0−1 )( z3−1 − z 0−1 ) 0 z 2−1 − z 0−1
0 z3−1 − z 0−1

( z 2−1 − z 0−1 )( z 2−1 + z1−1 + z 0−1 )

( z3−1 − z 0−1 )( z3−1 + z1−1 + z 0−1 )

z −1 − z −1 ( z 2−1 − z1−1 )( z 2−1 + z1−1 + z 0−1 )
= ( z1−1 − z 0−1 )( z 2−1 − z 0−1 )( z3−1 − z 0−1 ) 2−1 1−1
z3 − z1
( z3−1 − z1−1 )( z3−1 + z1−1 + z 0−1 )
= ( z1−1 − z 0−1 )( z 2−1 − z 0−1 )( z 3−1 − z 0−1 )( z 2−1 − z1−1 )( z 3−1 − z1−1 )( z 3−1 − z 2−1 )
=

−1
−1
∏ ( z k − z l ) . Hence, in the general case, det( D N ) =

3≥ k ≥ l ≥ 0

−1
−1
∏ ( z k − z l ). It follows

N − 1≥ k ≥ l ≥ 0

from this expression that the determinant det(D N ) is non-zero, i.e., D N is non-singular, if
the sampling points z k are distinct.
6.29 X ( z) = 1 − 2 z −1 + 3z −2 − 4 z −3 . Thus,
1
X NDFT [ 0] = X ( z 0 ) = 1 − 2⎛⎜ − ⎞⎟
⎝ 2⎠

−1

1
+ 3⎛⎜ − ⎞⎟
⎝ 2⎠
X NDFT [1] = X ( z1 ) = 1 − 2 + 3 − 4 = −2,
1
X NDFT [ 2] = X ( z 2 ) = 1 − 2⎛⎜ ⎞⎟
⎝2⎠
1
X NDFT [3] = X ( z 3 ) = 1 − 2⎛⎜ ⎞⎟
⎝ 3⎠

−1

−1

1
+ 3⎛⎜ ⎞⎟
⎝2⎠
1
+ 3⎛⎜ ⎞⎟
⎝ 3⎠

−2

−2

−2

1
− 4⎛⎜ − ⎞⎟
⎝ 2⎠

1
− 4⎛⎜ ⎞⎟
⎝2⎠
1
− 4⎛⎜ ⎞⎟
⎝ 3⎠

Not for sale.

−3

−3

−3

= 1 + 4 + 3 ⋅ 4 + 4 ⋅ 8 = 49,

= 1 − 4 + 3 ⋅ 4 − 4 ⋅ 8 = −23,
= 1 − 2 ⋅ 3 + 3 ⋅ 9 − 4 ⋅ 27 = −86.

161

I 0 ( z ) = (1 − z −1 )(1 − z −1 )(1 − z −1 ) = 1 −

11 −1
1
1
z + z − 2 − z − 3 ⇒ I 0 (− ) = 10,
6
6
2
1
1 −1
1 −1
1 −1
1 −1 1 − 2
1 −3
I1 ( z ) = (1 + z )(1 − z )(1 − z ) = 1 − z − z + z ⇒ I1 (1) = ,
2
2
3
3
4
12
2
1
2

1
3

2
1
1
5
1
1
1
I 2 ( z ) = (1 + z −1 )(1 − z −1 )(1 − z −1 ) = 1 − z −1 − z − 2 + z − 3 ⇒ I 2 ( ) = − ,
2
3
6
3
6
2
3
5
1
1
1
1
1
I 3 ( z ) = (1 + z −1 )(1 − z −1 )(1 − z −1 ) = 1 − z −1 − z − 2 + z − 3 ⇒ I 3 ( ) = .
2
2
4
4
3
2
49
−2
− 23
− 86
Therefore, X ( z ) =
I 0 ( z) +
I1 ( z ) +
I 2 (z) +
I 3 ( z)
10
1/ 2
− 2/3
5/2
= 4.9 I 0 ( z ) − 4 I1 ( z ) + 34.5I 2 ( z ) − 34.4 I 3 ( z ) = 1 − 2 z −1 + 3z −2 − 4 z −3 .

ˆ
6.30 (a) X ( z ) = ∑ x[ n]z − n . Let Xˆ ( z ) = log( X ( z )) ⇒ X ( z ) = e X ( z ) . Thus,

X ( e jω ) = e
(b) xˆ[ n] =

n = −∞
Xˆ (e jω )

.

(

)

1 π

jωn

− jω
).
∫ log X (e ) e dω. If x[n] is real, then X (e ) = X * (e
2π − π

(

)

(

)

Therefore, log X (e jω ) = log X * (e − j jω ) . .

xˆ * [ n] =

(

)

(

)

1 π
1 π

− jωn
− jω
dω =
) e − jωn dω
∫ log X * (e ) e
∫ log X (e
2π − π
2π − π

=

(

)

1 π

jωn
∫ log X (e ) e dω = xˆ[ n].
2π − π

(

xˆ[ n] + xˆ[ − n] 1 π

(c) xˆ ev [ n] =
=
∫ log X (e )
2
2π − π

=
Similarly, xˆ ev [ n] =

(

)

)

⎛ e jωn + e − jωn ⎞

⎟ dω

2

1 π

∫ log X (e ) cos(ωn)dω.
2π − π

(

xˆ[ n] − xˆ[ − n] 1 π

=
∫ log X (e )
2
2π − π

=

π

(

)

)⎛⎜⎜ e

j ωn

− e − jωn ⎞⎟

2

j

∫ log X (e ) sin(ωn)dω.
2π − π

6.31 x[ n] = aδ[ n] + bδ[ n − 1]. Thus, X ( z ) = Z{x[ n]} = a + bz −1. Also,

n

(b / a ) − n
z .
Xˆ ( z) = log(a + bz −1 ) = log(a ) + log(1 + b / az −1 ) = log(a) + ∑ (−1) n −1
n
n = −∞

Not for sale.

162

log( a ),
if n = 0,

n
(b / a )

Therefore, xˆ[ n] = ⎨(−1) n −1
, for n > 0,
n

0,
otherwise.
⎪⎩

6.32 (a) Xˆ ( z ) = log( K ) + ∑ log(1 − α k z −1 ) + ∑ log(1 − γ k z ) − ∑ log(1 − β k z −1 ) − ∑ log(1 − δ k z )
k =1
n
Nα ∞ α
k −n

= log( K ) − ∑ ∑

k =1 n =1 n

z

k =1
k =1
n
n
N
Nλ ∞ γ
Nδ ∞ δ n
β ∞ β
k n
k −n
k n

− ∑ ∑

k =1 n =1 n

z − ∑ ∑

k =1 n =1 n

z

− ∑ ∑

k =1 n =1 n

k =1

z .

log( K ),
n = 0,

n
n
N
Nα ∞ α
⎪⎪ β ∞ β
Thus, xˆ[ n] = ⎨ ∑ ∑ k − ∑ ∑ k , n > 0,
⎪ k =1 n =1 n k = 1 n =1 n
⎪ Nγ ∞ γ−n Nδ ∞ δ −n
⎪ ∑ ∑ k − ∑ ∑ k , n < 0.
k =1 n =1 n
⎩⎪k =1 n =1 n
(b) xˆ[ n] < N

r

n

as n → ∞, where r is the maximum value of α k , β k , γ k , and δ k for all
n
values of k, and N is a constant. Thus, xˆ[ n] is a decaying bounded sequence as n → ∞.

(c) From Part (a) if α k = β k = 0, then xˆ[ n] = 0 for all n > 0, and is thus anti-causal.
(d) If γ k = δ k = 0, then xˆ[ n] = 0 for all n < 0, and is thus causal.
6.33 If X ( z ) has no poles and zeros on the unit circle, then from Part (b) of Problem 6.32,
γ k = δ k = 0, then xˆ[ n] = 0 for all n < 0.
dXˆ ( z)
1 dX ( z)
dXˆ ( z )
dX ( z )
=
. Thus, z
Xˆ ( z ) = log (X ( z )). Therefore,
.
= zX ( z )
dz
X ( z) dz
dz
dz
n

Taking the inverse –transform we get nx[ n] = ∑ k xˆ[ k ]x[ n − k ], n ≠ 0. Or,
k =0

n −1 k

x[ n] = ∑

k =0 n

xˆ[ k ]x[ n − k ] + xˆ[ n]x[0]. Hence, xˆ[ n] =

x[ n] n −1 ⎛ k ⎞ xˆ[ k ]x[ n − k ]
, n ≠ 0.
− ∑ ⎜ ⎟
x[0] k = 0 ⎝ n ⎠
x[0]

For n = 0, xˆ[0] = Xˆ ( z ) z = ∞ = X ( z ) z = ∞ = log( x[0]). Thus,

Not for sale.

163

0,
n < 0,

xˆ[ n] = ⎨
log( x[0]),
n = 0,
⎪ x[ n] n −1 ⎛ k ⎞ xˆ[ k ]x[ n − k ]
− ∑ ⎜ ⎟
, n > 0.

x[0]
⎩ x[0] n = 0 ⎝ n ⎠

6.34

| H(e j5ω ) |

0
0.02π

0.06π

5

π

__
5

ω

6.35 Given the real part of a real, stable transfer function
∑ N a cos(iω) A(e jω )
,
H re (e jω ) = i = 0 i
=
∑ iN= 0 bi cos(iω) B(e jω )
the problem is to determine the transfer function H ( z ) =

(6-1)
−i
P( z ) ∑ iN= 0 pi z
.
=
D( z ) ∑ iN= 0 pi z − i

(a) H re (e jω ) = [ H (e jω ) + H * (e jω )] = [ H (e jω ) + H (e − jω )]
2
2
1

1

= [ H ( z ) + H ( z −1 )]
1
2

z = e jω

.

Substituting H ( z ) = P ( z ) / D( z ) in the above we get

H re (e jω ) =

1 P ( z ) D( z − 1 ) + P ( z − 1 ) D ( z )
2
D( z ) D( z − 1 )

,

(6-2)

z = e jω

which is Eq. (6.117).
(b) Comparing Eqs. (6-1) and (6-2) we get B(e jω ) = D( z ) D( z −1 )

A(e jω ) = [ P( z) D( z −1 ) + P( z −1 ) D( z)]
1
2

z = e jω

z = e jω

,

(6-3)

.

(6-4)

N

Now, D ( z ) is of the form D( z ) = Kz − N ∏ ( z − z i ),
i =1

where the zi ' s are the roots of B( z ) = B(e

)

e jω = z

(6-5)
inside the unit circle and K is a scalar

constant. Putting ω = 0 in Eq. (6-3) we get B(1) = [ D(1)]2 , or
Hence,
K = B(1) / ∏ iN=1 (1 − zi ).
Not for sale.

B(1) = K ∏ iN=1 (1 − zi ).
(6-6)

164

(c) By analytic continuation, Eq. (6-4) yields A( z ) = [ P( z ) D( z −1 ) + P( z −1 ) D( z )]. (6-7)
1
2

Substituting A( z ) =

1 N
i
−i
∑ ai ( z + z ) and the polynomial forms of P ( z ) and D ( z ), we get
2 i =0

N
N
N
N
⎛ N
i
−i
− i ⎞⎛
i⎞ ⎛
i ⎞⎛
−i ⎞
∑ ai ( z + z ) = ⎜⎜ ∑ p i z ⎟⎟⎜⎜ ∑ d i z ⎟⎟ + ⎜⎜ ∑ p i z ⎟⎟⎜⎜ ∑ d i z ⎟⎟ and equating the coefficients
i =0

⎠⎝ i = 0
⎠ ⎝ i =0
⎠⎝ i = 0
⎝i =0

of ( z i + z −i ) on both sides, we arrive at a set of N equations which can be solved for the
numerator coefficients pi of H ( z ).
For the given example, i.e., H re (e) =

1 + cos(ω) + cos(2ω)
, we observe
17 − 8 + cos(2ω)

A( z ) = 1 + ( z + z −1 ) + ( z 2 + z − 2 ).
1
2

1
2

Also, B( z ) = 17 − 4( z 2 + z −2 ), which has roots at z = ±

1
2

(6-8)

and z = ±2. Hence,

D( z ) = Kz − 2 ( z − )( z + ) = K ( z 2 − )z − 2 .
1
2

1
4

1
2

(6-9)

Also, from Eq. (6-6) we have K = 17 − 8 /(1 − ) = 4, so that D( z ) = 4 − z −2 .
1
4

Substituting Eqs. (6-8), (6-9) and P ( z ) = p0 + p1 z −1 + p2 z −2 in Eq. (6-7) we get

[

]

1 + ( z + z −1 ) + ( z 2 + z − 2 ) = ( p0 + p1 z −1 + p2 z − 2 )(4 − z 2 ) + ( p0 + p1 z + p2 z 2 )(4 − z − 2 ) .
2
2
1

1

Equating the coefficients of ( z i + z −i ) / 2, 0 ≤ i ≤ 2, on both sides we get
4 p0 − p2 = 1, 3 p1 = 1, 4 p2 − p0 = 1. Solving these equations we then arrive at

p0 = p1 = p2 = 1 / 3. Therefore, H ( z ) =

1 + z −1 + z −2
3(4 − z − 2 )

.

d M + d M − 1 z −1 + L + d 1 z − M + 1 + z − M

⇒ A(1) = 1 and A( −1) = −1 if M is odd.
1 + d1 z −1 + L + d M −1 z − M + 1 + d M z − M
In which case, G (1) = H (1) and G ( −1) = H (−1). If is even, then G (1) = H (1) and
G ( −1) = H (1). .

6.36 A( z ) =

6.37 H ( z ) = H1 ( z ) H 2 ( z ) + H 3 ( z ) = (1.2 + 3.3z −1 + 0.7 z −2 )(−4.1 − 2.5z −1 + 0.9 z −2 )
+ 2.3 + 4.3z −1 + 0.8 z −2 = −2.62 − 12.23z −1 − 9.24 z −2 + 1.22 z −3 + 0.63z −4 .

6.38 (a) (1 − 0.1z −1 + 0.14 z −2 + 0.49z −3 )Y ( z ) = (5 + 9.5z −1 + 1.4 z −2 − 24 z −3 ) X ( z ) ⇒

Not for sale.

165

Y (z)
5 + 9.5z −1 + 1.4 z −2 − 24 z −3
=
. Using Program 6_1.m we factorize H ( z )
X ( z) 1 − 0.1z −1 + 0.14 z − 2 + 0.49z − 3
and develop is pole-zero plot shown below:
H ( z) =

Numerator factors
1.00000000000000
3.10000000000000
1.00000000000000 -1.20000000000000

4.00000000000000
0

Denominator factors
1.00000000000000 -0.80000000000000
1.00000000000000
0.70000000000000

0.70000000000000
0

Gain constant
5
The factored form of H ( z ) is thus H ( z ) =

5(1 + 3.1z −1 + 4 z −2 )(1 − 1.2 z −1 )
(1 − 0.81z −1 + 0.7 z − 2 )(1 + 0.7 z −1 )

Imaginary Part

1
0.5
0
-0.5
-1
-1

0
Real Part

1

As all poles are inside the unit circle, H ( z ) is BIBO stable.
(b) (1 − 0 . 5 z − 1 + 0 . 1 z − 2 + 0 . 3 z − 3 − 0 . 0936 z − 4 )Y ( z )

= (5 + 16.5z −1 + 14.7z −2 − 22.4 z −3 − 33.6 z −4 ) X ( z ) ⇒

Y ( z) 5 + 16.5z −1 + 14.7z −2 − 22.4 z −3 − 33.6 z −4
. Using Program 6_1.m we
=
X ( z) 1 − 0.5z −1 + 0.1z − 2 + 0.3z − 3 − 0.0936z − 4
factorize H ( z ) and develop is pole-zero plot shown below:
H (z) =

Numerator factors
1.00000000000000
1.00000000000000
1.00000000000000

-1.20000000000000
3.10000000000001
1.39999999999999

0
4.00000000000001
0

Denominator factors
1.00000000000000
1.00000000000000
1.00000000000000

0.59950918226500
-0.80131282790906
-0.29819635435594

0
0.52021728677142
0

Not for sale.

166

Gain constant
5
The factored form of H ( z ) is thus
5(1 − 1.2 z −1 )(1 + 3.1z −1 + 4 z −2 )(1 + 1.4 z −1 )

H ( z) =

(1 + 0.5995 z −1 )(1 − 0.801313 z −1 + 0.520217 z − 2 )(1 − 0.2982 z −1 )
As all poles are inside the unit circle, H ( z ) is BIBO stable.

Imaginary Part

1
0.5
0
-0.5
-1
-2

-1

0
Real Part

1

6.39 A partial-fraction expansion of H ( z ) in z −1 using the M-file residuez yields

H ( z) = −

1.21212
1 − 0.4 z −1

+

2.21212(1 − 0.81781z −1 )
1 + 0.5z −1 + 0.3z − 2

. Comparing the denominator of the

quadratic factor with 1 − 2r cos(ω o ) z −1 + r 2 z −2 we get r = 0.3 = 0.54772 and
0.5
cos(ωo ) = −
, or ωo = 2.04478. Hence, from Table 6.1 we have
2 0.3

h[ n] = −1.21212(0.4) n µ[ n] + ( 0.3 ) n cos(2.04478n)µ[ n].
6.40 (a) A partial-fraction expansion of H ( z ) in z −1 using the M-file residuez yields
5
2
H ( z ) = −2 +

. Hence, from Table 6.1 we have
−1
1 + 0 .6 z
1 − 0 .3 z − 1

h[ n] = −2δ[ n] + 5(−0.6) n µ[ n] − 2(0.3) n µ[ n].
(b) x[ n] = 2.1(0.4) n µ[ n] + 0.3(−0.3) n µ[ n]. Its z –transform is thus given by
X ( z) =

2 .1
1 − 0 .4 z − 1

+

0 .3
1 + 0 .3 z − 1

=

2.4 + 0.51z −1
(1 − 0.4 z −1 )(1 + 0.3z −1 )

, z > 0.4. The z –transform of

⎤ ⎡1 − 3.3z −1 + 0.36 z − 2 ⎤
2.4 + 0.51z −1
the output is then given by Y ( z ) = ⎢
⎥⋅⎢
⎥.
⎢⎣ (1 − 0.4 z −1 )(1 + 0.3z −1 ) ⎥⎦ ⎢⎣1 + 0.3z −1 − 0.18 z − 2 ⎥⎦
A partial-fraction expansion of Y ( z ) in z −1 using the M-file residuez yields

Not for sale.

167

Y ( z) =

9.3
1 + 0.6z −1

(

16.8

+

1 − 0 .4 z − 1

12.3

1 − 0.3z −1

2.4
1 + 0.3z −1

, z > 0.6. Hence, from Table 6.1

)

we have y[ n] = 9.3(−0.6) n − 16.8(0.4) n + 12.3(0.3) n − 2.4(−0.3) n µ[ n].
6.41 (a) H ( z ) = Z{h[ n]} =

Y ( z) = H ( z) X ( z) =

1
1 + 0 .4 z − 1
1

, z > 0.4, X ( z ) = Z{x[ n]} =

(1 + 0.4 z −1 )(1 − 0.2 z −1 )

using the M-file residuez yields Y ( z ) =
2
3

1
1 − 0 .2 z − 1

, z > 0.4. Thus,

, z > 0.4. A partial-fraction expansion of
2/3

1 + 0 .4 z

−1

+

1/ 3
1 − 0 .2 z − 1

. Hence, from Table 6.1

1
3

y[ n] = (−0.4) n µ[ n] + (0.2) n µ[ n].
(b) H ( z ) = Z{h[ n]} =

Y ( z) = H ( z) X ( z) =

1
1 + 0 .2 z
1

−1

(1 + 0.2 z −1 ) 2

, z > 0.2, X ( z ) = Z{x[ n]} =

1
1 + 0 .2 z − 1

, z > 0.4. Thus,

, z > 0.2. Hence, from Table 6.1,

y[ n] = (n + 1)(−0.2) n µ[ n].
6.42 Y ( z ) = Z{y[ n]} =

2
1 + 0 .3 z

−1

, z > 0.3, X ( z ) = Z{x[ n]} =

4
1 − 0 .6 z − 1

, z > 0.2. Thus,

Y ( z ) 0.5(1 − 0.6 z −1 )
=
, z > 0.3. A partial-fraction expansion of using the M-file
X ( z)
1 + 0.3z −1
1 .5
residuez yields H ( z ) = −1 +
. Hence, from Table 6.1,
1 + 0 .3 z − 1
H (z) =

h[ n] = − δ[ n] + 1.5(−0.3) n µ[ n].
6.43 (a) Taking the –transform of both sides of the difference equation we get
Y (z)
2
Y ( z ) = 0.2 z −1Y ( z ) + 0.08z −2Y ( z ) + 2 X ( z ). Hence, H ( z ) =
=
.
X ( z ) 1 − 0.2 z −1 − 0.08 z − 2
(b) A partial-fraction expansion of using the M-file residuez yields
4/3
2/3
H ( z) =
+
. Hence, from Table 6.1,
−1
1 − 0 .4 z
1 + 0 .2 z − 1
4
3

2
3

h[ n] = (0.4) n µ[ n] + (−0.2) n µ[ n].
(c) Now S ( z ) = Z{s[ n]} = H ( z ) ⋅ Z{µ[ n]} =

Not for sale.

2
(1 − 0.2 z −1 − 0.08z − 2 )(1 − z −1 )

.

168

A partial-fraction expansion of using the M-file residuez yields
2.7778
0.8889
0.1111
S ( z) =

+
. Hence, from Table 6.1,
−1
−1
1− z
1 − 0 .4 z
1 + 0 .2 z − 1

s[ n] = 2.7778 µ[ n] − 0.8889(0.4) n µ[ n] + 0.1111(−0.2) n µ[ n].
6.44 H ( z) =

1 − z −2
1 − (1 + α) cos(ωc )z −1 + αz − 2

. Thus,

1 − e j 2ω

H ( e jω ) =

.
1 − (1 + α) cos(ωc ) e − jω + αe − j 2ω
2(1 − cos 2ω)
H ( e jω ) 2 =
2
2
1 + (cos ω c ) (1 + α ) + α 2 + 2α cos 2ω − 2 cos ω c (1 + α ) 2 cos ω

=

4 sin 2 ω
2

2

2

(1 + α) (cos ω − cos ωc ) + (1 − α) sin ω

. Note that H (e jω ) 2 is maximum when

cos ω = cos ωc , i.e, ω = ωc . Then H (e jω c ) 2 =

4 sin 2 ωc
(1 − α) 2 sin 2 ωc

=

4
(1 − α) 2

, and hence,

H (e jω c ) = 2 /(1 − α ).

6.45 H ( z ) = 1 − αz −R ⇒ H (e jω ) = 1 − αe − jωR . Then, H (e jω = 1 + α 2 − 2α cos(ωR).
H(e jω is maximum when cos( ωR ) = −1 and is minimum when cos(ωR ) = 1. The

maximum value of H (e jω ) is 1 + α , and the minimum value is 1 − α . H(e jω has R
peaks and R dips in the range 0 ≤ ω < 2π.
2 πk
(2 k + 1)π
and the dips are located at ω = ω k =
The peaks are located at ω = ω k =
,
R
R
0 ≤ k ≤ R − 1.
Phase Response
1

1.5

0.5

Amplitude

Magnitude Response
2

1
0.5
0

0

0.5

(

6.46 G (e jω ) = H (e jω )

1
ω/π

1.5

2

0
-0.5
-1

0

0.5

1

1.5

2

ω/ π

)3 = (1 − αe jωR )3 .
Not for sale.

169

M −1

6.47 G(e jω ) = ∑ α n e − jωn =
n=0

1 − α M e − jωM
1 − αe

− jω

. Note that G (e jω ) = H (e jω ) for

1− αM
. Hence, to make the dc value of the
1− α
magnitude response equal to unity, the impulse response should be multiplied by a constant
αn =

1
,
M

0 ≤ n ≤ M − 1. Now, G(e j 0 ) =

K = (1 − α ) /(1 − α M ) .

6.48 Y (e jω ) = X (e jω ) + αe − jωR Y (e jω ). Hence, H (e jω ) =

Y ( e jω )

=

1

. Maximum
X ( e jω ) 1 − α e − jωR
1
1
. There are R peaks and dips
value of H (e jω ) is
and the minimum value is
1− α
1+ α

in the range 0 ≤ ω ≤ 2π. The locations of the peaks and dips are given by
α
1 − αe − jωR = 1 ± α or, e − jωR = ± . The locations of the peaks are given by
α
(2π + 1)k
2πk
and the locations of the dips are given by ω = ωk =
,
ω = ωk =
R
R

0 ≤ k ≤ R − 1. Plots of the magnitude and phase responses of H (e jω ) for α = 0.8 and
R = 6 are shown below:
Phase Response
1

4

0.5

Amplitude

Magnitude Response
5

3
2
1
0

0

0.5

1
ω/π

2

-0.5
-1

0

0.5

1

1.5

2

ω/ π

b0 + b1e − jω + b2 e − j 2ω

(b0 e jω + b2 e − jω ) + b1

=
1 + a1e − jω + a 2 e − j 2ω
(e jω + a 2 e − jω ) + a1
b + (b0 + b2 ) cos ω + j (b0 − b2 ) sin ω
. Therefore,
= 1
a1 + (1 + a 2 ) cos ω + j (1 − a 2 ) sin ω

6.49 A(e

A(e

)=

1.5

0

)

2

[
b1 + (b0 + b2 )]2 cos 2 ω + (b0 − b2 ) 2 sin 2 ω
= 1. Hence, at ω = 0, we have
=
[a1 + (1 + a2 )]2 cos 2 ω + (1 − a2 ) 2 sin 2 ω

b1 + (b0 + b2 ) = ±[ a1 + (1 + a 2 )], and at ω = π / 2, we have b0 − b2 = ± (1 − a 2 ).
Solution #1: Consider b0 − b2 = 1 − a 2 . Choose b 0 = 1, − b 2 = 1 − a 2 , and b 2 = a 2 .
Substituting these values in b1 + (b0 + b2 ) = ±[ a1 + (1 + a 2 )], we get b1 = a1 . In this case,

Not for sale.

170

A(e

)=

1 + a1e − jω + a 2 e − j 2ω

1 + a1e − jω + a 2 e − j 2ω

= 1, a trivial solution.

Solution #2: Consider b0 − b2 = a 2 − 1. Choose b0 = a 2 and b2 = 1. Substituting these
values in b1 + (b0 + b2 ) = ±[ a1 + (1 + a 2 )], we get b1 = a1 . In this case,

A(e jω ) =

a 2 + a1e − jω + e − j 2ω
.
1 + a1e − jω + a 2 e − j 2ω

6.50 From Eq. (2.20), the input-output relation of a factor-of-2 up-sampler is given by
⎧ x[ n / 2], n = 0, ± 2, ± 4,K
x u [ n] = ⎨
otherwise.
⎩ 0,
The DTFT of x u [n] is therefore given by

Y (e jω ) = ∑ y[ n]e − jωn =
n = −∞

∑ x[ n / 2]e

− jωn

n = −∞
n even

=

∑ x[ m]e

− j 2ωm

m = −∞

= X (e j 2ω ) where X (e jω )

is the DTFT of x[n].
1
. Thus,
1 − 0.5 cos ω + j 0.5 sin ω
1 − 0.5e
1
1
H (e ± jπ / 4 ) =
=
= 0.6512 m j1.1907.
1 − 0.5 cos(± π / 4) + j 0.5 sin(± π / 4) 0.6464 ± j 0.3536

6.51 H (e jω ) =

1

− jω

=

Therefore, H (e ± jπ / 4 ) = 1.3572 and arg{H (e ± jπ / 4 )} = θ (± jπ / 4) = m1.0703.
Now, for an input x[ n] = sin(ω o n) µ[ n], the steady-state output is given by
y[ n] = H (e jωo ) sin (ωo n + θ (ωo )) which for ωo = π / 4 reduces to
π
π
y[ n] = H (e jπ / 4 ) sin⎛⎜ n + θ (π / 4) ⎞⎟ = 1.3572 sin⎛⎜ n − 1.0703 ⎞⎟.
4

⎝4

6.52 To guarantee the stability of G ( z ), the transformation z → F ( z ) should be such that the
unit circle remains inside the ROC after the mapping. If the points inside the unit circle
after the mapping remains inside the unit circle, G ( z ) will be causal and stable. On the
other hand, if the points inside the unit circle after the mapping move outside the unit
circle, G ( z ) will be stable but anti-causal. For example, the mapping z → − z will ensure
that G ( z ) will be causal and stable, whereas, the mapping z → z −1 will result in a G ( z )
that is stable, but anti-causal.
6.53

Not for sale.

171

K

6.54

−1
n
∑ h[ n] = 0.95 ∑ h[ n] . Since H ( z) = 1 /(1 − βz ), h[ n] = ( β ) µ[ n]. Thus,
2

n=0

1− β

2

n=0

2K

1− β

2

=

0.95
1− β

2

. Solving this equation for we get K = 0.5

log(0.05)
.
log( α )

6.55 Let the output of the predictor of Figure P6.3(a) be denoted by E ( z ). Then analysis of this
structure yields E ( z) = P( z)[U ( z) + E ( z)] and U ( z ) = X ( z ) − E ( z ). From the first equation
P( z)
we have E ( z ) =
U ( z ) which when substituted in the second equation yields
1 − P( z )
U ( z)
H (z) =
= 1 − P ( z).
X (z)
Analyzing Figure P6.3(b) we get Y ( z ) = V ( z ) + P ( z )Y ( z ) which leads to
1
Y ( z)
, which is seen to be the inverse of H ( z ).
G( z) =
=
V ( z ) 1 − P( z)
1
For P ( z) = h1 z −1 , H ( z ) = 1 − h1 z −1 and G ( z ) =
. Similarly, for
1 − h1 z −1
1
P( z ) = h1 z −1 + h2 z −2 , H ( z) = 1 − h1 z −1 − h2 z −2 and G ( z ) =
.
−1
1 − h1 z − h2 z − 2
6.56 Y ( z) = [H 0 ( z)F0 ( z) − H 0 (− z )F0 (− z)]X ( z). Since the output is a delayed replica of the
input, we must have H 0 ( z ) F0 ( z ) − H 0 ( − z ) F0 ( − z ) = z − r . But, H 0 ( z ) = 1 + αz −1 . Hence,
(1 + αz −1 ) F0 ( z ) − (1 − αz −1 )F0 ( − z ) = z −r . Let F0 ( z ) = a 0 + a1 z −1 . This implies,
2( a 0 α + a1 ) z −1 = z −r .

The solution is therefore, r = 1 and 2(a 0 α + a1 ) = 1. One possible solution is thus

a 0 = 1 / 2 and a1 = 1 / 4. Hence, F0 ( z ) = 0.25(1 + z −1 ).
6.57 H1 ( z ) = Z{h1[ n]} = 1.2 +

0 .5

0 .6

1 + 0 .5 z − 1 1 − 0 .2 z − 1
function of the inverse transform is thus

=

1.1 − 0.04 z −1 − 0.12 z −2
1 + 0.3z −1 − 0.1z − 2

. The transfer

1
1 + 0.3z −1 − 0.1z −2
(1 + 0.5z −1 )(1 − 0.2 z −1 )
=
. As both
=
H1 ( z ) 1.1 − 0.04 z −1 − 0.12 z − 2 1.1(1 − 0.349 z −1 )(1 + 0.3126 z −1 )
poles are inside the unit circle, H 2 ( z) is stable and causal with an ROC z > 0.349. A
H 2 ( z) =

partial-fraction expansion in obtained using the M-file residuez is

Not for sale.

172

0.498
0.42224464
1
+

. Hence,
1.2 1 − 0.34897 z −1 1 + 0.31261z −1
1
δ[ n] + 0.498(0.34897) n µ[ n] − 0.42224464(−0.31261) n µ[ n].
1 .2

H 2 (z) =

h2 [ n] =

6.58 Now, H ( z ) z = e jω = H (e jω ) = H (e jω ) e jθ(ω) . Denote H ' ( z) =

dH ( z )
.
dz

From the above we get ln H ( z ) z = e jω = ln H (e jω ) + jθ(ω). Therefore,

d H (e jω ) / dω
H ' ( z) dz
dθ(ω) d H (e ) / dω
=
− jτ g (ω).

=
+j
H ( z) dω z = e jω

H ( e jω )
H (e jω )

(6-A)

d H (e jω ) / dω
H ' ( z)
−j
Hence, τ g (ω) = − z
.
H ( z ) z = e jω
H ( e jω )

(6-B)

Replacing by in Eq, (6-A) we arrive at

d H (e jω ) / dω
'
(
)
H
z
τ g (ω) = − z −1
+j
.
H ( z ) z = e jω
H (e jω )
Adding Eqs. (6-B) and (6-C), and making use of the notation T ( z) = z

τ g (ω) = −

T ( z ) + T ( z −1 )
2

(6-C)

H ' ( z)
we finally get
H ( z)

.
z = e jω

M6.1 (a) The output data generated by Program 6_1 is as follows:
Numerator factors
1.00000000000000 -2.10000000000000
5.00000000000000
1.00000000000000 -0.40000000000000
0.90000000000000
Denominator factors
1.00000000000000
2.00000000000000
1.00000000000000 -0.20000000000000

4.99999999999999
0.40000000000001

Gain constant
0.50000000000000
Hence, G1 ( z ) =

0.5(1 − 2.1z −1 + 5z −2 )(1 + 2 z −1 + 0.9 z −2 )

(1 + 2 z −1 + 5z − 2 )(1 − 0.2 z −1 + 0.4 z − 2 )
The pole-zero plot of G1 ( z) is given below:

Not for sale.

.

173

Imaginary Part

2
1
0
-1
-2
-3

-2

-1

0
1
Real Part

2

3

There are 3 ROCs associated with G1 ( z) : R1 : z < 0.4 , R 2 : 0.4 < z < 5 ,
and R 3 : z > 5. The inverse z –transform corresponding to the ROC R1 is a leftsided sequence, the inverse z –transform corresponding to the ROC R 2 is a two-sided
sequence, and the inverse z –transform corresponding to the ROC R 3 is a right-sided
sequence.
(b) The output data generated by Program 6_1 is as follows:
Numerator factors
1.00000000000000
1.20000000000000
3.99999999999999
1.00000000000000 -0.50000000000000
0.90000000000001
Denominator factors
1.00000000000000
2.10000000000000
1.00000000000000
0.60000000000003
1.00000000000000
0.39999999999997

4.00000000000001
0
0

Gain constant
1
Hence, G2 ( z ) =

(1 + 1.2 z −1 + 4 z −2 )(1 − 0.5z −1 + 0.9 z −2 )

(1 + 2.1z −1 + 4 z − 2 )(1 + 0.6 z −1 )(1 + 0.4 z −1 )
The pole-zero plot of G2 ( z) is given below:

Imaginary Part

2
1
0
-1
-2
-2

-1

0
1
Real Part

2

There are 4 ROCs associated with G2 ( z) : R1 : z < 0.4, R 2 : 0.4 < z < 0.6,

Not for sale.

174

R 3 : 0.6 < z < 2, and R 4 : z > 2. The inverse z –transform corresponding to the ROC
R1 is a left-sided sequence, the inverse z –transform corresponding to the ROC R 2 is a
two-sided sequence, the inverse z –transform corresponding to the ROC R 3 is a twosided sequence, and the inverse z –transform corresponding to the ROC R 4 is a rightsided sequence.
M6.2 (a) The output data generated by Program 6_3 is as follows:
Residues
-3.33333333333333 3.33333333333333
Poles
-0.60000000000000 0.30000000000000
Constants
0
Hence, the partial-fraction expansion of X a (z ) is given by
10 / 3
10 / 3
X a ( z) = −
+
. The z –transform has poles at z = −0.6 and at
1 + 0 .6 z − 1 1 − 0 .3 z − 1
z = 0.3. Thus, it is associated with ROCs as given in the solution of Problem 6.20 which
also shows their corresponding inverse z –transform.
(b) The output data generated by Program 6_3 is as follows:
Residues
Columns 1 through 2
2.33333333333333
-3.66666666666667 + 0.00000008829151i
Column 3
4.33333333333333 - 0.00000008829151i
Poles
Columns 1 through 2
-0.60000000000000
0.30000000000000 - 0.00000000722385i
Column 3
0.30000000000000 + 0.00000000722385i
Constants
Hence, the partial-fraction expansion of X b (z) is given by
2.3333
3.66666
4.33333
. The z –transform has two poles
X b ( z) =

+
1 + 0.6 z −1 1 − 0.3z −1 (1 − 0.3z −1 ) 2
at z = −0.6 and one at z = 0.3. Thus, it is associated with ROCs as given in the solution
of Problem 6.20 which also shows their corresponding inverse z –transform.
M6.3 (a) X1 ( z) = 3 −

4
5+ z

−1

7
6+z

−1

= 3−

4/5
1 + (1 / 5)z

Not for sale.

−1

7/6
1 + (1 / 6)z −1

.

175

The output data generated by Program 6_4 is as follows:
Numerator polynomial coefficients
1.03333333333333
0.73333333333333 0.1000000000000
Denominator polynomial coefficients
1.00000000000000
0.36666666666667
Hence, X1 ( z) =

31 + 22 z

−1

+ 3z

−1

+ z −2

30 + 11z

.

1.4 + z −1

3

(b) X 2 ( z) = −2.5 +

0.03333333333333

−2

1 + 0.4 z −1 1 + 0.6 z − 2
3
0.7 − j 6454972243679 0.7 + j 6454972243679
= −2.5 +

.

1 + 0.4 z −1 1 − j 0.774596669z −1
1 + j 0.774596669z −1
The output data generated by Program 6_4 is as follows:
Numerator polynomial coefficients
-0.9000
-2.5600
-0.1000
-0.6000
Denominator polynomial coefficients
1.0000
0.4000
0.6000
0.2400
Hence, X b ( z) = −

0.9 + 2.56 z −1 + 0.1z −2 + 0.6 z −3
1.0 + 0.4 z −1 + 0.6 z − 2 + 0.24 z − 3
5

(c) X 3 ( z) =

6

+

+

.

−4

1 + 0.64 z −1 4 + 2 z −1 (4 + 2 z −1 ) 2
5
1.5
− 0.25
=
+
+
.
1 + 0.64 z −1 1 + 0.5z −1 (1 + 0.5z −1 ) 2
The output data generated by Program 6_4 is as follows:
Numerator polynomial coefficients
6.2500
6.5500
1.7300
0
Denominator polynomial coefficients
1.0000
1.6400
0.8900
0.1600
Hence, X 3 ( z ) =

6.25 + 6.55z −1 + 1.73z −2
1 + 1.64 z −1 + 0.89z − 2 + 0.16 z − 3

(d) X 4 ( z) = −5 +

2

+

.

z −1

4 + 3z −1 4 + 3z −1 + 0.9z − 2
0.5
j 0.4303
j 0.4303
.
= −5 +

+
1 + 0.75z −1 1 + (0.3750 - j0.2905)z −1 1 + (0.3750 + j0.2905)z −1
The output data generated by Program 6_4 is as follows:
Numerator polynomial coefficients
Not for sale.

176

-4.5000

-6.8750

-3.6375

-0.8438

Denominator polynomial coefficients
1.0000
1.5000
0.7875
0.1688
Hence, X 4 ( z) = −

4.5 + 6.875z −1 + 3.6375z −2 + 0.84375z −3
1 + 1.5z −1 + 0.7875z − 2 + 0.16875z − 3

.

M6.4 (a) The inverse z –transform of X a (z ) from it partial-fraction expansion form is thus
4

7

x a [ n] = 3δ[ n] − (−1 / 5) n µ[ n] − (−1 / 6) n µ[ n]. The first 10 samples of xa [n] obtained
5
6
by evaluating this expression in MATAB are given by
Columns 1 through 4
1.0333333333
0.3544444444 -0.0644074074
0.0118012345
Columns 5 through 8
-0.0021802057
0.0004060343

-0.0000762057

0.0000144076

Columns 9 through 10
-0.0000027426
0.0000005254
The first 10 samples of the inverse z–transform of the rational form of X a (z ) obtaining
using the M-file impz are identical to the samples given above.
(b) The inverse z –transform of X b (z) from it partial-fraction expansion form is thus
x b [ n] = −2.5δ[ n] + 3( −0.4) n µ[ n] − (0.7 − j 0.6454972243 679)( j 0.7745966692 41) n µ[ n]

− (0.7 + j 0.6454972243679)(− j 0.774596669241) n µ[ n] .
The first 10 samples of x b [n] obtained by evaluating this expression in MATAB are
given by
Columns 1 through 4
-0.900000000 -2.200000000

1.320000000

0.408000000

Columns 5 through 8
-0.427200000 -0.390720000

0.314688000

0.211084800

Columns 9 through 10
-0.179473920 -0.130386432
The first 10 samples of the inverse z–transform of the rational form of X b (z) obtaining
using the M-file impz are identical to the samples given above.
(c) The inverse z –transform of X c (z ) from it partial-fraction expansion form is thus
x c [ n] = 5( −0.64) n µ[ n] + 1.5(−0.5) n µ[ n] − 0.25( n + 1)( −0.5) n µ[ n].

The first 10 samples of x c [n] obtained by evaluating this expression in MATAB are
given by
Not for sale.

177

Columns 1 through 5
6.250000000 -3.700000000
0.854485800

2.235500000

-1.373220000

Columns 6 through 10
-0.536870912
0.3396911337 -0.215996076
0.137807801
-0.0881188675
The first 10 samples of the inverse z–transform of the rational form of X c (z ) obtaining
using the M-file impz are identical to the samples given above.
(d) The inverse z –transform of X d (z ) from it partial-fraction expansion form is thus

x 4 [ n] = −5 + 0.5(−0.75) n µ[ n] − j 0.430331483(−0.375 + j 0.29047375) n µ[ n]
+ j 0.430331483(−0.375 − j 0.29047375) n µ[ n].
The first 10 samples of x d [n] obtained by evaluating this expression in MATAB are
given by
Columns 1 through 5
4.50000000000000 -0.12499999991919
0.09374999993940 0.12656249997773
0.13710937500069
Columns 6 through 10
-0.12181640625721
0.09610839844318
0.05192523193410 -0.03790274963350

-0.07136938476820

The first 10 samples of the inverse z–transform of the rational form of X d (z ) obtaining
using the M-file impz are identical to the samples given above.
M6.5 To verify using MATLAB that H 2 ( z) =

H1 ( z ) =

1 + 0.3z −1 − 0.1z −2
1.1 − 0.04 z −1 − 0.12 z − 2

is the inverse of

1.1 − 0.04 z −1 − 0.12 z −2

, we determine the first 20 samples of h1[ n] and h2 [ n] ,
1 + 0.3z −1 − 0.1z − 2
and then form the convolution of these two sequences using the M-file conv. The first
samples of the convolution result are as follows:
Columns 1 through 9
1.0000
0 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 -0.0000
Columns 10 through 18
0.0000 -0.0000 -0.0000 -0.0000 -0.0000

0.0000 -0.0000

0.0000

0.0000

Columns 19 through 21
-0.0000 -0.0000 -0.0000
M6.6 % As an example try a sequence x = 0:24;
% calculate the actual uniform dft

Not for sale.

178

%
%
%
%
%

and then use these uniform samples
with this ndft program to get the
original sequence back
[X,w] = freqz(x,1,25,’whole’);
use freq = X and points = exp(j*w)

freq = input(‘The sample values = ‘);
points = input(‘Frequencies at which samples are taken = ‘);
L = 1;
len = length(points);
val = zeros(size(1,len));
L = poly(points);
for k = 1:len
if (freq(k) ~= 0)
xx = [1 –points(k)];
[yy, rr] = deconv(L,xx);
F(k,:) = yy;
Down = polyval(yy,points(k))*(points(k))*(points(k)^(-len+1));
F(k,:) = freq(k)/down*yy;
val = val+F(k,:);
end
end
coeff = val;

Not for sale.

179

Chapter 7
7.1

⎧⎪ 1 , 0 ≤ n ≤ M − 1,
Impulse response of the moving average filter is: h[ n] = ⎨ M
⎪⎩ 0,
otherwise.
Its frequency response is: H (e jω ) =

1 ⎛⎜ 1 − e − jMω ⎞⎟ 1 sin( Mω / 2) − jMω / 2
= ⋅
e
.
M ⎜⎝ 1 − e − jω ⎟⎠ M sin(ω / 2)

Now, for a BR transfer function, H (e jω ) ≤ 1, ∀ω. For the moving-average filter,
H ( e jω ) =

1 sin( Mω / 2)
. We shall show by induction that
M sin(ω / 2)

1 sin( Mω / 2)
≤ 1, ∀ω. Now, for
M sin(ω / 2)
M = 2, H (e jω ) =

1 sin(ω)
1 2 sin(ω / 2) cos(ω / 2)
=
= cos(ω / 2) ≤ 1, ∀ω.
2 sin(ω / 2) 2
sin(ω / 2)

We assume next that

1 sin( mω / 2)
≤ 1 for 1 ≤ m ≤ M − 1. We can express
m sin(ω / 2)

1 sin( Mω / 2)
1 sin (( M − 1)ω / 2 ) cos(ω / 2) + cos(( M − 1)ω / 2 )sin(ω / 2)
=
M sin(ω / 2)
M
sin(ω / 2)

1 ⎡ sin(( M − 1)ω / 2 )
cos(ω / 2) + cos(( M − 1)ω / 2 ) ⎥. Now, cos(mω / 2 ) ≤ 1 for

M⎣
sin(ω / 2)

1 sin( Mω / 2)
1
all values of m. Hence,
≤ [M − 1 + 1] ≤ 1.
M sin(ω / 2)
M

7.2

A( z ) =

1 − d1* z
(1 − d1* z )(1 − d1z*)
2
. Therefore,
. A( z ) = A( z ) A * ( z ) =
z − d1
( z − d1 )( z * − d1* )
2

1 − A( z ) =
2

=

( z − d1 )( z * − d1* ) − (1 − d1* z )(1 − d1z*)
( z − d1 )( z * − d1* )

2

2

2

z + d 1 − d1 z * − d1* z − 1 − d 1 z + d1z * + d1* z
( z − d1 )( z * − d1* )

⎧ > 0, if z > 1,

2⎪

2

=

2

( z − 1)(1 − d 1 )
z −d1

2

. Hence,

⎧ < 1, if z > 1,

2⎪

1 − A1 ( z ) ⎨= 0, if z = 1, Thus, A1 ( z ) ⎨= 1, if z = 1, Thus, Eq. (7.20) holds
⎪ < 0, if z < 1.
⎪ > 1, if z < 1.

for any first-order allpass function. A higher-order allpass function can br factored
into a product of first-order allpass functions. Since, Eq. (7.20) holds true each of
these factors individually, hence, it also holds true for the product.

Not for sale

177

7.3

An m –th order stable, real allpass transfer function can be expressed as a product

1 − di* z
. If di is
z − di
1 − di z
. Now,
complex, then A(z ) has another factor of the form Ai' ( z ) =
z − di*
of first-order allpass transfer functions of the form Ai ( z) =

Ai (e jω ) =

1 − d *i e jω
e

− di

= e − jω

Then, Ai (e jω ) = e − jω

(1 − d *i e jω )(1 − d *i e jω )
− jω

(1 − di e
)(1 − d *i e jω )
− jθ jω 2

(1 − α e

e

)

[1 + α 2 − 2α cos(θ − ω)]

{

. Let di = di e jθ = α e jθ .

. Therefore,

}

{

}

arg{Ai (e jω )} = − ω + arg (1 − α e − jθ e jω ) 2 = − ω + 2 arg (1 − α e − jθ e jω )

⎡ α sin(θ − ω) ⎤
= − ω + 2 tan −1 ⎢
⎥. We can show similarly,
⎣1 − α cos(θ − ω) ⎦
⎡ α sin(θ + ω) ⎤
arg{ Ai' (e jω )} = − ω + 2 tan −1 ⎢
⎥.
⎣1 − α cos(θ + ω) ⎦
⎡ α sin ω ⎤
If d i = α is real, then arg{Ai (e jω )} = − ω + 2 tan −1 ⎢
⎥. Now for real di ,
⎣1 − α cos ω ⎦
arg{A i (e j 0 )} − arg{A i (e jπ )} = {−0 + 2 tan −1 (−0)} − {− π + 2 tan −1 ( −0)} = π. For

complex di , arg{A i (e j 0 )} + arg{ A i' (e j 0 )} − arg{A i (e jπ )} − arg{ A i' (e jπ )} =
⎡ α sin θ ⎤
⎡ − α sin θ ⎤
= −0 + 2 tan −1 ⎢
− 0 + 2 tan −1 ⎢

⎣ 1 − α cos θ ⎦
⎣ 1 − α cos θ ⎦
⎡ α sin θ ⎤
⎡ α sin θ ⎤
+ π − 2 tan −1 ⎢
+ π − 2 tan −1 ⎢

⎥ = 2π. Now,
⎣ 1 + α cos θ ⎦
⎣ 1 + α cos θ ⎦
d
arg{A(e jω )} . Therefore,
τ ( ω) = −

(

π

π

0

0

)

j0

∫ τ (ω)dω = − ∫ d[arg{A(e )}] = arg{A(e ) − arg{A(e ). Since
m

π

i =1

0

arg{A(e jω ) = ∑ arg{Ai (e jω ), it follows then ∫ τ (ω)dω = mπ.
7.4

h[ n] = a1δ[ n − 2] − a2 δ[ n − 1] − a3δ[ n] + a 4 δ[ n + 1] − a5δ[ n + 2]. Therefore,
H (e jω ) = a1e j 2 ω − a 2 e jω − a3 + a 4 e − jω − a 5 e −2 ω

= a1 (cos 2ω + j sin 2ω) − a1 (cos ω + j sin ω) − a3 + a 4 (cos ω − j sin ω)
− a3 + a 4 (cos ω − j sin ω) − a5 (cos 2ω − j sin 2ω)
= (a1 − a5 ) cos 2ω − (a2 − a4 ) cos ω − a3 + j ((a1 + a5 ) sin 2ω − (a2 + a 4 ) sin ω). To

Not for sale

178

have a zero-phase frequency response, the imaginary part of must be equal to zero
for all values of ω. Hence, for a zero-phase response, a1 = − a5 and a2 = − a4 .
7.5

v[ n] = x[ − n] O
* h[ n], and u[ n] = v[ − n] = x[ n] O
* h[ − n]. Hence,

y[ n] = (h[ n] + h[ − n]) O
* x[ n]. Therefore, G (e ) = H (e ) + H * (e ). Thus, the

equivalent frequency response is real and has zero phase.
7.6

Now, AM ( z) = ±

z − M D*M (1 / z*)
, where
DM ( z)

D M ( z ) = 1 + d1z −1 + d 2 z −2 + L + d M z − M .
2

AM ( z ) = AM ( z ) A*M (1 / z*) =

Consider the first-order factor 1 + az −1 . Its square-magnitude function is given by
(1 + az −1 )(1 + az )

z = e jω

= (1 + a 2 ) + 2 a cos ω. We thus rewrite the given square-

magnitude function as

H (e

2

) = H ( z)H ( z

−1

)

z = e jω

Therefore, H ( z )H ( z −1 ) =

=

9[1.0625 + 0.25( z + z −1 )][1.49 − 0.7( z + z −1 )]
[1.36 + 0.6( z + z −1 )][1.64 + 0.8( z + z −1 )]

9(1 + 0.25z )(1 + 0.25z −1 )(1 − 0.7 z )(1 − 0.7 z −1 )

(1 + 0.6 z )(1 + 0.6 z −1 )(1 + 0.8 z )(1 + 0.8 z −1 )
As can be seen, there are 4 possible causal, stable transfer functions:

(i) H ( z ) =

3(1 + 0.25z −1 )(1 − 0.7 z −1 )

(iii) H ( z ) =

(1 + 0.6 z −1 )(1 + 0.8 z −1 )

3(1 + 0.25z )(1 − 0.7z )
(1 + 0.6 z −1 )(1 + 0.8z −1 )

,

, (ii) H ( z ) =

(iv)

.

3(1 + 0.25z )(1 − 0.7 z −1 )
(1 + 0.6 z −1 )(1 + 0.8 z −1 )

H (z) =

.
z = e jω

,

3(1 + 0.25z −1 )(1 − 0.7 z )
(1 + 0.6 z −1 )(1 + 0.8 z −1 )

.

The zero locations of the four FIR transfer functions are given below:
(ii)

(i)
1

1.5
Imaginary Part

Imaginary Part

7.7

D *M (1 / z*) D M ( z )

= 1.
D M ( z ) D *M (1 / z*)

0.5
0
-0.5

1
0.5
0
-0.5
-1
-1.5

-1
-1

-0.5

0
0.5
Real Part

-4

1

Not for sale

-3

-2
-1
Real Part

0

1

179

(iii)

(iv)
1

1

Imaginary Part

Imaginary Part

1.5
0.5
0
-0.5
-1

0.5
0
-0.5

-1.5
-1
-4

-3

-2
-1
Real Part

0

1

-1

-0.5

0
0.5
Real Part

1

1.5

The zeros of the transfer function of a linear-phase FIR filter exhibits mirror-image
symmetry with respect to the unit circle. The zeros of the transfer function of a
minimum-phase FIR filter are all inside the unit circle and that of a maximum-phase
FIR filter are outside the unit circle.
(a) The transfer functions of (iii) and (iv) have linear-phase as their zeros exhibit
mirror-image symmetry.
(b) The transfer function of (i) is minimum-phase as its zeros are inside the unit
circle.
(c) The transfer function of (ii) is maximum-phase as its zeros are inside the unit
circle.
7.9

G( z ) = 6(1 + 1.7 z −1 − 2 z −2 )(1 − 0.5z −1 ) = 6(1 + 2.5z −1 )(1 − 0.8 z −1 )(1 − 0.5z −1 )
= 6 + 7.2 z −1 − 17.1 z −2 + 6 z −3 .
(a) Other transfer functions having the same magnitude responses are:
⎛ 2.5 + z −1 ⎞
⎟⎟
(i) G2 ( z ) = 6(1 + 2.5z −1 )(1 − 0.8z −1 )(1 − 0.5z −1 )⎜⎜
⎝ 1+ 2.5z −1 ⎠
= 6(2.5 + z −1 )(1 − 0.8 z −1 )(1 − 0.5z −1 ) = 15 − 13.5 z −1 − 1.8 z −2 + 2.4 z −3 .
⎛ − 0.8 + z −1 ⎞
(ii) G3 ( z ) = 6(1 + 2.5z −1 )(1 − 0.8z −1 )(1 − 0.5z −1 )⎜⎜
⎟⎟
⎝ 1− 0.8 z −1 ⎠
= 6(1 + 2.5z −1 )(−0.8 + z −1 )(1 − 0.5z −1 ) = −4.8 − 3.6 z −1 + 18 z −2 − 7.56 z −3 .
(iii)

⎛ − 0.5 + z −1 ⎞
⎟⎟
G4 ( z ) = 6(1 + 2.5z −1 )(1 − 0.8 z −1 )(1 − 0.5z −1 )⎜⎜
⎝ 1− 0.5z −1 ⎠

= 6(1 + 2.5z −1 )(1 − 0.8 z −1 )(−0.5 + z −1 ) = −3 + 0.9 z −1 + 16.28 z −2 − 124 z −3 .
(iv)

⎛ 2.5 + z −1 ⎞⎛ − 0.8 + z −1 ⎞
G5 ( z ) = 6(1 + 2.5z −1 )(1 − 0.8 z −1 )(1 − 0.5z −1 )⎜⎜
⎟⎟⎜⎜
⎟⎟
⎝ 1+ 2.5z −1 ⎠⎝ 1− 0.8 z −1 ⎠

= 6(2.5 + z −1 )(−0.8 + z −1 )(1 − 0.5z −1 ) = −12 + 16.2 z −1 + 0.9 z −2 − 3 z −3 .
Not for sale

180

(v)

⎛ − 0.8 + z −1 ⎞⎛ − 0.5 + z −1 ⎞
G6 ( z ) = 6(1 + 2.5z −1 )(1 − 0.8z −1 )(1 − 0.5z −1 )⎜⎜
⎟⎟⎜⎜
⎟⎟
⎝ 1− 0.8 z −1 ⎠⎝ 1− 0.5z −1 ⎠

= 6(1 + 2.5z −1 )(−0.8 + z −1 )(−0.5 + z −1 ) = 2.4 − 1.8 z −1 − 13.5 z −2 + 15 z −3 .
⎛ 2.5 + z −1 ⎞⎛ − 0.5 + z −1 ⎞
(vi) G7 ( z ) = 6(1 + 2.5z −1 )(1 − 0.8z −1 )(1 − 0.5z −1 )⎜⎜
⎟⎟⎜⎜
⎟⎟
⎝ 1+ 2.5z −1 ⎠⎝ 1− 0.5z −1 ⎠
= 6(2.5 + z −1 )(1 − 0.8 z −1 )(−0.5 + z −1 ) = −7.5 + 18 z −1 − 3.6 z −2 − 4.8 z −3 .
⎛ 2.5 + z −1 ⎞⎛ − 0.8 + z −1 ⎞⎛ − 0.5 + z −1 ⎞
(vii) G8 ( z ) = 6(1 + 2.5z −1 )(1 − 0.8z −1 )(1 − 0.5z −1 )⎜⎜
⎟⎟⎜⎜
⎟⎟⎜⎜
⎟⎟
⎝ 1+ 2.5z −1 ⎠⎝ 1− 0.8 z −1 ⎠⎝ 1− 0.5z −1 ⎠
= 6(1 + 2.5z −1 )(−0.8 + z −1 )(−0.5 + z −1 ) = −6 − 17.1 z −1 + 7.2 z −2 + 6 z −3 .
(b) The minimum phase filter is G2 ( z), as all its zeros inside the unit circle.
Likewise, the maximum phase filter is G6 ( z ), as all its zeros outside the unit circle.
(c) The partial energy of impulse responses of each of the above transfer functions
for different values of k are given by:
k=0

k=1

k=2

k=3

G1(z)

36

87.84

380.25

416.25

G2(z)

225

407.25

410.49

416.25

G3(z)

23.04

36

360

416.25

G4(z)

9

9.81

272.25

416.25

G5(z)

144

406.44

407.25

416.25

G6(z)

5.76

9

191.25

416.25

G7(z)

56.25

380.25

393.25

416.25

G8(z)

36.00

328.41

380.25

416.25

The partial energy remains the same for values of k ≥ 3. From the above table it
n

2

n

2

can be seen that ∑ g k [ m] ≤ ∑ g2 [ m] , and

2

m=0

m=0

2

∑ g k [ m] = ∑ g2 [ m] = 416.25, 1 ≤ k ≤ 8, where G2 ( z) is the minimum

m=0

m=0

n

2

n

2

phase transfer function. Likewise, ∑ g k [ m] ≥ ∑ g6 [ m] , 1 ≤ k ≤ 8, where
m=0

m=0

G6 ( z ) is the maximum phase transfer function.
7.9 H1 ( z ) = 1 − 0.5z −1 + 0.8z −2 − 0.4 z −3 + 0.25z −4 − 0.125z −5 + 0.2 z −6 − 0.1z −7

Not for sale

181

= (1 − 0.8z −2 )(1 + z −1 + 0.5z −2 )(1 − z −1 + 0.5z −2 )(1 − 0.5z −1 ).
H 2 ( z ) = 0.5 + 0.25z −1 + 0.4 z −2 − 0.425z −3 + 0.75z −4 − 0.75z −5 + 0.6 z −6 − 0.2 z −7
= (1 − 0.8 z −2 )(0.5 + z −1 + z −2 )(1 − z −1 + 0.5z −2 )(1 − 0.5z −1 ).
H 3 ( z ) = −0.25 + 0.25z −1 + 0.175 z −2 + 0.7 z −3 − 0.45z −4 + 0.9 z −5 − 0.6 z −6 + 0.4 z −7

= (1 − 0.8 z −2 )(0.5 + z −1 + z −2 )(1 − z −1 + 0.5z −2 )(−0.5 + z −1 ).
H 4 ( z ) = −0.5 + z −1 − 0.4 z −2 + 0.8 z −3 − 0.125z −4 + 0.25z −5 − 0.1z −6 + 0.2 z −7
= (1 − 0.8 z −2 )(1 + z −1 + 0.5z −2 )(1 − z −1 + 0.5z −2 )(−0.5 + z −1 ).
H 5 ( z ) = −0.1 + 0.2 z −1 − 0.125 z −2 + 0.25z −3 − 0.4 z −4 + 0.8 z −5 − 0.5z −6 + z −7

= (−0.8 + z −2 )(0.5 + z −1 + z −2 )(0.5 − z −1 + z −2 )(−0.5 + z −1 ) = z −7 H1 ( z −1 ).
Each factor of H1 ( z) has roots inside the unit circle. Hence, H1 ( z) is a minimumphase transfer function. Since H 5 ( z ) is a mirror-image of H1 ( z) , it has all zeros
outside the unit circle and is thus a maximum-phase transfer function.
There are 11 other length- FIR filters having the same magnitude responses as that
of the above filters.
7.11 H (e jω ) = a1 + a 2 e − jω + a3 e − j 2 ω + a 4 e − j 3ω + a 5 e − j 4 ω + a6 e − j 5ω

[

= e − j 5ω / 2 (a1e j 5ω / 2 + a6 e − j 5ω / 2 ) + (a 2 e j 3ω / 2 + a 5 e − j 3ω / 2 ) + (a3 e jω / 2 + a 4 e − jω / 2 )
−j

2 2[( a

j (a1 − a6 ) sin⎛⎜ ⎞⎟ +
⎝ 2 ⎠
ω
ω

+ (a2 + a5 ) cos⎛⎜ ⎞⎟ + j (a2 − a5 ) sin⎛⎜ ⎞⎟ + (a3 + a 4 ) cos⎛⎜ ⎞⎟ + j (a3 − a 4 ) sin⎛⎜ ⎞⎟]
⎝2⎠
⎝2⎠
⎝ 2 ⎠
⎝ 2 ⎠

=e

⎛ 5ω ⎞
1 + a6 ) cos⎜ 2 ⎟ +
⎝ ⎠

It follows from the above that H (e jω ) will have linear-phase, i.e., constant group
delay, if the imaginary parts inside the square brackets are zero. Hence, for a
constant group delay we must have a1 = a6 , a2 = a5 , and a3 = a 4 .
7.12 The frequency response of the LTI discrete-time system is given by
H (e jω ) = a1e jkω − a2 e j ( k −1)ω + a 2 e j ( k − 3)ω − a1e j ( k − 4)ω
= e − j (2 − k )ω [ a1 (e j 2ω − e − j 2ω ) + a2 (e jω − e − jω )]
= − je

− j ( 2 − k )ω

2[ a1 sin(2ω) + a2 sin ω] = − e

π
2

− j ( 2 − k − )ω

2[ a1 sin(2ω) + a 2 sin ω].

π
2

So, for k = 2 − , the system will have a frequency response that is a real function
of ω .

Not for sale

182

]

7.13 H 1(e jω ) = α + β e − jω and H 2 (e jω ) =
H (e jω ) = H 1(e jω )H 2 (e jω ) =

1
1 + γ e − jω

α + β e − jω

1 − γe − jω

. Thus,

.

α + β e − jω α + β e jω α 2 + β 2 + 2αβ cos ω

=
= K2
2
− jω

1 − γe
1 − γe
1 + γ − 2 γ cos ω

2

H ( e j ω ) = H ( e jω ) H ( e − j ω ) =

if α 2 + β 2 = K 2 (1 + γ 2 ) and αβ = − K 2 γ, i.e., (α + β ) 2 = K 2 (1 − γ ) 2 .
α

7.14 Y (e jω ) = X (e jω ) e j arg X (e

jω )

. Hence, H (e jω ) =

Y ( e jω )
X (e jω )

= X ( e jω )

(α −1)

. Since

H (e jω ) is a real function of ω, it has zero phase.

(

)

7.15 H (e jω ) = h[0](1 + e − j 2ω ) + h[1]e − jω = e − jω h[0](e jω + e − jω ) + h[1]

= e − jω (2h[0] cos ω + h[1]). Thus, we require H (e j 0.3 ) = 2h[ 0] cos(0.3) + h[1] = 1
and H (e j 0.6 ) = 2h[ 0] cos(0.6) + h[1] = 0. Solving these two equations we get
h[0] = 3.8461 and h[1] = −6.3487.
7.16 (a) H (e jω ) = h[0] + h[1]e − jω − h[1]e − j 3ω − h[0]e − j 4ω

(

= e − j 2ω h[0](e j 2ω − e − j 2ω ) + h[1](e jω − e − jω )

)

= j 2e − j 2ω (h[0] sin(2ω) + h[1] sin(ω)). Therefore,
H (e jω ) = 2(h[ 0] sin( 2ω) + h[1] sin( ω)). Thus,
H (e j 0.3 ) = 2(h[0] sin(0.6) + h[1] sin(0.3)) = 0.3 and
H (e j 0.6 ) = 2(h[ 0] sin(1.2) + h[1] sin(0.6)) = 0.8. Solving these two equations we get

h[1] = −0.9873 and h[0] = 2.5686.
(b) H (e jω ) = j 2e − j 2ω (2.5686 sin(2ω) − 0.9873 sin(ω)). The plot of the magnitude
and phase responses are shown below.
7.17 (a) H (e jω ) = h[0] + h[1]e − jω − h[1]e − j 2ω − h[0]e − j 3ω

(

= e − j 3ω / 2 h[0](e j 3ω / 2 − e − j 3ω / 2 ) + h[1](e jω / 2 − e − jω / 2 )
= j 2e − j 3ω / 2 (h[0] sin(3ω / 2) + h[1] sin(ω / 2)). Therefore,

)

H (e jω ) = 2(h[ 0] sin(3ω / 2) + h[1] sin(ω / 2)). Thus,

Not for sale

183

H (e j 0.25 ) = 2(h[ 0] sin(0.375) + h[1] sin( 0.125)) = 0.2 and
H (e j 0.8 ) = 2(h[0] sin(1.2) + h[1] sin(0.4)) = 0.8. Solving these two equations we get

h[1] = 6.2573 and h[0] = −1.8569.
(b) H (e jω ) = j 2e − j 3ω / 2 (− 1.8569 sin(3ω / 2) + 6.2573 sin(ω / 2)).
7.18 (a) H (e jω ) = h[0] + h[1]e − jω + h[2]e − j 2 ω + h[1]e − j 3ω + h[0]e − j 4ω

(

)

= e − j 2ω h[0](e j 2ω + e − j 2ω ) + h[1](e jω + e − jω ) + h[2]
=e

− j 2ω

(2h[0] cos(2ω) + 2h[1] cos ω + h[2]) .

Therefore,

H (e jω ) = 2h[ 0] cos(2ω) + 2h[1] cos ω + h[ 2]. Thus,

H (e j 0.3 ) = 2 h[0] cos(0.6) + 2 h[1] cos(0.3) + h[2] = 1,
H (e j 0.5 ) = 2 h[0] cos(1.0) + 2 h[1] cos(0.5) + h[2] = 0, and
H (e j 0.8 ) = 2 h[0] cos(1.6) + 2 h[1] cos(0.8) + h[2] = 1. Solving these three equations

we get h[0] = 17.7761, h[1] = −58.7339 , and h[2] = 83.8786 .
(b) H (e jω ) = e − j 2ω (35.5522 cos(2ω) − 117.4677 cos ω + 83.8786 ). The plot of the
magnitude and phase responses are shown below.
4

2

Magnitude

1.5
1
0.5
0

0

0.2

0.4

0.6

0.8

1

ω

2
0
-2
-4

0

0.5

1

1.5

2

2.5

3

ω

7.19 H (e j 0 ) = 13, H (e j 3π / 4 ) = −3 − j 4, and H (e jπ ) = −3. Using the symmetry
property of the DTFT of a real sequence, we observe
H (e jπ / 4 ) = H * (e j 3π / 4 ) = −3 + j 4. Thus, the 4 –point DFT H [k ] of the length- 4

Not for sale

184

sequence {h[n]} is given by {H[ k ]} = {13, − 3 + j 4, − 3, − 3 − j 4}. Its 4 –point
1
1 ⎤ ⎡ 13 ⎤ ⎡ 1 ⎤
⎡h[0]⎤
⎡1 1
⎢ h[1] ⎥
⎢1 j − 1 − j ⎥ ⎢ − 3 + j 4⎥ ⎢ 2 ⎥
⎥= 1⎢
⎥⎢
⎥ = ⎢ ⎥.
inverse DFT is thus given by ⎢
⎢h[2]⎥ 4 ⎢1 − 1 1 − 1⎥ ⎢ − 3 ⎥ ⎢ 4 ⎥

⎥⎢
⎥ ⎢ ⎥
⎣ h[3]⎦
⎣1 − j − 1 j ⎦ ⎣ − 3 − j 4 ⎦ ⎣6 ⎦
Hence, H ( z ) = 1 + 2 z −1 + 4 z −2 + 6 z −3 .
7.20 Now, for a real, anti-symmetric sequence {h[ n]} of even length, H (e j 0 ) = 0.
Using the symmetry property of the DTFT of a real sequence, we observe also
H (e jπ / 4 ) = H * (e j 3π / 4 ) = −5 + j 5. Hence, the 4 –point DFT of the length- 4
sequence {h[ n]} is given by {H[ k ]} = {0, − 5 + j 5, 20, − 5 − j 5}. Its 4 –point
⎡h[0]⎤
1
1 ⎤ ⎡ 0 ⎤ ⎡ 2 .5 ⎤
⎡1 1
⎢ h[1] ⎥ 1 ⎢1 j − 1 − j ⎥ ⎢− 5 + j 5⎥ ⎢− 7.5⎥
inverse DFT is thus given by ⎢
=
.
=
h[2]⎥ 4 ⎢1 − 1 1 − 1⎥ ⎢ 20 ⎥ ⎢− 2.5⎥
⎢ h[3]⎥
⎢⎣1 − j − 1 j ⎥⎦ ⎢⎣ − 5 − j 5⎥⎦ ⎢⎣ 7.5 ⎥⎦

Hence, H ( z ) = 2.5 − 7.5 z −1 − 2.5 z −2 + 7.5 z −3 .
7.21 (a) H A (e jω ) = 0.5 − e − jω + 0.5e − j 2ω = e − jω (cos ω − 1) and
H B (e jω ) = 0.5 + e − jω + 0.5e − j 2ω = e − jω (cos ω + 1). Hence,
H A (e jω ) = cos ω − 1 and H B (e jω ) = cos ω + 1 . Plots of these two magnitude

functions is given below:

HB(e )

2

2

1.5

1.5
Magnitude

Magnitude

HA(e )

1

0.5

0.5
0

1

0

0.2

0.4

0.6

0.8

1

0

0

0.2

0.4

0.6

0.8

1

ω/ π

ω/ π

As can be seen from the above plots, h A [n] is a highpass filter and hB [n] is a
lowpass filter.
(b) hC [ n] = ( −1) n h A [ n] = h B [ n]. Hence, H C (e jω ) = H A (e j (ω + π ) ) = H B (e jω ).
HC (e jω ) is a shifted version of H A (e jω ) shifted by π radians.

Not for sale

185

7.22 h[ n] = h[ N − n]. Thus,
N −1

N −1

N −1

n=0

n=0

k =0

H ( z ) = ∑ h[ n]z − n = ∑ h[ N − n]z − n = ∑ h[ k ]z − ( N − k ) = z − N H ( z −1 ). As
H ( z ) has zeros in a mirror-image symmetry in the z –plane, G ( z ) = 1 / H ( z ) will
have poles outside the unit circle making it unstable.

7.23 (a) Since H ( z ) is a minimum-phase FIR transfer function, any other FIR transfer
function G ( z ) with the same magnitude response as that of H ( z ) can be expressed
as G ( z ) = H ( z ) A( z ) where A( z ) is an allpass function. Now, g[ 0] = lim G ( z ).
z→∞

Thus, g[0] = | lim G( z) |= | lim H ( z) A( z) |= | lim H ( z) | ⋅ | lim A( z ) |
z→∞

z →∞

z →∞

z→∞

≤ | lim H ( z ) |, as | lim A( z ) | < 1 (see Eq. (7.20)). Hence, g[0] ≤ h[0] .
z→∞

z→∞

(b) If λ1 is a zero of H ( z ), then λ1 < 1, since H ( z ) is a minimum-phase causal
stable transfer function with all zeros inside the unit circle. Let
H ( z ) = B( z )(1 − λ1 z −1 ). It follows that is also a minimum-phase causal transfer
function.
⎛ λ* − z −1 ⎞
⎟. If
Now, consider the transfer function F ( z) = B( z )( λ*1 − z −1 ) = H ( z )⎜ 1
⎜ 1 − λ z −1 ⎟
1

h[ n], b[ n], and f [ n] denote, respectively, the inverse –transform of H ( z ), B( z ), and
F ( z ), then we get
⎧⎪
b[0],
n = 0,

λ*1b[0],
n = 0,
h[ n] = ⎨
and f [ n] = ⎨ *
⎪⎩ λ1b[ n] − b[ n − 1], n ≥ 1.
⎩b[ n] − λ1b[ n − 1], n ≥ 1,
Consider
m

2

m

2

2

ε = ∑ h[ n] − ∑ f [ n] = b[0] − λ*1
n=0
2

h[ n] =
2

f [ n] =
Hence, ε

2

n=0

2

m

2

m

2

b[0] + ∑ h[ n] − ∑ f [ n] . Now,

n =1
n =1
2
2
2
b[ n] + λ1 b[ n − 1] − λ1 b[ n − 1]b * [ n] − λ*1 b * [ n − 1]b[ n], and
2
2
2
λ1 b[ n] + b[ n − 1] − λ1 b[ n − 1]b * [ n] − λ*1b * [ n − 1]b[ n].
m
2
2
2
2
2
2
= b[0] − λ1 b[0] + ∑ ⎛⎜ b[ n] − λ1 b[ n − 1] ⎞⎟

n =1
m
2
2
2
2
2
− ∑ ⎛⎜ λ1 b[ n] − b[ n − 1] ⎞⎟ = ⎛⎜ 1 − λ1 ⎞⎟ b[ m] .

n =1
m
m
m
m
2
2
2

2

Since λ1 < 1, ε > 0, i.e., ∑ h[ n] > ∑ f [ n] . Hence, ∑ h[ n] ≥ ∑ g[ n] .
n=0

n=0

Not for sale

n=0

n=0

186

7.24 (a) H1 ( z ) =

z 3 + 3z 2 + 2 z + 7
(2 z + 3)( z 2 + 0.5 z + 0.8)

=

( z + 3.0867)( z 2 − 0.0867 z + 2.2678)
(2 z + 3)( z 2 + 0.5 z + 0.8)

has a pole at

3
2

z = − , which is outside the unit circle. Hence, H1 ( z) is not a stable transfer
function. To arrive at a stable transfer function with an identical magnitude
2z+3

resulting in
response, we multiply H1 ( z) with the allpass function
2 + 3z
z 3 + 3z 2 + 2 z + 7

2z + 3
z 3 + 3z 2 + 2 z + 7
.
=
(2 z + 3)( z 2 + 0.5 z + 0.8) 2 + 3 z (2 + 3 z )( z 2 + 0.5 z + 0.8)

(b) H 2 ( z ) =

4 z 3 − 2 z 2 + 5z − 6
(1.5 z 2 + 3 z − 5)( z 2 − 0.3 z + 0.7)

=

4( z 2 + 0.4181+ 1.6338)( z − 0.9181)
1.5( z + 3.0817)( z −1.0817)( z 2 − 0.3 z + 0.7)

has

two poles outside the unit circle at z = −3.0817 and z = 1.0817 . Hence, H 2 ( z) is
not a stable transfer function. To arrive at a stable transfer function with an
identical magnitude response, we multiply H 2 ( z) with the allpass function
z 2 − 0.3 z + 0.7
0 .7 z 2 − 0 .3 z + 1

=

resulting in

4 z 3 − 2 z 2 + 5z − 6

z 2 − 0.3 z + 0.7

(1.5z 2 + 3z − 5)( z 2 − 0.3z + 0.7) 0.7 z 2 − 0.3 z + 1

4 z 3 − 2 z 2 + 5z − 6
(5 z 2 + 3 z −1.5)( z 2 − 0.3z + 0.7)

.

7.25 The transfer function of the simplest notch filter is given by
H ( z ) = (1 − e jωo z −1 )(1 − e − jωo z −1 ) = 1 − 2 cos(ωo ) z −1 + z − 2 . In the steady-state,
the output for an input x[ n] = cos(ωo n) is given by y[ n] = H (e
where θ (ωo ) = arg[ H (e

jω o

jω o

) cos(ωo n + θ (ωo ))

)].

(a) Comparing H ( z ) = 1 − z −1 + z −2 with H ( z ) as given above we conclude
cos(ωo ) = , or ωo = . Here, H1 (e jω o ) = 1 − e − jω o + e − j 2ω o
2
3
π

1

= 1 − e − jπ / 3 + e − j 2π / 3 = 1 − (0.5 + j 0.866) + (−0.5 + j 0.866) = 0.
Hence, y[ n] = H1 (e jωo ) cos(ωo n + θ (ωo )) = 0.
(b) Comparing H 2 ( z ) = 1 − 0.8 z −1 + z −2 with H ( z ) as given above we conclude
cos(ωo ) = 0.4 or ωo = 0.369π. Here,
H 2 (e jω o ) = 1 − e − jω o + e − j 2ω o = 1 − (0.32 + j 0.7332 ) + (−0.68 + j 0.7332 ) = 0.
Hence, y[ n] = H 2 (e

jω o

) cos(ωo n + θ (ωo )) = 0. .

Not for sale

187

(c) Comparing H 3 ( z ) = 1 − 1.6 z −1 + z −2 with H ( z ) as given above we conclude
cos(ωo ) = 0.8 or ωo = 0.2048π. Here,
H 2 (e jω o ) = 1 − e − jω o + e − j 2ω o = 1 − (1.2891 + j 0.9599 ) + (0.2801 + j 0.9599 ) = 0.
Hence, y[ n] = H 3 (e

jω o

) cos(ωo n + θ (ωo )) = 0.

Y (z)
Y (z)
7.26 From the figure, H 0 ( z ) = 0
= G L ( z )G H ( z 2 ), H1 ( z ) = 1
= G H ( z )G H ( z 2 ),
X ( z)
X ( z)
Y ( z)
Y ( z)
H 2 ( z) = 2
= G H ( z )G L ( z 2 ), H 3 ( z ) = 3
= G L ( z )G L ( z 2 ).
X ( z)
X (z)
For the magnitude responses of G L (z) and G H (z) shown below
GH(e jω)

G L(e jω)

1

1

0

π/4

0

π/2

π

ω

0

π/2

0
2

3π/4

π

ω

the magnitude responses of G L ( z 2 ) and G H ( z ) are as shown below
G L(e

j2ω

GH(e j2ω )

)

1

1

0

π/4

0

π/2

π

ω

0

π/2

0

3π/4

π

ω

Hence, the magnitude responses of the casacaded filters H i ( z ), 0 ≤ i ≤ 3, are as indicated
below:
H 0(e jω )

H 1(e jω )
1

1

0

0

π/4

π/2

π

ω

π/2

0

3π/4

π

ω

H 3(e jω )

H 2(e jω )

1

1

0

0

0

π/4

π/2

π

ω

0

Not for sale

0

π/4

π/2

π

ω

188

⎧1, 0 ≤ ω < ωc ,
7.27 G(e jω ) = H LP (e j (π − ω) ). Since, H LP (e jω ) = ⎨
⎩0, ωc ≤ ω < π,
⎧0, 0 ≤ ω < π − ωc ,
G ( e jω ) = ⎨
Hence, G(e jω ) is a highpass filter with a cutoff

<
1
,
π
ω
ω
π
.
c

frequency given by ω o = π − ω c . Also, since G( z) = H LP (− z), we have
g[ n] = (−1) n hLP [ n].
7.28 G( z ) = H LP (e jωo z ) + H LP (e − jωo z ). Hence, g[ n] = hLP [ n]e − jωo n + hLP [ n]e jωo n
= 2 h LP [ n ] cos( ω o n ). Thus, G ( z ) is a real coefficient bandpass filter with
a center frequency at ωo and a passband width of 2ω p .
7.29 F ( z ) = H LP (e jωo z ) + H LP (e − jωo z ) + H LP ( z ). Hence,
f [ n] = (1 + 2 cos(ωo n))hLP [ n]. Thus, F ( z ) is a real coefficient bandstop filter with
a center frequency at ωo and a stopband width of (π − 3ω p ) / 2.
7.30
V(z)

+

X(z)

hLP[n]

U(z)

+

Y(z)

( _1)n

( _1)n

From the above figure, we get V ( z) = X (− z), U ( z) = H LP ( z) X (− z), and
Y ( z)
Y ( z) = U (− z) = H LP (− z) X ( z). Hence, H eq ( z ) =
= H LP (− z ), which is the
X (z)
highpass filter of Problem 7.27.
7.31
2

V 0(z)
cos(ωo n) +

cos(ωo n)

+

U1 (z)
2

hLP[n]

Y(z)

V 1(z)

+

X(z)

hLP[n]

+

+

U0 (z)

sin(ωo n)

sin(ωo n)

From the above figure, we have u0 [ n] = 2 x[ n] cos(ωo n) = x[ n] e jω o n + x[ n] e − jω o n
or, U 0 (e jω ) = X (e j (ω + ω o ) ) + X (e j (ω − ω o ) ). Likewise,
U1 (e jω ) = j X (e j (ω + ω o ) ) − j X (e j (ω − ω o ) ). We also have

Not for sale

189

V0 ( z ) = H LP ( z ) X ( z e jω o ) + H LP ( z ) X ( z e − jω o ), and

V1 ( z ) = j H LP ( z ) X ( z e jω o ) − j H LP ( z ) X ( z e − jω o ). Therefore,
Y ( z) =

1
2

(V (z e
0

jω o

) (V (z e

) + V0 ( z e − jω o ) + j

1
2

1

jω o

)

) − V1 ( z e − jω o ) , which after

simplification yields

{H LP (z e jωo ) X (z e j 2ωo ) + H LP (z e jωo ) X (z) + H LP (z e − jωo ) X (z)
1
+ H LP ( z e − jω o ) X ( z e − j 2ω o )} − {H LP ( z e jω o ) X ( z e j 2ω o ) − H LP ( z e jω o ) X ( z )
4
− jω o
− jω o
+ H LP ( z e
) X ( z ) + H LP ( z e
) X ( z e − j 2ω o )}. Hence,

Y ( z) =

1
4

Y ( z) =

1
2

{H

H eq ( z ) =

1
2

LP ( z e

{H

jω o

LP ( z e

}

) + H LP ( z e − jω o ) X ( z ). Therefore,

jω o

}

) + H LP ( z e − jω o ) . Thus, the structure of Figure P7.4

implements the bandpass filter of Problem 7.28.

7.32

Not for sale

190

|H 0(e jω )|

π 0 __
π
_ __
M
M
|H 1(e jω )|

0

__
M

π

π

π

π

ω

ω

|H 2(e jω )|

0

__
M

ω

|HM _ 1(e jω )|

_ 2π
__
M

0

2( M _ 1)π
_____
M

ω

The output of the k –th filter is a bandpass signal occupying a bandwidth of 2π / M
and centered at ω = kπ / M. In general, the k –th filter H k (z ) has a complex
impulse response generating a complex output sequence. To realize a real
coefficient bandpass filter, one can combine the outputs of H k (z ) and H M − k (z ).
7.33 H 0 ( z ) = (1 + z −1 ). Thus, H 0 (e jω ) = cos(ω / 2). Now, G ( z ) = (H 0 ( z )) M .
1
2

2

Hence, G(e jω ) = H 0 (e jω )

2M

= (cos(ω / 2))2 M . The 3 –dB cutoff frequency

ωc of G ( z ) is thus given by (cos(ωc / 2))2 M = . Hence, ωc = 2 cos −1 (2 −1 / 2 M ).
1
2

2

7.34 H1 ( z ) = (1 − z −1 ). Thus, H1 (e jω ) = sin 2 (ω / 2). Let, F ( z ) = (H 0 ( z )) M . Then,
2
1

2

F (e jω ) = (sin(ω / 2))2 M . The 3 –dB cutoff frequency ωc of F ( z ) is thus given

Not for sale

191

by (sin(ωc / 2))2 M = , which yields, ωc = 2 sin −1 (2 −1 / 2 M ).
2
1

7.35 H LP ( z ) =

1 − α ⎛⎜ 1 + z −1 ⎞⎟
. Note that H LP (z) is stable if α < 1. Now,
2 ⎜⎝ 1 − αz −1 ⎟⎠

1 − sin(ωc ) cos 2 (ωc / 2) + sin 2 (ωc / 2) − 2 sin(ωc / 2) cos(ωc / 2)
α=
=
cos(ωc )
cos 2 (ωc / 2) − sin 2 (ωc / 2)

=

cos(ωc / 2) − sin(ωc / 2) 1 − tan(ωc / 2)
=
.
cos(ωc / 2) + sin(ωc / 2) 1 + tan(ωc / 2)

(7-a)

If 0 ≤ ω < π, then tan(ωc / 2) ≥ 0. Hence, α < 1.
7.36 From Eq. (7-a), α =

1 − tan(ωc / 2)
1− α
. Hence, tan(ωc / 2) =
.
1+ α
1 + tan(ωc / 2)

1 − sin(0.6)
= 0.5275. Substituting this value of α
cos(0.6)

1 + z −1
⎟.
in Eq.(7.71), we arrive at H LP ( z ) = 0.2363⎜
⎜ 1 − 0.5275 z −1 ⎟

7.37 (a) From Eq.(7.73b), we get α =

1 − sin(0.45)
= 0.0787. Substituting this value of
cos(0.45)

1 + z −1
⎟.
α in Eq.(7.71), we arrive at H LP ( z ) = 0.4607⎜
⎜ 1 − 0.0787 z −1 ⎟

1 + α ⎛⎜ 1 − z −1 ⎞⎟
1 + α ⎛⎜ 1 − e − jω ⎞⎟
7.38 H HP ( z ) =
. Thus, H HP (e jω ) =
. Thus,
2 ⎜⎝ 1 − αz −1 ⎟⎠
2 ⎜⎝ 1 − αe − jω ⎟⎠
2 ⎛ 1 + α ⎞2 ⎛
2 − 2 cos(ω) ⎞⎟
H HP (e jω ) = ⎜
. At –dB cutoff frequency ωc ,
⎟ ⎜⎜
⎝ 2 ⎠ ⎝ 1 + α 2 − 2α cos(ω) ⎟⎠
2
1

H HP (e jωc ) = which yields cos(ωc ) =
.
2
1 + α2
1 − sin(0.6)
7.39 (a) From Eq.(7.73b), we get α =
= 0.5275. Substituting this value of
cos(0.6)
⎛ 1 + z −1

⎟.
α in Eq.(7.74), we arrive at H HP ( z ) = 0.7637⎜
⎜ 1 − 0.5275z −1 ⎟

1 − sin(0.45)
(b) From Eq.(7.73b), we get α =
= 0.0787. Substituting this value of
cos(0.45)
(b) From Eq.(7.73b), we get α =

Not for sale

192

⎛ 1 + z −1

⎟.
α in Eq.(7.74), we arrive at H HP ( z ) = 0.5394⎜
⎜ 1 − 0.0787 z −1 ⎟

7.40 H ( z) =

1 − z −1
1 − kz

−1

(1 − cos ω) 2 + sin 2 ω

2

. Hence, H (e jω ) =

2

Now, H (e jπ ) =

4
(1 + k ) 2

2

2

2

(1 − k cos ω) + k sin ω

=

2 − 2 cos ω
1 + k 2 − 2 k cos ω

. Thus, the scaled transfer function is given by

1 + k ⎛⎜ 1 − z −1 ⎞⎟
. A plot of the magnitude responses of the scaled transfer
H (z) =
2 ⎜⎝ 1 − kz −1 ⎟⎠

function for k = 0.95, 0.9, and − 0.5 are given below:
1

←k=0.95
←k=0.9

Magnitude

0.8

k=-0.5→

0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/pi

7.41 H BP ( z ) =

1 − α ⎛⎜
1 − z−2
⎟. Thus,
2 ⎜⎝ 1 − β (1 + α )z −1 + αz − 2 ⎟⎠

H BP (e jω ) =
H BP (e jω )
=

2

1 − α ⎛⎜
1 − e − j 2ω
⎟. Hence,
2 ⎜⎝ 1 − β (1 + α )e − jω + αe − j 2ω ⎟⎠
=

(1 − α ) 2
2(1 − 2 cos(2ω))

2
1 + β 2 (1 + α ) 2 + α 2 + 2α cos(2ω) − 2 β (1 + α ) 2 cos(ω)

(1 − α) 2 sin 2 ω
(1 + α) 2 (cos ω − β) 2 + (1 − α ) 2 sin 2 ω

. At the center frequency ω o ,

2

H BP (e jωo ) = 1. Hence, (cosω o − β ) 2 = 0 or cosω o = β.
2

At the 3 –dB bandedges ω1 and ω2 , H BP (e jωi ) = , i = 1, 2. This imples
(1 + α ) 2 (cos ωi − β) 2 = (1 − α ) 2 sin 2 ωi ,

1
2

(7-c)

⎛1+ α ⎞
or, sin ωi = ±⎜
⎟ (cos ωi − β), i = 1, 2. Since, ω1 < ωo < ω2 , sin ω1 must have
⎝1− α ⎠
positive sign and sin ω2 must have negative sign, because otherwise, sin ω2 < 0

Not for sale

193

.

for ω2 in (0, π). Now, Eq. (7-c) can be rewritten as
2(1 + α 2 ) cos 2 ωi − 2β(1 + α) 2 cos ωi + β 2 (1 + α) 2 − (1 − α 2 ) = 0. Hence,

(1 + α) 2

β 2 (1 + α) 2 − (1 − α) 2
, and (cos ω1 )(cos ω2 ) =
.
2(1 + α 2 )
1 + α2
Denote ω3dB = ω2 − ω1. Then cos ω3dB = cos ω2 cos ω1 + sin ω2 sin ω1

cos ω1 + cos ω2 = β

2

(

)

⎛1+ α ⎞
2
.
= cos ω2 cos ω1 − ⎜
⎟ cos ω2 cos ω1 + β (cos ω2 + cos ω1 ) =
⎝1− α ⎠
1 + α2

7.42 (a) Using Eq.(7.76), we get β = cos( 0.55π) = −0.1564 . Next, from Eq.(7.78), we

get
= cos(0.25π) = 0.7071, or, equivalently, 0.7071α 2 − 2α + 0.7071 = 0.
2
1+ α
Solution of this quadratic equation yields α = 2.4142 and α = 0.4142.
Substituting α = 2.4142 and β = −0.1564 . in Eq.(7.77) we arrive at the
denominator polynomial of the transfer function H BP (z) as
D( z ) = 1 + 0.5340 z −1 + 2.4142 z −2 . Comparing with Eq. (7.136) we note
d1 = 0.5340 and d 2 = 2.4142. Since the condition of Eq. (7.139) is not satisfied,
the corresponding H BP (z) is unstable.
Substituting α = 0.4142 and β = −0.1564 . in Eq.(7.77) we arrive at the
denominator polynomial of the transfer function H BP (z) as
D( z ) = 1 + 0.2212 z −1 + 0.4142 z −2 . Comparing with Eq. (7.136) we note
d1 = 0.2212 and d 2 = 0.4142. Since the conditions of Eqs. (7.139) and (7.141) are
satisfied, the corresponding H BP (z) is a stable transfer function. Hence, the desired
transfer function is H BP ( z) =

0.2929(1 − z −2 )

1 + 0.2212 z −1 + 0.4142 z − 2

.

(b) Using Eq.(7.76), we get β = cos( 0.3π) = 0.5878 . Next, from Eq.(7.78), we get

= cos(0.3π) = 0.5878, or, equivalently, 0.5878 α 2 − 2α + 0.5878 = 0.
2
1+ α
Solution of this quadratic equation yields α = 3.0766 and α = 0.3249.
Substituting α = 3.0766 and β = 0.5878. in Eq.(7.77) we arrive at the denominator
polynomial of the transfer function H BP (z) as D( z ) = 1 + 2.3968 z −1 + 0.788 z −2 .
Comparing with Eq. (7.136) we note d1 = 2.3968 and d 2 = 0.788. Since the
condition of Eq. (7.141) is not satisfied, the corresponding H BP (z) is unstable.
Substituting α = 0.3249 and β = 0.5878. in Eq.(7.77) we arrive at the denominator
polynomial of the transfer function H BP (z) as D( z ) = 1 + 0.7788 z −1 + 0.3249z −2 .
Not for sale

194

Comparing with Eq. (7.136) we note d1 = 0.2212 and d 2 = 0.4142. Since the
conditions of Eqs. (7.139) and (7.141) are satisfied, the corresponding H BP (z) is a
stable transfer function. Hence, the desired transfer function is

0.3376(1 − z −2 )

H BP ( z ) =

.
1 + 0.7788z −1 + 0.3249z − 2
1 + α ⎛⎜ 1 − 2βz −1 + z − 2
7.43 H BS ( z ) =
2 ⎜⎝ 1 − β(1 + α)z −1 + αz − 2
H BS (e
=

⎟. Thus,

2

2 + 4β 2 − 8β cos(ω) + 2 cos(2ω)
⎛1+ α ⎞
) =⎜
⎟ ⋅
⎝ 2 ⎠ 1 + β 2 (1 + α) 2 + α 2 + 2α cos(2ω) − 2β(1 + α) 2 cos(ω)
2

(1 + α) 2 (cos ω − β) 2
(1 + α) 2 (cos ω − β) 2 + (1 − α ) 2 sin 2 ω

. At the center frequency ω o ,

2

H BS (e jωo ) = 0. Hence, (cosω o − β ) 2 = 0 or cosω o = β. At the 3 –dB
2

bandedges ω1 and ω2 , H BP (e jωi ) = , i = 1, 2. This leads to Eq. (7-c) given in
2
the solution of Problem 7.41. Hence, as in the solution of Problem 7.41,

.
ω3dB =
1 + α2
1

7.44 (a) Using Eq.(7.76), we get β = cos( 0.35π) = 0.454 . Next, from Eq.(7.78), we get

= cos(0.2 π) = 0.809, or, equivalently, 0.809 α 2 − 2α + 0.809 = 0. Solution
2
1+ α
of this quadratic equation yields α = 1.9627 and α = 0.3249. Substituting
α = 1.9627 and β = 0.454 in Eq.(7.80) we arrive at the denominator polynomial of
the transfer function H BS (z ) as D( z ) = 1 − 1.3451 z −1 + 1.9627 z −2 . Since the
condition of Eq. (7.139) is not satisfied, the corresponding H BS (z ) is unstable.
Substituting α = 0.3249 and β = 0.454 in Eq.(7.80) we arrive at the denominator
polynomial of the transfer function H BS (z ) as D( z ) = 1 + 0.6015 z −1 + 0.3249 z −2 . Since
the conditions of Eqs. (7.139) and (7.141) are satisfied, the corresponding H BS (z ) is a
stable transfer function. Hence, the desired transfer function is
H BS ( z ) =

0.6624(1 − 0.908 z −1 + z −2 )
1 + 0.6015 z −1 + 0.3249 z − 2

.

(b) Using Eq.(7.76), we get β = cos( 0.6 π) = −0.309. Next, from Eq.(7.78), we get

= cos(0.15π) = 0.891, or, equivalently, 0.891α 2 − 2α + 0.891 = 0. Solution
2
1+ α
Not for sale

195

of this quadratic equation yields α = 1.6319 and α = 0.6128 .
Substituting α = 1.6319 and β = −0.309 in Eq.(7.80) we arrive at the denominator
polynomial of the transfer function H BS (z ) as D( z ) = 1 + 2.345 z −1 + 1.6319 z −2 .
Since the condition of Eq. (7.139) is not satisfied, the corresponding H BS (z ) is
unstable.
Substituting α = 0.6128 and β = −0.309 in Eq.(7.80) we arrive at the denominator
polynomial of the transfer function H BS (z ) as D( z ) = 1 + 0.4974 z −1 + 0.6128 z −2 .
Since the conditions of Eqs. (7.139) and (7.141) are satisfied, the
corresponding H BS (z ) is a stable transfer function. Hence, the desired transfer
function is H BS ( z ) =
7.45

0.564(1 + 0.618 z −1 + z −2 )
1 + 0.4984 z −1 + 0.6128 z − 2

(1 − α) 2 (1 + cos ωc )
2

2(1 + α − 2α cos ωc )

.

= 2 −1 / K . Let C = 2 ( K −1) / K . Simplifying the first equation

we then get α 2 (cosωc +1 − C ) − 2α(1 + cosωc −C cosωc ) + 1 + cosωc −C = 0.
Solving the quadratic equation for α we obtain
α=

=
=

2(1 + (1 − C ) cos ωc ) ± 4(1 + (1 − C ) cos ωc )2 − 4(1 + cos ωc − C ) 2

(1 + (1 − C ) cos ωc ) ±

2(1 + cos ωc − C )

(2 + 2 cos ωc − C − C cos ωc )(C (1 − cos ωc ) )
1 + cos ωc − C

1 + (1 − C ) cos ωc ± sin ωc 2C − C 2
. For stability we require α < 1, hence the
1 + cos ωc − C

desired solution is α =

1 + (1 − C ) cos ωc − sin ωc 2C − C 2
.
1 + cos ωc − C
2

2 ⎛ 1 + α ⎞ 2 1 − e − jω
1 + α ⎛⎜ 1 − z −1 ⎞⎟
. H HP (e jω ) = ⎜
7.46 H HP ( z ) =
.

2 ⎜⎝ 1 − αz −1 ⎟⎠
⎝ 2 ⎠ 1 − α e − jω

H HP (e

)

2K

⎛1+ α ⎞
=⎜

⎝ 2 ⎠

2K

1 − e − jω
1 − αe

2K

− jω

the –dB cutoff frequency ωc , H HP (e jω c )

⎛1+ α ⎞
=⎜

⎝ 2 ⎠
2K

2K

2 K (1 − cos ω) K
2

(1 + α − 2α cos ω)

K

. At

= . Let C = 2 ( K −1) / K .
1
2

Simplifying the above equation we get
α 2 (1 − cosωc −C ) − 2α(1 − cosωc +C cosωc ) + 1 − cosωc −C = 0. Solving the
quadratic equation for α we obtain

Not for sale

196

α=

− 2(1 − cos ωc + C cos ωc ) ± 2 (1 − cos ωc + C cos ωc ) 2 − (1 − cos ωc − C ) 2
2(1 − cos ωc − C )

.

For stability we require α < 1, hence the desired solution is

sin ωc 2C − C 2 − (1 − cos ωc + C cos ωc )
.
α=
1 − cos ωc − C
K
1
7.47 (a) Analyzing Figure P7.6(a) we get Y ( z ) = ⎡ (1 + A1 ( z )) + (1 − A1 ( z ))⎤ X ( z ).
⎢⎣ 2
⎥⎦
2
Y (z) ⎛ 1 + K ⎞ ⎛ 1 − K ⎞
Hence, H ( z ) =
=⎜
⎟+⎜
⎟A1 ( z ).
X (z) ⎝ 2 ⎠ ⎝ 2 ⎠
⎛1+ K ⎞
⎛1− K ⎞
(b) Analyzing Figure P7.6(b) we get Y ( z ) = ⎜
⎟ X ( z) + ⎜
⎟A1 ( z ) X ( z ).
⎝ 2 ⎠
⎝ 2 ⎠
Y (z) ⎛ 1 + K ⎞ ⎛ 1 − K ⎞
Hence, H ( z ) =
=⎜
⎟+⎜
⎟A1 ( z ).
X (z) ⎝ 2 ⎠ ⎝ 2 ⎠
K
7.48 (a) Analyzing Figure P7.6(a) we get Y ( z ) = X ( z ) + (1 + A1 ( z ))X ( z ).
2
Y ( z) ⎛ 1 + K ⎞ ⎛ K ⎞
Therefore, H ( z ) =
=⎜
⎟ + ⎜ ⎟A1 ( z ).
X ( z) ⎝ 2 ⎠ ⎝ 2 ⎠
K
(b) Analyzing Figure P7.6(b) we get Y ( z ) = X ( z ) + (1 − A1 ( z ))X ( z ). Therefore,
2
Y ( z) ⎛ 1 + K ⎞ ⎛ K ⎞
=⎜
H ( z) =
⎟ − ⎜ ⎟A1 ( z ).
X ( z) ⎝ 2 ⎠ ⎝ 2 ⎠

7.49 H (e

⎧1, ω p1 ≤ ω ≤ ω p2 ,

) = ⎨0,
G(e jω ) = H (e j ( π − ω) ), This implies that the
0 ≤ ω ≤ ω s1 ,
⎪0,
ω s 2 ≤ ω < π.

frequency response of H ( − z ) is a shifted version of the frequency response of
H ( z ) , shifted by π radians. Therefore,

⎧1, π − ω p2 ≤ ω ≤ π − ω p1,

) = ⎨0,
G ( e jω ) = H ( e
Hence, H (− z ) is also a
0 ≤ ω ≤ π − ωs 2 ,
⎪0,
π − ωs1 ≤ ω < π.

bandpass filter with passband edges at π − ω p2 and π − ω p1 , and stopband edges at
j ( π − ω)

π − ω s 2 and π − ω s1 with π − ωs 2 < π − ω p1 < π − ω p1 < π − ωs1.
1 − α ⎛⎜ 1 + z −1 ⎞⎟
1 − α ⎛⎜ 1 − z −1 ⎞⎟
, G HP ( z) =
. Let β = − α. Then,
7.50 H LP ( z) =
2 ⎜⎝ 1 − αz −1 ⎟⎠
2 ⎜⎝ 1 + αz −1 ⎟⎠

Not for sale

197

G HP ( z) =

1 + β ⎛⎜ 1 − z −1 ⎞⎟
. Therefore, ωc = cos −1 ( β ) = cos −1 (− α ).

1

2 ⎝ 1 − βz ⎠

7.51 The magnitude responses of H ( z ), H (− z), H ( z 3 ), and H (− z 3 ) are shown below:
|H(e jω )|

|H(_ e jω )|

1

1

π
__
3

0

__
3

π

ω

0

|H(_ e j3ω )|

|H(e j3ω )|
1

π
__
3

__
3

ω

π

1

0 π__

9

π
__
3

__ 7π
__ 2π
__
9 3 9

ω

π

0

__ 4π
__ π
__
9 3 9

__
3

__ π
9

ω

The magnitude responses of H ( z)H ( z 3 ), H (− z) H ( z 3 ), and H ( z)H (− z 3 ) are shown
below:
|H(_ e jω )H(e j3ω )|

|H(e jω )H(e j3ω )|

1

|H(e jω )H(_ e j3ω)|

1

1

0 π__

9

π
__
3

__
3

π

ω

0

π
__
3

__ 7π
__
3 9

π

ω

0

__
__ π
9 3

__
3

π

ω

(
7.52 From Eq. (7.49) we observe that the amplitude response H (ω) of a Type 1 FIR
(
transfer function is a function of cos( ωn). Thus, H (ω + 2πk ) will be a function of
cos((ω + 2πk )n ) = cos(ωn + 2πkn) = cos(ωn) cos(2πkn) − sin(ωn) sin(2πkn) = cos(ωn).
(
Hence, H (ω) is a periodic function in ω with a period 2π.
(
Likewise, from Eq. (7.53) we observe that the amplitude response H (ω) of a Type
(
3 FIR transfer function is a function of sin( ωn). Thus, H (ω + 2πk ) will be a
function of sin((ω + 2πk )n ) = sin(ωn + 2πkn) = sin(ωn) cos(2πkn)
(
+ cos( ωn) sin( 2πkn) = sin( ωn) . Hence, H (ω) is a periodic function in ω with a
period 2π.
(
Next, from Eq. (7.52) we observe that the amplitude response H (ω) of a Type 2
(
1
FIR transfer function is a function of cos⎛⎜ ω(n − ) ⎞⎟. Thus, H (ω + 4πk ) will be a
2 ⎠

function of
1
1
1
1
1
cos⎛⎜ (ω + 4πk )(n − ) ⎞⎟ = cos⎛⎜ ω(n − ) + 4πk (n − ) ⎞⎟ = cos⎛⎜ ω(n − ) ⎞⎟ cos⎛⎜ 4πk (n − ) ⎞⎟
2 ⎠
2
2 ⎠
2 ⎠
2 ⎠

Not for sale

198

1
1
1
1
− sin⎛⎜ ω(n − ) ⎞⎟ sin⎛⎜ 4πk (n − ) ⎞⎟ = cos⎛⎜ ω(n − ) ⎞⎟ as cos⎛⎜ 4πk (n − ) ⎞⎟ = 1 and
2 ⎠
2 ⎠
2 ⎠
2 ⎠

(
1
sin⎛⎜ 4πk (n − ) ⎞⎟ = 0. Hence, H (ω) is a periodic function in ω with a period 4π.
2 ⎠

(
Finally, from Eq. (7.54) we observe that the amplitude response H (ω) of a Type
(
1
4 FIR transfer function is a function of sin⎛⎜ ω(n − ) ⎞⎟. Thus, H (ω + 4πk ) will be a
2 ⎠

function of
1
1
1
1
1
sin⎛⎜ (ω + 4πk )(n − ) ⎞⎟ = sin⎛⎜ ω(n − ) + 4πk (n − ) ⎞⎟ = sin⎛⎜ ω(n − ) ⎞⎟ cos⎛⎜ 4πk (n − ) ⎞⎟
2 ⎠
2
2 ⎠
2 ⎠
2 ⎠

1
1
1
1
+ cos⎛⎜ ω(n − ) ⎞⎟ sin⎛⎜ 4πk (n − ) ⎞⎟ = sin⎛⎜ ω(n − ) ⎞⎟ as cos⎛⎜ 4πk (n − ) ⎞⎟ = 1 and
2
2
2
2 ⎠

⎠ ⎝

(
1
sin⎛⎜ 4πk (n − ) ⎞⎟ = 0. Hence, H (ω) is a periodic function in ω with a period 4π.
2 ⎠

7.53 The remaining zeros are at: z 5 =

1
1
=
= 1.25; z 6 = z 2* = j; z 7 = z 3* = 2 + j 2;
z1 0.8

1
1
=
= 0.25 + j 0.25; z 9 = z8* = 0.25 − j 0.25; z10 = z *4 = −0.5 − j 0.3;
z3 2 − j 2
1
1
0 .5
0 .3
0 .5
0 .3
z11 =
=
=−
+j
, z12 = z11* = −
−j
.
z 4 − 0 .5 + j 0 .3
0.34
0.34
0.34
0.34

z8 =

12

H1 ( z) = ∏ (1 − z k z −1 ) = 1 − 2.6088z −1 + 1.7576 z − 2 + 11.0226 z − 3 + 5.6432 z − 4
k =1
−5

− 24.2166 z

+ 9.7711z −6 − 24.2166 z −7 + 5.6432 z −8 + 11.0226 z −9 + 1.7576 z −10

− 2.6088 z −11 + z −12 .

7.54 The remaining zeros are at: z 4 =

1
1
=
= 0.3226, z 5 = z 2* = −2 − j 4,
z1 3.1

1
1
=
= −0.1 − j 0.2, z 7 = −0.1 − j 0.2,
z2 − 2 + j 4
1
1
z8 = z 3* = 0.8 − j 0.4, z 9 =
=
= 1 − j 0.5, z10 = 1 + j 0.5, z11 = −1.
z 3 0 .8 + j 0 .4
z6 =

11

H 2 ( z) = ∏ (1 − z k z −1 ) = 1 − 1.8226 z −1 + 7.1039z − 2 − 79.4635z − 3 +182.19z − 4
k =1

− 111.2306 z −5 −111.2306 z −6 +182.19 z −7 −79.4635 z −8 +7.1039 z −9 −1.8226 z −10 + z −11 .

Not for sale

199

7.55 The remaining zeros are at: z 4 = z1* = 0.1 + j 0.599,
1
1
z5 =
=
= 0.2711 + j 1.6242, z 6 = z 5* = 0.2711 - j 1.6242,
z1 0.1 − j 0.599
1
z 7 = z 2* = − 0 . 3 − j 0 . 4 , z8 =
= −1.2 − j 1.6, z 9 = z8* = −1.2 + j 1.6,
z2
1
z 3 = 2, z10 =
= 0.5, z10 = 1, z11 = −1.
z3
H 3 ( z ) = 1 − 0 .2423 z − 1 + 1 .0076 z − 2 − 6 .5294 z − 3 + 1 .3338 z − 4 − 17 .2533 z − 5
+ 17.2533 z −7 − 1.3338 z −8 + 6.5294 z −9 − 1.0076 z −10 + 0.2423 z −11 − z −12 .

7.56 The remaining zeros are at: z 4 = 2.2 − j 3.4, z 5 =

1
= 0.1341 − j 0.2073,
z1

z 6 = z *5 = 0.1341 + j 0.2073,
z 2 = 0 . 6 + j 0 . 9 , z 7 = z 2* = 0 . 6 − j 0 . 9 , z8 =
z 9 = z8* = 0.5128 + j 0.7692, z 3 = −0.5, z10 =

1
= 0.5128 − j 0.7692,
z2

1
= −2, z11 = 1.
z3

H 4 ( z) = 1 − 5.3939z −1 + 19.446 z −2 − 5.0838z −3 − 51.8577z −4 + 119.7073z −5
− 119.7073 z −6 + 51.8577 z −7 + 5.0838 z −8 − 19.446 z −9 + 5.3939 z −10 − z −11 .

7.57 The magnitude responses of H ( z M ) (solid line) and F1 ( z ) (dashed line) are shown
below:
|F1(e jω)|

1

0

ω
__p
M

ω__s
M

|H(e jMω )|

s
__
M

__
M

ω

__
M

Hence, G1 ( z) = H ( z M )F1 ( z ) is a lowpass filter with a unity passband magnitude,
passband edge at ω p / M and stopband edge at ωs / M .
The magnitude responses of H ( z M ) (solid line) and F2 ( z) (dashed line) are shown
below:

Not for sale

200

1

|F2(e jω)|

|H(e jMω )|

0

ω__s
M

ω
__p
M

+ωs

__
s
__
__
M
M 2π _ ω M 2π + ω
__p
__p
M
M

s
__
M

ω

__
M

Hence, G2 ( z) = H ( z M )F2 ( z ) is a bandpass filter with a unity passband magnitude,
passband edges at (2π − ω p ) / M and (2π + ω p ) / M , and stopband edges at

(2π − ωs ) / M and (2π + ωs ) / M .
N

N

n=0

n=0

7.58 H ( z) = ∑ h[ n] z − n and H (e jω ) = ∑ h[ n] e − jωn . The frequency response will
exhibit generalized linear phase if it can be expressed in the form H (e jω )
(
(
= H (ω)e − jαωe − jβ , where H (ω), the amplitude function, is a real function of ω
and α and β are constants. We need to examine the case when the order N is
even and when N is odd separately. Without any loss of generality, assume first
5

N = 5. Then H ( z ) = ∑ h[ n] z − n , and
n=0
− jω

H (e jω ) = h[0] + h[1] e

+ h[2] e − j 2ω + h[3] e − j 3ω + h[ 4] e − j 4ω + h[5] e − j 5ω

= e − j 5ω / 2 h[ 0]e j 5ω / 2 + h[5]e − j 5ω / 2 + h[1] e j 3ω / 2 + h[ 4] e − j 3ω / 2 + h[ 2] e jω / 2

(

+ h[3] e − j 3ω / 2 = e − j 5ω / 2 (h[ 0] + h[ 5]) cos( 5ω / 2) + (h[1] + h[ 4]) cos(3ω / 2)

)

[

+ (h[ 2] + h[3]) cos(3ω / 2) + je − j 5ω / 2 (h[ 0] − h[ 5]) sin( 5ω / 2) + (h[1] − h[ 4]) sin(3ω / 2)

]
+ (h[2] − h[3]) cos(3ω / 2)] .

[

It follows then that if h[ n] = h[ 5 − n], 0 ≤ n ≤ 5, we

(
have H (e jω ) = e − j 5ω / 2 H (ω), where
(
H (ω) = 2h[0] cos(5ω / 2) + 2h[1] cos(3ω / 2) + 2h[2] cos(ω / 2), which is a real
function of and as a result, H (e jω ) has generalized phase.
Alternately, if h[ n] = − h[ 5 − n], 0 ≤ n ≤ 5, then we have
(
(
H (e jω ) = je − j 5ω / 2 H (ω) = e − j 5ω / 2 e jπ H (ω), where,
(
H (ω) = 2h[0] sin(5ω / 2) + 2h[1] sin(3ω / 2) + 2h[2] sin(ω / 2), which is a real
function of and as a result, H (e jω ) has generalized phase.
6

Next, assume N = 6. Then H ( z) = ∑ h[ n] z − n , and H (e jω ) = h[0] + h[1] e − jω
n=0

Not for sale

201

+ h[2] e − j 2 ω + h[3] e − j 3ω + h[ 4] e − j 4ω + h[5] e − j 5ω + h[6] e − j 6ω
= e − j 3ω (h[ 0] + h[6]) cos(3ω) + (h[1] + h[5]) cos( 2ω) + (h[ 2] + h[ 4]) cos(ω) + h[3]

[
+ je − j 3ω[(h[ 0] − h[6]) sin(3ω) + (h[1] − h[ 5]) sin( 2ω) + (h[ 2] − h[ 4]) sin( ω)].

]

(
Hence, it follows that if h[ n] = h[6 − n], 0 ≤ n ≤ 6, then H (e jω ) = e − j 3ω H (ω),
(
where H (ω) = 2h[0] cos(3ω) + 2h[1] cos(2ω) + 2h[2] cos(ω) + h[3], which is a real
function of and as a result, H (e jω ) has generalized phase.
Alternately, if h[ n] = − h[6 − n], 0 ≤ n ≤ 5, then we have
(
(
H (e jω ) = je − j 3ω H (ω) = e − j 3ω e jπ H (ω), where,
(
H (ω) = 2h[0] sin(3ω) + 2h[1] sin(2ω) + 2h[2] sin(ω), which is a real function of and
as a result, H (e jω ) has generalized phase.
7.59 Type 1:
Type 2:
Type 3:
Type 4:

h[ n] = {a
h[ n] = {a
h[ n] = {a
h[ n] = {a

−b
−b
−b
−b

−c
−c
−c
−c

d
d
d
d

e
e
e
e

d − c − b a}.
e d − c − b a}.
0 − e − d c b − a}.
− e − d c b − a}.

7.60 (a) Type 1: {h[ n]} = {1 − 3 − 4 6 8 6 − 4 − 3 1}. Hence,

H ( z ) = 1 − 3 z − 1 − 4 z − 2 + 6 z −3 + 8 z − 4 + 6 z − 5 − 4 z − 6 − 3 z − 7 + z −8 .
The zero plot obtained using the M-file zplane is shown below:

Imaginary Part

2
1
8

0
-1
-2
-1

0

1
2
Real Part

3

It can be seen from the above that a complex-conjugate zero pair on the unit circle
appear singly and real zeros appear in mirror-image symmetry. There are no zeros
at z = 1.
(b) Type 2: {h[ n]} = {1 − 3 − 4 6 8 8 6 − 4 − 3 1}. Hence,

H ( z ) = 1 − 3 z − 1 − 4 z − 2 + 6 z − 3 + 8 z −4 + 8 z − 5 + 6 z − 6 − 4 z − 7 − 3 z − 8 + z − 9 .
The zero plot obtained using the M-file zplane is shown below:

Not for sale

202

1.5

Imaginary Part

1
0.5
0

3

9

-0.5
-1
-1.5
-1

0

1
Real Part

2

3

From the above zero plot it can be seen that complex-conjugate zero pairs on the
unit circle appear singly and real zerosappear in mirror-image symmetry. There are
3 zeros at z = −1.
(c) Type 3: {h[ n]} = {1 − 3 − 4 6 8 0 − 8 − 6 4 3 − 1}. Hence,

H ( z) = 1 − 3z −1 − 4 z −2 + 6 z −3 + 8z −4 − 8z −6 − 6 z −7 + 4 z −8 + 3z −9 − z −10 .
The zero plot obtained using the M-file zplane is shown below:
1.5
Imaginary Part

1
0.5
10

0
-0.5
-1
-1.5
-1

0

1
2
Real Part

3

From the above zero plot it can be seen that complex-conjugate zero pairs on the
unit circle appear singly and real zeros appear in mirror-image symmetry. There is
one zero at z = −1. and one zero at z = 1.
(d) Type 4: {h[ n]} = {1 − 3 − 4 6 8 − 8 − 6 4 3 − 1}. Hence,

H ( z ) = 1 − 3 z − 1 − 4 z − 2 + 6 z − 3 + 8 z −4 + 8 z − 5 + 6 z − 6 − 4 z − 7 − 3 z − 8 + z − 9 .
The zero plot obtained using the M-file zplane is shown below:

Not for sale

203

1.5
Imaginary Part

1
0.5
9

0
-0.5
-1
-1.5
-1

0

1
2
Real Part

3

From the above zero plot it can be seen that complex-conjugate zeros appear in
mirror-image symmetry, a complex-conjugate zero pair on the unit circle appear
singly, and real zeros appear in mirror-image symmetry. There is one zero at z = 1.
7.61 H1 ( z) is of Type 1 and hence, it has a symmetric impulse response of odd length
2 N + 1. Let α be the constant term of H1 ( z). Then, the coefficient of the highest
power of z −1 of H1 ( z) is also α.
H 2 ( z) is of Type 2 and hence, it has a symmetric impulse response of even length
2M. Let β be the constant term of H 2 ( z). Then, the coefficient of the highest
power of z −1 of H 2 ( z) is also β .
H 3 ( z) is of Type 3 and hence, it has an anti-symmetric impulse response of odd
length 2 R + 1. Let γ be the constant term of H 3 ( z). Then, the coefficient of the
highest power of z −1 of H 3 ( z) is − γ.
H 4 ( z) is of Type 4 and hence, it has an anti-symmetric impulse response of even
length 2K . Let δ be the constant term of H 4 ( z). Then, the coefficient of the
highest power of z −1 of H 4 ( z) is − δ.
(a) The length of H1 ( z)H1 ( z) is (2 N + 1) + (2 N + 1) − 1 = 4 N + 1 which is odd.
The constant term of H1 ( z)H1 ( z) is α 2 and the coefficient of the highest power of
z −1 of H1 ( z)H1 ( z) is also α 2 . Hence, H1 ( z)H1 ( z) is of Type 1.

(b) The length of H1 ( z)H 2 ( z) is (2 N + 1) + (2 M ) − 1 = 2( N + M ) which is even.
The constant term of H1 ( z)H 2 ( z) is αβ and the coefficient of the highest power of
z −1 of H1 ( z)H 2 ( z ) is also αβ . Hence, H1 ( z )H 2 ( z ) is of Type 2.

(c) The length of H1 ( z)H 3 ( z) is (2 N + 1) + (2 R + 1) − 1 = 2( N + R ) + 1 which is
odd. The constant term of H1 ( z)H 3 ( z) is αγ and the coefficient of the highest
power of z −1 of H1 ( z)H 3 ( z) is − αγ. . Hence, H1 ( z)H 3 ( z) is of Type 3.

Not for sale

204

(d) The length of H1 ( z)H 4 ( z) is (2 N + 1) + (2 K ) − 1 = 2( N + K ) which is even.
The constant term of H1 ( z)H 4 ( z) is αδ and the coefficient of the highest power of
z −1 of H1 ( z)H 4 ( z) is − αδ . Hence, H1 ( z)H 4 ( z) is of Type 4.

(e) The length of H 2 ( z)H 2 ( z) is (2 M ) + (2 M ) − 1 = 4 M − 1 which is odd. The
constant term of H 2 ( z)H 2 ( z) is β 2 and the coefficient of the highest power of
z −1 of H 2 ( z)H 2 ( z) is also β 2 . Hence, H 2 ( z)H 2 ( z) is of Type 1.

(f) The length of H 3 ( z)H 3 ( z ) is (2 R + 1) + (2 R + 1) − 1 = 4 R + 1 which is odd. The
constant term of H 3 ( z)H 3 ( z ) is γ 2 and the coefficient of the highest power of
z −1 of H 3 ( z)H 3 ( z ) is also γ 2 . Hence, H 3 ( z)H 3 ( z ) is of Type 1.

(g) The length of H 4 ( z)H 4 ( z) is (2 K ) + (2 K ) − 1 = 4 K − 1 which is odd. The
constant term of H 4 ( z)H 4 ( z) is δ 2 and the coefficient of the highest power of
z −1 of H 4 ( z)H 4 ( z) is also δ 2 . Hence, H 4 ( z)H 4 ( z) is of Type 1.

(h) The length of H 2 ( z)H 3 ( z ) is (2 M ) + (2 R + 1) − 1 = 2( M + R ) which is even.
The constant term of H 2 ( z)H 3 ( z ) is βγ and the coefficient of the highest power of
z −1 of H 2 ( z)H 3 ( z ) is − βγ . Hence, H 2 ( z)H 3 ( z ) is of Type 4.

(i) The length of H 3 ( z)H 4 ( z) is (2 R + 1) + (2 K ) − 1 = 2( R + K ) which is even. The
constant term of H 3 ( z)H 4 ( z) is γδ and the coefficient of the highest power of
z −1 of H 3 ( z)H 4 ( z) is also γδ . Hence, H 3 ( z)H 4 ( z) is of Type 2.

7.62 (a) F 1( z ) = 2.1 − 3.5z −1 + 4.2 z −2 = 2.1(1 − 1.667z −1 + 2 z −2 ). F1( z) has complex
conjugate zeros at z = 0.8333 ± j1.1426. To generate a linear-phase transfer
function H ( z ) , we need to multiply F1( z) with the factor F 2 ( z) which has
complex-conjugate zeros situated in the z –plane with a mirror-image symmetry
with respect to the zeros of F1( z) . Hence, F 2 ( z) = 2 − 1.667z −1 + z −2 , resulting in

H ( z ) = F1 ( z )F2 ( z) = 2.1(2 − 5z −1 + 7.7778z −2 − 5z −3 + 2 z −4 )
= 4.2 − 10.5z −1 + 16.3333 z −2 − 10.5z −3 + 4.2 z −4 .

(b) F 1( z) = 1.4 + 5.2 z −1 − 2.2 z −2 + 3.3z −3

= 1.4(1 + 3.7143z −1 − 1.5714 z −2 + 2.3571z −3 ). F1( z) has complex conjugate zeros

Not for sale

205

at z = 0.2524 ± j 0.7035 . and a real zero at − 4.2192. To generate a linear-phase
transfer function H ( z ) , we need to multiply F1( z) with the factor F 2 ( z) which has
situated in the z –plane with a mirror-image symmetry with respect to the zeros of

F1( z) . Hence, F 2 ( z ) = 2.3571 − 1.5714 z −1 + 3.7143z −2 + z −3 , resulting in
H ( z) = F1 ( z)F2 ( z) = 1.4(2.3571 + 7.1837z −1 − 5.8265z −2 + 22.8214 z −3
− 5.8265z −4 + 7.1837z −5 + 2.3571z −6 ) = 3.3 + 10.0571z −1 − 8.1671z −2 + 31.95z −3
− 8.1571z −4 + 10.0571z −5 + 3.3z −6 .
N

7.63 We rewrite the polynomial H ( z ) in the form f ( z) = Kz − N ∏ ( z − λi ) . Its
i =1

'

logarithmic differential is given by

z

f ( z)
N N 1
or
=− +∑
f ( z)
z i =1 z − λi

N
N
f ' ( z)
z
1
.
= −N + ∑
= −N + ∑
f ( z)
z

λ
1

(
λ
/
z
)
i
i
i =1
i =1

For any | z |> max | λi | we can expand the above in a Taylor series as
i
N

[

]

f ′( z)
= − N + ∑ 1 + (λi / z) + (λi / z)2 + (λi / z)3 + ... , or
f ( z)
i =1
f ′(z)
z
= − N + N + (S1 / z) + (S2 / z)2 + (S3 / z)3 + ...
f ( z)
z

where S m =

N

∑ λi

m

(1)

.

i =1

Now set f ( z ) = h[0] + h[1]z −1 + h[2]z −2 + ... + h[ N ]z − N . So that

z

f ′( z)
h[1]z −1 + 2h[2]z −2 + ... + Nh[ N ]z − N
.
=−
f ( z)
h[0] + h[1]z −1 + h[2]z−2 + ... + h[ N ]z − N

(2)

Equations (1) and (2) pertain to the same quantity hence identically the right hand
sides are the same. Thus the following convolution holds true

{h[0],

h[1], h[2], K, h[ N ]}O
* {0, S1 , S 2 , S 3 , K} = −{h[1], 2h[ 2], K, Nh[ n]}

Hence the Newton Identities.
n1

n2

i =1

i =1

7.64 The root moments of Sm of H ( z) = K ∏ (1 − αi z −1 )∏ (1 − βi z −1 ) are defined as

Not for sale

206

Sm =

n1

∑ α im +

i =1

n2

∑ βim .

For real H ( z ), the complex roots occur in conjugate pairs

i =1

and hence, their corresponding powers are also in this form, thereby making their
sum entirely real.
(a) If H ( z ) is minimum-phase, then βi = 0 for all i, and αi < 1. Therefore as
m → ∞, S m will decrease exponentially.
n1

n2

i =1

i =1

(b) Write ln H ( z ) = ln K + ∑ ln(1 − αi z −1 ) + ∑ ln(1 − βi z −1 ) and then expand
ln H ( z ) in a Laurent series. The second summation can be re-expressed as
n2

∑ ln(1 − βi z

−1

)=

i =1

ln H ( z) = ln K +

n2

i =1

∑ ln( − βi z −1 )⎜⎜ 1 −

z
βi

⎟⎟. Hence,

n2

n1

n2

i =1

i =1

i =1

z
∑ ln(− βi ) − n 2 ln z + ∑ (1 − αi z −1 ) + ∑ (1 − β ). Now, βi appear
i

n2

in complex conjugate pairs, hence, ln K + ∑ ln(− βi ) can be written as ln K1 where
i =1

K1 is real.

(α z −1 )2 (αi z −1 )3
+
+ L⎥ and
However, ln(1 − αi z −1 ) = − ⎢αi z −1 + i
2
3
⎢⎣
⎥⎦
⎡ z ( z / β ) 2 ( z / β )3

z ⎞
i
i

+
+ L⎥. These two Taylor series expansions are
ln⎜ 1 − ⎟ = − ⎢ +
βi ⎠
2
3

⎣⎢ βi
⎦⎥
valid since αi z −1 < 1 and z / βi < 1 on z = 1. Thus,

ln H ( z ) = ln K1 −

N2
n1
n1
∞ ⎛ S N1

N2
m
m z − m + S − m z m ⎟ where S N1 =

α
and
S
=
βi− m .
∑⎜

m
i
−m

m
m =1 ⎝ m
i =1
i =1

(
On z = 1 we have H ( z ) z = e jω = H (ω) e jθ (ω) and therefore,

⎛ ∞ S N1 − jmω ∞ S N 2 jmω ⎞
(
⎟ + (− jn ω).
ln H (ω) + jθ (ω) = ln K1 − ⎜ ∑ m e
+ ∑ −m e
2
⎜ m =1 m

m
m =1

On equating real and imaginary parts of the equation we arrive at
∞ ⎛ S N1 + S N 2 ⎞
(
− m ⎟ cos(mω),
ln H (ω) = ln K1 − ∑ ⎜ m

m
m =1 ⎝

∞ ⎛ S N1 − S N 2 ⎞
− m ⎟ sin(mω).
θ (ω) = − n2 ω + ∑ ⎜ m

m
m =1 ⎝

Not for sale

207

(c) From the expression for the phase as given above, it can be seen that the second
term contributes non-linear components to the phase, whilst the only term which is
linear is the first (− n2 ω). Thus to have linear phase we must have (1) n2 ≠ 0 and (2)
N1
Sm
= S −Nm2 . The second condition means that the zeros of H ( z ) outside the unit circle
must be the same in number as those inside the unit circle and the zeros outside must be
at locations of the zeros inside the unit circle.

d + e − jω d1e jω / 2 + e − jω / 2
d + z −1
7.65 (a) A1 ( z ) = 1
. Thus, A1 (e jω ) = 1
=
1 + d1e − jω e jω / 2 + d1e − jω / 2
1 + d1z −1
α e jβ
=
= e j 2 β , where αe jβ = d1e jω / 2 + e − jω / 2 )
− jβ

αe

= (d1 + 1) cos(ω / 2) + j(d1 − 1) sin(ω / 2). Therefore, phase is given by
⎛ 1 − d1

θ (ω) = 2 β = −2 tan −1 ⎜⎜
tan(ω / 2) ⎟⎟. Now, for small values of x, tan( x ) ≅ x and
⎝ 1 + d1

tan −1 ( x ) ≅ x. Hence, the approximate expression for the phase at low frequencies

⎛ 1 − d1 ⎞ ω
⎛ 1 − d1 ⎞
⎟⎟ = −⎜⎜
⎟⎟ ω. Therefore, the approximate
is given by θ (ω) ≅ −2⎜⎜
⎝ 1 + d1 ⎠ 2
⎝ 1 + d1 ⎠
θ ( ω)
1 − d1
expression for the phase delay is given by τ p (ω) = −
samples.
=δ≅
ω
1 + d1
(b) For δ = 0.5 samples, d1 =

1− δ
1+ δ

=

0 .5
1.5

=

1
.
3

Then, A1 ( z) =

1
3

+ z −1

1 + z −1
1
3

. Thus, the

2 1− d
2
θ ( ω)
= ⎛⎜ 1 tan(ω / 2) ⎞⎟ = (0.5 tan(ω / 2)).
exact phase delay is given by τ p (ω) = −
ω
ω ⎝ 1+ d1
ω

For a sampling rate of 20 kHz, the normalized angular frequency equivalent to 1

kHz is ω o =

τ p (ω o ) =

10 3
20×10 3

2
ωo

=

1
20

= 0.05. The exact phase delay at ωo is thus

(0.5 tan(ωo / 2) =

2
(0.5 tan(0.025) = 0.500078 samples, which is
0.05

seen to be very close to the desired phase delay of 0.5 samples.
Phase delay, in samples

Phase delay, in samples

1
0.9
0.8
0.7
0.6
0.5

0

0.2

0.4

0.6

0.8

1

0.55
0.5
0.45
0.4

ω/π

0

0.05

0.1

0.15

0.2

ω/ π

Not for sale

208

d2 + d1e − jω + e − j 2 ω
d2 e jω + d1 + e − jω
7.66 A2 (e ) =
=
1 + d1e − jω + d2 e − j 2ω e jω + d1 + d2 e − jω
(d + d2 cos ω + cos ω) + j (d2 sin ω + sin ω) (d1 + (d2 + 1) cos ω) + j (d2 − 1) sin ω
=
.
= 1
(d1 + d2 cos ω + cos ω) + j (sin ω − d2 sin ω) (d1 + (d2 + 1) cos ω) − j (d2 − 1) sin ω
( d 2 −1) sin ω ⎞
Therefore, θ(ω) = 2 tan −1 ⎛⎜
⎟. Now,
⎝ d1 + ( d 2 +1) cos ω ⎠
θ(ω)
2
⎛ ( d 2 −1) sin ω ⎞ . For ω ≅ 0, sin ω = ω and
τ p (ω) = −
= − tan −1 ⎜

ω
ω
⎝ d1 + ( d 2 +1) cos ω ⎠
( d −1)ω ⎞
2
−1
cos ω = 1. Then, τ p (ω) = − tan −1 ⎛⎜ 2
⎟. Also, for x ≅ 0, tan x ≅ x.
ω
d
+
(
d
+
1
)
2
⎝ 1

2( d 2 −1)
2 ( d 2 −1)ω
2−δ ⎞
=−
. Now, substituting d1 = 2⎛⎜
Hence, τ p (ω) = −
⎟ and
ω d1 + ( d 2 +1)
d1 + d 2 +1
⎝ 1+ δ ⎠

d2 =

2( d 2 −1)
(2 − δ)(1− δ)
, we can easily show that −
= δ.
(2 + δ)(1+ δ)
d1 + d 2 +1

7.67 Since G ( z ) is non-minimum phase but causal, it will have some zeros outside the
unit circle. Let z = α be one such zero. We can then write G( z) = P( z)(1 − αz −1 )
−1 ⎞
⎛ 1− αz −1 ⎞

⎟. Note that ⎜ 1− αz

= P( z)(−α * + z −1 )⎜⎜
⎜ ( −α * + z −1 ) ⎟ is a stable first-order
−1 ⎟

⎝ ( −α * + z ) ⎠

allpass function. If we carry out this operation for all zeros of G ( z ) that are outside
the unit circle, we can write G ( z ) = H ( z ) A( z ) where H ( z ) will have all zeros inside
the unit circle and will thus be a minimum phase function and will be a product of
stable first-order allpass functions, and hence an allpass function.

(3z − 2.1)( z 2 + 2.5z + 5)
. In order to correct for magnitude distortion we
( z − 0.65)( z + 0.48)
require the transfer function G ( z ) to satisfy the following property
1
G(e jω ) =
. Hence, one possible solution is

H (e )

7.68 H ( z ) =

Gd ( z ) =

1
( z − 0.65)( z + 0.48)
. Note that the coefficients of the pole
=
H ( z) (3z − 2.1)(z 2 + 2.5z + 5)

factor ( z 2 + 2.5z + 5) in the denominator of Gd (z ) do not satisfy the condition of
Eq. (7.139) and hence, has roots outside the unit circle making Gd (z ) unstable.
To develop a stable transfer function with magnitude response same as Gd (z ) , we
multiply it with the stable allpass function

z 2 + 2.5z + 5

5z 2 + 2.5z + 1

Not for sale

resulting in the transfer

209

( z − 0.65)( z + 0.48)

function G( z) =

(3z − 2.1)(5z 2 + 2.5z + 1)

which is the desired stable solution

satisfying the condition G (e jω ) H (e jω ) = 1.
7.69 (a) G ( z ) = H ( z ) A( z ), where A( z ) is an allpass function. Then, g[ 0] = lim G ( z ).
z→∞

Hence,
g[0] = | lim G( z) | = | lim H ( z ) A( z) | = | lim H ( z) | | lim A( z) | ≤ | lim H ( z) |
z →∞

z →∞

z →∞

z →∞

z →∞

≤ | lim H ( z ) | because | lim A( z ) | < 1 because of Property 2 of stable allpass
z→∞

z→∞

function (see Eq. (7.20)). Hence, g[0] ≤ h[0] .
(b) If λ1 is a zero of H ( z ), then λ1 < 1, since H ( z ), is a minimum-phase causal
stable transfer function. As H ( z ) has all zeros inside the unit circle, we can write

H ( z ) = B( z)(1 − λ1z −1 ). It follows that B( z ) is also a minimum-phase causal
transfer function.
⎛ λ* − z −1 ⎞
⎟. If
Now consider the transfer function F ( z ) = B( z)( λ*1 − z −1 ) = H ( z)⎜ 1
⎜ 1 − λ z −1 ⎟
1

h[ n], b[ n], and f [ n] denote, respectively, the inverse z –transforms of H ( z ), B( z ),

b[0],
n = 0,

and
and F ( z ), then we get h[ n] = ⎨
⎩b[ n] − λ1b[ n − 1], n ≥ 1,
⎧⎪
λ*1b[0],
n = 0,
f [ n] = ⎨ *
⎪⎩ λ1b[ n] − b[ n − 1], n ≥ 1.
m

2

m

2

2

2

m

2

2

m

2

Consider ε = ∑ h[ n] − ∑ f [ n] = b[0] − λ*1 b[0] + ∑ h[ n] − ∑ f [ n] .
n=0

2

n=0
2

2

n =1

n =1

2

Now, h[ n] = b[ n] + λ1 b[ n − 1] − λ1b[ n − 1]b * [ n] − λ*1b * [ n − 1]b[ n], and
2

2

2

2

f [ n] = λ1 b[ n] + b[ n − 1] − λ1b[ n − 1]b * [ n] − λ*1b * [ n − 1]b[ n]. Hence,
m
m
2
2
2
2
2
2
2
2
2
ε = b[0] − λ1 b[0] + ∑ ⎛⎜ b[ n] + λ1 b[ n − 1] ⎞⎟ − ∑ ⎛⎜ λ1 b[ n] − b[ n − 1] ⎞⎟

n =1
n =1
2

m

2

2

m

2

= (1 − λ1 ) b[ m] . Since λ1 < 1, ε > 0, i.e., ∑ h[ n] > ∑ f [ n] . Hence,
n=0

m

2

m

n=0

2

∑ h[ n] ≥ ∑ g[ n] .

n=0

n=0

Not for sale

210

(2 z + 3)(4 z − 1)
3
has a zero at z = which is outside the unit circle and is
2
( z + 0.4)( z − 0.6)
thus a non-minimum phase transfer function. To develop a minimum phase transfer

7.70 H ( z) =

function G ( z ) such that G (e jω ) = H (e jω ) , we multiply with an allpass function

3z + 2
(3z + 2)(4 z − 1)
and arrive at G( z) =
which is minimum phase.
( z + 0.4)( z − 0.6)
2z + 3
The first 5 impulse response of H ( z ) are
{h[ n]} = {8 11.6 1.24 3.032 0.904}, 0 ≤ n ≤ 4, and
The first 5 impulse response of G ( z ) are
{g[ n]} = {12 7.4 2.36 2.248 1.016}, 0 ≤ n ≤ 4.
m

0

m

Σ h[ n]

n=0
m

Σ g[ n]

2
2

n= 0

64

1

2

3

4

198.56 200.0976 209.2906 210.1078

144 198.76 204.3296 209.3831 210.4154
m

m

2

2

It follows from the above that ∑ g[ n] > ∑ h[ n]
n=0

n=0

for m ≥ 1.

7.71 See Example 7.14.
(a) H BS ( z) = (1 + z − 2 )2 . Thus, H BP ( z ) = z − 2 − (1 + z − 2 )2 = − (1 − z − 2 )2 .
4
4
4
1

1

1

(b) H BS ( z) = (1 + z − 2 )(−1 + 6 z − 2 − z − 4 ). Thus,
16
1

H BP ( z ) = z − 4 −
=

1
(1 + z − 2 )(−1 + 6 z − 2
16

− z−4 ) =

1
(1 − 4 z − 2
16

+ 6 z − 4 − 4 z − 6 + z −8 )

1
−2 4
(1 − z ) .
16

(c) H BS ( z ) = (1 + z − 2 )2 (−3 + 14 z − 2 − 3z − 4 ). Thus,
32
1

H BP ( z) = z − 4 −

1
(1 + z − 2 )2 (−3 + 14 z − 2
32

− 3z − 4 ) =

1
(3 − 8z − 2
32

+ 10 z − 4 − 8 z − 6 + 3z − 8 ).

7.72 H 0 ( z ) = A0 ( z ) + A1 ( z), and H1 ( z) = A0 ( z) − A1 ( z), where A0 ( z) and A1 ( z) are
allpass functions of orders M and N , respectively. Hence, the orders of H 0 ( z)

z − M D0 ( z −1 )
and
and H1 ( z) are M + N . Now, we can write A0 ( z) =
D0 ( z)

Not for sale

211

A1 ( z) =

z − N D1 ( z −1 )
. Then,
D1 ( z )

H0 (z) =

P( z) z − M D0 ( z −1 ) D1 ( z ) + z − N D0 ( z) D1 ( z −1 )
=
and
D( z )
D0 ( z) D1 ( z )

H1 ( z) =

Q( z) z − M D0 ( z −1 ) D1 ( z) − z − N D0 ( z) D1 ( z −1 )
. Since P (z ) is of degree
=
D( z )
D0 ( z ) D1 ( z )

[

M + N and z − ( M + N ) P ( z −1 ) = z − ( M + N ) z M D0 ( z ) D1 ( z −1 ) + z N D0 ( z −1 ) D1 ( z )

]

= z − M D0 ( z −1 ) D1 ( z ) + z − N D0 ( z ) D1 ( z −1 ) = P ( z ). Hence, P (z ) is symmetric.
Similarly, one can show that Q(z ) is anti-symmetric.
1

1

7.73 H 0 ( z) = [ A0 ( z) + A1 ( z )] and H1 ( z) = [ A0 ( z) − A1 ( z)]. Thus,
2
2
H 0 ( z ) H 0 ( z −1 ) + H1 ( z ) H1 ( z −1 ) = [ A0 ( z) + A1 ( z)][ A0 ( z −1 ) + A1 ( z −1 )]
1
4
−1

+ [ A0 ( z ) − A1 ( z )][ A0 ( z −1 ) − A1 ( z )]
4
1

= [ A0 ( z ) A0 ( z −1 ) + A0 ( z) A1 ( z −1 ) + A1 ( z) A0 ( z −1 ) + A1 ( z) A1 ( z −1 )]
4
1

+ [ A0 ( z) A0 ( z −1 ) − A0 ( z) A1 ( z −1 ) − A1 ( z ) A0 ( z −1 ) + A1 ( z) A1 ( z −1 )]
4
1

2

2

= [ A0 ( z) A0 ( z −1 ) + A1 ( z ) A1 ( z −1 )] = 1. Thus, H 0 (e jω ) + H1 (e jω ) = 1
1
2

implying that H 0 ( z) and H1( z) form a power-complementary pair.
2

7.74 H 0 (e jω ) =

{A0 (e jω ) A0* (e jω ) + A1(e jω ) A0* (e jω ) + A0 (e jω ) A1* (e jω )
+ A1 (e jω ) A1* (e jω )}. Since A0 ( z) and A1 ( z) are allpass functions,
1
4

A0 (e jω ) = e jφ0 (ω) and A1 (e jω ) = e jφ1 (ω) . Therefore,
2

H 0 ( e jω ) =

{

}

1
2 + e j (φ0 (ω) − φ1 (ω)) + + e − j (φ0 (ω) − φ1 (ω)) ≤ 1 as maximum values of
4

and e − j (φ0 (ω) − φ1 (ω)) are 1. H 0 ( z) is stable since A0 ( z) and A1 ( z)
are stable allpass functions. Hence, H 0 ( z) is BR.
e

j ( φ0 ( ω)− φ1( ω))

1 M −1
1 M −1M −1
−1
−1
∑ Ak ( z). Thus, H ( z)H ( z ) = 2 ∑ ∑ Ak ( z) Ar ( z ). Hence,
M k =0
M k =0 r =0
2
M

1
M

1
1
j (φ (ω) − φr (ω))
H ( e jω ) =
≤ 1. Also, H (z ) is stable since
∑ ∑ e k
2
M r =0 k =0
Ai ( z ), 0 ≤ i ≤ M − 1, are stable allpass functions. Hence, H (z ) is BR.

7.75 H ( z) =

Not for sale

212

1

1

7.76 H BP ( z) = [1 − A( z)] and H BS ( z) = [1 + A( z)] where
2
2
A( z ) =

α − β (1 + α )z −1 + z −2

is an allpass function. Note
1 − β (1 + α )z −1 + αz − 2
H BP ( z ) + H BS ( z ) = A( z ) and from the solution of Problem 7.73,
2

2

H BP (e jω ) + H BS (e jω ) = 1. Hence, H BP (z) and H BS (z) are doublycomplementary pair.
7.77 H ( z )H ( z −1 ) + H (− z)H (− z −1 ) = K . On the unit circle, this reduces to

H (e jω )H (e − jω ) + H (−e jω )H (−e − jω ) = K , or, equivalently,
2

2

H (e jω ) + H (−e jω ) = K , as H (z ) is a real-coefficient transfer function. Now,
2

2

H (−e jω ) = H (e j (π + ω) ) . Hence, for ω = π / 2, the power-symmetric condition
2

2

reduces to H (e jπ / 2 ) + H (e j (π + π / 2) ) = K . Since H (z ) is a real-coefficient
transfer function, H (e jω )
2

2

is an even function of ω, and thus,
2

2

H (e jπ / 2 ) = H (e j (2π − π / 2) ) . As a result, 2 H (e jπ / 2 ) = K , from which we
obtain 10 log10 2 + 20 log10 H (e jπ / 2 ) = 10 log10 K , or,
20 log10 H (e jπ / 2 ) = 10 log10 K − 3 dB.

7.78 H ( z ) = A0 ( z 2 ) + z −1 A1 ( z 2 ). Therefore,
H ( z )H ( z −1 ) + H ( − z ) H ( − z −1 ) = [ A0 ( z 2 ) + z −1 A1 ( z 2 )][ A0 ( z −2 ) + zA1 ( z −2 )]

+ [ A0 (( − z )2 ) − z −1 A1(( − z )2 )] [ A0 (( − z )2 ) − z A1(( − z )2 )]
= A0 ( z 2 ) A0 ( z −2 ) + z −1 A1 ( z 2 ) A0 ( z −2 ) + z A0 ( z 2 ) A1 ( z −2 ) + A1 ( z 2 ) A1 ( z −2 )

+ A0 ( z 2 ) A0 ( z −2 ) − z −1 A1 ( z 2 ) A0 ( z −2 ) − z A0 ( z 2 ) A1 ( z −2 ) + A1 ( z 2 ) A1 ( z −2 ) = 4, as
A0 ( z 2 ) A0 ( z −2 ) = A1 ( z 2 ) A1 ( z −2 ) = 1.

Not for sale

213

7.79 H ( z) =

− 0.1 + 0.5z −1 + 0.05z −2 + 0.05z −3 + 0.5z −4 − 0.1z −5

where A( z) =

1 + 0.1z

−2

− 0.2 + 0.1z −1 + z −2
1 + 0.1z −1 − 0.2 z − 2

− 0.2 z

−4

= [ A( z 2 ) + z −1 ],
1
2

is a stable allpass function. Thus,

H ( z )H ( z −1 ) + H (− z )H (− z −1 ) = [ A( z 2 ) + z −1 ][ A( z − 2 ) + z ]
1
4

+ [ A( z 2 ) − z −1 ][ A( z − 2 ) − z] = 1.
1
4

7.80 (a) H a ( z ) = 1 − 2 z −1 + 4.5z −2 + 6 z −3 + z −4 + 0.5z −5 . Thus, H a ( z )H a ( z −1 )

= (1 − 2 z −1 + 4.5z −2 + 6 z −3 + z −4 + 0.5z −5 )(1 − 2 z + 4.5z 2 + 6 z 3 + z 4 + 0.5z 5 )
= 0.5z 5 + 6.25z 3 + 22.5z + 62.5 + 22.5z −1 + 6.25 z −3 + 0.5z −5 .

Next, we compute H a ( − z ) H a (− z −1 ) =

= (1 + 2 z −1 + 4.5z −2 − 6 z −3 + z −4 − 0.5z −5 )(1 + 2 z + 4.5z 2 − 6 z 3 + z 4 − 0.5z 5 )
= −0.5z 5 − 6.25 z 3 − 22.5z + 62.5 − 22.5z −1 − 6.25 z −3 − 0.5z −5 .

Hence, H a ( z ) H a ( z −1 ) + H a ( − z ) H a ( − z −1 ) = 125.
(b) H b ( z) = 1 + z −1 + z − 2 − z − 4 + 2 z − 5 . Thus, H a ( z )H a ( z −1 )
4
2
1
15
15
1
H b ( z)H b ( z −1 ) = ⎛⎜1 + z −1 + z − 2 − z − 4 + 2 z − 5 ⎞⎟⎛⎜1 + z + z 2 − z 4 + 2 z 5 ⎞⎟
2
4
4
2

⎠⎝

1

15

= 2 z 5 + 7 z 3 + 0.375 z + 20.3125 + 0.375 z −1 + 7 z −3 + 2 z −5 .

Next, we compute H b (− z )H b ( − z −1 ) =

= (1 − 0.5z −1 + 3.75z −2 − z −4 − 2 z −5 )(1 − 0.5z + 3.75z 2 − z 4 − 2 z 5 )
= −2 z 5 − 7 z 3 − 0.375 z + 20.3125 − 0.375 z −1 − 7 z −3 − 2 z −5 .

Hence, H a ( z )H a ( z −1 ) + H a ( − z ) H a (− z −1 ) = 40.625.
7.81 H ( z )H ( z −1 ) = a 2 (1 + bz −1 )(1 + bz) = a 2 bz + a 2 (1 + b 2 ) + a 2 bz −1 = cz + d + cz −1 .
Thus, c = a 2 b and d = a 2 (1 + b 2 ). Now, H ( z)H ( z −1 ) + H (− z)H (− z −1 )
cz + d + cz −1 − cz + d − cz −1 = 2 d. Therefore, 2 d = 2 a 2 (1 + b 2 ) = 1. This
1

condition is satisfied by a =
1
2

2(1+ b 2 )

b = −1, a = , and b = 10, a =

1
10

1
2

. For b = 1, a = . Other solutions include

.

Since H (z ) is a first-order causal FIR transfer function, G( z) = − z −1 H (− z −1 ) is
also a first-order causal FIR transfer function. Now,

Not for sale

214

H ( z )H ( z −1 ) + G( z)G( z −1 ) = H ( z )H ( z −1 ) + [ − z −1 H (− z −1 )][− zH (− z )]
= H ( z )H ( z −1 ) + H (− z )H (− z −1 ) = 1. Hence, H (z ) and G (z ) are power
complementary.
7.82 H ( z)H ( z −1 ) = (cz + d + cz −1 )[ d 2 z 2 + d1 (1 + d 2 )z + (1 + d12 + d 22 ) + d1 (1 + d 2 )z −1 + d 2 z −2 ]
Thus, H ( z)H ( z −1 ) + H (− z)H (− z −1 ) = 2[cd1 (1 + d 2 )z 2 + dd 2 z 2 + 2cd1 (1 + d 2 )

+ d (1 + d12 + d 22 ) + dd 2 z −2 + cd1 (1 + d 2 )z −2 ] = 1. Hence, we require
dd2 + cd1 (1 + d 2 ) = 0 and 2cd1 (1 + d 2 ) + + d (1 + d12 + d 22 ) = 1. Solving these two
d2
equations we arrive at c =
, and
d 2 (1 + d 2 )(2d 2 − 1 − d12 − d 22 )
1
1
d=−
. For d1 = d2 = 1, we get c = − and d = 1.
2
2 d 2 − 1 − d12 − d 22
Since H (z ) is a third-order causal FIR transfer function, G( z) = − z −3 H (− z −1 ) is
also a third-order causal FIR transfer function. Now,

H ( z )H ( z −1 ) + G( z )G( z −1 ) = H ( z )H ( z −1 ) + [ − z −3 H (− z −1 )][ − z 3 H (− z)]
= H ( z )H ( z −1 ) + H (− z )H (− z −1 ) = 1. Hence, H (z ) and G (z ) are power
complementary.
1

1

7.83 H 0 ( z) = [ A 0 ( z) + A1 ( z )] and H1 ( z) = [ A 0 ( z ) − A1 ( z)], where A0 ( z ) and A1 ( z)
2
2
are stable allpass transfer functions. From these two equations we obtain
H 0 ( z) + H1 ( z) = A0 ( z ), and H 0 ( z) − H1 ( z ) = A1 ( z). Moreover, we have
2

2

H 0 (e jω ) + H1 (e jω ) = 1. Choose G 0 ( z ) = H 02 ( z ) and G1 ( z) = − H12 ( z).
2

2

Hence, G 0 (e jω ) + G1 (e jω ) = H 0 (e jω ) + H 1(e jω ) = 1 .
7.84 (a) H1 ( z) = (1 + 3z −1 ). Therefore, H1 ( z)H1 ( z −1 ) = (1 + 3z −1 )(1 + 3z) ⇒
4
16
1

2

H 1 ( e jω ) =

1

2⎞
d ⎛
10 + 6 cos ω
3
⎜⎜ H1 (e jω ) ⎟⎟ = − sin ω < 0 for 0 ≤ ω < π.
. Thus,
8
dω ⎝
16

Thus, H1 (e jω ) is a montonically decreasing function of ω. The maximum value
of H1 (e jω ) = 1 is at ω = 0, and the minimum value is at ω = π. Hence, H1 ( z) is
BR.

(1 − 1.2 z −1 ). Therefore, H 2 ( z)H 2 ( z −1 ) =
(1 − 1.2 z −1 )(1 − 1.2 z) ⇒
(b) H 2 ( z) =
2.2
4.84
1

1

Not for sale

215

2

H 2 ( e jω ) =

2⎞
d ⎛
2.44 − 2.4 cos ω
2.4
⎜⎜ H 2 (e jω ) ⎟⎟ =
sin ω > 0 for
. Thus,
dω ⎝
4.84
⎠ 4.84

0 ≤ ω < π. Thus, H 2 (e jω ) is a montonically increasing function of ω. The
maximum value of H 2 (e jω ) = 1 is at ω = π, and the minimum value is at ω = 0.
Hence, H 2 ( z) is BR.

1 + α z −1
(1 + αz −1 )(1 − β z −1 )
G
z
(
)
=
(c) H 3 ( z ) =
and
= G1 ( z )G 2 ( z ), where 1
1+ α
(1 + α)(1 + β)
G2 ( z) =

2 1 + α 2 + 2α cos ω
1 − β z −1
. Now, G1 (e jω ) =
. Thus,
1+ β
(1 + α) 2

2⎞
d ⎛
2α sin ω
⎜⎜ G1 (e jω ) ⎟⎟ = −
< 0 for 0 ≤ ω < π as α > 0. As a result, G1 (e jω ) is
2
dω ⎝
(1 + α)

a monotonically decreasing decreasing function of ω. The maximum value of
G1 (e jω ) = 1 is at ω = 0, and the minimum value is at ω = π. Hence, G1 ( z) is BR.
2

Likewise, G 2 (e jω ) =

1 + β 2 − 2β cos ω
(1 + β) 2

. Thus,

2 ⎞ 2β sin ω
d ⎛
⎜⎜ G2 (e jω ) ⎟⎟ =
>0
dω ⎝
⎠ (1 + β) 2

for 0 ≤ ω < π as β > 0. Thus, H 2 (e jω ) is a montonically increasing function of

ω. The maximum value of G 2 (e jω ) = 1 is at ω = π, and the minimum value is at
ω = 0. Hence, G2 ( z) is BR. Therefore, H 3 ( z ) = G3 ( z )G2 ( z) is also BR.
(1 − 0.3z −1 )(1 + 0.2 z −1 )(1 − 0.5z −1 ) ⎛⎜ 1 − 0.3z −1 ⎞⎟⎛⎜ 1 + 0.2 z −1 ⎞⎟⎛⎜ 1 − 0.5z −1 ⎞⎟
=
.

⎟⎜
⎟⎜

2.34
1
.
3
1
.
2
1
.
5

⎠⎝
⎠⎝

Since each individual factor on the right-hand side is BR, H 4 ( z) is BR.
(d) H 4 ( z ) =

7.85 (a) H1 ( z) =

2.6 + 2.6 z −1
4.2 + z −1

A0 ( z) = 1 and A1 ( z ) =

=

1 ⎛⎜ 1 + 4.2 z −1 ⎞⎟ 1
1+
= ( A0 ( z ) + A1 ( z )), where
2 ⎜⎝
4.2 + z −1 ⎟⎠ 2

1 + 4.2 z −1

are stable allpass transfer functions. Therefore,
4.2 + z −1
H1 ( z) is BR (See solution of Problem 7.74).

Not for sale

216

(b) H 2 ( z) =

1.6 − 1.6 z −1

1 ⎛⎜ 1 + 4.2 z −1 ⎞⎟ 1
= 1−
= ( A ( z) − A1 ( z)), where
−1
−1 ⎟ 2 0

2
4.2 + z
4.2 + z ⎠

A0 ( z) = 1 and A1 ( z ) =

1 + 4.2 z −1

are stable allpass transfer functions. Therefore,
4.2 + z −1
H 2 ( z) is BR (See solution of Problem 7.74).
(c) H 3 ( z ) =

0.1(1 − z − 2 )
1 + 0.4 z −1 + 0.8z − 2

1 ⎛⎜
0.8 + 0.4 z −1 + z − 2
1−
2 ⎜⎝ 1 + 0.4 z −1 + 0.8z − 2

=

0.8 + 0.4 z −1 + z −2
1
are stable
= ( A0 ( z ) − A1 ( z )) , where A0 ( z) = 1 and A1 ( z) =
2
1 + 0.4 z −1 + 0.8z − 2
allpass transfer functions. Therefore, H 3 ( z) is BR (See solution of Problem 7.74).
(d) H 4 ( z ) =

4.5 + 2 z −1 + 4.5z − 2
5 + 2 z −1 + 4 z − 2

where A0 ( z) = 1 and A1 ( z ) =

=

1 ⎛⎜ 4 + 2 z −1 + 5z − 2
1+
2 ⎜⎝ 5 + 2 z −1 + 4 z − 2

⎞ 1
⎟ = ( A ( z) + A ( z)) ,
1
⎟ 2 0

4 + 2 z −1 + 5z −2

are stable allpass transfer functions.
5 + 2 z −1 + 4 z − 2
Therefore, H 4 ( z) is BR (See solution of Problem 7.74).

7.86 Since A1 ( z) and A2 ( z) are LBR, A1 (e jω ) = 1 and A2 (e jω ) = 1. Thus,

(

)

1
⎟ = A e − jφ 2 (ω) .
A1 (e jω ) = e jφ1 (ω) and A2 (e jω ) = e jφ 2 (ω) . Now, A1 ⎜
1
⎜ A ( e jω ) ⎟
⎝ 2

⎛ 1 ⎞
⎟⎟ is LBR.
Thus, A1 e − jφ 2 (ω) = 1. Thus, A1 ⎜⎜
⎝ A2 ( z) ⎠

(

)

⎛ G ( e jω ) + α ⎞
⎛ jφ(ω) + α ⎞
⎛ G( z) + α ⎞
⎟ = e jω ⎜ e

7.87 F ( z) = z⎜⎜
⎟⎟. Thus, F (e jω ) = e jω ⎜
j
ω
j
φ
(
ω
)

⎝ 1 + αG ( z ) ⎠
⎝ 1 + αG ( e ) ⎠
⎝ 1 + αe

since G (z ) is LBR. Therefore, F (e

=

(cos(φ(ω) + α )2 + (sin(φ(ω))2
(1 + α cos(φ(ω))2 + (α sin(φ(ω))2

of F (z ). Then G( z ) z = λ =

=

)

2

=

e jφ(ω) + α

2

1 + αe jφ(ω)

1 + 2α cos(φ(ω) + α 2
1 + 2α cos(φ(ω) + α 2

= 1. Let z = λ be a pole

F ( z ) − αz
1
= − , or, G(λ) = 1 / α . If α < 1, then
α
z − αF ( z ) z = λ

Not for sale

217

G(λ) > 1, which is satisfied by the LBR G (z ) if λ < 1. Hence, F (z ) is LBR. The
order of F (z ) is same as that of G (z ).
G (z ) can be realized in the form of a two-pair constrained by the transfer function
F (z ) as shown below:
Y2

X1
G(z)

F(z)
Y1

X2

To this end, we express G (z ) in terms of F (z ) arriving at
G( z) =

− α + z −1 F ( z )

=

C + D F (z)
, where A, B, C , and D are the chain
A + B F (z)

1 − α z −1 F ( z )
parameters of the two-pair. Comparing the above two expressions we get A = 1,

B = −α z −1 , C = −α, and D = z −1 . The corresponding transfer parameters are
given by t11 = −α, t 21 = 1, t12 = (1 − α 2 )z −1 and t 22 = α z −1.
1 ⎞
. Now, A(z ) being LBR, A(e jω ) = e jφ(ω) . Thus,
7.88 Let F ( z) = G⎛⎜

⎝ A( z ) ⎠
⎛ 1 ⎞
⎟ = G(e − jφ(ω) ). Since G (z ) is a BR function, G (e jω ) ≤ 1.
F (e jω ) = G⎜⎜

j
ω
⎝ A(e ) ⎠
⎛ 1 ⎞
⎟ ≤ 1.
Hence, F (e jω ) = G ⎜⎜
jω ⎟
⎝ A(e ) ⎠
Let z = ξ be a pole of F (z ) . Hence, F (z ) will be a BR function if ξ < 1. Let

z = λ be a pole of G (z ). Then this pole is mapped to the location z = ξ of F (z ) by
1
1
1
> 1 because of Eq.
= λ, or A(ξ) = . Hence, A(ξ) =
the relation
A( z ) z = ξ
λ
λ
(7.20). This implies, λ < 1. Thus, G (z ) is a BR function.
7.89 (a) H ( z) =

2.6(1 + z −1 )
4.2 + z −1

H ( z) + G( z) =

, G( z ) =

2.6(1 + z −1 )
4.2 + z

−1

+

H ( z)H ( z −1 ) + G( z )G( z −1 ) =
=

1.6(1 − z −1 )
4.2 + z −1

1.6(1 − z −1 )
4.2 + z

−1

=

. Now,

4.2 + z −1
4.2 + z

−1

= 1. Next,

) 2.6(1 + z) 1.6(1 − z −1 ) 1.6(1 − z)

+

4.2 + z
4.2 + z
4.2 + z −1
4.2 + z −1

2.6(1 + z

−1

(6.76 z + 13.52 + 6.76 z −1 ) + (−2.56 z + 5.12 − 2.56 z −1 )
4.2 z −1 + 18.64 + 4.2 z −1

Not for sale

=

4.2 z −1 + 18.64 + 4.2 z −1
4.2 z −1 + 18.64 + 4.2 z −1

218

= 1.

2

2

Thus, H (e jω ) + G(e jω ) = 1. Hence, H (z ) and G (z ) are both allpasscomplementary and power complementary. As a result, they are doubly
complementary.
(b) H ( z ) =

0.1(1 − z −2 )
1 + 0 .4 z − 1 + 0 .8 z − 2

H ( z ) + G( z ) =

, G( z) =

0.1(1 − z −2 )
1 + 0.4 z −1 + 0.8 z − 2

+

0 .9 + 0 .4 z −1 + 0 .9 z −2
1 + 0 .4 z − 1 + 0 .8 z − 2

0.9 + 0.4 z −1 + 0.9 z −2
1 + 0 .4 z − 1 + 0 .8 z − 2

=

. Now,
0 .8 + 0 .4 z −1 + z −2

1 + 0.4 z −1 + 0.8 z − 2

implying that and are allpass complementary. Next, H ( z )H ( z −1 ) + G ( z)G( z −1 )
=

0.1(1 − z −2 )

0.1(1 − z 2 )

1 + 0.4 z −1 + 0.8 z − 2 1 + 0.4 z + 0.8 z 2
=

=

+

0.9 + 0.4 z −1 + 0.9 z −2 0.9 + 0.4 z + 0.9 z 2

1 + 0 .4 z − 1 + 0 .8 z − 2
1 + 0.4 z + 0.8 z 2

( −0.01 z 2 + 0.02 − 0.01 z −2 ) + (0.81 z 2 + 0.72 z + 1.78 + 0.72 z −1 + 0.81 z −2 )
0.8 z 2 + 0.72 z + 1.8 + 0.72 z −1 + 0.8 z − 2

0.8 z 2 + 0.72 z + 1.8 + 0.72 z −1 + 0.8 z −2

= 1. Hence H (z ) and G (z ) are also
0.8 z 2 + 0.72 z + 1.8 + 0.72 z −1 + 0.8 z − 2
power complementary. As a result, they are doubly complementary.

2(2 + z −1 + 2 z −2 )

1
3 + 2 z −1 + 5z −2
= [1 + A( z)] where A( z ) =
is an
7.90 (a) H a ( z) =
2
5 + 2 z −1 + 3 z − 2
5 + 2 z −1 + 3 z − 2
allpass function. Hence, the power complementary transfer function of H a (z) is
given by G a ( z ) =

(b) H b ( z ) =

−1
−2 ⎤
−2

1
[1 − A(z)] = 1 ⎢1 − 3 + 2 z −1 + 5z − 2 ⎥ = 1 −−1z − 2 .
2
2 ⎢⎣ 5 + 2 z + 3z ⎥⎦ 5 + 2 z + 3z

3 + 7 .5 z −1 + 7 .5 z −2 + 3 z −3
8 + 8 z −1 + 4 z − 2 + z − 3

=

3 + 7 .5 z −1 + 7 .5 z − 2 + 3 z −3
2(1 + 0.5z −1 )(4 + 2 z −1 + z − 2 )

1 + 2 z −1 + 4 z −2
0.5 + z −1
1
and A1 ( z ) =
are
= [A0 ( z ) + A1 ( z )] where A0 ( z ) =
2
4 + 2 z −1 + z − 2
1 + 0.5z −1
allpass functions. Hence, the power complementary transfer function of H b (z) is
1
− 1 − 2.5z −1 + 2.5z −2 + z −3
.
given by Gb ( z ) = [A0 ( z) − A1 ( z )] =
2
8 + 8 z −1 + 4 z − 2 + z − 3
7.91 From Eq. (7.126) we have X1 = AY2 + BX 2 , Y1 = CY2 + DX 2 . From the first
1
B
equation, Y2 = X1 − X 2 . Substituting this in the second equation we get
A
A

Not for sale

219

C
B
⎛1

X 2 . Comparing the last two
Y1 = C ⎜ X1 − X 2 ⎟ + DX 2 = X1 +
A
A
A
⎝A

C
B
1
equations with Eq. (7.123) we arrive at t11 = , t12 =
, t 21 = , t 22 = − .
A
A
A
A

From Eq. (7.123) we have Y1 = t11 X1 + t12 X 2 , Y2 = t 21 X1 + t 22 X 2 . From the
t
1
Y2 . Substituting this in the first
second equation we get X1 = − 22 X 2 +
t 21
t 21
equation we get
⎛ t

t
t t −t t
1
Y1 = t11 ⎜⎜ − 22 X 2 +
Y2 ⎟⎟ + t12 X 2 = 11 Y2 + 12 21 11 22 X 2 . Comparing the
t 21 ⎠
t 21
t 21
⎝ t 21
last two equations with Eq. (7.126) we arrive at A =

t
, B = − 22 ,
t 21
t 21

1

t
t t −t t
C = 11 , D = 12 21 11 22 .
t 21
t 21
7.92 From Eq. (7.128a) we note t12 =

1
and t 21 = . Hence, t12 = t 21 imply
A
A

⎡ "⎤
⎡Y ' ⎤ ⎡k (1 − k 2 )z −1 ⎤ ⎡ X ' ⎤ ⎡Y " ⎤ ⎡k (1 − k 2 )z −1 ⎤ ⎡ X " ⎤
1
1 , 1 = 2
2
1 , where X1 =
7.93 ⎢ 1' ⎥ = ⎢ 1

⎢ "⎥

− k 2 z −1 ⎥⎦ ⎢⎣ X 2" ⎥⎦
− k1 z −1 ⎥⎦ ⎢⎣ X 2' ⎥⎦ ⎢⎣Y2" ⎥⎦ ⎢⎣ 1
⎢⎣ X 2 ⎥⎦
⎢⎣Y2 ⎥⎦ ⎢⎣ 1
⎡Y ' ⎤
⎢ 1' ⎥ . The transfer matrices of the two two-pairs are given by
⎣⎢Y2 ⎦⎥
⎡k
⎡k (1 − k 2 )z −1 ⎤
1
2
τ1= ⎢ 1
and
τ
=

2 ⎢
−1
− k1 z
⎥⎦
⎢⎣ 1
⎢⎣ 1

(1 − k 22 )z −1 ⎤
⎥. The corresponding chain
− k 2 z −1 ⎥⎦

⎡1
matrices are obtained using Eq. (7.128b) and are given by Γ 1 = ⎢
⎣⎢k1

k1 z −1 ⎤
⎥ and
z −1 ⎦⎥

⎡ 1 k z −1 ⎤
2
Γ2= ⎢
⎥. Therefore, the chain matrix of the Γ-cascade is given by
z −1 ⎥⎦
⎢⎣k 2
⎡ 1 k z −1 ⎤ ⎡ 1 k z −1 ⎤ ⎡1 + k k z −1 k z −1 + k z −1 ⎤
1
2
1 2
2
1
Γ1Γ 2 = ⎢
. Next using Eq.
=
−1 ⎥ ⎢
−1 ⎥ ⎢
−1
−1
−2 ⎥
k
z
k
z
k
k
z
k
k
z
z
+
+
2
1 2
⎦⎥
⎦⎥ ⎣⎢ 1
⎦⎥ ⎣⎢ 2
⎣⎢ 1
(7.128a) we arrive at the transfer matrix of the Γ-cascade as

Not for sale

220

⎡ k + k z −1
2
⎢ 1
⎢1 + k1k 2 z −1
τ =⎢
1

⎢ 1 + k k z −1
1 2

z − 2 (1 − k12 )(1 − k 22 ) ⎤

1 + k1k 2 z −1
.
−1
−1 ⎥
z (k 2 + k1 z ) ⎥

1 + k1k 2 z −1

⎡ Y ' ⎤ ⎡X " ⎤
k2 z −1 ⎤ ⎡ Y2" ⎤
,
where
⎢ 2' ⎥ = ⎢ 1" ⎥. The
−1 ⎥ ⎢ " ⎥
z
⎥⎦ ⎢⎣ X 2 ⎥⎦
⎢⎣ X 2 ⎥⎦ ⎢⎣Y1 ⎥⎦
⎡ 1 k z −1 ⎤
1
chain matrices of the two two-pairs are given by Γ 1 = ⎢
and
−1 ⎥
⎣⎢k1 z ⎦⎥

⎡X ' ⎤ ⎡ 1
7.94 ⎢ 1' ⎥ = ⎢
⎢⎣ Y1 ⎥⎦ ⎢⎣k1

⎡1
Γ2= ⎢
⎢⎣k 2

k1 z −1 ⎤ ⎡ Y2' ⎤
⎥⎢ ⎥
z −1 ⎥⎦ ⎢⎣ X 2' ⎥⎦

⎡X " ⎤ ⎡ 1
⎢ 1" ⎥ = ⎢
⎢⎣Y1 ⎥⎦ ⎢⎣k 2

k 2 z −1 ⎤
⎥. The corresponding transfer matrices are obtained using Eq.
z −1 ⎥⎦

⎡k (1 − k 2 )z −1 ⎤
⎡k (1 − k 2 )z −1 ⎤
1
2
2
(7.128a) and are given by τ 1 = ⎢ 1
and
τ
=
⎥. The

2 ⎢
−1
− k 2 z −1 ⎦⎥
− k1 z
⎦⎥
⎣⎢ 1
⎣⎢ 1
transfer matrix τ of the τ-cascade is therefore given by
⎡k (1 − k 2 )z −1 ⎤ ⎡k (1 − k 2 )z −1 ⎤
2
1
1
τ 2 τ1= ⎢ 2

−1 ⎥ ⎢
− k2 z
− k1z −1 ⎥⎦
⎥⎦ ⎢⎣ 1
⎢⎣ 1
⎡k k + z −1 (1 − k 2 ) z −1k (1 − k 2 ) − z − 2 k (1 − k 2 ) ⎤
2
2
1
1
2
. Using Eq. (7.128b) we thus
=⎢ 1 2
−1
−1
2
−1
2 −2 ⎥
k
k
z
z
k
z
k
k
z
(
1
)
(
1
)

+

1
2
1
2
1
⎦⎥
⎣⎢
arrive at the chain matrix of the τ-cascade as

− (k1k 2 z −1 + 1 − k12 )z −1 ⎤
1

k1 − k 2 z −1
k1 − k 2 z −1
Γ= ⎢
⎥.
−1
2
−2
+

k
k
z
k
(
1
)
z
1
2
2

−1
−1

k
k
z
k
k
z
1
2
1
2

7.95 (a) Analyzing Figure P7.10(a) we obtain Y2 = X1 − k m z −1 X 2 and
2 −1
Y1 = k m Y2 + z −1 X 2 = k m ( X1 − k m z −1 X 2 ) z −1 X 2 = k m X1 + (1 − k m
)z X 2 .
2 −1 ⎤
⎡k
(1 − k m
)z
m
=
Hence, the transfer matrix is given by τ ⎢
⎥. Using Eq. (7.128b)
− k m z −1 ⎦⎥
⎣⎢ 1

⎡1
we then arrive at the chain matrix Γ = ⎢
⎢⎣ k m

Not for sale

k m z −1 ⎤
⎥.
z −1 ⎥⎦

221

(b) Labeling V1 the output variable of the top left adder connected to input X1 we
then analyze Figure P7.10(b) and obtain V1 = k m ( X 1− z −1 X 2 ), Y1 = V1 + X 1
= (1 + k m ) X 1− k m z −1 X 2 , and Y2 = V1 + z −1 X 2 = k m X1 + (1 − k m ) z −1 X 2 . Hence,
⎡1 + k
− k m z −1 ⎤
m
the transfer matrix of the two-pair is given by τ = ⎢
⎥. Using
(1 − k m )z −1 ⎦⎥
⎣⎢ k m

⎡ 1

Eq. (7.128b) we then arrive at the chain matrix Γ = ⎢ k m
⎢1 − k m

⎣ km
7.96 Solving G( z) =

(1 − k m )z −1 ⎤

km
⎥.

1

km

k + z −1G( z )
for H (z ) we get H ( z) = m
.
z −1[1 − k m H ( z)]
1 + k m z −1G( z)

H ( z) − k m

For the constrained two-pair H ( z ) =

C + D ⋅ G( z)
. Comparing the last two
A + B ⋅ G( z)

equations we thus get C = k m , D = z −1 , A = 1, B = k m z −1 . Substituting these
values of the chain parameters in Eq. (7.128a) we get
C
1
B
2
t11 = = k m , t12 =
= z −1 (1 − k m
), t 21 = = 1, t 22 = − = − k m z −1 .
A
A
A
A
7.97 (a) H ( z ) =

α + z −1G ( z )

1 + αz −1G( z )

. For the constrained two-pair H ( z ) =

C + D ⋅ G( z)
.
A + B ⋅ G( z)

Comparing the last two equations we thus get C = α, D = z −1 , A = 1, B = αz −1 .
(b)
7.98 From the results of Problem 7.95, Part (a), we observe that the chain matrix of the
⎡ 1 k z −1 ⎤
i
i -th lattice two-pair is given by Γ i = ⎢
⎥, i = 1, 2, 3. Thus, the chain matrix
z −1 ⎥⎦
⎢⎣ ki
of the cascade of the three lattice two-pairs is given by
⎡ 1 k z −1 ⎤ ⎡ 1 k z −1 ⎤ ⎡ 1 k3 z −1 ⎤
1
2

⎥⎢
−1 ⎥ ⎢
z −1 ⎦⎥ ⎣⎢ k3
z −1 ⎦⎥
⎦⎥ ⎣⎢ k 2
⎣⎢ k1 z

⎡ 1 + k k z −1 + k z −1 (k + k z −1 )
k3 z −1 + k 2 z −1 + k1 z −1 (k 2 k3 + z − 2 ) ⎤
2 3
1
2
3
=⎢
.
−1
−1
−1
−1
−1
−1
−2 ⎥
⎢⎣ k1 (1 + k 2 k3 z ) + z (k 2 + k3 z ) k1 (k3 z + k 2 z ) + z (k 2 k3 + z )⎥⎦
C+D
From Eq. (7.135a) we obtain A3 ( z ) =
A+ B

Not for sale

222

1 + (k2 k3 + k1k2 + k3 )z −1 + (k1k3 + k 2 + k1k 2 k3 )z −2 + k1 z −3
which is seen to be
k1 + (k1k3 + k 2 + k1k 2 k3 )z −1 + (k 2 k3 + k1k 2 + k3 )z − 2 + z − 3
an allpass function.
=

7.99 Let

D( z) = 1 + d1 z −1 + d 2 z −2 = (1 − λ1 z −1 )(1 − λ2 z −1 ) = 1 − ( λ1 + λ2 )z −1 + λ1 λ2 z −2 .
Thus, d 2 = λ1 λ2 and d1 = −( λ1 + λ2 ). For stability, λi < 1, i = 1, 2. As a result,
d 2 = λ1 λ2 < 1.
Case 1: Complex poles: d 2 > 0. In this case, λ2 = λ*1 . Now,
λ1 , λ2 =

− d1 ± d12 − 4d 2

. Hence, λ1 and λ2 will be complex, if d12 < 4d 2 . In

2
d
j
4d 2 − d12 . Thus,
this case, λ1 = − 1 +
2 2
Consequently, if the poles are complex and
circle.

λ1

2

=

1
4

(d12 + 4d2 − d12 ) = d2 < 1.

d 2 < 1, then they are inside the unit

Case 2: Real poles. In this case we get − 1 < λi < 1, i = 1, 2. Since, λi < 1, it follows
then d1 < λ1 + λ2 < 2. Now, − 1 <

− d1 ± d12 − 4 d 2
2

< 1, or

± d12 − 4 d 2 < 2 + d1 . It is not possible to satisfy the inequality on the right hand
side with a minus sign in front of the square root as it would imply then d1 < −2.
Therefore,

d12 − 4d 2 < 2 + d1 , or d12 − 4 d 2 < 4 + d12 + 4d1 , or − d1 < 1 + d 2 .
Similarly,

− d1 ± d12 − 4d 2

(7-x)

< −1, or ± d12 − 4d 2 > −2 + d1 . Again it is not

2
possible to satisfy the inequality on the right hand side with a plus sign in front of
the square root as it would imply then d1 > 2. Therefore, − d12 − 4 d 2 > −2 + d1 ,
or

d12 − 4d 2 < 2 − d1 , or d12 − 4 d 2 < 4 + d12 − 4d1 , or equivalently,
d1 < 1 + d 2 .

(7-y)

Combining Eqs. (7-x) and (7-y) we get d1 < 1 + d 2 .
7.100 (a) D a ( z ) = 4(1 + 0.75z −1 + 0.25z −2 ) ⇒ d 1= 0.75, d 2 = 0.5. Since

d 2 = 0.5 < 1 and 1 + d 2 = 1.5, d1 = 0.75 < 1 + d 2 . Hence, both roots of Da (z ) are
inside the unit circle.

Not for sale

223

(b) Db ( z ) = 2(1 + 0.5z −1 + 0.5z −2 ) ⇒ d 1= 0.5, d 2 = 0.5. Since d 2 = 0.5 < 1 and
1 + d 2 = 1.5, d1 = 0.5 < 1 + d 2 . 1 + d 2 = 1.5, d1 = 0.75 < 1 + d 2 . Hence, both
roots of Db (z ) are inside the unit circle.
(c) Dc ( z) = 3(1 + z −1 − z − 2 ) ⇒ d 1= , d 2 = − . Since d 2 = > 1, at least
3
3
3
3
3
one root of Dc (z ) is outside the unit circle.
4

4

4

4

4

(d) Dd ( z) = 3(1 − z −1 − z − 2 ) ⇒ d 1= − , d 2 = − . Since d 2 = < 1 and
3
3
3
6
6
1

1

1

1

1

1

2

1 + d 2 = , d1 = < 1 + d 2 . Hence, both roots of Dd (z) are inside the unit circle.
3
3
7.101 (a) A3 ( z) =

0.25 + 0.5z −1 + 0.75z −2 + z −3
1 + 0.75z −1 + 0.5z − 2 + 0.25z − 3

(7.148) we arrive at A2 ( z ) =

+ z −1 + z − 2

1
3

2
3

1

. Here, k 2 = < 1. Continuing this
1
2
3
1 + z −1 + z − 2

0.5 + z

process we get A1 ( z) =

. Note k3 = 0.25 < 1. Using Eq.

3
−1

1 + 0.5z −1

3

. Finally, k1 = 0.5 < 1. Since ki < 1, i = 1, 2, 3,

H a (z) is stable.
(b) A3 ( z) =

− − z −1 + z − 2 + z − 3
1
3

2
3
2 −1
z
3

1+

arrive at A2 ( z) =

2
3
2 −2 1 −3
z − z
3
3
−1
−2

− 0.5 + 0.5z

+z

1 + 0.5z −1 − 0.5z − 2

process we get A1 ( z ) =

1 + z −1
1 + z −1

. Note k3 =

1
3

< 1. Using Eq. (7.148) we

. Here, k 2 = 0.5 < 1. Continuing this

. Finally, k1 = 1. Since k1 is not less than 1 ,

H b (z) is not stable.

(c) A4 ( z) =

− + z −1 + z − 2 + z − 3 + z − 4
1
6

1+

1
2
2 −1
z
3

we get A3 ( z) =

+

2
3
2 −2
z
3

+

2
3
1 −3 1 − 4
z − z
6
2
−1

0.6286 + 0.8z

+ 0.7714z −2 + z −3

1 + 0.7714z −1 + 0.8z − 2 + 0.6286z − 3

Using Eq. (7.148) we next get A2 ( z) =

1

. Note k 4 = < 1. Using Eq. (7.148)
6

. Note k3 = 0.6286 < 1.

0.5209 + 0.444z −1 + z −2

1 + 0.444z −1 + 0.5209z − 2

Not for sale

. Here,

224

k 2 = 0.209 < 1. Continuing this process we get A1 ( z) =

0.2919 + z −1

1 + 0.2919z −1
k1 = 0.2919 < 1. Since ki < 1, i = 1, 2, 3, 4, H c (z ) is stable.
(d) A4 ( z) =

0.2 + 0.4 z −1 + 0.6 z − 2 + 0.8 z − 3 + z − 4
1 + 0.8z −1 + 0.6 z − 2 + 0.4 z

Eq. (7.148) we get A3 ( z) =

2
3
−3

+ 0.2 z − 4

. Note k 4 = 0.2 < 1. Using

0.25 + 0.5z −1 + 0.75z −2 + z −3

. Note k3 = 0.25 < 1.

1 + 0.75z −1 + 0.5z − 2 + 0.25z − 3

Using Eq. (7.148) we next get A2 ( z ) =
Continuing this process we get A1 ( z) =

1
3

+ z −1 + z − 2

1+

2
3

2 −1 1 − 2
z + z
3
3
−1

0.5 + z

1 + 0.5z −1

. Finally,

1

. Here, k 2 = < 1.
3

. Finally, k1 = 0.5 < 1. Since

ki < 1, i = 1, 2, 3, 4, H d (z) is stable.
(e) A5 ( z) =

0.1 + 0.2 z
−1

−1

+ 0.3z −2 + 0.5z −3 + 0.7z −4 + z −5

1 + 0.7z + 0.5z − 2 + 0.3z − 3 + 0.2 z − 4 + 0.1z − 5
Using Eq. (7.148) we get
A4 ( z ) =

. Note k 5 = 0.1 < 1.

0.1313 + 0.2525z −1 + 0.4747 z −2 + 0.6869z −3 + z −4

1 + 0.6869z −1 + 0.4747 z − 2 + 0.2525z − 3 + 0.1313z − 4
k 4 = 01313 < 1. Using Eq. (7.148) we next get

A3 ( z) =

0.1652 + 0.4196z −1 + 0.6652z −2 + z −3

1 + 0.6652z −1 + 0.4196z − 2 + 0.1652z − 3

Eq. (7.148) we next get A2 ( z ) =

. Note

. Note k3 = 0.1652 < 1. Using

0.3185 + 0.6126z −1 + z −2

1 + 0.6126z −1 + 0.3185z − 2

k 2 = 0.3185 < 1. Continuing this process we get A1 ( z) =

. Here,

0.4646 + z −1

1 + 0.4646z −1
k1 = 0.4646 < 1. Since ki < 1, i = 1, 2, 3, 4, 5, H e (z) is stable.

7.102

(a) A5 ( z ) =

0.1 + 0.2 z
−1

−1

+ 0.4 z −2 + 0.6 z −3 + 0.8z −4 + z −5

1 + 0.8z + 0.6 z − 2 + 0.4 z − 3 + 0.2 z − 4 + 0.1z − 5
k 5 = 0.1 < 1. Using Eq. (7.148) we get

A4 ( z ) =

0.1212 + 0.3434z −1 + 0.5657 z −2 + 0.7879z −3 + z −4

1 + 0.7879z −1 + 0.5657 z − 2 + 0.3434z − 3 + 0.1212z − 4
k 4 = 0.1211 < 1. Using Eq. (7.148) we next get

Not for sale

. Finally,

. Note

. Note

225

A3 ( z) =

0.2516 + 0.5045z −1 + 0.7574z −2 + z −3

1 + 0.7574z −1 + 0.5045z − 2 + 0.2516z − 3

Eq. (7.148) we next get A2 ( z ) =

. Note k3 = 0.2516 < 1. Using

0.3351 + 0.6730z −1 + z −2

1 + 0.6730z −1 + 0.3351z − 2

. Here,

0.5041 + z −1

k 2 = 0.3351 < 1. Continuing this process we get A1 ( z ) =

. Finally,
1 + 0.5041z −1
k1 = 0.5041 < 1. Since ki < 1, i = 1, 2, 3, 4, 5, D a (z) has all roots inside the unit
circle..
(b) A5 ( z) =

− 0.25 + 0.5z −1 + 0.625z −2 + 0.75z −3 + 0.875z −4 + z −5

1 + 0.875z −1 + 0.75z − 2 + 0.625z − 3 + 0.5z − 4 − 0.25z − 5
k 5 = 0.25 < 1. Using Eq. (7.148) we get

A4 ( z ) =

0.7667 + 0.8667 z −1 + 0.9667 z −2 + 1.0667 z −3 + z −4

1 + 1.0667 z −1 + 0.9667 z − 2 + 0.8667 z − 3 + 0.7667 z − 4
k 4 = 0.7667 < 1. Using Eq. (7.148) we next get

A3 ( z ) =

0.1186 + 0.5472z −1 + 0.9757 z −2 + z −3

1 + 0.9757z −1 + 0.5472z − 2 + 0.1186z − 3

Eq. (7.148) we next get A2 ( z ) =

. Note

. Note

. Note k3 = 0.1186 < 1. Using

0.4376 + 0.9238z −1 + z −2

1 + 0.9238z −1 + 0.4376z − 2

k 2 = 0.4376 < 1. Continuing this process we get A1 ( z) =

. Here,

0.6426 + z −1

. Finally,
1 + 0.6426z −1
k1 = 0.6426 < 1. Since ki < 1, i = 1, 2, 3, 4, 5, D b (z) has all roots inside the unit
circle..
M7.1 The MATLAB code fragments used to simulate the FIR filter are
b = [3.8461 -6.3487 3.8461];zi = [0 0];
n=0:49;x1 = cos(0.3*n);x2=cos(0.6*n);
y = filter(b,1,x1+x2,zi);
The plot generated by the above program is shown below:
2

y[n]
x 1[n]
x 2[n]

Amplitude

1.5
1
0.5
0
-0.5
-1

0

10

20
30
Time index n

Not for sale

40

226

M7.2 The MATLAB code fragments used to simulate the FIR filter are:
b =[17.7761 -58.7339 83.8786 -58.7339 17.7761];
n = 0:49; x1 = cos(0.3*n);x2=cos(0.5*n);x3=cos(0.8*n);
y = filter(b,1,x1+x2+x3);
The plot generated by the above program is shown below:
3

Amplitude

2
1
0
-1
-2
-3

y[n]
x1[n]+x3[n]
0

10

20
30
Time index n

40

M7.3 The gain response of H (z ) is shown below:
20
0

Gain, dB

-20
-40
-60
-80
-100

0

0.2

0.4

0.6

0.8

1

ω/ π

M7.4 (a) The MATLAB code fragments used to evaluate H ( z)H ( z −1 ) + H (− z)H (− z −1 ) is
shown below:
n = [1 -2 3.5];n = [n fliplr(n)];
d = [8 0 -2 0 -1 0];
k = 0:5;n1 = (-1).^k.*n;
a = conv(n,fliplr(n))+conv(n1,fliplr(n1));
b = conv(d,fliplr(d));
The numerator and denominator of are H ( z)H ( z −1 ) + H (− z)H (− z −1 ) given by
the vectors a and b :
a =
0
-8
0 -14
0
69
0 -14
0
-8
0
b =
0
-8
0 -14
0
69
0 -14
0
-8
0
Hence H ( z )H ( z −1 ) + H (− z)H (− z −1 ) = 1 verifying the power-complementary
property of H (z ) and H ( − z ).

Not for sale

227

(b) The MATLAB code fragments used to evaluate H ( z)H ( z −1 ) + H (− z)H (− z −1 ) is
shown below:
n = [1 1.5 5.25 7.25];n = [n fliplr(n)];
d = [12 0 13 0 4.5 0 0.5 0];
k = 0:7;n1 = (-1).^k.*n;
a = conv(n,fliplr(n))+conv(n1,fliplr(n1));
b = conv(d,fliplr(d));
The numerator and denominator of are H ( z)H ( z −1 ) + H (− z)H (− z −1 ) given by
the vectors a and b :
a =
Columns 1 through 10
0
6.0000
0
60.5000
0 216.7500
0 333.5000
0 216.7500
Columns 11 through 15
0
60.5000
0
6.0000
0
b =
Columns 1 through 10
0
6.0000
0
60.5000
0 216.7500
0 333.5000
0 216.7500
Columns 11 through 15
0
60.5000
0
6.0000
0
Hence H ( z )H ( z −1 ) + H (− z)H (− z −1 ) = 1 verifying the power-complementary
property of H (z ) and H ( − z ).
M7.5 The magnitude and phase responses of H (z ) are shown below:
4

Magnitude

1
0.8
0.6
0.4

2
0
-2

0.2
0

0

0.2

0.4

0.6

0.8

1

-4

0

0.2

0.4

ω/π

0.6

0.8

1

ω/π

From the magnitude response plot given above it can be seen that represents a
highpass filter. The difference equation representation of H (z ) is given by
y[ n] + 0.7074 y[ n − 1] + 0.7976 y[ n − 2] + 0.2004 y[ n − 3]
= 0.2031x[ n] − 0.2588 x[ n − 1] + 0.2588 x[ n − 2] − 0.2031x[ n − 3].
M7.6 The magnitude and phase responses of H (z ) are shown below:

Not for sale

228

5

4

Magnitude

4
3
2
1
0

0

0.2

0.4

0.6

0.8

2
0
-2
-4

1

0

ω/π

0.2

0.4

0.6

0.8

1

ω/π

From the magnitude response plot given above it can be seen that represents a
bandpass filter. The difference equation representation of H (z ) is given by
y[ n] + 0.6402 y[n − 1] + 1.7497 y[ n − 2] + 0.5354 y[ n − 3] + 0.7015 y[ n − 4]
= 0.2031x[ n] − 0.2588 x[ n − 1] + 0.2588 x[ n − 2] − 0.2031x[ n − 3].
M7.7 Here K = 5. Using Eq. (7.85) we obtain first C = 1.7411. Then using Eq. (7.84)
we obtain α = −0.3779. From Eq. (7.71) we get the transfer function of the
lowpass filter as H LP ( z ) =

0.6889(1 + z -1 )

. A plot of the gain response of a
1 − 0.3779 z −1
cascade of 5 lowpass filters is shown below:
Passband Details
5

-1
-2
Gain, in dB

Gain, in dB

0
-5
-10
-15
-20
-25
-30

-3
-4
-5

0

0.2

0.4

0.6

0.8

1

ω/π

0.4 0.42 0.44 0.46 0.48 0.5 0.52
ω/π

M7.8 From the solution of Problem 7.46 we observe

sin ωc 2C − C 2 − (1 − cos ωc + C cosωc )
, where C = 2 ( K −1) / K .
α=
1 − cos ωc − C
Substituting K = 6 in the second equation we get C = 1.7818. Next, substituting
this value of C and ωc = 0.4 π in the first equation we arrive at α = 0.5946.
From Eq. (7.74) we get the transfer function of the lowpass filter as
H HP ( z ) =

0.7973(1 − z -1 )

. A plot of the gain response of a cascade of 6
1 − 0.5946 z −1
highpass filters is shown below:

Not for sale

229

Passband Details
5

-2
-4
Gain, in dB

Gain, in dB

0
-5
-10
-15
-20

-8
-10

-25
-30

-6

-12
0

0.2

0.4

0.6

0.8

0.2

1

0.3

0.4
ω/ π

ω/ π

0.5

0.6

M7.9 Using Eq. (7.73b) we obtain α = −0.1584. Sunstituting this value of α in Eqs.

0.4208(1 − z -1 )
.
and H HP ( z ) =
1 − 0.1584 z −1
1 − 0.1584 z −1
Plots of the magnitude responses of H LP (z ) and H HP (z ) along with the plot of
the magnitude response of H LP ( z ) + H HP ( z ) and plot of

(7.71) and (7.74) we get H LP ( z ) =

2

2

H LP (e jω ) + H HP(e jω )

0.5792(1 + z -1 )

are shown below verifying the doubly complementary

property of H LP (z ) and H HP (z ).
1

HLP(z)

HHP(z)
Magniitude

Magniitude

0.8
0.6
0.4
0.2
0

|HLP(e )+HHP(e )|

1
0.8
0.6
0.4
0.2

0

0.2

0.4

0.6

0.8

0

1

0

0.2

0.4

ω/ π
jω 2

Square Magniitude

0.6

0.8

1

ω/ π
jω 2

|HLP(e )| +|HHP(e )|

1
0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/ π

M7.10 From Eq. (7.78) we arrive at the quadratic equation α 2 − 2.8284 α + 1 = 0 whose
solution yields α = 0.4142 and α = 2.4142. A stable bandpass and bandstop
transfer function requires α < 1. Hence we choose α = 0.4142. Next, from Eq.

Not for sale

230

(7.76) we get β = 0.3090. Substituting these two parameters in Eqs. (7.77) and
(7.80) we obtain H BP ( z ) =
H BS ( z ) =

0.2929(1 − z −2 )
1 − 0.4370 z −1 + 0.4142 z − 2

and

0.7071(1 − 0.6180 z -1 + z −2 )

. Plots of the magnitude responses of
1 − 0.4370 z −1 + 0.4142 z − 2
H BP (z ) and H BS (z ) along with the plot of the magnitude response of
2

H BP ( z ) + H BS ( z ) and plot of H BP(e jω ) + H BS (e jω )

2

are shown below

verifying the doubly complementary property of H BP (z ) and H BS (z ).

0.8
0.6
0.4
0.2
0

|HBP(e )+HBS(e )|

1
Magniitude

Magniitude

HBS(z)

HBP(z)

1

0.8
0.6
0.4
0.2

0

0.2

0.4

0.6

0.8

0

1

0

0.2

0.4

ω/ π
jω 2

Square Magniitude

0.6

0.8

1

ω/ π
jω 2

|HBP(e )| +|HBS(e )|

1
0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/ π

M7.11 From Eq. (7.78) we arrive at the quadratic equation α 2 − 2.4721α + 1 = 0 whose
solution yields α = 0.5095 and α = 1.9626. A stable bandpass transfer function
requires α < 1. Hence we choose α = 0.5095. Next, from Eq. (7.76) we get
β = −0.3090. Substituting these two parameters in Eqs. (7.77) we obtain

H BP ( z ) =

0.2453(1 − z −2 )

1 + 0.4665 z −1 + 0.5095 z − 2
H BP (z ) is shown below:

. A plot of the magnitude response of

Not for sale

231

Magniitude

1
0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/ π

M7.12 From Eq. (7.78) we arrive at the quadratic equation α 2 − 2.4721α + 1 = 0 whose
solution yields α = 0.5095 and α = 1.9626. A stable bandstop transfer function
requires α < 1. Hence we choose α = 0.5095. Next, from Eq. (7.80) we get
β = −0.3090. Substituting these two parameters in Eqs. (7.77) we obtain

H BS ( z ) =

0.7548(1 + 0.6180 z −1 + z −2 )

1 + 0.4665 z −1 + 0.5095 z − 2
H BS (z ) is shown below:

. A plot of the magnitude response of

Magniitude

1
0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/ π

M7.13 A plot of the magnitude response of H BP ( z ) + H BS ( z ) and plot of H BP(e jω )
+ H BS (e jω )

2

2

are shown below verifying the doubly complementary property of

H BP (z ) and H BS (z ).

Not for sale

232

j

Square Magniitude

Magniitude

j

0.8
0.6
0.4
0.2
0

0

0.2

|HBP(e ω )| +|HBS(e ω )|
j

|HBP(e ω )+HBS(e ω )|

1

0.4

0.6

0.8

1

j

2

0.8
0.6
0.4
0.2
0

1

2

0

0.2

0.4

ω/ π

0.6

0.8

1

ω/ π

M7.14 (a)

jω 2

Square Magnitude

Magnitude

|H(e )+G(e )|

1
0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

0.8
0.6
0.4
0.2
0

1

jω 2

|H(e )| +|G(e )|

1

0

0.2

0.4

ω/π

0.6

A plot of the magnitude response of H ( z ) + G ( z ) and plot of H (e jω )

+ G ( e jω )

0.8

1

ω/π

2

2

are shown above verifying the doubly complementary property of H (z )

and G (z ).
(b)

jω 2

|H(e )+G(e )|
Square Magnitude

Magnitude

0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

jω 2

|H(e )| +|G(e )|

1

0.8

1

ω/π

1
0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

A plot of the magnitude response of H ( z ) + G ( z ) and plot of H (e jω )

+ G ( e jω )

0.8

1

ω/π

2

2

are shown above verifying the doubly complementary property of H (z )

and G (z ).
Not for sale

233

M7.15 (a) The pole-zero plot obtained using the M-file zplane is shown below:
Ha(z)
1
0.5

0.8
Magnitude

Imaginary Part

1

0
-0.5

0.6
0.4
0.2

-1
-1.5

-1

-0.5

0
0.5
Real Part

1

0

1.5

0

0.2

0.4

0.6

0.8

1

ω/π

It can be seen from the above pole-zero plot that the two poles of H a (z ) are
inside the unit circle and hence H a (z ) is stable. From the magnitude response
plot given above, we observe that H a (e jω ) ≤ 1 and hence, H a (z ) is a BR
function.
(b) The pole-zero plot obtained using the M-file zplane is shown below:
Hb(z)

1

0.8
Magnitude

Imaginary Part

1
0.5
0
-0.5

0.6
0.4
0.2

-1
-1.5

-1

-0.5

0
0.5
Real Part

1

1.5

0

0

0.2

0.4

0.6

0.8

1

ω/π

It can be seen from the above pole-zero plot that the three poles of H b (z ) are
inside the unit circle and hence H b (z ) is stable. From the magnitude response
plot given above, we observe that H a (e jω ) ≤ 1 and hence, H a (z ) is a BR
function.
M7.16 (a) The power-complementary transfer function G a ( z) = Q( z) / D( z ) to the
transfer function H a ( z ) = P( z ) / D( z) satisfy the relation G a ( z )G a ( z −1 )
= 1 − H a ( z ) H a ( z −1 ), or equivalently the relation Q( z )Q( z −1 ) = D( z ) D( z −1 )

− P( z)P ( z −1 ). Here, D( z ) = 1 + 0.4 z −1 + 0.6 z −2 and
P( z) = 0.8 + 0.4 z −1 + 0.8z −2 .
(b)

Not for sale

234

M7.17 (a) The pole-zero plot of H a (z ) obtained using the M-file zplane is shown
below. From this plot it can be seen that all 3 poles of H a (z ) are inside the unit
circle, and hence, H a (z ) is a stable transfer function.
Ha(z)

Imaginary Part

2
1
0
-1
-2
-2

0
Real Part

2

(b) The pole-zero plot of H b (z ) obtained using the M-file zplane is shown
below. From this plot it can be seen that one pole of H b (z ) is on the unit circle,
and hence, H b (z ) is an unstable transfer function.
Hb(z)

Imaginary Part

3
2
1
0
-1
-2
-3
-2

0

2
4
Real Part

6

8

(c) The pole-zero plot of H c (z ) obtained using the M-file zplane is shown
below. From this plot it can be seen that all 4 poles of H c (z ) are inside the unit
circle, and hence, H c (z ) is a stable transfer function.
Hc (z)

Imaginary Part

2
1
0
-1
-2
-3

-2

-1

0
1
Real Part

Not for sale

2

3

235

(d) The pole-zero plot of H d (z ) obtained using the M-file zplane is shown
below. From this plot it can be seen that all 4 poles of H d (z ) are inside the unit
circle, and hence, H d (z ) is a stable transfer function.
Hd(z)

Imaginary Part

1
0.5
4

0
-0.5
-1
-1.5

-1

-0.5

0
0.5
Real Part

1

1.5

(e) The pole-zero plot of H e (z ) obtained using the M-file zplane is shown
below. From this plot it can be seen that all 5 poles of H e (z ) are inside the unit
circle, and hence, H e (z ) is a stable transfer function.
He(z)

Imaginary Part

1
0.5
5

0
-0.5
-1
-1.5

-1

-0.5

0
0.5
Real Part

1

1.5

M7.18 (a) The output data generated by running Program 7_2 are:
The stability test parameters are
0.2500
0.3333
0.5000
stable =
1
Hence, H a (z ) is a stable transfer function.
(b) The output data generated by running Program 7_2 are:
The stability test parameters are
-0.3333
-0.5000
1.0000
stable =
0

Not for sale

236

Hence, H b (z ) is an unstable transfer function.
(c) The output data generated by running Program 7_2 are:
The stability test parameters are
-0.1667
0.6286
0.5209
0.2919
stable =
1
Hence, H c (z ) is a stable transfer function.
(d) The output data generated by running Program 7_2 are:
The stability test parameters are
0.2000
0.2500
0.3333
0.5000
stable =
1
Hence, H d (z ) is a stable transfer function.
(e) The output data generated by running Program 7_2 are:
The stability test parameters are
0.1000
0.1313
0.1652
0.3185
stable =
1

0.4646

Hence, H e (z ) is a stable transfer function.
M7.19 (a) The output data generated by running Program 7_2 are:
The stability test parameters are
0.1000
0.1212
0.2516
0.3351
stable =
1

0.5041

Hence, all roots of Da (z ) are inside the unit circle.
(b) The output data generated by running Program 7_2 are:
(c)
The stability test parameters are
-0.2500
0.7667
0.1186
0.4376
0.6426
stable =
1
Hence, all roots of Db (z ) are inside the unit circle.
Not for sale

237

k
M7.20 H ( z ) = − + ⎛⎜ 1 +
4 ⎝

k ⎞ −1
⎟z
4⎠

− z−2.
k
4

3
k=2
Magnitude

2.5
2
k=1
1.5
1

0

0.2

0.4

0.6

0.8

1

ω/π

k
k
k
k ⎞
k
k

M7.21 H ( z ) = − 2 + 1 z −1 + ⎜ 1 − 1 + 2 ⎟ z − 2 + 1 z − 3 − 2 z − 4 .
4
4
4
4 ⎠
4
4

5
k 1=2, k 2=-4

Magnitude

4
3
2

k 1=1, k 2=-2

1
0

0

0.2

0.4

0.6

0.8

1

ω/π

Not for sale

238

Chapter 8
8.1

Analyzing Figure P8.1, we get w[n] = A( X [n] + CDu[n]) and
y[n] = C ( ABx[n] + u[n]). A direct implementation of these two equations leads to
the structure shown below which has no delay-free loop.
A
w[n]

x[n]
AB
CD

y[n]

8.2

C

u[n]

(a) Figure P8.2(a) with input and output nodes removed and delays removed is as
shown below. There are two delay-free loops in the reduced structure shown
below. Analysis of this structure yields
W3

W1
α1

k1

U2

k2
W2 U 3

U1

(1): W1 = W3 + U 2 , (2): W2 = α1W1 + U1 , (3): W3 = k1W3 + k 2W2 + U 2 .
From Eq. (3) we get (4): (1 − k1 )W3 − k 2W2 = U 2 , and from Eq. (1) we get
W3 = W1 − U 2 . Substituting the last equation in Eq. (4) we get
(1 − k1 )W1 − k 2W2 = (1 − k1 )U 2 + U 3 and from Eq. (2) we have − α1W1 + W2 = U1.
k
1 − k1
1
U 2 + U 3 and
Solving the last two equations we arrive at W1 = 2 U 1 +

α (1 − k1 )
α
1 − k1
W2 =
U1 + 1
U 2 + 1 U 3 . A realization based on the above two

equations after insertion of the input and output nodes and the delays is shown
below which has no delay-free loops.

Y (z)

W1
(1_ k1)/∆
U2
k 2 /∆

α2
X(z)

_1

z

(1_ k1)/∆
U1

Not for sale

1/∆

α1
W2

_
z1

U3

239

(b) Figure P8.2(b) with input and output nodes removed and delays removed is as
shown below. There is one delay-free loop in the reduced structure shown below.
Analysis of this structure yields
δ

U1

W1 U2

W2

β
U3

(1): W1 = U1 + δW2 , and (2): W2 = U 2 + βW1 + U 3 . Substituting Eq. (1) in Eq. (2)
β
1
1
we get after some algebra (3): W2 = U 1 + U 2 + U 3 , where ∆ = 1 − βδ.

δ
βδ
δ
Substituting Eq. (3) in Eq. (1) we get (5): W1 = U 1 + U 1 + U 2 + U 3 . A

realization based on Eqs. (3) and (5) after insertion of the input and output nodes
and the delays is shown below which has no delay-free loop.
δ

X(z)

U1

1/∆ W
1

_1

z

β

1/∆

U2
_
z 1

U3

W2

Y(z)

γ

8.3 Analysis yields Y ( z ) = G ( z )( X ( z ) − C ( z )Y ( z ) ). Hence,
Y ( z)
G( z)
2
H ( z) =
=
=
. The overall transfer function has
X ( z ) 1 + G ( z )C ( z ) (1 + 2 K ) + 3 z −1
a pole at z = −

3
3
. Thus, the system is stable if
< 1, or, K > 1 or
1 + 2K
1 + 2K

K < −2.
8.4 From the results of Problem 8.3, we have H ( z ) =

G( z)
which can be
1 + G ( z )C ( z )

G( z) − H ( z)
. Substituting the expressions for G (z ) and
G( z)H ( z)
H (z ) in this expression we get

solved yielding C ( z ) =

Not for sale

240

C ( z) = −

z −1 (7.0 + 9.1z −1 + 6.87 z −2 + 1.84 z −3 + 0.04 z −4 )

. Pole-zero plots of
0.04(1 + 1.8 z −1 + 1.46 z − 2 + 0.56 z − 3 + 0.08 z − 4 )
G ( z ), C ( z ), and H (z ) obtained using zplane are shown below:
H(z)
1

0.5

0.5

Imaginary Part

Imaginary Part

G(z)
1

0
-0.5
-1

0
-0.5
-1

-1.5

-1

-0.5

0
0.5
Real Part

1

1.5

-1.5

-1

-0.5

0
0.5
Real Part

1

1.5

C(z)

Imaginary Part

1
0.5
2

0
-0.5
-1
-1.5

-1

-0.5

0
0.5
Real Part

1

1.5

Ac can be seen from the above ploe-zero plots, G ( z ), C ( z ), and H (z ) are stable
transfer functions.
8.5

The structure with internal variables is shown below. Analysis of this structure yields
K
_

_

z1

X(z)

z1

W 1(z) β

Y(z)
α

_1

W 2(z)

_1
_

z1

_

z1

W1 ( z ) = K X ( z ) − z −2Y ( z ), W2 ( z ) = − Kz −1 X ( z ) + z −1Y ( z ), and
Y ( z ) = β W1 ( z ) + αW2 ( z ) + Kz −2 X ( z ). Substituting the first two equations into the
third, and rearranging we get (1 − αz −1 + βz −2 )Y ( z ) = K (β − αz −1 + z −2 ) X ( z ).

Not for sale

241

Hence, the transfer function is given by H ( z ) =

K (β − αz −1 + z −2 )

.
1 − αz −1 + β z − 2
(a) Since the structure employs 2 delays and has a second order transfer function, it
is a canonic structure.

(b) and (c) We form
⎛ β − α z −1 + z − 2
H ( z ) H ( z −1 ) = K 2 ⎜
⎜ 1 − α z −1 + β z − 2

⎞⎛ β z − 2 − αz −1 + 1 ⎞
⎟ = K 2 . Therefore,
⎟⎜
⎟⎜ z − 2 − αz −1 + β ⎟

⎠⎝

H (e jω ) = K for all values of ω. Hence, H (e jω ) = 1 if K = 1.

8.6

The structure with internal variables is shown below. Analysis of this structure yields

+
_
z 1

W 1(z)

_k

Y(z)
2

_

k2
_
z 1

+

W2 (z)

k1

+
γ1

γ2
X(z)

W1 ( z ) = γ 2 X ( z ) + k 2Y ( z ) + z −1W2 ( z ), W2 ( z ) = γ1 X ( z ) − k1Y ( z ), and
Y ( z ) = z −1W1 ( z ) − k 2 z −1W2 ( z ). Substituting the second equation in the first and
rearranging we get W1 ( z ) = ( γ 2 + γ1 z −1 ) X ( z ) + (k 2 − k1 z −1 )Y ( z ). Substituting this
equation and the second equation in the third equation we get after some
rearrangement [1 − (k1 + 1)k 2 z −1 + k1 z −2 ]Y ( z ) = [( γ 2 − k 2 γ1 ) z −1 + γ1 z −2 ] X ( z ).
( γ 2 − k 2 γ 1 ) z −1 + γ 1 z −2
Y ( z)
Hence, the transfer function is given by H ( z ) =
=
.
X ( z ) 1 − (k1 + 1) k 2 z −1 + k1 z − 2
For stability, we must have k1 < 1 and (1 + k1 )k 2 < 1 + k1 , or k 2 < 1.
8.7

The structure with internal variables is shown on next page. Analysis of this structure
yields
S 0 ( z ) = X ( z ) − S1 ( z ), S1 ( z ) = z −1 S 2 ( z ) + z −1 S 5 ( z ), S 2 ( z ) = a3 S 0 ( z ) − z −1 S 2 ( z ),
S 3 ( z ) = a3 S 5 ( z ) − z −1 S 3 ( z ), S 4 ( z ) = S 0 ( z ) + z −1 S 3 ( z ), S 5 ( z ) = a1 S 4 ( z ) − z −1 S 5 ( z ),
and Y ( z ) = a 4 S 4 ( z ). Eliminating S 0 ( z ), S1 ( z ), S 2 ( z ), S 3 ( z ), S 4 ( z ) and S 5 (z )
from these equations we get after some algebra

Not for sale

242

S 0(z)

X(z)

S 4(z)

a4

a3

Y(z)

a1
_

1
S2 (z) _ 1 z
_1 _
S 3(z)
z1

_1

S 5(z)

_1

_1

z

a2
S 1(z)

a 4 (1 + z −1 ) 3
Y ( z)
H ( z) =
=
.
X ( z ) 1 + (3 + a1 + a3 ) z −1 + (3 + 2 a1 + 2a3 − a1a2 ) z − 2
+ (1 + a1 + a3 − a1a2 − a1a2 a3 ) z − 3
8.8

The structure with internal variables is shown below. Analysis of this structure yields
_1

X(z)

W 1(z)

W 3(z)
_1

z

R1 (z)
α1

β1

α0

_

z1

β2

α2

β3

R 2(z)

α3
_

z1

W2 (z)

Y(z)

_1

Y ( z ) = α 0 X ( z ) + β1 z −1 R1 ( z ), W1 ( z ) = X ( z ) − β 2 z −1 R2 ( z ),

R1 ( z ) = W1 ( z ) + α1β1 z −1 R1 ( z ), W2 ( z ) = β1 z −1 R1 ( z ) − β 3 z −1W3 ( z ),
R2 ( z ) = W2 ( z ) + α 2 β 2 z −1 R2 ( z ), W3 ( z ) = β 2 z −1 R2 ( z ) + α 3β 3 z −1W3 ( z ).

From the third equation we get W1 ( z ) = (1 − α1β1 z −1 ) R1 ( z ). From the sixth equation
we get W3 ( z ) =

β 2 z −1 R2 ( z )
α 3β 3 z

−1

. From the fifth equation we get R2 ( z ) =

W2 ( z )

1 − α 2 β 2 z −1

.

Rewriting the fourth equation we get W2 ( z ) + β 3 z −1W3 ( z ) = β1 z −1 R1 ( z ), in which we
substitute the expressions for W3 (z ) and W1 (z ) resulting in after some algebra
R2 ( z ) =

β1 z −1 (1 − α 3β 3 z −1 )

β 2 β 3 z −1 + (1 − α 2 β 2 z −1 )(1 − α 3β 3 z −1 )

R1 ( z ). Substituting this expression

for R2 (z ) and the expression for W1 (z ) in the second equation we arrive at

Not for sale

243

R1 ( z ) =

β 2 β 3 z −2 + (1 − α 2 β 2 z −1 )(1 − α 3β 3 z −1 )

β 2 β 3 z − 2 (1 − α1β1 z −1 ) + (1 − α1β1 z −1 )(1 − α 2 β 2 z −1 )(1 − α 3β 3 z −1 )

X ( z ).

β1β 2 z − 2 (1 − α 3β 3 z −1 )
Substituting the above in the first equation we finally arrive at
H ( z) =

β1β 2 β 3 z −3 + β1 z −1 (1 − α 2 β 2 z −1 )(1 − α 3β 3 z −1 )
Y ( z)
= α0 +
.
X ( z)
(1 − α1β1 z −1 )(1 − α 2 β 2 z −1 )(1 − α 3β 3 z −1 ) + β 2 β 3 z − 2 (1 − α1β1 z −1 )
β1β 2 z − 2 (1 − α 3β 3 z −1 )

8.9 The structure with internal variables is shown below. Analysis of this structure yields
U(z) _ k1

W(z)
X(z)
k1

(3): Y ( z ) =

Y(z)

k2

1
z–

(1): W ( z) = X ( z) + k1Y ( z), (2): U ( z ) =

1
z–

1
1 − z −1

W ( z ) + k 2Y ( z ), and

− k1

U ( z ). Substituting Eq. (2) in Eq. (3) we get
1 − z −1
− k1
k k
− k1 ⎛ 1

(4): Y ( z ) =
⎜⎜
W ( z ) − 1 2 Y ( z ).
W ( z ) + k 2Y ( z ) ⎟⎟ =
1
1

1
2
1− z ⎝1− z
1 − z −1
⎠ (1 − z )
Substituting Eq. (1) in Eq. (4) we then get
k1
k k
Y (z) = −
[ X ( z ) + k1Y ( z )] − 1 2 Y ( z )
(1 − z −1 ) 2
1 − z −1
=−

k1

(1 − z −1 ) 2

X ( z) −

⎡ k1 + k 2 − k 2 z −1 ⎤

⎥Y ( z ), or,
1 − z −1 ⎣⎢
1 − z −1
⎦⎥

k1

⎡ k1 (k1 + k 2 − k 2 z −1 ) ⎤
k1
X ( z ). Hence,
⎢1 +
⎥Y ( z ) = −

1
2
(1 − z )
(1 − z −1 ) 2
⎢⎣
⎥⎦
k1
Y (z)
.
H (z) =
=−
X ( z)
[1 + k1 (k1 + k 2 )] − (2 + k1k 2 )z −1 + z − 2

8.10

(a) A direct form realization of H ( z ) = 1 − 3.6 z −1 + 5.4 z −2 − 4.32 z −3 + 1.944 z −4

− 0.4666z −5 + 0.0467z −6 and its transformed structure are shown on next page:

Not for sale

244

_
z 1

_
z 1

x[n]

_ 3.6

_
z 1

+
z

_

5.4

+

–1

z
_

0.0467

z

0.4666

–1

z
_

1.944

0.0467

–1

z

–1

z

–1

y[n]

_ 3.6

5.4

4.32

y[n]

+

+

+

+

–1

_ 0.4666

1.944

4.32

_
z 1

_
z 1

_
z 1

x[n]

H ( z ) = (1 − 0.6 z −1 )(1 − 0.6 z −1 )(1 − 0.6 z −1 )(1 − 0.6 z −1 )(1 − 0.6 z −1 )(1 − 0.6 z −1 ) is
shown below:
x[n]

+
_1

+

+

_1

z

_1

z
_

_

0.6

z

_1

_

0.6

z
_

0.6

y[n]

_1

z

z
_

0.6

+

+

+
_1

_

0.6

0.6

H ( z ) = (1 − 1.2 z −1 + 0.36 z −2 )(1 − 1.2 z −1 + 0.36 z −2 )(1 − 1.2 z −1 + 0.36 z −2 ) is shown
below:
x[n]

+

+

+
_1

z

z

z
_ 0.2

_ 0.2

+

_ 0.2

+
z

z

z

0.36

0.36

+

_1

_1

_1

y[n]

_1

_1

0.36

H ( z ) = (1 − 1.8 z −1 + 1.08 z −2 − 0.216 z −3 )(1 − 1.8 z −1 + 1.08 z −2 − 0.216 z −3 ) is shown
below:
y[n]
x[n]
+
+
_
z 1

_

_
z 1

1.8

_

+

_1

+

_1

z

z
1.08

1.08

+
_
z 1

1.8

_ 0.216

Not for sale

+
_
z 1

_ 0.216

245

H ( z ) = (1 − 0.6 z −1 )(1 − 0.6 z −1 )(1 − 1.2 z −1 + 0.36 z −2 )(1 − 1.2 z −1 + 0.36 z −2 ) is shown
below:
x[n]

+
_1

z

_ 0.6

+

+

+

z

z

z

_ 0.2

_ 0.6

y[n]

_1

_1

_1

_ 0.2

+

+

_1

_1

z

z

0.36

0.36

All realizations given above require 6 multipliers and 6 two-input adders.
8.11 (a) A four branch polyphase decomposition is given by
H ( z) = E 0 ( z 4 ) + E1 ( z 4 )z −1 + E 2 ( z 4 )z −2 + E 3 ( z 4 )z −3 , where
E 0 ( z ) = h[0] + h[ 4]z −1 + h[8] z −2 , E1 ( z ) = h[1] + h[5] z −1 + h[9] z −2 ,
E 2 ( z ) = h[ 2] + h[6]z −1, E 3 ( z) = h[3] + h[7]z −1.

(b) A canonic four-branch polyphase realization of the above transfer function is
shown below:
h[3]
h[7]

+
_
z 1

h[2]
h[6]

x[n]

_
z 4

_
z 4

+

+
_
z 1

h[9]
h[5]

+

h[1]

_
z 1

h[8]
h[4]
h[0]

+

+

+

y[n]

8.12 (a) A three branch polyphase decomposition is given by
H ( z) = E 0 ( z 3 ) + E1 ( z 3 )z −1 + E 2 ( z 3 )z −2 , where

E 0 ( z) = h[0] + h[3]z −1 + h[6]z −2 + h[9]z −3 , E1 ( z ) = h[1] + h[ 4]z −1 + h[7]z −2 ,
E 2 ( z ) = h[ 2] + h[ 5]z −1 + h[8]z −2 .

Not for sale

246

(b) A canonic three branch polyphase realization of the above transfer function is
shown below:
h[2]
h[5]

+

h[8]
_
z 1
_3

_3

_3

z

z

z

h[7]
h[4]

+

+

h[1]
_
z 1

h[9]

h[6]

+

h[3]

+

+

h[0]

8.13 (a) A two branch polyphase decomposition is given by H ( z) = E 0 ( z 2 ) + E1 ( z 2 )z −1 ,

where E 0 ( z) = h[0] + h[2]z −1 + h[ 4]z −2 + h[6]z −3 + h[8]z −4 ,
E1 ( z ) = h[1] + h[3]z −1 + h[ 5]z −2 + h[ 7]z −3 + h[9]z −4 .

(b) A canonic two branch polyphase realization of the above transfer function is shown
below:
+
+
+
+
_
z 2

_
z 2

_
z 2

h[5]

h[3]

h[9]

h[7]

+

h[1]

h[8]

+

_
z 2

h[6]

+

x[n]

h[4]

h[2]

_
z 1

+

h[0]

y[n]

+
8.14

_
z 1

x[n]

_
z 1

_
z 1

+

+

+
_1

_1

_
z 1

_1

_
z 1

_
z 1

h[1]

h[0]

+

h[2]

+

y[n]

8.15

Not for sale

247

_
z 1

_
z 1

_
z 1

+

+

_
z 1

_
z 1

h[1]

h[0]

_1

_1

_
z 1

_
z 1

+

+

_1

_1

h[3]

h[2]

+

+

+

8.16 G ( z ) = z − N / 2 − H ( z ). A canonic realization of both G ( z ) and H ( z ) is shown below
for N = 6.
_
_
_
+
G(z)
z 1
z 1
z 1

+

+

+
_
z 1

_
z 1

h[1]

h[0]

_
z 1

h[2]

h[3]

+

+

_1

+

H(z)

8.17 Without any loss of generality, assume M = 5 which means N = 11. In this case, the
transfer function is given by H ( z )

[

]

= z − 5 h[ 5] + h[ 4]( z + z −1 ) + h[3]( z 2 + z − 2 ) + h[ 2]( z 3 + z − 3 ) + h[1]( z 4 + z − 4 ) + h[ 0]( z 5 + z − 5 ) .

Now, the recursion relation for the Chebyshev polynomial is given by
Tr ( x ) = 2 x Tr −1 ( x ) − Tr − 2 ( x ), r ≥ 2 with T0 ( x ) = 1 and T1 ( x ) = x. Hence,

T2 ( x ) = 2 x T1 ( x ) − T0 ( x ) = 2 x 2 − 1,
T3 ( x) = 2 x T2 ( x ) − T1 ( x ) = 2 x(2 x 2 − 1) − x = 4 x 3 − 3 x,
T4 ( x ) = 2 x T3 ( x ) − T2 ( x ) = 2 x(4 x 3 − 3 x) − (2 x 2 − 1) = 8 x 4 − 8 x 2 + 1,
T5 ( x ) = 2 x T4 ( x ) − T3 ( x ) = 2 x(8 x 4 − 8 x 2 + 1) − (4 x 3 − 3 x) = 16 x 5 − 20 x 3 + 5 x.
We can thus rewrite the expression inside the square brackets given above as
⎛ z + z −1 ⎞
⎛ z + z −1 ⎞
⎛ z + z −1 ⎞
⎛ z + z −1 ⎞
⎛ z + z −1 ⎞
h[ 5] + 2h[ 4]T1 ⎜

⎟ + 2h[ 0]T5 ⎜
⎟ + 2h[1]T4 ⎜
⎟ + 2h[2]T3 ⎜
⎟ + 2h[3]T2 ⎜
⎝ 2 ⎠
⎝ 2 ⎠
⎝ 2 ⎠
⎝ 2 ⎠
⎝ 2 ⎠
⎡ ⎛ z + z −1 ⎞ 3
⎡ ⎛ z + z −1 ⎞ 2 ⎤
⎛ z + z −1 ⎞ ⎤
⎛ z + z −1 ⎞
= h[5] + 2h[ 4]⎜
⎟⎥
⎟ − 3⎜
⎟ − 1⎥ + 2h[2]⎢ 4⎜
⎟ + 2h[3]⎢2⎜
⎝ 2 ⎠ ⎥⎦
⎝ 2 ⎠
⎢⎣ ⎝ 2 ⎠
⎥⎦
⎢⎣ ⎝ 2 ⎠
2
3

⎡ ⎛ z + z −1 ⎞ 4
⎡ ⎛ z + z −1 ⎞ 5
⎛ z + z −1 ⎞
⎛ z + z −1 ⎞
⎛ z + z −1 ⎞ ⎤

+ 2h[1]⎢8⎜

+
+

8
1
2
h
[
0
]
16
20

⎟ + 5⎜
⎟⎥
⎝ 2 ⎠
⎝ 2 ⎠
⎝ 2 ⎠ ⎥⎦
⎥⎦
⎢⎣ ⎝ 2 ⎠
⎢⎣ ⎝ 2 ⎠
5

n

⎛ z + z −1 ⎞
=
a[ n]⎜
⎟ , where a[ 0] = h[ 5] − 2h[3] + 2h[1], a[1] = 2h[ 4] − 6h[ 2] + 10 h[ 0],
2 ⎠

n=0

a[ 2] = 4h[3] − 16 h[1], a[3] = 8h[ 2] − 40 h[1], a[ 4] = 16 h[1], and a[5] = 32h[0]. A
realization of

Not for sale

248

H ( z) =

z−5

2

n

5

⎛ 1+ z −2 ⎞ − 3
⎛ z + z −1 ⎞
⎛ 1+ z − 2 ⎞ − 4
a[ n]⎜
⎟ z
⎟ = a[0]z − 5 + a[1]⎜
⎟ z + a[2]⎜
⎝ 2 ⎠
⎝ 2 ⎠
⎝ 2 ⎠
n=0

5

4

3

⎛ 1+ z − 2 ⎞
⎛ 1 + z − 2 ⎞ −1
⎛ 1+ z − 2 ⎞ − 2
+ a[3]⎜
⎟ is shown below:
⎟ z + a[5]⎜
⎟ z + a[ 4]⎜
⎝ 2 ⎠
⎝ 2 ⎠
⎝ 2 ⎠
_

_

1+z 2
______
2

x[n]

_
z 1

+

a[2]
_
z 1

1+z 2
______
2

+

a[5]

a[4]

a[3]
_
z 1

+

_

1+z 2
______
2

1+z 2
______
2

a[1]

a[0]

_

_

1+z 2
______
2

_
z 1

_
z 1

+

+

y[n]

P( z ) P1( z) P 2 ( z) P 3 ( z)
. Assume all zeros of P ( z ) and D ( z )
=

D( z ) D1 ( z) D2 ( z) D3 ( z)
are complex. Note that the numerator of the first stage can be one of the 3 factors,
P1 ( z), P2 ( z), and P3 ( z ). Likewise, the numerator of the second stage can be one of the
remaining 2 factors, and the numerator of the third stage is the remaining factor.
Similarly, that the denominator of the first stage can be one of the 3 factors,
D1 ( z), D2 ( z), and D3 ( z ). Likewise, the denominator of the second stage can be one
of the remaining 2 factors, and the denominator of the third stage is the remaining
factor. Hence, there are (3! ) 2 = 36 different types of cascade realizations.

8.18 Consider H ( z) =

If the zeros of P ( z ) and D ( z ) are all real, then P ( z ) has 6 real zeros and D ( z ) has 6
real zeros. In this case, then there are (6! ) 2 = 518400 different types of cascade
realizations.
8.19 H ( z ) =

K

∏ Dii (z).
i =1

P (z)

⎛K⎞
Here, the numerator of the first stage can be chosen in ⎜ ⎟ ways,
⎝1⎠

⎛ K −1 ⎞
the numerator of the second stage can be chosen in ⎜
⎟ ways, and so on, until there
⎝ 1 ⎠
is only one possible choice for the numerator of the K –th stage. Likewise, the
K
denominator of the first stage can be chosen in ⎛⎜ ⎞⎟ ways, the denominator of the
⎝1⎠
⎛ K − 1⎞
second stage can be chosen in ⎜
⎟ ways, and so on, until there is only one possible
⎝ 1 ⎠
choice for the denominator of the K –th stage. Hence, the total number of possible
2

2

2

2

2

K
K −1⎞ ⎛ K − 2 ⎞
⎛ 2 ⎞ ⎛ 1⎞
cascade realizations are equal to ⎛⎜ ⎞⎟ ⎛⎜
⎟ ⎜
⎟ L ⎜ ⎟ ⎜ ⎟ = ( K! ) 2 .
1
1
1
⎝ 1 ⎠ ⎝ 1⎠
⎝ ⎠ ⎝
⎠ ⎝

Not for sale

249

8.20 A canonic direct form II realization of H ( z ) =

3 − 5 z −2 + 3 z −3

1 − 4 z −1 + 3 z − 2 + 6 z − 4
on the left. Its transposed realization is shown below on the right.
x[n]

2

+

+

y[n]

2

x[n]

4

+

_
z 1

_5

3

_5

+

_3

+
_
z 1

_
z 1

3

3

_

4

+
_
z 1

_

y[n]

+
_
z 1

_
z 1

+

is shown below

+
_
z 1

_
z 1

_6

6

8.21 (a) A cascade canonic realization of H1( z) based on the decomposition

z −1 + 3 z − 2
H 1 ( z ) = 0.3⎜
⎜ 1 − 1.6 z −1 + 2.1z − 2

x[n]

⎞⎛ 1 − 0.8 z −1 ⎞
⎟ is shown below:
⎟⎜
⎟⎜ 1 − 0.75 z −1 ⎟

⎠⎝

0.3
z

–1

–1

z
_

0.75

y[n]
0.8

1.6

–1

z

_ 2.1

3

An alternate cascade canonic realization of H1( z) based on the decomposition
⎛ z −1 − 0.8 z − 2
H 1 ( z ) = 0.3⎜
⎜ 1 − 1.6 z −1 + 2.1z − 2

x[n]

⎞⎛ 1 + 3 z −1 ⎞
⎟ is shown below:
⎟⎜
⎟⎜ 1 − 0.75 z −1 ⎟

⎠⎝

0.3
z–1

–1

z

y[n]
0.75

3

1.6
_ 2.1

Not for sale

–1

z

_

0.8

250

(b) A cascade canonic realization of H 2 ( z) based on the decomposition

⎛ 3.1 + 0.85z −1 ⎞⎛ 1 − 4.5z −1 ⎞⎛
3 − 0 .5 z − 2
⎟ is shown below:
⎟⎜
⎟⎜
H 2 ( z ) = ⎜⎜
−1 ⎟⎜ 1 + 0.2 z −1 ⎟⎜ 1 + 0.5z −1 + 0.1z − 2 ⎟
1
0
.
15
z

⎠⎝
⎠⎝

3.1

3

x[n]
z–1

z–1

_

_ 0.2

0.85

0.15

–1

z

y[n]
_ 0.5

4.5

z

–1
_ 0.5

_ 0.1

An alternate cascade canonic realization of H 2 ( z) based on the decomposition
⎛ 1 − 4.5z −1 ⎞⎛ 3.1 + 0.85z −1 ⎞⎛

3 − 0 .5 z − 2
⎟⎜
⎟⎜
⎟ is shown below:
H 2 ( z ) = ⎜⎜

1

1

1

2
⎝ 1 − 0.15z ⎠⎝ 1 + 0.2 z
⎠⎝ 1 + 0.5z + 0.1z ⎠
3.1

3

x[n]
–1

z–1

z

_

0.15

4.5

–1

z

y[n]

_ 0.2

_ 0.5

0.85
_

–1

z

_ 0.5

0.1

(c) The factored form of the denominator of H 3 ( z ) is given by:
(1 + 3.1z −1 )(1 − 2.0246 z −1 )(1 − 0.3754 z −1 ). A cascade canonic realization of
H 3 ( z ) based on the decomposition
⎛ 1 .5 + 0 .7 z −1 ⎞ ⎛

5.2 − 8.4 z −1 + 7 z − 2
⎟⎜

H 3 ( z ) = ⎜⎜
−1 ⎟⎜ (1 − 2.0246 z −1 )(1 − 0.3754 z −1 ) ⎟
+
1
3
.
1
z

⎠⎝

1

1

2

⎛ 1.5 + 0.7 z ⎞⎛ 5.2 − 8.4 z + 7 z
⎟ is shown below:
⎟⎜
=⎜
⎜ 1 + 3.1z −1 ⎟⎜ 1 − 2.4 z −1 + 0.76 z − 2 ⎟

⎠⎝

1.5

5.2

x[n]

z
_ 3.1

–1

2.4
0.7

z

–1

z
_ 0.76

_ 8.4

y[n]

–1
7.0

An alternate cascade canonic realization of H 3 ( z ) based on the decomposition
⎛ 1 .5 + 0 . 7 z − 1 ⎞ ⎛

5 .2 − 8 .4 z − 1 + 7 z − 2
⎟⎜

H 3 ( z ) = ⎜⎜
−1 ⎟⎜ (1 + 3.1z −1 )(1 − 0.3754 z −1 ) ⎟
1
2
.
0246
z

⎠⎝

Not for sale

251

⎛ 1.5 + 0.7 z −1 ⎞⎛

5 .2 − 8 .4 z − 1 + 7 z − 2
⎟⎜
⎟ is shown below:
= ⎜⎜
− 1 ⎟⎜ 1 + 2.7246 z −1 − 1.1637 z − 1 ⎟

1
2
.
0246
z

⎠⎝

1.5

5.2

x[n]

z

–1

_ 2.7246

2.024

0.7

z–1

z

_ 8.4

y[n]

–1
7.0

1.1637

8.22 (a) A partial fraction expansion of H1( z) in z −1 obtained using the M-file residuez

is given by H1 ( z ) = 0.4571 +

− 0.0513

+

− 0.4059 + 0.9878z −1

. The Parallel Form I
1 − 0.75z −1
1 − 1.6 z −1 + 2.1z − 2
realization based on this expansion is shown on the next page on the left side.
0.4571

x[n]

_ 0.0513

+

+

y[n]

x[n]

+

_
z 1

_
z 1

0.75
_

0.0385

+

y[n]

0.4059

+

+

+

_
z 1

+

_

0.75

_
z 1

1.6

+

0.9878

0.3385

1.6

_
z 1

+
_
z 1

_ 2.1

_ 2.1

_ 0.8523

A partial fraction expansion of H1( z) in z obtained using the M-file residue is
given by H1 ( z ) =

− 0.0385z −1

+

0.3385z −1 − 0.8523z −2

. The Parallel Form II
1 − 0.75z −1
1 − 1.6 z −1 + 2.1z − 2
realization based on this expansion is shown above on the right side.
(b) A partial fraction expansion of H 2 ( z) in z −1 obtained using the M-file residuez
146.71
238.6
261.4905 + 79.8813z −1
+
+
. The
is given by H 2 ( z ) = −637.5 +
1 + 0.2 z −1 1 − 0.15z −1
1 + 0.5z −1 + 0.1z − 2
Parallel Form I realization based on this expansion is shown on the next page on the left
side.

Not for sale

252

146.71

+

+
_
z 1

_
z 1

_ 0.2

x[n]

0.15
238.6

+

+

y[n]

x[n]

+

_1

_
z 1

z
0.15

_

637.5

+

+
_

+

0.5

+

y[n]

9.3
_
z 1

261.4905

_ 50.8639

_ 0.5

79.8813

+

+

_
z 1

_1

z
_

_ 29.3414

_ 0.2

+
_
z 1

35.7904

0.1

_ 26.1491

_ 0.1

A partial fraction expansion of H 2 ( z) in z obtained using residue is given by
− 29.3414 z −1 35.7904 z −1 − 50.8639z −1 − 26.1491z −2
+
+
. The
H 2 ( z ) = 9.3 +
1 + 0.2 z −1
1 − 0.15z −1
1 + 0.5z −1 + 0.1z − 2
Parallel Form II realization based on this expansion is shown above on the right side.
(c) A partial fraction expansion of H 3 ( z ) in z −1 obtained using residuez is given
5.939
2.4696
− 2.6884
+
+
. The Parallel Form I
by H 3 ( z ) = 2.0798 +
1 + 3.1z −1 1 − 0.0246 z −1 1 − 0.3754 z −1
realization based on this expansion is shown below on the left side.
2.0798

7.8

5.939

+

+

_
z 1

_ 3.1

_ 3.1

x[n]

_
z 1

2.4696

+

+

y[n]

x[n]

_ 18.4109

+

_1

z

+
_
z 1

0.0246

0.0246

y[n]

5.0

_ 2.6884

+

+
_1

_
z 1

z

0.3754

_

1.0092

0.3754

Not for sale

253

A partial fraction expansion of H 3 ( z ) in z obtained using residue is given by
− 18.4109z −1
5.0 z −1
− 1.0092 z −1
+
+
H 3 ( z ) = 7.8 +
. The Parallel Form II
1 + 3.1z −1
1 − 0.0246 z −1 1 − 0.3754 z −1
realization based on this expansion is shown on the previous page on the right side.
8.23 A cascade realization of based on the factored form given by

⎞⎛ 0.44 + 0.362 z −1 + 0.02 z − 2 ⎞
z −1
⎟⎜
⎟ is shown below:
H ( z ) = ⎜⎜

− 1 + 0 .5 z − 2 ⎟ ⎜
−1
+

1
0
.
8
z
1
0
.
4
z

⎠⎝

x[n]

+

0.44

+

_

+

_
z 1

8.24 (a) H ( z ) =

y[n]

_
z 1

0.8

0.362

0.4
_
z 1

_

+

+

_
z 1

0.5

0.02

2 − 0.3z −1 0.4 + z −1
3
2.4 + 5.64 z −1 − 0.9z −2

=
.
1 + 0.5z −1 1 + 0.4 z −1 1 + 0.5z −1 1 + 1.4 z −1 + 0.65z − 2 + 0.1z − 3

(b) y[ n] = 2.4 x[ n] + 5.64 x[ n − 1] − 0.9 x[ n − 2] − 1.4 y[ n − 1] − 0.65 y[ n − 2] − 0.1y[ n − 3].
(c) A cascade realization of H (z ) is shown below:
x[n]

2

+

+

0.4

+

_
z 1
_ 0.5

+

3

+
_
z 1

_
z 1
_

_

0.3

0.4

y[n]

_ 0.5

(d) A partial-fraction expansion of H ( z ) in z −1 obtained using residuez is given by
− 62.4
− 277.2
342
+
+
H ( z) =
. The Parallel Form I realization

1

1
2
1 + 0.5z
(1 + 0.5z )
1 + 0.4 z −1
based on this expansion is shown on the next page.
(e) The inverse z –transform of the partial-fraction of H ( z ) given in Part (d) yields
h[ n] = 342(−0.5) n µ[ n] − 62.4( n + 1)( −0.5) n µ[ n] − 277.2( −0.4) n µ[ n].

Not for sale

254

_

+

277.2

_
z 1
_ 0.4

342

+

x[n]

+

y[n]

, X ( z) =

1

_
z 1
_ 0.5
_

+

62.4

_
z 1

+

_1

_ 0.25

8.25 (a) Y ( z ) =

1
1 − 0.4 z −1

0.3z −1
1 − 0.4 z −1

=

_
z 1

1 − 0.3z −1
1 − 0.4 z −1

1 − 0.2 z −1
Y ( z ) (1 − 0.3z −1 )(1 − 0.2 z −1 ) 1 − 0.5z −1 + 0.06 z −2
=
=
H (z) =
.
X (z)
1 − 0.4 z −1
1 − 0.4 z −1

. Thus,

(b) y[ n] = x[ n] − 0.5 x[ n − 1] + 0.06 x[ n − 2] + 0.4 y[ n − 1].

x[n]

+

_ 0.5

_
z 1

+

+

(c)
y[n]

0.06
_
z 1

0.4

A partial-fraction expansion of H ( z ) in z −1 using residuez is given by
0.125
whose realization yields the Parallel Form I
H ( z ) = 0.875z −1 − 0.15 +
1 − 0.4 z −1
structure as indicated on the next page.
(d)

(e) The inverse z -transform of H ( z ) yields
h[ n] = 0.875δ[ n − 1] − 0.15δ[ n] + 0.125(0.4) n µ[ n].

Not for sale

255

_1

+

x[n]

z
0.4

_
z 1

(f) X ( z ) =

1

0.4 z −1

=

_

0.15

+

0.125
y[n]

0.875

1 − 0.4 z −1

. Therefore, Y ( z ) = H ( z ) X ( z )
1 − 0.3z −1 1 − 0.3z −1
⎛ (1 − 0.3z −1 )(1 − 0.2 z −1 ) ⎞⎛ 1 − 0.4 z −1 ⎞
⎟⎜
⎟ = 1 − 0.2 z −1 , whose inverse z -transform
= ⎜⎜

1

1
1 − 0 .4 z

⎠ ⎝ 1 − 0 .3 z ⎠
yields y[ n] = δ[ n] − 0.2δ[ n − 1].
1 − 0.3z −1

8.26 Figure P8.12 can be seen to be a Parallel Form II structure. A partial-fraction
2 z 2 − 3.2 z − 18.6
in z obtained using the M-file residue is
expansion of H ( z ) =
z 2 − 2.6 z − 1.2
5 z −1
− 3z −1
. Comparing the coefficients of the
given by H ( z ) = 2 +
+
1 − 3z −1 1 + 0.4 z −1
expansion with the corresponding multiplier coefficients in Figure P8.12 we conclude
that the multiplier coefficient − 5 in the feed forward path should be 5 and the
multiplier coefficient − 0.5 in the feedback path should be replaced with − 0.4.
8.27 Figure P8.13 can be seen to be a Parallel Form I structure. A partial-fraction expansion
4 z 2 − 5.6 z
in z −1 obtained using the M-file residuez is given by
of H ( z ) =
2
z + 0.2 z − 0.08
⎛ − 1.5

12
−8
1
⎟⎟. Comparing the
+
= −8⎜⎜
H (z) =
+
1 + 0.4 z −1 1 − 0.2 z −1
⎝ 1 + 0.4 z −1 1 − 0.2 z −1 ⎠
coefficients of the expansion with the corresponding multiplier coefficients in Figure
P8.12 we conclude that the multiplier coefficient A has a value equal to − 8 and the
multiplier coefficient B has a value equal to − 0.4.
8.28 (a) The difference equation corresponding to the transfer function H ( z ) =

=

Y ( z)
X ( z)

(1 + α1 + α 2 )(1 + 2 z −1 + z −2 )

is given by
1 − α1z −1 + α2 z − 2
y[ n] + α1 y[ n − 1] − α2 y[ n − 2] = (1 + α1 + α2 )( x[ n] + 2 x[ n − 1] + x[ n − 2]), which can be
rewritten as y[ n] = ( x[ n] + 2 x[ n − 1] + x[ n − 2]) + α1 ( y[ n − 1] − x[ n] − 2 x[ n − 1] − x[ n − 2])
− α2 ( y[ n − 2] − x[ n] − 2 x[ n − 1] − x[ n − 2]). Denoting w[ n] = x[ n] + 2 x[ n − 1] + x[ n − 2],
the difference equation representation becomes

Not for sale

256

y[ n] = w[ n] + α1 ( y[ n − 1] − w[ n]) − α2 ( y[ n − 2] − w[ n]). A realization of H (z ) based on the
last two equations is as indicated below:
w[n]

_
z 1

α1

_

1

2

+

+

_1

z

y[n]

+

+

+

x[n]

_1

z

_

_
z 1

2

1

+
An interchange of the two stages leads to an equivalent realization shown below:
x[n]

+

+

+

α1

_

1

+

_1

z

_1

_1

z

_1

2

2

y[n]

+

_1

z

z

+
Finally, by delay sharing the above structure reduces to the canonic realization shown
below:
x[n]
y[n]
+
+
+
α1

_

1

+
_

_
z 1

2

+

_1

2

z

1

+
(b) The difference equation corresponding to the transfer function H ( z ) =

=

Y ( z)
X ( z)

(1 − α2 )(1 − z −2 )

is given by
1 − α1z −1 + α2 z − 2
y[ n] − α1 y[ n − 1] + α2 y[ n − 2] = (1 − α2 )( x[ n] − x[ n − 2]) which can be rewritten as
y[ n] = α1 y[ n − 1] − α2 y[ n − 2] + x[ n] − α2 x[ n] − x[ n − 2] + α2 x[ n − 2]
= α1 y[ n − 1] − α2 ( x[ n] − x[ n − 2] + y[ n − 2]) + ( x[ n] − x[ n − 2]). Denoting
w[ n] = x[ n] − x[ n − 2], we can rewrite the last equation as
y[ n] = α1 y[ n − 1] − α2 (w[ n] + y[ n − 2]) + w[ n].
A realization of H (z ) based on the last two equations is as shown on next page.

Not for sale

257

+

x[n]
_
z 1

w[n]

_1

+

+
2

y[n]
α1 _ 1
z

+

_
z 1

_
z 1

_

1

An interchange of the two stages leads to an equivalent realization shown below:
x[n]

+

+
2

+

α1 _ 1
z

z

_1

+
_
z 1

_

1

y[n]

_1

_
z 1

Finally, by delay sharing the above structure reduces to a canonic realization as shown
below:
x[n]

+

+
2

+

α1 _ 1
z

y[n]

_1

+
_
z 1

_1

8.29 (a)

+

+

X1
α1

α2

+
α3

_
z 1

Y1

α4

_
z 1

(b)
+

+

X1
α4

Y1
α2

_
z 1

α1

+

α3

_
z 1

Not for sale

258

C + D ⋅ H N −1 ( z )
Y
,
8.30 (a) From the structure of Figure P8.14 it follows that H N ( z ) = 1 =
X1 A + B ⋅ H N − 1 ( z )
C − A ⋅ H N ( z)
. Substituting the expression for H N ( z ) we
from which we get H N −1 ( z ) =
B ⋅ H N (z) − D
N
N
C ⎛⎜ 1 +
di z − i ⎞⎟ − A⎛⎜
pi ⎞⎟
=
1
=
0
i
i

then arrive at H N −1 ( z ) = ⎝
N
N
B⎛⎜
pi ⎞⎟ − D⎛⎜ 1 +
d z − i ⎞⎟
i =1 i

⎝ i=0 ⎠

=

(C − Ap 0 ) + (Cd1 − Ap1 )z −1 + K + (Cd N −1 − Ap N −1 )z − N + 1 + (Cd N − Ap N )z − N
( Bp 0 − D ) + ( Bp1 − Dd1 )z −1 + L + ( Bp N −1 − Dd N −1 )z − N + 1 + ( Bp N − Dd N )z − N

.

Substituting the values A = 1, B = d N z −1 , C = p0 , and D = p N z −1 , we get H N −1 ( z )
=

( p 0 − p 0 ) + ( p 0 d 1 − p1 ) z − 1 + L + ( p 0 d N − 1 − p N − 1 ) z − N + 1 + ( p 0 d N − p N ) z − N
(d N p0 − p N )z −1 + (d N p1 − p N d1 )z − 2 + L + (d N p N −1 − p N d N −1 )z − N
+ (d N p N − p N d N )z − N −1

=

( p 0 d 1 − p1 ) + L + ( p 0 d N − 1 − p N − 1 ) z − N + 2 + ( p 0 d N − p N ) z − N + 1

(d N p0 − p N ) + (d N p1 − p N d1 )z −1 + L + (d N p N −1 − p N d N −1 )z − N + 1

p ' + p1' z −1 + L + p 'N − 2 z − N + 2 + p 'N −1 z − N + 1
where
= 0
' −2z− N +2 + dN
' −1 z − N + 1
1 + d1' z −1 + L + d N

p k' =

p k d k +1 − pk +1
p d − pN dk
, 0 ≤ k ≤ N − 1, and d k' = k N
, 1 ≤ k ≤ N − 1.
p0 d N − p N
p0 d N − p N

(b) From the chain parameters, we obtain for the first stage t 11=

C
= p0 ,
A

1
B
= ( p N − p0 d N )z −1 , t 21= = 1, t 22 = − = − d N z −1 . The
A
A
A
corresponding input-output relations of the two-pair are given by
Y1 = p 0 X1 + ( p N − p 0 d N )z −1 X 2 = p 0 ( X1 − d N z −1 X 2 ) + p N z −1 X 2 ,
t12 =

Y2 = X1 − d N z −1 X 2 . Substituting the second equation into the first we rewrite it as
Y1 = p0Y2 + p N z −1 X 2 . A realization of the two-pair based on the last two equations is
therefore as indicated below:
X1

Y2
_d

N

p

0

Y1

p

z–1

X2

N

Not for sale

259

(c) Except for the first stage, all other stages require 2 multipliers. Hence, the total number
of multipliers needed to implement an N –th order transfer function H N (z) is 2 N + 1.
The total number of 2-input adders required is 2N , while the overall realization is canonic
requiring N delays.
8.31 (a) From G3 ( z) = H1 ( z) =

0.3z −1 + 0.66 z −2 − 0.72 z −3

, using Eq. (8.132a) and
1 − 2.35z −1 + 3.3z − 2 − 1.575z − 3
− 0.4167 − 0.9167z −1 + z −2
. Repeating the procedure we get
(8.132b) we get G2 ( z ) =
1 − 3.0063z −1 + 1.8562 z − 2
− 1.2232 + z −1
. From G 3 ( z), G 2 ( z ), and G1( z), we arrive at the cascaded lattice
G1( z) =
1 − 0.7357z −1
realization shown below:
_1.8562

1.575

_
_

_1.2232

0.4167

z–1

0.72

(b) From G 4 ( z ) = H 2 ( z ) =

0.7357

z

–1

z

–1

9.3 − 39.3z −1 − 13.025z −2 + 6.55z −3 + 1.9125z −4

, using Eq.
1 + 0.55z −1 + 0.095z − 2 − 0.01z − 3 − 0.003z − 4
− 22.8896 − 7.1679z −1 + 3.4235z −2 + z −3
.
(8.132a) and (8.132b) we get G3 ( z) =
1 + 0.4813z −1 + 0.0735z − 2 + 0.00027z − 3
3.8259 + 5.0744 z −1 + z −2
and
Repeating the procedure twice we get G2 ( z) =
1 + 0.4803z −1 + 0.0721z − 2
4.4704 + z −1
. From G 4 ( z ), G 3 ( z), G 2 ( z), and G1( z), we arrive at the cascaded
G1( z) =
1 + 0.1579z −1
lattice realization shown below:
0.003
9.3
1.9125

_

_ 0.0721

0.00027

_

22.8896
–1
z

(c) From G3 ( z ) = H 3 ( z ) =

4.7915
0.9179

3.8259
z–1

z–1

z–1

7.8 − 8.96 z −1 + 4.62 z −2 + 4.92 z −3

, using Eq. (8.132a) and
1 + 0.7 z −1 − 6.68z − 2 + 2.356 z − 3
1.07 − 4.209z −1 + z −2
. Repeating the procedure we get
(8.132b) we get G2 ( z) =
1 − 1.8209z −1 + 3.2364 z − 2

Not for sale

260

G1( z) =

0.9179 + z −1

. From G 3 ( z), G 2 ( z ), and G1( z), we arrive at the cascaded lattice
1 − 4.7915z −1
realization shown below:
_ 2.356

_3.2364

4.7915

7.8

1.07

0.9179

z

4.92

z–1

–1

z

–1

8.32 When H N (z) is an allpass transfer function of the form
H N ( z ) = AN ( z ) =

d N + d N −1 z −1 + L + d 1z − N + 1 + z − N
, then from Eq. (8.132a), the
1 + d 1z −1 + L + d N −1 z − N + 1 + d N z − N

numerator coefficients of H N −1 ( z ) are given by
− p k + 1 d N d k + 1 − d N − k −1
p d
=
, and from Eq. (8.132b) the denominator
p k' = 0 k + 1
2 −1
p0 d N − p N
dN
coefficients of H N −1 ( z) are given by
p N − k −1d N − d N − k − 1 d k + 1d N − d N − k − 1
'
dN
=
= pk' , implying H N −1 ( z) is an
− k −1 =
2 −1
p0 d N − pN
dN
allpass transfer function of order N − 1. Since here p N = 1 and p0 = d N , the lattice
structure of Problem 8.30 then reduces to the lattice structure employed in the Gray-Markel
realization procedure.
8.33 (a) Consider the realization of Type 1B allpass structure. From its transfer parameters
given in Eq. (8.50b) we arrive at Y1 = z −1 X1 + (1 + z −1 ) X 2 = z −1 ( X1 + X 2 ) + X 2 , and
Y2 = (1 − z −1 ) X1 − z −1 X 2 = X1 − z −1 ( X1 + X 2 ). A realization of the two-pair based on
these two equations is as shown below which leads to the structure of Figure 8.24(b).

X1

+

_1

_
z 1

X2

+
Y2

+

Y1

(b) From the transfer parameters of Type 1At allpass given in Eq. (8.50c) we obtain
Y1 = z −1 X1 + X 2 , and Y2 = (1 − z −2 ) X1 − z −1 X 2 = X1 − z −1 ( z −1 X1 + X 2 ) = X1 − z −1Y1 .
A realization of the two-pair based on these two equations is as shown on next page which
leads to the structure of Figure 8.24(c).

Not for sale

261

_
z 1

X1

+

Y1

_1

X2

z

Y2

_1

+

(c) From the transfer parameters of Type 1At allpass given in Eq. (8.50d) we obtain
Y1 = z −1 X 1+ (1 − z −1 ) X 2 = z −1 ( X1 − X 2 ) + X 2 , and Y2 = (1 + z −1 ) X1 − z −1 X 2

= X1 + z −1 ( X1 − X 2 ). A realization of the two-pair based on these two equations is as
shown below which leads to the structure of Figure 8.24(d).
_
z 1

+

X1

+

_1

Y2

X2

+

Y1

8.34 (a) A cascade connection of 4 Type 1A first-order allpass networks is shown below:
_
z 1

+

_
z 1

+

+

+

_1

_1

a

_
z 1

_
z 1

+

+

_1
_
z 1

b

_
z 1

+

+

_1
_
z 1

c

_
z 1

d

Simple block diagram manipulation of the above structure leads to:
_
z 1

+
_1

+

_
z 1

+

_
z 1

+

z

c

b
_1

+

+

_1

z

z

a

_
z 1

+

_1

_1

_
z 1

+

d

_1

_1

Finally, by delay sharing between adjacent allpass sections we arrive at the following
equivalent realization requiring now 5 delays compared to 8 delays in the direct realization
shown on the previous page.
+

+

_
z 1

+

+

_
z 1

+

+

_
z 1

_1
_
z 1

a
_1

+

d

c

b

_
z 1

+

_1

_1

(b) A cascade connection of 4 Type 1At first-order allpass networks is shown below which
requires 8 delays.
_
_
_
_
z 1 +
z 1 +
z 1 +
z 1 +
a

+

_
z 1
_1

b

+

_
z 1
_1

Not for sale

c

+

_
z 1
_1

d

+

_
z 1
_1

262

By delay sharing between adjacent allpass sections we arrive at the following equivalent
realization requiring now 5 delays.
_
_
_
_
+
+
z 1
z 1
z 1
+
z 1 +
a
_1

_
z 1

_1

_1

_1

+

+

+

+

d

c

b

8.35 The structure of Figure P8.15 with internal variables labeled is shown below:
–1
–1

X(z)

W(z)

z

a

U(z)

Y(z)
–1

Its analysis yields (1): W ( z ) = X ( z ) − z −1U ( z ), (2): U ( z ) = a W ( z ) + z −1U ( z ), and (3):
a
W ( z). Substituting Eq.
Y ( z ) = −W ( z ) + U ( z ). From Eq. (2) we obtain (4): U ( z ) =
1 − z −1
⎛ 1 − a − z −1 ⎞
⎟W ( z ). Substituting Eq. (4) in Eq. (1) we get
(4) in Eq. (3) we get (5): Y ( z ) = −⎜⎜
−1 ⎟

1
z

⎛ 1 − a − z −1 + a z −1 ⎞
⎟W ( z ). Finally, from Eqs. (5) and (6) we arrive at
(6) : X ( z ) = ⎜

−1
1
z

1
Y ( z) − (1 − a ) + z
.
H (z) =
=
X ( z) 1 − (1 − a)z −1
8.36 We realize A2 ( z ) =

d1d 2 + d1 z −1 + z −2
in the form of a constrained three-pair as
1 + + d1 z −1 + d1d 2 z − 2

indicated below:
X1

Y1

t11 t12 t13

Y2

t21 t22 t23
t31 t32 t33

_d

X 3 d2

X2

Y3

⎡ Y1 ⎤ ⎡ t 11 t 12
From the above figure, we have ⎢⎢Y2 ⎥⎥ = ⎢⎢t 21 t 22
⎢⎣Y3 ⎥⎦ ⎢⎣t 31 t 32

Not for sale

1

t 13 ⎤ ⎡ X1 ⎤
t 23 ⎥⎥ ⎢⎢ X 2 ⎥⎥, and X 2 = − d1Y 2,
t 33 ⎥⎦ ⎢⎣ X 3 ⎥⎦

263

Y
N ( z)
, where
X 3 = d 2Y 3. From these equations, we get after some algebra A2 ( z ) = 1 =
X1 D ( z )
D( z) = 1 + d1t 22 − d 2 t 33 + d1d 2 (t 23 t 32 − t 22 t 33 ), and
N ( z) = t11 + d1 (t11t 22 − t12 t 21 ) − d 2 (t11t 33 − t13 t 31 )
+ d1d 2 {t 21 (t12 t 33 − t13 t 32 ) + t 31 (t 22 t13 − t12 t 23 ) + t11 (t 23 t 32 − t 22 t 33 )}.
Comparing the denominator of the desired allpass transfer function with D (z ) we get

t 22 = z −1 , t 33 = 0, t 23 t 32 = z −2 . Next, comparing the numerator of the desired allpass
transfer function with N (z ) we get t11 = z −2 , t12 t 21 = z −1 ( z −2 − 1), t13 t 31 = 0, and
t 32 (t11t 23 − t 21t13 ) + t 31 (t 22 t13 − t12 t 23 ) = 1. Substituting the appropriate transfer
parameters from the previous equations into the last equation we simplify it to
t13 t 21t 32 + t 31t12 t 23 = z −4 − 1.
Since t13 t 31 = 0, either t13 = 0, or t 31 = 0. (Both cannot be simultaneously equal to zero,
as this will violate the condition t13 t 21t 32 + t 31t12 t 23 = z −4 − 1.
Consider the case t13 = 0. Then the above equation reduces to t 31t12 t 23 = z −4 − 1. From
this equation and t 23 t 32 = z −2 , it follows that t 32 = z −2 , t 23 = 1, t 31t12 = z −4 − 1
= ( z −1 − 1)( z −1 + 1)( z −2 + 1). There are four possible realizable sets of values of t 21 and

t 31 satisfying the last equation and t12 t 21 = z −1 ( z −2 − 1). These lead to 4 different
⎡ z−2
z − 2 − 1 0⎤

⎢ −1
realizable transfer matrices for the three-pair: Type 2A: ⎢ z
1 ⎥,
z −1
⎢z − 2 + 1
0⎥
z−2

2

1

2

z
z + 1 0⎤
z
z 1 − 1 0⎤

1⎥, Type 2C: ⎢ z −1 ( z −1 + 1)
1 ⎥,
Type 2B: ⎢ z −1 ( z −1 − 1)
z −1
z −1
⎢( z − 2 + 1)( z −1 − 1)
⎢( z − 2 + 1)( z −1 + 1)
0⎥
0⎥
z−2
z−2

2

1 0⎤
z

⎢ −1 − 2
Type 2D: ⎢ z ( z − 1) z −1 1 ⎥. A realization of each of the above Type 2 allpass
⎢ z−4 −1
z − 2 0⎥

structures is obtained by implementing its respective transfer matrix, and then constraining
the Y2 and X 2 variables through the multiplier d1 and constraining the Y3 and X 3

variables through the multiplier − d2 resulting in the four structures shown in Figure 8.25
of the text.
It can be easily shown that the allpass structures obtained for the case t 31 = 0 are precisely
the transpose of the structures of Figure 8.25.
8.37 A cascade of two Type 2D second-order allpass networks is shown below which requires 8
delays.
Not for sale

264

w[n _1]
x[n]
_1

+

_
z 1

_
z 1

a

b

+

w[n]

+

_
z 1

_
z 1

_1

c

+

_
z 1

y[n]

_1

_1

w[n _1]

+

d

+

_1

w[n _2]

w[n _2]

_
z 1

_1

z

z

By delay sharing between adjacent allpass sections we arrive at the following equivalent
realizations requiring now 6 delays.

x[n]
_1

w[n _ 2]

+

_
z 1

_
z 1

a

b

+

+

_1

w[n _1]

w[n]
_1

+

_
z 1

c

_
z 1

+

y[n]

d

+
_1

_
z 1

_
z 1

The minimum number of delays needed to implement a cascade of Type 2D second-order
allpass sections is thus 4 + 2( M − 1) = 2( M + 1).

d 2 + d1 z −1 + z −2
in the form of a constrained three-pair as indicated
1 + d1 z −1 + d 2 z − 2
in the solution of Problem 8.36. Comparing the numerator and the denominator of the
Type 3 allpass transfer function with N (z ) and D (z ) given in the solution of Problem 8.36
we arrive at
t 11= z −2 , t 22 = z −1 , t 22 = z −2 , t 33 = z −2 , t 23 t 32 = z −3 , t 12 t 212 = z −1 ( z −2 − 1),

8.38 We realize A2 ( z ) =

t 13 t 31= z −4 − 1, t 13 t 21t 32 + t 13 t 21t 23 = z −3 ( z −2 − 1) + z −1 ( z −4 − 1).
To solve the last four equations to arrive at a set of realizable set of values for the transfer
parameters, t12 , t 21 , t13 , t 31 , t 23 , and t 32 , we need to pre-select t 23 and t 32 satisfying the
condition t 23t 32 = z −3 , and then determine realizable values for the other transfer
parameters.
(a) Let t 23 = z −1 , t 32 = z −2 . There are 4 possible realizable sets of values of t12 , t 21 , t13 ,
and t 31 satisfying the constraints given above resulting in the following transfer matrices
of the three-pair:

Not for sale

265

⎡ z−2

z − 2 − 1 z − 2 − 1⎤
z−2
z −1 + 1 z −1 + 1⎤

Type 3A: ⎢ z −1
z −1
z −1 ⎥, Type 3B: ⎢ z −1 ( z −1 − 1)
z −1
z −1 ⎥,
⎢z − 2 + 1
⎢( z −1 − 1)( z − 2 + 1)
z−2
z−2 ⎥
z−2
z−2 ⎥

2

1

1

2

1
1 ⎤
z
z − 1 z − 1⎤
z

Type 3C: ⎢ z −1 ( z −1 + 1)
z −1
z −1 ⎥, Type 3D: ⎢ z −1 ( z − 2 − 1) z −1 z −1 ⎥.
⎢( z −1 + 1)( z − 2 + 1)
⎢ z−4 −1
z−2
z−2 ⎥
z−2 z−2 ⎥

(b) t 23 = z −2 , t 32 = z −1. This leads to 4 new realizations which are transpose of the 4
structures developed in Case (a).
(c) t 23 = 1, t 32 = z −3 . Here also there are 4 different realizations whose transfer matrices
are as given below:
⎡ z−2
z −1 ( z − 2 − 1) z − 2 − 1⎤

Type 3E: ⎢ 1
z −1
z −1 ⎥,
⎢z − 2 + 1
z−3
z−2 ⎥

2

1

1

z
z ( z + 1) z −1 + 1⎤

1 ⎥,
Type 3F: ⎢
z −1 − 1
z −1
⎢( z −1 − 1)( z − 2 + 1)
z−3
z−2 ⎥

2

1

1

1

⎡ z−2
1 ⎤
z
z ( z − 1) z − 1⎤
z −1

1 ⎥, Type 3H: ⎢ z − 2 − 1 z −1
1 ⎥.
Type 3G: ⎢
z −1 + 1
z −1
⎢( z −1 + 1)( z − 2 + 1)
⎢z − 4 − 1 z − 3 z − 2 ⎥
z −3
z−2 ⎥

(d) t 23 = z −3 , t 32 = 1. This leads to 4 new realizations which are transpose of the 4
structures developed in Case (b).
8.39 A cascade of two Type 3B second-order allpass networks is shown below which requires 8
delays.

+

+

_1

c

a
x[n]

_1

z

+

_1

_1

+

z

w[n]

_1

z

+

_
z 1

+

_1

z

b
_1

+

w[n _ 2]

Not for sale

z

d

w[n _1]

_
z 1

y[n]
_1

_1

_
z 1

+

266

By delay sharing between adjacent allpass sections we arrive at the following equivalent
realizations requiring now 7 delays.

+

+

_1

w[n _1]

a
x[n]

_1

z

_1

+

+

z

w[n]

_1

_1

c
_
z 1

+

z

z

d

w[n _ 2]
_1

_1

+

y[n]
_1

_
z 1

b

+

_
z 1

+

The minimum number of delays needed to implement a cascade of Type 3B second-order
allpass sections is thus 4 + 3( M − 1) = 3( M + 1).
8.40 From the transfer parameters given in Eq. (8.50d) we arrive at the input-output relations of
the two-pair as (1): Y1 = k m X1 + z −1 X 2 , and (2): Y2 = (1 − k m2 ) X1 − k m z −1 X 2 . A threemultiplier realization based on these two equations is shown below:
1 _ km2

+

X1

_k

km
Y1

Y2
m

_
z 1

+

X2

8.41 Eq. (2) in the solution of Problem 8.40 can be rewritten as:
2 X − k z −1 X = X − k (k X + z −1 X ). Substituting Eq. (1) of the
Y2 = X1 − k m
1
m
2
1
m m 1
2

solution of Problem 8.40 in this equation we then get (3): Y2 = X1 − k m Y1 . A
realization of the two-pair based on Eq. (3) and Eq. (1) of the solution of Problem 8.40
results in the lattice structure shown below:
X1

Y2
_k

m

km
Y1

z–1

X2

From Eqs. (1) and (2) of the solution of Problem 8.40 we arrive at the transfer
2,
parameters of the lattice two-pair given by t11 = k m , t12 = z −1 , t 21 = 1 − k m

t 22 = − k m z −1 . The corresponding chain parameters are obtained using Eq. (7.128b)

Not for sale

267

and are given by A =

km
k z −1
z −1
,B = m
,C=
,D=
.
2
2
2
2
1 − km
1 − km
1 − km
1 − km
1

X1 (z)

W 1 (z)
_k

_k

2

1

k1

k2
Y 1(z)

z

–1

z

–1

S 1 (z)

Hence, the input-output relation of the all-pole cascade lattice structure given above can
⎡ 1 k 2 z −1 ⎤ ⎡ 1 k1 z −1 ⎤ ⎡W1 ( z)⎤
⎡ X ( z )⎤
1
be expressed as: ⎢ 1 ⎥ =
⎥⎢
⎥⎢

⎥, from
z −1 ⎦⎥ ⎣⎢ k1 z −1 ⎦⎥ ⎣ S1 ( z ) ⎦
⎣ Y1 ( z ) ⎦ (1 − k 2 )(1 − k 2 ) ⎣⎢ k 2
2

which we obtain H r ( z ) =

1

(1 − k 22 )(1 − k12 )
W1 ( z )
.
=
X1 ( z ) 1 + k1 (1 + k 2 )z −1 + k 2 z − 2

X 1(z)

W 1 (z)
_k

_k

2

1

k1

k2

z–1

Y 1(z)

z–1

S 1 (z)

Likewise, from Section 8.8.1, we arrive at input-output relation of the all-pole cascade
⎡ X ( z )⎤ ⎡ 1 k 2 z −1 ⎤ ⎡ 1 k1 z −1 ⎤ ⎡W1 ( z )⎤
lattice structure given above as: ⎢ 1 ⎥ = ⎢
⎥⎢
⎥⎢
⎥, from
z −1 ⎥⎦ ⎢⎣ k1 z −1 ⎥⎦ ⎣ S1 ( z ) ⎦
⎣ Y1 ( z ) ⎦ ⎢⎣ k 2
W1 ( z)
1
.
=
X1 ( z) 1 + k1 (1 + k 2 )z −1 + k 2 z − 2
Now a second-order all-pole transfer function can also be expressed in terms of its
1
. Comparing the denominator of with that
poles as H ( z) =
1 − 2r cos(ω o )z −1 + r 2 z − 2

which we obtain H s ( z) =

of H s (z) and H r ( z), we observe k 2 = r 2 and k1 (1 + k 2 ) = −2r cos(ωo ). As r ≅ 1,
2r cos(ωo )
k1 = −
≅ − cos(ωo ). At the true resonance peak, we have
1+ r2
1
1
. Likewise, at ω = ωr , H s (e jωr ) =
. As
H (e ωr ) =
(1 − r 2 ) sin(ωo )
(1 − k 2 ) 1 = k12
r → 1, k 2 = r 2 → 1, and (1 − k12 )(1 − k 22 ) = (1 − k12 )(1 − k 2 )(1 + k 2 ) ≅ 2(1 − k12 )(1 − k 2 ).

Hence, at ω = ωr , H r (e jωr ) ≅

2(1 − k12 )(1 − k2 )
(1 − k 2 ) 1 − k12

= 2 1 − k12 = 2 sin(ωr ) . This

indicates that the gain at the peak resonance is approximately independent of the pole

Not for sale

268

radius as long as the pole angle is constant and the radius is close to 1.
3.9 + 2.3z −1 + z −2

8.42 (a) H1 ( z) =

. Choose A2 ( z ) =

0.5 + 0.3z −1 + z −2

. Note
1 + 0.3z −1 + 0.5z − 2
1 + 0.3z −1 + 0.5z − 2
k2 = A2 (∞) = d 2 = 0.5 < 1. Using Eq. (8.45) we next determine the coefficient d1' of
0.2 + z −1
the allpass transfer function A1 ( z) arriving at A1 ( z) =
. Here,
1 + 0.2 z −1
k1 = A1 (∞ ) = d1' = 0.2 < 1. Therefore, A2 ( z) and hence, H1 ( z) is stable.
The feed-forward coefficients are given by α1 = p2 = 1.0, α2 = p1 − α1d 1= 2.0,

α3 = p0 − α1d 2 −α2 d1' = 3.0. Final realization of H1 ( z) is as shown below:
X1
_

_

0.5

0.2
0.2

0.5

z

–1

z

–1
3

2

Yo

(b) H 2 ( z) =

2.6 + 0.74 z −1 + 3z −2

. Choose A2 ( z) =

0.4 − 0.42 z −1 + z −2

. Note
1 − 0.42 z −1 + 0.4 z − 2
1 − 0.42 z −1 + 0.4 z − 2
k2 = A2 (∞) = d 2 = 0.4 < 1. Using Eq. (8.45) we next determine the coefficient d1' of
− 0.3 + z −1
the allpass transfer function A1 ( z) arriving at A1 ( z ) =
. Here,
1 − 0.3z −1
k1 = A1 (∞ ) = d1' = 0.3 < 1. Therefore, A2 ( z) and hence, H 2 ( z) is stable.

The feed-forward coefficients are given by α1 = p2 = 3.0, α2 = p1 − α1d 1= 2.0,
α3 = p0 − α1d 2 −α2 d1' = 2.0. Final realization of H 2 ( z) is as shown below:
X1
_

0.4

_ 0.3

0.4

0.3

z–1

z
2

3

–1
2

Yo

(c) H 3 ( z) =

− 3 + 5.192 z −1 − 3.56 z −2 + 2 z −3
1 − 0.28 z −1 + 0.056 z − 2 + 0.4 z − 3

Not for sale

. Choose

269

A3 ( z) =

0.4 + 0.056 z −1 − 0.28z −2 + z −3

Note k3 = A3 (∞ ) = d3 = 0.4 < 1. Using
1 − 0.28z −1 + 0.056 z − 2 + 0.4 z − 3
Eq. (8.45) we next determine the coefficient d1' and d 2' of the allpass transfer function
0.2 − 0.36 z −1 + z −2

. Now k2 = A2 (∞) = d 2 = 0.2 < 1.
1 − 0.36 z −1 + 0.2 z − 2
Using Eq. (8.45) we next determine the coefficient d1" of the allpass transfer function

A2 ( z) arriving at A2 ( z) =

− 0.3 + z −1

A1 ( z) arriving at A1 ( z ) =

1 − 0.3z −1

. Here, k1 = A1 (∞ ) = d1' = 0.3 < 1. Therefore,

A3 ( z) and hence, H 3 ( z) is stable.
The feed-forward coefficients are given by α1 = p3 = 2.0, α 2 = p2 − α1d 1= −3.0,
α3 = p1 − α1d 2 −α2 d1' = 4.0, α 4 = p0 − α1d 3 −α2 d 2' − α3 d1" = −2.0. Final realization
of H 3 ( z) is as shown below:
X1
_ 0.4

_ 0.2

0.4

0.3
_

0.2
_1

_1

z

2

_

z1

z

_3

0.3

4

_2

Yo

(d) H 4 ( z) =

1.63 + 3.112 z −1 + 0.84 z −2 + 2 z −3

. Choose
1 + 0.42 z −1 + 0.056 z − 2 − 0.3z − 3
− 0.3 + 0.056 z −1 + 0.42 z −2 + z −3
A3 ( z) =
Note k3 = A3 (∞) = d3 = 0.3 < 1. Using
1 + 0.42 z −1 + 0.056 z − 2 − 0.3z − 3
Eq. (8.45) we next determine the coefficients d1' and d 2' of the allpass transfer function

A2 ( z) arriving at A2 ( z) =

0.2 + 0.48z −1 + z −2

.
1 + 0.48z −1 + 0.2 z − 2
Using Eq. (8.45) we next determine the coefficient

A1 ( z) arriving at A1 ( z) =

0.4 + z −1
1 + 0.4 z −1

Now k2 = A2 (∞) = d 2 = 0.2 < 1.
d1" of the allpass transfer function

. Here, k1 = A1 (∞ ) = d1' = 0.4 < 1. Therefore,

A3 ( z) and hence, H 4 ( z) is stable.
The feed-forward coefficients are given by α1 = p3 = 2.0, α 2 = p2 − α1d 1= 0,
α3 = p1 − α1d 2 −α2 d1' = 3.0, α 4 = p0 − α1d 3 −α2 d 2' − α3 d1" = 1.0. Final realization
of H 3 ( z) is as shown on the next page.

Not for sale

270

X1
_

0.4
_1
_

_1

z1

_1

z

0.6

z

3.2

2.2

Yo

2 + 0.6 z −1 − z −2 + 0.3z −3

(e) H 5 ( z ) =

. Choose
1 − 0.1z −1 + 0.18z − 2 + 0.08z − 3 − 0.08z − 4
− 0.08 + 0.08z −1 + 0.18 z −2 − 0.1z −3 + z −4
A4 ( z) =
. Note
1 − 0.1z −1 + 0.18 z − 2 + 0.08 z − 3 − 0.08z − 4
k 4 = A4 (∞) = d 4 = 0.08 < 1. Using Eq. (8.45) we next determine the coefficients

d1' , d 2' and d3' of the allpass transfer function A3 ( z) arriving at

A3 ( z) =

0.0725 + 0.195z −1 − 0.0942 z −2 + z −3

. Now
1 − 0.0942 z −1 + 0.1957z − 2 + 0.0725z − 3
k3 = A3 (∞) = d3' = 0.0725 < 1. Using Eq. (8.45) we next determine the coefficients d1"

and d 2" of the allpass transfer function A2 ( z) arriving at
A2 ( z ) =

0.2035 − 0.1090 z −1 + z −2
1 − 0.1090 z −1

+ 0.2035z − 2

. Now k 2 = A2 (∞) = d 2" = 0.2035 < 1. Using Eq.

(8.45) we next determine the coefficient d1' ' ' of the allpass transfer function A1 ( z)
arriving at A1 ( z ) =

− 0.009058 + z −1
1 − 0.0905z −1

. Here, k1 = A1 (∞) = d1' ' ' = 0.0905 < 1.

Therefore, A4 ( z) and hence, H 5 ( z ) is stable.
The feed-forward coefficients are given by α1 = p 4 = 0, α 2 = p3 − α1d 1= −0.3,

α3 = p2 − α1d 2 −α2 d1' = −1.0283, α 4 = p1 − α1d 3−α 2 d 2' − α 3 d1" = 0.5467,
α5 = p0 − α1d 4 −α2 d3' − α3 d 2" − α 4 d1' ' ' = 2.2805. Final realization of H 5 ( z ) is as
shown below:
X1
_

_ 0.0725

0.08

0.0725
_

0.3

0.0905

0.2035

_ 0.0905

_1

z1
_

0.2035

z
_

1.0283

_

z1

0.5467

_

z1

2.2805

Yo

8.43 (a) H1 ( z ) =

0.32 z −1 + 0.66 z −2 − 0.72 z −3
1 − 2.35z −1 + 3.36 z − 2 − 1.575z − 3

Not for sale

. Choose

271

− 1.575 + 3.36 z −1 − 2.35z −2 + z −3

. Note k3 = A3 (∞) = d3 = 1.575 > 1.
1 − 2.35z −1 + 3.36 z − 2 − 1.575z − 3
Therefore, A3 ( z) and hence, H 3 ( z) is unstable. Using Eq. (8.45) we next determine
A3 ( z) =

the coefficients d1' and d 2' of the allpass transfer function A2 ( z) arriving at
3.5105 − 2.6308z −1 + z −2
A2 ( z ) =
. Now k2 = A2 (∞) = d 2 = 3.5105 > 1. Using Eq.
1 − 2.6308z −1 + 3.5105z − 2
(8.45) we next determine the coefficient d1" of the allpass transfer function A1 ( z)
arriving at A1 ( z ) =

− 0.5833 + z −1
1 − 0.5833z −1

. Here, k1 = A1 (∞ ) = d1" = 0.5833 < 1.

The feed-forward coefficients are given by α1 = p3 = 2.0, α 2 = p2 − α1d 1= 0,
α3 = p1 − α1d 2 −α2 d1' = 3.0, α 4 = p0 − α1d 3 −α2 d 2' − α3 d1" = 1.0. Final realization
of H 3 ( z) is as shown below:
X1
_ 3.5105

1.575
_ 1.575

0.5833
_

3.5105
_1

0.5833

_1

z

_

z1

z

3

2

Yo

(b) H 2 ( z) =

9.3 − 39.3z −1 − 13.025z −2 + 6.552 z −3 + 1.9125z −4

. Choose
1 + 0.55z −1 + 0.095z − 2 − 0.01z − 3 − 0.003z − 4
− 0.003 − 0.01z −1 + 0.095z −2 + 0.55z −3 + z −4
A4 ( z) =
. Note
1 + 0.55z −1 + 0.095z − 2 − 0.01z − 3 − 0.003z − 4
k 4 = A4 (∞) = d 4 = 0.003 < 1. Using Eq. (8.45) we next determine the coefficients
d1' , d 2' and d3' of the allpass transfer function A3 ( z) arriving at
− 0.0084 + 0.0953z −1 + 0.55z −2 + z −3
A3 ( z) =
. Now
1 + 0.55z −1 + 0.0953z − 2 − 0.0084 z − 3

k3 = A3 (∞ ) = d3' = 0.0084 < 1. Using Eq. (8.45) we next determine the coefficients
d1" and d 2" of the allpass transfer function A2 ( z) arriving at
0.0999 + 0.5508z −1 + z −2
A2 ( z) =
. Now k2 = A2 (∞) = d 2 = 0.0999 < 1. Using Eq.
1 + 0.5508z −1 + 0.0999z − 2
(8.45) we next determine the coefficient d1" of the allpass transfer function A1 ( z)

arriving at A1 ( z ) =

0.5088 + z −1
1 + 0.5088z −1

. Here, k1 = A1 (∞ ) = d1" = 0.5088 < 1. As a result,

A3 ( z) , and hence, H 2 ( z) is stable.

Not for sale

272

The feed-forward coefficients are given by
α1 = p 4 = 1.9125, α 2 = p3 − α1d 1= 5.5001, α 3 = p 2 − α1d 2 −α 2 d1' = −16.2316,
α 4 = p1 − α1d 3−α 2 d 2' − α 3 d1" = −30.8644,
α 5 = p0 − α1d 4 −α 2 d 3' − α 3 d 3" − α 4 d 4' ' ' = 26.4295. Final realization of H 2 (z ) is as
shown below:
X1
_

0.003

0.0084

0.003

_ 0.0084

_

0.0999

_

_

z1

_ 30.8644

1.6985
_
z 1

_
z 1
_ 0.7198

+

0.1891

0.1574

+

+

_
z 1

y[n]

+

+
1.6985

0.2545

26.4295
Yo

. A direct form II realization of

+

1 − 0.1891z −1 + 0.7198 z − 2 − 0.1574 z − 3
H (z ) is shown below:

(b) H ( z ) =

z

_ 16.2316

0.2545(1 + 1.6985 z −1 + 1.6985 z −2 + z −3 )

x[n]

_1

z1

5.5001

1.9125

0.5088
0.5088

_

z1

8.44 (a) H ( z ) =

_

0.0999

0.2545(1 + z −1 )(1 + 0.6985 z −1 + z

−2

)

(1 − 0.2169 z −1 )(1 + 0.0278 z −1 + 0.7258 z
H (z ) is shown below:

−2

)

0.2545
x[n]

+

+

+

+

y[n]

_
z 1

_
z 1

_

+

0.6985

0.0278

0.2169

+
_
z 1

_ 0.7285

(c) We first form A3 ( z) =

− 0.1574 + 0.71985z −1 − 0.1891z −2 + z −3
1 − 0.1891z −1 + 0.7198z − 2 − 0.1574 z − 3

. Now

k3 = A3 (∞ ) = d3' = 0.1574 < 1. Using Eq. (8.45) we next determine the coefficients
d1" and d 2" of the allpass transfer function A2 ( z) arriving at

Not for sale

273

0.7075 − 0.0777z −1 + z −2

. Now k2 = A2 (∞) = d 2 = 0.7075 < 1. Using Eq.
1 − 0.0777z −1 + 0.7075z − 2
(8.45) we next determine the coefficient d1" of the allpass transfer function A1 ( z)
A2 ( z ) =

arriving at A1 ( z ) =

− 0.0455 + z −1

. Here, k1 = A1 (∞ ) = d1" = 0.0455 < 1.
1 − 0.0455z −1
The feed-forward coefficients are given by α1 = p3 = 0.2545, α2 = p2 − α1d 1= 0.4804,
α3 = p1 − α1d 2 −α2 d1' = 0.2864, α 4 = p0 − α1d 3 −α2 d 2' − α3 d1" = −0.0323. The
Gray-Markel realization of H (z ) is as shown below:
X1
_

0.1574
_ 0.1574

0.0455

0.7075

_ 0.0455

0.7075
_1

_1

z

0.4804

0.2545

_1

z

z

_ 0.0323

0.2864

Yo

(d) Using Eqs. (8.132a) and (8.132b) we determine the coefficients of H 2 ( z) :
1.6309 + 0.8456 z −1 + z −2
H 2 ( z) =
. Next, using Eqs. (8.132a) and (8.132b) we
1 + 0.0676 z −1 + 0.8529z − 2
− 1.8808 + z −1
determine the coefficients of H1 ( z) : H1 ( z) =
1 + 1.6718z −1
realization of using the method of Problem 8.30 is thus as shown below:
X1
_

0.1574

1.6309

0.2545
Y1

0.2545

0.8529

z

–1

z

_

1.6718

_

1.8808

–1

z

–1

A comparison of the hardware requirements of the four realizations is as follows:
Direct form: No. of multipliers = 7 and no. of 2-input adders = 6,
Cascade form: No. of multipliers = 10 and no. of 2-input adders = 9,
Gray-Markel realization: No. of multipliers = 5 and no. of 2-input adders = 6,
Cascaded lattice realization: No. of multipliers = 6 and no. of 2-input adders = 6.
Hence, the Gray-Markel realization has the smallest hardware requirements.
8.45 (a) A partial-fraction expansion of G (z ) is of the form
N /2

N /2
N /2
υi
υi
υi*
d
+
G( z ) = d +
. If we define H ( z ) = +
, then we can
*

z

λ
z
λ
2
z

λ
i
i
i
i =1
i =1
i =1

Not for sale

274

write G( z) = H ( z) + H* ( z), where H* ( z) represents the transfer function obtained from
H (z ) by conjugating its coefficients.
(b) In this case, the partial-fraction expansion of G (z ) is of the form
Nr

N /2

N /2

c
c
ρi
υi
υi*
, where N r and N c are the number
G( z) = d +
+
+
*
z

ξ
z

λ
z

λ
i
i
i
i =1
i =1
i =1
of real poles ξ i ’s and complex poles λi ’s, respectively, with residues ρi ’s and υi ’s.

We can thus decompose G (z ) as G( z) = H ( z) + H* ( z), where
N

N /2

c
r
ρi / 2
υi
d
.
H (z) = +
+
z

λ
2 i =1 z − ξ i
i
i =1

(c) An implementation of real coefficient G (z ) is thus simply a parallel connection of
two complex-coefficient filters characterized by transfer functions H (z ) and H* ( z) as
indicated in the figure below:
x[n]

x[n]

H(z)

real
imaginary

H* (z)

=0

However, for a real-valued input x[ n], the output of H (z ) is the complex conjugate of
that of H* ( z). As a result, two times the real part of the output of H (z ) is the desired
real-valued output sequence y[n] indicating that a single complex-coefficient filter
H (z ) is sufficient to realize G (z ) as indicated below:
2

x[n]

H(z)

y[n]

Y ( z)
A + jB
=
, we arrive at the difference equation
X ( z ) 1 + (α + j β ) z − 1
representation yre [ n] + j yim [ n] = −(α + jβ )(yre [ n − 1] + j yim [ n − 1]) + Ax[ n] + jBx[ n],
which is equivalent to a set of two difference equations involving all real variables and
real multiplier coefficients: yre [ n] = −α yre [ n − 1] + β yim [ n − 1] + A x[ n] and
yim [ n] = − β yre [ n − 1] − α yim [ n − 1] + B x[ n]. A realization of H (z ) based on the last
two equations is shown on next page.
To determine the transfer function Yre ( z) / X ( z ), we take the z -transform of the last

8.46 From H ( z) =

two difference equations and arrive at (1 + α z −1 )Yre ( z) − β z −1Yim ( z) = A X ( z) and

β z −1Yre ( z ) + (1 + α z −1 )Yim ( z ) = B X ( z). Solving these two equations we get

Not for sale

275

Yre ( z )
A + ( Aα + Bβ )z −1
, which is seen to be a second-order transfer
=
X ( z ) 1 + 2αz −1 + (α 2 + β 2 )z − 2
function.
A
y re [n]

x[n]

–1

z

B

y im [n]

z

–1

β

To determine the transfer function Yre ( z) / X ( z ), we take the z -transform of the last
two difference equations and arrive at (1 + α z −1 )Yre ( z) − β z −1Yim ( z) = A X ( z) and

β z −1Yre ( z ) + (1 + α z −1 )Yim ( z ) = B X ( z). Solving these two equations we get
Yre ( z )
A + ( Aα + Bβ )z −1
, which is seen to be a second-order transfer
=
X ( z ) 1 + 2αz −1 + (α 2 + β 2 )z − 2
function.
8.47 An m -th order complex allpass function is given by
* + α * z −1 + L + α * z − m + 1 + z − m
αm
m −1
1
. To generate an ( m − 1) –th
Am ( z ) =

1

2
1 + α 1 z + α 2 z + L + + α m −1 z − m + 1 + α m z − m
* ⎤
⎡ A (z) − km
Pm −1 ( z )
= z⎢ m
⎥.
Dm −1 ( z )
⎢⎣ 1 − k m Am ( z ) ⎦⎥
Substituting the expression for A m (z ) in the above equation we obtain after some
−2 + L + α * z − m + 1 + z − m
* + α * z −1 + α *
algebra Pm −1 ( z ) = z[α m
1
m −1
m−2z

order allpass function, we use the recursion Am −1 ( z) =

− m + 1 + α z − m )]
* (1 + α z −1 + α z −2 + L + α
− αm
1
2
m −1 z
m

* −1 − α m
* α1 ) + (α m
* − 2 − αm
* α 2 ) z − 1 + L + (α1* − α m
* α m − 1 ) z − m + 2 + (1 − α m ) z − m + 1 ,
= (α m
2

and Dm −1 ( z) = 1 + α1 z −1 + α 2 z −2 + L + α m −1 z − m +1 + α m z − m
* + α * z −1 + L + α * z − m + 1 + z − m )
− α m (α m
1
m −1
* − 2 ) z − 2 + L + (α m − 1 − α m α1* ) z − m + 1 .
* − 1 ) z − 1 + (α 2 − α m α m
= (1 − α m ) + (α1 − α m α m
Thus, A m −1 ( z) is a complex allpass function of order m − 1 and given by
2

Not for sale

276

Am −1 ( z ) =

−1 + L + β * z − m + 2 + z − m + 1
*
*
βm
−1 + β m − 2 z
1
, where
1
2

1 + β1 z + β 2 z + L + β m − 2 z − m + 2 + β m −1 z − m + 1

*
α − αm αm
− k 1 ≤ k ≤ m − 1.
,
βk = k
2
1 − αm

To develop a realization of A m (z ) we express A m (z ) in terms of Am −1 ( z ) :
k * − z −1 Am −1 ( z )
Y
, and compare it with Eq. (8.47) resulting in the
Am ( z ) = 1 = m
X1 1 + k m z −1 Am −1 ( z )
*,
following expressions for the transfer parameters of the two-pair: t11 = k m
* )z −1 . As in the case of the realization of a real
t 22 = − k m z −1 , and t12 t 21 = (1 − k m k m

allpass function, there are many possible choices for t12 and t 21. We choose
* )z −1 , t = 1. The corresponding input-output relations of the two-pair
t12 = (1 − k m k m
21
* X + (1 − k * k )z −1 X = k * ( X − k z −1 X ) + z −1 X and
are: Y1 = k m
1
m m
2
m 1
m
2
2

Y2 = X1 − k m z −1 X 2 . A realization of A m (z ) based on the above two-pair relations is
indicated below:
X1
_k

A m (z)

Y2
m

A m _1(z)

*
km

Y1

z

X2

–1

−1 ⎞
⎛ 1 + z −1 ⎞ 1 ⎛ 8 + 8 z −1 ⎞ 1 ⎛
⎟ = ⎜ 1 + 3 + 5z ⎟ = 1 ( A0 ( z ) + A1 ( z )),
⎟= ⎜
8.48 (a) H1 ( z ) = 4⎜
⎜ 5 + 3z −1 ⎟ 2 ⎜ 5 + 3z − 1 ⎟ 2 ⎜
5 + 3z −1 ⎟⎠ 2

3 + 5z −1
where A0 ( z) = 1 and A1 ( z ) =
.
5 + 3z − 1
−1 ⎞
⎛ 5 − 5z − 1 ⎞ 1 ⎛ 5 − 5z −1 ⎞ 1 ⎛
⎟ = ⎜ 1 − − 2 + 3z ⎟ = 1 ( A0 ( z ) − A1 ( z )),
⎟= ⎜
(b) H 2 ( z ) = ⎜⎜

−1 ⎟
−1 ⎟
3 − 2 z −1 ⎟⎠ 2
⎝ 6 − 4z ⎠ 2 ⎝ 3 − 2z ⎠ 2 ⎝
− 2 + 3z −1
where A0 ( z) = 1 and A1 ( z ) =
.
3 − 2 z −1
0.5414(1 − z −1 )(1 − 0.0757z −1 + z −2 )
1
= [A0 ( z ) − A1 ( z)], where
(c) H 3 ( z) =
1
1
2

(1 − 0.1768z )(1 − 0.004 z + 0.9061z ) 2

A0 ( z) =

0.9061 − 0.004 z −1 + z −2
1 − 0.004 z −1 + 0.9061z − 2

(d) H 4 ( z) =

and A1 ( z ) =

− 0.1768 + z −1
1 − 0.1768z −1

0.4547(1 + z −1 )(1 − 0.2859z −1 + z −2 )
(1 + 0.0712 z −1 )(1 − 0.0377z −1 + 0.8482 z − 2 )

Not for sale

=

.

1
[A0 ( z) − A1( z)] , where
2

277

A0 ( z ) =

0.0712 + z −1
1 + 0.0712 z −1

and A1 ( z) =

0.8482 − 0.0377z −1 + z −2
1 − 0.0377z −1 + 0.8482 z − 2

.

8.49 (a) From the given equation we get
y[ 2l] = h[ 0] x[ 2l ] + h[1] x[ 2l − 1] + h[ 2] x[ 2l − 2] + h[3] x[ 2l − 3] + h[ 4] x[ 2l − 4] + h[ 5] x[ 2l − 5].
y[ 2l + 1] = h[ 0] x[ 2l + 1] + h[1] x[ 2l] + h[ 2] x[ 2l − 1] + h[3] x[ 2l − 2] + h[ 4] x[ 2l − 3] + h[ 5] x[ 2l − 4].

Rewriting the above equations in matrix form we arrive at
⎡ y[2l] ⎤ ⎡h[0] 0 ⎤ ⎡ x[2l] ⎤ ⎡h[2] h[1] ⎤ ⎡ x[2l − 2]⎤
⎢ y[2l + 1]⎥ = ⎢ h[1] h[0]⎥ ⎢ x[2l + 1]⎥ + ⎢ h[3] h[2]⎥ ⎢ x[2l − 1] ⎥

⎦ ⎣
⎦ ⎣
⎦⎣

⎦⎣
⎡h[ 4] h[3]⎤ ⎡ x[2l − 4]⎤ ⎡0 h[5]⎤ ⎡ x[2l] ⎤
+⎢
⎥+⎢
⎥⎢
⎥,
⎥⎢
⎣ h[5] h[ 4]⎦ ⎣ x[2l − 3]⎦ ⎣0 0 ⎦ ⎣ x[2l + 1]⎦
which can be alternately expressed as Yl = H 0 X l + H1X l −1 + H 2 X l − 2 + H 3 X l − 3 ,

⎡ y[2l] ⎤
⎡ x[2l] ⎤
⎡h[0] 0 ⎤
⎡h[2] h[1] ⎤
where Yl = ⎢
, H1 = ⎢
, H0 = ⎢
, Xl = ⎢

⎥,
⎣ y[2l + 1]⎦
⎣ x[2l + 1]⎦
⎣ h[1] h[0]⎦
⎣ h[3] h[2]⎦
⎡h[ 4] h[3]⎤
⎡0 h[5]⎤
H2 = ⎢
, H3 = ⎢

⎥.
⎣ h[5] h[ 4]⎦
⎣0 0 ⎦
(b) Here we have

y[3l ] = h[ 0] x[3l] + h[1] x[3l − 1] + h[ 2] x[3l − 2] + h[3] x[3l − 3] + h[ 4] x[3l − 4] + h[ 5] x[3l − 5].
y[3l + 1] = h[0]x[3l + 1] + h[1]x[3l] + h[2]x[3l − 1] + h[3]x[3l − 2] + h[ 4]x[3l − 3] + h[5]x[3l − 4],
y[3l + 2] = h[0]x[3l + 2] + h[1]x[3l + 1] + h[2]x[3l] + h[3]x[3l − 1] + h[ 4]x[3l − 2] + h[5]x[3l − 3].

Rewriting the above equations in matrix form we arrive at
0 ⎤ ⎡ x[3l] ⎤ ⎡ h[3] h[2] h[1] ⎤ ⎡ x[3l − 3]⎤
⎡ y[3l] ⎤ ⎡h[0] 0
⎢ y[3l + 1] ⎥ = ⎢ h[1] h[0] 0 ⎥ ⎢ x[3l + 1] ⎥ + ⎢h[ 4] h[3] h[2]⎥ ⎢ x[3l − 2]⎥

⎥⎢
⎥ ⎢
⎥⎢
⎥ ⎢

⎢⎣ y[3l + 2]⎥⎦ ⎢⎣h[2] h[1] h[0]⎥⎦ ⎢⎣ x[3l + 2]⎥⎦ ⎢⎣ h[5] h[ 4] h[3]⎥⎦ ⎢⎣ x[3l − 1] ⎥⎦
⎡0 h[2] h[1] ⎤ ⎡ x[3l − 6]⎤
+ ⎢⎢0 0 h[2]⎥⎥ ⎢⎢ x[3l − 5]⎥⎥,
⎢⎣0 0
0 ⎥⎦ ⎢⎣ x[3l − 4]⎥⎦

which can be alternately expressed as Yl = H 0 X l + H1X l −1 + H 2 X l − 2 , where

0 ⎤
⎡ h[3] h[2] h[1] ⎤
⎡h[0] 0
⎡ x[3l] ⎤
⎡ y[3l] ⎤

Y l = ⎢ y[3l + 1] ⎥, X l = ⎢ x[3l + 1] ⎥, H 0 = ⎢ h[1] h[0] 0 ⎥, H1 = ⎢⎢h[ 4] h[3] h[2]⎥⎥,
⎢⎣ h[5] h[ 4] h[3]⎥⎦
⎢⎣h[2] h[1] h[0]⎥⎦
⎢⎣ x[3l + 2]⎥⎦
⎢⎣ y[3l + 2]⎥⎦

Not for sale

278

⎡0 h[2] h[1] ⎤
and H 2 = ⎢⎢0 0 h[2]⎥⎥.
⎢⎣0 0
0 ⎥⎦
(c) Following a procedure similar to that outlined in Parts (a) and (b) above, we can
show that here Yl = H 0 X l + H1X l −1 + H 2 X l − 2 , where

⎡ y[ 4l] ⎤
⎢ y[ 4l + 1] ⎥
⎥,
Yl = ⎢
⎢ y[ 4l + 2]⎥

⎣ y[ 4l + 3]⎦
⎡h[ 4] h[3]
⎢ h[5] h[ 4]
H1 = ⎢
⎢ 0
h[5]

0
⎣ 0

⎡h[0]
⎡ x[ 4l] ⎤
⎢ h[1]
⎢ x[ 4l + 1] ⎥

, H0 = ⎢
Xl =
⎢h[2]
⎢ x[ 4l + 2]⎥

⎣ h[3]
⎣ x[ 4l + 3]⎦
h[2] h[1] ⎤
⎡0 0 0
⎢0 0 0

h[3] h[2]⎥
, H2 = ⎢
⎢0 0 0
h[ 4] h[3]⎥

h[5] h[ 4]⎦
⎣0 0 0

0
h[0]
h[1]
h[2]
h[5]⎤
0 ⎥⎥
.
0 ⎥

0 ⎦

0
0 ⎤
0
0 ⎥⎥
,
h[0] 0 ⎥

h[1] h[0]⎦

8.50 (a) d 0 y[2l] + d1 y[2l − 1] + d 2 y[2l − 2] + d3 y[2l − 3] + d 4 y[2l − 4]
= p0 x[2l] + p1 x[2l − 1] + p2 x[2l − 2] + p3 x[2l − 3] + p4 x[2l − 4],
d 0 y[2l + 1] + d1 y[2l] + d 2 y[2l − 1] + d3 y[2l − 2] + d 4 y[2l − 3]
= p0 x[2l + 1] + p1 x[2l] + p2 x[2l − 1] + p3 x[2l − 2] + p 4 x[2l − 3].
Rewriting the above equations in matrix form we arrive at
⎡ d 0 0 ⎤ ⎡ y[2l] ⎤ ⎡ d 2 d1 ⎤ ⎡ y[2l − 2]⎤ ⎡ d 4 d3 ⎤ ⎡ y[2l − 4]⎤
⎥⎢
⎢d d ⎥ ⎢
⎥ + ⎢d
⎥ + ⎢ 0 d ⎥ ⎢ y[2l − 3]⎥
d
[
2
1
]
[
2
1
]
y
y
l
l
+

⎦ ⎣

1
0
3
2
4⎦⎣

0 ⎤ ⎡ x[2l] ⎤ ⎡ p2 p1 ⎤ ⎡ x[2l − 2]⎤ ⎡ p 4
⎡p
=⎢ 0
⎥⎢
⎥⎢
⎥+⎢
⎥+⎢
⎣ p1 p0 ⎦ ⎣ x[2l + 1]⎦ ⎣ p3 p2 ⎦ ⎣ x[2l − 1] ⎦ ⎣ 0
The above matrix equation can be alternately expressed as

p3 ⎤ ⎡ x[2l − 4]⎤
.
p 4 ⎥⎦ ⎢⎣ x[2l − 3]⎥⎦

⎡ y[2l] ⎤
D 0Yl + D1Yl −1 + D 2Yl − 2 = P 0 X l + P1 X l −1 + P 2 X l − 2 , where Yl = ⎢
⎥,
⎣ y[2l + 1]⎦
d1 ⎤
0⎤
d3 ⎤
⎡d
⎡d
⎡d
⎡ x[2l] ⎤
Xl = ⎢
, D0 = ⎢ 0
, D1 = ⎢ 2
, D2 = ⎢ 4

⎥,

⎣ x[2l + 1]⎦
⎣ 0 d4 ⎦
⎣ d1 d 0 ⎦
⎣ d3 d2 ⎦

⎡p
P0 = ⎢ 0
⎣ p1
⎡d0
(b) ⎢⎢ d1
⎢⎣ d 2

0⎤
⎡p
, P1 = ⎢ 2

p0 ⎦
⎣ p3
0
d0
d1

p1 ⎤
⎡p
, and P2 = ⎢ 4

p2 ⎦
⎣0

0 ⎤ ⎡ y[3l] ⎤ ⎡ d3
0 ⎥⎥ ⎢⎢ y[3l + 1] ⎥⎥ + ⎢⎢ d 4
d 0 ⎥⎦ ⎢⎣ y[3l + 2]⎥⎦ ⎢⎣ 0

d2
d3
d4

p3 ⎤
.
p4 ⎥⎦

d1 ⎤ ⎡ y[3l] − 3⎤ ⎡0 0 d 4 ⎤ ⎡ y[3l − 6]⎤
d 2 ⎥⎥ ⎢⎢ y[3l − 2]⎥⎥ + ⎢⎢0 0 0 ⎥⎥ ⎢⎢ y[3l − 5]⎥⎥
d3 ⎥⎦ ⎢⎣ y[3l − 1] ⎥⎦ ⎢⎣0 0 0 ⎥⎦ ⎢⎣ y[3l − 4]⎥⎦

Not for sale

279

0 ⎤ ⎡ x[3l] ⎤ ⎡ p3
⎡ p0 0

= ⎢ p1 p0 0 ⎥⎥ ⎢⎢ x[3l + 1] ⎥⎥ + ⎢⎢ p 4
⎢⎣ p2 p1 p0 ⎥⎦ ⎢⎣ x[3l + 2]⎥⎦ ⎢⎣ 0
which can be alternately expressed as
D 0 Yl + D1Yl −1 + D 2 Yl − 2

p2
p3
p4

p1 ⎤ ⎡ x[3l] − 3⎤ ⎡0 0
p2 ⎥⎥ ⎢⎢ x[3l − 2]⎥⎥ + ⎢⎢0 0
p3 ⎥⎦ ⎢⎣ x[3l − 1] ⎥⎦ ⎢⎣0 0

p4 ⎤ ⎡ x[3l − 6]⎤
0 ⎥⎥ ⎢⎢ x[3l − 5]⎥⎥ ,
0 ⎥⎦ ⎢⎣ x[3l − 4]⎥⎦

⎡ y[3l] ⎤
= P0 X l + P1X l −1 + P2 X l − 2 , where D 0 = ⎢⎢ y[3l + 1] ⎥⎥,
⎢⎣ y[3l + 2]⎥⎦
0
0⎤
⎡ d3 d 2 d1 ⎤
⎡0 0 d 4 ⎤

d 0 0 ⎥, D1 = ⎢ d 4 d3 d 2 ⎥, D 2 = ⎢⎢0 0 0 ⎥⎥,
⎢⎣ 0 d 4 d3 ⎥⎦
⎢⎣0 0 0 ⎥⎦
d1 d 0 ⎥⎦

⎡ x[3l] ⎤
⎡d0

P 0 = ⎢ x[3l + 1] ⎥, D 0 = ⎢⎢ d1
⎢⎣ x[3l + 2]⎥⎦
⎢⎣ d 2
0⎤
⎡ p0 0
⎡ p3

P0 = ⎢ p1 p0 0 ⎥, P1 = ⎢⎢ p 4
⎢⎣ p2 p1 p0 ⎥⎦
⎢⎣ 0

p2
p3
p4

p1 ⎤
⎡0 0

p2 ⎥, and P 2 = ⎢⎢0 0
⎢⎣0 0
p3 ⎥⎦

p4 ⎤
0 ⎥⎥.
0 ⎥⎦

0
0 ⎤ ⎡ y[ 4l] ⎤ ⎡ d 4 d3 d 2 d1 ⎤ ⎡ y[ 4l − 4]⎤
⎡d0 0
⎢d d
0
0 ⎥⎥ ⎢⎢ y[ 4l + 1] ⎥⎥ ⎢⎢ 0 d 4 d3 d 2 ⎥⎥ ⎢⎢ y[ 4l − 3]⎥⎥
1
0

+
(c)
⎢ d 2 d1 d 0 0 ⎥ ⎢ y[ 4l + 2]⎥ ⎢ 0
0 d 4 d3 ⎥ ⎢ y[ 4l − 2]⎥

⎥⎢
⎥⎢
⎥ ⎢

0
0 d 4 ⎦ ⎣ y[ 4l − 1] ⎦
⎣ d3 d 2 d1 d 0 ⎦ ⎣ y[ 4l + 3]⎦ ⎣ 0
0
0 ⎤ ⎡ x[ 4l] ⎤ ⎡ p 4 p3 p2 p1 ⎤ ⎡ x[ 4l − 4]⎤
⎡ p0 0
⎢p
p0 0
0 ⎥⎥ ⎢⎢ x[ 4l + 1] ⎥⎥ ⎢⎢ 0 p 4 p3 p2 ⎥⎥ ⎢⎢ x[ 4l − 3]⎥⎥
=⎢ 1
+
, which can be
⎢ p2 p1 p0 0 ⎥ ⎢ x[ 4l + 2]⎥ ⎢ 0
0 p 4 p3 ⎥ ⎢ x[ 4l − 2]⎥

⎥⎢
⎥⎢
⎥ ⎢

0
0 p4 ⎦ ⎣ x[ 4l − 1] ⎦
⎣ p3 p2 p1 p0 ⎦ ⎣ x[ 4l + 3]⎦ ⎣ 0
alternately written as D 0 Yl + D1Yl −1 + D 2 Yl − 2 = P0 X l + P1X l −1 + P2 X l − 2 , where

⎡d0
⎡ y[ 4l] ⎤
⎡ x[ 4l] ⎤
⎢d
⎢ y[ 4l + 1] ⎥
⎢ x[ 4l + 1] ⎥
⎥, X l = ⎢
⎥, D 0 = ⎢ 1
Yl = ⎢
⎢d2
⎢ y[ 4l + 2]⎥
⎢ x[ 4l + 2]⎥

⎣ y[ 4l + 3]⎦
⎣ x[ 4l + 3]⎦
⎣ d3
0
⎡ p0 0
⎡ d 4 d3 d 2 d1 ⎤

⎢0 d

p
p0 0
4 d3 d 2 ⎥
D1 = ⎢
, P0 = ⎢ 1
⎢ p2 p1 p0
⎢0
0 d 4 d3 ⎥

0
0 d4 ⎦
⎣0
⎣ p3 p2 p1

0
d0
d1
d2

0
0
d0
d1

0⎤
0 ⎥⎥
,
0⎥

d0 ⎦

0⎤
⎡ p4

⎢0
0⎥
, and P 1 = ⎢
⎢0
0⎥

p0 ⎦
⎣0

p3
p4
0
0

p2
p3
p4
0

p1 ⎤
p2 ⎥⎥
.
p3 ⎥

p4 ⎦

8.51 We first rewrite the second-order block difference equation
D 0 Yl + D1Yl −1 + D 2 Yl − 2 = P0 X l + P1X l −1 + P2 X l − 2 ,
as two separate equations: Wl = P0 X l + P1X l −1 + P2 X l − 2 , and

Not for sale

280

Yl = − D 0−1D1Yl −1 − D 0−1D 2 Yl − 2 + D 0−1Wl . As cascade realization of the IIR block

digital filter based on the last two equations is thus as shown below:
Wk

Xk

_1

D0

P0

Yk

D1

P1

D2

P2

By interchanging the locations of the two block sections in the above structure we arrive at
an equivalent realization as indicated below:
_1

Xk

D0

P0

Yk

D1

P1

P2

D2

Finally, by delay-sharing the above structure reduces to a canonic realization as shown
below:
Xk

_1

D0

P0

Yk

D1

P1

D2

P2

8.51 By setting α = ± β sin θ in Eq. (8.125), the state-space description of the sine-cosine

Not for sale

281

⎡ s [ n + 1] ⎤ ⎡cos ωo
generator reduces to ⎢ 1
⎥=⎢
⎣ s 2 [ n + 1]⎦ ⎢⎣ m
multiplier realization shown below:
cos ω o
s 2[n+1]

cos ω o
+
_ sin 2ω o
_1

z

8.53 Let −

± sin 2 ωo ⎤ ⎡ s 1[ n] ⎤
⎥⎢
⎥, which leads to the threecos ωo ⎥⎦ ⎣ s 2 [ n]⎦

s 2[n]

s 1[n+1]

_1

z

s 1[n]

_
+

− sin 2 ωo cos 2 ωo − 1
β sin ωo
α
=
= cos ωo − 1, and
= 1 + cos ωo . Then sin ωo =
α
β
cos ωo + 1 cos ωo + 1

β
sin ωo = cos ωo + 1. Substituting these values in Eq. (8.125) we arrive at
α
cos ωo − 1⎤ ⎡ s 1[ n] ⎤
⎡ s1[ n + 1] ⎤ ⎡ cos ωo
⎢ s [ n + 1]⎥ = ⎢cos ω + 1 cos ω ⎥ ⎢ s [ n]⎥. These equations can be alternately written
o
o ⎦⎣ 2
⎣ 2
⎦ ⎣

as s1[ n + 1] = cos ωo (s1[ n] + s 2 [ n]) − s 2 [ n], and s 2 [ n + 1] = cos ωo (s1[ n] + s 2 [ n]) + s 2 [ n].
A realization based on the last two equations results in a single-multiplier structure as
indicated below:

s 2[n+1]

_1

z

s 2[n]

_1

s 1[n+1]

_

z1

s 1[n]

cos ω o

⎡ s1[ n + 1] ⎤ ⎢0
8.54 ⎢
⎥=
⎣ s 2 [ n + 1]⎦ ⎢0

α(C − cos ωo ) ⎤
C
[
1
]
+
s
n
⎡ 1
⎤ ⎢

+⎢
β sin ωo
⎥ ⎢⎣ s 2 [ n + 1]⎥⎦ ⎢ β
0
− sin ωo

⎣⎢ α

α(1 − C cos ωo ) ⎤
⎥ ⎡ s1[ n] ⎤
β sin ωo
. If
⎥⎢
s 2 [ n]⎥⎦

cos ωo
⎦⎥

C = 1, then
⎡ s1[ n + 1] ⎤
⎢ s [ n + 1]⎥
⎣ 2

α(1 − cos ωo ) ⎤
1
⎡ s1[ n + 1] ⎤ ⎢

+⎢
β sin ωo
⎥ ⎢⎣ s 2 [ n + 1]⎥⎦ ⎢ β
0
− sin ωo

⎢⎣ α
β sin ω o . Then the above equation reduces to

0
=⎢

⎣0

Choose α =

α(1 − 1 cos ωo ) ⎤
β sin ωo ⎥ ⎡ s1[ n] ⎤.
⎥⎢

⎥ ⎣ s 2 [ n]⎦
cos ωo
⎦⎥

⎡ s1 [ n + 1] ⎤ ⎡0 1 − cos ωo ⎤ ⎡ s1 [ n + 1] ⎤ ⎡ 1 1 − cos ωo ⎤ ⎡ s1 [ n] ⎤
⎢ s [ n + 1]⎥ = ⎢
⎥ ⎢ s [ n + 1]⎥ + ⎢ − 1 cos ω ⎥ ⎢ s [ n]⎥.
0
⎦⎣ 2
o ⎦⎣ 2
⎣ 2
⎦ ⎣0
⎦ ⎣

A two-multiplier realization of the above equation is shown below:
Not for sale

282

_1

1 _ cos ω o
_

_

z1

s 2[n+1]
cos ω o

s 2[n]

s 1[n+1]

z1

s 1[n]

To arrive at a one-multiplier realization we observe that the two equations describing the
sine-cosine generator are given by
s1[n + 1] = (1 − cos ωo ) s2 [n + 1] + s1[n] + (1 − cos ωo ) s2 [n], and
s2 [n + 1] = − s1[n] + cosωo ⋅s2 [n]. Substituting the second equation in the first we arrive at
an alternate description in the form s1[n + 1] = − cos ωo ⋅ s2 [n + 1] + s2 [n], and
s2 [n + 1] = − s1[n] + cosωo ⋅s2 [n]. A realization of these two equations leads to the singlemultiplier structure shown below:
_1

s 2[n+1]

_

z1

s 2[n]

cos ω o

s 1[n+1]

_

z1

s 1[n]

_1

_1

z

8.55 From Figure P8.18(a), the input-output relation of the channel is given
H12 ( z )⎤ ⎡ X1 ( z ) ⎤
⎡Y (z) ⎤ ⎡ 1
. Likewise, the input-output relation of the
by ⎢ 1 ⎥ = ⎢
1 ⎥⎦ ⎢⎣ X 2 ( z )⎥⎦
⎣Y2 ( z )⎦ ⎣ H 21 ( z)
channel separation circuit of Figure P8.18(b) is given by
1
− G12 ( z )⎤ ⎡ Y1 ( z) ⎤
⎡ V1 ( z ) ⎤ ⎡
⎢V ( z )⎥ = ⎢ − G ( z )
⎥ ⎢Y ( z)⎥. Hence, the overall system is characterized by
1
21
⎣ 2 ⎦ ⎣
⎦⎣ 2 ⎦

H12 ( z)⎤ ⎡ X1 ( z) ⎤
1
− G12 ( z )⎤ ⎡ 1
⎡ V1 ( z ) ⎤ ⎡
⎢V ( z )⎥ = ⎢ − G ( z)

1
1 ⎥⎦ ⎢⎣ X 2 ( z)⎥⎦
21
⎣ 2 ⎦ ⎣
⎦ ⎣ H 21 ( z)
⎡1 − H 21 ( z)G12 ( z) H12 ( z ) − G12 ( z ) ⎤ ⎡ X1 ( z ) ⎤
=⎢
⎥⎢
⎥.
⎣ H 21 ( z ) − G21 ( z) 1 − H12 ( z)G21 ( z )⎦ ⎣ X 2 ( z )⎦
The cross-talk is eliminated if V1 ( z) is a function of either X1 ( z) or X 2 ( z), and similarly,
if V2 ( z) is a function of either X1 ( z) or X 2 ( z). From the above equation it follows that if
H 12 ( z) = G12 ( z), and H 21( z) = G 21( z), then V1 ( z) = (1 − H 21 ( z)G12 ( z))X1 ( z), and

−1 ( z ), and G ( z ) = H −1 ( z ),
V2 ( z) = (1 − H12 ( z)G21 ( z))X 2 ( z). Alternately, if G12 ( z ) = H 21
21
12

Not for sale

283

⎛ H ( z )H 21 ( z) − 1 ⎞
⎛ H ( z )H 21 ( z) − 1 ⎞
⎟⎟ X 2 ( z), and V2 ( z ) = ⎜⎜ 12
⎟⎟ X1 ( z).
then V1 ( z ) = ⎜⎜ 12
H 21 ( z)
H12 ( z)

M8.1 (a) Using the M-file factor, we factorize H1 ( z) resulting in

H1 ( z) = −0.24(1 − 2.5z −1 )(1 − 0.4 z −1 )(1 + z −1 + z − 2 )(1 + 0.8z −1 + 0.6 z − 2 ).
4
3

6
3

A cascade realization of H1 ( z) based on the above factored form is shown below:
_ 0.24

x[n]

+
z

z

z

z

_ 2.5

_1

_1

_1
_ 0.5

y[n]

+

+

+

_1

4/3

0.8

+

+

_1

_1

z

z

6/3

0.6

(b) Using the M-file factor, we factorize H 2 ( z) resulting in
H 2 ( z ) = 4(1 − 3.8508 z −1 )(1 + 2.5z −1 )(1 − 1.5403 z −1 )(1 + 0.4 z −1 )(1 − 0.649 z −1 )(1 − 0.2597 z −1 ).
A cascade realization of H 2 ( z) based on the above factored form is shown below:
4

x[n]

+
_1

+
_1

z

z
_

+

1.5403

z
_

0.4

y[n]

_1

z

z
_

2.5

3.8508

_1

_1

z

+

+

+

_1

0.649

_

0.2597

(c) Using the M-file factor, we factorize H 3 ( z) resulting in
H 3 ( z ) = −0.24(1 + z −1 )(1 − z −1 )(1 − 1.4024z −1 )(1 − 0.71306z −1 )(1 + 1.3488z −1 + z −2 ).

A cascade realization of H 3 ( z) based on the above factored form is shown below:
_ 0.24

+

x[n]
_1

z

+
_1

+
_1

z

z
_1

y[n]

_1

z

z
_ 1.4024

+

+
_1
_ 0.71306

1.3488

+
_1

z

(d) Using the M-file factor, we factorize H 4 ( z) resulting in
H 4 ( z ) = 4(1 + z −1 )(1 − z −1 )(1 − 4.6735 z −1 )(1 − 0.214 z −1 )(1 + 1.4675 z −1 + z −2 ).
A cascade realization of H 4 ( z) based on the above factored form is shown below:

Not for sale

284

4

+

x[n]
_1

+
_1

z

+
_1

z

z

_1

z

z

_1

y[n]

+

+
_1
_ 0.214

_ 4.6735

1.4675

+
_1

z

M8.2 (a) Using the M-file factor, we factorize the numerator and the denominator of G (z )

resulting in G ( z ) =

0.3901(1 + 1.4891z −1 + z −2 )(1 + 0.1582 z −1 + z −2 )
(1 − 0.06147 z −1 + 0.70426 z − 2 )(1 + 0.5653 z −1 + 0.2228 z − 2 )

.

(b) We can write

⎞⎛ 1 + 0.1582 z −1 + z − 2
1 + 1.4891z −1 + z − 2
⎟.

⎟⎜
G ( z ) = 0.3901
⎜ 1 − 0.06147 z −1 + 0.70426 z − 2 ⎟⎜ 1 + 0.5653 z −1 + 0.2228 z − 2 ⎟

⎠⎝
cascade realization of G (z ) based on the above decomposition is shown on next page:

A

0.3901

+

+

x[n]

+

_ 0.56527

1.4891

0.1582

+

+

+
_
z 1

_
z 1
_

y[n]

_
z 1

_
z 1

0.06147

+

+

_ 0.2228

0.70426

Alternately, we can write

⎞⎛
1 + 0.1582 z −1 + z − 2
1 + 1.4891z −1 + z − 2
⎟.
⎟⎜
G ( z ) = 0.3901⎜
⎜ 1 − 0.06147 z −1 + 0.70426 z − 2 ⎟⎜ 1 + 0.5653 z −1 + 0.2228 z − 2 ⎟

⎠⎝
A cascade realization of G (z ) based on the above decomposition is shown below:
0.3901
x[n]

+

+

+

0.1582

+

_ 0.56527

0.70426

1.4891

+

+
_
z 1

_
z 1
_

y[n]

_
z 1

_
z 1

0.06147

+

+

_ 0.2228

(c) A partial-fraction expansion of G (z ) in z −1 obtained using the M-file residuez is

given by G ( z ) = 2.4863 +

− 0.3481 − 0.11128 z −1
1 − 0.06147 z −1 + 0.70425 z − 2

Not for sale

+

− 1.7481 + 0.4094 z −1
1 + 0.5653 z −1 + 0.2228 z − 2

285

.

A Parallel Form I realization of G (z ) based on the above expansion is shown on the left
side top of next page.
(d) A partial-fraction expansion of G (z ) in z obtained using the M-file residue is given

− 0.1327 z −1 + 0.2451z −1

by G ( z ) = 0.3901 +

+

0.5787 z −1 + 0.3895 z −2

. A
1 − 0.06147 z −1 + 0.70425 z − 2 1 + 0.5653 z −1 + 0.2228 z − 2
Parallel Form II realization of G (z ) based on the above expansion is shown on the right
side top of next page.
M8.3 (a) Using the M-file factor, we factorize the numerator and the denominator of H (z )

resulting in H ( z ) =

0.3549(1 + 0.5958 z −1 + z −2 )(1 − 0.03169 z −1 + z −2 )
(1 + 0.6209 z −1 + 0.9845 z − 2 )(1 + 0.6313 z −1 + 0.5681z − 2 )

.

2.4863
0.3901

_ 0.3481

x[n]

+

+

+

y[n]
x[n]

_
z 1

+

+
_
z 1

0.06147
_1

_ 0.70425

_ 0.11128

+

z

_

_

+

y[n]

_
z 1

0.70425

0.2451

+

_1

_
z 1

z
_ 0.5653

_ 0.5653

+

+

1.7481

+

+

_ 0.1327

0.06147

0.4094

0.5787

+

_1

+

_1

z

z

_ 0.2228

_ 0.2228

0.3895

(b) We can write

⎞⎛ 1 − 0.03169 z −1 + z − 2 ⎞
1 + 0.5958 z −1 + z − 2
⎟.
⎟⎜
H ( z ) = 0.3549⎜
⎜ 1 + 0.620986 z −1 + 0.9845 z − 2 ⎟⎜ 1 + 0.6313 z −1 + 0.5683 z − 2 ⎟

⎠⎝
A cascade realization of H (z ) based on the above decomposition is shown on top of the
next page.
Alternately, we can write

⎛ 1 − 0.03169 z −1 + z − 2 ⎞⎛ 1 + 0.5958 z −1 + z − 2
⎟.
⎟⎜
H ( z ) = 0.3549⎜
⎜ 1 + 0.6209 z −1 + 0.9845 z − 2 ⎟⎜ 1 + 0.6313 z −1 + 0.5683 z − 2 ⎟

⎠⎝

Not for sale

286

0.3549

+

+

x[n]

+

+
_
z 1

_
z 1
_

+

y[n]

0.5958

0.6209

+

+

_ 0.03169

_ 0.6313

+
_
z 1

_
z 1
_ 0.9845

_ 0.5683

A cascade realization of H (z ) based on the decomposition given at the bottom of the
previous page is shown below:
0.3549

+

+

x[n]

+

+
_
z 1

_
z 1
_

+

_ 0.03169

0.6209

y[n]

+

+

_ 0.6313

0.5958

+
_
z 1

_
z 1
_ 0.9845

_ 0.5683

(c)
A partial-fraction expansion of G (z ) in z −1 obtained using the M-file
residuez is given by

− 0.0133 + 0.0094 z −1

H ( z ) = 0.6343 +

− 0.2661 − 0.4299 z −1

+
. A Parallel
1 + 0.6209 z −1 + 0.9845 z − 2 1 + 0.6313 z −1 + 0.5683 z − 2
Form I realization of G (z ) based on the above expansion is shown below on the left side.
0.6343

x[n]

_ 0.0133

+

0.3549

+

+

y[n]

x[n]

+

_1

_
z 1

z
_ 0.6209

_ 0.6209

0.0094

+

_
z 1
_

_

_ 0.2661

+

0.9845

y[n]

0.0131

_
z 1

z

_ 0.6313

_ 0.4299

+

+

+

_1

_ 0.6313

+

_
z 1

0.9845

+

0.0177

+

_ 0.2619

+

+

_1

_1

z

z
_ 0.5683

_ 0.5683

Not for sale

0.1512

287

(d) A partial-fraction expansion of G (z ) in z obtained using the M-file residue is given

− 0.0177 z −1 + 0.0131z −1

by G ( z ) = 0.3549 +

+

− 0.2619 z −1 + 0.1512 z −2

. A Parallel
1 + 0.6209 z −1 + 0.9845 z − 2 1 + 0.6313 z −1 + 0.5683 z − 2
Form II realization of G (z ) based on the above expansion is shown at the bottom of the
previous page on the right side.
M8.4

Lattice parameters are
0.1393
0.6973
0.3131

0.1569

Feedforward multipliers are
0.3901
0.4461
0.3212

0.0727

-0.0488

The Gray-Markel tapped cascaded lattice realization of G (z ) is shown on next page:
x[n]
_ 0.1569

_ 0.3131

0.3131

0.1569
_1

M8.5

0.1393

0.6973
_1

z

0.3901

_ 0.1393

_ 0.6973

0.4461

_1

_1

z

z

z

0.3212

0.0727

_ 0.0488

y[n]

Lattice parameters are
0.3171
0.9949
0.3986

0.5595

Feedforward multipliers are
03549
-0.2442
0.2642

-0.0082

-0.0066

The Gray-Markel tapped cascaded lattice realization of H (z ) is shown below:
x[n]
_ 0.5595

_ 0.3986

0.3986

0.5595
_1

0.3549

_ 0.9949

_ 0.3171

0.3171

0.9949
_1

z

z

_ 0.2442

0.2642

_1

z

_ 0.0082

_1

z

_ 0.0066

y[n]

M8.6 (a) (b) Program 8_7 generated the following error messages for the FIR transfer functions
of Parts (a) and (b):
??? Error using ==> signal\private\levdown
At least one of the reflection coefficients is equal to one.
The algorithm fails for this case.
(c) (d) Program 8_7 generated the following error messages for the FIR transfer functions
of Parts (c) and (d):
Warning: Divide by zero.
Lattice coefficients are
Not for sale

288

-1 NaN NaN NaN NaN NaN
M8.7 (a) Using the M-file tf2ca we arrive at the parallel allpass decomposition of G (z ) as:

G( z) =

1
2

A1 ( z ) =

[A0 ( z ) + A1 ( z )],

where A0 ( z ) =

0.8041 − 0.8171z −1 + z −2

− 0.2440 + 1.2092 z −1 − 1.1436 z −2 + z −3
1 − 1.1436 z −1 + 1.2092 z − 2 − 0.2440 z − 3

.

1 − 0.8171z −1 + 0.8041z − 2

(b) The power-complementary transfer function is then given by H ( z ) =

=

0.524(1 − 3.0011z −1 + 4.9948 z −2 − 4.9948 z −3 + 3.0011z −4 − z −5

1 − 1.9606 z −1 + 2.9473 z − 2 − 2.1511z − 3 + 1.1713 z − 4 − 0.1961z − 5
(c)
jω 2

Magnitude Square

and

1
2

[A0 ( z ) − A1 ( z )]

.

jω 2

|G(e )| +|H(e )|

1
jω 2

0.8

jω 2

|G(e )|

|H(e )|

0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/π

M8.8 (a) Using the M-file tf2ca we arrive at the parallel allpass decomposition of G (z ) as:

G( z) =

1
2

[A0 (z) − A1 (z)], where

A0 ( z ) =

0.9077 + 0.6205z −1 + z −2

1 + 0.6205z −1 + 0.9077z − 2
0.3325 + 1.2027z −1 + 0.9522 z −2 + z −3
A1 ( z) =
.
1 + 0.9522 z −1 + 1.2027z − 2 + 0.3325z − 3

and

(b) The power-complementary transfer function is then given by H ( z) =

=

(1 + 2.3334 z −1 + 3.7759 z −2 + 3.7759 z −3 + 2.3334 z −4 + z −5
1 + 1.5727 z −1 + 2.712 z − 2 + 1.9431z − 3 + 1.2979 z − 4 +0.3018 z − 5

1
2

[A0 (z) + A1 (z)]

.

(c)

Not for sale

289

|G(e ω | +|H(e ω |

Magnitude Square

j 2

1

|H(e ω |

|G(e ω |

j 2

0.8

j 2
j 2

0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/π

M8.9 The transfer function given cannot be realized as a parallel connection of two real allpass
sections as the degree difference between the two allpass transfer functions must be 2.
M8.10 The MATLAB program for the simulation of the sine-cosine generator of Problem 8.53
is given below:
s10 = 0.1; s20 = 0.1; a = 0.8;
y1 = zeros(1,50); y2 = y1;
for n = 1:50;
y1(n) = a*(s10 + s20) - s20;
y2(n) = a*(s10 +s20) + s10;
s10 = y1(n);
s20 = y2(n);
end
y1max = max(y1);y2max = max(y2);
k = 1:50;
stem (k-1,y1/y1max);
xlabel('Time index n');ylabel('Amplitude');
title('y_1[n]');
pause
stem(k-1,y2/y2max);
xlabel('Time index n');ylabel('Amplitude');
title('y_2[n]');

The plots generated by the above program for initial conditions s1[-1] s2[-1] =
0.1 are shown below:

Not for sale

290

y 2[n]
1

0.5

0.5
Amplitude

Amplitude

y 1[n]
1

0
-0.5
-1

0
-0.5

0

10

20
30
Time index n

40

50

-1

0

10

20
30
Time index n

40

50

The outputs are zero for zero initial conditions. Non-zero initial conditions of equal values
appear to have no effects on the outputs. However, unequal initial conditions have effects
on the amplitudes and phases of the two output sequences.
M8.11 The MATLAB program for the simulation of the sine-cosine generator of Problem 8.53
is given below:
s10 = 0.1; s20 = 1; a = 0.8;
y1 = zeros(1,50); y2 = y1;
for n = 1:50;
y1(n) = -s20 +a*s10;
y2(n) = -a*y1(n) + s10;
s10 = y1(n); s20 = y2(n);
end
y1max = max(y1);y2max = max(y2);
k = 1:50;
stem (k-1,y1/y1max);
xlabel('Time index n');ylabel('Amplitude');
title('y_1[n]');
pause
stem(k-1,y2/y2max);
xlabel('Time index n');ylabel('Amplitude');
title('y_2[n]');

The plots generated by the above program for initial conditions s1[-1] s2[-1] =
0.1 are shown on top of the next page.:
The outputs are zero for zero initial conditions. Non-zero initial conditions of equal values
appear to have no effects on the outputs. However, unequal initial conditions have effects
on the amplitudes and phases of the two output sequences.

Not for sale

291

y 1[n]

y 2[n]

1

1
0.5
Amplitude

Amplitude

0.5
0
-0.5
-1

0
-0.5

0

10

20
30
Time index n

40

50

Not for sale

-1

0

10

20
30
Time index n

40

50

292

Chapter 9
9.1

We obtain the solutions by using Eq. (9.3) and Eq. (9.4).
−α p / 20

= 1 − 10 − 0.21 / 20 = 0.0239, δ s = 10 − α s / 20 = 10 − 53 / 20 = 0.0022

−α p / 20

= 1 − 10 − 0.17 / 20 = 0.0194,

(a) δ p = 1 − 10
(b) δ p = 1 − 10

δ s = 10 −α s / 20 = 10 − 78 / 20 = 1.26 × 10 − 4 .

9.2

We obtain the solutions by using Eq. (9.3) and Eq. (9.4).
(a) α p = −20 log10 1 − δ p = −20 log10 (1 − 0.02) = 0.1755 dB,

(

)

(

)

α s = −20 log10 (δ s ) = −20 log10 (0.03) = 30.458 dB.

(b) α p = −20 log10 1 − δ p = −20 log10 (1 − 0.055) = 0.4914 dB,
α s = −20 log10 (δ s ) = −20 log10 (0.033) = 29.630 dB.

2

9.3

2 jω

2 jω
G( z ) = H 2 ( z ), or equivalently, G(e ) = H (e ). G(e ) = H (e ) = H (e ) .

Let δ p and δ s denote the passband and stopband ripples of H (e jω ) , respectively. Also,
let δ p,2 = 2δ p, and δ s,2 denote the passband and stopband ripples of G(e jω ) ,
respectively. Then δ p,2 = 1 − (1 − δ p ) 2 , and δ s,2 = (δ s ) 2 . For a cascade of M sections,
δ p, M = 1 − (1 − δ p ) M , and δ s,M = (δ s ) M .
9.4
HLP (e j )
HHP (e

1 p
1– p

1– p

1

0
– s

– p

)

p

s

s

j

š
p

s

–( – p ) –( – )
s

š

0
s

Therefore, the passband edge and the stopband edge of the highpass filter are given by
ω p,HP = π − ω p , and ω s,HP = π − ω s , respectively.
9.5

Note that G (z ) is a complex bandpass filter with a passband in the range 0 ≤ ω ≤ π . Its
passband edges are at ω p,BP = ωo ± ω p , and stopband edges at ω s, BP = ω o ± ω s . A real
coefficient bandpass transfer function can be generated according to

Not for sale

293

p

G BP ( z ) = H LP (e jωo z ) + H LP (e – jωo z ) which will have a passband in the range 0 ≤ ω ≤ π
and another passband in the range – π ≤ ω ≤ 0. However because of the overlap of the two
spectra a simple formula for the bandedges cannot be derived.
HLP (e

j

)

j

G(e )

1 p
1– p

1

1– p

s

s

0
– s

– p

š

0

s

p

o

9.6 (a) h p (t ) = ha (t ) ⋅ p(t ) where p(t ) =

n = −∞

h p (t)e stdt

p

o

p

o

s

ha (nT) (t nT) .

ha (t) e st dt and
nT)e st dt

h a (nT) (t

h a (nT)e snT .
n

g[n]z n

Comparing the above expression with G(z)
n

H p (s)

o

n

n

conclude that G(z)

š

o
s

∑ δ(t − nT ). Thus, h p (t)

ha (nT). Now, Ha (s)

We also have, g[n]
H p (s)

p

s

h a (nT)z n , we
n

.
1
ln z
T

We can also show that a Fourier series expansion of p(t) is given by
1
e j(2 kt / T) .
Therefore,
p(t)
Tk

h p (t)

1
Tk

e j (2 kt / T) h a (t)

H p (s)

1
Tk

Ha s

G(z)

j

1
Tk

ha (t) e j(2 kt / T) . Hence,

2 kt
. As a result, we have
T

1
Tk

Ha s

j

2 kt
.
1
T
s
ln z

(7-1)

T

e sT . If we express
e oT e j oT . Therefore,

(b) The transformation from the s -plane to z -plane is given by z

s

o

j o , then we can write z

re j

Not for sale

294

1, for o 1,
z
1, for o 1, Or in other words, a point in the left-half s -plane is mapped onto
1, for o 1.
a point inside the unit circle in the z -plane, a point in the right-half s -plane is mapped
onto a point outside the unit circle in the z -plane, and a point on the j -axis in the s plane is mapped onto a point on the unit circle in the z -plane. As a result, the mapping has
the desirable properties enumerated in Section 9.1.3.
(c) However, all points in the s -plane defined by s
mapped onto a single point in the z -plane as z
mapping is illustrated in the figure below

e

j o

o
oT

j(

e

o

j

2 k
, k
T

2 k
)T
T

j

e

0,,1,2,K , are

oT

e

j

oT

. The

Im z

T
š
T

1

1

Re z

– šT

– T
s -plane

z -plane

Note that the strip of width 2π / T in the s -plane for values of s in the range
/T
/ T is mapped into the entire z -plane, and so are the adjacent strips of
width 2π / T . The mapping is many-to-one with infinite number of such strips of width
2π / T .
It follows from the above figure and also from Eq. (7-1) that if the frequency response
1
j
, and there is no aliasing.
Ha (j ) 0 for
, then G(e )
H a ( j ), for
T
T
T

(d) For z
9.7

ej

e j T . Or equivalently,

T.

Assume ha (t ) is causal. Now, ha (t ) = H a (s )e st ds. Hence,

g[ n] = ha (nT ) = H a ( s ) e snT ds.

G( z ) =

n =0

g[ n]z −n =

∑∫

Therefore,

H a (s )e snT z −n ds = H a (s )

n =0

Not for sale

∑ z −n e snT ds = ∫ 1 − easT z −1 ds .
H (s )

n =0

295

Hence G( z ) =

⎡ H a (s) ⎤
Residues⎢
.
sT −1 ⎥

e
z
1

all poles of H a ( s )

A
. The transfer function has a pole at s = −α. Now
s+α

A
A
A
G( z) = Residue⎢
.
=
⎥=
sT

1
sT

1

α T −1
at s = – α ⎣⎢ (s + α)(1 − e z ) ⎦⎥ 1 − e z
1
e
z

s= – α

9.8 H a ( s ) =

9.9

(a) H a (s ) =

4(3s + 7)
2

(s + 2)(s + 4s + 5)

=

−4
2−6j
2+6j
+
+
s + 2 (s + 2 + j ) (s + 2 − j )

(− 12 ) ⋅ 12 .
4s − 4
4(s + 2 )
−4
−4
+
=
+
+
s + 2 (s + 2 )2 + 12 s + 2 (s + 2 )2 + 12 (s + 2 )2 + 12
Using Eqs. (9.58), (9.60), and (9.62), we get
=

G a (z ) =

−4
1− e

− 2T − 1

+

z

(

− 1 − 2T

1 − 2z e

−1 −2 T

z e

+

−4
1 − 0.5488 z −1
=

+

(

z

,

)

1 − 1.0486 z −1 + 0.3012 z −2
+

1 − 0.5488 z −1

(s

sin (T )

4 1 − 0.5243z −1

−4

(b) H b (s ) =

cos(T ) + e

− 4T − 2

cos(T ) + e − 4T z − 2

−1 − 2T

1 − 2z e
where T = 0.3. Therefore,

G a (z ) =

)

4 1 − z −1e − 2T cos(T )

+

4 − 1.935z −1
1 − 1.0486 z −1 + 0.3012 z −2

8s 2 + 37s + 56

)

=

0.1622 z −1
1 − 1.0486 z −1 + 0.3012 z − 2
.

2
3 + 0. 5 j
s − 0 .5 j
+
+
(s + 4 ) (s + 1 + 3 j ) (s + 1 − 3 j )

+ 2 s + 10 (s + 4 )
2
6s + 9
2
6(s + 1)
3(3)
=
+
=
+
.
+
2
2
2
s + 4 (s + 1) + 9 s + 4 (s + 1) + 3
(s + 1)2 + 32
Using Eqs. (9.58), (9.60), and (9.62), we get
Gb (z ) =

2

2
1 − e − 4T z − 1

+6

1 − 2 z −1e − T cos(3T ) + e − 2T z − 2

+3

where T = 0.3. Therefore,

Gb (z ) =

4
1 − 0.3012 z −1

+

1 − z −1e −T cos(3T )

z −1e −T sin (3T )

1 − 2 z −1e −T cos(3T ) + e − 2T z − 2

(

6 1 − 0.4605z −1

)

1 − 0.9210 z −1 + 0.5489z −2

Not for sale

+

,

(

3 0.5803z −1

)

1 − 0.9210 z −1 + 0.5489z −2

296

=

4
1 − 0.3012 z −1

(s + 2 )

1 − 0.9210 z −1 + 0.5489 z −2
4s + 1

− 3s + 2

(s 2 + 2s + 5)(s 2 + s + 4) (s 2 + 2s + 5) (s 2 + s + 4)

4(s + 2 )
2

6 − 1.0221z −1

s 3 + s 2 + 6 s + 14

(c) H c (s ) =
=

+

2

+1

+

− 7(1)

(s + 2 )

2

+

2

+1

=

+

− 3(s + 0.5)

(s + 0.5)2 + (

15 / 2

)

2

+

(s + 0.5)2 + (

Using Eqs. (9.58), (9.60), and (9.62), we get
Gc (z ) = (4 )

+ (− 3)

1 − e −2T cos(T )z −1

+ (− 7 )

1 − 2e − 2T cos(T )z −1 + e − 4T z − 2

(

)

1 − e −T / 2 cos 15T / 2 z −1

(

)

1 − 2e − 2T cos 15T / 2 z −1 + e − T z − 2

where T = 0.3. Hence,
Gc (z ) =

(

4 1 − 0.5243z −1

)

1 − 1.0486 z −1 + 0.3012 z −2

(

− 3 1 − 0.7195z −1

)

1 − 1.4390 z −1 + 0.7408z − 2
9.10 (a) G a (z ) =

3z
z − e −1.5

+

4z
z − e −1.8

=

+

(

15 / 2

1 − 2e − 2T cos(T )z −1 + e − 4T z − 2

+ 7 / 15

(

(

1 − 1.4390 z −1 + 0.7408z − 2

z − e − α1T

+

A2 z

z − e − α 2T

=

.

.
3
4
+
.
s + 7 .5 s + 9

ze −β T sin (λT )

(b) H b (s ) = Gb (z ) z = 4⎛ 1+ s ⎞ =

⎝ 1− s ⎠

2

1+ s ⎞⎤
⎡ 1+ s ⎞⎤ + 4

⎟⎥ + 3⎢4⎛⎜
s
1

⎠⎦
⎣ ⎝ 1− s ⎠⎥⎦
2

1+ s ⎞⎤
⎡ 1+ s ⎞⎤ + 6

⎟⎥ + 4 ⎢4⎛⎜
⎝ 1− s ⎠ ⎦
⎣ ⎝ 1− s ⎠⎥⎦

10 ⎡ 4⎛⎜

⎢⎣

54 ⎡ 4⎛⎜

⎢⎣

)

1 − 2e − 2T cos 15T / 2 z −1 + e − T z − 2

z 2 − 2 ze − β T cos(λT ) + e − 2β T
7.5
Since T = 0.2, λ = 7.5, β = 6, it follows H b (s ) =
(s + 6 )2 + 7.52

⎝ 1− s ⎠

)

e −T / 2 sin 15T / 2 z −1

)

0.8538z −1

z 2 − 2 ze −1.2 cos(1.5) + e − 2.4

⎢⎣

.

1 − 1.0486 z −1 + 0.3012 z −2

ze −1.2 sin (1.5)

9.11 (a) H a (s ) = G a (z ) z = 4⎛ 1+ s ⎞ =

)2

1.1353z −1

A1 z

4 ⎡4⎛⎜

)

e −2T sin (T )z −1

Since T = 0.2, α 1 = 7.5, α 2 9, A1 = 3, A2 = 4 , it follows H a (s ) =
(b) Gb (z ) =

(

7 / 15 15 / 2

3

=

.

(

16 s 2 + 3s + 4

).

75s 2 + 154 s + 91

2

1+ s ⎞⎤
⎡ 1+ s ⎞⎤ + 26 ⎡4⎛ 1+ s ⎞⎤ +18
⎟⎥ + 62 ⎢4⎛⎜

⎢⎣ ⎜⎝ 1− s ⎟⎠⎥⎦
⎝ 1− s ⎠⎦
⎣ ⎝ 1− s ⎠⎥⎦

2
⎛ 3⎡4⎛ 1+ s ⎞⎤ +1 ⎞⎛⎜ 12 ⎡4⎛ 1+ s ⎞⎤ − 4 ⎡4⎛ 1+ s ⎞⎤ + 8 ⎞⎟
⎜ ⎢ ⎜ 1− s ⎟⎥ ⎟⎜ ⎢ ⎜ 1− s ⎟⎥
⎢⎣ ⎜⎝ 1− s ⎟⎠⎥⎦ ⎟
⎠⎦ ⎠⎝ ⎣ ⎝
⎠⎦
⎝ ⎣ ⎝

Not for sale

297

,

=

(

− (5 + 3s ) 425s 2 + 846 s + 457

(

4(13s + 11) 27s 2 + 46 s + 23

9.12 For the impulse invariance design:

(

)

).

)(

)

ω p = Ω p T = 2 πFp T = 2π 0.45 × 10 3 0.3 × 10 − 3 = 0.27π
For the bilinear transformation method:
⎛ Ω pT ⎞
⎟ = 2 tan −1 Fp Tπ = 2 tan −1 0.45 × 10 3 ⋅ 0.3 × 10 − 3 ⋅ π = 0.2554 π
ω p = 2 tan −1 ⎜⎜

⎝ 2 ⎠
ωp
0.56 π
9.13 For the impulse invariance method: 2 πFp =
⇒ Fp = 1.4 kHz
=
T
0.2 × 10 −3
For the bilinear transformation method:
⎛ ωp ⎞ 1
1
⎛ 0.56 π ⎞
⎟⋅
Fp = tan⎜⎜
tan

= 1.923 kHz.
=

⎝ 2 ⎠ π 0.2 × 10 − 3
⎝ 2 ⎠ πT

(

)

(

(

)

)

9.14 The passband and the stopband edges of the analog lowpass filter are assumed to
Ω p = 0.25π and s 0.55 . The requirements to be satisfied by the analog
lowpass filter are thus 20log10 H a ( j 0.25π) ≥ −0.5 dB and

20log10 H a ( j 0.55π) ≤ −15 dB.
From α p = 20log10 ( 1 + ε 2 ) = 0.5 we obtain 2 0.1220184543. From
2
we obtain A 2 31.6227766 . From Eq. (4.32), the inverse
s 10 log10 (A ) 15

1
A2 − 1
=
= 15.841979 and from Eq. (4.31) the inverse
k1
ε
1 Ωs
= 2.2 . Substituting these values in Eq. (4.35) we
transition ratio is given by =
k Ωp

discrimination ratio is given by

obtain

N=

log10 (1 / k1 ) log10 (15.841979)
=
= 3.503885. We choose N = 4.
log10 (1 / k )
log10 (2.2)
2N

⎛Ωp ⎞

= ε 2 . Substituting the values of
From Eq. (4.33) we have ⎜⎜

⎝ Ωc ⎠
Ω c = 1.3007568(Ω p ) = 1.021612.

p,

N, and

2

we get

Using the statement [z,p,k] = buttap(4) we get the poles of the 4-th order
Butterworth analog filter with a 3-dB cutoff at 1 rad/s as p1 = −0.3827 + j 0.9239,
p2 = −0.3827 − j 0.9239, p3 = −0.9239 + j 0.3827, and p 4 = −0.9239 − j 0.3827.
Therefore,
Not for sale

298

H an (s ) =

1
1
=
.
(s − p1 )(s − p 2 )(s − p3 )(s − p 4 ) (s 2 + 0.7654s + 1)(s 2 + 1.8478s + 1)

Next we expand H an (s ) in a partial-fraction expansion using the M-file residue and
arrive at H an (s ) =

− 0.9238729 s − 0.7071323

+

0.9238729 s + 1.7071323

. We next
s 2 + 0.7654 s + 1
s 2 + 1.8478s + 1
denormalize H an (s ) to move the 3-dB cutoff frequency to c 1.021612 using the M-file
s

lp2lp resulting in H a ( s ) = H an ⎜
⎝ 1.021612 ⎠
0.943847s + 1.78174665
− 0.943847s − 0.738039
+
=
s 2 + 0.781947948s + 1.0437074244 s 2 + 1.887749436s + 1.0437074244
− 0.943847 s − 0.738039
0.943847s + 1.78174665
=
+
(s + 0.390974 ) 2 + (0.9438467) 2 (s + 0.94387471) 2 + (0.39090656 ) 2
Making use of the M-file bilinear we finally arrive at
G( z) =

− 0.943847 z 2 + 0.68178386 z
z 2 − 1.363567724 z + 0.4575139

9.15 The mapping is given by s =

s = σ o + jΩ o , z =

1
T

+

0.943847 z 2 − 0.25640047 z
z 2 − 0.77823439 z + 0.1514122

(1 − z −1 ) or equivalently, by z =

.

1
. For
1 − sT
1

1
2
. Therefore, z =
. Hence,
1 − σ o T − jΩ o T
(1 − σ o T ) 2 + (Ω o T ) 2

z < 1 for o

0. As a result, a stable H a (s ) results in a stable H (z ) after the
1
2
which is equal to 1 only for
transformation. However, for σ o = 0, z =
1 + (Ω o T ) 2
Ω o = 0. Hence, only the point Ω o = 0 on the jΩ -axis in the s -plane is mapped onto the
point z = 1 on the unit circle. Consequently, this mapping is not useful for the design of
digital filters via analog filter transformation.

9.16 For no aliasing T ≤

H1 (z)

π
. Figure below shows the magnitude responses of the digital filters
Ωc

and H 2 (z) .
j

H 2 (e j )

H1(e )

2
1

–š

0

š

Not for sale

–š

0

š

299

(a) The magnitude responses of the digital filters G1 ( z) and G2 ( z) are shown below:
G2 (e j )

G1(e j )

3/2
1
–š

0

š

–š

0

š

(b) As can be seen from the above G1 ( z) is a multi-band filter, whereas, G2 ( z) is a
highpass filter.
9.17 H a (s ) is causal and stable and H a (s )

s, Now, G( z) = H a (s ) s = 2 ⎛⎜ 1− z −1 ⎞⎟ . Thus,

1,

T ⎜ 1− z −1 ⎟

G (z ) is causal and stable. Now,

G ( e jω ) = H a ( s )

=
2 ⎛ 1− e jω ⎞⎟
s= ⎜
T ⎜ 1+ e jω ⎟

H a (s ) s = j 2 tan(ω / 2) = H a ( j
T

2
T

tan(ω / 2).

Therefore, G(e jω ) = H a ( j tan(ω / 2) ≤ 1 for all values of . Hence, G (z ) is a BR
T
2

function.
9.18 G ( z ) =

1+ α
1 − 2 βz − 1 + z − 2
. For β = cos ωo , the numerator of G (z ) becomes

2 1 − β (1 + α ) z −1 + αz − 2
1

2

j

1

j 0 1
z )

1 2cos 0 z
z
(1 e 0 z )(1 e
numerator of G(z N ) is then given by (1 e j
N

j

0z

N

which has roots at z

)(1 e

j 0 N
z )

e j

0.

The

whose roots are
j(2 n

)/ N

0
e
, and are given by z e
,
obtained by solving the equation z
N
0 n N 1. Hence G(z ) has N complex conjugate zero-pairs located on the unit
2 n
0 radians, 0 n N 1.
circle at angles of
N

0

0

/ 2, there are 2N equally spaced zeros on the unit circle starting at

9.19 (a) H ( z ) =

1
[1 +
2

/ 2N.

z −2 D1 ( z −1 ) z −2 D2 ( z −1 )
N ( z)
. We can write A4 ( z ) =

, where
A4 ( z )] =
D( z )
D1 ( z )
D2 ( z )

D1 ( z ) = 1 − β1 (1 + α1 )z −1 + α1 z −2 and D2 ( z ) = 1 − β2 (1 + α 2 )z −1 + α 2 z −2 . Therefore,
N ( z) =

1
2

[D (z)D (z) + z
1

2

−4

]

D1 ( z −1 ) D2 ( z −1 ) . Now,

Not for sale

300

z − 4 N ( z −1 ) =

1
2

[z

−4

]

D1 ( z −1 ) D2 ( z −1 ) + D1 ( z ) D2 ( z ) = N ( z ). Hence, N ( z ) is a symmetric

polynomial. It follows then
P( z) =

1
2

[(α1 − β1 (1 + α1 )z −1 + z − 2 )(α2 − β2 (1 + α2 )z −1 + z − 2 )

)(

(

+ 1 − β1 (1 + α1 )z −1 + α1 z − 2 1 − β2 (1 + α 2 )z −1 + zα 2 − 2

=

)]

1 + α 1α 2
[1 − (1 + α1 )(1 + α 2 )(β1 + β2 ) z −1 + 2[α1 + α 2 + β1β 2 (1 + α1 )(1 + α 2 )] z − 2
2
1 + α 1α 2
1 + α 1α 2
(1 + α1 )(1 + α 2 )(β1 + β 2 ) − 3

z + z − 4 ] = a(1 + b1 z −1 + b2 z −2 + b1 z −3 + z −3 ),
1 + α1α 2

(1 + α1 )(1 + α 2 )( β1 + β2 )
,
1 + α1α 2
2[α1 + α2 + β1 β2 (1 + α1 )(1 + α2 )]
,
b2 =
1 + α1α2
b1 = −

where

(b) a =

(7-a)
(7-b)

1 + α 1α 2
.
2

(7-c)

(c) for z = e jω , we can write N (e jω ) = a(1 + b1e − jω + b2 e − j 2ω + b1e − j 3ω + e − j 4ω )
= ae − j 2ω (b2 + 2b1 cos ω + 2 cos 2ω). Now, N (e jω ) = 0 for i = 1, 2. For i = 1, we get
b2 + 2b1 cos ω1 + 2 cos 2ω1 = 0,
(7-d)
for i = 2, we get

b2 + 2b1 cos ω2 + 2 cos 2ω2 = 0,

(7-e)

Solving Eqs. (7-d) and (7-e) we get b1 = −2(cos ω1 + cos ω2 )
b2 = 2(2 cos ω1 cos ω2 + 1).
and

(7-f)
(7-g)

(1 + α1 )(1 + α 2 )(β1 + β1 )
= 2(cos ω1 + cos ω2 ),
(7-h)
1 + α 1α 2
and from Eqs. (7-b) and (7-g) we have
2[α1 + α 2 + β1β 2 (1 + α1 )(1 + α 2 )]
= 2(2 cos ω1 cos ω2 + 1).
(7-i)
1 + α1α 2
1 − tan( B2 / 2)
1 − tan( B1 / 2)
, and after rearrangement we get
Substituting α1 =
and α 2 =
1 + tan( B2 / 2)
1 + tan( B1 / 2)
β + β = (cos ω + cos ω )[1 + tan( B / 2) tan( B / 2)] ∆ θ ,
(7-j)
From Eqs. (7-a) and (7-f) we have

1

2

1

2

Not for sale

1

2

= 1

301

β1β 2 = [1 + tan( B1 / 2) tan( B2 / 2)] cos ω1 cos ω2 ∆
= θ2 .

and

(7-k)

The above two nonlinear equations can be solved yielding
1

2
1

4 2
and 2

2.
2
1
(d) For the double notch filter with the following specifications: ω1 = 0.2π ,
ω2 = 0.6π , B1 = 0.2π, and B2 = 0.25π we get the following values for the
parameters of the notch filter transfer function:
α1 = 0.5095, α 2 = 0.4142, θ1 = 0.5673, θ 2 = −0.1491, β1 = 0.7628, and
1

β 2 = −0.1955 . H (z ) =

1
2

[1 + A4 (z )] :

1

2

Magnitude

0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

1
0
-1
-2

0

ω/ π

0.2

0.4

0.6

0.8

1

ω/ π

9.20 A zero (pole) of H LP (z) is given by the factor ( z − z k ). After applying the lowpass-tozˆ − α
− z k , and hence the new location of the
lowpass transformation, this factor becomes
1 − α zˆ
zero (pole) is given by the roots of the equation
α + zk
. For z k = −1,
zˆ − α − z k + α z k zˆ = (1 + α z k )zˆ − (α + z k ) = 0 or zˆ k =
1 + αzk
a −1
zˆ k =
= − 1.
1− a
9.21 The lowpass-to-bandpass transformation is given by z

b a ˆz zˆ 2
where a
1 a zˆ b zˆ 2

2
1

and

1
. A zero (pole) of H LP (z) is given by the factor ( z − z k ). . After applying the
1
b azˆ zˆ 2
zk , and hence, the
lowpass-to-bandpass transformation, this factor becomes
1 a zˆ b zˆ 2
b

new location of the zero (pole) of the bandpass transfer function is given by the roots of the

Not for sale

302

a(1 z k )
b zk
z
1 b zk
1 b zk

equation (1 b zk ) z2 a(1 zk ) z (b zk ) 0, or z 2
a(1 z k )
2(1 b z k )

solution is given by zˆ k

(

0.3404 1 + z −1

9.22 G LP (z ) =

a(1 z k )
2(1 b z k )

2

0, whose

b zk
. For z k
1 bz k

1, zˆ k

)2

, with ωc = 0.55π.
1 + 0.1842 z −1 + 0.1776 z −2
ˆ c = 0.42 π
H LP (z ) for ω
0.55π ⎞
0.42 π ⎞
0.55π − 0.42 π ⎞
tan⎛⎜
⎟ sin⎛⎜
⎟ − tan⎛⎜

2
⎝ 2 ⎠=
⎝ 2 ⎠
⎠ = 0.2030

α=
0
.
55
0
.
42
0
.
55
0
.
42
π
π
π
+
π
⎞⎟ sin⎛⎜
⎞⎟ + tan⎛⎜
⎞⎟
tan⎛⎜
2
⎝ 2 ⎠
⎝ 2 ⎠

H LP (z ) = G LP (z ) z −1 = zˆ −1 − α =
1− αzˆ −1

=

⎛ zˆ −1 − α ⎞

0.3404⎜⎜ 1+
−1 ⎟
ˆ
1

α
z

2

⎛ −1 ⎞
⎛ zˆ −1 − α ⎞
⎟⎟ + 0.1776⎜⎜ zˆ − α ⎟⎟
1 + 0.1842⎜⎜
⎝ 1− αzˆ −1 ⎠
⎝ 1− αzˆ −1 ⎠

0.21623 + 0.43245zˆ −1 + 0.21623zˆ −2
0.96993 − 0.2863zˆ −1 + 0.181416zˆ − 2

2

.

Gain Responses
0
GLP(z)
Gain in dB

-10
HLP(z)
-20
-30
-40
-50

9.23

0

0.2

0.4

0.6

0.8

1

0.55π + 0.47π ⎞
− cos⎛⎜

2

⎠ = 0.0317.
α=
0.55π − 0.47π ⎞
cos⎛⎜

2

H HP (z ) = G LP (z ) z −1 = −⎛⎜ zˆ −1 + α ⎞⎟ =
⎜ 1+ αzˆ −1 ⎟

0.3192 − 0.6383zˆ −1 + 0.3192 zˆ −2
0.9943 − 0.1097 zˆ −1 + 0.1728zˆ − 2

Not for sale

.

303

1.

Gain Responses
0

Gain, dB

-10
HHP(z)

-20

GLP(z)

-30
-40
-50

0

0.2

0.4

0.6

0.8

1

ω /π

ˆ c2 − ω
ˆ c1 , and α = cos(ω
ˆ c ) = 0.2790
9.24 ωc = ω
⎛ zˆ −1 − α ⎞ 0.2790 zˆ −1 − zˆ − 2
−1

1
⎟=
z = − zˆ ⎜
.
⎜ 1 − αzˆ −1 ⎟
1 − 0.2790 zˆ −1

0.6918 − 0.1968zˆ −1 − 0.6234 zˆ −2 − 0.1968zˆ −3 + 0.6918zˆ −4
H BP (z ) =
.
3.5588 − 2.6824 zˆ −1 + 3.046 zˆ −2 − 1.2546 zˆ −3 + zˆ −4
Gain Responses
0
HBP(z)
Gain, dB

-10

GLP(z)

-20
-30
-40
-50

0

0.2

0.4

0.6

0.8

1

ω /π

⎛ 0.52 π − 0.45π ⎞
sin⎜

2

⎠ = 0.1099.
ˆ p = 0.45π, and ω p = 0.52π. λ =
9.25 ω
⎛ 0.52π + 0.45π ⎞
sin⎜

2

2.873 − 5.622 zˆ −1 + 5.629zˆ −2 − 2.874 zˆ −3
.
H HP (z ) = G HP (z ) z = zˆ −1 −0.1099 =
−1
− 2 ˆ −3

1
ˆ
ˆ

z
+
z
+
z
9
.
7623
1
.
5359
6
.
711
1−0.1099 zˆ

Not for sale

304

Gain Responses
0
HHP(z)

Gain in dB

-10

GHP(z)

-20
-30
-40
-50

0

0.2

0.4

0.6

0.8

1

ω /π

1 − z −2
9.26 Eq. (7.79): H BP (z ) = 0.136728736⎜
⎜ 1 − 0.53353098z −1 + 0.726542528z −2

⎛ 0.5π − 0.4π ⎞
sin⎜

2

⎠ = −0.1584
ˆ 0 = 0.5π, and ω0 = 0.4 π. λ =
ω
⎛ 0.5π + 0.4π ⎞
sin ⎜

2

G BP (z ) = H BP (z ) z −1= zˆ −1 + 0.1584 =
1+ 0.1584

zˆ −1

0.1333 − 0.1333zˆ −2
0.933718 + 0.667122 zˆ −2

⎟.

.

Gain Responses
0
GBP(z)

HBP(z)

Gain in dB

-10
-20
-30
-40
-50

0

0.2

0.4

0.6

0.8

1

ω /π

B
1 − tan⎛⎜ ω ⎞⎟
⎛ 80 ⎞
⎛ 5 ⎞
⎝ 2 ⎠ = 0.9244
9.27 ω0 = 2 π⎜
⎟ = 0.4 π , Bω = 2 π⎜
⎟ = 0.025π. α =
B
⎝ 400 ⎠
⎝ 400 ⎠
1 + tan⎛⎜ ω ⎞⎟
⎝ 2 ⎠
β = cos(ω0 ) = 0.3090.

G (z ) =

=

1
2

(1 + α ) − 2β(1 + α )z −1 + (1 + α )z − 2
1 − β(1 + α )z −1 + αz − 2

0.9622 − 0.59465z
1 − 0.5946 z

−1

−1

+ 0.9622 z

+ 0.9244 z − 2

=

(

0.5 1.9244 − 1.1893 z −1 + 1.9244 z − 2
1 − 0.5946 z

−1

+ 0.9244 z

−2

)

−2

.

Not for sale

305

ˆ
ω −ω
sin⎛⎜ 0 0 ⎞⎟
⎝ 2 ⎠ = 0.2738.
α=
ˆ
ω +ω
sin⎛⎜ 0 0 ⎞⎟
⎝ 2 ⎠

50 ⎞
ˆ 0 = ⎛⎜
ω
⎟2 π = 0.25π.
⎝ 400 ⎠

H BS (z ) = G BS (z ) z −1 = zˆ −1 −α =

1.0301 − 1.45679zˆ −1 + 1.0301zˆ −2
1.06017 − 1.45674zˆ −1 + zˆ − 2

1− αzˆ −1

.

Gain Responses
0
HBS(z)

Gain in dB

-10

GBS(z)

-20
-30
-40
-50

0

0.2

0.4

0.6

0.8

1

ω /π

ˆ p = 0.45π and ω p = 0.52π.
9.28 ω
G HP (z ) =

(

0.2397 1 − 1.5858z −1 + 1.5858z −2 − z −3

1 + 0.3272 z −1 + 0.7459z −2 + 0.179z −3
⎛ ω p + ωˆ p ⎞
− cos⎜

2

⎠ = −0.0474.
α=
ˆ
ω

ω
⎛ p p⎞
cos⎜

⎝ 2 ⎠
H LP (z ) = G HP (z )

− zˆ −1 − α
z −1 =
1+ αzˆ −1

=

)

− 1.03399 − 1.45578zˆ −1 − 1.45578zˆ −2 − 1.03399zˆ −3
− 4.72684 + 2.52265zˆ −1 − 3.7756zˆ − 2 + zˆ −3
Gain Responses

0
HLP(z)

Gain, dB

-10

GHP(z)

-20
-30
-40
-50

0

0.2

0.4

0.6

0.8

1

ω /π

9.29 D = 6.9231 , hence, N = 6. A(z ) =

( )

z −6 D z −1
.
D (z )

Not for sale

306

Group delay
6

0.75

5.5

Magnitude

Magnitude response
0.8

0.7
0.65
0.6
0.55

5
4.5
4
3.5

0

0.2

0.4

0.6

0.8

3

1

0

0.2

0.4

ω/ π

9.30

0.6

0.8

1

ω/ π

From Eq. (2.141) y(nT ) = y((n − 1)T ) +

nT

∫ x(τ)dτ =y((n − 1)T ) + T ⋅ x((n − 1)T ), which

( n −1)T

reduces to y[ n] = y[ n − 1] + T ⋅ x[ n − 1]. Hence, the corresponding transfer function is
given by H R ( z ) =

Tz −1

1 − z −1

. From Eq. (2.120) y[ n] = y[ n − 1] +

T
(x[ n] + x[ n − 1]) .
2

T 1 + z −1
. From the
Hence, the corresponding transfer function is given by H T ( z) = ⋅
2 1 − z −1
plot given below it can be seen that the magnitude response of H int (z) lies between that
of H R ( z) and H T (z) .

Magnitude

1.5

1
HR(z)
0.5
Hint(z)

HT(z)
0
0.2

0.4

0.6

0.8

1

ω/ π

3

1

9.31 H N ( z ) = H R ( z ) + H T ( z ). From the plot given below it can be seen that the magnitude
4
4

response of H N (z) lies between that of H R (z) and H T (z) , and is much closer to that of
H int ( z) .

Not for sale

307

Magnitude

1.5

1
HR(z)
0.5

HN(z)
HT(z)

0
0.2

0.4

Hint(z)

0.6

0.8

1

ω/π

3 ⎛ z −1

1 ⎛ 1+ z −1 ⎞

8(1− z −1 )

1+ 7 z −1

⎟+ ⎜
,
⎟=
9.32 H N ( z ) = ⎜⎜
. Its inverse is given by H ( z ) =
4 ⎝ 1 − z −1 ⎟⎠ 8 ⎜⎝ 1− z −1 ⎟⎠ 8(1 − z −1 )
1+ 7 z −1
which is unstable as it has a pole at z = −7. A stable transfer function with the same
magnitude response is obtained by multiplying H (z ) with an allpass function
resulting in H IIR ( z) =

8(1− z −1 ) 1+ 7 z −1

1+ 7 z −1 7 + z −1

=

8(1− z −1 )
7 + z −1

1+ 7 z −1
7 + z −1

. A plot of the ideal differentiator

H DIF (z) with a frequency response given by Eq. (7.68) and that of the IIR differentiator
H IIR (z) is given below. As can be seen the magnitude response of H IIR (z) is very close
to that H DIF (z) .
3

HIIR(z)→

Magnitude

2.5
2
HDIF(z)

1.5
1
0.5
0

0

0.2

0.4

0.6

0.8

1

ω/π

M9.1 FT = 100 kHz, α p = 0.4 dB, Fp = 10 kHz, α s = 50 dB, & Fs = 30 kHz ,
ωp =

2 πF p
FT

= 0.628, ωs =

2 πFs
= 1.885.
FT

⎛ ωp
Let T = 2. Ω p = tan⎜⎜
⎝ 2

ω
⎟ = 1.376, and Ω s = tan⎛⎜ s ⎞⎟ = 1.376. Therefore,

⎝ 2 ⎠

⎛ 1 ⎞
1 Ωs
⎟ = −0.4. Hence, ε 2 = 0.096.
= 4.235. Now, 20 log10 ⎜
=

2 ⎟
k Ωp
⎝ 1+ ε ⎠

Not for sale

308

⎛1⎞
From 20 log10 ⎜ ⎟ = −50 we obtain A 2 = 100,000. Therefore,
⎝ A⎠
1
k1 = 9.798 × 10 −4 , or
= 1020.62. As a result,
k1

log10 (1 / k1 )
N=
= 4.80 → 5. Next, solving
log10 (1 / k )

⎛ Ωs
⎜⎜
⎝ Ωc

10

⎟⎟

= A 2 − 1 = 99,999 we get

Ωs

Ωc =

= 0.316Ω s = 0.435
99,9991 / 10
Using the M-file buttap, we determine the normalized analog Butterworth
transfer function of 5th order with a 3-dB cutoff frequency at Ω c = 1 , which is:
1
.
H an (s ) =
2
(s + 1) s + 0.618s + 1 s 2 + 1.618s + 1
Denormalize H an (s ) to move Ω c to 0.316:
1
s ⎞
H a (s ) = H an ⎛⎜
⎟=
⎝ 0.435 ⎠ (2.30s + 1)(5.28s 2 + 1.42s + 1)(5.28s 2 + 3.72s + 1)

(

=

)(

)

1
1 + 7.44s + 27.66s 2 + 63.58s 3 + 90.30s 4 + 64.12s 5

.

G (z ) = H a (s ) s = 2 ⎛⎜ 1− z −1 ⎞⎟
T ⎜ 1+ z −1 ⎟

=

0.0039 + 0.0197z −1 + 0.0394 z −2 + 0.0394z −3 + 0.0197z −4 + 0.0039z −5
1 − 2.3617z −1 + 2.6139z − 2 − 1.5492 z −3 + 0.4865z − 4 − 0.0637z − 5

.

Matlab code is as follows:
% Program M9.01
N = 5;
[z, p, k] = buttap(N);
[num, den] = zp2tf(z, p, k);
% s -> s/0.435
den = [64.12 90.30 63.58 27.66 7.44 1];
num = [0 0 0 0 0 1];
% compute z, p, and k
[z, p, k] = tf2zp(num, den);
% perform bilinear transformation with T = 2;
[zd, pd, kd] = bilinear(z, p, k, 1/2);
% get the digital transfer function
[n2, d2] = zp2tf(zd, pd, kd);
% get the frequency response
[h, w] = freqz(n2, d2, 512);
figure(1);
plot(w/pi, 20*log10(abs(h))); grid;
axis([0 1 -60 5]);
xlabel('\omega/\pi'); ylabel('Gain, dB');

Not for sale

309

title('Gain response');
figure(2);
plot(w/pi, unwrap(angle(h))); grid;
axis([0 1 -8 1]);
title('Phase response');
Gain response

Phase response

0

0

Gain, dB

-10
-20
-30
-40
-50
-60

0

0.2

0.4

0.6

0.8

1

-2
-4
-6
-8

ω/ π

0

0.2

0.4

0.6

0.8

1

ω/ π

M9.2 % Problem M9.02
Fp = input('Passband edge frequency in Hz = ');
Fs = input('Stopband edge frequency in Hz = ');
FT = input('Sampling frequency in Hz = ');
Rp = input('Passband ripple in dB = ');
Rs = input('Stopband minimum attenuation in dB = ');
Wp = 2*Fp/FT;
Ws = 2*Fs/FT;
[N, Wn] = buttord(Wp, Ws, Rp, Rs)
[b, a] = butter(N, Wn);
disp('Numerator polynomial');
disp(b)
disp('Denominator polynomial');
disp(a)
[h, w] = freqz(b, a, 512);
plot(w/pi, 20*log10(abs(h))); grid
axis([0 1 -60 5]);
xlabel('\omega/\pi'); ylabel('Magnitude, dB');
pause
plot(w/pi, unwrap(angle(h))); grid
axis([0 1 -8 1]);
M9.3 % Program #M9.03
close all;
clear;
clc;
% (a.) Ft = 1Hz;
[B, A] = besself(5, 0.5);

Not for sale

310

[BZ,AZ] = impinvar(B,A,1);
[h, w] = freqz(BZ, AZ, 512);
[Gd,W] = grpdelay(BZ,AZ,512);
figure(1);
plot(w/pi, 20*log10(abs(h)));
title('Gain response, , Sampling rate = 1 Hz');
xlabel('\omega/\pi'); ylabel('Gain, in dB');
figure(2);
plot(W/pi, (Gd));
title('Group delay, Sampling rate = 1 Hz');
xlabel('\omega/\pi'); ylabel('Delay, in samples');
% (b.) Ft = 2Hz;
[B, A] = besself(5, 0.5);
[BZ,AZ] = impinvar(B,A,2);
[h, w] = freqz(BZ, AZ, 512);
[Gd,W] = grpdelay(BZ,AZ,512);
figure(3);
plot(w/pi, 20*log10(abs(h)));
title('Gain response, Sampling rate = 2 Hz');
xlabel('\omega/\pi');ylabel('gain response');
figure(4);
plot(W/pi, (Gd));
title('Group delay, Sampling rate = 2 Hz');
xlabel('\omega/\pi'); ylabel('Delay, in samples');
(a) FT = 1 Hz
Group delay, Sampling rate = 1 Hz
8

0

6

Delay, in samples

Gain, in dB

Gain response, Sampling rate = 1 Hz
20

-20
-40
-60
-80

0

0.2

0.4

0.6

0.8

1

ω/ π

4
2
0
-2
-4

0

0.2

0.4

0.6

0.8

1

ω/ π

(b) FT = 2 Hz

Not for sale

311

Group delay, Sampling rate = 2 Hz
20

0

15

Delay, in samples

Gain, in dB

Gain response, Sampling rate = 2 Hz
20

-20
-40
-60

10
5
0

-80
0

0.2

0.4

0.6

0.8

-5

1

0

0.2

0.4

ω/ π

M9.4 ω p = 0.628, ω s = 1.885.

(

0.6

0.8

1

ω/ π

)

(

)

20 log10 G e 0.628 j ≥ −0.4, 20 log10 G e1.885 j ≤ −50.

Impulse invariance method, let T = 1 and assume no aliasing. Then
⎛ 1 ⎞
⎟ = −0.4 ⇒ ε 2 = 0.096 ,
20 log10 ⎜⎜

2
⎝ 1+ε ⎠
⎛1⎞
20 log10 ⎜ ⎟ = −50 ⇒ A 2 = 100,000 .
⎝ A⎠

A2 − 1

1
=
k1

= 1020.62 ,

ε2
⎛1⎞
log10 ⎜ ⎟
⎝ k1 ⎠ = 6.3 → 7
N=
1
log10 ⎛⎜ ⎞⎟
⎝k⎠
⎛ Ωs
⎜⎜
⎝ Ωc

1 1.885
=
= 3 . Hence,
k 0.628

10

⎟⎟

= A 2 − 1 = 99,999 or Ω c =

Ωs

(A − 1)
2

1 / 10

= 0.596 .

Using the M-file butter, we get the analog Butterworth transfer function H a (s )
of 7th order with a 3-dB cutoff frequency at Ω c .
H a (s ) is then transformed into a digital transfer function using the M-file
impinvar, which yields:

G (z ) =

0.001z −2 + 0.0034 z −3 + 0.0023z −4 + 0.0003z −5
1 − 4.38z −1 + 8.54 z −2 − 9.55z −3 + 6.57z −4 − 2.77z −5 + 0.66 z −6 − 0.07z −7

% Problem M9.04
[B, A] = butter(7, 0.596, 's');
[num, den] = impinvar(B, A, 1);

Not for sale

312

.

% get the frequency response
[h, w] = freqz(num, den, 512);
plot(w/pi, 20*log10(abs(h))); grid;
axis([0 1 -60 5]);
xlabel('\omega/\pi'); ylabel('Gain, in dB');
title('Gain response');
Gain response
0

Gain, in dB

-10
-20
-30
-40
-50
-60

0

0.2

0.4

0.6

0.8

1

ω/ π

M9.5 FT = 100 kHz, Fp = 10 kHz, Fs = 30 kHz, α p = 0.4 dB, α s = 50 dB
ωp =

2 πFp
FT

2πFs
= 0.6 π.
FT

= 0.2 π, and . ω s =

(

)

(

)

20 log10 G e 0.628 j ≥ −0.4, 20 log10 G e1.885 j ≤ −50.

Impulse invariance method:
Let T = 1 and assume no aliasing. In this case, the specifications for H a (s ) are

same as that for G (z ), i.e.,
Ω p = 0.628, Ω s = 1.885,

20 log10 H a (0.628 j ) ≥ −0.4, 20 log10 H a (1.885 j ) ≤ −50.
⎛ 1 ⎞
⎟ = −0.4 ⇒ ε 2 = 0.096 , and
Now, 20 log10 ⎜⎜

2
⎝ 1+ε ⎠
1
20 log10 ⎛⎜ ⎞⎟ = −50 ⇒ A 2 = 100,000 . Hence,
⎝ A⎠
1
=
k1

A2 − 1
ε

2

= 1020.62 and

1 1.885
=
= 3.
k 0.628

Order of the Type I Chebyshev filter is N =

cosh −1 (1 / k1 )
cosh −1 (1 / k )

= 4.324 → 5

Bilinear transformation method:
Let T = 2. Ω p = 0.325, Ω s = 1.376 . Here,

Not for sale

1 Ωs
=
= 4.235 , and
k Ωp

313

ε 2 = 0.096, A 2 = 100,000 . Thus,

1
= 1020.62 .
k1

Order of the Type I Chebyshev filter is N =

cosh −1 (1 / k1 )

= 3.59 → 4.
cosh −1 (1 / k )
Both designs meet the specifications, while the bilinear transformation method
meets with a filter of lower order. MATLAB code is as follows:

% Problem #M9.05
% Impulse invariance method
[z, p, k] = cheb1ap(5, 0.4);
[B, A] = zp2tf(z, p, k);
[BT, AT] = lp2lp(B, A, 0.628);
[num, den] = impinvar(BT, AT, 1);
[h, w] = freqz(num, den, 512);
figure(1);
plot(w/pi, 20*log10(abs(h))); grid;
axis([0 1 -60 5]);
xlabel('\omega/\pi'); ylabel('Gain, in dB');
title('Impulse Invariance Method');
figure(2);
plot(w/pi, unwrap(angle(h))); grid
axis([0 1 -8 1]);
title('Impulse Invariance Method');
% Bilinear transformation method
[z, p, k] = cheb1ap(4, 0.4);
[B, A] = zp2tf(z, p, k);
[BT, AT] = lp2lp(B, A, 0.325);
[num, den] = bilinear(BT, AT, 0.5);
[h, w] = freqz(num, den, 512);
figure(3);
plot(w/pi, 20*log10(abs(h))); grid;
axis([0 1 -60 5]);
xlabel('\omega/\pi'); ylabel('Gain, in dB');
title('Bilinear Transformation Method');
figure(4);
plot(w/pi, unwrap(angle(h))); grid
axis([0 1 -8 1]);
title('Bilinear Transformation Method');

Impulse Invariance Method:

Not for sale

314

Impulse Invariance Method

Impulse Invariance Method

0

0

Gain, in dB

-10
-20
-30
-40
-50
-60

0

0.2

0.4

0.6

0.8

-2
-4
-6
-8

1

0

0.2

0.4

ω/ π

0.6

0.8

1

ω/ π

Bilinear Transformation Method:
Bilinear Transformation Method

Bilinear Transformation Method
0

0

Gain, in dB

-10
-20
-30
-40
-50
-60

0

0.2

0.4

0.6

0.8

1

-2
-4
-6
-8

0

0.2

0.4

0.6

0.8

1

ω/ π

ω/ π

M9.6 % Problem M9.06
Wp = input('Normalized passband edge = ');
Ws = input('Normalized stopband edge = ');
Rp = input('Passband ripple in dB = ');
Rs = input('Minimum stopband attenuation in dB = ');
[N, Wn] = cheb1ord(Wp, Ws, Rp, Rs);
[b, a] = cheby1(N, Rp, Wn);
[h, omega] = freqz(b,a,256);
figure(1);
plot(omega/pi, 20*log10(abs(h)));grid;
xlabel('\omega/\pi'); ylabel('Gain, in dB');
title('Type I Chebyshev Filter');
axis([0 1 -60 5]);
figure(2);
plot(omega/pi, unwrap(angle(h))); grid
axis([0 1 -8 1]);
title('Type I Chebyshev Filter');

Not for sale

315

Type I Chebyshev Filter

Type I Chebyshev Filter

0

0

Gain, in dB

-10
-20
-30
-40
-50
-60

0

0.2

0.4

0.6

0.8

-2
-4
-6
-8

1

0

0.2

0.4

ω/π

0.6

0.8

1

ω/π

M9.7 % Problem M9.07
Wp = input(' Passband edge in radians = ');
Ws = input(' Stopband edge in radians = ');
Rp = input('Passband ripple in dB = ');
Rs = input('Stopband minimum attenuation in dB = ');
[N, Wn] = cheb1ord(Wp,Ws,Rp,Rs, 's');
[B, A] = cheby1(N, 0.5, Wn, 's');
[num, den] = impinvar(B, A, 1);
[h, omega] = freqz(num,den,256);
figure(1);
plot(omega/pi, 20*log10(abs(h)));grid;
xlabel('\omega/\pi'); ylabel('Gain, in dB');
title('Type I Chebyshev Filter');
axis([0 1 -60 5]);
figure(2);
plot(omega/pi, unwrap(angle(h))); grid
axis([0 1 -8 1]);
title('Type I Chebyshev Filter');
Type I Chebyshev Filter

Type I Chebyshev Filter

0

0

Gain, in dB

-10
-20
-30
-40
-50
-60

0

0.2

0.4

0.6

0.8

1

-2
-4
-6
-8

ω/ π

0

0.2

0.4

0.6

0.8

1

ω/ π

M9.8 Impulse invariance method:
2 log10 (4 / k1 )
1 Ωs
1
, where
=
= 3.00,
= 1020.62. Hence, N ≈
log10 (1 / ρ )
k Ωp
k1

Not for sale

316

k ' = 1 − k 2 , ρ0 =

1 − k'

(

2 1 + k'

),

ρ = ρ0 + 2 ρ05 + 15 ρ09 + 150 ρ13
0 . In our case,
k ' = 0.943, ρ0 = 0.0073, ρ = 0.0073 . Hence, N ≈ 3.38 → 4

Bilinear transformation method:
1
1 Ω
= 1020.62.
Ω p = 0.325, Ω s = 1.376, = s = 4.235.
k Ωp
k1

k ' = 0.972, ρ0 = 0.0036 , ρ = 0.0036 . Hence, N ≈ 2.96 → 3
We note that the filter design using the impulse invariance method does not meet
the specifications due to aliasing. Increasing N from 4 to 6 will meet the
specifications.
% Problem M9_08
% Impulse Invariance Method
[z, p, k] = ellipap(4, 0.4, 50);
[B, A] = zp2tf(z, p, k);
[BT, AT] = lp2lp(B, A, 0.628);
[num, den] = impinvar(BT, AT, 1);
[h, omega] = freqz(num,den,256);
figure(1);
plot(omega/pi, 20*log10(abs(h)));grid;
xlabel('\omega/\pi'); ylabel('Gain, in dB');
title('Impulse Invariance Method');
axis([0 1 -60 5]);
figure(2);
plot(omega/pi, unwrap(angle(h))); grid
axis([0 1 -8 1]);
title('Impulse Invariance Method');
% Bilinear Transformation Method
[z, p, k] = ellipap(4, 0.4, 50);
[B, A] = zp2tf(z, p, k);
[BT, AT] = lp2lp(B, A, 0.325);
[num, den] = bilinear(BT, AT, 0.5);
[h, omega] = freqz(num,den,256);
figure(3);
plot(omega/pi, 20*log10(abs(h)));grid;
xlabel('\omega/\pi'); ylabel('Gain, in dB');
title('Bilinear Transformation Method');
axis([0 1 -60 5]);
figure(4);
plot(omega/pi, unwrap(angle(h))); grid
axis([0 1 -8 1]);
Not for sale

317

title('Bilinear Transformation Method');
Impulse Invariance Method

Impulse Invariance Method
0

0

Gain, in dB

-10
-20
-30
-40
-50
-60

0

0.2

0.4

0.6

0.8

-2
-4
-6
-8

1

0

0.2

0.4

0.6

0.8

1

ω/ π

ω/ π
Bilinear Transformation Method

Bilinear Transformation Method

0
0

Gain, in dB

-10
-20
-30
-40
-50
-60

0

0.2

0.4

0.6

0.8

-2
-4
-6
-8

1

0

0.2

0.4

ω/ π

0.6

0.8

1

ω/ π

M9.9 % Program M9_09
Wp = input('Normalized passband edge = ');
Ws = input('Normalized stopband edge = ');
Rp = input('Passband ripple in dB = ');
Rs = input('Stopband ripple in dB = ');
[N, Wn] = ellipord(Wp, Ws, Rp, Rs);
[b, a] = ellip(N, Rp, Rs, Wn);
[h, omega] = freqz(b, a, 256);
plot(omega/pi, 20*log10(abs(h)));grid;
xlabel('\omega/\pi'); ylabel('Gain, in dB');
title('IIR Elliptic Lowpass Filter');
axis([0 1 -60 5]);
IIR Elliptic Lowpass Filter
0

Gain, in dB

-10
-20
-30
-40
-50
-60

0

0.2

0.4

0.6

0.8

1

ω/π

M9.10 FT = 1.5 MHz, Fp = 600 kHz, Fs = 210 kHz.

Not for sale

318

. α p = 0.4 dB, α s = 45 dB
(a) ω p =

2 πF p
FT

= 2.513, ωs =

2πFs
= 0.880.
FF

ˆ = tan⎛⎜ ωp ⎞⎟ = 3.076, Ω
ˆ = tan⎛⎜ ωs ⎞⎟ = 0.471.

p
s
⎝ 2⎠
⎝ 2 ⎠
Analog highpass specifications are thus:
ˆ = 3.076, Ω
ˆ = 0.471, α = 0.4 dB, α = 45 dB

p
s
p
s
ˆ

p
(b) Ω p = 1, Ω s =
= 6.531
ˆ

s
% Problem M9.10
close all;
clear;
clc;
[N, Wn] = ellipord(1, 6.531, 0.4, 45, 's');
[B, A] = ellip(N, 0.4, 45, Wn, 's');
[BT, AT] = lp2hp(B, A, 3.076);
[num, den] = bilinear(BT, AT, 0.5);
[h, omega] = freqz(num, den, 256);

figure(1);
plot(omega/pi, 20*log10(abs(h)));grid;
xlabel('\omega/\pi'); ylabel('Gain, in dB');
title('Elliptic Highpass Filter');
axis([0 1 -60 5]);
Elliptic Highpass Filter
0

Gain, in dB

-10
-20
-30
-40
-50
-60

0

0.2

0.4

0.6

0.8

1

ω/ π

(c) Analog lowpass transfer function coefficient are obtained by displaying B and A:
H LP (s ) =

0.0556 s 2 + 0.8329
s 3 + 1.3314 s 2 + 1.6434 s + 0.8329

.

Analog highpass transfer function coefficient are obtained by displaying BT and AT:

Not for sale

319

H HP (s ) =

s 3 + 0.6315 s
s 3 + 6.0693 s 2 + 15.1254 s + 34.9438

.

Digital highpass transfer function coefficient are obtained by displaying num and
den:
: G HP ( z ) =

0.0286 − 0.0415 z −1 + 0.0415 z −2 − 0.0286 z −3
1 + 1.9407 z −1 + 1.5613 z − 2 + 0.4356 z − 3

.

M9.11 FT = 9kHz, Fp1 = 1.2kHz, Fp2 = 2.2kHz, Fs1 = 650 Hz, Fs 2 = 3kHz,

α p = 0.8dB, αs = 31dB
(a) ω p1 =

ωs 2 =

2 πF p1
FT

= 0.8378 , ω p 2 =

2 πF p2
FT

2πFs1
= 0.4538 , and
FT

2πFs 2
= 2.094 , with
FT

⎛ ω p1 ⎞
⎛ ω p2
ˆ
ˆ

,
=
=

tan
=
0
.
445

tan
p1
p2
⎜ 2 ⎟
⎜ 2

ˆ = tan⎛⎜ ωs 2 ⎞⎟ = 1.731 .
and Ω
s2
⎝ 2 ⎠
ˆ
ˆ
B =Ω
−Ω
= 0.521,
ω

= 1.536 , ω s1 =

p2

ˆ = tan⎛⎜ ωs1 ⎞⎟ = 0.231 ,
⎟ = 0.966 , Ω
s1

⎝ 2 ⎠

p1

ˆ2 =Ω
ˆ Ω
ˆ
ˆ ˆ

0
p1 p2 = 0.430 ≠ Ω s1Ω s 2 = 0.40
ˆ 2 −Ω
ˆ2

s1
0
Therefore, we choose to lower the stopband edge to: Ω s =
= 3.13
ˆ
Ω s1 ⋅ Bω
(b) The analog lowpass specifications are thus:
Ω p = 1 rad, Ω s = 3.13 rad, α p = 0.8 dB, α s = 31 dB.

% Problem M9.11
[N, Wn] = cheb1ord(1, 3.13, 0.8, 31, 's');
[B, A] = cheby1(N, 0.8, Wn, 's');
[BT, AT] = lp2bp(B, A, sqrt(0.43), 0.521);
[num, den] = bilinear(BT, AT, 0.5);
[h, omega] = freqz(num, den, 256);
plot(omega/pi, 20*log10(abs(h)));grid;
xlabel('\omega/\pi'); ylabel('Gain,in dB');
title('Chebyshev I Bandpass Filter');
axis([0 1 -60 5]);

Not for sale

320

Chebyshev I Bandpass Filter
0

Gain,in dB

-10
-20
-30
-40
-50
-60

0

0.2

0.4

0.6

0.8

1

ω/π

(c) Analog lowpass transfer function coefficient are obtained by displaying B and A:
0.5559
H LP (s ) =
.
s 3 + 1.0719 s 2 + 1.3244 s + 0.5559

Analog bandpass transfer function coefficient are obtained by displaying BT and AT:
H BP (s ) =

0.0786 s 3
s 6 + 0.5584 s 5 + 1.6495 s 4 + 0.5589 s 3 + 0.7093 s 2 + 0.1033 s + 0.0795

.

Digital bandpass transfer function coefficient are obtained by displaying den and
num:
G BP ( z ) =

0.0169 − 0.0506 z −2 + 0.0506 z −4 − 0.0169 z −6
1 − 1.9799 z −1 + 3.3196 z − 2 − 3.1443 z − 3 + 2.6191 z − 4 + 1.1983 z − 5 + 0.476 z − 6

M9.12 FT = 8 kHz, Fp1 = 0.9 kHz, Fp2 = 2.1 kHz, Fs1 = 0.6 kHz, Fs 2 = 3 kHz,
α p = 1.5 dB, α s = 30 dB
(a) ω p1 =

ωs2 =

2πF p1
FT

= 0.707 , ω p 2 =

2πF p 2
FT

= 1.649 , ω s1 =

2πFs 2
= 2.356 , with
FT

ω
ω
ˆ = tan⎛⎜ p 2
ˆ = tan⎛⎜ p1 ⎞⎟ = 0.369 , Ω

p1
p
2
⎜ 2 ⎟
⎜ 2

ˆ = tan⎛⎜ ω s 2 ⎞⎟ = 2.414 .
and Ω
s2
⎝ 2 ⎠
ˆ −Ω
ˆ = 2.174
B =Ω
ω

2πFs1
= 0.471 , and
FT

s2

ˆ = tan⎛⎜ ω s1 ⎞⎟ = 0.240 ,
⎟⎟ = 1.081 , Ω
s1
⎝ 2 ⎠

s1

ˆ2 =Ω
ˆ Ω
ˆ = 0.399 ≠ Ω
ˆ Ω
ˆ = 0.579

0
p1
p2
s1
s2
Therefore, we choose to lower the passband:

Not for sale

321

.

ˆ ˆ
ˆ = Ω s 2 Ω s1 = 0.536

p1
ˆ

p2

The analog bandstop filter specifications are thus:

s1
s2
p1
p2

α p = 1.5 dB, α s = 30 dB.
(b) Analog prototype LP filter:
ˆ ⋅B

ω
p1
= 3.995 rad, α p = 1.5 dB, α s = 30 dB.
Ω s = 1 rad, Ω p =
ˆ 2 −Ω
ˆ2

0
p1
% Problem M9.12
close all;
clear;
clc;
[N, Wn] = ellipord(3.995, 1, 1.5, 30, 's');
[B, A] = ellip(N, 1.5, 30, Wn, 's');
[BT, AT] = lp2bs(B, A, sqrt(0.579), 2.174);
[num, den] = bilinear(BT, AT, 0.5);
[h, omega] = freqz(num, den, 256);
plot(omega/pi, 20*log10(abs(h)));grid;
xlabel('\omega/\pi'); ylabel('Gain, in dB');
title('Elliptic Bandstop Filter');
axis([0 1 -60 5]);
Elliptic Bandstop Filter
0

Gain, in dB

-10
-20
-30
-40
-50
-60

0

0.2

0.4

0.6

0.8

1

ω/ π

(c) Analog lowpass transfer function coefficient are obtained by displaying B and A:
H LP (s ) =

0.0316 s 2 + 12.6706
s 2 + 3.6122 s + 15.059

.

Analog bandstop transfer function coefficients are obtained by displaying BT and
AT:
H BS (s ) =

0.8414 s 4 + 0.9843 s 2 + 0.2821
s 4 + 0.5215 s 3 + 1.4719 s 2 + 0.3019 s + 0.3352

Not for sale

.

322

Digital bandstop transfer function coefficients are obtained by displaying num
and den:
G BS ( z ) =

0.5806 − 0.6163 z −1 + 1.3145 z −2 − 0.6163 z −3 + 0.5806 z −4
1 − 0.8534 z −1 + 1.3959 z − 2 − 0.6115 z − 3 + 0.5464 z − 4

.

M9.13 % Program M9.13
close all;
clear;
clc;
Wp = 0.7; Ws = 0.5;
Rp = 1; Rs = 32;
[N,Wn] = cheb1ord(Wp,Ws,Rp,Rs);
[b,a] = cheby1(N,Rp,Wn,'high');
[h,omega] = freqz(b,a,256);
plot (omega/pi,20*log10(abs(h)));grid;
xlabel('\omega/\pi'); ylabel('Gain, in dB');
title('Type I Chebyshev Highpass Filter');
[GdH,w] = grpdelay(b,a,512);
plot(w/pi,GdH); grid
xlabel('\omega/\pi'); ylabel('Delay, in samples');
title('Original Filter Group Delay');
F = 0.7:0.001:1;
g = grpdelay(b,a,F,2);
% Equalize the passband
Gd = max(g)-g;
% Design the allpass delay equalizer
[num,den,tau] = iirgrpdelay(2*N, F, [0.7 1], Gd);
[GdA,w] = grpdelay(num,den,512);
plot(w/pi,GdH+GdA); grid
xlabel('\omega/\pi');ylabel('Delay, in samples');
title('Group Delay Equalized Filter');
Group Delay Equalized Filter

Delay, in samples

40
30
20
10
0

0

0.2

0.4

0.6

0.8

1

ω/ π

M9.14 % Program M9.14
close all;
clear;

Not for sale

323

clc;
Wp = [0.45 0.65]; Ws = [0.3 0.75];
Rp = 1; Rs = 40;
[N,Wn] = buttord(Wp, Ws, Rp, Rs);
[b,a] = butter(N,Wn);
[h,omega] = freqz(b,a,256);
gain = 20*log10(abs(h));
plot (omega/pi,gain);grid;
xlabel('\omega/\pi'); ylabel('Gain, in dB');
title('IIR Butterworth Bandpass Filter');
[GdH,w] = grpdelay(b,a,512);
plot(w/pi,GdH); grid
xlabel('\omega/\pi'); ylabel('Delay, in samples');
title('Original Filter Group Delay');
F = 0.45:0.001:0.65;
g = grpdelay(b,a,F,2);
% Equalize the passband
Gd = max(g)-g;
% Design the allpass delay equalizer
[num,den,tau] = iirgrpdelay(2*N, F, [0.45 0.65], Gd);
[GdA,w] = grpdelay(num,den,512);
plot(w/pi,GdH+GdA); grid
xlabel('\omega/\pi');ylabel('Delay, in samples');
title('Group Delay Equalized Filter');
Group Delay Equalized Filter

Delay, in samples

60
50
40
30
20
10
0

0

0.2

0.4

0.6

0.8

1

ω/π

Not for sale

324

Chapter 10
10.1

To compute the filter orders, we use Kaiser’s formula of Eq. (10.3), Bellanger’s formula
of Eq. (10.4), and Hermann’s formula of Eq. (10.5).
Filter #1:

(

)

− 20 log10 0.0224 ⋅ 0.00012 − 13
= 157.097 ≈ 158
14.6(0.14375π − 0.10625π ) / 2π
− 2 log10 (10 ⋅ 0.0224 ⋅ 0.000112 )
Bellanger’s formula - N =
− 1 = 162.575 ≈ 163
3(0.14375π − 0.10625π ) / 2π
Hermann’s formula 2
D∞ (δ p , δ s ) = 0.005309(log10 0.0224 ) + 0.07114(log10 0.0224 ) − 0.4761 ⋅ log10 0.000112 −
Kaiser’s formula - N =

[0.00266(log

[

]

]

0.0224 ) + 0.5941(log10 0.0224 ) + 0.4278 = 2.8326
F (δ p , δ s ) = 11.01217 + 0.51244[log 10 0.0224 − log 10 0.000112 ] = 12.1913

N=

2

10

2.8326 − 12.1913[(0.14375π − 0.10625π ) / 2π ]2
= 150.8434 ≈ 151
(0.14375π − 0.10625π ) / 2π

Filter #2:

(

)

− 20 log10 0.017 ⋅ 0.034 − 13
= 33.186 ≈ 34
14.6(0.2875π − 0.2075π ) / 2π
− 2 log10 (10 ⋅ 0.017 ⋅ 0.034 )
Bellanger’s formula - N =
− 1 = 36.3012 ≈ 37
3(0.2875π − 0.2075π ) / 2π
Hermann’s formula 2
D∞ (δ p , δ s ) = 0.005309(log10 0.017 ) + 0.07114(log10 0.017 ) − 0.4761 ⋅ log10 0.034 −
Kaiser’s formula - N =

[0.00266(log

[

]

]

0.017 ) + 0.5941(log10 0.017 ) + 0.4278 = 1.474777
F (δ p , δ s ) = 11.01217 + 0.51244[log 10 0.017 − log 10 0.034] = 10.85791019

N=

2

10

1.474777 − 10.85791019[(0.2875π − 0.2075π ) / 2π ]2
= 36.435 ≈ 37
(0.2875π − 0.2075π ) / 2π

Filter #3:

(

)

− 20 log10 0.0411 ⋅ 0.0137 − 13
= 11.6107 ≈ 12
14.6(0.575π − 0.345π ) / 2π
− 2 log10 (10 ⋅ 0.0411 ⋅ 0.0137 )
Bellanger’s formula - N =
− 1 = 12.04 ≈ 13
3(0.575π − 0.345π ) / 2π
Hermann’s formula Kaiser’s formula - N k =

Not for sale

324

[

]

D∞ (δ p , δ s ) = 0.005309(log10 0.0411) + 0.07114(log10 0.0411) − 0.4761 ⋅ log10 0.0137 −

[0.00266(log

2

]

0.0411) + 0.5941(log10 0.0411) + 0.4278 = 1.4424
F (δ p , δ s ) = 11.01217 + 0.51244[log 10 0.0411 − log 10 0.0137 ] = 11.256666
2

10

1.4424 − 11.256666[(0.575π − 0.345π ) / 2π ]2
N=
= 11.248 ≈ 12
(0.575π − 0.345π ) / 2π
10.2

N = 75 and ω s − ω p = 0.05π and we assume δ s = δ p .
(a) Using Kaiser’s formula of Eq. (10.3):
⎛ 75[14.6⋅0.05π / 2 π ]+13 ⎞

− 20

= 10

δs
= 0.009577 ∴ α s = 40.375 dB.
(b) Using Bellanger’s formula of Eq. (10.4):
1/ 2

⎛ − 76⋅3⋅0.05π / 2π ⎞ ⎞

⎟⎟

2

δ s = ⎜ 0.1 ⋅ 10
= 0.0119 ∴ α s = 38.5 dB.

(c) Using Hermann’s formula of Eq. (10.5):
2
F = b1 ∴ D∞ (δ s ) = N (ω s − ω p ) / 2π + b1 (ω s − ω p ) / 2π

[

[

]

]

D∞ (δ s ) = a1 (log10 δ s ) + a 2 (log10 δ s ) + a3 (log10 δ s ) −

[a (log
4

10

2

δ s )2 + a5 (log10 δ s ) + a6

]

D∞ (δ s ) = a1 (log10 δ s ) + a 2 (log10 δ s ) + a3 (log10 δ s ) −
3

2

a 4 (log10 δ s ) − a5 (log10 δ s ) − a 6
2

D∞ (δ s ) = a1 (log10 δ s ) + (a 2 − a 4 )(log10 δ s ) + (a3 − a5 )(log10 δ s ) − a 6
Let x = (log10 δ s ) , and thus
3

2

D∞ (δ s ) = 0.005309 x 3 + 0.06848 x 2 − 1.0702 x − 0.4278 = 1.875697
Solving for x gives us three possible solutions:
x = −21.3787 , x = −1.94654, x = 10.4263
The most reasonable solution is the second. Therefore,
δ s = 0.0113 ∴ α s = 38.93

10.3

N = 75 and ω s − ω p = 0.05π = ∆ω

α s = 2.285(∆ω )N + 8 = 34.9 dB.
10.4

The ideal L -band digital filter H ML (z) has a frequency response given by
H ML (e jω ) = Ak , for ωk −1 ≤ ω ≤ ωk , 1 ≤ k ≤ L, and can be considered as sum of L ideal
k
k
0
bandpass filters with cutoff frequencies at ωc1 = ωk −1 and ωc 2 = ωk , where ωc1 = 0
L
and ωc 2 = π. Now from Eq. (10.47) the impulse response of an ideal bandpass filter is

Not for sale

325

sin(ωc 2 n) sin(ωc1n)
. Therefore ,

πn
πn
sin(ωk n) sin(ωk −1n)
k

[ n] =
h BP
. Hence,
πn
πn

given by h BP [ n] =

h ML [ n] =

L

k
[ n] =
h BP

k =1

L

⎛ sin(ωk n) sin(ωk −1n) ⎞

πn
πn

∑ Ak ⎜⎝

k =1

L −1
sin(ωk n) L −1 ⎛ sin(ωk n) sin(ωk −1n) ⎞
⎛ sin(ω1n) sin(0n) ⎞
= A1 ⎜

Ak
Ak ⎜

⎟+
πn ⎠ k = 2
πn
π
n
π
n
⎝ πn

k =2

⎛ sin(ω L n) sin(ω L −1 n) ⎞
+ AL ⎜

πn
πn

sin(ω1n) L −1 sin(ωk n) L −1 sin(ωk −1n)
sin(ωL −1n)
= A1
+

− AL
Ak
Ak
πn
πn
πn
πn
k =2
k =2

=

L −1

L
sin(ωk n)
sin(ωk −1n)

.
Ak
πn
π
n
k =2

Ak

k =1

sin(ωL n)
Since ωL = π, sin(ωL n) = 0. We add a term AL
to the first sum in the above
πn
expression and change the index range of the second sum, resulting in
h ML [ n] =

L

Ak

k =1

sin(ωk n) L −1
sin(ωk n)

Ak + 1
.
πn
π
n
k =1

Finally, since AL +1 = 0, we can add a term AL +1
L

sin(ωL n)
to the second sum. This leads to
πn

L
sin(ωk n) L
sin(ωk n)
sin(ωk n)
h ML [ n] =
Ak
Ak +1

=
( Ak − Ak + 1 )
.
π
n
π
n
π
n
k =1
k =1
k =1

10.5

⎧ j, − π < ω < 0,
H HT (e jω ) = ⎨
Therefore,
⎩− j, 0 < ω < π.
1
hHT [ n] =

1
=

0

∫ je

−π

jωn

0

∫ H HT (e

)e

j ωn

−π

1
dω −

π

1
dω +
H HT (e jω )e jωn dω

0

π

∫ je
0

1
For n = 0, h HT [0] =

0

−π

jωn

2
2 sin 2 (πn / 2)
(1 − cos(πn)) =
dω =
if n ≠ 0.
πn
2 πn

1
jdω −

π

∫ jdω = 0.
0

Not for sale

326

0,
if n = 0,

2
Hence, h HT [n] = ⎨ 2sin (πn / 2)
, if n ≠ 0.
⎪⎩
πn
Since hHT [ n] = −hHT [ − n], and the length of the truncated impulse response is odd, it is a Type
3 linear-phase FIR filter.

Magnitude

1
0.8
0.6
0.4

N=11
N=21
N=61

0.2
0

0

0.2

0.4

0.6

0.8

1

ω/π

From the frequency response plots given above, we observe the presence of ripples at the
bandedges due to the Gibbs phenomenon caused by the truncation of the impulse response.
10.6

H {x[ n]} =

∑ hHT [n − k ]x[k ] . Hence,

k = −∞

⎧⎪ jX (e jω ), − π < ω < 0,
F {H {x[ n]}} = H HT (e jω ) X (e jω ) = ⎨
⎪⎩− jX (e jω ), 0 < ω < π.
⎧⎪ j 4 X (e jω ),
− π < ω < 0,

=
Y
(
e
)
= X (e jω ).
(a) Let y[ n] = H {H {H {H {x[n]}}}}. Hence,

4
ω
j
⎪⎩(– j ) X (e ), 0 < ω < π ,
Therefore, y[ n] = x[ n].
(b) Define g[ n] = H {x[ n]}, and h * [ n] = x[ n]. Then
But from the Parseval's' relation in Table 3.4,
Therefore,

∑ l = –∞ H {x[l]}x[l] = ∑ l = – ∞ g[l]h * [l].
π

∑ l = –∞ g[l]h * [l] = 2π ∫– π G(e jω )G(e jω )dω.
1

π

∑ l = – ∞ H {x[l]}x[l] = 2π ∫− π H HT (e jω )X (e jω )X (e – jω )dω where
1

⎧ j, − π < ω < 0,
H HT (e jω ) = ⎨
Since the integrand H HT (e jω ) X (e jω ) X (e – jω ) is an odd

<
<
,
0
.
j
ω
π

function of ω,

π

∫− π H HT (e

) X (e jω ) X (e – jω )dω = 0. As a result,

Not for sale

∑ l = –∞ H {x[l]}x[l] = 0.

327

10.7

N

∑ hLP [n] z − n .

H LP ( z ) =

Its frequency response H LP (e jω ) is shown in Figure (a)

n=0

π
2

j (ω − )

π
2

j (ω + )

) + H LP (e
below. A plot of the frequency response H HT (e ) = H LP (e
is shown in Figure (b) below. It is evident from this figure that H HT (e jω ) is the
frequency response of an ideal Hilbert transformer. Therefore, we have

)

HLP (e jω )
1

π
_ __
2

π
__
2

0

π

ω

(a)
HHT (e jω )

π
__

HLP(e j(ω + 2 ) )

HLP(e j(ω

1

π
_ __
2

π
_ __
)
2 )

π
__
2

0

π

ω

(b)

H HT (e

=
=

N

∑ hLP

) = H LP

π
j (ω + )
2 )+
(e

π
− jn (ω + )
2
[ n] e

n=0
N

N

+

H LP (e

∑ hLP

π
2

j (ω − )

)

π
− jn(ω − )
2
[ n] e

n=0

∑ 2 hLP [n]e − jnω cos

n=0

N

=

(

∑ hLP [n]e − jnω e − jnπ / 2 + e jnπ / 2

n=0

)

( ). Now, for n odd, cos(nω / 2) = 0 and hence, we can drop all

2

odd terms in the above expression. Let N = 2 M with N even and let r = 2n. Then, we
can rewrite the above equation as H HT (e jω ) =

M

∑ 2hLP [2r ] cos(rπ)e − j 2rω .

The

r =0

corresponding transfer function of the Hilbert transformer is therefore given by

H HT ( z ) =

10.8

M

∑ 2hLP [2n] cos(nπ) z − 2n

n =0

=

M

∑ 2(−1) n hLP [2n]z − 2n .

n=0

H DIF (e jω )= jω. Hence,
j π
j ⎛⎜ ωe jωn e jωn
1 π
jωn
jωn
h DIF [ n] =
j ωe
dω=
dω =
ωe
+
2π − π
2π − π
2π ⎜⎝ jn
n2
cos(πn) sin(πn) cos(πn)
h DIF [ n] =
, if n ≠ 0.

=
n
n
πn 2

Not for sale

π
−π

. Therefore,

328

1
For n = 0 , h DIF [0] =

π

∫ jωdω = 0.

−π

n = 0,
⎧⎪ 0,
n
cos(
)
π
h
n
[
]
=

Hence, DIF
, | n | > 0. Since hDIF [ n] = −hDIF [ − n], the truncated impulse
⎪⎩ n
response is a Type 3 linear-phase FIR filter. The magnitude responses of the above
differentiator for several values of M are given below:
4
M=10
M=20
M=30

Magnitude

3

2

1

0

0

0.2

0.4

0.6

0.8

1

ω/π

10.9

N = 2 M + 1.

ω

for n = M ,
1− c ,

π

ˆh [ n] = ⎪⎨− sin (ωc (n − m)) , if n ≠ M , 0 ≤ n < N ,
HP
π (n − m)

otherwise.
⎪⎩
0,

Now, Hˆ HP ( z ) + Hˆ LP ( z ) =

=

hˆ HP [ n]z − n +

n = −∞
N −1

hˆ LP [ n]z − n =

n = −∞

N −1

N −1

n=0

n=0

hˆ HP [ n]z − n +

∑ hˆLP [n]z −n

∑ (hˆHP [n] + hˆLP [n])z −n .

n=0

⎧0, 0 ≤ n ≤ N–1, n ≠ M ,
But hˆ HP [ n] + hˆ LP [ n] = hˆ HP [ n] + hˆ LP [ n] = ⎨
,
n=M .
⎩1,
Hence, Hˆ HP ( z ) + Hˆ LP ( z ) = z – M , i.e. the two filters are delay-complementary.

⎧ ω , ω < ωc ,

10.10 H LLP (e ) = ⎨
Therefore,
⎩ 0, otherwise.

Not for sale

329

ωc

0

1 ⎜
j ωn
j ωn
hLLP [ n] =
dω + ωe
dω ⎟
− ωe

2π ⎜

0

⎝ − ωc
ω ⎞
0

⎡ ωe jωn e jωn ⎤ c ⎟
1 ⎜ ⎡ ωe jωn e jωn ⎤
+⎢
+
=
+
⎥ ⎟

⎜ −⎢
2
2π ⎜ ⎢⎣ jn
jn
n 2 ⎥⎦
n
⎥⎦ 0 ⎟

–ωc ⎣

jω n
− jω n
cos(ωc n) − 1
1 ⎛⎜ ωc e c − ωc e c
e jωc n + e − jωc n − 2 ⎞⎟ ωc
=
sin(ωc n) +
.
+
=
2
2

πn
2π ⎜
jn
π
n
n

⎧ω, ω < ωc ,

Hence,
10.11 H BLDIF (e ) = ⎨
⎩ 0, otherwise.
ωc ⎞

ωc

1
1 ⎜ ⎡ ωe jωn e jωn ⎤
jωn
+
h BLDIF [ n] =
dω =
ωe

⎜⎢

2

2π ⎜ ⎣⎢ jn

n

− ωc
−ωc ⎟⎠

1 ⎛⎜ ωc e
=
2π ⎜

jωc n

+ ωc e − jωc n e jωc n − e − jωc n
+
jn
n2

⎟ = – j ωc cos(ω n) + j 1 sin(ω n).
c
c

πn
πn 2

10.12 The frequency response of a causal ideal notch filter can thus be expressed as
(
(
H notch (e jω ) = H notch (ω) e jθ (ω) where H notch (ω) is the amplitude response which can

(
(
⎧ 1, 0 ≤ ω ≤ ωo ,
be expressed as H notch (ω) = ⎨
It follows then that H notch (ω) is
⎩− 1, ωo < ω < π.
(
related to the amplitude response H LP (ω) of the ideal lowpass filter with a cutoff at ωo
(
(
Hence, the impulse response of the ideal
through H notch (ω) = ±[ 2 H LP (ω) − 1].
notch filter is given by hnotch [ n] = ±[2hLP [ n] − δ[ n]], where
sin(ωo n)
, − ∞ < n < ∞. The magnitude responses of a length 41 notch filter
πn
with a notch frequency at ωo = 0.4π and its associated length-41 lowpass filter are shown
below.
hLP [ n] =

Not for sale

330

Amplitude

1.5

1

0.5

0

o

0.2

0

1

0.8

0.6

0.4

= 0.4š

/

1

10.13 Φ R =

( )2 dω , where H t (e jω ) = ∑ ht [n]e − jωn .
n=−M

∫ ( )
−π
π

M

H t e jω − H d e jω

∑ ht [n] − hd [n]

Using Parseval’s relation, we can write Φ R =

2

n = −∞
M

ht [n] − hd [n] +

=

2

n=−M

Now, Φ Haan =

− M −1

n = −∞

n = M +1

∑ hd2 [n] + ∑ hd2 [n] .

∑ hd [n]⋅ w Hann [n] − hd [n]

2

n = −∞

− M −1

⎛1 1
⎛ 2πn ⎞ ⎞
2
=
hd [n]⎜⎜ + cos⎜
hd [n] +
hd2 [n]
⎟ ⎟⎟ − hd [n] +
⎝ 2M + 1⎠ ⎠
⎝2 2
n=−M
n = −∞
n = M +1
Hence, Φ Excess = Φ R − Φ Haan
2

M

M

=

∑ hd [n]⋅ w R [n] − hd [n]

n=−M

=−

2

10.14 Φ R =

M

⎛1 1
⎛ 2πn ⎞ ⎞

hd [n] ⋅ ⎜⎜ + cos⎜
⎟ ⎟⎟ − hd [n]
+
2
1
M
2
2

⎠⎠

n= −M

hd [n] ⎛ 2πn ⎞ hd [n]
cos⎜
⎟−
2
2
1
2
M
+

n= −M
M

2

∑ ht [n] − hd [n] and Φ Hamm =
2

n = −∞

2

2

=

1
(1 + 2 M ) cos⎛⎜ 2πM ⎞⎟ − 1 .
2
⎝ 2M + 1⎠

∑ hd [n]⋅ w Hamm [n] − hd [n]

2

.

n = −∞
M

⎛ 2πn ⎞
Therefore, Φ Excess = Φ R − Φ Hamm = −
hd [n]⎜⎜ 0.46 cos⎜
⎟ − 0.46 ⎟⎟
⎝ 2M + 1 ⎠

n= −M

M

⎛ ⎛ 2πn ⎞ ⎞
=−
0.46hd [n]⎜⎜ cos⎜
⎟ − 1⎟⎟
+
M
2
1

⎠ ⎠

n=−M

2

2

2

⎛ 2πM ⎞
= 0.46(2 M + 1) cos⎜
⎟ −1 .
⎝ 2M + 1 ⎠

Not for sale

331

10.15 (a) ω p = 0.47π , ω s = 0.59π , δ p = 0.001 , δ s = 0.007 , ∆ω = 0.12π ,

α s = −20 log10 δ s = 43.1 dB
From Table 10.2, we see that for fixed-window functions, we can achieve the minimum
stopband attenuation by using Hann, Hamming, or Blackman windows. Hann will have
the lowest filter length.
3.11π
N Haan = ⎡2 M + 1⎤ = 53 since M =
= 25.917.
0.12π
Lowpass filter using Hann window
20
0

Magnitude Response

-20
-40
-60
-80
-100
-120
-140

0

0.1

0.2

0.3

0.4

0.5
ω/π

0.6

0.7

0.8

0.9

1

(b) ω p = 0.61π , ω s = 0.78π , δ p = 0.001 , δ s = 0.002 , ∆ω = 0.17π ,

α s = −20 log10 δ s = 54 dB
From Table 10.2, we see that for fixed-window functions, we can achieve the minimum
stopband attenuation by using either Hamming, or Blackman windows. Hamming will
have the lowest filter length.
3.32π
N Hamm = ⎡2 M + 1⎤ = 41 since M =
= 19.53 .
0.17π

Not for sale

332

Lowpass filter using Hamming window
20

Magnitude Response

0

-20

-40

-60

-80

-100

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

ω/π

10.16 ω p1 = 0.45π , ω p 2 = 0.65π , ω s1 = 0.3π , ω s 2 = 0.8π , ∆ω1 = ∆ω 2 = 0.15π ,

δ p = 0.01 , δ s1 = 0.008 , δ s 2 = 0.05

α s1 = −20 log10 δ s1 = 42 dB, α s 2 = −20 log10 δ s 2 = 26 dB
From Table 10.2, we see that the Hann window will have minimum length and meet the
minimum stopband attenuation.
3.11π
M=
= 21. Therefore, N = 43 .
0.15π

Not for sale

333

Bandpass filter using Hann window
20
0

Magnitude Response

-20
-40
-60
-80
-100
-120
-140

0

0.1

0.2

0.3

0.4

0.5
ω/π

0.6

0.7

0.8

0.9

1

10.17 ω p1 = 0.3π , ω p 2 = 0.8π , ω s1 = 0.45π , ω s 2 = 0.65π , ∆ω1 = ∆ω 2 = 0.15π ,

δ p1 = 0.05 , δ p 2 = 0.009 , δ s = 0.02 ,

α s = −20 log10 δ s = 34 dB
From Table 10.2, we see that the Hann window will have minimum length and meet the
minimum stopband attenuation.

Not for sale

334

Bandstop filter using Hann window
20

0

Magnitude Response

-20

-40

-60

-80

-100

-120

0

0.1

0.2

0.3

0.4

0.5
ω/π

0.6

0.7

0.8

0.9

1

10.18 Consider another filter with a frequency response G(e jω ) given by

0,
0 ≤ ω ≤ ωp ,

π (ω − ω p ) ⎞
⎪ − π sin⎛⎜
⎟,
ω p < ω ≤ ωs ,

⎪ 2∆ω ⎜⎝
∆ω

G (e ) = ⎨
+
(
)
π
ω
ω

p ⎞
⎪ −π

⎟, − ωs ≤ ω ≤ −ω p ,
sin
⎪ 2∆ω ⎜

ω

0
,
elsewhere.

Clearly G(e ) =

dH (e jω )
. Now,

⎛ π (ω − ω p ) ⎞
⎛ π (ω − ω p ) ⎞

⎜ ωs j⎜
ωs − j⎜
⎟ j n

⎟ jωn

π
ω
ω

−π ⎜
⎠ e ω dω

⎠e

1
e
d
ω
e
=

j ω j ωn
g[ n] =
G (e )e
dω 8π∆ωj ⎜

ωp
⎜⎜ ω p
−π

+

−ω p

⎛ π ( ω+ ω p ) ⎞
⎛ π ( ω+ ω p ) ⎞

−ω p − j ⎜

j⎜

∆ω ⎠ jωn
∆ω ⎠ jωn

e
e
dω −
e
e
dω ⎟

− ωs

− ωs

Not for sale

335

π ⎞⎤
π ⎞
π ⎞ ⎤

⎡ jω ⎛⎜ n − π ⎞⎟

jω p ⎜ n +
jω p ⎜ n −
⎟)
jπω p

⎜ jπω p ⎢ jωs ⎜ n + ⎟
s

ω

ω⎠ ⎥
ω
ω

e
−e
−1 ⎜
−e

⎥ − e ∆ω ⎢ e
e ∆ω ⎢
=

π ⎞
π ⎞
j8∆ω

j⎜ n +
j⎜ n −

∆ω ⎠
∆ω ⎠

π ⎞⎤
π ⎞ ⎤⎞

⎡ − jω ⎛⎜ n + π ⎞⎟
⎡ − jω ⎛⎜ n − π ⎞⎟
− jω s ⎜ n −
− jω s ⎜ n +
jπω p
jπω p
⎟ ⎟

p
p

⎝ ∆ω ⎠ ⎥ ⎟
⎝ ∆ω ⎠ − e
⎝ ∆ω ⎠ ⎥
⎝ ∆ω ⎠ − e

e
e
⎥ − e ∆ω ⎢

+ e ∆ω ⎢

⎥⎟
π
π

+

j
n
j
n

⎥ ⎟⎟

ω

ω

⎦⎠

− 1 ⎜ − sin(ωs n) − sin(ω p n) − sin(ωs n) − sin(ω p n) ⎟ = sin(ωs n) + sin(ω p n) ⎜ − 2π / ∆ω
=

4∆ωj
π ⎞
π ⎞
4∆ωj ⎜

π2
2

⎜n +

⎜n −

⎜n −
∆ω ⎠
∆ω ⎠

∆ω 2

sin(ωc n) cos(∆ωn / 2) ⎛⎜
1
⎟.
=
⎜ π 1 − (∆ω / π ) 2 n 2 ⎟
∆ωj

⎛ cos(∆ωn / 2) ⎞⎛ sin(ωc n) ⎞
j

⎟⎜
h
n
=
[
]
⎟.
Now, h[ n] = g[ n]. Therefore,
⎜ 1 − (∆ω / π ) 2 n 2 ⎟⎝ πn ⎠
n

(

(

)

)

⎛ 4πn ⎞ ⎤
⎛ 2πn ⎞
⎟ ⎥ w R [ n]
⎟ + γ cos⎜
10.19 wGC [ n] = ⎢α + β cos⎜
⎝ 2 M + 1 ⎠⎦
⎝ 2M + 1 ⎠

⎛ 2 πn ⎞ ⎞
⎛ 4 πn ⎞ ⎞ ⎤
⎛ j ⎛⎜ 2πn ⎞⎟
⎛ j ⎛⎜ 4πn ⎞⎟
− j⎜
− j⎜
⎟⎟
⎟ ⎟⎥

= ⎢ α + 2 β ⎜ e ⎝ 2 M + 1 ⎠ + e ⎝ 2 M + 1 ⎠ ⎟ + 2 γ⎜ e ⎝ 2 M + 1 ⎠ + e ⎝ 2 M + 1 ⎠ ⎟ ⎥ w R [ n ]

⎟⎥

⎠⎦

⎛ j ⎛⎜ ω − 2 π ⎞⎟ ⎞
⎛ j ⎛⎜ ω + 2 π ⎞⎟ ⎞

Hence, ΨGC (e jω ) = αΨR (e jω ) + 2 βΨR ⎜ e ⎝ 2 M + 1 ⎠ ⎟ + 2 βΨR ⎜ e ⎝ 2 M + 1 ⎠ ⎟

⎛ j ⎛⎜ ω − 4π ⎞⎟ ⎞
⎛ j ⎛⎜ ω + 4π ⎞⎟ ⎞

2 M +1 ⎠
2 M +1 ⎠

+ 2 γΨR ⎜ e
⎟ + 2 γΨR ⎜ e ⎝
⎟.

For the Hann window :

ΨHaan (e

) = 0.5ΨR (e

= 0.5,

= 0.5 and

= 0. Hence,

⎛ j ⎛⎜ ω − 2π ⎞⎟ ⎞
⎛ j ⎛⎜ ω + 2 π ⎞⎟ ⎞

2
+
1
2 M +1 ⎠
M
⎠ + 2 βΨ e ⎝
) + ΨR ⎜ e ⎝

R⎜

Not for sale

336

⎛ (2 M + 1)ω ⎞ sin⎛⎜ (2 M + 1)⎛ ω − π ⎞ ⎞⎟ sin⎛⎜ (2 M + 1)⎛ ω + π ⎞ ⎞⎟
⎟⎟

⎟⎟

sin⎜

2
⎝ 2 2M + 1 ⎠ ⎠
⎝ 2 2M + 1 ⎠ ⎠

.
+
+
= 0.5
sin(ω / 2)
π ⎞
π ⎞
⎛ω
⎛ω
sin⎜ −
sin⎜ +

⎝ 2 2M + 1 ⎠
⎝ 2 2M + 1 ⎠
For the Hamming window,

= 0.54,

= 0.46, and

= 0. Hence,

⎛ j ⎛⎜ ω − 2 π ⎞⎟ ⎞
⎛ j ⎛⎜ ω + 2π ⎞⎟ ⎞

ΨHam min g (e jω ) = 0.54ΨR (e jω ) + 0.92ΨR ⎜ e ⎝ 2 M + 1 ⎠ ⎟ + 0.92ΨR ⎜ e ⎝ 2 M + 1 ⎠ ⎟

π ⎞⎞
π ⎞⎞
⎛ω
⎛ω
⎛ (2 M + 1)ω ⎞
sin⎜⎜ (2 M + 1)⎜ +
sin⎜⎜ (2 M + 1)⎜ −
⎟ ⎟⎟
⎟ ⎟⎟
sin⎜

2
⎝ 2 2M + 1⎠ ⎠
⎝ 2 2M + 1 ⎠ ⎠

.
+ 0.92
+ 0.92
= 0.54
F
sin(ω / 2)
π ⎞
π ⎞
⎛ω
⎛ω
sin⎜ +
sin⎜ −

⎝ 2 2M + 1 ⎠
⎝ 2 2M + 1⎠
or the Blackmann window
= 0.42, = 0.5 and = 0.08
⎛ j ⎛⎜ ω − 2π ⎞⎟ ⎞
⎛ j ⎛⎜ ω + 2 π ⎞⎟ ⎞

2
+
1
2 M +1 ⎠
M
⎠ +Ψ e ⎝
ΨBlackman (e ) = 0.42ΨR (e ) + ΨR ⎜ e ⎝

R⎜

⎛ j ⎛⎜ ω− 4π ⎞⎟ ⎞
⎛ j ⎛⎜ ω− 4π ⎞⎟ ⎞

+ 0.16ΨR ⎜ e ⎝ 2 M +1 ⎠ ⎟ + 0.16ΨR ⎜ e ⎝ 2 M +1 ⎠ ⎟

⎛ (2 M + 1)ω ⎞ sin⎛⎜ (2 M + 1)⎛ ω − π ⎞ ⎞⎟ sin⎛⎜ (2 M + 1)⎛ ω + π ⎞ ⎞⎟
⎟⎟

⎟⎟

sin⎜

2
⎝ 2 2M + 1 ⎠⎠
⎝ 2 2M + 1 ⎠ ⎠

+
+
= 0.42
sin(ω / 2)
π ⎞
π ⎞
⎛ω
⎛ω
sin⎜ −
sin⎜ +

⎝ 2 2M + 1 ⎠
⎝ 2 2M + 1 ⎠

2π ⎞ ⎞
2π ⎞ ⎞
⎛ω
⎛ω
sin⎜⎜ (2 M + 1)⎜ −
sin⎜⎜ (2 M + 1)⎜ −
⎟ ⎟⎟
⎟ ⎟⎟
⎝ 2 2M + 1⎠ ⎠
⎝ 2 2M + 1⎠ ⎠

+ 0.16
+ 0.16
.
2π ⎞
2π ⎞
⎛ω
⎛ω
sin⎜ −
sin⎜ −

⎝ 2 2M + 1 ⎠
⎝ 2 2M + 1⎠
N

10.20 (a) H ( z ) = z − D ≈ ∑ h[n ]z − n = h[0] + h[1]z −1 + h[2]z − 2 + L + h[N ]z − N
n =0

We see that if xˆ a (t ) =

N2

N2

⎛ t − tl
⎜⎜
l = − N1 ⎝ t k − t l

∑ Pk (t )x[n + k ], then Pk (t ) = ∏

k = − N1

⎟⎟ for − N1 ≤ k ≤ N 2 .

l≠k

Not for sale

337

Here, we have Hˆ (z ) =

N

∑ h[n]z −n and the solution follows if

n=0

Pk (t ) = h[n] , N1 = 0 , N 2 = N , k = n , t = D , t l = k , and t k = n
Therefore, we have h[n] =

N

D−k

∏ n−k

for 0 ≤ n ≤ N .

k =0

k≠n

(b) N = 21, D = 90/13, L = 22. H (z ) ≈

21

h[n]z − n , where h[n] =

n=0

21

90 / 13 − k
.
n

k
k =0

k≠n

% Problem #10.20
D = 90/13;
N = 21;
for n = 0:N,
for k = 0:N,
if n ~= k,
tmp(n+1,k+1) = (D-k)/(n-k);
else
tmp(n+1,k+1) = 1;
end
end
end
h = prod(tmp');
[Gd,W] = grpdelay(h,1,512);
[H, w] = freqz(h,1,512);
figure(1);
plot(W/pi, Gd);
xlabel('\omega/\pi');
ylabel('Group Delay');
title('Group delay of z^-^D');
grid;
figure(2);
plot(w/pi, (abs(H)));
xlabel('\omega/\pi');
ylabel('Magnitude');
title('Magnitude response of z^-^D');
grid;

Not for sale

338

Group delay of z

Magnitude response of z

-D

-D

1.5

8

Magnitude

Group Delay

1.4
7.5

7

1.3
1.2
1.1
1

6.5

0.9
0

0.2

0.4

0.6

0.8

1

0

0.2

0.4

10.21 (a) x[ n] = s[ n] +

0.6

0.8

1

ω/π

ω/π

M

∑ Ak sin(kωo n +φ k ) = s[n] + r[n], where s[n] is the desired signal and

k =0

r[ n] =

M

∑ Ak sin(kωo n +φ k ) is the harmonic interference with fundamental

k =0

frequency ωo . Now, r[ n − D] =

M

∑ Ak sin[kωo (n − D) +φ k ]

k =0

=

M

∑ Ak sin(kωo n +φ k −2πk ) = r[n].

k =0

(b) y[ n] = x[ n] − x[ n − D] = s[ n] + r[ n] − s[ n − D] − r[ n − D] = s[ n] + r[ n] − s[ n − D] − r[ n]

= s[ n] − s[ n − D]. Hence, y[ n] does not contain any harmonic disturbances.
(c) H c ( z ) =

(1 − cos( Dω)) + j sin( Dω) .
1 − e − jDω
=
. Thus, H c (e jω ) =
−D −D
− D − jDω
1 − ρ D cos( Dω) + jρ D sin( Dω)
1− ρ e
1− ρ z
1 − z −D

Then, H c (e jω ) =

(

2(1 − cos( Dω))
D

1 − 2 ρ cos( Dω) + ρ
ρ = 0.99 is shown below:

2D

Not for sale

)

. A plot of H c (e jω ) for ωo = 0.22π and

339

Amplitude

1
0.8
0.6
0.4
0.2
0
-1

-0.5

0

0.5

1

ω/π

(d)
y[n]

x[n]
z–D

+

P( z)
1 − N ( z)
, with D = 2π / 0.18π , N = 18, and ρ = 0.98 .
=
Q( z) 1 − 0.98 D N ( z)
% Problem #10.23
close all;
clear;
clc;
D = 2/.18;
N = 18;
rho = 0.98;
for n = 0:N,
for k = 0:N,
if n ~= k,
tmp(n+1,k+1) = (D-k)/(n-k);
else
tmp(n+1,k+1) = 1;
end
end
end
h = prod(tmp');
[H,w] = freqz(h,1,1024);
Hc = (1-H)./(1-(rho^D)*H);
x = sqrt((2*(1-cos(D*w)))./(12*(rho^D)*cos(D*w)+rho^(2*D)));
%plot(w/pi, abs(Hc));
plot(w/pi, x); grid;
xlabel('\omega/\pi');
ylabel('Magnitude');
title(‘Comb filter using FIR fractional delay');

10.22 H c ( z) =

Not for sale

340

Comb filter using FIR fractional delay

Magnitude

1
0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/π

P( z)
D( z) − z −11 D( z −1 )
=
, with D = 2π / 0.18π = 11.11 and N = 11.
Q( z) D( z) − 0.98 90 / 13 z −11 D( z −1 )
% Problem #10.23
close all;
clear;
clc;
D = 2/0.18;
rho = 0.98;
N = floor(D);
for k = 1:N,
for n = 0:N,
p(n+1) = (D-N+n)/(D-N+k+n);
end
d(k) = ((-1)^k)*nchoosek(N,k)*prod(p);
end
[H,w] = freqz(fliplr(d)-d, fliplr(d)-(rho^D).*d , 512);
plot(w/pi, abs(H));grid;
xlabel('\omega/\pi');
ylabel('Magnitude');
title(Comb filter using allpass IIR fractional delay');

10.23 H c ( z) =

Comb filter using allpass IIR fractional delay

Magnitude

1
0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/π

Not for sale

341

−ω p ≤ ω ≤ ω p ,
1,

⎪ ⎛
⎪1 − ⎜ ω − ω p ⎞⎟, ω < ω ≤ ω ,
p
s
⎪⎪ ⎜ ωs − ω p ⎟

=
(
)

10.24 H LP e
⎪ ⎛⎜ ω + ω p ⎞⎟
⎪1 + ⎜ ω − ω ⎟, −ωs < ω ≤ ω p ,
p⎠
⎪ ⎝ s
⎪⎩
0,
elsewhere.

Now, for n ≠ 0, hLP [ n] =

1

π

∫ H LP (e

)e jωn dω

−π

−ωp
⎡ ωp
ωs
ω − ω p ⎞ jωn
ω + ωp

1 ⎢
jωn
⎜1 −
⎟e
⎜1 +
=
+
+
e
d
ω
d
ω

2π ⎢
∆ω ⎟⎠
∆ω
− ωs ⎝
ωp ⎝
⎢⎣ − ω p
−ω p
⎡ ωs

ωs
ω − ω p jωn
ω + ω p jωn ⎥
1 ⎢
jωn
=
e
dω −
e
dω +
e
dω⎥
2π ⎢
∆ω
∆ω
− ωs
ωp
⎢⎣ − ωs
⎥⎦

⎞ jωn ⎥
⎟e
dω⎥

⎥⎦

ωs
– ωp ⎫

jωn
jωn
⎪⎪
e jωn ⎤⎥
e jωn ⎤⎥
1 ⎪⎪ 2sin(ωs n)
1 ⎡⎢ (ω − ω p )e
1 ⎡⎢ (ω + ω p )e
+
+
=

+

2 ⎥
2 ⎥
jn
jn
2π ⎪
∆ω ⎢
∆ω ⎢
πn
n
n
⎦ – ωs ⎪⎪

⎦ ωp

⎪⎩

jω n
− jω p n

− ∆ωe − jωs n e
1 ⎪ 2sin(ωs n)
− e − jωs n ⎤ ⎪
1 ⎡ ∆ωe jωs n e jωs n − e p

=

+
+

2
2
2π ⎪
jn
πn
∆ω ⎢
jn
⎥⎪
n
n

⎦⎭

=

1 ⎧⎪ 2sin(ωs n)
2 ⎡ ∆ω sin(ωs n) cos(ωs n) − cos(ω p n) ⎤ ⎫⎪

+
⎥⎬

2π ⎪⎩
πn
∆ω ⎣⎢
n
n2
⎦⎥ ⎪⎭

1 ⎧ cos((ωc − ∆ω / 2)n) cos((ωc + ∆ω / 2)n) ⎫
1 ⎧⎪ cos(ω p n) cos(ωs n) ⎫⎪
=

∆ω ⎪⎩ πn 2
πn 2
πn 2

πn 2 ⎪⎭ ∆ω ⎩
2 sin(∆ωn / 2) sin(ωc n)
=
.
∆ωn
πn

=

1
Next, for n = 0, h LP [0] =

=

(

)

π

∫ H LP (e

) dω =

−π

1
(area under the curve)

ω
1 2 ωs + ω p
= c.
π

2

Not for sale

342

ωc

,
if n = 0,

π
h
n
[
]
=
⎨ 2sin(∆ωn / 2) sin(ω n)
Hence, LP
c

, if n ≠ 0.
∆ωn
πn

An alternate approach to solving this problem is as follows. Consider the frequency response
⎧ 0, − ω p ≤ ω ≤ ω p ,
⎪ 1

d H LP (e ) ⎪⎪ – ∆ω , ω p < ω < ωs ,
G ( e jω ) =
=⎨
Its inverse DTFT is given by

⎪ 1 , − ωs < ω < − ω p ,
⎪ ∆ω
⎪⎩
0, elsewhere.
1
g[ n] =

π

∫ G (e

)e

jωn

–π

1
dω =

– ωp

– ωs

1 jωn
1
e
dω −
∆ω

ωs

ωp

1 jωn
e

∆ω

−ωp
ωs ⎞

e jωn
1 ⎜ e jωn
1

⎟ =
=

cos(ω p n) − cos(ωs n) .
jn
π
2 ∆ω ⎜
jn
j
n

π
ω

− ωs
ω p ⎟⎠

1
j
cos(ω p n) − cos(ωs n)
Thus, hLP [ n] = g[ n] = 2
n
πn ∆ω
⎛⎛
1 ⎡ ⎛⎛
∆ω ⎞ ⎞
∆ω ⎞ ⎞ ⎤
cos⎜⎜ ⎜ ωc −
=
⎟n ⎟⎥
⎟ n ⎟⎟ − cos⎜⎜ ⎜ ωs +

2 ⎠ ⎠
2 ⎠ ⎟⎠ ⎦
⎝⎝
πn 2 ∆ω ⎣ ⎝ ⎝

(

(

=

)

)

2sin(∆ωn / 2) sin(ωc n)

, for n ≠ 0.
πn
∆ωn

For n = 0, hLP [ n] =

ωc
.
π

10.25 Consider the case when the transition region is approximated by a second order spline. In
this case the ideal frequency response can be constructed by convolving an ideal, notransition-band frequency response with a triangular pulse of width ∆ω = ωs − ω p ,

which in turn can be obtained by convolving two rectangular pulses of width ∆ω / 2 . In
the time domain this implies that the impulse response of a filter with transition band
approximated by a second order spline is given by the product of the impulse response of
an ideal low pass filter with no transition region and square of the impulse response of a
rectangular pulse. Now,

Not for sale

343

H LP (ideal ) [ n] =

sin(ωc n)
sin(∆ωn / 4)
and H rec [ n] =
. Hence,
πn
∆ωn / 4

H LP [ n] = H LP(ideal) [ n](H rec [ n])2 .

Thus for a lowpass filter with a transition region approximated by a second order spline
ωc

if n = 0,
,
⎪⎪
π
hLP [ n] = ⎨
2
⎪⎛⎜ sin(∆ωn / 4) ⎞⎟ sin(ωc n) , otherwise.
⎪⎩⎝ ∆ωn / 4 ⎠
πn
Similarly the frequency response of a lowpass filter with the transition region specified by a
P -th order spline can be obtained by convolving in the frequency domain an ideal filter with no
transition region with P rectangular pulses of width ∆ω / P. Hence,
H LP [ n] = H LP(ideal) [ n](H rec [ n]) P , where the rectangular pulse is of width ∆ω / P. Thus
ωc

,
if n = 0 ,
⎪⎪
π
hLP [ n] = ⎨
P
⎪⎛⎜ sin(∆ωn / 2 P ) ⎞⎟ sin(ωc n) , otherwise.
πn
⎩⎪⎝ ∆ωn / 2 P ⎠
(
(
10.26 From Step 2, we have G (e jω ) = G (ω)e − jNω = δ s( F ) e − jNω + F (ω)e − jNω .
(
The amplitude response G (ω) has been obtained by raising the amplitude response
(
F (ω) by δ s( F ) and hence, the filter G (z ) has double zeros in the stopband.

Not for sale

344

We may factorize as follows: G ( z ) = z − N H m ( z ) H m ( z −1 ), where H m (z ) is a real-coefficient
(
minimum-phase FIR lowpass filter with half the degree of the original H (z ). Since, G (ω) ≥ 0,
(
the amplitude response H m (ω) of the minimum-phase filter H m (z ) does not oscillate about
(
unity in the passband. Since the original frequency response was raised by δ s( F ) , H m (ω) must be
normalized by a factor 1 + δ s . Therefore,

(
For H m (ω) , we can see δs(F ) =

1 + δ p + δs
δp
2δ s
and δ (pF ) =
−1 = 1+
−1.
1 + δs
1 + δs
1 + δs

10.27 (a) N = 1 and hence, x a (t ) = α 0 + α1t. Without any loss of generality, for L = 5. we first
fit the data set {x[ k ]}, − 5 ≤ k ≤ 5, by the polynomial x a (t ) = α 0 + α1t with a minimum
mean-square error at t = −5, − 4,K , − 1, 0,1,K ,5, , and then replace x[ 0] with a new value
x[ 0 ] = x ( 0 ) = α 0 .

Now, the mean-square error is given by ε(α0 , α1 ) =

5

∑ (x[k ] − α0 − α1k )2 . We set

k = −5

5
5
∂ ε(α 0 , α1 )
∂ ε(α 0 , α1 )
= 0 and
= 0 which yields 11α 0 + α1 ∑ k = ∑ x[ k ], and
∂α 0
∂α1
k = −5 k = −5
5

5

5

k = −5

k = −5

k = −5

α0 ∑ k + α1 ∑ k 2 = ∑ k x[ k ].
From the first equation we get x[0] = α 0 =

5

1
x[ k ]. In the general case we thus
11 k = −5

Not for sale

345

5

1
have x[ n] = α0 =
x[ k ] which is a moving average filter of length 11.
11 k = −5
(b) N = 2, and hence, x a (t ) = α 0 + α1t + α 2 t 2 . Here, we fit the data set {x[ k ]},
− 5 ≤ k ≤ 5, by the polynomial x a (t ) = α 0 + α1t + α 2 t 2 with a minimum mean-square

error at t = −5, − 4,K , − 1, 0,1,K ,5, and then replace x[ 0] with a new value
x[0] = x a (0) = α 0 . The mean-square error is now given by
5

∑ (x[k ] − α0 − α1k − α2 k 2 ) .

ε (α 0 , α1 , α 2 ) =

2

We set

k = −5

∂ ε(α 0 , α1 , α 2 )
= 0,
∂α 0

∂ ε(α 0 , α1 , α 2 )
∂ ε(α 0 , α1 , α 2 )
= 0, and
= 0, which yields
∂α1
∂α 2
5

11α 0 + 110α 2 =

5

x[ k ], 110α1 =

k = −5

k x[ k ], 110α 0 + 1958α 2 =

k = −5

5

∑ k 2 x[k ].

From the

k = −5

first and the third equations we then get
5

1958

x[ k ] − 110

k = −5

5

∑ k 2 x[k ]

k = −5
(1958 × 11) − (110) 2

α0 =

5

∑(

)

1
=
89 − 5k 2 x[ k ].
429 k = −5

Hence, here we replace x[n] with a new value x[ n] = α0 which is a weighted combination of
the original data set {x[ k ]}, − 5 ≤ k ≤ 5 :
5

∑(

)

1
x[ n ] =
89 − 5k 2 x[ n − k ]
429 k = −5
=

1
(−36 x[ n + 5] + 9 x[ n + 4] + 44 x[ n + 3] + 69 x[ n + 2] + 84 x[ n + 1] + 89 x[ n]
429
+84 x[ n − 1] + 69x[ n − 2] + 44 x[ n − 3] + 9 x[ n − 4] − 36 x[ n − 5]).

(c) The impulse response of the FIR filter of Part (a) is given by
1
h1[ n] = {1 1 1 1 1 1 1 1 1 1 1},
11

whereas, the impulse response of the FIR filter of Part (b) is given by
1
h1[ n] =
{− 36 9 44 69 84 89 84 69 44 9 − 36}.
429
The corresponding frequency responses are given by
H1

( e jω ) =

5

5

∑(

)

1
1
e − jωk , and H 2 (e jω ) =
89 − 5k 2 e − jωk .
11 k = −5
429 k = −5

Not for sale

346

A plot of the magnitude responses of these two filters are shown below from which it can
be seen that the filter of Part (b) has a wider passband and thus provides smoothing over a
larger frequency range than the filter of Part (a).
1
H1(z)
H2(z)

Magnitude

0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/π

10.28 y[ n] =

1
{−3 x[ n − 7] − 6 x[ n − 6] − 5 x[n − 5] + 3 x[ n − 4] + 21x[ n − 3] + 46 x[ n − 2]
320
+ 67 x[ n − 1] + 74 x[ n] + 67 x[ n + 1] + 46 x[ n + 2] + 21x[ n + 3] + 3 x[ n + 4]

+ 5 x[ n + 5] − 6 x[ n − 6] − 3 x[ n + 7]}.

{

Y ( e jω )
1
=
− 3e − j 7ω − 6e − j 6ω − 5e − j 5ω + 3e − j 4ω + 21e − j 3ω
X (e jω ) 320
46e − j 2ω +67e − jω + 74 +67e jω + 46e j 2ω + 21e j 3ω + 3e j 4ω − 5e j 5ω − 6e j 6ω − 3e j 7ω .
1
{74 + 67cos ω + 46cos(2ω) + 21cos(3ω) + 3cos(4ω)– 5cos(5ω) – 6cos(6ω) – 3cos(7ω)}
=
160
– 5cos(5ω) – 6cos(6ω) – 3cos(7ω)}.

Hence, H 3 (e jω ) =

}

The magnitude response of the above FIR filter H 3 ( z) is plotted below (solid line) along
with that of the FIR filter H 2 ( z) of Part (b) of Problem 10.27 (dashed line). Note that
both filters have roughly the same passband but the Spencer's filter has very large
attenuation in the stopband and hence it provides better smoothing than the filter of Part
(b).
1
H3(z)
H2(z)

Magnitude

0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/π

Not for sale

347

10.29 (a) L = 3. P ( x ) = α1 x + α 2 x 2 + α 3 x 3 . Now P (0) = 0 is satisfied by the way P (x ) has
been defined. Also to ensure P (1) = 1 we require α1 + α 2 + α3 = 1 . Choose m = 1 and

n = 1. Since

dP ( x )
= 0, hence α1 + 2α 2 x + 3α3 x 2
= 0, implying α1 = 0 . Also
x =0
dx x = 0

dP( x )
= 0, hence α1 + 2α 2 + 3α3 = 0 . Thus solving the three equations:
dx x =1
α1 + α 2 + α3 = 1 , α1 = 0 ,and α1 + 2α 2 + 3α3 = 0

since

we arrive at

1

0,

2

3 , and

–2. Therefore, P ( x ) = 3 x 2 − 2 x 3 .

3

(b) L = 4. Hence, P ( x ) = α1 x + α 2 x 2 + α 3 x 3 + α 4 x 4 . Choose m = 2 and n = 1
(alternatively one can choose m = 1 and n = 2 for better stopband performance ). Then,
P (1) = 1 ⇒ α1 + α 2 + α3 + α 4 = 1. Also,

dP( x )
= 0 ⇒ α1 + 2α2 x + 3α3 x 2 + 4α 4 x 3 x = 0 = 0,
dx x = 0
d 2 P( x )
= 0 ⇒ 2α2 + 6α3 x + 12α 4 x 2 x = 0 = 0,
2
dx
x=0
dP( x )
= 0 ⇒ α1 + 2α2 + 3α3 + 4α 4 = 0.
dx x =1
Solving the above four simultaneous equations we get

1

0,

2

0,

3

4, and

4

– 3.

Therefore, P ( x ) = 4 x 3 − 3 x 4 .
(c) L = 5. Hence P ( x ) = α1 x + α 2 x 2 + α3 x 3 + α 4 x 4 + α 5 x 5 . Choose m = 2 and n = 2 .
Following a procedure similar to that in parts (a) and (b) we get 1 0, 2 0, 3 10,
4 –15, and
5 6.
M
(
10.30 From Eq. (7.102) we have H (ω) =
c[ k ] sin(ωk ). Now

k =1
M

M
(
H (π − ω) =
c[ k ] sin((π − ω)k ) = −

k =1

c[ k ] sin(ωk ) cos(π k ) =

k =1

M

∑ c[k ](−1) k +1 (ωk ).

k =1

M
M
(
(
Thus, H (ω) = H (π − ω) implies
c[ k ] sin(ωk ) =
c[ k ] (−1) k + 1 sin(ωk ), or

k =1

k =1

M

equivalently,
k = 2, 4, 6,K.

∑ (1 − (−1) k +1 )c[k ] sin(ωk ) = 0, which in turn implies that c[k ] = 0 for

k =1

Not for sale

348

1
2

But from Eq. (7.103) we have c[ k ] = 2 h[ M − k ], 1 ≤ k ≤ M , or, h[ k ] = c[ M − k ]. For k even,
1
2

i.e., k = 2 R, h[2 R ] = c[ M − 2 R] = 0 if M is even.
10.31 (a) H [ k ] = H (e jωk ) = H (e j 2πk / M ), 0 ≤ k ≤ M − 1. Thus, h[ n] =

M −1

1
− kn ,
H [ k ] WM
M k =0

where WM = e − j 2πk / M .
Now, H ( z) =

M −1

M −1 M −1
M −1
⎛ M −1

1
− kn z − n = 1
− kn z − n ⎟.
H [ k ] WM
H[ k ]⎜
WM

M n=0 k =0
M k =0
⎝ n=0

∑ ∑

h[ n] z − n =

n=0
M −1

We can write

=

− kn z − n
WM
n=0

− kn z − n
WM
n=0

=

− kn z − n − W − kM z − M
WM
M

n=0

∑ WM− kn z − n

n= M

(

− kn z − n = 1 − z − M
WM

n=0

)∑ WM− kn z − n = 1 −1 W− z− k z −1 .
−M

n=0

M

M −1

H[ k ]
1 − z− M
Therefore, H ( z) =
.
− k z −1
M

W
1
M
k =0

(b)
1
1 z 1

x[n]

1 z M

H[0]
M
H[1]
M

1
1 z 1e j2 / M

1
1 z

y[n]

H[M 1]
M

1e j2 (M 1)/ M

M −1
M −1

⎛ M −1
1− z−M
H[ k ]
1
− kn z − n ⎟.
H[ k ] ⎜
WM
=
(c) Note H ( z) =
− k −1

M
M k =0
k = 0 1 − WM z

⎝ n=0

Not for sale

349

⎛ M −1

− kn e − jωn ⎟. For
H[ k ] ⎜
WM

M k =0
⎝ n=0

M −1
M

1

⎛ 1
− kn e − 2πln / M ⎟
H[ k ] ⎜
WM
ω = j 2πl / M , we then get from the above H (e j 2πl / M ) =

⎜M
k =0

⎝ n=0
M −1
M

1
M

1
M

1

⎛ 1
⎛ 1
− ( k − l)n ⎟
− knW ln ⎟ =

H[ k ] ⎜
WM
H
[
k
]
W
. Using the identity of Eq.
=
M ⎟
M

⎜M
⎜M
k =0
⎠ k =0

⎝ n=0
⎝ n=0

1
On the unit circle the above reduces to H (e jω ) =

M −1

M −1
if l = k,
1
− ( k − l )n ⎧1,
=⎨
WM
(5.23) of text we observe that
M n=0
⎩0, otherwise.
Hence, H (e j 2 πl / M ) = H [ l].

10.32 (a) For Type 1 FIR filter,

H ( e jω ) =

e

−j

ω( M − 1)
2

H (e jω ) . Since in the frequency

2πk
,
M
0 ≤ k ≤ M − 1, therefore H [ k ] = H (e j 2πk / M ) = H d (e j 2πk / M ) e j 2πk ( M −1) / M ,

sampling approach we sample the DTFT H (e jω ) at M points given by ω =

0 ≤ k ≤ M − 1. Since the filter is of Type 1, M − 1 is even, thus, e j 2πk ( M −1) / 2 = 1. .
Moreover, h[n] being real, H (e jω ) = H * (e jω ). Thus, H (e jω ) = e jω( M −1) / 2 H (e jω ) ,

π ≤ ω < 2π. Hence,
⎧ H (e j 2 πk / M ) e − j 2 πk ( M −1) / 2 M ,
d

H[ k ] = ⎨
⎪ H d (e j 2πk / M ) e − j 2 π ( M − k )( M −1) / 2 M

k = 0,1, 2,K,
k=

M −1
,
2

M +1 M + 3
,
, K , M − 1.
2
2

(b) For the Type 2 FIR filter
M −1
⎧ H (e j 2πk / M ) e − j 2πk ( M −1) / 2 M ,
,
k = 0,1, 2,K,
d

2

M
H[ k ] = ⎨
0,
k= ,
2

M
j
k
M
j
M
k
M
M
π

π

2
/
2
(
)(
1
)
/
2
)e
, k = + 1,K, M − 1.
⎪ H d (e
2

10.33 (a) ω p = 0.55π = 1.72788. The frequency spacing between 2 consecutive DFT samples

is given by
ω = 2π

19

5
and
19

= 0.3307. The desired passband edge is between the frequency samples at
ω = 2π

6
.
19

Therefore, the 19-point DFT is given by

Not for sale

350

⎧e − j (2 π / 19) 9k , k = 0,1,K,5,13,14,K,18,
H[k ] = ⎨
0,
k = 6,K,12.

A 19-point IDFT of the above DFT samples yields the impulse response coefficients
given below in ascending powers of z −1 :

Columns 1 through 10
-0.0037
-0.0022
0.0224
-0.0316
-0.1128
0.2888

-0.0211
0.6316

Columns 11 through 19
0.2888
-0.1128
-0.0316
0.0224
-0.0022
-0.0037

0.0674

-0.0231

-0.0231

0.0674

-0.0211

(b)

Magnitude

1
0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/π

% Problem #10.33
close all;
clear;
clc;
ind = 1;
for k = 0:5,
H(ind) = exp(-i*2*pi*9*k/19);
ind = ind + 1;
end
for k = 6:12,
H(ind) = 0;
ind = ind + 1;
end
for k = 13:18,
H(ind) = exp(-i*2*pi*9*k/19);
ind = ind + 1;
end
h = ifft(H);
figure(1);
stem(real(h));
[FF, w] = freqz(h, 1, 512);

Not for sale

351

figure(2);
plot(w/pi, abs(FF)); axis([0 1 0 1.2]);grid;
ylabel('Gain, in dB'); xlabel('\omega/\pi');
10.34 (a) ω p = 0.35π = 1.09956. The frequency spacing between 2 consecutiveDFT samples

is given by

39

= 0.1611. The desired passband edge is between the frequency samples

6
7
and ω = 2π . Therefore, the 39-point DFT is given by
39
39
⎧e − j (2 π / 39)19 k , k = 0,1,K,6, 32, 33,K,38,

at ω = 2π

H[k ] = ⎨
0,
k = 7,K,31.

A 39-point IDFT of the above DFT samples yields the impulse response coefficients
given below in ascending powers of z −1 :

Columns 1 through 10
0.0006
0.0031
0.0017
0.0128
0.0146
-0.0034

-0.0054
-0.0237

Columns 11 through 20
-0.0192
0.0134
0.0405
-0.0249
0.1222
0.2870
Columns 21 through 30
0.2870
0.1222
-0.0249
0.0405
0.0134
-0.0192

-0.0091

0.0227
0.3590
-0.0753
-0.0237

Columns 31 through 39
-0.0034
0.0146
0.0128
0.0017
0.0031
0.0006

-0.0362

-0.0362

-0.0010

-0.0091

-0.0010

-0.0753

0.0227

-0.0054

(b)

Magnitude

1
0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/π

% Problem #10.34
close all;
clear;
clc;
ind = 1;
for k = 0:6,

Not for sale

352

H(ind) = exp(-i*2*pi*19*k/39);
ind = ind + 1;
end
for k = 7:31,
H(ind) = 0;
ind = ind + 1;
end
for k = 32:38,
H(ind) = exp(-i*2*pi*19*k/39);
ind = ind + 1;
end
h = ifft(H);
figure(1);
stem(real(h));
[FF, w] = freqz(h, 1, 512);
figure(2);
plot(w/pi, abs(FF)); axis([0 1 0 1.2]); grid;
xlabel('\omega/\pi'); ylabel('Gain, dB');
10.35 By expressing cos(ωn) = Tn (cos ω), where Tn (x ) Tn(x) is the n -th order Chebyshev
polynomial in x, we first rewrite Eq. (10.48) in the form:
M
M
(
H (ω) =
a[ n] cos(ωn) =
α n cos n (ω).

n=0

n=0

Therefore, we can rewrite Eq. (10.70) repeated below for convenience
P(ωi )[H (ωi ) − D(ωi )] = (−1)i ε, 1 ≤ i ≤ M + 2,
in a matrix form as
⎡1
⎤ ⎡ α 0 ⎤ ⎡ D(ω1 ) ⎤
cos(ω1 )
cos M (ω1 )
1 / P (ω1 )
L

⎥⎢

⎥ ⎢
L
− 1 / P (ω2 )
cos(ω2 )
cos M (ω2 )
⎢1
⎥ ⎢ α1 ⎥ ⎢ D(ω2 ) ⎥
⎢M
⎥⎢ M ⎥ = ⎢
⎥.
M
O
M
M

(−1) M / P (ω M + 1 ) ⎥ ⎢α M ⎥ ⎢ D(ω M + 1 ) ⎥
⎢1 cos(ω M + 1 ) L cos M (ω M + 1 )
⎢1 cos(ω
⎥⎢
M
M + 1 / P (ω

⎥ ⎢
M + 2 ) L cos (ω M + 2 ) ( −1)
M + 2 )⎦ ⎣ ε ⎦ ⎣ D(ω M + 2 )⎦

Note that the coefficients {αi } are different from the coefficients {a[i ]} of Eq. (10.70). To
determine the expression of
1
1
∆ = det M
1
1

we use Cramer's rule arriving at

, where

L
cos M (ω1 )
1 / P (ω1 )
M
L
− 1 / P (ω2 )
cos(ω2 )
cos (ω2 )
, and
M
O
M
M
cos(ω M +1 ) L cos M (ω M +1 )
(−1) M / P (ω M + 1 )
cos(ω M + 2 ) L cos M (ω M + 2 ) (−1) M + 1 / P (ω M + 2 )
cos(ω1 )

Not for sale

353

L
D(ω1 )
cos M (ω1 )
L
D(ω2 )
cos(ω2 )
cos M (ω2 )
..
M
O
M
M
cos(ω M + 1 ) L cos M (ω M + 1 ) D(ω M + 1 )
cos(ω M + 2 ) L cos M (ω M + 2 ) D(ω M + 2 )
cos(ω1 )

1
1
∆ε = M
1
1

Expanding both determinants using the last column we get ∆ ε =

M +2

∑ bi D(ωi +1 ) and
i =1

M +2

(−1) i −1

i =1

P(ωi )

∑ bi

∆=

1
1
M
bi = det 1
1

, where
cos(ω1 )
cos(ω2 )

cos 2 (ω1 )
cos 2 (ω2 )

L
L

cos M (ω1 )
cos M (ω2 )

M
cos(ωi −1 )
cos(ωi + 1 )

M
cos 2 (ωi −1 )
cos 2 (ωi + 1 )

O
L
L

cos M (ωi −1 )
cos M (ωi + 1 )

M
.

M
M
M
O
M
1 cos(ω M + 2 ) cos 2 (ω M + 2 ) L cos M (ω M + 2 )

The above matrix is seen to be a Vandermonde matrix and is determinant is given by

∏ (cos ωk − cos ωl ).

bi =

k ≠ l, k > l
k, l ≠ i

ci =

bi
Define ci =
. It can be shown by induction that
M +2
br

r =1
r ≠i

M +2

∏ cos ωi − cos ωn . Therefore,
1

n =1
n≠i

M +2

∑ bi D(ωi )

c D(ω1 ) + c2 D(ω2 ) + L + c M + 2 D(ω M + 2 )
= 1
.
M +2
c1
c2
c M + 2 (−1) M + 1
(−1) i

+L+
bi
P(ω1 ) P(ω2 )
P (ω M + 2 )
P
(
ω
)
i
i =1

ε = i =1

10.36

⎧ 1

W (ω) = ⎨ δ p
⎪δ
⎩ s

ω ∈ passband,
ω ∈ stopband.

⎧ 1 0 ≤ ω ≤ 0.45π,
W (ω) = ⎨
⎩4.5 0.55π ≤ ω ≤ π.

Not for sale

354

⎧ 1

10.37 W (ω) = ⎨ δ p
⎪δ
⎩ s

ω ∈ passband,

⎧ 1 0 ≤ ω ≤ 0.55π,
W (ω) = ⎨
⎩3.4 0.7π ≤ ω ≤ π.

ω ∈ stopband.

⎧ δp

⎪⎪ δs1
10.38 W (ω) = ⎨ 1
⎪ δp

⎪⎩ δs 2

0 ≤ ω ≤ ωs1 ,

0 ≤ ω ≤ 0.44π,
⎧1.43

ω p1 ≤ ω ≤ ω p2 , = ⎨ 1
0.55π ≤ ω ≤ 0.7π,
⎪ 5
0.82π ≤ ω ≤ π.

ωs 2 ≤ ω ≤ π ,

10.39 It follows from Eq. (10.22) that the impulse response of an ideal Hilbert transformer is an
antisymmetric sequence. If we truncate it to a finite number of terms between − M ≤ n ≤ M
the impulse response is of length (2 M + 1) which is odd. Hence the FIR Hilbert transformer
obtained by truncation and satisfying Eq. (10.90) cannot be satisfied by a Type 4 FIR filter.

10.40 (a) X ( z) =

N −1

∑ x[n] z − n .

Thus,

n=0

(
(
X ( z ) = X ( z ) z −1 = − α + z −1 =
(
1− αz −1

(
P( z ) =
(
D( z ) =

N −1

⎛ − α + z( −1 ⎞ P ( z( )

x[ n]⎜⎜
( −1 ⎟ = D( z( ) , where
1
α
z

n=0

N −1

∑ p[n] z( −1 = ∑ x[n](1 − αz( −1 ) N −1− n (−α + z( −1 ) n ,

and

n=0
N −1

∑ d[n]z( − n = (1 − αz( −1 ) N −1.

n=0

(
(
P( z )
(
(b) X[ k ] = X ( z ) ( j 2 πk / N =
(
z =e
D( z )

(
z = e j 2 πk / N

(
(
P[ k ]
(
= ( , where P[ k ] = P ( z ) z( = e j 2 πk / N is
D[ k ]

(
(
the N -point DFT of the sequence p[n] and D[ k ] = D( z ) z( = e j 2 πk / N is the N -point DFT

of the sequence d[n].
(c) Let P = [ p[0]

p[1] L p[ N − 1]]T and X = [x[0] x[1] L x[ N − 1]]T . Without

(
any loss of generality, assume N = 4 in which case P( z ) =

(
)
(
+ (− 3αx[ 0] + (1 + 2α 2 ) x[1] − α (2 + α 2 ) x[ 2] + 3α 2 x[3])z −1
(
+ (3α 2 x[ 0] − α(2 + α 2 ) x[1] + (1 + 2α 2 ) x[ 2] − 3α x[3])z − 2
=

3

∑ p[n] z( − n

n=0

x[ 0] − α x[1] + α 2 x[ 2] − α 3 x[3]

Not for sale

355

(

)

(
(
+ − α 3 x[ 0] + α 2 x[1] − α x[ 2] + α x[3] z − 3 . Equating like powers of z −1 we can write
P = Q ⋅ X, where P = [ p[0]

⎡ 1

− 3α
Q=⎢
⎢ 3α 2
⎢ 3
⎢⎣ − α

p[1] p[2] p[3]]T , X = [x[0] x[1] x[2] x[3]]T and

α2
−α
− α3 ⎤

− α(2 + α 2 ) 3α 2 ⎥
1 + 2α 2
.
− α (2 + α 2 )
− 3α ⎥
1 + 2α 2

α2
− α2
1 ⎥⎦

It can be seen that the elements q r , s , 0 ≤ r , s ≤ 3, q r,s , of the 4 4 matrix Q can be
determined as follows:
(i) The first row is given by q 0, s = (− α ) s ,
3!
(−α ) r , and
r! (3 − r )!
(iii) the remaining elements can be obtained using the recurrence relation
q r , s = qr −1, s −1 − αq r , s −1 + αq r −1, s .
In the general case, we only change the computation of the elements of the first column
( N − 1)!
(−α )r .
using the relation qr ,0 = N −1C r (− α ) r =
r! ( N − 1 − r )!
(ii) The first column is given by qr ,0 = 3C r (−α ) r =

M10.1 The impulse response coefficients of the truncated FIR highpass filter with cutoff
frequency at 0.4π can be generated using the following MATLAB statements:
n = -M:M;
num = -0.4*sinc(0.4*n);
num(M+1) = 0.6;

The magnitude responses of the truncated FIR highpass filter for two values of M are
shown below:

Magnitude

1
0.8
0.6
0.4
M=5
M=20

0.2
0

0

0.2

0.4

0.6

0.8

1

ω/π

Not for sale

356

M10.2 The impulse response coefficients of the truncated FIR bandpass filter with cutoff
frequencies at 0.7π and 0.3π can be generated using the following MATLAB statements:
n = -M:M;
num = 0.7*sinc(0.7*n) - 0.3*sinc(0.3*n);

The magnitude responses of the truncated FIR bandpass filter for two values of M are shown
below:

Magnitude

1
0.8
0.6
0.4
M=5
M=20

0.2
0

0

0.2

0.4

0.6

0.8

1

ω/π

M10.3 The impulse response coefficients of the truncated Hilbert transformer can be generated
using the following MATLAB statements:
n = 1:M;
c = 2*sin(pi*n/2).*sin(pi*n/2);b = c./(pi*n);
num = [-fliplr(b) 0 b];

The magnitude responses of the truncated Hilbert transformer for two values of M are
shown below:

Magnitude

1
0.8
0.6
0.4
M=5
M=20

0.2
0

0

0.2

0.4

0.6

0.8

1

ω/π

M10.4
% Problem #M10.4
% Cascade of 2 boxcar filters of length 4
K = 2;N = 4;b = firgauss(K,N);
figure(1);
stem(b);xlabel(‘n’);ylabel(‘h[n]’);
title('Cascade of 2 boxcar filters of length 4');

Not for sale

357

% Cascade of 4 boxcar filters of length 4
K = 4;N = 4;b = firgauss(K,N);
figure(2);
stem(b);xlabel(‘n’);ylabel(‘h[n]’);
title('Cascade of 4 boxcar filters of length 4');
% Cascade of 2 boxcar filters of length 12
K = 2;N = 12;b = firgauss(K,N);
figure(3);
stem(b);xlabel(‘n’);ylabel(‘h[n]’);
title('Cascade of 2 boxcar filters of length 12');
% Cascade of 4 boxcar filters of length 12
K = 4;N = 12;b = firgauss(K,N);
figure(4);
stem(b);xlabel(‘n’);ylabel(‘h[n]’);
title('Cascade of 4 boxcar filters of length 12');
Cascade of 2 boxcar filters of length 4

Cascade of 4 boxcar filters of length 4

4

50
40

3
h[n]

h[n]

30
2
1
0

20
10

1

2

3

4
n

5

6

0

7

0

5

10

1000

8

800

6

600

4

400

2

200

0

5

10

15

Cascade of 4 boxcar filters of length 12
1200

h[n]

h[n]

Cascade of 2 boxcar filters of length 12
12

0

10
n

15

20

25

0

0

n

10

20

30

40

50

n

We can see that by increasing either K or N, the approximation to a Gaussian function
gets better. It is noted that increasing the number of boxcar filters K to a number greater
than 3 greatly affects the Gaussian shape of the impulse response.
M10.5 % Problem #M10.05
N = 36;fc = 0.2*pi;
M = N/2;
n = -M:1:M;t = fc*n;
lp = fc*sinc(t);

Not for sale

358

b = 2*[lp(1:M) (lp(M+1) - 0.5) lp((M+2):N+1)];
bw = b.*hamming(N+1)';
[h2, w] = freqz(bw, 1, 512);
plot(w/pi, abs(h2));axis([0 1 0 1.2]);
xlabel('\omega/\pi');ylabel('Magnitude');
title(['\omega_c = ', num2str(fc), ', N = ', num2str(N)]);
ωc = 0.62832, N = 36

Magnitude

1
0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/ π

M10.6 D( x) = 3.2 x 2 + 4.05 x − 5.5. Its approximation over the range − 3 ≤ x ≤ 2 is given by
A( x) = a0 + a1 x. We want to minimize the peak value of the absolute error, i.e.,

minimize max 3.2 x 2 + 4.05 x − 5.5 − a0 − a1 x . Since there are 3 unknowns, a0 , a1 ,
− 3≤ x ≤ 2

and ε, we need 3 extremal points on x , which we arbitrarily choose as x1 = −3, x2 = 0,
and x3 = 2. We then solve the 3 linear equations: a0 + a1 x − ( −1) l ε = D ( xl ), l = 1, 2, 3.
⎡1 − 3 1 ⎤ ⎡a0 ⎤ ⎡11.15⎤
This leads to ⎢1 0 − 1⎥ ⎢ a1 ⎥ = ⎢ − 5.5 ⎥, whose solution yields a0 = 4.1, a1 = 0.85,
1 ⎥⎦ ⎢ ε ⎥ ⎢⎣ 15.4 ⎥⎦
⎢⎣1 2
⎣ ⎦
and ε = 9.6. A plot of the corresponding error E1 ( x) = 3.2 x 2 + 3.2 x − 9.6 is shown
below in Figure (c).
Result of second guess

10

10

5

5
Error

Error

Result of first guess

0

0

-5

-5

-10

-10

-3

-2

-1

0

1

2

x

-3

-2

-1

0

1

2

x

(c)

(d)

Not for sale

359

After looking at E1 (x) , we move the second extremal point x2 to the location where
E1 (x) is a minimum. The next extrtemal points are therefore given by x1 = −3,
x2 = −0.5, and x3 = 2. The new values of the unknowns are obtained by solving
1 ⎤ ⎡a0 ⎤ ⎡ 11.15 ⎤
⎡1 − 3
⎢1 − 0.5 − 1⎥ ⎢ a1 ⎥ = ⎢− 6.725⎥, which yields a0 = 3.7, a1 = 0.85, and ε = 10. A plot
2
1 ⎥⎦ ⎢ ε ⎥ ⎢⎣ 15.4 ⎥⎦
⎢⎣1
⎣ ⎦
of the corresponding error E 2 ( x) = 3.2 x 2 + 3.2 x − 9.2 is shown on the previous page in
Figure (d).
%
x
d
D
A
C

Problem #M10.06
= [-3 0 2];
= 3.2.*x.^2 + 4.05.*x-5.5;
= d';
= [1 -3 1;1 0 -1;1 2 1];
= inv(A)*D;

y = -3:0.05:2;
E = 3.2.*y.^2 + 4.05.*y - 5.5 - C(1) - C(2).*y;
% Results of first guess
figure(1);
plot(y,E);
axis([-3 2 -12 12]);
xlabel('x');
ylabel('Error');
title('Result of first guess');
hold on;
plot([-3 -3], [E(1) E(1)], 'o');
plot([2 2], [E(end) E(end)], 'o');
plot([0 0], [E(61) E(61)], 'o');
hold off;
x
d
D
A
C

=
=
=
=
=

[-3 -0.5 2];
3.2.*x.^2 + 4.05.*x-5.5;
d';
[1 -3 1;1 -0.5 -1;1 2 1];
inv(A)*D;

y = -3:0.05:2;
E = 3.2.*y.^2 + 4.05.*y - 5.5 - C(1) - C(2).*y;
% Results of second guess
figure(2);
plot(y,E);
axis([-3 2 -12 12]);
xlabel('x');
ylabel('Error');

Not for sale

360

title('Result of second guess');
hold on;
plot([-3 -3], [E(1) E(1)], 'o');
plot([2 2], [E(end) E(end)], 'o');
plot([-0.5 -0.5], [E(51) E(51)], 'o');
hold off;
M10.7 D(x ) = −5 x 3 − 0.2 x 2 + 8 x + 5.5 . Its approximation over the range − 2 ≤ x ≤ 2 is given by
A( x) = a0 + a1 x + a2 x 2 . We want to minimize the peak value of the absolute error, i.e.,

minimize max − 5 x 3 − 0.2 x 2 + 8 x + 5.5 − a0 − a1 x − a2 x 2 . Since there are 4
−2≤ x≤2

unknowns, a0 , a1 , a2 , and ε, we need 4 extremal points on x , which we arbitrarily
choose as x1 = −2, x2 = −1, x3 = 1, and x4 = 2. We then solve the 4 linear equations:
a0 + a1 x + a 2 x 2 − ( −1) l ε = D ( xl ), l = 1, 2, 3, 4. This leads to
⎡1 − 2 4 1 ⎤ ⎡a0 ⎤ ⎡ 28.7 ⎤
⎢1 − 1 1 − 1⎥ ⎢ a1 ⎥ = ⎢ 2.3 ⎥, whose solution yields a = 5.5, a = −7, a = −0.2,
0
1
2
⎢1 1 1 1 ⎥ ⎢a2 ⎥ ⎢ 8.3 ⎥

1
2
4
1
19
3

.
⎣⎢
⎦⎥ ⎣ ε ⎦ ⎣⎢
⎦⎥

and ε = 10. A plot of the corresponding error E1 ( x) = −5 x 3 + 15 x is shown below in
Figure (e). We observe that these values maximize the error (ε = 10).
Result of first guess
10

Error

5
0
-5
-10
-2

%
x
d
D
A
C

-1

0
x

1

2

Problem #M10.07
= [-2 -1 1 2];
= -5.*x.^3 - 0.2.*x.^2 + 8.*x + 5.5;
= d';
= [1 -2 4 1;1 -1 1 -1;1 1 1 1;1 2 4 -1];
= inv(A)*D;

y = -2:0.05:2;
E = -5.*y.^3 - 0.2.*y.^2 + 8.*y + 5.5 - C(1) - C(2).*y C(3).*y.^2;
% Results of first guess
figure(1);

Not for sale

361

plot(y,E);
axis([-2 2 -12 12]);
xlabel('x');
ylabel('Error');
title('Result of first guess');
hold on;
plot([-2 -2], [E(1) E(1)], 'o');
plot([2 2], [E(end) E(end)], 'o');
plot([-1 -1], [E(21) E(21)], 'o');
plot([1 1], [E(1) E(1)], 'o');
hold off;

12π
10π
, ωs =
, ωT =
18
18
18
% Problem #M10.8
wp = 4*(2*pi)/18;
ws = 6*(2*pi)/18;
wc = (wp + ws)/2;
dw = ws - wp;
% Hamming
M = ceil(3.32*pi/dw);N = 2*M+1;n = -M:M;
num = (6/18)*sinc(6*n/18);
wh = hamming(N)';b = num.*wh;

M10.8 ω p =

figure(1);
k=0:2*M:stem(k,b);
title('Impulse Response Coefficients');
xlabel('Time index n'); ylabel('Amplitude');
figure(2);
[h, w] = freqz(b,1,512);
plot(w/pi, 20*log10(abs(h))); grid;
xlabel('\omega/\pi'); ylabel('Gain, in dB');
title('Lowpass filter designed using Hamming window');
axis([0 1 -80 10]);
% Hann
M = ceil(3.11*pi/dw);N = 2*M+1;n = -M:M;
num = (6/18)*sinc(6*n/18);
wh = hann(N)';b = num.*wh;
figure(3);
k=0:2*M:stem(k,b);
title('Impulse Response Coefficients');
xlabel('Time index n'); ylabel('Amplitude');
figure(4);
[h, w] = freqz(b,1,512);
plot(w/pi, 20*log10(abs(h)));grid;
xlabel('\omega/\pi');ylabel('Gain, in dB');

Not for sale

362

title('Lowpass filter designed using Hann window');
axis([0 1 -80 10]);
% Blackman
M = ceil(5.56*pi/dw);N = 2*M+1;n = -M:M;
num = (6/18)*sinc(6*n/18);
wh = blackman(N)';b = num.*wh;
figure(5);
k=0:2*M:stem(k,b);
title('Impulse Response Coefficients');
xlabel('Time index n'); ylabel('Amplitude');
figure(6);
[h, w] = freqz(b,1,512);
plot(w/pi, 20*log10(abs(h)));grid;
xlabel('\omega/\pi');ylabel('Gain, in dB');
title('Lowpass filter designed using Blackman window');
axis([0 1 -80 10]);

Lowpass filter design using Hamming window: N = 31
Lowpass filter designed using Hamming window

Impulse Response Coefficients
0.4

0
Gain, in dB

Amplitude

0.3
0.2
0.1

-40
-60

0
-0.1

-20

-80
0

5

10
15
20
Time index n

25

30

0

0.2

0.4

0.6

0.8

1

ω/π

Lowpass filter design using Hann window: N = 29
Impulse Response Coefficients

Lowpass filter designed using Hann window

0.3

0

0.2

-20

Gain, in dB

Magnitude

0.4

0.1
0
-0.1

-40
-60

0

5

10
15
20
Time index n

25

-80

0

0.2

0.4

0.6

0.8

1

ω/π

Lowpass filter design using Blackman window: N = 53

Not for sale

363

Lowpass filter designed using Blackman window

0.3

0

0.2

-20

Gain, in dB

Magnitude

Impulse Response Coefficients

0.1

-60

0
-0.1

-40

0

10

20
30
Time index n

40

-80

50

0

0.2

0.4

0.6

0.8

1

ω/π

Comments: The Hann window method results in using the lowest filter order. All filters
meet the requirements of the specifications.
M10.9 α s = 42 , β = 0.5842(42 − 21)0.4 + 0.07886(42 − 21) = 3.631 using Eq. (10.41).
42 − 8
N=
using Eq. (10.42).
⎛ 2π ⎞
2.285⎜ ⎟
⎝ 18 ⎠
N = 42.627 ≥ 43 and we choose 44 since N must be even. M = 22.
% Problem #M10.9
beta = 3.631;N = 44;n = -N/2:N/2;
num = (6/18)*sinc(6*n/18);
wh = kaiser(N+1,beta)';b = num.*wh;
figure(1);
stem(b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel('Amplitude')
figure(2);
[h, w] = freqz(b,1,512);
plot(w/pi, 20*log10(abs(h)));grid;
xlabel('\omega/\pi');ylabel('Gain, in dB');
title('Lowpass filter designed using Kaiser window');
axis([0 1 -80 10]);
Impulse Response Coefficients

Lowpass filter designed using Kaiser window

0.4
0
Gain, in dB

Amplitude

0.3
0.2
0.1

-40
-60

0
-0.1

-20

0

10

20
30
Time index n

40

-80

0

0.2

0.4

0.6

0.8

1

ω/π

Not for sale

364

M10.10 ω p = 0.4π , ω s = 0.6π , α s = 42 dB, ω c = 0.5π , ∆ω = 0.2π
We will use the Hann window since it meets the requirements and has the lowest order
from Table 10.2.
3.11π
M=
= 15.55 → 16 ⇒ N = 32.
∆ω
% Problem M10.10
n = -16:16;
lp = 0.5*sinc(0.5*n);wh = hanning(33);
b = lp.*wh';
figure(1);
k=0:2*n;stem(k,b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel(‘Amplitude');
figure(2);
[h, w] = freqz(b,1,512);
plot(w/pi, 20*log10(abs(h)));grid;
xlabel('\omega/\pi');ylabel('Gain, in dB');
title('Lowpass filter designed using Hann window');
axis([0 1 -80 10]);
Impulse Response Coefficients

Lowpass filter designed using Hann window
0

0.5
Gain, in dB

Amplitude

0.4
0.3
0.2
0.1

-20
-40
-60

0
-0.1

0

5

10

15
20
Time index n

25

30

-80

0

0.2

0.4

0.6

0.8

1

ω/ π

M10.11 We use the same specifications from Problem M10.10, but we use the Dolph-Chebyshev
2.056(42 ) − 16.4
= 48.7 . We use N = 50, which is a much higher order
window. N =
2.285(0.2π )
than in Problem M10.10.
% Problem M10.11
n = -25:25;
lp = 0.5*sinc(0.5*n);wh = chebwin(51,42);
b = lp.*wh';
figure(1);
stem(b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel('Amplitude');
figure(2);
[h, w] = freqz(b,1,512);

Not for sale

365

plot(w/pi, 20*log10(abs(h)));grid;
xlabel('\omega/\pi');ylabel('Magnitude');
title('Filter designed using Dolph-Chebyshev window');
axis([0 1 -80 10]);
Impulse Response Coefficients

Filter designed using Dolph-Chebyshev window

0.6

0
Magnitude

Amplitude

0.4
0.2
0
-0.2

-20
-40
-60

0

10

20
30
Time index n

40

-80

50

0

0.2

0.4

0.6

0.8

1

ω/π

M10.12 n = -16:16;
b = fir1(32, 0.5, hanning(33));
figure(1);
stem(b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel('Amplitude');
figure(2);
[h, w] = freqz(b,1,512);
plot(w/pi, 20*log10(abs(h)));
grid;
xlabel('\omega/\pi');ylabel('Magnitude');
title(' Lowpass filter designed using Hann window');
axis([0 1 -80 10]);
Impulse Response Coefficients

Lowpass filter designed using Hann window

0.6
0
Magnitude

Amplitude

0.4
0.2
0
-0.2

-20
-40
-60

0

5

10

15
20
Time index n

25

30

-80

0

0.2

0.4

0.6

0.8

1

ω/ π

M10.13 N = 35;
for k = 1:N+1,
w(k) = 2*pi*(k-1)/(N+1);
if(w(k) >= 0.45*pi & w(k) <= 1.45*pi) H(k) = 1;
else H(k) = 0;
end
if (w(k) <= pi) phase(k) = i*exp(-i*w(k)*N/2);
else phase(k) = -i*exp(i*(2*pi-w(k))*N/2);

Not for sale

366

end
end
H = H.*phase;
f = ifft(H);
[FF, w] = freqz(f, 1, 512);
k = 0:N;
figure(1);
stem(k, real(f));
xlabel('Time index n');ylabel('Amplitude');
figure(2);
plot(w/pi,20*log10(abs(FF)));grid
xlabel('\omega/\pi');ylabel('Gain, dB');
axis([0 1 -50 5]);
0.5

0

Gain, dB

Amplitude

-10
0

-20
-30
-40

-0.5

0

5

10

15
20
25
Time index n

30

35

-50

0

0.2

0.4

0.6

0.8

1

ω/π

M10.14 N = 45;
L = N + 1;
for k = 1:L,
w = 2*pi*(k-1)/L;
if (w >= 0.5*pi & w <= 0.7*pi) H(k) = i*exp(-i*w*N/2);
elseif (w >= 1.3*pi & w <= 1.5*pi) H(k) = i*exp(i*(2*pi-w)*N/2);
else H(k) = 0;
end
end
f = ifft(H);
[FF, w] = freqz(f, 1, 512);
k = 0:N;
figure(1);
stem(k, real(f));
xlabel('Time index, n');ylabel('h[n]');
figure(2);
plot(w/pi, 20*log10(abs(FF)));grid;
ylabel('Gain, dB');xlabel('\omega/\pi');
axis([0 1 -50 5]);

Not for sale

367

0.2
0
-10
Gain, dB

h[n]

0.1
0
-0.1
-0.2

-20
-30
-40

0

10

20
30
Time index, n

-50

40

0

0.2

0.4

0.6

0.8

1

ω/ π

M10.15 ind = 1;
for k = 0:6,
H(ind) = exp(-i*2*pi*19*k/39);
ind = ind + 1;
end
k = 7;
H(ind) = 0.5*exp(-i*2*pi*19*k/39);
ind = ind + 1;
for k = 8:30,
H(ind) = 0;
ind = ind + 1;
end
k = 31;
H(ind) = 0.5*exp(-i*2*pi*19*k/39);
ind = ind + 1;
for k = 32:38,
H(ind) = exp(-i*2*pi*19*k/39);
ind = ind + 1;
end
h = ifft(H);
[FF, w] = freqz(h, 1, 512);
plot(w/pi, 20*log10(abs(FF)));grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
axis([0 1 -50 5]);
0

Gain, dB

-10
-20
-30
-40
-50

0

0.2

0.4

0.6

0.8

1

ω/ π

M10.16 ind = 1;

Not for sale

368

for k = 0:6,
H(ind) = exp(-i*2*pi*19*k/39);
ind = ind + 1;
end
k = 7;
H(ind) = (2/3)*exp(-i*2*pi*19*k/39);
ind = ind + 1;
k = 8;
H(ind) = (1/3)*exp(-i*2*pi*19*k/39);
ind = ind + 1;
for k = 9:29,
H(ind) = 0;
ind = ind + 1;
end
k = 30;
H(ind) = (1/3)*exp(-i*2*pi*19*k/39);
ind = ind + 1;
k = 31;
H(ind) = (2/3)*exp(-i*2*pi*19*k/39);
ind = ind + 1;
for k = 32:38,
H(ind) = exp(-i*2*pi*19*k/39);
ind = ind + 1;
end
h = ifft(H);
[FF, w] = freqz(h, 1, 512);
plot(w/pi, 20*log10(abs(FF)));grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
axis([0 1 -50 5]);
0

Gain, dB

-10
-20
-30
-40
-50

0

0.2

0.4

0.6

0.8

1

ω/π

12π
10π

, ωs =
, ωc =
, ∆ω =
18
18
18
18
wp = 4*(2*pi)/18;
ws = 6*(2*pi)/18;
wc = (wp + ws)/2;
dw = ws - wp;
% Hamming
M = ceil(3.32*pi/dw);
N = 2*M;
b = fir1(N, ws/(2*pi));

M10.17 ω p =

Not for sale

369

[H, w] = freqz(b,1,512);
figure(1);
stem(b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel('h[n]');
figure(2);
plot(w/pi, 20*log10(abs(H)));grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Lowpass filter designed using Hamming window');
axis([0 1 -80 10]);
% Hann
M = ceil(3.11*pi/dw);
N = 2*M;
b = fir1(N, ws/(2*pi), hanning(N+1));
[H, w] = freqz(b,1,512);
figure(3);
stem(b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel('h[n]');
figure(4);
plot(w/pi, 20*log10(abs(H)));grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Lowpass filter designed using Hann window');
axis([0 1 -80 10]);
% Blackman
M = ceil(5.56*pi/dw);
N = 2*M;
b = fir1(N, ws/(2*pi), blackman(N+1));
[H, w] = freqz(b,1,512);
figure(5);
stem(b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel('h[n]');
figure(6);
plot(w/pi, 20*log10(abs(H)));
grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Lowpass filter designed using Blackman window');
axis([0 1 -80 10]);
% Kaiser
beta = 3.631;
N = 44;
b = fir1(N, ws/(2*pi), kaiser(N+1, beta));
[H, w] = freqz(b,1,512);
figure(7);
stem(b);
title('Impulse Response Coefficients');
Not for sale

370

xlabel('Time index n');ylabel('h[n]');
figure(8);
plot(w/pi, 20*log10(abs(H)));grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Lowpass filter designed using Kaiser window');
axis([0 1 -80 10]);
(a) Hamming window using fir1
Impulse Response Coefficients

Lowpass filter designed using Hamming window

0.4
0

Gain, dB

0.3

0.1

-0.1

-20
-40
-60

0

0

5

10
15
20
Time index n

25

-80

30

0

0.2

0.4

0.6

0.8

1

ω/ π

(b) Hann window using fir1
Lowpass filter designed using Hann window

Impulse Response Coefficients
0.4

0

Gain, dB

0.3

h[n]

0.2
0.1

-0.1

-20
-40
-60

0

0

5

10
15
20
Time index n

-80

25

0

0.2

0.4

0.6

0.8

1

ω/ π

(c) Blackman window using fir1
Impulse Response Coefficients

Lowpass filter designed using Blackman window

0.4
0

Gain, dB

0.3
0.2
h[n]

h[n]

0.2

0.1

-40
-60

0
-0.1

-20

0

10

20
30
Time index n

40

50

-80

0

0.2

0.4

0.6

0.8

1

ω/ π

(d) Kaiser window using fir1
Not for sale

371

Impulse Response Coefficients

Lowpass filter designed using Kaiser window

0.4
0

Gain, dB

0.3

h[n]

0.2
0.1

-40
-60

0
-0.1

-20

0

10

20
30
Time index n

40

-80

0

0.2

0.4

0.6

0.8

1

ω/ π

M10.18 ω s = 0.4π , ω p = 0.55π , α p = 0.1 dB, α s = 42 dB, ∆ω = 0.15π , ωc = 0.475π
3.32 π
(a) Hamming: use Eq. (10.33): M =
= 22.133 → 23 ∴ N = 2 M = 46
0.15π
3.11π
(b) Hann: M =
= 20.733 → 21 ∴ N = 2 M = 42
0.15π
5.56 π
(c) Blackman: M =
= 37.067 → 38 ∴ N = 2 M = 76
0.15π
(d) Kaiser: δ s = 10 −α s / 20 = 0.00794
% Hamming
N = 46;
b = fir1(N, 0.475, 'high');
[H,w] = freqz(b,1,512);
figure(1);
stem(b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel('h[n]');
figure(2);
plot(w/pi, 20*log10(abs(H)));grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Highpass filter designed using Hamming window');
axis([0 1 -80 10]);
% Hann
N = 42;
b = fir1(N, 0.475, 'high', hanning(N+1));
[H,w] = freqz(b,1,512);
figure(3);
stem(b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel('h[n]');
figure(4);
plot(w/pi, 20*log10(abs(H)));grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Highpass filter designed using Hann window');
axis([0 1 -80 10]);
Not for sale

372

% Blackman
N = 76;
b = fir1(N, 0.475, 'high', blackman(N+1));
[H,w] = freqz(b,1,512);
figure(5);
stem(b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel('h[n]');
figure(6);
plot(w/pi, 20*log10(abs(H)));grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Highpass filter designed using Blackman window');
axis([0 1 -80 10]);
% Kaiser
ds = 0.00794;
[N,Wn,beta,type] = kaiserord([0.4 0.55],[1 0],[ds ds]);
b = fir1(N, 0.475,'high',kaiser(N+1,beta));
[H,w] = freqz(b,1,512);
figure(7);
stem(b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel('h[n]');
figure(8);
plot(w/pi, 20*log10(abs(H)));grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Highpass filter designed using Kaiser window');
axis([0 1 -80 10]);
(a) Hamming window using fir1
Impulse Response Coefficients

Highpass filter designed using Hamming window

0.6
0

Gain, dB

0.4

h[n]

0.2
0

-40
-60

-0.2
-0.4

-20

0

10

20
30
Time index n

40

-80

0

0.2

0.4

0.6

0.8

1

ω/ π

(b) Hann window using fir1

Not for sale

373

Impulse Response Coefficients

Highpass filter designed using Hann window

0.6
0
0.4
-20

Gain, dB

h[n]

0.2
0

-40
-60

-0.2
-0.4

0

10

20
30
Time index n

-80

40

0

0.2

0.4

0.6

0.8

1

ω/ π

(c) Blackman window using fir1
Highpass filter designed using Blackman window

Impulse Response Coefficients
0.6

0

Gain, dB

0.4

h[n]

0.2
0

-40
-60

-0.2
-0.4

-20

0

20

40
Time index n

-80

60

0

0.2

0.4

0.6

0.8

1

ω/ π

(d) Kaiser window using fir 1
Impulse Response Coefficients

Highpass filter designed using Kaiser window

0.4

0

0.2

-20

Gain, dB

h[n]

0.6

0
-0.2
-0.4

-40
-60

0

5

10

15
20
Time index n

25

30

-80

0

0.2

0.4

0.6

0.8

1

ω/ π

M10.19 ω p1 = 0.65π , ω p2 = 0.85π , ωs1 = 0.55π , ωs 2 = 0.75π , α p = 0.2 dB, α s = 42 dB
∆ω1 = ω p1 − ωs1 = 0.1π , ∆ω2 = ω p2 − ωs 2 = 0.1π = ∆ω
(a) Hamming window: M =
(b) Hann: M =

3.32π
= 33.2 → 34 ∴ N = 2 M = 68
0.1π

3.11π
= 31.1 → 32 ∴ N = 2 M = 64
0.1π

Not for sale

374

(c) Blackman: M =

5.56π
= 55.6 → 56 ∴ N = 2 M = 112
0.1π
− α / 20

(d) Kaiser: δs = 10 − α s / 20 = 0.00794 , δ p = 10 p
= 0.97724
% Problem #M10.19
% Hamming
N = 68;
b = fir1(N, [0.6 0.8]);
[H, w] = freqz(b,1,512);
figure(1);
stem(b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel('h[n]');
figure(2);
plot(w/pi, 20*log10(abs(H)));grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Bandpass filter designed using Hamming window');
axis([0 1 -80 10]);
% Hann
N = 64;
b = fir1(N, [0.6 0.8], hanning(N+1));
[H, w] = freqz(b,1,512);
figure(3);
stem(b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel('h[n]');
figure(4);
plot(w/pi, 20*log10(abs(H)));grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Bandpass filter designed using Hann window');
axis([0 1 -80 10]);
% Blackman
N = 112;
b = fir1(N, [0.6 0.8], blackman(N+1));
[H, w] = freqz(b,1,512);
figure(5);
stem(b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel('h[n]');
figure(6);
plot(w/pi, 20*log10(abs(H)));grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Bandpass filter designed using Blackman window');
axis([0 1 -80 10]);
% Kaiser
[N, Wn, beta, type] = kaiserord([0.6 0.8], [1 0], [0.97724
0.00794]);

Not for sale

375

b = fir1(2*N, [0.6 0.8], kaiser(2*N+1, beta));
[H, w] = freqz(b,1,512);
figure(7);
stem(b);
title('Impulse Response Coefficients');
xlabel('Time index n');ylabel('h[n]');
figure(8);
plot(w/pi, 20*log10(abs(H)));grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Bandpass filter designed using Kaiser window');
axis([0 1 -80 10]);

Hamming window using fir1

(a)

Bandpass filter designed using Hamming window

Impulse Response Coefficients
0.3

0

Gain, dB

0.2

h[n]

0.1
0

-40
-60

-0.1
-0.2

-20

0

10

20

30
40
Time index n

50

-80

60

0

0.2

Impulse Response Coefficients

0.8

1

Bandpass filter designed using Hann window

0.3
0.2

0

0.1

-20

Gain, dB

h[n]

0.6

ω/π

Hann window using fir1

(b)

0
-0.1
-0.2

(c)

0.4

-40
-60

0

10

20

30
40
Time index n

50

60

-80

0

0.2

0.4

0.6

0.8

1

ω/ π

Blackman window using fir1

Not for sale

376

Impulse Response Coefficients

Bandpass filter designed using Blackman window

0.2
0
-20

Gain, dB

h[n]

0.1
0

-40

-0.1
-0.2

-60

0

20

40
60
80
Time index n

-80

100

0

0.2

0.4

0.6

0.8

1

ω/ π

Kaiser window using fir1

(d)

Impulse Response Coefficients

Bandpass filter designed using Kaiser window

0.3
0

Gain, dB

0.2

h[n]

0.1
0

-40
-60

-0.1
-0.2

-20

0

10

20
30
Time index n

-80

40

0

0.2

0.4

0.6

0.8

1

ω/ π

2π ⋅ 15
= 0.4286 π
70
% Problem #M10.20
N = 31;
d1 = fir1(N-1,0.4286);d2 = fir1(N-1,0.4286,'high');
[h1,w] = freqz(d1,1,512);h2 = freqz(d2,1,w);
plot(w/pi,20*log10(abs(h1)),'-r',w/pi,20*log10(abs(h2)),'-b');grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Crossover pair');
axis([0 1 -80 10]);

M10.20 ωc =

Crossover pair

Gain, dB

0
-20
-40
-60
-80

0

0.2

0.4

0.6

0.8

1

ω/π

Not for sale

377

M10.21 ωc1 = 0.2494 , ωc 2 = 0.5442
% Problem #M10.21
N = 32;
d1 = fir1(N, 0.2494, hanning(33));
d2 = fir1(N, 0.5442, 'high', hanning(33));
d3 = -d1-d2;
d3(17) = 1-d1(17)-d2(17);
[h1,w] = freqz(d1,1,512);h2 = freqz(d2,1,w);h3 =
freqz(d3,1,w);
g1 = 20*log10(abs(h1))); g2 = 20*log10(abs(h2)));
g3 = 20*log10(abs(h3)));
plot(w/pi, g1,’—-b’,w/pi,g2,’-.g’,w/pi,g3,’-r’);grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Crossover triple');
axis([0 1 -80 10]);
Crossover triple

Gain, dB

0
-20
-40
-60
-80

0

0.2

0.4

0.6

0.8

1

ω/π

M10.22 % Problem #M10.22
fpts = [0 0.35 0.4 0.7 0.72 1];
mval = [0.2 0.2 1 1 0.6 0.6];
b = fir2(70, fpts, mval);
[H,w] = freqz(b,1,512);
figure(1);
plot(w/pi,abs(H));grid;
xlabel('\omega/\pi');ylabel('Magnitude');
title('Multilevel FIR filter');
axis([0 1 0 1.2]);
Multilevel FIR filter

Magnitude

1
0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/ π

Not for sale

378

M10.23 From Problem 10.36,
ω p = 0.45π , ω s = 0.6π , δ p = 0.2043 , δ s = 0.0454 and we assume that

0.45π ⋅ FT
= 0.45 and Fs = 0.6.

α p = −20 log10 1 − δ p = 1.985 dB, α s = −20 log10 (δ s ) = 26.858 dB.
After obtaining the length N using ‘remezord’, the specifications of the filter were not
met. We increased N to 11 to meet the specifications.
% Program #M10.23
Ft = 2;Fp = 0.45;Fs = 0.6;
ds = 0.0454;dp = 0.2043;
F = [Fp Fs];A = [1 0];DEV = [dp ds];
[N,Fo,Ao,W] = remezord(F,A,DEV,Ft);
b = remez(N,Fo,Ao,W);
[H,w] = freqz(b,1,512);
figure(1);
plot(w/pi, 20*log10(abs(H)));
xlabel('\omega/\pi');ylabel('Gain, dB');title('N = 9');
%axis([0 0.45 -3 3]);
N = 11;
b = remez(N,Fo, Ao, W);
[H,w] = freqz(b,1,512);
figure(2);
plot(w/pi, 20*log10(abs(H)));
xlabel('\omega/\pi');ylabel(Gain, dB);title('N = 11');

FT = 2. Therefore, F p =

(

)

Using remezord, we get N = 9. The corresponding gain response is shown in Figure (e)
below:
N = 11
20

0

0
Gain, dB

Gain, dB

N=9
20

-20
-40
-60
-80

-20
-40
-60

0

0.2

0.4

0.6

0.8

1

-80

ω/ π

0

0.2

0.4

0.6

0.8

1

ω/ π

(e)

(f)

However, specifications are not met in the passband with this filter, so we increase N to
11. The corresponding gain response is shown in Figure (f) above. The specifications
are now met.
Not for sale

379

M10.24 From problem 10.37,
ω p = 0.7π , ω s = 0.55π , δ p = 0.03808 , δ s = 0.0112 and we assume that

0.7π ⋅ FT
= 0.7 and Fs = 0.55.

α p = −20 log10 1 − δ p = 0.3372 dB, α s = −20 log10 (δ s ) = 39.016 dB.
Using ‘remezord’, we get an estimate of N = 20. However, using this order, the
specifications are not met in the stopband, so we need to increase N up to 23 to meet the
specifications.
% Program #M10.24
Ft = 2;Fp = 0.7;Fs = 0.55;
ds = 0.0112;dp = 0.03808;
F = [Fs Fp];A = [0 1];DEV = [ds dp];
[N,Fo,Ao,W] = remezord(F,A,DEV,Ft);
b = remez(N,Fo,Ao,W);
[H,w] = freqz(b,1,512);
figure(1);
plot(w/pi, 20*log10(abs(H)));
xlabel('\omega/\pi');ylabel('Gain, dB');title('N = 20');
N = 23;
b = remez(N,Fo,Ao,W);
[H,w] = freqz(b,1,512);
figure(2);
plot(w/pi,20*log10(abs(H)));
xlabel('\omega/\pi');ylabel('Gain,dB ');title('N = 23');

FT = 2. Therefore, F p =

(

)

N = 24

0

0

-20

-20
Gain, dB

Gain, dB

N = 20

-40

-60

-60
-80

-40

0

0.2

0.4

0.6

0.8

1

-80

0

0.2

0.4

0.6

0.8

1

ω/π

ω/ π

Note: As odd order symmetric FIR filters must have a gain of zero at the Nyquist
frequency. The order has been increased by one by remez.
M10.25 From Problem 10.38,
ω p1 = 0.55π , ω p 2 = 0.7π , ω s1 = 0.44 π , ω s 2 = 0.82π , δ p = 0.01 , δ s1 = 0.007 ,

(

)

δ s 2 = 0.002 . α p = −20 log10 1 − δ p = 0.087 dB,

Not for sale

380

α s1 = −20 log10 (δ s1 ) = 43 dB, α s 2 = −20 log10 (δ s 2 ) = 54 dB.
% Program #M10.25
Ft = 2;Fp1 = 0.55;Fp2 = 0.7;Fs1 = 0.44;Fs2 = 0.82;
ds1 = 0.007;ds2 = 0.002;dp = 0.01;
F = [Fs1 Fp1 Fp2 Fs2];A = [0 1 0];DEV = [ds1 dp ds2];
[N,Fo,Ao,W] = remezord(F,A,DEV,Ft);
b = remez(N,Fo,Ao,W);
[H, w] = freqz(b, 1, 512);
figure(1);
plot(w/pi,20*log10(abs(H)));grid;
xlabel('\omega/\pi');ylabel('Gain, dB');title('N = 39');
axis([0 1 -80 10]);
N = 41;
b = remez(N, Fo, Ao, W);
[H, w] = freqz(b, 1, 512);
figure(2);
plot(w/pi, 20*log10(abs(H)));
xlabel('\omega/\pi');ylabel('Gain, dB');title('N = 41');
axis([0 1 -80 10]);

Using remezord, we estimate the filter length to be N = 39. However, the minimum
stopband attenuation specifications are not met in both stopbands, so we increase N to 41
N = 41

0

0

-20

-20

Gain, dB

Gain, dB

N = 39

-40

-60

-60
-80

-40

0

0.2

0.4

0.6

0.8

1

-80

0

0.2

0.4

0.6

0.8

1

ω/ π

ω/ π

.
M10.26 % Program #M10.26
b = remez(29, [0 1], [0 pi], 'differentiator');
[H, w] = freqz(b, 1, 512);
plot(w/pi,abs(H));grid
xlabel('\omega/\pi');ylabel('Magnitude');
axis([0 1 0 pi]);

The magnitude response of the differentiator is given below:

Not for sale

381

3

Magnitude

2.5
2
1.5
1
0.5
0

0

0.2

0.4

0.6

0.8

1

ω/π

M10.27 % Program #M10.27
f = [0.02 0.05 0.07 0.95 0.97 1];
m = [0 0 1 1 0 0];
wt = [1 60 1];
b = remez(30, f, m, wt, 'hilbert');
[H,w] = freqz(b,1,512);
plot(w/pi,abs(H));grid
xlabel('\omega/\pi');ylabel('Magnitude');
axis([0 1 0 1.2]);

The magnitude response of the Hilbert transformer is shown below:

Magnitude

1
0.8
0.6
0.4
0.2
0

0

0.2

0.4

0.6

0.8

1

ω/ π

M10.28 ω p = 0.35π , ω s = 0.5π , R p = 1 dB, and Rs = 28 dB.
% Program #M10_28
% Design of a minimum-phase lowpass FIR filter
Wp = 0.35; Ws = 0.5; Rp = 1; Rs = 28;
% Desired ripple values of minimum-phase filter
dp = 1- 10^(-Rp/20); ds = 10^(-Rs/20);
% Compute ripple values of prototype linear-phase filter
Ds = (ds*ds)/(2 - ds*ds);
Dp = (1 + Ds)*((dp + 1)*(dp + 1) - 1);
% Estimate filter order
[N,fpts,mag,wt] = remezord([Wp Ws], [1 0], [Dp Ds]);
% Design the prototype linear-phase filter H(z)
[b,err,res] = remez(N,fpts,mag,wt);
K = N/2;
b1 = b(1:K);

Not for sale

382

% Design the linear-phase filter G(z)
lenerr= res.error(length(res.error));
c = [b1 (b(K+1) + lenerr)) fliplr(b1)]/(1+Ds);
zplane(c);title('Zeros of G(z)');pause
c1 = c(K+1:N+1);
[y, ssp, iter] = minphase(c1);
zplane(y);title(‘Zeros of the minimum-phase filter’);
pause
[hh,w] = freqz(y,1, 512);
% Plot the gain response of the minimum-phase filter
plot(w/pi, 20*log10(abs(hh)));grid
xlabel('\omega/\pi');ylabel('Gain, dB');
Zeros of G(z)

Zeros of the minimum-phase filter
1
Imaginary Part

Imaginary Part

2
1
2

0

20

-1

0.5

-0.5

-2
-6

10

0

-1
-4

-2
Real Part

0

-1.5

-1

-0.5

0
0.5
Real Part

1

1.5

20

Gain, dB

0
-20
-40
-60
-80

0

0.2

0.4

0.6

0.8

1

ω/ π

M10.29 % Program #M10.29
c = [2.4 6.76 26.15 68.43 186.83 326.51 565.53 678.95 805.24
678.95 565.53 326.51 186.83 68.43 26.15 6.76 2.4];
h = firminphase(c)

The coefficients of the minimum phase spectral factor are:
h = 7.6730 8.5329 18.0722 12.5696 12.6822 4.8388 2.0784 0.5332 0.3128
M10.30 % Program #M10.30
[h,g]=ifir(6,'low',[.1 .15],[.001 .001]);
[hh,w]=freqz(h,1,1024); hg=freqz(g,1,1024);
h = hh.*hg; % Compounded response
Fg = 20*log10(abs(hh)); Ig = 20*log10(abs(hg));
plot(w/pi,Fg,'-r',w/pi,Ig,'--b'); grid;
axis([0 1 -90 5]);

Not for sale

383

legend('F(z^6)','I(z)');
xlabel('\omega/\pi');title('Gain responses, in dB');
pause;
plot(w/pi,20*log10(abs(h))); grid;
axis([0 1 -90 5]);
xlabel('\omega/\pi');title('Gain response, in dB');
gtext('H_{IFIR}(z)');
Gain response, in dB

Gain responses, in dB
0

0

HIFIR(z)
6

-20

F(z )
I(z)

-20

-40

-40

-60

-60

-80

-80

0

0.2

0.4

0.6

0.8

0

1

0.2

0.4

0.6

0.8

1

ω/π

ω/ π

M10.31 % Program #M10.31
[h,g]=ifir(6,'high',[.9 .95],[.002 .004]);
[hh,w]=freqz(h,1,1024); hg=freqz(g,1,1024);
h = hh.*hg; % Compounded response
Fg = 20*log10(abs(hh)); Ig = 20*log10(abs(hg));
plot(w/pi,Fg,'-r',w/pi,Ig,'--b'); grid;
axis([0 1 -90 5]);
legend('F(z^6)','I(z)');
xlabel('\omega/\pi');title('Gain responses, in dB');
pause;
plot(w/pi,20*log10(abs(h))); grid;
axis([0 1 -90 5]);
xlabel('\omega/\pi');title('Gain response, in dB');
gtext('H_{IFIR}(z)');
Gain responses, in dB

Gain response, in dB

0

0
6

F(z )
I(z)

-20
-40

-40

-60

-60

-80

-80
0

0.2

HIFIR(z)

-20

0.4

0.6

0.8

1

ω/ π

0

0.2

0.4

0.6

0.8

1

ω/ π

Not for sale

384

M10.32 % Problem #M10.32
InpN = ceil(2*pi/(0.15*pi));
% Plotting function
% plots the result of using an equalizer of length InpN
%function [N] = plotfunc(InpN);
% Creating filters
Wfilt = ones(1, InpN);
Efilt = remezfunc(InpN, Wfilt);
% Plot running sum filter response
figure(1);
Wfilt = Wfilt/sum(Wfilt);
[hh, w] = freqz(Wfilt, 1, 512);
plot(w/pi, 20*log10(abs(hh)));
axis([0 1 -50 5]);grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Prefilter H(z)');
% Plot equalizer filter response
figure(2);
[hw, w] = freqz(Efilt, 1, 512);
plot(w/pi, 20*log10(abs(hw)));
axis([0 1 -50 5]);grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
title('Equalizer F(z)');
figure(3);
Cfilt = conv(Wfilt, Efilt);
[hc, w] = freqz(Cfilt, 1, 512);
plot(w/pi, 20*log10(abs(hc)));
axis([0 1 -80 5]);grid;
xlabel('\omega/\pi');ylabel('Gain, dB');
% Remez function using 1/P(z) as desired amplitude
% and P(z) as weighting
function [N] = remezfunc(Nin, Wfilt);
% Nin : number of tuples in the remez equalizer filter
% Wfilt : the prefilter
a = [0:0.001:0.999]; % The accuracy of the computation
w = a.*pi;wp = 0.05*pi;ws = 0.15*pi;
i = 1;n = 1;
for t = 1:(length(a)/2),

Not for sale

385

if w(2*t) < wp
pas(i) = w(2*t - 1);
pas(i+1) = w(2*t);
i = i+2;
end
if w(2*t-1) > ws
sto(n) = w(2*t - 1);
sto(n+1) = w(2*t);
n = n+2;
end
end
w = cat(2, pas, sto);
bi = length(w)/2;
for t1 = 1:bi,
bw(t1) = (w(2*t1) + w(2*t1-1))/2;
W(t1) = Weight(bw(t1), Wfilt, ws);
end
W = W/max(W);
for t2 = 1:length(w),
G(t2) = Hdr(w(t2), Wfilt, wp);
end
G = G/max(G);
N = remez(Nin, w/pi, G, W);
% Weighting function
function[Wout] = Weight(w, Wfilt, ws);
K = 22.8;
L = length(Wfilt);
Wtemp = 0;
Wsum = 0;
for k = 1:L,
Wtemp = Wfilt(k)*exp((k-1)*i*w);
Wsum = Wsum + Wtemp;
end
Wout = abs(Wsum);
if w > ws,
Wout = K*max(Wout);
end
% Desired function
function [Wout] = Hdr(w, Wfilt, ws);
if w <= ws,
L = length(Wfilt);
Wtemp = 0;
Wsum = 0;
for k = 1:L,
Wtemp = Wfilt(k)*exp(i*(k-1)*w);
Wsum = Wsum + Wtemp;
end
Wsum = abs(Wsum);
Wout = 1/Wsum;
Not for sale

386

else
Wout = 0;
end
Equalizer F(z)
0

-10

-10
Gain, dB

0

-20
-30
-40
-50

-20
-30
-40

0

0.2

0.4

0.6

0.8

-50

1

0

0.2

0.4

ω/ π

0.6

0.8

1

ω/ π
0
-20

Gain, dB

Gain, dB

Prefilter H(z)

-40
-60
-80

0

0.2

0.4

0.6

0.8

1

ω/ π

Not for sale

387

Chapter 11
11.1 Analysis yields
w1[ n] = β1{x[ n − 1] + α 2 w2 [ n − 1]}, w2 [ n] = w1[ n] + w3 [ n],
w3 [ n] = β1{α 2 w2 [ n − 1] + α 2 w 4 [ n − 1]}, w 4 [ n] = w3 [ n] + w5 [ n],
w5 [ n] = β3 {α 2 w 4 [ n − 1] + α3 w5 [ n − 1]}, y[ n] = w1[ n] + α 0 x[ n].
In matrix form the above equations can be written as
0
0
0 ⎤ ⎡ w1[n − 1] ⎤ ⎡ β1 x[n − 1]⎤
⎡ w1[n] ⎤ ⎡0 0 0 0 0 0 ⎤ ⎡ w1[n] ⎤ ⎡0 α1β1 0

⎢ w [n]⎥ ⎢1 0 1 0 0 0 ⎥ ⎢ w [n]⎥ ⎢0
0
0
0
0
0
0 ⎥⎥ ⎢⎢ w2 [n − 1]⎥⎥ ⎢⎢

⎢ 2 ⎥ ⎢
⎥⎢ 2 ⎥ ⎢

⎢ w3[n]⎥ ⎢0 0 0 0 0 0 ⎥ ⎢ w3 [n]⎥ ⎢0 α 2 β 2 0 α 2 β 2
0
0
0 ⎥ ⎢ w3 [n − 1]⎥ ⎢

⎥⎢
⎥+⎢

⎥=⎢
⎥⎢
⎥+⎢
0
0
0
0
0
0 ⎥ ⎢ w4 [n − 1]⎥ ⎢

⎢ w4 [n]⎥ ⎢0 0 1 0 1 0 ⎥ ⎢ w4 [n]⎥ ⎢0

⎢ w5 [n]⎥ ⎢0 0 0 0 0 0 ⎥ ⎢ w5 [n]⎥ ⎢0
0
0
0 α 2 β 3 α 3 β3 0 ⎥ ⎢ w5 [n − 1]⎥ ⎢

⎥⎢
⎥ ⎢

⎥ ⎢
⎥⎢
⎥ ⎢
0
0
0
0
0 ⎦ ⎣ y[n − 1] ⎦ ⎢⎣ α 0 x[n] ⎥⎦
⎣ y[n] ⎦ ⎣1 0 0 0 0 0 ⎦ ⎣ y[n] ⎦ ⎣0
⎡0 0 0 0 0 0 ⎤
⎢1 0 1 0 0 0 ⎥
Here the F matrix is given by: F = ⎢⎢0 0 0 0 0 0⎥⎥.
0 0 1 0 1 0
⎢0 0 0 0 0 0 ⎥
⎢⎣1 0 0 0 0 0⎥⎦
Since the F matrix contains nonzero entries above the main diagonal, the above set of
equations are not computable.
11.2 A computable set of equations are given by:
w1[ n] = β1{x[ n − 1] + α1w2 [ n − 1]}, w3[ n] = β1{α2 w2 [ n − 1] + α2 w4 [ n − 1]},

w5[ n] = β3 {α2 w4 [ n − 1] + α3w5[ n − 1]}, w2 [ n] = w1[ n] + w3[ n], y[ n] = w1[ n] + α0 x[ n] ,
w4 [ n] = w3[ n] + w5[ n] .
In matrix form the above equations can be written as
0
0
0 ⎤ ⎡ w1[n − 1] ⎤ ⎡ β1 x[n − 1]⎤
⎡ w1[n] ⎤ ⎡0 0 0 0 0 0 ⎤ ⎡ w1[n] ⎤ ⎡0 α1β1 0
⎢ w [ n]⎥ ⎢ 0 0 0 0 0 0 ⎥ ⎢ w [ n]⎥ ⎢ 0 α β 0 α β

0
0 ⎥⎥ ⎢⎢ w3 [n − 1]⎥⎥ ⎢⎢
0
2 2
2 2
⎢ 3 ⎥ ⎢

⎥⎢ 3 ⎥ ⎢
⎢ w5 [n]⎥ ⎢0 0 0 0 0 0 ⎥ ⎢ w5 [n]⎥ ⎢0

0
0 α 2 β 3 α 3 β 3 0 ⎥ ⎢ w5 [n − 1]⎥ ⎢
0
⎥+⎢

⎥=⎢
⎥+⎢
⎥⎢

⎥⎢
0
0
0
0
0 ⎥ ⎢ w2 [n − 1]⎥ ⎢
0
⎢ w2 [n]⎥ ⎢1 0 1 0 0 0 ⎥ ⎢ w2 [n]⎥ ⎢0

⎢ y[n] ⎥ ⎢1 0 0 0 0 0 ⎥ ⎢ y[n] ⎥ ⎢0

0
0
0
0
0 y[n − 1]
α 0 x[n] ⎥

⎥ ⎢
⎥ ⎢
⎥ ⎢
⎥⎢

⎥⎢
0
0
0
0
0 ⎦ ⎢⎣ w4 [n − 1]⎥⎦ ⎣
0
⎢⎣ w4 [n]⎥⎦ ⎣0 0 1 0 1 0 ⎦ ⎢⎣ w4 [n]⎥⎦ ⎣0

⎡0
⎢0
Here the F matrix is given by: F = ⎢⎢0
1
⎢1
⎢⎣0

0
0
0
0
0
0

0
0
0
1
0
1

0
0
0
0
0
0

Not for sale

0
0
0
0
0
1

0⎤
0⎥
0 ⎥.
0⎥
0⎥
0⎥⎦

388

Since the F matrix has no nonzero entries above the main diagonal, the new set of equations
are computable.
11.3 Analysis yields w1[ n] = β1 {α1w2 [ n − 1] + α 2 w 4 [ n − 1] + α3 w6 [ n − 1]}, w2 [ n] = w1[ n] + x[ n],
w3 [ n] = β2 {α2 w 4 [ n − 1] + α3 w6 [ n − 1]}, w 4 [ n] = w2 [ n] + w3 [ n], w 5 [ n] = β 3 {α3 w6 [ n − 1]},
w6 [ n] = w4 [ n] + w5[ n], y[ n] = α1w2 [ n] + α2 w 4 [ n] + α3 w6 [ n] + α 0 x[ n].
In matrix form the above set of equations are given by
⎡ w1[n] ⎤ ⎡0 0
⎢ w [ n ]⎥ ⎢ 1 0
⎢ 2 ⎥ ⎢
⎢ w3 [n]⎥ ⎢0 0

⎥ ⎢
⎢ w4 [n]⎥ = ⎢0 1
⎢ w5 [n]⎥ ⎢0 0

⎥ ⎢
⎢ w6 [n]⎥ ⎢0 0
⎢ y[n] ⎥ ⎢0 a

⎦ ⎣
1

0
0

0
0

0
0

0
0

0
1
0

0
0
0

0
0
0

0
0
0

0 1
0 a2

0 ⎤ ⎡ w1[n] ⎤ ⎡0 β1α1
0 ⎥⎥ ⎢ w2 [ n]⎥ ⎢0
0

⎥ ⎢
0 ⎥ ⎢ w3 [ n]⎥ ⎢0
0
⎥⎢
⎥ ⎢
0 ⎥ ⎢ w4 [ n]⎥ + ⎢0
0

0 w5 [n]
0
0
⎥⎢
⎥ ⎢
0 ⎥ ⎢ w6 [ n]⎥ ⎢0
0
0 ⎥⎦ ⎢⎣ y[ n] ⎥⎦ ⎢⎣0
0

1 0
0 a3

⎡0 0
⎢1 0
⎢0 0
Here the F matrix is given by: F = ⎢0 1
⎢0 0
⎢0 0
⎢⎣0 α1

β1α 2

0

β1α 3

0

0
0
0 β 2α 2

0
0
0 β 2α 3

0
0
0

0
0
0

0
0
0 β 3α 3
0
0

0

0

0

0

0 0
0 0
0 0
0 0
0 0
1 0
0 α3

0⎤
0⎥
0⎥
0 ⎥.
0⎥
0⎥
0 ⎥⎦

0 0
0 0
0 0
1 0
0 0
0 1
0 α2

0 ⎤ ⎡ w1[n − 1] ⎤ ⎡ 0 ⎤
0 ⎥⎥ ⎢⎢ w2 [n − 1]⎥⎥ ⎢⎢ x[ n] ⎥⎥
0 ⎥ ⎢ w3 [n − 1]⎥ ⎢ 0 ⎥

⎥⎢
⎥ ⎢
0 ⎥ ⎢ w4 [n − 1]⎥ + ⎢ 0 ⎥
0 ⎥ ⎢ w5 [n − 1]⎥ ⎢ 0 ⎥

⎥⎢
⎥ ⎢
0 ⎥ ⎢ w6 [ n − 1]⎥ ⎢ 0 ⎥
0 ⎥⎦ ⎢⎣ y[ n − 1] ⎥⎦ ⎢⎣α 0 x[n]⎥⎦

Since the diagonal of the F matrix has all zeros, and no nonzero entries above the main
diagonal, the new set of equations are computable.
11.4
α0
_1

x[n]

β1 z

_1

β2 α 1 z

1
β1 α 1 z

w1[n]

_1

β2 α 2 z

1

w2 [n]

β3 α 2 z

1
_1

_1

1

w5 [n]

w4 [n]

w3 [n]

y[n]

_1

1

β3 α 3 z

Reduced signal flow graph obtained by removing the branches going out of the input node
and the delay branches is:
1
w1[n]

1
1

1
w2 [n]

w3 [n]

w4 [n]

w5 [n]

y[n]

1

Not for sale

389

The node sets of the precedence graph are as follows:
{N 1 } = {w 1 [n], w 3 [n], w 5 [ n] }, {N 2 } = {w 2 [ n], w 4 [ n], wy[ n] }.
N1

N2

w1[n]

w2 [n]

w3 [n]

w4 [n]

w5 [n]

y[n]

In the above precedence graph, N 1 contains only outgoing nodes and N 2 contains only incoming
branches. The structure has no delay free loops. A valid computational algorithm is:

w1[ n] = β1{x[ n − 1] + α1w2 [ n − 1]}, w3[ n] = β1{α2 w2 [ n − 1] + α2 w4 [ n − 1]},
w5[ n] = β3 {α2 w4 [ n − 1] + α3w5[ n − 1]}, w2 [ n] = w1[ n] + w3[ n], y[ n] = w1[ n] + α0 x[ n] ,
w4 [ n] = w3[ n] + w5[ n] .

11.5 Reduced signal flow graph is:
1

1
w1[n]

1

1

w2 [n]

w3 [n]

α3

1

w4 [n]

w5 [n]

α2

y[n]

w6 [n]

α1

The node sets of the precedence graph are as follows: {N 1 } = {w1[ n], w3 [ n], w5 [ n]},
{N 2 } = {w2 [ n]}, {N 3 } = {w4 [ n]}, {N 4 } = {w6 [ n]},{N 5 } = {y[ n]}.
N1

w1[n]
w3 [n]

N2

N3

N4

N5

w2 [n]

w5 [n]

y[n]

w4 [n]

w6 [n]

A valid computational algorithm is:

w1[ n] = β1{α1w2 [ n − 1] + α2 w 4 [ n − 1] + α3 w6 [ n − 1]},
w3 [ n] = β2 {α2 w 4 [ n − 1] + α3 w6 [ n − 1]}, w 5 [ n] = β 3 {α3 w6 [ n − 1]},

w2 [ n] = w1[ n] + x[ n], w4 [ n] = w2 [ n] + w3 [ n], w6 [ n] = w4 [ n] + w5[ n],
y[ n] = α1w2 [ n] + α2 w 4 [ n] + α3 w6 [ n] + α 0 x[ n].
11.6 (a) Analysis yields w1[ n] = w2 [ n] − k1w1[ n − 1], w2 [ n] = w3 [ n] − k 2 s 2 [ n − 1],
w3 [ n] = x[ n] − k3 s3 [ n − 1], s 2 [ n] = k1w1[ n] + w1[ n − 1], s3 [ n] = k 2 w2 [ n] + s 2 [ n − 1],
y[ n] = k3 w3 [ n] + s3 [ n − 1]. In matrix form we have

Not for sale

390

⎡ w1 [ n ] ⎤ ⎡ 0 ⎤ ⎡ 0 1 0 0 0 0 ⎤ ⎡ w 1 [ n ] ⎤
⎢ w 2 [ n]⎥ ⎢ 0 ⎥ ⎢ 0 0 1 0 0 0⎥ ⎢ w 2 [ n]⎥
⎢ w [ n]⎥ ⎢ x[ n]⎥ ⎢ 0 0 0 0 0 0⎥ ⎢ w [ n]⎥
3
⎢ 3 ⎥=⎢
0 0 0 0 0 ⎥ ⎢ s [ n] ⎥
⎥ + ⎢k
2

⎢ s 2 [ n] ⎥ ⎢ 0 ⎥ ⎢ 1

⎢ s 3 [ n] ⎥ ⎢ 0 ⎥ ⎢ 0 k 2 0 0 0 0 ⎥ ⎢ s 3 [ n] ⎥
0 0⎦ ⎢ y[ n] ⎥
⎢⎣ y[ n] ⎥⎦ ⎣ 0 ⎦ 1
3 4044
⎣ 044044k2

43 ⎣
F
0
0
0 ⎤ ⎡ w1 [ n − 1] ⎤
⎡ − k1 0 0
⎢ 0
0 0 − k2
0
0 ⎥ ⎢ w 2 [ n − 1]⎥

⎢ 0
− k3 0 ⎥ ⎢ w 3 [ n − 1]⎥ .
0 0
0
+⎢
⎥ s [ n − 1]
0 0
0
0
0⎥ ⎢ 2

⎢ 1
s
n

[
1
]
0
0
0
1
0
0

⎥ 3
0
0
0
0
1
0

⎣1444442444443⎦ ⎣ y[ n − 1] ⎥⎦
G
As the diagonal elements of F-matrix are all zeros, there are no delay-free loops. However,
the above set of equations are not computable as there are non-zero elements above the
diagonal of F.
(b) The reduced signal flow-graph representation of Figure P11.3 is shown below:

From the above flow-graph we observe that the set composed of nodes with only outgoing
branches is N 1 = {w 3 [ n]}. The set of nodes with incoming branches from N 1 and
outgoing branches is N 2 = {w2 [ n]}. The set of nodes with incoming branches from N 1
and N 2 and outgoing branches is N 3 = {w1[ n]}. Finally, the set of nodes with only
incoming branches from N 1 , N 2 and N 3 , and no outgoing branches
N 4 = {s 2 [ n], s3 [ n], y[ n]}. Therefore, one possible ordered set of equations that is

⎡ w3 [ n]⎤ ⎡ 0 0 0 0 0 0⎤ ⎡ w3 [ n]⎤
⎢ w2 [ n]⎥ ⎢ 1 0 0 0 0 0⎥ ⎢ w2 [ n]⎥
⎢ w [ n] ⎥ ⎢ 0 1 0 0 0 0 ⎥ ⎢ w [ n] ⎥
computable is given by ⎢ 1 ⎥ = ⎢ 0 0 k 0 0 0⎥ ⎢ 1 ⎥
1
s [ n]
⎢ s 2 [ n] ⎥ ⎢
⎥⎢ 2 ⎥
k
0
0
0
0
0
2
⎢ s1[ n] ⎥ ⎢
⎥ ⎢ s1[ n] ⎥
k3 0 0 0 0 0⎦ ⎢ y[ n] ⎥
⎢⎣ y[ n] ⎥⎦ ⎣1

4444244443 ⎣
F
− k3 0⎤ ⎡ w3 [ n − 1]⎤
0
0
⎡0 0
⎢0 0
− k2
0
0
0⎥ ⎢ w2 [ n − 1]⎥

⎢0 0 − k
0
0
0⎥ ⎢ w1[ n − 1] ⎥ .
3
+⎢
⎥ s [ n − 1]
1
0
0
0⎥ ⎢ 2

⎢0 0

s
n
[
1
]
0
0
0
1
0
0

⎥ 1

0
0
1
0⎥⎦ ⎢⎣ y[ n − 1] ⎥⎦
⎢⎣0 0
1444442444443
G
Note: All elements on the diagonal and above diagonal of F are zeros.

Not for sale

391

p + p1 z −1 + p2 z −2
11.7 H ( z ) = 0
. From Eq. (11.18) we get,
1 − 3 z − 1 + 1 .5 z − 2
0
0 ⎤ ⎡ 1 ⎤ ⎡ 3.2 ⎤
⎡ p0 ⎤ ⎡3.2
⎢ p ⎥ = ⎢5.6 3.2
⎥ ⎢ − 3⎥ = ⎢ − 4.0⎥. Hence, H ( z ) = 3.2 − 4 z −1 − 5z −2 .
0
1
⎢ ⎥ ⎢
⎥⎢ ⎥ ⎢

1 − 3z −1 + 1.5z − 2
⎢⎣ p2 ⎥⎦ ⎢⎣7.0 5.6 3.2⎥⎦ ⎢⎣1.5 ⎥⎦ ⎢⎣ − 5.0 ⎥⎦

11.8 From Eqn (11.15),

⎡ d1 ⎤
⎡4
⎢ d ⎥ = − ⎢8

⎣ 2⎦
⎡ p0 ⎤ ⎡ 2
⎢ p ⎥ = ⎢− 2
⎢ 1⎥ ⎢
⎢⎣ p2 ⎥⎦ ⎢⎣ 4

− 2⎤
4 ⎥⎦

0
2
−2

11.9 From Eqn. (11.15)

⎡ d 1 ⎤ ⎡− 8
⎢ d ⎥ = ⎢ 12
⎢ 2⎥ ⎢
⎢⎣ d 3 ⎥⎦ ⎢⎣ 16
⎡ p0 ⎤ ⎡ 2
⎢p ⎥ ⎢
⎢ 1 ⎥ = ⎢− 4
⎢ p2 ⎥ ⎢ 8
⎢ ⎥ ⎢
⎣ p3 ⎦ ⎣ − 8
H ( z) =

0 0⎤
⎡ p0 ⎤ ⎡ 2
⎢ p ⎥ ⎢ −2 2 0 ⎥ 1
⎢ 1⎥ ⎢
⎥⎡ ⎤
⎢ p2 ⎥ = ⎢ 4 −2 2 ⎥ ⎢ d1 ⎥ and from Eqn (11.20),
⎢ ⎥ ⎢
⎥⎢ ⎥

0
8
4
2
⎢ ⎥ ⎢
⎥ ⎣⎢ d 2 ⎦⎥
⎢⎣ 0 ⎥⎦ ⎢⎣ −12 8 4 ⎥⎦
−1

⎡ 8 ⎤ ⎡ − 0.25⎤
⎢ − 12⎥ = ⎢ 3.5 ⎥. . Finally, from Eqn (11.21),

⎦ ⎣

0⎤ ⎡ 1 ⎤ ⎡ 2 ⎤
2 − 2.5z −1 + 11.5z −2
0⎥⎥ ⎢⎢ − 0.25⎥⎥ = ⎢⎢ − 2.5⎥⎥. Hence, H ( z) =
.
1 − 0.25z −1 + 3.5z − 2
2⎥⎦ ⎢⎣ 3.5 ⎥⎦ ⎢⎣ 11.5 ⎥⎦

⎡ p0 ⎤ ⎡ 2
⎢ ⎥ ⎢
⎢ p1 ⎥ ⎢ −4
⎢ p2 ⎥ ⎢ 8
⎢ ⎥ ⎢
⎢ p3 ⎥ = ⎢ −8
⎢ 0 ⎥ ⎢12
⎢ ⎥ ⎢
⎢ 0 ⎥ ⎢16
⎢ 0 ⎥ ⎢ −8
⎣ ⎦ ⎣

8
−8
12

− 4⎤
8 ⎥⎥
− 8 ⎥⎦

0
2
−4
8

0
0
2
−4

−1

0 0 0⎤
2 0 0 ⎥⎥ ⎡ ⎤
−4 2 0 ⎥ ⎢⎢ 1 ⎥⎥

8 −4 2 ⎥ ⎢ d1 ⎥ . Next, from Eqn. (11.20),
⎢ ⎥
−8 8 −4 ⎥ ⎢ d 2 ⎥

12 −8 8 ⎥ ⎢⎣ d3 ⎥⎦
16 12 −8⎥⎦

⎡ 12 ⎤ ⎡ 0.4 ⎤
⎢ 16 ⎥ = ⎢ − 4.8⎥. Finally, from Eqn. (11.21),

⎢ ⎥ ⎢
⎢⎣ − 8⎥⎦ ⎢⎣ − 7.4⎥⎦
0⎤ ⎡ 1 ⎤ ⎡ 2.0 ⎤
0⎥⎥ ⎢⎢ 0.4 ⎥⎥ ⎢⎢ − 3.2 ⎥⎥
=
. Hence,
0⎥ ⎢ − 4.8⎥ ⎢ − 3.2 ⎥
⎥⎢
⎥ ⎢

2⎦ ⎣ − 7.4 ⎦ ⎣ − 0.4⎦

2 − 3.2 z −1 − 3.2 z −2 − 0.4 z −3
1 + 0.4 z −1 − 4.8z − 2 − 7.4 z − 3

.

Not for sale

392

11.10

⎡ p0 ⎤ ⎡ 2 0 0
⎢ ⎥ ⎢
⎢ p1 ⎥ = ⎢ −4 2 0
⎢ p2 ⎥ ⎢ 4 −4 2
⎢ ⎥ ⎢
⎣ p3 ⎦ ⎣ −6 −6 −4

0⎤ ⎡ 1 ⎤ ⎡ 2 ⎤
0 ⎥⎥ ⎢⎢ −0.6 ⎥⎥ ⎢⎢ −5.2⎥⎥
=
. Hence, P ( z ) = 2 − 5.2 z −1 + 6.8 z −2 − 5.6 z −3 .
0 ⎥ ⎢ 0.2 ⎥ ⎢ 6.8 ⎥
⎥⎢
⎥ ⎢

2 ⎦ ⎣ 1.8 ⎦ ⎣ −5.6⎦

⎡ p0 ⎤ ⎡ 2 0 0 0
⎢ ⎥ ⎢
⎢ p1 ⎥ ⎢ −2 2 0 0
11.11 ⎢ p2 ⎥ = ⎢ 4 −2 2 0
⎢ ⎥ ⎢
⎢ p3 ⎥ ⎢ 6 4 −2 2
⎢⎣ p4 ⎥⎦ ⎢⎣ −8 6 4 −2

0⎤ ⎡ 1 ⎤ ⎡ 2 ⎤
0 ⎥⎥ ⎢⎢ 2 3 ⎥⎥ ⎢⎢ − 2 3 ⎥⎥
0⎥ ⎢ 0 ⎥ = ⎢ 2 2 3 ⎥ .
⎥⎢ ⎥ ⎢

0 ⎥ ⎢ −2 ⎥ ⎢ 4 2 3 ⎥
2 ⎥⎦ ⎢⎣ 5 3 ⎥⎦ ⎢⎣ −3 1 3 ⎥⎦

{ pi } = {2, − 2 3 , 2 2 3 , 4 2 3 , −3 13}

11.12 The k –th sample of an N –point DFT is given by X[ k ] =

N −1

∑ n = 0 x[n]WNnk .

Thus, the

computation of X [k ] requires N complex multiplications and N − 1 complex additions.
Now, each complex multiplication, in turn, requires 4 real multiplications and 2 real
complex multiplications needed to compute X [k ] require a total of 4 N real multiplications
and a total of 2 N − 2 real additions. Therefore, each sample of the N –point DFT involves
4 N real multiplications and 4 N − 2 real additions. Hence, the computation of all DFT
samples thus requires 4N 2 real multiplications and (4 N − 2) N real additions.
11.13 Let the two complex numbers be α = a + j b and β = c + j d. Thus, αβ = ( a + j b)(c + j d )
= ( ac − bd ) + j ( ad + bc ), which requires 4 real multiplications and 2 real additions. Consider
the product ( a + b)(c + d ), ac, and bd, which require 3 real multiplications and 2 real
additions. The imaginary part of αβ can be formed from ( a + b)(c + d ) − ac − bd = ad + bc,
which now requires 2 real additions. Likewise, the real part of αβ can be formed from
ac − bd requiring an additional real additions. Hence, the complex multiplication can be
computed using 3 real multiplications and 5 real additions.
11.14 Recall, WN = e − j 2π / N = c + j s, where c = cos( 2π / N ) and s = − sin( 2π / N ). Thus,
c 2 + s 2 = 1. Now, Ψ r +1 = W N ⋅Ψ r = (c + j s )(Re{Ψ r } + j Im{Ψ r })

= (c ⋅ Re{Ψr } − s ⋅ Im{Ψr }) + j (c ⋅ Im{Ψr } + s ⋅ Re{Ψr }).
Thus, Re{Ψr +1} = c ⋅ Re{Ψr } − s ⋅ Im{Ψr } and Im{Ψr +1} = c ⋅ Im{Ψr } + s ⋅ Re{Ψr }.
Figure P11.4 with internal node label is shown below. Its analysis yields
U

Re{ψr }
c _1
____
s

Re{ψr+1}
s

Im{ψ r }

c _1
____
s
Im{ψr+1}

Not for sale

393

c −1
Im{Ψr }, (2): Im{Ψr +1} = Im{Ψr } + sU , and
s
c −1
(3): Re{Ψr + 1} = U +
Im{Ψr + 1}. Substituting Eqs. (1) in Eq. (2) we get
s
⎛ c −1

(4): Im{Ψr + 1} = Im{Ψr } + s⎜
Im{Ψr } + Re{Ψr }⎟ = c ⋅ Im{Ψr } + s ⋅ Re{Ψr }. Next,
⎝ s

substituting Eqs. (1) and (4) in Eq. (3) we get (5):
c −1
(5): Re{Ψr + 1} =
(c ⋅ Im{Ψr } + s ⋅ Re{Ψr }) + c − 1 Im{Ψr } + Re{Ψr }
s
s
2
2
c −1
−s
=
Im{Ψr } + c ⋅ Re{Ψr } =
⋅ Im{Ψr } + c ⋅ Re{Ψr } = c ⋅ Re{Ψr } − s ⋅ Im c ⋅ Re{Ψr }.
s
s
It thus follows that the structure of Figure P11.4 implements the multiplication of a complex
signal with the twiddle factor using only 3 real multiplications.

(1): U = Re{Ψr } +

.
In the case of multiplication by complex twiddle factor W N−1 = c − js, we have
Ψ r +1 = W N−1 ⋅Ψ r = (c − j s )(Re{Ψ r } + j Im{Ψ r })

= (c ⋅ Re{Ψr } + s ⋅ Im{Ψr }) + j (c ⋅ Im{Ψr } − s ⋅ Re{Ψr }). Thus, here
Re{Ψr +1} = c ⋅ Re{Ψr } + s ⋅ Im{Ψr } and Im{Ψr +1} = c ⋅ Im{Ψr } − s ⋅ Re{Ψr }. It follows
then that the real and imaginary parts are simply obtained by reversing the sign of s of the
real and imaginary parts derived in the case of multiplication by W N . As a result, the
corresponding structure is obtained by cascading an inverter to the real multipliers in Figure
P11.4 as indicated below:
Re{ψr }

Im{ψ r }

U
_1
c _1
____
s

s

_1

_1

c _1
____
s

Re{ψr+1}

Im{ψr+1}

kFT
, where N = # of bins, and FT is the sampling
N
⎢ f N⎥
frequency. Inverting we have k ( f ) = ⎢
⎥. Therefore, the absolute difference from one of
⎣ FT ⎦
the given four tones (150 Hz, 375 Hz, 620 Hz, and 850 Hz) to the center of its bin is given by
F ⎢ f N⎥
dist ( N , f ) = f − T ⎢
⎥ . It follows from this equation that the distance goes to zero if
N ⎣ FT ⎦

11.15 The center frequency bin k : f c (k ) =

FT ⎢ f N ⎥
fN
is an integer.

⎥ = f or
N ⎣ FT ⎦
FT

Not for sale

394

fN
The total distance is reduced to zero if i is an integer for i = 1,K , 4. The minimum value
FT
of N for which this true is 500.
However, the total distance can be small, but nonzero, for significantly smaller values of N .
11.16 H k ( z ) =

1
1 − WN− k z −1

where V ( z ) =

. Hence, Y ( z ) =

1
1 − WN− k z −1

{

= 1,

WN−1 ,

1 − WN− k z −1

=

1 + z −N / 2
1 − WN− k z −1

. Or, in other words, y[ n] = v[ n] + v[ n −

Consider k = 1 : Then V ( z ) =
v[ n] = WN− n µ[ n] =

X ( z)

1
1 − WN−1 z −1

= V ( z ) + z − N / 2V ( z ),
N
].
2

. This implies,

N

− ( + 1)

1

2

N
/
2
2
, WN
, L⎬
⎨1, WN , WN , L, WN

WN− 2 ,

son on. Thus, v[ n −

L, − 1,
N
]=
2

{0,

− WN−1 ,

N
− ( + 1)

/
2
N
L , since WN
= −1, WN 2
= −WN−1 , and

}

}

0, 0, L, 1, WN−1 , WN− 2 , L . Hence,

N

− ( − 1)

1

2
2
y[ n] = v[ n] + v[ n −
, 0, 0, 0, L⎬.
⎨1, WN , WN , L, WN

1
N
Now, consider k = : V ( z ) =
. This implies,
2
1 − WN− N / 2 z −1
N
]=
2

v[ n] = WN− ( N / 2 )n µ[ n] =

N
] = 0, 0, 0, L,
2
N N
N
− ⋅

N
/
2
WN 2 2 = (−1)
, WN 2

Thus, v[ n −
Now,

{

N
N N
N
N

− ⋅
− ( + 1) ⋅

N

2
2
2
2
2
, WN
, L⎬.
⎨1, WN , WN , L, WN

}

0, 1, WN− N / 2 , WN− N , L .

N
N
− ( + 1)
N

2 = (−1) N / 2 (−1), etc. Hence,
= −1, WN = 1, WN 2

N
+1

N
/
2
y[ n] = ⎨1, − 1, 1, − 1, L, 1 + (−1)
, 1 + (−1) 2 , L, L⎬. Therefore, if N / 2
N
n= N / 2

n = +1
2

is even, y[ n] = ⎨1, − 1, 1, − 1, L,
− 2,
2,
2, L⎬, and if N / 2 is odd,
n = N / 2 n = N +1

2

Not for sale

395

y[ n] = ⎨1, − 1, 1, − 1, L,
0,
0,
0, L⎬,
n = N / 2 n = N +1

2

11.17

11.18

11.19 X[ k ] =

N −1

∑ x[n]WNnk

n=0

=

N
−1
r1

n=0

nkr1

x[ nr1 ]WN

+

N
−1
r1

n=0

nkr1 + k

x[ nr1 + 1]WN

+K+

Not for sale

N
−1
r1

n=0

nkr1 + (r1 −1)k

x[ nr1 + r 1−1]WN

396

⎛ N −1

r1 −1⎜ r1

nkr ⎟
=
x[nr 1+i ]W N 1 ⎟ W Nki .

i =0 ⎜ n=0
⎜ ( N / r1 ) − point DFT ⎟

Thus, if the ( N / r1 ) -point DFT has been calculated, we need at the first stage an additional
(r1 − 1) multiplications to compute one sample of the N –point DFT X [k ] and, as a result,
additional (r1 − 1) N multiplications are required to compute all N samples of the N –point
DFT. Decomposing it further, it follows that additional (r2 − 1) N multiplications are needed
at the second stage, and so on. Therefore, the total number of multiply (add) operations
ν
ri − ν ⎞⎟ N .
= (r1 − 1) N + (r2 − 1) N + K + (rν − 1) N = ⎛⎜

⎝ i =1

∑∑

11.20 An examination of the flow-graph of the 8-point DIT FFT algorithm shown in Figure 11.24
reveals that in the first stage all twiddle factors are WN0 = 1 and in the second stage the

twiddle factors are either WN0 = 1 or W NN / 4 = j. Hence, there are no twiddle factors with
nonunity magnitudes in the first two stages. In all succeeding stages, one of the twiddle
factors is WN0 = 1 and another one is W NN / 4 = j. The number of twiddle factors with
nonzero magnitudes in the i -th stage, i ≥ 3, is 2 i −1 − 2. Hence, the total number of twiddle
factors with nonzero magnitudes in an N-point radix-2 FFT algorithm is
⎛ ν

R (ν) = ⎜⎜
2 i −1 ⎟⎟ − 2(ν − 2), ν ≥ 3, where N = 2 ν.

⎝ i =3

11.21 Direct computation of M samples of an N -point DFT requires M 2 multiplications,

whereas, the Radix-2 FFT algorithm requires

N
2

log 2 N multiplications. In order for a N-

point radix-2 FFT algorithm to be computationally more efficient than a direct computation
⎢ N

of M samples of an N -point DFT, the following inequality must hold: M > ⎢
log 2 N ⎥.
⎣ 2

a) N = 512, M = 49, b) N = 1024, M = 72, c) N = 2048, M = 107
11.22 X ( z) = X 0 ( z 3 ) + z −1 X1 ( z 3 ) + z −2 X 2 ( z 3 ). Thus, the N –point DFT can be expressed as
X[ k ] = X 0 [〈 k 〉 N / 3 ] + WNk X1[〈 k 〉 N / 3 ] + WN2 k X 2 [〈 k 〉 N / 3 ]. Hence, the structural

interpretation of the first stage of the radix-3 DFT is as indicated below:

Not for sale

397

x[n]

3
z

+

X[k]

x 1[n] N_ -point X 1[< k >N/3 ] WNk
+
3
3
DFT

z

3

11.23 X[ k ] =

x 0[n] N
_ -point X 0 [< k >N/3 ]
3
DFT

2k
x 2[n] N_
X 2[< k >N/3 ] WN
-point
3
DFT

∑ x[n]W9nk = (x[0]W90⋅ k + x[3]W93k + x[6]W96k )
8

n=0

(

) (

+ x[1]W9k + x[ 4]W94 k + x[7]W97k + x[2]W92 k + x[5]W95k + x[8]W98k

)

(
) (
)
+ (x[2]W90 ⋅ k + x[5]W93k + x[8]W96 k )W92 k = G0 [〈 k 〉 3 ] + G1[〈 k 〉 3 ]W9k + G2 [〈 k 〉 3 ]W92 k , where

= x[0]W90 ⋅ k + x[3]W93k + x[6]W96 k + x[1]W90 ⋅ k + x[ 4]W93k + x[7]W96 k W9k

G0 [〈 k 〉 3 ] = x[0]W30 ⋅ k + x[3]W3k + x[6]W32 k , G1[〈 k 〉 3 ] = x[1]W30 ⋅ k + x[ 4]W3k + x[7]W32 k ,
G2 [〈 k 〉 3 ] = x[2]W30 ⋅ k + x[5]W3k + x[8]W32 k , are 3-point DFTs. A flow-graph representation
of this radix-3 DIT FFT computation scheme is shown below,

where the twiddle factors for computing the DFT samples are indicated below for a typical
DFT sample:

Not for sale

398

In the above diagram, the 3-point DFT computation is carried out as indicated below:

11.24 X[ k ] =

∑ x[n]W15nk = (x[0]W150⋅ k + x[3]W153k + x[6]W156k + x[9]W159k + x[12]W1512k )
14

n=0
k + x[ 4]W 4 k + x[ 7]W 7k + x[10]W 10 k + x[13]W 13k
+ x[1]W15
15
15
15
15
2
k
5
k
8
k
11
k
14 k
+ x[2]W15 + x[5]W15 + x[8]W15 + x[11]W15 + x[14]W15
k + G [ 〈 k 〉 ]W 2 k , where
= G0 [〈 k 〉 5 ] + G1[〈 k 〉 5 ]W15
2
5 15
0

k
k
G0 [〈 k 〉 5 ] = x[0]W5 + x[3]W5 + x[6]W52 k + x[9]W53k + x[12]W54 k ,
G1[〈 k 〉 5 ] = x[1]W50 ⋅ k + x[ 4]W5k + x[7]W52 k + x[10]W53k + x[13]W54 k , and
G2 [〈 k 〉 5 ] = x[2]W50 ⋅ k + x[5]W5k + x[8]W52 k + x[11]W53k + x[14]W54 k .

(
(

)
)

A flow-graph representation of this mixed-radix DIT FFT computation scheme is shown
below:

Not for sale

399

11.25

11.26 Now, by definition, q[ n] = Im{X [ n]} + j Re{X [ n]}. Its N -point DFT is given by

Q[ k ] =

N −1

∑ q[n]WNnk .

Thus,

n=0

Not for sale

400

(1): Re{Q[ k ]} =
(2): Im{Q[ k ]} =

N −1

⎛ 2πmk ⎞
⎛ 2πmk ⎞ ⎤
⎟ + Re{X[ n] sin⎜
⎟⎥ ,
N ⎠
⎝ N ⎠⎦

∑ ⎢⎣Im{X[n] cos⎜⎝

m=0
N −1

⎛ 2πmk ⎞
⎛ 2πmk ⎞ ⎤
⎟ + Re{X[ n] cos⎜
⎟⎥ .
N ⎠
⎝ N ⎠⎦

∑ ⎢⎣− Im{X[n] sin⎜⎝

m=0

N −1

1
X[ m]WN− mk . Hence,
From the definition of the inverse DFT we observe x[ k ] =
N m=0
(3): Re{x[ k ]} =

N −1

1
⎛ 2πmk ⎞
⎛ 2πmk ⎞ ⎤
Re{X[ m]} cos⎜
⎟ − Im{X[ m]}sin⎜
⎟ ⎥,

N m=0 ⎣
⎝ N ⎠
⎝ N ⎠⎦

N −1

1
⎛ 2πmk ⎞
⎛ 2πmk ⎞ ⎤
Im{X[ m]} cos⎜
⎟ + Re{X[ m]}sin⎜
⎟ ⎥.

N m=0 ⎣
⎝ N ⎠
⎝ N ⎠⎦
1
Comparing Eqs. (2) and (3) we get Re{x[ n]} = ⋅Im{Q[ k ]} k = n , and comparing Eqs. (1)
N
1
and (4) we get Im{x[ n]} = ⋅Re{Q[ k ]} k = n .
N

(4): Im{x[ k ]} =

if n = 0,
⎧ X[0],
Therefore,
11.27 r[ n] = X[〈− n〉 N ] = ⎨

X
N
n
n ≠ 0.
[
],
if

R[ k ] =

N −1

r[ n]WNnk
n=0
N −1

= X [ 0] +

= r[0] +

x[ n] =

r[ n]WNnk
n =1
N −1

= X [ 0] +

N −1

∑ X[ N − n]WNnk

n =1
N −1
X[ n]WN− nk =
X[ n]WN− nk
n=0
n =1

∑ X[n]WN( N − n)k = X[0] + ∑

n =1
1
N

N −1

= N ⋅ x[ k ]. Thus,

⋅R[ k ] k = n .

11.28 Let y[n] denote the result of convolving a length-L sequence x[n] with a length-N sequence
h[n]. The length of y[n] is then L + N – 1. Here L = 16 and N = 9, hence length of y[n] is 24.

Method #1: Direct linear convolution - For a length-L sequence x[n] and a length-N sequence
h[n],
# of real mult. = 2

N

9

n =1

n=1

∑ n + N (L − N − 1) = 2∑ n + 9(16 − 9 − 1) = 135 .

Method # 2: Linear convolution via circular convolution - Since y[n] is of length 24, to get
the correct result we need to pad both sequences with zeros to increase their lengths to 24
before carrying out the circular convolution.
# of real mult. = 24 ×24 = 576.

Not for sale

401

Method #3: Linear convolution via radix-2 FFT - The process involves computing the 16point FFT G[k ] of the length-16 complex sequence g[ n] = x[ n] + j he [ n], where he [n] is
obtained by zero-padding h[n] to length 16. Then, the 16-point DFTs, X [k ] and H e [k ], of
x[n] and he [n] , respectively, are recovered from G[k ] . Finally, the IDFT of the product
Y [ k ] = X[ k ] ⋅ H e [ k ] yields y[n].
Now, the first stage of the 16-point radix-2 FFT requires 0 complex multiplications, the
second stage requires 0 complex multiplications, the third stage requires 4 complex
multiplications, and the last stage requires 6 complex multiplications, resulting in a total of
10 complex multiplications.
# of complex mult. to implement G[k ] = 10
# of complex mult. to recover X [k ] and H e [k ] from G[k ] = 0
# of complex mult. to form Y [k ] = 16
# of complex mult. to form the IDFT of Y [k ] = 10
Hence, the total number of complex mult. = 36
A direct implementation of a complex multiplication requires 4 real multiplications resulting
in a total of 4× 36 = 144 real multiplications for Method #3. However, if a complex multiply
can be implemented using 3 real multiplies (see Problem 11.13), in which case Method #3
requires a total of 3× 36 = 108 real multiplications.
11.29 Let y[n] denote the result of convolving a length-L sequence x[n] with a length-N sequence
h[n]. The length of y[n] is then L + N – 1. Here, L = 16 and N = 10, hence length of y[n] is
25.

Method #1: Direct linear convolution - For a length-L sequence x[n] and a length-N sequence
h[n],
# of real mult. = 2

N

10

n=1

n=1

∑ n + N (L − N − 1) = 2∑ n + 10(16 − 10 − 1) = 160 .

Method # 2: Linear convolution via circular convolution - Since y[n] is of length 24, to get
the correct result we need to pad both sequences with zeros to increase their lengths to 24
before carrying out the circular convolution.
# of real mult. = 25 ×25 = 625.
Method #3: Linear convolution via radix-2 FFT - The process involves computing the 16point FFT G[k ] of the length-16 complex sequence g[ n] = x[ n] + j he [ n], where he [n] is
obtained by zero-padding h[n] to length 16. . Then, the 16-point DFTs, X [k ] and H e [k ], of
x[n] and he [n] , respectively, are recovered from G[k ] . Finally, the IDFT of the product
Y [ k ] = X[ k ] ⋅ H e [ k ] yields y[n].
Now, the first stage of the 16-point radix-2 FFT requires 0 complex multiplications, the
second stage requires 0 complex multiplications, the third stage requires 4 complex

Not for sale

402

multiplications, and the last stage requires 6 complex multiplications resulting in a total of 10
complex multiplications.
# of complex mult. to implement G[k ] = 10
# of complex mult. to recover X [k ] and H e [k ] from G[k ] = 0
# of complex mult. to form Y [k ] = 16
# of complex mult. to form the IDFT of Y [k ] = 10
Hence, the total number of complex mult. = 36
A direct implementation of a complex multiplication requires 4 real multiplications resulting
in a total of 4× 36 = 144 real multiplications for Method #3. However, if a complex multiply
can be implemented using 3 real multiplies (see Problem 11.13), in which case Method #3
requires a total of 3× 36 = 108 real multiplications.
11.30 (a) Since the impulse response of the filter is of length 72, the transform length N should be
greater than 72. If L denotes the number of input samples used for convolution, then L = N –
71. So for every L samples of the input sequence, an N -point DFT is computed and
multiplied with an N -point DFT of the impulse response sequence h[n] (which needs to be
computed only once), and finally an N -point inverse of the product sequence is evaluated.
Hence, the total number R M of complex multiplications required (assuming N is a powerN
2048 ⎤
( N log 2 N + N ) + log 2 N .
of-2) is given by R M = ⎡
2
⎢ N − 71 ⎥

It should be noted that in developing the above expression, multiplications due to twiddle
factors of values ±1 and ± j have not been excluded. The values of R M for different values
of N are as follows:
1) For N = 128, R M = 37312, 2) For N = 256, R M = 28672, 3) For N = 512, R M =
27904, 4) For N = 1024, R M = 38912.
Hence, N = 512 is the appropriate choice for the transform length requiring 27904 complex
multiplications or equivalently, 27904 × 3 = 83712 real multiplications.
Since the first stage of the FFT calculation process requires only multiplications by ±1, the
total number of complex multiplications for N = 128 is actually
N
N
2048 ⎤
RM = ⎡
( N log 2 N + N ) + log 2 N − = 27648, or equivalently, 27648× 3 = 82944
⎢ N − 71 ⎥
2
2
real multiplications.
(b) For direct convolution, # of real multiplications =
N

2

n=1

n + N ( L − N − 1) = 2

72

∑ n + 72(2048 − 72 − 1) = 147456 .

n =1

11.31 (a)

Not for sale

403

⎡1

⎢0
⎢0

⎢0
V8 = ⎢
⎢1
⎢0

⎢0
⎢0

⎡1

⎢1
⎢0

⎢0
V2 = ⎢
⎢0
⎢0

⎢0
⎢0

0 0 0 W80
1 0 0

0

0

0

W81

0

0 1 0

0

0

W82

0 0 1

0

0

0

0 0 0

W84

0

0

1 0 0

0

W85

0

0 1 0

0

0

W86

0 0 1

0

0

0

0 ⎤

0 ⎥
0 ⎥

W83 ⎥
⎥,
0 ⎥
0 ⎥

0 ⎥
W87 ⎥⎦

W80

0

0

0

0

0

W84

0

0

0

0

0

0

1 W80

0

0

0

0

W84

0

0

0
0
0

1

0

0

0

1

0

0

0

1

W80
W84

0

0

0

0

0

1

0

0

0

0

0

1

⎡1

⎢0
⎢1

⎢0
V4 = ⎢
⎢0
⎢0

⎢0
⎢0

0 ⎤

0 ⎥
0 ⎥

0 ⎥ E=
⎥,
0 ⎥
0 ⎥

W80 ⎥
W84 ⎥⎦

⎡1
⎢0
⎢0
⎢0
⎢0
⎢0
⎢0
⎢⎣0

0 W80

0

0 0

0

1

0

W82

0 0

0

0

W80

0

0 0

0

1

0

W82

0 0

0

0

0

0

1 0 W80

0

0

0

0 1

0

0

0

1 0 W80

0

0

0

0 1

0
0
0
0
1
0
0
0

0
0
1
0
0
0
0
0

0
0
0
0
0
0
1
0

0
1
0
0
0
0
0
0

0
0
0
0
0
1
0
0

0
0
0
1
0
0
0
0

0
0

0 ⎤

0 ⎥
0 ⎥

0 ⎥
⎥,
0 ⎥
W82 ⎥

0 ⎥
W82 ⎥⎦

0⎤
0⎥
0⎥
0 ⎥.
0⎥
0⎥
0⎥
1⎥⎦

As can be seen from the above, multiplication by each matrix Vk , k = 1, 2, 3, requires at most
8 complex multiplications.
(b) The transpose of the matrices given in Part (a) are as follows:
⎡ 1 0 0 0 1 0 0 0 ⎤
⎡ 1 0 1 0 0 0 0 0 ⎤
⎢ 0 1 0 0 0 1 0 0 ⎥
⎢ 0 1 0 1 0 0 0 0 ⎥
⎢ 0 0 1 0 0 0 1 0 ⎥
⎢W 0 0 W 0 0 0 0 0 0 ⎥
8
⎢ 0 0 0 1 0 0 0 1 ⎥
⎢ 8

2 0 W2 0
W
0
0 0 0 ⎥
0
0

⎢W

t
t
8
8
0
0
0
0
0
0
W
V8 = ⎢ 8
8
⎥, V4 = ⎢ 0 0 0 0 1 0 1 0 ⎥,
1
1
⎢ 0 W8 0 0 0 W8 0 0 ⎥
⎢ 0 0 0 0 0 1 0 1 ⎥
⎢ 0 0 W2 0 0 0 W2 0 ⎥
⎢ 0 0 0 0 W0 0 W0 0 ⎥
8
8
8
8

2
2⎥
⎢ 0 0 0 W3 0 0 0 W3⎥

W
W
0
0
0
0
0
0
8
8 ⎦

8
8 ⎦

⎡ 1 1 0 0 0 0 0 0 ⎤
⎢W 0 W 4 0 0 0 0 0 0 ⎥
8
⎢ 8

0
0
1 1 0 0 0 0 ⎥

0 0 W80 W84 0 0 0 0 ⎥ E t = E =
V2t = ⎢⎢
,
0 0 0 0 1 1 0 0 ⎥
⎢ 0 0 0 0 W0 W4 0 0 ⎥
8
8

⎢ 0 0 0 0 0 0 1 1 ⎥
⎢⎣ 0 0 0 0 0 0 W80 W84 ⎥⎦

⎡1
⎢0
⎢0
⎢0
⎢0
⎢0
⎢0
⎢⎣0

0
0
0
0
1
0
0
0

0
0
1
0
0
0
0
0

0
0
0
0
0
0
1
0

0
1
0
0
0
0
0
0

0
0
0
0
0
1
0
0

0
0
0
1
0
0
0
0

0⎤
0⎥
0⎥
0 ⎥.
0⎥
0⎥
0⎥
1⎥⎦

It is easy to show that the flow-graph representation of D8 = E t V2t V4t V8t is precisely the 8point DIF FFT algorithm of Figure 11.28.

Not for sale

404

11.32 X [2l] =

n+

N
2

N −1

n=0

x[ n]WN2 ln =

N
−1
2

N −1

n=0

n= N / 2

∑ x[n]WN2ln , 0 ≤ l ≤ 2 − 1.

x[ n]WN2ln +

N

Replacing n by

in the right-most sum we get

X [ 2l ] =

N
−1
2

N
−1
2

n=0

n=0

N
−1
2

∑ x[n]WN2ln + ∑ x[n + 2 ]WN2ln = ∑ ⎛⎜⎝ x[n] + x[n + 2 ]⎞⎟⎠WNln/ 2 , 0 ≤ l ≤ 2 − 1.
N

N

N

n=0

Likewise,

X[ 4l + 1] =

N
−1
4

n=0

where 0 ≤ l ≤

x[ n]WN( 4l + 1)n

N
4

+

+

n=

x[ n]WN( 4l + 1)n

3N
4

N
4

+

N
2

N −1

∑ x[n]WN(4l +1)n ,

n=

3N
4

N
2

in the third

in the fourth sum, we get

N
−1
4

N
−1
4

n=0
N
−1
4

n=0

∑ x[n]WN4lnWNn + ∑ x[n + 4 ]WN4lnWNn WNlN WNN / 4
N

N
−1
4

∑ x[n + 2 ]WN4lnWNn WN2lN WNN / 2 + ∑ x[n +
N

N
−1
4

n=0
WNN / 2

N

N

n=0

N
−1
4

N
−1
4

N
−1
4

n=0

n=0

3N
]WN4lnWNn WN3lN WN3 N / 4 .
4

= −1, and WN3 N / 4 = + j. Therefore,

∑ ⎧⎨⎩⎛⎜⎝ x[n] − x[n + 2 ]⎞⎟⎠ − j⎛⎜⎝ x[n + 4 ] − x[n +

Similarly, X [ 4l + 3] =

N
3 N ⎞ ⎫ n ln
]⎟ ⎬W W
, 0 ≤ l ≤ − 1.
4 ⎠⎭ N N / 4
4

∑ x[n]WN(4l + 3)n + ∑ x[n + 4 ]WN(4l + 3)nWN(4l + 3)N / 4
N

N
−1
4

∑ x[n + 2 ]WN(4l + 3)nWN(4l + 3)N / 2 + ∑ x[n +
N

n=0

=

x[ n]WN( 4l + 1)n

in the second sum, n by n +

Now, WNlN = WN2 lN = WN3lN = 1, WNN / 4 = − j,

+

n=

n=0

X [ 4l + 1] =

+

3N
−1
4

N
4

− 1. Replacing n by n +

sum, and n by n +

X [ 4l + 1] =

N
−1
2

n=0

N
−1
4

N
−1
4

n=0

n=0

3N
]WN( 4l + 3)nWN( 4l + 3)3 N / 4
4

∑ x[n]WN4lnWN3n + ∑ x[n + 4 ]WN4lnWN3nWNlN WN3N / 4
N

Not for sale

405

+

N
−1
4

N
−1
4

∑ x[n + 2 ]WN4lnWN3nWN2lN WN6 N / 4 + ∑ x[n + 4 ]WN4lnWN3nWN3lN WN9 N / 4
N

n=0

=

N
−1
4

n=0

∑ ⎧⎨⎩⎛⎜⎝ x[n] − x[n + 2 ]⎞⎟⎠ + j⎛⎜⎝ x[n + 4 ] − x[n +

n=0

N

N

N

N
3 N ⎞ ⎫ 3 n ln
]⎟ ⎬W WN / 4 , 0 ≤ l ≤ − 1.
4
4 ⎠⎭ N

The butterfly here is as shown below which is seen to require two complex multiplications.

11.33 From the flow-graph of the 8-point split-radix FFT algorithm given below it can be seen
that the total number of complex multiplications required is 2. On the other hand, the total
number of complex multiplications required for a standard DIF FFT algorithm is also 2.

11.34 If multiplications by ± j, ± 1 are ignored, the flow-graph shown below requires 8 complex
multiplications = 24 real multiplications. A radix-2 DIF 16-point FFT algorithm, on the
other hand requires 10 complex multiplications = 30 real multiplications.

Not for sale

406

0 ≤ n1 ≤ 3,
11.35 (a) N = 12. Choose N1 = 4 and N 2 = 3. Thus, n = n1 + 4 n2 , ⎧⎨
and
0
⎩ ≤ n 2 ≤ 2,
⎧ 0 ≤ k1 ≤ 3, The corresponding index mappings are indicated below:
k = 3k1 + k 2 , ⎨
⎩0 ≤ k 2 ≤ 2.
k
n
1
3
2
1
3
2
k2 1 0
n2 1 0
X[0] X[3] X[6] X[9]
x[0] x[1] x[2] x[3]
0
0
X[1] X[4] X[7] X[10]
x[4] x[5] x[6] x[7]
1
1
2 X[2] X[5] X[8] X[11]
2 x[8] x[9] x[10] x[11]

0 ≤ n1 ≤ 2,
N = 15. Choose N1 = 3 and N 2 = 5. Thus, n = n1 + 3n2 , ⎧⎨
and
⎩0 ≤ n2 ≤ 4,
⎧ 0 ≤ k1 ≤ 2, The corresponding index mappings are indicated below:
k = 5k1 + k 2 , ⎨
⎩0 ≤ k 2 ≤ 4.

(b)

n
n2 1 0
x[0]
0
x[3]
1
x[6]
2
3
x[9]
4

1
x[1]
x[4]
x[7]

2
x[2]
x[5]
x[8]

x[10] x[11]

x[12] x[13] x[14]

Not for sale

k
k2 1 0
X[0]
0
X[1]
1
X[2]
2
3
X[3]
4

1

2

X[5] X[10]
X[6] X[11]
X[7] X[12]
X[8] X[13]

X[4] X[9] X[14]

407

0 ≤ n1 ≤ 6,
(c) N = 21. Choose N1 = 7 and N 2 = 3. Thus, n = n1 + 7n2 , ⎧⎨
and
⎩0 ≤ n2 ≤ 2,
⎧ 0 ≤ k1 ≤ 6, The corresponding index mappings are indicated below:
k = 3k1 + k 2 , ⎨
⎩0 ≤ k 2 ≤ 2.
n
n2 1 0
x[0]
0
1
x[7]
2 x[14]

1
x[1]
x[8]

2

3

4

5

x[2] x[3] x[4] x[5]
x[9] x[10] x[11] x[12]
x[15] x[16] x[17] x[18] x[19]

6
x[6]
x[13]
x[20]

k
4
5
6
1
3
2
k2 1 0
X[18]
X[3]
X[15]
X[6] X[9] X[12]
X[0]
0
1
X[1] X[4] X[7] X[10] X[13] X[16] X[19]
2
X[2] X[5] X[8] X[11] X[14] X[17] X[20]

0 ≤ n1 ≤ 6,
(d) N = 35. Choose N1 = 7 and N 2 = 5. Thus, n = n1 + 7n2 , ⎧⎨
and
0
⎩ ≤ n 2 ≤ 4,
⎧ 0 ≤ k1 ≤ 6, The corresponding index mappings are indicated below:
k = 5k1 + k 2 , ⎨
⎩0 ≤ k 2 ≤ 4.
n
1
2
n2 1 0
x[0] x[1] x[2]
0
x[7] x[8] x[9]
1
2 x[14] x[15] x[16]
3 x[21] x[22] x[23]
4

3

4

5

x[3] x[4] x[5]
x[10] x[11] x[12]
x[17] x[18] x[19]

x[13]
x[20]

x[24] x[25] x[26]

x[27]

x[28] x[29] x[30] x[31] x[32] x[33]

x[34]

k
4
5
1
3
2
k2 1 0
X[0] X[5] X[10] X[15] X[20] X[25]
0
1
X[1] X[6] X[11] X[16] X[21] X[26]
2 X[2] X[7] X[12] X[17] X[22] X[27]
3 X[3] X[8] X[13] X[18] X[23] X[28]
4

6
x[6]

6
X[30]
X[31]
X[32]
X[33]

X[4] X[9] X[14] X[19] X[24] X[29] X[34]

11.36 (a) N = 12. Choose N1 = 4 and N 2 = 3. A = 3, B = 4, C = 3〈3 −1 〉 4 = 9,
⎧ 0 ≤ n1 ≤ 3,
D = 4〈 4 −1 〉 3 = 4. Thus, n = 〈3n1 + 4 n2 〉 12 , ⎨
k = 〈 9k1 + 4 k 2 〉 12 ,
⎩0 ≤ n2 ≤ 2,
The corresponding index mappings are indicated below:
n
1
3
2
n2 1 0
x[0] x[3] x[6] x[9]
0
x[4] x[7] x[10] x[1]
1
2 x[8] x[11] x[2] x[5]

⎧ 0 ≤ k1 ≤ 3,
⎨0 ≤ k ≤ 2.
2

k
1
3
2
k2 1 0
X[0] X[9] X[6] X[3]
0
X[4] X[1] X[10] X[7]
1
2 X[8] X[5] X[2] X[11]

(b) N = 15. Choose N1 = 3 and N 2 = 5. A = 5, B = 3, C = 5〈5 −1 〉 3 = 10, D = 3〈3 −1 〉 5 = 6.

Not for sale

408

0 ≤ n1 ≤ 2,
Thus, n = 〈 5n1 + 3n2 〉 15 , ⎧⎨
k = 〈10 k1 + 6 k 2 〉 15 ,
0
⎩ ≤ n 2 ≤ 4,
The corresponding index mappings are indicated below:
n
n2 1 0
x[0]
0
x[3]
1
x[6]
2
3 x[9]
4

1
2
x[5] x[10]
x[8] x[13]
x[11] x[1]
x[14] x[4]

x[12] x[2]

x[7]

⎧ 0 ≤ k1 ≤ 2,
⎨0 ≤ k ≤ 4.
2

k
1
2
k2 1 0
X[0] X[10] X[5]
0
X[6] X[1] X[11]
1
X[12] X[7] X[2]
2
3
X[3] X[13] X[8]
4 X[9] X[4] X[14]

(c) N = 21. Choose N1 = 7 and N 2 = 3. A = 3, B = 7, C = 3〈3 −1 〉 4 = 3 × 5 = 15,
⎧ 0 ≤ n1 ≤ 6,
D = 7〈 7 −1 〉 3 = 7 × 1 = 7. Thus, n = 〈3n1 + 7n2 〉 21 , ⎨
⎩0 ≤ n2 ≤ 2,
⎧ 0 ≤ k1 ≤ 6, The corresponding index mappings are indicated
k = 〈15k1 + 7k 2 〉 21 , ⎨
⎩0 ≤ k 2 ≤ 2.
below:
n
4
5
6
1
3
2
n2 1 0
x[18]
x[0] x[3] x[6] x[9] x[12] x[15]
0
1
x[7] x[10] x[13] x[16] x[19] x[1] x[4]
2 x[14] x[17] x[20] x[2] x[5] x[8] x[11]

k
4
5
6
1
3
2
k2 1 0
X[0] X[15] X[9] X[3] X[18] X[12] X[6]
0
1
X[7] X[1] X[16] X[10] X[4] X[19] X[13]
2
X[14] X[8] X[2] X[17] X[11] X[5] X[20]

(d) N = 35. Choose N1 = 7 and N 2 = 5. A = 5, B = 7, C = 5〈 5 −1 〉 7 = 5 × 3 = 15,
⎧ 0 ≤ n1 ≤ 6,
D = 7〈7 −1 〉 5 = 7 × 3 = 21. Thus, n = 〈 5n1 + 7n2 〉 35 , ⎨
⎩0 ≤ n2 ≤ 4,
⎧ 0 ≤ k1 ≤ 6, The corresponding index mappings are indicated
k = 〈15k1 + 21k 2 〉 35 , ⎨
⎩0 ≤ k 2 ≤ 4.
below:
n
1
3
2
n2 1 0
x[0] x[5] x[10] x[15]
0
x[7] x[12] x[17] x[22]
1
2 x[14] x[19] x[24] x[29]
3 x[21] x[26] x[31] x[1]
4

4

5

x[20] x[25]
x[27] x[32]
x[34] x[4]

6
x[30]
x[2]
x[9]

x[6] x[11]

x[16]

x[28] x[33] x[3] x[8] x[13] x[18]

x[23]

k
4
5
6
1
3
2
k2 1 0
X[0] X[15] X[30] X[10] X[25] X[5] X[20]
0
X[21] X[1] X[16] X[31] X[11] X[26] X[6]
1
2
X[7] X[22] X[2] X[17] X[32] X[12] X[27]
3 X[28] X[8] X[23] X[3] X[18] X[33] X[13]
4

X[14] X[29] X[9] X[24] X[4] X[19] X[34]

Not for sale

409

11.37 N = 12. Choose N1 = 4 and N 2 = 3. A = 3, B = 4, C = 3〈3 −1 〉 4 = 9, D = 4〈 4 −1 〉 3 = 4.
0 ≤ n1 ≤ 3,
⎧ 0 ≤ k1 ≤ 3,
Thus, n = 〈3n1 + 4 n2 〉 12 , ⎧⎨
k = 〈 9k1 + 4 k 2 〉 12 , ⎨
⎩0 ≤ n2 ≤ 2,
⎩0 ≤ k 2 ≤ 2.
The corresponding index mappings are indicated below:
n
1
3
2
n2 1 0
x[0] x[3] x[6] x[9]
0
x[4] x[7] x[10] x[1]
1
2 x[8] x[11] x[2] x[5]

k
1
3
2
k2 1 0
X[0] X[9] X[6] X[3]
0
X[4] X[1] X[10] X[7]
1
2 X[8] X[5] X[2] X[11]

0 ≤ n1 ≤ 3,
Alternately, k = 〈 9n1 + 4 n2 〉 12 , ⎧⎨
k = 〈3k1 + 4 k 2 〉 12 ,
⎩0 ≤ n2 ≤ 2,
The corresponding index mappings are indicated below:
n
1
3
2
n2 1 0
x[0] x[9] x[6] x[3]
0
x[4] x[1] x[10] x[7]
1
2 x[8] x[5] x[2] x[11]

⎧ 0 ≤ k1 ≤ 3,
⎨0 ≤ k ≤ 2.
2

k
1
3
2
k2 1 0
X[0] X[3] X[6] X[9]
0
X[4] X[7] X[10] X[1]
1
2 X[8] X[11] X[2] X[5]

Hence, X [2 k ] = Y [2 k ], and X [2k + 1] = Y [〈6 + (2k + 1)〉12 ], k = 0,1,K,5.
11.38 (a) N = 6. Choose N1 = 2 and N 2 = 3. A = 3, B = 2, C = 3〈3 −1 〉 2 = 3, D = 2〈 2 −1 〉 3 = 4.
0 ≤ n1 ≤ 1,
⎧ 0 ≤ k1 ≤ 1,
Thus, n = 〈3n1 + 2 n2 〉 6 , ⎧⎨
k = 〈3k1 + 4 k 2 〉 6 , ⎨
⎩0 ≤ n2 ≤ 2,
⎩0 ≤ k 2 ≤ 2.
The corresponding index mappings are indicated below:
n
1
n2 1 0
x[0] x[3]
0
x[2] x[5]
1
2 x[4] x[1]
x[0]
x[3]
x[2]
x[5]
x[4]
x[1]

n
1
k2 1 0
G[0,0] G[1,0]
0
G[0,1] G[1,1]
1
2 G[0,2] G[1,2]
G[0,0]

2-pt
DFT G[1,0]

k
1
k2 1 0
X[0] X[3]
0
X[4] X[1]
1
2 X[2] X[5]
X[0]

3-pt
DFT

X[4]
X[2]

G[0,1]

2-pt
DFT G[1,1]
G[0,2]

2-pt
DFT G[1,2]

Not for sale

X[3]

3-pt
DFT

X[1]
X[5]

410

(b) N = 10. Choose N1 = 2 and N 2 = 5. A = 5, B = 2, C = 5〈5 −1 〉 2 = 5, D = 2〈 2 −1 〉 5 = 6.
0 ≤ n1 ≤ 4,
⎧ 0 ≤ k1 ≤ 4,
Thus, n = 〈 5n1 + 2 n2 〉 10 , ⎧⎨
k = 〈 5k1 + 6 k 2 〉 10 , ⎨
n

,
0
2
2

⎩0 ≤ k 2 ≤ 2.
The corresponding index mappings are indicated below:
n
1
n2 1 0
x[0] x[5]
0
x[2] x[7]
1
x[4] x[9]
2
x[6] x[1]
3
x[8] x[3]
4
x[0]
x[5]
x[2]
x[7]

x[4]
x[9]
x[6]

x[1]
x[8]

x[3]

n
1
k2 1 0
G[0,0] G[1,0]
0
G[0,1] G[1,1]
1
G[0,2] G[1,2]
2
G[0,3] G[1,3]
3
G[1,4]
G[0,4]
4

k
1
k2 1 0
X[0] X[5]
0
X[6] X[1]
1
2 X[2] X[7]
3 X[8] X[3]
4 X[4] X[9]

G[0,0]

X[0]

2-pt
DFT G[1,0]
G[0,1]

2-pt
DFT G[1,1]

X[6]

5-pt
DFT

X[2]
X[8]

X[4]

G[0,2]

2-pt
DFT G[1,2]
X[5]

G[0,3]

2-pt
DFT G[1,3]
G[0,4]

2-pt
DFT G[1,4]

X[1]

5-pt
DFT

X[7]

X[3]
X[9]

(c) N = 12. Choose N1 = 3 and N 2 = 4. A = 4, B = 3, C = 4〈 4 −1 〉 3 = 4, D = 3〈3 −1 〉 4 = 3.
0 ≤ n1 ≤ 2,
⎧ 0 ≤ k1 ≤ 2,
Thus, n = 〈 4 n1 + 3n2 〉 12 , ⎧⎨
k = 〈 4 k1 + 3k 2 〉 12 , ⎨
n
0

3
,
2

⎩0 ≤ k 2 ≤ 3.
The corresponding index mappings are indicated below:
n
n2 1 0
x[0]
0
x[3]
1
2 x[6]
3 x[9]

1
2
x[4] x[8]
x[7] x[11]
x[10] x[2]
x[1] x[5]

k
1
2
n2 1 0
G[0,0] G[1,0] G[2,0]
0
G[0,1] G[1,1] G[2,1]
1
2 G[0,2] G[1,2] G[2,2]
3 G[0,3] G[1,3] G[2,3]

Not for sale

k
k2 1 0
X[0]
0
X[3]
1
2 X[6]
3 X[9]

1

2

X[4]
X[7]
X[10]
X[1]

X[8]
X[11]
X[2]
X[5]

411

G[0,0]

x[0]
x[4]
x[8]

3-pt
DFT G[2,0]

x[11]

G[2,2]

x[9]

G[0,3]

x[1]
x[5]

4-pt
DFT

3-pt G[1,2]

x[2]

X[6]

X[4]

G[0,2]

DFT

X[3]

X[9]

3-pt G[1,1]
DFT G[2,1]

x[6]
x[10]

4-pt
DFT

G[0,1]

x[3]
x[7]

X[0]

G[1,0]

X[7]
X[10]

X[1]
X[8]

4-pt
DFT

G[1,3]

3-pt
DFT G[2,3]

X[11]

X[2]
X[5]

(d) N = 15. Choose N1 = 5 and N 2 = 3. A = 3, B = 5, C = 3〈3 −1 〉 5 = 10,
⎧ 0 ≤ n1 ≤ 4,
⎧0 ≤ k1 ≤ 4,
D = 5〈 5 −1 〉 3 = 6. Thus, n = 〈 5n1 + 3n2 〉 10 , ⎨
k = 〈10 k1 + 6 k 2 〉 10 , ⎨
n
0

2
,
2

⎩0 ≤ k 2 ≤ 2.
The corresponding index mappings are indicated below:
n
n
1
2
3
2
4
1
3
k2 1 0
n2 1 0
G[0,0] G[1,0] G[2,0] G[3,0]
0
x[0] x[3] x[6] x[9] x[12]
0
1
x[5] x[8] x[11] x[14] x[2]
1
G[0,1] G[1,1] G[2,1] G[3,1]
x[1]
x[10]
x[13]
x[4]
G[0,2] G[1,2] G[2,2] G[3,2]
x[7]
2
2

k
k2 1 0
X[0]
0
1 X[10]
X[5]
2

1

2

3

4

X[6] X[12] X[3] X[9]
X[1] X[7] X[13] X[4]
X[11] X[2] X[8] X[14]

Not for sale

412

G[0,0]

x[0]

G[0,1]

x[3]

5-pt G[0,2]
DFT G[0,3]

x[6]
x[9]
x[12]

G[0,4]

G[1,0]
G[1,1]

x[5]
x[8]

5-pt G[1,2]
DFT G[1,3]

x[11]
x[14]
x[2]

X[0]

3-pt
DFT

X[5]

3-pt
DFT

x[10]
x[13]

5-pt G[2,2]
DFT G[2,3]

x[1]
x[4]

G[2,4]

x[7]

X[6]
X[1]
X[11]
X[12]

3-pt
DFT

X[7]
X[2]

G[1,4]
G[2,0]
G[2,1]

X[10]

X[3]

3-pt
DFT

X[13]

X[8]

3-pt
DFT

X[9]
X[4]
X[14]

11.39 Note that 1536 = 256 × 6. Now an N -point DFT , with N divisible by 6, can be computed

as follows: X[ k ] =

N −1

∑ x[n]WNnk = X 0 [〈 k〉 N / 6 ] + WNk X1[〈 k〉 N / 6 ] + WN2k X 2 [〈 k〉 N / 6 ]

n=0
k
3
+ WN X 3 [〈 k 〉 N / 6 ] + WN4 k X 4 [〈 k 〉 N / 6 ] + WN5k X 5 [〈 k 〉 N / 6 ], where
N
−1
6
X l [〈 k 〉 N / 6 ] =
x[6r + l]WNrk/ 6 , 0 ≤ l ≤ 5. For N = 1536, we thus get
r =0
k X [〈 k 〉
2k
3k
X[ k ] = X 0 [〈 k 〉 256 ] + W1536
1
256 ] + W1536 X 2 [ 〈 k 〉 256 ] + W1536 X 3 [ 〈 k 〉 256 ]
5111
rk ,
4
5
k
k
+ W1536 X 4 [〈 k 〉 256 ] + W1536 X 5 [〈 k 〉 256 ], where X l [〈 k 〉 256 ] =
x[6r + l]W256
r =0

x000[n] 256-point X000 [ k 256 ]
6
DFT

x[n]
z

6
z
z

x [n] 256-point X [ k 256 ]
010
6 010
DFT

6
z
6
z

x001[n] 256-point X001 [ k 256 ]
DFT

x011[n] 256-point X [
011 k 256 ]
DFT
x100[n]

]
X [
256-point 100 k 256
DFT

x101[n] 256-point X101 [ k 256 ]
6
DFT

Not for sale

0 ≤ l ≤ 5.

X[k]
k
W768
k
W768

k
W1536

k
W768
k
W768

413

Now an N-point FFT algorithm requires
multiplications and

N
6

N
6

N
2

log 2 N complex multiplications and N log 2 N

-point FFT algorithm requires

N
log 2 ( N / 6)
12

complex

log 2 ( N / 6) complex additions. In addition, we need 5N complex

multiplications and 5N complex additions to compute the N-point DFT X[k]. Hence, for N =
1536, the evaluation of X [k ] using six 256-point FFT modules requires

= 128 × log 2 (256) + 5 × 1536 = 8704 complex multiplications and

N
6

N
log 2 ( N / 6) + 5N
12

log 2 ( N / 6) + 5N

= 256 × log 2 (256) + 5 × 1536 = 9728 complex additions. It should be noted that a direct
computation of the 3072-point DFT would require 6431296 complex multiplications and
11.40 (a) # of zero-valued samples to be added is 512 – 498 = 14.
(b) Direct computation of a 512-point DFT of a length-498 sequence requires (498)2 =
248004 complex multiplications and 497 ×498 = 247506 complex additions.
(c) A 512-point Cooley-Tukey type FFT algorithm requires 256 × log 2 (512) = 2304 complex
multiplications and 512 × log 2 (512) = 46308 complex additions.
11.41 z l = α l . Hence, AoVo− l e jθ o e − jlφ o = α l . Since α is real, we have Ao = 1, Vo = 1 / α,
θ o = 0, φ o = 0.
11.42 (a) Y ( z ) = H ( z ) X ( z ) or y[ 0] + y[1]z −1 + y[ 2]z −2 = (h[ 0] + h[1]z −1 )( x[ 0] + x[1]z −1 ). Now,
Y ( z 0 ) = Y (−1) = y[0] − y[1] + y[2] = H (−1) X (−1) = (h[0] − h[1])( x[0] − x[1]),

Y ( z1 ) = Y (∞) = y[0] = H (∞) X (∞ ) = h[0]x[0],
Y ( z 2 ) = Y (1) = y[0] + y[1] + y[2] = H (1) X (1) = {h[0] + h[1])( x[0] + x{1}).
From Eqs. (6.114) and (6.115), we arrive at
I ( z)
I (z)
I ( z)
Y (z0 ) + 1
Y ( z1 ) + 2
Y ( z 2 ), where I 0( z ) = (1 − z1z −1 )(1 − z 2 z −1 ),
Y ( z) = 0
I 0 (z0 )
I1 ( z1 )
I 2 (z2 )

I 1( z ) = (1 − z 0 z −1 )(1 − z 2 z −1 ), I 2( z ) = (1 − z 0 z −1 )(1 − z1z −1 ). Therefore,
I 0 ( z)
(1 − z1 z −1 )(1 − z 2 z −1 )
1
=
= − z −1 (1 − z −1 ),
2
I 0 ( z 0 ) (1 − z1 z 0−1 )(1 − z 2 z 0−1 )
I1 ( z) (1 − z 0 z −1 )(1 − z 2 z −1 )
=
= (1 − z − 2 ), and

1

1
I1 ( z1 ) (1 − z 0 z1 )(1 − z 2 z1 )
I 2 ( z) (1 − z 0 z −1 )(1 − z1 z −1 ) 1 −1
=
= z (1 + z −1 ). Hence,
1
1

I 2 ( z 0 ) (1 − z 0 z 2 )(1 − z1 z 2 ) 2

Not for sale

414

Y ( z ) = − z −1 (1 − z −1 )Y ( z 0 ) + (1 − z − 2 )Y ( z1 ) + z −1 (1 + z −1 )Y ( z 2 )
1
2

1
2

1
1
1
1
= Y ( z1 ) + ⎛⎜ − Y ( z 0 ) + Y ( z 2 ) ⎞⎟ z −1 + ⎛⎜ Y ( z 0 ) − Y ( z1 ) + Y ( z 2 ) ⎞⎟ z − 2
2
2
⎝ 2

⎝2

1
1
= h[0]x[0] + ⎛⎜ − (h[0] − h[1])( x[0] − x[1]) + (h[0] + h[1])( x[0] + x[1]) ⎞⎟ z −1
2
⎝ 2

1
1
+ ⎛⎜ (h[0] − h[1])( x[0] − x[1]) − h[0]x[0] + (h[0] + h[1])( x[0] + x[1]) ⎞⎟ z − 2
2
⎝2

1

2
= h[ 0] x[ 0] + (h[ 0] x[1] + h[1] x[ 0])z + h[1] x[1]z .

Ignoring the multiplications by

1
,
2

computation of the coefficients of Y ( z ) require the values

of Y ( z 0 ), Y ( z1 ), and Y ( z2 ) , which can be evaluated using only 3 multiplications.
(b) Y ( z ) = H ( z ) X ( z ) or

(

)(

)

y[ 0] + y[1]z −1 + y[2]z − 2 + y[3]z − 3 + y[ 4]z − 4 = h[ 0] + h[1]z −1 + h[ 2]z − 2 x[ 0] + x[1]z −1 + x[ 2]z − 2 .
Now,
1
2

Y ( z 0 ) = Y (− ) = (h[0] − 2h[1] + 4h[2])( x[0] − 2 x[1] + 4 x[2]),

Y ( z1 ) = Y (−1) = (h[0] − h[1] + h[2])( x[0] − x[1] + x[2]), Y ( z2 ) = Y (∞) = h[0]x[0],
Y ( z3 ) = Y (1) = (h[0] + h[1] + h[2])( x[0] + x[1] + x[2]),
1
2

Y ( z 4 ) = Y ( ) = (h[0] + 2h[1] + 4h[2])( x[0] + 2 x[1] + 4 x[2]).
From Eqs. (6.114) and (6.115), we arrive at
I ( z)
I ( z)
I ( z)
I (z)
I ( z)
Y (z) = 0
Y (z0 ) + 1
Y ( z1 ) + 2
Y (z2 ) + 3
Y ( z3 ) + 4
Y ( z 4 ), where
I 0 (z0 )
I1 ( z1 )
I 2 (z2 )
I 3 ( z3 )
I 4 (z 4 )

I 0 ( z) = (1 − z1z −1 )(1 − z 2 z −1 )(1 − z3 z −1 )(1 − z 4 z −1 ),
I 1( z) = (1 − z0 z −1 )(1 − z 2 z −1 )(1 − z3 z −1 )(1 − z 4 z −1 ),
I 2 ( z) = (1 − z0 z −1 )(1 − z1z −1 )(1 − z3 z −1 )(1 − z 4 z −1 ),
I 3( z) = (1 − z 0 z −1 )(1 − z1z −1 )(1 − z 2 z −1 )(1 − z 4 z −1 ),
I 4( z) = (1 − z 0 z −1 )(1 − z1z −1 )(1 − z 2 z −1 )(1 − z3 z −1 ). Therefore,
I 0 ( z) (1 − z1z −1 )(1 − z2 z −1 )(1 − z3 z −1 )(1 − z4 z −1 ) 1 −1
1
=
= z (1 − z −1 )(1 − z − 2 ),

1

1
1
1
2
I 0 ( z0 ) (1 − z1z0 )(1 − z2 z0 )(1 − z3 z0 )(1 − z4 z0 ) 12

I1 ( z) (1 − z 0 z −1 )(1 − z 2 z −1 )(1 − z3 z −1 )(1 − z 4 z −1 )
1
2
=
= − z −1 (1 − z −1 )(1 − z − 2 ),
4
3
I1 ( z1 ) (1 − z 0 z1−1 )(1 − z 2 z1−1 )(1 − z3 z1−1 )(1 − z 4 z1−1 )
I 2 ( z) (1 − z 0 z −1 )(1 − z1 z −1 )(1 − z3 z −1 )(1 − z 4 z −1 )
1
=
= (1 − z − 2 )(1 − z − 2 ),
4
I 2 ( z 2 ) (1 − z 0 z 2−1 )(1 − z1 z 2−1 )(1 − z3 z 2−1 )(1 − z 4 z 2−1 )
I 3 ( z) (1 − z 0 z −1 )(1 − z1 z −1 )(1 − z 2 z −1 )(1 − z 4 z −1 ) 2 −1
1
=
= z (1 + z −1 )(1 − z − 2 ),
1
1
1
1

4
3
I 3 ( z3 ) (1 − z 0 z3 )(1 − z1 z3 )(1 − z 2 z3 )(1 − z 4 z3 )
Not for sale

415

(1 − z 0 z −1 )(1 − z1 z −1 )(1 − z 2 z −1 )(1 − z3 z −1 )
I 4 ( z)
1
1
=
= − z −1 (1 + z −1 )(1 − z − 2 ). Hence,
1
1
1
1

2
12
I 4 ( z 4 ) (1 − z 0 z 4 )(1 − z1 z 4 )(1 − z 2 z 4 )(1 − z3 z 4 )
Y ( z) =

1 −1
(z
12

− z − 2 − z − 3 + z − 4 )Y ( z 0 ) − ( z −1 − z − 2 − z − 3 + z − 4 )Y ( z1 )
1
2

+ (1 −

1
4
5 −2
z
4

1
4

2
3

+ z − 4 )Y ( z 2 ) + ( z −1 + z − 2 −
1
4

2
3

1
4
1 −3
z
4

− z − 4 )Y ( z3 )
1
4

1
1 −1 1 −1
+ z − z − 3 − z − 4 )Y ( z 4 )
(z
2
2
12
2
2
1
Y ( z1 ) + Y ( z3 ) − Y ( z 4 ) ⎞⎟ z −1
3
3
12

1
= Y ( z 2 ) + ⎛⎜ Y ( z 0 ) −
12

1
2
5
2
1
+ ⎛⎜ − Y ( z 0 ) + Y ( z1 ) − Y ( z 2 ) + Y ( z3 ) − Y ( z 4 ) ⎞⎟ z − 2
3
4
3
24
⎝ 24

1
1
1
1
+ ⎛⎜ − Y ( z 0 ) + Y ( z1 ) − Y ( z3 ) + Y ( z 4 ) ⎞⎟ z − 3
12
6
6
12

1
1
1
1
1
+ ⎛⎜ Y ( z 0 ) − Y ( z1 ) + Y ( z 2 ) − Y ( z3 ) + Y ( z 4 ) ⎞⎟ z − 4 .
6
4
6
24
⎝ 24

Substituting the expressions for Y ( z 0 ), Y ( z1 ), Y ( z 2 ), Y ( z 3 ), and Y ( z 4 ), in the above
equation, we then arrive at the expressions for the coefficients {y[ n]} in terms of the
coefficients {h[ n]} and {x[ n]}. Thus, y[0] = Y ( z 2 ) = h[0]x[0],
1
2
2
1
Y ( z 0 ) − Y ( z1 ) + Y ( z3 ) − Y ( z 4 ) = h[0]x[1] + h[1]x[0],
12
3
3
12
1
2
5
2
1
y[2] = − Y ( z 0 ) + Y ( z1 ) − Y ( z 2 ) + Y ( z3 ) − Y ( z 4 ) = h[0]x[2] + h[1]x[1] + h[2]x[0],
24
3
4
3
24
1
1
1
1
y[3] = − Y ( z 0 ) + Y ( z1 ) − Y ( z3 ) + Y ( z 4 ) = h[1]x[2] + h[2]x[1],
12
6
6
12
1
1
1
1
1
y[ 4] = Y ( z 0 ) − Y ( z1 ) + Y ( z 2 ) − Y ( z3 ) + Y ( z 4 ) = h[2]x[2].
24
6
4
6
24
1 2 5 1 1
1
Hence, ignoring the multiplications by , , , , , and , computation of the
12 3 4 4 6
24

y[1] =

coefficients of Y ( z ) require the values of Y ( z 0 ), Y ( z1 ), Y ( z 2 ), Y ( z3 ), and Y ( z 4 ), which can
be evaluated using only 5 multiplications.

(

)(

11.43 Y ( z ) = H ( z ) X ( z ) or y[ 0] + y[1]z − 1 + y[2]z − 2 = h[ 0] + h[1]z − 1 x[ 0] + x[1]z −1
h[ 0] x[0] + (h[ 0] x[1] + h[1] x[ 0]) z −1

)

+ h[ 2] x[[ 2]z −2 .

Hence, y[ 0] = h[ 0] x[ 0],
=
y[1] = h[ 0] x[1] + h[1] x[ 0], y[ 2] = h[1] x[1]. Now,
(h[ 0] + h[1])( x[ 0] + x[1]) − h[ 0] x[ 0] − h[1] x[1] = h[ 0] x[1] + h[1] x[0] = y[1]. As a result,
evaluation of H ( z ) X ( z ) requires the computation of 3 products, h[ 0] x[ 0], h[1] x[1] , and
(h[ 0] + h[1])( x[ 0] + x[1]) . In addition, it requires 4 additions, h[0] + h[1], x[ 0] + x[1], and
(h[ 0] + h[1])( x[ 0] + x[1]) − h[ 0] x[ 0] − h[1] x[1] .
11.43 Let the two length- N sequences be denoted by {h[ n]} and {x[ n]}. Denote the sequence

Not for sale

416

generated by the linear convolution of h[n] and x[n] as y[n]. Let H ( z) =

X ( z) =

N −1

∑ n = 0 x[n]z − n .

and

Rewrite H ( z ) and X ( z ) in the form H ( z) = H 0 ( z) + z − N / 2 H1 ( z), and

X ( z) = X 0 ( z) + z − N / 2 X1 ( z), where H 0 ( z ) =
( N / 2 ) −1

N −1

∑ n = 0 h[n]z − n

( N / 2) −1

∑n=0

h[ n]z − n ,

( N / 2) −1

∑ n = 0 h[n + 2 ]z − n , X 0 (z) = ∑ n = 0 x[n]z − n ,
( N / 2) −1
N
X1 ( z ) = ∑
x[ n + ]z − n . Therefore, we can write
n=0
2

H1 ( z ) =

N

(

)(

)

Y ( z) = H 0 ( z ) + z − N / 2 H1 ( z) X 0 ( z ) + z − N / 2 X1 ( z )
= H 0 ( z ) X 0 ( z) + z − N / 2 (H 0 ( z) X1 ( z) + H1 ( z) X 0 ( z)) + z − N H1 ( z) X1 ( z)
= Y0 ( z) + z − N / 2Y1 ( z ) + z − N Y2 ( z ), where Y0 ( z) = H 0 ( z) X 0 ( z),
Y1 ( z ) = H 0 ( z) X1 ( z ) + H1 ( z) X 0 ( z ), and Y2 ( z) = H1 ( z) X1 ( z). Note that Y0 ( z) and Y1 ( z) are
2

N
each, and hence, require ⎛⎜ ⎞⎟ multiplications each.
⎝2⎠
Now, we can write Y1 ( z ) = (H 0 ( z ) + H1 ( z ))(X 0 ( z ) + X1 ( z)) − Y0 ( z) − Y1 ( z ). Since,

products of two polynomials of degree

N
2

(H0 ( z) + H1( z))(X 0 ( z) + X1( z)) is a product of two polynomials of degree

N
2

each, it can be

2

N
computed using ⎛⎜ ⎞⎟ multiplications. As a result, Y ( z ) = H ( z ) X ( z ) can be computed using
⎝2⎠

2

N
N
is even, and
3⎛⎜ ⎞⎟ multiplications instead of N 2 multiplications. If N is a power-of-2,
2
⎝2⎠
the same procedure can be applied to compute Y0 ( z), Y1 ( z), and Y2 ( z), reducing further the
number of multiplications. This process can be continued until the sequences being
convolved are of length 1 each.

Let R ( N ) denote the total number of multiplications required to compute the linear
convolution of two N –length sequences. Then, in the method outlined above, we have
R ( N ) = 3 ⋅ R ( N / 2) with R (1) = 1. A solution of this equation is given by R ( N ) = 3 log 2 N .

11.45 The dynamic range of a signed B –bit integer η is given by − (2 ( B −1) − 1) ≤ η < (2 ( B −1) − 1)

which for B = 32 is given by − (2 31 − 1) ≤ η < (2 31 − 1).
(a) For E = 6 and M = 25, the value of a 32-bit floating-point number is given by
η = ( −1) s 2 E − 31 ( M ). Hence, the value of the largest number is ≈ 2 32 , and the value of the

smallest number is ≈ −2 32. The dynamic range is therefore ≈ 2 × 2 32.
(b) For E = 7 and M = 24, the value of a 32-bit floating-point number is given by
η = ( −1) s 2 E − 63 ( M ). Hence, the value of the largest number is ≈ 2 64 , and the value of the

Not for sale

417

smallest number is ≈ −2 64. The dynamic range is therefore ≈ 2 × 2 64.
For E = 8 and M = 23, the value of a 32-bit floating-point number is given by
η = ( −1) s 2 E −127 ( M ). Hence, the value of the largest number is ≈ 2128 , and the value of the

(c)

smallest number is ≈ −2128. The dynamic range is therefore ≈ 2 × 2128.
Hence, the dynamic range in a floating-point representation is much larger than that in a fixedpoint representation with the same wordlength.
11.46 A 32-bit floating-point number in the IEEE Format has E = 8 and M = 23. Also, the
exponent E is coded in a biased form as E − 127 with certain conventions for special cases
such as E = 0, 255, and M = 0 (See page 637 of text).

Now, a positive 32-bit floating-point number represented in the “normalized” form have an
exponent in the range 0 < E < 255, and is of the form η = ( −1) s 2 E −127 (1∆ M ). Hence, the
smallest positive number that can be represented will have E = 1, and M = 0 0 K 0 0,
1442443
22 bits

and has therefore a value given by 2 −126 ≅ 1.18 × 10 −38. For the largest positive number, E
= 254, and M = 1 1 K 1 1. Thus, here η = (−1) 0 2127 (1∆ 1 1 K 1 1) ≈ 2127 × 2
1442443
1442443
22 bits

22 bits

3.4 × 10 38.

Note: For representing numbers less than 2 −126 , IEEE format uses the “de-normalized”
form where E = 0, and η = ( −1) s 2 −126 (0 ∆ M ). In this case, the smallest positive number
that can be represented is given by η = (−1) 0 2126 (0 ∆ 0 0 K 0 1) ≈ 2 −149
1442443
22 bits

≅ 1.4013 × 10 −45.
11.47 For a two’s-complement binary fraction s ∆ a −1a − 2 K a − b , the decimal equivalent for

s = 0 is simply

∑ i =1 a − i 2 − i. For s = 1, the decimal equivalent is given by
b

b
b
b
⎛ b

i

b

i
b
b
i

(1 − a − i )2 + 2

=− 2 +
a − i 2 − 2 = −(1 − 2 ) +
a− i 2 − i − 2 − b

i =1
i =1
i =1
⎝ i =1

= −1 +
−s+

b

∑ a− i 2 − i.

i =1
b

Hence, the decimal equivalent of s ∆ a −1a − 2 K a − b is given by

∑ a − i 2 − i.
i =1

11.48 For a ones’-complement binary fraction s ∆ a −1a − 2 K a − b , the decimal equivalent for

Not for sale

418

s = 0 is simply

∑ i =1 a − i 2 − i. For s = 1, the decimal equivalent is given by
b

b
b
b
⎛ b

i

i
b
i

(1 − a − i )2

=− 2 +
a − i 2 = −(1 − 2 ) +
a− i 2 − i

i =1
i =1
i =1
⎝ i =1

= −1 +

b

∑ a− i 2 − i.
i =1

− s(1 − 2 − b ) +

Hence, the decimal equivalent of s ∆ a −1a − 2 K a − b is given by

b

∑ a − i 2 − i.
i =1

11.49 η1 = 0.7812510 = 0 ∆11001, η2 = 0.6875010 = 0 ∆10110, η3 = 0.5312510 = 1∆ 01111.
η1 + η2 = 0 ∆11001 + 0 ∆10110 = 1∆ 01111.
Dropping thee overflow bit and adding (η1 + η2 ) to η3 :
(η1 + η2 ) + n3 = 0 ∆ 01111 + 1∆ 01111 = 0 ∆11110 = 0.937510 ,
where we have dropped the carry bit in the MSB location. Note that the final sum is correct
inspite of the overflow.
⎛ e jωˆ + e − jωˆ ⎞
⎟,
β ⎜⎜

2

1

1
⎛ zˆ + zˆ ⎞
z+z
⎟. Now, H ( z ) let
which by analytic continuation can be expressed as
= α + β ⎜⎜

2
⎝ 2 ⎠
be a Type 1 linear-phase FIR transfer function of degree 2M. As indicated in Eq. (8.128),

11.49 The transformation cos ω = α + β cos ωˆ is equivalent to

H ( z ) can be expressed as H ( z) = z − M

given by H (e jω ) = e − jMω

e jω + e − jω
=α+
2

⎛ z + z −1 ⎞
M

a[ n]⎜⎜
n=0
2 ⎟

(

n

with a frequency response

∑ n = 0 a[n](cos ω) n , with H (ω) = ∑ n = 0 a[n](cos ω) n
M

M

denoting

the amplitude function or the zero-phase frequency response. The amplitude function or the
zero-phase frequency response of the transformed filter obtained by applying the mapping
(
M
a[ n](α + β cos ωˆ ) n . Or, equivalently,
cos ω = α + β cos ωˆ is therefore given by H (ωˆ ) =

∑n=0

the transfer function of the transformed filter is given by

H ( zˆ ) = z − M

M
zˆ + zˆ −1 ⎞⎟
a[ n]⎜⎜ α + β
n=0
2 ⎟⎠

n

A convenient way to realize H ( zˆ ) is to consider the

realization of the parent transfer function H ( z ) in the form of a Taylor structure as outlined in
Problem 8.17 which is obtained by expressing H ( z ) in the form

H ( z) =

n
−2 ⎞

+
1
M
z
⎟ . Similarly, the transformed filter can be realized by
a[ n]z − M + n ⎜⎜
n=0
2 ⎟

Not for sale

419

1 + z −2
1 + zˆ −2
in the Taylor structure by the block αzˆ −1 + β
.
2
2
Now, for a lowpass-to-lowpass transformation we can impose the condition
(
(
H (ωˆ )
= H (ω)
. This condition is met if α + β = 1 and 0 ≤ α < 1. In this case, the

replacing each block

ωˆ = 0

ω=0

transformation reduces to cos ω = α + (1 − α ) cos ωˆ . From the plot of the mapping given below, it
follows that as α is varied between 0 and 1, ωˆ c < ωc .
1

cos ω

0.5
0
-0.5
-1
-1

-0.5

0
^
cos ω

0.5

1

On the other hand if ωˆ c > ωc . is desired along with a lowpass-to-lowpass transformation, we
(
(
= H (ω)
. This condition is met if β = 1 + α and
can impose the condition H (ωˆ )
ωˆ = π

ω=π

− 1 < α ≤ 1. The corresponding transformation is now given by cos ω = α + (1 + α ) cos ωˆ . From
the plot of the mapping given below, it follows that as α is varied between 0 and 1, ωˆ c > ωc .
1

cos ω

0.5
0
-0.5
-1
-1

11.51 (a) X ( z ) =

N −1

x[n]z

n=0

−n

-0.5

0
^
cos ω

0.5

1

n
(
N −1
⎛ − α + z( −1 ⎞
(
( −1
⎟ = P( z() ,
. X ( z ) = X ( z ) z −1 = − α + z = ∑ x[n]⎜
⎜ 1 − αz( −1 ⎟
D( z )
(
n=0

1− αz −1

N −1
( N −1
(
(
(
where P( z ) = ∑ p[n]z − n = ∑ x[n](1 − α z −1 ) N −1− n (−α + z −1 ) n , and
n=0

n=0

(
D( z ) =

N −1

(

(

∑ d [n]z − n = (1 − αz −1 ) N −1.

n=0

Not for sale

420

(
(
(
(
P( z )
P[k ]
(
(
= ( , where P[k ] = P( z ) z( = e j 2 πk / N is
(b) X [k ] = X ( z ) z( = e j 2 πk / N =
(
D( z ) z( = e j 2 πk / N
D[k ]
(
(
the N –point DFT of the sequence p[n] and D[k ] = D( z ) z( = e j 2 πk / N is the N –point DFT
of the sequence d [n].
(c) Let P = [ p[0] p[1] L p[ N − 1]]t and X = [x[0] x[1] L x[ N − 1]]t . Without any
loss of generality, assume N = 4 in which case

(

)

3
(
(
P( z ) = ∑ p[n]z − n = x[0] − αx[1] + α 2 x[2] − α 3 x[3]

(
(
(

n=0

)

(
+ − 3αx[0] + (1 + 2α 2 ) x[1] − α(2 + α 2 ) x[2] + 3α 2 x[3] z −1
(
+ 3α 2 x[0] − α(2 + α 2 ) x[1] − (1 + α 2 ) x[2] − 3αx[3] z −2
(
(
+ − α 3 x[0] + α 2 x[1] − αx[2] + αx[3] z −3 . Equating like powers of z −1 we can write

−α
α2
− α 3 ⎤ ⎡ x[0]⎤
⎡ p[0]⎤ ⎢ 1

2
2
⎢ p[1] ⎥ ⎢ − 3α
− α(2 + α ) 3α 2 ⎥ ⎢ x[1] ⎥
1 + 2α
P = Q ⋅ X or ⎢
. It can be shown
=
p[2]⎥ ⎢ 2 − α(2 + α 2 )
− 3α ⎥ ⎢⎢ x[2]⎥⎥
1 + 2α 2
⎢ p[3]⎥ ⎢ 3α
⎥ x[3]⎦
3

α2
−α
1 ⎦⎣
⎣− α
that the elements q r , s , 0 ≤ r , s ≤ 3, of the 4 × 4 matrix Q can be determined as follows:
(i)

)

)

The first row is given by q0, s = (−α) s ,

3!
(−α ) r , and
r!(3 − r )!
(iii) The remaining elements can be obtained using the recurrence relation
q r , s = q r −1, s −1 − α q r , s −1 + α q r −1, s .

(ii)

The first column is given by q r ,0 = 3C r (−α ) r =

In the general case, we only change the computation of the elements of the first column
( N − 1)!
( −α ) r .
using the relation q r ,0 = N −1C r (−α) r =
r!( N − 1 − r )!

A 0 ( z) =

0.3038 + z −1

[

]

1
A ( z) − A1 ( z) ,
2 0
0.639 + 0.4012 z −1 + z −2

11.52 The highpass transfer function can be expressed as H ( z ) =

where

and A1 ( z) =
. The tunable highpass
1 + 0.3038z −1
1 + 0.4012 z −1 + 0.639z − 2
transfer function is obtained by applying the lowpass-to-lowpass transformation of Eq.
sin[(0.6π − ωˆ p ) / 2]
(11.115) where the tuning parameter is given by α =
. The tunable
sin[(0.6π + ωˆ p ) / 2]
highpass transfer function is then given by H ( z, α ) =

Not for sale

1
2

[Aˆ 0 (z) − Aˆ1 (z)], where from Eqn.

421

[ 0.3038 − 0.9077α ] + z −1
and from Eqn. (11.123)
(11.119) Aˆ 0 ( z ) =
1 + [ 0.3038 − 0.9077α ]z −1
[ 0.639 − 0.1448α ] + [0.4012 + 3.439α ]z −1 + z −2
.
Aˆ 1( z ) =
1 + +[ 0.4012 + 3.439α ]z −1 + [ 0.639 − 0.1448α ]z − 2
11.53

[

]

1
A ( z) + A1 ( z) ,
2 0
0.3554 − 0.4191 z −1 + z −2

The lowpass transfer function can be expressed as H ( z) =
− 0.1584 + z −1

where

. The tunable bandpass
and A1 ( z ) =
1 − 0.4191 z −1 + 0.3554 z − 2
1 − 0.1584 z −1
transfer function is obtained by applying the lowpass-to-bandpass transformation of Eq.
(11.124) where the tuning parameter is given by β = cos ωo with ωo denoting the center
frequency of the bandpass filter. The tunable bandpass transfer function is then given by
⎛ β + z −1 ⎞

− 0.1584 − z −1 ⎜⎜
−1 ⎟
1
+
β
z
1 ˆ

⎠ and
H ( z, α ) = A 0 ( z) + Aˆ 1 ( z) , where Aˆ 0 ( z ) =
2
⎛ β + z −1 ⎞
⎟⎟
1 + 0.1584 z −1 ⎜⎜
⎝ 1+ βz −1 ⎠

A 0 (z) =

[

Aˆ 1 ( z ) =

]

⎛ β + z −1 ⎞
⎛ β + z −1 ⎞
⎟⎟ + z − 2 ⎜⎜

0.3554 + 0.4191 z −1 ⎜⎜
−1
−1 ⎟
+
z
+
z
1
1
β
β

2

⎟⎟ + 0.3554 z − 2 ⎜⎜
⎟⎟
1 + 0.4191 z −1 ⎜⎜

1

1
z
z
+
β
+
β
1
1

β + z −1

β + z −1

M11.1 (a) Numerator coefficients =
[0.0528
0.0797
0.1295
Denominator coefficients =
[1.0000
-1.8107
2.4947

2

.

0.1295

0.0797

0.0528]

-1.8801

0.9537

-0.2336]

Impulse response samples
0.6

Amplitude

0.4
0.2
0
-0.2

0

5

10
15
20
Time index n

(b) Numerator coefficients =
[0.0084
-0.0335
0.0502
Denominator coefficients =

-0.0335

Not for sale

25

30

0.0084]

422

[1.0000

2.3741

2.7057

1.5917

0.4103]

Impulse response samples
0.4

Amplitude

0.2
0
-0.2
-0.4

0

5

10
15
20
Time index n

25

30

(c) Numerator coefficients = [0.0003 0 -0.0019 0 0.0057 0
-0.0095 0 0.0095 0 -0.0057 0 0.0019 0 -0.0003]
Denominator coefficients = [1.0000 1.7451
4.9282
6.1195
9.8134
9.2245
10.4323
7.5154
6.4091
3.4595
2.2601
0.8470 0.4167
0.0856
0.0299]
Impulse response samples
0.4

Amplitude

0.2
0
-0.2
-0.4

0

5

10
15
20
Time index n

25

30

M11.2 The modified MATLAB program is given below:
n = 0:60;
w = input('Normalized angular frequency vector = ');
num = input('Numerator coefficients = ');
den = input('Denominator coefficients = ');
x1 = cos(w(1)*pi*n); x2 = cos(w(2)*pi*n);
x = x1+x2;
% Generate the output sequence by filtering the input
y = filter(num,den,x);
% Plot the input and the output sequences
figure(1)
stem(n,x);
xlabel('Time index n'); ylabel('Amplitude');
title('Input sequence');
figure(2)
stem(n,y);
xlabel('Time index n'); ylabel('Amplitude');

Not for sale

423

title('Output sequence');

The plots generated by the above program for the filter of Example 9.14 for an input
composed of a sum of two sinusoidal sequences of angular frequencies, 0.3π and 0.6π, are
given below:
Output sequence
1

1

0.5
Amplitude

Amplitude

Input sequence
2

0
-1
-2

0
-0.5

0

10

20
30
40
Time index n

50

-1

60

0

10

20
30
40
Time index n

50

60

The blocking of the high-frequency signal by the lowpass filter can be demonstrated by
replacing the statement stem(n,x); with stem(n,x2); and the statement
y=filter(num,den,x); with y=filter(num,den,x2);. The plots of the highfrequency input signal and the corresponding output are shown below:
Output sequence
0.3

0.5

0.2
Amplitude

Amplitude

Input sequence
1

0
-0.5
-1

0.1
0
-0.1

0

10

20
30
40
Time index n

50

-0.2

60

0

10

20
30
40
Time index n

50

60

M11.3 The plots generated by the program of Exercise M11.2 for the filter of Example 9.15
for an input composed of a sum of two sinusoidal sequences of angular frequencies, 0.3π and
0.6π, are given below:
Output sequence
0.4

1

0.2
Amplitude

Amplitude

Input sequence
2

0
-1
-2

0
-0.2
-0.4

0

10

20
30
40
Time index n

50

60

-0.6

Not for sale

0

10

20
30
40
Time index n

50

60

424

The blocking of the low-frequency signal by the highpass filter can be demonstrated by
replacing the statement stem(n,x); with stem(n,x1); and the statement
y=filter(num,den,x); with y=filter(num,den,x1);. The plots of the lowfrequency input signal and the corresponding output are shown below:
Output sequence
0.2

0.5

0.1
Amplitude

Amplitude

Input sequence
1

0
-0.5
-1

0
-0.1

0

10

20
30
40
Time index n

50

60

-0.2

0

10

20
30
40
Time index n

50

60

M11.4 % The factors for the transfer function of
% the lowpass filter of Example 9.14 are
% num1 = [0.0528 0.0528 0];
% den1 = [1.0000 -0.4909 0];
% num2 = [1.0000 0.6040 1.0000];
% den2 = [1.0000 -0.7624 0.5390];
% num3 = [1.0000 -0.0949 1.0000];
% den3 = [1.0000 -0.5574 0.8828];
N = input(‘Total number of sections = ‘);
for k = 1:N;
num(k,:) = input('Numerator factor = ');
den(k,:) = input('Denominator factor = ');
end
n = 0:60;
w = input('Normalized angular frequency vector = ');
x1 = cos*w(1)*pi*n); x2 = cos*w(2)*pi*n);
x = x1 + x2;
% Plot the input sequence
figure(1)
stem(n,x);
xlabel('Time index n'); ylabel('Amplitude');
title('Input sequence');
si = [0 0];
for k = 1:N;
y(k,:) = filter(num(k,:),den(k,:),x,si);
x = y(k,:);
end
% Plot the output sequence
figure(2)
stem(n,y(N,:));
xlabel('Time index n'); ylabel('Amplitude');
title('Output sequence');

Not for sale

425

The plots generated by the above program for the filter of Example 9.14 for an input
composed of a sum of two sinusoidal sequences of angular frequencies, 0.3π and 0.6π, are
given below:
Output sequence

Input sequence

1

2

0.5
Amplitude

Amplitude

1
0

-0.5

-1
-2

M11.5 %
%
%
%
%
%

0

-1
0

10

20
30
40
Time index n

50

60

0

10

20
30
40
Time index n

50

60

The factors for the transfer function of
the highpass filter of Example 9.15 are
num1 = [0.0084 -0.0167 0.0084];
den1 = [1.0000
1.3101 0.5151];
num2 = [1.0000 -2.0000 1.0000];
den2 = [1.0000
1.0640 0.7966];

The plots generated by the program of Exercise M11.4 for the filter of Example 9.15 for an
input composed of a sum of two sinusoidal sequences of angular frequencies, 0.3π and 0.6π,
are given below:
Input sequence

Output sequence

2

0.4
0.2
Amplitude

Amplitude

1
0
-1
-2

0
-0.2
-0.4

0

10

20
30
40
Time index n

50

60

-0.6

0

10

20
30
40
Time index n

50

60

M11.6 To apply the function direct2 to filter a sum of two sinusoidal sequences, we replace
the statement y = filter(num,den,x,si); in the MATLAB program given in the
solution of Exercise M11.2 with the statement y = direct2(num,den,x,si);. The
plots generated by running the modified program for the data given in this problem are given
below:

Not for sale

426

Output sequence
1

1

0.5
Amplitude

Amplitude

Input sequence
2

0

-0.5

-1
-2

0

0

10

20
30
40
Time index n

50

60

-1

0

10

20
30
40
Time index n

50

60

M11.7 The MATLAB program that can be used to compute all DFT samples using the function
goertzel and the function fft is as follows:
clear
N = input('Desired DFT length = ');
x = input('Input sequence = ');
for j = 1:N
Y(j)=goertzel(x,j);
end
disp('DFT samples computed using goertzel are ') ;
disp(Y)
disp('DFT samples computed using fft are ') ;
X = fft(x); disp(X);
Results obtained for two input sequences of lengths 8 and 12, respectively, are given below:
Desired DFT length = 8
Input sequence = [1 2 3 4 4 3 2 1]
DFT samples computed using goertzel are
Columns 1 through 4
20.0000
-5.8284-2.4142i -0.0000-0.0000i -0.1716-0.4142i
Columns 5 through 8
0-0.0000i -0.1716+0.4142i
0.0000-0.0000i -5.8284+2.4142i
DFT samples computed using fft are
Columns 1 through 4
20.0000
-5.8284-2.4142i 0
Columns 5 through 8
0
-0.1716+0.4142i 0
Desired DFT length = 12
Input sequence = [1 2 4 8 10 12 7

-0.1716-0.4142i
-5.828 +2.4142i
3 -4 5 0

DFT samples computed using goertzel are
Columns 1 through 4
54.0000 -13.0622-21.0885i
1.5000+19.9186i
Columns 5 through 8

Not for sale

6]

-4.0000-2.0000i

427

4.5000+2.5981i
-0.9378-10.0885i

-0.9378+10.0885i

Columns 9 through 12
4.5000-2.5981i -4.0000+2.0000i
13.0622 +21.0885i

-18.0000+0.0000i

1.5000-19.9186i

DFT samples computed using fft are
54.0000 -13.0622-21.0885i
1.5000+19.9186i
Columns 5 through 8
4.5000+2.5981i
-0.9378-10.0885i

-0.9378+10.0885i

Columns 9 through 12
4.5000-2.5981i -4.0000+2.0000i
13.0622 +21.0885i

-

-4.0000-2.0000i

-18.0000+0.0000i

1.5000-19.9186i

-

M11.8 The MATLAB program that can be used to verify the plots of Figure 11.43 is given below:
%Program_11_8.m
[z,p,k] = ellip(5,0.5,40,0.4);
a = conv([1 -p(1)],[1 -p(2)]);b =[1 -p(5)];
c = conv([1 -p(3)],[1 -p(4)]);
w = 0:pi/255:pi;
alpha = 0;
an1 = a(2) + (a(2)*a(2) - 2*(1 + a(3)))*alpha;
an2 = a(3) + (a(3) -1)*a(2)*alpha;
g = b(2) - (1 - b(2)*b(2))*alpha;
cn1 = c(2) + (c(2)*c(2) - 2*(1 + c(3)))*alpha;
cn2 = c(3) + (c(3) -1)*c(2)*alpha;
a = [1 an1 an2];b = [1 g]; c = [1 cn1 cn2];
h1 = freqz(fliplr(a),a,w); h2 = freqz(fliplr(b),b,w);
h3 = freqz(fliplr(c),c,w);
ha = 0.5*(h1.*h2 + h3);ma = 20*log10(abs(ha));
alpha = 0.1;
an1 = a(2) + (a(2)*a(2) - 2*(1 + a(3)))*alpha;
an2 = a(3) + (a(3) -1)*a(2)*alpha;
g = b(2) - (1 - b(2)*b(2))*alpha;
cn1 = c(2) + (c(2)*c(2) - 2*(1 + c(3)))*alpha;
cn2 = c(3) + (c(3) -1)*c(2)*alpha;
a = [1 an1 an2];b = [1 g]; c = [1 cn1 cn2];