Anual Modern Digital Signal Processing by Roberto Christ I

http://ebook29.blogspot.
com
Chapter 1: Problem Solutions

Review of Signals and Systems
Signals
à Problem 1.1
a) x@nD = -0.5 ∆@n + 1D + ∆@nD + 0.5 ∆@n - 1D + ∆@n - 2D - 0.8 ∆@n - 3D
b) x@nD = -0.5 ∆@n + 5D + ∆@n + 4D + 0.5 ∆@n + 3D + ∆@n + 2D - 0.8 ∆@n + 1D
à Problem 1.2
a) I = e-1
b) I = e-1
c) I = 0 since the interval of integration does not include the point t = -1, where the impulse is
centered.
d) I = 1
e) I = cos2 H0.1 ΠL
f) I = e
g) let Λ = - 12 t, then t = -2 Λ and dt = -2 dΛ. Substitute in the integral to obtain

-¥
I = Ù H-2 ΛL2 ∆ IΛ + 12 M H-2 dΛL =
+¥
+¥
= Ù 8 Λ2 ∆ IΛ + 12 M dΛ = 2
-¥
h) let Λ = 3 t, then t = Λ 3 and dt = I 13 M dΛ. Then the integral becomes

+¥ 13
I = Ù eΛ3 ∆ HΛ - 1L I 13 M dΛ = e
3
-¥
http://ebook29.blogspot.com
2
http://ebook29.blogspot.com CurrentValue[FileName]
à Problem 1.3
Amplitude A = 2,
1
Period T0 = 0.01 sec , then frequency F0 = T = 100 Hz = 0.1 kHz
0
Phase Α = 0
Then the signal can be written as x HtL = 2 cos H200 ΠtL.
à Problem 1.4
a) Frequency F0 = 1 T0 = 1 I3 10-3 M = 13 103 Hz. Then
x HtL = 2.5 cos I 23 1000 Πt + 150 M

b) Digital Frequency Ω0 = 2 ΠF0 Fs = 2 Π 6 = Π 3 rad. Therefore the sampled sinusoid
becomes
x@nD = 2.5 cos I Π3 n + 150 M
à Problem 1.5
All sinusoids are distinct. In continuous time there is no ambiguity between

frequency and signal.
à Problem 1.6
First bring all frequencies within the interval -Π to Π. This yields
x2 @nD = 2 cos H1.5 Πn - 0.1 Π - 2 ΠnL = 2 cos H-0.5 Πn - 0.1 ΠL
= 2 cos H0.5 Πn + 0.1 ΠL
and also
x3 @nD = 2 cos H1.5 Πn + 0.1 Π - 2 ΠnL = 2 cos H-0.5 Πn + 0.1 ΠL
= 2 cos H0.5 Πn - 0.1 ΠL
Therefore we can see that x1 @nD = x2 @nD and x3 @nD = x4 @nD.
A different way of solving this problem is graphically. The frequency
plots for all four signals are shown next. The key point is to understand
that in discrete time, all frequencies within the interval -Π, Π yield complex Exponentials with dis-
tinct values:
CurrentValue[FileName]
http://ebook29.blogspot.com 3
x1 @nD Frequency Representation
4
Again you see that x1 @nD and x2 @nD have the same representation, and the same for x3 @nD and
x4 @nD.
à Problem 1.7
The sinusoid x@nD = 3 cos H1.9 Πn + 0.2 ΠL has the same samples, since
3 cos H0.1 Πn - 0.2 ΠL = 3 cos H-0.1 Πn + 2 Πn + 0.2 ΠL
à Problem 1.8
Problem 1.8
a) since 150 = 15 * Π 180 = Π 12, we can write

x HtL = 32 IejΠ12 ej100Πt + e-jΠ12 e-j100Πt M
b) x HtL = e-j0 .1 Π ej10Πt + ej0 .1 Π e-j10Πt

c) x HtL = -2.5 je-j0 .2 Π ej20Πt + 2.5 jej0 .2 Π e-j20Πt
d) x HtL = 5 jej1000Πt - 5 je-j1000Πt
e) x HtL = 1.5 je-j0 .2 Πt ej200Πt - 1.5 jej0 .2 Πt e-j200Πt
à Problem 1.9
a) since 2 + j = 2.2361 ej0 .4636 , we can write x HtL = 4.4722 cos H100 Πt + 0.4636L
b) since 1 + 2 ej0 .1 Π = 2.9672 ej0 .2098 , then x HtL = 5.9344 cos H10 Πt + 0.2098L
, Π , Π
c) the signal can be written as xHtL = 12 sinH5 Πt + 0.1 ΠL J 2 e j 4 e j10Πt + 2 e- j 4 e-j10Πt N and therefore
1
x HtL = ,
sin H5 Πt + 0.1 ΠL cos I10 Πt + Π4 M
2
à Problem 1.10

a) x HtL = j20Π x HtL
1
b) Ù x HtL dt = j20Π
x HtL
c) x Ht - 0.1L = e-j2Π x HtL = x HtL
à Problem 1.11
In order to solve this problem we have to decompose the signals into complex
exponentials.
a) x HtL = 2 cos H100 Πt + 0.2 ΠL = ej0 .2 Π ej100Πt + e-j0 .2 Π e-j100Πt
6
b) x HtL = 32 e-j0 .1 Π ej100Πt + 32 ej0 .1 Π e-j100Πt

3 3
- 2
j
ej0 .2 Π ej110Πt + 2
j
e-j0 .2 Π e-j110Πt
This becomes
x HtL = 32 e-j0 .1 Π ej100Πt + 32 ej0 .1 Π e-j100Πt

32 ej0 .7 Π ej110Πt + 32 e-j0 .7 Π e-j110Πt
c) x HtL = e-j0 .2 Π ej100Πt + ej0 .2 Π e-j100Πt

3 3
- 2
j
ej0 .1 Π ej100Πt + 2
j
e-j0 .1 Π e-j100Πt
c) x HtL = e-j0 .2 Π ej100Πt + ej0 .2 Π e-j100Πt

3 3
- 2
j
ej0 .1 Π ej100Πt + 2
j
e-j0 .1 Π e-j100Πt
Combining terms we obtain

x HtL = 0.9071 ej1 .1800 ej100t + 0.9071 e-j1 .1800 e-j100t
, Π ,
d) x HtL = 2 5 ej 4
ej1000Πt + 5 ej2 .67795 e-j1100Πt
e) x HtL = ej0t + 12 ej10Πt + 12 e-j10Πt - 2

1
j
1
ej20Πt + 2
j
e-j20Πt
This becomes
Π Π
x HtL = ej0t + 12 ej10Πt + 12 e-j10Πt 12 ej 2
ej20Πt + 12 e-j 2
e-j20Πt
8
Π Π
x HtL = ej0t + 12 ej10Πt + 12 e-j10Πt 12 ej 2
ej20Πt + 12 e-j 2
e-j20Πt
à Problem 1.12
Referring to the figure above:

a) Ω0 = Π 2
b) Ω0 = 2 Π 3
c) Ω0 = Π
d) Ω0 = 0.75 Π, since 1000 Π 800 = 1.25 Π > Π. We want the frequency to be between -Π and
Π , we use the fact that this sinusoid has the same samples as Ω0 = 2 Π - 1.25 Π = 0.75 Π.
d) Ω0 = 0.75 Π, since 1000 Π 800 = 1.25 Π > Π. We want the frequency to be between -Π and
Π , we use the fact that this sinusoid has the same samples as Ω0 = 2 Π - 1.25 Π = 0.75 Π.
e) Ω0 = 0, since 1000 Π 500 = 2 Π and therefore Ω0 = 2 Π - 2 Π = 0.
à Problem 1.13
a) Since Ω0 = 0.2 Π = 2 ΠF0 2000 we solve for the frequency as F0 = 200 Hz.
b) Sinusoids with frequencies F0 + Fs = 2.2 kHz, or Fs - F0 = 1.8 kHz have the same sample
values.
à Problem 1.14
a) Since
Π Π
x HtL = ej100Πt + e-j100Πt + 32 ej 2
ej120Πt + 32 e-j 2
e-j120Πt + 2 ej150Πt + 2 e-j150Πt
we obtain the frequency plot shown below.
b) The maximum frequency is FMAX = 75 Hz. Therefore the sampling frequency has to be such that
Fs > 150 Hz.
à Problem 1.15
The bandwidth is 4 kHz and therefore the sampling frequency has to be at least Fs = 8 kHz. This
means that we need 8000 samples for every second of data, and therefore
16 8000 = 128 kbytes sec. For one minute of data we need at least
60 120 = 7200 kbytes = 7.2 Mbytes to store the file.
10
Systems
à Problem 1.16
If a property is not specified it is assumed Linear, Time Invariant, Causal,

BIBO Stable.
b) Non Linear. In fact if you apply superposition:

x1 @nD ® S ® y1 @nD = 2 x1 @nD - 1
x2 @nD ® S ® y2 @nD = 2 x2 @nD - 1
and therefore
x@nD = x1 @nD + x2 @nD ® S ® y@nD = 2 Hx1 @nD + x2 @nDL - 1
which shows that the output is different from y1 @nD + y2 @nD = 2 Hx1 @nD + x2 @nDL - 2.
c) Time Varying, due to the time varying coefficient "t". Non BIBO Stable, since x HtL = u HtL , a
Bounded Input , yields y HtL = tu HtL, a Non Bounded Output;
d) Time Varying,due to the coefficient e-t ;
e) Non Linear. If x HtL = x1 HtL + x2 HtL then the output is y HtL = È x1 HtL + x2 HtL È
different from y1 HtL + y2 HtL = È x1 HtL È + È x2 HtL È;
f) Non Causal, since the output at any time t depends on a future input at time t + 1;
h) Non Causal since the output at time n depends on a future input at time n + 1;
i) Non Linear (due to the term x2 ). Time Varying (due to the time varying coefficient n). Not BIBO
Stable, since y@nD = nu@nD + ... , ie Non Bounded, when the input x@nD = u@nD, a Bounded
input.
à Problem 1.17
+¥
a) y@nD = S u@kD 0.5n-k u@n - kD.
k=-¥
If n ³ 0:
n n n+1
y@nD = S 0.5n-k = S 0.5k = 1-0.5 = 2 - 0.5n

1-0.5
k=0 k=0
If n < 0 then u@kD u@n - kD = 0 for all k, and therefore y@nD = 0.

Therefore y@nD = I2 - 0.5n M u@nD.
+¥
b) y@nD = S u@k - 2D 0.5n-k u@n - kD.
k=-¥
If n ³ 2:
n n
y@nD = S 0.5n-k = S 0.5k - 0.5n - 0.5n-1 =
k=2 k=0
n+1
= 1-0.5 - 0.5n - 0.5n-1 = 2 I1 - 0.5n-2 M

1-0.5
If n < 2 then u@k - 2D u@n - kD = 2 for all k, and therefore y@nD = 0.

Therefore y@nD = I2 - 0.5n-2 M u@n - 2D.
+¥
c) y@nD = S 0.8k u@kD 0.5n-k u@n - kD.
k=-¥
If n ³ 0:
n n n+1
y@nD = S 0.8k 0.5n-k = 0.5n S 1.6k = 0.5n 1-1.6 = 1.25 0.5n - 2 0.8n

1-1.6
k=0 k=0

Therefore y@nD = I1.25 0.5n - 2 0.8n M u@nD.
+¥
d) y@nD = S 1.2k u@-kD 0.5n-k u@n - kD
k=-¥
If n < 0 then u@-kD u@n - kD = u@n - kD. This implies

n n
y@nD = S 1.2k 0.5n-k = 0.5n S 2.4k =
k=-¥ k=-¥
n +¥
0.5 S 2.4-k =
k=-n
+¥ -n-1
0.5n K S 2.4-k - S 2.4-k O =
k=0 k=0
n
0.5n J
1 1-2.4
-1 -
-1 N =
1-2.4 1-2.4
n
= 1.7143 1.2
If n ³ 0 then u@-kD u@n - kD = u@-kD. This implies
0 0
y@nD = S 1.2k 0.5n-k = 0.5n S 2.4k =
k=-¥ k=-¥
n +¥
= 0.5 S 2.4-k = 1.7143 0.5n
k=0
Finally we can write the answer y@nD = 1.7143 1.2n u@-n - 1D + 1.7143 0.5n u@nD.
+¥ +¥
e) y@nD = S ej0 .2 Πn-k 0.5k u@kD = K S 0.5k e-j0 .2 Πk O ej0 .2 Πn
k=-¥ k=0
Applying the geometric series we obtain
12
+¥ +¥
e) y@nD = S ej0 .2 Πn-k 0.5k u@kD = K S 0.5k e-j0 .2 Πk O ej0 .2 Πn
k=-¥ k=0
Applying the geometric series we obtain

1
y@nD = 1-0.5

e-j0 .2 Π
ej0 .2 Πn = 1.5059 ej H0.2 Πn-0.4585L
+¥ n
f) y@nD = S ej0 .2 Πn-k u@n - kD 0.5k u@kD = K S 0.5k e-j0 .2 Πk O ej0 .2 Πn if n ³ 0.
k=-¥ k=0
Then, applying the geometric series we obtain
n+1
i 1-I0.5 e-j0 .2 Π M y
y@nD = j
j
-j0 .2 Π
z
z ej0 .2 Πn =
k 1-0.5 e {
= 1.5059 Iej H0.2 Πn-0.4585L - e-j0 .2 Π 0.5n+1 M
and y@nD = 0 when n < 0. Therefore
y@nD = 1.5059 Iej H0.2 Πn-0.4585L - e-j0 .2 Π 0.5n+1 M u@nD
g) Since x@nD = ej0 .2 Πn + e-j0 .2 Πn , using the solution to problem e) above we obtain
y@nD = 1.5059 Iej H0.2 Πn-0.4585 + e-j H0.2 Πn-0.4585 M =
= 3.0118 cos H0.2 Πn - 0.4585L
h) Since x@nD = ej0 .2 Πn u@nD + e-j0 .2 Πn u@nD, using the solution to problem f) above we obtain
y@nD = 3.0118 Icos H0.2 Πn - 0.4585L - cos H0.2 ΠL 0.5n+1 M u@nD
= 3.0118 Icos H0.2 Πn - 0.4585L - 1.2183 0.5n M u@nD
à Problem 1.18
a) h@nD = ∆@nD + ∆@n - 1D + ∆@n - 2D
b) y@nD = u@nD + u@n - 1D + u@n - 2D
c) y@nD = ej0 .5 Πn + ej0 .5 Πn Hn-1L + ej0 .5 Πn Hn-2L . Therefore, since e-j0 .5 Π = -j, we can write
y@nD = H1 - j - 1L ej0 .5 Πn = eHj0 .5 Πn-0.5 ΠL
d) y@nD = cos H0.5 Πn - 0.5 ΠL
à Problem 1.19
Problem 1.19
To determine the impulse response ∆@nD substitute x@nD = ∆@nD:
a) h@nD = 3 ∆@nD + 2 ∆@n - 1D - ∆@n - 2D
b) h@nD has to satisfy the recursion h@nD = h@n - 1D + ∆@nD. Therefore:
if n < 0 then h@nD = 0;
if n = 0 then h@0D = h@-1D + 1 = 1, since h@-1D = 0;
if n > 0 then h@nD = h@n - 1D that is to say h@nD is a constant, which must be equal to h@0D = 1.
Putting everything together you can verify that h@nD = u@nD;
c) h@nD = 2 ∆@n - 2D
For the next three problems recall that any signal x@nD can be written as
+¥
x@nD = S x@kD ∆@n - kD.
k=-¥
+¥
d) y@nD = S 0.5k ∆@n - kD = 0.5n u@nD
k=0
+¥
e) y@nD = S 0.5ÈkÈ ∆@n - kD = 0.5ÈnÈ
k=-¥
+¥ +¥
f) h@nD = S 0.5k-2 u@k - 5D ∆@n - k - 2D = S 0.5k u@k - 3D ∆@n - kD. Therefore the
k=-¥ k=-¥
impulse response is h@nD = 0.5n u@n - 3D.
à Problem 1.20
+¥ +¥
We have to check whether or not S Ë h@nD Ë < +¥. Recall that the geometric series S an < ¥ if
-¥ n=0
and only if È a È < 1.
+¥
a) S 0.5n < +¥ , then BIBO Stable;
n=0
+¥ -1 +¥ +¥
b) S 0.5n + S 0.5-n = S 0.5n + S 0.5-n < +¥ , then BIBO Stable;
n=0 n=-¥ n=0 n=1
+¥ -1 +¥ +¥
c) S 0.5n + S 0.5n = S 0.5n + S 0.5-n = +¥ then it is NOT BIBO Stable
n=0 n=-¥ n=0 n=1
0 +¥
d) S 0.5n = S 2n = +¥ then NOT BIBO Stable
n=-¥ n=0
+¥
e) S 1n = +¥, then NOT BIBO Stable
n=1
+¥ 1
f) S n
2 < +¥, then BIBO Stable
n=1
à
14
+¥1
f) S n
2 < +¥, then BIBO Stable
n=1
+¥
g) S 1 = +¥, then NOT BIBO Stable
n=0
+¥ +¥
h) S Ë ej0 .2 Πn Ë = S 1 = +¥, then NOT BIBO Stable
n=-¥ n=-¥
+¥ +¥
i) S Ì ej0 .2 Πn Ì = S 1 = +¥then NOT BIBO Stable
n=0 n=0
+¥ +¥
j) S Ì 0.8n ej0 .2 Πn Ì = S 0.8n < +¥ then BIBO Stable
n=0 n=0
+¥ +¥ -1 +¥ +¥
k) S Ë 0.8ÈnÈ cos H0.2 ΠnL Ì £ S 0.8n + S 0.8-n = S 0.8n + S 0.8n < +¥ then
n=-¥ n=0 n=-¥ n=0 n=1
BIBO Stable
0 +¥
l) S 2n = S 0.5n < +¥ then BIBO Stable
n=-¥ n=0
+¥
m) S 2n = +¥ then NOT BIBO Stable
n=0
z- Transforms
à Problem 1.21
+¥
a) X HzL = S 0.5n z-n = 1-0.5 1

z-1
z
= z-0.5
, ROC É 0.5 z-1 É < 1, ie È z È > 0.5;
n=0
-1 +¥
b) X HzL = S 0.5n z-n = S 0.5-n zn = 1-21

z
z
- 1 = - z-0.5
=, ROC É 0.5-1 z É < 1, ie
n=-¥ n=1
È z È < 0.5;
0 +¥
c) S 0.5n z-n = S 0.5-n zn = 1-21

z
, ROC È z È < 0.5;
n=-¥ n=0
+¥ n 1 z
d) X HzL = S ej0 .4 Π z-n = 1-e

j0 .4 Π z-1 = z-e j0 .4 Π , ROC É e
j0 .4 Π z-1 É < 1, ie È z È > 1;
n=0
-1 n +¥ n +¥ n +¥ n
e) X HzL = S ej0 .4 Π z-n + S ej0 .4 Π z-n = S e-j0 .4 Π zn + S ej0 .4 Π z-n . The first
n=-¥ n=0 n=1 n=0
series converges when È z È < 1 and the second when È z È > 1. Therefore there is no common
ROC and X HzL for this signal DOES NOT exist!
-1 +¥
+¥ +¥ 15
n n n n
e) X HzL = S ej0 .4 Π z-n + S ej0 .4 Π z-n = S e-j0 .4 Π zn + S ej0 .4 Π z-n . The first
n=-¥ n=0 n=1 n=0
series converges when È z È < 1 and the second when È z È > 1. Therefore there is no common
ROC and X HzL for this signal DOES NOT exist!
-1 n +¥ 1 n z
f) X HzL = S ej0 .4 Π z-n = S e-j0 .4 Π zn = 1-e -j0 .4 Π z - 1 = - z-e
j0 .4 Π , ROC

n=-¥ n=1
É e-j0 .4 Π z É < 1, ie È z È < 1;
-1 +¥ +¥ +¥
g) X HzL = S 0.5-n z-n + S 0.5n z-n = S 0.5n zn + S 0.5n z-n . See the two series:
n=-¥ n=0 n=1 n=0
+¥
S 0.5n zn = 1-0.51

z
z
- 1 = - z-2
, when È z È < 2;
n=1
+¥
S 0.5n z-n = 1-0.5 1

z-1
z
= z-0.5
, when ÈzÈ>0.5 .
n=0
Therefore the ROC has to be the intersection of the ROC’s, which yields
z z
X HzL = z-0.5
- z-2
, ROC: 0.5 < È z È < 2
h) X HzL = 1 + z-1 + z-2 , ROC: 0 < È z È;

i) X HzL = z + z2 + z3 , ROC È z È < +¥
à Problem 1.22
Recall
z
Z 8an u@nD< = z-a
ROC: È z È > È a È
z
Z 8an u@-n - 1D< = - z-a
ROC: È z È < È a È
a) x@nD = 2 0.5n-1 u@n - 1D, then X HzL = 2 z-1 z-0.5

z
2
= z-0.5

b) x@nD = 0.5n u@-n - 1D + ∆@nD, then X HzL = z-0.5

-z
-0.5
+ 1 = z-0.5

n z
d) x@nD = I0.5 ej0 .1 Π M u@nD, then X HzL = z-0.5

ej0 .1 Π

e) x@nD = H-0.5Ln u@nD, then X HzL = z+0.5

z

n n
f) x@nD = 12 I0.5 ej0 .1 Π M u@nD + 12 I0.5 ej0 .1 Π M u@nD, then
X HzL = 12 z-0.5 z
+ 12 z-0.5

ej0 .1 Π
z

e-j0 .1 Π
. Combining terms w obtain
2
z -cos H0.1 ΠL z
X HzL = z
2 -cos H0.1 ΠL+0.25
g) x@nD = 0.5n u@nD + 0.5-n u@-n - 1D, then X HzL = z-0.5

z
z
- z-2
-1.5 z
2 -2.5 z+1 , ROC:
= z
0.5 < È z È < 2
-1.5 z -1.5
h) Using the result in g) above, X HzL = z-1 z
2 -2.5 z+1 = z
2 -2.5 z+1
16
-1.5 z -1.5
h) Using the result in g) above, X HzL = z-1 z
2 -2.5 z+1 = z
2 -2.5 z+1
à Problem 1.23
j2 .6180 -j2 .6180
a) X
HzL
z
= 2z + 0.5774 e
+ 0.5774

z-ej2 .0944
e

z-e-j2 .0944
therefore
z z
X HzL = 1 + 0.5774 ej2 .6100 + 0.5774 e-j2 .6100
z-e j2 .0944 z - e .0944-j2
Since È z È > 1 all terms are causal. Therefore

x@nD = ∆@nD + 0.5774 ej H2.0944 n+2.6100L u@nD + 0.5774 e-j H2.0944 n+2.6100L u@nD
Combining terms we obtain
x@nD = ∆@nD + 1.0548 cos H2.0944 n + 2.6100L u@nD
b) Same X HzL as in the previous problem, but different ROC. All terms are noncausal,
and therefore
x@nD =
∆@nD - 0.5774 ej H2.0944 n+2.6100L u@-n - 1D - 0.5774 e-j H2.0944 n+2.6100L u@-n - 1D
and, after combining terms
x@nD = ∆@nD 11.0548 cos H2.0944 n + 2.6100L u@-n - 1D
c) X
HzL
z
= 0.5

z
2
- z-1
+ 1.5
z-2
z
, therefore X HzL = 0.5 - 2 z-1
z
+ 1.5 z-2
. Form the ROC all terms
are causal. Therefore
x@nD = 0.5 ∆@nD - 2 u@nD + 1.5 2n u@nD
d) Same X HzL with different ROC. In this case all terms are anticausal,
x@nD = 0.5 ∆@nD + 2 u@-n - 1D - 1.5 2n u@-n - 1D
z z
e) Same X HzL with different ROC. In this case z-1
yields a causal term, and z-2
yields an anticausal
term. Therefore
x@nD = 0.5 ∆@nD - 2 u@nD - 1.5 2n u@-n - 1D
0.5 j 0.5 j
f) X
HzL
z
1
= z
2 +1 =
z+j
-
z-j
z
therefore X HzL = 0.5 j z+j
z
- 0.5 j z-j
. From the ROC
È z È > 1 all terms are causal, therefore
Π Π
x@nD = 0.5 jej 2 n
u@nD - 0.5 je-j 2 n
u@nD
Combining terms:
x@nD = cos I Π2 n + Π2 M u@nD
g) Same X HzL but all anticausal:
Π Π
x@nD = -0.5 jej 2 n
u@-n - 1D + 0.5 je-j 2 n
u@-n - 1D
Combine terms:
Combine terms:
x@nD = cos I Π2 n - Π2 M u@-n - 1D
à Problem 1.24
a) H HzL = 1-0.5 1

z-1
z
= z-0.5
Since it is causal, the impulse response is h@nD = 0.5n u@nD
b) H HzL = 1-0.5 1

z
-2
= z-2
. Since H
HzL
z
-2
= z
Hz-2L
= 1z - z-2
1
z
then H HzL = 1 - z-2
. The impulse
response is anticausal, form the difference equation,
therefore the impulse response is h@nD = ∆@nD + 2n u@-n - 1D.
-1 -2 2 2
c) H HzL = 1+z +z
= z

1-z-2
+z+1
z2 -1
. Therefore H
HzL
z
z +z+1
= -1
=
2

z
+ 1.5
+ 0.5
z-1
which leads to the
z+1
z Iz -1M
z z
decomposition H HzL = -1 + 1.5
z-1
+ 0.5 . Since the
z+1
difference equation is causal, then the
impulse response is
causal, which leads to h@nD = -∆@nD + 1.5 u@nD + 0.5 H-1Ln u@nD.
1 z 2 H HzL z 0.2764 0.7236
d) H HzL = 1-z

-1 -z-2 = z 2 -z-1 , then
z
= z
2 -z-1 = z+0.6180
+ z-1.6180
and therefore
z z
H HzL = 0.2764 z+0.6180
+ 0.7236 z-1.6180
. The difference equation is causal and therefore the
impulse response is
h@nD = 0.2764 H-0.6180Ln u@nD + 0.7236 1.6180n u@nD
à Problem 1.25
z
a) The system is causal with transfer function H HzL = z-0.5
, with ROC È z È > 0.5. The unit
circle is in the ROC and therefore the system is BIBO stable.
The impulse response is h@nD = 0.5n u@nD.
z z
b) Now H HzL = -2 1-2

z
= z-0.5
, same as before. But since it is clearly anticausal (the output
depends
from future values only) the ROC is È z È < 0.5 and the impulse response
h@nD = -0.5n u@-n - 1D. The system is NOT BIBO Stable, since the unit circle is not within the
ROC.
z
c) If the input is x@nD = u@nD, its z-Transform is X HzL = z-1
with ROC È z È > 1.
When the system is causal then
z z z z
Y HzL = z-0.5

z-1
= - z-0.5
+ 2 z-1
, ROC È z È > 1
therefore y@nD = -0.5n u@nD + 2 u@nD;

When the system is anticausal, the z-Transform is the same, but with
different ROC given by the intersection of È z È > 1 and È z È < 0.5, which is empty. Therefore
in this case the output y@nD does not have a z-Transform.
18
When the system is anticausal, the z-Transform is the same, but with
different ROC given by the intersection of È z È > 1 and È z È < 0.5, which is empty. Therefore
in this case the output y@nD does not have a z-Transform.
à Problem 1.26
a) The difference equation can be implemented in a number of ways, so we have

not enough information to assess causality or the ROC of the transfer
function.
z -1 z z
b) The transfer function is H HzL = 1-2.5

z-1 +z-2
= z
2 -2.5 z+1 = Hz-2L

Hz-0.5L
. The possible ROC’s
are the following:
È z È > 2, the system is causal , and NOT BIBO stable, since the unit circle is not
in the ROC;
0.5 < È z È < 2, the system is non causal, but BIBO Stable, since the unit circle is in
the ROC;
È z È < 0.5, the system is anticausal, NOT BIBO Stable since the unit circle is not
in the ROC.
c) The step response has transfer function
z2 z z z
Y HzL = Hz-2L = 0.6667 z-0.5

Hz-0.5L Hz-1L
- 2 z-1
+ 1.333 z-2

When the system is causal the ROC is È z È > 2, then

y@nD = 0.66670 .5n u@nD - 2 u@nD + 1.333 2n u@nD;
When the system is non casual, the ROC is 1 < È z È < 2, then
y@nD = 0.66670 .5n u@nD - 2 u@nD - 1.333 2n u@-n - 1D;
When the system is anticausal, the ROC of the transfer function (0.5 < È z È) and the ROC of the
input ( È z È > 1) have empty intersection, and therefore the output has no z-Transform.
Frequency Response
à Problem 1.27
The transfer function of the system is
1 z
H HzL = 1-0.5

z-1
= z-0.5

Therefore the system is stable (pole at z = 0.5 inside the unit circle) and its frequency response is
Therefore the system is stable (pole at z = 0.5 inside the unit circle) and its frequency response is
e jΩ
H HΩL = e
jΩ -0.5
a) the input is a complex exponential with frequency Ω = 0.2 Π and therefore the corresponding
output is y@nD = H H0.2 ΠL ej0 .2 Πn . Substituting for the frequency response H HΩL we obtain
y@nD = 1.5059 e-j4585 ej0 .2 Πn = 1.5059 ej H0.2 Πn-0.4585L .
b) Since x@nD = ej0 .1 Π ej0 .5 Πn + e-j0 .1 Π e-j0 .5 Πn the response is
y@nD = H H0.5 ΠL ej0 .1 Π ej0 .5 Πn + H H-0.5 ΠL e-j0 .1 Π e-j0 .5 Πn . Substituting for H HΩL
this yields
y@nD = 0.8944 e-j0 .4636 ej0 .1 Π ej0 .5 Πn + 0.8944 ej0 .4636 e-j0 .1 Π e-j0 .5 Πn
= 0.4472 cos H0.5 Πn - 0.1495L
1
c) Since H H0L = 1-0.5
= 2 and H H0.3 ΠL = 1.2289 e-j5202 then
y@nD = 1 2 + 5 1.2289 cos H0.3 Πn - 0.5 Π - 0.5202L which yields
y@nD = 2 + 6.1443 cos H0.3 Πn - 2.0910L;
d) H H0L = 2 and H HΠL = 0.6667, then y@nD = 2 + 0.6667 H-1Ln ;
e) H H0.2 ΠL = 1.5059 e-j4585 , therefore
y@nD = 1.5059 sin H0.2 Πn - 0.4585L + 1.5059 cos H0.2 Πn - 0.4585L;
à Problem 1.28
The system has transfer function H HzL = Z 8h@nD< = Z 90.8n u@nD + 1.25n u@-n - 1D=. This
yields
z z
H HzL = z-0.8
- z-1.25
, ROC: 0.8 < È z È < 1.25
The unit circle is within the ROC therefore the system is stable and it
e jΩ e jΩ
has a Frequency Response H HΩL = e
jΩ -0.8 - e jΩ -1.25 .

a) H H0.2 ΠL = 1.0417, then y@nD = 1.0417 ej0 .2 Πn ;

b) H H0.5 ΠL = 0.2195, then y@nD = 0.4390 cos H0.5 Πn + 0.1 ΠL;
c) H H0L = 9, H H0.3 ΠL = 0.5146, then y@nD = 9 + 2.5731 cos H0.3 Πn - 0.5 ΠL;
d) H H0L = 9, H HΠL = 0.1111, therefore y@nD = 9 + 0.1111 H-1Ln ;
e) H H0.2L = 1.0417, then y@nD = 1.0417 sin H0.2 ΠnL + 1.0417 cos H0.2 ΠnL
à Problem 1.29
-2 z z
The system ha transfer function H HzL = 1-2

z
= z-0.5
and it is anticausal.Therefore the ROC is
given by È z È < 0.5 and the unit circle is NOT in the ROC. As a consequence the system is
UNSTABLE and the steady state frequency response does not exist.
à Problem 1.30
20
Problem 1.30
z
The transfer function is H HzL = 1+z -1 +z-2 , the system is causal and the poles are on the unit circle.

Therefore
ROC: È z È > 1, the unit circle is NOT in the ROC and the system is NOT stable. Again
the frequency response does not exist!
à Problem 1.31
a) From the graph H H0.1 ΠL = 6.5 e-j . Therefore the output is

y@nD = 13 cos H0.1 Πn - 1L
b) From the graph we obtain H H0L = 8, H H0.5 ΠL = 1.8 ejΠ and H HΠL = 0. Therefore the
output is
y@nD = 16 + 1.8 cos H0.5 Πn + ΠL
Time, z-, and Frequency Domains
à Problem 1.32
a) First we have to determine the digital frequency of the disturbance,
as Ω0 = 2 ΠF0 Fs = 0.4 Π. Therefore the filter must have at least two zeros at z1,2 = e±j0 .4 Π , and
its transfer function becomes
b z2 +b z+b
H HzL = b0 + b1 z-1 + b2 z-2 =
0
z2
1 2

Iz-ej0 .4 Π M Iz-e-j0 .4 Π M
= b0 I1 - 0.6180 z-1 + z-2 M
= b0
z2
The poles are both z = 0.

b) With the constraint that H H0L = 1, ie that H HzL = 1 when z = 1 we obtain an equation to solve
for the coefficient b0 . Therefore
H H1L = b0 H1 - 0.6180 + 1L = 1
which yields b0 = 0.7236. Therefore the difference equation of the filter becomes
y@nD = 0.7236 x@nD - 0.4472 x@n - 1D + 0.7236 x@nD
à Problem 1.33
If we keep the same zeros at z1,2 = e±j0 .4 Π , we add two poles close to the zeros as, say,
p1,2 = 0.95 e±j0 .4 Π , then the frequency response becomes more selective. In this way the
Transfer Function becomes
2
H HzL = b0 Hz-z 1 L Hz-z2 L

Hz-p L Hz-p L
z -0.6180 z+1
= b0 z
2 -0.5871 z+0.9025
1 2
In order to have the Frequency Response with H H0L = 1, we need to solve for the coefficient b0 from
the equation
1-0.6180+1
b0 1-0.5871+0.9025
=1
which yields b0 = 0.9518. Finally the difference equation for the filter becomes
y@nD = 0.5871 y@n - 1D - 0.9025 y@n - 2D +
0.9518 x@nD - 0.5883 x@n - 1D + 0.9518 x@n - 2D
à Problem 1.34
Since x@nD is periodic with period N, by definition x@nD = x@n - ND for all n, and therefore
x@nD - x@n - ND = 0 for all n. Therefore in this case the difference equation becomes
y@nD = -a1 y@n - 1D - a2 y@n - 2D + 0
and if the system is stable y@nD ® 0 from any initial condition.
a) When N = 4 the filter has Transfer Function
z -1 4
H HzL = b0 z
2 +a z+a
1 2
The zeros must be the solutions of the equation z4 = 1, ie z4 = ejk2Π with k integer. This yields
z = ejkΠ2 , for k = 0, 1, 2, 3, Therefore the four zeros are z = 1, j, -1, -j
b) Choose the poles close to the zeros, inside the unit circle, at z = Ρ, jΡ, -Ρ, -jΡ, with
0 < Ρ < 1, close to one.
à Problem 1.35
The digital frequencies of the disturbance are Ω1 = 2 Π100 1000 = 0.2 Π,
Ω2 = 2 Π150 1000 = 0.3 Π, Ω3 = 2 Π200 1000 = 0.4 Π.
a) zeros at z1,2 = e±j0 .2 Π , z3,4 = e±j0 .3 Π , z5,6 = e±j0 .4 Π , and poles at p1,2 = Ρe±j0 .2 Π ,
p3,4 = Ρe±j0 .3 Π , p5,6 = Ρe±j0 .4 Π with Ρ again positive, close to one. Say Ρ = 0.9.
b) From the zeros and the poles the transfer function becomes
6 5 4
z -3.4116 z +6.6287 z -7.9988 z +6.6287 z -3.4116 z+1 3 2
H HzL = b0 z
6 -3.0705 z5 +5.3692 z4 -5.8312 z3 +4.3491 z2 -2.0145 z+0.5314
22
6 5 4
z -3.4116 z +6.6287 z -7.9988 z +6.6287 z -3.4116 z+1 3 2
H HzL = b0 z
6 -3.0705 z5 +5.3692 z4 -5.8312 z3 +4.3491 z2 -2.0145 z+0.5314
If we want the DC gain to be one, ie H HΩL = 1 when Ω = 0, then b0 = 0.7664;

c) The difference equation then becomes
y@nD = 3.0705 y@n - 1D - 5.3692 y@n - 2D + 5.8312 y@n - 3D -
4.3491 y@n - 4D + 2.0145 y@n - 5D - 0.5314 y@n - 6D +
+ 0.7664 Hx@nD - 3.4116 x@n - 1D + 6.6287 x@n - 2D -
7.9988 x@n - 3D + 6.6287 x@n - 4D - 3.4116 x@n - 5D + x@n - 6DL
à Problem 1.36
The digital frequency of the signal we want to enhance is Ω0 = 2 Π 1.5 5 = 0.6 Π radians.
a) The poles are p1,2 = Ρe±j0 .6 Π with 0 < Ρ < 1, close to one. We can choose any zeros we like,
say z1,2 = ± 1 to attenuate the low frequencies (Ω = 0) and the high frequencies (Ω = Π).
b) Choose (say) Ρ = 0.9, and the transfer function becomes
z -1 2
H HzL = b0 z
2 -0.5562 z+0.81
We can choose b0 to satisfy any normalization we like. For example, if we want the
frequency response at frequency Ω0 to have unit magnitude, we impose È H H0.6 ΠL È = 1 which
yields b0 = 0.0950.
c) The difference equation becomes
y@nD = 0.5562 y@n - 1D - 0.81 y@n - 2D + 0.0950 x@nD - 0.0950 x@n - 2D
à Problem 1.37
a) The filter has zeros at z1,2 = ± j, poles at p1,2 = ± j0 .9. Its frequency response is as shown;
b) Notch Filter;
c) y@nD = -0.81 y@n - 2D + x@nD + x@n - 2D.
Fourier Analysis of Discrete Time Signals
à Problem 1.38
2 2Π
kn
a) Period N = 3, therefore X@kD = S x@nD e-j
3

, for k = 0, 1, 2. Just substitute for one
n=0
period as x@0D = 1, x@1D = 2, x@2D = 3 to obtain
X@0D = 6.000, X@1D = -1.5 + j0 .8660, X@2D = -1.5 - j0 .8660.
1 2Π
b) Period N = 2, then X@kD = S x@nD e-j
2
kn
= x@0D + H-1Lk x@1D, for k = 0, 1. Therefore
n=0
X@0D = 1 - 1 = 0, X@1D = 1 - H-1L = 2.
3 2Π 3
c) Period N = 4, then X@kD = S x@nD e-j
4
kn
= S x@nD H-jLkn , for k = 0, 1, 2, 3.
n=0 n=0
Substitute the numerical values of the sequence to obtain X@0D = X@2D = 0, X@1D = X@3D = 2 or,
in vector form,
X = @0, 2, 0, 2D
d) The signal x@nD = É cos I Π4 nM É is shown below.
Π
Plot of x@nD = É cos I 4 nM É
The period can be seen by inspection as N = 4 and therefore the expansion is of the form
3 Π
x@nD = 14 S X@kD e-j 2 kn
, with the DFS being
k=0
3
X@kD = S x@nD jkn , for k = 0, 1, 2, 3
n=0
Π
since ej 2
= j. Therefore
24
Π
since ej 2
= j. Therefore
X = @2.4142, 1, -0.4142, 1D
à Problem 1.39
0.2 z
The transfer function of the system is H HzL = z-0.8
, by inspection. The system is causal and stable
therefore we can define
the frequency response as
0.2 e jΩ
H HΩL = e
jΩ -0.8
a) The DFS of the signal is X = @6, -1.5 + j0 .866, -1.5 - j0 .866D. If we call y@nD the
output signal, it is going to be periodic with the same period N = 3 and DFS given by
Y@kD = H Ik 2N
Π
M X@kD, for k = 0, 1, 2. By substitution we find Y@0D = 1-0.80.2
H6L = 6,
2Π
j
0.2 e 3
H-1.5 + j0 .866L = -0.123 + j0 .1846 and
Y@1D =
2Π
j
e 3 -0.8
2Π
j2
0.2 e 3
Y@2D = H-1.5 - j0 .866L = -0.123 - j0 .1846. In vector form:
2Π
j2
e 3 -0.8
Y = @6, -0.123 + j0 .1846, -0.123 - j0 .1846D

b) Period N = 2, therefore Y@kD = H Ik 2
2
Π
M X@kD for k = 0, 1. This yields
k
0.2 H-1L
Y@kD =
k X@kD for k = 0, 1. Substitute the numerical values of X@kD to obtain
H-1L -0.8
Y = @0, 0.5D
c) Period N = 4, therefore Y@kD = H Ik 2
4
Π
M X@kD for k = 0, 1, 2, 3. This yields
k
0.2 H-jL
Y@kD =
k X@kD. Substitute for X@kD to obtain
H-jL -0.8
Y = @0, 0.2439 + j0 .1951, 0, 0.2439 - j0 .1951D
à Problem 1.40
3 2Π
kn
a) The length of the sequence is N = 4. Therefore the DFT becomes X@kD = S x@nD e-j
4

for
n=0
k = 0, 1, 2, 3. Substitute for the sequence to obtain
X = @0, 0, 4, 0D
3 2Π
kn
b) Length N = 4, then again X@kD = S x@nD e-j
4

for k = 0, ..., 3 to obtain
n=0
X = @4, 0, 0, 0D
2 2Π
kn
c) Length N = 3, then X@kD = S x@kD = e-j
3

for n = 0, 1, 2. This yields
n=0
2 2Π
kn
c) Length N = 3, then X@kD = S x@kD = e-j
3

for n = 0, 1, 2. This yields
n=0
X = @6.0000, -1.5000 + j0 .8660, -1.5000 - j0 .8660D

3 2Π
kn
d) Length N = 4, then again X@kD = S x@nD e-j
4

n=0
X = @0, 2, 0, 2D
3 2Π
kn
e) Length N = 4, then again X@kD = S x@nD e-j
4

n=0
X = @1 + j2 .4142, 1 - j2 .4142, 1 - j0 .4142 i, 1 + j0 .4142D

3 2Π
kn
f) Length N = 4, then again X@kD = S x@nD e-j
4

n=0
X = @-4.0961, 4.1904 - j2 .4938, 2.7687, 4.1904 + j2 .4938D

7 2Π
kn
g) Length N = 8, then X@kD = S x@nD e-j
8

n=0
X = @0, 7.0534 + j9 .7082, 0, 0, 0, 0, 0, 7.0534 - j9 .7082D
à Problem 1.41
+¥
Recall the definition X HΩL = DTFT 8x@nD< = S x@nD e-jΩn . Then:
n=-¥
+¥ -1 +¥
a) X HΩL = S 0.8ÈnÈ e-jΩn = S 0.8-n e-jΩn + S 0.8n e-jΩn . After some simple
n=-¥ n=-¥ n=0
manipulations:
+¥ +¥
X HΩL = S 0.8n ejΩn - 1 + S 0.8n e-jΩn = 1-0.8 1

ejΩ
- 1 + 1-0.8 1

e-jΩ
= 1-1.6 0.36

cos HΩL+0.64

n=0 n=0
An alternative way would be to use the z-Transforms. Since, in this case
x@nD = 1.25n u@-n - 1D + 0.8n u@nD

z z
the z-Transform is X HzL = - z-1.25
+ z-0.8
, ROC : 0.8 < È z È < 1.25
the unit circle È z È = 1 is within the ROC. Therefore just substitute z = ejΩ to obtain:
e jΩ e jΩ
X HΩL = - e
jΩ -1.25 + e
jΩ -0.8
and the result follows with some algebra.

+¥ jΩ
b) X HΩL = S 0.5n e-jΩn = 1-0.5 1

e-jΩ
e
= e
jΩ -0.5
n=0
c) x@nD = 1 when 0 £ n £ 4 and x@nD = 0 otherwise. Therefore
26
c) x@nD = 1 when 0 £ n £ 4 and x@nD = 0 otherwise. Therefore

4 -j5Ω
X HΩL = S e-jΩn = 1-e

1-e-j Ω
= e-j1 .5 Ω sin H2.5 ΩL

sin HΩL
n=0
magnitude È X HΩL È
d) x@nD = 1.5 ej0 .2 Π ej0 .1 Πn + 1.5 e-j0 .2 Π e-j0 .1 Πn .

Recall that DTFT 9ejΩ0 n = = 2 Π∆ HΩ - Ω0 L and therefore we obtain
X HΩL = 3 Πej0 .2 Π ∆ HΩ - 0.1 ΠL + 3 Πe-j0 .2 Π ∆ HΩ + 0.1 ΠL
X HΩL magnitude and phase
e) x@nD = -0.5 jej0 .5 Πn + 0.5 je-j0 .5 Πn + 0.5 ej0 .25 Πn + 0.5 e-j0 .25 Πn . Therefore the
DTFT becomes
X HΩL = Πe-j0 .5 Π ∆ HΩ - 0.5 ΠL + Πej0 .5 Π ∆ HΩ + 0.5 ΠL +
+Π ∆ HΩ - 0.25 ΠL + Π∆ HΩ + 0.25 ΠL
28
2
f) x@nD = 2 I-0.5 jej0 .5 Πn + 0.5 je-j0 .5 Πn M =
= 2 I-0.25 ejΠn - 0.25 e-jΠn + 2 0.25M =
= 1 - ejΠn
Therefore X HΩL = 2 Π∆ HΩL - 2 Π∆ HΩ - ΠL =
= 2 Π∆ HΩL + Πe-jΠ ∆ HΩ - ΠL + ΠejΠ ∆ HΩ + ΠL
Notice that we split the term 2 Π∆ HΩ - ΠL = Πe-jΠ ∆ HΩ - ΠL + ΠejΠ ∆ HΩ + Π) to preserve the
symmetry of the DTFT.

g) x@nD = H-1Ln = e-jΠn = 0.5 ejΠn + 0.5 e-jΠn therefore
X HΩL = Π∆ HΩ - ΠL + Π ∆ HΩ + ΠL
again we split it to preserve the symmetry of the DTFT.
30
Fourier Transform
à Problem 1.42
You know that
FT 8rect HtL< = sinc HFL
rectHtL sincHFL
2 2
1.5 1.5
1 1
0.5 0.5
t F
-3 -2 -1 1 2 3 -6 -4 -2 2 4 6
-0.5 -0.5
Apply the properties of the Fourier Transform:
a) FT 8rect H2 tL< = 12 sinc I F2 M
rectH2tL 0.5sincHF• 2L
2 2
1.5 1.5
1 1
0.5 0.5
t F
-3 -2 -1 1 2 3 -6 -4 -2 2 4 6
-0.5 -0.5
b) FT 8rect H2 t - 5L< = FT 8rect H2 Ht - 2.5LL< = e-j2ΠF 2.5 FT 8rect H2 tL< which

yields
FT 8rect H2 t - 5L< = e-j5ΠF 12 sinc I F2 M
c) FT 8rect H5 tL cos H20 ΠtL< = FT 9rect H5 tL 12 Iej2Π 10 t + e-j 2 Π 10 t M=. Since
FT 8rect H5 tL< = 15 sinc I F5 M then
1 1
FT 8rect H5 tL cos H20 ΠtL< = 10
sinc HF - 10L + 10
sinc HF + 10L
xHtL XHFL
1.5 0.2
1 0.15
0.5 0.1
t 0.05
-0.4-0.2
-0.5 0.2 0.4 F
-1 -15-10 -5
-0.05 5 10 15
-1.5 -0.1
d) FT 8sinc H0.1 tL<. By the duality property, FT 8sinc HtL< = rect H-FL. Since the "rect"
function is symmetric, we can write FT 8sinc HtL< = rect HFL. Therefore
32
d) FT 8sinc H0.1 tL<. By the duality property, FT 8sinc HtL< = rect H-FL. Since the "rect"
function is symmetric, we can write FT 8sinc HtL< = rect HFL. Therefore
FT 8sinc H0.1 tL< = 10 rect H10 FL
sincH0.1tL 10 rectH10FL
1.5 14
1.25 12
1 10
0.75 8
0.5 6
0.25 4
t 2
-20 -0.25
-10 10 20 F
-0.5 -0.4-0.2 0.2 0.4
e) FT 8sinc H0.1 tL cos H20 Πt + 0.1 ΠL< = . Therefore

ej0 .1 Π FT 9ej2Π10t sinc H0.1 ΠtL= + e-j0 .1 Π FT 9e-j2Π10t sinc H0.1 ΠtL=
X HFL = ej0 .1 Π 0.1 rect H10 HF - 10LL + e-j0 .1 Π 0.1 rect H10 HF + 10LL
1
f) Since FT 8rect HatL< = ÈaÈ sinc I Fa M , then IFT 9sinc I Fa M= = È a È rect HatL.
Therefore let a = 1 10 and then
IFT 8sinc H10 FL< = 0.1 rect H0.1 tL
g) IFT 8rect H10 FL< = 0.1 sinc H0.1 tL for the same reasons as in f);
+¥ +¥
h) Recall that FT : S ∆ Ht - nT0 L> = F0 S ∆ HF - kF0 L, where F0 = 1 T0 . Also recall that
n=-¥ k=-¥
+¥ +¥
S x HnT0 L ∆ Ht - nT0 L = x HtL S ∆ Ht - nT0 L.
n=-¥ n=-¥
With this in mind we can write

+¥ +¥
S sinc H0.2 nL ∆ Ht - 0.1 nL = sinc H2 tL S ∆ Ht - 0.1 nL, and therefore
n=-¥ n=-¥
+¥ +¥
FT : S sinc H0.2 nL ∆ Ht - 0.1 nL> = FT 8sinc H2 tL< * FT : S ∆ Ht - 0.1 nL>
n=-¥ n=-¥
+¥ +¥
= 0.5 rect H0.5 FL * 10 S ∆ HF - 10 kL = 5 S rect H0.5 HF - 10 kLL
k-¥ k-¥
An alternative way to arrive at the same answer is by using the "rep" and
"comb" operators. Recall that
FT 8combT0 x HtL< = F0 repF0 X HFL

In our case T0 = 0.1, F0 = 1 T0 = 10, x HtL = sinc H2 tL,
X HFL = FT 8sinc H2 tL< = 0.5 rect H0.5 FL. Therefore
In our case T0 = 0.1, F0 = 1 T0 = 10, x HtL = sinc H2 tL,

X HFL = FT 8sinc H2 tL< = 0.5 rect H0.5 FL. Therefore
FT 8comb0.1 sinc H2 tL< = 10 rep10 0.5 rect H0.5 FL
i) Again apply duality to obtain
FT 9repT0 x HtL= = F0 combF0 X HFL
+¥
Therefore FT : S rect Ht - 2 nL> = FT 8comb2 rect HtL< = 0.5 rep0.5 sinc HFL, and
n=-¥
therefore
+¥ +¥
FT : S rect Ht - 2 nL> = 0.5 S sinc H0.5 kL ∆ HF - 0.5 kL
n=-¥ k=-¥
j) By the same arguments as in the previous problem, we can write
+¥
FT : S rect Ht - nL> = FT 8comb1 rect HtL< = rep1 sinc HFL
n=-¥
and therefore
+¥ +¥
FT : S rect Ht - nL> = S sinc HkL ∆ HF - kL
n=-¥ k=-¥
+¥
But now recall that sinc HkL = 0 for all k ¹ 0. Also it is easy to see that S rect Ht - nL = 1
n=-¥
for all t. Therefore
FT 81< = ∆ HFL
since all other terms sinc HkL ∆ HF - kL , for k ¹ 0, in the summation, are zero.
à Problem 1.43
Let a signal x HtL have Fourier Transform as shown.
34
Sketch the following:

+¥ +¥
a) S x H0.01 nL ∆ Ht - 0.01 nL = x HtL S ∆ Ht - 0.01 nL = comb0.01 x HtL. There-
n=-¥ n=-¥
fore its Fourier Transform is given by
+¥
FT : S x H0.01 nL ∆ Ht - 0.01 nL> = 100 rep100 X HFL
n=-¥
shown below
+¥
b) S x Ht - 0.01 nL = rep0.01 x HtL. Therefore its Fourier Transform is 100 comb100 X HFL
n=-¥
shown below
c) FT 8x HtL cos H100 ΠtL< = 12 X IF - 50 + 12 X HF + 50L
2
d) Since cos2 H100 ΠtL = I 12 ej100Πt + 12 e-j100Πt M = 14 Iej200Πt + 2 + e-j200Πt M then
FT 9x HtL cos2 H100 ΠtL= = 12 X HFL + 14 X HF - 100L + 14 X HF + 100L
shown below.
e) From the properties, FT 8x H5 tL< = 15 X I F5 M shown below:
36
f) From the properties again, FT 8x H0.1 tL< = 10 X H10 FL, as shown:

Discrete Time Processing of Continuous Time Signals
Sampling
à Problem 2.1.
Problem:
Consider a sinusoidal signal

x HtL = 3 cos H1000 Πt + 0.1 ΠL
and let us sample it at a frequency Fs = 2 kHz.
a) Determine and expression for the sampled sequence x@nD = x HnTs L and determine its Discrete
Time Fourier Transform X HΩL = DTFT 8x@nD<;
b) Determine X HFL = FT 8x HtL<;
c) Recompute X HΩL from the X HFL and verify that you obtain the same expression as in a).
Solution:
a) x@nD = x HtL È t=nTs = 3 cos H0.5 Πn + 0.1 ΠL. Equivalently, using complex exponentials,
x@nD = 1.5 ej0 .1 Π ej0 .5 Πn + 1.5 e-j0 .1 Π e-j0 .5 Πn
Therefore its DTFT becomes
X HΩL = DTFT 8x@nD< = 3 Πej0 .1 Π ∆ IΩ - Π2 M + 3 Πe-j0 .1 Π ∆ IΩ + Π2 M
with -Π < Ω < Π
b) Since FT 9ej2ΠF0 t = = ∆ HF - F0 L then
X HFL = 1.5 ej0 .1 Π ∆ HF - 500L + 1.5 e-j0 .1 Π ∆ HF + 500L

for all F.
c) Recall that X HΩL = DTFT 8x@nD< and X HFL = FT 8x HtL< are related as
2
c) Recall that X HΩL = DTFT 8x@nD< and X HFL = FT 8x HtL< are related as
+¥
X HΩL = Fs S X HF - kFs L Ë F=ΩFs 2 Π
k=-¥
with Fs the sampling frequency. In this case there is no aliasing, since all
frequencies are contained within Fs 2 = 1 kHz. Therefore, in the interval -Π < Ω < +Π we can write
X HΩL = Fs X HFL É F=ΩFs 2 Π
with Fs = 2000 Hz. Substitute for X HFL from part b) to obtain

Ω Ω
X HΩL = 2000 I1.5 ej0 .1 Π ∆ I2000 2
Π
- 500M + 1.5 e-j0 .1 Π ∆ I2000 2
Π
+ 500MM
Now recall the property of the "delta" function: for any constant a ¹ 0,
1
∆ HatL = ÈaÈ
∆ I ta M
Therefore we can write
X HΩL = 3 Πej0 .1 Π ∆ IΩ - Π2 M + 3 Πe-j0 .1 Π ∆ IΩ + Π2 M
same as in b).
à Problem 2.2.
Problem
Repeat Problem 1 when the continuous time signal is

x HtL = 3 cos H3000 ΠtL
Solution
Following the same steps:

a) x@nD = 3 cos H1.5 ΠnL. Notice that now we have aliasing, since
Fs
F0 = 1500 Hz >
2
= 1000 Hz. Therefore, as shown in the figure below, there is an aliasing at
Fs - F0 = 2000 - 1500 Hz = 500 Hz. Therefore after sampling we have the same signal as in
Problem 1.1, and
everything follows.
à Problem 2.3.
Problem
For each X HFL = FT 8x HtL< shown, determine X HΩL = DTFT 8x@nD<, where x@nD = x HnTs L
is the sampled sequence. The Sampling frequency Fs is given for each case.
a) X HFL = ∆ HF - 1000L, Fs = 3000 Hz;
b) X HFL = ∆ HF - 500L + ∆ HF + 500L, Fs = 1200 Hz
F
c) X HFL = 3 rect I 1000
M, Fs = 2000 Hz;
F
d) X HFL = 3 rect I 1000
M, Fs = 1000 Hz;
e) X HFL = rect I F-3000

1000
M + rect I F+3000

1000
M, Fs = 3000 Hz;
Solution
For all these problems use the relation

+¥
X HΩL = Fs â X HΩ Fs 2 Π - kFs L
k=-¥
+¥ +¥
a) X HΩL = 3000 S ∆ IΩ 3000

2Π
- 1000 - k3000M = 2 Π S ∆ IΩ - 2
3
Π
- k2ΠM;
k=-¥ k=-¥
+¥
b) X HΩL = 1200 S ∆ IΩ 1200

2Π
- 500 - k1200M + ∆ IΩ 1200

2Π
+ 500 - k1200M
k=-¥
+¥
= 2 Π S ∆ HΩ - Ω0 - k2ΠL +http://ebook29.blogspot.com
∆ HΩ + Ω0 - k2ΠL
k=-¥
4
+¥
b) X HΩL = 1200 S ∆ IΩ 1200

2Π
- 500 - k1200M + ∆ IΩ 1200

2Π
+ 500 - k1200M
k=-¥
+¥
= 2 Π S ∆ HΩ - Ω0 - k2ΠL + ∆ HΩ + Ω0 - k2ΠL
k=-¥
Ω0 = 2 Π 500 1200 = Π 1.2;

+¥ +¥
c) X HΩL = 2000 3 S rect I Ω20002 Π

1000
- k 2000
M = 6000 S rect I Ω-k2Π
1000

Π
M shown below.
k=-¥ k=-¥
+¥ +¥
d) X HΩL = 1000 3 S rect I Ω10002 Π

1000
- k 1000
M = 3000 S rect I Ω-k2Π
1000

2Π
M shown below
k=-¥ k=-¥
+¥
e) X HΩL = 3000 S rect I Ω30002 Π-3000

1000
- k 3000
M + rect I Ω30002
1000
Π+3000

1000
- k 3000
M
1000
k=-¥
+¥
= 3000 S rect I 32ΩΠ - 3 - 3 k M + rect I 32ΩΠ + 3 - 3 kM
k=-¥
+¥
= 6000 S rect I Ω-k2Π

2 Π3
M
k=-¥
shown below.
à Problem 2.4.
Problem
In the system shown, let the sequence be

y@nD = 2 cos H0.3 Πn + Π 4L
and the sampling frequency be Fs = 4 kHz. Also let the low pass filter be ideal, with bandwidth
Fs 2.
a) Determine an expression for S HFL = FT 8s HtL<. Also sketch the frequency spectrum
(magnitude only) within the frequency
range -Fs < F < Fs ;
b) Determine the output signal y HtL.
Solution.
From what we have seen, recall that

1 F
S HFL = e-jΠFFs F sinc I F M Y HΩL É Ω=2 ΠFFs
s s
+¥
From Y HΩL = 2 Π S ejΠ4 ∆ HΩ - 0.3 Π - k2ΠL + e-jΠ4 ∆ HΩ + 0.3 Π - k2ΠL we obtain
k=-¥
Y HΩL È Ω=2 ΠFFs =

+¥ F F
2Π S ejΠ4 ∆ I2 Π F - 0.3 Π - k2ΠM + e-jΠ4 ∆ I2 Π F + 0.3 Π - k2ΠM
k=-¥ s s
F +¥
s
= 2 Π 2
Π
S ejΠ4 ∆ HF - 600 - k4000L +
k=-¥
F http://ebook29.blogspot.com
e-jΠ4 ∆ I2 Π
Fs
+ 600 - k4000M
6
Y HΩL È Ω=2 ΠFFs =

+¥ F F
2Π S ejΠ4 ∆ I2 Π F - 0.3 Π - k2ΠM + e-jΠ4 ∆ I2 Π F + 0.3 Π - k2ΠM
k=-¥ s s
F +¥
s
= 2 Π 2
Π
S ejΠ4 ∆ HF - 600 - k4000L +
k=-¥
F
e-jΠ4 ∆ I2 Π
Fs
+ 600 - k4000M
and then
+¥
S HFL = Fs S Ts e-jΠ H600+k4000L4000 sinc I 600+k4000

4000
M ejΠ4 ∆ HF - 600 - k4000L +
k=-¥
F
Ts e-jΠ H-600+k4000L4000 sinc I -600+k4000

4000
M e-jΠ4 ∆ I2 Π F + 600 - k4000M
s
where we used the fact that the ZOH has frequency response Ts e-jΠFFs sinc HF Fs L.
This can be simplified to
+¥ 3Π
S HFL = S H-1Lk e-j
20
3
sinc I 20
+ kM ejΠ4 ∆ HF - 600 - k4000L +
k=-¥
3Π
H-1Lk e-j
20

sinc I -3
+ kM e-jΠ4 ∆ HF + 600 - k4000L
20
In the interval -Fs = -4000 < F < Fs = 4000 we have only terms corresponding to
k = -1, 0, 1. The reader can verify that all other frequencies are outside this
interval. Therefore, for -4000 < F < +4000 we have
S HFL = 0.17 e-j2 .827 ∆ HF - 3400L + 0.9634 e-j0 .1 Π ∆ HF - 600L +
+ 0.9634 ej0 .1 Π ∆ HF + 600L + 0.17 ej2 .827 ∆ HF + 3400L
shown below.
b) Since the Low Pass Filter stops all the frequencies above Fs 2 the output signal y HtL has only
the frequencies at F = ± 600 Hz, and therefore
y HtL = IFT 90.9634 ej0 .1 Π ∆ HF + 600L + 0.9634 e-j0 .1 Π ∆ HF - 600L= =
= 2 0.9634 cos H1200 Πt - 0.1 ΠL
y HtL = IFT 90.9634 ej0 .1 Π ∆ HF + 600L + 0.9634 e-j0 .1 Π ∆ HF - 600L= =

= 2 0.9634 cos H1200 Πt - 0.1 ΠL
à Problem 2.5.
Problem
We want to digitize and store a signal on a CD, and then reconstruct it at

a later time. Let the signal x HtLbe
x HtL = 2 cos H500 ΠtL - 3 sin H1000 ΠtL + cos H1500 ΠtL
and let the sampling frequency be Fs = 2000 Hz.

a) Determine the continuous time signal y HtL after the reconstruction.
b) Notice that y HtL is not exactly equal x HtL. How could we reconstruct the signal x HtL exactly
from its samples x@nD?
Solution
a) Recall the formula, in absence of aliasing,

F
Y HFL = e-jΠFFs sinc I F M X HFL
s
with Fs = 2000 Hz being the sampling frequency. In this case there is no aliasing, since
the maximum frequency is 750 Hz smaller than Fs 2 = 1000 Hz. Therefore, each sinusoid at
frequency F has magnitude and phase scaled by the above expression. Define
jΠF
e- 2000

sin I 2000 ΠF
M
G HFL =
ΠF

2000
which yields
G@250D = 0.9745 e-j0 .392 , G@500D = 0.9003 e-j0 .785 , G@750D = 0.784 e-j1 .178
Finally, apply to each sinusoid to obtain.
8
Finally, apply to each sinusoid to obtain.

y HtL = 2 0.9745 cos H500 Πt - 0.392L - 3 0.9003 sin H1000 Πt - 0.785L +
+0.784 cos H1500 Πt - 1.178L
b) In order to compensate for the distortion we can design a filter with
s -F s F
frequency response 1 G HFL, when
2
< F <
2
.The magnitude would be as follows
à Problem 2.6.
Problem
In the system shown below, determine the output signal y HtL for each of the following input signals
x HtL. Assume the sampling frequency Fs = 5 kHz and the Low Pass Filter (LPF) to be ideal with
bandwidth Fs 2:
a) x HtL = ej2000Πt ;
b) x HtL = cos H2000 Πt + 0.15 ΠL;
c) x HtL = 2 cos H5000 ΠtL;
d) x HtL = 2 sin H5000 ΠtL;
e) x HtL = cos H2000 Πt + 0.1 ΠL - cos H5500 ΠtL.
Solution
Recall the frequency response, in case of no aliasing, is
jΠF
ΠF
ã- 5000

SinJ N
5000
GHFL =
ΠF

5000
with -2500 < F < 2500. Then:
a) G H1000L = 0.935 e-j0 .628 and then y HtL = 0.935 ej H2000 Πt-0.628L
b) Using the same number for 1000Hz we obtain
y HtL = 0.935 cos H2000 Π t + 0.15 Π - 0.628L
b) Using the same number for 1000Hz we obtain

y HtL = 0.935 cos H2000 Π t + 0.15 Π - 0.628L
c) G H2500L = 0.637 e-j1 .5708 , therefore y HtL = 2 0.637 cos H5000 Πt - 1.5708L
d) same: y HtL = 2 0.637 sin H5000 Πt - 1.5708L
e) the term cos H2 Π 2750 tL has aliasing, since it has a frequency above 2500 Hz. From the
figure, the aliased frequency is
Faliased = 5.00 - 2.75 = 2.25 kHz. Therefore it is as if the input signal were
x HtL = cos H2000 Πt + 0.1 ΠL - cos H4500 ΠtL. This yields G H1000L = 0.935 e-j0 .628
and G H2250L = 0.699 e-j0 .393 , and finally
y HtL = 0.935 cos H2000 Πt + 0.1 Π - 0.628L - 0.699 cos H4500 Πt - 1.41372L
à Problem 2.7.
Problem
Suppose in the DAC we want to use a linear interpolation between samples, as

shown in the figure below. We can call this reconstructor a First Order Hold,
since the equation of a line is a polynomial of degree one.
10
+¥
a) Show that y HtL = S x@nD g Ht - nTs L, with g HtL a triangular pulse as shown below;
n=-¥
b) Determine an expression for Y HFL = FT 8y HtL< in terms of Y HΩL = DTFT 8y@nD< and
G HFL = FT 8g HtL<;
c) In the figure below, let y@nD = 2 cos H0.8 ΠnL , the sampling frequency Fs = 10 kHz and the
filter be ideal with bandwidth Fs 2. Determine the output signal y HtL.
Solution
+¥
a) From the interpolation y HtL = S x@nD g Ht - nTs L and the definition of the interpolating
n=-¥
function g HtL we can see that y HtL is a sequence of straight lines. In particular if we look at any
interval
nTs £ t £ Hn + 1L Ts it is easy to see that only two terms in the summation are nonzero, as
y HtL = x@nD g Ht - nTs L + x@n + 1D g Ht - Hn + 1L Ts L, for nTs £ t £ Hn + 1L Ts
y HtL = x@nD g Ht - nTs L + x@n + 1D g Ht - Hn + 1L Ts L, for nTs £ t £ Hn + 1L Ts

This is shown in the figure below. Since g H± Ts L = 0 we can see that the line has to go through the
two points x@nD and x@n + 1D, and it yields the desired linear interpolation.
Interpolation by First Order Hold (FOH)
b) Taking the Fourier Transform we obtain

+¥
Y HFL = FT 8y HtL< = S x@nD G HFL e-j2ΠFnTs
n=-¥
= G HFL X HΩL È Ω=2 ΠFFs
where G HFL = FT 8g HtL<. Using the Fourier Transform tables, or the fact that (easy to verify)
1 t t
g HtL = T rect I T M * rect I T M
s s s
F 2 t F
we determine G HFL = Ts Isinc I F MM , since FT 9rect I T M= = Ts sinc I F M.
s s s
à Problem 2.8.
Problem
In the system below, let the sampling frequency be Fs = 10 kHz and the digital filter have difference
equation
y@nD = 0.25 Hx@nD + x@n - 1D + x@n - 2D + x@n - 3DL
Both analog filters (Antialiasing and Reconstruction) are ideal Low Pass
Filters (LPF) with bandwidth Fs 2.
12
a) Sketch the frequency response H HΩL of the digital filter (magnitude only);
b) Sketch the overall frequency
response Y HFL X HFL of the filter, in the analog domain (again magnitude only);
c) Let the
input signal be
x HtL = 3 cos H6000 Πt + 0.1 ΠL - 2 cos H12 000 ΠtL
Determine the output signal y HtL.
Solution.
-4
a) The transfer function of the filter is H HzL = 0.25 I1 + z-1 + z-2 + z-3 M = 0.25 1-z
, where
1-z-1
we applied the geometric sum. Therefore the frequency response is
-j4Ω
H HΩL = H HzL È z=ejΩ = 0.25 1-e

1-e-jΩ
= 0.25 e-j1 .5 Ω sin H2 ΩL

Ω
sin I 2 M
whose magnitude is shown below.
b) Recall that the overall frequency response is given by

Y HFL F

X HFL
= HH HΩL È Ω=2 ΠFFs L e-jΠFFs sinc I F M
s
In our case Fs = 10 kHz, and therefore we obtain
Y HFL
È€ € €€ È
X HFL
1.4
1.2
1
0.8
0.6
0.4
0.2
F
-4000 -2000 2000 4000
c) The input signal has two frequencies: F1 = 3 kHz < Fs 2, and F2 = 6 kHz > Fs 2, with
Fs = 10 kHz the sampling frequency. Therefore the antialising filter is going to stop
the second frequency, and the overall output is going to be
y HtL = 3 0.156 cos H6000 Π t + 0.1 Π + 0.1 Π - ΠL
= 0.467745 cos H6000 Π t + 0.62832L
since, at F = 3 kHz, Y HFL X HFL = 0.156 ej0 .1 Π .
Quantization Errors
à Problem 2.9
Problem
In the system below, let the signal x@nD be affected by some random error e@nD as shown. The error
is white, zero mean, with variance Σe2 = 1.0. Determine the variance of the error Ε@nD after the filter
for each of the following filters H HzL:
14
a) H HzL an ideal Low Pass Filter with bandwidth Π 4;

z
b) H HzL = z-0.5
;
c) y@nD = 14 Hs@nD + s@n - 1D + s@n - 2D + s@n - 3DL, with s@nD = x@nD + e@nD;
d) H HΩL = e-ÈΩÈ , for -Π < Ω < +Π.
Solution.
Recall the two relationships in the frequency and time domain:
i Π Ä
Ä
Ä
Ä
Ä
Ä y 2
ΣΕ2 j
j
=j 1 Ä
Ä H HΩL Ä
Ä 2 dΩz
z
j 2Π Ù ÄÄ
Ä
Ä
Ä
Ä
Ä
Ä z Σe
z
k -Π Ä Ä {
+¥
= J S Ë h@nD Ë 2N Σe2
-¥
i Π Ä
Ä
Ä
Ä
Ä
Ä y 2 j i 1 Π4 y 2 1 2
j Ä Ä z
a) ΣΕ2 j
=j 1

j2Π Ù Ä
Ä
Ä H HΩL Ä
Ä
Ä
2 dΩz
z j
z Σe = j
z j2Π Ù dΩz
z Σe = 4 Σe ;
z
Ä
Ä
-Π Ä
Ä
Ä
k Ä { k -Π4 {
b) the impulse response in this case is h@nD = 0.5n u@nD therefore
+¥ +¥
ΣΕ2 = J S Ë h@nD Ë 2N Σe2 = K S 0.52 n O Σe2 = 1-0.25
1
Σe2 = 43 Σe2
-¥ 0
c) in this case Ε@nD = 14 He@nD + e@n - 1D + e@n - 2D + e@n - 3DL . Therefore the impulse
response is
c) in this case Ε@nD = 14 He@nD + e@n - 1D + e@n - 2D + e@n - 3DL . Therefore the impulse
response is
h@nD = 14 H∆@nD + ∆@n - 1D + ∆@n - 2D + ∆@n - 3DL
and therefore
3 1 y
ΣΕ2 = i
j S
42
1
z Σe2 = 4
16 e
Σ2 = 14 Σe2
kn=0 {
i 1 Π -ÈwÈ y 2
d) ΣΕ2 = j
j
j 2Π Ù e
j dΩz
z 2
z Σe = 0.3045 Σe
z
k -Π {
à Problem 2.10.
Problem
A continuous time signal x HtL has a bandwidth FB = 10 kHz and it is sampled at Fs = 22 kHz,
using 8bits/sample. The signal is properly scaled so that È x@nD È < 128 for all n.
a) Determine your best estimate of the variance of the quantization error
Σe2 ;
b) We want to increase the sampling rate by 16 times. How many bits per
samples you would use in order to maintain the same level of quantization
error?
Solution
a) Since the signal is such that -128 < x@nD < 128 it has a range VMAX = 256. If we digitize it with
Q1 = 8 bits. we have 28 = 256 levels of quantization. Therefore each level has a range
D = VMAX 2Q1 = 256 256 = 1. Therefore the variance of the noise is Σe2 = 1 12 if we assume
uniform distribution.
b) If we increase the sampling rate as Fs2 = 16 Fs1 , the number of bits required for the same quanti-
zation error becomes
F
Q2 = Q1 + 12 log2 F
s1
= 8 + 12 H-4L = 6 bits sample
s2

Fourier Analysis of Discrete Time Signals
Problems on the DTFT: Definitions and Basic Properties
à Problem 3.1
Problem
Using the definition determine the DTFT of the following sequences. It it does not exist say why:
a) x@nD = 0.5n u@nD
b) x@nD = 0.5 n
c) x@nD = 2n u@−nD
d )x@nD = 0.5n u@−nD
e) x@nD = 2 n
f) x@nD = 3 H0.8L n
cos H0.1 πnL
Solution
a) Applying the geometric series
X HωL = Σ 0.5n e−jωn =

+∞ 1
n=0 1−0.5 e−jω
b) Applying the geometric series
2
X HωL = Σ 0.5n e−jωn + Σ 0.5−n e−jωn =

+∞ −∞
n=0 n=−1
+∞ +∞
Σ 0.5n e−jωn + Σ 0.5n ejωn − 1 = 1
1−0.5 e−jω
+ 1
1−0.5 ejω
−1
n=0 n=0
I−2.+ ω M I−0.5+ ωM
1.5 ω
= −
c) Applying the geometric series
X HωL = Σ 2n e−jωn = Σ 2−n ejωn =

−∞ ∞ 1
n=0 n=0 1−0.5 ejω
d) Applying the geometric series
X HωL = Σ 0.5n e−jωn = Σ 0.5−n ejωn = Σ 2n ejωn → does not converge

−∞ ∞ ∞
n=0 n=0 n=0
and the DTFT does not exist;

e) Applying the geometric series
X HωL = Σ 2n e−jωn + Σ 2− n e−jωn = Σ 2n e−jωn + Σ 2n ejωn → does not converge

+∞ −∞ ∞ +∞
n=0 n=−1 n=0 n=1
and the DTFT does not exist;

f) Expanding the cosine, we can write
x@nD = 3
2
0.8n ej0 .1 πn u@nD + 3
2
1.25n ej0 .1 πn u@−n − 1D +
3
0.8 n
e −j0 .1 πn u@nD + 3
1.25n e−j0 .1 πn u@−n − 1D
2 2
which becomes
I0.8 ej0 .1 π M u@nD + I1.25 ej0 .1 π M u@−n − 1D +
3 n 3 n
x@nD =
I0.8 e−j0 .1 π M u@nD + I1.25 e−j0 .1 π M u@−n − 1D
2 2
3 n 3 n
2 2
Taking the DTFT of every term, recall that
DTFT 8an u@nD< = Σ an e−jωn =

+∞ ejω
ejω −a
, if a <1
n=0
DTFT 8an u@−n − 1D< = Σ an e−jωn = Σ a−n ejωn − 1 = − eejω −a , if

−∞ +∞ jω
a >1
n=−1 n=0
Therefore we obtain
X HwL =
e jw e jw e jw e jw
+ - -
e -0.8äe j0 .1 p
jw
e -0.8äe-j0 .1 p
jw
e -1.25äe j0 .1 p
jw
e -1.25äe-j0 .1 p
jw
Problem 3.2
Problem
Given the fact that DTFT 80.8n u@nD< = 1 ë I1 − 0.8 e−jω M and using the properties, compute
the DTFT of the following sequences:
a) x@nD = 0.8n u@n − 2D
b) x@nD = 0.8n u@nD cos H0.1 πnL
c) x@nD = n0 .8n u@nD
d) x@nD = 0.8−n u@−nD
e) x@nD = :
0.8n if 0 ≤ n ≤ 5
0 otherwise
Solution
a) x@nD = 0.8n u@n − 2D = I0.82 M 0.8n−2 u@n − 2D. Therefore
X HωL = 0.64 e−j2ω 1

1−0.8 e−jω
b) x@nD = 0.8n u@nD cos H0.1 πnL = 1

2
I0.8n u@nD ej0 .1 πn + 0.8n u@nD e−j0 .1 πn M
which yields
X HωL = 1
I 1−0.8 e−j
1
Hω−0.1 πL + 1−0.8 e−j Hω+0.1 πL
1
M
c) X HωL =
2
1 d 1
j dω 1−0.8 e−jω
which yields
X HωL =
ω
I4.−5. ω M2
20.
d) X HωL = Σ 0.8−n u@−nD e−jωn = Σ 0.8n u@nD ejωn =

−∞ +∞ 1
n=0 n=0 1−0.8 ejω
e) x@nD = 0.8n u@nD − 0.8n−6 u@n − 6D which yields
X HωL =
I1−e−j6ω M
1−0.8 e−jω
4
Basic Problems on the DFT
à Problem 3.3
Problem
Compute the DFT of the following sequences

a) x = @1, 0, −1, 0D
b) x = @j, 0, j, 1D
c) x = @1, 1, 1, 1, 1, 1, 1, 1D
d) x@nD = cos H0.25 πnL, n = 0, ..., 7
e) x@nD = 0.9n , n = 0, ..., 7
Solution
a) N = 4 therefore w4 = e−j2πê4 = −j. Therefore

X@kD = Σ x@nD H−jLnk = 1 − H−jL2 k , k = 0, 1, 2, 3. This yields
3
X = @0, 2, 0, 2D
n=0
b) Similarly, X@kD = j + j H−jL2 k + H−jL3 k = j + j H−1Lk + jk , k = 0, ..., 3, which

yields
X = @1 + 2 j, j, −1 + 2 j, −jD
c) N = 8 and w8 = e−j2πê8 = e−jπê4 . Therefore, applying the geometric sum we obtain
1−Hw8 L8 k
X@kD = Σ Hw8 L =: 1−Hw8 Lk
7
nk = 0 when k ≠ 0
n=0
8 when k = 0
since HwN LN = Ie−j2πêN M = 1. Therefore

N
X = @8, 0, 0, 0, 0, 0, 0, 0D
d) Again N = 8 and w8 = e−jπê4 . Therefore
K Σ ej O=
7 π π 7 π π
X@kD = 1
2
4
n
e−j 4
nk
+ Σ e−j 4
n
e−j 4
nk
n=0 n=0
K Σ ej Hk−1L n H−k−1L n
O
7 π 7 π
1
= 2
4 + Σ ej 4
n=0 n=0
Applying the geometric sum we obtain

X = @0, 4, 0, 0, 0, 0, 4D
7 π
1−H0.9L8
e) X@kD = Σ 0.9n e−j 4
nk
= π , for k = 0, ..., 7 Substituting numerically we obtain
1−0.9 e−j
X = @5.69, 0.38 − 0.67 j, 0.31 − 0.28 j, 0.30 − 0.11 j,

k
n=0 4
0.30, 0.30 + 0.11 j, 0.31 + 0.28 j, 0.38 + 0.67 jD
à Problem 3.4
Problem
Let X = @1, j, −1, −jD, H = @0, 1, −1, 1D be the DFT's of two sequences x and h respec-
tively. Using the properties of the DFT (do not compute the sequences) determine the DFT's of the
following:
Solution
a) x@Hn − 1L4 D
Recall DFT@x@Hn − mLN DD = wN −km X@kD. In our case w4 = e−j2πê4 = −j and therefore
DFT@x@Hn − 1L4 DD = H−jL−k X@kD for k = 0, ..., 3
This yields DFT@x@Hn − 1L4 DD = @1, 1, 1, 1D
b) DFT@x@Hn + 3L4 DD = H−jL3 k X@kD = @1, −1, 1, −1D
c) Y@kD = H@kD X@kD from the property of circular convolution. This yields
Y = @0, j, 1, −jD
6
d) DFT 8H−1Ln x@nD< = Σ x@nD e−jπn e−j n Hk+2L

= X@Hk + 2L4 D
3 π 3 π
2
nk
= Σ x@nD e−j 2
where the H ..L4 has been inserted since X@kD is periodic with period N = 4. Finally
n=0 n=0
DFT 8H−1Ln x@nD< = @−1, 1, 0, 1D
e) DFT 8jn x@nD< = DFT :ej n Hk−1L

= X@Hk − 1L4 D. This yields
π 3 π
2
n
x@nD> = Σ x@nD e−j 2
DFT 8jn x@nD< = @−j, 1, j, −1D

n=0
f) x@H−nL4 D = @x@0D, x@3D, x@2D, x@1DD therefore
DFT 8x@H−nL4 D< = x@0D + Σ x@4 − nD H−jLnk

3
Hlet m = 4 − nL = x@0D + Σ x@mD H−jLH4−mL k

n=1
3
= x@0D + Σ x@mD H−jL−mk

m=1
3
Σ x@mD H−jLm H−kL = X@H−kL4 D

m=1
3
=
Finally: DFT 8x@H−nL4 D< = @1, −j, 1, jD.

m=0
g) DFT 8x@H2 − nL4 D< = DFT 8y@Hn − 2L4 D< where y@nD = x@H−nL4 D. Using the result in the
previous problem we obtain
DFT 8x@H2 − nL4 D< = H−jL−2 k X@H−kL4 D
= @1, j, 1, −jD
à Problem 3.5
Problem
Let x@nD, n = 0, ..., 7 be an 8-point sequence with DFT

X = @1, 1 − j, 1, 0, 1, 0, 1, 1 + jD
Using the properties of the DFT, determine the DFT of the following sequences:
Solution
a) DFT :x@nD e > = Σ x@nD e = X@Hk − 1L4 D

n Hk−1L
π 7 π π 7 π
j2 n j2 n −j2 nk −j2
8 8 e 8 = Σ x@nD e 8
n=0 n=0
Therefore DFT :x@nD e > = @1 + j, 1, 1 − j, 1, 0, 1, 0, 1D

π
j2 8
n
b) DFT 8δ@Hn − 2L8 D< = e since DFT 8δ@nD< = 1. Therefore

2π
−j 2k
DFT 8x@nD ⊗ δ@Hn − 2L8 D< = @1, −1 − j, −1, 0, 1, 0, −1, −1 + jD

8
à Problem 3.6
Problem
A 4-point sequence x has DFT X = @1, j, 1, −jD. Using the properties of the DFT determine the
DFT of the following sequences:
Solution
a) From the definition we can write
Y@kD = DFT 8H−1Ln x@nD< = Σ x@nD ejπn e−j

3 π
2
kn
n=0
Hk−2L n
= X@Hk − 2L4 D
3 π
= Σ x@nD e−j 2
n=0
The circular shift comes from the fact that X@kD is periodic with period 4, and therefore any shift is
going to be circular. Substituting for X@kD we obtain
DFT 8H−1Ln x@nD< = X@Hk − 2L4 D = @1, −j, 1, jD.
b) DFT 8x@Hn + 1L4 D< = H−jLk X@kD
c) DFT 8x@nD ⊗ δ@Hn − 2L4 D< = H−jL2 k X@kD
d) DFT 8x@H−nL4 D< = X@H−kL4 D = @1, −j, 1, jD
à Problem 3.7
Problem
Two finite sequences x = @x@0D, x@1D, x@2D, x@3DD and h = @h@0D, h@1D, h@2D, h@3DD have DFT's given by
X = DFT 8x< = @1, j, -1, - jD
H = DFT 8h< = @0, 1 + j, 1, 1 - jD
8
Using the properties of the DFT (do not compute x and h explicitly, compute the following:
a) DFT 8@x@3D, x@0D, x@1D, x@2D<D
b) DFT 8@h@0D, -h@1D, h@2D, -h@3D<
c) DFT 8h ≈ x<, where ≈ denotes circular convolution
d) DFT 8@x@0D, h@0D, x@1D, h@1D, x@2D, h@2D, x@3D, h@3D<
Solution
a) DFT 8@x@3D, x@0D, x@1D, x@2DD< = DFT 8x@Hn - 1L4 D< = w-k
4 X @kD with w4 = e
DFT 8x@3D, x@0D, x@1D, x@2D< = H- jL @1, j, -1, - jD = @1, 1, 1, 1D
-j2pê4
= - j Therefore
k
b) DFT 8@h@0D, -h@1D, h@2D, -h@3DD< =

DFT 8H-1Ln h@nD< = DFT 9e-j2H2 p ê4L n h@nD= = H@Hk - 2L4 D = @1, 1 - j, 0, 1 + jD
c) DFT 8h ≈ x< = H@kD X @kD = @0, -1 + j, -1, -1 - jD

d) Let y = @x@0D, h@0D, x@1D, h@1D, x@2D, h@2D, x@3D, h@3DD, with length N = 8. Therefore its DFT is
Y @kD = S y@nD wnk + S y@2 m + 1D wH2

7 3 3
2 mk m+1L k
8 = S y@2 mD w8 8
n=0 m=0 m=0
Y @kD = X @kD + wk8 H@kD, for k = 0, ..., 7
Y = B1, j + H1 + jL e- 4 , -1 - j, - j + H1 - jL e- , 1, j + H1 + jL e , -1 + j, - j + H1 - jL e 4 F
This yields
jp 3 jp 3 jp jp
4 4
à Problem 3.8
Problem
Two finite sequences h and x have the following DFT's:

X = DFT 8x< = @1, -2, 1, -2D
H = DFT 8h< = @1, j, 1, - jD
Let y = h ≈ x be the four point circular convolution of the two sequences. Using the properties of the
DFT (do not compute x@nD and h@nD),
a) determine DFT 8x@Hn - 1L4 D< and DFT 8h@Hn + 2L4 D<;
b) determine y@0D and y@1D.
Solution
a) DFT 8x@Hn - 1L4 D< = H- jLk X @kD = H- jL-k @1, -2, 1, -2D = @1, 2 j, -1, -2 jD. Similarly
DFT 8h@Hn + 2L4 D< = H- jL2 k H@kD = H-1Lk @1, j, 1, - jD = @1, - j, , 1, jD
Y @kD HH1L H0L + H jL H1 + jL + H-1L H1L + H- jHL 1 - jLL =

3
1 1 -3
b) y@0D = S
4 k=0
= 4 4
and
S Y @kD H- IH1L H0L + H jL H1 + jL H- jL-1 + H-1L H1L H- jL-2 + H- jL H1 - jL H- jL-3 M =

3
-1
jL-k =
1 1
y@1D = 4 k=0 4 4
à Problem 3.9
Problem
Let x be a finite sequence with DFT

X = DFT 8x< = @0, 1 + j, 1, 1 - jD
Using the properties of the DFT determine the DFT's of the following:
a) y@nD = e jHpê2L n x@nD
p
b) y@nD = cosI 2 nM x@nD
c) y@nD = x@Hn - 1L4 D

d) y@nD = @0, 0, 1, 0D ∆ x@nD with ∆ denoting circular convolution
Solution
a) Since e jHpê2L n x@nD = e jH2 pê4L n x@nD then DFT 9e jHpê2L n x@nD= = X @Hk - 1L4 D = @1 - j, 0, 1 + j, 1D
e jH2 pê4L n x@nD + e- jH2 pê4L n x@nD and therefore its DFT is
1 1
b) In this case y@nD =
X @Hk - 1L4 D + X @Hk + 1L4 D = 2 @1 - j, 0, 1 + j, 1D + 2 @1 + j, 1, 1 - j, 0D. Putting things together we
2 2
1 1 1 1
2 2
obtain
Y = A1, 1, 2 E
1 1
,
c) DFT 9x@Hn - 1L4 D = e- jH2 pê4L k X @kD = H- jLk X @kD and therefore
2
Y = @0, 1 - j, -1, 1 + jD
10
d) DFT 8x@Hn - 2L4 D< = H- jL2 k X @kD = H-1Lk X @kD = @0, -1, j, 1, -1 + jD
Problems on DFT: Manipulation of Properties and Derivation of Other

Properties
à Problem 3.11
Problem
You know that DFT 8@1, 2, 3, 4D< = @10, -2 + 2 j, -2, -2 - 2 jD. Use the minimum number of opera-
tions to compute the following transforms:
Solution
a) Let
x = @1, 2, 3, 4D
s = @1, 2, 3, 4, 1, 2, 3, 4D
y = @1, 2, 3, 4, 0, 0, 0, 0D
Then we can write y@nD = s@nD w@nD, n = 0, ..., 7 where
w = @1, 1, 1, 1, 0, 0, 0, 0D
and then, by a property of the DFT we can determine
Y @kD = DFT 8y@nD< = DFT 8s@nD w@nD< = W @kD ≈ S@kD
1
N
Let's relate the respective DFT's:
S@kD = DFT 8s< = S s@nD wn8 k

7
n=0
= S s@nD wn8 k + S s@n + 4D wHn+4L

3 3
k
8
n=0 n=0
= S s@nD wn8 k + H-1Lk S s@nD wn8 k

3 3
n=0 n=0
= I1 + H-1Lk M S s@nD wn8 k

3
n=0
This implies that
S@2 kD = 2 X @kD
S@2 k + 1D = 0
and therefore
S = DFT 8@1, 2, 3, 4, 1, 2, 3, 4D< = 2 @X @0D, 0, X @1D, 0, X @2D, 0, X @3D, 0D
Furthermore:
W = DFT 8w< = S =:
1-H-1Lk
3 if k = 1, ..., 7
w8n k 1-wk8
n=0
4 if k = 0
and therefore
W = @4, 1 - j2 .4142, 0 , 1 - j 0.4142, 0, 1 + j0 .4142, 0, 1 + j 2.4142D
Finally the result:
Y @kD = S@mD W @Hk - mL8 D

7
1
S
8 m=0
X @ pD W AHk - 2 pL8 E
3
1
= S
8 p=0
2
X @ pD W AHk - 2 pL8 E
3
1
= S
4 p=0
where we used the fact that S@mD ∫ 0 only for m even, and S@2 pD = 2 X @ pD.
The final result is simpler than it seems. In fact notice that W @2D = W @4D = W @6D = 0. This implies that
Y @0D = W @0D X @0D = X @0D = 10
1 1
Y @2D = W @0D X @1D = X @1D = -2 + 2 j

4 2
1 1
Y @4D = W @0D X @2D = X @2D = -2

4 2
1 1
4 2
and we have only to compute

Y @1D = HW @1D X @0D + W @7D X @1D + W @5D X @2D + W @3D X @3DL = -0.4142 - j 2.4142
1
Y @3D = HW @3D X @0D + W @1D X @1D + W @7D X @2D + W @5D X @3DL = 2.4142 - j 1.2426
4
1
All other terms are computed by symmetry as Y @8 - kD = Y * @kD, for k = 1, 2, 3.

4
b) Let y = @1, 0, 2, 0, 3, 0, 4, 0D. Then we can write
Y @kD = S y@nD wn8 k = S x@mD w28 m k

7 3
n=0 m=0
since y@2 mD = x@mD and y@2 m + 1D = 0. From the fact that
12
2p
-j 2
w28 = e 8 = w4
we can write
Y @kD = S x@mD wm
4 = X @kD, k = 0, 1, ..., 7
3
k
m=0
and therefore
Y = @10, -2 + 2 j, -2, -2 - 2 j, 10, -2 + 2 j, -2, -2 - 2 jD
à Problem 3.12
Problem
You know that the DFT of the sequence x = @x@0D, ..., x@N − 1DD is
X = @X@0D, ..., X@N − 1DD. Now consider two sequences:
s = @x@0D, ..., x@N − 1D, x@0D, ..., x@N − 1DD
y = Bx@0D, ..., x@N − 1D, 0, ... , 0F
both of length 2 N. Let S = DFT 8s< and Y = DFT@yD.

N times
a) Show that S@2 mD = 2 X@mD, and S@2 m + 1D = 0, for m = 0, ..., N − 1;

b) Show that Y@2 mD = X@mD for m = 0, ..., N − 1;
y@nD = s@nD w@nD where w = B1, ..., 1, 0, ..., 0F is the rectangular sequence.
c) Determine an expression for Y@2 m + 1D, k = 0, ..., N − 1. Use the fact that
N times N times
Solution
a) S@kD = Σ x@nD Hw2 N Lnk + Σ x@nD Hw2 N LHn+NL k

N−1 N−1
from the definition of s@nD. Now
−j H2 πê2 NL 2 N = −1. Also consider that w −Hj2πê2 NL = Hw L 2 . There-
n=0 n=0
1
2N = e
consider that wNk 2N = e N
fore we obtain
S@kD = I1 + H−1Lk M Σ x@nD Hw2 N Lnk = :

2 XA k2 E if k is even
N−1 0 if k is odd
n=0
b) Y@2 mD = Σ x@nD Hw2 N L2 m n . But again Hw2 N L2 = wN and therefore Y@2 mD = X@mD, for
N−1
n=0
m = 0, ..., N − 1
c) Since clearly y@nD = s@nD w@nD for all n = 0, ..., 2 N − 1, we can use the property of the
DFT which states

DFT 8s@nD w@nD< = 1
2N
S@kD ⊗ W@kD
where W@kD = DFT 8w@nD< = Σ Hw2 N Lnk =

1−Hw2 N LNk
1−Hw2 N Lk
N−1
. Therefore
n=0
Y@kD = 1 N−1
Σ
2 N m=0
X@mD H@Hk − 2 mL2 N D
where we take into account the result in part a). ie S@2 mD = X@mD and S@2 m + 1D = 0.
à Problem 3.13
Problem
want to determine the DFT of the sequence x = B0, 1, 2, 3, 4 , 5, 6, 7F with the zero index
A problem in many software packages is that you have access only to positive indexes. Suppose you
↑
n=0
as shown, and the DFT algorithm let you compute the DFT of a sequence like
x@0D, ..., x@N − 1D, what do you do?
Solution
Since x is effectively a period of a periodic sequence, we determine the DFT of one period starting at
n = 0, which yields
X = DFT 8@4, 5, 6, 7, 0, 1, 2, 3D<
à Problem 3.14
Problem
Let X@kD = DFT 8x@nD< with n, k = 0, ..., N − 1. Determine the relationships between X@kD
and the following DFT's:
a) DFT 8x∗ @nD<
b) DFT 8x@H−nLN D<
c) DFT 8Re 8x@nD<<
d) DFT 8Im 8x@nD<<
e) apply all the above properties to the sequence x = IDFT 8@1, −j, 2, 3 jD<
14
Solution
a) DFT 8x∗ @nD< = Σ x∗ @nD wnk

N = K Σ x@nD wN O = X @H−kLN D
N−1 N−1 ∗
−nk ∗
n=0 n=0
b) DFT 8x@H−nLN D< = Σ x@H−nLN D wnk −nk = X@H−kL D

N−1 N−1
N = Σ x@nD wN N
c) DFT 8Re 8x@nD<< = DFT 8x@nD< + DFT 8x∗ @nD< = X∗ @H−kLN D

n=0 n=0
1 1 1 1
X@kD +
d) DFT 8Im 8x@nD<< = DFT 8x@nD< − DFT 8x∗ @nD< = X∗ @H−kLN D

2 2 2 2
1 1 1 1
X@kD −
e) DFT 8x∗ @nD< = @1, −3 j, 2, jD

2j 2j 2j 2j
DFT 8x@H−nLN D< = @1, 3 j, 2, −jD

DFT 8Re 8x@nD<< = @1, −2 j, 2, 2 jD
DFT 8Im 8x@nD<< = @0, 1, 0, 1D
à Problem 3.15
Problem
Let x@nD = IDFT 8X@kD< for n, k = 0, ..., N − 1. Determine the relationship between x@nD
and the following IDFT's:
a) IDFT 8X∗ @kD<
b) IDFT 8X@H−kLN D<
c) IDFT 8Re 8X@kD<<
d) IDFT 8Im 8X@kD<<
e) apply all the above properties to the sequence X@kD = DFT 8@1, −2 j, +j, −4 jD<
Solution
8X∗ @kD< X∗ @kD K 1N Σ N O = x∗ @H−nLN D

∗
1 N−1 N−1
a) IDFT = Σ
N k=0
w−nk
N = X@kD wnk
k=0
b) IDFT 8X@H−kLN D< = 1 N−1

Σ
N k=0
X@H−kLN D w−nk
N = 1 N−1
Σ
N k=0 N = x@H−nLN D
X@kD wnk
c) IDFT 8Re 8X@kD<< = 1

2
IDFT 8X@kD< + 1
2
IDFT 8X∗ @kD< = 1
2
x@nD + 1
2
x∗ @H−nLN D
IDFT 8Im 8X@kD<< = IDFT 8X@kD< − IDFT 8X∗ @kD< = x∗ @H−nLN D

d)
1 1 1 1
2j 2j 2j
x@nD − 2j
e) IDFT 8X∗ @kD< = @1, 4 j, −j, 2 jD

IDFT 8X@H−kLN D< = @1, −4 j, j, −2 jD
IDFT 8Re 8X@kD<< = @1, j, 0, −jD
IDFT 8Im 8X@kD<< = @0, −3, 1, −3D
à Problem 3.16
Problem
Let x@nD be an infinite sequence, periodic with period N. This sequence is the input to a BIBO stable
LTI system with impulse response h@nD, −∞ < n < +∞. Say how you can use the DFT to deter-
mine the output of the system.
Solution
Call x = @x@0D, ..., x@N − 1DD one period of the signal, and let X = DFT 8x<. Then the whole
1 N−1
k2π
infinite sequence x@nD can be expressed as x@nD = Σ
N k=0
X@kD ej N
n
. Therefore the output
becomes
1 N−1
X@kD H I k2π M ej
k2π
n
y@nD = Σ
N k=0 N
N
We can see that the output sequence y@nD is also periodic, with the same period N given by
y@nD = IDFT 9H I k2π
N
M X@kD=, for n, k = 0, ..., N − 1
à Problem 3.17
Problem
Let x@nD be a periodic signal with one period given by B1, −2, 3 , −4, 5, −6F with the zero
↑
n=0
index as shown.
It is the input to a LTI system with impulse response h@nD = 0.8 n
. Determine one period of the
output sequence y@nD.
Solution
The input signal is periodic with period N = 6 and it can be written as
X @kD e
5 p
1 jk n
x@nD = S
6 k=0
3 , for all n,
where
16
X @kD = DFT 8@3, -4, 5, -6, 1, -2D< =

@-3.0, 3.0 + j 1.7321, -3.0 - j 5.1962, 21.0, -3.0 + j 5.1962, 3.0 - j 1.7321D
The frequency response of the system is given by
HHwL = DTFT 80.8n u@nD + 1.25n u@-n - 1D< =

e jw e jw
-
e -0.8
jw
e -1.25
jw
Therefore the response to the periodic signal x@nD becomes
H@kD X @kD e
5 p
1 jk n
y@nD = S
6 k=0
3 , for all n,
with
p p
jk jk
e e
H@kD = HHwL = - , k = 0, ..., 5
3 3
w=kpê3 jk
p
jk
p
e 3 -0.8 e 3 -1.25
Therefore one period of the output signal is determined as

y@nD = IDFT 8H@kD X @kD<, n = 0, ..., 5,
where
H@kD = @9.0, 0.4286, 0.1475, 0.1111, 0.1475, 0.4286D
Computing the IDFT we obtain
y@nD = @-3.8301, -4.6079, -3.8160, -5.4650, -4.6872, -4.5938D, n = 0, ..., 5
The DFT: Data Analysis
à Problem 3.18
Problem
A narrowband signal is sampled at 8 kHz and we take the DFT of 16 points as follows. Determine the
best estimate of the frequency of the sinusoid, its possible range of values and an estimate of the
amplitude:
X = @0.4889, 4.0267 − j24 .6698, 2.0054 − j5 .0782,

1.8607 − j2 .8478, 1.8170 − j1 .8421, 1.7983 − j1 .2136,
1.7893 − j0 .7471, 1.7849 − j0 .3576, 1.7837, 1.7849 + j0 .3576 ,
1.7893 + j0 .7471, 1.7983 + j1 .2136, 1.8170 + j1 .842,
1.8607 + j2 .8478, 2.0054 + j5 .0782, 4.0267 + j24 .6698D
Solution
Looking at the DFT we see that there is a peak for k = 1 and k = 15, withe the respective values
being X@1D = 4.0267 − j24 .6698, and X@15D = 4.0267 + j24 .6698. The magnitude is
Hk0 − 1L Ns < F < Hk0 + 1L Ns

X@1D = X@15D = 25.0259. Therefore the frequency is within the interval
F F
with k0 = 1. Fs = 8 kHz and N = 16. This yield an estimated frequency

8
0 < F < 2 16 kHz = 1 kHz.
à Problem 3.19
Problem
A real signal x HtL is sampled at 8 kHz and we store 256 samples x@0D, ..., x@255D. The
magnitude of the DFT X@kD has two sharp peaks at k = 15 and k = 241. What can you say about
the signal?
Solution
8 8
The signal has a dominant frequency component in the range 14 256
< F < 16 256
kHz, that is to
say 0.4375 kHz < F < 0.500 kHz.
à Problem 3.21
Problem
We have seen in the theory that the frequency resolution with the DFT is ∆F = 2 Fs ê N, with Fs the
sampling frequency and N the data length. Since the lower ∆F the better, by lowering the sampling
frequency Fs , while keeping the data length N constant, we get arbitrary good resolution! Do you buy
that? Why? or Why not?
18
Solution
It is true that by lowering the sampling frequency Fs and keeping the number of points N constant we
improve the resolution. In fact by so doing the sampling interval Ts gets longer and we get a longer
interval of data. However we cannot decrease the sampling frequency Fs indefinitely, since we have to
keep it above twice the bandwidth of the signal to prevent aliasing.
à Problem 3.23
We want to plot the DTFT of a sequence which is not in the tables. Using the DFT and an increasing
number of points sketch a plot of the DTFT (magnitude and phase) of each of the following sequences:
Solution
For every signal in this problem we need to take the DFT of the following sequence
xN = @x@0D, ..., x@N - 1D, x@- ND, ..., x@-1DD
of length 2 N. In other words the first N points correspond to positive indexes, while the last N points
correspond to negative indexes. By increasing the value of the parameter N we observe the conver-
gence of the DFT.
1
a) x@nD = n
u@n − 1D
The following figures show the DFT for different lengths N = 27 , 29 , 211 (magnitude and phase):
10
0
-3 -2 -1 0 1 2 3
-2
-3 -2 -1 0 1 2 3
N = 27
10
0
-3 -2 -1 0 1 2 3
-2
-3 -2 -1 0 1 2 3
N = 29
10
0
-3 -2 -1 0 1 2 3
-2
-3 -2 -1 0 1 2 3
N = 211
Notice that at w = 0 the DFT does not converge (the magnitude increases with N) while it converges
at all other frequencies. As a consequence, we can say that the DFT converges to the DTFT for all
w ∫ 0. This is expected, since the signal x@nD is NOT absolutely summable.
Also notice the discontinuity in the phase at ω = 0.
20
1
b) x@nD = n2
u@n − 1D
The following figures show the DFT for different lengths N = 27 , 211 (magnitude and phase):
1.5
0.5
0
-3 -2 -1 0 1 2 3
-2
-3 -2 -1 0 1 2 3
N = 27
1.5
0.5
0
-3 -2 -1 0 1 2 3
-2
-3 -2 -1 0 1 2 3
N = 211
Notice that the DFT converges everywhere, as expected since the signal is absolutely summable.
1
c) x@nD = n2 +1
The following figures show the DFT for different lengths N = 27 , 211 (magnitude and phase):
22
4
3
2
1
0
-3 -2 -1 0 1 2 3
-2
-3 -2 -1 0 1 2 3
N = 27
4
3
2
1
0
-3 -2 -1 0 1 2 3
-2
-3 -2 -1 0 1 2 3
N = 211
Notice that the DFT converges everywhere, as expected since the signal is absolutely summable.
1
d) x@nD = n +1
The following figures show the DFT for different lengths N = 27 , 29 , 211 (magnitude and phase):
20
15
10
0
-3 -2 -1 0 1 2 3
-2
-3 -2 -1 0 1 2 3
N = 27
20
15
10
0
-3 -2 -1 0 1 2 3
-2
-3 -2 -1 0 1 2 3
N = 29
24
20
15
10
0
-3 -2 -1 0 1 2 3
-2
-3 -2 -1 0 1 2 3
N = 211
Notice that at w = 0 the DFT does not converge (the magnitude increases with N) while it converges
at all other frequencies. As a consequence, we can say that the DFT converges to the DTFT for all
w ∫ 0. This is expected, since the signal x@nD is NOT absolutely summable.
Review Problems
à Problem 3.26
Problem.
In the system shown, let the continuous time signal x HtL have Fourier Transform as shown. Sketch
X HωL = DTFT 8x@nD< for the given sampling frequency.
Solution.
Recall that, for a sampled signal,
DTFT 8x@nD< = Fs Σ X HF − kFs L F=ωFs ê2 π

+∞
with x@nD = x HnTs L, and Ts = 1 ê Fs .

k=−∞
a) Fs = 8 kHz. Then clearly there is no aliasing, and therefore

X ( F + Fs ) X (F ) X ( F − Fs )
−8 − 4.5 − 3.5 − 3 3 3.5 4.5 8
we can write
X HωL = Fs X HFL F=ωFs ê2 π , for −π < ω ≤ π
In other words we just need to rescale the frequency axis (F → ω = 2 πF ê Fs ) and the vertical axis
(multiply by Fs ), as shown below. For the "delta" function recall that the value associated is not the
amplitude but it is the area. This implies
δ HF − F0 L F=ωFs ê2 π = δ I 2 sπ Hω − ω0 LM = 2π
δ Hω − ω0 L
F
where ω0 = 2 πF0 ê Fs .
Fs
X (ω )
8,000
4π 4π
7π 3π 3π 7π ω
−π −
8 − 4 8
π
4
b) Fs = 6 kHz. In this case the two "delta" functions are aliased, as shown below
26
∑ X ( F − kF )
k
s
Λ Λ
3 F (kHz)
6 − 3.5 = 2.5
(aliased)
The aliased frequency is at 6-3.5=2.5kHz. The DTFT of the sampled signal is then given in the figure
below.
X (ω )
6,000
4π 4π
5π 5π ω
−π −
6 6
π
c) Fs = 2 kHz. In this case the whole signal is aliased, not only the narrowband component as in the previous
case. In order to make the figure not too confusing, consider the parts of the signal separately.
For the "broadband" component of the signal, the repetition in frequency yields the following:
Λ Λ
−1 1 2 3 4 5 F (kHz)
Then summing all components we obtain the result shown. Just sum within one period (say
−1 kHz < F ≤ 1 kHz) and repeat periodically.
∑ X ( F − kF )
k
s
5
3
Λ Λ
2
3
−1 1 F (kHz)
For the "narrowband" component (the two "delta" functions) we obtain the following:
Λ 2
Λ
F (kHz)
− 3.5 − 1 1 3.5
3.5 − 2.0 = 1.5
− 3.5 + 2.0 = −1.5 − 3.5 + 2 × 2.0 = 0.5
3.5 − 2 × 2.0 = −0.5
axis ( Fs the vertical axis, and 2 π ê Fs the horizontal axis) to obtain X HωL:
Finally consider only the interval −1 kHz ≤ F < 1 kHz, put the two together and properly rescale the
10 , 000
3
X (ω )
Λ Λ
4, 000
3
2π
−π π π ω
2
28
à Problem 3.27
Problem.
The signal x HtL = 3 cos H2 πF1 t + 0.25 πL − 2 sin H2 πF2 t − 0.3 πL has frequencies
F1 = 3 kHz and F2 = 4 kHz. Determine the DTFT of the sampled sequence for the given sampling
frequencies:
Solution.
The continuous time signal can be written in terms of complex exponentials as

x HtL = 3 j0 .25 π j2πF1 t
2
e e + 32 e−j0 .25 π e−j2πF1 t +
−j0 .3 π j2πF2 t j0 .3 π −j2πF2 t
+je e − je e
Therefore its Fourier Transform becomes
X HFL = 3 j0 .25 π
δ HF − F1 L + 3 −j0 .25 π
HF + F1 L +
ej0 .2 π δ HF − F2 L + e−j0 .2 π δ HF + F2 L
2
e 2
e δ
+
which is shown below.
3
e − j 0.25π
X (F )
2 3
2 e j 0.25π
e − j 0.2π e j 0.2π
−4 −3 3 4 F (kHz )
X HωL = Fs X HFL F=ωFs ê2 π shown below.

a) Fs = 9 kHz. In this case there is no aliasing, and it is just a matter of computing
X (ω )
− j 0.25π
3πe 3πe j 0.25π
2πe − j 0.2π 2πe j 0.2π
−π −
8π − 2π
3
2π
3
8π π ω
9 9
b) Fs = 7 kHz. In this case there is aliasing as shown:
X (F )
3 − j 0.25π
2
e 3
e j 0.25π
2
e − j 0.2π e j 0.2π
−4 −3 3 4 F (kHz )
3 j 0.25π
e 3
e − j 0.25π
2
e j 0.2π e − j 0.2π 2
X ( F + Fs ) X ( F − Fs )
Summing the components within the interval −3.5 kHz ≤ F < 3.5 kHz we obtain only two "delta"
functions as
Σ X HF − kFs L = Ie−j0 .2 π + 3
ej0 .25 π M δ HF − 3, 000L +
e−j0 .25 π M δ HF + 3, 000L

k 2
+Iej0 .2 π + 3
2
Therefore
X HωL = 2 π 1.9285 ej0 .2477 δ Iω − 6π
7
M + 2 π 1.9285 e−j0 .2477 δ Iω + 6π
7
M
c) Fs = 5 kHz. The aliased components are computed from the figure below.
X (F )
3
2 e − j 0.25π 3
e j 0.25π
2
e − j 0.2π e j 0.2π
−4 −3 3 4 F (kHz )
e j 0.2π e − j 0.2π
3
2 e j 0.25π 3
e − j 0.25π
2
− 2 −1 1 2
X ( F + Fs ) X ( F − Fs )
Therefore within the interval −2.5 kHz ≤ F < 2.5 kHz we can write
30
Σ X HF − kFs L = 3
ej0 .25 π δ HF + 2000L + ej0 .2 π δ HF + 1000L +
+ 32 e−j0 .25 π δ HF − 2000L + e−j0 .2 π δ HF − 1000L

k 2
Therefore
X HωL = 3 πej0 .25 π δ Iω + 4π
M + 2 πej0 .2 π δ Iω + 2π
M +
+ 3 πe−j0 .25 π δ Iω − M + 2 πe−j0 .2 π δ Iω − M
5 5
4π 2π
5 5
for −π ≤ ω < π.
à Problem 3.28
Problem.
You want to compute the DCT of a finite set of data., but you have only the DFT. Can you still com-
pute it?
Solution.
From the definition of DCT-II,
XII @kD = C@kD Σ x@nD cos I πk H2 M

2 N−1 n+1L
N k=0 2N
Expanding the cosine term we obtain
Σ x@nD cos I πk H2 M=
N−1 kπ N−1 2π kπ N−1 2π
n+1L 1 j jk n 1 −j −jk n
2N 2
e 2N Σ x@nD e 2N + 2
e 2N Σ x@nD e 2N
k=0 k=0 k=0
Now form the sequence

x0 = @x@0D, x@1D, ..., x@N − 1D, 0, ..., 0D
of length 2 N, zero padded with N zeros, and call X0 @kD = DFT 8x0 @nD<, for
k = 0, ..., 2 N − 1. Then if the sequence x@nD is real it is easy to see that
Σ x@nD cos I πk H2 M= Re :e X0 @kD>, k = 0, ..., N − 1

N−1 kπ
n+1L 1 j 2N
k=0 2N 2
and therefore
XII @kD = C@kD Re :e X0 @kD>, k = 0, ..., N − 1.

kπ
1 j 2N
2N
à Problem 3.29
Problem
In MATLAB generate the sequence x@nD = n - 64 for n = 0, ..., 127.

a) Let X @kD = DFT 8x@nD<. For various values of L set to zero the "high frequency coefficients"
X @64 - L ê 2D =. .. = X @64 + L ê 2D = 0 and take the IDFT. Plot the results;
b) Let XDCT @kD = DCT 8x@nD<. For the same values of L set to zero the "high frequency coefficients"
XDCT @127 - LD =. .. = X @127D = 0. Take the IDCT for each case, and compare the reconstruction with
the previous case. Comment on the error.
Solution
X @64 - L ê 2D =. .. = X @64 + L ê 2D = 0 . Call XL @kD the DFT we obtain in this way, and
a) Take L = 20, 30, 40 and set L coefficients of the DFT to zero, as
`
xL @nD = IDFT 9X̀ L @kD=. Notice that xL @nD is still a real signal, since its DFT is still symmetric around
` `
the middle point.
Plots of xL @nD for L = 20, 30, 40 show that xL @nD is not a good approximation of x@nD. As expected the
` `
errors are concentrated at the edges of the signal.
b) Now let XDCT @kD = DCT 8x@nD< and set L coefficients to zero, for L = 20, 30, 40, as
XDCT @127 - LD =. .. = XDCT @127D = 0. Notice that in this case a real signal always yields a real DCT
32
and viceversa. This is due to the basis functions being "cosines" and not "complex exponentials" as in
the DFT. As a consequence we do not have to worry about symmetry to keep the signal real, as we did
for the DFT in a). The inverse DCT obtained for L = 20, 30, 40 are shown below. Notice that the error
is very small compared to the signal itself, and it is not concentrated at any particular point.
The DFT and Linear Algebra
à Problem 3.30
Problem.
An N ä N circulant matrix A is of the form

a@0D a@1D ... a@N - 1D
a@N - 1D a@0D ... a@N - 2D
A=
ª ∏ ∏ ª
a@1D ... a@N - 1D a@0D
Then, do the following:
Solution.
a) Verify that A@k, nD = a@Hn - kLN D, for n, k = 0, ..., N - 1.

Easy, by inspection. Just notice that the first row (k = 0) is the sequence a@0D, ..., a@N - 1D, and the
k - th row is the first row circularly shifted by k.
b) Verify that the eigenvectors are always given by ek = A1, wkN , w2Nk , ..., wHN-1L E , with
k T
N
-j2pêN
k = 0, ..., N - 1 and wN = e
By multiplying the matrix A by each vector ek we obtain the k -th component of the DFT of each
row is DFT 8a@Hn - kLN D< = wkN X @kD, where X @kD = DFT 8a@nD< is the DFT of the first row. Therefore
row. Since the k -th row is obtained by circularly shifting the first row k times, the DFT of the k -th
X @kD
wkN X @kD
A ek = = X @kD ek
ª
wHN-1L k
X @kD
which shows that ek is an eigenvector, and X @kD is the corresponding eigenvalue.

N
c) Determine a factorization A = E L E*T with L diagonal and E*T E = I.

"Pack" all eigenvectors e0 , ..., eN-1 in an N ä N matrix E as
E = @e0 , e1 , ..., eN-1 D
*T
First notice that E E = N since
k em = :
0 if k ∫ m
e*T
N if k = m
Then the matrix
@e0 , e1 , ..., eN-1 D
1 1
E= E=
N N
is such that E*T E = I.

Then use the fact that the vectors ek are eigenvectors, to obtain
X @0D 0
X @1D
... 0
@e0 , e1 , ..., eN-1 D = @e0 , e1 , ..., eN-1 D

1 1 0 ∏ 0
AE= A
... X @N - 1D
N N 0 ∏ ∏ 0
0 0
which can be written as
34
AE= EL
with L = diagHX @0D, ..., X @N - 1DL. Since E-1 = E*T we obtain the desired decomposition
A = E L E*T
tion between the two sequences. Show that you can write the vector y = @y@0D, ..., y@N - 1DDT in terms
d) Let h@nD and x@nD be two sequences of equal length N, and y@nD = h@nD ≈ x@nD be the circular convolu-
of the product y = H x where x = @x@0D, ..., x@N - 1DDT and H circulant. Show how to determine the
matrix H.
From the definition of circular convolution
y@nD = S h@Hn - kLN D x@kD = S H@k, nD x@kD

N-1 N-1
k=0 k=0
Therefore
y = HT x
with H being circulant.
Y @kD = H@kD X @kD, for k = 0, ..., N - 1, with X , H, Y being the DFT's of x, h, y respectively.
e) By writing the DFT in matrix form, and using the factorization of the matrix H, show that
Decomposing the circulant matrix we obtain H T = IE L E*T M = IE* L ET M and therefore

T
y = E* L ET x
Multiplying on the left by ET we obtain
ET y = L ET x
where L = @H@0D, ..., H@N - 1DD. Now
eT0 Y @0D eT0 X @0D
1 eT1 1 Y @1D 1 eT1 1 X @1D
ET y = y= , and ET x = x=
Y @N - 1D X @N - 1D
N ª N ª N ª N ª
eTN-1 eTN-1
where X @kD = DFT 8x@nD< and Y @kD = DFT 8y@nD<. Therefore, since L = diagHH@0D, ..., H 8N - 1DL we
obtain
Y @kD = H@kD X @kD, for k = 0, ..., N - 1

Digital Filters
Problems on Non Ideal Filters
à Problem 4.1
We want to design a Discrete Time Low Pass Filter for a voice signal. The
specifications are:
Passband Fp = 4 kHz, with 0.8 dB ripple;

Stopband FS = 4.5 kHz, with 50dB attenuation;
Sampling Frequency Fs = 22 kHz.
Determine a) the discrete time Passband and Stopband frequencies, b) the
maximum and minimum values of È H HΩL È in the Passband and the Stopband, where H HΩL is the
filter frequency response.
Solution
a) Recall the mapping from analog to digital frequency Ω = 2 ΠF Fs , with Fs the sampling fre-
quency. Then the passband and stopband frequencies become
Ωp = 2 Π 4 22 rad = 0.36 Π rad, Ωs = 2 Π 4.5 22 rad = 0.41 Π rad;
b) A 0.8 dB ripple means that the frequency response in the passband is within the
interval 1 ± ∆ where ∆ is such that 20 log10 H1 + ∆L = 0.8 This yields ∆ = 100.04 - 1 = 0.096.
Therefore the frequency response within the passband is within the
interval 0.9035 £ È H HΩL È < 1.096. Similarly in the stopband the maximum value is
É H HΩL É < 10-5020 = 0.0031
à Problem 4.2
A Digital Filter has frequency response H HΩL such that
0.95 £ È H HΩL È £ 1.05 for 0 £ Ω £ 0.3 Π
0 £ È H HΩL È £ 0.005 for 0.4 Π £ Ω £ Π
2
0.95 £ È H HΩL È £ 1.05 for 0 £ Ω £ 0.3 Π

0 £ È H HΩL È £ 0.005 for 0.4 Π £ Ω £ Π
Also let the sampling frequency be Fs = 8 kHz. Determine the Passband and Stopband frequencies in
kHz, the Passband
ripple and the Stopband attenuation in dB.
Solution
The passband ripple is given by 20 log10 H1.05L = 0.42 dB, and the attenuation in the stopband
-20 log10 0.005 = 46 dB. The analog passband frequency is 0.3 Π Fs 2 Π = 1.2 kHz and the
stopband 0.4 Π Fs 2 Π = 1.6 kHz
à Problem 4.3
A continuous time filter has frequency response
1
H HFL =
j2ΠF
1+ 1000

Determine the passband and stopband frequencies in Hz, assuming a passband

ripple of 1dB and attenuation of 40dB in the stopband. Also determine the
half power frequency Fc .
Solution.
A passband ripple of 1dB means that the frequency response is within the
interval 1 - ∆ £ È H HFL È £ 1 + ∆ with 20 log10 1 + ∆ = 1, which yields ∆ = 0.12. Since
1
È H HFL È = then we determine the passband from the equation
2 ΠF 2
$%%%%%%%%%%%%%%%%%%%%%%%%%%%
1+I 1000 M
1
= 1 - 0.12 = 0.88
2 ΠF 2
$%%%%%%%%%%%%%%%%%%%%%%%%%%%
1+I 1000 M
which yields F = 85.9Hz. Similarly for the stopband, we need to determine the frequency where
È H HFL È = 10-4020 = 0.01 which yields F = 15, 914 Hz Notice that this filter has a very long
transition region, as we can see
from the plot of its magnitude:
dB
FHHzL
5000 10000 15000 20000
-10
-20
-30
-40
à Problem 4.4
A Digital Filter is defined by the difference equation
y@nD = 0.99 y@n - 1D + x@nD
The filter is clearly recursive. Determine the impulse response h@nD.
a) Is the filter stable?
b) Would you classify it as Low Pass, Band Pass ... or what?
c) Would you feel comfortable in implementing this on a digital machine?
Solution
1 z
a) The filter is stable since its transfer function H HzL = 1-0.99

z-1
= z-0.99
has one pole at
z = 0.99;
b) It is a low pass filter since it has one pole close to z = 1, ie Ω = 0. This makes the frequency
response "large" at small frequencies. A plot
of its magnitude is as follows:
4
ÈGHwLÈ
40
30
20
10
w
0.5 1 1.5 2 2.5 3
à Problem 4.5
A simple averaging filter is defined as
y@nD = 1N Hx@n - 1D + ... + x@n - NDL

This is clearly an FIR Filter.
a) Let N = 4. Determine the transfer function, its zeros and poles;
b) Determine a general form for zeros and poles for any N;
c) By comparing y@nD and y@n - 1D determine a recursive implementation. Also the transfer func-
tion,
together with its zeros and poles of the recursive implementation. Looking at
this example, can we say that "any" recursive filter is IIR?
Solution
a) With N = 4 we obtain the transfer function H HzL = 14 Iz-1 + z-2 + z-3 + z-4 M. After normaliza-
tion this becomes
3 2
H HzL = 14 z
+z +z+1
z4

The are four poles at z = 0 and three zeros from the solution
4
z3 + z2 + z + 1 = 1-z

1-z
=0
Therefore the zeros must be such that z4 = 1, with the exclusion of z = 1. That is to say z4 = ejk2Π
for k = 1, 2, 3, and therefore the zeros are z = jk with k = 1, 2, 3, ie z = j, -1, -j.This is
shown in the z-plane below.
b) Since the transfer function is of the form

N
H HzL = 1N z
1-z
N H1-zL
2Π
the zeros are of the form z = ejk
N

, k = 1, ..., N - 1 and the poles are all at z = 0.
c) Since y@nD = 1N Hx@n - 1D + ... + x@n - NDL and
y@n - 1D = 1N Hx@n - 2D + ... + x@n - N - 1DL by comparing y@nD and y@n - 1D we see
that
y@nD = y@n - 1D + 1N x@n - 1D - 1N x@n - N - 1D
This yields the transfer function
-1 -N-1 N
H HzL = 1N z
-z
1-z-1
= 1N z
1-z
N H1-zL
as we saw before. This is an example of a recursive filter with finite

impulse response (FIR).
6
Problems on FIR Filters
à Problem 4.6
We want to design a Low Pass FIR Filter with the following characteristics:
Passband 10kHz,
Stopband 11kHz, with attenuation of 50dB,
Sampling frequency 44kHz
Determine the causal impulse response h@nD, and an expression for the phase within the passband.
Use one of the
standard windows listed in section 4.3.
Solution
First we have to determine the specifications in the digital freq. domain.
Passband: Ωp = 2 Π 10 44 = 0.4545 Π rad

Stopband: ΩS = 2 Π 11 44 = 0.5 Π rad
Therefore we choose the passband of the ideal filter as ΩC = 12 HΩp + ΩS L = 21

Π = 0.477 Π. We
44
need a Blackman window to satisfy the 50dB attenuation in the
stopband. With this window the transition region has a width of 12 Π N. Since we want a transition
region ΩS - ΩP = 2 Π 44 we determine the filter length N as
2Π

44
³ 12 ΠN
which yields N ³ 12 22 = 264. Therefore we choose N = 265 and a shift L = 132. Finally the
impulse response is
which yields N ³ 12 22 = 264. Therefore we choose N = 265 and a shift L = 132. Finally the
impulse response is
h@nD = hd @n - 132D wBlackman @nD
= sin H0.4545 Π Hn-132LL
wBlackman @nD

Π Hn-132L
which is shown below.
Within the passband the phase is linear and it is given by the expression
H HΩL = -Ω L = -132 Ω
à Problem 4.7
Repeat Problem 2.1 with an equiripple filter using the "remez" function
in Matlab. Plot the two frequency responses and compare the two filters in
terms of performance and complexity.
8
Solution
With Matlab we need first to determine the order of the filter. Use the
function "remezord" as follows:
@N, fo, mo, kD =

remezord H@10 000, 11 000D, @1, 0D, @delta, deltaD, 44 000L;
with delta = 10 ^ H-50 20L the maximum deviation corresponding to 50dB’s. This yields an
order N = 114, in the sense that the transfer function is of the form
H HzL = h@0D + h@1D z-1 + ... + h@114D z-114
The impulse response h@nD is obtained as
h = remez HN, fo, mo, kL
where fo, mo and k are from remezord.
Notice that the order of the equiripple filter N = 114 is considerably smaller than the order of the filter
designed with the
Blackman window in Problem 4.6.
Notice that the order of the equiripple filter N = 114 is considerably smaller than the order of the filter
designed with the
Blackman window in Problem 4.6.
à Problem 4.8
Repeat Problem 4.6 using the Kaiser window.
Solution
Wi the Kaiser window we have to determine the parameters N and Β from the specifications. In particu-
lar we want an
attenuation A = 50 dB which yields a factor Β from the expression
Β = 0.5842 HA - 21L0.4 + .07886 HA - 21L = 4.53
Also the filter length is determine from the expression
A-8 42
N ³ 2.285

DΩ
= 2.285 = 128.717

HΠ 22L
So we can choose N = 129 and L = 64. The frequency response of the filter therefore becomes
sin H0.4545 Π Hn-64LL

Π Hn-64L
wKaiser @nD
Its magnitude is shown below.
10
à Problem 4.9
We want to approximate a filter with frequency response
e-0.1ÈFÈ if É F É < 10 Hz
H HFL = :
0 if È F È > 11 Hz
Let the sampling frequency be Fs = 50 Hz, and the attenuation in the stopband be 40dB. Determine
the impulse
response of a FIR filter which approximates this frequency response. Plot the
frequency response in terms of magnitude and phase to verify that the
approximation holds.
Solution
In the digital domain, let Ω = 2 ΠF Fs and therefore F = ΩFs 2 Π. Therefore the filter’s desired
frequency response becomes
e-5ÈΩÈ2 Π if É Ω É < 2 Π 5 rad
H HΩL = :
0 if È F È > 2.2 Π 5 rad
The ideal filter therefore is going to have a frequency response Hd HΩL given by
e-5ÈΩÈ2 Π if É Ω É < 2.1 Π 5
Hd HΩL = :
0 otherwise
and the impulse response
+2.1 Π5
1
hd @nD = 2
Π Ù e-5ÈΩÈ2 Π ejΩn dΩ = 2.27405-0.795775 Cos@1.31947 nD+1. n Sin@1.31947 nD

5.68512+8.97758 n2

-2.1 Π5
Since we want 40dB attenuation in the stopband we can use a hamming

window, which has a transition region of width 8 Π N. The desired width is DΩ = 2 Π 50 and
therefore N is determined from the equation
2Π

50
³ 8
N
Π
and N ³ 50 4 = 200. Choose N = 201 and L = 100. This yields the impulse response
h@nD = hd @n - 100D whamming @nD
à Problem 4.10
A bandpass filter needs to be designed, to pass a signal within
frequencies 4 kHz and 8 kHz, with two transition regions not exceeding DF = 0.5 kHz . Also we
want the attenuation in the stopband not exceeding 50dB, and
the same error within the passband. Finally let the sampling frequency be Fs = 44 kHz.
a) Determine the impulse response of the ideal filter;
b) Design the
filter using the Kaiser window;
c) Design the filter using the equiripple
method.
Compare the two frequency responses.
Solution
a) In the digital frequency domain we want to design a bandpass filter

which passes the frequencies between 2 Π 4 44 = 2 Π 11 rad and 2 Π 8 44 = 4 Π 11 rad.
Therefore the impulse response of the ideal filter is
2nΠ 4nΠ
4 Π11 -2 Π11 -SinA
11
E+SinA
11
E
hd @nD = 1
ejΩn dΩ + 1
ejΩn dΩ = : if n ¹ 0
2Π Ù 2Π Ù nΠ
2 Π11 -4 Π11 0 if n = 0
b) From the formulas of the Kaiser window we determine the parameters Β and N as follows from the
attenuation A = 50 dB and the width of the stopband DΩ = 2 Π 0.5 44 = Π 44 rad. Recall the
formulas:
0.5842 HA - 21L0.4 + .07886 HA - 21L = 4.53
A-8
N ³ 2.285

DΩ
257.434
which yields a window of length N = 259 and a time delay L = 129. Finally thr impulse response
becomes
h@nD = hd @n - 129D wkaiser @nD
12
à Problem 4.11
You want to design a low pass filter with passband Fp = 2 kHz and stopband FS = 2.5 kHz, with
attenuation of at least 40dB. Let the sampling frequency be Fs = 10 kHz. Using the techniques you
know, determine the design with the least
number of coefficients.
Solution
First we translate the specifications into the digital frequency domain:
2
Passband Ωp = 2 Π 10
= 2
5
Π
Stopband ΩS = 2 Π 2.5

10
= Π2
DΩ = Π2 - 2
5
Π Π
= 10

We know three techniques:
a) Window based: from the desired attenuation we need a hamming window. From
the transition region
DΩ = 8
N
Π Π
³ 10

we obtain the length of the filter N = 81;
b) Kaiser window: applying the formulas with A = 40 and DΩ = 0.1 Π we obtain
40-8
N = 2.2850.1Π
= 45
c) Equiripple Filter: using the matlab function "remezord" we obtain the
order N = 39 which yields the lowest complexity. The corresponding frequency response
is shown below (magnitude only in dB’s).
à Problem 4.12
In the section on equiripple filters we have taken a few things for granted.
In this problem try to prove the following simple facts:
a) when the order of the filter N increases, the maximum error decreases;
b) the solution is unique, in
the sense that, for any given order N, there is only one impulse response h@0D, ..., h@NDwhich
minimizes the maximum error.
Solution
` ` `
a) Call hN = AhN @0D, ..., hN @NDE the optimal solution (in the minmax sense) of order N. Then
the vector
` ` `
hN+1 = AhN @0D, ..., hN @nD, 0E ¹ hN+1
represents an impulse response of order N + 1, not necessarily optimal. Therefore
` `
e HhN+1 L £ e HhN+1 L = e HhN L
The leftmost inequality is due to the fact that hN+1 is not the optimal solution of order N + 1, and the
rightmost equality is due to the fact
14
The leftmost inequality is due to the fact that hN+1 is not the optimal solution of order N + 1, and the
rightmost equality is due to the fact
`
HN+1 HΩL = HN HΩL for all Ω
` `
where HN+1 and HN are the frequency responses of hN+1 and hN .
` `
b) If there were two different solutions hN and gN of the same order N having the sign alternation
` `
property, the difference HN HΩL - GN HΩL would have N + 1 roots against the assumption of being
polynomials (in cos HΩL) of order N.
à Problem 4.13
A Hilbert Transform is a filter with frequency response
Hd HΩL = -jsign HΩL
with sign HΩL = ± 1 for Ω ¹ 0 being the signum function.
a) Plot the magnitude and phase of the filter;
b) Determine the impulse
response hd @nD;
c) Determine a causal approximation h@0D, ..., h@ND using a rectangular window. Plot the
magnitude of the frequency response
for various values of N, say N = 40, 60, 120. Does it converge everywhere? In this case what
would you call the
transition region?
Solution
a) Since È Hd HΩL È = È -jsign HΩL È = 1 for all Ω, and

-Π 2 if Ω > 0
Phase HH HΩLL = :
+Π 2 if Ω < 0
we obtain the plot below.
b) The impulse response becomes

Π nΠ 2
1 2 SinA E
hd @nD =
2Π Ù -jsign HΩL ejΩn dΩ = 2

nΠ
if n ¹ 0
-Π
and hd @0D = 0.
c) The plots of
+L
HL HΩL = S hd @nD e-jΩn
n=-L
are shown below (magnitude only) for L = 20, 40, 60 corresponding to the orders
N = 40, 60, 120 respectively. The transition region is defined around the frequency Ω = 0.
16
N =40
1.2
1
0.8
0.6
0.4
0.2
-3 -2 -1 1 2 3
N =60
1.2
1
0.8
0.6
0.4
0.2
-3 -2 -1 1 2 3
N =120
1.2
1
0.8
0.6
0.4
0.2
-3 -2 -1 1 2 3
Problems on IIR Filters
à Problem 4.14
Using the Bilinear Transformation, determine the order N and the cut off frequency Wc of the analog
prototype filter for the following discrete time design:
a) passband 8 kHz;
b) stopband 9 kHz;
c) passband ripple 0.5 dB;
d) stopband attenuation 40 dB;
e) sampling frequency Fs = 44 kHz.
Solution
First we define the problem in the digital frequency domain:

2 Π F p
= 1.1424 rad
Ωp =
F s
2 Π FS
ΩS =
Fs
= 1.2852 rad
Then we determine the specifications of the Analog Prototype:
pΩ
Wp = Fs 2 Tan I
2
M = 56 554.2 rad sec
SΩ
WS = Fs 2 Tan I
2
M = 65 876.0 rad sec
Now from the Passband Ripple we determine Ε as
∆p = 10-0.520 = 0.944
1 2
Ε = $%%%%%%%%%%%%%%%%%%%%%%%
I ∆ M - 1 = 0.349
p
18
∆p = 10-0.520 = 0.944
1 2
Ε = $%%%%%%%%%%%%%%%%%%%%%%%
I ∆ M - 1 = 0.349
p
Then we determine the order of the filter

1-∆s 2
log K
2 2
O
Ε ∆ s
N ³
W = 37.0762
s
2 log J W N
p
which yields N = 38. Finally we determine the cut off frequency of the filter as
Wc = Wp Ε1N = 58 141.4 rad sec
and therefore its frequency response becomes
1 1
H HWL =
W 2N
=
2 N
1+Ε2 J W N W
p
1+K
1N
O
IWp Ε M
The following plot shows the poles of the filter in the s-plane:
-5 -4 -3 -2 -1
-2
-4
-6
The frequency response (magnitude only) is shown next:
20000 40000 60000 80000 100000

-25
-50
-75
-100
-125
-150
-175
20
à Problem 4.15
A 4-th order Butterworth filter has cut off frequency Wc = 200 Π rad sec.
a) Determine the zeros and poles of the transfer function;
b) What would be its passband and stopband frequencies if we want 1dB
ripple in the passband and 40dB attenuation in the stopband?
c) If we apply
a Bilinear Transformation with sampling frequency Fs = 1 kHz, determine the zeros and poles in the
z- plane.
Solution
a) Recall that an N - th order Butterworth Filter has poles on a circle with radius
Wc = 200 Π rad sec spaced by an angle of 360 2 N = 360 8 = 45 degrees. The poles are
shown in the figure below
and they are given by p1,4 = 200 Πe±j5Π8 and p2,3 = 200 Πe±j7Π8 . All zeros are at s = ¥.
1
b) From the frequency response H HWL = $%%%%%%%%%%%%%%%%%%%%
8 % we solve for Wp and WS , as

W
1+I 200
Π M
H HWp L = 10-120 = 530.673 rad sec

H HWS L = 10-4020 = 1986.89 rad sec
c) Applying the formula for the Bilinear Transformation each pole is mapped
as http://ebook29.blogspot.com
c) Applying the formula for the Bilinear Transformation each pole is mapped
as
2
s+ T
s
z = -
2
s- T
s
This yields poles in the z-plane at 0.673045 ± j 0.433479 and 0.53675 ± j0 .143193 and
four zeros at z = -1.
à Problem 4.16
Repeat Problem 4.14 using a Chebyshev filter. Which one would you choose if
complexity is an issue?
Solution
The specifications of the prototype filter are the same, since they depend on
the original specifications of the filter and on the bilinear transformation.
Recall them here for convenience:
2 Π F p
= 1.1424 rad
Ωp =
F s
2 Π FS
ΩS =
Fs
= 1.2852 rad
p Ω
Wp = Fs 2 Tan I
2
M = 56 554.2 rad sec
S Ω
WS = Fs 2 Tan I
2
M = 65 876.0 rad sec
∆p = 10-120 = 0.944
∆S = 10-4020 = 0.01
1 2
Ε = $%%%%%%%%%%%%%%%%%%%%%%%
I ∆ M - 1 = 0.349
p
Then use the formulas for the Chebyshev Filter, from section 4.4, to
obtain N = 12 (the complexity of the filter) and the frequency response as shown
below.
22
Then use the formulas for the Chebyshev Filter, from section 4.4, to
obtain N = 12 (the complexity of the filter) and the frequency response as shown
below.
The filter order is determined from the formula

i "################################
1-∆2S IΕ2 +1M # +"##############
1-∆2S y
log j j
j
Ε ∆S
z
z
z
k {
N ³
i
j WS y
z
WS 2
log j j
j
j
W
+ $%%%%%%%%%%%%%%%%%%%%%%%
I
W
M - 1 z
z
z
z
j p p z
k {
which yields N = 12. Then the poles of the filter in the s-plane are computed as
!!!!!!!!!!!!!! ! 1N
i
j Ε2 + 1 + 1 y z
j
Β=j z
z ;
j
j z
z
Ε
k {
Wp IΒ2 + 1M
r1 = ;
2Β
Wp IΒ2 - 1M
r2 = ;
2Β
i Π H2 k + 1L Π y i Π H2 k + 1L Π y
p = r2 cos j
j + z
z + j r1 sin j j + z
z , k = 0, ..., N - 1
k 2N 2{ k 2N 2{
The poles in the s-plane are shown below. Notice the two different scales for
the Real and Imaginary axis
40000
20000
-8000 -6000 -4000 -2000
-20000
-40000
The plot of the frequency response is shown below. If you compare it with
the Butterworth filter in Problem 4.14, notice that you obtain the same
attenuation with a lower complexity (N = 12 for Chebyshev and N = 38 for Butterworth).
The plot of the frequency response is shown below. If you compare it with
the Butterworth filter in Problem 4.14, notice that you obtain the same
attenuation with a lower complexity (N = 12 for Chebyshev and N = 38 for Butterworth).
20000 40000 60000 80000 100000
-20
-40
-60
-80
-100
à Problem 4.17
We want to implement the analog filter with transfer function
2 s+1
H HsL = s
2 +s+1
by a discrete time approximation, using the Bilinear Transformation. Let

Fs = 10 Hz be the sampling frequency.
a) Determine zeros and poles of both the analog filter H HsL and the discrete time implementation
H HzL;
b) Determine the Linear Difference Equation of the discrete time
implementation;
c) Plot the frequency responses of the digital filter H HΩL and the analog filter H HWL. Verify that you
can obtain one from the other by the appropriate
frequency transformation.
Solution
a) The zeros of the analog system are s = -1 2, and s = ¥ (yes this is a zero too!). The poles are the
solution of s2 + s + 1 = 0 which yields s = e±j2Π3
Applying the mapping s® z
24
Applying the mapping s® z

2
s+
T s
z = -
2
s- T
s
we can verify that the zeros are mapped as

s = -1 2 ® z = 0.9512
s=¥ ® z = -1
and the poles
s = e±j2Π3 ® z = 0.947743 ± j 0.0822827
b) Since we know the zeros and the poles we can write the transfer function
to be
Hz+1L Hz-0.9512L
H HzL = K Hz-0.947743

-j 0.0822827L Hz-0.947743 +j 0.0822827L
with the constant K to be determined. Combining terms we obtain

2
H HzL = K zz
+ 0.0488 z-0.9512
2 -1.8955 z+ 0.9050
We can determine K by matching the value at one frequency component, say at s = 0 ® z = 1:

2 s+1 2
H HsL È s=0 2 +s+1 = 1
= s ® H HzL È z=1 = K zz
+ 0.0488 z-0.9512
2 -1.8955 z +0.9050 É z=1 = 10.2737 K
Equating the two we obtain K = 1 10.2737 = 0.0973361. Finally the transfer function
2
H HzL = 0.0973361 zz
+ 0.0488 z-0.9512
2 -1.8955 z+ 0.9050
and the difference equation

y@nD - 1.8955 y@n - 1D + 0.9050 y@n - 2D =
0.09763361 Hx@nD + 0.0488 x@n - 1D - 0.9512 x@n - 2DL
à Problem 4.18
An integrator has transfer function
H HsL = 1s
a) Determine a discrete time implementation using Euler’s approximation,
with sampling frequency Fs = 10 Hz. Sketch the frequency response, and verify that it approximates
the
integrator for low frequencies;
a) Determine a discrete time implementation using Euler’s approximation,
with sampling frequency Fs = 10 Hz. Sketch the frequency response, and verify that it approximates
the
integrator for low frequencies;
b) Same as a), using the Bilinear Transformation.
Solution
a) By the Euler approximation

-1
s = 1-z

T

s
we obtain
z
H HzL = 0.1 z-1

The frequency response is
e jΩ e jΩ
H HΩL = 0.1 e
jΩ -1 = 0.1 2

j e-jΩ2 sin HΩ2L
For Ω small, we can approximate ejΩ = 1 and sin HΩ 2L = Ω 2, and therefore we see that, for
small Ω the discrete time system behaves like an integrator with frequency
response
1 1
H HΩL = jΩ

F
= j2ΠF
É F=ΩFs 2 Π
s
The following plot compares the frequency response È H HΩL È with the ideal integrator. We can see
that in this case the Euler
approximation gives a good approximation.
60
40
20
0.5 1 1.5 2 2.5 3
-20
b) With the Bilinear Transformation we have

T -1
s 1+z
H HzL =
2 1-z-1
and the frequency response is

Ω
0.1 cos I 2 M
H HΩL =
2 j sin I Ω M
2
26
Ω
0.1 cos I 2 M
H HΩL =
2 j sin I Ω M
2
The following plot compares the frequency response È H HΩL È with the ideal integrator. Even here
we can see that in this case the
Euler approximation gives a good approximation. Notice that at z = -1 (ie Ω = Π), the frequency
response is zero, and in the dB plot it goes to -¥.
100
50
0.5 1 1.5 2 2.5 3
-50
-100
-150
à Problem 4.19
You want to design an analog Band Pass Filter which passes the frequencies in
the interval
5 kHz £ F £ 6 kHz
with 1dB ripple in the passband. Let the the filter be Butterworth with
order N = 4.
a) Determine the frequency transformation Low Pass to Band Pass you would
use;
b) Determine the frequency response of the corresponding Low Pass Filter,

together with its zeros and poles;
c) Determine the zeros and poles of the Band Pass Filter and its transfer
function.
b) Determine the frequency response of the corresponding Low Pass Filter,
together with its zeros and poles; http://ebook29.blogspot.com
CurrentValue[FileName] 27
c) Determine the zeros and poles of the Band Pass Filter and its transfer
function.
Solution
From the specifications we determine the lower and upper frequencies as
WL = 2 Π 5000
WU = 2 Π 6000
Then we choose a prototype Butterworth filter of order N = 4 and with arbitrary cut off frequency, say
WC = 1
which has poles at
äΠ ä H2 k+1L Π
pk = WC ã
2 ã

2 4 , k = 0, 1, 2, 3
In order to apply the proper transformation (Low Pass to Band Pass)
WC Is2 + WL WU M
q HsL =
s HWU - WL L
we compute the poles of the bandpass filter from the equations
q HsL = pk , k = 0, 1, 2, 3
Each equation is quadratic and it yields two solutions. As a total we have
2 4 = 8 poles for the bandpass filter which are given by
poles = 8-1303.33 + 37 418.3 ä, -1101.14 - 31 613.4 ä,
-3004.15 + 35 515.3 ä, -2800.76 - 33 110.8 ä, -3004.15 - 35 515.3 ä,
-2800.76 + 33 110.8 ä, -1303.33 - 37 418.3 ä, -1101.14 + 31 613.4 ä<
The filter has also four zeros at s = 0, due to the fact that q H0L = ¥. From the zeros and the poles we
determine the transfer function. The
magnitude of the frequency response is shown below
28
-16
6´10
-16
5´10
-16
4´10
-16
3´10
-16
2´10
-16
1´10
10000 20000 30000 40000 50000 60000
Problems Involving Material from Previous Chapters
à Problem 4.20
Recall that, when we apply a digital filter to a continuous time signal,
we need two analog filters: Anti-Aliasing and Reconstruction. Due to
hardware constraints and cost, these two filters cannot have a very large
order, and they represent a constraint in our design. Suppose we want to
design a digital filter for a signal having bandwidth of 8 kHz, and we have to use a 5 pole Butterworth
filter (commercially available
for low cost) for both antialiasing and reconstruction filters. Also we want
1dB passband ripple and 50dB attenuation in the stopband.
a) What do you
think would be the minimum sampling frequency we can have, with these two
analog filters. (Hint: recall what is the passband and what is the stopband
of the analog antialising and reconstruction filters in terms of signal
bandwidth and sampling frequency);
b) The 5 pole Butterworth filter you

buy, often is based on switched capacitor technology. This allows to select
the cut off frequency Wc fairly easily, by adjusting the frequency of an oscillator. Given the
sampling frequency you determined in question Q1, determine a suitable value
for the cut off frequency Wc of the filter.
Solution
a) From the specifications of the problem, we have to pass all frequencies

up to 8 kHz. This yields
Wp = 2 Π 8000 rad/sec
We want 1dB ripple in the passband. This gives a value of Ε as
Ε = 0.509
30
Ε = 0.509
Therefore the frequency response becomes
1
È H HWL È = $%%%%%%%%%%%%%%%%%%%%%%%

W 10

2
1+Ε J N Wp
with Ε and Wp as given. Now we have to find the stopband frequency, from the
requirement of 50dB attenuation. This leads to the equation
1
È H HWL È = = 10-5020
&'''''''''''''''''''''''''''''''''
W 10
1 + Ε2 I W M
p
which we solve for W .This yields the stopband frequency
WS =181950. rad/sec
or, in Hertz,
W S
FS = 2
Π
= 28 958.2 Hz
Now the sampling frequency Fs has to be such that
FS £ Fs - 8000 Hz
which yields a sampling frequency
Fs ³ FS + 8000 = 28 958.2 + 8000 = 36.9582 kHz
b) The cut-off frequency WC of the filter is obtained by solving
1 1
È H HWL È = =
&'''''''''''''''''''''''''''''''''
W 10 !!!!!
1 + Ε2 I W M 2
p
which yields WC = 57 537.7 rad sec. The figure below illustrates the problem.
à Problem 4.21
In your CD the data is sampled at 44.1kHz (CD quality), and we want to have a
good sound quality up to 21kHz. If you had a to use an analog Butterworth
filter as a reconstruction filter, what would be the order of your filter? Do
you think you can reasonably build a filter with that complexity? (I do not
think so either!) The chapter on Multi Rate DSP is going to show you how the
CD technology solves this problem.
Solution
Since we want the filter to pass the signal and reject all frequencies
above Fs 2, we can see that passband and stopband frequencies become
Wp = 2 Π 21 000 rad sec
WS = 2 Π44100 2 = 2 Π 22 050 rad sec
Assuming a 1dB passband ripple and 40dB attenuation in the stopband, this
would yield a frequency response of the form
1
È H HWL È = $%%%%%%%%%%%%%%%%%%%%%%%

W 2N %
2
1+Ε J N Wp
with Ε =0.509and WP as given. For an attenuation of 40dB we obtain N > 108 after solving
1
È H HWS L È = $%%%%%%%%%%%%%%%%%%%%%%%
2 N % £ 0.01

W
2 S
1+Ε J W N
p
32
1
È H HWS L È = $%%%%%%%%%%%%%%%%%%%%%%%
2 N % £ 0.01

W
2 S
1+Ε J W N
p
for the order N.

Digital Filter Implementation
State Space Realizations
à Problem 5.1.
Problem
Given the system with transfer function

2 z+1
H HzL = z
3 +2 z2 +z+1
determine:
a) the difference equation relating the input x@nD to the output y@nD;
b) a block diagram realization, together with its state space equations;
c) Its diagonal state space realization;
A realization made of blocks of first and second order only.
Solution
a) Call x@nD and y@nD the input and output sequences respectively. Then the difference equation
is
y@nD + 2 y@n - 1D + y@n - 2D + y@n - 3D = 2 x@n - 2D + x@n - 3D
b) Block Diagram representation for a Type I realization:
2
From the diagram we can determine the state space equations as
s1 @n + 1D = s2 @nD
s2 @n + 1D = s3 @nD
s3 @n + 1D = -s1 @nD - s2 @nD - 2 s3 @nD + x@nD
y@nD = s1 @nD + 2 s2 @nD
In matrix form these become
i 0 1 1 y i 0y
j
j z
z j
j z
z
s@n + 1D = j
j 0 0 1 z
j z s@nD + j
z j0z
j z x@nD
z
k -1 -1 -2 { k1{
y@nD = H 1 2 0 L s@nD
where s@nD = H s1 @nD s2 @nD s3 @nD LT being the state vector.
c) The eigenvalues and eigenvectors of the matrix A are the following:
T
Λ1 = -0.1226 + j 0.7449, q1 = @-0.6626 - j 0.2981, 0.3032 - j 0.4570, 0.3032 + j 0.2819D
T
Λ2 = -0.1226 - j 0.7449, q2 = @-0.6626 + j 0.2981, 0.3032 + j 0.4570, 0.3032 - j 0.2819D
Λ3 = -1.7549, q3 = @-0.2715, 0.4765, -0.8362D
Calling Q = @q1 , q2 , q3 D the 33 matrix of eigenvenvectors, the diagonal realization is given by the
matrices
ij -0.1226 + j 0.7449 0 0 yz
-1 jj zz
A = Q A Q = jj j 0 -0.1226 - j 0.7449 0 zz
j zz
k 0 0 -1.7549 {
T
B = Q-1 B = H0.3872 + j 0.3395, 0.3872 - j 0.3395, -1.1440L

C = H -0.0561 - j 1.2120, -0.0561 + j 1.2120, 0.6815L
ij -0.1226 + j 0.7449 0 0 yz
-1 jj zz
A = Q A Q = jjj 0 -0.1226 - j 0.7449 0 zz
zz
j
k 0 0 -1.7549 {
T
B = Q-1 B = H0.3872 + j 0.3395, 0.3872 - j 0.3395, -1.1440L

C = H -0.0561 - j 1.2120, -0.0561 + j 1.2120, 0.6815L
As you notice the eigenvalues are complex and in general we want to avoid
using complex operations if we do not need to. A better realization would be
to the define the transformation matrix Q using the real and imaginary parts of the eigenvectors. In this
way
define
ij -0.6626 -0.2981 -0.2715 yz
j z
Q = @Re 8q1 <, Im 8q1 <, q3 D = jjjj 0.3032 -0.4570 0.4765 zzzz
j z
k 0.3032 0.2819 -0.8362 {
and therefore we obtain

ij -0.1226 0.7449 0 yz
-1 j
j zz
j
A = Q A Q = jj 0.7449 -0.1226 0 zz
j zz
k 0 0 -1.7549 {
T
B = Q-1 B = H0.7743, -0.6790, -1.1440L

C = C Q = H-0.0561, -1.2120, 0.6815L
d) Factor the transfer function in terms of zeros and poles
1
2 Jz+ N 2 z+1 1
2
HHzL =
2
= J
2
N I Hz+1.7549L

M
Iz +0.2452 z+0.5699M Hz+1.7549L z +0.2452 z+0.5699
à Problem 5.2
Problem
Repeat problem 5.1 for the transfer function

5 z-2
H HzL = z
2 +z+1
Solution
4
Solution
a) Difference equation
y@nD = -y@n - 1D - y@n - 2D + 5 x@n - 1D - 2 x@n - 2D
b) The matrices for the state space realization
i 0 1 yz
A = jj z
k -1 -1 {
T
B = H0, 1L
C = H-2, 5L
D=0
c) Eigenvalues and eigenvectors

T
Λ1 = -0.5 + j 0.866, q1 = @0.6124 - j 0.3536, j 0.7071D
T
Λ2 = -0.5 - j 0.866 q2 = @0.6124 + j 0.3536, - j 0.7071D
Then the diagonal realization
i -0.5 + j 0.866 0 yz
A = Q-1 A Q = jj z
k 0 -0.5 - j 0.866 {
T
B = Q-1 B = H0.4082 - j 0.7071, 0.4082 + j 0.7071L

C = C Q = H -1.2247 + j 4.2426, -1.2247 - j 4.2426L
d) The filter is already a second order system with complex poles, and it
cannot be reduced any further.
à Problem 5.3
Problem

2
2 z +2 z+2
H HzL = z
4 -1.6 z3 0.64 z2 1.024 z-0.8192
Solution
a) The difference equation

y@nD =
1.6 y@n - 1D - 0.64 y@n - 2D - 1.24 y@n - 3D + 0.8192 y@n - 4D + 2 x@n - 2D + 2 x@n - 3D + 2 x@n - 4D
b) Matrices for the State Space equations:
b) Matrices for the State Space equations:

ij 0 1 0 0 yz
jj z
jj 0
j 0 1 0 zzzz
A = jj z
jj 0
jj 0 0 1 zzzz
z
k 0.8192 -1.024 -0.64 1.6 {
T
B = H0, 0, 0, 1L
C = H 2, 2, 2, 0L
c) Eigenvalues and eigenvectors:
Λ1 = 0.8 + j0 .8,
q1 = @0.3487 - j 0.2113, 0.4480 + j 0.1099, 0.2704 + j 0.4463, -0.1407 + j 0.5734D
Λ2 = 0.8 - j0 .8,
q2 = @0.3487 + j 0.2113, 0.4480 - j 0.1099, 0.2704 - j 0.4463, -0.1407 - j 0.5734D
Λ3 = 0.8, q3 = @0.6577, 0.5262, 0.4209, 0.3367D
Λ4 = -0.8, q4 = @0.6577, -0.5262 , 0.4209, -0.3367D
A real diagonal realization (with all real entries) is obtained from the
transformation matrix
Q = @Re 8q1 <, Im 8q1 <, q3 , q4 D

which yields
ij 0.8 0.8 0 0 yz
jj z
jj -0.8 0.8 0
j 0 zzzz
A = Q-1 A Q = jj z
jj 0
jj 0 0.8 0 zzzz
z
k 0 0 0 -0.8 {
T
B = Q-1 B = H-2.1353 , 0.1735, 1.4848, -0.2970 L

C = C Q = H2.1343, 0.6901, 3.2096, 1.1049L
d) Factorize numerator and denominator:
2 Iz2 +z+1M 2 Iz2 +z+1M 1 1
HHzL =
2
= K
2
O I z-0.8
M I z+0.8
M
Iz -1.6 z+1.28M Hz-0.8L Hz+0.8L z -1.6 z+1.28
à Problem 5.4
Problem
You want design a Low Pass Filter by appropriately placing zeros and
poles. You come up with the transfer function
2
Hz+1L
H HzL = K
3
Hz-0.95L
6
2
Hz+1L
H HzL = K
3
Hz-0.95L
a) Determine the value of the gain K so that È H HΩL È Ω=0 = 1;

b) Determine the difference equation associated to this system;
c)
Determine a Type I state space realization;
d) In the transfer function,
perturb the denominator coefficient of z0 by any value of your choice, smaller in magnitude than
10-3 . Is the system still stable? Would you trust this filter implement on
fixed point arithmetic?
e) Implement the filter as the cascade of low order sections (first or second
order). How can you guarantee stability in the presence of numerical
errors?
Solution
a) H HΩL È Ω=0 = H HzL È z=1 = K 22 0.053 = 1. Solve for K to find the gain K = 3.125 105
b) Expanding the transfer function as

2
H HzL = 10-5 -0.857375+2.7075
3.125+6.25 z+3.125 z

z-2.85 z2 +z3

we obtain the difference equation
y@nD - 2.85 y@n - 1D + 2.7075 y@n - 2D - 0.857375 y@n - 3D =
10-5 H3.125 x@n - 1D + 6.25 x@n - 2D + 3.125 x@n - 3DL
`
d) If we perturb the denominator polynomial to be D HzL = z3 - 2.85 z2 + 2.7075 z - 0.856
you can verify that one of the roots is outside the unit circle, and
therefore the system is unstable.
z+1 z+1 1
e) By simple factorization, we can write H HzL = K I z-0.96

M I z-0.96
M I z-0.95 M . Each section of
the filter is stable and it is easy to guarantee
stability of the whole system.
z+1 z+1 1
e) By simple factorization, we can write H HzL = K
I z-0.96 M
I z-0.96 M
I z-0.95 M . Each section of
the filter is stable and it is easy to guarantee
stability of the whole system.
à Problem 5.5
Problem
Given the system shown
Determine the State Space equations and the Transfer Function.
Solution
In this class of problems, it is easy to go from the block diagram to a state

space realization. The transfer function then comes easy.
To determine a state space realization call (say) s1 @nD and s2 @nD the output of each time delay as
shown in the figure. Then we can write:
s1 @n + 1D = -0.5 s1 @nD + 0.5 x@nD
s2 @n + 1D = -0.5 s1 @nD + 0.5 x@nD
y@nD = -0.5 s1 @nD + s1 @nD + s2 @nD + 0.5 x@nD
In matrix form it becomes:
-0.5 0 0.5
s@n + 1D = J N s@nD + J N x@nD
-0.5 0 0.5
y@nD = H 0.5 1 L s@nD + 0.5 x@nD
The Transfer function therefore becomes
z + 0.5 0 -1 0.5
H HzL = H0.5 1L J N J N + 0.5 =
0.5 z 0.5
1 z 0 0.5
= z
Hz+0.5L
H0.5 1L J NJ N + 0.5
-0.5 z + 0.5 0.5
8
z + 0.5 0 -1 0.5
H HzL = H0.5 1L J N J N + 0.5 =
0.5 z 0.5
1 z 0 0.5
= z
Hz+0.5L
H0.5 1L J NJ N + 0.5
-0.5 z + 0.5 0.5
0.75 z
= z
Hz+0.5L
+ 0.5 = 0.5 z+1

z+0.5
à Problem 5.6.
Problem
You have given the following program:

begin loop
read input x@nD
y[n]=v1+x[n];
output y[n]
v1_new=v1+v2- 0.5v3;
v2_new=v1- v2- v3+x[n];

v3_new=v1+v2+v3;
v1=v1_new
v2=v2_new
v3=v3_new
end loop
a) Determine a Block Diagram realization;
b) Determine the state space equations;
c) Determine the transfer function and the difference equation;
d) Is this system stable?
Solution
a) First notice that there are three state variables. The labels v1, v2, v3
refer to the current values at index n, while v1_new, v2_new, v3_new refer to
the values for the next iteration, ie at index n+1. Therefore we have the
following state space equations:
v1 @n + 1D = v1 @nD + v2 @nD - 0.5 v3 @nD

v2 @n + 1D = v1 @nD - v2 @nD - v3 @nD + x@nD
v3 @n + 1D = v1 @nD + v2 @nD + v3 @nD
y@nD = v1 @nD + x@nD
Therefore the state space matrices
ij 1 1 -0.5 yz
j z
A = jjjj 1 -1 -1 zzzz
j z
k1 1 1 {
T
B = H0, 1, 0L
C = H1, 0, 0L
D=1
This yields a transfer function
3 2
HHzL = z
-z +0.5 z+1.5
3 2

z -z -0.5 z+3
which is not stable, since there is at least one pole outside the unit circle
(all of them as a matter of fact!).
The difference equation is
y@nD = y@n - 1D + 0.5 y@n - 2D - 3 y@n - 3D + x@nD - x@n - 1D + 0.5 x@n - 2D + 1.5 x@n - 3D
à Problem 5.7
Problem
Repeat Problem 5.6 for the following program

begin loop
read input x@nD
y[n]=v1+x[n]; http://ebook29.blogspot.com
output y[n]
10
begin loop
read input x@nD
y[n]=v1+x[n];
output y[n]
v1=v1+v2- 0.5v3;
v2=v1- v2- v3+x[n];

v3=v1+v2+v3;
end loop
How is this different from the program in Problem 5.6?
Solution
This problem is slightly different from the previous, but it gives a totally
different answer. Since we do not assign different variables to the "new"
values, things get a bit mixed up. In this case the state space equations
become
v1 @n + 1D = v1 @nD + v2 @nD - 0.5 v3 @nD

v2 @n + 1D = v1 @n + 1D - v2 @nD - v3 @nD + x@nD
v3 @n + 1D = v1 @n + 1D + v2 @n + 1D + v3 @nD
After simple substitutions we obtain
v1 @n + 1D = v1 @nD + v2 @nD - 0.5 v3 @nD
v2 @n + 1D = v1 @nD - 1.5 v3 @nD + x@nD
v3 @n + 1D = 2 v1 @nD + v2 @nD - 2 v3 @nD
3 2
HHzL = z
+z +0.5 z+1.5
3 2

z +z -0.5 z
à Problem 5.8
Problem

begin loop
read input x@nD
v1_new=v1+v2- 0.5v3;http://ebook29.blogspot.com
v2_new=v1- v2- v3+x[n];
begin loop
read input x@nD
v1_new=v1+v2- 0.5v3;
v2_new=v1- v2- v3+x[n];
v3_new=v1+v2+v3;
v1=v1_new
v2=v2_new
v3=v3_new
y[n]=v1+x[n];
output
y[n]
end loop
How is this different from the program in Problem 5.6?
Solution
a) First notice that there are three state variables. The labels v1, v2, v3
refer to the current values at index n, while v1_new, v2_new, v3_new refer to
the values for the next iteration, ie at index n+1. Therefore we have the
following state space equations:
v1 @n + 1D = v1 @nD + v2 @nD - 0.5 v3 @nD

v2 @n + 1D = v1 @nD - v2 @nD - v3 @nD + x@nD
v3 @n + 1D = v1 @nD + v2 @nD + v3 @nD
The output equation becomes
y@nD = v1 @n + 1D + x@nD = v1 @nD + v2 @nD - 0.5 v3 @nD + x@nD
where we substituted for v1 @n + 1D.
Therefore the matrices for the state space equations become
ij 1 1 -0.5 yz
j z
j z
k1 1 1 {
T
B = H0, 1, 0L
C = H1, 1, -0.5L
D=1
12
ij 1 1 -0.5 yz
j z
j z
k1 1 1 {
T
B = H0, 1, 0L
C = H1, 1, -0.5L
D=1
z -2 z+33
HHzL =
3 2

z -z -0.5 z+3
which again is not stable.
The difference equation is
y@nD = y@n - 1D + 0.5 y@n - 2D - 3 y@n - 3D + x@nD - 2 x@n - 2D + 3 x@n - 3D
à Problem 5.9
Problem

begin loop
read input x@nD
y[n]=v1+v2+3x[n];
v1_new=v1+v2;
v2_new=v1- v2+x[n];
v1=v1_new
v2=v2_new
end loop
Solution
There are two state variables v1 @nD and v2 @nD. From the program we can write
v1 @n + 1D = v1 @nD + v2 @nD
v2 @n + 1D = v1 @nD - v2 @nD + x@nD
y@nD = v1 @nD + v2 @nD + 3 x@nD
v1 @n + 1D = v1 @nD + v2 @nD
v2 @n + 1D = v1 @nD - v2 @nD + x@nD
y@nD = v1 @nD + v2 @nD + 3 x@nD
The state space equations have matrices
i 1 1 yz
A = jj z
k 1 -1 {
T
B = H0, 1L
C = H1, 1L
D=3
Therefore the transfer function is given by
2
HHzL = 3
z +z-6
2

z -2
and the difference equation

y@nD = 2 y@n - 2D + 3 x@nD + x@n - 1D - 6 x@n - 2D
Lattice Implementation
à Problem 5.11
Problem
Given the transfer function

2
A HzL = I1 - 0.8 z-1 M

a) Determine its reverse A HzL. Also compute its zeros and poles. Compare them to the zeros and
poles of
A HzL;
b) Determine the reflection coefficients of A HzL, and the lattice realization.
Solution
a) By expansion we determine A HzL = 1 - 1.6 z-1 + 0.64 z-2 and therefore

A HzL = 0.64 - 1.6 z-1 + z-2
The transfer function A HzL has two zeros at z = 0.8 and two poles at z = 0, while the transfer

function A HzL has two zeros at z = 1 0.8 = 1.25 and two poles at z = 0;
14
The transfer function A HzL has two zeros at z = 0.8 and two poles at z = 0, while the transfer

function A HzL has two zeros at z = 1 0.8 = 1.25 and two poles at z = 0;
b) The reflection coefficients are determined by the recursive formula
Am HzL 1 1 -Km Am+1 HzL

B F =
2 B FB F
z-1 Am HzL 1-Km -Km 1 Am+1 HzL
Km = a
m+1 @m+1D
a @0D

m+1
Therefore
K1 = 0.64
1
A1 HzL =
2 IA HzL - 0.64 A HzLM = 1 - .97561 z-1
I1-0.64 M
K0 = -0.97561
A0 HzL = 1
à Problem 5.12
Problem
Repeat problem 5.11 for the following transfer function

A HzL = 1 - 1.8856 z-1 + 0.7728 z-2 + 0.8610 z-3
-1.1221 z-4 + 0.5398 z-5 - 0.1296 z-6
Solution
Applying the same reasoning as in Problem 5.11 we start with

A6 HzL = 1 - 1.8856 z-1 + 0.7728 z-2 + 0.8610 z-3 - 1.1221 z-4 + 0.5398 z-5 - 0.1296 z-6

A6 HzL = -0.1296 + 0.5398 z-1 - 1.1221 z-2 + 0.8610 z-3 + 0.7728 z-4 - 1.8856 z-5 + z-6
Therefore
a @6D
6
K5 = a
@0D
= -0.1296
6
is the first reflection coefficient. This yields

1
A5 HzL =
2
IA6 HzL - K5 A6 HzLM = 1 - 1.8467 z-1 + 0.6381 z-2 + 0.9892 z-3 - 1.0394 z-4 + 0.3005 z-5
1-K5
Continuing the same arguments we obtain

K4 = 0.3005
A4 HzL = 1 - 1.6866 z-1 + 0.3747 z-2 + 0.8766 z-3 - 0.5326 z-4

K3 = -0.5326
A3 HzL = 1 - 1.7028 z-1 + 0.8017 z-2 - 0.0303 z-3

K2 = -0.0303 http://ebook29.blogspot.com
-1 -2
K4 = 0.3005
A4 HzL = 1 - 1.6866 z-1 + 0.3747 z-2 + 0.8766 z-3 - 0.5326 z-4

K3 = -0.5326
A3 HzL = 1 - 1.7028 z-1 + 0.8017 z-2 - 0.0303 z-3

K2 = -0.0303
A2 HzL = 1 - 1.6800 z-1 + 0.7508 z-2

K1 = 0.7508
A1 HzL = 1 - 0.9596 z-1

K0 = -0.9596
A0 HzL = 1
à Problem 5.13
Problem
1
Given the transfer function H HzL = 1-0.81

z-2
determine a lattice realization.
Solution
From the denominator A HzL = 1 - 0.81 z-2 we obtain

K1 = -0.81
A1 HzL = 1
K0 =0
A0 HzL = 1
Therefore the IIR lattice implementation becomes as shown below.
16
à Problem 5.14
Problem
Repeat Problem 5.13 for the Transfer Function

1
H HzL = 1-1.8

z-1 +1.62 z-2 -0.729 z-3

Solution
From the denominator A HzL = 1 - 1.8 z-1 + 1.62 z-2 - 0.729 z-3 we obtain
K2 = -0.729
A2 HzL = 1 - 1.32111 z-1 + 0.656908 z-2
K1 = 0.656908
A1 HzL = 1 - 0.797337 z-1
K0 = -0.797337
A0 HzL = 1
The implementation is shown below.
à Problem 5.15
Problem

1
H HzL =
-1
3
I1-0.98 z M
Solution
Starting from the polynomial

3
A3 HzL = I1 - 0.98 z-1 M = 1 - 2.9400 z-1 + 2.8812 z-2 - 0.941192 z-3
3
A3 HzL = I1 - 0.98 z-1 M = 1 - 2.9400 z-1 + 2.8812 z-2 - 0.941192 z-3
we obtain
K2 = -0.941192
A2 HzL = 1 - 1.99932 z-1 + 0.999456 z-2
K1 = 0.999456
A1 HzL = 1 - 0.999932 z-1
K0 = -0.999932
Notice that the filter is very close to instability, and the above
computations require double precision arithmetic. The filter can be kept
stable even with round off errors, by making sure that the reflection
coefficients in magnitude do not exceed one.
à Problem 5.16
Problem
Determine the Lattice Realization of the Transfer Function of the system in

Problem 5.4.
Solution
The transfer function is

2
Hz+1L
2 z-1 I1+z-1 M
HHzL =
3
= 3
Hz-0.95L H1-0.95 zL
First we need to determine the Lattice expansion of the denominator. We

obtain:
A3 HzL = Hz - 0.95L3 = 1 - 2.85 z-1 + 2.7075 z-2 - 0.857375 z-3

This yields
K3 = -0.857375
A2 HzL = 1 - 1.99562 z-1 + 0.996501 z-2
K2 = 0.996501
A1 HzL = 1 - 0.999561 z-1
18
K2 = 0.996501
A1 HzL = 1 - 0.999561 z-1
K0 = -0.999561
Then we need to determine the coefficients c0 , c1 , c2 , c3 to solve the polynomial equation

2
z-1 I1 + z-1 M = c0 + c1 A1 HzL + c2 A2 HzL + c3 A3 HzL
This yields
c0 = 3.99213, c1 = 7.97128, c2 = 4.85, c3 = 1
The Lattice implementation is shown below.
à Problem 5.17
Problem
Given the transfer function

z +1 2
H HzL =
2
Hz-0.9L Hz+0.9L
a) Determine a lattice realization;
b) Determine the state space equations that implement the lattice

realization.
Solution
a) Again we write the transfer function as

z +z -1 -3
H HzL =
2
I1-0.9 z-1 M I1+0.9 z-1 M
Then we expand the denominator in terms of reflection coefficients:
A3 HzL = 1 - 0.9 z-1 - 0.81 z-2 + 0.729 z-3

K2 = 0.729
A2 HzL = 1 - 0.660557 z-1 - 0.328454 z-2
K1 = -0.328454
A1 HzL = 1 - 0.983636 z-1
K0 = -0.983636

z-1 + z-3 = c0 + c1 A1 HzL + c2 A2 HzL + c3 A3 HzL
This yields
c0 = 1.93176, c1 = 2.4045, c2 = 0.9, c3 = 1
The realization is shown below.
b) For the state space equations, define the state vector as

v@nD = Hv1 @nD, v2 @nD, v3 @nDLT
Then we obtain the following equations:
v1 @n + 1D = -K0 v1 @nD - K1 v2 @nD - K2 v3 @nD + x@nD
v2 @n + 1D = v1 @nD + K0 H-K0 v1 @nD - K1 v2 @nD - K2 v3 @nD + x@nDL
v3 @n + 1D = v2 @nD + K1 H-K2 v3 @nD + x@nDL
y@nD = c0 H-K0 v1 @nD - K1 v2 @nD - K2 v3 @nD + x@nDL +
+ c1 Hv1 @nD + K0 http://ebook29.blogspot.com
H-K0 v1 @nD - K1 v2 @nD - K2 v3 @nD + x@nDLL
+ c2 Hv2 @nD + K1 H-K2 v3 @nD + x@nDLL
20
v1 @n + 1D = -K0 v1 @nD - K1 v2 @nD - K2 v3 @nD + x@nD

v2 @n + 1D = v1 @nD + K0 H-K0 v1 @nD - K1 v2 @nD - K2 v3 @nD + x@nDL
v3 @n + 1D = v2 @nD + K1 H-K2 v3 @nD + x@nDL
y@nD = c0 H-K0 v1 @nD - K1 v2 @nD - K2 v3 @nD + x@nDL +
+ c1 Hv1 @nD + K0 H-K0 v1 @nD - K1 v2 @nD - K2 v3 @nD + x@nDLL
+ c2 Hv2 @nD + K1 H-K2 v3 @nD + x@nDLL
from which we can derive the usual state space equations.

However to implement this class of filters it might turn useful to use the
recursive structure shown below, and write the intermediate signals sm @nD and wm @nD.
In this case we obtain the input-output equations of all sections:
w2 @nD = -K2 v2 @nD + x@nD

s3 @nD = v2 @nD + K2 w2 @nD
w1 @nD = -K1 v1 @nD + w2 @nD
s2 @nD = v1 @nD + K1 w1 @nD
w0 @nD = -K0 v0 @nD + w1 @nD
s1 @nD = v0 @nD + K0 w0 @nD
s0 @nD = w0 @nD
then we compute the output:
y@nD = c0 s0 @nD + c1 s1 @nD + c2 s2 @nD + c3 s3 @nD
and finally we update the states
v2 @n + 1D = s2 @nD
v1 @n + 1D = s1 @nD
v0 @m + 1D = s0 @nD
à Problem 5.18
Problem 5.18
Problem
Repeat Problem 5.16 for the transfer function H HzL = Hz - 1L Hz + 0.9L2 .
Solution
Again we write the transfer function as

z -z -1 -2
H HzL =
-1
2
I1+0.9 z M
Then we expand the denominator in terms of reflection coefficients:
A2 HzL = 1 + 1.8 z-1 + 0.81 z-2

K1 = 0.81
A1 HzL = 1 + 0.994475 z-1
K0 = 0.994475

z-1 - z-2 = c0 + c1 A1 HzL + c2 A2 HzL
This yields
c0 = -1.97453, c1 = 2.8, c2 = -1.
à Problem 5.19
Problem
Repeat Problem 5.16 for the transfer function H HzL = 1 Iz4 - 0.4096M.
Solution
Again we write the transfer function as

z -4
H HzL = 1-0.4096

z-4
We then expand the denominator:

A4 HzL = 1 - 0.4096 z-4
à K3 = -0.4096
22
A4 HzL = 1 - 0.4096 z-4

K3 = -0.4096
At this point we obtain
A3 HzL = A1 HzL = 1
K2 = K1 = K0 = 0
Finally for the rest of the coefficients to solve the polynomial equation

z-4 = c0 + c1 A1 HzL + c2 A2 HzL + c3 A3 HzL + c4 A4 HzL
we obtain
c0 = 0.4096, c1 = 0, c2 = 0, c3 = 0, c4 = 1

Multirate Digital Signal Processing:
Fundamentals
Sampling, Upsampling and Downsampling
à Problem 6.1
Solution
From the definiton of downsampling,

y@nD = x@2 nD
a) y@nD = ∆@2 nD = ∆@nD
b) y@nD = ∆@2 n - 1D = 0
c) y@nD = H-1L2 n u@2 nD = u@nD
d) y@nD = ej0 .2 Πn
e) y@nD = ej0 .2 Πn u@2 nD = y@nD = ej0 .2 Πn u@nD
f) y@nD = 2 cos H0.4 ΠnL
g) y@nD = 2 cos H0.5 Π 2 nL = 2 cos HΠnL = H-1Ln
h) y@nD = 2 sin HΠnL = 0
i) y@nD = cos H2 ΠnL = 1
j) y@nD = 2 sin H2 ΠnL = 0
à Problem 6.2
Solution
Recall that Y HΩL = 12 IX I Ω2 M + X I Ω2 - ΠMM
a) X HΩL = 1 therefore Y HΩL = 12 H1 + 1L = 1, ie y@nD = ∆@nD
b) X HΩL = e-jΩ therefore Y HzL = 12 Ie-jΩ2 - e-jΩ2 M = 0
2
b) X HΩL = e-jΩ therefore Y HzL = 12 Ie-jΩ2 - e-jΩ2 M = 0

e jΩ 1 e jΩ2
-e 1 e -ejΩ2 +e +e jΩ j Ω2 e jΩ jΩ2 jΩ
c) X HΩL = e
jΩ +1 then Y HΩL =
2
I e
jΩ2 +1 + -e

jΩ2 +1 M =
2

ejΩ -1 jΩ -1 and there-
= e
fore y@nD = ∆@nD
d) X HΩL = 2 Π∆ HΩ - 0.1 ΠL for -Π < Ω £ Π, then
Y HΩL = 12 2 Π∆ I Ω2 - 0.1 ΠM + 12 2 Π∆ I Ω2 - 0.1 Π - ΠM for -Π < Ω2 £ +Π, that is to say for
-2 Π < Ω £ 2 Π. Therefore
Y HΩL = 2 Π ∆ HΩ - 0.2 ΠL + 2 Π ∆ HΩ - 0.2 Π - 2 ΠL
and therefore Y HΩL = 2 Π∆ HΩ - 0.2 ΠL for -Π < Ω £ +Π and y@nD = e-j0 .2 Πn .
e jΩ
e) X HΩL = e
jΩ -e-j0 .1 Π therefore
jΩ2 jΩ2
Y HΩL = 12 I e
e
jΩ2 -e-j0 .1 Π + -e
-e

jΩ2 -e-j0 .1 Π M
jΩ jΩ2 j0 .1 Π jΩ jΩ2 j0 .1 Π
= 12 e
+e e +e -e
ejΩ -ej0 .2 Π
e
jΩ
= 12 e
e
jΩ -ej0 .2 Π
which yields y@nD = ej0 .2 Πn u@nD

f) X HΩL = Π H∆ HΩ - 0.2 ΠL + ∆ HΩ + 0.2 ΠLL for -Π < Ω £ +Π. Then
Y HΩL = 12 Π I∆ I Ω2 - 0.2 ΠM + ∆ I Ω2 + 0.2 ΠM + ∆ I Ω2 - 0.2 Π - ΠM + ∆ I Ω2 + 0.2 Π - ΠMM
= Π H∆ HΩ - 0.4 ΠL + ∆ HΩ + 0.4 ΠL + ∆ HΩ - 0.4 Π - 2 ΠL + ∆ HΩ + 0.4 Π - 2 ΠLL
for -2 Π < Ω £ +2 Π. Therefore
Y HΩL = Π H∆ HΩ - 0.4 ΠL + ∆ HΩ + 0.4 ΠLL for -Π < Ω £ Π
and then y@nD = cos H0.4 ΠnL.
which yields y@nD = ej0 .2 Πn u@nD

g) X HΩL = Π H∆ HΩ - 0.5 ΠL + ∆ HΩ + 0.5 ΠLL for -Π < Ω £ +Π. Then
Y HΩL = 12 Π I∆ I Ω2 - 0.5 ΠM + ∆ I Ω2 + 0.5 ΠM + ∆ I Ω2 - 0.5 Π - ΠM + ∆ I Ω2 + 0.5 Π - ΠMM
= Π H∆ HΩ - ΠL + ∆ HΩ + ΠL + ∆ HΩ - Π - 2 ΠL + ∆ HΩ + Π - 2 ΠLL
Y HΩL = Π H∆ HΩ - ΠL + ∆ HΩ + ΠLL for -Π < Ω £ Π
and then y@nD = cos HΠnL.
h) X HΩL = -jΠ H∆ HΩ - 0.5 ΠL - ∆ HΩ + 0.5 ΠLL for -Π < Ω £ +Π. Then

Y HΩL =
-j 12 Π I∆ I Ω2 - 0.5 ΠM - ∆ I Ω2 + 0.5 ΠM + ∆ I Ω2 - 0.5 Π - ΠM - ∆ I Ω2 + 0.5 Π - ΠMM
= -jΠ H∆ HΩ - ΠL - ∆ HΩ + ΠL + ∆ HΩ - Π - 2 ΠL - ∆ HΩ + Π - 2 ΠLL
Y HΩL =
-j 12 Π I∆ I Ω2 - 0.5 ΠM - ∆ I Ω2 + 0.5 ΠM + ∆ I Ω2 - 0.5 Π - ΠM - ∆ I Ω2 + 0.5 Π - ΠMM
= -jΠ H∆ HΩ - ΠL - ∆ HΩ + ΠL + ∆ HΩ - Π - 2 ΠL - ∆ HΩ + Π - 2 ΠLL
Y HΩL = -jΠ H∆ HΩ - ΠL - ∆ HΩ + ΠLL = -jΠ H∆ HΩ - ΠL - ∆ HΩ + Π - 2 ΠL = 0 for
-Π < Ω £ Π
and then y@nD = 0.
i) X HΩL = Π H∆ HΩ - ΠL + ∆ HΩ + ΠLL for -Π < Ω £ +Π. Then
Y HΩL = 12 Π I∆ I Ω2 - ΠM + ∆ I Ω2 + ΠM + ∆ I Ω2 - Π - ΠM + ∆ I Ω2 + Π - ΠMM

= Π H∆ HΩ - 2 ΠL + ∆ HΩ + 2 ΠL + ∆ HΩ - 2 Π - 2 ΠL + ∆ HΩ + 2 Π - 2 ΠLL
Y HΩL = Π H∆ HΩ - 2 ΠL + ∆ HΩ + 2 ΠLL = 2 Π∆ HΩL for -Π < Ω £ Π
and then y@nD = 1.
j) X HΩL = -jΠ H∆ HΩ - ΠL - ∆ HΩ + ΠLL for -Π < Ω £ +Π. Then
Y HΩL = -j 12 Π I∆ I Ω2 - ΠM - ∆ I Ω2 + ΠM + ∆ I Ω2 - Π - ΠM - ∆ I Ω2 + Π - ΠMM

= -jΠ H∆ HΩ - 2 ΠL - ∆ HΩ + 2 ΠL + ∆ HΩ - 2 Π - 2 ΠL - ∆ HΩ + 2 Π - 2 ΠLL
Y HΩL = -jΠ H∆ HΩ - 2 ΠL - ∆ HΩ + 2 ΠLL = -jΠ H∆ HΩ - 2 ΠL - ∆ HΩ - 2 ΠL = 0 for
-Π < Ω £ Π
and then y@nD = 0.
à Problem 6.3
Solution
In all cases
Y HΩL = 12 X I Ω2 M + 12 X I Ω2 - ΠM for all Ω
4
Y HΩL = 12 X I Ω2 M + 12 X I Ω2 - ΠM for all Ω
Whene there is no aliasing, ie X HΩL = 0 for Π2 £ É Ω É £ Π then this relation simpilfies to
Y HΩL = 12 X I Ω2 M
a) B = Π5 < Π2 . Then there is no aliasing after downsampling and therefore Y HΩL is as shown below
b) B = Π2 . Then again there is no aliasing after downsampling and therefore Y HΩL is as shown below
c) B = 3
4
Π
. In this cases there is aliasing and we have to account for it. Best way
to do it is proceed in two steps: sampling and then downsampling.
The sampling operation yields

Y HΩL = 12 X HΩL + 12 X HΩ - ΠL
The figure below shows both 12 X HΩL and 12 X HΩ - ΠL.

Then downsampling yields Y HΩL = Y I Ω2 M, just a rescaling of the frequency axis. The final results is
shown
below.
d) B = Π. Same reasoning as in c). This time it is easy to see that

Y HΩL = 12 X HΩL + 12 X HΩ - ΠL = 12 for all Ω

and therefore Y HΩL = Y I Ω2 M = 12 for all Ω.
6

and therefore Y HΩL = Y I Ω2 M = 12 for all Ω.
à Problem 6.4
Solution
Recall that y@nD = xA n2 E ∆2 @nD = 12 xA n2 E I1 + H-1Ln M = 12 xA n2 E I1 + e-jΠn M
a) y@nD = ∆@nD;
b) y@nD = 12 IH-1Ln + 1M u@nD
c) y@nD = 12 ej0 .05 Πn + 12 ej H0.05 Π-ΠL n
d) y@nD = I 12 ej0 .05 Πn + 12 ej H0.05 Π-ΠL n M u@nD
e) y@nD = I 14 ej0 .05 Πn + 14 ej H0.05 Π-ΠL n M u@nD + I 14 e-j0 .05 Πn + 14 e-j H0.05 Π-ΠL n M u@nD
which becomes
y@nD = 12 cos H0.05 ΠnL u@nD + 2
1
cos H0.95 ΠnL u@nD
f) similarly y@nD = 12 cos H0.05 ΠnL + 2

1
cos H0.95 ΠnL
à Problem 6.5
Solution
Using the DTFT. For upsampling

S HΩL = X H2 ΩL
and downsampling
Y HΩL = 12 S I Ω2 M + 12 S I Ω2 - ΠM

Substitute for S HΩL to obtain
S I Ω2 M = X I2 Ω2 M = X HΩL
S I Ω2 - ΠM = X I2 I Ω2 - ΠMM = X HΩ - 2 ΠL
Therefore
Y HΩL = 12 X HΩL + 12 X HΩ - 2 ΠL = X HΩL
Y HΩL = 12 X HΩL + 12 X HΩ - 2 ΠL = X HΩL

from the periodicity of the DTFT.
à Problem 6.6
Solution
From the diagram it is easy to verify that

x@nD if n even
y@nD = :
0 if n odd
Therefore y@nD = x@nD ∆2 @nD, and Y HzL = 12 HX HzL + X H-zLL, or equivalently,
Y HΩL = 12 HX HΩL + X HΩ - ΠLL.
One way we can verify this is the following: call v@nD the output of the downsampler. Then
V HΩL = 12 IX I Ω2 M + X I Ω2 - ΠMM
Since y@nD is the output of the upsampler then Y HΩL V H2 ΩL = 12 HX HΩL + X HΩ - ΠLL as we
expect.
à Problem 6.7
Solution
The effect of modulation on the frequency spectrum is as follows
DTFT 8xM @nD< = 12 X HΩ - Ω0 L + 12 X HΩ + Ω0 L

where xM @nD = x@nD cos HΩ0 nL.
After upsampling by 2 the signal y@nD has DTFT
Y HΩL = XM H2 ΩL = 12 X H2 Ω - Ω0 L + 12 X H2 Ω + Ω0 L
8
Y HΩL = XM H2 ΩL = 12 X H2 Ω - Ω0 L + 12 X H2 Ω + Ω0 L
a) Ω0 = Π4 . Then XM HΩL and Y HΩL are shown below.
b) Ω0 = Π2 . Then XM HΩL and Y HΩL are shown below.
c) Ω0 = Π. Then XM HΩL and Y HΩL are shown below.
In this case notice the maximum amplitude of the DTFT being "one"
(rather then 1/2 as in the previous cases). This is due to the fact that X HΩ - ΠL = X HΩ + ΠL.
à Problem 6.8
Solution
In this problem we need to increase the sampling frequency from Fs = 8 kHz to Fs = 12 kHz, ie by a
factor 12

8
= 32 . Therefore with ideal filters the scheme is as shown below.
10
à Problem 6.9
Solution
In this case the Low Pass Filter H HΩL is a non ideal FIR filter. The whole problem is to choose the
correct
specifications for the filter.
The stopband has to be ΩS = Π3 , since the purpose of this filter is to stop the frequency artifacts
generated by the upsampling operation. The passband has to be decided on the

basis of the bandwirdth of the signal we want to pass and the desired
complexity of the filter. For a window based, recall that we need a hamming
window (from the desired attenuation) with transition region DΩ = 8 Π N. Therefore an FIR filter of
length N will have a passband Ωp = Π3 - 8
N
Π
.
à Problem 6.10
Solution
The digital signal has frequencies at Ω1 = ± 2 Π 6 = ± Π 3 rad and Ω2 = 2 Π 2 6 = ± 2 Π 3

rad. Therefore the output signal has frequencies at ± Π 3, ± 2 Π 3 and also at
Π 3 - Π = -2 Π 3, -Π 3 + Π = 2 Π 3, 2 Π 3 - Π = -Π 3 and -2 Π 3 + Π = Π 3. All
components at ± Π 3 and ± 2 Π 3 are going to sum with each other.
à Problem 6.11
Solution
a) H HzL = I2 - 4 z-3 - 6 z-6 M + z-1 I-3 - 5 z-3 + 2 z-6 M + z-2 I2 + 2 z-3 M
b) H HzL = I2 + 2 z-2 - 5 z-4 - 6 z-6 M + z-1 I-3 - 4 z-2 + 2 z-4 + 2 z-6 M
c) H HzL = I2 z2 + 1 + 2 z-2 - 5 z-4 - 6 z-6 M + z-1 I-z2 - 3 - 4 z-2 + 2 z-4 + 2 z-6 M
d) H HzL = I-z3 - 4 z-3 - 6 z-6 M + z-1 Iz3 + 2 - 5 z-3 + 2 z-6 M + z-2 I2 z6 - 3 z3 + 2 z-3 M
+¥ +¥ +¥
e) H HzL = S 0.5n z-n = S 0.52 n z-2 n + z-1 S 0.52 n+1 z-2 n whixh yields
n=0 n=0 n=0
1 0.5
H HzL = 1-0.25

z-2
+ z-1 1-0.25

z-2
z
f) H HzL = z-0.8
+¥ +¥ +¥
H HzL = S 0.8n z-n = S 0.82 n z-2 n + z-1 S 0.82 n+1 z-2 n . This becomes
n=0 n=0 n=0
1 0.8
H HzL = 1-0.64

z-2
+ z-1 1-0.64

z-2
1 0.8
H HzL = 1-0.64

z-2
+ z-1 1-0.64

z-2
The same result can be obtained by an alternative way:

z z+0.8 z 2 0.8 z 2
H HzL = z-0.8 -1
2 -0.64 + z
= z
z+0.8 z2 -0.64
amd you can verify that the two answers are the same.
g) You can verify that the general exprezzion for the polyphase terms is
+¥
Hk HzL = S h@nM - kD z-n
n=-¥
for k = 0, ..., M - 1. Applying this formula we obtain

Π
+¥ sin I 5 H3 n-kLM
Hk HzL = S
Π z-n for k = 0, 1, 2
n=-¥ 5 H3 n-kL
à Problem 6.12
Solution
a) From the transfer function of the filter H HzL = 1 + z-1 + 2 z-2 - z-3 + z-4 - z-5 + z-6
and M = 4 we obtain the decomposition
H HzL = H0 Iz4 M + z-1 H1 Iz4 M + z-2 H2 Iz4 M + z-3 H3 Iz4 M
with
H0 HzL = 1 + z-1
H1 HzL = 1 - z-1
H2 HzL = 2 + z-1
H3 HzL = -1
The block diagram of the system is shown below:
12
b) Using the same filters, the realization is as follows:
c) When M = 2 the polyphase decomposition becomes H0 HzL = 1 + 2 z-1 + z-2 + z-3 and
H HzL = 1 - z-1 - z-2
Therefore the system becomes as shown below:
Now notice that the cascade upsampler - downsampler (both by 2) is just an

identity. Also the cascede upsampler - time delay - downsampler as shown
gives an output of zero, no matter what the input is (easy to verify). This
is shown below:
Therefore the overall system looks like this one:
14
Applications of MultiRate
à Problem 6.13
Solution
a) In digital frequency, the passband is ΩP = 2 Π 25 6000 = Π 120 radians, and the stopband
is ΩS = 2 Π 30 6000 = Π 100 radians. As a consequence the transition region is
DΩ = ΩS - ΩP = 0.0017 Π. For a 60 dB attenuation we can use (say) a Kaiser window with parame-
ters
Β = 0.1102 H60 - 8.7L = 5.6533

60-8
N = 2.285

DΩ
= 4, 261.1
and therefore the order is 4, 262. This yields a total number of
4, 262 6, 000 = 25.572 106 multiplications and additions per second
b) Since we want to reject all frequencies above 30Hz, we can downsample
from the orginal sampling frequency (6kHz) down to 60Hz, ie by a factor D = 6, 000 60 = 100.
As seen in class, this can be done in three stages by factoring D = 100 as (say)
100 = 10 5 2
as shown below
Now the specifications of the filters become as follows:

H1 : Fp = 25 Hz, FS = 600 - 30 = 570 Hz, DΩ = 2 Π H570 - 25L 6000 = 0.1817 Π
From the transition bands we can determine the orders of the filters.
using the Kaiser window again we have the orders N1 , N2 , N3 of the threee filters to be. respectively,
From the transition bands we can determine the orders of the filters.
using the Kaiser window again we have the orders N1 , N2 , N3 of the threee filters to be. respectively,
60-8
N1 = 2.285

DΩ
= 40
60-8
N2 = 2.285

DΩ
= 34
60-8
N3 = 2.285

DΩ
= 87
Finally the total number of operations per second becomes:
ops sec = 40 6000 + 34 * 600 + 87 * 120 = 270, 840 = 0.28 106
which is reduction of a factor of more than 90. Also notice that we are not
even attempting to save even more in computation using the polyphase
decomposition!
à Problem 6.14
Solution
Q1) Recall the Butterworth filter frequency response:

2 1 1
É H HWL É =
W 2N
=
W 2N
1+J W N 1+Ε2 J W N
c p
Then for the reconstruction filter, the passband is Fp = 22 kHz = 2 Π 22, 000, that is to say
Wp = 44, 000 Π rad sec. The stopband has to be at half the sampling frequency, ie
FS = 44.1 2 = 22.05 kHz, that si to say WS = 44, 100 Π rad sec.
From the passband ripple we determine the factor Ε from
1

1+Ε2
= H1 - 0.01L2 = 0.9801 ® Ε2 = 0.0203 and therefore Ε = 0.1425
We determine the order from the requirement
É H HWS L É 2 1
= 0.012 = 10-4
=
44.1 2 N
1+0.0203 I
44
M
Solving for N gives a very large number, and the analog filter cannot be implemented.
Furthermore there will be a distortion in phase from the reconstruction
filter itself.
Q2) Now the filter has still the same expression, but with N = 4, since we are restricted to a 4 pole
filer. Therefore the filter is given
by
16
Q2) Now the filter has still the same expression, but with N = 4, since we are restricted to a 4 pole
filer. Therefore the filter is given
by
2 1
É H HWL É =
W 8
1+0.0203 I 44,000
Π M
Since we want the attenuation to be 40dB in the stopband, we can solve for
the stopband, as
2
1
8 = 0.01

W
= 10-4
1+0.0203 I 44,000
Π M
which yields
4 18
W S 10 -1

44,000 Π
= J 0.0203
N = 5.1470
Therefore the stopband of this filter is at FS = 5.1470 22 kHz = 113.24 kHz. This has to coin-
cide with half the sampling frequency, and therefore the
new sampling frequency is Fs = 2 113.24 kHz = 226.5 kHz. In other words we have to upsam-
ple at least by 226.5/44.1=5.1 times
before the Digital to Analog conversion.

DFT Filter Banks
à Problem 7.1
Solution
a) The transfer function F HzL is determined as F HzL = H Ize-j0 .2 Π M. In fact you can verify that
F HΩL = F HzL È z=ejΩ = H IejΩ e-j0 .2 Π M = H HΩ - 0.2 ΠL. Substituting for the z-Transform we
obtain
F HzL = 2 + ej0 .2 Π z-1 - ej0 .4 Π z-2 + 0.5 ej0 .6 Π z-3
This yields an impulse response
f@nD = 2 ∆@nD + ej0 .2 Π ∆@n - 1D - ej0 .4 Π ∆@n - 2D + 0.5 ej0 .6 Π ∆@n - 3D
Notice that it is computed as f@nD = h@nD ej0 .2 Πn , with h@nD the impulse response of H HzL.
b) By the same argument the transfer function can be determined as
G HzL = H Ize-j0 .2 Π M + H Izej0 .2 Π M. Therefore the transfer function becomes
G HzL = 4 + 2 cos H0.2 ΠL z-1 - 2 cos H0.4 ΠL z-2 + cos H0.6 ΠL z-3
= 4 + 1.61803 z-1 - 0.6180 z-2 - 0.3090 z-3
and the impulse response
g@nD = 4 ∆@nD + 1.61803 ∆@n - 1D - 0.6180 ∆@n - 2D - 0.3090 ∆@n - 3D

It is computed as g@nD = Iej0 .2 Πn + e-j0 .2 Πn M h@nD = 2 cos H0.2 ΠnL h@nD. Therefore
g@nD = 2 Real 8f@nD<.
à Problem 7.2
Solution
The prototype filter is a low pass filter with bandwidth Π 2, with impulse response
Π
sin I nM
2
h0 @nD =
Πn
w@nD
with w@nD the non causal window sequence. Since we need 50dB attenuation in the
stopband we use a Blackman window, which has a transition region DΩ = 12 Π N. Therefore we
determine the filter length N from the transition band as
2
with w@nD the non causal window sequence. Since we need 50dB attenuation in the
stopband we use a Blackman window, which has a transition region DΩ = 12 Π N. Therefore we
determine the filter length N from the transition band as
12 Π

N
£ 0.1 Π
This yields N = 121. The Low Pass Filter H0 HzL has the polyphase decomposition
H0 HzL = E0 Iz2 M + z-1 E1 Iz2 M
where
E0 HzL = S h0 @2 nD z-n = w@0D = 1
n
since h0 @2 nD = 0 for n ¹ 0. Similarly

Π
sin IΠn+ M
E1 @nD = S h0 @2 n + 1D z-n = S
Π H2 n+1L
2
w@2 n + 1D z-n
n n
à Problem 7.3
Solution
The prototype filter has frequency response H HΩL with bandwidth ΠM = Π8 . Therefore the non causal
impulse response of the prototype filter is
given by
Π
sin I 8 nM
h@nD =
Πn
w@nD
with w@nD being a window of length N + 1 = 21. For example let w@nD be a hamming window,
which has the expression
wAn - N2 E = 0.54 - 0.46 cos I 2
N
Π
nM for 0 £ n £ N
where we listed the causal expression generally found in most tables. From
this expression it is easy to see that
w@nD = 0.54 + 0.46 cos I 2

20
Π
nM for -10 £ n £ 10
Finally the expression of the impulse response h@nD becomes
Π
sin I 8 nM y
h@nD = i
j z
j
Πn
Π
z I0.54 + 0.46 cos I 10
nMM for -10 £ n £ 10
k {
and zero otherwise.
The transfer function of the prototype filter is then given by
The transfer function of the prototype filter is then given by
H HzL = -0.0018 z10 - 0.0014 z9 + 0.0000 z8 +

0.0047 z7 + 0.0149 z6 + 0.0318 z5 + 0.0543 z4 +
0.0794 z3 + 0.1027 z2 + 0.1191 z + 0.1250 +
0.1191 z-1 + 0.1027 z-2 + 0.0794 z-3 +
0.0543 z-4 + 0.0318 z-5 + 0.0149 z-6 + 0.0047 z-7 +
0.0000 z-8 - 0.0014 z-9 - 0.0018 z-10
The eight polyphase components of the prototype filter then become as

follows:
+¥
E-k Iz8 M = S h@8 n - kD z-8 n , for k = 0, 1, ..., 7
n=-¥
which yields
E0 Iz8 M = h@-8D z8 + h@0D z0 + h@8D z-8 = 0.1250
E-1 Iz8 M = h@-9D z8 + h@-1D z0 + h@7D z-8 = -0.0014 z8 + 0.1191 + 0.0047 z-8
E-2 Iz8 M = h@-10D z8 + h@-2D z0 + h@6D z-8 = -0.0018 z8 + 0.1027 + 0.0149 z-8
E-3 Iz8 M = h@-3D z0 + h@5D z-8 = 0.0794 + 0.0318 z-8
E-4 Iz8 M = h@-4D z0 + h@4D z-8 = 0.0543 + 0.0543 z-8
E-5 Iz8 M = h@-5D z0 + h@3D z-8 = 0.0318 + 0.0794 z-8
E-6 Iz8 M = h@-6D z0 + h@2D z-8 + h@10D z-16 = 0.0149 + 0.1027 z-8 - 0.0018 z-16
E-7 Iz8 M = h@-7D z0 + h@1D z-8 + h@9D z-16 = 0.0047 + 0.1191 z-8 - 0.0014 z-16
The implementation is shown below.
4
à Problem 7.4
Solution
a) , b), c) H HzL is M -Band, since h@4 nD = 0 for n ¹ 0;

d) H HzL is not M -Band since S H IΩ - k Π2 M is not a constant for all Ω, as shown below.
k
e) H HzL is M - Band since S H IΩ - k Π2 M is a constant as shown below.

k
à Problem 7.5
Solution
Let Ω1 = Π5 - DΩ and Ω2 = Π5 + DΩ for any 0 £ DΩ £ Π5 . Then H HzL is an M -Band filter with
A = 15 as shown below.
à Problem 7.6
Solution
With M = 16 the prototype filters for both Analysis and Synthesis have a bandwidth
Ωc = Π 16. From what we have seen about maximally decimated DFT Filter banks, if
we want to use FIR filters, the only possibility for perfect reconstruction
is that both filters h@nD and g@nD in the analysis and synthesis network have length M = 16. In this way
we would have
HHzL = h@0D + h@-1D z + ... + h@-15D z15
GHzL = g@0D + g@1D z-1 + ... + g@15D z-15
6
HHzL = h@0D + h@-1D z + ... + h@-15D z15

GHzL = g@0D + g@1D z-1 + ... + g@15D z-15
with the Perfect Reconstruction condition
1
h@nD g@nD = 16

What makes this problem a bit different from the standard FIR window based
design problem is the fact the filter order is odd, ie the total filter
length is 16, which is even. In Chapter 4 we have considered only the case
where the total filter length is odd as N = 2 L + 1. Although most of the time this is not a major restric-
tion, in this case
we have to design a filter with the precise length, and none of the filter
coefficients can be zero. In other words we cant use (say) a filter with
length 14, and assume h@-15D = 0, since this would require g@15D = ¥, clearly not feasible.
In order to design a filter with even length, we can call Hd HΩL the frequency response of an ideal Low
Pass Filter with bandwidth Ωc , and compute
Ω
+Ωc Ω
1
hd @nD = IDTFT :HHΩL e- j 2 > = 2

Π Ù
e- j 2 e jΩn dΩ
-Ωc
This leads to the impulse response

1
sinJΩc Jn- NN
2
hd @nD =
1
Π Jn- N
2
Now the goal is to find a linear phase approximation with a finite number of
coefficients. In particular let
` L
HL HΩL = S hd @nD e-jΩn
n=-L+1

` L-1 L
HL HΩL = S hd @-nD e jΩn + S hd @nD e-jΩn
n=0 n=1
It is easy to see that hd @-nD = hd @n + 1D and therefore we can write

` L L
HL HΩL = e-jΩ S hd @nD e jΩn + S hd @nD e-jΩn
n=1 n=1
Ω `
This shows that e j 2 H L HΩL is real, since
Ω ` Ω L Ω L
e j 2 HL HΩL = e- j 2 S hd @nD e jΩn + e- j 2 S hd @nD e-jΩn
n=1 n=1
Ω ` Ω L Ω L
e j 2 HL HΩL = e- j 2 S hd @nD e jΩn + e- j 2 S hd @nD e-jΩn
n=1 n=1
`
and therefore H L HΩL has linear phase. As a consequence a causal translation has linear phase
too, which leads to the linear phase FIR filter with frequency response
` 2 L-1
HL HΩL e-jΩHL-1L = S hd @n - L + 1D e-jΩn
n=0
Π
In our case, the bandwidth is Ωc = 16 , the filter order is 15 = 2 L - 1, which yields L = 8, and the FIR

filter becomes
Π 1
sinJ Jn-7- NN
16 2
hd @nD =
1
, for n = 0, ..., 15
Π Jn-7- N
2
without including the window. Finally the filters h@nD and g@nD of the analysis and synthesis networks
become
Π 1
sinJ Jn-7- NN
16 2
I0.54 - 0.46 cosI 2
h@-nD = hd @nD w@nD =
1 15
Π
nMM for 0 £ n £ N
Π Jn-7- N
2
and
16
g@nD = h@-nD , for 0 £ n £ N

In terms of the polyphase decomposition, every term is a constant, as
E-k HzL = h@-kD

Fk HzL = g@kD
for k = 0, ..., 15. It is just a matter of computing the coefficients to determine the final
result shown in the figure below for the analysis network.
8
à Problem 7.7
Solution
First we can verify that, in the the system below

v@nD if n even
w@nD = :
0 if n odd
n
which yields w@nD = v@nD ∆2 @nD = 12 Iv@nD + H-1L v@nDM. Therefore, as you recall,
WHΩL = 12 VHΩL + 21 VHΩ - ΠL

and therefore
WHzL = 12 VHzL + 12 VH-zL
WHzL = 12 VHzL + 12 VH-zL

Applying this result it is easy to see that
YHzL = GHzL I 12 HHzL XHzL + 12 HH-zL XH-zLM

= 12 GHzL HHzL XHzL + 12 GHzL HH-zL XH-zL
à Problem 7.8
Solution
In this case we have a filter bank with two filters. Therefore M = 2 and forperfect reconstruction the
filters have to be
HHzL = h@0D + h@-1D z
GHzL = g@0D + g@1D z-1
with the condition
h@0D g@0D = 12

h@-1D g@1D = 12
Then Let us see how to relate XHzL = Z 8x@nD< with YHzL = Z 8y@nD<. Applying the result from the previ-
ous problem we have
YHzL = GHzL I 12 HHzL XHzL + 12 HH-zL XH-zLM +
+ GH-zL I 12 HH-zL XHzL + 12 HHzL XH-zLM
which becomes
YHzL = 12 HGHzL HHzL + GH-zL HH-zLL XHzL + 12 HGHzL HH-zL + GH-zL HHzLL XH-zL
Now let’s see the two transfer functions XHzL ® YHzL and XH-zL ® YHzL with the perfect reconstruction
conditions above:
1

2
HG HzL H HzL + G H-zL H H-zLL =
1
IIg@0D + g@1D z-1 M Hh@0D + h@-1D zL + IIg@0D - g@1D z-1 M Hh@0D - h@-1D zLM =

2
=
1
I2 Hg@0D h@0D + g@1D h@-1DL + Hg@0D h@-1D - g@0D h@-1DL z + Hg@1D h@0D - g@1D h@0DL z-1 M
2
= 1 for all z
1
HG HzL H H-zL + G H-zL H HzLL =
2
1

2
IIg@0D + g@1D z-1 M Hh@0D - h@-1D zL + IIg@0D - g@1D z-1 M Hh@0D + h@-1D zLM =
=
1
I2 Hg@0D h@0D - g@1D h@-1DL + H-g@0D h@-1D + g@0D h@-1DL z + Hg@1D h@0D - g@1D h@0DL z-1 M
2
= 0 for all z
Therefore. as expected,
10
Therefore. as expected,
YHzL = XHzL
and the filter bank perfectly reconstructs the input signal.
à Problem 7.9
Solution
a) From the Problem 7.8, we can write YHzL in terms of the input signal XHzL and its alias XH-zL. The
aliasing comes from the downsampling operation.
In terms of the DTFT we can write
YHΩL = AHΩL XHΩL + BHΩL XHΩ - ΠL
where
AHΩL = 12 GHΩL HHΩL + 12 GHΩ - ΠL HHΩ - ΠL

BHΩL = 12 GHΩL HHΩ - ΠL + 12 GHΩ - ΠL HHΩL
In our case the two prototype filters have frequency response HHΩL = GHΩL as shown below.
Therefore:
2 2
I 10 M Ë Ω - 0.55 Π Ë

Π
2 +I 10 M Ë Ω - 0.45 Π Ë

Π
2 if 0.45 Π < È Ω È < 0.55 Π
A HΩL = :
1 otherwise
Analogously:
- É 10

Π
É 2 H È Ω È -0.55 ΠL H È Ω È -0.45 ΠL if 0.45 Π < È Ω È < 0.55 Π
B HΩL = :
0 otherwise
- É 10

Π
É 2 H È Ω È -0.55 ΠL H È Ω È -0.45 ΠL if 0.45 Π < È Ω È < 0.55 Π
B HΩL = :
0 otherwise
and the maximum value is at Ω = ± Π2 where the maximum is BI± Π2 M = 0.25, as shown below.
b) For the given signal

XHΩL = 20 Π∆HΩL + 2 Π∆HΩ - 0.2 ΠL + 2 Π∆HΩ + 0.2 ΠL - 3 Π∆HΩ - 0.7 ΠL - 3 Π∆HΩ + 0.7 ΠL, for
-Π £ Ω < Π
Therefore the reconstructed signal becomes
YHΩL = AHΩL XHΩL + BHΩL XHΩ - ΠL
with AHΩL, BHΩL as above, and
XHΩ - ΠL = 20 Π∆HΩ - ΠL + 2 Π∆HΩ + 0.8 ΠL + 2 Π∆HΩ - 0.8 ΠL - 3 Π∆HΩ + 0.3 ΠL - 3 Π∆HΩ - 0.3 ΠL
From the plot of AHΩL and BHΩL we can verify that
AH0L = AH±0.2 ΠL = AH±0.7 ΠL = 1
BH±ΠL = BH±0.8 ΠL = AH±0.3 ΠL = 0
and therefore, for the given signal, y@nD = x@nD.
à Problem 7.10
Solution
First let’s see what is the transfer function (or the frequency response) of
the system shown below.
Applying standard considerations we can see that
12
Applying standard considerations we can see that
YHΩL = 12 QI Ω2 M VI Ω2 M + 12 QI Ω2 - ΠM VI Ω2 - ΠM

Also, from the upsampler, VHΩL = XH2 ΩL, which implies
VI Ω2 M = XHΩL
VI Ω2 - ΠM = XI2 I Ω2 - ΠMM = XHΩ - 2 ΠL = XHΩL
using the periodicity of the DTFT. Therefore, substituting into the
expression for YHΩL we obtain
YHΩL = 12 IQI Ω2 M + QI Ω2 - ΠMM XHΩL

= Q0 HΩL XHΩL
Therefore the impulse response q0 @nD = IDTFT 8Q0 HΩL< is the impulse response q@nD downsampled by
two, ie
q0 @nD = q@2 nD
In other words from the polyphase decomposition
QHzL = Q0 Iz2 M + z-1 Q1 Iz2 M
where
Qk HzL = Z 8q@2 n + kD<
we can determine the transfer function Q0 HzL.
a) We want to determine the four transfer functions Yi HzL X j HzL for i, j = 0, 1. See each one separately:
Y HzL
0

X HzL
= 0 : since, in this case
1
QHzL = G1 HzL H0 HzL = 14 I1 - z-1 + z-2 - z-3 M 14 I1 + z1 + z2 + z3 M

1
= 16 I-z-3 - z-1 + z + z3 M

and therefore Q0 HzL = 0 since there are no even powers of z in QHzL.

Y HzL
1

X HzL
= 0 : since, in this case
0
QHzL = G0 HzL H1 HzL = 14 I1 + z-1 + z-2 + z-3 M 14 I1 - z1 + z2 - z3 M

1
= 16 Iz-3 + z-1 - z - z3 M

and therefore Q0 HzL = 0 since there are no even powers of z in QHzL
Y HzL 1
0

X HzL
= 16 I2 z-1 + 4 + 2 zM : since, in this case

0
Y HzL 1
0

X HzL

0
QHzL = G0 HzL H0 HzL = 14 I1 + z-1 + z-2 + z-3 M 14 I1 + z1 + z2 + z3 M

1
= 16 Iz-3 + 2 z-2 + 3 z-1 + 4 + 3 z + 2 z2 + z3 M

and the polyphase decomposition
1 1
QHzL = 16 I2 z-2 + 4 + 2 z2 M + z-1 16
Iz-2 + 3 + 3 z2 + z4 M

which yields
1
Q0 HzL = 16 I2 z-1 + 4 + 2 zM.

Y HzL 1
1

X HzL

1
QHzL = G1 HzL H1 HzL = 14 I1 - z-1 + z-2 - z-3 M 14 I1 - z1 + z2 - z3 M

1
= 16 I-z-3 + 2 z-2 - 3 z-1 + 4 - 3 z + 2 z2 - z3 M

and the polyphase decomposition
1 1
QHzL = 16 I2 z-2 + 4 + 2 z2 M + z-1 16
I-z-2 - 3 - 3 z2 - z4 M

which yields
1
Q0 HzL = 16 I2 z-1 + 4 + 2 zM.

b) Write all filters in terms of polyphase decompositions
G0 HzL = F0 Iz2 M + z-1 F1 Iz2 M
G1 HzL = G0 H-zL = F0 Iz2 M - z-1 F1 Iz2 M
From the given transfer function we have:
F0 HzL = F1 HzL = 14 I1 + z-1 M

Similarly for H0 HzL and H1 HzL
H0 HzL = E0 Iz2 M + zE1 Iz2 M
H1 HzL = H0 H-zL = E0 Iz2 M - zE1 Iz2 M
Now we can write all transfer functions in terms of the polyphase
components Ek and Fk . In particular for the cross talk transfer functions we can write
G0 HzL H1 HzL = IF0 Iz2 M + z-1 F1 Iz2 MM IE0 Iz2 M - zE1 Iz2 MM
= IF0 Iz2 M E0 Iz2 M - F1 Iz2 M E1 Iz2 MM +
+ z-1 IF1 Iz2 M E0 Iz2 M - z2 F0 Iz2 M E1 Iz2 MM
Therefore taking the first component of the polyphase decomposition we obtain

14
Therefore taking the first component of the polyphase decomposition we obtain
Y1 HzL

X0 HzL
= F0 HzL E0 HzL - F1 HzL E1 HzL
which is zero if we choose
E0 HzL = F1 HzL = 14 I1 + z-1 M

E1 HzL = F0 HzL = 14 I1 + z-1 M
à Problem 7.11
Solution
Write the filters in terms of the polyphase decompositions

G0 HzL = F0 Iz2 M + z-1 F1 Iz2 M
G1 HzL = G0 H-zL = F0 Iz2 M - z-1 F1 Iz2 M
Similarly for H0 HzL and H1 HzL
H0 HzL = E0 Iz2 M + zE1 Iz2 M
H1 HzL = H0 H-zL = E0 Iz2 M - zE1 Iz2 M
Therefore we can write
2
G0 HzL j F0 Iz M y
i z
K O = W2 j
j
j -1
z
z
z
G1 HzL z F1 Iz 2 M
k {
and
H0 HzL i
j E0 Iz2 M yz
K j
O = W2 j z
z
H1 HzL j 2 z
k zE 1 Iz M {
where the matrix
1 1
W2 = J N
1 -1
Performs the 2-point DFT . Putting things together we obtain the scheme as
shown below
Performs the 2-point DFT . Putting things together we obtain the scheme as
shown below
where we recognize the Parallel to Serial and the Serial to Parallel

conversions, which are the inverse of each other. Also the two DFT matrices
W2 are such that
W2 W2 = 2 I
with I the 2 2 identity matrix. Therefore the conditions for Perfect Reconstruction are
F0 HzL E0 HzL = 12

F1 HzL E1 HzL = 12
Since all transfer functions are polynomials, the only possibility is for
Fi HzL and Ei HzL to be constants, independent on z. For example, say
1
F0 HzL = E0 HzL = F1 HzL = E1 HzL = !!!!!

2
Therefore the filters are

1
I1 + z-1 M
G0 HzL = !!!!!

2
1
H0 HzL =
!!!!! H1 + zL
2
1
I1 - z-1 M
G1 HzL = G0 H-zL = !!!!!

2
1
H1 HzL = H0 H-zL =
!!!!! H1 - zL
2
16
à Problem 7.12
Solution
Using the transfer functions we determined in Problem 7.11 we obtain
1 1
G0 HzL H0 HzL = !!!!!
H1 + zL = 1 + z-1 12 I1 + z2 M
I1 + z-1 M !!!!!

2 2
1 1
G1 HzL H1 HzL =
!!!!! I1 - z-1 M
!!!!! H1 - zL = 1 + z-1 12 I-1 - z2 M
2 2
Therefore A00 HzL = A11 HzL = 1. Furthermore for the crosstalk:

1 1
G0 HzL H1 HzL = !!!!!
H1 - zL = 0 + z-1 12 I1 - z2 M
I1 + z-1 M !!!!!

2 2
1 1
G1 HzL H0 HzL =
!!!!! I1 - z-1 M
!!!!! H1 + zL = 0 + z-1 12 I-1 + z2 M
2 2
Therefore A01 HzL = A10 HzL = 0. Finally all the Bij HzL = 0 since there is no aliasing.
à Problem 7.13
Solution
As shown in Problem 7.10 we can write

Y0 HΩL

X0 HΩL
= 12 IG0 I Ω2 M H0 I Ω2 M + G0 I Ω2 - ΠM H0 I Ω2 - ΠMM
Y1 HΩL

X1 HΩL
From the statement of the problem we have
G0 HΩL = H0 HΩL = H HΩL
G1 HΩL = H1 HΩL = H HΩ - ΠL
Therefore the two frequency responses become
Y0 HΩL

X0 HΩL
= 12 I É H I Ω2 M É 2 + É H I Ω2 - ΠM É 2M
Y1 HΩL

X1 HΩL
= 12 I É H I Ω2 - ΠM É 2 + É H I Ω2 M É 2M
where H HΩL is shown below
Therefore:
2 2
Yi HΩL I 10 M Ë Ω2 - 0.55 Π Ë

Π
2 +I 10 M Ë Ω2 - 0.45 Π Ë

Π
2 if 0.45 Π < É Ω2 É < 0.55 Π

Xi HΩL
=:
1 otherwise
Similarly
Y0 HΩL

X1 HΩL
Y1 HΩL

X0 HΩL
which yields
Y0 HΩL

X1 HΩL
= 12 IH I Ω2 - ΠM H I Ω2 M + H I Ω2 M H I Ω2 - ΠMM
Y1 HΩL

X0 HΩL
= 12 IH I Ω2 M H I Ω2 - ΠM + H I Ω2 - ΠM H I Ω2 MM
Therefore
Y0 HΩL

X1 HΩL
= YX
1 HΩL
HΩL
= H I Ω2 M H I Ω2 - ΠM
0
à Problem 7.14
Solution
a) Let C HzL = 1, ie ideal channel. Then

G HzL H HzL = I1 + z-1 M H1 + zL = 2 + z-1 I1 + z2 M
G H-zL H H-zL = I1 - z-1 M H1 - zL = 2 + z-1 I-1 - z2 M
G HzL H H-zL = I1 + z-1 M H1 - zL = 0 + z-1 I1 - z2 M

G H-zL H HzL = I1 - z-1 M H1 + zL = 0 + z-1 I-1 + z2 M
and therefore
Yi HzL = 2 Xi HzL, for i = 0, 1
b) Let C HΩL = e-0.1ÈΩÈ for -Π £ Ω < Π. Then
Y0 HΩL

X0 HΩL
= 12 HG HΩL H HΩL C HΩL + G HΩ - ΠL H HΩ - ΠL C HΩ - ΠLL
18
Y0 HΩL

X0 HΩL
= 12 HG HΩL H HΩL C HΩL + G HΩ - ΠL H HΩ - ΠL C HΩ - ΠLL
Before proceeding any further notice that

e-0.1ÈΩ-ΠÈ if 0 £ Ω < Π
C HΩ - ΠL = :
e-0.1ÈΩ+2 Π-ΠÈ = e-0.1ÈΩ+ΠÈ if - Π £ Ω < 0
and we can write it as
C HΩ - ΠL = e0.1 HÈΩÈ-ΠL for -Π £ Ω < Π
Therefore
Y0 HΩL

X0 HΩL
= H1 + cos HΩLL e-0.1ÈΩÈ + H1 - cos HΩLL e0.1 HÈΩÈ-ΠL
0 Y HΩL
The magnitude É X
HΩL
É is shown below.
0
2.5
1.5
0.5
-3 -2 -1 1 2 3
Similarly
Y1 HΩL

X1 HΩL
= 12 HG HΩ - ΠL H HΩ - ΠL C HΩL + G HΩL H HΩL C HΩ - ΠLL
= H1 - cos HΩLL e-0.1ÈΩÈ + H1 + cos HΩLL e0.1 HÈΩÈ-ΠL
The magnitude É YX

1 HΩL
HΩL
É is shown below.
1
2.5
1.5
0.5
-3 -2 -1 1 2 3
and for the cross talk:

Y1 HΩL

X0 HΩL
= 12 HG HΩL H HΩ - ΠL C HΩL + G HΩ - ΠL H HΩL C HΩ - ΠLL
= -jsin HΩL e-0.1ÈΩÈ + jsin HΩL e0.1 HÈΩÈ-ΠL
Y0 HΩL

X1 HΩL
= 12 HG HΩ - ΠL H HΩL C HΩL + G HΩL H HΩ - ΠL C HΩ - ΠLL
= jsin HΩL e-0.1ÈΩÈ - jsin HΩL e0.1 HÈΩÈ-ΠL
Y HΩL
The magnitude of the crosstalk É YX
1 HΩL
HΩL
0
HΩL
É is shown below.
É = É X
0 1
2.5
1.5
0.5
-3 -2 -1 1 2 3

Maximally Decimated Filter Banks
à Problem 8.1
Solution
Consider the set V of vectors of the form v = a K O, with "a" complex. Then, using the standard
1
−2
rules of scalar and addition and multiplication,
a. v1 + v2 = a1 K O + a2 K O = Ha1 + a2 L K O ∈ V;
1 1 1
−2 −2 −2
b. αv = αa K O ∈ V;
1
−2
c. d. properties of addition and multiplication easy to verify.
Therefore V is a vector space.
à Problem 8.2
Solution
The set V of vectors of the form v = av0 + w0 with v0 , w0 any given non zero vector (v0 = K O
1
−2
and w0 = K O in this case), with again " a " scalar, is not a vector space. In fact, using standard
1
1
rules of vector operations, for any two vectors v1 , v2 ∈ V the sum
v1 + v2 = a1 v0 + w0 + a2 v0 + w0 = Ha1 + a2 L v0 + 2 w0
is clearly not in V.
à Problem 8.3
Solution
a) Let e1 = K O and e2 = K O form a basis for a two dimensional space. Then the dual basis has
1 2
−1 0
to be such that
2
He1 , e2 L = K O
∗T
e1 1 0
∗T 0 1
e2
Then
= He1 , e2 L =K O =K O
e∗T
1 −1 1 2 −1 0 −1
∗T −1 0 0.5 0.5
e2
Therefore, equating the rows we obtain
e1 = K O, e2 = K O
0 0.5
−1 0.5
b) Given any vector v (in this case v = K O) we can express it as

3
−4
v = a1 e1 + a2 e2
with
a1 = e2 v = H0, −1L K O=4

∗T 3
−4
v = H0.5, 0.5L K O = −0.5

∗T 3
a2 = e2
−4
à Problem 8.4
Solution
a) given two signals x1 HtL and x2 HtL continuous in 0 ≤ t ≤ 1 , any linear combination
a1 x1 HtL + a2 x2 HtL is still continuous on the same interval;
b) the inner product
< x, y > = Ÿ x∗ HtL y HtL dt = Ÿ cos HπtL e−t dt = 0.125844

1 1
0 0
c) the inner product
< x, y > = Ÿ x∗ HtL y HtL dt = Ÿ e−j 2 πt ej4πt dt = 0

1 1
0 0
Filter Banks
à Problem 8.5
Solution
Given the decomposition
K O=K OK O
a@kD 1 α x@2 kD
d@kD −α 1 x@2 k + 1D
let's see how we can reconstruct the signal x@nD by a synthesis network. First let's see a "brute force"
approach, and then verify that it fits within the theory developed in the chapter.
~
a[k ] xa [n]
G ( z −1 ) 2 2 G (z )
x[n] x[n]
~ H (z )
H ( z −1 ) 2 2
d [k ] xd [n]
a) From the decomposition we see that
K O= K OK O
x@2 kD 1 1 −α a@kD
x@2 k + 1D 1+α2 α 1 d@kD
b) Now the problem is how to write these two equations in terms of a synthesis network. First notice
that
x@nD = Σ δ@n − 2 kD x@2 kD + δ@n − 2 k − 1D x@2 k + 1D
k
Now substitute for x@2 kD and x@2 k + 1D from the above equations to obtain
x@nD = 1
Σ δ@n − 2 kD Ha@kD − αd@kDL +
2 Σ δ@n − 2 k − 1D Hα a@kD + d@kDL

1+α2 k
1
+ 1+α
k
Now rearrange terms as
4
x@nD = Σ I 1+α
1
2 δ@n − 2 kD +
α
δ@n − 1 − 2 kDM a@kD +
+Σ I 1+α
k 1+α2
−α 1
2 δ@n − 2 kD + 1+α2
δ@n − 1 − 2 kDM d@kD
k
This is of the form

x@nD = Σ g@n − 2 kD a@kD + Σ h@n − 2 kD d@kD
k k
as in the synthesis filter bank shown above. In particular

1 α
g@nD = 1+α2
δ@nD + 1+α2
δ@n − 1D
−α 1
h@nD = 1+α2
δ@nD + 1+α2
δ@n − 1D
which yields the two transfer functions

G HzL = 1
I1 + αz−1 M
H HzL = I−α + z−1 M
1+α2
1
1+α2
c) Now let's derive the same transfer functions from the Perfect Reconstruction conditions. Referring
to the analysis and synthesis section with real filters, shown below, we see that
G Iz−1 M = 1 + αz
H Iz−1 M = −α + z
Therefore we determine the two filters G HzL and H HzL from the conditions
G Iz−1 M G I−z−1 M G HzL H HzL
K O=K O
H Iz−1 M H I−z−1 M G H−zL H H−zL
2 0
0 2
which yields
G Iz−1 M G HzL + G I−z−1 M G H−zL = H1 + αzL G HzL + H1 − αzL G H−zL = 2
H Iz−1 M G HzL + H I−z−1 M G H−zL = H−α + zL G HzL + H−α − zL G H−zL = 0
Let G HzL = β0 + β1 z−1 and let's find the coefficients β0 and β1 . Substituting we obtain
H1 + αzL Iβ0 + β1 z−1 M + H1 − αzL Iβ0 − β1 z−1 M =
= 2 Hβ0 + αβ1 L = 2
in the first equation, and
H−α + zL Iβ0 + β1 z−1 M + H−α − zL Iβ0 − β1 z−1 M =
= 2 H−αβ0 + β1 L = 0
in the second equation.
This yields two equations in the unknowns β0 and β1
K OK O=K O
1 α β0 1
−α 1 β1 0
which yields
G HzL = 1
I1 + αz−1 M
Use the same procedure for H HzL. Let H HzL = β0 + β1 z−1 and we obtain
1+α2
G Iz−1 M H HzL + G I−z−1 M H H−zL = H1 + αzL H HzL + H1 − αzL H H−zL = 0

H Iz−1 M G HzL + H I−z−1 M G H−zL = H−α + zL H HzL + H−α − zL H H−zL = 2
This yields
K OK O=K O
1 α β0 0
−α 1 β1 1
and finally
H HzL = 1
1+α2
I−α + z−1 M
à Problem 8.6
Solution
. Then X HωL = 2 πδ Iω − M.
2π
j n 2π
Let x@nD = e 3
3
a[k ]
~
xa [n]
G ( z −1 ) 2 2 G (z )
x[n] x[n]
~ H (z )
H ( z −1 ) 2 2
xd [n]
d [k ]
Now from the figure we see that
Xa HωL = G HωL I 12 G H−ωL X HωL + 1
G H−ω + πL X Hω + πLM
Xd HωL = H HωL I 12 H H−ωL X HωL + H H−ω + πL X Hω + πLM
2
1
2
From Problem 8.5 recall

G Iz−1 M = 1 + αz
H Iz−1 M = −α + z
6
G HzL = 1
I1 + αz−1 M
H HzL = I−α + z−1 M
1+α2
1
1+α2
Therefore
Xa HωL = I1 + αe−jω M I1 + αejω M δ Iω − M
2 I1+α2 M
1 2π
+
I1 + αe−jω M I1 + αej Hω−πL M δ Iω + M

3
2 I1+α2 M
1 π
+ 3
which becomes
Xa HωL = J1 + αe N J1 + αe N δ Iω − M
2π 2π
2 I1+α2 M
1 −j j 2π
3 3
3
+
J1 + αe N J1 − αe N δ Iω + M
π π
2 I1+α2 M
1 j −j π
+ 3 3
3
Similarly
Xd HωL = I−α + e−jω M I−α + ejω M δ Iω − M
2 I1+α2 M
1 2π
+
I−α + e−jω M I−α + ej Hω−πL M δ Iω + M

3
2 I1+α2 M
1 π
+ 3
which becomes
Xd HωL = J−α + e N J−α + e N δ Iω − M

2π 2π
2 I1+α2 M
1 −j j 2π
3 3
3
+
J−α + e N J−α − e N δ Iω + M
π π
2 I1+α2 M
1 j −j π
+ 3 3
3
π
Now we can see that the two aliased components with frequency ω = 3
cancel each other.
à Problem 8.7
Solution
When all filters are ideal we can verify that

G HωL = G H−ωL = H H−ω + πL are all low pass
H HωL = H H−ωL = G H−ω + πL are all high pass
Also assume the filters to have magnitude 2 within the respective passbands.
a[k ]
~
xa [n]
G ( z −1 ) 2 2 G (z )
x[n] x[n]
~ H (z )
H ( z −1 ) 2 2
xd [n]
d [k ]
Therefore in the expression
Xa HωL = G HωL I 12 G H−ωL X HωL + 1
G H−ω + πL X Hω + πLM
Xd HωL = H HωL I 12 H H−ωL X HωL + H H−ω + πL X Hω + πLM
2
1
2
the aliased terms are zero, since

G HωL G H−ω + πL = H HωL H H−ω + πL = 0.
Also
G HωL G H−ωL + H HωL H H−ωL = 2 for all ω
which yields perfect reconstruction.
à Problem 8.8
Solution
Let G HzL = 1 + z−1 + z−2 .

Step 1: we determine G HzL from the fact that G Iz−1 M G HzL has to be half band filter satisfying the
relation
G Iz−1 M G HzL + G I−z−1 M G H−zL = 2
First we assume G HzL = g@0D + g@1D z + g@2D z2 . Then, substituting in G Iz−1 M G HzL we
obtain
Ig@0D + g@1D z−1 + g@2D z−2 M I1 + z−1 + z−2 M =
= g@0D + Hg@0D + g@1DL z−1 + Hg@0D + g@1D + g@2DL z−2 +
+Hg@1D + g@2DL z−3 + g@2D z−4
The coefficients of even power of z, excluding z0 , have to be zero. Therefore
8
g@0D = 2
g@0D + g@1D + g@2D = 0
g@2D = 0
This yields easily

g@0D = 2, g@1D = −2, g@2D = 0
and therefore
G HzL = 2 H1 − zL
Step 2: we determine the other filters from the relations
H HzL = G I−z−1 M z−N = 2 I1 + z−1 M z−N
H Iz−1 M = G H−zL zN = 2 I1 − z + z2 M zN
with N odd. For example let N = 1, to obtain the set of four filters
G HzL = 1 + z−1 + z−2
G HzL = 2 H1 − zL
H HzL = 2 Iz−1 + z−2 M
H HzL = 2 Iz−1 − z−2 + z−3 M
à Problem 8.9
Solution
From
Q2 HωL = 1 + 15
cos HωL − 5
cos3 HωL + 3
cos5 HωL
expand cos HωL =

8 4 8
1
ejω + 1
e−jω , and substitute z = ejω to obtain
Q2 HzL = 1 + Iz + z−1 M − Iz + z−1 M + Iz + z−1 M

2 2
15 5 3 3 5
16 32 256
With some simple algebra this can be written as
Q2 HzL = 3 z10 −25 z8 +150 z6 +256 z5 +150 z4 −25 z2 +3

256 z5
It has ten zeros at

z1,2 = 0.287251 ± j 0.152892
z3,4 = 2.71275 ± j 1.44389
z5,6,7,8,9,10 = −1
and ten poles

p1,2,3,4,5 = 0
p6,7,8,9,10 = ∞
As expected, also notice that z3,4 = 1 ë z1,2 .
Now we want to factor

Q2 HzL = G Iz−1 M G HzL
a) Orthogonal, ie G HzL = G HzL. In this way we assign to G HzLall poles and zeros inside the unit
circle and half the zeros at z = −1. Therefore
G HzL = K
Hz+1L3 Hz−0.287251 −j 0.152892L Hz−0.287251 +j 0.152892L
z5
with K a constant we need to determine. This leads to

G HzL = K I1 + 2.4255 z−1 + 1.38238 z−2 − 0.405839 z−3 −
−0.256835 z−4 + 0.105889 z−5 M
Without having to remember a lot of formulas, we can compute the constant K from (say)
Q2 HzL z=1 = G H1L G H1L
which yields
2 = K2 4.25112
and therefore K = 0.3327
b) Biorthogonal, ie G HzL ≠ G HzL. The decomposition is not unique, since we assign to G HzL some
(or all) zeros at z = −1, and the rest of the zeros to G HzL. In particular we have
G HzL = K Hz+1LN
G Iz−1 M = K
zM
Hz+1LN R2 HzL
zM
3
where N + N = 6, M + M = 5, K K= and
R2 HzL = Hz − 0.287251 − j 0.152892L Hz − 0.287251 + j 0.152892L

256
Hz − 2.71275 − j 1.44389L Hz − 2.71275 + j 1.44389L

which becomes
10
R2 HzL = z4 − 6 z3 + 38
z2 − 6 z + 1
Notice that all coefficients of R2 HzL are rational numbers.

3
à Problem 8.10
Solution
From Example 8.11, continued with 8.12, the transfer functions are
G HzL = 1 + z−1
G Iz−1 M = 1
H1 + zL
H HzL = 1 − z−1
2
H Iz−1 M = 1
2
H1 − zL
From the relation

Y HzL = G HzL IG Iz−1 M X HzL + G I−z−1 M X H−zLM +
+ H HzL IH Iz−1 M X HzL + H I−z−1 M X H−zLM
we obtain
Y HzL = I 12 I1 + z−1 M H1 + zL + 1
I1 − z−1 M H1 − zLM X HzL +
I 12 I1 + z−1 M H1 − zL + I1 − z−1 M H1 + zLM X H−zL
2
1
2
This yields
Y HzL = X HzL
as expected.
à Problem 8.11
Solution
From the recursion
2 p+1
cp = cp−1
Qp HωL = Qp−1 HωL + cos HωL H1 − cos HωLLp H1 + cos HωLLp
2p
cp
for p ≥ 1, with initial condition c0 = 1 and Q0 HωL = 1 + cos HωL we obtain

2 p+1
35
c3 = 16
and therefore
Q3 HωL = Q2 HωL + 5
cos HωL H1 − cos HωLL3 H1 + cos HωLL3
cos HωL − cos3 HωL + cos5 HωL +
16
15 5 3
= 1+
8 2 8
5
+ 16
After some algebra we obtain

Q3 HωL = 1 + 35
16
cos HωL − 55
16
cos3 HωL + 21
16
cos5 HωL − 5
16
cos7 HωL
from which we generate the Daubechies filter dB4.
Similarly
9 315
c4 = 8
c3 = 128
and therefore
Q4 HωL = Q3 HωL + 35
128
This yields
Q4 HωL = 1 + 315
cos HωL − 145
cos3 HωL + 189
cos5 HωL −
cos7 HωL + cos9 HωL
128 32 64
− 45
32
35
128
from which we generate the Daubechies filter dB5.
à Problem 8.12
Solution
From equation 8.39 and problem 8.9 we obtain

G HzL = 0.3327 I1 + 2.4255 z−1 + 1.38238 z−2 − 0.405839 z−3 −
−0.256835 z−4 + 0.105889 z−5 M
which yields
GHzL = 0.3327 + 0.806964 z-1 + 0.459919 z-2 - 0.135023 z-3 - 0.085449 z-4 + 0.0352293 z-5
Then the corresponding HHzL = GI- z-1 M z-N , with N odd (say N = 5) becomes
è
HHzL = -0.0352293 - 0.085449 z-1 + 0.135023 z-2 + 0.459919 z-3 - 0.806964 z-4 + 0.3327 z-5
12
Now we have to determine the rotation angles θp such that

Gp HzL Gp−1 HzL
= R Hθp L
Hp HzL z−2 Hp−1 HzL
with boundary condition G3 HzL = G HzL and H3 HzL = H HzL. Proceeding backward we obtain the
Gp−1 HzL Gp HzL

recursion
= R H−θp L
z−2 Hp−1 HzL Hp HzL
Starting with p = 3 we obtain

g2 @0D + g2 @1D z−1 + g2 @2D z−2 + ...
0 + 0 z−1 + h2 @0D z−2 + ...
=
R H−θ3 L
0.3327 + 0.806964 z−1 + ...
−0.0352293 + 0.085449 z−1 + ...
Therefore the angle θ3 has to provide the rotation
g2 @0D
K O = R H−θ3 L K O
0.3327
0 −0.0352293
which yields
θ3 = −0.105496 rad
and
G2 HzL = 0.33456 + 0.811475 z-1 + 0.443144 z-2 - 0.182702 z-3
H2 HzL = 0.182702 + 0.443144 z-1 - 0.811475 z-2 + 0.33456 z-3
Iterating again we obtain

g1 @0D + g1 @1D z−1 + g1 @2D z−2 + ...
0 + 0 z−1 + h1 @0D z−2 + ...
=
R H−θ2 L
0.33456 + 0.811475 z−1 + ...
0.182702 + 0.443144 z−1 + ...
and therefore
q2 = 0.499841 rad
which yields
G1 HzL = 0.381196 + 0.924591 z-1

H1 HzL = -0.924591 + 0.381196 z-1
This final stage is already in rotation format as
G1 HzL
= RH-q1 L -1
H1 HzL
1
z
with
q1 = atanI 0.381196 M = 1.17974 rad.
0.924591
Now retrace all the steps, and we can write the two transfer functions GHzL and HHzL as
GHzL 0.994 0.105 z-2 0.878 -0.479 z-2 0.381 -0.924 1

=
HHzL -0.105 0.994 z-2 0.479 0.878 z-2 0.924 0.381 z-1
1 0.106 z-2 1 -0.546 z-2 1 -2.4255 1
= 0.33267
-0.106 z-2 0.546 z-2 2.4255 1 z-1
which leads to the Lattice realization shown below.
a[k ]
2
2.4255 0.546 − 0.106
x[n]
0.332 − 2.4255 − 0.546 0.106
d [k ]
−1 −2 −2 2
z z z

Anual Modern Digital Signal Processing by Roberto Christ I

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Anual Modern Digital Signal Processing by Roberto Christ I

Transféré par

Droits d'auteur :

Formats disponibles

http://ebook29.blogspot.

Chapter 1: Problem Solutions

g) let Λ = - 12 t, then t = -2 Λ and dt = -2 dΛ. Substitute in the integral to obtain

h) let Λ = 3 t, then t = Λ  3 and dt = I 13 M dΛ. Then the integral becomes

a) Frequency F0 = 1  T0 = 1  I3  10-3 M = 13  103 Hz. Then

x HtL = 2.5 cos I 23 1000 Πt + 150 M

All sinusoids are distinct. In continuous time there is no ambiguity between

x1 @nD Frequency Representation

x2 @nD Frequency Representation

x3 @nD Frequency Representation

x4 @nD Frequency Representation

a) since 150 = 15 * Π  180 = Π  12, we can write

b) x HtL = e-j0 .1 Π ej10Πt + ej0 .1 Π e-j10Πt

c) x Ht - 0.1L = e-j2Π x HtL = x HtL

a) x HtL = 2 cos H100 Πt + 0.2 ΠL = ej0 .2 Π ej100Πt + e-j0 .2 Π e-j100Πt

b) x HtL = 32 e-j0 .1 Π ej100Πt + 32 ej0 .1 Π e-j100Πt

x HtL = 32 e-j0 .1 Π ej100Πt + 32 ej0 .1 Π e-j100Πt

c) x HtL = e-j0 .2 Π ej100Πt + ej0 .2 Π e-j100Πt

c) x HtL = e-j0 .2 Π ej100Πt + ej0 .2 Π e-j100Πt

Combining terms we obtain

e) x HtL = ej0t + 12 ej10Πt + 12 e-j10Πt - 2

Referring to the figure above:

If a property is not specified it is assumed Linear, Time Invariant, Causal,

b) Non Linear. In fact if you apply superposition:

If n < 0 then u@kD u@n - kD = 0 for all k, and therefore y@nD = 0.

If n < 0 then u@kD u@n - kD = 0 for all k, and therefore y@nD = 0.

If n < 2 then u@k - 2D u@n - kD = 2 for all k, and therefore y@nD = 0.

If n < 0 then u@kD u@n - kD = 0 for all k, and therefore y@nD = 0.

If n < 0 then u@-kD u@n - kD = u@n - kD. This implies

Applying the geometric series we obtain

h) X HzL = 1 + z-1 + z-2 , ROC: 0 < È z È;

a) x@nD = 2  0.5n-1 u@n - 1D, then X HzL = 2 z-1 z-0.5

b) x@nD = 0.5n u@-n - 1D + ∆@nD, then X HzL = z-0.5

e) x@nD = H-0.5Ln u@nD, then X HzL = z+0.5

g) x@nD = 0.5n u@nD + 0.5-n u@-n - 1D, then X HzL = z-0.5

Since È z È > 1 all terms are causal. Therefore

therefore y@nD = -0.5n u@nD + 2 u@nD;

a) The difference equation can be implemented in a number of ways, so we have

When the system is causal the ROC is È z È > 2, then

a) H H0.2 ΠL = 1.0417, then y@nD = 1.0417 ej0 .2 Πn ;

a) From the graph H H0.1 ΠL = 6.5 e-j . Therefore the output is

Time, z-, and Frequency Domains

The poles are both z = 0.

If we want the DC gain to be one, ie H HΩL = 1 when Ω = 0, then b0 = 0.7664;

Fourier Analysis of Discrete Time Signals

Y = @6, -0.123 + j0 .1846, -0.123 - j0 .1846D

Y = @0, 0.2439 + j0 .1951, 0, 0.2439 - j0 .1951D

X = @6.0000, -1.5000 + j0 .8660, -1.5000 - j0 .8660D

X = @1 + j2 .4142, 1 - j2 .4142, 1 - j0 .4142 i, 1 + j0 .4142D

X = @-4.0961, 4.1904 - j2 .4938, 2.7687, 4.1904 + j2 .4938D

X = @0, 7.0534 + j9 .7082, 0, 0, 0, 0, 0, 7.0534 - j9 .7082D

An alternative way would be to use the z-Transforms. Since, in this case

x@nD = 1.25n u@-n - 1D + 0.8n u@nD

and the result follows with some algebra.

c) x@nD = 1 when 0 £ n £ 4 and x@nD = 0 otherwise. Therefore

c) x@nD = 1 when 0 £ n £ 4 and x@nD = 0 otherwise. Therefore

d) x@nD = 1.5 ej0 .2 Π ej0 .1 Πn + 1.5 e-j0 .2 Π e-j0 .1 Πn .

X HΩL = 3 Πej0 .2 Π ∆ HΩ - 0.1 ΠL + 3 Πe-j0 .2 Π ∆ HΩ + 0.1 ΠL

X HΩL magnitude and phase

X HΩL magnitude and phase

X HΩL magnitude and phase

X HΩL magnitude and phase

Apply the properties of the Fourier Transform:

g) let Λ = - 12 t, then t = -2 Λ and dt = -2 dΛ. Substitute in the integral to obtain

h) let Λ = 3 t, then t = Λ 3 and dt = I 13 M dΛ. Then the integral becomes

a) Frequency F0 = 1 T0 = 1 I3 10-3 M = 13 103 Hz. Then

x HtL = 2.5 cos I 23 1000 Πt + 150 M

a) since 150 = 15 * Π 180 = Π 12, we can write

b) x HtL = 32 e-j0 .1 Π ej100Πt + 32 ej0 .1 Π e-j100Πt

x HtL = 32 e-j0 .1 Π ej100Πt + 32 ej0 .1 Π e-j100Πt

e) x HtL = ej0t + 12 ej10Πt + 12 e-j10Πt - 2

a) x@nD = 2 0.5n-1 u@n - 1D, then X HzL = 2 z-1 z-0.5

a) FT 8rect H2 tL< = 12 sinc I F2 M

In our case T0 = 0.1, F0 = 1 T0 = 10, x HtL = sinc H2 tL,

c) FT 8x HtL cos H100 ΠtL< = 12 X IF - 50 + 12 X HF + 50L

e) From the properties, FT 8x H5 tL< = 15 X I F5 M shown below:

Ω0 = 2 Π 500 1200 = Π 1.2;

Y HΩL È Ω=2 ΠFFs =

Y HΩL È Ω=2 ΠFFs =

a) H HzL an ideal Low Pass Filter with bandwidth Π 4;