Sie sind auf Seite 1von 8

University of Maryland at College Park

Department of Electrical and Computer Engineering


ENEE 624 Advanced Digital Signal Processing
Problem Set 1 Solutions
Spring 2003
Wednesday, February 5, 2003
Problem 1.1
(a) We have
Y (z) [z
1

5
2
+z] = X(z) H(z) =
1
z
1

5
2
+z
or, equivalently,
H(z) =
z
1
1
5
2
z
1
+z
2
.
Hence, N = 2, a
1
= 1, a
j
= 0, for j = 1, b
0
= 1, b
1
=
5
2
, b
2
= 1.
(b) Factoring the denominator of H(z) we get
H(z) =
z
1
_
1
1
2
z
1
_
(1 2z
1
)
.
Therefore, H(z) has poles at p
1
=
1
2
and p
2
= 2, and zeros at q
1
= and q
2
= 0. Also
H(z) =

2
3
1
1
2
z
1
+
2
3
1 2z
1
.
There are 3 possible regions of convergence. First, ROC
1
= {z : |z| > 2}, giving
h
1
[n] =
2
3
_
1
2
_
n
u[n] +
2
3
2
n
u[n]
=
2
3
_
2
n

1
2
n
_
u[n] .
This system is causal but not stable. Second, ROC
2
=
_
z :
1
2
< |z| < 2
_
, giving
h
2
[n] =
2
3
_
1
2
_
n
u[n]
2
3
2
n
u[n 1] .
This system is stable but not causal. Finally, ROC
3
=
_
z : |z| <
1
2
_
, giving
h
3
[n] =
2
3
_
1
2
_
n
u[n 1]
2
3
2
n
u[n 1]
=
2
3
_
1
2
n
2
n
_
u[n 1] .
This system is neither stable nor causal.
Problem Set 1 Solutions 2
Problem 1.2
If any of the systems S
1
, S
2
, S
3
, S
4
were LTI it would satisfy
e
jn/3
e
jn/3
where is a complex number (i.e., independent of n). We have
S
1
:
e
jn/3
= ne
jn/3
= n .
Thus S
1
is not LTI.
S
2
:
e
jn/3
= e
j7n/3
= e
j(2n+n/3)
= e
jn/3
= 1 .
Thus S
2
could be LTI.
S
3
:
e
jn/3
= e
jn/3
u[n] = u[n] .
Thus S
3
is not LTI.
S
4
:
e
jn/3
= e
jn/3
_
1
2 j

e
j2n/3
2 j
_
=
1
2 j

e
j2n/3
2 j
.
Thus S
4
is not LTI.
Problem Set 1 Solutions 3
Problem 1.3
H(z) = K
_
1 q
1
z
1
_ _
1 q
2
z
1
_
(1 p
1
z
1
) (1 p
2
z
1
)
= K
(z q
1
) (z q
2
)
(z p
1
) (z p
2
)
.
Let q
i
= r
i
e
j
i
, and p
i
=
i
e
j
i
, with r
i
,
i
0, for i = 1, 2. We also have
C(z) = H(z) H

(1/z

) .
So the zeros of C(z) are at q
1
, q
2
, 1/q

1
and 1/q

2
, and the poles at p
1
, p
2
, 1/p

1
and 1/p

2
. Hence,
each of |q
1
|, |q
2
|, |p
1
|, snd |p
2
| are either 3/4 or 4/3. Since H(z) is minimum-phase (i.e., all
poles and zeros are inside the unit circle), it must be that |q
1
| = |q
2
| = |p
1
| = |p
2
| = 3/4, i.e.,
q
i
=
3
4
e
j
i
, and p
i
=
3
4
e
j
i
. From the gure, since |H(e
j
)| has local minima at = /2 and
= /2, the phases of the zeros of H(z) are
1
= /2 and
2
= /2, i.e., q
1
= 3j/4 and
q
2
= 3j/4. Similarly, since |H(e
j
)| has local maxima at = 0 and = , the phases of
the poles of H(z) are
1
= 0 and
2
= , i.e., p
1
= 3/4 and p
2
= 3/4. It thus remains to
determine K. From the gure, |H(e
j0
)| = 1 (0 dB). Hence,
|H(1)| = 1 |K|
| (1 3j/4) (1 + 3j/4) |
| (1 3/4) (1 + 3/4) |
= 1 |K| =
7
25
,
and since h[n] is real-valued K = 7/25. So the two possible minimum-phase systems consistent
with the given information have transfer functions
H(z) =
7
25
_
1
3j
4
z
1
_ _
1 +
3j
4
z
1
_
_
1
3
4
z
1
_ _
1 +
3
4
z
1
_ ,
and
H(z) =
7
25
_
1
3j
4
z
1
_ _
1 +
3j
4
z
1
_
_
1
3
4
z
1
_ _
1 +
3
4
z
1
_ ,
and both have ROC
H
= {z : |z| > 3/4}.
Problem Set 1 Solutions 4
Problem 1.4
(a) Consider the cascade of two allpass systems with frequency response H
1
_
e
j
_
and H
2
_
e
j
_
,
respectively. Let H
_
e
j
_
denote the frequency response of the overall system. Since
H
1
_
e
j
_
and H
2
_
e
j
_
are allpass, we have
|H
1
_
e
j
_
| = c
1
, and |H
2
_
e
j
_
| = c
2
.
This implies that
|H
_
e
j
_
| = |H
1
_
e
j
_
| |H
2
_
e
j
_
| = c
1
c
2
,
which implies that the overall system is also allpass. Thus, the answer is no.
(b) Let H
1
_
e
j
_
= 1 for all , and
H
2
_
e
j
_
=
_
1 || < /2
1 /2 < ||
.
Then, clearly, |H
1
_
e
j
_
| = |H
2
_
e
j
_
| = 1, i.e., both H
1
_
e
j
_
and H
2
_
e
j
_
are allpass.
Let H
_
e
j
_
denote the frequency response of the parallel interconection system. Then,
H
_
e
j
_
= H
1
_
e
j
_
+H
2
_
e
j
_
=
_
2 || < /2
0 /2 < ||
,
which is clearly lowpass. Hence, the answer is yes.
(c)
H
_
e
j
_
= c
1
1
p

e
j
1 pe
j
= c
e
j
p

1 p

e
j
1 pe
j
= c
e
j
p

1 pe
j
1 pe
j
.
Thus
|H
_
e
j
_
| = |c|

e
j

|p

1 pe
j

|1 pe
j
|
=
|c|
|p|
,
i.e., independent of (and, therefore, allpass). In fact, via an analytic continuation
argument one can show that, in light of the fact that |H
_
e
j
_
| = c
o
, if H(z) has a pole
at p, then it must have a zero at 1/p

.
(d) It is more convenient to consider the unity-gain term from H(z) in part (c) [i.e., neglect
the constant-gain term c/p

].
H
1
_
e
j
_
=
e
j
p

1 pe
j
,
or,
H
1
(z) =
z
1
p

1 p z
1
,
Note that if p = 0, H
1
(z) = z
1
, and if p = , H
1
(z) = z, so that it covers all possible
rst-order factors. Hence, the general form is as follows:
H(z) =
N

i=1
z
1
p

i
1 p
i
z
1
for any p
1
, p
2
, , p
N
C {}.
Problem Set 1 Solutions 5
Problem 1.5
(a) The condition to avoid aliasing in the band (/T
1
, /T
1
) passed by the reconstruction
lter is
s
= 2/T
1
2
o
, or, equivalently, in terms of T
1
, T
1


o
.
(b) We rst note that X
c
(j) is nonzero and arbitrary only over the range
2o
3
< || <
o
.
The width of this range is 2
o
/3, i.e., considerably less than 2
o
. Note that by picking
the reconstruction lter to have frequency response
H
s
(j) =
_
K
2
o
3
< || <
o
0 otherwise
,
(where K is a gain that we will pick later), we eectively ensure that if X
o
(j) = X
c
(j)
for
2o
3
< || <
o
, then x
c
(t) = x
o
(t) (since both signals have no frequency components
outside this range). Consequently, all we need to do is vary
s
= 2/T
2
, from to 0,
and observe for what values of
s
(or T
2
) there is no aliasing in the bands of interest.
Clearly, there is no aliasing for
s
2
o
(case (a) in Fig. 1.5-2). Once we make, however,

s
just smaller than 2
o
we start having aliasing. In particular, as Fig. 1.5-2 case (b)
reveals, we have aliasing for
4
3

o
<
s
< 2
o
. However, there is no aliasing if
s
=
4
3

o
.
Furthermore, if
o

s

4
3

o
there is no aliasing (case (c) in Fig. 1.5-2). Finally,
although for
2
3

o
<
s
<
o
there is aliasing (case (d) in Fig. 1.5-2), at
2
3

o
=
s
there
is no aliasing (case (e) in Fig. 1.5-2). So we have no aliasing if and only if

s

__
2
3

o
_

o
,
4
3

o
_
[2
o
, )
_
which is equivalent to
T
2
A
__
0,

o
_

_
3
2
o
,
2

o
_

_
3

o
__
.
For any T
2
A we have
x
o
(t) =
K
T
2
x
c
(t)
so by using a gain K = T
2
in the reconstruction lter H
s
(j) we get the desired x
o
(t) =
x
c
(t). Clearly, the largest T
2
that allows reconstruction of x
c
(t) from x
2
[n] is T
2
=
3
o
.
The corresponding frequency response is shown in Figure 1.5-1.
H
s
(j)
-
6
0
3
o
3
o
2
o
3

o
2
o
3

Figure 1.5-1 Frequency response H


s
(j)
Problem Set 1 Solutions 6
Case (a):
s
2
o
X
s
(j)
-
6
0
e
e
e
e
e
ee

2
3

2
3

o

o

o
e
e
e
e
e
ee

s
Case (b):
4
3

o
<
s
< 2
o
X
s
(j)
-
6
0
e
e
e
e
e
ee

2
3

2
3

o

o

2
3

o
e
e
e
e
e
ee
2
3

s
Case (c):
o

s

4
3

o
X
s
(j)
-
6
e
e
e
e
e
ee

2
3

2
3

o

o

o
e
e
e
e
e
ee

2
s

o
e
e
e
e
e
ee

o
2
s
Case (d):
2
3

o
<
s
<
o
X
s
(j)
-
6
e
e
e
e
e
ee

2
3

2
3

o

o

2
s

2
3

o
e
e
e
e
e
ee
2
3

o
2
s

2
3

o
e
e
e
e
e
ee
2
3

s
Case (e):
s
=
2
3

o
X
s
(j)
-
6
0
e
e
e
e
e
ee

2
3

2
3

o

o

e
e
e
e
e
ee
e
e
e
e
e
ee
e
e
e
e
e
ee
Figure 1.5-2 Fourier transform X
s
(j)
Problem Set 1 Solutions 7
Problem 1.6
Let
x[n] y[n] , x
1
[n] y
1
[n] , x
2
[n] y
2
[n] . (1)
First consider x
1
[n] = 2 x[n]. As the system possesses the additivity property,
x
1
[n] = x[n] +x[n] y
1
[n] = y[n] +y[n] ,
so
2 x[n] 2 y[n] ,
and by induction,
nx[n] ny[n] , for every positive integer n. (2)
Next, let x
1
[n] = 0 = 0 + 0 = x
1
[n] +x
1
[n]. Then
x
1
[n] = 0 = x
1
[n] +x
1
[n] y
1
[n] = y
1
[n] +y
1
[n] = 2y
1
[n] y
1
[n] = 0 ,
This implies that
0 0 , (3)
which, when combined with (2), gives
nx[n] ny[n] , for every nonegative integer n. (4)
Next, let x
1
[n] = x[n] and x
2
[n] = x
1
[n] + x[n] = 0. Due to (3) and the additivity property
of the system we have
0 = x
2
[n] = x
1
[n] +x[n] 0 = y
2
[n] = y
1
[n] +y[n] y
1
[n] = y[n] .
So x[n] y[n], and by induction nx[n] ny[n], for every positive integer n, which,
when combined with (4), gives
nx[n] ny[n] , for every integer n. (5)
Finally, let x
1
[n] =
p
q
x[n], where p, q are arbitrary integers (clearly, q = 0). Note that from (5)
we have
p x[n] p y[n] , and q x
1
[n] q y
1
[n] .
However, p x[n] = q x
1
[n], so p y[n] = q y
1
[n] or, equivalently, y
1
[n] =
p
q
y[n], i.e.,
p
q
x[n]
p
q
y[n]. Summarizing, if the system possesses the additivity property, then
x[n] y[n] , implies x[n] y[n] , for any rational .
It can be shown, however, that the system does not necessarily possess the scaling property
for irrational (in which case the system is not homogeneous). Construction of a system that
satises additivity but not homogeneity is in fact nontrivial. One well-known example involves
the use of Hamel bases as a means of describing such a system.
Problem Set 1 Solutions 8
The interested reader may consult
Real Functions, by Casper Goman, pp 150151;
for details. In particular, the function f() dened at the end of the section on page 151 provides
one such example of a memoryless system
x[n] y[n] = f(x[n]) ,
where f() satises
f(x +y) = f(x) +f(y) but not f(x) = f(x) for all real .
Roughly speaking, the function f() is dened via a Hamel basis according to which any real
number is uniquely represented as a nite linear combination of the rational basis elements,
i.e., if
x

=
1
x

1
+
2
x

2
+ +
M
x

M
, with basis elements x

i
then
f(x

) =
1
f(x

1
) +
2
f(x

2
) + +
M
f(x

M
) .
In particular, by picking the elements of the basis x

1
, x

2
, and setting f(x

1
) = 1 and
f(x

2
) = 0, we obtain the desired function.