Digital Communications 5th Ed SolutionsChap2

Solutions Manual
for
Digital Communications, 5th Edition
(Chapter 2) 1
Prepared by
Kostas Stamatiou
January 11, 2008
c
PROPRIETARY MATERIAL. The
McGraw-Hill Companies, Inc. All rights reserved. No part of this
Manual may be displayed, reproduced or distributed in any form or by any means, without the prior written
permission of the publisher, or used beyond the limited distribution to teachers and educators permitted by
McGraw-Hill for their individual course preparation. If you are a student using this Manual, you are using
it without permission.
2
Problem 2.1
a.
x
(t) =
Hence :
=
=
x(a)
ta da
R
1 x(b)
t+b (db)
R
1 x(b)
t+b db
R
1 x(b)
(t)
tb db = x
x(t) = 1
=
x(a)
da
ta
where we have made the change of variables : b = a and used the relationship : x(b) = x(b).
b. In exactly the same way as in part (a) we prove :
x
(t) = x
(t)
c. x(t) = cos 0 t, so its Fourier transform is : X(f ) = 21 [(f f0 ) + (f + f0 )] , f0 = 20 .

Exploiting the phase-shifting property (2-1-4) of the Hilbert transform :
) = 1 [j(f f0 ) + j(f + f0 )] = 1 [(f f0 ) (f + f0 )] = F 1 {sin 2f0 t}
X(f
2
2j
Hence, x
(t) = sin 0 t.
d. In a similar way to part (c) :
x(t) = sin 0 t X(f ) =
1
) = 1 [(f f0 ) (f + f0 )]
[(f f0 ) (f + f0 )] X(f
2j
2
) = 1 [(f f0 ) + (f + f0 )] = F 1 {cos 20 t} x
X(f
(t) = cos 0 t
2
e. The positive frequency content of the new signal will be : (j)(j)X(f ) = X(f ), f > 0, while
the negative frequency content will be : j jX(f ) = X(f ), f < 0. Hence, since X(f
) = X(f ),
we have : x
(t) = x(t).
f. Since the magnitude
response of the Hilbert transformer is characterized by : |H(f )| = 1, we

have that : X(f ) = |H(f )| |X(f )| = |X(f )| . Hence :
Z
Z

2
|X(f )|2 df
X(f ) df =
c
McGraw-Hill Companies, Inc. All rights reserved. No part of this Manual may be displayed,
reproduced or distributed in any form or by any means, without the prior written permission of the publisher, or used beyond the
limited distribution to teachers and educators permitted by McGraw-Hill for their individual course preparation. If you are a
student using this Manual, you are using it without permission.
3
and using Parsevals relationship :
Z
x
(t)dt =
x2 (t)dt
g. From parts (a) and (b) above, weR note that if x(t) is even, x
(t) is odd and vice-versa. Therefore,
x(t)
x(t) is always odd and hence : x(t)
x(t)dt = 0.
Problem 2.2
1. Using relations
1
1
Xl (f f0 ) + Xl (f f0 )
2
2
1
1
Y (f ) = Yl (f f0 ) + Yl (f f0 )
2
2
X(f ) =
and Parsevals relation, we have

Z
Z
X(f )Y (f ) dt
x(t)y(t) dt =

Z
1
1
1
1
=
Xl (f f0 ) + Xl (f f0 )
Yl (f f0 ) + Yl (f f0 ) df
2
2
2
2
Z
Z
1
1
Xl (f f0 )Yl (f f0 ) df +
Xl (f f0 )Yl (f f0 ) df
=
4
4
Z
1
1
Xl (u)Yl (u) du + Xl (v)Y (v) dv
=
4
4

Z
1
Xl (f )Yl (f ) df
= Re
2

Z
1
xl (t)yl (t) dt
= Re
2
where we have used the fact that since Xl (f f0 ) and Yl (f f0 ) do not overlap, Xl (f
f0 )Yl (f f0 ) = 0 and similarly Xl (f f0 )Yl (f f0 ) = 0.
2. Putting y(t) = x(t) we get the desired result from the result of part 1.
c
4
Problem 2.3
A well-known result in estimation theory based on the minimum mean-squared-error criterion states
that the minimum of Ee is obtained when the error is orthogonal to each of the functions in the
series expansion. Hence :
#
Z "
K
X
s(t)
sk fk (t) fn (t)dt = 0,
n = 1, 2, ..., K
(1)
k=1
since the functions {fn (t)} are orthonormal, only the term with k = n will remain in the sum, so :
Z
s(t)fn (t)dt sn = 0,
n = 1, 2, ..., K
or:
sn =
s(t)fn (t)dt
n = 1, 2, ..., K
The corresponding residual error Ee is :

ih
i
R h
P
PK
s
f
(t)
s(t)
s
f
(t)
dt
Emin = s(t) K
k=1 k k
n=1 n n
=
2
|s(t)| dt
2
|s(t)| dt
= Es
PK
k=1 |sk |
R PK
k=1 sk fk (t)s (t)dt
R PK
k=1 sk fk (t)s
PK
n=1 sn
R h
s(t)
i
s
f
(t)
fn (t)dt
k
k
k=1
PK
(t)dt
where we have exploited relationship (1) to go from the second to the third step in the above
calculation.
Note : Relationship (1) can also be obtained by simple differentiation of the residual error with
respect to the coefficients {sn } . Since sn is, in general, complex-valued sn = an + jbn we have to
differentiate with respect to both real and imaginary parts :
ih
i
R h
PK
PK
d
d
s(t)
s
f
(t)
s(t)
s
f
(t)
dt = 0
E
=
k=1 k k
n=1 n n
dan e
dan
h
i
h
i
PK
PK
f (t) s(t)
a
f
(t)
s(t)
s
f
(t)
s
f
(t)
dt = 0
+
a
n
n
n
n
n
n
n n
n=1
n=1
2an
n
h
io
PK
(t) s(t)
Re
f
s
f
(t)
dt = 0
n
n=1 n n
h
io
n
PK
(t) s(t)
s
f
(t)
dt = 0,
Re
f
n
n
n
n=1
n = 1, 2, ..., K
c
5
where we have exploited the identity : (x + x ) = 2Re{x}. Differentiation of Ee with respect to bn
will give the corresponding relationship for the imaginary part; combining the two we get (1).
Problem 2.4
The procedure is very similar to the one for the real-valued signals described in the book (pages
33-37). The only difference is that the projections should conform to the complex-valued vector
space :
Z
c12=
s2 (t)f1 (t)dt
and, in general for the k-th function :

Z
sk (t)fi (t)dt,
cik =
i = 1, 2, ..., k 1
Problem 2.5
The first basis function is :
s4 (t) 1/ 3, 0 t 3
s4 (t)
g4 (t) = = =
E4
3
0,
o.w.
Then, for the second basis function :
c43
Hence :
2/3, 0 t 2
s3 (t)g4 (t)dt = 1/ 3 g3 (t) = s3 (t) c43 g4 (t) =

=
4/3, 2 t 3
0,
o.w
Z
1/
6,
0
g3 (t)
g3 (t) =
=
2/ 6, 2 t 3
E3
0,
o.w
R3
where E3 denotes the energy of g3 (t) : E3 = 0 (g3 (t))2 dt = 8/3.
For the third basis function :
Z
Z
s2 (t)g3 (t)dt = 0
s2 (t)g4 (t)dt = 0 and c32 =
c42 =
c
6
Hence :
g2 (t) = s2 (t) c42 g4 (t) c32 g3 (t) = s2 (t)
and
2, 0 t 1
1/
g2 (t)
g2 (t) = =
1/ 2, 1 t 2
E2
0,
o.w
R2
where : E2 = 0 (s2 (t))2 dt = 2.
Finally for the fourth basis function :
Z
Z
s1 (t)g4 (t)dt = 2/ 3, c31 =

c41 =
s1 (t)g3 (t)dt = 2/ 6, c21 = 0
Hence :
g1 (t) = s1 (t) c41 g4 (t) c31 g3 (t) c21 g2 (t) = 0 g1 (t) = 0
The last result is expected, since the dimensionality of the vector space generated by these signals
is 3. Based on the basis functions (g2 (t), g3 (t), g4 (t)) the basis representation of the signals is :

s4 = 0, 0, 3 E4 = 3
p

s3 = 0, 8/3, 1/ 3 E3 = 3

2, 0, 0 E2 = 2
s2 =

s1 = 2/ 6, 2/ 3, 0 E1 = 2
Problem 2.6
Consider the set of signals enl (t) = jnl (t), 1 n N , then by

definition of lowpass equivalent
signals and by Equations 2.2-49
and
2.2-54,
we
see
that
(t)s
are
2 times the lowpass equivalents
n
e
e
of nl (t)s and n (t)s are 2 times the lowpass equivalents of nl (t)s. We also note that since
n (t)s have unit energy, hnl (t), enl (t)i = hnl (t), jnl (t)i = j and since the inner product is pure
imaginary, we conclude that n (t) and en (t) are orthogonal. Using the orthonormality of the set
nl (t), we have
hnl (t), jml (t)i = jmn
and using the result of problem 2.2 we have
We also have
hn (t), em (t)i = 0
for all n, m
hn (t), m (t)i = 0 for all n 6= m

c
7
and
hen (t), em (t)i = 0 for all n 6= m
Using the fact that the energy in lowpass equivalent signal is twice the energy in the bandpass
signal we conclude that the energy in n (t)s and en (t)s is unity and hence the set of 2N signals
{n (t), en (t)} constitute an orthonormal set. The fact that this orthonormal set is sufficient for
expansion of bandpass signals follows from Equation 2.2-57.
Problem 2.7
Let x(t) = m(t) cos

less than f0 . Then
2f0 t where m(t) is real and lowpass with bandwidth
x(t)] = 2j M (f f0 ) + 2j M (f + f0 )
F [
x(t)] = j sgn(f ) 21 M (f f0 ) + 12 M (f + f0 ) and hence F [
where we have used that fact that M (f f0 ) = 0 for f < 0 and M (f + f0 ) = 0 for f > 0. This
shows that x
(t) = m(t) sin 2f0 t. Similarly we can show that Hilbert transform of m(t) sin 2f0 t is
m(t) cos 2f0 t. From above and Equation 2.2-54 we have
H[n (t)] = 2ni (t) sin 2f0 t + 2nq (t) cos 2f0 t = en (t)
Problem 2.8
R
For real-valued signals the correlation coefficients are given by : km = E1E sk (t)sm (t)dt and
k m

1/2
(e)
the Euclidean distances by : dkm = Ek + Em 2 Ek Em km
. For the signals in this problem :
E1 = 2, E2 = 2, E3 = 3, E4 = 3
2
6
12 = 0
13 =
23 = 0
24 = 0
14 = 26
34 = 31
and:
q
q
(e)
(e)
(e)
d12 = 2
d13 = 2 + 3 2 6 26 = 1 d14 = 2 + 3 + 2 6 26 = 3
(e)
(e)
d23 = 2 + 3 = 5
d24 = 5
q
(e)
d34 = 3 + 3 + 2 3 31 = 2 2
c
8
Problem 2.9
We know from Fourier transform properties that if a signal x(t) is real-valued then its Fourier
transform satisfies : X(f ) = X (f ) (Hermitian property). Hence the condition under which sl (t)
is real-valued is : Sl (f ) = Sl (f ) or going back to the bandpass signal s(t) (using 2-1-5):
S+ (fc f ) = S+
(fc + f )
The last condition shows that in order to have a real-valued lowpass signal sl (t), the positive frequency content of the corresponding bandpass signal must exhibit hermitian symmetry around the
center frequency fc . In general, bandpass signals do not satisfy this property (they have Hermitian
symmetry around f = 0), hence, the lowpass equivalent is generally complex-valued.
Problem 2.10
a. To show that the waveforms fn (t), n = 1, . . . , 3 are orthogonal we have to prove that:
Z
fm (t)fn (t)dt = 0,
m 6= n
Clearly:
c12 =
f1 (t)f2 (t)dt =
Z 2
f1 (t)f2 (t)dt
0
4
f1 (t)f2 (t)dt
f1 (t)f2 (t)dt +
2
Z
1
1 4
1
1 2
dt
dt = 2 (4 2)
=
4 0
4 2
4
4
= 0
=
Similarly:
c13 =
Z 1
1
=
4
= 0
and :
c23 =
1
dt
4
dt
1
4
f1 (t)f3 (t)dt
2
1
1
dt
4
f2 (t)f3 (t)dt =
Z 1
1
4
= 0
=
f1 (t)f3 (t)dt =
dt +
1
3
2
1
dt +
4
dt
f2 (t)f3 (t)dt
1
4
3
2
dt
1
4
dt
c
9
Thus, the signals fn (t) are orthogonal. It is also straightforward to prove that the signals have unit
energy :
Z
|fi (t)|2 dt = 1, i = 1, 2, 3
Hence, they are orthonormal.

b. We first determine the weighting coefficients
Z
x(t)fn (t)dt,
xn =
n = 1, 2, 3
x1 =
x2 =
x3
x(t)f1 (t)dt =
0
4
x(t)f2 (t)dt =
1
2
1
2
Z
0
4
dt +
1
2
dt
1
2
dt +
2
1
2
dt = 0
3
x(t)dt = 0
Z
Z
Z
Z
Z 4
1 2
1 3
1 4
1 1
dt
dt +
dt +
dt = 0
x(t)f3 (t)dt =
=
2 0
2 1
2 2
2 3
0
0
As it is observed, x(t) is orthogonal to the signal wavaforms fn (t), n = 1, 2, 3 and thus it can not
represented as a linear combination of these functions.
Problem 2.11
a. As an orthonormal set of basis functions we consider the set
1 0t<1
f1 (t) =
f2 (t) =
0 o.w
1 2t<3
f3 (t) =
f4 (t) =
0 o.w
1 1t<2
0 o.w
1 3t<4
0 o.w
In matrix notation, the four waveforms can be represented as
s1 (t)
2 1 1 1
s (t) 2
1
1
0
2

s3 (t) 1 1
1 1
s4 (t)
1 2 2
2
f1 (t)
f2 (t)
f3 (t)
f4 (t)
Note that the rank of the transformation matrix is 4 and therefore, the dimensionality of the
waveforms is 4
c
10
b. The representation vectors are
s1 =
s2
s3
s4
2 1 1 1
h
i
=
2 1 1 0
h
i
=
1 1 1 1
h
i
=
1 2 2 2
c. The distance between the first and the second vector is:
r h
i2
p

2
d1,2 = |s1 s2 | = 4 2 2 1 = 25
Similarly we find that :
d1,3 =
d1,4 =
d2,3 =
d2,4 =
|s1 s3
|s2 s3 |2
|2
|s1 s4 |2
|s2 s4 |2
r h

=
r h

=
r h

=
r h

=
r h

=
i2

1 0 2 0 = 5
i2

1 1 1 3 = 12
i2

3 2 0 1 = 14
i2

3 3 3 2 = 31
i2

0 1 3 3 = 19
Thus, the minimum distance between any pair of vectors is dmin = 5.

d3,4 =
|s3 s4 |2
Problem 2.12
As a set of orthonormal functions we consider the waveforms
1 0t<1
1 1t<2
f1 (t) =
f2 (t) =
0 o.w
0 o.w
The vector representation of the signals is
s1 =
s2 =
s3 =
s4
2 2 2
2 0 0
0 2 2
h
i
=
2 2 0
1 2t<3
f3 (t) =
0 o.w
c
11
Note that s3 (t) = s2 (t) s1 (t) and that the dimensionality of the waveforms is 3.
Problem 2.13
1. P (E2 ) = P (R2, R3, R4) = 3/7.

2. P (E3 |E2 ) =
P (E3 E2 )
P (E2 )
P (R2)
3/7
= 13 .
3. Here E4 = {R2, R4, B2, R1, B1} and P (E2 |E4 E3 ) =
P (E2 E3 E4 )
P (E3 E4 )
P (R2)
P (R2,B2,R1,B1)
= 14 .
4. E5 = {R2 , R4 , B2 }. We have P (E3 E5 ) = P (R2 , B2 ) = 27 and P (E3 ) = P (R1, R2, B1, B2) =
and P (E5 ) = 73 . Obviously P (E3 E5 ) 6= P (E3 )P (E5 ) and the events are not independent.
4
7
Problem 2.14
1. P (R) = P (A)P (R|A) + P (B)P (R|B) + P (C)P (R|C) = 0.2 0.05 + 0.3 0.1 + 0.5 0.15 =
0.01 + 0.03 + 0.075 = 0.115.
2. P (A|R) =
P (A)P (R|A)
P (R)
0.01
0.115
0.087.
Problem 2.15
The relationship holds for n = 2 (2-1-34) : p(x1 , x2 ) = p(x2 |x1 )p(x1 )

Suppose it holds for n = k, i.e : p(x1 , x2 , ..., xk ) = p(xk |xk1 , ..., x1 )p(xk1 |xk2 , ..., x1 ) ...p(x1 )
Then for n = k + 1 :
p(x1 , x2 , ..., xk , xk+1 ) = p(xk+1 |xk , xk1 , ..., x1 )p(xk , xk1 ..., x1 )
= p(xk+1 |xk , xk1 , ..., x1 )p(xk |xk1 , ..., x1 )p(xk1 |xk2 , ..., x1 ) ...p(x1 )
Hence the relationship holds for n = k + 1, and by induction it holds for any n.
c
12
Problem 2.16
1. Let T and R denote channel input and outputs respectively. Using Bayes rule we have
p(T = 0)p(R = A|T = 0)
p(T = 0)p(R = A|T = 0) + p(T = 1)p(R = A|T = 1)
0.4 16
=
0.4 61 + 0.6 13
1
=
4
p(T = 0|R = A) =
and therefore p(T = 1|R = A) = 43 , obviously if R = A is observed, the best decision would
be to declare that a 1 was sent, i.e., T = 1, because T = 1 is more probable that T = 0.
Similarly it can be verified that p(T = 0|R = B) = 47 and p(T = 0|R = C) = 14 . Therefore,
when the output is B, the best decision is 0 and when the output is C, the best decision is
T = 1. Therefore the decision function d can be defined as
(
1, R = A or C
d(R) =
0, R = B
This is the optimal decision scheme.
2. Here we know that a 0 is transmitted, therefore we are looking for p(error|T = 0), this is
the probability that the receiver declares a 1 was sent when actually a 0 was transmitted.
Since by the decision method described in part 1 the receiver declares that a 1 was sent when
R = A or R = C, therefore, p(error|T = 0) = p(R = A|T = 0) + p(R = C|T = 0) = 31 .
3. We have p(error|T = 0) = 31 , and p(error|T = 1) = p(R = B|T = 1) = 31 . Therefore, by the
total probability theorem
p(error) = p(T = 0)p(error|T = 0) + p(T = 1)p(error|T = 1)
1
1
= 0.4 + 0.6
3
3
1
=
3
Problem 2.17
Following the same procedure as in example 2-1-1, we prove :

yb
1
pX
pY (y) =
|a|
a
c
13
Problem 2.18
Relationship (2-1-44) gives :

pY (y) =
1
3a [(y b) /a]2/3
pX
"
yb
a
1/3 #
X is a gaussian r.v. with zero mean and unit variance : pX (x) = 12 ex

Hence :
2/3
1
21 ( yb
a )
e
pY (y) =
3a 2 [(y b) /a]2/3
2 /2
pdf of Y
0.5
0.45
0.4
a=2
0.35
b=3
0.3
0.25
0.2
0.15
0.1
0.05
0
10
0
y
10
Problem 2.19
1) The random variable X is Gaussian with zero mean and variance 2 = 108 . Thus p(X > x) =
Q( x ) and
4
10
4
p(X > 10 ) = Q
= Q(1) = .159
104

4 104
= Q(4) = 3.17 105
p(X > 4 104 ) = Q
104
p(2 104 < X 104 ) = 1 Q(1) Q(2) = .8182
2)

p(X > 104 , X > 0)
p(X > 104 )
.159
=
=
= .318
p(X > 104 X > 0) =
p(X > 0)
p(X > 0)
.5
c
14
Problem 2.20
1) y = g(x) = ax2 . Assume without loss of generality that a > 0. Then, if y < 0 the equation
2 has no real solutions and f (y) = 0. If y > 0 there are two solutions to the system, namely
y = axp
Y
x1,2 = y/a. Hence,
fY (y) =
=
=
fX (x2 )
fX (x1 )
+
|g (x1 )| |g (x2 )|
p
p
fX ( y/a) fX ( y/a)
p
p
+
2a y/a
2a y/a
y
1
e 2a2

2
ay 2
2) The equation y = g(x) has no solutions if y < b. Thus FY (y) and fY (y) are zero for y < b. If
b y b, then for a fixed y, g(x) < y if x < y; hence FY (y) = FX (y). If y > b then g(x) b < y
for every x; hence FY (y) = 1. At the points y = b, FY (y) is discontinuous and the discontinuities
equal to
FY (b+ ) FY (b ) = FX (b)
and
FY (b+ ) FY (b ) = 1 FX (b)
The PDF of y = g(x) is
fY (y) = FX (b)(y + b) + (1 FX (b))(y b) + fX (y)[u1 (y + b) u1 (y b)]

y2
b
1
= Q
e 22 [u1 (y + b) u1 (y b)]
((y + b) + (y b)) +
2 2
3) In the case of the hard limiter
p(Y = b) = p(X < 0) = FX (0) =
1
2
p(Y = a) = p(X > 0) = 1 FX (0) =
1
2
Thus FY (y) is a staircase function and

fY (y) = FX (0)(y b) + (1 FX (0))(y a)
c
15
4) The random variable y = g(x) takes the values yn = xn with probability
p(Y = yn ) = p(an X an+1 ) = FX (an+1 ) FX (an )
Thus, FY (y) is a staircase function with FY (y) = 0 if y < x1 and FY (y) = 1 if y > xN . The PDF
is a sequence of impulse functions, that is
fY (y) =
N
X
i=1
[FX (ai+1 ) FX (ai )] (y xi )
N h
a i
X
ai
i+1
Q
Q
(y xi )
=
i=1
Problem 2.21
For n odd, xn is odd and since the zero-mean Gaussian PDF is even their product is odd. Since
the integral
odd function over the interval [, ] is zero, we obtain E[X n ] = 0 for n odd.
R of an
n
Let In = x exp(x2 /2 2 )dx. Obviously In is a constant and its derivative with respect to x
is zero, i.e.,

Z
x2
x2
d
1
In =
nxn1 e 22 2 xn+1 e 22 dx = 0
dx
which results in the recursion
In+1 = n 2 In1
This is true forall n. Now let n = 2k 1, we will have I2k = (2k 1) 2 I2k2 , with the initial
condition I0 = 2 2 . Substituting we have
I2 = 2 2 2
I4 = 3 2 I2 = 3 4 2 2
I6 = 5 3 2 I4 = 5 3 6 2 2
I8 = 7 2 I6 = 7 5 3 8 2 2
.. ..
.=.
and in general if I2k = (2k 1)(2k 3)(2k 5)

3 1 2k 2 2 , then I2k+2 = (2k + 1)2 I2k =
(2k + 1)(2k 1)(2k 3)(2k 5) 3 1 2k+2 2 2 . Using the fact that E[X 2k ] = I2k / 2 2 ,
we obtain
In = 1 3 5 (n 1) n
for n even.
c
16
Problem 2.22
a. Since (Xr , Xi ) are statistically independent :

pX (xr , xi ) = pX (xr )pX (xi ) =
1 (x2r +x2i )/22

e
2 2
Also :
Yr + jYi = (Xr + Xi )ej
Xr + Xi = (Yr + jYi ) ej = Yr cos + Yi sin + j(Yr sin + Yi cos )
X = Y cos + Y sin
r
r
i
Xi = Yr sin + Yi cos
The Jacobian of the above transformation is :

Xr Xi
Y
cos sin
Yr
r
J =
=
Xi
Xr

sin cos
Yi
Yi
Hence, by (2-1-55) :

=1

pY (yr , yi ) = pX ((Yr cos + Yi sin ) , (Yr sin + Yi cos ))

2
2
2
1
e(yr +yi )/2
=
22
b. Y = AX and X = A1 Y
Now, pX (x) =
2
1
ex x/2
(22 )n/2
(the covariance matrix M of the random variables x1 , ..., xn is
M = 2 I, since they are i.i.d) and J = 1/| det(A)|. Hence :

pY (y) =
1 1
2
1
ey (A ) A y/2
2
n/2
| det(A)|
(2 )
For the pdfs of X and Y to be identical we require that :

| det(A)| = 1 and (A1 ) A1 = I = A1 = A
Hence, A must be a unitary (orthogonal) matrix .
Problem 2.23
Since we are dealing with linear combinations of jointly Gaussian random variables, it is clear
that Y is jointly Gaussian. We clearly have mY = E[AX] = AmX . This means that Y mY =
A (X mX ). Also note that

CY = E (Y mY )(Y mY ) = E A (X mX ) (X mX ) A
c
17
resulting in CY = ACX A .
Problem 2.24
a.
#
"n
n
h Pn
i
Y
Y

n
jvY
E ejvX = X (ejv )
ejvxi =
Y (jv) = E e
= E ejv i=1 xi = E
i=1
i=1
But,
pX (x) = p(x 1) + (1 p)(x) X (ejv ) = 1 + p + pejv
Y (jv) = 1 + p + pejv
n
b.
E(Y ) = j
dY (jv)
|v=0 = jn(1 p + pejv )n1 jpejv |v=0 = np
dv
and
E(Y 2 ) =

d2 Y (jv)
d
jv n1 jv
|
=
jn(1
p
+
pe
)
pe
= np + np(n 1)p
v=0
v=0
d2 v
dv
E(Y 2 ) = n2 p2 + np(1 p)
Problem 2.25
1. In the figure shown below

c
18
R=x 2
let us consider the region u > x, v > x shown as the colored region extending to infinity, call
u2 +v 2
this region R, and let us integrate e 2 over this region. We have

ZZ
ZZ
u2 +v 2
r2
e 2 r dr d
e 2 du dv =
R
r2
re
x 2
dr

2
r2
=
e
2
x 2
2
= ex
2
where we have used the fact that region R is included in the region outside the quarter circle
as shown in the figure. On the other hand we have
Z
Z
ZZ
2
2
2
v2
u2
u +v
2
e
e 2 dv
du dv =
du
e
x
=
=
Z
u2
du

2
2Q(x)
2
= 2 (Q(x))2
From the above relations we conclude that
2 (Q(x))2
and therefore, Q(x) 12 e
x2
2
x2
e
2
c
19
2. In
R
x
uv
y2
2
dy
y2
define u = e
y2
2
and dv =
dy
y2
and use the integration by parts relation

2
y2
v du. We have v = y1 and du = ye

Z
Now note that
u dv =
y2
x2
Z
2
dy e 2
e 2
y2
=
2Q(x)
e
dy =
2
y
y
x
x
2
y2
dy. Therefore
y2
2
dy
y2
> 0 which results in

x2
x2
e 2
1
e 2
2Q(x) > 0 Q(x) <
x
2x
On the other hand, note that

Z
which results in
or,
2 1+x
Q(x) >
x2
y2
1
dy
< 2
2
y
x
y2
dy =
2
Q(x)
x2
x2
e 2
2
2Q(x) < 2 Q(x)
x
x
x2
e 2
x
which results in
x2
x
e 2
Q(x) >
2(1 + x2 )
3. From
we have
x2
x2
x
1
e 2 < Q(x) <
e 2
2(1 + x2 )
2x
2
2
1
1
x2
x2
e
<
Q(x)
<
e
2( x1 + x)
2x
As x becomes large x1 in the denominator of the left hand side becomes small and the two
bounds become equal, therefore for large x we have
Q(x)
x2
1
e 2
2x
Problem 2.26
c
20
1. FYn (y) = P [Yn y] = 1 P [Yn > y] = 1 P [x1 > y, X2 > y, . . . , Xn > y] = 1R (P [X > y])n
A
where we have used the independence of Xi s in the last step. But P [X > y] = y A1 dy = Ay
A .
Therefore, FYn (y) = 1
(Ay)n
An ,
and fYn (y) =
d
dy FYn (y)
n1
= n (Ay)
An
, 0 < y < A.
2.
y n1
n
1
A
A
ny n

1
=
1 Ay
nA

y n
1
=
ey y > 0
1 Ay
n
f (y) =
Problem 2.27
i
h
(jv1 , jv2 , jv3 , jv4 ) = E ej(v1 x1 +v2 x2 +v3 x3 +v4 x4 )
E (X1 X2 X3 X4 ) = (j)4
4 (jv1 , jv2 , jv3 , jv4 )

|v1 =v2 =v3 =v4 =0
v1 v2 v3 v4
From (2-1-151) of the text, and the zero-mean property of the given rvs :
1
(jv) = e 2 v Mv
where v = [v1 , v2 , v3 , v4 ] , M = [ij ] .
We obtain the desired result by bringing the exponent to a scalar form and then performing
quadruple differentiation. We can simplify the procedure by noting that :
1
(jv)
= i ve 2 v Mv
vi
where i = [i1 , i2 , i3 , i4 ] . Also note that :

j v
vi
Hence :
= ij = ji
4 (jv1 , jv2 , jv3 , jv4 )

|V=0 = 12 34 + 23 14 + 24 13
v1 v2 v3 v4
c
21
Problem 2.28
1) By Chernov bound, for t > 0,

P [X ] et E[etX ] = et X (t)
This is true for all t > 0, hence
ln P [X ] min [t + ln X (t)] = max [t ln X (t)]
t0
t0
2) Here
ln P [Sn ] = ln P [Y n] max [tn ln Y (t)]
t0
where Y = X1 + X2 + + Xn , and Y (t) = E[eX1 +X2 ++Xn ] = [X (t)]n . Hence,

ln P [Sn ] = max n [t ln X (t)] = nI()
t0
1
P [Sn ] enI()
n
R
1
d
X (t) = 0 etx ex dx = 1t
as long as t < 1. I() = maxt0 (t + ln(1 t)), hence dt
(t + ln(1

1
1
t)) = 0 and t = . Since 0, t 0 and also obviously t < 1. I() = 1 + ln 1 =

1 ln , using the large deviation theorem
ln P [Sn ] = en(1ln )+o(n) = n en(1)+o(n)
Problem 2.29
For the central chi-square with n degress of freedom :

(jv) =
Now :
1
(1 j2v 2 )n/2
jn 2
d(jv)
d(jv)
=
|v=0 = n 2
E (Y ) = j
n/2+1
dv
dv
(1 j2v 2 )

d2 (jv)
2n 4 (n/2 + 1)
d2 (jv)
2
=
E
Y
=
|v=0 = n(n + 2) 2
dv 2
dv 2
(1 j2v 2 )n/2+2

The variance is Y2 = E Y 2 [E (Y )]2 = 2n 4
For the non-central chi-square with n degrees of freedom :
(jv) =
1
(1
j2v 2 )n/2
2
2
ejvs /(1j2v )
c
22
where by definition : s2 =
Pn
2
i=1 mi
"
#
2
2
d(jv)
jn 2
js2
=
+
ejvs /(1j2v )
n/2+1
n/2+2
2
2
dv
(1 j2v )
(1 j2v )
2
2
Hence, E (Y ) = j d(jv)
dv |v=0 = n + s
#
"
2
2
d2 (jv)
s2 (n + 4) 2 ns2 2
s4
n 4 (n + 2)
+
+
ejvs /(1j2v )
=
2
n/2+2
n/2+3
n/2+4
2
2
2
dv
(1 j2v )
(1 j2v )
(1 j2v )
Hence,

d2 (jv)
E Y2 =
|v=0 = 2n 4 + 4s2 2 + n 2 + s2
2
dv
and

Y2 = E Y 2 [E (Y )]2 = 2n 4 + 4 2 s2
Problem 2.30
The Cauchy r.v. has : p(x) =
a/
x2 +a2 ,
<x<
a.
E (X) =
since p(x) is an even function.
E X
Note that for large x,
x2
x2 +a2
xp(x)dx = 0
a
x p(x)dx =
=
x2
dx
x2 + a2
1 (i.e non-zero value). Hence,

E X 2 = , 2 =
b.
(jv) = E
jvX
a/ jvx
e dx =
x2 + a2
a/
ejvx dx
(x + ja) (x ja)
This integral can be evaluated by using the residue theorem in complex variable theory. Then, for
v0:

a/ jvx
(jv) = 2j
e
= eav
x + ja
x=ja
For v < 0 :
(jv) = 2j
a/ jvx
e
x ja
= eav v
x=ja
c
23
Therefore :
(jv) = ea|v|
Note: an alternative way to find the characteristic function is to use the Fourier transform relationship between p(x), (jv) and the Fourier pair :
eb|t|
1
c
, c = b/2, f = 2v
2
c + f2
Problem 2.31
Since R0 and R1 are independent fR0 ,R1 (r0 , r1 ) = fR0 (r0 )fR1 (r1 ) and
r 2 +r 2
2
r0 r1
r1 2 2 12 2 0
e
I
e
, r0 , r1 0
4
0
2
fR0 ,R1 (r0 , r1 ) =

0,
otherwise.
Now
P (R0 > R1 ) =
=
ZZ
r0 >r1
Z
f (r0 , r1 ) dr1 dr0

dr1
f (r0 , r1 ) dr0
Z

Z
fR0 (r0 ) dr0 dr1
fR1 (r1 )
=
r1
0
Z

Z
2
r0 r02
2
e
dr0 dr1
=
fR1 (r1 )
2
r1
0

Z
r2
02
2
fR1 (r1 ) e
=
dr1
0
=
=
Now using the change of variable y =
=
=
=
r1
P (R0 > R1 ) =
r1
r2
12
2
fR1 (r1 ) dr1
2
1
r1 r1 2 +2r
2
2
dr1
I
e
0
2
2
2r1 and letting s = 2 we obtain

Z
sy 2s2 +y2 dy
y
I0
e 22
2
2
2
2
0
Z

2
2
s
+y
y
1 s2
sy 2 2
e 2
I
e 2 dy
0
2
2
2
0
1 s22
e 2
2
1 22
e 4
2
c
24
where we have used the fact that

pdf.
R
0
y
2 I0
s +y
sy 2 2
2 e
2
dy = 1 because it is the integral of a Rician
Problem 2.32
1. The joint pdf of a, b is :

pab (a, b) = pxy (a mr , b mi ) = px (a mr )py (b mi ) =
1 12 [(amr )2 +(bmi )2 ]
e 2
2 2
a2 + b2 , = tan 1 b/a a = u cos , b = u sin The Jacobian of the transformation is

a/u a/
= u, hence :

: J(a, b) =

b/u b/
2. u =
u 12 [(u cos mr )2 +(u sin mi )2 ]

e 2
2 2
u 12 [u2 +M 2 2uM cos()]
e 2
2 2
pu (u, ) =
=
where we have used the transformation :
q
M = m2 + m2
m = M cos
r
r
i
= tan 1 m /m
mi = M sin
i
r
3.
pu (u) =
pu (u, )d
Z 2
2
1
u u2 +M
2
e 22 [2uM cos()] d
e 2
2
2
0
Z 2
2
u u2 +M
1
2
e 22
euM cos()/ d
2
2 0
2

u u2 +M
e 22 Io uM/ 2
2
=
=
=
c
25
Problem 2.33
a. Y =
1
n
Pn
i=1 Xi ,
Xi (jv) = ea|v|
n
n
h 1 Pn
in
h v i Y
h
i Y
Y (jv) = E ejv n i=1 Xi =
E ej n Xi =
Xi (jv/n) = ea|v|/n = ea|v|
i=1
i=1
b. Since Y (jv) = Xi (jv) pY (y) = pXi (xi ) pY (y) =
a/
y 2 +a2 .
, which is not Gaussian ; hence, the central limit theorem does not
c. As n , pY (y) = y2a/
+a2
hold. The reason is that the Cauchy distribution does not have a finite variance.
Problem 2.34

Since Z and Zej have the same pdf, we have hE[Z] = E Zeij = ej E[Z] for all . Putting

t
or E ZZ t = e2j E ZZ t , for
= gives E[Z] = 0. We also have E ZZ t = E Zej Zej

all . Putting = 2 gives E ZZ t = 0. Since Z is zero-mean and E ZZ t = 0, we conclude that
it is proper.
Problem 2.35
Using Equation 2.6-29 we note that for the zero-mean proper case if W = ej Z, it is suf1
1
z. But CW = [W W H ] =
w = z H CZ
ficient to show that det(CW ) = det(CZ ) and w H CW
1
1
j
j
H
H
zej =
w = ej z H CZ
E[e Ze Z ] = E[ZZ ] = CZ , hence det(CW ) = det(CZ ). Similarly, w H CW
1
z. Substituting into Equation 2.6-29, we conclude that p(w) = p(z).
z H CZ
Problem 2.36
Since Z is proper, we have E[(Z E(Z))(Z E(Z))t ] = 0. Let W = AZ + b, then

E[(W E(W ))(W E(W ))t ] = AE[(Z E(Z))(Z E(Z))t ]At = 0
c
26
hence W is proper.
Problem 2.37
We assume that x(t), y(t), z(t) are real-valued stochastic processes. The treatment of complexvalued processes is similar.
a.
Rzz ( ) = E {[x(t + ) + y(t + )] [x(t) + y(t)]} = Rxx ( ) + Rxy ( ) + Ryx ( ) + Ryy ( )
b. When x(t), y(t) are uncorrelated :

Rxy ( ) = E [x(t + )y(t)] = E [x(t + )] E [y(t)] = mx my
Similarly :
Ryx ( ) = mx my
Hence :
Rzz ( ) = Rxx ( ) + Ryy ( ) + 2mx my
c. When x(t), y(t) are uncorrelated and have zero means :

Rzz ( ) = Rxx ( ) + Ryy ( )
Problem 2.38
The power spectral density of the random process x(t) is :

Z
Rxx ( )ej2f d = N0 /2.
Sxx (f ) =
The power spectral density at the output of the filter will be :

Syy (f ) = Sxx (f )|H(f )|2 =
N0
|H(f )|2
2
c
27
Hence, the total power at the output of the filter will be :
Z
Z
N0
N0
Syy (f )df =
Ryy ( = 0) =
(2B) = N0 B
|H(f )|2 df =
2
2
Problem 2.39
2Bt
. From the result of
The power spectral density of X(t) corresponds to : Rxx (t) = 2BN0 sin2Bt
Problem 2.14 :

2
2
2
2
2 2 sin 2Bt
Ryy ( ) = Rxx (0) + 2Rxx ( ) = (2BN0 ) + 8B N0
2Bt
Also :
2
Syy (f ) = Rxx
(0)(f ) + 2Sxx (f ) Sxx (f )
The following figure shows the power spectral density of Y (t) :

(2BN0 )2 (f )
2N 2 B
0
ZZ

Z

Z

Z

Z

Z
2B
2B
Problem 2.40
X1
MX = E [(X mx )(X mx ) ] , X =
X2 , mx is the corresponding vector of mean values.
X3
Then :
MY = E [(Y my )(Y my ) ]
= E [A(X mx )(A(X mx )) ]
= E [A(X mx )(X mx ) A ]
= AE [(X mx )(X mx ) ] A
= AMx A
c
28
Hence :
MY =
11
11 + 13
422
11 + 31
11 + 13 + 31 + 33
Problem 2.41
Y (t) = X 2 (t), Rxx ( ) = E [x(t + )x(t)]

Ryy ( ) = E [y(t + )y(t)] = E x2 (t + )x2 (t)
Let X1 = X2 = x(t), X3 = X4 = x(t + ). Then, from problem 2.7 :
E (X1 X2 X3 X4 ) = E (X1 X2 ) E (X3 X4 ) + E (X1 X3 ) E (X2 X4 ) + E (X1 X4 ) E (X2 X3 )

Hence :
2
2
Ryy ( ) = Rxx
(0) + 2Rxx
( )
Problem 2.42
pR (r) =
2
(m)
We know that
Hence :

m m 2m1 mr 2 /
r
e
, X = 1 R

1
x
: pX (x) = 1/
p
.
R
1/
pX (x) =
2
2
1
2 m m 2m1 m(x)2 /
e
x
=
mm x2m1 emx
(m)
(m)
1/
Problem 2.43
The transfer function of the filter is :

H(f ) =
1
1
1/jC
=
=
R + 1/jC
jRC + 1
j2f RC + 1
c
29
a.
Sxx (f ) = 2 Syy (f ) = Sxx (f ) |H(f )|2 =
b.
Ryy ( ) = F
2
{Sxx (f )} =
RC
1 2
( RC )
2
(2RC)2 f 2 + 1
1
RC
(2f )2
ej2f df
Let : a = RC, v = 2f. Then :

Ryy ( ) =
2
2RC
a/ jv
2 a| |
2 | |/RC
e
dv
=
e
=
e
a2 + v 2
2RC
2RC
where the last integral is evaluated in the same way as in problem P-2.9 . Finally :

2
E Y 2 (t) = Ryy (0) =
2RC
Problem 2.44
If SX (f ) = 0 for |f | > W, then SX (f )ej2f a is also bandlimited. The corresponding autocorrelation

function can be represented as (remember that SX (f ) is deterministic) :
n
X
sin 2W 2W
n

RX (
RX ( a) =
a)
(1)
n
2W
2W 2W
n=
Let us define :
We must show that :

or
"
n
n sin 2W t 2W

X(
)
X(t) =
n
2W
2W
t
2W
n=
h
i
2
E |X(t) X(t)|
=0
m

X
m sin 2W t 2W

X(
E X(t) X(t)
X(t)
)
m
2W 2W t 2W
m=
!#
=0
(2)
First we have :
n
h

X
n m sin 2W t 2W
m i
m

RX (
E X(t) X(t) X(
) = RX (t
)
)
n
2W
2W
2W
2W t 2W
n=
c
30
But the right-hand-side of this equation is equal
of (1) with a = m/2W.
h to zero by application
i
Since this is true for any m, it follows that E X(t) X(t) X(t) = 0. Also
E
h
n

i
X
sin 2W t 2W
n

RX (
X(t) X(t) X(t) = RX (0)
t)
n
2W
2W t 2W
n=
Again, by applying (1) with a = t anf = t, we observe that the right-hand-side of the equation is
also zero. Hence (2) holds.
Problem 2.45
R
2
Q(x) = 12 x et /2 dt = P [N x] , where N is a Gaussian r.v with zero mean and unit variance.
From the Chernoff bound :

(1)
P [N x] evx E evN
where v is the solution to :

E N evN xE evN = 0
Now :
E evN
and
Hence (2) gives :
1
2
= ev
2 /2
= ev
2 /2
(2)
vt t2 /2 dt
e e
1
2
(tv)2 /2 dt
e

2
d
E evN = vev /2
E N evN =
dv
v = x
and then :
2
(1) Q(x) ex ex
2 /2
Q(x) ex
2 /2
Problem 2.46
Since H(0) =
h(n)
= 0 my = mx H(0) = 0
c
31
The autocorrelation of the output sequence is
Ryy (k) =
XX
i
h(i)h(j)Rxx (k j + i) =
x2
h(i)h(k + i)
i=
where the last equality stems from the autocorrelation function of X(n) :
2 , j = k + i
x
Rxx (k j + i) = x2 (k j + i) =
0,
o.w.
Hence, Ryy (0) = 6x2 , Ryy (1) = Ryy (1) = 4x2 , Ryy (2) = Ryy (2) = x2 , Ryy (k) = 0 otherwise.
Finally, the frequency response of the discrete-time system is :
H(f ) =
=
=
j2f n
h(n)e
1 2ej2f + ej4f
1 ej2f
2
= ej2f ejf ejf

= 4ejf sin 2 f
2
which gives the power density spectrum of the output :

Syy (f ) = Sxx (f )|H(f )|2 = x2 16 sin 4 f = 16x2 sin 4 f
Problem 2.47
|k|
1
R(k) =
2
c
32
The power density spectrum is
S(f ) =
=
=
j2f k
k= R(k)e
P1
k=

1 k j2f k
e
2
1 j2f k k
)
k=0 ( 2 e
1
1ej2f /2
2cos 2f
5/4cos 2f
3
54 cos 2f
k=0

1 k j2f k
e
2
1 j2f k
)
k=0 ( 2 e
1
1ej2f /2
Problem 2.48
We will denote the discrete-time process by the subscript d and the continuous-time (analog) process
by the subscript a. Also, f will denote the analog frequency and fd the discrete-time frequency.
a.
Rd (k) = E [X (n)X(n + k)]
= E [X (nT )X(nT + kT )]
= Ra (kT )
Hence, the autocorrelation function of the sampled signal is equal to the sampled autocorrelation
function of X(t).
b.
Rd (k) = Ra (kT ) =
=
=
=
j2f kT df
Sa (F )e
R (2l+1)/2T
l= (2l1)/2T
R 1/2T
l= 1/2T
R 1/2T P
1/2T
Sa (F )ej2f kT df
Sa (f +
l= Sa (f
l
j2F kT df
T )e

+ Tl ) ej2F kT df
c
33
Let fd = f T. Then :
Rd (k) =
1/2
1/2
"
1 X
Sa ((fd + l)/T ) ej2fd k dfd
T
(1)
l=
We know that the autocorrelation function of a discrete-time process is the inverse Fourier transform
of its power spectral density
Rd (k) =
Comparing (1),(2) :
Sd (fd ) =
1/2
1/2
Sd (fd )ej2fd k dfd
1 X
fd + l
)
Sa (
T
T
(2)
(3)
l=
c. From (3) we conclude that :

Sd (fd ) =
1
fd
Sa ( )
T
T
iff :
Sa (f ) = 0,
f : |f | > 1/2T
Otherwise, the sum of the shifted copies of Sa (in (3)) will overlap and aliasing will occur.
Problem 2.49
u(t) = X cos 2f t Y sin 2f t

E [u(t)] = E(X) cos 2f t E(Y ) sin 2f t
and :
Ruu (t, t + ) = E {[X cos 2f t Y sin 2f t] [X cos 2f (t + ) Y sin 2f (t + )]}

= E X 2 [cos 2f (2t + ) + cos 2f ] + E Y 2 [ cos 2f (2t + ) + cos 2f ]
E (XY ) sin 2f (2t + )
For u(t) to be wide-sense stationary, we must have : E [u(t)] =constant and Ruu (t, t + ) = Ruu ( ).
We note that if E(X) = E(Y ) = 0, and E(XY ) = 0 and E(X 2 ) = E(Y 2 ), then the above
requirements for WSS hold; hence these conditions are necessary. Conversely, if any of the above
c
34
conditions does not hold, then either E [u(t)] 6=constant, or Ruu (t, t + ) 6= Ruu ( ). Hence, the
conditions are also necessary.
Problem 2.50
a.
Ra ( ) =
=
=
j2f df
Sa (f )e
RW
ej2f df
sin 2W
By applying the result in problem 2.21, we have

Rd (k) = fa (kT ) =
b. If T =
1
2W
, then :
sin 2W kT
kT
2W = 1/T,
k=0
Rd (k) =
0,
otherwise
Thus, the sequence X(n) is a white-noise sequence. The fact that this is the minimum value of
T can be shown from the following figure of the power spectral density of the sampled process:
fs W
fs
fs + W
fs W
fs
fs + W
We see that the maximum sampling rate fs that gives a spectrally flat sequence is obtained when :
W = fs W fs = 2W T =
1
2W
c
35
|f |
c. The triangular-shaped spectrum S(f ) = 1
W , |f | W may be obtained by convolving the rectangular-shaped spectrum S1 (f ) = 1/ W , |f | W/2. Hence, R( ) = R12 ( ) =

1 sin W 2
.Therefore, sampling X(t) at a rate T1 = W samples/sec produces a white sequence
W
with autocorrelation function :

k=0
sin k 2 W,
1 sin W kT 2
=W
=
Rd (k) =
0, otherwise
W
kT
k
Problem 2.51
Lets denote : y(t) = fk (t)fj (t).Then :

Z
Z
fk (t)fj (t)dt =
y(t)dt = Y (f )|f =0
where Y (f ) is the Fourier transform of y(t). Since : y(t) = fk (t)fj (t) Y (f ) = Fk (f ) Fj (f ).

But :
Z
1 j2f k/2W
fk (t)ej2f t dt =
Fk (f ) =
e
2W
Then :
Y (f ) = Fk (f ) Fj (f ) =
and at f = 0 :
Y (f )|f =0 =
=
=
Fk (a) Fj (f a)da
Fk (a) Fj (a)da
R
1 2 j2a(kj)/2W
da
2W
e
1/2W, k = j
0,
k=
6 j
Problem 2.52
Beq =
1
G
|H(f )|2 df
For the filter shown in Fig. P2-12 we have G = 1 and

Z
|H(f )|2 df = B
Beq =
0
c
36
For the lowpass filter shown in Fig. P2-16 we have
H(f ) =
1
1
|H(f )|2 =
1 + j2f RC
1 + (2f RC)2
So G = 1 and
2
0 |H(f )| df
R
1
2
2 |H(f )| df
1
4RC
Beq =
=
=
where the last integral is evaluated in the same way as in problem P-2.9 .
Problem 2.53
a.
E [z(t)z(t + )] = E [{x(t + ) + jy(t + t)} {x(t) + jy(t)}]
= E [x(t)x(t + )] E [y(t)y(t + )] + jE [x(t)y(t + )]

+E [y(t)x(t + )]
= Rxx ( ) Ryy ( ) + j [Ryx ( ) + Rxy ( )]

But Rxx ( ) = Ryy ( )and Ryx ( ) = Rxy ( ). Therefore :
E [z(t)z(t + )] = 0
b.
V =
z(t)dt

E V2 =
E [z(a)z(b)] dadb = 0
0
from the result in (a) above. Also :
E (V V ) =
=
=
RT RT
0
RT RT
RT
0
E [z(a)z (b)] dadb

N0 (a b)dadb
N0 da = N0 T
c
37
Problem 2.54
E [x(t + )x(t)] = A2 E [sin (2fc (t + ) + ) sin (2fc t + )]

=
A2
2
cos 2fc
A2
2 E [cos (2fc (2t
+ ) + 2)]
where the last equality follows from the trigonometric identity :

sin A sin B = 12 [cos(A B) cos(A + B)] . But :
E [cos (2fc (2t + ) + 2)] =
=
Hence :
R 2
0
1
2
cos (2fc (2t + ) + 2) p()d

R 2
0 cos (2fc (2t + ) + 2) d = 0
E [x(t + )x(t)] =
A2
cos 2fc
2
Problem 2.55
1) We have E[Z(t)] = E[X(t)] + jE[Y (t)] = 0 + j0 = 0 and

RZ (t + , t) = E [(X(t + ) + jY (t + )) (X(t) jY (t))]
= RX ( ) + RY ( )
= 2RX ( )
because E [X(t + )Y (t)] = E [Y (t + )X(t)] = E[X(t + )]E[Y (t)] = 0 (by independence)
h
iand
f
1
=
N0 2W
therefore Z(t) is obviously stationary. We also note that RX ( ) = RY ( ) = F
2W N0 sinc(2W )
2) To compute
the power spectral density of Z(t), we have SZ (f ) = F [2RX ( )] = 2SX (f ) =

f
2N0 2W . Note that (t) is a rectangular pulse defined as
3) E[Zj ] = E
hR
1, |t| < 1
(t) = 21 , |t| = 1
0, otherwise.
i
R
(t) dt =
Z(t)R
j
E[Z(t)]Rj (t) dt = 0 since Z(t) is zero-mean. For the
c
38
correlation we have
E[Zj Zk ]
Z
Z(s)Rj (s) ds
Z (t)Rk (t) dt
=E
Z
Z
RZ (s t)Rj (s)Rk (t) dsdt
=

Z
Z
RZ (s t)Rj (s) ds dt ()
Rk (t)
=
R
Using Parsevals Theorem, x(t)y (t) dt = X(f )Y (f ) df , we have (Sj (f ) is the Fourier
transform of Rj (t)).

Z
Z
f
ej2f t 2N0
RZ (s t)Rj (s) ds =
Sj (f ) df
2W
Z W
a
N0 ej2f t Sj (f ) df
=2
ZW
b
N0 ej2f t Sj (f ) df
=2

f
where (a) is due to the fact that 2W
is zero outside the [W, W ] interval and (b) follows from
Rj (t) being bandlimited to [W, W ]. From above we have

Z
Z
j2f t
RZ (s t)Rj (s) ds = 2N0
e
Sj (f ) df
= 2N0 Rj (t)
Substituting this result in equation (**) we have

Z
N0 Rj (t)Rk (t) dt
E[Zj Zk ] = 2
(
2N0 , j = k
=
0,
j 6= k
This shows that Zj s are Gaussian random variables (since they are the result of linear operation
on a Gaussian process) with mean zero and variance 2N0 , i.e., Zj N (0, 2N0 ). Also note that for
j 6= k, Zj and Zk are independent since they are Gaussian and uncorrelated.
4) This is done similar to part 3 (lengthy but straightforward) and the result is that for any k, Zkr
2 ) = E(Z 2 ) = N and
and Zki are zero-mean, independent Gaussian random variables with E(Zkr
0
ki
therefore the random vector (Z1r , Z1i , Z2r , Z2i , , Znr , Zni ) is a 2n-dimensional Gaussian vector
with independent zero-mean components each having variance N0 . In standard notation
(Z1r , Z1i , Z2r , Z2i , , Znr , Zni ) N (0, N0 I)
where 0 is a 2n-dimensional zero vector and I is a 2n 2n identity matrix.
c
39
5) We have
E[Z(t)Z
k ] = E[(Z(t)
E[Z(t)Zk ]
where we have used

E[Zj Zk ]
N
X
Zj Rj (t))Zk ]
j=1
2N0 Rk (t)
(
2N0 , j = k
0,
j 6= k
Now we have
E[Z(t)Zk ]

Z
Z (s)Rk (s) ds
= E Z(t)
Z
RZ (t s)Rk (s) ds
=
Rk (s)RZ
(s t) ds
=

Z
f
j2f t
Sk (f )e
N0
=2
df
2W
Z W
2N0 Sk (f )ej2f t dt
=
W
Z
a
Sk (f )ej2f t df
= 2N0
= 2N0 Rk (t)
(a): because Rk (t) is bandlimited to [W, W ].
From above it follows that E[Z(t)Z

k ] = 0 for all k = 1, 2, , N . This means that the error term
is independent of the projections.
Problem 2.56
1. SX (f ) = | j sgn(f )|2 SX (f ) = SX (f ), hence RX ( ) = RX ( ).

X ( ).
2. SX X (f ) = SX (f )(j sgn(f )) = j sgn(f )SX (f ), therefore, RX X ( ) = R
h

i
+ ) X(t) j X(t)
3. RZ ( ) = E X(t + ) + j X(t
, expanding we have

RZ ( ) = RX ( ) + RX ( ) j RX X ( ) RXX
( )
c
40
X ( ) is an odd function (since it is
Using RX ( ) = RX ( ), and the fact that RX X ( ) = R
the HT of an even signal) we have RXX
( ) = RX X
( ), we have
( ) = RX X
X ( )
RZ ( ) = 2RX ( ) j2RX X ( ) = 2RX ( ) + j2R
Taking FT of both sides we have
SZ (f ) = 2SX (f ) + j2 (j sgn(f )SX (f )) = 2 (1 + sgn(f )) SX (f ) = 4SX (f )u1 (f )
4. We have
i
h
RXl (t + , t) = E Z(t + )ej2f0 (t+ ) Z (t)ej2f0 t
= ej2f0 RZ ( )
This shows that Xl (t) is WSS (we already know it is zero-mean). Taking FT, we have
SXl (f ) = SZ (f f0 ) = 4SX (f f0 )u
f0 ), this
1 (f j2f
shows that Xl (t) is lowpass. Also from
1
1
0
above RX ( ) = 2 Re [RZ (t)] = 2 Re RXl ( )e
. This shows that RXl ( ) is twice the LP
equivalent of RX ( ).
Problem 2.57
1) The power spectral density Sn (f ) is depicted in the following figure. The output bandpass
process has non-zero power content for frequencies in the band 49 106 |f | 51 106 . The
power content is
P
49106
10
51106
49106

8
= 10
51106
2
= 2 10
f
1+ 8
10
df +
51106
10
49106
49106

16 1 2
+ 10
+ 10
51106
f
1 8
10
df
51106
51106

16 1 2

x
x
10
2 49106
49106
8
!a10a
!
!
aa
!!
aa
!
!
aa
!
a
!
5107
5107
108
2) The output process N (t) can be written as

N (t) = Nc (t) cos(250 106 t) Ns (t) sin(250 106 t)
c
41
where Nc (t) and Ns (t) are the in-phase and quadrature components respectively, given by
(t) sin(250 106 t)
Nc (t) = N (t) cos(250 106 t) + N
(t) cos(250 106 t) N (t) sin(250 106 t)
Ns (t) = N
The power content of the in-phase component is given by
(t)|2 ] sin2 (250 106 t)
E[|Nc (t)|2 ] = E[|N (t)|2 ] cos2 (250 106 t) + E[|N
= E[|N (t)|2 ] = 2 102
(t)|2 ]. Similarly we find that E[|Ns (t)|2 ] =

where we have used the fact that E[|N (t)|2 ] = E[|N
2
2 10 .
3) The power spectral density of Nc (t) and Ns (t) is
S (f 50 106 ) + S (f + 50 106 ) |f | 50 106

N
N
SNc (f ) = SNs (f ) =
0
otherwise
SNc (f ) is depicted in the next figure. The power content of SNc (f ) can now be found easily as
Z 106
108 df = 2 102
PNc = PNs =
106
108
106
106
4) The power spectral density of the output is given by

SY (f ) = SX (f )|H(f )|2 = 106 (|f | 49 106 )(108 1016 |f |) for 49 106 |f | 51 106
Hence, the power content of the output is
Z 49106
6
PY = 10 (
(f 49 106 )(108 + 1016 f )df )
51106
Z 51106
+106 (
49106
(f 49 106 )(108 1016 f )df )
4
= 106 (2 104 102 )
3
The power spectral density of the in-phase and quadrature components of the output process is
given by

SYc (f ) = SYs (f ) = 106 ( (f + 50 106 ) 49 106 108 1016 (f + 50 106 ) )

+106 ( (f 50 106 ) 49 106 108 + 1016 (f 50 106 ) )
= 106 (2 1016 f 2 + 102 )
c
42
for |f | 106 and zero otherwise. The power content of the in-phase and quadrature component is
PYc = PYs
= 106
6
= 10
106
106
(2 1016 f 2 + 102 )df
106
106

2
f
+ 10 f
)
3 106
106
16 1 3
(2 10
4
= 106 (2 104 102 ) = PY
3
c

Digital Communications 5th Ed SolutionsChap2

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Digital Communications 5th Ed SolutionsChap2

Hochgeladen von

Copyright:

Verfügbare Formate

Solutions Manual

c. x(t) = cos 0 t, so its Fourier transform is : X(f ) = 21 [(f f0 ) + (f + f0 )] , f0 = 20 .

and Parsevals relation, we have

The corresponding residual error Ee is :

k=1 sk fk (t)s (t)dt

and, in general for the k-th function :

The first basis function is :

Then, for the second basis function :

s3 (t)g4 (t)dt = 1/ 3 g3 (t) = s3 (t) c43 g4 (t) =

s1 (t)g4 (t)dt = 2/ 3, c31 =

s1 (t)g3 (t)dt = 2/ 6, c21 = 0

Consider the set of signals enl (t) = jnl (t), 1 n N , then by

hn (t), m (t)i = 0 for all n 6= m

Let x(t) = m(t) cos

Hence, they are orthonormal.

a. As an orthonormal set of basis functions we consider the set

In matrix notation, the four waveforms can be represented as

Similarly we find that :

Thus, the minimum distance between any pair of vectors is dmin = 5.

As a set of orthonormal functions we consider the waveforms

The vector representation of the signals is

1. P (E2 ) = P (R2, R3, R4) = 3/7.

3. Here E4 = {R2, R4, B2, R1, B1} and P (E2 |E4 E3 ) =

The relationship holds for n = 2 (2-1-34) : p(x1 , x2 ) = p(x2 |x1 )p(x1 )

Following the same procedure as in example 2-1-1, we prove :

Relationship (2-1-44) gives :

X is a gaussian r.v. with zero mean and unit variance : pX (x) = 12 ex

p(Y = a) = p(X > 0) = 1 FX (0) =

Thus FY (y) is a staircase function and

[FX (ai+1 ) FX (ai )] (y xi )

which results in the recursion

and in general if I2k = (2k 1)(2k 3)(2k 5)

a. Since (Xr , Xi ) are statistically independent :

1 (x2r +x2i )/22

Xr + Xi = (Yr + jYi ) ej = Yr cos + Yi sin + j(Yr sin + Yi cos )

The Jacobian of the above transformation is :

pY (yr , yi ) = pX ((Yr cos + Yi sin ) , (Yr sin + Yi cos ))

(the covariance matrix M of the random variables x1 , ..., xn is

M = 2 I, since they are i.i.d) and J = 1/| det(A)|. Hence :

For the pdfs of X and Y to be identical we require that :

1. In the figure shown below

this region R, and let us integrate e 2 over this region. We have

and use the integration by parts relation

v du. We have v = y1 and du = ye

Now note that

> 0 which results in

On the other hand, note that

and fYn (y) =

4 (jv1 , jv2 , jv3 , jv4 )

where i = [i1 , i2 , i3 , i4 ] . Also note that :

4 (jv1 , jv2 , jv3 , jv4 )

1) By Chernov bound, for t > 0,

where Y = X1 + X2 + + Xn , and Y (t) = E[eX1 +X2 ++Xn ] = [X (t)]n . Hence,

t)) = 0 and t = . Since 0, t 0 and also obviously t < 1. I() = 1 + ln 1 =

For the central chi-square with n degress of freedom :

The Cauchy r.v. has : p(x) =

1 (i.e non-zero value). Hence,

fR0 ,R1 (r0 , r1 ) =

f (r0 , r1 ) dr1 dr0

fR1 (r1 ) dr1

2r1 and letting s = 2 we obtain

where we have used the fact that

dy = 1 because it is the integral of a Rician

1. The joint pdf of a, b is :