Solutions to Chapter 11 Exercises

August 21, 2012

Exercise 11.1
(a) The impulse response of the integrator is h(t) = rect( t−ttoo /2 ). Hence the
transfer function is
H(f ) = to sinc(f to )e−j2πf to .
The PSD of the output is
No t2o
SY,Y (f ) = SX,X (f )|H(f )|2 = sinc2 (f to ).
2
(b) Taking the inverse transforms of the above result
RY,Y (τ ) = F −1 [SY,Y (f )]
No to −1
= F [to sinc2 (f to )]
2
No to
= tri(t/to ).
2
The total power in the output is:
No to
PY = RY,Y (0) = .
2
(c)The noise equivalent bandwidth, Bneq , will satisfy
No t2o No to
· 2Bneq = PY =
2 2
1
⇒ Bneq = .
2to

1
Exercise 11.2
(a)

RY,Y (τ ) = E[Y (t)Y (t + τ )]

 
X(t) − X(t − to ) X(t + τ ) − X(t + τ − to )
= E ·
to to
E[X(t)X(t + τ )] E[X(t)X(t + τ − to )]
= −
t2o t2o
E[X(t − to )X(t + τ )] E[X(t − to )X(t + τ − to )]
− +
t2o t2o
2RX,X (τ ) − RX,X (τ + to ) − RX,X (τ − to )
= .
t2o
(b)
 
2RX,X (τ ) − RX,X (τ + to ) − RX,X (τ − to )
SY,Y (f ) = F [RY,Y (τ )] = F
t2o
2SX,X (f ) − SX,X (f )ej2πf to − SX,X (f )e−j2πf to
=
t2o
SX,X (f )
= 2
[2 − ej2πf to − e−j2πf to ]
to
SX,X (f )
= [2 − 2 cos(2πf to )]
t2o
4SX,X (f ) 2
= sin (πf to )
t2o
= (2πf )2 SX,X (f )sinc2 (f to ).

(c) As to → 0 the relationship betwen the input and output becomes Y (t) =
dX(t)
dt
which can be expressed in terms of the transfer function H(f ) = j2πf .
Hence we expect

SY,Y (f ) = |j2πf |2 SX,X (f ) = (2πf )2 SX,X (f ).

Since
lim sinc2 (f to ) = 1,
to →0

this is exactly the relationship we get from the result of part (b) as to → 0.

2
Exercise 11.3
RY,Y (τ ) = E[Y (t)Y (t + τ )] = E[cX(t) · cX(t + τ )] = c2 RX,X (τ ).

Exercise 11.4
(a)
RY,Y (τ ) = E[Y (t)Y (t + τ )]
= E[(X(t) + X(t − to ) + X(t − 2to ))
· (X(t + τ ) + X(t + τ − to ) + X(t + τ − 2to ))]
= 3RX,X (τ ) + 2RX,X (τ − to ) + 2RX,X (τ + to )
+ RX,X (τ − 2to ) + RX,X (τ + 2to ).
(b)
E[Y 2 (t)] = RY,Y (0)
= 3RX,X (0) + 2RX,X (−to ) + 2RX,X (to ) + RX,X (−2to ) + RX,X (2to )
= 3RX,X (0) + 4RX,X (to ) + 2RX,X (2to ).

Exercise 11.5
Y (t) = a(S 2 (t) + N 2 (t) + 2S(t)N (t))
E[Y (t)] = aE[S 2 (t)] + aE[N 2 (t)] + 2aE[S(t)N (t)]
= a(σS2 + σN 2
).
RY,Y (τ ) = a E[(S (t) + N 2 (t) + 2S(t)N (t))
2 2

(S 2 (t + τ ) + N 2 (t + τ ) + 2S(t + τ )N (t + τ ))]
= a2 E[S 2 (t)S 2 (t + τ )] + a2 E[N 2 (t)N 2 (t + τ )]
+ 4a2 E[S(t)S(t + τ )N (t)N (t + τ )] + a2 E[S 2 (t)N 2 (t + τ )]
+ 2a2 E[S 2 (t)S(t + τ )N (t + τ )] + a2 E[N 2 (t)S 2 (t + τ )]
+ 2a2 E[N 2 (t)N (t + τ )S(t + τ )] + 2a2 E[S(t)S 2 (t + τ )N (t)]
+ 2a2 E[S(t)N (t)N 2 (t + τ )]
= a2 [RS,S
2
(0) + 2RS,S2 2
(τ ) + RN,N (0) + 2RN,N2
(τ )
+ 4RS,S (τ )RN,N (τ ) + RS,S (0)RN,N (0) + RN,N (0)RS,S (0)]
= a2 [(RS,S (0) + RN,N (0))2 + 2(RS,S (τ ) + RN,N (τ ))2 ]
= a2 [(σS2 + σN 2 2
) + 2(RS,S (τ ) + RN,N (τ ))2 ].

3
Exercise 11.6
For the given PSD’s,
A2 −j2πfc τ A2
RS,S (τ ) = F −1 [SS,S (f )] = (e + ej2πfc τ ) = cos(2πfc τ )
4 2
No B
RN,N (τ ) = F −1 [SN,N (f )] = sinc(Bτ )(e−j2πfc τ + ej2πfc τ )
2
= No Bsinc(Bτ ) cos(2πfc τ )
From these we determine that
A2
σS2 = RS,S (0) = ,
2
2
σN = RN,N (0) = No B.
The autocorrelation of Y (t) is then
RY,Y (τ ) = a2 [(σS2 + σN2 2
) + 2(RS,S (τ ) + RN,N (τ ))2 ]
" 2 2 #
2
 2
A A
= a2 + No B + 2 + No Bsinc(Bτ ) cos2 (2πfc τ )
2 2
"
2
A2

= a2 + No B
2
 2  #
A
+ 2 + A2 No Bsinc(Bτ ) + (No B)2 sinc2 (Bτ ) (1 + cos(4πfc τ ))
4
Taking FT’s, the PSD of Y (t) is then
" 2
A2
 2 
2 A 2 2
SY,Y (f ) = a + No B δ(f ) + 2 δ(f ) + A No rect(f /B) + No Btri(f /B)
2 4
 #
1 1
∗ δ(f ) + δ(f − 2fc ) + δ(f + 2fc )
2 2
 2 
2 A
= a 2
+ A No B + (No B) δ(f ) + A2 No rect(f /B) + No2 Btri(f /B)
2
2
A2
    
2 f − 2fc f + 2fc
+ (δ(f − 2fc ) + δ(f + 2fc )) + A No rect + rect
4 B B
    
f − 2fc f + 2fc
+ No2 B tri + tri .
B B

4
Exercise 11.7
From example 10.5, the PSD of the random telegraph process is
A A c
SX,X (f ) = δ(f ) + · 2 .
4 4 c + (πf )2
The output PSD is given by

SY,Y (f ) = SX,X (f )|H(f )|2 .

For the filter given,

b
h(t) = be−at u(t) ↔ H(f ) = .
a + j2πf
Therefore, the output PSD is
b2
  
A A c
SY,Y (f ) = δ(f ) + · 2
4 4 c + (πf )2 a2 + (2πf )2
Ab2 Acb2 1
= 2
δ(f ) + · 2 .
4a 4 (c + (πf ) )(a2 + (2πf )2 )
2

Exercise 11.8
 
f
SX,X (f ) = F [RX,X (τ )] = F [2Afo sinc(2fo τ )] = Arect
2fo
 
f
H(f ) = F [h(t)] = F [2f1 sinc(2f1 t)] = rect
2f1
       
2 f 2 f f f
SY,Y (f ) = SX,X (f )|H(f )| = Arect rect = Arect rect .
2fo 2f1 2fo 2f1
Case I (f0 ≥ f1 ):
 
f
SY,Y (f ) = Arect ↔ RY,Y (τ ) = 2Af1 sinc(2f1 τ ).
2f1
Case I (f0 ≤ f1 ):
 
f
SY,Y (f ) = Arect ↔ RY,Y (τ ) = 2Afo sinc(2fo τ ).
2fo

5
Exercise 11.9
    
2 a 2 f − fc f + fc
SY,Y (f ) = SX,X (f )|H(f )| = ·b rect + rect ,
1 + (f /fo )2 f2 − f1 f2 − f1
where fc = (f1 + f2 )/2. Assuming f2 − f1 << fo , the input PSD will be
approximately constant over the passband of the filter. In that case, the
output PSD will be well approximated by

ab2
    
2 f − fc f + fc
SY,Y (f ) ≈ SX,X (fc )|H(f )| = · rect + rect
1 + (fc /fo )2 f2 − f1 f2 − f1
2
2ab (f2 − f1 )
⇒ = sinc((f2 − f1 )τ ) cos(ωc τ ).
1 + (fc /fo )2

Exercise 11.10
(a)
1
|H(f )|2 = = 1 + f 2 = (1 + jf )(1 − jf ).
SX,X (f )
Take H(f ) to be H(f ) = 1 + jf so that

1 d
h(t) = δ(t) + δ(t).
2π dt

(b)

_
_

y(t)
+
+
x(t )

6
Exercise 11.11
1 t 1
h(t) = exp(− )u(t) ↔ H(f ) = .
RC RC 1 + j2πf RC
(a) The crosscorrelation between the input and output is
τ
RX,Y (τ ) = RX,X (τ ) ∗ h(τ ) = exp(− )u(τ ).
RC
Therefore, RX,X (τ ) = 0 for τ < 0. As a result, Y (t) is independent of X(t1 )
for t < t1 .
(b) The autocorrelation of the output is found as follows:
1
SY,Y (f ) ∝ |H(f )|2 =
1 + (2πf RC)2
|τ |
⇒ RY,Y (τ ) ∝ exp(− ).
RC
Since RY,Y (τ ) is never zero, Y (t) is not independent of Y (t1 ) for any t.
(c) For the finite time integrator
 
t − to /2
h(t) = rect ↔ H(f ) = to sinc(f to )e−jπf to .
to

The crosscorrelation between the input and the output is

 
τ − to /2
RX,Y (τ ) = RX,X (τ ) ∗ h(τ ) = rect .
to

This will be zero when τ < 0 or when τ > to . Hence Y (t) is independent of
X(t1 ) for t < t1 or t > t1 + to . The autocorrelation of the output is found
according to:

SY,Y (f ) ∝ |H(f )|2 ∝ sinc2 (f to )

⇒ RY,Y (τ ) ∝ tri(τ /to ).

This will be zero when |τ | > to . Hence Y (t) is independent of Y (t1 ) when
|t − t1 | > to .

7
Exercise 11.12
(a)
RY1 ,Y2 (τ ) = E[Y1 (t)Y2 (t + τ )]
ˆ ˆ
= E[ h1 (u)N (t − u)du h2 (v)N (t + τ − v)dv]
ˆ ˆ
= h1 (u)h2 (v)E[N (t − u)N (t + τ − v)]dudv
ˆ ˆ
= h1 (u)h2 (v)RN,N (τ − v + u)dudv
ˆ ˆ
= h1 (u)h2 (v)δ(τ − v + u)dudv
ˆ
= h1 (u)h2 (u + τ )du
= h1 (−τ ) ∗ h2 (τ ).
(b)
SY1 ,Y2 (f ) = F [h1 (−τ ) ∗ h2 (τ )] = H1∗ (f )H2 (f ).
(c) The two processes are independent at the same sampling times if RY1 ,Y2 (0) =
0. From the results of part (a), this constraint is expressed in the time domain
as ˆ
RY1 ,Y2 (0) = 0 ⇒ h1 (t)h2 (t)dt = 0.

In words, the impulse responses of the filters must be orthogonal. Trans-

forming this to the frequency domain, the constraint becomes
ˆ ˆ
SY1 ,Y2 (f )df = 0 ⇒ H1∗ (f )H2 (f )df = 0.

Hence the transfer functions must be orthogonal.

(d) The two processes are independent at arbitrary sampling times if RY1 ,Y2 (τ ) =
0 for all τ . This leads to
ˆ
h1 (t)h2 (t + τ )dt = 0.

In the frequency domain, we must have SY1 ,Y2 (f ) = 0 for all f . This leads to
H1∗ (f )H2 (f ) = 0.
In this case, the transfer functions must be non-overlapping in frequency.

8
Exercise 11.13

RY1 ,Y2 (τ ) = E[Y1 (t)Y2 (t + τ )]

ˆ ˆ 
= E h1 (α)X1 (t − α)dα h2 (β)X2 (t − β)dβ
ˆ ˆ
= h1 (α)h2 (β)E[X1 (t − α)X2 (t − β)]dαdβ
ˆ ˆ
= h1 (α)h2 (β)RX1 ,X2 (τ + α − β)dαdβ.

Exercise 11.14
The given function is not a valid autocorrelation since |RX,X [k]| is not less
than RX,X [0] for all k.

Exercise 11.15

E[Z 2 [n]] = E[(X[n + i] − Y [n])2 ]

= E[X 2 [n + i]] + E[Y 2 [n]] − 2E[X[n + i]Y [n]]
= RX,X [0] + RY,Y [0] − 2RX,Y [−i].

Exercise 11.16
RX,Y [k] = h[k] ∗ RX,X [k].
Since RX,X [k] is given to be a discrete-time Delta function, the above sim-
plifies to
RX,Y [k] = h[k] = an u[n].

9
Exercise 11.17
∞
X
SX,X (f ) = RX,X [k]e−j2πf k
k=−∞
X∞
= ab|k| e−j2πf k
k=−∞
" ∞ 0
#
X X
= a −1+ bk e−j2πf k + b−k e−j2πf k
k=0 k=−∞
" #
1 1
= a −1+ −j2πf
+
1 − be 1 − bej2πf
" #
1 − b2
= a .
1 + b2 − 2b cos(2πf )

Exercise 11.18
Let X[n] be the input sequence and Y [n] the output sequence. Then

SY,Y (f ) = SX,X (f )|H(f )|2 .

Since SX,X (f ) is real and non-negative and |H(f )|2 is real and non-negative,
then SY,Y (f ) will be also.

Exercise 11.19
For this filter, the impulse response is
(
1/n, k = 0, 1, . . . , n − 1,
h [k] =
0, otherwise.

The corresponding transfer function is

−1
nP
1
H (f ) = e−j2πkf
n
k=0

10
(a) The output autocorrelation, RY,Y [k], is

RY Y [k] = RXX [k] ∗ h[k] ∗ h [−k]

= δ [k] ∗ h [k] ∗ h [−k]
= h [k] ∗ h [−k]
(
n−|k−n+1|
n2
, k = 0, 1, 2, . . . , 2n − 2,
=
0, otherwise.

(b) The output power spectral density is

SY Y (f ) = SXX (f ) | H (f ) |2 =| H (f ) |2
n−1
1 P
= | e−j2πkf |2
n
k=0
 
nP−1
1 
= n+2 (n − m) cos (2πmf )
n2
m=1

11
Exercise 11.20
Note: There is a typo in the problem statement. RXX [n] should be a|n| .
For a process whose autocorrelation is RXX [n] = a|n| , the corresponding power
spectral density is
1 − a2
SXX (f ) =
1 + a2 − 2a cos (2πf )
(a) In order for the output of the filter to be white we need
SY Y (f ) = SXX (f ) | H (f ) |2 = 1
Therefore
−1 1 + a2 − 2a cos (2πf )
| H (f ) |2 = SXX (f ) =
1 − a2
or in terms of z-transforms
1 + a2 − az − az −1
H (z) H z −1


=
1 − a2
This leads to
1 − az 1 − ae−j2πf
H (z) = √ or H (f ) = √ .
1 − a2 1 − a2

12
(b) The corresponding impulse response is
 1
 √1−a2 ,
 n = 0,
−a
h [n] = √
2, n = 1,
 1−a

0, otherwise.

Exercise 11.21
b
h(t) = be−at u(t) ↔ H(f ) = .
a + j2πf

ˆ ∞ ˆ ˆ
2 1 ∞ 2 1 ∞ 2
|H(f )| df = |H(f )| df = h (t)dt
0 2 −∞ 2 −∞
ˆ
b2 ∞ −2at b2
= e dt = .
2 0 4a
|H(0)|2 = b 2
ˆ ∞
1 1
⇒ Bneq = 2
|H(f )|2 df = .
|H(0| 0 4a

Exercise 11.22
4
H(f ) = ↔ h(t) = 4e−10t u(t).
10 + j2πf
The noise equivalent BW is found as follows:
ˆ ∞ ˆ ˆ
2 1 ∞ 2 1 ∞ 2
|H(f )| df = |H(f )| df = h (t)dt
2 −∞ 2 −∞
0
ˆ ∞
2
= 8 e−20t dt = .
0 5
4
|H(0)|2 =
25 ˆ ∞
1 5
⇒ Bneq = 2
|H(f )|2 df = .
|H(0| 0 2

The 3dB BW occurs when

1 5
|H(f3 )|2 = |H(0)|2 ⇒ f3 = .
2 π

13
The ratio is then
Bneq 5/2 π
= = .
f3 5/π 2

Exercise 11.23
SY,Y (f ) 2
SY,Y (f ) = SX,X (f )|H(f )|2 ⇒ |H(f )|2 = = .
SX,X (f ) (2πf ) + 102
2

(a) The 3dB bandwidth occurs when 2πf3 = 10 ⇒ f3 = 5/π.

(b) The gain at zero frequency is |H(0)|2 = 2/100 or -17dB.
(c) √ √
∗ 2 2
H(f )H (f ) = .
10 + j2πf 10 − j2πf
To make the filter causal, we choose
√
2 √
H(f ) = ↔ h(t) = 2e−10t u(t).
10 + j2πf

This can be achieved with an RC lowpass filter (and an attenuator with a

loss of 17dB).

Exercise 11.24
For a filter with an impulse response of h(t) = te−t u(t), the d.c. gain is
ˆ ∞ ˆ ∞
H (0) = h (t) dt = te−t dt = 1.
−∞ 0

Also, the energy in the transfer function (or impulse response) is

ˆ ∞ ˆ ∞ ˆ ∞
1
2
| H (f ) | df = 2
h (t) dt = t2 e−2t dt = .
−∞ −∞ 0 4

Therefore, the noise equivalent bandwidth is

ˆ ∞
1 2 1 1 1
Bneq = | H (f ) | df = · = Hz.
2|H(0)|2 −∞ 2 4 8

14
0.1 Exercise 11.25
(a) The filter is a BPF with a passband around f0 = 1kHz.
(b) For a BPF the noise equivalent bandwidth is
ˆ ∞
1
Bneq = | H (f ) |2 df.
| H (f0 ) |2 0
For the given filter, the passband gain is
1 + j2πf0
H (f0 ) =
(1 + j2πf0 )2 + (2nf0 )2
1 + ω02 1
| H (f0 ) |2 = 2
' since ω0  1
1 + 4ω0 4
The energy in the transfer function is
ˆ ∞ ˆ ∞ ˆ ∞
2 1 2 1 1 + ω2
| H (f ) | df = | H (ω) | dω = dω.
0 2π 0 2π 0 (1 + ω02 − ω 2 )2 + 4ω 2

The value of this integral (computed numerically) is 0.1244. ∴The bandwidth

is
0.1244
Bneq = = 0.4975 Hz.
1/4

Exercise 11.26
The d.c. gain is calculated as
ˆ ∞
H (0) = h (t) dt,
−∞

while the energy is (assuming h(t) is real)

ˆ ∞ ˆ ∞
2
| H (f ) | df = h2 (t) dt.
−∞ −∞

Then
´∞
−∞
h2 (t) dt
Bneq = h´ i2 .
∞
2 −∞
h (t) dt

15
Exercise 11.27
Since H (f ) = sinc2 (f ) ←→ h (t) = tri (t) ,then

H (0) = sinc2 (0) = 1,

ˆ ˆ ˆ
2 2
| H (f ) | df = h (t) dt = tri2 (t) dt =
2
3
2/3 1
⇒ Bneq = = Hz.
2·1 3

Exercise 11.28
For this filter, the transfer function and the impulse response are
j2πf RC 1
H(f ) = ↔ h(t) = δ(t) − exp(−t/RC)u(t).
1 + j2πf RC RC
The input mean is

µX (t) = E[A sin(ωc t + Θ) + N (t)] = 0.

Therfore the output mean is also zero. The input PSD is

A2 No
SX,X (f ) = (δ(f − fc ) + δ(f + fc )) + .
4 2
The PSD at the output of the filter is then

SY,Y (f ) = SX,X (f )|H(f )|2

 2
(2πf RC)2
 
A No
= (δ(f − fc ) + δ(f + fc )) +
4 2 1 + (2πf RC)2
A2 (2πfc RC)2 No (2πf RC)2
= (δ(f − f c ) + δ(f + f c )) + .
4 1 + (2πfc RC)2 2 1 + (2πf RC)2
Taking inverse Fourier Transforms, the corresponding autocorrelation func-
tion of the output is
A2 (2πfc RC)2
 
No 1
RY,Y (τ ) = cos(ωc τ ) + δ(τ ) − exp(−|τ /RC|)
4 1 + (2πfc RC)2 2 2RC
The first term in the expression above is due to the signal component at the
output and the second term is due to the noise component at the output.

16
The variance of the output is RY,Y (0) = ∞. While the output corresponding
to the signal part has a finite power given by

A2 (2πfc RC)2
PSo = ,
2 1 + (2πfc RC)2

the output due to the noise has infinite power and hence the SNR is zero.
The output autocorrelation function is shown in the figure

1.5

0.5
RY,Y(τ)

−0.5

−1

−1.5
−4 −3 −2 −1 0 1 2 3 4
τ

Exercise 11.29
For this circuit, the transfer function is

1 s 1 f2
H(s) = · 1 s ⇒ |H(f )|2 = .
C LC
+ RC
+ s2 (2πC)2 (fo2 − f 2 )2 + (f a/π)2

The corresponding impulse response is

1 ωo −at
h(t) = e cos(ωn t + φ)u(t),
C ωn

17
where
1
a = ,
2RC
r
1
ωo = ,
LC
p
ωn = ωo2 − a2 ,
φ = tan− 1(a/ωn ).

(a) To compute the autocorrelation of the output, we break the input into
the signal part and the noise part (noting that they are independent). Since
the signal input is a sinusoid, the signal output will be a sinusoid at the same
frequency scaled in amplituded by |H(fc )| and shifted in phase by ∠H(fc ).
Hence the signal part of the output is

so (t) = A|H(fc )| sin(ωc t + Θ + ∠H(fc )).

The autocorrelation of such a sinusoidal signal is

A2 |H(fc )|2
Rso ,so (τ ) = cos(ωc τ ).
2
The corresponding spectrum is:
A2 |H(fc )|2
Sso ,so (f ) = [δ(f − fc ) + δ(f − fc )].
4
The noise part of the output will have an autocorrelation function given by
ˆ ˆ ˆ
No No
Rno ,no (τ ) = δ(τ + α − β)h(α)h(β)dαdβ = h(t)h(t + τ )dt.
2 2
Plugging in the impulse response from above and caluclating the correlation
integral above (after lengthy algebraic calculations) results in
 2  
No −a|τ | ωo
Rno ,no (τ ) = e − a cos(ωn τ ) − a sin(ωn |τ |) .
8C 2 ωn2 a
The corresponding power spectrum is
No f2
Sno ,no (f ) = .
8π 2 C 2 (fo2 − f 2 )2 + (f b/π)2

18
The overall output power spectrum is the sum of the signal part and the
noise part. Hence,

A2 |H(fc )|2 No f2
So,o (f ) = [δ(f − fc ) + δ(f − fc )] + 2 2 2 .
4 8π C (fo − f 2 )2 + (f b/π)2

(b) The input PSD is

A2 No
SX,X (f ) = [δ(f − fc ) + δ(f + fc )] + .
4 2
Hence the output PSD is

A2 A2 No
SY,Y (f ) = SX,X (f )|H(f )|2 = |H(fc )|2 δ(f − fc ) + |H(−fc )|2 δ(f + fc ) + |H(f )|2
4 4 2
A2 No
= |H(fc )|2 [δ(f − fc ) + δ(f + fc )] + |H(f )|2
4 2
(c) The output power due to the signal, Ps , and the noise, Pn are:

A2 A2 A2 fc2
Ps = [|H(fc )|2 + |H(−fc )|2 ] = |H(fc )|2 = 2 2 2 ,
4 2 8π C (fo − fc2 )2 + (fc /2πRC)2
No ωo2 1 No ωo2 − a2
 
a No No R
Pn = Rno ,no (0) = − = = = .
8C 2 ωn2 a ωo2 8C 2 ωn2 a 8C 2 a 4C

The signal to noise ratio is then

Ps A 2 fc fc /2πRC
SN R = = .
Pn No π (fo − fc )2 + (fc /2πRC)2
2 2

This SNR is maximized by choosing (assuming fc 6= fo )

fc
RC = .
2π|fo2 − fc2 |

Exercise 11.30
Th signal output is
ˆ t ˆ t
so (t) = s(v)dv = e−v u(v)dv.
t−to t−to

19
At t = to , so (to ) = 1 − e−to and hence the instantaneous signal power at that
point in time is
Pso (to ) = (1 − e−to )2 .
The noise output is ˆ t
No (t) = N (u)du.
t−to

The corresponding noise power is

ˆ t ˆ t
PNo = RNo ,No (0) = E[No2 (t)] = E[N (u)N (v)]dudv
t−to t−to
ˆ t ˆ t ˆ
No No t No to
= δ(u − v)dudv = dv = .
2 t−to t−to 2 t−to 2
Therefore the SNR is
PSo 2 (1 − e−to )2
SN R = = .
PN o No to
This expression is optimized when to = 1.26 (this was determined numerically
using MATLAB).

Exercise 11.31
The signal out of the filter will be
ˆ ∞
s0 (t) = si (t − u) h (u) du.
−∞

With h(t) a one-sided exponential and si (t) a square pulse, then

(
t1 (1 − exp (−t/t1 )) , 0 ≤ t < t0 ,
s0 (t) =
t1 (exp (t0 /t1 ) − 1) exp (−t/t1 ) , t > t0 .

The maximum value of the signal component is

s0 (t0 ) = t1 (1 − exp (−t0 /t1 ))

∴ The maximum signal power is

s20 (t0 ) = t21 (1 − exp (−t0 /t1 ))2

20
The noise power (at any point in time) will be
ˆ ˆ ˆ
N0 ∞ N0 ∞ 2 u ∞
 
2 2t N0 t1
Pn = | H (f ) | df = h (t) dt = exp − dt = .
2 −∞ 2 ∞ 2 0 t1 4
∴ the maximum signal-to-noise ratio is

t21 (1 − exp (−t0 /t1 ))2

SN R = N0 t1
∝ t1 (1 − exp (−t0 /t1 ))2
4

To find the optimal value of signal-to-noise ratio, take dtd1 SNR= 0 and solve
for t1
  2      
d t0 t0 t0 t0
SNR = 1 − exp − + 2t1 1 − exp − exp − −
dt1 t1 t1 t1 t12
       
t0 t0 2t0 t0
= 1 − exp − 1 − exp − − exp −
t t1 t1 t
  1       1
t0 2t0 t0
= 1 − exp − 1− 1+ exp − =0
t1 t1 t1
The first term will be zero when t1 = ∞and represents a minima of the
signal-to-noise
ratio. The second term will be zero when
 
2t0 t0
1+ = exp
t1 t1
This occurs when
t0
= 1.2565
t1
⇒ t1 = 795.9µ sec

Exercise 11.32
In both cases, since the impulse response of the filter is symmetrical about
zero (even), the output of the filter (signal part) will be maximum at time
t = 12 . At this point in time the signal output will be
  ˆ  
1 1
s0 = h t− s (t) dt
2 2

21
(a) For the rectangular filter the signal output will be (assuming ta < 1)
  ˆ ta + 1  
1 2 2 2 πta
s0 = sin (πt) dt = sin
2 1
2
− ta
2
π 2

while the noise output power will be

ˆ
N0 ∞ 2 N0 ta
Pn = h (t) dt = .
2 −∞ 2
The signal-to-noise ratio is then
2 2
sin2 πta


 
8 πta
SNR = π
N0 ta
2
∝ sin2
2
N0 ta π 2 2

The signal-to-noise ratio will be maximum when

sin πt2a


d
SNR = (πta cos (πta /2) − sin (πta /2)) = 0
dta t2a
While there are many solutions to this equation, we are interested in the one
where0 < ta < 1which occurs when
 
πta
πta = tan ⇒ ta = 0.7420
2
Note: There is no need to consider the case when ta > 1since the signal power
does not increase as ta increases beyond ta = 1while the noise power does.
(b) For the triangular filter, the signal output will be (assuming tb > 12 )
ˆ 1
 t − 21 
       
1 2 1 1 1
s0 = sin (πt) 1 −   dt = 1− −
2 0 tb  π tb 2 π
while the noise power will be
ˆ ˆ   2
N0 ∞ 2 N0 tb t 2tb
Pn = h (t) dt = 1 −   dt = .
2 −∞ 2 −tb tb 3
The signal-to-noise ratio is then

h
2 2 1 1 1

i2
− −


2
π
1 tb 2 π 12 tb − 21 − π1
SN R = = .
N0 tb /3 N0 π 2 t3b

22
The maximum value occurs at tb = 3 21 − π1 = 0.5451.

(c) The ratio of the two signal-to-noise ratios is

2
12(tb −( 12 − π1 ))

2
SN Rb N0 π 2 t3
b
3 ta tb − 12 − π1
= = 3
SN Ra 8 sin2 (πta /2) 2 tb sin2 (πta /2)
N0 π 2 ta

Using the optimal values of ta = 0.7420 & tb = 0.5451 leads to

SN Rb
= 1.0746 = 0.31dB
SN Ra
∴ The triangular filter produces an output signal-to-noise ratio that is 0.3dB
larger than the rectangular filter.

Exercise 11.33
The matched filter impulse response and the output of this filter when the
signal is at the input are shown in the figure.

1 3

2.5
MF Impulse Response

0.5 2

1.5
MF output

0 1

0.5

−0.5 0

−0.5
−1 −1
0 1 2 3 0 2 4 6
t t

Exercise 11.34
h(t) = s(to − t). In this case, s(to − t) = s(t) so the impulse response of the
matched filter is the same as the signal itself.

23
Exercise 11.35
In order to simplify this problem, we decompose the matched filter into
two stages. That is the impulse response of the matched filter is expressed
as h(t) = h1 (t) ∗ h2 (t). Equivalently, in the frequency domain, H(f ) =
H1 (f )H2 (f ). The input to the first filter is s(t) + N (t) where the noise has
some (non-white) PSD given by, SN,N (f ). Express the output of the first
stage of the matched filter (input to the second stage) as s1 (t) + N1 (t) where
s1 (t) = s(t) ∗ h1 (t),
N1 (t) = N (t) ∗ h1 (t),
SN1 ,N1 (f ) = SN,N (f )|H1 (f )|2 .
the transfer function of the first stage is chosen such that |H1 (f )| ∝
If p
1/ SN,N (f ), then the noise process N1 (t) will be white. In that case, the
input to the second stage is simply a known signal s1 (t) plus white noise.
Hence the second stage should be matched to the signal s1 (t). That is,
h2 (t) = s1 (to − t) ⇔ H2 (f ) = S1∗ (f )e−j2πf to = S ∗ (f )H1∗ (f )e−j2πf to
Putting these two stages together, we get the overall matched filter response
to be
S ∗ (f ) −j2πf to
H(f ) = H1 (f )H2 (f ) = S ∗ (f )|H1 (f )|2 e−j2πf to = e .
SN,N (f )

Exercise 11.36
´ t2
The noise output is N = N (t) dt. The variance is
0
ˆ t2 ˆ t2 
σ2
 2
= E N =E N (t) N (u) dtdu
0 0
ˆ t2 ˆ t2
= E [N (t) N (u)] dtdu
0
ˆ ˆ
0
ˆ
N0 t2 t2 N0 t2 N0 t2
= δ (t − u) dtdu = dt = .
2 0 0 2 0 2
For the case of t2 < t,the signal output will be s0 = t2 and the signal-to-noise
ratio is
t22 2t2
SNR = N0 t2
=
2
N0

24
For the case of t2 > t1, the signal output will be s0 = t1 and the signal-to-noise
ratio is
t21 2t21
SNR = N0 t2
=
2
N0 t2
2t1
When t2 = t1 (optimal case) the signal-to-noise ratio is N0
. The loss due to
the incorrect integration time is
(
SN Rt2 t2 /t1 , t2 < t1 ,
=
SN Rt1 t1 /t2 , t2 > t1 .

This loss will be within 0.25dB if 94.4µ sec < t2 < 105.9µ sec.

Exercise 11.37
(a) The Wiener-Hopf equation is the same as (11.59) in the text with X(t)
replaced with Z(t). In this case, the W-H equation is
ˆ ∞
h(u)RZ,Z (τ − u)du = RZ,Z (τ ).
∞

(b)
2 2
RZ,Z (τ ) = e−|τ k ↔ SZ,Z (f ) = = .
1 + (2πf )2 (1 + j2πf )(1 − j2πf )
The transfer function of the whitening filter is then
1
H1 (f ) = √ (1 + j2πf ).
2
The crosscorrelation between the input and output of the whitening filter is
then found according to
√
∗ 2 1 2
SZ,Z̃ (f ) = SZ,Z (f )H1 (f ) = 2
√ (1 − j2πf ) =
1 + (2πf ) 2 1 + j2πf
√ −τ
RZ,Z̃ (τ ) = 2e u(τ ).

The impulse response of the second stage is then chosen according to

√
h2 (τ ) = RZ,Z̃ (τ )u(τ − to ) = 2e−τ u(τ − to ).

25
The corresponding transfer function is
√ −t
2e o −j2πf to
H2 (f ) = e .
1 + j2πf
Putting the two stages together produces
H(f ) = H1 (f )H2 (f ) = e−to e−j2πf to ↔ h(t) = e−to δ(t − to ).
(c) The MSE is given by
E[2 (t)] = E[(Z(t + to ) − Z(t))2 ]
= RZ,Z (0)(1 + e−2to ) − 2e−to RZ,Z (to )
= 1 + e−2to − 2e−2to
= 1 − e−2to .

Exercise 11.38
(a) In this problem, the error in the estimator is
(t1 ) = Z(t1 ) − Y (t1 ) = Z(t1 ) − aZ(t0 ) − bZ(t2 ).
Using the orthogonality principle, the error must be orthogonal to the ob-
served data. Therefore
E[Z(to )(t1 )] = E[Z(to )(Z(t1 ) − aZ(t0 ) − bZ(t2 ))]
= R(t1 − t0 ) − aR(0) − bR(t2 − t0 ) = 0
E[Z(t2 )(t1 )] = E[Z(t2 )(Z(t1 ) − aZ(t0 ) − bZ(t2 ))]
= R(t2 − t1 ) − aR(t2 − t0 ) − bR(0) = 0
To simplify the remaining calculations, define the following vectors and ma-
trices:
 
Z(t0 )
z = ,
Z(t2 )
 
a
w = ,
b
 
R(t1 − t0 )
p = E[Z(t1 )z] = ,
R(t2 − t1 )
 
T R(0) R(t2 − t0 )
R = E[zz ] =
R(t2 − t0 ) R(0)

26
Then the equations above developed from the orthogonality principle reduce
to the matrix equation Rw = p with the resulting solution w = R−1 p.

(b) The MSE is

E[2 (t1 )] = E[(Z(t1 ) − zT w)2 ]

= R(0) − 2pT w + wT Rw
= R(0) − pT w
= R(0) − pT R−1 p.

Exercise 11.39
(a) In this problem, the error in the estimator is

(t2 ) = Z(t2 ) − Y (t2 ) = Z(t2 ) − aZ(t0 ) − bZ(t1 ).

Using the orthogonality principle, the error must be orthogonal to the ob-
served data. Therefore

E[Z(to )(t2 )] = E[Z(t0 )(Z(t2 ) − aZ(t0 ) − bZ(t1 ))]

= R(t2 − t0 ) − aR(0) − bR(t1 − t0 ) = 0
E[Z(t1 )(t2 )] = E[Z(t1 )(Z(t2 ) − aZ(t0 ) − bZ(t1 ))]
= R(t2 − t1 ) − aR(t1 − t0 ) − bR(0) = 0

To simplify the remaining calculations, define the following vectors and ma-
trices:
 
Z(t0 )
z = ,
Z(t1 )
 
a
w = ,
b
 
R(t2 − t0 )
p = E[Z(t2 )z] = ,
R(t2 − t1 )
 
T R(0) R(t1 − t0 )
R = E[zz ] =
R(t1 − t0 ) R(0)

Then the equations above developed from the orthogonality principle reduce
to the matrix equation Rw = p with the resulting solution w = R−1 p.

27
(b) Using R(τ ) = c exp(−dτ ) where c and d are positive constants, we get

   
R(t2 − t0 ) exp(−d(t1 − t0 ))
p = = c exp(−d(t2 − t1 )) ,
R(t2 − t1 ) 1
   
R(0) R(t1 − t0 ) 1 exp(−d(t1 − t0 ))
R = =c ,
R(t1 − t0 ) R(0) exp(−d(t1 − t0 )) 1
 
−1 0
w = R p= .
exp(−d(t2 − t1 ))
Hence, a = 0 and the more distant sample should not be used.

Exercise 11.40
(a) For the smoothing problem, the transfer function of the Wiener filter is
(see (11.53) in the text)
SS,S (f )
H(f ) = .
SS,S (f ) + SN,N (f )
For the given spectra, this works out to be
2
50+(2πf )2 2
H(f ) = 2 = .
40 + 50+(2πf )2
2002 + 40(2πf )2
The corresponding impulse response is
r
1 2002
h(t) = √ exp(− |t|).
40 · 2002 40
(b) The PSD of the output of the Weiner filter is

SY,Y (f ) = SX,X (f )|H(f )|2

 2
SS,S (f )
= (SS,S (f ) + SN,N (f ))
SS,S (f ) + SN,N (f )
2
SS,S (f )
=
SS,S (f ) + SN,N (f )
 2
2
50+(2πf )2
= 2 .
40 + 50+(2πf )2

28
Exercise 11.41
For a discrete time process whose autocorrelation is of the form
A
R[k] = p|k| ,
1 − p2
the corresponding PSD is
A
S(f ) = .
1+ p2 + 2p cos(2πf )
Hence the signal and noise PSDs are
10 50
SS,S (f ) = = ,
1.01 + 0.2 cos(2πf ) 5.05 + cos(2πf )
100 200
SN,N (f ) = =
1.0625 + 0.5 cos(2πf ) 2.125 + cos(2πf ).
a) The transfer function of the Wiener filter is (see (11.53) in the text)

SS,S (f )
H(f ) = .
SS,S (f ) + SN,N (f )
For the given spectra, this works out to be
106.25 + 50 cos(2πf )
H(f ) = .
1116.25 + 250 cos(2πf )
b) The PSD of the output of the Weiner filter is

SY,Y (f ) = SX,X (f )|H(f )|2

 2
SS,S (f )
= (SS,S (f ) + SN,N (f ))
SS,S (f ) + SN,N (f )
2
SS,S (f )
=
SS,S (f ) + SN,N (f )
 2
50
5.05+cos(2πf )
= 50 200 .
5.05+cos(2πf )
+ 2.125+cos(2πf ).

29
c) The input signal and noise powers are given by

10
PSin = RS,S [0] = = 10.101 = 10.04dB,
1 − (1/10)2
100
PNin = RN,N [0] = = 106.667 = 20.28dB,
1 − (1/4)2

Hence, the input SNR is 10.04 − 20.28 = −10.24dB.

The signal out of the filter has a PSD given by
SSo ,So (f ) = SS,S (f )|H(f )|2
and the resulting signal output power is
ˆ 1/2
PSout = SS,S (f )|H(f )|2 df.
−1/2

Similarly, for the noise component,

ˆ 1/2
PNout = SN,N (f )|H(f )|2 df.
−1/2

We used MATLAB to calculate these integrals resulting in

PSout = 0.0854 = −10.67dB, PNout = 12516 = 40.97dB.
Hence the output SNR is −10.67 − 40.97 = −51.66dB. The Wiener filter has
substantially reduced the SNR.

Exercise 11.42
(a)
SXX (f ) = SY Y (f ) = L.P. {SN N (f − f0 ) + SN N (f + f0 )}
= tri (f /B) .
(b)
SXY (f ) = jL.P. {SN N (f − f0 ) + SN N (f + f0 )}
(
j 1 − Bf ,


0 < f < B,
= f


j −1 + B , −B < f < 0.

30
Exercise 11.43
(a)

N 2 (t) = (X (t) cos (ω0 t) − Y (t) sin (ω0 t))2 =

= X 2 (t) cos2 (ω0 t) − Y 2 (t) sin2 (ω0 t) − 2X (t) Y (t) sin (ω0 t) cos (ω0 t)
   
2 1 1 2 1 1
= X (t) + cos (2ω0 t) + Y (t) − cos (2ω0 t) − X (t) Y (t) sin (2ω0 t)
2 2 2 2
After the low- pass filter removes the 2ω0 terms, the output is
1 1 1 1
Z (t) = L.P. N 2 (t) = X 2 (t) + Y 2 (t) = |X (t) + jY (t)|2 = |GN (t)|2

2 2 2 2
(b)
  
1 2 1 2 1 2 1 2
RZZ (t, t + t) = E X (t) + Y (t) X (t + τ ) + Y (t + τ )
2 2 2 2
1  2  1 
E X (t) X 2 (t + τ ) + E X 2 (t) Y 2 (t + τ )

=
4 4
1  2 1
E X (t + τ ) Y 2 (t) + E Y 2 (t) Y 2 (t + τ )
  
+
4 4
Using the Gaussian moment factoring theorem we find

E X 2 (t) X 2 (t + τ ) = RXX
2
(0) + 2RXX (τ ) = B 2 + Bsinc2 (Bτ )
 

E Y (t) Y 2 (t + τ ) = RY2 Y (0) + 2RY Y (τ ) = B 2 + Bsinc2 (Bτ )

 2 

E X 2 (t) Y 2 (t + τ ) = RXX (0) RY Y (0) + 2RXY (t) = B 2 + RXY (τ )

 

E X 2 (t + τ ) Y (t) = RXX (0) RY Y (0) − 2RXY (τ ) = B 2 − RXY (τ )

 

31
Adding these results and dividing by 4 results in
B
RZZ (t, t + t) = B 2 + sinc2 (B t)
2

Exercise 11.44

E [GZ (t) GZ (t + t)] = E [(X [t] + jY (t)) (X (t + t) + jY (t + t))]

= RXX (t) − RY Y (t) + jRXY (t) + jRXY (-t)

Since RXX (t) = RY Y (t)and RXY (t)is odd, then E [GZ (t) GZ (t + t)] = 0.

32