Professional Documents
Culture Documents
Analog
Communications
Problems and Solutions
Analog Communications
Kasturi Vasudevan
Analog Communications
Problems and Solutions
123
Kasturi Vasudevan
Electrical Engineering
Indian Institute of Technology Kanpur
Kanpur, Uttar Pradesh, India
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2021
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of
illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publishers, the authors, and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publishers nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publishers remain neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
To my family
Preface
vii
viii Preface
In spite of my best efforts, some errors might have gone unnoticed. Suggestions
for improving the book are welcome.
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
xi
About the Author
xiii
Notation
xv
Chapter 1
Signals and Systems
Using the properties of the Dirac-delta function, prove the inverse Fourier trans-
form, that is, ∞
g(t) = G( f ) exp (j 2π f t) d f. (1.2)
f =−∞
δ(t) 1
∞
⇒ exp (j 2π f t) d f = δ(t)
f =−∞
∞
⇒ exp (j 2π f (t − x)) d f = δ(t − x). (1.4)
f =−∞
© The Editor(s) (if applicable) and The Author(s), under exclusive license 1
to Springer Nature Switzerland AG 2021
K. Vasudevan, Analog Communications,
https://doi.org/10.1007/978-3-030-50337-6_1
2 1 Signals and Systems
Thus proved.
2. (Simon Haykin 1983) Consider a pulse-like function g(t) that consists of a small
number of straight-line segments. Suppose that this function is differentiated with
respect to time twice so as to generate a sequence of weighted delta functions as
shown by
d 2 g(t)
= ki δ(t − ti ), (1.6)
dt 2 i
(b) Using the above procedure find the Fourier transform of the pulse in Fig. 1.1.
d 2 g(t)
= g1 (t) (j 2π f )2 G( f ) = G 1 ( f ) (say). (1.9)
dt 2
Consequently
G1( f )
G( f ) = − . (1.10)
4π 2 f 2
−tb −ta ta tb
1 Signals and Systems 3
Therefore
1 −j 2π f ti
G( f ) = − ki e . (1.12)
4π 2 f 2 i
d 2 g(t) A
= (δ(t + tb ) − δ(t + ta ) − δ(t − ta ) + δ(t − tb )) . (1.13)
dt 2 tb − ta
A
G( f ) = − (exp(j 2π f tb ) − exp(j 2π f ta )
(tb − ta )4π 2 f 2
− exp(−j 2π f ta ) + exp(−j 2π f tb ))
A
⇒ G( f ) = − (cos (2π f tb ) − cos (2π f ta ))
(tb − ta )2π 2 f 2
A
⇒ G( f ) = sin (π f (tb + ta )) sin (π f (tb − ta )) , (1.14)
(tb − ta )π 2 f 2
−tb −ta ta tb
dg(t)/dt
A/(tb − ta )
t
d2 g(t)/dt2
A/(tb − ta )
t
4 1 Signals and Systems
0 T0
2T
3. (Simon Haykin 1983) Using the above procedure compute the complex Fourier
series representation of the periodic train of triangular pulses shown in Fig. 1.3.
Here T < T0 /2. Hence also find the Fourier transform of g p (t).
• Solution: Consider the pulse:
g p (t) for −T0 /2 < t < T0 /2
g(t) = (1.16)
0 elsewhere.
Clearly
d 2 g(t) A
= (δ(t + T ) − 2δ(t) + δ(t − T )) . (1.17)
dt 2 T
Hence
A
G( f ) = − (exp(j 2π f T ) − 2 + exp(−j 2π f T ))
4π 2 f 2T
AT
= sin2 (π f T )
π2 f 2 T 2
= AT sinc2 ( f T ). (1.18)
where
1 Signals and Systems 5
1 ∞
cn = g(t) exp (−j 2πnt/T0 ) dt
T0 t=−∞
1
= G(n/T0 ). (1.20)
T0
exp (j 2π f c t) δ( f − f c ) (1.22)
∞
AT n
G p( f ) = sinc2 (nT /T0 )δ f − . (1.23)
T0 n=−∞ T0
where we have made use of the fact that sinusoids integrated over one period
is equal to zero. In computing the autocorrelation, we have lost the phase
information.
Rg (τ ) |G( f )|2
d Rg (τ )
⇒ j 2π f |G( f )|2 . (1.31)
dτ
1 Signals and Systems 7
Let
d Rg (τ )
= g1 (τ ). (1.32)
dτ
Substituting for g1 (t) and G 1 ( f ) in the above equation we get the required
result.
7. A sinusoidal signal of the form A cos(2π f 0 t), A > 0, is full wave rectified to
obtain g p (t).
(a) Using the complex Fourier series representation of g p (t) and the Parseval’s
power theorem, compute
2
2 2 2
S=9 1 + 2 + ··· + + · · · , (1.35)
π 3 (1 − 4n 2 )2
where
T1 /2
1
cn = g p (t) exp(−j 2πnt/T1 ) dt (1.37)
T1 −T1 /2
Hence
T1 /2
A
cn = |cos(2π f 0 t)| exp(−j 2πnt/T1 ) dt
T1 −T1 /2
T1 /2
2A
= 2 cos(2π f 0 t) cos(2πnt/T1 ) dt
2T1 0
A T1 /2
= [cos(2π( f 0 − n/T1 )t) + cos(2π( f 0 + n/T1 )t)] dt
T1 0
A
= [sin(2π( f 0 − n/T1 )T1 /2)/(2π( f 0 − n/T1 ))
T1
+ sin(2π( f 0 + n/T1 )T1 /2)/(2π( f 0 + n/T1 ))]
AT1
= [cos(nπ)/(2π( f 0 T1 − n))
T1
+ cos(nπ)/(2π( f 0 T1 + n))]
A(−1)n
= [1/(1 − 2n) + 1/(1 + 2n)]
π
2 A(−1)n
= . (1.39)
π(1 − 4n 2 )
c−n = cn . (1.40)
where we have used (1.40). Comparing (1.35) with the last equation of (1.41),
we obtain
A = 3. (1.42)
S = 4.5. (1.44)
2
g p1 (t) = 2cn exp( j 2πnt/T1 ), (1.45)
n=−2
n=0
where we have assumed that the gain of the bandpass filter is 2 and cn is given
by (1.39). The signal power at the bandpass filter output is
2
P= 4 |cn |2
n=−2
n=0
= 3.372. (1.46)
8. (Simon Haykin 1983) Determine the autocorrelation of the Gaussian pulse given
by
1 πt 2
g(t) = exp − 2 . (1.47)
t0 t0
e−πt e−π f
2 2
1 1
⇒ e−πt e−π f .
2 2
(1.48)
t0 t0
G 2 ( f ) = e−2π f
2 2
t0
= ψg ( f ) (say). (1.50)
10 1 Signals and Systems
−W 0 W
Once again we use the time scaling property of the Fourier transform, namely,
1
√ e−πt /(2t0 ) ,
2 2
Rg (t) = (1.52)
|t0 | 2
which is the required autocorrelation of the Gaussian pulse. Observe the |t0 |
in the denominator of (1.52), since Rg (0) must be positive.
9. (Rodger and William 2002) Assume that the Fourier transform of x(t) has the
shape as shown in Fig. 1.4. Determine and plot the spectrum of each of the
following signals:
(a) x1 (t) = (3/4)x(t)
+ (1/4)j x̂(t)
(b) x2 (t) =
(3/4)x(t) + (3/4)j x̂(t) e j 2π f0 t
(c) x3 (t) = (3/4)x(t) + (1/4)j x̂(t) e j 2πW t ,
where f 0 W and x̂(t) denotes the Hilbert transform of x(t).
• Solution: We know that
Therefore
X 1 ( f ) = (3/4)X ( f ) + (1/4)sgn ( f )X ( f )
⎧
⎨ X( f ) for f > 0
= (3/4)A for f = 0 . (1.54)
⎩
(1/2)X ( f ) for f < 0
Fig. 1.5 X 1 ( f ) X1 (f )
A
3A/4
A/2
−W 0 W
Fig. 1.6 X 2 ( f ) X2 (f )
3A/2
3A/4
f0 f0 + W
Then
M( f ) = (3/4)X ( f ) + (3/4)sgn ( f )X ( f )
⎧
⎨ (3/2)X ( f ) for f > 0
= (3/4)A for f = 0 . (1.56)
⎩
0 for f < 0
Then
M( f ) = (3/4)X ( f ) + (1/4)sgn( f ) X ( f )
⎧
⎨ X( f ) for f > 0
= (3/4)A for f = 0 . (1.59)
⎩
(1/2)X ( f ) for f < 0
12 1 Signals and Systems
Fig. 1.7 X 3 ( f ) X3 (f )
A
3A/4
A/2
0 W 2W
find the output by convolving the complex envelopes. Assume that the impulse
response of the filter is of the form:
where gc (t) and gs (t) extend over the frequency range [−W, W ], where W
f c . The complex envelope of g(t) is defined as
Thus
x(t) = x̃(t)e j 2π f0 t . (1.68)
In order to facilitate computation of the filter output using the complex envelopes,
it is given that the filter impulse response is to be represented by
1 −αt
h̃(t) = αe u(t). (1.71)
2
Thus
h(t) = 2h̃(t)e j 2π f0 t . (1.72)
Since u(t − τ ) = 0 for t < τ , and since −T /2 < τ < T /2, it is clear that ỹ(t) =
0, and hence y(t) = 0 for t < −T /2.
Now, for −T /2 < t < T /2, we have
14 1 Signals and Systems
α t
ỹ(t) = e j 2π f τ e−α(t−τ ) dτ
2 τ =−T /2
t
α
= e−αt eτ (α+j 2π f ) dτ
2 τ =−T /2
α
= e−αt et (α+j 2π f ) − e−T /2(α+j 2π f )
2(α + j 2π f )
α
j 2π f t
= e − e−α(t+T /2)−j π f T . (1.75)
2(α + j 2π f )
Let
2π f
θ1 = tan−1
α
θ2 = π f T
r = α2 + (2π f )2 . (1.76)
Y (f )
38.4
19.2
4
f
Fig. 1.8 Y ( f )
Sketch the spectrum at the output labeling all the important frequencies and
amplitudes.
• Solution: The Fourier transform of the output is
Y ( f ) = X ( f ) + 0.2X ( f ) X ( f ). (1.82)
The spectrum of Y ( f ) can be found out by inspection and is shown in Fig. 1.8.
12. (Simon Haykin 1983) Let R12 (τ ) denote the cross-correlation function of two
energy signals g1 (t) and g2 (t).
(a) Using Fourier transforms, show that
∞
(m+n)
R12 (τ ) = (−1) n
h 1 (t)h ∗2 (t − τ ) dt, (1.83)
t=−∞
where
denote the mth and nth derivatives of g1 (t) and g2 (t), respectively.
16 1 Signals and Systems
g1 (t) g2 (t)
2
2
t t
−3 0 3
−2
−3 −1 0 3 1
(b) Use the above relation with m = 1 and n = 0 to evaluate and sketch the
cross-correlation function R12 (τ ) of the pulses g1 (t) and g2 (t) shown in
Fig. 1.9.
Let
Similarly
Using the fact that multiplication in the frequency domain is equivalent to con-
volution in the time domain, we get
(m+n)
R12 (t) = (−1)n h 1 (t) h ∗2 (−t)
∞
(m+n)
⇒ R12 (τ ) = (−1) n
h 1 (t)h ∗2 (t − τ ) dt. (1.88)
t=−∞
Hence proved. For the signal in Fig. 1.9, using m = 1 and n = 0, we get
g2 (t)
2
t
0
−2
(1)
R12 (τ )
4
τ
−4
R12 (τ )
−8
−6 −4 −2 6 4 2 0
−3 −1 1 3
Thus
∞
(1)
R12 (τ ) = h 1 (t)h 2 (t − τ ) dt
t=−∞
= 2g2 (−τ − 3) − 2g2 (3 − τ ). (1.90)
(1)
R12 (τ ) and R12 (τ ) are plotted in Fig. 1.10.
13. Let R12 (τ ) denote the cross-correlation function of two energy signals g1 (t) and
g2 (t).
(a) Using Fourier transforms, show that
∞
(m+n)
R12 (τ ) = (−1) n
h 1 (t)h ∗2 (t − τ ) dt, (1.91)
t=−∞
18 1 Signals and Systems
1 1
t t
0 1
0 1 2 2
−1
where
denote the mth and nth derivatives of g1 (t) and g2 (t), respectively.
(b) Use the above relation with m = 1 and n = 0 to evaluate and sketch the
cross-correlation function R12 (τ ) of the pulses g1 (t) and g2 (t) shown in
Fig. 1.11.
Let
Similarly
Hence proved. For the signal in Fig. 1.11, using m = 1 and n = 0, we get
1 Signals and Systems 19
Thus
∞
(1)
R12 (τ ) = h 1 (t)h 2 (t − τ ) dt
t=−∞
∞
= [δ(t) + δ(t − 1) − 2δ(t − 2)] g2 (t − τ )
t=−∞
= g2 (−τ ) + g2 (1 − τ ) − 2g2 (2 − τ ). (1.98)
(1)
R12 (τ ) and R12 (τ ) are plotted in Fig. 1.12.
g2 (t)
1
t
−1
(1)
R12 (τ )
2
1
τ
−3
R12 (τ )
1
τ
−2
−2 −1 0 1 2
14. (Simon Haykin 1983) Let x(t) and y(t) be the input and output signals of a linear
time-invariant filter. Using Rayleigh’s energy theorem, show that if the filter is
stable and the input signal x(t) has finite energy, then the output signal y(t) also
has finite energy.
• Solution: Let H ( f ) denote the Fourier transform of h(t). We have
∞
H( f ) = h(t)e−j 2π f t dt
t=−∞
∞
⇒ |H ( f )| = h(t)e −j 2π f t
dt
t=−∞
∞
≤ h(t)e−j 2π f t dt
t=−∞
∞
⇒ |H ( f )| ≤ |h(t)| dt, (1.99)
t=−∞
Let
where we have used the Rayleigh’s energy theorem. Using the fact that
Y ( f ) = H ( f )X ( f ) (1.103)
Since the input signal has finite energy, so does the output signal.
1 Signals and Systems 21
15. (Simon Haykin 1983) Prove the following properties of the complex exponential
Fourier series representation, for a real-valued periodic signal g p (t):
(a) If the periodic function g p (t) is even, that is, g p (−t) = g p (t), then the Fourier
coefficient cn is purely real and an even function of n.
(b) If g p (t) is odd, that is, g p (−t) = −g p (t), then cn is purely imaginary and
an odd function of n.
(c) If g p (t) has half-wave symmetry, that is, g p (t ± T0 /2) = −g p (t), where T0
is the period of g p (t), then cn consists of only odd order terms.
• Solution: The Fourier series for any periodic signal g p (t) is given by
∞
g p (t) = cn e j 2πnt/T0
n=−∞
∞
= a0 + 2 an cos(2πnt/T0 ) + bn sin(2πnt/T0 ), (1.105)
n=1
where
⎧
⎨ an − j bn for n > 0
cn = a0 for n = 0 (1.106)
⎩
a−n + j b−n for n < 0.
Note that an and bn are real-valued, since g p (t) is real-valued. To prove the
first part, we note that
∞
g p (−t) = cn e−j 2πnt/T0 . (1.107)
n=−∞
−∞
g p (−t) = c−m e j 2πmt/T0
m=∞
∞
= c−m e j 2πmt/T0 . (1.108)
m=−∞
Since g p (−t) = g p (t), comparing (1.105) and (1.108) we must have c−m = cm
(even function of m). Moreover, from (1.106), cm must be purely real.
To prove the second part, we note from (1.105) and (1.108) that c−m = −cm
(odd function of m) and moreover from (1.106) it is clear that cm must be
purely imaginary.
To prove the third part, we note that
22 1 Signals and Systems
∞
g p (t ± T0 /2) = cn e j 2πn(t±T0 /2)/T0
n=−∞
∞
= cn (−1)n e j 2πnt/T0 . (1.109)
n=−∞
16. (Simon Haykin 1983) A signal x(t) of finite energy is applied to a square-law
device whose output y(t) is given by
x(t) X ( f ). (1.111)
Therefore
∞
Y( f ) = X (α)X ( f − α) dα
α=−∞
W
= X (α)X ( f − α) dα, (1.112)
α=−W
−W ≤ f −α≤ W (1.113)
−W 0 W
X(−2W − α)
f = −2W
−3W −W 0 W
X(2W − α)
f = 2W
−3W −W 0 W 3W
X(−α)
f =0
−W 0 W
Now
f −B f +B
X ( f ) = A rect − A rect
2B 2B
2 AB sinc(2Bt)e j 2π Bt − 2 AB sinc(2Bt)e−j 2π Bt
⇒ X ( f ) j 4 AB sinc(2Bt) sin(2π Bt). (1.116)
Note that
∞
x̂(t) dt = −j A = X̂ (0) = 0. (1.119)
t=−∞
18. Consider the system shown in Fig. 1.15. Assume that the current in branch AB
is zero. The voltage at point A is v1 (t).
(a) Find out the time-domain expression that relates v1 (t) and vi (t).
1Ω v1 (t)
A B vo (t) = dv1 (t)/dt
+
2F d
vi (t) = 3 cos(6πt) i(t) dt
(volts)
−
(b) Using the Fourier transform, find out the relation between V1 ( f ) and Vi ( f ).
(c) Find the relation between Vo ( f ) and Vi ( f ).
(d) Compute the power dissipated when the output voltage vo (t) is applied
across a 1 resistor.
However
dv1 (t)
i(t) = C . (1.122)
dt
Thus
dv1 (t)
vi (t) = RC + v1 (t). (1.123)
dt
Taking the Fourier transform of both sides, we get
Vi ( f ) = RCj 2π f V1 ( f ) + V1 ( f )
Vi ( f )
⇒ V1 ( f ) = . (1.124)
1 + j 2π f RC
It is given that
dv1 (t)
vo (t) =
dt
⇒ Vo ( f ) = j 2π f V1 ( f )
j 2π f Vi ( f )
= . (1.125)
1 + j 2π f RC
Now
3
Vi ( f ) = [δ( f − 3) + δ( f + 3)] . (1.126)
2
Therefore
3 j 6π 3 −j 6π
Vo ( f ) = δ( f − 3) + δ( f + 3) . (1.127)
2 1 + j 12π 2 1 − j 12π
0.5 H
v1 (t)
A B
vo (t) = dv1 (t)/dt
+
+
d
i(t) dt
vi (t) = 5 sin(4πt)
2Ω
(volts)
−
−
2 × 9 36π 2
P = |c1 |2 + |c2 |2 = ≈ 9/8 W. (1.129)
4 1 + 144π 2
19. Consider the system shown in Fig. 1.16. Assume that the current in branch AB
is zero. The voltage at point A is v1 (t).
(a) Find out the time-domain expression that relates vi (t) and i(t).
(b) Using the Fourier transform, find out the relation between V1 ( f ) and Vi ( f ).
(c) Find the relation between Vo ( f ) and Vi ( f ).
(d) Compute vo (t).
(e) Compute the power dissipated when the output voltage vo (t) is applied
across a 1/2 resistor.
Vi ( f ) = R I ( f ) + j 2π f L I ( f )
Vi ( f )
⇒ I( f ) = . (1.131)
R + j 2π f L
Therefore
V1 ( f ) = R I ( f )
Vi ( f )
=
1 + j 2π f L/R
Vi ( f )
= , (1.132)
1 + j π f /2
1 Signals and Systems 27
Since
5
Vi ( f ) = [δ( f − 2) − δ( f + 2)] (1.134)
2j
we get
2π (−2π)
Vo ( f ) = 5δ( f − 2) − 5δ( f + 2) . (1.135)
1 + jπ 1 − jπ
where
10π
c1 =
1 + jπ
= Aej φ
10π
c2 =
1 − jπ
= Ae−j φ , (1.137)
where
10π
A= √
1 + π2
φ = − tan−1 (π). (1.138)
P = |c1 |2 + |c2 |2
= 2 A2
200π 2
= W. (1.140)
1 + π2
20. Consider a complex-valued signal g(t). Let g1 (t) = g ∗ (−t). Let g1(n) (t) denote
the nth derivative of g1 (t). Consider another signal g2 (t) = g (n) (t).
Is g2∗ (−t) = g1(n) (t)? Justify your answer using Fourier transforms.
• Solution: Let G( f ) denote the Fourier transform of g(t). Then we have
Therefore
g1(n) (t) (j 2π f )n G 1 ( f )
⇒ g1(n) (t) (j 2π f )n G ∗ ( f ). (1.142)
21. (Simon Haykin 1983) Consider N stages of the RC-lowpass filter as illustrated
in Fig. 1.17.
(a) Compute the magnitude response |Vo ( f )/Vi ( f )| of the overall cascade con-
nection. The current drawn by the buffers is zero.
R R R
V1 (f ) V2 (f ) V2 (f ) Vo (f )
V1 (f )
Vi (f ) Buffer Buffer
C C C
However,
dv2 (t)
i(t) = C . (1.146)
dt
Therefore (1.145) can be rewritten as
dv2 (t)
v1 (t) = RC + v2 (t). (1.147)
dt
Taking the Fourier transform of both sides, we get
V1 ( f ) = (j 2π f RC + 1)V2 ( f )
V2 ( f ) 1
⇒ = . (1.148)
V1 ( f ) 1 + j 2π f RC
where
(n) d n f (x)
f (0) = . (1.153)
d x n x=0
1
f (x) = (1.154)
(1 + 4π 2 τ02 x) N /2
with x = f 2 . We have
f (0) = 1
−N 4π 2 τ02
(1)
f (0) =
2 (1 + 4π 2 τ02 x) N /2+1 x=0
−N
= (4π 2 τ02 )
2
2 2 2
(2) N N 4π τ0
f (0) = +1
2 2 (1 + 4π 2 τ02 x) N /2+2
x=0
2
N 2 2
≈ 4π τ0 , (1.155)
2
N N
+1≈ . (1.156)
2 2
Generalizing (1.155), we can obtain the nth derivative of the Maclaurin series
as
N N N
f (n) (0) = (−1)n + 1 ... +n−1
2 2 2
2 2 n
4π τ0
×
(1 + 4π 2 τ02 x) N /2+n
x=0
n
N
≈ (−1)n 4π 2 τ02 , (1.157)
2
N N
+n−1≈ . (1.158)
2 2
Thus f (x) in (1.154) can be written as
1 Signals and Systems 31
|H(f )|
(b)
4
3
f
∠H(f )
(c)
π
−π
y y2
f (x) = 1 − x + x2 − · · ·
1! 2!
= exp (−x y) , (1.159)
where
Finally, substituting for x we get the desired result which is Gaussian. Observe
that, in order to satisfy the condition n N , the nth term of the series in
(1.159) must tend to zero for n N . This can happen only when
xy < 1
⇒ (N /2)4π R C f 2 < 1
2 2 2
1
⇒ |f| < √ , (1.161)
π RC 2N
∞
g p (t) a0 δ( f ) + an [δ( f − n/T ) + δ( f + n/T )] , (1.165)
n=1
g(t)
gp (t)
(a)
2
t
−5T 5T
16
0 16
T
Rg (τ )
(b)
10T /4
τ
−10T 10T
16
0 16
T
Rgp (τ )
(c)
10/4
1
τ
−10T 6T 10T
16
0 16 16
T
Rgp (τ ) (resultant)
(d)
10/4
1
τ
−10T 6T 10T
16
0 16 16
T
g p (t) for −T /2 < t < T /2
g(t) = (1.166)
0 elsewhere.
∞
1
Rg p (τ ) = Rg (τ + mT ), (1.167)
T m=−∞
Clearly, the filter output is a dc component plus the first and second harmonics.
The gain of the dc component is 4 and the phase shift is 0. Hence, the output
dc signal is 4a0 = 5.
34 1 Signals and Systems
0 T /2 T
Similarly, the gain of the first harmonic is also 4 and the phase shift is −2π/5.
Finally, the gain of the second harmonic is 3 and the phase shift is −4π/5.
Thus, the filter output is
16
x(t) = 5 + sin(5π/8) cos(2πt/T − 2π/5)
π
6
+ sin(10π/8) cos(4πt/T − 4π/5). (1.168)
π
23. Consider the periodic waveform g p (t) given in Fig. 1.20, where T denotes the
period.
(a) Compute the real Fourier series representation of g p (t). Give the expression
for the coefficient of the nth term.
(b) Compute the Fourier transform of g p (t).
Similarly,
T−
1
an = g p (t) cos(2πnt/T ) dt
T t=0−
T−
1
= [2δ(t) + δ(t − T /2)] cos(2πnt/T ) dt
T t=0−
1
= [2 + cos(nπ)] for n ≥ 1. (1.171)
T
The Fourier transform of g p (t) is given by
∞
g p (t) a0 δ( f ) + an [δ( f − n/T ) + δ( f + n/T )] , (1.172)
n=1
24. If g(t) has the Fourier transform G( f ), compute the inverse Fourier transform
of g(a f − f 0 ) for
(a) a = 0,
(b) a = 0.
• Solution: From the duality property of the Fourier transform, we know that
g( f ) G(−t). (1.173)
In other words
∞
G(−t) = g( f ) e j 2π f t d f. (1.174)
f =−∞
Let
a f − f0 = x
⇒ a d f = d x. (1.176)
|a| d f = −d x. (1.177)
36 1 Signals and Systems
g(t) G( f )
1
⇒ h(t) = g(at) H ( f ) = G( f /a)
|a|
1
⇒ h(t − t0 /a) = g(at − t0 ) H ( f )e−j 2π f t0 /a = G( f /a)e−j 2π f t0 /a .
|a|
(1.180)
Since
p( f ) P(−t)
1
⇒ g(a f − f 0 ) G(−t/a)ej 2πt f0 /a . (1.181)
|a|
25. (Simon Haykin 1983) Let Rg (t) denote the autocorrelation of an energy signal
g(t).
(a) Using Fourier transforms, show that
∞
Rg(m+n) (τ ) = (−1) n
g2 (t)g1∗ (t − τ ) dt, (1.182)
t=−∞
A
t
T 2T 3T 4T
(b) Using the above relation, evaluate and sketch the autocorrelation of the signal
in Fig. 1.21. You can use m = 1 and n = 0.
Similarly let
Thus, using the fact that multiplication in the frequency domain is equivalent
to convolution in the time domain, we get
Rg(m+n) (t) = (−1)n g2 (t) g1∗ (−t)
∞
⇒ Rg(m+n) (τ ) = (−1)n g2 (t)g1∗ (t − τ ) dt. (1.187)
t=−∞
Hence proved. For the signal in Fig. 1.21, using m = 1 and n = 0 we get
g1 (t) = g(t)
g2 (t) = g (1) (t)
= A (δ(t) − 2δ(t − 2T ) + 2δ(t − 3T ) − δ(t − 4T )) . (1.188)
38 1 Signals and Systems
(1)
Rg (t)
5A2
A2
t
−A2
−5A2
Rg (t)
4A2 T
A2 T
t
−A2 T
−4T −2T 0 T 3T
(1)
Fig. 1.22 Plot of Rg (t) and Rg (t)
Hence
we get
26. Let Rg (t) denote the autocorrelation of an energy signal g(t) shown in Fig. 1.23.
(a) Express the first derivative of Rg (t) in terms of the derivative of g(t).
(b) Draw the first derivative of Rg (t). Show all the steps.
1 Signals and Systems 39
1
t
−1 0 2
where
−2 0 1
s(t + 1)
1
t
−3 −2 −1 0 1
s(t − 2)
1
t
−3 −2 −1 0 1 2 3
dRg (t)/dt
0 1 2 3 t
−3 −2 −1
−2
−3
−4
∞
R12 (τ ) = g1 (t)g2∗ (t − τ ) dt
t=−∞
= g1 (t) g2∗ (−t)
G 1 ( f )G ∗2 ( f ), (1.198)
where “” denotes convolution and we have used the fact that convolution in
the time domain is equivalent to multiplication in the frequency domain. Thus,
we have
1 Signals and Systems 41
0 1 2 3 t
−3 −2 −1
−2
−3
−4
Rg (t)
−3 −2 −1 0 1 2 3
∞
R12 (τ ) exp (−j 2π f τ ) dτ = G 1 ( f )G ∗2 ( f )
τ =−∞
∞
⇒ R12 (τ ) = G 1 (0)G ∗2 (0)
τ =−∞
∞
= g1 (t) dt
t=−∞
∞ ∗
g2 (t) dt . (1.199)
t=−∞
42 1 Signals and Systems
Thus proved.
28. (Simon Haykin 1983) Consider two periodic signals g p1 (t) and g p2 (t), both of
period T0 . Show that the cross-correlation function R12 (τ ) satisfies the Fourier
transform pair:
∞
1
R12 (τ ) G 1 (n/T0 )G ∗2 (n/T0 )δ( f − n/T0 ), (1.202)
T02 n=−∞
where G 1 (n/T0 ) and G 2 (n/T0 ) are the Fourier transforms of the generating
functions for the periodic functions g p1 (t) and g p2 (t), respectively.
• Solution: For periodic signals we know that
T0 /2
1
R12 (τ ) = g p1 (t)g ∗p2 (t − τ ) dt. (1.203)
T0 t=−T0 /2
Define
g p1 (t) for −T0 /2 ≤ t < T0 /2
g1 (t) =
0 elsewhere
g p2 (t) for −T0 /2 ≤ t < T0 /2
g2 (t) = (1.204)
0 elsewhere.
Note that g1 (t) and g2 (t) are the generating functions of g p1 (t) and g p2 (t),
respectively. Hence, we have the following relationships:
1 Signals and Systems 43
∞
g p1 (t) = g1 (t − mT0 )
m=−∞
∞
g p2 (t) = g2 (t − mT0 ). (1.205)
m=−∞
Using the first equation of (1.204) and the second equation of (1.205), we get
T0 /2 ∞
1
R12 (τ ) = g1 (t) g2∗ (t − τ − mT0 ) dt
T0 t=−T0 /2 m=−∞
∞ ∞
1
⇒ R12 (τ ) = g1 (t) g2∗ (t − τ − mT0 ) dt. (1.206)
T0 t=−∞ m=−∞
Note that
(a) R12 (τ ) is periodic with period T0 .
(b) The span of Rg1 g2 (τ ) may exceed T0 , hence the summation in (1.207) may
result in aliasing (overlapping). Therefore
1
R12 (τ ) = Rg g (τ ) for −T0 /2 ≤ τ < T0 /2. (1.208)
T0 1 2
where
T0 /2
1
cn = R12 (τ ) exp (−j 2πnτ /T0 ) dτ . (1.210)
T0 τ =−T0 /2
Substituting
t − τ − mT0 = x (1.213)
Combining the summation and the second integral and rearranging terms, we
get
∞
1
cn = 2 g1 (t) exp (−j 2πnt/T0 ) dt
T t=−∞
0∞
g2∗ (x) exp (j 2πnx/T0 ) d x
x=−∞
= G 1 (n/T0 )G ∗2 (n/T0 ). (1.215)
Hence
∞
1
R12 (τ ) = G 1 (n/T0 )G ∗2 (n/T0 ) exp (j 2πnτ /T0 )
T02 n=−∞
∞
1
G 1 (n/T0 )G ∗2 (n/T0 )δ( f − n/T0 ). (1.216)
T02 n=−∞
Thus proved.
29. Compute the Fourier transform of
d x(t)
, (1.217)
dt
where the “hat” denotes the Hilbert transform. Assume that the Fourier transform
of x(t) is X ( f ).
• Solution: We have
1 Signals and Systems 45
−2 −1 1 2
d x(t)
j 2π f X ( f )
dt
d x(t)
⇒ −j sgn ( f ) (j 2π f ) X ( f )
dt
= 2π| f |X ( f ). (1.218)
Evaluate the three bounds on |G( f )| for the pulse shown in Fig. 1.26.
• Solution: We start from the Fourier transform relation:
∞
G( f ) = g(t)e−j 2π f t dt. (1.221)
f =−∞
Similarly, we have
46 1 Signals and Systems
∞
dg(t) −j 2π f t
j 2π f G( f ) = e dt
f =−∞ dt
∞
dg(t)
⇒ |j 2π f G( f )| ≤
dt dt (1.223)
f =−∞
and
∞
d 2 g(t) −j 2π f t
(j 2π f )2 G( f ) = 2
e dt
f =−∞ dt
∞ 2
d g(t)
⇒ (j 2π f )2 G( f ) ≤
dt 2 dt. (1.224)
f =−∞
The various derivatives of g(t) are shown in Fig. 1.27. For the given pulse
2
|g(t)| dt = 3
t=−2
2
dg(t)
dt dt = 2
t=−2
2+ 2
d g(t)
−
dt 2 dt = 4. (1.225)
t=−2
−2 −1 1 2
dg(t)
dt
1 2 t
−2 −1
−1
d2 g(t)
dt2
−1 1 t
2
−2
−1
1 Signals and Systems 47
3.5
2.5
2 |G(f)|
2/(2*pi*|f|)
1.5 4/(2*pi*f)^2
3
1
0.5
-0.5
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
f
|G( f )| ≤ 3
2
|G( f )| ≤
2π| f |
4
|G( f )| ≤ . (1.226)
(2π f )2
These bounds on |G( f )| are shown plotted in Fig. 1.28. It can be shown that
cos(2π f ) − cos(4π f )
G( f ) = . (1.227)
2π 2 f 2
31. (Simon Haykin 1983) A signal that is popularly used in communication systems
is the raised cosine pulse. Consider a periodic sequence of these pulses as shown
in Fig. 1.29. Compute the first three terms (n = 0, 1, 2) of the (real) Fourier
series expansion.
• Solution: The (real) Fourier series expansion of a periodic signal g p (t) can be
written as follows:
∞
g p (t) = a0 + 2 an cos(2πnt/T0 ) + bn sin(2πnt/T0 ). (1.228)
n=1
48 1 Signals and Systems
2.5
1 + cos(2πt)
2
1.5
gp (t)
0.5
-0.5
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
t
4
a1 =
3π
1
a2 = . (1.231)
4
1 Signals and Systems 49
32. (Simon Haykin 1983) Any function g(t) can be expressed as the sum of ge (t)
and go (t), where
1
ge (t) = [g(t) + g(−t)]
2
1
go (t) = [g(t) − g(−t)] . (1.232)
2
(a) Sketch the even and odd parts of
t 1
g(t) = A rect − . (1.233)
T 2
which is nothing but rect (t/T ) shifted by T /2. This is illustrated in Fig. 1.30,
along with the even and odd parts of g(t). We know that
0 T
ge (t)
A/2
−T 0 T
go (t)
A/2
−T t
0
T
−A/2
50 1 Signals and Systems
t
A rect AT sinc ( f T ) . (1.235)
T
A
ge (t) (2T ) sinc ( f (2T )) = AT sinc (2 f T ) . (1.236)
2
Similarly
A A
go (t) T sinc ( f T ) e−j 2π f T /2 − T sinc ( f T ) ej 2π f T /2
2 2
= −j AT sinc ( f T ) sin(π f T ). (1.237)
Now
Using the Fourier transform properties of time scaling and time shifting, we
have
1 Signals and Systems 51
1
x1 (t) = g(at) G( f /a) = X 1 ( f )
|a|
⇒ x1 (t − t0 ) = g[a(t − t0 )] = y(t) X 1 ( f )e−j 2π f t0
1
= G( f /a)e−j 2π f t0
|a|
= Y( f )
1 −π f 2 /a 2 −j 2π f t0
= e e . (1.241)
|a|
1 −2π f 2 /a 2
|Y ( f )|2 = e = H( f ) (say). (1.242)
a2
where
a
c= √ . (1.244)
2
1
√ e−πa t /2 .
2 2
h(t) = (1.245)
|a| 2
34. Consider the periodic pulse train in Fig. 1.31a. Here T ≤ T0 /2.
(a) Using the Fourier transform property of differentiation in the time domain,
compute the complex Fourier series representation of g p (t) in Fig. 1.31a.
Hence, also find the Fourier transform of g p (t).
(b) If g p (t) is passed through a filter having the magnitude and phase response
as depicted in Fig. 1.31b, c, find the output y(t).
• Solution: Consider the pulse:
g p (t) for −T0 /2 < t < T0 /2
g(t) = (1.246)
0 elsewhere.
52 1 Signals and Systems
−T 0 T T0
|H(f )|
(b)
1
f
−3/(2T0 ) 0 3/(2T0 )
∠H(f )
(c)
π/2
3/(2T0 ) f
0
−3/(2T0 )
−π/2
Clearly
d 2 g(t) A
= (δ(t + T ) − 2δ(t) + δ(t − T )) . (1.247)
dt 2 T
Hence
A
G( f ) = − (exp(j 2π f T ) − 2 + exp(−j 2π f T ))
4π 2
f 2T
AT
= 2 2 2 sin2 (π f T )
π f T
= AT sinc2 ( f T ). (1.248)
where
1 ∞
cn = g(t) exp (−j 2πnt/T0 ) dt
T0 t=−∞
1
= G(n/T0 ). (1.250)
T0
1 Signals and Systems 53
exp (j 2π f c t) δ( f − f c ) (1.252)
∞
AT n
G p( f ) = sinc (nT /T0 )δ f −
2
. (1.253)
T0 n=−∞ T0
The filter output will have the frequency components 0, ±1/T0 . From
Fig. 1.31b, c, we note that
H (0) = 1
1
H (1/T0 ) = e−j π/3
3
1 j π/3
H (−1/T0 ) = e . (1.254)
3
Hence
AT AT
y(t) = + sinc2 (T /T0 )e j (2πt/T0 −π/3)
T0 3T0
AT
+ sinc2 (T /T0 )e j (−2πt/T0 +π/3)
3T0
AT 2 AT
= + sinc2 (T /T0 ) cos(2πt/T0 − π/3). (1.255)
T0 3T0
sin(t) H T 1 − cos(t)
−→ . (1.256)
t t
• Solution: We start with the familiar Fourier transform pair:
Substituting A = B = 1, we get
1 HT t
−→ . (1.262)
1+t 2 1 + t2
1 1
exp (−|t|) . (1.263)
2 1 + (2π f )2
1 1
exp (−|2πt|) . (1.264)
2 (2π)(1 + f 2 )
1
π exp (−|2π f |) = g(t) (say). (1.265)
1 + t2
1 Signals and Systems 55
Now the Hilbert transform of g(t) in the above equation is best obtained in
the frequency domain. Thus
∞
ĝ(t) = −j sgn ( f )π exp (−|2π f |) exp (j 2π f t) d f
f =−∞
∞
= −j π exp (−2π f (1 − j t)) d f
f =0
0
+ jπ exp (2π f (1 + j t)) d f
f =−∞
−1 1
= −j π + jπ
−2π(1 − j t) 2π(1 + j t)
t
= . (1.266)
1 + t2
Thus proved.
37. (Simon Haykin 1983) Prove the following Hilbert transform:
HT 1 t − 1/2
rect (t) −→ − ln . (1.267)
π t + 1/2
• Solution: This problem can be easily solved in the time domain. From the
basic definition of the Hilbert transform, we have
1 ∞ g(τ )
ĝ(t) = dτ
π τ =−∞ t − τ
1/2
1 1
= dτ
π τ =−1/2 t − τ
−1 t − 1/2
= ln . (1.268)
π t + 1/2
Thus proved.
38. (Simon Haykin 1983) Let ĝ(t) denote the Hilbert transform of a real-valued
energy signal g(t). Show that the cross-correlation functions of g(t) and ĝ(t) are
given by
Rgĝ (τ ) = − R̂g (τ )
Rĝg (τ ) = R̂g (τ ), (1.269)
Here
g1 (t) = g(t) G( f )
g2 (t) = ĝ(t) −j sgn ( f )G( f )
∗
⇒ g2 (−t) = ĝ ∗ (−t) = ĝ(−t) j sgn ( f )G ∗ ( f ), (1.271)
Rg (τ ) |G( f )|2
⇒ R̂g (τ ) −j sgn ( f ) |G( f )|2 . (1.273)
where g1 (t) and g2 (t) are real-valued signals. Hence, ĝ1 (t) and ĝ2 (t) are also
real-valued. Next, we note that the real part of the integral is equal to the
integral of the real part. Hence, the right-hand side of (1.275) becomes
∞
1
g1 (t)g2 (t) + ĝ1 (t)ĝ2 (t) dt. (1.278)
2 t=−∞
Thus (1.275) is proved. Now, let us see what happens when g2 (t) is replaced
by g2 (−t). Let
g3 (t) = g2 (−t)
⇒ g3+ (t) = g3 (t) + j ĝ3 (t)
= g2 (−t) + j ĝ2 (−t)
∗
⇒ g3+ (t) = g3 (t) − j ĝ3 (t)
= g2 (−t) − j ĝ2 (−t). (1.283)
58 1 Signals and Systems
Clearly (1.275) is still valid with g2+ (t) replaced by g3+ (t) as given in (1.283),
that is,
∞ ∞
1 ∗
{g1+ (t)} {g3+ (t)} dt = g1+ (t)g3+ (t) dt . (1.284)
t=−∞ 2 t=−∞
where g̃(t) is the complex envelope of g(t). Thus, (1.276) follows immediately.
40. Let a narrowband signal be expressed in the form:
Taking the Fourier transform of both sides in the above equation, we get
1
Gc( f ) = G 1 ( f ) + G ∗1 (− f )
2
1
Gs ( f ) = G 1 ( f ) − G ∗1 (− f ) . (1.289)
2j
G1( f ) = G +( f + fc )
∗
⇒ G 1 (− f ) = G ∗+ (− f + f c ). (1.290)
1 Signals and Systems 59
1
Gc( f ) = G + ( f + f c ) + G ∗+ (− f + f c )
2
1
Gs ( f ) = G + ( f + f c ) − G ∗+ (− f + f c ) . (1.291)
2j
41. (Simon Haykin 1983) The duration of a signal provides a measure for describing
the signal as a function of time. The bandwidth of a signal provides a measure
for describing its frequency content. There is no unique set of definitions for
the duration and bandwidth. However, regardless of the definition we find that
their product is always a constant. The choice of a particular set of definitions
merely changes the value of this constant. This problem is intended to explore
these issues.
(a) The root mean square (rms) bandwidth of a lowpass energy signal g(t) is
defined by
∞ 0.5
f =−∞ f
2
|G( f )|2 d f
Wrms = ∞ . (1.292)
f =−∞
|G( f )|2 d f
(b) Consider an energy signal g(t) for which |G( f )| is defined for all frequencies
from −∞ to ∞ and symmetric about f = 0 with its maximum value at
f = 0. The equivalent rectangular bandwidth is defined by
∞
f =−∞
|G( f )|2 d f
Weq = . (1.295)
2|G(0)|2
Show that
• Solution: To prove
g1 (t) = tg(t)
dg(t)
g2 (t) = . (1.300)
dt
Thus, the right-hand side of Schwarz’s inequality in (1.299) becomes
∞ ∞ dg(t) 2
4 t 2 |g(t)|2 dt
dt dt. (1.301)
t=−∞ t=−∞
Let
dg(t)
g3 (t) = j 2π f G( f ) = G 3 ( f ). (1.302)
dt
Then, by Rayleigh’s energy theorem
∞ ∞
|g3 (t)|2 dt = |G 3 ( f )|2 d f
t=−∞ f =−∞
∞
dg(t) 2
∞
⇒ dt = 4π 2 f 2 |G( f )|2 d f. (1.303)
dt
t=−∞ f =−∞
Let us now consider the square root of the left-hand side of Schwarz’s inequal-
ity in (1.299). We have
∞
∞
∗ dg(t) dg(t) ∗ dg(t)
t g (t) + g(t) dt = t g ∗ (t)
t=−∞ dt dt t=−∞ dt
∗
dg (t)
+ g(t) dt
dt
∞
d
= t g(t)g ∗ (t) dt
t=−∞ dt
∞
d
= t |g(t)|2 dt.
t=−∞ dt
(1.305)
In the above equation, we have used the fact (this can also be proved using
Fourier transforms) that
∗
dg(t) dg ∗ (t)
= . (1.306)
dt dt
Let
g4 (t) = |g(t)|2 G 4 ( f )
dg4 (t)
⇒ g5 (t) = j 2π f G 4 ( f ) = G 5 ( f )
dt
j dG 5 ( f )
⇒ tg5 (t)
2π d f
∞
j dG 5 ( f )
⇒ tg5 (t) dt =
t=−∞ 2π d f f =0
∞
j d
⇒ tg5 (t) dt = (j 2π f G 4 ( f ))
t=−∞ 2π d f f =0
∞
d
⇒ t |g(t)|2 dt = −G 4 (0)
t=−∞ dt
∞ ∞
d
⇒ t |g(t)|2 dt = − |g(t)|2 dt. (1.307)
t=−∞ dt t=−∞
Taking the square root of (1.304) and (1.308), we get the desired result in
(1.298).
To prove
Using the above inequality and the Rayleigh’s energy theorem which states
that
∞ ∞
|g(t)|2 dt = |G( f )|2 d f, (1.312)
t=−∞ f =−∞
42. x(t) is a bandpass signal whose Fourier transform is shown in Fig. 1.32. h(t) is
an LTI system having a bandpass frequency response of the form
−fc 0 fc fc + W
−fc − W −fc + W fc − W
1 Signals and Systems 63
f 2 for | f | < B
Hc ( f ) = , (1.315)
0 otherwise
X ( f ) = 4[X c ( f − f c ) + X c ( f + f c )]
8xc (t) cos(2π f c t)
= x(t). (1.316)
Similarly
Hc ( f − f c ) + Hc ( f + f c )
H( f ) = . (1.317)
2
Now
Y ( f ) = X ( f )H ( f )
= 2[X c ( f − f c )Hc ( f − f c ) + X c ( f + f c )Hc ( f + f c )].
(1.318)
Let
G c ( f ) = X c ( f )Hc ( f )
2
f X c ( f ) for | f | < W
=
0 otherwise
−1 d 2 xc (t)
4π 2 dt 2
= gc (t). (1.319)
Then
43. Two periodic signals g p1 (t) and g p2 (t) are depicted in Fig. 1.33. Let
gp2 (t)
2
1
t
0 1 2 3 4
Let
d 2 g p (t)
g p4 (t) = (1.324)
dt 2
gp2 (t)
2
1
t
0 1 2 3 4
gp (t)
3
1
t
dgp (t)/dt
3
−1
−2
d2 gp (t)/dt2 3
1 1
t
−5
Now
∞
g p4 (t) = g4 (t − kT )
k=−∞
∞
G4( f ) e−j 2π f kT
k=−∞
∞
1
= G 4 (n/T )δ( f − n/T ), (1.326)
T n=−∞
66 1 Signals and Systems
where T = 4 s.
44. Using Fourier transform properties and/or any other relation, compute
∞ A sinc ( f T ) e−j π f T − A e−j 4π f T 2
d f. (1.327)
j 2π f
f =−∞
Clearly state which Fourier transform property and/or relation is being used. The
constant of any integration may be assumed to be zero.
• Solution: We know from Parseval’s relation that
∞ ∞
|G( f )|2 d f = |g(t)|2 dt, (1.328)
f =−∞ t=−∞
where
g(t) G( f ) (1.329)
Let
G 1 ( f ) = j 2π f G( f )
= A sinc ( f T ) e−j π f T − A e−j 4π f T
g1 (t)
= dg(t)/dt
A t − T /2
= rect − Aδ(t − 2T ). (1.331)
T T
state which Fourier transform property and/or relation is being used and show
all the steps.
1 Signals and Systems 67
2T t
T
A
g(t)
T 2T
e−πt e−π f .
2 2
(1.333)
g(t) G( f ), (1.334)
1
g(at) G( f /a), (1.335)
|a|
where a = 0. Therefore
1
e−2πt √ e−π f /2
2 2
2
1
e−3πt √ e−π f /3 .
2 2
(1.336)
3
Moreover
2
2e−2πt e−3πt √ e−π f /2 e−π f /3 .
2 2 2 2
(1.337)
6
68 1 Signals and Systems
Using
∞
g(t) dt = G(0) (1.338)
t=−∞
√
the area under 2e−2πt e−3πt is 2/ 6.
2 2
47. Let G( f ) denote the Fourier transform of g(t). It is given that G(0) is finite and
non-zero. Does g(t) contain a dc component? Justify your answer.
• Solution: If g(t) has a dc component, then G( f ) contains a Dirac-delta function
at f = 0. However, it is given that G(0) is finite. Therefore, g(t) does not
contain any dc component.
48. Consider the following: if CONDITION then STATEMENT. The CONDITION
is given to be necessary.
Which of the following statement(s) are correct (more than one statement may
be correct).
(a) CONDITION is true implies STATEMENT is always true.
(b) CONDITION is true implies STATEMENT is always false.
(c) CONDITION is true implies STATEMENT may be true or false.
(d) CONDITION is false implies STATEMENT is always true.
(e) CONDITION is false implies STATEMENT is always false.
(f) CONDITION is false implies STATEMENT may be true or false.
∞
1
S= (1.340)
n=1
n
does not exist even though the nth term goes to zero as n → ∞.
1 Signals and Systems 69
50. State Parseval’s power theorem for periodic signals in terms of its complex
Fourier series coefficients.
• Solution: Let g p (t) be a periodic signal with period T0 . Then g p (t) can be
represented in the form of a complex Fourier series as follows:
∞
g p (t) = cn e j 2πnt/T0 . (1.341)
n=−∞
51. State and derive the Poisson sum formula. Hence show that
∞
∞
1 n
e j 2π f nT0 = δ f − . (1.343)
n=−∞
T0 n=−∞ T0
• Solution: Let g p (t) be a periodic signal with period T0 . Then g p (t) can be
expressed in the form of a complex Fourier series as follows:
∞
g p (t) = cn e j 2πnt/T0 , (1.344)
n=−∞
70 1 Signals and Systems
where
T0
1
cn = g p (t)e−j 2πnt/T0 dt. (1.345)
T0 t=0
Note that
∞
g p (t) = g(t − nT0 ). (1.347)
n=−∞
cn T0 = G(n/T0 ). (1.349)
which proves the Poisson sum formula. Taking the Fourier transform of both
sides of (1.350) we obtain
∞
∞
−j 2π f nT0 1 n
G( f ) e = G δ( f − n/T0 ). (1.351)
n=−∞
T0 n=−∞ T0
52. Clearly define the unit step function. Derive the Fourier transform of the unit
step function.
• Solution: The unit step function is defined as
t
u(t) = δ(τ ) dτ
τ =−∞
⎧
⎨ 0 for τ < 0
= 1/2 for τ = 0 , (1.353)
⎩
1 for τ > 0
where
⎧
⎨ 1 for t > 0
sgn (t) = 0 for t = 0 (1.356)
⎩
−1 for t < 0.
−j 4π f
G( f ) = . (1.357)
a2 + (2π f )2
1
sgn (t) . (1.358)
jπ f
We know that
1
2U ( f ) − δ( f ) =
jπ f
1 δ( f )
⇒ U( f ) = + , (1.360)
j 2π f 2
1
H( f ) = . (1.363)
a + j 2π f
Taking the limit a → 0 in (1.363), we obtain from (1.362), the Fourier trans-
form of the unit step as
1
u(t) , (1.364)
j 2π f
which is different from (1.360). However, we note from (1.364) that the right-
hand side is purely imaginary, which implies that the unit step is an odd
function, which is incorrect. Therefore, the Fourier transform of the unit step
is given by (1.360).
53. Compute ∞
sinc3 (3t) dt, (1.365)
t=−∞
where
sin(πt)
sinc (t) = . (1.366)
πt
• Solution: We know that
1
rect ( f /3) sinc (3t). (1.369)
3
Using the property that multiplication in the time domain corresponds to con-
volution in the frequency domain, we get
1
G( f ) = (rect ( f /3) rect ( f /3) rect ( f /3)) sinc3 (3t) = g(t),
27
(1.370)
where “” denotes convolution. Using the property
∞
g(t) dt = G(0), (1.371)
t=−∞
we get
∞
1
sinc3 (3t) dt = rect ( f /3) 3 [(1 − | f |/3) rect ( f /6)] f =0
t=−∞ 27
3/2
1
= (1 − |α|/3) dα
9 α=−3/2
2 3/2
= (1 − α/3) dα
9 α=0
1
= . (1.372)
4
54. A signal g(t) has Fourier transform given by
Compute ĝ(t).
• Solution: Clearly
δ(t + t0 ) − δ(t − t0 )
g(t) = , (1.374)
2j
where δ(·) is the Dirac-delta function. The impulse response of the Hilbert
transformer is
1
h(t) = . (1.375)
πt
74 1 Signals and Systems
Therefore
24
g(t) = . (1.377)
9 + (2πt)2
2ab
ae−b|t| . (1.378)
b2 + (2π f )2
2ab
G( f ) = ae−b| f | = g(t). (1.379)
b2 + (2πt)2
1/(4T )
t
−T /3 0 T /3 2T /3
−1
= 4j
(3 − j 2πt)
1
+ 4j
(3 + j 2πt)
16πt
= . (1.381)
9 + (2πt)2
56. Consider the signal g(t) in Fig. 1.36. Which of the following statement(s) are
correct?
(a)
5
g(t) g(t)|t=−T /3 = . (1.382)
48T
(b) The Fourier transform of g(t) g(t) is real-valued and even.
(c) The g(t) g(−t) is real-valued and even.
(d) g(t) δ(3t) = g(t)/3.
Here “” denotes convolution.
1/(4T )
τ
−T /3 0 T /3 2T /3
g(τ − T /3)
1/(2T )
1/(4T )
τ
0 2T /3 T
g(−τ − T /3)
1/(2T )
1/(4T )
τ
−T −2T /3 0
∞
g(t) δ(3t) = δ(3τ )g(t − τ ) dτ
τ =−∞
∞
1
= δ(α)g(t − α/3) dα
α=−∞ 3
g(t)
= . (1.385)
3
References
1. Show that the magnitude of the correlation coefficient |ρ| is always less than or
equal to unity.
• Solution: For any real number a, we have
E (a(X − m X ) − (Y − m Y ))2 ≥ 0
⇒ a 2 σ 2X − 2a E [(X − m X )(Y − m Y )] + σY2 ≥ 0. (2.1)
Hence proved.
2. (Papoulis 1991) Using characteristic functions, show that
where E[X i X j ] = Ci j and the random variables X i are jointly normal (Gaussian)
with zero mean.
• Solution: The joint characteristic function of X 1 , X 2 , X 3 , and X 4 is given by
Expanding the exponential in the form of a power series and considering only
the fourth power, we have
© The Editor(s) (if applicable) and The Author(s), under exclusive license 77
to Springer Nature Switzerland AG 2021
K. Vasudevan, Analog Communications,
https://doi.org/10.1007/978-3-030-50337-6_2
78 2 Random Variables and Random Processes
E e j (v1 X 1 +···+v4 X 4 )
1
= · · · + E (v1 X 1 + · · · + v4 X 4 )4 + · · ·
4!
24
= ··· + E[X 1 X 2 X 3 X 4 ]v1 v2 v3 v4 + · · · . (2.5)
4!
Now, let
W = v1 X 1 + · · · + v4 X 4 . (2.6)
Then
E[W ] = 0
E[W 2 ] = vi v j Ci j = σw2 . (2.7)
i, j
= e− 2 vi v j Ci j
1
i, j . (2.8)
X if X ≥ 0
Y = (2.10)
0 if X < 0.
Y = g(X ). (2.11)
2 Random Variables and Random Processes 79
X = g −1 (Y ). (2.12)
dy 1 for X > 0
= (2.14)
dx 0 for X < 0.
where k is a constant such that the pdf of Y integrates to unity. It is easy to see
that
∞
f Y (y) dy = k + 1/2 = 1
y=−∞
⇒ k = 1/2, (2.17)
Y
-0.2
x1 x2 + dx2 x
-0.4 2
x1 + dx1
-0.6
-0.8
-1
0 1 2 3 4 5 6
dy dy
= (2.19)
d x1 d x x=x1
1/W for 0 ≤ f ≤ W
fF ( f ) = (2.23)
0 elsewhere.
Is X (t) WSS?
• Solution: Let us first compute the mean value of X (t).
W
1
E[X (t)] = sin(2π Ft) d F
W f =0
−1
= (cos(2πW t) − 1) (2.24)
2W πt
where A is a Gaussian distributed random variable with zero mean and variance
σ 2A . This random process is applied to an ideal integrator producing an output
Y (t) defined by
t
Y (t) = X (τ ) dτ . (2.26)
τ =0
(a) Determine the probability density function of the output Y (t) at a particular
time tk .
(b) Determine whether Y (t) is WSS.
(c) Determine whether Y (t) is ergodic in the mean and in the autocorrelation.
A
Y (tk ) = sin(2π f c tk ). (2.27)
2π f c
Since all the terms in the above equation excepting A are constants for the
time instant tk , Y (tk ) is also a Gaussian distributed random variable with mean
and variance given by
E[A]
E[Y (tk )] = sin(2π f c tk ) = 0
2π f c
E[A2 ] 2
E[Y 2 (tk )] = sin (2π f c tk )
4π 2 f c2
σ 2A
= sin2 (2π f c tk ). (2.28)
4π 2 f c2
82 2 Random Variables and Random Processes
Since the variance is a function of time, Y (tk ) is not WSS. Hence it is not
ergodic in the autocorrelation. However, it can be shown that the time-averaged
mean is zero. Hence Y (t) is ergodic in the mean.
(a) Determine the joint pdf of the random variables Z (t1 ) and Z (t2 ).
(b) Is Z (t) WSS?
• Solution: The random process Z (t) has mean and autocorrelation given by
E[X ] = E[Y ] = 0
E[X 2 ] = E[Y 2 ] = 1
E[X Y ] = E[X ]E[Y ] = 0. (2.31)
f Y1 , Y2 (y1 , y2 )
1
=
2πσ1 σ2 1 − ρ2
σ 2 (y1 − m 1 )2 − 2σ1 σ2 ρ(y1 − m 1 )(y2 − m 2 ) + σ12 (y2 − m 2 )2
× exp − 2 ,
2σ12 σ22 (1 − ρ2 )
(2.32)
2 Random Variables and Random Processes 83
where
E[y1 ] = m 1
E[y2 ] = m 2
E[(y1 − m 1 )2 ] = σ12
E[(y2 − m 2 )2 ] = σ22
E[(y1 − m 1 )(y2 − m 2 )]/(σ1 σ2 ) = ρ. (2.33)
8. Using ensemble averaging, find the mean and the autocorrelation of the random
process given by
∞
X (t) = Sk p(t − kT − α), (2.35)
k=−∞
where for the given binary phase shift keying (BPSK) constellation
R X (τ ) = E[X (t)X (t − τ )]
⎡ ⎤
∞ ∞
= E⎣ Sk p(t − kT − α) S j p(t − τ − j T − α)⎦
k=−∞ j=−∞
∞
∞
= E[Sk S j ]E[ p(t − kT − α) p(t − τ − j T − α)]
k=−∞ j=−∞
∞
∞
= A2 δ K (k − j)
k=−∞ j=−∞
1 T
× p(t − kT − α) p(t − τ − j T − α) dα
T α=0
∞
A2 T
= p(t − kT − α) p(t − τ − kT − α) dα, (2.38)
k=−∞
T α=0
1 for k = j
δ K (k − j) = (2.39)
0 for k = j
x = t − kT − α. (2.40)
∞
A2 t−kT
R X (τ ) = p(x) p(x − τ ) d x. (2.41)
k=−∞
T x=t−kT −T
9. (Haykin 1983) The square wave x(t) of constant amplitude A, period T0 , and
delay td represents the sample function of a random process X (t). This is illus-
trated in Fig. 2.2. The delay is random and is described by the pdf
2 Random Variables and Random Processes 85
T0
td
(a) Determine the mean and autocorrelation of X (t) using ensemble averaging.
(b) Determine the mean and autocorrelation of X (t) using time averaging.
(c) Is X (t) WSS?
(d) Is X (t) ergodic in the mean and the autocorrelation?
t − kT0 − td = x (2.46)
we get
∞ t−kT0
1
E[X (t)] = p(x) d x. (2.47)
T0 k=−∞ x=t−kT0 −T0
86 2 Random Variables and Random Processes
A
E[X (t)] = . (2.49)
2
The autocorrelation can be computed as
∞
E[X (t)X (t − τ )] = E p(t − kT0 − td )
k=−∞
⎤
∞
p(t − τ − j T0 − td )⎦
j=−∞
∞ ∞
1
=
T0 k=−∞ j=−∞
T0
p(t − kT0 − td ) p(t − τ − j T0 − td ) dtd . (2.50)
td =0
Substituting
t − kT0 − td = x (2.51)
we get
∞ ∞
1
E[X (t)X (t − τ )] =
T0 k=−∞ j=−∞
t−kT0
p(x) p(x + kT0 − τ − j T0 ) d x. (2.52)
x=t−kT0 −T0
Let
∞ ∞
1
E[X (t)X (t − τ )] =
T0 k=−∞ m=−∞
t−kT0
p(x) p(x + mT0 − τ ) d x. (2.54)
x=t−kT0 −T0
Now we interchange the order of summation and combine the summation over
k and the integral to obtain
∞ ∞
1
E[X (t)X (t − τ )] = p(x) p(x + mT0 − τ ) d x
T0 m=−∞ x=−∞
∞
1
= R p (τ − mT0 ) = R X (τ ), (2.55)
T0 m=−∞
p(t)
(a)
A
T0 /2 T0
Rp (τ )
(b)
A2 T0 /2
−T0 /2 T0 /2
RX (τ )
(c)
A2 /2
−T0 /2 T0 /2
Fig. 2.3 Computing the autocorrelation of a periodic wave with random timing phase
A
t
0
T0 /4
td
T0
Therefore, comparing (2.55) and (2.57) we find that X (t) is ergodic in the
autocorrelation.
10. A signal x(t) with period T0 and delay td represents the sample function of a
random process X (t). This is illustrated in Fig. 2.4. The delay td is a random
variable which is uniformly distributed in [0, T0 ).
(a) Determine the mean and autocorrelation of X (t) using ensemble averaging.
(b) Determine the mean and autocorrelation of X (t) using time averaging.
(c) Is X (t) WSS?
(d) Is X (t) ergodic in the mean and the autocorrelation?
2 Random Variables and Random Processes 89
t − kT0 − td = x (2.62)
we get
∞ t−kT0
1
E[X (t)] = p(x) d x. (2.63)
T0 k=−∞ x=t−kT0 −T0
3A
E[X (t)] = . (2.65)
8
The autocorrelation can be computed as
∞
E[X (t)X (t − τ )] = E p(t − kT0 − td )
k=−∞
90 2 Random Variables and Random Processes
⎤
∞
p(t − τ − j T0 − td )⎦
j=−∞
∞ ∞
1
=
T0 k=−∞ j=−∞
T0
p(t − kT0 − td ) p(t − τ − j T0 − td ) dtd(2.66)
.
td =0
Substituting
t − kT0 − td = x (2.67)
we get
∞ ∞
1
E[X (t)X (t − τ )] =
T0 k=−∞ j=−∞
t−kT0
p(x) p(x + kT0 − τ − j T0 ) d x.(2.68)
x=t−kT0 −T0
Let
Now we interchange the order of summation and combine the summation over
k and the integral to obtain
∞ ∞
1
E[X (t)X (t − τ )] = p(x) p(x + mT0 − τ ) d x
T0 m=−∞ x=−∞
∞
1
= R p (τ − mT0 ) = R X (τ ), (2.71)
T0 m=−∞
p(t)
(a)
2A
A
t
T0 /8 T0 /4
Rp (τ )
(b)
5A2 T0 /8
2A2 T0 /8
τ
−T0 /4 0 T0 /4
−T0 /8 T0 /8
RX (τ )
(c)
5A2 /8
2A2 /8
−T0 /4 T0 /4
−T0 /8 T0 /8
Fig. 2.5 Computing the autocorrelation of a periodic wave with random timing phase
Therefore, comparing (2.71) and (2.73) we find that X (t) is ergodic in the
autocorrelation.
11. For two jointly Gaussian random variables X and Y with means m X and m Y ,
variances σ 2X and σY2 and coefficient of correlation ρ, compute the conditional
pdf of X given Y . Hence compute the values of E[X |Y ] and var(X |Y ).
• Solution: The joint pdf of two real-valued Gaussian random variables X and
Y is given by
1
f X, Y (x, y) =
2πσ X σY 1 − ρ2
σY2 (x − m X )2 − 2σ X σY ρ(x − m X )(y − m Y ) + σ 2X (y − m Y )2
× exp − .
2σ 2X σY2 (1 − ρ2 )
(2.76)
Therefore
2 Random Variables and Random Processes 93
f X Y (x, y)
f X |Y (x|y) =
f Y (y)
1
=
σ X 2π(1 − ρ2 )
σ 2 A2 − 2σ X σY ρAB + ρ2 σ 2X B 2
× exp − Y , (2.78)
2σ 2X σY2 (1 − ρ2 )
A = x − mX
B = y − mY . (2.79)
1
f X |Y (x|y) =
σ X 2π(1 − ρ2 )
(σY (x − m X ) − ρσ X (y − m Y ))2
× exp −
2σ 2X σY2 (1 − ρ2 )
1
=
σ X 2π(1 − ρ2 )
(x − (m X + ρ(σ X /σY )(y − m Y )))2
× exp − . (2.80)
2σ 2X (1 − ρ2 )
12. If
1 (x − m)2
f X (x) = √ exp − (2.82)
σ 2π 2σ 2
compute (a) E (X − m)2n and (b) E (X − m)2n−1 .
• Solution: Clearly
E (X − m)2n−1 = 0 (2.83)
since the integrand is an odd function. To compute E[(X − m)2n ] we use the
method of integration by parts. We have
94 2 Random Variables and Random Processes
∞
1
(x − m)2n−1 (x − m)e−(x−m) /(2σ 2 )
2
E (X − m)2n = √ d x.
σ 2π x=−∞
(2.84)
The first function is taken as (x − m)2n−1 . The integral of the second function
is
(x − m)2 (x − m)2
(x − m) exp − d x = −σ exp −
2
. (2.85)
x 2σ 2 2σ 2
13. X is a uniformly distributed random variable between zero and one. Find out the
transformation Y = g(X ) such that Y is a Rayleigh distributed random variable.
You can assume that the mapping between X and Y is one-to-one and Y mono-
tonically increases with X . There should not be any unknown constants in your
answer. The Rayleigh pdf is given by
y y2
f Y (y) = 2 exp − 2 for y ≥ 0. (2.88)
σ 2σ
1 for 0 ≤ x ≤ 1
f X (x) = . (2.89)
0 elsewhere
f Y (y) dy = f X (x) d x
Y X
⇒ f Y (y) dy = f X (x) d x. (2.90)
y=0 0
2 Random Variables and Random Processes 95
1 − e−Y /(2σ 2 )
2
=X
⇒ Y = −2σ 2 ln(1 − X ).
2
(2.91)
14. Given that Y is a Gaussian distributed random variable with zero mean and
variance σ 2 , use the Chernoff bound to compute the probability that Y ≥ δ for
δ > 0.
• Solution: We know that
P[Y ≥ δ] ≤ E eμ(Y −δ) , (2.92)
where μ needs to be found out such that the RHS is minimized. Hence we set
d μ(Y −δ)
E e = 0. (2.93)
dμ
15. If two Gaussian distributed random variables, X and Y are uncorrelated, are they
statistically independent? Justify your statement.
• Solution: Since X and Y are uncorrelated
E[(X − m x )(Y − m Y )] = 0
⇒ ρ = E[(X − m X )(Y − m Y )]/σ X σY = 0. (2.96)
SN (f ) (Watts/Hz)
1
f (Hz)
−7 −5 −4 0 4 5 7
1 1
√ e−(x−m X ) /(2σ X ) √ e−(y−m Y ) /(2σY )
2 2 2 2
f X Y (x, y) =
σ X 2π σY 2π
= f X (x) f Y (y). (2.97)
S N ( f − f c ) + S N ( f + f c ) for | f | < f c
S Nc ( f ) = S Ns ( f ) = (2.98)
0 otherwise.
The psd of the in-phase and quadrature components is plotted in Fig. 2.7.
We also know that the cross-spectral densities are given by
j [S N ( f + f c ) − S N ( f − f c )] for | f | < f c
S Nc Ns ( f ) =
0 otherwise.
= −S Ns Nc ( f ). (2.99)
x for x > 0
z =
2
(2.100)
−x for x < 0.
SN (f ) (Watts/Hz)
1
S2 (f ) S1 (f )
f (Hz)
−7 −5 −4 0 4 5 7
Watts/Hz
S1 (f + fc )
1
0.5
f (Hz)
0
0
S2 (f − fc )
1
0.5
f (Hz)
0
SNc (f ) = SNs (f )
0.5
f (Hz)
0
f (Hz)
−2 −1 0 1 2
d x for x > 0
2z dz = (2.101)
−d x for x < 0.
SN (f ) (Watts/Hz)
1
S2 (f ) S1 (f )
f (Hz)
−7 −5 −4 0 4 5 7
Watts/Hz
S1 (f + fc )
1
0.5
f (Hz)
0
f (Hz)
0
−0.5
−1
−S2 (f − fc )
SNc Ns (f )/j
0.5
f (Hz)
0
−0.5
f (Hz)
−2 −1 0 1 2
1.8
1.6
1.4
1.2
1
Z
0.8
0.6
0.4
0.2
0
-1 -0.5 0 0.5 1 1.5 2 2.5 3
X
√
Fig. 2.9 The transformation Z = |X |
f X (x)
f Z (z) =
|dz/d x| x=z 2
1/4
⇒ f Z (z) =
1/(2z)
z √
= for 1 ≤ z ≤ 3. (2.104)
2
To summarize
z for 0 ≤ z ≤ √
1
f Z (z) = (2.105)
z
2
for 1 ≤ z ≤ 3.
π + 2 tan−1 (x)
FX (x) = for − ∞ < x < ∞ (2.106)
2π
find the pdf of the random variable Z given by
X 2 for X ≥ 0
Z= (2.107)
−1 for X < 0
100 2 Random Variables and Random Processes
FX (−∞) = 0
FX (∞) = 1. (2.108)
d FX (x)
f X (x) =
dx
1
= . (2.109)
π(1 + x 2 )
Note that
dz 2x for x ≥ 0
=
dx 0 for x < 0
√
dz 2 z for x ≥ 0
⇒ = (2.110)
dx 0 for x < 0.
Therefore f Z (z) must have a delta function at Z = −1. Further, the mapping
from X to Z is one-to-one (injective) in the range 0 ≤ x < ∞. Hence, the pdf
of Z is given by
kδ(z + 1) for z = −1
f Z (z) = (2.112)
f X (x)/(dz/d x)|x=√z for 0 ≤ z < ∞,
kδ(z + 1) for z = −1
f Z (z) = √ (2.113)
1/(2π(1 + z) z) for 0 ≤ z < ∞.
Since
∞
f Z (z) dz = P(Z > 0) = P(X > 0) = 1 − FX (0) = 1/2 (2.114)
z=0
19. (Papoulis 1991) Let X be a nonnegative continuous random variable and let a
be any positive constant. Prove the Markov inequality given by
2 Random Variables and Random Processes 101
g(X)
1
−2 −1 0 1 2
where m X = E[X ].
• Solution: Consider a function g(X ) given by
1 for X > a
g(X ) = (2.116)
0 for X < a
20. Let a random variable X have a uniform pdf over −2 ≤ X ≤ 2. Compute the
pdf of Y = g(X ), where
2X 2 for |X | ≤ 1
g(X ) = (2.118)
3|X | − 1 for 1 ≤ |X | ≤ 2.
• Solution: We first note that the mapping g(x) is many-to-one. This is illustrated
in Fig. 2.11. Let y and x denote particular values taken by the RVs Y and X ,
respectively. Therefore we have
102 2 Random Variables and Random Processes
Note that
21. A Gaussian distributed random variable X having zero mean and variance σ 2X is
transformed by a square-law device defined by Y = X 2 .
(a) Compute the pdf of Y .
(b) Compute E[Y ].
• Solution: We first note that the mapping is many-to-one. Thus the probability
that Y lies in the range [y, y + dy] is equal to the probability that X lies in the
range [x1 , x1 + d x1 ] and [x2 , x2 + d x2 ], as illustrated in Fig. 2.12. Observe
that d x2 is negative. Mathematically
where
dy dy
= (2.125)
d x1 d x x=x1
√ √
with x1 = y and x2 = − y. Since
2 Random Variables and Random Processes 103
4.5
3.5
2.5
Y
1.5 y + dy
y
1
0.5
x2 + dx2 x2 x1 x1 + dx1
0
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
X
1
√ e−x /(2σ X )
2 2
f X (x) =
σ X 2π
dy
= 2x (2.126)
dx
we have
1
e−y/(2σ X )
2
f Y (y) = √ for y ≥ 0. (2.127)
σ X 2π y
Finally
A = X (t)
B = X (t + τ + τ p ) − X (t + τ ) (2.131)
in we obtain
However (2.134) cannot be less than zero; it can only be equal to zero. There-
fore
R X (τ + τ p ) = R X (τ ) (2.135)
23. Let X and Y be two independent and uniformly distributed random variables
over [−a, a].
(a) Compute the pdf of Z = X + Y .
(b) If a = π and X , Y , Z denote the phase over [−π, π], recompute the pdf of
Z . Assume that X and Y are independent.
• Solution: We note that the pdf of the sum of two independent random variables
is equal to the convolution of the individual pdfs. Therefore
−a 0 a
fZ (z)
(b) 1/(2a)
z
−2a 0 2a
fZ (z)
(c) 1/(2a)
z
−π 0 π
fZ (z)
(d) 1/(2a)
z
−π 0 π
This is illustrated in Fig. 2.13c. The resulting pdf is shown in Fig. 2.13d. Thus
we find that the modified pdf of Z is also uniform in [−π, π].
24. Let X and Y be two independent RVs with pdfs given by
1/4 for − 2 ≤ x ≤ 2
f X (x) = (2.138)
0 otherwise
and
(a) Find A.
(b) Find the pdf of Z = 3X + 4Y .
• Solution: We have
∞
f Y (y) dy = 1
y=0
∞
⇒A e−3y dy = 1
y=0
⇒ A = 3. (2.140)
Let U = 3X . Then
1/12 for − 6 ≤ u ≤ 6
fU (u) = (2.141)
0 otherwise.
f V (v) dv = f Y (y) dy
f Y (y)
⇒ f V (v) =
dv/dy y=v/4
3
⇒ f V (v) = e−3v/4 for 0 ≤ v < ∞
4
3
= e−3v/4 S(v), (2.142)
4
where S(v) denotes the modified unit step function defined by
1 for v ≥ 0
S(v) = (2.143)
0 for v < 0.
25. (Haykin 1983) Consider two linear filters h 1 (t) and h 2 (t) connected in cascade
as shown in Fig. 2.14. Let X (t) be a WSS process with autocorrelation R X (τ ).
(a) Find the autocorrelation function of Y (t).
(b) Find the cross-correlation function RV Y (t).
• Solution: Let
Then
Hence
RY (τ ) = E [Y (t)Y (t − τ )]
∞ ∞
=E g(α)X (t − α) dα g(β)X (t − τ − β) dβ
α=−∞ β=−∞
∞ ∞
= g(α)g(β)R X (τ + β − α) dα dβ. (2.148)
α=−∞ β=−∞
RV Y (τ ) = E [V (t)Y (t − τ )]
∞
=E h 1 (α)X (t − α) dα
α=−∞
∞
× h 2 (β)V (t − τ − β) dβ
β=−∞
∞ ∞
=E h 1 (α)h 2 (β)X (t − α)
α=−∞ β=−∞
∞
× h 1 (γ)X (t − τ − β − γ) dγ dα dβ
γ=−∞
108 2 Random Variables and Random Processes
∞ ∞ ∞
= h 1 (α)h 2 (β)h 1 (γ)
α=−∞ β=−∞ γ=−∞
× R X (τ + β + γ − α) dγ dα dβ. (2.149)
26. (Haykin 1983) Consider a pair of WSS processes X (t) and Y (t). Show that their
cross-correlations R X Y (τ ) and RY X (τ ) have the following properties:
(a) RY X (τ ) = R X Y (−τ ).
(b) |R X Y (τ )| ≤ 21 [R X (0) + RY (0)].
• Solution: We know that
RY X (τ ) = E [Y (t)X (t − τ )] . (2.150)
RY X (τ ) = E [Y (α + τ )X (α).] . (2.151)
Substituting τ = −β we get
Thus the first part is proved. To prove the second part we note that
E (X (t) ± Y (t − τ ))2 ≥0
1
⇒ E X (t)2 + E Y (t − τ )2 ≥ ∓E [X (t)Y (t − τ )]
2
1
⇒ [R X (0) + RY (0)] ≥ ∓R X Y (τ )
2
1
⇒ [R X (0) + RY (0)] ≥ |R X Y (τ )|. (2.153)
2
27. (Haykin 1983) Given that a stationary random process X (t) has the autocorre-
lation function R X (τ ) and power spectral density S X ( f ) show that
(a) The autocorrelation function of d X (t)/dt is equal to minus the second
derivative of R X (τ ).
(b) The power spectral density of d X (t)/dt is equal to 4π 2 f 2 S X ( f ).
(c) If S X ( f ) = 2 rect ( f /W ), compute the power of d X (t)/dt.
• Solution: Let
d X (t)
Y (t) = . (2.154)
dt
It is clear that Y (t) can be obtained by passing X (t) through an ideal differ-
entiator. We know that the Fourier transform of an ideal differentiator is
2 Random Variables and Random Processes 109
H ( f ) = j 2π f. (2.155)
SY ( f ) = S X ( f )|H ( f )|2
= 4π 2 f 2 S X ( f )
∞
⇒ RY (τ ) = SY ( f ) exp (j 2π f τ ) d f
f =−∞
∞
= 4π 2 f 2 S X ( f ) exp (j 2π f τ ) d f. (2.156)
f =−∞
Thus, the second part of the problem is proved. To prove the first part, we note
that
∞
R X (τ ) = S X ( f ) exp (j 2π f τ ) d f
f =−∞
∞
d R X (τ )
2
⇒ = −4π 2 f 2 S X ( f ) exp (j 2π f τ ) d f. (2.157)
dτ 2 f =−∞
d 2 R X (τ )
RY (τ ) = − . (2.158)
dτ 2
1
f
−f0 f0
110 2 Random Variables and Random Processes
(a) Determine R X (τ ).
(b) Determine the dc power in X (t).
(c) Determine the ac power in X (t).
(d) What sampling-rates will give uncorrelated samples of X (t)? Are the sam-
ples statistically independent?
• Solution: The psd of X (t) can be written as
|f|
S X ( f ) = δ( f ) + 1 − . (2.160)
f0
We know that
Applying duality
|f|
A2 f 0 1 − A2 f 02 sinc2 (t f 0 ). (2.162)
f0
Thus
R X (τ ) = 1 + f 0 sinc2 ( f 0 τ ). (2.164)
E[Z (t)] = A
R Z (τ ) = E[Z (t)Z (t − τ )] = A2 + RY (τ ). (2.166)
Hence the DC power (contributed by the delta function of the psd) is unity.
The AC power (contributed by the triangular part of the psd) is f 0 .
The covariance of X (t) is
It is clear that
f 0 for τ = 0
K X (τ ) = (2.169)
0 for τ = n(k/ f 0 ), n, k = 0,
where n and k are positive integers. Thus, when X (t) is sampled at a rate
equal to f 0 /k, the samples are uncorrelated. However the samples may not be
statistically independent.
29. The random variable X has a uniform distribution over 0 ≤ x ≤ 2. For the ran-
dom process defined by V (t) = 6e X t compute
(a) E[V (t)]
(t)V(t − τ )]
(b) E[V
(c) E V 2 (t)
Hence
112 2 Random Variables and Random Processes
RY (τ )
1
τ
E V 2 (t) = E [V (t)V (t − τ )]τ =0
9 4t
= e −1 . (2.172)
t
where G(t) and X (t) have zero mean and are independent of each other. The
autocorrelation of Y (t) can be written as
RY (τ ) = E[Y (t)Y (t − τ )]
= E[(A + G(t) + X (t))(A + G(t − τ ) + X (t − τ ))]
= A2 + RG (τ ) + R X (τ ), (2.174)
RG (τ ) = E[G(t)G(t − τ )]
= E[G(t)G(t − τ − kT0 )]
= RG (τ + kT0 ). (2.175)
2 Random Variables and Random Processes 113
Therefore RG (τ ) is also
√ periodic with a period of kT0 . Moreover, since A is
given to be equal to 3/2, we have A2 = 3/2. Therefore the DC power is
3/2.
By inspecting Fig. 2.16 we conclude that G(t) and X (t) have the autocor-
relation as indicated in Fig. 2.17a, b, respectively. Thus, the power in the
periodic component is RG (0) = 0.5 and the power in the random component
is R X (0) = 1.
Recall that a periodic signal with period T0 and random timing phase td can
be expressed as a random process:
∞
G(t) = p(t − kT0 − td ), (2.176)
k=−∞
A2
R X (τ ) = R p (τ ), (2.179)
T0
(a) RG (τ )
0.5
−3T −T T 3T τ
0
−2T 2T
−0.5
RX (τ )
(b)
1
−T 0 T
g(t) p(t)
√ T
(c) 0.5 √
0.5
T
t T0 t
0 0
√
− 0.5 √
− 0.5
T
T
td
x(t) p(t)
(d) √
2 √
2
T
T
T
t t
0 0
T0
√
− 2
T
td
Fig. 2.17 a Autocorrelation of G(t). b Autocorrelation of X (t). c Sample function of G(t) and the
corresponding p(t). d Sample function of X (t) and the corresponding p(t)
2 Random Variables and Random Processes 115
Identify h(t).
(b) Obtain a general expression for the autocorrelation of Y (t) in terms of the
autocorrelation of X (t) and the autocorrelation of h(t).
(c) Using the above relation, compute RY (τ ).
where
1 t
h(t) = rect . (2.183)
T T
We know that
∞ ∞
RY (τ ) = h(α)h(β)R X (τ − α + β) dα dβ. (2.184)
α=−∞ β=−∞
Let
α − β = λ. (2.185)
Therefore
116 2 Random Variables and Random Processes
1 T
|λ|
RY (τ ) = R X (τ − λ) 1 − dλ. (2.189)
T λ=−T T
32. Let a random process X (t) be applied at the input of a filter with impulse response
h(t). Let the output process be denoted by Y (t).
(a) Show that
RY X (τ ) = h(τ ) R X (τ ). (2.190)
Therefore
RY X (τ ) = E[Y (t)X (t − τ )]
∞
=E h(α)X (t − α)X (t − τ ) dα . (2.192)
α=−∞
Taking the Fourier transform of both sides and noting that convolution in the
time-domain is equivalent to multiplication in the frequency domain, we get
SY X ( f ) = H ( f )S X ( f ). (2.194)
33. Let X (t1 ) and X (t2 ) be two random variables obtained by observing the random
process X (t) at times t1 and t2 , respectively. It is given that X (t1 ) and X (t2 ) are
uncorrelated. Does this imply that X (t) is WSS?
• Solution: Since X (t1 ) and X (t2 ) are uncorrelated we have
−f2 −f1 0 f1 f2
W
where m X (t1 ) and m X (t2 ) are the means at times t1 and t2 , respectively.
From the above equation, it is clear that X (t) need not be WSS.
34. The voltage at the output of a noise generator, whose statistics is known to be a
WSS Gaussian process, is measured with a DC voltmeter and a true rms (root
mean square) meter. The DC voltmeter reads 3 V. The rms meter, when it is AC
coupled (DC component is removed), reads 2 V.
(a) Find out the pdf of the voltage at any time.
(b) What would be the reading of the rms meter if it is DC coupled?
• Solution: The DC meter reads the average (mean) value of the voltage. Hence
the mean of the voltage is m = 3 V.
When the rms meter is AC coupled, it reads the standard deviation (σ). Hence
σ = 2 V.
Thus the pdf of the voltage takes the form
1
√ e−(v−m) /(2σ ) .
2 2
pV (v) = (2.196)
σ 2π
We know that
35. (Ziemer and Tranter 2002) This problem demonstrates the non-uniqueness of
the representation of narrowband noise. Here we show that the in-phase and
quadrature components depend on the choice of the carrier frequency f c .
A narrowband noise process of the form
(b) f c = f 2
(c) f c = ( f 1 + f 2 )/2
• Solution: It is given that
f 2 − f 1 = W, (2.199)
where f 1 , f 2
W . We know that
S N ( f − f c ) + S N ( f + f c ) for | f | < f c
S Nc ( f ) =
0 otherwise.
= S Ns ( f ). (2.200)
The psd of Nc (t) and Ns (t) for f c = f 1 is indicated in Fig. 2.19. The psd of
Nc (t) and Ns (t) for f c = f 2 is shown in Fig. 2.20. The psd of Nc (t) and Ns (t)
for f c = ( f 1 + f 2 )/2 is shown in Fig. 2.21.
36. Consider the system used for generating narrowband noise, as shown in Fig. 2.22.
The Fourier transform of the bandpass filter is also shown. Assume that the
narrowband noise representation is of the form
The input W (t) is a zero mean, WSS, white Gaussian noise process with a psd
equal to 2 × 10−8 W/Hz.
(a) Sketch and label the psd of Nc (t) when f c = 22 MHz. Show the steps, start-
ing from S N ( f ).
(b) Write down the pdf of the random variable X obtained by observing Nc (t)
at t = 2 s.
(c) Sketch and label the cross-spectral density S Nc Ns ( f ) when f c = 22 MHz.
(d) Derive the expression for the correlation coefficient (ρ(τ )) between Nc (t)
and Ns (t − τ ) for f c = 22 MHz.
(e) For what carrier frequency f c , is ρ(τ ) = 0 for all values of τ ?
Assume the following:
R Nc Ns (τ ) = E[Nc (t)Ns (t − τ )]
R Nc (τ ) = E[Nc (t)Nc (t − τ )]. (2.202)
S N ( f ) = SW ( f )|H ( f )|2
N0
= |H ( f )|2 . (2.203)
2
2 Random Variables and Random Processes 119
SN (f )
N0 /2
−f2 −f1 0 f1 f2
W
SN (f − f1 )
N0 /2
2f1 f
−W 0 f1 + f2
SN (f + f1 )
N0 /2
−2f1 f
−f1 − f2 0 W
SNc (f ) = SNs (f )
N0 /2
−W 0 W
1
√ e−(x−m) /(2σ ) .
2 2
f X (x) = (2.204)
σ 2π
Here
120 2 Random Variables and Random Processes
SN (f )
N0 /2
−f2 −f1 0 f1 f2
W
SN (f − f2 )
N0 /2
f1 + f2 f
0 W 2f2
SN (f + f2 )
N0 /2
−f1 − f2 f
−2f2 −W 0
SNc (f ) = SNs (f )
N0 /2
−W 0 W
m= 0
∞
σ2 = S Nc ( f ) d f = 0.68. (2.205)
f =−∞
where
2 Random Variables and Random Processes 121
SN (f )
N0 /2
−f2 −f1 0 f1 f2
W
SN (f − fc )
N0 /2
−W/2 0 W/2 f1 + fc f2 + fc
SN (f + fc )
N0 /2
SNc (f ) = SNs (f )
N0 /2
−W/2 0 W/2
W (t) N (t)
H(f )
H(f )
3
2
f (MHz)
SN (f ) (×10−8 )
18
8
f (MHz)
SN (f − fc ) (×10−8 )
18
f (MHz)
−2 −1 0 1 43 44 45 46
SN (f + fc ) (×10−8 )
18
8
f (MHz)
26
8
f (MHz)
−2 −1 0 1 2
f 1 = 0.5 × 106
f 2 = 1.5 × 106
B = 106
A1 = 10 × 10−8
A2 = 8 × 10−8 . (2.207)
SN (f ) (×10−8 )
18
8
f (MHz)
SN (f − fc ) (×10−8 )
18
f (MHz)
−2 −1 0 1 43 44 45 46
SN (f + fc ) (×10−8 )
18
8
f (MHz)
10
8
0
−8
−10
f (MHz)
−2 −1 0 1 2
R Nc Ns (τ ) = jA1 B sinc (Bt) e j 2π f1 t − e−j 2π f1 t
+ jA2 B sinc (Bt) e j 2π f2 t − e−j 2π f2 t
= −sinc (Bt) [0.2 sin(2π f 1 t) + 0.16 sin(2π f 2 t)] . (2.208)
−2T −T 0 T 2T
where σ 2 = 0.68.
R Nc Ns (τ ) = 0 for all τ when f c = 22.5 MHz. Hence ρ(τ ) = 0 for all τ when
f c = 22.5 MHz.
37. Consider the function R(τ ) as illustrated in Fig. 2.25. Is it a valid autocorrelation
function? Justify your answer.
• Solution: Note that
t t
R(τ ) = rect + rect
2T 4T
2T sinc (2 f T ) + 4T sinc (4 f T ). (2.210)
Let 2π f T = θ. Hence
sin(θ) sin(2θ)
R(τ ) 2T + 4T
θ 2θ
sin(θ)
2T (1 + 2 cos(θ))
θ
= S(θ). (2.211)
Clearly, S(θ) is negative for 3π/2 ≤ |θ| < 2π, as illustrated in Fig. 2.26 (sin(θ)
is negative and cos(θ) is positive). Therefore S(θ) cannot be the power spectral
density, and R(τ ) is not a valid autocorrelation.
3
S(θ)
-1
-6 -4 -2 0 2 4 6
θ
−W W
which can take negative values. Since the power spectral density cannot be
negative, R(τ ) is not a valid autocorrelation, even though it is an even function.
39. (Haykin 1983) A pair of noise processes are related by
E[N2 (t)N2 (t − τ )]
= E [(N1 (t) cos(2π f c t + θ) − N1 (t) sin(2π f c t + θ))
(N1 (t − τ ) cos(2π f c (t − τ ) + θ) − N1 (t − τ ) sin(2π f c (t − τ ) + θ))]
126 2 Random Variables and Random Processes
SN2 (f )
a/2
−fc − W −fc fc fc + W
R N1 (τ ) cos(2π f c τ ) R N1 (τ ) cos(2π f c τ )
= +
2 2
R N1 (τ ) sin(2π f c τ ) R N1 (τ ) sin(2π f c τ )
+ − .
2 2
= R N1 (τ ) cos(2π f c τ ). (2.215)
Hence
S N1 ( f − f c ) + S N1 ( f + f c )
S N2 ( f ) = . (2.216)
2
The psd for N2 (t) is illustrated in Fig. 2.28.
A2
SX ( f ) = f F ( f ). (2.220)
2
When f is a constant, say equal to f c , then
A2
R X (τ ) = cos(2π f c τ )
2
A2
⇒ SX ( f ) = [δ( f − f c ) + δ( f + f c )] . (2.221)
4
41. (Haykin 1983) A real-valued stationary Gaussian process X (t) with mean m X ,
variance σ 2X and power spectral density S X ( f ) is passed through two real-valued
linear time invariant filters h 1 (t) and h 2 (t), yielding the output processes Y (t)
and Z (t), respectively.
(a) Determine the joint pdf of Y (t) and Z (t − τ ).
(b) State the conditions, in terms of the frequency response of h 1 (t) and
h 2 (t), that are sufficient to ensure that Y (t) and Z (t − τ ) are statistically
independent.
• Solution: The mean values of Y (t) and Z (t − τ ) are given by
∞
E[Y (t)] = E h 1 (α)X (t − α) dα
α=−∞
∞
= mX h 1 (α) dα
α=−∞
= m X H1 (0)
= mY
∞
E[Z (t − τ )] = m X h 2 (α) dα
α=−∞
= m X H2 (0)
= mZ. (2.222)
128 2 Random Variables and Random Processes
= σY2
E[(Z (t − τ ) − m Z )2 ] = E Z 2 (t) − m 2Z
∞
= S X ( f )|H2 ( f )|2 d f − m 2Z
f =−∞
= σ 2Z . (2.223)
E [(Y (t) − m Y ) (Z (t − τ ) − m Z )]
= E[Y (t)Z (t − τ )] − m Y m Z
∞ ∞
= E h 1 (α)X (t − α) dα h 2 (β)X (t − τ − β) dβ
α=−∞ β=−∞
− mY m Z
∞ ∞
= h 1 (α)h 2 (β)R X (τ + β − α) dα dβ − m Y m Z
α=−∞ β=−∞
= K Y Z (τ ). (2.224)
K Y Z (τ )
ρ= . (2.225)
σY σ Z
Now, the joint pdf of two Gaussian random variables y1 and y2 is given by
pY1 , Y2 (y1 , y2 )
1
=
2πσ1 σ2 1 − ρ2
σ 2 (y1 − m 1 )2 − 2σ1 σ2 ρ(y1 − m 1 )(y2 − m 2 ) + σ12 (y2 − m 2 )2
× exp − 2 .
2σ12 σ22 (1 − ρ2 )
(2.226)
The joint pdf of Y (t) and Z (t − τ ) can be similarly found out by substituting
the appropriate values from (2.222)–(2.225). The random variables Y (t) and
Z (t − τ ) are uncorrelated if their covariance is zero. This is possible when
(a) H1 (0) = 0 or H2 (0) = 0 AND
2 Random Variables and Random Processes 129
White
Bandpass Lowpass
noise N (t) Output
filter filter
W (t) N1 (t)
H1 (f ) H2 (f )
cos(2πfc t)
H1 (f ) H2 (f )
2B
1 1
f f
−fc fc
2B
(b) The product H1 ( f )H2 ( f ) = 0 for all f . This can be easily shown by
computing the Fourier transform of K Y Z (τ ), which is equal to (assuming
m Y m Z = 0)
K Y Z (τ ) H1 ( f )H2∗ ( f )S X ( f ). (2.227)
42. (Haykin 1983) Consider a white Gaussian noise process of zero mean and psd
N0 /2 which is applied to the input of the system shown in Fig. 2.29.
(a) Find the psd of the random process at the output of the system.
(b) Find the mean and variance of this output.
We also know that the psd of n c (t) and n s (t) are given by
S N ( f − f c ) + S N ( f + f c ) for − B ≤ f ≤ B
S Nc ( f ) = S Ns ( f ) = (2.229)
0 elsewhere,
N0 for − B ≤ f ≤ B
R Nc (τ ) S Nc ( f ) = (2.230)
0 elsewhere.
130 2 Random Variables and Random Processes
SN (f ) SN1 (f )
2B
N0 /2
N0 /4
f f
−fc fc
2B
(a) (b)
Fig. 2.30 a Noise psd at the output of H1 ( f ). b Noise psd at the output of H2 ( f )
H2 ( f )
n(t) cos(2π f c t) −→ n c (t)/2 (2.231)
whose autocorrelation is
n c (t) n c (t − τ ) R Nc (τ ) SN ( f )
E = c . (2.232)
2 2 4 4
The psd of noise at the output of H2 ( f ) is illustrated in Fig. 2.30b. The output
noise is zero-mean Gaussian with variance
N0 N0 B
× 2B = . (2.233)
4 2
43. (Haykin 1983) Consider a narrowband noise process N (t) with its Hilbert trans-
form denoted by N̂ (t). Show that the cross-correlation functions of N (t) and
N̂ (t) are given by
R N N̂ (τ ) = − R̂ N (τ )
R N̂ N (τ ) = R̂ N (τ ), (2.234)
R X (τ ) = e−2ν|τ | , (2.236)
SY ( f ) = S X ( f )|H ( f )|2
ν 1
= 2 ×
ν +π f 2 2 1 + (2π f RC)2
A B
= 2 + . (2.239)
ν + π2 f 2 1 + (2π f RC)2
ν
A=
1 − 4R 2 C 2 ν 2
−4R 2 C 2 ν
B= . (2.240)
1 − 4R 2 C 2 ν 2
The autocorrelation of the output random process is the inverse Fourier trans-
form of (2.239), and is given by
αe−αx x > 0
f X (x) =
0 otherwise
βe−β y y > 0
f Y (y) = (2.242)
0 otherwise,
When α = β we have
z
f Z (z) = α2 e−αz da
2 −αz a=0
zα e for z > 0
= (2.244)
0 otherwise.
46. A WSS random process X (t) with psd N0 /2 is passed through a first-order RC
highpass filter. Determine the psd and autocorrelation of the filter output.
• Solution: The transfer function of the first-order RC highpass filter is
2 Random Variables and Random Processes 133
j 2π f RC
H( f ) = . (2.245)
1 + j 2π f RC
Let Y (t) denote the random process at the filter output. Therefore the psd of
the filter output is
N0 (2π f RC)2
SY ( f ) = ×
2 1 + (2π f RC)2
N0 1
= 1− . (2.246)
2 1 + (2π f RC)2
1 −|t| 1
e . (2.247)
2 1 + (2π f )2
g(t) G( f )
1
⇒ g(at) G( f /a) (2.248)
|a|
1 −|t|/(RC) RC
e . (2.249)
2 1 + (2π f RC)2
N0 N0 −|t|/(RC)
RY (τ ) = δ(τ ) − e . (2.250)
2 4RC
47. Let X be a random variable having pdf
• Solution: We have
∞
2 ae−bx = 1
x=0
⇒ 2a = b. (2.252)
134 2 Random Variables and Random Processes
Now
48. A narrowband noise process N (t) has zero mean and autocorrelation function
R N (τ ). Its psd S N ( f ) is centered about ± f c . Its quadrature components Nc (t)
and Ns (t) are defined by
R Nc (τ ) = R Ns (τ ) = R N (τ ) cos(2π f c τ ) + R̂ N (τ ) sin(2π f c τ )
R Nc Ns (τ ) = −R Ns Nc (τ ) = R N (τ ) sin(2π f c τ ) − R̂ N (τ ) cos(2π f c τ ).
(2.255)
R Nc (τ ) = E [Nc (t)Nc (t − τ )]
= E N (t) cos(2π f c t) + N̂ (t) sin(2π f c t)
N (t − τ ) cos(2π f c (t − τ )) + N̂ (t − τ ) sin(2π f c (t − τ ))
R N (τ )
= [cos(2π f c τ ) + cos(4π f c t − 2π f c τ )]
2
R (τ )
+ N N̂ [sin(4π f c t − 2π f c τ ) − sin(2π f c τ )]
2
R (τ )
+ N̂ N [sin(4π f c t − 2π f c τ ) + sin(2π f c τ )]
2
R (τ )
+ N̂ [cos(2π f c τ ) − cos(4π f c t − 2π f c τ )] . (2.256)
2
2 Random Variables and Random Processes 135
Now
R N̂ N (τ ) = −R N N̂ (τ ) = R̂ N (τ ). (2.258)
R Nc Ns (τ )
= E [Nc (t)Ns (t − τ )]
= E N (t) cos(2π f c t) + N̂ (t) sin(2π f c t)
N̂ (t − τ ) cos(2π f c (t − τ )) − N (t − τ ) sin(2π f c (t − τ ))
R N N̂ (τ )
= [cos(2π f c τ ) + cos(4π f c t − 2π f c τ )]
2
R N (τ )
− [sin(4π f c t − 2π f c τ ) − sin(2π f c τ )]
2
R (τ )
+ N̂ [sin(4π f c t − 2π f c τ ) + sin(2π f c τ )]
2
R (τ )
− N̂ N [cos(2π f c τ ) − cos(4π f c t − 2π f c τ )] . (2.260)
2
Once again substituting (2.257) and (2.258) in (2.260) we get
N0
RY (τ ) = δ(τ ). (2.262)
2
The autocorrelation of the output is given by
∞ ∞
R X (τ ) = E h(α)Y (t − α) dα h(β)Y (t − τ − β) dβ
α=−∞ β=−∞
∞ ∞
= h(α)h(β)RY (τ + β − α) dα dβ
α=−∞ β=−∞
∞
N0 ∞
= h(α)h(β)δ(τ + β − α) dα dβ
2 α=−∞ β=−∞
∞
N0
= h(α)h(α − τ ) dα
2 α=−∞
N0
= (h(t) h(−t)) . (2.263)
2
The transfer function must satisfy the following relationship:
N0
SX ( f ) = |H ( f )|2 . (2.264)
2
50. White Gaussian noise W (t) of zero mean and psd N0 /2 is applied to a first-
order RC-lowpass filter, producing noise N (t). Determine the pdf of the random
variable obtained by observing N (t) at time tk .
• Solution: Clearly, the first-order RC lowpass filter is stable and linear, hence
N (t) is also a Gaussian random process. Hence N (tk ) is a Gaussian distributed
random variable with pdf given by
1
√ e−(x−m) /(2σ ) ,
2 2
p N (tk ) (x) = (2.265)
σ 2π
1/(j 2π f C)
H( f ) =
R + 1/(j 2π f C)
1
=
1 + j 2π f RC
1
⇒ |H ( f )|2 = . (2.268)
1 + (2π f RC)2
Hence
N0 1 −1 ∞ N0
σ 2 = E N 2 (tk ) = tan (2π f RC) f =−∞ = .(2.269)
2 2π RC 4RC
51. Let X and Y be two independent random variables. X is uniformly distributed
between [−1, 1] and Y is uniformly distributed between [−2, 2]. Find the pdf
of Z = max(X, Y ).
• Solution: We begin by noting that Z lies in the range [−1, 2]. Next, we com-
pute the cumulative distribution function of Z and proceed to compute the pdf
by differentiating the distribution function.
To compute the cumulative distribution function, we note that
X if X > Y
Z= (2.270)
Y if Y > X.
1/2 for − 1 ≤ x ≤ 1
f X (x) =
0 otherwise
1/4 for − 2 ≤ y ≤ 2
f Y (y) = (2.271)
0 otherwise.
We also note that since the pdfs of X and Y are different in the range [−1, 1]
and [1, 2], we expect the distribution function of Z to be different in these
two regions.
Using the fact that X and Y are independent and the events X > Y and Y > X
are mutually exclusive we have for −1 ≤ z ≤ 1
138 2 Random Variables and Random Processes
m Z = E[Z ] = am X + bm Y . (2.275)
The variance of Z is
E (Z − m Z )2 = E Z 2 − m 2Z
= a 2 E X 2 + b2 E Y 2 + 2abE[X ]E[Y ]
− (am X + bm Y )2
2 Random Variables and Random Processes 139
= a 2 (σ 2 + m 2X ) + b2 (σ 2 + m 2Y ) + 2abm X m Y
− (am X + bm Y )2
= σ 2 (a 2 + b2 ). (2.276)
X if X < Y
Z= (2.277)
Y if Y < X.
1/2 for − 1 ≤ x ≤ 1
f X (x) =
0 otherwise
1/4 for − 2 ≤ y ≤ 2
f Y (y) = (2.278)
0 otherwise.
Moreover, since the pdfs of X and Y are different in the region [−2, −1]
and [−1, 1], we also expect the cumulative distribution function of Z to be
different in these two regions.
Let us first consider the region [−2, −1]. Using the fact that X and Y are
independent and the events X < Y and Y < X are mutually exclusive we
have for −2 ≤ z ≤ −1
1/4 for − 2 ≤ z ≤ −1
f Z (z) = . (2.283)
(1/8) [−2z + 3] for − 1 ≤ z ≤ 1
54. (Haykin 1983) Let X (t) be a stationary Gaussian process with zero mean and
variance σ 2 and autocorrelation R X (τ ). This process is applied to an ideal half-
wave rectifier. The random process at the rectifier output is denoted by Y (t).
(a) Compute the mean value of Y (t).
(b) Compute the autocorrelation of Y (t). Make use of the result
∞ ∞ 1 − θ cot(θ)
uv exp −u 2 − v 2 − 2uv cos(θ) du dv = .
(2.284)
u=0 v=0 4 sin2 (θ)
X (t) if X (t) ≥ 0
Y (t) = g(X ) = (2.285)
0 if X (t) < 0.
E [Y (t)Y (t − τ )] . (2.287)
X (t) = α
X (t − τ ) = β. (2.288)
The joint pdf of two real-valued Gaussian random variables Y1 and Y2 is given
by
f Y1 , Y2 (y1 , y2 )
1
=
2πσ1 σ2 1 − ρ2
σ 2 (y1 − m 1 )2 − 2σ1 σ2 ρ(y1 − m 1 )(y2 − m 2 ) + σ12 (y2 − m 2 )2
× exp − 2 .
2σ12 σ22 (1 − ρ2 )
(2.290)
E[α] = E[β] = 0
E[α2 ] = E[β 2 ] = σ 2
E[αβ] R X (τ )
ρ = ρ(τ ) = = . (2.292)
σ 2 σ2
Thus the autocorrelation of Y (t) can be written as
∞ ∞
1 α2 − 2ραβ + β 2
RY (τ ) = αβ exp − dα dβ.
2πσ 2 1 − ρ2 α=0 β=0 2σ 2 (1 − ρ2 )
(2.293)
Let
1
A=
2σ 2 (1 − ρ2 )
1
B=
2σ 2 (1 − ρ2 )
u = Aα
v = −Bβ. (2.294)
Thus
∞ ∞
2σ 2 (1 − ρ2 )3/2
RY (τ ) = uv exp −(u 2 + 2uvρ + v 2 ) du dv
π u=0 v=0
(2.295)
Using (2.284) we get
σ 2
RY (τ ) = 1 − ρ2 − ρ cos−1 (ρ) . (2.296)
2π
55. (Haykin 1983) Let X (t) be a real-valued zero-mean stationary Gaussian process
with autocorrelation function R X (τ ). This process is applied to a square-law
device. Denote the output process as Y (t). Compute the mean and autocovariance
of Y (t).
• Solution: It is given that
Let
X (t) = X 1
X (t − τ ) = X 2 . (2.300)
We know that
where E[X i X j ] = Ci j and the random variables X i are jointly normal (Gaus-
sian) with zero mean. For the given problem we note that X 3 = X 1 and
X 4 = X 2 . Thus
RY (τ ) = 2R 2X (τ ) + R 2X (0). (2.302)
U = |X |
V = −Y. (2.304)
Z = U + V, (2.305)
where U and V are independent. Clearly Z lies in the range [−8, −3]. More-
over, the pdf of Z is given by the convolution of the pdfs of U and V and is
equal to
∞
f Z (z) = fU (α) f V (z − α) dα. (2.306)
α=−∞
144 2 Random Variables and Random Processes
fX (x)
(a)
1/5
x
−1 0 4
fY (y)
(b)
1
y
0 7 8
fU (u)
(c)
2/5
1/5
u
0 1 4
fV (v)
(d)
1
v
−8 −7 0
The various steps in the convolution of U and V are illustrated in Fig. 2.32.
We have
f Z (−8) = 0
f Z (−7) = 2/5
f Z (−6) = 1/5
f Z (−4) = 1/5
f Z (−3) = 0. (2.307)
0 1 4
fV (−8 − α)
(b)
1
α
−1 0
fV (−7 − α)
(c)
1
α
0 1
fV (−6 − α)
(d)
1
α
0 1 2
fV (−4 − α)
(e)
1
α
0 3 4
fV (−3 − α)
(f)
1
α
0 4 5
2/5
1/5
z
−8 −7 −6 −4 −3 0
146 2 Random Variables and Random Processes
0.2
0.15
fU (u)
0.1
0.05
0
0 2 4 6 8 10
u
and
3 2
f Y (y) = δ(y + 1) + δ(y − 7). (2.309)
5 5
Now, the pdf of V = |Y | is
3 2
f V (v) = δ(v − 1) + δ(v − 7). (2.310)
5 5
Z = U + V. (2.312)
Since X and Y are independent, U and V are also independent. Moreover, the
pdf of Z is given by the convolution of the pdfs of U and V and is equal to
∞
f Z (z) = fU (α) f V (z − α) dα
α=−∞
3 2
= fU (z − 1) + fU (z − 7). (2.313)
5 5
The pdf of Z is shown in Fig. 2.35.
2 Random Variables and Random Processes 147
0.1
0.08
fZ (z)
0.06
0.04
0.02
0
0 2 4 6 8 10 12 14 16
z
58. Let X (t) be a zero-mean Gaussian random process with psd N0 /2. X (t) is passed
through a filter with impulse response h(t) = exp(−πt 2 ), followed by an ideal
differentiator. Let the output process be denoted by Y (t). Determine the pdf of
Y (1).
• Solution: At the outset, we emphasize that Y (t) can be considered to be a
random process as well as a random variable. Recall that a random variable is
obtained by observing a random process at a particular instant of time.
Note that both h(t) and the ideal differentiator are LTI filters. Hence Y (t) is
also a Gaussian random process. Since X (t) is WSS, so is Y (t). Therefore the
random variable Y (t) has a Gaussian pdf, independent of t. It only remains
to compute the mean and the variance of Y (t). Since X (t) is zero mean, so is
Y (t). Therefore
H ( f ) = exp(−π f 2 ). (2.315)
The overall frequency response of h(t) in cascade with the ideal differentiator
is
G( f ) = j 2π f H ( f ). (2.316)
The variance of the random variable Y (t) (this is also the power of the random
process Y (t)) is
σY2 = E |Y (t)|2
∞
= SY ( f ) d f
f =−∞
√
2
= N0 π. (2.318)
4
Hence the pdf of the random variable Y (t) is
1
pY (y) = √ exp(−y 2 /(2σY2 )) (2.319)
σY 2π
independent of t, and is hence also the pdf of the random variable Y (1).
59. Let X 1 and X 2 be two independent random variables. X 1 is uniformly distributed
between [−1, 1] and X 2 has a triangular pdf between [−3, 3] with a peak at
zero. Find the pdf of Z = max(X 1 , X 2 ).
• Solution: Let us consider a more general problem given by
Z = max(X 1 , . . . , X n ), (2.320)
Therefore
d
f Z (z) = P(Z < z)
dz
d
= [P(X 1 < z) . . . P(X n < z)] . (2.322)
dz
Moreover, it can be seen that Z lies in the range [−1, 3]. In order to compute
(2.323), we need to partion the range of Z into three intervals, given by [−1, 0],
[0, 1] and [1, 3]. Now
2 Random Variables and Random Processes 149
⎧
⎨ (z + 1)/2 for − 1 ≤ z ≤ 0
P(X 1 < z) = (z + 1)/2 for 0 ≤ z ≤ 1 , (2.324)
⎩
1 for 1 ≤ z ≤ 3
Z = min(X 1 , . . . , X n ), (2.328)
Therefore
d
f Z (z) = P(Z < z)
dz
d
= [1 − P(Z > z)]
dz
150 2 Random Variables and Random Processes
d
= [1 − P(X 1 > z) . . . P(X n > z)]
dz
d
= [1 − (1 − P(X 1 < z)) . . . (1 − P(X n < z))] . (2.330)
dz
Moreover, it can be seen that Z lies in the range [−3, 1]. In order to com-
pute (2.331), we need to partion the range of Z into three intervals, given by
[−3, −1], [−1, 0] and [0, 1]. Now
References
1. (Haykin 1983) Suppose that a non-linear device is available for which the output
current i 0 and the input voltage vi are related by:
where a1 and a3 are constants. Explain how this device may be used to provide
(a) a product modulator (b) an amplitude modulator.
• Solution: Let
a3 A3c
i 0 (t) = a1 (Ac cos(π f c t) + m(t)) + (3 cos(π f c t) + cos(3π f c t))
4
3
+ a3 A2c (1 + cos(2π f c t)m(t)) + 3a3 Ac cos(π f c t)m 2 (t)
2
+ a3 m 3 (t)
3 3
= a1 + a3 A2c m(t) + a3 m 3 (t) + a1 Ac + a3 A3c cos(π f c t)
2 4
3
+ 3a3 Ac cos(π f c t)m 2 (t) + a3 A2c m(t) cos(2π f c t)
2
a3 A3c
+ cos(3π f c t). (3.3)
4
Using the fact that multiplication in the time domain is equivalent to convo-
lution in the frequency domain, we know that m 2 (t) occupies the frequency
© The Editor(s) (if applicable) and The Author(s), under exclusive license 153
to Springer Nature Switzerland AG 2021
K. Vasudevan, Analog Communications,
https://doi.org/10.1007/978-3-030-50337-6_3
154 3 Amplitude Modulation
−3W −W 0 W 3W fc /2 fc 3fc /2
fc /2 + 2W
fc − W
3
m(t) s(t) = a A2
2 3 c
cos(2πfc t)m(t)
band [−2W, +2W ] and m 3 (t) occupies [−3W, 3W ]. The Fourier transform
of i 0 (t) is illustrated in Fig. 3.1. To extract the DSBSC component at f c using
a bandpass filter, the following conditions must be satisfied:
f c /2 + 2W < f c − W
f c + W < 3 f c /2. (3.4)
The procedure for obtaining the DSBSC signal is illustrated in Fig. 3.2.
The procedure for obtaining the AM signal using two identical nonlinear
devices and bandpass filters is illustrated in Fig. 3.3. The amplitude sensitivity
is controlled by A0 .
Note that with this method, the amplitude sensitivity and the carrier power
can be independently controlled. Moreover, the amplitude sensitivity is inde-
pendent of a3 , which is not under user control (a3 is device dependent).
2. (Haykin 1983) In this problem we consider the switching modulator shown
in Fig. 3.4. Assume that the carrier wave c(t) applied to the diode is large in
amplitude compared to |m(t)|, so that the diode acts like an ideal switch, that is
3 Amplitude Modulation 155
s1 (t)
Ac cos(πfc t) Nonlinear
BPF
device
m(t) AM
3
s1 (t) = a A2
2 3 c
cos(2πfc t)m(t)
signal
3
s2 (t) = a A A2
2 3 0 c
cos(2πfc t)
Ac cos(πfc t) Nonlinear
BPF
device s2 (t)
A0
v1 (t) c(t) > 0
v2 (t) = (3.6)
0 c(t) < 0.
where
∞
1 2 (−1)n−1
g p (t) = + cos(2π f c (2n − 1)t). (3.8)
2 π n=1 2n − 1
• Solution: We have
Ac
v2 (t) = cos(2π f c t)
2
∞
Ac (−1)n−1
+ (cos(4π(n − 1) f c t) + cos(4πn f c t))
π n=1 2n − 1
∞
m(t) 2 (−1)n−1
+ + cos(2π(2n − 1) f c t)m(t). (3.9)
2 π n=1 2n − 1
Ac 2
cos(2π f c t) + cos(2π f c t)m(t). (3.10)
2 π
3. (Haykin 1983) Consider the AM signal
x = 2π f m t. (3.13)
2
v(x)
1.5
0.5
0
-3 -2 -1 0 1 2 3
x
3 Amplitude Modulation 157
Now
Since v(x) is an even function, it has a Fourier series representation given by:
∞
v(x) = a0 + 2 an cos(nx), (3.16)
n=1
where
π−π/3 π
2 2
a0 = (1 + 2 cos(x)) d x − (1 + 2 cos(x)) d x
2π x=0 2π π−π/3
√
1 2 3
= + . (3.17)
3 π
Similarly
π−π/3
2
an = (1 + 2 cos(x)) cos(nx) d x
2π
π
x=0
2
− (1 + 2 cos(x)) cos(nx) d x
2π π−π/3
2 sin(2nπ/3) sin(2(n − 1)π/3) sin(2(n + 1)π/3)
= + + .
π n n−1 n+1
(3.18)
3.5
2.5
v(x)
1.5
0.5
0
-6 -4 -2 0 2 4 6
x
Fig. 3.6 Output of the ideal envelope detector for α = π/2
x = 2π f m t. (3.21)
Note that v(x) is periodic with a period 2π. If the real Fourier series represen-
tation of v(x) is to contain only dc and cosine terms, we require:
α = π/2. (3.22)
Therefore
v(x) = |1 + 3 cos(x)|
∞
= a0 + 2 an cos(nx). (3.23)
n=1
3 Amplitude Modulation 159
Let
v(x1 ) = |1 + 3 cos(x1 )|
=0
⇒ x1 = π − cos−1 (1/3)
= 1.911 rad. (3.24)
Now
x1 π
2 2
a0 = (1 + 3 cos(x)) d x − (1 + 3 cos(x)) d x
2π x=0 2π x1
4x1 − 2π + 12 sin(x1 )
=
2π
= 2.02 (3.25)
and
x1
2
an = (1 + 3 cos(x)) cos(nx) d x
2π
π x=0
2
− (1 + 3 cos(x)) cos(nx) d x
2π x1
2 sin(nx1 ) 3 sin((n − 1)x1 ) 3 sin((n + 1)x1 )
= + + for n > 1
π n π n−1 π n+1
(3.26)
and
x1
2
a1 = (1 + 3 cos(x)) cos(x) d x
2π
π x=0
2
− (1 + 3 cos(x)) cos(x) d x
2π x1
2 3 3 sin(2x1 ) 3
= sin(x1 ) + x1 + − . (3.27)
π π π 2 2
5. (Haykin 1983) The AM signal
is applied to the system in Fig. 3.7. Assuming that |ka m(t)| < 1 for all t and m(t)
is bandlimited to [−W, W ], and that the carrier frequency f c > 2W , which show
that m(t) can be obtained from the square rooter output, v3 (t).
160 3 Amplitude Modulation
−W 0 W f
• Solution: We have
A2c
v1 (t) = [1 + cos(4π f c t)] [1 + ka m(t)]2 . (3.29)
2
We also have (assuming an ideal LPF)
A2c
v2 (t) = [1 + ka m(t)]2 . (3.30)
2
Therefore
Ac
v3 (t) = √ [1 + ka m(t)] . (3.31)
2
6. (Haykin 1983) Consider a message signal m(t) with spectrum shown in Fig. 3.8,
with W = 1 kHz. This message is DSB-SC modulated using a carrier of the form
Ac cos(2π f c t), producing the signal s(t). The modulated signal is next applied
to a coherent detector with carrier A0 cos(2π f c t). Determine the spectrum of the
detector output when the carrier is (a) f c = 1.25 kHz (b) f c = 0.75 kHz. Assume
that the LPF in the demodulator is ideal with unity gain. What is the lowest carrier
frequency for which there is no aliasing (no overlap in the frequency spectrum)
in the modulated signal s(t)?
• Solution: We have
−W W
A0 cos(2πfc t)
Fig. 3.9 A coherent demodulator for DSB-SC signals. An ideal LPF is assumed
G(f )
Ac A0 /2
Ac A0 /4
f kHz
Ac A0 /2
f kHz
−1 0 1
Fig. 3.10 Spectrum at the outputs of the multiplier and the LPF for f c = 1.25
2 fc − W > W
⇒ f c > W. (3.34)
162 3 Amplitude Modulation
G(f )
Ac A0 /2
Ac A0 /4
f kHz
H(f )
Ac A0 /2
Ac A0 /4
Ac A0 /8
f kHz
−1 −0.5 0 0.5 1
Fig. 3.11 Spectrum at the outputs of the multiplier and the LPF for f c = 0.75
The carrier power is 100 W and the power efficiency (ratio of sideband power
to total power) is 40%. Compute A, B and the modulation factor μ.
• Solution: The general expression for a tone modulated AM signal is:
A = Ac
μA
B=
2
f c = 200 Hz
f c − f m = 180 Hz
f c + f m = 220 Hz. (3.37)
3 Amplitude Modulation 163
A2
= 100W
2 √
⇒ A = 200. (3.38)
Since the given AM wave is tone modulated, the power efficiency is given by:
μ2 2
=
2 + μ2 5
⇒ μ = 1.155. (3.39)
μA
B= = 8.165. (3.40)
2
8. Consider a modulating wave m(t) such that (1 + ka m(t)) > 0 for all t. Assume
that the spectrum of m(t) is zero for | f | > W . Let
Let
2π f c t = x. (3.44)
1 1
Kn = sin((n − 1)π/2) + sin((n + 1)π/2). (3.47)
π(n − 1) π(n + 1)
2 Ac
v(t) = Ac [1 + ka m(t)]a0 = [1 + ka m(t)]. (3.48)
π
9. (Haykin 1983) This problem is related to the 2two-stage approach for generating
SSB (frequency discrimination method) signals. A voice signal m(t) occupying
the frequency band 0.3–3.4 kHz is to be SSB modulated with only the upper
sideband transmitted. Assume the availability of bandpass filters which provide
an attenuation of 50 dB in a transition band that is one percent of the center
frequency of the bandpass filter, as illustrated in Fig. 3.12a. Assume that the first
stage eliminates the lower sideband and the product modulator in the second
stage uses a carrier frequency of 11.6 MHz. The message spectrum is shown
in Fig. 3.12b and the spectrum of the final SSB signal must be as shown in
Fig. 3.12c.
Find the range of the carrier frequencies that can be used by the product modulator
in the first stage, so that the unwanted sideband is attenuated by no less than 50 dB.
• Solution: Firstly, we note that if only a single stage were was employed, the
required Q-factor of the BPF would be very high, since
centre frequency
Q-factor ≈ . (3.49)
transition bandwidth
Secondly, we would like to align the center frequency of the bandpass filter
in the center of the message band to be transmitted. The transmitted message
band could be the upper or the lower sideband. Let us denote the carrier
frequency of the first stage by f 1 and that of the second stage by f 2 . Then, the
3 Amplitude Modulation 165
20 log |H(f )| dB
0
(a)
−50
f
−fc fc
M (f ) 0.01fc S(f )
(b) (c)
f f
Fig. 3.12 a Magnitude response of the bandpass filters. b Message spectrum. c Spectrum of the
final SSB signal
center frequency of the BPF at the second stage is given by (see Fig. 3.13):
1
f c2 = [ f2 + f1 + fa + f2 + f1 + fb ] . (3.50)
2
In the above equation:
f 2 = 11600 kHz
f a = 0.3 kHz
f b = 3.4 kHz. (3.51)
The transition bandwidth of the second BPF is 0.01 f c2 . The actual transition
bandwidth at the output of the first SSB modulator is 2( f 1 + f a ). We require
that
2( f 1 + f a ) ≥ 0.01 f c2
⇒ f 1 (1 − 0.005) ≥ 58 + 0.01 × 0.925 − 0.3
f 1 ≥ 57.999 kHz. (3.52)
1
f c1 = [ f1 + fa + f1 + fb ]
2
= f 1 + 1.85. (3.53)
166 3 Amplitude Modulation
M (f )
−fb −fa 0 fa fb
(a)
S1 (f )
Fig. 3.13 a Message spectrum. b SSB signal at the output of the first stage. c SSB signal at the
output of the second stage
The transition bandwidth of the first BPF is 0.01 f c1 and the actual transition
bandwidth of the input message is 2 f a . We require that:
2 f a ≥ 0.01 f c1
⇒ f 1 ≤ 58.15 kHz. (3.54)
10. (Haykin 1983) This problem is related to the 2two-stage approach for generating
SSB signals. A voice signal m(t) occupying the frequency band 0.3–3 kHz is to
be SSB modulated with only the lower sideband transmitted. Assume the avail-
ability of bandpass filters which provide an attenuation of 60 dB in a transition
band that is two percent of the center frequency of the bandpass filter, as illus-
trated in Fig. 3.14a. Assume that the first stage eliminates the upper sideband and
the product modulator in the second stage uses a carrier frequency of 1 MHz.
3 Amplitude Modulation 167
20 log |H(f )| dB
0
(a)
−60
f
−fc fc
0.02fc
M (f ) S(f )
(b) (c)
f f
−3 −0.3 0 0.3 3 0
Fig. 3.14 a Magnitude response of the bandpass filters. b Message spectrum. c Spectrum of the
final SSB signal
The message spectrum is shown in Fig. 3.14b and the spectrum of the final SSB
signal must be as shown in Fig. 3.14c.
(a) Find the range of carrier frequencies that can be used by the product mod-
ulator in first stage, so that the unwanted sideband is attenuated by no less
than 60 dB.
(b) Write down the expression of the SSB signal s(t) at the output of the second
stage, in terms of m(t). Clearly specify the carrier frequency of s(t) in terms
of the carrier frequency of the two product modulators.
• Solution: Firstly, we note that if only a single stage were was employed, the
required Q-factor of the BPF would be very high, since
centre frequency
Q-factor ≈ . (3.55)
transition bandwidth
Secondly, we would like to align the center frequency of the bandpass filter
in the center of the message band to be transmitted. The transmitted message
band could be the upper or the lower sideband. Let us denote the carrier
frequency of the first stage by f 1 and that of the second stage by f 2 . Then, the
center frequency of the BPF at the second stage is given by (see Fig. 3.15):
1
f c2 = [ f2 + f1 − fb + f2 + f1 − fa ] . (3.56)
2
In the above equation:
168 3 Amplitude Modulation
M (f )
(a)
−fb −fa 0 fa fb
S1 (f )
(b)
−f1 + fa −f1 + fb f1 − fb f1 − fa
S(f )
(c)
−f2 − f1 + fb f2 + f1 − fb
−f2 − f1 + fa f2 + f1 − fa
Fig. 3.15 a Message spectrum. b SSB signal at the output of the first stage. c SSB signal at the
output of the second stage
f 2 = 1000 kHz
f a = 0.3 kHz
f b = 3.0 kHz. (3.57)
The transition bandwidth of the second BPF is 0.02 f c2 . The actual transition
bandwidth at the output of the first SSB modulator is 2( f 1 − f b ). We require
that
2( f 1 − f b ) ≥ 0.02 f c2
⇒ f 1 − f b ≥ 0.01[ f 2 + f 1 − 0.5( f a + f b )]
f 1 ≥ 13.1146 kHz. (3.58)
Envelope
Input Output
BPF
detector
cos(2πf0 t)
Variable
frequency oscillator
M1 (f ) M2 (f )
fc1 fc2
2W 2W
1
f c1 = [ f1 − fa + f1 − fb ]
2
= f 1 − 1.65. (3.59)
The transition bandwidth of the first BPF is 0.02 f c1 and the actual transition
bandwidth of the input message is 2 f a . We require that:
2 f a ≥ 0.02 f c1
⇒ f 1 ≤ 31.65 kHz. (3.60)
where f c = f 2 + f 1 − f a + f a = f 2 + f 1 .
11. Consider the receiver in Fig. 3.16, for a frequency division multiplexing (FDM)
system. An FDM system is similar to the simultaneous radio broadcast by many
stations. Assume that the BPF is ideal with a bandwidth of 2W = 10 kHz.
(a) Now consider the input signal in Fig. 3.17. Assume that the input signals are
amplitude modulated, f 0 = 1.4 MHz, f c1 = 1.1 MHz, and f c2 = 1.7 MHz.
170 3 Amplitude Modulation
Assume that the BPF is centered at 300 kHz. Find out the expression for the
signal at the output of the BPF. The frequency f c2 is called the image of f c1 .
(b) Now assume that f c1 = 1.1 MHz is the desired carrier and f c2 = 1.7 MHz
is to be rejected. Assume that f 0 can only be varied between f c1 and f c2 and
that the center frequency of the BPF cannot be less than 200 kHz (this is
required for proper envelope detection) and cannot be greater than 300 kHz
(so that the Q-factor of the BPF is within practical limits). Find out the
permissible range of values f 0 can take. Also find out the corresponding
permissible range of values of the center frequency of the BPF.
A1
s(t) = [1 + ka1 m 1 (t)] cos(2π( f 0 − f c1 )t) + cos(2π( f 0 + f c1 )t)
2
A2
+ [1 + ka2 m 2 (t)] cos(2π( f c2 − f 0 )t) + cos(2π( f c2 + f 0 )t) .
2
(3.63)
A1
y(t) = [1 + ka1 m 1 (t)] cos(2π( f 0 − f c1 )t)
2
A2
+ [1 + ka2 m 2 (t)] cos(2π( f c2 − f 0 )t). (3.64)
2
Now consider Fig. 3.18. Denote the two difference (IF) frequencies by x and
B − x, where B = f c2 − f c1 . To reject B − x, we must have:
B − x − x > 2W
⇒ x < 295. (3.65)
Thus, the BPF center frequency can vary between 200 and 295 kHz. Corre-
spondingly, f 0 can vary between 1.3 and 1.395 MHz.
12. Consider the Costas loop for AM signals as shown in Fig. 3.19. Let the received
signal r (t) be given by:
x B−x
fc1 f0 fc2
x B−x
2W
Fig. 3.18 Figure to illustrate the condition to be satisfied by the two difference frequencies, so that
one of them is rejected by the BPF
demodulated signal
LPF
In-phase arm
x1 (t) z1 (T )
Integrator
cos(2πfc t + α)
phase
r(t) y (T )
VCO
detector
sin(2πfc t + α)
Integrator
x2 (t) z2 (T )
Quadrature arm
where w(t) is zero-mean additive white Gaussian noise (AWGN) with psd N0 /2.
The random variable z 1 (T ) is computed as:
T
1
z 1 (T ) = x1 (t) dt. (3.67)
T 0
z 2 (T )
y (T ) = . (3.68)
z 1 (T )
172 3 Amplitude Modulation
Assume that:
(a) θ and α are uniformly distributed random variables.
(b) α − θ is a constant and close to zero.
(c) w(t) and α are statistically independent.
(d) m(t) has zero -mean (zero dc).
(e) T is large, hence the integrator output is a dc term plus noise.
(f) The signal-to-noise ratio (SNR) at the input to the phase detector is high.
Compute the mean and the variance of the random variables z 1 (T ), z 2 (T ) and
y (T ).
• Solution: We have
Ac
x1 (t) = [1 + ka m(t)] [cos(α − θ) + cos(4π f c t + α + θ)] + a1 (t)
2
Ac
x2 (t) = [1 + ka m(t)] [sin(α − θ) + sin(4π f c t + α + θ)] + a2 (t)
2
, (3.69)
where
Ac
z 1 (T ) = cos(α − θ) + b1
2
Ac
z 2 (T ) = sin(α − θ) + b2 , (3.73)
2
where
3 Amplitude Modulation 173
1 T
b1 = a1 (t) dt
T t=0
T
1
, b2 = a2 (t) dt. (3.74)
T t=0
Then
1 T
E[b1 ] = E[a1 (t)] dt
T t=0
=0
= E[b2 ]. (3.75)
In the above equation we have used the Rayleigh’s energy theorem. Thus:
Ac
E[z 1 (T )] = cos(α − θ)
2
Ac
E[z 2 (T )] = sin(α − θ)
2
N0
var [z 1 (T )] =
4T
N0
var [z 2 (T )] = . (3.77)
4T
The phase detector output is:
M (f )
f (kHz)
wWhere we have made the high SNR approximation. Hence, the mean and
variance of y (T ) is:
E[y (T )]] = α − θ
4 N0
var [y (T )] = 2 . (3.79)
Ac 4T
13. Consider a message signal m(t) whose spectrum extends over 300 Hz to 3.4 kHz,
as illustrated in Fig. 3.20. This message is SSB modulated to obtain:
M1 (f )
(a)
Ac /2
f (kHz)
M2 (f )
(b)
Ac /2
f (kHz)
Ac
s2 (t) = m(t) cos(2π f t) − m̂(t) sin(2π f t) . (3.82)
2
When f is positive, s2 (t) is an SSB signal with carrier frequency f and
upper sideband transmitted. This is illustrated in Fig. 3.21a, for f = 10 Hz.
When f is negative, s2 (t) is an SSB signal with carrier frequency f and
lower sideband transmitted. This is illustrated in Fig. 3.21b with f = −10.
15. For the message signal shown in Fig. 3.22 compute the power efficiency (the
ratio of sideband power to total power) in terms of the amplitude sensitivity ka ,
A1 , A2 , and T .
The message is periodic with period T , has zero -mean, and is AM modulated
176 3 Amplitude Modulation
0 t
T
−A2
0 t1 t
T
−A2
according to:
Assume that T 1/ f c .
1 1
t1 A1 = (T − t1 )A2
2 2
A2 T
⇒ t1 = , (3.87)
A1 + A2
m(t) AC Am(t)
LPF
amplifier
c(t)
c(t)
1
t
0
−1
T
1
Pm = m 2 (t) dt, (3.90)
T t=0
where
A1 t/t1 for 0 < t < t1
m(t) = (3.91)
A2 (t − T )/(T − t1 ) for t1 < t < T .
16. (Haykin 1983) Figure 3.24 shows the block diagram of a chopper-stabilized dc
amplifier. It uses a multiplier and an ac amplifier, which shifts the spectrum of
the input signal from the vicinity of zero frequency to the vicinity of the carrier
frequency ( f c ). The signal at the output of the ac amplifier is then coherently
demodulated. The carrier c(t) is a square wave as indicated in the figure.
178 3 Amplitude Modulation
(a) Specify the frequency response of the ac amplifier so that there is no distor-
tion in the LPF output, assuming that m(t) is bandlimited to −W < f < W .
Specify also the relation between f c and W .
(b) Determine the overall gain of the system ( A), assuming that the ac amplifier
has a gain of K and the lowpass filter is ideal having unity gain.
• Solution: The square wave c(t) can be written as:
∞
4 (−1)n−1
c(t) = cos(2π f c (2n − 1)t). (3.93)
π n=1 2n − 1
The ac amplifier can have the frequency response of an ideal bandpass filter,
with a gain of K and bandwidth f c − W < | f | < f c + W . Observe that for
no aliasing of the spectrum centered at f c , we require:
fc + W < 3 fc − W
⇒ W < fc . (3.94)
4K
s(t) = m(t) cos(2π f c t). (3.95)
π
The output of the lowpass filter would be:
LPF 16K 8K
s(t)c(t) −→ m(t) = 2 m(t). (3.96)
2π 2 π
17. (Haykin 1983) Consider the phase discrimination method of generating an SSB
signal. Let the message be given by:
m(t) = a Am cos(2π f m t)
m̂(t) = b Am sin(2π f m t). (3.99)
(c) The carrier signal applied to the product modulators are not in phase quadra-
ture, that is:
c1 (t) = cos(2π f c t)
c2 (t) = sin(2π f c t + δ). (3.100)
Thus
Am /2
fc + fm
R1
Am /2
δ
Am /2
R2 fc − fm
δ
Am /2
δ
Am /2
Again from the phasor diagram in Fig. 3.26, we see that the ratio of the ampli-
tude of the undesired sideband to the desired sideband is R2 /R1 where R1 and
R2 are given by (3.103).
Am /2
fc + fm
R1
Am /2
δ
Am /2
fc − fm
δ Am /2
δ
R2
Am /2
Therefore
√
s(t) = Am 2 cos(2π f m t − π/4) cos(2π f c t)
√
+ Am 2 sin(2π f m t − π/4) [sin(2π f c t) + a sin(4π f c t)]
√
= Am 2 cos (2π ( f c − f m ) t − π/4)
a Am
+ √ cos (2π(2 f c − f m )t + π/4)
2
a Am
− √ cos (2π(2 f c + f m )t − π/4) . (3.110)
2
182 3 Amplitude Modulation
gp (t)
A
−T 0 T T0
H(f )
2
f
−4/T0 4/T0
a 2 A2m /2
R=
2 A2m /2
a2
= . (3.111)
2
19. Consider the system shown in Fig. 3.27. The input signal g p (t) is periodic with
a period of T0 . Assume that T /T0 = 0.5. The signal g p (t) is passed through an
ideal lowpass filter, that has a gain of two in the passband, as indicated in the
figure. The LPF output m 1 (t) is further passed through a dc blocking capacitor
to yield the message m(t). The capacitor acts as a short for the frequencies of
interest. This message is used to generate an AM signal given by
Clearly
d 2 g(t) A
2
= (δ(t + T ) − 2δ(t) + δ(t − T )) . (3.114)
dt T
3 Amplitude Modulation 183
Hence
A
G( f ) = − (exp(j 2π f T ) − 2 + exp(−j 2π f T ))
4π 2
f 2T
AT
= 2 2 2 sin2 (π f T )
π f T
= AT sinc2 ( f T ). (3.115)
where
1 ∞
cn = g(t) exp (−j 2πnt/T0 ) dt
T0 t=−∞
1
= G(n/T0 ). (3.117)
T0
where we have used the fact that T /T0 = 0.5. The output of the dc blocking
capacitor is
8A 8A
m(t) = cos(2πt/T0 ) + 2 cos(6πt/T0 ), . (3.119)
π 2 9π
where we have used the fact that the LPF has a gain of two in the frequency
range [−4/T0 , 4/T0 ]. The minimum value of m(t) is
80 A
m min (t) = − (3.120)
9π 2
and occurs at nT0 + T0 /2. To prevent overmodulation we must have:
1 + ka m min (t) ≥ 0
9π 2
⇒ ka ≤ . (3.121)
80 A
184 3 Amplitude Modulation
f (Hz) f (Hz)
m(t) y(t)
H(f )
2 cos(10000πt)
20. The system shown in Fig. 3.28 is used for scrambling audio signals. The output
y(t) is the scrambled version of the input m(t). The spectrum (Fourier transform)
of m(t) and the lowpass filter (H ( f )) are as shown in the figure.
(a) Draw the spectrum of y(t). Label all important points on the x- and y-axis.
(b) The output y(t) corresponds to a particular modulation scheme (i.e., AM,
FM, SSB with lower/upper sideband transmitted, VSB, DSB-SC). Which
modulation scheme is it?
(c) Write down the precise expression for y(t) in terms of m(t), corresponding
to the spectrum computed in part (a).
(d) Suggest a method for recovering m(t) (not km(t), where k is a constant)
from y(t).
Y1 ( f ) = M( f − f c ) + M( f + f c ) (3.123)
Fig. 3.29 a Y1 ( f ). b Y ( f ). Y1 (f )
c Procedure for recovering
m(t) A
2A
−fc 0 fc
y(t) m(t)
LPF
cos(2πfc t)
(c)
k
Y( f ) = [M( f − f c ) + M( f + f c )]
2
k
= − sgn( f − f c )M( f − f c ) − sgn( f + f c )M( f + f c )
2
k k
= M( f − f c )[1 − sgn( f − f c )] + M( f + f c )[1 + sgn( f + f c )].
2 2
(3.125)
Comparing the above equation with Fig. 3.29b, we conclude that k = 2. There-
fore
21. Figure 3.30 shows the block diagram of a DSB-SC modulator. Note that the
oscillator generates cos3 (2π f c t). The message spectrum is also shown.
(a) Draw and label the spectrum (Fourier transform) of y(t).
186 3 Amplitude Modulation
cos3 (2πfc t)
M (f )
−B 0 B
(b) Sketch the Fourier transform of a filter such that the output signal is
m(t) cos(2π f c t), where m(t) is the message. The filter should pass only
the required signal and reject all other frequencies.
(c) What is the minimum usable value of f c ?
3 1
y(t) = m(t) cos(2π f c t) + m(t) cos(6π f c t). (3.127)
4 4
Therefore
3 1
Y( f ) = [M( f − f c ) + M( f + f c )] + [M( f − 3 f c ) + M( f + 3 f c )] .
8 8
(3.128)
The spectrum of y(t) and the bandpass filter is shown in Fig. 3.31. We must
also have the following relationship:
3 fc − B ≥ fc + B
⇒ f c ≥ B. (3.129)
22. Consider the receiver of Fig. 3.32. Both bandpass filters are assumed to be ideal
with unity gain in the passband. The passband of BPF1 is in the range 500–
1500 kHz. The bandwidth of BPF2 is 10 kHz and it selects only the difference
frequency component. The input consists of a series of AM signals of bandwidth
10 kHz and spaced 10 kHz apart, occupying the band 500–1500 kHz as shown
in the figure.
When the LO frequency is 1505 kHz, the output signal is m 1 (t).
(a) Compute the centere frequency of BPF2.
3 Amplitude Modulation 187
M (f )
−B 0 B
Y (f )
4/3
3/8
1/8
f
LO
M1 (f ) M2 (f )
f kHz
(b) What should be the LO frequency so that the last message (occupying the
band 1490–1500 kHz) is selected.
(c) Can the LO frequency be equal to 1 MHz and the center frequency of BPF2
equal to 495 kHz? Justify your answer.
• Solution: When the LO frequency is 1505 kHz, the difference frequency for
m 1 (t) is 1505 − 505 = 1000 kHz. Therefore, the center frequency of BPF2
must be 1000 kHz.
When the LO frequency is 1495 + 1000 = 2495 kHz, the last message is
selected.
When the LO frequency is 1 MHz and the center frequency of BPF2 is
188 3 Amplitude Modulation
equal to 495 MHz, then the output signal would be the sum of the first mes-
sage (1000 − 505 = 495 kHz) and the last message (image at 1495 − 1000 =
495 kHz). Hence this combination should not be used.
23. (Haykin 1983) Consider a multiplex system in which four input signals m 0 (t),
m 1 (t), m 2 (t), and m 3 (t) are, respectively, multiplied by the carrier waves
3
s(t) = m i (t)ci (t). (3.131)
i=0
All the messages are bandlimited to [−W, W ]. At the receiver, the ith message
is recovered as follows:
LPF
s(t)ci (t) −→ m i (t) (3.132)
where the LPF is ideal with unity gain in the frequency band [−W, W ].
(a) Compute αi and βi , 0 ≤ i ≤ 3 for this to be feasible.
(b) Determine the minimum separation between f a and f b .
provided
| f b − f a | ≥ 2W
fa ≥ W
f b ≥ W. (3.134)
α1 = β1 = π/2. (3.135)
3 Amplitude Modulation 189
The two equations in (3.136) imply that if α2 is in the 1st quadrant then β2
must be in the 3rd quadrant. Otherwise, if α2 is in the 2nd quadrant, then β2
must be in the 4th quadrant. Let us take α2 = π/4. Then β2 = 5π/4.
Finally we consider c3 (t). We get the relations:
cos(α3 ) = − sin(α3 )
cos(β3 ) = − sin(β3 ). (3.138)
Thus we conclude that α3 and β3 must be an odd multiple of π/4 and must
lie in the 2nd/4th quadrant and 4th/2nd quadrant, respectively. If we take
α3 = 3π/4 then β3 = 7π/4. Note that there are infinite solutions to αi and βi
(1 ≤ i ≤ 3).
24. Consider the system shown in Fig. 3.33a. The signal s(t) is given by:
Complex
multiplication
s(t)
y(t)
Ideal |M (f )|
Hilbert ŝ(t) x(t)
transformer
Local
f
oscillator
fl
−W 0 W
(a) (b)
Both m 1 (t) and m 2 (t) are real-valued and bandlimited to [−W, W ]. The carrier
frequency f c W . The inputs to the complex multiplier are s(t) + j ŝ(t) and
x(t) (which could be real or complex-valued and depends on the local oscillator
frequency fl ). The multiplier output is the complex-valued signal y(t). Let
Therefore
If
then
The message m(t) is real-valued, does not have a dc component and its one-
sided bandwidth is W
f c . Explain how km(t), where k is a constant, can be
recovered from s(t) using a Hilbert transformer and other components.
• Solution: When s(t) is passed through a Hilbert transformer, its output is:
Now
The message m(t) can be recovered using the block diagram in Fig. 3.35. Note
that
26. Consider the Costas loop shown in Fig. 3.36. The input signal s(t) is
Complex
multiplication
s(t)
y(t) Ac ka m(t)
Ideal DC blocking
Hilbert ŝ(t) capacitor
transformer
e−j (2πfc t+θ−π/2)
xI (t)
LPF
2 cos(2πfc t + θ)
s(t) y
∞
VCO t=−∞
2 sin(2πfc t + θ)
LPF
xQ (t)
M (f )
2
f
−W 0 W
• Solution: Clearly
Hence
∞
y = 2 sin(2θ) m 2 (t) dt
t=−∞
∞
= 2 sin(2θ) |M( f )|2 d f
f =−∞
2
W
−2 f
= 4 sin(2θ) +2 df
f =0 W
16
= W sin(2θ) (3.153)
3
3 Amplitude Modulation 193
f (kHz)
where we have used the Rayleigh’s energy theorem. It is clear that y = 0 when
θ = kπ/2, where k is an integer.
27. It is desired to transmit 400 voice signals using frequency division multiplexing
(FDM). The voice signals are SSB modulated with lower sideband transmitted.
The spectrum of each of the voice signals is illustrated in Fig. 3.37.
(a) What is the minimum carrier spacing required to multiplex all the signals,
such that there is no overlap of spectra.
(b) Using a single- stage approach and assuming a carrier spacing of 7 kHz and
a minimum carrier frequency of 200 kHz, determine the lower and upper
frequency limits occupied by the FDM signal containing 400 voice signals.
(c) Determine the spectrum (lower and upper frequency limits) occupied by the
100th voice signal.
(d) In the two-stage approach, the first stage uses SSB with lower sideband
transmitted. In the first stage, the voice signals are grouped into L blocks,
with each block containing K voice signals. The carrier frequencies in each
block in the first stage is given by 7n kHz, for 1 ≤ n ≤ K . Note that the
FDM signal using the two-stage approach must be identical to the single-
stage approach in (b).
i. Sketch the spectrum of each block at the output of the first stage.
ii. Specify the modulation required for each of the blocks in the second
stage.
iii. How many carriers are required to generate the composite FDM signal?
Express your answer in terms of L and K .
iv. Determine L and K such that the number of carriers is minimized.
v. Give the expression for the carrier frequencies required to modulate
each block in the second stage.
The spectrum of the 400th voice signal ends at 2993 − 0.3 = 2992.7 kHz.
Therefore, the spectrum of the FDM signal extends from 196.7 to 2992.7 kHz,
that is, 196.7 ≤ | f | ≤ 2992.7 kHz.
The carrier for the 100th voice signal is at
Hence, the spectrum of the 100th voice signal extends from 893 − 3.3 =
889.7 kHz to 893 − 0.3 = 892.7 kHz, that is, 889.7 ≤ | f | ≤ 892.7 kHz.
The spectrum of each block at the output of the first stage is shown in Fig. 3.38b.
The modulation required for each of the blocks in the second stage is SSB
with upper sideband transmitted.
For the second approach, let us first evaluate the number of carriers required
to obtain each block in the first stage. Clearly, the number of carriers required
is K . In the next stage, L carriers are required to translate each block to the
appropriate frequency band. Thus, the total number of carriers required is
L + K.
Now, we need to minimize L + K subject to the constraint L K = 400. The
problem can be restated as
400
min + K. (3.156)
K K
Differentiating with respect to K and setting the result to zero, we get the
solution as K = 20, L = 20.
The carrier frequencies required in the second stage is given by 200 − 7 +
7K (l − 1) for 1 ≤ l ≤ L.
The block diagram of the two-stage approach is given in Fig. 3.38a.
28. It is desired to transmit 500 voice signals using frequency division multiplexing
(FDM). The voice signals are DSB-SC modulated. The spectrum of each of the
voice signals is illustrated in Fig. 3.39.
(a) How many carrier signals are required to multiplex all the 500 voice signals?
(b) What is the minimum carrier spacing required to multiplex all the signals,
such that there is no overlap of spectra.
(c) Using a single- stage approach and assuming a carrier spacing of 10 kHz and
a minimum carrier frequency of 300 kHz, determine the lower and upper
frequency limits occupied by the FDM signal containing 500 voice signals.
(d) Determine the spectrum (lower and upper frequency limits) occupied by the
150th voice signal.
(e) It is desired to reduce the number of carriers using a two-stage approach.
The first stage uses DSB-SC modulation. In the first stage, the voice signals
are grouped into L blocks, with each block containing K voice signals. The
carrier frequencies in each block in the first stage is given by 10n kHz, for
3 Amplitude Modulation 195
ml, 1 (t)
(a) sl (t) Sl (f )
SSB
for 1 ≤ l ≤ L
fc = 7 kHz
ml, K (t)
SSB
fc = 7K kHz
1st stage
s1 (t)
SSB
fc = 200 − 7 kHz
Final FDM signal
sL (t)
SSB
fc = 200 − 7 + 7K(L − 1) kHz
2nd stage
(b) Sl (f )
(kHz)
f
0
Fig. 3.38 a Two-stage approach for obtaining the final FDM signal. b Spectrum of each block at
the output of the first stage
f (kHz)
−4 0 4
196 3 Amplitude Modulation
1 ≤ n ≤ K . Note that the FDM signal using the two-stage approach must
be identical to the single-stage approach in (c).
i. Sketch the spectrum of each block at the output of the first stage.
ii. Specify the modulation required for each of the blocks in the second
stage, such that, the minimum carrier frequency used in the second
stage, is closest to 300 kHz.
iii. How many carriers are required to generate the composite FDM signal?
Express your answer in terms of L and K .
iv. Determine L and K such that the number of carriers is minimized.
v. Give the expression for the carrier frequencies required to modulate
each block in the second stage.
• Solution: The number of carriers required to multiplex all the voice signals is
500.
The minimum carrier spacing required is 8 kHz.
If the minimum carrier frequency is 300 kHz, then the spectrum of the first
voice signal starts at 300 − 4 = 296 kHz. The carrier for the 500th voice signal
is at
The spectrum of the 500th voice signal ends at 5290 + 4 = 5294 kHz. There-
fore, the spectrum of the FDM signal extends from 296 to 5294 kHz, that is
296 ≤ | f | ≤ 5294 kHz.
The carrier for the 150th voice signal is at
Hence the spectrum of the 150th voice signal extends from 1790 − 4 =
1786 kHz to 1790 + 4 = 1794 kHz, that is 1786 ≤ | f | ≤ 1794 kHz.
The spectrum of each block at the output of the first stage is shown in Fig. 3.40b.
The modulation required for each of the blocks in the second stage is SSB.
In order to ensure that the minimum carrier frequency in the second stage is
closest to 300 kHz, the upper sideband must be transmitted. Thus, the mini-
mum carrier frequency in the second stage is 300 − 10 = 290 kHz.
For the second approach, let us first evaluate the number of carriers required
to obtain each block in the first stage. Clearly, the number of carriers required
is K . In the next stage, L carriers are required to translate each block to the
appropriate frequency band. Thus, the total number of carriers required is
L + K.
Now, we need to minimize L + K subject to the constraint L K = 500. The
problem can be restated as
500
min + K. (3.159)
K K
3 Amplitude Modulation 197
ml, 1 (t)
(a) sl (t) Sl (f )
DSBSC
for 1 ≤ l ≤ L
fc = 10 kHz
ml, K (t)
(c) L K L+K
DSBSC
fc = 10K kHz
2 250 252
1st stage
4 125 129
s1 (t) 10 50 60
SSB 20 25 45
fc = 300 − 10 kHz Final
25 20 45
FDM
50 10 60
signal
sL (t) 125 4 129
SSB 250 2 252
fc = 300 − 10 + 10K(L − 1) kHz
2nd stage
(b)
Sl (f )
(kHz)
f
0
Fig. 3.40 a Two-stage approach for obtaining the final FDM signal. b Spectrum of each block at
the output of the first stage. c Various possible values of L and K
Differentiating with respect to K and setting the result to zero, we get the
solution as K = 22.36, which is not an integer. Hence, we need to obtain the
solution manually, as given in Fig. 3.40c. We find that, there are two sets of
solutions, L = 20, K = 25, and L = 25, K = 20.
The carrier frequencies required in the second stage is given by 300 − 10 +
10K (l − 1) for 1 ≤ l ≤ L.
The block diagram of the two-stage approach is given in Fig. 3.40a.
29. Let m(t) be a signal having Fourier transform M( f ), with M(0) = 0. When
m(t) is passed through a filter with impulse response h(t), the Fourier transform
of the output can be written as
198 3 Amplitude Modulation
⎧
⎨ M( f ) e−j θ for f > 0
Y( f ) = 0 for f = 0 (3.160)
⎩
M( f ) e j θ for f < 0
Determine h(t).
• Solution: We know that if (this was done in class)
sin(θ)
h(t) = cos(θ)δ(t) + . (3.163)
πt
30. Consider the DSB-SC signal
Assume that the energy signal m(t) occupies the frequency band [−W, W ].
Now, s(t) is applied to a square law device given by
The output y(t) is applied to an ideal bandpass filter with a passband transfer
function equal to 1/( f ), midband frequency of ±2 f c and bandwidth f .
Assume that f → 0. All signals are real-valued.
(a) Determine the spectrum of y(t).
(b) Find the relation between f c and W for no aliasing in the spectrum of y(t).
(c) Find the expression for the signal v(t) at the BPF output.
• Solution: We have
A2c
y(t) = [1 + cos(4π f c t)] m 2 (t)
2
A2 A2
⇒ Y ( f ) = c G( f ) + c [G( f − 2 f c ) + G( f + 2 f c )] , (3.166)
2 4
where
3 Amplitude Modulation 199
G( f ) = M( f ) M( f )
M( f ) m(t). (3.167)
2 f c − 2W > 2W
⇒ f c > 2W. (3.168)
A2c
V( f ) = G(0) [δ( f − 2 f c ) + δ( f + 2 f c )] . (3.169)
4
Therefore
A2c
v(t) = G(0) cos(4π f c t). (3.170)
2
Now
∞
G( f ) = M(x)M( f − x) d x
x=−∞
∞
⇒ G(0) = M(x)M(−x) d x. (3.171)
x=−∞
M(−x) = M ∗ (x)
∞
⇒ G(0) = |M(x)|2 d x
x=−∞
= E, (3.172)
where we have used the Rayleigh’s energy theorem. Thus (3.170) reduces to:
A2c
v(t) = E cos(4π f c t). (3.173)
2
31. (Haykin 1983) Consider the quadrature -carrier multiplex system shown in
Fig. 3.41. The multiplexed signal s(t) is applied to a communication channel
of frequency response H ( f ). The channel output is then applied to the receiver
input. Here f c denotes the carrier frequency and the message spectra extends
over [−W, W ]. Find
200 3 Amplitude Modulation
Ac cos(2πfc t)
s(t) y(t)
H(f )
Ac sin(2πfc t)
m2 (t)
Ac m1 (t)
LPF
y(t) 2 cos(2πfc t)
2 sin(2πfc t)
Ac m2 (t)
LPF
Let
Y ( f ) = S( f )H ( f ) y(t). (3.175)
Now
3 Amplitude Modulation 201
Y ( f − fc ) = H ( f − fc )
M1 ( f − 2 f c ) + M1 ( f ) M2 ( f − 2 f c ) − M2 ( f )
+
2 2j
Y ( f + fc ) = H ( f + fc )
M1 ( f ) + M1 ( f + 2 f c ) M2 ( f ) − M2 ( f + 2 f c )
+
2 2j
(3.177)
From the above equation, it is clear that to recover M1 ( f ) from the upper LPF
(G 1 ( f ) = M1 ( f )) we require
H ( f − fc ) = H ( f + fc ) for −W ≤ f ≤ W
∗
⇒ H ( fc − f ) = H ( f + fc ) for −W ≤ f ≤ W (3.179)
(a) Show that an envelope detector may be used to recover the sum signal
m r (t) + m l (t) from the quadrature multiplexed signal. How would you min-
imize the signal distortion produced by the envelope detector.
(b) Show that a coherent detector can recover the difference m l (t) − m r (t).
z 1 (t) = Ac m 1 (t)
= Ac (V0 + m l (t) + m r (t)). (3.186)
z 2 (t) = Ac m 2 (t)
= Ac (m l (t) − m r (t)). (3.187)
The message signals m l (t) and m r (t) typically have zero dc, hence Ac V0 can
be removed by a dc blocking capacitor.
33. (Haykin 1983) Using the message signal
1
m(t) = , (3.190)
1 + t2
m(0) = 1
∞
= , M( f ) d f (3.191)
f =−∞
where M( f ) is the Fourier transform of m(t). Hence the AM signal with 50%
modulation is given by:
0.5
s(t) = Ac1 1 + cos(2π f c t). (3.192)
1 + t2
Ac1 Ac1
S( f ) = [δ( f − f c ) + δ( f + f c )] + [M( f − f c ) + M( f + f c )] .
2 4
(3.193)
Since
∞
S( f ) d f = 1, (3.194)
f =−∞
we require
Ac1 Ac1
[1 + 1] + [1 + 1] = 1
2 4
⇒ Ac1 = 2/3, (3.195)
204 3 Amplitude Modulation
where we have used (3.191). Note that shifting the spectrum of M( f ) does
not change the area. The DSB-SC modulated signal is given by:
Ac2
s(t) = cos(2π f c t). (3.196)
1 + t2
Ac2
S( f ) = [M( f − f c ) + M( f + f c )] . (3.197)
2
Again, due to (3.191) and (3.194) we have
Ac2
[1 + 1] = 1
2
⇒ Ac2 = 1. (3.198)
1 HT t
. (3.199)
1+t 2 1 + t2
Therefore, the SSB modulated signal with upper sideband transmitted is given
by:
1 t
s(t) = Ac3 cos(2π f c t) − sin(2π f c t) . (3.200)
1 + t2 1 + t2
Ac3
S( f ) = M( f − f c ) 1 + sgn( f − f c )
2
Ac3
+ M( f + f c ) 1 − sgn( f + f c ) . (3.201)
2
Note that m(t) is real-valued and an even function of time. Hence, M( f ) is
also real-valued and an even function of frequency. Therefore from (3.191)
∞
M( f ) d f = 1/2. (3.202)
f =0
2 × 0.5Ac3 2 × 0.5Ac3
+ =1
2 2
⇒ Ac3 = 1. (3.203)
3 Amplitude Modulation 205
M (f )
2
f (kHz)
−3 −0.2 0 0.2 3
M1 (f )
(a)
Ac
f (kHz)
M2 (f )
(b)
Ac
f (kHz)
Ac
s2 (t) = m(t) cos(2π f t) − m̂(t) sin(2π f t) . (3.207)
2
When f is positive, s2 (t) is an SSB signal with carrier frequency f and
upper sideband transmitted. This is illustrated in Fig. 3.43a, for f = 20 Hz.
When f is negative, s2 (t) is an SSB signal with carrier frequency f and
lower sideband transmitted. This is illustrated in Fig. 3.43b with f = −10
Hz.
35. Consider the series RLC circuit in Fig. 3.44. The resonant frequency of the circuit
is 1 MHz and the Q-factor is 100. The input signal vi (t) is given by:
Q = 2π f c L/R
fc
= (3.209)
−3 dB bandwidth
μAc
vi (t) = Ac cos(2π f c t) + [cos(2π( f c − f m )t) + cos(2π( f c + f m )t)]
2
(3.210)
Vo (ω) R
= H (ω) = , (3.211)
Vi (ω) R + j ωL − j/(ωC)
ω = ωc + ω (3.212)
R
H (ω) =
R + j (ωc + ω)L − j/((ωc + ω)C)
R
≈
R + j ωL + jω/(ωc2 C)
1
=
1 + j ωL/R + jω/(ωc2 RC)
1
= , (3.213)
1 + j (2ω)L/R
ωc L = 1/(ωc C)
1
≈ 1 − ω/ωc when ω
ωc . (3.214)
1 + ω/ωc
Since the Q-factor of the filter is high, we can assume that 2ω is the 3-dB
bandwidth of the filter. Hence
ωc
ω = = 2π × 5000 rad/s, (3.215)
2Q
1
H (ωc + ω) = √ e−j π/4
2
1 j π/4
H (ωc − ω) = √ e
2
H (ωc ) = 1. (3.216)
μAc
vo (t) = Ac cos(2π f c t) + √ {cos [2π( f c − f m )t + π/4]
2 2
+ cos [2π( f c + f m )t − π/4]}
μ
= Ac 1 + √ cos (2π f m t − π/4) cos (2π f c t) . (3.217)
2
36. Let su (t) denote the SSB wave obtained by transmitting only the upper sideband,
that is
su (t) = Ac m(t) cos(2π f c t) − m̂(t) sin(2π f c t) , (3.218)
1
m(t) = su (t) cos(2π f c t) + ŝu (t) sin(2π f c t) . (3.220)
Ac
37. (Haykin 1983) A method that is used for carrier recovery in SSB modulation
systems involves transmitting two pilot frequencies that are appropriately posi-
tioned with respect to the transmitted sideband. This is shown in Fig. 3.45a for
the case where the lower sideband is transmitted. Here, the two pilot frequencies
are defined by:
f1 = fc − W − f
f 2 = f c + f, (3.221)
S(f )
(a)
−fc + W fc − W
(b)
Frequency
Narrowband
s(t) v1 (t) v3 (t) v4 (t)
filter Lowpass divide
centered filter
by n + 2
at f1
Narrowband
v2 (t)
filter
centered
at f2
Narrowband
Output v5 (t)
filter
centered
at fc
Fig. 3.45 Carrier recovery for SSB signals with lower sideband transmitted
W
n= (3.222)
f
v1 (t) = A1 cos(2π f 1 t + φ1 )
v2 (t) = A2 cos(2π f 2 t + φ2 ). (3.223)
−φ1
φ2 = . (3.224)
1+n
(b) For the case when only the upper sideband is transmitted, the two pilot
frequencies are
210 3 Amplitude Modulation
f1 = fc − f
f 2 = f c + W + f. (3.225)
How would you modify the carrier recovery scheme in order to deal with this
case. What is the corresponding relation between φ1 and φ2 for the output
to be proportional to the carrier signal?
A1 A2
v3 (t) = cos(2π( f 2 − f 1 )t + φ2 − φ1 ). (3.226)
2
Substituting for f 1 and f 2 we get:
A1 A2
v3 (t) = cos(2π(n + 2)W t/n + φ2 − φ1 )
2
A1 A2
= cos(2π(n + 2)W t/n + φ2 − φ1 + 2πk)
2
, (3.227)
A1 A2
v4 (t) = cos(2πW (t − t0 )/n)
2
A1 A2
= cos(2πW t/n + (φ2 − φ1 )/(n + 2) + 2πk/(n + 2))
2
A1 A2
= cos(2π f t + (φ2 − φ1 )/(n + 2) + 2πk/(n + 2)).
2
(3.230)
A1 A22
v4 (t)v2 (t) = cos(2π f t + (φ2 − φ1 )/(n + 2))
2
× cos(2π( f c + f )t + φ2 ). (3.231)
A1 A22
v5 (t) = cos(2π f c t + φ2 − (φ2 − φ1 )/(n + 2)), (3.232)
4
which is proportional to the carrier frequency when
φ2 − (φ2 − φ1 )/(n + 2) = 0
−φ1
⇒ φ2 = . (3.233)
n+1
A1 A2
v3 (t) = cos(2π( f 2 − f 1 )t + φ2 − φ1 )
2
A1 A2
= cos(2π(n + 2)W t/n + φ2 − φ1 + 2πk), (3.234)
2
for 0 ≤ k < n + 2. Using similar arguments, the output of the frequency
divider is (for k = 0):
A1 A2
v4 (t) = cos(2πW t/n + (φ2 − φ1 )/(n + 2))
2
A1 A2
= cos(2π f t + (φ2 − φ1 )/(n + 2)). (3.235)
2
The output of the second multiplier is given by:
A21 A2
v4 (t)v1 (t) = cos(2π f t + (φ2 − φ1 )/(n + 2))
2
× cos(2π( f c − f )t + φ1 ). (3.236)
212 3 Amplitude Modulation
S(f )
(a)
−fc fc
(b)
Frequency
Narrowband
s(t) v2 (t) v3 (t) v4 (t)
filter Lowpass divide
centered filter
by n + 2
at f2
Narrowband
v1 (t)
filter
centered
at f1
Narrowband
Output v5 (t)
filter
centered
at fc
Fig. 3.46 Carrier recovery for SSB signals with upper sideband transmitted
A21 A2
v5 (t) = cos(2π f c t), (3.237)
4
provided
φ2 − φ1
+ φ1 = 0
n+2
⇒ φ2 = −(n + 1)φ1 . (3.238)
M (f )
(a)
2
−fb −fa 0 fa fb
(b)
s(t) m(t)
Filter1 Filter2
2 cos(2πfc t) x
Fig. 3.47 a Spectrum of the message signal. b Scheme to recover the message
Then
−j
−j
m(t) sin(2π f c t) [M( f − f c ) − M( f + f c )]
2
= S2 ( f ) (3.242)
S( f ) = S1 ( f ) − S2 ( f ), (3.243)
39. It is desired to transmit 100 voice signals using frequency division multiplexing
(FDM). The voice signals are SSB modulated, with the upper sideband trans-
mitted. The spectrum of each of the voice signals is illustrated in Fig. 3.49.
214 3 Amplitude Modulation
M (f )
2
(a)
−fb −fa 0 fa fb
S1 (f )
j
(b)
−fc + fa fc + fa
f
0
−fc fc
−fc − fa fc − fa
−j
S2 (f )
(c) j
fc − fa fc + fa
f
0
−fc fc
−fc − fa −fc + fa
−j
S(f )
2j
(d) −fc + fb
−fc + fa
f
0
−fc fc
fc − fa
fc − fb
−2 j
(e)
1
−fb 0 fb
s(t) m̂(t) −m(t) m(t)
LPF HT
2 cos(2πfc t) −1
Fig. 3.48 a M( f ). b S1 ( f ). c S2 ( f ). d S( f )
3 Amplitude Modulation 215
f (kHz)
(a) What is the minimum carrier spacing required to multiplex all the signals,
such that there is no overlap of spectra.
(b) Using a single- stage approach and assuming a carrier spacing of 4 kHz and
a minimum carrier frequency of 100 kHz, determine the lower and upper
frequency limits occupied by the FDM signal containing 100 voice signals.
(c) In the two-stage approach, both stages use SSB with upper sideband trans-
mitted. In the first stage, the voice signals are grouped into L blocks, each
containing K voice signals. The carrier frequencies in each block in the first
stage is given by 4n kHz, for 0 ≤ n ≤ K − 1. Note that the FDM signal
using the two-stage approach must be identical to the single-stage approach
in (b).
i. Sketch the spectrum of each block in the first stage.
ii. How many carriers are required to generate the FDM signal? Express
your answer in terms of L and K .
iii. Determine L and K such that the number of carriers is minimized.
iv. Give the expression for the carrier frequencies required to modulate
each block in the 2nd stage.
The spectrum of the 100th voice signal starts at 496.3 kHz and ends at
499.3 kHz. Therefore, the spectrum of the composite FDM signal extends
from 100.3 to 499.3 kHz, that is 100.3 ≤ | f | ≤ 499.3 kHz.
The spectrum of each block in the first stage is shown in Fig. 3.50b.
For the second approach, let us first evaluate the number of carriers required
to obtain each block in the first stage. Clearly, the number of carriers required
is K − 1, since the first message in each block is not modulated at all. In the
next stage, L carriers are required to translate each block to the appropriate
frequency band. Thus, the total number of carriers required is L + K − 1.
Now, we need to minimize L + K − 1 subject to the constraint L K = 100.
The problem can be restated as
216 3 Amplitude Modulation
ml, 1 (t)
(a)
ml, 2 (t)
sl (t) Sl (f )
SSB
for 1 ≤ l ≤ L
fc = 4 kHz
ml, K (t)
SSB
fc = 4(K − 1) kHz
1st stage
s1 (t)
SSB
fc = 100 kHz
Final FDM signal
sL (t)
SSB
fc = 100 + 4K(L − 1) kHz
2nd stage
(b) Sl (f )
(kHz)
f
−7.3 −4.3 −3.3 0 3.3 4.3 7.3
Fig. 3.50 a Two-stage approach for obtaining the final FDM signal. b Spectrum of each block at
the output of the first stage
100
min + K − 1. (3.245)
K K
Differentiating with respect to K and setting the result to zero, we get the
solution as K = 10, L = 10.
The carrier frequencies required to modulate each block in the second stage
is given by 100 + 4K l for 0 ≤ l ≤ L − 1.
The block diagram of the two-stage approach is given in Fig. 3.50a.
40. Consider the modified switching modulator shown in Fig. 3.51. Note that
A1 cos3 (2πfc t)
where
∞
1 2 (−1)n−1
g p (t) = + cos(2π f c (2n − 1)t). (3.247)
2 π n=1 2n − 1
• Solution: We have
Ac
v2 (t) = cos(2π f c t)
2
∞
Ac (−1)n−1
+ (cos(4π(n − 1) f c t) + cos(4πn f c t))
π n=1 2n − 1
∞
m(t) 2 (−1)n−1
+ + cos(2π(2n − 1) f c t)m(t). (3.248)
2 π n=1 2n − 1
2
− cos(6π f c t)m(t), (3.249)
3π
which occurs at n = 2. Similarly
A1
A1 cos3 (2π f c t) = [3 cos(2π f c t) + cos(6π f c t)] , (3.250)
4
which has a component at f c and 3 f c . Thus it is clear that in order to extract the
components at 3 f c , the spectrum of the BPF must be as indicated in Fig. 3.52.
The BPF output is given by:
218 3 Amplitude Modulation
H(f )
k
A1 k 2k
s(t) = cos(6π f c t) − m(t) cos(6π f c t)
4 3π
k A1 8
= 1− m(t) cos(6π f c t). (3.251)
4 3π A1
8
max |m(t)| = 1
3π A1
16
⇒ =1
3π A1
⇒ A1 = 1.7. (3.252)
k 2 A21
= 10
32
⇒ k = 10.5. (3.253)
Assume that the envelope detector is implemented using a diode and an RC filter,
as shown in Fig. 3.53. Determine the upper limit on RC such that the capacitor
voltage follows the envelope. Assume that RC 1/ f c , e−x ≈ 1 − x for small
values of x and f c to be very large.
s(t) Rs
v(t)
C R
2
a(t0 )
1.5
a(t)
v(t)
1
0.5
-0.5
s(t)
-1
-1.5
-2
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
t
Fig. 3.54 Plots of s(t), the envelope a(t) and the capacitor voltage v(t)
since |μ| < 1. Consider any point a(t0 ) on the envelope at time t0 , which
coincides with the carrier peak. The capacitor starts discharging from a(t0 )
upto approximately the next carrier peak. The capacitor voltage v(τ ) can be
written as:
We require that the absolute value of the slope of the capacitor voltage must
exceed that of the envelope at time t0 . The absolute value of the slope of the
envelope at t0 is
da(t)
= μAc (2π f m ) |sin(2π f m t0 )| . (3.258)
dt
t=t0
Note that since t0 must coincide with the carrier peak, we must have
k
t0 = (3.259)
fc
for integer values of k. From (3.257) and (3.258) we get the required condition
as
a(t0 )
≥ μAc (2π f m ) |sin(2π f m t0 )|
RC
1 + μ cos(2π f m t0 )
⇒ RC ≤ (3.260)
2πμ f m | sin(2π f m t0 )|
1
RC
, (3.263)
W
where W is the one-sided bandwidth of the message. We find that (3.263) is
valid when μ is close to unity and f m is replaced by W in (3.262).
H(f )
Similarly
where
222 3 Amplitude Modulation
j [H ( f − f c ) − H ( f + f c )] for −W ≤ f ≤ W
Hs ( f ) = (3.269)
0 otherwise.
Ac dm(t)
ss (t) = − . (3.272)
πW dt
1
H( f ) = , (3.274)
1 + j f / f0
1 fm
f0 = = . (3.275)
2π RC 2
Clearly
H(f )
(a)
4
H(f − fc )
(b)
4
H(f + fc )
(c)
4
Hs (f )/j
(d)
4
W f
−W
−4
Hs (f )/j
(e)
4
W f
−W
−4
m(t)
2Ac cos(2πfc t) +
s(t)
Hs (f )
−(Ac /2) sin(2πfc t) +
−
z(t)
where
θ = − tan−1 (2)
1
|H ( f m )| = √ . (3.277)
5
|ka |B|H ( f m )| ≤ 1
√
5
⇒ |ka | ≤ (3.278)
B
with the constraint that ka = 0. In order to compute the power of s(t) we note
that
s(t) = Ac cos(2π f c t)
B
+ Ac ka √ cos(2π( f c − f m )t − θ)
2 5
B
+ Ac ka √ cos(2π( f c + f m )t + θ). (3.279)
2 5
Thus the power of s(t) is just the sum of the power of the individual sinusoids
and is equal to:
3 Amplitude Modulation 225
−fb −fa 0 fa fb
(b)
s(t) m(t)
H(f )
2 sin(2πfc t)
A2c A2 k 2 B 2
P= + c a . (3.280)
2 20
44. An AM signal
• Solution: Clearly, the signal at the BPF output is SSB modulated with lower
sideband transmitted. Therefore, the expression for s(t) is
k
S( f ) = M( f − f c ) 1 − sgn ( f − f c )
2
k
+ M( f + f c ) 1 + sgn ( f + f c ) . (3.284)
2
226 3 Amplitude Modulation
S(f )
3Ac ka /2
Inphase channel
filter
cos(2πf0 t) cos(2πfc t)
m(t) SSB signal
sin(2πf0 t) sin(2πfc t)
e(t)
filter
Quadrature channel
M (f )
1
−fb −fa fa fb
Fig. 3.61 Weaver’s method of generating SSB signals transmitting the upper sideband
3Ac ka
k= . (3.285)
2
In order to recover the message, we need a lowpass filter with unity gain and
passband [− f b , f b ] in cascade with an ideal Hilbert transformer followed by
a gain of −1/k. These three elements can be combined into a single filter
whose frequency response is given by
j (1/k)sgn( f ) for | f | < f b
H( f ) = (3.286)
0 otherwise.
45. (Haykin 1983) Consider the block diagram in Fig. 3.61. The message m(t) is
bandlimited to f a ≤ | f | ≤ f b . The auxiliary carrier applied to the first pair of
3 Amplitude Modulation 227
fa + fb
f0 = . (3.287)
2
The lowpass filters are identical and can be assumed to be ideal with unity
gain and cutoff frequency equal to ( f b − f a )/2. The carrier frequency f c >
( f b − f a )/2.
(a) Plot the spectra of the complex signals a(t) + j b(t), c(t) + j d(t) and the
real-valued signal e(t).
(b) Write down the expression for e(t) in terms of m(t).
(c) How would you modify Fig. 3.61 so that only the lower sideband is trans-
mitted.
• Solution: We have
Thus
m 1 (t) M1 ( f ) = M( f − f 0 ). (3.289)
The plot of M1 ( f ) is given in Fig. 3.62b. Note that m 1 (t) cannot be regarded as
a pre-envelope, since it has non-zero negative frequencies. The various edge
frequencies in M1 ( f ) are given by:
f 1 = ( f b − f a )/2
f 2 = ( f b + 3 f a )/2
f 3 = ( f a + 3 f b )/2. (3.290)
Now the complex-valued signal m 1 (t) gets convolved with a real-valued low-
pass filter (the lowpass filters have been stated to be identical, hence they
can be considered to be a single real-valued lowpass filter), resulting in a
complex-valued output m 2 (t). The plot of M2 ( f ) is shown in Fig. 3.62c. Let
−fb −fa fa fb
(b) M1 (f )
1
−f1 f1 f2 f3
(c) M2 (f )
1
f
−f1 f1
(d) E(f )
0.5
f
−fc fc
fc + f1
fc − f1
1
e(t) [M2 ( f + f c ) + M2 (− f + f c )] = E( f ) (say). (3.293)
2
E( f ) is plotted in Fig. 3.62d.
Let
f c1 = f c − f 1 − f a . (3.294)
Then
In order to transmit the lower sideband, the first product modulator in the
quadrature arm must be fed with − sin(2π f 0 t).
46. (Haykin 2001) The spectrum of a voice signal m(t) is zero outside the inter-
val f a < | f | < f b . To ensure communication privacy, this signal is applied to a
3 Amplitude Modulation 229
filter filter
• Solution: The block diagram of the system is shown in Fig. 3.63. The output
of the first product modulator is given by:
1
V1 ( f ) = [M( f − f c ) + M( f + f c )] . (3.296)
2
The output of the highpass filter is an SSB signal with the upper sideband
transmitted. This is illustrated in Fig. 3.64. Hence:
1
v2 (t) = m(t) cos(2π f c t) − m̂(t) sin(2π f c t) . (3.297)
2
The output of the second product modulator is given by:
1
s(t) = m(t) cos(2π f b t) + m̂(t) sin(2π f b t) . (3.299)
4
230 3 Amplitude Modulation
M (f )
−fb −fa 0 fa fb
V1 (f )
1 Highpass filter
0.5
−fc − fb −fc 0 fc fc + fb
V2 (f )
0.5
−fc − fb −fc 0 fc fc + fb
S(f )
0.25
−fb 0 fb
fa − fb fb − fa
S1 (f ) = M (f )
1/16
−fb −fa 0 fa fb
It is clear that s(t) is an SSB signal with lower sideband transmitted and carrier
frequency f b . This is illustrated in Fig. 3.64.
If s(t) is fed to the scrambler, the output would be similar to (3.299), that is:
1
s1 (t) = s(t) cos(2π f b t) + ŝ(t) sin(2π f b t) . (3.300)
4
In other words, we transmit the lower sideband of s(t) with carrier frequency
f b . Thus
1
s1 (t) = m(t) (3.301)
16
which is again shown in Fig. 3.64.
Am Ac
s(t) = [a cos(2π( f c + f m )t) + (1 − a) cos(2π( f c − f m )t)] (,3.303)
2
where 0 ≤ a ≤ 1 is a constant, representing the attenuation of the upper side
frequency.
(a) From the canonical representation of a bandpass signal, find the quadrature
component of s(t).
(b) The VSB signal plus a carrier Ac cos(2π f c t), is passed through an envelope
detector. Determine the distortion produced by the quadrature component.
(c) What is the value of a for which the distortion is maximum.
(d) What is the value of a for which the distortion is minimum.
Am Ac
s(t) = [cos(2π f c t) cos(2π f m t) + (1 − 2a) sin(2π f c t) sin(2π f m t)] .
2
(3.304)
By comparing with the canonical representation of a bandpass signal, we see
that the quadrature component is:
Am Ac
− (1 − 2a) sin(2π f m t). (3.305)
2
After the addition of the carrier, the modified signal is:
232 3 Amplitude Modulation
filter meter
A cos(2πfc t)
Variable
frequency
oscillator
Am
s(t) = Ac cos(2π f c t) 1 + cos(2π f m t)
2
Am Ac
+ (1 − 2a) sin(2π f c t) sin(2π f m t). (3.306)
2
The envelope is given by:
Am
E(t) = Ac 1 + cos(2π f m t) 1 + D(t), (3.307)
2
48. (Haykin 1983) Figure 3.65 shows the block diagram of a heterodyne spectrum
analyzer. The oscillator has an amplitude A and operates over the range f 0 to
f 0 + W where ± f 0 is the midband frequency of the BPF and g(t) extends over
the frequency band [−W, W ]. Assume that the BPF bandwidth f
W and
f 0 W and that the passband response of the BPF is unity. Determine the value
of the energy meter output for an input signal g(t), for a particular value of the
oscillator frequency, say f c . Assume that g(t) is a real-valued energy signal.
• Solution: Let us denote the output of the product modulator by x(t). Hence
A2
E= |G( f 0 − f c )|2 f (3.312)
2
where we have used the fact that g(t) is real-valued, hence
vD
+ −
D
v1 (t) v2 (t)
iD
iD
R
vD
0 Vγ
i − v characteristics of D
• Solution: We have
v1 (t) − Vγ when D is ON
v2 (t) =
0 when D is OFF
v1 (t) − Vγ when v1 (t) ≥ Vγ
⇒ v2 (t) =
0 when v1 (t) < Vγ
v1 (t) − Vγ when Ac cos(2π f c t) ≥ Vγ
⇒ v2 (t) = (3.314)
0 when Ac cos(2π f c t) < Vγ
where we have assumed that v1 (t) ≈ Ac cos(2π f c t). The output voltage v2 (t)
can also be written as:
where
1 when Ac cos(2π f c t) ≥ Vγ
g p (t) = (3.316)
0 when Ac cos(2π f c t) < Vγ
1.5
1 gp (t)
Amplitude
0.5
0
cos(2πfc t)
−T0 T0
-0.5
T
-1
-1.5
-2
-1.5 -1 -0.5 0 0.5 1 1.5
fc t
√
Fig. 3.67 Plot of g p (t) and cos(2π f c t) for Vγ /Ac = 1/ 2
3 Amplitude Modulation 235
Ac cos(2π f c T0 ) = Vγ
√
⇒ cos(2π f c T0 ) = 1/ 2
⇒ 2π f c T0 = π/4
⇒ T0 /T = 1/8, (3.317)
∞
2πnt 2πnt
g p (t) = a0 + 2 an cos + bn sin , (3.318)
n=1
T T
where
T /2
1
a0 = g p (t) dt
T −T /2
T0
1
= dt
T −T0
= 1/4. (3.319)
∞
1 1
g p (t) = +2 sin(nπ/4) cos(2πnt/T )
4 n=1
nπ
√
1 2 1
= + cos(2π f c t) + cos(4π f c t) + · · · (3.321)
4 π π
Substituting (3.321) in (3.315) and collecting terms corresponding to the AM
signal centered at f c , we obtain:
236 3 Amplitude Modulation
√
Ac 2
v2 (t) = cos(2π f c t) + m(t) cos(2π f c t)
4 √ π
2 Ac
− Vγ cos(2π f c t) + cos(2π f c t) + · · · (3.322)
π 2π
Using a bandpass signal centered at f c and two-sided bandwidth equal to W
we obtain the desired AM signal as:
√ √
Ac Vγ 2 Ac 2
s(t) = − + + m(t) cos(2π f c t)
4 π 2π π
√
1 1 1 2
= Ac − + + m(t) cos(2π f c t)
4 π 2π π Ac
√
1 1 2
= Ac − + m(t) cos(2π f c t)
4 2π π Ac
0.4501582
= Ac 0.0908451 + m(t) cos(2π f c t)
Ac
4.9552301
= 0.0908451Ac 1 + m(t) cos(2π f c t). (3.323)
Ac
50. A signal x(t) = A sin(2000πt) is a full wave rectified and passed through an
ideal lowpass filter (LPF) having a bandwidth [−4.5, 4.5] kHz and a gain of 2 in
the passband. Let the LPF output be denoted by m(t). Find y(t) = m(t) + j m̂(t)
and sketch its Fourier transform. Compute the power of y(t).
• Solution: Note that the period of x(t) is 1 ms. The full wave rectified sine wave
can be represented by the complex Fourier series coefficients as follows:
∞
|x(t)| = cn e j 2πnt/T1 , (3.324)
n=−∞
where ω0 = 2π/T0 , with T0 = 2T1 = 1 ms. The above equation can be rewrit-
ten as:
T1
A j ω0 t
cn = e − e−j ω0 t e−j 2πnt/T1 dt
2 j T1 t=0
3 Amplitude Modulation 237
0 1/T1 2/T1
2
m(t) = 2 cn e j 2πnt/T1
n=−2
= 2c0 + 4c1 cos(ω1 t) + 4c2 cos(2ω1 t), (3.327)
Note the absence of c0 in (3.328), since the Hilbert transformer blocks the dc
component. Therefore
51. Explain the principle of operation of the Costas loop for DSB-SC signals given by
s(t) = Ac m(t) cos(2π f c t). Draw the block diagram. Clearly identify the phase
discriminator. Explain phase ambiguity.
• Solution: See Fig. 3.69. We assume that m(t) is real-valued with energy E and
Fourier transform M( f ). Clearly
and
1 2 2
x3 (t) = A m (t) sin(2φ). (3.333)
2 c
Note that
∞
m (t)
2
M(α)M( f − α) dα
α=−∞
= G( f ). (3.334)
Therefore
∞
G(0) = |M(α)|2 dα
α=−∞
=E (3.335)
lim H1 ( f ) = δ( f ). (3.336)
f →0
Hence
X 4 ( f ) = X 3 ( f )δ( f )
1
= A2c sin(2φ)Eδ( f )
2
1
A2c sin(2φ)E
2
= x4 (t). (3.337)
Thus, the (dc) control signal x4 (t) determines the phase of the voltage con-
trolled oscillator (VCO). Clearly, x4 (t) = 0 when
3 Amplitude Modulation 239
x1 (t)
H(f )
Phase discriminator
2 cos(2πfc t + φ)
2 sin(2πfc t + φ)
H(f )
x2 (t)
H1 (f )
H(f )
1 1/Δf
f f
−W 0 W
Δf
ae−t
t
0
T /5 T
−7
2φ = nπ
⇒ φ = nπ/2. (3.338)
according to:
Assume that T 1/ f c .
• Solution: Since the message is zero- mean we must have:
T
m(t) dt = 0
t=0
T /5
1 T
⇒ ae−t dt =
×7× T −
t=0 2 5
14T
⇒a= . (3.340)
5(1 − e−T /5 )
Let
1 T /5 2 −2t
P1 = a e dt
T t=0
a2
= 1 − e−2T /5 . (3.344)
2T
Let
T
1
P2 = (mt + c)2 dt
T t=T /5
1 T 2 2
= m t + 2mct + c2 dt
T t=T /5
T
1 m2t 3
= + mct 2 + c2 t
T 3 t=T /5
3 Amplitude Modulation 241
m2T 2 1 1 1
= 1− + mT c 1 − +c 1−
2
3 125 25 5
2
35 124 24 4
= − +
4 3 × 125 25 5
= 13.0667, (3.345)
where
35
m=
4T
c = −mT. (3.346)
Now
Pm = P1 + P2 . (3.347)
References
The variation of the beat (difference) frequency ( f t (t) − fr (t)) with time is
plotted in Fig. 4.2b. Note that the number of beat cycles over the time duration
1/ f 0 is given by
t1 +1/ f 0
N = | f t (t) − fr (t)| dt
t=t1
= |area of ABCD| + |area of DEFG| . (4.3)
Note that
f1 = fc + f − f2 . (4.4)
© The Editor(s) (if applicable) and The Author(s), under exclusive license 243
to Springer Nature Switzerland AG 2021
K. Vasudevan, Analog Communications,
https://doi.org/10.1007/978-3-030-50337-6_4
244 4 Frequency Modulation
frequency
1/f0
fc + Δf
fc t
fc − Δf
ft (t) τ fr (t)
τ
frequency
1/f0 (a)
fc + Δf
Y
f2
t3 t4
t
fc
X
t1 t2 t5
fc − Δf
τ
ft (t) − fr (t)
(b)
E F
f1
A D G
t
−f1
B C
1/(2f0 )
Fig. 4.2 Variation of the instantaneous difference frequency with time in an FM radar
y = mt + c, (4.5)
4 Frequency Modulation 245
where
m = 4 f 0 f. (4.6)
At t = t1 + τ we have
f c + f = 4 f 0 f (t1 + τ ) + c. (4.7)
At t = t1 we have
f 2 = 4 f 0 f t1 + c
= fc + f − 4 f0 f τ . (4.8)
Therefore
f1 = fc + f − f2
= 4 f0 f τ . (4.9)
Now
1τ 1
N =4 f1 + 2 f1 −τ
22 2 f0
1 − 2τ f 0
= τ f1 + f1
f0
≈ τ f1 + f1 / f0 . (4.10)
N f 0 = f 1 (1 + τ f 0 )
≈ f1 = 4 f0 f τ , (4.11)
which is proportional to τ and hence twice the distance between the target and
the radar. In other words,
τ = 2x/c, (4.12)
where x is the distance between the target and the radar and c is the velocity
of light. Thus the FM radar can be used for ranging.
2. (Haykin 1983) The instantaneous frequency of a cosine wave is equal to fc − f
for |t| < T /2 and f c for |t| > T /2. Determine the spectrum of this signal.
• Solution: The Fourier transform of this signal is given by
246 4 Frequency Modulation
−T /2
S( f ) = cos(2π f c t)e−j 2π f t dt
t=−∞
T /2
+ cos(2π( f c + f )t)e−j 2π f t dt
t=−T /2
∞
+ cos(2π f c t)e−j 2π f t dt
t=T /2
∞
= cos(2π f c t)e−j 2π f t dt
t=−∞
T /2
+ (cos(2π( f c + f )t) − cos(2π f c t)) e−j 2π f t dt
t=−T /2
1
= [δ( f − f c ) + δ( f + f c )]
2
T
+ [sinc (( f − f c − f )T ) + sinc (( f + f c + f )T )]
2
T
− [sinc (( f − f c )T ) + sinc (( f + f c )T )] . (4.13)
2
3. (Haykin 1983) Single sideband modulation may be viewed as a hybrid form of
amplitude modulation and frequency modulation. Evaluate the envelope and the
instantaneous frequency of an SSB wave, in terms of the message signal and its
Hilbert transform, for the two cases:
(a) When only the upper sideband is transmitted.
(b) When only the lower sideband is transmitted.
Assume that the message signal is m(t), the carrier amplitude is Ac /2 and carrier
frequency is f c .
• Solution: The SSB signal can be written as
Ac
s(t) = m(t) cos(2π f c t) ± m̂(t) sin(2π f c t) . (4.14)
2
When the upper sideband is to be transmitted, the minus sign is used and when
the lower sideband is to be transmitted, the plus sign is used. The envelope of
s(t) is given by
Ac 2
a(t) = m (t) + m̂ 2 (t) (4.15)
2
is independent of whether the upper or lower sideband is transmitted. Note
that s(t) can be written as
The plus sign in the above equation is used when the upper sideband is trans-
mitted and the minus sign is used when the lower sideband is transmitted. The
total instantaneous phase is given by
1 dθtot (t)
f tot (t) = . (4.19)
2π dt
When the upper sideband is transmitted, the total instantaneous frequency is
given by
where m (t) denotes the derivative of m(t) and m̂ (t) denotes the derivative
of m̂(t). When the lower sideband is transmitted, the total instantaneous fre-
quency is given by
(a) Determine the envelope of s(t). What is the ratio of the maximum to the
minimum value of this envelope.
(b) Determine the total average power of the narrowband FM signal. Determine
the total average power in the sidebands.
(c) Assuming that s(t) in (4.22) can be written as
expand θ(t) in the form of a Maclaurin series. Assume that β < 0.3. What
is the power ratio of the third harmonic to the fundamental component.
Therefore, the maximum and the minimum values of the envelope are given
by
Amax = Ac 1 + β 2
Amin = Ac
Amax
⇒ = 1 + β2. (4.25)
Amin
A2c 2β 2 A2c
Ptot = + . (4.26)
2 8
The total average power in the sidebands is equal to
2β 2 A2c
Pmes = . (4.27)
8
Assuming that β 1, the narrowband FM signal can be written as
where θ(t) denotes the instantaneous phase of the message component and is
given by
Now, the Maclaurin series expansion of tan−1 (x) is (ignoring higher terms)
x3
tan−1 (x) ≈ x − . (4.30)
3
Thus
1
θ(t) = β sin(2π f m t) − β 3 sin3 (2π f m t). (4.31)
3
Using the fact that
3 sin(θ) − sin(3θ)
sin3 (θ) = (4.32)
4
(4.31) becomes
4 Frequency Modulation 249
Wideband FM
m(t) Narrowband Frequency signal
BPF
FM modulator multiplier n1 fc = 104 MHz
Frequency
f0 = 1 MHz multiplier n2
θ(t) ≈ β sin(2π f m t)
1
− β 3 (3 sin(2π f m t) − sin(2π(3 f m )t))
12
β3 β3
= β− sin(2π f m t) + sin(2π(3 f m )t). (4.33)
4 12
5. (Proakis and Salehi 2005) To generate wideband FM, we can first generate a
narrowband FM signal, and then use frequency multiplication to spread the signal
bandwidth. This is illustrated in Fig. 4.3, which is called the Armstrong-type FM
modulator. The narrowband FM has a frequency deviation of 1 kHz.
(a) If the frequency of the first oscillator is 1 MHz, determine n 1 and n 2 that
is necessary to generate an FM signal at a carrier frequency of 104 MHz
and a maximum frequency deviation of 75 kHz. The BPF allows only the
difference frequency component.
(b) If the error in the carrier frequency f c for the wideband FM signal is to
be within ±200 Hz, determine the maximum allowable error in the 1 MHz
oscillator.
75
n1 = = 75. (4.35)
1
Consequently, the carrier frequency at the output of the first frequency multi-
plier is 75 MHz. However, the required carrier frequency is 104 MHz. Hence,
what we now require is a frequency translation. Therefore, we must have
250 4 Frequency Modulation
1 × n 2 − 75 = 104 MHz
⇒ n 2 = 179. (4.36)
Let us assume that the error in f 0 is x Hz. Hence, the error in the final output
carrier frequency is
n 2 x − n 1 x = ±200 Hz
⇒ x = ±1.92 Hz. (4.37)
6. Figure 4.4 shows the Fourier transform of a frequency discriminator for negative
frequencies. Here BT denotes the transmission bandwidth of the FM signal and
f c denotes the carrier frequency.
The block diagram of the system proposed for frequency demodulation is also
shown, where
t
s(t) = Ac cos 2π f c t + 2πk f m(τ ) dτ , (4.38)
τ =0
H (− f ) = H ∗ ( f ), (4.39)
4 Frequency Modulation 251
H(f )/j
πaBT
−fc − BT /2 fc − BT /2 f
0
−fc −fc + BT /2 fc fc + BT /2
−πaBT
where
We know that the Fourier transform of the complex envelope of h(t) is given
by
H ( f + f c ) for − BT /2 < f < BT /2
H̃ ( f ) = , (4.43)
0 elsewhere
which reduces to
j 2πa f for − BT /2 < f < BT /2
H̃ ( f ) = . (4.44)
0 elsewhere
Let us denote the Fourier transform of the complex envelope of the output by
Ỹ ( f ). Then
252 4 Frequency Modulation
Ỹ ( f ) = H̃ ( f ) S̃( f )
= j 2π f a S̃( f ), (4.45)
where S̃( f ) is the Fourier transform of the complex envelope of s(t). Note
that
t
s̃(t) = Ac exp j 2πk f m(τ ) dτ . (4.46)
τ =0
d s̃(t)
ỹ(t) = a
dt
t
= j a Ac 2πk f m(t) exp j 2πk f m(τ ) dτ
τ =0
t
= 2a Ac πk f m(t) exp j 2πk f m(τ ) dτ + j π/2 . (4.47)
τ =0
Therefore
t
y(t) = 2πa Ac k f m(t) cos 2π f c t + 2πk f m(τ ) dτ + π/2 . (4.48)
τ =0
The output is
7. Figure 4.6 shows the Fourier transform of a frequency discriminator for negative
frequencies. Here BT denotes the transmission bandwidth of the FM signal and
f c denotes the carrier frequency.
The block diagram of the system proposed for frequency demodulation is also
shown, where
t
s(t) = Ac cos 2π f c t + 2πk f m(τ ) dτ , (4.50)
τ =0
−2πaBT
H (− f ) = H ∗ ( f ), (4.51)
H(f )/j
2πaBT
−fc − BT /2 −fc + BT /2 f
0
−fc fc − BT /2 fc fc + BT /2
−2πaBT
where
We know that the Fourier transform of the complex envelope of h(t) is given
by
H ( f + f c ) for − BT /2 < f < BT /2
H̃ ( f ) = , (4.55)
0 elsewhere
which reduces to
−j 2πa( f − BT /2) for − BT /2 < f < BT /2
H̃ ( f ) = . (4.56)
0 elsewhere
Let us denote the Fourier transform of the complex envelope of the output by
Ỹ ( f ). Then
Ỹ ( f ) = H̃ ( f ) S̃( f )
= −j 2πa( f − BT /2) S̃( f ), (4.57)
where S̃( f ) is the Fourier transform of the complex envelope of s(t). Note
that
t
s̃(t) = Ac exp j 2πk f m(τ ) dτ . (4.58)
τ =0
d s̃(t)
ỹ(t) = −a + j πa BT s̃(t)
dt
t
= −j a Ac 2πk f m(t) exp j 2πk f m(τ ) dτ
τ =0
t
+j a Ac π BT exp j 2πk f m(τ ) dτ
τ =0
t
= j πa Ac BT − 2k f m(t) exp j 2πk f m(τ ) dτ . (4.59)
τ =0
Therefore
t
y(t) = πa Ac [BT − 2k f m(t)] cos 2π f c t + 2πk f m(τ ) dτ + π/2 .
τ =0
(4.60)
The output is
4 Frequency Modulation 255
Therefore, the proposed system can be used for frequency demodulation pro-
vided
2k f m(t)/BT < 1. (4.62)
8. (Haykin 1983) Consider the frequency demodulation scheme shown in Fig. 4.8
in which the incoming FM signal is passed through a delay line that produces a
delay of T such that 2π f c T = π/2. The delay-line output is subtracted from the
incoming FM signal and the resulting output is envelope detected. This demod-
ulator finds wide application in demodulating microwave FM waves. Assuming
that
2π f m T 1. (4.64)
Let
θ(t) = β sin(2π f m t)
α(t) = θ(t) − 2π f m T β cos(2π f m t). (4.66)
where
sin(θ(t)) + cos(α(t))
tan(φ(t)) = (4.68)
cos(θ(t)) − sin(α(t))
2π f m T 1
β < 1
⇒ θ(t) − α(t) 1
⇒ sin(θ(t) − α(t)) ≈ θ(t) − α(t). (4.70)
Note that a(t) > 0 for all t. The message signal cos(2π f m t) can be recovered
by passing a(t) through a capacitor.
9. Consider the frequency demodulation scheme shown in Fig. 4.9 in which the
incoming FM signal is passed through a delay line that produces a delay of T
such that 2π f c T = π/2. The delay-line output is subtracted from the incoming
FM signal and the resulting output is envelope detected. This demodulator finds
wide application in demodulating microwave FM waves. Assuming that
t
s(t) = Ac cos 2π f c t + 2πk f m(τ ) dτ (4.71)
τ =0
2π f T 1 (4.72)
where f denotes the frequency deviation. Assume also that m(t) is constant
over any interval T .
• Solution: The signal x(t) can be written as
where
t−T
s(t − T ) = Ac cos 2π f c (t − T ) + 2πk f m(τ ) dτ
τ =0
t−T
= Ac cos 2π f c t − π/2 + 2πk f m(τ ) dτ
τ =0
t−T
= Ac sin 2π f c t + 2πk f m(τ ) dτ . (4.74)
τ =0
Now
t−T t t
2πk f m(τ ) dτ = 2πk f m(τ ) dτ − 2πk f m(τ ) dτ
τ =0 τ =0 τ =t−T
t
= 2πk f m(τ ) dτ − 2πk f T m(t)
τ =0
= θ(t) − 2πk f T m(t)
= α(t) (say), (4.75)
where
t
2πk f m(τ ) dτ = θ(t). (4.76)
τ =0
where
258 4 Frequency Modulation
sin(θ(t)) + cos(α(t))
tan(φ(t)) = (4.78)
cos(θ(t)) − sin(α(t))
f = max |k f m(t)|
2π f T 1 (given)
⇒ |θ(t) − α(t)| 1
⇒ sin(θ(t) − α(t)) ≈ θ(t) − α(t). (4.80)
Note that a(t) > 0 for all t. The message signal m(t) can be recovered by
passing a(t) through a capacitor.
10. A tone-modulated FM signal of the form:
is passed through an ideal unity gain BPF with center frequency equal to the
carrier frequency and bandwidth equal to 3 f m (±1.5 f m on either side of the
carrier), yielding the signal z(t).
(a) Derive the expression for z(t).
(b) Assuming that z(t) is of the form
where Jn (β) denotes the nth order Bessel function of the first kind and
argument β.
• Solution: The input FM signal can be written as
4 Frequency Modulation 259
where
which is periodic with a period 1/ f m . Hence, s̃(t) can be expanded in the form
of a complex Fourier series given by
∞
s̃(t) = cn e j 2πn fm t , (4.86)
n=−∞
where
1/(2 f m )
cn = f m s̃(t)e−j 2πn fm t dt
t=−1/(2 f m )
1/(2 fm )
= −j Ac f m e j β cos(2π fm t) e−j 2πn fm t dt. (4.87)
t=−1/(2 f m )
Let
2π f m t = π/2 − x
⇒ 2π f m dt = −d x. (4.88)
Thus
−π/2
j Ac
cn = e j(β sin(x)−n(π/2−x)) d x. (4.89)
2π x=3π/2
Therefore
260 4 Frequency Modulation
∞
s̃(t) = Ac J−n (β)e j (2πn fm t−(n+1)π/2) . (4.91)
n=−∞
Thus
∞
s(t) = Ac J−n (β) cos(2π f c t + 2πn f m t − (n + 1)π/2). (4.92)
n=−∞
we have
The phase is
J0 (β)
θ(t) = − tan−1 . (4.97)
2J1 (β) cos(2π f m t)
11. (Haykin 1983) The bandwidth of an FM signal extends over both sides of the
carrier frequency. However, in the single sideband version of FM, it is possible
to transmit either the upper or the lower sideband.
(a) Assuming that the FM signal is given by
explain how we can transmit only the upper sideband. Express your result in
terms of complex envelope of s(t) and Hilbert transforms. Assume that for
all practical purposes, s(t) is bandlimited to f c − BT /2 < | f | < f c + BT /2
and f c
BT .
(b) Verify your answer for single-tone FM modulation when
• Solution: Recall that in SSB modulation with upper sideband transmitted, the
signal is given by
where m̂(t) is the Hilbert transform of the message m(t), which is typically
bandlimited between [−W, W ] and f c
W . Observe that m(t) in (4.101)
can be complex. For the case of the FM signal given by
where
s̃(t) is the Hilbert transform of s̃(t).
Now in the given example
−1
s̃(t) = Ac Jn (β) (cos(2πn f m t + π/2) + j sin(2πn f m t + π/2))
n=−∞
∞
+Ac Jn (β) (cos(2πn f m t − π/2) + j sin(2πn f m t − π/2))
n=1
−1
= Ac Jn (β) (− sin(2πn f m t) + j cos(2πn f m t))
n=−∞
∞
+Ac Jn (β) (sin(2πn f m t) − j cos(2πn f m t)) (4.107)
n=1
Thus, we find that only the upper sideband is transmitted, which verifies our
result.
12. (Haykin 1983) Consider the message signal as shown in Fig. 4.10, which is used
to frequency modulate a carrier. Assume a frequency sensitivity of k f Hz/V and
that the FM signal is given by
t
−T0 /4 0 T0 /4
−1
T0 /2 T0 /2
where
β−n β+n
cn = an sinc + bn sinc , (4.112)
2 2
T0
− A + 2πk f =A
2
T0
⇒ A = πk f . (4.114)
2
t
−T0 /4 0 T0 /4
−1
T0 /2 T0 /2
f (t)
(b)
fc + k f
fc t
−T0 /4 0 T0 /4
fc − k f
T0 /2 T0 /2
φ(t)
(c)
t
0
−T0 /4 T0 /4 3T0 /4
−A
T0 /2 T0 /2
Therefore
4 Frequency Modulation 265
3T0 /4
1
cn = s̃(t)e−j 2πnt/T0 dt
T0 t=−T0 /4
T0 /4
Ac
= e j 2πt (k f −n/T0 )
T0 t=−T0 /4
3T0 /4
Ac
+ e j πβ e−j 2πt (k f +n/T0 ) . (4.117)
T0 t=T0 /4
Now
3nπ 3π nπ
− = − + π n − nπ = − − nπ. (4.120)
2 2 2
Similarly
nπ π nπ
− = − + π n − nπ = − nπ. (4.121)
2 2 2
Substituting (4.120) and (4.121) in (4.119), we get
Ac −j nπ β+n
II = e sinc
2 2
Ac β+n
= (−1)n sinc . (4.122)
2 2
Therefore
Ac β−n Ac β+n
cn = sinc + (−1)n sinc (4.123)
2 2 2 2
Ac
an =
2
Ac
bn = (−1)n . (4.124)
2
266 4 Frequency Modulation
τ
fc + Δf
ft (t) fr (t)
fc
fc − Δf
1/f0
The beat (difference) frequency is f t (t) − fr (t). The number of beat cycles
over the time duration 1/ f 0 is given by
4 Frequency Modulation 267
1/ f 0
N= | f t (t) − fr (t)| dt. (4.128)
t=0
Note that
fr (t) = f c + f sin(2π f 0 (t − τ ))
= f c + f [sin(2π f 0 t) cos(2π f 0 τ ) − cos(2π f 0 t) sin(2π f 0 τ )]
≈ f c + f [sin(2π f 0 t) − 2π f 0 τ cos(2π f 0 t)] . (4.129)
Therefore
N f 0 = 4τ f f 0 . (4.132)
cos(2πfc t)
Ac
sin(2πfc t)
Ac
S( f ) = [δ( f − f c ) + δ( f + f c )]
2
Ac β p
− [δ( f − f c + f m ) − δ( f + f c − f m )]
4j
Ac β p
− [δ( f − f c − f m ) − δ( f + f c + f m )] . (4.135)
4j
which is periodic with period 1/ f m and hence can be represented in the form
of a Fourier series as follows:
∞
s̃(t) = cn exp( j 2πn f m t). (4.139)
n=−∞
Let
2π f m t = π/2 − x. (4.141)
Then
−π/2
−Ac
cn = exp( j k p Am sin(x) − j (nπ/2 − nx)) d x. (4.142)
2π x=3π/2
Since the above integrand is periodic with a period of 2π, we can write
π
Ac
cn = exp(−j nπ/2) exp( j k p Am sin(x) + j nx)) d x. (4.143)
2π x=−π
We know that
π
1
Jn (β) = exp( j β sin(x) − j nx)) d x, (4.144)
2π x=−π
where Jn (β) is the nth-order Bessel function of the first kind and argument β.
Thus
x(t) = Ac J0 (β p ) cos(2π f c t)
+Ac J−1 (β p ) cos(2π( f c + f m )t − π/2)
+Ac J1 (β p ) cos(2π( f c − f m )t + π/2)
= Ac J0 (β p ) cos(2π f c t)
+Ac J−1 (β p ) sin(2π( f c + f m )t)
−Ac J1 (β p ) sin(2π( f c − f m )t). (4.147)
However
Hence
x(t) = Ac J0 (β p ) cos(2π f c t)
−Ac J1 (β p ) sin(2π( f c + f m )t)
−Ac J1 (β p ) sin(2π( f c − f m )t)
= Ac J0 (β p ) cos(2π f c t)
−2 Ac J1 (β p ) sin(2π f c t) cos(2π f m t)
= a(t) cos(2π f c t + θi (t)). (4.149)
−1 2J1 (β p )
φi (t) = 2π f c t + tan cos(2π f m t) . (4.151)
J0 (β p )
1 d 2J1 (β p )
f i (t) = f c + tan−1 cos(2π f m t)
2π dt J0 (β p )
2J1 (β p )/J0 (β p )
= fc − f m sin(2π f m t)
1 + (2J1 (β p )/J0 (β p ))2 cos2 (2π f m t)
2J1 (β p )J0 (β p )
= fc − 2 f m sin(2π f m t).
J0 (β p ) + (2J1 (β p ))2 cos2 (2π f m t)
(4.152)
16. (Haykin 1983) A carrier wave is frequency modulated using a sinusoidal signal
of frequency f m and amplitude Am .
(a) Determine the values of the modulation index β for which the carrier com-
ponent of the FM signal is reduced to zero.
(b) In a certain experiment conducted with f m = 1 kHz and increasing Am start-
ing from zero volts, it is found that the carrier component of the FM signal
is reduced to zero for the first time when Am = 2 V. What is the frequency
sensitivity of the modulator? What is the value of Am for which the carrier
component is reduced to zero for the second time?
• Solution: From the table of Bessel functions, we see that J0 (β) is equal to zero
for
β = 2.44
β = 5.52
β = 8.65
β = 11.8. (4.153)
The value of Am for which the carrier component goes to zero for the second
time is equal to
β fm
Am =
kf
5.52
⇒ Am =
1.22
= 4.52 V. (4.155)
272 4 Frequency Modulation
Ac
3
X( f ) = Jn (2)δ( f − f c − n f m ) + δ( f + f c + n f m ), (4.157)
2 n=−3
0.6
0.35 0.35
0.3
0.08
fc − 3fm fc − fm
fc − 2fm fc fc + fm fc + 3fm
−0.08
fc + 2fm
−0.6
f 1 = k p Am f m , (4.160)
BT, PM = 2( f 1 + f m )
= 2 f m (k p Am + 1)
≈ 2 f m k p Am , (4.162)
which varies linearly with f m . However, in the case of FM, the frequency
deviation is
f 2 = k f Am , (4.163)
BT, FM = 2( f 2 + f m )
= 2(k f Am + f m ). (4.164)
19. Figure 4.15 shows the block diagram of a system. Here s(t) and h(t) are FM
signals, given by
s(t)
274 4 Frequency Modulation
(a) Using the method of complex envelopes, determine the output of h(t).
(b) Determine the envelope of the output of h(t).
• Solution: Let the signal at the output of the multiplier be denoted by v1 (t).
Then
1 j π fm t
h̃(t) = e , (4.169)
2
where we have assumed that
h(t) = 2h̃(t)e j 2π fc t . (4.170)
1
| ỹ(t)| = |G( f m )|. (4.174)
2
Hence, the envelope of the output signal is proportional to the magnitude
response of g(t) evaluated at f = f m .
20. (Haykin 1983) Figure 4.16 shows the block diagram of the transmitter and
receiver for stereophonic FM. The input signals l(t) and r (t) represent left-
hand and right-hand audio signals. The difference signal x1 (t) = l(t) − r (t) is
DSB-SC modulated as shown in the figure, with f c = 25 kHz. The DSB-SC
wave, x2 (t) = l(t) + r (t) and the pilot carrier are summed to produce the com-
posite signal m(t). The composite signal m(t) is used to frequency modulate
a carrier and the resulting FM signal is transmitted. Assume that f 2 = 20 kHz,
f 1 = 200 Hz.
(a) Sketch the spectrum of m(t). Label all the important points on the x- and
y-axes. The spectrums of l(t) and r (t) are shown in Fig. 4.16.
(b) Assuming that the frequency deviation of the FM signal is 90 kHz, find the
transmission bandwidth of the FM signal using Carson’s rule.
(c) In the receiver block diagram determine the signal y(t), the input-output
characteristics of the device, and the specifications of filter1, filter2, filter3,
and filter4 in the frequency domain. Assume ideal filter characteristics.
l(t) + x1 (t)
Transmitter
−
cos(4πfc t)
Stereophonic
r(t)
Freq Pilot FM signal
carrier s(t)
doubler
source
+
+ m(t) FM
cos(2πfc t) modulator
x2 (t)
f f
Receiver
s(t) FM m(t) x2 (t)
Filter1
demodulator
y(t) cos(2πfc t)
Filter2 Device Filter3
2 cos(4πfc t)
x1 (t)
Filter4
M (f )
0.5
−f6 −f5 −f4 −f3 f3 f4 f5 f6
0 f
−2fc −fc −f2 −f1 f1 f2 fc 2fc
−0.5
f3 = 2fc − f2 f5 = 2fc + f1
f4 = 2fc − f1 f6 = 2fc + f2
π
t × 10−6 sec
0
2 3 5 6
BT = 2( f + W ), (4.176)
s(t) = A sin(θ(t)),
where θ(t) is a periodic waveform as shown in Fig. 4.18. The message signal
has zero mean. Compute the carrier frequency.
• Solution: We know that the instantaneous frequency is given by
1 dθ
f (t) =
2π dt
= f c + k f m(t), (4.177)
which is plotted in Fig. 4.19b. Observe that f (t) is also periodic. Since m(t)
has zero mean, the mean value of f (t) is f c , where
25 × 2 + 50 × 1
fc = × 104
3
100
= × 104 Hz. (4.178)
3
22. A 10 kHz periodic square wave g p (t) is applied to a first-order R L lowpass filter
as shown in Fig. 4.20. It is given that R/(2πL) = 10 kHz. The output signal m(t)
is FM modulated with frequency deviation equal to 75 kHz.
Determine the bandwidth of the FM signal s(t), using Carson’s rule. Ignore those
278 4 Frequency Modulation
(a)
θ(t) (radians)
2π
π
t × 10−6 sec
0
2 3 5 6
(b)
f (t) (Hz)
50 × 104
25 × 104
t × 10−6 sec
0
3 2 6 5
gp (t)
L
gp (t) m(t) s(t)
1/T = 10 kHz FM
B modulator
t
R
0
−B
T /2 T /2
harmonic terms in m(t) whose (absolute value of the) amplitude is less than 1%
of the fundamental.
• Solution: We know that
∞
4B (−1)m
g p (t) = cos [2π(2m + 1) f 0 t] , (4.179)
π m=0 (2m + 1)
R
H (ω) =
R + j ωL
1
= , (4.180)
1 + j ω/ω0
4 Frequency Modulation 279
where
1
Am = |H ((2m + 1)ω0 )| = (4.182)
1 + (2m + 1)2
and
4B 1 1 4B 1
< 0.01 √
π (2m + 1) 1 + (2m + 1)2 π 2
1 1
⇒ < 0.00707 for m > 0. (4.184)
(2m + 1) 1 + (2m + 1)2
FM
Frequency
signal
multiplier
n2
Narrowband Frequency
Message
Mixer
FM multiplier
modulator n1
Oscillator Oscillator
Thus, the frequency deviation of the output FM signal is 60 kHz. The mod-
ulation index is 60/5 = 12. The frequency separation in the spectrum of the
output FM signal is unchanged at 5 kHz.
24. (Haykin 1983) Figure 4.21 shows the block diagram of a wideband frequency
modulator using the indirect method. Note that a mixer is essentially a multiplier,
followed by a bandpass filter which allows only the difference frequency com-
ponent. This transmitter is used to transmit audio signals in the range 100 Hz to
15 kHz. The narrowband frequency modulator is supplied with a carrier of fre-
quency f 1 = 0.1 MHz. The second oscillator supplies a frequency of 9.5 MHz.
The system specifications are as follows: carrier frequency at the transmitter
output f c = 100 MHz with frequency deviation, f = 75 kHz. Maximum mod-
ulation index at the output of the narrowband frequency modulator is 0.2 rad.
(a) Calculate the frequency multiplication ratios n 1 and n 2 .
(b) Specify the value of the carrier frequency at the output of the first frequency
multiplier.
• Solution: It is given that the frequency deviation of the output FM wave is
equal to 75 kHz. Note that for a tone-modulated FM signal, the frequency
deviation is related to the frequency of the modulating wave as
f = β fm . (4.188)
4 Frequency Modulation 281
75
n1n2 =
0.02
= 3750. (4.190)
The carrier frequency at the output of the first multiplier is 0.1n 1 MHz. The
carrier frequency at the output of the second multiplier is
n 1 = 75
n 2 = 50. (4.192)
(C). The capacitance in the varactor diode is related to the voltage V (t) applied
across its terminals by
Ci (t) = 100/ V (t) pF. (4.194)
1 1
f i (t) = √ . (4.195)
2π L(C + Ci (t))
1 1
f i (t) = √ . (4.196)
2π L(C + Ci (t))
1 1
fc =
√
2π L(C + C0 )
⇒ C + C0 = 126.65 pF
⇒ C0 = 26.651 pF
⇒ 100/ Vb = 26.651
⇒ Vb = 14.078 V. (4.197)
Since the final carrier frequency is 64 MHz and the frequency multiplication
factor is 64. Thus, the modulation index at the VCO output is
5
β= = 0.078. (4.198)
64
Thus, the FM signal at the VCO output is narrowband. The instantaneous
frequency of oscillation at the VCO output is
1 1
f i (t) = . (4.199)
2π L(C + 100(Vb + Vm sin(2π f m t))−0.5 )
Since Vb
Vm the instantaneous frequency can be approximated as
4 Frequency Modulation 283
1 1
f i (t) ≈
2π L(C + 100V −0.5 (1 − V /(2V ) sin(2π f t)))
b m b m
1 1
= √ . (4.200)
2π L(C + C0 (1 − Vm /(2Vb ) sin(2π f m t)))
1 1
f i (t) = √
2π L(C + C0 − C0 Vm /(2Vb ) sin(2π f m t)))
1 1
= √
2π L(C + C0 )
1
√
1 − C0 Vm /(2Vb (C + C0 )) sin(2π f m t)
C0 Vm
≈ fc 1 + sin(2π f m t) . (4.201)
4Vb (C + C0 )
C0 Vm f c
β= = 0.078. (4.202)
4Vb f m (C + C0 )
Vm = 0.2087 V. (4.203)
1 1
f i (t) = √ . (4.205)
2π L(C + Ci (t))
1 1
f i (t) = √ . (4.206)
2π L(C + Ci (t))
1 1
fc =
√
2π L(C + C0 )
⇒ C + C0 = 253.303 pF
⇒ C0 = 53.303 pF
⇒ 100/ Vb = 53.303
⇒ Vb = 3.5196 V. (4.207)
Since the final carrier frequency is 128 MHz, the frequency multiplication
factor is 128/2 = 64. Thus, the modulation index at the VCO output is
6
β= = 0.09375. (4.208)
64
Thus, the FM signal at the VCO output is narrowband. The instantaneous
frequency of oscillation at the VCO output is
1 1
f i (t) = . (4.209)
2π L(C + 100(Vb + Vm sin(2π f m t))−0.5 )
Since Vb
Vm the instantaneous frequency can be approximated as
4 Frequency Modulation 285
1 1
f i (t) ≈
2π L(C + 100V −0.5 (1 − V /(2V ) sin(2π f t)))
b m b m
1 1
= √ . (4.210)
2π L(C + C0 (1 − Vm /(2Vb ) sin(2π f m t)))
1 1
f i (t) = √
2π L(C + C0 − C0 Vm /(2Vb ) sin(2π f m t)))
1 1
= √
2π L(C + C0 )
1
√
1 − C0 Vm /(2Vb (C + C0 )) sin(2π f m t)
C0 Vm
≈ fc 1 + sin(2π f m t) . (4.211)
4Vb (C + C0 )
C0 Vm f c
β= = 0.09375. (4.212)
4Vb f m (C + C0 )
Vm = 0.0627 V. (4.213)
is applied to the system shown in Fig. 4.24. Assume that the resistance R is small
compared to the impedance of C for all significant frequency components of s(t)
and the envelope detector does not load the filter. Determine the resulting signal
at the envelope detector output assuming that k f |m(t)| < f c for all t.
• Solution: The transfer function of the highpass filter is
R
H( f ) =
R + 1/(j 2π f C)
≈ j 2π f RC (4.215)
provided
286 4 Frequency Modulation
1
R . (4.216)
2π f C
Thus, over the range of frequencies in which (4.216) is valid, the highpass
filter acts like an ideal differentiator. Hence, the output of the highpass filter
is given by
ds(t)
x(t) = RC
dt
t
= −RC Ac 2π f c + 2πk f m(t) sin 2π f c t + 2πk f m(τ ) dτ .
τ =0
(4.217)
where ω = 20,000 rad/s. Compute the frequency deviation and the modulation
index.
• Solution: The total instantaneous frequency is
5k f = 50 kHz. (4.222)
4 Frequency Modulation 287
10πk f
β= = 5π. (4.224)
ω
29. (Haykin 1983) Suppose that the received signal in an FM system contains some
residual amplitude modulation as shown by
where a(t) > 0 and f c is the carrier frequency. The phase φ(t) is related to the
modulating signal m(t) by
t
φ(t) = 2πk f m(τ ) dτ . (4.226)
τ =0
ds(t)
s (t) = = a (t) cos(2π f c t + φ(t))
dt
−a(t) sin(2π f c t + φ(t))(2π f c + φ (t)). (4.227)
|φ (t)|
|a (t)| (4.229)
where we have assumed that [ f c + k f m(t)] > 0. Thus, we see that there is
distortion due to a(t), at the envelope detector output.
30. (Haykin 1983) Let
where a(t) > 0, be applied to a hard limiter whose output z(t) is defined by
z(t) = sgn[s(t)]
+1 for s(t) > 0
= (4.233)
−1 for s(t) < 0.
(a) Show that z(t) can be expressed in the form of a Fourier series as follows:
∞
4 (−1)n
z(t) = cos[2π f c t (2n + 1) + (2n + 1)φ(t)]. (4.234)
π n=0 2n + 1
(b) Compute the output when z(t) is applied to an ideal bandpass filter with cen-
ter frequency f c and bandwidth BT , where BT is the transmission bandwidth
of s(t) in the absence of amplitude modulation. Assume that f c
BT .
Let
This is illustrated in Fig. 4.25. Hence, z(t) can be written in the form of a Fourier
series with respect to α(t) as follows:
4 Frequency Modulation 289
cos(α(t)) z(α(t))
0.5
-0.5
-1
-6 -4 -2 0 2 4 6
α(t)
∞
z(α(t)) = 2 an cos(nα(t)), (4.239)
n=1
where
π
1
an = z(α(t)) cos(nα(t)) dα(t)
2π α(t)=−π
π
1
= z(α(t)) cos(nα(t)) dα(t). (4.240)
π α(t)=0
Thus
290 4 Frequency Modulation
∞
4(−1)m
z(x) = cos((2m + 1)x)
m=0
π(2m + 1)
∞
4(−1)m
⇒ z(α(t)) = cos((2m + 1)α(t))
m=0
π(2m + 1)
∞
4(−1)m
= cos((2m + 1)2π f c t + (2m + 1)φ(t)).
m=0
π(2m + 1)
(4.242)
Observe that the mth harmonic has a carrier frequency at (2m + 1) f c and band-
width (2m + 1)BT , where BT is the bandwidth of s(t) with amplitude modula-
tion removed, that is, a(t) = Ac . If the fundamental component at m = 0 is to
be extracted from z(α(t)), we require
BT BT
< (2m + 1) f c − (2m + 1)
fc +
2 2
BT 1
⇒ 1+ < fc for m > 0. (4.243)
2 m
f c > BT . (4.244)
Now if z(α(t)) is passed through an ideal bandpass filter with center frequency
f c and bandwidth BT , the output is (assuming (4.244) is satisfied)
4
y(t) = cos(2π f c t + φ(t)), (4.245)
π
which has no amplitude modulation.
31. The message signal
4
sin(t B)
m(t) = A (4.246)
t
BT = 2( f + W ), (4.247)
4 Frequency Modulation 291
where
f = k f A4 B 4 (4.249)
A
Asinc(t B) rect( f /B). (4.250)
B
Time scaling by 1/π, we obtain
Aπ
Asinc(t B/π) rect( f π/B)
B
sin(t B) Aπ
⇒A rect( f π/B)
tB B
sin(t B)
⇒A Aπ rect( f π/B), (4.251)
t
32. Explain the principle of operation of the PLL demodulator for FM signals. Draw
the block diagram and clearly state the signal model and assumptions.
• Solution: Consider the block diagram in Fig. 4.26. Here s(t) denotes the input
FM signal (bandlimited to [ f c − BT /2, f c + BT /2]; f c is the carrier fre-
quency, BT is the bandwidth of s(t)) given by
where
t
φ1 (t) = 2πk f m(τ ) dτ , (4.254)
τ =−∞
Mixer
Loop
s(t) e(t) v(t)
LPF filter
h(t)
r(t)
VCO
Fig. 4.26 Block diagram of the phase locked loop (PLL) demodulator for FM signals
where
t
φ2 (t) = 2πkv v(τ ) dτ , (4.256)
τ =−∞
where kv is the frequency sensitivity of the VCO in Hz/V and v(·) is the control
signal at the VCO input. The lowpass filter eliminates the sum frequency
component at the multiplier output and allows only the difference frequency
component. Hence
where
K 0 = km kv Ac Av Hz. (4.260)
j 2π f e ( f ) = j 2π f 1 ( f ) − 2πK 0 e ( f )H ( f )
j f 1 ( f )
⇒ e ( f ) =
j f + K0 H ( f )
1 ( f )
⇒ e ( f ) = . (4.264)
1 + K 0 H ( f )/(j f )
Now φ1 (t) is bandlimited to [−W, W ]. It will be shown later that φ2 (t) is also
bandlimited to [−W, W ]. Therefore, both φe (t) and h(t) are bandlimited to
[−W, W ]. If
K0 H ( f )
1 for | f | < W (4.265)
f
j f 1 ( f )
e ( f ) = . (4.266)
K0 H ( f )
E( f ) = Ac Av km e ( f ) (4.267)
and
294 4 Frequency Modulation
V ( f ) = E( f )H ( f )
= Ac Av km j f 1 ( f )/K 0
= (j f /kv ) 1 ( f ). (4.268)
1 dφ1 (t)
v(t) =
2πkv dt
kf
= m(t). (4.269)
kv
From (4.269), it is clear that v(t) and φ2 (t) are bandlimited to [−W, W ].
References
is transmitted over a channel that adds additive Gaussian noise with psd shown
in Fig. 5.1. The message spectrum extends over [−4, 4] kHz and the carrier
frequency is 200 kHz. Assuming that the average power of S(t) is 10 W and
coherent detection, determine the output SNR of the receiver.
Assume that the IF filter is ideal with unity gain in the passband and zero for
other frequencies, and the narrowband representation of noise at the IF filter
output is
© The Editor(s) (if applicable) and The Author(s), under exclusive license 295
to Springer Nature Switzerland AG 2021
K. Vasudevan, Analog Communications,
https://doi.org/10.1007/978-3-030-50337-6_5
296 5 Noise in Analog Modulation
10−6
f (kHz)
-400 0 400
0.49
f (kHz)
= 8000 × 10−6 W.
(5.5)
Note that
where we have used the fact that the cross spectral density S Nc Ns ( f ) is an odd
function, therefore R Nc Ns (0) = 0.
The power of the modulated message signal is
filter
2 cos(2πfc t + Θ(t))
The psd of M(t) extends over [−W, W ] with power P. The LPF is ideal with
unity gain in [−W, W ].
• Solution: The output of the multiplier is
We now use the following relations (assuming |(t)| 1 for all t):
4 (t)
(1 − cos((t)))2 ≈
4
E Nc2 (t) = 2N0 W
E Ns2 (t) = 2N0 W
E M 2 (t) = P. (5.14)
Thus
E (Y0 (t) − Y (t))2
A2 P
= c E 4 (t) + 2N0 W E cos2 ((t)) + sin2 ((t))
4
3A2c Pσ4
= + 2N0 W, (5.15)
4
where we have used the fact that if X is a zero-mean Gaussian random variable
with variance σ 2
E X 2n = 1 × 3 × · · · × (2n − 1)σ 2n . (5.16)
3. (Haykin 1983) Let the message M(t) be transmitted using SSB modulation. The
psd of M(t) is
a| f |
for | f | < W
SM ( f ) = W (5.17)
0 elsewhere,
where a and W are constants. Let the transmitted signal be of the form:
since
E Nc2 (t) = E Ns2 (t) = E N 2 (t)
= N0 W
2
E cos (θ ) + sin (θ ) = 1.
2
(5.23)
A2c aW a A2c
SNR O = = . (5.24)
N0 W N0
The output signal power is A2c and the output noise power is 2N0 W . Hence
A2c
SNR O = . (5.28)
2N0 W
5. (Haykin 1983) A frequency division multiplexing (FDM) system uses SSB mod-
ulation to combine 12 independent voice channels and then uses frequency mod-
ulation to transmit the composite signal. Each voice signal has an average power
P and occupies the frequency band [−4, 4] kHz. Only the lower sideband is
transmitted. The modulated voice signals used for the first stage of modulation
are defined by
for 1 ≤ k ≤ 12, where f 0 = 4 kHz. Note that E[Mk2 (t)] = P. The received sig-
nal consists of the transmitted FM signal plus zero-mean white Gaussian noise
of psd N0 /2.
Assume that the output of the FM receiver is given by
where
12
S(t) = Sk (t). (5.31)
k=1
The overall signal at the output of the first stage of modulation is given by
12
S(t) = Sk (t). (5.34)
k=1
The bandwidth of S(t) extends over [−48, 48] kHz. The signal S(t) is given
as input to the second stage, which is a frequency modulator.
The output of the FM receiver is given by
where Ac is the amplitude of the transmitted FM signal. The noise power in the
kth received voice band is
kB
PNk = 2 S No ( f ) d f
(k−1)B
2N0 B 3 2
= 3k − 3k + 1 , (5.37)
3A2c
Pk = k 2f A2k P. (5.38)
The output SNR for the kth SSB modulated voice signal is
3k 2f A2k P A2c
SNR O, k = . (5.39)
2N0 B 3 (3k 2 − 3k + 1)
where C is a constant.
302 5 Noise in Analog Modulation
2 cos(2πfc t + θ)
6. Consider the system shown in Fig. 5.4. Here, H ( f ) is an ideal LPF with unity
gain in the band [−W, W ]. Carefully follow the two procedures outlined below:
(a) Let
Therefore
RY (τ ) = E[Y (t)Y (t − τ )]
1
= R M (τ ) 1 + cos(4π f c τ ) . (5.43)
2
Hence
1
SY ( f ) = S M ( f ) + [S M ( f − 2 f c ) + S M ( f + 2 f c )] . (5.44)
4
The psd of Z (t) is
S Z ( f ) = SY ( f )|H ( f )|2
= S M ( f ). (5.45)
5 Noise in Analog Modulation 303
Hence
RY (τ ) = 2R X (τ ) cos(2π f c τ )
⇒ SY ( f ) = S X ( f − f c ) + S X ( f + f c ). (5.47)
Therefore
S Z ( f ) = SY ( f )|H ( f )|2
= [S X ( f − f c ) + S X ( f + f c )]|H ( f )|2 . (5.48)
we get
1
R X (τ ) = R M (τ ) cos(2π f c τ )
2
1
⇒ S X ( f ) = [S M ( f − f c ) + S M ( f + f c )] . (5.50)
4
Substituting the above value of S X ( f ) into (5.48) we get
1
SZ ( f ) = [S M ( f − 2 f c ) + S M ( f ) + S M ( f ) + S M ( f + 2 f c )] |H ( f )|2
4
1
= S M ( f ). (5.51)
2
The reason for the difference in the psd using the two procedures is that in the
first case we are doing coherent demodulation.
However, in the second case coherent demodulation is not assumed. In fact, the
local oscillator supplying any arbitrary phase α would have given the result in
(5.51).
7. Consider the communication system shown in Fig. 5.5. The message is assumed
to be a random process X (t) given by
∞
X (t) = Sk p(t − kT − α), (5.52)
k=−∞
304 5 Noise in Analog Modulation
W (t)
−3 −1 1 3
The symbols Sk are drawn from a 4-ary constellation as indicated in Fig. 5.5.
The symbols are independent, that is
Also assume that Sk and α are independent. The term W (t) denotes an additive
white noise process with psd N0 /2. The LPF is ideal with unity gain in the
bandwidth [−W, W ]. Assume that the LPF does not distort the message.
(a) Derive the expression for the autocorrelation and the psd of X (t).
(b) Compute the signal-to-noise ratio at the LPF output.
(c) What should be the value of W so that the SNR at the LPF output is maximized
without distorting the message?
R X (τ ) = E[X (t)X (t − τ )]
⎡ ⎤
∞ ∞
= E⎣ Si p(t − i T − α) S j p(t − τ − j T − α)⎦
i=−∞ j=−∞
∞
∞
= E[Si S j ]E[ p(t − i T − α) p(t − τ − j T − α)].
i=−∞ j=−∞
(5.56)
5 Noise in Analog Modulation 305
Now
where
1 for i = j
δ K (i − j) = (5.58)
0 for i = j
Similarly
Let
t − iT − α = z (5.60)
Thus
∞ ∞
5
R X (τ ) = δ K (i − j)
T i=−∞ j=−∞
t−i T
p(z) p(z + i T − τ − j T ) dz
z=t−i T −T
∞ t−i T
5
= p(z) p(z − τ ) dz
T i=−∞ z=t−i T −T
∞
5
= p(z) p(z − τ ) dz
T z=−∞
5
= R pp (τ ), (5.61)
T
where R pp (τ ) is the autocorrelation of p(t). The power spectral density of X (t)
is given by
5
SX ( f ) = |P( f )|2
T
= 5T rect ( f T )
5 sinc (t/T ). (5.62)
306 5 Noise in Analog Modulation
f1 = fc − W
f2 = fc + W
k 1
(a)
−f2 −f1 0 f1 f2 −W 0 W
Ac cos(2πfc t + θ)
SM (f ) (watts/kHz)
(b) 5
f (kHz)
−4 0 4
Therefore
N0
PN = × 2W = N0 W. (5.65)
2
Therefore
5
SNR O = . (5.66)
N0 W
8. Consider the coherent DSB-SC receiver shown in Fig. 5.6a. The signal R(t) is
given by
5 Noise in Analog Modulation 307
where M(t) has the psd as shown in Fig. 5.6b, θ is a uniformly distributed random
variable in [0, 2π ), and W (t) is a zero-mean random process with psd
A2 P
E S 2 (t) = A2c E M 2 (t) E cos2 (2π f c t + θ ) = c = 160, (5.70)
2
where
4
P = E M (t) = 2
S M ( f ) d f = 20 W. (5.71)
f =−4
Therefore
Ac = 4. (5.72)
S N ( f ) = ak 2 f 2 for f c − W ≤ | f | ≤ f c + W. (5.74)
k A2c M(t) Ac Ac
V (t) = + Nc (t) cos(θ ) + Ns (t) sin(θ ). (5.75)
2 2 2
308 5 Noise in Analog Modulation
filter
2 cos(2πfc t + Θ(t))
A2c 2 2 A2
E Nc (t) E cos (θ ) + E Ns2 (t) E sin2 (θ ) = c E N 2 (t) .
4 4
(5.77)
Now
fc +W
E N 2 (t) = 2ak 2 f2df
f = f c −W
4ak 2
2
= 3 fc W + W 3 . (5.78)
3
Therefore, the noise power at the receiver output becomes
ak 2 A2c 2
3 f c W + W 3 = 4.4373k 2 . (5.79)
3
The SNR at the receiver output in dB is
The psd of M(t) extends over [−W, W ] with power P. The LPF is ideal with
unity gain in [−W, W ].
We now use the following relations (assuming |(t)| 1 for all t):
4 (t)
(1 − cos((t)))2 ≈
4
E Nc2 (t) = 2N0 W
E Ns2 (t) = 2N0 W
E M 2 (t) = P. (5.86)
Thus
E (Y0 (t) − Y (t))2
A2 P
= c E 4 (t) + 2N0 W E cos2 ((t)) + sin2 ((t))
4
A2c Pδ 4
= + 2N0 W, (5.87)
20
where we have used the fact that
310 5 Noise in Analog Modulation
δ
4 1
E (t) = α 4 dα
2δ α=−δ
δ4
= . (5.88)
5
10. (Haykin 1983) Consider a phase modulation (PM) system, with the received
signal at the output of the IF filter given by
where
Assume that the carrier-to-noise ratio of x(t) to be high, the message power to
be P, the message bandwidth to extend over [−W, W ], and the transmission
bandwidth of the PM signal to be BT . The psd of n(t) is equal to N0 /2 for
f c − BT /2 ≤ | f | ≤ f c + BT /2 and zero elsewhere.
(a) Find the output SNR. Show all the steps.
(b) Determine the figure-of-merit of the system.
(c) If the PM system uses a pair of pre-emphasis and de-emphasis filters defined
by
Then the received signal x(t) can be written as (see Fig. 5.8):
θ(t)
5 Noise in Analog Modulation 311
where
r (t) sin(ψ(t) − φ(t))
θ (t) = φ(t) + tan−1 . (5.94)
Ac + r (t) cos(ψ(t) − φ(t))
The phase detector consists of the cascade of a hard limiter, bandpass filter,
frequency discriminator, and an integrator. The hard limiter and bandpass filter
remove envelope variations in x(t) to yield
assuming that
x3 (t) = θ (t)
r (t) sin(ψ(t) − φ(t))
≈ φ(t) +
Ac
= φ(t) + n s (t)/Ac , (5.98)
The noise power at the output of the postdetection (baseband) lowpass filter is
2N0 W/A2c . The output SNR is
k 2p P A2c
SNR O = . (5.100)
2N0 W
The average power of the PM signal is A2c /2 and the average noise power in the
message bandwidth is N0 W . Therefore, the channel signal-to-noise ratio is
A2c
SNRC = . (5.101)
2N0 W
312 5 Noise in Analog Modulation
SNR O
= k 2p P. (5.102)
SNRC
When pre-emphasis and de-emphasis is used, the noise psd at the output of the
de-emphasis filter is
N0
S No ( f ) = | f | < W. (5.103)
A2c (1 + ( f / f 0 )2 )
W/ f 0
D= . (5.105)
tan−1 (W/ f 0 )
11. (Haykin 1983) Suppose that the transfer functions of the pre-emphasis and de-
emphasis filters of an FM system are scaled as follows:
jf
H pe ( f ) = k 1 +
f0
1 1
Hde ( f ) = . (5.106)
k 1 + j f / f0
The scaling factor k is chosen so that the average power of the emphasized signal
is the same as the original message M(t).
(a) Find the value of k that satisfies this requirement for the case when the psd
of the message is
1/ 1 + ( f / f 0 )2 for − W ≤ f ≤ W
SM ( f ) = (5.107)
0 otherwise.
(b) What is the corresponding value of the improvement factor obtained by using
this pair of pre-emphasis and de-emphasis filters.
• Solution: The message power is
5 Noise in Analog Modulation 313
W
P= SM ( f ) d f
f =−W
−1 W
= 2 f 0 tan . (5.108)
f0
k 2 (W/ f 0 )3
D=
3[(W/ f 0 ) − tan−1 (W/ f 0 )]
(W/ f 0 )2 tan−1 (W/ f 0 )
= . (5.112)
3[(W/ f 0 ) − tan−1 (W/ f 0 )]
n(t)
Assume that
(a) The bandwidth of the LPF is large enough so as to reject only those compo-
nents centered at 2 f c .
(b) The channel signal-to-noise ratio at the receiver input is high.
(c) The psd of n(t) is flat with a height of N0 /2 in the range f c − W ≤ | f | ≤
fc + W .
Compute the output SNR.
A2c μ2
P= . (5.118)
4
The noise power is
2N0 W
PN = . (5.119)
2
The output SNR is
A2c μ2
SNR O = . (5.120)
4N0 W
where A is a constant and W (t) is a zero-mean WSS random process with psd
N0 /2. The signal X (t) is passed through a first-order RC-lowpass filter. Find
the expression for the output SNR, with the dc component at the LPF output
regarded as the signal of interest.
• Solution: The signal component at the output of the lowpass filter is A. Assum-
ing that 1/ f 0 = 2π RC, the noise power at the LPF output is
∞
N0 1 N0
PN = df = . (5.122)
2 f =−∞ 1 + ( f / f0 ) 2 4RC
4RC A2
SNR O = . (5.123)
N0
14. Consider the modified receiver for detecting DSB-SC signals, as illustrated in
Fig. 5.10. Note that in this receiver configuration, the IF filter is absent, hence
w(t) is zero-mean AWGN with psd N0 /2. The DSB-SC signal s(t) is given by
cos(2πfc t + θ)
316 5 Noise in Analog Modulation
N0 N0 W
PN = 2W = . (5.126)
4 2
The signal component at the LPF output is
A2c P
SNR O = . (5.129)
2N0 W
A2 P
E s 2 (t) = c . (5.130)
2
The average noise power in the message bandwidth for baseband transmis-
sion is
N0
PN1 = (2W ) = N0 W. (5.131)
2
Thus, the channel SNR is
5 Noise in Analog Modulation 317
A2c P
SNRC = . (5.132)
2N0 W
f IF (= 455 kHz)
Q≈
bandwidth of message ( = 10 kHz)
= 45.5, (5.134)
which is reasonable.
Note that the IF filter cannot reject image stations. The image stations are
rejected by the RF bandpass filter. The frequency spacing between the image
stations is 2 f IF , which is quite large, hence the Q-factor requirement of the RF
bandpass filter is not very high.
15. Consider an FM demodulator in the presence of noise. The input signal to the
demodulator is given by
t
X (t) = Ac cos 2π f c t + 2π k f M(τ ) dτ + N (t) (5.135)
τ =0
where N (t) denotes narrowband noise process with psd as illustrated in Fig. 5.11,
and BT is the bandwidth of the FM signal. The psd of the message is
a f 2 for | f | < W
SM ( f ) = (5.136)
0 otherwise.
(a) Write down the expression for the signal at the output of the FM discriminator
(cascade of a differentiator, envelope detector, dc blocking capacitor, and a
gain of 1/(2π )). No derivation is required.
(b) Compute the SNR at the output of the FM demodulator.
318 5 Noise in Analog Modulation
SN (f )
N0 /2
−fc 0 fc
−fc − BT /2 fc + BT /2
−fc + BT /2 fc − BT /2
where Ns (t) has the same statistical properties as Ns (t). Here Ns (t) denotes
the quadrature component of N (t), as given by
Bandpass Lowpass
r(t) mo (t) + no (t)
filter filter
H1 (f ) H2 (f )
2 cos(2πfc t)
2a A2c k 2f
SNR O = . (5.143)
N0
16. (Haykin 1983) Consider the AM receiver shown in Fig. 5.12 where
where w(t) has zero mean and power spectral density N0 /2 and m(t) is WSS
with psd S M ( f ). Note that H1 ( f ) and H2 ( f ) in Fig. 5.12 are nonideal filters,
h 1 (t) and h 2 (t) are real-valued, and
h 1 (t) = 2 h̃ 1 (t) e j 2π fc t . (5.145)
Note that m o (t), n o (t) and m(t) are real-valued. All symbols have their usual
meaning.
320 5 Noise in Analog Modulation
where
S0
SM ( f ) = | f | < ∞, f c
f 0 (5.150)
1 + ( f / f 0 )2
and
1 for | f | ≤ B, f c
B
H( f ) = (5.151)
0 otherwise
find B such that E is minimized. Ignore the effects of aliasing due to the 2 f c
component during demodulation.
(b)
(c)
S0
SM ( f ) = | f | < ∞, f c
f 0 . (5.154)
1 + ( f / f 0 )2
(d)
1 for | f | ≤ B, f c
B
H( f ) = (5.155)
0 otherwise.
where H̃1 ( f ) denotes the Fourier transform of the complex envelope, h̃ 1 (t).
Substituting (5.156) in (5.153) we obtain
that is, h̃ 1 (t) is real-valued. The signal component at the bandpass filter output
is
y(t) = ỹ(t) e j 2π fc t , (5.158)
where
where
Let us now turn our attention to the noise component. The noise psd at the
bandpass filter output is
322 5 Noise in Analog Modulation
N0
SN ( f ) = |H1 ( f )|2 . (5.165)
2
The narrowband noise at the bandpass filter output is given by
where we have used (5.165), (5.156), and (5.157). Therefore (5.168) becomes
2
S No ( f ) = N0 |H2 ( f )|2 H̃1 ( f )
= N0 |H ( f )|2 , (5.170)
where we have assumed that the noise and message are independent, and that
the noise has zero mean. Simplifying (5.171) we obtain
5 Noise in Analog Modulation 323
∞
E = S Mo ( f ) + S N o ( f ) + S M ( f )
f =−∞
− S M ( f )H ( f ) − S M ( f )H ∗ ( f ) d f
∞
= |1 − H ( f )|2 S M ( f ) + N0 |H ( f )|2 d f, (5.172)
f =−∞
3.2
2.8
2.6
2.4
2.2
E
1.8
1.6
1.4
1.2
0 1 2 3 4 5 6 7 8 9 10
B (Hz)
W (t) C
dE 1
= −2S0 + 2N0
dB 1 + (B/ f 0 )2
=0
S0
⇒ B = f0 − 1. (5.176)
N0
17. Consider the block diagram in Fig. 5.14. Here S(t) is a narrowband FM signal
given by
S(t) = Ac cos(2π f c t + θ )
β Ac
− cos(2π f 1 t + θ − α)
2
β Ac
+ cos(2π f 2 t + θ + α), (5.178)
2
where
f1 = fc − fm
f2 = fc + fm . (5.179)
R S (τ ) = E [S(t)S(t − τ )]
A2
= c cos(2π f c τ )
2
β 2 A2c
+ cos(2π f 1 τ )
8
β 2 A2c
+ cos(2π f 2 τ ). (5.180)
8
The psd of S(t) is the Fourier transform of R S (τ ) in (5.180), and is given by
A2c
SS ( f ) = [δ( f − f c ) + δ( f + f c )]
4
β 2 A2c
+ [δ( f − f 1 ) + δ( f + f 1 )]
16
β 2 A2c
+ [δ( f − f 2 ) + δ( f + f 2 )] . (5.181)
16
The psd of S1 (t) is
A2c
SS1 ( f ) = |H ( f c )|2 [δ( f − f c ) + δ( f + f c )]
4
β 2 A2c
+ |H ( f 1 )|2 [δ( f − f 1 ) + δ( f + f 1 )]
16
β 2 A2c
+ |H ( f 2 )|2 [δ( f − f 2 ) + δ( f + f 2 )] , (5.182)
16
where
1
|H ( f )|2 =
1 + (2π f RC)2
⇒ |H ( f )|2 = |H (− f )|2 . (5.183)
326 5 Noise in Analog Modulation
Reference
12σ 2
SNR Q = , (6.1)
2
where the step-size is given by
2xmax
= . (6.2)
2n
Here xmax = 5σ is the maximum input the quantizer can handle and n is the
number of bits used to encode any representation level. Substituting for ,
the SNR Q becomes
12 × 22n
SNR Q = = 104
100
⇒ n = 8.17. (6.3)
Since n has to be an integer and SNR Q must be greater than 40 dB, we take
n = 9.
Now, the number of samples obtained in 10 s is 8 × 104 . The number of bits
obtained is 9 × 8 × 104 , which is the storage capacity required for the speech
signal.
© The Editor(s) (if applicable) and The Author(s), under exclusive license 327
to Springer Nature Switzerland AG 2021
K. Vasudevan, Analog Communications,
https://doi.org/10.1007/978-3-030-50337-6_6
328 6 Pulse Code Modulation
56
= 8 MHz. (6.4)
7
Hence the maximum message bandwidth is 4 MHz.
For a sinusoidal signal of amplitude A, the power is A2 /2. Since the sinusoid
is stated to be at full-load, xmax = A. The step-size is
2A
= . (6.5)
27
The mean-square quantization error is
2
σ 2Q = . (6.6)
12
Hence the SNR Q is
6 × 1282
SNR Q = ≡ 43.91 dB. (6.7)
4
3. Twenty-four voice signals are sampled uniformly and then time division multi-
plexed (TDM). The sampling operation uses flat-top samples with 1 µs duration.
The multiplexing operation includes provision for synchronization by adding
an extra pulse also of 1 µs duration. The highest frequency component of each
voice signal is 3.4 kHz.
(a) Assuming a sampling-rate of 8 kHz and uniform spacing between pulses,
calculate the spacing (the time gap between the ending of a pulse and the
starting of the next pulse) between successive pulses of the multiplexed
signal.
(b) Repeat your calculation using Nyquist-rate sampling.
• Solution: Since the sampling-rate is 8 kHz, the time interval between two con-
secutive samples of the same message is 106 /8000 = 125 µs. Thus, we can
visualize a “frame” of duration 125 µs containing samples of the 24 voice
signals plus the extra synchronization pulse. Thus the spacing between the
starting points of 2 consecutive pulses is 125/25 = 5 µs. Since the pulse-
width is 1 µs, the spacing between consecutive pulses is 4 µs.
6 Pulse Code Modulation 329
1 0 1 1 0 0 0 1
A
0 T
h(t)
A
0 1
t
Constellation
T
330 6 Pulse Code Modulation
∞
X (t) = Sk δ(t − kT − α), (6.8)
k=−∞
where m S denotes the mean value of the symbols in the constellation and is
equal to
M
mS = P(Si )Si , (6.10)
i=1
where Si denotes the ith symbol in the constellation and P(Si ) denotes the
probability of occurrence of Si , and M is the number of symbols in the con-
stellation. In the given problem M = 2, P(Si ) = 0.5, and S1 = 0 and S2 = 1,
hence m S = 0.5.
The output Y (t) is given by
∞
Y (t) = Sk h(t − kT − α), (6.11)
k=−∞
where h(t) is the impulse response of the transmit filter. We know that the
psd of Y (t) is given by
∞
P − m 2S m2
SY ( f ) = |H ( f )|2 + 2S |H (n/T )|2 δ( f − n/T ), (6.12)
T T n=−∞
M
P= P(Si )Si2 = 0.5 (6.13)
i=1
H ( f ) = e−j π f T AT sinc ( f T )
⇒ |H ( f )|2 = A2 T 2 sinc2 ( f T ). (6.14)
Therefore
6 Pulse Code Modulation 331
∞
A2 T 2 sinc2 ( f T ) A2
SY ( f ) = + sinc2 (nT /T )δ( f − n/T ). (6.15)
4T 4 k=−∞
Since
1 for f = 0
sinc( f T ) = (6.16)
0 for f = n/T , n = 0
and δ( f − n/T ) is not defined for f = n/T , it is assumed that the product
sinc(n)δ( f − n/T ) is zero for n = 0. The reason is as follows. Consider an
analogy in the time-domain. If aδ(t) is input to a filter with impulse response
h(t), the output is ah(t). This implies that if a = 0, then the output is also
zero, which in turn is equivalent to zero input. Thus we conclude that 0δ( f )
is equivalent to zero.
Thus
A2 T sinc2 ( f T ) A2
SY ( f ) = + δ( f ). (6.17)
4 4
6. Consider the 15-segment piecewise linear characteristic used for μ-law com-
panding (μ = 255), shown in Table 6.1. Assume that the input signal lies in the
range [−8159, 8159] mV. Compute the representation level corresponding to an
input of 500 mV. Note that in Table 6.1 the representation levels are not properly
scaled, hence c(xmax ) = xmax .
• Solution: From Table 6.1 we see that 500 mV lies in segment 4a. The first
representation level corresponding to segment 4a is 64. The step-size for 4a is
32. Hence the end point of the first uniform segment in 4a is 479 + 32 = 511.
Thus the representation level is 64.
332 6 Pulse Code Modulation
0.50
0.25
x
3α α
7α 15α = 1
ln(1 + μx)
c(x) = for 0 ≤ x ≤ 1. (6.18)
ln(1 + μ)
The projections of all segments along the y-axis are spaced uniformly. For the
segments in the first quadrant, the projections of the segments along the x-axis
are such that the length of the projection is double that of the previous segment.
(a) Compute μ.
(b) Let e(x) denote the difference between the values obtained by the μ-law
and that obtained by the 8-segment approximation. Determine e(x) for the
second segment in the first quadrant.
• Solution: Let the projection of the first segment in the first quadrant along
the x-axis be denoted by α. Then the projection of the second segment in the
first quadrant along the x-axis is 2α and so on. The projection of each of the
segments along the y-axis is of length 1/4 since c(1) = 1. This is illustrated
in Fig. 6.3. Thus
α + 2α + 4α + 8α = 1
⇒ α = 1/15. (6.19)
Also
ln(1 + μα)
= 1/4
ln(1 + μ)
ln(1 + μ3α)
= 1/2
ln(1 + μ)
⇒μ= 1/α
= 15. (6.20)
6 Pulse Code Modulation 333
−3 −1 1 0 3
Hence
The autocorrelation of the input is given by: R X (0) = 1, R X (1) = 0.8, R X (2) =
0.6.
Compute the optimum forward prediction coefficients and the prediction gain.
σ 2X 1
= 2 = 2.8125. (6.26)
σ 2E σE
9. Consider the message pdf shown in Fig. 6.4. The decision thresholds of a non-
uniform quantizer are at 0, ±1, and ±3.
Compute SNR Q at the output of the expander.
• Solution: Since the area under the pdf must be unity, we must have A = 1/3.
Moreover
334 6 Pulse Code Modulation
−3 0 3
The representation levels at the output of the expander are at ±0.5 and ±2.
Here f X (x) cannot be considered a constant in any decision region, therefore
1 3
σ 2Q =2 (x − 1/2) f X (x) d x + 2
2
(x − 2)2 f X (x) d x
x=0 x=1
1
−x 4 4x 3 13x 2 x
=2 + − +
36 27 72 12
x=0
4 3
−x 7x 3 16x 2 4x
+2 + − +
36 27 18 3 x=1
= 0.19444. (6.28)
• Solution: Since the area under the pdf is unity, we must have A = 1/3. Due to
symmetry of the pdf, one of the decision thresholds is x1 = 0. The other two
decision thresholds are x0 = −3 and x2 = 3. The corresponding representa-
tion levels are y1 and y2 which are related by y1 = −y2 . Thus the variance of
the quantization error is
3
σ 2Q = 2 (x − y2 )2 f X (x) d x, (6.29)
x=0
t = kT
R
w(t) x(t) xk 2nd -order x̂k
prediction filter
C
+ −
ek
3 3
x f X (x) d x = x(−x/9 + 1/3) d x
x=0 x=0
3
= (−x 3 /27 + x 2 /6)x=0
= 1/2. (6.31)
Thus y2 = 1.
The minimum variance of the quantization error is
3
σ 2Q, min =2 (x − 1)2 f X (x) d x
x=0
3
=2 (x − 1)2 (−x/9 + 1/3) d x
x=0
= 0.5. (6.33)
11. Consider the block diagram in Fig. 6.6. Assume that w(t) is zero-mean AWGN
with psd N0 /2 = 0.5 × 10−4 W/Hz. Assume that RC = 10−4 s and T = RC/4 s.
Let
(a) Compute p1 and p2 so that the prediction error variance, E[ek2 ], is mini-
mized.
(b) What is the minimum value of the prediction error variance?
N0
SX ( f ) = |H ( f )|2
2
N0 /2
. (6.35)
1 + (2π f RC)2
Now
1 1 −t/RC
H( f ) = e u(t). (6.36)
1 + j 2π f RC RC
N0
E[x(t)x(t − τ )] = R X (τ ) = (h(τ ) h(−τ ))
2
N0 ∞
= h(t)h(t − τ ) dt
2 t=−∞
∞
N0
= e−t/RC u(t)
2(RC)2 t=−∞
× e(τ −t)/RC u(t − τ ) dt. (6.38)
When τ > 0
∞
N0
E[x(t)x(t − τ )] = e−t/RC e(τ −t)/RC dt
2(RC)2 t=τ
N0 τ
= exp − . (6.39)
4RC RC
When τ < 0
∞
N0
E[x(t)x(t − τ )] = e−t/RC e(τ −t)/RC dt
2(RC)2 t=0
N0 τ
= exp . (6.40)
4RC RC
Thus R X (τ ) is equal to
N0 |τ |
E[x(t)x(t − τ )] = R X (τ ) = exp − . (6.41)
4RC RC
R X (0) = 0.25
R X (1) = 0.25 e−1/4 = 0.1947
R X (2) = 0.25 e−1/2 = 0.1516. (6.43)
We know that
R X (0) R X (1) p1 R X (1)
= . (6.44)
R X (1) R X (0) p2 R X (2)
12. A delta modulator is designed to operate on speech signals limited to 3.4 kHz.
The specifications of the modulator are
(a) Sampling-rate is ten times the Nyquist-rate of the speech signal.
(b) Step-size = 100 mV.
The modulator is tested with a 1 kHz sinusoidal signal. Determine the max-
imum amplitude of this test signal required to avoid slope overload.
• Solution
0.1 f s
Am <
2π f c
0.1 × 68
=
2π
= 1.08 V. (6.46)
(a) Compute the representation levels of the optimum 2-level Lloyd-Max quan-
tizer.
(b) What is the minimum value of σ 2Q ?
338 6 Pulse Code Modulation
• Solution: Since the area under the pdf is unity, we must have
∞
2a e−x d x = 1
x=0
⇒ a = 1/2. (6.48)
Thus y2 = 1.
The minimum variance of the quantization error is
∞
σ 2Q, min = 2 (x − 1)2 f X (x) d x
x=0
= 1. (6.53)
14. A message signal having a pdf shown in Fig. 6.7 is applied to a μ-law compressor.
The compressor characteristic is given by
xmax
c(|x|) = ln (1 + μ|x|/xmax ) for 0 ≤ |x| ≤ xmax . (6.54)
ln(1 + μ)
Derive the expression for the signal-to-quantization noise ratio at the expander
output.
6 Pulse Code Modulation 339
−xmax 0 xmax
Assume that the number of representation levels is L and the overload level is
xmax .
3L 2 σ 2X
SNR Q = xmax . (6.55)
−2 d x
x=−xmax f X (x)(dc/d x)
2
xmax
Since
B ln (1 + μx/xmax ) for 0 ≤ x ≤ xmax
c(x) = (6.57)
−B ln (1 − μx/xmax ) for −xmax ≤ x ≤ 0,
where
xmax
B= (6.58)
ln(1 + μ)
we have
dc μ 1
= for 0 ≤ |x| ≤ xmax . (6.59)
dx ln(1 + μ) 1 + μ|x|/xmax
where
2
ln(1 + μ)
2
K =
xmax μ
C = K xmax . (6.62)
L2
SNR Q = . (6.64)
2I
15. A message signal having a pdf shown in Fig. 6.8 is applied to a uniform mid-step
quantizer also shown in the same figure. Compute a and the pdf of the quantizer
output Y . Assume that the overload level of the quantizer is ±3.
Output (Y )
Input (X)
−3 0 3
2xmax 6
= = = 2. (6.66)
L 3
The various quantizer levels are indicated in Fig. 6.9. Thus
Therefore
342 6 Pulse Code Modulation
Output (Y )
Input (X)
−2
−3 −1 0 1 3
a/4
x
−10 −2 0 2 10
13 1 13
f Y (y) = δ(y + 2) + δ(y) + δ(y − 2). (6.68)
27 27 27
16. The probability density function f X (x) of a message signal is given in Fig. 6.10.
The message is quantized by a 3-bit quantizer such that all the eight reconstruc-
tion levels occur with equal probability.
(a) Determine a.
(b) Determine the power of the message signal.
(c) Compute the decision thresholds.
(d) For each partition cell, compute the reconstruction level such that the quan-
tization noise power is minimized.
(e) Determine the overall quantization noise power.
• Solution: Since the area under the pdf is unity we must have
6 Pulse Code Modulation 343
a
2 2a + 8 =1
4
1
⇒a= . (6.69)
8
The power in the message is
10
2 x 2 f X (x) d x = 21.33. (6.70)
x=0
Since the message pdf is symmetric, we expect the decision thresholds also
to be symmetric about the origin. Let us denote the decision thresholds on
the positive x-axis as x0 , x1 , x2 , x3 , and x4 . The decision thresholds along the
negative x-axis are −x1 , −x2 , −x3 , and −x4 . Note that x4 = 10 and x0 = 0.
Let us first compute x3 . Since each of the reconstruction levels occur with
probability equal to 1/8, we have
4
Q=2 Qi . (6.73)
i=1
−16 0 16
which is the centroid of each partition cell. For the given problem f X (x) is
constant for each partition cell, therefore
xi−1 + xi
yi = . (6.75)
2
Therefore the reconstruction levels are at y1 = 1/2, y2 = 3/2, y3 = 4, and
y4 = 8.
The overall quantization noise power is
Q = 2[Q 1 + Q 2 + Q 3 + Q 4 ]
= 2[2/96 + 2/6]
= 17/24
= 0.70833. (6.76)
17. The probability density function f X (x) of a message signal is given in Fig. 6.11.
The message is quantized by a 3-bit quantizer such that all the eight reconstruc-
tion levels occur with equal probability.
(a) Determine a.
(b) Determine the power of the message signal.
(c) Compute the decision thresholds.
(d) For each partition cell, compute the reconstruction level such that the quan-
tization noise power is minimized.
(e) Determine the overall quantization noise power.
• Solution: Consider Fig. 6.12. Since the area under the pdf is unity we must
have
1
2× × 16a = 1
2
1
⇒a= . (6.77)
16
The power in the message is
6 Pulse Code Modulation 345
−16 0 x1 x3 16
x2
x0 x4
16 16
1 x
2 x f X (x) d x = 2
2
x 2
− dx
x=0 x=0 16 256
128
= . (6.78)
3
Since the message pdf is symmetric, we expect the decision thresholds also
to be symmetric about the origin. Let us denote the decision thresholds on
the positive x-axis as x0 , x1 , x2 , x3 , and x4 . The decision thresholds along the
negative x-axis are −x1 , −x2 , −x3 , and −x4 . Note that x4 = 16 and x0 = 0.
Let us first compute x3 . Since each of the reconstruction levels occur with
probability equal to 1/8, we have
Similarly we get
P(x2 ≤ x ≤ x3 ) = 1/8
1 1 1 x2
⇒ + − (x3 − x2 ) = 1/8
2 32 16 256
⇒ x2 = 4.6862915 (6.80)
and
P(0 ≤ x ≤ x1 ) = 1/8
1 1 1 x1
⇒ + − x1 = 1/8
2 16 16 256
⇒ x1 = 2.1435935. (6.81)
axis are −y1 , −y2 , −y3 , and −y4 . The quantization noise power for the ith
partition cell is
xi
Qi = (x − yi )2 f X (x) d x for 1 ≤ i ≤ 4. (6.82)
x=xi−1
4
Q=2 Qi . (6.83)
i=1
which is the centroid of each partition cell. Therefore the reconstruction levels
are at
y1 = 1.0461463
y2 = 3.3721317
y3 = 6.2483887
y4 = 10.666667. (6.85)
which evaluates to
6 Pulse Code Modulation 347
Q 1 = 0.0477823
Q 2 = 0.0671179
Q 3 = 0.1132596
Q 4 = 0.4444444. (6.87)
Therefore
Q = 2(Q 1 + Q 2 + Q 3 + Q 4 )
= 1.3452084. (6.88)
18. A speech signal has a total duration of 20 s. It is sampled at a rate of 8 kHz and
then encoded. The SNR Q must be greater than 50 dB. Calculate the minimum
storage capacity required to accommodate this digitized speech signal. Assume
that the speech signal has a Gaussian pdf with zero mean and variance σ 2 and
the overload factor is 6. Assume that the quantizer is uniform and of the mid-rise
type.
• Solution: We know that
12σ 2
SNR Q = , (6.89)
2
where the step-size is given by
2xmax
= . (6.90)
2n
Here, xmax = 6σ is the maximum input the quantizer can handle and n is the
number of bits used to encode any representation level. Substituting for ,
the SNR Q becomes
12 × 22n
SNR Q = = 105
144
⇒ n = 10.097. (6.91)
Since n has to be an integer and SNR Q must be greater than 50 dB, we take
n = 11.
Now, the number of samples obtained in 20 s is 16 × 104 . The number of
bits obtained is 11 × 16 × 104 = 176 × 104 , which is the storage capacity
required for the speech signal.
19. A DPCM system uses a second-order predictor of the form
The autocorrelation of the input is given by: R X (0) = 2, R X (1) = 1.8, R X (2) =
1.6.
Compute the optimum prediction coefficients and the prediction gain.
• Solution: We know that
R X (0) R X (1) p1 R X (1)
= . (6.93)
R X (1) R X (0) p2 R X (2)
σ 2X 2
= 2 = 5.28. (6.95)
σ 2E σE
• Solution: Clearly
E[X (t)X (t − τ )] = R X (τ )
= E [Y cos(2π f c t + θ)Y cos(2π f c (t − τ ) + θ)]
= E Y 2 E [cos(2π f c t + θ) cos(2π f c (t − τ ) + θ)]
3
= cos(2π f 0 τ ). (6.98)
2
The psd is
6 Pulse Code Modulation 349
3
SX ( f ) = [δ( f − f 0 ) + δ( f + f 0 )] . (6.99)
4
The signal power is 3/2. The quantization noise power is
2
σ 2Q = , (6.100)
12
where
2m max 6
= = n, (6.101)
2n 2
where m max = 3 denotes the maximum amplitude of the random process and
n is the number of bits per sample.
Therefore the SQNR is
3 22n
SQNR =×
2 3
= 22n−1
⇒ 10 log10 (22n−1 ) ≥ 40
⇒ n ≥ 7.14
⇒ n = 8. (6.102)
21. The pdf of a full-load message signal is uniformly distributed in [−6, 6]. Com-
pute the SNR Q in dB, due to a 2-bit, uniform mid-rise quantizer.
• Solution: The signal power is
6
σ 2X = x 2 f X (x) d x
x=−6
6
2
= x2 dx
12 x=0
= 12. (6.103)
2xmax
=
4
2×6
=
4
= 3. (6.104)
4
xi
σ 2Q = (x − yi )2 f X (x) d x
i=1 x=xi−1
3 6
2 2
= (x − 3/2) d x +
2
(x − 9/2)2 d x
12 x=0 12 x=3
= 3/4. (6.105)
Hence we get
σ 2X
SNR Q =
σ 2Q
= 16
⇒ SNR Q (in dB) = 10 log10 (16)
= 12.0412 dB. (6.106)
u
Gc = , (6.108)
r, min
where u is the step-size of the uniform quantizer and r, min is the minimum
step-size of the robust quantizer having the same overload level (xmax ) and
representation levels (L). Now
2xmax
u = . (6.109)
L
Similarly, for the kth segment of the robust quantizer we have
dc(x) 2xmax
= , (6.110)
dx k Lk
where k is the length of the kth segment along the x-axis (input). Note that
the length of each segment along the y-axis is identical and equal to 2xmax /L.
From (6.110) it is clear that k is minimum when dc/d x is maximum, which
6 Pulse Code Modulation 351
for μ = 255.
23. The A-law compressor characteristic is given by
c(|x|) A|x|/(K xmax ) for 0 ≤ |x|/xmax ≤ 1/A
= (6.112)
xmax (1 + ln(A|x|/xmax ))/K for 1/A ≤ |x|/xmax ≤ 1,
where K = 1 + ln(A). Derive and compute the companding gain for A = 87.56.
u
Gc = , (6.113)
r, min
where u is the step-size of the uniform quantizer and r, min is the minimum
step-size of the robust quantizer having the same overload level (xmax ) and
representation levels (L). Now
2xmax
u = . (6.114)
L
Similarly, for the kth segment of the robust quantizer we have
dc(x) 2xmax
= , (6.115)
d x k Lk
where k is the length of the kth segment along the x-axis (input). Note that
the length of each segment along the y-axis is identical and equal to 2xmax /L.
From (6.115) it is clear that k is minimum when dc/d x is maximum, which
occurs at the origin. Substituting (6.114) and (6.115) in (6.113) we get
dc(x)
Gc =
d x x=0
A
=
(1 + ln(A))
= 16.000514 (6.116)
for A = 87.56.
352 6 Pulse Code Modulation
3 Δk
Op Ip (x)
representation
level (yk )
Expander 12
Op
Fig. 6.13 Illustrating the transfer characteristics of the compressor and expander
24. (Haykin 1988) derived the ideal compressor characteristic. Clearly state the
objective of the compressor. Can the ideal compressor be used in practice? Give
reasons.
• Solution: Recall that in the case of a uniform quantizer, the SNR Q is directly
proportional to the input signal power. Therefore, if the input signal power
decreases, the SNR Q also decreases. However, in the case of a robust quan-
tizer, the SNR Q remains constant over a wide range of input signal power. In
order to achieve this feature, the robust quantizer uses a compressor. This is
shown in Fig. 6.13. For ease of illustration, the compressor characteristic is
assumed to be piecewise-linear. The compressor and uniform quantizer are
located at the transmitter, whereas the expander is located at the receiver.
Note that in the case of the compressor, both input and output amplitudes
are continuous. However, in the case of the expander, both input and output
amplitudes are discrete.
The representation level yk is related to the decision thresholds xk and xk+1
as follows:
6 Pulse Code Modulation 353
1
yk = (xk + xk+1 ). (6.117)
2
The quantization error for the kth decision region is defined as
qk = x − yk , (6.118)
This ensures that the quantization error is uniformly distributed in [−k /2,
k /2].
(c) The input does not overload the compressor.
Now, the variance of the quantization error in the kth interval is
L−1
σ 2Q = Pk σ 2Q, k
k=0
1
L−1
= Pk 2k , (6.124)
12 k=0
where
1
L−1 2
4xmax
σ 2Q = Pk 2 . (6.126)
12 k=0 L (dc(x)/d x)2k
P → f X (x) d x
k
dc(x) dc(x)
→ (6.127)
dx k dx
Therefore
σ 2X
SNR Q =
σ 2Q
xmax
1
= x 2 f X (x) d x (6.129)
σ 2Q x=−xmax
dc(x) K
=
dx x
⇒ c(x) = K ln(x) + C0 for x > 0, (6.130)
Note that when x < 0, (6.119) needs to be applied. The ideal c(x) is not used
in practice, since c(x) → −∞ as x → 0+ .
25. The probability density function of a message signal is
Let the decision thresholds in the 1st iteration be given by x0, 1 , . . . , x4, 1 . We
have
x0, 1 = −∞
y1, 0 + y2, 0
x1, 1 =
2
= −2.5
y2, 0 + y3, 0
x2, 1 =
2
=0
y3, 0 + y4, 0
x3, 1 =
2
= 2.5
x4, 1 = ∞. (6.135)
y1, 1 = −2.83
= −y4, 1
y2, 1 = −0.33194
= −y3, 1 . (6.137)
References
Simon Haykin. Digital Communications. John Wiley & Sons, first edition, 1988.
K. Vasudevan. Digital Communications and Signal Processing, Second edition (CDROM included).
Universities Press (India), Hyderabad, www.universitiespress.com, 2010.
Chapter 7
Signaling Through AWGN Channel
1. For the transmit filter with impulse response given in Fig. 7.1, draw the output of
the matched filter. Assume that the matched filter has an impulse response p(−t).
where ak denotes symbols from a binary constellation. The final PCM signal
y(t) is
© The Editor(s) (if applicable) and The Author(s), under exclusive license 357
to Springer Nature Switzerland AG 2021
K. Vasudevan, Analog Communications,
https://doi.org/10.1007/978-3-030-50337-6_7
358 7 Signaling Through AWGN Channel
−0.5
0 T /2 T
−T /4
−T −T /2 0 T /2 T
∞
y(t) = ak p(t − kT /4). (7.4)
k=−∞
Note that the symbol-rate of ak is 4/T , that is four times the input bit-rate.
Draw the constellation and write down the sequence of symbols that are
generated corresponding to an input bit bk = 0 and an input bit bk = 1.
(b) The received signal is given by
where y(t) is the transmitted signal and w(t) is a sample function of a zero-
mean AWGN process with psd N0 /2. Compute the mean and variance of
z 1 and z 2 , given that 1 (x1 (t)) was transmitted in the interval [0, T ]. Also
compute cov(z 1 , z 2 ).
(c) Derive the detection rule for the optimal detector.
(d) Derive the average probability of error.
7 Signaling Through AWGN Channel 359
bk
Input bits Bit y1 (t) Transmit y(t)
manipulator filter
(1s and 0s) PCM signal
from quantizer
Transmitter
x1 (t) x2 (t)
A/2 A/2
t t
−A/2 −A/2
3T
0 T
2
T 0 −
T
4 4
T
z1
x1 (−t)
Optimum
u(t) = y(t) + w(t)
detector
z2
x2 (−t)
t = kT
Receiver
Given that 1 (x1 (t)) has been transmitted in the interval [0, T ], the received
signal can be written as
z 1 = u(t) x1 (−t)|t=0
∞
= u(τ )x1 (−t + τ ) dτ
τ =−∞ t=0
T
= u(t)x1 (t) dt
t=0
2
A T
= + w1 , (7.10)
4
where
T
w1 = w(t)x1 (t) dt. (7.11)
t=0
Similarly
z 2 = u(t) x2 (−t)|t=0
T
= u(t)x2 (t) dt
t=0
= w2 , (7.12)
where we have used the fact that x1 (t) and x2 (t) are orthogonal and
T
w2 = w(t)x2 (t) dt. (7.13)
t=0
Since w(t) is zero mean, w1 and w2 are also zero mean. Since x1 (t) and x2 (t)
are LTI filters, w1 and w2 are Gaussian distributed RVs. Moreover
T T
E[w1 w2 ] = E w(t)x1 (t) dt w(τ )x2 (τ ) dτ
t=0 τ =0
T T
N0
= x1 (t)x2 (τ ) δ(t − τ ) dt dτ
t=0 τ =0 2
T
N0
= x1 (t)x2 (t) dt
t=0 2
= 0. (7.14)
7 Signaling Through AWGN Channel 361
Thus w1 and w2 are uncorrelated, and being Gaussian, they are also statistically
independent. This implies that z 1 and z 2 are also statistically independent. The
variance of w1 and w2 is
T T
E w12 = E w(t)x1 (t) dt w(τ )x1 (τ ) dτ
t=0 τ =0
T T
N0
= x12 (t) δ(t − τ ) dt dτ
t=0 τ =0 2
T
N0
= x12 (t) dt
t=0 2
N0 A2 T
=
8
= E[w22 ]
= var(z 1 ) = var(z 2 ) = σ 2 (say). (7.15)
A2 T
E[z 1 ] = = m 1, 1 (say)
4
E[z 2 ] = 0 = m 2, 1 (say), (7.16)
E[z 1 ] = 0 = m 1, 0 (say)
A2 T
E[z 2 ] = = m 2, 0 (say). (7.17)
4
Let
T
z = z1 z2 . (7.18)
and decides in favor of that bit for which the probability is maximum. Using
Bayes’ rule we have
f Z (z| j)P( j)
P( j|z) = , (7.20)
f Z (z)
362 7 Signaling Through AWGN Channel
max f Z 1 (z 1 | j) f Z 2 (z 2 | j)
j
1 (z 1 − m 1, j )2 + (z 2 − m 2, j )2
⇒ max exp − . (7.22)
j 2πσ 2 2σ 2
(z 1 − m 1, 0 )2 + (z 2 − m 2, 0 )2 < (z 1 − m 1, 1 )2 + (z 2 − m 2, 1 )2 . (7.24)
Let
e1 = m 1, 1 − m 1, 0
e2 = m 2, 1 − m 2, 0
e1 + e22 = d 2 .
2
(7.25)
Let
Then
= σ 2Z (say). (7.28)
3. Consider the passband PAM system in Fig. 7.4. The bits 1 and 0 from the quantizer
are equally likely. The signal b(t) is given by
∞
b(t) = ak δ(t − kT ), (7.30)
k=−∞
0 1
−d = A0 0 d = A1
Constellation
Input bit
Transmit
stream Bit b(t) y(t) s(t)
filter
from manipulator
quantizer p(t)
(1s and 0s)
Transmitter cos(2πfc t + θ)
t = kT
Output
Matched
s1 (t) y1 (t) r(kT ) Optimum bit
filter
stream
detector
p(−t)
cos(2πfc t + θ) Receiver
(from which the root-raised cosine spectrum is obtained) is α = 0.5. Assume that
the bit-rate 1/T = 1 kbps, the energy of p(t) = 2, θ is a uniformly distributed
random variable in [0, 2π] and θ and w(t) are statistically independent. The
received signal is given by
B = 2 × 500(1 + α)
= 1500 Hz. (7.33)
y(t)
s1 (t) cos(2π f c t) = (1 + cos(4π f c t)) + w1 (t), (7.34)
2
where w1 (t) is
∞
1
y1 (t) = ai p(t − kT ) + w1 (t). (7.37)
2 i=−∞
The MF output is
Note that g(t) is a pulse corresponding to the raised cosine spectrum, hence it
satisfies the Nyquist criterion for zero intersymbol interference (ISI). The MF
output sampled at time kT is
r k = ak + z k . (7.41)
N0 N0
E[z k2 ] = g(0) = = σ2 (say). (7.42)
4 2
At time kT given that ak = A j was transmitted, the mean value of rk is
E[rk |A j ] = A j . (7.43)
Note that
1
E[rk ] = E[rk |A j ]P(A j ) = 0. (7.44)
j=0
In other words, the unconditional mean is zero. At time kT , the MAP detector
computes the probabilities P(A j |rk ) for j = 0, 1 and decides in favor of that
symbol for which the probability is maximum. Using Bayes’ rule, the MAP
detector can be re-written as
366 7 Signaling Through AWGN Channel
f Rk |A j (rk |A j )P(A j )
max for j = 0, 1, (7.45)
j f Rk (rk )
1
max √ e−(rk −A j ) /(2σ )
2 2
for j = 0, 1. (7.47)
j σ 2π
Given that 1 was transmitted at time kT , the ML detector makes an error when
where
t = kT
w(t) z(t) z(kT )
p(−t)
−d1 /2
1
e−zk /(2σ ) dz k
2 2
P(z k < −d1 /2) = √
σ 2π zk =−∞
⎛ ⎞
2
1 d
= erfc ⎝ 1 ⎠
2 8σ 2
⎛ ⎞
1 d2 ⎠
= erfc ⎝ . (7.52)
2 N0
5. Consider the block diagram in Fig. 7.5. Here w(t) is a sample function of a zero-
mean AWGN process with psd N0 /2. Let p(t) denote the pulse corresponding to
the root-raised cosine spectrum. Let P( f ) denote the Fourier transform of p(t)
and let the energy of p(t) be 2. Assume that w(t) is WSS.
Compute E[z(kT )z(kT − mT )].
• Solution: The psd of z(t) is
N0
SZ ( f ) = |P( f )|2 , (7.53)
2
where |P( f )|2 has a raised-cosine spectrum. Note that since w(t) is WSS,
z(t) is also WSS. This implies that the autocorrelation of z(t) is
N0
E[z(t)z(t − τ )] = R Z (τ ) = g(τ ), (7.54)
2
where g(τ ) is the inverse Fourier transform of the raised cosine spectrum.
Therefore
N0
E[z(kT )z(kT − mT )] = R Z (mT ) = g(mT )
2
N0 for m = 0
= (7.55)
0 otherwise
Index
A Compressor, 338
Ac coupled, 117 Condition
Aliasing, 32 necessary, 68
Amplifier sufficient, 69
ac, 177 Conjugate
dc, 177 complex, 36
chopper stabilized, 177 symmetry, 319
Amplitude modulation (AM)
Constellation
amplitude sensitivity, 154
BPSK, 83
modulation factor, 162
Convolution, 12
overmodulation, 182
power efficiency, 162 Correlation
residual, 287 auto, 5
Attenuation, 164 coefficient, 77
cross, 15
Costas loop, 170
B phase ambiguity, 238
Bandpass signal phase discriminator, 238
canonical representation, 12 Cross covariance, 128
Bayes’ rule, 361 Cross-spectral density, 96
Bessel function, 258 Cumulative distribution function, 137
C
Capacitor, 182 D
Carrier Dc coupled, 117
frequency, 56
Dc voltmeter, 117
recovery, 208
Delta function
spacing, 193
Dirac, 1
Carrier-to-noise ratio, 299
Centroid, 344 Kronecker, 84
Characteristic function, 77 Delta modulator, 337
Chernoff bound, 95 Differentiator, 147
Communication privacy, 228 Dirichlet’s conditions, 69
Companding Discriminant, 77
A-law, 351 Distortion, 178
µ-law, 331 DPCM, 333
© The Editor(s) (if applicable) and The Author(s), under exclusive license 369
to Springer Nature Switzerland AG 2021
K. Vasudevan, Analog Communications,
https://doi.org/10.1007/978-3-030-50337-6
370 Index
R
N Radio, 169
Noise Random process, 80
AWGN, 171 cyclostationary, 87
Gaussian, 118 Random timing phase, 83
generator, 117 Random variable, 77
narrowband, 96 discrete, 83
white, 118 Gaussian, 77
Non-linear device, 153 independent, 95
Rayleigh distributed, 94
transformation of a, 78
O
uncorrelated, 95
Orthogonality, 188
uniformly distributed, 79
Oscillator, 126
Rayleigh’s energy theorem, 20
local, 295
Rectifier
full wave, 7
P half wave, 78
Parseval’s power theorem, 7 Response
Phase, 6 magnitude, 32
ambiguity, 210 phase, 32
error, 297 RLC circuit
Phasor, 179 series, 206
Poisson sum formula, 69 resonant frequency, 206
Power spectral density (psd), 96
Prediction
coefficients, 333 S
gain, 333 Sampling, 327
Probability density function (pdf), 78 Nyquist-rate, 328
conditional, 92 Schwarz’s inequality, 45
joint, 82 Scrambling, 184
Pulse amplitude modulation (PAM), 329 Sideband, 162
Pulse shaping lower, 164
Nyquist, 329 upper, 164
Signal
audio, 184
Q energy, 15
Quadrant, 189 narrowband, 56
Quadratic equation, 77 NRZ unipolar, 329
Quantizer periodic, 21
decision threshold, 333 power, 9
372 Index
speech, 327 T
digitized, 327 Time division multiplexed (TDM), 328
Transition bandwidth, 165
Signal-to-noise ratio (SNR), 172
True rms meter, 117
Slope overload, 337
SNR Q , 327
Spectrum, 10 U
RC, 363 Unit step function, 71
roll-off factor, 363
RRC, 363 V
square-law device, 22 Varactor diode, 281
Square rooter, 159 bias voltage, 282
Superheterodyne, 317 Voltage controlled oscillator (VCO), 238
Surjective mapping, 98
System W
nonlinear, 15 Wide sense stationary (WSS), 81