DRILL PROBLEMS

Problem 2.1
Evaluate the Fourier transform of the damped sinusoidal wave g ( t ) = exp ( – t ) sin ( 2π f c t )u ( t )
where u(t) is the unit step function

Solution
The Fourier transform of g(t) is

∞
G( f ) = ∫0 exp ( –t ) sin ( 2π f c t ) sin ( – j2π f c t ) dt
1 ∞
= ------ ∫ exp ( – t ) [ exp ( j2π f c t ) – exp ( – j2π f c t ) ] exp ( – j2πft ) dt
2j 0
1 ∞
= ------ ∫ [ exp ( j2π ( f c – f )t – t ) ] dt
2j 0
1 1 1 ∞
= ------ ---------------------------------------- exp ( j2π ( f c - f )t-t ) + ---------------------------------------- exp ( ( – j2π ( f c + f )t )-t )
2 j j2π ( f c – f ) – 1 j2π ( f c – f ) + 1 t=0

= ------  ---------------------------------------- + ----------------------------------------

1 1 1
2 j  j2π ( f – f ) – 1 j2π ( f – f ) + 1
c c
1  ( j2π ( f c – f ) + 1 ) + ( j2π ( f c – f ) – 1 )
= ------  ------------------------------------------------------------------------------------------------
-
2 j 2
1 + 4π ( f c – f )
2

2π f c
= -----------------------------------------
2
-
2
1 + 4π ( f – f c )

Problem 2.2
Determine the inverse Fourier transform of the frequency function G(f) defined by the amplitude
and phase spectra shown in Fig. 2.5.

|G(f)| arg[G(f)]

1.0

π
2
f f
-W 0 W -W 0 W
− π
2

IADC 1
Solution
0 jπ ⁄ 2 W – j π ⁄ 2 j2πft
∫–W e df + ∫ e
j2πft
g(t) = ⋅e e df
0
π 0 π W
j  --- + 2πft j  – --- + 2πft
1 2  1  2 
= -----------e + -----------e
j2πt j2πt
f = –W f =0
π π
j  --- -2πWt j  – --- - j2πWt
1  2  jπ ⁄ 2 1  – jπ ⁄ 2  2 
= -----------  e –e  + -----------  e –e 
j2πt   j2πt  
1 – jπ ⁄ 2 jπ ⁄ 2 1 – j2πWt jπ ⁄ 2 – j π ⁄ 2
= ----------- ( e –e ) + -----------e (e –e )
j2πt j2πt
1 1 – j2πWt 1 – j2πWt
= – ----- + ----- e = ----- ( e – 1)
πt πt πt

1
Note: If we let W → ∞ , G ( f ) → j sgn ( t ) , the inverse of which – ----
- . This result agrees with the
πt
limiting value of the solution for W = ∞ .

Problem 2.3
Suppose g(t) is real valued with a complex-valued Fourier transform G(f). Explain how the rule of
Eq. (2.31) can be satisfied by such a signal.

Solution
With G(f) being complex valued, we may express it as
G ( f ) = G r ( f ) + jG i ( f )
where Gr(f) is the real part of G(f) and Gi(f) is its imaginary part. Hence,
G ( 0 ) = G r ( 0 ) + jG i ( 0 ) .
According to Eq. (2.31) in the text,
∞
∫–∞ g ( t ) dt = G r ( 0 ) + jG i ( 0 )
With g(t) being real valued, this condition can only be satisfied if the imaginary part Gi(0) is zero.

Problem 2.4
Continuing with Problem 2.3, explain how the rule of Eq. (2.32) can be satisfied by the signal g(t)
described therein.

Solution
∞
Since g(t) is real valued, it follows that the integral ∫–∞ G ( f ) d f must likewise be real valued. For
this condition to be satisfied, the imaginary part of G(f) must be an odd function of f.

IADC 2
Problem 2.5
Develop the detailed steps that show that the modulation and convolution theorems are indeed the
dual of each other.

Solution
The modulation theorem states that
∞
g 1 ( t )g 2 ( t ) ∫–∞ G 1 ( λ ) G 2 ( f – λ ) dλ (1)

To apply the duality theorem, we say that if

g 1 ( t )g 2 ( t ) X ( f ) , then
X( f ) g 1 ( – f )g 2 ( – f )
For the problem at hand, we may therefore write
∞
∫–∞ G 1 ( λ ) G 2 ( f – λ ) dλ g 1 ( – f )g 2 ( – f ) (2)

Next, we apply Eq. (2.21), which states that if g ( t ) G ( f ) then g ( – t ) G ( – f ) . Hence,

applying this rule to Eq. (2), we may write
∞
∫–∞ G 1 ( λ ) G 2 ( λ – t ) dλ g 1 ( f )g 2 ( f )

which is a statement of the convolution theorem, with G 1 ( t ) g 1 ( f ) and G 2 ( t ) g2( f ) .

Problem 2.6
Develop the detailed steps involved in deriving Eq. (2.53), starting from Eq. (2.51).

Solution
According to Eq. (2.51),
∞
∫–∞ g 1 ( τ ) g 2 ( t – τ ) dτ G 1 ( f )G 2 ( f )

According to Eq. (2.21), if g ( t ) G ( f ) , then g ( – t ) G ( – f ) . Hence, applying this rule to the

problem at hand, we may write
∞
∫–∞ g 1 ( τ ) g 2 ( τ – t ) dτ G 1 ( f )G 2 ( – f )

Next, we note that if we complex conjugate the term g2(τ - t), then the conjugation theorem of Eq.
(2.22) teaches us that
∞
∫–∞ g 1 ( τ ) g 2 ( τ – t ) dτ
* *
G 1 ( f )G 2 ( – f )

which is the desired result, except for the fact that we have interchanged the roles of variables t
and τ.

Problem 2.7
Prove the following properties of the convolution process:

IADC 3
(a) The commutative property:
g 1 ( t )★g 2 ( t ) = g 2 ( t )★g 1 ( t )

Proof:
∞
g 1 ( t )★g 2 ( t ) = ∫–∞ g 1 ( τ ) g 2 ( t – τ ) dτ
∞
= ∫–∞ g 2 ( t – τ ) g 1 ( τ ) dτ
Replace t - τ with λ. That is, τ = t - λ. Hence
–∞
g 1 ( t )★g 2 ( t ) = – ∫ g 2 ( λ ) g 1 ( t – λ ) dλ
+∞
∞
= ∫–∞ g 2 ( λ ) g 1 ( t – λ ) dλ
= g 2 ( t )★g 1 ( t )

(b) The associative property:

g 1 ( t )★ [ g 2 ( t )★g 3 ( t ) ] = [ g 1 ( t )★g 2 ( t ) ]★g 3 ( t )

Proof:
.
Let
x ( t ) = g 2 ( t )★g 3 ( t )
∞
= ∫–∞ g 2 ( τ ) g 3 ( t – τ ) dτ
Hence
∞
I ( t ) = g 1 ( t )★ [ g 2 ( t )★g 3 ( t ) ] = ∫–∞ g 1 ( λ ) x ( t – λ ) dλ








∞ ∞
= ∫–∞ g 1 ( λ ) ∫–∞ g 2 ( τ ) g 3 ( t – τ – λ ) dτ dλ (1)

Replace τ + λ with µ; that is, τ = µ - λ. Hence, keeping λ fixed, we may write

∞ ∞
I (t ) = ∫–∞ g 1 ( λ ) ∫–∞ g 2 ( µ – λ ) g 3 ( t – µ ) dµ dλ (2)

∞
With µ fixed, the integral ∫–∞ g 1 ( λ ) g 2 ( µ – λ ) dλ is recognized as the convolution of g1(µ) and
g2(µ), as shown by
∞
g 12 ( µ ) = ∫–∞ g 1 ( λ ) g 2 ( µ – λ ) dλ = g 1 ( µ ) ★g 2 ( µ )

We may therefore rewrite Eq. (1) as

IADC 4
 ∞
I (t ) = ∫–∞ g 12 ( µ ) g 3 ( t – µ ) dµ = g 12 ( t ) ★g 3 ( t ) = [ g 1 ( t ) ★g 2 ( t ) ]★g 3 ( t )

(c) The distributive property:

g 1 ( t )★ [ g 2 ( t ) + g 3 ( t ) ] = g 1 ( t )★g 2 ( t ) + g 1 ( t )★g 3 ( t )

Proof:
∞
g 1 ( t )★ [ g 2 ( t ) + g 3 ( t ) ] = ∫–∞ g 1 ( τ ) [ g 2 ( t – τ ) + g 3 ( t – τ ) ] dτ
∞ ∞
= ∫–∞ g 1 ( τ ) g 2 ( t – τ ) dτ + ∫–∞ g 1 ( τ ) g 3 ( t – τ ) dτ
= g 1 ( t )★g 2 ( t ) + g 1 ( t )★g 3 ( t )

Problem 2.8
Considering the sinc pulse sinc(t), show that
∞
∫–∞ sin c
2
( t ) dt = 1

Solution
This integral may be viewed as
∞
I = ∫–∞ sin c ( t ) ⋅ sin c ( t ) dt
which, in light of Rayleigh’s energy theorem, may also be expressed as
∞
∫–∞ F [ sin c ( t ) ]
2
I = df

From Eq. (2.25) in the text, we have

F [ sin ct ] = rect ( f )
Hence,
∞
∫–∞ rect
2
I = ( f )d f
1⁄2
∫–1 ⁄ 2 1
2
= df

= 1

Problem 2.9
Determine the Fourier transform of the squared sinusoidal signals:
2
(i) g ( t ) = cos ( 2π f c t )
2
(ii) g ( t ) = sin ( 2π f c t )

IADC 5
Solution
(i) Using the trigonometric identity
2 1
cos θ = --- ( 1 + cos 2θ )
2
we may express g(t) as
1
g ( t ) = --- ( 1 + cos 4π f c t )
2
Hence,
1 1 1
G ( f ) = --- δ ( f ) + --- δ ( f – 2 f c ) + --- δ ( f + 2 f c )
2 4 4
(ii) Next, using the trigonometric identity
2 1
sin θ = --- ( 1 – cos 2θ )
2
we may write
2 1
sin ( 2π f c t ) = --- ( 1 – cos 4π f c t )
2
Hence,
1 1 1
G ( f ) = --- δ ( f ) – --- δ ( f – f c ) – --- δ ( f + f c )
2 4 4

Problem 2.10
Consider the function

g ( t ) = δ  t + --- – δ  t – ---
1 1
 2  2
1
which consists of two delta functions at t = ± --- . The integration of g(t) with respect to time t
2
yields the unit rectangular function rect(t). Using Eq. (2.79), show that
rect ( t ) sin c ( f )

Solution
To begin, consider the transform pair
δ(t ) 1
Hence, the Fourier transform of g(t) is
G ( f ) = exp ( jπft ) – exp ( – j πft )
from which we readily deduce that G(0). Hence, applying Eq. (2.79) in the text yields
1
F [ rect ( t ) ] = ------------ [ exp ( jπf ) – exp ( – jπf ) ]
j2πf
sin ( πf )
= ------------------- = sin c ( f )
πf
where we have used the identity
1 jπf – jπf
sin ( πf ) = ------ ( e – e )
2j

IADC 6
Problem 2.11
Using the Euler formula
1
cos x = --- exp [ ( jx ) + exp ( – jx ) ]
2
reformulate Eqs. (2.91) and (2.92) in terms of cosinusoidal functions.

Solution
∞ ∞ –1

∑ exp ( j2πn f 0 t ) = ∑ exp ( j2πn f 0 t ) + 1 + ∑ exp ( j2πn f 0 t )

n=-∞ n=1 n=-∞
∞
= 1+ ∑ [ exp ( j2πn f 0 t ) + exp ( – j 2πn f 0 t ) ]
n=1
∞
= 1 + 2 ∑ cos ( 2πn f 0 t )
n=1
We may therefore reformulate Eq. (2.91) as
∞ ∞

∑ δ ( t – mT 0 ) = f 0 + 2 f 0 ∑ cos ( 2πm f 0 t )
m=-∞ n=1
where f0 = 1/T.
Similarly, we may write
∞ ∞

∑ cos ( j2πm f 0 t ) = 1 + 2 ∑ cos ( 2πm f 0 t )

m=-∞ m=1
Hence, we may reformulate Eq. (2.92) as
∞ ∞
1 + 2 ∑ cos ( 2πm f 0 t ) = T 0 ∑ δ( f – n f 0)
m=1 n=-∞

Problem 2.12
Discuss the following two issues, citing examples for your answers:
(a) Is it possible for a linear time-invariant system to be causal but unstable?
(b) Is it possible for such a system to be noncausal but stable?

Solution
(a) It is possible for a system to be causal but unstable. Causality means that the impulse response
of the system h(t) must be zero for negative t. Instability means that the BIBO criterion
∞
∫–∞ h ( t ) dt < ∞
is violated. Such a system could be represented by the impulse response

IADC 7
  for t ≤ 0
h(t ) =  0
 exp ( t ) for t > 0
(b) By the same token, it is possible for the system to be stable but noncausal. In this second case,
we may cite the impulse response
 
h ( t ) =  exp ( t ) for t ≤ 0
 0 for t > 0 
What does Problem 2.12 teach us?
The problem teaches us that the properties of stability and causality are independent.

Problem 2.13
The impulse response of a linear system is defined by the Gaussian function
 t2 
h ( t ) = exp  – -------2-
 2τ 
where the parameter τ is used to adjust duration of the impulse response. Determine the frequency
response of the system.

Solution
From Eq. (2.40) in the text, recall that
2 2
exp ( – πt ) exp ( – π f )
Next, from the dilation property of the Fourier transform described in Eq. (2.20), recall that if
h(t ) H ( f ) , then

----- H  ---
1 f
h ( at )
a  a
where a is the dilation parameter. For the problem at hand, we have
1 1
a = ------ ---
2π τ
Accordingly, the frequency response of the system is
2 2 2
H( f ) = 2π τ exp ( – 2π τ f )

Problem 2.14
A tapped-delay-line filter consists of N weights, where N is odd. It is symmetric with respect to
the center tap, that is, the weights satisfy the condition
w n = w N -1-n , 0≤n≤N –1
(a) Find the amplitude response of the filter.
(b) Show that this filter has a linear phase response. What is the implication of this property?

Solution
The impulse response of the filter is

IADC 8
 N -1
h(t ) = ∑ wn δ ( t – n∆τ )
n=0
Hence, the frequency response of the filter is
N -1
H( f ) = ∑ wn exp ( – j2πnf∆τ )
n=0
To illustrate, consider the example of N = 5. Then
H ( f ) = w 0 + w 1 exp ( – j2πf∆τ ) + w 2 exp ( – j4πf∆τ ) + w 3 exp ( – j6πf∆τ ) + w 4 exp ( – j8πf∆τ )
= exp ( – j4πf∆τ )[w 0 exp ( j4πf∆τ ) + w 1 exp ( j2πf∆τ ) + w 2 + w 3 exp ( – j 2πf∆τ )
+ w 4 exp ( – j 4πf∆τ )] (1)
For this example, the symmetry condition
w n = w N -1-n for 0 < n < N-1
reads as
w n = w 4-n for 0 < n < 4
Hence, w0 = w4 and w1 = w3. Accordingly, we may rewrite Eq. (1) as
H ( f ) = exp ( – j4πf∆τ )[w 0 exp ( j4πf∆τ ) + exp ( – j 4πf∆τ )
+ w 1 ( exp ( j2πf∆τ ) ) + exp ( – j2πf∆τ )
+ w2 ]
= exp ( – j 4πf∆τ ) [ 2w 0 cos ( 4πf∆τ ) + 2w 1 ( 2πf∆τ ) + w 2 ]
We may therefore generalize this result as
N -1
---------- – 1
2
H ( f ) = exp  – j2π  ---------- f∆τ w N -1 + 2
N -1
  2   ---------- ∑ w n cos ( 2πnf∆τ )
2 n=0

(a) The amplitude response of the filter is therefore

N -1
---------- – 1
2
H ( f ) = w N -1 + 2
----------
∑ w n cos ( 2πnf∆τ )
2 n=0
(b) The phase response of the filter is therefore

arg ( H ( f ) ) = exp  – j2π  ---------- f∆τ

N -1
  2  
which is linear with respect to the frequency f. The implication of this condition is that except
for a delay, there is no phase distortion produced by the filter.

Problem 2.15
Derive the relationship of Eq. (2.142) between the two cross-correlation factors Rxy(τ) and Ryx(τ).

IADC 9
Solution
By definition
∞
∫–∞ y ( t )x
*
R yx ( τ ) = ( t – τ ) dt
Complex conjugate both sides of the equation:
∞
∫–∞ x ( t – τ ) y
* *
R yx ( τ ) = ( t ) dt
Next, replace τ with -τ:
∞
∫–∞ x ( t + τ ) y
* *
R yx ( – τ ) = ( t ) dt
Finally, replace t + τ with t, which is equivalent to replacing t with t - τ; we therefore (since dt
remains unchanged)
∞
∫–∞ x ( t ) y
* *
R yx ( – τ ) = ( t – τ ) dt = R xy ( τ )

Problem 2.16
Consider the decaying exponential pulse
 exp ( – at ) t > 0

g ( t ) =  1, t=0

 0, t<0
Determine the energy spectral density of the pulse g(t).

Solution
The Fourier transform of g(t) is (see Eq. (2.12) in the text
1
G ( f ) = ----------------------
a + j2πf
The energy spectral density of the pulse is therefore
2
Eg( f ) = G( f )
1
= --------------------------
2 2 2
-, –∞ < f < ∞
a + 4π f

Problem 2.17
Repeat Problem 2.16 for the double exponential pulse
 exp ( – at ), t>0

g ( t ) =  1, t=0

 exp ( at ), t<0

Solution
The Fourier transform of g(t) is (see Eq. (2.16))

IADC 10
 2a
G ( f ) = --------------------------
2 2 2
-
a + 4π f
The energy spectral density of the double exponential pulse is
2
4a
E g ( f ) = ----------------------------------2-, –∞ < f < ∞
2 2 2
( a + 4π f )

Problem 2.18
In an implicit sense, Eq. (2.153) embodies Parseval’s power theorem, which states that for a
periodic signal x(t) we have
∞
1 T ⁄2
--- ∫ ∑
2 2
x ( t ) dt = X (n f 0)
T –T ⁄ 2
n=-∞
where T is the period of the signal, f0 is the fundamental frequency, and X(nf0) is the Fourier trans-
form of x(t) evaluated at the frequency nf0. Prove this theorem.

Solution
Adapting Eq. (2.86) to the problem at hand, we may write
∞
xT ( t ) = f 0 ∑ X ( n f 0 ) exp ( j2πn f 0 t ) (1)
n=-∞
where
 T T 

x T ( t ) =  x ( t ), – --- ≤ t ≤ ---
2 2

 0, otherwise 
1
f 0 = ---
T
and X(nf0) is the Fourier transform of g(t), evaluated at the frequency f = nf0. Using Eq. (1) to
evaluate the integral
1 T ⁄2
--- ∫
2
x ( t ) dt
T –T ⁄ 2
we write
∞ ∞
T ⁄2   
I = f 0∫  ∑ X ( n f 0 ) exp ( j2πn f 0 t )  ∑ X ( m f 0 ) exp ( – j 2πm f 0 t ) dt
*
–T ⁄ 2    m=-∞ 
n=-∞
∞
T ⁄2
∑ X (n f 0) X (m f 0)∫
*
= f0 exp ( j2π ( n – m ) f 0 t ) dt (2)
–T ⁄ 2
n=-∞
To evaluate the integral on the right-hand side of Eq. (2), we write
T ⁄2 1 T ⁄2
∫–T ⁄ 2 exp ( j2π ( n –m ) f 0 t ) dt = -------------------------------
j2π ( n – m ) f 0
- exp ( j2π ( n – m ) f 0 t ) t = –T ⁄ 2

IADC 11
 1
= -------------------------------- [ exp ( ( jπ ( n – m ) ) – exp – jπ ( n – m ) ) ]
j2π ( n – m ) f 0
1
= ---------------------------- sin ( π ( n – m ) ) (3)
π(n – m) f 0
Whenever the indices n and m are assigned different integer values, Eq. (3) assumes the value
zero. On the other hand, whenever the indices are assigned the same integer value, the integral in
Eq. (3) assumes the limiting value
1 sin ( π ( n – m ) ) 1
------ lim ---------------------------------- = ------
f 0 n=m π(n – m) f0
Accordingly, we may simplify Eq. (3) as
 1
T ⁄2  ------ , n = m
∫–T ⁄ 2 exp ( j2π ( n –m ) f 0 t ) dt =  f 0 (4)
 0, otherwise
Hence, substituting Eq. (4) into (2), we get
∞

∑
2
I = X (n f 0) (5)
n=-∞
We finally write
∞
1 T ⁄2
--- ∫ ∑
2 2
x ( t ) dt = X (n f 0)
T –T ⁄ 2
n=-∞
which is the desired result.

IADC 12
ADDITIONAL PROBLEMS

Problem 2.19

(a) The half-cosine pulse g(t) of Fig. 2.40(a) may be considered as the product of the rectangular
function rect(t/T) and the sinusoidal wave Acos(πt/T). Since

rect  ---
t
T sin c ( fT )
T
πt
A cos  -------- δ  f – ------- + δ  f + -------
A 1 1
T 2  2T   2T 
and multiplication in the time domain is transformed into convolution in the frequency
domain, it follows that
A 1 
G ( f ) = [ T sin c ( fT ) ]★  --- δ  f – ------- + f + ------- 
1
 2T  
2 2T
where ★ denotes convolution. Therefore, noting that

sin c ( fT ) ★δ  f – ------- = sin c T  f – -------

1 1
 2T   2T 

sin c ( fT ) ★δ  f + ------- = sin c T  f + -------

1 1
 2T   2T 
we obtain the desired result

G ( f ) = -------- sin c  fT – --- + sin c  fT + ---

AT 1 1
2  2  2

(b) The half-sine pulse of Fig. 2.40(b) may be obtained by shifting the half-cosine pulse to the
right by T/2 seconds. Since a time shift of T/2 seconds is equivalent to multiplication by
exp(-jπfT) in the frequency domain, it follows that the Fourier transform of the half-sine pulse
is

G ( f ) = -------- sin c  fT – --- + sin c  fT + --- exp ( – jπfT )

AT 1 1
2  2   2

(c) The Fourier transform of a half-sine pulse of duration aT is equal to

-------------- sin c  afT – --- + sin c  afT + --- exp ( – jπfaT )

a AT 1 1
2  2   2

(d) The Fourier transform of the negative half-sine pulse shown in Fig. 2.40(c) is obtained from
the result by putting a = -1, and multiplying the result by -1, and so we find that its Fourier
transform is equal to

– -------- sin c  fT + --- + sin c  fT – --- exp ( jπfT )

AT 1 1
2  2   2

IADC 13
(e) The full-sine pulse of Fig. 2.40(d) may be considered as the superposition of the half-sine
pulses shown in parts (b) and (c) of the figure. The Fourier transform of this pulse is therefore

G ( f ) = -------- sin c  fT – --- + sin c  fT + --- [ exp ( – j πfT ) – exp ( jπfT ) ]

AT 1 1
2  2  2

= – jAT sin c  fT – --- + sin c  fT + --- sin ( πfT )

1 1
 2  2
π π
sin  πfT – --- sin  πfT + ---
 2  2
= – jAT -------------------------------- + -------------------------------- sin ( πfT )
π π
πfT – --- πfT + ---
2 2

cos ( πfT ) cos ( πfT )

= – j AT – ----------------------- + ----------------------- sin ( πfT )
π π
πfT – --- πfT + ---
2 2
sin ( 2πfT ) sin ( 2πfT )
= jAT ------------------------- – -------------------------
2πfT – π 2πfT + π
sin ( 2πfT – π ) sin ( 2πfT + π )
= jAT – ----------------------------------- + -----------------------------------
2πfT – π 2πfT + π
= jAT [ sin c ( 2 fT + 1 ) – sin c ( 2 fT – 1 ) ]

Problem 2.20

(a) The even part ge(t) of a pulse g(t) is given by

1
g e ( t ) = --- [ g ( t ) + g ( – t ) ]
2

Therefore, for g ( t ) = Arect  --- – --- we obtain

t 1
 T 2

g e ( t ) = --- rect  --- – --- + rect  – --- – ---

A t 1 t 1
2  T 2  T 2

= --- rect  -------

A t
2  2T 
which is shown illustrated in Fig. 1:
g(t)
A

t
0 T

ge(t)
A/2
Figure 1
t
-T T

IADC 14
 The odd part of g(t) is defined by
1
g o ( t ) = --- [ g ( t ) – g ( – t ) ]
2

= --- rect  --- – --- – rect  – --- – ---

A t 1 t 1
2  T 2   T 2
which is illustrated in Fig. 2:
go(t)
A
-T 2
t
T
- A
2
Figure 2

(b) The Fourier transform of the even part is

G e ( f ) = AT sin c ( 2 fT )
The Fourier transform of the odd part is
AT AT
G o ( f ) = -------- sin c ( fT ) exp ( – jπfT ) – -------- sin c ( fT ) exp ( jπfT )
2 2
AT
= -------- sin c ( fT ) sin ( πfT )
j

Problem 2.21

Express g(t) as
g (t ) = g1(t ) + g2(t )
where
1 0  πu 2
g 1 ( t ) = --- ∫ exp  – --------
- du
τ t-T  τ2 

1 t+T  πu 2
g 2 ( t ) = ∫ exp  – --------
--
- - du
τ 0  τ2 
Therefore,
1 0  πu 2
g 1 ( t+T ) = ∫ exp  – --------
--
- - du
τ t  τ2 

1 t  πu 2
= – --- ∫ exp  – --------
- du
τ 0  τ2 

1 t  πu 2 1
= – --- ∫ exp  – --------
- du + ---
τ –∞  τ  2 2
where we have made use of the fact that

IADC 15
1 t  πu 2 1
--- ∫ exp  – --------- du = ---
τ –∞  τ  2 2
Similarly
t πu 2
g 2 ( t-T ) = ∫ exp  – --------
- du
0  τ2 

1 t  πu 2 1
= --- ∫ exp  – --------
- du = ---
τ –∞  τ  2 2
Next, noting the following four relationships
F [ g 1 ( t+T ) ] = G 1 ( f ) exp ( j2πfT )
F [ g 2 ( t-T ) ] = G 2 ( f ) exp ( – j 2πfT )

 πt 2 2 2
exp  – ------- τ exp ( – πτ f )
 τ2 
t 1 1
∫–∞ g ( u ) du
= ------------G ( f ) + --- G ( 0 )δ ( f )
j2πf 2
we find that taking the Fourier transforms of g1(t+T) and g2(t-T) respectively yields
1 2 2
G 1 ( f ) exp ( j2πfT ) = – ------------ exp ( – πτ f )
j2πf
1 2 2
G 2 ( f ) exp ( – j 2πfT ) = ------------ exp ( – πτ f )
j2πf
Therefore,
1 2 2
G 1 ( f ) = ------------ exp ( – πτ f ) exp ( – j2πfT )
j2πf
1 2 2
G 2 ( f ) = ------------ exp ( – πτ f ) exp ( j2πfT )
j2πf
Thus the Fourier transform of g(t) is
G ( f ) = G1 ( f ) + G2( f )
1 2 2
= ------------ exp ( – πτ f ) [ exp ( – j 2πfT ) + exp ( j2πfT ) ]
j2πf
1 2 2
= ------ exp ( – πτ f ) sin ( 2πfT )
πf
2 2
= 2T exp ( – πτ f ) sin c ( 2 fT )
When τ approaches zero, G(f) approaches the limiting value 2Tsinc(2fT), which corresponds to
the Fourier transform of a rectangular pulse of unit amplitude and duration 2T, which is correct.

IADC 16
Problem 2.22

∞
(a) G ( f ) = ∫–∞ g ( t ) exp ( – j2πft ) dt
0 ∞
= ∫–∞ g ( t ) exp ( – j2πft ) dt + ∫0 g ( t ) exp ( – j2πft ) dt
0 0
= ∫–∞ g ( t ) cos ( 2πft ) dt – ∫–∞ jg ( t ) sin ( 2πft ) dt
∞ ∞
+ ∫ g ( t ) cos ( 2πft ) dt – ∫ jg ( t ) sin ( 2πft ) dt
0 0
If g(t) is even, then g(t) = g(-t). Hence,
0 ∞
∫–∞ g ( t ) cos ( 2πft ) dt = ∫0 g ( t ) cos ( 2πft ) dt
0 ∞
∫–∞ g ( t ) sin ( 2πft ) dt = – ∫ g ( t ) sin ( 2πft ) dt
0
and so
∞
G ( f ) = 2 ∫ g ( t ) cos ( 2πft ) dt . which is purely real.
0
If, on the other hand, g(t) is odd, g(t) = -g(-t). Hence,
0 ∞
∫–∞ g ( t ) sin ( 2πft ) dt = ∫0 g ( t ) sin ( 2πft ) dt
0 ∞
∫–∞ g ( t ) cos ( 2πft ) dt = – ∫ g ( t ) cos ( 2πft ) dt
0
and thus
∞
G ( f ) = – 2 j ∫ g ( t ) sin ( 2πft ) dt which is purely imaginary.
0

(b) The Fourier transform of g(t) is defined by

∞
G( f ) = ∫–∞ g ( t ) exp ( – j2πft ) dt
Differentiating both sides of this relation n times with respect to f:
n
d G( f ) n ∞ n
------------------
df
n
- = ( – j2π ) ∫– ∞
t exp ( – j2πft ) dt (1)

That is,
n
n j  nd G( f )
 -----
t g(t ) - ------------------
-
 2π n
df

(c) Putting f = 0 in Eq. (1), we get

∞ j  n (n)
 -----
∫–∞
n
t g ( t ) d t = - G (0)
 2π

IADC 17
 n
(n) d G( f )
where G ( f ) = -------------------
d fn

(d) Since
* *
g2(t ) G2( – f )
it follows that
∞
∫–∞ G 1 ( λ )G 2 ( λ – f ) dλ
* *
g 1 ( t )g 2 ( t )
From this result we deduce the Fourier transform
∞
∫–∞ g 1 ( t )g 2 ( t ) exp ( – j2πft ) dt
* *
F [ g 1 ( t )g 2 ( t ) ] =
∞
∫–∞ G 1 ( λ )G 2 ( λ – f ) dλ
*
= (2)
Setting f = 0 in Eq. (2), we get the desired relation
∞ ∞
∫–∞ g 1 ( t )g 2 ( t ) ∫–∞ G 1 ( λ )G 2 ( λ ) dλ
* *
=

IADC 18
Problem 2.23

We are given the following inequalities:

∞
G( f ) ≤ ∫ g 1 ( t ) dt
–∞
∞
j2πfG ( f ) ≤ ∫ dg ( t ) dt
-------------
–∞ dt
∞ 2
j2π f G ( f ) ≤ ∫
2 d g(t)
---------------- dt
–∞ dt
Considering the triangular pulse g(t) of Fig. 2.41 in the text, its first and second derivatives with
respect to time t are illustrated in Fig. 1:
g(t)
A

t
-T 0 T

dg(t)
dt
A/T
T
t
-T 0
-A/T

d2g(t)
A/T A/T
dt2

t
-T 0 T
Figure 1
-2A/T

We thus have
∞
∫–∞ g ( t ) dt = AT
∞ dg ( t )
∫–∞ ------------
dt
- dt = 2A
2
∞ d g(t) ∞ A
∫–∞ ---------------
dt
- dt = ∫–∞ ---T δ ( t + T ) – 2δ ( t ) + δ ( t – T ) dt
4A
= -------
T
The bounds on the amplitude spectrum |G(f)| are therefore as follows:
G ( f ) ≤ AT

IADC 19
 A
G ( f ) ≤ ----------
π f
A
G ( f ) ≤ ---------------
2 2
-
π f T
which are shown plotted in Fig. 2.
|G(f)|

AT
Bound (1)

Bound (2)
Bound (3)

f
-2 1 0 1 2
T T T T
Actual
Figure 2 spectrum

The actual amplitude spectrum of the triangular pulse is given by

2
G ( f ) = AT sin c ( fT )
which is also plotted in Fig. 1. From this figure we see that bounds (1) and (3) define boundaries
on the actual spectrum |G(f)|.

Problem 2.24

(a) The convolution of g1(t) and g2(t) is defined by

∞
g 1 ( t )★g 2 ( t ) = ∫–∞ g 1 ( τ )g 2 ( t – τ ) dτ (1)
Differentiating both sides of Eq. (1) with respect to time t:
d ∞ d
----- [ g 1 ( t )★g 2 ( t ) ] = ∫ g ( τ ) ----- g ( t – τ ) dτ
dt –∞ dt 2
∞ d
= ∫ g 1 ( τ ) ------------------ g ( t – τ ) dτ
–∞ d (t – τ) 2
d
= g 1 ( t )★ ----- g ( t )
dt 2
Since convolution is commutative, we may also write
d d
----- [ g 1 ( t )★g 2 ( t ) ] = ----- g ( t ) ★g 2 ( t )
dt dt 1
In other words, the derivative of a convolution produce of two signals is equivalent to the
convolution of one of the signals and the derivative of the other.

IADC 20
(b) Changing variables in Eq. (1), we may write
∞
g 1 ( t )★g 2 ( t ) = ∫–∞ g 1 ( λ )g 2 ( t – λ ) dλ (2)
Integrating both sides of Eq. (2) with respect to t:
t t ∞
∫–∞ [ g 1 ( τ )★g 2 ( τ ) ] dτ = ∫–∞ ∫–∞ g 1 ( λ )g 2 ( τ – λ )λ dτ
Interchanging the order of integration and rearranging terms:
t ∞ ∞
∫–∞ [ g 1 ( τ )★g 2 ( τ ) ] dτ = ∫– ∞ g1(λ)∫
–∞
g 2 ( τ – λ )τ dλ d (3)
Recognizing that
t t-λ
∫–∞ g 2 ( τ – λ ) dλ = ∫ – ∞ g 2 ( τ ) dτ
we may rewrite Eq. (3) as
t ∞ t-λ
∫–∞ [ g 1 ( τ )★g 2 ( τ ) ] dτ = ∫– ∞ g1(λ)∫
–∞
g 2 ( τ ) dτ dλ

t
= g 1 ( t )★ ∫–∞ g 2 ( τ ) dτ
In other words, the integral of a convolution product of two signals is equivalent to the
convolution of one of the signals and the integral of the other.

Problem 2.25

Express y(t) as
2
y(t ) = x (t )
= x ( t )x ( t )
Since multiplication in the time domain corresponds to convolution in the frequency domain, we
may express the Fourier transform of y(t) as
∞
Y( f ) = ∫–∞ X ( λ ) X ( f – λ ) dλ
where X(f) is the Fourier transform of x(t). However, X(f) is zero for |f| > W. Therefore,
W
Y( f ) = ∫–W X ( λ ) X ( f – λ ) dλ
In this integral we note that X(f - λ) is limited to -W < f - λ < W. When λ = -W, we find that
-2W < f < 0. When λ = W, we find that 0 < f < 2W. Accordingly, the Fourier transform Y(f) is
limited to the frequency interval -2W < f < 2W.

IADC 21
Problem 2.26

(a) Consider a rectangular pulse g(t) of duration T and amplitude 1/T, centered at t = 0, as shown
in Fig. 1:

g(t)
1/T
area = 1

t
Figure 1 -T 0 T
2 2
The Fourier transform of g(t) is
sin ( πfT )
G ( f ) = ----------------------
πfT
As the duration T approaches zero, g(t) approaches a delta function, and so we find that in the
limit:
sin ( πfT )
lim G ( f ) = lim ---------------------- = 1
T →0 T → 0 πfT

(b) Consider next the sinc pulse 2W sinc(2Wt) of unit area, as shown in Fig. 2:
g(t)
2W

t
0 1
- 3 - 1 - 1 1 - 3
2W 2W 2W W W
W

The Fourier transform of g(t) is

G ( f ) = rect  ---------
f
 2W 
which has unit amplitude and width 2W, centered at f = 0. As W approaches infinity, g(t)
approaches a delta funct6ion, and the corresponding Fourier transform becomes equal to unity
for all f.

Problem 2.27

The G(f) is in the form of a unit step function defined in the frequency domain, as shown in Fig. 1
1.0

1
2
.
f
Figure 1 0

Now, for a unit step function defined in the time domain, we have

IADC 22
 1 1
u(t ) ------------ + --- δ ( f )
j2πf 2
Applying the duality property of the Fourier transform to this relation, we get
1 1
– ----------- + --- δ ( f ) u( f )
j2πt 2
where we have used the fact that δ(-t) = δ(t). Therefore, the time function g(t) whose Fourier
transform is depicted in Fig. 1, is given by
1 1
g ( t ) = – ----------- + --- δ ( t )
j2πt 2

Problem 2.28

2
d g(t)
(a) Taking the Fourier transform of both sides of ---------------
2
- = ∑ k i δ ( t – t i ) , we get
dt i

∑ k i exp ( – j2πf t i )
2
( j2πf ) G ( f ) =
i
–1
2 2∑ i
Therefore, G(f) = --------------- k exp ( – j2πf t i )
4π f i

(b) Differentiating the trapezoidal pulse of Fig. 2.42 twice, we get:

d2g(t)/dt2

-ta ta
-tb 0 tb

Hence,
–A
G ( f ) = ----------------------------------
2 2
- [ exp ( j2πf t b ) – exp ( j2πf t a ) – exp ( j2πf t a ) – + exp ( j2πf t b ) ]
4π f ( t b – t a )
–A
- [ cos ( j2πf t b ) – cos ( j2πf t a ) ]
= ----------------------------------
2 2
2π f ( t b – t a )
–A
- sin [ πf ( t b – t a ) sin πf ( t b + t a ) ]
= ----------------------------------
2 2
2π f ( t b – t a )

Problem 2.29

(a) From part (b) of Problem 2.28, we have

A
2 2
- sin [ πf ( t b – t a ) ] sin [ πf ( t b + t a ) ]
G ( f ) = ------------------------------- (1)
π f (tb – ta)
As tb approaches ta, we get the following result:

IADC 23
 1
lim 2
- sin [ πf ( t b – t a ) ] = 1
----------------------------
tb → ta π f (tb – ta)
and
lim sin [ πf ( t b + t a ) ] = sin ( πf t a )
tb → ta

Accordingly, the Fourier transform of Eq. (1) approaches the limiting value
A
lim G ( f ) = ------ sin ( 2πf t a )
tb → ta πf
sin ( 2πf t a )
= 2t a A ---------------------------
2πf t a
= 2t a A sin c ( 2πf t a ) (2)
which is the desired result.

(b) The limiting Fourier transform of Eq. (2) is recognized as the Fourier transform of a
rectangular pulse of amplitude A and duration T = 2ta.

Problem 2.30

The transfer function H(f) and impulse response h(t) of a linear time-invariant filter are related by
∞
H( f ) = ∫–∞ h ( t ) exp ( – j2πft ) dt
Applying a special form of Schwarz’s inequality (see Appendix 5), we may write
∞
H( f ) ≤ ∫ h ( t ) exp ( – j2πft ) dt
–∞
Since exp ( – j2πft ) = 1 , we may simplify this relation as
∞
H( f ) ≤ ∫ h ( t ) dt
–∞
If the filter is stable, the impulse response is absolutely integrable:
∞
∫–∞ h ( t ) dt < ∞
Therefore, the amplitude response of a stable filter is bounded for every value of the frequency f,
as shown by
H( f ) = ∞
According to Rayleigh’s energy theorem, the energy of the input signal x(t) is given by
∞ ∞
∫–∞ x ( t ) ∫–∞
2 2
Ex = dt = X( f ) df
and the energy of the output signal y(t) is
∞ ∞
∫–∞ ∫–∞ Y ( f )
2 2
Ey = y ( t ) dt = df
The Fourier transforms Y(f) and X(f) are related by

IADC 24
Y ( f ) = H( f )X ( f )
Therefore,
∞
∫–∞ H ( f )
2 2
Ey = X( f ) df (1)
For a stable filter, we may express |H(f)| in the form K|Hn(f)| where K is a scaling factor equal to
the maximum value of |H(f)| and |Hn(f)| < 1 for all f. Thus, we may rewrite Eq. (1) in the form:
2 ∞
∫–∞ H n ( f )
2 2
Ey = K X( f ) df
Since |Hn(f)| < 1 for all f, it follows that
∞ ∞
∫–∞ H ( f ) X( f ) df ≤ ∫
2 2 2
X( f ) df
–∞
or equivalently
2 ∞
∫– ∞
2
Ey ≤ K X( f ) df
If the input signal has finite energy, we then have
∞
∫–∞
2
X( f ) df = ∞

Accordingly, we find that E y < ∞ , which means that the output signal y(t) also has finite energy.

Problem 2.31

(a) The transfer function of the ith stage of the system of Fig. 2.43 is
1
H i ( f ) = -----------------------------
1 + j2πfRC
1
= --------------------------- , T0 = RC
1 + j2πf τ 0
where it is assumed that the buffer amplifier has a constant gain of unity. The overall transfer
function of the system is therefore
N
H( f ) = ∏ H i( f )
i=1

1
= -----------------------------------
N
-
( 1 + j2πf τ 0 )
The corresponding amplitude response is
1
H ( f ) = --------------------------------------------- (1)
2 N ⁄2
[ 1 + ( 2πf τ 0 ) ]

IADC 25
(b) Let
2
2 T
τ 0 = -------------
2
-
4π N
Then, we may rewrite Eq. (1) for the amplitude response as
1 –N ⁄ 2
2
H ( f ) = 1 + ---- ( fT )
N
In the limit, as N approaches infinity we have
1 –N ⁄ 2
2
H ( f ) = lim 1 + ---- ( fT )
N →∞ N
N 1 2
= exp ---- ⋅ ---- ( fT )
2 N
2 2
= exp  – -------------
f T
 2 

Problem 2.32

(a) The integrator output is

t
y(t ) = ∫t-T x ( τ ) dτ
Let x ( t ) X ( f ) ; then
∞
x(t ) = ∫–∞ X ( f ) exp ( j2πft ) d f
Therefore,
t ∞
y(t ) = ∫t-T ∫–∞ X ( f ) exp ( j2πfτ ) d f dτ

Interchanging the order of integration:

∞ t
y(t ) = ∫–∞ X ( f ) ∫t-T exp ( j2πfτ ) dτ df

∞
= ∫–∞ [ TX ( f ) sin c ( fT ) exp ( – jπfT ) ] exp ( j2πft ) d f
The Fourier transform of the integrator output is therefore
Y ( f ) = TX ( f ) sin c ( fT ) exp ( – jπfT ) (1)
Equation (1) shows that y(t) can be obtained by passing the input signal x(t) through a linear
filter whose transfer function is equal to Tsinc(fT)exp(-jπfT).

IADC 26
(b) The amplitude response of this filter is shown in Fig. 1:
|H(f)|
T

Figure 1 0
f
-3 -2 -1 1 2 3
T T T T T T

The approximation with an ideal low-pass filter of bandwidth 1/T, gain T, and delay T/2, is
shown dashed in Fig. 1. The response of this ideal filter to a unit step function applied at t = 0
is given by
2π  T 
------ t – ---
T T  2  sin λ
y ideal ( t ) = --- ∫ ----------- dλ
π –∞ λ
At time t = T. we therefore have
T π sin λ
y ideal ( t ) = --- ∫ ----------- dλ
π –∞ λ
T 0 sin λ π sin λ
= ---
π ∫–∞ ----------
λ
- dλ + ∫ ----------- dλ
0 λ
T
= --- [ Si ( ∞ ) + Si ( π ) ]
π
T π
= ---  --- + 1.85
π2 
= 1.09T (2)
On the other hand, the output of the ideal integrator to a unit step function, evaluated at time
t = T, is given by
T
y(T ) = ∫0 u ( τ ) dτ
= T (3)
Thus, comparing Eqs. (2) and (3) we see that the ideal low-pass filter output exceeds the ideal
integrator output by only nine percent for T = 1.

Problem 2.33

The half cosine pulse in Fig. 2.33(a) is

πt
g ( t ) = Arect  --- cos  -----
t
T T
Fourier transforming both sides gives
sin [ πfT ]  1 1 
G ( f ) = AT ---------------------- ★  --- δ  f – ------- + δ  f + ------- 
1
πfT  2T   2T  
2

IADC 27
 π π
AT sin πfT – --- AT sin πfT + ---
2 2
= ----------------------------------------- + -----------------------------------------
π π
2  πfT – --- 2  πfT + ---
 2   2
2 AT cos ( πfT )
= ----------------------------------------------------
π ( 1 – 2 fT ) ( 1 + 2 fT )
Therefore, the energy density of g(t) is
2 2 2 2 2 2
4 A T cos ( πfT )
2 4 A T cos ( πfT )
Ψ ( f ) = G ( f ) = -----------------------------------------
2
- = -----------------------------------------
2
- (1)
2 2 2 2 2 2
π (1 – 4 f T ) π ( 4T f – 1 )
Consider next the half-sine pulse in Fig. 2.33(b), which is the same as that of Fig. 2.33(b) shifted
to the right by T/2. This time-shift corresponds to multiplication by exp(-j2πfT), which has unit
amplitude for all f. Therefore, both pulses have exactly the same energy density defined in Eq. (1).

Problem 2.34

The autocorrelation function of a deterministic signal g(t) is defined by

∞
Rg(τ) = ∫–∞ g ( t )g ( t – τ ) dt (1)
This formula applies to a real-valued signal, which is satisfied by all three signals specified under
parts (a) through (c).

(a) g(t) = exp(-at)u(t), u(t): unit step function

Applying Eq. (1) yields
∞
Rg(τ) = ∫τ exp ( –at ) exp ( –a ( t – τ ) ) dt
∞
= exp ( aτ ) ∫ exp ( – 2at ) dt
τ

1 ∞
= exp ( aτ ) – ------ exp ( – 2at )
2a t=τ
1
= ------ exp ( – aτ )
2a
which is depicted in Fig. 1
Rg(t)
1
2a

τ
0
Figure 1

IADC 28
(b) g(t) = exp(-a|t|)
which is sketched in Fig. 2(a). Part (b) of the figure sketches g(t - τ) = exp(-a|t - τ|)
g(t)
1.0

(a)
t
0

g(t-τ)
1.0

(b)
Figure 2
t
0 τ

In light of Fig. 1, applying Eq. (1):

∞ τ
Rg(τ) = ∫τ exp ( – at ) exp ( – a ( t – τ ) ) dt + ∫ exp ( – at ) exp ( – a ( t – τ ) ) dt
0
τ
+ ∫ exp ( at ) exp ( a ( t – τ ) ) dt
0
1 1
= ------ exp ( – at ) + τ exp ( – at ) + ------ exp ( – aτ )
2a 2a

=  --- + τ exp ( – aτ )
1
a 
which is sketched in Fig. 3.
Rg(τ) 1
a

Figure 3 τ
0

(c) g(t) = exp(-at)u(t) - exp(at)u(-t)

IADC 29
 which is sketched in Fig. 4(a). Part (b) of the figure sketches g(t - τ)
g(t)

1.0

0
(a) t

-1

g(t-τ)

1.0

0
(b) t

-1.0
Figure 4
In light of Fig. 4, applying Eq. (1) for τ > 0:
∞
Rg(τ) = ∫τ exp ( –at ) exp ( –a ( t – τ ) ) dt
τ
+ ∫ exp ( – at ) [ – exp ( a ( t – τ ) ) ] dt
0
0
+∫ [ – exp ( – at ) ] [ – exp ( a ( t – τ ) ) ] dt
–∞
1 1
= ------ exp ( – aτ ) – τ exp ( – aτ ) + ------ exp ( – aτ )
2a 2a

=  --- – τ exp ( – aτ ) ,
1
τ>0
a 
Similarly, for τ < 0 we have

R g ( τ ) =  --- + τ exp ( aτ )
1
a 
Summing up these two results:

  1--- – τ exp ( – aτ ), τ ≥ 0
 a 
Rg(τ) = 
  1--- + τ exp ( aτ ),
 τ≤0
a 


IADC 30
 which is sketched in Fig. 5.
Rg(τ)
1
a

-2/a -1/a 2/a

τ
0 1/a

- 1 exp(-2)
a

Figure 5

Problem 2.35

Applying the formula for the autocorrelation function

∞
∫–∞ g ( t )g
*
Rg(τ) = ( t – τ ) dt
to the specified signal
1  πt 2
g ( t ) = ---- exp  – -------, –∞ < t < ∞
t0  t2  0
we get
∞ 1 π 2
∫–∞ t---2- exp
2
Rg(τ) = ---2- ( t + ( t – τ ) ) dt
0 t0

1 ∞  π 2
= ---2- ∫
2
exp  – ---2- ( 2t – 2tτ + τ ) dt
t 0 –∞  t 0

1 ∞  π  τ π τ
2
= ---2- ∫ exp  – ---2- 2t – ------- – ---2- ----- dt
t 0 –∞  t 0  2 t 0 2

 πτ 2 ∞  π τ 2
= ---2- exp  – -------2- ∫ exp  – ---2-  2t – ------- dt
1
(1)
t0  2t 0  –∞  t 0  2

τ
Let x = ----  2t – ------- , and therefore (for fixed τ)
1
t0  2
t0
dt = ------- dx
2
We may then rewrite Eq. (1) as
1  πτ 2 ∞
R g ( τ ) = ----------- exp  – -------2- ∫ exp ( – πx ) d x
2
(2)
2t 0  2t 0  –∞
Recognizing that

IADC 31
 ∞
∫–∞ exp ( –πx
2
) dx = 1
we find that Eq. (2) simplifies to
1  πτ 2
R g ( τ ) = ----------- exp  – -------2-
2t 0  2t 0 
which has the same form as the bell-shaped Gaussian curve:
Rg(τ) 1
2 t0

τ
0

Problem 2.36

We are given the Fourier transform

G ( f ) = sin c ( f )
Using the transform pair
2
Rg(τ) G( f )
we may therefore express the autocorrelation function Rg(τ) as the inverse Fourier transform of
sinc2(f). From Eq. (2.43) in the text, we readily deduce that Rg(τ) has the triangular form
Rg(τ)
1.0

τ
-1 0 1

Problem 2.37

Recognizing two facts:

1. Rg(τ) = |G(f)|2, and
2. the spectrum G(f) is invariant to a time shift,
we infer that a signal g(t) and the time-shifted version g(t - t0) for any t0 will have exactly the
same autocorrelation function.

Problem 2.38

(a) We are given the power signal

g ( t ) = A 0 + A 1 cos ( 2π f 1 t + θ 1 ) + A 2 cos ( 2π f 2 t )

IADC 32
 The three components of g(t) are uncorrelated with each other. Therefore, the power spectral
density of g(t) is the sum of the power spectral densities of the three constituent components,
as shown by
2 2 2
A A1 A2
S g ( f ) = ------ δ ( f ) + ------ [ δ ( f – f 1 ) + δ ( f + f 1 ) ] + ------ [ δ ( f – f 2 ) + δ ( f + f 2 ) ]
2 4 4
Correspondingly, the autocorrelation function Rg(τ) is given by
2 2 2
A A1 A2
R g ( τ ) = ------ + ------ cos ( 2π f 1 τ ) + ------ cos ( 2π f 2 τ )
2 2 2
(Here we are postulating a fundamental result that, as with energy signals, the autocorrelation
function and power spectral density of a power signal constitute a Fourier-transform pair).

2
A
(b) R g ( 0 ) = ------ .
2

(c) In calculating the autocorrelation function, information about the phase shifts θ1 and θ2 is
completely lost.

Problem 2.39

We will determine the autocorrelation function of the signal g(t) depicted in Fig. 2.45 by
proceeding on a segment-by-segment basis:

1. The maximum value of Rg(τ) occurs at τ = 0, for then g(t) and g(t-τ) overlap exactly, yielding
2 2
Rg(0) = 3( A )(T ) = 3 A T

2. For 0 < τ < (T/2), we have the picture depicted in Fig. 1. From this figure, we obtain
–T ⁄ 2 – ( T ⁄ 2 )+τ
Rg(τ) = ∫–( 3T ⁄ 2 )+τ ( + A ) ( + A ) dt + ∫
–T ⁄ 2
( + A ) ( - A ) dt
–T ⁄ 2 ( T ⁄ 2 )+τ
+∫ ( + A ) ( + A ) dt + ∫ ( - A ) ( + A ) dt
– ( T ⁄ 2 )+τ T ⁄2
3T ⁄ 2
+∫ ( - A ) ( - A ) dt
( T ⁄ 2 )+τ
2 2 2 2 2
= A (T – τ) – A τ + A (T – τ) – A τ + A (T – τ)
2
= A ( 3T – 5τ ) , 0 < τ < (T ⁄ 2)
where the use of |τ| is invoked in light of the symmetric property of the autocorrelation
function.

3. Next, for (T/2) < τ < T, we have the picture depicted in Fig. 2, from which we obtain
–T ⁄ 2 – ( T ⁄ 2 )+τ
Rg(τ) = ∫–( 3T ⁄ 2 )+τ ( - A ) ( - A ) dt + ∫
–T ⁄ 2
( + A ) ( - A ) dt

IADC 33
 T ⁄2 ( T ⁄ 2 )+τ
+∫ ( + A ) ( + A ) dt + ∫ ( - A ) ( + A ) dt
– ( T ⁄ 2 )+τ T ⁄2
3T ⁄ 2
+∫ ( - A ) ( - A ) dt
( T ⁄ 2+τ )
2 2 2 2 2
= A (T – τ) – A τ + A (T – τ) – A τ + A (T – τ)
2 T
= A ( 3T – 5τ ) , --- < τ < T
2

4. Next, for T < τ < 3T/2, we have the picture depicted in Fig. 3, from which we obtain
T ⁄2 – ( T ⁄ 2 ) +τ
Rg(τ) = ∫–( 3T ⁄ 2 )+τ ( + A ) ( - A ) dt + ∫
T ⁄2
( - A ) ( - A ) dt
3T ⁄ 2
+∫ ( - A ) ( + A ) dt
– ( T ⁄ 2 )+τ
2 2 2
= – A ( 2T – τ ) + A ( – T + τ ) – A ( 2T – τ )
2
= A ( – 5T + 3τ ) for T < |τ| < 3T/2

5. For (3T/2) < τ < 2T we have the picture depicted in Fig. 4, from which we obtain
T ⁄2 – ( T ⁄ 2 ) +τ 3T ⁄ 2
Rg(τ) = ∫–( 3T ⁄ 2 )+τ ( + A ) ( - A ) dt + ∫T ⁄ 2 ( - A ) ( - A ) dt + ∫
– ( T ⁄ 2 )+τ
( - A ) ( + A ) dt
2 2 2
= – A ( 2T – τ ) + A ( – T + τ ) – A ( 2T – τ )
2
= A ( – 5T + 3τ ) for (3T/2) + |τ| < 2T

g(t)

-3T/2 -T/2 T/2 3T/2 t

0

-A

g(t - τ)
A
τ
0
t

-A

Figure 1: -(T/2) < τ < (T/2)

IADC 34
 g(t)

-3T/2 -T/2 0 T/2 3T/2 t

-A

g(t - τ)
τ A

t
0

−A

Figure 2: (T/2) < τ < (3T/2)

g(t)
A

-3T/2 -T/2 T/2 3T/2

t
0

-A

τ τ

g(t - τ)
A

t
0

-A

Figure 3: T < τ < (3T/2)

IADC 35
 g(t)
A

-3T/2 -T/2 T/2 3T/2

t
0

-A

g(t - τ)
A

t
0

-A

Figure 4: (3T/2) < τ < 2T

g(t)
A

-3T/2 -T/2 T/2 3T/2

t
0

-A

g(t - τ)
A
τ
t
0

-A

Figure 5: 2T < τ < 3T

6. Next, for 2T < τ < 3T we have the picture depicted in Fig. 5, from which we obtain
3T ⁄ 2
Rg(τ) = ∫–3T ⁄ 2+τ ( - A ) ( - A ) dt
2
= A ( 3T – τ ) for 2T < |τ| < 3T

IADC 36
7. Finally, for |τ| > 3T, we find that R g ( τ ) = 0 .
Putting all these pieces together, we get the autocorrelation function Rg(τ) plotted in Fig. 6, which
is symmetric about the origin τ = 0.

Rg(τ)

3A2T

T
τ
0 T/2 2T

2A2T

Figure 6: Plot of the autocorrelation function Rg(tτ)

IADC 37
Problem 2.40

We start with the Fourier-transform pair

Rg(τ) Ψg( f ) (1)
where Rg(τ) denotes the autocorrelation function of energy signal g(t) and Ψ g ( f ) denotes the
corresponding energy spectral density. Differentiating Rg(τ) with respect to the lag τ changes Eq.
(1) into the form
dR g ( τ )
----------------- j2πfΨ g ( f ) (2)
dτ
Next, we apply Rayleigh’s energy theorem, which, in the context of the problem at hand, yields
∞ dR ( τ ) 2 ∞
∫–∞ ----------------- dτ = ∫ j2πfΨ g ( f ) d f
g 2
dτ –∞

2 ∞ 2
∫–∞ f
2
= 4π Ψg( f )d f (3)
Since g(t) is real valued, Rg(τ) will likewise be real valued; hence we may write
dR g ( τ ) 2 dR g ( τ ) 2
----------------- = ----------------
dτ
-
dτ
Moreover, by definition,
2 2
Ψg( f ) = G( f )
where G(f) is the Fourier transform of g(t). Accordingly, we may rewrite Eq. (3) as follows:
∞ dR g ( τ ) 2 2 ∞
∫–∞ dτ ∫–∞ G ( f ) d f
4
----------------
- d τ = 4π (4)

which is the desired result.

dR g ( τ ) 2
Note: In the first publication of the book, the power 2 in ----------------- was missed out in error.
dτ
Problem 2.41

To determine the cross-correlation function R12(τ) of the two pulses g1(t) and g2(t) in Fig. 2.41,
we proceed on a stage-by-stage basis as follows:

1. For τ = 0, we have
∞
R 12 ( 0 ) = ∫–∞ g 1 ( t )g 2 ( t ) dt
–1 +1 +3
= ∫–3 ( 1.0 ) ( – 1.0 ) dt + ∫ ( 1.0 ) ( 1.0 ) dt + ∫ ( 1.0 ) ( – 1.0 ) dt
–1 –3
= – 2 + 2 – 2 = –2

2. For 0 < τ < 2.0, we have the picture depicted in Fig. 1, from which we obtain

IADC 38
 3
R 12 ( τ ) = ∫1+τ ( 1 ) ( –1 ) dt
= τ – 2, 0<τ<2
g1(t)
1.0

t
-3 0 3

g2(t − τ)

t
0
τ

Figure 1: 0 < τ < 2.0

3. Next, for 2 < τ < 4 we have the picture depicted in Fig. 2, from which we obtain
g1(t)
1.0

t
-3 3

g2(t − τ)

+1.0
t

Figure 2 −1.0

– 1+τ 3
R 12 ( τ ) = ∫–3+τ ( 1 ) ( –1 ) dt + ∫–1+τ ( 1 ) ( 1 ) dt
= – 2 + (4 – τ)
= 2 – τ, 2<τ<4

4. For 4 < τ < 6 we have the picture depicted in Fig. 3, from which we obtain:
g1(t)
1.0

t
-3 3

τ
g2(t - τ)

1.0
t
Figure 3 -1.0

IADC 39
 3
R 12 ( τ ) = ∫–3+τ ( 1 ) ( –1 ) dt
= τ – 6, 4<τ<6

5. Finally, for τ > 6 we find that R12(τ) is zero.

Putting all these results together, we get the cross-correlation function R12(τ) plotted in Fig. 4.

R12(τ)

-6 -4 -2 0 2 4 6
τ

-2
Figure 4

For the problem at hand, R12(τ) is symmetric about the origin τ = 0. Hence, the pulses defined in
Fig. 2.46 satisfy the property
R 12 ( τ ) = R 21 ( τ )
Moreover, the fact that R12(0) is nonzero implies that the pulses g1(t) and g2(t) are not orthogonal.

Problem 2.42

The convolution of energy signals g1(t) and g2(t) is defined by

∞
g 1 ( t )★g 2 ( t ) = ∫–∞ g 1 ( τ )g 2 ( t – τ ) dt
With g1(t) delayed by t1 and g2(t) delayed by t2, we have
∞
g 1 ( t – t 1 )★g 2 ( t – t 2 ) = ∫–∞ g 1 ( τ – t 1 )g 2 ( t – t 2 – τ ) dτ (1)
The cross-correlation of g1(t) and g2(t) is defined by
∞
R 12 ( τ ) = ∫–∞ g 1 ( t )g 2 ( t – τ ) dt
When g1(t) is delayed by t1 and g2(t) is delayed by t2, we have
∞
R 12 ( τ ) = ∫–∞ g 1 ( t – t 1 )g 2 ( t – t 2 – τ ) dt (2)
Comparing Eqs. (1) and (2), we make the following observations:
1. The convolution integral in Eq. (1) involves the dummy variable τ as the integration variable,
whereas the cross-correlation function of Eq. (2) involves time t as the integration variable.
2. By virtue of the difference identified under point 1, the arguments of g1 and g2 in Eq. (1) are
additive, in which case the time delays t1 and t2 are additive. On the other hand, the argument
of g2 is subtracted from the argument of g2 in Eq. (2); hence, the time delays t1 and t2 are
subtractive.

IADC 40
Problem 2.43

(a) The relationship between the Fourier transform X(f) of an energy signal g(t), its
autocorrelation function Rx(τ), and energy spectral density Ψ x ( f ) is illustrated by the flow-
graph in Fig. 1.
∞
∫ x(t)x*(t-τ)dt
x(t) −∞ Rx(τ)

F[.]
F-1[.]

X(f) F[.]

F[.] F-1[.] | . |2

Ψx(f) = |X(f)|2

Figure 1

(b) One way of calculating the autocorrelation function Rx(τ) from the Fourier transform X(f) is to
proceed as follows (in accordance with Fig. 1):
• Take the squared magnitude of X(f), obtaining the energy spectral density Ψx(f) = |X(f)|2.
• Take the inverse Fourier transform of Ψx(f) to obtain the autocorrelation function Rx(τ) as
desired.
Another way of calculating Rx(τ) from X(f) is to proceed as follows (again in accordance with
Fig. 1:
• Use the inverse Fourier transform to calculate x(t) from X(f).
• Then use the formula
∞
Rx(τ) = ∫–∞ x ( t )x* ( t – τ ) dt
to calculate Rx(τ) for prescribed lag τ.

Problem 2.44

The autocorrelation function of a power signal is the inverse Fourier transform of the power
spectral density. Given the power spectral density of Fig. 2.47, we write
∞
Rx(τ) = ∫–∞ S x ( f ) exp ( j2πfτ ) dτ
–1 1
= ∫–2 1 ⋅ exp ( j2πfτ ) dτ + ∫–1 2 ⋅ exp ( j2πfτ ) dτ
2
+ ∫ 1 ⋅ exp ( j2πfτ ) dτ
1
1 –1 –1 2
= ------------ { exp ( j2πfτ ) f =-2 + 2 exp ( j2πfτ ) f =-1 + exp ( j2πfτ ) f =1 }
j2πτ
1
= ------------{ exp ( – j 2πfτ ) – exp ( – j 4πfτ ) + 2 [ exp ( j2πτ ) – exp ( – j 2πτ ) ]
j2πτ

IADC 41
 + exp ( j4πτ ) – exp ( j2πτ )}
1
= ------ { sin ( 4πτ ) – sin ( 2πτ ) + 2 sin ( 2πτ ) }
πτ
1
= ------ { sin ( 4πτ ) + sin ( 2πτ ) }
πτ
The value of the autocorrelation function Rx(τ) at τ = 0 is given by
sin ( 4πτ ) sin ( 2πτ )
R x ( 0 ) = lim 4 ⋅ ---------------------- + 2 ⋅ ----------------------
τ→0 4πτ 2πτ
= 4+2 = 6 (1)
As a check, Rx(0) equals the total area under the power spectral density Sx(f). For the example of
Fig. 2.47, we write
Rx(0) = x + 2 × 2 + x
= 6
which checks the result computed in Eq. (1).

Problem 2.45

To proceed with the Fourier series expansion of the square wave specified in Fig. 2.48, we note
the following by inspection of the square wave:
• The square wave is symmetric about the origin t = 0.
• The average value is therefore zero.
• The expansion consists of cosine terms only.
Accordingly, we may write
∞
g(t) = ∑ an cos ( 2πn f 0 t )
n=1
where f0 = 1/T0 = 1/4. The coefficient an is defined by
1 T0 ⁄ 2
a n = --- ∫ g ( t ) cos ( 2πn f 0 t ) dt
T –T 0 ⁄ 2
πnt πnt
= --- ∫ ( – 1 ) cos  -------- dt + --- ∫ ( 1 ) cos  -------- dt
1 –1 1 1
4 –2  2  4 –1  2 
πnt
+ --- ∫ ( – 1 ) cos  -------- dt
1 2
4 1  2 
πnt πnt
= --- ∫ cos  -------- dt – ∫ cos  -------- dt
1 1 2
2 0  2  1  2 
πnt 1 πnt 2
= --- ⋅ ------ sin  -------- – sin  --------
1 2
2 nπ  2  t=0
 2  t=1

= ------ 2 sin  ------ – sin ( nπ )

1 nπ
nπ  2
Since sin(nπ) = 0 for integer values of n, the formula for an reduces to

IADC 42
a n = ------ sin  ------
2 nπ
nπ  2 

= sin c  --- ,
n
n = 1,3,5,...
 2
The Fourier series expansion of the square wave is thus defined by
2 1 1
g ( t ) = --- cos ( 2π f 0 t ) – --- cos ( 6π f 0 t ) + --- cos ( 10π f 0 t ) – …
π 3 5
Correspondingly, the Fourier transform of the square wave is defined by
1 1 1
G ( f ) = --- δ ( f – f 0 ) + δ ( f + f 0 ) – --- δ ( f – 3 f 0 ) – --- δ ( f + 3 f 0 )
π 3 3
1 1
+ --- δ ( f – 5 f 0 ) + --- δ ( f – 5 f 0 ) – … (1)
5 5
Next, we make use of the following formula. Given a periodic signal with the Fourier series
expansion
∞
G( f ) = ∑ cn [ δ ( f – n f 0 ) + δ ( f – n f 0 ) ] (2)
n=-∞
we may express the power spectral density of the signal by (see the solution to Problem 2.46)
∞

∑
2
S g( f ) = cn [ δ ( f – n f 0 ) + δ ( f – n f 0 ) ] (3)
n=-∞
Hence, in light of Eqs. (2) and (3), we may use Eq. (1) pertaining to the square wave of Fig. 2.48
to write
1 1 1
S g ( f ) = ----2-[δ ( f – f 0 ) + δ ( f + f 0 ) – --- δ ( f – 3 f 0 ) – --- δ ( f + 3 f 0 )
π 9 9
1 1
+ ------ δ ( f – 5 f 0 ) + ------ δ ( f + 5 f 0 ) – …] (4)
25 25
The inverse Fourier transform of Sg(f) defined in Eq. (4) yields the autocorrelation function
2 1 1
R x ( τ ) = ----2- cos ( 2π f 0 t ) – --- cos ( 6π f 0 t ) + ------ cos ( 10π f 0 t ) – … (5)
π 9 25
which follows the symmetric triangular waveform plotted in Fig. 1.
Rx(τ)
1.0

-4 -3 -2 -1 1 2 3 4

1.0
Figure 1

IADC 43
The average power of the square wave is defined by
1 T0 ⁄ 2 2
P g = ------ ∫ g ( t ) dt
T 0 –T 0 ⁄ 2
1 2
∫–2 ( 1 )
2
= --- dt
4
= 1

Additional Note: A follow-up exercise for the reader is to demonstrate that the triangular wave of
Fig. 1 has the Fourier series representation given in Eq. (5); see Eq. (4) of the solution to Problem
2.49.

Problem 2.46

We are given the two periodic signals

∞

∑ c2, n exp  – --------------

-
j2πnt
g p1 ( t ) =
T0 
n=-∞
∞

∑ c2, n exp  – --------------

-
j2πnt
g p2 ( t ) =
T0 
n=-∞
Applying the formula for autocorrelation function
1 T0 ⁄ 2
R 12 ( τ ) = ------ ∫
*
g p1 ( t )g p2 ( t – τ ) dt
T 0 –T 0 ⁄ 2
we get
∞ ∞
1 T0 ⁄ 2  
c 1, n exp  – --------------- c 2, n exp  --------------- ( t – τ ) dt
j2πnt j2πnt
R 12 ( τ ) = ------ ∫ ∑ ∑
*
T 0 –T 0 ⁄ 2   T0   T0 

n=-∞ m=-∞
∞ ∞
T0 ⁄ 2
exp  --------- ( m – n )t dt
1 j2π
∑ ∑ c 1, n c 2, n exp ( – j2πmτ ⁄ T 0 ) ∫
*
= ------ (1)
T0 –T 0 ⁄ 2  T0 
m=-∞ m=-∞
Consider the integral
1 T0 ⁄ 2
exp  --------- ( m-n )t dt = ------------------------ [ exp ( jπ ( m-n ) ) – exp ( – jπ ( m-n ) ) ]
j2π 1
------ ∫
T 0 –T 0 ⁄ 2  T0  j2π ( m-n )
1
= ------------------ sin ( π ( m-n ) )
π ( m-n )
= sin c ( m-n )
 for m = n 
=  1 (2)
 2 otherwise 
Using Eq. (2) in (1), we get the simplified result

IADC 44
 ∞

∑ c1, n c2, n exp ( – j2πnt ⁄ T 0 )

*
R 12 ( τ ) = (3)
n=-∞
Finally, taking the Fourier transform of R12(τ) defined in Eq. (3), we get
S 12 ( f ) = F [ R 12 ( τ ) ]
∞

∑ c1, n c2, n δ  f – -----

-
* n
=
T 0
n=-∞
We may thus finally write
∞

∑ c1, n c2, n δ  f – T-----0-

* n
R 12 ( τ ) (4)
n=-∞
Note: For the special case of gp1(t) = gp2(t) = gp(t), Eq. (4) reduces to the simple form
∞
c n δ  f – ------
n
∑
2
R 12 ( τ )
 T 0
n=-∞

Problem 2.47

(a) We are given the autocorrelation formula

1 T0 ⁄ 2
R g p ( τ ) = ------ ∫
*
g p ( t )g p ( t – τ ) dt (1)
T 0 –T 0 ⁄ 2
For the problem at hand, gp(t) is defined by
g p ( t ) = A cos ( 2π f c t + θ ) (2)
The use of Eq. (2) in (1) yields
2
A T0 ⁄ 2
R g p ( τ ) = ------ ∫ cos ( 2π f c t + θ ) cos ( 2π f c t – 2π f c τ + θ ) dt
T 0 –T 0 ⁄ 2
2
A T0 ⁄ 2
= --------- ∫ cos ( 4π f c t – 2π f c τ + 2θ ) + cos ( 2π f c τ ) dt
2T 0 –T 0 ⁄ 2
2
A
= ------ cos ( 2π f c τ ) (3)
2
The waveform of R g p ( τ ) is plotted in Fig. 1. Examining this figure, we observe two important
points:
1. Information on the phase θ of the sinusoidal wave gp(t) is lost in calculating the
autocorrelation function.

IADC 45
2. The autocorrelation function R g p ( τ ) has the same waveform as the input sinusoidal signal
gp(t).
Rg (τ)
p
A/2

−A2
2
Figure 1

(b) From Fig. 1, we readily find that

2
A
R g p ( 0 ) = ------
2
which is an intuitively satisfying result.

Problem 2.48

(a) The square wave gp(t) has the waveform plotted in Fig. 1.
gp(t)

t
T0 T 0 0 T0 T0
- - - -
2 4 2 2

Figure 1
To apply the formula for the autocorrelation function
1 T0 ⁄ 2
R g p ( τ ) = ------ ∫
*
g ( t )g p ( t – τ ) dt (1)
T 0 –T 0 ⁄ 2 p
we proceed in stages, as pictured in Fig. 2.
1. For τ = 0. gp(t) and g *p ( t – τ ) overlap each other completely. Hence,
1 T0 ⁄ 4 2
R 12 ( 0 ) = ------ ∫ A dt
T 0 –T 0 ⁄ 4
2
A
= ------ (2)
2
2. For 0 < τ < T0/4, we find from Fig. 2(a) that
1 T0 ⁄ 4 2
R g p ( τ ) = ------ ∫ T A dt
T 0 – -----0- +τ
4

IADC 46
 2
A T 0 
= ------ ------ – τ (3)
T 0 2 
3. For T0/4 < τ < T0/2, we find from Fig. 2(b) that
1 –T 0 ⁄ 4 2
R g p ( τ ) = ------ ∫ 3T A dt
T 0 – --------0- +τ
4
2
A T
= ------  -----0- – τ (4)
T 0 2 

g(t)
A

t
-T0/2 -T0/4 0 T0/4 T0/2

g(t - τ)
τ A

t
-T0/2 0 T0/2

(a) 0 < τ < T0/4

g(t)
A

t
-T0/2 -T0/4 0 T0/4 T0/2

g(t - τ)
τ

t
0

Figure 2 (b) T0/4 < τ < T0/2

Piercing these results together, we get the picture depicted in Fig. 3 for R g p ( τ ) , where we
have used the following facts:
1. R g p ( τ ) is periodic with exactly the same period as the square wave gp(t).
2. R g p ( τ ) is an even function of the lag τ.
Rg (τ)
p

A2/2

A2/4

τ
3T0 T0 T0 T 0 T0 T0 T0 3T0
- - - 0
2 2 4 4 2 2
Figure 3

IADC 47
Notes on Problems 2.47 and 2.48
In defining the formula for the autocorrelation function R g p ( τ ) for a periodic signal gp(t) in
Problem 2.47, the gp(t) is formulated in its most general form: complex fourier series expansion,
assuming that gp(t) is complex valued. However, in presenting the solutions to Problems 2.47 and
2.48, we did not make use of the complex Fourier series for gp(t). The reason for doing so is two-
fold:
• The sinusoidal wave in Problem 2.47 and square wave in Problem 2.48 are real valued.
• The solutions were handled by making direct use of the rather simple forms of both input
signals gp(t).

Problem 2.49

We first note that for periodic signals, the power spectral density and autocorrelation function
form a Fourier-transform pair:
Rg p(τ) S g p( f )

(a) For the sinusoidal wave

g p ( t ) = A cos ( 2π f c t + θ )
the autocorrelation function was derived in the solution to Problem 2.47 and repeated here for
convenience of presentation:
2
A
R g p ( τ ) = ------ cos ( 2π f c τ ) (1)
2
Hence, applying the Fourier transform to R g p ( τ ) yields the power spectral density
2
A
S g p ( f ) = ------ [ δ ( f – f c ) + δ ( f + f c ) ] (2)
4
(b) For the square wave of Problem 2.48, the autocorrelation function R g p ( τ ) is pictured in Fig. 3
in the solution to that problem. Specifically, one period of R g p ( τ ) is defined by

 2
 A T0 T0
 -------------  1 – -----
2τ
- for 0 < τ < ------
 4  T 0 2
Rg p(τ) =  (3)
 A T 2
T0
 ------------0-  1 + -----
2τ
- for – ------ < τ < 0
 4  T 0 2

From the Fourier-transform pairs table of Appendix 6, we find that the Fourier transform of
the triangular pulse
 t
 ----, t ≤T
g(t) =  1 – T

 0, t ≥0

IADC 48
 is Tsinc2(fT). Adapting this pair to the problem at hand, we may state that the Fourier
transform of the triangular pulse defined in Eq. (3) is
2 2
A T0 2 f T0
G ( f ) = ------------- sin c  ----------
8  2 
Finally, applying Eq. (2.88) in the text book to the problem at hand, we get the desired result
∞ 2 2
A T 2 n f 0 T 0
S g p( f ) = f 0 ∑ ------------0- sin c  ---------------
- δ( f – n f 0)
8  2 
n=-∞
2 2 ∞
A T0
sin c  --- δ ( f – n f 0 )
2 n
= -------------
8 ∑  2
(4)
n=-∞

Problem 2.50

(a) In the flow graph of Fig. 2.47 in the text book for the 8-point FFT algorithm, the incoming
data sequence gn is in normal order, but the transform sequence produced by the algorithm is
in bit-reversed order. In Fig. 1, we show another implementation of the FFT algorithm, in
which the transform sequence is in normal order, but the incoming data sequence is in bit-
reversed order. In other words, the new flow graph of Fig. 1 portrays the decimation-in-time
version of the FFT algorithm.
The flow graph of Fig. 1 shown here is obtained from the flow graph of Fig. 2.46 in the
text as follows:
• All the nodes that are horizontally adjacent to G4 in Fig. 2.46 are interchanged with all the
nodes that are horizontally adjacent to G1.
• In a similar way, all the nodes that are horizontally adjacent to G6 in Fig. 2.46 are
interchanged with the nodes that are horizontally adjacent to G3.
• The nodes that are horizontally adjacent to G0, G2, G5 and G7 are left unchanged.

(b) Comparing the decimation-in-frequency version of the FFT algorithm in Fig. 2.46 of the text
with the decimation-in-time version of the algorithm presented here as Fig. 1, we may make
the following observations:
• In the decimation-in-frequency version, the input data sequence is in normal order and the
output transform sequence is in bit-reversed order. The reverse applies to the decimation-
in-time version: the input data sequence is in bit-reversed order and the output transform
sequence is in normal order.
• Both versions of the FFT algorithm contain the same number of butterflies, but in reversed
order.
• The computational complexity is the same for both versions of the algorithm.

IADC 49
 In the final analysis, it is merely a matter of choice as to which particular version of the FFT
algorithm is used to compute the discrete Fourier transform in practice.
Date sequence Transform sequence

g0 o o o o o G0

W0

g4 o o o o o G1
W4

W0

g2 o o o o o G2
W4

W2 W0

g6 o o o o o G3
W6 W4

g1 o o o o o G4
-1

W1

g5 o o o o o G5
W5 -1
W2

g3 o o o o o G6
W6 -1

W3 W2

g7 o o o o o G7
W7 W6 -1
Stage I Stage II Stage III

Figure 1: Decimation-in-time FFT algorithm

Problem 2.51

(a) According to Schwarz’s inequality (see Appendix 5)

2
 ∞  ∞ ∞
 ∫ [ g 1 ( t )g 2 ( t ) + g 1 ( t )g 2 ( t ) ] dt  ≤ 4 ∫ g 1 ( t ) dt ∫ g 2 ( t ) dt
* * 2 2
(1)
 –∞  –∞ –∞

Define
g 1 ( t ) = tg ( t )
dg ( t )
g 2 ( t ) = -------------
dt
Then, for the left-hand side of the inequality (1) we may write

IADC 50
 *
∞ dg ( t ) d g (t ) ∞ d 2
∫– ∞ ∫–∞ dt
*
t g ( t ) ------------
- + tg ( t ) ---------------
- d t = t ---
- g ( t ) dt
dt dt
Integrating by parts:
∞ d 2 2 ∞ ∞
∫–∞ dt ∫–∞ g ( t ) dt
2
t ---- g ( t ) d t = [ g ( t ) ] – ∞ –

Assuming that g ( t ) → 0 faster than 1 ⁄ t as t → ∞ , we get

∞ d 2 ∞
∫–∞ dt ∫–∞ g ( t ) dt
2
t ---- g ( t ) d t =
Therefore, using Schwarz’s inequality we have
∞ 2 ∞ ∞
2 dg ( t ) 2 dt
∫–∞ ≤ 4∫ t g ( t ) dt ∫
2 2
g ( t ) dt -------------
–∞ –∞ dt
We now recognize that
dg ( t )
------------- j2πfG ( f ) from the differentiation property of the Fourier transform
dt
∞ ∞
∫–∞ ∫– ∞ G ( f )
2 2
g ( t ) dt = d f from Rayleigh’s energy theorem
Hence,
∞ 2 ∞ 2
∫–∞ t g ( t ) dt ∫–∞ ( 2πf ) G ( f ) d f 1
2 2

- ⋅ ---------------------------------------------------
---------------------------------
∞ ∞
- ≥ ---
4
∫ g ( t ) dt ∫ G( f ) d f
2 2
–∞ –∞
Using the root mean-square (rms) definitions of bandwidth and duration, we may write
2 2 2 1
4π T rms W rms ≥ ---
4
That is,
1
T rms W rms ≥ ------
4π

(b) For the Gaussian pulse g(t) = exp(-πt2), we have G(f) = exp(--πf2). Therefore,
∞ ∞
∫–∞ ∫–∞ exp ( –2πt
2 2
g ( t ) dt = ) dt

1 ∞
= ------- ∫ exp ( – 2πt ) dt
2
2 –∞
1
= -------
2
∞ 2 ∞
∫–∞ t ∫–∞ t
2 2 2
g ( t ) dt = exp ( – 2πt ) dt

t ∞ t ∞
+ ------ ∫ exp ( – 2πt ) dt
2 2
= – ------ exp ( – 2πt )
4π – ∞ 4π – ∞

IADC 51
 1
= --------------
4 2π
Therefore, we may write
2 1⁄2
T rms =  --------------
1
= ----------
 4 2π 2π
Similarly,
∞ 1
∫–∞ G ( f ) d f = ------2-
2

∞ 2 1
∫–∞ f G ( f ) d f = -------------
2
-
4 2π
1
W rms = ----------
2 π
We thus find that for the Gaussian pulse exp(-πt2),
1
T rms W rms = ------
4π
and the relation connecting Trms and Wrms is therefore satisfied with the equality sign.

Problem 2.52

sin t
(a) g ( t ) = ---------
t
The Hilbert transform of sint/t is
1 ∞ g(τ)
ĝ ( t ) = --- ∫ ----------- dτ
π –∞ t – τ
1 ∞ sin ( τ )
= --- ∫ ------------------ dτ
π –∞ τ ( t – τ )
1 ∞ 1
= ----- ∫  --- + ---------- sin τ dτ
1
πt –∞  τ t – τ
1 ∞ sin τ 1 ∞ sin τ
= ----- ∫ ---------- dτ + ----- ∫ ---------- dτ (10
πt –∞ τ πt –∞ t – τ
We now note that (see the mathematical tables of Appendix 6)
∞
∫–∞ sin c ( t ) dt = 1
Therefore, for the first integral of Eq. (1) we have
∞ sin τ
∫–∞ ---------
τ
- dτ = π
Next, for the second integral of Eq. (1) we write
∞ sin τ ∞ sin ( t – τ )
∫–∞ t – τ
---------
- d τ = ∫–∞ ----------------------
τ
- dτ

IADC 52
 cos τ
∞ sin ( τ )
= sin t ∫ ----------- dτ – cos t ∫∞ --------------- dτ
–∞ τ – ∞ τ
= – π cos t
We thus obtain the Hilbert transform
1
ĝ ( t ) = --- ( 1 – cos t )
t

(b) g ( t ) = rect ( t )

 1, t < ---
1
 2
= 
 0, t > ---
1
 2

The Hilbert transform of rect(t) is given by
1 1⁄2 1
ĝ ( t ) = --- P ∫ ---------- dτ
π –1 ⁄ 2 t – τ
where P denotes the “principal value of”. When t < -1/2 the singularity in the integrand is
below the range of integration and the significant values of t - τ are negative. We then have
1 1⁄2
ĝ ( t ) = – --- [ ln ( t – τ ) ] –1 ⁄ 2
π
t–1⁄2
= – --- ln  ------------------ ,
1 1
t < – --- (2)
π t+1⁄2  2
where ln denotes the natural
When t > 1/2, the singularity is above the range of integration and the significant values of
t - τ are positive. The corresponding value of ĝ ( t ) is
1 1⁄2
ĝ ( t ) = – --- [ ln ( t – τ ) ] –1 ⁄ 2
π
t–1⁄2
= – --- ln  ------------------ ,
1 1
t > ---
π  t + 1 ⁄ 2 2
1 1
For the case when – --- < t < --- , we write
2 2
1 t-ε dτ 1⁄2 dτ
ĝ ( t ) = --- lim
π ε→0 ∫–1 ⁄ 2 ------
t-τ ∫t-ε
-+ -------
t-τ
1 t-ε 1⁄2
= --- lim { [ – ln ( t – τ ) ] –1 ⁄ 2 + [ – ln ( t – τ ) ] t+ε }
π ε→0
ε t–1⁄2
= --- lim – ln  ------------------ – ln  – -----------------
1
π ε→0  t + 1 ⁄ 2  ε 

IADC 53
 1
1 --- – t 1 1
= – --- ln 2 – --- < t < ---
π ------------ , 2 2
(3)
1
--- + 6
2
We finally combine the results of Eqs. (2) and (3) by expressing the Hilbert transform ĝ ( t )
for all t as follows
1 t – (1 ⁄ 2)
ĝ ( t ) = – --- ln ----------------------
π t + ( 1 ⁄ 2 -)

(c) g ( t ) = δ ( t )
The Hilbert transform of the delta function δ(t) is
1 ∞ δ(τ)
ĝ ( t ) = --- ∫ ---------- dτ
π –∞ t – τ
Using the sifting property of the delta function, we get the desired result
1
ĝ ( t ) = -----
πt

1
(d) g ( t ) = ------------2-
1+t
The Hilbert transform of 1/(t + t2) is
1 ∞ 1
ĝ ( t ) = --- ∫ ----------------------------------
- dτ
π –∞ ( 1 + τ ) ( t – τ )
2

∞ t+τ
 -------------
- + ---------- dτ
1 1
= ---------------------
π(1 + t )
- ∫
2 –∞ 
1+τ
2 t – τ
(4)

But (see the mathematical tables of Appendix 6)

∞ 1
∫–∞ 1-------------
+τ
- dτ = π
2

∞ τ
∫–∞ 1-------------
+τ
- dτ = 0
2

∞ 1
∫–∞ --------------
(t – τ)
- dτ = 0

Therefore, Eq. (4) reduces to

t
ĝ ( t ) = -------------2-
1+τ

Problem 2.53

We are given the one-sided frequency function

IADC 54
  
 exp ( – f ), f >0

G ( f ) =  1--- , f= 0 (1)
 2
 0, f <0
 
Applying the inverse Fourier transform to Eq. (1) yields the corresponding time function
∞
g(t) = ∫–∞ G ( f ) exp ( j2πft ) d f
∞
= ∫0 exp ( – f ) exp ( j2πft ) d f
∞
= ∫0 exp ( f ( j2πt – 1 ) ) d f
1 ∞
= ------------------------- exp ( f ( j2πt – 1 ) ) f =0
– 1 + j2πt
1
= -------------------- (2)
1 – j2πt
Expressing g(t) in terms of its real and imaginary parts:
1 2πt
g ( t ) = ------------------------2- + j ------------------------2-
1 + ( 2πt ) 1 + ( 2πt )
Hence, for the real and imaginary parts of g(t) we may write
1
g r ( t ) = ------------------------2- (3)
1 + ( 2πt )
2πt
g i ( t ) = ------------------------2- (4)
1 + ( 2πt )
According to the last entry of the table in Problem 2.52, we find that t/(1 + t2) is the Hilbert
transform of 1/(1 + t2). From this pair, we readily see that the imaginary part gi(t) defined in Eq.
(4) is indeed the Hilbert transform of the real part gr(t) defined in Eq. (3).

Problem 2.54

From the hilbert transform

1 ∞ g(τ)
ĝ ( t ) = --- ∫ ----------- dτ
π –∞ t – τ
we first note that the Hilbert transform ĝ ( t ) is obtained by convolving the time function g(t) with
1/(πt). Next, we note that the Fourier transform 1/(πt) is defined by
1 1
F ----- = --- sgn ( f )
πt j
Accordingly, the process of Hilbert transformation is equivalent to passing the time function
through a linear time-invariant system whose transfer function is defined by

IADC 55
 1
H ( f ) = --- sgn ( f ) (1)
j
from which we readily find that
Ĝ ( f ) = H ( f )G ( f )
1
= --- sgn ( f )G ( f )
j
In particular, Eq. (1) shows that the Hilbert transformer satisfies the following pair of conditions:

(a) Magnitude response

H ( f ) = 1 for all f

(b) Phase response

 
 – 90 o for f > 0
arg [ H ( f ) ] = 
 +90 o for f < 0
 
which are the conditions specified under parts (a) and (b) of the problem statement.

To address part (c) of the statement, we note that the impulse response of the Hilbert transformer,
namely,
–1
h(t ) = F [ H ( f )]
1
= -----
πt
is noncausal. Specifically, it has the dependence on time pictured in Fig. 1. This figure shows that
h(t) is infinitely large at t = 0 and nonzero for t < 0. The impulse response h(t) is not physically
realizable for the simple reason that it requires the impulse response to be infinitely large at time
t = 0. Note also that it violates causality since h(t) is nonzero for negative time.
Another way of stressing the fact that the Hilbert transform is non-physically realizable is to
recognize that it is impossible to build a linear system that has a constant magnitude response for
all frequencies, and a constant phase response of -90o for all positive frequencies and +90o for
negative frequencies. We could approximately satisfy these two requirements over a finite band of
frequencies, but not for a frequency band of infinite extent.

IADC 56
IADC 57