103

Chapter 5

Continuous-Time Fourier Transform

5.1 Introduction
The Fourier series provides an extremely useful representation for periodic signals. Often, however, we need to deal
with signals that are not periodic. A more general tool than the Fourier series is needed in this case. In this chapter,
we will introduce a tool for representing arbitrary (i.e., possibly aperiodic) signals, known as the Fourier transform.

5.2 Development of the Continuous-Time Fourier Transform

As demonstrated earlier, the Fourier series is an extremely useful signal representation. Unfortunately, this signal
representation can only be used for periodic signals, since a Fourier series is inherently periodic. Many signals,
however, are not periodic. Therefore, one might wonder if we can somehow use the Fourier series to develop a
representation for aperiodic signals. As it turns out, this is possible. In order to understand why, we must make
the following key observation. An aperiodic signal can be viewed as a periodic signal with a period of infinity. By
viewing an aperiodic signal as this limiting case of a periodic signal where the period is infinite, we can use the Fourier
series to develop a more general signal representation that can be used in the aperiodic case. (In what follows, our
development of the Fourier transform is not completely rigorous, as we assume that various integrals, summations,
and limits converge. Such assumptions are not valid in all cases. Our development is mathematically sound, however,
provided that the Fourier transform of the signal being considered exists.)
Suppose that we have an arbitrary signal x(t) that is not necessarily periodic. Let us define the signal xT (t) as
(
x(t) for − T2 ≤ t < T2
xT (t) = (5.1)
0 otherwise.

In other words, xT (t) is identical to x(t) over the interval − T2 ≤ t < T2 and zero otherwise. Let us now repeat the
portion of xT (t) for − T2 ≤ t < T2 to form a periodic signal x̃(t) with period T . That is, we define x̃(t) as

x̃(t) = xT (t) for − T2 ≤ t < T

2 and x̃(t) = x̃(t + T ).

In Figures 5.1 and 5.2, we provide illustrative examples of the signals x(t), xT (t), and x̃(t).
Before proceeding further, we make two important observations that we will use later. First, from the definition of
xT (t), we have
lim xT (t) = x(t). (5.2)
T →∞

Second, from the definition of xT (t) and x̃(t), we have

lim x̃(t) = x(t). (5.3)

T →∞

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

104 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

Now, let us consider the signal x̃(t). Since x̃(t) is periodic, we can represent it using a Fourier series as

∞
x̃(t) = ∑ ak e jkω0 t , (5.4)
k=−∞

where
Z T /2
1
ak = x̃(t)e− jkω0 t dt (5.5)
T −T /2

and ω0 = 2π /T . Since xT (t) = x̃(t) for − T2 ≤ t < T2 , we can rewrite (5.5) as

Z T /2
1
ak = xT (t)e− jkω0 t dt.
T −T /2

Furthermore, since xT (t) = 0 for t < − T2 and t ≥ T2 , we can rewrite the preceding expression for ak as
Z ∞
1
ak = xT (t)e− jkω0 t dt.
T −∞

Substituting this expression for ak into (5.4) and rearranging, we obtain the following Fourier series representation for
x̃(t):

∞  Z ∞ 
1 − jkω0 τ
x̃(t) = ∑ xT (τ )e d τ e jkω0 t
k=−∞ T −∞
∞  Z 
ω0 ∞ − jkω0 τ
= ∑ 2π −∞ T x (τ )e d τ e jkω0 t
k=−∞
Z ∞ 
1 ∞ − jkω0 τ
= ∑ −∞ xT (τ )e d τ e jkω0 t ω0 .
2π k=−∞

Substituting the above expression for x̃(t) into (5.3), we obtain

∞ Z ∞

1
x(t) = lim ∑ xT (τ )e− jkω0 τ d τ e jkω0 t ω0 . (5.6)
T →∞ 2π −∞
k=−∞

Now, we must evaluate the above limit. As T → ∞, we have that ω0 → 0. Thus, in the limit above, kω0 becomes a
continuous variable which we denote as ω , ω0 becomes the infinitesimal d ω , and the summation becomes an integral.
This is illustrated in Figure 5.3. Also, as T → ∞, we have that xT (t) → x(t). Combining these results, we can rewrite
(5.6) to obtain
Z ∞
1
x(t) = X(ω )e jω t d ω ,
2π −∞

where
Z ∞
X(ω ) = x(t)e− jω t dt.
−∞

Thus, we have found a representation of the arbitrary signal x(t) in terms of complex sinusoids at all frequencies. We
call this the Fourier transform representation of the signal x(t).

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.2. DEVELOPMENT OF THE CONTINUOUS-TIME FOURIER TRANSFORM 105

x(t)

t
−T −T1 − T2 0 T
2 T1 T
(a)

xT (t)

t
−T −T1 − T2 0 T
2 T1 T
(b)

x̃(t)

··· ···

t
−T −T1 − T2 0 T
2 T1 T
(c)

Figure 5.1: Example of signals used in derivation of Fourier transform representation (where T1 > T2 ).

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

106 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

x(t)

t
− T2 −T1 0 T1 T
2

(a)

xT (t)

t
−T − T2 −T1 0 T1 T
2 T
(b)

x̃(t)

··· ···

t
−T −T1 0 T1 T
(c)

Figure 5.2: Example of signals used in derivation of Fourier transform representation (where T1 < T2 ).

R∞ − jωτ d τ
XT (ω ) = −∞ xT (τ )e

XT (kω0 )e jkω0 t

ω0
ω
kω0 (k + 1)ω0

Figure 5.3: Integral obtained in the derivation of the Fourier transform representation.

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.3. DEFINITION OF THE CONTINUOUS-TIME FOURIER TRANSFORM 107

5.3 Definition of the Continuous-Time Fourier Transform

In the previous section, we derived the Fourier transform representation of an arbitrary signal. This representation
expresses a signal in terms of complex sinusoids at all frequencies. Given a signal x(t), its Fourier transform repre-
sentation is given by
Z ∞
1
x(t) = X(ω )e jω t d ω , (5.7a)
2π −∞

where
Z ∞
X(ω ) = x(t)e− jω t dt. (5.7b)
−∞

We refer to (5.7b) as the Fourier transform analysis equation and (5.7a) as the Fourier transform synthesis equa-
tion.
The quantity X(ω ) is called the Fourier transform of x(t). That is, the Fourier transform of the signal x(t), denoted
as F {x(t)} or X(ω ), is defined as Z ∞
F {x(t)} = X(ω ) = x(t)e− jω t dt. (5.8)
−∞
Similarly, the inverse Fourier transform of X(ω ) is given by
Z ∞
1
F −1
{X(ω )} = x(t) = X(ω )e jω t d ω . (5.9)
2π −∞

If a signal x(t) has the Fourier transform X(ω ) we often denote this as
F
x(t) ←→ X(ω ).

As a matter of terminology, x(t) and X(ω ) are said to be a Fourier transform pair.
Example 5.1 (Fourier transform of the unit-impulse function). Find the Fourier transform X(ω ) of the signal x(t) =
Aδ (t − t0 ). Then, from this result, write the Fourier transform representation of x(t).
Solution. From the definition of the Fourier transform, we can write

X(ω ) = F {x(t)}
= F {Aδ (t − t0 )}
Z ∞
= Aδ (t − t0 )e− jω t dt
−∞
Z ∞
=A δ (t − t0 )e− jω t dt.
−∞

Using the sifting property of the unit-impulse function, we can simplify the above result to obtain

X(ω ) = Ae− jω t0 .

Thus, we have shown that

F
Aδ (t − t0 ) ←→ Ae− jω t0 .

From the Fourier transform analysis and synthesis equations, we have that the Fourier transform representation of x(t)
is given by
Z ∞
1
x(t) = X(ω )e jω t d ω
2π −∞

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

108 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

where

X(ω ) = Ae− jω t0 .

Example 5.2 (Inverse Fourier transform of the unit-impulse function). Find the inverse Fourier transform of X(ω ) =
2π Aδ (ω − ω0 ).
Solution. From the definition of the inverse Fourier transform, we can write
Z ∞
1
F −1 {2π Aδ (ω − ω0 )} = 2π Aδ (ω − ω0 )e jω t d ω
2π −∞
Z ∞
=A δ (ω − ω0 )e jω t d ω .
−∞

Using the sifting property of the unit-impulse function, we can simplify the preceding equation to obtain

F −1 {2π Aδ (ω − ω0 )} = Ae jω0 t .

Thus, we have that

F
Ae jω0 t ←→ 2π Aδ (ω − ω0 ).

Example 5.3 (Fourier transform of the rectangular pulse). Find the Fourier transform X(ω ) of the signal x(t) = rectt.
Solution. From the definition of the Fourier transform, we can write
Z ∞
X(ω ) = F {x(t)} = [rectt]e− jω t dt.
−∞

From the definition of the rectangular pulse function, we can simplify this equation as follows:
Z 1/2
X(ω ) = [rectt]e− jω t dt
−1/2
Z 1/2
= e− jω t dt.
−1/2

Evaluating the integral and simplifying, we obtain

  1/2
1
X(ω ) = − e− jω t
jω −1/2
1 jω /2
= [e − e− jω /2 ]
jω
1 ω
= [2 j sin ]
jω 2
2
= sin ω /2
ω
= sinc ω /2.

Thus, we have shown that

F
rectt ←→ sinc ω /2.

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.4. CONVERGENCE OF THE CONTINUOUS-TIME FOURIER TRANSFORM 109

5.4 Convergence of the Continuous-Time Fourier Transform

When deriving the Fourier transform representation earlier, we implicitly made some assumptions about the conver-
gence of the integrals and other expressions involved. These assumptions are not always valid. For this reason, a more
careful examination of the convergence properties of the Fourier transform is in order.
Suppose that we have an arbitrary signal x(t). This signal has the Fourier transform representation x̂(t) given by
Z ∞
1
x̂(t) = X(ω )e jω t d ω ,
2π −∞

where
Z ∞
X(ω ) = x(t)e− jω t dt.
−∞

Now, we need to concern ourselves with the convergence properties of this representation. In other words, we want to
know when x̂(t) is a valid representation of x(t). In our earlier derivation of the Fourier transform, we relied heavily
on the Fourier series. Therefore, one might expect that the convergence of the Fourier transform representation is
closely related to the convergence properties of Fourier series. This is, in fact, the case. The convergence properties of
the Fourier transform are very similar to the convergence properties of the Fourier series (as studied in Section 4.4).
The first important result concerning convergence relates to finite-energy signals as stated by the theorem below.
R∞ 2
Theorem 5.1 (Convergence of Fourier transform (finite-energy case)). If a signal x(t) is of finite energy (i.e., −∞ |x(t)| dt <
∞), then its Fourier transform representation converges in the MSE sense.

In other words, if x(t) is of finite energy, then

Z ∞
E= |x̂(t) − x(t)|2 dt = 0.
−∞

Although x(t) and x̂(t) may differ at individual points, the energy E in the difference is zero.
The other important result concerning convergence that we shall consider relates to what are known as the Dirichlet
conditions. The Dirichlet conditions for the signal x(t) are as follows:
R∞
1. The signal x(t) is absolutely integrable (i.e., −∞ |x(t)| dt < ∞).

2. The signal x(t) has a finite number of maxima and minima on any finite interval.

3. The signal x(t) has a finite number of discontinuities on any finite interval, and each discontinuity is itself finite.

For a signal satisfying the Dirichlet conditions, we have the important convergence result stated below.

Theorem 5.2 (Convergence of Fourier transform (Dirichlet case)). If a signal x(t) satisfies the Dirichlet conditions,
then its Fourier transform representation x̂(t) converges pointwise for all t, except at points of discontinuity. Further-
more, at each discontinuity point t = ta , we have that

x̂(ta ) = 12 [x(ta+ ) + x(ta− )],

where x(ta− ) and x(ta+ ) denote the values of the signal x(t) on the left- and right-hand sides of the discontinuity,
respectively.

In other words, if a signal x(t) satisfies the Dirichlet conditions, then the Fourier transform representation x̂(t) of
x(t) converges to x(t) for all t, except at points of discontinuity where x̂(t) instead converges to the average of x(t) on
the two sides of the discontinuity.
The finite-energy and Dirichlet conditions mentioned above are only sufficient conditions for the convergence of
the Fourier transform representation. They are not necessary. In other words, a signal may violate these conditions
and still have a valid Fourier transform representation.

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

110 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

x(t)

t
− 21 1
2

Figure 5.4: Signal x(t).

Example 5.4. Consider R ∞the function x(t) shown in Figure 5.4. Let x̂(t) denote the Fourier transform representation
of x(t) (i.e., x̂(t) = 21π −∞ X(ω )e jω t d ω , where X(ω ) denotes the Fourier transform of x(t)). Determine the values
x̂(− 12 ) and x̂( 12 ).

Solution. We begin by observing that x(t) satisfies the Dirichlet conditions. Consequently, Theorem 5.2 applies.
Thus, we have that
h i
− +
x̂(− 21 ) = 12 x(− 21 ) + x(− 12 ) = 21 (0 + 1) = 21 and
h i
− +
x̂( 12 ) = 12 x( 21 ) + x( 21 ) = 21 (1 + 0) = 12 .

5.5 Properties of the Continuous-Time Fourier Transform

The Fourier transform has a number of important properties. In the sections that follow, we introduce several of these
properties. For convenience, these properties are also later summarized in Table 5.1 (on page 123).

5.5.1 Linearity
F F
If x1 (t) ←→ X1 (ω ) and x2 (t) ←→ X2 (ω ), then
F
a1 x1 (t) + a2 x2 (t) ←→ a1 X1 (ω ) + a2 X2 (ω ),

where a1 and a2 are arbitrary complex constants. This is known as the linearity property of the Fourier transform.
To prove the above property, we proceed as follows:
Z ∞
F {a1 x1 (t) + a2 x2 (t)} = [a1 x1 (t) + a2 x2 (t)]e− jω t dt
−∞
Z ∞ Z ∞
= a1 x1 (t)e− jω t dt + a2 x2 (t)e− jω t dt
−∞ −∞
Z ∞ Z ∞
− jω t
= a1 x1 (t)e dt + a2 x2 (t)e− jω t dt
−∞ −∞
= a1 F {x1 (t)} + a2 F {x2 (t)}
= a1 X1 (ω ) + a2 X2 (ω ).

Thus, we have shown that the linearity property holds.

Example 5.5 (Linearity property of the Fourier transform). Find the Fourier transform X(ω ) of the signal x(t) =
A cos ω0t.

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.5. PROPERTIES OF THE CONTINUOUS-TIME FOURIER TRANSFORM 111

Solution. We recall that cos α = 21 [e jα + e− jα ] for any real α . Thus, we can write

X(ω ) = F {x(t)}
= F {A cos ω0t}
= F { A2 [e jω0 t + e− jω0 t ]}.

Then, we use the linearity property of the Fourier transform to obtain

X(ω ) = A2 F {e jω0 t } + A2 F {e− jω0 t }.

F F
From Example 5.2, we know that e jω0 t ←→ 2πδ (ω − ω0 ) and e− jω0 t ←→ 2πδ (ω + ω0 ). Thus, we can further simplify
the above expression for X(ω ) as follows:

X(ω ) = A2 [2πδ (ω + ω0 )] + A2 [2πδ (ω − ω0 )]

= Aπ [δ (ω + ω0 ) + δ (ω − ω0 )].

Thus, we have shown that

F
A cos ω0t ←→ Aπ [δ (ω + ω0 ) + δ (ω − ω0 )].

F 2
Example 5.6 (Fourier transform of the unit-step function). Suppose that sgnt ←→ jω . Find the Fourier transform
X(ω ) of the signal x(t) = u(t).

Solution. First, we observe that x(t) can be expressed in terms of the signum function as

x(t) = u(t) = 21 + 21 sgnt.

Taking the Fourier transform of both sides of this equation yields

X(ω ) = F {x(t)} = F { 21 + 12 sgnt}.

Using the linearity property of the Fourier transform, we can rewrite this as

X(ω ) = 21 F {1} + 12 F {sgnt}.

F F 2
From Example 5.5 (with ω0 = 0), we know that 1 ←→ 2πδ (ω ). Also, we are given that sgnt ←→ jω . Using these
facts, we can rewrite the expression for X(ω ) as

X(ω ) = 21 [2πδ (ω )] + 21 ( j2ω )

1
= πδ (ω ) + jω .

Thus, we have shown that

F 1
u(t) ←→ πδ (ω ) + jω .

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

112 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

5.5.2 Time-Domain Shifting

F
If x(t) ←→ X(ω ), then
F
x(t − t0 ) ←→ e− jω t0 X(ω ),
where t0 is an arbitrary real constant. This is known as the time-shifting property of the Fourier transform.
To prove the above property, we proceed as follows. To begin, we have
Z ∞
F {x(t − t0 )} = x(t − t0 )e− jω t dt.
−∞

Now, we use a change of variable. Let λ = t − t0 so that t = λ + t0 and dt = d λ . Performing the change of variable
and simplifying, we obtain
Z ∞
F {x(t − t0 )} = x(λ )e− jω (λ +t0 ) d λ
−∞
Z ∞
= x(λ )e− jωλ e− jω t0 d λ
−∞
Z ∞
= e− jω t0 x(λ )e− jωλ d λ
−∞
= e− jω t0 F {x(t)}
= e− jω t0 X(ω ).

Thus, we have proven that the time-shifting property holds.

Example 5.7 (Time-domain shifting property of the Fourier transform). Find the Fourier transform X(ω ) of the signal
x(t) = A cos(ω0t + θ ).

Solution. Let v(t) = A cos ω0t so that x(t) = v(t + ωθ0 ). From Example 5.5, we have that

F {v(t)} = V (ω ) = F {A cos ω0t}

= Aπ [δ (ω + ω0 ) + δ (ω − ω0 )].

From the definition of v(t) and the time-shifting property of the Fourier transform, we have

X(ω ) = F {x(t)}
= e jωθ /ω0 V (ω )
= e jωθ /ω0 Aπ [δ (ω + ω0 ) + δ (ω − ω0 )].

Thus, we have shown that

F
A cos(ω0t + θ ) ←→ Aπ e jωθ /ω0 [δ (ω + ω0 ) + δ (ω − ω0 )].

5.5.3 Frequency-Domain Shifting

F
If x(t) ←→ X(ω ), then
F
e jω0 t x(t) ←→ X(ω − ω0 ),
where ω0 is an arbitrary real constant. This is known as the frequency-domain shifting property of the Fourier trans-
form.

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.5. PROPERTIES OF THE CONTINUOUS-TIME FOURIER TRANSFORM 113

To prove the above property, we proceed as follows. From the definition of the Fourier transform and straightfor-
ward algebraic manipulation, we can write
Z ∞
F {e jω0 t x(t)} = e jω0 t x(t)e− jω t dt
−∞
Z ∞
= x(t)e− j(ω −ω0 )t dt
−∞
= X(ω − ω0 ).

Thus, we have shown that the frequency-domain shifting property holds.

Example 5.8 (Frequency-domain shifting property of the Fourier transform). Find the Fourier transform X(ω ) of the
signal x(t) = (cos ω0t)(cos 20π t).

Solution. Recall that cos α = 12 [e jα + e− jα ] for any real α . Using this relationship and the linearity property of the
Fourier transform, we can write

X(ω ) = F {x(t)}
= F {(cos ω0t)( 21 )[e j20π t + e− j20π t ]}
= F { 12 e j20π t cos ω0t + 21 e− j20π t cos ω0t}
= 12 F {e j20π t cos ω0t} + 12 F {e− j20π t cos ω0t}.

F
From Example 5.5 (where we showed cos ω0t ←→ π [δ (ω − ω0 ) + δ (ω + ω0 )]) and the frequency-domain shifting
property of the Fourier transform, we have
1 1
X(ω ) = 2 [π [δ (v − ω0 ) + δ (v + ω0 )]]|v=ω −20π + 2 [π [δ (v − ω0 ) + δ (v + ω0 )]]|v=ω +20π
1 1
= 2 (π [δ (ω + ω0 − 20π ) + δ (ω − ω0 − 20π )]) + 2 (π [δ (ω + ω0 + 20π ) + δ (ω − ω0 + 20π )])
π
= 2 [δ (ω + ω0 − 20π ) + δ (ω − ω0 − 20π ) + δ (ω + ω0 + 20π ) + δ (ω − ω0 + 20π )] .

5.5.4 Time- and Frequency-Domain Scaling

F
If x(t) ←→ X(ω ), then
F 1 ω 
x(at) ←→ X ,
|a| a
where a is an arbitrary nonzero real constant. This is known as the time/frequency-scaling property of the Fourier
transform.
To prove the above property, we proceed as follows. From the definition of the Fourier transform, we can write
Z ∞
F {x(at)} = x(at)e− jω t dt.
−∞

Now, we use a change of variable. Let λ = at so that t = λ /a and dt = d λ /a. Performing the change of variable (and
being mindful of the change in the limits of integration), we obtain
(R
∞
x(λ )e− j(ω /a)λ ( a1 )d λ for a > 0
F {x(at)} = R−∞ −∞ − j(ω /a)λ ( 1 )d λ
∞ x(λ )e a for a < 0
( R
1 ∞
x(λ )e− j(ω /a)λ d λ for a > 0
= a 1−∞ R∞ − j( ω /a) λ
− a −∞ x(λ )e d λ for a < 0.

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

114 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

Combining the two cases (i.e., for a < 0 and a > 0), we obtain
Z
1 ∞
F {x(at)} = x(λ )e− j(ω /a)λ d λ
|a| −∞
1 ω 
= X .
|a| a
Thus, we have shown that the time/frequency-scaling property holds.
Example 5.9 (Time scaling property of the Fourier transform). Find the Fourier transform X(ω ) of the signal x(t) =
rect(at).
Solution. Let v(t) = rect(t) so that x(t) = v(at). From Example 5.3, we know that

F {v(t)} = V (ω ) = F {rectt} = sinc ω /2. (5.10)

From the definition of v(t) and the time-scaling property of the Fourier transform, we have

X(ω ) = F {x(t)}
1 ω 
= V .
|a| a

Substituting the expression for V (ω ) in (5.10) into the preceding equation, we have
1 ω
X(ω ) = sinc .
|a| 2a
Thus, we have shown that
F 1 ω
rect(at) ←→ |a| sinc 2a .

5.5.5 Conjugation
F
If x(t) ←→ X(ω ), then
F
x∗ (t) ←→ X ∗ (−ω ).
This is known as the conjugation property of the Fourier transform.
A proof of the above property is quite simple. From the definition of the Fourier transform, we have
Z ∞
F {x∗ (t)} = x∗ (t)e− jω t dt.
−∞

From the properties of conjugation, we can rewrite this equation as

Z ∞ ∗ ∗
F {x∗ (t)} = x∗ (t)e− jω t dt
−∞
Z ∞ ∗
 ∗
= [x(t)∗ ]∗ e− jω t dt
−∞
Z ∞ ∗
= x(t)e jω t dt
−∞
∗
= X (−ω ).

Thus, we have shown that the conjugation property holds.

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.5. PROPERTIES OF THE CONTINUOUS-TIME FOURIER TRANSFORM 115

Example 5.10 (Fourier transform of a real signal). Show that the Fourier transform X(ω ) of any real signal x(t) must
satisfy X(ω ) = X ∗ (−ω ), and this condition implies that |X(ω )| = |X(−ω )| and arg X(ω ) = − arg X(−ω ) (i.e., |X(ω )|
and arg X(ω ) are even and odd functions, respectively).

Solution. From the conjugation property of the Fourier transform, we have

F {x∗ (t)} = X ∗ (−ω ).

Since x(t) is real, x(t) = x∗ (t) which implies

F {x(t)} = X ∗ (−ω ),

or equivalently

X(ω ) = X ∗ (−ω ). (5.11)

Taking the magnitude of both sides of (5.11) and observing that |z| = |z∗ | for any complex z, we have

|X(ω )| = |X ∗ (−ω )|
= |X(−ω )| .

Taking the argument of both sides of (5.11) and observing that arg z∗ = − arg z for any complex z, we have

arg X(ω ) = arg X ∗ (−ω )

= − arg X(−ω ).

5.5.6 Duality
F
If x(t) ←→ X(ω ), then
F
X(t) ←→ 2π x(−ω )
This is known as the duality property of the Fourier transform. This property follows from the similarity/symmetry
in the definition of the forward and inverse Fourier transforms. That is, the forward Fourier transform equation given
by (5.8) and the inverse Fourier transform equation given by (5.9) are identical except for a factor of 2π and different
sign in the parameter for the exponential function.
To prove the above property, we proceed as follows. From the Fourier transform synthesis equation, we have
Z ∞
1
x(t) = X(ω )e jω t d ω .
2π −∞

Substituting −t for t, we obtain

Z ∞
1
x(−t) = X(ω )e− jω t d ω .
2π −∞

Now, we multiply both sides of the equation by 2π to yield

Z ∞
2π x(−t) = X(ω )e− jω t d ω
−∞
= F {X(t)}.

Thus, we have shown that the duality property holds.

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

116 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

F
Example 5.11 (Fourier transform of the sinc function). Given that rect(t) ←→ sinc ω /2, find the Fourier transform
X(ω ) of the signal x(t) = sinc(t/2).

Solution. From the given Fourier transform pair and the duality property, we have that

X(ω ) = F {sinct/2} = 2π rect(−ω ).

Since rect(−ω ) = rect(ω ), we can simplify this to obtain

X(ω ) = 2π rect(ω ).

Thus, we have shown that

F
sinct/2 ←→ 2π rect ω .

5.5.7 Time-Domain Convolution

F F
If x1 (t) ←→ X1 (ω ) and x2 (t) ←→ X2 (ω ), then
F
x1 (t) ∗ x2 (t) ←→ X1 (ω )X2 (ω ).

This is known as the time-domain convolution property of the Fourier transform.

The proof of this property is as follows. From the definition of the Fourier transform and convolution, we have
Z ∞
F {x1 (t) ∗ x2 (t)} = [x1 (t) ∗ x2 (t)] e− jω t dt
−∞
Z ∞ Z ∞ 
= x1 (τ )x2 (t − τ )d τ e− jω t dt
−∞ −∞
Z ∞Z ∞
= x1 (τ )x2 (t − τ )e− jω t d τ dt.
−∞ −∞

Changing the order of integration, we obtain

Z ∞Z ∞
F {x1 (t) ∗ x2 (t)} = x1 (τ )x2 (t − τ )e− jω t dtd τ .
−∞ −∞

Now, we use a change of variable. Let λ = t − τ so that t = λ + τ and d λ = dt. Applying the change of variable and
simplifying, we obtain
Z ∞Z ∞
F {x1 (t) ∗ x2 (t)} = x1 (τ )x2 (λ )e− jω (λ +τ ) d λ d τ
−∞ −∞
Z ∞Z ∞
= x1 (τ )x2 (λ )e− jωλ e− jωτ d λ d τ
−∞ −∞
Z ∞ Z ∞ 
= x1 (τ )e− jωτ x2 (λ )e− jωλ d λ d τ
−∞ −∞
Z ∞  Z ∞ 
= x1 (τ )e− jωτ d τ x2 (λ )e− jωλ d λ
−∞ −∞
= F {x1 (t)}F {x2 (t)}
= X1 (ω )X2 (ω ).

Thus, we have shown that the time-domain convolution property holds.

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.5. PROPERTIES OF THE CONTINUOUS-TIME FOURIER TRANSFORM 117

Example 5.12 (Time-domain convolution property of the Fourier transform). With the aid of Table 5.2, find the
Fourier transform X(ω ) of the signal x(t) = x1 (t) ∗ x2 (t) where

x1 (t) = e−2t u(t) and

x2 (t) = u(t).

Solution. Let X1 (ω ) and X2 (ω ) denote the Fourier transforms of x1 (t) and x2 (t), respectively. From the time-domain
convolution property of the Fourier transform, we know that

X(ω ) = F {x1 (t) ∗ x2 (t)} = X1 (ω )X2 (ω ). (5.12)

From Table 5.2, we know that

X1 (ω ) = F {x1 (t)} = F {e−2t u(t)} = 1

2+ jω and
1
X2 (ω ) = F {x2 (t)} = F {u(t)} = πδ (ω ) + jω .

Substituting these expressions for X1 (ω ) and X2 (ω ) into (5.12), we obtain

X(ω ) = [ 2+1jω ](πδ (ω ) + 1

jω )
π 1 1
= 2+ jω δ (ω ) + jω ( 2+ jω )
π 1
= 2+ jω δ (ω ) + j2ω −ω 2
π 1
= 2 δ (ω ) + j2ω −ω 2 .

5.5.8 Frequency-Domain Convolution

F F
If x1 (t) ←→ X1 (ω ) and x2 (t) ←→ X2 (ω ), then
Z ∞
F 1 1
x1 (t)x2 (t) ←→ 2π X1 (ω ) ∗ X2 (ω ) = 2π
X1 (θ )X2 (ω − θ )d θ .
−∞

This is known as the frequency-domain convolution (or time-domain multiplication) property of the Fourier transform.
To prove the above property, we proceed as follows. From the definition of the inverse Fourier transform, we have
  Z  
1 1 ∞ 1
F −1 X1 (ω ) ∗ X2 (ω ) = X1 (ω ) ∗ X2 (ω ) e jω t d ω
2π 2π −∞ 2π
Z Z ∞ 
1 ∞ 1
= X1 (λ )X2 (ω − λ )d λ e jω t d ω
2π −∞ −∞ 2π
Z Z
1 ∞ ∞ 1
= X1 (λ )X2 (ω − λ )e jω t d λ d ω .
2π −∞ −∞ 2π

Reversing the order of integration, we obtain

  Z Z
1 1 ∞ ∞ 1
F −1
X1 (ω ) ∗ X2 (ω ) = X1 (λ )X2 (ω − λ )e jω t d ω d λ .
2π 2π −∞ −∞ 2π

Now, we employ a change of variable. Let v = ω − λ so that ω = v + λ and dv = d λ . Applying the change of variable

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

118 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

and simplifying yields

  Z Z
1 1 ∞ ∞ 1
F −1 X1 (ω ) ∗ X2 (ω ) = X1 (λ )X2 (v)e j(v+λ )t dvd λ
2π 2π −∞ −∞ 2π
Z Z
1 ∞ ∞ 1
= X1 (λ )X2 (v)e jvt e jλ t dvd λ
2π −∞ −∞ 2π
Z  Z∞ 
1 ∞ jλ t 1 jvt
= X1 (λ )e X2 (v)e dv d λ
2π −∞ 2π −∞
 Z∞  Z ∞ 
1 jλ t 1 jvt
= X1 (λ )e d λ X2 (v)e dv
2π −∞ 2π −∞
 Z∞  Z ∞ 
1 jω t 1 jω t
= X1 (ω )e d ω X2 (ω )e d ω
2π −∞ 2π −∞
= F −1 {X1 (ω )}F −1 {X2 (ω )}
= x1 (t)x2 (t).

Thus, we have shown that the frequency-domain convolution property holds.

Example 5.13 (Frequency-domain convolution property). Suppose that we have the signal

y(t) = x(t) cos ωct

where ωc is a nonzero real constant. Find the Fourier transform Y (ω ) of the signal y(t) in terms of X(ω ) = F {x(t)}.

Solution. Taking the Fourier transform of both sides of the above equation for y(t), we have

Y (ω ) = F {x(t) cos ωct}.

Using the frequency-domain convolution property of the Fourier transform, we can write

Y (ω ) = F {x(t) cos ωct}

1
= 2π F {x(t)} ∗ F {cos ωc t}
1
= 2π X(ω ) ∗ [π [δ (ω − ωc ) + δ (ω + ωc )]]
1
= 2 X(ω ) ∗ [δ (ω − ωc ) + δ (ω + ωc )]
1
= 2 [X(ω ) ∗ δ (ω − ωc ) + X(ω ) ∗ δ (ω + ωc )]
Z ∞ Z ∞ 
1
= 2 X(λ )δ (ω − λ − ωc )d λ + X(λ )δ (ω − λ + ωc )d λ
−∞ −∞
Z ∞ Z ∞ 
1
= 2 X(λ )δ (λ − ω + ωc )d λ + X(λ )δ (λ − ω − ωc )d λ
−∞ −∞
1
= 2 [X(ω − ωc ) + X(ω + ωc )]
1 1
= 2 X(ω − ωc ) + 2 X(ω + ωc ).

5.5.9 Time-Domain Differentiation

F
If x(t) ←→ X(ω ), then
dx(t) F
←→ jω X(ω ).
dt
This is known as the time-domain differentiation property of the Fourier transform.

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.5. PROPERTIES OF THE CONTINUOUS-TIME FOURIER TRANSFORM 119

To prove the above property, we begin by using the definition of the inverse Fourier transform to write
Z ∞
1
x(t) = X(ω )e jω t d ω .
2π −∞

Now, we differentiate both sides of the preceding equation with respect to t and simplify to obtain
Z
dx(t) 1 ∞
= X(ω )( jω )e jω t d ω
dt 2π −∞
= F −1 { jω X(ω )}.

Thus, we have shown that the time-differentiation property holds. In passing, we also note that by repeating the above
argument, we have
dn F
x(t) ←→ ( jω )n X(ω ).
dt n
d
Example 5.14 (Time-domain differentiation property). Find the Fourier transform X(ω ) of the signal x(t) = dt δ (t).

Solution. Taking the Fourier transform of both sides of the above equation for x(t) yields

X(ω ) = F { dtd δ (t)}.

Using the time-domain differentiation property of the Fourier transform, we can write

X(ω ) = F { dtd δ (t)}

= jω F {δ (t)}
= jω (1)
= jω .

5.5.10 Frequency-Domain Differentiation

F
If x(t) ←→ X(ω ), then
d
F
tx(t) ←→ j X(ω ).
dω
This is known as the frequency-domain differentiation property of the Fourier transform.
To prove the above property, we proceed as follows. From the definition of the Fourier transform, we can write
Z ∞
X(ω ) = x(t)e− jω t dt.
−∞

Now, we differentiate both sides of this equation with respect to ω and simplify to obtain
Z ∞
d
X(ω ) = x(t)(− jt)e− jω t dt
dω −∞
Z ∞
= −j tx(t)e− jω t dt
−∞
= − jF {tx(t)}.

Multiplying both sides of the preceding equation by j yields

d
j X(ω ) = F {tx(t)}.
dω
Thus, we have shown that the frequency-domain differentiation property holds.

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

120 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

Example 5.15 (Frequency-domain differentiation property). Find the Fourier transform X(ω ) of the signal x(t) =
t cos ω0t where ω0 is a nonzero real constant.

Solution. Taking the Fourier transform of both sides of the equation for x(t) yields

X(ω ) = F {t cos ω0t}.

From the frequency-domain differentiation property of the Fourier transform, we can write

X(ω ) = F {t cos ω0t}

= j ddω [F {cos ω0t}]
= j ddω [π [δ (ω − ω0 ) + δ (ω + ω0 )]]
= jπ ddω [δ (ω − ω0 ) + δ (ω + ω0 )]
= jπ ddω δ (ω − ω0 ) + jπ ddω δ (ω + ω0 ).

5.5.11 Time-Domain Integration

F
If x(t) ←→ X(ω ), then
Z t
F 1
x(τ )d τ ←→ X(ω ) + π X(0)δ (ω ).
−∞ jω
This is known as the time-domain integration property of the Fourier transform.
The above property can be proven as follows. First, we observe that
Z t
x(τ )d τ = x(t) ∗ u(t).
−∞

Taking the Fourier transform of both sides of the preceding equation and using the time-domain convolution property
of the Fourier transform, we have
Z t 
F x(τ )d τ = F {x(t) ∗ u(t)}
−∞
= X(ω )F {u(t)}. (5.13)

F 1
From Example 5.6, we know that u(t) ←→ πδ (ω ) + jω . Using this fact, we can rewrite (5.13) as
Z t

1
F x(τ )d τ = X(ω )[πδ (ω ) + jω ]
−∞
1
= jω X(ω ) + π X(ω )δ (ω ).

From the equivalence property of the unit-impulse function, we have

Z t 
F x(τ )d τ = j1ω X(ω ) + π X(0)δ (ω ).
−∞

Thus, we have shown that the time-domain integration property holds.

Example 5.16 (Time-domain integration property of the Fourier transform). Use the time-domain integration property
of the Fourier transform in order to find the Fourier transform X(ω ) of the signal x(t) = u(t).

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.5. PROPERTIES OF THE CONTINUOUS-TIME FOURIER TRANSFORM 121

Solution. We begin by observing that x(t) can be expressed in terms of an integral as follows:
Z t
x(t) = u(t) = δ (τ )d τ .
−∞

Now, we consider the Fourier transform of x(t). We have

X(ω ) = F {u(t)}
Z t 
=F δ (τ )d τ .
−∞

From the time-domain integration property, we can write

1
X(ω ) = F {δ (t)} + π [F {δ (t)}]|ω =0 δ (ω )
jω
1
= (1) + π (1)δ (ω )
jω
1
= + πδ (ω ).
jω
F 1
Thus, we have shown that u(t) ←→ jω + πδ (ω ).

5.5.12 Parseval’s Relation

R∞ 2 F
Recall that the energy of a signal f (t) is given by −∞ | f (t)| dt. If x(t) ←→ X(ω ), then
Z ∞ Z ∞
1
|x(t)|2 dt = |X(ω )|2 d ω . (5.14)
−∞ 2π −∞

That is, the energy of x(t) and energy of X(ω ) are equal within a scaling factor of 2π . This relationship is known as
Parseval’s relation.
To prove the above relationship, we proceed as follows. Consider the left-hand side of (5.14) which we can write
as
Z ∞ Z ∞
|x(t)|2 dt = x(t)x∗ (t)dt
−∞ −∞
Z ∞
= x(t)F −1 {F {x∗ (t)}}dt.
−∞

F
From the conjugation property of the Fourier transform, we have that x∗ (t) ←→ X ∗ (−ω ). So, we can rewrite the
above equation as
Z ∞ Z ∞
|x(t)|2 dt = x(t)F −1 {X ∗ (−ω )}dt
−∞ −∞
Z ∞  Z∞ 
1
= x(t) X ∗ (−ω )e jω t d ω dt.
−∞ 2π −∞

Now, we employ a change of variable (i.e., replace ω by −ω ) to obtain

Z ∞ Z ∞  Z∞ 
1
|x(t)|2 dt = x(t) X ∗ (ω )e− jω t d ω dt
−∞ −∞ 2π −∞
Z Z
1 ∞ ∞
= x(t)X ∗ (ω )e− jω t d ω dt.
2π −∞ −∞

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

122 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

Reversing the order of integration and simplifying, we have

Z ∞ Z Z
1 ∞ ∞
2
|x(t)| dt = x(t)X ∗ (ω )e− jω t dtd ω
−∞ 2π −∞ −∞
Z Z ∞ 
1 ∞ ∗
= X (ω ) x(t)e− jω t dt d ω
2π −∞ −∞
Z
1 ∞ ∗
= X (ω )X(ω )d ω
2π −∞
Z ∞
1
= |X(ω )|2 d ω .
2π −∞
Thus, Parseval’s relation holds.
Example 5.17 (Energy of the sinc signal). Suppose that we have the signal x(t) = sinct/2 and F {x(t)} = 2π rect ω .
Compute the energy of the signal x(t).
Solution. We could directly compute the energy of the signal x(t) as
Z ∞
E= |x(t)|2 dt
−∞
Z ∞
= |sinct/2|2 dt.
−∞

This integral is not so easy to compute, however. Instead, we use Parseval’s relation to write
Z
1 ∞
E= |X(ω )|2 d ω
2π −∞
Z
1 ∞
= |2π rect ω |2 d ω
2π −∞
Z
1 1/2
= (2π )2 d ω
2π −1/2
Z 1/2
= 2π dω
−1/2
1/2
= 2π [ω ]|−1/2
= 2π [ 21 + 21 ]
= 2π .
Thus, we have
Z ∞
2
E= sinc 2t dt = 2π .
−∞

5.6 Continuous-Time Fourier Transform of Periodic Signals

The Fourier transform can be applied to both aperiodic and periodic signals. In what follows, we derive a method for
computing the Fourier transform of a periodic signal.
Suppose that we have a periodic signal x(t) with period T . Since x(t) is periodic, we can express it using a Fourier
series as
∞
x(t) = ∑ ak e jkω0 t (5.15)
k=−∞

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.6. CONTINUOUS-TIME FOURIER TRANSFORM OF PERIODIC SIGNALS 123

Table 5.1: Fourier Transform Properties

Property Time Domain Frequency Domain
Linearity a1 x1 (t) + a2 x2 (t) a1 X1 (ω ) + a2 X2 (ω )
Time-Domain Shifting x(t − t0 ) e− jω t0 X(ω )
Frequency-Domain Shifting e jω0 t x(t) X(ω − ω0 )
1 ω


Time/Frequency-Domain Scaling x(at) |a| X a
Conjugation x∗ (t) X ∗ (−ω )
Duality X(t) 2π x(−ω )
Time-Domain Convolution x1 (t) ∗ x2 (t) X1 (ω )X2 (ω )
1
Frequency-Domain Convolution x1 (t)x2 (t) 2π X1 (ω ) ∗ X2 (ω )
d
Time-Domain Differentiation dt x(t) jω X(ω )
Frequency-Domain Differentiation tx(t) j ddω X(ω )
Rt 1
Time-Domain Integration −∞ x(τ )d τ jω X(ω ) + π X(0)δ (ω )

Property
R∞ 2 1 R∞ 2
Parseval’s Relation −∞ |x(t)| dt = 2π −∞ |X(ω )| d ω

where
Z
1
ak = x(t)e− jkω0 t dt. (5.16)
T T

Let us define the signal xT (t) to contain a single period of x(t) as follows:
(
x(t) for − T2 ≤ t < T2
xT (t) = (5.17)
0 otherwise.
Let XT (ω ) denote the Fourier transform of xT (t). Consider the expression for ak in (5.16). Since xT (t) = x(t) for a
single period of x(t) and is zero otherwise, we can rewrite (5.16) as
Z
1 ∞
ak = xT (t)e− jkω0 t dt
T −∞
1
= XT (kω0 ). (5.18)
T
Now, let us consider the Fourier transform of x(t). By taking the Fourier transform of both sides of (5.15), we
obtain
( )
∞
F {x(t)} = F ∑ ak e jkω0 t
k=−∞
" #
Z ∞ ∞
jkω0 t
= ∑ ak e e− jω t dt.
−∞ k=−∞

Reversing the order of summation and integration, we have

∞ Z ∞
F {x(t)} = ∑ ak e jkω0 t e− jω t dt
k=−∞ −∞
∞
= ∑ ak F {e jkω0 t }. (5.19)
k=−∞

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

124 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

x(t)

··· ···
t
− 23 − 21 1
2
3
2

Figure 5.5: Periodic function x(t).

From Example 5.2, we know that F {e jλ t } = 2πδ (ω − λ ). So, we can simplify (5.19) to obtain
∞
F {x(t)} = ∑ ak [2πδ (ω − kω0 )]
k=−∞
∞
= ∑ 2π ak δ (ω − kω0 ). (5.20)
k=−∞

Thus, the Fourier transform of a periodic function is a series of impulse functions located at integer multiples of
the fundamental frequency ω0 . The weight of each impulse is 2π times the corresponding Fourier series coefficient.
Furthermore, by substituting (5.18) into (5.20), we have
∞
F {x(t)} = ∑ 2π [ T1 XT (kω0 )]δ (ω − kω0 )
k=−∞
∞
= ∑ ω0 XT (kω0 )δ (ω − kω0 ). (5.21)
k=−∞

This provides an alternative expression for the Fourier transform of x(t) in terms of the Fourier transform XT (ω ) of a
single period of x(t).
In summary, we have shown that the periodic signal x(t) with period T and frequency ω0 = 2Tπ has the Fourier
transform X(ω ) given by
∞
X(ω ) = ∑ 2π ak δ (ω − kω0 ). (5.22)
k=−∞

(as in (5.20)), or equivalently,

∞
X(ω ) = ∑ ω0 XT (kω0 )δ (ω − kω0 ) (5.23)
k=−∞

(as in (5.21), where ak is the Fourier series coefficient sequence of x(t), XT (ω ) is the Fourier transform of xT (t), and
xT (t) is a function equal to x(t) over a single period and zero elsewhere (e.g., as in (5.17)). Furthermore, we have also
shown that the Fourier series coefficients ak of x(t) are related to XT (ω ) by the equation

1
ak = XT (kω0 ).
T
Thus, we have that the Fourier series coefficient sequence ak of the periodic signal x(t) is produced by sampling the
Fourier transform of xT (t) at integer multiples of the fundamental frequency ω0 and scaling the resulting sequence by
1
T.

Example 5.18. Consider the periodic function x(t) with period T = 2 as shown in Figure 5.5. Using the Fourier
transform, find the Fourier series representation of x(t).

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.7. FOURIER TRANSFORMS 125

Table 5.2: Fourier Transform Pairs

Pair x(t) X(ω )
1 δ (t) 1
1
2 u(t) πδ (ω ) + jω
3 1 2πδ (ω )
2
4 sgn(t) jω
5 e j ω0 t 2πδ (ω − ω0 )
6 cos ω0t π [δ (ω − ω0 ) + δ (ω + ω0 )]
π
7 sin ω0t j [δ (ω − ω0 ) − δ (ω + ω0 )]
8 rect(t/T ) T sinc(ω T /2)
B ω
9 π sinc Bt rect 2B
10 e−at u(t), Re{a} > 0 1
a+ jω
(n−1)!
11 t n−1 e−at u(t), Re{a} > 0 (a+ jω )n
T 2
12 tri(t/T ) 2 sinc (T ω /4)

2π
Solution. We have that ω0 = T = π . Let y(t) = rectt (i.e., y(t) corresponds to a single period of the periodic function
x(t)). Thus, we have that
∞
x(t) = ∑ y(t − 2k).
k=−∞

Taking the Fourier transform of y(t), we obtain

Y (ω ) = F {rectt} = sinc ω /2.
Now, we seek to find the Fourier series representation of x(t), which has the form
∞
x(t) = ∑ ck e jkω0 t .
k=−∞

Using the Fourier transform, we compute ck as follows:

ck = T1 Y (kω0 )
= 12 sinc( kω2 0 )
= 12 sinc(kπ /2).

5.7 Fourier Transforms

Throughout this chapter, we have derived a number of Fourier transform pairs. Some of these and other important
transform pairs are listed in Table 5.2. Using the various Fourier transform properties listed in Table 5.1 and the
Fourier transform pairs listed in Table 5.2, we can determine (more easily) the Fourier transform of more complicated
signals.
F F
Example 5.19. Suppose that x(t) ←→ X(ω ), x1 (t) ←→ X1 (ω ), and
d2
x1 (t) = x(t − 2).
dt 2
Express X1 (ω ) in terms of X(ω ).

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

126 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

Solution. Let v1 (t) = x(t − 2). From the definition of v1 (t) and the time-shifting property of the Fourier transform,
we have
V1 (ω ) = e− j2ω X(ω ). (5.24)
From the definition of v1 (t), we have
d2
x1 (t) = v1 (t).
dt 2
Thus, from the time-differentiation property of the Fourier transform, we can write
X1 (ω ) = ( jω )2V1 (ω )
= −ω 2V1 (ω ). (5.25)
Combining (5.24) and (5.25), we obtain
X1 (ω ) = −ω 2 e− j2ω X(ω ).

F F
Example 5.20. Suppose that x(t) ←→ X(ω ), x1 (t) ←→ X1 (ω ), and
x1 (t) = x(at − b),
where a is a nonzero real constant and b is a real constant. Express X1 (ω ) in terms of X(ω ).
Solution. We rewrite x1 (t) as
x1 (t) = v1 (at)

where

v1 (t) = x(t − b).

We now take the Fourier transform of both sides of each of the preceding equations. Using the time-shifting property
of the Fourier transform, we can write
V1 (ω ) = e− jbω X(ω ). (5.26)
Using the time-scaling property of the Fourier transform, we can write
1 ω 
X1 (ω ) = V1 . (5.27)
|a| a
Substituting the expression for V1 (ω ) in (5.26) into (5.27), we obtain
1 − j(b/a)ω  ω 
X1 (ω ) = e X .
|a| a

Example 5.21. Suppose that we have the periodic signal x(t) given by
∞
x(t) = ∑ x0 (t − kT )
k=−∞

where a single period of x(t) is given by

x0 (t) = A rect(2t/T ).
Find the Fourier transform X(ω ) of the signal x(t).

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.7. FOURIER TRANSFORMS 127

Solution. From (5.23), we know that

( )
∞
F {x(t)} = F ∑ x0 (t − kT )
k=−∞
∞
= ∑ ω0 X0 (kω0 )δ (ω − kω0 ).
k=−∞

So, we need to find X0 (ω ). This quantity is computed (by using the linearity property of the Fourier transform and
Table 5.2) as follows:

X0 (ω ) = F {x0 (t)}
= F {A rect(2t/T )}
= AF {rect(2t/T )}
AT
= 2 sinc(ω T /4).

Thus, we have that

∞
X(ω ) = ∑ ω0 AT
2 sinc(kω0 T /4)δ (ω − kω0 )
k=−∞
∞
= ∑ π A sinc( π2k )δ (ω − kω0 ).
k=−∞

Example 5.22. Suppose that we have the signal x(t) given by

Z t
x(t) = e−(3+ j2)τ u(τ )d τ .
−∞

Find the Fourier transform X(ω ) of the signal x(t).

Solution. We can rewrite x(t) as
Z t
x(t) = v1 (τ )d τ , (5.28)
−∞

where

v1 (t) = e− j2t v2 (t), and (5.29)

−3t
v2 (t) = e u(t). (5.30)

From (5.28) and the time-domain integration property of the Fourier transform, we have
1
X(ω ) = jω V1 (ω ) + π V1 (0)δ (ω ). (5.31)

From (5.29) and the frequency-domain shifting property of the Fourier transform, we have

V1 (ω ) = V2 (ω + 2). (5.32)

From (5.30) and Table 5.2 (i.e., the entry for F {e−at u(t)}), we have

1
V2 (ω ) = . (5.33)
3 + jω

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

128 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

Combining (5.31), (5.32), and (5.33), we obtain

1
X(ω ) = V1 (ω ) + π V1 (0)δ (ω )
jω
1
= V2 (ω + 2) + π V2 (2)δ (ω )
jω
   
1 1 1
= +π δ (ω ).
jω 3 + j(ω + 2) 3 + j2

Example 5.23. Let X(ω ) and Y (ω ) denote the Fourier transforms of x(t) and y(t), respectively. Suppose that y(t) =
x(t) cos at, where a is a nonzero real constant. Find an expression for Y (ω ) in terms of X(ω ).
Solution. Essentially, we need to take the Fourier transform of both sides of the given equation. There are two obvious
ways in which to do this. One is to use the time-domain multiplication property of the Fourier transform, and another
is to use the frequency-domain shifting property. We will solve this problem using each method in turn in order to
show that the two approaches do not involve an equal amount of effort.
F IRST SOLUTION ( USING AN UNENLIGHTENED APPROACH ). We use the time-domain multiplication property.
Taking the Fourier transform of both sides of the given equation, we obtain

Y (ω ) = F {x(t) cos at}

1
= 2π [X(ω ) ∗ F {cos at}]
1
= 2π [X(ω ) ∗ [π (δ (ω − a) + δ (ω + a))]]
1
= 2 [X(ω ) ∗ [δ (ω − a) + δ (ω + a)]]
1
= 2 [X(ω ) ∗ δ (ω − a) + X(ω ) ∗ δ (ω + a)]
Z ∞ Z ∞ 
1
= 2 X(τ )δ (ω − τ − a)d τ + X(τ )δ (ω − τ + a)d τ
−∞ −∞
1
= 2 [X(ω − a) + X(ω + a)]
1 1
= 2 X(ω − a) + 2 X(ω + a).

Note that the above solution is identical to the one appearing earlier in Example 5.13 on page 118.
S ECOND SOLUTION ( USING AN ENLIGHTENED APPROACH ). We use the frequency-domain shifting property.
Taking the Fourier transform of both sides of the given equation, we obtain

Y (ω ) = F {x(t) cos at}

= F { 21 (e jat + e− jat )x(t)}
= 12 F {e jat x(t)} + 21 F {e− jat x(t)}
= 12 X(ω − a) + 21 X(ω + a).

C OMMENTARY. Clearly, of the above two solution methods, the second approach is simpler and much less error
prone. Generally, the use of the time-domain multiplication property tends to lead to less clean solutions, as this forces
a convolution to be performed in the frequency domain and convolution is often best avoided if possible.
Example 5.24 (Fourier transform of an even signal). Show that if a signal x(t) is even, then its Fourier transform
X(ω ) is even.
Solution. From the definition of the Fourier transform, we have
Z ∞
X(ω ) = x(t)e− jω t dt.
−∞

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.8. FREQUENCY SPECTRA OF SIGNALS 129

Since x(t) is even, we can rewrite this as

Z ∞
X(ω ) = x(−t)e− jω t dt.
−∞

Now, we employ a change of variable. Let λ = −t so that d λ = −dt. Applying the change of variable, we obtain
Z −∞
X(ω ) = x(λ )e jωλ (−1)d λ
∞
Z −∞
=− x(λ )e jωλ d λ
∞
Z ∞
= x(λ )e jωλ d λ
−∞
Z ∞
= x(λ )e− j(−ω )λ d λ
−∞
= X(−ω ).

Therefore, X(ω ) is even.

Example 5.25 (Fourier transform of an odd signal). Show that if a signal x(t) is odd, then its Fourier transform X(ω )
is odd.

Solution. From the definition of the Fourier transform, we have

Z ∞
X(ω ) = x(t)e− jω t dt.
−∞

Since x(t) is odd, we can rewrite this as

Z ∞
X(ω ) = −x(−t)e− jω t dt.
−∞

Now, we employ a change of variable. Let λ = −t so that d λ = −dt. Applying this change of variable, we obtain
Z −∞
X(ω ) = −x(λ )e− jω (−λ ) (−1)d λ
∞
Z −∞
= x(λ )e jωλ d λ
∞
Z ∞
=− x(λ )e− j(−ω )λ d λ
−∞
= −X(−ω ).

Therefore, X(ω ) is odd.

5.8 Frequency Spectra of Signals

The Fourier transform representation expresses a signal as a function of complex sinusoids at all frequencies. In this
sense, the Fourier transform representation captures information about the frequency content of a signal. For example,
suppose that we have a signal x(t) with Fourier transform X(ω ). If X(ω ) is nonzero for some frequency ω0 , then the
signal x(t) contains some information at the frequency ω0 . On the other hand, if X(ω ) is zero at the frequency ω0 , then
the signal has no information at that frequency. In this way, the Fourier transform representation provides a means for
measuring the frequency content of a signal. This distribution of information in a signal over different frequencies is
referred to as the frequency spectrum of the signal. That is, X(ω ) is the frequency spectrum of x(t).

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

130 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

To gain further insight into the role played by the Fourier transform X(ω ) in the context of the frequency spectrum
of x(t), it is helpful to write the Fourier transform representation of x(t) with X(ω ) expressed in polar form as follows:
Z ∞
x(t) = 1
2π X(ω )e jω t d ω
−∞
Z ∞
= 1
2π |X(ω )| e j arg X(ω ) e jω t d ω
−∞
Z ∞
= 1
2π |X(ω )| e j[ω t+arg X(ω )] d ω .
−∞

In effect, the quantity |X(ω )| is a weight that determines how much the complex sinusoid at frequency ω contributes
to the integration result x(t). Perhaps, this can be more easily seen if we express the above integral as the limit of
a sum, derived from an approximation of the integral using the area of rectangles. Expressing x(t) in this way, we
obtain
∞
x(t) = lim 1
∑ ∆ω |X(k∆ω )| e j[k∆ω t+arg X(k∆ω )]
∆ω →0 2π
k=−∞
∞
′ ′
= lim 1
∑ ∆ω X(ω ′ ) e j[ω t+arg X(ω )] ,
∆ω →0 2π
k=−∞

where ω ′ = k∆ω . From the last line of the above equation, the kth term in the summation (associated with the
frequency ω ′ = k∆ω ) corresponds to a complex sinusoid with fundamental frequency ω ′ that has had its amplitude
scaled by a factor of |X(ω ′ )| and has been time-shifted by an amount that depends on arg X(ω ′ ). For a given ω ′ =
k∆ω (which is associated with the kth term in the summation), the larger |X(ω ′ )| is, the larger the amplitude of its
′
corresponding complex sinusoid e jω t will be, and therefore the larger the contribution the kth term will make to the
overall summation. In this way, we can use |X(ω ′ )| as a measure of how much information a signal x(t) has at the
frequency ω ′ .
Note that, since the Fourier transform X(ω ) is a function of a real variable (namely, ω ), a signal can, in the most
general case, have information at any arbitrary real frequency. This is different from the case of frequency spectra
in the Fourier series context (which deals only with periodic signals), where a signal can only have information at
certain specific frequencies (namely, at integer multiples of the fundamental frequency). There is no inconsistency
here, however. As we saw in Section 5.6, in the case of periodic signals the Fourier transform will also be zero, except
possibly at integer multiples of the fundamental frequency.
Since the frequency spectrum is complex (in the general case), it is usually represented using two plots, one
showing the magnitude and one showing the argument of X(ω ). We refer to |X(ω )| as the magnitude spectrum of
the signal x(t). Similarly, we refer to arg X(ω ) as the phase spectrum of the signal x(t). In the special case that X(ω )
is a real-valued (or purely imaginary) function, we usually plot the frequency spectrum directly on a single graph.
Consider the signal x(t) = sgn(t − 1). We can show that the Fourier transform X(ω ) of this signal is X(ω ) =
2 − jω
jω e . In this case, X(ω ) is neither purely real nor purely imaginary, so we use two separate graphs to represent
the frequency spectrum of the signal. We plot the magnitude spectrum and phase spectrum as shown in Figures 5.6(a)
and (b), respectively.
Consider the signal x(t) = δ (t). This signal has the Fourier transform X(ω ) = 1. Since, in this case, X(ω ) is a
real-valued function, we can plot the frequency spectrum X(ω ) on a single graph, as shown in Figure 5.7.
Consider the signal x(t) = sinct/2. This signal has the Fourier transform X(ω ) = 2π rect ω . Since, in this case,
X(ω ) is a real-valued function, we can plot the frequency spectrum X(ω ) on a single graph, as shown in Figure 5.8.
Suppose that we have a signal x(t) with Fourier transform X(ω ). If x(t) is real, then

|X(ω )| = |X(−ω )| and

arg X(ω ) = − arg X(−ω )

(i.e., |X(ω )| is an even function and arg X(ω ) is an odd function). (Earlier, these identities were shown to hold. See
Example 5.10 for a detailed proof.)

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.8. FREQUENCY SPECTRA OF SIGNALS 131

arg X(ω )

|X(ω )|

6 π
2

4 ω
− π2 π
2
2
− π2

ω
−3 −2 −1 0 1 2 3
(a)

(b)

Figure 5.6: Frequency spectrum of the time-shifted signum signal. (a) Magnitude spectrum and (b) Phase spectrum.

X(ω )

1
··· ···

Figure 5.7: Frequency spectrum of the unit-impulse function.

X(ω )

2π

ω
− 21 1
2

Figure 5.8: Frequency spectrum of the sinc function.

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

132 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

Example 5.26 (Frequency spectrum of  the shifted

 signum signal). Suppose that we have the signal x(t) = sgn(t −
2
1) with the Fourier transform X(ω ) = jω e − j ω . Find and plot the magnitude and phase spectra of this signal.
Determine at what frequency (or frequencies) the signal x(t) has the most information.

Solution. First, we must find the magnitude spectrum |X(ω )|. From the expression for X(ω ), we can write

2 − jω
|X(ω )| = jω e

= 2
jω e− jω
2
= jω
2
= |ω | .

Next, we find the phase spectrum arg X(ω ). From the expression for X(ω ), we can write
n o
arg X(ω ) = arg j2ω e− jω
= arg e− jω + arg 2
jω
2
= −ω + arg jω

= −ω + arg(− ωj2 )
(
− π − ω for ω > 0
= π 2
2 −ω for ω < 0
= − π2 sgn ω − ω .

Note that
(
− π2 for ω > 0
arg 2
jω = arg(− ωj2 ) = π
2 for ω < 0.

Finally, using numerical calculation, we can plot the graphs of the functions |X(ω )| and arg X(ω ) to obtain the results
shown previously in Figures 5.6(a) and (b). Since |X(ω )| is largest for ω = 0, the signal x(t) has the most information
at the frequency 0.

Example 5.27 (Frequency spectra of images). The human visual system is more sensitive to the phase spectrum of
an image than its magnitude spectrum. This can be aptly demonstrated by separately modifying the magnitude and
phase spectra of an image, and observing the effect. Below, we consider two variations on this theme.
Consider the potatohead and hongkong images shown in Figures 5.9(a) and (b), respectively. Replacing the
magnitude spectrum of the potatohead image with the magnitude spectrum of the hongkong image (and leaving the
phase spectrum unmodified), we obtain the new image shown in Figure 5.9(c). Although changing the magnitude
spectrum has led to distortion, the basic essence of the original image has not been lost. On the other hand, replacing
the phase spectrum of the potatohead image with the phase spectrum of the hongkong image (and leaving the mag-
nitude spectrum unmodified), we obtain the image shown in Figure 5.9(d). Clearly, by changing the phase spectrum
of the image, the fundamental nature of the image has been altered, with the new image more closely resembling the
hongkong image than the original potatohead image.
A more extreme scenario is considered in Figure 5.10. In this case, we replace each of the magnitude and phase
spectra of the potatohead image with random data, with this data being taken from the image consisting of random
noise shown in Figure 5.10(b). When we completely replace the magnitude spectrum of the potatohead image with
random values, we can still recognize the resulting image in Figure 5.10(c) as a very grainy version of the original
potatohead image. On the other hand, when the phase spectrum of the potatohead image is replaced with random
values, all visible traces of the original potatohead image are lost in the resulting image in Figure 5.10(d).

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.9. BANDWIDTH OF SIGNALS 135

|X(ω )|

ω
−B B

Figure 5.11: Bandwidth of a signal.

5.9 Bandwidth of Signals

A signal x(t) with Fourier transform X(ω ) is said to be bandlimited if, for some nonnegative real constant B, X(ω ) =
0 for all ω satisfying |ω | > B. We sometimes refer to B as the bandwidth of the signal x(t). An illustrative example
is provided in Figure 5.11.
One can show that a signal cannot be both time limited and bandlimited. To help understand why this is so we
recall the time/frequency scaling property of the Fourier transform. From this property, we know that as we compress
a signal x(t) along the t-axis, its Fourier transform X(ω ) will expand along the ω -axis. Similarly, a compression of
the Fourier transform X(ω ) along the ω -axis corresponds to an expansion of x(t) along the t-axis. Clearly, there is an
inverse relationship between the time-extent and bandwidth of a signal.

5.10 Frequency Response of LTI Systems

Suppose that we have a LTI system with input x(t), output y(t), and impulse response h(t). Such a system is depicted
in Figure 5.12. The behavior of such a system is governed by the equation

y(t) = x(t) ∗ h(t). (5.34)

Let X(ω ), Y (ω ), and H(ω ) denote the Fourier transforms of x(t), y(t), and h(t), respectively. Taking the Fourier
transform of both sides of (5.34) yields

Y (ω ) = F {x(t) ∗ h(t)}.

From the time-domain convolution property of the Fourier transform, however, we can rewrite this as

Y (ω ) = X(ω )H(ω ). (5.35)

This result provides an alternative way of viewing the behavior of an LTI system. That is, we can view the system as
operating in the frequency domain on the Fourier transforms of the input and output signals. In other words, we have
a system resembling that in Figure 5.13. In this case, however, the convolution operation from the time domain is
replaced by multiplication in the frequency domain. The frequency spectrum (i.e., Fourier transform) of the output is
the product of the frequency spectrum (i.e., Fourier transform) of the input and the frequency spectrum (i.e., Fourier
transform) of the impulse response. As a matter of terminology, we refer to H(ω ) as the frequency response of the
system. The system behavior is completely characterized by the frequency response H(ω ). If we know the input,
we can compute its Fourier transform X(ω ), and then determine the Fourier transform Y (ω ) of the output. Using the
inverse Fourier transform, we can then determine the output y(t).
In the general case, H(ω ) is a complex-valued function. Thus, we can represent H(ω ) in terms of its magnitude
|H(ω )| and argument arg H(ω ). We refer to |H(ω )| as the magnitude response of the system. Similarly, we call
arg H(ω ) the phase response of the system.

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

136 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

x(t) y(t)
h(t)

Figure 5.12: Time-domain view of LTI system.

X(ω ) Y (ω )
H(ω )

Figure 5.13: Frequency-domain view of LTI system.

H(ω )

2π

ω
− 21 1
2

Figure 5.14: Frequency response of example system.

From (5.35), we can write

|Y (ω )| = |X(ω )H(ω )|
= |X(ω )| |H(ω )| and (5.36)

argY (ω ) = arg(X(ω )H(ω ))

= arg X(ω ) + arg H(ω ). (5.37)

From (5.36), we can see that the magnitude spectrum of the output equals the magnitude spectrum of the input times
the magnitude spectrum of the impulse response. From (5.37), we have that the phase spectrum of the output equals
the phase spectrum of the input plus the phase spectrum of the impulse response.
Since the frequency response H(ω ) is simply the frequency spectrum of the impulse response h(t), for the reasons
explained in Section 5.8, if h(t) is real, then

|H(ω )| = |H(−ω )| and

arg H(ω ) = − arg H(−ω )

(i.e., the magnitude response |H(ω )| is an even function and the phase response arg H(ω ) is an odd function).
Example 5.28. Suppose that we have a LTI system with impulse response h(t) = sinct/2. The frequency response of
the system is H(ω ) = 2π rect ω . In this particular case, H(ω ) is real. So, we can plot the frequency response H(ω )
on a single graph as shown in Figure 5.14.

5.11 Frequency Response and Differential Equation Representations of LTI

Systems
Many LTI systems of practical interest can be represented using an Nth-order linear differential equation with constant
coefficients. Suppose that we have such a system with input x(t) and output y(t). Then, the input-output behavior of

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.11. FREQUENCY RESPONSE AND DIFFERENTIAL EQUATION REPRESENTATIONS OF LTI
SYSTEMS 137

the system is given by an equation of the form

N M
dk dk
∑ bk dt k
y(t) = ∑ dt k x(t)
ak
k=0 k=0

(where M ≤ N). Let X(ω ) and Y (ω ) denote the Fourier transforms of x(t) and y(t), respectively. Taking the Fourier
transform of both sides of this equation yields
( ) ( )
N M
dk dk
F ∑ bk k y(t) = F ∑ ak k x(t) .
k=0 dt k=0 dt
Using the linearity property of the Fourier transform, we can rewrite this as
N  k  M  k 
d d
∑ bk F dt k y(t) = ∑ ak F dt k x(t) .
k=0 k=0

Using the time-differentiation property of the Fourier transform, we can re-express this as
N M
∑ bk ( jω )kY (ω ) = ∑ ak ( jω )k X(ω ).
k=0 k=0

Then, factoring we have

N M
Y (ω ) ∑ bk ( jω )k = X(ω ) ∑ ak ( jω )k .
k=0 k=0

Rearranging this equation, we find the frequency response H(ω ) of the system to be
Y (ω ) ∑M ak ( jω )k ∑M k k
k=0 ak j ω
H(ω ) = = Nk=0 = N
.
X(ω ) ∑k=0 bk ( jω )k ∑k=0 bk jk ω k
Observe that, for a system of the form considered above, the frequency response is a rational function—hence, our
interest in rational functions.
Example 5.29 (Resistors, inductors, and capacitors). The basic building blocks of many electrical networks are
resistors, inductors, and capacitors. The resistor, shown in Figure 5.15(a), is governed by the relationship
F
v(t) = Ri(t) ←→ V (ω ) = RI(ω )
where R, v(t) and i(t) denote the resistance of, voltage across, and current through the resistor, respectively. The
inductor, shown in Figure 5.15(b), is governed by the relationship
F
v(t) = L dtd i(t) ←→ V (ω ) = jω LI(ω )

or equivalently
Z t
1 F 1
i(t) = L v(τ )d τ ←→ I(ω ) = jω L V (ω )
−∞

where L, v(t), and i(t) denote the inductance of, voltage across, and current through the inductor, respectively. The
capacitor, shown in Figure 5.15(c), is governed by the relationship
Z t
1 F 1
v(t) = C i(τ )d τ ←→ V (ω ) = jω C I(ω )
−∞

or equivalently
F
i(t) = C dtd v(t) ←→ I(ω ) = jω CV (ω )
where C, v(t), and i(t) denote the capacitance of, voltage across, and current through the capacitor, respectively.

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

138 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

R L C
i(t) + − i(t) + − i(t) + −

v(t) v(t) v(t)

(a) (b) (c)

Figure 5.15: Basic electrical components. (a) Resistor, (b) inductor, and (c) capacitor.

Example 5.30 (Simple RL network). Suppose that we have the RL network shown in Figure 5.16 with input v1 (t)
and output v2 (t). This system is LTI, since it can be characterized by a linear differential equation with constant
coefficients. (a) Find the frequency response H(ω ) of the system. (b) Find the response v2 (t) of the system to the
input v1 (t) = sgnt.

Solution. (a) From basic circuit analysis, we can write

v1 (t) = Ri(t) + L dtd i(t) and (5.38)

v2 (t) = L dtd i(t). (5.39)

(Recall that the voltage v(t) across an inductor L is related to the current i(t) through the inductor as v(t) = L dtd i(t).)
Taking the Fourier transform of (5.38) and (5.39) yields

V1 (ω ) = RI(ω ) + jω LI(ω )
= (R + jω L)I(ω ) and (5.40)

V2 (ω ) = jω LI(ω ). (5.41)

From (5.40) and (5.41), we have

V2 (ω )
H(ω ) =
V1 (ω )
jω LI(ω )
=
(R + jω L)I(ω )
jω L
= . (5.42)
R + jω L
Thus, we have found the frequency response of the system.
(b) Now, suppose that v1 (t) = sgnt (as given). Taking the Fourier transform of the input v1 (t) (with the aid of
Table 5.2), we have
2
V1 (ω ) = . (5.43)
jω
From the definition of the system, we know

V2 (ω ) = H(ω )V1 (ω ). (5.44)

Substituting (5.43) and (5.42) into (5.44), we obtain

  
jω L 2
V2 (ω ) =
R + jω L jω
2L
= .
R + jω L

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.12. ENERGY SPECTRAL DENSITY 139

R
i(t)

v1 (t) L v2 (t)

Figure 5.16: Simple RL network.

Taking the inverse Fourier transform of both sides of this equation, we obtain
 
2L
v2 (t) = F −1
R + jω L
 
2
= F −1
R/L + jω
 
1
= 2F −1 .
R/L + jω
Using Table 5.2, we can simplify to obtain
v2 (t) = 2e−(R/L)t u(t).
Thus, we have found the response v2 (t) to the input v1 (t) = sgnt.

5.12 Energy Spectral Density

Suppose that we have a signal x(t) with finite energy E and Fourier transform X(ω ). By definition, the energy
contained in x(t) is given by
Z ∞
E= |x(t)|2 dt.
−∞

We can use Parseval’s relation (5.14) to express E in terms of X(ω ) as

Z ∞
1
E= |X(ω )|2 d ω
2π −∞
Z ∞
= 1
2π |X(ω )|2 d ω .
−∞

Thus, the energy E is given by

Z ∞
E= Ex (ω )d ω
−∞

where

Ex (ω ) = 1
2π |X(ω )|2 .
We refer to Ex (ω ) as the energy spectral density of the signal x(t). The quantity Ex (ω ) indicates how the energy in
the signal x(t) is distributed as a function of frequency (in units of energy per rad/unit-time). For example, the energy
contributed by frequency components in range ω1 ≤ ω ≤ ω2 is simply given by
Z ω2
Ex (ω )d ω .
ω1

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

140 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

Example 5.31. Compute the energy spectral density Ex (ω ) of the signal x(t) = sinct/2. Determine the amount of
energy contained in the frequency components in the range |ω | ≤ 14 . Also, determine the total amount of energy in
the signal.

Solution. First, we compute the Fourier transform X(ω ) of x(t). We obtain X(ω ) = 2π rect ω . Next, we find the
energy spectral density function Ex (ω ) as follows:

Ex (ω ) = 1
2π |X(ω )|2
= 1
2π |2π rect ω |2
= 2π rect2 ω
= 2π rect ω .

Let E1 denote the energy contained in the signal for frequencies |ω | ≤ 41 . Then, we have
Z 1/4
E1 = Ex (ω )d ω
−1/4
Z 1/4
= 2π rect ω d ω
−1/4
Z 1/4
= 2π d ω
−1/4
= π.

Let E denote the total amount of energy in the signal. We can compute E as follows:
Z ∞
E= Ex (ω )d ω
−∞
Z ∞
= 2π rect ω d ω
−∞
Z 1/2
= 2π d ω
−1/2
= 2π .

5.13 Power Spectral Density

Suppose that we have a signal x(t) with finite power P and Fourier transform X(ω ). Let us define xT (t) as the
following truncated (or “windowed”) version of x(t):

xT (t) = x(t) rect(t/T ).

Also, let XT (ω ) denote the Fourier transform of xT (t). From this definition of xT (t), we have that

lim xT (t) = x(t).

T →∞

The average power of the signal xT (t) is given by

Z ∞
1
P = lim |xT (t)|2 dt.
T →∞ T −∞

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.14. FILTERING 141

Using Parseval’s relation (5.14), we can rewrite this expression for P in terms of X(ω ) to obtain
 Z 
1 1 ∞ 2
P = lim |XT (ω )| d ω
T →∞ T 2π −∞
 Z ∞ 
1 2
= lim |XT (ω )| d ω .
T →∞ 2π T −∞

Interchanging the order of the limit operation and integration, we obtain

Z ∞
1
P= lim |XT (ω )|2 d ω ].
−∞ T →∞ 2π T

Thus, we have that

Z ∞
P= Sx (ω )d ω
−∞

where
1
Sx (ω ) = lim |XT (ω )|2 .
T →∞ 2π T
We refer to Sx (ω ) as the power spectral density of the signal x(t). This quantity indicates how the power in the signal
x(t) is distributed as a function of frequency (in units of power per rad/unit-time). We can, therefore, determine the
amount of power contained in frequency components over the range ω1 ≤ ω ≤ ω2 as
Z ω2
Sx (ω )d ω .
ω1

5.14 Filtering
In some applications, we want to change the relative amplitude of the frequency components of a signal or possibly
eliminate some frequency components altogether. This process of modifying the frequency components of a signal
is referred to as filtering. Various types of filters exist. One type is frequency-selective filters. Frequency selective
filters pass some frequencies with little or no distortion, while significantly attenuating other frequencies. Several
basic types of frequency-selective filters include: lowpass, highpass, and bandpass.
An ideal lowpass filter eliminates all frequency components with a frequency greater than some cutoff frequency,
while leaving the remaining frequency components unaffected. Such a filter has a frequency response of the form
(
1 for |ω | ≤ ωc
H(ω ) =
0 otherwise,

where ωc is the cutoff frequency. A plot of this frequency response is given in Figure 5.17(a).
The ideal highpass filter eliminates all frequency components with a frequency less than some cutoff frequency,
while leaving the remaining frequency components unaffected. Such a filter has a frequency response of the form
(
1 for |ω | ≥ ωc
H(ω ) =
0 otherwise,

where ωc is the cutoff frequency. A plot of this frequency response is given in Figure 5.17(b).
An ideal bandpass filter eliminates all frequency components that do not lie in its passband, while leaving the
remaining frequency components unaffected. Such a filter has a frequency response of the form
(
1 for ωc1 ≤ |ω | ≤ ωc2
H(ω ) =
0 otherwise,

where the limits of the passband are ωc1 and ωc2 . A plot of this frequency response is given in Figure 5.17(c).

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

142 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

H(ω )

ω
−ωc ωc
(a)

H(ω )

1
··· ···

ω
−ωc ωc
(b)

H(ω )

ω
−ωc2 −ωc1 ωc1 ωc2
(c)

Figure 5.17: Frequency responses of (a) ideal lowpass, (b) ideal highpass, and (c) ideal bandpass filters.

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.14. FILTERING 143

Example 5.32 (Ideal filters). For each of the following impulse responses, find and plot the frequency response of the
corresponding system:

hLP (t) = ωπc sinc ωct,

hHP (t) = δ (t) − ωπc sinc ωct, and
2 ωb
hBP (t) = π [sinc ωb t] cos ωa t,

where ωc , ωa , and ωb are positive real constants. In each case, identify the type of frequency-selective filter to which
the system corresponds.

Solution. In what follows, let us denote the input and output of the system as x(t) and y(t), respectively. Also, let
X(ω ) and Y (ω ) denote the Fourier transforms of x(t) and y(t), respectively.
First, let us consider the system with impulse response hLP (t). The frequency response HLP (ω ) of the system is
simply the Fourier transform of the impulse response hLP (t). Thus, we have

HLP (ω ) = F {hLP (t)}

= F { ωπc sinc ωct}
= ωπc F {sinc ωct}
h i
= ωπc ωπc rect 2ωωc
= rect 2ωωc
(
1 for |ω | ≤ ωc
=
0 otherwise.

The frequency response HLP (ω ) is plotted in Figure 5.18(a). Since Y (ω ) = HLP (ω )X(ω ) and HLP (ω ) = 0 for |ω | >
ωc , Y (ω ) will contain only those frequency components in X(ω ) that lie in the frequency range |ω | ≤ ωc . In other
words, only the lower frequency components from X(ω ) are kept. Thus, the system represents a lowpass filter.
Second, let us consider the system with impulse response hHP (t). The frequency response HHP (ω ) of the system
is simply the Fourier transform of the impulse response hHP (t). Thus, we have

HHP (ω ) = F {hHP (t)}

= F {δ (t) − ωπc sinc ωct}
= F {δ (t)} − ωπc F {sinc ωct}
= 1 − ωπc [ ωπc rect 2ωωc ]
= 1 − rect 2ωωc
(
1 for |ω | ≥ ωc
=
0 otherwise.

The frequency response HHP (ω ) is plotted in Figure 5.18(b). Since Y (ω ) = HHP (ω )X(ω ) and HHP (ω ) = 0 for
|ω | < ωc , Y (ω ) will contain only those frequency components in X(ω ) that lie in the frequency range |ω | ≥ ωc . In
other words, only the higher frequency components from X(ω ) are kept. Thus, the system represents a highpass filter.
Third, let us consider the system with impulse response hBP (t). The frequency response HBP (ω ) of the system is

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

144 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

simply the Fourier transform of the impulse response hBP (t). Thus, we have

HBP (ω ) = F {hBP (t)}

= F { 2ωπ b [sinc ωbt] cos ωat}
ωb
= π F {[sinc ωb t](2 cos ωa t)}
ωb j ωa t
= π F {[sinc ωb t][e + e− jωa t ]}
ωb  j ωa t

= π F {e sinc ωbt} + F {e− jωa t sinc ωbt}
h i
ωb
= π
π
ωb rect ω2−ωωa + ωπ rect ω2+ωωa
b b b
ω −ωa ω +ωa
= rect + rect
2 ωb 2 ωb
(
1 for ωa − ωb ≤ |ω | ≤ ωa + ωb
=
0 otherwise.

The frequency response HBP (ω ) is plotted in Figure 5.18(c). Since Y (ω ) = HBP (ω )X(ω ) and HBP (ω ) = 0 for |ω | <
ωa − ωb or |ω | > ωa + ωb , Y (ω ) will contain only those frequency components in X(ω ) that lie in the frequency range
ωa − ωb ≤ |ω | ≤ ωa + ωb . In other words, only the middle frequency components of X(ω ) are kept. Thus, the system
represents a bandpass filter.

Example 5.33 (Lowpass filtering). Suppose that we have a LTI system with input x(t), output y(t), and impulse
response h(t), where

h(t) = 300 sinc 300π t.

Using frequency-domain methods, find the response y(t) of the system to the input x(t) = x1 (t), where

x1 (t) = 12 + 43 cos 200π t + 21 cos 400π t − 41 cos 600π t.

Solution. To begin, we must find the frequency spectrum X1 (ω ) of the signal x1 (t). Computing X1 (ω ), we have

X1 (ω ) = F 21 + 43 cos 200π t + 21 cos 400π t − 14 cos 600π t
= 21 F {1} + 34 F {cos 200π t} + 21 F {cos 400π t} − 41 F {cos 600π t}
= 21 [2πδ (ω )] + 34π [δ (ω + 200π ) + δ (ω − 200π )] + π2 [δ (ω + 400π ) + δ (ω − 400π )]
− π4 [δ (ω + 600π ) + δ (ω − 600π )]
= − π4 δ (ω + 600π ) + π2 δ (ω + 400π ) + 34π δ (ω + 200π ) + πδ (ω ) + 34π δ (ω − 200π )
+ π2 δ (ω − 400π ) − π4 δ (ω − 600π ).

A plot of the frequency spectrum X1 (ω ) is shown in Figure 5.19(a). Using the results of Example 5.32, we can
determine the frequency response H(ω ) of the system to be

H(ω ) = F {300 sinc 300π t}

ω
= rect 2(300 π)
(
1 for |ω | ≤ 300π
=
0 otherwise.

The frequency response H(ω ) is shown in Figure 5.19(b). The frequency spectrum Y (ω ) of the output can be com-
puted as

Y (ω ) = H(ω )X(ω )
3π 3π
= 4 δ (ω + 200π ) + πδ (ω ) + 4 δ (ω − 200π ).

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.14. FILTERING 145

HLP (ω )

ω
−ωc ωc
(a)

HHP (ω )

1
··· ···

ω
−ωc ωc
(b)

HBP (ω )

ω
−ωa − ωb −ωa −ωa + ωb ωa − ωb ωa ωa + ωb
(c)

Figure 5.18: Frequency responses of systems from example.

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

146 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

X1 (ω )

π
3π
4
3π
4 H(ω )
π π
2 2 1

ω
−300π 300π
ω
−600π −400π−200π 200π 400π 600π (b)
− π4 − π4

(a)

Y (ω )

π
3π 3π
4 4

ω
−200π 200π
(c)

Figure 5.19: Frequency spectra for lowpass filtering example.

The frequency spectrum Y (ω ) is shown in Figure 5.19(c). Taking the inverse Fourier transform of Y (ω ) yields

y(t) = F −1 34π δ (ω + 200π ) + πδ (ω ) + 34π δ (ω − 200π )
= π F −1 {δ (ω )} + 43 F −1 {π [δ (ω + 200π ) + δ (ω − 200π )]}
= π ( 21π ) + 43 cos 200π t
= 21 + 43 cos 200π t.

Example 5.34 (Bandpass filtering). Suppose that we have a LTI system with input x(t), output y(t), and impulse
response h(t), where
h(t) = (200 sinc 100π t) cos 400π t.
Using frequency-domain methods, find the response y(t) of the system to the input x(t) = x1 (t) where x1 (t) is as
defined in Example 5.33.
Solution. From Example 5.33, we already know the frequency spectrum X1 (ω ). In particular, we previously found
that
X1 (ω ) = − π4 δ (ω + 600π ) + π2 δ (ω + 400π ) + 34π δ (ω + 200π ) + πδ (ω ) + 34π δ (ω − 200π )
+ π2 δ (ω − 400π ) − π4 δ (ω − 600π ).

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.15. SAMPLING AND INTERPOLATION 147

The frequency spectrum X1 (ω ) is shown in Figure 5.20(a). Now, we compute the frequency response H(ω ) of the
system. Using the results of Example 5.32, we can determine H(ω ) to be

H(ω ) = F {(200 sinc 100π t) cos 400π t}

= rect ω2(100
−400π ω +400π
π ) + rect 2(100π )
(
1 for 300π ≤ |ω | ≤ 500π
=
0 otherwise.

The frequency response H(ω ) is shown in Figure 5.20(b). The frequency spectrum Y (ω ) of the output is given by

Y (ω ) = H(ω )X(ω )
= π2 δ (ω + 400π ) + π2 δ (ω − 400π ).

Taking the inverse Fourier transform, we obtain


y(t) = F −1 π2 δ (ω + 400π ) + π2 δ (ω − 400π )
= 21 F −1 {π [δ (ω + 400π ) + δ (ω − 400π )]}
= 12 cos 400π t.

5.15 Sampling and Interpolation

Often, we encounter situations in which we would like to process a continuous-time signal in the discrete-time domain
or vice versa. For example, we might have a continuous-time audio signal that we would like to process using a digital
computer (which is a discrete-time system), or we might have a discrete-time audio signal that we wish to play on a
loudspeaker (which is a continuous-time system). Clearly, some means is needed to link the continuous- and discrete-
time domains. This connection is established through processes known as sampling and interpolation. In what follows,
we will formally introduce these processes and study them in some detail.
Sampling allows us to create a discrete-time signal from a continuous-time signal. Although sampling can be
performed in many different ways, the most commonly used scheme is periodic sampling. With this scheme, a
sequence y[n] of samples is obtained from a continuous-time signal x(t) according to the relation

y[n] = x(nT ) for all integer n, (5.45)

where T is a positive real constant. As a matter of terminology, we refer to T as the sampling period, and ωs = 2π /T
as the (angular) sampling frequency. A system such as that described by (5.45) is known as an ideal continuous-to-
discrete-time (C/D) converter, and is shown diagrammatically in Figure 5.21. An example of periodic sampling is
shown in Figure 5.22. In Figure 5.22(a), we have the original continuous-time signal x(t). This signal is then sampled
with sampling period T = 10, yielding the sequence y[n] in Figure 5.22(b).
Interpolation allows us to construct a continuous-time signal from a discrete-time signal. In effect, this process is
one of assigning values to a signal between its sample points. Although there are many different ways in which to per-
form interpolation, we will focus our attention in subsequent sections on one particular scheme known as bandlimited
interpolation. Interpolation produces a continuous-time signal x̂(t) from a sequence y[n] according to the relation

x̂(t) = f (y[n]),

where f is some function of the sample values y[n]. The precise form of the function f depends on the particular
interpolation scheme employed. The interpolation process is performed by a system known as an ideal discrete-to-
continuous-time (D/C) converter, as shown in Figure 5.23.

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

148 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

X1 (ω )

π
3π 3π H(ω )
4 4

π π
2 2
1

ω
−500π −300π 300π 500π
ω
−600π −400π−200π 200π 400π 600π (b)
− π4 − π4

(a)

Y (ω )

π π
2 2

ω
−400π 400π
(c)

Figure 5.20: Frequency spectra for bandpass filtering example.

x(t) C/D y[n]

(with sampling
period T )

Figure 5.21: Ideal C/D converter.

x(t) y[n]

4 4

3 3

2 2

1 1

t n
0 10 20 30 40 50 60 70 0 1 2 3 4 5 6 7

(a) (b)

Figure 5.22: Periodic sampling of signal.

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.15. SAMPLING AND INTERPOLATION 149

y[n] D/C x̂(t)

(with sampling
period T )

Figure 5.23: Ideal D/C converter.

In the absence of any constraints, a continuous-time signal cannot usually be uniquely determined from a sequence
of its equally-spaced samples. In other words, the sampling process is not generally invertible. Consider, for example,
the continuous-time signals x1 (t) and x2 (t) given by
x1 (t) = 0 and
x2 (t) = sin(2π t).
If we sample each of these signals with the sampling period T = 1, we obtain the respective sequences
y1 [n] = x1 (nT ) = x1 (n) = 0 and
y2 [n] = x2 (nT ) = sin(2π n) = 0.
Thus, y1 [n] = y2 [n] for all n, although x1 (t) 6= x2 (t) for all noninteger t. This example trivially shows that if no
constraints are placed upon a continuous-time signal, then the signal cannot be uniquely determined from its samples.
Fortunately, under certain circumstances, a continuous-time signal can be recovered exactly from its samples. In
particular, in the case that the signal being sampled is bandlimited, we can show that a sequence of its equally-spaced
samples uniquely determines the signal if the sampling period is sufficiently small. This result, known as the sampling
theorem, is of paramount importance in the study of signals and systems.

5.15.1 Sampling
In order to gain some insight into sampling, we need a way in which to mathematically model this process. To this
end, we employ the simple model for the ideal C/D converter shown in Figure 5.24. In short, we may view the process
of sampling as impulse train modulation followed by conversion of an impulse train to a sequence of sample values.
More specifically, to sample a signal x(t) with sampling period T , we first multiply the signal x(t) by the periodic
impulse train p(t) to obtain
s(t) = x(t)p(t) (5.46)
where
∞
p(t) = ∑ δ (t − kT ).
k=−∞
Then, we take the weights of successive impulses in s(t) to form a sequence y[n] of samples. The sampling frequency
is given by ωs = 2π /T . As a matter of terminology, p(t) is referred to as a sampling function. From the diagram, we
can see that the signals s(t) and y[n], although very closely related, have some key differences. The impulse train s(t)
is a continuous-time signal that is zero everywhere except at integer multiples of T (i.e., at sample points), while y[n]
is a discrete-time signal, defined only on integers with its values corresponding to the weights of successive impulses
in s(t). The various signals involved in sampling are illustrated in Figure 5.25.
In passing, we note that the above model of sampling is only a mathematical convenience. That is, the model
provides us with a relatively simple way in which to study the mathematical behavior of sampling. The above model,
however, is not directly useful as a means for actually realizing sampling in a real world system. Obviously, the im-
pulse train employed in the above model poses some insurmountable problems as far as implementation is concerned.
Now, let us consider the above model of sampling in more detail. In particular, we would like to find the relation-
ship between the frequency spectra of the original signal x(t) and its impulse-train sampled version s(t). Since p(t) is
T -periodic, it can be represented in terms of a Fourier series as
∞
p(t) = ∑ ck e jkωs t . (5.47)
k=−∞

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

150 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

ideal C/D converter

∞
p(t) = ∑ δ (t − kT )
k=−∞

x(t) s(t) convert from y[n]

× impulse train
to sequence

Figure 5.24: Model of ideal C/D converter.

x(t)

p(t)
4

3 1 1 1 1

1 t
0 T 2T 3T
t (b)
0 T 2T 3T
(a)

s(t) y[n]

x(T ) x(T )
4 4

3 x(3T ) 3 x(2T ) x(3T )

x(2T )

2 x(0) 2 x(0)

1 1

t n
0 T 2T 3T 0 1 2 3
(c) (d)

Figure 5.25: The various signals involved in the sampling process. (a) The original continuous-time signal x(t). (b)
The sampling function p(t). (c) The impulse-modulated signal s(t). (d) The discrete-time signal y[n].

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.15. SAMPLING AND INTERPOLATION 151

Using the Fourier series analysis equation, we calculate the coefficients ck to be

Z T /2
ck = 1
T p(t)e− jkωs t dt
−T /2
Z T /2
= 1
T δ (t)e− jkωst dt
−T /2
1
= T
ωs
= 2π . (5.48)

Substituting (5.47) and (5.48) into (5.46), we obtain

∞
ωs jkωs t
s(t) = x(t) ∑ 2π e
k=−∞
∞
ωs
= 2π ∑ x(t)e jkωs t .
k=−∞

Taking the Fourier transform of s(t) yields

∞
ωs
S(ω ) = 2π ∑ X(ω − kωs ). (5.49)
k=−∞

Thus, the spectrum of the impulse-train sampled signal s(t) is a scaled sum of an infinite number of shifted copies of
the spectrum of the original signal x(t).
Now, we consider a simple example to further illustrate the behavior of the sampling process in the frequency
domain. Suppose that we have a signal x(t) with the Fourier transform X(ω ) where |X(ω )| = 0 for |ω | > ωm (i.e.,
x(t) is bandlimited). To simplify the visualization process, we will assume X(ω ) has the particular form shown in
Figure 5.26(a). In what follows, however, we only actually rely on the bandlimited nature of x(t) and not the particular
shape of X(ω ). So, the results that we derive in what follows generally apply to any bandlimited signal. Let S(ω )
denote the Fourier transform of s(t). From (5.49), we know that S(ω ) is formed by the superposition of an infinite
number of shifted copies of X(ω ). Upon more careful consideration, we can see that two distinct situations can arise.
That is, the shifted copies of X(ω ) used to form S(ω ) can either: 1) overlap or, 2) not overlap. These two cases are
illustrated in Figures 5.26(b) and 5.26(c), respectively. From these graphs, we can see that the shifted copies of X(ω )
will not overlap if

ωm < ωs − ωm and − ωm > −ωs + ωm

or equivalently

ωs > 2ωm .

Consider the case in which the copies of the original spectrum X(ω ) in S(ω ) do not overlap, as depicted in
Figure 5.26(b). In this situation, the spectrum X(ω ) of the original signal is clearly discernable in the spectrum S(ω ).
In fact, one can see that the original spectrum X(ω ) can be obtained directly from S(ω ) through a lowpass filtering
operation. Thus, the original signal x(t) can be exactly recovered from s(t).
Now, consider the case in which copies of the original spectrum X(ω ) in S(ω ) do overlap. In this situation,
multiple frequencies in the spectrum X(ω ) of the original signal are mapped to the same frequency in S(ω ). This phe-
nomenon is referred to as aliasing. Clearly, aliasing leads to individual periods of S(ω ) having a different shape than
the original spectrum X(ω ). When aliasing occurs, the shape of the original spectrum X(ω ) is no longer discernable
from S(ω ). Consequently, we are unable to recover the original signal x(t) from s(t) in this case.

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

152 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

X(ω )

ω
−ωm 0 ωm
(a)

S(ω )

1
T

··· ···

ω
−ωs − ωm −ωs −ωs + ωm −ωm 0 ωm ωs − ωm ωs ωs + ωm
(b)

S(ω )

1
T

··· ···

ω
0 ωm
ωs − ωm ωs
(c)

Figure 5.26: Effect of impulse-train sampling on frequency spectrum. (a) Spectrum of original signal x(t). (b) Spec-
trum of s(t) in the absence of aliasing. (c) Spectrum of s(t) in the presence of aliasing.

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.15. SAMPLING AND INTERPOLATION 153

ideal D/C converter

y[n] convert from s(t) x̂(t)

sequence to H(ω )
impulse train

Tω
H(ω ) = T rect 2π

Figure 5.27: Model of ideal D/C converter.

5.15.2 Interpolation and Reconstruction of a Signal From Its Samples

Interpolation allows us to construct a continuous-time signal from a discrete-time one. This process is essentially
responsible for determining the value of a continuous-time signal between sample points. Except in very special
circumstances, it is not generally possible to exactly reproduce a continuous-time signal from its samples. Although
many interpolation schemes exist, we shall focus our attention shortly on one particular scheme. The interpolation
process can be modeled with the simple ideal D/C converter system, shown in Figure 5.27.
Recall the ideal C/D converter of Figure 5.24. Since the process of converting an impulse train to a sequence is
invertible, we can reconstruct the original signal x(t) from its sampled version y[n] if we can somehow recover x(t)
from s(t). Let us suppose now that x(t) is bandlimited. As we saw in the previous section, we can recover x(t) from
s(t) provided that x(t) is bandlimited and sampled at a sufficiently high rate so as to avoid aliasing. In the case that
aliasing does not occur, we can reconstruct the original continuous-time signal x(t) from y[n] using the ideal D/C
converter shown in Figure 5.27. In what follows, we will derive a formula for computing the original continuous-time
signal x̂(t) from its samples y[n]. Consider the model of the D/C converter. We have a lowpass filter with frequency
response
    (
Tω 2π ω T |ω | < ω2s
H(ω ) = T rect = rect =
2π ωs ωs 0 otherwise

and impulse response

 πt  ω t 
s
h(t) = sinc = sinc .
T 2
First, we convert the sequence y[n] to the impulse train s(t) to obtain
∞
s(t) = ∑ y[n]δ (t − nT ).
n=−∞

Then, we filter the resulting signal s(t) with the lowpass filter having impulse response h(t), yielding
x̂(t) = s(t) ∗ h(t)
Z ∞
= s(τ )h(t − τ )d τ
−∞
Z ∞ ∞
=
−∞
h(t − τ ) ∑ y[n]δ (τ − nT )d τ
n=−∞
∞ Z ∞
= ∑ y[n]
−∞
h(t − τ )δ (τ − nT )d τ
n=−∞
∞
= ∑ y[n]h(t − nT )
n=−∞
∞
= ∑ y[n] sinc( Tπ (t − nT )).
n=−∞

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

154 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

If x(t) is bandlimited and aliasing is avoided, x̂(t) = x(t) and we have a formula for exactly reproducing x(t) from its
samples y[n].

5.15.3 Sampling Theorem

In the preceding sections, we have established the important result given by the theorem below.
Theorem 5.3 (Sampling Theorem). Let x(t) be a signal with Fourier transform X(ω ), and suppose that |X(ω )| = 0
for all ω satisfying |ω | > ωM (i.e., x(t) is bandlimited to the interval [−ωM , ωM ]). Then, x(t) is uniquely determined
by its samples y[n] = x(nT ) for n = 0, ±1, ±2, . . ., if
ωs > 2ωM , (5.50)
where ωs = 2π /T . In particular, if (5.50) is satisfied, we have that
∞
x(t) = ∑ y[n] sinc( Tπ (t − nT )),
n=−∞

or equivalently (i.e., rewritten in terms of ωs instead of T ),

∞
x(t) = ∑ y[n] sinc( ω2s t − π n).
n=−∞

As a matter of terminology, we refer to (5.50) as the Nyquist condition (or Nyquist criterion). Also, we call
ωs /2 the Nyquist frequency and 2ωM the Nyquist rate. It is important to note that the Nyquist condition is a strict
inequality. Therefore, to ensure aliasing does not occur in the most general case, one must choose the sampling rate
larger than the Nyquist rate. One can show, however, that if the frequency spectrum does not have impulses at the
Nyquist frequency, it is sufficient to sample at exactly the Nyquist rate.
Example 5.35. Let x(t) denote a continuous-time audio signal with Fourier transform X(ω ). Suppose that |X(ω )| = 0
for all |ω | ≥ 44100π . Determine the largest period T with which x(t) can be sampled that will allow x(t) to be exactly
recovered from its samples.
Solution. The signal x(t) is bandlimited to frequencies in the range (−ωm , ωm ), where ωm = 44100π . From the
sampling theorem, we know that the minimum sampling rate required is given by
ωs = 2ωm
= 2(44100π )
= 88200π .
Thus, the largest permissible sampling period is given by
2π
T= ωs
2π
= 88200π
1
= 44100 .

Although the sampling theorem provides an upper bound on the sampling rate that holds in the case of arbitrary
bandlimited signals, in some special cases it may be possible to employ an even smaller sampling rate. This point is
further illustrated by way of the example below.
Example 5.36. Suppose that we have a signal x(t) with the Fourier transform X(ω ) shown in Figure 5.28 (where
ωc ≫ ωa ). (a) Using the sampling theorem directly, determine the largest permissible sampling period T that will
allow x(t) to be exactly reconstructed from its samples. (b) Explain how one can exploit the fact that X(ω ) = 0 for a
large portion of the interval [−ωc − ωa , ωc + ωa ] in order to reduce the rate at which x(t) must be sampled.

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.16. AMPLITUDE MODULATION 155

Solution. (a) The signal x(t) is bandlimited to (−ωm , ωm ), where ωm = ωc + ωa . Thus, the minimum sampling rate
required is given by

ωs = 2ωm
= 2(ωc + ωa )
= 2ωc + 2ωa .

and the maximum sampling period is calculated as

2π
T= ωs
2π
= 2ωc +2ωa
π
= ωc +ωa .

(b) We can modulate and lowpass filter x(t) in order to compress all of its spectral information into the frequency
range [−2ωa , 2ωa ], yielding the signal x1 (t). That is, we have

x1 (t) = [x(t) cos([ωc − ωa ]t)] ∗ h(t)

where
 
4 ωa F ω
h(t) = π sinc(2ωat) ←→ H(ω ) = 2 rect 4 ωa .

This process can be inverted (by modulation and filtering) to obtain x(t) from x1 (t). In particular, we have that

x(t) = (x1 (t) cos([ωc − ωa ]t)) ∗ h2 (t)

where
 
F
h2 (t) = δ (t) − 2(ωcπ−ωa ) sinc([ωc − ωa ]t) ←→ H2 (ω ) = 2 − 2 rect ω
4(ωc −ωa ) .

Let X1 (ω ) denote the Fourier transform of x1 (t). The Fourier transform X1 (ω ) is as shown in Figure 5.29. Apply-
ing the sampling theorem to x1 (t) we find that the minimum sampling rate is given by

ωs = 2(2ωa )
= 4ωa

and the largest sampling period is given by

2π
T= ωs
2π
= 4 ωa
π
= 2 ωa .

Since ωc ≫ ωa (by assumption), this new sampling period is larger than the one computed in the first part of this
problem.

5.16 Amplitude Modulation

In communication systems, we often need to transmit a signal using a frequency range that is different from that of the
original signal. For example, voice/audio signals typically have information in the range of 0 to 20 kHz. Often, it is not
practical to transmit such a signal using its original frequency range. Two potential problems with such an approach

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

156 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

X(ω )

ω
−ωc − ωa −ωc −ωc + ωa 0 ωc − ωa ωc ωc + ωa

Figure 5.28: Frequency spectrum of signal x(t).

X1 (ω )

ω
−2ωa 0 2ωa

Figure 5.29: Frequency spectrum of signal x1 (t).

are: 1) interference and 2) constraints on antenna length. Since many signals are broadcast over the airwaves, we
need to ensure that no two transmitters use the same frequency bands in order to avoid interference. Also, in the case
of transmission via electromagnetic waves (e.g., radio waves), the length of antenna required becomes impractically
large for the transmission of relatively low frequency signals. For the preceding reasons, we often need to change the
frequency range associated with a signal before transmission. In what follows, we consider one possible scheme for
accomplishing this. This scheme is known as amplitude modulation.
Amplitude modulation (AM) is used in many communication systems. Numerous variations on amplitude modu-
lation are possible. Here, we consider two of the simplest variations: double-side-band/suppressed-carrier (DSB/SC)
and single-side-band/suppressed-carrier (SSB/SC).

5.16.1 Modulation With a Complex Sinusoid

Suppose that we have the communication system shown in Figure 5.30. First, let us consider the transmitter in
Figure 5.30(a). The transmitter is a system with input x(t) and output y(t). Mathematically, the behavior of this
system is given by
y(t) = x(t)c1 (t) (5.51)
where
c1 (t) = e jωc t .
Let X(ω ), Y (ω ), and C1 (ω ) denote the Fourier transforms of x(t), y(t), and c1 (t), respectively. Taking the Fourier
transform of both sides of (5.51), we obtain
Y (ω ) = F {c1 (t)x(t)}

= F e jωc t x(t)
= X(ω − ωc ). (5.52)

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.16. AMPLITUDE MODULATION 157

c1 (t) = e jωc t c2 (t) = e− jωc t

x(t) y(t) y(t) x̂(t)

× ×

(a) (b)

Figure 5.30: Simple communication system. (a) Transmitter and (b) receiver.
.

X(ω ) C1 (ω ) C2 (ω )

1 2π 2π

ω ω ω
−ω b ωb ωc −ω c

(a) (b) (c)

Y (ω ) X̂(ω )

1 1

ω ω
ωb ωc − ωb ωc ωc + ωb −ω b ωb

(d) (e)

Figure 5.31: Frequency spectra for modulation with a complex sinusoid.

.

Thus, the frequency spectrum of the output is simply the frequency spectrum of the input shifted by ωc . The relation-
ship between the frequency spectra of the input and output is illustrated in Figure 5.31. Clearly, the output signal has
been shifted to a different frequency range as desired. Now, we need to determine whether the receiver can recover
the original signal x(t) from the transmitted signal y(t).
Now, let us consider the receiver shown in Figure 5.30(b). The receiver is a system with input y(t) and output x̂(t).
Mathematically, this system is given by
x̂(t) = y(t)c2 (t) (5.53)
where
c2 (t) = e− jωc t
(i.e., c2 (t) = c∗1 (t)). In order for the communication system to be useful, we need for the received signal x̂(t) to be
equal to the original signal x(t) from the transmitter. Let Y (ω ), X̂(ω ), and C2 (ω ) denote the Fourier transform of y(t),
x̂(t), and c2 (t), respectively. Taking the Fourier transform of both sides of (5.53), we obtain
X̂(ω ) = F {c2 (t)y(t)}
= F {e− jωc t y(t)}
= Y (ω + ωc ).

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

158 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

Substituting the expression for Y (ω ) in (5.52) into this equation, we obtain

X̂(ω ) = X([ω + ωc ] − ωc )
= X(ω ).

Since X̂(ω ) = X(ω ), we have that the received signal x̂(t) is equal to the original signal x(t) from the transmitter.
Thus, the communication system has the desired behavior. The relationship between the frequency spectra of the
various signals in the AM system is illustrated in Figure 5.31.
Although the above result is quite interesting mathematically, it does not have direct practical application. The
difficulty here is that c1 (t), c2 (t), and y(t) are complex signals, and we cannot realize complex signals in the physical
world. This communication system is not completely without value, however, as it leads to the development of the
practically useful system that we consider next.

5.16.2 DSB/SC Amplitude Modulation

Now, let us consider the communication system shown in Figure 5.32. This system is known as a double-side-
band/suppressed-carrier (DSB/SC) amplitude modulation (AM) system. This system is very similar to the one in
Figure 5.30. In the new system, however, we have replaced the complex sinusoid c1 (t) with a real sinusoid c(t). The
new system also requires that the input signal x(t) be bandlimited to frequencies in the interval [−ωb , ωb ] and that
ωb < ωc0 < 2ωc − ωb . The reasons for this restriction will become clear after having studied this system in more
detail.
Consider the transmitter shown in Figure 5.32(a). The transmitter is a system with input x(t) and output y(t).
Mathematically, we can describe the behavior of the system as

y(t) = x(t)c(t) (5.54)

where c(t) = cos ωct. (Note that we can rewrite c(t) as c(t) = 21 [e jωc t + e− jωc t ].) Taking the Fourier transform of both
sides of (5.54), we obtain

Y (ω ) = F {x(t)c(t)}

= F 21 [e jωc t + e− jωc t ]x(t)
 
= 21 F {e jωc t x(t)} + F {e− jωc t x(t)}
= 21 [X(ω − ωc ) + X(ω + ωc )] . (5.55)

Thus, the frequency spectrum of the output is the average of two shifted versions of the frequency spectrum of the
input. The relationship between the frequency spectra of the input and output is illustrated in Figure 5.33(d). Observe
that we have managed to shift the frequency spectrum of the input signal into a different range of frequencies for
transmission as desired. Now, we must determine whether the receiver can recover the original signal x(t).
Consider the receiver shown in Figure 5.32(b). The receiver is a system with input y(t) and output x̂(t). Let Y (ω ),
V (ω ) and X̂(ω ) denote the Fourier transforms of y(t), v(t) and x̂(t), respectively. Then, the input-output behavior of
the system is characterized by the equations

v(t) = c(t)y(t) and (5.56)

X̂(ω ) = H(ω )V (ω ) (5.57)

where
(
2 for |ω | ≤ ωc0
H(ω ) =
0 otherwise

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.17. EQUALIZATION 159

and ωb < ωc0 < 2ωc − ωb . Taking the Fourier transform of both sides of (5.56) yields

V (ω ) = F {c(t)y(t)}
  
= F 12 e jωc t + e− jωc t y(t)
   
= 21 F e jωc t y(t) + F e− jωc t y(t)
= 21 [Y (ω − ωc ) +Y (ω + ωc )].

Substituting the expression for Y (ω ) in (5.55) into this equation, we obtain

 
V (ω ) = 12 12 [X([ω − ωc ] − ωc ) + X([ω − ωc ] + ωc )] + 12 [X([ω + ωc ] − ωc ) + X([ω + ωc ] + ωc )]
= 21 X(ω ) + 14 X(ω − 2ωc ) + 41 X(ω + 2ωc ). (5.58)

The relationship between V (ω ) and X(ω ) is depicted graphically in Figure 5.33(e). Substituting the above expression
for V (ω ) into (5.57) and simplifying, we obtain

X̂(ω ) = H(ω )V (ω )
 
= H(ω ) 21 X(ω ) + 14 X(ω − 2ωc ) + 41 X(ω + 2ωc )
= 12 H(ω )X(ω ) + 41 H(ω )X(ω − 2ωc ) + 14 H(ω )X(ω + 2ωc )
= 12 [2X(ω )] + 41 (0) + 14 (0)
= X(ω ).

(In the above simplification, since H(ω ) = 2 rect( 2ωωc0 ) and ωb < ωc0 < 2ωc − ωb , we were able to deduce that
H(ω )X(ω ) = 2X(ω ), H(ω )X(ω − 2ωc ) = 0, and H(ω )X(ω + 2ωc ) = 0.) The relationship between X̂(ω ) and X(ω )
is depicted in Figure 5.33(f). Thus, we have that X̂(ω ) = X(ω ) which implies x̂(t) = x(t). So, we have recovered
the original signal x(t) at the receiver. This system has managed to shift x(t) into a different frequency range before
transmission and then recover x(t) at the receiver. This is exactly what we wanted to accomplish.

5.16.3 SSB/SC Amplitude Modulation

By making a minor modification to the DSB/SC amplitude modulation system, we can reduce the bandwidth require-
ments of the system by half. The resulting system is referred to as a SSB/SC amplitude modulation system. This
modified system is illustrated in Figure 5.34. In this new system, G(ω ) and H(ω ) are given by
(
1 for |ω | ≥ ωc
G(ω ) =
0 otherwise, and

(
4 for |ω | ≤ ωc0
H(ω ) =
0 otherwise.

Let X(ω ), Y (ω ), Q(ω ), V (ω ), X̂(ω ), and C(ω ) denote the Fourier transforms of x(t), y(t), q(t), v(t), x̂(t), and c(t),
respectively. Figure 5.35 depicts the transformations the signal undergoes as it passes through the system. Again, the
output from the receiver is equal to the input to the transmitter.

5.17 Equalization
Often, we find ourselves faced with a situation where we have a system with a particular frequency response that is
undesirable for the application at hand. As a result, we would like to change the frequency response of the system
to be something more desirable. This process of modifying the frequency response in this way is referred to as

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

160 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

c(t) = cos ωct c(t) = cos ωct H(ω ) = 2 rect( 2ωωc0 )

x(t) y(t) y(t) v(t) x̂(t)

× × H(ω )

(a) (b)

Figure 5.32: DSB/SC amplitude modulation system. (a) Transmitter and (b) receiver.

X(ω )

1 C(ω ) H(ω )

π π 2

ω ω
−ω c ωc −ωc0 ωc0
ω (b) (c)
−ω b ωb

(a)
Y (ω )
1
2

ω
−2ωc −ω c − ω b −ω c −ωc + ωb −ω b ωb ωc − ωb ωc ωc + ωb 2ωc

(d)
V (ω )
1
2

1
4

ω
−2ωc − ωb −2ωc −2ωc + ωb −ω b ωb 2ωc − ωb 2ωc 2ωc + ωb

(e)
X̂(ω )

ω
−2ωc −ω c −ω b ωb ωc 2ωc

(f)

Figure 5.33: Signal spectra for DSB/SC amplitude modulation.

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.17. EQUALIZATION 161

c(t) = cos ωct G(ω ) = 1 − rect( 2ωωc ) c(t) = cos ωct H(ω ) = 4 rect( 2ωωc0 )

x(t) q(t) y(t) y(t) v(t) x̂(t)

× G(ω ) × H(ω )

(a) (b)

Figure 5.34: SSB/SC amplitude modulation system. (a) Transmitter and (b) receiver.

X(ω )

1 C(ω ) G(ω ) H(ω )

π π 4
··· 1 ···

ω ω ω
−ωc ωc −ωc ωc −ωc0 ωc0

ω (b) (c) (d)

−ωb ωb

(a)
Q(ω )
1
2

ω
−2ωc −ω c − ω b −ω c −ω c + ω b −ω b ωb ωc − ωb ωc ωc + ωb 2ωc

(e)
Y (ω )
1
2

ω
−2ωc −ω c − ω b −ω c −ω c + ω b −ω b ωb ωc − ωb ωc ωc + ωb 2ωc

(f)
V (ω )
1
2

1
4

ω
−2ωc − ωb −2ωc −2ωc + ωb −ω b ωb 2ωc − ωb 2ωc 2ωc + ωb

(g)
X̂(ω )

ω
−2ωc −ω c −ω b ωb ωc 2ωc

(h)

Figure 5.35: Signal spectra for SSB/SC amplitude modulation.

Version: 2013-09-11 Copyright c 2013 Michael D. Adams
162 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

Input Output X(ω ) Y (ω )

Ho (ω ) He (ω ) Ho (ω )

(a) (b)

Figure 5.36: Equalization example. (a) Original system. (b) New system with equalization.

X(ω ) Y (ω )
G(ω ) H(ω )

Figure 5.37: Equalization system.

equalization. Essentially, equalization is just a filtering operation, where the filtering is applied with the specific goal
of obtaining a more desirable frequency response.
Let us now examine the mathematics behind equalization. Consider the LTI system with frequency response
Ho (ω ) as shown in Figure 5.36(a). Suppose that the frequency response Ho (ω ) is undesirable for some reason (i.e.,
the system does not behave in a way that is good for the application at hand). Consequently, we would instead like to
have a system with frequency response Hd (ω ). In effect, we would like to somehow change the frequency response
Ho (ω ) of the original system to Hd (ω ). This can be accomplished by using another system called an equalizer.
More specifically, consider the new system shown in Figure 5.36(b) which consists of a LTI equalizer with frequency
response He (ω ) connected in series with the original system having frequency response Ho (ω ). From the block
diagram, we have

Y (ω ) = H(ω )X(ω ),

where H(ω ) = Ho (ω )He (ω ). In effect, we want to force H(ω ) to be equal to Hd (ω ) so that the overall (i.e., series-
d (ω )
interconnected) system has the frequency response desired. So, we choose the equalizer to be such that He (ω ) = H Ho (ω ) .
Then, we have

H(ω ) = Ho (ω )He (ω )
 
Hd (ω )
= Ho (ω )
Ho (ω )
= Hd (ω ).

Thus, the system in Figure 5.36(b) has the frequency response Hd (ω ) as desired.
Equalization is used in many applications. In real-world communication systems, equalization is used to eliminate
or minimize the distortion introduced when a signal is sent over a (nonideal) communication channel. In audio
applications, equalization can be employed to emphasize or de-emphasize certain ranges of frequencies. For example,
often we like to boost the bass (i.e., emphasize the low frequencies) in the audio output of a stereo.
Example 5.37 (Communication channel equalization). Suppose that we have a LTI communication channel with
frequency response H(ω ) = 3+1jω . Unfortunately, this channel has the undesirable effect of attenuating higher fre-
quencies. Find the frequency response G(ω ) of an equalizer that when connected in series with the communication
channel yields an ideal (i.e., distortionless) channel. The new system with equalization is shown in Figure 5.37.
Solution. An ideal communication channel has a frequency response equal to one for all frequencies. Consequently,
we want H(ω )G(ω ) = 1 or equivalently G(ω ) = 1/H(ω ). Thus, we conclude that
1 1
G(ω ) = = 1
= 3 + jω .
H(ω ) 3+ jω

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.18. PROBLEMS 163

5.18 Problems
5.1 Using the Fourier transform analysis equation, find the Fourier transform of each of the following signals:
(a) x(t) = Aδ (t − t0 ) where t0 and A are real and complex constants, respectively;
(b) x(t) = rect(t − t0 ) where t0 is a constant;
(c) x(t) = e−4t u(t − 1);
(d) x(t) = 3[u(t) − u(t − 2)]; and
(e) x(t) = e−|t| .

5.2 Use a Fourier transform table and properties of the Fourier transform to find the Fourier transform of each of
the signals below.
(a) x(t) = cos(t − 5);
(b) x(t) = e− j5t u(t + 2);
(c) x(t) = [cost]u(t);
(d) x(t) = 6[u(t) − u(t − 3)];
(e) x(t) = 1/t;
(f) x(t) = t rect(2t);
(g) x(t) = e− j3t sin(5t − 2);
(h) x(t) = cos(5t − 2);
(i) x(t) = e− j2t 3t+1
1
;
R 5t −τ −1
(j) x(t) = −∞ e u(τ − 1)d τ ;
(k) x(t) = (t + 1) sin(5t − 3);
(l) x(t) = (sin 2π t)δ (t − π2 );
(m) x(t) = e− jt 3t−2
1
;
j5t
(n) x(t) = e (cos 2t)u(t); and
(o) x(t) = e− j2t sgn(−t − 1).

5.3 Compute the Fourier transform X(ω ) of the signal x(t) given by
∞
x(t) = ∑ ak δ (t − kT ),
k=0

where a is a constant satisfying |a| < 1. (Hint: Recall the formula for the sum of an infinite geometric series.
b
That is, b + br + br2 + . . . = 1−r if |r| < 1.)

5.4 The ideal Hilbert transformer is a LTI system with the frequency response

− j for ω > 0

H(ω ) = 0 for ω = 0


j for ω < 0.

This type of system is useful in a variety of signal processing applications (e.g., SSB/SC amplitude modulation).
By using the duality property of the Fourier transform, find the impulse response h(t) of this system.
F F
5.5 Given that x(t) ←→ X(ω ) and y(t) ←→ Y (ω ), express Y (ω ) in terms of X(ω ) for each of the following:
(a) y(t) = x(at − b) where a and b are constants and a 6= 0;
R 2t
(b) y(t) = −∞ x(τ )d τ ;
Rt 2
(c) y(t) = −∞ x (τ )d τ ;
(d) y(t) = dtd [x(t) ∗ x(t)];
(e) y(t) = tx(2t − 1);
(f) y(t) = e j2t x(t − 1);

∗
(g) y(t) = te− j5t x(t) ;

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

164 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

   
(h) y(t) = dtd x(t) ∗ e− jt x(t) ;
R 3t ∗
(i) y(t) = −∞ x (τ − 1)d τ ;
(j) y(t) = [cos(3t
 −
 1)] x(t);
(k) y(t) = dtd x(t) sin(t − 2);
(l) y(t) = tx(t) sin 3t; and
(m) y(t) = e j7t [x(λ ) ∗ x(λ )]|λ =t−1 .
5.6 Find the Fourier transform of each of the periodic signals shown below.

x(t)
1 1
x(t)
1
··· ···
t ··· ···
−4 −3 −2 −1 1 2 3 4

t
−4 −3 −2 −1 0 1 2 3 4

−1 −1 (b)

(a)

5.7 Using the time-domain convolution property of the Fourier transform, compute the convolution h(t) = h1 (t) ∗
h2 (t) where
h1 (t) = 2000 sinc(2000π t) and h2 (t) = δ (t) − 1000 sinc(1000π t).

5.8 Compute the energy contained in the signal x(t) = 200 sinc(200π t).
5.9 Compute the frequency spectrum of each of the signals specified below. In each case, also find and plot the
corresponding magnitude and phase spectra.
(a) x(t) = e−at u(t), where a is a positive real constant; and
(b) x(t) = sinc t−1
200 .

5.10 Suppose that we have the LTI systems defined by the differential/integral equations given below, where x(t) and
y(t) denote the system input and output, respectively. Find the frequency response of each of these systems.
2
(a) dtd 2 y(t) + 5 dtd y(t) + y(t) + 3 dtd x(t) − x(t) = 0; and
Rt
(b) dtd y(t) + 2y(t) + −∞ 3y(τ )d τ + 5 dtd x(t) − x(t) = 0.
5.11 Suppose that we have the LTI systems with the frequency responses given below. Find the differential equation
that characterizes each of these systems.
jω
(a) H(ω ) = ; and
1 + jω
jω + 21
(b) H(ω ) = .
− jω 3 − 6ω 2 + 11 jω + 6
5.12 Suppose that we have a LTI system with input x(t) and output y(t), and impulse response h(t), where
h(t) = δ (t) − 300 sinc 300π t.
Using frequency-domain methods, find the response y(t) of the system to the input x(t) = x1 (t), where
x1 (t) = 12 + 43 cos 200π t + 12 cos 400π t − 14 cos 600π t.

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.18. PROBLEMS 165

5.13 Consider the LTI system with input v0 (t) and output v1 (t) as shown in the figure below, where R = 1 and L = 1.

R
i(t)
+ +

v0 (t) L v1 (t)

− −

(a) Find the frequency response H(ω ) of the system.

(b) Determine the magnitude and phase responses of the system.
(c) Find the impulse response h(t) of the system.
1
5.14 Consider the LTI system with input v0 (t) and output v1 (t) as shown in the figure below, where R = 1, C = 1000 ,
1
and L = 1000 .

L C
i(t)
+ +

v0 (t) R v1 (t)

− −

(a) Find the frequency response H(ω ) of the system.

(b) Use a computer to plot the magnitude and phase responses of the system.
(c) From the plots in part (b), identify the type of ideal filter that this system approximates.
5.15 Let x(t) be a real signal with Fourier transform X(ω ) satisfying X(ω ) = 0 for |ω | > ωb . We use amplitude
modulation to produce the signal y(t) = x(t) sin ωct. Note that ωc ≫ ωb . In order to recover the original signal
x(t), it is proposed that the system shown in the figure below be used. The frequency response H(ω ) is given
by
(
2 for |ω | < ωb
H(ω ) =
0 otherwise.

cos ωc t

y(t) v(t) x̂(t)

× H(ω )

Let Y (ω ), V (ω ), and X̂(ω ) denote the Fourier transforms of y(t), v(t), and x̂(t), respectively.
(a) Find an expression for Y (ω ) in terms of X(ω ). Find an expression for X̂(ω ) in terms of V (ω ). Find a
simplified expression for X̂(ω ).
(b) Compare x̂(t) and x(t). Comment on the utility of the proposed system.
5.16 When discussing DSB/SC amplitude modulation, we saw that a system of the form shown below in Figure A
is often useful. In practice, however, the multiplier unit needed by this system is not always easy to implement.
For this reason, we sometimes employ a system like that shown below in Figure B. In this second system, we
sum the sinusoidal carrier and modulating signal x(t) and then pass the result through a nonlinear squaring
device (i.e., v2 (t) = [v1 (t)]2 ).

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

166 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

cos ωc t cos ωc t

x(t) y(t) x(t) v1 (t) v2 (t) y(t)

× + (·)2 H(ω )

(a) (b)

Let X(ω ), V1 (ω ), and V2 (ω ) denote the Fourier transforms of x(t), v1 (t), and v2 (t), respectively. Suppose that
X(ω ) = 0 for |ω | > ωb (i.e., x(t) is bandlimited).
(a) Find an expression for v1 (t), v2 (t) and V2 (ω ). (Hint: If X(ω ) = 0 for |ω | > ωb , then using the time-domain
convolution property of the Fourier transform, we can deduce that the Fourier transform of x2 (t) is zero for
|ω | > 2ωb .)
(b) Determine the frequency response H(ω ) required for the system shown in Figure B to be equivalent to the
system in Figure A. State any assumptions made with regard to the relationship between ωc and ωb . (Hint: It
might be helpful to sketch X(ω ) and V2 (ω ) for the case of some simple X(ω ). Then, compare V2 (ω ) to X(ω )
in order to deduce your answer.)

5.17 Consider the system with input x(t) and output y(t) as shown in Figure A below. The frequency response H(ω )
is that of an ideal Hilbert transformer, which is given by

H(ω ) = − j sgn ω .

Let X(ω ), Y (ω ), V1 (ω ), V2 (ω ), and V3 (ω ) denote the Fourier transforms of x(t), y(t), v1 (t), v2 (t), and v3 (t),
respectively.

cos ωc t

X1 (ω )
v1 (t)
×
x(t) y(t) 1
+
v2 (t)
H(ω ) ×
v3 (t)
ω
−ωb ωb

sin ωc t (b)
(a)

(a) Suppose that X(ω ) = 0 for |ω | > ωb , where ωb ≪ ωc . Find expressions for V1 (ω ), V2 (ω ), V3 (ω ), and Y (ω )
in terms of X(ω ).
(b) Suppose that X(ω ) = X1 (ω ) where X1 (ω ) is as shown in Figure B. Sketch V1 (ω ), V2 (ω ), V3 (ω ), and Y (ω )
in this case.
(c) Draw the block diagram of a system that could be used to recover x(t) from y(t).

5.18 Consider the system shown below in Figure A with input x(t) and output x̂(t), where
(
2 for |ω | ≤ 100π
G(ω ) =
0 otherwise.

Let X(ω ), X̂(ω ), Y (ω ), and Q(ω ) denote the Fourier transforms of x(t), x̂(t), y(t), and q(t), respectively.

Copyright c 2013 Michael D. Adams Version: 2013-09-11

5.19. MATLAB PROBLEMS 167

X1 (ω )
sin 1000π t sin 1000π t
1

x(t) y(t) q(t) x̂(t)

× × G(ω )

(a) ω
−100π 0 100π
(b)

(a) Suppose that X(ω ) = 0 for |ω | > 100π . Find expressions for Y (ω ), Q(ω ), and X̂(ω ) in terms of X(ω ).
(b) If X(ω ) = X1 (ω ) where X1 (ω ) is as shown in Figure B, sketch Y (ω ), Q(ω ), and X̂(ω ).

5.19 Consider the system shown below in Figure A with input x(t) and output y(t). Let X(ω ), P(ω ), S(ω ), H(ω ),
and Y (ω ) denote the Fourier transforms of x(t), p(t), s(t), h(t), and y(t), respectively. Suppose that
∞
n 1 ω
p(t) = ∑ δ (t − 1000 ) and H(ω ) = 1000 rect( 2000 π ).
n=−∞

(a) Derive an expression for S(ω ) in terms of X(ω ). Derive an expression for Y (ω ) in terms of S(ω ) and H(ω ).
(b) Suppose that X(ω ) = X1 (ω ), where X1 (ω ) is as shown in Figure B. Using the results of part (a), plot S(ω )
and Y (ω ). Indicate the relationship (if any) between the input x(t) and output y(t) of the system.
(c) Suppose that X(ω ) = X2 (ω ), where X2 (ω ) is as shown in Figure C. Using the results of part (a), plot S(ω )
and Y (ω ). Indicate the relationship (if any) between the input x(t) and output y(t) of the system.

p(t)

x(t) s(t) y(t)

× h(t)

(a)
X1 (ω ) X2 (ω )

1 1

ω ω
−1000π 1000π −2000π 2000π
(b) (c)

5.19 MATLAB Problems

5.101 (a) Consider a frequency response H(ω ) of the form
M−1
∑k=0 ak ω k
H(ω ) = N−1
,
∑k=0 bk ω k

where ak and bk are complex constants. Write a MATLAB function called freqw that evaluates a function of
the above form at an arbitrary number of specified points. The function should take three input arguments: 1) a
vector containing the ak coefficients, 2) a vector containing the bk coefficients, 3) a vector containing the values
of ω at which to evaluate H(ω ). The function should generate two return values: 1) a vector of function values,
and 2) a vector of points at which the function was evaluated. If the function is called with no output arguments

Version: 2013-09-11 Copyright c 2013 Michael D. Adams

168 CHAPTER 5. CONTINUOUS-TIME FOURIER TRANSFORM

(i.e., the nargout variable is zero), then the function should plot the magnitude and phase responses before
returning. [Hint: The polyval function may be helpful.]
(b) Use the function developed in part (a) to plot the magnitude and phase responses of the system with the
frequency response

16.0000
H(ω ) = .
1.0000ω 4 − j5.2263ω 3 − 13.6569ω 2 + j20.9050ω + 16.0000

For each of the plots, use the frequency range [−5, 5].
(c) What type of ideal frequency-selective filter does this system approximate?

5.102 Consider the filter associated with each of the frequency responses given below. In each case, plot the magnitude
and phase responses of the filter, and indicate what type of ideal frequency-selective filter it best approximates.
ωb3
(a) H(ω ) = where ωb = 1;
( jω )3 + 2ωb ( jω )2 + 2ωb ( jω ) + ωb3
( jω )5
(b) H(ω ) = ; and
( jω )5 + 17.527635( jω )4 + 146.32995( jω )3 + 845.73205( jω )2 + 2661.6442( jω ) + 7631.0209
13.104406( jω )3
(c) H(ω ) = .
( jω ) + 3.8776228( jω ) + 34.517979( jω ) + 75.146371( jω )3 + 276.14383( jω )2 + 248.16786( jω ) + 512
6 5 4
Hint: Use the freqs function with s = jω to compute the frequency response. The abs, angle, linspace,
plot, xlabel, ylabel, and print functions may also prove useful for this problem.

5.103 (a) Use the butter and besself functions to design a tenth-order Butterworth lowpass filter and tenth-order
Bessel lowpass filter, each with a cutoff frequency of 10 rad/s.
(b) For each of the filters designed in part (a), plot the magnitude and phase responses using a linear scale for
the frequency axis. In the case of the phase response, plot the unwrapped phase (as this will be helpful later in
part (d) of this problem). (Hint: The freqs and unwrap functions may be helpful.)
(c) Consider the magnitude responses for each of the filters. Recall that an ideal lowpass filter has a magnitude
response that is constant in the passband. Which of the two filters more closely approximates this ideal behav-
ior?
(d) Consider the phase responses for each of the filters. An ideal lowpass filter has a phase response that is a
linear function. Which of the two filters has a phase response that best approximates a linear (i.e., straight line)
function in the passband?

Copyright c 2013 Michael D. Adams Version: 2013-09-11