Lecture Notes

for Engineering Mathematics

Chapter 6
The Fourier Integral and Fourier Transforms

By
Wenson Chang
Department of Electric Engineering
National Cheng-Kung University

July, 2015
c
Copyright 2015 by Wenson Chang
Contents

6 The Fourier Integral and Fourier Transforms 1

6.1 The Fourier Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
6.2 Fourier Cosine and Sine Integrals . . . . . . . . . . . . . . . . . . . . 4
6.3 The Complex Fourier Integral and the Fourier Transform . . . . . . . 7
6.4 Additional Properties and Applications . . . . . . . . . . . . . . . . . 18
6.4.1 The Fourier Transform of a Derivative . . . . . . . . . . . . . 18
6.4.2 Frequency Differentiation . . . . . . . . . . . . . . . . . . . . . 21
6.4.3 The Fourier Transform of an Integral . . . . . . . . . . . . . . 22
6.4.4 Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
6.4.5 Filtering and the Dirac Delta Function . . . . . . . . . . . . . 28
6.4.6 The Windowed Fourier Transform . . . . . . . . . . . . . . . . 29
6.4.7 The Shannon Sampling Theorem . . . . . . . . . . . . . . . . 34
6.4.8 Lowpass and Bandpass Filters . . . . . . . . . . . . . . . . . . 36
6.5 The Fourier Cosine and Sine Transform . . . . . . . . . . . . . . . . . 40
6.6 The Finite Fourier Cosine and Sine Transforms . . . . . . . . . . . . 43

Bibliography 47
 1

CHAPTER 6

The Fourier Integral and Fourier

Transforms

6.1 The Fourier Integral

In this chapter, we should first know the conditions for using Fourier transform in
stead of using Fourier series.

• Fourier series:

– If f (x) is defined on an interval [−L, L], we may be able to represent it, at

least at ”most” points on this interval, by a Fourier series.

– If f is periodic, then we may be able to represent it by a Fourier series on

intervals along the entire real line.

• Fourier transform:

– Now suppose f (x) is defined for all x, but is not periodic, i.e. −∞ < x <
∞. Then we cannot represent f (x) by a Fourier series over the entire line.

– However, we may still be able to write a representation in terms of sines

and cosines, using an integral instead of a summation with the conditions:
R∞
∗ f is absolutely integrable, which means that −∞ |f (x)|dx converges.
∗ f is piecewise smooth on every interval [−L, L].

1
The main reference, i.e. the text book, of this course is [1]; Whereas there are still some good
references for this course [2–7].
2 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

Write the Fourier series of f on an arbitrary interval [−L, L], with the integral for-
mulas for the coefficients included:
Z L X∞  Z 
1 1 L
f (ξ)dξ + f (ξ) cos (nπξ/L)dξ cos (nπx/L)
2L −L n=1
L −L
 Z L  
1
+ f (ξ) sin (nπξ/L)dξ sin (nπx/L) .
L −L

We want to let L → ∞ to obtain a representation of f (x) over the whole line.

• let ωn = nπ/L and ωn − ωn−1 = π/L = ∆ω. Then the Fourier series on [−L, L]
can be written
Z L  ∞ Z L 
1 1X
f (ξ)dξ ∆ω + f (ξ) cos (ωn ξ)dξ cos (ωn x)
2π −L π n=1 −L

Z L  
+ f (ξ) sin (ωn ξ)dξ sin (ωn x) ∆ω . (6.1)
−L

• Now let L → ∞, causing ∆ω → 0. In the last expression,

Z L 
1
f (ξ)dξ ∆ω → 0
2π −L

RL
because by assumption −L
f (ξ)dξ converges.

• The other terms in the expression (15.1) resemble a Riemann sum for a definite
integral, and we assert that, in the limit as L → ∞ and ∆ω → 0, this expression
approaches the limit
Z Z ∞  Z ∞  
1 ∞
f (ξ) cos (ωξ)dξ cos (ωx) + f (ξ) sin (ωξ)dξ sin (ωx) dω .
π 0 −∞ −∞

This is the Fourier integral of f on the real line. Under the assumption made
about f , this integral converges to
1
(f (x− ) + f (x+ ))
2
at each x. In particular, if f is continuous at x, then this integral converges to
f (x).

• Often this Fourier integral is written

Z ∞
[Aω cos (ωx) + Bω sin (ωx)]dω , (6.2)
0
6.1. THE FOURIER INTEGRAL 3


1 for − 1 ≤ x ≤ 1
FIGURE 6.1: f (x) = .
0 for |x| > 1

in which the Fourier integral coefficients of f are

Z
1 ∞
Aω = f (ξ) cos (ωξ)dξ
π −∞
and Z ∞
1
Bω = f (ξ) sin (ωξ)dξ .
π −∞

This Fourier integral representation of f (x) is entirely analogous to a Fourier

R∞ P
series on an interval, with 0 ...dω replacing ∞ n=1 , and having integral formulas
for the coefficients.

Example 6.1. Let (

1 for − 1 ≤ x ≤ 1
f (x) = .
0 for |x| > 1
R∞
• Figure 15.1 is a graph of f . Certainly f is piecewise smooth, and −∞
|f (x)|dx
converges. The Fourier coefficients of f are
Z
1 1 2 sin (ω)
Aω = cos (ωξ)dξ =
π −1 πω
and Z ∞
1
Bω = f (ξ) sin (ωξ)dξ = 0 .
π −∞

The Fourier integral of f is

Z ∞
2 sin (ω)
cos (ωx)dω .
0 πω

• Because f is piecewise smooth, this converges to 12 (f (x− ) + f (x+ )) for all x.

More explicitly

Z ∞

 1 for − 1 < x < 1
2 sin (ω) 1
cos (ωx)dω = for x = ±1 .
0 πω 

2
0 for |x| > 1
4 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

Compact Expression of Fourier Integral

There is an another expression for the Fourier integral of a function that we will
sometimes find convenient. Write
Z ∞ Z  Z ∞
∞ 
1
[Aω cos (ωx) + Bω sin (ωx)]dω = f (ξ) cos (ωξ)dξ cos (ωx)
0 0 π −∞
 Z ∞  
1
+ f (ξ) sin (ωξ)dξ sin (ωx) dω
π −∞
Z Z
1 ∞ ∞
= f (ξ)[cos (ωξ) cos (ωx)
π 0 −∞

+ sin (ωξ) sin (ωx)]dξdω

Z Z
1 ∞ ∞
= f (ξ) cos (ω(ξ − x))dξdω .
π 0 −∞

(6.3)

Of course, this integral has the same convergence properties as the integral expression
(6.2), since it is just a rearrangement of that integral.

6.2 Fourier Cosine and Sine Integrals

R∞
If f is piecewise smooth on the half-line [0, ∞), and 0
|f (ξ)|dξ converges, then we
can write a Fourier cosine or sine integral for f that is completely analogous to sine
and cosine expansions of a function on an interval [0, L].

Fourier Cosine

• To write a cosine integral, extend f to an even function fe defined on the whole

line by setting (
f (x) for x ≥ 0
fe (x) = .
f (−x) for x < 0

– Since fe is an even function, its Fourier integral has only cosine terms.

• Since fe (x) = f (x) for x ≥ 0, this cosine integral can be defined to be the
Fourier cosine integral of f on [0, ∞). The coefficient of fe in its Fourier integral
expansion is Z
1 ∞
fe (ξ) cos (ωξ)dξ
π −∞
6.2. FOURIER COSINE AND SINE INTEGRALS 5

and this is Z ∞
2
f (ξ) cos (ωξ)dξ
π 0

• The above suggests the following definition.

Definition 6.1. Fourier Cosine Integral

R∞
Let f be defined on [0, ∞) and let 0
|f (ξ)|dξ converge. The Fourier cosine in-
tegral of f is
Z ∞
Aω cos (ωx)dω ,
0

in which
Z ∞
2
Aω = f (ξ) cos (ωξ)dξ . 
π 0

• By applying the convergence theorem to the integral expansion of fe , we find

that:

– If f is piecewise continuous on each interval [0, L], then its cosine integral
expansion converges to 12 (f (x− ) + f (x+ )) for each x > 0, and to f(0) for
x = 0.

– In particular, at any positive x at which f is continuous, the cosine integral

converges to f (x).

Fourier Sine Integral

• By extending f to an odd function fo , i.e.

(
f (x) for x ≥ 0
fo (x) = .
−f (−x) for x < 0

Similar to what we did with series, we obtain a Fourier integral for fo containing
only sine terms.

• Since fo (x) = f (x) for x ≥ 0, this gives a sine integral for f on [0, ∞).

Definition 6.2. Fourier Sine Integral

6 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS
R∞
Let f be defined on [0, ∞) and let 0
|f (ξ)|dξ converge. The Fourier sine inte-
gral of f is
Z ∞
Aω sin (ωx)dω
0
in which
Z ∞
2
Aω = f (ξ) sin (ωξ)dξ . 
π 0

• If f is piecewise smooth on every interval [0, L]:

– This integral converges to 12 (f (x− ) + f (x+ )) on (0, ∞).

– This Fourier sine integral converges to 0 at x = 0.

Example 6.2. Laplace’s Integrals

Fourier Cosine

• Let f (x) = e−kx or x ≥ 0, with k a positive constant. Then f is continuously

differentiable on any interval [0, L], and integrable, which means
Z ∞
1
e−kx dx =
0 k

• For the Fourier cosine integral, compute the coefficients

Z
2 ∞ −kξ 2 k
Aω = e cos (ωξ)dξ = .
π 0 π k + ω2
2

Note that we can have two methods to conduct this calculation.

– By partial integration: Let u = e−kξ and dv = cos(wξ)dξ, which leads to

du = −ke−kξ dξ and v = w1 sin(wξ). Then, you will need a 2nd round of
partial integration and next get the results.
– Rewrite cos(wξ) as (ejwξ + e−jwξ )/2. After some derivations, the result can
be reached.

• The Fourier cosine integral representation of f converges to e−kx for x ≥ 0:

Z
−kx 2k ∞ 1
e = cos (ωx)dω .
π 0 k2 + ω2
Fourier Sine
6.3. THE COMPLEX FOURIER INTEGRAL AND THE FOURIER TRANSFORM7

• For the sine integral, compute the following equation using the same method as
above Z
2 ∞ −kξ 2 ω
Bω = e sin (kξ)dξ = .
π 0 π k + ω2
2

• The sine integral converges to e−kx for x > 0 and to 0 for x = 0:

Z
−kx 2 ∞ ω
e = sin (ωx)dω for x > 0
π 0 k + ω2
2

• These integral representations are called Laplace’s integrals because Aω is 2/π

times the Laplace transform of sin(kx), while Bω is 2/π times the Laplace
transform of cos(kx). Note that the general form of Laplace transform is
Z ∞
L[f ](s) = f (t)e−st dt .  (6.4)
0

6.3 The Complex Fourier Integral and the

Fourier Transform
It is sometimes convenient to have a complex form of the Fourier integral. This
complex setting will prove a natural platform from which to develop the Fourier
transform.
R∞
• Suppose f is piecewise smooth on each interval [−L, L], and that −∞
|f (x)|dx
converges. Then, at any x,
Z ∞ Z ∞
1 1
(f (x− ) + f (x+ )) = f (ξ) cos (ω(ξ − x))dξdω ,
2 π 0 −∞

by the expression (6.3).

• Insert the complex exponential form of the cosine function into this expression
to write
Z ∞ Z ∞
1 1 1
(f (x− ) + f (x+ )) = f (ξ) (eiω(ξ−x) + e−iω(ξ−x) )dξdω
2 π0 −∞ 2
Z ∞Z ∞
1
= f (ξ)eiω(ξ−x)dξdω +
2π 0 −∞
Z ∞Z ∞
1
f (ξ)e−iω(ξ−x)dξdω .
2π 0 −∞
8 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

• In the first integral on the last line, put ω = −w to get

Z 0 Z ∞
1 − + 1
(f (x ) + f (x )) = f (ξ)e−iω(ξ−x)dξdω +
2 2π −∞ −∞
Z ∞Z ∞
1
f (ξ)e−iω(ξ−x) dξdω . (6.5)
2π 0 −∞

• Now write the variable of integration in the next to last integral as ω again and
combine these two integrals to write
Z ∞ Z ∞
1 1
(f (x− ) + f (x+ )) = f (ξ)e−iω(ξ−x) dξdω . (6.6)
2 2π −∞ −∞

• This is the complex Fourier integral representation of f on the real line. If we

R∞
let Cω = −∞ f (t)e−iωt dt, then this integral is
Z ∞
1 − + 1
(f (x ) + f (x )) = Cω eiωx dω .
2 2π −∞
We call Cω the complex Fourier integral coefficient of f , which is the so-called
Fourier transform. To summarize, we can now formally define the Fourier
transform as following.

Definition 6.3. Fourier Transform

R∞
Suppose −∞
|f (x)|dx converges. Then the Fourier transform of f is defined to
be the function

Z ∞
F[f ](ω) = f (t)e−iωt dt .  (6.7)
−∞

• Furthermore, if f is continuous, and f ′ is piecewise continuous on every interval

[−L, L], then the Fourier integral of f represents f :
Z ∞
1
f (t) = fˆ(ω)eiωt dω . (6.8)
2π −∞
We can therefore use equation (6.8) as an inverse Fourier transform, retrieving f
from fˆ. Also, we write ℑ−1 [fˆ] = f if ℑ[f ] = fˆ; And (6.7) and (6.8) constitute a
transform pair for the Fourier transform. As we expect of any integral transform,
F is linear:
F[αf + βg] = αF[f ] + βF[g] .
6.3. THE COMPLEX FOURIER INTEGRAL AND THE FOURIER TRANSFORM9

• In the context of the Fourier transform, the amplitude spectrum is often taken
to be a graph of |fˆ(ω)| [or written as |F[f ](ω)|].

Example 6.3. Let f (x) = e−a|x| for all real x, with a a positive constant. We will
compute the complex Fourier integral representation of f .

• First, we have (
e−ax for x ≥ 0
f (x) =
eax for x < 0
Further, Z Z Z
∞ 0 ∞
ax 2
f (x)dx = e dx + e−ax dx =
−∞ −∞ 0 a
• Now compute
Z ∞
Cω = e−a|t| e−iωt dt
−∞
Z 0 Z ∞
at −iωt
= e e dt + e−at e−iωt dt
−∞ 0
Z 0 Z ∞
(a−iω)t
= e dt + e−(a+iω)t dt
−∞ 0

 0  ∞
1 (a−iω)t −1 −(a+iω)t
= e + e
a − iω −∞ a + iω 0
 
1 1 2a
= + = 2 .
a + iω a − iω a + ω2

• At last, we obtain the complex Fourier integral representation of f is

Z
−a|x| a ∞ 1
e = 2 2
eiωx dω . 
π −∞ a + ω

Example 6.4. Let a be a positive constant. Then

2a
F[e−a|t| ](ω) =
a2 + ω2
This follows immediately from Example 6.3, where we calculated the Fourier integral
coefficient Cω of e−a|t| . This coefficient is the Fourier transform of f . 

Example 6.5. Let a and k be positive numbers, and let

(
k for − a ≤ x < a
f (x) = .
0 for t < −a and for t ≥ a
10 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

FIGURE 6.2

• This pulse function can be written in terms of the Heaviside function as

f (t) = k[H(t + a) − H(t − a)] ,
and is graphed in Figure 6.2.

• The Fourier transform of f is

Z ∞
F[f ](ω) = fˆ(ω) = f (t)e−iωt dt
−∞
Z a  a
−iωt −k iωt
= ke dt = e
−a iω −a

k −iωa 2k
= − [e − eiωa ] = sin (aω) . 
iω ω
Example 6.6. Let
(
1 − |t| for − 1 ≤ t ≤ 1
f (t) =
0 for t > 1 and for t < −1
Then f is continuous and absolutely integrable, and f ′ is piecewise continuous.
• Compute
Z ∞
fˆ(ω) = f (t)e−iωt dt
−∞
Z 1
2(1 − cos (ω))
= (1 − |t|)e−iωt dt = .
−1 ω2
This is the Fourier coefficient Cω in the complex Fourier expansion of f (t).

• If we want to go the other way, then by equation (6.8),

Z ∞
1
f (t) = fˆ(ω)eiωt dω
2π −∞
Z
1 ∞ (1 − cos (ω)) iωt
= e dω .
π −∞ ω2
6.3. THE COMPLEX FOURIER INTEGRAL AND THE FOURIER TRANSFORM11

 
 
FIGURE 6.3: Graph of fˆ(ω) = √ 1
a2 +ω 2
, with f (t) = H(t)e−αt .

• We can verify this by explicitly carrying out this integration. A software package
yields
Z
1 ∞ (1 − cos (ω)) iωt
e dω
π −∞ ω2

= πtsignum(t + 1) + πsignum(t + 1) + πtsignum(t − 1)

−πsignum(t − 1) − 2signum(t)

in which 

 1 for ω > 0
signum(ω) = 0 for ω = 0


−1 for ω < 0
This expression is equal to 1 − |t| for −1 ≥ t ≥ 1, and 0 for t > 1 and for
t < −1, verifying the integral for the inverse. 

Example 6.7. If f (t) = H(t)e−at , then fˆ(ω) = 1/(a + iω), so

  1
ˆ 
f (ω) = √ 2
a + ω2

Figure 6.3 shows a graph of |fˆ(ω)|. This graph is the amplitude spectrum of f . 

Example 6.8. The amplitude spectrum of the function f of Example 15.5 is a graph
of    
ˆ   sin (ω) 

f (ω) = 2k  
ω 
shown in Figure 6.4. 
R∞
In the following theorem, we assume that −∞ |f (t)|dt converges, and, for the
inverse version, that f is continuous and f ′ is piecewise continuous on each [−L, L].
12 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

FIGURE 6.4

Theorem 6.1. Time Shifting

If t0 is a real number, then

F[f (t − t0 )](ω) = e−iωt0 fˆ(w) . 

Proof 6.1.
Z ∞
F[f (t − t0 )](ω) = f (t − t0 )e−iωt dt
−∞
Z ∞
−iωt0
= e f (t − t0 )e−iω(t−t0 ) dt .
−∞

Let u = t − t0 to write
Z ∞
−iωt0
F[f (t − t0 )](ω) = e f (u)e−iωu du = e−iωt0 fˆ(ω) . 
−∞

Note that the inverse version of the time-shifting theorem is:

F−1 [e−iωt0 F (ω)](t) = f (t − t0 ) (6.9)

Example 6.9. Suppose we want the Fourier transform of the pulse of amplitude 6
which turns on at time 3 and off at time 7. This is the function
(
0 for t < 3 and for t ≥ 7
g(t) =
6 for 3 ≤ t < 7

shown in Figure 6.5.

• We can certainly compute ĝ(ω) by integration. But we can also observe that
the midpoint of the pulse (that is, of the nonzero part) occurs when t = 5.
6.3. THE COMPLEX FOURIER INTEGRAL AND THE FOURIER TRANSFORM13


0 for t < 3 and for t ≥ 7
FIGURE 6.5: g(t) = .
6 for 3 ≤ t < 7

FIGURE 6.6: The function of Figure 15.5 shifted 5 units to the left.

• Shift the graph 5 units to the left to center the pulse at zero (Figure 6.6).
Calling this shifted pulse f , then f (t) = g(t + 5).

• Shift f five units to the right again just get us back to g:

g(t) = f (t − 5) .

• The point to this is that we already know the Fourier transform of f from
Example 6.5:
sin (2ω)
F[f (t)](ω) = 12
ω
Note that from Example 6.5, we obtain
2k
k[H(t + a) − H(t − a)] ⇋ sin(aw) .
w

• By the time-shifting theorem,

sin (2ω)
F[g(t)](ω) = F[f (t − 5)](ω) = 12e−5iω .
ω

Example 6.10. Suppose we want

 
−1 e2iω
F
5 + iω
14 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

FIGURE 6.7: Graph of H(t + 2)e−5(t+2) .

• The presence of the exponential factor suggests the inverse version of the time-
shifting theorem.

• Put t0 = −2 in equation (6.9) to write

 
−1 e2iω
F = f (t − (−2)) = f (t + 2) ,
5 + iω

where  
−1 1
f (t) = F = H(t)e−5t .
5 + iω
Note that from Example 6.7, we know that

1
H(t)e−at ⇋ .
a + iw

• Therefore  
−1 e2iω
F = f (t + 2) = H(t + 2)e−5(t+2) .
5 + iω
A graph of this function is shown in Figure 6.7. 

The next result is reminiscent of the first shifting theorem for the Laplace
transform.

Theorem 6.2. Frequency Shifting

If ω0 is any real number, then

F[eiω0 t f (t)] = fˆ(ω − ω0 ) . 

6.3. THE COMPLEX FOURIER INTEGRAL AND THE FOURIER TRANSFORM15

Proof 6.2.
Z ∞
iωt
F[e f (t)](ω) = eiω0 t f (t)e−iωt dt
−∞
Z ∞
= f (t)e−i(ω−ω0 )t dt = fˆ(ω − ω0 ) . 
−∞

The inverse version of the frequency-shifting theorem is

F−1 [fˆ(ω − ω0 )](t) = eiω0 t f (t) .

Theorem 6.3. Scaling

If a is a nonzero real number, then

1 ˆ ω 
F[f (at)](ω) = f .
|a| a

Proof 6.3. The proof can be conducted by two parts.

• a > 0: it is intuitive, as following.

Z ∞ Z ∞
−jwt 1
f (t)e dt = f (at)e−j(w/a)(at) d(at)
−∞ −∞ a
Z
1 ∞
= f (at)e−j(w/a)(at) d(at)
a −∞
1 ˆ ω 
= f .
a a

• a < 0:
Z ∞ Z −∞
−jwt 1
f (t)e dt = f (at)e−j(w/a)(at) d(at)
−∞ ∞ a
Z ∞
−1
= f (at)e−j(w/a)(at) d(at)
−∞ a
Z ∞
1
= f (at)e−j(w/a)(at) d(at)
|a| −∞

1 ˆ ω 
= f .
|a| a
Summarize these two parts, we can obtain the results as wished. 
16 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

The inverse transform version of this result is

h  ω i
F−1 fˆ (t) = |a|f (at) .
a
Theorem 6.4. Time Reversal

F[f (−t)](ω) = fˆ(−ω) . 

Proof 6.4. The proof is straight forward by putting a = −1 into Thm. 6.3. 

Example 6.11. We know from Example 6.6 that, if

(
1 − |t| for − 1 ≤ t ≤ 1
f (t) = ,
0 for t > 1 and for t < −1

then
1 − cos (ω)
fˆ(ω) = 2
ω2
Let (
−1 1
1 − |7t| for 7
≤t≤ 7
g(t) = f (7t) = 1 −1
.
0 for t > 7
and for t < 7

Then
1 ω 
ĝ(ω) = F[f (7t)](ω) = fˆ
7 7
ω


2 1 − cos (ω/7) 1 − cos 7
= 2
= 14 .
7 (ω/7) ω2

Theorem 6.5. Symmetry

F[fˆ(t)](ω) = 2πf (−ω) . 

Proof 6.5. The inverse Fourier transform is expressed as

Z ∞
1
f (t) = fˆ(w)eiwt dw .
2π −∞
Now, let t = −w̄ and multiply the above equation by 2π then
Z ∞
2πf (−w̄) = fˆ(w)e−iww̄ dw ,
−∞
6.3. THE COMPLEX FOURIER INTEGRAL AND THE FOURIER TRANSFORM17

which can be rewritten as

Z ∞
2πf (−w̄) = fˆ(t)e−itw̄ dt .
−∞

Put w̄ = w and have

Z ∞
2πf (−w) = fˆ(t)e−iwt dt = F[fˆ(t)] . 
−∞

Example 6.12. Let

(
4 − t2 for − 2 ≤ t ≤ 2
f (t) = .
0 for t > 2 and for t < −2
Figure 6.8 shows a graph of f . The Fourier transform of f is
Z ∞ Z 2
ˆ
f (ω) = f (t)e−iωt
dω = (4 − t2 )e−iωt dt
−∞ −2

sin (2ω) − 2ω cos (2ω)

= 4
ω3
• Note that
Z 2
t2 e−iωt dt (6.10)
−2

can be calculated by using partial integration method for two times. Firstly, let
u = t2 and dv = e−iwt dt. Secondly, let u = t and dv = e−iwt dt.

• In this example, f (−t) = f (t), so exchanging −ω for ω should not make any
difference in fˆ(ω), and we can see that this is indeed the case. Also, applying
the Symmetry property on this example gives
 
sin (2t) − 2t cos (2t) π
F 3
(w) = (4 − w 2 ) .  (6.11)
t 2

Theorem 6.6. Modulation

If ω0 is a real number, then

1
F[f (t) cos (ω0 t)](ω) = [fˆ(ω + ω0 ) + fˆ(ω − ω0 )]
2
and

1
F[f (t) sin (ω0 t)](ω) = i[fˆ(ω + ω0 ) − fˆ(ω − ω0 )] . 
2
18 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS


4 − t2 for − 2 ≤ t ≤ 2
FIGURE 6.8: f (t) = .
0 for |t| > 2

Proof 6.6. Put cos(ωt) = 12 (eiω0 t +e−iω0 t ) and use the linearity of F and the frequency-
shifting theorem to get
 
1 iω0 t 1 −iω0 t
F[f (t) cos (ω0 t)](ω) = F e f (t) + e f (t) (ω)
2 2
1 iω0 t 1
= F[e f (t)](ω) + F[e−iω0 t f (t)](ω)
2 2
1 ˆ
= [f (ω + ω0 ) + fˆ(ω − ω0 )]
2
1
The second conclusion is proved similarly, using sin(ωt) = 2i
(eiω0 t − e−iω0 t ). 

6.4 Additional Properties and Applications of

the Fourier Transform
6.4.1 The Fourier Transform of a Derivative
In using the Fourier transform to solve differential equations, we need an expression
relating the transform of f ′ to that of f . The following theorem provides such a
relationship for derivatives of any order, and is called the operational rule for the
Fourier transform. A similar issue arises for any integral transform when it is to
be used in connection with differential equations (as in Theorem 3.5 and 3.6 for the
Laplace transform).
Recall that the k th derivative of f is denoted f (k) . As a convenience we may
let k = 0 in this symbol, with the understanding that f (0) = f .

Theorem 6.7. Differentiation in the Time Variable

6.4. ADDITIONAL PROPERTIES AND APPLICATIONS 19

Let n be a positive integer. Suppose f (n−1) is continuous, and f (n) is piecewise con-
R∞
tinuous on each interval [−L, L]. Suppose −∞ |f (n−1) (t)|dt converges. Suppose

lim f (k) (t) = lim f (k) (t) = 0

t→∞ t→−∞

for k = 0, 1, ..., n − 1. Then

F[f (n) (t)](ω) = (iω)n fˆ(ω) . 

Proof 6.7.

• Begin with the first derivative. Integrating by parts with u = e−iwt , dv = f ′ (t)dt,
du = −iwe−iwt and v = f (t), we have
Z ∞
′
F[f ](ω) = f ′ (t)e−iωt dt
−∞
Z ∞
= [f (t)e−iωt ]∞
−∞ − f (t)(−iω)e−iωt dt .
−∞

• Now e−iωt = cos(ωt) − i sin(ωt) has magnitude 1, and by assumption,

lim f (t) = lim f (t) = 0 .

t→∞ t→−∞

Therefore Z ∞
′
F[f (t)](ω) = iω f (t)e−iωt dt = (iω)fˆ(ω) .
−∞

• The conclusion for higher derivatives follows by inducation on n and the fact
that.
d (n−1)
f (n) (t) = f (t) . 
dt

Theorem 6.8.

Suppose f is continuous on the real line, except for jump discontinuities at t1 , ..., tM .
R∞
Let f ′ be piecewise continuous on every [−L, L]. Assume that −∞ |f (t)|dt converges,
and that

lim f (t) = lim f (t) = 0 .

t→∞ t→−∞
20 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

FIGURE 6.9: The function f has a jump discontinuity at tj .

Then

M
X
F[f ](ω) = iω fˆ(ω) −
′
[f (t+ −
j ) − f (tj )]e
−itj ω
.
j=1

Proof 6.8.

• Suppose first that f has a single jump discontinuity, at t1 .

– In the event of more jump discontinuities, the argument proceeds along the
same lines, but includes more of the type of the calculation we are about
to do.

• Integrate by parts:
Z ∞
′
F[f ](ω) = f ′ (t)e−iωt dt
−∞
Z t1 Z ∞
′ −iωt
= f (t)e dt + f ′ (t)e−iωt dt
−∞ t1

(u = e−iwt , dv = f ′ (t)dt, du = −iwe−iwt and v = f (t))

Z t1
−iωt t1
= [f (t)e ]−∞ − f (t)(−iω)e−iωt dt
−∞
Z ∞
+[f (t)e−iωt ]∞
t1 − (−iω) f (t)e−iωt dt
t1
Z ∞
= f (t−
1 )e
−it1 ω
− f (t+
1 )e
−it1 ω
+ iω f (t)e−iωt dt
−∞

= iω fˆ(ω) − [f (t+ −
1 ) − f (t1 )]e
−it1 ω
.
6.4. ADDITIONAL PROPERTIES AND APPLICATIONS 21

FIGURE 6.10: y(t) = − 81 e−4|t| .

Example 6.13. Solve

y ′ − 4y = H(t)e−4t ,

in which H is the Heaviside function. Thus the differential equation is

(
′ e−4t for t ≥ 0
y − 4y = .
0 for t < 0

• Apply the Fourier transform to the differential equation to get

F[y ′](ω) − 4ŷ(ω) = F[H(t)e−4t ](ω) .

• Using Theorem 6.7 and the fact that H(t)e−at ⇋ 1/(a+ iw), write this equation
as
1
iω ŷ(ω) − 4ŷ(ω) = .
4 + iω

• Solve for ŷ(ω) to obtain

−1
ŷ(ω) = .
16 + ω 2

• With e−a|t| ⇋ 2a/(a2 + w 2 ). The solution is

 
−1 −1 1 −4|t|
y(t) = F (t) = − e ,
16 + ω 2 8

which is graphed in Figure 6.10. 

6.4.2 Frequency Differentiation

In this context, differentiation of fˆ(ω) with respect to ω is called frequency differen-
tiation. We will now relate derivatives of fˆ(ω) and f (t).
22 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

Theorem 6.9. Frequency Differentiation

Let n be a positive integer. Let f be piecewise continuous on [−L, L] for every positive
R∞
number L, and assume that −∞ |tn f (t)|dt converges. Then

dn ˆ
F[tn f (t)](ω) = in f (ω) . 
dω n

In particular, under the conditions of the theorem,

d ˆ d2
F[tf (t)](ω) = i f (ω) and F[t2 f (t)](ω) = − 2 fˆ(ω) .
dω dω
Proof 6.9. We will prove the theorem for n = 1. The argument of larger n is similar.
Apply Leibniz’s rule for differentiation under the integral to write
Z ∞ Z ∞
d ˆ d −iωt ∂
f (ω) = f (t)e dt = [f (t)e−iωt ]dt
dω dω −∞ −∞ ∂ω
Z ∞ Z ∞
−iωt
= f (t)(−it)e dt = −i [tf (t)]e−iωt dt
−∞ −∞

= −iF[tf (t)](ω) . 

Example 6.14. Suppose we want to compute F[t2 e−5|t| ]. Recall from Example 6.4
that
2a
F[e−a|t| ](ω) = 2 .
a + ω2
Then
10
F[e−5|t| ](ω) =
25 + ω 2
By the frequency differentiation theorem,
2
 
2 −5|t| 2 d 10 25 − 3ω 2
F[t e ](ω) = i = 20 .
dω 2 25 + ω 2 (25 + ω 2)3

6.4.3 The Fourier Transform of an Integral

Theorem 6.10.
6.4. ADDITIONAL PROPERTIES AND APPLICATIONS 23
R∞
Let f be piecewise continuous on every interval [−L, L]. Suppose −∞
|f (t)|dt con-
verges. Suppose fˆ(0) = 0. Then
Z t 
1 ˆ
F f (τ )dτ (ω) = f (ω) . 
−∞ iω

Rt
Proof 6.10. Let g(t) = −∞
f (τ )dτ . then g ′(t) = f (t) for any t at which f is
continuous, and g(t) → 0 as t → − ∞. Further,
Z ∞
lim g(t) = f (τ )dτ = fˆ(0) = 0
t→∞ −∞

We can therefore apply Theorem 6.7 to g to obtain

fˆ(ω) = F[f (t)](ω) = F[g ′ (t)](ω)

Z t 
= iωF[g(t)](ω) = iωF f (τ )dτ (ω)
−∞

This is equivalent to the conclusion to be proved. 

6.4.4 Convolution
We saw a convolution for the Laplace transform in Chapter 3. We will now discuss
convolution for the Fourier transform.

Definition 6.4. Convolution

Let f and g be functions defined on the real line. Then f has a convolution with g if
Rb Rb
1. a
f (t)dt and a
g(t)dt exist for every interval [a, b].

2. For every real number t,

Z ∞
|f (t − τ )g(τ )|dτ
−∞

converges.

In this event, we define the convolution f ∗ g of f with g to be the function given by

Z ∞
(f ∗ g)(t) = f (t − τ )g(τ )dτ . 
−∞
24 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

Theorem 6.11.

Suppose f has a convolution with g. Then

1. (Commutativity of Convolution) g has a convolution with f , and f ∗ g =

g ∗ f.

2. (Linearity) If f and g both have convolutions with h, and α and β are real
numbers, then αf + βg also has a convolution with h, and

(αf + βg) ∗ h = α(f ∗ h) + β(g ∗ h) . 

Proof 6.11. For (1), let z = t − τ to write

Z ∞
f ∗ g(t) = f (t − τ )g(τ )dτ
−∞
Z −∞ Z ∞
= f (z)g(t − z)(−1)dz = g(t − z)f (z)dz = g ∗ f (t)
∞ −∞

Conclusion (2) follows from elementary properties of integrals, given that the integrals
involved converge. 

Theorem 6.12.

R∞
Suppose f and g are bounded and continuous on the real line, and that −∞
|f (t)|dt
R∞
and −∞ |g(t)|dt both converge. Then,

1. Z ∞ Z ∞ Z ∞
f ∗ g(t)dt = f (t)dt g(t)dt .
−∞ −∞ −∞

2. (Time Convolution)
f[
∗ g(ω) = fˆ(ω)ĝ(ω) .

3. (Frequency Convolution)

1 ˆ
f\
(t)g(t)(ω) = (f ∗ ĝ)(ω) . 
2π

Proof 6.12.
6.4. ADDITIONAL PROPERTIES AND APPLICATIONS 25

1. Write
Z ∞ Z ∞ Z ∞ 
f ∗ g(t)dt = f (t − τ )g(τ )dτ dt
−∞ −∞ −∞
Z ∞ Z ∞  Z ∞ Z ∞ 
= f (t − τ )g(τ )dt dτ = f (t − τ )dt g(τ )dτ
−∞ −∞ −∞ −∞

assuming the validity of this interchange of the order of the integration. Now,
Z ∞ Z ∞
f (t − τ )dt = f (t)dt
−∞ −∞

for any real number τ . Therefore

Z ∞ Z ∞ Z ∞ 
f ∗ g(t)dt = f (t)dt g(τ )dτ
−∞ −∞ −∞
Z ∞ Z ∞ Z ∞ Z ∞
= f (t)dt g(τ )dτ = f (t)dt g(t)dt .
−∞ −∞ −∞ −∞

2. Begin by letting F (t) = e−iωt f (t) and G(t) = e−iωt g(t) for real t and ω. Then
Z ∞
f[
∗ g(ω) = f ∗ g(t)eiωt dt
−∞
Z ∞ Z ∞ 
= f (t − τ )g(τ )dτ e−iωt dt
−∞ −∞
Z ∞ Z ∞ 
−iωt
= e f (t − τ )g(τ )dτ dt
−∞ −∞
Z ∞ Z ∞ 
−iω(t−τ ) iωτ
= e f (t − τ )e g(τ )dτ dt .
−∞ −∞

Now recognize that the integral in large parentheses in the last line is the con-
volution of F with G. Then, by (1) of this theorem applied to F and G,
Z ∞ Z ∞ Z ∞
f[
∗ g(ω) = F ∗ G(t)dt = F (t)dt G(t)dt
−∞ −∞ −∞
Z ∞ Z ∞
= f (t)e −iωt
dt g(t)e−iωt dt = fˆ(ω)ĝ(ω) .
−∞ −∞

3. We can have three versions for this proof. Firstly, let’s begin from the t-domain
26 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

as what follows.
Z ∞
f\
(t)g(t)(w) = f (t)g(t)e−jwtdt
−∞
Z ∞  Z ∞ 
1 ˆ ∗ jw ∗ t
= f (w )e dw g(t)e−jwtdt
∗
−∞ 2π −∞
Z ∞ Z ∞ 
1
= fˆ(w ∗ ) g(t)e −j(w−w ∗ )t
dt dw ∗
2π −∞ −∞
| {z }
ĝ(w−w ∗ )

Z ∞
1
= fˆ(w ∗ )ĝ(w − w ∗ )dw ∗
2π −∞

1 ˆ
= (f ∗ ĝ)(w) . (6.12)
2π
Secondly, we can also begin from the w-domain as what follows.
Z ∞
1 ˆ 1
f ∗ ĝ(w) = fˆ(w − w ∗ )ĝ(w ∗ )dw ∗
2π 2π −∞
Z ∞ Z ∞
1 ˆ ∗ ∗
= f (w − w ) g(t)e−iw t dt dw ∗
2π −∞
| −∞ {z }
ĝ(w ∗ )

Z ∞ Z −∞
1
fˆ(w − w ∗ )ei(w−w )t (−1)d(w − w ∗ ) g(t)e−iwt dt
∗
=
2π −∞ ∞ | {z }
=dw ∗
Z ∞ Z ∞
1
fˆ(w − w ∗)ei(w−w )t d(w − w ∗) g(t)e−iwt dt
∗
=
−∞ 2π −∞
| {z }
f (t)

Z ∞
= f (t)g(t)e−iwtdt = f\
(t)g(t)(ω) . (6.13)
−∞

1 ˆ
Thirdly, we can begin with from taking the inverse Fourier transform of 2π f∗
ĝ(w).
Z ∞ Z ∞ Z ∞ 
1 1 ˆ 1 1
jwt
f ∗ ĝ(w)e dw = fˆ(w − w )ĝ(w )dw ejwtdw
∗ ∗ ∗
2π −∞ 2π 2π −∞ 2π −∞
 
Z ∞
1
= fˆ(w − w ∗)ej(w−w )t d(w − w ∗ ) ×
∗

2π −∞ | {z }
dw
 Z ∞ 
1 ∗ jw ∗ t ∗
ĝ(w )e dw = f (t)g(t) . 
2π −∞
6.4. ADDITIONAL PROPERTIES AND APPLICATIONS 27

Example 6.15. Suppose we want to compute

 
−1 1
F .
(4 + ω 2 )(9 + ω 2 )
• The inverse version of conclusion (2) tell us that
 
1
F −1
(t) = F−1 [fˆ(ω)ĝ(ω)](t) = f ∗ g(t) .
(4 + ω 2 )(9 + ω 2 )

• Thus, we compute the individual transforms as what follows.

 
1 1 −2|t| 1
2
=F e = fˆ(ω) with f (t) = e−2|t| .
4+ω 4 4
and  
1 1 −3|t| 1
=F e = ĝ(ω) with g(t) = e−3|t| .
9 + ω2 6 6
• Then, we have
  Z ∞
−1 1 1 −2|t| 1 −3|t| 1
F (t) = e ∗ e = e−2|t−τ | e−3|τ | dτ .
(4 + ω 2 )(9 + ω 2 ) 4 6 24 −∞

• If t > 0, then
Z 0 Z t Z ∞
−2|t−τ | −3|τ | −2|t−τ | −3|τ |
24[f ∗ g(t)] = e e dτ + e e dτ + e−2|t−τ | e−3|τ | dτ
−∞ 0 t
Z 0 Z t Z ∞
−2(t−τ ) 3τ −2(t−τ ) −3τ
= e e dτ + e e dτ + e−2(t−τ ) e−3τ dτ
−∞ 0 t

6 −2t 4 −3t
= e − e .
5 5
• If t < 0, then
Z t Z 0 Z ∞
−2|t−τ | −3|τ | −2|t−τ | −3|τ |
24[f ∗ g(t)] = e e dτ + e e dτ + e−2|t−τ | e−3|τ | dτ
−∞ t 0
Z t Z 0 Z ∞
−2(t−τ ) −3τ 2(t−τ ) 3τ
= e e dτ + e e dτ + e2(t−τ ) e−3τ dτ
−∞ t 0

4 6
= − e3t + e2t .
5 5
• Therefore
   
−1 1 1 6 −2|t| 4 −3|t|
F (t) = e − e
(4 + ω 2 )(9 + ω 2) 24 5 5
1 −2|t| 1
= e − e−3|t| . 
20 30
28 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

1
FIGURE 6.11: y = 2a
[H(t + a) − H(t − a)].

6.4.5 Filtering and the Dirac Delta Function

A Dirac delta function is a pulse of infinite magnitude having infinitely short duration.
One way to describe such an object mathematically is to form a short pulse
1
[H(t + a) − H(t − a)] ,
2a
as shown in Figure 6.11. Take the limit as the width of the pulse approaches zero:
1
δ(t) = lim [H(t + a) − H(t − a)] .
a→0 2a

Fourier Transform of the Dirac Delta Function

First, if we take the Fourier transform of the pulse, we get

Z a a
−iωt 1 −iωt
F[H(t + a) − H(t − a)] = e dt = − e
−a iω −a

1 iaω sin (aω)

= (e − e−iaω ) = 2 .
iω ω
• By interchanging the limit and the operation of taking the transform, we have
 
1
F[δ(t)](ω) = F lim [H(t + a) − H(t − a)] (ω)
a→0 2a

1
= lim F[H(t + a) − H(t − a)](ω)
a→0 2a

sin (aω) ω cos(aω)

= lim = lim =1 .
|a→0 aω {za→0 ω }
by L’Hôpital’s Rule

• Further, putting δ(t) formally through the convolution, we have

F[δ ∗ f ] = F[δ]F[f ] = F[f ]

and
F[f ∗ δ] = F[f ]F[δ] = F[f ] ,
6.4. ADDITIONAL PROPERTIES AND APPLICATIONS 29

suggesting that
δ∗f =f ∗δ =f .

Theorem 6.13. Filtering

If f has a Fourier transform and is continuous at t0 , then

Z ∞
f (t)δ(t − t0 )dt = f (t0 ) . 
−∞

This result can be modified to allow for a jump discontinuity of f at t0 . In

this event, we get
Z ∞
1

f (t)δ(t − t0 )dt = f (t+ −
0 ) + f (t0 ) .
−∞ 2
Also, this can be proved as what follows.
Z ∞ Z ∞
1
f (t)δ(t − t0 )dt = lim f (t) [H(t − t0 + a) − H(t − t0 − a)] dt
−∞ a→0 −∞ 2a
Z t0 +a Z t0 +a Z t−
1 1 1 0
= lim f (t)dt = lim f (t)dt + lim f (t)dt
a→0 2a t0 −a a→0 2a t+ a→0 2a t0 −a
0

1  1 −
t0 +a t
= lim F (t) t+ + lim F (t) t00 −a
a→0 2a 0 a→0 2a
 
1 F (t0 + a) − F (t+
0) F (t−
0 ) − F (t0 − a)
= lim + lim
2 a→0 a a→0 a
1 + 
= f (t0 ) + f (t−
0) .
2

6.4.6 The Windowed Fourier Transform

Suppose f is a signal. This means that f is a function that is defined over the real
R∞
line, and has finite energy −∞ |f (t)|2 dt.

• In analyzing f (t), we sometimes want to localize its frequency content with

respect to the time variable. We have mentioned that fˆ(ω) carries informa-
tion about the frequency of the signal. However, fˆ(ω) does not particularize
information to specific time intervals, since
Z ∞
fˆ(ω) = f (t)e−iωt dt
−∞

and this integration is over all time.

30 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

FIGURE 6.12: Typical window function with compact support [0, T ].

• Hence the picture we obtain does not contain information about specific times,
but instead enables us only to compute the total amplitude spectrum |fˆ(ω)|.
If we think of f (t) as a piece of music being played over time, we would have
to wait until the entire piece was done before even computing this amplitude
spectrum. However, we can obtain a picture of the frequency content of f (t)
within given time intervals by windowing the function before taking its Fourier
transform.

• To do this, we first need a window function g, which is a function taking on

nonzero values only on some closed interval, often [0, T ] or −T, T . Figures 6.12
and 6.13 show typical graphs of such functions, one on [0, T ] and the other on
[−T, T ]. The interval is called the support of g, and in this case that we are
dealing with closed intervals, we say that g has compact support. The function
g has zero values outside of this support interval.

• We window a function f with g by forming the product g(t)f (t), which vanishes
outside of [−T, T ]. The windowed Fourier transform (with respect to the choice
of g) is
Z ∞
Fwin [f ](ω) = fd
win (ω) = f (t)g(t)e−iωt dt
−∞
Z T
= f (t)g(t)e−iωt dt
−T

Example 6.16. Consider the window function

(
1 for − 4 ≤ t ≤ 4
g(t) =
0 for |t| > 4
having compact support [−4, 4].

• This function is graphed in Figure 6.14a, with the vertical segments at t = ±4

included to emphasize this interval.
6.4. ADDITIONAL PROPERTIES AND APPLICATIONS 31

FIGURE 6.13: Typical window function with compact support [−T, T ].

(a) (b) (c)


1 for |t| ≤ 4
FIGURE 6.14: (a) Window function g(t) = (b) f (t) = t sin (t) (c)
0 for |t| > 4
f windowed with g.

• Let f (t) = t sin(t), shown in Figure 6.14b.

• To window f with g, form the product g(t)f (t), shown in Figure 6.14c.

• This windowed function vanishes outside the support of g. For this choice of g,
windowing has the effect of turning the signal f (t) on at time −4 and turning
it off at t = 4. 

Example 6.17. Let f (t) = 6e−|t| . Then

Z ∞
12
fˆ(ω) = 6e−|t| e−iωt dt = .
−∞ 1 + ω2

• Use the window function

(
1 for − 2 ≤ t ≤ 2
g(t) = .
0 for |t| > 2

Figure 6.15 shows a graph of the windowed function g(t)f (t).

32 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

• The windowed Fourier transform of f is

Z ∞
d
fwin (ω) = 6e−|t| g(t)e−iωt dt
−∞
Z 2
= 6e−|t| e−iωt dt
−2

−2e−2 cos2 (ω) + e−2 + e−2 ω sin (2ω) + 1

= 12 .
1 + ω2
This gives the frequency content of the signal f in the time interval −2 ≤ t ≤
2. 

Shifted Window Function

Often we use a shifted window function. Suppose the support of g is [−T, T ].

• If t0 > 0, then the graph of g(t − t0 ) is the graph of g(t) shifted t0 units to the
right. Now
(
f (t)g(t − t0 ) for t0 − T ≤ t ≤ t0 + T
f (t)g(t − t0 ) = .
0 for t < t0 − T and for t > t0 + T

Figure 6.16a through 6.16d illustrate this process.

• In this case we take the Fourier transform of the shifted windowed signal to be

f\
win,t0 (ω) = F[f (t)g(t − t0 )](ω)

Z t0 +T
= f (t)g(t − t0 )e−iωt dt .
t0 −T

This gives the frequency content of the signal in the time interval t0 − T, t0 + T .


−|t| 1 for |t| ≤ 2
FIGURE 6.15: f (t) = 6e windowed with g(t) = .
0 for |t| > 2
6.4. ADDITIONAL PROPERTIES AND APPLICATIONS 33

(a) (b)

(c) (d)

FIGURE 6.16: (a) A window function g on [−T, T ] (b) Shifted window function
g(t − t0 ) (c) Typical signal f (t) (d) g(t − t0 )f (t).

Center and Radius of the Window Function

• If g is the window function, the center of g is defined to be the point

R∞
t|g(t)|2dt
tC = R−∞
∞ .
−∞
|g(t)|2dt
The number R∞ !1/2
−∞
(t − tC )2 |g(t)|2dt
tR = R∞
−∞
|g(t)|2dt
is the radius of the window function. The width of the window function is 2tR ,
and is referred to as the RMS duration of the window.

• When we deal with the Fourier transform of the window function, then similar
terminology applies:
R∞
ω|ĝ(ω)|2 dω
center of ĝ = ωC = R−∞
∞
−∞
|ĝ(ω)|2 dω
and R∞ !1/2
−∞
(ω − ωC )2 |ĝ(ω)|2dω
radius of ĝ = ωR = R∞ .
−∞
|ĝ(ω)|2dω
The width of ĝ is 2ωR , a number referred to as the RMS bandwidth of the
window function.
34 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

6.4.7 The Shannon Sampling Theorem

We will derive the Shannon sampling theorem, which states that a band-limited signal
can be reconstructed from certain sampled values.

• A signal f is band-limited if its Fourier transform fˆ has compact support (has

nonzero values only on a closed interval of finite length). This means that, for
some L,
fˆ(ω) = 0 if |ω| > L .
In this event L is the bandwidth of the signal. The total frequency content of
such a signal f lies in the band [−L, L].

• Assume that we can recover f (t) for all real t from its transform:
Z ∞
1
f (t) = fˆ(ω)eiωt dω
2π −∞
Z L
1
= fˆ(ω)eiωt dω . (6.14)
2π −L
| {z }
Because f is band-limited

• First recall the definition of “Complex Fourier series” as follows. Let f have
fundamental period p. Let ω0 = 2π/p. Then the complex Fourier Series of f is
∞
X
dn einω0 x
n=−∞

where
Z p/2
1
dn = f (t)e−inω0 t dt
p −p/2

for n = 0, ±1, ±2, .... The numbers dn are the complex Fourier coefficients of f .
Then, put this aside for the moment and write the complex Fourier series for
fˆ(ω) on [−L, L]:
∞
X
ˆ
f (ω) = cn enπiω/L (6.15)
n=−∞

where Z L
1
cn = fˆ(ω)e−nπiω/L dω .
2L −L

Comparing cn with f (t) in equation (6.14), we conclude that

 
π −nπ
cn = f .
L L
6.4. ADDITIONAL PROPERTIES AND APPLICATIONS 35

• Substitute this into equation (6.15) to get

X∞  
π −nπ
fˆ(ω) = f enπiω/L .
n=−∞
L L

Since n takes on all integer values in this summation, we can replace n with −n
to write
π X  nπ  −nπiω/L
∞
fˆ(ω) = f e .
L n=−∞ L

• Now substitute this series for fˆ(ω) into equation (15.8) to get
Z π ∞
X  nπ 
1 π
f (t) = f e−nπiω/L eiωt dω
2π L −π n=−∞ L

Interchange the sum and the integral to get

Z
1 X  nπ  L iω(t−nπ/L)
∞
f (t) = f e dω
2L n=−∞ L −L

∞
1 X 1
= f (nπ/L) [eiω(t−nπ/L) ]L−L
2L n=−∞ i(t − nπ/L)

∞
1 X 1
= f (nπ/L) (ei(Lt−nπ) − e−i(Lt−nπ) )
2L n=−∞ i(t − nπ/L)

∞
X 1 1 i(Lt−nπ)
= f (nπ/L) (e − e−i(Lt−nπ) )
n=−∞
Lt − nπ 2i

∞
X sin (Lt − nπ)
= f (nπ/L) . (6.16)
| {z } Lt − nπ
n=−∞
sample at t = nπ/L

Note that f (nπ/L) is the sample at t = nπ/L. Once the values of f (t) are
known for these times, then equation (6.16) reconstructs the entire signal. This
is actually the way engineers convert digital signals to analog signals, with
application to technology such as that involved in making compact disk.

Shannon Sampling Theorem

Equation (6.16) is known as the Shannon sampling theorem.

• We will encounter it again when we discuss wavelets. In the case L = π, the

36 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

0.8

0.6

Asinc(t)
0.4

0.2

−0.2

−0.4
−5 −4 −3 −2 −1 0 1 2 3 4 5
t

(a) (b)

FIGURE 6.17: (a) A general rectangular pulse; (b) A general sinc fucntion sinc(t).

sampling theorem has the simple form

∞
X sin (π(t − n))
f (t) = f (n)
n=−∞
π(t − n)

∞
X
= f (n)sinc(t − n) (6.17)
n=−∞

and f (n) represents the n-th sample of the original signal; Also sinc(t) is called
the Shannon sampling function.

• One should note that the rectangular pulse AΠ(t/T ) can be Fourier transformed
to be a sinc function AT sinc(f T ). Examples of the general rectangular AΠ(t/T )
and sinc function sinc(t) can be found in Figure 6.17.

6.4.8 Lowpass and Bandpass Filters

Consider a signal f , not necessarily band-limited. However, we assume that the signal
has finite energy, so Z ∞
|f (t)|2dt
−∞

is finite. Such functions are called square integrable.

Lowpass Filters

• If f is not band-limited, we can replace f with a band-limited signal fω0 for the
range [−ω0 , ω0 ] as
(
fˆ(ω) for − ω0 ≤ ω ≤ ω0
fc
ω0 (ω) = .
0 for |ω| > ω0
6.4. ADDITIONAL PROPERTIES AND APPLICATIONS 37

FIGURE 6.18: Graph of χ[−ω0 ,ω0 ] .

This defines the transform of the function fω0 , from which we recover fω0 by
the inverse Fourier transform:
Z ∞ Z ω0
1 c 1
fω0 (t) = iωt
fω (ω)e dω = fc iωt
ω (ω)e dω .
2π −∞ 0 2π −ω0 0

• The process of applying the lowpass filter is carried out mathematically by

multiplying by an appropriate function (essentially windowing). Define the
characteristic function χI of an interval I by
(
1 if t is in I
χI (t) = .
0 if t is a real number that is not in I

Now observe that

fc ˆ
ω0 (ω) = χ[−ω0 ,ω0 ] (ω)f (ω) . (6.18)

or, more succintly,

fc ˆ
ω0 = χ[−ω0 ,ω0 ] f .

In this context, χ[−ω0 ,ω0 ] is called the transfer function. Its graph is shown in
Figure 6.18.

• The inverse Fourier transform of the transfer function is

Z ω0
−1 1 sin (ω0 t)
F [χ[−ω0 ,ω0 ] ](t) = eiωt dω =
2π −ω0 πt

whose graph is given in Figure 6.19. In the case that ω0 = π, this is the function,
evaluated at t − n instead of t, which occurs in the Shannon sampling formula
(6.17) that reconstructs f (t) from sampled values f (n) on the integers. For this
reason sin(ω0 t)/πt is called the Shannon sampling function.

• If ϕ(t) is the filter function, then the effect of filtering a function f by ϕ is a

new function g defined by
Z ∞
g(t) = (ϕ ∗ f )(t) = ϕ(ξ)f (t − ξ)dξ .
−∞
38 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

FIGURE 6.19: Graph of sin (ω0 t)/πt for ω0 = 2.7.

Taking the Fourier transform of this equation, we have

ĝ(ω) = ϕ̂(ω)fˆ(ω) .

We can now formulate equation (15.12) as

 
sin (ω0 t)
fω0 (t) = ∗ f (t) .
πt
This gives the lowpass filtering of f as the convoluton of the Shannon sampling
function with f .

Bandpass Filters

• In the kind of bandpass filtering, we want to filter out the effects of the signal
outside of given bandwidths. A band-limited signal f can be decomposed into
a sum of signals, each of which carries the information content of f within a
certain given frequency band. Let f be a band-limited signal of bandwidth Ω.
Consider a finite increasing sequence of frequencies,

0 < ω1 < ω2 < ... < ωn = Ω .

For j = 1, ..., N, define a bandwidth filter function βj by means of its transfer

function:
βˆj = χ[−ωj ,−ωj−1 ] + χ[ωj−1 ,ωj ]
This transfer function, which is a sum of characteristic functions of frequency
intervals, is graphed in Figure 6.20.

• The bandwidth filter function βj (t), which filters the frequency content of f (t)
outside of the frequency range [ωj−1 , ωj ], is obtained by taking the inverse
Fourier transform of βˆj (ω). We get
sin (ωj t) − sin (ωj−1 t)
βj (t) = ,
πt
6.4. ADDITIONAL PROPERTIES AND APPLICATIONS 39

FIGURE 6.20: χ[−ωj ,−ωj−1 ] + χ[ωj−1 ,ωj ] .

sin (ωj t)−sin (ωj−1 t)

FIGURE 6.21: βj (t) = πt
with ωj = 2.2 and ωj−1 = 1.7.

whose graph is shown in Figure 6.21.

• Now define functions

 
sin (ω0 t)
f0 (t) = ∗f (t)
πt

and, for j = 1, 2, ..., N,

fj (t) = (βj ∗ f )(t) .

Decomposition of Signal into Specific Frequency Components

Then, for j = 1, 2, ..., N, each fj (t) carries the content of the signal f (t) in the
frequency range ωj−1 ≤ ω ≤ ωj , while f0 (t) carries the content in [0, ω0 ], which is the
low-frequency range of f (t). Further,

f (t) = f0 (t) + f1 (t) + f2 (t) + ... + fN (t) (6.19)

giving a decomposition of the signal into components carrying the information of the
signal for specific frequency intervals.
40 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

6.5 The Fourier Cosine and Sine Transform

Fourier Cosine Transform
R∞
Suppose f (t) is piecewise smooth on each interval [0, L] and 0
|f (t)|dt converges.
Then for each t at which f is continuous,
Z ∞
f (t) = aω cos (ωt)dω ,
0

where Z
2 ∞
aω = f (t) cos (ωt)dt .
π 0
Based on these two equations, we make the following.

Definition 6.5. Fourier Cosine Transform

The Fourier cosine transform of f is defined by

Z ∞
Fc [f ](ω) = f (t) cos (ωt)dt .  (6.20)
0

Often we will denote Fc [f ](ω) = fˆC (ω).

Notice that
π
fˆC (ω) = aω
2
and that Z
2 ∞ ˆ
f (t) = fC (ω) cos (ωt)dω . (6.21)
π 0
The integrals in expressions (6.20) and (6.21) form the transform pair for the Fourier
cosine transform, which complies with the linearity.

FC [αf + βg] = αFC [f ] + βFC [g] .

Example 6.18. Let K be a positive number, and let

(
1 for 0 ≤ t ≤ K
f (t) = .
0 for t > K
The Fourier transform of f is
Z ∞
fˆC (ω) = f (t) cos (ωt)dt
0
Z K
sin (Kω)
= cos (ωt)dt = .
0 ω
6.5. THE FOURIER COSINE AND SINE TRANSFORM 41

Fourier Sine Transform

Similar to the Fourier cosine transform, we can also define the Fourier sine transform
as following.
Definition 6.6. Fourier Sine Transform

The Fourier sine transform of f is defined by

Z ∞
Fs [f ](ω) = f (t) sin (ωt)dt . 
0

We also denote this as fˆS (ω). If f is continuous at t > 0, then the Fourier integral
sine representation is Z ∞
f (t) = bω sin (ωt)dω ,
0
where Z ∞
2
bω = f (t) sin (ωt)dt .
π 0
Since
π
fˆS (ω) = bω
2
then Z
2 ∞ ˆ
f (t) = fS (ω) sin (ωt)dω .
π 0
The linearity of the transform says

FS [αf + βg] = αFS [f ] + βFS [g] .

Example 6.19. With f the function of Example 6.18,

Z ∞
ˆ
fS (ω) = f (t) sin (ωt)dt
0
Z K
1
= sin (ωt)dt = [1 − cos (Kω)] . 
0 ω

When these transforms are used to solve differential equations, the following
operational rules play a key role.
Theorem 6.14. Operational Rules

R∞
Let f and f ′ be continuous on every [0, L], and let 0 |f (t)|dt converge. Suppose
f (t) → 0 and f ′ (t) → 0 as t → ∞. Suppose f ′′ is piecewise continuous on every
interval [0, L]. Then,
42 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

1.

FC [f ′′ (t)](ω) = −ω 2 fˆC (ω) − f ′ (0) ;

2.

FS [f ′′ (t)](ω) = −ω 2 fˆS (ω) + ωf (0) . 

The operational formula dictates which transform is used to solve a given problem.

• If we seek a function f (t), for 0 ≤ t < ∞, and f (0) is specified, then we might
consider a Fourier sine transform.

• If, however, information is given about f ′ (0), then the cosine transform might
be appropriate.

Proof 6.14. Given that limt→∞ f (t) = 0 and limt→∞ f ′ (t) = 0. In the following
proofs, we should use integration by parts for two times.

• For the first conclusion of this theorem.

Z ∞
′′
FC [f (t)](w) = f ′′ (t) cos(wt)dt
|0 {z }
Let u = cos(wt) and dv = f ′′ (t)dt

Z ∞
′
= f (t) cos(wt) |∞
0 +w f ′ (t) sin(wt)dt
|0 {z }
Let u = sin(wt) and dv = f ′ (t)dt

Z ∞
′
= f (t) cos(wt) |∞
0 + wf (t) sin(wt) |∞
0 −w 2
f (t) cos(wt)dt
0

= 0 − f ′ (0) + 0 − 0 − w 2 fˆC (w)

= −w 2 fˆC (w) − f ′ (0) .

6.6. THE FINITE FOURIER COSINE AND SINE TRANSFORMS 43

• For the second conclusion of this theorem.

Z ∞
′′
FS [f (t)](w) = f ′′ (t) sin(wt)dt
|0 {z }
Let u = sin(wt) and dv = f ′′ (t)dt

Z ∞
′
= f (t) sin(wt) |∞
0 −w f ′ (t) cos(wt)dt
0
| {z }
Let u = cos(wt) and dv = f ′ (t)dt

Z ∞
′
= f (t) sin(wt) |∞
0 − wf (t) cos(wt) |∞
0 −w 2
f (t) sin(wt)dt
0

= 0 − 0 − 0 + wf (0) − w 2 fˆS (w)

= −w 2 fˆS (w) − wf (0) . 

6.6 The Finite Fourier Cosine and Sine

Transforms
The Fourier transform, cosine transform and sine transform are all motivated by the
respective integral representations of a function. If we employ essentially the same
line of reasoning, but using Fourier cosine and sine series instead of integrals, we
obtain what are called finite transforms.
Suppose f is piecewise smooth on [0, π]. Then here comes the finite Fourier
Cosine and Sine transforms.

Finite Fourier Cosine Transforms

Definition 6.7. Finite Fourier Cosine Transform

The finite Fourier cosine transform of f is defined by

Z π
C[f ](n) = f˜C (n) = f (x) cos (nx)dx
0

for n = 0, 1, 2, ... . 

• If f is continuous at x in [0, π], then f (x) has the Fourier cosine series repre-
sentation
X∞
1
f (x) = a0 + an cos (nx) ,
2 n=1
44 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

where Z π
2 2˜
an = f (x) cos (nx)dx = fC (n) .
π 0 π
Then ∞
1˜ 2X˜
f (x) = fC (0) + fC (n) cos (nx)
π π n=1
an inversion-type of expression from which we can recover f (x) from the finite
Fourier cosine transform of f .

• Note that one may curiously compare the finite cosine Fourier transform with
the Fourier cosine series, which is written as
X∞
1 nπt
f (t) = a0 + an cos( );
2 n=1
L
Z ∞
2 nπt
an = f (t) cos( )dt ,
L 0 L
where an is called the coefficient of the Fourier cosine series.

Finite Fourier Sine Transforms

By the same token, we can define a finite sine transform.

Definition 6.8. Finite Fourier Sine Transform

The finite Fourier sine transform of f is defined by

Z π
S[f ](n) = f˜S (n) = f (x) sin (nx)dx
0
for n = 1, 2, ... . 

For 0 < x < π, if f is continuous at x, then the sine series representation is

∞
2X˜
f (x) = fS (n) sin (nx) ,
π n=1

an inversion formula for the finite sine transform.

Example 6.20. Let f (x) = x2 for 0 ≥ x ≥ π. For the finite cosine transform,
compute Z π
1
f˜C (0) = x2 dx = π 3
0 3
and, for n = 1, 2, ...,
Z π
(−1)n
f˜C (n) = x2 cos (nx)dx = 2π .
0 n2
6.6. THE FINITE FOURIER COSINE AND SINE TRANSFORMS 45

For the finite sine transform, compute

Z π
(−1)n [2 − n2 π 2 ] − 2
f˜S (n) = x2 sin (nx)dx = .
0 n3

One should note the above calculation can be conducted by using the partial integra-
tion techniques for two times. For example, in calculating f˜C (n), we can firstly let
u = x2 and dv = cos(nx)dx; And then let u = x and dv = sin(nx)dx . .

Here are the fundamental operational rules for these transforms.

Theorem 6.15. Operational Rules

Let f and f ′ be continuous on [0, π], and let f ′′ be piecewise continuous. Then

1.
C[f ′′ ](n) = −n2 f˜C (n) − f ′ (0) + (−1)n f ′ (π)

for n = 1, 2, ..., and

2.
S[f ′′ ](n) = −n2 f˜S (n) + nf (0) − n(−1)n f (π)

for n = 1, 2, ... . 

We will see applications of these finite transforms when we discuss partial

differential equations.

Proof 6.15.

• For the first conclusion:

Z π
˜′′
fC (n) = f ′′ (x) cos(nx)dx
|0 {z }
Let u = cos(nx) and dv = f ′′ (x)dx

Z π
′
= f (x) cos(nx)|π0 +n f ′ (x) sin(nx)dx
|0 {z }
(Let u = sin(nx) and dv = f ′ (x)dx)

Z π
′
= f (x) cos(nx)|π0 + nf (x) sin(nx)|π0 −n 2
f (x) cos(nx)dx
0

= (−1)n f ′ (π) − f ′ (0) − n2 f˜C (n) .

46 CHAPTER 6. THE FOURIER INTEGRAL AND FOURIER TRANSFORMS

• For the second conclusion:

Z π
˜′′
fS (n) = f ′′ (x) sin(nx)dx
|0 {z }
(Let u = sin(nx) and dv = f ′′ (x)dx)

Z π
′
= f (x) sin(nx)|π0 −n f ′ (x) cos(nx)dx
|0 {z }
(Let u = cos(nx) and dv = f ′ (x)dx)

Z π
′
= f (x) sin(nx)|π0 − nf (x) cos(nx)|π0 −n 2
f (x) sin(nx)dx
0

= −n(−1)n f ′ (π) + nf (0) − n2 f˜S (n) . 

 47

Bibliography

[1] P. V. O’Neil, Advanced Engineering Mathematics, international student edition ed.

Thomson, 2007.

[2] ——, Advanced Engineering Mathematics, 7th ed. Thomson, 2012.

[3] ——, Advanced Engineering Mathematics, international student edition ed.

Thomson, 2012.

[4] E. Kreyszig, Advanced Engineering Mathematics, 10th ed. John Wiley & Sons,
INC., 2011.

[5] D. G. Zill and W. S. Wright, Advanced Engineering Mathematics, 5th ed. Jones
and Bartlett Pulisher, 2014.

[6] M. D. Greenberg, Advanced Engineering Mathematics, 5th ed. Prentice Hall

International, INC., 1998.

[7] J. D. Faires and B. T. Faires, Calculus, 2nd ed. MCGraw-Hill, INC., 1993.