THE FOURIER TRANSFORM

INTRODUCTION and Definitions

The Fourier Transform, especially after treatments of the theory of the Laplace transform
and the theory of Fourier series and integrals, may be introduced from two perspectives.
On one hand, we may introduce it from the perspective of a special case of the Laplace
transform for a function f ( x ) that is defined on the entire real axis, and in which the
parameter s is pure imaginary ( s = jt ), with the angular frequency  permitted to
assume all values on the entire interval −    . On the other hand, we may choose to
introduce it from the perspective of a complex form of the Fourier integral of a non-
periodic function of a function f ( x ) , again defined on the entire real axis. Our treatment
here will use the latter approach, as it appears to relate, physically, much more closely to
other electrical and mechanical engineering concepts, to establish the basic definition.
When we come to investigations of properties of the transform, then we shall frequently
have occasions to refer to the corresponding or analogous properties of the Laplace
transform. Further, for reasons of closeness to electrical or mechanical variables, it will be
more convenient to adopt "t " , which will frequently denote “time”, in place of the usual
" x " for the independent variable in our writings.

To begin then, we recall from our discussions of the extension of the Fourier series to the
representation of non-periodic functions that a function f (t ) , defined on the entire real axis,
−  t  , may be represented by means of the Fourier integral

f (t ) =   A( ) cos t + B( )sin t  d , −     , (2.1.1)
0

where

1
A( ) =
 
−
f ( ) cos  d , −     , (2.1.2)

and

1 EM
 
1
B ( ) =
 
−
f ( ) sin  d , −     , (2.1.3)

provided,
1. f (t ) and f (t ) are piecewise continuous in every interval, and

2. 
−
f (t ) dt exists.

Note that in the last two statements (2.1.2) and (2.1.3, we have used " " simply as a dummy
variable of integration so as not to get confused with "t " used to specify the domain of
definition of the non-periodic function f (t ) of interest. We shall reinstate "t " once we have
established the desired definition and no further confusion is apparent.

Then by introducing statements (2.1.2) and (2.1.3) in (2.1.1), we may clearly rewrite the
Fourier integral as


 1   1   
f (t ) =    f ( ) cos  d  cos t +   f ( )sin  d  sin t  d .
  −
0     −  

On combining the two integrals it becomes


1  
f (t ) =    f ( ) ( cos  cos t + sin  sin t ) d  d .
0
 − 

If now we exploit the well-known trigonometric relationship

cos  cos  + sin  sin  = cos( −  ),

we may of course then recast the Fourier integral alternatively as


1  
f (t ) =    f ( ) cos  (t −  )d  d .
0 
 − 
j − j
e +e
By virtue of the Euler’s identity cos  = , we may quite obviously re-write the
2
Fourier integral further as

1   1  
   

     f ( )e
j ( t − ) − j ( t − )
f (t ) = f ( ) e d  d  + d  d . (2.1.4)
2 0  −  2 0  − 

By replacing  with ( − ) in the second term and thereby changing the direction of
integration with respect to  , we may recast the integral in this second term as

2 EM
 
  
 
 0
 
0  −      f ( )e
− j ( t − ) j ( t − ) j ( t − )
f ( ) e d  d  = − f ( ) e d  d  = d  d .
 0  −  −  − 

Upon using this result in (2.1.4) and combining the resulting two terms, we find that

1

 
−  − f ( )e d  d .
j ( t − )
f (t ) = (2.1.5)
2

On splitting the exponential quantity e j ( t − ) in the integrand as product, we may finally re-
write the Fourier integral in the equivalent complex form as


 1 

f (t ) = −  2 
−
f ( )e − j ) d  e jt d .

(2.1.6)

It is evident that, once performed and the limits have been substituted, the inner integral

1
2 − f ( )e − j ) d yields a function of  . If we denote this function by F ( ) , then the

complex form of the Fourier integral clearly takes the shape



 F ( )e
jt
f (t ) = d , (2.1.7)
−

with

1
F ( ) =
2 
−
f ( )e − j d . (2.1.8)

Reinstating "t " in place of " " , we may, without further confusion, re-write the latter
expression as

1
F ( ) =
2 
−
f (t )e − jt dt. (2.1.9)

Equations (2.1.7) and (2.1.9) constitute what is commonly known in the literature as the
Fourier Transform Pair. Equation (2.1.9) is known as the Fourier transform F ( ) of the
function f (t ) and for emphasis is thus sometimes symbolised by F  f (t ) . Similarly, f (t )
in equation (2.1.7) is called the Inverse Fourier Transform of F ( ) and is thus again for
reasons of emphasis sometimes denoted by F −1 F ( ) .

A word of caution is necessary at this point though. In the literature there is actually no
universal agreement over the convention that should be used to symbolise what is meant
by the Fourier Transform and what is meant by the Inverse Fourier Transform. Thus, some

3 EM
 1
authors prefer to write the coefficient in equation (2.1.7), rather than in (2.1.9), and
2
thereby have

1
 F ( )e
jt
f (t ) = d (2.1.10)
2 −

as the Inverse Fourier transform and


F ( ) = 
−
f (t )e − jt dt. (2.1.11)

as the Fourier transform. Others, on the other hand, partition the coefficient equally
between the two expressions and thereby have

1
 F ( )e
jt
f (t ) = d (2.1.12)
2 −

as the Inverse Fourier transform and


1
F ( ) =
2

−
f (t )e − jt dt. (2.1.13)

Further still, there is also variation in the location of the factor e jt and e − jt , in such a way
that the former features in the Fourier Transform, while the latter features in the Inverse
Transform. It is thus important to be aware of these variations when referring to other
sources. Of these we shall henceforth use versions (2.1.10) and (2.1.11) throughout, so that
we may easily relate some of the results here to those already obtained during our
consideration of the Laplace transform.

Turning to expression (2.1.11), we note the similarity between it and the definition of the
Laplace transform of f (t ) :

L  f (t ) = F ( s) =  f (t )e− st dt. (2.1.14)
0

We see that, apart from the limits of integration, the substitution j = s in equation
(2.1.11) yields the Laplace transform. There indeed exists an important relationship
between the two transforms, which will be discussed further subsequently. We should also
note here that although formula (2.1.10) provides a means of recovering a function f (t )
from its transform, evaluation of the integral is frequently very difficult to evaluate. Other
approaches may therefore have to be resorted to.

To conclude this section, let us consider the following examples.

Example 2.1.1: Find the Fourier Transform of the function f (t ) = u (t )e −t , where

u (t ) is the unit step function.

4 EM
Solution: The function f (t ) = u (t )e−t is depicted in Figure 2.1.1 below.

Figure 2.1.1: Graph of

Using equation (2.1.11), its Fourier transform is given by

 

 dt =  e −t e − jt dt
− jt
F ( ) = f (t )e (since f (t ) = 0 for t  0)
− 0

 
 e − (1+ j )t  1
= e − (1+ j ) t
dt =   = (since e− (1+ j )t → 0 as t → ).
0  −(1 + j )  0 1 + j
Hence
F u (t )e−t  = F ( ) =
1
. (2.1.15)
1 + j

Example 2.1.2: Find the Fourier integral representation of the function defined by
 1, − 1  t  1,
f (t ) =  (2.1.16)
0, t  1.

Solution: Using equation (2.1.10), we have


1
f (t ) = 
2 −
F ( )e jt d  (2.1.17)

where
 1
F ( ) =  f (t )e − jt
dt =  1  e − jt dt , (since f (t ) is zero outside  −1,1)
− −1
1
 e− jt  e j − e − j 2 j sin 
=  = = (by virtue of Euler's identity) (2.1.18)
 − j  −1 j j
2sin 
= .

Putting this result back in (2.1.17), we then have

5 EM
 
1 2sin 
f (t ) =
2 
−

e jt d  (2.1.19)

2sin 
is the required integral representation. We note that F ( ) = is the Fourier

sin 
transform of f (t ) , defined in equation (2.1.16). Because the function arises very

frequently in applications, it has been given the special name of sinc function.

Just as with Laplace transforms, tables of Fourier transforms have been compiled for
reference. Table 2.1.1 contains some of the common transforms.

Table 2.1.1 Common Fourier Transforms

f (t ) F ( )

f (t ) = Au (t )e− t , a  0 A
F ( ) =
 + j
 1, −   t   , 2sin 
f (t ) =  F ( ) =
0, otherwise 
f (t ) = A (a constant) F ( ) = 2 ( )

f (t ) = u (t ) A (a constant)  1
F ( ) = A   ( ) − 
 
f (t ) =  (t ) F ( ) = 1

f (t ) =  (t −  ) F ( ) = e j

f (t ) = cos at F ( ) =  ( ( + a) +  ( − a) )

f (t ) = sin at 
F ( ) = ( ( + a) −  ( − a) )
j
f (t ) = e
− t
,  0 2
F ( ) =
 + 2
2

 1, t  0 F ( ) =
2
f (t ) = sgn(t ) = 
−1, t  0 j
1 F ( ) = − j sgn( )
f (t ) =
t
f (t ) = e− at 
2

F ( ) e −
2
/ 4a

6 EM
2.2 Some Properties of the Fourier Transform
As indicated in the preceding section, a number of the properties of Laplace transforms that
have been discussed previously hold for Fourier transforms. Here we shall revisit two of the
related theorems, namely, the Linearity and two Shift Theorems.

2.2.1 Linearity

Theorem 2.2.1: If f and g are functions of t and  is a constant, then

F  f + g = F ( f ) + F  g , (2.2.1)
and
F  f  =  F  f  . (2.2.2)

Both of these properties follow directly from the definition and linearity properties of
integrals, as already seen with regard to Laplace transforms. They emphasise that the
Fourier operator F is linear just like the Laplace operator L. To illustrate, let us consider
an example.

Example 2.2.2: Find the Fourier transform of

f (t ) = u (t )e−t + u (t )e −2t . (2.2.3)

Solution: In Example 2.1.1, we saw that

F u (t )e−t  =
1
. (2.2.4)
1 + j
Furthermore,
 
F u (t )e −2t  =
1
 u (t )e e = e
−2 t − jt − (2 + jt )
= . (2.2.5)
− −
2 + j
Therefore,
F u (t )e−t + u (t )e−2t  =
1 1
+ (by liinearity)
1 + j 2 + j
(2.2.6)
2 + j + 1 + j 2 j
= = .
(1 + j )( 2 + j ) 2 −  2 + 3 j

2.2.2 The First Shift Theorem

Theorem 2.2.2: If F ( ) is the Fourier Transform of f (t ) , then

F e j t f (t ) = F ( −  ) (2.2.7)
where  is a constant.
Let demonstrate the application in the following example.

7 EM
Example 2.2.3: a) Find the Fourier Transform of
 3, − 2  t  2,
f (t ) = 
0, otherwise.

b) Use the First Shift Theorem to find the Fourier transform of e− jt f (t ) .

c) Verify the result in b) by computing the Fourier transform of e− jt f (t )
directly.

Solution:
a) From definition we have
 2
2
 e− jt   e−2 j − e2 j   e2 j − e−2 j 
F ( ) =  dt = 3  e
− jt − jt
f (t )e dt = 3   = 3  = 6 
− −2  − j  −2  − j   2 j 
sin 2
=6 .


sin 2
b) From part a), we already have F ( ) = 6 . Upon invoking the First Shift

Theorem, we then find that
sin 2( + 1)
F e− jt f (t ) = F ( + 1) = 6 .
 +1

c) By direct computation,
 2
2
 e− j (1+ ) t 
F e − jt
f (t ) = e
− jt
f (t )e − jt
dt = 3  e − j (1+ ) t
dt = 3  
− −2  − j (1 +  )  −2
−2 j (1+ ) 2 j (1+ )
e −e   e2 j (1+ ) − e −2 j (1+ )  sin 2( + 1)
= 3  = 6  =6 ,
 − j (1 +  )   2 j (1 +  )  (1 +  )
as in part b).

Example 2.2.4:
Use the First Shift Theorem to find the function whose Fourier transform is
1
F ( ) = ,
3 + j ( − 2)
given that F u (t )e − mt  =
1
, m  0.
m + j
Solution:
From the given hint, we have F u (t )e−3t  =
1
. Therefore, by the Shift Theorem
3 + j
with  = 2, we get

8 EM
 F e2 jt u (t )e−3t  =
1
.
3 + j ( − 2)
1
Consequently, the function whose Fourier transform is F ( ) = is thus
3 + j ( − 2)
f (t ) = u (t )e − (3−2 j ) t .

Example 2.2.5:
a) Find the Fourier transform of
 e−3t , t  0,
f (t ) =  3t
e , t  0.
6
b) Deduce the function whose Fourier transform is G ( ) = .
10 + 2 +  2

Solution: From definition, we have

 0 

 f (t )e − jt dt =  e e dt +  e e dt
3t − jt −3t − jt
F ( ) =
− − 0

 (3− j ) t 0 
0
e   e− (3+ j ) t 
= e dt +  e
(3− j ) t
dt =  − (3+ j ) t
 + 
− 0  3 − j  −  −(3 + j )  0
1 1 6
= + = .
3 − j 3 + j 9 +  2

On the other hand, we have

6 6
G ( ) = = = F ( + 1).
10 + 2 +  2
( + 1) 2 + 9
On using the First Shift Theorem, we may thus write F ( + 1) will be the Fourier transform
6
of e− jt f (t ). The required function whose Fourier is G ( ) = is thus
10 + 2 +  2
 e− (3+ j )t , t  0
f (t ) =  (3− j )t
e , t  0.

2.2.3 The Second Shift Theorem

Theorem 2.2.2: If F ( ) is the Fourier Transform of f (t ) , then

F  f (t −  ) = e− j F ( ) (2.2.8)
where  is a constant.

We again illustrate the theorem by the following example.

Example 2.2.6:

9 EM
 
1, t  1, 2sin 
a) Given that when f (t ) =  , we have F ( ) = , apply the Second

 0, t  1, 
 1, 1  t  3,
Shift Theorem to find the Fourier transform of g (t ) = 
0, otherwise.
b) Verify the result by direct computation.

Solution:
 1, 1  t  3,
a) The function g (t ) =  is sketched in the Figure 2.2.2 below.
0, otherwise.

Figure 2.2.2: The Graph of


1, t  1,
This function is clearly f (t ) =  translated 2 units to the right, that is,
0, t  1,

g (t ) = f (t − 2)
2sin 
But we already have F  f (t ) = F ( ) = . Therefore,

2e−2 j sin 
F  g (t ) = F  f (t − 2) = e −2 j F ( ) = .


b) To verify this result by direct computation, we evaluate

 3
3
 e − jt  e − j − e −3 j
F  g (t ) =  g (t )e
− j t
dt =  e − jt
dt =   =
− 1  − j 1 j
j − j j
e −e
− j 2 t  − j 2 t  e − e − j 
=e   = 2e  
 j   2 j 
2e − j 2t sin 
= ,

as previously found.

10 EM
2.3 Spectra - Amplitude versus Frequency and Phase versus
Frequency Plots

In Fourier analysis of periodic functions it is established that although a periodic function

physically exists in the time or spatial domain, it can be regarded as comprising of components
with a variety of temporal or spatial frequencies, corresponding to the respective harmonics.
b
The amplitudes ck = ak2 + bk2 and phases k = arctan k featuring in the complex form
ak
representations of these components are obtained from the Fourier coefficients ak and bk in
the trigonometric expressions for the harmonics. Such a specification of the period function
in terms of the amplitudes and phases of the frequency components is said to constitute what
is popularly known in the literature as the frequency domain description, in contrast with
the initial time domain definition. Plots of amplitude versus frequency and phase and
frequency, taken together, are known as the spectrum of a function. Periodic functions are
then found to have discrete or line spectra, as the various amplitudes and phases are defined
only at discrete integer values k = 0, 1, 2, . In other words, the spectra assume non-zero
values only at certain discrete integral frequencies. Consequently, only a discrete set of
frequencies is required to synthesize a periodic function. In contrast, when analysing non-
periodic phenomena via the Fourier technique (Fourier integral representations) we find that,
in general, a continuous range of frequencies is required to synthesize the corresponding non-
periodic functions. Instead of discrete spectra we thus have continuous spectra. The
amplitude of the spectrum in these situations is given by the modulus F ( ) of the Fourier
transform F ( ) , while the phase is given by the argument arg( F ( )) of the Fourier
transform. We demonstrate this in the following example.

Example 2.3.1: Obtain and sketch the spectrum of the of the non-periodic function

 1, t  1,
f (t ) =  (2.3.1)
0, t  1.

Solution: In Example 2.1.2, we already found that the Fourier transform of this function is
twice the sinc function, that is,
2sin 
F ( ) = (2.3.2)

In this case the Fourier transform is thus purely real. The spectrum of f (t ) is thus depicted
2sin 
by plotting F ( ) = against the angular frequency  , as illustrated in Figure 2.3.1

below.

11 EM
 Figure 2.3.1: Spectrum of

Note the by expressing sin  in its Maclaurin’s form, we find that

sin  1 3 5 7  2 k +1 
lim = lim   − + − + + (−1) k + 
 →0   →0  (2k + 1)!
 3! 3! 3! 
 2 4 6  2k 
= lim 1 − + − + + (−1) k
+ .
 →0
 3! 3! 3! (2k + 1)! 
= 1.
Hence the magnitude of the zero frequency component of the spectrum is 2.

To illustrate further the concept of spread (extent) and location (point of concentration) of
the spectrum of a given non-periodic, we consider the following example, that you will
encounter subsequently in your course of Telecommunications Engineering.

2.4 The t −  Duality Principle

Sometimes it is possible to use the so-called t −  duality principle to deduce the Fourier
transform of a function based on the time domain equation of a function whose structure is
the same as the Fourier transform of some known time domain function, and vice versa. Below
we briefly outline this principle and illustrate with a simple example.

Recall from the definition of the Fourier integral, that


1
 F ( )e d  ,
j t
f (t ) = (2.4.1)
2 −
where

12 EM
 
F ( ) = 
−
f (t )e − jt dt. (2.4.2)

is the Fourier transform of f (t ). But in equation (2.4.1) we know that  is simply a dummy
variable of integration that we can replace with another dummy variable without affecting
the value of the integral. Consequently, we could choose to re-write this equation as

1
f (t ) =
2 − F ( )e j t d  . (2.4.3)

Then, from equation (2.4.3), replacing t by − , we find that


1
f (− ) =
2 −  F ( )e − j d  . (2.4.4)

Since  , as stated earlier, is simply a dummy variable we can replace it by t , thereby reinstating
the latter. As a result we then get

1
f (− ) =
2 −  F (t )e − jt dt. (2.4.5)

1
We easily recognise the right hand side as times the Fourier transform of F (t ). We thus
2
have the following result:

The t −  Duality Principle

If F ( ) is the Fourier transform of f (t ) , then f (− ) is the Fourier transform of F (t ).

Thus, in Example 2.3.1, we saw that the non-periodic function


 1, t  1,
f (t ) =  (2.4.6)
0, t  1.

has the Fourier transform
2sin 
F ( ) = (2.4.7)

This is depicted in Figure 2.4.1

Figure 2.4.1: Illustrating the Duality Principle

From the duality principle we immediately deduce that

13 EM
  2sin t 
F  = 2 f (− ) = 2 f ( ). (2.4.8)
 t 
Note that in writing the right-hand most equality we have invoked the fact that f (t ) is an
even function, as seen in Figure 2.4.2 below.

Figure 2.4.2: Illustrating the Duality Principle

Example 2.4.2
1
Given that the Fourier transform of f (t ) = u (t )e−t is F ( ) = , use the duality
1 + j
1
principle to deduce the Fourier transform of g (t ) = .
1 + jt

Solution
1
Since we know that F ( ) = is the Fourier transform of f (t ) = u (t )e−t , the duality
1 + j
1
principle informs us that G( ) = 2 u (− )e is the Fourier transform of g (t ) = .
1 + jt

14 EM
2.5 Fourier Transforms of Some Special
Functions
It will be recalled from our introductory statements about frequency spectra of signals, we saw
that the Fourier transform tells us about the frequency content of a signal. If we were to find
the Fourier transform of a signal composed of only one frequency component, such as
f (t ) = sin t , from physical considerations normally expect the exercise of finding the Fourier
transform to yield a spectrum containing only that one single frequency. Unfortunately,
however, if we attempt to find the Fourier transform of, say, f (t ) = sin t , through the formal
definition


 sin te
− jt
F ( ) = dt , (2.5.1)
−
we run into problems, since the integral cannot be evaluated in the usual sense due to the

fact that sint oscillates indefinitely as t → . Indeed, we find that 
−
sin t dt diverges and

hence one of the conditions for the existence of an integral representation of a non-periodic
is thus absent.

2.5.1 The Fourier Transforms of the Delta Function,  (t −  )

Using properties of delta function discussed during consideration of the Laplace Transform
and the definition of the Fourier transform, we find that

F  (t −  ) =   (t −  )e
− jt
dt (2.5.2)
−
But from the sifting property of the delta function, we know that



−
f (t ) (t −  )e − jt dt = f ( ) (2.5.3)

Therefore,

F  (t −  ) =   (t −  )e
− jt
dt = e − j . (2.5.4)
−
In particular, for  = 0, we have

F  (t ) =   (t )e
− jt
dt = 1. (2.5.5)
−
The latter result is depicted in Figure 2.5.1.

By applying the duality principle, we find that

F 1 = 2 (− ) = 2 ( ), (2.5.6)
after considering that  (− ) =  ( ). This is illustrated in Figure 2.5.2. Physically, the
function F (t ) = 1 can be interpreted as a direct current waveform. This confirms that a d.c.
signal has one frequency component, namely, zero.

15 EM
 Figure 2.5.1: The Spectrum of

Figure 2.5.2: The Spectrum of

2.5.2 The Fourier Transforms of Some Functions

The result in equation (2.5.4), which for convenience we reproduce here as

F  (t −  ) =   (t −  )e
− jt
dt = e − j (2.5.7)
−
is very useful in deducing the Fourier transform of some periodic functions. The underlying
principle here is that since a periodic function can be represented as a superposition of sine
and cosine functions, it follows by the linearity property of the Fourier transform operator
that once we know the Fourier transforms of the individual sine and cosine functions the
result Fourier transform of the original periodic function can be obtained by superposition.
Thus, specifically, since on the basis of the duality principle, equation (2.5.4) yields the
Fourier transforms of e ja and e − ja as F e j  = 2 ( − a) and F e− j  = 2 ( +  )
, respectively, we find that the Fourier transform of cos at is
 e j a + e− ja  1
 = F e  + F e 
1
F cos at = F  j a − j a

 2  2 2 (2.5.8)
=   ( − a) +  ( + a) .
Likewise, we find that

16 EM
  e ja − e− ja  1
F e ja  − F e− ja 
1
F sin at = F  =
 2j  2j 2j
(2.5.9)

 ( − a) −  ( + a).
=
j
These two results thus provide the necessary building blocks for the construction of Fourier
transforms of periodic functions by means of the superposition principle.

2.6 The Relationship between the Fourier

Transform and the Laplace Transform
We have already remarked earlier that there are similarities between the Laplace transform and
the Fourier transform. But there are also some differences between the two. To identify some
of these differences, let us have a closer look at the defining equations

F  f (t ) =  f (t )e − jt dt (2.6.1)
−
and

L  f (t ) =  f (t )e− st dt. (2.6.2)
0
With the parameter in Laplace transform a complex number given by
s =  + j, (2.6.3)
we can of course recast the Laplace transform as

L  f (t ) =  f (t )e− t e− jt dt. (2.6.4)
0
Thus, now do we not only have a difference in the lower limits of integration, but also the
appearance of an additional factor e− t in the Laplace transform. For   0 this represents an
exponentially decaying factor, the presence of which means that the integral exist for a wider
variety of functions that the corresponding Fourier integral. To illustrate this point in greater
depth, consider the following example.

Example 2.6.1: Ascertain the whether or not the function

f (t ) = u (t )e3t (2.6.5)
has Laplace and Fourier transforms and under what conditions.

Solution
From the definition and by direct integration, for Laplace transform we have
  
 e− ( s −3)t 
L  f (t ) = L u (t )e  =  u(t )e 1
3t 3t
e dt =  e
− st − ( s −3) t
dt =   = , (2.6.6)
0 0  − ( s − 3) 0 s − 3

provided s  3.

17 EM
For the Fourier transform, on the other hand, we have
  
 e(3− j )t 
F  f (t ) = F u (t )e 3t
 =  u(t )e 3t
e − jt
dt =  e (3− j ) t
dt =  
− 0  (3 − j )  − (2.6.7)
1 1
= lim e3t e − jt = lim e3t ( cos t − j sin t )  ,
3 − j t → 3 − j t →
which clearly does not exist.

Clearly, therefore, f (t ) = u (t )e3t has a Laplace transform, but not a Fourier transform.

Example 2.6.2:
Comment on the result of setting s = j in Laplace transforms of the functions
f (t ) = u (t )e−2t and g (t ) = u (t )e2t .

Solution:
First, we recall that
L u (t )e − st  = and L u (t )e st  =
1 1
. (2.6.8)
s+2 s−2
Replacing s by j in both expressions we find that
1 1 1 1
→ and → .
s+2 2 + j s−2 j − 2
But we also know that

F u (t )e −2t =
1
2 + j
. 
Thus, replacement of s by j in the Laplace transform of f (t ) = u (t )e−2t yields its
Fourier Transform. On the other hand, however, we know that the Fourier transform,
 
F u(t )e2t , of the function f (t ) = u (t )e−2t does not exist, despite the replacement of
1
s by j in its Laplace transform yielding ! Consequently we must not simply
j − 2
1
interpret as the Fourier transform of g (t ) = u (t )e2t .
j − 2

Notwithstanding the foregoing example, the Fourier transform does possess certain
advantages over the Laplace transform. Thus, for instance, while the Laplace transform can
only be applied to functions that are initially zero, that is, are zero for t  0, the Fourier
transform is applicable to functions with domain −  t  . It important to note this
difference, particularly in situations where t does not represent time but, say, some spatial
variable, because we then often have to deal with negative values of t .

Finally, we should point out that although the inverse Fourier transform

18 EM
 
1
F  F ( ) =
−1
 F ( )e jt d  (2.6.9)
2 −
appears to require techniques of ordinary integration, while the corresponding Laplace
transform requires advanced techniques in the theory of functions of complex variable, we
should note that this is not necessarily always an advantage as it may seem because it is often
very difficult to perform the required integration of the inverse Fourier transform analytically.

19 EM
2.7 Convolution and Correlation
A concept, reminiscent of theory of the Laplace transform and which we encountered in our
earlier coverage, that also surfaces in connection with Fourier transforms is Convolution.
Convolution is a technique that is extremely important in signal and image processing. It
provides a means of computing the response or output of a system to an arbitrary input signal
if the impulse response (encountered in discussions of applications of the Laplace transform)
is known. It will be recalled that the impulse response is the response of the system to an
impulse function. Convolving the input signal with the impulse response results in the
response of the system to the arbitrary input. The second concept that also surfaces in
applications of the Fourier transform is Correlation. Correlation is a technique that can be
used to determine the time delay between a transmitted and a received signal as often occurs
in radar and sonar (under-water) detection equipment.

2.7.1 Convolution and the Convolution Theorem

If f (t ) and g (t ) are two piecewise continuous functions, their convolution, which, as already
introduced previously, is denoted by f  g , is defined as

f g = 
−
f ( ) g (t −  )d (2.4.9)

where f  g is a function t , which is often shown explicitly by writing ( f  g )(t ). As

articulated in the Chapter on Laplace transform, convolution is an integral with respect to the
dummy variable  . Although, in general, this dummy variable will, span the interval
−    , in many instances this interval is reduced.

Because convolution is commutative, that is, f  g = g  f , the convolution can be ritten

equivalently as

f g = 
−
f (t −  ) g ( )d (2.4.10)

In cases where f (t ) and g (t ) are zero for t  0, this expression reduces to that for Laplace
transform, namely,
t
f  g =  f (t −  ) g ( )d (2.4.11)
0

The Convolution Theorem then states that convolution in time is corresponds to

multiplication in frequency, as stated formally here below

Convolution Theorem:
If F ( ) = F  f (t ) is the Fourier transform of f (t ), while G( ) = F  g (t ) is the Fourier
transform of g (t ), then
F  f  g = F  f (t )F g (t ) = F ( )G( ). (2.4.12)

20 EM
This theorem thus gives a technique of computing the convolution of two functions using the
Fourier transform, since by the inverse Fourier transform formula we then find that
F −1 F ( )G( ) = f  g. (2.4.13)
Thus, it is possible to find the convolution of two functions f (t ) and g (t ) using the Fourier
transform through the following steps:
• Step 1:
Find the corresponding Fourier transforms F ( ) = F  f (t ) and G( ) = F  g (t ) ;
• Step 2:
Multiply these two Fourier transforms to obtain F ( )G( ) = F  f (t ) F  g (t ) ;
• Step 3:
Find the inverse Fourier transform of the product F −1 F ( )G( ) = f  g.
This is in fact the process used for finding convolutions using a computer.

A graphical representation of convolution is often useful in as far it can throw some light
on the underlying process and help us to determine the appropriate limits of integration. To
illustrate these points, let us consider the following example.

Example 2.7.1
a) Using the defining formula for convolution, compute the convolution f  g ,
given that f (t ) = u (t )e−t and g (t ) = u (t )e −2t , where u (t ) is the unit step function.
b) Verify the Convolution Theorem for these two functions.

Solution
The convolution of f and g is given by

f g = 
−
f (t −  ) g ( )d .

Graphs of f (t ) = u (t )e−t and g (t ) = u (t )e −2t are sketched in Figure 2.7.1.

Figure 2.7.1: Graphs of and

21 EM
Evaluating a convolution at the beginning can be difficult to conceptualise, and so we shall
develop the procedure and hence solution in stages.

The first step is to understand and provide interpretation to the functions that feature in the
integrand. Thus, if f (t ) = u (t )e−t , then f ( ) = u ( )e− . Likewise, if and g (t ) = u (t )e −2t , then
g ( ) = u ( )e−2 . On the other hand, g ( − ) is obtained by reflecting g ( ) in the vertical axis,
as depicted in Figure 2.7.2 (a). The process of reflection about the axis is sometimes called
folding in signal processing.

Figure 2.7.1(a): Graphs of f (t ) = u (t )e−t and g (t ) = u (t )e −2t

The folded graph can then be translated (shifted along the horizontal axis) by t units the left
or right by changing the argument of g to g (t −  ).

Figure 2.7.2(b): Graphs of and

If t is negative the graph moves to the left as shown in Figure 2.7.2 (b) above. On the hand, if
t is positive, the graph to the right, as shown in Figure 2.7.2 (c) below. In Figure 2.7.2, we have
superimposed the graphs of f ( ) , shown red, and g (t −  ), shown blue, for t being zero,
negative and positive. Quite clearly, when the graphs do not overlap, as in Figure 2.7.2 (a), the
product f ( ) g (t −  ) and hence the convolution, must be zero. To identify when the graphs
over lap, we examine the domains t  0 and t  0 separately.

22 EM
 Figure 2.7.2(c): Graphs of and

For t  0, there is clearly no overlap and it follows that f  g = 0.

When t  0, there is clearly an overlap from  = 0 to  = t , that is, 0    t. Hence,

t t
f  g =  e− e−2(t − ) d = e−2t  e d = e−2t et  = e −2t et − 1 = e −t − e −2t . (2.4.14)
t

0
0 0

The complete expression for the convolution then has the form
 e−t − e−2t , t  0,
f g =  (2.4.15)
0, t  0.
Using the unit step function, we may re-write this expression more compactly as
f  g = u(t ) ( e−t − e−2t ) (2.4.16)

In order to verify the Convolution Theorem for the functions f (t ) = u (t )e−t and
g (t ) = u (t )e −2t , from Table of Fourier transforms presented at the beginning of the Chapter,
we have
F ( ) = F u (t )e −t  = , G ( ) = F u (t )e−2t  =
1 1
.
1 + j 2 + j
Their product is therefore
1
F ( )G ( ) = .
(1 + j )(2 + j )
Expressing this product as partial fractions, we have
1 1 1
F ( )G ( ) = = − .
(1 + j )(2 + j ) 1 + j 2 + j
Taking the inverse transform of this product and exploiting the linearity property, we get

23 EM
  1   1 1 
F −1  F ( )G ( ) = F −1  =F
−1
 − 
 (1 + j )(2 + j )  1 + j 2 + j 
 1  −1  1 
= F −1  −F 
−t −2 t
= e −e ,
1 + j   2 + j 
as previously obtained, and thereby verify the convolution theorem for the two functions.

Example 2.7.2
c) Using the defining formula for convolution, compute the convolution f  g ,
 1, − 1  t  1,
given that f (t ) =  and g (t ) = u (t )e−t ,
0, elsewhere,
where u (t ) is the unit step function.

d) Verify the Convolution Theorem for these two functions.

Solution
The convolution of f and g is given by

f g =
−
 f (t −  ) g ( )d .
−2 t
Graphs of f (t ) = u (t )e−t and g (t ) = u (t )e are sketched in Figure 2.7.3.

2.7.3: Graphs of the “top-hat” function and

Evaluating a convolution at the beginning can be difficult to conceptualise, and so we shall
develop the procedure and hence solution in stages.

24 EM
 2.7.4 (a): Graphs of the “top-hat” function and its reflection about vertical axis

Like previously, the function g ( − ) is obtained by reflecting or folding the graph of the
function g ( ) in the vertical axis. This folding is shown in Figure 2.7.4 (a) above. The folded
graph can than be translated a distance t to the left or to the right by changing the argument
of the function g to g (t −  ) . If t is negative, the Figure 2.7.4 (a) moves to the left, whereas if
it is positive it moves to the right. The resulting graphs are shown in Figures 2.7.4 (b) – (f).

2.7.4 (b): Graphs of the “top-hat” function and

2.7.4 (c): Graphs of the “top-hat” function and

25 EM
 2.7.4 (d): Graphs of the “top-hat” function and

2.7.4 (e): Graphs of the “top-hat” function and

2.7.4 (f): Graphs of the “top-hat” function and

The convolution is then the integral of the “top-hat” function f ( ) and g (t −  ). We have
superimposed the graphs of f ( ) in Figure 2.7.4 (b) – (f). For values of  where there is
no overlap this product, and hence the convolution is zero. Inspection of the graphs shows
that for t  −1, as in Figure 2.7.4 (b), there is no overlap and so f ( ) g (t −  ) = 0. Hence
f  g = 0, t  −1.

26 EM
When −1  t  1 , as in Figure 2.7.4(d), there is an overlap, and hence a non-zero product.
This overlap occurs for values of  = −1 to  = t , that is, the interval −1    t. Within
this interval, f ( ) = 1 and g (t −  ) = e− (t − ) .

There then follows that if −1  t  1

t t
f g = e  e d = e
− ( t − )  t
d = e −t −t
e  = e−t et − 1 = 1 − e−t .
0
−1 0

When t  1, the graphs overlap, but only for values of  in the range −1    1. So
1 t
f  g =  e − (t − ) d = e − t  e d = e − t e  = e − t e1 − e −1  .
1

−1
−1 −1

The complete result of this convolution is thus given by

 0, t  −1,

( f  g ) (t ) = 1 − e , − 1  t  1,
−1−t

 −t 1 1
e ( e − e ) , t  1.

To verify the convolution theorem for the functions

 1, − 1  t  1,
f (t ) = 
0, elsewhere,
and
g (t ) = u (t )e−t ,
we must demonstrate that
F ( f  g ) (t ) = F  f (t )F g (t ) = F ()G() .
Since from the table of Fourier transform tables we already known that
2sin 
F ( ) = F  f (t ) = ,

and
1
G ( ) = F (t ) = .
1 + j
we must in fact demonstrate that

2sin 
− ( f  g ) (t ) e dt =  (1 + j ) .
− jt

To this we first observe that the convolution is defined differently over the various portions
of the domain of integration. Consequently, we must find the Fourier transform of the
transform by integrating over each part of the domain separately. In other words, we must
obtain the Fourier transform of the convolution by means of the expression
 −1 1 

 ( f  g ) (t ) e − jt dt =  ( f  g ) (t ) e − jt dt +  ( f  g ) (t ) e − jt dt +  ( f  g ) (t ) e − jt dt.
− − −1 1

27 EM
Substituting the pertinent expressions for the various integrands, we find that
−1 

 ( f  g ) (t ) e dt =  0 e − jt dt = 0,

− jt

− −1

1 1 1 1

 ( f  g ) (t ) e dt =  1 − e dt =  e e

− jt − jt − jt − (1+ j ) t
− (1+ t )
e dt −e −1
dt
−1 −1 −1 −1
1 1
 e − jt  −1  e
− (1+ j ) t

=  −e  
 − j  −1  −(1 + j )  −1
 e j − e − j  −1  e − (1+ j ) − e(1+ j ) 
= +e  ,
 j   1 + j 
 

 ( f  g ) (t ) e
− jt
dt =  ( e1 − e −1 ) e − t e − jt dt
1 1

= ( e1 − e −1 )  e − (1+ j ) t dt
1

 e − (1+ j ) t 
= − (e − e )  1 −1

 (1 + j ) 1
e − j (1+ j )
= ( e1 − e −1 ) .
1 + j
On adding the three components, we then get


e j − e − j −1  e
− (1+ j )
− e (1+ j )  − (1+ j )

 + (e − e )
−1 e
 ( f  g ) (t ) e
− jt
dt = +e  1

−
j  1 + j  1 + j
e j − e − j e(1+ j ) e − (1+ j )
= − e −1 + e1
j 1 + j 1 + j
e j − e − j e j e − j
= − +
j 1 + j 1 + j
e j − e − j 2 j  e j − e − j 
= −  
j 1 + j  2j 
2sin  2 j sin 
= −
 1 + j )
2sin 
=
(1 + j )
2sin  1
=  = F ( )G ( ),
 1 + j

28 EM
as asserted in the convolution theorem.

2.7.2 Correlation and the Correlation Theorem

If f (t ) and g (t ) are two piecewise continuous functions, their correlation, (sometimes called
cross-correlation), is for lack of a better notation we shall in this text symbolise by f  g ,
is defined as

f g = 
−
f ( ) g ( − t )d (2.7.9)

Note that mathematically this formula is virtually the same as that for the convolution, except
for the important difference that the function g is not folded, that is, we have g ( − t ) rather
than g (t −  ) . Correlation, unlike convolution, is not exactly commutative in the strict sense,
but it can be written alternatively as

f g = 
−
f ( + t ) g ( )d (2.7.10)

It is important to note in computation that it is the argument of f , which is t +  .

The corresponding Correlation Theorem states as follows:

Correlation Theorem
If F  f (t ) = F ( ) and F  g (t ) = G( ) , then the Fourier transform of the correlation of
f (t ) and g (t ) is given by
F  f  g = F ( )G(− ) (2.7.11)

This theorem thus gives a technique of computing the correlation of two functions using the
Fourier transform, since by the inverse Fourier transform formula we then find that
F −1 F ( )G(− ) = f  g. (2.7.12)
Thus, it is possible to find the correlation of two functions f (t ) and g (t ) using the Fourier
transform through the following steps:
• Step 1:
Find the corresponding Fourier transforms F ( ) = F  f (t ) and G( ) = F  g (t ) ;
• Step 2:
Replace the argument of G( ) = F  g (t ) by − and then multiply by
F ( ) = F  f (t ) to obtain F ( )G (− );
• Step 3:
Find the inverse Fourier transform of the product F −1 F ( )G(− ) = f  g.
Like for the case of convolution, this is in fact the process used for finding convolutions
using a computer.

29 EM
A graphical representation of correlation is also often useful in as far it can throw some
light on the underlying process and help us to determine the appropriate limits of
integration. To illustrate these points, let us consider the following example.

Example 2.7.3
e) Using the defining formula for correlation, compute the correlation f  g ,
given that f (t ) = u (t )e−t and g (t ) = u (t )e −2t , where u (t ) is the unit step function.
f) Verify the Correlation Theorem for these two functions.

Solution
The correlation of f and g is given by

f g =
−
 f (t −  ) g ( )d .

Graphs of f (t ) = u (t )e−t and g (t ) = u (t )e

−2 t
are sketched in Figure 2.7.5.

Figure 2.7.5: Graphs of and

Evaluating a correlation can also be difficult to conceptualise, and may require development
of the procedure and hence solution in stages.

The first step again is to provide interpretation to the functions that feature in the integrand.
Thus, if f (t ) = u (t )e−t , then f ( ) = u ( )e− . Likewise, if and g (t ) = u (t )e −2t , then
g ( ) = u ( )e−2 . As there is no folding, we do not need to give an interpretation to the
function g ( − ) , as was the case for convolution. Instead, we simply look at g ( − t ) as the
function g ( ) translated to the left or right a distance t along the horizontal axis. If t  0 the
translation is to the left and if t  0 , the translation is to the right, as depicted in Figures 2.7.6(a)
and (b).

Where the graphs do not overlap, the product f ( ) g ( − t ) , and hence the correlation is zero.
Specifically, when t  0 , the graphs overlap for 0    . Hence,

30 EM
    
 e−3  1 2t
f  g =  f (t −  ) g ( )d =  e e − −2( −t )
d = e 2t
e
−3
d = e 
2t
 = e .
0 0 0  −3  0 3

When t  0, the graphs will overlap for t    . Hence,

Figure 2.7.6(a): Graphs of and

Figure 2.7.6(b): Graphs of and

   
 e−3  1 −t
f  g =  f (t −  ) g ( )d =  e e − −2( −t )
d = e 2t
e
−3
d = e 
2t
 = e .
t t t  −3  t 3

To verify the correlation theorem for the functions f (t ) = u (t )e−t and g (t ) = u (t )e −2t , from
Table of Fourier transforms, we again have

F ( ) = F u (t )e −t  = , G ( ) = F u (t )e−2t  =
1 1
.
1 + j 2 + j
Therefore
1
F ( )G (− ) = .
(1 + j )(2 − j )

31 EM
On the other hand, taking the Fourier transform of the correlation expressions obtained
above, we get


F ( f  g ) (t ) =  ( f  g ) (t )  e − jt

−
0  0 
1 1 1 1
= 
3 −
e 2t e − jt dt +  e − t e − jt dt =  e (2− j )t d +  e − (1+ j )t d
30 3 − 30
0 
1  e(2− j )t  1  e − (1+ j ) t  1 1
=   −   = +
3  2 − j  − 3  1 + j  0 3(2 − j ) 3(1 + j )
1
= = F ( )G (− ),
(1 + j )(2 − j )

as expected.

It is instructive to observe the following points from the foregoing example. First, we note
that for both t  0 and t  0, we have
lim ( f  g ) (t ) = 0 .
t →

Secondly, for t  0 we have

 e 2t  1
lim ( f  g ) (t ) = lim   = ,
t →0 t →0
 3  3
and for t  0 we have
 e−t  1
lim ( f  g ) (t ) = lim   = .
t →0 t →0
 3  3
On the other hand, since t = 0 corresponds to a complete overlap of the two functions over
the domain 0     , if we multiply f ( ) and g ( ) , and then integrate over this domain of
complete overlap, we find that
  
 e−3t  1
0 f (t ) g (t )dt = 0 e dt = − 3  = 3 .
−3t

0
It thus follows that when there is no overlap we have a zero correlation and when there is a
complete overlap we get maximum correlation. When there is partial overlap the correlation
will lie some where between zero and the maximum value. It is this feature that underlies the
application of the concept of correlation in signal detection in radar and sonar systems.

32 EM