1|Page

EEE504:
DIGITAL SIGNAL PROCESSING (DSP)
LECTURE NOTES
DISCRETE FOURIER (DFT) AND FAST FOURIER TRANSFORM (FFT)

By:
Olorunniwo O.
Dept. of Electronic & Electrical Engineering
Obafemi Awolowo University
Ile-Ife
 2|Page

1. Introduction : Background on Fourier Transform

2. Development of DFT
• Aliasing issue
3. Implementation of FFT

1. INTRODUCTION: Background on Fourier Transform

1.1 Historical Perspective
Fourier analysis is a family of mathematical techniques which is based on the decomposition of signals
into sinusoids and the synthesis of these back into the original form. Why? 1. It provides a method of
mapping signals (and indeed the visualization) between time-domain and frequency. 2. The Fourier
representation of signals in frequency domain affords an alternative view of discrete-time signals. 3. It
presents an efficient computation of certain time-domain operations, namely: convolution and
correlation.
Fourier analysis is named after Jean Baptiste Joseph Fourier (1768-1830), French mathematician and
physicist. Amongst other contributors to this field, Fourier is honored for his mathematical discoveries
and insight into the practical usefulness of this technique. He presented a paper in 1807 to the Institut de
France on the use of sinusoids to represent temperature distribution.
Moreover, tools for Fourier analysis consist of discrete Fourier series (DFS), discrete Fourier transforms
(DFT), discrete–time Fourier transform (DTFT), and the fast Fourier transforms (FFT). Table 1 outlines
the four types of transforms. Most DSP processors are able to compute FFT with software algorithms
for major demanding applications. Note: Fourier analysis is associated with both discrete and continuous
signals— periodic or aperiodic.
Table 1

1.2 Mathematical Overview

1.2.1 Fourier Series
Consider this: A periodic signal with period T , which is continuous, such that

f (t ) = f (t ± nT ) n = 0,1,2, K
In addition to being periodic, assume f (t ) has only a finite number of discontinuities in any finite period,
then the integral
α +T
∫α f (t ) dt
 3|Page

is finite, where α is an arbitrary real number, and f (t ) could be expanded into infinite trigonometric
series
a0
f (t ) = + a1 cos ωt + a2 cos 2ωt + K + b1 sin ωt + b2 sin 2ωt + K , and ω = 2π T
2
The above trigonometric series is referred to as the Fourier series. Compactly, the Fourier series is
expressed as
a0 ∞
f (t ) = + ∑ (an cos nωt + bn sin ωt )
2 n=1
And,
2 α +T
a0 =
T ∫α s (t )

2 α +T
T ∫α
ak = s (t ) cos kωt dt

2 α +T
bk = ∫ s (t ) sin kωt dt
T α

In summary, the function f (t ) could be described completely in terms of the coefficients,

a o , a1 , a 2 , K , b1 , b2 , K known as the harmonic terms. These coefficients represent a frequency
domain description of the signal.

1.2.2 Fourier Transform

The Fourier series expresses a periodic wave in terms of a d.c. value and a series of sinusoidal signals. The
sinusoidal components have frequencies equal to the multiple frequency of the original wave, called
harmonics of the fundamental frequency. Hence, the Fourier transform of the continuous function
f (t ) is defined as F (s ) ,
∞
F (s ) = ∫ f (t )e
− st
dt
−∞
The Fourier transform allows non-periodic signals to be decomposed into its frequency components. In
addition, F (s ) is normally complex, and it is expressed both in magnitude and phase of the frequency
components. Further, the Fourier transform is sometimes referred to as the Fourier spectrum or simply
the spectrum, where
F (s ) = magnitude or amplitude spectrum ,and
∠F (s ) = phase spectrum

The inverse Fourier transform (IFT) operates on the frequency spectrum of the signal to recover in the
time variation, the signal shape. The IFT is given by
∞
1
f (t ) = ∫ F (s )e
st
ds
2π −∞
Surprisingly, though, there exist some functions without Fourier transform—e.g. pure alternating current,.
sin t , step function H (t ) , an impulse function, δ (t ) to mention a few. Therefore, conditions for the
existence of Fourier transform are:
1. The integral of f (t ) from − ∞ to ∞ must exist;
 4|Page

2. Any discontinuities in f (t ) are finite.

If the above conditions are met, it suggests that the Fourier integrals converge, and the mathematical
expressions are physically realizable.
Six common transform pairs for references are listed below:
∞ ∞
If f (t ) = e ∫e ∫ e e ds = e
−πt 2
, then F (s ) = −πt 2 −πs 2 2 2
− st −πs + st −πt
1. e dt = e ⇔
−∞ −∞
∞ ∞
sin πt
2. If f (t ) = sinc t = , then F (s ) = ∫ sinc t e −st dt = ∏(s ) ⇔ ∫ ∏ (s )e +st ds = sinc x Note:
πt −∞ −∞

1 s 〈 12
Rectangle function, ∏(s ) = 
0 s 〉 12
∞ ∞
If f (t ) = sinc 2 t , then F (s ) = ∫ sinc t e dt = Λ(s ) ⇔ ∫ Λ(s )e
2 − st + st
3. ds = sinc 2 t
−∞ −∞

1 − s s 〈1
Note: triangle function, Λ (s ) = 
0 s 〉1
2
What does this imply? It postulates that 1. The transform of the Gaussian function, e −πt , is the same
Gaussian function. 2. The transform of the sinc function is the unit rectangle function, ∏ . 3. The
transform of the sinc function is the triangle function, Λ , of unit height and area. Table 2 illustrates
some of the Fourier transform pairs.

1.2.3 Basic Theorems in Fourier Transforms

a. Similarity Theorem: If f ( x ) has the Fourier transform F (s ) , then f (ax) has the Fourier transform
1
F (s a ) .
a
b. Addition Theorem: If f ( x ) and g ( x ) have the corresponding Fourier transform F (s ) and G(s ) , then
f (x ) + g ( x) has the Fourier transform F (s ) + G(s ) .
This theorem reflects the suitability of the Fourier transform for dealing with linear problems.
c. Shift Theorem: If f ( x ) has the Fourier transform F (s ) , then f ( x − a ) has the Fourier transform
e − j 2πas F (s ) . This implies that when a given function is shifted in the positive direction by an
amount a, no Fourier component changes in amplitude. The Fourier component changes are
confined to the phase.
d. Modulation Theorem: If f ( x ) has the Fourier transform F (s ) ,then the Fourier transform is
 ω  1  ω 
F s −
1
2  + 2 F s + 
 2π   2π 
e. Convolution Theorem: If f ( x ) and g ( x ) have the corresponding Fourier transform F (s ) and G(s ) , then
f (x ) ∗ g (x ) has the Fourier transform F (s )G(s ) . This implies that the convolution of two functions is
simply the multiplication of their transforms.
f. Derivative Theorem: If f ( x ) has the Fourier transform F (s ) , then f ' (x ) has the Fourier transform
j 2πsF (s ) . Since taking the differentiation of a function multiples it transform by j 2πs , it implies
 5|Page

differentiation enhances the higher frequencies, attenuates the lower frequencies, and suppresses any
zero-frequency component.

Table2
Signal f ( x ) Fourier Transform F (s )

1.2.4 Discrete-Time Fourier Transform

Many continuous functions with independent variables have obtainable Fourier transforms. However,
other existing function values are given only as discrete values of the independent variable. This is the
case of observable systems, with physical measurements made at regular time intervals. Thus, discrete-
time Fourier transform is the Fourier representation of sequences.
Assume for an arbitrary input sequence x[n] , a Fourier representation using a Fourier integral is of the
form,
1 π
x[n] = ∫ π X (e )e dω , and
j ω ω
j n

2π −
 6|Page

∞
( )
X e jω = ∑ x[n]e − jωn
−∞

In addition to this, there exists a synthesis formula referred to as the inverse Fourier transform. This
represent x[n ] as a superposition of infinitesimally small complex sinusoids of the form,
1
2π
( )
X e jω e jωn dω

( )
with ω ranging over an interval of length 2π and with X e jω determining the relative amount of each
complex sinusoid. Hence, as stated earlier, the Fourier transform is a complex-valued function of ω and
( )
the quantities X e jω and ∠X e ( ) are the magnitude and phase of the Fourier transform.
jω

1 π
X (e )e
2π ∫
Note: Determining the class of signals that can be represented by x[n] = jω jωn
dω is
−π

1
equivalent to considering the convergence of the infinite sum given by
2π
( )
X e jω e jωn dω . That is to say,

1. If the sequence x[n ] is absolutely summable, then the Fourier transform of the sequence
X (e jω ) exists.
2. The series can be shown to converge to a continuous function of ω .

All stable sequences are absolutely summable, therefore, all stable sequences have Fourier transforms.
And by extension, any stable system would have a finite and a continuous frequency response. In
actual fact, absolute summability is a sufficient condition for the existence of Fourier
representation, and it also guarantees convergence.

Symmetry Properties of the Discrete-Time Fourier Transform

The symmetry properties of the Fourier Transform are often useful in simplifying transform problems.
Table 3 summarizes the symmetry properties of the Fourier Transform for a sequence signal x[n ] .
Definitions
1. A conjugate-symmetric sequence x e [n ] is defined as a sequence for which x e [n ] = x e∗ [− n ] ;
2. A conjugate-antisymmetric sequence x o [n ] is defined as a sequence for which x o [n ] = − x o∗ [− n ], where
* denotes complex conjugation.

Any sequence x[n ] can be expressed as a sum of a conjugate-symmetric and conjugate-antisymmetric

sequence, such that,
x[n ] = x e [n ] + x o [n ]
where, x e [n ] = (x[n] + x [− n]) = x [n]
1
2
∗ ∗
e

and , xo [n] = (x[n] + x [− n]) = − x [− n]

1
2
∗ ∗
o

Moreover, a real sequence that is conjugate-symmetric such that x e [n ] = x e [− n ] is called an even

sequence, while the conjugate-antisymmetric such that x o [n ] = − x o [− n ] is called an odd sequence.
( )
Thus, a Fourier transform X e jω can be decomposed into a sum of conjugate-symmetric and
conjugate-antisymmetric functions, such as,

( ) ( )
X e jω = X e e jω + X o e jω ( )
 7|Page

Table3

( ) [X (e ) + X (e )]
X e e jω = 1
2
jω ∗ jω

Xo (e ) = [X (e ) − X (e )]
jω 1
2
jω ∗ jω

In addition to the Symmetry properties, a variety of theorems govern the operations on the sequences to
obtain a transform. These are similar to the theorems for the Fourier transform of continuous-time
signals. Table 4 summarizes Discrete-time Fourier transform properties.

2 Development of Discrete Fourier Transform (DFT)

A periodic function is function that is describable by a sequence of coefficients at discrete intervals.
However, various questions of discreteness arise in connection with periodic functions. In this case,
discrete Fourier transform (DFT) may be viewed as a special case of continuous frequency. The DFT
may be regarded as a string of equally spaced delta functions. That is, the DFT itself is sequence, rather
than a function. By definition, consider a time series f [τ ] of a discrete time, τ ; if the sampling interval is
T and the first sample of interest occurs at t = t 0 , then,

f [τ ] = v(t o + τT ) τ = 0,1,2,3,..., N − 1
 8|Page

Table4

1 N −1
and there exist F [υ ] = ∑ f [τ ]e − j 2π (υ N )τ
N τ =0
This implies the DFT approximates the Fourier transform of the function underlying the data samples,
since it provides only for a finite set of discrete frequencies. The DFT is readily implemented using the
MATLAB® using the fft( ).
Further, given the discrete transform, F [υ ] , it is possible to recover the time series f [τ ] with the aid of
the inverse relationship,
N −1
f [τ ] = ∑ F [υ ]e j 2π (v N )τ
υ =0

The inverse summation of the Fourier transform is performed by the ifft ().

3 Implementation of Fast Fourier Transform

In 1965 a method of computing discrete Fourier transforms suddenly became widely known and
revolutionized many fields where onerous computing was an impediment to progress. The computational
algorithms which have come to be known as fast Fourier transform (FFT) is based upon the
decomposition algorithm of the discrete Fourier transforms of a sequence length N into successively
smaller discrete Fourier transforms.
 9|Page

Consider a method to understanding FFT: A sequence of N elements may be divided into two shorter
sequences of N/2 elements each placing the even-numbered elements into first sequence and the odd-
numbered ones into the second. For example, {8 7 6 5 4 3 2 1} could be split into {8 6 4 2} and {7 5 3
1}. Each of the sequences possesses a DFT. That is,

{8 7 6 5 4 3 2 1 } = {8 0 6 0 4 0 2 0} + {0 7 0 5 0 3 0 1}

Using the stretching and shift theorem, the desired DFT can be obtained such that, if
1. {P Q R S } corresponds to the sequence {7 5 3 1},
then, {0 7 0 5 0 3 0 1} would correspond to the stretched sequence
1
2
{P Q R S P Q R S } .
2. Similarly, if {8 6 4 2} has {A B C D} as its corresponding sequence,
, then {8 0 6 0 4 0 2 0} would have a corresponding stretched sequence
1
2
{A B C D A B C D}
3. Hence, the shift theorem states,
4. {0 7 0 5 0 3 0 1} ⊃ 12 {P WQ W 2 R W 3 S W 4 P W 5Q W 6 R W 7 S }
5. The DFT is simply the addition of both derivations [(2)+(4)].
The multiplication by W means rotation by one Nth of a revolution in the complex plane, so the effect
of the shift is to apply a phase delay that increases progressively along the sequences of the elements
{P Q R S P Q R S }.
Noticeably, the transformation with N=8 has been broken down into two transformations with N=4,
which potentially represents a 50% saving time.

Figure 1: Fast Fourier Transform algorithm

3.1 Applications of the FFT

FFT finds application in subjects such as x-ray diffraction and radio interferometry, wavelet transforms,
waveform decomposition and filters, envelopes detection and dynamic spectrum, to mention a few. In
these fields, the introduction of the FFT merely speeds up the known Fourier techniques.

Review Questions
1. Determine the discrete spectrum of a periodic sequence f [τ ] with period N=4 given by
f [τ ] = {0 1 1 0} for τ = 0,1,2,3 .
 10 | P a g e

N −1
1
Note: The DFT of a periodic signal is defined as F [υ ] =
N
∑ f [τ ]e
τ=0
− j 2π (υ N )τ
, and the spectrum

is F [υ ] .
2

2. Determine the Fourier Transforms of the following waveforms

a. b.

c. d.

3. A volcano on the floor of the Pacific Ocean erupted near an inhabited island, causing the sea
surface to rise and fall, reaching a maximum height of about 10 m. The height was recorded by
the captain of a vessel standing offshore, using a sextant to determine the distance from the
water to the top of the cliff. Later examination showed that the height h(t ) could be represented
approximately by
(
h(t ) = 11 sin (45° − 72°t ) exp − t 2 5 )
Where, h is in meters and t is in minutes. The volcano erupts from time to time.
(a) Obtain the Fourier transform h ( f ) of h(t )
(b) At what frequency, in cycles per minute, will the excitation be at a maximum?
Assume the Fourier transforms of F (cos πx ) = C(s ) ≡ 12 δ (s + 1
2
) + 12 δ (s − 12 ) and
F (sin πx ) = jI1 (s ) ≡ 12 δ (s + 1
2
) − 12 δ (s − 12 )
4. Outline step-by-step, using the stretching and shift theorems, the FFT algorithm.
i. In what fields of endeavor is the FFT applicable today?
ii. What is the major advantage of FFT implementation over DFT techniques?

5. List types of Fourier Transform and examples of signals it operates on.

State the following theorems
a. addition theorem
b. modulation theorem
c. derivative theorem
d. shift theorem

i. If the height of a pulse wave is 70%, and the width is doubled, what would be the corresponding
effects on its Fourier transform waveform? What would be the % height of the waveform?
 11 | P a g e

In radio and television applications, a harmonic carrier wave is modulated by an envelope. Sketch
the Fourier transform of a modulated signal. Describe briefly the spectrum of the envelope. What
is the inverse Fourier Transform of the waveform in P5.

P5