American International University-Bangladesh

Department of Electrical and Electronic Engineering

Digital Signal Processing Lab


Experiment no. 3(Final term): Discrete Fourier Series (DFS) and Discrete Fourier
Transform


Objective:
- To understand the Discrete Fourier Series (DFS) and Discrete Fourier Transform
(DFT)

Introduction
The analysis of real world signals is a fundamental problem for many engineers and
scientists, especially for electrical engineers since almost every real world signal is
changed into electrical signals by means of transducers, e.g., accelerometers in
mechanical engineering, EEG electrodes and blood pressure probes in biomedical
engineering, seismic transducers in Earth Sciences, antennas in electromagnetics, and
microphones in communication engineering, etc.

Traditional way of observing and analyzing signals is to view them in time domain.
Baron Jean Baptiste Fourier, more than a century ago, showed that any waveform that
exists in the real world can be represented (i.e., generated) by adding up sine waves.
Since then, we have been able to build (break down) our real world time signal in terms
of (into) these sine waves. It is shown that the combination of sine waves is unique; any
real world signal can be represented by only one combination of sine waves.

The Fourier transform (FT) has been widely used in circuit analysis and synthesis, from
filter design to signal processing, image reconstruction, stochastic modeling to non-
destructive measurements. The FT has also been widely used in electromagnetics from
antenna theory to radiowave propagation modeling, radar cross-section prediction to
multi-sensor system system design.

Discrete Fourier Transform (DFT)

If a discrete signal x(n) is periodic with a period of N. It means the signal satisfies
the following condition

) ( ) ( kN n x n x + =
, where k is integer and N is the fundamental period of the sequence. Then this signal can
be expressed as

) ........( .......... .......... ) (
1
) (
1
0
2
A e k X
N
n x
N
k
kn
N
j

=
=


,where ,......... 2 , 1 , 0 = n and ,......... 2 , 1 , 0 ), ( = k k X are
called the discrete Fourier transform co-efficients, which is given by


) .( .......... .......... .......... ) ( ) (
1
0
2
B e n x k X
N
n
kn
N
j

=


, where ,......... 2 , 1 , 0 = k . The discrete Fourier series co-efficients X(k) is
itself a 9complex-valued) periodic sequence with fundamental period of N. Using
N
j
N
e W
2

=
, we can express Equations (A) and (B) as

) .( .......... .......... .......... ) ( ) (
1
0
C W n x k X
N
n
nk
N

=
=

) ........( .......... .......... ) (
1
) (
1
0
D W k X
N
n x
N
k
nk
N

=



Now, we will define a Matlab function that implement Discrete Fourier Transform of a
periodic sequence.

function [Xk]=dfs(xn,N)
% Compute Discrete Fourier Series Co-efficients
% [Xk]=DFS co-efficients array over 0<=k<=N-1
% xn= one of period of the periodic sequence
% N= fundamental period
n=[0:1:N-1]
k=[0:1:N-1]
WN=exp(-j*2*pi/N);
nk=n'*k
WNnk=WN.^nk
Xk=xn*WNnk

Example 1: Find the DFT of a periodic discrete signal defined by
x(n)={ . 0,1,2,3,0,1,2,3,0,1,2,3.}


Solution: For the given signal, the fundamental period is N=4. The signal can be defined
one period as xn={0,1,2,3}.

By using the above function, we can compute the DFS by using the following code

>> xn=[0 1 2 3]

xn =

0 1 2 3

>> N=4

N =

4

>> xk=dfs(xn,N)


xk = 6.0000 -2.0000 + 2.0000i -2.0000 - 0.0000i -2.0000 - 2.0000i


Now, we will define another function that will implement the Inverse Discrete Fourier
Transform (IDFT).

function [Xk]=idfs(Xk,N)
% Compute Inverse Discrete Fourier Series Co-efficients
% [xn]=one of period of the periodic sequence
% [Xk]=DFS co-efficients error
% N= fundamental period
n=[0:1:N-1]
k=[0:1:N-1]
WN=exp(-j*2*pi/N);
nk=n'*k;
WNnk=WN.^(-nk);
Xk=(Xk*WNnk)/N

We can use this function to compute one period of signal is the DFS co-efficients are
given as follows:
>> N=4

N =

4

>> Xk=[6.0000 -2.0000+2.0000i -2.0000-0.0000i -2.0000-2.0000i]

>> xn=idfs(xk,N)

The output of the code will be
xn = 0.0000 - 0.0000i 1.0000 - 0.0000i 2.0000 - 0.0000i 3.0000 + 0.0000i

The xn shows one period of the discrete signal.

Example 2: A periodic signal is given by
1 ) ( = n x
mNnmN+L-1
0 = mN+Ln(m+1)N-1
where m=0,1, 2, , N is the fundamental period, L/N is the duty cycle
Find the DFS of the signal and plot the magnitude spectrum for different L and N.

The following Matlab Code will find the DFS and plot the magnitude spectrum
L=5;
N=20;
n=[-N:N+1];
x1=[ones(1,L), zeros(1,N-L+1)];
x=[x1 x1]
subplot(221)
stem(n,x)
title('Signal')
% find the DFS with period N=20 and duty cycle =5
k=[-N/2:N/2]
xn=[ones(1,L), zeros(1,N-L)]
Xk=dfs(xn,N);
magXk=abs([Xk(N/2+1:N) Xk(1:N/2+1)])
subplot(222)
stem(k,magXk)
title('DFS with N=20, L=5')

% find the DFS with period N=50 and duty cycle=5
N=50
L=5
k=[-N/2:N/2]
xn=[ones(1,L), zeros(1,N-L)]
Xk=dfs(xn,N);
magXk=abs([Xk(N/2+1:N) Xk(1:N/2+1)])
subplot(223)
stem(k,magXk)
title('DFS with N=50, L=5')

% find the DFS with period N=100 and duty cycle=5
N=200
L=5
k=[-N/2:N/2]
xn=[ones(1,L), zeros(1,N-L)]
Xk=dfs(xn,N);
magXk=abs([Xk(N/2+1:N) Xk(1:N/2+1)])
subplot(224)
stem(k,magXk)
title('DFS with N=200, L=5')

20 0 20 40
0
0.2
0.4
0.6
0.8
1
Signal
10 5 0 5 10
0
1
2
3
4
5
DFS with N=20, L=5
40 20 0 20 40
0
1
2
3
4
5
DFS with N=50, L=5
100 50 0 50 100
0
1
2
3
4
5
DFS with N=200, L=5



Several interesting observations can be made from the plots. The envelopes of the DFS
co-efficients of square wave is looks like a sinc functions. The amplitude at k=0 is equal
to L. The students should now find the spectrum of the signal by changing the duty cycle
L=5 to L=10

Discrete Fourier Transform(DFT)

To compute the Fourier transform numerically on a computer, discretization plus
numerical integration are required. This is an approximation of the true (i.e.,
mathematical), analytically-defined FT in a synthetic (digital) environment, and is called
Discrete Fourier Transformation (DFT). The DFT of a continuous time signal sampled
over the period of T, with a sampling rate of t can be given as

=

=
1
0
2
) ( ) (
N
n
t n f m j
e t n s
N
T
f m S

(4)
where f=1/T, and, is valid at frequencies up to f
max
= 1/(2t).


The Discrete Fourier Series (DFS) is practically equivalent to the Discrete Fourier
Transform (DFT) when 0nN-1. [The students are asked to check the text book for
proof] Therefore, the implementation of DFT should be similar to that of DFS. Hence
the earlier dfs and idfs functions can be renamed as dft and idft as follows:

function [Xk]=dft(xn,N)
% Compute Discrete Fourier Series Co-efficients
% [Xk]=DFS co-efficients array over 0<=k<=N-1
% xn= one of period of the periodic sequence
% N= fundamental period
n=[0:1:N-1]
k=[0:1:N-1]
WN=exp(-j*2*pi/N);
nk=n'*k
WNnk=WN.^nk
Xk=xn*WNnk


function [Xk]=idft(Xk,N)
% Compute Inverse Discrete Fourier Series Co-efficients
% [xn]=one of period of the periodic sequence
% [Xk]=DFS co-efficients error
% N= fundamental period
n=[0:1:N-1]
k=[0:1:N-1]
WN=exp(-j*2*pi/N);
nk=n'*k;
WNnk=WN.^(-nk);
Xk=(Xk*WNnk)/N



Example 3: Let x(n) be a 4-point sequence define by
X(n) = 1 0n3
= 0, otherwise

a. Compute the Discrete Time Fourier Transform (DTFT) of x(n)
b. Compute the 4-point DFT
c. Compute the 8-point, 16-point and 64 point DFT of x(n)
d. Compare the results of all DFTs


The DTFT of x(n) is defined by

=
3
0
) ( ) (
n
jn j
e n x e X



3 2
1

+ + + = e e e
j j


The following piece of MatLab code will compute and plot the magnitude of DTFT. In
addition 4-point, 8-point and 16-point DFTs are also calculated and plotted for
comparison.

w=0:pi/200:2*pi
H=1+exp(-j*w)+exp(-j*2*w)+exp(-j*3*w)
subplot(221)
plot(w,abs(H))
title('Magnitude of DTFT of x(n)')
axis([0 6 0 4])

% computer the 4-point DFT
x=[1 1 1 1]
N=4
n=0:1:(N-1)
X4=dft(x,4)
subplot(222)
stem(n,abs(X4))
title('Magnitude of 4-point DFT')
axis([0 3 0 4])


% compute the 16-point DFT
N=16
x=[ones(1,4) zeros(1,12)] % Note that x has been zero padded
n=0:1:(N-1)
X8=dft(x,N)
subplot(223)
stem(n,abs(X8))
hold on
plot(n,abs(X8),'-.k')
title('Magnitude of 16-point DFT')
axis([0 16 0 4])


% compute 32 point DFT
N=32
x=[ones(1,4) zeros(1,28)] % Note that x has been zero padded
n=0:1:(N-1)
X8=dft(x,N)
subplot(224)
stem(n,abs(X8))
hold on
plot(n,abs(X8),'-.k')
title('Magnitude of 32-point DFT')
axis([0 32 0 4])


The Magnitude and the phase of DTFT of x(n) are shown in the following figures

0 2 4 6
0
1
2
3
4
Magnitude of DTFT of x(n)
0 1 2 3
0
1
2
3
4
Magnitude of 4point DFT
0 5 10 15
0
1
2
3
4
Magnitude of 16point DFT
0 10 20 30
0
1
2
3
4
Magnitude of 32point DFT




Now, we should compute and plot 4-point DFT, 8-point DFT and 16-point DFT by using
the following MatLab code

Note: Base on the plots we can conclude that zero-padding is an operation in which
more zeros are appended with the original sequence. The resulting longer DFT
provides closely resemblances to the DTFT of the original sequence.