Ecr305 L8 SP 2017

Digital Signal Processing ECR 305_L8
Lecture – 8 Objectives
N 1
X  k   DFT [ x(n)]   x  n WNnk
n 0
1 N 1
x  n   IDFT [ X (k )]   X  k WN nk ,
N k 0
2
j
WN  e N
To learn and understand
 DFT - the frequency domain
Discrete Fourier Transform sampling & reconstruction of

discrete-time signal.
& Fast Fourier Transform

 FFT – an efficient computation
technique of DFT.
IUB Dr. Abdur Razzak 1
Introduction ECR 305_L8
In lecture-4 & 5, we have studied transform-domain

representations of discrete signals.
 The discrete-time Fourier transform (DTFT) provides the frequency-
domain () representation for absolutely summable sequences (see
lecture-5).
 The z-transform provides a generalized frequency-domain (z)
representation for arbitrary sequences (see lecture-4).
These transforms have two features in common:

 First, the transforms are defined for infinite-length sequences.
 Second, and the most important, they are functions of continuous
variables ( or z).
From the numerical computation viewpoint (or from MATLAB's

viewpoint), these two features are troublesome because one has
to evaluate infinite sums at uncountably infinite frequencies
Introduction (contd..) ECR 305_L8
To use MATLAB, we have to truncate sequences and then

evaluate the expressions at finitely many points, which are the
approximations to the exact calculations. In other words, the
discrete-time Fourier transform and the z-transform are not
numerically computable transforms.
Therefore we need a numerically computable transform, which is
obtained by sampling the discrete-time Fourier transform in the
frequency domain. This is known Discrete Fourier Transform.
From Fourier analysis we know that a periodic sequence can
always be represented by a linear combination of harmonically
related complex exponentials (which is a form of sampling). This
gives us the Discrete Fourier Series (DFS) to finite-duration
sequence which leads to a new transform, called DFT.
Discrete Fourier Series ECR 305_L8
Let us consider a periodic sequence
x n  ~
~ x n  kN 
where N is the fundamental period of the sequence.
From Fourier analysis, we know that the periodic function can be

synthesized as a linear combinations of complex exponentials
whose frequencies are multiples (or harmonics) of the
fundamental frequency (which in our case is 2/N).
From the frequency-domain periodicity of the discrete-time

Fourier transform, we conclude that there are a finite number of
2
harmonics; the frequencies are { k , k = 0, 1, 2,……, N–1}.
N
Discrete Fourier Series (contd..) ECR 305_L8
Therefore, the discrete Fourier series representation of a periodic

sequence ~x n  can be expressed as
N 1 2
x n    X k e
~ 1 ~ j kn
N
, n  0,1,2,..., N  1 IDFS
N k 0
where X k ,
~
k  0,1,2,..., N  1  are called the discrete
Fourier series coefficients, which are given by
N 1 2

j
X k   x n e
~ ~ nk
N
, k  0,1,2,..., N  1 DFS
n 0

2
j
Using WN  e N , previous 2 equations can be expressed as
 Analysis or a DFS equation:
N 1
ˆ DFS~
X k   x n    ~
x n WNnk
~
n 0
 Synthesis or an inverse DFS equation:
 
N 1
x n  ˆ IDFS X k   X k WN
~ 1 ~
~
N

k 0
 nk

In matrix form, the DFS and IDFS equations can be rewritten as
X N  WN x N 1 *
x N  WN-1X N  WN X N
N
where the matrix WN called a DFS matrix given by

WN ˆ WNkn 
 x0   X 0  
0 k , n  N 1
n
 x1   X 1 
1 1 
xN    XN   
1 ...
 WN1 ... WN N 1   :   
k 1 :
    
 :
 xN - 1  XN  1
: ... 
  N 1  N 12 
1 WN ... WN 

Example-1 ECR 305_L8
Find DFS representation of ~x n   ....,0,1,2,3,0,1,2,3,0,1,2,3,..... .


Solution: Given N = 4. So W4  e  j 2 / 4
e  j / 2
 cos / 2  j sin  / 2   j
3
X k    ~
x n W4nk ,
~
Now k  0, 1, 2, 3
n 0
3
X 0   ~
x n  j   ~x 0  ~
x 1  ~
x 2  ~
x 3  0  1  2  3  6
Hence ~ n0
n 0
3
X 1   ~
x n  j   0  1 j   2 j   3 j   2  2 j
~
Similarly, n 2 3
n 0
3
X 2    x n  j   0  1 j   2 j   3 j   2
~ ~ 2n 2 4 6
n 0
3
X 3   ~
x n  j   0  1 j   2 j   3 j   2  2 j
~ 3n 3 6 9
n 0

Relation to z-transform ECR 305_L8
x n  by periodically repeating x(n)

Construct a periodic sequence ~
with a period N,
~x n , 0  n  N 1
xn   
 0, elsewhere
x n  is given by,
The DFS of ~
2 2 n
N 1 j N 1  
X k    ~
x n e   xn  e
~ nk j k

N N
n 0 n 0  
Comparing with z-transform, X  z   n0
N 1 ~
x n  z n
,we have
X k   X z  z e j 2N k
~
which means that DFS k  represents N evenly spaced samples

~
X
of the z-transform X(z) around the unit circle.
Relation to DTFT ECR 305_L8
The DTFT of x(n) is given by

N 1
X    n 0
xn e  jn
Comparing DFS & IDFS equations, we have

N 1 2 N 1
 
j
X k   x n e x n e  jn  X     2 k
~ ~ nk
~
N

n 0 n 0 2 N
 k
N
2 2
Let 1 ˆ and k ˆ k  k1 , then the DFS
N N
 
X k   X e jk  X e jk1  
which means that the DFS is obtained by evenly sampling the
2𝜋
DTFT with the sampling interval 𝑁 .
Discrete Fourier transform ECR 305_L8
The discrete Fourier series provided us a mechanism for numerically

computing the discrete-time Fourier transform. It also alerted us to a potential
problem of aliasing in the time domain.
Mathematics dictates that the sampling of the discrete-time Fourier transform

result in a periodic sequence x(n). But most of the signals in practice are not
periodic. They are likely to be of finite duration. How can we develop a
numerically computable Fourier representation for such signals?
Theoretically, we can take care of this problem by defining a periodic signal

whose primary shape is that of the finite-duration signal and then using the
DFS on this periodic signal.
Practically, we define a new transform called the Discrete Fourier Transform

(DFT), which is the primary period of the DFS. This DFT is the ultimate
numerically computable Fourier transform for arbitrary finite-duration
sequences.
Discrete Fourier transform (contd..) ECR 305_L8
Discrete Fourier transform (DFT) of an N-point sequence:
 X k , 0  k  N  1
~
X k  ˆ DFTxn   
 0, elsewhere
N 1
or X k    xn  WNnk , 0  k  N 1
n 0
Inverse discrete Fourier transform (IDFT) of an N-point

sequence:
1 N 1
x  n    X  k  WN nk , 0  n  N 1
N k 0

Properties of DFT ECR 305_L8
Linearity: DFTax1 n  bx2 n  a DFTx1 n  b DFTx2 n
Circular folding: If an N-point sequence is folded, then the

result would not be an N-point sequence, and it would not be
possible to compute DFT. Therefore, we use the modulo-N
operation on the argument (-n) and define folding by
 x0, n0
x n N 
 xN  n , 1  n  N 1
The DFT of a circular folding is given by

 X 0 , k 0
DFTx n N   X  k N 
 X N  k , 1  k  N  1
Properties of DFT (contd..) ECR 305_L8
Circular shifting: An N-point DFT of a finite duration sequence

x(n) of length  N is equivalent to the N-point DFT of a periodic
sequence xp(n), which is obtained by periodically extending x(n),

x p n    xn  rN 
r  
Now suppose that we shift xp(n) by k units to the right,


xp n   x p n  k    xn  k  rN 
r  
 xp n , 0  n  N  1
Then the circular shift is x n   
0, otherwise
Note: Counter clockwise  +ve, clockwise  -ve direction.

Properties of DFT (contd..) ECR 305_L8
Circular convolution: Consider 2 N-point sequences x1(n) and

x2(n). The respective N-point DFTs are
N 1 2
X 1 k    x1 n e
j nk
N
, k  0,1,2,..., N  1
n 0
N 1 2
X 2 k    x2 n e
j nk
N
, k  0,1,2,..., N  1
n 0
Then X 3 k   X 1 k X 2 k 
Then the IDFT of X3(k), called the circular convolution, given by

N 1
x3 m    x nx m  n
n 0
1 2 N m  0,1,2,..., N  1

Determine the circular shifting of the sequence by 2 units:

x n   1,2,3,4
Solution:


Example-2 (contd..) ECR 305_L8

Determine the circular folding of the sequence:

x n   1,3,5,7,9,7,5,3,1


Determine the circular convolution of the following two

sequences: x n   2,1,2,1 x2 n   1,2,3,4
1
 
Solution:
N 1
x3 m    x nx m  n
n 0
1 2 N
where m  0,1,2,..., N  1
x3 m   14,16,14,16

N 1
x3 m    x nx m  n
n 0
1 2 N
where m  0,1,2,..., N  1
x3 m   14,16,14,16

Fast Fourier Transform ECR 305_L8
Although DFT is a computable transform, the numerical

computation of the DFT is time consuming and its straightforward
implementation is very inefficient especially for large sequences.
Therefore, several algorithms have been developed to efficiently

compute the DFT. These are collectively called Fast Fourier
Transform (FFT) algorithms.
In 1965, Cooley and Tukey showed a procedure to substantially

reduce the amount of computations involved in the DFT, which
led to the explosion of the application of the DFT including in the
digital signal processing area and to the development of other
efficient algorithms. These are collectively known as Fast Fourier
Transform (FFT).
Fast Fourier Transform (contd..) ECR 305_L8
Consider an N-point sequence x(n). Its N-point DFT is given by

N 1
X k    xn WNnk , 0  k  N 1
n 0
where WNnk  e  j 2 / N. To obtain one sample of X(k), we need N complex

multiplications and (N–1) complex addition.
Hence to obtain a complete set of DFT coefficients, we need N2

complex multiplications and N(N-1)  N2 complex additions. Also one
has to store N2 complex coefficients  WNnk  or generate internally at an
extra cost.
Therefore, the number of DFT computation for an N-point sequence

depends quadratically on N, i.e. N4 . Generally processing time for one
addition is much less than for one multiplication.

Fast Fourier Transform (contd..) ECR 305_L8
Goal of an efficient computation: In an efficiently designed

algorithm the number of computation should be constant per data
sample, and therefore the total number of computation should be
linear with respect to N.
The quadratic dependence of N can be reduced by realizing that

most of the computations (which are done again and again) can
be eliminated using the periodicity property
WNnk  WNk n  N   WNn k  N 
and the symmetry property WNnk  N / 2  WNnk
of the factor W .
nk
N

Compute a 4-point DFT of x(n) = {1, 1, 1, 1} and develop an

efficient algorithm for its computation.
3
X k    xn W4nk , 0  k  3; W4  e  j 2 / 4   j
n 0
The above computation can be done in the matrix form

 X 0 W40 W40 W40 W40   x0
 X 1   0 1 2 3 
x 1 
 W4 W 4 W4 W4   
 X 2 W4 W4 W4 W4   x2
0 2 4 6
   0 9  
 X 3 
  4
W W4
3
W4
6
W 
4  
x 3 
which requires 16 complex multiplications.

Efficient approach: Using periodicity
W40  W44  W48  1; W41  W45  W49   j
W42  W46  1; W43  W47   j
Putting these values into previous equation, we have

 X 0 1 1 1 1   x0
 X 1  1  j  1 j   x1 
    
 X 2 1  1 1  1   x2
     
 X 3  1 j  1  j   x 3
Using symmetry we obtain

X 0   x0   x1  x2   x3  x0   x2   x1  x3
 
g1 g2
X 1  x0   jx1  x2   jx3  x0   x2   jx1  x3

 
h1 h2
X 2   x0   x1  x2   x3  x0   x2   x1  x3

 
g1 g2
X 3  x0   jx1  x2   jx3  x0   x2   jx1  x3

 
h1 h2

4-point DFT (contd..) ECR 305_L8
step 1 step 2
Hence an efficient algorithm is
g1  x0  x2 X 0  g1  g 2
g 2  x1  x3 X 1  h1  jh2
h1  x0  x2 X 2  g1  g 2
h2  x1  x3 X 3  h1  jh2
These requires only 2 complex multiplications, which is a

considerably smaller number. A signal flow graph structure of
this algorithm is shown below:

Assignment-7 (due on next class) ECR305_L8
Problems : 7.8, 7.12, 7.14, 7.17, 7.24, 7.25

MATLAB implementation ECR 305_L8
MATLAB
Examples

MATLAB implementation of DFS ECR 305_L8
The DFS is a numerically computable representation for which

we need two nested for … end loop (one for summation and
another for DFS coefficients), which is not efficient in MATLAB.
An efficient implementation in MATLAB would be to use a

matrix-vector multiplication same as we used for DTFT.
~
Let ~
x and X denote the column vectors corresponding to the
primary periods of sequences and ~ , respectively.
x n 
~ X k 

MATLAB implementation of DFS (contd..) ECR 305_L8
Then DFS and IDFS equations can be rewritten as

~
X  WN ~
x
~ 1 *~
x  WN X
N
where the matrix WN called a DFS matrix given by
n
1 1 ... 1 
k 1 ... WN N 1 

WN ˆ WNkn 0 k , n  N 1  
 :
WN1
: ... 
  N 1  N 12 
1 WN ... WN 

dfs function ECR 305_L8
~
X  WN ~
x function [Xk] = dfs(xn,N)
% -------------------------------------------------------------------------
% Compute discrete Fourier series coefficients
% Xk : DFS coefficient array over 0 <= k <= N-1
% xn : one period of periodic signal over 0 <= n <= N-1
 
% N : fundamental period of xn
WN ˆ WNkn 0k , n N 1
% -------------------------------------------------------------------------
n = 0:N-1; % row vector for n
k = 0:N-1; % row vector for k
WN = exp(-j*2*pi/N); % Wn factor
nk = n'*k; % N by N matrix of nk values
WNnk = WN.^nk; % DFS matrix
2 Xk = xn*WNnk; % row vector for DFS coefficient
j
WN  e N
% ------------------------------------------------------------------

idfs function ECR 305_L8
~ 1 ~ function [xn] = idfs(Xk,N)

x  WN* X
N %=============================================
% Compute inverse discrete Fourier series coefficients
% Xk : DFS coefficient array over 0 <= k <= N-1
% one period of periodic signal over 0 <= n <= N-1
 
% N : fundamental period of xn
% -------------------------------------------------------------------------
WNnk = WN.^(-nk); % DFS matrix
2
j xn = (1/n).*(Xk*WNnk); % row vector for DFS coefficient
WN  e N
% -------------------------------------------------------------------------

MATLAB example 5.2b ECR 305_L8
A periodic square wave sequence is given by
1, mN  n  mN  L  1
x n   
~
0 mN  L  n  m  1N  1
where N is the fundamental period and L/N is the duty cycle.
Plot the magnitude X k  for L = 7 and N = 60.

~

MATLAB example 5.2b (contd..) ECR 305_L8
% File name: ex5p2.m

% Definition
L = 7; N = 60; k = -N/2:N/2; xn = [ones(1,L), zeros(1,N-L)];
% Calculation
Xk = dfs(xn,N); magXk = abs([Xk(N/2+1:N) Xk(1:N/2+1)]);
% Plotting
Subplot(2,1,1); stem(k, magXk);
title('DFS of square wave'); xlabel('k'); ylabel('Xtilde(k)');

MATLAB implementation of DFT ECR 305_L8
Then DFT and IDFT equations can be rewritten as
X  WN x
1 *
x  WN X
N
where the matrix WN called a DFT matrix given by
n
1 1 ... 1 
k 1 ... WN N 1 

WN ˆ WNkn 0 k , n  N 1  
 :
WN1
: ... 
  N 1  N 12 
1 WN ... WN 

dft function ECR 305_L8
X  WN x function [Xk] = dft(xn,N)
% -------------------------------------------------------------------------
% Compute discrete Fourier transform
% Xk : DFT coefficient array over 0 <= k <= N-1
% xn : N-point finite duration sequence
 
% N : Length of DFT
% -------------------------------------------------------------------------
WNnk = WN.^nk; % DFT matrix
2 Xk = xn*WNnk; % row vector for DFT coefficient
j
WN  e N
% ------------------------------------------------------------------

idft function ECR 305_L8
1 * function [xn] = idft(Xk,N)

x WN X
N %=============================================
% Compute inverse discrete Fourier transform
% Xk : DFT coefficient array over 0 <= k <= N-1
% xn: N-point sequence over 0 <= n <= N-1
 
% N : Length of DFT
% -------------------------------------------------------------------------
WNnk = WN.^(-nk); % DTS matrix
2
j xn = (1/N).*(Xk*WNnk); % row vector for DFT coefficient
WN  e N
% -------------------------------------------------------------------------

MATLAB example 5.6 ECR 305_L8
Let x(n) be an N-point sequence: 1, 0n3

xn   
0, otherwise
(a) Compute the DTFT and plot its magnitude and phase.
(b) Compute the 4-point DFT of x(n).
% Definition % Plotting
Subplot(2,1,1); stem(m,magXk);
n1=0; n2=3; n=n1:n2; M = 200; k = 0:M; title('Magnitude','fontsize',15);
w=(4*pi/M)*k; N = 4; m = 0:N-1; xlabel('k','fontsize',15);
x = [1,1,1,1]; xn = [1,1,1,1]; ylabel('X(k)','fontsize',15);
grid; hold on
% DTFT plot(w/pi,magX,'--'); axis([0 4 0 5]);
X = dtft(x,n,k); magX=abs(X);
Subplot(2,1,2); stem(m,PhaXk);
angX=angle(X)*180/pi; title('Angle','fontsize',15);
xlabel('k','fontsize',15);
% DFT ylabel('Degrees','fontsize',15);
Xk = dft(xn,N); magXk = abs(Xk); grid; hold on
PhaXk = angle(Xk)*180/pi; plot(w/pi,angX,'--'); axis([0 4 -200 200]);

MATLAB example 5.6 (contd..) ECR 305_L8

MATLAB example 5.7 ECR 305_L8
How can we obtain further samples of X(ej)?
Ans: x n   1,1,1,1,0,0,0,0 <= By padding zeros


% Definition % Plotting
Subplot(2,1,1); stem(m,magXk);
n1=0; n2=7; n=n1:n2; M = 200; k = 0:M; title('Magnitude','fontsize',15);
w=(8*pi/M)*k; N = 8; m = 0:N-1; xlabel('k','fontsize',15);
x = xn = [ones(1,4),zeros(1,4)]; ylabel('X(k)','fontsize',15);
grid; hold on
% DTFT plot(w/pi,magX,'--'); axis([0 4 0 5]);
X = dtft(x,n,k); magX=abs(X);
Subplot(2,1,2); stem(m,PhaXk);
angX=angle(X)*180/pi; title('Angle','fontsize',15);
% DFT ylabel('Degrees','fontsize',15);
Xk = dft(xn,N); magXk = abs(Xk); grid; hold on
PhaXk = angle(Xk)*180/pi; plot(w/pi,angX,'--'); axis([0 4 -200 200]);


How can we obtain further samples of X(ej)?

Ans: x n   1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0


MATLAB problem 5.12 ECR 305_L8
(a) Design a MATLAB function to implement an N-point circular

folding operation x2(n) = x1((-n))N.
(b) determine the circular folding of the sequence:
x1(n) = {1,3,5,7,9,-7,-5,-3,-1}
function x2 = circfold(x1,N) % Plotting

n = 0:N-1; subplot(2,1,1)
x2 = x1(mod(-n,N)+1); stem(0:N-1,x1); title ('x_1(n)','fontsize',15);
xlabel('n','fontsize',15); ylabel('x(n)','fontsize',15);
% Function
x1 = [1,3,5,7,9,-7,-5,-3,-1]; subplot(2,1,2)
N = length(x1); stem(0:N-1,x2); title ('x_2(n)','fontsize',15);
% Circular fold of x1 xlabel('n','fontsize',15);
x2 = circfold(x1,N); ylabel('x((-n))_N','fontsize',15);

MATLAB problem 5.12 (contd..) ECR 305_L8

MATLAB example ECR 305_L8
Let x(n) = 10(0.9)n, 0  n  10. Plot x((-n))11.
% File name: 5p12b.m % Plotting
% Function x subplot(2,1,1)
n = 0:10; stem(n,real(Y))
x = 10*(0.9).^n; title ('Real{DFT[(x((-n))_1_1]}','fontsize',15);
% Circular fold x
y = circfold(x,11); subplot(2,1,2)
stem(n,imag(Y))
% DFT of y title ('Imaginary{DFT[(x((-n))_1_1]}','fontsize',15);
Y = dft(y,11); xlabel('k','fontsize',15);

MATLAB example (contd..) ECR 305_L8

MATLAB example 5.6-fft ECR 305_L8
Let x(n) be an N-point sequence: 1, 0n3

xn   
0, otherwise
(a) Compute the DTFT and plot its magnitude and phase.
(b) Compute the 4-point DFT of x(n).
% File name: ex5p6fft.m % Plotting
Subplot(2,1,1);
% Definition stem(m,magXk');
N = 4; m = 0:N-1; title('Magnitude','fontsize',15);
% Function ylabel('X(k)','fontsize',15);
xn = [1,1,1,1];
Subplot(2,1,2);
% Calculation stem(m,PhaXk');
Xk = fft(xn,N); title('Angle','fontsize',15);
magXk = abs(Xk); xlabel('k','fontsize',15);
PhaXk = angle(Xk)*180/pi; ylabel('Degrees','fontsize',15);

MATLAB example 5.6-fft (contd..) ECR 305_L8

References ECR 305_L8
1. John G. Proakis, Digital Signal Processing, Pearson, 4th

Edition, Seventh Impression, 2011. (pp. 449–562)
2. Vinay K. Ingle, and John G. Proakis, Digital Signal

Processing using MATLAB, Thomson Learning
Bookware Companion Series, 2007. (pp. 118–185)

Next class.. ECR 305_L8
Digital Filter
Structure

Ecr305 L8 SP 2017

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ecr305 L8 SP 2017

Uploaded by

Copyright:

Available Formats

Digital Signal Processing ECR 305_L8

Discrete Fourier Transform sampling & reconstruction of

& Fast Fourier Transform

In lecture-4 & 5, we have studied transform-domain

These transforms have two features in common:

From the numerical computation viewpoint (or from MATLAB's

To use MATLAB, we have to truncate sequences and then

Let us consider a periodic sequence

From Fourier analysis, we know that the periodic function can be

From the frequency-domain periodicity of the discrete-time

Therefore, the discrete Fourier series representation of a periodic

Fourier series coefficients, which are given by

IUB Dr. Abdur Razzak 5

 Analysis or a DFS equation:

 Synthesis or an inverse DFS equation:

IUB Dr. Abdur Razzak 6

In matrix form, the DFS and IDFS equations can be rewritten as

IUB Dr. Abdur Razzak 7

Find DFS representation of ~x n   ....,0,1,2,3,0,1,2,3,0,1,2,3,..... .

IUB Dr. Abdur Razzak 8

x n  by periodically repeating x(n)

which means that DFS k  represents N evenly spaced samples

The DTFT of x(n) is given by

Comparing DFS & IDFS equations, we have

The discrete Fourier series provided us a mechanism for numerically

Mathematics dictates that the sampling of the discrete-time Fourier transform

Theoretically, we can take care of this problem by defining a periodic signal

Practically, we define a new transform called the Discrete Fourier Transform

Discrete Fourier transform (DFT) of an N-point sequence:

Inverse discrete Fourier transform (IDFT) of an N-point

IUB Dr. Abdur Razzak 12

Linearity: DFTax1 n  bx2 n  a DFTx1 n  b DFTx2 n

Circular folding: If an N-point sequence is folded, then the

The DFT of a circular folding is given by

Circular shifting: An N-point DFT of a finite duration sequence

Now suppose that we shift xp(n) by k units to the right,

Note: Counter clockwise  +ve, clockwise  -ve direction.

Circular convolution: Consider 2 N-point sequences x1(n) and

Then the IDFT of X3(k), called the circular convolution, given by

IUB Dr. Abdur Razzak 15

Determine the circular shifting of the sequence by 2 units:

IUB Dr. Abdur Razzak 16

IUB Dr. Abdur Razzak 17

Determine the circular folding of the sequence:

IUB Dr. Abdur Razzak 18

Determine the circular convolution of the following two

Although DFT is a computable transform, the numerical

Therefore, several algorithms have been developed to efficiently

In 1965, Cooley and Tukey showed a procedure to substantially

Consider an N-point sequence x(n). Its N-point DFT is given by

where WNnk  e  j 2 / N. To obtain one sample of X(k), we need N complex

Hence to obtain a complete set of DFT coefficients, we need N2

Therefore, the number of DFT computation for an N-point sequence

IUB Dr. Abdur Razzak 22

Goal of an efficient computation: In an efficiently designed

The quadratic dependence of N can be reduced by realizing that

WNnk  WNk n  N   WNn k  N 

and the symmetry property WNnk  N / 2  WNnk

IUB Dr. Abdur Razzak 23

Compute a 4-point DFT of x(n) = {1, 1, 1, 1} and develop an

The above computation can be done in the matrix form

which requires 16 complex multiplications.

Putting these values into previous equation, we have