You are on page 1of 7

THE FRACTIONAL FOURIER TRANSFORM AND APPLICATIONS

INTRODUCTION
Fast Fourier Transform(FFT) 1algorithm computes

Discrete Fourier Transform(DFT)


Discrete Convolution
Trigonometric Interpolation

The term fast Fourier transform (FFT) refers to an efficient implementation of the
discrete Fourier transform (DFT) for highly composite transform lengths .
Algorithms to compute the discrete Fourier transform (DFT) in
better, for approximate algorithms), where
and and

O(x)

means on the order of

When computing the DFT as a set of


computational complexity is

( N 2)

O( N lg N ) time (or

is the number of discrete points

x .
inner products of length

1 https://ccrma.stanford.edu/~jos/st/Fast_Fourier_Transform_FFT.html

each, the

1. Generalization of DFT
The Fourier transform of a continuous-time signal

x (t)

may be defined as

X ( )= x (t)e jt dt , ( , )

The DFT, on the other hand, replaces the infinite integral with a finite sum:
N1

X ( k )= x (t n)e j t , k=0,1,2, , N 1,
k n

n=0

The sampling interval

is also called the sampling period.

it is common to write the DFT and its inverse in the more pure form below, obtained
by setting

T =1

in the previous definition:

N 1

X ( k ) = x ( n ) e j 2 nk / N , k =0,1,2, , N 1
n=0

1
x ( n )=
N

N 1

X (k ) e j 2 nk / N , n=0,1,2, , N 1,
n=0

where
the

x (n) denotes the input signal at time (sample)

n , and

X (k)

denotes

k th spectral sample.

2. CONVOLUTION2
Theorem: For any

, y CN

x y X Y
Proof:

This is perhaps the most important single Fourier theorem of all. It is the basis of a
large number of FFT applications. Since an FFT provides a fast Fourier transform, it
also provides fast convolution, thanks to the convolution theorem. It turns out that
using an FFT to perform convolution is really more efficient in practice only for
reasonably long convolutions, such as

100

. For much longer convolutions, the

savings become enormous compared with ``direct'' convolution. This happens


because direct convolution requires on the order of

N2

operations (multiplications

and additions), while FFT-based convolution requires on the order of


operations, where

lg N

denotes the logarithm-base-2 of

N.

2.1 Discrete Convolution


2 https://ccrma.stanford.edu/~jos/mdft/Convolution_Theorem.html

N lg N

You should be familiar with Discrete-Time Convolution, which tells us that given two

x [n] , the system's input, and

discrete-time signals

h[ n] , the system's

response, we define the output of the system as

y [ n ] =x [n]h[n]

x [k ]h [nk ],

k=

When we are given two DFTs (finite-length sequences usually of length

N , we

cannot just multiply them together as we do in the above convolution formula, often
referred to as linear convolution. Because the DFTs are periodic, they have nonzero
values for n N
for

and thus the multiplication of these two DFTs will be nonzero

n N ,. We need to define a new type of convolution operation that will result


n={0,1, , N 1 } . This

in our convolved signal being zero outside of the range

idea led to the development of circular convolution, also called cyclic or periodic
convolution.

2.2 Circular Convolution


What happens when we multiply two DFT's together, where
of

Y [k]

is the DFT

y [n ] ?

Y [ k ] =F [ k ] H [ k ]
when

0 k N 1

Using the DFT synthesis formula for

y [n ]=

1
N

N 1

F [k ] H [k ]e

y [n ]

2
kn
N

k=0

N1

And then applying the analysis formula

1
y [n ]=
N

N 1 N 1

( j)

f [m]e

m=0 m=0

2
kn
N

H [k ]e

2
kn
N

( j)

F [ k ]= f [m]e

2
kn
N

m=0

N 1

1
f [m](
N
m=0

N1

H [k ]e
k=0

2
k(nm )
N

where we can reduce the second summation found in the above equation into

h [ ( ( nm ) )N ]=

1
N

N1

k=0

H [k ]e

2
k(nm)
N

N 1

y [ n ] = f [m]h [ ( ( nm ) )N ]
m=0

which equals circular convolution! When we have


we get:

y [ n ] = y [ n ] h[n]

0 n N1

in the above, then

TRIGONOMETRIC INTERPOLATION
Periodic function:

f ( t )=f ( t +T ) , t ( , ) .

we can consider

f ( t +2 ) =f ( t )

Idea: use trigonometric functions to interpolate a periodic function


at equally distributed

n+1

points

x j=

f ( x ) , x [0,2 ]

i
2.
n+1

Trigonometric interpolation:

a
[ k cos ( kx ) +b k sin ( kx )],
a0 M
P ( x )= +
2 k=1

or
M

P ( x )=

k=M

c k e ikx ,

where

n
if n is even
M= 2
n+ 1
, if n is odd
2

use interpolation condition to determine the coefficient

x
x
f ( j )
P( j )=

ck

, we have

FRACTIONAL FOURIER TRANSFORM


Given a time signal s(t), its FrFT is de_ned as
S_(u) = F_fs(t)g =
Z1
1

K(_; u; t) s(t) dt; (24)


where the kernel function is
K(_; u; t) =
8<
:
1i cot _
2_

_1=2 expfi cot _


2 (u2 + t2)g expfi u
sin _tg; _ 6= n_;
_(t u); _ = n2_;
_(t + u); _ + _ = n2_;
(25)
and where _ is called the order of the transform, which can also be interpreted as an
angle
of rotation in the time-frequency plane.
For _ = 0 (i.e for n = 0 in Equation 25), we have S0(u) = s(u) which is the timedomain signal itself. For _ = _=2, we have cot _ = 0 and sin _ = 1, and S_=2(u)
reduces
to the classic Fourier transform S(u) = S(f) of s(t). It is insightful to re-interpret the
classic FT as a rotation of the time signal through an angle of _=2 in the timefrequency
plane. A rotation through some other angle gives a generalised or `fractional
frequency'
spectrum of the signal at that angle, with the fractional frequency denoted by u1.

You might also like