The Gabor Transform, STFT and CWT Invertibility, and

generalized Parseval s like theorem

Kartik Kariya (02307923), B RajaSekhar (02307046), C.H.Ravichandrababu (02307033)

1.1 Parseval s Theorem

Parseval s theorem states that while going from time domain to frequency domain inner
product (dot products) are preserved. In other words it states that while transforming from
one domain to other (e.g. time to frequency) energy is preserved.
In context to the Fourier transform the Parseval s theorem can be stated as
1
x(t ) y (t )dt x( w) y ( w)dw
2
s
where, x(t) x(w) and y(t) y (w) are Fourier transform Pair.

For any other transform if we prove that Parseval s like relationship exists then the
inverse transform relations also exist. In following sections, Gabor Transform, Short
Time Fourier Transform (STFT) and Wavelet Transforms are defined and it is shown that
Parseval s like relationship exists for them.

1.2 Gabor Transform:

Fourier transform (FT) tells whether a certain frequency component exists or not. This
information is independent of where in time this component appears. It is therefore very
important to know whether a signal is stationary or not, prior to processing it with the FT.
Therefore in order to analyze the non stationary waveforms the windowing methods were
used so that time localization can be done to some extent. In continuation of these efforts
the Gabor Transform uses Gaussian window. The following section gives definition of
Gabor Transform and proves its invertibility by showing that Parseval s like relationship
exists for the Gabor Transform.
The Fourier Transform of a function f (t ) L2 ( R) , namely f (w) , reveals the spectral

information of the signal. To extract information of the spectrum f (w) from a local
observation of the signal f(t), a time window is needed. Breaking the signal into segments
of period T is equivalent to multiplying it by a rectangular window. This is, in effect,
convolution in the frequency domain of the Fourier Transforms of the signal and the
window. In order to avoid aliasing effects the Fourier Transform of the window should
satisfy following properties.
1. Main lobe must be narrow.
2. Side lobes must be suppressed.
These properties are contradictory for a finite length window. In order to reduce the main
lobe width length need to be increased, but this would lead to poor time localization.
Gaussian window is optimal in this sense. More over, it does not have any side lobes
since its Fourier Transform is again Gaussian. Gaussian function is used as localizing
function.

1.3 The Gaussian Function and its Fourier Transform

The Uncertainty principle gives us the lower bound on the time frequency bandwidth
product. The principle states that for any function f(t)
2 2 1
t w
4
This inequality is reached if
dx(t )
t x(t) = (1)
dt
where, is constant
dx(t ) 1
t.dt
x(t )
Solving this differential equation we get,
1 2
ln x(t) = t +C
2
x(t)= C0 exp(t2/2 ) s
x(t) will be a finite energy signal iff is negative
=- 2
x(t)= C0 exp(-t2/2 2) for all real (2)

The solution for this equation is nothing but Gaussian function. This shows that that the
Gaussian function is optimal for its time-frequency bandwidth product, giving
localization in both time and frequency domain.

The Fourier transform of Gaussian function can be found as follows.

Let g(t) be a Gaussian function.
g (t) = C0 exp(-t2 /2 2)

dg (t ) t
- exp (-t2 /2 ) where = 2
dt

dg (t ) j
{ -j t. g(t) } (3)
dt
Taking Fourier Transform for equation (3) we will get the following result in frequency
domain.
j dg ( )
j g( ) =
d
dg ( )
=- g( )
d
dg ( )
=- d
g( )
 2 2 2
g( ) = C exp(- /2) = C exp(- /2) (4)

Thus Fourier transform of Gaussian is also a Gaussian function.

1.4 Gabor Transform Definition and its invertibility.

Gabor Transform of a function x(t ) L2 ( R) is defined by the following equation.

X(b, ) x(t) g(t - b) e -jwt dt (5)

X(b, ) localizes the Fourier Transform of the function x(t) around t=b. We can interpret
it as follows.

Let g 0(t) = g(t-b) ejwt

Then the Gabor Transform is then nothing but the inner product of x(t) and g 0(t).

X(b, ) = < x(t), g 0(t)> (6)

From Parseval s Identity, dot product of two functions in time domain is same as dot
product of same functions in its frequency domain with a scaling factor.

< x(t), g 0(t)> = 1/(2 ) < x(w) , g w0 ( w) > (7)

Here the function g( ) is real function, so its complex conjugate will also be same
function. We are interested in finding Fourier Transform of g 0(t) which is g 0( ), and it
is given as follows.

g 0( ) = g(t-b) ejw0t e-jwt dt (8)

change of variable method through which keep t-b=k

g 0( ) = g(k) exp(j 0(b+k)) exp(-j (b+k)) dk

= exp(j 0b) exp(-j b) g(k) exp(j( 0- )k) dk

= exp(j 0b) exp(-j b) g(k) exp( j( - 0)k) dk

= exp(j 0b) exp(-j b) g( - 0)

Hence for Gabor Transform the Parsval s like theorem is proved as

X(b, ) = < x(t), g(t-b) exp(j 0t) > = 1/(2 ) < x( ), exp(j 0b) exp(-j b) g( - 0)> (9)

Thus the Gabor transform is invertible. Graphically above equation can be represented as
below. As shown in below in time and frequency domain b and w interchange their roles.
 Spectrum

b t w0 w

1.5 Short Time Fourier Transform (STFT)

The Gabor Transform studied earlier is the special case of Short Time Fourier Transform
where window function is chosen as Gaussian Function.

1.6 STFT Definition and its invertibility

The STFT is given as
jwt
X STFT (b, w) x(t )v(t b)e dt

where, v is window function.

The inverse relation ship is given as

x(t ) A X STFT (b, w)v(t b)e jwt dwdb

1.6.1 Proof
The invertibility of STFT is shown by proving that Parsval s like relationship exists for
STFT.
i.e. we need to prove that following relationship exists

x(t ) y (t ) A X STFT (b, w) y STFT (b, w)dbdw

RHS of above equation can be expanded as

jwt jw
RHS A { x(t )v(t b)e dt}{ y ( )v( b)e d }dbdw

Assuming order of integration can be interchanged, integrating with w first we have

e jw( t ) dw 2 ( t)
Integrating with b
2
v(t b)v(t b)db v(t b)
Above equation is nothing but energy in function v and which is constant.
Hence
RHS x(t ) y (t )dt
t

Thus Parsval s like relationship exists for STFT and hence it is invertible with equation
given above.

1.7 Continuous Wavelet Transform (CWT):

The continuous wavelet transform is given by equation:
1 t b
W f (b, a ) f (t ) ( )dt
a a

Above is the analysis equation of continuous wavelet transform. In following section we

try to investigate whether similar synthesis equation exists. Alternately we are trying to
answer if Parsval s like theorem holds true for the wavelet transform which states that
inner product in time domain is equivalent to that of in transform domain.

Theorem
The Parsval s like identity for Wavelet Transform is given as
da
W f (b, a )W g (b, a ) 2 db C f (t )g (t )dt .1
0 a
where,
2
( w)
C dw
w
Proof:
Consider the equation
1 t b
W f (b, a ) f (t ) ( )dt
a a
using the standard Parsval s theorem,
1 t b
W f (b, a ) f ( w) ( )dw ..2
2 a a
where,
t b t b jw
( ) ( )e dw
a a
putting t b
a
t b jwb j ( aw )
( ) ae ( )e dw
a
t b
) ae jwb ( aw)
( 3
a
Using equation 2 and 3 and LHS of equation becomes
 LHS =
1 jwb 1 jwb da
( { x ( w1 ) ae ( aw1 )) dw1 }( { y ( w ) ae ( aw 2 )) dw 2 } db
a 0b
2 2 a2

1 jwb 1 jwb
= ( { x( w1 )e ( aw1 ))dw1}( { y ( w 2 )e ( aw2 ))dw2 }dadb
a 0b
2 2

Integrating with b , we have

1
e j ( w2 w1 ) b
db ( w2 w1 )
2

1 1 jwb
LHS = ( { x( w1 ) ( w2 w1 ) ( aw1 ))dw1 )( { y ( w1 )e ( aw2 ))dw2 }da
a 0
2 2

We have
2 2
d
( aw) da ( ) (obtained by putting aw = )

=C (Assuming integral converges)

Thus LHS becomes
C 1 C
LHS = x( w) y ( w) dw x(t ) y (t )
2 2 2

1.8 Admissibility Condition

Thus it is shown that Parsval s like relationship exist for WT exist subjected to condition
that
2
dw
( w)
w

Two cases of the above integral, one with limits - to 0, and other with limits 0 to + ,
are to be satisfied, but since for real (t), magnitude is symmetric on positive and
negative sides, showing any one will suffice.
2
dw
( w)
0 w
This is called the admissibility condition for mother wavelet .
This document was created with Win2PDF available at http://www.daneprairie.com.
The unregistered version of Win2PDF is for evaluation or non-commercial use only.