DFT and FFT

Discrete Fourier
Transform
Let x(t) be a periodic function of time having a time period T0, then
the fundamental frequency of x(t) is
2
0
= T0
2
0
= T0
0 =
2T0
The signal x(t) may be expressed in terms of the Fourier series as
0
x(t) = a0 +nn= (0 a n cosn
t + b sin
n t)
x(t) = a +
0
n t)
(
a
n
n=
1
cosn t + b sin
0
x(t) = a +
0
(
a
n
n=
1
cosn t + b sin
0
n t) 1 2
T0
1
a0 =
x(t)d(
0t) = 0x(t)dt
0
0
2
T
where
the average value
x(t) = a +
0
(
a
n
n=
1
cosn t + b sin
0
n t) 1 2
T0
1
a0 =
x(t)d(
0t) = 0x(t)dt
0
0
2
T
T0
2
1
an = 2 0 x(t)cosn0td(0 t) =0 0
T
cosn0tdt
where
the average value
x(t)
for n = 1, 2, 3,
x(t) = a +
0
(
a
n
n=
1
cosn t + b sin
0
n t) 1 2
T0
1
the average value
a0 =
x(t)d(
0t) = 0x(t)dt
0
0
2
T
T0
2
1
an = 2 0 x(t)cosn0td(0 t) =0 0 x(t)
T for n = 1, 2, 3,
cosn0tdt
T0
2
1
bn = 2 0 x(t)sinn0td(0 t) =
x(t)
0
0
T for n = 1, 2, 3,

sinn0tdt
where
x(t) = a +
0
(
a
n
n=
1
cosn t + b sin
0
n t) 1 2
T0
1
the average value
a0 =
x(t)d(
0t) = 0x(t)dt
0
0
2
T
T0
2
1
an = 2 0 x(t)cosn0td(0 t) =0 0 x(t)
T for n = 1, 2, 3,
cosn0tdt
T0
2
1
bn = 2 0 0 x(t)sinn0td(0 t) =
x(t)
0
T for n = 1, 2, 3,

sinn0tdt
a s are known as
cosine coefficients and b s are known as sine coefficients.
where
Relation (2) may be rewritten as
an
x(t) = a0 a +b ( 2
n
n
an
n=1
+
+b
2
cosn0 t
2
n
(b
)a
+b
sinn0
t)
an
x(t) = a0 a +b ( 2
n
n
an
n=1
+
+b
cosn0 t
2
n
(b
)a
+b
sinn0
t)
or
x(t) = C0 + n=1 Cn cos(n0t +n

)
[cos(A+ B) = cosAcosB sin

AsinB]
n
an
-bn
an
x(t) = a0 a +b ( 2
n
n
an
n=1
+
+b
(b
)a
cosn0 t
2
n
+b
sinn0
t)
or

)

AsinB]
where C =
C= a
n =
2
tan
+b
0 0
n n
n
an
-bn
an
x(t) = a0 a +b ( 2
n
n
an
n=1
+
+b
(b
)a
cosn0 t
2
n
+b
sinn0
t)
or

)

AsinB]
where C =
C= a
n =
2
0 0
+b
tan
n n
n
an
-bn
Cn, n = 1, 2, 3, is the amplitude and n, n = 1, 2, 3, is the phase

of the nth harmonic. C0 is the average value.
Expressing cosine and sine terms of relation (2) in terms of their complex
exponential values as
e
x(t) = a0 + [an
(
n=1
jn0t
jn0t
jn0t
+e
) +ebn
2 2j (
jn0t
)]
jn0t
x(t) = a0 + n= [an
1
(
or
x(t) = a + n=1 [e
)]
jn0t
jn0t
jn0t
+e
) +ebn
2 2j (
jn0t
)]
an jbn
jn0t an + jbn
( 2 2 )+e
(
jn0t
x(t) = a0 + n= [an
1
(
or
or
+ [F
e
x(t) = a + n=1 [e
)]
x(t) = F0
jn0t
n=1
jn0t
+e
) +ebn
2 2j (
jn0t
)]
an jbn
jn0t an + jbn
( 2 2 )+e
(
jn0t
jn0t
+ Fn
e
jn0t
jn0t
x(t) = a0 + n= [an
1
(
or
or
+ [F
e
x(t) = a + n=1 [e
)]
x(t) = F0
jn0t
F0 = a0
,
jn0t
+e
) +ebn
2 2j (
jn0t
)]
an jbn
jn0t an + jbn
( 2 2 )+e
(
jn0t
n=1
where
jn0t
an
Fn =
jbn 2
(
+ Fn
e
jn0t
an +
) and Fn = (
jbn 2
Now
x(t) = F0 + [Fn
n=1 e
where
F0 = a0
,
Here F
an
Fn =
jbn 2
(
= F n , conjugate of Fn.
jn0t
+ Fjnn0t
e
an +
) and Fn = (
jbn 2
Now
x(t) = F0 + [Fn e jn+0tFnejn0t
n=1
F0 = a0
where
,
Here F
an
Fn =
jbn 2
(
an +
) and Fn = (
jbn 2
= F n , conjugate of Fn.
Relation (4) may be expressed as
x(t) = F0 + n=1 Fn
jn0
t
+e F
n=1
jn0
t
Now
x(t) = F 0 +
Fe
n
n=1
t
0
jn
n=1
et
0
jn
Now
x(t) = F 0 +
Fe
n
n=1
t
0
x(t ) = F0 + n F n
=1
e
jn
jn
t
0
F
+ F
e
n=1
n=
1
et
jn
jn
t
0
Now
x(t) = F 0 +
Fe
n
n=1
t
0
x(t ) = F0 + n F n
=1
e
jn
jn
t
0
F
+ F
e
n=1
n=
1
et
jn
jn
t
0
Now
x(t) = F 0 +
Fe
n
n=1
t
0
x(t ) = F0 + n F n
=1
e
jn
jn
t
0
F
+ F
e
n=1
n=
1
et
jn
jn
t
0
Now
x(t) = F 0 +
Fe
n
jn
n=1
t
0
x(t ) = F0 + n F n
=1
e
jn
t
0
F
+ F
e
n=1
n=
1
jn 0
t
Hence, we can write,
x(t ) =
F e
n = n
et
jn
jn
t
0
Now
x(t) = F 0 +
Fe
n
jn
n=1
t
0
x(t ) = F0 + n F n
=1
e
jn
t
0
F
+ F
e
n=1
n=
1
jn 0
t
x(t ) =
et
jn
jn
t
0
F e
n = n
Thus x(t) may be expressed in terms of Complex Fourier Series in relation

(5). Here Fn is known as the Complex Fourier coefficient.
Now
x(t) = F 0 +
Fe
n
jn
n=1
t
0
x(t ) = F0 + n F n
=1
e
jn
t
0
F
+ F
e
n=1
n=
1
jn 0
t
x(t ) =
et
jn
F e
n = n
Variation of Fn coefficients with n
jn
t
0
Now
x(t) = F 0 +
Fe
n
jn
n=1
t
0
x(t ) = F0 + n F n
=1
e
jn
t
0
F
+ F
e
n=1
n=
1
jn 0
t
x(t ) =
et
jn
jn
t
0
F e
n = n
Variation of Fn coefficients with n
= Fn
The amplitudes Cns of relation (3) may be related to Fns as
C=F
00
, the average value
C=F
00
and
, the average value
C =2F
, for n = 1, 2, 3,
n n
the amplitude of the nth harmonic.
C=F
00
and
, the average value
C =2F
, for n = 1, 2, 3,
n n
the amplitude of the nth harmonic.
and
j(F n nF
n = tan Fn +n
) ( )
the phase ofF
the nth harmonic.
1
From relation (4), Fn may be expressed as
an jbn
F =( 2
)
n
an jbn
F =( 2
)
Substituting
expressions of an and bn from relation (2)
n
1
Fn =
T0
T0
jn
x(t)e
dt
t
an jbn
F =( 2
)
Substituting
n
1
Fn =
T0
T0
jn
x(t)e
dt
t
For aperiodic signals, the time period T0 becomes infinite, and the Fourier
transform of an aperiodic signal x(t) is defined as
X ( ) =
dt
x(t )e jt
an jbn
F =( 2
)
Substituting
n
1
Fn =
T0
T0
jn
x(t)e
dt
t
For aperiodic signals, the time period T0 becomes infinite, and the Fourier
transform of an aperiodic signal x(t) is defined as
X ( ) = x(t )e jt
dt
Reference: Network Analysis by

Valkenburg
Let xk be a periodic discrete sequence obtained from a periodic signal x(t)

with a time period T0.

Let N number of samples be available in the time period T0 with a sampling
interval .
T 0 = N

Let N number of samples be available in the time period T0 with a sampling
interval .
T 0 = N
Using rectangular rule for integration, the Fourier coefficients may be obtained as
From relation (2)
1
a0 =
T0
0 0
)dt T
x(t
From relation (2)
1
a0 =
1
a0 =
N 1
0 k =0 k
T0
0 0
)dt T
x(t
From relation (2)
a0 =
1
T0
1
a0 =
or
a
0
N 1
0 k =0 k
N 1
Nk =0 k
T0
dt
0
x(t )
From relation (2)
a0 =
1
T0
1
a0 =
or
0 k =0 k
N 1
k =0
x
k
dt
0
N 1
T0
1
x =
k
N
1
k
=0
x(t )
Similarly,
From relation (2)
2
an =
T0
x(t ) cos
n0tdt T
0 0
Similarly,
From relation (2)
2
an =
an
=
N
1
2
N
x
k =0
T0
x(t ) cos
n0tdt T
0 0
cos n
(k )
N
(using the substitutions: 0 =
2 2
=
T0 N
and T0 =
and
t = k
in relation (2))
Similarly,
From relation (2)
2
an =
an
=
N
1
2
N
x
k =0
T0
x(t ) cos
n0tdt T
0 0
cos n
(k )
N
a n = 2 N 1 kx cos

N k=
N
2kn0
0 = 2 =2
(using the substitutions:
or
T0 N
and T0 =
and
t = k
in relation (2))
and
From relation (2)
2
bn =
T0
x(t ) sin
n0tdt T
0 0
and
From relation (2)
2
bn =
N 1
bn
=
2
N
x
k =0
T0
x(t ) sin
n0tdt T
0 0
2
sin n
(k )
N
(using the substitutions: 0 =
2 2
=
T0 N
and T0 =
and
t = k
in relation (2))
and
From relation (2)
2
bn =
N 1
bn
=
2
N
x
k =0
T0
x(t ) sin
n0tdt T
0 0
2
sin n
(k )
N
bn = 2 N 1k x sin
N k =0
N
2kn
2 2
0 =
=
(using the substitutions:
or
T0 N
and T0 =
and
t = k
in relation (2))
Now from relations (11) and

(12),
an
jbn 2
2kn
2kn
=
N k = 0 xk cos N j sin N
N 1

(12),
an
jbn 2
2kn
2kn
=
N jn 2k
1 1
= xk e N
N
N 1
k =0

(12),
an
jbn 2
2kn
2kn
=
N jn 2k
1 1
= xk e N
N
N 1
k =0
N
or
2
(a
jb
x e
jn 2k
N
N
1
)k k=
=0

(12),
an
jbn 2
2kn
2kn
=
N jn 2k
1 1
= xk e N
N
N 1
k =0
N
or
2
(a
Substituting
jb
)k k=
x e
Na0 = X
N1 0
jn N
Xn = k=0 xk e
2k
jn 2k
N
N
1
=0
and
N
(a jb ) = X , for n = 1, 2, 3,
n
n
2 n

for n = 0, 1, 2,
N1
From relation (13)
jn N
Xn = k=0 xk e
2k
for n = 0, 1, 2,
N1
From relation (13)
jn N
Xn = k=0 xk e
2k
for n = 0, 1, 2,
Now, for n N, let n = N + m, for m = 0, 1, 2,
N 1
j (N +m ) 2k
N

Xn =
xk
e
k
=0
jn N
2k
N1
From relation (13)
Xn = k=0 xk e
for n = 0, 1, 2,
N 1
j (N +m ) 2k
N

Xn =
xk
e
k
=0
N1
or
m+N
j(2k
= xk e .e
)
k=0
jm N
2k
jn N
2k
N1
From relation (13)
Xn = k=0 xk e
for n = 0, 1, 2,
N 1
j (N +m ) 2k
N

Xn =
xk
e
k
=0
N1
or
m+N
jm N
2k
j(2k
= xk e .e
)
k=0
or
2
jm k
N
N
1
m+N
e
=
k=0
xk
= Xm
jn N
2k
N1
From relation (13)
Xn = k=0 xk e
for n = 0, 1, 2,
N 1
j (N +m ) 2k
N

Xn =
xk
e
k
=0
N1
or
m+N
jm N
2k
j(2k
= xk e .e
)
k=0
or
N
1
2
jm k
N
= Xm
e is periodic with a period N.

Therefore X
m+N
= k=0 xk
n
Then, within one period (i.e. for n = 0, 1, 2, , N-1),

N 1
X n = xk
e
k
=0
jn
2k

N
, for n = 0, 1, 2, ., N-1

N 1
X n = k =0 xk
e
jn
2k

N
, for n = 0, 1, 2, ., N-1
Relation (15) is known as the Discrete Fourier Transform (DFT) of a finite

sequence xk, k = 0, 1, 2, , N-1.

N 1
X n = k =0 xk
e
jn
2k

N
, for n = 0, 1, 2, ., N-1

sequence xk, k = 0, 1, 2, , N-1.
Xn is the discrete Fourier coefficient.

N 1
X n = k =0 xk
e
jn
2k

N
, for n = 0, 1, 2, ., N-1

sequence xk, k = 0, 1, 2, , N-1.
Relation (14) represents the periodicity property of DFT. Xn repeats at the Nth
harmonic.

N 1
X n = k =0 xk
e
jn
2k

N
, for n = 0, 1, 2, ., N-1

sequence xk, k = 0, 1, 2, , N-1.
harmonic.
The frequency corresponding to the Nth harmonic
N
1
=
= =f
T0 N
N
, the sampling
frequency.

N 1
X n = k =0 xk
e
jn
2k

N
, for n = 0, 1, 2, ., N-1

sequence xk, k = 0, 1, 2, , N-1.
harmonic.
The frequency corresponding to the Nth harmonic
N
1
=
=
=
T0 N
N
f
, the sampling
frequency.
Thus Xn repeats at the sampling frequency fs.
The Discrete Fourier Transform (DFT) of a finite sequence xk, k = 0, 1, 2, , N1 is defined as

N 1
X n = xk
e
k
=0
jn
2k

N
, for n = 0, 1, 2, ., N-1

N 1
X n = xk
e
k
=0
jn
2k

N
, for n = 0, 1, 2, ., N-1
Amplitude Cn (c.f. relation (3)) is related to Xn as
1
C =N
0
, the average value

N 1
jn
X n = xk
e
k
=0
2k

N
, for n = 0, 1, 2, ., N-1
Amplitude Cn (c.f. relation (3)) is related to Xn as
1
C =N
0
and
X2
Cn
Xn
N
=
, the average value
, for n = 1, 2, 3,
Relation (15):
By multiplying
1
e
N
N
1
k
k
=0
jn 2l
N
Xn = x
2k
jn
, for n = 0, 1, 2, ., N-1
on both sides of relation (15) and summing up from n = 0 to N-1 with 0 l N

N 1
1
Xn
=
N n0
e
2l
jn N
N 1 N
1
jn 2Nk
1
= N n = 0 k xk
=0
e
1 N11 N
= N n =0 kx=0 e k
.e
2l
jn N
jn (l k )
2
N
By multiplying
1
e
N
N
1
Relation
(15):
n
k
k
=0
jn 2l
N
X
e
2k
jn
, for n = 0, 1, 2, ., N-1

N 1
1
Xn
=
N n0
e
2l
jn N
N 1 N
1
jn 2Nk
1
= N n = 0 k xk
=0
e
1 N11 N
= N n =0 kx=0 e k
Now, changing the order of

summation,
1
N
N 1
n=0
X
n
e
2l
jn N

N
1
k =0
k 1
N
N
1
n=0
jn
.e
2l
jn N
jn (l k )
2
N
2 (l k )
N
By multiplying
1
e
N
N
1
Relation
(15):
n
k
k
=0
jn 2l
N
X
e
2k
jn
, for n = 0, 1, 2, ., N-1

N 1
1
Xn
=
N n0
e
2l
jn N
N 1 N
1
jn 2Nk
1
= N n = 0 k xk
=0
e
1 N11 N
= N n =0 kx=0 e k
Now, changing the order of

summation,
1
N
N 1
n=0
X
n
e
2l
jn N

N
1
k =0
k 1
N
N
1
n=0
jn
.e
2l
jn N
jn (l k )
2
N
2 (l k )
N
N 1
jn
)
2 (l k
, when
N
(l k) =
Now, in n=0
pN
e
where p is a positive integer, the expression
becomes
N 1
e
n=0
jn 2 p
N 1
jn
)
2 (l k
, when
N
(l k) =
N 1
Now, in n=0
pN
e
becomes
N 1 N 1
As np is another integer, it
becomes
n=0 en=0
j 2 (np )
e
pn =
0
jn 2
= 1 =
N 1
jn
)
2 (l k
, when
N
(l k) =
N 1
Now, in n=0
pN
e
becomes
N 1 N 1
As np is another integer, it
becomes
n=0 e
j 2 (np )
n=0
e
pn =
0
jn 2
= 1 =
In the present case, as l and k are limited

N within 0 and (N 1), the possible
value of p is zero, i.e. when (l - k) = 0 or l = k, the summation becomes N.
N 1
Now, in n=0
jn
)
2 (l k
, let
N
2 (l k
=
) N
N 1
Now, in n=0
jn
)
2 (l k
, let
N
2 (l k
=
) N
N 1
jn
2 (l k
Then the summation becomes
n=0 e
N 1
=
N
e
jn
n=0
N 1
jn
)
2 (l k
Now, in n=0
, let
N
2 (l k
=
) N
N 1
jn
2 (l k
n=0 e
It may be expressed as
N1 N
jn
n=0 m=1
(m1)
=e
, where m = n + 1
N 1
=
N
e
jn
n=0
N 1
jn
)
2 (l k
, let
N
Now, in n=0
2 (l k
=
) N
N 1
jn
2 (l k
n=0 e
It may be expressed as
N1 N
e =e
( )
jn
n=0
m1
, where m = n + 1
m=1
= em=1 .e
j
jm
N 1
=
N
e
jn
n=0
N1 N
or
jn
n=0 m=1
j
= e
jm
.e
= e (cosm + j sinm
j m=1
N1 N
or
jn
n=0 m=1
j
= e
jm
.e
= e (cosm + j sinm
j m=1
N
N
sin
) sin
2
2
N
+1
N +1
j
cos 2 +
sin 2
=e
sin
j
2
sin2
N1 N
or
jn
n=0 m=1
j
= e
jm
.e
= e (cosm + j sinm
j m=1
N
N
sin
) sin
2
2
N
+1
N +1
j
cos 2 +
sin 2
=e
sin
j
2
sin2
sin
N +1
j 2
2
e
j
=e
sin2
or
N
1
e
n=
0
jn
sin 2 j N2+1
= e
sine2
N
sin
N 1
j
2
e 2
=
sin 2
or
N
1
e
n=
0
jn
sin 2 j N2+1
= e
sine2
N
sin
N 1
j
2
e 2
=
sin 2
jN
2
e e
jN
2
j
j
2 e 2
jN
2j
e
2j
2
j
2
or
N
1
e
n=
0
jn
sin 2 j N2+1
= e
sine2
N
sin
N 1
j
2
e 2
=
sin 2
jN
2
e e
jN
2
j
2 e 2
jN
2j
e
2
j
e
2j
jN
ej
1
1
N
1
or
1
= e j 1
jN
jn
n=
0
Putting the value of

,
N1
jn2
n=0
(lk
N
j 2 (lk )
= e (lk
j 2
e ) 1N
1
N
1
or
1
= e j 1
jN
jn
n=
0

,
N1
jn2
n=0
(lk
N
j 2 (lk )
= e (lk
j 2
e ) 1N
1
Now for l k, the summation is zero.
N
1
or
1
= e j 1
jN
jn
n=
0

,
N1
jn2
n=0
(lk
N
j 2 (lk )
= e (lk
j 2
e ) 1N
1
Now for l k, the summation is zero.
0
form.
0
And for l = k, it becomes indeterminate
N1
jn2
k )
Thus,
n=0 e
considering
(l
N
= N, forl =
k
0 l, k N
N1
jn2
k )
Thus,
n=0 e
considering
(l
N
= N, forl =
k
= 0,forl
k
0 l, k N
N1
jn2
k )
Thus,
n=0 e
(l
N
= N, forl =
k
= 0,forl
k
0 l, k N
considering
Thus all terms on the right hand side of relation (17) vanishes except
when l = k.
1
N
N
1
Xn
e
n=0
jn 2l
N
N 1
1
= k =0 xk N

N
1
n=0
jn
2 (l k )
N
N1
jn2
k )
Thus,
(l
n=0 e
= N, forl =
k
= 0,forl
k
0 l, k N
considering
Thus all terms on the right hand side of relation (17) vanishes except
when l = k.
1
N
Therefore,
1
N
N 1
2l
jn N
Xn
e n=0
N 1
n =0 Xe
jn 2l
N 1
k =0
N 1
N e
n =0
N

for l = 0,1,2,,N= xl N = xl ,

1
jn
2 (l k )
N
1
N
N 1
n=0
2l
jn N
Xn
N

for l = 0,1,2,,N= xl N = xl ,

1
1
N
N 1
2l
jn N
Xn
n=0
N

for l = 0,1,2,,N= xl N = xl ,

1
Now, changing the suffix l to

k,
N
2 k
1 1
xk =
X
N n=0 e
jn N
, for k = 0,1,2,,N1
1
N
N 1
2l
jn N
Xn
N

for l = 0,1,2,,N-1
= xl N = xl ,

n=0

k,
N
2 k
1 1
xk =
X
N n=0 e
jn N
, for k = 0,1,2,,N1
Relation (18) is known as the Inverse Discrete Fourier Transform

(IDFT).
1
N
N 1
2l
jn N
Xn
n=0
N

for l = 0,1,2,,N= xl N = xl ,

1

k,
N
2 k
1 1
xk =
X
N n=0 e
jn N
, for k = 0,1,2,,N1
Relation (18) is known as the Inverse Discrete Fourier Transform

(IDFT).
Relations (15) and (18) are called N-point DFT pair.
N-point DFT:
N 1
jn 2Nk
X n=
k = 0xk e
for n = 0,1,2,, N-1
N-point IDFT:
1 N 1
xk =
Xn
N n=0 e
2k
jn N
, for k = 0,1,2,,N-1
N-point DFT:
N 1
jn 2Nk
X n=
k = 0xk e
for n = 0,1,2,, N-1
N-point IDFT:
1 N 1
xk =
Xn
N n=0 e
2k
jn N
, for k = 0,1,2,,N-1
j 2

N

Replacing the expression

e takes the form
by the term WN, the DFT pair
N 1
xW
x
1 X
=N
W
X n = k =0
k
k
nk
N
1
n=0 n
for n = 0,1,2,, N-1
nk
for k = 0,1,2,,N-1
N
1
xW
x
1 X
=N
W
X n = k =0
k
k
nk
N
1
n=0 n
for n = 0,1,2,, N-1
nk
for k = 0,1,2,,N-1
j 2
N

WN =
e
a complex operator (twiddle factor), which rotates any vector through
where
(2
Radians.
)
N
N
1
k
N
k =0
1 X
=N
W
X =
nk
x W , for n = 0,1,2,, N-
N
1
1
xk
n=0 n
nk
for k = 0,1,2,,N-1
j 2
N

WN =
e
a complex operator (twiddle factor), which rotates any vector through
where
Radians.
)
N
2 n =
Here,
number.
harmonic number and k = sample
Periodicity
Periodicity
From relation (19),
N 1
=
X n = k =0 xkWN , for n = 0,1,2,, N-1, where WN
e
nk
2
j
Periodicity
From relation (19),
N 1
nk
X n = k =0 xkWN , for n = 0,1,2,, N-1, where

Then,
N 1
n + pN
(n +
= x
W
=
N
k=0
=
n
x kW
N 1
k=0
pN )k
k
N
=
WN e
j 2

N
for p = 0 , 1, 2 ,
.
nk
, as
W
pNk
N
= W NN ( pk )
Periodicity
From relation (19),
N 1
nk

Then,
N 1
n + pN
= x
W
=
N
k=0
n + pN
pN )k
x kW
N 1
k=0
i.e.
(n +
k
N
=
WN e
j 2

N
for p = 0 , 1, 2 ,
.
nk
, as
W
pNk
N
= Xn n for p = 0, 1, 2, .
= W NN ( pk )
Periodicity
From relation (19),
N 1
nk

Then,
N 1
n + pN
= x
W
=
N
k=0
n + pN
pN )k
x kW
N 1
k=0
i.e.
(n +
k
N
=
WN e
j 2

N
for p = 0 , 1, 2 ,
.
nk
, as
W
pNk
N
= Xn n for p = 0, 1, 2, .
= W NN ( pk )
Thus Xn is periodic with a period N, i.e. the pNth harmonic or at the p times
sampling frequency, the DFT repeats.
Symmetry
Symmetry
From relation (19),
N 1
=
X n = k =0 xkWN , for n = 0,1,2,, N-1, where WN
e
nk
2
j
Symmetry
From relation (19),
N 1
nk

Then,
X pN n
=
=
WN e
N 1
k
k= k
N01
xW
( pN n )
= xk W nk , as W
k=0
N
= Xn
for p = 0,1,2, .
NpNk N
=W
N ( pk )
j 2

N
Symmetry
From relation (19),
N 1
nk

Then,
X pN n
=
=
WN e
j 2

N
N 1
k
k= k
N01
xW
( pN n )
= xk W nk , as W
k=
N
0
for p = 0,1,2, .
NpNk N
=W
N ( pk )
= X n , conjugate of Xn, if xk is a real sequence.
Symmetry
From relation (19),
N 1
nk

Then,
X pN n
=
=
WN e
j 2

N
N 1
k
k= k
N01
xW
for p = 0,1,2, .
( pN n )
= xk W nk , as W
k=
N
0
NpNk N
=W
N ( pk )

Thus,
pN n
=
X
, for
.
n
p = 0,1,2,
Symmetry
From relation (19),
N 1
nk

Then,
X pN n
=
=
WN e
j 2

N
N 1
k
k= k
N01
xW
for p = 0,1,2, .
( pN n )
= xk W nk , as W
k=
N
0
NpNk N
=W
N ( pk )

Thus,
pN n
For p = 0, X
n
=
X
, for
.
n
=Xn
p = 0,1,2,
Symmetry
From relation (19),
N 1
nk

Then,
X pN n
=
=
WN e
j 2

N
N 1
k
k= k
N01
xW
for p = 0,1,2, .
( pN n )
= xk W nk , as W
NpNk N
=W
k=
N
0
N ( pk )

Thus,
pN n
For p = 0, X
n
=
X
, for
.
p = 0,1,2,
= X n and for p = 1, X N = X n
n
From relation (19),

N 1
for n = 0,1,2,, N-1, where WN =

X n = k =0 xkWN ,
e
nk
2
j
From relation (19),

N 1
for n = 0,1,2,, N-1, where WN =

X n = k =0 xkWN ,
e
nk
2
j
It may be represented in matrix form as
[X ] = [W nk ][Xk
]
nN
where
nk [Xn] and [Xk
are
column matrices
N1
and matrix.
[W is an NN] square
N
[X ] = [W nk ][Xk
]
nN
[X ] = [W nk ][Xk
]
X
nN
Here,
0

[Xn ]= X1

XN1
x0

[Xkx]1=

x
N1
[X ] = [W nk ][Xk
]
X
nN
Here,
and
X1
[Xn ]=

XN1
N0
W
0
[W ]
nk
N
x0

[Xkx]1=

x
N1
0N
1
N2
W
WN
WN
W
W
W
N2 (N
(N 1)
N0
(N 1)
N
1)
W0 N
=
N
W
W 0
N
(N 1)(N 1)
WN
[X ] = [W nk ][Xk
]
nN
[X ] = [W nk ][X
For N] = 4, relation (23) becomes
nN
W0 4 W0 4
0
W0
1
2
W
W
4
4
X
1 4
2 =
40
4 2 4
W
W
W
W
X 4 4 0
3
6
W
W
X3
4
4

(Frequency)W
4
x
W4 0
3
W4 x1

46 2
W x
W 9 x3
4
(Time)
0
[X ] = [W nk ][X
nN
W0 4 W0 4
0
W0
1
2
W
W
4
4
X
1 4
2 =
40
4 2 4
W
W
W
W
X 4 4 0
3
6
W
W
X3
4
4

(Frequency)W
(Time)
4
x
W4 0
3
W4 x1

46 2
W x
W 9 x3
Hence, computation of X0 requires 4 complex multiplications and 4 complex

additions.
[X ] = [W nk ][X
nN
W0 4 W0 4
0
W0
1
2
W
W
4
4
X
1 4
2 =
40
4 2 4
W
W
W
W
X 4 4 0
3
6
W
W
X3
4
4

(Frequency)W
(Time)
4
x
W4 0
3
W4 x1

46 2
W x
W 9 x3

additions.
In general, execution of relation (23) requires N2 complex multiplications
and N2 complex additions.
[X ] = [W nk ][X
nN
W0 4 W0 4
0
W0
1
2
W
W
4
4
X
1 4
2 =
40
4 2 4
W
W
W
W
X 4 4 0
3
6
W
W
X3
4
4

(Frequency)W
(Time)
4
x
W4 0
3
W4 x1

46 2
W x
W 9 x3

additions.
and N2 complex additions. Thus computational load increases rapidly with
increasing N.
[X ] = [W nk ][X
nN
W0 4 W0 4
0
W0
1
2
W
W
4
4
X
1 4
2 =
40
4 2 4
W
W
W
W
X 4 4 0
3
6
W
W
X3
4
4

(Frequency)W
(Time)
4
x
W4 0
3
W4 x1

46 2
W x
W 9 x3

additions.
and N2 complex additions. Thus computational load increases rapidly with
increasing N. Fast Fourier Transform (FFT) algorithms allow computation
of DFT with reduced computational burden.
From relation (19),

N 1
X n =k
=0
x k WN
nk
, for n = 0,1,2,, N-1
From relation (19),

N 1
nk
, for n = 0,1,2,, N-1
X n = k =0
Assuming
xkWN N to be a power of 2, N-point data sequence xk in relation (19)
may be split into two N/2 point data sequences as follows:
From relation (19),

N 1
nk
, for n = 0,1,2,, N-1
X n = k =0
Assuming
N
X = k =0 x W
n
nk
k
N
N
1
xW
N
knk=
2
k
N
nk
From relation (19),
0,1,2,, N-1
, for n =
N 1
X n = k =0
Assuming
N
X = k =0 x W
n
nk
k
N
N
1
xW
N
knk=
2
21
21
= k=0 xkNWnk +
k=0
k
N
N
x k+W
N
2
N
n k+
nk
From relation (19),
0,1,2,, N-1
, for n =
N 1
X n = k =0
Assuming
N
X = k =0 x W
n
nk
k
N
N
1
xW
N
knk=
2
21
21
= k=0 xkNWnk +
N
2
1
k=0
N
x k+W
N
k
N
nk
= k=0 xk=0
W
+
W
N
k
N
N
n k+
2
N
nN
2 1
N
nk
x k+W
or
N
1
2
nk
nN
2 1
2
k
N
N
=
x
W
+
W
k=0
k=0
n
nk
N
x k+W
N
2
or
N
1
2
nk
k
=
x
W +
k=0
N
n
W nN jn 2
Now,
N
W
=e
2
N
2 1
Then,
N 2
nN
2 1
N2
k=0
nk
N
x k+W
N
2
=
=
e
jn
(1)n
Xn =
x kW nkN + N(1N) xk + W nk
k =0
2
or
N
1
2
nk
k
=
x
W +
k=0
N
n
W nN jn 2
Now,
N
W
=e
2
N
2 1
Then,
or
N 2
nN
2 1
N2
k=0
nk
N
x k+W
N
2
=
=
e
jn
(1)n
Xn =
x kW nkN + N(1N) xk + W nk
k =0
2
N
2 1
k
N
n +N
X =
x + (1) k x
k =0
nk
W
2
n
or
N
2 1
k
N
n +N
X =
x + (1) k x
k =0
nk
W
2
n
N
2 1
k
N
n +N
X =
x + (1) k x
k =0
nk
W
and odd harmonics,
Now, splitting (or decimating) Xn into2 even
or
N
2 1
k
N
n +N
X =
x + (1) k x
k =0
nk
W
and odd harmonics,
or
for even harmonics, n = 2p, for p = 0,1,2,,(N/2-1) and
N
2 1
k
N
n +N
X =
x + (1) k x
k =0
nk
W
and odd harmonics,
or
for odd harmonics, n = 2p+1, for p = 0,1,2,,(N/2-1).
N
2 1
k
N
n +N
X =
x + (1) k x
k =0
nk
W
and odd harmonics,
or

for odd harmonics, n = 2p+1, for p = 0,1,2,,(N/2-1). For
even harmonics,
N
2
1
X = x
k =0
pk
2p
+ xk
+N
N
2
W
N
2 1
k
N
n +N
X =
x + (1) k x
k =0
nk
W
and odd harmonics,
or

for odd harmonics, n = 2p+1, for p = 0,1,2,,(N/2-1). For even
harmonics,
N
2
1
2p
X = x
k =0
pk
N
2
+ xk + N W
N
2 1
= x + k+
x N
k =0
2 2
pk
N
2 pk
as WN
pk
= WN2

X = k =0 xk +
2p
x
N
2 1
Now
k
+
pk
W
N
N
2 2

pk
W
X = k =0 xk + k N N
2p
+
2 2
x
gk =k x+ x, for k = 0,1,2,(N/2-1)
N
2 1
Now
Let,
k+
N
2

pk
W
Now, X
N
=
x
+
k
k
=0
2p
k + 2 N2
x
g
=
x+ x, for k = 0,1,2,(N/2-1)
Let,
k
k
N
2 1
N
k+ 2
N
2 1
Then,
X 2 p=
pk
g WN 2
k =0 k

pk
W
Now, X
N
=
x
+
k
k
=0
2p
k + 2 N2
x
g
=
x+ x, for k = 0,1,2,(N/2-1)
Let,
k
k
N
2 1
N
k+ 2
N
2 1
Then,
X 2 p=
pk
g WN 2
k =0 k
This is an N/2 point DFT sequence gk, k = 0,1,2,.(N/2 - 1)
Now, for odd harmonics [c.f. relation

(24)],
X 2 p+1= x k
(2
k +N
x
NW
p+1
)
k
2
k =0
N
2 1

(24)],
X 2 p+1= x k
(2
k +N
x
NW
p+1
)
k
2
k =0
N
2 1
N
2
1
= xk
k =0
2 pk
k +xN
WNkW

(24)],
X 2 p+1= x k
(2
k +N
x
NW
p+1
)
k
2
k =0
N
2 1
N
2
1
= xk
k =0
2 pk
N
2
1
k
k
= =0 x
k +xN
+N
WNkW
k
pk
2
WW
as
W 2 pk= W
N N
pk
2

(24)],
X 2 p+1= x k
(2
k +N
x
NW
p+1
)
k
2
k =0
N
2 1
N
2
1
Let,
= xk k +xN N WNkW
k =0
2 pk
N
2
2
1
2 pk
as
k
pk
W
=W
k
N
N
+N
k
N N
2
= k=0 x x W W

/
k
N
2 for k = 0,1,2,,(N/2-1)
gk =k x x k +W
pk
2
N
2 1
Then,
X 2 p=+1
k =gN W
pk
2
N
2 1
Then,
X 2 p=+1
k =gN W
pk
This is an N/2 point DFT sequence
N
2 1
Then,
X 2 p=+1
k =gN W
pk
Thus an N-point DFT may be split into two N/2-point

DFTs.
N
2 1
Then,
X 2 p=+1
k =gN W
pk
Thus an N-point DFT may be split into two N/2-point DFTs.

This process of splitting may be continued up to 2-point transforms as N is
a power of 2.
Let N = 4. Then from relation (25),

N
1
2
X 2 p=
k
k =0g W
N
2 1
X2p =
pk
gWk
k =0
pk
N
2
, for p = 0,1,2,,(N/2-1)
N
2

N
1
2
X 2 p=
Where,
k
k =0g W
N
2 1
X2p =
pk
gWk
k =0
pk
, for p = 0,1,2,,(N/2-1)
N
2
gk =k x+
x
,
N
k+
2
for k = 0,1,2,,(N/2-1)
N
2
pk
k
k =0g W
Where,
or,
X =
, for p = 0,1,2,,(N/2-1)
N
2
gk =k x+
x
1
2p
X2p =
pk
gWk
k =0
N
1
2
X 2 p=
N
2 1
,
N
k+
2
pk
gW
k =0
, for
0,1
for k = 0,1,2,,(N/2-1)
p=
N
2
pk
k
k =0g W
Where,
or,
X =
gk =k x+
x
and,
, for p = 0,1,2,,(N/2-1)
N
2
1
2p
X2p =
pk
gWk
k =0
N
1
2
X 2 p=
N
2 1
,
N
k+
2
pk
gW
k =0
, for
x
g k = x+
k
k+
2
for k = 0,1,2,,(N/2-1)
p=
0,1
k=
for 0,1
N
2
Now,
and,
X 2 p=
pk
k
k =0g W
x
g k = x+
k
, for
k+
2
,
for
k=
0,1
p=
0,1
Now,
and,
X 2 p=
k
k =0g W
x
g k = x+
k
Then for
1
pk
p = 0,
, for
k+
2
,
for
p=
0,1
k=
0,1
X 0 =k =0 gkk =0W2 0 =k g0 =1 g+
Now,
and,
X 2 p=
k
k =0g W
x
g k = x+
k
Then for
p = 0,
pk
, for
k+
2
,
for
p=
0,1
k=
0,1
X 0 =k =0 gkk =0W2 0 =k g0 =1 g+
g
and for p =
1,
2 1
X 2 = k =0 g kW k2 = g0 W2 0 +1 g W
= g0 +
( 1)g =0 g 1
1
g
X 2 p+1
Now from relation (26),
N
1
2
=
gW
k
pk
/
=0
X 2 p+1
1 2
=
g WN
k
pk
k /
=0
, for p = 0,1,2,,(N/2-1)
N
2
X 2 p+1
N
1
2
=
gW
k
pk
/
=0
X 2 p+1
1 2
=
g WN
k
pk
k /
=0
where
, for p = 0,1,2,,(N/2-1)
k
k
= x x k+
g
/
N
2
WNk , for k = 0,1,2,,(N/2-1)
N
2
X 2 p+1
N
1
2
=
gW
k
pk
/
=0
X 2 p+1
1 2
=
g WN
k
pk
=0
where
, for p = 0,1,2,,(N/2-1)
k /
k
k
= x x k+
g
/
N
2
WNk , for k = 0,1,2,,(N/2-1)
or,
X =
2 p+1
k =0
gpk W
k
, for
p = 0,1
N
2
X 2 p+1
N
1
2
=
gW
k
pk
/
=0
X 2 p+1
1 2
=
g WN
k
pk
, for p = 0,1,2,,(N/2-1)
k /
=0
k
k
= x x k+
g
/
N
2
where
WNk , for k = 0,1,2,,(N/2-1)
or,
and,
X =
gpk W
, for
= (x x )W
g
,
2 p+1
k
=0
p = 0,1
k+
2
for
k=
0,1
N
2
Now,
and,
X =
gpk W
, for
= (x x )W
g
,
2 p+1
k
=0
p = 0,1
k+
2
for
k=
0,1
Now,
and,
X =
gpk W
, for
= (x x )W
g
,
2 p+1
k
=0
Then for
1
k+
2
p = 0,1
for
k=
0,1
p=
0,
X 1 = kk =0 g2 /W0 0
/
g
=g/+
Now,
and,
X =
gpk W
, for
= (x x )W
g
,
2 p+1
k
=0
k+
2
for
k=
0,1
p=
0,
Then for
1
X 1 = kk =0 g2 /W 0
/
g for p =
and
1
p = 0,1
=0 g / 1+
1,
X 3 = kk g2 /W k = g0 / +1
/ /
=0
/
1
g W= g g
0
Now from relations (27) and (30),

for
k =
0,1

for
g = x+ x
0
k =
0,1

for
g = x+ x
0
g = x+ x
1
k =
0,1

for
k =
0,1
g = x+ x
0
g = x+ x
1
0
0 = (x
x
)
W
= x x= x+ x
g
/
2
W 00
2
4
0
2
2
4

for
k =
0,1
g = x+ x
0
g = x+ x
1
0
0 = (x
x
)
W
= x x= x+ x
g
/
2
W 00
2
4
0
2
2
4
1
1
1
1
1 = (x
x
)
W
=
x
W
x
W
=
x
W
+
g
/
x W3 3 1 4 3 4 1 4 3 4 1 4

for
k =
0,1
g = x+ x
0
g = x+ x
1
= (x x ) = x x = x + x W 2 0
g
2
0
/
2
4
0
W
= (x0 2 x 4) = x W 1 x W 1 = x W 1 + x
1
g
/
3
W
1
4
3
4
1
4
W1
3
4
Then in matrix
1
3 form,
4
g0
0
x
0
1
0
1

1 0 2
1 1x
g 1 0
0 2x
=
4
0/
g 1 0 W
1
3
0 W x
g / 0 W
4
1

4
3
0
From relations (28), (29), (31) and (32),
X = g+ g
0 0
X = g+ g
0 0
X = g g
2 0
X = g+ g
0 0
X = g g
2 0
= g / + g /1
X = g+ g
0 0
X = g g
2 0
= g / + g /1
= g / g /3
X = g+ g
0
X = g g
X
= g / + g /1
= g / g /3
Then in matrix form,
0
g
0 1 1 0
X 2 1
1 0
g1
00
X 1 =
0
/
0 0 1 1 g
X
1 g/
3
1
00 1
From relations (33) and (34),
0
0
1 0 2
1
1x
g 1 0
0 2x
=
4
0/
g 1 0 W
1
3
0 W x
g / 0 W
4
1

4
3
1 0 1
Time
history
Natural order
0
0
1 0 2
1
1x
g 1 0
0 2x
=
4
0/
g 1 0 W
1
3
0 W x
g / 0 W
4
1

4
3
X
0
g
Frequency
0 1 1 0
0

0 g1
X2
1
0
1
X 1 =
0
/
0
0
1
1
/
X
1 g
3
1
0
0
1
Bit reversed order
1 0 1
Time
history
X
0

X2
0
0 g1
1 0
=
1
0

/
X
0 0 1 1 g /
X
1 g
3
1
Bit reversed order
0 0 1
Frequency
1 0
X
0

X2
0
0 g1
1 0
=
1
0

/
X
0 0 1 1 g /
X
1 g
3
1
Bit reversed order
0 0 1
Frequency
since,
1 0
X 0 = X00
X =X

X2 10= X
X =X
3 11
1
01
in terms of binary bits

g
1
0 1 0
1
g 0
1 0
0/
4
g
0 1 W
2
= 1
g / 0 W 0
1
x
0

1 x1
0 x 2

3
W x
4 3
0
X
0
0 1 1

10
X 2 1
=
X 1 0
0
1
X
3 0 0
1
g
0

0 g1
1 g / 0
1 g /
1

g
1
0 1 0
1
g 0
1 0
0/
4
g
0 1 W
2
= 1
g / 0 W 0
1
x
0

1 x1
0 x 2

3
W x
4 3
0
X
0
0 1 1

10
X 2 1
=
X 1 0
0
1
X
3 0 0
1
g
0

0 g1
1 g / 0
1 g /
1

g
1
0 1 0
1
g 0
1 0
0/
4
g
0 1 W
2
= 1
g / 0 W 0
1
x
0

1 x1
0 x 2

3
W x
4 3
0
X
0
0 1 1

10
X 2 1
=
X 1 0
0
1
X
3 0 0
1
Here, number of iterations = 2
g
0

0 g1
1 g / 0
1 g /
1

g
1
0 1 0
1
g 0
1 0
0/
4
g
0 1 W
2
= 1
g / 0 W 0
1
x
0

1 x1
0 x 2

3
W x
4 3
0
X
0
0 1 1

10
X 2 1
=
X 1 0
0
1
X
3 0 0
1
g
0

0 g1
1 g / 0
1 g /
1
Number of iterations = M, where M = log2N [as N = 2M], here N = 4 and M = 2
Each iteration involves N/2 number of butterfly computations.

Computation of g1 and g1 / may be represented as:
x1
g1
w4
1
x3
w4
w43 = - w41
-1
/
g1
w4
w4
w4
1
-j

x1
g1
1
x3
w4
w43 = - w41
/
g1
x1
computation
butterfly
x3
g1
g1
w4
-1
w4
w4
1
-j
w4

x1
g1
x1
g1
w4
1
x3
w4
w43 = - w41
/
g1
x3
w4
-1
w4
w4
w4
1
-1
1/
-j

x1
x1
g1
g1
w4
1
x3
w4
w43 = - w41
/
1
g
w4
-1
w4
w4
1
-1
w4
-j
This involves two complex additions and one complex multiplication.

x1
x1
g1
g1
w4
1
x3
w4
w43 = - w41
/
1
g
w4
-1
w4
w4
1
-1
w4
-j
This involves two complex additions and one complex multiplication.

This is true for all butterflies.
The procedure can be summarized as,

No. of iterations = M = log 2 N

NM N
Total no. of butterflies = 2= 2log N 2

NM N
No. of complex multiplications per butterfly = 1

NM N
No. of complex additions per butterfly = 2

NM N
NM N
Total no. of complex multiplications = 2= 2log N2

NM N
NM N
Total no. of complex multiplications = 2= 2log N2
Total no. of complex additions = NM = N log 2
N
Computation of each butterfly may be carried out in-place to reduce memory

requirement as follows:
A
Butterfly
A
B

A
Butterfly
A
B
B
-1

A
Butterfly
A
B
B
-1
T=AB
A = A +
B B = T
W

A
Butterfly
A
B
T=AB
A = A +
B B = T
T
-1 W
W
Here T is a scratch-pad variable and W is the twiddle

factor.
The above algorithm for the computation of FFT of sequence xk, k = 0,1,2,.,(N-1)
may be called radix-2 decimation-in-frequency in-place FFT algorithm.
Here, N should be a power of 2.
Here, N should be a power of 2.
Similarly, radix-2 decimation-in-time in-place FFT algorithm may be derived with
same computation load.
DFT
FFT
complex
complex
complex
complex
additions
multiplications
additions
multiplications
16
16
64
64
24
12
16
256
256
64
32
32
1024
1024
160
80
DFT
FFT
complex
complex
complex
complex
additions
multiplications
additions
multiplications
16
16
64
64
24
12
16
256
256
64
32
32
1024
1024
160
80
DFT
FFT
complex
complex
complex
complex
additions
multiplications
additions
multiplications
16
16
64
64
24
12
16
256
256
64
32
32
1024
1024
160
80
Reference:
Proakis and Manolakis, Introduction to Digital Signal processing.
Relations (25) and (26) may be split further (i.e. decimated) into N/2-point
DFTs as follows:
2p
g W pk
k=0
N
2
N
2
1
2 p+1
W
= gk/pk
N
k
=0
DFTs as follows:
In relation (25), splitting N/2-point sequence
gk
into two N/4-point

sequences,
2p
N
2
1
g W
k=0
pk
N
2
DFTs as follows:
gk
into two N/4-point

sequences,
N
4
X 2p=
k=
0
k
gW
pk
2p
g W
k=0
N
1
2
+ g W
pk
N
2
N
2
1
k=
N
4
N
2
pk
N
2
DFTs as follows:
gk
into two N/4-point

sequences,
N
4
X 2p=
k
gW
k=
0
N
1
4
k=
0
pk
N
1
2
+ g W
pk
g kW N +
2
2p
g W
k=0
pk
N
2
N
2
1
N
k=
N 4
1
4
N
g kW
+
N
k=0
N
2
p k+
N
4
2
pk
N
2
DFTs as follows:
gk
into two N/4-point

sequences,
N
4
X 2p=
=
=
pk
k
gW
k=
0
N
1
4
k=
0
N
1
4
k=
0
N
k=
N 4
1
4
g kW N +
2
pk
k
gW
N
1
2
+ g W
pk
N
2
2p
g W
k=0
pk
N
2
N
2
1
N
g kW
+
N
k=0
N
1
4
+ g
k=0
N
2
p k+
N
4
2
pk
N
W
k+
N
2
pN
4
N
2
pk
N
2
Now,
X2
p
N
4 1 4
pk
gW +
k =0 k k=0 N2
g k + N4
2
pN
4
N
2
pk
W W
N
Now,
X2
p
Here,
pN
4
N
2
N
4 1 4
N
pk
gW +
k =0 k k=0 N2
=
(1)p
g k + N4
2
pN
4
N
2
pk
W W
N
Now,
X2
p
Here,
pN
4
N
2
N
4 1
pk
gkWN2 +
k =0
=
(1)p
X2 p =
k
=0
N
4 1
Therefore,
N
1
4
k =0
g k + N4
pN
4
N
2
pk
W W
N
+ ( 1)p g N N pk
k+
W
Now,
X2
p
Here,
pN
4
N
2
Therefore,
N
4 1
N
1
4
pk
gkWN2 +
k =0
=
(1)p
X2 p =
k
=0
N
4 1
k =0
g k + N4
pN
4
N
2
pk
W W
N
+ ( 1)p g N N pk
k+
W
Nowgsplitting X2p into even and odd harmonics,
Now,
X2
p
Here,
pN
4
N
2
Therefore,
N
4 1
N
1
4
pk
gkWN2 +
k =0
=
(1)p
X2 p =
k
=0
N
4 1
k =0
g k + N4
pN
4
N
2
pk
W W
N
+ ( 1)p g N N pk
k+
W

for even harmonics,
p = 2r
, for r = 0,1,2,,(N/4-1)
Now,
X2
p
Here,
pN
4
N
2
Therefore,
N
4 1
N
1
4
pk
gkWN2 +
k =0
=
(1)p
X2 p =
k
=0
k =0
g k + N4
pN
4
N
2
pk
W W
N
+ ( 1)p g N N pk
k+
W
N
4 1

for even harmonics,
p = 2r
and for odd harmonics,

(N/4-1)
, for r = 0,1,2,,(N/4-1)
p = (2r +1) , for r = 0,1,2,,
Now, for even harmonics,
X 4r=
k = 0
g
N
4 1
+g
k
k+
2 rk
W
N
X 4r=
k = 0
N
g
4 1
=
k=0
g
N
4 1
+g
k
2 rk
W
N
N
k+
4
2
rk
+ g N W N
k+
4
4

2 rk
+
g
W
N
N
X 4r= k
k+
k = 0
N
g
4
2
4 1
rk
+ g N W N
= k
k+
4
4
k
=
0

g
Let hk =k g +
,
for
k = 0,1,2,,(N/4-1)
k+
N
g
4
N
4 1

2 rk
+
g
W
N
N
X 4r= k
k+
k = 0
N
g
4
2
4 1
rk
+ g N W N
= k
k+
4
4
k
=
0

g
Let hk =k g +
,
for
k = 0,1,2,,(N/4-1)
k+
N 4
g
N
N
4 1
4 1
Then,
X 4r=
k=
0
k
hW
rk
, for r = 0,1,2,,(N/4-1)
N
4

2 rk
+
g
W
N
N
X 4r= k
k+
k = 0
N
g
4
2
4 1
rk
+ g N W N
= k
k+
4
4
k
=
0

g
Let hk =k g +
,
for
k = 0,1,2,,(N/4-1)
k+
N 4
g
N
N
4 1
4 1
Then,
X 4r=
k=
0
k
hW
rk
, for r = 0,1,2,,(N/4-1)
N
4
This is an N/4-point DFT of sequence h , k = 0,1,2,,(N/4k

1)
Now, for odd harmonics,
X 4 r +=
2
N
4 1
gk
k=
g
0
( 2 r +1 )
k +N W
k N2
4
X 4 r +=
2
( 2 r +1 )
k
k
+
W
k=
N k N2
4
g
0
N
4 1
g
W N2 k Nrk
k
N
+
=
k
W4
k=0
N
4 1
X 4 r +=
2
Let
( 2 r +1 )
k
k
+
W
k=
N k N2
4
g
0
N
4 1
g
W N2 k Nrk
k
N
+
=
k
W4
k=0
/
2k
N
4 1
hk = gk g Nk + W4N
for k = 0,1,2,,(N/4-1)
( 2 r +1 )
k
k
+
W
k=
N k N2
4
g
0
N
4 1
g
W N2 k Nrk
k
N
+
=
k
W4
k=0
/
2k
h
N
4 1
X 4 r +=
2
Let
gk g N W4 N
k + 4
=
for k = 0,1,2,,(N/4-1)
N
4 1
Then,
X =
4r+
2
hW
k =0
rk k
N
4
, for r = 0,1,2,,(N/4-1)
( 2 r +1 )
k
k
+
W
k=
N k N2
4
g
0
N
4 1
g
W N2 k Nrk
k
N
+
=
k
W4
k=0
/
2k
h
N
4 1
X 4 r +=
2
Let
gk g N W4 N
k + 4
=
for k = 0,1,2,,(N/4-1)
N
4 1
Then,
X =
4r+
2
hW
k =0
rk k
, for r = 0,1,2,,(N/4-1)
N
4
This is an N/4-point DFT of sequence

(N/4-1)
k/
, k = 0,1,2,,
( 2 r +1 )
k
k
+
W
k=
N k N2
4
g
0
N
4 1
g
W N2 k Nrk
k
N
+
=
k
W4
k=0
/
2k
h
N
4 1
X 4 r +=
2
Let
gk g N W4 N
k + 4
=
for k = 0,1,2,,(N/4-1)
N
4 1
Then,
X =
4r+
2
hW
k =0
rk k
, for r = 0,1,2,,(N/4-1)
N
4
This is an N/4-point DFT of sequence k/

, k = 0,1,2,,
h
(N/4-1)
Thus the N/2-point DFT as represented in relation (25), may be split into
two N/4-point DFTs, as represented in relations (36) and (37).
Similarly, the N/2-point DFT in relation (26)

may be split into two even and odd harmonic
N/4- point DFTs as follows:
2 p+1
gW
k
N
2
1
k
=0
N
/
pk

For even harmonics,
N
X 4r+1=
1 4
l W rkk
k=0
N
4
, for r = 0,1,2,,(N/4-1)
2 p+1
gW
k
N
2
1
k
=0
/
pk

gW
For even harmonics,

N
X 4r+1=
1 4
l W rkk
k=0
2 p+1
N
2
1
k
=0
/
pk
, for r = 0,1,2,,(N/4-1)
N
4
This is an N/4-point DFT where
l = g/
k
+g
k
/
N
k+
4
, for k = 0,1,2,,(N/4-1)
X
Similarly for odd harmonics,
gW
k
X 4r
+3
N
1
4
=
k =0 l WN
rk
k/
, for r = 0,1,2,,(N/4-1)
2 p+1
N
2
1
k
=0
/
pk
X
Similarly for odd harmonics,
gW
k
X 4r
+3
N
1
4
=
k =0 l WN
rk
k/
, for r = 0,1,2,,(N/4-1)
This is another N/4-point DFT where
l/ k = / gk
k+
g
/
N
WN2k , for k = 0,1,2,,(N/4

1)
2 p+1
N
2
1
k
=0
/
pk
Let N = 8 (= 23) for 8-point

FFT.

FFT.
2p
N
2
1
g W
k=0
In relation (25),
pk
N
2

FFT.
, for k = 0,1,2,,(N/2-1)
k+
N
2
2p
g W
k=0
In relation (25),
g k =k x+ x
N
2
1
pk
N
2

FFT.
2p
g W
k=0
In relation (25),
g k =k x+ x
N
2
1
, for k = 0,1,2,,(N/2-1)
k+
N
2
Possible values of k are k = 0,1,2,3.
pk
N
2

FFT.
2p
g W
k=0
In relation (25),
g k =k x+ x
N
2
1
pk
N
2
, for k = 0,1,2,,(N/2-1)
k+
N
2

Then,
g 0 = x0 + x4
g = x + x
1
g 2 = 2x + 6x
g3 =
x+
x
3
7
X
Now in relation (26),
2 p+1
gW
k
N
2
1
k
=0
N
/
pk
X
2 p+1
gW

gk = kx
x
/
N k
k+ 2
, for k = 0,1,2,,(N/2-1)
N
2
1
k
=0
N
/
pk
X
2 p+1
gW

gk = kx
x
/
, for k = 0,1,2,,(N/2-1)
N k
k+ 2
N
2
1
k
=0
N
/
pk
X
2 p+1
gW

gk = kx
x
/
N
2
1
k
=0
/
pk
, for k = 0,1,2,,(N/2-1)
N k
k+ 2

Then,
= (x0 4 x )8W 0
g
/
g / = (x x ) 1
W2 = (x2 6 x )8W 2
1
1
5
8
g
/
3
=
(
x
x
)
W
3
3
7
8
g
/
g
0
x0
4

= x
+
x
g
g
=
x3
x
x2
x7
g0
g
x
g
/
3
(x
)W
(x
0
x
=
)W
=
x
3
1
(x
)W
(x
)
=
x
8
4
8 2
W 83
5
0
g
0
g
1
x0
4

= x
+
x
g0
g
x
/
3
(x
)W
(x
0
x
=
)W
8
4
8 2
W 83
5
0
1
1
g
= x
+
g
=
(x
From2 relations
x2
6 (40) and (41), signal flow graph for computations of
x
)W
gg03 =x03 gx / may

2
2
6
be represented as:
+
3 and
3
7
g
=
(x
)
7
g
0
g
1
x0
4

= x
+
x
g0
g
x
/
3
(x
)W
(x
0
x
=
)W
8
4
8 2
W 83
5
0
1
1
g
= x
+
g
=
(x
From2 relations
x2
6 (40) and (41), signal flow graph for computations of
x
)W
gg0 =x gx0/ may

2
2
6
be represented as:
+
3
7 3
g
=
(x
3 3 and
)
7
hk = gk +
N
k+
4
g
, for k = 0,1,2,,(N/4-1)
hk = gk +
N
k+
4
g
, for k = 0,1,2,,(N/4-1)
Possible values of K are K = 0,1.
hk = gk +
N
k+
4
g
, for k = 0,1,2,,(N/4-1)

Then,
h0 =
0 g+
2 g
h = g+ g
1
1
3
And from relation (37),

h = gk
g
/
k
k+
N
W
k
2
N
, for k = 0,1,2,,(N/4-1)

h = gk
4
g
/
k
k+
N
WN2 k , for k = 0,1,2,,(N/4-
1)

h = gk
4
g
/
k
k+
N
WN2 k , for k = 0,1,2,,(N/4-
1)
Then,
= (g0 2 g )8W 0
h
2
3 g )8W
/ = (g
1
1
h
/
0
lk =k g/ +
g/
, for k = 0,1,2,,(N/4-1)
k+
N
4
lk =k g/ +
g/
, for k = 0,1,2,,(N/4-1)
k+
N
4
lk =k g/ +
g/
, for k = 0,1,2,,(N/4-1)
k+
N
4

Then,
l0 =0 g/ 2+
/
gl1/ =1 g 3+
g/
/
/ /
2k
k
k
N
l = g gk+N 4W
, for k = 0,1,2,,(N/4-1)
/
/ /
2k
k
k
l = g 4 gk+N W N
, for k = 0,1,2,,(N/4-1)
/
/ /
2k
k
k
l = g 4 gk+N W N
, for k = 0,1,2,,(N/4-1)

Then,
= (g 0/ 2g / )
l/
W
= (g 1/ 3g / )
1
l/
W
0
2
8
=g +g

0
0
2
h =g
+g
1
1
3
h0/
h/
)W
1
=00 (g 2g )W
= (g 8 g 2
l0 =0/ g +2/

lg = g /
+/
g
1
1
3
l 0/
W
l1/
=
=/ (0 g / g2 ) 0
(g1/ 38/ )W 28
g
l 0/ =/ (0 g / g2 ) 0
h0/ =00 (g 2g )W
=g +g
l0 =0/ g +2/

/
/
2
8
W
0
0
2
l / =/ (g / g8 ) 2
h
=
(
g
g
+/
gl = g
g
h =g
+g
W
)W
1
1
3
1
1
3
1
1
3
8
1
1
3
8
From relations (42), (43), (44) and (45), signal flow graph for computation of
h
h01, h01, l
0 1
and l
/
0 1
may be represented as:
l 0/ =/ (0 g / g2 ) 0
h0/ =00 (g 2g )W
=g +g
l0 =0/ g +2/

/
/
2
8
W
0
0
2
l / =/ (g / g8 ) 2
h
=
(
g
g
+/
gl = g
g
h =g
+g
W
)W
1
1
3
1
1
3
1
1
3
8
1
1
3
8
From relations (42), (43), (44) and (45), signal flow graph for computation of
h
h01, h01, l
0 1
and l
/
0 1
may be represented as:
Now, from relation (36),

possible values of r are r = 0,1.
X 4r
=
N
1
4
k=0
hW
rkk
N
4
, for r = 0,1,2,,(N/4-1)

Then,
0 0
2
X 0 = 0h W2 0 +1 h W
= h1 + h
X = h W 0+ h W =1 h h
4 0
X 4r
=
N
1
4
k=0
hW
rkk
, for r = 0,1,2,,(N/4-1)
N
4

X 4r
=
N
1
4
k=0
hW
Then,
rkk
, for r = 0,1,2,,(N/4-1)
N
4
X 0 = 0h W2 0 + h1 W 2 = h+
0 h1
0
0
1
X = h W + h W = h h
4 0from2relation
1
2
0
And,
(37),
N
1
4
X 4 r + 2 = h /W
k=0
rk
,
k
N
4
for r = 0,1,2,,(N/4-1)

X 4r
=
N
1
4
k=0
hW
Then,
rkk
, for r = 0,1,2,,(N/4-1)
N
4
X 0 = 0h W2 0 + h1 W2 = 0h+ h 1
0
0
1
X = h W + h W = h h
4 0from2 relation
1
2
0
And,
(37),
N
1
4
X 4 r + 2 = h /W
k=0
Then,
X
= h /W 0 + h /W 0 + h1 /
=2 h / 20 / 00 2 / 1 1
X
= h W + h W h1 /
=2 h / 60
0
2
1
rk
,
k
for r = 0,1,2,,(N/4-1)
N
4
From relation (38),

N
4 1
X 4r =
rk
k
l
W
k
=0
+1
N
4
, for r = 0,1,2,,(N/4-1)
From relation (38),
N
4 1
X 1 = 0l W2 0 +1 l W2 0 =0 l
1
0
1
X
=
l
W
+
l
W
=
l
l
+l
, for r = 0,1,2,,(N/4-1)
N
4
Then,
X 4r =
rk
k
l
W
k
=0
+1
From relation (38),
N
4 1
X 4r =
rk
k
l
W
k
=0
+1
, for r = 0,1,2,,(N/4-1)
N
4
Then,
X 1 = 0l W2 0 +1 l W2 0 =0 l
1
0
1
X
=
l
W
+
l
W
=
l
l
+l
5 0from2 relation
1
2(39),0
And,
N
4 1
X 4 r +3=
rk
l W N4
k =0
k
, for r = 0,1,2,,(N/4-1)
From relation (38),
N
4 1
X 4r =
rk
k
l
W
k
=0
+1
, for r = 0,1,2,,(N/4-1)
N
4
Then,
X 1 = 0l W2 0 +1 l W2 0 =0 l
1
0
1
X
=
l
W
+
l
W
=
l
l
+l
5 0from2 relation
1
2(39),0
And,

Then,
X 3 = 0l /W2 0 +1 l /W2 0 =0 l /
1
/= l /W 0 + l /W1 = l /
X
+ 7l 0 2 1 2 0
1
l/
N
4 1
X 4 r +3=
rk
l W N4
, for r = 0,1,2,,(N/4-1)
k =0
k
0
X 0 = 0h W 02 + hW
=2 h +
1
0
1
0
1
+ hW
=
hX = h W
h 4 h0
2
1
2
0
0
0
X 1 = 1l W
+2 l W 1 = l 2 + l
0
0
X = l W0
l 5
l 2
0
1
+lW
0
X 2 = 0h/W 02 + h/W
= 2h/ + h/
1
0
1
X = /h W 0/ + / h W 1 =/ h
6h 0
2
1
2
0
/
0
/
0
/
1 l W
X3 =
+l W
=2 l + l
0
2
1
0
X = /l W
7
l 2
0
1
1
0
/
+ /l W
1
=/ l
0
0
X 0 = 0h W 02 + hW
=2 h +
1
0
1
0
1
+ hW
=
hX = h W
h 4 h0
2
1
2
0
0
0
X 1 = 1l W
+2 l W 1 = l 2 + l
0
0
0
X 2 = 0h/W 02 + h/W
= 2h/ + h/
1
0
1
X = /h W 0/ + / h W 1 =/ h
6h 0
2
1
2
0
/
0
/
0
/
1 l W
X3 =
+l W
=2 l + l
0
2
1
0
1
/X
X = l W0
+lW
=
= /l W 0/ + / l W 1 =/ l
7
l 5
l 2
l 2
0
1
2
0
0
1
2
0
From relations
(46), (47), (48) and (49), signal1 flow
graph for computation of
1
X 07 may be represented as:
0
X 0 = 0h W 02 + hW
=2 h +
1
0
1
0
1
+ hW
=
hX = h W
h 4 h0
2
1
2
0
0
0
X 1 = 1l W
+2 l W 1 = l 2 + l
0
0
0
X 2 = 0h/W 02 + h/W
= 2h/ + h/
1
0
1
X = /h W 0/ + / h W 1 =/ h
6h 0
2
1
2
0
/
0
/
0
/
1 l W
X3 =
+l W
=2 l + l
0
2
1
0
1
/X
X = l W0
+lW
=
= /l W 0/ + / l W 1 =/ l
7
l 5
l 2
l 2
0
1
2
0
0
1
2
0
From relations
(46), (47), (48) and (49), signal1 flow
1
0
X 0 = 0h W 02 + hW
=2 h +
1
0
1
0
1
+ hW
=
hX = h W
h 4 h0
2
1
2
0
0
0
X 1 = 1l W
+2 l W 1 = l 2 + l
0
0
0
X 2 = 0h/W 02 + h/W
= 2h/ + h/
1
0
1
X = /h W 0/ + / h W 1 =/ h
6h 0
2
1
2
0
/
0
/
0
/
1 l W
X3 =
+l W
=2 l + l
0
2
1
0
1
/X
X = l W0
+lW
=
= /l W 0/ + / l W 1 =/ l
7
l 5
l 2
l 2
0
1
2
0
0
1
2
0
From relations
(46), (47), (48) and (49), signal1 flow
1
not in natural order, hence bit-reversal

should be carried out to bring it in
natural order
Bit Reversal procedure
Bit Reversal procedure
Complete Signal Flow Graph
Complete Signal Flow Graph
Time
history
Frequency
C ***Subroutine to compute radix-2 FFT***

C Decimation-in-frequency in-place algorithm
SUBROUTINE FFT(A,N,INV)
C N: Dimension of Array (must be a power of
2) C A: Complex array containing data
sequence C DFT coefficients are returned in
the array
C INV = 0 for forward FFT C INV = 1 for
inverse FFT
C ***Subroutine to compute radix-2 FFT***

C Decimation-in-frequency in-place algorithm
SUBROUTINE FFT(A,N,INV)
C N: Dimension of Array (must be a power of
2) C A: Complex array containing data
sequence C DFT coefficients are returned in
the array
C INV = 0 for forward FFT C INV = 1 for
inverse FFT
DIMENSION A(N) COMPLEX T,W,A
IF (INV.EQ.0) GO TO 8
C Divide sequence by N for inverse FFT DO 7
I=1,N
7
A(I)=A(I)/CMPLX(FLOAT(N),
8
0.0) S=-1.0
IF (INV.EQ.1) S=1.0
C Calculate number of iterations

C M: Number of iterations (log(N) to the base 2)
M=1
K=N
2
K=K/2
IF (K.EQ.1) GO TO 1
M=M+1
GO TO 2
C Calculate number of iterations

C M: Number of iterations (log(N) to the base 2)
M=1
K=N
2
K=K/2
IF (K.EQ.1) GO TO 1
M=M+1
GO TO 2
C Compute for each iteration
C NP: Number of points in each partition
1
NB=N
DO 3 I=1,M NP=NB
NB=NP/2
PHI=3.14159265/FLOAT
(NB)

1
NB=N
DO 3 I=1,M NP=NB
NB=NP/2
(NB)
C Calculate the twiddle factor W for each butterfly
C NB: Number of butterflies for each partition
DO 3 J=1,NB ARG=FLOAT(J-1)*PHI
W=CMPLX(COS(ARG),S*SIN(ARG))

1
NB=N
DO 3 I=1,M NP=NB
NB=NP/2
(NB)
C Compute butterfly for each partition DO 3
K=NP,N,NP
J1=K-NP+J J2=J1+NB T=A(J1)-A(J2)
A(J1)=A(J1)+A(J2) A(J2)=T*W
3 CONTINUE

1
NB=N
DO 3 I=1,M NP=NB
NB=NP/2
(NB)
K=NP,N,NP
A(J1)=A(J1)+A(J2) A(J2)=T*W
3 CONTINUE

1
NB=N
DO 3 I=1,M NP=NB
NB=NP/2
(NB)
K=NP,N,NP
A(J1)=A(J1)+A(J2) A(J2)=T*W
3 CONTINUE
Nested DO loop
C Bit reversal operation

N2=N/2
N1=N-1
J=1
DO 4 I=1,N1
IF (I.GE.J) GO TO 5
T=A(J)
A(J)=A(I) A(I)=T
5
6
K=N2
IF (K.GE.J) GO TO 4
J=J-K
K=K/2 GO TO 6
J=J+K RETURN
END
C Bit reversal operation

N2=N/2
N1=N-1
J=1
DO 4 I=1,N1
IF (I.GE.J) GO TO 5
T=A(J)
A(J)=A(I) A(I)=T
5
6
K=N2
IF (K.GE.J) GO TO 4
J=J-K
K=K/2 GO TO 6
J=J+K RETURN
END
During the bit-reversal operation, N/2 DFT coefficients remain unchanged and
the remaining N/2 coefficients are exchanged in place as required.
Computation of amplitude spectrum of a finite real data sequence
data
sequence xk,
k=0, 1, ,
(N-1)
(N must be a
power of 2)
complex
array
of
length N
from xk, k=0,
1, , (N-1)
Compute
N- point
FFT
Compute
amplitudes
Cn, n=0, 1,
, ((N/2)-1)
data
sequence xk,
k=0, 1, ,
(N-1)
complex
array
of
length N
from xk, k=0,
1, , (N-1)
(N must be a
power of 2)
1
C =N X
0
, the average value
Compute
N- point
FFT
Compute
amplitudes
Cn, n=0, 1,
, ((N/2)-1)
data
sequence xk,
k=0, 1, ,
(N-1)
complex
array
of
length N
from xk, k=0,
1, , (N-1)
Compute
N- point
FFT
Compute
amplitudes
Cn, n=0, 1,
, ((N/2)-1)
(N must be a
power of 2)
1
C =N X
, the average value
2
C =N X
, for n = 1,2,,(N/2-1), the nth harmonic amplitude.
and
The range of frequency may be expressed as fs/2 where fs is the

sampling
frequency 1 .
=
The range of frequency may be expressed as fs/2 where fs is the

sampling
frequency 1 .
=
The frequency resolution may be estimated as f0 where f0 is the fundamental

frequency (=1/T0 ), where T0 is the time period of fundamental frequency and also
the width of the analysis window.
FORTRAN program for computation of amplitude spectrum

C ***Amplitude spectrum analysis program using FFT***
DIMENSION A(1024),B(1024),C(512),PHASE(512)
COMPLEX A
CHARACTER*64 FNAME WRITE(*,10)
10
20
200
FORMAT(1X,'Enter file name - '\)

READ(*,20)FNAME
FORMAT(A) OPEN(2,
FILE=FNAME) READ(2,*,
END=100)(B(I),I=1,1024) 100 N=I1
CLOSE(2) WRITE(*,200)N
FORMAT(1X,'Data points = ',I4)
15
24
5
25
30
300
DO 15 I=1,10
IF(N-2**I)24,25,15 CONTINUE WRITE(*,5)
FORMAT(1X,'Incorrect size - it must be a power of 2')
STOP
DO 30 I=1,N A(I)=CMPLX(B(I),0.0)/CMPLX(FLOAT(N),
0.0) WRITE(*,300)
FORMAT(1X,'FFT analysis in progress') CALL FFT(A,N,
0)
40
80
NA=N/2 C(1)=CABS
(A(1)) DO 40 I=2,NA
C(I)=CABS(A(I))*2.0
D=180.0/3.141592654
DO 80 I=2,NA
R=REAL(A(I))
X=AIMAG(A(I))
ALPHA=ATAN2(X,R)
PHASE(I)=D*ALPHA
60
70
75
85
90
WRITE(*,60)
FORMAT('0','Harmonic no.',7X,'Amplitude',12X,'Phase (deg)')
WRITE(*,70)
FORMAT(1X,'------------',7X,'---------',12X,'-----------'//) NB=0
WRITE(*,75)NB,C(1) FORMAT(5X,I3,9X,1P,E13.6) DO 85
I=2,NA
NB=I-1 WRITE(*,90)NB,C(I),PHASE(I)
FORMAT(5X,I3,9X,1P,E13.6,9X,E13.6) END
FFT-based digital filtering of a finite real data sequence
convolution
xk
input
sequence
Digital
Filter
yk
output
sequence
convolution
xk
input
sequence
Digital
Filter
yk
output
sequence
yk = hk * xk, hk is the impulse sequence of the digital filter
convolution
Digital
xk
yk
Filter
input
sequence
output
sequence
multiplication
xk
FFT
Xn
Hn
Yn
IFFT
yk
convolution
Digital
xk
yk
Filter
input
sequence
output
sequence
multiplication
xk
FFT
Xn
Hn
Yn
IFFT
yk
Yn = Hn . Xn, Hn is the complex gain of the digital filter

m u lt ip l i c a t i o n
xk
FFT
Xn
Hn
Yn
IF F T
yk
Yn = H n . X n

m u lt ip l i c a t i o n
xk
FFT
Xn
N -point R eal
data sequence
xk, k=0, 1, ,
(N -1)
Yn
Hn
Form N-point
array
input
sequence
((N/2)+1)-point
Com pute
Filter specification
(upto folding frequency)
yk
Yn = H n . X n
Xn
FFT
m ultiplier
Yn
Form com plex

length N using
com plex
conjugates
Frequency
response
Com pute
N- point
IFFT
gain and phase

characteristic
IF F T
H
n
output
sequence
yk,
k=0, 1, ,
(N -1)

N -p o in t R ea
da
l ta sequence
x k, k = 0 , 1 ,
,
(N -1 )
in p u t
seq u en ce
Fo rm N-po in t
arra y
Com
pute
Xn
FFT
m u ltip lier
Yn
IFFT
((N /2 )+ 1 )-p o in
F o rm co m p
lex
len g th N usin
g co m p lex
co n ju g a tes
t g a in a nd p h
a se ch
aracteristic
F ilter sp ecifica tio n (u

p to fo ld in g freq u en cy)
Hn
N/2
F req u en cy
resp o n se
Co m p u
te
N- po in t
Hn
seq u en ce
yk ,
k = 0 , 1 ,
, (N -1
)

N -p o in t R ea
da
l ta sequence
x k, k = 0 , 1 ,
,
(N -1 )
in p u t
seq u en ce
Fo rm N-po in t
arra y
Com
pute
Xn
FFT
Yn
m u ltip lier
IFFT
((N /2 )+ 1 )-p o in
F o rm co m p
lex
len g th N usin
g co m p lex
co n ju g a tes
t g a in a nd p h
a se ch
aracteristic
F ilter sp ecifica tio n (u

p to fo ld in g freq u en cy)
Hn
Co m p u
te
N- po in t
Hn
F req u en cy
resp o n se
Hn
N/2
N/2
N-1 n
seq u en ce
yk ,
k = 0 , 1 ,
, (N -1
)
FORTRAN program for FFT-based digital filtering

C ***FFT based digital filter program*** DIMENSION X(1024),
A(513),PH(513) COMPLEX H(1024),CX(1024)
CHARACTER*64 FNAME1,FNAME2 WRITE(*,20)
20
30
40
FORMAT(1X,'Enter input file name - '\)

READ(*,30)FNAME1
FORMAT(A) WRITE(*,40)
FORMAT(1X,'Enter output file name - '\)
READ(*,30)FNAME2
50
60
70
375
380
400
OPEN(1,FILE=FNAME1) READ(1,*,END=50)(X(I),
I=1,1024) GOTO 60
I=I-1 N=I
WRITE(*,70)N
FORMAT(1X,'Data points = ',I4) DO 375 I=1,10
IF(N-2**I)380,390,375 CONTINUE WRITE(*,400)
FORMAT(1X,'Incorrect size of data') STOP

390
100
11
0
WRITE(*,100)N/2+1
FORMAT(1X,'Enter filter gain (',I3,' points)') READ(*,*)(A(I),I=1,
N/2+1)
PH(1)=0.0 WRITE(*,110)N/2
FORMAT(1X,'Enter filter phase (',I3,' points)') READ(*,*)(PH(I),
I=2,N/2+1)

390
100
WRITE(*,100)N/2+1
N/2+1)
PH(1)=0.0 WRITE(*,110)N/2
I=2,N/2+1)
11
0
C Form complex filter gain array H
(1)=CMPLX(A(1)*N,0.0)
DO 200 I=2,N/2+1
200
H(I)=CMPLX(A(I)*N*COS(PH(I))/2.0,A(I)*N*SIN(PH(I))/2.0)
N /2

390
100
WRITE(*,100)N/2+1
N/2+1)
PH(1)=0.0 WRITE(*,110)N/2
I=2,N/2+1)
11
0
C Form complex filter gain array H
(1)=CMPLX(A(1)*N,0.0)
DO 200 I=2,N/2+1
200
H(I)=CMPLX(A(I)*N*COS(PH(I))/2.0,A(I)*N*SIN(PH(I))/2.0)
C Form rest of the gain array by complex conjugate
DO 340 I=2,N/2
J=N+2-I
340
H(J)=CONJG(H(I))
H
N /2
N -1

C Form complex input data array DO 350 I=1,N
350 CX(I)=CMPLX(X(I),0.0)
xk

350 CX(I)=CMPLX(X(I),0.0) C Compute FFT
CALL FFT(CX,N,0)
xk
FFT
Xn

CALL FFT(CX,N,0)
C Perform filtering in frequency domain DO 360 I=1,N
360 CX(I)=CX(I)*H(I)
Multiplication
xk
FFT
Xn
Hn
Yn

CALL FFT(CX,N,0)
C Perform filtering in frequency domain DO 360 I=1,N
360 CX(I)=CX(I)*H(I) C Back to time
CALL FFT(CX,N,1) DO 370 I=1,N
370 X(I)=REAL(CX(I))
Multiplication
xk
FFT
Xn
Hn
Yn
IFFT
yk

C Save output
IF(FNAME1.EQ.FNAME2)CLOSE(1) OPEN(2,FILE=FNAME2,
STATUS='NEW') WRITE(2,*)(X(I),I=1,N)
END
J
J
a

DFT and FFT

Caricato da

Informazioni sul documento

Descrizione originale:

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

DFT and FFT

Caricato da

Copyright:

Formati disponibili

Discrete Fourier

The signal x(t) may be expressed in terms of the Fourier series as

the average value

the average value

Relation (2) may be rewritten as

Relation (2) may be rewritten as

x(t) = C0 + n=1 Cn cos(n0t +n

[cos(A+ B) = cosAcosB sin

Relation (2) may be rewritten as

x(t) = C0 + n=1 Cn cos(n0t +n

[cos(A+ B) = cosAcosB sin

Relation (2) may be rewritten as

x(t) = C0 + n=1 Cn cos(n0t +n

[cos(A+ B) = cosAcosB sin

Cn, n = 1, 2, 3, is the amplitude and n, n = 1, 2, 3, is the phase

x(t) = F0 + [Fn e jn+0tFnejn0t

Relation (4) may be expressed as

Hence, we can write,

Hence, we can write,

Thus x(t) may be expressed in terms of Complex Fourier Series in relation

Hence, we can write,

Variation of Fn coefficients with n

Hence, we can write,

Variation of Fn coefficients with n

The amplitudes Cns of relation (3) may be related to Fns as

, the average value

The amplitudes Cns of relation (3) may be related to Fns as

, the average value

the amplitude of the nth harmonic.

The amplitudes Cns of relation (3) may be related to Fns as

, the average value

the amplitude of the nth harmonic.

From relation (4), Fn may be expressed as

From relation (4), Fn may be expressed as

From relation (4), Fn may be expressed as

From relation (4), Fn may be expressed as

Reference: Network Analysis by

Let xk be a periodic discrete sequence obtained from a periodic signal x(t)

Let xk be a periodic discrete sequence obtained from a periodic signal x(t)

Let xk be a periodic discrete sequence obtained from a periodic signal x(t)

From relation (2)

From relation (2)

From relation (2)

From relation (2)

From relation (2)

From relation (2)

(using the substitutions: 0 =

From relation (2)

From relation (2)

From relation (2)

(using the substitutions: 0 =

From relation (2)

Now from relations (11) and

Now from relations (11) and

Now from relations (11) and

Now from relations (11) and

From relation (13)

From relation (13)

Now, for n N, let n = N + m, for m = 0, 1, 2,

From relation (13)

Now, for n N, let n = N + m, for m = 0, 1, 2,

From relation (13)

Now, for n N, let n = N + m, for m = 0, 1, 2,