You are on page 1of 4

The Fourier Transform

and its Applications


The Fourier Transform:
F(s) =
_

f(x)e
i2sx
dx
The Inverse Fourier Transform:
f(x) =
_

F(s)e
i2sx
ds
Symmetry Properties:
If g(x) is real valued, then G(s) is Hermitian:
G(s) = G

(s)
If g(x) is imaginary valued, then G(s) is Anti-Hermitian:
G(s) = G

(s)
In general:
g(x) = e(x) + o(x) = e
R
(x) + ie
I
(x) + o
R
(x) + io
I
(x)
G(s) = E(s) + O(s) = E
R
(s) + iE
I
(s) + iO
I
(s) + O
R
(s)
Convolution:
(g h)(x)

=
_

g()h(x )d
Autocorrelation: Let g(x) be a function satisfying
_

|g(x)|
2
dx < (nite energy) then

g
(x)

= (g

g)(x)

=
_

g()g

( x)d
= g(x) g

(x)
Cross correlation: Let g(x) and h(x) be functions with
nite energy. Then
(g

h)(x)

=
_

()h( + x)d
=
_

( x)h()d
= (h

g)

(x)
The Delta Function: (x)
Scaling: (ax) =
1
|a|
(x)
Sifting:
_

(x a)f(x)dx = f(a)
_

(x)f(x + a)dx = f(a)


Convolution: (x) f(x) = f(x)
Product: h(x)(x) = h(0)(x)

2
(x) - no meaning
(x) (x) = (x)
Fourier Transform of (x): F{(x)} = 1
Derivatives:

(n)
(x)f(x)dx = (1)
n
f
(n)
(0)

(x) f(x) = f

(x)
x(x) = 0
x

(x) = (x)
Meaning of [h(x)]:
[h(x)] =

i
(x x
i
)
|h

(x
i
)|
The Shah Function: III(x)
Sampling: III(x)g(x) =

n=
g(n)(x n)
Replication: III(x) g(x) =

n=
g(x n)
Fourier Transform: F{III(x)} = III(s)
Scaling: III(ax) =

(ax n) =
1
|a|

(x
n
a
)
Even and Odd Impulse Pairs
Even: II(x) =
1
2
(x +
1
2
) +
1
2
(x
1
2
)
Odd: I
I
(x) =
1
2
(x +
1
2
)
1
2
(x
1
2
)
Fourier Transforms: F{II(x)} = cos s
F{I
I
(x)} = i sin s
Fourier Transform Theorems
Linearity: F{f(x) + g(x)} = F(s) + G(s)
Similarity: F{g(ax)} =
1
|a|
G(
s
a
)
Shift: F{g(x a)} = e
i2as
G(s)
F{g(ax b)} =
1
|a|
e
i2s
b
a
G(
s
a
)
Rayleighs:
_

|g(x)|
2
dx =
_

|G(s)|
2
ds
Power:
_

f(x)g

(x)dx =
_

F(s)G

(s)ds
Modulation:
F{g(x)cos(2s
0
x)} =
1
2
[G(s s
0
) + G(s + s
0
)]
Convolution: F{f g} = F(s)G(s)
1
Autocorrelation: F{g

g} = |G(s)|
2
Cross Correlation: F{g

f} = G

(s)F(s)
Derivative:
F{g

(x)} = i2sG(s)
F{g
(n)
(x)} = (i2s)
n
G(s)
F{x
n
g(x)} = (
i
2
)
n
G
(n)
(s)
Fourier Integral: If g(x) is of bounded variation and
is absolutely integrable, then
F
1
{F{g(x)}} =
1
2
[g(x
+
) + g(x

)]
Moments:
_

f(x)dx = F(0)
_

xf(x)dx =
i
2
F

(0)
_

x
n
f(x)dx = (
i
2
)
n
F
(n)
(0)
Miscellaneous:
If F{g(x)} = G(s) then F{G(x)} = g(s)
and F{g

(x)} = G

(s)
F
__
x

g()d
_
=
1
2
G(0)(s) +
G(s)
i2s
Function Widths
Equivalent Width
W
f

=
_

f(x)dx
f(0)
=
F(0)
f(0)
=
F(0)
_

F(s)ds
=
1
W
F
Autocorrelation Width
W
f

=
_

f dx
f

f |
x=0
=
|F(0)|
2
_

|F(s)|
2
ds
=
1
W
|F|
2
Standard Deviation of Instantaneous Power: x
(x)
2

=
_

x
2
|f(x)|
2
dx
_

|f(x)|
2
dx

__

x|f(x)|
2
dx
_

|f(x)|
2
dx
_
2
(s)
2

=
_

s
2
|F(s)|
2
ds
_

|F(s)|
2
ds

__

s|F(s)|
2
ds
_

|F(s)|
2
ds
_
2
Uncertainty Relation: (x)(s)
1
4
Central Limit Theorem
Given a function f(x), if F(s) has a single absolute
maximum at s = 0; and, for suciently small |s|,
F(s) a cs
2
where 0 < a < and 0 < c < ,
then:
lim
n
[

nf(

nx)]
n
a
n
=
_
a
2
e

a
c
x
2
and
[f(x)]
n

a
n+
1
2
n
1
2
_

c
e

a
cn
x
2
Linear Systems
For a linear system w(t) = S[v(t)] with response h(t, )
to a unit impulse at time :
S[v
1
(t) + v
2
(t)] = S[v
1
(t)] + S[v
2
(t)]
w(t) =
_

v()h(t, )d
If such a system is time-invariant, then:
w(t ) = S[v(t )]
and
w(t) =
_

v()h(t )d
= (v h)(t)
The eigenfunctions of any linear time-invariant system
are e
i2f0t
, since for a system with transfer function H(s),
the response to an input of v(t) = e
i2f0t
is given by:
w(t) = H(f
0
)e
i2f0t
.
Sampling Theory
g(x) = III(
x
X
)g(x)
= X

n=
g(nX)(x nX)
2

G(s) = XIII(Xs) G(s)


=

n=
G(s
n
X
)
Whittaker-Shannon-Kotelnikov Theorem: For a bandlim-
ited function g(x) with cuto frequencies s
c
, and with
no discrete sinusoidal components at frequency s
c
,
g(x) =

n=
g(
n
2s
c
)sinc[2s
c
(x
n
2s
c
)]
Fourier Tranforms for Periodic Functions
For a function p(x) with period L, let f(x) = p(x) (
x
L
).
Then
p(x) = f(x)

n=
(x nL)
P(s) =
1
L

n=
F(
n
L
)(s
n
L
)
The complex fourier series representation:
p(x) =

n=
c
n
e
i2
n
L
x
where
c
n
=
1
L
F(
n
L
)
=
1
L
_
L/2
L/2
p(x)e
i2
n
L
x
dx
The Discrete Fourier Transform
Let g(x) be a physical process, and let f(x) = g(x) for 0
x L, f(x) = 0 otherwise. Suppose f(x) is approximately
bandlimited to B Hz, so we sample f(x) every 1/2B
seconds, obtaining N = 2BL samples.
The Discrete Fourier Transform:
F
m
=
N1

n=0
f
n
e
i
2mn
N
for m = 0, . . . , N 1
The Inverse Discrete Fourier Transform:
f
n
=
1
N
N1

m=0
F
m
e
i
2mn
N
for n = 0, . . . , N 1
Convolution:
h
n
=

N1
k=0
f
k
g
nk
for n = 0, . . . , N 1
where f, g are periodic
Serial Product:
h
n
=

N1
k=0
f
k
g
nk
for n = 0, . . . , 2N 2
where f, g are not periodic
DFT Theorems
Linearity: DFT {f
n
+ g
n
} = F
m
+ G
m
Shift: DFT {f
nk
} = F
m
e
i
2
N
km
(f periodic)
Parsevals:

N1
n=0
f
n
g

n
=
1
N

N1
m=0
F
m
G

m
Convolution: F
m
G
m
= DFT {

N1
k=0
f
k
g
nk
}
The Hilbert Transform
The Hilbert Transform of f(x):
F
Hi
(x)

=
1

f()
x
d (CPV)
The Inverse Hilbert Transform:
f(x) =
1

F
Hi
()
x
d + f
DC
(CPV)
Impulse response:
1
x
Transfer function: i sgn(s)
Causal functions: A causal function g(x) has Fourier
Transform G(s) = R(s) + iI(s), where I(s) =
H{R(s)}.
Analytic signals: The analytic signal representation
of a real-valued function v(t) is given by:
z(t)

= F
1
{2H(s)V (s)}
= v(t) iv
Hi
(t)
Narrow Band Signals: g(t) = A(t) cos[2f
0
t + (t)]
Analytic approx: z(t) A(t)e
i[2f0t+(t)]
Envelope: A(t) =| z(t) |
Phase: arg[g(t)] = 2f
0
t + (t)
Instantaneous freq: f
i
= f
0
+
1
2
d
dt
(t)
The Two-Dimensional Fourier Transform
F(s
x
, s
y
) =
_

f(x, y)e
i2(sxx+syy)
dxdy
The Inverse Two-Dimensional Fourier Transform:
f(x, y) =
_

F(s
x
, s
y
)e
i2(sxx+syy)
ds
x
ds
y
The Hankel Transform (zero order):
F(q) = 2
_

0
f(r)J
0
(2rq)rdr
3
The Inverse Hankel Transform (zero order):
f(r) = 2
_

0
F(q)J
0
(2rq)qdq
Projection-Slice Theorem: The 1-D Fourier transform
P

(s) of any projection p

(x

) through g(x, y) is identi-


cal with the 2-D transform G(s
x
, s
y
) of g(x, y), evaluated
along a slice through the origin in the 2-D frequency do-
main, the slice being at angle to the x-axis. i.e.:
P

(s) = G(s cos , s sin )


Reconstruction by Convolution and Backprojection:
g(x, y) =
_

0
F
1
{| s | P

(s)}d
=
_

0
f

(xcos + y sin )d
where f

(x

) = (2s
2
c
sinc2s
c
x

s
2
c
sinc
2
s
c
x

) p

(x

)
compiled by John Jackson
4