1 Fourier Transform
If f L
1
[0, 2] is a periodic function with period 2 then its Fourier coe-
cients are given by
f(n) =
_
2
0
f(t)e
int
dt (1.1)
This gives us the the CTDF Fourier Transform of a function f L
1
[0, 2] as
a function
f : Z C
where the value
f(n) of this function at any point n Z is as dened in
(1.1). We shall now consider functions f(t) dened over R = (, ), i.e.,
f : R R
Now since there is no periodicity of f, the frequencies need not be restricted
over the discrete set Z, but we should take the frequencies to vary over R.
We then can, analogous to (1.1), dene the Continuous Time Continuous
Frequency (CTCF) Fourier Transform as,
f() =
_
f(t)e
it
dt ; (, ) (1.2)
This is called the Fourier Transform of f.
While dening the CTDF Fourier Transform we found that the integral on
the right hand side of (1.1) is well dened if f L
1
[0, 2]. Analogously,
we see that the integral on the right hand side of (1.2) is well dened if
f L
1
(R), i.e., if
_
|f(t)|dt < (1.3)
This is because, we have
f(t)e
it
dt
|f(t)e
it
|dt
=
_
|f(t)|dt since |e
it
| = 1
< (since f L
1
(R))
Thus we see that the Fourier Transform is well dened if f L
1
(R). We
have,
1
Denition 1.1
The Fourier Transform (CTCF) for any function f L
1
(R) is dened as
f() =
_
f(t)e
it
dt
(A signal f(t) satisfying (1.3) is called a STABLE SIGNAL).
2 Properties of L
1
(R) Fourier Transforms
We observe some important properties :
1.
f L
1
(R) =
_
_
i)
f is dened for every R and is a continuous function on R
ii)|
f()|
_
|f(t)|dt R
Thus the Fourier Transform of an L
1
(R) function is a bounded contin-
uous function.
2. Inversion Formula
If f L
1
(R) is such that its Fourier transform
f() is also in L
1
(R),
then the function f(t) can be recovered from its Fourier transform using
the inversion formula,
f(t) =
1
2
_
f()e
it
d (2.4)
3. Fourier transform of the Fourier Transform:
If f L
1
(R) is such that its Fourier transform
f() is also in L
1
(R),
then we can dene the Fourier transform of the Fourier transform. We
have
(
f)() =
_
f(t)e
it
dt
= 2
_
1
2
_
f(t)e
it
dt
_
= 2f() (by the inversion formula (2.4) above)
2
4. Riemann-Lebesgue Lemma:
Recall that when we were dealing with Fourier series and the CTDF
Fourier transform, we observed that
lim
|n|
f(n)
= 0
Similarly in the case of Fourier transforms we have
lim
||
f()
= 0
This is called the Riemann-Lebesgue Lemma.
5. Let f(t) L
1
(R) and dene g(t) = f(t) where is any real number.
Then we have
g() =
_
g(t)e
(it)
dt
=
_
f(t )e
(it)
dt
=
_
f(u)e
(i(u+))
du (by change of variables u = t )
= e
(i)
_
f(u)e
(iu)
du
= e
(i)
f()
Thus the Fourier transform f(t ) is given by e
(i)
f()
6. Let f L
1
(R). Consider the function g(t) = e
(it)
f(t) where is any
real number. Then we have
g() =
_
g(t)e
(it)
dt
=
_
e
(it)
f(t)e
(it)
dt
=
_
f(t)e
(i()t)
dt
=
f( )
Thus the Fourier transform of e
(it)
f(t) is given by
f( )
3
7. For f L
1
(R) dene
f
a
(t) = f(at) where a is a real constant and a = 0
Then we have
_
|f
a
(t)|dt =
_
|f(at)|dt
=
_
|f(u)|du (by changing the variable of integration as u = at)
< (since f L
1
(R))
Thus
f
a
L
1
(R)
and hence we can dene its Fourier transform. We have
Case i: a > 0
f
a
() =
_
f
a
(t)e
(it)
dt
=
_
f(at)e
(it)
dt
=
1
a
_
f(u)e
(i
u
a
)
du (by change of variables at = u)
=
1
a
_
f(u)e
(i
a
u)
du
=
1
a
f
_
a
_
=
1
|a|
f
_
a
_
(since |a| = a when a > 0)
Case ii: a < 0
f
a
() =
_
f
a
(t)e
(it)
dt
=
_
f(at)e
(it)
dt
=
1
a
_
f(u)e
(i
u
a
)
du (by change of variables at = u)
(Note a < 0 in this case)
4
=
1
a
_
f(u)e
(i
a
u)
=
1
a
f
_
a
_
=
1
|a|
f
_
a
_
(since |a| = a when a < 0)
Combining the two cases we have
(For a = 0, the Fourier transform of f(at) is
1
|a|
f
_
a
_
(2.5)
3 Example
Example 3.1 Consider the indicator function, I
of the interval [, ] de-
ned as
I
(t) =
_
1 if t [, ]
0 if t [, ]
(3.1)
The graph of the function is shown below:
1
Figure 1
I
(t)
We have
_
|I
(t)|dt =
_
dt
= 2
<
Thus I
(t) L
1
(R) and hence its Fourier transform is well dened. We have
f() =
_
(t)e
(it)
dt
5
=
_
e
(it)
dt
=
_
e
(it)
i
_
t=
t=
=
e
(i)
e
(i)
i
=
2
e
(i)
e
(i)
2i
= 2
sin()
= 2sinc()
where
sinc(x) =
sin x
x
(3.2)
is the cardinal sine function. Thus we have
() = 2sinc() (3.3)
From the properties of the Fourier transform we mentioned above we get
Fourier transform of I
(t ) is e
(i)
I
() = e
(i)
2sinc() (3.4)
Fourier transform of e
(it)
I
(t) is
I
( ) = 2sinc( ) (3.5)
4 Fourier Plancheral Transform
In the case of CTDF Fourier Transform , we were dealing with functions
dened on a nite interval [0, 2], and in that case whenever f L
2
[0, 2] we
also had f L
1
[0, 2] and hence the CTDF Fourier Transform of signals with
nite energy, i.e. for f L
2
[0, 2], was automatically dened. However,
this process of dening the Fourier Transform of signals (over R) with nite
energy, i.e. for functions f L
2
(R), is not so straight forward since a function
f L
2
(R) need not necessarily be in L
1
(R), i.e., signals with nite energy
need not necessarily be stable.
Let us look at some examples:
6
Consider the function f(t) dened as follows:
f(t) =
_
_
1
(t)
3
4
for 0 < t 1
0 for t outside the interval (0, 1]
(4.1)
We have
_
|f(t)|dt =
_
1
0
dt
(t)
3
4
=
_
4(t)
1
4
_
1
t=0
= 4
<
Hence f L
1
(R). However we have
_
|f(t)|
2
dt =
_
1
0
dt
(t)
3
2
=
_
2(t)
1
2
_
1
t=0
=
<
Hence we have f L
2
(R). Thus this function is in L
1
(R) but not in L
2
(R).
On the other hand consider the function f(t) dened as
f(t) =
_
_
1
(t)
3
4
for 1 t <
0 for t outside the interval [1, )
(4.2)
Then we have
_
|f(t)|dt =
_
1
dt
(t)
3
4
=
_
4(t)
1
4
_
t=1
=
7
Hence f L
1
(R). However we have
_
|f(t)|
2
dt =
_
1
dt
(t)
3
2
=
_
2(t)
1
2
_
t=1
= 2
<
Hence we have f L
2
(R). Thus this function is not in L
1
(R) but in L
2
(R).
Hence if a function f(t) is in L
2
(R) it does not mean that it must necessarily
be in L
1
(R). Further the function I
(t), the indicator function of the interval
[, ] is in both L
1
(R) and L
2
(R). Thus we have the following picture:
L
1
(R) L
2
(R)
L
1
(R)
L
2
(R) L
1
(R) \ L
2
(R) L
2
(R) \ L
1
(R)
Each one of the three pieces above are nonempty. For example the function
f(t) dened in (4.1) is in L
1
(R)\L
2
(R), the function I
(t) is in L
1
(R)
L
2
(R)
and the function f(t) dened in (4.2) is in L
2
(R) \ L
1
(R). We shall now see
the structure of these pieces and how this helps us to dene the Fourier
Pancheral transform of a function in L
2
(R).
STEP 1:
The rst step towards dening the L
2
(R) Fourier transform is to observe that
for those L
2
(R) functions which are also in L
1
(R), that is for those functions
in L
1
(R)
L
2
(R), we can dene the L
1
(R) Fourier transform which we denote
as usual by
f()
STEP 2:
The second step is to show that the L
1
(R) Fourier transform of a function in
8
L
1
(R)
L
2
(R) is in L
2
(R), that is
f L
1
(R)
L
2
(R) =
f() L
1
(R) (4.3)
STEP 3:
The third step is to show that every L
2
(R) function can be approximated
as closely as we want by a function in L
1
(R)
L
2
(R). This amounts to the
following:
f L
2
(R) = a sequence {f
n
} L
1
(R)
L
2
(R) such that f
n
L
2
(R)
f
(4.4)
(The convergence in (4.4) above is the RMS convergence).
STEP 4:
For the sequence {f
n
} we can dene the L
1
(R) Fourier transforms
_
f
n
_
. All
these
f
n
are in L
2
(R) by (4.3) above.
STEP 5:
The next step is to show that this sequence of Fourier transforms converges
in L
2
(R) to a function in L
2
(R), that is
g L
2
(R) such that
f
n
L
2
(R)
g (4.5)
(Once again the convergence above is RMS convergence)
STEP 6:
The next important step is to show that the g we obtained is independent of
the sequence {f
n
} chosen in L
1
(R)
L
2
(R) to converge in L
2
(R) to f (as in
STEP 3 above). This means that
{h
n
} any sequence L
1
(R)
L
2
(R) and h
n
L
2
(R)
f
=
h
n
L
2
(R)
g, the same g as obtained in STEP 5 (4.6)
STEP 7:
It is therefore natural to dene the L
2
(R) Fourier Transform
f of f as this
function g
STEP 8:
If f L
1
(R)
L
2
(R) then we can dene both its L
1
(R) Fourier transform
9
and the L
2
(R) Fourier transform. The next step is to show that these two
are the same
STEP 9:
Since we can choose any sequence in L
1
(R)
L
2
(R) to approximate we now
look at a particular sequence. Given f L
2
(R) dene
f
n
(t) =
_
f(t) for n t n
0 for |t| > n
(4.7)
Then {f
n
} is in L
1
(R)
L
2
(R) and by STEP 6 we have
f
n
L
2
(R)
f
Since
f
n
() =
_
f(t)e
(it)
dt
we write nally the L
2
(R) Fourier transform of f as
f() =
l.i.m
n
_
n
n
f(t)e
(it)
dt (4.7)
(l.i.m stands for the fact that this is in the sense of RMS limit). This is called
the Fourier Plancheral Transform.
5 Properties of L
2
(R) Fourier Transform
We shall now look at some properties of the Fourier-Plancheral transform of
L
2
(R) functions.
1. The Fourier transform of a function in L
2
(R) is also in L
2
(R). We have
f L
2
(R) =
f L
2
(R) (5.1)
2. If f L
2
(R) and
f is its Fourier transform then we can recover f from
f by the INVERSION FORMULA
f(t) =
1
2
_
f()e
(it)
d (5.2)
10
3. Since
f is also in L
2
(R) we can also dene the Fourier transform of the
Fourier transform. We have
f() = 2f() (5.3)
4. From the above we see that
f() = 4
2
f() (5.4)
This means that the Fourier transform F is a transformation
F : L
2
(R) L
2
(R)
which has the property that
F
4
= 4
2
I (5.5)
where I is the identity transformation.
5. Analogous to the L
1
(R) Fourier transform we have the following prop-
erties of L
2
(R) Fourier transform:
Function Fourier Transform
f(t)
f()
e
(it)
f(t)
f( )
f(t ) e
(i)
f()
f(at)
1
|a|
f
_
a
_
a R and a = 0
11
6 Example
Example 6.1 We have seen that the function I
(t), the indicator function
of the interval [, ] is in L
1
(R) and hence we found its L
1
(R) Fourier
transform as (see (3.3))
() = 2sinc()
This function is also in L
2
(R) since
_
|I
(t)|
2
dt =
_
dt
= 2
<
Hence we can have its L
2
(R) Fourier transform also and it will also be the
same function 2sinc() as above. Therefore sinc() is in L
2
(R). Hence
the Fourier transform of the function
1
2
I
(t) is the function sinc(). Using
this fact and the properties of the Fourier transform we get the following
Fourier transforms:
Function Fourier Transform
I
(t) 2sinc()
1
2
I
(t) sinc()
sinc(t) I
()
sinc((t )) e
i
I
()
(where R)
12
Using the above Fourier Transforms, for any two integers m, n Z we have
by Plancherals theorem
_
sinc((t n))sinc((t m))dt =
1
2
_
e
(in)
e
(im)
d
=
1
2
_
e
i(mn)
d
=
_
0 if m = n
1 if m = n
Thus we see that the sequence of functions {sinc((t n))}
nZ
is an or-
thonormal set in L
2
(R)
Example 6.2 Let a > 0 be a positive real number. Consider the function
F
+
a
(t) dened as
F
+
a
(t) =
_
_
e
at
for t > 0
1 for t = 0
0 for t < 0
(6.1)
We have
_
|f(t)|dt =
_
0
e
at
dt
=
_
e
at
a
t=0
_
=
1
a
<
Thus F
+
a
L
1
(R). Similarly we have
_
|f(t)|
2
dt =
_
0
e
2at
dt
=
_
e
2at
2a
_
t=0
=
1
2a
<
13
Thus F
+
a
L
2
(R). Hence we have F
+
a
L
1
(R)
L
2
(R). So we can dene
both its L
1
(R) and L
2
(R) Fourier Transforms and both will be same. Let us
nd the Fourier Transform.
F
+
a
() =
_
F
+
a
(t)e
(it)
dt
=
_
0
e
at
e
(it)
dt
=
_
0
e
(a+i)t
dt
=
_
e
(a+i)t
(a + i)
_
t=0
=
1
a + i
Next, consider the function F
a
(t) dened as
F
a
(t) =
_
_
0 for t > 0
1 for t = 0
e
at
for t < 0
(6.2)
We have
_
|f(t)|dt =
_
0
e
at
dt
=
_
e
at
a
_
0
t=
=
1
a
<
Thus F
a
L
1
(R). Similarly we have
_
|f(t)|
2
dt =
_
0
e
2at
dt
=
_
e
2at
2a
_
0
t=
=
1
2a
<
14
Thus F
a
L
2
(R). Hence we have F
a
L
1
(R)
L
2
(R). So we can dene
both its L
1
(R) and L
2
(R) Fourier Transforms and both will be same. Let us
nd the Fourir Transform.
a
() =
_
a
(t)e
(it)
dt
=
_
0
e
at
e
(it)
dt
=
_
0
e
(ai)t
dt
=
_
e
(ai)t
(a i)
_
0
t=
=
1
a i
We shall next consider the functions
h
+
a
(t) = aF
+
a
(t) (6.3)
=
_
_
ae
at
for t > 0
a for t = 0
0 for t < 0
(6.4)
and
h
a
(t) = aF
a
(t) (6.5)
=
_
_
0 for t > 0
a for t = 0
ae
at
for t < 0
(6.6)
We then have their Fourier transforms as
h
+
a
() =
a
a + i
(6.7)
a
() =
a
a i
(6.8)
Now consider the function
H
a
(t) =
h
+
a
(t) + h
a
(t)
2
(6.9)
=
_
_
ae
at
2
for t > 0
a for t = 0
ae
at
2
for t < 0
(6.10)
15
We have its Fourier transform given by
H
a
() =
1
2
_
a
a + i
+
a
a i
_
(6.11)
=
a
2
a
2
+
2
(6.12)
We see that as a the function H
a
(t) approaches the delta function
and its Fourier transform
H
a
() approaches 1. Hence we dene the Fourier
transform of the delta function as 1
Example 6.3 Consider the function
f(t) = e
t
2
We shall do some formal calculations to get the Fourier transform of this
function. We have,
f() =
_
e
t
2
e
it
dt
Dierentiating both sides with respect to and interchanging dierentiation
and integration on the right hand side, we get,
d
d
f() =
_
e
t
2
e
it
(it)dt
=
_
e
it
_
e
t
2
(2t)
_
i
2
__
dt
=
i
2
_
e
it
_
2te
t
2
_
dt
which by integration by parts becomes,
d
d
_
f()
_
=
i
2
_
e
t
2
e
it
_
t=
t=
e
t
2
(i)e
it
dt
=
i
2
(i)
_
e
t
2
e
it
dt
=
2
f()
Solving this dierential equation for
f() we get
f() = Ae
2
4
16
where A is a constant. Now we have,
f(0) =
_
e
t
2
dt
=
_
Using this above we get
A =
_
Hence we get nally,
f() =
_
2
4
7 Sampling Theorem
Let f(t) L
2
(R) be such that
f() = 0 for || > (7.1)
(Note that using (7.1) we get
f() L
2
(R) and hence in L
2
[, ) and hence
L
1
[, ] and hence in L
1
(R). Hence its Fourier transform namely 2f()
and consequently F is continuous at all points in R).
Consider the 2 periodic function F(t) dened as
F(t) =
f(t) for t < (7.2)
Note that
F(t) is equal to
f(t) for t < (7.3)
Since F(t) is periodic we can expand this in a Fourier series as
F(t) =
1
2
n=
F
n
e
(int)
(7.4)
where
F
n
=
_
F(t)e
(int)
dt (7.5)
17
Now using (7.3) in (7.5) we get
F
n
=
_
f(t)e
(int)
dt
=
_
f()e
(in)
d
=
_
f()e
(in)
d (using (7.1))
= 2
_
1
2
_
f()e
(in)
d
_
= 2f(n) (by the inversion formula) (7.6)
Thus we have
F
n
= 2f(n) (7.7)
Hence substituting in (7.4) we get
F(t) =
1
2
n=
2f(n)e
(int)
=
n=
f(n)e
(int)
=
n=
f(n)e
(int)
Since this is valid for t [, ), using (7.3) we get
f() =
n=
f(n)e
(in)
for in [, ) (7.8)
Now by the inversion formula we have
f(t) =
1
2
_
f()e
(it)
d
=
1
2
_
f()e
(it)
d (by (7.1))
=
1
2
_
n=
f(n)e
(in)
_
e
(it)
d (by (7.11))
18
Interchanging the integral and summation we get
f(t) =
1
2
n=
f(n)
__
e
(i(tn))
d
_
=
1
2
n=
f(n)
_
e
(i(tn))
i(t n)
_
=
=
=
1
2
n=
f(n)
e
i(tn)
e
i(tn)
i(t n)
=
n=
f(n)
e
i(tn)
e
i(tn)
2i(t n)
=
n=
f(n)
sin((t n))
(t n)
=
n=
f(n)sinc((t n))
Thus we have
Theorem 7.1 If f L
2
(R) is such that it satises (7.1) then
f(t) =
n=
f(n)sinc((t n)) (7.10)
(The convergence above being in L
2
(R) sense)
Let > 0 and consider a function,
f(t) L
2
(R) such that
f() = 0 for || > (7.11)
We then dene
g(t) = f
_
t
_
(7.12)
Then we have
g() =
f
_
_
(7.13)
19
By (7.14) we get that
g() = 0 if
> that is if
|| >
Hence by Theorem 7.1 above we get
g(t) =
n=
g(n)sinc((t n)) (7.14)
Hence we get from (7.15) and (7.17)
f(t) = g
_
t)
_
=
n=
g(n)sinc
_
t n
__
=
n=
g(n)sinc (t n)
=
n=
f
_
n
_
sinc (t n)
Thus we have
Theorem 7.2 Sampling Theorem
If f L
2
(R) and
f() = 0 for || > , (where > 0), then
f(t) =
n=
f
_
n
_
sinc (t n) (7.15)
20
