EE5323 Signals and Systems

Fourier Transform
Mohamed A. Taha
c PSUT, July 19, 2008
1 Fourier Transform
In chapter 3, we have discussed the Fourier series representation. We showed
that the Fourier series representation is usually performed for periodic signals.
Generally, we need models (equations) to represent non-periodic signals
such as speech signals. For example, a single token word is a non-periodic
signal because it does not repeat itself. The way we treat it as a periodic
signal with period innity.
If we assume T , then we can produce equations for non-periodic
signals
Recall, for periodic signal, we had
a
k
=
1
T
_
T
x(t)e
jk
0
t
dt (1)
and
x(t) =

k=
a
k
e
jk
0
t
(2)
If T
0
=
2
T
0. In this case, we can replace
0
with d,
which represents a very small (inntesemal) value of
lim
T
1
T
=
d
2
1
Now, k
0
can also be replaced by which is a continuous frequency
variable. If we wok with the analysis equation (1)
a() =
d
2

x(t)e
jt
dt
If we divide both sides by
d
2
2
a()
d
=

x(t)e
jt
dt
Dene X() = 2
a()
d

X() =

x(t)e
jt
dt
which is called the Fourier Transform of non - periodic signal x(t). To
recover the signal x(t) from X()
x(t) =
1
2

X()e
jt
d
which is called the Inverse Fourier Transform.
Usually, we use the following shorthand notation for Fourier transform
pair
X() = F {x(t)} =

x(t)e
jt
dt
x(t) = F
1
{X()} =
1
2

X()e
jt
d
Example 1.1 Let x(t) = e
at
u(t), a > 0
X() =

x(t)e
jt
dt
=

_
0
e
(a+j)t
=
1
j +a

So, we can write
e
at
u(t)
FT

1
j +a
Note that X() is a complex valued function, and so, it has magnitude
and phase
|X()| =
_
1

2
+a
2
{X()} = tan
1
(

a
)
We can re write X() in polar form as
X() =
_
1

2
+a
2
e
j tan
1
(

a
)
Example 1.2 Let x(t) = rect(t)
X() =

x(t)e
jt
dt
=
1
2
_

1
2
e
jt
dt
=
1
j
_
e

1
2

1
2
= sinc(

2
)

rect(t) sinc(

2
)
Note that in example 1.2, we used the relation
sinc(x) =
sin(x)
x
If we set substitute x with
x

, gives
sinc(
x

) =
sin(x)
x
Example 1.3 Let x(t) = (t)
X() =

x(t)e
jt
dt
=

(t)e
jt
dt
= 1
It should be noted in example 1.3, that
T
2
_
T
1
x(t)(t t
0
)dt =
_
x(t
0
) T
1
< t
0
< T
2
0 Otherwise
2 Fourier Transform of LTI Systems
Consider an LTI system with impulse response h(t). If an input signal x(t)
is applied to the system with corresponding output y(t).
y(t) = x(t) h(t) =

x()h(t )
Y () =

y(t)e
jt
dt
=

x()h(t )de
jt
dt
=

x()

h(t )dtd
=

x()e
j

h(t )e
j(t)
dt
. .
H()
d
=

x()e
j
H()d
= X()H()
This means that convolution in time domain is multiplication in frequency
domain
x(t) h(t) Y () = X()H()
H() is called the system transfer function or the system frequency response.
Consider the cascaded system shown in Figure 1
Figure 1: Cascaded system
In this system
y(t) = x(t) h
1
(t) h
2
(t)
Y () = X() H
1
()H
2
()
. .
H()
where
X() = F {x(t)}
H
1
() = F {h
1
(t)}
H
2
() = F {h
2
(t)}
The overall frequency response is H() with magnitude and phase
|H()| = |H
1
()| |H
2
()|
H() = H
1
() +H
2
()
3 Properties of the Fourier Transform
1. Linearity
If
x
1
(t) X
1
()
x
2
(t)
FT
X
2
()
Then
x
1
(t) +x
2
(t)
FT
X
1
() +X
2
()
2. Time Reversal If
x(t) X()
Then
x(t)
FT
X()
3. Time Shift
If
x(t)
FT
X()
Then
x(t t
0
)
FT
X()e
jt
0
4. Time Scaling
x(at)
FT

1
|a|
X(

a
)
5. Conjugate Symmetry
Assume x(t) is real with
x(t) X()
Then
X() = X

()
6. Duality
Assume that
g(t)
FT
f()
Then
f(t) 2g()
7. Frequency Shift
If
x(t)
FT
X()
Then
x(t)e
j
0
t
FT
X(
0
)
8. Time Domain Multiplication If
x(t) X()
y(t)
FT
Y ()
Then
x(t)y(t)
FT

1
2
X() Y ()
9. Time Domain Convolution
x(t)y(t)
FT
X()Y ()
10. Time Domain Dierentiation If
x(t) X()
Then
dx(t)
dt
FT
jX()
11. Parsevals Theorem If
x(t) X()

|x(t)|
2
dt
. .
Time domain energy
=
1
2

|X()|
2
d
. .
Frequency domain energy
Example 3.1 Let x(t) = (t T)
If we nd the Fourier Trnsform for (t)
(t) 1
Applying time shift property
(t T)
FT
1 e
jT
Example 3.2 Let x(t) = e
a|t|
The signal x(t) can be written as
x(t) = e
at
u(t) +e
at
u(t)
e
at
u(t)
FT

1
j +a
Using time reversal property we can nd
e
at
u(t)
FT

1
j +a
Applying linearity, we have
e
a|t|
FT

1
j +a
+
1
j +a
=
2a

2
+a
2
Example 3.3 Let x(t) = rect(
tT
P
)
rect(t)
FT
sinc(

2
)
x(t) is a rect(t) signal, time scaled by P and shifted to the right by time T
rect(
t
P
)
FT
Psinc(
P
2
)
rect(
t T
P
)
FT
Psinc(
P
2
)e
jT
Fourier transform pairs for important functions are shown in table 1
4 Continuous Time Fourier Transform (CTFT)
of Periodic Signals
Consider the CTFT
X() =

k=
2a
k
( k
0
)
Note: This is the CTFT of the periodic signal x(t) with Fourier series
coecients a
k
.
The magnitude CTFT of the signal is shown in Figure 2
Figure 2: Magnitude CTFT of the periodic signal
The inverse CTFT is given by
x(t) =
1
2

k=
X()e
jt
d
Substitute X() in the above expression gives
x(t) =

k=
a
k
e
jk
0
t
which represents a Fourier series expansion for periodic signal x(t).
Notes: CTFT of Periodic Signals
Forms impulses in frequency domain.
Impulses are spaced at fundamental frequency
0
Magnitude of the impulses is proportional to a
k
It is called a Comb structure
Example 4.1 Let x(t)
FT
X(). We need to nd Fourier transform of
y(t) = x(t)

k=
(t kT)
Y () = F
_
x(t)

k=
(t kT)
_
=
1
2
X() F
_

k=
(t kT)
_
=
1
2
X()
_

k=
( k
0
))
_
=

0
2
X()

k=
( k
0
)
=
1
T

k=
X( k
0
)
This is a periodic replication of X() with period T =

0
2
.
Assume that X() is limited to ||

0
2
as shown in Figure reflimited

0
0
/2

0
/2
|X()|
Figure 3: Limited X()
Then
1
T

k=
X( k
0
)
is shown in Figure 4

0
/2
0
/2 3
0
/2 3
0
/2
... ...
Figure 4:
1
T

k=
X( k
0
)
As shown in the previous Figure, there is no overlap between the replicated
spectra, hence, there is no loss of information about x(t). This is called the
ideal sampling.
Example 4.2 Frequency selective ltering is accomplished with an LTI sys-
tem whose frequency response H(j) passes the desired range of frequencies
and signicantly attenuates frequencies outside the desired range. For ex-
ample, consider the ideal low pass lter shown in Figure 5 with frequency
response
H(j) =
_
1 || <
c
0 || >
c
If we look at this lter in the time domain, we can say that the lter (Low
pass) passes signals with frequencies less than or equal
c
, thats why
c
is
called the cuto frequency of the lter.
The impulse response (inverse Fourier transform) of the lter is shown
in Figure 6 and is given by
h(t) =
sin
c
t
t
In frequency domain, this lter is ideal since it has innite slope to cut
the frequencies beyond |
c
|. On the other hand, in time domain, the lter

c

c

0
H(j)
passband stopband stopband
Figure 5: Lowpass Filter H(j)
h(t)
t

c
/
/
c
/
c

Figure 6: Lowpass Filter impulse response h(t)
has non-zero impulse response for t < 0, and hence non-causal. In practice,
this lter (ideal lter) should be avoided. In addition, the impulse response
contains oscillation which is undesirable in most applications. Finally, there
is a trade-o between frequency and time domains, as if we need good fre-
quency characteristic, there should be some scaries in time characteristics.
A compromise between frequency and time is the solution.
Example 4.3 Consider the LTI system with impulse response
h(t) = e
t
u(t)
The frequency response of this system is
H(j) =
1
j + 1
This system can be implemented with simple RC Circuit. The impulse re-
sponse and magnitude of the frequency response are shown in Figure 7. This
lter does not have strong frequency selectivity as the ideal lter discussed
above, but it is causal, and has an impulse response that decays monotoni-
cally (without oscillations).
Example 4.4 Consider the response of an LTI system with impulse response
h(t) = e
at
u(t), a > 0
to the input signal
x(t) = e
bt
u(t), b > 0
Rather than computing y(t) = x(t) h(t) directly, let us transform the
problem into the frequency domain.
X(j) =
1
b +j
and
H(j) =
1
a +j
Note that Y (j) = X(j)H(j)
0 1 2 3 4 5 6
0
0.2
0.4
0.6
0.8
1
4 3 2 1 0 1 2 3 4
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Impulse Response
Time (sec)
A
m
p
l
i
t
u
d
e
|H(j)|

Figure 7: Non-Ideal Lowpass Filter
Y (j) =
1
(a +j)(b +j)
To determin the output y(t) we just nd F
1
{Y (j)}
If a = b, we can perform the Partial Fraction Expansion as follows
Y (j) =
1
(a +j)(b +j)
=
A
a +j
+
B
b +j
To nd A
A = [(a +j)Y (j)]
j=a
=
1
b a
To nd B
B = [(b +j)Y (j)]
j=b
=
1
a b
Y (j) =
1
ba
a +j
. .
1
ba
e
at
u(t)
+
1
ab
b +j
. .
1
ab
e
bt
u(t)
y(t) =
1
b a
_
e
at
e
bt
_
u(t)
5 Linear Constant-Coecients Dierential Equa-
tions
Generally the dierential equation can expressed as
N

k=0
a
k
d
k
y(t)
dt
k
=
M

k=0
b
k
d
k
x(t)
dt
k
Applying Fourier transform to both sides
N

k=0
a
k
(j)
k
Y () =
M

k=0
b
k
(j)
k
X()
Dene H() =
Y ()
X()
H() =
M

k=0
b
k
(j)
k
N

k=0
a
k
(j)
k
Which is a rational function of j
Rational Function H(j) =
NUM(j)
DEN(j)
Example 5.1 Consider a stable LTI system characterized by the dierential
equation
dy(t)
dt
+ay(t) = x(t), a > 0
Taking the Fourier transform of both sides
(j)Y (j) +aY (j) = X(j)
The frequency response H(j) =
Y (j)
X(j)
H(j) =
1
j +a
To nd the impulse response of the system h(t), we take the inverse
Fourier transform of H(j)
h(t) = e
at
u(t)
5.1 Soultion using Matlab
In Matlab :
NUM can be dened as a polynomial coecients in j in descending
order.
DEN can be dened as a polynomial coecients in j in descending
order
H = freqs(NUM,DEN,w); is used to compute H(j) numerically.
h = impulse(NUM,DEN); is used to obtain the impulse response h(t)
numerically.
In the previous example:
H(j) =
1
j +a
Let a = 0.5
System function H(j) and impulse response computation using Matlab
The output of this script is shown in Figure 8
4 3 2 1 0 1 2 3 4
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10 12
0
0.2
0.4
0.6
0.8
Impulse Response
Time (sec)
A
m
p
l
i
t
u
d
e
0.5/(j+0.5)
h(t)=e
0.5 t
u(t)
H(j)=
H(j)
Figure 8: H(j) and h(t) using Matlab
6 Homework
1. Use the Fourier transform analysis equation (1) to calculate the Fourier
transforms of:
(a) e
2(t1)
u(t 1)
(b) e
2|t1|
(c) (t + 1) + (t 1)
(d)
d
dt
{u(2 t) +u(t 2)}
Sketch and label the magnitude of each Fourier transform.
2. Determine the Fourier transform of each of the following periodic sig-
nals:
(a) sin(2t +

4
)
(b) 1 + cos(6t +

8
)
3. Use the Fourier transform synthesis equation (2) to determine the in-
verse Fourier transforms of :
(a) X
1
(j) = 2() +( 4) +( + 4)
(b) X
2
(j) =
_

_
2 0 2
2 -2 0
0 || > 2
4. Use the Fourier transform synthesis equation to determine the inverse
Fourier transform of X() = |X(j)| e
X(j)
, where
|X(j)| = 2 {u( + 3) u( 3)}
X(j) =
3
2
+
Use your answer to determine the values of t for which x(t) = 0.
5. Given that x(t) has the Fourier transform X(j), express the Fourier
transform of the signals listed below in terms of X(j)
(a) x
1
(t) = x(1 t) +x(1 t)
(b) x
2
(t) = x(3t 6)
(c) x
3
(t) =
d
2
dt
2
x(t 1)
6. Consider the signal
x(t) =
_

_
0 t <
1
2
t +
1
2

1
2
t
1
2
1 t >
1
2
(a) Use the dierentiation and integration properties and the Fourier
transform pair for the rectangular pulse to nd a closed-form ex-
pression for X(j).
(b) What is the Fourier transform of g(t) = x(t)
1
2
.
(a) Use the Fourier transform properties to help determine the Fourier
transform of the following signal
x(t) = t
_
sin t
t
_
2
7. Use Parsevals relation and the result of the previous part to determine
the numerical values of
A =

t
2
_
sin t
t
_
4
dt
8. Given the relationships
y(t) = x(t) h(t)
and
g(t) = x(3t) h(3t)
and given that x(t) has Fourier transform X(j) and h(t) has Fourier
transform H(j), use Fourier transform properties to show that g(t)
has the form of
g(t) = Ay(Bt)
Determine the values of A and B.
9. Let x(t) be a signal whose Fourier transform is
X(j) = () + ( ) +( 5)
and let
h(t) = u(t) u(t 2)
(a) Is x(t) periodic ?
(b) Is x(t) h(t) periodic?
(c) Can the convolution of two aperiodic signals be periodic?
10. Consider a signal x(t) with Fourier transform X(j). Suppose we are
given the following facts:
(a) x(t) is real and nonnegative
(b) F
1
{(1 +j)X(j)} = Ae
2t
u(t) , A is independent of t
(c)

|X(j)|
2
d = 2
Determine a closed-form expression for x(t).
11. Consider a causal LTI system with frequency response
H(j) =
1
j + 3
For a particular input x(t) this system is observed to produce the output
y(t) = e
3t
u(t) = e
4t
u(t)
Determine x(t) and plot it. Use Matlab to obtain x(t) and plot the
obtained result. Compare the obtained results.
x(t) X()
1 2()
u(t) () +
1
j
(t) 1
(t t
0
) e
jt
0
rect(
t

) sinc(

2
) =
2 sin

2

sinc(

B
t

) rect(

2
B
)
e
j
0
t
2(
0
)

k=
a
k
e
jk
0
t
2

k=
X( k
0
)
cos
0
t [(
0
) +( +
0
)]
sin
0
t

j
[(
0
) ( +
0
)]
cos
0
t rect(
t

) sinc
(
0
)
2
e
at
u(t), {a} > 0
1
j+a
te
at
u(t), {a} > 0 (
1
j+a
)
2
t
n1
(n1)!
e
at
u(t), {a} > 0
1
(a+j)
n
e
a|t|
, a > 0
2a
a
2
+
2
|t| e
a|t|
, {a} > 0
4aj
a
2
+
2
1
a
2
+t
2
, {a} > 0

a
e
a||
t
a
2
+t
2
, {a} > 0
je
a||
2a
e
at
2
, a > 0
_

a
e

2
4a
(
t

) sinc
2
2

k=
(t kT)
2
T

k=
(
2k
T
)
Table 1: Fourier Transform Pairs