SIGNALS AND SYSTEMS USING MATLAB
Chapter 5 Frequency Analysis: The Fourier Transform
Luis F. Chaparro
From the Fourier Series to the Fourier Transform
Aperiodic signal x(t) can be thought of as periodic signal x(t) with infinite fundamental
period. From Fourier series of x(t) and limiting process we obtain Fourier transform pair
x(t)
X ()
x(t) is transformed into X () in the frequencydomain by the
Z
Fourier transform:
X () =
x(t)e jt dt
while X () is transformed into x(t) in the timedomain by the
Z
1
Inverse Fourier Transform:
x(t) =
X ()e jt d
2
2 / 25
Existence of the Fourier Transform
For X () to exist, x(t) must be absolutely integrable
|X ()|
|x(t)e
jt
Z
|dt =
|x(t)|dt <
ROC of X (s) = L[x(t)] contains the j-axis then
F[x(t)] = L[x(t)]|s=j =
x(t)e jt dt
= X (s) |s=j
Duality between time and frequency allows computation of Fourier transforms
3 / 25
Example: Fourier transform from Laplace transform
1
(a) x1(t) = u(t), X1(s) = , ROC : > 0, j-axis not included
s
X () cannot be obtained
(b) x2(t) = e 2t u(t),
X2() =
X2(s) =
1
, ROC : > 2
s +2
1
1
|s=j =
s +2
j + 2
(c) x3(t) = e |t|,
1
1
+
, ROC : 1 < < 1
s + 1 s + 1
2
2
=
=
1 (j)2 1 + 2
X3(s) =
X3() = X3(s)|s=j
4 / 25
Inverse proportionality of time and frequency
Support of X () is inversely proportional to the support of x(t)
If x(t) has a Fourier transform X () and 6= 0 is a real number, then x(t)
is a contracted signal when > 1;
is a contracted and reflected signal when ( < 1);
is an expanded signal when 0 < < 1;
is a reflected and expanded signal when 1 < < 0; or
is a reflected signal when = 1
and
1
X
x(t)
||
5 / 25
1
0.8
0.6
0.6
x2(t)
x1(t)
1
0.8
0.4
0.2
0.2
0
t
X2()
X1()
0.4
0
40
30
20
10
10
20
30
40
1
50
50
0
t
1
50
50
Fourier transform of pulses x1 (t) = u(t + 0.5) u(t 0.5), (left) and x2 (t) = u(t + 2) u(t 2)
(right). Notice the wider the pulse the more concentrated in frequency its Fourier transform
6 / 25
Example: x(t) = u(t) u(t 1) vs x1(t) = x(2t)
x(t), x1(t)
1
x(t)
x1(t)
0.5
0
0
0.2
0.4
0.6
0.8
|X()|,|X1()
0.5
0
50
|X()|
|X1()|
50
1 e s
X (s) =
, ROC : whole s-plane
s
e j/2 (e j/2 e j/2 ) sin(/2) j/2
X () =
=
e
infinite support
2j/2
/2
x1 (t) = x(2t) = u(2t) u(2t 1) = u(t) u(t 0.5)
e j/4 (e j/4 e j/4 ) 1 sin(/4) j/4 1
=
e
= X (/2)
X1 () =
j
2 /4
2
7 / 25
Duality
x(t)
X (t)
X ()
2x()
Example:
A(t)
2A() = 2A()
Example:
(t 0) + (t + 0)
2 cos(0t)
x(t) = cos(0t)
e j0 + e j0 = 2 cos(0)
2[( + 0) + ( 0)]
X () = [( + 0) + ( 0)]
8 / 25
10
10
X1()
x1(t)
8
6
4
2
0
1
0.5
0
t
0.5
100
50
50
100
0.5
0.5
60
X()=2x1()
x(t)=X1(t)
10
0
100
50
0
t
50
100
40
20
0
1
Duality to find Fourier transform of x(t) = 10sinc(0.5t)
9 / 25
Modulation
Frequency shift:
x(t)
x(t)e j0t
X ()
X ( 0)
Modulation:
modulated signal x(t) cos(0t)
0.5 [X ( 0) + X ( + 0)]
y1(t)
1
0.5
0
0.5
5
t (sec)
1
<Y1()
|Y1()|
1
0.8
0.6
0.4
0.2
0
0
1
20
20
20
20
Modulated signal y1 (t) = e |t| cos(10t), its magnitude and phase spectra
10 / 25
Fourier transform of periodic signals
Represent periodic signal x(t), of period T0, by its Fourier series:
X
X
jk0 t
x(t) =
Xk e
X () =
2Xk ( k0)
k
Example: Periodic x(t) with period x1(t) = r (t) 2r (t 0.5) + r (t 1), fundamental
frequency 0 = 2
e 0.5s 0.5s
1
0.5s
s
0.5s
e
+e
=
2+e
X1(s) = 2 1 2e
s
s2
Fourier coefficients :
2
1
k sin (k/2)
Xk = X1(s)|s=j2k = (1)
, k 6= 0, X0 = 0.5
T0
2k 2
X
X () = 2X0() +
2Xk ( 2k)
k=,6=0
11 / 25
Parsevals energy relation
For aperiodic signal x(t) with energy Ex < :
Energy conservation in time and frequency
Z
Z
1
Ex =
|X ()|2d
|x(t)|2dt =
2
|X ()|2 energy density: energy at each of the frequencies . Plot |X ()|2 vs is
called the energy spectrum of x(t), and displays how the energy of the signal is
distributed over frequency
Example: Impulse x(t) = (t) is not finite energy signal
X () = F[(t)]
Z =1
1
Ex =
|X ()|2d
2
12 / 25
Symmetry of spectral representations
x(t) realvalued signal
X () = F[x(t)] = |X ()|e jX () = Re[X ()] + jIm[X ()]
|X ()| = |X ()|,
X () = X (),
Re[X ()] = Re[X ()] (even functions of )
Im[X ()] = Im[X ()] (odd functions of )
Spectra
|X ()| vs
X () vs
|X ()|2 vs
Magnitude Spectrum
Phase Spectrum
Energy/Power Spectrum.
13 / 25
Example:
(a) x1 (t) = u(t) u(t 1), let z(t) = x1 (t + 0.5)
sin(/2)
Z () =
(real)
/2
X1 () = e j0.5 Z ()
sin(/2)
|X1 ()| =
/2
X1 () = Z () 0.5 =
(b) x2 (t) = e t u(t), X2 () =
|X2 ()| =
1 ,
1+2
0.5
Z () 0
0.5 Z () < 0
1
1 + j
(X2 ()) = tan1 ,
14 / 25
x1(t)
0.8
0.6
0.4
0.2
0
0.2
t (sec)
3
<X1()
|X1()|
0.8
0.6
0.4
0.2
0
1
2
0
0.2
3
20
20
20
Pulse x1 (t) = u(t) u(t 1) and its magnitude and phase spectra.
15 / 25
20
Convolution and filtering
Input x(t) (periodic or aperiodic) of stable LTI system has Fourier transform X ()
system has frequency response H(j) = F[h(t)], h(t) impulse response
output is convolution integral y (t) = (x h)(t), with Fourier transform
Y () = X () H(j)
If input x(t) is periodic the output is also periodic of the same fundamental period,
and with Fourier transform
Y () =
2 Xk H(jk 0)( k0)
k=
where {Xk } are the Fourier series coefficients of x(t) and 0 its fundamental
frequency.
16 / 25
Example: Windowing
rectangular window w(t) = u(t + ) u(t ), > 0
windowed signal y (t) = w(t)x(t)
Z
Z
1
1
y (t) = w(t)
X ()e jt d =
X ()w(t)e jt d
2
2
| {z
}
x(t)
Z
Z
1
1
Y () =
X ()F[w(t)e jt ]d =
X ()W( )d
2
2
y (t) = x(t)w(t)
1
2 sin()
convolution of X () and W () =
2
17 / 25
Ideal filtering
Filtering: to pass desired frequency component and to attenuate undesirable components
Ideal filters: (top-left clockwise) low-pass, band-pass, band-eliminating and high-pass
18 / 25
Issues with ideal filters:
Noncausal
Paley-Wiener integral condition causal and stable filter with frequency response H(j) should
satisfy small
Z
| log(H(j))|
d <
1 + 2
Example: Gibbs phenomenon
Passing x(t) through ideal low-pass filter
1 c c , N0 < c < (N + 1)0
H(j) =
0 otherwise
X
X () =
2Xk ( k0 )
k=
The output of the filter with 2N + 1 Fourier coefficients
"
xN (t) = F 1 [X ()H(j)] = F 1
N
X
#
2Xk ( k0 )
k=N
= [x h](t),
h(t) sinc function
Convolution around the discontinuities of x(t) causes ringing before and after them, independent of
the value of N
19 / 25
Example: RLC circuit, R = 1 , L = 1 H, and C= 1 F, and IC zero
+ vR (t)
+
L
+ vL (t)
C
vi (t)
vC (t)
1
VC (s)
= 2
lowpass: output vc (t), Hlp (s) =
Vi (s)
s +s +1
VL(s)
s2
highpass: output vL(t), Hhp (s) =
=
Vi (s) s 2 + s + 1
VR (s)
s
bandpass: output vR (t), Hbp (s) =
= 2
Vi (s)
s +s +1
VcL(s) s 2 + 1
bandstop: output vcL(t), Hbs (s) =
Vi (s) s 2 + s + 1
20 / 25
Frequency Response from Poles and Zeros
G (s) = K
s z
,
s p
zero z, pole p, gain K 6= 0
s-plane
j0
P! (0 )
! 0)
Z(
Frequency response of G (s) at frequency 0
~ (0 )
~
Z
~ (0 )P(
~ 0 ))
jK |Z (0 )| j(Z
G (j0 ) = K
= |K |e
e
.
~ 0)
~ 0 )|
P(
|P(
~ (0 )|
|Z
~ 0 )|
|P(
~ (0 ) P(
~ 0)
Phase response G (j0 ) = K + Z
Magnitude response |G (j0 )| = |K |
21 / 25
Example: Frequency response of high-pass filter
Vr (s)
s
=
Vs (s) s + 1
~ ()
j
Z
=
H(j) =
~
1 + j P()
~ () from s = 0 to j
vector Z
~
vector P()
from s = 1 to j
H(s) =
~ ()
Z
~
P()
~ ()/P()
~
H(j) = Z
0e j/2
1e j0
0e j/2
j/2
1e
e j/2
2e j/4
e j/2
22 / 25
0.707e j/4
1e j0
Spectrum analyzer
LP F
BP F1
x(t)
Power
measurement
Power
measurement
..
.
BP FN
Px (0)
Px (1 )
..
.
Power
measurement
Px (N )
Bank-of-filter spectrum analyzer: the frequency response of the bankoffilters is that of an allpass
filter covering the desired range of frequencies
23 / 25
Basic Properties of the Fourier Transform
1
Expansion/contraction
x(t), 6= 0
X
||
Reflection
x(t)
X ()
Z
Z
1
2
Parsevals
Ex =
|X ()|2 d
|x(t)| dt
Ex =
2
Duality
X (t)
2x()
d n x(t)
Differentiation
, n1
(j)n X ()
n
Z dtt
X ()
+ X (0)()
Integration
x(t 0 )dt 0
j
Shifting
x(t ), e j0 t x(t)
e j X (), X ( 0 )
Modulation
x(t) cos(c t)
0.5[X ( c ) + X ( + c )]
X
X
Periodic
x(t) =
Xk e jk0 t
X () =
2Xk ( k0 )
k
Symmetry
Convolution
x(t) real
z(t) = [x y ](t)
24 / 25
|X ()| = |X ()|,
X () = X ()
Z () = X ()Y ()
Fourier Transform Pairs
cos(0 t),
<t <
1, e j
1
1
+ (),
+ ()
j
j
2
j
A
2A(),
j + a
A
(j + a)2
2a
a 2 + 2
[( 0 ) + ( + 0 )]
sin(0 t),
<t <
j[( 0 ) ( + 0 )]
(t), (t )
u(t), u(t)
sgn(t) = 2[u(t) 0.5]
A, Ae at u(t), a > 0
Ate at u(t), a > 0
e a|t| , a > 0
p(t) = A[u(t + ) u(t )],
25 / 25
2A
sin( )