Signals and Systems

Spring 2008
Lecture #5
(2/20/2008)

• Covers O & W pp. 177-201

• Complex Exponentials as Eigenfunctions of LTI Systems
• Fourier Series representation of CT periodic signals
• How do we calculate the Fourier coefficients?
• Convergence and Gibbs’ Phenomenon
“Figures and images used in these lecture notes by permission,
copyright 1997 by Alan V. Oppenheim and Alan S. Willsky”

The eigenfunctions φk(t) and their properties

(Focus on CT systems now, but results apply to DT systems as well.)

!k (t) System " k (t)

{k !{

eigenvalue eigenfunction

Eigenfunction in → same function out with a “gain”

From the superposition property of LTI systems:

x(t) = " ak !k (t) LTI y(t ) = " # k ak! k (t)

k k

Now the task of finding response of LTI systems is to determine λk.

The solution is simple, general, and insightful.

2
 Two Key Questions

1. What are the eigenfunctions of a general LTI system?

2. What kinds of signals can be expressed as

superpositions of these eigenfunctions?

A specific LTI system can have more than one type of eigenfunction
Ex. #1: Identity system

x(t) ! (t) x(t) " ! (t) = x(t)

Any function is an eigenfunction for this LTI system.

Ex. #2: A delay

x(t) ! (t " T) x(t " T )

Any periodic function x(t) = x(t+T) is an eigenfunction.

Ex. #3: h(t) even

cos !t h(t) = h("t) y(t)
$ $
y(t) = %#$ h(! ) cos[" (t # ! )]d! = % h(! )[cos"t & cos"! + sin "t & sin"! )]d!
#$
$
= cos "t h(! )cos "!d!
%14 — cos"t is an eigenfunction
#$
442444 3
H( j" )
4
Complex Exponentials are the only Eigenfunctions of any LTI Systems

+#
x(t) = e
st
h(t) y(t) = $ h(! )e s( t "! )d!
"#

= [$ +#

"# ]
h(! )e "s! d! est

e st
H (s){
={

eigenvalue eigenfunction

"
x[n] = z n h[n] y[n] = # h[m]z n!m
m= !"

= [# ]
"
m = !"
h[m]z! m z n

zn
H( z){
={

eigenvalue eigenfunction

System Functions H(s) or H(z) – Eigenvalues for

now, and more about them later
est H(s)est
CT: x(t) h(t) y(t)
"
H (s) = #!" h(t)e! st dt

x(t) = $ ak e sk t %
%& y(t ) = $ ak H(sk )e sk t

zn H(z)zn
DT: x[n] h[n] y[n]

"
H(z) = # n=!" h[n]z! n
x[n] = # ak znk $% y[n] = # ak H (zk )zkn
$

6
 Question 2. What kinds of signals can we
represent as “sums” of complex exponentials?

For Now: Focus on restricted sets of complex exponentials

CT: s = j! - purely imaginary ,

i.e. signals of the form e j!t
DT:
z = e j! , i.e. signals of the form e j!n
"
CT & DT Fourier Series and Transforms

Joseph Fourier (1768-1830)

8
 Fourier Series Representation of CT Periodic Signals
x(t) = x(t + T ) for all t

- smallest such T is the fundamental period

2"
- !o = is the fundamental frequency
T

e j!t periodic with period T " ! = k! o

#
+% +%
x(t) = & k =$% ak e jk! 0 t = & k = $% ak e jk 2't / T = x(t + T )

-periodic with period T

- {ak} are the Fourier (series) coefficients
- k = 0 DC
- k = ±1 first harmonic
- k = ±2 second harmonic

• For real periodic signals, there are two other commonly used
forms for CT Fourier series:
$
x(t) = ao + % [! k cosk " o t + # k sink " o t ]
k =1

or
$
x(t) = ao + % [& k cos(k " ot + ' k )]
k =1

• Because of the eigenfunction property of ejωt, we will only use

the complex exponential form in 6.003.

–– A nontrivial consequence of this is that we need to include

terms for both positive and negative frequencies:

e jk! 0 t , e " jk! 0 t

10
 Question: How do we find the Fourier coefficients ak?
Let’s first take a detour by studying a three-dimensional vector:
r
A = Ax xˆ + Ay yˆ + Az zˆ,
xˆ , yˆ , and zˆ are unit vectors.

How do we find the coefficients Ax, Ay, and Az?

r r r
Easy,
! Ax = A • x, Ay = A • y, and Az = A • zˆ .
ˆ ˆ
Az
Project the vector onto the x-, y-, and z-axis.
r
Why does it work this way? A
! Ay
Orthogonality:
Ax
yˆ • xˆ = zˆ • yˆ = xˆ • zˆ " 0
!
11

Back to Fourier series

Now let’s imagine that x(t) is a vector in a space of ∞-
dimensions and e jk" o t (k = 0, ± 1, ± 2, ...) is a unit vector in this
space. (Such a space does exist, it is called Hilbert space, just
nobody lives there.) Then,
+$
a2
x(t) =!% ak e jk" o t , with the analog y x(t)
k = #$
r
x(t) & A
a1
jk" o t
e & xˆ , yˆ , zˆ !
a0
ak & Ax , Ay , Az

Now all we need to do is to “project” x(t) onto the e jk" o t # axis ,

then we will obtain the coefficients ak. How to do that? Again,
! need orthogonality.

12
!
 Obtaining the Fourier series coefficients
Orthogonality in the Hilbert space:
1 1 '1, k = n
T
% T
e jk" o t # e$ jn" o t dt =
T
% T
e j(k$n )" o t dt = (
)0, k & n
= *[k $ n] .
( %T = integral over any interval of length T, and the operation of
1
% e$ jn" o t dt # is to take an "inner product" with e jn" o t )
T T

Now if we "project" x(t) onto e jn" o t by taking the operation :

+&
1 1
!
T
% T
x(t) # e $ jn" o t
dt =
T
% 'a e
T k
j(k$n )" o t
dt,
k = $&

then only one term (an ) will be nonzero. That is :

1
T
% T
x(t) # e$ jn" o t dt = an .
13

Finally
1
T
% T
x(t) " e# jn$ o t dt = an .

&

CT Fourier Series Pair ($ o = 2' /T)

+(
x(t) = )a e k
jk$ o t
(Synthesis equation)
k=#(

1
ak = % x(t)e# jk$ o t dt (Analysis equation)
T T

14
 Example: Periodic Square Wave

1 T /2 1 T1 / 2 " jk# t
ak =
T
$ "T / 2
x(t)e" jk# o t dt =
T
$ "T1 / 2
e o
dt

1 T1 2%
= [e" jk# o t ] (# o = )
" jk# oT "T1 T
sin(k# oT1 )
=
k%

!
15

Convergence of CT Fourier Series

• How can the Fourier series (composed of continuous sine and
cos functions) for the square wave (with many discontinuities)
possibly make sense?

• The key is: What do we mean by

+#
jk! 0 t
x(t) = $ ae k ?
k ="#

• One useful notion for engineers: there is no energy in the

difference +#
jk" 0 t
e(t) = x(t) ! $ae
k = !#
k

2
% | e(t) |
T
dt = 0

(just need x(t) to have finite energy per period)

16
 Under a different, but reasonable set of conditions
(the Dirichlet conditions)

Condition 1. x(t) is absolutely integrable over one period, i. e.

" | x(t) | dt < !

T

Condition 2. In a finite time interval,

x(t) has a finite number
of maxima and minima.
Ex. An example that violates
Condition 2.
x(t) = sin(2" /t) 0 < t # 1
Condition 3. In a finite time interval, x(t) has only a finite
number of discontinuities.
Ex. An example that violates
! Condition 3.

17

The Dirichlet Conditions (cont.)

Dirichlet conditions are met for most of the signals we will
encounter in the real world. Then
$
1
%a e k
jk" o t
=
2
[x(t + 0) + x(t # 0)]
k=#$

– The Fourier series = x(t) at points where x(t) is continuous

– The Fourier series = “midpoint” at points of discontinuity

!

• Still, convergence has some interesting characteristics:

N
jk!0 t
x N (t) = #ae k
k="N

– As N → ∞, xN(t) exhibits Gibbs’ phenomenon at points

of discontinuity
18
•Demo: Fourier Series for CT square wave
(Gibbs phenomenon).

19

Next lecture covers:

O & W pp. 202-221

20