Generalized Fourier Series

Definitions
1. A set of functions {i(t)} is said to be orthogonal over an interval [a,b] iff
Ai ( 0 or ) if i = j
  (t ) (t )dt 
b
*
i j
a
0

2. A set of functions is said to be complete over an interval [a,b] if any waveform with a finite
number of discontinuities on [a,b] can be expressed as a linear combination of members of that
set.

Theorem: The Fourier Series

Given: {i(t)} is a complete, orthogonal set on [a,b],
b
 | f (t ) | dt   , and
2
a
f(t) has a finite number of discontinuities on [a,b].

Then: f (t )   K i *  i (t ) over [a,b]
i 1

  f (t ) *  (t )dt
b
*
i
Where: K i  a

  (t ) *  (t )dt
b
*
i j
a

Proof:
Since {i(t)} is a complete set on [a,b]

f (t )   K i *  j (t ) over [a,b]
i 1

Multiplying both sides of the equation by  *j (t ) yields


f (t ) *  *j (t )   K i *  i (t ) *j (t )
i 1

Now integrating both sides over [a,b]

  f (t ) *  (t )dt    K
b  b 

*
j i *  i (t ) *  *j (t ) dt
a a
 i 1 

J. N. Denenberg 10/18/2013
Generalized Fourier Series Page 2

Reversing the order of integration and summation (linear operators) and moving the constants out of
the integral leaves

  f (t ) *  (t )dt   K *   (t ) *  (t )dt
b  b
* *
j j i j
a a
j 1

  (t ) (t )dt  0 for i  j so we are left with only the ith term of the sum
b
*
But i j
a

  f (t ) *  (t )dt  K *   (t ) *  (t )dt
b b
* *
j i i i
a a

Solving for Ki

  f (t ) *  (t )dt
b
*
i
Ki  a

  (t ) *  (t )dt
b
*
i j
a

Q. E. D.

J. N. Denenberg 10/18/2013
Generalized Fourier Series Page 3

Examples of Complete, Orthogonal Sets

{i(t)} [a,b]

{ 1, sin(n0t), cos(n0t) } ( 1  n   ) [ -T/2, T/2] ( T = 2/0 )

{ exp(jn0t) } ( -   n   ) [ -T/2, T/2] ( T = 2/0 )

Pn(t) – The Legendre Polynomials [ -1, 1 ]

P0(t) = 1
P1(t)= t
P2(t)= ½( 3t2 – 1)
P3(t)= ½( 5t3 – 3t) …

  *t 
J 0  n  - The 0th Order Bessel Function [0, a]
 a 
n is the nth root of J0(t) = 0

0 = 2.41, 1 = 5.51, 2 = 8.64, 3 = 11.79, …

  (1) k * ( x / 2) ( 2*k ) 
J 0 ( x)    

k 0  K!2 

J. N. Denenberg 10/18/2013
Generalized Fourier Series Page 4

Example: The Sawtooth or Sweep function

f(t)
f(t) = (A/T)*t
1.2 t in [0, T)
A
1

0.8

0.6
... 0.4 ...
0.2

0
-1.5 -1
-T -0.5 0 0.5 1
T 1.5 2
2T 2.5

Use { 1, sin(n0t), cos(n0t) } (0 = 2*/T)

A 
 2n *     2n *  
f (t )  * t  a0   a n * cos * t    bn * sin *t 
T n 1  T  n1  T 
First find a 0 0
0

 2n  
 2n 
f (t ) *1 dt   a0 *1 dt   a n *  cos t  dt   bn *  sin
T T T T
0 0
n 1
0
 T  n 1
0
 T 
t  dt

note: the second and third integrals are zero since the area under a sine or cosine over a
integer number of cycles is always zero. They have equal areas above and below the
axis.
T
 a0 * t
t 0

 a0 * T
Solving for a0
1 TA
T 0 T
a0  * t dt

A t2
a0 
T
* t 0
T2 2
a0 = A/2 Note: This is the average or DC value of the waveform

J. N. Denenberg 10/18/2013
Generalized Fourier Series Page 5

Now find bn for n  1

again
A 
 2n *     2n *  
* t  a0   a n * cos * t    bn * sin *t 
T n 1  T  n1  T 

 2m *  
Now multiply both sides by sin * t  and integrate over the interval
 T 

T  2m *   T  2m *  
0
f (t ) * sin 
 T
* t  dt   a 0 * sin 
 0
 T
* t  dt 


 2n   2m *  

T
a n *  cos t  * sin  * t  dt 
n 1
0
 T   T 

 2n   2m *  

T
bn *  sin  t  * sin  * t  dt
n 1
0
 T   T 
The first integral is zero as before; the second is also zero since
cos( x) * sin( y)  0.5 * sin( x  y)  sin( x  y) HW: check it out
The third integral uses another trigonometric identity
sin( x) * sin( y)  0.5 * cos( x  y)  cos( x  y)
The (x+y) terms again all yield zero and all of the (x-y) terms are also zero except for the one where
m=n. This yields
T  2m *   T  2 * 0 * 
0
f (t ) * sin
 T
* t  dt  bm  cos
 0
 T
* t  dt

But cos(0) = 1 and the integral becomes T and we have
1 T  2m *  
bm 
T 0
f (t ) * sin
 T
* t  dt

or
A T  2m *  
bm 
T2 
0
t * sin
 T
* t  dt

Integrating this requires the use of integration by parts

 udv  uv   vdu
We want to get rid of the t in the integral so let u = t and dv be the rest. This yields
A   2m *    T T  2m *   T 
bm  t  cos    cos
T
* t  * *t * dt 
T2   T   2m *   T  2m *  
0 0

J. N. Denenberg 10/18/2013
Generalized Fourier Series Page 6

The integral is once again zero for the usual reason. The first term becomes
A    2m *   
bm  T  cos * T    0
T * 2 * m *    T  
Simplifying further

*  cos2m *  
A
bm 
2 * m *
or
A
bm 
2 * m *

HW: Show that the an are all zero for n  0

J. N. Denenberg 10/18/2013
Generalized Fourier Series Page 7

The Complex Exponential Series: Another form of the Sine-Cosine Series

Euler’s Formula
 jx  cosx   j sin x 
where
j  1
and
  2.718282 - the base of ln
This also means that
 jx    jx
cos( x) 
2
and
 jx    jx
sin( x) 
2j
The unit circle represents exp(x) as a unit vector of magnitude 1 and angle 
j

sin(x) 
cos(x) 1

The complex exponential is therefore closely related to the Phasor analysis used in AC circuits and

f (t )   C n *  jn 0t
n 1

HW: find the complex exponential form for the Fourier Series of the sawtooth example
function.
Note: the answer should be consistent with
an  jbn
for n  0
Cn  2
a n  jb n
for n  0
2

C0  a0

J. N. Denenberg 10/18/2013
Generalized Fourier Series Page 8

Note: Both the Sine-Cosine and Complex Exponential sets consist of members that are periodic
functions. Each of their members has a frequency, which is a multiple of the fundamental
frequency that has a period, T, equal to the length of the region of orthogonality. Each
element is also periodic with period T and therefore their sum is also periodic.
Although we made no general statement in the theory about the behavior of the original
function outside of the region of orthogonality, the Fourier series will only represent a
periodic function. It, however, will represent a periodic function over all time.
Further note: at discontinuities, there will be some residual error

J. N. Denenberg 10/18/2013