Chapter 4 – Frequency Analysis:

The Fourier Series

Richard Heusdens

December 1, 2017 1

EE2S11
 Fourier
Fourier
Fourier series
series
series

Jean Baptiste Joseph Fourier (1768-1830)

Jean Baptiste Joseph Fourier (1768-1830)

Fourier’s idea: idea:

Fourier’s
Fourier’s idea:
any periodic
any function
periodic can can
function be writ-
be rewrit-
any periodic
ten asten function
a weighted can
sum of be rewrit-
sines and and
ten as a weighted
as a weighted sum sum of
of sines andsines
cosines of di↵erent frequencies.
cosinescosines of di↵erent
of di↵erent frequencies.
frequencies.

(not exactly
(not exactly true) true)

December 2, 2015 2
December
December 1, 2017
2, 2015 2 2

EE2S11EE2S11
EE2S11
 Fourier series
1
X 2⇡
x(t) = c0 + 2 (ck cos(k⌦0 t) + dk sin(k⌦0 t)) , ⌦0 =
T0
k=1
period
When Fourier submitted his paper in 1807, the committee (which in-
cluded Lagrange, Laplace, Malus and Legendre, among others) con-
cluded:

... the manner in which the author arrives at these equations is not
exempt of difficulties and [...] his analysis to integrate them still
leaves something to be desired on the score of generality and even
rigour.

December 1, 2017 3

EE2S11
 Fourier series

Fourier’s paper never got published, until

some 15 years later, when Fourier wrote
his own book, The Analytical Theory of
Heat (Fourier 1822).

In that book, Fourier extended his finding

to non-periodic signals, stating that such
a signal can be represented by a weighted
integral of a series of sine and cosine func-
tions. Such an integral is termed the
Fourier transform.

December 1, 2017 4

EE2S11
 What in this chapter?

• Eigenfunctions and LTI systems

• Complex and trigonometric Fourier series
• Spectrum of periodic signals
• Fourier series and Laplace transform

• Properties of Fourier series

• Convergence of Fourier series

December 1, 2017 5

EE2S11
 Eigenfunctions revisited

Consider a LTI system with input signal x(t) = es0 t , t 2 R:

Z 1
y(t) = h(⌧ )es0 (t ⌧ ) d⌧
1 transfer function
Z 1
= e s0 t h(⌧ )e s0 ⌧ d⌧ = es0 t H(s0 )
1
| {z }
H(s0 )

assuming H(s0 ) exists (s0 2 ROC)

• es0 t is called an eigenfunction of the LTI system

December 1, 2017 6

EE2S11
 Eigenfunctions revisited

Special case:
• es0 t = ej⌦0 t : harmonic signal ) y(t) = H(j⌦0 )ej⌦0 t

The function H(j⌦) is called the frequency response of the LTI sys-
tem:
y(t) = H(j⌦0 )ej⌦0 t = |H(j⌦0 )|ej(⌦0 t+\H(j⌦0 ))
• magnitude response |H(j⌦0 )| modifies the magnitude of ej⌦0 t

• phase response \H(j⌦0 ) modifies the phase of ej⌦0 t

December 1, 2017 7

EE2S11
 Fourier series and transform

Express x as a linear combination of harmonics ej⌦k t :

1
X 1
X
x(t) = Xk ej⌦k t ) y(t) = H(j⌦k )Xk ej⌦k t
k= 1 k= 1
Fourier series
or
Fourier transform
Z 1 Z 1
x(t) = X(⌦)ej⌦t d⌦ ) y(t) = H(j⌦)X(⌦)ej⌦t d⌦
1 1

December 1, 2017 8

EE2S11
 Why study this special case?

• Fourier transform is probably the most widely applied signal

processing tool in science and engineering
• Ties together two of the most used phenomenas known to en-
gineers: those of time and frequency

• Time and frequency are dual domains

• Many signal manipulations are done in the frequency domain
including filtering, sampling, modulation, etc.
• Harmonic signals appear naturally in many applications

• Steady-state analysis
December 1, 2017 9

EE2S11
 Fourier series
Can we represent/approximate any signal by sinusoids?
Approximation
Inputusing
Approximation signal9
using 21sinusoids
3
4
5
6
7
8 sinusoid Error
Errorusing 9
using0
21sinusoids
3
4
5
6
7
8 sinusoid
0.4 0.4

0.35 0.3

0.3
0.2

0.25
0.1
x(t), x N(t)

x(t), x N(t)
0.2
0
0.15
-0.1
0.1

-0.2
0.05

0 -0.3

-0.05 -0.4
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
t (sec) t (sec)

December 1, 2017 10

EE2S11
 Fourier series

Key questions:
• Does any signal has such a representation?
• How should we choose the frequencies of the constituent sinu-
soids?

• How do we find the weights?

• How many do we need, finite or infinite many?
• How does the sequence converge (in norm, pointwise, uni-
formly)?

December 1, 2017 11

EE2S11
 Orthonormal system
R2
x
(
e2 1, k=m
(ek , em ) =
(x, e2 ) e1 0, k 6= m
(x, e1 )

Pythagoras’ theorem
2
X 2
X
x= (x, ek )ek , kxk2 = |(x, ek )|2
k=1 k=1

December 1, 2017 12

EE2S11
 Orthonormal system

• The functions { k (t), t 2 [a, b]} are called orthonormal (or-

thogonal and normalized) if
Z b (
⇤
0 k 6= m
( k , m) = k (t) m (t)dt =
a 1 k=m

• To ensure the existence of the norm and the inner product,

we assume the functions have finite energy. That is, k 2
L2 ([a, b]) where
Z
L2 (E) = {f : |f (t)|2 dt < 1}
E
December 1, 2017 13

EE2S11
 Orthonormal system

Theorem: Let { k }1 k=1 denote a complete orthonormal system in

L2 (E) and let x 2 L2 (E). Then
1
X
x= (x, k) k
k=1

Moreover, we have (Parseval’s identity)

1
X
kxk2 = |(x, k )|
2

k=1

December 1, 2017 14

EE2S11
 Complex Fourier series

The Fourier series representation of a periodic signal x(t) of period

T0 is given by
X1
jk⌦0 t 2⇡
x(t) = Xk e , ⌦0 =
T0
k= 1

with Fourier coefficients Xk

Z t0 +T0
1 jk⌦0 t
Xk = x(t)e dt, k2Z
T0 t0
1
X
Moreover, we have (Parseval’s identity): kxk2 = T0 |Xk |2
k= 1
December 1, 2017 15

EE2S11
 Complex Fourier series

• Fourier series determine the frequency components of periodic

signals and how the power is distributed over the frequency
components, called the spectrum
• The spectrum of periodic signals is discrete (line spectrum)

• For real signals, the spectrum is conjugate symmetric:

Z t0 +T0
1
X k = x(t)ejk⌦0 t dt = Xk⇤
T0 t0

• As a consequence, |X k| = |Xk | and \X k = \Xk

| {z } | {z }
even symmetric odd symmetric
December 1, 2017 16

EE2S11
 Complex Fourier series

Example (periodic pulse train):

x(t)
2

t
T0 0 T0

• DC (average) value of 1

December 1, 2017 17

EE2S11
 Complex Fourier series

• Complex Fourier series

Z T0 /2 Z T0 /4
1 2
Xk = x(t)e jk⌦0 t dt = e jk⌦0 t
dt
T0 T0 /2 T0 T0 /4
8
> T0 /4
>
< 1
e jk⌦0 t , k 6= 0
= jk⇡ T0 /4
>
>
: 1, k =0
8
< 2 sin(k⇡/2), k 6= 0
= k⇡
:
1, k=0
December 1, 2017 18

EE2S11
 Complex Fourier series

Fourier series is given by

2 j⌦0 t 2 j⌦0 t 2 j3⌦0 t 2 j3⌦0 t
x(t) = 1 + e + e e e ···
⇡ ⇡ 3⇡ 3⇡

EE2S11 11
 Complex Fourier series

Note that |Xk | = |X k| and that \Xk = \X k.

Hence, we can rewrite the Fourier series as

2 j⌦0 t 2 j⌦0 t 2 j3⌦0 t 2 j3⌦0 t
x(t) = 1 + e + e e e ···
⇡ ⇡ 3⇡ 3⇡
4 4
= 1 + cos(⌦0 t) cos(3⌦0 t) + · · ·
⇡ 3⇡

December 1, 2017 20

EE2S11
 Trigonometric Fourier series

The Fourier series representation of a periodic signal x(t) of period

T0 is given by
1
X 2⇡
x(t) = c0 + 2 (ck cos(k⌦0 t) + dk sin(k⌦0 t)) , ⌦0 =
T0
k=0

with Fourier coefficients ck and dk

Z t0 +T0
1
ck = x(t) cos(k⌦0 t)dt, k = 0, 1, 2, · · ·
T0 t0
Z t0 +T0
1
dk = x(t) sin(k⌦0 t)dt, k = 1, 2, · · ·
T0 t0
December 1, 2017 21

EE2S11
 Trigonometric Fourier series

Observe that
1 j
• ck = (Xk + X k ), dk = (Xk X k)
2 2
• Xk = ck jdk , X k = ck + jdk
q ✓ ◆
dk
• |Xk | = c2k + d2k , \Xk = tan 1
ck
• if x even symmetric (x(t) = x( t)), all dk s are zero
• if x odd symmetric (x(t) = x( t)), all ck s are zero

December 1, 2017 22

EE2S11
 Trigonometric Fourier series

Example (periodic pulse train):

x(t)
2

t
T0 0 T0

• DC (average) value of 1
• x(t) is even symmetric (dk s are zero)

December 1, 2017 23

EE2S11
 Trigonometric Fourier series

• Trigonometric Fourier series

Z T0 /2 Z T0 /4
1 4
ck = x(t) cos(k⌦0 t)dt = cos(k⌦0 t)dt
T0 T0 /2 T0 0
8
< 2 sin(k⇡/2), k 6= 0
= k⇡
:
1, k=0

Hence, the Fourier series is given by

4 4
x(t) = 1 + cos(⌦0 t) cos(3⌦0 t) + · · ·
⇡ 3⇡
December 1, 2017 24

EE2S11
 Fourier series

Some observations:
• Notice that
lim Xk = lim ck = 0
k!1 k!1

Riemann-Lebesgue lemma: If x 2 L2 (E), then

Z
lim x(t)e jk⌦0 t dt = 0
k!±1 E
Z Z
lim x(t) cos(k⌦0 t)dt = lim x(t) sin(k⌦0 t)dt = 0
k!1 E k!1 E

December 1, 2017 25

EE2S11
 Fourier series

Some observations:
• Notice that
lim Xk = lim ck = 0
k!1 k!1
1
and that the decay is of order O(k )

• x(T0 /4) = 2 but the Fourier series yields 1 at t = T0 /4 ???

• The Fourier series seems to have convergence problems around

discontinuities (Gibb’s phenomena)

December 1, 2017 26

EE2S11
 Gibb’s Phenomena

4 4
x(t) = 1 + cos(⌦0 t) cos(3⌦0 t) + · · ·
⇡ 3⇡
Approximation
Approximation
Approximation
Approximationusing
using
using
using100
10
50
21sinusoids
3
4
5
6
7
8
9 sinusoid
sinusoids
sinusoids
2.5

1.5
x(t), xN (t)

0.5

-0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t (sec)

December 1, 2017 27

EE2S11
 Convergence of Fourier series

Some (important) remarks:

1
X
• Let y(t) = Xk ejk⌦0 t denotes the Fourier series represen-
k= 1
tation of x(t). Then kx yk = 0. That is
Z
|x(t) y(t)|2 dt = 0
E

• We have convergence in norm, which does not imply that it

converges pointwise to x(t)

December 1, 2017 28

EE2S11
 Relation Laplace transform

Let x1 (t) be defined as

(
x(t), t 2 [t0 , t0 + T0 ]
x1 (t) =
0, otherwise
one fundamental period
Then Z t0 +T0
st
X1 (s) = x1 (t)e dt.
t0
1
The Fourier coefficients Xk are given by Xk = X1 (s)
T0 s=jk⌦0
Z t0 +T0
1 jk⌦0 t
Xk = x1 (t)e dt
T0 t0
December 1, 2017 29

EE2S11
 Properties of Fourier Series

F
• Time-shifting: y(t) = x(t + ⌧ ) ! Yk = ejk⌦0 ⌧ Xk

Direct approach:
Z T0 Z T0
1 jk⌦0 t 1 jk⌦0 (s ⌧ )
Yk = x(t + ⌧ )e dt = x(s)e ds
T0 0 T0 0
jk⌦0 ⌧ Z T0
e jk⌦0 s
= x(s)e ds = ejk⌦0 ⌧ Xk
T0 0

Using Laplace (see Table 3.1):

1 1 ⇣ s⌧ ⌘
Yk = Y1 (s) = e X1 (s) = ejk⌦0 ⌧ Xk
T0 s=jk⌦0 T 0 s=jk⌦0
December 1, 2017 30

EE2S11
 Properties of Fourier Series

jm⌦0 t F
• Frequency-shifting: y(t) = x(t)e ! Yk = Xk+m

Direct approach:
Z T0 Z T0
1 jm⌦0 t jk⌦0 t 1 j(k+m)⌦0 t
Yk = x(t)e e dt = x(t)e dt = Xk+m
T0 0 T0 0

Using Laplace (see Table 3.1):

1 1
Yk = Y1 (s) = X1 (s + jm⌦0 ) = Xk+m
T0 s=jk⌦0 T0 s=jk⌦0

December 1, 2017 31

EE2S11
 Properties of Fourier Series

dx(t) F
• Di↵erentiation: y(t) = ! Yk = jk⌦0 Xk
dt
Direct approach:
Z T0 Z T0
1 1
Yk = x0 (t)e jk⌦0 t
dt = e jk⌦0 t d(x(t))
T0 0 T0 0
T
Z T0
1 jk⌦0 t
0 jk⌦0
= x(t)e + x(t)e jk⌦0 t dt = jk⌦0 Xk
T0 0 T0 0
Using Laplace (see Table 3.1):
1 1 ⇣ ⌘
Yk = Y1 (s) = sX1 (s) = jk⌦0 Xk
T0 s=jk⌦0 T0 s=jk⌦0

December 1, 2017 32

EE2S11
 Convergence of the Fourier Series

• Fourier series can be defined for functions x 2 L1 (E) L2 (E)

• The pointwise convergence of a Fourier series is a rather com-
plicated problem
(Dirichlet conditions)
– Dirichlet (1829) showed that if x 2 L1 (E) and has a finite
number of discontinuities and extrema, then the Fourier
series converges everywhere to the local average
– Kolmogorov (1926) has given an example of a function in
L1 (E) in which the Fourier series diverges everywhere!
– Carleson (1966) proved that if x 2 L2 (E), then the
Fourier series converges for almost all t to x(t)
December 1, 2017 33

EE2S11
 Convergence of the Fourier Series

Dirichlet’s theorem: if x 2 L1 ([0, T0 ]), then

Z T0
1
SN (t) = x(u)DN (u t)du
T0 0
50

where
sin((N + 12 )⌦0 t) 40
N = 20
DN (t) =
sin( 12 ⌦0 t)
30

20

is called Dirichlet’s kernel 10

-10
-10 -5 0 5 10

December 1, 2017 34

EE2S11
 Convergence of the Fourier Series

Conclusion:
• Convergence of Fourier series depends on local behaviour

• For large N , the Dirichlet kernel becomes a -function:

1
lim SN (t0 ) = x(t0 ) + x(t+
0)
N !1 2
If x is continuous at t = t0 , then lim SN (t0 ) = x(t0 )
N !1

In conclusion, the Fourier series converges in norm (we have equality

in L2 ([0, T0 ])), but we only have pointwise convergence at points
where x is continuous!
December 1, 2017 35

EE2S11
 Convergence of the Fourier Series
⌦0 = 1
Example: x(t) = t, t 2 [ ⇡, ⇡]. Since x is odd, ck = 0 for all k and
Z ⇡
1
dk = t sin(kt)dt
2⇡ ⇡
⇡ Z ⇡
1 1 1
= t cos(kt) + cos(kt)dt = ( 1)k+1
2⇡k ⇡ 2⇡k ⇡ k
| {z }
=0

Hence, the Fourier series becomes

2 2 2 2
sin(t) sin(2t) + sin(3t) sin(4t) + · · ·
1 2 3 4
December 1, 2017 36

EE2S11
 Convergence of the Fourier Series

1
• Notice that lim dk = 0 and that the decay is O(k )
k!1

• x(⇡) = ⇡ but substituting t = ⇡ in the Fourier series yields 0

dk
1

-1

-2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

k
December 1, 2017 37

EE2S11
 Convergence of the
Fourier Fourier Series
Series
4 4

2 2

0 0

2 2

4 4
π π/2 0 π/2 π π π/2 0 π/2 π

4 4

2 2

0 0

2 2

4 4
π π/2 0 π/2 π π π/2 0 π/2 π
December 1, 2017
April 28, 2015 38
14

EE2S11
EE2S31
 Convergence of the Fourier Series

Example: x(t) = |t|, t 2 [ ⇡, ⇡]. Since x is even, dk = 0 for all k

Z ⇡
1
ck = t cos(kt)dt
⇡ 0
Z ⇡ (
⇡ 0, k = 2, 4, . . .
(k 6= 0) 1 1
= t sin(kt) sin(kt)dt =
k⇡ k⇡ 0 2
| {z 0
} ⇡k2 , k odd
=0

and Z ⇡
1 ⇡
c0 = tdt =
⇡ 0 2

December 1, 2017 39

EE2S11
 Convergence of the Fourier Series

The Fourier series becomes

✓ ◆
⇡ 4 cos(t) cos(3t) cos(5t)
2
+ 2
+ 2
+ ···
2 ⇡ 1 3 5
2
Notice that lim ck = 0 and that the decay is of order O(k )
k!1

2
ck
1

-1

-2
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
k
December 1, 2017 40

EE2S11
 Convergence of the
Fourier Fourier Series
Series
4 4

3 3

2 2

1 1

0 0
π π/2 0 π/2 π π π/2 0 π/2 π

4 4

3 3

2 2

1 1

0 0
π π/2 0 π/2 π π π/2 0 π/2 π
December 1, 2017
April 28, 2015 41
19

EE2S11
EE2S31
 Convergence of the Fourier Series

• If x is p times di↵erentiable and all derivatives are in L1 (E),

then
F
x(p) (t) ! (jk⌦0 )p Xk

• Applying the Riemann-Lebesgue lemma on x(p) , we conclude

that
lim (k⌦0 )p Xk = 0
k!±1

so that regularity of x translates to rapid decay of Xk

This explains the faster decay of the Fourier coefficients of the func-
tion x(t) = |t| as compared to those of x(t) = t.
December 1, 2017 42

EE2S11
 Convergence of the Fourier Series

December 1, 2017 43

EE2S11
 What have we accomplished?

• Response of LTI systems to periodic signals (eigenfunction prop-

erty)
• Harmonic (sinusoidal) representation of periodic/finite-length
signals

• Spectrum of periodic/finite-length signals

• Connection between Fourier and Laplace
• Convergence properties of Fourier series

December 1, 2017 44

EE2S11
 Where do we go?

• Extension of Fourier representation for aperiodic/infinite-length

signals
• Unification of spectral theory for periodic and aperiodic signals
• Connection between Fourier and Laplace transforms

• Duality relation time and frequency domain

• Convolution and filtering
• Relation between pole/zero locations and frequency response

December 1, 2017 45

EE2S11